> -----Original Message-----
> From: Vivekanandan, Balasubramani
> <balasubramani.vivekanandan@xxxxxxxxx>
> Sent: Thursday, October 20, 2022 7:42 AM
> To: Srivatsa, Anusha <anusha.srivatsa@xxxxxxxxx>; intel-
> gfx@xxxxxxxxxxxxxxxxxxxxx
> Subject: Re: [PATCH 1/2] drm/i915/display: Do both crawl and
> squash when changing cdclk
>
> On 13.10.2022 16:32, Anusha Srivatsa wrote:
> > From: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
> >
> > For MTL, changing cdclk from between certain frequencies has both
> > squash and crawl. Use the current cdclk config and the new(desired)
> > cdclk config to construtc a mid cdclk config.
> > Set the cdclk twice:
> > - Current cdclk -> mid cdclk
> > - mid cdclk -> desired cdclk
> >
> > v2: Add check in intel_modeset_calc_cdclk() to avoid cdclk change via
> > modeset for platforms that support squash_crawl sequences(Ville)
> >
> > v3: Add checks for:
> > - scenario where only slow clock is used and cdclk is actually 0
> > (bringing up display).
> > - PLLs are on before looking up the waveform.
> > - Squash and crawl capability checks.(Ville)
> >
> > Signed-off-by: Anusha Srivatsa <anusha.srivatsa@xxxxxxxxx>
> > Signed-off-by: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
>
> Please add the Bspec number.
Will do.
>
> > ---
> > drivers/gpu/drm/i915/display/intel_cdclk.c | 157
> > +++++++++++++++++----
> > 1 file changed, 128 insertions(+), 29 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/display/intel_cdclk.c
> > b/drivers/gpu/drm/i915/display/intel_cdclk.c
> > index ad401357ab66..430b4cb0a8ab 100644
> > --- a/drivers/gpu/drm/i915/display/intel_cdclk.c
> > +++ b/drivers/gpu/drm/i915/display/intel_cdclk.c
> > @@ -1675,7 +1675,7 @@ static u32 cdclk_squash_waveform(struct
> drm_i915_private *dev_priv,
> > const struct intel_cdclk_vals *table = dev_priv->display.cdclk.table;
> > int i;
> >
> > - if (cdclk == dev_priv->display.cdclk.hw.bypass)
> > + if (cdclk == dev_priv->display.cdclk.hw.bypass || cdclk == 0)
> > return 0;
> >
> > for (i = 0; table[i].refclk; i++)
> > @@ -1689,37 +1689,72 @@ static u32 cdclk_squash_waveform(struct
> drm_i915_private *dev_priv,
> > return 0xffff;
> > }
> >
> > -static void bxt_set_cdclk(struct drm_i915_private *dev_priv,
> > - const struct intel_cdclk_config *cdclk_config,
> > - enum pipe pipe)
> > +static int cdclk_squash_divider(u16 waveform) {
> > + return hweight16(waveform ?: 0xffff); }
> > +
> > +static bool cdclk_crawl_and_squash(struct drm_i915_private *i915,
> > + const struct intel_cdclk_config
> *old_cdclk_config,
> > + const struct intel_cdclk_config
> *new_cdclk_config,
> > + struct intel_cdclk_config *mid_cdclk_config)
> {
>
> I was thinking of asking to rename this function to a more descriptive one,
> but then I myself was not able to come up with one.
> For a fresh eyes, it is difficult to make out what this function is actually doing.
> Can you please add a summary as a comment above this function pointing
> out what is mid_cdclk and whats the meaning of its return value.
I thought the commit message was explaining the needful. But I agree if it is indeed confusing, adding comments in code would help, will churn that change.
>
> > + u16 old_waveform = cdclk_squash_waveform(i915, old_cdclk_config-
> >cdclk);
> > + u16 new_waveform = cdclk_squash_waveform(i915,
> new_cdclk_config->cdclk);
> > + u16 mid_waveform;
> > + int size = 16;
> > + int div = 2;
> > +
> > + /* Return if both Squash and Crawl are not present */
> > + if (!HAS_CDCLK_CRAWL(i915) || !has_cdclk_squasher(i915))
> > + return false;
>
> Can cdclk_squasher feature availability be also made a part of device_info
> structure like cdclk_crawl and create a macro similar to HAS_CDCLK_CRAWL?
> Like Ville said it looks bit weird. Also we would avoid adding platform checks
> inside has_cdclk_squasher() function like it is done now in your second
> patch.
Yes will make that change.
>
> > +
> > + /* Return if Squash only or Crawl only is the desired action */
> > + if (old_cdclk_config->vco <= 0 || new_cdclk_config->vco <= 0 ||
> > + old_cdclk_config->vco == new_cdclk_config->vco ||
> > + old_waveform == new_waveform)
> > + return false;
> > +
> > + *mid_cdclk_config = *new_cdclk_config;
> > +
> > + /* If moving to a higher cdclk(squash) the mid cdclk config
> > + * should have the new (squash) waveform.
> > + * If moving to a lower cdclk (crawl) the mid cdclk config
> > + * should have the new vco.
> > + */
> > +
> > + if (cdclk_squash_divider(new_waveform) >
> cdclk_squash_divider(old_waveform)) {
> > + mid_cdclk_config->vco = old_cdclk_config->vco;
> > + mid_waveform = new_waveform;
> > + } else {
> > + mid_cdclk_config->vco = new_cdclk_config->vco;
> > + mid_waveform = old_waveform;
> > + }
> > +
> > + mid_cdclk_config->cdclk =
> DIV_ROUND_CLOSEST(cdclk_squash_divider(mid_waveform) *
> > + mid_cdclk_config->vco, size
> * div);
> > +
> > + /* make sure the mid clock came out sane */
> > +
> > + drm_WARN_ON(&i915->drm, mid_cdclk_config->cdclk <
> > + min(old_cdclk_config->cdclk, new_cdclk_config->cdclk));
> > + drm_WARN_ON(&i915->drm, mid_cdclk_config->cdclk >
> > + i915->display.cdclk.max_cdclk_freq);
> > + drm_WARN_ON(&i915->drm, cdclk_squash_waveform(i915,
> mid_cdclk_config->cdclk) !=
> > + mid_waveform);
> > +
> > + return true;
> > +}
> > +
> > +static void _bxt_set_cdclk(struct drm_i915_private *dev_priv,
> > + const struct intel_cdclk_config *cdclk_config,
> > + enum pipe pipe)
> > {
> > int cdclk = cdclk_config->cdclk;
> > int vco = cdclk_config->vco;
> > u32 val;
> > u16 waveform;
> > int clock;
> > - int ret;
> > -
> > - /* Inform power controller of upcoming frequency change. */
> > - if (DISPLAY_VER(dev_priv) >= 11)
> > - ret = skl_pcode_request(&dev_priv->uncore,
> SKL_PCODE_CDCLK_CONTROL,
> > - SKL_CDCLK_PREPARE_FOR_CHANGE,
> > - SKL_CDCLK_READY_FOR_CHANGE,
> > - SKL_CDCLK_READY_FOR_CHANGE, 3);
> > - else
> > - /*
> > - * BSpec requires us to wait up to 150usec, but that leads to
> > - * timeouts; the 2ms used here is based on experiment.
> > - */
> > - ret = snb_pcode_write_timeout(&dev_priv->uncore,
> > -
> HSW_PCODE_DE_WRITE_FREQ_REQ,
> > - 0x80000000, 150, 2);
> > - if (ret) {
> > - drm_err(&dev_priv->drm,
> > - "Failed to inform PCU about cdclk change (err %d,
> freq %d)\n",
> > - ret, cdclk);
> > - return;
> > - }
> >
> > if (HAS_CDCLK_CRAWL(dev_priv) && dev_priv->display.cdclk.hw.vco
> > 0 && vco > 0) {
> > if (dev_priv->display.cdclk.hw.vco != vco) @@ -1772,6
> +1807,44 @@
> > static void bxt_set_cdclk(struct drm_i915_private *dev_priv,
> >
> > if (pipe != INVALID_PIPE)
> >
> intel_crtc_wait_for_next_vblank(intel_crtc_for_pipe(dev_priv,
> > pipe));
> > +}
> > +
> > +static void bxt_set_cdclk(struct drm_i915_private *dev_priv,
> > + const struct intel_cdclk_config *cdclk_config,
> > + enum pipe pipe)
> > +{
>
> bxt_set_cdclk() is now bloated as it is reused for new platforms with new
> features resulting in too many condition checks. I see it is now time to switch
> to a new set_cdclk() function. I would prefer a new function for platforms
> starting from which supports squash/crawl feature and add this new
> crawl_and_squash feature implementation to the same function. But
> definitely I dislike using bxt_set_cdclk for MTL.
>
> > + struct intel_cdclk_config mid_cdclk_config;
> > + int cdclk = cdclk_config->cdclk;
> > + int ret;
> > +
> > + /* Inform power controller of upcoming frequency change. */
> > + if (DISPLAY_VER(dev_priv) >= 11)
> > + ret = skl_pcode_request(&dev_priv->uncore,
> SKL_PCODE_CDCLK_CONTROL,
> > + SKL_CDCLK_PREPARE_FOR_CHANGE,
> > + SKL_CDCLK_READY_FOR_CHANGE,
> > + SKL_CDCLK_READY_FOR_CHANGE, 3);
> > + else
> > + /*
> > + * BSpec requires us to wait up to 150usec, but that leads to
> > + * timeouts; the 2ms used here is based on experiment.
> > + */
> > + ret = snb_pcode_write_timeout(&dev_priv->uncore,
> > +
> HSW_PCODE_DE_WRITE_FREQ_REQ,
> > + 0x80000000, 150, 2);
> > + if (ret) {
> > + drm_err(&dev_priv->drm,
> > + "Failed to inform PCU about cdclk change (err %d,
> freq %d)\n",
> > + ret, cdclk);
> > + return;
> > + }
> > +
> > + if (cdclk_crawl_and_squash(dev_priv, &dev_priv->display.cdclk.hw,
> > + cdclk_config, &mid_cdclk_config)) {
> > + _bxt_set_cdclk(dev_priv, &mid_cdclk_config, pipe);
> > + _bxt_set_cdclk(dev_priv, cdclk_config, pipe);
> > + } else {
> > + _bxt_set_cdclk(dev_priv, cdclk_config, pipe);
> > + }
> >
> > if (DISPLAY_VER(dev_priv) >= 11) {
> > ret = snb_pcode_write(&dev_priv->uncore,
> SKL_PCODE_CDCLK_CONTROL,
> > @@ -1944,6 +2017,27 @@ void intel_cdclk_uninit_hw(struct
> drm_i915_private *i915)
> > skl_cdclk_uninit_hw(i915);
> > }
> >
> > +static bool intel_cdclk_can_crawl_and_squash(struct drm_i915_private
> *i915,
> > + const struct intel_cdclk_config *a,
> > + const struct intel_cdclk_config *b) {
> > + u16 old_waveform;
> > + u16 new_waveform;
> > +
> > + if (a->vco == 0 || b->vco == 0)
> > + return false;
> > +
> > + if (HAS_CDCLK_CRAWL(i915) && has_cdclk_squasher(i915)) {
> > + old_waveform = cdclk_squash_waveform(i915, a->cdclk);
> > + new_waveform = cdclk_squash_waveform(i915, b->cdclk);
> > + } else {
> > + return false;
> > + }
> > +
> > + return a->vco != b->vco &&
> > + old_waveform != new_waveform; }
> > +
> > static bool intel_cdclk_can_crawl(struct drm_i915_private *dev_priv,
> > const struct intel_cdclk_config *a,
> > const struct intel_cdclk_config *b) @@ -
> 2750,9 +2844,14 @@ int
> > intel_modeset_calc_cdclk(struct intel_atomic_state *state)
> > pipe = INVALID_PIPE;
> > }
> >
> > - if (intel_cdclk_can_squash(dev_priv,
> > - &old_cdclk_state->actual,
> > - &new_cdclk_state->actual)) {
> > + if (intel_cdclk_can_crawl_and_squash(dev_priv,
> > + &old_cdclk_state->actual,
> > + &new_cdclk_state->actual)) {
> > + drm_dbg_kms(&dev_priv->drm,
> > + "Can change cdclk via crawler and squasher\n");
> > + } else if (intel_cdclk_can_squash(dev_priv,
> > + &old_cdclk_state->actual,
> > + &new_cdclk_state->actual)) {
>
> In the bxt_set_cdclk(), we perform crawl_and_squash only if neither crawl
> and squash alone can't accomplish cdclk change. So move the
> intel_cdclk_can_crawl_and_squash() check to after the checks for crawl and
> squash individually.
> This would just make sure the logs reflect how actually the cdclk is changed.
As @ville.syrjala@xxxxxxxxxxxxxxx mentioned , this ordering is the right one to follow. We do not want MTL to fall into can_squash() for cases where it should actually be doing crawl_and_squash()
Thanks for the feedback!
Anusha
> Regards,
> Bala
>
> > drm_dbg_kms(&dev_priv->drm,
> > "Can change cdclk via squasher\n");
> > } else if (intel_cdclk_can_crawl(dev_priv,
> > --
> > 2.25.1
> >
