Re: [PATCH 8/9] drm/i915/display: Wait for scl start instead of dsb_wait_vblanks

["Ville Syrjälä" <ville.syrjala@linux.intel.com>]

On Mon, Sep 22, 2025 at 04:49:46PM +0300, Ville Syrjälä wrote:
> On Mon, Sep 22, 2025 at 04:44:19PM +0300, Ville Syrjälä wrote:
> > On Mon, Sep 22, 2025 at 07:04:49PM +0530, Nautiyal, Ankit K wrote:
> > > 
> > > On 9/22/2025 3:56 PM, Ville Syrjälä wrote:
> > > > On Sun, Sep 21, 2025 at 10:05:34AM +0530, Ankit Nautiyal wrote:
> > > >> Until LNL, intel_dsb_wait_vblanks() waits for the undelayed vblank start.
> > > >> However, from PTL onwards, it waits for the start of the safe window,
> > > >> defined by the number of lines programmed in TRANS_SET_CONTEXT_LATENCY.
> > > >> This change was introduced to move the SCL window out of the vblank region,
> > > >> supporting modes with higher refresh rates and smaller vblanks.
> > > >>
> > > >> As a result, on PTL+ platforms, the DSB wait for vblank completes exactly
> > > >> SCL lines earlier than the undelayed vblank start. Since we use
> > > >> intel_dsb_wait_vblanks() to time the send push operation, this causes
> > > >> issues when SCL lines are non-zero.
> > > >>
> > > >> Instead of relying on the helper, instruct the DSB to wait from
> > > >> (undelayed vblank start - SCL) to (delayed vblank start - SCL) before
> > > >> sending the push. This approach works for both pre-PTL and PTL+ platforms.
> > > >>
> > > >> Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
> > > >> ---
> > > >>   drivers/gpu/drm/i915/display/intel_display.c |  2 +-
> > > >>   drivers/gpu/drm/i915/display/intel_dsb.c     | 16 ++++++++++++++++
> > > >>   drivers/gpu/drm/i915/display/intel_dsb.h     |  2 ++
> > > >>   3 files changed, 19 insertions(+), 1 deletion(-)
> > > >>
> > > >> diff --git a/drivers/gpu/drm/i915/display/intel_display.c b/drivers/gpu/drm/i915/display/intel_display.c
> > > >> index bfeec3706f35..8d78037d5a2a 100644
> > > >> --- a/drivers/gpu/drm/i915/display/intel_display.c
> > > >> +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > > >> @@ -7265,7 +7265,7 @@ static void intel_atomic_dsb_finish(struct intel_atomic_state *state,
> > > >>   				new_crtc_state->dsb_color);
> > > >>   
> > > >>   	if (new_crtc_state->use_dsb && !intel_color_uses_chained_dsb(new_crtc_state)) {
> > > >> -		intel_dsb_wait_vblanks(new_crtc_state->dsb_commit, 1);
> > > >> +		intel_dsb_wait_for_scl_start(state, new_crtc_state->dsb_commit);
> > > >>   
> > > >>   		intel_vrr_send_push(new_crtc_state->dsb_commit, new_crtc_state);
> > > >>   		intel_dsb_wait_for_scl_lines(state, new_crtc_state->dsb_commit);
> > > >> diff --git a/drivers/gpu/drm/i915/display/intel_dsb.c b/drivers/gpu/drm/i915/display/intel_dsb.c
> > > >> index 400dcc87a992..e94a05cc8c82 100644
> > > >> --- a/drivers/gpu/drm/i915/display/intel_dsb.c
> > > >> +++ b/drivers/gpu/drm/i915/display/intel_dsb.c
> > > >> @@ -826,6 +826,22 @@ void intel_dsb_wait_for_scl_lines(struct intel_atomic_state *state,
> > > >>   	intel_dsb_wait_usec(dsb, usecs);
> > > >>   }
> > > >>   
> > > >> +void intel_dsb_wait_for_scl_start(struct intel_atomic_state *state,
> > > >> +				  struct intel_dsb *dsb)
> > > >> +{
> > > >> +	struct intel_crtc *crtc = dsb->crtc;
> > > >> +	const struct intel_crtc_state *crtc_state =
> > > >> +		intel_pre_commit_crtc_state(state, crtc);
> > > >> +	int undelayed_vblank_start = crtc_state->hw.adjusted_mode.crtc_vdisplay;
> > > >> +	int delayed_vblank_start = crtc_state->hw.adjusted_mode.crtc_vblank_start;
> > > >> +	int start, end;
> > > >> +
> > > >> +	start = undelayed_vblank_start - crtc_state->set_context_latency;
> > > >> +	end = delayed_vblank_start - crtc_state->set_context_latency;
> > > > For these we perhaps want something like:
> > > >
> > > > intel_vrr_safe_window_start()
> > > > {
> > > > 	if (ptl+)
> > > > 		return crtc_vdisplay - set_context_latency;
> > > > 	else
> > > > 		return crtc_vdisplay;
> > > > }
> > > >
> > > > intel_vrr_vmin_safe_window_end()
> > > > {
> > > > 	intel_vrr_vmin_vblank_start() - set_context_latency;
> > > > }
> > > >
> > > >> +
> > > >> +	intel_dsb_wait_scanline_out(state, dsb, start, end);
> > > > And I suspect we want to do this just before the usec wait in
> > > > intel_dsb_wait_vblank_delay() (for the VRR case only). No need
> > > > to bother higher level code with this, I think.
> > > 
> > > Ok sure. I will re-arrange this and use the suggested functions for VRR 
> > > case.
> > > 
> > > On quick checking, it seems we need to use start -1 in :
> > > 
> > > intel_dsb_wait_scanline_out(state, dsb, start -1, end);
> > > 
> > > I tested with initializing with bigger value (like 5 instead of 1).
> > > 
> > >  From the dsb utility which you had shared, also we see that it gives 
> > > (vactive - scl - 1) as the safe window live status change for PTL.
> > 
> > The hardware scanline numbers are always off by one (or two) when
> > compared to our more sensible software numbers (see
> > intel_crtc_scanline_offset()). On this level we're always
> > thinking in terms of the sotfware numbers, and dsb_scanline_to_hw()
> > will then do the necessary adjustment for us.
> 
> I suppose to make our lives easier we could make intel_display_poller
> adjust the scanline numbers in a similar fashion. Though we might
> want to make that optional so that we can also have it use the
> raw hardware numbers as well. I'll see if I can cook up something...

Hmm, this might be a bit harder than I initially though because
everything would need the %vtotal to be entirely correct. So I'd
need to do a bunch of readout first to determine the vtotal
(or vmax) correctly. And if we get that wrong then we'd
potentially report garbage results.

Hmm, I suppose I could skip the %vtotal, and just not apply the
adjusment in the 'wrap' test at all. That might be the only test
that really cares about scanlines around the vtotal->0 point. And
I guess I could keep the adjustment behind a command line option
so that you sort of have to know what you're doing when enabling
it. I'll need to think about this a bit more.

-- 
Ville Syrjälä
Intel