| Message ID | 20200527200022.28003-1-stanislav.lisovskiy@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v1] drm/i915: Minor link training logic fixes for dp_mst | expand |
On Wed, May 27, 2020 at 11:00:22PM +0300, Stanislav Lisovskiy wrote: > First of all *_needs_link_retraining function should return > false is link_train is set to true but not false. > > Also if we detect channel eq problem when checking mst status > we simply bail out, without setting link_train to false again, > which might end up in a situation that we don't do link retraining > when needed. > > There were some issues, when we had several problems with dp mst > and at the same time the log was floode by messages about > "channel eq not ok, need retraining" however the actual training > seems to be never done. > > Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> > --- > drivers/gpu/drm/i915/display/intel_dp.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c > index 1768731678a1..9288dc1f8914 100644 > --- a/drivers/gpu/drm/i915/display/intel_dp.c > +++ b/drivers/gpu/drm/i915/display/intel_dp.c > @@ -5627,6 +5627,7 @@ intel_dp_check_mst_status(struct intel_dp *intel_dp) > drm_dbg_kms(&i915->drm, > "channel EQ not ok, retraining\n"); > need_retrain = true; > + intel_dp->link_trained = false; > } > > drm_dbg_kms(&i915->drm, "got esi %3ph\n", esi); > @@ -5654,7 +5655,7 @@ intel_dp_needs_link_retrain(struct intel_dp *intel_dp) > { > u8 link_status[DP_LINK_STATUS_SIZE]; > > - if (!intel_dp->link_trained) > + if (intel_dp->link_trained) intel_dp->link_trained is set when we trained the link during a modeset, it doesn't mean that the link status is good, as you seem to interpret it. With this change I don't see how we would retrain the link when this is called from intel_dp_short_pulse(). Could you describe more the failing scenario? > return false; > > /* > -- > 2.24.1.485.gad05a3d8e5 >
On Wed, May 27, 2020 at 11:26:49PM +0300, Imre Deak wrote: > On Wed, May 27, 2020 at 11:00:22PM +0300, Stanislav Lisovskiy wrote: > > First of all *_needs_link_retraining function should return > > false is link_train is set to true but not false. > > > > Also if we detect channel eq problem when checking mst status > > we simply bail out, without setting link_train to false again, > > which might end up in a situation that we don't do link retraining > > when needed. > > > > There were some issues, when we had several problems with dp mst > > and at the same time the log was floode by messages about > > "channel eq not ok, need retraining" however the actual training > > seems to be never done. > > > > Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> > > --- > > drivers/gpu/drm/i915/display/intel_dp.c | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c > > index 1768731678a1..9288dc1f8914 100644 > > --- a/drivers/gpu/drm/i915/display/intel_dp.c > > +++ b/drivers/gpu/drm/i915/display/intel_dp.c > > @@ -5627,6 +5627,7 @@ intel_dp_check_mst_status(struct intel_dp *intel_dp) > > drm_dbg_kms(&i915->drm, > > "channel EQ not ok, retraining\n"); > > need_retrain = true; > > + intel_dp->link_trained = false; > > } > > > > drm_dbg_kms(&i915->drm, "got esi %3ph\n", esi); > > @@ -5654,7 +5655,7 @@ intel_dp_needs_link_retrain(struct intel_dp *intel_dp) > > { > > u8 link_status[DP_LINK_STATUS_SIZE]; > > > > - if (!intel_dp->link_trained) > > + if (intel_dp->link_trained) > > intel_dp->link_trained is set when we trained the link during a modeset, > it doesn't mean that the link status is good, as you seem to interpret > it. With this change I don't see how we would retrain the link when this > is called from intel_dp_short_pulse(). Could you describe more the > failing scenario? One reason we wouldn't retrain when needed is that the sink is not seen as connected in intel_dp_retrain_link(). > > > return false; > > > > /* > > -- > > 2.24.1.485.gad05a3d8e5 > > > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
On Wed, May 27, 2020 at 11:00:22PM +0300, Stanislav Lisovskiy wrote: > First of all *_needs_link_retraining function should return > false is link_train is set to true but not false. > > Also if we detect channel eq problem when checking mst status > we simply bail out, without setting link_train to false again, > which might end up in a situation that we don't do link retraining > when needed. > > There were some issues, when we had several problems with dp mst > and at the same time the log was floode by messages about > "channel eq not ok, need retraining" however the actual training > seems to be never done. > > Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> > --- > drivers/gpu/drm/i915/display/intel_dp.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c > index 1768731678a1..9288dc1f8914 100644 > --- a/drivers/gpu/drm/i915/display/intel_dp.c > +++ b/drivers/gpu/drm/i915/display/intel_dp.c > @@ -5627,6 +5627,7 @@ intel_dp_check_mst_status(struct intel_dp *intel_dp) > drm_dbg_kms(&i915->drm, > "channel EQ not ok, retraining\n"); > need_retrain = true; > + intel_dp->link_trained = false; > } > > drm_dbg_kms(&i915->drm, "got esi %3ph\n", esi); > @@ -5654,7 +5655,7 @@ intel_dp_needs_link_retrain(struct intel_dp *intel_dp) > { > u8 link_status[DP_LINK_STATUS_SIZE]; > > - if (!intel_dp->link_trained) > + if (intel_dp->link_trained) This is not correct. Since link_trained is set when link training is completed as part of a complete modeset. If link training is not done, like at hotplug, then in that case we should not retrain since the pipe has not been configured for this new hotplug and link training has not been done. Retraining is expected to happen only in cases where there is a short pulse or a spurious long pulse when link training through modeset is already complete and hence the old logic of returnin a false when !intel_dp->link_trained is correct. Regards Manasi > return false; > > /* > -- > 2.24.1.485.gad05a3d8e5 >
On Wed, May 27, 2020 at 11:49:55PM +0300, Imre Deak wrote: > On Wed, May 27, 2020 at 11:26:49PM +0300, Imre Deak wrote: > > On Wed, May 27, 2020 at 11:00:22PM +0300, Stanislav Lisovskiy wrote: > > > First of all *_needs_link_retraining function should return > > > false is link_train is set to true but not false. > > > > > > Also if we detect channel eq problem when checking mst status > > > we simply bail out, without setting link_train to false again, > > > which might end up in a situation that we don't do link retraining > > > when needed. > > > > > > There were some issues, when we had several problems with dp mst > > > and at the same time the log was floode by messages about > > > "channel eq not ok, need retraining" however the actual training > > > seems to be never done. > > > > > > Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> > > > --- > > > drivers/gpu/drm/i915/display/intel_dp.c | 3 ++- > > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c > > > index 1768731678a1..9288dc1f8914 100644 > > > --- a/drivers/gpu/drm/i915/display/intel_dp.c > > > +++ b/drivers/gpu/drm/i915/display/intel_dp.c > > > @@ -5627,6 +5627,7 @@ intel_dp_check_mst_status(struct intel_dp *intel_dp) > > > drm_dbg_kms(&i915->drm, > > > "channel EQ not ok, retraining\n"); > > > need_retrain = true; > > > + intel_dp->link_trained = false; > > > } > > > > > > drm_dbg_kms(&i915->drm, "got esi %3ph\n", esi); > > > @@ -5654,7 +5655,7 @@ intel_dp_needs_link_retrain(struct intel_dp *intel_dp) > > > { > > > u8 link_status[DP_LINK_STATUS_SIZE]; > > > > > > - if (!intel_dp->link_trained) > > > + if (intel_dp->link_trained) > > > > intel_dp->link_trained is set when we trained the link during a modeset, > > it doesn't mean that the link status is good, as you seem to interpret > > it. With this change I don't see how we would retrain the link when this > > is called from intel_dp_short_pulse(). Could you describe more the > > failing scenario? > > One reason we wouldn't retrain when needed is that the sink is not seen > as connected in intel_dp_retrain_link(). But mst is always connected. So maybe intel_dp_needs_link_retrain() does an early return because intel_dp_get_link_status() or intel_dp_link_params_valid() fails? > > > > > > return false; > > > > > > /* > > > -- > > > 2.24.1.485.gad05a3d8e5 > > > > > _______________________________________________ > > Intel-gfx mailing list > > Intel-gfx@lists.freedesktop.org > > https://lists.freedesktop.org/mailman/listinfo/intel-gfx > _______________________________________________ > Intel-gfx mailing list > Intel-gfx@lists.freedesktop.org > https://lists.freedesktop.org/mailman/listinfo/intel-gfx
On Wed, May 27, 2020 at 01:54:27PM -0700, Manasi Navare wrote: > On Wed, May 27, 2020 at 11:00:22PM +0300, Stanislav Lisovskiy wrote: > > First of all *_needs_link_retraining function should return > > false is link_train is set to true but not false. > > > > Also if we detect channel eq problem when checking mst status > > we simply bail out, without setting link_train to false again, > > which might end up in a situation that we don't do link retraining > > when needed. > > > > There were some issues, when we had several problems with dp mst > > and at the same time the log was floode by messages about > > "channel eq not ok, need retraining" however the actual training > > seems to be never done. > > > > Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> > > --- > > drivers/gpu/drm/i915/display/intel_dp.c | 3 ++- > > 1 file changed, 2 insertions(+), 1 deletion(-) > > > > diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c > > index 1768731678a1..9288dc1f8914 100644 > > --- a/drivers/gpu/drm/i915/display/intel_dp.c > > +++ b/drivers/gpu/drm/i915/display/intel_dp.c > > @@ -5627,6 +5627,7 @@ intel_dp_check_mst_status(struct intel_dp *intel_dp) > > drm_dbg_kms(&i915->drm, > > "channel EQ not ok, retraining\n"); > > need_retrain = true; > > + intel_dp->link_trained = false; > > } > > > > drm_dbg_kms(&i915->drm, "got esi %3ph\n", esi); > > @@ -5654,7 +5655,7 @@ intel_dp_needs_link_retrain(struct intel_dp *intel_dp) > > { > > u8 link_status[DP_LINK_STATUS_SIZE]; > > > > - if (!intel_dp->link_trained) > > + if (intel_dp->link_trained) > > This is not correct. Since link_trained is set when link training is completed as part of a > complete modeset. If link training is not done, like at hotplug, then in that case we should > not retrain since the pipe has not been configured for this new hotplug and link training > has not been done. > Ok, I was confusing the link training and retraining - assuming those are same procedure. So we can't retrain until the training has been completed. Thanks for clarification - will dig further then. > Retraining is expected to happen only in cases where there is a short pulse or a spurious long pulse > when link training through modeset is already complete and hence the old logic of returnin a false > when !intel_dp->link_trained is correct. Yes, that is the scenario which I was suspecting to be happening in our recent MST failures. We sometimes get a lot of short pulses requiring rettraining however it doesn't happen in practice. Stan > > Regards > Manasi > > > return false; > > > > /* > > -- > > 2.24.1.485.gad05a3d8e5 > >
diff --git a/drivers/gpu/drm/i915/display/intel_dp.c b/drivers/gpu/drm/i915/display/intel_dp.c index 1768731678a1..9288dc1f8914 100644 --- a/drivers/gpu/drm/i915/display/intel_dp.c +++ b/drivers/gpu/drm/i915/display/intel_dp.c @@ -5627,6 +5627,7 @@ intel_dp_check_mst_status(struct intel_dp *intel_dp) drm_dbg_kms(&i915->drm, "channel EQ not ok, retraining\n"); need_retrain = true; + intel_dp->link_trained = false; } drm_dbg_kms(&i915->drm, "got esi %3ph\n", esi); @@ -5654,7 +5655,7 @@ intel_dp_needs_link_retrain(struct intel_dp *intel_dp) { u8 link_status[DP_LINK_STATUS_SIZE]; - if (!intel_dp->link_trained) + if (intel_dp->link_trained) return false; /*
First of all *_needs_link_retraining function should return false is link_train is set to true but not false. Also if we detect channel eq problem when checking mst status we simply bail out, without setting link_train to false again, which might end up in a situation that we don't do link retraining when needed. There were some issues, when we had several problems with dp mst and at the same time the log was floode by messages about "channel eq not ok, need retraining" however the actual training seems to be never done. Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com> --- drivers/gpu/drm/i915/display/intel_dp.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)