Skip to contentSkip to conversation

drm: mxsfb: fix runtime PM handling

PATCH In review Anson Huang 4 participants 9 messages Jul 13, 2018

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm: mxsfb: fix runtime PM handling

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

---
 drivers/gpu/drm/mxsfb/mxsfb_drv.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/gpu/drm/mxsfb/mxsfb_drv.c +1 -2
index ffe5137..1ba179b 100644
@@ -235,7 +235,6 @@ static int mxsfb_load(struct drm_device *drm, unsigned long flags)
 
 	pm_runtime_get_sync(drm->dev);
 	ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
-	pm_runtime_put_sync(drm->dev);
 
 	if (ret < 0) {
 		dev_err(drm->dev, "Failed to install IRQ handler\n");
@@ -264,6 +263,7 @@ static int mxsfb_load(struct drm_device *drm, unsigned long flags)
 err_irq:
 	drm_panel_detach(mxsfb->panel);
 err_vblank:
+	pm_runtime_put_sync(drm->dev);
 	pm_runtime_disable(drm->dev);
 
 	return ret;
@@ -279,7 +279,6 @@ static void mxsfb_unload(struct drm_device *drm)
 	drm_kms_helper_poll_fini(drm);
 	drm_mode_config_cleanup(drm);
 
-	pm_runtime_get_sync(drm->dev);
 	drm_irq_uninstall(drm);
 	pm_runtime_put_sync(drm->dev);
 
Signature · 1 line
2.7.4
maintainer
#1 lore raw
Quoted 14 lines · Anson Huang

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

Quoted 33 lines
---
 drivers/gpu/drm/mxsfb/mxsfb_drv.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/gpu/drm/mxsfb/mxsfb_drv.c +1 -2
index ffe5137..1ba179b 100644
@@ -235,7 +235,6 @@ static int mxsfb_load(struct drm_device *drm, unsigned long flags)
 
 	pm_runtime_get_sync(drm->dev);
 	ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
-	pm_runtime_put_sync(drm->dev);
 
 	if (ret < 0) {
 		dev_err(drm->dev, "Failed to install IRQ handler\n");
@@ -264,6 +263,7 @@ static int mxsfb_load(struct drm_device *drm, unsigned long flags)
 err_irq:
 	drm_panel_detach(mxsfb->panel);
 err_vblank:
+	pm_runtime_put_sync(drm->dev);
 	pm_runtime_disable(drm->dev);
 
 	return ret;
@@ -279,7 +279,6 @@ static void mxsfb_unload(struct drm_device *drm)
 	drm_kms_helper_poll_fini(drm);
 	drm_mode_config_cleanup(drm);
 
-	pm_runtime_get_sync(drm->dev);
 	drm_irq_uninstall(drm);
 	pm_runtime_put_sync(drm->dev);
 
Signature · 2 lines
Best regards,
Marek Vasut
#2 lore raw

Anson Huang
Best Regards!

Quoted 20 lines

-----Original Message-----
From: Marek Vasut [mailto:[email protected]]
Sent: Friday, July 13, 2018 5:03 PM
To: Anson Huang <[email protected]>; [email protected]; [email protected]; [email protected]
Cc: dl-linux-imx <[email protected]>
Subject: Re: [PATCH] drm: mxsfb: fix runtime PM handling

Quoted 11 lines · Anson Huang

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 3 lines
Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

I think drm driver should have somewhere to implement the display
disable case, like when fb0 is blank (echo 1 > /sys/class/graphics/fb0/blank), then lcdif can be powered gated, and
when display is back on (unblank), lcdif needs to be re-initialization and display will be on, current implementation is incorrect, with kernel booting up, lcdif is NOT working at all.

The runtime power gating case can be implemented later to save power, need lcdif experts to do it. Thanks.

Anson.

Quoted 42 lines
Quoted 30 lines
---
 drivers/gpu/drm/mxsfb/mxsfb_drv.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/gpu/drm/mxsfb/mxsfb_drv.c +0 -1
index ffe5137..1ba179b 100644
@@ -235,7 +235,6 @@ static int mxsfb_load(struct drm_device *drm,
 unsigned long flags)
 
 	pm_runtime_get_sync(drm->dev);
 	ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
-	pm_runtime_put_sync(drm->dev);
 
 	if (ret < 0) {

dev_err(drm->dev, "Failed to install IRQ handler\n"); @@ -264,6
+263,7 @@ static int mxsfb_load(struct drm_device *drm, unsigned long flags)
err_irq:

drivers/gpu/drm/mxsfb/mxsfb_drv.c +1 -0
 	drm_panel_detach(mxsfb->panel);
 err_vblank:
+	pm_runtime_put_sync(drm->dev);
 	pm_runtime_disable(drm->dev);
 
 	return ret;

drivers/gpu/drm/mxsfb/mxsfb_drv.c static void mxsfb_unload(struct drm_device *drm) · line 284

Quoted 7 lines
drivers/gpu/drm/mxsfb/mxsfb_drv.c +0 -1
@@ -279,7 +279,6 @@ static void mxsfb_unload(struct drm_device *drm)
 	drm_kms_helper_poll_fini(drm);
 	drm_mode_config_cleanup(drm);
 
-	pm_runtime_get_sync(drm->dev);
 	drm_irq_uninstall(drm);
 	pm_runtime_put_sync(drm->dev);
 

--
Best regards,
Marek Vasut

maintainer
#3 lore raw

On 07/13/2018 11:06 AM, Anson Huang wrote:
[...]

Quoted 13 lines
Quoted 13 lines · Anson Huang
Quoted 11 lines

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

I think drm driver should have somewhere to implement the display disable case, like when fb0 is blank (echo 1 > /sys/class/graphics/fb0/blank),

Isn't this just the fbdev emulation on top of drm/kms ? I think this stuff can be compiled out completely.

Quoted 4 lines

then lcdif can be powered gated, and
when display is back on (unblank), lcdif needs to be re-initialization and display will be on, current implementation is incorrect, with kernel booting up, lcdif is NOT working at all.

It works fine on MX6SX , so I think this is isolated to MX6SL ? I'm CCing Stefan, he might have some valuable feedback here.

Quoted 47 lines

The runtime power gating case can be implemented later to save power, need lcdif experts to do it. Thanks.

Anson.

Quoted 42 lines
Quoted 30 lines
---
 drivers/gpu/drm/mxsfb/mxsfb_drv.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/gpu/drm/mxsfb/mxsfb_drv.c +0 -1
index ffe5137..1ba179b 100644
@@ -235,7 +235,6 @@ static int mxsfb_load(struct drm_device *drm,
 unsigned long flags)
 
 	pm_runtime_get_sync(drm->dev);
 	ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
-	pm_runtime_put_sync(drm->dev);
 
 	if (ret < 0) {

dev_err(drm->dev, "Failed to install IRQ handler\n"); @@ -264,6
+263,7 @@ static int mxsfb_load(struct drm_device *drm, unsigned long flags)
err_irq:

drivers/gpu/drm/mxsfb/mxsfb_drv.c +1 -0
 	drm_panel_detach(mxsfb->panel);
 err_vblank:
+	pm_runtime_put_sync(drm->dev);
 	pm_runtime_disable(drm->dev);
 
 	return ret;

drivers/gpu/drm/mxsfb/mxsfb_drv.c static void mxsfb_unload(struct drm_device *drm) · line 284

Quoted 7 lines
drivers/gpu/drm/mxsfb/mxsfb_drv.c +0 -1
@@ -279,7 +279,6 @@ static void mxsfb_unload(struct drm_device *drm)
 	drm_kms_helper_poll_fini(drm);
 	drm_mode_config_cleanup(drm);
 
-	pm_runtime_get_sync(drm->dev);
 	drm_irq_uninstall(drm);
 	pm_runtime_put_sync(drm->dev);
 

--
Best regards,
Marek Vasut

Signature · 2 lines
Best regards,
Marek Vasut
#4 lore raw
Quoted 25 lines · Marek Vasut

On 07/13/2018 11:06 AM, Anson Huang wrote:
[...]

Quoted 13 lines
Quoted 13 lines · Anson Huang
Quoted 11 lines

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

I think drm driver should have somewhere to implement the display disable case, like when fb0 is blank (echo 1 > /sys/class/graphics/fb0/blank),

Isn't this just the fbdev emulation on top of drm/kms ? I think this stuff can be compiled out completely.

Quoted 7 lines
Quoted 4 lines

then lcdif can be powered gated, and
when display is back on (unblank), lcdif needs to be re-initialization and display will be on, current implementation is incorrect, with kernel booting up, lcdif is NOT working at all.

It works fine on MX6SX , so I think this is isolated to MX6SL ? I'm CCing Stefan, he might have some valuable feedback here.

