On Tue, 2022-11-29 at 14:00 +0000, Souza, Jose wrote:
> On Tue, 2022-11-29 at 09:51 +0200, Jouni Högander wrote:
> > Currently we are observing occasionally display flickering or
> > complete
> > freeze. This is narrowed down to be caused by single full frame
> > update
> > (SFF).
> >
> > SFF bit after it's written gets cleared by HW in subsequent vblank
> > i.e. when the update is sent to the panel. SFF bit is required to
> > be
> > written together with partial frame update (PFU) bit. After the SFF
> > bit gets cleared by the HW psr2 man trk ctl register still contains
> > PFU bit. If there is subsequent update for any reason we will end
> > up
> > having selective update/fetch configuration where start line is 0
> > and
> > end line is 0.
> >
>
> How did you get this information(start and end line 0)?
If you consider what is written into psr2 man trk ctl register in case
of full frame update in intel_psr2_program_trans_man_trk_ctl and in
_psr_flush_handle:
SFF = 1
PFU = 1
Start line = 0
End line = 0
On next vblank SFF is cleared by the hw. After that we have:
PFU = 1
Start line = 0
End line = 0
which is basically selective update with start and endline as 0. If
there is an update with this configration we are observing
freeze/flicker. I can use CFF instead of SFF as a workaround. I also
checked that configuring start and end lines as a full frame is also
fixing the issue. I choose to come out with the first one.
>
> > Also selective fetch configuration for the planes is
> > not properly performed. This seems to be causing problems with some
> > panels.
> >
> > Using CFF without SFF doesn't work either because it may happen
> > that
> > psr2 man track ctl register is overwritten by next update before
> > vblank triggers sending the update. This is causing problems to
> > psr_invalidate/flush. Using CFF and SFF together solves the
> > problems
> > as SFF is cleared only by HW in subsequent vblank.
>
> This looks dangerous, have you checked with HW engineers if setting
> both could cause any issue?
Yes, this make sense, I will check this.
> At the SFF write you could get what is the current vblank counter and
> properly handle future PSR2_MAN_TRK_CTL writes.
Is vblank counter updated with that granularity? E.g. if you have
psr_invalidate/psr_flush sequence there is an update, but vblank
interrupts are not enabled? In legacy cursor update there is also an
update without vblank interrupts getting enabled. How vblank counter
works when vblank interrupt is disabled?
>
> >
> > Fix the flickering/freeze issue by adding continuous full frame
> > with
> > single full frame update and switch to partial frame update only
> > when
> > selective update area is properly calculated and configured.
> >
> > This is also workaround for HSD 14014971508
>
> Please use versions in your patches, you had 2 patches in the
> previous approach with the same subject but no versioning and no
> information about what
> changed between versions.
First one was sent to trybot and then to intel-gfx. In this one I
changed the subject. So I was thinking it's ok to leave out versioning.
I will add versioning when sending again.
>
> >
> > Cc: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
> > Cc: José Roberto de Souza <jose.souza@xxxxxxxxx>
> > Cc: Mika Kahola <mika.kahola@xxxxxxxxx>
> >
> > Reported-by: Lee Shawn C <shawn.c.lee@xxxxxxxxx>
> > Signed-off-by: Jouni Högander <jouni.hogander@xxxxxxxxx>
> > ---
> > drivers/gpu/drm/i915/display/intel_psr.c | 19 ++++++++++---------
> > 1 file changed, 10 insertions(+), 9 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/display/intel_psr.c
> > b/drivers/gpu/drm/i915/display/intel_psr.c
> > index 5b678916e6db..88388201684e 100644
> > --- a/drivers/gpu/drm/i915/display/intel_psr.c
> > +++ b/drivers/gpu/drm/i915/display/intel_psr.c
> > @@ -1510,7 +1510,8 @@ static void psr_force_hw_tracking_exit(struct
> > intel_dp *intel_dp)
> > PSR2_MAN_TRK_CTL(intel_dp-
> > >psr.transcoder),
> > man_trk_ctl_enable_bit_get(dev_priv)
> > |
> >
> > man_trk_ctl_partial_frame_bit_get(dev_priv) |
> > -
> > man_trk_ctl_single_full_frame_bit_get(dev_priv));
> > +
> > man_trk_ctl_single_full_frame_bit_get(dev_priv) |
> > +
> > man_trk_ctl_continuos_full_frame(dev_priv));
> >
> > /*
> > * Display WA #0884: skl+
> > @@ -1624,11 +1625,8 @@ static void psr2_man_trk_ctl_calc(struct
> > intel_crtc_state *crtc_state,
> > val |= man_trk_ctl_partial_frame_bit_get(dev_priv);
> >
> > if (full_update) {
> > - /*
> > - * Not applying Wa_14014971508:adlp as we do not
> > support the
> > - * feature that requires this workaround.
> > - */
> > val |=
> > man_trk_ctl_single_full_frame_bit_get(dev_priv);
> > + val |= man_trk_ctl_continuos_full_frame(dev_priv);
> > goto exit;
> > }
> >
> > @@ -2307,12 +2305,15 @@ static void _psr_flush_handle(struct
> > intel_dp *intel_dp)
> > /* can we turn CFF off? */
> > if (intel_dp->psr.busy_frontbuffer_bits ==
> > 0) {
> > u32 val =
> > man_trk_ctl_enable_bit_get(dev_priv) |
> > -
> > man_trk_ctl_partial_frame_bit_get(dev_priv) |
> > -
> > man_trk_ctl_single_full_frame_bit_get(dev_priv);
> > + man_trk_ctl_partial_frame_b
> > it_get(dev_priv) |
> > + man_trk_ctl_single_full_fra
> > me_bit_get(dev_priv) |
> > + man_trk_ctl_continuos_full_
> > frame(dev_priv);
>
> style.
>
> >
> > /*
> > - * turn continuous full frame off
> > and do a single
> > - * full frame
> > + * turn continuous full frame off
> > and do a single full frame. Still
> > + * keep cff bit enabled as we don't
> > have proper SU configuration in
> > + * case update is sent for any
> > reason after sff bit gets cleared by
> > + * the HW on next vblank.
>
>
> turn off and keep bit enabled?! makes no sense this comment.
>
> > */
> > intel_de_write(dev_priv,
> > PSR2_MAN_TRK_CTL(intel_dp->psr.transcoder),
> > val);
>
