Re: [ucounts] 59ec71575a: will-it-scale.per_process_ops -10.3% regression
From: Eric W. Biederman
Date: Wed Jan 12 2022 - 14:09:14 EST
kernel test robot <oliver.sang@xxxxxxxxx> writes:
> Greeting,
>
> FYI, we noticed a -10.3% regression of will-it-scale.per_process_ops due to commit:
I am having a bit of a challenge reading this report. Does the -10.3%
mean that will-it-scale.per_process_ops is performing 70873 operations
per second instead of 78995 aka -10.3% fewer operations per second?
The per-profile.self-cycles-pp.do_dec_rlimits_put_ucounts now takes +2.6
more cycles, and perf-profile.self.cycles.pp.inc_rlimit_get_ucounts now
takes +3.8 more cycles.
Which if I read this correctly is a regression report about two
functions taking 3 or 4 cycles more after the code was rearranged inside
of them.
Is that correct? Is this a regression report saying those two functions
are taking a few cycles more which leads to
will-it-scale.per_process_ops not being able to perform as many
operations per second?
Given the change in the commit mentioned I think this is all down to a
bug fix causing the code to be rearranged and a compiler optimizations.
I don't see much room to do anything about this.
Eric
> commit: 59ec71575ab440cd5ca0aa53b2a2985b3639fad4 ("ucounts: Fix rlimit max values check")
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
>
> in testcase: will-it-scale
> on test machine: 144 threads 4 sockets Intel(R) Xeon(R) Gold 5318H CPU @ 2.50GHz with 128G memory
> with following parameters:
>
> nr_task: 100%
> mode: process
> test: signal1
> cpufreq_governor: performance
> ucode: 0x7002302
>
> test-description: Will It Scale takes a testcase and runs it from 1 through to n parallel copies to see if the testcase will scale. It builds both a process and threads based test in order to see any differences between the two.
> test-url: https://github.com/antonblanchard/will-it-scale
>
>
>
> If you fix the issue, kindly add following tag
> Reported-by: kernel test robot <oliver.sang@xxxxxxxxx>
>
>
> Details are as below:
> -------------------------------------------------------------------------------------------------->
>
>
> To reproduce:
>
> git clone https://github.com/intel/lkp-tests.git
> cd lkp-tests
> sudo bin/lkp install job.yaml # job file is attached in this email
> bin/lkp split-job --compatible job.yaml # generate the yaml file for lkp run
> sudo bin/lkp run generated-yaml-file
>
> # if come across any failure that blocks the test,
> # please remove ~/.lkp and /lkp dir to run from a clean state.
>
> =========================================================================================
> compiler/cpufreq_governor/kconfig/mode/nr_task/rootfs/tbox_group/test/testcase/ucode:
> gcc-9/performance/x86_64-rhel-8.3/process/100%/debian-10.4-x86_64-20200603.cgz/lkp-cpl-4sp1/signal1/will-it-scale/0x7002302
>
> commit:
> v5.16-rc2
> 59ec71575a ("ucounts: Fix rlimit max values check")
>
> v5.16-rc2 59ec71575ab440cd5ca0aa53b2a
> ---------------- ---------------------------
> %stddev %change %stddev
> \ | \
> 11375458 -10.3% 10205843 will-it-scale.144.processes
> 78995 -10.3% 70873 will-it-scale.per_process_ops
> 11375458 -10.3% 10205843 will-it-scale.workload
> 35.26 +4.6% 36.90 turbostat.RAMWatt
> 1.177e+10 -10.1% 1.058e+10 perf-stat.i.branch-instructions
> 87846548 -6.7% 81932988 perf-stat.i.branch-misses
> 43.42 +2.7 46.09 perf-stat.i.cache-miss-rate%
> 91079467 ± 2% +27.6% 1.162e+08 perf-stat.i.cache-misses
> 2.086e+08 +20.7% 2.517e+08 perf-stat.i.cache-references
> 7.77 +11.8% 8.68 perf-stat.i.cpi
> 5126 -21.1% 4044 perf-stat.i.cycles-between-cache-misses
> 1.814e+10 -10.6% 1.621e+10 ± 2% perf-stat.i.dTLB-loads
> 1.17e+10 -11.1% 1.04e+10 ± 2% perf-stat.i.dTLB-stores
> 53823400 -22.3% 41823499 ± 16% perf-stat.i.iTLB-load-misses
> 6e+10 -9.9% 5.405e+10 perf-stat.i.instructions
> 0.13 -10.5% 0.12 perf-stat.i.ipc
> 290.39 -10.4% 260.05 ± 2% perf-stat.i.metric.M/sec
> 5934502 ± 3% +38.4% 8213006 perf-stat.i.node-load-misses
> 504440 ± 3% +34.1% 676425 ± 2% perf-stat.i.node-loads
> 3.48 +34.0% 4.66 perf-stat.overall.MPKI
> 0.75 +0.0 0.77 ± 2% perf-stat.overall.branch-miss-rate%
> 43.63 +2.5 46.13 perf-stat.overall.cache-miss-rate%
> 7.78 +11.8% 8.70 perf-stat.overall.cpi
> 5128 -21.1% 4045 perf-stat.overall.cycles-between-cache-misses
> 0.13 -10.5% 0.11 perf-stat.overall.ipc
> 1.174e+10 -10.1% 1.055e+10 perf-stat.ps.branch-instructions
> 87583795 -6.8% 81664372 perf-stat.ps.branch-misses
> 90844134 ± 2% +27.5% 1.158e+08 perf-stat.ps.cache-misses
> 2.082e+08 +20.6% 2.511e+08 perf-stat.ps.cache-references
> 1.809e+10 -10.7% 1.616e+10 ± 2% perf-stat.ps.dTLB-loads
> 1.167e+10 -11.1% 1.037e+10 ± 2% perf-stat.ps.dTLB-stores
> 53794169 -22.4% 41750328 ± 16% perf-stat.ps.iTLB-load-misses
> 5.984e+10 -10.0% 5.388e+10 perf-stat.ps.instructions
> 5915765 ± 3% +38.4% 8188054 perf-stat.ps.node-load-misses
> 542552 ± 4% +30.8% 709510 ± 3% perf-stat.ps.node-loads
> 1.826e+13 -10.5% 1.634e+13 perf-stat.total.instructions
> 48.83 -8.5 40.29 perf-profile.calltrace.cycles-pp.security_task_kill.do_send_specific.do_tkill.__x64_sys_tgkill.do_syscall_64
> 48.72 -8.5 40.21 perf-profile.calltrace.cycles-pp.apparmor_task_kill.security_task_kill.do_send_specific.do_tkill.__x64_sys_tgkill
> 66.93 -4.7 62.21 perf-profile.calltrace.cycles-pp.__x64_sys_tgkill.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 66.86 -4.7 62.16 perf-profile.calltrace.cycles-pp.do_tkill.__x64_sys_tgkill.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 66.62 -4.7 61.93 perf-profile.calltrace.cycles-pp.do_send_specific.do_tkill.__x64_sys_tgkill.do_syscall_64.entry_SYSCALL_64_after_hwframe
> 21.22 -4.1 17.12 perf-profile.calltrace.cycles-pp.aa_get_task_label.apparmor_task_kill.security_task_kill.do_send_specific.do_tkill
> 95.18 -1.5 93.70 perf-profile.calltrace.cycles-pp.raise
> 2.11 -1.1 0.96 ±100% perf-profile.calltrace.cycles-pp.__setup_rt_frame.arch_do_signal_or_restart.exit_to_user_mode_prepare.syscall_exit_to_user_mode.do_syscall_64
> 1.70 -0.9 0.78 ±100% perf-profile.calltrace.cycles-pp.copy_fpstate_to_sigframe.__setup_rt_frame.arch_do_signal_or_restart.exit_to_user_mode_prepare.syscall_exit_to_user_mode
> 1.48 -0.8 0.70 ±100% perf-profile.calltrace.cycles-pp.__fpu_restore_sig.fpu__restore_sig.restore_sigcontext.__x64_sys_rt_sigreturn.do_syscall_64
> 92.89 -0.8 92.13 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.raise
> 92.54 -0.7 91.80 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 1.04 -0.5 0.50 ±100% perf-profile.calltrace.cycles-pp.restore_fpregs_from_user.__fpu_restore_sig.fpu__restore_sig.restore_sigcontext.__x64_sys_rt_sigreturn
