Re: [linux-next:master] [mm/filemap] 0b9c0aeba9: pts.graphics-magick.Swirl.iterations_per_minute 8.6% regression
Fujunjie <[email protected]> Wed, 24 Jun 2026 16:01:44 +0800
| Newsgroups | dev.linux.lists.oe-lkp,org.kernel.vger.linux-fsdevel,org.kvack.linux-mm |
|---|---|
| Message-ID | <[email protected]> |
On 6/17/2026 10:26 PM, kernel test robot wrote:
>
>
> Hello,
>
> kernel test robot noticed a 8.6% regression of pts.graphics-magick.Swirl.iterations_per_minute on:
>
>
> commit: 0b9c0aeba938aad9964f855df00bf929b83a484d ("mm/filemap: count only the faulting address as a mmap hit")
> https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git master
>
> [still regression on linux-next/master ec039126b7fac4e3af35ebccaa7c6f9b6875ba81]
>
> testcase: pts
> config: x86_64-rhel-9.4
> compiler: gcc-14
> test machine: 256 threads 2 sockets Intel(R) Xeon(R) 6767P CPU @ 2.4GHz (Granite Rapids) with 256G memory
> parameters:
>
> test: graphics-magick-2.2.0
> option_a: Swirl
> cpufreq_governor: performance
>
>
>
>
> If you fix the issue in a separate patch/commit (i.e. not just a new version of
> the same patch/commit), kindly add following tags
> | Reported-by: kernel test robot <[email protected]>
> | Closes: https://lore.kernel.org/oe-lkp/[email protected]
>
>
> Details are as below:
> -------------------------------------------------------------------------------------------------->
>
>
> The kernel config and materials to reproduce are available at:
> https://download.01.org/0day-ci/archive/20260617/[email protected]
>
> =========================================================================================
> compiler/cpufreq_governor/kconfig/option_a/rootfs/tbox_group/test/testcase:
> gcc-14/performance/x86_64-rhel-9.4/Swirl/debian-12-x86_64-phoronix/lkp-gnr-2sp3/graphics-magick-2.2.0/pts
>
> commit:
> 95b8e43226 ("mm: use zone lock guard in __offline_isolated_pages()")
> 0b9c0aeba9 ("mm/filemap: count only the faulting address as a mmap hit")
>
> 95b8e432265f61bd 0b9c0aeba938aad9964f855df00
> ---------------- ---------------------------
> %stddev %change %stddev
> \ | \
> 814.83 -8.6% 745.00 pts.graphics-magick.Swirl.iterations_per_minute
> 55127 ± 5% -10.3% 49460 ± 3% pts.time.involuntary_context_switches
> 2153 ± 5% +5894.6% 129074 ± 6% pts.time.major_page_faults
> 12775652 -8.5% 11686096 pts.time.minor_page_faults
> 15099 -8.1% 13876 pts.time.percent_of_cpu_this_job_got
> 2124 -4.0% 2039 pts.time.system_time
> 9036 -9.2% 8207 pts.time.user_time
> 4500789 -4.4% 4301829 pts.time.voluntary_context_switches
> 12315 ± 4% +10.6% 13616 ± 4% meminfo.PageTables
> 1828588 ± 5% -10.9% 1629373 ± 2% numa-numastat.node1.numa_hit
> 1828671 ± 5% -10.9% 1629233 ± 2% numa-vmstat.node1.numa_hit
> 1.34 ± 3% +0.2 1.55 mpstat.cpu.all.iowait%
> 0.10 ± 3% +0.0 0.12 ± 3% mpstat.cpu.all.soft%
> 78.37 -6.7% 73.12 mpstat.max_utilization_pct
> 2.44 ± 66% +4.8 7.24 ± 58% perf-profile.calltrace.cycles-pp.sw_perf_event_destroy.__free_event.perf_event_release_kernel.perf_release.__fput
> 2.98 ± 59% +5.8 8.81 ± 33% perf-profile.calltrace.cycles-pp.__free_event.perf_event_release_kernel.perf_release.__fput.task_work_run
> 2.44 ± 66% +4.8 7.24 ± 49% perf-profile.children.cycles-pp.sw_perf_event_destroy
> 2.98 ± 59% +5.8 8.81 ± 33% perf-profile.children.cycles-pp.__free_event
> 3149 ± 2% +7.2% 3376 ± 2% proc-vmstat.nr_page_table_pages
> 443579 -1.9% 435060 proc-vmstat.nr_shmem
> 3584637 ± 3% -7.2% 3324877 proc-vmstat.numa_hit
> 3319585 ± 3% -7.8% 3060170 proc-vmstat.numa_local
> 3.533e+08 -13.0% 3.075e+08 proc-vmstat.pgalloc_normal
> 13271607 -7.4% 12294786 proc-vmstat.pgfault
> 3.527e+08 -13.0% 3.069e+08 proc-vmstat.pgfree
> 2086 ± 5% +6047.5% 128277 ± 6% proc-vmstat.pgmajfault
> 166303 -8.6% 152057 proc-vmstat.thp_fault_alloc
> 28.46 ± 2% +1.0 29.44 perf-stat.i.cache-miss-rate%
> 3164 ± 6% +24.3% 3933 ± 2% perf-stat.i.cpu-migrations
> 28.98 ± 7% +5995.2% 1766 ± 6% perf-stat.i.major-faults
> 0.44 +2.6% 0.46 perf-stat.overall.MPKI
> 0.01 +0.0 0.01 perf-stat.overall.branch-miss-rate%
> 1719 -1.6% 1692 perf-stat.overall.cycles-between-cache-misses
> 59.55 +1.9 61.43 perf-stat.overall.node-load-miss-rate%
> 3138 ± 6% +23.7% 3880 ± 2% perf-stat.ps.cpu-migrations
> 28.03 ± 7% +5980.2% 1704 ± 6% perf-stat.ps.major-faults
> 5.496e+13 -8.5% 5.028e+13 perf-stat.total.instructions
> 1.41 ± 36% +83.3% 2.58 ± 11% perf-sched.sch_delay.max.ms.perf_trace_sched_switch.do_wait.kernel_wait4.do_syscall_64.entry_SYSCALL_64_after_hwframe
> 1.17 ± 34% -82.8% 0.20 ±169% perf-sched.sch_delay.max.ms.perf_trace_sched_switch.preempt_schedule_thunk.affine_move_task.__set_cpus_allowed_ptr.__sched_setaffinity
> 7.34 +11.2% 8.15 perf-sched.total_wait_and_delay.average.ms
> 366839 -7.5% 339473 perf-sched.total_wait_and_delay.count.ms
> 7.32 +11.2% 8.14 perf-sched.total_wait_time.average.ms
> 3.58 +24.1% 4.45 ± 2% perf-sched.wait_and_delay.avg.ms.perf_trace_sched_switch.futex_do_wait.__futex_wait.futex_wait.do_futex
> 0.38 ± 2% +48.1% 0.57 ± 2% perf-sched.wait_and_delay.avg.ms.perf_trace_sched_switch.io_schedule.folio_wait_bit_common.filemap_fault.__do_fault
> 4.64 ±101% +192.3% 13.57 ± 23% perf-sched.wait_and_delay.avg.ms.perf_trace_sched_switch.preempt_schedule_thunk._raw_spin_unlock.__do_huge_pmd_anonymous_page.__handle_mm_fault
> 7.14 ± 45% +73.8% 12.41 ± 3% perf-sched.wait_and_delay.avg.ms.perf_trace_sched_switch.preempt_schedule_thunk.lock_vma_under_rcu.do_user_addr_fault.exc_page_fault
> 339.44 ± 3% +18.1% 400.91 ± 3% perf-sched.wait_and_delay.avg.ms.perf_trace_sched_switch.schedule_hrtimeout_range_clock.poll_schedule_timeout.constprop.0
> 58862 -8.3% 53992 perf-sched.wait_and_delay.count.perf_trace_sched_switch.futex_do_wait.__futex_wait.futex_wait.do_futex
> 294758 -7.6% 272232 perf-sched.wait_and_delay.count.perf_trace_sched_switch.io_schedule.folio_wait_bit_common.filemap_fault.__do_fault
> 113.17 ± 4% -12.1% 99.50 ± 2% perf-sched.wait_and_delay.count.perf_trace_sched_switch.schedule_hrtimeout_range_clock.poll_schedule_timeout.constprop.0
> 7.56 ±100% +191.1% 22.02 ± 3% perf-sched.wait_and_delay.max.ms.perf_trace_sched_switch.preempt_schedule_thunk._raw_spin_unlock.__do_huge_pmd_anonymous_page.__handle_mm_fault
> 15.87 ± 3% +36.3% 21.64 perf-sched.wait_and_delay.max.ms.perf_trace_sched_switch.preempt_schedule_thunk.cpu_stop_queue_work.stop_one_cpu.migrate_task_to
> 14.18 ± 45% +67.2% 23.71 ± 6% perf-sched.wait_and_delay.max.ms.perf_trace_sched_switch.preempt_schedule_thunk.lock_vma_under_rcu.do_user_addr_fault.exc_page_fault
> 3.55 +24.2% 4.41 ± 2% perf-sched.wait_time.avg.ms.perf_trace_sched_switch.futex_do_wait.__futex_wait.futex_wait.do_futex
> 0.37 ± 2% +49.8% 0.56 ± 2% perf-sched.wait_time.avg.ms.perf_trace_sched_switch.io_schedule.folio_wait_bit_common.filemap_fault.__do_fault
> 2.43 ± 14% +91.5% 4.66 ± 16% perf-sched.wait_time.avg.ms.perf_trace_sched_switch.irqentry_exit.asm_exc_page_fault.[unknown].[unknown]
> 6.68 ± 50% +103.0% 13.55 ± 23% perf-sched.wait_time.avg.ms.perf_trace_sched_switch.preempt_schedule_thunk._raw_spin_unlock.__do_huge_pmd_anonymous_page.__handle_mm_fault
> 8.24 ± 8% +50.0% 12.36 ± 3% perf-sched.wait_time.avg.ms.perf_trace_sched_switch.preempt_schedule_thunk.lock_vma_under_rcu.do_user_addr_fault.exc_page_fault
> 339.33 ± 3% +18.1% 400.81 ± 3% perf-sched.wait_time.avg.ms.perf_trace_sched_switch.schedule_hrtimeout_range_clock.poll_schedule_timeout.constprop.0
> 15.55 ± 3% +52.0% 23.65 ± 14% perf-sched.wait_time.max.ms.perf_trace_sched_switch.irqentry_exit.asm_exc_page_fault.[unknown].[unknown]
> 14.33 ± 6% +53.7% 22.03 ± 5% perf-sched.wait_time.max.ms.perf_trace_sched_switch.irqentry_exit.asm_sysvec_call_function_single.[unknown]
> 15.24 ± 5% +44.4% 22.01 ± 5% perf-sched.wait_time.max.ms.perf_trace_sched_switch.preempt_schedule_irq.irqentry_exit.asm_sysvec_apic_timer_interrupt.do_user_addr_fault
> 12.27 ± 44% +79.4% 22.01 ± 3% perf-sched.wait_time.max.ms.perf_trace_sched_switch.preempt_schedule_thunk._raw_spin_unlock.__do_huge_pmd_anonymous_page.__handle_mm_fault
> 15.86 ± 3% +36.3% 21.63 perf-sched.wait_time.max.ms.perf_trace_sched_switch.preempt_schedule_thunk.cpu_stop_queue_work.stop_one_cpu.migrate_task_to
> 14.53 ± 3% +45.3% 21.10 ± 3% perf-sched.wait_time.max.ms.perf_trace_sched_switch.preempt_schedule_thunk.do_read_fault.do_fault.__handle_mm_fault
> 15.90 ± 5% +41.6% 22.51 ± 2% perf-sched.wait_time.max.ms.perf_trace_sched_switch.preempt_schedule_thunk.lock_vma_under_rcu.do_user_addr_fault.exc_page_fault
> 10.27 ± 61% +100.7% 20.61 ± 2% perf-sched.wait_time.max.ms.perf_trace_sched_switch.preempt_schedule_thunk.stop_two_cpus.migrate_swap.task_numa_migrate
> 4282234 -17.3% 3541405 sched_debug.cfs_rq:/.avg_vruntime.avg
> 4735201 ± 2% -13.9% 4076046 ± 4% sched_debug.cfs_rq:/.avg_vruntime.max
> 4115256 -17.0% 3417248 ± 2% sched_debug.cfs_rq:/.avg_vruntime.min
> 423.15 ± 2% -9.6% 382.71 sched_debug.cfs_rq:/.runnable_avg.avg
> 420.53 ± 2% -9.6% 380.35 sched_debug.cfs_rq:/.util_avg.avg
> 1190 ± 4% -9.6% 1077 ± 3% sched_debug.cfs_rq:/.util_avg.max
> 286.15 ± 7% -14.4% 244.91 ± 7% sched_debug.cfs_rq:/.util_est.avg
