[Bug]spurious PEBS NMI triggered by non-precise events

[XiongZhang]

The following perf command can easily trigger the pebs warning or spurious NMI error on Skylake/Broadwell/Haswell platforms.
sudo perf record -e 'cpu/umask=0x04,event=0xc4/pp,cycles,branches,ref-cycles,cache-misses,cache-references' --call-graph fp -b -c1000 -a
Also NMI watchdog must be enabled to reproduce this issue.
Here is the dump.
[ 113.452176] Call Trace:
[ 113.452178] <NMI> [<ffffffff813c3a2e>] dump_stack+0x63/0x85
[ 113.452188] [<ffffffff810a46f2>] warn_slowpath_common+0x82/0xc0
[ 113.452190] [<ffffffff810a483a>] warn_slowpath_null+0x1a/0x20
[ 113.452193] [<ffffffff8100fe2e>] intel_pmu_drain_pebs_nhm+0x2be/0x320
[ 113.452197] [<ffffffff8100caa9>] intel_pmu_handle_irq+0x279/0x460
[ 113.452204] [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
[ 113.452208] [<ffffffff811f290d>] ? vunmap_page_range+0x20d/0x330
[ 113.452211] [<ffffffff811f2f11>] ? unmap_kernel_range_noflush+0x11/0x20
[ 113.452216] [<ffffffff8148379f>] ? ghes_copy_tofrom_phys+0x10f/0x2a0
[ 113.452218] [<ffffffff814839c8>] ? ghes_read_estatus+0x98/0x170
[ 113.452224] [<ffffffff81005a7d>] perf_event_nmi_handler+0x2d/0x50
[ 113.452230] [<ffffffff810310b9>] nmi_handle+0x69/0x120
[ 113.452233] [<ffffffff810316f6>] default_do_nmi+0xe6/0x100
[ 113.452236] [<ffffffff810317f2>] do_nmi+0xe2/0x130
[ 113.452240] [<ffffffff817aea71>] end_repeat_nmi+0x1a/0x1e
[ 113.452243] [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
[ 113.452246] [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
[ 113.452249] [<ffffffff810639b6>] ? native_write_msr_safe+0x6/0x40
[ 113.452250] <<EOE>> <IRQ> [<ffffffff81006df8>] ? x86_perf_event_set_period+0xd8/0x180
[ 113.452255] [<ffffffff81006eec>] x86_pmu_start+0x4c/0x100
[ 113.452258] [<ffffffff8100722d>] x86_pmu_enable+0x28d/0x300
[ 113.452263] [<ffffffff811994d7>] perf_pmu_enable.part.81+0x7/0x10
[ 113.452267] [<ffffffff8119cb70>] perf_mux_hrtimer_handler+0x200/0x280
[ 113.452270] [<ffffffff8119c970>] ? __perf_install_in_context+0xc0/0xc0
[ 113.452273] [<ffffffff8110f92d>] __hrtimer_run_queues+0xfd/0x280
[ 113.452276] [<ffffffff811100d8>] hrtimer_interrupt+0xa8/0x190
[ 113.452278] [<ffffffff81199080>] ? __perf_read_group_add.part.61+0x1a0/0x1a0
[ 113.452283] [<ffffffff81051bd8>] local_apic_timer_interrupt+0x38/0x60
[ 113.452286] [<ffffffff817af01d>] smp_apic_timer_interrupt+0x3d/0x50
[ 113.452290] [<ffffffff817ad15c>] apic_timer_interrupt+0x8c/0xa0
[ 113.452291] <EOI> [<ffffffff81199080>] ? __perf_read_group_add.part.61+0x1a0/0x1a0
[ 113.452298] [<ffffffff81123de5>] ? smp_call_function_single+0xd5/0x130
[ 113.452300] [<ffffffff81123ddb>] ? smp_call_function_single+0xcb/0x130
[ 113.452303] [<ffffffff81199080>] ? __perf_read_group_add.part.61+0x1a0/0x1a0
[ 113.452306] [<ffffffff8119765a>] event_function_call+0x10a/0x120
[ 113.452308] [<ffffffff8119c660>] ? ctx_resched+0x90/0x90
[ 113.452311] [<ffffffff811971e0>] ? cpu_clock_event_read+0x30/0x30
[ 113.452313] [<ffffffff811976d0>] ? _perf_event_disable+0x60/0x60
[ 113.452315] [<ffffffff8119772b>] _perf_event_enable+0x5b/0x70
[ 113.452318] [<ffffffff81197388>] perf_event_for_each_child+0x38/0xa0
[ 113.452320] [<ffffffff811976d0>] ? _perf_event_disable+0x60/0x60
[ 113.452322] [<ffffffff811a0ffd>] perf_ioctl+0x12d/0x3c0
[ 113.452326] [<ffffffff8134d855>] ? selinux_file_ioctl+0x95/0x1e0
[ 113.452330] [<ffffffff8124a3a1>] do_vfs_ioctl+0xa1/0x5a0
[ 113.452334] [<ffffffff81036d29>] ? sched_clock+0x9/0x10
[ 113.452336] [<ffffffff8124a919>] SyS_ioctl+0x79/0x90
[ 113.452338] [<ffffffff817ac4b2>] entry_SYSCALL_64_fastpath+0x1a/0xa4
[ 113.452340] --[ end trace aef202839fe9a71d ]--
[ 113.452611] Uhhuh. NMI received for unknown reason 2d on CPU 2.
[ 113.453880] Do you have a strange power saving mode enabled?

[XiongZhang (xiong-y-zhang)]

One commit from v4.6 fix this issue, please back port it to 16.04
c3d266c perf/x86/intel: Fix PEBS warning by only restoring active PMU in pmi

[Tim Gardner (timg-tpi)]

I think this will have to wait for a 16.10 HWE kernel. The number of prerequisite commits are just too many at this late stage.

[Leann Ogasawara (leannogasawara)]

The v4.8 based kernel has been uploaded to the Yakkety 16.10 archive. I'm setting this to Fix Released.

git describe --contains c3d266c
v4.6-rc1~165^2~8

commit c3d266c8a9838cc141b69548bc3b1b18808ae8c4
Author: Kan Liang <email address hidden>
Date: Thu Mar 3 18:07:28 2016 -0500

    perf/x86/intel: Fix PEBS warning by only restoring active PMU in pmi