Tested v3 on Meteor Lake (Core Ultra 5 125H, vpu_37xx firmware 20260219)
on top of v7.4-rc1 (d48723711bae):

- 4 kexec runs with the NPU runtime-suspended (one from a 7.0 kernel,
  three from 7.4): every run hit ATS_ERR during probe, recovery_work was
  not queued, probe failed with -ETIMEDOUT, and there was no oops
- FLR plus rebind brings the NPU back, and it autosuspends afterwards
- debugfs force_recovery after a successful probe recovers the NPU, and
  it returns to runtime-suspended

None of the runs had probe succeed after the error IRQ, so the deferred
recovery in ivpu_pm_enable() isn't covered by this.

One question: v2 had "Cc: [email protected] # v6.10+" and v3 dropped
the annotation. disable_work()/enable_work() only exist since v6.10 and
the Fixes: commit is in v6.3, so should the "# v6.10+" come back?

Tested-by: Daksh Sahni <[email protected]>

On Tue, Nov 10, 2026 at 5:45 PM Karol Wachowski
<[email protected]> wrote:
>
> An error IRQ that fires after ivpu_boot() enables interrupts queues
> recovery_work during probe. If probe then fails, the device is torn
> down while the work runs, and the work faults on the unmapped BARs:
>
>   BUG: unable to handle page fault for address: ffffcfc6680200f4
>   Workqueue: events_unbound ivpu_pm_recovery_work [intel_vpu]
>   RIP: 0010:ivpu_hw_ip_ipc_rx_addr_get+0xc/0xb0 [intel_vpu]
>   Call Trace:
>    ivpu_wait_for_ready+0x155/0x220 [intel_vpu]
>    ivpu_boot+0x9e/0x270 [intel_vpu]
>    ivpu_resume+0xa4/0xd0 [intel_vpu]
>    ivpu_pm_reset_complete+0x2c/0x90 [intel_vpu]
>    ivpu_pm_recovery_work+0xf6/0x140 [intel_vpu]
>
> If probe succeeds, the recovery runs concurrently with the rest of
> probe, and its pm_runtime_disable() discards the initial autosuspend
> request, leaving the NPU active until first use. Any recovery that
> starts while the NPU is idle loses the pending autosuspend the same way.
>
> Keep recovery_work disabled from ivpu_pm_init() until ivpu_pm_enable().
> A trigger during probe still runs diagnostics and masks interrupts, and
> ivpu_pm_enable() runs the deferred recovery before the initial
> autosuspend request. A trigger racing with ivpu_pm_enable() can cause
> a second, harmless recovery. Request autosuspend at the end of
> ivpu_pm_reset_complete() so the NPU suspends again after every reset.
>
> Fixes: 852be13f3bd3 ("accel/ivpu: Add PM support")
> Cc: [email protected]
> Reported-by: Daksh Sahni <[email protected]>
> Closes: https://lore.kernel.org/r/[email protected]
> Signed-off-by: Karol Wachowski <[email protected]>
> ---
>
> Notes:
>     v3:
>     - Request autosuspend at the end of ivpu_pm_reset_complete().
>       pm_runtime_disable() in ivpu_pm_reset_begin() cancels a pending
>       autosuspend request and pm_runtime_enable() does not re-issue it, so
>       a recovery racing with the initial autosuspend request in
>       ivpu_pm_enable() left the NPU active until first use (Sashiko).
>       A trigger racing with ivpu_pm_enable() can still queue recovery
>       twice. The second run is a full reset that leaves the device in a
>       consistent state, and autosuspend is now re-requested after each
>       reset.
>     - Replace pm_runtime_mark_last_busy() with pm_request_autosuspend() in
>       ivpu_pm_reset_complete(); it marks the device busy itself.
>     - Shorten the commit message.
>     - v2: 
> https://lore.kernel.org/r/[email protected]
>
>     v2:
>     - Rework of Daksh's "accel/ivpu: Disable recovery work on probe failure":
>       https://lore.kernel.org/r/[email protected]
>     - Keep recovery_work disabled from ivpu_pm_init() until ivpu_pm_enable()
>       instead of disabling it on the probe error path, which is too late as
>       recovery may already run concurrently with ivpu_boot() (Sashiko)
>     - Run a recovery requested during probe from ivpu_pm_enable() and wait
>       for it before the initial autosuspend request, so the NPU is reset
>       before use and the autosuspend is not lost when probe succeeds
>
>     v1:
>     - Link: 
> https://lore.kernel.org/r/[email protected]
>
>  drivers/accel/ivpu/ivpu_pm.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/accel/ivpu/ivpu_pm.c b/drivers/accel/ivpu/ivpu_pm.c
> index aa813262253e..e4b79c946ca0 100644
> --- a/drivers/accel/ivpu/ivpu_pm.c
> +++ b/drivers/accel/ivpu/ivpu_pm.c
> @@ -150,8 +150,8 @@ static void ivpu_pm_reset_complete(struct ivpu_device 
> *vdev)
>         up_write(&vdev->pm->reset_lock);
>         atomic_set(&vdev->pm->reset_pending, 0);
>
> -       pm_runtime_mark_last_busy(vdev->drm.dev);
>         pm_runtime_enable(vdev->drm.dev);
> +       pm_request_autosuspend(vdev->drm.dev);
>  }
>
>  static void ivpu_pm_recovery_work(struct work_struct *work)
> @@ -415,6 +415,7 @@ void ivpu_pm_init(struct ivpu_device *vdev)
>         atomic_set(&pm->engine_reset_counter, 0);
>
>         INIT_WORK(&pm->recovery_work, ivpu_pm_recovery_work);
> +       disable_work(&pm->recovery_work);
>         INIT_DELAYED_WORK(&pm->job_timeout_work, ivpu_job_timeout_work);
>
>         if (ivpu_disable_recovery)
> @@ -439,6 +440,12 @@ void ivpu_pm_enable(struct ivpu_device *vdev)
>  {
>         struct device *dev = vdev->drm.dev;
>
> +       enable_work(&vdev->pm->recovery_work);
> +       if (atomic_read(&vdev->pm->reset_pending)) {
> +               queue_work(system_dfl_wq, &vdev->pm->recovery_work);
> +               flush_work(&vdev->pm->recovery_work);
> +       }
> +
>         pm_runtime_allow(dev);
>         pm_runtime_put_autosuspend(dev);
>  }
>
> base-commit: 54d3ff1b5ac08d0e28029b93a1719ddd9ff670a4
> --
> 2.43.0
>

Reply via email to