- 28 Jul, 2016 40 commits
-
-
Mel Gorman authored
As reclaim is now per-node based, convert zone_reclaim to be node_reclaim. It is possible that a node will be reclaimed multiple times if it has multiple zones but this is unavoidable without caching all nodes traversed so far. The documentation and interface to userspace is the same from a configuration perspective and will will be similar in behaviour unless the node-local allocation requests were also limited to lower zones. Link: http://lkml.kernel.org/r/1467970510-21195-24-git-send-email-mgorman@techsingularity.netSigned-off-by:
Mel Gorman <mgorman@techsingularity.net> Acked-by:
Vlastimil Babka <vbabka@suse.cz> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Acked-by:
Johannes Weiner <hannes@cmpxchg.org> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Michal Hocko <mhocko@kernel.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Rik van Riel <riel@surriel.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Mel Gorman authored
The ac_classzone_idx is used as the basis for waking kswapd and that is based on the preferred zoneref. If the preferred zoneref's first zone is lower than what is available on other nodes, it's possible that kswapd is woken on a zone with only higher, but still eligible, zones. As classzone_idx is strictly adhered to now, it causes a problem because eligible pages are skipped. For example, node 0 has only DMA32 and node 1 has only NORMAL. An allocating context running on node 0 may wake kswapd on node 1 telling it to skip all NORMAL pages. Link: http://lkml.kernel.org/r/1467970510-21195-23-git-send-email-mgorman@techsingularity.netSigned-off-by:
Hillf Danton <hillf.zj@alibaba-inc.com> Acked-by:
-
Mel Gorman authored
kswapd is woken when zones are below the low watermark but the wakeup decision is not taking the classzone into account. Now that reclaim is node-based, it is only required to wake kswapd once per node and only if all zones are unbalanced for the requested classzone. Note that one node might be checked multiple times if the zonelist is ordered by node because there is no cheap way of tracking what nodes have already been visited. For zone-ordering, each node should be checked only once. Link: http://lkml.kernel.org/r/1467970510-21195-22-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
As reclaim is now node-based, it follows that page write activity due to page reclaim should also be accounted for on the node. For consistency, also account page writes and page dirtying on a per-node basis. After this patch, there are a few remaining zone counters that may appear strange but are fine. NUMA stats are still per-zone as this is a user-space interface that tools consume. NR_MLOCK, NR_SLAB_*, NR_PAGETABLE, NR_KERNEL_STACK and NR_BOUNCE are all allocations that potentially pin low memory and cannot trivially be reclaimed on demand. This information is still useful for debugging a page allocation failure warning. Link: http://lkml.kernel.org/r/1467970510-21195-21-git-send-email-mgorman@techsingularity.netSigned-off-by:
Michal Hocko <mhocko@suse.com> Cc: Hillf Danton <hillf.zj@alibaba-inc.com> Acked-by:
-
Mel Gorman authored
There are now a number of accounting oddities such as mapped file pages being accounted for on the node while the total number of file pages are accounted on the zone. This can be coped with to some extent but it's confusing so this patch moves the relevant file-based accounted. Due to throttling logic in the page allocator for reliable OOM detection, it is still necessary to track dirty and writeback pages on a per-zone basis. [mgorman@techsingularity.net: fix NR_ZONE_WRITE_PENDING accounting] Link: http://lkml.kernel.org/r/1468404004-5085-5-git-send-email-mgorman@techsingularity.net Link: http://lkml.kernel.org/r/1467970510-21195-20-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
NR_FILE_PAGES is the number of file pages. NR_FILE_MAPPED is the number of mapped file pages. NR_ANON_PAGES is the number of mapped anon pages. This is unhelpful naming as it's easy to confuse NR_FILE_MAPPED and NR_ANON_PAGES for mapped pages. This patch renames NR_ANON_PAGES so we have NR_FILE_PAGES is the number of file pages. NR_FILE_MAPPED is the number of mapped file pages. NR_ANON_MAPPED is the number of mapped anon pages. Link: http://lkml.kernel.org/r/1467970510-21195-19-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Reclaim makes decisions based on the number of pages that are mapped but it's mixing node and zone information. Account NR_FILE_MAPPED and NR_ANON_PAGES pages on the node. Link: http://lkml.kernel.org/r/1467970510-21195-18-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Historically dirty pages were spread among zones but now that LRUs are per-node it is more appropriate to consider dirty pages in a node. Link: http://lkml.kernel.org/r/1467970510-21195-17-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Working set and refault detection is still zone-based, fix it. Link: http://lkml.kernel.org/r/1467970510-21195-16-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Memcg needs adjustment after moving LRUs to the node. Limits are tracked per memcg but the soft-limit excess is tracked per zone. As global page reclaim is based on the node, it is easy to imagine a situation where a zone soft limit is exceeded even though the memcg limit is fine. This patch moves the soft limit tree the node. Technically, all the variable names should also change but people are already familiar by the meaning of "mz" even if "mn" would be a more appropriate name now. Link: http://lkml.kernel.org/r/1467970510-21195-15-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Earlier patches focused on having direct reclaim and kswapd use data that is node-centric for reclaiming but shrink_node() itself still uses too much zone information. This patch removes unnecessary zone-based information with the most important decision being whether to continue reclaim or not. Some memcg APIs are adjusted as a result even though memcg itself still uses some zone information. [mgorman@techsingularity.net: optimization] Link: http://lkml.kernel.org/r/1468588165-12461-2-git-send-email-mgorman@techsingularity.net Link: http://lkml.kernel.org/r/1467970510-21195-14-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
