Re: [PATCH v4 2/3] drm/rockchip: Add optional support for CRTC gamma LUT

From: Ezequiel Garcia
Date: Wed Oct 09 2019 - 20:46:08 EST


Hello Sean,

Thanks for the thourough review.

On Wed, 9 Oct 2019 at 15:01, Sean Paul <sean@xxxxxxxxxx> wrote:
>
> On Tue, Oct 08, 2019 at 08:00:37PM -0300, Ezequiel Garcia wrote:
> > Add an optional CRTC gamma LUT support, and enable it on RK3288.
> > This is currently enabled via a separate address resource,
> > which needs to be specified in the devicetree.
> >
> > The address resource is required because on some SoCs, such as
> > RK3288, the LUT address is after the MMU address, and the latter
> > is supported by a different driver. This prevents the DRM driver
> > from requesting an entire register space.
> >
> > The current implementation works for RGB 10-bit tables, as that
> > is what seems to work on RK3288.
> >
> > Signed-off-by: Ezequiel Garcia <ezequiel@xxxxxxxxxxxxx>
>
> Hey Ezequiel,
> Just a few comments on the actual content of the patch as opposed to my higher
> level comments yesterday. I think we're almost there, thanks for sticking this
> out!
>
> > ---
> > Changes from v3:
> > * Move to atomic_enable and atomic_begin,
> > as discussed with Sean Paul.
> > * Dropped the Reviewed-bys.
> > Changes from v2:
> > * None.
> > Changes from v1:
> > * drop explicit linear LUT after finding a proper
> > way to disable gamma correction.
> > * avoid setting gamma is the CRTC is not active.
> > * s/int/unsigned int as suggested by Jacopo.
> > * only enable color management and set gamma size
> > if gamma LUT is supported, suggested by Doug.
> > * drop the reg-names usage, and instead just use indexed reg
> > specifiers, suggested by Doug.
> > Changes from RFC:
> > * Request (an optional) address resource for the LUT.
> > * Drop support for RK3399, which doesn't seem to work
> > out of the box and needs more research.
> > * Support pass-thru setting when GAMMA_LUT is NULL.
> > * Add a check for the gamma size, as suggested by Ilia.
> > * Move gamma setting to atomic_commit_tail, as pointed
> > out by Jacopo/Laurent, is the correct way.
> > ---
> > drivers/gpu/drm/rockchip/rockchip_drm_fb.c | 1 +
> > drivers/gpu/drm/rockchip/rockchip_drm_vop.c | 125 ++++++++++++++++++++
> > drivers/gpu/drm/rockchip/rockchip_drm_vop.h | 5 +
> > drivers/gpu/drm/rockchip/rockchip_vop_reg.c | 2 +
> > 4 files changed, 133 insertions(+)
> >
> > diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_fb.c b/drivers/gpu/drm/rockchip/rockchip_drm_fb.c
> > index ca01234c037c..697ee04b85cf 100644
> > --- a/drivers/gpu/drm/rockchip/rockchip_drm_fb.c
> > +++ b/drivers/gpu/drm/rockchip/rockchip_drm_fb.c
> > @@ -17,6 +17,7 @@
> > #include "rockchip_drm_drv.h"
> > #include "rockchip_drm_fb.h"
> > #include "rockchip_drm_gem.h"
> > +#include "rockchip_drm_vop.h"
>
> Leftover from the previous version?
>

Yup.

> >
> > static const struct drm_framebuffer_funcs rockchip_drm_fb_funcs = {
> > .destroy = drm_gem_fb_destroy,
> > diff --git a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
> > index 613404f86668..85c1269a1218 100644
> > --- a/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
> > +++ b/drivers/gpu/drm/rockchip/rockchip_drm_vop.c
> > @@ -139,6 +139,7 @@ struct vop {
> >
> > uint32_t *regsbak;
> > void __iomem *regs;
> > + void __iomem *lut_regs;
> >
> > /* physical map length of vop register */
> > uint32_t len;
> > @@ -1048,6 +1049,84 @@ static bool vop_crtc_mode_fixup(struct drm_crtc *crtc,
> > return true;
> > }
> >
> > +static bool vop_dsp_lut_is_enable(struct vop *vop)
>
> *enabled
>

Good catch.

> > +{
> > + return vop_read_reg(vop, 0, &vop->data->common->dsp_lut_en);
> > +}
> > +
> > +static void vop_crtc_write_gamma_lut(struct vop *vop, struct drm_crtc *crtc)
> > +{
> > + struct drm_color_lut *lut = crtc->state->gamma_lut->data;
> > + unsigned int i;
> > +
> > + for (i = 0; i < crtc->gamma_size; i++) {
> > + u32 word;
> > +
> > + word = (drm_color_lut_extract(lut[i].red, 10) << 20) |
> > + (drm_color_lut_extract(lut[i].green, 10) << 10) |
> > + drm_color_lut_extract(lut[i].blue, 10);
> > + writel(word, vop->lut_regs + i * 4);
> > + }
> > +}
> > +
> > +static void vop_crtc_gamma_set(struct vop *vop, struct drm_crtc *crtc,
> > + struct drm_crtc_state *old_crtc_state)
> > +{
> > + unsigned int idle;
> > + int ret;
> > +
>
> How about:
>
> if (!vop->lut_regs)
> return;
>
> here and then you can remove that condition above the 2 callsites
>

Yes, sounds good.

