[Intel-gfx] [PATCH 5/6] drm/i915: Adjust seamless_m_n flag behaviour
Ville Syrjälä
ville.syrjala at linux.intel.com
Wed Aug 30 06:58:51 UTC 2023
On Wed, Aug 30, 2023 at 06:51:25AM +0000, Golani, Mitulkumar Ajitkumar wrote:
> Hi Ville,
>
> > -----Original Message-----
> > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > Sent: 30 August 2023 10:47
> > To: Golani, Mitulkumar Ajitkumar
> > <mitulkumar.ajitkumar.golani at intel.com>
> > Cc: intel-gfx at lists.freedesktop.org
> > Subject: Re: [Intel-gfx] [PATCH 5/6] drm/i915: Adjust seamless_m_n flag
> > behaviour
> >
> > On Tue, Aug 29, 2023 at 01:48:18PM +0000, Golani, Mitulkumar Ajitkumar
> > wrote:
> > > Hi Ville,
> > >
> > > Thanks for the inputs.
> > >
> > > I encountered an interesting observation while validating the changes.
> > > In scenarios where VRR is by default ON from the panel, I noticed that
> > during the first-time enabling of VRR, a full modeset is required due to a
> > fastset requirement mismatch, as indicated in the logs.
> >
> > Which logs?
> >
> > > However, I also observed that after this, the function
> > intel_crtc_needs_modeset returns 0, which typically implies a fastset
> > operation.
> >
> > Any fastset always starts out as a full modeset, it will then be promoted (or
> > demoted?) to a fastset by intel_crtc_check_fastset() if possible.
> >
>
> Thank you for your explanation. Considering the underrun issues observed during validation,
> it seems plausible that for the initial computation of VRR parameters, a full modeset might be required
> to ensure the accurate establishment of the baseline configuration. I appreciate your insight on this matter.
Are you saying that these patches alone are causing
underruns when enabling VRR?
>
> Regards,
> Mitul
>
> > >
> > > Considering that both conditions can't hold simultaneously, there seems
> > to be an inconsistency with underrun errors.
> > > Could you kindly shed some light on this situation? I'm trying to align the
> > log messages with the observed behavior and the function's return value.
> > >
> > > Your insights would be greatly appreciated.
> > >
> > > Regards,
> > > Mitul
> > >
> > > > -----Original Message-----
> > > > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > Sent: 29 August 2023 13:51
> > > > To: Golani, Mitulkumar Ajitkumar
> > > > <mitulkumar.ajitkumar.golani at intel.com>
> > > > Cc: intel-gfx at lists.freedesktop.org
> > > > Subject: Re: [Intel-gfx] [PATCH 5/6] drm/i915: Adjust seamless_m_n
> > > > flag behaviour
> > > >
> > > > On Mon, Aug 28, 2023 at 04:58:49PM +0000, Golani, Mitulkumar
> > > > Ajitkumar
> > > > wrote:
> > > > > Hi Ville,
> > > > >
> > > > > > -----Original Message-----
> > > > > > From: Intel-gfx <intel-gfx-bounces at lists.freedesktop.org> On
> > > > > > Behalf Of Ville Syrjala
> > > > > > Sent: 28 August 2023 11:12
> > > > > > To: intel-gfx at lists.freedesktop.org
> > > > > > Subject: [Intel-gfx] [PATCH 5/6] drm/i915: Adjust seamless_m_n
> > > > > > flag behaviour
> > > > > >
> > > > > > From: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > > >
> > > > > > Make the seamless_m_n flag more like the update_pipe fastset
> > > > > > flag, ie. the flag will only be set if we need to do the
> > > > > > seamless M/N update, and in all other cases the flag is cleared.
> > > > > > Also rename the flag to update_m_n to make it more clear it's
> > similar to update_pipe.
> > > > > >
> > > > > > I believe special casing seamless_m_n like this makes sense as
> > > > > > it also affects eg. vblank evasion. We can potentially avoid
> > > > > > some vblank evasion tricks, simplify some checks, and hopefully
> > > > > > will help with
> > > > the VRR vs. M/N mess.
