From 28ee86a9b0f2e980beeb637da4f787065c74a39e Mon Sep 17 00:00:00 2001 From: Sebastian Huber Date: Wed, 27 Apr 2016 11:58:19 +0200 Subject: Import DPAA driver snapshot Imported from Freescale Linux repository git://git.freescale.com/ppc/upstream/linux.git commit 2774c204cd8bfc56a200ff4dcdfc9cdf5b6fc161. Linux compatibility layer is partly from FreeBSD. --- linux/drivers/soc/fsl/qbman/bman-debugfs.c | 121 + linux/drivers/soc/fsl/qbman/bman.c | 692 ++++++ linux/drivers/soc/fsl/qbman/bman.h | 542 +++++ linux/drivers/soc/fsl/qbman/bman_api.c | 1123 +++++++++ linux/drivers/soc/fsl/qbman/bman_portal.c | 399 ++++ linux/drivers/soc/fsl/qbman/bman_priv.h | 136 ++ linux/drivers/soc/fsl/qbman/bman_test.c | 60 + linux/drivers/soc/fsl/qbman/bman_test.h | 34 + linux/drivers/soc/fsl/qbman/bman_test_api.c | 188 ++ linux/drivers/soc/fsl/qbman/bman_test_thresh.c | 216 ++ linux/drivers/soc/fsl/qbman/bman_utils.c | 76 + linux/drivers/soc/fsl/qbman/dpaa_resource.c | 363 +++ linux/drivers/soc/fsl/qbman/dpaa_sys.h | 292 +++ linux/drivers/soc/fsl/qbman/qman-debugfs.c | 1317 +++++++++++ linux/drivers/soc/fsl/qbman/qman.c | 1106 +++++++++ linux/drivers/soc/fsl/qbman/qman.h | 1133 +++++++++ linux/drivers/soc/fsl/qbman/qman_api.c | 3026 ++++++++++++++++++++++++ linux/drivers/soc/fsl/qbman/qman_driver.c | 87 + linux/drivers/soc/fsl/qbman/qman_portal.c | 796 +++++++ linux/drivers/soc/fsl/qbman/qman_priv.h | 293 +++ linux/drivers/soc/fsl/qbman/qman_test.c | 61 + linux/drivers/soc/fsl/qbman/qman_test.h | 44 + linux/drivers/soc/fsl/qbman/qman_test_api.c | 222 ++ linux/drivers/soc/fsl/qbman/qman_test_stash.c | 540 +++++ linux/drivers/soc/fsl/qbman/qman_utils.c | 309 +++ 25 files changed, 13176 insertions(+) create mode 100644 linux/drivers/soc/fsl/qbman/bman-debugfs.c create mode 100644 linux/drivers/soc/fsl/qbman/bman.c create mode 100644 linux/drivers/soc/fsl/qbman/bman.h create mode 100644 linux/drivers/soc/fsl/qbman/bman_api.c create mode 100644 linux/drivers/soc/fsl/qbman/bman_portal.c create mode 100644 linux/drivers/soc/fsl/qbman/bman_priv.h create mode 100644 linux/drivers/soc/fsl/qbman/bman_test.c create mode 100644 linux/drivers/soc/fsl/qbman/bman_test.h create mode 100644 linux/drivers/soc/fsl/qbman/bman_test_api.c create mode 100644 linux/drivers/soc/fsl/qbman/bman_test_thresh.c create mode 100644 linux/drivers/soc/fsl/qbman/bman_utils.c create mode 100644 linux/drivers/soc/fsl/qbman/dpaa_resource.c create mode 100644 linux/drivers/soc/fsl/qbman/dpaa_sys.h create mode 100644 linux/drivers/soc/fsl/qbman/qman-debugfs.c create mode 100644 linux/drivers/soc/fsl/qbman/qman.c create mode 100644 linux/drivers/soc/fsl/qbman/qman.h create mode 100644 linux/drivers/soc/fsl/qbman/qman_api.c create mode 100644 linux/drivers/soc/fsl/qbman/qman_driver.c create mode 100644 linux/drivers/soc/fsl/qbman/qman_portal.c create mode 100644 linux/drivers/soc/fsl/qbman/qman_priv.h create mode 100644 linux/drivers/soc/fsl/qbman/qman_test.c create mode 100644 linux/drivers/soc/fsl/qbman/qman_test.h create mode 100644 linux/drivers/soc/fsl/qbman/qman_test_api.c create mode 100644 linux/drivers/soc/fsl/qbman/qman_test_stash.c create mode 100644 linux/drivers/soc/fsl/qbman/qman_utils.c (limited to 'linux/drivers/soc/fsl/qbman') diff --git a/linux/drivers/soc/fsl/qbman/bman-debugfs.c b/linux/drivers/soc/fsl/qbman/bman-debugfs.c new file mode 100644 index 00000000..09f5a283 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman-debugfs.c @@ -0,0 +1,121 @@ +#include + +#include + +/* Copyright 2010 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_priv.h" + +static struct dentry *dfs_root; /* debugfs root directory */ + +/* Query Buffer Pool State */ + +static int query_bp_state_show(struct seq_file *file, void *offset) +{ + int ret; + struct bm_pool_state state; + int i, j; + u32 mask; + + memset(&state, 0, sizeof(state)); + ret = bman_query_pools(&state); + if (ret) { + seq_printf(file, "Error %d\n", ret); + return ret; + } + + seq_puts(file, "bp_id free_buffers_avail bp_depleted\n"); + for (i = 0; i < 2; i++) { + mask = 0x80000000; + for (j = 0; j < 32; j++) { + seq_printf(file, + " %-2u %-3s %-3s\n", + (i * 32) + j, + state.as.state.__state[i] & mask ? "no" : "yes", + state.ds.state.__state[i] & mask ? "yes" : "no"); + mask >>= 1; + } + } + + return 0; +} + +static int query_bp_state_open(struct inode *inode, struct file *file) +{ + return single_open(file, query_bp_state_show, NULL); +} + +static const struct file_operations query_bp_state_fops = { + .owner = THIS_MODULE, + .open = query_bp_state_open, + .read = seq_read, + .release = single_release, +}; + +static int __init bman_debugfs_init(void) +{ + int ret = 0; + struct dentry *d; + + dfs_root = debugfs_create_dir("bman", NULL); + if (dfs_root == NULL) { + pr_err("Cannot create dir\n"); + return -ENOMEM; + } + + d = debugfs_create_file("query_bp_state", + S_IRUGO, + dfs_root, + NULL, + &query_bp_state_fops); + if (d == NULL) { + ret = -ENOMEM; + pr_err("Cannot create query_bp_state\n"); + goto _return; + } + + return 0; + +_return: + debugfs_remove_recursive(dfs_root); + + return ret; +} + +static void __exit bman_debugfs_exit(void) +{ + debugfs_remove_recursive(dfs_root); +} + +module_init(bman_debugfs_init); +module_exit(bman_debugfs_exit); + +MODULE_LICENSE("Dual BSD/GPL"); diff --git a/linux/drivers/soc/fsl/qbman/bman.c b/linux/drivers/soc/fsl/qbman/bman.c new file mode 100644 index 00000000..35459bc7 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman.c @@ -0,0 +1,692 @@ +#include + +#include + +/* Copyright (c) 2009 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_priv.h" + +/* Last updated for v00.79 of the BG */ + +struct bman; + +/* Register offsets */ +#define REG_POOL_SWDET(n) (0x0000 + ((n) * 0x04)) +#define REG_POOL_HWDET(n) (0x0100 + ((n) * 0x04)) +#define REG_POOL_SWDXT(n) (0x0200 + ((n) * 0x04)) +#define REG_POOL_HWDXT(n) (0x0300 + ((n) * 0x04)) +#define REG_POOL_CONTENT(n) (0x0600 + ((n) * 0x04)) +#define REG_FBPR_FPC 0x0800 +#define REG_ECSR 0x0a00 +#define REG_ECIR 0x0a04 +#define REG_EADR 0x0a08 +#define REG_EDATA(n) (0x0a10 + ((n) * 0x04)) +#define REG_SBEC(n) (0x0a80 + ((n) * 0x04)) +#define REG_IP_REV_1 0x0bf8 +#define REG_IP_REV_2 0x0bfc +#define REG_FBPR_BARE 0x0c00 +#define REG_FBPR_BAR 0x0c04 +#define REG_FBPR_AR 0x0c10 +#define REG_SRCIDR 0x0d04 +#define REG_LIODNR 0x0d08 +#define REG_ERR_ISR 0x0e00 /* + "enum bm_isr_reg" */ + +/* Used by all error interrupt registers except 'inhibit' */ +#define BM_EIRQ_IVCI 0x00000010 /* Invalid Command Verb */ +#define BM_EIRQ_FLWI 0x00000008 /* FBPR Low Watermark */ +#define BM_EIRQ_MBEI 0x00000004 /* Multi-bit ECC Error */ +#define BM_EIRQ_SBEI 0x00000002 /* Single-bit ECC Error */ +#define BM_EIRQ_BSCN 0x00000001 /* pool State Change Notification */ + +/* BMAN_ECIR valid error bit */ +#define PORTAL_ECSR_ERR (BM_EIRQ_IVCI) + +union bman_ecir { + u32 ecir_raw; + struct { + u32 __reserved1:4; + u32 portal_num:4; + u32 __reserved2:12; + u32 numb:4; + u32 __reserved3:2; + u32 pid:6; + } __packed info; +}; + +union bman_eadr { + u32 eadr_raw; + struct { + u32 __reserved1:5; + u32 memid:3; + u32 __reserved2:14; + u32 eadr:10; + } __packed info; +}; + +struct bman_hwerr_txt { + u32 mask; + const char *txt; +}; + +#define BMAN_HWE_TXT(a, b) { .mask = BM_EIRQ_##a, .txt = b } + +static const struct bman_hwerr_txt bman_hwerr_txts[] = { + BMAN_HWE_TXT(IVCI, "Invalid Command Verb"), + BMAN_HWE_TXT(FLWI, "FBPR Low Watermark"), + BMAN_HWE_TXT(MBEI, "Multi-bit ECC Error"), + BMAN_HWE_TXT(SBEI, "Single-bit ECC Error"), + BMAN_HWE_TXT(BSCN, "Pool State Change Notification"), +}; +#define BMAN_HWE_COUNT (sizeof(bman_hwerr_txts)/sizeof(struct bman_hwerr_txt)) + +struct bman_error_info_mdata { + u16 addr_mask; + u16 bits; + const char *txt; +}; + +#define BMAN_ERR_MDATA(a, b, c) { .addr_mask = a, .bits = b, .txt = c} +static const struct bman_error_info_mdata error_mdata[] = { + BMAN_ERR_MDATA(0x03FF, 192, "Stockpile memory"), + BMAN_ERR_MDATA(0x00FF, 256, "SW portal ring memory port 1"), + BMAN_ERR_MDATA(0x00FF, 256, "SW portal ring memory port 2"), +}; +#define BMAN_ERR_MDATA_COUNT \ + (sizeof(error_mdata)/sizeof(struct bman_error_info_mdata)) + +/* Add this in Kconfig */ +#define BMAN_ERRS_TO_UNENABLE (BM_EIRQ_FLWI) + +/** + * bm_err_isr__ - Manipulate global interrupt registers + * @v: for accessors that write values, this is the 32-bit value + * + * Manipulates BMAN_ERR_ISR, BMAN_ERR_IER, BMAN_ERR_ISDR, BMAN_ERR_IIR. All + * manipulations except bm_err_isr_[un]inhibit() use 32-bit masks composed of + * the BM_EIRQ_*** definitions. Note that "bm_err_isr_enable_write" means + * "write the enable register" rather than "enable the write register"! + */ +#define bm_err_isr_status_read(bm) \ + __bm_err_isr_read(bm, bm_isr_status) +#define bm_err_isr_status_clear(bm, m) \ + __bm_err_isr_write(bm, bm_isr_status, m) +#define bm_err_isr_enable_read(bm) \ + __bm_err_isr_read(bm, bm_isr_enable) +#define bm_err_isr_enable_write(bm, v) \ + __bm_err_isr_write(bm, bm_isr_enable, v) +#define bm_err_isr_disable_read(bm) \ + __bm_err_isr_read(bm, bm_isr_disable) +#define bm_err_isr_disable_write(bm, v) \ + __bm_err_isr_write(bm, bm_isr_disable, v) +#define bm_err_isr_inhibit(bm) \ + __bm_err_isr_write(bm, bm_isr_inhibit, 1) +#define bm_err_isr_uninhibit(bm) \ + __bm_err_isr_write(bm, bm_isr_inhibit, 0) + +#ifndef __rtems__ +static u16 bman_pool_max; +#else /* __rtems__ */ +/* FIXME */ +extern u16 bman_ip_rev; +extern u16 bman_pool_max; +#endif /* __rtems__ */ + +/* + * TODO: unimplemented registers + * + * BMAN_POOLk_SDCNT, BMAN_POOLk_HDCNT, BMAN_FULT, + * BMAN_VLDPL, BMAN_EECC, BMAN_SBET, BMAN_EINJ + */ + +/* Encapsulate "struct bman *" as a cast of the register space address. */ + +static struct bman *bm_create(void *regs) +{ + return (struct bman *)regs; +} + +static inline u32 __bm_in(struct bman *bm, u32 offset) +{ + return ioread32be((void *)bm + offset); +} +static inline void __bm_out(struct bman *bm, u32 offset, u32 val) +{ + iowrite32be(val, (void*) bm + offset); +} +#define bm_in(reg) __bm_in(bm, REG_##reg) +#define bm_out(reg, val) __bm_out(bm, REG_##reg, val) + +static u32 __bm_err_isr_read(struct bman *bm, enum bm_isr_reg n) +{ + return __bm_in(bm, REG_ERR_ISR + (n << 2)); +} + +static void __bm_err_isr_write(struct bman *bm, enum bm_isr_reg n, u32 val) +{ + __bm_out(bm, REG_ERR_ISR + (n << 2), val); +} + +static void bm_get_version(struct bman *bm, u16 *id, u8 *major, u8 *minor) +{ + u32 v = bm_in(IP_REV_1); + *id = (v >> 16); + *major = (v >> 8) & 0xff; + *minor = v & 0xff; +} + +static u32 __generate_thresh(u32 val, int roundup) +{ + u32 e = 0; /* co-efficient, exponent */ + int oddbit = 0; + + while (val > 0xff) { + oddbit = val & 1; + val >>= 1; + e++; + if (roundup && oddbit) + val++; + } + DPA_ASSERT(e < 0x10); + return val | (e << 8); +} + +static void bm_set_pool(struct bman *bm, u8 pool, u32 swdet, u32 swdxt, + u32 hwdet, u32 hwdxt) +{ + DPA_ASSERT(pool < bman_pool_max); + + bm_out(POOL_SWDET(pool), __generate_thresh(swdet, 0)); + bm_out(POOL_SWDXT(pool), __generate_thresh(swdxt, 1)); + bm_out(POOL_HWDET(pool), __generate_thresh(hwdet, 0)); + bm_out(POOL_HWDXT(pool), __generate_thresh(hwdxt, 1)); +} + +static void bm_set_memory(struct bman *bm, u64 ba, int prio, u32 size) +{ + u32 exp = ilog2(size); + /* choke if size isn't within range */ + DPA_ASSERT((size >= 4096) && (size <= 1073741824) && + is_power_of_2(size)); + /* choke if '[e]ba' has lower-alignment than 'size' */ + DPA_ASSERT(!(ba & (size - 1))); + bm_out(FBPR_BARE, upper_32_bits(ba)); + bm_out(FBPR_BAR, lower_32_bits(ba)); + bm_out(FBPR_AR, (prio ? 0x40000000 : 0) | (exp - 1)); +} + +/*****************/ +/* Config driver */ +/*****************/ + +/* We support only one of these. */ +static struct bman *bm; + +/* And this state belongs to 'bm' */ +#ifndef __rtems__ +static dma_addr_t fbpr_a; +static size_t fbpr_sz; + +static int bman_fbpr(struct reserved_mem *rmem) +{ + fbpr_a = rmem->base; + fbpr_sz = rmem->size; + + WARN_ON(!(fbpr_a && fbpr_sz)); + + return 0; +} +RESERVEDMEM_OF_DECLARE(bman_fbpr, "fsl,bman-fbpr", bman_fbpr); +#else /* __rtems__ */ +static DPAA_NOCACHENOLOAD_ALIGNED_REGION(fbpr, 16777216); +#define fbpr_a ((uintptr_t)&fbpr[0]) +#define fbpr_sz sizeof(fbpr) +#endif /* __rtems__ */ + +int bm_pool_set(u32 bpid, const u32 *thresholds) +{ + if (!bm) + return -ENODEV; + bm_set_pool(bm, bpid, thresholds[0], thresholds[1], + thresholds[2], thresholds[3]); + return 0; +} +EXPORT_SYMBOL(bm_pool_set); + +static void log_edata_bits(u32 bit_count) +{ + u32 i, j, mask = 0xffffffff; + + pr_warn("ErrInt, EDATA:\n"); + i = bit_count/32; + if (bit_count%32) { + i++; + mask = ~(mask << bit_count%32); + } + j = 16-i; + pr_warn(" 0x%08x\n", bm_in(EDATA(j)) & mask); + j++; + for (; j < 16; j++) + pr_warn(" 0x%08x\n", bm_in(EDATA(j))); +} + +static void log_additional_error_info(u32 isr_val, u32 ecsr_val) +{ + union bman_ecir ecir_val; + union bman_eadr eadr_val; + + ecir_val.ecir_raw = bm_in(ECIR); + /* Is portal info valid */ + if (ecsr_val & PORTAL_ECSR_ERR) { + pr_warn("ErrInt: SWP id %d, numb %d, pid %d\n", + ecir_val.info.portal_num, ecir_val.info.numb, + ecir_val.info.pid); + } + if (ecsr_val & (BM_EIRQ_SBEI|BM_EIRQ_MBEI)) { + eadr_val.eadr_raw = bm_in(EADR); + pr_warn("ErrInt: EADR Memory: %s, 0x%x\n", + error_mdata[eadr_val.info.memid].txt, + error_mdata[eadr_val.info.memid].addr_mask + & eadr_val.info.eadr); + log_edata_bits(error_mdata[eadr_val.info.memid].bits); + } +} + +/* BMan interrupt handler */ +static irqreturn_t bman_isr(int irq, void *ptr) +{ + u32 isr_val, ier_val, ecsr_val, isr_mask, i; + + ier_val = bm_err_isr_enable_read(bm); + isr_val = bm_err_isr_status_read(bm); + ecsr_val = bm_in(ECSR); + isr_mask = isr_val & ier_val; + + if (!isr_mask) + return IRQ_NONE; + + for (i = 0; i < BMAN_HWE_COUNT; i++) { + if (bman_hwerr_txts[i].mask & isr_mask) { + pr_warn("ErrInt: %s\n", bman_hwerr_txts[i].txt); + if (bman_hwerr_txts[i].mask & ecsr_val) { + log_additional_error_info(isr_mask, ecsr_val); + /* Re-arm error capture registers */ + bm_out(ECSR, ecsr_val); + } + if (bman_hwerr_txts[i].mask & BMAN_ERRS_TO_UNENABLE) { + pr_devel("Un-enabling error 0x%x\n", + bman_hwerr_txts[i].mask); + ier_val &= ~bman_hwerr_txts[i].mask; + bm_err_isr_enable_write(bm, ier_val); + } + } + } + bm_err_isr_status_clear(bm, isr_val); + + return IRQ_HANDLED; +} + +u32 bm_pool_free_buffers(u32 bpid) +{ + return bm_in(POOL_CONTENT(bpid)); +} +EXPORT_SYMBOL(bm_pool_free_buffers); + +#ifndef __rtems__ +static ssize_t show_fbpr_fpc(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", bm_in(FBPR_FPC)); +}; + +static ssize_t show_pool_count(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + u32 data; + int i; + + if (kstrtoint(dev_attr->attr.name, 10, &i)) + return -EINVAL; + data = bm_in(POOL_CONTENT(i)); + return snprintf(buf, PAGE_SIZE, "%d\n", data); +}; + +static ssize_t show_err_isr(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "0x%08x\n", bm_in(ERR_ISR)); +}; + +static ssize_t show_sbec(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + int i; + + if (sscanf(dev_attr->attr.name, "sbec_%d", &i) != 1) + return -EINVAL; + return snprintf(buf, PAGE_SIZE, "%u\n", bm_in(SBEC(i))); +}; + +static DEVICE_ATTR(err_isr, S_IRUSR, show_err_isr, NULL); +static DEVICE_ATTR(fbpr_fpc, S_IRUSR, show_fbpr_fpc, NULL); + +/* Didn't use DEVICE_ATTR as 64 of this would be required. + * Initialize them when needed. */ +static char *name_attrs_pool_count; /* "xx" + null-terminator */ +static struct device_attribute *dev_attr_buffer_pool_count; + +static DEVICE_ATTR(sbec_0, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_1, S_IRUSR, show_sbec, NULL); + +static struct attribute *bman_dev_attributes[] = { + &dev_attr_fbpr_fpc.attr, + &dev_attr_err_isr.attr, + NULL +}; + +static struct attribute *bman_dev_ecr_attributes[] = { + &dev_attr_sbec_0.attr, + &dev_attr_sbec_1.attr, + NULL +}; + +static struct attribute **bman_dev_pool_count_attributes; + +/* root level */ +static const struct attribute_group bman_dev_attr_grp = { + .name = NULL, + .attrs = bman_dev_attributes +}; +static const struct attribute_group bman_dev_ecr_grp = { + .name = "error_capture", + .attrs = bman_dev_ecr_attributes +}; +static struct attribute_group bman_dev_pool_countent_grp = { + .name = "pool_count", +}; + +static int of_fsl_bman_remove(struct platform_device *ofdev) +{ + sysfs_remove_group(&ofdev->dev.kobj, &bman_dev_attr_grp); + return 0; +}; +#endif /* __rtems__ */ + +static int of_fsl_bman_probe(struct platform_device *ofdev) +{ + int ret, err_irq, i; + struct device *dev = &ofdev->dev; + struct device_node *node = dev->of_node; + struct resource res; + u32 __iomem *regs; + u16 id; + u8 major, minor; + + if (!of_device_is_available(node)) + return -ENODEV; + + ret = of_address_to_resource(node, 0, &res); + if (ret) { + dev_err(dev, "Can't get %s property 'reg'\n", node->full_name); + return ret; + } + regs = devm_ioremap(dev, res.start, res.end - res.start + 1); + if (!regs) + return -ENXIO; + + bm = bm_create(regs); + + bm_get_version(bm, &id, &major, &minor); + dev_info(dev, "Bman ver:%04x,%02x,%02x\n", id, major, minor); + if ((major == 1) && (minor == 0)) + bman_pool_max = 64; + else if ((major == 2) && (minor == 0)) + bman_pool_max = 8; + else if ((major == 2) && (minor == 1)) + bman_pool_max = 64; + else + dev_warn(dev, "unknown Bman version, default to rev1.0\n"); +#ifdef __rtems__ + bman_ip_rev = (u16)((major << 8) | minor); +#endif /* __rtems__ */ + + + bm_set_memory(bm, fbpr_a, 0, fbpr_sz); + + err_irq = of_irq_to_resource(node, 0, NULL); + if (err_irq == NO_IRQ) { + dev_info(dev, "Can't get %s property 'interrupts'\n", + node->full_name); + return -ENODEV; + } + ret = devm_request_irq(dev, err_irq, bman_isr, IRQF_SHARED, "bman-err", + node); + if (ret) { + dev_err(dev, "devm_request_irq() failed %d for '%s'\n", + ret, node->full_name); + return ret; + } + /* Disable Buffer Pool State Change */ + bm_err_isr_disable_write(bm, BM_EIRQ_BSCN); + /* Write-to-clear any stale bits, (eg. starvation being asserted prior + * to resource allocation during driver init). */ + bm_err_isr_status_clear(bm, 0xffffffff); + /* Enable Error Interrupts */ + bm_err_isr_enable_write(bm, 0xffffffff); + +#ifndef __rtems__ + ret = sysfs_create_group(&dev->kobj, &bman_dev_attr_grp); + if (ret) + goto done; + ret = sysfs_create_group(&dev->kobj, &bman_dev_ecr_grp); + if (ret) + goto del_group_0; + + name_attrs_pool_count = devm_kmalloc(dev, + sizeof(char) * bman_pool_max * 3, GFP_KERNEL); + if (!name_attrs_pool_count) + goto del_group_1; + + dev_attr_buffer_pool_count = devm_kmalloc(dev, + sizeof(struct device_attribute) * bman_pool_max, GFP_KERNEL); + if (!dev_attr_buffer_pool_count) + goto del_group_1; + + bman_dev_pool_count_attributes = devm_kmalloc(dev, + sizeof(struct attribute *) * (bman_pool_max + 1), GFP_KERNEL); + if (!bman_dev_pool_count_attributes) + goto del_group_1; + + for (i = 0; i < bman_pool_max; i++) { + ret = scnprintf((name_attrs_pool_count + i * 3), 3, "%d", i); + if (!ret) + goto del_group_1; + dev_attr_buffer_pool_count[i].attr.name = + (name_attrs_pool_count + i * 3); + dev_attr_buffer_pool_count[i].attr.mode = S_IRUSR; + dev_attr_buffer_pool_count[i].show = show_pool_count; + bman_dev_pool_count_attributes[i] = + &dev_attr_buffer_pool_count[i].attr; + } + bman_dev_pool_count_attributes[bman_pool_max] = NULL; + + bman_dev_pool_countent_grp.attrs = bman_dev_pool_count_attributes; + + ret = sysfs_create_group(&dev->kobj, &bman_dev_pool_countent_grp); + if (ret) + goto del_group_1; + + goto done; + +del_group_1: + sysfs_remove_group(&dev->kobj, &bman_dev_ecr_grp); +del_group_0: + sysfs_remove_group(&dev->kobj, &bman_dev_attr_grp); +done: + if (ret) + dev_err(dev, "Cannot create dev attributes ret=%d\n", ret); +#else /* __rtems__ */ + (void)i; +#endif /* __rtems__ */ + + return ret; +}; + +#ifndef __rtems__ +static const struct of_device_id of_fsl_bman_ids[] = { + { + .compatible = "fsl,bman", + }, + {} +}; + +static struct platform_driver of_fsl_bman_driver = { + .driver = { + .name = KBUILD_MODNAME, + .of_match_table = of_fsl_bman_ids, + }, + .probe = of_fsl_bman_probe, + .remove = of_fsl_bman_remove, +}; + +builtin_platform_driver(of_fsl_bman_driver); +#else /* __rtems__ */ +#include +#include +#include +#include +#include + +static struct bm_portal_config bman_configs[NR_CPUS]; + +u16 bman_ip_rev; + +u16 bman_pool_max; + +SYSINIT_REFERENCE(irqs); + +static void +bman_sysinit(void) +{ + const char *fdt = bsp_fdt_get(); + struct device_node dn; + struct platform_device ofdev = { + .dev = { + .of_node = &dn, + .base = (uintptr_t)&qoriq + } + }; + const char *name; + int cpu_count = (int)rtems_get_processor_count(); + int cpu; + int ret; + int node; + int parent; + + qoriq_reset_qman_and_bman(); + qoriq_clear_ce_portal(&qoriq_bman_portal[0][0], + sizeof(qoriq_bman_portal[0])); + qoriq_clear_ci_portal(&qoriq_bman_portal[1][0], + sizeof(qoriq_bman_portal[1])); + + memset(&dn, 0, sizeof(dn)); + + name = "fsl,bman"; + node = fdt_node_offset_by_compatible(fdt, 0, name); + if (node < 0) + panic("bman: no bman in FDT"); + + dn.full_name = name; + dn.offset = node; + ret = of_fsl_bman_probe(&ofdev); + if (ret != 0) + panic("bman: probe failed"); + + name = "fsl,bman-portal"; + node = fdt_node_offset_by_compatible(fdt, 0, name); + if (node < 0) + panic("bman: no portals in FDT"); + parent = fdt_parent_offset(fdt, node); + if (parent < 0) + panic("bman: no parent of portals in FDT"); + node = fdt_first_subnode(fdt, parent); + + dn.full_name = name; + dn.offset = node; + + for (cpu = 0; cpu < cpu_count; ++cpu) { + struct bm_portal_config *pcfg = &bman_configs[cpu]; + struct bman_portal *portal; + struct resource res; + + if (node < 0) + panic("bman: missing portal in FDT"); + + ret = of_address_to_resource(&dn, 0, &res); + if (ret != 0) + panic("bman: no portal CE address"); + pcfg->addr_virt[0] = (__iomem void *) + ((uintptr_t)&qoriq_bman_portal[0][0] + (uintptr_t)res.start); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[0] >= + (uintptr_t)&qoriq_bman_portal[0][0]); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[0] < + (uintptr_t)&qoriq_bman_portal[1][0]); + + ret = of_address_to_resource(&dn, 1, &res); + if (ret != 0) + panic("bman: no portal CI address"); + pcfg->addr_virt[1] = (__iomem void *) + ((uintptr_t)&qoriq_bman_portal[0][0] + (uintptr_t)res.start); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[1] >= + (uintptr_t)&qoriq_bman_portal[1][0]); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[1] < + (uintptr_t)&qoriq_bman_portal[2][0]); + + pcfg->public_cfg.irq = of_irq_to_resource(&dn, 0, NULL); + if (pcfg->public_cfg.irq == NO_IRQ) + panic("bman: no portal interrupt"); + + pcfg->public_cfg.cpu = cpu; + bman_depletion_fill(&pcfg->public_cfg.mask); + + portal = bman_create_affine_portal(pcfg); + if (portal == NULL) + panic("bman: cannot create portal"); + + bman_p_irqsource_add(portal, BM_PIRQ_RCRI | BM_PIRQ_BSCN); + + node = fdt_next_subnode(fdt, node); + dn.offset = node; + } + + bman_seed_bpid_range(0, bman_pool_max); +} +SYSINIT(bman_sysinit, SI_SUB_CPU, SI_ORDER_FIRST, bman_sysinit, NULL); +#endif /* __rtems__ */ diff --git a/linux/drivers/soc/fsl/qbman/bman.h b/linux/drivers/soc/fsl/qbman/bman.h new file mode 100644 index 00000000..c9879386 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman.h @@ -0,0 +1,542 @@ +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_priv.h" + +extern u16 bman_pool_max; + +/***************************/ +/* Portal register assists */ +/***************************/ + +/* Cache-inhibited register offsets */ +#define BM_REG_RCR_PI_CINH 0x0000 +#define BM_REG_RCR_CI_CINH 0x0004 +#define BM_REG_RCR_ITR 0x0008 +#define BM_REG_CFG 0x0100 +#define BM_REG_SCN(n) (0x0200 + ((n) << 2)) +#define BM_REG_ISR 0x0e00 +#define BM_REG_IIR 0x0e0c + +/* Cache-enabled register offsets */ +#define BM_CL_CR 0x0000 +#define BM_CL_RR0 0x0100 +#define BM_CL_RR1 0x0140 +#define BM_CL_RCR 0x1000 +#define BM_CL_RCR_PI_CENA 0x3000 +#define BM_CL_RCR_CI_CENA 0x3100 + +/* BTW, the drivers (and h/w programming model) already obtain the required + * synchronisation for portal accesses via lwsync(), hwsync(), and + * data-dependencies. Use of barrier()s or other order-preserving primitives + * simply degrade performance. Hence the use of the __raw_*() interfaces, which + * simply ensure that the compiler treats the portal registers as volatile (ie. + * non-coherent). */ + +/* Cache-inhibited register access. */ +#define __bm_in(bm, o) __raw_readl((bm)->addr_ci + (o)) +#define __bm_out(bm, o, val) __raw_writel((val), (bm)->addr_ci + (o)) +#define bm_in(reg) __bm_in(&portal->addr, BM_REG_##reg) +#define bm_out(reg, val) __bm_out(&portal->addr, BM_REG_##reg, val) + +/* Cache-enabled (index) register access */ +#define __bm_cl_touch_ro(bm, o) dcbt_ro((bm)->addr_ce + (o)) +#define __bm_cl_touch_rw(bm, o) dcbt_rw((bm)->addr_ce + (o)) +#define __bm_cl_in(bm, o) __raw_readl((bm)->addr_ce + (o)) +#define __bm_cl_out(bm, o, val) \ + do { \ + u32 *__tmpclout = (bm)->addr_ce + (o); \ + __raw_writel((val), __tmpclout); \ + dcbf(__tmpclout); \ + } while (0) +#define __bm_cl_invalidate(bm, o) dcbi((bm)->addr_ce + (o)) +#define bm_cl_touch_ro(reg) __bm_cl_touch_ro(&portal->addr, BM_CL_##reg##_CENA) +#define bm_cl_touch_rw(reg) __bm_cl_touch_rw(&portal->addr, BM_CL_##reg##_CENA) +#define bm_cl_in(reg) __bm_cl_in(&portal->addr, BM_CL_##reg##_CENA) +#define bm_cl_out(reg, val) __bm_cl_out(&portal->addr, BM_CL_##reg##_CENA, val) +#define bm_cl_invalidate(reg)\ + __bm_cl_invalidate(&portal->addr, BM_CL_##reg##_CENA) + +/* Cyclic helper for rings. FIXME: once we are able to do fine-grain perf + * analysis, look at using the "extra" bit in the ring index registers to avoid + * cyclic issues. */ +static inline u8 bm_cyc_diff(u8 ringsize, u8 first, u8 last) +{ + /* 'first' is included, 'last' is excluded */ + if (first <= last) + return last - first; + return ringsize + last - first; +} + +/* Portal modes. + * Enum types; + * pmode == production mode + * cmode == consumption mode, + * Enum values use 3 letter codes. First letter matches the portal mode, + * remaining two letters indicate; + * ci == cache-inhibited portal register + * ce == cache-enabled portal register + * vb == in-band valid-bit (cache-enabled) + */ +enum bm_rcr_pmode { /* matches BCSP_CFG::RPM */ + bm_rcr_pci = 0, /* PI index, cache-inhibited */ + bm_rcr_pce = 1, /* PI index, cache-enabled */ + bm_rcr_pvb = 2 /* valid-bit */ +}; +enum bm_rcr_cmode { /* s/w-only */ + bm_rcr_cci, /* CI index, cache-inhibited */ + bm_rcr_cce /* CI index, cache-enabled */ +}; + + +/* --- Portal structures --- */ + +#define BM_RCR_SIZE 8 + +struct bm_rcr { + struct bm_rcr_entry *ring, *cursor; + u8 ci, available, ithresh, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + u32 busy; + enum bm_rcr_pmode pmode; + enum bm_rcr_cmode cmode; +#endif +}; + +struct bm_mc { + struct bm_mc_command *cr; + struct bm_mc_result *rr; + u8 rridx, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + enum { + /* Can only be _mc_start()ed */ + mc_idle, + /* Can only be _mc_commit()ed or _mc_abort()ed */ + mc_user, + /* Can only be _mc_retry()ed */ + mc_hw + } state; +#endif +}; + +struct bm_addr { + void __iomem *addr_ce; /* cache-enabled */ + void __iomem *addr_ci; /* cache-inhibited */ +}; + +struct bm_portal { + struct bm_addr addr; + struct bm_rcr rcr; + struct bm_mc mc; + struct bm_portal_config config; +} ____cacheline_aligned; + +/* --- RCR API --- */ + +/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */ +#define RCR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(BM_RCR_SIZE << 6))) + +/* Bit-wise logic to convert a ring pointer to a ring index */ +static inline u8 RCR_PTR2IDX(struct bm_rcr_entry *e) +{ + return ((uintptr_t)e >> 6) & (BM_RCR_SIZE - 1); +} + +/* Increment the 'cursor' ring pointer, taking 'vbit' into account */ +static inline void RCR_INC(struct bm_rcr *rcr) +{ + /* NB: this is odd-looking, but experiments show that it generates + * fast code with essentially no branching overheads. We increment to + * the next RCR pointer and handle overflow and 'vbit'. */ + struct bm_rcr_entry *partial = rcr->cursor + 1; + + rcr->cursor = RCR_CARRYCLEAR(partial); + if (partial != rcr->cursor) + rcr->vbit ^= BM_RCR_VERB_VBIT; +} + +static inline int bm_rcr_init(struct bm_portal *portal, enum bm_rcr_pmode pmode, + __maybe_unused enum bm_rcr_cmode cmode) +{ + /* This use of 'register', as well as all other occurrences, is because + * it has been observed to generate much faster code with gcc than is + * otherwise the case. */ + register struct bm_rcr *rcr = &portal->rcr; + u32 cfg; + u8 pi; + + rcr->ring = portal->addr.addr_ce + BM_CL_RCR; + rcr->ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1); + pi = bm_in(RCR_PI_CINH) & (BM_RCR_SIZE - 1); + rcr->cursor = rcr->ring + pi; + rcr->vbit = (bm_in(RCR_PI_CINH) & BM_RCR_SIZE) ? BM_RCR_VERB_VBIT : 0; + rcr->available = BM_RCR_SIZE - 1 + - bm_cyc_diff(BM_RCR_SIZE, rcr->ci, pi); + rcr->ithresh = bm_in(RCR_ITR); +#ifdef CONFIG_FSL_DPA_CHECKING + rcr->busy = 0; + rcr->pmode = pmode; + rcr->cmode = cmode; +#endif + cfg = (bm_in(CFG) & 0xffffffe0) | (pmode & 0x3); /* BCSP_CFG::RPM */ + bm_out(CFG, cfg); + return 0; +} + +static inline void bm_rcr_finish(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + u8 pi = bm_in(RCR_PI_CINH) & (BM_RCR_SIZE - 1); + u8 ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1); + + DPA_ASSERT(!rcr->busy); + if (pi != RCR_PTR2IDX(rcr->cursor)) + pr_crit("losing uncommited RCR entries\n"); + if (ci != rcr->ci) + pr_crit("missing existing RCR completions\n"); + if (rcr->ci != RCR_PTR2IDX(rcr->cursor)) + pr_crit("RCR destroyed unquiesced\n"); +} + +static inline struct bm_rcr_entry *bm_rcr_start(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + + DPA_ASSERT(!rcr->busy); + if (!rcr->available) + return NULL; +#ifdef CONFIG_FSL_DPA_CHECKING + rcr->busy = 1; +#endif + dcbz_64(rcr->cursor); + return rcr->cursor; +} + +static inline void bm_rcr_abort(struct bm_portal *portal) +{ + __maybe_unused register struct bm_rcr *rcr = &portal->rcr; + + DPA_ASSERT(rcr->busy); +#ifdef CONFIG_FSL_DPA_CHECKING + rcr->busy = 0; +#endif +} + +static inline struct bm_rcr_entry *bm_rcr_pend_and_next( + struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + + DPA_ASSERT(rcr->busy); + DPA_ASSERT(rcr->pmode != bm_rcr_pvb); + if (rcr->available == 1) + return NULL; + rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit; + dcbf_64(rcr->cursor); + RCR_INC(rcr); + rcr->available--; + dcbz_64(rcr->cursor); + return rcr->cursor; +} + +static inline void bm_rcr_pci_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + + DPA_ASSERT(rcr->busy); + DPA_ASSERT(rcr->pmode == bm_rcr_pci); + rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit; + RCR_INC(rcr); + rcr->available--; + hwsync(); + bm_out(RCR_PI_CINH, RCR_PTR2IDX(rcr->cursor)); +#ifdef CONFIG_FSL_DPA_CHECKING + rcr->busy = 0; +#endif +} + +static inline void bm_rcr_pce_prefetch(struct bm_portal *portal) +{ + __maybe_unused register struct bm_rcr *rcr = &portal->rcr; + + DPA_ASSERT(rcr->pmode == bm_rcr_pce); + bm_cl_invalidate(RCR_PI); + bm_cl_touch_rw(RCR_PI); +} + +static inline void bm_rcr_pce_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + + DPA_ASSERT(rcr->busy); + DPA_ASSERT(rcr->pmode == bm_rcr_pce); + rcr->cursor->__dont_write_directly__verb = myverb | rcr->vbit; + RCR_INC(rcr); + rcr->available--; + lwsync(); + bm_cl_out(RCR_PI, RCR_PTR2IDX(rcr->cursor)); +#ifdef CONFIG_FSL_DPA_CHECKING + rcr->busy = 0; +#endif +} + +static inline void bm_rcr_pvb_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_rcr *rcr = &portal->rcr; + struct bm_rcr_entry *rcursor; + + DPA_ASSERT(rcr->busy); + DPA_ASSERT(rcr->pmode == bm_rcr_pvb); + lwsync(); + rcursor = rcr->cursor; + rcursor->__dont_write_directly__verb = myverb | rcr->vbit; + dcbf_64(rcursor); + RCR_INC(rcr); + rcr->available--; +#ifdef CONFIG_FSL_DPA_CHECKING + rcr->busy = 0; +#endif +} + +static inline u8 bm_rcr_cci_update(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + u8 diff, old_ci = rcr->ci; + + DPA_ASSERT(rcr->cmode == bm_rcr_cci); + rcr->ci = bm_in(RCR_CI_CINH) & (BM_RCR_SIZE - 1); + diff = bm_cyc_diff(BM_RCR_SIZE, old_ci, rcr->ci); + rcr->available += diff; + return diff; +} + +static inline void bm_rcr_cce_prefetch(struct bm_portal *portal) +{ + __maybe_unused register struct bm_rcr *rcr = &portal->rcr; + + DPA_ASSERT(rcr->cmode == bm_rcr_cce); + bm_cl_touch_ro(RCR_CI); +} + +static inline u8 bm_rcr_cce_update(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + u8 diff, old_ci = rcr->ci; + + DPA_ASSERT(rcr->cmode == bm_rcr_cce); + rcr->ci = bm_cl_in(RCR_CI) & (BM_RCR_SIZE - 1); + bm_cl_invalidate(RCR_CI); + diff = bm_cyc_diff(BM_RCR_SIZE, old_ci, rcr->ci); + rcr->available += diff; + return diff; +} + +static inline u8 bm_rcr_get_ithresh(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + + return rcr->ithresh; +} + +static inline void bm_rcr_set_ithresh(struct bm_portal *portal, u8 ithresh) +{ + register struct bm_rcr *rcr = &portal->rcr; + + rcr->ithresh = ithresh; + bm_out(RCR_ITR, ithresh); +} + +static inline u8 bm_rcr_get_avail(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + + return rcr->available; +} + +static inline u8 bm_rcr_get_fill(struct bm_portal *portal) +{ + register struct bm_rcr *rcr = &portal->rcr; + + return BM_RCR_SIZE - 1 - rcr->available; +} + +/* --- Management command API --- */ + +static inline int bm_mc_init(struct bm_portal *portal) +{ + register struct bm_mc *mc = &portal->mc; + + mc->cr = portal->addr.addr_ce + BM_CL_CR; + mc->rr = portal->addr.addr_ce + BM_CL_RR0; + mc->rridx = (__raw_readb(&mc->cr->__dont_write_directly__verb) & + BM_MCC_VERB_VBIT) ? 0 : 1; + mc->vbit = mc->rridx ? BM_MCC_VERB_VBIT : 0; +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = mc_idle; +#endif + return 0; +} + +static inline void bm_mc_finish(struct bm_portal *portal) +{ + __maybe_unused register struct bm_mc *mc = &portal->mc; + + DPA_ASSERT(mc->state == mc_idle); +#ifdef CONFIG_FSL_DPA_CHECKING + if (mc->state != mc_idle) + pr_crit("Losing incomplete MC command\n"); +#endif +} + +static inline struct bm_mc_command *bm_mc_start(struct bm_portal *portal) +{ + register struct bm_mc *mc = &portal->mc; + + DPA_ASSERT(mc->state == mc_idle); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = mc_user; +#endif + dcbz_64(mc->cr); + return mc->cr; +} + +static inline void bm_mc_abort(struct bm_portal *portal) +{ + __maybe_unused register struct bm_mc *mc = &portal->mc; + + DPA_ASSERT(mc->state == mc_user); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = mc_idle; +#endif +} + +static inline void bm_mc_commit(struct bm_portal *portal, u8 myverb) +{ + register struct bm_mc *mc = &portal->mc; + struct bm_mc_result *rr = mc->rr + mc->rridx; + + DPA_ASSERT(mc->state == mc_user); + lwsync(); + mc->cr->__dont_write_directly__verb = myverb | mc->vbit; + dcbf(mc->cr); + dcbit_ro(rr); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = mc_hw; +#endif +} + +static inline struct bm_mc_result *bm_mc_result(struct bm_portal *portal) +{ + register struct bm_mc *mc = &portal->mc; + struct bm_mc_result *rr = mc->rr + mc->rridx; + + DPA_ASSERT(mc->state == mc_hw); + /* The inactive response register's verb byte always returns zero until + * its command is submitted and completed. This includes the valid-bit, + * in case you were wondering... */ + if (!__raw_readb(&rr->verb)) { + dcbit_ro(rr); + return NULL; + } + mc->rridx ^= 1; + mc->vbit ^= BM_MCC_VERB_VBIT; +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = mc_idle; +#endif + return rr; +} + +/* --- Portal interrupt register API --- */ + +static inline int bm_isr_init(__always_unused struct bm_portal *portal) +{ + return 0; +} + +static inline void bm_isr_finish(__always_unused struct bm_portal *portal) +{ +} + +#define SCN_REG(bpid) BM_REG_SCN((bpid) / 32) +#define SCN_BIT(bpid) (0x80000000 >> (bpid & 31)) +static inline void bm_isr_bscn_mask(struct bm_portal *portal, u8 bpid, + int enable) +{ + u32 val; + + DPA_ASSERT(bpid < bman_pool_max); + + /* REG_SCN for bpid=0..31, REG_SCN+4 for bpid=32..63 */ + val = __bm_in(&portal->addr, SCN_REG(bpid)); + if (enable) + val |= SCN_BIT(bpid); + else + val &= ~SCN_BIT(bpid); + __bm_out(&portal->addr, SCN_REG(bpid), val); +} + +static inline u32 __bm_isr_read(struct bm_portal *portal, enum bm_isr_reg n) +{ + return __bm_in(&portal->addr, BM_REG_ISR + (n << 2)); +} + +static inline void __bm_isr_write(struct bm_portal *portal, enum bm_isr_reg n, + u32 val) +{ + __bm_out(&portal->addr, BM_REG_ISR + (n << 2), val); +} + +/* Buffer Pool Cleanup */ +static inline int bm_shutdown_pool(struct bm_portal *p, u32 bpid) +{ + struct bm_mc_command *bm_cmd; + struct bm_mc_result *bm_res; + int aq_count = 0; + bool stop = false; + + while (!stop) { + /* Acquire buffers until empty */ + bm_cmd = bm_mc_start(p); + bm_cmd->acquire.bpid = bpid; + bm_mc_commit(p, BM_MCC_VERB_CMD_ACQUIRE | 1); + while (!(bm_res = bm_mc_result(p))) + cpu_relax(); + if (!(bm_res->verb & BM_MCR_VERB_ACQUIRE_BUFCOUNT)) { + /* Pool is empty */ + /* TBD : Should we do a few extra iterations in + case some other some blocks keep buffers 'on deck', + which may also be problematic */ + stop = true; + } else + ++aq_count; + } + return 0; +} diff --git a/linux/drivers/soc/fsl/qbman/bman_api.c b/linux/drivers/soc/fsl/qbman/bman_api.c new file mode 100644 index 00000000..cdfcebbd --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_api.c @@ -0,0 +1,1123 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman.h" + +/* Compilation constants */ +#define RCR_THRESH 2 /* reread h/w CI when running out of space */ +#define IRQNAME "BMan portal %d" +#define MAX_IRQNAME 16 /* big enough for "BMan portal %d" */ +#define FSL_DPA_PORTAL_SHARE 1 /* Allow portals to be shared */ + +struct bman_portal { + struct bm_portal p; + /* 2-element array. pools[0] is mask, pools[1] is snapshot. */ + struct bman_depletion *pools; + int thresh_set; + unsigned long irq_sources; + u32 slowpoll; /* only used when interrupts are off */ +#ifdef FSL_DPA_CAN_WAIT_SYNC + struct bman_pool *rcri_owned; /* only 1 release WAIT_SYNC at a time */ +#endif +#ifdef FSL_DPA_PORTAL_SHARE + raw_spinlock_t sharing_lock; /* only used if is_shared */ +#ifndef __rtems__ + int is_shared; + struct bman_portal *sharing_redirect; +#endif /* __rtems__ */ +#endif + /* When the cpu-affine portal is activated, this is non-NULL */ + const struct bm_portal_config *config; + /* 64-entry hash-table of pool objects that are tracking depletion + * entry/exit (ie. BMAN_POOL_FLAG_DEPLETION). This isn't fast-path, so + * we're not fussy about cache-misses and so forth - whereas the above + * members should all fit in one cacheline. + * BTW, with 64 entries in the hash table and 64 buffer pools to track, + * you'll never guess the hash-function ... */ + struct bman_pool *cb[64]; + char irqname[MAX_IRQNAME]; + /* Track if the portal was alloced by the driver */ + u8 alloced; +}; + + +#ifdef FSL_DPA_PORTAL_SHARE +/* For an explanation of the locking, redirection, or affine-portal logic, + * please consult the QMan driver for details. This is the same, only simpler + * (no fiddly QMan-specific bits.) */ +#ifndef __rtems__ +#define PORTAL_IRQ_LOCK(p, irqflags) \ + do { \ + if ((p)->is_shared) \ + raw_spin_lock_irqsave(&(p)->sharing_lock, irqflags); \ + else \ + local_irq_save(irqflags); \ + } while (0) +#define PORTAL_IRQ_UNLOCK(p, irqflags) \ + do { \ + if ((p)->is_shared) \ + raw_spin_unlock_irqrestore(&(p)->sharing_lock, \ + irqflags); \ + else \ + local_irq_restore(irqflags); \ + } while (0) +#else /* __rtems__ */ +#define PORTAL_IRQ_LOCK(p, irqflags) \ + raw_spin_lock_irqsave(&(p)->sharing_lock, irqflags) +#define PORTAL_IRQ_UNLOCK(p, irqflags) \ + raw_spin_unlock_irqrestore(&(p)->sharing_lock, irqflags) +#endif /* __rtems__ */ +#else +#define PORTAL_IRQ_LOCK(p, irqflags) local_irq_save(irqflags) +#define PORTAL_IRQ_UNLOCK(p, irqflags) local_irq_restore(irqflags) +#endif + +#ifndef __rtems__ +static cpumask_t affine_mask; +static DEFINE_SPINLOCK(affine_mask_lock); +#endif /* __rtems__ */ +static DEFINE_PER_CPU(struct bman_portal, bman_affine_portal); +static inline struct bman_portal *get_raw_affine_portal(void) +{ + return &get_cpu_var(bman_affine_portal); +} +#ifdef FSL_DPA_PORTAL_SHARE +static inline struct bman_portal *get_affine_portal(void) +{ + struct bman_portal *p = get_raw_affine_portal(); + +#ifndef __rtems__ + if (p->sharing_redirect) + return p->sharing_redirect; +#endif /* __rtems__ */ + return p; +} +#else +#define get_affine_portal() get_raw_affine_portal() +#endif +static inline void put_affine_portal(void) +{ + put_cpu_var(bman_affine_portal); +} +static inline struct bman_portal *get_poll_portal(void) +{ + return this_cpu_ptr(&bman_affine_portal); +} +#define put_poll_portal() + +/* GOTCHA: this object type refers to a pool, it isn't *the* pool. There may be + * more than one such object per BMan buffer pool, eg. if different users of the + * pool are operating via different portals. */ +struct bman_pool { + struct bman_pool_params params; + /* Used for hash-table admin when using depletion notifications. */ + struct bman_portal *portal; + struct bman_pool *next; + /* stockpile state - NULL unless BMAN_POOL_FLAG_STOCKPILE is set */ + struct bm_buffer *sp; + unsigned int sp_fill; +#ifdef CONFIG_FSL_DPA_CHECKING + atomic_t in_use; +#endif +}; + +/* (De)Registration of depletion notification callbacks */ +static void depletion_link(struct bman_portal *portal, struct bman_pool *pool) +{ + __maybe_unused unsigned long irqflags; + + pool->portal = portal; + PORTAL_IRQ_LOCK(portal, irqflags); + pool->next = portal->cb[pool->params.bpid]; + portal->cb[pool->params.bpid] = pool; + if (!pool->next) + /* First object for that bpid on this portal, enable the BSCN + * mask bit. */ + bm_isr_bscn_mask(&portal->p, pool->params.bpid, 1); + PORTAL_IRQ_UNLOCK(portal, irqflags); +} +static void depletion_unlink(struct bman_pool *pool) +{ + struct bman_pool *it, *last = NULL; + struct bman_pool **base = &pool->portal->cb[pool->params.bpid]; + __maybe_unused unsigned long irqflags; + + PORTAL_IRQ_LOCK(pool->portal, irqflags); + it = *base; /* <-- gotcha, don't do this prior to the irq_save */ + while (it != pool) { + last = it; + it = it->next; + } + if (!last) + *base = pool->next; + else + last->next = pool->next; + if (!last && !pool->next) { + /* Last object for that bpid on this portal, disable the BSCN + * mask bit. */ + bm_isr_bscn_mask(&pool->portal->p, pool->params.bpid, 0); + /* And "forget" that we last saw this pool as depleted */ + bman_depletion_unset(&pool->portal->pools[1], + pool->params.bpid); + } + PORTAL_IRQ_UNLOCK(pool->portal, irqflags); +} + +/* In the case that the application's core loop calls qman_poll() and + * bman_poll(), we ought to balance how often we incur the overheads of the + * slow-path poll. We'll use two decrementer sources. The idle decrementer + * constant is used when the last slow-poll detected no work to do, and the busy + * decrementer constant when the last slow-poll had work to do. */ +#define SLOW_POLL_IDLE 1000 +#define SLOW_POLL_BUSY 10 +static u32 __poll_portal_slow(struct bman_portal *p, u32 is); + +/* Portal interrupt handler */ +static irqreturn_t portal_isr(__always_unused int irq, void *ptr) +{ + struct bman_portal *p = ptr; + u32 clear = p->irq_sources; + u32 is = bm_isr_status_read(&p->p) & p->irq_sources; + + clear |= __poll_portal_slow(p, is); + bm_isr_status_clear(&p->p, clear); + return IRQ_HANDLED; +} + + +struct bman_portal *bman_create_portal( + struct bman_portal *portal, + const struct bm_portal_config *config) +{ + struct bm_portal *__p; + const struct bman_depletion *pools = &config->public_cfg.mask; + int ret; + u8 bpid = 0; + + if (!portal) { + portal = kmalloc(sizeof(*portal), GFP_KERNEL); + if (!portal) + return portal; + portal->alloced = 1; + } else + portal->alloced = 0; + + __p = &portal->p; + + /* prep the low-level portal struct with the mapped addresses from the + * config, everything that follows depends on it and "config" is more + * for (de)reference... */ + __p->addr.addr_ce = config->addr_virt[DPA_PORTAL_CE]; + __p->addr.addr_ci = config->addr_virt[DPA_PORTAL_CI]; + if (bm_rcr_init(__p, bm_rcr_pvb, bm_rcr_cce)) { + pr_err("RCR initialisation failed\n"); + goto fail_rcr; + } + if (bm_mc_init(__p)) { + pr_err("MC initialisation failed\n"); + goto fail_mc; + } + if (bm_isr_init(__p)) { + pr_err("ISR initialisation failed\n"); + goto fail_isr; + } + portal->pools = kmalloc(2 * sizeof(*pools), GFP_KERNEL); + if (!portal->pools) + goto fail_pools; + portal->pools[0] = *pools; + bman_depletion_init(portal->pools + 1); + while (bpid < bman_pool_max) { + /* Default to all BPIDs disabled, we enable as required at + * run-time. */ + bm_isr_bscn_mask(__p, bpid, 0); + bpid++; + } + portal->slowpoll = 0; +#ifdef FSL_DPA_CAN_WAIT_SYNC + portal->rcri_owned = NULL; +#endif +#ifdef FSL_DPA_PORTAL_SHARE + raw_spin_lock_init(&portal->sharing_lock); +#ifndef __rtems__ + portal->is_shared = config->public_cfg.is_shared; + portal->sharing_redirect = NULL; +#endif /* __rtems__ */ +#endif + memset(&portal->cb, 0, sizeof(portal->cb)); + /* Write-to-clear any stale interrupt status bits */ + bm_isr_disable_write(__p, 0xffffffff); + portal->irq_sources = 0; + bm_isr_enable_write(__p, portal->irq_sources); + bm_isr_status_clear(__p, 0xffffffff); + snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, config->public_cfg.cpu); + if (request_irq(config->public_cfg.irq, portal_isr, 0, portal->irqname, + portal)) { + pr_err("request_irq() failed\n"); + goto fail_irq; + } +#ifndef __rtems__ + if ((config->public_cfg.cpu != -1) && + irq_can_set_affinity(config->public_cfg.irq) && + irq_set_affinity(config->public_cfg.irq, + cpumask_of(config->public_cfg.cpu))) { + pr_err("irq_set_affinity() failed\n"); + goto fail_affinity; + } +#endif /* __rtems__ */ + + /* Need RCR to be empty before continuing */ + ret = bm_rcr_get_fill(__p); + if (ret) { + pr_err("RCR unclean\n"); + goto fail_rcr_empty; + } + /* Success */ + portal->config = config; + + bm_isr_disable_write(__p, 0); + bm_isr_uninhibit(__p); + return portal; +fail_rcr_empty: +#ifndef __rtems__ +fail_affinity: +#endif /* __rtems__ */ + free_irq(config->public_cfg.irq, portal); +fail_irq: + kfree(portal->pools); +fail_pools: + bm_isr_finish(__p); +fail_isr: + bm_mc_finish(__p); +fail_mc: + bm_rcr_finish(__p); +fail_rcr: + if (portal->alloced) + kfree(portal); + return NULL; +} + +struct bman_portal *bman_create_affine_portal( + const struct bm_portal_config *config) +{ + struct bman_portal *portal; + + portal = &per_cpu(bman_affine_portal, config->public_cfg.cpu); + portal = bman_create_portal(portal, config); +#ifndef __rtems__ + if (portal) { + spin_lock(&affine_mask_lock); + cpumask_set_cpu(config->public_cfg.cpu, &affine_mask); + spin_unlock(&affine_mask_lock); + } +#endif /* __rtems__ */ + return portal; +} + + +#ifndef __rtems__ +struct bman_portal *bman_create_affine_slave(struct bman_portal *redirect, + int cpu) +{ +#ifdef FSL_DPA_PORTAL_SHARE + struct bman_portal *p = &per_cpu(bman_affine_portal, cpu); + + BUG_ON(p->config); + BUG_ON(p->is_shared); + BUG_ON(!redirect->config->public_cfg.is_shared); + p->irq_sources = 0; + p->sharing_redirect = redirect; + put_affine_portal(); + return p; +#else + BUG(); + return NULL; +#endif +} +#endif /* __rtems__ */ + +void bman_destroy_portal(struct bman_portal *bm) +{ + const struct bm_portal_config *pcfg = bm->config; + + bm_rcr_cce_update(&bm->p); + bm_rcr_cce_update(&bm->p); + + free_irq(pcfg->public_cfg.irq, bm); + + kfree(bm->pools); + bm_isr_finish(&bm->p); + bm_mc_finish(&bm->p); + bm_rcr_finish(&bm->p); + bm->config = NULL; + if (bm->alloced) + kfree(bm); +} + +const struct bm_portal_config *bman_destroy_affine_portal(void) +{ + struct bman_portal *bm = get_raw_affine_portal(); + const struct bm_portal_config *pcfg; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (bm->sharing_redirect) { + bm->sharing_redirect = NULL; + put_affine_portal(); + return NULL; + } + bm->is_shared = 0; +#endif /* __rtems__ */ +#endif + pcfg = bm->config; + bman_destroy_portal(bm); +#ifndef __rtems__ + spin_lock(&affine_mask_lock); + cpumask_clear_cpu(pcfg->public_cfg.cpu, &affine_mask); + spin_unlock(&affine_mask_lock); +#endif /* __rtems__ */ + put_affine_portal(); + return pcfg; +} + +/* When release logic waits on available RCR space, we need a global waitqueue + * in the case of "affine" use (as the waits wake on different cpus which means + * different portals - so we can't wait on any per-portal waitqueue). */ +static DECLARE_WAIT_QUEUE_HEAD(affine_queue); + +static u32 __poll_portal_slow(struct bman_portal *p, u32 is) +{ + struct bman_depletion tmp; + u32 ret = is; + + /* There is a gotcha to be aware of. If we do the query before clearing + * the status register, we may miss state changes that occur between the + * two. If we write to clear the status register before the query, the + * cache-enabled query command may overtake the status register write + * unless we use a heavyweight sync (which we don't want). Instead, we + * write-to-clear the status register then *read it back* before doing + * the query, hence the odd while loop with the 'is' accumulation. */ + if (is & BM_PIRQ_BSCN) { + struct bm_mc_result *mcr; + __maybe_unused unsigned long irqflags; + unsigned int i, j; + u32 __is; + + bm_isr_status_clear(&p->p, BM_PIRQ_BSCN); + while ((__is = bm_isr_status_read(&p->p)) & BM_PIRQ_BSCN) { + is |= __is; + bm_isr_status_clear(&p->p, BM_PIRQ_BSCN); + } + is &= ~BM_PIRQ_BSCN; + PORTAL_IRQ_LOCK(p, irqflags); + bm_mc_start(&p->p); + bm_mc_commit(&p->p, BM_MCC_VERB_CMD_QUERY); + while (!(mcr = bm_mc_result(&p->p))) + cpu_relax(); + tmp = mcr->query.ds.state; + PORTAL_IRQ_UNLOCK(p, irqflags); + for (i = 0; i < 2; i++) { + int idx = i * 32; + /* tmp is a mask of currently-depleted pools. + * pools[0] is mask of those we care about. + * pools[1] is our previous view (we only want to + * be told about changes). */ + tmp.__state[i] &= p->pools[0].__state[i]; + if (tmp.__state[i] == p->pools[1].__state[i]) + /* fast-path, nothing to see, move along */ + continue; + for (j = 0; j <= 31; j++, idx++) { + struct bman_pool *pool = p->cb[idx]; + int b4 = bman_depletion_get(&p->pools[1], idx); + int af = bman_depletion_get(&tmp, idx); + + if (b4 == af) + continue; + while (pool) { + pool->params.cb(p, pool, + pool->params.cb_ctx, af); + pool = pool->next; + } + } + } + p->pools[1] = tmp; + } + + if (is & BM_PIRQ_RCRI) { + __maybe_unused unsigned long irqflags; + + PORTAL_IRQ_LOCK(p, irqflags); + bm_rcr_cce_update(&p->p); +#ifdef FSL_DPA_CAN_WAIT_SYNC + /* If waiting for sync, we only cancel the interrupt threshold + * when the ring utilisation hits zero. */ + if (p->rcri_owned) { + if (!bm_rcr_get_fill(&p->p)) { + p->rcri_owned = NULL; + bm_rcr_set_ithresh(&p->p, 0); + } + } else +#endif + bm_rcr_set_ithresh(&p->p, 0); + PORTAL_IRQ_UNLOCK(p, irqflags); + wake_up(&affine_queue); + bm_isr_status_clear(&p->p, BM_PIRQ_RCRI); + is &= ~BM_PIRQ_RCRI; + } + + /* There should be no status register bits left undefined */ + DPA_ASSERT(!is); + return ret; +} + +const struct bman_portal_config *bman_get_portal_config(void) +{ + struct bman_portal *p = get_affine_portal(); + const struct bman_portal_config *ret = &p->config->public_cfg; + + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(bman_get_portal_config); + +u32 bman_irqsource_get(void) +{ + struct bman_portal *p = get_raw_affine_portal(); + u32 ret = p->irq_sources & BM_PIRQ_VISIBLE; + + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(bman_irqsource_get); + +int bman_p_irqsource_add(struct bman_portal *p, __maybe_unused u32 bits) +{ + __maybe_unused unsigned long irqflags; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (p->sharing_redirect) + return -EINVAL; +#endif /* __rtems__ */ +#endif + PORTAL_IRQ_LOCK(p, irqflags); + set_bits(bits & BM_PIRQ_VISIBLE, &p->irq_sources); + bm_isr_enable_write(&p->p, p->irq_sources); + PORTAL_IRQ_UNLOCK(p, irqflags); + return 0; +} +EXPORT_SYMBOL(bman_p_irqsource_add); + +int bman_irqsource_add(__maybe_unused u32 bits) +{ + struct bman_portal *p = get_raw_affine_portal(); + int ret = bman_p_irqsource_add(p, bits); + + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(bman_irqsource_add); + +int bman_irqsource_remove(u32 bits) +{ + struct bman_portal *p = get_raw_affine_portal(); + __maybe_unused unsigned long irqflags; + u32 ier; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (p->sharing_redirect) { + put_affine_portal(); + return -EINVAL; + } +#endif /* __rtems__ */ +#endif + /* Our interrupt handler only processes+clears status register bits that + * are in p->irq_sources. As we're trimming that mask, if one of them + * were to assert in the status register just before we remove it from + * the enable register, there would be an interrupt-storm when we + * release the IRQ lock. So we wait for the enable register update to + * take effect in h/w (by reading it back) and then clear all other bits + * in the status register. Ie. we clear them from ISR once it's certain + * IER won't allow them to reassert. */ + PORTAL_IRQ_LOCK(p, irqflags); + bits &= BM_PIRQ_VISIBLE; + clear_bits(bits, &p->irq_sources); + bm_isr_enable_write(&p->p, p->irq_sources); + ier = bm_isr_enable_read(&p->p); + /* Using "~ier" (rather than "bits" or "~p->irq_sources") creates a + * data-dependency, ie. to protect against re-ordering. */ + bm_isr_status_clear(&p->p, ~ier); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return 0; +} +EXPORT_SYMBOL(bman_irqsource_remove); + +#ifndef __rtems__ +const cpumask_t *bman_affine_cpus(void) +{ + return &affine_mask; +} +EXPORT_SYMBOL(bman_affine_cpus); +#endif /* __rtems__ */ + +u32 bman_poll_slow(void) +{ + struct bman_portal *p = get_poll_portal(); + u32 ret; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (unlikely(p->sharing_redirect)) + ret = (u32)-1; + else +#endif /* __rtems__ */ +#endif + { + u32 is = bm_isr_status_read(&p->p) & ~p->irq_sources; + + ret = __poll_portal_slow(p, is); + bm_isr_status_clear(&p->p, ret); + } + put_poll_portal(); + return ret; +} +EXPORT_SYMBOL(bman_poll_slow); + +/* Legacy wrapper */ +void bman_poll(void) +{ + struct bman_portal *p = get_poll_portal(); + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (unlikely(p->sharing_redirect)) + goto done; +#endif /* __rtems__ */ +#endif + if (!(p->slowpoll--)) { + u32 is = bm_isr_status_read(&p->p) & ~p->irq_sources; + u32 active = __poll_portal_slow(p, is); + + if (active) + p->slowpoll = SLOW_POLL_BUSY; + else + p->slowpoll = SLOW_POLL_IDLE; + } +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ +done: +#endif /* __rtems__ */ +#endif + put_poll_portal(); +} +EXPORT_SYMBOL(bman_poll); + +static const u32 zero_thresholds[4] = {0, 0, 0, 0}; + +struct bman_pool *bman_new_pool(const struct bman_pool_params *params) +{ + struct bman_pool *pool = NULL; + u32 bpid; + + if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID) { + if (bman_alloc_bpid(&bpid)) + return NULL; + } else { + if (params->bpid >= bman_pool_max) + return NULL; + bpid = params->bpid; + } +#ifdef CONFIG_FSL_BMAN + if (params->flags & BMAN_POOL_FLAG_THRESH) { + if (bm_pool_set(bpid, params->thresholds)) + goto err; + } +#else + if (params->flags & BMAN_POOL_FLAG_THRESH) + goto err; +#endif + pool = kmalloc(sizeof(*pool), GFP_KERNEL); + if (!pool) + goto err; + pool->sp = NULL; + pool->sp_fill = 0; + pool->params = *params; +#ifdef CONFIG_FSL_DPA_CHECKING + atomic_set(&pool->in_use, 1); +#endif + if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID) + pool->params.bpid = bpid; + if (params->flags & BMAN_POOL_FLAG_STOCKPILE) { + pool->sp = kmalloc(sizeof(struct bm_buffer) * BMAN_STOCKPILE_SZ, + GFP_KERNEL); + if (!pool->sp) + goto err; + } + if (pool->params.flags & BMAN_POOL_FLAG_DEPLETION) { + struct bman_portal *p = get_affine_portal(); + + if (!p->pools || !bman_depletion_get(&p->pools[0], bpid)) { + pr_err("Depletion events disabled for bpid %d\n", bpid); + goto err; + } + depletion_link(p, pool); + put_affine_portal(); + } + return pool; +err: +#ifdef CONFIG_FSL_BMAN + if (params->flags & BMAN_POOL_FLAG_THRESH) + bm_pool_set(bpid, zero_thresholds); +#endif + if (params->flags & BMAN_POOL_FLAG_DYNAMIC_BPID) + bman_release_bpid(bpid); + if (pool) { + kfree(pool->sp); + kfree(pool); + } + return NULL; +} +EXPORT_SYMBOL(bman_new_pool); + +void bman_free_pool(struct bman_pool *pool) +{ +#ifdef CONFIG_FSL_BMAN + if (pool->params.flags & BMAN_POOL_FLAG_THRESH) + bm_pool_set(pool->params.bpid, zero_thresholds); +#endif + if (pool->params.flags & BMAN_POOL_FLAG_DEPLETION) + depletion_unlink(pool); + if (pool->params.flags & BMAN_POOL_FLAG_STOCKPILE) { + if (pool->sp_fill) + pr_err("Stockpile not flushed, has %u in bpid %u.\n", + pool->sp_fill, pool->params.bpid); + kfree(pool->sp); + pool->sp = NULL; + pool->params.flags ^= BMAN_POOL_FLAG_STOCKPILE; + } + if (pool->params.flags & BMAN_POOL_FLAG_DYNAMIC_BPID) + bman_release_bpid(pool->params.bpid); + kfree(pool); +} +EXPORT_SYMBOL(bman_free_pool); + +const struct bman_pool_params *bman_get_params(const struct bman_pool *pool) +{ + return &pool->params; +} +EXPORT_SYMBOL(bman_get_params); + +static noinline void update_rcr_ci(struct bman_portal *p, u8 avail) +{ + if (avail) + bm_rcr_cce_prefetch(&p->p); + else + bm_rcr_cce_update(&p->p); +} + +int bman_rcr_is_empty(void) +{ + __maybe_unused unsigned long irqflags; + struct bman_portal *p = get_affine_portal(); + u8 avail; + + PORTAL_IRQ_LOCK(p, irqflags); + update_rcr_ci(p, 0); + avail = bm_rcr_get_fill(&p->p); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return avail == 0; +} +EXPORT_SYMBOL(bman_rcr_is_empty); + +static inline struct bm_rcr_entry *try_rel_start(struct bman_portal **p, +#ifdef FSL_DPA_CAN_WAIT + __maybe_unused struct bman_pool *pool, +#endif + __maybe_unused unsigned long *irqflags, + __maybe_unused u32 flags) +{ + struct bm_rcr_entry *r; + u8 avail; + + *p = get_affine_portal(); + PORTAL_IRQ_LOCK(*p, (*irqflags)); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) && + (flags & BMAN_RELEASE_FLAG_WAIT_SYNC))) { + if ((*p)->rcri_owned) { + PORTAL_IRQ_UNLOCK(*p, (*irqflags)); + put_affine_portal(); + return NULL; + } + (*p)->rcri_owned = pool; + } +#endif + avail = bm_rcr_get_avail(&(*p)->p); + if (avail < 2) + update_rcr_ci(*p, avail); + r = bm_rcr_start(&(*p)->p); + if (unlikely(!r)) { +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) && + (flags & BMAN_RELEASE_FLAG_WAIT_SYNC))) + (*p)->rcri_owned = NULL; +#endif + PORTAL_IRQ_UNLOCK(*p, (*irqflags)); + put_affine_portal(); + } + return r; +} + +#ifdef FSL_DPA_CAN_WAIT +static noinline struct bm_rcr_entry *__wait_rel_start(struct bman_portal **p, + struct bman_pool *pool, + __maybe_unused unsigned long *irqflags, + u32 flags) +{ + struct bm_rcr_entry *rcr = try_rel_start(p, pool, irqflags, flags); + + if (!rcr) + bm_rcr_set_ithresh(&(*p)->p, 1); + return rcr; +} + +static noinline struct bm_rcr_entry *wait_rel_start(struct bman_portal **p, + struct bman_pool *pool, + __maybe_unused unsigned long *irqflags, + u32 flags) +{ + struct bm_rcr_entry *rcr; +#ifndef FSL_DPA_CAN_WAIT_SYNC + pool = NULL; +#endif +#ifndef __rtems__ + if (flags & BMAN_RELEASE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (rcr = __wait_rel_start(p, pool, irqflags, flags))); + else +#endif /* __rtems__ */ + wait_event(affine_queue, + (rcr = __wait_rel_start(p, pool, irqflags, flags))); + return rcr; +} +#endif + +/* to facilitate better copying of bufs into the ring without either (a) copying + * noise into the first byte (prematurely triggering the command), nor (b) being + * very inefficient by copying small fields using read-modify-write */ +struct overlay_bm_buffer { + u32 first; + u32 second; +}; + +static inline int __bman_release(struct bman_pool *pool, + const struct bm_buffer *bufs, u8 num, u32 flags) +{ + struct bman_portal *p; + struct bm_rcr_entry *r; + struct overlay_bm_buffer *o_dest; + struct overlay_bm_buffer *o_src = (struct overlay_bm_buffer *)&bufs[0]; + __maybe_unused unsigned long irqflags; + u32 i = num - 1; + +#ifdef FSL_DPA_CAN_WAIT + if (flags & BMAN_RELEASE_FLAG_WAIT) + r = wait_rel_start(&p, pool, &irqflags, flags); + else + r = try_rel_start(&p, pool, &irqflags, flags); +#else + r = try_rel_start(&p, &irqflags, flags); +#endif + if (!r) + return -EBUSY; + /* We can copy all but the first entry, as this can trigger badness + * with the valid-bit. Use the overlay to mask the verb byte. */ + o_dest = (struct overlay_bm_buffer *)&r->bufs[0]; + o_dest->first = (o_src->first & 0x0000ffff) | + (((u32)pool->params.bpid << 16) & 0x00ff0000); + o_dest->second = o_src->second; + if (i) + copy_words(&r->bufs[1], &bufs[1], i * sizeof(bufs[0])); + bm_rcr_pvb_commit(&p->p, BM_RCR_VERB_CMD_BPID_SINGLE | + (num & BM_RCR_VERB_BUFCOUNT_MASK)); +#ifdef FSL_DPA_CAN_WAIT_SYNC + /* if we wish to sync we need to set the threshold after h/w sees the + * new ring entry. As we're mixing cache-enabled and cache-inhibited + * accesses, this requires a heavy-weight sync. */ + if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) && + (flags & BMAN_RELEASE_FLAG_WAIT_SYNC))) { + hwsync(); + bm_rcr_set_ithresh(&p->p, 1); + } +#endif + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & BMAN_RELEASE_FLAG_WAIT) && + (flags & BMAN_RELEASE_FLAG_WAIT_SYNC))) { +#ifndef __rtems__ + if (flags & BMAN_RELEASE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->rcri_owned != pool)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, (p->rcri_owned != pool)); + } +#endif + return 0; +} + +int bman_release(struct bman_pool *pool, const struct bm_buffer *bufs, u8 num, + u32 flags) +{ + int ret = 0; + +#ifdef CONFIG_FSL_DPA_CHECKING + if (!num || (num > 8)) + return -EINVAL; + if (pool->params.flags & BMAN_POOL_FLAG_NO_RELEASE) + return -EINVAL; +#endif + /* Without stockpile, this API is a pass-through to the h/w operation */ + if (!(pool->params.flags & BMAN_POOL_FLAG_STOCKPILE)) + return __bman_release(pool, bufs, num, flags); +#ifdef CONFIG_FSL_DPA_CHECKING + if (!atomic_dec_and_test(&pool->in_use)) { + pr_crit("Parallel attempts to enter bman_released() detected."); + panic("only one instance of bman_released/acquired allowed"); + } +#endif + /* This needs some explanation. Adding the given buffers may take the + * stockpile over the threshold, but in fact the stockpile may already + * *be* over the threshold if a previous release-to-hw attempt had + * failed. So we have 3 cases to cover; + * 1. we add to the stockpile and don't hit the threshold, + * 2. we add to the stockpile, hit the threshold and release-to-hw, + * 3. we have to release-to-hw before adding to the stockpile + * (not enough room in the stockpile for case 2). + * Our constraints on thresholds guarantee that in case 3, there must be + * at least 8 bufs already in the stockpile, so all release-to-hw ops + * are for 8 bufs. Despite all this, the API must indicate whether the + * given buffers were taken off the caller's hands, irrespective of + * whether a release-to-hw was attempted. */ + while (num) { + /* Add buffers to stockpile if they fit */ + if ((pool->sp_fill + num) < BMAN_STOCKPILE_SZ) { + copy_words(pool->sp + pool->sp_fill, bufs, + sizeof(struct bm_buffer) * num); + pool->sp_fill += num; + num = 0; /* --> will return success no matter what */ + } + /* Do hw op if hitting the high-water threshold */ + if ((pool->sp_fill + num) >= BMAN_STOCKPILE_HIGH) { + ret = __bman_release(pool, + pool->sp + (pool->sp_fill - 8), 8, flags); + if (ret) { + ret = (num ? ret : 0); + goto release_done; + } + pool->sp_fill -= 8; + } + } +release_done: +#ifdef CONFIG_FSL_DPA_CHECKING + atomic_inc(&pool->in_use); +#endif + return ret; +} +EXPORT_SYMBOL(bman_release); + +static inline int __bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs, + u8 num) +{ + struct bman_portal *p = get_affine_portal(); + struct bm_mc_command *mcc; + struct bm_mc_result *mcr; + __maybe_unused unsigned long irqflags; + int ret; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = bm_mc_start(&p->p); + mcc->acquire.bpid = pool->params.bpid; + bm_mc_commit(&p->p, BM_MCC_VERB_CMD_ACQUIRE | + (num & BM_MCC_VERB_ACQUIRE_BUFCOUNT)); + while (!(mcr = bm_mc_result(&p->p))) + cpu_relax(); + ret = mcr->verb & BM_MCR_VERB_ACQUIRE_BUFCOUNT; + if (bufs) + copy_words(&bufs[0], &mcr->acquire.bufs[0], + num * sizeof(bufs[0])); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (ret != num) + ret = -ENOMEM; + return ret; +} + +int bman_acquire(struct bman_pool *pool, struct bm_buffer *bufs, u8 num, + u32 flags) +{ + int ret = 0; + +#ifdef CONFIG_FSL_DPA_CHECKING + if (!num || (num > 8)) + return -EINVAL; + if (pool->params.flags & BMAN_POOL_FLAG_ONLY_RELEASE) + return -EINVAL; +#endif + /* Without stockpile, this API is a pass-through to the h/w operation */ + if (!(pool->params.flags & BMAN_POOL_FLAG_STOCKPILE)) + return __bman_acquire(pool, bufs, num); +#ifdef CONFIG_FSL_DPA_CHECKING + if (!atomic_dec_and_test(&pool->in_use)) { + pr_crit("Parallel attempts to enter bman_acquire() detected."); + panic("only one instance of bman_released/acquired allowed"); + } +#endif + /* Only need a h/w op if we'll hit the low-water thresh */ + if (!(flags & BMAN_ACQUIRE_FLAG_STOCKPILE) && + (pool->sp_fill <= (BMAN_STOCKPILE_LOW + num))) { + /* refill stockpile with max amount, but if max amount + * isn't available, try amount the user wants */ + int bufcount = 8; + + ret = __bman_acquire(pool, pool->sp + pool->sp_fill, bufcount); + if (ret < 0 && bufcount != num) { + bufcount = num; + /* Maybe buffer pool has less than 8 */ + ret = __bman_acquire(pool, pool->sp + pool->sp_fill, + bufcount); + } + if (ret < 0) + goto hw_starved; + DPA_ASSERT(ret == bufcount); + pool->sp_fill += bufcount; + } else { +hw_starved: + if (pool->sp_fill < num) { + ret = -ENOMEM; + goto acquire_done; + } + } + copy_words(bufs, pool->sp + (pool->sp_fill - num), + sizeof(struct bm_buffer) * num); + pool->sp_fill -= num; + ret = num; +acquire_done: +#ifdef CONFIG_FSL_DPA_CHECKING + atomic_inc(&pool->in_use); +#endif + return ret; +} +EXPORT_SYMBOL(bman_acquire); + +int bman_flush_stockpile(struct bman_pool *pool, u32 flags) +{ + u8 num; + int ret; + + while (pool->sp_fill) { + num = ((pool->sp_fill > 8) ? 8 : pool->sp_fill); + ret = __bman_release(pool, pool->sp + (pool->sp_fill - num), + num, flags); + if (ret) + return ret; + pool->sp_fill -= num; + } + return 0; +} +EXPORT_SYMBOL(bman_flush_stockpile); + +int bman_query_pools(struct bm_pool_state *state) +{ + struct bman_portal *p = get_affine_portal(); + struct bm_mc_result *mcr; + __maybe_unused unsigned long irqflags; + + PORTAL_IRQ_LOCK(p, irqflags); + bm_mc_start(&p->p); + bm_mc_commit(&p->p, BM_MCC_VERB_CMD_QUERY); + while (!(mcr = bm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & BM_MCR_VERB_CMD_MASK) == BM_MCR_VERB_CMD_QUERY); + *state = mcr->query; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return 0; +} +EXPORT_SYMBOL(bman_query_pools); + +#ifdef CONFIG_FSL_BMAN +u32 bman_query_free_buffers(struct bman_pool *pool) +{ + return bm_pool_free_buffers(pool->params.bpid); +} +EXPORT_SYMBOL(bman_query_free_buffers); + +int bman_update_pool_thresholds(struct bman_pool *pool, const u32 *thresholds) +{ + u32 bpid; + + bpid = bman_get_params(pool)->bpid; + + return bm_pool_set(bpid, thresholds); +} +EXPORT_SYMBOL(bman_update_pool_thresholds); +#endif + +int bman_shutdown_pool(u32 bpid) +{ + struct bman_portal *p = get_affine_portal(); + __maybe_unused unsigned long irqflags; + int ret; + + PORTAL_IRQ_LOCK(p, irqflags); + ret = bm_shutdown_pool(&p->p, bpid); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(bman_shutdown_pool); + +const struct bm_portal_config * +bman_get_bm_portal_config(const struct bman_portal *portal) +{ +#ifndef __rtems__ + return portal->sharing_redirect ? NULL : portal->config; +#else /* __rtems__ */ + return portal->config; +#endif /* __rtems__ */ +} diff --git a/linux/drivers/soc/fsl/qbman/bman_portal.c b/linux/drivers/soc/fsl/qbman/bman_portal.c new file mode 100644 index 00000000..f9fd022c --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_portal.c @@ -0,0 +1,399 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_priv.h" + +/* + * Global variables of the max portal/pool number this BMan version supported + */ +u16 bman_ip_rev; +EXPORT_SYMBOL(bman_ip_rev); + +u16 bman_pool_max; +EXPORT_SYMBOL(bman_pool_max); + +/* After initialising cpus that own shared portal configs, we cache the + * resulting portals (ie. not just the configs) in this array. Then we + * initialise slave cpus that don't have their own portals, redirecting them to + * portals from this cache in a round-robin assignment. */ +static struct bman_portal *shared_portals[NR_CPUS] __initdata; +static int num_shared_portals __initdata; +static int shared_portals_idx __initdata; + +static LIST_HEAD(unused_pcfgs); +static void *affine_bportals[NR_CPUS]; + +#ifndef __rtems__ +static const int flags[] = {0, _PAGE_GUARDED | _PAGE_NO_CACHE}; +#else /* __rtems__ */ +static const int flags[] = {0, 0}; +#endif /* __rtems__ */ + +static struct bm_portal_config * __init get_pcfg(struct list_head *list) +{ + struct bm_portal_config *pcfg; + + if (list_empty(list)) + return NULL; + pcfg = list_entry(list->prev, struct bm_portal_config, list); + list_del(&pcfg->list); + + return pcfg; +} + +static struct bman_portal * __init init_pcfg(struct bm_portal_config *pcfg) +{ + struct bman_portal *p = bman_create_affine_portal(pcfg); + + if (p) { +#ifdef CONFIG_FSL_DPA_PIRQ_SLOW + bman_p_irqsource_add(p, BM_PIRQ_RCRI | BM_PIRQ_BSCN); +#endif + pr_info("Portal %sinitialised, cpu %d\n", + pcfg->public_cfg.is_shared ? "(shared) " : "", + pcfg->public_cfg.cpu); + affine_bportals[pcfg->public_cfg.cpu] = p; + } else + pr_crit("Portal failure on cpu %d\n", pcfg->public_cfg.cpu); + + return p; +} + +static void __init init_slave(int cpu) +{ + struct bman_portal *p; + + p = bman_create_affine_slave(shared_portals[shared_portals_idx++], cpu); + if (!p) + pr_err("Slave portal failure on cpu %d\n", cpu); + else + pr_info("Portal %s initialised, cpu %d\n", "(slave) ", cpu); + if (shared_portals_idx >= num_shared_portals) + shared_portals_idx = 0; + affine_bportals[cpu] = p; +} + +/* Bootarg "bportals=[...]" has the same syntax as "qportals=", and so the + * parsing is in dpaa_sys.h. The syntax is a comma-separated list of indexes + * and/or ranges of indexes, with each being optionally prefixed by "s" to + * explicitly mark it or them for sharing. + * Eg; + * bportals=s0,1-3,s4 + * means that cpus 1,2,3 get "unshared" portals, cpus 0 and 4 get "shared" + * portals, and any remaining cpus share the portals that are assigned to cpus 0 + * or 4, selected in a round-robin fashion. (In this example, cpu 5 would share + * cpu 0's portal, cpu 6 would share cpu4's portal, and cpu 7 would share cpu + * 0's portal.) */ +static struct cpumask want_unshared __initdata; /* cpus requested without "s" */ +static struct cpumask want_shared __initdata; /* cpus requested with "s" */ + +static int __init parse_bportals(char *str) +{ + return parse_portals_bootarg(str, &want_shared, &want_unshared, + "bportals"); +} +__setup("bportals=", parse_bportals); + +static void __cold bman_offline_cpu(unsigned int cpu) +{ + struct bman_portal *p = (struct bman_portal *)affine_bportals[cpu]; + const struct bm_portal_config *pcfg; + + if (p) { + pcfg = bman_get_bm_portal_config(p); + if (pcfg) + irq_set_affinity(pcfg->public_cfg.irq, cpumask_of(0)); + } +} + +#ifdef CONFIG_HOTPLUG_CPU +static void __cold bman_online_cpu(unsigned int cpu) +{ + struct bman_portal *p = (struct bman_portal *)affine_bportals[cpu]; + const struct bm_portal_config *pcfg; + + if (p) { + pcfg = bman_get_bm_portal_config(p); + if (pcfg) + irq_set_affinity(pcfg->public_cfg.irq, cpumask_of(cpu)); + } +} + +static int __cold bman_hotplug_cpu_callback(struct notifier_block *nfb, + unsigned long action, void *hcpu) +{ + unsigned int cpu = (unsigned long)hcpu; + + switch (action) { + case CPU_ONLINE: + case CPU_ONLINE_FROZEN: + bman_online_cpu(cpu); + break; + case CPU_DOWN_PREPARE: + case CPU_DOWN_PREPARE_FROZEN: + bman_offline_cpu(cpu); + } + + return NOTIFY_OK; +} + +static struct notifier_block bman_hotplug_cpu_notifier = { + .notifier_call = bman_hotplug_cpu_callback, +}; +#endif /* CONFIG_HOTPLUG_CPU */ + +static int __cold bman_portal_probe(struct platform_device *of_dev) +{ + struct device *dev = &of_dev->dev; + struct device_node *node = dev->of_node; + struct bm_portal_config *pcfg; + int i, irq, ret; + + if (!of_device_is_available(node)) + return -ENODEV; + + if (of_device_is_compatible(node, "fsl,bman-portal-1.0") || + of_device_is_compatible(node, "fsl,bman-portal-1.0.0")) { + bman_ip_rev = BMAN_REV10; + bman_pool_max = 64; + } else if (of_device_is_compatible(node, "fsl,bman-portal-2.0") || + of_device_is_compatible(node, "fsl,bman-portal-2.0.8")) { + bman_ip_rev = BMAN_REV20; + bman_pool_max = 8; + } else if (of_device_is_compatible(node, "fsl,bman-portal-2.1.0") || + of_device_is_compatible(node, "fsl,bman-portal-2.1.1") || + of_device_is_compatible(node, "fsl,bman-portal-2.1.2") || + of_device_is_compatible(node, "fsl,bman-portal-2.1.3")) { + bman_ip_rev = BMAN_REV21; + bman_pool_max = 64; + } + + pcfg = devm_kmalloc(dev, sizeof(*pcfg), GFP_KERNEL); + if (!pcfg) { + dev_err(dev, "Can't allocate portal config\n"); + return -ENOMEM; + } + + for (i = DPA_PORTAL_CE; i <= DPA_PORTAL_CI; i++) { + ret = of_address_to_resource(node, i, pcfg->addr_phys + i); + if (ret < 0) { + dev_err(dev, "Can't get %s property 'reg::%d'\n", + node->full_name, i); + return ret; + } + ret = devm_request_resource(dev, &iomem_resource, + pcfg->addr_phys + i); + if (ret < 0) + return ret; + pcfg->addr_virt[i] = devm_ioremap_prot(dev, + pcfg->addr_phys[i].start, + resource_size(pcfg->addr_phys + i), + flags[i]); + if (!pcfg->addr_virt[i]) + return -ENXIO; + } + + pcfg->public_cfg.cpu = -1; + + irq = irq_of_parse_and_map(node, 0); + if (irq == NO_IRQ) { + dev_err(dev, "Can't get %s property 'interrupts'\n", + node->full_name); + return -ENXIO; + } + pcfg->public_cfg.irq = irq; + + bman_depletion_fill(&pcfg->public_cfg.mask); + + list_add_tail(&pcfg->list, &unused_pcfgs); + + return 0; +}; + +static int __cold bman_portal_remove(struct platform_device *of_dev) +{ + return 0; +}; + +static const struct of_device_id bman_portal_ids[] = { + { + .compatible = "fsl,bman-portal", + }, + {} +}; +MODULE_DEVICE_TABLE(of, bman_portal_ids); + +static struct platform_driver bman_portal_driver = { + .driver = { + .name = KBUILD_MODNAME, + .of_match_table = bman_portal_ids, + }, + .probe = bman_portal_probe, + .remove = bman_portal_remove, +}; + +static int __init bman_portal_driver_register(struct platform_driver *drv) +{ + int _errno; + struct cpumask slave_cpus; + struct cpumask unshared_cpus = *cpu_none_mask; + struct cpumask shared_cpus = *cpu_none_mask; + LIST_HEAD(unshared_pcfgs); + LIST_HEAD(shared_pcfgs); + struct bm_portal_config *pcfg; + struct bman_portal *p; + int cpu; + struct cpumask offline_cpus; + + _errno = platform_driver_register(drv); + if (_errno < 0) + return _errno; + +/* Initialise the BMan driver. The meat of this function deals with portals. The + * following describes the flow of portal-handling, the code "steps" refer to + * this description; + * 1. Portal configs are parsed from the device-tree into 'unused_pcfgs', with + * ::cpu==-1. Regions and interrupts are mapped (but interrupts are not + * bound). + * 2. The "want_shared" and "want_unshared" lists (as filled by the + * "bportals=[...]" bootarg) are processed, allocating portals and assigning + * them to cpus, placing them in the relevant list and setting ::cpu as + * appropriate. If no "bportals" bootarg was present, the defaut is to try to + * assign portals to all online cpus at the time of driver initialisation. + * Any failure to allocate portals (when parsing the "want" lists or when + * using default behaviour) will be silently tolerated (the "fixup" logic in + * step 3 will determine what happens in this case). + * 3. Do fixups relative to cpu_online_mask(). If no portals are marked for + * sharing and sharing is required (because not all cpus have been assigned + * portals), then one portal will marked for sharing. Conversely if no + * sharing is required, any portals marked for sharing will not be shared. It + * may be that sharing occurs when it wasn't expected, if portal allocation + * failed to honour all the requested assignments (including the default + * assignments if no bootarg is present). + * 4. Unshared portals are initialised on their respective cpus. + * 5. Shared portals are initialised on their respective cpus. + * 6. Each remaining cpu is initialised to slave to one of the shared portals, + * which are selected in a round-robin fashion. + */ + /* Step 2. */ + for_each_possible_cpu(cpu) { + if (cpumask_test_cpu(cpu, &want_shared)) { + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &shared_pcfgs); + cpumask_set_cpu(cpu, &shared_cpus); + } + if (cpumask_test_cpu(cpu, &want_unshared)) { + if (cpumask_test_cpu(cpu, &shared_cpus)) + continue; + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &unshared_pcfgs); + cpumask_set_cpu(cpu, &unshared_cpus); + } + } + if (list_empty(&shared_pcfgs) && list_empty(&unshared_pcfgs)) { + /* Default, give an unshared portal to each online cpu */ + for_each_possible_cpu(cpu) { + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &unshared_pcfgs); + cpumask_set_cpu(cpu, &unshared_cpus); + } + } + /* Step 3. */ + cpumask_andnot(&slave_cpus, cpu_possible_mask, &shared_cpus); + cpumask_andnot(&slave_cpus, &slave_cpus, &unshared_cpus); + if (cpumask_empty(&slave_cpus)) { + /* No sharing required */ + if (!list_empty(&shared_pcfgs)) { + /* Migrate "shared" to "unshared" */ + cpumask_or(&unshared_cpus, &unshared_cpus, + &shared_cpus); + cpumask_clear(&shared_cpus); + list_splice_tail(&shared_pcfgs, &unshared_pcfgs); + INIT_LIST_HEAD(&shared_pcfgs); + } + } else { + /* Sharing required */ + if (list_empty(&shared_pcfgs)) { + /* Migrate one "unshared" to "shared" */ + pcfg = get_pcfg(&unshared_pcfgs); + if (!pcfg) { + pr_crit("No portals available!\n"); + return 0; + } + cpumask_clear_cpu(pcfg->public_cfg.cpu, &unshared_cpus); + cpumask_set_cpu(pcfg->public_cfg.cpu, &shared_cpus); + list_add_tail(&pcfg->list, &shared_pcfgs); + } + } + /* Step 4. */ + list_for_each_entry(pcfg, &unshared_pcfgs, list) { + pcfg->public_cfg.is_shared = 0; + p = init_pcfg(pcfg); + } + /* Step 5. */ + list_for_each_entry(pcfg, &shared_pcfgs, list) { + pcfg->public_cfg.is_shared = 1; + p = init_pcfg(pcfg); + if (p) + shared_portals[num_shared_portals++] = p; + } + /* Step 6. */ + if (!cpumask_empty(&slave_cpus)) + for_each_cpu(cpu, &slave_cpus) + init_slave(cpu); + pr_info("Portals initialised\n"); + cpumask_andnot(&offline_cpus, cpu_possible_mask, cpu_online_mask); + for_each_cpu(cpu, &offline_cpus) + bman_offline_cpu(cpu); + +#ifdef CONFIG_HOTPLUG_CPU + register_hotcpu_notifier(&bman_hotplug_cpu_notifier); +#endif + + bman_seed_bpid_range(0, bman_pool_max); + + return 0; +} + +module_driver(bman_portal_driver, + bman_portal_driver_register, platform_driver_unregister); diff --git a/linux/drivers/soc/fsl/qbman/bman_priv.h b/linux/drivers/soc/fsl/qbman/bman_priv.h new file mode 100644 index 00000000..e87f17a3 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_priv.h @@ -0,0 +1,136 @@ +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include "dpaa_sys.h" + +#include + +/* used by CCSR and portal interrupt code */ +enum bm_isr_reg { + bm_isr_status = 0, + bm_isr_enable = 1, + bm_isr_disable = 2, + bm_isr_inhibit = 3 +}; + +#ifdef CONFIG_FSL_BMAN +/* Set depletion thresholds associated with a buffer pool. Requires that the + * operating system have access to BMan CCSR (ie. compiled in support and + * run-time access courtesy of the device-tree). */ +int bm_pool_set(u32 bpid, const u32 *thresholds); +#define BM_POOL_THRESH_SW_ENTER 0 +#define BM_POOL_THRESH_SW_EXIT 1 +#define BM_POOL_THRESH_HW_ENTER 2 +#define BM_POOL_THRESH_HW_EXIT 3 + +/* Read the free buffer count for a given buffer */ +u32 bm_pool_free_buffers(u32 bpid); + +#endif /* CONFIG_FSL_BMAN */ + +#if defined(CONFIG_FSL_BMAN_PORTAL) || defined(CONFIG_FSL_BMAN_PORTAL_MODULE) +/* Revision info (for errata and feature handling) */ +#define BMAN_REV10 0x0100 +#define BMAN_REV20 0x0200 +#define BMAN_REV21 0x0201 +extern u16 bman_ip_rev; /* 0 if uninitialised, otherwise BMAN_REVx */ + +struct bm_portal_config { + /* Corenet portal addresses; + * [0]==cache-enabled, [1]==cache-inhibited. */ + __iomem void *addr_virt[2]; +#ifndef __rtems__ + struct resource addr_phys[2]; + /* Allow these to be joined in lists */ + struct list_head list; +#endif /* __rtems__ */ + /* User-visible portal configuration settings */ + struct bman_portal_config public_cfg; +}; + +/* Hooks from bman_driver.c in to bman_high.c */ +struct bman_portal *bman_create_portal( + struct bman_portal *portal, + const struct bm_portal_config *config); +struct bman_portal *bman_create_affine_portal( + const struct bm_portal_config *config); +struct bman_portal *bman_create_affine_slave(struct bman_portal *redirect, + int cpu); +void bman_destroy_portal(struct bman_portal *bm); + +const struct bm_portal_config *bman_destroy_affine_portal(void); + +/* Stockpile build constants. The _LOW value: when bman_acquire() is called and + * the stockpile fill-level is <= _LOW, an acquire is attempted from h/w but it + * might fail (if the buffer pool is depleted). So this value provides some + * "stagger" in that the bman_acquire() function will only fail if lots of bufs + * are requested at once or if h/w has been tested a couple of times without + * luck. The _HIGH value: when bman_release() is called and the stockpile + * fill-level is >= _HIGH, a release is attempted to h/w but it might fail (if + * the release ring is full). So this value provides some "stagger" so that + * ring-access is retried a couple of times prior to the API returning a + * failure. The following *must* be true; + * BMAN_STOCKPILE_HIGH-BMAN_STOCKPILE_LOW > 8 + * (to avoid thrashing) + * BMAN_STOCKPILE_SZ >= 16 + * (as the release logic expects to either send 8 buffers to hw prior to + * adding the given buffers to the stockpile or add the buffers to the + * stockpile before sending 8 to hw, as the API must be an all-or-nothing + * success/fail.) + */ +#define BMAN_STOCKPILE_SZ 16u /* number of bufs in per-pool cache */ +#define BMAN_STOCKPILE_LOW 2u /* when fill is <= this, acquire from hw */ +#define BMAN_STOCKPILE_HIGH 14u /* when fill is >= this, release to hw */ + +/*************************************************/ +/* BMan s/w corenet portal, low-level i/face */ +/*************************************************/ + +/* Used by all portal interrupt registers except 'inhibit' + * This mask contains all the "irqsource" bits visible to API users + */ +#define BM_PIRQ_VISIBLE (BM_PIRQ_RCRI | BM_PIRQ_BSCN) + +/* These are bm__(). So for example, bm_disable_write() means "write + * the disable register" rather than "disable the ability to write". */ +#define bm_isr_status_read(bm) __bm_isr_read(bm, bm_isr_status) +#define bm_isr_status_clear(bm, m) __bm_isr_write(bm, bm_isr_status, m) +#define bm_isr_enable_read(bm) __bm_isr_read(bm, bm_isr_enable) +#define bm_isr_enable_write(bm, v) __bm_isr_write(bm, bm_isr_enable, v) +#define bm_isr_disable_read(bm) __bm_isr_read(bm, bm_isr_disable) +#define bm_isr_disable_write(bm, v) __bm_isr_write(bm, bm_isr_disable, v) +#define bm_isr_inhibit(bm) __bm_isr_write(bm, bm_isr_inhibit, 1) +#define bm_isr_uninhibit(bm) __bm_isr_write(bm, bm_isr_inhibit, 0) + +const struct bm_portal_config * +bman_get_bm_portal_config(const struct bman_portal *portal); +#endif /* CONFIG_FSL_BMAN_PORTAL* */ diff --git a/linux/drivers/soc/fsl/qbman/bman_test.c b/linux/drivers/soc/fsl/qbman/bman_test.c new file mode 100644 index 00000000..154b7374 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_test.c @@ -0,0 +1,60 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_test.h" + +MODULE_AUTHOR("Geoff Thorpe"); +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_DESCRIPTION("BMan testing"); + +static int test_init(void) +{ +#ifdef CONFIG_FSL_BMAN_TEST_API + int loop = 1; + + while (loop--) + bman_test_api(); +#endif +#ifdef CONFIG_FSL_BMAN_TEST_THRESH + bman_test_thresh(); +#endif + return 0; +} + +static void test_exit(void) +{ +} + +module_init(test_init); +module_exit(test_exit); diff --git a/linux/drivers/soc/fsl/qbman/bman_test.h b/linux/drivers/soc/fsl/qbman/bman_test.h new file mode 100644 index 00000000..9c51c38b --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_test.h @@ -0,0 +1,34 @@ +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_priv.h" + +void bman_test_api(void); +void bman_test_thresh(void); diff --git a/linux/drivers/soc/fsl/qbman/bman_test_api.c b/linux/drivers/soc/fsl/qbman/bman_test_api.c new file mode 100644 index 00000000..5585bdf6 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_test_api.c @@ -0,0 +1,188 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_test.h" + +/*************/ +/* constants */ +/*************/ + +#define PORTAL_OPAQUE ((void *)0xf00dbeef) +#define POOL_OPAQUE ((void *)0xdeadabba) +#define NUM_BUFS 93 +#define LOOPS 3 +#define BMAN_TOKEN_MASK 0x00FFFFFFFFFFLLU + +/***************/ +/* global vars */ +/***************/ + +static struct bman_pool *pool; +static int depleted; +static struct bm_buffer bufs_in[NUM_BUFS] ____cacheline_aligned; +static struct bm_buffer bufs_out[NUM_BUFS] ____cacheline_aligned; +static int bufs_received; + +/* Predeclare the callback so we can instantiate pool parameters */ +static void depletion_cb(struct bman_portal *, struct bman_pool *, void *, int); + +/**********************/ +/* internal functions */ +/**********************/ + +static void bufs_init(void) +{ + int i; + + for (i = 0; i < NUM_BUFS; i++) + bm_buffer_set64(&bufs_in[i], 0xfedc01234567LLU * i); + bufs_received = 0; +} + +static inline int bufs_cmp(const struct bm_buffer *a, const struct bm_buffer *b) +{ + if ((bman_ip_rev == BMAN_REV20) || (bman_ip_rev == BMAN_REV21)) { + + /* On SoCs with BMan revison 2.0, BMan only respects the 40 + * LS-bits of buffer addresses, masking off the upper 8-bits on + * release commands. The API provides for 48-bit addresses + * because some SoCs support all 48-bits. When generating + * garbage addresses for testing, we either need to zero the + * upper 8-bits when releasing to BMan (otherwise we'll be + * disappointed when the buffers we acquire back from BMan + * don't match), or we need to mask the upper 8-bits off when + * comparing. We do the latter. + */ + if ((bm_buffer_get64(a) & BMAN_TOKEN_MASK) + < (bm_buffer_get64(b) & BMAN_TOKEN_MASK)) + return -1; + if ((bm_buffer_get64(a) & BMAN_TOKEN_MASK) + > (bm_buffer_get64(b) & BMAN_TOKEN_MASK)) + return 1; + } else { + if (bm_buffer_get64(a) < bm_buffer_get64(b)) + return -1; + if (bm_buffer_get64(a) > bm_buffer_get64(b)) + return 1; + } + + return 0; +} + +static void bufs_confirm(void) +{ + int i, j; + + for (i = 0; i < NUM_BUFS; i++) { + int matches = 0; + + for (j = 0; j < NUM_BUFS; j++) + if (!bufs_cmp(&bufs_in[i], &bufs_out[j])) + matches++; + BUG_ON(matches != 1); + } +} + +/********/ +/* test */ +/********/ + +static void depletion_cb(struct bman_portal *__portal, struct bman_pool *__pool, + void *pool_ctx, int __depleted) +{ + BUG_ON(__pool != pool); + BUG_ON(pool_ctx != POOL_OPAQUE); + depleted = __depleted; +} + +void bman_test_api(void) +{ + struct bman_pool_params pparams = { + .flags = BMAN_POOL_FLAG_DEPLETION | BMAN_POOL_FLAG_DYNAMIC_BPID, + .cb = depletion_cb, + .cb_ctx = POOL_OPAQUE, + }; + int i, loops = LOOPS; + + bufs_init(); + + pr_info(" --- Starting high-level test ---\n"); + + pool = bman_new_pool(&pparams); + BUG_ON(!pool); + + /*******************/ + /* Release buffers */ + /*******************/ +do_loop: + i = 0; + while (i < NUM_BUFS) { + u32 flags = BMAN_RELEASE_FLAG_WAIT; + int num = 8; + + if ((i + num) > NUM_BUFS) + num = NUM_BUFS - i; + if ((i + num) == NUM_BUFS) + flags |= BMAN_RELEASE_FLAG_WAIT_SYNC; + if (bman_release(pool, bufs_in + i, num, flags)) + panic("bman_release() failed\n"); + i += num; + } + + /*******************/ + /* Acquire buffers */ + /*******************/ + while (i > 0) { + int tmp, num = 8; + + if (num > i) + num = i; + tmp = bman_acquire(pool, bufs_out + i - num, num, 0); + BUG_ON(tmp != num); + i -= num; + } + i = bman_acquire(pool, NULL, 1, 0); + BUG_ON(i > 0); + + bufs_confirm(); + + if (--loops) + goto do_loop; + + /************/ + /* Clean up */ + /************/ + bman_free_pool(pool); + pr_info(" --- Finished high-level test ---\n"); +} diff --git a/linux/drivers/soc/fsl/qbman/bman_test_thresh.c b/linux/drivers/soc/fsl/qbman/bman_test_thresh.c new file mode 100644 index 00000000..c0f045be --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_test_thresh.c @@ -0,0 +1,216 @@ +#include + +#include + +/* Copyright 2010 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "bman_test.h" + +/* Test constants */ +#define TEST_NUMBUFS 129728 +#define TEST_EXIT 129536 +#define TEST_ENTRY 129024 + +struct affine_test_data { + struct task_struct *t; + int cpu; +#ifndef __rtems__ + int expect_affinity; +#endif /* __rtems__ */ + int drain; + int num_enter; + int num_exit; + struct list_head node; + struct completion wakethread; + struct completion wakeparent; +}; + +static void cb_depletion(struct bman_portal *portal, + struct bman_pool *pool, + void *opaque, + int depleted) +{ + struct affine_test_data *data = opaque; + int c = smp_processor_id(); + + pr_info("%s(): bpid=%d, depleted=%d, cpu=%d, original=%d\n", __func__, + bman_get_params(pool)->bpid, !!depleted, c, data->cpu); + /* We should be executing on the CPU of the thread that owns the pool if + * and that CPU has an affine portal (ie. it isn't slaved). */ +#ifndef __rtems__ + BUG_ON((c != data->cpu) && data->expect_affinity); + BUG_ON((c == data->cpu) && !data->expect_affinity); +#endif /* __rtems__ */ + if (depleted) + data->num_enter++; + else + data->num_exit++; +} + +/* Params used to set up a pool, this also dynamically allocates a BPID */ +static const struct bman_pool_params params_nocb = { + .flags = BMAN_POOL_FLAG_DYNAMIC_BPID | BMAN_POOL_FLAG_THRESH, + .thresholds = { TEST_ENTRY, TEST_EXIT, 0, 0 } +}; + +/* Params used to set up each cpu's pool with callbacks enabled */ +static struct bman_pool_params params_cb = { + .bpid = 0, /* will be replaced to match pool_nocb */ + .flags = BMAN_POOL_FLAG_DEPLETION, + .cb = cb_depletion +}; + +static struct bman_pool *pool_nocb; +static LIST_HEAD(threads); + +static int affine_test(void *__data) +{ + struct bman_pool *pool; + struct affine_test_data *data = __data; + struct bman_pool_params my_params = params_cb; + + pr_info("Thread %d: starting\n", data->cpu); + /* create the pool */ + my_params.cb_ctx = data; + pool = bman_new_pool(&my_params); + BUG_ON(!pool); + complete(&data->wakeparent); + wait_for_completion(&data->wakethread); + init_completion(&data->wakethread); + + /* if we're the drainer, we get signalled for that */ + if (data->drain) { + struct bm_buffer buf; + int ret; + + pr_info("Thread %d: draining...\n", data->cpu); + do { + ret = bman_acquire(pool, &buf, 1, 0); + } while (ret > 0); + pr_info("Thread %d: draining done.\n", data->cpu); + complete(&data->wakeparent); + wait_for_completion(&data->wakethread); + init_completion(&data->wakethread); + } + + /* cleanup */ + bman_free_pool(pool); + while (!kthread_should_stop()) + cpu_relax(); + pr_info("Thread %d: exiting\n", data->cpu); + return 0; +} + +static struct affine_test_data *start_affine_test(int cpu, int drain) +{ + struct affine_test_data *data = kmalloc(sizeof(*data), GFP_KERNEL); + + if (!data) + return NULL; + data->cpu = cpu; +#ifndef __rtems__ + data->expect_affinity = cpumask_test_cpu(cpu, bman_affine_cpus()); +#endif /* __rtems__ */ + data->drain = drain; + data->num_enter = 0; + data->num_exit = 0; + init_completion(&data->wakethread); + init_completion(&data->wakeparent); + list_add_tail(&data->node, &threads); + data->t = kthread_create(affine_test, data, "threshtest%d", cpu); +#ifndef __rtems__ + BUG_ON(IS_ERR(data->t)); +#else /* __rtems__ */ + BUG_ON(data->t == NULL); +#endif /* __rtems__ */ + kthread_bind(data->t, cpu); + wake_up_process(data->t); + return data; +} + +void bman_test_thresh(void) +{ + int loop = TEST_NUMBUFS; + int ret, num_cpus = 0; + struct affine_test_data *data, *drainer = NULL; + + pr_info("%s(): Start\n", __func__); + + /* allocate a BPID and seed it */ + pool_nocb = bman_new_pool(¶ms_nocb); + BUG_ON(!pool_nocb); + while (loop--) { + struct bm_buffer buf; + + bm_buffer_set64(&buf, 0x0badbeef + loop); + ret = bman_release(pool_nocb, &buf, 1, + BMAN_RELEASE_FLAG_WAIT); + BUG_ON(ret); + } + while (!bman_rcr_is_empty()) + cpu_relax(); + pr_info("%s(): Buffers are in\n", __func__); + + /* create threads and wait for them to create pools */ + params_cb.bpid = bman_get_params(pool_nocb)->bpid; +#ifndef __rtems__ + for_each_cpu(loop, cpu_online_mask) { +#else /* __rtems__ */ + for (loop = 0; loop < rtems_get_processor_count(); ++loop) { +#endif /* __rtems__ */ + data = start_affine_test(loop, drainer ? 0 : 1); + BUG_ON(!data); + if (!drainer) + drainer = data; + num_cpus++; + wait_for_completion(&data->wakeparent); + } + + /* signal the drainer to start draining */ + complete(&drainer->wakethread); + wait_for_completion(&drainer->wakeparent); + init_completion(&drainer->wakeparent); + + /* tear down */ + list_for_each_entry_safe(data, drainer, &threads, node) { + complete(&data->wakethread); + ret = kthread_stop(data->t); + BUG_ON(ret); + list_del(&data->node); + /* check that we get the expected callbacks (and no others) */ + BUG_ON(data->num_enter != 1); + BUG_ON(data->num_exit != 0); + kfree(data); + } + bman_free_pool(pool_nocb); + + pr_info("%s(): Done\n", __func__); +} diff --git a/linux/drivers/soc/fsl/qbman/bman_utils.c b/linux/drivers/soc/fsl/qbman/bman_utils.c new file mode 100644 index 00000000..c6fa0b33 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/bman_utils.c @@ -0,0 +1,76 @@ +#include + +#include + +/* Copyright 2009 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "dpaa_sys.h" + +#include + +/* BMan APIs are front-ends to the common code */ + +static DECLARE_DPAA_RESOURCE(bpalloc); /* BPID allocator */ + +/* BPID allocator front-end */ + +int bman_alloc_bpid_range(u32 *result, u32 count, u32 align, int partial) +{ + return dpaa_resource_new(&bpalloc, result, count, align, partial); +} +EXPORT_SYMBOL(bman_alloc_bpid_range); + +static int bp_cleanup(u32 bpid) +{ + return bman_shutdown_pool(bpid) == 0; +} +void bman_release_bpid_range(u32 bpid, u32 count) +{ + u32 total_invalid = dpaa_resource_release(&bpalloc, + bpid, count, bp_cleanup); + + if (total_invalid) + pr_err("BPID range [%d..%d] (%d) had %d leaks\n", + bpid, bpid + count - 1, count, total_invalid); +} +EXPORT_SYMBOL(bman_release_bpid_range); + +void bman_seed_bpid_range(u32 bpid, u32 count) +{ + dpaa_resource_seed(&bpalloc, bpid, count); +} +EXPORT_SYMBOL(bman_seed_bpid_range); + +int bman_reserve_bpid_range(u32 bpid, u32 count) +{ + return dpaa_resource_reserve(&bpalloc, bpid, count); +} +EXPORT_SYMBOL(bman_reserve_bpid_range); diff --git a/linux/drivers/soc/fsl/qbman/dpaa_resource.c b/linux/drivers/soc/fsl/qbman/dpaa_resource.c new file mode 100644 index 00000000..3f7b74bf --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/dpaa_resource.c @@ -0,0 +1,363 @@ +#include + +#include + +/* Copyright 2009 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#if defined(CONFIG_FSL_BMAN_PORTAL) || \ + defined(CONFIG_FSL_BMAN_PORTAL_MODULE) || \ + defined(CONFIG_FSL_QMAN_PORTAL) || \ + defined(CONFIG_FSL_QMAN_PORTAL_MODULE) +#include "dpaa_sys.h" + +/* The allocator is a (possibly-empty) list of these */ +struct dpaa_resource_node { + struct list_head list; + u32 base; + u32 num; + /* refcount and is_alloced are only set + when the node is in the used list */ + unsigned int refcount; + int is_alloced; +}; + +#ifdef DPAA_RESOURCE_DEBUG +#define DPRINT pr_info +static void DUMP(struct dpaa_resource *alloc) +{ + int off = 0; + char buf[256]; + struct dpaa_resource_node *p; + + pr_info("Free Nodes\n"); + list_for_each_entry(p, &alloc->free, list) { + if (off < 255) + off += snprintf(buf + off, 255-off, "{%d,%d}", + p->base, p->base + p->num - 1); + } + pr_info("%s\n", buf); + + off = 0; + pr_info("Used Nodes\n"); + list_for_each_entry(p, &alloc->used, list) { + if (off < 255) + off += snprintf(buf + off, 255-off, "{%d,%d}", + p->base, p->base + p->num - 1); + } + pr_info("%s\n", buf); +} +#else +#define DPRINT(x...) +#define DUMP(a) +#endif + +int dpaa_resource_new(struct dpaa_resource *alloc, u32 *result, + u32 count, u32 align, int partial) +{ + struct dpaa_resource_node *i = NULL, *next_best = NULL, + *used_node = NULL; + u32 base, next_best_base = 0, num = 0, next_best_num = 0; + struct dpaa_resource_node *margin_left, *margin_right; + + *result = (u32)-1; + DPRINT("alloc_range(%d,%d,%d)\n", count, align, partial); + DUMP(alloc); + /* If 'align' is 0, it should behave as though it was 1 */ + if (!align) + align = 1; + margin_left = kmalloc(sizeof(*margin_left), GFP_KERNEL); + if (!margin_left) + goto err; + margin_right = kmalloc(sizeof(*margin_right), GFP_KERNEL); + if (!margin_right) { + kfree(margin_left); + goto err; + } + spin_lock_irq(&alloc->lock); + list_for_each_entry(i, &alloc->free, list) { + base = (i->base + align - 1) / align; + base *= align; + if ((base - i->base) >= i->num) + /* alignment is impossible, regardless of count */ + continue; + num = i->num - (base - i->base); + if (num >= count) { + /* this one will do nicely */ + num = count; + goto done; + } + if (num > next_best_num) { + next_best = i; + next_best_base = base; + next_best_num = num; + } + } + if (partial && next_best) { + i = next_best; + base = next_best_base; + num = next_best_num; + } else + i = NULL; +done: + if (i) { + if (base != i->base) { + margin_left->base = i->base; + margin_left->num = base - i->base; + list_add_tail(&margin_left->list, &i->list); + } else + kfree(margin_left); + if ((base + num) < (i->base + i->num)) { + margin_right->base = base + num; + margin_right->num = (i->base + i->num) - + (base + num); + list_add(&margin_right->list, &i->list); + } else + kfree(margin_right); + list_del(&i->list); + kfree(i); + *result = base; + } + spin_unlock_irq(&alloc->lock); +err: + DPRINT("returning %d\n", i ? num : -ENOMEM); + DUMP(alloc); + if (!i) + return -ENOMEM; + + /* Add the allocation to the used list with a refcount of 1 */ + used_node = kmalloc(sizeof(*used_node), GFP_KERNEL); + if (!used_node) + return -ENOMEM; + used_node->base = *result; + used_node->num = num; + used_node->refcount = 1; + used_node->is_alloced = 1; + list_add_tail(&used_node->list, &alloc->used); + return (int)num; +} +EXPORT_SYMBOL(dpaa_resource_new); + +/* Allocate the list node using GFP_ATOMIC, because we *really* want to avoid + * forcing error-handling on to users in the deallocation path. */ +static void _dpaa_resource_free(struct dpaa_resource *alloc, u32 base_id, + u32 count) +{ + struct dpaa_resource_node *i, + *node = kmalloc(sizeof(*node), GFP_ATOMIC); + + BUG_ON(!node); + DPRINT("release_range(%d,%d)\n", base_id, count); + DUMP(alloc); + BUG_ON(!count); + spin_lock_irq(&alloc->lock); + + node->base = base_id; + node->num = count; + list_for_each_entry(i, &alloc->free, list) { + if (i->base >= node->base) { + /* BUG_ON(any overlapping) */ + BUG_ON(i->base < (node->base + node->num)); + list_add_tail(&node->list, &i->list); + goto done; + } + } + list_add_tail(&node->list, &alloc->free); +done: + /* Merge to the left */ + i = list_entry(node->list.prev, struct dpaa_resource_node, list); + if (node->list.prev != &alloc->free) { + BUG_ON((i->base + i->num) > node->base); + if ((i->base + i->num) == node->base) { + node->base = i->base; + node->num += i->num; + list_del(&i->list); + kfree(i); + } + } + /* Merge to the right */ + i = list_entry(node->list.next, struct dpaa_resource_node, list); + if (node->list.next != &alloc->free) { + BUG_ON((node->base + node->num) > i->base); + if ((node->base + node->num) == i->base) { + node->num += i->num; + list_del(&i->list); + kfree(i); + } + } + spin_unlock_irq(&alloc->lock); + DUMP(alloc); +} + +static void dpaa_resource_free(struct dpaa_resource *alloc, u32 base_id, + u32 count) +{ + struct dpaa_resource_node *i = NULL; + + spin_lock_irq(&alloc->lock); + + /* First find the node in the used list and decrement its ref count */ + list_for_each_entry(i, &alloc->used, list) { + if (i->base == base_id && i->num == count) { + --i->refcount; + if (i->refcount == 0) { + list_del(&i->list); + spin_unlock_irq(&alloc->lock); + if (i->is_alloced) + _dpaa_resource_free(alloc, base_id, + count); + kfree(i); + return; + } + spin_unlock_irq(&alloc->lock); + return; + } + } + /* Couldn't find the allocation */ + pr_err("Attempt to free ID 0x%x COUNT %d that wasn't alloc'd or reserved\n", + base_id, count); + spin_unlock_irq(&alloc->lock); +} + +/* Same as free but no previous allocation checking is needed */ +void dpaa_resource_seed(struct dpaa_resource *alloc, u32 base_id, u32 count) +{ + _dpaa_resource_free(alloc, base_id, count); +} +EXPORT_SYMBOL(dpaa_resource_seed); + +/* Like 'new' but specifies the desired range, returns -ENOMEM if the entire + * desired range is not available, or 0 for success + */ +int dpaa_resource_reserve(struct dpaa_resource *alloc, u32 base, u32 num) +{ + struct dpaa_resource_node *i = NULL, *used_node; + + DPRINT("alloc_reserve(%d,%d)\n", base, num); + DUMP(alloc); + + spin_lock_irq(&alloc->lock); + + /* Check for the node in the used list. + If found, increase it's refcount */ + list_for_each_entry(i, &alloc->used, list) { + if ((i->base == base) && (i->num == num)) { + ++i->refcount; + spin_unlock_irq(&alloc->lock); + return 0; + } + if ((base >= i->base) && (base < (i->base + i->num))) { + /* This is an attempt to reserve a region that was + already reserved or alloced with a different + base or num */ + pr_err("Cannot reserve %d - %d, it overlaps with" + " existing reservation from %d - %d\n", + base, base + num - 1, i->base, + i->base + i->num - 1); + spin_unlock_irq(&alloc->lock); + return -1; + } + } + /* Check to make sure this ID isn't in the free list */ + list_for_each_entry(i, &alloc->free, list) { + if ((base >= i->base) && (base < (i->base + i->num))) { + /* yep, the reservation is within this node */ + pr_err("Cannot reserve %d - %d, it overlaps with" + " free range %d - %d and must be alloced\n", + base, base + num - 1, + i->base, i->base + i->num - 1); + spin_unlock_irq(&alloc->lock); + return -1; + } + } + /* Add the allocation to the used list with a refcount of 1 */ + used_node = kmalloc(sizeof(*used_node), GFP_KERNEL); + if (!used_node) { + spin_unlock_irq(&alloc->lock); + return -ENOMEM; + + } + used_node->base = base; + used_node->num = num; + used_node->refcount = 1; + used_node->is_alloced = 0; + list_add_tail(&used_node->list, &alloc->used); + spin_unlock_irq(&alloc->lock); + return 0; +} +EXPORT_SYMBOL(dpaa_resource_reserve); + +/* This is a sort-of-conditional dpaa_resource_free() routine. Eg. when + * releasing FQIDs (probably from user-space), it can filter out those + * that aren't in the OOS state (better to leak a h/w resource than to + * crash). This function returns the number of invalid IDs that were not + * released. +*/ +u32 dpaa_resource_release(struct dpaa_resource *alloc, + u32 id, u32 count, int (*is_valid)(u32 id)) +{ + int valid_mode = 0; + u32 loop = id, total_invalid = 0; + + while (loop < (id + count)) { + int isvalid = is_valid ? is_valid(loop) : 1; + + if (!valid_mode) { + /* We're looking for a valid ID to terminate an invalid + * range */ + if (isvalid) { + /* We finished a range of invalid IDs, a valid + * range is now underway */ + valid_mode = 1; + count -= (loop - id); + id = loop; + } else + total_invalid++; + } else { + /* We're looking for an invalid ID to terminate a + * valid range */ + if (!isvalid) { + /* Release the range of valid IDs, an unvalid + * range is now underway */ + if (loop > id) + dpaa_resource_free(alloc, id, + loop - id); + valid_mode = 0; + } + } + loop++; + } + /* Release any unterminated range of valid IDs */ + if (valid_mode && count) + dpaa_resource_free(alloc, id, count); + return total_invalid; +} +EXPORT_SYMBOL(dpaa_resource_release); +#endif /* CONFIG_FSL_*MAN_PORTAL* */ diff --git a/linux/drivers/soc/fsl/qbman/dpaa_sys.h b/linux/drivers/soc/fsl/qbman/dpaa_sys.h new file mode 100644 index 00000000..85f87800 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/dpaa_sys.h @@ -0,0 +1,292 @@ +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#ifndef __DPAA_SYS_H +#define __DPAA_SYS_H + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#ifdef CONFIG_HOTPLUG_CPU +#include +#endif + +#include +#ifdef __rtems__ +#include +#include +#include +#include +#include +#include +#define DPAA_NOCACHENOLOAD_ALIGNED_REGION(designator, size) \ + BSP_NOCACHENOLOAD_SUBSECTION(designator) __aligned(size) \ + uint8_t designator[size] +#endif /* __rtems__ */ + +struct dpaa_resource { + struct list_head free; + spinlock_t lock; + struct list_head used; +}; + +#define DECLARE_DPAA_RESOURCE(name) \ +struct dpaa_resource name = { \ + .free = { \ + .prev = &name.free, \ + .next = &name.free \ + }, \ + .lock = __SPIN_LOCK_UNLOCKED(name.lock), \ + .used = { \ + .prev = &name.used, \ + .next = &name.used \ + } \ +} + +int dpaa_resource_new(struct dpaa_resource *alloc, u32 *result, + u32 count, u32 align, int partial); +u32 dpaa_resource_release(struct dpaa_resource *alloc, + u32 id, u32 count, int (*is_valid)(u32 id)); +void dpaa_resource_seed(struct dpaa_resource *alloc, u32 base_id, u32 count); +int dpaa_resource_reserve(struct dpaa_resource *alloc, u32 base, u32 num); + +/* When copying aligned words or shorts, try to avoid memcpy() */ +#define CONFIG_TRY_BETTER_MEMCPY + +/* For 2-element tables related to cache-inhibited and cache-enabled mappings */ +#define DPA_PORTAL_CE 0 +#define DPA_PORTAL_CI 1 + +/* Misc inline assists */ + +/* TODO: NB, we currently assume that hwsync() and lwsync() imply compiler + * barriers and that dcb*() won't fall victim to compiler or execution + * reordering with respect to other code/instructions that manipulate the same + * cacheline. */ +#define hwsync() __asm__ __volatile__ ("sync" : : : "memory") +#ifndef __rtems__ +#define lwsync() __asm__ __volatile__ (stringify_in_c(LWSYNC) : : : "memory") +#else /* __rtems__ */ + #ifdef __PPC_CPU_E6500__ + #define lwsync() ppc_light_weight_synchronize() + #else + #define lwsync() ppc_synchronize_data() + #endif +#endif /* __rtems__ */ +#define dcbf(p) __asm__ __volatile__ ("dcbf 0,%0" : : "r" (p) : "memory") +#define dcbt_ro(p) __asm__ __volatile__ ("dcbt 0,%0" : : "r" (p)) +#define dcbt_rw(p) __asm__ __volatile__ ("dcbtst 0,%0" : : "r" (p)) +#define dcbi(p) dcbf(p) +#ifdef CONFIG_PPC_E500MC +#define dcbzl(p) __asm__ __volatile__ ("dcbzl 0,%0" : : "r" (p)) +#define dcbz_64(p) dcbzl(p) +#define dcbf_64(p) dcbf(p) +/* Commonly used combo */ +#define dcbit_ro(p) \ + do { \ + dcbi(p); \ + dcbt_ro(p); \ + } while (0) +#else +#define dcbz(p) __asm__ __volatile__ ("dcbz 0,%0" : : "r" (p)) +#define dcbz_64(p) \ + do { \ + dcbz((u32)p + 32); \ + dcbz(p); \ + } while (0) +#define dcbf_64(p) \ + do { \ + dcbf((u32)p + 32); \ + dcbf(p); \ + } while (0) +/* Commonly used combo */ +#define dcbit_ro(p) \ + do { \ + dcbi(p); \ + dcbi((u32)p + 32); \ + dcbt_ro(p); \ + dcbt_ro((u32)p + 32); \ + } while (0) +#endif /* CONFIG_PPC_E500MC */ + +static inline u64 mfatb(void) +{ + u32 hi, lo, chk; + + do { + hi = mfspr(SPRN_ATBU); + lo = mfspr(SPRN_ATBL); + chk = mfspr(SPRN_ATBU); + } while (unlikely(hi != chk)); + return ((u64)hi << 32) | (u64)lo; +} + +#ifdef CONFIG_FSL_DPA_CHECKING +#define DPA_ASSERT(x) WARN_ON(!(x)) +#else +#define DPA_ASSERT(x) +#endif + +#ifdef CONFIG_TRY_BETTER_MEMCPY +static inline void copy_words(void *dest, const void *src, size_t sz) +{ + u32 *__dest = dest; + const u32 *__src = src; + size_t __sz = sz >> 2; + + BUG_ON((unsigned long)dest & 0x3); + BUG_ON((unsigned long)src & 0x3); + BUG_ON(sz & 0x3); + while (__sz--) + *(__dest++) = *(__src++); +} +#else +#define copy_words memcpy +#endif + +/* RB-trees */ + +/* We encapsulate RB-trees so that its easier to use non-linux forms in + * non-linux systems. This also encapsulates the extra plumbing that linux code + * usually provides when using RB-trees. This encapsulation assumes that the + * data type held by the tree is u32. */ + +struct dpa_rbtree { + struct rb_root root; +}; +#define DPA_RBTREE { .root = RB_ROOT } + +static inline void dpa_rbtree_init(struct dpa_rbtree *tree) +{ + tree->root = RB_ROOT; +} + +#define IMPLEMENT_DPA_RBTREE(name, type, node_field, val_field) \ +static inline int name##_push(struct dpa_rbtree *tree, type *obj) \ +{ \ + struct rb_node *parent = NULL, **p = &tree->root.rb_node; \ + while (*p) { \ + u32 item; \ + parent = *p; \ + item = rb_entry(parent, type, node_field)->val_field; \ + if (obj->val_field < item) \ + p = &parent->rb_left; \ + else if (obj->val_field > item) \ + p = &parent->rb_right; \ + else \ + return -EBUSY; \ + } \ + rb_link_node(&obj->node_field, parent, p); \ + rb_insert_color(&obj->node_field, &tree->root); \ + return 0; \ +} \ +static inline void name##_del(struct dpa_rbtree *tree, type *obj) \ +{ \ + rb_erase(&obj->node_field, &tree->root); \ +} \ +static inline type *name##_find(struct dpa_rbtree *tree, u32 val) \ +{ \ + type *ret; \ + struct rb_node *p = tree->root.rb_node; \ + while (p) { \ + ret = rb_entry(p, type, node_field); \ + if (val < ret->val_field) \ + p = p->rb_left; \ + else if (val > ret->val_field) \ + p = p->rb_right; \ + else \ + return ret; \ + } \ + return NULL; \ +} + +#ifndef __rtems__ +/* Bootargs */ + +/* QMan has "qportals=" and BMan has "bportals=", they use the same syntax + * though; a comma-separated list of items, each item being a cpu index and/or a + * range of cpu indices, and each item optionally be prefixed by "s" to indicate + * that the portal associated with that cpu should be shared. See bman_driver.c + * for more specifics. */ +static int __parse_portals_cpu(const char **s, unsigned int *cpu) +{ + *cpu = 0; + if (!isdigit(**s)) + return -EINVAL; + while (isdigit(**s)) + *cpu = *cpu * 10 + (*((*s)++) - '0'); + return 0; +} +static inline int parse_portals_bootarg(char *str, struct cpumask *want_shared, + struct cpumask *want_unshared, + const char *argname) +{ + const char *s = str; + unsigned int shared, cpu1, cpu2, loop; + +keep_going: + if (*s == 's') { + shared = 1; + s++; + } else + shared = 0; + if (__parse_portals_cpu(&s, &cpu1)) + goto err; + if (*s == '-') { + s++; + if (__parse_portals_cpu(&s, &cpu2)) + goto err; + if (cpu2 < cpu1) + goto err; + } else + cpu2 = cpu1; + for (loop = cpu1; loop <= cpu2; loop++) + cpumask_set_cpu(loop, shared ? want_shared : want_unshared); + if (*s == ',') { + s++; + goto keep_going; + } else if ((*s == '\0') || isspace(*s)) + return 0; +err: + pr_crit("Malformed %s argument: %s, offset: %lu\n", argname, str, + (unsigned long)s - (unsigned long)str); + return -EINVAL; +} +#endif /* __rtems__ */ +#endif /* __DPAA_SYS_H */ diff --git a/linux/drivers/soc/fsl/qbman/qman-debugfs.c b/linux/drivers/soc/fsl/qbman/qman-debugfs.c new file mode 100644 index 00000000..ecdb6be6 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman-debugfs.c @@ -0,0 +1,1317 @@ +#include + +#include + +/* Copyright 2010 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ +#include "qman_priv.h" + +#define MAX_FQID (0x00ffffff) +#define QM_FQD_BLOCK_SIZE 64 +#define QM_FQD_AR (0xC10) + +static u32 fqid_max; +static u64 qman_ccsr_start; +static u64 qman_ccsr_size; + +static const char * const state_txt[] = { + "Out of Service", + "Retired", + "Tentatively Scheduled", + "Truly Scheduled", + "Parked", + "Active, Active Held or Held Suspended", + "Unknown State 6", + "Unknown State 7", + NULL, +}; + +static const u8 fqd_states[] = { + QM_MCR_NP_STATE_OOS, QM_MCR_NP_STATE_RETIRED, QM_MCR_NP_STATE_TEN_SCHED, + QM_MCR_NP_STATE_TRU_SCHED, QM_MCR_NP_STATE_PARKED, + QM_MCR_NP_STATE_ACTIVE}; + +struct mask_to_text { + u16 mask; + const char *txt; +}; + +struct mask_filter_s { + u16 mask; + u8 filter; +}; + +static const struct mask_filter_s mask_filter[] = { + {QM_FQCTRL_PREFERINCACHE, 0}, + {QM_FQCTRL_PREFERINCACHE, 1}, + {QM_FQCTRL_HOLDACTIVE, 0}, + {QM_FQCTRL_HOLDACTIVE, 1}, + {QM_FQCTRL_AVOIDBLOCK, 0}, + {QM_FQCTRL_AVOIDBLOCK, 1}, + {QM_FQCTRL_FORCESFDR, 0}, + {QM_FQCTRL_FORCESFDR, 1}, + {QM_FQCTRL_CPCSTASH, 0}, + {QM_FQCTRL_CPCSTASH, 1}, + {QM_FQCTRL_CTXASTASHING, 0}, + {QM_FQCTRL_CTXASTASHING, 1}, + {QM_FQCTRL_ORP, 0}, + {QM_FQCTRL_ORP, 1}, + {QM_FQCTRL_TDE, 0}, + {QM_FQCTRL_TDE, 1}, + {QM_FQCTRL_CGE, 0}, + {QM_FQCTRL_CGE, 1} +}; + +static const struct mask_to_text fq_ctrl_text_list[] = { + { + .mask = QM_FQCTRL_PREFERINCACHE, + .txt = "Prefer in cache", + }, + { + .mask = QM_FQCTRL_HOLDACTIVE, + .txt = "Hold active in portal", + }, + { + .mask = QM_FQCTRL_AVOIDBLOCK, + .txt = "Avoid Blocking", + }, + { + .mask = QM_FQCTRL_FORCESFDR, + .txt = "High-priority SFDRs", + }, + { + .mask = QM_FQCTRL_CPCSTASH, + .txt = "CPC Stash Enable", + }, + { + .mask = QM_FQCTRL_CTXASTASHING, + .txt = "Context-A stashing", + }, + { + .mask = QM_FQCTRL_ORP, + .txt = "ORP Enable", + }, + { + .mask = QM_FQCTRL_TDE, + .txt = "Tail-Drop Enable", + }, + { + .mask = QM_FQCTRL_CGE, + .txt = "Congestion Group Enable", + }, + { + .mask = 0, + .txt = NULL, + } +}; + +static const char *get_fqd_ctrl_text(u16 mask) +{ + int i = 0; + + while (fq_ctrl_text_list[i].txt != NULL) { + if (fq_ctrl_text_list[i].mask == mask) + return fq_ctrl_text_list[i].txt; + i++; + } + return NULL; +} + +static const struct mask_to_text stashing_text_list[] = { + { + .mask = QM_STASHING_EXCL_CTX, + .txt = "FQ Ctx Stash" + }, + { + .mask = QM_STASHING_EXCL_DATA, + .txt = "Frame Data Stash", + }, + { + .mask = QM_STASHING_EXCL_ANNOTATION, + .txt = "Frame Annotation Stash", + }, + { + .mask = 0, + .txt = NULL, + }, +}; + +static int user_input_convert(const char __user *user_buf, size_t count, + unsigned long *val) +{ + char buf[12]; + + if (count > sizeof(buf) - 1) + return -EINVAL; + if (copy_from_user(buf, user_buf, count)) + return -EFAULT; + buf[count] = '\0'; + if (kstrtoul(buf, 0, val)) + return -EINVAL; + return 0; +} + +struct line_buffer_fq { + u32 buf[8]; + u32 buf_cnt; + int line_cnt; +}; + +static void add_to_line_buffer(struct line_buffer_fq *line_buf, u32 fqid, + struct seq_file *file) +{ + line_buf->buf[line_buf->buf_cnt] = fqid; + line_buf->buf_cnt++; + if (line_buf->buf_cnt == 8) { + /* Buffer is full, flush it */ + if (line_buf->line_cnt != 0) + seq_puts(file, ",\n"); + seq_printf(file, "0x%06x,0x%06x,0x%06x,0x%06x,0x%06x," + "0x%06x,0x%06x,0x%06x", + line_buf->buf[0], line_buf->buf[1], line_buf->buf[2], + line_buf->buf[3], line_buf->buf[4], line_buf->buf[5], + line_buf->buf[6], line_buf->buf[7]); + line_buf->buf_cnt = 0; + line_buf->line_cnt++; + } +} + +static void flush_line_buffer(struct line_buffer_fq *line_buf, + struct seq_file *file) +{ + if (line_buf->buf_cnt) { + int y = 0; + + if (line_buf->line_cnt != 0) + seq_puts(file, ",\n"); + while (y != line_buf->buf_cnt) { + if (y+1 == line_buf->buf_cnt) + seq_printf(file, "0x%06x", line_buf->buf[y]); + else + seq_printf(file, "0x%06x,", line_buf->buf[y]); + y++; + } + line_buf->line_cnt++; + } + if (line_buf->line_cnt) + seq_putc(file, '\n'); +} + +static struct dentry *dfs_root; /* debugfs root directory */ + +/* Query Frame Queue Non Programmable Fields */ + +struct query_fq_np_fields_data_s { + u32 fqid; +}; +static struct query_fq_np_fields_data_s query_fq_np_fields_data = { + .fqid = 1, +}; + +static int query_fq_np_fields_show(struct seq_file *file, void *offset) +{ + int ret; + struct qm_mcr_queryfq_np np; + struct qman_fq fq; + + fq.fqid = query_fq_np_fields_data.fqid; + ret = qman_query_fq_np(&fq, &np); + if (ret) + return ret; + /* Print state */ + seq_printf(file, "Query FQ Non Programmable Fields Result fqid 0x%x\n", + fq.fqid); + seq_printf(file, " force eligible pending: %s\n", + (np.state & QM_MCR_NP_STATE_FE) ? "yes" : "no"); + seq_printf(file, " retirement pending: %s\n", + (np.state & QM_MCR_NP_STATE_R) ? "yes" : "no"); + seq_printf(file, " state: %s\n", + state_txt[np.state & QM_MCR_NP_STATE_MASK]); + seq_printf(file, " fq_link: 0x%x\n", np.fqd_link); + seq_printf(file, " odp_seq: %u\n", np.odp_seq); + seq_printf(file, " orp_nesn: %u\n", np.orp_nesn); + seq_printf(file, " orp_ea_hseq: %u\n", np.orp_ea_hseq); + seq_printf(file, " orp_ea_tseq: %u\n", np.orp_ea_tseq); + seq_printf(file, " orp_ea_hptr: 0x%x\n", np.orp_ea_hptr); + seq_printf(file, " orp_ea_tptr: 0x%x\n", np.orp_ea_tptr); + seq_printf(file, " pfdr_hptr: 0x%x\n", np.pfdr_hptr); + seq_printf(file, " pfdr_tptr: 0x%x\n", np.pfdr_tptr); + seq_printf(file, " is: ics_surp contains a %s\n", + (np.is) ? "deficit" : "surplus"); + seq_printf(file, " ics_surp: %u\n", np.ics_surp); + seq_printf(file, " byte_cnt: %u\n", np.byte_cnt); + seq_printf(file, " frm_cnt: %u\n", np.frm_cnt); + seq_printf(file, " ra1_sfdr: 0x%x\n", np.ra1_sfdr); + seq_printf(file, " ra2_sfdr: 0x%x\n", np.ra2_sfdr); + seq_printf(file, " od1_sfdr: 0x%x\n", np.od1_sfdr); + seq_printf(file, " od2_sfdr: 0x%x\n", np.od2_sfdr); + seq_printf(file, " od3_sfdr: 0x%x\n", np.od3_sfdr); + return 0; +} + +static int query_fq_np_fields_open(struct inode *inode, + struct file *file) +{ + return single_open(file, query_fq_np_fields_show, NULL); +} + +static ssize_t query_fq_np_fields_write(struct file *f, + const char __user *buf, size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + if (val > MAX_FQID) + return -EINVAL; + query_fq_np_fields_data.fqid = (u32)val; + return count; +} + +static const struct file_operations query_fq_np_fields_fops = { + .owner = THIS_MODULE, + .open = query_fq_np_fields_open, + .read = seq_read, + .write = query_fq_np_fields_write, + .release = single_release, +}; + +/* Frame Queue Programmable Fields */ + +struct query_fq_fields_data_s { + u32 fqid; +}; + +static struct query_fq_fields_data_s query_fq_fields_data = { + .fqid = 1, +}; + +static int query_fq_fields_show(struct seq_file *file, void *offset) +{ + int ret; + struct qm_fqd fqd; + struct qman_fq fq; + int i = 0; + + memset(&fqd, 0, sizeof(struct qm_fqd)); + fq.fqid = query_fq_fields_data.fqid; + ret = qman_query_fq(&fq, &fqd); + if (ret) + return ret; + seq_printf(file, "Query FQ Programmable Fields Result fqid 0x%x\n", + fq.fqid); + seq_printf(file, " orprws: %u\n", fqd.orprws); + seq_printf(file, " oa: %u\n", fqd.oa); + seq_printf(file, " olws: %u\n", fqd.olws); + + seq_printf(file, " cgid: %u\n", fqd.cgid); + + if ((fqd.fq_ctrl & QM_FQCTRL_MASK) == 0) + seq_puts(file, " fq_ctrl: None\n"); + else { + i = 0; + seq_puts(file, " fq_ctrl:\n"); + while (fq_ctrl_text_list[i].txt != NULL) { + if ((fqd.fq_ctrl & QM_FQCTRL_MASK) & + fq_ctrl_text_list[i].mask) + seq_printf(file, " %s\n", + fq_ctrl_text_list[i].txt); + i++; + } + } + seq_printf(file, " dest_channel: %u\n", fqd.dest.channel); + seq_printf(file, " dest_wq: %u\n", fqd.dest.wq); + seq_printf(file, " ics_cred: %u\n", fqd.ics_cred); + seq_printf(file, " td_mant: %u\n", fqd.td.mant); + seq_printf(file, " td_exp: %u\n", fqd.td.exp); + + seq_printf(file, " ctx_b: 0x%x\n", fqd.context_b); + + seq_printf(file, " ctx_a: 0x%llx\n", qm_fqd_stashing_get64(&fqd)); + /* Any stashing configured */ + if ((fqd.context_a.stashing.exclusive & 0x7) == 0) + seq_puts(file, " ctx_a_stash_exclusive: None\n"); + else { + seq_puts(file, " ctx_a_stash_exclusive:\n"); + i = 0; + while (stashing_text_list[i].txt != NULL) { + if ((fqd.fq_ctrl & 0x7) & stashing_text_list[i].mask) + seq_printf(file, " %s\n", + stashing_text_list[i].txt); + i++; + } + } + seq_printf(file, " ctx_a_stash_annotation_cl: %u\n", + fqd.context_a.stashing.annotation_cl); + seq_printf(file, " ctx_a_stash_data_cl: %u\n", + fqd.context_a.stashing.data_cl); + seq_printf(file, " ctx_a_stash_context_cl: %u\n", + fqd.context_a.stashing.context_cl); + return 0; +} + +static int query_fq_fields_open(struct inode *inode, + struct file *file) +{ + return single_open(file, query_fq_fields_show, NULL); +} + +static ssize_t query_fq_fields_write(struct file *f, + const char __user *buf, size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + if (val > MAX_FQID) + return -EINVAL; + query_fq_fields_data.fqid = (u32)val; + return count; +} + +static const struct file_operations query_fq_fields_fops = { + .owner = THIS_MODULE, + .open = query_fq_fields_open, + .read = seq_read, + .write = query_fq_fields_write, + .release = single_release, +}; + +/* Query WQ lengths */ + +struct query_wq_lengths_data_s { + union { + u16 channel_wq; /* ignores wq (3 lsbits) */ + struct { + u16 id:13; /* qm_channel */ + u16 __reserved:3; + } __packed channel; + }; +}; +static struct query_wq_lengths_data_s query_wq_lengths_data; +static int query_wq_lengths_show(struct seq_file *file, void *offset) +{ + int ret; + struct qm_mcr_querywq wq; + int i; + + memset(&wq, 0, sizeof(struct qm_mcr_querywq)); + wq.channel.id = query_wq_lengths_data.channel.id; + ret = qman_query_wq(0, &wq); + if (ret) + return ret; + seq_printf(file, "Query Result For Channel: 0x%x\n", wq.channel.id); + for (i = 0; i < 8; i++) + /* mask out upper 4 bits since they are not part of length */ + seq_printf(file, " wq%d_len : %u\n", i, wq.wq_len[i] & 0x0fff); + return 0; +} + +static int query_wq_lengths_open(struct inode *inode, + struct file *file) +{ + return single_open(file, query_wq_lengths_show, NULL); +} + +static ssize_t query_wq_lengths_write(struct file *f, + const char __user *buf, size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + if (val > 0xfff8) + return -EINVAL; + query_wq_lengths_data.channel.id = (u16)val; + return count; +} + +static const struct file_operations query_wq_lengths_fops = { + .owner = THIS_MODULE, + .open = query_wq_lengths_open, + .read = seq_read, + .write = query_wq_lengths_write, + .release = single_release, +}; + +/* Query CGR */ + +struct query_cgr_s { + u8 cgid; +}; +static struct query_cgr_s query_cgr_data; + +static int query_cgr_show(struct seq_file *file, void *offset) +{ + int ret; + struct qm_mcr_querycgr cgrd; + struct qman_cgr cgr; + int i, j; + u32 mask; + + memset(&cgr, 0, sizeof(cgr)); + memset(&cgrd, 0, sizeof(cgrd)); + cgr.cgrid = query_cgr_data.cgid; + ret = qman_query_cgr(&cgr, &cgrd); + if (ret) + return ret; + seq_printf(file, "Query CGR id 0x%x\n", cgr.cgrid); + seq_printf(file, " wr_parm_g MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n", + cgrd.cgr.wr_parm_g.MA, cgrd.cgr.wr_parm_g.Mn, + cgrd.cgr.wr_parm_g.SA, cgrd.cgr.wr_parm_g.Sn, + cgrd.cgr.wr_parm_g.Pn); + + seq_printf(file, " wr_parm_y MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n", + cgrd.cgr.wr_parm_y.MA, cgrd.cgr.wr_parm_y.Mn, + cgrd.cgr.wr_parm_y.SA, cgrd.cgr.wr_parm_y.Sn, + cgrd.cgr.wr_parm_y.Pn); + + seq_printf(file, " wr_parm_r MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n", + cgrd.cgr.wr_parm_r.MA, cgrd.cgr.wr_parm_r.Mn, + cgrd.cgr.wr_parm_r.SA, cgrd.cgr.wr_parm_r.Sn, + cgrd.cgr.wr_parm_r.Pn); + + seq_printf(file, " wr_en_g: %u, wr_en_y: %u, we_en_r: %u\n", + cgrd.cgr.wr_en_g, cgrd.cgr.wr_en_y, cgrd.cgr.wr_en_r); + + seq_printf(file, " cscn_en: %u\n", cgrd.cgr.cscn_en); + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) { + seq_puts(file, " cscn_targ_dcp:\n"); + mask = 0x80000000; + for (i = 0; i < 32; i++) { + if (cgrd.cgr.cscn_targ & mask) + seq_printf(file, " send CSCN to dcp %u\n", + (31 - i)); + mask >>= 1; + } + + seq_puts(file, " cscn_targ_swp:\n"); + for (i = 0; i < 4; i++) { + mask = 0x80000000; + for (j = 0; j < 32; j++) { + if (cgrd.cscn_targ_swp[i] & mask) + seq_printf(file, " send CSCN to swp" + " %u\n", (127 - (i * 32) - j)); + mask >>= 1; + } + } + } else { + seq_printf(file, " cscn_targ: %u\n", cgrd.cgr.cscn_targ); + } + seq_printf(file, " cstd_en: %u\n", cgrd.cgr.cstd_en); + seq_printf(file, " cs: %u\n", cgrd.cgr.cs); + + seq_printf(file, " cs_thresh_TA: %u, cs_thresh_Tn: %u\n", + cgrd.cgr.cs_thres.TA, cgrd.cgr.cs_thres.Tn); + + seq_printf(file, " mode: %s\n", + (cgrd.cgr.mode & QMAN_CGR_MODE_FRAME) ? + "frame count" : "byte count"); + seq_printf(file, " i_bcnt: %llu\n", qm_mcr_querycgr_i_get64(&cgrd)); + seq_printf(file, " a_bcnt: %llu\n", qm_mcr_querycgr_a_get64(&cgrd)); + + return 0; +} + +static int query_cgr_open(struct inode *inode, struct file *file) +{ + return single_open(file, query_cgr_show, NULL); +} + +static ssize_t query_cgr_write(struct file *f, const char __user *buf, + size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + if (val > 0xff) + return -EINVAL; + query_cgr_data.cgid = (u8)val; + return count; +} + +static const struct file_operations query_cgr_fops = { + .owner = THIS_MODULE, + .open = query_cgr_open, + .read = seq_read, + .write = query_cgr_write, + .release = single_release, +}; + +/* Test Write CGR */ + +struct test_write_cgr_s { + u64 i_bcnt; + u8 cgid; +}; +static struct test_write_cgr_s test_write_cgr_data; + +static int testwrite_cgr_show(struct seq_file *file, void *offset) +{ + int ret; + struct qm_mcr_cgrtestwrite result; + struct qman_cgr cgr; + u64 i_bcnt; + + memset(&cgr, 0, sizeof(struct qman_cgr)); + memset(&result, 0, sizeof(struct qm_mcr_cgrtestwrite)); + cgr.cgrid = test_write_cgr_data.cgid; + i_bcnt = test_write_cgr_data.i_bcnt; + ret = qman_testwrite_cgr(&cgr, i_bcnt, &result); + if (ret) + return ret; + seq_printf(file, "CGR Test Write CGR id 0x%x\n", cgr.cgrid); + seq_printf(file, " wr_parm_g MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n", + result.cgr.wr_parm_g.MA, result.cgr.wr_parm_g.Mn, + result.cgr.wr_parm_g.SA, result.cgr.wr_parm_g.Sn, + result.cgr.wr_parm_g.Pn); + seq_printf(file, " wr_parm_y MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n", + result.cgr.wr_parm_y.MA, result.cgr.wr_parm_y.Mn, + result.cgr.wr_parm_y.SA, result.cgr.wr_parm_y.Sn, + result.cgr.wr_parm_y.Pn); + seq_printf(file, " wr_parm_r MA: %u, Mn: %u, SA: %u, Sn: %u, Pn: %u\n", + result.cgr.wr_parm_r.MA, result.cgr.wr_parm_r.Mn, + result.cgr.wr_parm_r.SA, result.cgr.wr_parm_r.Sn, + result.cgr.wr_parm_r.Pn); + seq_printf(file, " wr_en_g: %u, wr_en_y: %u, we_en_r: %u\n", + result.cgr.wr_en_g, result.cgr.wr_en_y, result.cgr.wr_en_r); + seq_printf(file, " cscn_en: %u\n", result.cgr.cscn_en); + seq_printf(file, " cscn_targ: %u\n", result.cgr.cscn_targ); + seq_printf(file, " cstd_en: %u\n", result.cgr.cstd_en); + seq_printf(file, " cs: %u\n", result.cgr.cs); + seq_printf(file, " cs_thresh_TA: %u, cs_thresh_Tn: %u\n", + result.cgr.cs_thres.TA, result.cgr.cs_thres.Tn); + + /* Add Mode for Si 2 */ + seq_printf(file, " mode: %s\n", + (result.cgr.mode & QMAN_CGR_MODE_FRAME) ? + "frame count" : "byte count"); + + seq_printf(file, " i_bcnt: %llu\n", + qm_mcr_cgrtestwrite_i_get64(&result)); + seq_printf(file, " a_bcnt: %llu\n", + qm_mcr_cgrtestwrite_a_get64(&result)); + seq_printf(file, " wr_prob_g: %u\n", result.wr_prob_g); + seq_printf(file, " wr_prob_y: %u\n", result.wr_prob_y); + seq_printf(file, " wr_prob_r: %u\n", result.wr_prob_r); + return 0; +} + +static int testwrite_cgr_open(struct inode *inode, struct file *file) +{ + return single_open(file, testwrite_cgr_show, NULL); +} + +static const struct file_operations testwrite_cgr_fops = { + .owner = THIS_MODULE, + .open = testwrite_cgr_open, + .read = seq_read, + .release = single_release, +}; + + +static int testwrite_cgr_ibcnt_show(struct seq_file *file, void *offset) +{ + seq_printf(file, "i_bcnt: %llu\n", test_write_cgr_data.i_bcnt); + return 0; +} +static int testwrite_cgr_ibcnt_open(struct inode *inode, struct file *file) +{ + return single_open(file, testwrite_cgr_ibcnt_show, NULL); +} + +static ssize_t testwrite_cgr_ibcnt_write(struct file *f, const char __user *buf, + size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + test_write_cgr_data.i_bcnt = val; + return count; +} + +static const struct file_operations teswrite_cgr_ibcnt_fops = { + .owner = THIS_MODULE, + .open = testwrite_cgr_ibcnt_open, + .read = seq_read, + .write = testwrite_cgr_ibcnt_write, + .release = single_release, +}; + +static int testwrite_cgr_cgrid_show(struct seq_file *file, void *offset) +{ + seq_printf(file, "cgrid: %u\n", (u32)test_write_cgr_data.cgid); + return 0; +} +static int testwrite_cgr_cgrid_open(struct inode *inode, struct file *file) +{ + return single_open(file, testwrite_cgr_cgrid_show, NULL); +} + +static ssize_t testwrite_cgr_cgrid_write(struct file *f, const char __user *buf, + size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + if (val > 0xff) + return -EINVAL; + test_write_cgr_data.cgid = (u8)val; + return count; +} + +static const struct file_operations teswrite_cgr_cgrid_fops = { + .owner = THIS_MODULE, + .open = testwrite_cgr_cgrid_open, + .read = seq_read, + .write = testwrite_cgr_cgrid_write, + .release = single_release, +}; + +/* Query Congestion State */ + +static int query_congestion_show(struct seq_file *file, void *offset) +{ + int ret; + struct qm_mcr_querycongestion cs; + int i, j, in_cong = 0; + u32 mask; + + memset(&cs, 0, sizeof(struct qm_mcr_querycongestion)); + ret = qman_query_congestion(&cs); + if (ret) + return ret; + seq_puts(file, "Query Congestion Result\n"); + for (i = 0; i < 8; i++) { + mask = 0x80000000; + for (j = 0; j < 32; j++) { + if (cs.state.