> > + /*
> > + * In order to write the LUT to the internal memory,
> > + * we need to first make sure the dsp_lut_en bit is cleared.
> > + */
> > + spin_lock(&vop->reg_lock);
> > + VOP_REG_SET(vop, common, dsp_lut_en, 0);
> > + vop_cfg_done(vop);
> > + spin_unlock(&vop->reg_lock);
> > +
> > + /*
> > + * If the CRTC is not active, dsp_lut_en will not get cleared.
> > + * Apparently we still need to do the above step to for
> > + * gamma correction to be disabled.
> > + */
> > + if (!crtc->state->active)
> > + return;
> > +

I have realized that the above might no longer be needed,
given we are now using atomic_enable and atomic_begin.

Not sure if the CRTC is supposed to clear its GAMMA
when disabled.

> > + ret = readx_poll_timeout(vop_dsp_lut_is_enable, vop,
> > + idle, !idle, 5, 30 * 1000);
> > + if (ret) {
> > + DRM_DEV_ERROR(vop->dev, "display LUT RAM enable timeout!\n");
> > + return;
> > + }
> > +
> > + if (crtc->state->gamma_lut &&
> > + (!old_crtc_state->gamma_lut || (crtc->state->gamma_lut->base.id !=
> > + old_crtc_state->gamma_lut->base.id))) {
>
> Silly question, but isn't the second part of this check redundant since you need
> color_mgmt_changed || active_changed to get into this function?
>
> So maybe invert the conditional here and exit early (to save a level of
> indentation in the block below):
>

I took this from malidp_atomic_commit_update_gamma. I _believe_
the rational for this is that color_mgmt_changed can be set by re-setting
the gamma property, to the same property. But I admit I haven't
tested it's the case.

OTOH, it won't really affect much to re-write the table, if the user
requested a change.

> if (!crtc->state->gamma_lut)
> return;
>

In any case, inverting the condition makes sense.

> spin_lock(&vop->reg_lock);
>
> vop_crtc_write_gamma_lut(vop, crtc);
> VOP_REG_SET(vop, common, dsp_lut_en, 1);
> vop_cfg_done(vop);
>
> spin_unlock(&vop->reg_lock);
>
> > +
> > + spin_lock(&vop->reg_lock);
> > +
> > + vop_crtc_write_gamma_lut(vop, crtc);
> > + VOP_REG_SET(vop, common, dsp_lut_en, 1);
> > + vop_cfg_done(vop);
> > +
> > + spin_unlock(&vop->reg_lock);
> > + }
> > +}
> > +
> > +static void vop_crtc_atomic_begin(struct drm_crtc *crtc,
> > + struct drm_crtc_state *old_crtc_state)
> > +{
> > + struct vop *vop = to_vop(crtc);
> > +
> > + /*
> > + * Only update GAMMA if the 'active' flag is not changed,
> > + * otherwise it's updated by .atomic_enable.
> > + */
> > + if (vop->lut_regs && crtc->state->color_mgmt_changed &&
> > + !crtc->state->active_changed)
> > + vop_crtc_gamma_set(vop, crtc, old_crtc_state);
> > +}
> > +
> > static void vop_crtc_atomic_enable(struct drm_crtc *crtc,
> > struct drm_crtc_state *old_state)
> > {
> > @@ -1075,6 +1154,14 @@ static void vop_crtc_atomic_enable(struct drm_crtc *crtc,
> > return;
> > }
> >
> > + /*
> > + * If we have a GAMMA LUT in the state, then let's make sure
> > + * it's updated. We might be coming out of suspend,
> > + * which means the LUT internal memory needs to be re-written.
> > + */
> > + if (vop->lut_regs && crtc->state->gamma_lut)
> > + vop_crtc_gamma_set(vop, crtc, old_state);
> > +
> > mutex_lock(&vop->vop_lock);
> >
> > WARN_ON(vop->event);
> > @@ -1191,6 +1278,26 @@ static void vop_wait_for_irq_handler(struct vop *vop)
> > synchronize_irq(vop->irq);
> > }
> >
> > +static int vop_crtc_atomic_check(struct drm_crtc *crtc,
> > + struct drm_crtc_state *crtc_state)
> > +{
> > + struct vop *vop = to_vop(crtc);
> > +
> > + if (vop->lut_regs && crtc_state->color_mgmt_changed &&
> > + crtc_state->gamma_lut) {
> > + unsigned int len;
> > +
> > + len = drm_color_lut_size(crtc_state->gamma_lut);
> > + if (len != crtc->gamma_size) {
> > + DRM_DEBUG_KMS("Invalid LUT size; got %d, expected %d\n",
> > + len, crtc->gamma_size);
> > + return -EINVAL;
> > + }
>
> Overflow is avoided in drm_mode_gamma_set_ioctl(), so I don't think you need
> this function.
>

But that only applies to the legacy path. Isn't this needed to ensure
a gamma blob
has the right size?

Thanks,
Ezequiel