| Message ID | 20231016053002.756205-8-ying.huang@intel.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | mm: PCP high auto-tuning | expand |
Hello,
kernel test robot noticed a 8.4% improvement of will-it-scale.per_process_ops on:
commit: ba6149e96007edcdb01284c1531ebd49b4720f72 ("[PATCH -V3 7/9] mm: tune PCP high automatically")
url: https://github.com/intel-lab-lkp/linux/commits/Huang-Ying/mm-pcp-avoid-to-drain-PCP-when-process-exit/20231017-143633
base: https://git.kernel.org/cgit/linux/kernel/git/gregkh/driver-core.git 36b2d7dd5a8ac95c8c1e69bdc93c4a6e2dc28a23
patch link: https://lore.kernel.org/all/20231016053002.756205-8-ying.huang@intel.com/
patch subject: [PATCH -V3 7/9] mm: tune PCP high automatically
testcase: will-it-scale
test machine: 224 threads 4 sockets Intel(R) Xeon(R) Platinum 8380H CPU @ 2.90GHz (Cooper Lake) with 192G memory
parameters:
nr_task: 16
mode: process
test: page_fault2
cpufreq_governor: performance
Details are as below:
-------------------------------------------------------------------------------------------------->
The kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20231031/202310311001.edbc5817-oliver.sang@intel.com
=========================================================================================
compiler/cpufreq_governor/kconfig/mode/nr_task/rootfs/tbox_group/test/testcase:
gcc-12/performance/x86_64-rhel-8.3/process/16/debian-11.1-x86_64-20220510.cgz/lkp-cpl-4sp2/page_fault2/will-it-scale
commit:
9f9d0b0869 ("mm: add framework for PCP high auto-tuning")
ba6149e960 ("mm: tune PCP high automatically")
9f9d0b08696fb316 ba6149e96007edcdb01284c1531
---------------- ---------------------------
%stddev %change %stddev
\ | \
0.29 +0.0 0.32 mpstat.cpu.all.usr%
1434135 ± 2% +15.8% 1660688 ± 4% numa-meminfo.node0.AnonPages.max
22.97 +2.0% 23.43 turbostat.RAMWatt
213121 ± 5% -19.5% 171478 ± 7% meminfo.DirectMap4k
8031428 +12.0% 8998346 meminfo.Memused
9777522 +14.3% 11178004 meminfo.max_used_kB
4913700 +8.4% 5326025 will-it-scale.16.processes
307105 +8.4% 332876 will-it-scale.per_process_ops
4913700 +8.4% 5326025 will-it-scale.workload
1.488e+09 +8.5% 1.614e+09 proc-vmstat.numa_hit
1.487e+09 +8.4% 1.612e+09 proc-vmstat.numa_local
1.486e+09 +8.3% 1.609e+09 proc-vmstat.pgalloc_normal
1.482e+09 +8.3% 1.604e+09 proc-vmstat.pgfault
1.486e+09 +8.3% 1.609e+09 proc-vmstat.pgfree
2535424 ± 2% +6.2% 2693888 ± 2% proc-vmstat.unevictable_pgs_scanned
0.04 ± 9% +62.2% 0.06 ± 20% perf-sched.wait_and_delay.avg.ms.exit_to_user_mode_loop.exit_to_user_mode_prepare.irqentry_exit_to_user_mode.asm_exc_page_fault
85.33 ± 7% +36.1% 116.17 ± 8% perf-sched.wait_and_delay.count.exit_to_user_mode_loop.exit_to_user_mode_prepare.irqentry_exit_to_user_mode.asm_exc_page_fault
475.33 ± 3% +24.8% 593.33 ± 4% perf-sched.wait_and_delay.count.worker_thread.kthread.ret_from_fork.ret_from_fork_asm
0.16 ± 17% +449.1% 0.87 ± 39% perf-sched.wait_and_delay.max.ms.exit_to_user_mode_loop.exit_to_user_mode_prepare.irqentry_exit_to_user_mode.asm_exc_page_fault
0.03 ± 10% +94.1% 0.07 ± 26% perf-sched.wait_time.avg.ms.__cond_resched.__alloc_pages.__folio_alloc.vma_alloc_folio.do_cow_fault
0.04 ± 9% +62.2% 0.06 ± 20% perf-sched.wait_time.avg.ms.exit_to_user_mode_loop.exit_to_user_mode_prepare.irqentry_exit_to_user_mode.asm_exc_page_fault
0.16 ± 17% +449.1% 0.87 ± 39% perf-sched.wait_time.max.ms.exit_to_user_mode_loop.exit_to_user_mode_prepare.irqentry_exit_to_user_mode.asm_exc_page_fault
14.01 +6.0% 14.85 perf-stat.i.MPKI
5.79e+09 +3.6% 6.001e+09 perf-stat.i.branch-instructions
0.20 ± 2% +0.0 0.21 ± 2% perf-stat.i.branch-miss-rate%
12098037 ± 2% +8.5% 13122446 ± 2% perf-stat.i.branch-misses
82.90 +2.1 85.03 perf-stat.i.cache-miss-rate%
4.005e+08 +9.8% 4.399e+08 perf-stat.i.cache-misses
4.83e+08 +7.1% 5.174e+08 perf-stat.i.cache-references
2.29 -3.2% 2.22 perf-stat.i.cpi
164.08 -9.0% 149.33 perf-stat.i.cycles-between-cache-misses
7.091e+09 +4.2% 7.392e+09 perf-stat.i.dTLB-loads
0.97 +0.0 1.01 perf-stat.i.dTLB-store-miss-rate%
40301594 +8.8% 43829422 perf-stat.i.dTLB-store-misses
4.121e+09 +4.4% 4.302e+09 perf-stat.i.dTLB-stores
83.96 +2.6 86.59 perf-stat.i.iTLB-load-miss-rate%
10268085 ± 3% +23.0% 12628681 ± 3% perf-stat.i.iTLB-load-misses
2.861e+10 +3.7% 2.966e+10 perf-stat.i.instructions
2796 ± 3% -15.7% 2356 ± 3% perf-stat.i.instructions-per-iTLB-miss
0.44 +3.3% 0.45 perf-stat.i.ipc
984.67 +9.6% 1078 perf-stat.i.metric.K/sec
78.05 +4.2% 81.29 perf-stat.i.metric.M/sec
4913856 +8.4% 5329060 perf-stat.i.minor-faults
1.356e+08 +10.6% 1.499e+08 perf-stat.i.node-loads
32443508 +7.6% 34908277 perf-stat.i.node-stores
4913858 +8.4% 5329062 perf-stat.i.page-faults
14.00 +6.0% 14.83 perf-stat.overall.MPKI
0.21 ± 2% +0.0 0.22 ± 2% perf-stat.overall.branch-miss-rate%
82.92 +2.1 85.02 perf-stat.overall.cache-miss-rate%
2.29 -3.1% 2.21 perf-stat.overall.cpi
163.33 -8.6% 149.29 perf-stat.overall.cycles-between-cache-misses
0.97 +0.0 1.01 perf-stat.overall.dTLB-store-miss-rate%
84.00 +2.6 86.61 perf-stat.overall.iTLB-load-miss-rate%
2789 ± 3% -15.7% 2350 ± 3% perf-stat.overall.instructions-per-iTLB-miss
0.44 +3.2% 0.45 perf-stat.overall.ipc
1754985 -4.7% 1673375 perf-stat.overall.path-length
5.771e+09 +3.6% 5.981e+09 perf-stat.ps.branch-instructions
12074113 ± 2% +8.4% 13094204 ± 2% perf-stat.ps.branch-misses
3.992e+08 +9.8% 4.384e+08 perf-stat.ps.cache-misses
4.814e+08 +7.1% 5.157e+08 perf-stat.ps.cache-references
7.068e+09 +4.2% 7.367e+09 perf-stat.ps.dTLB-loads
40167519 +8.7% 43680173 perf-stat.ps.dTLB-store-misses
4.107e+09 +4.4% 4.288e+09 perf-stat.ps.dTLB-stores
10234325 ± 3% +23.0% 12587000 ± 3% perf-stat.ps.iTLB-load-misses
2.852e+10 +3.6% 2.956e+10 perf-stat.ps.instructions
4897507 +8.4% 5310921 perf-stat.ps.minor-faults
1.351e+08 +10.5% 1.494e+08 perf-stat.ps.node-loads
32335421 +7.6% 34789913 perf-stat.ps.node-stores
4897509 +8.4% 5310923 perf-stat.ps.page-faults
8.623e+12 +3.4% 8.912e+12 perf-stat.total.instructions
9.86 ± 3% -8.4 1.49 ± 5% perf-profile.calltrace.cycles-pp.rmqueue_bulk.__rmqueue_pcplist.rmqueue.get_page_from_freelist.__alloc_pages
8.11 ± 3% -7.5 0.58 ± 8% perf-profile.calltrace.cycles-pp._raw_spin_lock_irqsave.rmqueue_bulk.__rmqueue_pcplist.rmqueue.get_page_from_freelist
8.10 ± 3% -7.5 0.58 ± 8% perf-profile.calltrace.cycles-pp.native_queued_spin_lock_slowpath._raw_spin_lock_irqsave.rmqueue_bulk.__rmqueue_pcplist.rmqueue
7.52 ± 3% -6.4 1.15 ± 5% perf-profile.calltrace.cycles-pp.free_pcppages_bulk.free_unref_page_list.release_pages.tlb_batch_pages_flush.zap_pte_range
7.90 ± 4% -6.4 1.55 ± 4% perf-profile.calltrace.cycles-pp.free_unref_page_list.release_pages.tlb_batch_pages_flush.zap_pte_range.zap_pmd_range
5.78 ± 4% -5.8 0.00 perf-profile.calltrace.cycles-pp._raw_spin_lock_irqsave.free_pcppages_bulk.free_unref_page_list.release_pages.tlb_batch_pages_flush
5.78 ± 4% -5.8 0.00 perf-profile.calltrace.cycles-pp.native_queued_spin_lock_slowpath._raw_spin_lock_irqsave.free_pcppages_bulk.free_unref_page_list.release_pages
10.90 ± 3% -5.3 5.59 ± 2% perf-profile.calltrace.cycles-pp.get_page_from_freelist.__alloc_pages.__folio_alloc.vma_alloc_folio.do_cow_fault
10.57 ± 3% -5.3 5.26 ± 3% perf-profile.calltrace.cycles-pp.rmqueue.get_page_from_freelist.__alloc_pages.__folio_alloc.vma_alloc_folio
10.21 ± 3% -5.3 4.94 ± 3% perf-profile.calltrace.cycles-pp.__rmqueue_pcplist.rmqueue.get_page_from_freelist.__alloc_pages.__folio_alloc
11.18 ± 3% -5.3 5.91 ± 2% perf-profile.calltrace.cycles-pp.__folio_alloc.vma_alloc_folio.do_cow_fault.do_fault.__handle_mm_fault
11.15 ± 3% -5.3 5.88 ± 2% perf-profile.calltrace.cycles-pp.__alloc_pages.__folio_alloc.vma_alloc_folio.do_cow_fault.do_fault
11.56 ± 3% -5.2 6.37 ± 2% perf-profile.calltrace.cycles-pp.vma_alloc_folio.do_cow_fault.do_fault.__handle_mm_fault.handle_mm_fault
9.76 ± 3% -4.3 5.50 ± 6% perf-profile.calltrace.cycles-pp.release_pages.tlb_batch_pages_flush.zap_pte_range.zap_pmd_range.unmap_page_range
