Re: [Intel-gfx] [PATCH v18 4/8] drm/i915: Introduce more *_state_changed indicators

["Lisovskiy, Stanislav" <stanislav.lisovskiy@intel.com>]

[-- Attachment #1.1: Type: text/plain, Size: 15875 bytes --]

>> But this patch is not about that - it is about how we signal/determine
>> that some change has to be written at commit stage.
>>As you remember when we were discussed offline, I just wanted to have
>> some expicit way to mark if some global state subsystem had changed,
>>without having to do any additional checks, because imho all the checks
>> we should do during atomic check, while commit simply applies what has
>> to be applied.
>>
>>If you are really against having those boolean or any other way to be
>>able so simply mark some stage object "dirty" (just like mem pages
>>analogy) then would vote at least to have some helper functions to do
>>that.
>>i.e smth like:
>>
>>bool pipe_sagv_mask_changed(..)


>This is just a !=, no? Don't see a function really making it any more clear.


it is more compact at least(imho) and also it makes it more clear why

we do this "!=".

>>
>> bool ddb_state_changed(...)

>So far I don't see any real need to check for that.


The idea behind this is that we need to recompute SAGV only

when either active pipes had changed or wm/ddb allocations had changed(lets

say we now use a different mode or less planes and so on).


Currently we have _no_ flag that indicates if ddb/wm had changed and

recomputing SAGV everytime "just in case" looks redundant.


To me it looks easier rather than having comparisons with implicit

meaning.

Would be even nicer to unify that idea in a sense that any global object

like bw_state, dbuf_state can be checked if it's state had changed and to

have it as some helper functions for that in intel_global_state.c or something like that.



If you can propose a better way, please do: I think you got the idea,

what I mean.

________________________________
From: Ville Syrjälä <ville.syrjala@linux.intel.com>
Sent: Friday, February 28, 2020 6:12:36 PM
To: Lisovskiy, Stanislav
Cc: intel-gfx@lists.freedesktop.org; Roper, Matthew D
Subject: Re: [PATCH v18 4/8] drm/i915: Introduce more *_state_changed indicators

On Fri, Feb 28, 2020 at 08:56:58AM +0000, Lisovskiy, Stanislav wrote:
> On Thu, 2020-02-27 at 18:12 +0200, Ville Syrjälä wrote:
> > On Tue, Feb 25, 2020 at 04:57:33PM +0200, Stanislav Lisovskiy wrote:
> > > The reasoning behind this is such that current dependencies
> > > in the code are rather implicit in a sense, we have to constantly
> > > check a bunch of different bits like state->modeset,
> > > state->active_pipe_changes, which sometimes can indicate counter
> > > intuitive changes.
> > >
> > > By introducing more fine grained state change tracking we achieve
> > > better readability and dependency maintenance for the code.
> > >
> > > For example it is no longer needed to evaluate active_pipe_changes
> > > to understand if there were changes for wm/ddb - lets just have
> > > a correspondent bit in a state, called ddb_state_changed.
> > >
> > > active_pipe_changes just indicate whether there was some pipe added
> > > or removed. Then we evaluate if wm/ddb had been changed.
> > > Same for sagv/bw state. ddb changes may or may not affect if out
> > > bandwidth constraints have been changed.
> > >
> > > v2: Add support for older Gens in order not to introduce
> > > regressions
> > >
> > > Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
> > > ---
> > >  drivers/gpu/drm/i915/display/intel_atomic.c   |  2 ++
> > >  drivers/gpu/drm/i915/display/intel_bw.c       | 28 ++++++++++++++-
> > > -
> > >  drivers/gpu/drm/i915/display/intel_display.c  | 16 ++++++----
> > >  .../drm/i915/display/intel_display_types.h    | 32 ++++++++++++---
> > > ----
> > >  drivers/gpu/drm/i915/intel_pm.c               |  5 ++-
> > >  5 files changed, 62 insertions(+), 21 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/display/intel_atomic.c
> > > b/drivers/gpu/drm/i915/display/intel_atomic.c
> > > index d043057d2fa0..0db9c66d3c0f 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_atomic.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_atomic.c
> > > @@ -525,6 +525,8 @@ void intel_atomic_state_clear(struct
> > > drm_atomic_state *s)
> > >    state->dpll_set = state->modeset = false;
> > >    state->global_state_changed = false;
> > >    state->active_pipes = 0;
> > > + state->ddb_state_changed = false;
> > > + state->bw_state_changed = false;
> >
> > Not really liking these.
> >
> > After some pondering I was thinking along the lines of something
> > simple
> > like this:
> >
> > struct bw_state {
> >      u8 sagv_reject;
> > };
> >
> > bw_check()
> > {
> >      for_each_crtc_in_state() {
> >              if (sagv_possible(crtc_state))
> >                      new->sagv_reject &= ~BIT(pipe);
> >              else
> >                      new->sagv_reject |= BIT(pipe);
> >      }
> >
> >      calculate new->qgv_mask
> > }
>
> This is exactly what's done in the next patch, except
> that I store pipe, which are allowed to have SAGV, i.e:

I think inverted mask idea leads to neater code because then we
don't have to care which pipes are actually present in the hw
and which are fused off/not present:

sagv_reject == 0 -> SAGV possible
sagv_reject != 0 -> SAGV not possible

>
>  struct intel_bw_state {
>        struct intel_global_state base;
>
> +     /*
> +      * Contains a bit mask, used to determine, whether
> correspondent
> +      * pipe allows SAGV or not.
> +      */
> +     u8 pipe_sagv_mask;
> +
> +     /*
> +      * Used to determine if we already had calculated
> +      * SAGV mask for this state once.
> +      */
> +     bool sagv_calculated;
> +
> +     /*
> +      * Contains final SAGV decision based on current mask,
> +      * to prevent doing the same job over and over again.
> +      */
> +     bool can_sagv;
> +
>
> Also the mask is calculated almost exactly same way:
>
> static void icl_compute_sagv_mask(struct intel_atomic_state *state)
> +{
> +     struct intel_crtc *crtc;
> +     struct intel_crtc_state *new_crtc_state;
> +     int i;
> +     struct intel_bw_state *new_bw_state =
> intel_bw_get_state(state);
> +
> +     if (IS_ERR(new_bw_state)) {
> +             WARN(1, "Could not get bw_state\n");
> +             return;
> +     }
> +
> +     for_each_new_intel_crtc_in_state(state, crtc,
> +                                      new_crtc_state, i) {
> +             if (skl_can_enable_sagv_on_pipe(state, crtc->pipe))
> +                     new_bw_state->pipe_sagv_mask |= BIT(crtc-
> >pipe);
> +             else
> +                     new_bw_state->pipe_sagv_mask &= ~BIT(crtc-
> >pipe);
> +     }
> +}
>
> But this patch is not about that - it is about how we signal/determine
> that some change has to be written at commit stage.
> As you remember when we were discussed offline, I just wanted to have
> some expicit way to mark if some global state subsystem had changed,
> without having to do any additional checks, because imho all the checks
> we should do during atomic check, while commit simply applies what has
> to be applied.
>
> If you are really against having those boolean or any other way to be
> able so simply mark some stage object "dirty" (just like mem pages
> analogy) then would vote at least to have some helper functions to do
> that.
> i.e smth like:
>
> bool pipe_sagv_mask_changed(..)

This is just a !=, no? Don't see a function really making it any more clear.

>
> bool ddb_state_changed(...)

So far I don't see any real need to check for that.

