Re: [PATCH v2 2/4] perf: Enqueue SIGTRAP always via task_work.

[Marco Elver <elver@google.com>]

On Tue, 12 Mar 2024 at 19:08, Sebastian Andrzej Siewior
<bigeasy@linutronix.de> wrote:
>
> A signal is delivered by raising irq_work() which works from any context
> including NMI. irq_work() can be delayed if the architecture does not
> provide an interrupt vector. In order not to lose a signal, the signal
> is injected via task_work during event_sched_out().
>
> Instead going via irq_work, the signal could be added directly via
> task_work. The signal is sent to current and can be enqueued on its
> return path to userland instead of triggering irq_work. A dummy IRQ is
> required in the NMI case to ensure the task_work is handled before
> returning to user land. For this irq_work is used. An alternative would
> be just raising an interrupt like arch_send_call_function_single_ipi().
>
> During testing with `remove_on_exec' it become visible that the event
> can be enqueued via NMI during execve(). The task_work must not be kept
> because free_event() will complain later. Also the new task will not
> have a sighandler installed.
>
> Queue signal via task_work. Remove perf_event::pending_sigtrap and
> and use perf_event::pending_work instead. Raise irq_work in the NMI case
> for a dummy interrupt. Remove the task_work if the event is freed.
>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> ---
>  include/linux/perf_event.h |  3 +--
>  kernel/events/core.c       | 45 +++++++++++++++++---------------------
>  2 files changed, 21 insertions(+), 27 deletions(-)
>
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index d2a15c0c6f8a9..24ac6765146c7 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -781,7 +781,6 @@ struct perf_event {
>         unsigned int                    pending_wakeup;
>         unsigned int                    pending_kill;
>         unsigned int                    pending_disable;
> -       unsigned int                    pending_sigtrap;
>         unsigned long                   pending_addr;   /* SIGTRAP */
>         struct irq_work                 pending_irq;
>         struct callback_head            pending_task;
> @@ -959,7 +958,7 @@ struct perf_event_context {
>         struct rcu_head                 rcu_head;
>
>         /*
> -        * Sum (event->pending_sigtrap + event->pending_work)
> +        * Sum (event->pending_work + event->pending_work)
>          *
>          * The SIGTRAP is targeted at ctx->task, as such it won't do changing
>          * that until the signal is delivered.
> diff --git a/kernel/events/core.c b/kernel/events/core.c
> index c7a0274c662c8..e9926baaa1587 100644
> --- a/kernel/events/core.c
> +++ b/kernel/events/core.c
> @@ -2283,21 +2283,6 @@ event_sched_out(struct perf_event *event, struct perf_event_context *ctx)
>                 state = PERF_EVENT_STATE_OFF;
>         }
>
> -       if (event->pending_sigtrap) {
> -               bool dec = true;
> -
> -               event->pending_sigtrap = 0;
> -               if (state != PERF_EVENT_STATE_OFF &&
> -                   !event->pending_work) {
> -                       event->pending_work = 1;
> -                       dec = false;
> -                       WARN_ON_ONCE(!atomic_long_inc_not_zero(&event->refcount));
> -                       task_work_add(current, &event->pending_task, TWA_RESUME);
> -               }
> -               if (dec)
> -                       local_dec(&event->ctx->nr_pending);
> -       }
> -
>         perf_event_set_state(event, state);
>
>         if (!is_software_event(event))
> @@ -6741,11 +6726,6 @@ static void __perf_pending_irq(struct perf_event *event)
>          * Yay, we hit home and are in the context of the event.
>          */
>         if (cpu == smp_processor_id()) {
> -               if (event->pending_sigtrap) {
> -                       event->pending_sigtrap = 0;
> -                       perf_sigtrap(event);
> -                       local_dec(&event->ctx->nr_pending);
> -               }
>                 if (event->pending_disable) {
>                         event->pending_disable = 0;
>                         perf_event_disable_local(event);
> @@ -9592,14 +9572,17 @@ static int __perf_event_overflow(struct perf_event *event,
>
>                 if (regs)
>                         pending_id = hash32_ptr((void *)instruction_pointer(regs)) ?: 1;
> -               if (!event->pending_sigtrap) {
> -                       event->pending_sigtrap = pending_id;
> +               if (!event->pending_work) {
> +                       event->pending_work = pending_id;
>                         local_inc(&event->ctx->nr_pending);
> -                       irq_work_queue(&event->pending_irq);
> +                       WARN_ON_ONCE(!atomic_long_inc_not_zero(&event->refcount));
> +                       task_work_add(current, &event->pending_task, TWA_RESUME);
> +                       if (in_nmi())
> +                               irq_work_queue(&event->pending_irq);

Some brief code comments here would help having to dig through git
history to understand this.

>                 } else if (event->attr.exclude_kernel && valid_sample) {
>                         /*
>                          * Should not be able to return to user space without
> -                        * consuming pending_sigtrap; with exceptions:
> +                        * consuming pending_work; with exceptions:
>                          *
>                          *  1. Where !exclude_kernel, events can overflow again
>                          *     in the kernel without returning to user space.
> @@ -9609,7 +9592,7 @@ static int __perf_event_overflow(struct perf_event *event,
>                          *     To approximate progress (with false negatives),
>                          *     check 32-bit hash of the current IP.
>                          */
> -                       WARN_ON_ONCE(event->pending_sigtrap != pending_id);
> +                       WARN_ON_ONCE(event->pending_work != pending_id);
>                 }
>
>                 event->pending_addr = 0;
> @@ -13049,6 +13032,13 @@ static void sync_child_event(struct perf_event *child_event)
>                      &parent_event->child_total_time_running);
>  }
>
> +static bool task_work_cb_match(struct callback_head *cb, void *data)
> +{
> +       struct perf_event *event = container_of(cb, struct perf_event, pending_task);
> +
> +       return event == data;
> +}
> +
>  static void
>  perf_event_exit_event(struct perf_event *event, struct perf_event_context *ctx)
>  {
> @@ -13088,6 +13078,11 @@ perf_event_exit_event(struct perf_event *event, struct perf_event_context *ctx)
>                  * Kick perf_poll() for is_event_hup();
>                  */
>                 perf_event_wakeup(parent_event);
> +               if (event->pending_work &&
> +                   task_work_cancel_match(current, task_work_cb_match, event)) {

Brief comment which case this covers would be good.

> +                       put_event(event);
> +                       local_dec(&event->ctx->nr_pending);
> +               }
>                 free_event(event);
>                 put_event(parent_event);
>                 return;
> --
> 2.43.0
>