staging/lustre/llite: update ras window correctly
authorBobi Jam <bobijam.xu@intel.com>
Thu, 3 Nov 2016 01:25:02 +0000 (21:25 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Mon, 7 Nov 2016 10:36:46 +0000 (11:36 +0100)
When stride-RA hit case miss, we only reset normal sequential
read-ahead window, but not reset the stride IO to avoid the overhead
of re-detecting stride IO. While when the normal RA window is set
to not insect with the stride-RA window, when we try to increase
the stride-RA window length later, the presumption does not hold.

This patch resets the stride IO as well in this case.

Signed-off-by: Bobi Jam <bobijam.xu@intel.com>
Reviewed-on: http://review.whamcloud.com/23032
Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-8683
Reviewed-by: wangdi <di.wang@intel.com>
Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com>
Signed-off-by: Oleg Drokin <green@linuxhacker.ru>
Reviewed-by: James Simmons <jsimmons@infradead.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/staging/lustre/lustre/llite/rw.c

index d2515a8005d3641fb7be6732ff234ab6fbe07d7b..e34017d7c8da7b94f63e1ce2bfcfc743cc835171 100644 (file)
@@ -809,13 +809,20 @@ static void ras_update(struct ll_sb_info *sbi, struct inode *inode,
                if (ra_miss) {
                        if (index_in_stride_window(ras, index) &&
                            stride_io_mode(ras)) {
-                               /*If stride-RA hit cache miss, the stride dector
-                                *will not be reset to avoid the overhead of
-                                *redetecting read-ahead mode
-                                */
                                if (index != ras->ras_last_readpage + 1)
                                        ras->ras_consecutive_pages = 0;
                                ras_reset(inode, ras, index);
+
+                               /* If stride-RA hit cache miss, the stride
+                                * detector will not be reset to avoid the
+                                * overhead of redetecting read-ahead mode,
+                                * but on the condition that the stride window
+                                * is still intersect with normal sequential
+                                * read-ahead window.
+                                */
+                               if (ras->ras_window_start <
+                                   ras->ras_stride_offset)
+                                       ras_stride_reset(ras);
                                RAS_CDEBUG(ras);
                        } else {
                                /* Reset both stride window and normal RA