>
> Stan
>
> >
> > >  }
> > >
> > >  struct intel_crtc_state *
> > > diff --git a/drivers/gpu/drm/i915/display/intel_bw.c
> > > b/drivers/gpu/drm/i915/display/intel_bw.c
> > > index bdad7476dc7b..d5be603b8b03 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_bw.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_bw.c
> > > @@ -424,9 +424,27 @@ int intel_bw_atomic_check(struct
> > > intel_atomic_state *state)
> > >    struct intel_crtc *crtc;
> > >    int i, ret;
> > >
> > > - /* FIXME earlier gens need some checks too */
> > > - if (INTEL_GEN(dev_priv) < 11)
> > > + /*
> > > +  * For earlier Gens let's consider bandwidth changed if ddb
> > > requirements,
> > > +  * has been changed.
> > > +  */
> > > + if (INTEL_GEN(dev_priv) < 11) {
> > > +         if (state->ddb_state_changed) {
> > > +                 bw_state = intel_bw_get_state(state);
> > > +                 if (IS_ERR(bw_state))
> > > +                         return PTR_ERR(bw_state);
> > > +
> > > +                 ret = intel_atomic_lock_global_state(&bw_state-
> > > >base);
> > > +                 if (ret)
> > > +                         return ret;
> > > +
> > > +                 DRM_DEBUG_KMS("Marking bw state changed for
> > > atomic state %p\n",
> > > +                               state);
> > > +
> > > +                 state->bw_state_changed = true;
> > > +         }
> > >            return 0;
> > > + }
> > >
> > >    for_each_oldnew_intel_crtc_in_state(state, crtc,
> > > old_crtc_state,
> > >                                        new_crtc_state, i) {
> > > @@ -447,7 +465,7 @@ int intel_bw_atomic_check(struct
> > > intel_atomic_state *state)
> > >                old_active_planes == new_active_planes)
> > >                    continue;
> > >
> > > -         bw_state  = intel_bw_get_state(state);
> > > +         bw_state = intel_bw_get_state(state);
> > >            if (IS_ERR(bw_state))
> > >                    return PTR_ERR(bw_state);
> > >
> > > @@ -468,6 +486,10 @@ int intel_bw_atomic_check(struct
> > > intel_atomic_state *state)
> > >    if (ret)
> > >            return ret;
> > >
> > > + DRM_DEBUG_KMS("Marking bw state changed for atomic state %p\n",
> > > state);
> > > +
> > > + state->bw_state_changed = true;
> > > +
> > >    data_rate = intel_bw_data_rate(dev_priv, bw_state);
> > >    num_active_planes = intel_bw_num_active_planes(dev_priv,
> > > bw_state);
> > >
> > > diff --git a/drivers/gpu/drm/i915/display/intel_display.c
> > > b/drivers/gpu/drm/i915/display/intel_display.c
> > > index 3031e64ee518..137fb645097a 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_display.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_display.c
> > > @@ -15540,8 +15540,10 @@ static void
> > > intel_atomic_commit_tail(struct intel_atomic_state *state)
> > >             * SKL workaround: bspec recommends we disable the SAGV
> > > when we
> > >             * have more then one pipe enabled
> > >             */
> > > -         if (!intel_can_enable_sagv(state))
> > > -                 intel_disable_sagv(dev_priv);
> > > +         if (state->bw_state_changed) {
> > > +                 if (!intel_can_enable_sagv(state))
> > > +                         intel_disable_sagv(dev_priv);
> > > +         }
> > >
> > >            intel_modeset_verify_disabled(dev_priv, state);
> > >    }
> > > @@ -15565,7 +15567,7 @@ static void intel_atomic_commit_tail(struct
> > > intel_atomic_state *state)
> > >            intel_encoders_update_prepare(state);
> > >
> > >    /* Enable all new slices, we might need */
> > > - if (state->modeset)
> > > + if (state->ddb_state_changed)
> > >            icl_dbuf_slice_pre_update(state);
> > >
> > >    /* Now enable the clocks, plane, pipe, and connectors that we
> > > set up. */
> > > @@ -15622,7 +15624,7 @@ static void intel_atomic_commit_tail(struct
> > > intel_atomic_state *state)
> > >    }
> > >
> > >    /* Disable all slices, we don't need */
> > > - if (state->modeset)
> > > + if (state->ddb_state_changed)
> > >            icl_dbuf_slice_post_update(state);
> > >
> > >    for_each_oldnew_intel_crtc_in_state(state, crtc,
> > > old_crtc_state, new_crtc_state, i) {
> > > @@ -15641,8 +15643,10 @@ static void
> > > intel_atomic_commit_tail(struct intel_atomic_state *state)
> > >    if (state->modeset)
> > >            intel_verify_planes(state);
> > >
> > > - if (state->modeset && intel_can_enable_sagv(state))
> > > -         intel_enable_sagv(dev_priv);
> > > + if (state->bw_state_changed) {
> > > +         if (intel_can_enable_sagv(state)
> > > +                 intel_enable_sagv(dev_priv);
> > > + }
> > >
> > >    drm_atomic_helper_commit_hw_done(&state->base);
> > >
> > > diff --git a/drivers/gpu/drm/i915/display/intel_display_types.h
> > > b/drivers/gpu/drm/i915/display/intel_display_types.h
> > > index 0d8a64305464..12b47ba3c68d 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_display_types.h
> > > +++ b/drivers/gpu/drm/i915/display/intel_display_types.h
> > > @@ -471,16 +471,6 @@ struct intel_atomic_state {
> > >
> > >    bool dpll_set, modeset;
> > >
> > > - /*
> > > -  * Does this transaction change the pipes that are
> > > active?  This mask
> > > -  * tracks which CRTC's have changed their active state at the
> > > end of
> > > -  * the transaction (not counting the temporary disable during
> > > modesets).
> > > -  * This mask should only be non-zero when intel_state->modeset
> > > is true,
> > > -  * but the converse is not necessarily true; simply changing a
> > > mode may
> > > -  * not flip the final active status of any CRTC's
> > > -  */
> > > - u8 active_pipe_changes;
> > > -
> > >    u8 active_pipes;
> > >
> > >    struct intel_shared_dpll_state shared_dpll[I915_NUM_PLLS];
> > > @@ -494,10 +484,30 @@ struct intel_atomic_state {
> > >    bool rps_interactive;
> > >
> > >    /*
> > > -  * active_pipes
> > > +  * active pipes
> > >     */
> > >    bool global_state_changed;
> > >
> > > + /*
> > > +  * Does this transaction change the pipes that are
> > > active?  This mask
> > > +  * tracks which CRTC's have changed their active state at the
> > > end of
> > > +  * the transaction (not counting the temporary disable during
> > > modesets).
> > > +  * This mask should only be non-zero when intel_state->modeset
> > > is true,
> > > +  * but the converse is not necessarily true; simply changing a
> > > mode may
> > > +  * not flip the final active status of any CRTC's
> > > +  */
> > > + u8 active_pipe_changes;
> > > +
> > > + /*
> > > +  * More granular change indicator for ddb changes
> > > +  */
> > > + bool ddb_state_changed;
> > > +
> > > + /*
> > > +  * More granular change indicator for bandwidth state changes
> > > +  */
> > > + bool bw_state_changed;
> > > +
> > >    /* Number of enabled DBuf slices */
> > >    u8 enabled_dbuf_slices_mask;
> > >
> > > diff --git a/drivers/gpu/drm/i915/intel_pm.c
> > > b/drivers/gpu/drm/i915/intel_pm.c
> > > index 409b91c17a7f..ac4b317ea1bf 100644
> > > --- a/drivers/gpu/drm/i915/intel_pm.c
> > > +++ b/drivers/gpu/drm/i915/intel_pm.c
> > > @@ -3894,7 +3894,7 @@ skl_ddb_get_pipe_allocation_limits(struct
> > > drm_i915_private *dev_priv,
> > >     * that changes the active CRTC list or do modeset would need
> > > to
> > >     * grab _all_ crtc locks, including the one we currently hold.
> > >     */
> > > - if (!intel_state->active_pipe_changes && !intel_state->modeset)
> > > {
> > > + if (!intel_state->ddb_state_changed) {
> > >            /*
> > >             * alloc may be cleared by clear_intel_crtc_state,
> > >             * copy from old state to be sure
> > > @@ -5787,6 +5787,9 @@ static int skl_wm_add_affected_planes(struct
> > > intel_atomic_state *state,
> > >                    return PTR_ERR(plane_state);
> > >
> > >            new_crtc_state->update_planes |= BIT(plane_id);
> > > +
> > > +         DRM_DEBUG_KMS("Marking ddb state changed for atomic
> > > state %p\n", state);
> > > +         state->ddb_state_changed = true;
> > >    }
> > >
> > >    return 0;
> > > --
> > > 2.24.1.485.gad05a3d8e5
> >
> >

--
Ville Syrjälä
Intel

[-- Attachment #1.2: Type: text/html, Size: 34926 bytes --]

[-- Attachment #2: Type: text/plain, Size: 160 bytes --]

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx