Re: [Intel-gfx] [PATCH v6 10/11] drm/i915: Add intel_update_bigjoiner handling.

["Navare, Manasi" <manasi.d.navare@intel.com>]

On Tue, Sep 22, 2020 at 01:27:35PM +0300, Ville Syrjälä wrote:
> On Mon, Sep 21, 2020 at 02:18:33PM -0700, Navare, Manasi wrote:
> > On Mon, Sep 14, 2020 at 12:21:26PM -0700, Navare, Manasi wrote:
> > > On Thu, Sep 03, 2020 at 10:23:35PM +0300, Ville Syrjälä wrote:
> > > > On Wed, Jul 15, 2020 at 03:42:21PM -0700, Manasi Navare wrote:
> > > > > From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> > > > > 
> > > > > Enabling is done in a special sequence and so should plane updates
> > > > > be. Ideally the end user never notices the second pipe is used,
> > > > > so use the vblank evasion to cover both pipes.
> > > > > 
> > > > > This way ideally everything will be tear free, and updates are
> > > > > really atomic as userspace expects it.
> > > > > 
> > > > > ****This needs to be checked if it still works since lot of refactoring
> > > > > in skl_commit_modeset_enables
> > > > > 
> > > > > v2:
> > > > > * Manual Rebase (Manasi)
> > > > > * Refactoring on intel_update_crtc and enable_crtc and removing
> > > > > special trans_port_sync_update (Manasi)
> > > > > 
> > > > > Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
> > > > > Signed-off-by: Manasi Navare <manasi.d.navare@intel.com>
> > > > > ---
> > > > >  drivers/gpu/drm/i915/display/intel_display.c | 120 +++++++++++++++++--
> > > > >  drivers/gpu/drm/i915/display/intel_sprite.c  |  25 +++-
> > > > >  drivers/gpu/drm/i915/display/intel_sprite.h  |   3 +-
> > > > >  3 files changed, 129 insertions(+), 19 deletions(-)
> > > > > 
> > > > > diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> > > > > index a1011414da6d..00b26863ffc6 100644
> > > > > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > > > > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > > > > @@ -15656,7 +15656,7 @@ static void intel_update_crtc(struct intel_atomic_state *state,
> > > > >  	else
> > > > >  		i9xx_update_planes_on_crtc(state, crtc);
> > > > >  
> > > > > -	intel_pipe_update_end(new_crtc_state);
> > > > > +	intel_pipe_update_end(new_crtc_state, NULL);
> > > > >  
> > > > >  	/*
> > > > >  	 * We usually enable FIFO underrun interrupts as part of the
> > > > > @@ -15754,6 +15754,52 @@ static void intel_commit_modeset_disables(struct intel_atomic_state *state)
> > > > >  	}
> > > > >  }
> > > > >  
> > > > > +static void intel_update_bigjoiner(struct intel_crtc *crtc,
> > > > > +				   struct intel_atomic_state *state,
> > > > > +				   struct intel_crtc_state *old_crtc_state,
> > > > > +				   struct intel_crtc_state *new_crtc_state)
> > > > > +{
> > > > > +	struct drm_i915_private *dev_priv = to_i915(state->base.dev);
> > > > > +	bool modeset = needs_modeset(new_crtc_state);
> > > > > +	struct intel_crtc *slave = new_crtc_state->bigjoiner_linked_crtc;
> > > > > +	struct intel_crtc_state *new_slave_crtc_state =
> > > > > +		intel_atomic_get_new_crtc_state(state, slave);
> > > > > +
> > > > > +	if (modeset) {
> > > > > +		/* Enable slave first */
> > > > > +		intel_crtc_update_active_timings(new_slave_crtc_state);
> > > > > +		dev_priv->display.crtc_enable(state, slave);
> > > > > +
> > > > > +		/* Then master */
> > > > > +		intel_crtc_update_active_timings(new_crtc_state);
> > > > > +		dev_priv->display.crtc_enable(state, crtc);
> > > > > +
> > > > > +		/* vblanks work again, re-enable pipe CRC. */
> > > > > +		intel_crtc_enable_pipe_crc(crtc);
> > > > > +
> > > > > +	} else {
> > > > > +		intel_pre_plane_update(state, crtc);
> > > > > +		intel_pre_plane_update(state, slave);
> > > > > +
> > > > > +		if (new_crtc_state->update_pipe)
> > > > > +			intel_encoders_update_pipe(state, crtc);
> > > > > +	}
> > > > > +
> > > > > +	/*
> > > > > +	 * Perform vblank evasion around commit operation, and make sure to
> > > > > +	 * commit both planes simultaneously for best results.
> > > > > +	 */
> > > > > +	intel_pipe_update_start(new_crtc_state);
> > > > > +
> > > > > +	commit_pipe_config(state, crtc);
> > > > > +	commit_pipe_config(state, slave);
> > > > > +
> > > > > +	skl_update_planes_on_crtc(state, crtc);
> > > > > +	skl_update_planes_on_crtc(state, slave);
> > > > > +
