| Message ID | 20191129185836.2789-3-hdegoede@redhat.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | drm/i915/vlv_dsi: Control panel and backlight enable GPIOs on BYT | expand |
Hi Hans, thank you for your patch! On Fri, Nov 29, 2019 at 7:58 PM Hans de Goede <hdegoede@redhat.com> wrote: > - /* GPIO Desc for CRC based Panel control */ > + /* GPIO Desc for panel and backlight control */ > struct gpio_desc *gpio_panel; > + struct gpio_desc *gpio_backlight; I think what happens here is that you reimplement drivers/video/backlight/gpio_backlight.c The existing power control GPIO also reimplements drivers/regulator/fixed.c in a sense but I am under the impression that x86 in general and ACPI in particular has a problem with the regulator subsystem so I am uncertain about that one. When I look at the code I get more confused because nominally panels should have their own drivers in drivers/gpu/drm/panel/* especially DSI panels, and the panels control their own GPIOs or regulators for power and backlight. I was under the impression that Heikki's and Dmitry's recent additions to software nodes would make it possible to actually spawn devices like the GPIO backlight and/or fixed regulator and put references to them so that the drivers can pick them up from the respective frameworks and manage them? Maybe I misunderstood things here though, I am a bit under the impression that elder DRM drivers are considered "elder gods" and do not need to use separate panel drivers, backlight abstraction etc, and in that case just go ahead, I guess. But I suspect some separation would help the day the i915 driver wants to reuse some really complex DSI panel from drivers/gpu/drm/panel/* though. Yours, Linus Walleij
On Mon, 02 Dec 2019, Linus Walleij <linus.walleij@linaro.org> wrote: > Hi Hans, > > thank you for your patch! > > On Fri, Nov 29, 2019 at 7:58 PM Hans de Goede <hdegoede@redhat.com> wrote: > >> - /* GPIO Desc for CRC based Panel control */ >> + /* GPIO Desc for panel and backlight control */ >> struct gpio_desc *gpio_panel; >> + struct gpio_desc *gpio_backlight; > > I think what happens here is that you reimplement > drivers/video/backlight/gpio_backlight.c > > The existing power control GPIO also reimplements > drivers/regulator/fixed.c in a sense but I am under the > impression that x86 in general and ACPI in particular > has a problem with the regulator subsystem so I am > uncertain about that one. > > When I look at the code I get more confused because > nominally panels should have their own drivers in > drivers/gpu/drm/panel/* especially DSI panels, and the > panels control their own GPIOs or regulators for power > and backlight. > > I was under the impression that Heikki's and Dmitry's > recent additions to software nodes would make it > possible to actually spawn devices like the GPIO backlight > and/or fixed regulator and put references to them so > that the drivers can pick them up from the respective > frameworks and manage them? > > Maybe I misunderstood things here though, I am a bit > under the impression that elder DRM drivers are > considered "elder gods" and do not need to use separate > panel drivers, backlight abstraction etc, and in that > case just go ahead, I guess. > > But I suspect some separation > would help the day the i915 driver wants to reuse some > really complex DSI panel from drivers/gpu/drm/panel/* > though. Sadly I don't think that's going to happen, though. For i915 the panels are described by VBT, or Video BIOS Tables. We don't really know the make or model of the panels, we just get the timings and sequences etc. from there. There's no info to probe a separate panel driver. So we just have a "generic" DSI driver that's part of i915 which uses the data from the VBT to drive all the possible panels. We do use some of the drm DSI framework as a library to abstract things a bit better, but that's about it. The sequences should include details about GPIOs to toggle as well, so in that sense I feel like the gpio calls should be bolted there. BR, Jani.
