- 20 May, 2016 40 commits
-
-
Mel Gorman authored
Signed-off-by:
Mel Gorman <mgorman@techsingularity.net> Acked-by:
Vlastimil Babka <vbabka@suse.cz> Cc: Jesper Dangaard Brouer <brouer@redhat.com> Signed-off-by:
Andrew Morton <akpm@linux-foundation.org> Signed-off-by:
Linus Torvalds <torvalds@linux-foundation.org>
-
Mel Gorman authored
zonelist here is a copy of a struct field that is used once. Ditch it. Signed-off-by:
-
Mel Gorman authored
The number of zones skipped to a zone expiring its fair zone allocation quota is irrelevant. Convert to bool. Signed-off-by:
-
Mel Gorman authored
alloc_flags is a bitmask of flags but it is signed which does not necessarily generate the best code depending on the compiler. Even without an impact, it makes more sense that this be unsigned. Signed-off-by:
-
Mel Gorman authored
Pageblocks have an associated bitmap to store migrate types and whether the pageblock should be skipped during compaction. The bitmap may be associated with a memory section or a zone but the zone is looked up unconditionally. The compiler should optimise this away automatically so this is a cosmetic patch only in many cases. Signed-off-by:
-
Mel Gorman authored
__dec_zone_state is cheaper to use for removing an order-0 page as it has fewer conditions to check. The performance difference on a page allocator microbenchmark is; 4.6.0-rc2 4.6.0-rc2 optiter-v1r20 decstat-v1r20 Min alloc-odr0-1 382.00 ( 0.00%) 381.00 ( 0.26%) Min alloc-odr0-2 282.00 ( 0.00%) 275.00 ( 2.48%) Min alloc-odr0-4 233.00 ( 0.00%) 229.00 ( 1.72%) Min alloc-odr0-8 203.00 ( 0.00%) 199.00 ( 1.97%) Min alloc-odr0-16 188.00 ( 0.00%) 186.00 ( 1.06%) Min alloc-odr0-32 182.00 ( 0.00%) 179.00 ( 1.65%) Min alloc-odr0-64 177.00 ( 0.00%) 174.00 ( 1.69%) Min alloc-odr0-128 175.00 ( 0.00%) 172.00 ( 1.71%) Min alloc-odr0-256 184.00 ( 0.00%) 181.00 ( 1.63%) Min alloc-odr0-512 197.00 ( 0.00%) 193.00 ( 2.03%) Min alloc-odr0-1024 203.00 ( 0.00%) 201.00 ( 0.99%) Min alloc-odr0-2048 209.00 ( 0.00%) 206.00 ( 1.44%) Min alloc-odr0-4096 214.00 ( 0.00%) 212.00 ( 0.93%) Min alloc-odr0-8192 218.00 ( 0.00%) 215.00 ( 1.38%) Min alloc-odr0-16384 219.00 ( 0.00%) 216.00 ( 1.37%) Signed-off-by: -
Mel Gorman authored
The page allocator iterates through a zonelist for zones that match the addressing limitations and nodemask of the caller but many allocations will not be restricted. Despite this, there is always functional call overhead which builds up. This patch inlines the optimistic basic case and only calls the iterator function for the complex case. A hindrance was the fact that cpuset_current_mems_allowed is used in the fastpath as the allowed nodemask even though all nodes are allowed on most systems. The patch handles this by only considering cpuset_current_mems_allowed if a cpuset exists. As well as being faster in the fast-path, this removes some junk in the slowpath. The performance difference on a page allocator microbenchmark is; 4.6.0-rc2 4.6.0-rc2 statinline-v1r20 optiter-v1r20 Min alloc-odr0-1 412.00 ( 0.00%) 382.00 ( 7.28%) Min alloc-odr0-2 301.00 ( 0.00%) 282.00 ( 6.31%) Min alloc-odr0-4 247.00 ( 0.00%) 233.00 ( 5.67%) Min alloc-odr0-8 215.00 ( 0.00%) 203.00 ( 5.58%) Min alloc-odr0-16 199.00 ( 0.00%) 188.00 ( 5.53%) Min alloc-odr0-32 191.00 ( 0.00%) 182.00 ( 4.71%) Min alloc-odr0-64 187.00 ( 0.00%) 177.00 ( 5.35%) Min alloc-odr0-128 185.00 ( 0.00%) 175.00 ( 5.41%) Min alloc-odr0-256 193.00 ( 0.00%) 184.00 ( 4.66%) Min alloc-odr0-512 207.00 ( 0.00%) 197.00 ( 4.83%) Min alloc-odr0-1024 213.00 ( 0.00%) 203.00 ( 4.69%) Min alloc-odr0-2048 220.00 ( 0.00%) 209.00 ( 5.00%) Min alloc-odr0-4096 226.00 ( 0.00%) 214.00 ( 5.31%) Min alloc-odr0-8192 229.00 ( 0.00%) 218.00 ( 4.80%) Min alloc-odr0-16384 229.00 ( 0.00%) 219.00 ( 4.37%) perf indicated that next_zones_zonelist disappeared in the profile and __next_zones_zonelist did not appear. This is expected as the micro-benchmark would hit the inlined fast-path every time. Signed-off-by: -
Mel Gorman authored
zone_statistics has one call-site but it's a public function. Make it static and inline. The performance difference on a page allocator microbenchmark is; 4.6.0-rc2 4.6.0-rc2 statbranch-v1r20 statinline-v1r20 Min alloc-odr0-1 419.00 ( 0.00%) 412.00 ( 1.67%) Min alloc-odr0-2 305.00 ( 0.00%) 301.00 ( 1.31%) Min alloc-odr0-4 250.00 ( 0.00%) 247.00 ( 1.20%) Min alloc-odr0-8 219.00 ( 0.00%) 215.00 ( 1.83%) Min alloc-odr0-16 203.00 ( 0.00%) 199.00 ( 1.97%) Min alloc-odr0-32 195.00 ( 0.00%) 191.00 ( 2.05%) Min alloc-odr0-64 191.00 ( 0.00%) 187.00 ( 2.09%) Min alloc-odr0-128 189.00 ( 0.00%) 185.00 ( 2.12%) Min alloc-odr0-256 198.00 ( 0.00%) 193.00 ( 2.53%) Min alloc-odr0-512 210.00 ( 0.00%) 207.00 ( 1.43%) Min alloc-odr0-1024 216.00 ( 0.00%) 213.00 ( 1.39%) Min alloc-odr0-2048 221.00 ( 0.00%) 220.00 ( 0.45%) Min alloc-odr0-4096 227.00 ( 0.00%) 226.00 ( 0.44%) Min alloc-odr0-8192 232.00 ( 0.00%) 229.00 ( 1.29%) Min alloc-odr0-16384 232.00 ( 0.00%) 229.00 ( 1.29%) Signed-off-by: -
Mel Gorman authored
