[Intel-gfx] [PATCH] drm/i915/display: Communicate display power demands to pcode more accurately
Lisovskiy, Stanislav
stanislav.lisovskiy at intel.com
Tue Mar 7 11:15:06 UTC 2023
On Tue, Mar 07, 2023 at 12:23:59PM +0200, Jani Nikula wrote:
> On Tue, 07 Mar 2023, "Lisovskiy, Stanislav" <stanislav.lisovskiy at intel.com> wrote:
> > On Mon, Mar 06, 2023 at 08:14:35PM +0200, Jani Nikula wrote:
> >> On Mon, 27 Feb 2023, Stanislav Lisovskiy <stanislav.lisovskiy at intel.com> wrote:
> >> > Display to communicate display pipe count/CDCLK/voltage configuration
> >> > to Pcode for more accurate power accounting for gen >= 12.
> >> > Existing sequence is only sending the voltage value to the Pcode.
> >> > Adding new sequence with current cdclk associate with voltage value masking.
> >> > Adding pcode request when any pipe power well will disable or enable.
> >> >
> >> > v2: - Make intel_cdclk_need_serialize static to make CI compiler happy.
> >> >
> >> > Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy at intel.com>
> >> > ---
> >> > drivers/gpu/drm/i915/display/intel_cdclk.c | 174 +++++++++++++++++++--
> >> > drivers/gpu/drm/i915/display/intel_cdclk.h | 2 +
> >> > drivers/gpu/drm/i915/i915_reg.h | 14 ++
> >> > 3 files changed, 176 insertions(+), 14 deletions(-)
> >> >
> >> > diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c b/drivers/gpu/drm/i915/display/intel_cdclk.c
> >> > index 084a483f9776..df5a21f6a393 100644
> >> > --- a/drivers/gpu/drm/i915/display/intel_cdclk.c
> >> > +++ b/drivers/gpu/drm/i915/display/intel_cdclk.c
> >> > @@ -1932,10 +1932,10 @@ static void bxt_set_cdclk(struct drm_i915_private *dev_priv,
> >> > * NOOP - No Pcode communication needed for
> >> > * Display versions 14 and beyond
> >> > */;
> >> > - else if (DISPLAY_VER(dev_priv) >= 11)
> >> > + else if (DISPLAY_VER(dev_priv) >= 11 && !IS_DG2(dev_priv))
> >> > ret = snb_pcode_write(&dev_priv->uncore, SKL_PCODE_CDCLK_CONTROL,
> >> > cdclk_config->voltage_level);
> >> > - else
> >> > + if (DISPLAY_VER(dev_priv) < 11) {
> >> > /*
> >> > * The timeout isn't specified, the 2ms used here is based on
> >> > * experiment.
> >> > @@ -1946,7 +1946,7 @@ static void bxt_set_cdclk(struct drm_i915_private *dev_priv,
> >> > HSW_PCODE_DE_WRITE_FREQ_REQ,
> >> > cdclk_config->voltage_level,
> >> > 150, 2);
> >> > -
> >> > + }
> >> > if (ret) {
> >> > drm_err(&dev_priv->drm,
> >> > "PCode CDCLK freq set failed, (err %d, freq %d)\n",
> >> > @@ -2242,6 +2242,40 @@ void intel_cdclk_dump_config(struct drm_i915_private *i915,
> >> > cdclk_config->voltage_level);
> >> > }
> >> >
> >> > +static void intel_pcode_notify(struct drm_i915_private *i915,
> >> > + u8 voltage_level,
> >> > + u8 active_pipe_count,
> >> > + u8 cdclk,
>
> How can that be u8?
Agree, cdclk uses 10 bits according to BSpec, for the rest u8 is enough.
Thanks for noticing.
>
> >> > + bool cdclk_update_valid,
> >> > + bool pipe_count_update_valid)
>
> Also not super happy about the flags arguments... could have cdclk == 0
> means not valid and int active_pipe_count == -1 means not valid. Maybe.
Those are actually called that way, because BSpec calls those bits so.
In fact it is more like "should we update this" flag.
As I understand if it is not set PCode will simply ignore those bits in
notification.
The calling site should determine whether this bit need to be set or not.
We set the to true only if correspodent state got changed.
I think those are validated already, so we shouldn't expect any bogus values
there(otherwise we are screwed in quite many other places), however cdclk == 0
actually would be valid and this bit would be then set.
Active pipe count is always amount of "1"'s in cdclk_state->active_pipes mask, so it shouldn't
ever become -1. Or if it becomes, once again, this function would be the least
of our problems.