10.18 ± 3% -4.2 5.95 ± 5% perf-profile.calltrace.cycles-pp.tlb_batch_pages_flush.zap_pte_range.zap_pmd_range.unmap_page_range.unmap_vmas
15.40 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.__munmap
15.40 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.__munmap
15.40 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.__munmap
15.40 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.do_vmi_munmap.__vm_munmap.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe
15.40 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.do_vmi_align_munmap.do_vmi_munmap.__vm_munmap.__x64_sys_munmap.do_syscall_64
15.40 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe.__munmap
15.40 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.__vm_munmap.__x64_sys_munmap.do_syscall_64.entry_SYSCALL_64_after_hwframe.__munmap
15.39 ± 3% -3.7 11.70 perf-profile.calltrace.cycles-pp.unmap_region.do_vmi_align_munmap.do_vmi_munmap.__vm_munmap.__x64_sys_munmap
14.08 ± 3% -3.6 10.49 perf-profile.calltrace.cycles-pp.zap_pte_range.zap_pmd_range.unmap_page_range.unmap_vmas.unmap_region
14.10 ± 3% -3.6 10.52 perf-profile.calltrace.cycles-pp.unmap_vmas.unmap_region.do_vmi_align_munmap.do_vmi_munmap.__vm_munmap
14.10 ± 3% -3.6 10.52 perf-profile.calltrace.cycles-pp.unmap_page_range.unmap_vmas.unmap_region.do_vmi_align_munmap.do_vmi_munmap
14.10 ± 3% -3.6 10.52 perf-profile.calltrace.cycles-pp.zap_pmd_range.unmap_page_range.unmap_vmas.unmap_region.do_vmi_align_munmap
1.60 ± 2% -0.7 0.86 ± 6% perf-profile.calltrace.cycles-pp.__list_del_entry_valid_or_report.rmqueue_bulk.__rmqueue_pcplist.rmqueue.get_page_from_freelist
0.96 ± 3% -0.4 0.56 ± 3% perf-profile.calltrace.cycles-pp.free_pcppages_bulk.free_unref_page_list.release_pages.tlb_batch_pages_flush.tlb_finish_mmu
1.00 ± 4% -0.4 0.62 ± 4% perf-profile.calltrace.cycles-pp.free_unref_page_list.release_pages.tlb_batch_pages_flush.tlb_finish_mmu.unmap_region
1.26 ± 4% -0.1 1.11 ± 2% perf-profile.calltrace.cycles-pp.release_pages.tlb_batch_pages_flush.tlb_finish_mmu.unmap_region.do_vmi_align_munmap
1.28 ± 3% -0.1 1.16 ± 3% perf-profile.calltrace.cycles-pp.tlb_batch_pages_flush.tlb_finish_mmu.unmap_region.do_vmi_align_munmap.do_vmi_munmap
1.28 ± 4% -0.1 1.17 ± 2% perf-profile.calltrace.cycles-pp.tlb_finish_mmu.unmap_region.do_vmi_align_munmap.do_vmi_munmap.__vm_munmap
0.60 ± 3% -0.0 0.57 perf-profile.calltrace.cycles-pp.__mem_cgroup_charge.do_cow_fault.do_fault.__handle_mm_fault.handle_mm_fault
0.55 ± 3% +0.0 0.60 perf-profile.calltrace.cycles-pp.__perf_sw_event.do_user_addr_fault.exc_page_fault.asm_exc_page_fault.testcase
0.73 ± 3% +0.1 0.79 ± 2% perf-profile.calltrace.cycles-pp.lock_vma_under_rcu.do_user_addr_fault.exc_page_fault.asm_exc_page_fault.testcase
0.68 ± 3% +0.1 0.78 ± 3% perf-profile.calltrace.cycles-pp.page_remove_rmap.zap_pte_range.zap_pmd_range.unmap_page_range.unmap_vmas
0.57 ± 7% +0.1 0.71 ± 8% perf-profile.calltrace.cycles-pp.__mod_lruvec_page_state.folio_add_new_anon_rmap.set_pte_range.finish_fault.do_cow_fault
1.41 ± 3% +0.1 1.55 perf-profile.calltrace.cycles-pp.sync_regs.asm_exc_page_fault.testcase
0.77 ± 4% +0.2 0.93 ± 5% perf-profile.calltrace.cycles-pp.folio_add_new_anon_rmap.set_pte_range.finish_fault.do_cow_fault.do_fault
0.94 ± 3% +0.2 1.12 ± 3% perf-profile.calltrace.cycles-pp.lru_add_fn.folio_batch_move_lru.folio_add_lru_vma.set_pte_range.finish_fault
0.36 ± 70% +0.2 0.57 perf-profile.calltrace.cycles-pp.__perf_sw_event.handle_mm_fault.do_user_addr_fault.exc_page_fault.asm_exc_page_fault
1.26 ± 5% +0.2 1.47 ± 3% perf-profile.calltrace.cycles-pp.filemap_get_entry.shmem_get_folio_gfp.shmem_fault.__do_fault.do_cow_fault
1.61 ± 5% +0.3 1.87 ± 3% perf-profile.calltrace.cycles-pp.shmem_get_folio_gfp.shmem_fault.__do_fault.do_cow_fault.do_fault
1.75 ± 5% +0.3 2.05 ± 3% perf-profile.calltrace.cycles-pp.shmem_fault.__do_fault.do_cow_fault.do_fault.__handle_mm_fault
1.86 ± 4% +0.3 2.17 ± 2% perf-profile.calltrace.cycles-pp.__do_fault.do_cow_fault.do_fault.__handle_mm_fault.handle_mm_fault
