On Wed, 2020-06-03 at 23:33 +0300, Ville Syrjälä wrote:
> On Wed, Jun 03, 2020 at 12:43:07PM -0700, José Roberto de Souza wrote:
> > Hours Of Battery Life is a new GEN12+ power-saving feature that allows
> > supported motherboards to use a special voltage swing table for eDP
> > panels that uses less power.
> >
> > So here if supported by HW, OEM will set it in VBT and i915 will try
> > to train link with HOBL vswing table if link training fails it fall
> > back to the original table.
> >
> > Just not sure if DP compliance should also use this new voltage swing
> > table too, cced some folks that worked in DP compliance.
> >
> > BSpec: 49291
> > BSpec: 49399
> > Cc: Ville Syrjälä <ville.syrjala@xxxxxxxxxxxxxxx>
> > Cc: Animesh Manna <animesh.manna@xxxxxxxxx>
> > Cc: Manasi Navare <manasi.d.navare@xxxxxxxxx>
> > Signed-off-by: José Roberto de Souza <jose.souza@xxxxxxxxx>
> > ---
> > drivers/gpu/drm/i915/display/intel_ddi.c | 48 +++++++++++++++++--
> > .../drm/i915/display/intel_display_types.h | 2 +
> > .../drm/i915/display/intel_dp_link_training.c | 20 +++++++-
> > drivers/gpu/drm/i915/i915_drv.h | 2 +
> > drivers/gpu/drm/i915/i915_reg.h | 2 +
> > 5 files changed, 69 insertions(+), 5 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/i915/display/intel_ddi.c b/drivers/gpu/drm/i915/display/intel_ddi.c
> > index 236f3762b6f9..57174a111976 100644
> > --- a/drivers/gpu/drm/i915/display/intel_ddi.c
> > +++ b/drivers/gpu/drm/i915/display/intel_ddi.c
> > @@ -692,6 +692,10 @@ static const struct cnl_ddi_buf_trans tgl_combo_phy_ddi_translations_dp_hbr2[] =
> > { 0x6, 0x7F, 0x3F, 0x00, 0x00 }, /* 900 900 0.0 */
> > };
> >
> > +static const struct cnl_ddi_buf_trans tgl_combo_phy_ddi_translations_edp_hbr2_hobl[] = {
> > + { 0x6, 0x7F, 0x3F, 0x00, 0x00 }
> > +};
> > +
> > static const struct ddi_buf_trans *
> > bdw_get_buf_trans_edp(struct drm_i915_private *dev_priv, int *n_entries)
> > {
> > @@ -2301,14 +2305,51 @@ static void cnl_ddi_vswing_sequence(struct intel_encoder *encoder,
> > intel_de_write(dev_priv, CNL_PORT_TX_DW5_GRP(port), val);
> > }
> >
> > +/*
> > + * If supported return HOBL vswing table and set registers to enable HOBL
> > + * otherwise returns NULL and unset registers to enable HOBL.
> > + */
> > +static const struct cnl_ddi_buf_trans *
> > +hobl_get_combo_buf_trans(struct drm_i915_private *dev_priv,
> > + struct intel_encoder *encoder, int type, int rate,
> > + u32 level, int *n_entries)
> > +{
> > + const u32 hobl_en = EDP4K2K_MODE_OVRD_EN | EDP4K2K_MODE_OVRD_OPTIMIZED;
> > + enum phy phy = intel_port_to_phy(dev_priv, encoder->port);
> > + struct intel_dp *intel_dp;
> > +
> > + if (!HAS_HOBL(dev_priv) || type != INTEL_OUTPUT_EDP)
> > + return NULL;
>
> Not a real fan of the "hobl" name. It just sounds like nonsense. Also
> bspec doesn't use that term at all. It only appears in the vbt spec.
> Not sure if there's a better one though.
Maybe power_optimized_edp?
>
> > +
> > + intel_dp = enc_to_intel_dp(encoder);
> > + if (!intel_dp->try_hobl || rate > 540000) {
> > + intel_de_rmw(dev_priv, ICL_PORT_CL_DW10(phy), hobl_en, 0);
>
> I would vote for just doing this programming unconditionally in the normal
> sequence.
Thought about that but intel_combo_phy_power_up_lanes() that program this ICL_PORT_CL_DW10 is called right after tgl_ddi_vswing_sequence().
>
> > + return NULL;
> > + }
> > +
> > + drm_dbg_kms(&dev_priv->drm, "Enabling HOBL in PHY %c\n", phy_name(phy));
> > + drm_WARN_ON_ONCE(&dev_priv->drm, level > 0);
> > +
> > + intel_de_rmw(dev_priv, ICL_PORT_CL_DW10(phy), hobl_en, hobl_en);
> > + /* Same table applies to TGL, RKL and DG1 */
> > + *n_entries = ARRAY_SIZE(tgl_combo_phy_ddi_translations_edp_hbr2_hobl);
> > + return tgl_combo_phy_ddi_translations_edp_hbr2_hobl;
> > +}
> > +
> > static void icl_ddi_combo_vswing_program(struct drm_i915_private *dev_priv,
> > - u32 level, enum phy phy, int type,
> > - int rate)
> > + struct intel_encoder *encoder,
> > + u32 level, enum phy phy, int type,
> > + int rate)
>
> If we're passing in the encoder then a bunch of this other stuff is
> redundant.
Okay
>
> > {
> > const struct cnl_ddi_buf_trans *ddi_translations = NULL;
> > u32 n_entries, val;
> > int ln;
> >
> > + ddi_translations = hobl_get_combo_buf_trans(dev_priv, encoder, type,
> > + rate, level, &n_entries);
> > + if (ddi_translations)
> > + goto hobl_found;
>
> Why not just put it into tgl_get_combo_buf_trans(). Hmm. I guess to not
> upset .voltage_max(). This feels a bit hackish, but I don't have better
> ideas for now.
Exactly.
>
> > +
> > if (INTEL_GEN(dev_priv) >= 12)
> > ddi_translations = tgl_get_combo_buf_trans(dev_priv, type, rate,
> > &n_entries);
> > @@ -2321,6 +2362,7 @@ static void icl_ddi_combo_vswing_program(struct drm_i915_private *dev_priv,
> > if (!ddi_translations)
> > return;
> >
> > +hobl_found:
> > if (level >= n_entries) {
> > drm_dbg_kms(&dev_priv->drm,
> > "DDI translation not found for level %d. Using %d instead.",
> > @@ -2428,7 +2470,7 @@ static void icl_combo_phy_ddi_vswing_sequence(struct intel_encoder *encoder,
> > intel_de_write(dev_priv, ICL_PORT_TX_DW5_GRP(phy), val);
> >
> > /* 5. Program swing and de-emphasis */
> > - icl_ddi_combo_vswing_program(dev_priv, level, phy, type, rate);
> > + icl_ddi_combo_vswing_program(dev_priv, encoder, level, phy, type, rate);
> >
> > /* 6. Set training enable to trigger update */
> > val = intel_de_read(dev_priv, ICL_PORT_TX_DW5_LN0(phy));
> > diff --git a/drivers/gpu/drm/i915/display/intel_display_types.h b/drivers/gpu/drm/i915/display/intel_display_types.h
> > index 4b0aaa3081c9..f8943b67819d 100644
> > --- a/drivers/gpu/drm/i915/display/intel_display_types.h
> > +++ b/drivers/gpu/drm/i915/display/intel_display_types.h
> > @@ -1375,6 +1375,8 @@ struct intel_dp {
> >
> > /* Display stream compression testing */
> > bool force_dsc_en;
> > +
> > + bool try_hobl;
> > };
> >
> > enum lspcon_vendor {
> > diff --git a/drivers/gpu/drm/i915/display/intel_dp_link_training.c b/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> > index b9e4ee2dbddc..88f366bb28d7 100644
> > --- a/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> > +++ b/drivers/gpu/drm/i915/display/intel_dp_link_training.c
> > @@ -52,12 +52,24 @@ static u8 dp_voltage_max(u8 preemph)
> > void intel_dp_get_adjust_train(struct intel_dp *intel_dp,
> > const u8 link_status[DP_LINK_STATUS_SIZE])
> > {
> > + struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
> > u8 v = 0;
> > u8 p = 0;
> > int lane;
> > u8 voltage_max;
> > u8 preemph_max;
> >
> > + if (intel_dp->try_hobl) {
> > + /*
> > + * Do not adjust, try now with the regular table using VSwing 0
> > + * and pre-emp 0
> > + */
>
> What if the sink is still asking for vswing 0 + preemph 0? The spec is
> rather ambiguous when it comes to this stuff.
As it will fallback to regular table vswing 0 + preemph 0 that is not a issue.
>
> The table also doesn't specify the vswing/preemph for which we should
> use this optimized value. Your interpretation of 0+0 seems like the most
> sensible thing, but given that the VBT can also specifiy the fast link
> training vswing/preemph as something else (and maybe there was also
> something like this for normal link training?) I'm not 100% sure.
Yeah don't make much sense it not be vswing 0 + preemph 0 but lets wait for BSpec clarification then.
>
> Hmm. Actually noticed that all the eDP tables are missing the
> vswing/preemph levels (they do have the raw mV/dB values but not the
> DP spec levels). I filed a few issues in the hopes of clarification.
>
> > + intel_dp->try_hobl = false;
> > + drm_dbg_kms(&dev_priv->drm, "HOBL vswing table failed link "
> > + "training, switching back to regular table\n");
> > + return;
> > + }
> > +
> > for (lane = 0; lane < intel_dp->lane_count; lane++) {
> > v = max(v, drm_dp_get_adjust_request_voltage(link_status, lane));
> > p = max(p, drm_dp_get_adjust_request_pre_emphasis(link_status, lane));
> > @@ -103,9 +115,13 @@ intel_dp_set_link_train(struct intel_dp *intel_dp,
> > }
> >
> > static bool
> > -intel_dp_reset_link_train(struct intel_dp *intel_dp,
> > - u8 dp_train_pat)
> > +intel_dp_reset_link_train(struct intel_dp *intel_dp, u8 dp_train_pat)
> > {
> > + struct drm_i915_private *dev_priv = dp_to_i915(intel_dp);
> > +
> > + if (intel_dp_is_edp(intel_dp) && dev_priv->vbt.edp.hobl)
> > + intel_dp->try_hobl = true;
>
> If it failed once does it make sense to keep trying to use it?
It could pass in a different bit rate and would be to much complicated keep track of that.
Thanks for the review, lets wait for the BSpec clarifications that you asked.
>
> > +
> > memset(intel_dp->train_set, 0, sizeof(intel_dp->train_set));
> > intel_dp_set_signal_levels(intel_dp);
> > return intel_dp_set_link_train(intel_dp, dp_train_pat);
> > diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
> > index 2336c9231eef..c7e7df17eef2 100644
> > --- a/drivers/gpu/drm/i915/i915_drv.h
> > +++ b/drivers/gpu/drm/i915/i915_drv.h
> > @@ -1687,6 +1687,8 @@ IS_SUBPLATFORM(const struct drm_i915_private *i915,
> > #define INTEL_DISPLAY_ENABLED(dev_priv) \
> > (drm_WARN_ON(&(dev_priv)->drm, !HAS_DISPLAY(dev_priv)), !i915_modparams.disable_display)
> >
> > +#define HAS_HOBL(dev_priv) (INTEL_GEN(dev_priv) >= 12)
> > +
> > static inline bool intel_vtd_active(void)
> > {
> > #ifdef CONFIG_INTEL_IOMMU
> > diff --git a/drivers/gpu/drm/i915/i915_reg.h b/drivers/gpu/drm/i915/i915_reg.h
> > index 578cfe11cbb9..d4611171f075 100644
> > --- a/drivers/gpu/drm/i915/i915_reg.h
> > +++ b/drivers/gpu/drm/i915/i915_reg.h
> > @@ -1896,6 +1896,8 @@ static inline bool i915_mmio_reg_valid(i915_reg_t reg)
> > #define PWR_DOWN_LN_3_1_0 (0xb << 4)
> > #define PWR_DOWN_LN_MASK (0xf << 4)
> > #define PWR_DOWN_LN_SHIFT 4
> > +#define EDP4K2K_MODE_OVRD_EN (1 << 3)
> > +#define EDP4K2K_MODE_OVRD_OPTIMIZED (1 << 2)
> >
> > #define ICL_PORT_CL_DW12(phy) _MMIO(_ICL_PORT_CL_DW(12, phy))
> > #define ICL_LANE_ENABLE_AUX (1 << 0)
> > --
> > 2.27.0
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