>
> >> > +{
> >> > + int ret;
> >> > + u32 update_mask = 0;
> >> > +
> >> > + if (DISPLAY_VER(i915) < 12)
> >> > + return;
> >> > +
> >> > + update_mask = DISPLAY_TO_PCODE_UPDATE_MASK(cdclk, active_pipe_count, voltage_level);
> >> > +
> >> > + if (cdclk_update_valid)
> >> > + update_mask |= DISPLAY_TO_PCODE_CDCLK_VALID;
> >> > +
> >> > + if (pipe_count_update_valid)
> >> > + update_mask |= DISPLAY_TO_PCODE_PIPE_COUNT_VALID;
> >> > +
> >> > + ret = skl_pcode_request(&i915->uncore, SKL_PCODE_CDCLK_CONTROL,
> >> > + SKL_CDCLK_PREPARE_FOR_CHANGE |
> >> > + update_mask,
> >> > + SKL_CDCLK_READY_FOR_CHANGE,
> >> > + SKL_CDCLK_READY_FOR_CHANGE, 3);
> >> > + if (ret) {
> >> > + drm_err(&i915->drm,
> >> > + "Failed to inform PCU about display config (err %d)\n",
> >> > + ret);
> >> > + return;
> >>
> >> Superfluous return.
> >>
> >> > + }
> >> > +}
> >> > +
> >> > /**
> >> > * intel_set_cdclk - Push the CDCLK configuration to the hardware
> >> > * @dev_priv: i915 device
> >> > @@ -2311,6 +2345,98 @@ static void intel_set_cdclk(struct drm_i915_private *dev_priv,
> >> > }
> >> > }
> >> >
> >> > +/**
> >> > + * intel_cdclk_power_usage_to_pcode_pre_notification: display to pcode notification
>
> This doesn't need to be a kernel-doc comment (and it's a malformed one
> at that). Ditto for the others.
>
> >> > + * before the enabling power wells.
> >> > + * send notification with cdclk, number of pipes, voltage_level.
> >> > + * @state: intel atomic state
> >> > + */
> >> > +void intel_cdclk_power_usage_to_pcode_pre_notification(struct intel_atomic_state *state)
> >>
> >> This could be static?
> >>
> >> Not happy with the name, really, but at least it could use a
> >> verb. Notify instead of notification.
> >
> > Any hint for better naming?
>
> Maybe just intel_cdclk_pcode_pre_notify(),
> intel_cdclk_pcode_post_notify(), and then intel_cdclk_pcode_notify() as
> the shared thing?
Sorry Jani, but I think we would be going in circles then, because initially you were stating that:
"The function name should indicate what is being
notified. Pre-notification of what? Post-notification of what?
Functions in this file should be prefixed intel_cdclk_*."
At this point it was called "intel_display_to_pcode_pre_notification", the prefix comment
of course is now changed.
So if I change it back to "pcode_pre_notify" thing, that would lose information
about what its doing once again.
I can for sure do that, but then what is the guarantee it won't get a comment about
"lack of informativeness" once again? :)
Stan
>
> >
> >>
> >> > +{
> >> > + struct drm_i915_private *dev_priv = to_i915(state->base.dev);
> >>
> >> No new dev_priv please.
> >>
> >> > + const struct intel_cdclk_state *old_cdclk_state =
> >> > + intel_atomic_get_old_cdclk_state(state);
> >> > + const struct intel_cdclk_state *new_cdclk_state =
> >> > + intel_atomic_get_new_cdclk_state(state);
> >> > + unsigned int cdclk = 0; u8 voltage_level, num_active_pipes = 0;
> >> > + bool change_cdclk, update_pipe_count;
> >> > +
> >> > + if (!intel_cdclk_changed(&old_cdclk_state->actual,
> >> > + &new_cdclk_state->actual) &&
> >> > + (new_cdclk_state->active_pipes ==
> >> > + old_cdclk_state->active_pipes))
> >> > + return;
> >> > +
> >> > + /* According to "Sequence Before Frequency Change", voltage level set to 0x3 */
> >> > + voltage_level = DISPLAY_TO_PCODE_VOLTAGE_MAX;
> >> > +
> >> > + change_cdclk = new_cdclk_state->actual.cdclk != old_cdclk_state->actual.cdclk;
> >> > + update_pipe_count = hweight8(new_cdclk_state->active_pipes) >
> >> > + hweight8(old_cdclk_state->active_pipes);
> >> > +
> >> > + /*
> >> > + * According to "Sequence Before Frequency Change",
> >> > + * if CDCLK is increasing, set bits 25:16 to upcoming CDCLK,
> >> > + * if CDCLK is decreasing or not changing, set bits 25:16 to current CDCLK,
> >> > + * which basically means we choose the maximum of old and new CDCLK, if we know both
> >> > + */
> >> > + if (change_cdclk)
> >> > + cdclk = max(new_cdclk_state->actual.cdclk, old_cdclk_state->actual.cdclk);
> >> > +
> >> > + /*
> >> > + * According to "Sequence For Pipe Count Change",
> >> > + * if pipe count is increasing, set bits 25:16 to upcoming pipe count
> >> > + * (power well is enabled)
> >> > + * no action if it is decreasing, before the change
> >> > + */
> >> > + if (update_pipe_count)
> >> > + num_active_pipes = hweight8(new_cdclk_state->active_pipes);
> >> > +
> >> > + intel_pcode_notify(dev_priv, voltage_level, num_active_pipes, cdclk,
> >> > + change_cdclk, update_pipe_count);
> >> > +}
> >> > +
> >> > +/* intel_cdclk_power_usage_to_pcode_post_notification: after frequency change sending
> >> > + * voltage_level, active pipes, current CDCLK frequency.
> >> > + * @state: intel atomic state
> >> > + */
> >> > +void intel_cdclk_power_usage_to_pcode_post_notification(struct intel_atomic_state *state)
> >>
> >> Ditto about static & naming.
> >>
> >> > +{
> >> > + struct drm_i915_private *dev_priv = to_i915(state->base.dev);
> >>
> >> No new dev_priv please.
> >>
> >> > + const struct intel_cdclk_state *new_cdclk_state =
> >> > + intel_atomic_get_new_cdclk_state(state);
> >> > + const struct intel_cdclk_state *old_cdclk_state =
> >> > + intel_atomic_get_old_cdclk_state(state);
> >> > + unsigned int cdclk = 0; u8 voltage_level, num_active_pipes = 0;
> >> > + bool update_cdclk, update_pipe_count;
> >> > +
> >> > + /* According to "Sequence After Frequency Change", set voltage to used level */
> >> > + voltage_level = new_cdclk_state->actual.voltage_level;
> >> > +
> >> > + update_cdclk = new_cdclk_state->actual.cdclk != old_cdclk_state->actual.cdclk;
> >> > + update_pipe_count = hweight8(new_cdclk_state->active_pipes) <
> >> > + hweight8(old_cdclk_state->active_pipes);
> >> > +
> >> > + /*
> >> > + * According to "Sequence After Frequency Change",
> >> > + * set bits 25:16 to current CDCLK
> >> > + */
> >> > + if (update_cdclk)
> >> > + cdclk = new_cdclk_state->actual.cdclk;
> >> > +
> >> > + /*
> >> > + * According to "Sequence For Pipe Count Change",
> >> > + * if pipe count is decreasing, set bits 25:16 to current pipe count,
> >> > + * after the change(power well is disabled)
> >> > + * no action if it is increasing, after the change
> >> > + */
> >> > + if (update_pipe_count)
> >> > + num_active_pipes = hweight8(new_cdclk_state->active_pipes);
> >> > +
> >> > + intel_pcode_notify(dev_priv, voltage_level, num_active_pipes, cdclk,
> >> > + update_cdclk, update_pipe_count);
> >> > +}
> >> > +
> >> > /**
> >> > * intel_set_cdclk_pre_plane_update - Push the CDCLK state to the hardware
> >> > * @state: intel atomic state
> >> > @@ -2321,7 +2447,7 @@ static void intel_set_cdclk(struct drm_i915_private *dev_priv,
> >> > void