kswapd scans from highest to lowest for a zone that requires balancing. This was necessary when reclaim was per-zone to fairly age pages on lower zones. Now that we are reclaiming on a per-node basis, any eligible zone can be used and pages will still be aged fairly. This patch avoids reclaiming excessively unless buffer_heads are over the limit and it's necessary to reclaim from a higher zone than requested by the waker of kswapd to relieve low memory pressure. [hillf.zj@alibaba-inc.com: Force kswapd reclaim no more than needed] Link: http://lkml.kernel.org/r/1466518566-30034-12-git-send-email-mgorman@techsingularity.net Link: http://lkml.kernel.org/r/1467970510-21195-13-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Reclaim may stall if there is too much dirty or congested data on a node. This was previously based on zone flags and the logic for clearing the flags is in two places. As congestion/dirty tracking is now tracked on a per-node basis, we can remove some duplicate logic. Link: http://lkml.kernel.org/r/1467970510-21195-12-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Direct reclaim iterates over all zones in the zonelist and shrinking them but this is in conflict with node-based reclaim. In the default case, only shrink once per node. Link: http://lkml.kernel.org/r/1467970510-21195-11-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
kswapd goes through some complex steps trying to figure out if it should stay awake based on the classzone_idx and the requested order. It is unnecessarily complex and passes in an invalid classzone_idx to balance_pgdat(). What matters most of all is whether a larger order has been requsted and whether kswapd successfully reclaimed at the previous order. This patch irons out the logic to check just that and the end result is less headache inducing. Link: http://lkml.kernel.org/r/1467970510-21195-10-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
The balance gap was introduced to apply equal pressure to all zones when reclaiming for a higher zone. With node-based LRU, the need for the balance gap is removed and the code is dead so remove it. [vbabka@suse.cz: Also remove KSWAPD_ZONE_BALANCE_GAP_RATIO] Link: http://lkml.kernel.org/r/1467970510-21195-9-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Patch "mm: vmscan: Begin reclaiming pages on a per-node basis" started thinking of reclaim in terms of nodes but kswapd is still zone-centric. This patch gets rid of many of the node-based versus zone-based decisions. o A node is considered balanced when any eligible lower zone is balanced. This eliminates one class of age-inversion problem because we avoid reclaiming a newer page just because it's in the wrong zone o pgdat_balanced disappears because we now only care about one zone being balanced. o Some anomalies related to writeback and congestion tracking being based on zones disappear. o kswapd no longer has to take care to reclaim zones in the reverse order that the page allocator uses. o Most importantly of all, reclaim from node 0 with multiple zones will have similar aging and reclaiming characteristics as every other node. Link: http://lkml.kernel.org/r/1467970510-21195-8-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
kswapd checks all eligible zones to see if they need balancing even if it was woken for a lower zone. This made sense when we reclaimed on a per-zone basis because we wanted to shrink zones fairly so avoid age-inversion problems. Ideally this is completely unnecessary when reclaiming on a per-node basis. In theory, there may still be anomalies when all requests are for lower zones and very old pages are preserved in higher zones but this should be the exceptional case. Link: http://lkml.kernel.org/r/1467970510-21195-7-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
This patch makes reclaim decisions on a per-node basis. A reclaimer knows what zone is required by the allocation request and skips pages from higher zones. In many cases this will be ok because it's a GFP_HIGHMEM request of some description. On 64-bit, ZONE_DMA32 requests will cause some problems but 32-bit devices on 64-bit platforms are increasingly rare. Historically it would have been a major problem on 32-bit with big Highmem:Lowmem ratios but such configurations are also now rare and even where they exist, they are not encouraged. If it really becomes a problem, it'll manifest as very low reclaim efficiencies. Link: http://lkml.kernel.org/r/1467970510-21195-6-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Zone padding separates write-intensive fields used by page allocation, compaction and vmstats but the comments are a little misleading and need clarification. Link: http://lkml.kernel.org/r/1467970510-21195-5-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
