On Mon, Nov 06, 2023 at 11:02:32PM +0200, Imre Deak wrote:
> On Mon, Nov 06, 2023 at 10:39:25PM +0200, Ville Syrjälä wrote:
> > On Mon, Oct 30, 2023 at 05:58:26PM +0200, Imre Deak wrote:
> > > Atm, the BW allocated for an MST stream doesn't take into account the
> > > DSC control symbol (EOC) and data alignment overhead on the local (first
> > > downstream) MST link (reflected by the data M/N/TU values) and - besides
> > > the above overheads - the FEC symbol overhead on 8b/10b remote
> > > (after a downstream branch device) MST links.
> > >
> > > In addition the FEC overhead used on the local link is a fixed amount,
> > > which only applies to certain modes, but not enough for all modes; add a
> > > code comment clarifying this.
> > >
> > > Fix the above by calculating the data M/N values with the total BW
> > > overhead (not including the SSC overhead, since this isn't enabled by
> > > the source device) and using this the PBN and TU values for the local
> > > link and PBN for remote links (including SSC, since this is mandatory
> > > for links after downstream branch devices).
> > >
> > > For now keep the current fixed FEC overhead as a minimum, since this is
> > > what bspec requires for audio functionality.
> > >
> > > Calculate the effective link BW in a clearer way, applying the channel
> > > coding efficiency based on the coding type. The calculation was correct
> > > for 8b/10b, but not for 128b/132b links; this patch leaves the behavior
> > > for this unchanged, leaving the fix for a follow-up.
> > >
> > > v2:
> > > - Fix TU size programmed to the HW, making it match the payload size
> > > programmed to the payload table.
> > >
> > > Reviewed-by: Stanislav Lisovskiy <stanislav.lisovs...@intel.com> (v1)
> > > Signed-off-by: Imre Deak <imre.d...@intel.com>
> > > ---
> > > drivers/gpu/drm/i915/display/intel_dp_mst.c | 107 +++++++++++++++-----
> > > 1 file changed, 82 insertions(+), 25 deletions(-)
> > >
> > > diff --git a/drivers/gpu/drm/i915/display/intel_dp_mst.c
> > > b/drivers/gpu/drm/i915/display/intel_dp_mst.c
> > > index dcbc5d3aa7bc3..05b2d5d547c85 100644
> > > --- a/drivers/gpu/drm/i915/display/intel_dp_mst.c
> > > +++ b/drivers/gpu/drm/i915/display/intel_dp_mst.c
> > > @@ -66,6 +66,63 @@ static int intel_dp_mst_check_constraints(struct
> > > drm_i915_private *i915, int bpp
> > > return 0;
> > > }
> > >
> > > +static int intel_dp_mst_bw_overhead(const struct intel_crtc_state
> > > *crtc_state,
> > > + const struct intel_connector *connector,
> > > + bool ssc, bool dsc, int bpp)
> > > +{
> > > + const struct drm_display_mode *adjusted_mode =
> > > + &crtc_state->hw.adjusted_mode;
> > > + unsigned long flags = DRM_DP_BW_OVERHEAD_MST;
> > > + int dsc_slice_count = 0;
> > > + int overhead;
> > > +
> > > + flags |= intel_dp_is_uhbr(crtc_state) ? DRM_DP_BW_OVERHEAD_UHBR : 0;
> > > + flags |= ssc ? DRM_DP_BW_OVERHEAD_SSC_REF_CLK : 0;
> > > + flags |= crtc_state->fec_enable ? DRM_DP_BW_OVERHEAD_FEC : 0;
> > > +
> > > + if (dsc) {
> > > + flags |= DRM_DP_BW_OVERHEAD_DSC;
> > > + /* TODO: add support for bigjoiner */
> > > + dsc_slice_count = intel_dp_dsc_get_slice_count(connector,
> > > +
> > > adjusted_mode->clock,
> > > +
> > > adjusted_mode->hdisplay,
> > > + false);
> > > + }
> > > +
> > > + overhead = drm_dp_bw_overhead(crtc_state->lane_count,
> > > + adjusted_mode->hdisplay,
> > > + dsc_slice_count,
> > > + to_bpp_x16(bpp),
> > > + flags);
> > > +
> > > + /*
> > > + * TODO: clarify whether a minimum required by the fixed FEC overhead
> > > + * in the bspec audio programming sequence is required here.