__state[i] & mask) { + in_cong = 1; + seq_printf(file, " cg %u: %s\n", (i*32)+j, + "in congestion"); + } + mask >>= 1; + } + } + if (!in_cong) + seq_puts(file, " All congestion groups not congested.\n"); + return 0; +} + +static int query_congestion_open(struct inode *inode, struct file *file) +{ + return single_open(file, query_congestion_show, NULL); +} + +static const struct file_operations query_congestion_fops = { + .owner = THIS_MODULE, + .open = query_congestion_open, + .read = seq_read, + .release = single_release, +}; + +/* QMan register */ + +struct qman_register_s { + u32 val; +}; +static struct qman_register_s qman_register_data; + +static void init_ccsrmempeek(void) +{ + struct device_node *dn; + const u32 *regaddr_p; + + dn = of_find_compatible_node(NULL, NULL, "fsl,qman"); + if (!dn) { + pr_info("No fsl,qman node\n"); + return; + } + regaddr_p = of_get_address(dn, 0, &qman_ccsr_size, NULL); + if (!regaddr_p) { + of_node_put(dn); + return; + } + qman_ccsr_start = of_translate_address(dn, regaddr_p); + of_node_put(dn); +} +/* This function provides access to QMan ccsr memory map */ +static int qman_ccsrmempeek(u32 *val, u32 offset) +{ + void __iomem *addr; + u64 phys_addr; + + if (!qman_ccsr_start) + return -EINVAL; + + if (offset > qman_ccsr_size - sizeof(u32)) + return -EINVAL; + + phys_addr = qman_ccsr_start + offset; + addr = ioremap(phys_addr, sizeof(u32)); + if (!addr) { + pr_err("%s(): ioremap() failed\n", __func__); + return -EINVAL; + } + *val = in_be32(addr); + iounmap(addr); + return 0; +} + +static int qman_ccsrmempeek_show(struct seq_file *file, void *offset) +{ + u32 b; + + qman_ccsrmempeek(&b, qman_register_data.val); + seq_printf(file, "QMan register offset = 0x%x\n", + qman_register_data.val); + seq_printf(file, "value = 0x%08x\n", b); + + return 0; +} + +static int qman_ccsrmempeek_open(struct inode *inode, struct file *file) +{ + return single_open(file, qman_ccsrmempeek_show, NULL); +} + +static ssize_t qman_ccsrmempeek_write(struct file *f, const char __user *buf, + size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + /* multiple of 4 */ + if (val > qman_ccsr_size - sizeof(u32)) { + pr_info("Input 0x%lx > 0x%llx\n", + val, qman_ccsr_size - sizeof(u32)); + return -EINVAL; + } + if (val & 0x3) { + pr_info("Input 0x%lx not multiple of 4\n", val); + return -EINVAL; + } + qman_register_data.val = val; + return count; +} + +static const struct file_operations qman_ccsrmempeek_fops = { + .owner = THIS_MODULE, + .open = qman_ccsrmempeek_open, + .read = seq_read, + .write = qman_ccsrmempeek_write, +}; + +/* QMan state */ + +static int qman_fqd_state_show(struct seq_file *file, void *offset) +{ + struct qm_mcr_queryfq_np np; + struct qman_fq fq; + struct line_buffer_fq line_buf; + int ret, i; + u8 *state = file->private; + u32 qm_fq_state_cnt[ARRAY_SIZE(fqd_states)]; + + memset(qm_fq_state_cnt, 0, sizeof(qm_fq_state_cnt)); + memset(&line_buf, 0, sizeof(line_buf)); + + seq_printf(file, "List of fq ids in state: %s\n", state_txt[*state]); + + for (i = 1; i < fqid_max; i++) { + fq.fqid = i; + ret = qman_query_fq_np(&fq, &np); + if (ret) + return ret; + if (*state == (np.state & QM_MCR_NP_STATE_MASK)) + add_to_line_buffer(&line_buf, fq.fqid, file); + /* Keep a summary count of all states */ + if ((np.state & QM_MCR_NP_STATE_MASK) < ARRAY_SIZE(fqd_states)) + qm_fq_state_cnt[(np.state & QM_MCR_NP_STATE_MASK)]++; + } + flush_line_buffer(&line_buf, file); + + for (i = 0; i < ARRAY_SIZE(fqd_states); i++) { + seq_printf(file, "%s count = %u\n", state_txt[i], + qm_fq_state_cnt[i]); + } + return 0; +} + +static int qman_fqd_state_open(struct inode *inode, struct file *file) +{ + return single_open(file, qman_fqd_state_show, inode->i_private); +} + +static const struct file_operations qman_fqd_state_fops = { + .owner = THIS_MODULE, + .open = qman_fqd_state_open, + .read = seq_read, +}; + +static int qman_fqd_ctrl_show(struct seq_file *file, void *offset) +{ + struct qm_fqd fqd; + struct qman_fq fq; + u32 fq_en_cnt = 0, fq_di_cnt = 0; + int ret, i; + struct mask_filter_s *data = file->private; + const char *ctrl_txt = get_fqd_ctrl_text(data->mask); + struct line_buffer_fq line_buf; + + memset(&line_buf, 0, sizeof(line_buf)); + seq_printf(file, "List of fq ids with: %s :%s\n", + ctrl_txt, (data->filter) ? "enabled" : "disabled"); + for (i = 1; i < fqid_max; i++) { + fq.fqid = i; + memset(&fqd, 0, sizeof(struct qm_fqd)); + ret = qman_query_fq(&fq, &fqd); + if (ret) + return ret; + if (data->filter) { + if (fqd.fq_ctrl & data->mask) + add_to_line_buffer(&line_buf, fq.fqid, file); + } else { + if (!(fqd.fq_ctrl & data->mask)) + add_to_line_buffer(&line_buf, fq.fqid, file); + } + if (fqd.fq_ctrl & data->mask) + fq_en_cnt++; + else + fq_di_cnt++; + } + flush_line_buffer(&line_buf, file); + + seq_printf(file, "Total FQD with: %s : enabled = %u\n", + ctrl_txt, fq_en_cnt); + seq_printf(file, "Total FQD with: %s : disabled = %u\n", + ctrl_txt, fq_di_cnt); + return 0; +} + +/* QMan ctrl CGE, TDE, ORP, CTX, CPC, SFDR, BLOCK, HOLD, CACHE */ +static int qman_fqd_ctrl_open(struct inode *inode, struct file *file) +{ + return single_open(file, qman_fqd_ctrl_show, inode->i_private); +} + +static const struct file_operations qman_fqd_ctrl_fops = { + .owner = THIS_MODULE, + .open = qman_fqd_ctrl_open, + .read = seq_read, +}; + +/* QMan ctrl summary */ + +/* QMan summary state */ + +static int qman_fqd_non_prog_summary_show(struct seq_file *file, void *offset) +{ + struct qm_mcr_queryfq_np np; + struct qman_fq fq; + int ret, i; + u32 qm_fq_state_cnt[ARRAY_SIZE(fqd_states)]; + + memset(qm_fq_state_cnt, 0, sizeof(qm_fq_state_cnt)); + + for (i = 1; i < fqid_max; i++) { + fq.fqid = i; + ret = qman_query_fq_np(&fq, &np); + if (ret) + return ret; + /* Keep a summary count of all states */ + if ((np.state & QM_MCR_NP_STATE_MASK) < ARRAY_SIZE(fqd_states)) + qm_fq_state_cnt[(np.state & QM_MCR_NP_STATE_MASK)]++; + } + + for (i = 0; i < ARRAY_SIZE(fqd_states); i++) { + seq_printf(file, "%s count = %u\n", state_txt[i], + qm_fq_state_cnt[i]); + } + return 0; +} + +static int qman_fqd_prog_summary_show(struct seq_file *file, void *offset) +{ + struct qm_fqd fqd; + struct qman_fq fq; + int ret, i, j; + u32 qm_prog_cnt[ARRAY_SIZE(mask_filter)/2]; + + memset(qm_prog_cnt, 0, sizeof(qm_prog_cnt)); + + for (i = 1; i < fqid_max; i++) { + memset(&fqd, 0, sizeof(struct qm_fqd)); + fq.fqid = i; + ret = qman_query_fq(&fq, &fqd); + if (ret) + return ret; + /* Keep a summary count of all states */ + for (j = 0; j < ARRAY_SIZE(mask_filter); j += 2) + if ((fqd.fq_ctrl & QM_FQCTRL_MASK) & + mask_filter[j].mask) + qm_prog_cnt[j/2]++; + } + for (i = 0; i < ARRAY_SIZE(mask_filter) / 2; i++) { + seq_printf(file, "%s count = %u\n", + get_fqd_ctrl_text(mask_filter[i*2].mask), + qm_prog_cnt[i]); + } + return 0; +} + +static int qman_fqd_summary_show(struct seq_file *file, void *offset) +{ + int ret; + + /* Display summary of non programmable fields */ + ret = qman_fqd_non_prog_summary_show(file, offset); + if (ret) + return ret; + seq_puts(file, "-----------------------------------------\n"); + /* Display programmable fields */ + ret = qman_fqd_prog_summary_show(file, offset); + return ret; +} + +static int qman_fqd_summary_open(struct inode *inode, struct file *file) +{ + return single_open(file, qman_fqd_summary_show, NULL); +} + +static const struct file_operations qman_fqd_summary_fops = { + .owner = THIS_MODULE, + .open = qman_fqd_summary_open, + .read = seq_read, +}; + +/* QMan destination work queue */ + +struct qman_dest_wq_s { + u16 wq_id; +}; +static struct qman_dest_wq_s qman_dest_wq_data = { + .wq_id = 0, +}; + +static int qman_fqd_dest_wq_show(struct seq_file *file, void *offset) +{ + struct qm_fqd fqd; + struct qman_fq fq; + int ret, i; + u16 *wq, wq_id = qman_dest_wq_data.wq_id; + struct line_buffer_fq line_buf; + + memset(&line_buf, 0, sizeof(line_buf)); + /* use vmalloc : need to allocate large memory region and don't + * require the memory to be physically contiguous. */ + wq = vzalloc(sizeof(u16) * (0xFFFF+1)); + if (!wq) + return -ENOMEM; + + seq_printf(file, "List of fq ids with destination work queue id" + " = 0x%x\n", wq_id); + + for (i = 1; i < fqid_max; i++) { + fq.fqid = i; + memset(&fqd, 0, sizeof(struct qm_fqd)); + ret = qman_query_fq(&fq, &fqd); + if (ret) { + vfree(wq); + return ret; + } + if (wq_id == fqd.dest_wq) + add_to_line_buffer(&line_buf, fq.fqid, file); + wq[fqd.dest_wq]++; + } + flush_line_buffer(&line_buf, file); + + seq_puts(file, "Summary of all FQD destination work queue values\n"); + for (i = 0; i < 0xFFFF; i++) { + if (wq[i]) + seq_printf(file, "Channel: 0x%x WQ: 0x%x WQ_ID: 0x%x, " + "count = %u\n", i >> 3, i & 0x3, i, wq[i]); + } + vfree(wq); + return 0; +} + +static ssize_t qman_fqd_dest_wq_write(struct file *f, const char __user *buf, + size_t count, loff_t *off) +{ + int ret; + unsigned long val; + + ret = user_input_convert(buf, count, &val); + if (ret) + return ret; + if (val > 0xFFFF) + return -EINVAL; + qman_dest_wq_data.wq_id = val; + return count; +} + +static int qman_fqd_dest_wq_open(struct inode *inode, struct file *file) +{ + return single_open(file, qman_fqd_dest_wq_show, NULL); +} + +static const struct file_operations qman_fqd_dest_wq_fops = { + .owner = THIS_MODULE, + .open = qman_fqd_dest_wq_open, + .read = seq_read, + .write = qman_fqd_dest_wq_write, +}; + +/* QMan Intra-Class Scheduling Credit */ + +static int qman_fqd_cred_show(struct seq_file *file, void *offset) +{ + struct qm_fqd fqd; + struct qman_fq fq; + int ret, i; + u32 fq_cnt = 0; + struct line_buffer_fq line_buf; + + memset(&line_buf, 0, sizeof(line_buf)); + seq_puts(file, "List of fq ids with Intra-Class Scheduling Credit > 0" + "\n"); + + for (i = 1; i < fqid_max; i++) { + fq.fqid = i; + memset(&fqd, 0, sizeof(struct qm_fqd)); + ret = qman_query_fq(&fq, &fqd); + if (ret) + return ret; + if (fqd.ics_cred > 0) { + add_to_line_buffer(&line_buf, fq.fqid, file); + fq_cnt++; + } + } + flush_line_buffer(&line_buf, file); + + seq_printf(file, "Total FQD with ics_cred > 0 = %d\n", fq_cnt); + return 0; +} + +static int qman_fqd_cred_open(struct inode *inode, struct file *file) +{ + return single_open(file, qman_fqd_cred_show, NULL); +} + +static const struct file_operations qman_fqd_cred_fops = { + .owner = THIS_MODULE, + .open = qman_fqd_cred_open, + .read = seq_read, +}; + +/* helper macros used in qman_debugfs_init */ +#define QMAN_DBGFS_ENTRY(name, mode, parent, data, fops) \ + do { \ + d = debugfs_create_file(name, \ + mode, parent, \ + data, \ + fops); \ + if (d == NULL) { \ + ret = -ENOMEM; \ + goto _return; \ + } \ + } while (0) + +/* dfs_root as parent */ +#define QMAN_DBGFS_ENTRY_ROOT(name, mode, data, fops) \ + QMAN_DBGFS_ENTRY(name, mode, dfs_root, data, fops) + +/* fqd_root as parent */ +#define QMAN_DBGFS_ENTRY_FQDROOT(name, mode, data, fops) \ + QMAN_DBGFS_ENTRY(name, mode, fqd_root, data, fops) + +/* fqd state */ +#define QMAN_DBGFS_ENTRY_FQDSTATE(name, index) \ + QMAN_DBGFS_ENTRY_FQDROOT(name, S_IRUGO, \ + (void *)&mask_filter[index], &qman_fqd_ctrl_fops) + +static int __init qman_debugfs_init(void) +{ + int ret = 0; + struct dentry *d, *fqd_root; + u32 reg; + + fqid_max = 0; + init_ccsrmempeek(); + if (qman_ccsr_start) { + if (!qman_ccsrmempeek(®, QM_FQD_AR)) { + /* extract the size of the FQD window */ + reg = reg & 0x3f; + /* calculate valid frame queue descriptor range */ + fqid_max = (1 << (reg + 1)) / QM_FQD_BLOCK_SIZE; + } + } + + dfs_root = debugfs_create_dir("qman", NULL); + fqd_root = debugfs_create_dir("fqd", dfs_root); + if (dfs_root == NULL || fqd_root == NULL) { + pr_err("Cannot create dir\n"); + return -ENOMEM; + } + + if (fqid_max) { + QMAN_DBGFS_ENTRY_ROOT("ccsrmempeek", S_IRUGO | S_IWUGO, + NULL, &qman_ccsrmempeek_fops); + } + QMAN_DBGFS_ENTRY_ROOT("query_fq_np_fields", S_IRUGO | S_IWUGO, + &query_fq_np_fields_data, &query_fq_np_fields_fops); + + QMAN_DBGFS_ENTRY_ROOT("query_fq_fields", S_IRUGO | S_IWUGO, + &query_fq_fields_data, &query_fq_fields_fops); + + QMAN_DBGFS_ENTRY_ROOT("query_wq_lengths", S_IRUGO | S_IWUGO, + &query_wq_lengths_data, &query_wq_lengths_fops); + + QMAN_DBGFS_ENTRY_ROOT("query_cgr", S_IRUGO | S_IWUGO, + &query_cgr_data, &query_cgr_fops); + + QMAN_DBGFS_ENTRY_ROOT("query_congestion", S_IRUGO, + NULL, &query_congestion_fops); + + QMAN_DBGFS_ENTRY_ROOT("testwrite_cgr", S_IRUGO, + NULL, &testwrite_cgr_fops); + + QMAN_DBGFS_ENTRY_ROOT("testwrite_cgr_cgrid", S_IRUGO | S_IWUGO, + NULL, &teswrite_cgr_cgrid_fops); + + QMAN_DBGFS_ENTRY_ROOT("testwrite_cgr_ibcnt", S_IRUGO | S_IWUGO, + NULL, &teswrite_cgr_ibcnt_fops); + + /* Create files with fqd_root as parent */ + + QMAN_DBGFS_ENTRY_FQDROOT("stateoos", S_IRUGO, + (void *)&fqd_states[QM_MCR_NP_STATE_OOS], &qman_fqd_state_fops); + + QMAN_DBGFS_ENTRY_FQDROOT("state_retired", S_IRUGO, + (void *)&fqd_states[QM_MCR_NP_STATE_RETIRED], + &qman_fqd_state_fops); + + QMAN_DBGFS_ENTRY_FQDROOT("state_tentatively_sched", S_IRUGO, + (void *)&fqd_states[QM_MCR_NP_STATE_TEN_SCHED], + &qman_fqd_state_fops); + + QMAN_DBGFS_ENTRY_FQDROOT("state_truly_sched", S_IRUGO, + (void *)&fqd_states[QM_MCR_NP_STATE_TRU_SCHED], + &qman_fqd_state_fops); + + QMAN_DBGFS_ENTRY_FQDROOT("state_parked", S_IRUGO, + (void *)&fqd_states[QM_MCR_NP_STATE_PARKED], + &qman_fqd_state_fops); + + QMAN_DBGFS_ENTRY_FQDROOT("state_active", S_IRUGO, + (void *)&fqd_states[QM_MCR_NP_STATE_ACTIVE], + &qman_fqd_state_fops); + + QMAN_DBGFS_ENTRY_FQDSTATE("cge_enable", 17); + + QMAN_DBGFS_ENTRY_FQDSTATE("cge_disable", 16); + + QMAN_DBGFS_ENTRY_FQDSTATE("tde_enable", 15); + + QMAN_DBGFS_ENTRY_FQDSTATE("tde_disable", 14); + + QMAN_DBGFS_ENTRY_FQDSTATE("orp_enable", 13); + + QMAN_DBGFS_ENTRY_FQDSTATE("orp_disable", 12); + + QMAN_DBGFS_ENTRY_FQDSTATE("ctx_a_stashing_enable", 11); + + QMAN_DBGFS_ENTRY_FQDSTATE("ctx_a_stashing_disable", 10); + + QMAN_DBGFS_ENTRY_FQDSTATE("cpc_enable", 9); + + QMAN_DBGFS_ENTRY_FQDSTATE("cpc_disable", 8); + + QMAN_DBGFS_ENTRY_FQDSTATE("sfdr_enable", 7); + + QMAN_DBGFS_ENTRY_FQDSTATE("sfdr_disable", 6); + + QMAN_DBGFS_ENTRY_FQDSTATE("avoid_blocking_enable", 5); + + QMAN_DBGFS_ENTRY_FQDSTATE("avoid_blocking_disable", 4); + + QMAN_DBGFS_ENTRY_FQDSTATE("hold_active_enable", 3); + + QMAN_DBGFS_ENTRY_FQDSTATE("hold_active_disable", 2); + + QMAN_DBGFS_ENTRY_FQDSTATE("prefer_in_cache_enable", 1); + + QMAN_DBGFS_ENTRY_FQDSTATE("prefer_in_cache_disable", 0); + + QMAN_DBGFS_ENTRY_FQDROOT("summary", S_IRUGO, + NULL, &qman_fqd_summary_fops); + + QMAN_DBGFS_ENTRY_FQDROOT("wq", S_IRUGO | S_IWUGO, + NULL, &qman_fqd_dest_wq_fops); + + QMAN_DBGFS_ENTRY_FQDROOT("cred", S_IRUGO, + NULL, &qman_fqd_cred_fops); + + return 0; + +_return: + debugfs_remove_recursive(dfs_root); + + return ret; +} + +static void __exit qman_debugfs_exit(void) +{ + debugfs_remove_recursive(dfs_root); +} + +module_init(qman_debugfs_init); +module_exit(qman_debugfs_exit); + +MODULE_LICENSE("Dual BSD/GPL"); diff --git a/linux/drivers/soc/fsl/qbman/qman.c b/linux/drivers/soc/fsl/qbman/qman.c new file mode 100644 index 00000000..aa188888 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman.c @@ -0,0 +1,1106 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_priv.h" + +#include + +/* Last updated for v00.800 of the BG */ + +/* Register offsets */ +#define REG_QCSP_LIO_CFG(n) (0x0000 + ((n) * 0x10)) +#define REG_QCSP_IO_CFG(n) (0x0004 + ((n) * 0x10)) +#define REG_QCSP_DD_CFG(n) (0x000c + ((n) * 0x10)) +#define REG_DD_CFG 0x0200 +#define REG_DCP_CFG(n) (0x0300 + ((n) * 0x10)) +#define REG_DCP_DD_CFG(n) (0x0304 + ((n) * 0x10)) +#define REG_DCP_DLM_AVG(n) (0x030c + ((n) * 0x10)) +#define REG_PFDR_FPC 0x0400 +#define REG_PFDR_FP_HEAD 0x0404 +#define REG_PFDR_FP_TAIL 0x0408 +#define REG_PFDR_FP_LWIT 0x0410 +#define REG_PFDR_CFG 0x0414 +#define REG_SFDR_CFG 0x0500 +#define REG_SFDR_IN_USE 0x0504 +#define REG_WQ_CS_CFG(n) (0x0600 + ((n) * 0x04)) +#define REG_WQ_DEF_ENC_WQID 0x0630 +#define REG_WQ_SC_DD_CFG(n) (0x640 + ((n) * 0x04)) +#define REG_WQ_PC_DD_CFG(n) (0x680 + ((n) * 0x04)) +#define REG_WQ_DC0_DD_CFG(n) (0x6c0 + ((n) * 0x04)) +#define REG_WQ_DC1_DD_CFG(n) (0x700 + ((n) * 0x04)) +#define REG_WQ_DCn_DD_CFG(n) (0x6c0 + ((n) * 0x40)) /* n=2,3 */ +#define REG_CM_CFG 0x0800 +#define REG_ECSR 0x0a00 +#define REG_ECIR 0x0a04 +#define REG_EADR 0x0a08 +#define REG_ECIR2 0x0a0c +#define REG_EDATA(n) (0x0a10 + ((n) * 0x04)) +#define REG_SBEC(n) (0x0a80 + ((n) * 0x04)) +#define REG_MCR 0x0b00 +#define REG_MCP(n) (0x0b04 + ((n) * 0x04)) +#define REG_MISC_CFG 0x0be0 +#define REG_HID_CFG 0x0bf0 +#define REG_IDLE_STAT 0x0bf4 +#define REG_IP_REV_1 0x0bf8 +#define REG_IP_REV_2 0x0bfc +#define REG_FQD_BARE 0x0c00 +#define REG_PFDR_BARE 0x0c20 +#define REG_offset_BAR 0x0004 /* relative to REG_[FQD|PFDR]_BARE */ +#define REG_offset_AR 0x0010 /* relative to REG_[FQD|PFDR]_BARE */ +#define REG_QCSP_BARE 0x0c80 +#define REG_QCSP_BAR 0x0c84 +#define REG_CI_SCHED_CFG 0x0d00 +#define REG_SRCIDR 0x0d04 +#define REG_LIODNR 0x0d08 +#define REG_CI_RLM_AVG 0x0d14 +#define REG_ERR_ISR 0x0e00 /* + "enum qm_isr_reg" */ +#define REG_REV3_QCSP_LIO_CFG(n) (0x1000 + ((n) * 0x10)) +#define REG_REV3_QCSP_IO_CFG(n) (0x1004 + ((n) * 0x10)) +#define REG_REV3_QCSP_DD_CFG(n) (0x100c + ((n) * 0x10)) + +/* Assists for QMAN_MCR */ +#define MCR_INIT_PFDR 0x01000000 +#define MCR_get_rslt(v) (u8)((v) >> 24) +#define MCR_rslt_idle(r) (!rslt || (rslt >= 0xf0)) +#define MCR_rslt_ok(r) (rslt == 0xf0) +#define MCR_rslt_eaccess(r) (rslt == 0xf8) +#define MCR_rslt_inval(r) (rslt == 0xff) + +/* Corenet initiator settings. Stash request queues are 4-deep to match cores + ability to snarf. Stash priority is 3, other priorities are 2. */ +#define FSL_QMAN_CI_SCHED_CFG_SRCCIV 4 +#define FSL_QMAN_CI_SCHED_CFG_SRQ_W 3 +#define FSL_QMAN_CI_SCHED_CFG_RW_W 2 +#define FSL_QMAN_CI_SCHED_CFG_BMAN_W 2 + +struct qman; + +/* Follows WQ_CS_CFG0-5 */ +enum qm_wq_class { + qm_wq_portal = 0, + qm_wq_pool = 1, + qm_wq_fman0 = 2, + qm_wq_fman1 = 3, + qm_wq_caam = 4, + qm_wq_pme = 5, + qm_wq_first = qm_wq_portal, + qm_wq_last = qm_wq_pme +}; + +/* Follows FQD_[BARE|BAR|AR] and PFDR_[BARE|BAR|AR] */ +enum qm_memory { + qm_memory_fqd, + qm_memory_pfdr +}; + +/* Used by all error interrupt registers except 'inhibit' */ +#define QM_EIRQ_CIDE 0x20000000 /* Corenet Initiator Data Error */ +#define QM_EIRQ_CTDE 0x10000000 /* Corenet Target Data Error */ +#define QM_EIRQ_CITT 0x08000000 /* Corenet Invalid Target Transaction */ +#define QM_EIRQ_PLWI 0x04000000 /* PFDR Low Watermark */ +#define QM_EIRQ_MBEI 0x02000000 /* Multi-bit ECC Error */ +#define QM_EIRQ_SBEI 0x01000000 /* Single-bit ECC Error */ +#define QM_EIRQ_PEBI 0x00800000 /* PFDR Enqueues Blocked Interrupt */ +#define QM_EIRQ_IFSI 0x00020000 /* Invalid FQ Flow Control State */ +#define QM_EIRQ_ICVI 0x00010000 /* Invalid Command Verb */ +#define QM_EIRQ_IDDI 0x00000800 /* Invalid Dequeue (Direct-connect) */ +#define QM_EIRQ_IDFI 0x00000400 /* Invalid Dequeue FQ */ +#define QM_EIRQ_IDSI 0x00000200 /* Invalid Dequeue Source */ +#define QM_EIRQ_IDQI 0x00000100 /* Invalid Dequeue Queue */ +#define QM_EIRQ_IECE 0x00000010 /* Invalid Enqueue Configuration */ +#define QM_EIRQ_IEOI 0x00000008 /* Invalid Enqueue Overflow */ +#define QM_EIRQ_IESI 0x00000004 /* Invalid Enqueue State */ +#define QM_EIRQ_IECI 0x00000002 /* Invalid Enqueue Channel */ +#define QM_EIRQ_IEQI 0x00000001 /* Invalid Enqueue Queue */ + +/* QMAN_ECIR valid error bit */ +#define PORTAL_ECSR_ERR (QM_EIRQ_IEQI | QM_EIRQ_IESI | QM_EIRQ_IEOI | \ + QM_EIRQ_IDQI | QM_EIRQ_IDSI | QM_EIRQ_IDFI | \ + QM_EIRQ_IDDI | QM_EIRQ_ICVI | QM_EIRQ_IFSI) +#define FQID_ECSR_ERR (QM_EIRQ_IEQI | QM_EIRQ_IECI | QM_EIRQ_IESI | \ + QM_EIRQ_IEOI | QM_EIRQ_IDQI | QM_EIRQ_IDFI | \ + QM_EIRQ_IFSI) + +union qman_ecir { + u32 ecir_raw; + struct { + u32 __reserved:2; + u32 portal_type:1; + u32 portal_num:5; + u32 fqid:24; + } __packed info; +}; + +union qman_ecir2 { + u32 ecir2_raw; + struct { + u32 portal_type:1; + u32 __reserved:21; + u32 portal_num:10; + } __packed info; +}; + +union qman_eadr { + u32 eadr_raw; + struct { + u32 __reserved1:4; + u32 memid:4; + u32 __reserved2:12; + u32 eadr:12; + } __packed info; + struct { + u32 __reserved1:3; + u32 memid:5; + u32 __reserved:8; + u32 eadr:16; + } __packed info_rev3; +}; + +struct qman_hwerr_txt { + u32 mask; + const char *txt; +}; + +#define QMAN_HWE_TXT(a, b) { .mask = QM_EIRQ_##a, .txt = b } + +static const struct qman_hwerr_txt qman_hwerr_txts[] = { + QMAN_HWE_TXT(CIDE, "Corenet Initiator Data Error"), + QMAN_HWE_TXT(CTDE, "Corenet Target Data Error"), + QMAN_HWE_TXT(CITT, "Corenet Invalid Target Transaction"), + QMAN_HWE_TXT(PLWI, "PFDR Low Watermark"), + QMAN_HWE_TXT(MBEI, "Multi-bit ECC Error"), + QMAN_HWE_TXT(SBEI, "Single-bit ECC Error"), + QMAN_HWE_TXT(PEBI, "PFDR Enqueues Blocked Interrupt"), + QMAN_HWE_TXT(ICVI, "Invalid Command Verb"), + QMAN_HWE_TXT(IFSI, "Invalid Flow Control State"), + QMAN_HWE_TXT(IDDI, "Invalid Dequeue (Direct-connect)"), + QMAN_HWE_TXT(IDFI, "Invalid Dequeue FQ"), + QMAN_HWE_TXT(IDSI, "Invalid Dequeue Source"), + QMAN_HWE_TXT(IDQI, "Invalid Dequeue Queue"), + QMAN_HWE_TXT(IECE, "Invalid Enqueue Configuration"), + QMAN_HWE_TXT(IEOI, "Invalid Enqueue Overflow"), + QMAN_HWE_TXT(IESI, "Invalid Enqueue State"), + QMAN_HWE_TXT(IECI, "Invalid Enqueue Channel"), + QMAN_HWE_TXT(IEQI, "Invalid Enqueue Queue") +}; +#define QMAN_HWE_COUNT (sizeof(qman_hwerr_txts)/sizeof(struct qman_hwerr_txt)) + +struct qman_error_info_mdata { + u16 addr_mask; + u16 bits; + const char *txt; +}; + +#define QMAN_ERR_MDATA(a, b, c) { .addr_mask = a, .bits = b, .txt = c} +static const struct qman_error_info_mdata error_mdata[] = { + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 0"), + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 1"), + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 2"), + QMAN_ERR_MDATA(0x01FF, 24, "FQD cache tag memory 3"), + QMAN_ERR_MDATA(0x0FFF, 512, "FQD cache memory"), + QMAN_ERR_MDATA(0x07FF, 128, "SFDR memory"), + QMAN_ERR_MDATA(0x01FF, 72, "WQ context memory"), + QMAN_ERR_MDATA(0x00FF, 240, "CGR memory"), + QMAN_ERR_MDATA(0x00FF, 302, "Internal Order Restoration List memory"), + QMAN_ERR_MDATA(0x01FF, 256, "SW portal ring memory"), +}; +#define QMAN_ERR_MDATA_COUNT \ + (sizeof(error_mdata)/sizeof(struct qman_error_info_mdata)) + +/* Add this in Kconfig */ +#define QMAN_ERRS_TO_UNENABLE (QM_EIRQ_PLWI | QM_EIRQ_PEBI) + +/** + * qm_err_isr__ - Manipulate global interrupt registers + * @v: for accessors that write values, this is the 32-bit value + * + * Manipulates QMAN_ERR_ISR, QMAN_ERR_IER, QMAN_ERR_ISDR, QMAN_ERR_IIR. All + * manipulations except qm_err_isr_[un]inhibit() use 32-bit masks composed of + * the QM_EIRQ_*** definitions. Note that "qm_err_isr_enable_write" means + * "write the enable register" rather than "enable the write register"! + */ +#define qm_err_isr_status_read(qm) \ + __qm_err_isr_read(qm, qm_isr_status) +#define qm_err_isr_status_clear(qm, m) \ + __qm_err_isr_write(qm, qm_isr_status, m) +#define qm_err_isr_enable_read(qm) \ + __qm_err_isr_read(qm, qm_isr_enable) +#define qm_err_isr_enable_write(qm, v) \ + __qm_err_isr_write(qm, qm_isr_enable, v) +#define qm_err_isr_disable_read(qm) \ + __qm_err_isr_read(qm, qm_isr_disable) +#define qm_err_isr_disable_write(qm, v) \ + __qm_err_isr_write(qm, qm_isr_disable, v) +#define qm_err_isr_inhibit(qm) \ + __qm_err_isr_write(qm, qm_isr_inhibit, 1) +#define qm_err_isr_uninhibit(qm) \ + __qm_err_isr_write(qm, qm_isr_inhibit, 0) + +/* + * TODO: unimplemented registers + * + * Keeping a list here of QMan registers I have not yet covered; + * QCSP_DD_IHRSR, QCSP_DD_IHRFR, QCSP_DD_HASR, + * DCP_DD_IHRSR, DCP_DD_IHRFR, DCP_DD_HASR, CM_CFG, + * QMAN_EECC, QMAN_SBET, QMAN_EINJ, QMAN_SBEC0-12 + */ + +/* Encapsulate "struct qman *" as a cast of the register space address. */ + +static struct qman *qm_create(void *regs) +{ + return (struct qman *)regs; +} + +static inline u32 __qm_in(struct qman *qm, u32 offset) +{ + return ioread32be((void *)qm + offset); +} +static inline void __qm_out(struct qman *qm, u32 offset, u32 val) +{ + iowrite32be(val, (void *)qm + offset); +} +#define qm_in(reg) __qm_in(qm, REG_##reg) +#define qm_out(reg, val) __qm_out(qm, REG_##reg, val) + +static u32 __qm_err_isr_read(struct qman *qm, enum qm_isr_reg n) +{ + return __qm_in(qm, REG_ERR_ISR + (n << 2)); +} + +static void __qm_err_isr_write(struct qman *qm, enum qm_isr_reg n, u32 val) +{ + __qm_out(qm, REG_ERR_ISR + (n << 2), val); +} + +static void qm_set_dc(struct qman *qm, enum qm_dc_portal portal, + int ed, u8 sernd) +{ + DPA_ASSERT(!ed || (portal == qm_dc_portal_fman0) || + (portal == qm_dc_portal_fman1)); + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + qm_out(DCP_CFG(portal), (ed ? 0x1000 : 0) | (sernd & 0x3ff)); + else + qm_out(DCP_CFG(portal), (ed ? 0x100 : 0) | (sernd & 0x1f)); +} + +static void qm_set_wq_scheduling(struct qman *qm, enum qm_wq_class wq_class, + u8 cs_elev, u8 csw2, u8 csw3, u8 csw4, u8 csw5, + u8 csw6, u8 csw7) +{ + qm_out(WQ_CS_CFG(wq_class), ((cs_elev & 0xff) << 24) | + ((csw2 & 0x7) << 20) | ((csw3 & 0x7) << 16) | + ((csw4 & 0x7) << 12) | ((csw5 & 0x7) << 8) | + ((csw6 & 0x7) << 4) | (csw7 & 0x7)); +} + +static void qm_set_hid(struct qman *qm) +{ + qm_out(HID_CFG, 0); +} + +static void qm_set_corenet_initiator(struct qman *qm) +{ + qm_out(CI_SCHED_CFG, + 0x80000000 | /* write srcciv enable */ + (FSL_QMAN_CI_SCHED_CFG_SRCCIV << 24) | + (FSL_QMAN_CI_SCHED_CFG_SRQ_W << 8) | + (FSL_QMAN_CI_SCHED_CFG_RW_W << 4) | + FSL_QMAN_CI_SCHED_CFG_BMAN_W); +} + +static void qm_get_version(struct qman *qm, u16 *id, u8 *major, u8 *minor) +{ + u32 v = qm_in(IP_REV_1); + *id = (v >> 16); + *major = (v >> 8) & 0xff; + *minor = v & 0xff; +} + +static void qm_set_memory(struct qman *qm, enum qm_memory memory, u64 ba, + int enable, int prio, int stash, u32 size) +{ + u32 offset = (memory == qm_memory_fqd) ? REG_FQD_BARE : REG_PFDR_BARE; + u32 exp = ilog2(size); + /* choke if size isn't within range */ + DPA_ASSERT((size >= 4096) && (size <= 1073741824) && + is_power_of_2(size)); + /* choke if 'ba' has lower-alignment than 'size' */ + DPA_ASSERT(!(ba & (size - 1))); + __qm_out(qm, offset, upper_32_bits(ba)); + __qm_out(qm, offset + REG_offset_BAR, lower_32_bits(ba)); + __qm_out(qm, offset + REG_offset_AR, + (enable ? 0x80000000 : 0) | + (prio ? 0x40000000 : 0) | + (stash ? 0x20000000 : 0) | + (exp - 1)); +} + +static void qm_set_pfdr_threshold(struct qman *qm, u32 th, u8 k) +{ + qm_out(PFDR_FP_LWIT, th & 0xffffff); + qm_out(PFDR_CFG, k); +} + +static void qm_set_sfdr_threshold(struct qman *qm, u16 th) +{ + qm_out(SFDR_CFG, th & 0x3ff); +} + +static int qm_init_pfdr(struct qman *qm, u32 pfdr_start, u32 num) +{ + u8 rslt = MCR_get_rslt(qm_in(MCR)); + + DPA_ASSERT(pfdr_start && !(pfdr_start & 7) && !(num & 7) && num); + /* Make sure the command interface is 'idle' */ + if (!MCR_rslt_idle(rslt)) + panic("QMAN_MCR isn't idle"); + + /* Write the MCR command params then the verb */ + qm_out(MCP(0), pfdr_start); + /* TODO: remove this - it's a workaround for a model bug that is + * corrected in more recent versions. We use the workaround until + * everyone has upgraded. */ + qm_out(MCP(1), (pfdr_start + num - 16)); + lwsync(); + qm_out(MCR, MCR_INIT_PFDR); + /* Poll for the result */ + do { + rslt = MCR_get_rslt(qm_in(MCR)); + } while (!MCR_rslt_idle(rslt)); + if (MCR_rslt_ok(rslt)) + return 0; + if (MCR_rslt_eaccess(rslt)) + return -EACCES; + if (MCR_rslt_inval(rslt)) + return -EINVAL; + pr_crit("Unexpected result from MCR_INIT_PFDR: %02x\n", rslt); + return -ENODEV; +} + +/*****************/ +/* Config driver */ +/*****************/ + +/* We support only one of these */ +static struct qman *qm; +#ifndef __rtems__ +static struct device_node *qm_node; +#endif /* __rtems__ */ + +/* And this state belongs to 'qm'. It is set during fsl_qman_init(), but used + * during qman_init_ccsr(). */ +#ifndef __rtems__ +static dma_addr_t fqd_a, pfdr_a; +static size_t fqd_sz, pfdr_sz; + +static int qman_fqd(struct reserved_mem *rmem) +{ + fqd_a = rmem->base; + fqd_sz = rmem->size; + + WARN_ON(!(fqd_a && fqd_sz)); + + return 0; +} +RESERVEDMEM_OF_DECLARE(qman_fqd, "fsl,qman-fqd", qman_fqd); +#else /* __rtems__ */ +static DPAA_NOCACHENOLOAD_ALIGNED_REGION(fqd, 4194304); +#define fqd_a ((uintptr_t)&fqd[0]) +#define fqd_sz sizeof(fqd) +static DPAA_NOCACHENOLOAD_ALIGNED_REGION(pfdr, 33554432); +#define pfdr_a ((uintptr_t)&pfdr[0]) +#define pfdr_sz sizeof(pfdr) +#endif /* __rtems__ */ + +size_t qman_fqd_size(void) +{ + return fqd_sz; +} + +#ifndef __rtems__ +static int qman_pfdr(struct reserved_mem *rmem) +{ + pfdr_a = rmem->base; + pfdr_sz = rmem->size; + + WARN_ON(!(pfdr_a && pfdr_sz)); + + return 0; +} +RESERVEDMEM_OF_DECLARE(qman_fbpr, "fsl,qman-pfdr", qman_pfdr); + +/* Parse the property to extract the memory location and size and + * memblock_reserve() it. If it isn't supplied, memblock_alloc() the default + * size. Also flush this memory range from data cache so that QMAN originated + * transactions for this memory region could be marked non-coherent. + */ +static __init int parse_mem_property(struct device_node *node, + dma_addr_t *addr, size_t *sz, int zero) +{ + int ret; + + /* If using a "zero-pma", don't try to zero it, even if you asked */ + if (zero && of_find_property(node, "zero-pma", &ret)) { + pr_info(" it's a 'zero-pma', not zeroing from s/w\n"); + zero = 0; + } + + if (zero) { + /* map as cacheable, non-guarded */ + void __iomem *tmpp = ioremap_prot(*addr, *sz, 0); + + memset_io(tmpp, 0, *sz); + flush_dcache_range((unsigned long)tmpp, + (unsigned long)tmpp + *sz); + iounmap(tmpp); + } + + return 0; +} +#else /* __rtems__ */ +#include +#endif /* __rtems__ */ + +/* TODO: + * - there is obviously no handling of errors, + * - the calls to qm_set_memory() hard-code the priority and CPC-stashing for + * both memory resources to zero. + */ +static int __init fsl_qman_init(struct device_node *node) +{ +#ifndef __rtems__ + struct resource res; +#endif /* __rtems__ */ + u32 __iomem *regs; +#ifndef __rtems__ + const char *s; + int ret, standby = 0; +#endif /* __rtems__ */ + u16 id; + u8 major, minor; + +#ifndef __rtems__ + ret = of_address_to_resource(node, 0, &res); + if (ret) { + pr_err("Can't get %s property 'reg'\n", node->full_name); + return ret; + } + s = of_get_property(node, "fsl,hv-claimable", &ret); + if (s && !strcmp(s, "standby")) + standby = 1; + if (!standby) { + ret = parse_mem_property(node, &fqd_a, &fqd_sz, 1); + BUG_ON(ret); + ret = parse_mem_property(node, &pfdr_a, &pfdr_sz, 0); + BUG_ON(ret); + } +#else /* __rtems__ */ + memset((void *)fqd_a, 0, fqd_sz); +#endif /* __rtems__ */ + /* Global configuration */ +#ifndef __rtems__ + regs = ioremap(res.start, res.end - res.start + 1); +#else /* __rtems__ */ + regs = (u32 __iomem *)&qoriq.qman; +#endif /* __rtems__ */ + qm = qm_create(regs); +#ifndef __rtems__ + qm_node = node; +#endif /* __rtems__ */ + qm_get_version(qm, &id, &major, &minor); + pr_info("Ver: %04x,%02x,%02x\n", id, major, minor); + if (!qman_ip_rev) { + if ((major == 1) && (minor == 0)) { + pr_err("Rev1.0 on P4080 rev1 is not supported!\n"); +#ifndef __rtems__ + iounmap(regs); +#endif /* __rtems__ */ + return -ENODEV; + } else if ((major == 1) && (minor == 1)) + qman_ip_rev = QMAN_REV11; + else if ((major == 1) && (minor == 2)) + qman_ip_rev = QMAN_REV12; + else if ((major == 2) && (minor == 0)) + qman_ip_rev = QMAN_REV20; + else if ((major == 3) && (minor == 0)) + qman_ip_rev = QMAN_REV30; + else if ((major == 3) && (minor == 1)) + qman_ip_rev = QMAN_REV31; + else { + pr_warn("Unknown version, default to rev1.1\n"); + qman_ip_rev = QMAN_REV11; + } + } + +#ifndef __rtems__ + if (standby) { + pr_info(" -> in standby mode\n"); + return 0; + } +#endif /* __rtems__ */ + return 0; +} + +int qman_have_ccsr(void) +{ + return qm ? 1 : 0; +} + +#ifndef __rtems__ +__init void qman_init_early(void) +{ + struct device_node *dn; + int ret; + + for_each_compatible_node(dn, NULL, "fsl,qman") { + if (qm) + pr_err("%s: only one 'fsl,qman' allowed\n", + dn->full_name); + else { + if (!of_device_is_available(dn)) + continue; + + ret = fsl_qman_init(dn); + BUG_ON(ret); + } + } +} +#else /* __rtems__ */ +#include + +static void +qman_sysinit(void) +{ + const char *fdt = bsp_fdt_get(); + struct device_node dn; + const char *name; + int node; + int ret; + + memset(&dn, 0, sizeof(dn)); + + name = "fsl,qman"; + node = fdt_node_offset_by_compatible(fdt, 0, name); + if (node < 0) + panic("qman: no qman in FDT"); + + dn.full_name = name; + dn.offset = node; + + ret = fsl_qman_init(&dn); + if (ret != 0) + panic("qman: init 1 failed"); + + ret = qman_init_ccsr(&dn); + if (ret != 0) + panic("qman: init CCSR failed"); + + ret = qman_init(&dn); + if (ret != 0) + panic("qman: init 2 failed"); + + ret = qman_resource_init(); + if (ret != 0) + panic("qman: resource init failed"); +} +SYSINIT(qman_sysinit, SI_SUB_CPU, SI_ORDER_SECOND, qman_sysinit, NULL); +#endif /* __rtems__ */ + +static void log_edata_bits(u32 bit_count) +{ + u32 i, j, mask = 0xffffffff; + + pr_warn("ErrInt, EDATA:\n"); + i = bit_count/32; + if (bit_count%32) { + i++; + mask = ~(mask << bit_count%32); + } + j = 16-i; + pr_warn(" 0x%08x\n", qm_in(EDATA(j)) & mask); + j++; + for (; j < 16; j++) + pr_warn(" 0x%08x\n", qm_in(EDATA(j))); +} + +static void log_additional_error_info(u32 isr_val, u32 ecsr_val) +{ + union qman_ecir ecir_val; + union qman_eadr eadr_val; + + ecir_val.ecir_raw = qm_in(ECIR); + /* Is portal info valid */ + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) { + union qman_ecir2 ecir2_val; + + ecir2_val.ecir2_raw = qm_in(ECIR2); + if (ecsr_val & PORTAL_ECSR_ERR) { + pr_warn("ErrInt: %s id %d\n", + ecir2_val.info.portal_type ? "DCP" : "SWP", + ecir2_val.info.portal_num); + } + if (ecsr_val & (FQID_ECSR_ERR | QM_EIRQ_IECE)) + pr_warn("ErrInt: ecir.fqid 0x%x\n", ecir_val.info.fqid); + + if (ecsr_val & (QM_EIRQ_SBEI|QM_EIRQ_MBEI)) { + eadr_val.eadr_raw = qm_in(EADR); + pr_warn("ErrInt: EADR Memory: %s, 0x%x\n", + error_mdata[eadr_val.info_rev3.memid].txt, + error_mdata[eadr_val.info_rev3.memid].addr_mask + & eadr_val.info_rev3.eadr); + log_edata_bits( + error_mdata[eadr_val.info_rev3.memid].bits); + } + } else { + if (ecsr_val & PORTAL_ECSR_ERR) { + pr_warn("ErrInt: %s id %d\n", + ecir_val.info.portal_type ? "DCP" : "SWP", + ecir_val.info.portal_num); + } + if (ecsr_val & FQID_ECSR_ERR) + pr_warn("ErrInt: ecir.fqid 0x%x\n", ecir_val.info.fqid); + + if (ecsr_val & (QM_EIRQ_SBEI|QM_EIRQ_MBEI)) { + eadr_val.eadr_raw = qm_in(EADR); + pr_warn("ErrInt: EADR Memory: %s, 0x%x\n", + error_mdata[eadr_val.info.memid].txt, + error_mdata[eadr_val.info.memid].addr_mask + & eadr_val.info.eadr); + log_edata_bits(error_mdata[eadr_val.info.memid].bits); + } + } +} + +/* QMan interrupt handler */ +static irqreturn_t qman_isr(int irq, void *ptr) +{ + u32 isr_val, ier_val, ecsr_val, isr_mask, i; + + ier_val = qm_err_isr_enable_read(qm); + isr_val = qm_err_isr_status_read(qm); + ecsr_val = qm_in(ECSR); + isr_mask = isr_val & ier_val; + + if (!isr_mask) + return IRQ_NONE; + for (i = 0; i < QMAN_HWE_COUNT; i++) { + if (qman_hwerr_txts[i].mask & isr_mask) { + pr_warn("ErrInt: %s\n", qman_hwerr_txts[i].txt); + if (qman_hwerr_txts[i].mask & ecsr_val) { + log_additional_error_info(isr_mask, ecsr_val); + /* Re-arm error capture registers */ + qm_out(ECSR, ecsr_val); + } + if (qman_hwerr_txts[i].mask & QMAN_ERRS_TO_UNENABLE) { + pr_devel("Un-enabling error 0x%x\n", + qman_hwerr_txts[i].mask); + ier_val &= ~qman_hwerr_txts[i].mask; + qm_err_isr_enable_write(qm, ier_val); + } + } + } + qm_err_isr_status_clear(qm, isr_val); + return IRQ_HANDLED; +} + +static int __bind_irq(struct device_node *node) +{ + int ret, err_irq; + + err_irq = of_irq_to_resource(node, 0, NULL); + if (err_irq == NO_IRQ) { + pr_info("Can't get %s property 'interrupts'\n", + node->full_name); + return -ENODEV; + } + ret = request_irq(err_irq, qman_isr, IRQF_SHARED, "qman-err", node); + if (ret) { + pr_err("request_irq() failed %d for '%s'\n", + ret, node->full_name); + return -ENODEV; + } + /* Write-to-clear any stale bits, (eg. starvation being asserted prior + * to resource allocation during driver init). */ + qm_err_isr_status_clear(qm, 0xffffffff); + /* Enable Error Interrupts */ + qm_err_isr_enable_write(qm, 0xffffffff); + return 0; +} + +int qman_init_ccsr(struct device_node *node) +{ + int ret; + +#ifndef __rtems__ + if (!qman_have_ccsr()) + return 0; + if (node != qm_node) + return -EINVAL; +#endif /* __rtems__ */ + /* FQD memory */ + qm_set_memory(qm, qm_memory_fqd, fqd_a, 1, 0, 0, fqd_sz); + /* PFDR memory */ + qm_set_memory(qm, qm_memory_pfdr, pfdr_a, 1, 0, 0, pfdr_sz); + qm_init_pfdr(qm, 8, pfdr_sz / 64 - 8); + /* thresholds */ + qm_set_pfdr_threshold(qm, 512, 64); + qm_set_sfdr_threshold(qm, 128); + /* clear stale PEBI bit from interrupt status register */ + qm_err_isr_status_clear(qm, QM_EIRQ_PEBI); + /* corenet initiator settings */ + qm_set_corenet_initiator(qm); + /* HID settings */ + qm_set_hid(qm); + /* Set scheduling weights to defaults */ + for (ret = qm_wq_first; ret <= qm_wq_last; ret++) + qm_set_wq_scheduling(qm, ret, 0, 0, 0, 0, 0, 0, 0); + /* We are not prepared to accept ERNs for hardware enqueues */ + qm_set_dc(qm, qm_dc_portal_fman0, 1, 0); + qm_set_dc(qm, qm_dc_portal_fman1, 1, 0); + /* Initialise Error Interrupt Handler */ + ret = __bind_irq(node); + if (ret) + return ret; + return 0; +} + +#define LIO_CFG_LIODN_MASK 0x0fff0000 +void qman_liodn_fixup(u16 channel) +{ + static int done; + static u32 liodn_offset; + u32 before, after; + int idx = channel - QM_CHANNEL_SWPORTAL0; + + if (!qman_have_ccsr()) + return; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + before = qm_in(REV3_QCSP_LIO_CFG(idx)); + else + before = qm_in(QCSP_LIO_CFG(idx)); + if (!done) { + liodn_offset = before & LIO_CFG_LIODN_MASK; + done = 1; + return; + } + after = (before & (~LIO_CFG_LIODN_MASK)) | liodn_offset; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + qm_out(REV3_QCSP_LIO_CFG(idx), after); + else + qm_out(QCSP_LIO_CFG(idx), after); +} + +#define IO_CFG_SDEST_MASK 0x00ff0000 +int qman_set_sdest(u16 channel, unsigned int cpu_idx) +{ + int idx = channel - QM_CHANNEL_SWPORTAL0; + u32 before, after; + + if (!qman_have_ccsr()) + return -ENODEV; + + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) { + before = qm_in(REV3_QCSP_IO_CFG(idx)); + /* Each pair of vcpu share the same SRQ(SDEST) */ + cpu_idx /= 2; + after = (before & (~IO_CFG_SDEST_MASK)) | (cpu_idx << 16); + qm_out(REV3_QCSP_IO_CFG(idx), after); + } else { + before = qm_in(QCSP_IO_CFG(idx)); + after = (before & (~IO_CFG_SDEST_MASK)) | (cpu_idx << 16); + qm_out(QCSP_IO_CFG(idx), after); + } + return 0; +} + +#define MISC_CFG_WPM_MASK 0x00000002 +int qm_set_wpm(int wpm) +{ + u32 before; + u32 after; + + if (!qman_have_ccsr()) + return -ENODEV; + + before = qm_in(MISC_CFG); + after = (before & (~MISC_CFG_WPM_MASK)) | (wpm << 1); + qm_out(MISC_CFG, after); + return 0; +} + +int qm_get_wpm(int *wpm) +{ + u32 before; + + if (!qman_have_ccsr()) + return -ENODEV; + + before = qm_in(MISC_CFG); + *wpm = (before & MISC_CFG_WPM_MASK) >> 1; + return 0; +} + +#ifdef CONFIG_SYSFS + +#define DRV_NAME "fsl-qman" + +static ssize_t show_pfdr_fpc(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(PFDR_FPC)); +}; + +static ssize_t show_dlm_avg(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + u32 data; + int i; + + if (sscanf(dev_attr->attr.name, "dcp%d_dlm_avg", &i) != 1) + return -EINVAL; + data = qm_in(DCP_DLM_AVG(i)); + return snprintf(buf, PAGE_SIZE, "%d.%08d\n", data>>8, + (data & 0x000000ff)*390625); +}; + +static ssize_t set_dlm_avg(struct device *dev, + struct device_attribute *dev_attr, const char *buf, size_t count) +{ + unsigned long val; + int i; + + if (sscanf(dev_attr->attr.name, "dcp%d_dlm_avg", &i) != 1) + return -EINVAL; + if (kstrtoul(buf, 0, &val)) { + dev_dbg(dev, "invalid input %s\n", buf); + return -EINVAL; + } + qm_out(DCP_DLM_AVG(i), val); + return count; +}; + +static ssize_t show_pfdr_cfg(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(PFDR_CFG)); +}; + +static ssize_t set_pfdr_cfg(struct device *dev, + struct device_attribute *dev_attr, const char *buf, size_t count) +{ + unsigned long val; + + if (kstrtoul(buf, 0, &val)) { + dev_dbg(dev, "invalid input %s\n", buf); + return -EINVAL; + } + qm_out(PFDR_CFG, val); + return count; +}; + +static ssize_t show_sfdr_in_use(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(SFDR_IN_USE)); +}; + +static ssize_t show_idle_stat(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(IDLE_STAT)); +}; + +static ssize_t show_ci_rlm_avg(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + u32 data = qm_in(CI_RLM_AVG); + + return snprintf(buf, PAGE_SIZE, "%d.%08d\n", data>>8, + (data & 0x000000ff)*390625); +}; + +static ssize_t set_ci_rlm_avg(struct device *dev, + struct device_attribute *dev_attr, const char *buf, size_t count) +{ + unsigned long val; + + if (kstrtoul(buf, 0, &val)) { + dev_dbg(dev, "invalid input %s\n", buf); + return -EINVAL; + } + qm_out(CI_RLM_AVG, val); + return count; +}; + +static ssize_t show_err_isr(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + return snprintf(buf, PAGE_SIZE, "0x%08x\n", qm_in(ERR_ISR)); +}; + + +static ssize_t show_sbec(struct device *dev, + struct device_attribute *dev_attr, char *buf) +{ + int i; + + if (sscanf(dev_attr->attr.name, "sbec_%d", &i) != 1) + return -EINVAL; + return snprintf(buf, PAGE_SIZE, "%u\n", qm_in(SBEC(i))); +}; + +static DEVICE_ATTR(pfdr_fpc, S_IRUSR, show_pfdr_fpc, NULL); +static DEVICE_ATTR(pfdr_cfg, S_IRUSR, show_pfdr_cfg, set_pfdr_cfg); +static DEVICE_ATTR(idle_stat, S_IRUSR, show_idle_stat, NULL); +static DEVICE_ATTR(ci_rlm_avg, (S_IRUSR|S_IWUSR), + show_ci_rlm_avg, set_ci_rlm_avg); +static DEVICE_ATTR(err_isr, S_IRUSR, show_err_isr, NULL); +static DEVICE_ATTR(sfdr_in_use, S_IRUSR, show_sfdr_in_use, NULL); + +static DEVICE_ATTR(dcp0_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); +static DEVICE_ATTR(dcp1_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); +static DEVICE_ATTR(dcp2_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); +static DEVICE_ATTR(dcp3_dlm_avg, (S_IRUSR|S_IWUSR), show_dlm_avg, set_dlm_avg); + +static DEVICE_ATTR(sbec_0, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_1, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_2, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_3, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_4, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_5, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_6, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_7, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_8, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_9, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_10, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_11, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_12, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_13, S_IRUSR, show_sbec, NULL); +static DEVICE_ATTR(sbec_14, S_IRUSR, show_sbec, NULL); + +static struct attribute *qman_dev_attributes[] = { + &dev_attr_pfdr_fpc.attr, + &dev_attr_pfdr_cfg.attr, + &dev_attr_idle_stat.attr, + &dev_attr_ci_rlm_avg.attr, + &dev_attr_err_isr.attr, + &dev_attr_dcp0_dlm_avg.attr, + &dev_attr_dcp1_dlm_avg.attr, + &dev_attr_dcp2_dlm_avg.attr, + &dev_attr_dcp3_dlm_avg.attr, + /* sfdr_in_use will be added if necessary */ + NULL +}; + +static struct attribute *qman_dev_ecr_attributes[] = { + &dev_attr_sbec_0.attr, + &dev_attr_sbec_1.attr, + &dev_attr_sbec_2.attr, + &dev_attr_sbec_3.attr, + &dev_attr_sbec_4.attr, + &dev_attr_sbec_5.attr, + &dev_attr_sbec_6.attr, + &dev_attr_sbec_7.attr, + &dev_attr_sbec_8.attr, + &dev_attr_sbec_9.attr, + &dev_attr_sbec_10.attr, + &dev_attr_sbec_11.attr, + &dev_attr_sbec_12.attr, + &dev_attr_sbec_13.attr, + &dev_attr_sbec_14.attr, + NULL +}; + +/* root level */ +static const struct attribute_group qman_dev_attr_grp = { + .name = NULL, + .attrs = qman_dev_attributes +}; +static const struct attribute_group qman_dev_ecr_grp = { + .name = "error_capture", + .attrs = qman_dev_ecr_attributes +}; + +static int of_fsl_qman_remove(struct platform_device *ofdev) +{ + sysfs_remove_group(&ofdev->dev.kobj, &qman_dev_attr_grp); + return 0; +}; + +static int of_fsl_qman_probe(struct platform_device *ofdev) +{ + int ret; + struct device *dev = &ofdev->dev; + + ret = sysfs_create_group(&dev->kobj, &qman_dev_attr_grp); + if (ret) + goto done; + ret = sysfs_add_file_to_group(&dev->kobj, + &dev_attr_sfdr_in_use.attr, qman_dev_attr_grp.name); + if (ret) + goto del_group_0; + ret = sysfs_create_group(&dev->kobj, &qman_dev_ecr_grp); + if (ret) + goto del_group_0; + + goto done; + +del_group_0: + sysfs_remove_group(&dev->kobj, &qman_dev_attr_grp); +done: + if (ret) + dev_err(dev, "Cannot create dev attributes ret=%d\n", ret); + return ret; +}; + +static const struct of_device_id of_fsl_qman_ids[] = { + { + .compatible = "fsl,qman", + }, + {} +}; + +static struct platform_driver of_fsl_qman_driver = { + .driver = { + .name = DRV_NAME, + .of_match_table = of_fsl_qman_ids, + }, + .probe = of_fsl_qman_probe, + .remove = of_fsl_qman_remove, +}; + +builtin_platform_driver(of_fsl_qman_driver); + +#endif /* CONFIG_SYSFS */ diff --git a/linux/drivers/soc/fsl/qbman/qman.h b/linux/drivers/soc/fsl/qbman/qman.h new file mode 100644 index 00000000..331db7c7 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman.h @@ -0,0 +1,1133 @@ +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_priv.h" + +/* Portal register assists */ + +/* Cache-inhibited register offsets */ +#define QM_REG_EQCR_PI_CINH 0x0000 +#define QM_REG_EQCR_CI_CINH 0x0004 +#define QM_REG_EQCR_ITR 0x0008 +#define QM_REG_DQRR_PI_CINH 0x0040 +#define QM_REG_DQRR_CI_CINH 0x0044 +#define QM_REG_DQRR_ITR 0x0048 +#define QM_REG_DQRR_DCAP 0x0050 +#define QM_REG_DQRR_SDQCR 0x0054 +#define QM_REG_DQRR_VDQCR 0x0058 +#define QM_REG_DQRR_PDQCR 0x005c +#define QM_REG_MR_PI_CINH 0x0080 +#define QM_REG_MR_CI_CINH 0x0084 +#define QM_REG_MR_ITR 0x0088 +#define QM_REG_CFG 0x0100 +#define QM_REG_ISR 0x0e00 +#define QM_REG_IIR 0x0e0c +#define QM_REG_ITPR 0x0e14 + +/* Cache-enabled register offsets */ +#define QM_CL_EQCR 0x0000 +#define QM_CL_DQRR 0x1000 +#define QM_CL_MR 0x2000 +#define QM_CL_EQCR_PI_CENA 0x3000 +#define QM_CL_EQCR_CI_CENA 0x3100 +#define QM_CL_DQRR_PI_CENA 0x3200 +#define QM_CL_DQRR_CI_CENA 0x3300 +#define QM_CL_MR_PI_CENA 0x3400 +#define QM_CL_MR_CI_CENA 0x3500 +#define QM_CL_CR 0x3800 +#define QM_CL_RR0 0x3900 +#define QM_CL_RR1 0x3940 + +/* BTW, the drivers (and h/w programming model) already obtain the required + * synchronisation for portal accesses via lwsync(), hwsync(), and + * data-dependencies. Use of barrier()s or other order-preserving primitives + * simply degrade performance. Hence the use of the __raw_*() interfaces, which + * simply ensure that the compiler treats the portal registers as volatile (ie. + * non-coherent). */ + +/* Cache-inhibited register access. */ +#define __qm_in(qm, o) __raw_readl((qm)->addr_ci + (o)) +#define __qm_out(qm, o, val) __raw_writel((val), (qm)->addr_ci + (o)) +#define qm_in(reg) __qm_in(&portal->addr, QM_REG_##reg) +#define qm_out(reg, val) __qm_out(&portal->addr, QM_REG_##reg, val) + +/* Cache-enabled (index) register access */ +#define __qm_cl_touch_ro(qm, o) dcbt_ro((qm)->addr_ce + (o)) +#define __qm_cl_touch_rw(qm, o) dcbt_rw((qm)->addr_ce + (o)) +#define __qm_cl_in(qm, o) __raw_readl((qm)->addr_ce + (o)) +#define __qm_cl_out(qm, o, val) \ + do { \ + u32 *__tmpclout = (qm)->addr_ce + (o); \ + __raw_writel((val), __tmpclout); \ + dcbf(__tmpclout); \ + } while (0) +#define __qm_cl_invalidate(qm, o) dcbi((qm)->addr_ce + (o)) +#define qm_cl_touch_ro(reg) __qm_cl_touch_ro(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_touch_rw(reg) __qm_cl_touch_rw(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_in(reg) __qm_cl_in(&portal->addr, QM_CL_##reg##_CENA) +#define qm_cl_out(reg, val) __qm_cl_out(&portal->addr, QM_CL_##reg##_CENA, val) +#define qm_cl_invalidate(reg)\ + __qm_cl_invalidate(&portal->addr, QM_CL_##reg##_CENA) + +/* Cache-enabled ring access */ +#define qm_cl(base, idx) ((void *)base + ((idx) << 6)) + +/* Cyclic helper for rings. FIXME: once we are able to do fine-grain perf + * analysis, look at using the "extra" bit in the ring index registers to avoid + * cyclic issues. */ +static inline u8 qm_cyc_diff(u8 ringsize, u8 first, u8 last) +{ + /* 'first' is included, 'last' is excluded */ + if (first <= last) + return last - first; + return ringsize + last - first; +} + +/* Portal modes. + * Enum types; + * pmode == production mode + * cmode == consumption mode, + * dmode == h/w dequeue mode. + * Enum values use 3 letter codes. First letter matches the portal mode, + * remaining two letters indicate; + * ci == cache-inhibited portal register + * ce == cache-enabled portal register + * vb == in-band valid-bit (cache-enabled) + * dc == DCA (Discrete Consumption Acknowledgment), DQRR-only + * As for "enum qm_dqrr_dmode", it should be self-explanatory. + */ +enum qm_eqcr_pmode { /* matches QCSP_CFG::EPM */ + qm_eqcr_pci = 0, /* PI index, cache-inhibited */ + qm_eqcr_pce = 1, /* PI index, cache-enabled */ + qm_eqcr_pvb = 2 /* valid-bit */ +}; +enum qm_dqrr_dmode { /* matches QCSP_CFG::DP */ + qm_dqrr_dpush = 0, /* SDQCR + VDQCR */ + qm_dqrr_dpull = 1 /* PDQCR */ +}; +enum qm_dqrr_pmode { /* s/w-only */ + qm_dqrr_pci, /* reads DQRR_PI_CINH */ + qm_dqrr_pce, /* reads DQRR_PI_CENA */ + qm_dqrr_pvb /* reads valid-bit */ +}; +enum qm_dqrr_cmode { /* matches QCSP_CFG::DCM */ + qm_dqrr_cci = 0, /* CI index, cache-inhibited */ + qm_dqrr_cce = 1, /* CI index, cache-enabled */ + qm_dqrr_cdc = 2 /* Discrete Consumption Acknowledgment */ +}; +enum qm_mr_pmode { /* s/w-only */ + qm_mr_pci, /* reads MR_PI_CINH */ + qm_mr_pce, /* reads MR_PI_CENA */ + qm_mr_pvb /* reads valid-bit */ +}; +enum qm_mr_cmode { /* matches QCSP_CFG::MM */ + qm_mr_cci = 0, /* CI index, cache-inhibited */ + qm_mr_cce = 1 /* CI index, cache-enabled */ +}; + +/* --- Portal structures --- */ + +#define QM_EQCR_SIZE 8 +#define QM_DQRR_SIZE 16 +#define QM_MR_SIZE 8 + +struct qm_eqcr { + struct qm_eqcr_entry *ring, *cursor; + u8 ci, available, ithresh, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + u32 busy; + enum qm_eqcr_pmode pmode; +#endif +}; + +struct qm_dqrr { + const struct qm_dqrr_entry *ring, *cursor; + u8 pi, ci, fill, ithresh, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + enum qm_dqrr_dmode dmode; + enum qm_dqrr_pmode pmode; + enum qm_dqrr_cmode cmode; +#endif +}; + +struct qm_mr { + const struct qm_mr_entry *ring, *cursor; + u8 pi, ci, fill, ithresh, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + enum qm_mr_pmode pmode; + enum qm_mr_cmode cmode; +#endif +}; + +struct qm_mc { + struct qm_mc_command *cr; + struct qm_mc_result *rr; + u8 rridx, vbit; +#ifdef CONFIG_FSL_DPA_CHECKING + enum { + /* Can be _mc_start()ed */ + qman_mc_idle, + /* Can be _mc_commit()ed or _mc_abort()ed */ + qman_mc_user, + /* Can only be _mc_retry()ed */ + qman_mc_hw + } state; +#endif +}; + +#define QM_PORTAL_ALIGNMENT ____cacheline_aligned + +struct qm_addr { + void __iomem *addr_ce; /* cache-enabled */ + void __iomem *addr_ci; /* cache-inhibited */ +}; + +struct qm_portal { + /* In the non-CONFIG_FSL_DPA_CHECKING case, the following stuff up to + * and including 'mc' fits within a cacheline (yay!). The 'config' part + * is setup-only, so isn't a cause for a concern. In other words, don't + * rearrange this structure on a whim, there be dragons ... */ + struct qm_addr addr; + struct qm_eqcr eqcr; + struct qm_dqrr dqrr; + struct qm_mr mr; + struct qm_mc mc; +} QM_PORTAL_ALIGNMENT; + +/* --- EQCR API --- */ + +/* Bit-wise logic to wrap a ring pointer by clearing the "carry bit" */ +#define EQCR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_EQCR_SIZE << 6))) + +/* Bit-wise logic to convert a ring pointer to a ring index */ +static inline u8 EQCR_PTR2IDX(struct qm_eqcr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_EQCR_SIZE - 1); +} + +/* Increment the 'cursor' ring pointer, taking 'vbit' into account */ +static inline void EQCR_INC(struct qm_eqcr *eqcr) +{ + /* NB: this is odd-looking, but experiments show that it generates fast + * code with essentially no branching overheads. We increment to the + * next EQCR pointer and handle overflow and 'vbit'. */ + struct qm_eqcr_entry *partial = eqcr->cursor + 1; + + eqcr->cursor = EQCR_CARRYCLEAR(partial); + if (partial != eqcr->cursor) + eqcr->vbit ^= QM_EQCR_VERB_VBIT; +} + +static inline int qm_eqcr_init(struct qm_portal *portal, + enum qm_eqcr_pmode pmode, + unsigned int eq_stash_thresh, + int eq_stash_prio) +{ + /* This use of 'register', as well as all other occurrences, is because + * it has been observed to generate much faster code with gcc than is + * otherwise the case. */ + register struct qm_eqcr *eqcr = &portal->eqcr; + u32 cfg; + u8 pi; + + eqcr->ring = portal->addr.addr_ce + QM_CL_EQCR; + eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + qm_cl_invalidate(EQCR_CI); + pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + eqcr->cursor = eqcr->ring + pi; + eqcr->vbit = (qm_in(EQCR_PI_CINH) & QM_EQCR_SIZE) ? + QM_EQCR_VERB_VBIT : 0; + eqcr->available = QM_EQCR_SIZE - 1 - + qm_cyc_diff(QM_EQCR_SIZE, eqcr->ci, pi); + eqcr->ithresh = qm_in(EQCR_ITR); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; + eqcr->pmode = pmode; +#endif + cfg = (qm_in(CFG) & 0x00ffffff) | + (eq_stash_thresh << 28) | /* QCSP_CFG: EST */ + (eq_stash_prio << 26) | /* QCSP_CFG: EP */ + ((pmode & 0x3) << 24); /* QCSP_CFG::EPM */ + qm_out(CFG, cfg); + return 0; +} + +static inline unsigned int qm_eqcr_get_ci_stashing(struct qm_portal *portal) +{ + return (qm_in(CFG) >> 28) & 0x7; +} + +static inline void qm_eqcr_finish(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 pi = qm_in(EQCR_PI_CINH) & (QM_EQCR_SIZE - 1); + u8 ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + + DPA_ASSERT(!eqcr->busy); + if (pi != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("losing uncommited EQCR entries\n"); + if (ci != eqcr->ci) + pr_crit("missing existing EQCR completions\n"); + if (eqcr->ci != EQCR_PTR2IDX(eqcr->cursor)) + pr_crit("EQCR destroyed unquiesced\n"); +} + +static inline struct qm_eqcr_entry *qm_eqcr_start_no_stash(struct qm_portal + *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + DPA_ASSERT(!eqcr->busy); + if (!eqcr->available) + return NULL; + + +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 1; +#endif + dcbz_64(eqcr->cursor); + return eqcr->cursor; +} + +static inline struct qm_eqcr_entry *qm_eqcr_start_stash(struct qm_portal + *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci; + + DPA_ASSERT(!eqcr->busy); + if (!eqcr->available) { + old_ci = eqcr->ci; + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + if (!diff) + return NULL; + } +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 1; +#endif + dcbz_64(eqcr->cursor); + return eqcr->cursor; +} + +static inline void qm_eqcr_abort(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + + DPA_ASSERT(eqcr->busy); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline struct qm_eqcr_entry *qm_eqcr_pend_and_next( + struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + DPA_ASSERT(eqcr->busy); + DPA_ASSERT(eqcr->pmode != qm_eqcr_pvb); + if (eqcr->available == 1) + return NULL; + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + dcbf(eqcr->cursor); + EQCR_INC(eqcr); + eqcr->available--; + dcbz_64(eqcr->cursor); + return eqcr->cursor; +} + +#define EQCR_COMMIT_CHECKS(eqcr) \ +do { \ + DPA_ASSERT(eqcr->busy); \ + DPA_ASSERT(eqcr->cursor->orp == (eqcr->cursor->orp & 0x00ffffff)); \ + DPA_ASSERT(eqcr->cursor->fqid == (eqcr->cursor->fqid & 0x00ffffff)); \ +} while (0) + +static inline void qm_eqcr_pci_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + EQCR_COMMIT_CHECKS(eqcr); + DPA_ASSERT(eqcr->pmode == qm_eqcr_pci); + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + EQCR_INC(eqcr); + eqcr->available--; + dcbf(eqcr->cursor); + hwsync(); + qm_out(EQCR_PI_CINH, EQCR_PTR2IDX(eqcr->cursor)); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline void qm_eqcr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + + DPA_ASSERT(eqcr->pmode == qm_eqcr_pce); + qm_cl_invalidate(EQCR_PI); + qm_cl_touch_rw(EQCR_PI); +} + +static inline void qm_eqcr_pce_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + EQCR_COMMIT_CHECKS(eqcr); + DPA_ASSERT(eqcr->pmode == qm_eqcr_pce); + eqcr->cursor->__dont_write_directly__verb = myverb | eqcr->vbit; + EQCR_INC(eqcr); + eqcr->available--; + dcbf(eqcr->cursor); + lwsync(); + qm_cl_out(EQCR_PI, EQCR_PTR2IDX(eqcr->cursor)); +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline void qm_eqcr_pvb_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + struct qm_eqcr_entry *eqcursor; + + EQCR_COMMIT_CHECKS(eqcr); + DPA_ASSERT(eqcr->pmode == qm_eqcr_pvb); + lwsync(); + eqcursor = eqcr->cursor; + eqcursor->__dont_write_directly__verb = myverb | eqcr->vbit; + dcbf(eqcursor); + EQCR_INC(eqcr); + eqcr->available--; +#ifdef CONFIG_FSL_DPA_CHECKING + eqcr->busy = 0; +#endif +} + +static inline u8 qm_eqcr_cci_update(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci = eqcr->ci; + + eqcr->ci = qm_in(EQCR_CI_CINH) & (QM_EQCR_SIZE - 1); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + return diff; +} + +static inline void qm_eqcr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_eqcr *eqcr = &portal->eqcr; + + qm_cl_touch_ro(EQCR_CI); +} + +static inline u8 qm_eqcr_cce_update(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + u8 diff, old_ci = eqcr->ci; + + eqcr->ci = qm_cl_in(EQCR_CI) & (QM_EQCR_SIZE - 1); + qm_cl_invalidate(EQCR_CI); + diff = qm_cyc_diff(QM_EQCR_SIZE, old_ci, eqcr->ci); + eqcr->available += diff; + return diff; +} + +static inline u8 qm_eqcr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + return eqcr->ithresh; +} + +static inline void qm_eqcr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + eqcr->ithresh = ithresh; + qm_out(EQCR_ITR, ithresh); +} + +static inline u8 qm_eqcr_get_avail(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + return eqcr->available; +} + +static inline u8 qm_eqcr_get_fill(struct qm_portal *portal) +{ + register struct qm_eqcr *eqcr = &portal->eqcr; + + return QM_EQCR_SIZE - 1 - eqcr->available; +} + +/* --- DQRR API --- */ + +/* FIXME: many possible improvements; + * - look at changing the API to use pointer rather than index parameters now + * that 'cursor' is a pointer, + * - consider moving other parameters to pointer if it could help (ci) + */ + +#define DQRR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_DQRR_SIZE << 6))) + +static inline u8 DQRR_PTR2IDX(const struct qm_dqrr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_DQRR_SIZE - 1); +} + +static inline const struct qm_dqrr_entry *DQRR_INC( + const struct qm_dqrr_entry *e) +{ + return DQRR_CARRYCLEAR(e + 1); +} + +static inline void qm_dqrr_set_maxfill(struct qm_portal *portal, u8 mf) +{ + qm_out(CFG, (qm_in(CFG) & 0xff0fffff) | + ((mf & (QM_DQRR_SIZE - 1)) << 20)); +} + +static inline int qm_dqrr_init(struct qm_portal *portal, + const struct qm_portal_config *config, + enum qm_dqrr_dmode dmode, + __maybe_unused enum qm_dqrr_pmode pmode, + enum qm_dqrr_cmode cmode, u8 max_fill) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u32 cfg; + + /* Make sure the DQRR will be idle when we enable */ + qm_out(DQRR_SDQCR, 0); + qm_out(DQRR_VDQCR, 0); + qm_out(DQRR_PDQCR, 0); + dqrr->ring = portal->addr.addr_ce + QM_CL_DQRR; + dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->ci = qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); + dqrr->cursor = dqrr->ring + dqrr->ci; + dqrr->fill = qm_cyc_diff(QM_DQRR_SIZE, dqrr->ci, dqrr->pi); + dqrr->vbit = (qm_in(DQRR_PI_CINH) & QM_DQRR_SIZE) ? + QM_DQRR_VERB_VBIT : 0; + dqrr->ithresh = qm_in(DQRR_ITR); +#ifdef CONFIG_FSL_DPA_CHECKING + dqrr->dmode = dmode; + dqrr->pmode = pmode; + dqrr->cmode = cmode; +#endif + /* Invalidate every ring entry before beginning */ + for (cfg = 0; cfg < QM_DQRR_SIZE; cfg++) + dcbi(qm_cl(dqrr->ring, cfg)); + cfg = (qm_in(CFG) & 0xff000f00) | + ((max_fill & (QM_DQRR_SIZE - 1)) << 20) | /* DQRR_MF */ + ((dmode & 1) << 18) | /* DP */ + ((cmode & 3) << 16) | /* DCM */ +#ifndef __rtems__ + 0xa0 | /* RE+SE */ +#endif /* __rtems__ */ + (0 ? 0x40 : 0) | /* Ignore RP */ + (0 ? 0x10 : 0); /* Ignore SP */ + qm_out(CFG, cfg); + qm_dqrr_set_maxfill(portal, max_fill); + return 0; +} + +static inline void qm_dqrr_finish(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; +#ifdef CONFIG_FSL_DPA_CHECKING + if ((dqrr->cmode != qm_dqrr_cdc) && + (dqrr->ci != DQRR_PTR2IDX(dqrr->cursor))) + pr_crit("Ignoring completed DQRR entries\n"); +#endif +} + +static inline const struct qm_dqrr_entry *qm_dqrr_current( + struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + if (!dqrr->fill) + return NULL; + return dqrr->cursor; +} + +static inline u8 qm_dqrr_cursor(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + return DQRR_PTR2IDX(dqrr->cursor); +} + +static inline u8 qm_dqrr_next(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->fill); + dqrr->cursor = DQRR_INC(dqrr->cursor); + return --dqrr->fill; +} + +static inline u8 qm_dqrr_pci_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u8 diff, old_pi = dqrr->pi; + + DPA_ASSERT(dqrr->pmode == qm_dqrr_pci); + dqrr->pi = qm_in(DQRR_PI_CINH) & (QM_DQRR_SIZE - 1); + diff = qm_cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi); + dqrr->fill += diff; + return diff; +} + +static inline void qm_dqrr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->pmode == qm_dqrr_pce); + qm_cl_invalidate(DQRR_PI); + qm_cl_touch_ro(DQRR_PI); +} + +static inline u8 qm_dqrr_pce_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + u8 diff, old_pi = dqrr->pi; + + DPA_ASSERT(dqrr->pmode == qm_dqrr_pce); + dqrr->pi = qm_cl_in(DQRR_PI) & (QM_DQRR_SIZE - 1); + diff = qm_cyc_diff(QM_DQRR_SIZE, old_pi, dqrr->pi); + dqrr->fill += diff; + return diff; +} + +static inline void qm_dqrr_pvb_update(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + const struct qm_dqrr_entry *res = qm_cl(dqrr->ring, dqrr->pi); + + DPA_ASSERT(dqrr->pmode == qm_dqrr_pvb); + /* when accessing 'verb', use __raw_readb() to ensure that compiler + * inlining doesn't try to optimise out "excess reads". */ +#ifdef __rtems__ + dcbi(res); +#endif /* __rtems__ */ + if ((__raw_readb(&res->verb) & QM_DQRR_VERB_VBIT) == dqrr->vbit) { + dqrr->pi = (dqrr->pi + 1) & (QM_DQRR_SIZE - 1); + if (!dqrr->pi) + dqrr->vbit ^= QM_DQRR_VERB_VBIT; + dqrr->fill++; + } +} + +static inline void qm_dqrr_cci_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cci); + dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1); + qm_out(DQRR_CI_CINH, dqrr->ci); +} + +static inline void qm_dqrr_cci_consume_to_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cci); + dqrr->ci = DQRR_PTR2IDX(dqrr->cursor); + qm_out(DQRR_CI_CINH, dqrr->ci); +} + +static inline void qm_dqrr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cce); + qm_cl_invalidate(DQRR_CI); + qm_cl_touch_rw(DQRR_CI); +} + +static inline void qm_dqrr_cce_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cce); + dqrr->ci = (dqrr->ci + num) & (QM_DQRR_SIZE - 1); + qm_cl_out(DQRR_CI, dqrr->ci); +} + +static inline void qm_dqrr_cce_consume_to_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cce); + dqrr->ci = DQRR_PTR2IDX(dqrr->cursor); + qm_cl_out(DQRR_CI, dqrr->ci); +} + +static inline void qm_dqrr_cdc_consume_1(struct qm_portal *portal, u8 idx, + int park) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + DPA_ASSERT(idx < QM_DQRR_SIZE); + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + ((park ? 1 : 0) << 6) | /* PK */ + idx); /* DCAP_CI */ +} + +static inline void qm_dqrr_cdc_consume_1ptr(struct qm_portal *portal, + const struct qm_dqrr_entry *dq, + int park) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + u8 idx = DQRR_PTR2IDX(dq); + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + DPA_ASSERT((dqrr->ring + idx) == dq); + DPA_ASSERT(idx < QM_DQRR_SIZE); + qm_out(DQRR_DCAP, (0 << 8) | /* DQRR_DCAP::S */ + ((park ? 1 : 0) << 6) | /* DQRR_DCAP::PK */ + idx); /* DQRR_DCAP::DCAP_CI */ +} + +static inline void qm_dqrr_cdc_consume_n(struct qm_portal *portal, u16 bitmask) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + qm_out(DQRR_DCAP, (1 << 8) | /* DQRR_DCAP::S */ + ((u32)bitmask << 16)); /* DQRR_DCAP::DCAP_CI */ +} + +static inline u8 qm_dqrr_cdc_cci(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + return qm_in(DQRR_CI_CINH) & (QM_DQRR_SIZE - 1); +} + +static inline void qm_dqrr_cdc_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + qm_cl_invalidate(DQRR_CI); + qm_cl_touch_ro(DQRR_CI); +} + +static inline u8 qm_dqrr_cdc_cce(struct qm_portal *portal) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode == qm_dqrr_cdc); + return qm_cl_in(DQRR_CI) & (QM_DQRR_SIZE - 1); +} + +static inline u8 qm_dqrr_get_ci(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc); + return dqrr->ci; +} + +static inline void qm_dqrr_park(struct qm_portal *portal, u8 idx) +{ + __maybe_unused register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc); + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + (1 << 6) | /* PK */ + (idx & (QM_DQRR_SIZE - 1))); /* DCAP_CI */ +} + +static inline void qm_dqrr_park_current(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + DPA_ASSERT(dqrr->cmode != qm_dqrr_cdc); + qm_out(DQRR_DCAP, (0 << 8) | /* S */ + (1 << 6) | /* PK */ + DQRR_PTR2IDX(dqrr->cursor)); /* DCAP_CI */ +} + +static inline void qm_dqrr_sdqcr_set(struct qm_portal *portal, u32 sdqcr) +{ + qm_out(DQRR_SDQCR, sdqcr); +} + +static inline u32 qm_dqrr_sdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_SDQCR); +} + +static inline void qm_dqrr_vdqcr_set(struct qm_portal *portal, u32 vdqcr) +{ + qm_out(DQRR_VDQCR, vdqcr); +} + +static inline u32 qm_dqrr_vdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_VDQCR); +} + +static inline void qm_dqrr_pdqcr_set(struct qm_portal *portal, u32 pdqcr) +{ + qm_out(DQRR_PDQCR, pdqcr); +} + +static inline u32 qm_dqrr_pdqcr_get(struct qm_portal *portal) +{ + return qm_in(DQRR_PDQCR); +} + +static inline u8 qm_dqrr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_dqrr *dqrr = &portal->dqrr; + + return dqrr->ithresh; +} + +static inline void qm_dqrr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + qm_out(DQRR_ITR, ithresh); +} + +static inline u8 qm_dqrr_get_maxfill(struct qm_portal *portal) +{ + return (qm_in(CFG) & 0x00f00000) >> 20; +} + +/* --- MR API --- */ + +#define MR_CARRYCLEAR(p) \ + (void *)((unsigned long)(p) & (~(unsigned long)(QM_MR_SIZE << 6))) + +static inline u8 MR_PTR2IDX(const struct qm_mr_entry *e) +{ + return ((uintptr_t)e >> 6) & (QM_MR_SIZE - 1); +} + +static inline const struct qm_mr_entry *MR_INC(const struct qm_mr_entry *e) +{ + return MR_CARRYCLEAR(e + 1); +} + +static inline int qm_mr_init(struct qm_portal *portal, enum qm_mr_pmode pmode, + enum qm_mr_cmode cmode) +{ + register struct qm_mr *mr = &portal->mr; + u32 cfg; + + mr->ring = portal->addr.addr_ce + QM_CL_MR; + mr->pi = qm_in(MR_PI_CINH) & (QM_MR_SIZE - 1); + mr->ci = qm_in(MR_CI_CINH) & (QM_MR_SIZE - 1); + mr->cursor = mr->ring + mr->ci; + mr->fill = qm_cyc_diff(QM_MR_SIZE, mr->ci, mr->pi); + mr->vbit = (qm_in(MR_PI_CINH) & QM_MR_SIZE) ? QM_MR_VERB_VBIT : 0; + mr->ithresh = qm_in(MR_ITR); +#ifdef CONFIG_FSL_DPA_CHECKING + mr->pmode = pmode; + mr->cmode = cmode; +#endif + cfg = (qm_in(CFG) & 0xfffff0ff) | + ((cmode & 1) << 8); /* QCSP_CFG:MM */ + qm_out(CFG, cfg); + return 0; +} + +static inline void qm_mr_finish(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + if (mr->ci != MR_PTR2IDX(mr->cursor)) + pr_crit("Ignoring completed MR entries\n"); +} + +static inline const struct qm_mr_entry *qm_mr_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + if (!mr->fill) + return NULL; + return mr->cursor; +} + +static inline u8 qm_mr_cursor(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + return MR_PTR2IDX(mr->cursor); +} + +static inline u8 qm_mr_next(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + DPA_ASSERT(mr->fill); + mr->cursor = MR_INC(mr->cursor); + return --mr->fill; +} + +static inline u8 qm_mr_pci_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + u8 diff, old_pi = mr->pi; + + DPA_ASSERT(mr->pmode == qm_mr_pci); + mr->pi = qm_in(MR_PI_CINH); + diff = qm_cyc_diff(QM_MR_SIZE, old_pi, mr->pi); + mr->fill += diff; + return diff; +} + +static inline void qm_mr_pce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mr *mr = &portal->mr; + + DPA_ASSERT(mr->pmode == qm_mr_pce); + qm_cl_invalidate(MR_PI); + qm_cl_touch_ro(MR_PI); +} + +static inline u8 qm_mr_pce_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + u8 diff, old_pi = mr->pi; + + DPA_ASSERT(mr->pmode == qm_mr_pce); + mr->pi = qm_cl_in(MR_PI) & (QM_MR_SIZE - 1); + diff = qm_cyc_diff(QM_MR_SIZE, old_pi, mr->pi); + mr->fill += diff; + return diff; +} + +static inline void qm_mr_pvb_update(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + const struct qm_mr_entry *res = qm_cl(mr->ring, mr->pi); + + DPA_ASSERT(mr->pmode == qm_mr_pvb); + /* when accessing 'verb', use __raw_readb() to ensure that compiler + * inlining doesn't try to optimise out "excess reads". */ + if ((__raw_readb(&res->verb) & QM_MR_VERB_VBIT) == mr->vbit) { + mr->pi = (mr->pi + 1) & (QM_MR_SIZE - 1); + if (!mr->pi) + mr->vbit ^= QM_MR_VERB_VBIT; + mr->fill++; + res = MR_INC(res); + } + dcbit_ro(res); +} + +static inline void qm_mr_cci_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_mr *mr = &portal->mr; + + DPA_ASSERT(mr->cmode == qm_mr_cci); + mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1); + qm_out(MR_CI_CINH, mr->ci); +} + +static inline void qm_mr_cci_consume_to_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + DPA_ASSERT(mr->cmode == qm_mr_cci); + mr->ci = MR_PTR2IDX(mr->cursor); + qm_out(MR_CI_CINH, mr->ci); +} + +static inline void qm_mr_cce_prefetch(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mr *mr = &portal->mr; + + DPA_ASSERT(mr->cmode == qm_mr_cce); + qm_cl_invalidate(MR_CI); + qm_cl_touch_rw(MR_CI); +} + +static inline void qm_mr_cce_consume(struct qm_portal *portal, u8 num) +{ + register struct qm_mr *mr = &portal->mr; + + DPA_ASSERT(mr->cmode == qm_mr_cce); + mr->ci = (mr->ci + num) & (QM_MR_SIZE - 1); + qm_cl_out(MR_CI, mr->ci); +} + +static inline void qm_mr_cce_consume_to_current(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + DPA_ASSERT(mr->cmode == qm_mr_cce); + mr->ci = MR_PTR2IDX(mr->cursor); + qm_cl_out(MR_CI, mr->ci); +} + +static inline u8 qm_mr_get_ci(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + return mr->ci; +} + +static inline u8 qm_mr_get_ithresh(struct qm_portal *portal) +{ + register struct qm_mr *mr = &portal->mr; + + return mr->ithresh; +} + +static inline void qm_mr_set_ithresh(struct qm_portal *portal, u8 ithresh) +{ + qm_out(MR_ITR, ithresh); +} + +/* --- Management command API --- */ + +static inline int qm_mc_init(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + + mc->cr = portal->addr.addr_ce + QM_CL_CR; + mc->rr = portal->addr.addr_ce + QM_CL_RR0; + mc->rridx = (__raw_readb(&mc->cr->__dont_write_directly__verb) & + QM_MCC_VERB_VBIT) ? 0 : 1; + mc->vbit = mc->rridx ? QM_MCC_VERB_VBIT : 0; +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_idle; +#endif + return 0; +} + +static inline void qm_mc_finish(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mc *mc = &portal->mc; + + DPA_ASSERT(mc->state == qman_mc_idle); +#ifdef CONFIG_FSL_DPA_CHECKING + if (mc->state != qman_mc_idle) + pr_crit("Losing incomplete MC command\n"); +#endif +} + +static inline struct qm_mc_command *qm_mc_start(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + + DPA_ASSERT(mc->state == qman_mc_idle); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_user; +#endif + dcbz_64(mc->cr); + return mc->cr; +} + +static inline void qm_mc_abort(struct qm_portal *portal) +{ + __maybe_unused register struct qm_mc *mc = &portal->mc; + + DPA_ASSERT(mc->state == qman_mc_user); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_idle; +#endif +} + +static inline void qm_mc_commit(struct qm_portal *portal, u8 myverb) +{ + register struct qm_mc *mc = &portal->mc; + struct qm_mc_result *rr = mc->rr + mc->rridx; + + DPA_ASSERT(mc->state == qman_mc_user); + lwsync(); + mc->cr->__dont_write_directly__verb = myverb | mc->vbit; + dcbf(mc->cr); + dcbit_ro(rr); +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_hw; +#endif +} + +static inline struct qm_mc_result *qm_mc_result(struct qm_portal *portal) +{ + register struct qm_mc *mc = &portal->mc; + struct qm_mc_result *rr = mc->rr + mc->rridx; + + DPA_ASSERT(mc->state == qman_mc_hw); + /* The inactive response register's verb byte always returns zero until + * its command is submitted and completed. This includes the valid-bit, + * in case you were wondering... */ + if (!__raw_readb(&rr->verb)) { + dcbit_ro(rr); + return NULL; + } + mc->rridx ^= 1; + mc->vbit ^= QM_MCC_VERB_VBIT; +#ifdef CONFIG_FSL_DPA_CHECKING + mc->state = qman_mc_idle; +#endif + return rr; +} + +/* --- Portal interrupt register API --- */ + +static inline int qm_isr_init(__always_unused struct qm_portal *portal) +{ + return 0; +} + +static inline void qm_isr_finish(__always_unused struct qm_portal *portal) +{ +} + +static inline void qm_isr_set_iperiod(struct qm_portal *portal, u16 iperiod) +{ + qm_out(ITPR, iperiod); +} + +static inline u32 __qm_isr_read(struct qm_portal *portal, enum qm_isr_reg n) +{ + return __qm_in(&portal->addr, QM_REG_ISR + (n << 2)); +} + +static inline void __qm_isr_write(struct qm_portal *portal, enum qm_isr_reg n, + u32 val) +{ + __qm_out(&portal->addr, QM_REG_ISR + (n << 2), val); +} diff --git a/linux/drivers/soc/fsl/qbman/qman_api.c b/linux/drivers/soc/fsl/qbman/qman_api.c new file mode 100644 index 00000000..e838d08f --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_api.c @@ -0,0 +1,3026 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman.h" + +/* Compilation constants */ +#define DQRR_MAXFILL 15 +#define EQCR_ITHRESH 4 /* if EQCR congests, interrupt threshold */ +#define IRQNAME "QMan portal %d" +#define MAX_IRQNAME 16 /* big enough for "QMan portal %d" */ +#define QMAN_POLL_LIMIT 32 +#define QMAN_PIRQ_DQRR_ITHRESH 12 +#define QMAN_PIRQ_MR_ITHRESH 4 +#define QMAN_PIRQ_IPERIOD 100 +#define FSL_DPA_PORTAL_SHARE 1 /* Allow portals to be shared */ +/* Divide 'n' by 'd', rounding down if 'r' is negative, rounding up if it's + * positive, and rounding to the closest value if it's zero. NB, this macro + * implicitly upgrades parameters to unsigned 64-bit, so feed it with types + * that are compatible with this. NB, these arguments should not be expressions + * unless it is safe for them to be evaluated multiple times. Eg. do not pass + * in "some_value++" as a parameter to the macro! */ +#define ROUNDING(n, d, r) \ + (((r) < 0) ? div64_u64((n), (d)) : \ + (((r) > 0) ? div64_u64(((n) + (d) - 1), (d)) : \ + div64_u64(((n) + ((d) / 2)), (d)))) + +/* Lock/unlock frame queues, subject to the "LOCKED" flag. This is about + * inter-processor locking only. Note, FQLOCK() is always called either under a + * local_irq_save() or from interrupt context - hence there's no need for irq + * protection (and indeed, attempting to nest irq-protection doesn't work, as + * the "irq en/disable" machinery isn't recursive...). */ +#define FQLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_lock(&__fq478->fqlock); \ + } while (0) +#define FQUNLOCK(fq) \ + do { \ + struct qman_fq *__fq478 = (fq); \ + if (fq_isset(__fq478, QMAN_FQ_FLAG_LOCKED)) \ + spin_unlock(&__fq478->fqlock); \ + } while (0) + +static inline void fq_set(struct qman_fq *fq, u32 mask) +{ + set_bits(mask, &fq->flags); +} +static inline void fq_clear(struct qman_fq *fq, u32 mask) +{ + clear_bits(mask, &fq->flags); +} +static inline int fq_isset(struct qman_fq *fq, u32 mask) +{ + return fq->flags & mask; +} +static inline int fq_isclear(struct qman_fq *fq, u32 mask) +{ + return !(fq->flags & mask); +} + +struct qman_portal { + struct qm_portal p; + unsigned long bits; /* PORTAL_BITS_*** - dynamic, strictly internal */ + unsigned long irq_sources; + u32 use_eqcr_ci_stashing; + u32 slowpoll; /* only used when interrupts are off */ + struct qman_fq *vdqcr_owned; /* only 1 volatile dequeue at a time */ +#ifdef FSL_DPA_CAN_WAIT_SYNC + struct qman_fq *eqci_owned; /* only 1 enqueue WAIT_SYNC at a time */ +#endif +#ifdef FSL_DPA_PORTAL_SHARE + raw_spinlock_t sharing_lock; /* only used if is_shared */ +#ifndef __rtems__ + int is_shared; + struct qman_portal *sharing_redirect; +#endif /* __rtems__ */ +#endif + u32 sdqcr; + int dqrr_disable_ref; + /* A portal-specific handler for DCP ERNs. If this is NULL, the global + * handler is called instead. */ + qman_cb_dc_ern cb_dc_ern; + /* When the cpu-affine portal is activated, this is non-NULL */ + const struct qm_portal_config *config; +#ifndef __rtems__ + /* This is needed for providing a non-NULL device to dma_map_***() */ + struct platform_device *pdev; +#endif /* __rtems__ */ + struct dpa_rbtree retire_table; + char irqname[MAX_IRQNAME]; + /* 2-element array. cgrs[0] is mask, cgrs[1] is snapshot. */ + struct qman_cgrs *cgrs; + /* linked-list of CSCN handlers. */ + struct list_head cgr_cbs; + /* list lock */ + spinlock_t cgr_lock; + /* track if memory was allocated by the driver */ + u8 alloced; +}; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ +#define PORTAL_IRQ_LOCK(p, irqflags) \ + do { \ + if ((p)->is_shared) \ + raw_spin_lock_irqsave(&(p)->sharing_lock, irqflags); \ + else \ + local_irq_save(irqflags); \ + } while (0) +#define PORTAL_IRQ_UNLOCK(p, irqflags) \ + do { \ + if ((p)->is_shared) \ + raw_spin_unlock_irqrestore(&(p)->sharing_lock, \ + irqflags); \ + else \ + local_irq_restore(irqflags); \ + } while (0) +#else /* __rtems__ */ +#define PORTAL_IRQ_LOCK(p, irqflags) \ + raw_spin_lock_irqsave(&(p)->sharing_lock, irqflags) +#define PORTAL_IRQ_UNLOCK(p, irqflags) \ + raw_spin_unlock_irqrestore(&(p)->sharing_lock, irqflags) +#endif /* __rtems__ */ +#else +#define PORTAL_IRQ_LOCK(p, irqflags) local_irq_save(irqflags) +#define PORTAL_IRQ_UNLOCK(p, irqflags) local_irq_restore(irqflags) +#endif + +/* Global handler for DCP ERNs. Used when the portal receiving the message does + * not have a portal-specific handler. */ +static qman_cb_dc_ern cb_dc_ern; + +#ifndef __rtems__ +static cpumask_t affine_mask; +static DEFINE_SPINLOCK(affine_mask_lock); +static u16 affine_channels[NR_CPUS]; +#endif /* __rtems__ */ +static DEFINE_PER_CPU(struct qman_portal, qman_affine_portal); +#ifndef __rtems__ +void *affine_portals[NR_CPUS]; +#endif /* __rtems__ */ + +/* "raw" gets the cpu-local struct whether it's a redirect or not. */ +static inline struct qman_portal *get_raw_affine_portal(void) +{ + return &get_cpu_var(qman_affine_portal); +} +/* For ops that can redirect, this obtains the portal to use */ +#ifdef FSL_DPA_PORTAL_SHARE +static inline struct qman_portal *get_affine_portal(void) +{ + struct qman_portal *p = get_raw_affine_portal(); + +#ifndef __rtems__ + if (p->sharing_redirect) + return p->sharing_redirect; +#endif /* __rtems__ */ + return p; +} +#else +#define get_affine_portal() get_raw_affine_portal() +#endif +/* For every "get", there must be a "put" */ +static inline void put_affine_portal(void) +{ + put_cpu_var(qman_affine_portal); +} +/* Exception: poll functions assume the caller is cpu-affine and in no risk of + * re-entrance, which are the two reasons we usually use the get/put_cpu_var() + * semantic - ie. to disable pre-emption. Some use-cases expect the execution + * context to remain as non-atomic during poll-triggered callbacks as it was + * when the poll API was first called (eg. NAPI), so we go out of our way in + * this case to not disable pre-emption. */ +static inline struct qman_portal *get_poll_portal(void) +{ + return this_cpu_ptr(&qman_affine_portal); +} +#define put_poll_portal() + +/* This gives a FQID->FQ lookup to cover the fact that we can't directly demux + * retirement notifications (the fact they are sometimes h/w-consumed means that + * contextB isn't always a s/w demux - and as we can't know which case it is + * when looking at the notification, we have to use the slow lookup for all of + * them). NB, it's possible to have multiple FQ objects refer to the same FQID + * (though at most one of them should be the consumer), so this table isn't for + * all FQs - FQs are added when retirement commands are issued, and removed when + * they complete, which also massively reduces the size of this table. */ +IMPLEMENT_DPA_RBTREE(fqtree, struct qman_fq, node, fqid); + +/* This is what everything can wait on, even if it migrates to a different cpu + * to the one whose affine portal it is waiting on. */ +static DECLARE_WAIT_QUEUE_HEAD(affine_queue); + +static inline int table_push_fq(struct qman_portal *p, struct qman_fq *fq) +{ + int ret = fqtree_push(&p->retire_table, fq); + + if (ret) + pr_err("ERROR: double FQ-retirement %d\n", fq->fqid); + return ret; +} + +static inline void table_del_fq(struct qman_portal *p, struct qman_fq *fq) +{ + fqtree_del(&p->retire_table, fq); +} + +static inline struct qman_fq *table_find_fq(struct qman_portal *p, u32 fqid) +{ + return fqtree_find(&p->retire_table, fqid); +} + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +static void **qman_fq_lookup_table; +static size_t qman_fq_lookup_table_size; + +int qman_setup_fq_lookup_table(size_t num_entries) +{ + num_entries++; + /* Allocate 1 more entry since the first entry is not used */ + qman_fq_lookup_table = vzalloc((num_entries * sizeof(void *))); + if (!qman_fq_lookup_table) + return -ENOMEM; + qman_fq_lookup_table_size = num_entries; + pr_info("Allocated lookup table at %p, entry count %lu\n", + qman_fq_lookup_table, (unsigned long)qman_fq_lookup_table_size); + return 0; +} + +/* global structure that maintains fq object mapping */ +static DEFINE_SPINLOCK(fq_hash_table_lock); + +static int find_empty_fq_table_entry(u32 *entry, struct qman_fq *fq) +{ + u32 i; + + spin_lock(&fq_hash_table_lock); + /* Can't use index zero because this has special meaning + * in context_b field. */ + for (i = 1; i < qman_fq_lookup_table_size; i++) { + if (qman_fq_lookup_table[i] == NULL) { + *entry = i; + qman_fq_lookup_table[i] = fq; + spin_unlock(&fq_hash_table_lock); + return 0; + } + } + spin_unlock(&fq_hash_table_lock); + return -ENOMEM; +} + +static void clear_fq_table_entry(u32 entry) +{ + spin_lock(&fq_hash_table_lock); + BUG_ON(entry >= qman_fq_lookup_table_size); + qman_fq_lookup_table[entry] = NULL; + spin_unlock(&fq_hash_table_lock); +} + +static inline struct qman_fq *get_fq_table_entry(u32 entry) +{ + BUG_ON(entry >= qman_fq_lookup_table_size); + return qman_fq_lookup_table[entry]; +} +#endif + +/* In the case that slow- and fast-path handling are both done by qman_poll() + * (ie. because there is no interrupt handling), we ought to balance how often + * we do the fast-path poll versus the slow-path poll. We'll use two decrementer + * sources, so we call the fast poll 'n' times before calling the slow poll + * once. The idle decrementer constant is used when the last slow-poll detected + * no work to do, and the busy decrementer constant when the last slow-poll had + * work to do. */ +#define SLOW_POLL_IDLE 1000 +#define SLOW_POLL_BUSY 10 +static u32 __poll_portal_slow(struct qman_portal *p, u32 is); +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit); + +/* Portal interrupt handler */ +static irqreturn_t portal_isr(__always_unused int irq, void *ptr) +{ + struct qman_portal *p = ptr; + /* + * The CSCI source is cleared inside __poll_portal_slow(), because + * it could race against a Query Congestion State command also given + * as part of the handling of this interrupt source. We mustn't + * clear it a second time in this top-level function. + */ + u32 clear = QM_DQAVAIL_MASK; + u32 is = qm_isr_status_read(&p->p) & p->irq_sources; + /* DQRR-handling if it's interrupt-driven */ + if (is & QM_PIRQ_DQRI) { + clear |= QM_PIRQ_DQRI; + __poll_portal_fast(p, QMAN_POLL_LIMIT); + } + /* Handling of anything else that's interrupt-driven */ + clear |= __poll_portal_slow(p, is); + qm_isr_status_clear(&p->p, clear); + return IRQ_HANDLED; +} + +/* This inner version is used privately by qman_create_affine_portal(), as well + * as by the exported qman_stop_dequeues(). */ +static inline void qman_stop_dequeues_ex(struct qman_portal *p) +{ + unsigned long irqflags __maybe_unused; + PORTAL_IRQ_LOCK(p, irqflags); + if (!