zone_statistics has more branches than it really needs to take an unlikely GFP flag into account. Reduce the number and annotate the unlikely flag. The performance difference on a page allocator microbenchmark is; 4.6.0-rc2 4.6.0-rc2 nocompound-v1r10 statbranch-v1r10 Min alloc-odr0-1 417.00 ( 0.00%) 419.00 ( -0.48%) Min alloc-odr0-2 308.00 ( 0.00%) 305.00 ( 0.97%) Min alloc-odr0-4 253.00 ( 0.00%) 250.00 ( 1.19%) Min alloc-odr0-8 221.00 ( 0.00%) 219.00 ( 0.90%) Min alloc-odr0-16 205.00 ( 0.00%) 203.00 ( 0.98%) Min alloc-odr0-32 199.00 ( 0.00%) 195.00 ( 2.01%) Min alloc-odr0-64 193.00 ( 0.00%) 191.00 ( 1.04%) Min alloc-odr0-128 191.00 ( 0.00%) 189.00 ( 1.05%) Min alloc-odr0-256 200.00 ( 0.00%) 198.00 ( 1.00%) Min alloc-odr0-512 212.00 ( 0.00%) 210.00 ( 0.94%) Min alloc-odr0-1024 219.00 ( 0.00%) 216.00 ( 1.37%) Min alloc-odr0-2048 225.00 ( 0.00%) 221.00 ( 1.78%) Min alloc-odr0-4096 231.00 ( 0.00%) 227.00 ( 1.73%) Min alloc-odr0-8192 234.00 ( 0.00%) 232.00 ( 0.85%) Min alloc-odr0-16384 234.00 ( 0.00%) 232.00 ( 0.85%) Signed-off-by: -
Mel Gorman authored
The PageAnon check always checks for compound_head but this is a relatively expensive check if the caller already knows the page is a head page. This patch creates a helper and uses it in the page free path which only operates on head pages. With this patch and "Only check PageCompound for high-order pages", the performance difference on a page allocator microbenchmark is; 4.6.0-rc2 4.6.0-rc2 vanilla nocompound-v1r20 Min alloc-odr0-1 425.00 ( 0.00%) 417.00 ( 1.88%) Min alloc-odr0-2 313.00 ( 0.00%) 308.00 ( 1.60%) Min alloc-odr0-4 257.00 ( 0.00%) 253.00 ( 1.56%) Min alloc-odr0-8 224.00 ( 0.00%) 221.00 ( 1.34%) Min alloc-odr0-16 208.00 ( 0.00%) 205.00 ( 1.44%) Min alloc-odr0-32 199.00 ( 0.00%) 199.00 ( 0.00%) Min alloc-odr0-64 195.00 ( 0.00%) 193.00 ( 1.03%) Min alloc-odr0-128 192.00 ( 0.00%) 191.00 ( 0.52%) Min alloc-odr0-256 204.00 ( 0.00%) 200.00 ( 1.96%) Min alloc-odr0-512 213.00 ( 0.00%) 212.00 ( 0.47%) Min alloc-odr0-1024 219.00 ( 0.00%) 219.00 ( 0.00%) Min alloc-odr0-2048 225.00 ( 0.00%) 225.00 ( 0.00%) Min alloc-odr0-4096 230.00 ( 0.00%) 231.00 ( -0.43%) Min alloc-odr0-8192 235.00 ( 0.00%) 234.00 ( 0.43%) Min alloc-odr0-16384 235.00 ( 0.00%) 234.00 ( 0.43%) Min free-odr0-1 215.00 ( 0.00%) 191.00 ( 11.16%) Min free-odr0-2 152.00 ( 0.00%) 136.00 ( 10.53%) Min free-odr0-4 119.00 ( 0.00%) 107.00 ( 10.08%) Min free-odr0-8 106.00 ( 0.00%) 96.00 ( 9.43%) Min free-odr0-16 97.00 ( 0.00%) 87.00 ( 10.31%) Min free-odr0-32 91.00 ( 0.00%) 83.00 ( 8.79%) Min free-odr0-64 89.00 ( 0.00%) 81.00 ( 8.99%) Min free-odr0-128 88.00 ( 0.00%) 80.00 ( 9.09%) Min free-odr0-256 106.00 ( 0.00%) 95.00 ( 10.38%) Min free-odr0-512 116.00 ( 0.00%) 111.00 ( 4.31%) Min free-odr0-1024 125.00 ( 0.00%) 118.00 ( 5.60%) Min free-odr0-2048 133.00 ( 0.00%) 126.00 ( 5.26%) Min free-odr0-4096 136.00 ( 0.00%) 130.00 ( 4.41%) Min free-odr0-8192 138.00 ( 0.00%) 130.00 ( 5.80%) Min free-odr0-16384 137.00 ( 0.00%) 130.00 ( 5.11%) There is a sizable boost to the free allocator performance. While there is an apparent boost on the allocation side, it's likely a co-incidence or due to the patches slightly reducing cache footprint. Signed-off-by: -
Mel Gorman authored
Another year, another round of page allocator optimisations focusing this time on the alloc and free fast paths. This should be of help to workloads that are allocator-intensive from kernel space where the cost of zeroing is not nceessraily incurred. The series is motivated by the observation that page alloc microbenchmarks on multiple machines regressed between 3.12.44 and 4.4. Second, there is discussions before LSF/MM considering the possibility of adding another page allocator which is potentially hazardous but a patch series improving performance is better than whining. After the series is applied, there are still hazards. In the free paths, the debugging checking and page zone/pageblock lookups dominate but there was not an obvious solution to that. In the alloc path, the major contributers are dealing with zonelists, new page preperation, the fair zone allocation and numerous statistic updates. The fair zone allocator is removed by the per-node LRU series if that gets merged so it's nor a major concern at the moment. On normal userspace benchmarks, there is little impact as the zeroing cost is significant but it's visible aim9 4.6.0-rc3 4.6.0-rc3 vanilla deferalloc-v3 Min page_test 828693.33 ( 0.00%) 887060.00 ( 7.04%) Min brk_test 4847266.67 ( 0.00%) 4966266.67 ( 2.45%) Min exec_test 1271.00 ( 0.00%) 1275.67 ( 0.37%) Min fork_test 12371.75 ( 0.00%) 12380.00 ( 0.07%) The overall impact on a page allocator microbenchmark for a range of orders and number of pages allocated in a batch is 4.6.0-rc3 4.6.0-rc3 vanilla deferalloc-v3r7 Min alloc-odr0-1 428.00 ( 0.00%) 316.00 ( 26.17%) Min alloc-odr0-2 314.00 ( 0.00%) 231.00 ( 26.43%) Min alloc-odr0-4 256.00 ( 0.00%) 192.00 ( 25.00%) Min alloc-odr0-8 222.00 ( 0.00%) 166.00 ( 25.23%) Min alloc-odr0-16 207.00 ( 0.00%) 154.00 ( 25.60%) Min alloc-odr0-32 197.00 ( 0.00%) 148.00 ( 24.87%) Min alloc-odr0-64 193.00 ( 0.00%) 144.00 ( 25.39%) Min alloc-odr0-128 191.00 ( 0.00%) 143.00 ( 25.13%) Min alloc-odr0-256 203.00 ( 0.00%) 153.00 ( 24.63%) Min alloc-odr0-512 212.00 ( 0.00%) 165.00 ( 22.17%) Min alloc-odr0-1024 221.00 ( 0.00%) 172.00 ( 22.17%) Min alloc-odr0-2048 225.00 ( 0.00%) 179.00 ( 20.44%) Min alloc-odr0-4096 232.00 ( 0.00%) 185.00 ( 20.26%) Min alloc-odr0-8192 235.00 ( 0.00%) 187.00 ( 20.43%) Min alloc-odr0-16384 236.00 ( 0.00%) 188.00 ( 20.34%) Min alloc-odr1-1 519.00 ( 0.00%) 450.00 ( 13.29%) Min alloc-odr1-2 391.00 ( 0.00%) 336.00 ( 14.07%) Min alloc-odr1-4 313.00 ( 0.00%) 268.00 ( 14.38%) Min alloc-odr1-8 277.00 ( 0.00%) 235.00 ( 15.16%) Min alloc-odr1-16 256.00 ( 0.00%) 218.00 ( 14.84%) Min alloc-odr1-32 252.00 ( 0.00%) 212.00 ( 15.87%) Min alloc-odr1-64 244.00 ( 0.00%) 206.00 ( 15.57%) Min alloc-odr1-128 244.00 ( 0.00%) 207.00 ( 15.16%) Min alloc-odr1-256 243.00 ( 0.00%) 207.00 ( 14.81%) Min alloc-odr1-512 245.00 ( 0.00%) 209.00 ( 14.69%) Min alloc-odr1-1024 248.00 ( 0.00%) 214.00 ( 13.71%) Min alloc-odr1-2048 253.00 ( 0.00%) 220.00 ( 13.04%) Min alloc-odr1-4096 258.00 ( 0.00%) 224.00 ( 13.18%) Min alloc-odr1-8192 261.00 ( 0.00%) 229.00 ( 12.26%) Min alloc-odr2-1 560.00 ( 0.00%) 753.00 (-34.46%) Min alloc-odr2-2 424.00 ( 0.00%) 351.00 ( 17.22%) Min alloc-odr2-4 339.00 ( 0.00%) 393.00 (-15.93%) Min alloc-odr2-8 298.00 ( 0.00%) 246.00 ( 17.45%) Min alloc-odr2-16 276.00 ( 0.00%) 227.00 ( 17.75%) Min alloc-odr2-32 271.00 ( 0.00%) 221.00 ( 18.45%) Min alloc-odr2-64 264.00 ( 0.00%) 217.00 ( 17.80%) Min alloc-odr2-128 264.00 ( 0.00%) 217.00 ( 17.80%) Min alloc-odr2-256 264.00 ( 0.00%) 218.00 ( 17.42%) Min alloc-odr2-512 269.00 ( 0.00%) 223.00 ( 17.10%) Min alloc-odr2-1024 279.00 ( 0.00%) 230.00 ( 17.56%) Min alloc-odr2-2048 283.00 ( 0.00%) 235.00 ( 16.96%) Min alloc-odr2-4096 285.00 ( 0.00%) 239.00 ( 16.14%) Min alloc-odr3-1 629.00 ( 0.00%) 505.00 ( 19.71%) Min alloc-odr3-2 472.00 ( 0.00%) 374.00 ( 20.76%) Min alloc-odr3-4 383.00 ( 0.00%) 301.00 ( 21.41%) Min alloc-odr3-8 341.00 ( 0.00%) 266.00 ( 21.99%) Min alloc-odr3-16 316.00 ( 0.00%) 248.00 ( 21.52%) Min alloc-odr3-32 308.00 ( 0.00%) 241.00 ( 21.75%) Min alloc-odr3-64 305.00 ( 0.00%) 241.00 ( 20.98%) Min alloc-odr3-128 308.00 ( 0.00%) 244.00 ( 20.78%) Min alloc-odr3-256 317.00 ( 0.00%) 249.00 ( 21.45%) Min alloc-odr3-512 327.00 ( 0.00%) 256.00 ( 21.71%) Min alloc-odr3-1024 331.00 ( 0.00%) 261.00 ( 21.15%) Min alloc-odr3-2048 333.00 ( 0.00%) 266.00 ( 20.12%) Min alloc-odr4-1 767.00 ( 0.00%) 572.00 ( 25.42%) Min alloc-odr4-2 578.00 ( 0.00%) 429.00 ( 25.78%) Min alloc-odr4-4 474.00 ( 0.00%) 346.00 ( 27.00%) Min alloc-odr4-8 422.00 ( 0.00%) 310.00 ( 26.54%) Min alloc-odr4-16 399.00 ( 0.00%) 295.00 ( 26.07%) Min alloc-odr4-32 392.00 ( 0.00%) 293.00 ( 25.26%) Min alloc-odr4-64 394.00 ( 0.00%) 293.00 ( 25.63%) Min alloc-odr4-128 405.00 ( 0.00%) 305.00 ( 24.69%) Min alloc-odr4-256 417.00 ( 0.00%) 319.00 ( 23.50%) Min alloc-odr4-512 425.00 ( 0.00%) 326.00 ( 23.29%) Min alloc-odr4-1024 426.00 ( 0.00%) 329.00 ( 22.77%) Min free-odr0-1 216.00 ( 0.00%) 178.00 ( 17.59%) Min free-odr0-2 152.00 ( 0.00%) 125.00 ( 17.76%) Min free-odr0-4 120.00 ( 0.00%) 99.00 ( 17.50%) Min free-odr0-8 106.00 ( 0.00%) 85.00 ( 19.81%) Min free-odr0-16 97.00 ( 0.00%) 80.00 ( 17.53%) Min free-odr0-32 92.00 ( 0.00%) 76.00 ( 17.39%) Min free-odr0-64 89.00 ( 0.00%) 74.00 ( 16.85%) Min free-odr0-128 89.00 ( 0.00%) 73.00 ( 17.98%) Min free-odr0-256 107.00 ( 0.00%) 90.00 ( 15.89%) Min free-odr0-512 117.00 ( 0.00%) 108.00 ( 7.69%) Min free-odr0-1024 125.00 ( 0.00%) 118.00 ( 5.60%) Min free-odr0-2048 132.00 ( 0.00%) 125.00 ( 5.30%) Min free-odr0-4096 135.00 ( 0.00%) 130.00 ( 3.70%) Min free-odr0-8192 137.00 ( 0.00%) 130.00 ( 5.11%) Min free-odr0-16384 137.00 ( 0.00%) 131.00 ( 4.38%) Min free-odr1-1 318.00 ( 0.00%) 289.00 ( 9.12%) Min free-odr1-2 228.00 ( 0.00%) 207.00 ( 9.21%) Min free-odr1-4 182.00 ( 0.00%) 165.00 ( 9.34%) Min free-odr1-8 163.00 ( 0.00%) 146.00 ( 10.43%) Min free-odr1-16 151.00 ( 0.00%) 135.00 ( 10.60%) Min free-odr1-32 146.00 ( 0.00%) 129.00 ( 11.64%) Min free-odr1-64 145.00 ( 0.00%) 130.00 ( 10.34%) Min free-odr1-128 148.00 ( 0.00%) 134.00 ( 9.46%) Min free-odr1-256 148.00 ( 0.00%) 137.00 ( 7.43%) Min free-odr1-512 151.00 ( 0.00%) 140.00 ( 7.28%) Min free-odr1-1024 154.00 ( 0.00%) 143.00 ( 7.14%) Min free-odr1-2048 156.00 ( 0.00%) 144.00 ( 7.69%) Min free-odr1-4096 156.00 ( 0.00%) 142.00 ( 8.97%) Min free-odr1-8192 156.00 ( 0.00%) 140.00 ( 10.26%) Min free-odr2-1 361.00 ( 0.00%) 457.00 (-26.59%) Min free-odr2-2 258.00 ( 0.00%) 224.00 ( 13.18%) Min free-odr2-4 208.00 ( 0.00%) 223.00 ( -7.21%) Min free-odr2-8 185.00 ( 0.00%) 160.00 ( 13.51%) Min free-odr2-16 173.00 ( 0.00%) 149.00 ( 13.87%) Min free-odr2-32 166.00 ( 0.00%) 145.00 ( 12.65%) Min free-odr2-64 166.00 ( 0.00%) 146.00 ( 12.05%) Min free-odr2-128 169.00 ( 0.00%) 148.00 ( 12.43%) Min free-odr2-256 170.00 ( 0.00%) 152.00 ( 10.59%) Min free-odr2-512 177.00 ( 0.00%) 156.00 ( 11.86%) Min free-odr2-1024 182.00 ( 0.00%) 162.00 ( 10.99%) Min free-odr2-2048 181.00 ( 0.00%) 160.00 ( 11.60%) Min free-odr2-4096 180.00 ( 0.00%) 159.00 ( 11.67%) Min free-odr3-1 431.00 ( 0.00%) 367.00 ( 14.85%) Min free-odr3-2 306.00 ( 0.00%) 259.00 ( 15.36%) Min free-odr3-4 249.00 ( 0.00%) 208.00 ( 16.47%) Min free-odr3-8 224.00 ( 0.00%) 186.00 ( 16.96%) Min free-odr3-16 208.00 ( 0.00%) 176.00 ( 15.38%) Min free-odr3-32 206.00 ( 0.00%) 174.00 ( 15.53%) Min free-odr3-64 210.00 ( 0.00%) 178.00 ( 15.24%) Min free-odr3-128 215.00 ( 0.00%) 182.00 ( 15.35%) Min free-odr3-256 224.00 ( 0.00%) 189.00 ( 15.62%) Min free-odr3-512 232.00 ( 0.00%) 195.00 ( 15.95%) Min free-odr3-1024 230.00 ( 0.00%) 195.00 ( 15.22%) Min free-odr3-2048 229.00 ( 0.00%) 193.00 ( 15.72%) Min free-odr4-1 561.00 ( 0.00%) 439.00 ( 21.75%) Min free-odr4-2 418.00 ( 0.00%) 318.00 ( 23.92%) Min free-odr4-4 339.00 ( 0.00%) 269.00 ( 20.65%) Min free-odr4-8 299.00 ( 0.00%) 239.00 ( 20.07%) Min free-odr4-16 289.00 ( 0.00%) 234.00 ( 19.03%) Min free-odr4-32 291.00 ( 0.00%) 235.00 ( 19.24%) Min free-odr4-64 298.00 ( 0.00%) 238.00 ( 20.13%) Min free-odr4-128 308.00 ( 0.00%) 251.00 ( 18.51%) Min free-odr4-256 321.00 ( 0.00%) 267.00 ( 16.82%) Min free-odr4-512 327.00 ( 0.00%) 269.00 ( 17.74%) Min free-odr4-1024 326.00 ( 0.00%) 271.00 ( 16.87%) Min total-odr0-1 644.00 ( 0.00%) 494.00 ( 23.29%) Min total-odr0-2 466.00 ( 0.00%) 356.00 ( 23.61%) Min total-odr0-4 376.00 ( 0.00%) 291.00 ( 22.61%) Min total-odr0-8 328.00 ( 0.00%) 251.00 ( 23.48%) Min total-odr0-16 304.00 ( 0.00%) 234.00 ( 23.03%) Min total-odr0-32 289.00 ( 0.00%) 224.00 ( 22.49%) Min total-odr0-64 282.00 ( 0.00%) 218.00 ( 22.70%) Min total-odr0-128 280.00 ( 0.00%) 216.00 ( 22.86%) Min total-odr0-256 310.00 ( 0.00%) 243.00 ( 21.61%) Min total-odr0-512 329.00 ( 0.00%) 273.00 ( 17.02%) Min total-odr0-1024 346.00 ( 0.00%) 290.00 ( 16.18%) Min total-odr0-2048 357.00 ( 0.00%) 304.00 ( 14.85%) Min total-odr0-4096 367.00 ( 0.00%) 315.00 ( 14.17%) Min total-odr0-8192 372.00 ( 0.00%) 317.00 ( 14.78%) Min total-odr0-16384 373.00 ( 0.00%) 319.00 ( 14.48%) Min total-odr1-1 838.00 ( 0.00%) 739.00 ( 11.81%) Min total-odr1-2 619.00 ( 0.00%) 543.00 ( 12.28%) Min total-odr1-4 495.00 ( 0.00%) 433.00 ( 12.53%) Min total-odr1-8 440.00 ( 0.00%) 382.00 ( 13.18%) Min total-odr1-16 407.00 ( 0.00%) 353.00 ( 13.27%) Min total-odr1-32 398.00 ( 0.00%) 341.00 ( 14.32%) Min total-odr1-64 389.00 ( 0.00%) 336.00 ( 13.62%) Min total-odr1-128 392.00 ( 0.00%) 341.00 ( 13.01%) Min total-odr1-256 391.00 ( 0.00%) 344.00 ( 12.02%) Min total-odr1-512 396.00 ( 0.00%) 349.00 ( 11.87%) Min total-odr1-1024 402.00 ( 0.00%) 357.00 ( 11.19%) Min total-odr1-2048 409.00 ( 0.00%) 364.00 ( 11.00%) Min total-odr1-4096 414.00 ( 0.00%) 366.00 ( 11.59%) Min total-odr1-8192 417.00 ( 0.00%) 369.00 ( 11.51%) Min total-odr2-1 921.00 ( 0.00%) 1210.00 (-31.38%) Min total-odr2-2 682.00 ( 0.00%) 576.00 ( 15.54%) Min total-odr2-4 547.00 ( 0.00%) 616.00 (-12.61%) Min total-odr2-8 483.00 ( 0.00%) 406.00 ( 15.94%) Min total-odr2-16 449.00 ( 0.00%) 376.00 ( 16.26%) Min total-odr2-32 437.00 ( 0.00%) 366.00 ( 16.25%) Min total-odr2-64 431.00 ( 0.00%) 363.00 ( 15.78%) Min total-odr2-128 433.00 ( 0.00%) 365.00 ( 15.70%) Min total-odr2-256 434.00 ( 0.00%) 371.00 ( 14.52%) Min total-odr2-512 446.00 ( 0.00%) 379.00 ( 15.02%) Min total-odr2-1024 461.00 ( 0.00%) 392.00 ( 14.97%) Min total-odr2-2048 464.00 ( 0.00%) 395.00 ( 14.87%) Min total-odr2-4096 465.00 ( 0.00%) 398.00 ( 14.41%) Min total-odr3-1 1060.00 ( 0.00%) 872.00 ( 17.74%) Min total-odr3-2 778.00 ( 0.00%) 633.00 ( 18.64%) Min total-odr3-4 632.00 ( 0.00%) 510.00 ( 19.30%) Min total-odr3-8 565.00 ( 0.00%) 452.00 ( 20.00%) Min total-odr3-16 524.00 ( 0.00%) 424.00 ( 19.08%) Min total-odr3-32 514.00 ( 0.00%) 415.00 ( 19.26%) Min total-odr3-64 515.00 ( 0.00%) 419.00 ( 18.64%) Min total-odr3-128 523.00 ( 0.00%) 426.00 ( 18.55%) Min total-odr3-256 541.00 ( 0.00%) 438.00 ( 19.04%) Min total-odr3-512 559.00 ( 0.00%) 451.00 ( 19.32%) Min total-odr3-1024 561.00 ( 0.00%) 456.00 ( 18.72%) Min total-odr3-2048 562.00 ( 0.00%) 459.00 ( 18.33%) Min total-odr4-1 1328.00 ( 0.00%) 1011.00 ( 23.87%) Min total-odr4-2 997.00 ( 0.00%) 747.00 ( 25.08%) Min total-odr4-4 813.00 ( 0.00%) 615.00 ( 24.35%) Min total-odr4-8 721.00 ( 0.00%) 550.00 ( 23.72%) Min total-odr4-16 689.00 ( 0.00%) 529.00 ( 23.22%) Min total-odr4-32 683.00 ( 0.00%) 528.00 ( 22.69%) Min total-odr4-64 692.00 ( 0.00%) 531.00 ( 23.27%) Min total-odr4-128 713.00 ( 0.00%) 556.00 ( 22.02%) Min total-odr4-256 738.00 ( 0.00%) 586.00 ( 20.60%) Min total-odr4-512 753.00 ( 0.00%) 595.00 ( 20.98%) Min total-odr4-1024 752.00 ( 0.00%) 600.00 ( 20.21%) This patch (of 27): order-0 pages by definition cannot be compound so avoid the check in the fast path for those pages. [akpm@linux-foundation.org: use unlikely(order) in free_pages_prepare(), per Vlastimil] Signed-off-by: -
Michal Hocko authored
Right now the oom reaper will clear TIF_MEMDIE only for tasks which were successfully reaped. This is the safest option because we know that such an oom victim would only block forward progress of the oom killer without a good reason because it is highly unlikely it would release much more memory. Basically most of its memory has been already torn down. We can relax this assumption to catch more corner cases though. The first obvious one is when the oom victim clears its mm and gets stuck later on. oom_reaper would back of on find_lock_task_mm returning NULL. We can safely try to clear TIF_MEMDIE in this case because such a task would be ignored by the oom killer anyway. The flag would be cleared by that time already most of the time anyway. The less obvious one is when the oom reaper fails due to mmap_sem contention. Even if we clear TIF_MEMDIE for this task then it is not very likely that we would select another task too easily because we haven't reaped the last victim and so it would be still the #1 candidate. There is a rare race condition possible when the current victim terminates before the next select_bad_process but considering that oom_reap_task had retried several times before giving up then this sounds like a borderline thing. After this patch we should have a guarantee that the OOM killer will not be block for unbounded amount of time for most cases. Signed-off-by:
Michal Hocko <mhocko@suse.com> Cc: Raushaniya Maksudova <rmaksudova@parallels.com> Cc: Michael S. Tsirkin <mst@redhat.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: David Rientjes <rientjes@google.com> Cc: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp> Cc: Daniel Vetter <daniel.vetter@intel.com> Cc: Oleg Nesterov <oleg@redhat.com> Signed-off-by:
-
Michal Hocko authored
If either the current task is already killed or PF_EXITING or a selected task is PF_EXITING then the oom killer is suppressed and so is the oom reaper. This patch adds try_oom_reaper which checks the given task and queues it for the oom reaper if that is safe to be done meaning that the task doesn't share the mm with an alive process. This might help to release the memory pressure while the task tries to exit. [akpm@linux-foundation.org: fix nommu build] Signed-off-by:
-
Michal Hocko authored
__alloc_pages_may_oom is the central place to decide when the out_of_memory should be invoked. This is a good approach for most checks there because they are page allocator specific and the allocation fails right after for all of them. The notable exception is GFP_NOFS context which is faking did_some_progress and keep the page allocator looping even though there couldn't have been any progress from the OOM killer. This patch doesn't change this behavior because we are not ready to allow those allocation requests to fail yet (and maybe we will face the reality that we will never manage to safely fail these request). Instead __GFP_FS check is moved down to out_of_memory and prevent from OOM victim selection there. There are two reasons for that - OOM notifiers might release some memory even from this context as none of the registered notifier seems to be FS related - this might help a dying thread to get an access to memory reserves and move on which will make the behavior more consistent with the case when the task gets killed from a different context. Keep a comment in __alloc_pages_may_oom to make sure we do not forget how GFP_NOFS is special and that we really want to do something about it. Note to the current oom_notifier users: The observable difference for you is that oom notifiers cannot depend on any fs locks because we could deadlock. Not that this would be allowed today because that would just lockup machine in most of the cases and ruling out the OOM killer along the way. Another difference is that callbacks might be invoked sooner now because GFP_NOFS is a weaker reclaim context and so there could be reclaimable memory which is just not reachable now. That would require GFP_NOFS only loads which are really rare and more importantly the observable result would be dropping of reconstructible object and potential performance drop which is not such a big deal when we are struggling to fulfill other important allocation requests. Signed-off-by: -
Vitaly Kuznetsov authored
CONFIG_MEMORY_HOTPLUG_DEFAULT_ONLINE specifies the default value for the memory hotplug onlining policy. Add a command line parameter to make it possible to override the default. It may come handy for debug and testing purposes. Signed-off-by:
Vitaly Kuznetsov <vkuznets@redhat.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Dan Williams <dan.j.williams@intel.com> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: David Vrabel <david.vrabel@citrix.com> Cc: David Rientjes <rientjes@google.com> Cc: Igor Mammedov <imammedo@redhat.com> Cc: Lennart Poettering <lennart@poettering.net> Signed-off-by:
-
Vitaly Kuznetsov authored
This patchset continues the work I started with commit 31bc3858 ("memory-hotplug: add automatic onlining policy for the newly added memory"). Initially I was going to stop there and bring the policy setting logic to userspace. I met two issues on this way: 1) It is possible to have memory hotplugged at boot (e.g. with QEMU). These blocks stay offlined if we turn the onlining policy on by userspace. 2) My attempt to bring this policy setting to systemd failed, systemd maintainers suggest to change the default in kernel or ... to use tmpfiles.d to alter the policy (which looks like a hack to me): https://github.com/systemd/systemd/pull/2938 Here I suggest to add a config option to set the default value for the policy and a kernel command line parameter to make the override. This patch (of 2): Introduce config option to set the default value for memory hotplug onlining policy (/sys/devices/system/memory/auto_online_blocks). The reason one would want to turn this option on are to have early onlining for hotpluggable memory available at boot and to not require any userspace actions to make memory hotplug work. [akpm@linux-foundation.org: tweak Kconfig text] Signed-off-by:
-
Hugh Dickins authored
I've just discovered that the useful-sounding has_transparent_hugepage() is actually an architecture-dependent minefield: on some arches it only builds if CONFIG_TRANSPARENT_HUGEPAGE=y, on others it's also there when not, but on some of those (arm and arm64) it then gives the wrong answer; and on mips alone it's marked __init, which would crash if called later (but so far it has not been called later). Straighten this out: make it available to all configs, with a sensible default in asm-generic/pgtable.h, removing its definitions from those arches (arc, arm, arm64, sparc, tile) which are served by the default, adding #define has_transparent_hugepage has_transparent_hugepage to those (mips, powerpc, s390, x86) which need to override the default at runtime, and removing the __init from mips (but maybe that kind of code should be avoided after init: set a static variable the first time it's called). Signed-off-by:
Hugh Dickins <hughd@google.com> Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Cc: Andres Lagar-Cavilla <andreslc@google.com> Cc: Yang Shi <yang.shi@linaro.org> Cc: Ning Qu <quning@gmail.com> Cc: Mel Gorman <mgorman@techsingularity.net> Cc: Konstantin Khlebnikov <koct9i@gmail.com> Acked-by:
David S. Miller <davem@davemloft.net> Acked-by: Vineet Gupta <vgupta@synopsys.com> [arch/arc] Acked-by: Gerald Schaefer <gerald.schaefer@de.ibm.com> [arch/s390] Acked-by:
Ingo Molnar <mingo@kernel.org> Signed-off-by:
-
Hugh Dickins authored
Whatever huge pagecache implementation we go with, file rmap locking must be added to anon rmap locking, when mremap's move_page_tables() finds a pmd_trans_huge pmd entry: a simple change, let's do it now. Factor out take_rmap_locks() and drop_rmap_locks() to handle the locking for make move_ptes() and move_page_tables(), and delete the VM_BUG_ON_VMA which rejected vm_file and required anon_vma. Signed-off-by:
-
Hugh Dickins authored
Remove move_huge_pmd()'s redundant new_vma arg: all it was used for was a VM_NOHUGEPAGE check on new_vma flags, but the new_vma is cloned from the old vma, so a trans_huge_pmd in the new_vma will be as acceptable as it was in the old vma, alignment and size permitting. Signed-off-by:
-
Hugh Dickins authored
Provide /proc/sys/vm/stat_refresh to force an immediate update of per-cpu into global vmstats: useful to avoid a sleep(2) or whatever before checking counts when testing. Originally added to work around a bug which left counts stranded indefinitely on a cpu going idle (an inaccuracy magnified when small below-batch numbers represent "huge" amounts of memory), but I believe that bug is now fixed: nonetheless, this is still a useful knob. Its schedule_on_each_cpu() is probably too expensive just to fold into reading /proc/meminfo itself: give this mode 0600 to prevent abuse. Allow a write or a read to do the same: nothing to read, but "grep -h Shmem /proc/sys/vm/stat_refresh /proc/meminfo" is convenient. Oh, and since global_page_state() itself is careful to disguise any underflow as 0, hack in an "Invalid argument" and pr_warn() if a counter is negative after the refresh - this helped to fix a misaccounting of NR_ISOLATED_FILE in my migration code. But on recent kernels, I find that NR_ALLOC_BATCH and NR_PAGES_SCANNED often go negative some of the time. I have not yet worked out why, but have no evidence that it's actually harmful. Punt for the moment by just ignoring the anomaly on those. Signed-off-by:
-
Andres Lagar-Cavilla authored
Although shmem_fault() has been careful to count a major fault to vm_mm, shmem_getpage_gfp() has been careless in charging a remote access fault to current->mm owner's memcg instead of to vma->vm_mm owner's memcg: that is inconsistent with all the mem_cgroup charging on remote access faults in mm/memory.c. Fix it by passing fault_mm along with fault_type to shmem_get_page_gfp(); but in that case, now knowing the right mm, it's better for it to handle the PGMAJFAULT updates itself. And let's keep this clutter out of most callers' way: change the common shmem_getpage() wrapper to hide fault_mm and fault_type as well as gfp. Signed-off-by:
Andres Lagar-Cavilla <andreslc@google.com> Signed-off-by:
-
Hugh Dickins authored
Make a few cleanups in mm/shmem.c, before going on to complicate it. shmem_alloc_page() will become more complicated: we can't afford to to have that complication duplicated between a CONFIG_NUMA version and a !CONFIG_NUMA version, so rearrange the #ifdef'ery there to yield a single shmem_swapin() and a single shmem_alloc_page(). Yes, it's a shame to inflict the horrid pseudo-vma on non-NUMA configurations, but eliminating it is a larger cleanup: I have an alloc_pages_mpol() patchset not yet ready - mpol handling is subtle and bug-prone, and changed yet again since my last version. Move __SetPageLocked, __SetPageSwapBacked from shmem_getpage_gfp() to shmem_alloc_page(): that SwapBacked flag will be useful in future, to help to distinguish different cases appropriately. And the SGP_DIRTY variant of SGP_CACHE is hard to understand and of little use (IIRC it dates back to when shmem_getpage() returned the page unlocked): kill it and do the necessary in shmem_file_read_iter(). But an arm64 build then complained that info may be uninitialized (where shmem_getpage_gfp() deletes a freshly alloced page beyond eof), and advancing to an "sgp <= SGP_CACHE" test jogged it back to reality. Signed-off-by:
-
Hugh Dickins authored
v3.16 commit 07a42788 ("mm: shmem: avoid atomic operation during shmem_getpage_gfp") rightly replaced one instance of SetPageSwapBacked by __SetPageSwapBacked, pointing out that the newly allocated page is not yet visible to other users (except speculative get_page_unless_zero- ers, who may not update page flags before their further checks). That was part of a series in which Mel was focused on tmpfs profiles: but almost all SetPageSwapBacked uses can be so optimized, with the same justification. Remove ClearPageSwapBacked from __read_swap_cache_async() error path: it's not an error to free a page with PG_swapbacked set. Follow a convention of __SetPageLocked, __SetPageSwapBacked instead of doing it differently in different places; but that's for tidiness - if the ordering actually mattered, we should not be using the __variants. There's probably scope for further __SetPageFlags in other places, but SwapBacked is the one I'm interested in at the moment. Signed-off-by:
-
Hugh Dickins authored
Konstantin Khlebnikov pointed out (nearly four years ago, when lumpy reclaim was removed) that lru_size can be updated by -nr_taken once per call to isolate_lru_pages(), instead of page by page. Update it inside isolate_lru_pages(), or at its two callsites? I chose to update it at the callsites, rearranging and grouping the updates by nr_taken and nr_scanned together in both. With one exception, mem_cgroup_update_lru_size(,lru,) is then used where __mod_zone_page_state(,NR_LRU_BASE+lru,) is used; and we shall be adding some more calls in a future commit. Make the code a little smaller and simpler by incorporating stat update in lru_size update. The exception was move_active_pages_to_lru(), which aggregated the pgmoved stat update separately from the individual lru_size updates; but I still think this a simplification worth making. However, the __mod_zone_page_state is not peculiar to mem_cgroups: so better use the name update_lru_size, calls mem_cgroup_update_lru_size when CONFIG_MEMCG. Signed-off-by:
-
Hugh Dickins authored
Though debug kernels have a VM_BUG_ON to help protect from misaccounting lru_size, non-debug kernels are liable to wrap it around: and then the vast unsigned long size draws page reclaim into a loop of repeatedly doing nothing on an empty list, without even a cond_resched(). That soft lockup looks confusingly like an over-busy reclaim scenario, with lots of contention on the lru_lock in shrink_inactive_list(): yet has a totally different origin. Help differentiate with a custom warning in mem_cgroup_update_lru_size(), even in non-debug kernels; and reset the size to avoid the lockup. But the particular bug which suggested this change was mine alone, and since fixed. Make it a WARN_ONCE: the first occurrence is the most informative, a flurry may follow, yet even when rate-limited little more is learnt. Signed-off-by:
-
Konstantin Khlebnikov authored
Nobody uses it. Signed-off-by:
Konstantin Khlebnikov <koct9i@gmail.com> Signed-off-by:
-
Joonsoo Kim authored
node_page_state() manually adds statistics per each zone and returns total value for all zones. Whenever we add a new zone, we need to consider this function and it's really troublesome. Make it handle all zones by itself. Signed-off-by:
Joonsoo Kim <iamjoonsoo.kim@lge.com> Reviewed-by:
Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com> Cc: Rik van Riel <riel@redhat.com> Cc: Johannes Weiner <hannes@cmpxchg.org> Cc: Mel Gorman <mgorman@suse.de> Cc: Laura Abbott <lauraa@codeaurora.org> Cc: Minchan Kim <minchan@kernel.org> Cc: Marek Szyprowski <m.szyprowski@samsung.com> Cc: Michal Nazarewicz <mina86@mina86.com> Cc: Vlastimil Babka <vbabka@suse.cz> Signed-off-by:
-
Joonsoo Kim authored
nr_free_highpages() manually adds statistics per each highmem zone and returns a total value for them. Whenever we add a new highmem zone, we need to consider this function and it's really troublesome. Make it handle all highmem zones by itself. Signed-off-by:
-
Joonsoo Kim authored
ZONE_MOVABLE could be treated as highmem so we need to consider it for accurate statistics. And, in following patches, ZONE_CMA will be introduced and it can be treated as highmem, too. So, instead of manually adding stat of ZONE_MOVABLE, looping all zones and check whether the zone is highmem or not and add stat of the zone which can be treated as highmem. Signed-off-by:
-
Joonsoo Kim authored
ZONE_MOVABLE could be treated as highmem so we need to consider it for accurate calculation of dirty pages. And, in following patches, ZONE_CMA will be introduced and it can be treated as highmem, too. So, instead of manually adding stat of ZONE_MOVABLE, looping all zones and check whether the zone is highmem or not and add stat of the zone which can be treated as highmem. Signed-off-by:
-
Joonsoo Kim authored
There is a system thats node's pfns are overlapped as follows: -----pfn--------> N0 N1 N2 N0 N1 N2 Therefore, we need to care this overlapping when iterating pfn range. mark_free_pages() iterates requested zone's pfn range and unset all range's bitmap first. And then it marks freepages in a zone to the bitmap. If there is an overlapping zone, above unset could clear previous marked bit and reference to this bitmap in the future will cause the problem. To prevent it, this patch adds a zone check in mark_free_pages(). Signed-off-by:
-
Joonsoo Kim authored
There is a system thats node's pfns are overlapped as follows: -----pfn--------> N0 N1 N2 N0 N1 N2 Therefore, we need to care this overlapping when iterating pfn range. There are one place in page_owner.c that iterates pfn range and it doesn't consider this overlapping. Add it. Without this patch, above system could over count early allocated page number before page_owner is activated. Signed-off-by:
-
Joonsoo Kim authored
There is a system thats node's pfns are overlapped as follows: -----pfn--------> N0 N1 N2 N0 N1 N2 Therefore, we need to care this overlapping when iterating pfn range. There are two places in vmstat.c that iterates pfn range and they don't consider this overlapping. Add it. Without this patch, above system could over count pageblock number on a zone. Signed-off-by:
-
Joonsoo Kim authored
__offline_isolated_pages() and test_pages_isolated() are used by memory hotplug. These functions require that range is in a single zone but there is no code to do this because memory hotplug checks it before calling these functions. To avoid confusing future user of these functions, this patch adds comments to them. Signed-off-by:
-
Joonsoo Kim authored
This patchset deals with some problematic sites that iterate pfn ranges. There is a system thats node's pfns are overlapped as follows: -----pfn--------> N0 N1 N2 N0 N1 N2 Therefore, we need to take care of this overlapping when iterating pfn range. I audit many iterating sites that uses pfn_valid(), pfn_valid_within(), zone_start_pfn and etc. and others looks safe to me. This is a preparation step for a new CMA implementation, ZONE_CMA (https://lkml.org/lkml/2015/2/12/95), because it would be easily overlapped with other zones. But, zone overlap check is also needed for the general case so I send it separately. This patch (of 5): alloc_gigantic_page() uses alloc_contig_range() and this requires that the requested range is in a single zone. To satisfy this requirement, add this check to pfn_range_valid_gigantic(). Signed-off-by:
-
Andrew Morton authored
It's huge. Uninlining it saves 206 bytes per callsite. Shaves 4924 bytes from the x86_64 allmodconfig vmlinux. [akpm@linux-foundation.org: coding-style fixes] Cc: Steve Capper <steve.capper@arm.com> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Signed-off-by:
-
Chanho Min authored
is_highmem() can be simplified by use of is_highmem_idx(). This patch removes redundant code and will make it easier to maintain if the zone policy is changed or a new zone is added. (akpm: saves me 25 bytes of text per is_highmem() callsite) Signed-off-by:
Chanho Min <chanho.min@lge.com> Reviewed-by:
Dan Williams <dan.j.williams@intel.com> Cc: Michal Hocko <mhocko@kernel.org> Signed-off-by:
-
Vlastimil Babka authored
The goal of direct compaction is to quickly make a high-order page available for the pending allocation. Within an aligned block of pages of desired order, a single allocated page that cannot be isolated for migration means that the block cannot fully merge to a buddy page that would satisfy the allocation request. Therefore we can reduce the allocation stall by skipping the rest of the block immediately on isolation failure. For async compaction, this also means a higher chance of succeeding until it detects contention. We however shouldn't completely sacrifice the second objective of compaction, which is to reduce overal long-term memory fragmentation. As a compromise, perform the eager skipping only in direct async compaction, while sync compaction (including kcompactd) remains thorough. Testing was done using stress-highalloc from mmtests, configured for order-4 GFP_KERNEL allocations: 4.6-rc1 4.6-rc1 before after Success 1 Min 24.00 ( 0.00%) 27.00 (-12.50%) Success 1 Mean 30.20 ( 0.00%) 31.60 ( -4.64%) Success 1 Max 37.00 ( 0.00%) 35.00 ( 5.41%) Success 2 Min 42.00 ( 0.00%) 32.00 ( 23.81%) Success 2 Mean 44.00 ( 0.00%) 44.80 ( -1.82%) Success 2 Max 48.00 ( 0.00%) 52.00 ( -8.33%) Success 3 Min 91.00 ( 0.00%) 92.00 ( -1.10%) Success 3 Mean 92.20 ( 0.00%) 92.80 ( -0.65%) Success 3 Max 94.00 ( 0.00%) 93.00 ( 1.06%) We can see that success rates are unaffected by the skipping. 4.6-rc1 4.6-rc1 before after User 2587.42 2566.53 System 482.89 471.20 Elapsed 1395.68 1382.00 Times are not so useful metric for this benchmark as main portion is the interfering kernel builds, but results do hint at reduced system times. 4.6-rc1 4.6-rc1 before after Direct pages scanned 163614 159608 Kswapd pages scanned 2070139 2078790 Kswapd pages reclaimed 2061707 2069757 Direct pages reclaimed 163354 159505 Reduced direct reclaim was unintended, but could be explained by more successful first attempt at (async) direct compaction, which is attempted before the first reclaim attempt in __alloc_pages_slowpath(). Compaction stalls 33052 39853 Compaction success 12121 19773 Compaction failures 20931 20079 Compaction is indeed more successful, and thus less likely to get deferred, so there are also more direct compaction stalls. Page migrate success 3781876 3326819 Page migrate failure 45817 41774 Compaction pages isolated 7868232 6941457 Compaction migrate scanned 168160492 127269354 Compaction migrate prescanned 0 0 Compaction free scanned 2522142582 2326342620 Compaction free direct alloc 0 0 Compaction free dir. all. miss 0 0 Compaction cost 5252 4476 The patch reduces migration scanned pages by 25% thanks to the eager skipping. [hughd@google.com: prevent nr_isolated_* from going negative] Signed-off-by: -
Vlastimil Babka authored
Compaction drains the local pcplists each time migration scanner moves away from a cc->order aligned block where it isolated pages for migration, so that the pages freed by migrations can merge into higher orders. The detection is currently coarser than it could be. The cc->last_migrated_pfn variable should track the lowest pfn that was isolated for migration. But it is set to the pfn where isolate_migratepages_block() starts scanning, which is typically the first pfn of the pageblock. There, the scanner might fail to isolate several order-aligned blocks, and then isolate COMPACT_CLUSTER_MAX in another block. This would cause the pcplists drain to be performed, although the scanner didn't yet finish the block where it isolated from. This patch thus makes cc->last_migrated_pfn handling more accurate by setting it to the pfn of an actually isolated page in isolate_migratepages_block(). Although practical effects of this patch are likely low, it arguably makes the intent of the code more obvious. Also the next patch will make async direct compaction skip blocks more aggressively, and draining pcplists due to skipped blocks is wasteful. Signed-off-by:
-
Vlastimil Babka authored
Compaction code has accumulated numerous instances of manual calculations of the first (inclusive) and last (exclusive) pfn of a pageblock (or a smaller block of given order), given a pfn within the pageblock. Wrap these calculations by introducing pageblock_start_pfn(pfn) and pageblock_end_pfn(pfn) macros. [vbabka@suse.cz: fix crash in get_pfnblock_flags_mask() from isolate_freepages():] Signed-off-by:
-
