lkml.org 
[lkml]   [2024]   [Jan]   [24]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH 1/3] watchdog/softlockup: low-overhead detection of interrupt storm
From


On 2024/1/24 09:43, Liu Song wrote:
>
> 在 2024/1/23 20:12, Bitao Hu 写道:
>> The following softlockup is caused by interrupt storm, but it cannot be
>> identified from the call tree. Because the call tree is just a snapshot
>> and doesn't fully capture the behavior of the CPU during the soft lockup.
>>    watchdog: BUG: soft lockup - CPU#28 stuck for 23s! [fio:83921]
>>    ...
>>    Call trace:
>>      __do_softirq+0xa0/0x37c
>>      __irq_exit_rcu+0x108/0x140
>>      irq_exit+0x14/0x20
>>      __handle_domain_irq+0x84/0xe0
>>      gic_handle_irq+0x80/0x108
>>      el0_irq_naked+0x50/0x58
>>
>> Therefore,I think it is necessary to report CPU utilization during the
>> softlockup_thresh period (report once every sample_period, for a total
>> of 5 reportings), like this:
>>    watchdog: BUG: soft lockup - CPU#28 stuck for 23s! [fio:83921]
>>    CPU#28 Utilization every 4s during lockup:
>>      #1: 0.0% system, 0.0% softirq, 100.0% hardirq, 0.0% idle
>>      #2: 0.0% system, 0.0% softirq, 100.0% hardirq, 0.0% idle
>>      #3: 0.0% system, 0.0% softirq, 100.0% hardirq, 0.0% idle
>>      #4: 0.0% system, 0.0% softirq, 100.0% hardirq, 0.0% idle
>>      #5: 0.0% system, 0.0% softirq, 100.0% hardirq, 0.0% idle
>>    ...
>>
>> This would be helpful in determining whether an interrupt storm has
>> occurred or in identifying the cause of the softlockup. The criteria for
>> determination are as follows:
>>    a. If the hardirq utilization is high, then interrupt storm should be
>>    considered and the root cause cannot be determined from the call tree.
>>    b. If the softirq utilization is high, then we could analyze the call
>>    tree but it may cannot reflect the root cause.
>>    c. If the system utilization is high, then we could analyze the root
>>    cause from the call tree.
>>
>> Signed-off-by: Bitao Hu <yaoma@linux.alibaba.com>
>> ---
>>   kernel/watchdog.c | 58 +++++++++++++++++++++++++++++++++++++++++++++++
>>   1 file changed, 58 insertions(+)
>>
>> diff --git a/kernel/watchdog.c b/kernel/watchdog.c
>> index 81a8862295d6..9fad10e0a147 100644
>> --- a/kernel/watchdog.c
>> +++ b/kernel/watchdog.c
>> @@ -23,6 +23,8 @@
>>   #include <linux/sched/debug.h>
>>   #include <linux/sched/isolation.h>
>>   #include <linux/stop_machine.h>
>> +#include <linux/kernel_stat.h>
>> +#include <linux/math64.h>
>>   #include <asm/irq_regs.h>
>>   #include <linux/kvm_para.h>
>> @@ -441,6 +443,58 @@ static int is_softlockup(unsigned long touch_ts,
>>       return 0;
>>   }
>> +#ifdef CONFIG_IRQ_TIME_ACCOUNTING
>> +static DEFINE_PER_CPU(u64, cpustat_old[NR_STATS]);
>> +static DEFINE_PER_CPU(u64, cpustat_diff[5][NR_STATS]);
>> +static DEFINE_PER_CPU(int, cpustat_tail);
>> +
>> +static void update_cpustat(void)
>> +{
>> +    u64 *old = this_cpu_ptr(cpustat_old);
>> +    u64 (*diff)[NR_STATS] = this_cpu_ptr(cpustat_diff);
>> +    int tail = this_cpu_read(cpustat_tail), i;
>> +    struct kernel_cpustat kcpustat;
>> +    u64 *cpustat = kcpustat.cpustat;
>> +
>> +    kcpustat_cpu_fetch(&kcpustat, smp_processor_id());
>> +    for (i = 0; i < NR_STATS; i++) {
>> +        diff[tail][i] = cpustat[i] - old[i];
>> +        old[i] = cpustat[i];
>> +    }
>> +    this_cpu_write(cpustat_tail, (tail + 1) % 5);
> The number 5 here is related to the 5 in cpustat_diff[5], and it is
> recommended to use a macro definition instead of using the number 5
> directly.
In the "set_sample_period" function, the "sample_period" is hardcoded to
be 1/5 of the "softlockup_thresh", therefore I define the length of the
array here as 5.
>> +}
>> +
>> +static void print_cpustat(void)
>> +{
>> +    int i, j, k;
>> +    u64 a[5][NR_STATS], b[5][NR_STATS];
> Use define instead of the literal number 5.
Same as above
>> +    u64 (*diff)[NR_STATS] = this_cpu_ptr(cpustat_diff);
>> +    int tail = this_cpu_read(cpustat_tail);
>> +    u32 period_us = sample_period / 1000;
> Use NSEC_PER_USEC
Sure.
>> +
>> +    for (i = 0; i < 5; i++) {
>> +        for (j = 0; j < NR_STATS; j++) {
>> +            a[i][j] = 100 * (diff[i][j] / 1000);
>> +            b[i][j] = 10 * do_div(a[i][j], period_us);
>> +            do_div(b[i][j], period_us);
>> +        }
>> +    }
>> +    printk(KERN_CRIT "CPU#%d Utilization every %us during lockup:\n",
> better use "pr_crit", and was the original intent here microseconds (us)
> or milliseconds (ms)?
Using "pr_crit", each line will have a "watchdog:" prefix, which I think
might not look very neat.
The intended unit here is seconds(s).

>> +        smp_processor_id(), period_us/1000000);
> better use "period_us /NSEC_PER_MSEC"?
Sure, I will use USEC_PER_SEC here.
>> +    for (k = 0, i = tail; k < 5; k++, i = (i + 1) % 5) {
>
> It seems that only i and j are necessary, k is not essential.
Sure, I will remove the variable 'k'.
>
>> +        printk(KERN_CRIT "\t#%d: %llu.%llu%% system,\t%llu.%llu%%
>> softirq,\t"
>> +            "%llu.%llu%% hardirq,\t%llu.%llu%% idle\n", k+1,
>> +            a[i][CPUTIME_SYSTEM], b[i][CPUTIME_SYSTEM],
>> +            a[i][CPUTIME_SOFTIRQ], b[i][CPUTIME_SOFTIRQ],
>> +            a[i][CPUTIME_IRQ], b[i][CPUTIME_IRQ],
>> +            a[i][CPUTIME_IDLE], b[i][CPUTIME_IDLE]);
>> +    }
>> +}
>> +#else
>> +static inline void update_cpustat(void) { }
>> +static inline void print_cpustat(void) { }
>> +#endif
>> +
>>   /* watchdog detector functions */
>>   static DEFINE_PER_CPU(struct completion, softlockup_completion);
>>   static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
>> @@ -504,6 +558,9 @@ static enum hrtimer_restart
>> watchdog_timer_fn(struct hrtimer *hrtimer)
>>        */
>>       period_ts = READ_ONCE(*this_cpu_ptr(&watchdog_report_ts));
>> +    /* update cpu usage stat */
> The function name already indicates that it involves graphs, so the
> comment here appears superfluous.
> If a comment is absolutely necessary, please provide more detailed
> information.
Sure.
>> +    update_cpustat();
>> +
>>       /* Reset the interval when touched by known problematic code. */
>>       if (period_ts == SOFTLOCKUP_DELAY_REPORT) {
>>           if (unlikely(__this_cpu_read(softlockup_touch_sync))) {
>> @@ -539,6 +596,7 @@ static enum hrtimer_restart
>> watchdog_timer_fn(struct hrtimer *hrtimer)
>>           pr_emerg("BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n",
>>               smp_processor_id(), duration,
>>               current->comm, task_pid_nr(current));
>> +        print_cpustat();
>>           print_modules();
>>           print_irqtrace_events(current);
>>           if (regs)

\
 
 \ /
  Last update: 2024-05-27 14:31    [W:0.109 / U:0.172 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site