| Message ID | 20171115163157.14372-1-maarten.lankhorst@linux.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On Wed, Nov 15, 2017 at 05:31:56PM +0100, Maarten Lankhorst wrote: > The watermarks it should calculate against are the old optimal watermarks. > The currently active crtc watermarks are pure fiction, and are invalid in > case of a nonblocking modeset, page flip enabling/disabling planes or any > other reason. > > When the crtc is disabled or during a modeset the intermediate watermarks > don't need to be programmed separately, and could be directly assigned > to the optimal watermarks. > > CXSR must always be disabled in the intermediate case for modesets, else > we get a WARN for vblank wait timeout. > > Also rename crtc_state to new_crtc_state, to distinguish it from the old state. > > Changes since v1: > - Use intel_atomic_get_old_crtc_state. (ville) > Changes since v2: > - Always unset cxsr during modeset. > > Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> I was going to try and figure out how/if these get rid of the unclaimed reg warns, but I didn't quite get that far. I did spot a few other buglets in the wm code though (I'll send fixes for those at some point). Anyways, these patches make sense to me, so for the series Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> > --- > drivers/gpu/drm/i915/intel_pm.c | 24 ++++++++++++++++++------ > 1 file changed, 18 insertions(+), 6 deletions(-) > > diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c > index 8c69ec9eb6ee..f904bf73dbd6 100644 > --- a/drivers/gpu/drm/i915/intel_pm.c > +++ b/drivers/gpu/drm/i915/intel_pm.c > @@ -2026,16 +2026,27 @@ static void vlv_atomic_update_fifo(struct intel_atomic_state *state, > > static int vlv_compute_intermediate_wm(struct drm_device *dev, > struct intel_crtc *crtc, > - struct intel_crtc_state *crtc_state) > + struct intel_crtc_state *new_crtc_state) > { > - struct vlv_wm_state *intermediate = &crtc_state->wm.vlv.intermediate; > - const struct vlv_wm_state *optimal = &crtc_state->wm.vlv.optimal; > - const struct vlv_wm_state *active = &crtc->wm.active.vlv; > + struct vlv_wm_state *intermediate = &new_crtc_state->wm.vlv.intermediate; > + const struct vlv_wm_state *optimal = &new_crtc_state->wm.vlv.optimal; > + struct intel_atomic_state *intel_state = > + to_intel_atomic_state(new_crtc_state->base.state); > + const struct intel_crtc_state *old_crtc_state = > + intel_atomic_get_old_crtc_state(intel_state, crtc); > + const struct vlv_wm_state *active = &old_crtc_state->wm.vlv.optimal; > int level; > > + if (!new_crtc_state->base.active || drm_atomic_crtc_needs_modeset(&new_crtc_state->base)) { > + *intermediate = *optimal; > + > + intermediate->cxsr = false; > + goto out; > + } > + > intermediate->num_levels = min(optimal->num_levels, active->num_levels); > intermediate->cxsr = optimal->cxsr && active->cxsr && > - !crtc_state->disable_cxsr; > + !new_crtc_state->disable_cxsr; > > for (level = 0; level < intermediate->num_levels; level++) { > enum plane_id plane_id; > @@ -2054,12 +2065,13 @@ static int vlv_compute_intermediate_wm(struct drm_device *dev, > > vlv_invalidate_wms(crtc, intermediate, level); > > +out: > /* > * If our intermediate WM are identical to the final WM, then we can > * omit the post-vblank programming; only update if it's different. > */ > if (memcmp(intermediate, optimal, sizeof(*intermediate)) != 0) > - crtc_state->wm.need_postvbl_update = true; > + new_crtc_state->wm.need_postvbl_update = true; > > return 0; > } > -- > 2.15.0 > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
Op 17-11-17 om 14:31 schreef Ville Syrjälä: > On Wed, Nov 15, 2017 at 05:31:56PM +0100, Maarten Lankhorst wrote: >> The watermarks it should calculate against are the old optimal watermarks. >> The currently active crtc watermarks are pure fiction, and are invalid in >> case of a nonblocking modeset, page flip enabling/disabling planes or any >> other reason. >> >> When the crtc is disabled or during a modeset the intermediate watermarks >> don't need to be programmed separately, and could be directly assigned >> to the optimal watermarks. >> >> CXSR must always be disabled in the intermediate case for modesets, else >> we get a WARN for vblank wait timeout. >> >> Also rename crtc_state to new_crtc_state, to distinguish it from the old state. >> >> Changes since v1: >> - Use intel_atomic_get_old_crtc_state. (ville) >> Changes since v2: >> - Always unset cxsr during modeset. >> >> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> > I was going to try and figure out how/if these get rid of the unclaimed > reg warns, but I didn't quite get that far. I did spot a few other > buglets in the wm code though (I'll send fixes for those at some point). > > Anyways, these patches make sense to me, so for the series > Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Seems Chris Wilson already beat us to it.. https://intel-gfx-ci.01.org/tree/drm-tip/igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b.html was gone with commit 1a1f12872edcd5e425b668a35fb23548cfa918ef Author: Chris Wilson <chris@chris-wilson.co.uk <mailto:chris@chris-wilson.co.uk>> Date: Tue Nov 7 14:03:38 2017 +0000 drm/i915: Prevent unbounded wm results in g4x_compute_wm() But I think this patch also fixes it in a slightly different way, never using garbage for intermediaries. :-) Thanks for review, pushed. Cheers, ~Maarten
On Fri, Nov 17, 2017 at 03:47:58PM +0100, Maarten Lankhorst wrote: > Op 17-11-17 om 14:31 schreef Ville Syrjälä: > > On Wed, Nov 15, 2017 at 05:31:56PM +0100, Maarten Lankhorst wrote: > >> The watermarks it should calculate against are the old optimal watermarks. > >> The currently active crtc watermarks are pure fiction, and are invalid in > >> case of a nonblocking modeset, page flip enabling/disabling planes or any > >> other reason. > >> > >> When the crtc is disabled or during a modeset the intermediate watermarks > >> don't need to be programmed separately, and could be directly assigned > >> to the optimal watermarks. > >> > >> CXSR must always be disabled in the intermediate case for modesets, else > >> we get a WARN for vblank wait timeout. > >> > >> Also rename crtc_state to new_crtc_state, to distinguish it from the old state. > >> > >> Changes since v1: > >> - Use intel_atomic_get_old_crtc_state. (ville) > >> Changes since v2: > >> - Always unset cxsr during modeset. > >> > >> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> > > I was going to try and figure out how/if these get rid of the unclaimed > > reg warns, but I didn't quite get that far. I did spot a few other > > buglets in the wm code though (I'll send fixes for those at some point). > > > > Anyways, these patches make sense to me, so for the series > > Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> > Seems Chris Wilson already beat us to it.. > > https://intel-gfx-ci.01.org/tree/drm-tip/igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b.html > was gone with > > commit 1a1f12872edcd5e425b668a35fb23548cfa918ef > Author: Chris Wilson <chris@chris-wilson.co.uk <mailto:chris@chris-wilson.co.uk>> > Date: Tue Nov 7 14:03:38 2017 +0000 > > drm/i915: Prevent unbounded wm results in g4x_compute_wm() That patch should be nop. So this is a very surprising result. > > But I think this patch also fixes it in a slightly different way, never using garbage for intermediaries. :-) > > Thanks for review, pushed. > > Cheers, > ~Maarten
Op 17-11-17 om 15:53 schreef Ville Syrjälä: > On Fri, Nov 17, 2017 at 03:47:58PM +0100, Maarten Lankhorst wrote: >> Op 17-11-17 om 14:31 schreef Ville Syrjälä: >>> On Wed, Nov 15, 2017 at 05:31:56PM +0100, Maarten Lankhorst wrote: >>>> The watermarks it should calculate against are the old optimal watermarks. >>>> The currently active crtc watermarks are pure fiction, and are invalid in >>>> case of a nonblocking modeset, page flip enabling/disabling planes or any >>>> other reason. >>>> >>>> When the crtc is disabled or during a modeset the intermediate watermarks >>>> don't need to be programmed separately, and could be directly assigned >>>> to the optimal watermarks. >>>> >>>> CXSR must always be disabled in the intermediate case for modesets, else >>>> we get a WARN for vblank wait timeout. >>>> >>>> Also rename crtc_state to new_crtc_state, to distinguish it from the old state. >>>> >>>> Changes since v1: >>>> - Use intel_atomic_get_old_crtc_state. (ville) >>>> Changes since v2: >>>> - Always unset cxsr during modeset. >>>> >>>> Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> >>> I was going to try and figure out how/if these get rid of the unclaimed >>> reg warns, but I didn't quite get that far. I did spot a few other >>> buglets in the wm code though (I'll send fixes for those at some point). >>> >>> Anyways, these patches make sense to me, so for the series >>> Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> >> Seems Chris Wilson already beat us to it.. >> >> https://intel-gfx-ci.01.org/tree/drm-tip/igt@kms_pipe_crc_basic@suspend-read-crc-pipe-b.html >> was gone with >> >> commit 1a1f12872edcd5e425b668a35fb23548cfa918ef >> Author: Chris Wilson <chris@chris-wilson.co.uk <mailto:chris@chris-wilson.co.uk>> >> Date: Tue Nov 7 14:03:38 2017 +0000 >> >> drm/i915: Prevent unbounded wm results in g4x_compute_wm() > That patch should be nop. So this is a very surprising result. > Oh well maybe the watermarks initially read out were being garbage. Garbage in, garbage out. :-)
diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c index 8c69ec9eb6ee..f904bf73dbd6 100644 --- a/drivers/gpu/drm/i915/intel_pm.c +++ b/drivers/gpu/drm/i915/intel_pm.c @@ -2026,16 +2026,27 @@ static void vlv_atomic_update_fifo(struct intel_atomic_state *state, static int vlv_compute_intermediate_wm(struct drm_device *dev, struct intel_crtc *crtc, - struct intel_crtc_state *crtc_state) + struct intel_crtc_state *new_crtc_state) { - struct vlv_wm_state *intermediate = &crtc_state->wm.vlv.intermediate; - const struct vlv_wm_state *optimal = &crtc_state->wm.vlv.optimal; - const struct vlv_wm_state *active = &crtc->wm.active.vlv; + struct vlv_wm_state *intermediate = &new_crtc_state->wm.vlv.intermediate; + const struct vlv_wm_state *optimal = &new_crtc_state->wm.vlv.optimal; + struct intel_atomic_state *intel_state = + to_intel_atomic_state(new_crtc_state->base.state); + const struct intel_crtc_state *old_crtc_state = + intel_atomic_get_old_crtc_state(intel_state, crtc); + const struct vlv_wm_state *active = &old_crtc_state->wm.vlv.optimal; int level; + if (!new_crtc_state->base.active || drm_atomic_crtc_needs_modeset(&new_crtc_state->base)) { + *intermediate = *optimal; + + intermediate->cxsr = false; + goto out; + } + intermediate->num_levels = min(optimal->num_levels, active->num_levels); intermediate->cxsr = optimal->cxsr && active->cxsr && - !crtc_state->disable_cxsr; + !new_crtc_state->disable_cxsr; for (level = 0; level < intermediate->num_levels; level++) { enum plane_id plane_id; @@ -2054,12 +2065,13 @@ static int vlv_compute_intermediate_wm(struct drm_device *dev, vlv_invalidate_wms(crtc, intermediate, level); +out: /* * If our intermediate WM are identical to the final WM, then we can * omit the post-vblank programming; only update if it's different. */ if (memcmp(intermediate, optimal, sizeof(*intermediate)) != 0) - crtc_state->wm.need_postvbl_update = true; + new_crtc_state->wm.need_postvbl_update = true; return 0; }
The watermarks it should calculate against are the old optimal watermarks. The currently active crtc watermarks are pure fiction, and are invalid in case of a nonblocking modeset, page flip enabling/disabling planes or any other reason. When the crtc is disabled or during a modeset the intermediate watermarks don't need to be programmed separately, and could be directly assigned to the optimal watermarks. CXSR must always be disabled in the intermediate case for modesets, else we get a WARN for vblank wait timeout. Also rename crtc_state to new_crtc_state, to distinguish it from the old state. Changes since v1: - Use intel_atomic_get_old_crtc_state. (ville) Changes since v2: - Always unset cxsr during modeset. Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com> --- drivers/gpu/drm/i915/intel_pm.c | 24 ++++++++++++++++++------ 1 file changed, 18 insertions(+), 6 deletions(-)