[PATCH v5 0/6] perf: Make SIGTRAP and __perf_pending_irq() work on RT.

Sebastian Andrzej Siewior posted 6 patches 1 year, 5 months ago
[PATCH v5 0/6] perf: Make SIGTRAP and __perf_pending_irq() work on RT.
Posted by Sebastian Andrzej Siewior 1 year, 5 months ago
Hi,

Arnaldo reported that "perf test sigtrap" fails on PREEMPT_RT. Sending
the signal gets delayed until event_sched_out() which then uses
task_work_add() for its delivery. This breaks on PREEMPT_RT because the
signal is delivered with disabled preemption.

While looking at this, I also stumbled upon __perf_pending_irq() which
requires disabled interrupts but this is not the case on PREEMPT_RT.

This series aim to address both issues while not introducing a new issue
at the same time ;)
Any testing is appreciated.

v4…v5: https://lore.kernel.org/all/20240624152732.1231678-1-bigeasy@linutronix.de/
   - Add TWA_NMI_CURRENT as notify mode for task_work_add() and use it.
     PeterZ pointed out that the current version is not NMI safe.

v3…v4: https://lore.kernel.org/all/20240322065208.60456-1-bigeasy@linutronix.de/
   - Rebased on top of Frederic's series
      (https://lore.kernel.org/all/20240621091601.18227-1-frederic@kernel.org)
   - Frederick pointed out that perf_pending_task() needs to
     perf_swevent_get_recursion_context() in order not to recurse if
     something within perf_swevent_.*_recursion_context() triggers a
     software event. To address this, the counters have been moved to
     the task_struct (#3 + #4) and preemt_disable() has been replaced
     with a RCU-read lock (#5).
   - The remaning logic same that means the event is pushed to task-work
     instead of delivering from IRQ-work. The series was tested with
     remove_on_exec as suggested by Marco Elver: On PREEMPT_RT a single
     invocation passes, 100 parallel invocations report (for some)
     unexpected SIGTRAPs and timeouts. This also observed on !RT
     (without the series) with a higher task-count.

v2…v3: https://lore.kernel.org/all/20240312180814.3373778-1-bigeasy@linutronix.de/
    - Marco suggested to add a few comments
      - Added a comment to __perf_event_overflow() to explain why irq_work
        is raised in the in_nmi() case.
      - Added a comment to perf_event_exit_event() to explain why the
        pending event is deleted.

v1…v2: https://lore.kernel.org/all/20240308175810.2894694-1-bigeasy@linutronix.de/
    - Marco pointed me to the testsuite that showed two problems:
      - Delayed task_work from NMI / missing events.
        Fixed by triggering dummy irq_work to enforce an interrupt for
	the exit-to-userland path which checks task_work
      - Increased ref-count on clean up/ during exec.
        Mostly addressed by the former change. There is still a window
	if the NMI occurs during execve(). This is addressed by removing
	the task_work before free_event().
      The testsuite (remove_on_exec) fails sometimes if the event/
      SIGTRAP is sent before the sighandler is installed.

Sebastian
Re: [PATCH v5 0/6] perf: Make SIGTRAP and __perf_pending_irq() work on RT.
Posted by Peter Zijlstra 1 year, 5 months ago

Thanks, I'll go queue it for perf/core once the robot gets back to me on
the current pile.
Re: [PATCH v5 0/6] perf: Make SIGTRAP and __perf_pending_irq() work on RT.
Posted by Sebastian Andrzej Siewior 1 year, 5 months ago
On 2024-07-05 09:55:11 [+0200], Peter Zijlstra wrote:
> 
> 
> Thanks, I'll go queue it for perf/core once the robot gets back to me on
> the current pile.

Thank you.

Sebastian
Re: [PATCH v5 0/6] perf: Make SIGTRAP and __perf_pending_irq() work on RT.
Posted by Arnaldo Carvalho de Melo 1 year, 5 months ago
On Thu, Jul 04, 2024 at 07:03:34PM +0200, Sebastian Andrzej Siewior wrote:
> Hi,
> 
> Arnaldo reported that "perf test sigtrap" fails on PREEMPT_RT. Sending
> the signal gets delayed until event_sched_out() which then uses
> task_work_add() for its delivery. This breaks on PREEMPT_RT because the
> signal is delivered with disabled preemption.
> 
> While looking at this, I also stumbled upon __perf_pending_irq() which
> requires disabled interrupts but this is not the case on PREEMPT_RT.
> 
> This series aim to address both issues while not introducing a new issue
> at the same time ;)
> Any testing is appreciated.

Were should I apply this patch? The v4 series was applied to
linux-rt-devel/linux-6.10.y-rt IIRC

- Arnaldo
 
> v4…v5: https://lore.kernel.org/all/20240624152732.1231678-1-bigeasy@linutronix.de/
>    - Add TWA_NMI_CURRENT as notify mode for task_work_add() and use it.
>      PeterZ pointed out that the current version is not NMI safe.
> 
> v3…v4: https://lore.kernel.org/all/20240322065208.60456-1-bigeasy@linutronix.de/
>    - Rebased on top of Frederic's series
>       (https://lore.kernel.org/all/20240621091601.18227-1-frederic@kernel.org)
>    - Frederick pointed out that perf_pending_task() needs to
>      perf_swevent_get_recursion_context() in order not to recurse if
>      something within perf_swevent_.*_recursion_context() triggers a
>      software event. To address this, the counters have been moved to
>      the task_struct (#3 + #4) and preemt_disable() has been replaced
>      with a RCU-read lock (#5).
>    - The remaning logic same that means the event is pushed to task-work
>      instead of delivering from IRQ-work. The series was tested with
>      remove_on_exec as suggested by Marco Elver: On PREEMPT_RT a single
>      invocation passes, 100 parallel invocations report (for some)
>      unexpected SIGTRAPs and timeouts. This also observed on !RT
>      (without the series) with a higher task-count.
> 
> v2…v3: https://lore.kernel.org/all/20240312180814.3373778-1-bigeasy@linutronix.de/
>     - Marco suggested to add a few comments
>       - Added a comment to __perf_event_overflow() to explain why irq_work
>         is raised in the in_nmi() case.
>       - Added a comment to perf_event_exit_event() to explain why the
>         pending event is deleted.
> 
> v1…v2: https://lore.kernel.org/all/20240308175810.2894694-1-bigeasy@linutronix.de/
>     - Marco pointed me to the testsuite that showed two problems:
>       - Delayed task_work from NMI / missing events.
>         Fixed by triggering dummy irq_work to enforce an interrupt for
> 	the exit-to-userland path which checks task_work
>       - Increased ref-count on clean up/ during exec.
>         Mostly addressed by the former change. There is still a window
> 	if the NMI occurs during execve(). This is addressed by removing
> 	the task_work before free_event().
>       The testsuite (remove_on_exec) fails sometimes if the event/
>       SIGTRAP is sent before the sighandler is installed.
> 
> Sebastian
> 
Re: [PATCH v5 0/6] perf: Make SIGTRAP and __perf_pending_irq() work on RT.
Posted by Sebastian Andrzej Siewior 1 year, 5 months ago
On 2024-07-04 16:45:18 [-0300], Arnaldo Carvalho de Melo wrote:
> Were should I apply this patch? The v4 series was applied to
> linux-rt-devel/linux-6.10.y-rt IIRC

Revert the v4 series, apply this one. Wait a few hours and there should
be a new RT release with this done ;)

