tree: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf/urgent
head: 60f52ab61c7dc0a991125903ae06a35d1812698c
commit: 6d3edaae16c6c7d238360f2841212c2b26774d5e [4/5] x86/perf/amd: Resolve NMI latency issues for active PMCs
config: i386-tinyconfig (attached as .config)
compiler: gcc-7 (Debian 7.3.0-1) 7.3.0
reproduce:
git checkout 6d3edaae16c6c7d238360f2841212c2b26774d5e
# save the attached .config to linux build tree
make ARCH=i386
All error/warnings (new ones prefixed by >>):
arch/x86//events/amd/core.c: In function 'amd_pmu_handle_irq':
arch/x86//events/amd/core.c:538:10: error: 'NMI_DONE' undeclared (first use in this function); did you mean 'EM_NONE'?
return NMI_DONE;
^~~~~~~~
EM_NONE
arch/x86//events/amd/core.c:538:10: note: each undeclared identifier is reported only once for each function it appears in
>> arch/x86//events/amd/core.c:542:9: error: 'NMI_HANDLED' undeclared (first use in this function); did you mean 'IRQ_HANDLED'?
return NMI_HANDLED;
^~~~~~~~~~~
IRQ_HANDLED
>> arch/x86//events/amd/core.c:543:1: warning: control reaches end of non-void function [-Wreturn-type]
}
^
vim +542 arch/x86//events/amd/core.c
493
494 /*
495 * Because of NMI latency, if multiple PMC counters are active or other sources
496 * of NMIs are received, the perf NMI handler can handle one or more overflowed
497 * PMC counters outside of the NMI associated with the PMC overflow. If the NMI
498 * doesn't arrive at the LAPIC in time to become a pending NMI, then the kernel
499 * back-to-back NMI support won't be active. This PMC handler needs to take into
500 * account that this can occur, otherwise this could result in unknown NMI
501 * messages being issued. Examples of this is PMC overflow while in the NMI
502 * handler when multiple PMCs are active or PMC overflow while handling some
503 * other source of an NMI.
504 *
505 * Attempt to mitigate this by using the number of active PMCs to determine
506 * whether to return NMI_HANDLED if the perf NMI handler did not handle/reset
507 * any PMCs. The per-CPU perf_nmi_counter variable is set to a minimum of the
508 * number of active PMCs or 2. The value of 2 is used in case an NMI does not
509 * arrive at the LAPIC in time to be collapsed into an already pending NMI.
510 */
511 static int amd_pmu_handle_irq(struct pt_regs *regs)
512 {
513 struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
514 int active, handled;
515
516 /*
517 * Obtain the active count before calling x86_pmu_handle_irq() since
518 * it is possible that x86_pmu_handle_irq() may make a counter
519 * inactive (through x86_pmu_stop).
520 */
521 active = __bitmap_weight(cpuc->active_mask, X86_PMC_IDX_MAX);
522
523 /* Process any counter overflows */
524 handled = x86_pmu_handle_irq(regs);
525
526 /*
527 * If a counter was handled, record the number of possible remaining
528 * NMIs that can occur.
529 */
530 if (handled) {
531 this_cpu_write(perf_nmi_counter,
532 min_t(unsigned int, 2, active));
533
534 return handled;
535 }
536
537 if (!this_cpu_read(perf_nmi_counter))
> 538 return NMI_DONE;
539
540 this_cpu_dec(perf_nmi_counter);
541
> 542 return NMI_HANDLED;
> 543 }
544
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
On 4/3/19 9:47 AM, kbuild test robot wrote:
> tree: https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf/urgent
> head: 60f52ab61c7dc0a991125903ae06a35d1812698c
> commit: 6d3edaae16c6c7d238360f2841212c2b26774d5e [4/5] x86/perf/amd: Resolve NMI latency issues for active PMCs
> config: i386-tinyconfig (attached as .config)
> compiler: gcc-7 (Debian 7.3.0-1) 7.3.0
> reproduce:
> git checkout 6d3edaae16c6c7d238360f2841212c2b26774d5e
> # save the attached .config to linux build tree
> make ARCH=i386
>
> All error/warnings (new ones prefixed by >>):
>
> arch/x86//events/amd/core.c: In function 'amd_pmu_handle_irq':
> arch/x86//events/amd/core.c:538:10: error: 'NMI_DONE' undeclared (first use in this function); did you mean 'EM_NONE'?
> return NMI_DONE;
> ^~~~~~~~
> EM_NONE
> arch/x86//events/amd/core.c:538:10: note: each undeclared identifier is reported only once for each function it appears in
>>> arch/x86//events/amd/core.c:542:9: error: 'NMI_HANDLED' undeclared (first use in this function); did you mean 'IRQ_HANDLED'?
> return NMI_HANDLED;
> ^~~~~~~~~~~
> IRQ_HANDLED
>>> arch/x86//events/amd/core.c:543:1: warning: control reaches end of non-void function [-Wreturn-type]
> }
> ^
>
Looks like I need to change that include back to asm/nmi.h. I should have
looked closer at linux/nmi.h after the checkpatch message.
I'll send a patch to address it.
Thanks,
Tom
> vim +542 arch/x86//events/amd/core.c
>
> 493
> 494 /*
> 495 * Because of NMI latency, if multiple PMC counters are active or other sources
> 496 * of NMIs are received, the perf NMI handler can handle one or more overflowed
> 497 * PMC counters outside of the NMI associated with the PMC overflow. If the NMI
> 498 * doesn't arrive at the LAPIC in time to become a pending NMI, then the kernel
> 499 * back-to-back NMI support won't be active. This PMC handler needs to take into
> 500 * account that this can occur, otherwise this could result in unknown NMI
> 501 * messages being issued. Examples of this is PMC overflow while in the NMI
> 502 * handler when multiple PMCs are active or PMC overflow while handling some
> 503 * other source of an NMI.
> 504 *
> 505 * Attempt to mitigate this by using the number of active PMCs to determine
> 506 * whether to return NMI_HANDLED if the perf NMI handler did not handle/reset
> 507 * any PMCs. The per-CPU perf_nmi_counter variable is set to a minimum of the
> 508 * number of active PMCs or 2. The value of 2 is used in case an NMI does not
> 509 * arrive at the LAPIC in time to be collapsed into an already pending NMI.
> 510 */
> 511 static int amd_pmu_handle_irq(struct pt_regs *regs)
> 512 {
> 513 struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> 514 int active, handled;
> 515
> 516 /*
> 517 * Obtain the active count before calling x86_pmu_handle_irq() since
> 518 * it is possible that x86_pmu_handle_irq() may make a counter
> 519 * inactive (through x86_pmu_stop).
> 520 */
> 521 active = __bitmap_weight(cpuc->active_mask, X86_PMC_IDX_MAX);
> 522
> 523 /* Process any counter overflows */
> 524 handled = x86_pmu_handle_irq(regs);
> 525
> 526 /*
> 527 * If a counter was handled, record the number of possible remaining
> 528 * NMIs that can occur.
> 529 */
> 530 if (handled) {
> 531 this_cpu_write(perf_nmi_counter,
> 532 min_t(unsigned int, 2, active));
> 533
> 534 return handled;
> 535 }
> 536
> 537 if (!this_cpu_read(perf_nmi_counter))
> > 538 return NMI_DONE;
> 539
> 540 this_cpu_dec(perf_nmi_counter);
> 541
> > 542 return NMI_HANDLED;
> > 543 }
> 544
>
> ---
> 0-DAY kernel test infrastructure Open Source Technology Center
> https://lists.01.org/pipermail/kbuild-all Intel Corporation
>