[Intel-gfx] [PATCH 2/2] drm/i915: Move modeset state checker calls.

[Maarten Lankhorst]

Op 08-04-16 om 01:23 schreef Matt Roper:
> On Wed, Mar 23, 2016 at 02:58:07PM +0100, Maarten Lankhorst wrote:
>> The modeset state checker no longer has full access to the hardware,
>> instead it should only check affected crtc's.
>>
>> Looking for disabled stuff can be checked immediately after all crtc
>> disables have completed, while each enabled crtc can be checked right
>> after being enabled.
>>
>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst at linux.intel.com>
>> ---
>>  drivers/gpu/drm/i915/intel_display.c | 24 +++++-------------------
>>  1 file changed, 5 insertions(+), 19 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/i915/intel_display.c b/drivers/gpu/drm/i915/intel_display.c
>> index 4148b262f2a7..b6a75aa32e5a 100644
>> --- a/drivers/gpu/drm/i915/intel_display.c
>> +++ b/drivers/gpu/drm/i915/intel_display.c
>> @@ -13027,28 +13027,13 @@ check_disabled_dpll_state(struct drm_device *dev)
>>  }
>>  
>>  static void
>> -intel_modeset_check_disabled(struct drm_device *dev,
>> -			     struct drm_atomic_state *old_state)
>> +intel_modeset_check_disabled(struct drm_device *dev)
>>  {
>>  	check_encoder_state(dev);
>>  	check_connector_state(dev, NULL);
>>  	check_disabled_dpll_state(dev);
>>  }
>>  
>> -static void
>> -intel_modeset_check_state(struct drm_device *dev,
>> -			  struct drm_atomic_state *old_state)
>> -{
>> -	struct drm_crtc_state *old_crtc_state;
>> -	struct drm_crtc *crtc;
>> -	int i;
>> -
>> -	for_each_crtc_in_state(old_state, crtc, old_crtc_state, i)
>> -		intel_modeset_check_crtc(crtc, old_crtc_state, crtc->state);
>> -
>> -	intel_modeset_check_disabled(dev, old_state);
>> -}
>> -
>>  static void update_scanline_offset(struct intel_crtc *crtc)
>>  {
>>  	struct drm_device *dev = crtc->base.dev;
>> @@ -13616,6 +13601,8 @@ static int intel_atomic_commit(struct drm_device *dev,
>>  		if (dev_priv->display.modeset_commit_cdclk &&
>>  		    intel_state->dev_cdclk != dev_priv->cdclk_freq)
>>  			dev_priv->display.modeset_commit_cdclk(state);
>> +
>> +		intel_modeset_check_disabled(dev);
> As noted on patch #1, since we're checking all disabled state (and not
> just the things that are newly-disabled by this transaction), we may not
> hold full locks over everything that's disabled here and might be racing
> with other updates.  But we're no worse off than we were before.
>
> As a minor behavior change, it looks like we're also no longer calling
> this when we're doing non-modeset pipe updates, but that should be fine.
>
This is true, but enabling a pipe requires connection_mutex, which is always held if we do a modeset anyway..