Some time ago I sent a patch which tries to implement "proper" runtime PM for lcdif: https://patchwork.kernel.org/patch/10449761/

It seems to have slipped through the cracks, can somebody please take a look at it?

This only actually matters on SOCs where the display power domain is gated, however:

* On imx6sx the DISP power domain is currently not defined in DTS. Defining requires dealing with pcie being spread in multiple PDs.
* On imx6sl there is a nasty errata so the PD was marked always-on. This was a few days ago: https://lkml.org/lkml/2018/7/11/427
* imx6sll is new, evk dts only in next.

The lcdif block is also used in some imx8 chips so runtime PM should be dealt with eventually.

maintainer
#5 lore raw
In reply to Marek Vasut
Quoted 25 lines · Marek Vasut

On 07/13/2018 11:06 AM, Anson Huang wrote:
[...]

Quoted 13 lines
Quoted 13 lines · Anson Huang
Quoted 11 lines

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

I think drm driver should have somewhere to implement the display disable case, like when fb0 is blank (echo 1 > /sys/class/graphics/fb0/blank),

Isn't this just the fbdev emulation on top of drm/kms ? I think this stuff can be compiled out completely.

Quoted 8 lines
Quoted 4 lines

then lcdif can be powered gated, and
when display is back on (unblank), lcdif needs to be re-initialization and display will be on, current implementation is incorrect, with kernel booting up, lcdif is NOT working at all.

It works fine on MX6SX , so I think this is isolated to MX6SL ? I'm CCing Stefan, he might have some valuable feedback here.

Yeah not sure, but putting it in mxsfb_load seems wrong.

MXSFB uses struct drm_simple_display_pipe_funcs which does have enable/disable callbacks, probably closer to what we want...

--
Stefan

Quoted 47 lines
Quoted 47 lines

The runtime power gating case can be implemented later to save power, need lcdif experts to do it. Thanks.

Anson.

Quoted 42 lines
Quoted 30 lines
---
 drivers/gpu/drm/mxsfb/mxsfb_drv.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
drivers/gpu/drm/mxsfb/mxsfb_drv.c +0 -1
index ffe5137..1ba179b 100644
@@ -235,7 +235,6 @@ static int mxsfb_load(struct drm_device *drm,
 unsigned long flags)
 
 	pm_runtime_get_sync(drm->dev);
 	ret = drm_irq_install(drm, platform_get_irq(pdev, 0));
-	pm_runtime_put_sync(drm->dev);
 
 	if (ret < 0) {

dev_err(drm->dev, "Failed to install IRQ handler\n"); @@ -264,6
+263,7 @@ static int mxsfb_load(struct drm_device *drm, unsigned long flags)
err_irq:

drivers/gpu/drm/mxsfb/mxsfb_drv.c +1 -0
 	drm_panel_detach(mxsfb->panel);
 err_vblank:
+	pm_runtime_put_sync(drm->dev);
 	pm_runtime_disable(drm->dev);
 
 	return ret;

drivers/gpu/drm/mxsfb/mxsfb_drv.c static void mxsfb_unload(struct drm_device *drm) · line 284

Quoted 7 lines
drivers/gpu/drm/mxsfb/mxsfb_drv.c +0 -1
@@ -279,7 +279,6 @@ static void mxsfb_unload(struct drm_device *drm)
 	drm_kms_helper_poll_fini(drm);
 	drm_mode_config_cleanup(drm);
 
-	pm_runtime_get_sync(drm->dev);
 	drm_irq_uninstall(drm);
 	pm_runtime_put_sync(drm->dev);
 

--
Best regards,
Marek Vasut

maintainer
#6 lore raw
Quoted 31 lines · Stefan Agner
Quoted 30 lines · Marek Vasut

On 07/13/2018 11:06 AM, Anson Huang wrote:
[...]

Quoted 13 lines
Quoted 13 lines · Anson Huang
Quoted 11 lines

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

I think drm driver should have somewhere to implement the display disable case, like when fb0 is blank (echo 1 > /sys/class/graphics/fb0/blank),

Isn't this just the fbdev emulation on top of drm/kms ? I think this stuff can be compiled out completely.

Quoted 4 lines

then lcdif can be powered gated, and
when display is back on (unblank), lcdif needs to be re-initialization and display will be on, current implementation is incorrect, with kernel booting up, lcdif is NOT working at all.

Quoted 8 lines
Quoted 3 lines

It works fine on MX6SX , so I think this is isolated to MX6SL ? I'm CCing Stefan, he might have some valuable feedback here.

Yeah not sure, but putting it in mxsfb_load seems wrong.

MXSFB uses struct drm_simple_display_pipe_funcs which does have enable/disable callbacks, probably closer to what we want...

Seems to me like this is not something that should be hacked around in the mxsfb driver in the first place, but somewhere else.

Signature · 2 lines
Best regards,
Marek Vasut
maintainer
#7 lore raw
Quoted 36 lines · Marek Vasut
Quoted 35 lines · Stefan Agner
Quoted 33 lines · Marek Vasut

On 07/13/2018 11:06 AM, Anson Huang wrote:
[...]

Quoted 13 lines
Quoted 13 lines · Anson Huang
Quoted 11 lines

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

I think drm driver should have somewhere to implement the display disable case, like when fb0 is blank (echo 1 > /sys/class/graphics/fb0/blank),

Isn't this just the fbdev emulation on top of drm/kms ? I think this stuff can be compiled out completely.

Quoted 4 lines

then lcdif can be powered gated, and
when display is back on (unblank), lcdif needs to be re-initialization and display will be on, current implementation is incorrect, with kernel booting up, lcdif is NOT working at all.

It works fine on MX6SX , so I think this is isolated to MX6SL ? I'm CCing Stefan, he might have some valuable feedback here.

Quoted 7 lines
Quoted 4 lines

Yeah not sure, but putting it in mxsfb_load seems wrong.

MXSFB uses struct drm_simple_display_pipe_funcs which does have enable/disable callbacks, probably closer to what we want...

Seems to me like this is not something that should be hacked around in the mxsfb driver in the first place, but somewhere else.

By somewhere else you mean in the DRM stack?

Yeah not sure, currently pm seems to be handled on driver level only, see
grep -r -e pm_runtime_get_sync drivers/gpu/drm

--
Stefan

maintainer
#8 lore raw
Quoted 39 lines · Stefan Agner
Quoted 38 lines · Marek Vasut
Quoted 37 lines · Stefan Agner
Quoted 25 lines · Marek Vasut

On 07/13/2018 11:06 AM, Anson Huang wrote:
[...]

Quoted 13 lines
Quoted 13 lines · Anson Huang
Quoted 11 lines

As display power domain is combined with lcdif node on some i.MX platforms like i.MX6SL, when lcdif driver is enabled, the mxsfb_load is called to enable runtime pm, and a pair of pm_runtime_get_sync and pm_runtime_put_sync are also called, that will cause generic power domain driver to disable lcdif power domain and lcdif is no longer working, the lcdif power should ONLY be turned off when display is disabled, so move the pm_runtime_put_sync to mxsfb_unload and remove the pm_runtime_get_sync in mxsfb_unload as well, in this way, when display is enabled, the lcdif power will be always ON until the display is disabled.

Quoted 6 lines
Quoted 3 lines
Quoted 1 line

Doesn't this also mean the block will always be on, thus wasting power ?

I think drm driver should have somewhere to implement the display disable case, like when fb0 is blank (echo 1 > /sys/class/graphics/fb0/blank),

Isn't this just the fbdev emulation on top of drm/kms ? I think this stuff can be compiled out completely.

Quoted 8 lines
Quoted 4 lines

then lcdif can be powered gated, and
when display is back on (unblank), lcdif needs to be re-initialization and display will be on, current implementation is incorrect, with kernel booting up, lcdif is NOT working at all.

It works fine on MX6SX , so I think this is isolated to MX6SL ? I'm CCing Stefan, he might have some valuable feedback here.

Yeah not sure, but putting it in mxsfb_load seems wrong.

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

MXSFB uses struct drm_simple_display_pipe_funcs which does have enable/disable callbacks, probably closer to what we want...

Seems to me like this is not something that should be hacked around in the mxsfb driver in the first place, but somewhere else.

By somewhere else you mean in the DRM stack?

Rather the clock driver or somesuch. Maybe the power domains need to be modeled in the DT properly, but I didn't dig in deep enough.

Quoted 7 lines

Yeah not sure, currently pm seems to be handled on driver level only, see
grep -r -e pm_runtime_get_sync drivers/gpu/drm

--
Stefan

Signature · 2 lines
Best regards,
Marek Vasut