From mboxrd@z Thu Jan 1 00:00:00 1970 From: Oleg Drokin Date: Wed, 2 Nov 2016 21:25:02 -0400 Subject: [lustre-devel] [PATCH 12/14] staging/lustre/llite: update ras window correctly In-Reply-To: <1478136304-867780-1-git-send-email-green@linuxhacker.ru> References: <1478136304-867780-1-git-send-email-green@linuxhacker.ru> Message-ID: <1478136304-867780-13-git-send-email-green@linuxhacker.ru> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: Greg Kroah-Hartman , devel@driverdev.osuosl.org, Andreas Dilger , James Simmons Cc: Linux Kernel Mailing List , Lustre Development List , Bobi Jam , Oleg Drokin From: Bobi Jam 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 Reviewed-on: http://review.whamcloud.com/23032 Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-8683 Reviewed-by: wangdi Reviewed-by: Jinshan Xiong Signed-off-by: Oleg Drokin --- drivers/staging/lustre/lustre/llite/rw.c | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/drivers/staging/lustre/lustre/llite/rw.c b/drivers/staging/lustre/lustre/llite/rw.c index d2515a8..e34017d 100644 --- a/drivers/staging/lustre/lustre/llite/rw.c +++ b/drivers/staging/lustre/lustre/llite/rw.c @@ -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 -- 2.7.4