drm: bridge: fsl-ldb: fixup mode on freq mismatch
Subsystems (3)
PATCH 1/1 drm: bridge: fsl-ldb: fixup mode on freq mismatch
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
To overcome this, check this condition in mode_fixup() and adapt the pixel clock accordingly.
--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)
drivers/gpu/drm/bridge/fsl-ldb.c +36 -4
@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb); } static int fsl_ldb_attach(struct drm_bridge *bridge,
@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode) +{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge); +unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb); +unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb); + +if (adjusted_mode->clock != pclk) { +dev_warn (fsl_ldb->dev,"Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n" , + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable (struct drm_bridge *bridge,struct drm_bridge_state *old_bridge_state) {
@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,
Signature · 1 line
2.43.0
Luca Ceresoli
reviewer+Cc: Miquèl, who is actively working on imx8mp video clock rates.
Quoted 83 lines · On Tue, 26 Nov 2024 16:45:54 +0100 Nikolaus Voss <[email protected]> wrote:
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>Signed-off-by Nikolaus Voss <[email protected]>--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +36
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb); } static int fsl_ldb_attach(struct drm_bridge *bridge,@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode) +{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge); +unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb); +unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb); + +if (adjusted_mode->clock != pclk) { +dev_warn (fsl_ldb->dev,"Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n" , + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable (struct drm_bridge *bridge,struct drm_bridge_state *old_bridge_state) {@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,
Signature · 3 lines
Luca Ceresoli, Bootlin Embedded Linux and Kernel engineering https://bootlin.com
Quoted 40 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>Signed-off-by Nikolaus Voss <[email protected]>--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb); } static int fsl_ldb_attach(struct drm_bridge *bridge,
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 128
Quoted 6 lines
drivers/gpu/drm/bridge/fsl-ldb.c +3
-0@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode)
The driver uses atomic callbacks. Please use .atomic_check() instead.
Quoted 40 lines
drivers/gpu/drm/bridge/fsl-ldb.c +27
-0+{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge); +unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb); +unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb); + +if (adjusted_mode->clock != pclk) { +dev_warn (fsl_ldb->dev,"Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n" , + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable (struct drm_bridge *bridge,struct drm_bridge_state *old_bridge_state) {@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,Signature · 1 line
2.43.0
Signature · 2 lines
With best wishes Dmitry
Hi Dmitry,
Quoted 44 lines · Dmitry Baryshkov
Quoted 43 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>Signed-off-by Nikolaus Voss <[email protected]>--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb); } static int fsl_ldb_attach(struct drm_bridge *bridge,@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); }
Quoted 5 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 128
Quoted 3 lines
drivers/gpu/drm/bridge/fsl-ldb.c +3
-0+static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode)The driver uses atomic callbacks. Please use .atomic_check() instead.
So it is okay to modify drm_crtc_state->adjusted_mode in .atomic_check()? I chose .mode_fixup() because the function name and args make it more obvious what is done there. Btw, the API reference doesn't say this call is deprecated.
A second thought:
Maybe it would be a good idea to reject modes which result in an adjusted mode pclk that is not within certain boundaries, even if this patch doesn't do it yet. As I see it, that would be only possible in mode_fixup().
Quoted 43 lines
Quoted 41 lines
drivers/gpu/drm/bridge/fsl-ldb.c +27
-1+{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge); +unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb); +unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb); + +if (adjusted_mode->clock != pclk) { +dev_warn (fsl_ldb->dev,"Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n" , + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable (struct drm_bridge *bridge,struct drm_bridge_state *old_bridge_state) {@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state, --2.43.0
Signature · 1 line
Nikolaus Voss
Quoted 47 lines · Nikolaus Voss
Hi Dmitry,
Quoted 44 lines · Dmitry Baryshkov
Quoted 43 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>Signed-off-by Nikolaus Voss <[email protected]>--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb); } static int fsl_ldb_attach(struct drm_bridge *bridge,@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); }
Quoted 7 lines
Quoted 5 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 128
Quoted 3 lines
drivers/gpu/drm/bridge/fsl-ldb.c +3
-0+static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode)The driver uses atomic callbacks. Please use .atomic_check() instead.
So it is okay to modify
drm_crtc_state->adjusted_modein .atomic_check()? I
Yes. samsung-dsim, anx7625 do that (I stopped checking after the second one).
Quoted 3 lines
chose .mode_fixup() because the function name and args make it more obvious what is done there. Btw, the API reference doesn't say this call is deprecated.
It's not deprecated. But as the driver is using atomic calls (vs legacy calls) it makes more sense to use atomic_check() too.
Quoted 4 lines
A second thought:
Maybe it would be a good idea to reject modes which result in an adjusted mode pclk that is not within certain boundaries, even if this patch doesn't do it yet. As I see it, that would be only possible inmode_fixup().
atomic_check() can definitely reject whatever is being stuffed to it.
Quoted 48 lines
Quoted 44 lines
Quoted 42 lines
drivers/gpu/drm/bridge/fsl-ldb.c +27
-1+{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge); +unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb); +unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb); + +if (adjusted_mode->clock != pclk) { +dev_warn (fsl_ldb->dev,"Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n" , + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable (struct drm_bridge *bridge,struct drm_bridge_state *old_bridge_state) {@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state, --2.43.0
Signature · 1 line
Nikolaus Voss
Signature · 2 lines
With best wishes Dmitry
Quoted 10 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>
It looks like stable is not effectively Cc'ed.
