Re: [PATCH v6] hung_task: Deduplicate identical hang reports using explicit blocker tracking

[Aaron Tomlin]

On Sun, Jul 19, 2026 at 11:02:57PM -0700, Andrew Morton wrote:
> On Sun, 19 Jul 2026 12:13:05 -0400 Aaron Tomlin <atomlin@xxxxxxxxxxx> wrote:
> 
> > Currently, during severe lock contention, multiple tasks can hang while
> > waiting on the exact same resource. The khungtaskd kthread
> > indiscriminately reports every single instance with a stack trace.
> > This can roll the kernel ring buffer and prematurely exhaust the
> > kernel.hung_task_warnings budget. Consequently, the kernel is left
> > entirely blind to subsequent, unrelated deadlocks.
> 
> This does seem something we'd like to address.
> 
> > To preserve the warning budget and ring buffer without sacrificing
> > observability, this patch introduces a two-tier deduplication mechanism:
> > 
> 
> Thanks.  Hopefully Petr and Lance will be able to review this version.
> 
> AI review might have found an issue:
> 	https://sashiko.dev/#/patchset/20260719161305.428947-1-atomlin@xxxxxxxxxxx

Hi Andrew,

Thank you for you feedback.

This is indeed a legitimate issue.

I have moved the panic threshold check out of hung_task_info() and placing
it directly into check_hung_uninterruptible_tasks() right after
this_round_count is incremented.

This ensures that:

    1.  The panic threshold is correctly evaluated for all hung tasks,
        regardless of whether they have already been reported

    2.  If the panic threshold is reached, hung_task_call_panic is set to
        true, ensuring the watchdog successfully panics and dumps the
        system state

    3.  We still completely suppress the redundant stack traces for
        already-reported tasks (i.e., preventing log spam and preserving
        the global warning budget)


Kind regards,
-- 
Aaron Tomlin