On Fri, Nov 15, 2019 at 12:02:00AM +0000, Mihail Atanassov wrote:
> Hi James,
>
> On Thursday, 14 November 2019 08:37:24 GMT james qian wang (Arm Technology
> China) wrote:
> > Komeda HW can support side by side, which splits the internal display
> > processing to two single halves (LEFT/RIGHT) and handle them by two
> > pipelines separately.
> > komeda "side by side" is enabled by DT property: "side_by_side_master",
> > once DT configured side by side, komeda need to verify it with HW's
> > configuration, and assemble it for the further usage.
>
> A few problems I see with this approach:
> - This property doesn't scale to >2 pipes;
> - Our HW is capable of dynamically switching between SBS and non-SBS
> modes, with this DT property you're effectively denying the opportunity
> to use the second pipe when the first one can be satisfied with
> 4 planes and 1px/clk.
>
> If we only want to fix the first problem, then at least we need this
> to be a property of the pipeline node with a phandle linking slave to
> master (or bidirectional).
I had consider this way before, but consider we have no product (now
and in next 2/3 years) can support >2 pipes. So for DT I decide to
focus on current, but you may see I add two side_by_side flags.
- mdev->side_by_side.
- crtc->side_by_side.
And beside the DT parse we use mdev->side_by_side, the real SBS
operation actually based on crtc->side_by_side, then once the HW
changed, we only need to update the SBS assemble/decision code, but no
need to update the real sbs logic.
thanks
James
> For the second, why not do the SBS decision at modeset time?
> If the first CRTC has dual-link output and the commit:
> - only drives one CRTC
> - uses up to 4 planes
> - doesn't meet clk requirements without SBS but does with SBS
> then we can switch SBS on dynamically.
> And we can tweak that decision with power use in mind later on since
> there's no user-visible knob.
Yes, you're right, current implementation just use simplest way to
show the feature, and for dynamic enable/disable sbs will be added
when we have the real usage case.
> We can still keep a DT property if we have a use case for it (e.g.
> forcing SBS on for some reason), but we might want to name it slightly
> more conservatively then, so it doesn't imply that we never do SBS
> when it's not there.
>
> Lastly, maintaining that property in combination with the dynamic
> modeset-time SBS decision tree means extra code for more or less the
> same functionality. <2c>I'm not 100% sure it's worth it.</2c>
I think we'd add such support when we have the real use case. :)
> >
> > v3: Correct a typo.
> >
> > Signed-off-by: James Qian Wang (Arm Technology China)
> > <james.qian.w...@arm.com>
> > ---
> > .../gpu/drm/arm/display/komeda/komeda_crtc.c | 13 ++++-
> > .../gpu/drm/arm/display/komeda/komeda_dev.c | 3 ++
> > .../gpu/drm/arm/display/komeda/komeda_dev.h | 9 ++++
> > .../gpu/drm/arm/display/komeda/komeda_kms.h | 3 ++
> > .../drm/arm/display/komeda/komeda_pipeline.c | 50 +++++++++++++++++--
> > .../drm/arm/display/komeda/komeda_pipeline.h | 1 +
> > 6 files changed, 73 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c
> > b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c
> > index 1c452ea75999..cee9a1692e71 100644
> > --- a/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c
> > +++ b/drivers/gpu/drm/arm/display/komeda/komeda_crtc.c
> > @@ -561,21 +561,30 @@ int komeda_kms_setup_crtcs(struct komeda_kms_dev *kms,
> > kms->n_crtcs = 0;
> >
> > for (i = 0; i < mdev->n_pipelines; i++) {
> > + /* if sbs, one komeda_dev only can represent one CRTC */
> > + if (mdev->side_by_side && i != mdev->side_by_side_master)
> > + continue;
> > +
> > crtc = &kms->crtcs[kms->n_crtcs];
> > master = mdev->pipelines[i];
> >
> > crtc->master = master;
> > crtc->slave = komeda_pipeline_get_slave(master);
> > + crtc->side_by_side = mdev->side_by_side;
> >
> > if (crtc->slave)
> > sprintf(str, "pipe-%d", crtc->slave->id);
> > else
> > sprintf(str, "None");
> >
> > - DRM_INFO("CRTC-%d: master(pipe-%d) slave(%s).\n",
> > - kms->n_crtcs, master->id, str);
> > + DRM_INFO("CRTC-%d: master(pipe-%d) slave(%s) sbs(%s).\n",
> > + kms->n_crtcs, master->id, str,
> > + crtc->side_by_side ? "On" : "Off");
> >
> > kms->n_crtcs++;
> > +
> > + if (mdev->side_by_side)
> > + break;
> > }
> >
> > return 0;
> > diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_dev.c
> > b/drivers/gpu/drm/arm/display/komeda/komeda_dev.c
> > index 4e46f650fddf..c3fa4835cb8d 100644
> > --- a/drivers/gpu/drm/arm/display/komeda/komeda_dev.c
> > +++ b/drivers/gpu/drm/arm/display/komeda/komeda_dev.c
> > @@ -178,6 +178,9 @@ static int komeda_parse_dt(struct device *dev, struct
> > komeda_dev *mdev)
> > }
> > }
> >
> > + mdev->side_by_side = !of_property_read_u32(np, "side_by_side_master",
> > + &mdev->side_by_side_master);
> > +
> > return ret;
> > }
> >
> > diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_dev.h
> > b/drivers/gpu/drm/arm/display/komeda/komeda_dev.h
> > index d406a4d83352..471604b42431 100644
> > --- a/drivers/gpu/drm/arm/display/komeda/komeda_dev.h
> > +++ b/drivers/gpu/drm/arm/display/komeda/komeda_dev.h
> > @@ -183,6 +183,15 @@ struct komeda_dev {
> >
> > /** @irq: irq number */
> > int irq;
> > + /**
> > + * @side_by_side:
> > + *
> > + * on sbs the whole display frame will be split to two halves (1:2),
> > + * master pipeline handles the left part, slave for the right part
> > + */
> > + bool side_by_side;
>
> That's a duplicate of the one in komeda_crtc. You don't need both.
>
> > + /** @side_by_side_master: master pipe id for side by side */
> > + int side_by_side_master;
>
> As I detailed above, this should be on the crtc, otherwise we can't
> scale to >2 pipes.
>
> >
> > /** @lock: used to protect dpmode */
> > struct mutex lock;
> > diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_kms.h
> > b/drivers/gpu/drm/arm/display/komeda/komeda_kms.h
> > index 456f3c435719..ae6654fe95e2 100644
> > --- a/drivers/gpu/drm/arm/display/komeda/komeda_kms.h
> > +++ b/drivers/gpu/drm/arm/display/komeda/komeda_kms.h
> > @@ -76,6 +76,9 @@ struct komeda_crtc {
> > */
> > struct komeda_pipeline *slave;
> >
> > + /** @side_by_side: if the master and slave works on side by side mode */
> > + bool side_by_side;
> > +
> > /** @slave_planes: komeda slave planes mask */
> > u32 slave_planes;
> >
> > diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c
> > b/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c
> > index 452e505a1fd3..104e27cc1dc3 100644
> > --- a/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c
> > +++ b/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.c
> > @@ -326,14 +326,56 @@ static void komeda_pipeline_assemble(struct
> > komeda_pipeline *pipe)
> > struct komeda_pipeline *
> > komeda_pipeline_get_slave(struct komeda_pipeline *master)
> > {
> > - struct komeda_component *slave;
> > + struct komeda_dev *mdev = master->mdev;
> > + struct komeda_component *comp, *slave;
> > + u32 avail_inputs;
> > +
> > + /* on SBS, slave pipeline merge to master via image processor */
> > + if (mdev->side_by_side) {
> > + comp = &master->improc->base;
> > + avail_inputs = KOMEDA_PIPELINE_IMPROCS;
> > + } else {
> > + comp = &master->compiz->base;
> > + avail_inputs = KOMEDA_PIPELINE_COMPIZS;
> > + }
> >
> > - slave = komeda_component_pickup_input(&master->compiz->base,
> > - KOMEDA_PIPELINE_COMPIZS);
> > + slave = komeda_component_pickup_input(comp, avail_inputs);
> >
> > return slave ? slave->pipeline : NULL;
> > }
> >
> > +static int komeda_assemble_side_by_side(struct komeda_dev *mdev)
> > +{
> > + struct komeda_pipeline *master, *slave;
> > + int i;
> > +
> > + if (!mdev->side_by_side)
> > + return 0;
> > +
> > + if (mdev->side_by_side_master >= mdev->n_pipelines) {
> > + DRM_ERROR("DT configured side by side master-%d is invalid.\n",
> > + mdev->side_by_side_master);
> > + return -EINVAL;
> > + }
> > +
> > + master = mdev->pipelines[mdev->side_by_side_master];
> > + slave = komeda_pipeline_get_slave(master);
> > + if (!slave || slave->n_layers != master->n_layers) {
> > + DRM_ERROR("Current HW doesn't support side by side.\n");
> > + return -EINVAL;
> > + }
> > +
> > + if (!master->dual_link) {
> > + DRM_DEBUG_ATOMIC("SBS can not work without dual link.\n");
> > + return -EINVAL;
> > + }
> > +
> > + for (i = 0; i < master->n_layers; i++)
> > + master->layers[i]->sbs_slave = slave->layers[i];
> > +
> > + return 0;
> > +}
> > +
> > int komeda_assemble_pipelines(struct komeda_dev *mdev)
> > {
> > struct komeda_pipeline *pipe;
> > @@ -346,7 +388,7 @@ int komeda_assemble_pipelines(struct komeda_dev *mdev)
> > komeda_pipeline_dump(pipe);
> > }
> >
> > - return 0;
> > + return komeda_assemble_side_by_side(mdev);
> > }
> >
> > void komeda_pipeline_dump_register(struct komeda_pipeline *pipe,
> > diff --git a/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.h
> > b/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.h
> > index ac8725e24853..20a076cce635 100644
> > --- a/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.h
> > +++ b/drivers/gpu/drm/arm/display/komeda/komeda_pipeline.h
> > @@ -237,6 +237,7 @@ struct komeda_layer {
> > * not the source buffer.
> > */
> > struct komeda_layer *right;
> > + struct komeda_layer *sbs_slave;
> > };
> >
> > struct komeda_layer_state {
> >
>
>
> --
> Mihail
>
>
_______________________________________________
dri-devel mailing list
dri-devel@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/dri-devel
