When the watchdog determines that the current soft lockup is due
to an interrupt storm based on CPU utilization, reporting the
most frequent interrupts could be good enough for further
troubleshooting.
Below is an example of interrupt storm. The call tree does not
provide useful information, but we can analyze which interrupt
caused the soft lockup by comparing the counts of interrupts.
[ 2987.488075] watchdog: BUG: soft lockup - CPU#9 stuck for 23s! [kworker/9:1:214]
[ 2987.488607] CPU#9 Utilization every 4s during lockup:
[ 2987.488941] #1: 0% system, 0% softirq, 100% hardirq, 0% idle
[ 2987.489357] #2: 0% system, 0% softirq, 100% hardirq, 0% idle
[ 2987.489771] #3: 0% system, 0% softirq, 100% hardirq, 0% idle
[ 2987.490186] #4: 0% system, 0% softirq, 100% hardirq, 0% idle
[ 2987.490601] #5: 0% system, 0% softirq, 100% hardirq, 0% idle
[ 2987.491034] CPU#9 Detect HardIRQ Time exceeds 50%. Most frequent HardIRQs:
[ 2987.491493] #1: 330985 irq#7(IPI)
[ 2987.491743] #2: 5000 irq#10(arch_timer)
[ 2987.492039] #3: 9 irq#91(nvme0q2)
[ 2987.492318] #4: 3 irq#118(virtio1-output.12)
..
[ 2987.492728] Call trace:
[ 2987.492729] __do_softirq+0xa8/0x364
Signed-off-by: Bitao Hu <[email protected]>
---
kernel/watchdog.c | 156 ++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 156 insertions(+)
diff --git a/kernel/watchdog.c b/kernel/watchdog.c
index 71d5b6dfa358..26dc1ad86276 100644
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -18,6 +18,9 @@
#include <linux/init.h>
#include <linux/kernel_stat.h>
#include <linux/math64.h>
+#include <linux/irq.h>
+#include <linux/irqdesc.h>
+#include <linux/bitops.h>
#include <linux/module.h>
#include <linux/sysctl.h>
#include <linux/tick.h>
@@ -410,13 +413,153 @@ static void print_cpustat(void)
}
}
+#define HARDIRQ_PERCENT_THRESH 50
+#define NUM_HARDIRQ_REPORT 5
+static DECLARE_BITMAP(softlockup_hardirq_cpus, CONFIG_NR_CPUS);
+static DEFINE_PER_CPU(u32 *, hardirq_counts);
+static DEFINE_PER_CPU(int, actual_nr_irqs);
+struct irq_counts {
+ int irq;
+ u32 counts;
+};
+
+/* Tabulate the most frequent interrupts. */
+static void tabulate_irq_count(struct irq_counts *irq_counts, int irq, u32 counts, int rank)
+{
+ int i;
+ struct irq_counts new_count = {irq, counts};
+
+ for (i = 0; i < rank; i++) {
+ if (counts > irq_counts[i].counts)
+ swap(new_count, irq_counts[i]);
+ }
+}
+
+/*
+ * If the hardirq time exceeds HARDIRQ_PERCENT_THRESH% of the sample_period,
+ * then the cause of softlockup might be interrupt storm. In this case, it
+ * would be useful to start interrupt counting.
+ */
+static bool need_counting_irqs(void)
+{
+ u8 util;
+ int tail = __this_cpu_read(cpustat_tail);
+
+ tail = (tail + NUM_HARDIRQ_REPORT - 1) % NUM_HARDIRQ_REPORT;
+ util = __this_cpu_read(cpustat_util[tail][STATS_HARDIRQ]);
+ return util > HARDIRQ_PERCENT_THRESH;
+}
+
+static void start_counting_irqs(void)
+{
+ int i;
+ struct irq_desc *desc;
+ u32 *counts = __this_cpu_read(hardirq_counts);
+ int cpu = smp_processor_id();
+
+ if (!test_bit(cpu, softlockup_hardirq_cpus)) {
+ /*
+ * nr_irqs has the potential to grow at runtime. We should read
+ * it and store locally to avoid array out-of-bounds access.
+ */
+ __this_cpu_write(actual_nr_irqs, nr_irqs);
+ counts = kmalloc_array(__this_cpu_read(actual_nr_irqs),
+ sizeof(u32),
+ GFP_ATOMIC);
+ if (!counts)
+ return;
+ for (i = 0; i < __this_cpu_read(actual_nr_irqs); i++) {
+ desc = irq_to_desc(i);
+ if (!desc)
+ continue;
+ counts[i] = desc->kstat_irqs ?