> - Arnaldo

Sebastian
Re: [PATCH v5 0/6] perf: Make SIGTRAP and __perf_pending_irq() work on RT.
Posted by Arnaldo Carvalho de Melo 1 year, 5 months ago
On Thu, Jul 04, 2024 at 04:45:22PM -0300, Arnaldo Carvalho de Melo wrote:
> On Thu, Jul 04, 2024 at 07:03:34PM +0200, Sebastian Andrzej Siewior wrote:
> > Hi,
> > 
> > Arnaldo reported that "perf test sigtrap" fails on PREEMPT_RT. Sending
> > the signal gets delayed until event_sched_out() which then uses
> > task_work_add() for its delivery. This breaks on PREEMPT_RT because the
> > signal is delivered with disabled preemption.
> > 
> > While looking at this, I also stumbled upon __perf_pending_irq() which
> > requires disabled interrupts but this is not the case on PREEMPT_RT.
> > 
> > This series aim to address both issues while not introducing a new issue
> > at the same time ;)
> > Any testing is appreciated.
> 
> Were should I apply this patch? The v4 series was applied to
> linux-rt-devel/linux-6.10.y-rt IIRC

Looking at linux-rt-devel/linux-6.10.y-rt I see:

commit ca8b27c51f0962f8fb59e5acb23e0af791fb5c04
Author: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Date:   Tue Jun 25 10:56:32 2024 +0200

    perf: Update the perf series
    
    This is an all-in-one patch integrating the following changes:
    
    - Merging Frederick's "Fix leaked sigtrap events" series as of v4 which
      is a dependency.
    
    - Update the "perf test sigtrap" fixup to v4 as posted.
    
    Link: https://lore.kernel.org/20240621091601.18227-1-frederic@kernel.org
    Link: https://lore.kernel.org/20240624152732.1231678-1-bigeasy@linutronix.de
    Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

But I think that is v4, right?

- Arnaldo
>  
> > v4…v5: https://lore.kernel.org/all/20240624152732.1231678-1-bigeasy@linutronix.de/
> >    - Add TWA_NMI_CURRENT as notify mode for task_work_add() and use it.
> >      PeterZ pointed out that the current version is not NMI safe.
> > 
> > v3…v4: https://lore.kernel.org/all/20240322065208.60456-1-bigeasy@linutronix.de/
> >    - Rebased on top of Frederic's series
> >       (https://lore.kernel.org/all/20240621091601.18227-1-frederic@kernel.org)
> >    - Frederick pointed out that perf_pending_task() needs to
> >      perf_swevent_get_recursion_context() in order not to recurse if
> >      something within perf_swevent_.*_recursion_context() triggers a
> >      software event. To address this, the counters have been moved to
> >      the task_struct (#3 + #4) and preemt_disable() has been replaced
> >      with a RCU-read lock (#5).
> >    - The remaning logic same that means the event is pushed to task-work
> >      instead of delivering from IRQ-work. The series was tested with
> >      remove_on_exec as suggested by Marco Elver: On PREEMPT_RT a single
> >      invocation passes, 100 parallel invocations report (for some)
> >      unexpected SIGTRAPs and timeouts. This also observed on !RT
> >      (without the series) with a higher task-count.
> > 
> > v2…v3: https://lore.kernel.org/all/20240312180814.3373778-1-bigeasy@linutronix.de/
> >     - Marco suggested to add a few comments
> >       - Added a comment to __perf_event_overflow() to explain why irq_work
> >         is raised in the in_nmi() case.
> >       - Added a comment to perf_event_exit_event() to explain why the
> >         pending event is deleted.
> > 
> > v1…v2: https://lore.kernel.org/all/20240308175810.2894694-1-bigeasy@linutronix.de/
> >     - Marco pointed me to the testsuite that showed two problems:
> >       - Delayed task_work from NMI / missing events.
> >         Fixed by triggering dummy irq_work to enforce an interrupt for
> > 	the exit-to-userland path which checks task_work
> >       - Increased ref-count on clean up/ during exec.
> >         Mostly addressed by the former change. There is still a window
> > 	if the NMI occurs during execve(). This is addressed by removing
> > 	the task_work before free_event().
> >       The testsuite (remove_on_exec) fails sometimes if the event/
> >       SIGTRAP is sent before the sighandler is installed.
> > 
> > Sebastian
> >