Need a Fixes tag?
Quoted 30 lines
Signed-off-by Nikolaus Voss <[email protected]>--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb); } static int fsl_ldb_attach(struct drm_bridge *bridge,
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 130
Quoted 8 lines
drivers/gpu/drm/bridge/fsl-ldb.c +5
-0@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode) +{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge);
Why const?
If no const, then ...
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 132
Quoted 2 lines
drivers/gpu/drm/bridge/fsl-ldb.c +2
-0+unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb);
... this could be
unsigned long requested_link_freq =
fsl_ldb_link_frequency(fsl_ldb, mode->clock);
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 141
Quoted 9 lines
drivers/gpu/drm/bridge/fsl-ldb.c +9
-0+unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb);
I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10 ("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")
Quoted 17 lines
drivers/gpu/drm/bridge/fsl-ldb.c +13
-0+ +if (adjusted_mode->clock != pclk) { +dev_warn (fsl_ldb->dev,"Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n" , + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable (struct drm_bridge *bridge,struct drm_bridge_state *old_bridge_state) {
drivers/gpu/drm/bridge/fsl-ldb.c fsl_ldb_mode_valid(struct drm_bridge *bridge, · line 317
Quoted 7 lines
drivers/gpu/drm/bridge/fsl-ldb.c +1
-0@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,
--
Regards,
Liu Ying
Hi Liu,
Quoted 7 lines · Liu Ying
Quoted 6 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
Quoted 7 lines
Quoted 4 lines
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>It looks like stable is not effectively Cc'ed.
Need a Fixes tag?
I will include
in v2.
Quoted 48 lines
Quoted 35 lines
Signed-off-by Nikolaus Voss <[email protected]>--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb);}
static int fsl_ldb_attach(struct
drm_bridge*bridge,drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge · line 131
Quoted 9 lines
drivers/gpu/drm/bridge/fsl-ldb.c +5
-0@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode) +{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge);Why const?
If no const, then ...
Quoted 8 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 132
Quoted 2 lines
drivers/gpu/drm/bridge/fsl-ldb.c +2
-0+unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb);... this could be
unsigned longrequested_link_freq=fsl_ldb_link_frequency(fsl_ldb,
mode->clock);
I used fsl_ldb_link_freq_factor(fsl_ldb) to point out the symmetry to recalculate pclk = freq / fsl_ldb_link_freq_factor(fsl_ldb) below:
Quoted 18 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 141
Quoted 10 lines
drivers/gpu/drm/bridge/fsl-ldb.c +9
-0+unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb);I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")
I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch. With slightly adapted pixel clocks though.
The driver already warns if LDB and PCLK don't match, because this is a hardware constraint and violating this will lead to no image or image distortion. With the patch the driver tries to fix that.
I don't know if it works for all or at least most panels, but for my panels it does. And if the clocks match, by chance or by refconfiguring the PLL base clock, the patch does nothing.
Quoted 29 lines
Quoted 19 lines
drivers/gpu/drm/bridge/fsl-ldb.c +13
-0+ +if (adjusted_mode->clock != pclk) { + dev_warn(fsl_ldb->dev, "Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n", + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable(struct
drm_bridge*bridge,drivers/gpu/drm/bridge/fsl-ldb.c +0
-0struct drm_bridge_state *old_bridge_state) {drivers/gpu/drm/bridge/fsl-ldb.c fsl_ldb_mode_valid(struct drm_bridge *bridge, · line 317
Quoted 7 lines
drivers/gpu/drm/bridge/fsl-ldb.c +1
-0@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,--
Regards,
Liu Ying
Signature · 1 line
Nikolaus Voss
Quoted 3 lines · Nikolaus Voss
[You don't often get email from [email protected]. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
Hi Liu,
Hi,
Quoted 77 lines · Liu Ying
Quoted 7 lines · Nikolaus Voss
Quoted 6 lines
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
Quoted 7 lines
Quoted 4 lines
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>It looks like stable is not effectively Cc'ed.
Need a Fixes tag?I will include
Fixes 463db5c2ed4a ("drm: bridge: ldb: Implement simple Freescalei.MX8MP LDB bridge")
in v2.Quoted 55 lines
Quoted 36 lines
Signed-off-by Nikolaus Voss <[email protected]>---
drivers/gpu/drm/bridge/fsl-ldb.c | 40
++++++++++++++++++++++++++++----1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ + return fsl_ldb_is_dual(fsl_ldb) ? 3500 : 7000; +} + static unsigned long fsl_ldb_link_frequency(struct fsl_ldb *fsl_ldb,int clock) { - if (fsl_ldb_is_dual(fsl_ldb)) - return clock * 3500; - else - return clock * 7000; + return clock * fsl_ldb_link_freq_factor(fsl_ldb);}
static int fsl_ldb_attach(struct
drm_bridge*bridge,drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge · line 131
Quoted 9 lines
drivers/gpu/drm/bridge/fsl-ldb.c +5
-0@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, + const struct drm_display_mode *mode, + struct drm_display_mode *adjusted_mode) +{ + const struct fsl_ldb *fsl_ldb = to_fsl_ldb(bridge);Why const?