+ *this_cpu_ptr(desc->kstat_irqs) : 0;
+ }
+ __this_cpu_write(hardirq_counts, counts);
+ set_bit(cpu, softlockup_hardirq_cpus);
+ }
+}
+
+static void stop_counting_irqs(void)
+{
+ u32 *counts = __this_cpu_read(hardirq_counts);
+ int cpu = smp_processor_id();
+
+ if (test_bit(cpu, softlockup_hardirq_cpus)) {
+ kfree(counts);
+ counts = NULL;
+ __this_cpu_write(hardirq_counts, counts);
+ clear_bit(cpu, softlockup_hardirq_cpus);
+ }
+}
+
+static void print_irq_counts(void)
+{
+ int i;
+ struct irq_desc *desc;
+ u32 counts_diff;
+ u32 *counts = __this_cpu_read(hardirq_counts);
+ int cpu = smp_processor_id();
+ struct irq_counts irq_counts_sorted[NUM_HARDIRQ_REPORT] = {
+ {-1, 0}, {-1, 0}, {-1, 0}, {-1, 0},
+ };
+
+ if (test_bit(cpu, softlockup_hardirq_cpus)) {
+ for_each_irq_desc(i, desc) {
+ if (!desc)
+ continue;
+ /*
+ * We need to bounds-check in case someone on a different CPU
+ * expanded nr_irqs.
+ */
+ if (i < __this_cpu_read(actual_nr_irqs))
+ counts_diff = desc->kstat_irqs ?
+ *this_cpu_ptr(desc->kstat_irqs) - counts[i] : 0;
+ else
+ counts_diff = desc->kstat_irqs ?
+ *this_cpu_ptr(desc->kstat_irqs) : 0;
+ tabulate_irq_count(irq_counts_sorted, i, counts_diff,
+ NUM_HARDIRQ_REPORT);
+ }
+ /*
+ * We do not want the "watchdog: " prefix on every line,
+ * hence we use "printk" instead of "pr_crit".
+ */
+ printk(KERN_CRIT "CPU#%d Detect HardIRQ Time exceeds %d%%. Most frequent HardIRQs:\n",
+ smp_processor_id(), HARDIRQ_PERCENT_THRESH);
+ for (i = 0; i < NUM_HARDIRQ_REPORT; i++) {
+ if (irq_counts_sorted[i].irq == -1)
+ break;
+ desc = irq_to_desc(irq_counts_sorted[i].irq);
+ if (desc && desc->action)
+ printk(KERN_CRIT "\t#%u: %-10u\tirq#%d(%s)\n",
+ i+1, irq_counts_sorted[i].counts,
+ irq_counts_sorted[i].irq, desc->action->name);
+ else
+ printk(KERN_CRIT "\t#%u: %-10u\tirq#%d\n",
+ i+1, irq_counts_sorted[i].counts,
+ irq_counts_sorted[i].irq);
+ }
+ /*
+ * If the hardirq time is less than HARDIRQ_PERCENT_THRESH% in the last
+ * sample_period, then we suspect the interrupt storm might be subsiding.
+ */
+ if (!need_counting_irqs())
+ stop_counting_irqs();
+ }
+}
+
static void report_cpu_status(void)
{
print_cpustat();
+ print_irq_counts();
}
#else
static inline void update_cpustat(void) { }
static inline void report_cpu_status(void) { }
+static inline bool need_counting_irqs(void) { return false; }
+static inline void start_counting_irqs(void) { }
+static inline void stop_counting_irqs(void) { }
#endif
/*
@@ -520,6 +663,18 @@ static int is_softlockup(unsigned long touch_ts,
unsigned long now)
{
if ((watchdog_enabled & WATCHDOG_SOFTOCKUP_ENABLED) && watchdog_thresh) {
+ /*
+ * If period_ts has not been updated during a sample_period, then
+ * in the subsequent few sample_periods, period_ts might also not
+ * be updated, which could indicate a potential softlockup. In
+ * this case, if we suspect the cause of the potential softlockup
+ * might be interrupt storm, then we need to count the interrupts
+ * to find which interrupt is storming.
+ */
+ if (time_after_eq(now, period_ts + get_softlockup_thresh() / 5) &&
+ need_counting_irqs())
+ start_counting_irqs();
+
/* Warn about unreasonable delays. */
if (time_after(now, period_ts + get_softlockup_thresh()))
return now - touch_ts;
@@ -542,6 +697,7 @@ static DEFINE_PER_CPU(struct cpu_stop_work, softlockup_stop_work);
static int softlockup_fn(void *data)
{
update_touch_ts();
+ stop_counting_irqs();
complete(this_cpu_ptr(&softlockup_completion));
return 0;
--
2.37.1 (Apple Git-137.1)
Hi,
On 2024/2/7 05:42, Doug Anderson wrote:
> Hi,
>
> On Tue, Feb 6, 2024 at 1:59 AM Bitao Hu <[email protected]> wrote:
>>
>> diff --git a/kernel/watchdog.c b/kernel/watchdog.c
>> index 71d5b6dfa358..26dc1ad86276 100644
>> --- a/kernel/watchdog.c
>> +++ b/kernel/watchdog.c
>> @@ -18,6 +18,9 @@
>> #include <linux/init.h>
>> #include <linux/kernel_stat.h>
>> #include <linux/math64.h>
>> +#include <linux/irq.h>
>> +#include <linux/irqdesc.h>
>> +#include <linux/bitops.h>
>
> These are still not sorted alphabetically. "irq.h" and "irqdesc.h"
> should go between "init.h" and "kernel_stat.h". "bitops.h" is trickier
> because the existing headers are not quite sorted. Probably the best
> would be to fully sort them. They should end up like this:
>
> #include <linux/bitops.h>
> #include <linux/cpu.h>
> #include <linux/init.h>
> #include <linux/irq.h>
> #include <linux/irqdesc.h>
> #include <linux/kernel_stat.h>
> #include <linux/kvm_para.h>
> #include <linux/math64.h>
> #include <linux/mm.h>
> #include <linux/module.h>
> #include <linux/nmi.h>
> #include <linux/stop_machine.h>
> #include <linux/sysctl.h>
> #include <linux/tick.h>
>
> #include <linux/sched/clock.h>
> #include <linux/sched/debug.h>
> #include <linux/sched/isolation.h>
>
> #include <asm/irq_regs.h>
>
Sorry, I misunderstood your point, thinking that they should only be
added between "init.h" and "module.h". I will arrange them in the
alphabetical order as you suggested.
>
>> +static void start_counting_irqs(void)
>> +{
>> + int i;
>> + struct irq_desc *desc;
>> + u32 *counts = __this_cpu_read(hardirq_counts);
>> + int cpu = smp_processor_id();
>> +
>> + if (!test_bit(cpu, softlockup_hardirq_cpus)) {
>
> I don't think you need "softlockup_hardirq_cpus", do you? Just read
> "actual_nr_irqs" and see if it's non-zero? ...or read "hardirq_counts"
> and see if it's non-NULL?
Sure, the existing variables are sufficient for making a determination.
And may be I should swap it to make the decision logic here clearer,
like this (untested)?
bool is_counting_started(void)
{
return !!__this_cpu_read(hardirq_counts);
}
if (!is_counting_started()) {
>
>
>> + /*
>> + * nr_irqs has the potential to grow at runtime. We should read
>> + * it and store locally to avoid array out-of-bounds access.
>> + */
>> + __this_cpu_write(actual_nr_irqs, nr_irqs);
>
> nit: IMO store nr_irqs in a local variable to avoid all of the
> "__this_cpu_read" calls everywhere. Then just write it once from your
> local variable.
OK.
>
>
>> + counts = kmalloc_array(__this_cpu_read(actual_nr_irqs),
>> + sizeof(u32),
>> + GFP_ATOMIC);
>
> should use "kcalloc()" so the array is zeroed. That way if the set of
> non-NULL "desc"s changes between calls you don't end up reading
> uninitialized memory.
OK, I will use "kcalloc()" here.
>
>
>> +static void stop_counting_irqs(void)
>> +{
>> + u32 *counts = __this_cpu_read(hardirq_counts);
>> + int cpu = smp_processor_id();
>> +
>> + if (test_bit(cpu, softlockup_hardirq_cpus)) {
>> + kfree(counts);
>> + counts = NULL;
>> + __this_cpu_write(hardirq_counts, counts);
>
> nit: don't really need to set the local "counts" to NULL. Just:
>
> __this_cpu_write(hardirq_counts, NULL);
>
> ...and actually if you take my advice above and get rid of
> "softlockup_hardirq_cpus" then this function just becomes:
>
> kfree(__this_cpu_read(hardirq_counts));
> __this_cpu_write(hardirq_counts, NULL);
>
> Since kfree() handles when you pass it NULL...
OK.
>
>
>> +static void print_irq_counts(void)
>> +{
>> + int i;
>> + struct irq_desc *desc;
>> + u32 counts_diff;
>> + u32 *counts = __this_cpu_read(hardirq_counts);
>> + int cpu = smp_processor_id();
>> + struct irq_counts irq_counts_sorted[NUM_HARDIRQ_REPORT] = {
>> + {-1, 0}, {-1, 0}, {-1, 0}, {-1, 0},
>> + };
>> +
>> + if (test_bit(cpu, softlockup_hardirq_cpus)) {
>> + for_each_irq_desc(i, desc) {
>> + if (!desc)
>> + continue;
>
> The "if" test above isn't needed. The "for_each_irq_desc()" macro
> already checks for NULL.
Thanks for your reminder.
>
>
>
>> + /*
>> + * We need to bounds-check in case someone on a different CPU
>> + * expanded nr_irqs.
>> + */
>> + if (i < __this_cpu_read(actual_nr_irqs))
>> + counts_diff = desc->kstat_irqs ?
>> + *this_cpu_ptr(desc->kstat_irqs) - counts[i] : 0;
>> + else
>> + counts_diff = desc->kstat_irqs ?
>> + *this_cpu_ptr(desc->kstat_irqs) : 0;
>
> Why do you need to test "kstat_irqs" for 0?
Although "alloc_desc" wil allocate both "desc" and "kstat_irqs" at the
same time, I refer to the usage of "kstat_irqs" in "show_interrupts"
from kernel/irq/proc.c, where it does perform a check.
kernel/irq/proc.c: show_interrupts
for_each_online_cpu(j)
seq_printf(p, "%10u ", desc->kstat_irqs ?
*per_cpu_ptr(desc->kstat_irqs, j) : 0);
I'm not sure why this is necessary, so I copied it as it is. Can we skip
the check in "print_irq_counts"?
> duplicate the math. In other words, I'd expect this (untested):
>
> if (i < __this_cpu_read(actual_nr_irqs))
> count = counts[i];
> else
> count = 0;
> counts_diff = *this_cpu_ptr(desc->kstat_irqs) - count;
Agree.
>
> I guess I'd also put "__this_cpu_read(actual_nr_irqs)" in a local
> variable like you do with counts...
Hi,
On Tue, Feb 6, 2024 at 10:19 PM Bitao Hu <[email protected]> wrote:
>
> Hi,
>
> >> +static void start_counting_irqs(void)
> >> +{
> >> + int i;
> >> + struct irq_desc *desc;
> >> + u32 *counts = __this_cpu_read(hardirq_counts);
> >> + int cpu = smp_processor_id();
> >> +
> >> + if (!test_bit(cpu, softlockup_hardirq_cpus)) {
> >
> > I don't think you need "softlockup_hardirq_cpus", do you? Just read
> > "actual_nr_irqs" and see if it's non-zero? ...or read "hardirq_counts"
> > and see if it's non-NULL?
> Sure, the existing variables are sufficient for making a determination.
> And may be I should swap it to make the decision logic here clearer,
> like this (untested)?
>
> bool is_counting_started(void)
> {
> return !!__this_cpu_read(hardirq_counts);
> }
>
> if (!is_counting_started()) {
If you insist I guess I wouldn't object, but I don't feel it's
necessary. The whole point is just to know if you've already allocated
memory, right? ...and just checking to see if the pointer is non-NULL
or the array-size is non-zero feels pretty clear to me.
> >> + /*
> >> + * We need to bounds-check in case someone on a different CPU
> >> + * expanded nr_irqs.
> >> + */
> >> + if (i < __this_cpu_read(actual_nr_irqs))
> >> + counts_diff = desc->kstat_irqs ?
> >> + *this_cpu_ptr(desc->kstat_irqs) - counts[i] : 0;
> >> + else
> >> + counts_diff = desc->kstat_irqs ?
> >> + *this_cpu_ptr(desc->kstat_irqs) : 0;
> >
> > Why do you need to test "kstat_irqs" for 0?
> Although "alloc_desc" wil allocate both "desc" and "kstat_irqs" at the
> same time, I refer to the usage of "kstat_irqs" in "show_interrupts"
> from kernel/irq/proc.c, where it does perform a check.
Ah, I see. I hadn't noticed that you were testing the pointer before
dereferencing it. OK, seems fine to keep this check. I guess that
would make it this (untested):
if (desc->kstat_irqs) {
counts_diff = *this_cpu_ptr(desc->kstat_irqs);
if (i < __this_cpu_read(actual_nr_irqs))
counts_diff -= counts[i];
} else {
counts_diff = 0;
}