On Fri, Sep 24, 2021 at 05:10:44PM +0000, Souza, Jose wrote:
> On Fri, 2021-09-24 at 17:35 +0300, Ville Syrjälä wrote:
> > On Thu, Sep 23, 2021 at 12:46:11PM -0700, José Roberto de Souza wrote:
> > > Alderlake-P was getting 'max time under evasion' messages when PSR2
> > > was enabled, this is due PIPE_SCANLINE/PIPEDSL returning 0 over a
> > > period of time longer than VBLANK_EVASION_TIME_US.
> > >
> > > For PSR1 we had the same issue so intel_psr_wait_for_idle() was
> > > implemented to wait for PSR1 to get into idle state but nothing was
> > > done for PSR2.
> > >
> > > For PSR2 we can't only wait for idle state as PSR2 tends to keep
> > > into sleep state that means it is ready to send selective updates.
> > >
> > > To do so it was necessary to add intel_wait_for_condition(), this
> > > takes as parameter a function that will return true when the desidered
> > > condition is meet.
> > >
> > > Cc: Gwan-gyeong Mun <gwan-gyeong....@intel.com>
> > > Signed-off-by: José Roberto de Souza <jose.so...@intel.com>
> > > ---
> > > .../drm/i915/display/intel_display_debugfs.c | 3 +-
> > > drivers/gpu/drm/i915/display/intel_psr.c | 64 ++++++++++++-------
> > > drivers/gpu/drm/i915/i915_reg.h | 5 +-
> > > drivers/gpu/drm/i915/intel_uncore.c | 47 ++++++++++++++
> > > drivers/gpu/drm/i915/intel_uncore.h | 7 ++
> > > 5 files changed, 100 insertions(+), 26 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/display/intel_display_debugfs.c
> > > b/drivers/gpu/drm/i915/display/intel_display_debugfs.c
> > > index 68f4ba8c46e75..662596adb1da6 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_display_debugfs.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_display_debugfs.c
> > > @@ -303,8 +303,7 @@ psr_source_status(struct intel_dp *intel_dp, struct
> > > seq_file *m)
> > > };
> > > val = intel_de_read(dev_priv,
> > > EDP_PSR2_STATUS(intel_dp->psr.transcoder));
> > > - status_val = (val & EDP_PSR2_STATUS_STATE_MASK) >>
> > > - EDP_PSR2_STATUS_STATE_SHIFT;
> > > + status_val = REG_FIELD_GET(EDP_PSR2_STATUS_STATE_MASK, val);
> > > if (status_val < ARRAY_SIZE(live_status))
> > > status = live_status[status_val];
> > > } else {
> > > diff --git a/drivers/gpu/drm/i915/display/intel_psr.c
> > > b/drivers/gpu/drm/i915/display/intel_psr.c
> > > index 19a96d3c4acf4..a2e4ef42be60a 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_psr.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_psr.c
> > > @@ -1796,15 +1796,33 @@ void intel_psr_post_plane_update(const struct
> > > intel_atomic_state *state)
> > > _intel_psr_post_plane_update(state, crtc_state);
> > > }
> > >
> > > -/**
> > > - * psr_wait_for_idle - wait for PSR1 to idle
> > > - * @intel_dp: Intel DP
> > > - * @out_value: PSR status in case of failure
> > > - *
> > > - * Returns: 0 on success or -ETIMEOUT if PSR status does not idle.
> > > - *
> > > - */
> > > -static int psr_wait_for_idle(struct intel_dp *intel_dp, u32 *out_value)
> > > +static bool _is_psr2_read_for_pipe_update(void *data)
> > > +{
> > > + struct intel_dp *intel_dp = data;
> > > + struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
> > > + u32 val;
> > > +
> > > + val = intel_uncore_read_fw(&dev_priv->uncore,
> > > + EDP_PSR2_STATUS(intel_dp->psr.transcoder));
> > > + val &= EDP_PSR2_STATUS_STATE_MASK;
> > > +
> > > + return val == EDP_PSR2_STATUS_STATE_SLEEP || val ==
> > > EDP_PSR2_STATUS_STATE_IDLE;
> > > +}
> > > +
> > > +static int _psr2_ready_for_pipe_update_locked(struct intel_dp *intel_dp)
> > > +{
> > > + struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
> > > + unsigned int fw;
> > > +
> > > + fw = intel_uncore_forcewake_for_reg(&dev_priv->uncore,
> > > +
> > > EDP_PSR2_STATUS(intel_dp->psr.transcoder),
> > > + FW_REG_READ);
> > > + return intel_wait_for_condition(&dev_priv->uncore,
> > > + _is_psr2_read_for_pipe_update,
> > > + intel_dp, fw, 50);
> > > +}
> > > +
> > > +static int _psr1_ready_for_pipe_update_locked(struct intel_dp *intel_dp)
> > > {
> > > struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
> > >
> > > @@ -1814,15 +1832,13 @@ static int psr_wait_for_idle(struct intel_dp
> > > *intel_dp, u32 *out_value)
> > > * exit training time + 1.5 ms of aux channel handshake. 50 ms is
> > > * defensive enough to cover everything.
> > > */
> > > - return __intel_wait_for_register(&dev_priv->uncore,
> > > -
> > > EDP_PSR_STATUS(intel_dp->psr.transcoder),
> > > - EDP_PSR_STATUS_STATE_MASK,
> > > - EDP_PSR_STATUS_STATE_IDLE, 2, 50,
> > > - out_value);
> > > + return intel_de_wait_for_clear(dev_priv,
> > > + EDP_PSR_STATUS(intel_dp->psr.transcoder),
> > > + EDP_PSR_STATUS_STATE_MASK, 50);
> > > }
> > >
> > > /**
> > > - * intel_psr_wait_for_idle - wait for PSR1 to idle
> > > + * intel_psr_wait_for_idle - wait for PSR be ready for a pipe update
> > > * @new_crtc_state: new CRTC state
> > > *
> > > * This function is expected to be called from pipe_update_start() where
> > > it is
> > > @@ -1839,19 +1855,23 @@ void intel_psr_wait_for_idle(const struct
> > > intel_crtc_state *new_crtc_state)
> > > for_each_intel_encoder_mask_with_psr(&dev_priv->drm, encoder,
> > > new_crtc_state->uapi.encoder_mask)
> > > {
> > > struct intel_dp *intel_dp = enc_to_intel_dp(encoder);
> > > - u32 psr_status;
> > > + int ret;
> > >
> > > mutex_lock(&intel_dp->psr.lock);
> > > - if (!intel_dp->psr.enabled || intel_dp->psr.psr2_enabled) {
> > > +
> > > + if (!intel_dp->psr.enabled) {
> > > mutex_unlock(&intel_dp->psr.lock);
> > > continue;
> > > }
> > >
> > > - /* when the PSR1 is enabled */
> > > - if (psr_wait_for_idle(intel_dp, &psr_status))
> > > - drm_err(&dev_priv->drm,
> > > - "PSR idle timed out 0x%x, atomic update may
> > > fail\n",
> > > - psr_status);
> > > + if (intel_dp->psr.psr2_enabled)
> > > + ret = _psr2_ready_for_pipe_update_locked(intel_dp);
> > > + else
> > > + ret = _psr1_ready_for_pipe_update_locked(intel_dp);
> > > +
> > > + if (ret)
> > > + drm_err(&dev_priv->drm, "PSR wait timed out, atomic
> > > update may fail\n");
> > > +
> > > mutex_unlock(&intel_dp->psr.lock);
> > > }
> > > }
> > > diff --git a/drivers/gpu/drm/i915/i915_reg.h
> > > b/drivers/gpu/drm/i915/i915_reg.h
> > > index cad84c3b864bf..a827f5bf973cb 100644
> > > --- a/drivers/gpu/drm/i915/i915_reg.h
> > > +++ b/drivers/gpu/drm/i915/i915_reg.h
> > > @@ -4698,8 +4698,9 @@ enum {
> > > #define _PSR2_STATUS_A 0x60940
> > > #define _PSR2_STATUS_EDP 0x6f940
> > > #define EDP_PSR2_STATUS(tran) _MMIO_TRANS2(tran,
> > > _PSR2_STATUS_A)
> > > -#define EDP_PSR2_STATUS_STATE_MASK (0xf << 28)
> > > -#define EDP_PSR2_STATUS_STATE_SHIFT 28
> > > +#define EDP_PSR2_STATUS_STATE_MASK REG_GENMASK(31, 28)
> > > +#define EDP_PSR2_STATUS_STATE_SLEEP
> > > REG_FIELD_PREP(EDP_PSR2_STATUS_STATE_MASK, 0x3)
> > > +#define EDP_PSR2_STATUS_STATE_IDLE
> > > REG_FIELD_PREP(EDP_PSR2_STATUS_STATE_MASK, 0x0)
> > >
> > > #define _PSR2_SU_STATUS_A 0x60914
> > > #define _PSR2_SU_STATUS_EDP 0x6f914
> > > diff --git a/drivers/gpu/drm/i915/intel_uncore.c
> > > b/drivers/gpu/drm/i915/intel_uncore.c
> > > index 678a99de07fee..1b3ea7318c2d5 100644
> > > --- a/drivers/gpu/drm/i915/intel_uncore.c
> > > +++ b/drivers/gpu/drm/i915/intel_uncore.c
> > > @@ -2383,6 +2383,28 @@ int __intel_wait_for_register_fw(struct
> > > intel_uncore *uncore,
> > > #undef done
> > > }
> > >
> > > +static int __intel_wait_for_condition_fw(bool (*func)(void *data), void
> > > *data,
> > > + unsigned int fast_timeout_us,
> > > + unsigned int slow_timeout_ms)
> > > +{
> > > +#define done (func(data))
> > > + int ret;
> > > +
> > > + /* Catch any overuse of this function */
> > > + might_sleep_if(slow_timeout_ms);
> > > + GEM_BUG_ON(fast_timeout_us > 20000);
> > > + GEM_BUG_ON(!fast_timeout_us && !slow_timeout_ms);
> > > +
> > > + ret = -ETIMEDOUT;
> > > + if (fast_timeout_us && fast_timeout_us <= 20000)
> > > + ret = _wait_for_atomic(done, fast_timeout_us, 0);
> > > + if (ret && slow_timeout_ms)
> > > + ret = wait_for(done, slow_timeout_ms);
> >
> > Is there a particular reason for these complicated wrappers
> > instead of just using wait_for() directly?
>
> Just replicated what __intel_wait_for_register_fw() do.
> I guess the first one sleep less for cases that condition is meet in a few
> usecs.
wait_for() should already start with a short delay with
exponential backoff the longer it has to wait. So unless there
is a demonstrable problem with wait_for() I don't think we
should be adding even more complex wait macros.
--
Ville Syrjälä
Intel
