OnePlus 6 fixups
Subsystems (7)
This series contains a few fixes for the OnePlus 6 and 6T.
The display panel driver for the devices currently implements support for the reset pin.
This behaviour is fine on the 6 but on the 6T it seems to break the panel. This suggests an issue with the initialisation sequence however we haven't been able to find a way to bring up the panel without this. In the mean time, removing the reset capabilities solves the issues for us and doesn't introduce any other problems.
We also add a guard around the RMTFS memory region, this is implemented by the msm_sharedmem driver in the vendor kernel, but not a feature supported by mainline just yet. This is to workaround limitations in Qualcomms XPU hardware which can trigger false violations and crash the device if memory is allocated close to the RMTFS region.
Finally, enable the IPA node, modemmanager is gaining support for QRTR modems and with that we can have mobile data working on the device!
^Caleb
PATCH 1/4 arm64: dts: qcom: sdm845-oneplus-common: remove panel reset gpio
Resetting the panel causes issues on fajita, it is also completely unnecessary for normal use so lets just not bother.
--- arch/arm64/boot/dts/qcom/sdm845-oneplus-common.dtsi | 2 -- 1 file changed, 2 deletions(-)
arch/arm64/boot/dts/qcom/sdm845-oneplus-common.dtsi +0 -2
@@ -314,8 +314,6 @@ display_panel: panel@0 { vddio-supply = <&vreg_l14a_1p88>; - reset-gpios = <&tlmm 6 GPIO_ACTIVE_LOW>; - pinctrl-names = "default"; pinctrl-0 = <&panel_reset_pins &panel_te_pin &panel_esd_pin>;
Signature · 1 line
2.30.2
PATCH 2/4 drm: panel: sofef00: remove reset GPIO handling
In reply to Caleb ConnollyResetting the panel on fajita causes it to never come back, we aren't quite sure why this is so for now lets remove reset handling as it is effectively broken. It is also not needed on enchilada.
--- drivers/gpu/drm/panel/panel-samsung-sofef00.c | 26 +++---------------- 1 file changed, 4 insertions(+), 22 deletions(-)
drivers/gpu/drm/panel/panel-samsung-sofef00.c +4 -22
@@ -23,7 +23,6 @@ struct sofef00_panel {struct drm_panel panel;struct mipi_dsi_device *dsi;struct regulator *supply; -struct gpio_desc *reset_gpio;const struct drm_display_mode *mode;bool prepared; };
@@ -42,16 +41,6 @@ struct sofef00_panel *to_sofef00_panel(struct drm_panel *panel)return ret; \ }while (0 ) -static void sofef00_panel_reset(struct sofef00_panel *ctx) -{ -gpiod_set_value_cansleep (ctx->reset_gpio,0 ); -usleep_range (5000 ,6000 ); -gpiod_set_value_cansleep (ctx->reset_gpio,1 ); -usleep_range (2000 ,3000 ); -gpiod_set_value_cansleep (ctx->reset_gpio,0 ); -usleep_range (12000 ,13000 ); -} - static int sofef00_panel_on(struct sofef00_panel *ctx) {struct mipi_dsi_device *dsi = ctx->dsi;
@@ -132,12 +121,9 @@ static int sofef00_panel_prepare(struct drm_panel *panel)return ret; } -sofef00_panel_reset (ctx); - ret =sofef00_panel_on (ctx);if (ret <0 ) {dev_err (dev,"Failed to initialize panel: %d\n" , ret); -gpiod_set_value_cansleep (ctx->reset_gpio,1 );return ret; }
@@ -155,8 +141,11 @@ static int sofef00_panel_unprepare(struct drm_panel *panel)return 0 ; ret =sofef00_panel_off (ctx); -if (ret <0 ) + +if (ret <0 ) {dev_err (dev,"Failed to un-initialize panel: %d\n" , ret); +return ret; + }regulator_disable (ctx->supply);
@@ -276,13 +265,6 @@ static int sofef00_panel_probe(struct mipi_dsi_device *dsi)return ret; } - ctx->reset_gpio =devm_gpiod_get (dev,"reset" ,GPIOD_OUT_HIGH ); -if (IS_ERR (ctx->reset_gpio)) { - ret =PTR_ERR (ctx->reset_gpio); -dev_warn (dev,"Failed to get reset-gpios: %d\n" , ret); -return ret; - } - ctx->dsi = dsi;mipi_dsi_set_drvdata (dsi, ctx);
Signature · 1 line
2.30.2
Re: [PATCH 2/4] drm: panel: sofef00: remove reset GPIO handling
Quoted 52 lines · Caleb Connolly
Resetting the panel on fajita causes it to never come back, we aren't quite sure why this is so for now lets remove reset handling as it is effectively broken. It is also not needed on enchilada.