> > > > > +	intel_pipe_update_end(new_crtc_state, new_slave_crtc_state);
> > > > > +}
> > > > 
> > > > I think this should ideally all go away and just the normal logic
> > > > in commit_modeset_enables() should deal with it. Just like it does
> > > > for port sync/mst pipe dependencies.
> > > >
> > > 
> > > Yes I think so too. Except for the intel_pipe_update_end where
> > > now we send the new_slave_crtc_state() so thats still something I need to figure
> > > how it will work in normal code without special bigjoiner handling.
> > > 
> > > I think the 2p2p transcoder ports sync code initially had a special trans port sync handling
> > > function and thats why this was written this way but with the new code we just use
> > > the regular modeset_enables function with no special handling
> > > 
> > > Manasi
> > >  
> > > > > +
> > > > >  static void intel_commit_modeset_enables(struct intel_atomic_state *state)
> > > > >  {
> > > > >  	struct intel_crtc_state *new_crtc_state;
> > > > > @@ -15772,15 +15818,22 @@ static void intel_commit_modeset_enables(struct intel_atomic_state *state)
> > > > >  static void skl_commit_modeset_enables(struct intel_atomic_state *state)
> > > > >  {
> > > > >  	struct drm_i915_private *dev_priv = to_i915(state->base.dev);
> > > > > -	struct intel_crtc *crtc;
> > > > > +	struct intel_crtc *crtc, *slave;
> > > > >  	struct intel_crtc_state *old_crtc_state, *new_crtc_state;
> > > > >  	struct skl_ddb_entry entries[I915_MAX_PIPES] = {};
> > > > > +	struct skl_ddb_entry new_entries[I915_MAX_PIPES] = {};
> > > > >  	u8 update_pipes = 0, modeset_pipes = 0;
> > > > > +	const struct intel_crtc_state *slave_crtc_state;
> > > > >  	int i;
> > > > >  
> > > > >  	for_each_oldnew_intel_crtc_in_state(state, crtc, old_crtc_state, new_crtc_state, i) {
> > > > >  		enum pipe pipe = crtc->pipe;
> > > > >  
> > > > > +		if (new_crtc_state->bigjoiner_slave) {
> > > > > +			/* We're updated from master */
> > > > > +			continue;
> > > > > +		}
> > > > > +
> > > > >  		if (!new_crtc_state->hw.active)
> > > > >  			continue;
> > > > >  
> > > > > @@ -15791,6 +15844,34 @@ static void skl_commit_modeset_enables(struct intel_atomic_state *state)
> > > > >  		} else {
> > > > >  			modeset_pipes |= BIT(pipe);
> > > > >  		}
> > > > > +
> > > > > +		if (new_crtc_state->bigjoiner) {
> > > > > +			slave = new_crtc_state->bigjoiner_linked_crtc;
> > > > > +			slave_crtc_state =
> > > > > +				intel_atomic_get_new_crtc_state(state,
> > > > > +								slave);
> > > > > +
> > > > > +			/* put both entries in */
> > > > > +			new_entries[i].start = new_crtc_state->wm.skl.ddb.start;
> > > > > +			new_entries[i].end = slave_crtc_state->wm.skl.ddb.end;
> > > > > +		} else {
> > > > > +			new_entries[i] = new_crtc_state->wm.skl.ddb;
> > > > > +		}
> > > > > +
> > > > > +		/* ignore allocations for crtc's that have been turned off during modeset. */
> > > > > +		if (needs_modeset(new_crtc_state))
> > > > > +			continue;
> > > > > +
> > > > > +		if (old_crtc_state->bigjoiner) {
> > > > > +			slave = old_crtc_state->bigjoiner_linked_crtc;
> > > > > +			slave_crtc_state =
> > > > > +				intel_atomic_get_old_crtc_state(state, slave);
> > > > > +
> > > > > +			entries[i].start = old_crtc_state->wm.skl.ddb.start;
> > > > > +			entries[i].end = slave_crtc_state->wm.skl.ddb.end;
> > > > > +		} else {
> > > > > +			entries[i] = old_crtc_state->wm.skl.ddb;
> > > > > +		}
> > > > 
> > > > Why is this here? Can't see why the current code wouldn't work just fine
> > > > for bigjoiner too.
> > > >
> > 
> > Ville, could you provide inputs on how intel_pipe_update_end() should change so that we can use
> > the current code, now this takes an additional input new_slave_crtc_state
> 
> I would not change it at all. What I would do as the first step is to
> treat the pipes entirely separately, just like we do for port sync/etc.
> Later we can think what would be needed to make 100% sure they update
> atomically.

Okay yes so just do the updates sequentially first do all slaves then master like
we do for trans port sync and see what happens without changing
anything special in pipe_update_start and pipe_update_end functions ?

Manasi
> 
> -- 
> Ville Syrjälä
> Intel
_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx