drm/bridge: dw-hdmi-qp: Compute ACR CTS for unlisted TMDS rates
Subsystems (5)
PATCH 1/1 drm/bridge: dw-hdmi-qp: Compute ACR CTS for unlisted TMDS rates
common_tmds_cts_table[] holds only six TMDS character rates (25.175, 25.2, 27, 54, 74.25 and 148.5 MHz), so dw_hdmi_qp_find_cts() returns 0 for everything else. dw_hdmi_qp_set_cts_n() then clears the CTS override enable and programs a value of 0, leaving the sink with no CTS to regenerate the audio clock from.
Any deep colour link falls into this gap: a 10 bpc RK3576 HDMI output runs at 185625000 Hz (148.5 MHz * 1.25), which is absent from both tables. N is computed dynamically and comes out correct at 6144, but AUDPKT_ACR_CONTROL1 reads back as 0.
Give CTS the same dynamic fallback that N already has, using the formula from the Audio chapter of the HDMI specification, and drop the -ENOENT returned into an unsigned int for the unlisted sample rates.
---
This was found after testing unrelated patches from DetlevC that rename the RK audio cards to see the impact in Kodi. RK3588 had audio output, while RK3576 did not. I'd not used an RK3576 board for a while so tasked Claude to help triage the problem, and this was the finding. The problem appears to have been exposed since Kodi reworked plane selection logic and support for 10bpc planes; earlier Kodi/LibreELEC images were using 8bpc planes thus avoiding the problem.
drivers/gpu/drm/bridge/synopsys/dw-hdmi-qp.c | 45 ++++++++++++-------- 1 file changed, 27 insertions(+), 18 deletions(-)
drivers/gpu/drm/bridge/synopsys/dw-hdmi-qp.c +27 -18
@@ -12,6 +12,7 @@#include <linux/export.h> #include <linux/i2c.h> #include <linux/irq.h> +#include <linux/math64.h> #include <linux/minmax.h> #include <linux/module.h> #include <linux/mutex.h>
@@ -307,8 +308,15 @@ static unsigned int dw_hdmi_qp_find_n(struct dw_hdmi_qp *hdmi, unsigned long pixreturn dw_hdmi_qp_compute_n (hdmi, pixel_clk, sample_rate); } +static unsigned int dw_hdmi_qp_compute_cts (unsigned long pixel_clk, +unsigned long sample_rate, +unsigned int n) +{ +return div64_u64 ((u64 )pixel_clk * n,128ULL * sample_rate); +} +static unsigned int dw_hdmi_qp_find_cts (struct dw_hdmi_qp *hdmi,unsigned long pixel_clk, -unsigned long sample_rate) +unsigned long sample_rate,unsigned int n) {const struct dw_hdmi_audio_tmds_cts *tmds_cts =NULL ;int i;
@@ -320,23 +328,24 @@ static unsigned int dw_hdmi_qp_find_cts(struct dw_hdmi_qp *hdmi, unsigned long p } } -if (!tmds_cts) -return 0 ; - -switch (sample_rate) { -case 32000 : -return tmds_cts->cts_32k; -case 44100 : -case 88200 : -case 176400 : -return tmds_cts->cts_44k1; -case 48000 : -case 96000 : -case 192000 : -return tmds_cts->cts_48k; -default : -return -ENOENT ; +if (tmds_cts) { +switch (sample_rate) { +case 32000 : +return tmds_cts->cts_32k; +case 44100 : +case 88200 : +case 176400 : +return tmds_cts->cts_44k1; +case 48000 : +case 96000 : +case 192000 : +return tmds_cts->cts_48k; + } } + +dev_dbg (hdmi->dev,"Rate %lu missing; compute CTS dynamically\n" , pixel_clk); + +return dw_hdmi_qp_compute_cts (pixel_clk, sample_rate, n); } static void dw_hdmi_qp_set_audio_interface(struct dw_hdmi_qp *hdmi,
@@ -471,7 +480,7 @@ static void dw_hdmi_qp_set_sample_rate(struct dw_hdmi_qp *hdmi, unsigned long lounsigned int n, cts; n =dw_hdmi_qp_find_n (hdmi, tmds_char_rate, sample_rate); - cts =dw_hdmi_qp_find_cts (hdmi, tmds_char_rate, sample_rate); + cts =dw_hdmi_qp_find_cts (hdmi, tmds_char_rate, sample_rate, n);dw_hdmi_qp_set_cts_n (hdmi, cts, n); }
Signature · 1 line
2.43.0
Hi,
Quoted 18 lines · Christian Hewitt
common_tmds_cts_table[] holds only six TMDS character rates (25.175, 25.2, 27, 54, 74.25 and 148.5 MHz), so
dw_hdmi_qp_find_cts()returns 0 for everything else.dw_hdmi_qp_set_cts_n()then clears the CTS override enable and programs a value of 0, leaving the sink with no CTS to regenerate the audio clock from.Any deep colour link falls into this gap: a 10 bpc RK3576 HDMI output runs at 185625000 Hz (148.5 MHz * 1.25), which is absent from both tables. N is computed dynamically and comes out correct at 6144, but
AUDPKT_ACR_CONTROL1reads back as 0.Give CTS the same dynamic fallback that N already has, using the formula from the Audio chapter of the HDMI specification, and drop the
-ENOENTreturned into an unsigned int for the unlisted sample rates.Fixes fd0141d1a8a2a ("drm/bridge: synopsys: Add audio support for dw-hdmi-qp")Assisted-by Claude:claude-opus-5Signed-off-by Christian Hewitt <[email protected]>
Quoted 8 lines
---
This was found after testing unrelated patches from DetlevC that rename the RK audio cards to see the impact in Kodi. RK3588 had audio output, while RK3576 did not. I'd not used an RK3576 board for a while so tasked Claude to help triage the problem, and this was the finding. The problem appears to have been exposed since Kodi reworked plane selection logic and support for 10bpc planes; earlier Kodi/LibreELEC images were using 8bpc planes thus avoiding the problem.
You are looking for this series:
https://lore.kernel.org/linux-rockchip/[email protected]/
Greetings,
-- Sebastian
Quoted 85 lines
drivers/gpu/drm/bridge/synopsys/dw-hdmi-qp.c | 45 ++++++++++++-------- 1 file changed, 27 insertions(+), 18 deletions(-)drivers/gpu/drm/bridge/synopsys/dw-hdmi-qp.c +27
-18@@ -12,6 +12,7 @@#include <linux/export.h> #include <linux/i2c.h> #include <linux/irq.h> +#include <linux/math64.h> #include <linux/minmax.h> #include <linux/module.h> #include <linux/mutex.h> @@ -307,8 +308,15 @@ static unsigned int dw_hdmi_qp_find_n(struct dw_hdmi_qp *hdmi, unsigned long pixreturn dw_hdmi_qp_compute_n (hdmi, pixel_clk, sample_rate); } +static unsigned int dw_hdmi_qp_compute_cts (unsigned long pixel_clk, +unsigned long sample_rate, +unsigned int n) +{ +return div64_u64 ((u64 )pixel_clk * n,128ULL * sample_rate); +} +static unsigned int dw_hdmi_qp_find_cts (struct dw_hdmi_qp *hdmi,unsigned long pixel_clk, -unsigned long sample_rate) +unsigned long sample_rate,unsigned int n) {const struct dw_hdmi_audio_tmds_cts *tmds_cts =NULL ;int i;@@ -320,23 +328,24 @@ static unsigned int dw_hdmi_qp_find_cts(struct dw_hdmi_qp *hdmi, unsigned long p } } -if (!tmds_cts) -return 0 ; - -switch (sample_rate) { -case 32000 : -return tmds_cts->cts_32k; -case 44100 : -case 88200 : -case 176400 : -return tmds_cts->cts_44k1; -case 48000 : -case 96000 : -case 192000 : -return tmds_cts->cts_48k; -default : -return -ENOENT ; +if (tmds_cts) { +switch (sample_rate) { +case 32000 : +return tmds_cts->cts_32k; +case 44100 : +case 88200 : +case 176400 : +return tmds_cts->cts_44k1; +case 48000 : +case 96000 : +case 192000 : +return tmds_cts->cts_48k; + } } + +dev_dbg (hdmi->dev,"Rate %lu missing; compute CTS dynamically\n" , pixel_clk); + +return dw_hdmi_qp_compute_cts (pixel_clk, sample_rate, n); } static void dw_hdmi_qp_set_audio_interface(struct dw_hdmi_qp *hdmi,@@ -471,7 +480,7 @@ static void dw_hdmi_qp_set_sample_rate(struct dw_hdmi_qp *hdmi, unsigned long lounsigned int n, cts; n =dw_hdmi_qp_find_n (hdmi, tmds_char_rate, sample_rate); - cts =dw_hdmi_qp_find_cts (hdmi, tmds_char_rate, sample_rate); + cts =dw_hdmi_qp_find_cts (hdmi, tmds_char_rate, sample_rate, n);dw_hdmi_qp_set_cts_n (hdmi, cts, n); }Signature · 1 line
2.43.0
Quoted 32 lines · Sebastian Reichel
Hi,
Quoted 18 lines · Christian Hewitt
common_tmds_cts_table[] holds only six TMDS character rates (25.175, 25.2, 27, 54, 74.25 and 148.5 MHz), so
dw_hdmi_qp_find_cts()returns 0 for everything else.dw_hdmi_qp_set_cts_n()then clears the CTS override enable and programs a value of 0, leaving the sink with no CTS to regenerate the audio clock from.Any deep colour link falls into this gap: a 10 bpc RK3576 HDMI output runs at 185625000 Hz (148.5 MHz * 1.25), which is absent from both tables. N is computed dynamically and comes out correct at 6144, but
AUDPKT_ACR_CONTROL1reads back as 0.Give CTS the same dynamic fallback that N already has, using the formula from the Audio chapter of the HDMI specification, and drop the
-ENOENTreturned into an unsigned int for the unlisted sample rates.Fixes fd0141d1a8a2a ("drm/bridge: synopsys: Add audio support for dw-hdmi-qp")Assisted-by Claude:claude-opus-5Signed-off-by Christian Hewitt <[email protected]>Quoted 8 lines
---
This was found after testing unrelated patches from DetlevC that rename the RK audio cards to see the impact in Kodi. RK3588 had audio output, while RK3576 did not. I'd not used an RK3576 board for a while so tasked Claude to help triage the problem, and this was the finding. The problem appears to have been exposed since Kodi reworked plane selection logic and support for 10bpc planes; earlier Kodi/LibreELEC images were using 8bpc planes thus avoiding the problem.
Quoted 3 lines
You are looking for this series:
Ahh, I missed that. I have now tested that patch (and will reply to it with a TB) so please ignore this patch. Thanks!
Christian