0.17 ±141% +0.4 0.58 ± 3% perf-profile.calltrace.cycles-pp.xas_load.filemap_get_entry.shmem_get_folio_gfp.shmem_fault.__do_fault
2.60 ± 3% +0.5 3.14 ± 5% perf-profile.calltrace.cycles-pp._compound_head.zap_pte_range.zap_pmd_range.unmap_page_range.unmap_vmas
4.51 ± 3% +0.7 5.16 perf-profile.calltrace.cycles-pp._raw_spin_lock.__pte_offset_map_lock.finish_fault.do_cow_fault.do_fault
4.65 ± 3% +0.7 5.32 perf-profile.calltrace.cycles-pp.__pte_offset_map_lock.finish_fault.do_cow_fault.do_fault.__handle_mm_fault
1.61 ± 3% +1.9 3.52 ± 6% perf-profile.calltrace.cycles-pp.native_queued_spin_lock_slowpath._raw_spin_lock_irqsave.folio_lruvec_lock_irqsave.folio_batch_move_lru.folio_add_lru_vma
0.85 ± 2% +1.9 2.77 ± 13% perf-profile.calltrace.cycles-pp._raw_spin_lock_irqsave.folio_lruvec_lock_irqsave.release_pages.tlb_batch_pages_flush.zap_pte_range
0.84 ± 2% +1.9 2.76 ± 13% perf-profile.calltrace.cycles-pp.native_queued_spin_lock_slowpath._raw_spin_lock_irqsave.folio_lruvec_lock_irqsave.release_pages.tlb_batch_pages_flush
0.85 ± 2% +1.9 2.78 ± 12% perf-profile.calltrace.cycles-pp.folio_lruvec_lock_irqsave.release_pages.tlb_batch_pages_flush.zap_pte_range.zap_pmd_range
1.71 ± 3% +1.9 3.64 ± 6% perf-profile.calltrace.cycles-pp.folio_lruvec_lock_irqsave.folio_batch_move_lru.folio_add_lru_vma.set_pte_range.finish_fault
1.70 ± 2% +1.9 3.63 ± 6% perf-profile.calltrace.cycles-pp._raw_spin_lock_irqsave.folio_lruvec_lock_irqsave.folio_batch_move_lru.folio_add_lru_vma.set_pte_range
3.31 ± 2% +2.2 5.52 ± 5% perf-profile.calltrace.cycles-pp.folio_batch_move_lru.folio_add_lru_vma.set_pte_range.finish_fault.do_cow_fault
3.46 ± 2% +2.2 5.71 ± 5% perf-profile.calltrace.cycles-pp.folio_add_lru_vma.set_pte_range.finish_fault.do_cow_fault.do_fault
4.47 ± 2% +2.4 6.90 ± 4% perf-profile.calltrace.cycles-pp.set_pte_range.finish_fault.do_cow_fault.do_fault.__handle_mm_fault
9.22 ± 2% +3.1 12.33 ± 2% perf-profile.calltrace.cycles-pp.finish_fault.do_cow_fault.do_fault.__handle_mm_fault.handle_mm_fault
44.13 ± 3% +3.2 47.34 perf-profile.calltrace.cycles-pp.do_cow_fault.do_fault.__handle_mm_fault.handle_mm_fault.do_user_addr_fault
44.27 ± 3% +3.2 47.49 perf-profile.calltrace.cycles-pp.do_fault.__handle_mm_fault.handle_mm_fault.do_user_addr_fault.exc_page_fault
45.63 ± 2% +3.3 48.95 perf-profile.calltrace.cycles-pp.__handle_mm_fault.handle_mm_fault.do_user_addr_fault.exc_page_fault.asm_exc_page_fault
0.00 +3.4 3.37 ± 2% perf-profile.calltrace.cycles-pp.__list_del_entry_valid_or_report.__rmqueue_pcplist.rmqueue.get_page_from_freelist.__alloc_pages
46.88 ± 3% +3.4 50.29 perf-profile.calltrace.cycles-pp.handle_mm_fault.do_user_addr_fault.exc_page_fault.asm_exc_page_fault.testcase
49.40 ± 2% +3.6 53.03 perf-profile.calltrace.cycles-pp.do_user_addr_fault.exc_page_fault.asm_exc_page_fault.testcase
49.59 ± 2% +3.7 53.24 perf-profile.calltrace.cycles-pp.exc_page_fault.asm_exc_page_fault.testcase
59.06 ± 2% +4.5 63.60 perf-profile.calltrace.cycles-pp.asm_exc_page_fault.testcase
56.32 ± 3% +4.6 60.89 perf-profile.calltrace.cycles-pp.testcase
20.16 ± 3% +4.9 25.10 perf-profile.calltrace.cycles-pp.copy_page.do_cow_fault.do_fault.__handle_mm_fault.handle_mm_fault
16.66 ± 3% -8.8 7.83 ± 8% perf-profile.children.cycles-pp._raw_spin_lock_irqsave
16.48 ± 3% -8.8 7.66 ± 8% perf-profile.children.cycles-pp.native_queued_spin_lock_slowpath
9.90 ± 3% -8.4 1.50 ± 5% perf-profile.children.cycles-pp.rmqueue_bulk
8.92 ± 3% -6.7 2.18 ± 2% perf-profile.children.cycles-pp.free_unref_page_list
8.47 ± 3% -6.7 1.74 ± 4% perf-profile.children.cycles-pp.free_pcppages_bulk
10.96 ± 3% -5.3 5.64 ± 2% perf-profile.children.cycles-pp.get_page_from_freelist
10.62 ± 3% -5.3 5.30 ± 2% perf-profile.children.cycles-pp.rmqueue
10.26 ± 3% -5.3 4.97 ± 3% perf-profile.children.cycles-pp.__rmqueue_pcplist
11.24 ± 3% -5.3 5.96 ± 2% perf-profile.children.cycles-pp.__alloc_pages
11.18 ± 3% -5.3 5.92 ± 2% perf-profile.children.cycles-pp.__folio_alloc
