Zswap is a cache which compresses the pages that are being swapped out
and stores them into a dynamically allocated RAM-based memory pool.
Experiments have shown that around 10-20% of pages stored in zswap are
same-filled pages (i.e. contents of the page are all same), but these
pages are handled as normal pages by compressing and allocating memory
in the pool.
This patch adds a check in zswap_frontswap_store() to identify
same-filled page before compression of the page. If the page is a
same-filled page, set zswap_entry.length to zero, save the same-filled
value and skip the compression of the page and alloction of memory in
zpool. In zswap_frontswap_load(), check if value of zswap_entry.length
is zero corresponding to the page to be loaded. If zswap_entry.length
is zero, fill the page with same-filled value. This saves the
decompression time during load.
On a ARM Quad Core 32-bit device with 1.5GB RAM by launching and
relaunching different applications, out of ~64000 pages stored in zswap,
~11000 pages were same-value filled pages (including zero-filled pages)
and ~9000 pages were zero-filled pages.
An average of 17% of pages(including zero-filled pages) in zswap are
same-value filled pages and 14% pages are zero-filled pages. An average
of 3% of pages are same-filled non-zero pages.
The below table shows the execution time profiling with the patch.
Baseline With patch % Improvement
-----------------------------------------------------------------
*Zswap Store Time 26.5ms 18ms 32%
(of same value pages)
*Zswap Load Time
(of same value pages) 25.5ms 13ms 49%
-----------------------------------------------------------------
On Ubuntu PC with 2GB RAM, while executing kernel build and other test
scripts and running multimedia applications, out of 360000 pages stored
in zswap 78000(~22%) of pages were found to be same-value filled pages
(including zero-filled pages) and 64000(~17%) are zero-filled pages. So
an average of %5 of pages are same-filled non-zero pages.
The below table shows the execution time profiling with the patch.
Baseline With patch % Improvement
-----------------------------------------------------------------
*Zswap Store Time 91ms 74ms 19%
(of same value pages)
*Zswap Load Time 50ms 7.5ms 85%
(of same value pages)
-----------------------------------------------------------------
*The execution times may vary with test device used.
Dan said:
: I did test this patch out this week, and I added some instrumentation to
: check the performance impact, and tested with a small program to try to
: check the best and worst cases.
:
: When doing a lot of swap where all (or almost all) pages are same-value, I
: found this patch does save both time and space, significantly. The exact
: improvement in time and space depends on which compressor is being used,
: but roughly agrees with the numbers you listed.
:
: In the worst case situation, where all (or almost all) pages have the
: same-value *except* the final long (meaning, zswap will check each long on
: the entire page but then still have to pass the page to the compressor),
: the same-value check is around 10-15% of the total time spent in
: zswap_frontswap_store(). That's a not-insignificant amount of time, but
: it's not huge. Considering that most systems will probably be swapping
: pages that aren't similar to the worst case (although I don't have any
: data to know that), I'd say the improvement is worth the possible
: worst-case performance impact.
[srividya.dr@samsung.com: add memset_l instead of for loop]
Link: http://lkml.kernel.org/r/20171018104832epcms5p1b2232e2236258de3d03d1344dde9fce0@epcms5p1
Signed-off-by: Srividya Desireddy <srividya.dr@samsung.com>
Acked-by: Dan Streetman <ddstreet@ieee.org>
Cc: Seth Jennings <sjenning@redhat.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: Dinakar Reddy Pathireddy <dinakar.p@samsung.com>
Cc: SHARAN ALLUR <sharan.allur@samsung.com>
Cc: RAJIB BASU <rajib.basu@samsung.com>
Cc: JUHUN KIM <juhunkim@samsung.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Timofey Titovets <nefelim4ag@gmail.com>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
slab_state is being set to "UP" in create_kmalloc_caches(), and later on
we set it again in kmem_cache_init_late(), but slab_state does not
change in the meantime.
Remove the redundant assignment from kmem_cache_init_late().
And unless I overlooked anything, the same goes for "slab_state = FULL".
slab_state is set to "FULL" in kmem_cache_init_late(), but it is later
being set again in cpucache_init(), which gets called from
do_initcall_level(). So remove the assignment from cpucache_init() as
well.
Link: http://lkml.kernel.org/r/20171215134452.GA1920@techadventures.net
Signed-off-by: Oscar Salvador <osalvador@techadventures.net>
Acked-by: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
calculate_alignment() function is only used inside slab_common.c. So
make it static and let the compiler do more optimizations.
After this patch there's a small improvement in text and data size.
$ gcc --version
gcc (GCC) 7.2.1 20171128
Before:
text data bss dec hex filename
9890457 3828702 1212364 14931523 e3d643 vmlinux
After:
text data bss dec hex filename
9890437 3828670 1212364 14931471 e3d60f vmlinux
Also I fixed a style problem reported by checkpatch.
WARNING: Missing a blank line after declarations
#53: FILE: mm/slab_common.c:286:
+ unsigned long ralign = cache_line_size();
+ while (size <= ralign / 2)
Link: http://lkml.kernel.org/r/20171210080132.406-1-bhlee.kernel@gmail.com
Signed-off-by: Byongho Lee <bhlee.kernel@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We should not reuse the dirty bh in jbd2 directly due to the following
situation:
1. When removing extent rec, we will dirty the bhs of extent rec and
truncate log at the same time, and hand them over to jbd2.
2. The bhs are submitted to jbd2 area successfully.
3. The write-back thread of device help flush the bhs to disk but
encounter write error due to abnormal storage link.
4. After a while the storage link become normal. Truncate log flush
worker triggered by the next space reclaiming found the dirty bh of
truncate log and clear its 'BH_Write_EIO' and then set it uptodate in
__ocfs2_journal_access():
ocfs2_truncate_log_worker
ocfs2_flush_truncate_log
__ocfs2_flush_truncate_log
ocfs2_replay_truncate_records
ocfs2_journal_access_di
__ocfs2_journal_access // here we clear io_error and set 'tl_bh' uptodata.
5. Then jbd2 will flush the bh of truncate log to disk, but the bh of
extent rec is still in error state, and unfortunately nobody will
take care of it.
6. At last the space of extent rec was not reduced, but truncate log
flush worker have given it back to globalalloc. That will cause
duplicate cluster problem which could be identified by fsck.ocfs2.
Sadly we can hardly revert this but set fs read-only in case of ruining
atomicity and consistency of space reclaim.
Link: http://lkml.kernel.org/r/5A6E8092.8090701@huawei.com
Fixes: acf8fdbe6a ("ocfs2: do not BUG if buffer not uptodate in __ocfs2_journal_access")
Signed-off-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Yiwen Jiang <jiangyiwen@huawei.com>
Reviewed-by: Changwei Ge <ge.changwei@h3c.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
ocfs2 supports trimming the underlying disk via the fstrim command. But
there is a problem, ocfs2 is a shared disk cluster file system, if the
user configures a scheduled fstrim job on each file system node, this
will trigger multiple nodes trimming a shared disk simultaneously, which
is very wasteful for CPU and IO consumption. This also might negatively
affect the lifetime of poor-quality SSD devices.
So we introduce a trimfs dlm lock to communicate with each other in this
case, which will make only one fstrim command to do the trimming on a
shared disk among the cluster. The fstrim commands from the other nodes
should wait for the first fstrim to finish and return success directly,
to avoid running the same trim on the shared disk again.
Link: http://lkml.kernel.org/r/1513228484-2084-2-git-send-email-ghe@suse.com
Signed-off-by: Gang He <ghe@suse.com>
Reviewed-by: Changwei Ge <ge.changwei@h3c.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
A crash issue was reported by John Lightsey with a call trace as follows:
ocfs2_split_extent+0x1ad3/0x1b40 [ocfs2]
ocfs2_change_extent_flag+0x33a/0x470 [ocfs2]
ocfs2_mark_extent_written+0x172/0x220 [ocfs2]
ocfs2_dio_end_io+0x62d/0x910 [ocfs2]
dio_complete+0x19a/0x1a0
do_blockdev_direct_IO+0x19dd/0x1eb0
__blockdev_direct_IO+0x43/0x50
ocfs2_direct_IO+0x8f/0xa0 [ocfs2]
generic_file_direct_write+0xb2/0x170
__generic_file_write_iter+0xc3/0x1b0
ocfs2_file_write_iter+0x4bb/0xca0 [ocfs2]
__vfs_write+0xae/0xf0
vfs_write+0xb8/0x1b0
SyS_write+0x4f/0xb0
system_call_fastpath+0x16/0x75
The BUG code told that extent tree wants to grow but no metadata was
reserved ahead of time. From my investigation into this issue, the root
cause it that although enough metadata is not reserved, there should be
enough for following use. Rightmost extent is merged into its left one
due to a certain times of marking extent written. Because during
marking extent written, we got many physically continuous extents. At
last, an empty extent showed up and the rightmost path is removed from
extent tree.
Add a new mechanism to reuse extent block cached in dealloc which were
just unlinked from extent tree to solve this crash issue.
Criteria is that during marking extents *written*, if extent rotation
and merging results in unlinking extent with growing extent tree later
without any metadata reserved ahead of time, try to reuse those extents
in dealloc in which deleted extents are cached.
Also, this patch addresses the issue John reported that ::dw_zero_count
is not calculated properly.
After applying this patch, the issue John reported was gone. Thanks for
the reproducer provided by John. And this patch has passed
ocfs2-test(29 cases) suite running by New H3C Group.
[ge.changwei@h3c.com: fix static checker warnning]
Link: http://lkml.kernel.org/r/63ADC13FD55D6546B7DECE290D39E373F29196AE@H3CMLB12-EX.srv.huawei-3com.com
[akpm@linux-foundation.org: brelse(NULL) is legal]
Link: http://lkml.kernel.org/r/1515479070-32653-2-git-send-email-ge.changwei@h3c.com
Signed-off-by: Changwei Ge <ge.changwei@h3c.com>
Reported-by: John Lightsey <john@nixnuts.net>
Tested-by: John Lightsey <john@nixnuts.net>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Joseph Qi <jiangqi903@gmail.com>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If metadata is corrupted such as 'invalid inode block', we will get
failed by calling 'mount()' and then set filesystem readonly as below:
ocfs2_mount
ocfs2_initialize_super
ocfs2_init_global_system_inodes
ocfs2_iget
ocfs2_read_locked_inode
ocfs2_validate_inode_block
ocfs2_error
ocfs2_handle_error
ocfs2_set_ro_flag(osb, 0); // set readonly
In this situation we need return -EROFS to 'mount.ocfs2', so that user
can fix it by fsck. And then mount again. In addition, 'mount.ocfs2'
should be updated correspondingly as it only return 1 for all errno.
And I will post a patch for 'mount.ocfs2' too.
Link: http://lkml.kernel.org/r/5A4302FA.2010606@huawei.com
Signed-off-by: Jun Piao <piaojun@huawei.com>
Reviewed-by: Alex Chen <alex.chen@huawei.com>
Reviewed-by: Joseph Qi <jiangqi903@gmail.com>
Reviewed-by: Changwei Ge <ge.changwei@h3c.com>
Reviewed-by: Gang He <ghe@suse.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When some nodes of cluster face with TCP connection fault, ocfs2 will
pick up a quorum to continue to work and other nodes will be fenced by
resetting host.
In order to decide which node should be fenced, ocfs2 leverages
o2quo_state::qs_holds. If that variable is reduced to zero, then a try
to decide if fence local node is performed. However, under a specific
scenario that local node is not disconnected from others at the same
time, above method has a problem to reduce ::qs_holds to zero.
Because, o2net 90s idle timer corresponding to different nodes is
triggered one after another.
node 2 node 3
90s idle timer elapses
clear ::qs_conn_bm
set hold
40s is passed
90 idle timer elapses
clear ::qs_conn_bm
set hold
still up timer elapses
clear hold (NOT to zero )
90s idle timer elapses AGAIN
still up timer elapses.
clear hold
still up timer elapses
To solve this issue, a node which has already be evicted from
::qs_conn_bm can't set hold again and again invoked from idle timer.
Link: http://lkml.kernel.org/r/63ADC13FD55D6546B7DECE290D39E373F1F3F93B@H3CMLB12-EX.srv.huawei-3com.com
Signed-off-by: Yang Zhang <zhang.yangB@h3c.com>
Signed-off-by: Changwei Ge <ge.changwei@h3c.com>
Cc: Mark Fasheh <mfasheh@versity.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pavel Tatashin
Srividya Desireddy
Yang Shi
Andrew Morton
Miles Chen
Balasubramani Vivekanandan
Oscar Salvador
Byongho Lee
piaojun
Changwei Ge
Gang He
alex chen
Yang Zhang