Re: [PATCH v2 5/7] drm/msm/dsi: Account for DSC's bits_per_pixel having 4 fractional bits

From: Dmitry Baryshkov
Date: Wed Oct 05 2022 - 17:06:49 EST


On 05/10/2022 23:58, Marijn Suijten wrote:
On 2022-10-05 22:31:43, Dmitry Baryshkov wrote:
On Wed, 5 Oct 2022 at 21:17, Marijn Suijten
<marijn.suijten@xxxxxxxxxxxxxx> wrote:

drm_dsc_config's bits_per_pixel field holds a fractional value with 4
bits, which all panel drivers should adhere to for
drm_dsc_pps_payload_pack() to generate a valid payload. All code in the
DSI driver here seems to assume that this field doesn't contain any
fractional bits, hence resulting in the wrong values being computed.
Since none of the calculations leave any room for fractional bits or
seem to indicate any possible area of support, disallow such values
altogether.

Fixes: b9080324d6ca ("drm/msm/dsi: add support for dsc data")
Signed-off-by: Marijn Suijten <marijn.suijten@xxxxxxxxxxxxxx>
---
drivers/gpu/drm/msm/dsi/dsi_host.c | 25 +++++++++++++++++--------
1 file changed, 17 insertions(+), 8 deletions(-)

diff --git a/drivers/gpu/drm/msm/dsi/dsi_host.c b/drivers/gpu/drm/msm/dsi/dsi_host.c
index f42794cdd4c1..4717d49d76be 100644
--- a/drivers/gpu/drm/msm/dsi/dsi_host.c
+++ b/drivers/gpu/drm/msm/dsi/dsi_host.c
@@ -33,7 +33,7 @@

#define DSI_RESET_TOGGLE_DELAY_MS 20

-static int dsi_populate_dsc_params(struct drm_dsc_config *dsc);
+static int dsi_populate_dsc_params(struct msm_dsi_host *msm_host, struct drm_dsc_config *dsc);

static int dsi_get_version(const void __iomem *base, u32 *major, u32 *minor)
{
@@ -908,6 +908,7 @@ static void dsi_timing_setup(struct msm_dsi_host *msm_host, bool is_bonded_dsi)
u32 va_end = va_start + mode->vdisplay;
u32 hdisplay = mode->hdisplay;
u32 wc;
+ int ret;

DBG("");

@@ -943,7 +944,9 @@ static void dsi_timing_setup(struct msm_dsi_host *msm_host, bool is_bonded_dsi)
/* we do the calculations for dsc parameters here so that
* panel can use these parameters
*/
- dsi_populate_dsc_params(dsc);
+ ret = dsi_populate_dsc_params(msm_host, dsc);
+ if (ret)
+ return;

/* Divide the display by 3 but keep back/font porch and
* pulse width same
@@ -1769,7 +1772,7 @@ static char bpg_offset[DSC_NUM_BUF_RANGES] = {
2, 0, 0, -2, -4, -6, -8, -8, -8, -10, -10, -12, -12, -12, -12
};

-static int dsi_populate_dsc_params(struct drm_dsc_config *dsc)
+static int dsi_populate_dsc_params(struct msm_dsi_host *msm_host, struct drm_dsc_config *dsc)
{
int mux_words_size;
int groups_per_line, groups_total;
@@ -1780,6 +1783,12 @@ static int dsi_populate_dsc_params(struct drm_dsc_config *dsc)
int data;
int final_value, final_scale;
int i;
+ u16 bpp = dsc->bits_per_pixel >> 4;
+
+ if (dsc->bits_per_pixel & 0xf) {
+ DRM_DEV_ERROR(&msm_host->pdev->dev, "DSI does not support fractional bits_per_pixel\n");
+ return -EINVAL;
+ }

dsc->rc_model_size = 8192;
dsc->first_line_bpg_offset = 12;
@@ -1801,7 +1810,7 @@ static int dsi_populate_dsc_params(struct drm_dsc_config *dsc)
}

dsc->initial_offset = 6144; /* Not bpp 12 */
- if (dsc->bits_per_pixel != 8)
+ if (bpp != 8)
dsc->initial_offset = 2048; /* bpp = 12 */

mux_words_size = 48; /* bpc == 8/10 */
@@ -1824,14 +1833,14 @@ static int dsi_populate_dsc_params(struct drm_dsc_config *dsc)
* params are calculated
*/
groups_per_line = DIV_ROUND_UP(dsc->slice_width, 3);
- dsc->slice_chunk_size = DIV_ROUND_UP(dsc->slice_width * dsc->bits_per_pixel, 8);
+ dsc->slice_chunk_size = DIV_ROUND_UP(dsc->slice_width * bpp, 8);

I'd still prefer if we can get closer to drm_dsc_compute_rc_parameters().
The mentioned function has the following code:

vdsc_cfg->slice_chunk_size = DIV_ROUND_UP(vdsc_cfg->slice_width *

vdsc_cfg->bits_per_pixel,
(8 * 16));

Fwiw this discussion happened in dsi_update_dsc_timing() where a similar
calculation was the sole user of bits_per_pixel. This function,
dsi_populate_dsc_params(), has more uses of bits_per_pixel so it made
more sense to compute and document this "discrepancy" up front.
drm_dsc_compute_rc_parameters() doesn't document where this "16" comes
from, unfortunately (requiring knowledge of the contents of the struct).

In fact, could you please take a look if we can switch to using this
function and drop our code?

There's alread a:

/* FIXME: need to call drm_dsc_compute_rc_parameters() so that rest of
* params are calculated
*/

And it was trivial to replace everything below that comment with this
function call; I have not checked the math in detail but it assigns
_every_ field that was also assigned here, and the resulting values
provide an identical DCS PPS (which I happened to be printing to compare
to downstream, leading to find all the issues solved in this series) and
working phone screen.

Makes me wonder why this wasn't caught in review and replaced from the
get-go...

Good question. Partially it was because everybody wanted to get DSC support in to unblock other features. Thus DSC supporting code received several bumps afterwards.

Do you want me to do this in a v3, before applying this fractional-bits
fix? At that point this becomes the only user of bits_per_pixel:

Yes, please. This sounds like a perfect solution.


dsc->initial_offset = 6144; /* Not bpp 12 */
if (bpp != 8)
dsc->initial_offset = 2048; /* bpp = 12 */

Note that intel_vdsc.c has the exact same code right where they fill
vdsc_cfg->initial_offset:

int bpp = vdsc_cfg->bits_per_pixel >> 4;

I'm inclined to leave this as it is.

- Marijn

--
With best wishes
Dmitry