From: Yujie Liu <yujie.liu@intel.com>
To: Josh Poimboeuf <jpoimboe@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
<oe-lkp@lists.linux.dev>, <lkp@intel.com>,
<linux-kernel@vger.kernel.org>,
Thomas Gleixner <tglx@linutronix.de>,
Daniel Sneddon <daniel.sneddon@linux.intel.com>,
<ying.huang@intel.com>, <feng.tang@intel.com>,
<fengwei.yin@intel.com>
Subject: Re: [linus:master] [x86/syscall] 1e3ad78334: will-it-scale.per_process_ops 1.4% improvement
Date: Mon, 22 Apr 2024 15:41:27 +0800 [thread overview]
Message-ID: <ZiYUpxCTyh1uEgeO@yujie-X299> (raw)
In-Reply-To: <20240419073346.xlpx4qaocbo6bhip@treble>
Hi Josh,
On Fri, Apr 19, 2024 at 12:33:46AM -0700, Josh Poimboeuf wrote:
> On Fri, Apr 19, 2024 at 01:49:26PM +0800, kernel test robot wrote:
> > Hi Linus,
> >
> > We noticed that commit 1e3ad78334a6 caused performance fluctuations in
> > various micro benchmarks. The perf stat metrics related with branch
> > instructions do have noticeable changes, which may be an expected
> > result of this commit. We are sending this report to provide these data
> > and hope it can be helpful for the awareness of overall impact or any
> > further investigation. Thanks.
> >
> > kernel test robot noticed a 1.4% improvement of will-it-scale.per_process_ops on:
> >
> > commit: 1e3ad78334a69b36e107232e337f9d693dcc9df2 ("x86/syscall: Don't force use of indirect calls for system calls")
> > https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
>
> Thanks, these are significant regressions.
First we need to clarify that by running this specific will-it-scale
futex4 benchmark on a Skylake machine, we observed a +1.4% performance
improvement, not a regression.
> Since this is on Skylake (with IBRS enabled, presumably) I'd expect that
> these regressions are fixed by my "Only harden syscalls when needed"
> patch. I'm planning on posting a new version of that tomorrow, but v3
> [*] should be good enough to fix it. Could you run these tests on the
> same Skylake system with my patch added?
The v3 patch [*] cannot be applied on commit 1e3ad78334a6. Seems the
code base has changed a lot, so we are not able to directly compare
1e3ad78334a6 and 1e3ad78334a6+v3_patch.
The patch is good to apply on v6.9-rc4, so we tested v6.9-rc4 and
v6.9-rc4+v3_patch. Here are the test results for your reference:
Skylake
=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/mode/test/cpufreq_governor:
lkp-skl-fpga01/will-it-scale/debian-12-x86_64-20240206.cgz/x86_64-rhel-8.3/gcc-13/16/process/futex4/performance
commit:
0cd01ac5dcb1 ("x86/bugs: Change commas to semicolons in 'spectre_v2' sysfs file")
1e3ad78334a6 ("x86/syscall: Don't force use of indirect calls for system calls")
v6.9-rc4
v6.9-rc4+v3_patch
0cd01ac5dcb1 1e3ad78334a6 v6.9-rc4 v6.9-rc4+v3_patch
---------------- --------------------------- --------------------------- ---------------------------
%stddev %change %stddev %change %stddev %change %stddev
\ | \ | \ | \
1362315 +1.4% 1381406 +1.5% 1382652 +0.5% 1369778 will-it-scale.per_process_ops
21797058 +1.4% 22102512 +1.5% 22122442 +0.5% 21916453 will-it-scale.workload
0.04 ± 7% -7.4% 0.04 -6.1% 0.04 ± 2% -4.0% 0.04 perf-stat.i.MPKI
1.98e+09 +19.2% 2.36e+09 +19.2% 2.359e+09 +1.7% 2.014e+09 perf-stat.i.branch-instructions
1.47 -1.2 0.30 -1.2 0.30 ± 3% -0.0 1.45 perf-stat.i.branch-miss-rate%
30820475 -70.4% 9118612 -71.0% 8945551 +0.5% 30985854 perf-stat.i.branch-misses
7767463 -1.2% 7676829 -1.0% 7686158 -1.3% 7664542 perf-stat.i.cache-references
3.45 -4.4% 3.30 -4.4% 3.30 -0.4% 3.43 perf-stat.i.cpi
1.504e+10 +5.1% 1.58e+10 +5.2% 1.582e+10 +1.2% 1.522e+10 perf-stat.i.instructions
0.29 +4.5% 0.31 +4.6% 0.31 +0.4% 0.29 perf-stat.i.ipc
1.01 ±100% -0.6% 1.00 ±100% +104.1% 2.06 +0.3% 1.01 ±100% perf-stat.i.metric.K/sec
0.05 ± 2% -4.2% 0.04 -3.9% 0.04 ± 2% +0.4% 0.05 perf-stat.overall.MPKI
1.56 -1.2 0.39 -1.2 0.38 -0.0 1.54 perf-stat.overall.branch-miss-rate%
3.43 -4.3% 3.28 -4.4% 3.28 -0.5% 3.41 perf-stat.overall.cpi
0.29 +4.5% 0.30 +4.6% 0.30 +0.5% 0.29 perf-stat.overall.ipc
208138 +3.4% 215312 +3.5% 215474 +0.5% 209279 perf-stat.overall.path-length
1.973e+09 +19.2% 2.353e+09 +19.1% 2.351e+09 +1.8% 2.008e+09 perf-stat.ps.branch-instructions
30729762 -70.4% 9109071 -71.0% 8918595 +0.6% 30911752 perf-stat.ps.branch-misses
7745419 -1.1% 7663567 -1.1% 7663740 -1.3% 7647834 perf-stat.ps.cache-references
1.499e+10 +5.1% 1.575e+10 +5.2% 1.577e+10 +1.2% 1.517e+10 perf-stat.ps.instructions
4.537e+12 +4.9% 4.759e+12 +5.1% 4.767e+12 +1.1% 4.587e+12 perf-stat.total.instructions
12.23 -0.6 11.60 -0.6 11.64 -0.0 12.21 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
10.09 -0.6 9.51 -0.5 9.56 -0.1 10.01 perf-profile.calltrace.cycles-pp.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
22.31 -0.4 21.88 -0.4 21.94 +0.0 22.36 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.syscall
19.15 +0.2 19.30 +0.2 19.38 -0.1 19.04 perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.syscall
9.25 +0.2 9.43 +0.0 9.25 -0.0 9.23 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.syscall
8.79 +0.2 9.02 +0.3 9.07 -0.1 8.72 perf-profile.calltrace.cycles-pp.do_futex.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
7.13 +0.2 7.36 +0.3 7.41 -0.1 7.07 perf-profile.calltrace.cycles-pp.__futex_wait.futex_wait.do_futex.__x64_sys_futex.do_syscall_64
8.37 +0.3 8.63 +0.3 8.68 -0.1 8.28 perf-profile.calltrace.cycles-pp.futex_wait.do_futex.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe
12.38 -0.6 11.78 -0.5 11.84 -0.0 12.38 perf-profile.children.cycles-pp.do_syscall_64
10.12 -0.5 9.57 -0.5 9.63 -0.1 10.04 perf-profile.children.cycles-pp.__x64_sys_futex
22.63 -0.4 22.20 -0.4 22.24 +0.0 22.65 perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
0.48 ± 2% -0.0 0.46 -0.0 0.47 ± 2% -0.0 0.46 perf-profile.children.cycles-pp.get_futex_key
19.34 +0.1 19.49 +0.2 19.57 -0.1 19.24 perf-profile.children.cycles-pp.syscall_return_via_sysret
0.00 +0.2 0.18 ± 2% +0.2 0.18 ± 3% +0.0 0.00 perf-profile.children.cycles-pp.x64_sys_call
9.11 +0.2 9.29 +0.0 9.12 +0.0 9.12 perf-profile.children.cycles-pp.entry_SYSCALL_64
8.88 +0.2 9.11 +0.3 9.16 -0.1 8.81 perf-profile.children.cycles-pp.do_futex
7.13 +0.2 7.36 +0.3 7.41 -0.1 7.07 perf-profile.children.cycles-pp.__futex_wait
8.43 +0.3 8.70 +0.3 8.75 -0.1 8.34 perf-profile.children.cycles-pp.futex_wait
1.20 -0.7 0.47 -0.7 0.46 ± 3% -0.0 1.20 ± 2% perf-profile.self.cycles-pp.__x64_sys_futex
1.46 -0.2 1.27 -0.2 1.26 ± 2% +0.0 1.48 ± 2% perf-profile.self.cycles-pp.do_syscall_64
0.51 -0.1 0.44 -0.1 0.45 ± 2% +0.0 0.52 perf-profile.self.cycles-pp.do_futex
0.38 ± 5% -0.1 0.32 ± 4% -0.1 0.32 ± 5% +0.0 0.39 ± 7% perf-profile.self.cycles-pp.syscall_exit_to_user_mode
0.48 ± 2% -0.0 0.45 -0.0 0.45 ± 2% -0.0 0.46 ± 2% perf-profile.self.cycles-pp.get_futex_key
1.21 +0.0 1.24 ± 2% +0.0 1.23 ± 3% -0.0 1.18 perf-profile.self.cycles-pp.futex_wait
0.09 ± 14% +0.0 0.12 ± 8% +0.0 0.13 ± 6% +0.0 0.12 ± 5% perf-profile.self.cycles-pp.syscall_exit_to_user_mode_prepare
0.00 +0.1 0.15 ± 2% +0.2 0.15 ± 3% +0.0 0.00 perf-profile.self.cycles-pp.x64_sys_call
7.97 +0.1 8.12 -0.0 7.95 -0.0 7.96 perf-profile.self.cycles-pp.entry_SYSCALL_64
19.28 +0.2 19.44 +0.2 19.53 -0.1 19.21 perf-profile.self.cycles-pp.syscall_return_via_sysret
10.43 +0.2 10.60 +0.2 10.59 +0.0 10.46 perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
0.72 ± 3% +0.2 0.94 ± 3% +0.2 0.93 ± 4% +0.0 0.74 perf-profile.self.cycles-pp.__futex_wait
> Also it would be helpful to see the same tests on Cascade/Ice Lake, or
> some other system for which the 'spectre_v2' sysfs vulnerabilities file
> shows "BHI: SW loop". On such a system it shouldn't matter whether my
> patch is added as it won't disable Linus' syscall change. But it would
> be very helpful to see the performance impact of that combination.
The test results on Cascade/Ice Lake are as follows:
Intel Xeon Platinum 8260L (Cascade Lake)
=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/mode/test/cpufreq_governor:
lkp-csl-2sp3/will-it-scale/debian-12-x86_64-20240206.cgz/x86_64-rhel-8.3/gcc-13/16/process/futex4/performance
commit:
0cd01ac5dcb1 ("x86/bugs: Change commas to semicolons in 'spectre_v2' sysfs file")
1e3ad78334a6 ("x86/syscall: Don't force use of indirect calls for system calls")
v6.9-rc4
v6.9-rc4+v3_patch
0cd01ac5dcb1 1e3ad78334a6 v6.9-rc4 v6.9-rc4+v3_patch
---------------- --------------------------- --------------------------- ---------------------------
%stddev %change %stddev %change %stddev %change %stddev
\ | \ | \ | \
3237910 -0.3% 3229309 -10.1% 2911031 -11.0% 2882769 will-it-scale.per_process_ops
51806565 -0.3% 51668961 -10.1% 46576504 -11.0% 46124311 will-it-scale.workload
0.02 ± 7% -6.4% 0.02 ± 3% -9.5% 0.02 ± 2% -2.4% 0.02 ± 12% perf-stat.i.MPKI
4.649e+09 +17.4% 5.459e+09 +76.1% 8.186e+09 +75.4% 8.156e+09 perf-stat.i.branch-instructions
0.72 -0.6 0.15 ± 4% -0.6 0.12 -0.6 0.12 ± 2% perf-stat.i.branch-miss-rate%
34188248 -74.0% 8872232 ± 3% -69.9% 10285664 -70.0% 10244122 perf-stat.i.branch-misses
1.70 -4.2% 1.63 -8.0% 1.56 -8.3% 1.56 perf-stat.i.cpi
3.326e+10 +3.6% 3.444e+10 +9.1% 3.628e+10 +8.2% 3.599e+10 perf-stat.i.instructions
0.59 +4.3% 0.61 +8.7% 0.64 +9.0% 0.64 perf-stat.i.ipc
0.18 ± 16% -11.5% 0.16 ± 22% -33.9% 0.12 ± 46% -58.6% 0.08 ± 49% perf-stat.i.major-faults
0.02 ± 7% -6.3% 0.02 ± 4% -11.0% 0.02 ± 3% -2.3% 0.02 ± 13% perf-stat.overall.MPKI
0.74 -0.6 0.16 ± 3% -0.6 0.13 -0.6 0.13 perf-stat.overall.branch-miss-rate%
1.70 -4.1% 1.63 -8.0% 1.56 -8.3% 1.56 perf-stat.overall.cpi
0.59 +4.3% 0.61 +8.7% 0.64 +9.0% 0.64 perf-stat.overall.ipc
193210 +3.9% 200708 +21.4% 234594 +21.5% 234812 perf-stat.overall.path-length
4.633e+09 +17.4% 5.441e+09 +76.1% 8.159e+09 +75.4% 8.129e+09 perf-stat.ps.branch-instructions
34084869 -74.0% 8860998 ± 2% -69.9% 10274305 -70.0% 10220106 perf-stat.ps.branch-misses
3.315e+10 +3.6% 3.433e+10 +9.1% 3.616e+10 +8.2% 3.587e+10 perf-stat.ps.instructions
0.18 ± 16% -11.5% 0.16 ± 22% -33.8% 0.12 ± 46% -58.6% 0.08 ± 49% perf-stat.ps.major-faults
1.001e+13 +3.6% 1.037e+13 +9.2% 1.093e+13 +8.2% 1.083e+13 perf-stat.total.instructions
18.55 -0.3 18.23 -1.1 17.45 -1.1 17.46 perf-profile.calltrace.cycles-pp.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
1.82 -0.1 1.74 -0.2 1.60 -0.2 1.57 perf-profile.calltrace.cycles-pp.futex_q_unlock.futex_wait_setup.__futex_wait.futex_wait.do_futex
3.58 -0.1 3.51 -0.5 3.11 -0.5 3.11 perf-profile.calltrace.cycles-pp.futex_get_value_locked.futex_wait_setup.__futex_wait.futex_wait.do_futex
17.39 -0.1 17.32 -1.1 16.32 -1.1 16.30 perf-profile.calltrace.cycles-pp.do_futex.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
0.68 ± 2% -0.0 0.66 -0.1 0.60 ± 2% -0.1 0.60 ± 2% perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_safe_stack.syscall
2.73 -0.0 2.72 -0.3 2.40 -0.3 2.40 perf-profile.calltrace.cycles-pp.__get_user_nocheck_4.futex_get_value_locked.futex_wait_setup.__futex_wait.futex_wait
0.60 ± 2% +0.0 0.60 ± 2% -0.0 0.57 -0.0 0.59 ± 2% perf-profile.calltrace.cycles-pp.hrtimer_interrupt.__sysvec_apic_timer_interrupt.sysvec_apic_timer_interrupt.asm_sysvec_apic_timer_interrupt.cpuidle_enter_state
0.00 +0.0 0.00 +6.3 6.26 +6.2 6.22 perf-profile.calltrace.cycles-pp.clear_bhb_loop.syscall
0.61 ± 2% +0.0 0.61 ± 2% -0.0 0.58 -0.0 0.60 ± 2% perf-profile.calltrace.cycles-pp.__sysvec_apic_timer_interrupt.sysvec_apic_timer_interrupt.asm_sysvec_apic_timer_interrupt.cpuidle_enter_state.cpuidle_enter