Signed-off-by Caleb Connolly <[email protected]>--- drivers/gpu/drm/panel/panel-samsung-sofef00.c | 26 +++---------------- 1 file changed, 4 insertions(+), 22 deletions(-)drivers/gpu/drm/panel/panel-samsung-sofef00.c +0
-14@@ -23,7 +23,6 @@ struct sofef00_panel {struct drm_panel panel;struct mipi_dsi_device *dsi;struct regulator *supply; -struct gpio_desc *reset_gpio;const struct drm_display_mode *mode;bool prepared; };@@ -42,16 +41,6 @@ struct sofef00_panel *to_sofef00_panel(struct drm_panel *panel)return ret; \ }while (0 ) -static void sofef00_panel_reset(struct sofef00_panel *ctx) -{ -gpiod_set_value_cansleep (ctx->reset_gpio,0 ); -usleep_range (5000 ,6000 ); -gpiod_set_value_cansleep (ctx->reset_gpio,1 ); -usleep_range (2000 ,3000 ); -gpiod_set_value_cansleep (ctx->reset_gpio,0 ); -usleep_range (12000 ,13000 ); -} - static int sofef00_panel_on(struct sofef00_panel *ctx) {struct mipi_dsi_device *dsi = ctx->dsi;@@ -132,12 +121,9 @@ static int sofef00_panel_prepare(struct drm_panel *panel)return ret; } -sofef00_panel_reset (ctx); - ret =sofef00_panel_on (ctx);if (ret <0 ) {dev_err (dev,"Failed to initialize panel: %d\n" , ret); -gpiod_set_value_cansleep (ctx->reset_gpio,1 );return ret; }
drivers/gpu/drm/panel/panel-samsung-sofef00.c static int sofef00_panel_unprepare(struct drm_panel *panel) · line 147
Quoted 8 lines
drivers/gpu/drm/panel/panel-samsung-sofef00.c +3
-1@@ -155,8 +141,11 @@ static int sofef00_panel_unprepare(struct drm_panel *panel)return 0 ; ret =sofef00_panel_off (ctx); -if (ret <0 ) + +if (ret <0 ) {dev_err (dev,"Failed to un-initialize panel: %d\n" , ret); +return ret;
This early return seems unrelated to what's described in the commit message.
Also as a general comment, what do you expect the software should do if you return an error in the unprepare path? Can we if this happens just continue to display stuff on the display?
drivers/gpu/drm/panel/panel-samsung-sofef00.c static int sofef00_panel_probe(struct mipi_dsi_device *dsi) · line 279
Quoted 9 lines
drivers/gpu/drm/panel/panel-samsung-sofef00.c +1
-1+ }regulator_disable (ctx->supply);@@ -276,13 +265,6 @@ static int sofef00_panel_probe(struct mipi_dsi_device *dsi)return ret; } - ctx->reset_gpio =devm_gpiod_get (dev,"reset" ,GPIOD_OUT_HIGH );
If I understand your description in the cover letter this works for one of the devices using the panel, but not the other.
So how about using devm_gpiod_get_optional() instead. That will give you NULL back if the property isn't defined and make all the gpiod operations nops.
Regards,
Bjorn
Quoted 13 lines
drivers/gpu/drm/panel/panel-samsung-sofef00.c +0
-6-if (IS_ERR (ctx->reset_gpio)) { - ret =PTR_ERR (ctx->reset_gpio); -dev_warn (dev,"Failed to get reset-gpios: %d\n" , ret); -return ret; - } - ctx->dsi = dsi;mipi_dsi_set_drvdata (dsi, ctx);Signature · 1 line
2.30.2
Re: [PATCH 2/4] drm: panel: sofef00: remove reset GPIO handling
In reply to Caleb ConnollyQuoted 3 lines · Caleb Connolly
Resetting the panel on fajita causes it to never come back, we aren't quite sure why this is so for now lets remove reset handling as it is effectively broken. It is also not needed on enchilada.
Where are the device trees for these platforms?
Are there schematics so we/someone with access to the schematics can figure it out?
It's not something trivial like forgotten to put GPIO_ACTIVE_LOW on the gpio phandle?
drivers/gpu/drm/panel/panel-samsung-sofef00.c struct sofef00_panel *to_sofef00_panel(struct drm_panel *panel) · line 53
Quoted 9 lines
drivers/gpu/drm/panel/panel-samsung-sofef00.c +0
-9-static void sofef00_panel_reset (struct sofef00_panel *ctx) -{ -gpiod_set_value_cansleep (ctx->reset_gpio,0 ); -usleep_range (5000 ,6000 ); -gpiod_set_value_cansleep (ctx->reset_gpio,1 ); -usleep_range (2000 ,3000 ); -gpiod_set_value_cansleep (ctx->reset_gpio,0 ); -usleep_range (12000 ,13000 ); -}
This code looks a bit weird to begin with, I don't see the point with the first setting the gpiod low, either it was low already or high already neither matters, driving it asserted will assert it either way.
No big deal though.
Quoted 1 line
-
ctx->reset_gpio=devm_gpiod_get(dev, "reset", GPIOD_OUT_HIGH);
As Bjorn says: just use *_optional and comment the line out in the device tree if there is a problem.
Yours,
Linus Walleij