[Intel-gfx] [PATCH 15/26] drm/i915: add media well to VLV force wake routines

[Jesse Barnes]

On Wed, 6 Mar 2013 20:28:07 +0200
Ville Syrjälä <ville.syrjala at linux.intel.com> wrote:

> On Fri, Mar 01, 2013 at 02:08:31PM -0800, Jesse Barnes wrote:
> > We could split this out into a separate routine at some point as an
> > optimization.
> 
> BTW did anyone try to gang wakeup thing instead?

Not afaik.

> 
> > 
> > Signed-off-by: Jesse Barnes <jbarnes at virtuousgeek.org>
> > ---
> >  drivers/gpu/drm/i915/i915_reg.h |    2 ++
> >  drivers/gpu/drm/i915/intel_pm.c |   11 ++++++++---
> >  2 files changed, 10 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> > index 1c6e066..558c6d1 100644
> > --- a/drivers/gpu/drm/i915/i915_reg.h
> > +++ b/drivers/gpu/drm/i915/i915_reg.h
> > @@ -4271,6 +4271,8 @@
> >  #define  FORCEWAKE				0xA18C
> >  #define  FORCEWAKE_VLV				0x1300b0
> >  #define  FORCEWAKE_ACK_VLV			0x1300b4
> > +#define  FORCEWAKE_MEDIA_VLV			0x1300b8
> > +#define  FORCEWAKE_ACK_MEDIA_VLV		0x1300bc
> >  #define  FORCEWAKE_ACK_HSW			0x130044
> >  #define  FORCEWAKE_ACK				0x130090
> >  #define  VLV_GTLC_WAKE_CTRL			0x130090
> > diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
> > index 3e976f4..e3947cb 100644
> > --- a/drivers/gpu/drm/i915/intel_pm.c
> > +++ b/drivers/gpu/drm/i915/intel_pm.c
> > @@ -4423,10 +4423,15 @@ static void vlv_force_wake_get(struct drm_i915_private *dev_priv)
> >  		DRM_ERROR("Timed out waiting for forcewake old ack to clear.\n");
> >  
> >  	I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
> > +	I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV, _MASKED_BIT_ENABLE(FORCEWAKE_KERNEL));
> >  
> >  	if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_VLV) & 1),
> >  			    FORCEWAKE_ACK_TIMEOUT_MS))
> > -		DRM_ERROR("Timed out waiting for forcewake to ack request.\n");
> > +		DRM_ERROR("Timed out waiting for GT to ack forcewake request.\n");
> > +
> > +	if (wait_for_atomic((I915_READ_NOTRACE(FORCEWAKE_ACK_MEDIA_VLV) & 1),
> > +			    FORCEWAKE_ACK_TIMEOUT_MS))
> > +		DRM_ERROR("Timed out waiting for media to ack forcewake request.\n");
> 
> Based on some of my recent patches the '& 1's should be
> '& FORCEWAKE_KERNEL'.

And that was based on my misunderstanding of MT forcewake.  I thought
it was so the BIOS or AMT could do forcewake, but it's actually meant
for multiple kernel thread accesses.  Since we don't do that, simply
using the lowest bit all the time is fine.

> >  	__gen6_gt_wait_for_thread_c0(dev_priv);
> >  }
> > @@ -4434,8 +4439,8 @@ static void vlv_force_wake_get(struct drm_i915_private *dev_priv)
> >  static void vlv_force_wake_put(struct drm_i915_private *dev_priv)
> >  {
> >  	I915_WRITE_NOTRACE(FORCEWAKE_VLV, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
> > -	/* something from same cacheline, but !FORCEWAKE_VLV */
> > -	POSTING_READ(FORCEWAKE_ACK_VLV);
> > +	I915_WRITE_NOTRACE(FORCEWAKE_MEDIA_VLV, _MASKED_BIT_DISABLE(FORCEWAKE_KERNEL));
> > +	/* The below doubles as a POSTING_READ */
> 
> Are we sure? ;)
> 
> >  	gen6_gt_check_fifodbg(dev_priv);

Well it does a read first thing so I think so.

-- 
Jesse Barnes, Intel Open Source Technology Center