11.57 ± 3% -5.2 6.37 ± 2% perf-profile.children.cycles-pp.vma_alloc_folio
11.19 ± 3% -4.4 6.82 ± 5% perf-profile.children.cycles-pp.release_pages
11.46 ± 3% -4.3 7.12 ± 5% perf-profile.children.cycles-pp.tlb_batch_pages_flush
15.52 ± 3% -3.7 11.81 perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
15.52 ± 3% -3.7 11.81 perf-profile.children.cycles-pp.do_syscall_64
15.41 ± 3% -3.7 11.70 perf-profile.children.cycles-pp.__munmap
15.40 ± 3% -3.7 11.70 perf-profile.children.cycles-pp.do_vmi_munmap
15.40 ± 3% -3.7 11.70 perf-profile.children.cycles-pp.do_vmi_align_munmap
15.40 ± 3% -3.7 11.70 perf-profile.children.cycles-pp.__x64_sys_munmap
15.40 ± 3% -3.7 11.70 perf-profile.children.cycles-pp.__vm_munmap
15.39 ± 3% -3.7 11.70 perf-profile.children.cycles-pp.unmap_region
14.10 ± 3% -3.6 10.52 perf-profile.children.cycles-pp.unmap_vmas
14.10 ± 3% -3.6 10.52 perf-profile.children.cycles-pp.unmap_page_range
14.10 ± 3% -3.6 10.52 perf-profile.children.cycles-pp.zap_pmd_range
14.10 ± 3% -3.6 10.52 perf-profile.children.cycles-pp.zap_pte_range
2.60 ± 3% -2.0 0.56 ± 4% perf-profile.children.cycles-pp.__free_one_page
1.28 ± 3% -0.1 1.17 ± 2% perf-profile.children.cycles-pp.tlb_finish_mmu
0.15 ± 19% -0.1 0.08 ± 14% perf-profile.children.cycles-pp.get_mem_cgroup_from_mm
0.61 ± 3% -0.0 0.58 ± 2% perf-profile.children.cycles-pp.__mem_cgroup_charge
0.11 ± 6% -0.0 0.08 ± 7% perf-profile.children.cycles-pp.__mod_zone_page_state
0.25 ± 4% +0.0 0.26 perf-profile.children.cycles-pp.error_entry
0.15 ± 3% +0.0 0.17 ± 4% perf-profile.children.cycles-pp.free_unref_page_commit
0.12 ± 8% +0.0 0.14 ± 3% perf-profile.children.cycles-pp.__mem_cgroup_uncharge_list
0.18 ± 3% +0.0 0.20 ± 4% perf-profile.children.cycles-pp.access_error
0.07 ± 5% +0.0 0.09 ± 7% perf-profile.children.cycles-pp.task_tick_fair
0.04 ± 45% +0.0 0.06 ± 7% perf-profile.children.cycles-pp.page_counter_try_charge
0.30 ± 4% +0.0 0.32 perf-profile.children.cycles-pp.down_read_trylock
0.27 ± 3% +0.0 0.30 ± 2% perf-profile.children.cycles-pp.up_read
0.15 ± 8% +0.0 0.18 ± 3% perf-profile.children.cycles-pp.mem_cgroup_update_lru_size
0.02 ±142% +0.1 0.07 ± 29% perf-profile.children.cycles-pp.ret_from_fork_asm
0.44 ± 2% +0.1 0.49 ± 3% perf-profile.children.cycles-pp.mas_walk
0.46 ± 4% +0.1 0.52 ± 2% perf-profile.children.cycles-pp.__mod_memcg_lruvec_state
0.67 ± 3% +0.1 0.73 ± 4% perf-profile.children.cycles-pp.lock_mm_and_find_vma
0.42 ± 3% +0.1 0.48 ± 2% perf-profile.children.cycles-pp.free_swap_cache
0.43 ± 4% +0.1 0.49 ± 2% perf-profile.children.cycles-pp.free_pages_and_swap_cache
0.30 ± 5% +0.1 0.37 ± 3% perf-profile.children.cycles-pp.xas_descend
0.86 ± 3% +0.1 0.92 perf-profile.children.cycles-pp.___perf_sw_event
0.73 ± 3% +0.1 0.80 perf-profile.children.cycles-pp.lock_vma_under_rcu
0.40 ± 2% +0.1 0.47 perf-profile.children.cycles-pp.__mod_node_page_state
0.01 ±223% +0.1 0.09 ± 12% perf-profile.children.cycles-pp.shmem_get_policy
0.53 ± 2% +0.1 0.62 ± 2% perf-profile.children.cycles-pp.__mod_lruvec_state
1.09 ± 3% +0.1 1.18 perf-profile.children.cycles-pp.__perf_sw_event
0.50 ± 5% +0.1 0.60 ± 3% perf-profile.children.cycles-pp.xas_load
0.68 ± 3% +0.1 0.78 ± 3% perf-profile.children.cycles-pp.page_remove_rmap
1.45 ± 3% +0.1 1.60 perf-profile.children.cycles-pp.sync_regs
0.77 ± 4% +0.2 0.93 ± 5% perf-profile.children.cycles-pp.folio_add_new_anon_rmap
0.84 ± 5% +0.2 1.02 ± 7% perf-profile.children.cycles-pp.__mod_lruvec_page_state
0.96 ± 4% +0.2 1.15 ± 3% perf-profile.children.cycles-pp.lru_add_fn
1.27 ± 5% +0.2 1.48 ± 3% perf-profile.children.cycles-pp.filemap_get_entry
1.62 ± 4% +0.3 1.88 ± 3% perf-profile.children.cycles-pp.shmem_get_folio_gfp
1.75 ± 5% +0.3 2.06 ± 3% perf-profile.children.cycles-pp.shmem_fault
1.87 ± 4% +0.3 2.18 ± 2% perf-profile.children.cycles-pp.__do_fault
2.19 ± 2% +0.3 2.51 perf-profile.children.cycles-pp.native_irq_return_iret
2.64 ± 4% +0.5 3.18 ± 6% perf-profile.children.cycles-pp._compound_head
4.62 ± 3% +0.6 5.26 perf-profile.children.cycles-pp._raw_spin_lock
