Re: [PATCH v1] drm/xe: Introduce xe_wedge

[Raag Jadav <raag.jadav@intel.com>]

On Thu, Aug 27, 2026 at 07:38:14PM +0200, Michal Wajdeczko wrote:
> On 8/25/2026 1:42 PM, Raag Jadav wrote:
> > Consolidates all wedging implementation into a dedicated xe_wedge
> > component. While at it, add a worker to schedule the wedge handling to

...

> > +void xe_device_set_wedged_method(struct xe_device *xe, unsigned long method)
> 
> as we are adding/renaming functions, should we combine two old:
> 
> 	xe_device_declare_wedged(xe)
> and	xe_device_set_wedged_method(xe, method)
> 
> functions into single new:
> 
> 	xe_wedge_declare(xe, method)
> 
> as we either already know the recovery method or can use
> default one

This was already on my backlog after FLR, including other improvements
to xe_gt_declare_wedged().

...

> > +static void wedged_work(struct work_struct *work)
> > +{
> > +	struct xe_device *xe = container_of(work, struct xe_device, wedged.work);
> > +	struct xe_gt *gt;
> > +	u8 id;
> > +
> > +	for_each_gt(gt, xe, id)
> > +		xe_gt_declare_wedged(gt);
> 
> is it ok to do that in the async worker?

There are mutexes deep inside it. I have a cleanup in mind but it
requires landing FLR series.

...

> > +int xe_device_validate_wedged_mode(struct xe_device *xe, unsigned int mode)
> 
> maybe this can be static - all we need is to move here also
> debugfs stuff that adds "wedged_mode" attribute

Shouldn't it be the other way around? Keep it consolidated into debugfs?

Raag