[PATCH] drm/xe: Add NULL check before deferencing pointer

[Singh, Apoorva]

> -----Original Message-----
> From: Jani Nikula <jani.nikula at linux.intel.com>
> Sent: Monday, August 19, 2024 2:28 PM
> To: Singh, Apoorva <apoorva.singh at intel.com>; intel-
> xe at lists.freedesktop.org
> Cc: Singh, Apoorva <apoorva.singh at intel.com>
> Subject: Re: [PATCH] drm/xe: Add NULL check before deferencing pointer
> 
> On Mon, 19 Aug 2024, apoorva.singh at intel.com wrote:
> > From: Apoorva Singh <apoorva.singh at intel.com>
> 
> *dereferencing typo in the subject. You could also be more specific, that
> subject could refer to anything anywhere in the driver. If you find another
> one, what's its subject going to be?

Thank you for pointing out. I will take care in next revision.
> 
> > Add NULL check before deferencing of lrc->bo in
> > xe_lrc_snapshot_capture().
> >
> > Free the dynamically allocated memory for snapshot to avoid memory
> > leak
> 
> We've seen this patch or its variants many times now. Where's the
> changelog? What's going on?

Patch variants happened due to change in subject as per addressing the comments by Nirmoy and Rodrigo.
But finally, its concluded to not remove the if (lrc->bo) from above and instead add in the position mentioned in this patch.

> 
> > Signed-off-by: Apoorva Singh <apoorva.singh at intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_lrc.c | 32 ++++++++++++++++++--------------
> >  1 file changed, 18 insertions(+), 14 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/xe/xe_lrc.c b/drivers/gpu/drm/xe/xe_lrc.c
> > index 974a9cd8c379..e31d9bc64ff6 100644
> > --- a/drivers/gpu/drm/xe/xe_lrc.c
> > +++ b/drivers/gpu/drm/xe/xe_lrc.c
> > @@ -1652,20 +1652,24 @@ struct xe_lrc_snapshot
> *xe_lrc_snapshot_capture(struct xe_lrc *lrc)
> >  	if (lrc->bo && lrc->bo->vm)
> >  		xe_vm_get(lrc->bo->vm);
> >
> > -	snapshot->context_desc = xe_lrc_ggtt_addr(lrc);
> > -	snapshot->indirect_context_desc =
> xe_lrc_indirect_ring_ggtt_addr(lrc);
> > -	snapshot->head = xe_lrc_ring_head(lrc);
> > -	snapshot->tail.internal = lrc->ring.tail;
> > -	snapshot->tail.memory = xe_lrc_ring_tail(lrc);
> > -	snapshot->start_seqno = xe_lrc_start_seqno(lrc);
> > -	snapshot->seqno = xe_lrc_seqno(lrc);
> > -	snapshot->lrc_bo = xe_bo_get(lrc->bo);
> > -	snapshot->lrc_offset = xe_lrc_pphwsp_offset(lrc);
> > -	snapshot->lrc_size = lrc->bo->size - snapshot->lrc_offset;
> > -	snapshot->lrc_snapshot = NULL;
> > -	snapshot->ctx_timestamp = xe_lrc_ctx_timestamp(lrc);
> > -	snapshot->ctx_job_timestamp = xe_lrc_ctx_job_timestamp(lrc);
> > -	return snapshot;
> > +	if (lrc->bo) {
> 
> Please handle error cases in if branches with early returns, and do the happy
> day scenario in the top indentation level.	

Sure.

Thanks,
Apoorva
> 
> 
> BR,
> Jani.
> 
> 
> > +		snapshot->context_desc = xe_lrc_ggtt_addr(lrc);
> > +		snapshot->indirect_context_desc =
> xe_lrc_indirect_ring_ggtt_addr(lrc);
> > +		snapshot->head = xe_lrc_ring_head(lrc);
> > +		snapshot->tail.internal = lrc->ring.tail;
> > +		snapshot->tail.memory = xe_lrc_ring_tail(lrc);
> > +		snapshot->start_seqno = xe_lrc_start_seqno(lrc);
> > +		snapshot->seqno = xe_lrc_seqno(lrc);
> > +		snapshot->lrc_bo = xe_bo_get(lrc->bo);
> > +		snapshot->lrc_offset = xe_lrc_pphwsp_offset(lrc);
> > +		snapshot->lrc_size = lrc->bo->size - snapshot->lrc_offset;
> > +		snapshot->lrc_snapshot = NULL;
> > +		snapshot->ctx_timestamp = xe_lrc_ctx_timestamp(lrc);
> > +		snapshot->ctx_job_timestamp =
> xe_lrc_ctx_job_timestamp(lrc);
> > +		return snapshot;
> > +	}
> > +	kfree(snapshot);
> > +	return NULL;
> >  }
> >
> >  void xe_lrc_snapshot_capture_delayed(struct xe_lrc_snapshot
> > *snapshot)
> 
> --
> Jani Nikula, Intel