Skip to contentSkip to conversation

ARM STM/i.MX: Allocate memory for framebuffer during probe

PATCH In review Gregory CLEMENT 2 participants 7 messages Jan 4, 2011 – Jan 13, 2011

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/4 ARM STM/i.MX: Allocate memory for framebuffer during probe

With current code when the framebuffer is register, screen_base is not set yet. So when we want to access framebuffer from mmap we get a pointer to 0x0 instead of getting the pointer to the framebuffer address. This patch fix this bug by allocating memory for framebuffer during probe just before registering framebuffer driver.

---
 drivers/video/stm.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)
drivers/video/stm.c +10 -0
index f0abe4c..8a9acff 100644
@@ -480,6 +480,7 @@ static int stmfb_probe(struct device_d *hw_dev)
 {
 	struct imx_fb_videomode *pdata = hw_dev->platform_data;
 	int ret;
+	unsigned size;
 
 	/* just init */
 	fbi.info.priv = &fbi;
@@ -497,6 +498,15 @@ static int stmfb_probe(struct device_d *hw_dev)
 	fbi.info.yres = fbi.info.mode->yres;
 	fbi.info.bits_per_pixel = 16;
 
+	size = calc_line_length(fbi.info.mode->xres, fbi.info.bits_per_pixel) *
+		fbi.info.mode->yres;
+
+	ret = stmfb_memory_mmgt(&fbi.info, size);
+	if (ret != 0) {
+		dev_err(hw_dev, "Cannot allocate framebuffer memory\n");
+		return ret;
+	}
+
 	ret = register_framebuffer(&fbi.info);
 	if (ret != 0) {
 		dev_err(hw_dev, "Failed to register framebuffer\n");
Signature · 9 lines
1.7.0.4




_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#1 lore raw
Quoted 5 lines · Gregory CLEMENT

With current code when the framebuffer is register, screen_base is not set yet. So when we want to access framebuffer from mmap we get a pointer to 0x0 instead of getting the pointer to the framebuffer address. This patch fix this bug by allocating memory for framebuffer during probe just before registering framebuffer driver.

Are you sure? See below

Quoted 23 lines
---
 drivers/video/stm.c |   10 ++++++++++
 1 files changed, 10 insertions(+), 0 deletions(-)
drivers/video/stm.c +3 -0
index f0abe4c..8a9acff 100644
@@ -480,6 +480,7 @@ static int stmfb_probe(struct device_d *hw_dev)
 {
 	struct imx_fb_videomode *pdata = hw_dev->platform_data;
 	int ret;
+	unsigned size;
 
 	/* just init */
 	fbi.info.priv = &fbi;
@@ -497,6 +498,15 @@ static int stmfb_probe(struct device_d *hw_dev)
 	fbi.info.yres = fbi.info.mode->yres;
 	fbi.info.bits_per_pixel = 16;
 
+	size = calc_line_length(fbi.info.mode->xres, fbi.info.bits_per_pixel) *
+		fbi.info.mode->yres;

drivers/video/stm.c static int stmfb_probe(struct device_d *hw_dev) · line 512

Quoted 10 lines
drivers/video/stm.c +7 -0
+
+	ret = stmfb_memory_mmgt(&fbi.info, size);
+	if (ret != 0) {
+		dev_err(hw_dev, "Cannot allocate framebuffer memory\n");
+		return ret;
+	}
+
 	ret = register_framebuffer(&fbi.info);
 	if (ret != 0) {
 		dev_err(hw_dev, "Failed to register framebuffer\n");

This probe function sets up the default mode. And register_framebuffer() itself calls stmfb_activate_var() which calls stmfb_memory_mmgt(). So, when the call to register_framebuffer() returns the 'screen_base' should no longer be NULL.

jbe

Signature · 9 lines
Pengutronix e.K.                              | Juergen Beisert             |
Linux Solutions for Science and Industry      | Phone: +49-8766-939 228     |
Vertretung Sued/Muenchen, Germany             | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686              | http://www.pengutronix.de/  |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#2 lore raw
Quoted 8 lines · Juergen Beisert
Quoted 5 lines · Gregory CLEMENT

With current code when the framebuffer is register, screen_base is not set yet. So when we want to access framebuffer from mmap we get a pointer to 0x0 instead of getting the pointer to the framebuffer address. This patch fix this bug by allocating memory for framebuffer during probe just before registering framebuffer driver.

Are you sure? See below

It was the value I get at runtime at least.

Quoted 2 lines
Quoted 2 lines

[...]

Quoted 5 lines

This probe function sets up the default mode. And register_framebuffer() itself calls stmfb_activate_var() which calls stmfb_memory_mmgt(). So, when the call to register_framebuffer() returns the 'screen_base' should no longer be NULL.

But in register_framebuffer first map_base get its value from screen_base and it is only after that fb_activate_var is called. And the problem is that mmap will not get screen_base but map_base and this one is set too early. Extract from register_framebuffer() in fb.c:

	info->cdev.dev->map_base = (unsigned long)info->screen_base;
map_base get its value here ------^
	info->cdev.dev->size = info->cdev.size;

	dev = &info->dev;
	dev->priv = info;
	dev->id = id;

	sprintf(dev->name, "fb");

	register_device(&info->dev);
	dev_add_param(dev, "enable", fb_enable_set, NULL, 0);
	dev_set_param(dev, "enable", "0");

	if (info->num_modes && (info->mode_list != NULL) &&
			(info->fbops->fb_activate_var != NULL)) {
		dev_add_param(dev, "mode_name", fb_setup_mode, NULL, 0);
		dev_set_param(dev, "mode_name", info->mode_list[0].name);

fb_activate_var is called here ------------------------^

This code is new for me, maybe I missed something.

Quoted 3 lines

jbe

Signature · 9 lines
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#3 lore raw
Quoted 44 lines · Gregory CLEMENT
Quoted 8 lines · Juergen Beisert
Quoted 5 lines · Gregory CLEMENT

With current code when the framebuffer is register, screen_base is not set yet. So when we want to access framebuffer from mmap we get a pointer to 0x0 instead of getting the pointer to the framebuffer address. This patch fix this bug by allocating memory for framebuffer during probe just before registering framebuffer driver.

Are you sure? See below

It was the value I get at runtime at least.

Quoted 1 line
Quoted 1 line

[...]

Quoted 4 lines

This probe function sets up the default mode. And register_framebuffer() itself calls stmfb_activate_var() which calls stmfb_memory_mmgt(). So, when the call to register_framebuffer() returns the 'screen_base' should no longer be NULL.

But in register_framebuffer first map_base get its value from screen_base and it is only after that fb_activate_var is called. And the problem is that mmap will not get screen_base but map_base and this one is set too early. Extract from register_framebuffer() in fb.c:

	info->cdev.dev->map_base = (unsigned long)info->screen_base;
map_base get its value here ------^
	info->cdev.dev->size = info->cdev.size;

	dev = &info->dev;
	dev->priv = info;
	dev->id = id;

	sprintf(dev->name, "fb");

	register_device(&info->dev);
	dev_add_param(dev, "enable", fb_enable_set, NULL, 0);
	dev_set_param(dev, "enable", "0");

	if (info->num_modes && (info->mode_list != NULL) &&
			(info->fbops->fb_activate_var != NULL)) {
		dev_add_param(dev, "mode_name", fb_setup_mode, NULL, 0);
		dev_set_param(dev, "mode_name", info->mode_list[0].name);
Quoted 1 line

fb_activate_var is called here ------------------------^

And this call will call driver's stmfb_activate_var() (and(!) stmfb_memory_mmgt()).
But I will check it again.

jbe

Signature · 9 lines
Pengutronix e.K.                              | Juergen Beisert             |
Linux Solutions for Science and Industry      | Phone: +49-8766-939 228     |
Vertretung Sued/Muenchen, Germany             | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686              | http://www.pengutronix.de/  |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#4 lore raw
Quoted 41 lines · Juergen Beisert
Quoted 40 lines · Gregory CLEMENT
Quoted 8 lines · Juergen Beisert
Quoted 5 lines · Gregory CLEMENT

With current code when the framebuffer is register, screen_base is not set yet. So when we want to access framebuffer from mmap we get a pointer to 0x0 instead of getting the pointer to the framebuffer address. This patch fix this bug by allocating memory for framebuffer during probe just before registering framebuffer driver.

Are you sure? See below

It was the value I get at runtime at least.

Quoted 1 line
Quoted 1 line

[...]

Quoted 4 lines

This probe function sets up the default mode. And register_framebuffer() itself calls stmfb_activate_var() which calls stmfb_memory_mmgt(). So, when the call to register_framebuffer() returns the 'screen_base' should no longer be NULL.

But in register_framebuffer first map_base get its value from screen_base and it is only after that fb_activate_var is called. And the problem is that mmap will not get screen_base but map_base and this one is set too early. Extract from register_framebuffer() in fb.c:

	info->cdev.dev->map_base = (unsigned long)info->screen_base;
map_base get its value here ------^
	info->cdev.dev->size = info->cdev.size;

	dev = &info->dev;
	dev->priv = info;
	dev->id = id;

	sprintf(dev->name, "fb");

	register_device(&info->dev);
	dev_add_param(dev, "enable", fb_enable_set, NULL, 0);
	dev_set_param(dev, "enable", "0");
Quoted 8 lines
Quoted 5 lines
	if (info->num_modes && (info->mode_list != NULL) &&
			(info->fbops->fb_activate_var != NULL)) {
		dev_add_param(dev, "mode_name", fb_setup_mode, NULL, 0);
		dev_set_param(dev, "mode_name", info->mode_list[0].name);

fb_activate_var is called here ------------------------^

And this call will call driver's stmfb_activate_var() (and(!) stmfb_memory_mmgt()).

Yes I totally agree with this. But at this time it's too late because changing screen_base won't change map_base. And a call to memmap is a call to generic_memmap_rw which will return map_base.
I didn't find anywhere in video driver code a place where map_base is updated, that's why I think that once it get its value it won't change.

Quoted 4 lines

But I will check it again.

jbe

Signature · 9 lines
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#5 lore raw
Quoted 48 lines · Gregory CLEMENT
Quoted 47 lines · Juergen Beisert
Quoted 45 lines · Gregory CLEMENT
Quoted 8 lines · Juergen Beisert
Quoted 5 lines · Gregory CLEMENT

With current code when the framebuffer is register, screen_base is not set yet. So when we want to access framebuffer from mmap we get a pointer to 0x0 instead of getting the pointer to the framebuffer address. This patch fix this bug by allocating memory for framebuffer during probe just before registering framebuffer driver.

Are you sure? See below

It was the value I get at runtime at least.

Quoted 1 line
Quoted 1 line

[...]

Quoted 4 lines

This probe function sets up the default mode. And register_framebuffer() itself calls stmfb_activate_var() which calls stmfb_memory_mmgt(). So, when the call to register_framebuffer() returns the 'screen_base' should no longer be NULL.

But in register_framebuffer first map_base get its value from screen_base and it is only after that fb_activate_var is called. And the problem is that mmap will not get screen_base but map_base and this one is set too early. Extract from register_framebuffer() in fb.c:

	info->cdev.dev->map_base = (unsigned long)info->screen_base;
map_base get its value here ------^
	info->cdev.dev->size = info->cdev.size;

	dev = &info->dev;
	dev->priv = info;
	dev->id = id;

	sprintf(dev->name, "fb");

	register_device(&info->dev);
	dev_add_param(dev, "enable", fb_enable_set, NULL, 0);
	dev_set_param(dev, "enable", "0");

	if (info->num_modes && (info->mode_list != NULL) &&
			(info->fbops->fb_activate_var != NULL)) {
		dev_add_param(dev, "mode_name", fb_setup_mode, NULL, 0);
		dev_set_param(dev, "mode_name", info->mode_list[0].name);

fb_activate_var is called here ------------------------^

Quoted 11 lines
Quoted 2 lines

And this call will call driver's stmfb_activate_var() (and(!) stmfb_memory_mmgt()).

Yes I totally agree with this. But at this time it's too late because changing screen_base won't change map_base. And a call to memmap is a call to generic_memmap_rw which will return map_base.
I didn't find anywhere in video driver code a place where map_base is updated, that's why I think that once it get its value it won't change.

Quoted 2 lines

But I will check it again.

Had you had some time to check it?
Will you take this patch ?
If you have any comments I am willing to take them in account to make a new version of this patch if it necessary.

Regards,

Quoted 4 lines
Quoted 2 lines

jbe

Signature · 9 lines
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#6 lore raw
Quoted 58 lines · Gregory CLEMENT
Quoted 57 lines · Gregory CLEMENT
Quoted 41 lines · Juergen Beisert
Quoted 40 lines · Gregory CLEMENT
Quoted 8 lines · Juergen Beisert
Quoted 5 lines · Gregory CLEMENT

With current code when the framebuffer is register, screen_base is not set yet. So when we want to access framebuffer from mmap we get a pointer to 0x0 instead of getting the pointer to the framebuffer address. This patch fix this bug by allocating memory for framebuffer during probe just before registering framebuffer driver.

Are you sure? See below

It was the value I get at runtime at least.

Quoted 1 line
Quoted 1 line

[...]

Quoted 4 lines

This probe function sets up the default mode. And register_framebuffer() itself calls stmfb_activate_var() which calls stmfb_memory_mmgt(). So, when the call to register_framebuffer() returns the 'screen_base' should no longer be NULL.

But in register_framebuffer first map_base get its value from screen_base and it is only after that fb_activate_var is called. And the problem is that mmap will not get screen_base but map_base and this one is set too early. Extract from register_framebuffer() in fb.c:

	info->cdev.dev->map_base = (unsigned long)info->screen_base;
map_base get its value here ------^
	info->cdev.dev->size = info->cdev.size;

	dev = &info->dev;
	dev->priv = info;
	dev->id = id;

	sprintf(dev->name, "fb");

	register_device(&info->dev);
	dev_add_param(dev, "enable", fb_enable_set, NULL, 0);
	dev_set_param(dev, "enable", "0");
Quoted 8 lines
Quoted 5 lines
	if (info->num_modes && (info->mode_list != NULL) &&
			(info->fbops->fb_activate_var != NULL)) {
		dev_add_param(dev, "mode_name", fb_setup_mode, NULL, 0);
		dev_set_param(dev, "mode_name", info->mode_list[0].name);

fb_activate_var is called here ------------------------^

And this call will call driver's stmfb_activate_var() (and(!) stmfb_memory_mmgt()).

Yes I totally agree with this. But at this time it's too late because changing screen_base won't change map_base. And a call to memmap is a call to generic_memmap_rw which will return map_base.
I didn't find anywhere in video driver code a place where map_base is updated, that's why I think that once it get its value it won't change.

Quoted 3 lines
Quoted 1 line
Quoted 1 line

But I will check it again.

Had you had some time to check it?

Not yet, sorry.

Quoted 3 lines

Will you take this patch ?
If you have any comments I am willing to take them in account to make a new version of this patch if it necessary.

Maybe I can check it tomorrow.

jbe

Signature · 9 lines
Pengutronix e.K.                              | Juergen Beisert             |
Linux Solutions for Science and Industry      | Phone: +49-8766-939 228     |
Vertretung Sued/Muenchen, Germany             | Fax:   +49-5121-206917-5555 |
Amtsgericht Hildesheim, HRA 2686              | http://www.pengutronix.de/  |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox