summaryrefslogtreecommitdiffstats
path: root/c/src/lib/libbsp/powerpc/motorola_powerpc/include
diff options
context:
space:
mode:
authorTill Straumann <strauman@slac.stanford.edu>2005-11-04 01:00:33 +0000
committerTill Straumann <strauman@slac.stanford.edu>2005-11-04 01:00:33 +0000
commitbe45f8e250cde1ab212f61ddd2726a6ecaa9a0b7 (patch)
tree67eb4ad57221cb36b869e82203346154e9e5731b /c/src/lib/libbsp/powerpc/motorola_powerpc/include
parent2005-11-03 <strauman@slac.stanford.edu> (diff)
downloadrtems-be45f8e250cde1ab212f61ddd2726a6ecaa9a0b7.tar.bz2
2005-11-03 <strauman@slac.stanford.edu>
* Makefile.am, include/bsp.h: Added new shared pretaskinghook.c and zerobss.c files to list to be made. Added some explanations about CPU <-> PCI <-> VME address mapping issues.
Diffstat (limited to 'c/src/lib/libbsp/powerpc/motorola_powerpc/include')
-rw-r--r--c/src/lib/libbsp/powerpc/motorola_powerpc/include/bsp.h70
1 files changed, 61 insertions, 9 deletions
diff --git a/c/src/lib/libbsp/powerpc/motorola_powerpc/include/bsp.h b/c/src/lib/libbsp/powerpc/motorola_powerpc/include/bsp.h
index a1d73541e2..10bac559a2 100644
--- a/c/src/lib/libbsp/powerpc/motorola_powerpc/include/bsp.h
+++ b/c/src/lib/libbsp/powerpc/motorola_powerpc/include/bsp.h
@@ -32,13 +32,63 @@
#define CONFIGURE_INTERRUPT_STACK_MEMORY (16 * 1024)
+/*
+ * diagram illustrating the role of the configuration
+ * constants
+ * PCI_MEM_WIN0: CPU starting addr where PCI memory space is visible
+ * PCI_MEM_BASE: CPU address of PCI mem addr. zero. (regardless of this
+ * address being 'visible' or not!).
+ * _VME_A32_WIN0_ON_PCI: PCI starting addr of the 1st window to VME
+ * _VME_A32_WIN0_ON_VME: VME address of that same window
+ *
+ * AFAIK, only PreP boards have a non-zero PCI_MEM_BASE (i.e., an offset between
+ * CPU and PCI addresses). The mvme2300 'ppcbug' firmware configures the PCI
+ * bus using PCI base addresses! I.e., drivers need to add PCI_MEM_BASE to
+ * the base address read from PCI config.space in order to translate that
+ * into a CPU address.
+ *
+ * NOTE: VME addresses should NEVER be translated using these constants!
+ * they are strictly for BSP internal use. Drivers etc. should use
+ * the translation routines int VME.h (BSP_vme2local_adrs/BSP_local2vme_adrs).
+ *
+ * CPU ADDR PCI_ADDR VME ADDR
+ *
+ * 00000000 XXXXXXXX XXXXXXXX
+ * ^ ^ ........
+ * | |
+ * | | e.g., RAM XXXXXXXX
+ * | | 00000000
+ * | | ......... ^
+ * | | (possible offset |
+ * | | between pci and XXXXXXXX | ......
+ * | | cpu addresses) |
+ * | v |
+ * | PCI_MEM_BASE -------------> 00000000 --------------- |
+ * | ........ ........ ^ |
+ * | invisible | |
+ * | ........ from CPU | |
+ * v | |
+ * PCI_MEM_WIN0 ============= first visible PCI addr | |
+ * | |
+ * pci devices pci window | |
+ * visible here v v
+ * mapped by ========== _VME_A32_WIN0_ON_PCI ======= _VME_A32_WIN0_ON_VME
+ * vme window
+ * VME devices hostbridge mapped by
+ * visible here universe
+ * =====================================================
+ *
+ */
+
/* fundamental addresses for BSP (CHRPxxx and PREPxxx are from libcpu/io.h) */
#if defined(mvme2100)
#define _IO_BASE CHRP_ISA_IO_BASE
#define _ISA_MEM_BASE CHRP_ISA_MEM_BASE
/* address of our ram on the PCI bus */
#define PCI_DRAM_OFFSET CHRP_PCI_DRAM_OFFSET
+/* offset of pci memory as seen from the CPU */
#define PCI_MEM_BASE 0
+/* where (in CPU addr. space) does the PCI window start */
#define PCI_MEM_WIN0 0x80000000
#else
@@ -102,6 +152,10 @@
*/
extern unsigned int BSP_mem_size;
/*
+ * Start of the heap
+ */
+extern unsigned int BSP_heap_start;
+/*
* PCI Bus Frequency
*/
extern unsigned int BSP_bus_frequency;
@@ -126,15 +180,13 @@ extern int BSP_connect_clock_handler (void);
/* clear hostbridge errors
*
- * enableMCP: whether to enable MCP checkstop / machine check interrupts
- * on the hostbridge and in HID0.
- *
- * NOTE: HID0 and MEREN are left alone if this flag is 0
- *
- * quiet : be silent
- *
- * RETURNS : raven MERST register contents (lowermost 16 bits), 0 if
- * there were no errors
+ * NOTE: The routine returns always (-1) if 'enableMCP==1'
+ * [semantics needed by libbspExt] if the MCP input is not wired.
+ * It returns and clears the error bits of the PCI status register.
+ * MCP support is disabled because:
+ * a) the 2100 has no raven chip
+ * b) the raven (2300) would raise machine check interrupts
+ * on PCI config space access to empty slots.
*/
extern unsigned long _BSP_clear_hostbridge_errors(int enableMCP, int quiet);