HI, On 02-12-2019 12:53, Jani Nikula wrote: > On Mon, 02 Dec 2019, Linus Walleij <linus.walleij@linaro.org> wrote: >> Hi Hans, >> >> thank you for your patch! >> >> On Fri, Nov 29, 2019 at 7:58 PM Hans de Goede <hdegoede@redhat.com> wrote: >> >>> - /* GPIO Desc for CRC based Panel control */ >>> + /* GPIO Desc for panel and backlight control */ >>> struct gpio_desc *gpio_panel; >>> + struct gpio_desc *gpio_backlight; >> >> I think what happens here is that you reimplement >> drivers/video/backlight/gpio_backlight.c >> >> The existing power control GPIO also reimplements >> drivers/regulator/fixed.c in a sense but I am under the >> impression that x86 in general and ACPI in particular >> has a problem with the regulator subsystem so I am >> uncertain about that one. >> >> When I look at the code I get more confused because >> nominally panels should have their own drivers in >> drivers/gpu/drm/panel/* especially DSI panels, and the >> panels control their own GPIOs or regulators for power >> and backlight. >> >> I was under the impression that Heikki's and Dmitry's >> recent additions to software nodes would make it >> possible to actually spawn devices like the GPIO backlight >> and/or fixed regulator and put references to them so >> that the drivers can pick them up from the respective >> frameworks and manage them? >> >> Maybe I misunderstood things here though, I am a bit >> under the impression that elder DRM drivers are >> considered "elder gods" and do not need to use separate >> panel drivers, backlight abstraction etc, and in that >> case just go ahead, I guess. >> >> But I suspect some separation >> would help the day the i915 driver wants to reuse some >> really complex DSI panel from drivers/gpu/drm/panel/* >> though. > > Sadly I don't think that's going to happen, though. > > For i915 the panels are described by VBT, or Video BIOS Tables. We don't > really know the make or model of the panels, we just get the timings and > sequences etc. from there. There's no info to probe a separate panel > driver. So we just have a "generic" DSI driver that's part of i915 which > uses the data from the VBT to drive all the possible panels. > > We do use some of the drm DSI framework as a library to abstract things > a bit better, but that's about it. > > The sequences should include details about GPIOs to toggle as well, so > in that sense I feel like the gpio calls should be bolted there. Hmm, yes that makes sense, the problem really is that the VBT sequences for DSI panels on BYT are incomplete (*) and the MIPI_SEQ_POWER_ON / _OFF and MIPI_SEQ_BACKLIGHT_ON / _OFF sequences do not containt the "instructions" to toggle the panel-enable, resp. backlight enable GPIOs as they do on CHT and newer devices. So if we see all this of a shortcoming of the VBT tables, then it makes sense to move all special handling for this to intel_dsi_vbt.c, so we put the adding of the lookup table, the getting of the pins and the controlling of the pins all in there. Jani, is that what you have in mind ? This would also solve the problem of Linus and Andy not liking the polution of the pinctrl driver with the adding of the platform specific lookups; and then we could only add the lookups when the VBT says the SoC is used for backlight control, leaving things as is (not even adding the lookups) when the PMIC is used for backlight control. There is only one problem, currently is is not possible to unregister a mapping added with pinctrl_register_mappings and the i915 driver is typically a module which can be unloaded and I believe actually is unloaded as part of the i915 CI. pinctrl_register_mappings copies the passed in mapping, but it is a shallow copy, so it contains pointers to the modules const segment and we do not want to re-add another copy of the mapping when the module loads a second time. Fixing this is easy though, there already is a pinctrl_unregister_map() function, we just need to export it so that the i915 driver can remove the mapping when it is unbound. Linus would exporting this function be ok with you? Linus, question what is the purpose of the "dupping" / shallow copying of the argument passed to pinctrl_register_map ? Since it is shallow the mem for any pointers contained within there need to be kept around by the caller, so why not let the caller keep the pinctrl_map struct itself around too? If we are going to export pinctrl_unregister_map() we need to make it do the right thing for dupped maps too, we can just store the dup flag in struct pinctrl_maps. So this is easy, but I wonder if we cannot get rid of the dupping all together ? Regards, Hans *) I suspect these are all version 2 VBT sequences (I guess I should check this) and according to the comment in vlv_dsi.c which documents the sequences, for v2 these sequences are not used, e.g. instead of executing the MIPIPanelPowerOn sequence the "panel-on" step is described simply as "power on". and the MIPIBacklightOn/Off sequences are described as being VBT v3+ only
On Mon, Dec 2, 2019 at 4:49 PM Hans de Goede <hdegoede@redhat.com> wrote: > There is only one problem, currently is is not possible to > unregister a mapping added with pinctrl_register_mappings > and the i915 driver is typically a module which can be unloaded > and I believe actually is unloaded as part of the i915 CI. > > pinctrl_register_mappings copies the passed in mapping, but > it is a shallow copy, so it contains pointers to the modules > const segment and we do not want to re-add another copy of > the mapping when the module loads a second time. > > Fixing this is easy though, there already is a pinctrl_unregister_map() > function, we just need to export it so that the i915 driver can > remove the mapping when it is unbound. > > Linus would exporting this function be ok with you? Yep! > Linus, question what is the purpose of the "dupping" / shallow > copying of the argument passed to pinctrl_register_map ? The initial commit contained this comment later removed: + /* + * Make a copy of the map array - string pointers will end up in the + * kernel const section anyway so these do not need to be deep copied. + */ The use was to free up memory for platforms using boardfiles with a gazillion variants and huge pin control tables, so these could be marked __initdata and discarded after boot. As the strings would anyway stay around we didn't need to deep copy. See for example in arch/arm/mach-u300/core.c static struct pinctrl_map __initdata u300_pinmux_map[] > Since > it is shallow the mem for any pointers contained within there need > to be kept around by the caller, so why not let the caller keep > the pinctrl_map struct itself around too? So the strings will be kept around because the kernel can't get rid of strings. (Yeah it is silly, should haven been fixed ages ago, but not by me, haha :) > If we are going to export pinctrl_unregister_map() we need to make it > do the right thing for dupped maps too, we can just store the dup flag > in struct pinctrl_maps. So this is easy, but I wonder if we cannot > get rid of the dupping all together ? Maybe ... I don't know. What do you think? I suppose you could make u300 crash if you do that. Yours, Linus Walleij
Hi, On 11-12-2019 01:24, Linus Walleij wrote: > On Mon, Dec 2, 2019 at 4:49 PM Hans de Goede <hdegoede@redhat.com> wrote: > >> There is only one problem, currently is is not possible to >> unregister a mapping added with pinctrl_register_mappings >> and the i915 driver is typically a module which can be unloaded >> and I believe actually is unloaded as part of the i915 CI. >> >> pinctrl_register_mappings copies the passed in mapping, but >> it is a shallow copy, so it contains pointers to the modules >> const segment and we do not want to re-add another copy of >> the mapping when the module loads a second time. >> >> Fixing this is easy though, there already is a pinctrl_unregister_map() >> function, we just need to export it so that the i915 driver can >> remove the mapping when it is unbound. >> >> Linus would exporting this function be ok with you? > > Yep! > >> Linus, question what is the purpose of the "dupping" / shallow >> copying of the argument passed to pinctrl_register_map ? > > The initial commit contained this comment later removed: > > + /* > + * Make a copy of the map array - string pointers will end up in the > + * kernel const section anyway so these do not need to be deep copied. > + */ > > The use was to free up memory for platforms using boardfiles > with a gazillion variants and huge pin control tables, so these > could be marked __initdata and discarded after boot. > As the strings would anyway stay around we didn't need to > deep copy. > > See for example in arch/arm/mach-u300/core.c > static struct pinctrl_map __initdata u300_pinmux_map[] > >> Since >> it is shallow the mem for any pointers contained within there need >> to be kept around by the caller, so why not let the caller keep >> the pinctrl_map struct itself around too? > > So the strings will be kept around because the kernel can't get > rid of strings. (Yeah it is silly, should haven been fixed ages > ago, but not by me, haha :) > >> If we are going to export pinctrl_unregister_map() we need to make it >> do the right thing for dupped maps too, we can just store the dup flag >> in struct pinctrl_maps. So this is easy, but I wonder if we cannot >> get rid of the dupping all together ? > > Maybe ... I don't know. What do you think? I suppose you could > make u300 crash if you do that. I've prepared a patch which makes pinctrl_register_mappings remember if the mapping is dupped or not (store the dup value in struct pinctrl_maps); and which modifies pinctrl_unregister_map() to do the right thing depending on the stored dup value. I still need to test the new series and then I will post it. Regards, Hans
diff --git a/drivers/gpu/drm/i915/display/intel_dsi.h b/drivers/gpu/drm/i915/display/intel_dsi.h index b15be5814599..6ff1b59b0f6f 100644 --- a/drivers/gpu/drm/i915/display/intel_dsi.h +++ b/drivers/gpu/drm/i915/display/intel_dsi.h @@ -45,8 +45,9 @@ struct intel_dsi { struct intel_dsi_host *dsi_hosts[I915_MAX_PORTS]; intel_wakeref_t io_wakeref[I915_MAX_PORTS]; - /* GPIO Desc for CRC based Panel control */ + /* GPIO Desc for panel and backlight control */ struct gpio_desc *gpio_panel; + struct gpio_desc *gpio_backlight; struct intel_connector *attached_connector; diff --git a/drivers/gpu/drm/i915/display/vlv_dsi.c b/drivers/gpu/drm/i915/display/vlv_dsi.c index fc63e2f6a511..bd081a3bb54b 100644 --- a/drivers/gpu/drm/i915/display/vlv_dsi.c +++ b/drivers/gpu/drm/i915/display/vlv_dsi.c @@ -24,6 +24,7 @@ */ #include <linux/gpio/consumer.h> +#include <linux/pinctrl/consumer.h> #include <linux/slab.h> #include <drm/drm_atomic_helper.h> @@ -797,7 +798,7 @@ static void intel_dsi_pre_enable(struct intel_encoder *encoder, if (!IS_GEMINILAKE(dev_priv)) intel_dsi_prepare(encoder, pipe_config); - /* Power on, try both CRC pmic gpio and VBT */ + /* Power on, try both the panel-enable GPIO and the MIPI seq. */ if (intel_dsi->gpio_panel) gpiod_set_value_cansleep(intel_dsi->gpio_panel, 1); intel_dsi_vbt_exec_sequence(intel_dsi, MIPI_SEQ_POWER_ON); @@ -843,6 +844,9 @@ static void intel_dsi_pre_enable(struct intel_encoder *encoder, } intel_panel_enable_backlight(pipe_config, conn_state); + /* Backlight on, try both the backlight-enable GPIO and the MIPI seq. */ + if (intel_dsi->gpio_backlight) + gpiod_set_value_cansleep(intel_dsi->gpio_backlight, 1); intel_dsi_vbt_exec_sequence(intel_dsi, MIPI_SEQ_BACKLIGHT_ON); } @@ -859,7 +863,10 @@ static void intel_dsi_disable(struct intel_encoder *encoder, DRM_DEBUG_KMS("\n"); + /* Backlight off, try both the enable GPIO and the MIPI seq. */ intel_dsi_vbt_exec_sequence(intel_dsi, MIPI_SEQ_BACKLIGHT_OFF); + if (intel_dsi->gpio_backlight) + gpiod_set_value_cansleep(intel_dsi->gpio_backlight, 0); intel_panel_disable_backlight(old_conn_state); /* @@ -943,7 +950,7 @@ static void intel_dsi_post_disable(struct intel_encoder *encoder, /* Assert reset */ intel_dsi_vbt_exec_sequence(intel_dsi, MIPI_SEQ_ASSERT_RESET); - /* Power off, try both CRC pmic gpio and VBT */ + /* Power off, try both the panel-enable GPIO and the MIPI seq. */ intel_dsi_msleep(intel_dsi, intel_dsi->panel_off_delay); intel_dsi_vbt_exec_sequence(intel_dsi, MIPI_SEQ_POWER_OFF); if (intel_dsi->gpio_panel) @@ -1542,6 +1549,8 @@ static void intel_dsi_encoder_destroy(struct drm_encoder *encoder) /* dispose of the gpios */ if (intel_dsi->gpio_panel) gpiod_put(intel_dsi->gpio_panel); + if (intel_dsi->gpio_backlight) + gpiod_put(intel_dsi->gpio_backlight); intel_encoder_destroy(encoder); } @@ -1821,6 +1830,7 @@ void vlv_dsi_init(struct drm_i915_private *dev_priv) struct intel_connector *intel_connector; struct drm_connector *connector; struct drm_display_mode *current_mode, *fixed_mode; + struct pinctrl *pinctrl; enum port port; DRM_DEBUG_KMS("\n"); @@ -1921,7 +1931,7 @@ void vlv_dsi_init(struct drm_i915_private *dev_priv) vlv_dphy_param_init(intel_dsi); /* - * In case of BYT with CRC PMIC, we need to use GPIO for + * In case of BYT or CHT with CRC PMIC, we need to use GPIO for * Panel control. */ if ((IS_VALLEYVIEW(dev_priv) || IS_CHERRYVIEW(dev_priv)) && @@ -1934,6 +1944,34 @@ void vlv_dsi_init(struct drm_i915_private *dev_priv) intel_dsi->gpio_panel = NULL; } } + /* + * In case of BYT (and only BYT) and using the SoC for PWM, we need + * to use the SoC's GPIOs for panel and backlight enable. When the + * GOP did not initialize the panel (HDMI inserted) we may need to + * also change the pinmux for the SoC's PWM0 pin from GPIO to PWM. + */ + if (IS_VALLEYVIEW(dev_priv) && + dev_priv->vbt.dsi.config->pwm_blc == PPS_BLC_SOC) { + intel_dsi->gpio_panel = gpiod_get(dev->dev, + "soc_panel_enable", + GPIOD_OUT_HIGH); + if (IS_ERR(intel_dsi->gpio_panel)) { + DRM_ERROR("Failed to own gpio for panel control\n"); + intel_dsi->gpio_panel = NULL; + } + + intel_dsi->gpio_backlight = gpiod_get(dev->dev, + "soc_backlight_enable", + GPIOD_OUT_HIGH); + if (IS_ERR(intel_dsi->gpio_backlight)) { + DRM_ERROR("Failed to own gpio for backlight control\n"); + intel_dsi->gpio_backlight = NULL; + } + + pinctrl = devm_pinctrl_get_select(dev->dev, "soc_pwm0"); + if (IS_ERR(pinctrl)) + DRM_ERROR("Failed to set pinmux to PWM\n"); + } drm_connector_init(dev, connector, &intel_dsi_connector_funcs, DRM_MODE_CONNECTOR_DSI);
On Bay Trail devices the MIPI power on/off sequences for DSI LCD panels do not control the LCD panel- and backlight-enable GPIOs. So far, when the VBT indicates we should use the SoC for backlight control, we have been relying on these GPIOs being configured as output and driven high by the Video BIOS (GOP) when it initializes the panel. This does not work when the device is booted with a HDMI monitor connected as then the GOP will initialize the HDMI instead of the panel, leaving the panel black, even though the i915 driver tries to output an image to it. Likewise on some device-models when the GOP does not initialize the DSI panel it also leaves the mux of the PWM0 pin in generic GPIO mode instead of muxing it to the PWM controller. This commit makes the DSI code control the SoC GPIOs for panel- and backlight-enable on BYT, when the VBT indicates the SoC should be used for backlight control. It also ensures that the PWM0 pin is muxed to the PWM controller in this case. This fixes the LCD panel not lighting up on various devices when booted with a HDMI monitor connected. This has been tested to fix this on the following devices: Peaq C1010 Point of View MOBII TAB-P800W Point of View MOBII TAB-P1005W Terra Pad 1061 Yours Y8W81 Signed-off-by: Hans de Goede <hdegoede@redhat.com> --- drivers/gpu/drm/i915/display/intel_dsi.h | 3 +- drivers/gpu/drm/i915/display/vlv_dsi.c | 44 ++++++++++++++++++++++-- 2 files changed, 43 insertions(+), 4 deletions(-)