This moves the LRU lists from the zone to the node and related data such as counters, tracing, congestion tracking and writeback tracking. Unfortunately, due to reclaim and compaction retry logic, it is necessary to account for the number of LRU pages on both zone and node logic. Most reclaim logic is based on the node counters but the retry logic uses the zone counters which do not distinguish inactive and active sizes. It would be possible to leave the LRU counters on a per-zone basis but it's a heavier calculation across multiple cache lines that is much more frequent than the retry checks. Other than the LRU counters, this is mostly a mechanical patch but note that it introduces a number of anomalies. For example, the scans are per-zone but using per-node counters. We also mark a node as congested when a zone is congested. This causes weird problems that are fixed later but is easier to review. In the event that there is excessive overhead on 32-bit systems due to the nodes being on LRU then there are two potential solutions 1. Long-term isolation of highmem pages when reclaim is lowmem When pages are skipped, they are immediately added back onto the LRU list. If lowmem reclaim persisted for long periods of time, the same highmem pages get continually scanned. The idea would be that lowmem keeps those pages on a separate list until a reclaim for highmem pages arrives that splices the highmem pages back onto the LRU. It potentially could be implemented similar to the UNEVICTABLE list. That would reduce the skip rate with the potential corner case is that highmem pages have to be scanned and reclaimed to free lowmem slab pages. 2. Linear scan lowmem pages if the initial LRU shrink fails This will break LRU ordering but may be preferable and faster during memory pressure than skipping LRU pages. Link: http://lkml.kernel.org/r/1467970510-21195-4-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Mel Gorman authored
Node-based reclaim requires node-based LRUs and locking. This is a preparation patch that just moves the lru_lock to the node so later patches are easier to review. It is a mechanical change but note this patch makes contention worse because the LRU lock is hotter and direct reclaim and kswapd can contend on the same lock even when reclaiming from different zones. Link: http://lkml.kernel.org/r/1467970510-21195-3-git-send-email-mgorman@techsingularity.netSigned-off-by:
Minchan Kim <minchan@kernel.org> Acked-by:
-
Mel Gorman authored
Patchset: "Move LRU page reclaim from zones to nodes v9" This series moves LRUs from the zones to the node. While this is a current rebase, the test results were based on mmotm as of June 23rd. Conceptually, this series is simple but there are a lot of details. Some of the broad motivations for this are; 1. The residency of a page partially depends on what zone the page was allocated from. This is partially combatted by the fair zone allocation policy but that is a partial solution that introduces overhead in the page allocator paths. 2. Currently, reclaim on node 0 behaves slightly different to node 1. For example, direct reclaim scans in zonelist order and reclaims even if the zone is over the high watermark regardless of the age of pages in that LRU. Kswapd on the other hand starts reclaim on the highest unbalanced zone. A difference in distribution of file/anon pages due to when they were allocated results can result in a difference in again. While the fair zone allocation policy mitigates some of the problems here, the page reclaim results on a multi-zone node will always be different to a single-zone node. it was scheduled on as a result. 3. kswapd and the page allocator scan zones in the opposite order to avoid interfering with each other but it's sensitive to timing. This mitigates the page allocator using pages that were allocated very recently in the ideal case but it's sensitive to timing. When kswapd is allocating from lower zones then it's great but during the rebalancing of the highest zone, the page allocator and kswapd interfere with each other. It's worse if the highest zone is small and difficult to balance. 4. slab shrinkers are node-based which makes it harder to identify the exact relationship between slab reclaim and LRU reclaim. The reason we have zone-based reclaim is that we used to have large highmem zones in common configurations and it was necessary to quickly find ZONE_NORMAL pages for reclaim. Today, this is much less of a concern as machines with lots of memory will (or should) use 64-bit kernels. Combinations of 32-bit hardware and 64-bit hardware are rare. Machines that do use highmem should have relatively low highmem:lowmem ratios than we worried about in the past. Conceptually, moving to node LRUs should be easier to understand. The page allocator plays fewer tricks to game reclaim and reclaim behaves similarly on all nodes. The series has been tested on a 16 core UMA machine and a 2-socket 48 core NUMA machine. The UMA results are presented in most cases as the NUMA machine behaved similarly. pagealloc --------- This is a microbenchmark that shows the benefit of removing the fair zone allocation policy. It was tested uip to order-4 but only orders 0 and 1 are shown as the other orders were comparable. 4.7.0-rc4 4.7.0-rc4 mmotm-20160623 nodelru-v9 Min total-odr0-1 490.00 ( 0.00%) 457.00 ( 6.73%) Min total-odr0-2 347.00 ( 0.00%) 329.00 ( 5.19%) Min total-odr0-4 288.00 ( 0.00%) 273.00 ( 5.21%) Min total-odr0-8 251.00 ( 0.00%) 239.00 ( 4.78%) Min total-odr0-16 234.00 ( 0.00%) 222.00 ( 5.13%) Min total-odr0-32 223.00 ( 0.00%) 211.00 ( 5.38%) Min total-odr0-64 217.00 ( 0.00%) 208.00 ( 4.15%) Min total-odr0-128 214.00 ( 0.00%) 204.00 ( 4.67%) Min total-odr0-256 250.00 ( 0.00%) 230.00 ( 8.00%) Min total-odr0-512 271.00 ( 0.00%) 269.00 ( 0.74%) Min total-odr0-1024 291.00 ( 0.00%) 282.00 ( 3.09%) Min total-odr0-2048 303.00 ( 0.00%) 296.00 ( 2.31%) Min total-odr0-4096 311.00 ( 0.00%) 309.00 ( 0.64%) Min total-odr0-8192 316.00 ( 0.00%) 314.00 ( 0.63%) Min total-odr0-16384 317.00 ( 0.00%) 315.00 ( 0.63%) Min total-odr1-1 742.00 ( 0.00%) 712.00 ( 4.04%) Min total-odr1-2 562.00 ( 0.00%) 530.00 ( 5.69%) Min total-odr1-4 457.00 ( 0.00%) 433.00 ( 5.25%) Min total-odr1-8 411.00 ( 0.00%) 381.00 ( 7.30%) Min total-odr1-16 381.00 ( 0.00%) 356.00 ( 6.56%) Min total-odr1-32 372.00 ( 0.00%) 346.00 ( 6.99%) Min total-odr1-64 372.00 ( 0.00%) 343.00 ( 7.80%) Min total-odr1-128 375.00 ( 0.00%) 351.00 ( 6.40%) Min total-odr1-256 379.00 ( 0.00%) 351.00 ( 7.39%) Min total-odr1-512 385.00 ( 0.00%) 355.00 ( 7.79%) Min total-odr1-1024 386.00 ( 0.00%) 358.00 ( 7.25%) Min total-odr1-2048 390.00 ( 0.00%) 362.00 ( 7.18%) Min total-odr1-4096 390.00 ( 0.00%) 362.00 ( 7.18%) Min total-odr1-8192 388.00 ( 0.00%) 363.00 ( 6.44%) This shows a steady improvement throughout. The primary benefit is from reduced system CPU usage which is obvious from the overall times; 4.7.0-rc4 4.7.0-rc4 mmotm-20160623nodelru-v8 User 189.19 191.80 System 2604.45 2533.56 Elapsed 2855.30 2786.39 The vmstats also showed that the fair zone allocation policy was definitely removed as can be seen here; 4.7.0-rc3 4.7.0-rc3 mmotm-20160623 nodelru-v8 DMA32 allocs 28794729769 0 Normal allocs 48432501431 77227309877 Movable allocs 0 0 tiobench on ext4 ---------------- tiobench is a benchmark that artifically benefits if old pages remain resident while new pages get reclaimed. The fair zone allocation policy mitigates this problem so pages age fairly. While the benchmark has problems, it is important that tiobench performance remains constant as it implies that page aging problems that the fair zone allocation policy fixes are not re-introduced. 4.7.0-rc4 4.7.0-rc4 mmotm-20160623 nodelru-v9 Min PotentialReadSpeed 89.65 ( 0.00%) 90.21 ( 0.62%) Min SeqRead-MB/sec-1 82.68 ( 0.00%) 82.01 ( -0.81%) Min SeqRead-MB/sec-2 72.76 ( 0.00%) 72.07 ( -0.95%) Min SeqRead-MB/sec-4 75.13 ( 0.00%) 74.92 ( -0.28%) Min SeqRead-MB/sec-8 64.91 ( 0.00%) 65.19 ( 0.43%) Min SeqRead-MB/sec-16 62.24 ( 0.00%) 62.22 ( -0.03%) Min RandRead-MB/sec-1 0.88 ( 0.00%) 0.88 ( 0.00%) Min RandRead-MB/sec-2 0.95 ( 0.00%) 0.92 ( -3.16%) Min RandRead-MB/sec-4 1.43 ( 0.00%) 1.34 ( -6.29%) Min RandRead-MB/sec-8 1.61 ( 0.00%) 1.60 ( -0.62%) Min RandRead-MB/sec-16 1.80 ( 0.00%) 1.90 ( 5.56%) Min SeqWrite-MB/sec-1 76.41 ( 0.00%) 76.85 ( 0.58%) Min SeqWrite-MB/sec-2 74.11 ( 0.00%) 73.54 ( -0.77%) Min SeqWrite-MB/sec-4 80.05 ( 0.00%) 80.13 ( 0.10%) Min SeqWrite-MB/sec-8 72.88 ( 0.00%) 73.20 ( 0.44%) Min SeqWrite-MB/sec-16 75.91 ( 0.00%) 76.44 ( 0.70%) Min RandWrite-MB/sec-1 1.18 ( 0.00%) 1.14 ( -3.39%) Min RandWrite-MB/sec-2 1.02 ( 0.00%) 1.03 ( 0.98%) Min RandWrite-MB/sec-4 1.05 ( 0.00%) 0.98 ( -6.67%) Min RandWrite-MB/sec-8 0.89 ( 0.00%) 0.92 ( 3.37%) Min RandWrite-MB/sec-16 0.92 ( 0.00%) 0.93 ( 1.09%) 4.7.0-rc4 4.7.0-rc4 mmotm-20160623 approx-v9 User 645.72 525.90 System 403.85 331.75 Elapsed 6795.36 6783.67 This shows that the series has little or not impact on tiobench which is desirable and a reduction in system CPU usage. It indicates that the fair zone allocation policy was removed in a manner that didn't reintroduce one class of page aging bug. There were only minor differences in overall reclaim activity 4.7.0-rc4 4.7.0-rc4 mmotm-20160623nodelru-v8 Minor Faults 645838 647465 Major Faults 573 640 Swap Ins 0 0 Swap Outs 0 0 DMA allocs 0 0 DMA32 allocs 46041453 44190646 Normal allocs 78053072 79887245 Movable allocs 0 0 Allocation stalls 24 67 Stall zone DMA 0 0 Stall zone DMA32 0 0 Stall zone Normal 0 2 Stall zone HighMem 0 0 Stall zone Movable 0 65 Direct pages scanned 10969 30609 Kswapd pages scanned 93375144 93492094 Kswapd pages reclaimed 93372243 93489370 Direct pages reclaimed 10969 30609 Kswapd efficiency 99% 99% Kswapd velocity 13741.015 13781.934 Direct efficiency 100% 100% Direct velocity 1.614 4.512 Percentage direct scans 0% 0% kswapd activity was roughly comparable. There were differences in direct reclaim activity but negligible in the context of the overall workload (velocity of 4 pages per second with the patches applied, 1.6 pages per second in the baseline kernel). pgbench read-only large configuration on ext4 --------------------------------------------- pgbench is a database benchmark that can be sensitive to page reclaim decisions. This also checks if removing the fair zone allocation policy is safe pgbench Transactions 4.7.0-rc4 4.7.0-rc4 mmotm-20160623 nodelru-v8 Hmean 1 188.26 ( 0.00%) 189.78 ( 0.81%) Hmean 5 330.66 ( 0.00%) 328.69 ( -0.59%) Hmean 12 370.32 ( 0.00%) 380.72 ( 2.81%) Hmean 21 368.89 ( 0.00%) 369.00 ( 0.03%) Hmean 30 382.14 ( 0.00%) 360.89 ( -5.56%) Hmean 32 428.87 ( 0.00%) 432.96 ( 0.95%) Negligible differences again. As with tiobench, overall reclaim activity was comparable. bonnie++ on ext4 ---------------- No interesting performance difference, negligible differences on reclaim stats. paralleldd on ext4 ------------------ This workload uses varying numbers of dd instances to read large amounts of data from disk. 4.7.0-rc3 4.7.0-rc3 mmotm-20160623 nodelru-v9 Amean Elapsd-1 186.04 ( 0.00%) 189.41 ( -1.82%) Amean Elapsd-3 192.27 ( 0.00%) 191.38 ( 0.46%) Amean Elapsd-5 185.21 ( 0.00%) 182.75 ( 1.33%) Amean Elapsd-7 183.71 ( 0.00%) 182.11 ( 0.87%) Amean Elapsd-12 180.96 ( 0.00%) 181.58 ( -0.35%) Amean Elapsd-16 181.36 ( 0.00%) 183.72 ( -1.30%) 4.7.0-rc4 4.7.0-rc4 mmotm-20160623 nodelru-v9 User 1548.01 1552.44 System 8609.71 8515.08 Elapsed 3587.10 3594.54 There is little or no change in performance but some drop in system CPU usage. 4.7.0-rc3 4.7.0-rc3 mmotm-20160623 nodelru-v9 Minor Faults 362662 367360 Major Faults 1204 1143 Swap Ins 22 0 Swap Outs 2855 1029 DMA allocs 0 0 DMA32 allocs 31409797 28837521 Normal allocs 46611853 49231282 Movable allocs 0 0 Direct pages scanned 0 0 Kswapd pages scanned 40845270 40869088 Kswapd pages reclaimed 40830976 40855294 Direct pages reclaimed 0 0 Kswapd efficiency 99% 99% Kswapd velocity 11386.711 11369.769 Direct efficiency 100% 100% Direct velocity 0.000 0.000 Percentage direct scans 0% 0% Page writes by reclaim 2855 1029 Page writes file 0 0 Page writes anon 2855 1029 Page reclaim immediate 771 1628 Sector Reads 293312636 293536360 Sector Writes 18213568 18186480 Page rescued immediate 0 0 Slabs scanned 128257 132747 Direct inode steals 181 56 Kswapd inode steals 59 1131 It basically shows that kswapd was active at roughly the same rate in both kernels. There was also comparable slab scanning activity and direct reclaim was avoided in both cases. There appears to be a large difference in numbers of inodes reclaimed but the workload has few active inodes and is likely a timing artifact. stutter ------- stutter simulates a simple workload. One part uses a lot of anonymous memory, a second measures mmap latency and a third copies a large file. The primary metric is checking for mmap latency. stutter 4.7.0-rc4 4.7.0-rc4 mmotm-20160623 nodelru-v8 Min mmap 16.6283 ( 0.00%) 13.4258 ( 19.26%) 1st-qrtle mmap 54.7570 ( 0.00%) 34.9121 ( 36.24%) 2nd-qrtle mmap 57.3163 ( 0.00%) 46.1147 ( 19.54%) 3rd-qrtle mmap 58.9976 ( 0.00%) 47.1882 ( 20.02%) Max-90% mmap 59.7433 ( 0.00%) 47.4453 ( 20.58%) Max-93% mmap 60.1298 ( 0.00%) 47.6037 ( 20.83%) Max-95% mmap 73.4112 ( 0.00%) 82.8719 (-12.89%) Max-99% mmap 92.8542 ( 0.00%) 88.8870 ( 4.27%) Max mmap 1440.6569 ( 0.00%) 121.4201 ( 91.57%) Mean mmap 59.3493 ( 0.00%) 42.2991 ( 28.73%) Best99%Mean mmap 57.2121 ( 0.00%) 41.8207 ( 26.90%) Best95%Mean mmap 55.9113 ( 0.00%) 39.9620 ( 28.53%) Best90%Mean mmap 55.6199 ( 0.00%) 39.3124 ( 29.32%) Best50%Mean mmap 53.2183 ( 0.00%) 33.1307 ( 37.75%) Best10%Mean mmap 45.9842 ( 0.00%) 20.4040 ( 55.63%) Best5%Mean mmap 43.2256 ( 0.00%) 17.9654 ( 58.44%) Best1%Mean mmap 32.9388 ( 0.00%) 16.6875 ( 49.34%) This shows a number of improvements with the worst-case outlier greatly improved. Some of the vmstats are interesting 4.7.0-rc4 4.7.0-rc4 mmotm-20160623nodelru-v8 Swap Ins 163 502 Swap Outs 0 0 DMA allocs 0 0 DMA32 allocs 618719206 1381662383 Normal allocs 891235743 564138421 Movable allocs 0 0 Allocation stalls 2603 1 Direct pages scanned 216787 2 Kswapd pages scanned 50719775 41778378 Kswapd pages reclaimed 41541765 41777639 Direct pages reclaimed 209159 0 Kswapd efficiency 81% 99% Kswapd velocity 16859.554 14329.059 Direct efficiency 96% 0% Direct velocity 72.061 0.001 Percentage direct scans 0% 0% Page writes by reclaim 6215049 0 Page writes file 6215049 0 Page writes anon 0 0 Page reclaim immediate 70673 90 Sector Reads 81940800 81680456 Sector Writes 100158984 98816036 Page rescued immediate 0 0 Slabs scanned 1366954 22683 While this is not guaranteed in all cases, this particular test showed a large reduction in direct reclaim activity. It's also worth noting that no page writes were issued from reclaim context. This series is not without its hazards. There are at least three areas that I'm concerned with even though I could not reproduce any problems in that area. 1. Reclaim/compaction is going to be affected because the amount of reclaim is no longer targetted at a specific zone. Compaction works on a per-zone basis so there is no guarantee that reclaiming a few THP's worth page pages will have a positive impact on compaction success rates. 2. The Slab/LRU reclaim ratio is affected because the frequency the shrinkers are called is now different. This may or may not be a problem but if it is, it'll be because shrinkers are not called enough and some balancing is required. 3. The anon/file reclaim ratio may be affected. Pages about to be dirtied are distributed between zones and the fair zone allocation policy used to do something very similar for anon. The distribution is now different but not necessarily in any way that matters but it's still worth bearing in mind. VM statistic counters for reclaim decisions are zone-based. If the kernel is to reclaim on a per-node basis then we need to track per-node statistics but there is no infrastructure for that. The most notable change is that the old node_page_state is renamed to sum_zone_node_page_state. The new node_page_state takes a pglist_data and uses per-node stats but none exist yet. There is some renaming such as vm_stat to vm_zone_stat and the addition of vm_node_stat and the renaming of mod_state to mod_zone_state. Otherwise, this is mostly a mechanical patch with no functional change. There is a lot of similarity between the node and zone helpers which is unfortunate but there was no obvious way of reusing the code and maintaining type safety. Link: http://lkml.kernel.org/r/1467970510-21195-2-git-send-email-mgorman@techsingularity.netSigned-off-by: -
Mel Gorman authored
The helper early_page_nid_uninitialised() has been dead since commit 974a786e ("mm, page_alloc: remove MIGRATE_RESERVE") so remove the dead code. Link: http://lkml.kernel.org/r/1468008031-3848-2-git-send-email-mgorman@techsingularity.netSigned-off-by:
-
Michal Hocko authored
Commit c0ff7453 ("cpuset,mm: fix no node to alloc memory when changing cpuset's mems") has added TIF_MEMDIE and PF_EXITING check but it is checking the flag on the current task rather than the given one. This doesn't make much sense and it is actually wrong. If the current task which updates the nodemask of a cpuset got killed by the OOM killer then a part of the cpuset cgroup processes would have incompatible nodemask which is surprising to say the least. The comment suggests the intention was to skip oom victim or an exiting task so we should be checking the given task. But even then it would be layering violation because it is the memory allocator to interpret the TIF_MEMDIE meaning. Simply drop both checks. All tasks in the cpuset should simply follow the same mask. Link: http://lkml.kernel.org/r/1467029719-17602-3-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
freezing_slow_path() is checking TIF_MEMDIE to skip OOM killed tasks. It is, however, checking the flag on the current task rather than the given one. This is really confusing because freezing() can be called also on !current tasks. It would end up working correctly for its main purpose because __refrigerator will be always called on the current task so the oom victim will never get frozen. But it could lead to surprising results when a task which is freezing a cgroup got oom killed because only part of the cgroup would get frozen. This is highly unlikely but worth fixing as the resulting code would be more clear anyway. Link: http://lkml.kernel.org/r/1467029719-17602-2-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Ganesh Mahendran authored
The caller __alloc_pages_direct_compact() already checked (order == 0) so there's no need to check again. Link: http://lkml.kernel.org/r/1465973568-3496-1-git-send-email-opensource.ganesh@gmail.comSigned-off-by:
Ganesh Mahendran <opensource.ganesh@gmail.com> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com> Cc: Michal Hocko <mhocko@suse.com> Cc: Michal Nazarewicz <mina86@mina86.com> Cc: Minchan Kim <minchan@kernel.org> Cc: Anshuman Khandual <khandual@linux.vnet.ibm.com> Signed-off-by:
-
Johannes Weiner authored
Commit 23047a96 ("mm: workingset: per-cgroup cache thrash detection") added a page->mem_cgroup lookup to the cache eviction, refault, and activation paths, as well as locking to the activation path, and the vm-scalability tests showed a regression of -23%. While the test in question is an artificial worst-case scenario that doesn't occur in real workloads - reading two sparse files in parallel at full CPU speed just to hammer the LRU paths - there is still some optimizations that can be done in those paths. Inline the lookup functions to eliminate calls. Also, page->mem_cgroup doesn't need to be stabilized when counting an activation; we merely need to hold the RCU lock to prevent the memcg from being freed. This cuts down on overhead quite a bit: 23047a96 063f6715e77a7be5770d6081fe ---------------- -------------------------- %stddev %change %stddev \ | \ 21621405 +- 0% +11.3% 24069657 +- 2% vm-scalability.throughput [linux@roeck-us.net: drop unnecessary include file] [hannes@cmpxchg.org: add WARN_ON_ONCE()s] Link: http://lkml.kernel.org/r/20160707194024.GA26580@cmpxchg.org Link: http://lkml.kernel.org/r/20160624175101.GA3024@cmpxchg.orgReported-by:
Ye Xiaolong <xiaolong.ye@intel.com> Signed-off-by:
Guenter Roeck <linux@roeck-us.net> Signed-off-by:
-
zhong jiang authored
We need to assure the comment is consistent with the code. [akpm@linux-foundation.org: coding-style fixes] Link: http://lkml.kernel.org/r/1466171914-21027-1-git-send-email-zhongjiang@huawei.comSigned-off-by:
zhong jiang <zhongjiang@huawei.com> Signed-off-by:
-
Michal Hocko authored
"mm, oom: fortify task_will_free_mem" has dropped task_lock around task_will_free_mem in oom_kill_process bacause it assumed that a potential race when the selected task exits will not be a problem as the oom_reaper will call exit_oom_victim. Tetsuo was objecting that nommu doesn't have oom_reaper so the race would be still possible. The code would be racy and lockup prone theoretically in other aspects without the oom reaper anyway so I didn't considered this a big deal. But it seems that further changes I am planning in this area will benefit from stable task->mm in this path as well. So let's drop find_lock_task_mm from task_will_free_mem and call it from under task_lock as we did previously. Just pull the task->mm != NULL check inside the function. Link: http://lkml.kernel.org/r/1467201562-6709-1-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
The only case where the oom_reaper is not triggered for the oom victim is when it shares the memory with a kernel thread (aka use_mm) or with the global init. After "mm, oom: skip vforked tasks from being selected" the victim cannot be a vforked task of the global init so we are left with clone(CLONE_VM) (without CLONE_SIGHAND). use_mm() users are quite rare as well. In order to help forward progress for the OOM killer, make sure that this really rare case will not get in the way - we do this by hiding the mm from the oom killer by setting MMF_OOM_REAPED flag for it. oom_scan_process_thread will ignore any TIF_MEMDIE task if it has MMF_OOM_REAPED flag set to catch these oom victims. After this patch we should guarantee forward progress for the OOM killer even when the selected victim is sharing memory with a kernel thread or global init as long as the victims mm is still alive. Link: http://lkml.kernel.org/r/1466426628-15074-11-git-send-email-mhocko@kernel.orgSigned-off-by:
Oleg Nesterov <oleg@redhat.com> Cc: Vladimir Davydov <vdavydov@virtuozzo.com> Cc: David Rientjes <rientjes@google.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Signed-off-by:
-
Michal Hocko authored
oom_reaper relies on the mmap_sem for read to do its job. Many places which might block readers have been converted to use down_write_killable and that has reduced chances of the contention a lot. Some paths where the mmap_sem is held for write can take other locks and they might either be not prepared to fail due to fatal signal pending or too impractical to be changed. This patch introduces MMF_OOM_NOT_REAPABLE flag which gets set after the first attempt to reap a task's mm fails. If the flag is present after the failure then we set MMF_OOM_REAPED to hide this mm from the oom killer completely so it can go and chose another victim. As a result a risk of OOM deadlock when the oom victim would be blocked indefinetly and so the oom killer cannot make any progress should be mitigated considerably while we still try really hard to perform all reclaim attempts and stay predictable in the behavior. Link: http://lkml.kernel.org/r/1466426628-15074-10-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
The 0-day robot has encountered the following: Out of memory: Kill process 3914 (trinity-c0) score 167 or sacrifice child Killed process 3914 (trinity-c0) total-vm:55864kB, anon-rss:1512kB, file-rss:1088kB, shmem-rss:25616kB oom_reaper: reaped process 3914 (trinity-c0), now anon-rss:0kB, file-rss:0kB, shmem-rss:26488kB oom_reaper: reaped process 3914 (trinity-c0), now anon-rss:0kB, file-rss:0kB, shmem-rss:26900kB oom_reaper: reaped process 3914 (trinity-c0), now anon-rss:0kB, file-rss:0kB, shmem-rss:26900kB oom_reaper: reaped process 3914 (trinity-c0), now anon-rss:0kB, file-rss:0kB, shmem-rss:27296kB oom_reaper: reaped process 3914 (trinity-c0), now anon-rss:0kB, file-rss:0kB, shmem-rss:28148kB oom_reaper is trying to reap the same task again and again. This is possible only when the oom killer is bypassed because of task_will_free_mem because we skip over tasks with MMF_OOM_REAPED already set during select_bad_process. Teach task_will_free_mem to skip over MMF_OOM_REAPED tasks as well because they will be unlikely to free anything more. Analyzed by Tetsuo Handa. Link: http://lkml.kernel.org/r/1466426628-15074-9-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
task_will_free_mem is rather weak. It doesn't really tell whether the task has chance to drop its mm. 98748bd7 ("oom: consider multi-threaded tasks in task_will_free_mem") made a first step into making it more robust for multi-threaded applications so now we know that the whole process is going down and probably drop the mm. This patch builds on top for more complex scenarios where mm is shared between different processes - CLONE_VM without CLONE_SIGHAND, or in kernel use_mm(). Make sure that all processes sharing the mm are killed or exiting. This will allow us to replace try_oom_reaper by wake_oom_reaper because task_will_free_mem implies the task is reapable now. Therefore all paths which bypass the oom killer are now reapable and so they shouldn't lock up the oom killer. Link: http://lkml.kernel.org/r/1466426628-15074-8-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
Currently oom_kill_process skips both the oom reaper and SIG_KILL if a process sharing the same mm is unkillable via OOM_ADJUST_MIN. After "mm, oom_adj: make sure processes sharing mm have same view of oom_score_adj" all such processes are sharing the same value so we shouldn't see such a task at all (oom_badness would rule them out). We can still encounter oom disabled vforked task which has to be killed as well if we want to have other tasks sharing the mm reapable because it can access the memory before doing exec. Killing such a task should be acceptable because it is highly unlikely it has done anything useful because it cannot modify any memory before it calls exec. An alternative would be to keep the task alive and skip the oom reaper and risk all the weird corner cases where the OOM killer cannot make forward progress because the oom victim hung somewhere on the way to exit. [rientjes@google.com - drop printk when OOM_SCORE_ADJ_MIN killed task the setting is inherently racy and we cannot do much about it without introducing locks in hot paths] Link: http://lkml.kernel.org/r/1466426628-15074-7-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
vforked tasks are not really sitting on any memory. They are sharing the mm with parent until they exec into a new code. Until then it is just pinning the address space. OOM killer will kill the vforked task along with its parent but we still can end up selecting vforked task when the parent wouldn't be selected. E.g. init doing vfork to launch a task or vforked being a child of oom unkillable task with an updated oom_score_adj to be killable. Add a new helper to check whether a task is in the vfork sharing memory with its parent and use it in oom_badness to skip over these tasks. Link: http://lkml.kernel.org/r/1466426628-15074-6-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
oom_score_adj is shared for the thread groups (via struct signal) but this is not sufficient to cover processes sharing mm (CLONE_VM without CLONE_SIGHAND) and so we can easily end up in a situation when some processes update their oom_score_adj and confuse the oom killer. In the worst case some of those processes might hide from the oom killer altogether via OOM_SCORE_ADJ_MIN while others are eligible. OOM killer would then pick up those eligible but won't be allowed to kill others sharing the same mm so the mm wouldn't release the mm and so the memory. It would be ideal to have the oom_score_adj per mm_struct because that is the natural entity OOM killer considers. But this will not work because some programs are doing vfork() set_oom_adj() exec() We can achieve the same though. oom_score_adj write handler can set the oom_score_adj for all processes sharing the same mm if the task is not in the middle of vfork. As a result all the processes will share the same oom_score_adj. The current implementation is rather pessimistic and checks all the existing processes by default if there is more than 1 holder of the mm but we do not have any reliable way to check for external users yet. Link: http://lkml.kernel.org/r/1466426628-15074-5-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
Currently we have two proc interfaces to set oom_score_adj. The legacy /proc/<pid>/oom_adj and /proc/<pid>/oom_score_adj which both have their specific handlers. Big part of the logic is duplicated so extract the common code into __set_oom_adj helper. Legacy knob still expects some details slightly different so make sure those are handled same way - e.g. the legacy mode ignores oom_score_adj_min and it warns about the usage. This patch shouldn't introduce any functional changes. Link: http://lkml.kernel.org/r/1466426628-15074-4-git-send-email-mhocko@kernel.orgSigned-off-by:
-
Michal Hocko authored
Oleg has pointed out that can simplify both oom_adj_{read,write} and oom_score_adj_{read,write} even further and drop the sighand lock. The main purpose of the lock was to protect p->signal from going away but this will not happen since ea6d290c ("signals: make task_struct->signal immutable/refcountable"). The other role of the lock was to synchronize different writers, especially those with CAP_SYS_RESOURCE. Introduce a mutex for this purpose. Later patches will need this lock anyway. Suggested-by: -
Michal Hocko authored
Series "Handle oom bypass more gracefully", V5 The following 10 patches should put some order to very rare cases of mm shared between processes and make the paths which bypass the oom killer oom reapable and therefore much more reliable finally. Even though mm shared outside of thread group is rare (either vforked tasks for a short period, use_mm by kernel threads or exotic thread model of clone(CLONE_VM) without CLONE_SIGHAND) it is better to cover them. Not only it makes the current oom killer logic quite hard to follow and reason about it can lead to weird corner cases. E.g. it is possible to select an oom victim which shares the mm with unkillable process or bypass the oom killer even when other processes sharing the mm are still alive and other weird cases. Patch 1 drops bogus task_lock and mm check from oom_{score_}adj_write. This can be considered a bug fix with a low impact as nobody has noticed for years. Patch 2 drops sighand lock because it is not needed anymore as pointed by Oleg. Patch 3 is a clean up of oom_score_adj handling and a preparatory work for later patches. Patch 4 enforces oom_adj_score to be consistent between processes sharing the mm to behave consistently with the regular thread groups. This can be considered a user visible behavior change because one thread group updating oom_score_adj will affect others which share the same mm via clone(CLONE_VM). I argue that this should be acceptable because we already have the same behavior for threads in the same thread group and sharing the mm without signal struct is just a different model of threading. This is probably the most controversial part of the series, I would like to find some consensus here. There were some suggestions to hook some counter/oom_score_adj into the mm_struct but I feel that this is not necessary right now and we can rely on proc handler + oom_kill_process to DTRT. I can be convinced otherwise but I strongly think that whatever we do the userspace has to have a way to see the current oom priority as consistently as possible. Patch 5 makes sure that no vforked task is selected if it is sharing the mm with oom unkillable task. Patch 6 ensures that all user tasks sharing the mm are killed which in turn makes sure that all oom victims are oom reapable. Patch 7 guarantees that task_will_free_mem will always imply reapable bypass of the oom killer. Patch 8 is new in this version and it addresses an issue pointed out by 0-day OOM report where an oom victim was reaped several times. Patch 9 puts an upper bound on how many times oom_reaper tries to reap a task and hides it from the oom killer to move on when no progress can be made. This will give an upper bound to how long an oom_reapable task can block the oom killer from selecting another victim if the oom_reaper is not able to reap the victim. Patch 10 tries to plug the (hopefully) last hole when we can still lock up when the oom victim is shared with oom unkillable tasks (kthreads and global init). We just try to be best effort in that case and rather fallback to kill something else than risk a lockup. This patch (of 10): Both oom_adj_write and oom_score_adj_write are using task_lock, check for task->mm and fail if it is NULL. This is not needed because the oom_score_adj is per signal struct so we do not need mm at all. The code has been introduced by 3d5992d2 ("oom: add per-mm oom disable count") but we do not do per-mm oom disable since c9f01245 ("oom: remove oom_disable_count"). The task->mm check is even not correct because the current thread might have exited but the thread group might be still alive - e.g. thread group leader would lead that echo $VAL > /proc/pid/oom_score_adj would always fail with EINVAL while /proc/pid/task/$other_tid/oom_score_adj would succeed. This is unexpected at best. Remove the lock along with the check to fix the unexpected behavior and also because there is not real need for the lock in the first place. Link: http://lkml.kernel.org/r/1466426628-15074-2-git-send-email-mhocko@kernel.orgSigned-off-by:Vladimir Davydov <vdavydov@virtuozzo.com> Acked-by:
-