> 2.38 ± 5% -0.4 1.96 ± 4% perf-profile.calltrace.cycles-pp.aa_may_signal.apparmor_task_kill.security_task_kill.do_send_specific.do_tkill
> 1.50 -0.4 1.11 ± 23% perf-profile.calltrace.cycles-pp.__entry_text_start.raise
> 1.56 -0.2 1.40 ± 5% perf-profile.calltrace.cycles-pp.restore_sigcontext.__x64_sys_rt_sigreturn.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 1.52 -0.2 1.36 ± 5% perf-profile.calltrace.cycles-pp.fpu__restore_sig.restore_sigcontext.__x64_sys_rt_sigreturn.do_syscall_64.entry_SYSCALL_64_after_hwframe
> 2.20 +0.3 2.54 ± 8% perf-profile.calltrace.cycles-pp.__x64_sys_rt_sigreturn.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 0.62 ± 3% +0.5 1.12 ± 26% perf-profile.calltrace.cycles-pp.restore_altstack.__x64_sys_rt_sigreturn.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 1.69 ± 2% +0.6 2.25 ± 9% perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.handler
> 1.70 ± 2% +0.6 2.26 ± 9% perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.handler
> 1.68 ± 2% +0.6 2.24 ± 9% perf-profile.calltrace.cycles-pp.exit_to_user_mode_prepare.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.handler
> 1.68 ± 2% +0.6 2.25 ± 9% perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.handler
> 0.00 +0.6 0.64 ± 15% perf-profile.calltrace.cycles-pp.__set_current_blocked.__x64_sys_rt_sigreturn.do_syscall_64.entry_SYSCALL_64_after_hwframe
> 0.00 +0.8 0.84 ± 36% perf-profile.calltrace.cycles-pp._raw_spin_lock_irq.do_sigaltstack.restore_altstack.__x64_sys_rt_sigreturn.do_syscall_64
> 0.00 +0.9 0.89 ± 34% perf-profile.calltrace.cycles-pp.do_sigaltstack.restore_altstack.__x64_sys_rt_sigreturn.do_syscall_64.entry_SYSCALL_64_after_hwframe
> 1.08 ± 4% +1.0 2.11 ± 4% perf-profile.calltrace.cycles-pp._raw_spin_lock_irq.__set_current_blocked.sigprocmask.__x64_sys_rt_sigprocmask.do_syscall_64
> 2.65 +1.3 3.90 ± 3% perf-profile.calltrace.cycles-pp.__x64_sys_rt_sigprocmask.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 4.24 +1.3 5.56 ± 4% perf-profile.calltrace.cycles-pp.handler
> 1.76 ± 2% +1.3 3.11 ± 4% perf-profile.calltrace.cycles-pp.sigprocmask.__x64_sys_rt_sigprocmask.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 1.70 ± 2% +1.4 3.06 ± 4% perf-profile.calltrace.cycles-pp.__set_current_blocked.sigprocmask.__x64_sys_rt_sigprocmask.do_syscall_64.entry_SYSCALL_64_after_hwframe
> 20.17 ± 2% +2.5 22.62 perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 19.97 ± 2% +2.5 22.47 perf-profile.calltrace.cycles-pp.exit_to_user_mode_prepare.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.raise
> 16.45 ± 2% +3.8 20.23 perf-profile.calltrace.cycles-pp.__send_signal.do_send_sig_info.do_send_specific.do_tkill.__x64_sys_tgkill
> 16.19 ± 2% +3.8 20.00 perf-profile.calltrace.cycles-pp.__sigqueue_alloc.__send_signal.do_send_sig_info.do_send_specific.do_tkill
> 15.60 ± 2% +3.8 19.44 perf-profile.calltrace.cycles-pp.inc_rlimit_get_ucounts.__sigqueue_alloc.__send_signal.do_send_sig_info.do_send_specific
> 17.39 ± 2% +3.9 21.29 perf-profile.calltrace.cycles-pp.do_send_sig_info.do_send_specific.do_tkill.__x64_sys_tgkill.do_syscall_64
> 48.84 -8.5 40.29 perf-profile.children.cycles-pp.security_task_kill
> 48.81 -8.5 40.27 perf-profile.children.cycles-pp.apparmor_task_kill