> > > + */
> > > + return max(overhead, intel_dp_bw_fec_overhead(crtc_state->fec_enable));
> > > +}
> > > +
> > > +static void intel_dp_mst_compute_m_n(const struct intel_crtc_state
> > > *crtc_state,
> > > + const struct intel_connector *connector,
> > > + bool ssc, bool dsc,
> > > + int bpp,
> > > + struct intel_link_m_n *m_n)
> > > +{
> > > + const struct drm_display_mode *adjusted_mode =
> > > + &crtc_state->hw.adjusted_mode;
> > > + int overhead = intel_dp_mst_bw_overhead(crtc_state,
> > > + connector,
> > > + ssc, dsc, bpp);
> > > +
> > > + intel_link_compute_m_n(bpp, crtc_state->lane_count,
> > > + adjusted_mode->crtc_clock,
> > > + crtc_state->port_clock,
> > > + overhead,
> > > + m_n);
> > > +
> > > + m_n->tu = DIV_ROUND_UP_ULL(mul_u32_u32(m_n->data_m, 64), m_n->data_n);
> > > +}
> > > +
> > > static int intel_dp_mst_find_vcpi_slots_for_bpp(struct intel_encoder
> > > *encoder,
> > > struct intel_crtc_state
> > > *crtc_state,
> > > int max_bpp,
> > > @@ -106,14 +163,34 @@ static int
> > > intel_dp_mst_find_vcpi_slots_for_bpp(struct intel_encoder *encoder,
> > > crtc_state->lane_count);
> > >
> > > for (bpp = max_bpp; bpp >= min_bpp; bpp -= step) {
> > > + struct intel_link_m_n remote_m_n;
> > > + int link_bpp;
> > > +
> > > drm_dbg_kms(&i915->drm, "Trying bpp %d\n", bpp);
> > >
> > > ret = intel_dp_mst_check_constraints(i915, bpp, adjusted_mode,
> > > crtc_state, dsc);
> > > if (ret)
> > > continue;
> > >
> > > - crtc_state->pbn =
> > > drm_dp_calc_pbn_mode(adjusted_mode->crtc_clock,
> > > - bpp << 4);
> > > + link_bpp = dsc ? bpp :
> > > + intel_dp_output_bpp(crtc_state->output_format, bpp);
> > > +
> > > + intel_dp_mst_compute_m_n(crtc_state, connector, false, dsc,
> > > link_bpp,
> > > + &crtc_state->dp_m_n);
> > > + intel_dp_mst_compute_m_n(crtc_state, connector, true, dsc,
> > > link_bpp,
> > > + &remote_m_n);
> > > +
> > > + /*
> > > + * The TU size programmed to the HW determines which slots in
> > > + * an MTP frame are used for this stream, which needs to match
> > > + * the payload size programmed to the first downstream branch
> > > + * device's payload table.
> > > + */
> > > + crtc_state->dp_m_n.tu = remote_m_n.tu;
> >
> > The fact we use the "remote" value here is because the mst manager code
> > assumes the two numbers are the same, right? Should perhaps highlight
> > that fact a bit better.
>
> Yes, mst core assumes atm that the allocation on the MST link between
> the source and the first downstream branch device - the payload size
> programmed with direct AUX writes to this branch device's payload table
> - and the corresponding allocation for farther links - the PBN contained
> in the ALLOCATE_PAYLOAD side-band message - are the same (except for
> units). The former allocation could be reduced for a source not
> enabling SSC.
>
> > Maybe we want a WARN_ON(remote_m_n.tu < dp_m_n.tu) here as well?
>
> Ok, makes sense.
>
> > > +
> > > + crtc_state->pbn =
> > > DIV_ROUND_UP_ULL(mul_u32_u32(mst_state->pbn_div * 64,
> > > +
> > > remote_m_n.data_m),
> > > + remote_m_n.data_n);
>
> The above can be actually simplified to
> crtc_state->pbn = crtc_state->dp_m_n.tu * mst_state->pbn_div;
I was wondering if it's safer to stick to remote_n_m here, in case
we manage to untangle the two values and then somehow forget to
change this. But I suppose we'd have to track both values anyway
somewhere, so maybe that's just nonsense.
>
> > >
> > > slots = drm_dp_atomic_find_time_slots(state, &intel_dp->mst_mgr,
> > > connector->port,
> > > @@ -122,6 +199,8 @@ static int
> > > intel_dp_mst_find_vcpi_slots_for_bpp(struct intel_encoder *encoder,
> > > return slots;
> > >
> > > if (slots >= 0) {
> > > + drm_WARN_ON(&i915->drm, slots != remote_m_n.tu);
> > > +
> > > ret = drm_dp_mst_atomic_check(state);
> > > /*
> > > * If we got slots >= 0 and we can fit those based on
> > > check
> > > @@ -155,10 +234,7 @@ static int intel_dp_mst_compute_link_config(struct
> > > intel_encoder *encoder,
> > > struct drm_connector_state
> > > *conn_state,
> > > struct link_config_limits *limits)
> > > {
> > > - const struct drm_display_mode *adjusted_mode =
> > > - &crtc_state->hw.adjusted_mode;
> > > int slots = -EINVAL;
> > > - int link_bpp;
> > >
> > > /*
> > > * FIXME: allocate the BW according to link_bpp, which in the case of
> > > @@ -173,16 +249,6 @@ static int intel_dp_mst_compute_link_config(struct
> > > intel_encoder *encoder,
> > > if (slots < 0)
> > > return slots;
> > >
> > > - link_bpp = intel_dp_output_bpp(crtc_state->output_format,
> > > crtc_state->pipe_bpp);
> > > -
> > > - intel_link_compute_m_n(link_bpp,
> > > - crtc_state->lane_count,
> > > - adjusted_mode->crtc_clock,
> > > - crtc_state->port_clock,
> > > - intel_dp_bw_fec_overhead(crtc_state->fec_enable),
> > > - &crtc_state->dp_m_n);
> > > - crtc_state->dp_m_n.tu = slots;
> > > -
> > > return 0;
> > > }
> > >
> > > @@ -194,8 +260,6 @@ static int
> > > intel_dp_dsc_mst_compute_link_config(struct intel_encoder *encoder,
> > > struct intel_connector *connector =
> > > to_intel_connector(conn_state->connector);
> > > struct drm_i915_private *i915 = to_i915(connector->base.dev);
> > > - const struct drm_display_mode *adjusted_mode =
> > > - &crtc_state->hw.adjusted_mode;
> > > int slots = -EINVAL;
> > > int i, num_bpc;
> > > u8 dsc_bpc[3] = {};
> > > @@ -270,14 +334,6 @@ static int
> > > intel_dp_dsc_mst_compute_link_config(struct intel_encoder *encoder,
> > > return slots;
> > > }
> > >
> > > - intel_link_compute_m_n(crtc_state->dsc.compressed_bpp,
> > > - crtc_state->lane_count,
> > > - adjusted_mode->crtc_clock,
> > > - crtc_state->port_clock,
> > > - intel_dp_bw_fec_overhead(crtc_state->fec_enable),
> > > - &crtc_state->dp_m_n);
> > > - crtc_state->dp_m_n.tu = slots;
> > > -
> > > return 0;
> > > }
> > > static int intel_dp_mst_update_slots(struct intel_encoder *encoder,
> > > @@ -980,6 +1036,7 @@ intel_dp_mst_mode_valid_ctx(struct drm_connector
> > > *connector,
> > > if (ret)
> > > return ret;
> > >
> > > + /* TODO: also check if compression would allow for the mode */
> > > if (mode_rate > max_rate || mode->clock > max_dotclk ||
> > > drm_dp_calc_pbn_mode(mode->clock, min_bpp << 4) > port->full_pbn) {
> > > *status = MODE_CLOCK_HIGH;
> > > --
> > > 2.39.2
> >
> > --
> > Ville Syrjälä
> > Intel
--
Ville Syrjälä
Intel
