libblock: Add rtems_bdbuf_peek()

Adds a peek function that allows (for example) a file system to suggest
the next blocks that should be used for read ahead. This can increase
the read speed of fragmented files.

Update #3689

4 files changed, 109 insertions, 24 deletions

diff --git a/cpukit/include/rtems/bdbuf.h b/cpukit/include/rtems/bdbuf.h
index fbb4fc05e9..4d11a47619 100644
--- a/cpukit/include/rtems/bdbuf.h
+++ b/cpukit/include/rtems/bdbuf.h
@@ -540,6 +540,29 @@ rtems_bdbuf_read (
 );
 
 /**
+ * @brief Give a hint which blocks should be cached next.
+ *
+ * Provide a hint to the read ahead mechanism which blocks should be cached
+ * next. This overwrites the default linear pattern. You should use it in (for
+ * example) a file system to tell bdbuf where the next part of a fragmented file
+ * is. If you know the length of the file, you can provide that too.
+ *
+ * Before you can use this function, the rtems_bdbuf_init() routine must be
+ * called at least once to initialize everything. Otherwise you might get
+ * unexpected results.
+ *
+ * @param dd [in] The disk device.
+ * @param block [in] Linear media block number.
+ * @param nr_blocks [in] Number of consecutive blocks that can be pre-fetched.
+ */
+void
+rtems_bdbuf_peek (
+  rtems_disk_device *dd,
+  rtems_blkdev_bnum block,
+  uint32_t nr_blocks
+);
+
+/**
  * Release the buffer obtained by a read call back to the cache. If the buffer
  * was obtained by a get call and was not already in the cache the release
  * modified call should be used. A buffer released with this call obtained by a
diff --git a/cpukit/include/rtems/diskdevs.h b/cpukit/include/rtems/diskdevs.h
index 85d157dcd5..527a8cc955 100644
--- a/cpukit/include/rtems/diskdevs.h
+++ b/cpukit/include/rtems/diskdevs.h
@@ -59,6 +59,11 @@ typedef int (*rtems_block_device_ioctl)(
 #define RTEMS_DISK_READ_AHEAD_NO_TRIGGER ((rtems_blkdev_bnum) -1)
 
 /**
+ * @brief Size value to set number of blocks based on config and disk size.
+ */
+#define RTEMS_DISK_READ_AHEAD_SIZE_AUTO (0)
+
+/**
  * @brief Block device read-ahead control.
  */
 typedef struct {
@@ -71,7 +76,8 @@ typedef struct {
    * @brief Block value to trigger the read-ahead request.
    *
    * A value of @ref RTEMS_DISK_READ_AHEAD_NO_TRIGGER will disable further
-   * read-ahead requests since no valid block can have this value.
+   * read-ahead requests (except the ones triggered by @a rtems_bdbuf_peek)
+   * since no valid block can have this value.
    */
   rtems_blkdev_bnum trigger;
 
@@ -82,6 +88,14 @@ typedef struct {
    * be arbitrary.
    */
   rtems_blkdev_bnum next;
+
+  /**
+   * @brief Size of the next read-ahead request in blocks.
+   *
+   * A value of @ref RTEMS_DISK_READ_AHEAD_SIZE_AUTO will try to read the rest
+   * of the disk but at most the configured max_read_ahead_blocks.
+   */
+  uint32_t nr_blocks;
 } rtems_blkdev_read_ahead;
 
 /**
@@ -110,11 +124,17 @@ typedef struct {
   /**
    * @brief Read-ahead transfer count.
    *
-   * Each read-ahead transfer may read multiple blocks.
+   * Each read-ahead transfer may read multiple blocks. This counts all
+   * transfers (including peeks).
    */
   uint32_t read_ahead_transfers;
 
   /**
+   * @brief Read-ahead transfers caused by a peek.
+   */
+  uint32_t read_ahead_peeks;
+
+  /**
    * @brief Count of blocks transfered from the device.
    */
   uint32_t read_blocks;
diff --git a/cpukit/libblock/src/bdbuf.c b/cpukit/libblock/src/bdbuf.c
index a7d471507c..8aadf9c8ae 100644
--- a/cpukit/libblock/src/bdbuf.c
+++ b/cpukit/libblock/src/bdbuf.c
@@ -2019,6 +2019,23 @@ rtems_bdbuf_read_ahead_reset (rtems_disk_device *dd)
 }
 