> 66.95 -4.7 62.22 perf-profile.children.cycles-pp.__x64_sys_tgkill
> 66.87 -4.7 62.16 perf-profile.children.cycles-pp.do_tkill
> 66.62 -4.7 61.93 perf-profile.children.cycles-pp.do_send_specific
> 21.31 ± 2% -4.1 17.17 perf-profile.children.cycles-pp.aa_get_task_label
> 95.20 -1.0 94.16 perf-profile.children.cycles-pp.raise
> 1.49 -0.6 0.91 ± 55% perf-profile.children.cycles-pp.__fpu_restore_sig
> 2.38 ± 5% -0.4 1.96 ± 4% perf-profile.children.cycles-pp.aa_may_signal
> 1.92 -0.2 1.73 ± 4% perf-profile.children.cycles-pp.restore_sigcontext
> 2.12 -0.2 1.93 perf-profile.children.cycles-pp.__setup_rt_frame
> 1.72 -0.2 1.56 perf-profile.children.cycles-pp.copy_fpstate_to_sigframe
> 1.52 -0.2 1.37 ± 4% perf-profile.children.cycles-pp.fpu__restore_sig
> 1.00 -0.1 0.89 ± 2% perf-profile.children.cycles-pp._copy_from_user
> 1.00 -0.1 0.88 perf-profile.children.cycles-pp.__entry_text_start
> 1.04 -0.1 0.93 ± 7% perf-profile.children.cycles-pp.restore_fpregs_from_user
> 0.89 -0.1 0.78 ± 2% perf-profile.children.cycles-pp.syscall_return_via_sysret
> 0.52 -0.1 0.46 ± 2% perf-profile.children.cycles-pp.copy_user_enhanced_fast_string
> 0.49 -0.1 0.44 perf-profile.children.cycles-pp.copy_user_generic_unrolled
> 0.64 -0.0 0.59 ± 5% perf-profile.children.cycles-pp.__might_fault
> 0.12 ± 4% -0.0 0.08 ± 27% perf-profile.children.cycles-pp.syscall_exit_to_user_mode_prepare
> 0.27 ± 2% -0.0 0.24 ± 2% perf-profile.children.cycles-pp.__clear_user
> 0.19 ± 2% -0.0 0.16 ± 10% perf-profile.children.cycles-pp.__might_sleep
> 0.25 -0.0 0.23 ± 2% perf-profile.children.cycles-pp.__get_user_nocheck_8
> 0.20 ± 3% -0.0 0.18 ± 2% perf-profile.children.cycles-pp.__put_user_nocheck_4
> 0.14 ± 3% -0.0 0.12 ± 3% perf-profile.children.cycles-pp.syscall_enter_from_user_mode
> 0.15 -0.0 0.14 ± 3% perf-profile.children.cycles-pp.entry_SYSCALL_64_safe_stack
> 0.15 ± 3% -0.0 0.13 ± 3% perf-profile.children.cycles-pp.__x64_sys_getpid
> 0.63 ± 5% +0.1 0.75 ± 7% perf-profile.children.cycles-pp._raw_spin_lock_irqsave
> 0.95 +0.1 1.08 ± 7% perf-profile.children.cycles-pp.fpu__clear_user_states
> 0.63 ± 4% +0.2 0.83 ± 11% perf-profile.children.cycles-pp.fpregs_mark_activate
> 1.20 +0.4 1.59 perf-profile.children.cycles-pp.native_irq_return_iret
> 0.66 ± 5% +0.4 1.08 ± 13% perf-profile.children.cycles-pp.signal_setup_done
> 3.37 +0.5 3.86 ± 2% perf-profile.children.cycles-pp.__x64_sys_rt_sigreturn
> 0.63 ± 3% +0.5 1.13 ± 26% perf-profile.children.cycles-pp.restore_altstack
> 0.35 ± 5% +0.5 0.89 ± 34% perf-profile.children.cycles-pp.do_sigaltstack
> 1.88 ± 4% +0.6 2.50 ± 2% perf-profile.children.cycles-pp.recalc_sigpending
> 3.17 +0.9 4.08 ± 5% perf-profile.children.cycles-pp.handler
> 2.67 +1.2 3.92 ± 3% perf-profile.children.cycles-pp.__x64_sys_rt_sigprocmask
> 1.77 ± 2% +1.3 3.12 ± 4% perf-profile.children.cycles-pp.sigprocmask
> 2.77 ± 2% +2.0 4.77 ± 7% perf-profile.children.cycles-pp.__set_current_blocked
> 2.49 ± 3% +2.0 4.54 perf-profile.children.cycles-pp._raw_spin_lock_irq
> 15.12 ± 2% +2.6 17.72 perf-profile.children.cycles-pp.do_dec_rlimit_put_ucounts
> 16.73 ± 2% +2.7 19.40 ± 2% perf-profile.children.cycles-pp.dequeue_signal
> 17.48 ± 2% +2.7 20.15 ± 2% perf-profile.children.cycles-pp.get_signal
> 21.90 ± 2% +3.0 24.92 perf-profile.children.cycles-pp.syscall_exit_to_user_mode