(p->dqrr_disable_ref++)) + qm_dqrr_set_maxfill(&p->p, 0); + PORTAL_IRQ_UNLOCK(p, irqflags); +} + +static int drain_mr_fqrni(struct qm_portal *p) +{ + const struct qm_mr_entry *msg; +loop: + msg = qm_mr_current(p); + if (!msg) { + /* if MR was full and h/w had other FQRNI entries to produce, we + * need to allow it time to produce those entries once the + * existing entries are consumed. A worst-case situation + * (fully-loaded system) means h/w sequencers may have to do 3-4 + * other things before servicing the portal's MR pump, each of + * which (if slow) may take ~50 qman cycles (which is ~200 + * processor cycles). So rounding up and then multiplying this + * worst-case estimate by a factor of 10, just to be + * ultra-paranoid, goes as high as 10,000 cycles. NB, we consume + * one entry at a time, so h/w has an opportunity to produce new + * entries well before the ring has been fully consumed, so + * we're being *really* paranoid here. */ + u64 now, then = mfatb(); + + do { + now = mfatb(); + } while ((then + 10000) > now); + msg = qm_mr_current(p); + if (!msg) + return 0; + } + if ((msg->verb & QM_MR_VERB_TYPE_MASK) != QM_MR_VERB_FQRNI) { + /* We aren't draining anything but FQRNIs */ + pr_err("Found verb 0x%x in MR\n", msg->verb); + return -1; + } + qm_mr_next(p); + qm_mr_cci_consume(p, 1); + goto loop; +} + +struct qman_portal *qman_create_portal( + struct qman_portal *portal, + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs) +{ + struct qm_portal *__p; +#ifndef __rtems__ + char buf[16]; +#endif /* __rtems__ */ + int ret; + u32 isdr; + + if (!portal) { + portal = kmalloc(sizeof(*portal), GFP_KERNEL); + if (!portal) + return portal; + portal->alloced = 1; + } else + portal->alloced = 0; + + __p = &portal->p; + +#ifndef __rtems__ + portal->use_eqcr_ci_stashing = ((qman_ip_rev >= QMAN_REV30) ? + 1 : 0); +#else /* __rtems__ */ + portal->use_eqcr_ci_stashing = 0; +#endif /* __rtems__ */ + + /* prep the low-level portal struct with the mapped addresses from the + * config, everything that follows depends on it and "config" is more + * for (de)reference... */ + __p->addr.addr_ce = config->addr_virt[DPA_PORTAL_CE]; + __p->addr.addr_ci = config->addr_virt[DPA_PORTAL_CI]; + /* + * If CI-stashing is used, the current defaults use a threshold of 3, + * and stash with high-than-DQRR priority. + */ + if (qm_eqcr_init(__p, qm_eqcr_pvb, + portal->use_eqcr_ci_stashing ? 3 : 0, 1)) { + pr_err("EQCR initialisation failed\n"); + goto fail_eqcr; + } + if (qm_dqrr_init(__p, config, qm_dqrr_dpush, qm_dqrr_pvb, + qm_dqrr_cdc, DQRR_MAXFILL)) { + pr_err("DQRR initialisation failed\n"); + goto fail_dqrr; + } + if (qm_mr_init(__p, qm_mr_pvb, qm_mr_cci)) { + pr_err("MR initialisation failed\n"); + goto fail_mr; + } + if (qm_mc_init(__p)) { + pr_err("MC initialisation failed\n"); + goto fail_mc; + } + if (qm_isr_init(__p)) { + pr_err("ISR initialisation failed\n"); + goto fail_isr; + } + /* static interrupt-gating controls */ + qm_dqrr_set_ithresh(__p, QMAN_PIRQ_DQRR_ITHRESH); + qm_mr_set_ithresh(__p, QMAN_PIRQ_MR_ITHRESH); + qm_isr_set_iperiod(__p, QMAN_PIRQ_IPERIOD); + portal->cgrs = kmalloc(2 * sizeof(*cgrs), GFP_KERNEL); + if (!portal->cgrs) + goto fail_cgrs; + /* initial snapshot is no-depletion */ + qman_cgrs_init(&portal->cgrs[1]); + if (cgrs) + portal->cgrs[0] = *cgrs; + else + /* if the given mask is NULL, assume all CGRs can be seen */ + qman_cgrs_fill(&portal->cgrs[0]); + INIT_LIST_HEAD(&portal->cgr_cbs); + spin_lock_init(&portal->cgr_lock); + portal->bits = 0; + portal->slowpoll = 0; +#ifdef FSL_DPA_CAN_WAIT_SYNC + portal->eqci_owned = NULL; +#endif +#ifdef FSL_DPA_PORTAL_SHARE + raw_spin_lock_init(&portal->sharing_lock); +#ifndef __rtems__ + portal->is_shared = config->public_cfg.is_shared; + portal->sharing_redirect = NULL; +#endif /* __rtems__ */ +#endif + portal->sdqcr = QM_SDQCR_SOURCE_CHANNELS | QM_SDQCR_COUNT_UPTO3 | + QM_SDQCR_DEDICATED_PRECEDENCE | QM_SDQCR_TYPE_PRIO_QOS | + QM_SDQCR_TOKEN_SET(0xab) | QM_SDQCR_CHANNELS_DEDICATED; + portal->dqrr_disable_ref = 0; + portal->cb_dc_ern = NULL; +#ifndef __rtems__ + sprintf(buf, "qportal-%d", config->public_cfg.channel); + portal->pdev = platform_device_alloc(buf, -1); + if (!portal->pdev) + goto fail_devalloc; + if (dma_set_mask(&portal->pdev->dev, DMA_BIT_MASK(40))) + goto fail_devadd; + ret = platform_device_add(portal->pdev); + if (ret) + goto fail_devadd; +#endif /* __rtems__ */ + dpa_rbtree_init(&portal->retire_table); + isdr = 0xffffffff; + qm_isr_disable_write(__p, isdr); + portal->irq_sources = 0; + qm_isr_enable_write(__p, portal->irq_sources); + qm_isr_status_clear(__p, 0xffffffff); + snprintf(portal->irqname, MAX_IRQNAME, IRQNAME, config->public_cfg.cpu); + if (request_irq(config->public_cfg.irq, portal_isr, 0, portal->irqname, + portal)) { + pr_err("request_irq() failed\n"); + goto fail_irq; + } +#ifndef __rtems__ + if ((config->public_cfg.cpu != -1) && + irq_can_set_affinity(config->public_cfg.irq) && + irq_set_affinity(config->public_cfg.irq, + cpumask_of(config->public_cfg.cpu))) { + pr_err("irq_set_affinity() failed\n"); + goto fail_affinity; + } +#endif /* __rtems__ */ + + /* Need EQCR to be empty before continuing */ + isdr ^= QM_PIRQ_EQCI; + qm_isr_disable_write(__p, isdr); + ret = qm_eqcr_get_fill(__p); + if (ret) { + pr_err("EQCR unclean\n"); + goto fail_eqcr_empty; + } + isdr ^= (QM_PIRQ_DQRI | QM_PIRQ_MRI); + qm_isr_disable_write(__p, isdr); + if (qm_dqrr_current(__p) != NULL) { + pr_err("DQRR unclean\n"); + qm_dqrr_cdc_consume_n(__p, 0xffff); + } + if (qm_mr_current(__p) != NULL) { + /* special handling, drain just in case it's a few FQRNIs */ + if (drain_mr_fqrni(__p)) { + const struct qm_mr_entry *e = qm_mr_current(__p); + + pr_err("MR unclean, MR VERB 0x%x, rc 0x%x\n, addr 0x%x", + e->verb, e->ern.rc, e->ern.fd.addr_lo); + goto fail_dqrr_mr_empty; + } + } + /* Success */ + portal->config = config; + qm_isr_disable_write(__p, 0); + qm_isr_uninhibit(__p); + /* Write a sane SDQCR */ + qm_dqrr_sdqcr_set(__p, portal->sdqcr); + return portal; +fail_dqrr_mr_empty: +fail_eqcr_empty: +#ifndef __rtems__ +fail_affinity: +#endif /* __rtems__ */ + free_irq(config->public_cfg.irq, portal); +fail_irq: +#ifndef __rtems__ + platform_device_del(portal->pdev); +fail_devadd: + platform_device_put(portal->pdev); +fail_devalloc: +#endif /* __rtems__ */ + kfree(portal->cgrs); +fail_cgrs: + qm_isr_finish(__p); +fail_isr: + qm_mc_finish(__p); +fail_mc: + qm_mr_finish(__p); +fail_mr: + qm_dqrr_finish(__p); +fail_dqrr: + qm_eqcr_finish(__p); +fail_eqcr: + return NULL; +} + +struct qman_portal *qman_create_affine_portal( + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs) +{ + struct qman_portal *res; + struct qman_portal *portal; + + portal = &per_cpu(qman_affine_portal, config->public_cfg.cpu); + res = qman_create_portal(portal, config, cgrs); + if (res) { +#ifndef __rtems__ + spin_lock(&affine_mask_lock); + cpumask_set_cpu(config->public_cfg.cpu, &affine_mask); + affine_channels[config->public_cfg.cpu] = + config->public_cfg.channel; + affine_portals[config->public_cfg.cpu] = portal; + spin_unlock(&affine_mask_lock); +#endif /* __rtems__ */ + } + return res; +} + +#ifndef __rtems__ +/* These checks are BUG_ON()s because the driver is already supposed to avoid + * these cases. */ +struct qman_portal *qman_create_affine_slave(struct qman_portal *redirect, + int cpu) +{ +#ifdef FSL_DPA_PORTAL_SHARE + struct qman_portal *p = &per_cpu(qman_affine_portal, cpu); + + /* Check that we don't already have our own portal */ + BUG_ON(p->config); + /* Check that we aren't already slaving to another portal */ + BUG_ON(p->is_shared); + /* Check that 'redirect' is prepared to have us */ + BUG_ON(!redirect->config->public_cfg.is_shared); + /* These are the only elements to initialise when redirecting */ + p->irq_sources = 0; + p->sharing_redirect = redirect; + affine_portals[cpu] = p; + return p; +#else + BUG(); + return NULL; +#endif +} +#endif /* __rtems__ */ + +void qman_destroy_portal(struct qman_portal *qm) +{ + const struct qm_portal_config *pcfg; + + /* Stop dequeues on the portal */ + qm_dqrr_sdqcr_set(&qm->p, 0); + + /* NB we do this to "quiesce" EQCR. If we add enqueue-completions or + * something related to QM_PIRQ_EQCI, this may need fixing. + * Also, due to the prefetching model used for CI updates in the enqueue + * path, this update will only invalidate the CI cacheline *after* + * working on it, so we need to call this twice to ensure a full update + * irrespective of where the enqueue processing was at when the teardown + * began. */ + qm_eqcr_cce_update(&qm->p); + qm_eqcr_cce_update(&qm->p); + pcfg = qm->config; + + free_irq(pcfg->public_cfg.irq, qm); + + kfree(qm->cgrs); + qm_isr_finish(&qm->p); + qm_mc_finish(&qm->p); + qm_mr_finish(&qm->p); + qm_dqrr_finish(&qm->p); + qm_eqcr_finish(&qm->p); + +#ifndef __rtems__ + platform_device_del(qm->pdev); + platform_device_put(qm->pdev); +#endif /* __rtems__ */ + + qm->config = NULL; + if (qm->alloced) + kfree(qm); +} + +const struct qm_portal_config *qman_destroy_affine_portal(void) +{ + /* We don't want to redirect if we're a slave, use "raw" */ + struct qman_portal *qm = get_raw_affine_portal(); + const struct qm_portal_config *pcfg; +#ifndef __rtems__ + int cpu; +#endif /* __rtems__ */ + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (qm->sharing_redirect) { + qm->sharing_redirect = NULL; + put_affine_portal(); + return NULL; + } + qm->is_shared = 0; +#endif /* __rtems__ */ +#endif + pcfg = qm->config; +#ifndef __rtems__ + cpu = pcfg->public_cfg.cpu; +#endif /* __rtems__ */ + + qman_destroy_portal(qm); + +#ifndef __rtems__ + spin_lock(&affine_mask_lock); + cpumask_clear_cpu(cpu, &affine_mask); + spin_unlock(&affine_mask_lock); +#endif /* __rtems__ */ + put_affine_portal(); + return pcfg; +} + +const struct qman_portal_config *qman_p_get_portal_config(struct qman_portal *p) +{ + return &p->config->public_cfg; +} +EXPORT_SYMBOL(qman_p_get_portal_config); + +const struct qman_portal_config *qman_get_portal_config(void) +{ + struct qman_portal *p = get_affine_portal(); + const struct qman_portal_config *ret = qman_p_get_portal_config(p); + + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_get_portal_config); + +/* Inline helper to reduce nesting in __poll_portal_slow() */ +static inline void fq_state_change(struct qman_portal *p, struct qman_fq *fq, + const struct qm_mr_entry *msg, u8 verb) +{ + FQLOCK(fq); + switch (verb) { + case QM_MR_VERB_FQRL: + DPA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_ORL)); + fq_clear(fq, QMAN_FQ_STATE_ORL); + table_del_fq(p, fq); + break; + case QM_MR_VERB_FQRN: + DPA_ASSERT((fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_sched)); + DPA_ASSERT(fq_isset(fq, QMAN_FQ_STATE_CHANGING)); + fq_clear(fq, QMAN_FQ_STATE_CHANGING); + if (msg->fq.fqs & QM_MR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (msg->fq.fqs & QM_MR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + fq->state = qman_fq_state_retired; + break; + case QM_MR_VERB_FQPN: + DPA_ASSERT(fq->state == qman_fq_state_sched); + DPA_ASSERT(fq_isclear(fq, QMAN_FQ_STATE_CHANGING)); + fq->state = qman_fq_state_parked; + } + FQUNLOCK(fq); +} + +static u32 __poll_portal_slow(struct qman_portal *p, u32 is) +{ + const struct qm_mr_entry *msg; + + if (is & QM_PIRQ_CSCI) { + struct qman_cgrs rr, c; + struct qm_mc_result *mcr; + struct qman_cgr *cgr; + unsigned long irqflags __maybe_unused; + + spin_lock_irqsave(&p->cgr_lock, irqflags); + /* + * The CSCI bit must be cleared _before_ issuing the + * Query Congestion State command, to ensure that a long + * CGR State Change callback cannot miss an intervening + * state change. + */ + qm_isr_status_clear(&p->p, QM_PIRQ_CSCI); + qm_mc_start(&p->p); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + /* mask out the ones I'm not interested in */ + qman_cgrs_and(&rr, (const struct qman_cgrs *) + &mcr->querycongestion.state, &p->cgrs[0]); + /* check previous snapshot for delta, enter/exit congestion */ + qman_cgrs_xor(&c, &rr, &p->cgrs[1]); + /* update snapshot */ + qman_cgrs_cp(&p->cgrs[1], &rr); + /* Invoke callback */ + list_for_each_entry(cgr, &p->cgr_cbs, node) + if (cgr->cb && qman_cgrs_get(&c, cgr->cgrid)) + cgr->cb(p, cgr, qman_cgrs_get(&rr, cgr->cgrid)); + spin_unlock_irqrestore(&p->cgr_lock, irqflags); + } + +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (is & QM_PIRQ_EQCI) { + unsigned long irqflags; + + PORTAL_IRQ_LOCK(p, irqflags); + p->eqci_owned = NULL; + PORTAL_IRQ_UNLOCK(p, irqflags); + wake_up(&affine_queue); + } +#endif + + if (is & QM_PIRQ_EQRI) { + unsigned long irqflags __maybe_unused; + + PORTAL_IRQ_LOCK(p, irqflags); + qm_eqcr_cce_update(&p->p); + qm_eqcr_set_ithresh(&p->p, 0); + PORTAL_IRQ_UNLOCK(p, irqflags); + wake_up(&affine_queue); + } + + if (is & QM_PIRQ_MRI) { + struct qman_fq *fq; + u8 verb, num = 0; +mr_loop: + qm_mr_pvb_update(&p->p); + msg = qm_mr_current(&p->p); + if (!msg) + goto mr_done; + verb = msg->verb & QM_MR_VERB_TYPE_MASK; + /* The message is a software ERN iff the 0x20 bit is set */ + if (verb & 0x20) { + switch (verb) { + case QM_MR_VERB_FQRNI: + /* nada, we drop FQRNIs on the floor */ + break; + case QM_MR_VERB_FQRN: + case QM_MR_VERB_FQRL: + /* Lookup in the retirement table */ + fq = table_find_fq(p, msg->fq.fqid); + BUG_ON(!fq); + fq_state_change(p, fq, msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, msg); + break; + case QM_MR_VERB_FQPN: + /* Parked */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(msg->fq.contextB); +#else + fq = (void *)(uintptr_t)msg->fq.contextB; +#endif + fq_state_change(p, fq, msg, verb); + if (fq->cb.fqs) + fq->cb.fqs(p, fq, msg); + break; + case QM_MR_VERB_DC_ERN: + /* DCP ERN */ + if (p->cb_dc_ern) + p->cb_dc_ern(p, msg); + else if (cb_dc_ern) + cb_dc_ern(p, msg); + else + pr_crit_once("Leaking DCP ERNs!\n"); + break; + default: + pr_crit("Invalid MR verb 0x%02x\n", verb); + } + } else { + /* Its a software ERN */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(msg->ern.tag); +#else + fq = (void *)(uintptr_t)msg->ern.tag; +#endif + fq->cb.ern(p, fq, msg); + } + num++; + qm_mr_next(&p->p); + goto mr_loop; +mr_done: + qm_mr_cci_consume(&p->p, num); + } + /* + * QM_PIRQ_CSCI has already been cleared, as part of its specific + * processing. If that interrupt source has meanwhile been re-asserted, + * we mustn't clear it here (or in the top-level interrupt handler). + */ + return is & (QM_PIRQ_EQCI | QM_PIRQ_EQRI | QM_PIRQ_MRI); +} + +/* remove some slowish-path stuff from the "fast path" and make sure it isn't + * inlined. */ +static noinline void clear_vdqcr(struct qman_portal *p, struct qman_fq *fq) +{ + p->vdqcr_owned = NULL; + FQLOCK(fq); + fq_clear(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + wake_up(&affine_queue); +} + +/* Look: no locks, no irq_save()s, no preempt_disable()s! :-) The only states + * that would conflict with other things if they ran at the same time on the + * same cpu are; + * + * (i) setting/clearing vdqcr_owned, and + * (ii) clearing the NE (Not Empty) flag. + * + * Both are safe. Because; + * + * (i) this clearing can only occur after qman_volatile_dequeue() has set the + * vdqcr_owned field (which it does before setting VDQCR), and + * qman_volatile_dequeue() blocks interrupts and preemption while this is + * done so that we can't interfere. + * (ii) the NE flag is only cleared after qman_retire_fq() has set it, and as + * with (i) that API prevents us from interfering until it's safe. + * + * The good thing is that qman_volatile_dequeue() and qman_retire_fq() run far + * less frequently (ie. per-FQ) than __poll_portal_fast() does, so the nett + * advantage comes from this function not having to "lock" anything at all. + * + * Note also that the callbacks are invoked at points which are safe against the + * above potential conflicts, but that this function itself is not re-entrant + * (this is because the function tracks one end of each FIFO in the portal and + * we do *not* want to lock that). So the consequence is that it is safe for + * user callbacks to call into any QMan API *except* qman_poll() (as that's the + * sole API that could be invoking the callback through this function). + */ +static inline unsigned int __poll_portal_fast(struct qman_portal *p, + unsigned int poll_limit) +{ + const struct qm_dqrr_entry *dq; + struct qman_fq *fq; + enum qman_cb_dqrr_result res; + unsigned int limit = 0; + +loop: + qm_dqrr_pvb_update(&p->p); + dq = qm_dqrr_current(&p->p); + if (!dq) + goto done; + if (dq->stat & QM_DQRR_STAT_UNSCHEDULED) { + /* VDQCR: don't trust contextB as the FQ may have been + * configured for h/w consumption and we're draining it + * post-retirement. */ + fq = p->vdqcr_owned; + /* We only set QMAN_FQ_STATE_NE when retiring, so we only need + * to check for clearing it when doing volatile dequeues. It's + * one less thing to check in the critical path (SDQCR). */ + if (dq->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_clear(fq, QMAN_FQ_STATE_NE); + /* this is duplicated from the SDQCR code, but we have stuff to + * do before *and* after this callback, and we don't want + * multiple if()s in the critical path (SDQCR). */ + res = fq->cb.dqrr(p, fq, dq); + if (res == qman_cb_dqrr_stop) + goto done; + /* Check for VDQCR completion */ + if (dq->stat & QM_DQRR_STAT_DQCR_EXPIRED) + clear_vdqcr(p, fq); + } else { + /* SDQCR: contextB points to the FQ */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + fq = get_fq_table_entry(dq->contextB); +#else + fq = (void *)(uintptr_t)dq->contextB; +#endif + /* Now let the callback do its stuff */ + res = fq->cb.dqrr(p, fq, dq); + /* The callback can request that we exit without consuming this + * entry nor advancing; */ + if (res == qman_cb_dqrr_stop) + goto done; + } + /* Interpret 'dq' from a driver perspective. */ + /* Parking isn't possible unless HELDACTIVE was set. NB, + * FORCEELIGIBLE implies HELDACTIVE, so we only need to + * check for HELDACTIVE to cover both. */ + DPA_ASSERT((dq->stat & QM_DQRR_STAT_FQ_HELDACTIVE) || + (res != qman_cb_dqrr_park)); + /* Defer just means "skip it, I'll consume it myself later on" */ + if (res != qman_cb_dqrr_defer) + qm_dqrr_cdc_consume_1ptr(&p->p, dq, (res == qman_cb_dqrr_park)); + /* Move forward */ + qm_dqrr_next(&p->p); + /* Entry processed and consumed, increment our counter. The callback can + * request that we exit after consuming the entry, and we also exit if + * we reach our processing limit, so loop back only if neither of these + * conditions is met. */ + if ((++limit < poll_limit) && (res != qman_cb_dqrr_consume_stop)) + goto loop; +done: + return limit; +} + +u32 qman_irqsource_get(void) +{ + /* "irqsource" and "poll" APIs mustn't redirect when sharing, they + * should shut the user out if they are not the primary CPU hosting the + * portal. That's why we use the "raw" interface. */ + struct qman_portal *p = get_raw_affine_portal(); + u32 ret = p->irq_sources & QM_PIRQ_VISIBLE; + + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_irqsource_get); + +int qman_p_irqsource_add(struct qman_portal *p, u32 bits __maybe_unused) +{ + __maybe_unused unsigned long irqflags; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (p->sharing_redirect) + return -EINVAL; +#endif /* __rtems__ */ +#endif + PORTAL_IRQ_LOCK(p, irqflags); + set_bits(bits & QM_PIRQ_VISIBLE, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + PORTAL_IRQ_UNLOCK(p, irqflags); + return 0; +} +EXPORT_SYMBOL(qman_p_irqsource_add); + +int qman_irqsource_add(u32 bits __maybe_unused) +{ + struct qman_portal *p = get_raw_affine_portal(); + int ret; + + ret = qman_p_irqsource_add(p, bits); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_irqsource_add); + +int qman_p_irqsource_remove(struct qman_portal *p, u32 bits) +{ + __maybe_unused unsigned long irqflags; + u32 ier; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (p->sharing_redirect) { + put_affine_portal(); + return -EINVAL; + } +#endif /* __rtems__ */ +#endif + /* Our interrupt handler only processes+clears status register bits that + * are in p->irq_sources. As we're trimming that mask, if one of them + * were to assert in the status register just before we remove it from + * the enable register, there would be an interrupt-storm when we + * release the IRQ lock. So we wait for the enable register update to + * take effect in h/w (by reading it back) and then clear all other bits + * in the status register. Ie. we clear them from ISR once it's certain + * IER won't allow them to reassert. */ + PORTAL_IRQ_LOCK(p, irqflags); + bits &= QM_PIRQ_VISIBLE; + clear_bits(bits, &p->irq_sources); + qm_isr_enable_write(&p->p, p->irq_sources); + ier = qm_isr_enable_read(&p->p); + /* Using "~ier" (rather than "bits" or "~p->irq_sources") creates a + * data-dependency, ie. to protect against re-ordering. */ + qm_isr_status_clear(&p->p, ~ier); + PORTAL_IRQ_UNLOCK(p, irqflags); + return 0; +} +EXPORT_SYMBOL(qman_p_irqsource_remove); + +int qman_irqsource_remove(u32 bits) +{ + struct qman_portal *p = get_raw_affine_portal(); + int ret; + + ret = qman_p_irqsource_remove(p, bits); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_irqsource_remove); + +#ifndef __rtems__ +const cpumask_t *qman_affine_cpus(void) +{ + return &affine_mask; +} +EXPORT_SYMBOL(qman_affine_cpus); + +u16 qman_affine_channel(int cpu) +{ + if (cpu < 0) { + struct qman_portal *portal = get_raw_affine_portal(); + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + BUG_ON(portal->sharing_redirect); +#endif /* __rtems__ */ +#endif + cpu = portal->config->public_cfg.cpu; + put_affine_portal(); + } + BUG_ON(!cpumask_test_cpu(cpu, &affine_mask)); + return affine_channels[cpu]; +} +EXPORT_SYMBOL(qman_affine_channel); +#endif /* __rtems__ */ + +void *qman_get_affine_portal(int cpu) +{ +#ifndef __rtems__ + return affine_portals[cpu]; +#else /* __rtems__ */ + return &per_cpu(qman_affine_portal, cpu); +#endif /* __rtems__ */ +} +EXPORT_SYMBOL(qman_get_affine_portal); + +int qman_p_poll_dqrr(struct qman_portal *p, unsigned int limit) +{ + int ret; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (unlikely(p->sharing_redirect)) + ret = -EINVAL; + else +#endif /* __rtems__ */ +#endif + { + BUG_ON(p->irq_sources & QM_PIRQ_DQRI); + ret = __poll_portal_fast(p, limit); + } + return ret; +} +EXPORT_SYMBOL(qman_p_poll_dqrr); + +int qman_poll_dqrr(unsigned int limit) +{ + struct qman_portal *p = get_poll_portal(); + int ret; + + ret = qman_p_poll_dqrr(p, limit); + put_poll_portal(); + return ret; +} +EXPORT_SYMBOL(qman_poll_dqrr); + +u32 qman_p_poll_slow(struct qman_portal *p) +{ + u32 ret; + +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (unlikely(p->sharing_redirect)) + ret = (u32)-1; + else +#endif /* __rtems__ */ +#endif + { + u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources; + + ret = __poll_portal_slow(p, is); + qm_isr_status_clear(&p->p, ret); + } + return ret; +} +EXPORT_SYMBOL(qman_p_poll_slow); + +u32 qman_poll_slow(void) +{ + struct qman_portal *p = get_poll_portal(); + u32 ret; + + ret = qman_p_poll_slow(p); + put_poll_portal(); + return ret; +} +EXPORT_SYMBOL(qman_poll_slow); + +/* Legacy wrapper */ +void qman_p_poll(struct qman_portal *p) +{ +#ifdef FSL_DPA_PORTAL_SHARE +#ifndef __rtems__ + if (unlikely(p->sharing_redirect)) + return; +#endif /* __rtems__ */ +#endif + if ((~p->irq_sources) & QM_PIRQ_SLOW) { + if (!(p->slowpoll--)) { + u32 is = qm_isr_status_read(&p->p) & ~p->irq_sources; + u32 active = __poll_portal_slow(p, is); + + if (active) { + qm_isr_status_clear(&p->p, active); + p->slowpoll = SLOW_POLL_BUSY; + } else + p->slowpoll = SLOW_POLL_IDLE; + } + } + if ((~p->irq_sources) & QM_PIRQ_DQRI) + __poll_portal_fast(p, QMAN_POLL_LIMIT); +} +EXPORT_SYMBOL(qman_p_poll); + +void qman_poll(void) +{ + struct qman_portal *p = get_poll_portal(); + + qman_p_poll(p); + put_poll_portal(); +} +EXPORT_SYMBOL(qman_poll); + +void qman_p_stop_dequeues(struct qman_portal *p) +{ + qman_stop_dequeues_ex(p); +} +EXPORT_SYMBOL(qman_p_stop_dequeues); + +void qman_stop_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + + qman_p_stop_dequeues(p); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_stop_dequeues); + +void qman_p_start_dequeues(struct qman_portal *p) +{ + unsigned long irqflags __maybe_unused; + + PORTAL_IRQ_LOCK(p, irqflags); + DPA_ASSERT(p->dqrr_disable_ref > 0); + if (!(--p->dqrr_disable_ref)) + qm_dqrr_set_maxfill(&p->p, DQRR_MAXFILL); + PORTAL_IRQ_UNLOCK(p, irqflags); +} +EXPORT_SYMBOL(qman_p_start_dequeues); + +void qman_start_dequeues(void) +{ + struct qman_portal *p = get_affine_portal(); + + qman_p_start_dequeues(p); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_start_dequeues); + +void qman_p_static_dequeue_add(struct qman_portal *p, u32 pools) +{ + unsigned long irqflags __maybe_unused; + + PORTAL_IRQ_LOCK(p, irqflags); + pools &= p->config->public_cfg.pools; + p->sdqcr |= pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); + PORTAL_IRQ_UNLOCK(p, irqflags); +} +EXPORT_SYMBOL(qman_p_static_dequeue_add); + +void qman_static_dequeue_add(u32 pools) +{ + struct qman_portal *p = get_affine_portal(); + + qman_p_static_dequeue_add(p, pools); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_static_dequeue_add); + +void qman_p_static_dequeue_del(struct qman_portal *p, u32 pools) +{ + unsigned long irqflags __maybe_unused; + + PORTAL_IRQ_LOCK(p, irqflags); + pools &= p->config->public_cfg.pools; + p->sdqcr &= ~pools; + qm_dqrr_sdqcr_set(&p->p, p->sdqcr); + PORTAL_IRQ_UNLOCK(p, irqflags); +} +EXPORT_SYMBOL(qman_p_static_dequeue_del); + +void qman_static_dequeue_del(u32 pools) +{ + struct qman_portal *p = get_affine_portal(); + + qman_p_static_dequeue_del(p, pools); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_static_dequeue_del); + +u32 qman_p_static_dequeue_get(struct qman_portal *p) +{ + return p->sdqcr; +} +EXPORT_SYMBOL(qman_p_static_dequeue_get); + +u32 qman_static_dequeue_get(void) +{ + struct qman_portal *p = get_affine_portal(); + u32 ret = qman_p_static_dequeue_get(p); + + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_static_dequeue_get); + +void qman_p_dca(struct qman_portal *p, struct qm_dqrr_entry *dq, + int park_request) +{ + qm_dqrr_cdc_consume_1ptr(&p->p, dq, park_request); +} +EXPORT_SYMBOL(qman_p_dca); + +void qman_dca(struct qm_dqrr_entry *dq, int park_request) +{ + struct qman_portal *p = get_affine_portal(); + + qman_p_dca(p, dq, park_request); + put_affine_portal(); +} +EXPORT_SYMBOL(qman_dca); + +/* Frame queue API */ + +static const char *mcr_result_str(u8 result) +{ + switch (result) { + case QM_MCR_RESULT_NULL: + return "QM_MCR_RESULT_NULL"; + case QM_MCR_RESULT_OK: + return "QM_MCR_RESULT_OK"; + case QM_MCR_RESULT_ERR_FQID: + return "QM_MCR_RESULT_ERR_FQID"; + case QM_MCR_RESULT_ERR_FQSTATE: + return "QM_MCR_RESULT_ERR_FQSTATE"; + case QM_MCR_RESULT_ERR_NOTEMPTY: + return "QM_MCR_RESULT_ERR_NOTEMPTY"; + case QM_MCR_RESULT_PENDING: + return "QM_MCR_RESULT_PENDING"; + case QM_MCR_RESULT_ERR_BADCOMMAND: + return "QM_MCR_RESULT_ERR_BADCOMMAND"; + } + return ""; +} + +int qman_create_fq(u32 fqid, u32 flags, struct qman_fq *fq) +{ + struct qm_fqd fqd; + struct qm_mcr_queryfq_np np; + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) { + int ret = qman_alloc_fqid(&fqid); + + if (ret) + return ret; + } + spin_lock_init(&fq->fqlock); + fq->fqid = fqid; + fq->flags = flags; + fq->state = qman_fq_state_oos; + fq->cgr_groupid = 0; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + if (unlikely(find_empty_fq_table_entry(&fq->key, fq))) + return -ENOMEM; +#endif + if (!(flags & QMAN_FQ_FLAG_AS_IS) || (flags & QMAN_FQ_FLAG_NO_MODIFY)) + return 0; + /* Everything else is AS_IS support */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + fqd = mcr->queryfq.fqd; + mcc = qm_mc_start(&p->p); + mcc->queryfq_np.fqid = fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYFQ_NP); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("QUERYFQ_NP failed: %s\n", mcr_result_str(mcr->result)); + goto err; + } + np = mcr->queryfq_np; + /* Phew, have queryfq and queryfq_np results, stitch together + * the FQ object from those. */ + fq->cgr_groupid = fqd.cgid; + switch (np.state & QM_MCR_NP_STATE_MASK) { + case QM_MCR_NP_STATE_OOS: + break; + case QM_MCR_NP_STATE_RETIRED: + fq->state = qman_fq_state_retired; + if (np.frm_cnt) + fq_set(fq, QMAN_FQ_STATE_NE); + break; + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + fq->state = qman_fq_state_sched; + if (np.state & QM_MCR_NP_STATE_R) + fq_set(fq, QMAN_FQ_STATE_CHANGING); + break; + case QM_MCR_NP_STATE_PARKED: + fq->state = qman_fq_state_parked; + break; + default: + DPA_ASSERT(NULL == "invalid FQ state"); + } + if (fqd.fq_ctrl & QM_FQCTRL_CGE) + fq->state |= QMAN_FQ_STATE_CGR_EN; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return 0; +err: + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (flags & QMAN_FQ_FLAG_DYNAMIC_FQID) + qman_release_fqid(fqid); + return -EIO; +} +EXPORT_SYMBOL(qman_create_fq); + +void qman_destroy_fq(struct qman_fq *fq, u32 flags __maybe_unused) +{ + + /* We don't need to lock the FQ as it is a pre-condition that the FQ be + * quiesced. Instead, run some checks. */ + switch (fq->state) { + case qman_fq_state_parked: + DPA_ASSERT(flags & QMAN_FQ_DESTROY_PARKED); + case qman_fq_state_oos: + if (fq_isset(fq, QMAN_FQ_FLAG_DYNAMIC_FQID)) + qman_release_fqid(fq->fqid); +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + clear_fq_table_entry(fq->key); +#endif + return; + default: + break; + } + DPA_ASSERT(NULL == "qman_free_fq() on unquiesced FQ!"); +} +EXPORT_SYMBOL(qman_destroy_fq); + +u32 qman_fq_fqid(struct qman_fq *fq) +{ + return fq->fqid; +} +EXPORT_SYMBOL(qman_fq_fqid); + +void qman_fq_state(struct qman_fq *fq, enum qman_fq_state *state, u32 *flags) +{ + if (state) + *state = fq->state; + if (flags) + *flags = fq->flags; +} +EXPORT_SYMBOL(qman_fq_state); + +int qman_init_fq(struct qman_fq *fq, u32 flags, struct qm_mcc_initfq *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + u8 res, myverb = (flags & QMAN_INITFQ_FLAG_SCHED) ? + QM_MCC_VERB_INITFQ_SCHED : QM_MCC_VERB_INITFQ_PARKED; + + if ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + if (opts && (opts->we_mask & QM_INITFQ_WE_OAC)) { + /* And can't be set at the same time as TDTHRESH */ + if (opts->we_mask & QM_INITFQ_WE_TDTHRESH) + return -EINVAL; + } + /* Issue an INITFQ_[PARKED|SCHED] management command */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + ((fq->state != qman_fq_state_oos) && + (fq->state != qman_fq_state_parked)))) { + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return -EBUSY; + } + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initfq = *opts; + mcc->initfq.fqid = fq->fqid; + mcc->initfq.count = 0; + /* If the FQ does *not* have the TO_DCPORTAL flag, contextB is set as a + * demux pointer. Otherwise, the caller-provided value is allowed to + * stand, don't overwrite it. */ + if (fq_isclear(fq, QMAN_FQ_FLAG_TO_DCPORTAL)) { + dma_addr_t phys_fq; + + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTB; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + mcc->initfq.fqd.context_b = fq->key; +#else + mcc->initfq.fqd.context_b = (u32)(uintptr_t)fq; +#endif + /* and the physical address - NB, if the user wasn't trying to + * set CONTEXTA, clear the stashing settings. */ + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_CONTEXTA)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_CONTEXTA; + memset(&mcc->initfq.fqd.context_a, 0, + sizeof(mcc->initfq.fqd.context_a)); + } else { +#ifndef __rtems__ + phys_fq = dma_map_single(&p->pdev->dev, fq, sizeof(*fq), + DMA_TO_DEVICE); +#else /* __rtems__ */ + phys_fq = (dma_addr_t)fq; +#endif /* __rtems__ */ + qm_fqd_stashing_set64(&mcc->initfq.fqd, phys_fq); + } + } + if (flags & QMAN_INITFQ_FLAG_LOCAL) { + mcc->initfq.fqd.dest.channel = p->config->public_cfg.channel; + if (!(mcc->initfq.we_mask & QM_INITFQ_WE_DESTWQ)) { + mcc->initfq.we_mask |= QM_INITFQ_WE_DESTWQ; + mcc->initfq.fqd.dest.wq = 4; + } + } + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return -EIO; + } + if (opts) { + if (opts->we_mask & QM_INITFQ_WE_FQCTRL) { + if (opts->fqd.fq_ctrl & QM_FQCTRL_CGE) + fq_set(fq, QMAN_FQ_STATE_CGR_EN); + else + fq_clear(fq, QMAN_FQ_STATE_CGR_EN); + } + if (opts->we_mask & QM_INITFQ_WE_CGID) + fq->cgr_groupid = opts->fqd.cgid; + } + fq->state = (flags & QMAN_INITFQ_FLAG_SCHED) ? + qman_fq_state_sched : qman_fq_state_parked; + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return 0; +} +EXPORT_SYMBOL(qman_init_fq); + +int qman_schedule_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_parked) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTERFQ_SCHED management command */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state != qman_fq_state_parked))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_SCHED); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_SCHED); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_sched; +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_schedule_fq); + +int qman_retire_fq(struct qman_fq *fq, u32 *flags) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int rval; + u8 res; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_sched)) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_oos))) { + rval = -EBUSY; + goto out; + } + rval = table_push_fq(p, fq); + if (rval) + goto out; + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_RETIRE); + res = mcr->result; + /* "Elegant" would be to treat OK/PENDING the same way; set CHANGING, + * and defer the flags until FQRNI or FQRN (respectively) show up. But + * "Friendly" is to process OK immediately, and not set CHANGING. We do + * friendly, otherwise the caller doesn't necessarily have a fully + * "retired" FQ on return even if the retirement was immediate. However + * this does mean some code duplication between here and + * fq_state_change(). */ + if (likely(res == QM_MCR_RESULT_OK)) { + rval = 0; + /* Process 'fq' right away, we'll ignore FQRNI */ + if (mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) + fq_set(fq, QMAN_FQ_STATE_NE); + if (mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT) + fq_set(fq, QMAN_FQ_STATE_ORL); + else + table_del_fq(p, fq); + if (flags) + *flags = fq->flags; + fq->state = qman_fq_state_retired; + if (fq->cb.fqs) { + /* Another issue with supporting "immediate" retirement + * is that we're forced to drop FQRNIs, because by the + * time they're seen it may already be "too late" (the + * fq may have been OOS'd and free()'d already). But if + * the upper layer wants a callback whether it's + * immediate or not, we have to fake a "MR" entry to + * look like an FQRNI... */ + struct qm_mr_entry msg; + + msg.verb = QM_MR_VERB_FQRNI; + msg.fq.fqs = mcr->alterfq.fqs; + msg.fq.fqid = fq->fqid; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + msg.fq.contextB = fq->key; +#else + msg.fq.contextB = (u32)(uintptr_t)fq; +#endif + fq->cb.fqs(p, fq, &msg); + } + } else if (res == QM_MCR_RESULT_PENDING) { + rval = 1; + fq_set(fq, QMAN_FQ_STATE_CHANGING); + } else { + rval = -EIO; + table_del_fq(p, fq); + } +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return rval; +} +EXPORT_SYMBOL(qman_retire_fq); + +int qman_oos_fq(struct qman_fq *fq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret = 0; + u8 res; + + if (fq->state != qman_fq_state_retired) + return -EINVAL; +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_BLOCKOOS)) || + (fq->state != qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_ALTER_OOS); + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } + fq->state = qman_fq_state_oos; +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_oos_fq); + +int qman_fq_flow_control(struct qman_fq *fq, int xon) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret = 0; + u8 res; + u8 myverb; + + if ((fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired) || + (fq->state == qman_fq_state_parked)) + return -EINVAL; + +#ifdef CONFIG_FSL_DPA_CHECKING + if (unlikely(fq_isset(fq, QMAN_FQ_FLAG_NO_MODIFY))) + return -EINVAL; +#endif + /* Issue a ALTER_FQXON or ALTER_FQXOFF management command */ + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + FQLOCK(fq); + if (unlikely((fq_isset(fq, QMAN_FQ_STATE_CHANGING)) || + (fq->state == qman_fq_state_parked) || + (fq->state == qman_fq_state_oos) || + (fq->state == qman_fq_state_retired))) { + ret = -EBUSY; + goto out; + } + mcc = qm_mc_start(&p->p); + mcc->alterfq.fqid = fq->fqid; + mcc->alterfq.count = 0; + myverb = xon ? QM_MCC_VERB_ALTER_FQXON : QM_MCC_VERB_ALTER_FQXOFF; + + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + + res = mcr->result; + if (res != QM_MCR_RESULT_OK) { + ret = -EIO; + goto out; + } +out: + FQUNLOCK(fq); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_fq_flow_control); + +int qman_query_fq(struct qman_fq *fq, struct qm_fqd *fqd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *fqd = mcr->queryfq.fqd; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} +EXPORT_SYMBOL(qman_query_fq); + +int qman_query_fq_np(struct qman_fq *fq, struct qm_mcr_queryfq_np *np) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->queryfq.fqid = fq->fqid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *np = mcr->queryfq_np; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res == QM_MCR_RESULT_ERR_FQID) + return -ERANGE; + else if (res != QM_MCR_RESULT_OK) + return -EIO; + return 0; +} +EXPORT_SYMBOL(qman_query_fq_np); + +int qman_query_wq(u8 query_dedicated, struct qm_mcr_querywq *wq) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res, myverb; + + PORTAL_IRQ_LOCK(p, irqflags); + myverb = (query_dedicated) ? QM_MCR_VERB_QUERYWQ_DEDICATED : + QM_MCR_VERB_QUERYWQ; + mcc = qm_mc_start(&p->p); + mcc->querywq.channel.id = wq->channel.id; + qm_mc_commit(&p->p, myverb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == myverb); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *wq = mcr->querywq; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERYWQ failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} +EXPORT_SYMBOL(qman_query_wq); + +int qman_testwrite_cgr(struct qman_cgr *cgr, u64 i_bcnt, + struct qm_mcr_cgrtestwrite *result) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->cgrtestwrite.cgid = cgr->cgrid; + mcc->cgrtestwrite.i_bcnt_hi = (u8)(i_bcnt >> 32); + mcc->cgrtestwrite.i_bcnt_lo = (u32)i_bcnt; + qm_mc_commit(&p->p, QM_MCC_VERB_CGRTESTWRITE); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_CGRTESTWRITE); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *result = mcr->cgrtestwrite; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) { + pr_err("CGR TEST WRITE failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} +EXPORT_SYMBOL(qman_testwrite_cgr); + +int qman_query_cgr(struct qman_cgr *cgr, struct qm_mcr_querycgr *cgrd) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + mcc->querycgr.cgid = cgr->cgrid; + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCGR); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCC_VERB_QUERYCGR); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *cgrd = mcr->querycgr; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERY_CGR failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} +EXPORT_SYMBOL(qman_query_cgr); + +int qman_query_congestion(struct qm_mcr_querycongestion *congestion) +{ + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + + PORTAL_IRQ_LOCK(p, irqflags); + qm_mc_start(&p->p); + qm_mc_commit(&p->p, QM_MCC_VERB_QUERYCONGESTION); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCC_VERB_QUERYCONGESTION); + res = mcr->result; + if (res == QM_MCR_RESULT_OK) + *congestion = mcr->querycongestion; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + if (res != QM_MCR_RESULT_OK) { + pr_err("QUERY_CONGESTION failed: %s\n", mcr_result_str(res)); + return -EIO; + } + return 0; +} +EXPORT_SYMBOL(qman_query_congestion); + +/* internal function used as a wait_event() expression */ +static int set_p_vdqcr(struct qman_portal *p, struct qman_fq *fq, u32 vdqcr) +{ + unsigned long irqflags __maybe_unused; + int ret = -EBUSY; + + PORTAL_IRQ_LOCK(p, irqflags); + if (!p->vdqcr_owned) { + FQLOCK(fq); + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + goto escape; + fq_set(fq, QMAN_FQ_STATE_VDQCR); + FQUNLOCK(fq); + p->vdqcr_owned = fq; + ret = 0; + } +escape: + PORTAL_IRQ_UNLOCK(p, irqflags); + if (!ret) + qm_dqrr_vdqcr_set(&p->p, vdqcr); + return ret; +} + +static int set_vdqcr(struct qman_portal **p, struct qman_fq *fq, u32 vdqcr) +{ + int ret; + + *p = get_affine_portal(); + ret = set_p_vdqcr(*p, fq, vdqcr); + put_affine_portal(); + return ret; +} + +#ifdef FSL_DPA_CAN_WAIT +static int wait_p_vdqcr_start(struct qman_portal *p, struct qman_fq *fq, + u32 vdqcr, u32 flags) +{ + int ret = 0; + +#ifndef __rtems__ + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + ret = wait_event_interruptible(affine_queue, + !(ret = set_p_vdqcr(p, fq, vdqcr))); + else +#endif /* __rtems__ */ + wait_event(affine_queue, !(ret = set_p_vdqcr(p, fq, vdqcr))); + return ret; +} + +static int wait_vdqcr_start(struct qman_portal **p, struct qman_fq *fq, + u32 vdqcr, u32 flags) +{ + int ret = 0; + +#ifndef __rtems__ + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + ret = wait_event_interruptible(affine_queue, + !(ret = set_vdqcr(p, fq, vdqcr))); + else +#endif /* __rtems__ */ + wait_event(affine_queue, !