Re: [PATCH v2 1/2] drm/xe/gt: Synchronize GT reset with device unbind

[Matthew Brost <matthew.brost@intel.com>]

On Fri, Oct 31, 2025 at 11:17:02AM -0500, Lucas De Marchi wrote:
> On Fri, Oct 31, 2025 at 08:39:17AM -0700, Matthew Brost wrote:
> > On Thu, Oct 30, 2025 at 08:41:24PM +0530, Balasubramani Vivekanandan wrote:
> > > When unbinding wait for any GT reset in progress to complete. Unbinding
> > > will release the mmio mapping but mmio operations are performed during
> > > GT reset causing Kernel panic.
> > > 
> > > Cc: Lucas De Marchi <lucas.demarchi@intel.com>
> > > Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
> > > ---
> > > v2:
> > >  - Use the managed resource release function to wait for GT reset during
> > >    unbind (Lucas)
> > > ---
> > >  drivers/gpu/drm/xe/xe_gt.c | 2 ++
> > >  1 file changed, 2 insertions(+)
> > > 
> > > diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> > > index 89808b33d0a8..d0f8c40bc51e 100644
> > > --- a/drivers/gpu/drm/xe/xe_gt.c
> > > +++ b/drivers/gpu/drm/xe/xe_gt.c
> > > @@ -607,6 +607,8 @@ static void xe_gt_fini(void *arg)
> > >  	struct xe_gt *gt = arg;
> > >  	int i;
> > > 
> > > +	disable_work_sync(&gt->reset.worker);
> > 
> > We recently made a change [1] so if GT reset worker is queued it holds a PM ref.
> > 
> > So if disable_work_sync returns true [2], I think a PM ref needs to be dropped here.
> > 
> > Matt
> > 
> > [1] https://patchwork.freedesktop.org/patch/682622/?series=156292&rev=1
> > [2] https://elixir.bootlin.com/linux/v6.17.6/source/kernel/workqueue.c#L4471
> 
> yep, it looks like there are 2 v2 of this series. I replied that
> yesterday in the other thread:
> 
> https://lore.kernel.org/all/faewe43nslu62dt5donrrddkbzspnwicrsrm3llwp4xg4tdq7l@zxc7u6vd5ceh
> 

+1, we have the same comment.

Matt

> Lucas De Marchi
> 
> > 
> > > 
> > >  	for (i = 0; i < XE_ENGINE_CLASS_MAX; ++i)
> > >  		xe_hw_fence_irq_finish(&gt->fence_irq[i]);
> > > 
> > > --
> > > 2.43.0
> > >