- 13 Jan, 2012 40 commits
-
-
Mel Gorman authored
Short summary: There are severe stalls when a USB stick using VFAT is used with THP enabled that are reduced by this series. If you are experiencing this problem, please test and report back and considering I have seen complaints from openSUSE and Fedora users on this as well as a few private mails, I'm guessing it's a widespread issue. This is a new type of USB-related stall because it is due to synchronous compaction writing where as in the past the big problem was dirty pages reaching the end of the LRU and being written by reclaim. Am cc'ing Andrew this time and this series would replace mm-do-not-stall-in-synchronous-compaction-for-thp-allocations.patch. I'm also cc'ing Dave Jones as he might have merged that patch to Fedora for wider testing and ideally it would be reverted and replaced by this series. That said, the later patches could really do with some review. If this series is not the answer then a new direction needs to be discussed because as it is, the stalls are unacceptable as the results in this leader show. For testers that try backporting this to 3.1, it won't work because there is a non-obvious dependency on not writing back pages in direct reclaim so you need those patches too. Changelog since V5 o Rebase to 3.2-rc5 o Tidy up the changelogs a bit Changelog since V4 o Added reviewed-bys, credited Andrea properly for sync-light o Allow dirty pages without mappings to be considered for migration o Bound the number of pages freed for compaction o Isolate PageReclaim pages on their own LRU list This is against 3.2-rc5 and follows on from discussions on "mm: Do not stall in synchronous compaction for THP allocations" and "[RFC PATCH 0/5] Reduce compaction-related stalls". Initially, the proposed patch eliminated stalls due to compaction which sometimes resulted in user-visible interactivity problems on browsers by simply never using sync compaction. The downside was that THP success allocation rates were lower because dirty pages were not being migrated as reported by Andrea. His approach at fixing this was nacked on the grounds that it reverted fixes from Rik merged that reduced the amount of pages reclaimed as it severely impacted his workloads performance. This series attempts to reconcile the requirements of maximising THP usage, without stalling in a user-visible fashion due to compaction or cheating by reclaiming an excessive number of pages. Patch 1 partially reverts commit 39deaf85 to allow migration to isolate dirty pages. This is because migration can move some dirty pages without blocking. Patch 2 notes that the /proc/sys/vm/compact_memory handler is not using synchronous compaction when it should be. This is unrelated to the reported stalls but is worth fixing. Patch 3 checks if we isolated a compound page during lumpy scan and account for it properly. For the most part, this affects tracing so it's unrelated to the stalls but worth fixing. Patch 4 notes that it is possible to abort reclaim early for compaction and return 0 to the page allocator potentially entering the "may oom" path. This has not been observed in practice but the rest of the series potentially makes it easier to happen. Patch 5 adds a sync parameter to the migratepage callback and gives the callback responsibility for migrating the page without blocking if sync==false. For example, fallback_migrate_page will not call writepage if sync==false. This increases the number of pages that can be handled by asynchronous compaction thereby reducing stalls. Patch 6 restores filter-awareness to isolate_lru_page for migration. In practice, it means that pages under writeback and pages without a ->migratepage callback will not be isolated for migration. Patch 7 avoids calling direct reclaim if compaction is deferred but makes sure that compaction is only deferred if sync compaction was used. Patch 8 introduces a sync-light migration mechanism that sync compaction uses. The objective is to allow some stalls but to not call ->writepage which can lead to significant user-visible stalls. Patch 9 notes that while we want to abort reclaim ASAP to allow compation to go ahead that we leave a very small window of opportunity for compaction to run. This patch allows more pages to be freed by reclaim but bounds the number to a reasonable level based on the high watermark on each zone. Patch 10 allows slabs to be shrunk even after compaction_ready() is true for one zone. This is to avoid a problem whereby a single small zone can abort reclaim even though no pages have been reclaimed and no suitably large zone is in a usable state. Patch 11 fixes a problem with the rate of page scanning. As reclaim is rarely stalling on pages under writeback it means that scan rates are very high. This is particularly true for direct reclaim which is not calling writepage. The vmstat figures implied that much of this was busy work with PageReclaim pages marked for immediate reclaim. This patch is a prototype that moves these pages to their own LRU list. This has been tested and other than 2 USB keys getting trashed, nothing horrible fell out. That said, I am a bit unhappy with the rescue logic in patch 11 but did not find a better way around it. It does significantly reduce scan rates and System CPU time indicating it is the right direction to take. What is of critical importance is that stalls due to compaction are massively reduced even though sync compaction was still allowed. Testing from people complaining about stalls copying to USBs with THP enabled are particularly welcome. The following tests all involve THP usage and USB keys in some way. Each test follows this type of pattern 1. Read from some fast fast storage, be it raw device or file. Each time the copy finishes, start again until the test ends 2. Write a large file to a filesystem on a USB stick. Each time the copy finishes, start again until the test ends 3. When memory is low, start an alloc process that creates a mapping the size of physical memory to stress THP allocation. This is the "real" part of the test and the part that is meant to trigger stalls when THP is enabled. Copying continues in the background. 4. Record the CPU usage and time to execute of the alloc process 5. Record the number of THP allocs and fallbacks as well as the number of THP pages in use a the end of the test just before alloc exited 6. Run the test 5 times to get an idea of variability 7. Between each run, sync is run and caches dropped and the test waits until nr_dirty is a small number to avoid interference or caching between iterations that would skew the figures. The individual tests were then writebackCPDeviceBasevfat Disable THP, read from a raw device (sda), vfat on USB stick writebackCPDeviceBaseext4 Disable THP, read from a raw device (sda), ext4 on USB stick writebackCPDevicevfat THP enabled, read from a raw device (sda), vfat on USB stick writebackCPDeviceext4 THP enabled, read from a raw device (sda), ext4 on USB stick writebackCPFilevfat THP enabled, read from a file on fast storage and USB, both vfat writebackCPFileext4 THP enabled, read from a file on fast storage and USB, both ext4 The kernels tested were 3.1 3.1 vanilla 3.2-rc5 freemore Patches 1-10 immediate Patches 1-11 andrea The 8 patches Andrea posted as a basis of comparison The results are very long unfortunately. I'll start with the case where we are not using THP at all writebackCPDeviceBasevfat 3.1.0-vanilla rc5-vanilla freemore-v6r1 isolate-v6r1 andrea-v2r1 System Time 1.28 ( 0.00%) 54.49 (-4143.46%) 48.63 (-3687.69%) 4.69 ( -265.11%) 51.88 (-3940.81%) +/- 0.06 ( 0.00%) 2.45 (-4305.55%) 4.75 (-8430.57%) 7.46 (-13282.76%) 4.76 (-8440.70%) User Time 0.09 ( 0.00%) 0.05 ( 40.91%) 0.06 ( 29.55%) 0.07 ( 15.91%) 0.06 ( 27.27%) +/- 0.02 ( 0.00%) 0.01 ( 45.39%) 0.02 ( 25.07%) 0.00 ( 77.06%) 0.01 ( 52.24%) Elapsed Time 110.27 ( 0.00%) 56.38 ( 48.87%) 49.95 ( 54.70%) 11.77 ( 89.33%) 53.43 ( 51.54%) +/- 7.33 ( 0.00%) 3.77 ( 48.61%) 4.94 ( 32.63%) 6.71 ( 8.50%) 4.76 ( 35.03%) THP Active 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) +/- 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) Fault Alloc 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) +/- 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) Fault Fallback 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) +/- 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) 0.00 ( 0.00%) The THP figures are obviously all 0 because THP was enabled. The main thing to watch is the elapsed times and how they compare to times when THP is enabled later. It's also important to note that elapsed time is improved by this series as System CPu time is much reduced. writebackCPDevicevfat 3.1.0-vanilla rc5-vanilla freemore-v6r1 isolate-v6r1 andrea-v2r1 System Time 1.22 ( 0.00%) 13.89 (-1040.72%) 46.40 (-3709.20%) 4.44 ( -264.37%) 47.37 (-3789.33%) +/- 0.06 ( 0.00%) 22.82 (-37635.56%) 3.84 (-6249.44%) 6.48 (-10618.92%) 6.60 (-10818.53%) User Time 0.06 ( 0.00%) 0.06 ( -6.90%) 0.05 ( 17.24%) 0.05 ( 13.79%) 0.04 ( 31.03%) +/- 0.01 ( 0.00%) 0.01 ( 33.33%) 0.01 ( 33.33%) 0.01 ( 39.14%) 0.01 ( 25.46%) Elapsed Time 10445.54 ( 0.00%) 2249.92 ( 78.46%) 70.06 ( 99.33%) 16.59 ( 99.84%) 472.43 ( 95.48%) +/- 643.98 ( 0.00%) 811.62 ( -26.03%) 10.02 ( 98.44%) 7.03 ( 98.91%) 59.99 ( 90.68%) THP Active 15.60 ( 0.00%) 35.20 ( 225.64%) 65.00 ( 416.67%) 70.80 ( 453.85%) 62.20 ( 398.72%) +/- 18.48 ( 0.00%) 51.29 ( 277.59%) 15.99 ( 86.52%) 37.91 ( 205.18%) 22.02 ( 119.18%) Fault Alloc 121.80 ( 0.00%) 76.60 ( 62.89%) 155.40 ( 127.59%) 181.20 ( 148.77%) 286.60 ( 235.30%) +/- 73.51 ( 0.00%) 61.11 ( 83.12%) 34.89 ( 47.46%) 31.88 ( 43.36%) 68.13 ( 92.68%) Fault Fallback 881.20 ( 0.00%) 926.60 ( -5.15%) 847.60 ( 3.81%) 822.00 ( 6.72%) 716.60 ( 18.68%) +/- 73.51 ( 0.00%) 61.26 ( 16.67%) 34.89 ( 52.54%) 31.65 ( 56.94%) 67.75 ( 7.84%) MMTests Statistics: duration User/Sys Time Running Test (seconds) 3540.88 1945.37 716.04 64.97 1937.03 Total Elapsed Time (seconds) 52417.33 11425.90 501.02 230.95 2520.28 The first thing to note is the "Elapsed Time" for the vanilla kernels of 2249 seconds versus 56 with THP disabled which might explain the reports of USB stalls with THP enabled. Applying the patches brings performance in line with THP-disabled performance while isolating pages for immediate reclaim from the LRU cuts down System CPU time. The "Fault Alloc" success rate figures are also improved. The vanilla kernel only managed to allocate 76.6 pages on average over the course of 5 iterations where as applying the series allocated 181.20 on average albeit it is well within variance. It's worth noting that applies the series at least descreases the amount of variance which implies an improvement. Andrea's series had a higher success rate for THP allocations but at a severe cost to elapsed time which is still better than vanilla but still much worse than disabling THP altogether. One can bring my series close to Andrea's by removing this check /* * If compaction is deferred for high-order allocations, it is because * sync compaction recently failed. In this is the case and the caller * has requested the system not be heavily disrupted, fail the * allocation now instead of entering direct reclaim */ if (deferred_compaction && (gfp_mask & __GFP_NO_KSWAPD)) goto nopage; I didn't include a patch that removed the above check because hurting overall performance to improve the THP figure is not what the average user wants. It's something to consider though if someone really wants to maximise THP usage no matter what it does to the workload initially. This is summary of vmstat figures from the same test. 3.1.0-vanilla rc5-vanilla freemore-v6r1 isolate-v6r1 andrea-v2r1 Page Ins 3257266139 1111844061 17263623 10901575 161423219 Page Outs 81054922 30364312 3626530 3657687 8753730 Swap Ins 3294 2851 6560 4964 4592 Swap Outs 390073 528094 620197 790912 698285 Direct pages scanned 1077581700 3024951463 1764930052 115140570 5901188831 Kswapd pages scanned 34826043 7112868 2131265 1686942 1893966 Kswapd pages reclaimed 28950067 4911036 1246044 966475 1497726 Direct pages reclaimed 805148398 280167837 3623473 2215044 40809360 Kswapd efficiency 83% 69% 58% 57% 79% Kswapd velocity 664.399 622.521 4253.852 7304.360 751.490 Direct efficiency 74% 9% 0% 1% 0% Direct velocity 20557.737 264745.137 3522673.849 498551.938 2341481.435 Percentage direct scans 96% 99% 99% 98% 99% Page writes by reclaim 722646 529174 620319 791018 699198 Page writes file 332573 1080 122 106 913 Page writes anon 390073 528094 620197 790912 698285 Page reclaim immediate 0 2552514720 1635858848 111281140 5478375032 Page rescued immediate 0 0 0 87848 0 Slabs scanned 23552 23552 9216 8192 9216 Direct inode steals 231 0 0 0 0 Kswapd inode steals 0 0 0 0 0 Kswapd skipped wait 28076 786 0 61 6 THP fault alloc 609 383 753 906 1433 THP collapse alloc 12 6 0 0 6 THP splits 536 211 456 593 1136 THP fault fallback 4406 4633 4263 4110 3583 THP collapse fail 120 127 0 0 4 Compaction stalls 1810 728 623 779 3200 Compaction success 196 53 60 80 123 Compaction failures 1614 675 563 699 3077 Compaction pages moved 193158 53545 243185 333457 226688 Compaction move failure 9952 9396 16424 23676 45070 The main things to look at are 1. Page In/out figures are much reduced by the series. 2. Direct page scanning is incredibly high (264745.137 pages scanned per second on the vanilla kernel) but isolating PageReclaim pages on their own list reduces the number of pages scanned significantly. 3. The fact that "Page rescued immediate" is a positive number implies that we sometimes race removing pages from the LRU_IMMEDIATE list that need to be put back on a normal LRU but it happens only for 0.07% of the pages marked for immediate reclaim. writebackCPDeviceext4 3.1.0-vanilla rc5-vanilla freemore-v6r1 isolate-v6r1 andrea-v2r1 System Time 1.51 ( 0.00%) 1.77 ( -17.66%) 1.46 ( 2.92%) 1.15 ( 23.77%) 1.89 ( -25.63%) +/- 0.27 ( 0.00%) 0.67 ( -148.52%) 0.33 ( -22.76%) 0.30 ( -11.15%) 0.19 ( 30.16%) User Time 0.03 ( 0.00%) 0.04 ( -37.50%) 0.05 ( -62.50%) 0.07 ( -112.50%) 0.04 ( -18.75%) +/- 0.01 ( 0.00%) 0.02 ( -146.64%) 0.02 ( -97.91%) 0.02 ( -75.59%) 0.02 ( -63.30%) Elapsed Time 124.93 ( 0.00%) 114.49 ( 8.36%) 96.77 ( 22.55%) 27.48 ( 78.00%) 205.70 ( -64.65%) +/- 20.20 ( 0.00%) 74.39 ( -268.34%) 59.88 ( -196.48%) 7.72 ( 61.79%) 25.03 ( -23.95%) THP Active 161.80 ( 0.00%) 83.60 ( 51.67%) 141.20 ( 87.27%) 84.60 ( 52.29%) 82.60 ( 51.05%) +/- 71.95 ( 0.00%) 43.80 ( 60.88%) 26.91 ( 37.40%) 59.02 ( 82.03%) 52.13 ( 72.45%) Fault Alloc 471.40 ( 0.00%) 228.60 ( 48.49%) 282.20 ( 59.86%) 225.20 ( 47.77%) 388.40 ( 82.39%) +/- 88.07 ( 0.00%) 87.42 ( 99.26%) 73.79 ( 83.78%) 109.62 ( 124.47%) 82.62 ( 93.81%) Fault Fallback 531.60 ( 0.00%) 774.60 ( -45.71%) 720.80 ( -35.59%) 777.80 ( -46.31%) 614.80 ( -15.65%) +/- 88.07 ( 0.00%) 87.26 ( 0.92%) 73.79 ( 16.22%) 109.62 ( -24.47%) 82.29 ( 6.56%) MMTests Statistics: duration User/Sys Time Running Test (seconds) 50.22 33.76 30.65 24.14 128.45 Total Elapsed Time (seconds) 1113.73 1132.19 1029.45 759.49 1707.26 Similar test but the USB stick is using ext4 instead of vfat. As ext4 does not use writepage for migration, the large stalls due to compaction when THP is enabled are not observed. Still, isolating PageReclaim pages on their own list helped completion time largely by reducing the number of pages scanned by direct reclaim although time spend in congestion_wait could also be a factor. Again, Andrea's series had far higher success rates for THP allocation at the cost of elapsed time. I didn't look too closely but a quick look at the vmstat figures tells me kswapd reclaimed 8 times more pages than the patch series and direct reclaim reclaimed roughly three times as many pages. It follows that if memory is aggressively reclaimed, there will be more available for THP. writebackCPFilevfat 3.1.0-vanilla rc5-vanilla freemore-v6r1 isolate-v6r1 andrea-v2r1 System Time 1.76 ( 0.00%) 29.10 (-1555.52%) 46.01 (-2517.18%) 4.79 ( -172.35%) 54.89 (-3022.53%) +/- 0.14 ( 0.00%) 25.61 (-18185.17%) 2.15 (-1434.83%) 6.60 (-4610.03%) 9.75 (-6863.76%) User Time 0.05 ( 0.00%) 0.07 ( -45.83%) 0.05 ( -4.17%) 0.06 ( -29.17%) 0.06 ( -16.67%) +/- 0.02 ( 0.00%) 0.02 ( 20.11%) 0.02 ( -3.14%) 0.01 ( 31.58%) 0.01 ( 47.41%) Elapsed Time 22520.79 ( 0.00%) 1082.85 ( 95.19%) 73.30 ( 99.67%) 32.43 ( 99.86%) 291.84 ( 98.70%) +/- 7277.23 ( 0.00%) 706.29 ( 90.29%) 19.05 ( 99.74%) 17.05 ( 99.77%) 125.55 ( 98.27%) THP Active 83.80 ( 0.00%) 12.80 ( 15.27%) 15.60 ( 18.62%) 13.00 ( 15.51%) 0.80 ( 0.95%) +/- 66.81 ( 0.00%) 20.19 ( 30.22%) 5.92 ( 8.86%) 15.06 ( 22.54%) 1.17 ( 1.75%) Fault Alloc 171.00 ( 0.00%) 67.80 ( 39.65%) 97.40 ( 56.96%) 125.60 ( 73.45%) 133.00 ( 77.78%) +/- 82.91 ( 0.00%) 30.69 ( 37.02%) 53.91 ( 65.02%) 55.05 ( 66.40%) 21.19 ( 25.56%) Fault Fallback 832.00 ( 0.00%) 935.20 ( -12.40%) 906.00 ( -8.89%) 877.40 ( -5.46%) 870.20 ( -4.59%) +/- 82.91 ( 0.00%) 30.69 ( 62.98%) 54.01 ( 34.86%) 55.05 ( 33.60%) 20.91 ( 74.78%) MMTests Statistics: duration User/Sys Time Running Test (seconds) 7229.81 928.42 704.52 80.68 1330.76 Total Elapsed Time (seconds) 112849.04 5618.69 571.11 360.54 1664.28 In this case, the test is reading/writing only from filesystems but as it's vfat, it's slow due to calling writepage during compaction. Little to observe really - the time to complete the test goes way down with the series applied and THP allocation success rates go up in comparison to 3.2-rc5. The success rates are lower than 3.1.0 but the elapsed time for that kernel is abysmal so it is not really a sensible comparison. As before, Andrea's series allocates more THPs at the cost of overall performance. writebackCPFileext4 3.1.0-vanilla rc5-vanilla freemore-v6r1 isolate-v6r1 andrea-v2r1 System Time 1.51 ( 0.00%) 1.77 ( -17.66%) 1.46 ( 2.92%) 1.15 ( 23.77%) 1.89 ( -25.63%) +/- 0.27 ( 0.00%) 0.67 ( -148.52%) 0.33 ( -22.76%) 0.30 ( -11.15%) 0.19 ( 30.16%) User Time 0.03 ( 0.00%) 0.04 ( -37.50%) 0.05 ( -62.50%) 0.07 ( -112.50%) 0.04 ( -18.75%) +/- 0.01 ( 0.00%) 0.02 ( -146.64%) 0.02 ( -97.91%) 0.02 ( -75.59%) 0.02 ( -63.30%) Elapsed Time 124.93 ( 0.00%) 114.49 ( 8.36%) 96.77 ( 22.55%) 27.48 ( 78.00%) 205.70 ( -64.65%) +/- 20.20 ( 0.00%) 74.39 ( -268.34%) 59.88 ( -196.48%) 7.72 ( 61.79%) 25.03 ( -23.95%) THP Active 161.80 ( 0.00%) 83.60 ( 51.67%) 141.20 ( 87.27%) 84.60 ( 52.29%) 82.60 ( 51.05%) +/- 71.95 ( 0.00%) 43.80 ( 60.88%) 26.91 ( 37.40%) 59.02 ( 82.03%) 52.13 ( 72.45%) Fault Alloc 471.40 ( 0.00%) 228.60 ( 48.49%) 282.20 ( 59.86%) 225.20 ( 47.77%) 388.40 ( 82.39%) +/- 88.07 ( 0.00%) 87.42 ( 99.26%) 73.79 ( 83.78%) 109.62 ( 124.47%) 82.62 ( 93.81%) Fault Fallback 531.60 ( 0.00%) 774.60 ( -45.71%) 720.80 ( -35.59%) 777.80 ( -46.31%) 614.80 ( -15.65%) +/- 88.07 ( 0.00%) 87.26 ( 0.92%) 73.79 ( 16.22%) 109.62 ( -24.47%) 82.29 ( 6.56%) MMTests Statistics: duration User/Sys Time Running Test (seconds) 50.22 33.76 30.65 24.14 128.45 Total Elapsed Time (seconds) 1113.73 1132.19 1029.45 759.49 1707.26 Same type of story - elapsed times go down. In this case, allocation success rates are roughtly the same. As before, Andrea's has higher success rates but takes a lot longer. Overall the series does reduce latencies and while the tests are inherency racy as alloc competes with the cp processes, the variability was included. The THP allocation rates are not as high as they could be but that is because we would have to be more aggressive about reclaim and compaction impacting overall performance. This patch: Commit 39deaf85 ("mm: compaction: make isolate_lru_page() filter-aware") noted that compaction does not migrate dirty or writeback pages and that is was meaningless to pick the page and re-add it to the LRU list. What was missed during review is that asynchronous migration moves dirty pages if their ->migratepage callback is migrate_page() because these can be moved without blocking. This potentially impacted hugepage allocation success rates by a factor depending on how many dirty pages are in the system. This patch partially reverts 39deaf85 to allow migration to isolate dirty pages again. This increases how much compaction disrupts the LRU but that is addressed later in the series. Signed-off-by:
Mel Gorman <mgorman@suse.de> Reviewed-by:
Andrea Arcangeli <aarcange@redhat.com> Reviewed-by:
Rik van Riel <riel@redhat.com> Reviewed-by:
Minchan Kim <minchan.kim@gmail.com> Cc: Dave Jones <davej@redhat.com> Cc: Jan Kara <jack@suse.cz> Cc: Andy Isaacson <adi@hexapodia.org> Cc: Nai Xia <nai.xia@gmail.com> Cc: Johannes Weiner <jweiner@redhat.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Tao Ma authored
In trace_mm_vmscan_lru_isolate(), we don't output 'file' information to the trace event and it is a bit inconvenient for the user to get the real information(like pasted below). mm_vmscan_lru_isolate: isolate_mode=2 order=0 nr_requested=32 nr_scanned=32 nr_taken=32 contig_taken=0 contig_dirty=0 contig_failed=0 'active' can be obtained by analyzing mode(Thanks go to Minchan and Mel), So this patch adds 'file' to the trace event and it now looks like: mm_vmscan_lru_isolate: isolate_mode=2 order=0 nr_requested=32 nr_scanned=32 nr_taken=32 contig_taken=0 contig_dirty=0 contig_failed=0 file=0 Signed-off-by:
Tao Ma <boyu.mt@taobao.com> Acked-by:
KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com> Reviewed-by:
KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Mel Gorman <mel@csn.ul.ie> Reviewed-by:
Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@redhat.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Signed-off-by:
-
Shaohua Li authored
Put the tail subpages of an isolated hugepage under splitting in the lru reclaim head as they supposedly should be isolated too next. Queues the subpages in physical order in the lru for non isolated hugepages under splitting. That might provide some theoretical cache benefit to the buddy allocator later. Signed-off-by:
Shaohua Li <shaohua.li@intel.com> Signed-off-by:
-
Shaohua Li authored
We have tlb_remove_tlb_entry to indicate a pte tlb flush entry should be flushed, but not a corresponding API for pmd entry. This isn't a problem so far because THP is only for x86 currently and tlb_flush() under x86 will flush entire TLB. But this is confusion and could be missed if thp is ported to other arch. Also convert tlb->need_flush = 1 to a VM_BUG_ON(!tlb->need_flush) in __tlb_remove_page() as suggested by Andrea Arcangeli. The __tlb_remove_page() function is supposed to be called after tlb_remove_xxx_tlb_entry() and we can catch any misuse. Signed-off-by:
-
Shaohua Li authored
change_protection() will do TLB flush later, don't need duplicate tlb flush. Signed-off-by:
-
Shaohua Li authored
Improve the error code path. Delete unnecessary sysfs file for example. Also remove the #ifdef xxx to make code better. Signed-off-by:
-
Bob Liu authored
No need for two CONFIG_MEMORY_HOTPLUG blocks. Signed-off-by:
Bob Liu <lliubbo@gmail.com> Acked-by:
Michal Hocko <mhocko@suse.cz> Cc: Johannes Weiner <hannes@cmpxchg.org> Acked-by:
-
Bob Liu authored
If there is a zone below ZONE_NORMAL has present_pages, we can set node state to N_NORMAL_MEMORY, no need to loop to end. Signed-off-by:
-
Bob Liu authored
We already have for_each_node(node) define in nodemask.h, better to use it. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Now, at LRU handling, memory cgroup needs to do complicated works to see valid pc->mem_cgroup, which may be overwritten. This patch is for relaxing the protocol. This patch guarantees - when pc->mem_cgroup is overwritten, page must not be on LRU. By this, LRU routine can believe pc->mem_cgroup and don't need to check bits on pc->flags. This new rule may adds small overheads to swapin. But in most case, lru handling gets faster. After this patch, PCG_ACCT_LRU bit is obsolete and removed. [akpm@linux-foundation.org: remove unneeded VM_BUG_ON(), restore hannes's christmas tree] [akpm@linux-foundation.org: clean up code comment] [hughd@google.com: fix NULL mem_cgroup_try_charge] Signed-off-by:
Johannes Weiner <hannes@cmpxchg.org> Cc: Ying Han <yinghan@google.com> Signed-off-by:
Hugh Dickins <hughd@google.com> Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This is a preparation before removing a flag PCG_ACCT_LRU in page_cgroup and reducing atomic ops/complexity in memcg LRU handling. In some cases, pages are added to lru before charge to memcg and pages are not classfied to memory cgroup at lru addtion. Now, the lru where the page should be added is determined a bit in page_cgroup->flags and pc->mem_cgroup. I'd like to remove the check of flag. To handle the case pc->mem_cgroup may contain stale pointers if pages are added to LRU before classification. This patch resets pc->mem_cgroup to root_mem_cgroup before lru additions. [akpm@linux-foundation.org: fix CONFIG_CGROUP_MEM_CONT=n build] [hughd@google.com: fix CONFIG_CGROUP_MEM_RES_CTLR=y CONFIG_CGROUP_MEM_RES_CTLR_SWAP=n build] [akpm@linux-foundation.org: ksm.c needs memcontrol.h, per Michal] [hughd@google.com: stop oops in mem_cgroup_reset_owner()] [hughd@google.com: fix page migration to reset_owner] Signed-off-by:
-
KAMEZAWA Hiroyuki authored
This patch simplifies LRU handling of racy case (memcg+SwapCache). At charging, SwapCache tend to be on LRU already. So, before overwriting pc->mem_cgroup, the page must be removed from LRU and added to LRU later. This patch does spin_lock(zone->lru_lock); if (PageLRU(page)) remove from LRU overwrite pc->mem_cgroup if (PageLRU(page)) add to new LRU. spin_unlock(zone->lru_lock); And guarantee all pages are not on LRU at modifying pc->mem_cgroup. This patch also unfies lru handling of replace_page_cache() and swapin. Signed-off-by: -
KAMEZAWA Hiroyuki authored
This patch is a clean up. No functional/logical changes. Because of commit ef6a3c63 ("mm: add replace_page_cache_page() function") , FUSE uses replace_page_cache() instead of add_to_page_cache(). Then, mem_cgroup_cache_charge() is not called against FUSE's pages from splice. So now, mem_cgroup_cache_charge() gets pages that are not on the LRU with the exception of PageSwapCache pages. For checking, WARN_ON_ONCE(PageLRU(page)) is added. Signed-off-by:
-
David Rientjes authored
The oom killer relies on logic that identifies threads that have already been oom killed when scanning the tasklist and, if found, deferring until such threads have exited. This is done by checking for any candidate threads that have the TIF_MEMDIE bit set. For memcg ooms, candidate threads are first found by calling task_in_mem_cgroup() since the oom killer should not defer if there's an oom killed thread in another memcg. Unfortunately, task_in_mem_cgroup() excludes threads if they have detached their mm in the process of exiting so TIF_MEMDIE is never detected for such conditions. This is different for global, mempolicy, and cpuset oom conditions where a detached mm is only excluded after checking for TIF_MEMDIE and deferring, if necessary, in select_bad_process(). The fix is to return true if a task has a detached mm but is still in the memcg or its hierarchy that is currently oom. This will allow the oom killer to appropriately defer rather than kill unnecessarily or, in the worst case, panic the machine if nothing else is available to kill. Signed-off-by:
David Rientjes <rientjes@google.com> Acked-by:
-
Michal Hocko authored
If we are not able to allocate tree nodes for all NUMA nodes then we should release those that were allocated. Signed-off-by:
-
Bob Liu authored
There are multiple places which need to get the swap_cgroup address, so add a helper function: static struct swap_cgroup *swap_cgroup_getsc(swp_entry_t ent, struct swap_cgroup_ctrl **ctrl); to simplify the code. Signed-off-by: -
Johannes Weiner authored
mem_cgroup_uncharge_page() is only called on either freshly allocated pages without page->mapping or on rmapped PageAnon() pages. There is no need to check for a page->mapping that is not an anon_vma. Signed-off-by:
Johannes Weiner <jweiner@redhat.com> Acked-by:
-
Johannes Weiner authored
All callsites pass in freshly allocated pages and a valid mm. As a result, all checks pertaining to the page's mapcount, page->mapping or the fallback to init_mm are unneeded. Signed-off-by:
-
Johannes Weiner authored
lookup_page_cgroup() is usually used only against pages that are used in userspace. The exception is the CONFIG_DEBUG_VM-only memcg check from the page allocator: it can run on pages without page_cgroup descriptors allocated when the pages are fed into the page allocator for the first time during boot or memory hotplug. Include the array check only when CONFIG_DEBUG_VM is set and save the unnecessary check in production kernels. Signed-off-by:
-
Johannes Weiner authored
Pages have their corresponding page_cgroup descriptors set up before they are used in userspace, and thus managed by a memory cgroup. The only time where lookup_page_cgroup() can return NULL is in the CONFIG_DEBUG_VM-only page sanity checking code that executes while feeding pages into the page allocator for the first time. Remove the NULL checks against lookup_page_cgroup() results from all callsites where we know that corresponding page_cgroup descriptors must be allocated, and add a comment to the callsite that actually does have to check the return value. [hughd@google.com: stop oops in mem_cgroup_update_page_stat()] Signed-off-by:
-
Johannes Weiner authored
The fault accounting functions have a single, memcg-internal user, so they don't need to be global. In fact, their one-line bodies can be directly folded into the caller. And since faults happen one at a time, use this_cpu_inc() directly instead of this_cpu_add(foo, 1). Signed-off-by:
Balbir Singh <bsingharora@gmail.com> Cc: David Rientjes <rientjes@google.com> Cc: Hugh Dickins <hughd@google.com> Signed-off-by:
-
Johannes Weiner authored
Signed-off-by:
-
Johannes Weiner authored
The memcg argument of oom_kill_task() hasn't been used since 341aea2b 'oom-kill: remove boost_dying_task_prio()'. Kill it. Signed-off-by:
-
Ying Han authored
The two memcg stats pgpgin/pgpgout have different meaning than the ones in vmstat, which indicates that we picked a bad naming for them. It might be late to change the stat name, but better documentation is always helpful. Signed-off-by:
Ying Han <yinghan@google.com> Acked-by:
-
Zhu Yanhai authored
It should be memsw.max_usage_in_bytes. This typo has been there for a really long time. Signed-off-by:
Zhu Yanhai <gaoyang.zyh@taobao.com> Acked-by:
-
Johannes Weiner authored
Only the ratelimit checks themselves have to run with preemption disabled, the resulting actions - checking for usage thresholds, updating the soft limit tree - can and should run with preemption enabled. Signed-off-by:
Yong Zhang <yong.zhang0@gmail.com> Tested-by:
Luis Henriques <henrix@camandro.org> Tested-by:
-
KAMEZAWA Hiroyuki authored
In split_huge_page(), mem_cgroup_split_huge_fixup() is called to handle page_cgroup modifcations. It takes move_lock_page_cgroup() and modifies page_cgroup and LRU accounting jobs and called HPAGE_PMD_SIZE - 1 times. But thinking again, - compound_lock() is held at move_accout...then, it's not necessary to take move_lock_page_cgroup(). - LRU is locked and all tail pages will go into the same LRU as head is now on. - page_cgroup is contiguous in huge page range. This patch fixes mem_cgroup_split_huge_fixup() as to be called once per hugepage and reduce costs for spliting. [akpm@linux-foundation.org: fix typo, per Michal] Signed-off-by: -
Johannes Weiner authored
To find the page corresponding to a certain page_cgroup, the pc->flags encoded the node or section ID with the base array to compare the pc pointer to. Now that the per-memory cgroup LRU lists link page descriptors directly, there is no longer any code that knows the struct page_cgroup of a PFN but not the struct page. [hughd@google.com: remove unused node/section info from pc->flags fix] Signed-off-by:
Kirill A. Shutemov <kirill@shutemov.name> Cc: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com> Cc: Michal Hocko <mhocko@suse.cz> Cc: "Kirill A. Shutemov" <kirill@shutemov.name> Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> Cc: Balbir Singh <bsingharora@gmail.com> Cc: Ying Han <yinghan@google.com> Cc: Greg Thelen <gthelen@google.com> Cc: Michel Lespinasse <walken@google.com> Cc: Rik van Riel <riel@redhat.com> Cc: Minchan Kim <minchan.kim@gmail.com> Cc: Christoph Hellwig <hch@infradead.org> Signed-off-by:
-
Johannes Weiner authored
Now that all code that operated on global per-zone LRU lists is converted to operate on per-memory cgroup LRU lists instead, there is no reason to keep the double-LRU scheme around any longer. The pc->lru member is removed and page->lru is linked directly to the per-memory cgroup LRU lists, which removes two pointers from a descriptor that exists for every page frame in the system. Signed-off-by:
-
Johannes Weiner authored
Having a unified structure with a LRU list set for both global zones and per-memcg zones allows to keep that code simple which deals with LRU lists and does not care about the container itself. Once the per-memcg LRU lists directly link struct pages, the isolation function and all other list manipulations are shared between the memcg case and the global LRU case. Signed-off-by:
-
Johannes Weiner authored
The global per-zone LRU lists are about to go away on memcg-enabled kernels, global reclaim must be able to find its pages on the per-memcg LRU lists. Since the LRU pages of a zone are distributed over all existing memory cgroups, a scan target for a zone is complete when all memory cgroups are scanned for their proportional share of a zone's memory. The forced scanning of small scan targets from kswapd is limited to zones marked unreclaimable, otherwise kswapd can quickly overreclaim by force-scanning the LRU lists of multiple memory cgroups. Signed-off-by:
-
Johannes Weiner authored
root_mem_cgroup, lacking a configurable limit, was never subject to limit reclaim, so the pages charged to it could be kept off its LRU lists. They would be found on the global per-zone LRU lists upon physical memory pressure and it made sense to avoid uselessly linking them to both lists. The global per-zone LRU lists are about to go away on memcg-enabled kernels, with all pages being exclusively linked to their respective per-memcg LRU lists. As a result, pages of the root_mem_cgroup must also be linked to its LRU lists again. This is purely about the LRU list, root_mem_cgroup is still not charged. The overhead is temporary until the double-LRU scheme is going away completely. Signed-off-by:
-
Johannes Weiner authored
Memory cgroup limit reclaim and traditional global pressure reclaim will soon share the same code to reclaim from a hierarchical tree of memory cgroups. In preparation of this, move the two right next to each other in shrink_zone(). The mem_cgroup_hierarchical_reclaim() polymath is split into a soft limit reclaim function, which still does hierarchy walking on its own, and a limit (shrinking) reclaim function, which relies on generic reclaim code to walk the hierarchy. Signed-off-by:
-
Johannes Weiner authored
Memory cgroup limit reclaim currently picks one memory cgroup out of the target hierarchy, remembers it as the last scanned child, and reclaims all zones in it with decreasing priority levels. The new hierarchy reclaim code will pick memory cgroups from the same hierarchy concurrently from different zones and priority levels, it becomes necessary that hierarchy roots not only remember the last scanned child, but do so for each zone and priority level. Until now, we reclaimed memcgs like this: mem = mem_cgroup_iter(root) for each priority level: for each zone in zonelist: reclaim(mem, zone) But subsequent patches will move the memcg iteration inside the loop over the zones: for each priority level: for each zone in zonelist: mem = mem_cgroup_iter(root) reclaim(mem, zone) And to keep with the original scan order - memcg -> priority -> zone - the last scanned memcg has to be remembered per zone and per priority level. Furthermore, global reclaim will be switched to the hierarchy walk as well. Different from limit reclaim, which can just recheck the limit after some reclaim progress, its target is to scan all memcgs for the desired zone pages, proportional to the memcg size, and so reliably detecting a full hierarchy round-trip will become crucial. Currently, the code relies on one reclaimer encountering the same memcg twice, but that is error-prone with concurrent reclaimers. Instead, use a generation counter that is increased every time the child with the highest ID has been visited, so that reclaimers can stop when the generation changes. Signed-off-by: -
Johannes Weiner authored
Memory cgroup hierarchies are currently handled completely outside of the traditional reclaim code, which is invoked with a single memory cgroup as an argument for the whole call stack. Subsequent patches will switch this code to do hierarchical reclaim, so there needs to be a distinction between a) the memory cgroup that is triggering reclaim due to hitting its limit and b) the memory cgroup that is being scanned as a child of a). This patch introduces a struct mem_cgroup_zone that contains the combination of the memory cgroup and the zone being scanned, which is then passed down the stack instead of the zone argument. [akpm@linux-foundation.org: coding-style fixes] Signed-off-by:
-
Johannes Weiner authored
The traditional zone reclaim code is scanning the per-zone LRU lists during direct reclaim and kswapd, and the per-zone per-memory cgroup LRU lists when reclaiming on behalf of a memory cgroup limit. Subsequent patches will convert the traditional reclaim code to reclaim exclusively from the per-memory cgroup LRU lists. As a result, using the predicate for which LRU list is scanned will no longer be appropriate to tell global reclaim from limit reclaim. This patch adds a global_reclaim() predicate to tell direct/kswapd reclaim from memory cgroup limit reclaim and substitutes it in all places where currently scanning_global_lru() is used for that. Signed-off-by:
-
Johannes Weiner authored
The memcg naturalization series: Memory control groups are currently bolted onto the side of traditional memory management in places where better integration would be preferrable. To reclaim memory, for example, memory control groups maintain their own LRU list and reclaim strategy aside from the global per-zone LRU list reclaim. But an extra list head for each existing page frame is expensive and maintaining it requires additional code. This patchset disables the global per-zone LRU lists on memory cgroup configurations and converts all its users to operate on the per-memory cgroup lists instead. As LRU pages are then exclusively on one list, this saves two list pointers for each page frame in the system: page_cgroup array size with 4G physical memory vanilla: allocated 31457280 bytes of page_cgroup patched: allocated 15728640 bytes of page_cgroup At the same time, system performance for various workloads is unaffected: 100G sparse file cat, 4G physical memory, 10 runs, to test for code bloat in the traditional LRU handling and kswapd & direct reclaim paths, without/with the memory controller configured in vanilla: 71.603(0.207) seconds patched: 71.640(0.156) seconds vanilla: 79.558(0.288) seconds patched: 77.233(0.147) seconds 100G sparse file cat in 1G memory cgroup, 10 runs, to test for code bloat in the traditional memory cgroup LRU handling and reclaim path vanilla: 96.844(0.281) seconds patched: 94.454(0.311) seconds 4 unlimited memcgs running kbuild -j32 each, 4G physical memory, 500M swap on SSD, 10 runs, to test for regressions in kswapd & direct reclaim using per-memcg LRU lists with multiple memcgs and multiple allocators within each memcg vanilla: 717.722(1.440) seconds [ 69720.100(11600.835) majfaults ] patched: 714.106(2.313) seconds [ 71109.300(14886.186) majfaults ] 16 unlimited memcgs running kbuild, 1900M hierarchical limit, 500M swap on SSD, 10 runs, to test for regressions in hierarchical memcg setups vanilla: 2742.058(1.992) seconds [ 26479.600(1736.737) majfaults ] patched: 2743.267(1.214) seconds [ 27240.700(1076.063) majfaults ] This patch: There are currently two different implementations of iterating over a memory cgroup hierarchy tree. Consolidate them into one worker function and base the convenience looping-macros on top of it. Signed-off-by:
-
KAMEZAWA Hiroyuki authored
Commit ef6a3c63 ("mm: add replace_page_cache_page() function") added a function replace_page_cache_page(). This function replaces a page in the radix-tree with a new page. WHen doing this, memory cgroup needs to fix up the accounting information. memcg need to check PCG_USED bit etc. In some(many?) cases, 'newpage' is on LRU before calling replace_page_cache(). So, memcg's LRU accounting information should be fixed, too. This patch adds mem_cgroup_replace_page_cache() and removes the old hooks. In that function, old pages will be unaccounted without touching res_counter and new page will be accounted to the memcg (of old page). WHen overwriting pc->mem_cgroup of newpage, take zone->lru_lock and avoid races with LRU handling. Background: replace_page_cache_page() is called by FUSE code in its splice() handling. Here, 'newpage' is replacing oldpage but this newpage is not a newly allocated page and may be on LRU. LRU mis-accounting will be critical for memory cgroup because rmdir() checks the whole LRU is empty and there is no account leak. If a page is on the other LRU than it should be, rmdir() will fail. This bug was added in March 2011, but no bug report yet. I guess there are not many people who use memcg and FUSE at the same time with upstream kernels. The result of this bug is that admin cannot destroy a memcg because of account leak. So, no panic, no deadlock. And, even if an active cgroup exist, umount can succseed. So no problem at shutdown. Signed-off-by:
-
Jason Baron authored
The current epoll code can be tickled to run basically indefinitely in both loop detection path check (on ep_insert()), and in the wakeup paths. The programs that tickle this behavior set up deeply linked networks of epoll file descriptors that cause the epoll algorithms to traverse them indefinitely. A couple of these sample programs have been previously posted in this thread: https://lkml.org/lkml/2011/2/25/297. To fix the loop detection path check algorithms, I simply keep track of the epoll nodes that have been already visited. Thus, the loop detection becomes proportional to the number of epoll file descriptor and links. This dramatically decreases the run-time of the loop check algorithm. In one diabolical case I tried it reduced the run-time from 15 mintues (all in kernel time) to .3 seconds. Fixing the wakeup paths could be done at wakeup time in a similar manner by keeping track of nodes that have already been visited, but the complexity is harder, since there can be multiple wakeups on different cpus...Thus, I've opted to limit the number of possible wakeup paths when the paths are created. This is accomplished, by noting that the end file descriptor points that are found during the loop detection pass (from the newly added link), are actually the sources for wakeup events. I keep a list of these file descriptors and limit the number and length of these paths that emanate from these 'source file descriptors'. In the current implemetation I allow 1000 paths of length 1, 500 of length 2, 100 of length 3, 50 of length 4 and 10 of length 5. Note that it is sufficient to check the 'source file descriptors' reachable from the newly added link, since no other 'source file descriptors' will have newly added links. This allows us to check only the wakeup paths that may have gotten too long, and not re-check all possible wakeup paths on the system. In terms of the path limit selection, I think its first worth noting that the most common case for epoll, is probably the model where you have 1 epoll file descriptor that is monitoring n number of 'source file descriptors'. In this case, each 'source file descriptor' has a 1 path of length 1. Thus, I believe that the limits I'm proposing are quite reasonable and in fact may be too generous. Thus, I'm hoping that the proposed limits will not prevent any workloads that currently work to fail. In terms of locking, I have extended the use of the 'epmutex' to all epoll_ctl add and remove operations. Currently its only used in a subset of the add paths. I need to hold the epmutex, so that we can correctly traverse a coherent graph, to check the number of paths. I believe that this additional locking is probably ok, since its in the setup/teardown paths, and doesn't affect the running paths, but it certainly is going to add some extra overhead. Also, worth noting is that the epmuex was recently added to the ep_ctl add operations in the initial path loop detection code using the argument that it was not on a critical path. Another thing to note here, is the length of epoll chains that is allowed. Currently, eventpoll.c defines: /* Maximum number of nesting allowed inside epoll sets */ #define EP_MAX_NESTS 4 This basically means that I am limited to a graph depth of 5 (EP_MAX_NESTS + 1). However, this limit is currently only enforced during the loop check detection code, and only when the epoll file descriptors are added in a certain order. Thus, this limit is currently easily bypassed. The newly added check for wakeup paths, stricly limits the wakeup paths to a length of 5, regardless of the order in which ep's are linked together. Thus, a side-effect of the new code is a more consistent enforcement of the graph depth. Thus far, I've tested this, using the sample programs previously mentioned, which now either return quickly or return -EINVAL. I've also testing using the piptest.c epoll tester, which showed no difference in performance. I've also created a number of different epoll networks and tested that they behave as expectded. I believe this solves the original diabolical test cases, while still preserving the sane epoll nesting. Signed-off-by:
Jason Baron <jbaron@redhat.com> Cc: Nelson Elhage <nelhage@ksplice.com> Cc: Davide Libenzi <davidel@xmailserver.org> Signed-off-by:
-
Sasha Levin authored
When a user with the CAP_SYS_RESOURCE cap tries to F_SETPIPE_SZ a pipe with size bigger than kmalloc() can alloc it spits out an ugly warning: ------------[ cut here ]------------ WARNING: at mm/page_alloc.c:2095 __alloc_pages_nodemask+0x5d3/0x7a0() Pid: 733, comm: a.out Not tainted 3.2.0-rc1+ #4 Call Trace: warn_slowpath_common+0x75/0xb0 warn_slowpath_null+0x15/0x20 __alloc_pages_nodemask+0x5d3/0x7a0 __get_free_pages+0x12/0x50 __kmalloc+0x12b/0x150 pipe_set_size+0x75/0x120 pipe_fcntl+0xf8/0x140 do_fcntl+0x2d4/0x410 sys_fcntl+0x66/0xa0 system_call_fastpath+0x16/0x1b ---[ end trace 432f702e6db7b5ee ]--- Instead, make kcalloc() handle the overflow case and fail quietly. [akpm@linux-foundation.org: switch to sizeof(*bufs) for 80-column niceness] Signed-off-by:Sasha Levin <levinsasha928@gmail.com> Cc: Alexander Viro <viro@zeniv.linux.org.uk> Acked-by:
Pekka Enberg <penberg@kernel.org> Signed-off-by:
-