> 4282232 -17.3% 3541404 sched_debug.cfs_rq:/.zero_vruntime.avg
> 4735201 ± 2% -13.9% 4076017 ± 4% sched_debug.cfs_rq:/.zero_vruntime.max
> 4115256 -17.0% 3417248 ± 2% sched_debug.cfs_rq:/.zero_vruntime.min
> 4263217 -16.9% 3544690 sched_debug.cfs_rq:/init.scope.se->vruntime.avg
> 4263686 -16.8% 3545475 sched_debug.cfs_rq:/init.scope.se->vruntime.max
> 4262749 -16.9% 3543905 sched_debug.cfs_rq:/init.scope.se->vruntime.min
> 3220012 -14.8% 2744219 sched_debug.cfs_rq:/system.slice.avg_vruntime.avg
> 3545518 -10.9% 3160659 ± 3% sched_debug.cfs_rq:/system.slice.avg_vruntime.max
> 3158026 -15.2% 2679207 sched_debug.cfs_rq:/system.slice.avg_vruntime.min
> 448.40 ± 2% -9.4% 406.19 sched_debug.cfs_rq:/system.slice.runnable_avg.avg
> 448.42 ± 2% -9.4% 406.17 sched_debug.cfs_rq:/system.slice.se->avg.runnable_avg.avg
> 445.86 ± 2% -9.3% 404.18 sched_debug.cfs_rq:/system.slice.se->avg.util_avg.avg
> 1190 ± 4% -9.6% 1077 ± 3% sched_debug.cfs_rq:/system.slice.se->avg.util_avg.max
> 17612 -8.0% 16195 sched_debug.cfs_rq:/system.slice.se->sum_exec_runtime.avg
> 17072 -8.4% 15640 sched_debug.cfs_rq:/system.slice.se->sum_exec_runtime.min
> 4282329 -17.3% 3541763 sched_debug.cfs_rq:/system.slice.se->vruntime.avg
> 4734174 ± 2% -13.9% 4074191 ± 4% sched_debug.cfs_rq:/system.slice.se->vruntime.max
> 4115719 -17.0% 3417711 ± 2% sched_debug.cfs_rq:/system.slice.se->vruntime.min
> 445.87 ± 2% -9.3% 404.21 sched_debug.cfs_rq:/system.slice.util_avg.avg
> 1190 ± 4% -9.6% 1077 ± 3% sched_debug.cfs_rq:/system.slice.util_avg.max
> 3220012 -14.8% 2744218 sched_debug.cfs_rq:/system.slice.zero_vruntime.avg
> 3545518 -10.9% 3160649 ± 3% sched_debug.cfs_rq:/system.slice.zero_vruntime.max
> 3158026 -15.2% 2679207 sched_debug.cfs_rq:/system.slice.zero_vruntime.min
> 3201153 -14.7% 2731311 sched_debug.cfs_rq:/system.slice/containerd.service.se->vruntime.avg
> 3249048 -13.6% 2808484 ± 3% sched_debug.cfs_rq:/system.slice/containerd.service.se->vruntime.max
> 3153595 -14.8% 2685843 sched_debug.cfs_rq:/system.slice/containerd.service.se->vruntime.min
> 56.08 ± 17% -29.3% 39.67 ± 9% sched_debug.cfs_rq:/system.slice/containerd.service.tg_load_avg.avg
> 128.83 ± 17% -25.5% 96.00 ± 24% sched_debug.cfs_rq:/system.slice/containerd.service.tg_load_avg.max
> 25.42 ± 31% -34.4% 16.67 ± 30% sched_debug.cfs_rq:/system.slice/containerd.service.tg_load_avg.min
> 27.90 ± 11% -35.1% 18.10 ± 26% sched_debug.cfs_rq:/system.slice/containerd.service.tg_load_avg.stddev
> 17409 -8.0% 16022 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.avg_vruntime.avg
> 16929 -8.3% 15524 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.avg_vruntime.min
> 157.14 ± 19% -60.4% 62.15 ± 57% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.left_deadline.avg
> 17300 -23.2% 13285 ± 44% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.left_deadline.max
> 1631 ± 9% -45.1% 895.75 ± 47% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.left_deadline.stddev
> 157.13 ± 19% -60.5% 62.14 ± 57% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.left_vruntime.avg
> 17298 -23.2% 13283 ± 44% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.left_vruntime.max
> 1631 ± 9% -45.1% 895.61 ± 47% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.left_vruntime.stddev
> 504.25 ± 3% -8.3% 462.57 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.load_avg.avg
> 157.13 ± 19% -60.5% 62.14 ± 57% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.right_vruntime.avg
> 17298 -23.2% 13283 ± 44% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.right_vruntime.max
> 1631 ± 9% -45.1% 895.61 ± 47% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.right_vruntime.stddev
> 449.13 ± 2% -9.2% 407.68 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.runnable_avg.avg
> 448.89 ± 2% -9.3% 407.10 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->avg.runnable_avg.avg
> 446.27 ± 2% -9.2% 405.04 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->avg.util_avg.avg
> 1190 ± 4% -9.5% 1078 ± 3% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->avg.util_avg.max
> 17428 -8.0% 16038 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->sum_exec_runtime.avg
> 16954 -8.3% 15542 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->sum_exec_runtime.min
> 3220022 -14.8% 2744222 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->vruntime.avg
> 3545518 -10.9% 3160659 ± 3% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->vruntime.max
> 3158026 -15.2% 2679207 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.se->vruntime.min
> 117383 -8.5% 107420 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.tg_load_avg.avg
> 93409 ± 4% -12.8% 81492 ± 2% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.tg_load_avg.min
> 11146 ± 12% +23.7% 13782 ± 4% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.tg_load_avg.stddev
> 504.19 ± 2% -7.4% 467.07 ± 2% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.tg_load_avg_contrib.avg
> 446.38 ± 2% -9.2% 405.51 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.util_avg.avg
> 1191 ± 4% -9.5% 1078 ± 3% sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.util_avg.max
> 17409 -8.0% 16021 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.zero_vruntime.avg
> 16929 -8.3% 15524 sched_debug.cfs_rq:/system.slice/lkp-bootstrap.service.zero_vruntime.min
> 3219034 -15.3% 2727012 sched_debug.cfs_rq:/system.slice/redis-server.service.se->vruntime.avg
> 3233948 -15.3% 2739777 sched_debug.cfs_rq:/system.slice/redis-server.service.se->vruntime.max
> 3203580 -15.4% 2711785 sched_debug.cfs_rq:/system.slice/redis-server.service.se->vruntime.min
>
>
>
>
> 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.
>
>
Thanks for the report.
The large increase in major faults does look related to this mmap_miss
accounting change. The patch avoids counting all fault-around-installed
neighbouring PTEs as mmap readahead hits, but this workload may have
useful spatial locality where that feedback matters.
I will look into this and try to narrow down a suitable fix.
Thanks,
fujunjie