If no const, then ...drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 132
Quoted 2 lines
drivers/gpu/drm/bridge/fsl-ldb.c +2
-0+ unsigned long requested_link_freq = + mode->clock * fsl_ldb_link_freq_factor(fsl_ldb);... this could be
unsigned longrequested_link_freq=
Quoted 6 lines
Quoted 2 lines
fsl_ldb_link_frequency(fsl_ldb, mode->clock);I used
fsl_ldb_link_freq_factor(fsl_ldb)to point out the symmetry to recalculate
pclk = freq /fsl_ldb_link_freq_factor(fsl_ldb)below:
I would prefer to avoid this open coding of using fsl_ldb_link_freq_factor() than achieve some sort of symmetry.
Quoted 25 lines
Quoted 19 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 141
Quoted 11 lines
drivers/gpu/drm/bridge/fsl-ldb.c +9
-0+ unsigned long freq = clk_round_rate(fsl_ldb->clk, requested_link_freq); + + if (freq != requested_link_freq) { + /* + * this will lead to flicker and incomplete lines on + * the attached display, adjust the CRTC clock + * accordingly. + */ + int pclk = freq / fsl_ldb_link_freq_factor(fsl_ldb);I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")
Quoted 8 lines
I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.
If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
Quoted 6 lines
The driver already warns if LDB and PCLK don't match, because this is a hardware constraint and violating this will lead to no image or image distortion. With the patch the driver tries to fix that.
I don't know if it works for all or at least most panels, but for my
This patch is supposed to work for all panels and LVDS display bridges, but I don't think it does due to the problematic pixel clock rates I mentioned before.
Quoted 37 lines
panels
it does. And if the clocks match, by chance or by refconfiguring the PLL base
clock, the patch does nothing.Quoted 29 lines
Quoted 26 lines
drivers/gpu/drm/bridge/fsl-ldb.c +13
-0+ + if (adjusted_mode->clock != pclk) { + dev_warn(fsl_ldb->dev, "Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n", + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + + return true; +} +static void
fsl_ldb_atomic_enable(struct drm_bridge *bridge, struct drm_bridge_state *old_bridge_state){drivers/gpu/drm/bridge/fsl-ldb.c +1
-0@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge, static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,Signature · 2 lines
Regards, Liu YingSignature · 1 line
Nikolaus Voss
Signature · 2 lines
Regards, Liu Ying
On 12/3/24 3:22 AM, Liu Ying wrote:
[...]
Quoted 17 lines
Quoted 17 lines
Quoted 7 lines
I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
Quoted 7 lines
Quoted 4 lines
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
I temporarily agree.
I also currently use DTOs for various panels including their PLL setting, but in the end, I think/hope the work of Miquel and co. is going to make that PLL setting part unnecessary.
Quoted 26 lines · Marek Vasut
On 12/3/24 3:22 AM, Liu Ying wrote:
[...]
Quoted 22 lines
Quoted 21 lines
Quoted 7 lines
I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.
Quoted 7 lines
Quoted 2 lines
If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
I temporarily agree.
I also currently use DTOs for various panels including their PLL setting, but in the end, I think/hope the work of Miquel and co. is going to make that PLL setting part unnecessary.
That is exactly what my patch is about. I want to use one DT for all panels and store the panel's timing in EDID EEPROM.
Signature · 1 line
Nikolaus Voss
Quoted 29 lines · Nikolaus Voss
Quoted 28 lines · Marek Vasut
On 12/3/24 3:22 AM, Liu Ying wrote:
[...]
Quoted 24 lines
Quoted 13 lines
Quoted 7 lines
I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")Quoted 8 lines
I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
Quoted 8 lines
Quoted 5 lines
I temporarily agree.
I also currently use DTOs for various panels including their PLL setting, but in the end, I think/hope the work of Miquel and co. is going to make that PLL setting part unnecessary.
That is exactly what my patch is about. I want to use one DT for all panels
Right
Quoted 1 line
and store the panel's timing in EDID EEPROM.
Oh, that is a new one. Does the EDID EEPROM store the entirety of 'struct display_timing {}' somehow , or is that a custom format ?
Hi Marek,
Quoted 36 lines · Marek Vasut
Quoted 35 lines · Nikolaus Voss
Quoted 33 lines · Marek Vasut
On 12/3/24 3:22 AM, Liu Ying wrote:
[...]
Quoted 17 lines
Quoted 17 lines
Quoted 7 lines
I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
Quoted 7 lines
Quoted 4 lines
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
I temporarily agree.
I also currently use DTOs for various panels including their PLL setting, but in the end, I think/hope the work of Miquel and co. is going to make that PLL setting part unnecessary.
Quoted 8 lines
Quoted 2 lines
That is exactly what my patch is about. I want to use one DT for all panels
Right
Quoted 1 line
and store the panel's timing in EDID EEPROM.
Oh, that is a new one. Does the EDID EEPROM store the entirety of 'struct
display_timing{}' somehow , or is that a custom format ?
Well, sort of ;-). VESA has taken care of this 30 years ago
(https://en.wikipedia.org/wiki/Extended_Display_Identification_Data).
DRM handles this with drm_get_edid() and siblings, e.g. :
@@ -86,16 +92,36 @@ static int panel_lvds_get_modes(struct drm_panel *panel,
{
struct panel_lvds *lvds = to_panel_lvds(panel);
struct drm_display_mode *mode;+ int num = 0;
+
+ /* probe EDID if a DDC bus is available */
+ if (lvds->ddc) {
+ pm_runtime_get_sync(lvds->dev);
+
+ if (!lvds->edid)
+ lvds->edid = drm_get_edid(connector, lvds->ddc);
+
+ if (lvds->edid)
+ num += drm_add_edid_modes(connector, lvds->edid);
+
+ pm_runtime_mark_last_busy(lvds->dev);
+ pm_runtime_put_autosuspend(lvds->dev);
+ }
panel-simple.c does that in mainline, I added it to panel-lvds.c. The kernel subdir tools/edid has some code to generate the EEPROM data from timings and flags.
We keep the DDC EEPROM on a small adapter glued to to back of the panel so we can replace the usually short-lived panel with a successor.
Signature · 1 line
Nikolaus Voss
Quoted 1 line · Nikolaus Voss
Hi Marek,
Hi,
Quoted 35 lines
Quoted 35 lines
Quoted 24 lines
Quoted 24 lines
Quoted 24 lines
Quoted 13 lines
Quoted 7 lines
I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")Quoted 8 lines
I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
Quoted 8 lines
Quoted 5 lines
I temporarily agree.
I also currently use DTOs for various panels including their PLL setting, but in the end, I think/hope the work of Miquel and co. is going to make that PLL setting part unnecessary.
That is exactly what my patch is about. I want to use one DT for all panels
Right
Quoted 8 lines
Quoted 3 lines
Quoted 1 line
and store the panel's timing in EDID EEPROM.
Oh, that is a new one. Does the EDID EEPROM store the entirety of 'struct
display_timing{}' somehow , or is that a custom format ?Well, sort of ;-). VESA has taken care of this 30 years ago
(https://en.wikipedia.org/wiki/Extended_Display_Identification_Data).DRM handles this with
drm_get_edid()and siblings, e.g. :
EDID can not encode all the information in struct display_timing {} , or can it ?
I think what you would be missing are bus_flags , bus_format and possibly the single/dual link and channel (odd/even) mapping, won't you ?
Quoted 25 lines
(quoted diff) +15
-0@@ -86,16 +92,36 @@ static int panel_lvds_get_modes(struct drm_panel *panel, { struct panel_lvds *lvds = to_panel_lvds(panel); struct drm_display_mode *mode; + int num = 0; + + /* probe EDID if a DDC bus is available */ + if (lvds->ddc) { + pm_runtime_get_sync(lvds->dev); + + if (!lvds->edid) + lvds->edid = drm_get_edid(connector, lvds->ddc); + + if (lvds->edid) + num += drm_add_edid_modes(connector, lvds->edid); + + pm_runtime_mark_last_busy(lvds->dev); + pm_runtime_put_autosuspend(lvds->dev); + }
panel-simple.cdoes that in mainline, I added it topanel-lvds.c. The kernel subdirtools/edidhas some code to generate the EEPROM data from timings and flags.
Quoted 2 lines
We keep the DDC EEPROM on a small adapter glued to to back of the panel so we can replace the usually short-lived panel with a successor.
OK
Hi Marek,
Quoted 43 lines · Marek Vasut
Quoted 42 lines · Nikolaus Voss
Quoted 30 lines
Quoted 30 lines
Quoted 29 lines
Quoted 17 lines
Quoted 17 lines
Quoted 7 lines
I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
Quoted 7 lines
Quoted 4 lines
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
I temporarily agree.
I also currently use DTOs for various panels including their PLL setting, but in the end, I think/hope the work of Miquel and co. is going to make that PLL setting part unnecessary.
Quoted 8 lines
Quoted 2 lines
That is exactly what my patch is about. I want to use one DT for all panels
Right
Quoted 1 line
and store the panel's timing in EDID EEPROM.
Oh, that is a new one. Does the EDID EEPROM store the entirety of 'struct
display_timing{}' somehow , or is that a custom format ?Well, sort of ;-). VESA has taken care of this 30 years ago
(https://en.wikipedia.org/wiki/Extended_Display_Identification_Data).
Quoted 8 lines
Quoted 1 line
DRM handles this with
drm_get_edid()and siblings, e.g. :EDID can not encode all the information in struct
display_timing{} , or can it ?I think what you would be missing are
bus_flags,bus_formatand possibly the single/dual link and channel (odd/even) mapping, won't you ?
Yes, that's right. I use the vendor block for bus_flags and bus_format now, but that's not standard and not portable of course.
My first idea was to store the DT overlay in the display EEPROM but a standard 1k EEPROM is too small for that.
Signature · 1 line
Nikolaus Voss
On 12/9/24 9:46 AM, Nikolaus Voss wrote:
Hi,
Quoted 16 lines
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
and store the panel's timing in EDID EEPROM.
Oh, that is a new one. Does the EDID EEPROM store the entirety of 'struct
display_timing{}' somehow , or is that a custom format ?Well, sort of ;-). VESA has taken care of this 30 years ago
(https://en.wikipedia.org/wiki/Extended_Display_Identification_Data).Quoted 8 lines
Quoted 1 line
DRM handles this with
drm_get_edid()and siblings, e.g. :EDID can not encode all the information in struct
display_timing{} , or can it ?I think what you would be missing are
bus_flags,bus_formatand possibly the single/dual link and channel (odd/even) mapping, won't you ?
Quoted 5 lines
Yes, that's right. I use the vendor block for
bus_flagsandbus_formatnow, but that's not standard and not portable of course.My first idea was to store the DT overlay in the display EEPROM but a standard 1k EEPROM is too small for that.