(ret = set_vdqcr(p, fq, vdqcr))); + return ret; +} +#endif + +int qman_p_volatile_dequeue(struct qman_portal *p, struct qman_fq *fq, + u32 flags __maybe_unused, u32 vdqcr) +{ + int ret; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) + return -EINVAL; + if (vdqcr & QM_VDQCR_FQID_MASK) + return -EINVAL; + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + return -EBUSY; + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_WAIT) + ret = wait_p_vdqcr_start(p, fq, vdqcr, flags); + else +#endif + ret = set_p_vdqcr(p, fq, vdqcr); + if (ret) + return ret; + /* VDQCR is set */ +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_FINISH) { +#ifndef __rtems__ + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + /* NB: don't propagate any error - the caller wouldn't + * know whether the VDQCR was issued or not. A signal + * could arrive after returning anyway, so the caller + * can check signal_pending() if that's an issue. */ + wait_event_interruptible(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_volatile_dequeue); + +int qman_volatile_dequeue(struct qman_fq *fq, u32 flags __maybe_unused, + u32 vdqcr) +{ + struct qman_portal *p; + int ret; + + if ((fq->state != qman_fq_state_parked) && + (fq->state != qman_fq_state_retired)) + return -EINVAL; + if (vdqcr & QM_VDQCR_FQID_MASK) + return -EINVAL; + if (fq_isset(fq, QMAN_FQ_STATE_VDQCR)) + return -EBUSY; + vdqcr = (vdqcr & ~QM_VDQCR_FQID_MASK) | fq->fqid; +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_WAIT) + ret = wait_vdqcr_start(&p, fq, vdqcr, flags); + else +#endif + ret = set_vdqcr(&p, fq, vdqcr); + if (ret) + return ret; + /* VDQCR is set */ +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_VOLATILE_FLAG_FINISH) { +#ifndef __rtems__ + if (flags & QMAN_VOLATILE_FLAG_WAIT_INT) + /* NB: don't propagate any error - the caller wouldn't + * know whether the VDQCR was issued or not. A signal + * could arrive after returning anyway, so the caller + * can check signal_pending() if that's an issue. */ + wait_event_interruptible(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, + !fq_isset(fq, QMAN_FQ_STATE_VDQCR)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_volatile_dequeue); + +static noinline void update_eqcr_ci(struct qman_portal *p, u8 avail) +{ + if (avail) + qm_eqcr_cce_prefetch(&p->p); + else + qm_eqcr_cce_update(&p->p); +} + +int qman_eqcr_is_empty(void) +{ + unsigned long irqflags __maybe_unused; + struct qman_portal *p = get_affine_portal(); + u8 avail; + + PORTAL_IRQ_LOCK(p, irqflags); + update_eqcr_ci(p, 0); + avail = qm_eqcr_get_fill(&p->p); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return avail == 0; +} +EXPORT_SYMBOL(qman_eqcr_is_empty); + +void qman_set_dc_ern(qman_cb_dc_ern handler, int affine) +{ + if (affine) { + unsigned long irqflags __maybe_unused; + struct qman_portal *p = get_affine_portal(); + + PORTAL_IRQ_LOCK(p, irqflags); + p->cb_dc_ern = handler; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + } else + cb_dc_ern = handler; +} +EXPORT_SYMBOL(qman_set_dc_ern); + +static inline struct qm_eqcr_entry *try_p_eq_start(struct qman_portal *p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + u8 avail; + + PORTAL_IRQ_LOCK(p, (*irqflags)); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { + if (p->eqci_owned) { + PORTAL_IRQ_UNLOCK(p, (*irqflags)); + return NULL; + } + p->eqci_owned = fq; + } +#endif + if (p->use_eqcr_ci_stashing) { + /* + * The stashing case is easy, only update if we need to in + * order to try and liberate ring entries. + */ + eq = qm_eqcr_start_stash(&p->p); + } else { + /* + * The non-stashing case is harder, need to prefetch ahead of + * time. + */ + avail = qm_eqcr_get_avail(&p->p); + if (avail < 2) + update_eqcr_ci(p, avail); + eq = qm_eqcr_start_no_stash(&p->p); + } + + if (unlikely(!eq)) { +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) + p->eqci_owned = NULL; +#endif + PORTAL_IRQ_UNLOCK(p, (*irqflags)); + return NULL; + } + if (flags & QMAN_ENQUEUE_FLAG_DCA) + eq->dca = QM_EQCR_DCA_ENABLE | + ((flags & QMAN_ENQUEUE_FLAG_DCA_PARK) ? + QM_EQCR_DCA_PARK : 0) | + ((flags >> 8) & QM_EQCR_DCA_IDXMASK); + eq->fqid = fq->fqid; +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + eq->tag = fq->key; +#else + eq->tag = (u32)(uintptr_t)fq; +#endif + eq->fd = *fd; + return eq; +} + +static inline struct qm_eqcr_entry *try_eq_start(struct qman_portal **p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + + *p = get_affine_portal(); + eq = try_p_eq_start(*p, irqflags, fq, fd, flags); + if (!eq) + put_affine_portal(); + return eq; +} + +#ifdef FSL_DPA_CAN_WAIT +static noinline struct qm_eqcr_entry *__wait_eq_start(struct qman_portal **p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq = try_eq_start(p, irqflags, fq, fd, flags); + + if (!eq) + qm_eqcr_set_ithresh(&(*p)->p, EQCR_ITHRESH); + return eq; +} +static noinline struct qm_eqcr_entry *wait_eq_start(struct qman_portal **p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (eq = __wait_eq_start(p, irqflags, fq, fd, flags))); + else +#endif /* __rtems__ */ + wait_event(affine_queue, + (eq = __wait_eq_start(p, irqflags, fq, fd, flags))); + return eq; +} +static noinline struct qm_eqcr_entry *__wait_p_eq_start(struct qman_portal *p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq = try_p_eq_start(p, irqflags, fq, fd, flags); + + if (!eq) + qm_eqcr_set_ithresh(&p->p, EQCR_ITHRESH); + return eq; +} +static noinline struct qm_eqcr_entry *wait_p_eq_start(struct qman_portal *p, + unsigned long *irqflags __maybe_unused, + struct qman_fq *fq, + const struct qm_fd *fd, + u32 flags) +{ + struct qm_eqcr_entry *eq; + +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (eq = __wait_p_eq_start(p, irqflags, fq, fd, flags))); + else +#endif /* __rtems__ */ + wait_event(affine_queue, + (eq = __wait_p_eq_start(p, irqflags, fq, fd, flags))); + return eq; +} +#endif + +int qman_p_enqueue(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags) +{ + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_p_eq_start(p, &irqflags, fq, fd, flags); + else +#endif + eq = try_p_eq_start(p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_enqueue); + +int qman_enqueue(struct qman_fq *fq, const struct qm_fd *fd, u32 flags) +{ + struct qman_portal *p; + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_eq_start(&p, &irqflags, fq, fd, flags); + else +#endif + eq = try_eq_start(&p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_enqueue); + +int qman_p_enqueue_orp(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum) +{ + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_p_eq_start(p, &irqflags, fq, fd, flags); + else +#endif + eq = try_p_eq_start(p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Process ORP-specifics here */ + if (flags & QMAN_ENQUEUE_FLAG_NLIS) + orp_seqnum |= QM_EQCR_SEQNUM_NLIS; + else { + orp_seqnum &= ~QM_EQCR_SEQNUM_NLIS; + if (flags & QMAN_ENQUEUE_FLAG_NESN) + orp_seqnum |= QM_EQCR_SEQNUM_NESN; + else + /* No need to check 4 QMAN_ENQUEUE_FLAG_HOLE */ + orp_seqnum &= ~QM_EQCR_SEQNUM_NESN; + } + eq->seqnum = orp_seqnum; + eq->orp = orp->fqid; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_ORP | + ((flags & (QMAN_ENQUEUE_FLAG_HOLE | QMAN_ENQUEUE_FLAG_NESN)) ? + 0 : QM_EQCR_VERB_CMD_ENQUEUE) | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + PORTAL_IRQ_UNLOCK(p, irqflags); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_enqueue_orp); + +int qman_enqueue_orp(struct qman_fq *fq, const struct qm_fd *fd, u32 flags, + struct qman_fq *orp, u16 orp_seqnum) +{ + struct qman_portal *p; + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_eq_start(&p, &irqflags, fq, fd, flags); + else +#endif + eq = try_eq_start(&p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* Process ORP-specifics here */ + if (flags & QMAN_ENQUEUE_FLAG_NLIS) + orp_seqnum |= QM_EQCR_SEQNUM_NLIS; + else { + orp_seqnum &= ~QM_EQCR_SEQNUM_NLIS; + if (flags & QMAN_ENQUEUE_FLAG_NESN) + orp_seqnum |= QM_EQCR_SEQNUM_NESN; + else + /* No need to check 4 QMAN_ENQUEUE_FLAG_HOLE */ + orp_seqnum &= ~QM_EQCR_SEQNUM_NESN; + } + eq->seqnum = orp_seqnum; + eq->orp = orp->fqid; + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_ORP | + ((flags & (QMAN_ENQUEUE_FLAG_HOLE | QMAN_ENQUEUE_FLAG_NESN)) ? + 0 : QM_EQCR_VERB_CMD_ENQUEUE) | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_enqueue_orp); + +int qman_p_enqueue_precommit(struct qman_portal *p, struct qman_fq *fq, + const struct qm_fd *fd, u32 flags, + qman_cb_precommit cb, void *cb_arg) +{ + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_p_eq_start(p, &irqflags, fq, fd, flags); + else +#endif + eq = try_p_eq_start(p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* invoke user supplied callback function before writing commit verb */ + if (cb(cb_arg)) { + PORTAL_IRQ_UNLOCK(p, irqflags); + return -EINVAL; + } + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_p_enqueue_precommit); + +int qman_enqueue_precommit(struct qman_fq *fq, const struct qm_fd *fd, + u32 flags, qman_cb_precommit cb, void *cb_arg) +{ + struct qman_portal *p; + struct qm_eqcr_entry *eq; + unsigned long irqflags __maybe_unused; + +#ifdef FSL_DPA_CAN_WAIT + if (flags & QMAN_ENQUEUE_FLAG_WAIT) + eq = wait_eq_start(&p, &irqflags, fq, fd, flags); + else +#endif + eq = try_eq_start(&p, &irqflags, fq, fd, flags); + if (!eq) + return -EBUSY; + /* invoke user supplied callback function before writing commit verb */ + if (cb(cb_arg)) { + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return -EINVAL; + } + /* Note: QM_EQCR_VERB_INTERRUPT == QMAN_ENQUEUE_FLAG_WAIT_SYNC */ + qm_eqcr_pvb_commit(&p->p, QM_EQCR_VERB_CMD_ENQUEUE | + (flags & (QM_EQCR_VERB_COLOUR_MASK | QM_EQCR_VERB_INTERRUPT))); + /* Factor the below out, it's used from qman_enqueue_orp() too */ + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); +#ifdef FSL_DPA_CAN_WAIT_SYNC + if (unlikely((flags & QMAN_ENQUEUE_FLAG_WAIT) && + (flags & QMAN_ENQUEUE_FLAG_WAIT_SYNC))) { +#ifndef __rtems__ + if (flags & QMAN_ENQUEUE_FLAG_WAIT_INT) + wait_event_interruptible(affine_queue, + (p->eqci_owned != fq)); + else +#endif /* __rtems__ */ + wait_event(affine_queue, (p->eqci_owned != fq)); + } +#endif + return 0; +} +EXPORT_SYMBOL(qman_enqueue_precommit); + +int qman_modify_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + struct qman_portal *p = get_affine_portal(); + unsigned long irqflags __maybe_unused; + u8 res; + u8 verb = QM_MCC_VERB_MODIFYCGR; + + PORTAL_IRQ_LOCK(p, irqflags); + mcc = qm_mc_start(&p->p); + if (opts) + mcc->initcgr = *opts; + mcc->initcgr.cgid = cgr->cgrid; + if (flags & QMAN_CGR_FLAG_USE_INIT) + verb = QM_MCC_VERB_INITCGR; + qm_mc_commit(&p->p, verb); + while (!(mcr = qm_mc_result(&p->p))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == verb); + res = mcr->result; + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return (res == QM_MCR_RESULT_OK) ? 0 : -EIO; +} +EXPORT_SYMBOL(qman_modify_cgr); + +#define TARG_MASK(n) (0x80000000 >> (n->config->public_cfg.channel - \ + QM_CHANNEL_SWPORTAL0)) +#define PORTAL_IDX(n) (n->config->public_cfg.channel - QM_CHANNEL_SWPORTAL0) + +static u8 qman_cgr_cpus[__CGR_NUM]; + +int qman_create_cgr(struct qman_cgr *cgr, u32 flags, + struct qm_mcc_initcgr *opts) +{ + unsigned long irqflags __maybe_unused; + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret; + struct qman_portal *p; + + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + preempt_disable(); + p = get_affine_portal(); + qman_cgr_cpus[cgr->cgrid] = smp_processor_id(); + preempt_enable(); + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + cgr->chan = p->config->public_cfg.channel; + spin_lock_irqsave(&p->cgr_lock, irqflags); + + /* if no opts specified, just add it to the list */ + if (!opts) + goto add_list; + + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) + goto release_lock; + if (opts) + local_opts = *opts; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = + QM_CGR_TARG_UDP_CTRL_WRITE_BIT | PORTAL_IDX(p); + else + /* Overwrite TARG */ + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ | + TARG_MASK(p); + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + goto release_lock; +add_list: + list_add(&cgr->node, &p->cgr_cbs); + + /* Determine if newly added object requires its callback to be called */ + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* we can't go back, so proceed and return success, but screen + * and wail to the log file */ + pr_crit("CGR HW state partially modified\n"); + ret = 0; + goto release_lock; + } + if (cgr->cb && cgr_state.cgr.cscn_en && qman_cgrs_get(&p->cgrs[1], + cgr->cgrid)) + cgr->cb(p, cgr, 1); +release_lock: + spin_unlock_irqrestore(&p->cgr_lock, irqflags); + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_create_cgr); + +int qman_create_cgr_to_dcp(struct qman_cgr *cgr, u32 flags, u16 dcp_portal, + struct qm_mcc_initcgr *opts) +{ + unsigned long irqflags __maybe_unused; + struct qm_mcc_initcgr local_opts; + int ret; + + if ((qman_ip_rev & 0xFF00) < QMAN_REV30) { + pr_warn("This version doesn't support to send CSCN to DCP portal\n"); + return -EINVAL; + } + /* We have to check that the provided CGRID is within the limits of the + * data-structures, for obvious reasons. However we'll let h/w take + * care of determining whether it's within the limits of what exists on + * the SoC. + */ + if (cgr->cgrid >= __CGR_NUM) + return -EINVAL; + + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + if (opts) + local_opts = *opts; + + local_opts.cgr.cscn_targ_upd_ctrl = QM_CGR_TARG_UDP_CTRL_WRITE_BIT | + QM_CGR_TARG_UDP_CTRL_DCP | dcp_portal; + local_opts.we_mask |= QM_CGR_WE_CSCN_TARG; + + /* send init if flags indicate so */ + if (opts && (flags & QMAN_CGR_FLAG_USE_INIT)) + ret = qman_modify_cgr(cgr, QMAN_CGR_FLAG_USE_INIT, + &local_opts); + else + ret = qman_modify_cgr(cgr, 0, &local_opts); + + return ret; +} +EXPORT_SYMBOL(qman_create_cgr_to_dcp); + +int qman_delete_cgr(struct qman_cgr *cgr) +{ + unsigned long irqflags __maybe_unused; + struct qm_mcr_querycgr cgr_state; + struct qm_mcc_initcgr local_opts; + int ret = 0; + struct qman_cgr *i; + struct qman_portal *p = get_affine_portal(); + + if (cgr->chan != p->config->public_cfg.channel) { + pr_crit("Attempting to delete cgr from different portal " + "than it was create: create 0x%x, delete 0x%x\n", + cgr->chan, p->config->public_cfg.channel); + ret = -EINVAL; + goto put_portal; + } + memset(&local_opts, 0, sizeof(struct qm_mcc_initcgr)); + spin_lock_irqsave(&p->cgr_lock, irqflags); + list_del(&cgr->node); + /* + * If there are no other CGR objects for this CGRID in the list, update + * CSCN_TARG accordingly + */ + list_for_each_entry(i, &p->cgr_cbs, node) + if ((i->cgrid == cgr->cgrid) && i->cb) + goto release_lock; + ret = qman_query_cgr(cgr, &cgr_state); + if (ret) { + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); + goto release_lock; + } + /* Overwrite TARG */ + local_opts.we_mask = QM_CGR_WE_CSCN_TARG; + if ((qman_ip_rev & 0xFF00) >= QMAN_REV30) + local_opts.cgr.cscn_targ_upd_ctrl = PORTAL_IDX(p); + else + local_opts.cgr.cscn_targ = cgr_state.cgr.cscn_targ & + ~(TARG_MASK(p)); + ret = qman_modify_cgr(cgr, 0, &local_opts); + if (ret) + /* add back to the list */ + list_add(&cgr->node, &p->cgr_cbs); +release_lock: + spin_unlock_irqrestore(&p->cgr_lock, irqflags); +put_portal: + put_affine_portal(); + return ret; +} +EXPORT_SYMBOL(qman_delete_cgr); + +#ifndef __rtems__ +struct cgr_comp { + struct qman_cgr *cgr; + struct completion completion; +}; + +static int qman_delete_cgr_thread(void *p) +{ + struct cgr_comp *cgr_comp = (struct cgr_comp *)p; + int res; + + res = qman_delete_cgr((struct qman_cgr *)cgr_comp->cgr); + complete(&cgr_comp->completion); + + return res; +} + +void qman_delete_cgr_safe(struct qman_cgr *cgr) +{ + struct task_struct *thread; + struct cgr_comp cgr_comp; + + preempt_disable(); + if (qman_cgr_cpus[cgr->cgrid] != smp_processor_id()) { + init_completion(&cgr_comp.completion); + cgr_comp.cgr = cgr; + thread = kthread_create(qman_delete_cgr_thread, &cgr_comp, + "cgr_del"); + + if (likely(!IS_ERR(thread))) { + kthread_bind(thread, qman_cgr_cpus[cgr->cgrid]); + wake_up_process(thread); + wait_for_completion(&cgr_comp.completion); + preempt_enable(); + return; + } + } + qman_delete_cgr(cgr); + preempt_enable(); +} +EXPORT_SYMBOL(qman_delete_cgr_safe); +#endif /* __rtems__ */ + +int qman_set_wpm(int wpm_enable) +{ + return qm_set_wpm(wpm_enable); +} +EXPORT_SYMBOL(qman_set_wpm); + +int qman_get_wpm(int *wpm_enable) +{ + return qm_get_wpm(wpm_enable); +} +EXPORT_SYMBOL(qman_get_wpm); + + +/* Cleanup FQs */ +static int qm_shutdown_fq(struct qm_portal **portal, int portal_count, + u32 fqid) +{ + + struct qm_mc_command *mcc; + struct qm_mc_result *mcr; + u8 state; + int orl_empty, fq_empty, i, drain = 0; + u32 result; + u32 channel, wq; + + /* Determine the state of the FQID */ + mcc = qm_mc_start(portal[0]); + mcc->queryfq_np.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_QUERYFQ_NP); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ_NP); + state = mcr->queryfq_np.state & QM_MCR_NP_STATE_MASK; + if (state == QM_MCR_NP_STATE_OOS) + return 0; /* Already OOS, no need to do anymore checks */ + + /* Query which channel the FQ is using */ + mcc = qm_mc_start(portal[0]); + mcc->queryfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_QUERYFQ); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == QM_MCR_VERB_QUERYFQ); + + /* Need to store these since the MCR gets reused */ + channel = mcr->queryfq.fqd.dest.channel; + wq = mcr->queryfq.fqd.dest.wq; + + switch (state) { + case QM_MCR_NP_STATE_TEN_SCHED: + case QM_MCR_NP_STATE_TRU_SCHED: + case QM_MCR_NP_STATE_ACTIVE: + case QM_MCR_NP_STATE_PARKED: + orl_empty = 0; + mcc = qm_mc_start(portal[0]); + mcc->alterfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_ALTER_RETIRE); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_RETIRE); + result = mcr->result; /* Make a copy as we reuse MCR below */ + + if (result == QM_MCR_RESULT_PENDING) { + /* Need to wait for the FQRN in the message ring, which + will only occur once the FQ has been drained. In + order for the FQ to drain the portal needs to be set + to dequeue from the channel the FQ is scheduled on */ + const struct qm_mr_entry *msg; + const struct qm_dqrr_entry *dqrr = NULL; + int found_fqrn = 0; + u16 dequeue_wq = 0; + + /* Flag that we need to drain FQ */ + drain = 1; + + if (channel >= qm_channel_pool1 && + channel < (qm_channel_pool1 + 15)) { + /* Pool channel, enable the bit in the portal */ + dequeue_wq = (channel - + qm_channel_pool1 + 1)<<4 | wq; + } else if (channel < qm_channel_pool1) { + /* Dedicated channel */ + dequeue_wq = wq; + } else { + pr_info("Cannot recover FQ 0x%x, it is " + "scheduled on channel 0x%x", + fqid, channel); + return -EBUSY; + } + /* Set the sdqcr to drain this channel */ + if (channel < qm_channel_pool1) + for (i = 0; i < portal_count; i++) + qm_dqrr_sdqcr_set(portal[i], + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_DEDICATED); + else + for (i = 0; i < portal_count; i++) + qm_dqrr_sdqcr_set( + portal[i], + QM_SDQCR_TYPE_ACTIVE | + QM_SDQCR_CHANNELS_POOL_CONV + (channel)); + while (!found_fqrn) { + /* Keep draining DQRR while checking the MR*/ + for (i = 0; i < portal_count; i++) { + qm_dqrr_pvb_update(portal[i]); + dqrr = qm_dqrr_current(portal[i]); + while (dqrr) { + qm_dqrr_cdc_consume_1ptr( + portal[i], dqrr, 0); + qm_dqrr_pvb_update(portal[i]); + qm_dqrr_next(portal[i]); + dqrr = qm_dqrr_current( + portal[i]); + } + /* Process message ring too */ + qm_mr_pvb_update(portal[i]); + msg = qm_mr_current(portal[i]); + while (msg) { + if ((msg->verb & + QM_MR_VERB_TYPE_MASK) + == QM_MR_VERB_FQRN) + found_fqrn = 1; + qm_mr_next(portal[i]); + qm_mr_cci_consume_to_current( + portal[i]); + qm_mr_pvb_update(portal[i]); + msg = qm_mr_current(portal[i]); + } + cpu_relax(); + } + } + } + if (result != QM_MCR_RESULT_OK && + result != QM_MCR_RESULT_PENDING) { + /* error */ + pr_err("qman_retire_fq failed on FQ 0x%x, result=0x%x\n", + fqid, result); + return -1; + } + if (!(mcr->alterfq.fqs & QM_MCR_FQS_ORLPRESENT)) { + /* ORL had no entries, no need to wait until the + ERNs come in */ + orl_empty = 1; + } + /* Retirement succeeded, check to see if FQ needs + to be drained */ + if (drain || mcr->alterfq.fqs & QM_MCR_FQS_NOTEMPTY) { + /* FQ is Not Empty, drain using volatile DQ commands */ + fq_empty = 0; + do { + const struct qm_dqrr_entry *dqrr = NULL; + u32 vdqcr = fqid | QM_VDQCR_NUMFRAMES_SET(3); + qm_dqrr_vdqcr_set(portal[0], vdqcr); + + /* Wait for a dequeue to occur */ + while (dqrr == NULL) { + qm_dqrr_pvb_update(portal[0]); + dqrr = qm_dqrr_current(portal[0]); + if (!dqrr) + cpu_relax(); + } + /* Process the dequeues, making sure to + empty the ring completely */ + while (dqrr) { + if (dqrr->fqid == fqid && + dqrr->stat & QM_DQRR_STAT_FQ_EMPTY) + fq_empty = 1; + qm_dqrr_cdc_consume_1ptr(portal[0], + dqrr, 0); + qm_dqrr_pvb_update(portal[0]); + qm_dqrr_next(portal[0]); + dqrr = qm_dqrr_current(portal[0]); + } + } while (fq_empty == 0); + } + for (i = 0; i < portal_count; i++) + qm_dqrr_sdqcr_set(portal[i], 0); + + /* Wait for the ORL to have been completely drained */ + while (orl_empty == 0) { + const struct qm_mr_entry *msg; + + qm_mr_pvb_update(portal[0]); + msg = qm_mr_current(portal[0]); + while (msg) { + if ((msg->verb & QM_MR_VERB_TYPE_MASK) == + QM_MR_VERB_FQRL) + orl_empty = 1; + qm_mr_next(portal[0]); + qm_mr_cci_consume_to_current(portal[0]); + qm_mr_pvb_update(portal[0]); + msg = qm_mr_current(portal[0]); + } + cpu_relax(); + } + mcc = qm_mc_start(portal[0]); + mcc->alterfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result != QM_MCR_RESULT_OK) { + pr_err("OOS after drain Failed on FQID 0x%x, result 0x%x\n", + fqid, mcr->result); + return -1; + } + return 0; + case QM_MCR_NP_STATE_RETIRED: + /* Send OOS Command */ + mcc = qm_mc_start(portal[0]); + mcc->alterfq.fqid = fqid; + qm_mc_commit(portal[0], QM_MCC_VERB_ALTER_OOS); + while (!(mcr = qm_mc_result(portal[0]))) + cpu_relax(); + DPA_ASSERT((mcr->verb & QM_MCR_VERB_MASK) == + QM_MCR_VERB_ALTER_OOS); + if (mcr->result) { + pr_err("OOS Failed on FQID 0x%x\n", fqid); + return -1; + } + return 0; + case QM_MCR_NP_STATE_OOS: + /* Done */ + return 0; + } + return -1; +} + +int qman_shutdown_fq(u32 fqid) +{ + struct qman_portal *p; + unsigned long irqflags __maybe_unused; + int ret; + struct qm_portal *low_p; + + p = get_affine_portal(); + PORTAL_IRQ_LOCK(p, irqflags); + low_p = &p->p; + ret = qm_shutdown_fq(&low_p, 1, fqid); + PORTAL_IRQ_UNLOCK(p, irqflags); + put_affine_portal(); + return ret; +} + +const struct qm_portal_config *qman_get_qm_portal_config( + struct qman_portal *portal) +{ +#ifndef __rtems__ + return portal->sharing_redirect ? NULL : portal->config; +#else /* __rtems__ */ + return portal->config; +#endif /* __rtems__ */ +} diff --git a/linux/drivers/soc/fsl/qbman/qman_driver.c b/linux/drivers/soc/fsl/qbman/qman_driver.c new file mode 100644 index 00000000..6923504e --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_driver.c @@ -0,0 +1,87 @@ +#include + +#include + +/* Copyright 2013 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_priv.h" + +#include + +static int __init early_qman_init(void) +{ + struct device_node *dn; + u32 is_portal_available; + + qman_init(); + + is_portal_available = 0; + for_each_compatible_node(dn, NULL, "fsl,qman-portal") { + if (of_device_is_available(dn)) { + is_portal_available = 1; + break; + } + } + + if (!qman_have_ccsr() && is_portal_available) { + struct qman_fq fq = {.fqid = 1}; + struct qm_mcr_queryfq_np np; + int err, retry = CONFIG_FSL_QMAN_INIT_TIMEOUT; + struct timespec nowts, diffts, startts = current_kernel_time(); + + /* Loop while querying given fqid succeeds or time out */ + while (1) { + err = qman_query_fq_np(&fq, &np); + if (!err) { + /* success, control-plane has configured QMan */ + break; + } else if (err != -ERANGE) { + pr_err("I/O error, continuing anyway\n"); + break; + } + nowts = current_kernel_time(); + diffts = timespec_sub(nowts, startts); + if (diffts.tv_sec > 0) { + if (!retry--) { + pr_err("Time out, control-plane dead?\n"); + break; + } + pr_warn("Polling for the control-plane (%d)\n", + retry); + } + } + } + + qman_resource_init(); + + return 0; +} +subsys_initcall(early_qman_init); diff --git a/linux/drivers/soc/fsl/qbman/qman_portal.c b/linux/drivers/soc/fsl/qbman/qman_portal.c new file mode 100644 index 00000000..c74059b1 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_portal.c @@ -0,0 +1,796 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_priv.h" +#ifdef __rtems__ +#include +#endif /* __rtems__ */ + +/* Enable portal interupts (as opposed to polling mode) */ +#define CONFIG_FSL_DPA_PIRQ_SLOW 1 +#define CONFIG_FSL_DPA_PIRQ_FAST 1 + +/* Global variable containing revision id (even on non-control plane systems + * where CCSR isn't available) */ +u16 qman_ip_rev; +EXPORT_SYMBOL(qman_ip_rev); +u16 qm_channel_pool1 = QMAN_CHANNEL_POOL1; +EXPORT_SYMBOL(qm_channel_pool1); +u16 qm_channel_caam = QMAN_CHANNEL_CAAM; +EXPORT_SYMBOL(qm_channel_caam); +u16 qm_channel_pme = QMAN_CHANNEL_PME; +EXPORT_SYMBOL(qm_channel_pme); +u16 qm_channel_dce = QMAN_CHANNEL_DCE; +EXPORT_SYMBOL(qm_channel_dce); +u16 qman_portal_max; +EXPORT_SYMBOL(qman_portal_max); + +#ifndef __rtems__ +/* For these variables, and the portal-initialisation logic, the + * comments in bman_driver.c apply here so won't be repeated. */ +static struct qman_portal *shared_portals[NR_CPUS]; +static int num_shared_portals; +static int shared_portals_idx; +static LIST_HEAD(unused_pcfgs); +#endif /* __rtems__ */ + +/* A SDQCR mask comprising all the available/visible pool channels */ +static u32 pools_sdqcr; + +#define STR_ERR_NOPROP "No '%s' property in node %s\n" +#define STR_ERR_CELL "'%s' is not a %d-cell range in node %s\n" +#define STR_FQID_RANGE "fsl,fqid-range" +#define STR_POOL_CHAN_RANGE "fsl,pool-channel-range" +#define STR_CGRID_RANGE "fsl,cgrid-range" + +/* A "fsl,fqid-range" node; release the given range to the allocator */ +static __init int fsl_fqid_range_init(struct device_node *node) +{ + int ret; + const u32 *range = of_get_property(node, STR_FQID_RANGE, &ret); + + if (!range) { + pr_err(STR_ERR_NOPROP, STR_FQID_RANGE, node->full_name); + return -EINVAL; + } + if (ret != 8) { + pr_err(STR_ERR_CELL, STR_FQID_RANGE, 2, node->full_name); + return -EINVAL; + } + qman_seed_fqid_range(range[0], range[1]); + pr_info("FQID allocator includes range %d:%d\n", + range[0], range[1]); + return 0; +} + +/* A "fsl,pool-channel-range" node; add to the SDQCR mask only */ +static __init int fsl_pool_channel_range_sdqcr(struct device_node *node) +{ + int ret; + const u32 *chanid = of_get_property(node, STR_POOL_CHAN_RANGE, &ret); + + if (!chanid) { + pr_err(STR_ERR_NOPROP, STR_POOL_CHAN_RANGE, node->full_name); + return -EINVAL; + } + if (ret != 8) { + pr_err(STR_ERR_CELL, STR_POOL_CHAN_RANGE, 1, node->full_name); + return -EINVAL; + } + for (ret = 0; ret < chanid[1]; ret++) + pools_sdqcr |= QM_SDQCR_CHANNELS_POOL_CONV(chanid[0] + ret); + return 0; +} + +/* A "fsl,pool-channel-range" node; release the given range to the allocator */ +static __init int fsl_pool_channel_range_init(struct device_node *node) +{ + int ret; + const u32 *chanid = of_get_property(node, STR_POOL_CHAN_RANGE, &ret); + + if (!chanid) { + pr_err(STR_ERR_NOPROP, STR_POOL_CHAN_RANGE, node->full_name); + return -EINVAL; + } + if (ret != 8) { + pr_err(STR_ERR_CELL, STR_POOL_CHAN_RANGE, 1, node->full_name); + return -EINVAL; + } + qman_seed_pool_range(chanid[0], chanid[1]); + pr_info("Pool channel allocator includes range %d:%d\n", + chanid[0], chanid[1]); + return 0; +} + +/* A "fsl,cgrid-range" node; release the given range to the allocator */ +static __init int fsl_cgrid_range_init(struct device_node *node) +{ + struct qman_cgr cgr; + int ret, errors = 0; + const u32 *range = of_get_property(node, STR_CGRID_RANGE, &ret); + + if (!range) { + pr_err(STR_ERR_NOPROP, STR_CGRID_RANGE, node->full_name); + return -EINVAL; + } + if (ret != 8) { + pr_err(STR_ERR_CELL, STR_CGRID_RANGE, 2, node->full_name); + return -EINVAL; + } + qman_seed_cgrid_range(range[0], range[1]); + pr_info("CGRID allocator includes range %d:%d\n", + range[0], range[1]); + for (cgr.cgrid = 0; cgr.cgrid < __CGR_NUM; cgr.cgrid++) { + ret = qman_modify_cgr(&cgr, QMAN_CGR_FLAG_USE_INIT, NULL); + if (ret) + errors++; + } + if (errors) + pr_err("Warning: %d error%s while initialising CGRs %d:%d\n", + errors, (errors > 1) ? "s" : "", range[0], range[1]); + return 0; +} + +static void qman_get_ip_revision(struct device_node *dn) +{ +#ifdef __rtems__ + struct device_node of_dns; +#endif /* __rtems__ */ + u16 ip_rev = 0; + + for_each_compatible_node(dn, NULL, "fsl,qman-portal") { + if (!of_device_is_available(dn)) + continue; + if (of_device_is_compatible(dn, "fsl,qman-portal-1.0") || + of_device_is_compatible(dn, "fsl,qman-portal-1.0.0")) { + pr_err("Rev1.0 on P4080 rev1 is not supported!\n"); + BUG_ON(1); + } else if (of_device_is_compatible(dn, "fsl,qman-portal-1.1") || + of_device_is_compatible(dn, "fsl,qman-portal-1.1.0")) { + ip_rev = QMAN_REV11; + qman_portal_max = 10; + } else if (of_device_is_compatible(dn, "fsl,qman-portal-1.2") || + of_device_is_compatible(dn, "fsl,qman-portal-1.2.0")) { + ip_rev = QMAN_REV12; + qman_portal_max = 10; + } else if (of_device_is_compatible(dn, "fsl,qman-portal-2.0") || + of_device_is_compatible(dn, "fsl,qman-portal-2.0.0")) { + ip_rev = QMAN_REV20; + qman_portal_max = 3; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.0.0")) { + ip_rev = QMAN_REV30; + qman_portal_max = 50; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.0.1")) { + ip_rev = QMAN_REV30; + qman_portal_max = 25; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.0")) { + ip_rev = QMAN_REV31; + qman_portal_max = 50; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.1")) { + ip_rev = QMAN_REV31; + qman_portal_max = 25; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.2")) { + ip_rev = QMAN_REV31; + qman_portal_max = 18; + } else if (of_device_is_compatible(dn, + "fsl,qman-portal-3.1.3")) { + ip_rev = QMAN_REV31; + qman_portal_max = 10; + } else { + pr_warn("Unknown version in portal node, default to rev1.1\n"); + ip_rev = QMAN_REV11; + qman_portal_max = 10; + } + + if (!qman_ip_rev) { + if (ip_rev) { + qman_ip_rev = ip_rev; + } else { + pr_warn("Unknown version, default to rev1.1\n"); + qman_ip_rev = QMAN_REV11; + } + } else if (ip_rev && (qman_ip_rev != ip_rev)) + pr_warn("Revision = 0x%04x, but portal '%s' has 0x%04x\n", + qman_ip_rev, dn->full_name, ip_rev); + if (qman_ip_rev == ip_rev) + break; + } +} + +#ifndef __rtems__ +/* Parse a portal node, perform generic mapping duties and return the config. It + * is not known at this stage for what purpose (or even if) the portal will be + * used. */ +static struct qm_portal_config * __init parse_pcfg(struct device_node *node) +{ + struct qm_portal_config *pcfg; + const u32 *channel; + int irq, ret; + struct resource res; + + pcfg = kzalloc(sizeof(*pcfg), GFP_KERNEL); + if (!pcfg) + return NULL; + + /* + * This is a *horrible hack*, but the IOMMU/PAMU driver needs a + * 'struct device' in order to get the PAMU stashing setup and the QMan + * portal [driver] won't function at all without ring stashing + * + * Making the QMan portal driver nice and proper is part of the + * upstreaming effort + */ + pcfg->dev.bus = &platform_bus_type; + pcfg->dev.of_node = node; +#ifdef CONFIG_IOMMU_API + pcfg->dev.archdata.iommu_domain = NULL; +#endif + + ret = of_address_to_resource(node, DPA_PORTAL_CE, + &pcfg->addr_phys[DPA_PORTAL_CE]); + if (ret) { + pr_err("Can't get %s property 'reg::CE'\n", node->full_name); + goto err; + } + ret = of_address_to_resource(node, DPA_PORTAL_CI, + &pcfg->addr_phys[DPA_PORTAL_CI]); + if (ret) { + pr_err("Can't get %s property 'reg::CI'\n", node->full_name); + goto err; + } + + channel = of_get_property(node, "fsl,qman-channel-id", &ret); + if (!channel || (ret != 4)) { + pr_err("Can't get %s property 'fsl,qman-channel-id'\n", + node->full_name); + goto err; + } + pcfg->public_cfg.channel = *channel; + pcfg->public_cfg.cpu = -1; + irq = irq_of_parse_and_map(node, 0); + if (irq == NO_IRQ) { + pr_err("Can't get %s property 'interrupts'\n", node->full_name); + goto err; + } + pcfg->public_cfg.irq = irq; +#ifdef CONFIG_FSL_QMAN_CONFIG + /* We need the same LIODN offset for all portals */ + qman_liodn_fixup(pcfg->public_cfg.channel); +#endif + + pcfg->addr_virt[DPA_PORTAL_CE] = ioremap_prot( + pcfg->addr_phys[DPA_PORTAL_CE].start, + resource_size(&pcfg->addr_phys[DPA_PORTAL_CE]), + 0); + pcfg->addr_virt[DPA_PORTAL_CI] = ioremap_prot( + pcfg->addr_phys[DPA_PORTAL_CI].start, + resource_size(&pcfg->addr_phys[DPA_PORTAL_CI]), + _PAGE_GUARDED | _PAGE_NO_CACHE); + + return pcfg; +err: + kfree(pcfg); + return NULL; +} + +static struct qm_portal_config *get_pcfg(struct list_head *list) +{ + struct qm_portal_config *pcfg; + + if (list_empty(list)) + return NULL; + pcfg = list_entry(list->prev, struct qm_portal_config, list); + list_del(&pcfg->list); + return pcfg; +} +#endif /* __rtems__ */ + +static void portal_set_cpu(struct qm_portal_config *pcfg, int cpu) +{ +#ifdef CONFIG_FSL_PAMU + int ret; + int window_count = 1; + struct iommu_domain_geometry geom_attr; + struct pamu_stash_attribute stash_attr; + + pcfg->iommu_domain = iommu_domain_alloc(&platform_bus_type); + if (!pcfg->iommu_domain) { + pr_err("%s(): iommu_domain_alloc() failed", __func__); + goto _no_iommu; + } + geom_attr.aperture_start = 0; + geom_attr.aperture_end = + ((dma_addr_t)1 << min(8 * sizeof(dma_addr_t), (size_t)36)) - 1; + geom_attr.force_aperture = true; + ret = iommu_domain_set_attr(pcfg->iommu_domain, DOMAIN_ATTR_GEOMETRY, + &geom_attr); + if (ret < 0) { + pr_err("%s(): iommu_domain_set_attr() = %d", __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_domain_set_attr(pcfg->iommu_domain, DOMAIN_ATTR_WINDOWS, + &window_count); + if (ret < 0) { + pr_err("%s(): iommu_domain_set_attr() = %d", __func__, ret); + goto _iommu_domain_free; + } + stash_attr.cpu = cpu; + stash_attr.cache = PAMU_ATTR_CACHE_L1; + ret = iommu_domain_set_attr(pcfg->iommu_domain, + DOMAIN_ATTR_FSL_PAMU_STASH, + &stash_attr); + if (ret < 0) { + pr_err("%s(): iommu_domain_set_attr() = %d", + __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_domain_window_enable(pcfg->iommu_domain, 0, 0, 1ULL << 36, + IOMMU_READ | IOMMU_WRITE); + if (ret < 0) { + pr_err("%s(): iommu_domain_window_enable() = %d", + __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_attach_device(pcfg->iommu_domain, &pcfg->dev); + if (ret < 0) { + pr_err("%s(): iommu_device_attach() = %d", + __func__, ret); + goto _iommu_domain_free; + } + ret = iommu_domain_set_attr(pcfg->iommu_domain, + DOMAIN_ATTR_FSL_PAMU_ENABLE, + &window_count); + if (ret < 0) { + pr_err("%s(): iommu_domain_set_attr() = %d", + __func__, ret); + goto _iommu_detach_device; + } + +_no_iommu: +#endif +#ifdef CONFIG_FSL_QMAN_CONFIG + if (qman_set_sdest(pcfg->public_cfg.channel, cpu)) +#endif + pr_warn("Failed to set the stash request queue\n"); + + return; + +#ifdef CONFIG_FSL_PAMU +_iommu_detach_device: + iommu_detach_device(pcfg->iommu_domain, NULL); +_iommu_domain_free: + iommu_domain_free(pcfg->iommu_domain); + pcfg->iommu_domain = NULL; +#endif +} + +static struct qman_portal *init_pcfg(struct qm_portal_config *pcfg) +{ + struct qman_portal *p; + +#ifndef __rtems__ + pcfg->iommu_domain = NULL; +#endif /* __rtems__ */ + portal_set_cpu(pcfg, pcfg->public_cfg.cpu); + p = qman_create_affine_portal(pcfg, NULL); + if (p) { + u32 irq_sources = 0; + /* Determine what should be interrupt-vs-poll driven */ +#ifdef CONFIG_FSL_DPA_PIRQ_SLOW + irq_sources |= QM_PIRQ_EQCI | QM_PIRQ_EQRI | QM_PIRQ_MRI | + QM_PIRQ_CSCI; +#endif +#ifdef CONFIG_FSL_DPA_PIRQ_FAST + irq_sources |= QM_PIRQ_DQRI; +#endif + qman_p_irqsource_add(p, irq_sources); + pr_info("Portal %sinitialised, cpu %d\n", +#ifndef __rtems__ + pcfg->public_cfg.is_shared ? "(shared) " : "", +#else /* __rtems__ */ + "", +#endif /* __rtems__ */ + pcfg->public_cfg.cpu); + } else + pr_crit("Portal failure on cpu %d\n", pcfg->public_cfg.cpu); + return p; +} + +#ifndef __rtems__ +static void init_slave(int cpu) +{ + struct qman_portal *p; + struct cpumask oldmask = *tsk_cpus_allowed(current); + + set_cpus_allowed_ptr(current, get_cpu_mask(cpu)); + p = qman_create_affine_slave(shared_portals[shared_portals_idx++], cpu); + if (!p) + pr_err("Slave portal failure on cpu %d\n", cpu); + else + pr_info("Portal (slave) initialised, cpu %d\n", cpu); + set_cpus_allowed_ptr(current, &oldmask); + if (shared_portals_idx >= num_shared_portals) + shared_portals_idx = 0; +} + +static struct cpumask want_unshared __initdata; +static struct cpumask want_shared __initdata; + +static int __init parse_qportals(char *str) +{ + return parse_portals_bootarg(str, &want_shared, &want_unshared, + "qportals"); +} +__setup("qportals=", parse_qportals); + +static void qman_portal_update_sdest(const struct qm_portal_config *pcfg, + unsigned int cpu) +{ + struct pamu_stash_attribute stash_attr; + int ret; + + if (pcfg->iommu_domain) { + stash_attr.cpu = cpu; + stash_attr.cache = PAMU_ATTR_CACHE_L1; + ret = iommu_domain_set_attr(pcfg->iommu_domain, + DOMAIN_ATTR_FSL_PAMU_STASH, &stash_attr); + if (ret < 0) { + pr_err("Failed to update pamu stash setting\n"); + return; + } + } +#ifdef CONFIG_FSL_QMAN_CONFIG + if (qman_set_sdest(pcfg->public_cfg.channel, cpu)) +#endif + pr_warn("Failed to update portal's stash request queue\n"); +} + +static void qman_offline_cpu(unsigned int cpu) +{ + struct qman_portal *p; + const struct qm_portal_config *pcfg; + + p = (struct qman_portal *)affine_portals[cpu]; + if (p) { + pcfg = qman_get_qm_portal_config(p); + if (pcfg) { + irq_set_affinity(pcfg->public_cfg.irq, cpumask_of(0)); + qman_portal_update_sdest(pcfg, 0); + } + } +} +#endif /* __rtems__ */ + +#ifdef CONFIG_HOTPLUG_CPU +static void qman_online_cpu(unsigned int cpu) +{ + struct qman_portal *p; + const struct qm_portal_config *pcfg; + + p = (struct qman_portal *)affine_portals[cpu]; + if (p) { + pcfg = qman_get_qm_portal_config(p); + if (pcfg) { + irq_set_affinity(pcfg->public_cfg.irq, cpumask_of(cpu)); + qman_portal_update_sdest(pcfg, cpu); + } + } +} + +static int qman_hotplug_cpu_callback(struct notifier_block *nfb, + unsigned long action, void *hcpu) +{ + unsigned int cpu = (unsigned long)hcpu; + + switch (action) { + case CPU_ONLINE: + case CPU_ONLINE_FROZEN: + qman_online_cpu(cpu); + break; + case CPU_DOWN_PREPARE: + case CPU_DOWN_PREPARE_FROZEN: + qman_offline_cpu(cpu); + default: + break; + } + return NOTIFY_OK; +} + +static struct notifier_block qman_hotplug_cpu_notifier = { + .notifier_call = qman_hotplug_cpu_callback, +}; +#endif /* CONFIG_HOTPLUG_CPU */ + +#ifdef __rtems__ +#include +static struct qm_portal_config qman_configs[NR_CPUS]; +static void +qman_init_portals(void) +{ + const char *fdt = bsp_fdt_get(); + struct device_node dn; + const char *name; + int cpu_count = (int)rtems_get_processor_count(); + int cpu; + int ret; + int node; + int parent; + + memset(&dn, 0, sizeof(dn)); + + name = "fsl,qman-portal"; + node = fdt_node_offset_by_compatible(fdt, 0, name); + if (node < 0) + panic("qman: no portals in FDT"); + parent = fdt_parent_offset(fdt, node); + if (parent < 0) + panic("qman: no parent of portals in FDT"); + node = fdt_first_subnode(fdt, parent); + + dn.full_name = name; + dn.offset = node; + + qoriq_clear_ce_portal(&qoriq_qman_portal[0][0], + sizeof(qoriq_qman_portal[0])); + qoriq_clear_ci_portal(&qoriq_qman_portal[1][0], + sizeof(qoriq_qman_portal[1])); + + for (cpu = 0; cpu < cpu_count; ++cpu) { + struct qm_portal_config *pcfg = &qman_configs[cpu]; + struct qman_portal *portal; + struct resource res; + const u32 *channel; + + if (node < 0) + panic("qman: missing portal in FDT"); + + ret = of_address_to_resource(&dn, 0, &res); + if (ret != 0) + panic("qman: no portal CE address"); + pcfg->addr_virt[0] = (__iomem void *) + ((uintptr_t)&qoriq_qman_portal[0][0] + (uintptr_t)res.start); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[0] >= + (uintptr_t)&qoriq_qman_portal[0][0]); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[0] < + (uintptr_t)&qoriq_qman_portal[1][0]); + + ret = of_address_to_resource(&dn, 1, &res); + if (ret != 0) + panic("qman: no portal CI address"); + pcfg->addr_virt[1] = (__iomem void *) + ((uintptr_t)&qoriq_qman_portal[0][0] + (uintptr_t)res.start); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[1] >= + (uintptr_t)&qoriq_qman_portal[1][0]); + BSD_ASSERT((uintptr_t)pcfg->addr_virt[1] < + (uintptr_t)&qoriq_qman_portal[2][0]); + + pcfg->public_cfg.irq = of_irq_to_resource(&dn, 0, NULL); + if (pcfg->public_cfg.irq == NO_IRQ) + panic("qman: no portal interrupt"); + + channel = of_get_property(&dn, "fsl,qman-channel-id", &ret); + if (channel == NULL || ret != 4) + panic("qman: no portal channel ID"); + pcfg->public_cfg.channel = *channel; + + pcfg->public_cfg.cpu = cpu; + pcfg->public_cfg.pools = pools_sdqcr; + + portal = init_pcfg(pcfg); + if (portal == NULL) + panic("qman: cannot create portal"); + + node = fdt_next_subnode(fdt, node); + dn.offset = node; + } +} +#endif /* __rtems__ */ +#ifndef __rtems__ +__init int qman_init(void) +{ + struct cpumask slave_cpus; + struct cpumask unshared_cpus = *cpu_none_mask; + struct cpumask shared_cpus = *cpu_none_mask; + LIST_HEAD(unshared_pcfgs); + LIST_HEAD(shared_pcfgs); + struct device_node *dn; + struct qm_portal_config *pcfg; + struct qman_portal *p; + int cpu, ret; + struct cpumask offline_cpus; + + /* Initialise the QMan (CCSR) device */ + for_each_compatible_node(dn, NULL, "fsl,qman") { + if (!qman_init_ccsr(dn)) + pr_info("Err interrupt handler present\n"); + else + pr_err("CCSR setup failed\n"); + } +#else /* __rtems__ */ +int +qman_init(struct device_node *dn) +{ + struct device_node of_dns; + int ret; +#endif /* __rtems__ */ +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP + /* Setup lookup table for FQ demux */ + ret = qman_setup_fq_lookup_table(qman_fqd_size()/64); + if (ret) + return ret; +#endif + + /* Get qman ip revision */ + qman_get_ip_revision(dn); + if ((qman_ip_rev & 0xff00) >= QMAN_REV30) { + qm_channel_pool1 = QMAN_CHANNEL_POOL1_REV3; + qm_channel_caam = QMAN_CHANNEL_CAAM_REV3; + qm_channel_pme = QMAN_CHANNEL_PME_REV3; + } + + /* Parse pool channels into the SDQCR mask. (Must happen before portals + * are initialised.) */ + for_each_compatible_node(dn, NULL, "fsl,pool-channel-range") { + ret = fsl_pool_channel_range_sdqcr(dn); + if (ret) + return ret; + } + +#ifndef __rtems__ + memset(affine_portals, 0, sizeof(void *) * num_possible_cpus()); + /* Initialise portals. See bman_driver.c for comments */ + for_each_compatible_node(dn, NULL, "fsl,qman-portal") { + if (!of_device_is_available(dn)) + continue; + pcfg = parse_pcfg(dn); + if (pcfg) { + pcfg->public_cfg.pools = pools_sdqcr; + list_add_tail(&pcfg->list, &unused_pcfgs); + } + } + for_each_possible_cpu(cpu) { + if (cpumask_test_cpu(cpu, &want_shared)) { + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &shared_pcfgs); + cpumask_set_cpu(cpu, &shared_cpus); + } + if (cpumask_test_cpu(cpu, &want_unshared)) { + if (cpumask_test_cpu(cpu, &shared_cpus)) + continue; + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &unshared_pcfgs); + cpumask_set_cpu(cpu, &unshared_cpus); + } + } + if (list_empty(&shared_pcfgs) && list_empty(&unshared_pcfgs)) { + for_each_possible_cpu(cpu) { + pcfg = get_pcfg(&unused_pcfgs); + if (!pcfg) + break; + pcfg->public_cfg.cpu = cpu; + list_add_tail(&pcfg->list, &unshared_pcfgs); + cpumask_set_cpu(cpu, &unshared_cpus); + } + } + cpumask_andnot(&slave_cpus, cpu_possible_mask, &shared_cpus); + cpumask_andnot(&slave_cpus, &slave_cpus, &unshared_cpus); + if (cpumask_empty(&slave_cpus)) { + if (!list_empty(&shared_pcfgs)) { + cpumask_or(&unshared_cpus, &unshared_cpus, + &shared_cpus); + cpumask_clear(&shared_cpus); + list_splice_tail(&shared_pcfgs, &unshared_pcfgs); + INIT_LIST_HEAD(&shared_pcfgs); + } + } else { + if (list_empty(&shared_pcfgs)) { + pcfg = get_pcfg(&unshared_pcfgs); + if (!pcfg) { + pr_crit("No portals available!\n"); + return 0; + } + cpumask_clear_cpu(pcfg->public_cfg.cpu, &unshared_cpus); + cpumask_set_cpu(pcfg->public_cfg.cpu, &shared_cpus); + list_add_tail(&pcfg->list, &shared_pcfgs); + } + } + list_for_each_entry(pcfg, &unshared_pcfgs, list) { + pcfg->public_cfg.is_shared = 0; + p = init_pcfg(pcfg); + } + list_for_each_entry(pcfg, &shared_pcfgs, list) { + pcfg->public_cfg.is_shared = 1; + p = init_pcfg(pcfg); + if (p) + shared_portals[num_shared_portals++] = p; + } + if (!cpumask_empty(&slave_cpus)) + for_each_cpu(cpu, &slave_cpus) + init_slave(cpu); +#else /* __rtems__ */ + qman_init_portals(); +#endif /* __rtems__ */ + pr_info("Portals initialised\n"); +#ifndef __rtems__ + cpumask_andnot(&offline_cpus, cpu_possible_mask, cpu_online_mask); + for_each_cpu(cpu, &offline_cpus) + qman_offline_cpu(cpu); +#endif /* __rtems__ */ +#ifdef CONFIG_HOTPLUG_CPU + register_hotcpu_notifier(&qman_hotplug_cpu_notifier); +#endif + return 0; +} + +__init int qman_resource_init(void) +{ +#ifdef __rtems__ + struct device_node of_dns; +#endif /* __rtems__ */ + struct device_node *dn; + int ret; + + /* Initialise FQID allocation ranges */ + for_each_compatible_node(dn, NULL, "fsl,fqid-range") { + ret = fsl_fqid_range_init(dn); + if (ret) + return ret; + } + /* Initialise CGRID allocation ranges */ + for_each_compatible_node(dn, NULL, "fsl,cgrid-range") { + ret = fsl_cgrid_range_init(dn); + if (ret) + return ret; + } + /* Parse pool channels into the allocator. (Must happen after portals + * are initialised.) */ + for_each_compatible_node(dn, NULL, "fsl,pool-channel-range") { + ret = fsl_pool_channel_range_init(dn); + if (ret) + return ret; + } + + return 0; +} diff --git a/linux/drivers/soc/fsl/qbman/qman_priv.h b/linux/drivers/soc/fsl/qbman/qman_priv.h new file mode 100644 index 00000000..f04bd476 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_priv.h @@ -0,0 +1,293 @@ +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include "dpaa_sys.h" + +#include +#include +#include + +/* Congestion Groups */ + +/* This wrapper represents a bit-array for the state of the 256 QMan congestion + * groups. Is also used as a *mask* for congestion groups, eg. so we ignore + * those that don't concern us. We harness the structure and accessor details + * already used in the management command to query congestion groups. + */ +struct qman_cgrs { + struct __qm_mcr_querycongestion q; +}; +static inline void qman_cgrs_init(struct qman_cgrs *c) +{ + memset(c, 0, sizeof(*c)); +} +static inline void qman_cgrs_fill(struct qman_cgrs *c) +{ + memset(c, 0xff, sizeof(*c)); +} +static inline int qman_cgrs_get(struct qman_cgrs *c, int num) +{ + return QM_MCR_QUERYCONGESTION(&c->q, num); +} +static inline void qman_cgrs_set(struct qman_cgrs *c, int num) +{ + c->q.__state[__CGR_WORD(num)] |= (0x80000000 >> __CGR_SHIFT(num)); +} +static inline void qman_cgrs_unset(struct qman_cgrs *c, int num) +{ + c->q.__state[__CGR_WORD(num)] &= ~(0x80000000 >> __CGR_SHIFT(num)); +} +static inline int qman_cgrs_next(struct qman_cgrs *c, int num) +{ + while ((++num < __CGR_NUM) && !qman_cgrs_get(c, num)) + ; + return num; +} +static inline void qman_cgrs_cp(struct qman_cgrs *dest, + const struct qman_cgrs *src) +{ + *dest = *src; +} +static inline void qman_cgrs_and(struct qman_cgrs *dest, + const struct qman_cgrs *a, const struct qman_cgrs *b) +{ + int ret; + u32 *_d = dest->q.__state; + const u32 *_a = a->q.__state; + const u32 *_b = b->q.__state; + + for (ret = 0; ret < 8; ret++) + *(_d++) = *(_a++) & *(_b++); +} +static inline void qman_cgrs_xor(struct qman_cgrs *dest, + const struct qman_cgrs *a, const struct qman_cgrs *b) +{ + int ret; + u32 *_d = dest->q.__state; + const u32 *_a = a->q.__state; + const u32 *_b = b->q.__state; + + for (ret = 0; ret < 8; ret++) + *(_d++) = *(_a++) ^ *(_b++); +} + +/* used by CCSR and portal interrupt code */ +enum qm_isr_reg { + qm_isr_status = 0, + qm_isr_enable = 1, + qm_isr_disable = 2, + qm_isr_inhibit = 3 +}; + +struct qm_portal_config { + /* Corenet portal addresses; + * [0]==cache-enabled, [1]==cache-inhibited. */ + __iomem void *addr_virt[2]; +#ifndef __rtems__ + struct resource addr_phys[2]; + struct device dev; + struct iommu_domain *iommu_domain; + /* Allow these to be joined in lists */ + struct list_head list; +#endif /* __rtems__ */ + /* User-visible portal configuration settings */ + struct qman_portal_config public_cfg; +}; + +/* Revision info (for errata and feature handling) */ +#define QMAN_REV11 0x0101 +#define QMAN_REV12 0x0102 +#define QMAN_REV20 0x0200 +#define QMAN_REV30 0x0300 +#define QMAN_REV31 0x0301 +extern u16 qman_ip_rev; /* 0 if uninitialised, otherwise QMAN_REVx */ + +extern u16 qman_portal_max; + +#ifdef CONFIG_FSL_QMAN_CONFIG +/* Hooks from qman_driver.c to qman_config.c */ +int qman_init_ccsr(struct device_node *node); +void qman_liodn_fixup(u16 channel); +int qman_set_sdest(u16 channel, unsigned int cpu_idx); +size_t qman_fqd_size(void); +#endif + +int qm_set_wpm(int wpm); +int qm_get_wpm(int *wpm); + +/* Hooks from qman_driver.c in to qman_high.c */ +struct qman_portal *qman_create_portal( + struct qman_portal *portal, + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs); + +struct qman_portal *qman_create_affine_portal( + const struct qm_portal_config *config, + const struct qman_cgrs *cgrs); +struct qman_portal *qman_create_affine_slave(struct qman_portal *redirect, + int cpu); +const struct qm_portal_config *qman_destroy_affine_portal(void); +void qman_destroy_portal(struct qman_portal *qm); + +/* This CGR feature is supported by h/w and required by unit-tests and the + * debugfs hooks, so is implemented in the driver. However it allows an explicit + * corruption of h/w fields by s/w that are usually incorruptible (because the + * counters are usually maintained entirely within h/w). As such, we declare + * this API internally. */ +int qman_testwrite_cgr(struct qman_cgr *cgr, u64 i_bcnt, + struct qm_mcr_cgrtestwrite *result); + +#ifdef CONFIG_FSL_QMAN_FQ_LOOKUP +/* If the fq object pointer is greater than the size of context_b field, + * than a lookup table is required. */ +int qman_setup_fq_lookup_table(size_t num_entries); +#endif + + +/*************************************************/ +/* QMan s/w corenet portal, low-level i/face */ +/*************************************************/ + +/* Note: most functions are only used by the high-level interface, so are + * inlined from qman.h. The stuff below is for use by other parts of the + * driver. */ + +/* For qm_dqrr_sdqcr_set(); Choose one SOURCE. Choose one COUNT. Choose one + * dequeue TYPE. Choose TOKEN (8-bit). + * If SOURCE == CHANNELS, + * Choose CHANNELS_DEDICATED and/or CHANNELS_POOL(n). + * You can choose DEDICATED_PRECEDENCE if the portal channel should have + * priority. + * If SOURCE == SPECIFICWQ, + * Either select the work-queue ID with SPECIFICWQ_WQ(), or select the + * channel (SPECIFICWQ_DEDICATED or SPECIFICWQ_POOL()) and specify the + * work-queue priority (0-7) with SPECIFICWQ_WQ() - either way, you get the + * same value. + */ +#define QM_SDQCR_SOURCE_CHANNELS 0x0 +#define QM_SDQCR_SOURCE_SPECIFICWQ 0x40000000 +#define QM_SDQCR_COUNT_EXACT1 0x0 +#define QM_SDQCR_COUNT_UPTO3 0x20000000 +#define QM_SDQCR_DEDICATED_PRECEDENCE 0x10000000 +#define QM_SDQCR_TYPE_MASK 0x03000000 +#define QM_SDQCR_TYPE_NULL 0x0 +#define QM_SDQCR_TYPE_PRIO_QOS 0x01000000 +#define QM_SDQCR_TYPE_ACTIVE_QOS 0x02000000 +#define QM_SDQCR_TYPE_ACTIVE 0x03000000 +#define QM_SDQCR_TOKEN_MASK 0x00ff0000 +#define QM_SDQCR_TOKEN_SET(v) (((v) & 0xff) << 16) +#define QM_SDQCR_TOKEN_GET(v) (((v) >> 16) & 0xff) +#define QM_SDQCR_CHANNELS_DEDICATED 0x00008000 +#define QM_SDQCR_SPECIFICWQ_MASK 0x000000f7 +#define QM_SDQCR_SPECIFICWQ_DEDICATED 0x00000000 +#define QM_SDQCR_SPECIFICWQ_POOL(n) ((n) << 4) +#define QM_SDQCR_SPECIFICWQ_WQ(n) (n) + +/* For qm_dqrr_vdqcr_set(): use FQID(n) to fill in the frame queue ID */ +#define QM_VDQCR_FQID_MASK 0x00ffffff +#define QM_VDQCR_FQID(n) ((n) & QM_VDQCR_FQID_MASK) + +/* For qm_dqrr_pdqcr_set(); Choose one MODE. Choose one COUNT. + * If MODE==SCHEDULED + * Choose SCHEDULED_CHANNELS or SCHEDULED_SPECIFICWQ. Choose one dequeue TYPE. + * If CHANNELS, + * Choose CHANNELS_DEDICATED and/or CHANNELS_POOL() channels. + * You can choose DEDICATED_PRECEDENCE if the portal channel should have + * priority. + * If SPECIFICWQ, + * Either select the work-queue ID with SPECIFICWQ_WQ(), or select the + * channel (SPECIFICWQ_DEDICATED or SPECIFICWQ_POOL()) and specify the + * work-queue priority (0-7) with SPECIFICWQ_WQ() - either way, you get the + * same value. + * If MODE==UNSCHEDULED + * Choose FQID(). + */ +#define QM_PDQCR_MODE_SCHEDULED 0x0 +#define QM_PDQCR_MODE_UNSCHEDULED 0x80000000 +#define QM_PDQCR_SCHEDULED_CHANNELS 0x0 +#define QM_PDQCR_SCHEDULED_SPECIFICWQ 0x40000000 +#define QM_PDQCR_COUNT_EXACT1 0x0 +#define QM_PDQCR_COUNT_UPTO3 0x20000000 +#define QM_PDQCR_DEDICATED_PRECEDENCE 0x10000000 +#define QM_PDQCR_TYPE_MASK 0x03000000 +#define QM_PDQCR_TYPE_NULL 0x0 +#define QM_PDQCR_TYPE_PRIO_QOS 0x01000000 +#define QM_PDQCR_TYPE_ACTIVE_QOS 0x02000000 +#define QM_PDQCR_TYPE_ACTIVE 0x03000000 +#define QM_PDQCR_CHANNELS_DEDICATED 0x00008000 +#define QM_PDQCR_CHANNELS_POOL(n) (0x00008000 >> (n)) +#define QM_PDQCR_SPECIFICWQ_MASK 0x000000f7 +#define QM_PDQCR_SPECIFICWQ_DEDICATED 0x00000000 +#define QM_PDQCR_SPECIFICWQ_POOL(n) ((n) << 4) +#define QM_PDQCR_SPECIFICWQ_WQ(n) (n) +#define QM_PDQCR_FQID(n) ((n) & 0xffffff) + +/* Used by all portal interrupt registers except 'inhibit' + * Channels with frame availability + */ +#define QM_PIRQ_DQAVAIL 0x0000ffff + +/* The DQAVAIL interrupt fields break down into these bits; */ +#define QM_DQAVAIL_PORTAL 0x8000 /* Portal channel */ +#define QM_DQAVAIL_POOL(n) (0x8000 >> (n)) /* Pool channel, n==[1..15] */ +#define QM_DQAVAIL_MASK 0xffff +/* This mask contains all the "irqsource" bits visible to API users */ +#define QM_PIRQ_VISIBLE (QM_PIRQ_SLOW | QM_PIRQ_DQRI) + +/* These are qm__(). So for example, qm_disable_write() means "write + * the disable register" rather than "disable the ability to write". */ +#define qm_isr_status_read(qm) __qm_isr_read(qm, qm_isr_status) +#define qm_isr_status_clear(qm, m) __qm_isr_write(qm, qm_isr_status, m) +#define qm_isr_enable_read(qm) __qm_isr_read(qm, qm_isr_enable) +#define qm_isr_enable_write(qm, v) __qm_isr_write(qm, qm_isr_enable, v) +#define qm_isr_disable_read(qm) __qm_isr_read(qm, qm_isr_disable) +#define qm_isr_disable_write(qm, v) __qm_isr_write(qm, qm_isr_disable, v) +/* TODO: unfortunate name-clash here, reword? */ +#define qm_isr_inhibit(qm) __qm_isr_write(qm, qm_isr_inhibit, 1) +#define qm_isr_uninhibit(qm) __qm_isr_write(qm, qm_isr_inhibit, 0) + +#ifdef CONFIG_FSL_QMAN_CONFIG +int qman_have_ccsr(void); +#else +#define qman_have_ccsr 0 +#endif + +#ifndef __rtems__ +__init int qman_init(void); +#else /* __rtems__ */ +int qman_init(struct device_node *dn); +#endif /* __rtems__ */ +__init int qman_resource_init(void); + +extern void *affine_portals[NR_CPUS]; +const struct qm_portal_config *qman_get_qm_portal_config( + struct qman_portal *portal); diff --git a/linux/drivers/soc/fsl/qbman/qman_test.c b/linux/drivers/soc/fsl/qbman/qman_test.c new file mode 100644 index 00000000..18c04482 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_test.c @@ -0,0 +1,61 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_test.h" + +MODULE_AUTHOR("Geoff Thorpe"); +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_DESCRIPTION("QMan testing"); + +static int test_init(void) +{ + int loop = 1; + + while (loop--) { +#ifdef CONFIG_FSL_QMAN_TEST_STASH + qman_test_stash(); +#endif +#ifdef CONFIG_FSL_QMAN_TEST_API + qman_test_api(); +#endif + } + return 0; +} + +static void test_exit(void) +{ +} + +module_init(test_init); +module_exit(test_exit); diff --git a/linux/drivers/soc/fsl/qbman/qman_test.h b/linux/drivers/soc/fsl/qbman/qman_test.h new file mode 100644 index 00000000..0b34a670 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_test.h @@ -0,0 +1,44 @@ +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +void qman_test_stash(void); +void qman_test_api(void); diff --git a/linux/drivers/soc/fsl/qbman/qman_test_api.c b/linux/drivers/soc/fsl/qbman/qman_test_api.c new file mode 100644 index 00000000..63a6d11d --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_test_api.c @@ -0,0 +1,222 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_test.h" + +/*************/ +/* constants */ +/*************/ + +#define CGR_ID 27 +#define POOL_ID 2 +#define FQ_FLAGS QMAN_FQ_FLAG_DYNAMIC_FQID +#define NUM_ENQUEUES 10 +#define NUM_PARTIAL 4 +#define PORTAL_SDQCR (QM_SDQCR_SOURCE_CHANNELS | \ + QM_SDQCR_TYPE_PRIO_QOS | \ + QM_SDQCR_TOKEN_SET(0x98) | \ + QM_SDQCR_CHANNELS_DEDICATED | \ + QM_SDQCR_CHANNELS_POOL(POOL_ID)) +#define PORTAL_OPAQUE ((void *)0xf00dbeef) +#define VDQCR_FLAGS (QMAN_VOLATILE_FLAG_WAIT | QMAN_VOLATILE_FLAG_FINISH) + +/*************************************/ +/* Predeclarations (eg. for fq_base) */ +/*************************************/ + +static enum qman_cb_dqrr_result cb_dqrr(struct qman_portal *, + struct qman_fq *, + const struct qm_dqrr_entry *); +static void cb_ern(struct qman_portal *, struct qman_fq *, + const struct qm_mr_entry *); +static void cb_fqs(struct qman_portal *, struct qman_fq *, + const struct qm_mr_entry *); + +/***************/ +/* global vars */ +/***************/ + +static struct qm_fd fd, fd_dq; +static struct qman_fq fq_base = { + .cb.dqrr = cb_dqrr, + .cb.ern = cb_ern, + .cb.fqs = cb_fqs +}; +static DECLARE_WAIT_QUEUE_HEAD(waitqueue); +static int retire_complete, sdqcr_complete; + +/**********************/ +/* internal functions */ +/**********************/ + +/* Helpers for initialising and "incrementing" a frame descriptor */ +static void fd_init(struct qm_fd *__fd) +{ + qm_fd_addr_set64(__fd, 0xabdeadbeefLLU); + __fd->format = qm_fd_contig_big; + __fd->length29 = 0x0000ffff; + __fd->cmd = 0xfeedf00d; +} + +static void fd_inc(struct qm_fd *__fd) +{ + u64 t = qm_fd_addr_get64(__fd); + int z = t >> 40; + + t <<= 1; + if (z) + t |= 1; + qm_fd_addr_set64(__fd, t); + __fd->length29--; + __fd->cmd++; +} + +/* The only part of the 'fd' we can't memcmp() is the ppid */ +static int fd_cmp(const struct qm_fd *a, const struct qm_fd *b) +{ + int r = (qm_fd_addr_get64(a) == qm_fd_addr_get64(b)) ? 0 : -1; + + if (!r) + r = a->format - b->format; + if (!r) + r = a->opaque - b->opaque; + if (!r) + r = a->cmd - b->cmd; + return r; +} + +/********/ +/* test */ +/********/ + +static void do_enqueues(struct qman_fq *fq) +{ + unsigned int loop; + + for (loop = 0; loop < NUM_ENQUEUES; loop++) { + if (qman_enqueue(fq, &fd, QMAN_ENQUEUE_FLAG_WAIT | + (((loop + 1) == NUM_ENQUEUES) ? + QMAN_ENQUEUE_FLAG_WAIT_SYNC : 0))) + panic("qman_enqueue() failed\n"); + fd_inc(&fd); + } +} + +void qman_test_api(void) +{ + u32 flags; + int res; + struct qman_fq *fq = &fq_base; + + pr_info("%s(): Starting\n", __func__); + fd_init(&fd); + fd_init(&fd_dq); + + /* Initialise (parked) FQ */ + if (qman_create_fq(0, FQ_FLAGS, fq)) + panic("qman_create_fq() failed\n"); + if (qman_init_fq(fq, QMAN_INITFQ_FLAG_LOCAL, NULL)) + panic("qman_init_fq() failed\n"); + + /* Do enqueues + VDQCR, twice. (Parked FQ) */ + do_enqueues(fq); + pr_info("VDQCR (till-empty);\n"); + if (qman_volatile_dequeue(fq, VDQCR_FLAGS, + QM_VDQCR_NUMFRAMES_TILLEMPTY)) + panic("qman_volatile_dequeue() failed\n"); + do_enqueues(fq); + pr_info("VDQCR (%d of %d);\n", NUM_PARTIAL, NUM_ENQUEUES); + if (qman_volatile_dequeue(fq, VDQCR_FLAGS, + QM_VDQCR_NUMFRAMES_SET(NUM_PARTIAL))) + panic("qman_volatile_dequeue() failed\n"); + pr_info("VDQCR (%d of %d);\n", NUM_ENQUEUES - NUM_PARTIAL, + NUM_ENQUEUES); + if (qman_volatile_dequeue(fq, VDQCR_FLAGS, + QM_VDQCR_NUMFRAMES_SET(NUM_ENQUEUES - NUM_PARTIAL))) + panic("qman_volatile_dequeue() failed\n"); + + do_enqueues(fq); + pr_info("scheduled dequeue (till-empty)\n"); + if (qman_schedule_fq(fq)) + panic("qman_schedule_fq() failed\n"); + wait_event(waitqueue, sdqcr_complete); + + /* Retire and OOS the FQ */ + res = qman_retire_fq(fq, &flags); + if (res < 0) + panic("qman_retire_fq() failed\n"); + wait_event(waitqueue, retire_complete); + if (flags & QMAN_FQ_STATE_BLOCKOOS) + panic("leaking frames\n"); + if (qman_oos_fq(fq)) + panic("qman_oos_fq() failed\n"); + qman_destroy_fq(fq, 0); + pr_info("%s(): Finished\n", __func__); +} + +static enum qman_cb_dqrr_result cb_dqrr(struct qman_portal *p, + struct qman_fq *fq, + const struct qm_dqrr_entry *dq) +{ + if (fd_cmp(&fd_dq, &dq->fd)) { + pr_err("BADNESS: dequeued frame doesn't match;\n"); + BUG(); + } + fd_inc(&fd_dq); + if (!(dq->stat & QM_DQRR_STAT_UNSCHEDULED) && !fd_cmp(&fd_dq, &fd)) { + sdqcr_complete = 1; + wake_up(&waitqueue); + } + return qman_cb_dqrr_consume; +} + +static void cb_ern(struct qman_portal *p, struct qman_fq *fq, + const struct qm_mr_entry *msg) +{ + panic("cb_ern() unimplemented"); +} + +static void cb_fqs(struct qman_portal *p, struct qman_fq *fq, + const struct qm_mr_entry *msg) +{ + u8 verb = (msg->verb & QM_MR_VERB_TYPE_MASK); + + if ((verb != QM_MR_VERB_FQRN) && (verb != QM_MR_VERB_FQRNI)) + panic("unexpected FQS message"); +#ifndef __rtems__ + pr_info("Retirement message received\n"); +#endif /* __rtems__ */ + retire_complete = 1; + wake_up(&waitqueue); +} diff --git a/linux/drivers/soc/fsl/qbman/qman_test_stash.c b/linux/drivers/soc/fsl/qbman/qman_test_stash.c new file mode 100644 index 00000000..a3ca6603 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_test_stash.c @@ -0,0 +1,540 @@ +#include + +#include + +/* Copyright 2009 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_test.h" + +#include +#ifdef __rtems__ +#include +#undef msleep +#define msleep(x) usleep((x) * 1000) +#define L1_CACHE_BYTES 64 +#endif /* __rtems__ */ + +/* Algorithm: + * + * Each cpu will have HP_PER_CPU "handlers" set up, each of which incorporates + * an rx/tx pair of FQ objects (both of which are stashed on dequeue). The + * organisation of FQIDs is such that the HP_PER_CPU*NUM_CPUS handlers will + * shuttle a "hot potato" frame around them such that every forwarding action + * moves it from one cpu to another. (The use of more than one handler per cpu + * is to allow enough handlers/FQs to truly test the significance of caching - + * ie. when cache-expiries are occurring.) + * + * The "hot potato" frame content will be HP_NUM_WORDS*4 bytes in size, and the + * first and last words of the frame data will undergo a transformation step on + * each forwarding action. To achieve this, each handler will be assigned a + * 32-bit "mixer", that is produced using a 32-bit LFSR. When a frame is + * received by a handler, the mixer of the expected sender is XOR'd into all + * words of the entire frame, which is then validated against the original + * values. Then, before forwarding, the entire frame is XOR'd with the mixer of + * the current handler. Apart from validating that the frame is taking the + * expected path, this also provides some quasi-realistic overheads to each + * forwarding action - dereferencing *all* the frame data, computation, and + * conditional branching. There is a "special" handler designated to act as the + * instigator of the test by creating an enqueuing the "hot potato" frame, and + * to determine when the test has completed by counting HP_LOOPS iterations. + * + * Init phases: + * + * 1. prepare each cpu's 'hp_cpu' struct using on_each_cpu(,,1) and link them + * into 'hp_cpu_list'. Specifically, set processor_id, allocate HP_PER_CPU + * handlers and link-list them (but do no other handler setup). + * + * 2. scan over 'hp_cpu_list' HP_PER_CPU times, the first time sets each + * hp_cpu's 'iterator' to point to its first handler. With each loop, + * allocate rx/tx FQIDs and mixer values to the hp_cpu's iterator handler + * and advance the iterator for the next loop. This includes a final fixup, + * which connects the last handler to the first (and which is why phase 2 + * and 3 are separate). + * + * 3. scan over 'hp_cpu_list' HP_PER_CPU times, the first time sets each + * hp_cpu's 'iterator' to point to its first handler. With each loop, + * initialise FQ objects and advance the iterator for the next loop. + * Moreover, do this initialisation on the cpu it applies to so that Rx FQ + * initialisation targets the correct cpu. + */ + +/* helper to run something on all cpus (can't use on_each_cpu(), as that invokes + * the fn from irq context, which is too restrictive). */ +struct bstrap { + void (*fn)(void); + atomic_t started; +}; +static int bstrap_fn(void *__bstrap) +{ + struct bstrap *bstrap = __bstrap; + + atomic_inc(&bstrap->started); + bstrap->fn(); + while (!kthread_should_stop()) + msleep(1); + return 0; +} +static int on_all_cpus(void (*fn)(void)) +{ + int cpu; + +#ifndef __rtems__ + for_each_cpu(cpu, cpu_online_mask) { +#else /* __rtems__ */ + for (cpu = 0; cpu < (int)rtems_get_processor_count(); ++cpu) { +#endif /* __rtems__ */ + struct bstrap bstrap = { + .fn = fn, + .started = ATOMIC_INIT(0) + }; + struct task_struct *k = kthread_create(bstrap_fn, &bstrap, + "hotpotato%d", cpu); + int ret; + +#ifndef __rtems__ + if (IS_ERR(k)) +#else /* __rtems__ */ + if (k == NULL) +#endif /* __rtems__ */ + return -ENOMEM; + kthread_bind(k, cpu); + wake_up_process(k); + /* If we call kthread_stop() before the "wake up" has had an + * effect, then the thread may exit with -EINTR without ever + * running the function. So poll until it's started before + * requesting it to stop. */ + while (!atomic_read(&bstrap.started)) + msleep(10); + ret = kthread_stop(k); + if (ret) + return ret; + } + return 0; +} + +struct hp_handler { + + /* The following data is stashed when 'rx' is dequeued; */ + /* -------------- */ + /* The Rx FQ, dequeues of which will stash the entire hp_handler */ + struct qman_fq rx; + /* The Tx FQ we should forward to */ + struct qman_fq tx; + /* The value we XOR post-dequeue, prior to validating */ + u32 rx_mixer; + /* The value we XOR pre-enqueue, after validating */ + u32 tx_mixer; + /* what the hotpotato address should be on dequeue */ + dma_addr_t addr; + u32 *frame_ptr; + + /* The following data isn't (necessarily) stashed on dequeue; */ + /* -------------- */ + u32 fqid_rx, fqid_tx; + /* list node for linking us into 'hp_cpu' */ + struct list_head node; + /* Just to check ... */ + unsigned int processor_id; +} ____cacheline_aligned; + +struct hp_cpu { + /* identify the cpu we run on; */ + unsigned int processor_id; + /* root node for the per-cpu list of handlers */ + struct list_head handlers; + /* list node for linking us into 'hp_cpu_list' */ + struct list_head node; + /* when repeatedly scanning 'hp_list', each time linking the n'th + * handlers together, this is used as per-cpu iterator state */ + struct hp_handler *iterator; +}; + +/* Each cpu has one of these */ +static DEFINE_PER_CPU(struct hp_cpu, hp_cpus); + +/* links together the hp_cpu structs, in first-come first-serve order. */ +static LIST_HEAD(hp_cpu_list); +static spinlock_t hp_lock = __SPIN_LOCK_UNLOCKED(hp_lock); + +static unsigned int hp_cpu_list_length; + +/* the "special" handler, that starts and terminates the test. */ +static struct hp_handler *special_handler; +static int loop_counter; + +/* handlers are allocated out of this, so they're properly aligned. */ +static struct kmem_cache *hp_handler_slab; + +/* this is the frame data */ +#ifndef __rtems__ +static void *__frame_ptr; +#endif /* __rtems__ */ +static u32 *frame_ptr; +#ifndef __rtems__ +static dma_addr_t frame_dma; +#endif /* __rtems__ */ + +/* the main function waits on this */ +static DECLARE_WAIT_QUEUE_HEAD(queue); + +#define HP_PER_CPU 2 +#define HP_LOOPS 8 +/* 80 bytes, like a small ethernet frame, and bleeds into a second cacheline */ +#define HP_NUM_WORDS 80 +/* First word of the LFSR-based frame data */ +#define HP_FIRST_WORD 0xabbaf00d + +static inline u32 do_lfsr(u32 prev) +{ + return (prev >> 1) ^ (-(prev & 1u) & 0xd0000001u); +} + +static void allocate_frame_data(void) +{ + u32 lfsr = HP_FIRST_WORD; + int loop; +#ifndef __rtems__ + struct platform_device *pdev = platform_device_alloc("foobar", -1); + + if (!pdev) + panic("platform_device_alloc() failed"); + if (platform_device_add(pdev)) + panic("platform_device_add() failed"); + __frame_ptr = kmalloc(4 * HP_NUM_WORDS, GFP_KERNEL); + if (!__frame_ptr) + panic("kmalloc() failed"); + frame_ptr = (void *)(((unsigned long)__frame_ptr + 63) & + ~(unsigned long)63); +#else /* __rtems__ */ + frame_ptr = rtems_heap_allocate_aligned_with_boundary(4 * HP_NUM_WORDS, 64, 0); + if (frame_ptr == NULL) + panic("rtems_heap_allocate_aligned_with_boundary() failed"); +#endif /* __rtems__ */ + for (loop = 0; loop < HP_NUM_WORDS; loop++) { + frame_ptr[loop] = lfsr; + lfsr = do_lfsr(lfsr); + } +#ifndef __rtems__ + frame_dma = dma_map_single(&pdev->dev, frame_ptr, 4 * HP_NUM_WORDS, + DMA_BIDIRECTIONAL); + platform_device_del(pdev); + platform_device_put(pdev); +#endif /* __rtems__ */ +} + +static void deallocate_frame_data(void) +{ +#ifndef __rtems__ + kfree(__frame_ptr); +#endif /* __rtems__ */ +} + +static inline void process_frame_data(struct hp_handler *handler, + const struct qm_fd *fd) +{ + u32 *p = handler->frame_ptr; + u32 lfsr = HP_FIRST_WORD; + int loop; + + if (qm_fd_addr_get64(fd) != handler->addr) + panic("bad frame address"); + for (loop = 0; loop < HP_NUM_WORDS; loop++, p++) { + *p ^= handler->rx_mixer; + if (*p != lfsr) + panic("corrupt frame data"); + *p ^= handler->tx_mixer; + lfsr = do_lfsr(lfsr); + } +} + +static enum qman_cb_dqrr_result normal_dqrr(struct qman_portal *portal, + struct qman_fq *fq, + const struct qm_dqrr_entry *dqrr) +{ + struct hp_handler *handler = (struct hp_handler *)fq; + + process_frame_data(handler, &dqrr->fd); + if (qman_enqueue(&handler->tx, &dqrr->fd, 0)) + panic("qman_enqueue() failed"); + return qman_cb_dqrr_consume; +} + +static enum qman_cb_dqrr_result special_dqrr(struct qman_portal *portal, + struct qman_fq *fq, + const struct qm_dqrr_entry *dqrr) +{ + struct hp_handler *handler = (struct hp_handler *)fq; + + process_frame_data(handler, &dqrr->fd); + if (++loop_counter < HP_LOOPS) { + if (qman_enqueue(&handler->tx, &dqrr->fd, 0)) + panic("qman_enqueue() failed"); + } else { + pr_info("Received final (%dth) frame\n", loop_counter); + wake_up(&queue); + } + return qman_cb_dqrr_consume; +} + +static void create_per_cpu_handlers(void) +{ + struct hp_handler *handler; + int loop; + struct hp_cpu *hp_cpu = this_cpu_ptr(&hp_cpus); + + hp_cpu->processor_id = smp_processor_id(); + spin_lock(&hp_lock); + list_add_tail(&hp_cpu->node, &hp_cpu_list); + hp_cpu_list_length++; + spin_unlock(&hp_lock); + INIT_LIST_HEAD(&hp_cpu->handlers); + for (loop = 0; loop < HP_PER_CPU; loop++) { + handler = kmem_cache_alloc(hp_handler_slab, GFP_KERNEL); + if (!handler) + panic("kmem_cache_alloc() failed"); + handler->processor_id = hp_cpu->processor_id; +#ifndef __rtems__ + handler->addr = frame_dma; +#else /* __rtems__ */ + handler->addr = (dma_addr_t)frame_ptr; +#endif /* __rtems__ */ + handler->frame_ptr = frame_ptr; + list_add_tail(&handler->node, &hp_cpu->handlers); + } +} + +static void destroy_per_cpu_handlers(void) +{ + struct list_head *loop, *tmp; + struct hp_cpu *hp_cpu = this_cpu_ptr(&hp_cpus); + + spin_lock(&hp_lock); + list_del(&hp_cpu->node); + spin_unlock(&hp_lock); + list_for_each_safe(loop, tmp, &hp_cpu->handlers) { + u32 flags; + struct hp_handler *handler = list_entry(loop, struct hp_handler, + node); + if (qman_retire_fq(&handler->rx, &flags)) + panic("qman_retire_fq(rx) failed"); + BUG_ON(flags & QMAN_FQ_STATE_BLOCKOOS); + if (qman_oos_fq(&handler->rx)) + panic("qman_oos_fq(rx) failed"); + qman_destroy_fq(&handler->rx, 0); + qman_destroy_fq(&handler->tx, 0); + qman_release_fqid(handler->fqid_rx); + list_del(&handler->node); + kmem_cache_free(hp_handler_slab, handler); + } +} + +static inline u8 num_cachelines(u32 offset) +{ + u8 res = (offset + (L1_CACHE_BYTES - 1)) + / (L1_CACHE_BYTES); + if (res > 3) + return 3; + return res; +} +#define STASH_DATA_CL \ + num_cachelines(HP_NUM_WORDS * 4) +#define STASH_CTX_CL \ + num_cachelines(offsetof(struct hp_handler, fqid_rx)) + +static void init_handler(void *__handler) +{ + struct qm_mcc_initfq opts; + struct hp_handler *handler = __handler; + + BUG_ON(handler->processor_id != smp_processor_id()); + /* Set up rx */ + memset(&handler->rx, 0, sizeof(handler->rx)); + if (handler == special_handler) + handler->rx.cb.dqrr = special_dqrr; + else + handler->rx.cb.dqrr = normal_dqrr; + if (qman_create_fq(handler->fqid_rx, 0, &handler->rx)) + panic("qman_create_fq(rx) failed"); + memset(&opts, 0, sizeof(opts)); + opts.we_mask = QM_INITFQ_WE_FQCTRL | QM_INITFQ_WE_CONTEXTA; + opts.fqd.fq_ctrl = QM_FQCTRL_CTXASTASHING; + opts.fqd.context_a.stashing.data_cl = STASH_DATA_CL; + opts.fqd.context_a.stashing.context_cl = STASH_CTX_CL; + if (qman_init_fq(&handler->rx, QMAN_INITFQ_FLAG_SCHED | + QMAN_INITFQ_FLAG_LOCAL, &opts)) + panic("qman_init_fq(rx) failed"); + /* Set up tx */ + memset(&handler->tx, 0, sizeof(handler->tx)); + if (qman_create_fq(handler->fqid_tx, QMAN_FQ_FLAG_NO_MODIFY, + &handler->tx)) + panic("qman_create_fq(tx) failed"); +} + +static void init_phase2(void) +{ + int loop; + u32 fqid = 0; + u32 lfsr = 0xdeadbeef; + struct hp_cpu *hp_cpu; + struct hp_handler *handler; + + for (loop = 0; loop < HP_PER_CPU; loop++) { + list_for_each_entry(hp_cpu, &hp_cpu_list, node) { + int ret; + + if (!loop) + hp_cpu->iterator = list_first_entry( + &hp_cpu->handlers, + struct hp_handler, node); + else + hp_cpu->iterator = list_entry( + hp_cpu->iterator->node.next, + struct hp_handler, node); + /* Rx FQID is the previous handler's Tx FQID */ + hp_cpu->iterator->fqid_rx = fqid; + /* Allocate new FQID for Tx */ + ret = qman_alloc_fqid(&fqid); + if (ret) + panic("qman_alloc_fqid() failed"); + hp_cpu->iterator->fqid_tx = fqid; + /* Rx mixer is the previous handler's Tx mixer */ + hp_cpu->iterator->rx_mixer = lfsr; + /* Get new mixer for Tx */ + lfsr = do_lfsr(lfsr); + hp_cpu->iterator->tx_mixer = lfsr; + } + } + /* Fix up the first handler (fqid_rx==0, rx_mixer=0xdeadbeef) */ + hp_cpu = list_first_entry(&hp_cpu_list, struct hp_cpu, node); + handler = list_first_entry(&hp_cpu->handlers, struct hp_handler, node); + BUG_ON((handler->fqid_rx != 0) || (handler->rx_mixer != 0xdeadbeef)); + handler->fqid_rx = fqid; + handler->rx_mixer = lfsr; + /* and tag it as our "special" handler */ + special_handler = handler; +} + +static void init_phase3(void) +{ + int loop; + struct hp_cpu *hp_cpu; + + for (loop = 0; loop < HP_PER_CPU; loop++) { + list_for_each_entry(hp_cpu, &hp_cpu_list, node) { + if (!loop) + hp_cpu->iterator = list_first_entry( + &hp_cpu->handlers, + struct hp_handler, node); + else + hp_cpu->iterator = list_entry( + hp_cpu->iterator->node.next, + struct hp_handler, node); + preempt_disable(); + if (hp_cpu->processor_id == smp_processor_id()) + init_handler(hp_cpu->iterator); + else + smp_call_function_single(hp_cpu->processor_id, + init_handler, hp_cpu->iterator, 1); + preempt_enable(); + } + } +} + +static void send_first_frame(void *ignore) +{ + u32 *p = special_handler->frame_ptr; + u32 lfsr = HP_FIRST_WORD; + int loop; + struct qm_fd fd; + + BUG_ON(special_handler->processor_id != smp_processor_id()); + memset(&fd, 0, sizeof(fd)); + qm_fd_addr_set64(&fd, special_handler->addr); + fd.format = qm_fd_contig_big; + fd.length29 = HP_NUM_WORDS * 4; + for (loop = 0; loop < HP_NUM_WORDS; loop++, p++) { + if (*p != lfsr) + panic("corrupt frame data"); + *p ^= special_handler->tx_mixer; + lfsr = do_lfsr(lfsr); + } + pr_info("Sending first frame\n"); + if (qman_enqueue(&special_handler->tx, &fd, 0)) + panic("qman_enqueue() failed"); +} + +void qman_test_stash(void) +{ +#ifndef __rtems__ + if (cpumask_weight(cpu_online_mask) < 2) { + pr_info("%s(): skip - only 1 CPU\n", __func__); + return; + } +#endif /* __rtems__ */ + + pr_info("%s(): Starting\n", __func__); + + hp_cpu_list_length = 0; + loop_counter = 0; + hp_handler_slab = kmem_cache_create("hp_handler_slab", + sizeof(struct hp_handler), L1_CACHE_BYTES, + SLAB_HWCACHE_ALIGN, NULL); + if (!hp_handler_slab) + panic("kmem_cache_create() failed"); + + allocate_frame_data(); + + /* Init phase 1 */ + pr_info("Creating %d handlers per cpu...\n", HP_PER_CPU); + if (on_all_cpus(create_per_cpu_handlers)) + panic("on_each_cpu() failed"); + pr_info("Number of cpus: %d, total of %d handlers\n", + hp_cpu_list_length, hp_cpu_list_length * HP_PER_CPU); + + init_phase2(); + + init_phase3(); + + preempt_disable(); + if (special_handler->processor_id == smp_processor_id()) + send_first_frame(NULL); + else + smp_call_function_single(special_handler->processor_id, + send_first_frame, NULL, 1); + preempt_enable(); + + wait_event(queue, loop_counter == HP_LOOPS); + deallocate_frame_data(); + if (on_all_cpus(destroy_per_cpu_handlers)) + panic("on_each_cpu() failed"); + kmem_cache_destroy(hp_handler_slab); + pr_info("%s(): Finished\n", __func__); +} diff --git a/linux/drivers/soc/fsl/qbman/qman_utils.c b/linux/drivers/soc/fsl/qbman/qman_utils.c new file mode 100644 index 00000000..5b85f037 --- /dev/null +++ b/linux/drivers/soc/fsl/qbman/qman_utils.c @@ -0,0 +1,309 @@ +#include + +#include + +/* Copyright 2008 - 2015 Freescale Semiconductor, Inc. + * + * Redistribution and use in source and binary forms, with or without + * modification, are permitted provided that the following conditions are met: + * * Redistributions of source code must retain the above copyright + * notice, this list of conditions and the following disclaimer. + * * Redistributions in binary form must reproduce the above copyright + * notice, this list of conditions and the following disclaimer in the + * documentation and/or other materials provided with the distribution. + * * Neither the name of Freescale Semiconductor nor the + * names of its contributors may be used to endorse or promote products + * derived from this software without specific prior written permission. + * + * ALTERNATIVELY, this software may be distributed under the terms of the + * GNU General Public License ("GPL") as published by the Free Software + * Foundation, either version 2 of that License or (at your option) any + * later version. + * + * THIS SOFTWARE IS PROVIDED BY Freescale Semiconductor ``AS IS'' AND ANY + * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED + * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE + * DISCLAIMED. IN NO EVENT SHALL Freescale Semiconductor BE LIABLE FOR ANY + * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES + * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; + * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND + * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT + * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS + * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. + */ + +#include "qman_priv.h" + +/* --- FQID Pool --- */ + +struct qman_fqid_pool { + /* Base and size of the FQID range */ + u32 fqid_base; + u32 total; + /* Number of FQIDs currently "allocated" */ + u32 used; + /* Allocation optimisation. When 'usedfqid_base = fqid_start; + pool->total = num; + pool->used = 0; + pool->next = 0; + pool->bits = kzalloc(QNUM_BYTES(num), GFP_KERNEL); + if (!pool->bits) { + kfree(pool); + return NULL; + } + /* If num is not an even multiple of QLONG_BITS (or even 8, for + * byte-oriented searching) then we fill the trailing bits with 1, to + * make them look allocated (permanently). */ + for (i = num + 1; i < QNUM_BITS(num); i++) + set_bit(i, pool->bits); + return pool; +} +EXPORT_SYMBOL(qman_fqid_pool_create); + +int qman_fqid_pool_destroy(struct qman_fqid_pool *pool) +{ + int ret = pool->used; + + kfree(pool->bits); + kfree(pool); + return ret; +} +EXPORT_SYMBOL(qman_fqid_pool_destroy); + +int qman_fqid_pool_alloc(struct qman_fqid_pool *pool, u32 *fqid) +{ + int ret; + + if (pool->used == pool->total) + return -ENOMEM; + *fqid = pool->fqid_base + pool->next; + ret = test_and_set_bit(pool->next, pool->bits); + BUG_ON(ret); + if (++pool->used == pool->total) + return 0; + pool->next = find_next_zero_bit(pool->bits, pool->total, pool->next); + if (pool->next >= pool->total) + pool->next = find_first_zero_bit(pool->bits, pool->total); + BUG_ON(pool->next >= pool->total); + return 0; +} +EXPORT_SYMBOL(qman_fqid_pool_alloc); + +void qman_fqid_pool_free(struct qman_fqid_pool *pool, u32 fqid) +{ + int ret; + + fqid -= pool->fqid_base; + ret = test_and_clear_bit(fqid, pool->bits); + BUG_ON(!ret); + if (pool->used-- == pool->total) + pool->next = fqid; +} +EXPORT_SYMBOL(qman_fqid_pool_free); + +u32 qman_fqid_pool_used(struct qman_fqid_pool *pool) +{ + return pool->used; +} +EXPORT_SYMBOL(qman_fqid_pool_used); + +static DECLARE_DPAA_RESOURCE(fqalloc); /* FQID allocator */ +static DECLARE_DPAA_RESOURCE(qpalloc); /* pool-channel allocator */ +static DECLARE_DPAA_RESOURCE(cgralloc); /* CGR ID allocator */ + +/* FQID allocator front-end */ + +int qman_alloc_fqid_range(u32 *result, u32 count, u32 align, int partial) +{ + return dpaa_resource_new(&fqalloc, result, count, align, partial); +} +EXPORT_SYMBOL(qman_alloc_fqid_range); + +static int fq_cleanup(u32 fqid) +{ + return qman_shutdown_fq(fqid) == 0; +} + +void qman_release_fqid_range(u32 fqid, u32 count) +{ + u32 total_invalid = dpaa_resource_release(&fqalloc, + fqid, count, fq_cleanup); + + if (total_invalid) + pr_err("FQID range [%d..%d] (%d) had %d leaks\n", + fqid, fqid + count - 1, count, total_invalid); +} +EXPORT_SYMBOL(qman_release_fqid_range); + +int qman_reserve_fqid_range(u32 fqid, u32 count) +{ + return dpaa_resource_reserve(&fqalloc, fqid, count); +} +EXPORT_SYMBOL(qman_reserve_fqid_range); + +void qman_seed_fqid_range(u32 fqid, u32 count) +{ + dpaa_resource_seed(&fqalloc, fqid, count); +} +EXPORT_SYMBOL(qman_seed_fqid_range); + +/* Pool-channel allocator front-end */ + +int qman_alloc_pool_range(u32 *result, u32 count, u32 align, int partial) +{ + return dpaa_resource_new(&qpalloc, result, count, align, partial); +} +EXPORT_SYMBOL(qman_alloc_pool_range); + +static int qpool_cleanup(u32 qp) +{ + /* We query all FQDs starting from + * FQID 1 until we get an "invalid FQID" error, looking for non-OOS FQDs + * whose destination channel is the pool-channel being released. + * When a non-OOS FQD is found we attempt to clean it up */ + struct qman_fq fq = { + .fqid = 1 + }; + int err; + + do { + struct qm_mcr_queryfq_np np; + + err = qman_query_fq_np(&fq, &np); + if (err) + /* FQID range exceeded, found no problems */ + return 1; + if ((np.state & QM_MCR_NP_STATE_MASK) != QM_MCR_NP_STATE_OOS) { + struct qm_fqd fqd; + + err = qman_query_fq(&fq, &fqd); + BUG_ON(err); + if (fqd.dest.channel == qp) { + /* The channel is the FQ's target, clean it */ + if (qman_shutdown_fq(fq.fqid) != 0) + /* Couldn't shut down the FQ + so the pool must be leaked */ + return 0; + } + } + /* Move to the next FQID */ + fq.fqid++; + } while (1); +} + +void qman_release_pool_range(u32 qp, u32 count) +{ + u32 total_invalid = dpaa_resource_release(&qpalloc, + qp, count, qpool_cleanup); + + if (total_invalid) { + /* Pool channels are almost always used individually */ + if (count == 1) + pr_err("Pool channel 0x%x had %d leaks\n", + qp, total_invalid); + else + pr_err("Pool channels [%d..%d] (%d) had %d leaks\n", + qp, qp + count - 1, count, total_invalid); + } +} +EXPORT_SYMBOL(qman_release_pool_range); + +void qman_seed_pool_range(u32 poolid, u32 count) +{ + dpaa_resource_seed(&qpalloc, poolid, count); + +} +EXPORT_SYMBOL(qman_seed_pool_range); + +int qman_reserve_pool_range(u32 poolid, u32 count) +{ + return dpaa_resource_reserve(&qpalloc, poolid, count); +} +EXPORT_SYMBOL(qman_reserve_pool_range); + + +/* CGR ID allocator front-end */ + +int qman_alloc_cgrid_range(u32 *result, u32 count, u32 align, int partial) +{ + return dpaa_resource_new(&cgralloc, result, count, align, partial); +} +EXPORT_SYMBOL(qman_alloc_cgrid_range); + +static int cqr_cleanup(u32 cgrid) +{ + /* We query all FQDs starting from + * FQID 1 until we get an "invalid FQID" error, looking for non-OOS FQDs + * whose CGR is the CGR being released. + */ + struct qman_fq fq = { + .fqid = 1 + }; + int err; + + do { + struct qm_mcr_queryfq_np np; + + err = qman_query_fq_np(&fq, &np); + if (err) + /* FQID range exceeded, found no problems */ + return 1; + if ((np.state & QM_MCR_NP_STATE_MASK) != QM_MCR_NP_STATE_OOS) { + struct qm_fqd fqd; + + err = qman_query_fq(&fq, &fqd); + BUG_ON(err); + if ((fqd.fq_ctrl & QM_FQCTRL_CGE) && + (fqd.cgid == cgrid)) { + pr_err("CRGID 0x%x is being used by FQID 0x%x," + " CGR will be leaked\n", + cgrid, fq.fqid); + return 1; + } + } + /* Move to the next FQID */ + fq.fqid++; + } while (1); +} + +void qman_release_cgrid_range(u32 cgrid, u32 count) +{ + u32 total_invalid = dpaa_resource_release(&cgralloc, + cgrid, count, cqr_cleanup); + if (total_invalid) + pr_err("CGRID range [%d..%d] (%d) had %d leaks\n", + cgrid, cgrid + count - 1, count, total_invalid); +} +EXPORT_SYMBOL(qman_release_cgrid_range); + +void qman_seed_cgrid_range(u32 cgrid, u32 count) +{ + dpaa_resource_seed(&cgralloc, cgrid, count); + +} +EXPORT_SYMBOL(qman_seed_cgrid_range); -- cgit v1.2.3