 static void
+rtems_bdbuf_read_ahead_add_to_chain (rtems_disk_device *dd)
+{
+  rtems_status_code sc;
+  rtems_chain_control *chain = &bdbuf_cache.read_ahead_chain;
+
+  if (rtems_chain_is_empty (chain))
+  {
+    sc = rtems_event_send (bdbuf_cache.read_ahead_task,
+                           RTEMS_BDBUF_READ_AHEAD_WAKE_UP);
+    if (sc != RTEMS_SUCCESSFUL)
+      rtems_bdbuf_fatal (RTEMS_BDBUF_FATAL_RA_WAKE_UP);
+  }
+
+  rtems_chain_append_unprotected (chain, &dd->read_ahead.node);
+}
+
+static void
 rtems_bdbuf_check_read_ahead_trigger (rtems_disk_device *dd,
                                       rtems_blkdev_bnum  block)
 {
@@ -2026,18 +2043,8 @@ rtems_bdbuf_check_read_ahead_trigger (rtems_disk_device *dd,
       && dd->read_ahead.trigger == block
       && !rtems_bdbuf_is_read_ahead_active (dd))
   {
-    rtems_status_code sc;
-    rtems_chain_control *chain = &bdbuf_cache.read_ahead_chain;
-
-    if (rtems_chain_is_empty (chain))
-    {
-      sc = rtems_event_send (bdbuf_cache.read_ahead_task,
-                             RTEMS_BDBUF_READ_AHEAD_WAKE_UP);
-      if (sc != RTEMS_SUCCESSFUL)
-        rtems_bdbuf_fatal (RTEMS_BDBUF_FATAL_RA_WAKE_UP);
-    }
-
-    rtems_chain_append_unprotected (chain, &dd->read_ahead.node);
+    dd->read_ahead.nr_blocks = RTEMS_DISK_READ_AHEAD_SIZE_AUTO;
+    rtems_bdbuf_read_ahead_add_to_chain(dd);
   }
 }
 
@@ -2112,6 +2119,24 @@ rtems_bdbuf_read (rtems_disk_device   *dd,
   return sc;
 }
 
+void
+rtems_bdbuf_peek (rtems_disk_device *dd,
+                  rtems_blkdev_bnum block,
+                  uint32_t nr_blocks)
+{
+  rtems_bdbuf_lock_cache ();
+
+  if (bdbuf_cache.read_ahead_enabled && nr_blocks > 0)
+  {
+    rtems_bdbuf_read_ahead_reset(dd);
+    dd->read_ahead.next = block;
+    dd->read_ahead.nr_blocks = nr_blocks;
+    rtems_bdbuf_read_ahead_add_to_chain(dd);
+  }
+
+  rtems_bdbuf_unlock_cache ();
+}
+
 static rtems_status_code
 rtems_bdbuf_check_bd_and_lock_cache (rtems_bdbuf_buffer *bd, const char *kind)
 {
@@ -2952,18 +2977,33 @@ rtems_bdbuf_read_ahead_task (rtems_task_argument arg)
 
         if (bd != NULL)
         {
-          uint32_t transfer_count = dd->block_count - block;
+          uint32_t transfer_count = dd->read_ahead.nr_blocks;
+          uint32_t blocks_until_end_of_disk = dd->block_count - block;
           uint32_t max_transfer_count = bdbuf_config.max_read_ahead_blocks;
 
-          if (transfer_count >= max_transfer_count)
-          {
-            transfer_count = max_transfer_count;
-            dd->read_ahead.trigger = block + transfer_count / 2;
-            dd->read_ahead.next = block + transfer_count;
-          }
-          else
-          {
-            dd->read_ahead.trigger = RTEMS_DISK_READ_AHEAD_NO_TRIGGER;
+          if (transfer_count == RTEMS_DISK_READ_AHEAD_SIZE_AUTO) {
+            transfer_count = blocks_until_end_of_disk;
+
+            if (transfer_count >= max_transfer_count)
+            {
+              transfer_count = max_transfer_count;
+              dd->read_ahead.trigger = block + transfer_count / 2;
+              dd->read_ahead.next = block + transfer_count;
+            }
+            else
+            {
+              dd->read_ahead.trigger = RTEMS_DISK_READ_AHEAD_NO_TRIGGER;
+            }
+          } else {
+            if (transfer_count > blocks_until_end_of_disk) {
+              transfer_count = blocks_until_end_of_disk;
+            }
+
+            if (transfer_count > max_transfer_count) {
+              transfer_count = max_transfer_count;
+            }
+
+            ++dd->stats.read_ahead_peeks;
           }
 
           ++dd->stats.read_ahead_transfers;
diff --git a/cpukit/libblock/src/blkdev-print-stats.c b/cpukit/libblock/src/blkdev-print-stats.c
index 8edf24fd8c..539ff08157 100644
--- a/cpukit/libblock/src/blkdev-print-stats.c
+++ b/cpukit/libblock/src/blkdev-print-stats.c
@@ -47,6 +47,7 @@ void rtems_blkdev_print_stats(
      " READ HITS            | %" PRIu32 "\n"
      " READ MISSES          | %" PRIu32 "\n"
      " READ AHEAD TRANSFERS | %" PRIu32 "\n"
+     " READ AHEAD PEEKS     | %" PRIu32 "\n"
      " READ BLOCKS          | %" PRIu32 "\n"
      " READ ERRORS          | %" PRIu32 "\n"
      " WRITE TRANSFERS      | %" PRIu32 "\n"
@@ -59,6 +60,7 @@ void rtems_blkdev_print_stats(
      stats->read_hits,
      stats->read_misses,
      stats->read_ahead_transfers,
+     stats->read_ahead_peeks,
      stats->read_blocks,
      stats->read_errors,
      stats->write_transfers,