[Intel-gfx] [PATCH 4/5] drm/i915: Let number of workarounds more clear

[Chris Wilson]

On Fri, Sep 26, 2014 at 05:16:57PM +0300, Mika Kuoppala wrote:
> Jani Nikula <jani.nikula at linux.intel.com> writes:
> 
> > On Fri, 26 Sep 2014, Mika Kuoppala <mika.kuoppala at linux.intel.com> wrote:
> >> Rodrigo Vivi <rodrigo.vivi at intel.com> writes:
> >>
> >>> This helps when including or removing cs workarounds.
> >>> Signed-off-by: Rodrigo Vivi <rodrigo.vivi at intel.com>
> >>> ---
> >>>  drivers/gpu/drm/i915/intel_ringbuffer.c | 16 ++++++++++++----
> >>>  1 file changed, 12 insertions(+), 4 deletions(-)
> >>>
> >>> diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
> >>> index 7c3d17a..39fbea6 100644
> >>> --- a/drivers/gpu/drm/i915/intel_ringbuffer.c
> >>> +++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
> >>> @@ -694,6 +694,7 @@ static int bdw_init_workarounds(struct intel_engine_cs *ring)
> >>>  	int ret;
> >>>  	struct drm_device *dev = ring->dev;
> >>>  	struct drm_i915_private *dev_priv = dev->dev_private;
> >>> +	int wa_amount;
> >>>  
> >>>  	/*
> >>>  	 * workarounds applied in this fn are part of register state context,
> >>> @@ -704,10 +705,11 @@ static int bdw_init_workarounds(struct intel_engine_cs *ring)
> >>>  	memset(dev_priv->intel_wa_regs, 0, sizeof(dev_priv->intel_wa_regs));
> >>>  
> >>>  	/*
> >>> -	 * update the number of dwords required based on the
> >>> -	 * actual number of workarounds applied
> >>> +	 * update the number of workarounds when adding or removing was
> >>> +	 * in order the have propper dwords
> >>>  	 */
> >>> -	ret = intel_ring_begin(ring, 24);
> >>> +	wa_amount = 8;
> >>> +	ret = intel_ring_begin(ring, 3 * wa_amount);
> >>>  	if (ret)
> >>>  		return ret;
> >>>  
> >>
> >> I have a bit mixed feelings with this patch as I have tripped
> >> around here myself recently.
> >>
> >> I think we should just drop this patch and use:
> >> ret = intel_ring_begin(ring, 3 * <amount>) on the subsequent
> >> patches that modify the amount.
> >>
> >> This way the intel_ring_begin will always be the one that
> >> shows in a diff. And raises alarm if it doesnt. We don't want
> >> this to be too transparent and looking too easy for the next reader.
> >> As the reviewer is the only and last line of defense ensuring
> >> symmetry between intel_ring_begin and amount of emits.
> >
> > Considering intel_ring_emit_wa() adds all the needed information to
> > dev_priv->intel_wa_regs and dev_priv->num_wa_regs anyway, we could
> > trivially split this into constructing dev_priv->intel_wa_regs first,
> > and doing intel_ring_begin() and the intel_ring_emit()s next, and not
> > worry about this again.
> 
> My RFC series of fixing the workarounds after reset/suspend does this.

And so did mine! :-p
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre