[Intel-gfx] [PATCH v2 14/14] drm/i915/fbc: Reallocate cfb if we need more of it

[Maarten Lankhorst]

Op 28-11-2019 om 16:59 schreef Ville Syrjälä:
> On Thu, Nov 28, 2019 at 04:48:04PM +0100, Maarten Lankhorst wrote:
>> Op 27-11-2019 om 21:12 schreef Ville Syrjala:
>>> From: Ville Syrjälä <ville.syrjala at linux.intel.com>
>>>
>>> The code assumes we can omit the cfb allocation once fbc
>>> has been enabled once. That's nonsense. Let's try to
>>> reallocate it if we need to.
>>>
>>> The code is still a mess, but maybe this is enough to get
>>> fbc going in some cases where it initially underallocates
>>> the cfb and there's no full modeset to fix it up.
>>>
>>> Cc: Daniel Drake <drake at endlessm.com>
>>> Cc: Paulo Zanoni <paulo.r.zanoni at intel.com>
>>> Cc: Jian-Hong Pan <jian-hong at endlessm.com>
>>> Cc: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
>>> Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
>>> ---
>>>  drivers/gpu/drm/i915/display/intel_fbc.c | 22 +++++++++++++++-------
>>>  1 file changed, 15 insertions(+), 7 deletions(-)
>>>
>>> diff --git a/drivers/gpu/drm/i915/display/intel_fbc.c b/drivers/gpu/drm/i915/display/intel_fbc.c
>>> index c976698b0729..928059a5da80 100644
>>> --- a/drivers/gpu/drm/i915/display/intel_fbc.c
>>> +++ b/drivers/gpu/drm/i915/display/intel_fbc.c
>>> @@ -672,6 +672,14 @@ static void intel_fbc_update_state_cache(struct intel_crtc *crtc,
>>>  		cache->fence_id = -1;
>>>  }
>>>  
>>> +static bool intel_fbc_cfb_size_changed(struct drm_i915_private *dev_priv)
>>> +{
>>> +	struct intel_fbc *fbc = &dev_priv->fbc;
>>> +
>>> +	return intel_fbc_calculate_cfb_size(dev_priv, &fbc->state_cache) >
>>> +		fbc->compressed_fb.size * fbc->threshold;
>>> +}
>>> +
>>>  static bool intel_fbc_can_activate(struct intel_crtc *crtc)
>>>  {
>>>  	struct drm_i915_private *dev_priv = to_i915(crtc->base.dev);
>>> @@ -757,8 +765,7 @@ static bool intel_fbc_can_activate(struct intel_crtc *crtc)
>>>  	 * we didn't get any invalidate/deactivate calls, but this would require
>>>  	 * a lot of tracking just for a specific case. If we conclude it's an
>>>  	 * important case, we can implement it later. */
>>> -	if (intel_fbc_calculate_cfb_size(dev_priv, &fbc->state_cache) >
>>> -	    fbc->compressed_fb.size * fbc->threshold) {
>>> +	if (intel_fbc_cfb_size_changed(dev_priv)) {
>>>  		fbc->no_fbc_reason = "CFB requirements changed";
>>>  		return false;
>>>  	}
>>> @@ -1112,12 +1119,12 @@ void intel_fbc_enable(struct intel_crtc *crtc,
>>>  	mutex_lock(&fbc->lock);
>>>  
>>>  	if (fbc->crtc) {
>>> -		WARN_ON(fbc->crtc == crtc && !crtc_state->enable_fbc);
>>> -		goto out;
>>> -	}
>>> +		if (fbc->crtc != crtc ||
>>> +		    !intel_fbc_cfb_size_changed(dev_priv))
>>> +			goto out;
>>>  
>>> -	if (!crtc_state->enable_fbc)
>>> -		goto out;
>>> +		__intel_fbc_disable(dev_priv);
>>> +	}
>>>  
>>>  	WARN_ON(fbc->active);
>>>  
>>> @@ -1130,6 +1137,7 @@ void intel_fbc_enable(struct intel_crtc *crtc,
>>>  	if (intel_fbc_alloc_cfb(dev_priv,
>>>  				intel_fbc_calculate_cfb_size(dev_priv, cache),
>>>  				fb->format->cpp[0])) {
>>> +		cache->plane.visible = false;
>>>  		fbc->no_fbc_reason = "not enough stolen memory";
>>>  		goto out;
>>>  	}
>> Makes sense, unfortunately kms_cursor_legacy starts failing on this series. :(
>>
>> For 1-11, 14
>>
>> Reviewed-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
>>
>> We should probably get rid of the FBC disable on frontbuffer disable as well. I had some patches but nothing upstream-worthy yet. :(
> How would we get rid of the disable there? By triggering nukes at some
> predefined interval? Doesn't sound all that great.
Not touching FBC on frontbuffer write at all, and forcing userspace to use the dirtyfb api. I think the whole implicit tracking should be removed.
>
>> 12  and 13 need more thought for now, kms_cursor_legacy is failing.
> Already posted the v2 that fixes it.
>
>