Understood. I had the same problem, in the end I went for custom encoding in the EEPROM but the amount of panels was limited in my case.
Indeed, DTO does not fit the EEPROM and EDID is not really fitting too well to DPI/LVDS panels, it has too many fields that are specific to regular pluggable panels and not useful on DPI/LVDS ones.
Hi Marek,
Quoted 22 lines · Marek Vasut
Quoted 16 lines · Nikolaus Voss
Quoted 7 lines
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
and store the panel's timing in EDID EEPROM.
Oh, that is a new one. Does the EDID EEPROM store the entirety of 'struct
display_timing{}' somehow , or is that a custom format ?Well, sort of ;-). VESA has taken care of this 30 years ago
(https://en.wikipedia.org/wiki/Extended_Display_Identification_Data).Quoted 8 lines
Quoted 1 line
DRM handles this with
drm_get_edid()and siblings, e.g. :EDID can not encode all the information in struct
display_timing{} , or can it ?I think what you would be missing are
bus_flags,bus_formatand possibly the single/dual link and channel (odd/even) mapping, won't you ?Quoted 5 lines
Yes, that's right. I use the vendor block for
bus_flagsandbus_formatnow, but that's not standard and not portable of course.My first idea was to store the DT overlay in the display EEPROM but a standard 1k EEPROM is too small for that.
Quoted 7 lines
Understood. I had the same problem, in the end I went for custom encoding in the EEPROM but the amount of panels was limited in my case.
Indeed, DTO does not fit the EEPROM and EDID is not really fitting too well to DPI/LVDS panels, it has too many fields that are specific to regular pluggable panels and not useful on DPI/LVDS ones.
The curse of backward-compatibility ;-). Nevertheless, we are doing well with EDID for years now. 95% is implemented in standard kernel, you just need a few quirks.
Signature · 1 line
Nikolaus Voss
Quoted 124 lines · Liu Ying
Quoted 3 lines · Nikolaus Voss
[You don't often get email from [email protected]. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
Hi Liu,
Hi,
Quoted 78 lines · Liu Ying
Quoted 7 lines · Nikolaus Voss
Quoted 6 lines
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
Quoted 7 lines
Quoted 4 lines
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>It looks like stable is not effectively Cc'ed.
Need a Fixes tag?I will include
Fixes 463db5c2ed4a ("drm: bridge: ldb: Implement simple Freescalei.MX8MP LDB bridge")
in v2.Quoted 56 lines
Quoted 37 lines
Signed-off-by Nikolaus Voss <[email protected]>---
drivers/gpu/drm/bridge/fsl-ldb.c | 40
++++++++++++++++++++++++++++----1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ + return fsl_ldb_is_dual(fsl_ldb) ? 3500 : 7000; +} + static unsigned long fsl_ldb_link_frequency(struct fsl_ldb *fsl_ldb,int clock) { - if (fsl_ldb_is_dual(fsl_ldb)) - return clock * 3500; - else - return clock * 7000; + return clock * fsl_ldb_link_freq_factor(fsl_ldb);}
static int fsl_ldb_attach(struct
drm_bridge*bridge,drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge · line 131
Quoted 9 lines
drivers/gpu/drm/bridge/fsl-ldb.c +5
-0@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, + const struct drm_display_mode *mode, + struct drm_display_mode *adjusted_mode) +{ + const struct fsl_ldb *fsl_ldb = to_fsl_ldb(bridge);Why const?
If no const, then ...drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 132
Quoted 2 lines
drivers/gpu/drm/bridge/fsl-ldb.c +2
-0+ unsigned long requested_link_freq = + mode->clock * fsl_ldb_link_freq_factor(fsl_ldb);... this could be
unsigned longrequested_link_freq=Quoted 6 lines
Quoted 2 lines
fsl_ldb_link_frequency(fsl_ldb, mode->clock);I used
fsl_ldb_link_freq_factor(fsl_ldb)to point out the symmetry to recalculate
pclk = freq /fsl_ldb_link_freq_factor(fsl_ldb)below:I would prefer to avoid this open coding of using
fsl_ldb_link_freq_factor()than achieve some sort of symmetry.Quoted 29 lines
Quoted 19 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 141
Quoted 11 lines
drivers/gpu/drm/bridge/fsl-ldb.c +9
-0+ unsigned long freq = clk_round_rate(fsl_ldb->clk, requested_link_freq); + + if (freq != requested_link_freq) { + /* + * this will lead to flicker and incomplete lines on + * the attached display, adjust the CRTC clock + * accordingly. + */ + int pclk = freq / fsl_ldb_link_freq_factor(fsl_ldb);I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
Quoted 7 lines
Quoted 4 lines
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
Yes, but I would have to calculate the PLL clock by hand with knowledge of the internal signal flow (at least the clock ratio ;-)).
Quoted 11 lines
Quoted 7 lines
The driver already warns if LDB and PCLK don't match, because this is a hardware constraint and violating this will lead to no image or image distortion. With the patch the driver tries to fix that.
I don't know if it works for all or at least most panels, but for my
This patch is supposed to work for all panels and LVDS display bridges, but I don't think it does due to the problematic pixel clock rates I mentioned before.
To my understanding, the current driver works for nearly no panel, unless you explicitly code the correct PLL freq into the device tree. The basic problem is, that the driver does not _enforce_ the correct clock ratio although the hardware strictly _requires_ it.
There are two options in enforcing the correct ratio: forcing the ldb clock to the right rate, down to the clock source; this is Miquel's approach. Or getting the nearest possible ldb clock with the current PLL freq and modify the pixel clock accordingly, that's my approach.
Of course, never modifying the required pixel clock and still setting the correct clock ratio is the golden solution. But in most cases, this is not necessary as panels usually accept deviations from the ideal pixel clock without any problems. What doesn't work is pixel and ldb clock not having the correct ratio. And this is what the patch is about.
Quoted 33 lines
Quoted 33 lines
panels
it does. And if the clocks match, by chance or by refconfiguring the PLL base
clock, the patch does nothing.Quoted 27 lines
Quoted 27 lines
drivers/gpu/drm/bridge/fsl-ldb.c +3
-0+ + if (adjusted_mode->clock != pclk) { + dev_warn(fsl_ldb->dev, "Adjusted pixel clk tomatch LDB clk (%d
kHz -> %d kHz)!\n",
+adjusted_mode->clock, pclk);drivers/gpu/drm/bridge/fsl-ldb.c +9
-0+ + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + + return true; +} +static void
fsl_ldb_atomic_enable(struct drm_bridge *bridge, struct drm_bridge_state *old_bridge_state){drivers/gpu/drm/bridge/fsl-ldb.c +1
-0@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge, static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,
Signature · 1 line
Nikolaus Voss
Quoted 129 lines · Nikolaus Voss
Quoted 128 lines · Liu Ying
Quoted 3 lines · Nikolaus Voss
[You don't often get email from [email protected]. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]
Hi Liu,
Hi,
Quoted 77 lines · Liu Ying
Quoted 7 lines · Nikolaus Voss
Quoted 6 lines
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
Quoted 7 lines
Quoted 4 lines
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>It looks like stable is not effectively Cc'ed.
Need a Fixes tag?I will include
Fixes 463db5c2ed4a ("drm: bridge: ldb: Implement simple Freescalei.MX8MP LDB bridge")
in v2.Quoted 55 lines
Quoted 36 lines
Signed-off-by Nikolaus Voss <[email protected]>--- drivers/gpu/drm/bridge/fsl-ldb.c | 40 ++++++++++++++++++++++++++++---- 1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +6
-4@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor (const struct fsl_ldb *fsl_ldb) +{ +return fsl_ldb_is_dual (fsl_ldb) ?3500 :7000 ; +} +static unsigned long fsl_ldb_link_frequency (struct fsl_ldb *fsl_ldb,int clock) { -if (fsl_ldb_is_dual (fsl_ldb)) -return clock *3500 ; -else -return clock *7000 ; +return clock *fsl_ldb_link_freq_factor (fsl_ldb);}
static int fsl_ldb_attach(struct
drm_bridge*bridge,drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge · line 131
Quoted 9 lines
drivers/gpu/drm/bridge/fsl-ldb.c +5
-0@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge, bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, +const struct drm_display_mode *mode, +struct drm_display_mode *adjusted_mode) +{ +const struct fsl_ldb *fsl_ldb =to_fsl_ldb (bridge);Why const?
If no const, then ...drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 132
Quoted 2 lines
drivers/gpu/drm/bridge/fsl-ldb.c +2
-0+unsigned long requested_link_freq = + mode->clock *fsl_ldb_link_freq_factor (fsl_ldb);... this could be
unsigned longrequested_link_freq=Quoted 6 lines
Quoted 2 lines
fsl_ldb_link_frequency(fsl_ldb,
mode->clock);I used
fsl_ldb_link_freq_factor(fsl_ldb)to point out the symmetry to recalculate
pclk = freq /fsl_ldb_link_freq_factor(fsl_ldb)below:I would prefer to avoid this open coding of using
fsl_ldb_link_freq_factor()than achieve some sort of symmetry.Quoted 33 lines
Quoted 19 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 141
Quoted 11 lines
drivers/gpu/drm/bridge/fsl-ldb.c +9
-0+unsigned long freq =clk_round_rate (fsl_ldb->clk, requested_link_freq); + +if (freq != requested_link_freq) { +/* +* this will lead to flicker and incomplete lines on +* the attached display, adjust the CRTC clock +* accordingly. +*/ +int pclk = freq /fsl_ldb_link_freq_factor (fsl_ldb);I doubt that pixel clock tree cannot find appropriate division ratios for some pixel clock rates, especially for dual-link LVDS on i.MX8MP and i.MX93 platforms, because PLL clock rate should be 7x faster than pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5 folder between "ldb" clock and pixel clock can be met. That means the PLL clock rate needs to be explicitly set first for this case.
Quoted 5 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate the base PLL clock by hand and store it in the device tree.
In my test system, I have three different LVDS panels with EDID EEPROM, none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.
Quoted 5 lines
Quoted 2 lines
If each of the three LVDS panels has only one display mode, you may assign the PLL clock rates in DT overlays for the panels.
Yes, but I would have to calculate the PLL clock by hand with knowledge of the internal signal flow (at least the clock ratio ;-)).
To support a shared PLL between LVDS display pipeline and MIPI DSI display pipeline with i.MX8MP + separate PLLs, I'm afraid you have to assign the PLL clock rate(s) in DT with knowledge of the clock ratios. See my patch series for details.
https://lore.kernel.org/all/[email protected]/
Quoted 9 lines
Quoted 9 lines
Quoted 8 lines
The driver already warns if LDB and PCLK don't match, because this is a hardware constraint and violating this will lead to no image or image distortion. With the patch the driver tries to fix that.
