Skip to contentSkip to conversation

e3e3698c7b: will-it-scale.per_thread_ops 2.6% improvement

jpoimboe:uaccess-hardening kernel test robot 1 participant 1 message Oct 28, 2024

lore
Share & tools
lore

kernel test robot

author build bot
#0 lore raw

Hello,

kernel test robot noticed a 2.6% improvement of will-it-scale.per_thread_ops on:

commit: e3e3698c7b891811bb1133cc7fde9a30c8b0e00a ("x86/uaccess: Avoid barrier_nospec() in copy_from_user()")
https://git.kernel.org/cgit/linux/kernel/git/jpoimboe/linux.git uaccess-hardening

testcase: will-it-scale
config: x86_64-rhel-8.3
compiler: gcc-12
test machine: 256 threads 2 sockets GENUINE INTEL(R) XEON(R) (Sierra Forest) with 128G memory parameters:

	nr_task: 100%
	mode: thread
	test: eventfd1
	cpufreq_governor: performance

Details are as below:
-------------------------------------------------------------------------------------------------->

The kernel config and materials to reproduce are available at:
https://download.01.org/0day-ci/archive/20241028/[email protected]

=========================================================================================
compiler/cpufreq_governor/kconfig/mode/nr_task/rootfs/tbox_group/test/testcase:
gcc-12/performance/x86_64-rhel-8.3/thread/100%/debian-12-x86_64-20240206.cgz/lkp-srf-2sp1/eventfd1/will-it-scale

commit:
v6.12-rc2
e3e3698c7b ("x86/uaccess: Avoid barrier_nospec() in copy_from_user()")

       v6.12-rc2 e3e3698c7b891811bb1133cc7fd 
---------------- --------------------------- 
         %stddev     %change         %stddev
             \          |                \  
 6.236e+08            +2.6%  6.395e+08        will-it-scale.256.threads
   2435801            +2.6%    2497931        will-it-scale.per_thread_ops
 6.236e+08            +2.6%  6.395e+08        will-it-scale.workload
 1.676e+11            +2.5%  1.718e+11        perf-stat.i.branch-instructions
      0.81            -2.6%       0.79        perf-stat.i.cpi
 8.218e+11            +2.7%  8.437e+11        perf-stat.i.instructions
      1.24            +2.7%       1.28        perf-stat.i.ipc
      0.80            -2.6%       0.78        perf-stat.overall.cpi
      1.24            +2.7%       1.28        perf-stat.overall.ipc
  1.67e+11            +2.5%  1.712e+11        perf-stat.ps.branch-instructions
 8.191e+11            +2.7%  8.409e+11        perf-stat.ps.instructions
 2.489e+14            +2.7%  2.557e+14        perf-stat.total.instructions
      9.35            -2.3        7.08        perf-profile.calltrace.cycles-pp._copy_from_user.eventfd_write.vfs_write.ksys_write.do_syscall_64
     12.94            -2.2       10.74        perf-profile.calltrace.cycles-pp.eventfd_write.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe
     15.55            -2.2       13.38        perf-profile.calltrace.cycles-pp.vfs_write.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe.write
     21.51            -2.0       19.47        perf-profile.calltrace.cycles-pp.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe.write
     24.17            -2.0       22.18        perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.write
     25.37            -2.0       23.40        perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.write
     53.54            -1.3       52.28        perf-profile.calltrace.cycles-pp.write
      1.01            +0.0        1.03        perf-profile.calltrace.cycles-pp.x64_sys_call.do_syscall_64.entry_SYSCALL_64_after_hwframe.write
      0.69            +0.0        0.71        perf-profile.calltrace.cycles-pp.syscall_exit_to_user_mode.do_syscall_64.entry_SYSCALL_64_after_hwframe.read
      1.00            +0.0        1.02        perf-profile.calltrace.cycles-pp.x64_sys_call.do_syscall_64.entry_SYSCALL_64_after_hwframe.read
      1.83            +0.1        1.88        perf-profile.calltrace.cycles-pp.fput.ksys_write.do_syscall_64.entry_SYSCALL_64_after_hwframe.write
      2.03            +0.1        2.09        perf-profile.calltrace.cycles-pp.entry_SYSRETQ_unsafe_stack.read
      1.78            +0.1        1.84        perf-profile.calltrace.cycles-pp.fput.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe.read
      2.03            +0.1        2.09        perf-profile.calltrace.cycles-pp.entry_SYSRETQ_unsafe_stack.write
      2.09            +0.1        2.15        perf-profile.calltrace.cycles-pp._raw_spin_lock_irq.eventfd_read.vfs_read.ksys_read.do_syscall_64

0.58 ± 4% +0.1 0.66 ± 2% perf-profile.calltrace.cycles-pp.testcase
15.59 +0.3 15.86 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.read
8.82 +0.3 9.16 perf-profile.calltrace.cycles-pp._copy_to_iter.eventfd_read.vfs_read.ksys_read.do_syscall_64
15.58 +0.4 15.94 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64.write
12.34 +0.4 12.74 perf-profile.calltrace.cycles-pp.eventfd_read.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe
19.36 +0.6 19.97 perf-profile.calltrace.cycles-pp.vfs_read.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe.read
25.33 +0.8 26.10 perf-profile.calltrace.cycles-pp.ksys_read.do_syscall_64.entry_SYSCALL_64_after_hwframe.read
27.97 +0.8 28.80 perf-profile.calltrace.cycles-pp.do_syscall_64.entry_SYSCALL_64_after_hwframe.read
29.16 +0.9 30.02 perf-profile.calltrace.cycles-pp.entry_SYSCALL_64_after_hwframe.read
57.34 +1.5 58.82 perf-profile.calltrace.cycles-pp.read
9.68 -2.3 7.42 perf-profile.children.cycles-pp._copy_from_user
13.11 -2.2 10.91 perf-profile.children.cycles-pp.eventfd_write
15.78 -2.2 13.61 perf-profile.children.cycles-pp.vfs_write
21.73 -2.0 19.69 perf-profile.children.cycles-pp.ksys_write
53.02 -1.3 51.70 perf-profile.children.cycles-pp.write
52.44 -1.2 51.29 perf-profile.children.cycles-pp.do_syscall_64
54.76 -1.1 53.64 perf-profile.children.cycles-pp.entry_SYSCALL_64_after_hwframe
0.32 ± 2% +0.0 0.35 ± 2% perf-profile.children.cycles-pp.read@plt
2.06 +0.0 2.10 perf-profile.children.cycles-pp.x64_sys_call
0.97 ± 3% +0.1 1.08 perf-profile.children.cycles-pp.testcase
3.66 +0.1 3.77 perf-profile.children.cycles-pp.fput
4.37 +0.1 4.49 perf-profile.children.cycles-pp._raw_spin_lock_irq
8.83 +0.3 9.17 perf-profile.children.cycles-pp._copy_to_iter
12.43 +0.4 12.84 perf-profile.children.cycles-pp.eventfd_read
20.48 +0.4 20.90 perf-profile.children.cycles-pp.entry_SYSCALL_64
16.41 +0.4 16.84 perf-profile.children.cycles-pp.entry_SYSRETQ_unsafe_stack
19.41 +0.6 20.02 perf-profile.children.cycles-pp.vfs_read
25.35 +0.8 26.12 perf-profile.children.cycles-pp.ksys_read
56.65 +1.5 58.12 perf-profile.children.cycles-pp.read
9.64 -2.3 7.38 perf-profile.self.cycles-pp._copy_from_user
0.70 +0.0 0.71 perf-profile.self.cycles-pp.ksys_read
0.73 +0.0 0.75 perf-profile.self.cycles-pp.rw_verify_area
2.01 +0.0 2.04 perf-profile.self.cycles-pp.x64_sys_call
1.66 +0.0 1.70 perf-profile.self.cycles-pp.do_syscall_64
2.03 +0.0 2.07 perf-profile.self.cycles-pp.entry_SYSCALL_64_after_hwframe
0.91 +0.1 0.98 perf-profile.self.cycles-pp.testcase
3.59 +0.1 3.71 perf-profile.self.cycles-pp.fput
4.20 +0.1 4.32 perf-profile.self.cycles-pp._raw_spin_lock_irq
8.08 +0.2 8.24 perf-profile.self.cycles-pp.write
5.59 +0.2 5.77 perf-profile.self.cycles-pp.vfs_read
9.74 +0.2 9.93 perf-profile.self.cycles-pp.entry_SYSCALL_64
8.08 +0.2 8.30 perf-profile.self.cycles-pp.read
8.77 +0.3 9.10 perf-profile.self.cycles-pp._copy_to_iter
16.36 +0.4 16.79 perf-profile.self.cycles-pp.entry_SYSRETQ_unsafe_stack

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.

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki