[Intel-gfx] [PATCH 1/2] drm/i915/guc: Don't check CT descriptor status before CT write / read

[Jani Nikula]

On Fri, 21 Jan 2022, Matthew Brost <matthew.brost at intel.com> wrote:
> On Fri, Jan 21, 2022 at 09:28:46AM +0200, Jani Nikula wrote:
>> On Thu, 20 Jan 2022, Matthew Brost <matthew.brost at intel.com> wrote:
>> > Don't check CT descriptor status, unless CONFIG_DRM_I915_DEBUG_GUC is
>> > set, before CT write / read as this could result in a read across the
>> > PCIe bus thus adding latency to every CT write / read. On well behavied
>> > systems this vaue should always read as zero. For some reason it doesn't
>> > the CT channel is broken and will eventually recover from a GT reset,
>> > albeit the GT reset will not be triggered immediately by seeing that
>> > descriptor status is non-zero.
>> >
>> > v2:
>> >  (CI)
>> >   - Fix build error (hide corrupted label in write function behind
>> >     CONFIG_DRM_I915_DEBUG_GUC)
>> >
>> > Signed-off-by: Matthew Brost <matthew.brost at intel.com>
>> > ---
>> >  drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c | 6 ++++++
>> >  1 file changed, 6 insertions(+)
>> >
>> > diff --git a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
>> > index de89d40abd38d..948cf31429412 100644
>> > --- a/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
>> > +++ b/drivers/gpu/drm/i915/gt/uc/intel_guc_ct.c
>> > @@ -379,8 +379,10 @@ static int ct_write(struct intel_guc_ct *ct,
>> >  	u32 *cmds = ctb->cmds;
>> >  	unsigned int i;
>> >  
>> > +#ifdef CONFIG_DRM_I915_DEBUG_GUC
>> >  	if (unlikely(desc->status))
>> >  		goto corrupted;
>> > +#endif
>> 
>> Please don't add #ifdefs inline. You can use
>> IS_ENABLED(CONFIG_DRM_I915_DEBUG_GUC) in if statements, but otherwise
>> the code needs to be split out to a separate function.
>> 
>
> Sure, but I feel like I've actually been by someone else to not use the
> IS_ENABLED macro and use ifdefs inlines...

Citation needed.

Basically never use #if/#ifdef inline. Only use them at the top level
like this:

#if IS_ENABLED(CONFIG_FOO)
static int bar(void)
{
	/* implementation with foo */
}
#else
static int bar(void)
{
	/* implementation without foo */
}
#endif

Sometimes you can avoid the above boilerplate with IS_ENABLED() inline:

	if (IS_ENABLED(CONFIG_FOO))
		...

Basically if you think #if/#ifdef inline is the easiest, you need to
refactor the code to do it cleanly without them.

BR,
Jani.


>
> Matt
>
>> BR,
>> Jani.
>> 
>> >  
>> >  	GEM_BUG_ON(tail > size);
>> >  
>> > @@ -445,11 +447,13 @@ static int ct_write(struct intel_guc_ct *ct,
>> >  
>> >  	return 0;
>> >  
>> > +#ifdef CONFIG_DRM_I915_DEBUG_GUC
>> >  corrupted:
>> >  	CT_ERROR(ct, "Corrupted descriptor head=%u tail=%u status=%#x\n",
>> >  		 desc->head, desc->tail, desc->status);
>> >  	ctb->broken = true;
>> >  	return -EPIPE;
>> > +#endif
>> >  }
>> >  
>> >  /**
>> > @@ -815,8 +819,10 @@ static int ct_read(struct intel_guc_ct *ct, struct ct_incoming_msg **msg)
>> >  	if (unlikely(ctb->broken))
>> >  		return -EPIPE;
>> >  
>> > +#ifdef CONFIG_DRM_I915_DEBUG_GUC
>> >  	if (unlikely(desc->status))
>> >  		goto corrupted;
>> > +#endif
>> >  
>> >  	GEM_BUG_ON(head > size);
>> 
>> -- 
>> Jani Nikula, Intel Open Source Graphics Center

-- 
Jani Nikula, Intel Open Source Graphics Center