On 9/22/2025 7:34 PM, Ville Syrjälä wrote:
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 <[email protected]>
---
   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.

I will look into this utility and try other things out too.

This does look impressive for debug and understanding whats going on specifically with DSB.


As for the patch, I will use the suggested helpers for start and end lines and move this "wait" just before the vblank_delay usecs for VRR case.


Regards,

Ankit



Reply via email to