4.67 ± 3% +0.7 5.34 perf-profile.children.cycles-pp.__pte_offset_map_lock
3.32 ± 2% +2.2 5.54 ± 5% perf-profile.children.cycles-pp.folio_batch_move_lru
3.47 ± 2% +2.2 5.72 ± 5% perf-profile.children.cycles-pp.folio_add_lru_vma
4.49 ± 2% +2.4 6.92 ± 4% perf-profile.children.cycles-pp.set_pte_range
9.25 ± 2% +3.1 12.36 ± 2% perf-profile.children.cycles-pp.finish_fault
2.25 ± 2% +3.1 5.36 ± 2% perf-profile.children.cycles-pp.__list_del_entry_valid_or_report
44.16 ± 3% +3.2 47.37 perf-profile.children.cycles-pp.do_cow_fault
44.28 ± 3% +3.2 47.50 perf-profile.children.cycles-pp.do_fault
45.66 ± 2% +3.3 48.98 perf-profile.children.cycles-pp.__handle_mm_fault
46.91 ± 2% +3.4 50.33 perf-profile.children.cycles-pp.handle_mm_fault
49.44 ± 2% +3.6 53.08 perf-profile.children.cycles-pp.do_user_addr_fault
49.62 ± 2% +3.6 53.27 perf-profile.children.cycles-pp.exc_page_fault
2.70 ± 3% +4.1 6.75 ± 8% perf-profile.children.cycles-pp.folio_lruvec_lock_irqsave
55.26 ± 2% +4.2 59.44 perf-profile.children.cycles-pp.asm_exc_page_fault
58.13 ± 3% +4.6 62.72 perf-profile.children.cycles-pp.testcase
20.19 ± 3% +4.9 25.14 perf-profile.children.cycles-pp.copy_page
16.48 ± 3% -8.8 7.66 ± 8% perf-profile.self.cycles-pp.native_queued_spin_lock_slowpath
2.53 ± 3% -2.0 0.54 ± 3% perf-profile.self.cycles-pp.__free_one_page
0.12 ± 4% -0.1 0.05 ± 46% perf-profile.self.cycles-pp.rmqueue_bulk
0.14 ± 19% -0.1 0.08 ± 14% perf-profile.self.cycles-pp.get_mem_cgroup_from_mm
0.10 ± 3% -0.0 0.08 ± 10% perf-profile.self.cycles-pp.__mod_zone_page_state
0.13 ± 5% +0.0 0.14 ± 2% perf-profile.self.cycles-pp.free_unref_page_commit
0.13 ± 3% +0.0 0.14 ± 3% perf-profile.self.cycles-pp.exc_page_fault
0.15 ± 5% +0.0 0.17 ± 4% perf-profile.self.cycles-pp.__pte_offset_map
0.04 ± 44% +0.0 0.06 ± 6% perf-profile.self.cycles-pp.page_counter_try_charge
0.18 ± 3% +0.0 0.20 ± 4% perf-profile.self.cycles-pp.access_error
0.30 ± 3% +0.0 0.32 ± 2% perf-profile.self.cycles-pp.down_read_trylock
0.16 ± 6% +0.0 0.18 perf-profile.self.cycles-pp.set_pte_range
0.26 ± 2% +0.0 0.29 ± 3% perf-profile.self.cycles-pp.up_read
0.15 ± 8% +0.0 0.18 ± 4% perf-profile.self.cycles-pp.folio_add_lru_vma
0.15 ± 8% +0.0 0.18 ± 3% perf-profile.self.cycles-pp.mem_cgroup_update_lru_size
0.22 ± 6% +0.0 0.26 ± 5% perf-profile.self.cycles-pp.__alloc_pages
0.32 ± 6% +0.0 0.36 ± 3% perf-profile.self.cycles-pp.shmem_get_folio_gfp
0.28 ± 5% +0.0 0.32 ± 4% perf-profile.self.cycles-pp.do_cow_fault
0.14 ± 7% +0.0 0.18 ± 6% perf-profile.self.cycles-pp.shmem_fault
0.34 ± 5% +0.0 0.38 ± 4% perf-profile.self.cycles-pp.__mod_memcg_lruvec_state
0.00 +0.1 0.05 perf-profile.self.cycles-pp.__cond_resched
0.44 ± 3% +0.1 0.49 ± 4% perf-profile.self.cycles-pp.page_remove_rmap
0.41 ± 3% +0.1 0.47 ± 3% perf-profile.self.cycles-pp.free_swap_cache
0.75 ± 3% +0.1 0.81 ± 2% perf-profile.self.cycles-pp.___perf_sw_event
0.91 ± 2% +0.1 0.98 ± 2% perf-profile.self.cycles-pp.__handle_mm_fault
0.29 ± 6% +0.1 0.36 ± 3% perf-profile.self.cycles-pp.xas_descend
0.38 ± 2% +0.1 0.45 ± 2% perf-profile.self.cycles-pp.__mod_node_page_state
0.01 ±223% +0.1 0.09 ± 8% perf-profile.self.cycles-pp.shmem_get_policy
0.58 ± 3% +0.1 0.66 ± 2% perf-profile.self.cycles-pp.release_pages
0.44 ± 4% +0.1 0.54 ± 3% perf-profile.self.cycles-pp.lru_add_fn
1.44 ± 3% +0.1 1.59 perf-profile.self.cycles-pp.sync_regs
2.18 ± 2% +0.3 2.50 perf-profile.self.cycles-pp.native_irq_return_iret
4.36 ± 3% +0.4 4.76 perf-profile.self.cycles-pp.testcase
2.61 ± 4% +0.5 3.14 ± 5% perf-profile.self.cycles-pp._compound_head
4.60 ± 3% +0.6 5.23 perf-profile.self.cycles-pp._raw_spin_lock
2.23 ± 2% +3.1 5.34 ± 2% perf-profile.self.cycles-pp.__list_del_entry_valid_or_report
20.10 ± 3% +4.9 25.02 perf-profile.self.cycles-pp.copy_page
Disclaimer:
Results have been estimated based on internal Intel analysis and are provided
for informational purposes only. Any difference in system hardware or software
design or configuration may affect actual performance.
diff --git a/include/linux/gfp.h b/include/linux/gfp.h index 665edc11fb9f..5b917e5b9350 100644 --- a/include/linux/gfp.h +++ b/include/linux/gfp.h @@ -320,6 +320,7 @@ extern void page_frag_free(void *addr); #define free_page(addr) free_pages((addr), 0) void page_alloc_init_cpuhp(void); +int decay_pcp_high(struct zone *zone, struct per_cpu_pages *pcp); void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp); void drain_all_pages(struct zone *zone); void drain_local_pages(struct zone *zone); diff --git a/mm/page_alloc.c b/mm/page_alloc.c index 1fb2c6ebde9c..8382ad2cdfd4 100644 --- a/mm/page_alloc.c +++ b/mm/page_alloc.c @@ -2157,6 +2157,40 @@ static int rmqueue_bulk(struct zone *zone, unsigned int order, return i; } +/* + * Called from the vmstat counter updater to decay the PCP high. + * Return whether there are addition works to do. + */ +int decay_pcp_high(struct zone *zone, struct per_cpu_pages *pcp) +{ + int high_min, to_drain, batch; + int todo = 0; + + high_min = READ_ONCE(pcp->high_min); + batch = READ_ONCE(pcp->batch); + /* + * Decrease pcp->high periodically to try to free possible + * idle PCP pages. And, avoid to free too many pages to + * control latency. This caps pcp->high decrement too. + */ + if (pcp->high > high_min) { + pcp->high = max3(pcp->count - (batch << CONFIG_PCP_BATCH_SCALE_MAX), + pcp->high - (pcp->high >> 3), high_min); + if (pcp->high > high_min) + todo++; + } + + to_drain = pcp->count - pcp->high; + if (to_drain > 0) { + spin_lock(&pcp->lock); + free_pcppages_bulk(zone, to_drain, pcp, 0); + spin_unlock(&pcp->lock); + todo++; + } + + return todo; +} + #ifdef CONFIG_NUMA /* * Called from the vmstat counter updater to drain pagesets of this @@ -2318,14 +2352,13 @@ static bool free_unref_page_prepare(struct page *page, unsigned long pfn, return true; } -static int nr_pcp_free(struct per_cpu_pages *pcp, int high, bool free_high) +static int nr_pcp_free(struct per_cpu_pages *pcp, int batch, int high, bool free_high) { int min_nr_free, max_nr_free; - int batch = READ_ONCE(pcp->batch); - /* Free everything if batch freeing high-order pages. */ + /* Free as much as possible if batch freeing high-order pages. */ if (unlikely(free_high)) - return pcp->count; + return min(pcp->count, batch << CONFIG_PCP_BATCH_SCALE_MAX); /* Check for PCP disabled or boot pageset */ if (unlikely(high < batch)) @@ -2340,7 +2373,7 @@ static int nr_pcp_free(struct per_cpu_pages *pcp, int high, bool free_high) * freeing of pages without any allocation. */ batch <<= pcp->free_factor; - if (batch < max_nr_free && pcp->free_factor < CONFIG_PCP_BATCH_SCALE_MAX) + if (batch <= max_nr_free && pcp->free_factor < CONFIG_PCP_BATCH_SCALE_MAX) pcp->free_factor++; batch = clamp(batch, min_nr_free, max_nr_free); @@ -2348,28 +2381,48 @@ static int nr_pcp_free(struct per_cpu_pages *pcp, int high, bool free_high) } static int nr_pcp_high(struct per_cpu_pages *pcp, struct zone *zone, - bool free_high) + int batch, bool free_high) { - int high = READ_ONCE(pcp->high_min); + int high, high_min, high_max; - if (unlikely(!high || free_high)) + high_min = READ_ONCE(pcp->high_min); + high_max = READ_ONCE(pcp->high_max); + high = pcp->high = clamp(pcp->high, high_min, high_max); + + if (unlikely(!high)) return 0; - if (!test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) - return high; + if (unlikely(free_high)) { + pcp->high = max(high - (batch << CONFIG_PCP_BATCH_SCALE_MAX), + high_min); + return 0; + } /* * If reclaim is active, limit the number of pages that can be * stored on pcp lists */ - return min(READ_ONCE(pcp->batch) << 2, high); + if (test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) { + pcp->high = max(high - (batch << pcp->free_factor), high_min); + return min(batch << 2, pcp->high); + } + + if (pcp->count >= high && high_min != high_max) { + int need_high = (batch << pcp->free_factor) + batch; + + /* pcp->high should be large enough to hold batch freed pages */ + if (pcp->high < need_high) + pcp->high = clamp(need_high, high_min, high_max); + } + + return high; } static void free_unref_page_commit(struct zone *zone, struct per_cpu_pages *pcp, struct page *page, int migratetype, unsigned int order) { - int high; + int high, batch; int pindex; bool free_high = false; @@ -2384,6 +2437,7 @@ static void free_unref_page_commit(struct zone *zone, struct per_cpu_pages *pcp, list_add(&page->pcp_list, &pcp->lists[pindex]); pcp->count += 1 << order; + batch = READ_ONCE(pcp->batch); /* * As high-order pages other than THP's stored on PCP can contribute * to fragmentation, limit the number stored when PCP is heavily @@ -2394,14 +2448,15 @@ static void free_unref_page_commit(struct zone *zone, struct per_cpu_pages *pcp, free_high = (pcp->free_factor && (pcp->flags & PCPF_PREV_FREE_HIGH_ORDER) && (!(pcp->flags & PCPF_FREE_HIGH_BATCH) || - pcp->count >= READ_ONCE(pcp->batch))); + pcp->count >= READ_ONCE(batch))); pcp->flags |= PCPF_PREV_FREE_HIGH_ORDER; } else if (pcp->flags & PCPF_PREV_FREE_HIGH_ORDER) { pcp->flags &= ~PCPF_PREV_FREE_HIGH_ORDER; } - high = nr_pcp_high(pcp, zone, free_high); + high = nr_pcp_high(pcp, zone, batch, free_high); if (pcp->count >= high) { - free_pcppages_bulk(zone, nr_pcp_free(pcp, high, free_high), pcp, pindex); + free_pcppages_bulk(zone, nr_pcp_free(pcp, batch, high, free_high), + pcp, pindex); } } @@ -2685,24 +2740,38 @@ struct page *rmqueue_buddy(struct zone *preferred_zone, struct zone *zone, return page; } -static int nr_pcp_alloc(struct per_cpu_pages *pcp, int order) +static int nr_pcp_alloc(struct per_cpu_pages *pcp, struct zone *zone, int order) { - int high, batch, max_nr_alloc; + int high, base_batch, batch, max_nr_alloc; + int high_max, high_min; - high = READ_ONCE(pcp->high_min); - batch = READ_ONCE(pcp->batch); + base_batch = READ_ONCE(pcp->batch); + high_min = READ_ONCE(pcp->high_min); + high_max = READ_ONCE(pcp->high_max); + high = pcp->high = clamp(pcp->high, high_min, high_max); /* Check for PCP disabled or boot pageset */ - if (unlikely(high < batch)) + if (unlikely(high < base_batch)) return 1; + if (order) + batch = base_batch; + else + batch = (base_batch << pcp->alloc_factor); + /* - * Double the number of pages allocated each time there is subsequent - * allocation of order-0 pages without any freeing. + * If we had larger pcp->high, we could avoid to allocate from + * zone. */ + if (high_min != high_max && !test_bit(ZONE_RECLAIM_ACTIVE, &zone->flags)) + high = pcp->high = min(high + batch, high_max); + if (!order) { - max_nr_alloc = max(high - pcp->count - batch, batch); - batch <<= pcp->alloc_factor; + max_nr_alloc = max(high - pcp->count - base_batch, base_batch); + /* + * Double the number of pages allocated each time there is + * subsequent allocation of order-0 pages without any freeing. + */ if (batch <= max_nr_alloc && pcp->alloc_factor < CONFIG_PCP_BATCH_SCALE_MAX) pcp->alloc_factor++; @@ -2733,7 +2802,7 @@ struct page *__rmqueue_pcplist(struct zone *zone, unsigned int order, do { if (list_empty(list)) { - int batch = nr_pcp_alloc(pcp, order); + int batch = nr_pcp_alloc(pcp, zone, order); int alloced; alloced = rmqueue_bulk(zone, order, diff --git a/mm/vmstat.c b/mm/vmstat.c index 00e81e99c6ee..2f716ad14168 100644 --- a/mm/vmstat.c +++ b/mm/vmstat.c @@ -814,9 +814,7 @@ static int refresh_cpu_vm_stats(bool do_pagesets) for_each_populated_zone(zone) { struct per_cpu_zonestat __percpu *pzstats = zone->per_cpu_zonestats; -#ifdef CONFIG_NUMA struct per_cpu_pages __percpu *pcp = zone->per_cpu_pageset; -#endif for (i = 0; i < NR_VM_ZONE_STAT_ITEMS; i++) { int v; @@ -832,10 +830,12 @@ static int refresh_cpu_vm_stats(bool do_pagesets) #endif } } -#ifdef CONFIG_NUMA if (do_pagesets) { cond_resched(); + + changes += decay_pcp_high(zone, this_cpu_ptr(pcp)); +#ifdef CONFIG_NUMA /* * Deal with draining the remote pageset of this * processor @@ -862,8 +862,8 @@ static int refresh_cpu_vm_stats(bool do_pagesets) drain_zone_pages(zone, this_cpu_ptr(pcp)); changes++; } - } #endif + } } for_each_online_pgdat(pgdat) {
The target to tune PCP high automatically is as follows, - Minimize allocation/freeing from/to shared zone - Minimize idle pages in PCP - Minimize pages in PCP if the system free pages is too few To reach these target, a tuning algorithm as follows is designed, - When we refill PCP via allocating from the zone, increase PCP high. Because if we had larger PCP, we could avoid to allocate from the zone. - In periodic vmstat updating kworker (via refresh_cpu_vm_stats()), decrease PCP high to try to free possible idle PCP pages. - When page reclaiming is active for the zone, stop increasing PCP high in allocating path, decrease PCP high and free some pages in freeing path. So, the PCP high can be tuned to the page allocating/freeing depth of workloads eventually. One issue of the algorithm is that if the number of pages allocated is much more than that of pages freed on a CPU, the PCP high may become the maximal value even if the allocating/freeing depth is small. But this isn't a severe issue, because there are no idle pages in this case. One alternative choice is to increase PCP high when we drain PCP via trying to free pages to the zone, but don't increase PCP high during PCP refilling. This can avoid the issue above. But if the number of pages allocated is much less than that of pages freed on a CPU, there will be many idle pages in PCP and it is hard to free these idle pages. 1/8 (>> 3) of PCP high will be decreased periodically. The value 1/8 is kind of arbitrary. Just to make sure that the idle PCP pages will be freed eventually. On a 2-socket Intel server with 224 logical CPU, we run 8 kbuild instances in parallel (each with `make -j 28`) in 8 cgroup. This simulates the kbuild server that is used by 0-Day kbuild service. With the patch, the build time decreases 3.5%. The cycles% of the spinlock contention (mostly for zone lock) decreases from 11.0% to 0.5%. The number of PCP draining for high order pages freeing (free_high) decreases 65.6%. The number of pages allocated from zone (instead of from PCP) decreases 83.9%. Signed-off-by: "Huang, Ying" <ying.huang@intel.com> Suggested-by: Mel Gorman <mgorman@techsingularity.net> Suggested-by: Michal Hocko <mhocko@suse.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Vlastimil Babka <vbabka@suse.cz> Cc: David Hildenbrand <david@redhat.com> Cc: Johannes Weiner <jweiner@redhat.com> Cc: Dave Hansen <dave.hansen@linux.intel.com> Cc: Pavel Tatashin <pasha.tatashin@soleen.com> Cc: Matthew Wilcox <willy@infradead.org> Cc: Christoph Lameter <cl@linux.com> --- include/linux/gfp.h | 1 + mm/page_alloc.c | 119 ++++++++++++++++++++++++++++++++++---------- mm/vmstat.c | 8 +-- 3 files changed, 99 insertions(+), 29 deletions(-)