| Message ID | E1SCEK7-0008CY-LJ@tytso-glaptop.cam.corp.google.com |
|---|---|
| State | Accepted, archived |
| Headers | show |
On Mon, Mar 26, 2012 at 11:07 AM, Theodore Ts'o <tytso@mit.edu> wrote: > > Ext4 commits for 3.3 merge window; mostly cleanups and bug fixes Ted, is there any chance that the dirty handling (or something else) has some serious latency bug? I'm compiling an allmodconfig kernel, and my mouse is (sometimes) very erratic with bad latency. And the biggest changes in the kernel I'm running on that machine now (HEAD=529b73fc0a97) and the previous boot of that machine (HEAD=e22057c85993) are the ext4 changes. (Sure, there are tons of other changes in between, but they tend to be ARM changes or drivers that are irrelevant). It may be that the bad interactive behavior started earlier, but I'd have expected to notice it - it really is pretty bad, and I've been doing allmodconfig builds pretty regularly. (Adding Jan and Fengguang to the cc too, in case it's that inode_sync_wait() change that *should* be a no-op, but maybe somebody overlooked something) NOTE! Latency normally is perfectly fine. Then *occasionally* something bad happens, and it's horrible for a while. I have no idea what triggers it. It's a regular "make -j16" of an x86-64 allmodconfig build, and that should *not* be nearly enough to make this machine stutter noticeably enough for the mouse pointer to not move smoothly. I have no real data for this, though - just a "mouse pointer occasionally freezing up". So I'm looking for a potential "hmm, that makes me wonder if..." kind of reaction. Linus -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Mar 28, 2012 at 4:36 PM, Linus Torvalds <torvalds@linux-foundation.org> wrote: > > I have no real data for this, though - just a "mouse pointer > occasionally freezing up". So I'm looking for a potential "hmm, that > makes me wonder if..." kind of reaction. Heh. It may have been that my wireless mouse was having some issue syncing properly with the receiver. I unplugged and replugged the receiver, and I seem unable to reproduce the problem. So this may have been just an actual hardware latency issue, rather than the kernel having bad latency. But if it made you go "hmm, I should check xyz", please do go ahead and check it anyway, just in case ;) Linus -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Wed, Mar 28, 2012 at 04:45:55PM -0700, Linus Torvalds wrote: > On Wed, Mar 28, 2012 at 4:36 PM, Linus Torvalds > <torvalds@linux-foundation.org> wrote: > > > > I have no real data for this, though - just a "mouse pointer > > occasionally freezing up". So I'm looking for a potential "hmm, that > > makes me wonder if..." kind of reaction. > > Heh. It may have been that my wireless mouse was having some issue > syncing properly with the receiver. I unplugged and replugged the > receiver, and I seem unable to reproduce the problem. So this may have > been just an actual hardware latency issue, rather than the kernel > having bad latency. Yeah my wireless mouse also becomes insensitive nowadays (perhaps running low on battery power?) > But if it made you go "hmm, I should check xyz", please do go ahead > and check it anyway, just in case ;) I just booted into the new kernel, did some compile testing and it looks good. Anyway the classical "mouse pointer freezing up" problems are often caused by waits in direct page reclaim and can be confirmed with large allocstall # the number of direct reclaims nr_vmscan_write # the number of pageout()s numbers in /proc/vmstat FYI I'm testing desktop responsiveness remotely (as you know it could be a painful test). The test scheme is to run make -j16 (or any other workloads) as well as a small script to start some X apps