72.70 +0.0 72.72 +0.1 72.78 +0.1 72.80 perf-profile.calltrace.cycles-pp.syscall
1.78 +0.0 1.80 -0.2 1.59 -0.2 1.58 perf-profile.calltrace.cycles-pp._raw_spin_lock.futex_q_lock.futex_wait_setup.__futex_wait.futex_wait
16.67 +0.0 16.71 -1.1 15.61 -1.1 15.59 perf-profile.calltrace.cycles-pp.futex_wait.do_futex.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe
20.50 +0.0 20.55 -0.5 20.04 -0.5 20.01 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
21.76 +0.0 21.81 -0.5 21.22 -0.5 21.24 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.syscall
2.07 +0.1 2.13 -0.2 1.90 -0.2 1.91 perf-profile.calltrace.cycles-pp.futex_hash.futex_q_lock.futex_wait_setup.__futex_wait.futex_wait
0.76 +0.1 0.84 ± 3% -0.0 0.74 -0.0 0.74 perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
5.09 +0.1 5.17 -0.5 4.60 -0.5 4.62 perf-profile.calltrace.cycles-pp.futex_q_lock.futex_wait_setup.__futex_wait.futex_wait.do_futex
7.85 +0.1 7.94 -0.7 7.10 -0.8 7.07 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.syscall
0.91 ± 2% +0.1 1.04 +0.0 0.92 +0.0 0.92 perf-profile.calltrace.cycles-pp.get_futex_key.futex_wait_setup.__futex_wait.futex_wait.do_futex
39.86 +0.2 40.02 -4.4 35.46 -4.4 35.48 perf-profile.calltrace.cycles-pp.entry_SYSRETQ_unsafe_stack.syscall
14.13 +0.2 14.35 -0.9 13.20 -1.0 13.18 perf-profile.calltrace.cycles-pp.__futex_wait.futex_wait.do_futex.__x64_sys_futex.do_syscall_64
12.44 +0.3 12.70 -1.1 11.33 -1.1 11.34 perf-profile.calltrace.cycles-pp.futex_wait_setup.__futex_wait.futex_wait.do_futex.__x64_sys_futex
18.62 -0.3 18.30 -1.1 17.57 -1.0 17.58 perf-profile.children.cycles-pp.__x64_sys_futex
17.59 -0.1 17.46 -1.0 16.54 -1.1 16.52 perf-profile.children.cycles-pp.do_futex
1.82 -0.1 1.74 -0.2 1.60 -0.2 1.57 perf-profile.children.cycles-pp.futex_q_unlock
3.19 -0.1 3.13 -0.4 2.77 -0.4 2.77 perf-profile.children.cycles-pp.__get_user_nocheck_4
0.68 ± 2% -0.0 0.66 -0.1 0.60 ± 2% -0.1 0.60 ± 2% perf-profile.children.cycles-pp.entry_SYSCALL_64_safe_stack
3.58 -0.0 3.57 -0.4 3.16 -0.4 3.16 perf-profile.children.cycles-pp.futex_get_value_locked
0.80 -0.0 0.79 ± 4% -0.0 0.76 ± 2% -0.1 0.75 ± 3% perf-profile.children.cycles-pp.hrtimer_interrupt
0.81 -0.0 0.80 ± 4% -0.0 0.77 ± 2% -0.0 0.76 ± 3% perf-profile.children.cycles-pp.__sysvec_apic_timer_interrupt
0.46 ± 2% -0.0 0.45 ± 5% -0.0 0.43 ± 2% -0.0 0.42 ± 4% perf-profile.children.cycles-pp.tick_nohz_handler
0.35 -0.0 0.34 -0.0 0.30 ± 3% -0.0 0.30 ± 2% perf-profile.children.cycles-pp.testcase
0.66 -0.0 0.65 ± 4% -0.0 0.62 -0.0 0.62 ± 2% perf-profile.children.cycles-pp.__hrtimer_run_queues
0.38 ± 2% -0.0 0.38 ± 5% -0.0 0.36 -0.0 0.35 ± 5% perf-profile.children.cycles-pp.update_process_times
0.14 ± 5% -0.0 0.14 ± 5% -0.0 0.12 ± 4% -0.0 0.12 ± 5% perf-profile.children.cycles-pp.amd_clear_divider
0.00 +0.0 0.00 +6.3 6.32 +6.3 6.29 perf-profile.children.cycles-pp.clear_bhb_loop
0.28 ± 6% +0.0 0.28 ± 4% -0.0 0.25 ± 3% -0.0 0.24 ± 2% perf-profile.children.cycles-pp.syscall_exit_to_user_mode_prepare
0.90 +0.0 0.91 ± 3% -0.1 0.80 -0.1 0.81 perf-profile.children.cycles-pp.syscall_exit_to_user_mode
1.88 +0.0 1.90 ± 2% -0.2 1.69 -0.2 1.68 perf-profile.children.cycles-pp._raw_spin_lock
0.17 ± 4% +0.0 0.20 ± 2% -0.1 0.12 ± 3% -0.1 0.12 ± 7% perf-profile.children.cycles-pp.futex_setup_timer
20.64 +0.0 20.69 -0.5 20.18 -0.4 20.21 perf-profile.children.cycles-pp.do_syscall_64
21.90 +0.0 21.94 -0.5 21.41 -0.5 21.43 perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
2.07 +0.1 2.13 -0.2 1.90 -0.2 1.91 perf-profile.children.cycles-pp.futex_hash
5.13 +0.1 5.20 -0.5 4.64 -0.5 4.64 perf-profile.children.cycles-pp.entry_SYSCALL_64
16.84 +0.1 16.91 -1.1 15.73 -1.1 15.71 perf-profile.children.cycles-pp.futex_wait
5.30 +0.1 5.37 -0.5 4.79 -0.5 4.80 perf-profile.children.cycles-pp.futex_q_lock
0.91 ± 2% +0.1 1.05 +0.0 0.92 +0.0 0.92 perf-profile.children.cycles-pp.get_futex_key
42.79 +0.2 42.98 -4.6 38.15 -4.6 38.18 perf-profile.children.cycles-pp.entry_SYSRETQ_unsafe_stack
14.13 +0.2 14.36 -0.9 13.20 -1.0 13.18 perf-profile.children.cycles-pp.__futex_wait
12.58 +0.3 12.83 -1.1 11.44 -1.1 11.44 perf-profile.children.cycles-pp.futex_wait_setup
0.00 +0.4 0.41 ± 2% +0.6 0.56 ± 2% +0.6 0.58 ± 3% perf-profile.children.cycles-pp.x64_sys_call
4.04 -0.3 3.77 -0.7 3.36 -0.7 3.33 perf-profile.self.cycles-pp.syscall
1.03 ± 2% -0.3 0.76 ± 2% -0.0 1.00 -0.0 1.02 perf-profile.self.cycles-pp.__x64_sys_futex
0.88 -0.3 0.62 +0.0 0.91 +0.0 0.91 perf-profile.self.cycles-pp.do_futex
2.50 -0.1 2.42 -0.2 2.30 -0.2 2.31 perf-profile.self.cycles-pp.futex_wait
1.74 -0.1 1.68 -0.2 1.55 -0.2 1.52 perf-profile.self.cycles-pp.futex_q_unlock
3.18 -0.1 3.12 -0.4 2.76 -0.4 2.76 perf-profile.self.cycles-pp.__get_user_nocheck_4
0.54 -0.1 0.48 ± 3% -0.1 0.43 -0.1 0.44 ± 3% perf-profile.self.cycles-pp.syscall_exit_to_user_mode
1.48 -0.0 1.45 ± 2% +0.2 1.69 ± 2% +0.2 1.67 perf-profile.self.cycles-pp.__futex_wait
0.68 ± 2% -0.0 0.66 -0.1 0.60 ± 2% -0.1 0.60 ± 2% perf-profile.self.cycles-pp.entry_SYSCALL_64_safe_stack
0.35 -0.0 0.34 -0.0 0.30 ± 3% -0.0 0.30 ± 2% perf-profile.self.cycles-pp.testcase
0.00 +0.0 0.00 +6.3 6.26 +6.2 6.22 perf-profile.self.cycles-pp.clear_bhb_loop
1.33 +0.0 1.33 -0.1 1.23 -0.1 1.22 perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
1.44 +0.0 1.44 -0.1 1.30 -0.1 1.30 perf-profile.self.cycles-pp.futex_q_lock
1.03 +0.0 1.05 ± 2% +0.2 1.22 +0.3 1.28 perf-profile.self.cycles-pp.do_syscall_64
1.80 +0.0 1.84 ± 2% -0.2 1.62 -0.2 1.62 perf-profile.self.cycles-pp._raw_spin_lock
2.42 +0.0 2.46 -0.2 2.19 -0.2 2.20 perf-profile.self.cycles-pp.entry_SYSCALL_64
0.38 ± 6% +0.1 0.44 +0.0 0.38 -0.0 0.38 ± 3% perf-profile.self.cycles-pp.futex_get_value_locked
2.00 +0.1 2.06 -0.2 1.83 -0.2 1.84 perf-profile.self.cycles-pp.futex_hash
0.21 ± 6% +0.1 0.28 ± 4% +0.0 0.25 ± 3% +0.0 0.24 ± 2% perf-profile.self.cycles-pp.syscall_exit_to_user_mode_prepare
1.11 ± 3% +0.1 1.22 ± 3% -0.0 1.08 ± 2% -0.0 1.10 perf-profile.self.cycles-pp.futex_wait_setup
0.90 ± 2% +0.1 1.04 +0.0 0.92 +0.0 0.92 ± 2% perf-profile.self.cycles-pp.get_futex_key
42.61 +0.2 42.81 -4.6 38.00 -4.6 38.02 perf-profile.self.cycles-pp.entry_SYSRETQ_unsafe_stack
0.00 +0.4 0.41 ± 2% +0.6 0.55 ± 2% +0.5 0.52 ± 3% perf-profile.self.cycles-pp.x64_sys_call
Intel Xeon Gold 6346 (Ice Lake)
=========================================================================================
tbox_group/testcase/rootfs/kconfig/compiler/nr_task/mode/test/cpufreq_governor:
lkp-icl-2sp9/will-it-scale/debian-12-x86_64-20240206.cgz/x86_64-rhel-8.3/gcc-13/16/process/futex4/performance
commit: 0cd01ac5dcb1 ("x86/bugs: Change commas to semicolons in 'spectre_v2' sysfs file") 1e3ad78334a6 ("x86/syscall: Don't force use of indirect calls for system calls") v6.9-rc4 v6.9-rc4+v3_patch
0cd01ac5dcb1 1e3ad78334a6 v6.9-rc4 v6.9-rc4+v3_patch
---------------- --------------------------- --------------------------- ---------------------------
%stddev %change %stddev %change %stddev %change %stddev
\ | \ | \ | \
7907214 -1.8% 7763496 -15.4% 6686457 -15.5% 6678350 will-it-scale.per_process_ops
1.265e+08 -1.8% 1.242e+08 -15.4% 1.07e+08 -15.5% 1.069e+08 will-it-scale.workload
1.112e+10 +16.0% 1.29e+10 +67.3% 1.86e+10 +68.0% 1.868e+10 perf-stat.i.branch-instructions
0.06 ± 2% -0.0 0.06 ± 2% -0.0 0.05 -0.0 0.05 perf-stat.i.branch-miss-rate%
6858604 ± 2% +0.6% 6900573 ± 2% +8.4% 7434422 +7.7% 7388238 perf-stat.i.branch-misses
0.72 -2.0% 0.71 -2.7% 0.70 -2.7% 0.70 perf-stat.i.cpi
8.004e+10 +2.1% 8.17e+10 +2.8% 8.231e+10 +2.8% 8.232e+10 perf-stat.i.instructions
1.38 +2.1% 1.41 +2.8% 1.42 +2.8% 1.42 perf-stat.i.ipc
0.06 ± 2% -0.0 0.05 ± 2% -0.0 0.04 -0.0 0.04 perf-stat.overall.branch-miss-rate%
0.72 -2.0% 0.71 -2.8% 0.70 -2.8% 0.70 perf-stat.overall.cpi
1.38 +2.1% 1.41 +2.8% 1.42 +2.8% 1.42 perf-stat.overall.ipc
190470 +3.9% 197929 +21.7% 231786 +21.8% 231973 perf-stat.overall.path-length
1.108e+10 +16.0% 1.286e+10 +67.3% 1.854e+10 +68.0% 1.862e+10 perf-stat.ps.branch-instructions
6893534 ± 2% +0.5% 6924998 ± 2% +8.3% 7462919 +7.5% 7410265 perf-stat.ps.branch-misses
7.978e+10 +2.1% 8.143e+10 +2.8% 8.204e+10 +2.8% 8.205e+10 perf-stat.ps.instructions
2.41e+13 +2.0% 2.459e+13 +2.9% 2.48e+13 +2.9% 2.479e+13 perf-stat.total.instructions
48.06 -2.8 45.31 -9.9 38.20 ± 3% -9.1 38.94 perf-profile.calltrace.cycles-pp.__futex_wait.futex_wait.do_futex.__x64_sys_futex.do_syscall_64
42.93 -2.5 40.41 -8.8 34.12 ± 4% -8.1 34.84 perf-profile.calltrace.cycles-pp.futex_wait_setup.__futex_wait.futex_wait.do_futex.__x64_sys_futex
56.45 -2.4 54.10 -12.0 44.44 -11.4 45.05 perf-profile.calltrace.cycles-pp.futex_wait.do_futex.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe
58.78 -2.3 56.48 -12.5 46.31 -11.9 46.86 perf-profile.calltrace.cycles-pp.do_futex.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
61.14 -2.3 58.86 -12.9 48.20 -12.5 48.67 perf-profile.calltrace.cycles-pp.__x64_sys_futex.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
71.08 -1.4 69.71 -12.1 58.96 -12.1 58.95 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.syscall
68.07 -1.2 66.88 -11.8 56.28 -11.8 56.26 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
17.13 -1.1 16.06 -3.8 13.38 ± 7% -2.9 14.20 ± 5% perf-profile.calltrace.cycles-pp.futex_q_lock.futex_wait_setup.__futex_wait.futex_wait.do_futex
100.03 -0.8 99.22 -1.0 99.01 -1.4 98.60 perf-profile.calltrace.cycles-pp.syscall
15.22 -0.8 14.42 -2.7 12.53 ± 5% -2.7 12.52 perf-profile.calltrace.cycles-pp.futex_get_value_locked.futex_wait_setup.__futex_wait.futex_wait.do_futex
12.02 -0.6 11.37 -2.1 9.89 ± 6% -2.1 9.92 perf-profile.calltrace.cycles-pp.__get_user_nocheck_4.futex_get_value_locked.futex_wait_setup.__futex_wait.futex_wait
3.12 ± 9% -0.5 2.61 -0.9 2.22 ± 10% -1.0 2.08 ± 7% perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
7.38 ± 2% -0.4 6.99 -1.9 5.44 ± 5% -1.6 5.76 ± 5% perf-profile.calltrace.cycles-pp.futex_hash.futex_q_lock.futex_wait_setup.__futex_wait.futex_wait
5.12 -0.3 4.78 -0.9 4.20 ± 10% -0.6 4.47 ± 5% perf-profile.calltrace.cycles-pp._raw_spin_lock.futex_q_lock.futex_wait_setup.__futex_wait.futex_wait
4.99 ± 2% -0.3 4.66 -1.0 4.00 ± 5% -1.2 3.79 ± 6% perf-profile.calltrace.cycles-pp.futex_q_unlock.futex_wait_setup.__futex_wait.futex_wait.do_futex
3.02 ± 3% -0.2 2.80 -0.9 2.17 -0.8 2.25 ± 3% perf-profile.calltrace.cycles-pp.get_futex_key.futex_wait_setup.__futex_wait.futex_wait.do_futex
1.58 ± 3% -0.1 1.51 -0.3 1.29 ± 10% -0.4 1.22 ± 7% perf-profile.calltrace.cycles-pp.syscall_return_via_sysret.syscall
0.94 ± 3% -0.1 0.87 -0.2 0.75 ± 9% -0.2 0.70 ± 8% perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode_prepare.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
0.69 ± 3% -0.0 0.65 -0.2 0.47 ± 46% -0.3 0.36 ± 71% perf-profile.calltrace.cycles-pp.amd_clear_divider.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
0.68 ± 3% -0.0 0.66 -0.3 0.38 ± 71% -0.2 0.45 ± 45% perf-profile.calltrace.cycles-pp.entry_SYSRETQ_unsafe_stack.syscall
0.00 +0.0 0.00 +15.2 15.21 ± 2% +14.8 14.83 perf-profile.calltrace.cycles-pp.clear_bhb_loop.syscall
1.04 ± 2% +0.1 1.13 ± 2% -0.1 0.96 ± 3% -0.0 1.00 ± 5% perf-profile.calltrace.cycles-pp.testcase
1.57 +0.1 1.70 -0.1 1.48 ± 9% -0.0 1.54 ± 5% perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_safe_stack.syscall
0.00 +1.3 1.29 +1.6 1.62 ± 8% +1.3 1.34 ± 4% perf-profile.calltrace.cycles-pp.x64_sys_call.do_syscall_64.entry_SYSCALL_64_after_hwframe.syscall
16.18 +1.6 17.80 -0.6 15.63 ± 7% +0.0 16.22 ± 6% perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.syscall
48.52 -2.8 45.75 -10.0 38.57 ± 2% -9.2 39.30 perf-profile.children.cycles-pp.__futex_wait
44.20 -2.6 41.60 -9.1 35.12 ± 3% -8.4 35.84 perf-profile.children.cycles-pp.futex_wait_setup
57.11 -2.4 54.75 -12.1 44.98 -11.5 45.58 perf-profile.children.cycles-pp.futex_wait
59.22 -2.3 56.91 -12.7 46.54 -12.1 47.10 perf-profile.children.cycles-pp.do_futex
61.79 -2.3 59.51 -13.0 48.74 -12.6 49.19 perf-profile.children.cycles-pp.__x64_sys_futex
69.05 -1.5 67.59 -12.1 56.90 -12.2 56.85 perf-profile.children.cycles-pp.do_syscall_64
71.36 -1.4 70.00 -11.9 59.44 -11.9 59.43 perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
17.82 -1.1 16.70 -3.9 13.94 ± 7% -3.0 14.79 ± 5% perf-profile.children.cycles-pp.futex_q_lock
14.54 -0.8 13.76 -2.6 11.96 ± 5% -2.6 11.98 perf-profile.children.cycles-pp.futex_get_value_locked
13.16 -0.7 12.46 -2.3 10.83 ± 5% -2.3 10.84 perf-profile.children.cycles-pp.__get_user_nocheck_4
3.96 ± 3% -0.5 3.46 -1.0 2.95 ± 10% -1.2 2.78 ± 7% perf-profile.children.cycles-pp.syscall_exit_to_user_mode
7.61 ± 2% -0.4 7.21 -2.0 5.62 ± 4% -1.6 5.96 ± 5% perf-profile.children.cycles-pp.futex_hash
5.35 -0.4 5.00 -1.0 4.40 ± 10% -0.7 4.67 ± 5% perf-profile.children.cycles-pp._raw_spin_lock
5.22 ± 2% -0.3 4.88 -1.0 4.19 ± 5% -1.3 3.97 ± 6% perf-profile.children.cycles-pp.futex_q_unlock
3.26 ± 3% -0.2 3.02 -0.9 2.35 ± 2% -0.8 2.44 ± 3% perf-profile.children.cycles-pp.get_futex_key
98.50 -0.1 98.40 +0.1 98.60 +0.1 98.55 perf-profile.children.cycles-pp.syscall
1.81 ± 3% -0.1 1.73 -0.3 1.47 ± 10% -0.4 1.40 ± 7% perf-profile.children.cycles-pp.syscall_return_via_sysret
1.17 ± 3% -0.1 1.09 -0.2 0.94 ± 9% -0.3 0.87 ± 8% perf-profile.children.cycles-pp.syscall_exit_to_user_mode_prepare
0.93 ± 3% -0.1 0.86 ± 2% -0.2 0.73 ± 10% -0.2 0.70 ± 8% perf-profile.children.cycles-pp.amd_clear_divider
0.16 ± 2% -0.0 0.15 -0.0 0.15 ± 5% -0.0 0.15 ± 4% perf-profile.children.cycles-pp.__sysvec_apic_timer_interrupt
0.16 ± 3% -0.0 0.14 ± 3% -0.0 0.14 ± 3% -0.0 0.14 ± 5% perf-profile.children.cycles-pp.hrtimer_interrupt
9.15 ± 2% -0.0 9.14 -1.5 7.65 ± 7% -1.6 7.54 ± 3% perf-profile.children.cycles-pp.entry_SYSRETQ_unsafe_stack
0.65 ± 4% -0.0 0.64 -0.1 0.54 ± 7% -0.1 0.54 ± 2% perf-profile.children.cycles-pp.futex_setup_timer
0.00 +0.0 0.00 +15.4 15.40 ± 2% +15.0 15.01 perf-profile.children.cycles-pp.clear_bhb_loop
0.62 ± 2% +0.0 0.66 ± 2% -0.1 0.56 ± 5% -0.0 0.58 ± 5% perf-profile.children.cycles-pp.syscall@plt
1.45 +0.1 1.57 -0.1 1.33 ± 2% -0.1 1.39 ± 5% perf-profile.children.cycles-pp.testcase
1.59 +0.1 1.72 -0.1 1.49 ± 9% -0.0 1.56 ± 5% perf-profile.children.cycles-pp.entry_SYSCALL_64_safe_stack
9.45 +0.9 10.37 -0.3 9.12 ± 7% +0.0 9.45 ± 5% perf-profile.children.cycles-pp.entry_SYSCALL_64
0.00 +1.5 1.51 +1.8 1.81 ± 8% +1.5 1.51 ± 4% perf-profile.children.cycles-pp.x64_sys_call
12.90 -0.7 12.23 -2.3 10.63 ± 5% -2.3 10.64 perf-profile.self.cycles-pp.__get_user_nocheck_4
7.38 ± 2% -0.4 6.99 -1.9 5.43 ± 5% -1.6 5.76 ± 5% perf-profile.self.cycles-pp.futex_hash
5.09 -0.4 4.70 -1.0 4.11 ± 8% -0.7 4.35 ± 6% perf-profile.self.cycles-pp.futex_q_lock
5.11 -0.3 4.78 -0.9 4.21 ± 10% -0.6 4.47 ± 5% perf-profile.self.cycles-pp._raw_spin_lock
4.86 ± 2% -0.3 4.54 -0.9 3.92 ± 5% -1.1 3.71 ± 7% perf-profile.self.cycles-pp.futex_q_unlock
3.68 -0.2 3.45 -0.1 3.62 ± 9% -0.1 3.56 ± 7% perf-profile.self.cycles-pp.do_syscall_64
3.02 ± 3% -0.2 2.80 -0.9 2.16 ± 2% -0.8 2.25 ± 3% perf-profile.self.cycles-pp.get_futex_key
4.33 ± 3% -0.2 4.14 -0.9 3.45 ± 6% -0.9 3.45 ± 2% perf-profile.self.cycles-pp.__futex_wait
4.17 -0.2 3.99 ± 3% -0.9 3.32 -0.9 3.32 perf-profile.self.cycles-pp.futex_wait_setup
2.09 ± 3% -0.2 1.93 -0.4 1.64 ± 9% -0.5 1.56 ± 7% perf-profile.self.cycles-pp.syscall_exit_to_user_mode
1.39 ± 2% -0.1 1.29 -0.3 1.13 ± 2% -0.3 1.12 ± 2% perf-profile.self.cycles-pp.futex_get_value_locked
1.81 ± 3% -0.1 1.73 -0.3 1.47 ± 10% -0.4 1.40 ± 7% perf-profile.self.cycles-pp.syscall_return_via_sysret
0.94 ± 3% -0.1 0.88 -0.2 0.75 ± 10% -0.2 0.70 ± 8% perf-profile.self.cycles-pp.syscall_exit_to_user_mode_prepare
0.46 ± 2% -0.0 0.43 ± 2% -0.1 0.37 ± 11% -0.1 0.34 ± 6% perf-profile.self.cycles-pp.amd_clear_divider
0.43 ± 4% +0.0 0.43 ± 2% -0.1 0.36 ± 7% -0.1 0.36 ± 2% perf-profile.self.cycles-pp.futex_setup_timer
0.00 +0.0 0.00 +15.2 15.22 ± 2% +14.8 14.81 perf-profile.self.cycles-pp.clear_bhb_loop
8.92 ± 2% +0.0 8.92 -1.4 7.47 ± 7% -1.6 7.36 ± 2% perf-profile.self.cycles-pp.entry_SYSRETQ_unsafe_stack
0.20 +0.0 0.22 ± 3% -0.0 0.18 ± 8% -0.0 0.20 ± 7% perf-profile.self.cycles-pp.syscall@plt
2.76 +0.0 2.81 -0.4 2.38 ± 10% -0.5 2.27 ± 7% perf-profile.self.cycles-pp.__x64_sys_futex
1.90 ± 3% +0.0 1.95 -0.7 1.23 ± 11% -0.7 1.22 ± 8% perf-profile.self.cycles-pp.do_futex
2.50 ± 2% +0.1 2.61 +0.1 2.56 ± 8% +0.1 2.60 ± 4% perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
1.24 +0.1 1.35 ± 2% -0.1 1.14 ± 3% -0.0 1.19 ± 5% perf-profile.self.cycles-pp.testcase
1.59 +0.1 1.72 -0.1 1.49 ± 9% -0.0 1.56 ± 5% perf-profile.self.cycles-pp.entry_SYSCALL_64_safe_stack
2.72 +0.2 2.94 ± 2% -0.1 2.61 ± 9% -0.0 2.69 ± 6% perf-profile.self.cycles-pp.entry_SYSCALL_64
8.15 ± 3% +0.4 8.56 -2.0 6.18 ± 10% -2.1 6.04 ± 3% perf-profile.self.cycles-pp.futex_wait
11.95 +1.0 12.92 -1.0 10.97 ± 4% -0.6 11.35 ± 4% perf-profile.self.cycles-pp.syscall
0.00 +1.3 1.30 +1.6 1.62 ± 8% +1.3 1.33 ± 4% perf-profile.self.cycles-pp.x64_sys_call
BTW, we did observe some regressions by running other benchmarks on
commit 1e3ad78334a6, but these regressions are on Ice Lake, not Skylake.
Please kindly contact us if you are interested in looking into them.
stress-ng.null.ops_per_sec -4.0% regression on Intel Xeon Gold 6346 (Ice Lake)
unixbench.fsbuffer.throughput -1.4% regression on Intel Xeon Gold 6346 (Ice Lake)
Thanks,
Yujie
>
> [*] https://lkml.kernel.org/lkml/eda0ec65f4612cc66875aaf76e738643f41fbc01.1713296762.git.jpoimboe@kernel.org
>
> --
> Josh
>
prev parent reply other threads:[~2024-04-22 7:48 UTC|newest]
Thread overview: 3+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-04-19 5:49 [linus:master] [x86/syscall] 1e3ad78334: will-it-scale.per_process_ops 1.4% improvement kernel test robot
2024-04-19 7:33 ` Josh Poimboeuf
2024-04-22 7:41 ` Yujie Liu [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=ZiYUpxCTyh1uEgeO@yujie-X299 \
--to=yujie.liu@intel.com \
--cc=daniel.sneddon@linux.intel.com \
--cc=feng.tang@intel.com \
--cc=fengwei.yin@intel.com \
--cc=jpoimboe@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=lkp@intel.com \
--cc=oe-lkp@lists.linux.dev \
--cc=tglx@linutronix.de \
--cc=torvalds@linux-foundation.org \
--cc=ying.huang@intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).