[PATCH] drm/xe: Add wait for completion after gt force reset

[Gupta, Anshuman]

> -----Original Message-----
> From: Intel-xe <intel-xe-bounces at lists.freedesktop.org> On Behalf Of Rodrigo
> Vivi
> Sent: Friday, December 15, 2023 11:02 PM
> To: Poosa, Karthik <karthik.poosa at intel.com>
> Cc: intel-xe at lists.freedesktop.org
> Subject: Re: [PATCH] drm/xe: Add wait for completion after gt force reset
> 
> On Fri, Dec 15, 2023 at 11:43:04AM -0500, Rodrigo Vivi wrote:
> > On Fri, Dec 15, 2023 at 10:45:41AM +0530, Karthik Poosa wrote:
> > > Wait for gt reset to complete before returning from force_reset
> > > sysfs call. Without this igt test freq_reset_multiple fails
> > > sporadically in case xe_guc_pc is not started.
> >
> > \o/ I knew I was missing something there. Thanks for finding that out.
> >
> > >
> > > Testcase: igt at xe_guc_pc@freq_reset_multiple
> > > Signed-off-by: Karthik Poosa <karthik.poosa at intel.com>
> > > ---
> > >  drivers/gpu/drm/xe/xe_gt.c         |  3 +++
> > >  drivers/gpu/drm/xe/xe_gt_debugfs.c | 10 ++++++++++
> > >  drivers/gpu/drm/xe/xe_gt_types.h   |  3 +++
> > >  3 files changed, 16 insertions(+)
> > >
> > > diff --git a/drivers/gpu/drm/xe/xe_gt.c b/drivers/gpu/drm/xe/xe_gt.c
> > > index dfd9cf01a5d5..eb7552b6dfa5 100644
> > > --- a/drivers/gpu/drm/xe/xe_gt.c
> > > +++ b/drivers/gpu/drm/xe/xe_gt.c
> > > @@ -65,6 +65,7 @@ struct xe_gt *xe_gt_alloc(struct xe_tile *tile)
> > >
> > >  	gt->tile = tile;
> > >  	gt->ordered_wq = alloc_ordered_workqueue("gt-ordered-wq", 0);
> > > +	init_completion(&gt->reset_done);
> > >
> > >  	return gt;
> > >  }
> > > @@ -647,6 +648,8 @@ static int gt_reset(struct xe_gt *gt)
> > >  	xe_device_mem_access_put(gt_to_xe(gt));
> > >  	XE_WARN_ON(err);
> > >
> > > +	complete(&gt->reset_done);
> > > +
> > >  	xe_gt_info(gt, "reset done\n");
> > >
> > >  	return 0;
> > > diff --git a/drivers/gpu/drm/xe/xe_gt_debugfs.c
> > > b/drivers/gpu/drm/xe/xe_gt_debugfs.c
> > > index c4b67cf09f8f..49b30937a28b 100644
> > > --- a/drivers/gpu/drm/xe/xe_gt_debugfs.c
> > > +++ b/drivers/gpu/drm/xe/xe_gt_debugfs.c
> > > @@ -23,6 +23,8 @@
> > >  #include "xe_uc_debugfs.h"
> > >  #include "xe_wa.h"
> > >
> > > +#define XE_GT_RESET_TIMEOUT_MS	(msecs_to_jiffies(5*1000))
> >
> > 5s seems to much, but anyway, this is a debugfs function, so, no if 5s
> > fail, then we do have bigger problems.
> >
> > perhaps we could get the timeout as input?
How should we get that, I guess debugfs or modparam, sysfs will require a uapi user ?
Thanks,
Anshuman.
> >
> > anyway, it is better to have protected like this than the current racy
> > one that we have.
> >
> > Reviewed-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
> 
> ops, let me tune down my over excitement here.
> We need to address the feedback from Badal that was given at intel-gfx ml.
> 
> >
> > > +
> > >  static struct xe_gt *node_to_gt(struct drm_info_node *node)  {
> > >  	return node->info_ent->data;
> > > @@ -58,9 +60,17 @@ static int hw_engines(struct seq_file *m, void
> > > *data)  static int force_reset(struct seq_file *m, void *data)  {
> > >  	struct xe_gt *gt = node_to_gt(m->private);
> > > +	struct xe_device *xe = gt_to_xe(gt);
> > > +	unsigned long timeout;
> > >
> > >  	xe_gt_reset_async(gt);
> > >
> > > +	timeout = wait_for_completion_timeout(&gt->reset_done,
> XE_GT_RESET_TIMEOUT_MS);
> > > +	if (timeout == 0) {
> > > +		drm_err(&xe->drm, "gt reset timed out");
> > > +		return -ETIMEDOUT;
> > > +	}
> > > +
> > >  	return 0;
> > >  }
> > >
> > > diff --git a/drivers/gpu/drm/xe/xe_gt_types.h
> > > b/drivers/gpu/drm/xe/xe_gt_types.h
> > > index f74684660475..6f2fb9e3cfea 100644
> > > --- a/drivers/gpu/drm/xe/xe_gt_types.h
> > > +++ b/drivers/gpu/drm/xe/xe_gt_types.h
> > > @@ -358,6 +358,9 @@ struct xe_gt {
> > >  		/** @oob: bitmap with active OOB workaroudns */
> > >  		unsigned long *oob;
> > >  	} wa_active;
> > > +
> > > +	/** @reset_done : Completion of GT reset */
> > > +	struct completion reset_done;
> > >  };
> > >
> > >  #endif
> > > --
> > > 2.25.1
> > >