> >> > intel_set_cdclk_pre_plane_update(struct intel_atomic_state *state)
> >> > {
> >> > - struct drm_i915_private *dev_priv = to_i915(state->base.dev);
> >> > + struct drm_i915_private *i915 = to_i915(state->base.dev);
> >> > const struct intel_cdclk_state *old_cdclk_state =
> >> > intel_atomic_get_old_cdclk_state(state);
> >> > const struct intel_cdclk_state *new_cdclk_state =
> >> > @@ -2332,11 +2458,14 @@ intel_set_cdclk_pre_plane_update(struct intel_atomic_state *state)
> >> > &new_cdclk_state->actual))
> >> > return;
> >> >
> >> > + if (DISPLAY_VER(i915) >= 12)
> >> > + intel_cdclk_power_usage_to_pcode_pre_notification(state);
> >> > +
> >> > if (pipe == INVALID_PIPE ||
> >> > old_cdclk_state->actual.cdclk <= new_cdclk_state->actual.cdclk) {
> >> > - drm_WARN_ON(&dev_priv->drm, !new_cdclk_state->base.changed);
> >> > + drm_WARN_ON(&i915->drm, !new_cdclk_state->base.changed);
> >> >
> >> > - intel_set_cdclk(dev_priv, &new_cdclk_state->actual, pipe);
> >> > + intel_set_cdclk(i915, &new_cdclk_state->actual, pipe);
> >> > }
> >> > }
> >> >
> >> > @@ -2350,7 +2479,7 @@ intel_set_cdclk_pre_plane_update(struct intel_atomic_state *state)
> >> > void
> >> > intel_set_cdclk_post_plane_update(struct intel_atomic_state *state)
> >> > {
> >> > - struct drm_i915_private *dev_priv = to_i915(state->base.dev);
> >> > + struct drm_i915_private *i915 = to_i915(state->base.dev);
> >> > const struct intel_cdclk_state *old_cdclk_state =
> >> > intel_atomic_get_old_cdclk_state(state);
> >> > const struct intel_cdclk_state *new_cdclk_state =
> >> > @@ -2361,11 +2490,14 @@ intel_set_cdclk_post_plane_update(struct intel_atomic_state *state)
> >> > &new_cdclk_state->actual))
> >> > return;
> >> >
> >> > + if (DISPLAY_VER(i915) >= 12)
> >> > + intel_cdclk_power_usage_to_pcode_post_notification(state);
> >> > +
> >> > if (pipe != INVALID_PIPE &&
> >> > old_cdclk_state->actual.cdclk > new_cdclk_state->actual.cdclk) {
> >> > - drm_WARN_ON(&dev_priv->drm, !new_cdclk_state->base.changed);
> >> > + drm_WARN_ON(&i915->drm, !new_cdclk_state->base.changed);
> >> >
> >> > - intel_set_cdclk(dev_priv, &new_cdclk_state->actual, pipe);
> >> > + intel_set_cdclk(i915, &new_cdclk_state->actual, pipe);
> >> > }
> >> > }
> >> >
> >> > @@ -2871,6 +3003,21 @@ int intel_cdclk_init(struct drm_i915_private *dev_priv)
> >> > return 0;
> >> > }
> >> >
> >> > +static bool intel_cdclk_need_serialize(struct drm_i915_private *i915,
> >> > + const struct intel_cdclk_state *old_cdclk_state,
> >> > + const struct intel_cdclk_state *new_cdclk_state)
> >> > +{
> >> > + /*
> >> > + * We need to poke hw for gen >= 12, because we notify PCode if
> >> > + * pipe power well count changes.
> >> > + */
> >> > + return intel_cdclk_changed(&old_cdclk_state->actual,
> >> > + &new_cdclk_state->actual) ||
> >> > + ((DISPLAY_VER(i915) >= 12 &&
> >> > + hweight8(old_cdclk_state->active_pipes) !=
> >> > + hweight8(new_cdclk_state->active_pipes)));
> >>
> >> That's getting a bit unweildy for one expression.
> >>
> >> > +}
> >> > +
> >> > int intel_modeset_calc_cdclk(struct intel_atomic_state *state)
> >> > {
> >> > struct drm_i915_private *dev_priv = to_i915(state->base.dev);
> >> > @@ -2892,8 +3039,7 @@ int intel_modeset_calc_cdclk(struct intel_atomic_state *state)
> >> > if (ret)
> >> > return ret;
> >> >
> >> > - if (intel_cdclk_changed(&old_cdclk_state->actual,
> >> > - &new_cdclk_state->actual)) {
> >> > + if (intel_cdclk_need_serialize(dev_priv, old_cdclk_state, new_cdclk_state)) {
> >> > /*
> >> > * Also serialize commits across all crtcs
> >> > * if the actual hw needs to be poked.
> >> > @@ -2905,9 +3051,9 @@ int intel_modeset_calc_cdclk(struct intel_atomic_state *state)
> >> > old_cdclk_state->force_min_cdclk != new_cdclk_state->force_min_cdclk ||
> >> > intel_cdclk_changed(&old_cdclk_state->logical,
> >> > &new_cdclk_state->logical)) {
> >> > - ret = intel_atomic_lock_global_state(&new_cdclk_state->base);
> >> > - if (ret)
> >> > - return ret;
> >> > + ret = intel_atomic_lock_global_state(&new_cdclk_state->base);
> >> > + if (ret)
> >> > + return ret;
>
> Accidental indentation change?
>
> >> > } else {
> >> > return 0;
> >> > }
> >> > diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.h b/drivers/gpu/drm/i915/display/intel_cdclk.h
> >> > index 51e2f6a11ce4..fa356adc61d9 100644
> >> > --- a/drivers/gpu/drm/i915/display/intel_cdclk.h
> >> > +++ b/drivers/gpu/drm/i915/display/intel_cdclk.h
> >> > @@ -64,6 +64,8 @@ bool intel_cdclk_needs_modeset(const struct intel_cdclk_config *a,
> >> > const struct intel_cdclk_config *b);
> >> > void intel_set_cdclk_pre_plane_update(struct intel_atomic_state *state);
> >> > void intel_set_cdclk_post_plane_update(struct intel_atomic_state *state);
> >> > +void intel_cdclk_power_usage_to_pcode_pre_notification(struct intel_atomic_state *state);
> >> > +void intel_cdclk_power_usage_to_pcode_post_notification(struct intel_atomic_state *state);
> >> > void intel_cdclk_dump_config(struct drm_i915_private *i915,
> >> > const struct intel_cdclk_config *cdclk_config,
> >> > const char *context);
> >> > diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> >> > index c1efa655fb68..9f786952585b 100644
> >> > --- a/drivers/gpu/drm/i915/i915_reg.h
> >> > +++ b/drivers/gpu/drm/i915/i915_reg.h
> >> > @@ -6526,6 +6526,20 @@
> >> > #define ICL_PCODE_MEM_SS_READ_GLOBAL_INFO (0x0 << 8)
> >> > #define ICL_PCODE_MEM_SS_READ_QGV_POINT_INFO(point) (((point) << 16) | (0x1 << 8))
> >> > #define ADL_PCODE_MEM_SS_READ_PSF_GV_INFO ((0) | (0x2 << 8))
> >> > +#define DISPLAY_TO_PCODE_CDCLK_MAX 0x28D
> >> > +#define DISPLAY_TO_PCODE_VOLTAGE_MASK REG_GENMASK(1, 0)
> >> > +#define DISPLAY_TO_PCODE_VOLTAGE_MAX DISPLAY_TO_PCODE_VOLTAGE_MASK
> >> > +#define DISPLAY_TO_PCODE_CDCLK_VALID REG_BIT(27)
> >> > +#define DISPLAY_TO_PCODE_PIPE_COUNT_VALID REG_BIT(31)
> >> > +#define DISPLAY_TO_PCODE_CDCLK_MASK REG_GENMASK(25, 16)
> >> > +#define DISPLAY_TO_PCODE_PIPE_COUNT_MASK REG_GENMASK(30, 28)
> >> > +#define DISPLAY_TO_PCODE_CDCLK(x) REG_FIELD_PREP(DISPLAY_TO_PCODE_CDCLK_MASK, (x))
> >> > +#define DISPLAY_TO_PCODE_PIPE_COUNT(x) REG_FIELD_PREP(DISPLAY_TO_PCODE_PIPE_COUNT_MASK, (x))
> >> > +#define DISPLAY_TO_PCODE_VOLTAGE(x) REG_FIELD_PREP(DISPLAY_TO_PCODE_VOLTAGE_MASK, (x))
> >> > +#define DISPLAY_TO_PCODE_UPDATE_MASK(cdclk, num_pipes, voltage_level) \
> >> > + (DISPLAY_TO_PCODE_CDCLK(cdclk)) | \
> >> > + (DISPLAY_TO_PCODE_PIPE_COUNT(num_pipes)) | \
> >> > + (DISPLAY_TO_PCODE_VOLTAGE(voltage_level))
> >> > #define ICL_PCODE_SAGV_DE_MEM_SS_CONFIG 0xe
> >> > #define ICL_PCODE_REP_QGV_MASK REG_GENMASK(1, 0)
> >> > #define ICL_PCODE_REP_QGV_SAFE REG_FIELD_PREP(ICL_PCODE_REP_QGV_MASK, 0)
> >>
> >> --
> >> Jani Nikula, Intel Open Source Graphics Center
>
> --
> Jani Nikula, Intel Open Source Graphics Center
More information about the Intel-gfx
mailing list