I don't know if it works for all or at least most panels, but for my
Quoted 8 lines
Quoted 3 lines
This patch is supposed to work for all panels and LVDS display bridges, but I don't think it does due to the problematic pixel clock rates I mentioned before.
To my understanding, the current driver works for nearly no panel, unless you explicitly code the correct PLL freq into the device tree. The basic problem is, that the driver does not _enforce_ the correct clock ratio although the hardware strictly _requires_ it.
My above reply applies here, too.
Quoted 5 lines
There are two options in enforcing the correct ratio: forcing the ldb clock to the right rate, down to the clock source; this is Miquel's approach. Or getting the nearest possible ldb clock with the current PLL freq and modify the pixel clock accordingly, that's my approach.
With the PLL clock rate(s) assigned in DT, the pixel clock and LDB clock are kind of deterministic.
Quoted 6 lines
Of course, never modifying the required pixel clock and still setting the correct clock ratio is the golden solution. But in most cases, this is not necessary as panels usually accept deviations from the ideal pixel clock without any problems. What doesn't work is pixel and ldb clock not having the correct ratio. And this is what the patch is about.
As I said before, this patch is supposed to work for all panels and LVDS display bridges, but I don't think it does. You may try to change your single-link LVDS panels to virtual dual-link LVDS panels in DT by adding the second LVDS link. I don't think the pixel clock rate and/or LDB clock rate after fixup would work for the virtual panels.
Looking into your patch more carefully, it blindly fixes the pixel clock rate up, which means the PLL clock rate might be changed when LCDIF driver sets the fixed-up pixel clock rate because the pixel clock (IMX8MP_CLK_MEDIA_DISP2_PIX and IMX8MP_CLK_MEDIA_DISP2_PIX_ROOT) has the CLK_SET_RATE_PARENT flag set. That means the previous rounded LDB clock rate gotten from clk_round_rate() function call against fsl_ldb->clk might works no more.
Note that the mode_fixup/atomic_check callbacks should validate every atomic commit correctly. Once the commit passes the check successfully, the commit should just work. But, it looks like your patch might fail to filter out invalid atomic commits due to the clock constraints.
Quoted 34 lines
Quoted 33 lines
Quoted 33 lines
panels
it does. And if the clocks match, by chance or by refconfiguring the PLL base
clock, the patch does nothing.Quoted 26 lines
Quoted 26 lines
drivers/gpu/drm/bridge/fsl-ldb.c +13
-0+ +if (adjusted_mode->clock != pclk) { + dev_warn(fsl_ldb->dev, "Adjusted pixel clk to match LDB clk (%d kHz -> %d kHz)!\n", + adjusted_mode->clock, pclk); + + adjusted_mode->clock = pclk; + adjusted_mode->crtc_clock = pclk; + } + } + +return true; +} +static void fsl_ldb_atomic_enable(struct
drm_bridge*bridge,
structdrm_bridge_state*old_bridge_state)
{drivers/gpu/drm/bridge/fsl-ldb.c +1
-0@@ -280,6 +311,7 @@ fsl_ldb_mode_valid(struct drm_bridge *bridge,static const struct drm_bridge_funcs funcs = { .attach = fsl_ldb_attach, + .mode_fixup = fsl_ldb_mode_fixup, .atomic_enable = fsl_ldb_atomic_enable, .atomic_disable = fsl_ldb_atomic_disable, .atomic_duplicate_state = drm_atomic_helper_bridge_duplicate_state,
Signature · 2 lines
Regards, Liu Ying
Am Montag, dem 02.12.2024 um 11:13 +0100 schrieb Nikolaus Voss:
Quoted 138 lines
Hi Liu,
Quoted 7 lines · Liu Ying
Quoted 6 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
Quoted 7 lines
Quoted 4 lines
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>It looks like stable is not effectively Cc'ed.
Need a Fixes tag?I will include
Fixes 463db5c2ed4a ("drm: bridge: ldb: Implement simple Freescalei.MX8MP LDB bridge")
in v2.Quoted 54 lines
Quoted 40 lines
Signed-off-by Nikolaus Voss <[email protected]>---
drivers/gpu/drm/bridge/fsl-ldb.c | 40
++++++++++++++++++++++++++++----1 file changed, 36 insertions(+), 4 deletions(-)drivers/gpu/drm/bridge/fsl-ldb.c +1
-0@@ -104,12 +104,14 @@ static inline struct fsl_ldb *to_fsl_ldb(struct drm_bridge *bridge) return container_of(bridge, struct fsl_ldb, bridge); } +static unsigned int fsl_ldb_link_freq_factor(const struct
fsl_ldb*fsl_ldb)drivers/gpu/drm/bridge/fsl-ldb.c +4
-0+{ + return fsl_ldb_is_dual(fsl_ldb) ? 3500 : 7000; +} +static unsigned long fsl_ldb_link_frequency(struct
fsl_ldb*fsl_ldb,
int clock)
{
- if (fsl_ldb_is_dual(fsl_ldb))
- return clock * 3500;
- else
- return clock * 7000;
+ return clock *fsl_ldb_link_freq_factor(fsl_ldb); }static int fsl_ldb_attach(struct
drm_bridge*bridge,drivers/gpu/drm/bridge/fsl-ldb.c +0
-0@@ -121,6 +123,35 @@ static int fsl_ldb_attach(struct drm_bridge *bridge,drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge · line 133
Quoted 10 lines
drivers/gpu/drm/bridge/fsl-ldb.c +5
-0bridge, flags); } +static bool fsl_ldb_mode_fixup (struct drm_bridge *bridge, + const struct drm_display_mode *mode, + struct drm_display_mode *adjusted_mode) +{ + const struct fsl_ldb *fsl_ldb = to_fsl_ldb(bridge);Why const?
If no const, then ...Quoted 8 lines
drivers/gpu/drm/bridge/fsl-ldb.c static int fsl_ldb_attach(struct drm_bridge *bridge, · line 132
Quoted 2 lines
drivers/gpu/drm/bridge/fsl-ldb.c +2
-0+ unsigned long requested_link_freq = + mode->clock * fsl_ldb_link_freq_factor(fsl_ldb);... this could be
unsigned longrequested_link_freq=
fsl_ldb_link_frequency(fsl_ldb, mode->clock);I used
fsl_ldb_link_freq_factor(fsl_ldb)to point out the symmetry to recalculate
pclk = freq /fsl_ldb_link_freq_factor(fsl_ldb)below:Quoted 25 lines
Quoted 4 lines
drivers/gpu/drm/bridge/fsl-ldb.c +1
-0+ unsigned long freq = clk_round_rate(fsl_ldb->clk,
requested_link_freq);
+
+ if (freq !=requested_link_freq) {Quoted 8 lines
+ /*
+ * this will lead to flicker and incomplete lines on
+ * the attached display, adjust the CRTC clock + * accordingly.
+ */
+ int pclk = freq /fsl_ldb_link_freq_factor(fsl_ldb);I doubt that pixel clock tree cannot find appropriate division ratios
for some pixel clock rates, especially for dual-link LVDS on i.MX8MP
and i.MX93 platforms, because PLL clock rate should be 7x faster than
pixel clock rate and 2x faster than "ldb" clock rate so that the 3.5
folder between "ldb" clock and pixel clock can be met. That means the
PLL clock rate needs to be explicitly set first for this case.Quoted 6 lines
Can you assign the PLL clock rate in DT to satisfy the "ldb" and pixel
clock rates like the below commit does, if you use a LVDS panel?
4fbb73416b10("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")I probably could. The point of my patch is you don't have to know in advance which LVDS panel is connected, and you don't have to calculate
the base PLL clock by hand and store it in the device tree.In my test system, I have three different LVDS panels with EDID EEPROM,
none of which worked with the stock driver, but all work with this patch.
With slightly adapted pixel clocks though.The driver already warns if LDB and PCLK don't match, because this is a
hardware constraint and violating this will lead to no image or image distortion. With the patch the driver tries to fix that.
Quoted 6 lines
I don't know if it works for all or at least most panels, but for my panels
it does. And if the clocks match, by chance or by refconfiguring the PLL
base
clock, the patch does nothing.
just as a data point, I'm very happy with this patch as it make my 51,2 MHz lvds panel just work without doing anything else. I'll happily carry it along for v6.12 if it's not going into stable :)
martin
Quoted 7 lines · Liu Ying
Quoted 6 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
Quoted 7 lines
Quoted 4 lines
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>It looks like stable is not effectively Cc'ed.
Need a Fixes tag?
Isn't this fix effectively superseded by series
[PATCH 0/5] clk: Fix simple video pipelines on i.MX8
?
Quoted 13 lines · Marek Vasut
Quoted 12 lines · Liu Ying
Quoted 10 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>
Quoted 7 lines
Quoted 2 lines
It looks like stable is not effectively Cc'ed.
Need a Fixes tag?Isn't this fix effectively superseded by series
[PATCH 0/5] clk: Fix simple video pipelines on i.MX8
?
Maybe. I wasn't aware of the series. Looking at it, the change is rather complex and not suitable for the stable series.
My intention was to get a simple fix which doesn't potentially break anything. It wouldn't even break the patch series you mentioned.
Signature · 1 line
Nikolaus Voss
Quoted 20 lines · Nikolaus Voss
Quoted 19 lines · Marek Vasut
Quoted 7 lines · Liu Ying
Quoted 6 lines · Nikolaus Voss
LDB clock has to be a fixed multiple of the pixel clock. As LDB and pixel clock are derived from different clock sources (at least on imx8mp), this constraint cannot be satisfied for any pixel clock, which leads to flickering and incomplete lines on the attached display.
Quoted 7 lines
Quoted 4 lines
To overcome this, check this condition in
mode_fixup()and adapt the pixel clock accordingly.Cc <[email protected]>It looks like stable is not effectively Cc'ed.
Need a Fixes tag?Isn't this fix effectively superseded by series
[PATCH 0/5] clk: Fix simple video pipelines on i.MX8
Quoted 7 lines
Quoted 1 line
?
Maybe. I wasn't aware of the series. Looking at it, the change is rather complex and not suitable for the stable series.
My intention was to get a simple fix which doesn't potentially break anything. It wouldn't even break the patch series you mentioned.
I know, the proper fix is indeed complex and not yet fully figured out.
The fix for stable for existing hardware is similar to this commit I think:
4fbb73416b10 ("arm64: dts: imx8mp-phyboard-pollux: Set Video PLL1 frequency to 506.8 MHz")