and cycle through the windows. By checking if there are second(s) long gaps in the below progress messages, I confirm that the desktop is running smoothly. time window title 157.68 A swapping-desktop-test 157.73 A *Unsaved Document 1 - gedit 157.76 A Restore Session - Iceweasel 157.80 A Sudoku 157.83 A LibreOffice 3.4 157.91 A System Settings 158.01 A urxvt 158.01 A xeyes 158.06 A bay:/home/wfg/swapping-desktop-test - ZSH 158.08 A Xpdf: /usr/share/doc/shared-mime-info/shared-mime-info-spec.pdf 158.10 A bay:/home/wfg/swapping-desktop-test - ZSH 158.17 A OpenOffice.org 158.20 A OpenOffice.org 158.21 A OpenOffice.org 158.22 A OpenOffice.org 158.24 A OpenOffice.org 158.25 A Dictionary 158.27 A bay:/home/wfg/swapping-desktop-test - ZSH 158.30 A Tetravex 158.34 A Chess 158.37 A System Monitor 158.41 A Mines 158.50 A Robots 158.62 A Iagno 158.68 A Four-in-a-row 158.75 A Five or More 158.79 A Klondike 158.82 A Klotski 158.90 A Mahjongg - Easy 158.93 A Tali 159.00 A Desktop Help 159.08 A Home Thanks, Fengguang -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Linus Torvalds <torvalds@linux-foundation.org> writes: > On Mon, Mar 26, 2012 at 11:07 AM, Theodore Ts'o <tytso@mit.edu> wrote: >> >> Ext4 commits for 3.3 merge window; mostly cleanups and bug fixes > > Ted, is there any chance that the dirty handling (or something else) > has some serious latency bug? There's at least a serious scalability bug. That new extent cache statistic count in the super block is showing up badly on IO profiles on larger systems. -Andi
The following changes since commit 62aa2b537c6f5957afd98e29f96897419ed5ebab: Linux 3.3-rc2 (2012-01-31 13:31:54 -0800) are available in the git repository at: git://git.kernel.org/pub/scm/linux/kernel/git/tytso/ext4.git ext4_for_linus for you to fetch changes up to 9d547c35799a4ddd235f1565cec2fff6c9263504: vfs: remove unused superblock helpers (2012-03-21 22:34:55 -0400) ---------------------------------------------------------------- Ext4 commits for 3.3 merge window; mostly cleanups and bug fixes The changes to export dirty_writeback_interval are from Artem's s_dirt cleanup patch series. The same is true of the change to remove the s_dirt helper functions which never got used by anyone in-tree. I've run these changes by Al Viro, and am carrying them so that Artem can more easily fix up the rest of the file systems during the next merge window. (Originally we had hopped to remove the use of s_dirt from ext4 during this merge window, but his patches had some bugs, so I ultimately ended dropping them from the ext4 tree.) ---------------------------------------------------------------- Allison Henderson (1): ext4: correct ext4_punch_hole return codes Artem Bityutskiy (5): ext4: do not mark superblock as dirty unnecessarily ext4: write superblock only once on unmount ext4: remove useless s_dirt assignment mm: export dirty_writeback_interval vfs: remove unused superblock helpers Bobi Jam (1): ext4: expand commit callback and Curt Wohlgemuth (3): ext4: mark possibly unused variable in ext4_mb_normalize_request() ext4: don't release page refs in ext4_end_bio() ext4: add comments to definition of ext4_io_end_t Dan Carpenter (2): ext4: remove an unneeded NULL check in __ext4_check_dir_entry() ext4: using PTR_ERR() on the wrong variable in ext4_ext_migrate() Darrick J. Wong (1): ext4: update s_free_{inodes,blocks}_count during online resize Eric Sandeen (4): jbd2: clear BH_Delay & BH_Unwritten in journal_unmap_buffer ext4: remove the journal=update mount option ext4: remove the resize mount option ext4: avoid deadlock on sync-mounted FS w/o journal Heiko Carstens (1): ext4: fix balloc.c printk-format-warning Jan Kara (8): jbd2: split updating of journal superblock and marking journal empty jbd2: protect all log tail updates with j_checkpoint_mutex jbd2: issue cache flush after checkpointing even with internal journal jbd2: fix BH_JWrite setting in checkpointing code jbd2: declare __jbd2_journal_temp_unlink_buffer() static jbd2: remove always true condition in __journal_try_to_free_buffer() jbd2: remove bh_state lock from checkpointing code jbd2: cleanup journal tail after transaction commit Jeff Moyer (3): ext4: fix race between unwritten extent conversion and truncate ext4: clean up the flags passed to __blockdev_direct_IO ext4: fix race between sync and completed io work Joe Perches (3): ext4: remove redundant "EXT4-fs: " from uses of ext4_msg ext4: add no_printk argument validation, fix fallout ext4: avoid output message interleaving in ext4_error_<foo>() Lukas Czerner (9): ext4: ignore EXT4_INODE_JOURNAL_DATA flag with delalloc ext4: fix resize when resizing within single group ext4: rewrite punch hole to use ext4_ext_remove_space() ext4: remove unused code from ext4_ext_map_blocks() ext4: give more helpful error message in ext4_ext_rm_leaf() ext4: fix start and len arguments handling in ext4_trim_fs() ext4: fix trimmed block count accunting ext4: always set then trimmed blocks count into len ext4: remove restrictive checks for EOFBLOCKS_FL Santosh Nayak (1): ext4: Fix endianness bug when reading the MMP block Seiji Aguchi (1): jbd2: add drop_transaction/update_superblock_end tracepoints Theodore Ts'o (18): ext4: fold ext4_claim_inode into ext4_new_inode ext4: fix race when setting bitmap_uptodate flag ext4: fix INCOMPAT feature codepoint reservation for INLINEDATA ext4: remove EXT4_MB_{BITMAP,BUDDY} macros ext4: remove deprecation warnings for minix_df and grpid ext4: remove Opt_ignore ext4: remove the I_VERSION mount flag and use the super_block flag instead ext4: simplify handling of the errors=* mount options ext4: unify handling of mount options which have been removed ext4: use a table-driven handler for mount options ext4: move ext4_show_options() after parse_options() ext4: make ext4_show_options() be table-driven ext4: add debugging /proc file showing file system options ext4: ignore mount options supported by ext2/3 (but have since been removed) ext4: try to deprecate noacl and noxattr_user mount options ext4: check for zero length extent ext4: remove trailing newlines from ext4_msg() and ext4_error() messages ext4: change some printk() calls to use ext4_msg() instead Yongqiang Yang (3): jbd2: allocate transaction from separate slab cache jbd2: rename functions which initialize slab caches jbd2: use KMEM_CACHE instead of kmem_cache_create() Zheng Liu (2): ext4: remove unneeded variable in ext4_xattr_check_block() ext4: format flag in dx_probe() Documentation/filesystems/ext4.txt | 8 - fs/ext4/balloc.c | 63 ++- fs/ext4/dir.c | 13 +- fs/ext4/ext4.h | 34 +- fs/ext4/ext4_extents.h | 4 +- fs/ext4/ext4_jbd2.h | 128 ++++- fs/ext4/extents.c | 330 +++++------- fs/ext4/fsync.c | 2 + fs/ext4/ialloc.c | 260 ++++----- fs/ext4/inode.c | 95 ++-- fs/ext4/mballoc.c | 342 +++++------- fs/ext4/mballoc.h | 20 +- fs/ext4/migrate.c | 2 +- fs/ext4/mmp.c | 4 +- fs/ext4/namei.c | 2 +- fs/ext4/page-io.c | 18 +- fs/ext4/resize.c | 37 +- fs/ext4/super.c | 1075 ++++++++++++++++-------------------- fs/ext4/xattr.c | 25 +- fs/jbd2/checkpoint.c | 140 +---- fs/jbd2/commit.c | 47 ++- fs/jbd2/journal.c | 361 ++++++++----- fs/jbd2/recovery.c | 5 +- fs/jbd2/revoke.c | 12 +- fs/jbd2/transaction.c | 48 ++- include/linux/fs.h | 13 - include/linux/jbd2.h | 12 +- include/linux/journal-head.h | 2 + include/trace/events/jbd2.h | 29 +- mm/page-writeback.c | 2 + 30 files changed, 1522 insertions(+), 1611 deletions(-) -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html