> > > > > >
> > > > > > Signed-off-by: Ville Syrjälä <ville.syrjala at linux.intel.com>
> > > > > > ---
> > > > > > drivers/gpu/drm/i915/display/intel_atomic.c | 1 +
> > > > > > drivers/gpu/drm/i915/display/intel_crtc.c | 2 +-
> > > > > > drivers/gpu/drm/i915/display/intel_display.c | 22 +++++++++++------
> > --
> > > > > > .../drm/i915/display/intel_display_types.h | 2 +-
> > > > > > drivers/gpu/drm/i915/display/intel_dp.c | 2 +-
> > > > > > 5 files changed, 17 insertions(+), 12 deletions(-)
> > > > > >
> > > > > > diff --git a/drivers/gpu/drm/i915/display/intel_atomic.c
> > > > > > b/drivers/gpu/drm/i915/display/intel_atomic.c
> > > > > > index 7cf51dd8c056..aaddd8c0cfa0 100644
> > > > > > --- a/drivers/gpu/drm/i915/display/intel_atomic.c
> > > > > > +++ b/drivers/gpu/drm/i915/display/intel_atomic.c
> > > > > > @@ -259,6 +259,7 @@ intel_crtc_duplicate_state(struct drm_crtc
> > *crtc)
> > > > > > drm_property_blob_get(crtc_state->post_csc_lut);
> > > > > >
> > > > > > crtc_state->update_pipe = false;
> > > > > > + crtc_state->update_m_n = false;
> > > > > > crtc_state->disable_lp_wm = false;
> > > > > > crtc_state->disable_cxsr = false;
> > > > > > crtc_state->update_wm_pre = false; diff --git
> > > > > > a/drivers/gpu/drm/i915/display/intel_crtc.c
> > > > > > b/drivers/gpu/drm/i915/display/intel_crtc.c
> > > > > > index 1992e7060263..a04076064f02 100644
> > > > > > --- a/drivers/gpu/drm/i915/display/intel_crtc.c
> > > > > > +++ b/drivers/gpu/drm/i915/display/intel_crtc.c
> > > > > > @@ -510,7 +510,7 @@ static void
> > > > > > intel_crtc_vblank_evade_scanlines(struct
> > > > > > intel_atomic_state *state,
> > > > > > * M/N is double buffered on the transcoder's undelayed
> > vblank,
> > > > > > * so with seamless M/N we must evade both vblanks.
> > > > > > */
> > > > > > - if (new_crtc_state->seamless_m_n &&
> > > > > > intel_crtc_needs_fastset(new_crtc_state))
> > > > > > + if (new_crtc_state->update_m_n)
> > > > > > *min -= adjusted_mode->crtc_vblank_start -
> > > > > > adjusted_mode->crtc_vdisplay; }
> > > > > >
> > > > > > diff --git a/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > b/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > index 632f1f58df9e..6196ef76390b 100644
> > > > > > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > > > > > @@ -5170,7 +5170,7 @@ intel_pipe_config_compare(const struct
> > > > > > intel_crtc_state *current_config,
> > > > > > PIPE_CONF_CHECK_X(lane_lat_optim_mask);
> > > > > >
> > > > > > if (HAS_DOUBLE_BUFFERED_M_N(dev_priv)) {
> > > > > > - if (!fastset || !pipe_config->seamless_m_n)
> > > > > > + if (!fastset || !pipe_config->update_m_n)
> > > > > > PIPE_CONF_CHECK_M_N(dp_m_n);
> > > > > > } else {
> > > > > > PIPE_CONF_CHECK_M_N(dp_m_n);
> > > > > > @@ -5307,7 +5307,7 @@ intel_pipe_config_compare(const struct
> > > > > > intel_crtc_state *current_config,
> > > > > > if (IS_G4X(dev_priv) || DISPLAY_VER(dev_priv) >= 5)
> > > > > > PIPE_CONF_CHECK_I(pipe_bpp);
> > > > > >
> > > > > > - if (!fastset || !pipe_config->seamless_m_n) {
> > > > > > + if (!fastset || !pipe_config->update_m_n) {
> > > > > > PIPE_CONF_CHECK_I(hw.pipe_mode.crtc_clock);
> > > > > > PIPE_CONF_CHECK_I(hw.adjusted_mode.crtc_clock);
> > > > > > }
> > > > > > @@ -5402,6 +5402,7 @@ int intel_modeset_all_pipes(struct
> > > > > > intel_atomic_state *state,
> > > > > >
> > > > > > crtc_state->uapi.mode_changed = true;
> > > > > > crtc_state->update_pipe = false;
> > > > > > + crtc_state->update_m_n = false;
> > > > > >
> > > > > > ret = drm_atomic_add_affected_connectors(&state-
> > >base,
> > > > > > &crtc->base);
> > > > > > @@ -5519,13 +5520,14 @@ static void
> > > > > > intel_crtc_check_fastset(const struct intel_crtc_state *old_crtc_sta {
> > > > > > struct drm_i915_private *i915 =
> > > > > > to_i915(old_crtc_state->uapi.crtc-
> > > > > > >dev);
> > > > > >
> > > > > > - if (!intel_pipe_config_compare(old_crtc_state,
> > new_crtc_state,
> > > > > > true)) {
> > > > > > + if (!intel_pipe_config_compare(old_crtc_state,
> > new_crtc_state,
> > > > > > true))
> > > > > > drm_dbg_kms(&i915->drm, "fastset requirement not
> > met,
> > > > forcing
> > > > > > full modeset\n");
> > > > > > + else
> > > > > > + new_crtc_state->uapi.mode_changed = false;
> > > > > >
> > > > > > - return;
> > > > > > - }
> > > > > > + if (intel_crtc_needs_modeset(new_crtc_state))
> > > > > > + new_crtc_state->update_m_n = false;
> > > > > >
> > > > > > - new_crtc_state->uapi.mode_changed = false;
> > > > > > if (!intel_crtc_needs_modeset(new_crtc_state))
> > > > > > new_crtc_state->update_pipe = true; } @@ -6240,6
> > +6242,7
> > > > @@ int
> > > > > > intel_atomic_check(struct drm_device *dev,
> > > > > > if
> > (intel_cpu_transcoders_need_modeset(state,
> > > > > > BIT(master))) {
> > > > > > new_crtc_state-
> > >uapi.mode_changed = true;
> > > > > > new_crtc_state->update_pipe =
> > false;
> > > > > > + new_crtc_state->update_m_n =
> > false;
> > > > > > }
> > > > > > }
> > > > > >
> > > > > > @@ -6252,6 +6255,7 @@ int intel_atomic_check(struct drm_device
> > > > *dev,
> > > > > > if
> > (intel_cpu_transcoders_need_modeset(state,
> > > > > > trans)) {
> > > > > > new_crtc_state-
> > >uapi.mode_changed = true;
> > > > > > new_crtc_state->update_pipe =
> > false;
> > > > > > + new_crtc_state->update_m_n =
> > false;
> > > > > > }
> > > > > > }
> > > > > >
> > > > > > @@ -6259,6 +6263,7 @@ int intel_atomic_check(struct drm_device
> > > > *dev,
> > > > > > if (intel_pipes_need_modeset(state,
> > new_crtc_state-
> > > > > > >bigjoiner_pipes)) {
> > > > > > new_crtc_state-
> > >uapi.mode_changed = true;
> > > > > > new_crtc_state->update_pipe =
> > false;
> > > > > > + new_crtc_state->update_m_n =
> > false;
> > > > > > }
> > > > > > }
> > > > > > }
> > > > > > @@ -6437,7 +6442,7 @@ static void intel_pipe_fastset(const
> > > > > > struct intel_crtc_state *old_crtc_state,
> > > > > > IS_BROADWELL(dev_priv) || IS_HASWELL(dev_priv))
> > > > > > hsw_set_linetime_wm(new_crtc_state);
> > > > > >
> > > > > > - if (new_crtc_state->seamless_m_n)
> > > > > > + if (new_crtc_state->update_m_n)
> > > > > > intel_cpu_transcoder_set_m1_n1(crtc,
> > new_crtc_state-
> > > > > > >cpu_transcoder,
> > > > > > &new_crtc_state-
> > >dp_m_n); }
> > > > @@ -6573,8 +6578,7 @@
> > > > > > static void intel_update_crtc(struct intel_atomic_state *state,
> > > > > > *
> > > > > > * FIXME Should be synchronized with the start of vblank
> > > > > > somehow...
> > > > > > */
> > > > > > - if (vrr_enabling(old_crtc_state, new_crtc_state) ||
> > > > > > - (new_crtc_state->seamless_m_n &&
> > > > > > intel_crtc_needs_fastset(new_crtc_state)))
> > > > > > + if (vrr_enabling(old_crtc_state, new_crtc_state) ||
> > > > > > +new_crtc_state->update_m_n)
> > > > > > intel_crtc_update_active_timings(new_crtc_state,
> > > > > > new_crtc_state-
> > >vrr.enable);
> > > > >
> > > > > In the context where a Push has already been sent, should the
> > > > > update
> > > > occur when we enable VRR itself? I'm curious about the rationale for
> > > > not updating it immediately upon enabling VRR.
> > > >
> > > > The active timings should really be updated synchronously with the
> > > > hardware latching the new values. But that is actually impossible so
> > > > some race conditions will always remain.
> > > >
> > > > I haven't actually verified how the hardware behaves when we enable
> > VRR.
> > > > I assume the VRR_CTL will get latched at the next start of vblank as well.
> > > > But we should really try to confirm that on actual hardware.
> > > >
> > > > The case where we have VRR already enabled while updating M/N might
> > > > go badly with the current order of doing things as the vblank irq
> > > > may trigger immediately upon push. So for that case it would be
> > > > better to update the active timings before the push. Or perhaps we
> > > > should just make sure VRR is always disabled around M/N updates...
> > > >
> > > > >
> > > > > Regards,
> > > > > Mitul
> > > > > > b/drivers/gpu/drm/i915/display/intel_display_types.h
> > > > > > index 731f2ec04d5c..2367b030b469 100644
> > > > > > --- a/drivers/gpu/drm/i915/display/intel_display_types.h
> > > > > > +++ b/drivers/gpu/drm/i915/display/intel_display_types.h
> > > > > > @@ -1083,6 +1083,7 @@ struct intel_crtc_state {
> > > > > >
> > > > > > unsigned fb_bits; /* framebuffers to flip */
> > > > > > bool update_pipe; /* can a fast modeset be performed? */
> > > > > > + bool update_m_n; /* update M/N seamlessly during
> > fastset? */
> > > > > > bool disable_cxsr;
> > > > > > bool update_wm_pre, update_wm_post; /* watermarks are
> > > > updated */
> > > > > > bool fifo_changed; /* FIFO split is changed */ @@ -1195,7
> > > > > > +1196,6 @@ struct intel_crtc_state {
> > > > > > /* m2_n2 for eDP downclock */
> > > > > > struct intel_link_m_n dp_m2_n2;
> > > > > > bool has_drrs;
> > > > > > - bool seamless_m_n;
> > > > > >
> > > > > > /* PSR is supported but might not be enabled due the lack
> > of
> > > > > > enabled planes */
> > > > > > bool has_psr;
> > > > > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > > b/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > > index 05694e0b6143..ceab5b26b5e8 100644
> > > > > > --- a/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > > +++ b/drivers/gpu/drm/i915/display/intel_dp.c
> > > > > > @@ -2535,7 +2535,7 @@ intel_dp_drrs_compute_config(struct
> > > > > > intel_connector *connector,
> > > > > > int pixel_clock;
> > > > > >
> > > > > > if (has_seamless_m_n(connector))
> > > > > > - pipe_config->seamless_m_n = true;
> > > > > > + pipe_config->update_m_n = true;
> > > > > >
> > > > > > if (!can_enable_drrs(connector, pipe_config,
> > downclock_mode)) {
> > > > > > if (intel_cpu_transcoder_has_m2_n2(i915,
> > pipe_config-
> > > > > > >cpu_transcoder))
> > > > > > --
> > > > > > 2.41.0
> > > > >
> > > >
> > > > --
> > > > Ville Syrjälä
> > > > Intel
> >
> > --
> > Ville Syrjälä
> > Intel
--
Ville Syrjälä
Intel
More information about the Intel-gfx
mailing list