> 21.34 ± 2% +3.0 24.39 perf-profile.children.cycles-pp.arch_do_signal_or_restart
> 21.67 ± 2% +3.1 24.74 perf-profile.children.cycles-pp.exit_to_user_mode_prepare
> 16.48 ± 2% +3.8 20.25 perf-profile.children.cycles-pp.__send_signal
> 16.20 ± 2% +3.8 20.01 perf-profile.children.cycles-pp.__sigqueue_alloc
> 15.60 ± 2% +3.8 19.44 perf-profile.children.cycles-pp.inc_rlimit_get_ucounts
> 17.40 ± 2% +3.9 21.30 perf-profile.children.cycles-pp.do_send_sig_info
> 21.19 -4.1 17.07 perf-profile.self.cycles-pp.aa_get_task_label
> 24.97 -4.0 21.02 perf-profile.self.cycles-pp.apparmor_task_kill
> 0.81 -0.4 0.40 ± 85% perf-profile.self.cycles-pp.restore_fpregs_from_user
> 2.37 ± 5% -0.4 1.95 ± 4% perf-profile.self.cycles-pp.aa_may_signal
> 0.89 -0.1 0.78 ± 2% perf-profile.self.cycles-pp.syscall_return_via_sysret
> 0.88 -0.1 0.77 perf-profile.self.cycles-pp.raise
> 0.93 -0.1 0.84 ± 2% perf-profile.self.cycles-pp.copy_fpstate_to_sigframe
> 0.54 -0.1 0.47 ± 5% perf-profile.self.cycles-pp.fpu__clear_user_states
> 0.50 -0.1 0.44 ± 2% perf-profile.self.cycles-pp.copy_user_enhanced_fast_string
> 0.36 ± 2% -0.1 0.31 ± 8% perf-profile.self.cycles-pp.__setup_rt_frame
> 0.47 -0.1 0.42 perf-profile.self.cycles-pp.copy_user_generic_unrolled
> 0.16 ± 3% -0.0 0.11 ± 26% perf-profile.self.cycles-pp.kmem_cache_free
> 0.38 -0.0 0.34 ± 2% perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
> 0.43 -0.0 0.38 perf-profile.self.cycles-pp.__entry_text_start
> 0.22 -0.0 0.19 ± 3% perf-profile.self.cycles-pp.__x64_sys_rt_sigprocmask
> 0.17 ± 2% -0.0 0.14 ± 10% perf-profile.self.cycles-pp.__might_sleep
> 0.25 -0.0 0.22 ± 3% perf-profile.self.cycles-pp.__get_user_nocheck_8
> 0.19 ± 2% -0.0 0.17 ± 2% perf-profile.self.cycles-pp.__put_user_nocheck_4
> 0.16 ± 2% -0.0 0.14 ± 4% perf-profile.self.cycles-pp.__send_signal
> 0.17 ± 2% -0.0 0.15 ± 3% perf-profile.self.cycles-pp.__clear_user
> 0.22 ± 2% -0.0 0.20 ± 3% perf-profile.self.cycles-pp.kmem_cache_alloc
> 0.13 -0.0 0.11 ± 3% perf-profile.self.cycles-pp.syscall_enter_from_user_mode
> 0.15 -0.0 0.13 ± 3% perf-profile.self.cycles-pp.entry_SYSCALL_64_safe_stack
> 0.12 ± 2% -0.0 0.10 ± 4% perf-profile.self.cycles-pp._copy_from_user
> 0.10 ± 4% -0.0 0.09 ± 5% perf-profile.self.cycles-pp.handler
> 0.14 ± 3% -0.0 0.12 ± 5% perf-profile.self.cycles-pp.get_signal
> 0.11 -0.0 0.09 ± 5% perf-profile.self.cycles-pp.__might_fault
> 0.08 ± 5% -0.0 0.07 perf-profile.self.cycles-pp.__sigqueue_alloc
> 0.09 ± 3% -0.0 0.08 ± 4% perf-profile.self.cycles-pp.restore_sigcontext
> 0.63 ± 5% +0.1 0.75 ± 7% perf-profile.self.cycles-pp._raw_spin_lock_irqsave
> 0.60 ± 5% +0.2 0.81 ± 11% perf-profile.self.cycles-pp.fpregs_mark_activate
> 1.20 +0.4 1.59 perf-profile.self.cycles-pp.native_irq_return_iret
> 1.48 ± 5% +0.7 2.22 ± 14% perf-profile.self.cycles-pp.recalc_sigpending
> 2.48 ± 3% +2.0 4.53 perf-profile.self.cycles-pp._raw_spin_lock_irq
> 15.12 ± 2% +2.6 17.72 perf-profile.self.cycles-pp.do_dec_rlimit_put_ucounts
> 15.60 ± 2% +3.8 19.44 perf-profile.self.cycles-pp.inc_rlimit_get_ucounts
>
>
>
>
> 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.
>
>
> ---
> 0DAY/LKP+ Test Infrastructure Open Source Technology Center
> https://lists.01.org/hyperkitty/list/lkp@xxxxxxxxxxxx Intel Corporation
>
> Thanks,
> Oliver Sang