Re: [PATCH] drm/i915: Stop using flush_scheduled_work on driver remove

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 




On 23/09/2022 17:16, Ville Syrjälä wrote:
On Fri, Sep 23, 2022 at 03:29:34PM +0100, Tvrtko Ursulin wrote:
From: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>

Kernel is trying to eliminate callers of flush_scheduled_work so lets
try to accommodate.

We currently call it from intel_modeset_driver_remove_noirq on the driver
remove path but the comment next to it does not tell me what exact work it
wants to flush.

I can spot three (or four) works using the system_wq:

   ..hotplug.reenable_work
   ..hotplug.hotplug_work

Looks like we at least try to shoot those down via
intel_irq_uninstall()
  ->intel_hpd_cancel_work()
   ->cancel_delayed_work_sync()

But I'm not sure how broken the hpd disable path is here.
I know hpd cancel vs. irq disable has some known ordering
issues during suspend at least, some of which I think may
have gotten fixed recently. But hpd cancel is still a bit
of a mess in general.

Here we at least do cancel all the hpd works after irqs
have been disabled, so I don't think any further flushing
should help with whatever races we have left in there.

   ..psr.dc3co_work

I think the whole dc3co thing should be disabled atm,
so nothing should ever schedule this. We should
probably garbage collect the whole thing...

   ..crtc->drrs.work

That one should have been killed in
intel_display_driver_unregister()
  ->drm_atomic_helper_shutdown()
   ->...
    ->intel_drrs_deactivate()
     ->cancel_delayed_work_sync()

So if I replace it with intel_hpd_cancel_work() that appears would handle
the first two. What about the other two?

Other stuff that comes to mind is the pps vdd_off work.
But looks like that should get taken down in the
encoder->destroy() hook at the latest (via
intel_mode_config_cleanup()).

psr.work at least has a cancel_work_sync() in intel_psr_disable(),
so should hopefully get killed the same way as drrs.

opregion.asle_work seems to get cancelled from the unregister path.

The ones that look broken to me are dmc.work and fbc underrun_work.

Right, so I missed some and things are a bit more complicated. Okay to leave this with you, even if on a backlog?

Regards,

Tvrtko



Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
Cc: Jani Nikula <jani.nikula@xxxxxxxxx>
Cc: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
Cc: Tetsuo Handa <penguin-kernel@xxxxxxxxxxxxxxxxxxx>
---
I am clueless about the display paths and only send this because Jani
convinced me to send a patch to kick off the discussion. No expectations
whatsoever this is correct or complete.
---
  drivers/gpu/drm/i915/display/intel_display.c | 2 +-
  1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
index 2d0018ae34b1..0eb72530a003 100644
--- a/drivers/gpu/drm/i915/display/intel_display.c
+++ b/drivers/gpu/drm/i915/display/intel_display.c
@@ -8980,7 +8980,7 @@ void intel_modeset_driver_remove_noirq(struct drm_i915_private *i915)
  	intel_unregister_dsm_handler();
/* flush any delayed tasks or pending work */
-	flush_scheduled_work();
+	intel_hpd_cancel_work(i915);
intel_hdcp_component_fini(i915); --
2.34.1




[Index of Archives]     [AMD Graphics]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux