lkml.org 
[lkml]   [2010]   [Aug]   [25]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    Patch in this message
    /
    Date
    From
    SubjectRe: [PATCH -v3] perf, x86: try to handle unknown nmis with running perfctrs
    On 20.08.10 11:25:10, Ingo Molnar wrote:

    > > Ingo Molnar <mingo@elte.hu> wrote:
    > >
    > > >
    > > >it's not working so well, i'm getting:
    > > >
    > > > Uhhuh. NMI received for unknown reason 00 on CPU 9.
    > > > Do you have a strange power saving mode enabled?
    > > > Dazed and confused, but trying to continue
    > > >
    > > >on a nehalem box, after a perf top and perf stat run.
    >
    > FYI, it does not trigger on an AMD box.

    Ingo,

    do you mean it does not trigger false positives on AMD? Both patches
    applied on top of current tip/perf/urgent (c6db67c) are working on the
    systems I have.

    You might use the debug patch below for diagnostics.

    -Robert

    --

    From 1bbb5aa64e96360529c34a593a072e1a84114f04 Mon Sep 17 00:00:00 2001
    From: Robert Richter <robert.richter@amd.com>
    Date: Wed, 11 Aug 2010 18:14:00 +0200
    Subject: [PATCH] debug

    Signed-off-by: Robert Richter <robert.richter@amd.com>
    ---
    arch/x86/kernel/cpu/perf_event.c | 54 ++++++++++++++++++++++++++++++++++++-
    1 files changed, 52 insertions(+), 2 deletions(-)

    diff --git a/arch/x86/kernel/cpu/perf_event.c b/arch/x86/kernel/cpu/perf_event.c
    index dd2fceb..059ef09 100644
    --- a/arch/x86/kernel/cpu/perf_event.c
    +++ b/arch/x86/kernel/cpu/perf_event.c
    @@ -1203,10 +1203,43 @@ void perf_events_lapic_init(void)
    struct pmu_nmi_state {
    unsigned int marked;
    int handled;
    + u64 timestamp;
    };

    static DEFINE_PER_CPU(struct pmu_nmi_state, nmi);

    +struct nmi_debug {
    + int cpu;
    + unsigned int this_nmi;
    + unsigned int marked;
    + int handled;
    + u64 timestamp;
    + u64 delta;
    +};
    +
    +static DEFINE_PER_CPU(struct nmi_debug[16], nmi_debug);
    +
    +static void nmi_handler_debug(void)
    +{
    + struct nmi_debug *debug;
    + int i;
    +
    + if (!printk_ratelimit())
    + return;
    +
    + for (i = 0; i < 16; i++) {
    + debug = &__get_cpu_var(nmi_debug)[i];
    + printk(KERN_EMERG
    + "cpu #%d, nmi #%d, marked #%d, handled = %d, time = %llu, delta = %llu\n",
    + debug->cpu,
    + debug->this_nmi,
    + debug->marked,
    + debug->handled,
    + debug->timestamp,
    + debug->delta);
    + }
    +}
    +
    static int __kprobes
    perf_event_nmi_handler(struct notifier_block *self,
    unsigned long cmd, void *__args)
    @@ -1214,6 +1247,8 @@ perf_event_nmi_handler(struct notifier_block *self,
    struct die_args *args = __args;
    unsigned int this_nmi;
    int handled;
    + struct nmi_debug *debug;
    + u64 timestamp;

    if (!atomic_read(&active_events))
    return NOTIFY_DONE;
    @@ -1224,9 +1259,11 @@ perf_event_nmi_handler(struct notifier_block *self,
    break;
    case DIE_NMIUNKNOWN:
    this_nmi = percpu_read(irq_stat.__nmi_count);
    - if (this_nmi != __get_cpu_var(nmi).marked)
    + if (this_nmi != __get_cpu_var(nmi).marked) {
    + nmi_handler_debug();
    /* let the kernel handle the unknown nmi */
    return NOTIFY_DONE;
    + }
    /*
    * This one is a PMU back-to-back nmi. Two events
    * trigger 'simultaneously' raising two back-to-back
    @@ -1242,10 +1279,21 @@ perf_event_nmi_handler(struct notifier_block *self,
    apic_write(APIC_LVTPC, APIC_DM_NMI);

    handled = x86_pmu.handle_irq(args->regs);
    + this_nmi = percpu_read(irq_stat.__nmi_count);
    +
    + debug = &__get_cpu_var(nmi_debug)[0xf & this_nmi];
    + debug->cpu = smp_processor_id();
    + debug->this_nmi = this_nmi;
    + debug->marked = __get_cpu_var(nmi).marked;
    + debug->handled = handled;
    + rdtscll(timestamp);
    + debug->delta = timestamp - __get_cpu_var(nmi).timestamp;
    + __get_cpu_var(nmi).timestamp = timestamp;
    + debug->timestamp = timestamp;
    +
    if (!handled)
    return NOTIFY_DONE;

    - this_nmi = percpu_read(irq_stat.__nmi_count);
    if ((handled > 1) ||
    /* the next nmi could be a back-to-back nmi */
    ((__get_cpu_var(nmi).marked == this_nmi) &&
    @@ -1262,6 +1310,8 @@ perf_event_nmi_handler(struct notifier_block *self,
    */
    __get_cpu_var(nmi).marked = this_nmi + 1;
    __get_cpu_var(nmi).handled = handled;
    + debug->marked = __get_cpu_var(nmi).marked;
    + debug->handled = handled;
    }

    return NOTIFY_STOP;
    --
    1.7.1.1

    --
    Advanced Micro Devices, Inc.
    Operating System Research Center



    \
     
     \ /
      Last update: 2010-08-25 12:13    [W:0.056 / U:88.688 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site