Skip to contentSkip to conversation

BCM2835 / simple framebuffer support

PATCH In review Andre Heider 3 participants 20 messages Nov 4, 2013 – Nov 14, 2013

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

v2 cover BCM2835 / simple framebuffer support

This is v2 of my patch set posted on 10/24.
It adds framebuffer support for the RPi and provides an additional config knob to set up the kernel's simplefb driver. The active barebox framebuffer is basis for that feature, so that other platforms can make use of it.

This simplefb option depends on the bcm2835 fb driver since drivers need to be aware of simplefb: framebuffers need to stay configured and cannot be teared down.

The RPi case is rather simple in this regard. The allocated framebuffer lies within the memory range of the VideoCore of the ARM/VC memory split. Hence, there is no memory range that needs to be reserved as far as the ARM side of things is concerned.
This likely needs to be added once other fb driver want to use simplefb.

Changes since v1:
* renamed "pitch" to "line_length" to match the kernel
* the bcm2835 framebuffer driver is now standalone
* configuring simplefb for the kernel is based on the active barebox framebuffer

Andre Heider (6):
fb: add a line_length value to struct fb_info
gui: convert graphic utils to respect line_length
gui: convert the bmp renderer to respect line_length
ARM: bcm2835: add missing mbox overscan response field
video: add a BCM2835 framebuffer driver
video: set up the kernel's simple framebuffer driver

 arch/arm/boards/raspberry-pi/rpi.c        |   1 +
 arch/arm/mach-bcm2835/include/mach/core.h |   5 +
 arch/arm/mach-bcm2835/include/mach/mbox.h |   1 +
 drivers/video/Kconfig                     |  14 +++
 drivers/video/Makefile                    |   2 +
 drivers/video/bcm2835.c                   | 136 ++++++++++++++++++++++++
 drivers/video/fb.c                        |  11 +-
 drivers/video/simplefb.c                  | 171 ++++++++++++++++++++++++++++++
 include/fb.h                              |   1 +
 lib/gui/bmp.c                             |   8 +-
 lib/gui/graphic_utils.c                   |  10 +-
 11 files changed, 349 insertions(+), 11 deletions(-)
 create mode 100644 drivers/video/bcm2835.c
 create mode 100644 drivers/video/simplefb.c
Signature · 7 lines
1.8.3.2


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

PATCH v2 1/6 fb: add a line_length value to struct fb_info

Add support for framebuffers with noncontiguous horizontal lines.

Video drivers can set this value if the hardware requires it. In case a driver does not set it, the current value of xres * (bpp / 8) is used instead.

---
 drivers/video/fb.c | 11 +++++++++--
 include/fb.h       |  1 +
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/video/fb.c +9 -2
index 420e4e3..0159994 100644
@@ -72,12 +72,16 @@ static int fb_setup_mode(struct device_d *dev, struct param_d *param,
 
 	info->xres = info->mode->xres;
 	info->yres = info->mode->yres;
+	info->line_length = 0;
 
 	ret = info->fbops->fb_activate_var(info);
 
+	if (!info->line_length)
+		info->line_length = info->xres * (info->bits_per_pixel >> 3);
+
 	if (!ret) {
 		dev->resource[0].start = (resource_size_t)info->screen_base;
-		info->cdev.size = info->xres * info->yres * (info->bits_per_pixel >> 3);
+		info->cdev.size = info->line_length * info->yres;
 		dev->resource[0].end = dev->resource[0].start + info->cdev.size - 1;
 		dev_param_set_generic(dev, param, val);
 	} else
@@ -122,9 +126,12 @@ int register_framebuffer(struct fb_info *info)
 
 	dev = &info->dev;
 
+	if (!info->line_length)
+		info->line_length = info->xres * (info->bits_per_pixel >> 3);
+
 	info->cdev.ops = &fb_ops;
 	info->cdev.name = asprintf("fb%d", id);
-	info->cdev.size = info->xres * info->yres * (info->bits_per_pixel >> 3);
+	info->cdev.size = info->line_length * info->yres;
 	info->cdev.dev = dev;
 	info->cdev.priv = info;
 	dev->resource = xzalloc(sizeof(struct resource));
include/fb.h +1 -0
index df4ba8e..cf9ffa0 100644
@@ -94,6 +94,7 @@ struct fb_info {
 	u32 xres;			/* visible resolution		*/
 	u32 yres;
 	u32 bits_per_pixel;		/* guess what			*/
+	u32 line_length;		/* length of a line in bytes	*/
 
 	u32 grayscale;			/* != 0 Graylevels instead of colors */
 
Signature · 7 lines
1.8.3.2


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

Re: [PATCH v2 1/6] fb: add a line_length value to struct fb_info

Quoted 17 lines · Andre Heider

Add support for framebuffers with noncontiguous horizontal lines.

Video drivers can set this value if the hardware requires it. In case a driver does not set it, the current value of xres * (bpp / 8) is used instead.

---
 drivers/video/fb.c | 11 +++++++++--
 include/fb.h       |  1 +
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/video/fb.c +0 -0
index 420e4e3..0159994 100644

drivers/video/fb.c static int fb_setup_mode(struct device_d *dev, struct param_d *param, · line 81

Quoted 10 lines
drivers/video/fb.c +4 -0
@@ -72,12 +72,16 @@ static int fb_setup_mode(struct device_d *dev, struct param_d *param,
 
 	info->xres = info->mode->xres;
 	info->yres = info->mode->yres;
+	info->line_length = 0;
 
 	ret = info->fbops->fb_activate_var(info);
 
+	if (!info->line_length)
+		info->line_length = info->xres * (info->bits_per_pixel >> 3);
+

Isn't this already set from the register_framebuffer function?
You wanna see that somebody changed this value in info->fbops->fb_activate_var(info)?

Then I would do nothing here.

- Alex

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

#3 lore raw

Re: [PATCH v2 1/6] fb: add a line_length value to struct fb_info

Quoted 25 lines · Alexander Aring
Quoted 24 lines · Andre Heider

Add support for framebuffers with noncontiguous horizontal lines.

Video drivers can set this value if the hardware requires it. In case a driver does not set it, the current value of xres * (bpp / 8) is used instead.

---
 drivers/video/fb.c | 11 +++++++++--
 include/fb.h       |  1 +
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/video/fb.c +1 -0
index 420e4e3..0159994 100644
@@ -72,12 +72,16 @@ static int fb_setup_mode(struct device_d *dev, struct param_d *param,
 
 	info->xres = info->mode->xres;
 	info->yres = info->mode->yres;
+	info->line_length = 0;
 
 	ret = info->fbops->fb_activate_var(info);
 
Quoted 7 lines

drivers/video/fb.c static int fb_setup_mode(struct device_d *dev, struct param_d *param, · line 81

Quoted 3 lines
drivers/video/fb.c +3 -0
+	if (!info->line_length)
+		info->line_length = info->xres * (info->bits_per_pixel >> 3);
+

Isn't this already set from the register_framebuffer function?
You wanna see that somebody changed this value in info->fbops->fb_activate_var(info)?

Then I would do nothing here.

It is set from register_framebuffer, but the mode setup invalidates it, So Andre sets it to 0 first and updates it after calling into the driver if necessary. The code looks correct. Maybe I misunderstand you?

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

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

Re: [PATCH v2 1/6] fb: add a line_length value to struct fb_info

Hi Sascha,

Quoted 32 lines · Sascha Hauer
Quoted 31 lines · Alexander Aring
Quoted 17 lines · Andre Heider

Add support for framebuffers with noncontiguous horizontal lines.

Video drivers can set this value if the hardware requires it. In case a driver does not set it, the current value of xres * (bpp / 8) is used instead.

---
 drivers/video/fb.c | 11 +++++++++--
 include/fb.h       |  1 +
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/video/fb.c +0 -0
index 420e4e3..0159994 100644

drivers/video/fb.c static int fb_setup_mode(struct device_d *dev, struct param_d *param, · line 81

Quoted 10 lines
drivers/video/fb.c +4 -0
@@ -72,12 +72,16 @@ static int fb_setup_mode(struct device_d *dev, struct param_d *param,
 
 	info->xres = info->mode->xres;
 	info->yres = info->mode->yres;
+	info->line_length = 0;
 
 	ret = info->fbops->fb_activate_var(info);
 
+	if (!info->line_length)
+		info->line_length = info->xres * (info->bits_per_pixel >> 3);
+

Isn't this already set from the register_framebuffer function?
You wanna see that somebody changed this value in info->fbops->fb_activate_var(info)?

Quoted 6 lines
Quoted 1 line

Then I would do nothing here.

It is set from register_framebuffer, but the mode setup invalidates it, So Andre sets it to 0 first and updates it after calling into the driver if necessary. The code looks correct. Maybe I misunderstand you?

ok. Was a hard day and night for me ;-)
Maybe I was too tired and look all day into another code and other things ;-).

I think I need a coffee or beer now.

I mean we have the same code in register_framebuffer:

if (!info->line_length)

        info->line_length = info->xres * (info->bits_per_pixel >> 3);

I don't know the exactly context to call fb_setup_mode but then we set it twice and we don't need to check if the value has changed from fb_activate_var function.

It's only a nitpick, but this code is correct. Of course.

- Alex

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

#5 lore raw

Re: [PATCH v2 1/6] fb: add a line_length value to struct fb_info

Hi Alexander,

Quoted 48 lines · Alexander Aring

Hi Sascha,

Quoted 32 lines · Sascha Hauer
Quoted 31 lines · Alexander Aring
Quoted 17 lines · Andre Heider

Add support for framebuffers with noncontiguous horizontal lines.

Video drivers can set this value if the hardware requires it. In case a driver does not set it, the current value of xres * (bpp / 8) is used instead.

---
 drivers/video/fb.c | 11 +++++++++--
 include/fb.h       |  1 +
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/video/fb.c +0 -0
index 420e4e3..0159994 100644

drivers/video/fb.c static int fb_setup_mode(struct device_d *dev, struct param_d *param, · line 81

Quoted 10 lines
drivers/video/fb.c +4 -0
@@ -72,12 +72,16 @@ static int fb_setup_mode(struct device_d *dev, struct param_d *param,
 
 	info->xres = info->mode->xres;
 	info->yres = info->mode->yres;
+	info->line_length = 0;
 
 	ret = info->fbops->fb_activate_var(info);
 
+	if (!info->line_length)
+		info->line_length = info->xres * (info->bits_per_pixel >> 3);
+

Isn't this already set from the register_framebuffer function?
You wanna see that somebody changed this value in info->fbops->fb_activate_var(info)?

Quoted 6 lines
Quoted 1 line

Then I would do nothing here.

It is set from register_framebuffer, but the mode setup invalidates it, So Andre sets it to 0 first and updates it after calling into the driver if necessary. The code looks correct. Maybe I misunderstand you?

ok. Was a hard day and night for me ;-)
Maybe I was too tired and look all day into another code and other things ;-).

I think I need a coffee or beer now.

Quoted 8 lines

I mean we have the same code in register_framebuffer:

if (!info->line_length)

        info->line_length = info->xres * (info->bits_per_pixel >> 3);

I don't know the exactly context to call fb_setup_mode but then we set it twice and we don't need to check if the value has changed from fb_activate_var function.

With this patch we have a new line_length value, but fb drivers don't have to set it, its optional. What you see in register_framebuffer() is a default value for the initial mode. The framework sets it in case the driver doesn't.

But we can also change modes, so if a fb driver doesn't update line_length in the fb_activate_var callback, we have a stale line_length value from the old mode.
So I set it to zero before the callback and check afterwards if the driver set it. Without that there would be no way to distinguish if the value is valid for the new mode or a stale one from the prior mode.

Thanks,
Andre

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

#6 lore raw

Re: [PATCH v2 1/6] fb: add a line_length value to struct fb_info

Quoted 65 lines · Andre Heider

Hi Alexander,

Quoted 48 lines · Alexander Aring

Hi Sascha,

Quoted 32 lines · Sascha Hauer
Quoted 31 lines · Alexander Aring
Quoted 17 lines · Andre Heider

Add support for framebuffers with noncontiguous horizontal lines.

Video drivers can set this value if the hardware requires it. In case a driver does not set it, the current value of xres * (bpp / 8) is used instead.

---
 drivers/video/fb.c | 11 +++++++++--
 include/fb.h       |  1 +
 2 files changed, 10 insertions(+), 2 deletions(-)
drivers/video/fb.c +0 -0
index 420e4e3..0159994 100644

drivers/video/fb.c static int fb_setup_mode(struct device_d *dev, struct param_d *param, · line 81

Quoted 10 lines
drivers/video/fb.c +4 -0
@@ -72,12 +72,16 @@ static int fb_setup_mode(struct device_d *dev, struct param_d *param,
 
 	info->xres = info->mode->xres;
 	info->yres = info->mode->yres;
+	info->line_length = 0;
 
 	ret = info->fbops->fb_activate_var(info);
 
+	if (!info->line_length)
+		info->line_length = info->xres * (info->bits_per_pixel >> 3);
+

Isn't this already set from the register_framebuffer function?
You wanna see that somebody changed this value in info->fbops->fb_activate_var(info)?

Quoted 6 lines
Quoted 1 line

Then I would do nothing here.

It is set from register_framebuffer, but the mode setup invalidates it, So Andre sets it to 0 first and updates it after calling into the driver if necessary. The code looks correct. Maybe I misunderstand you?

ok. Was a hard day and night for me ;-)
Maybe I was too tired and look all day into another code and other things ;-).

I think I need a coffee or beer now.

Quoted 8 lines

I mean we have the same code in register_framebuffer:

if (!info->line_length)

        info->line_length = info->xres * (info->bits_per_pixel >> 3);

I don't know the exactly context to call fb_setup_mode but then we set it twice and we don't need to check if the value has changed from fb_activate_var function.

With this patch we have a new line_length value, but fb drivers don't have to set it, its optional. What you see in register_framebuffer() is a default value for the initial mode. The framework sets it in case the driver doesn't.

Quoted 7 lines

But we can also change modes, so if a fb driver doesn't update line_length in the fb_activate_var callback, we have a stale line_length value from the old mode.
So I set it to zero before the callback and check afterwards if the driver set it. Without that there would be no way to distinguish if the value is valid for the new mode or a stale one from the prior mode.

ok, thanks for the little explaination.

- Alex

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

#7 lore raw

PATCH v2 2/6 gui: convert graphic utils to respect line_length

In reply to Andre Heider
---
 lib/gui/graphic_utils.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)
lib/gui/graphic_utils.c +5 -5
index 95687df..300c525 100644
@@ -167,18 +167,18 @@ void rgba_blend(struct fb_info *info, struct image *img, void* buf, int height,
 {
 	unsigned char *adr;
 	int x, y;
-	int xres;
+	int line_length;
 	int img_byte_per_pixel = 3;
 	void *image;
 
 	if (is_rgba)
 		img_byte_per_pixel++;
 
-	xres = info->xres;
+	line_length = info->line_length;
 
 	for (y = 0; y < height; y++) {
-		adr = buf + ((y + starty) * xres + startx) *
-				(info->bits_per_pixel >> 3);
+		adr = buf + (y + starty) * line_length +
+				startx * (info->bits_per_pixel >> 3);
 		image = img->data + (y * img->width *img_byte_per_pixel);
 
 		for (x = 0; x < width; x++) {
@@ -219,7 +219,7 @@ int fb_open(const char * fbdev, struct screen *sc, bool offscreen)
 	sc->s.y = 0;
 	sc->s.width = sc->info.xres;
 	sc->s.height = sc->info.yres;
-	sc->fbsize = sc->s.width * sc->s.height * (sc->info.bits_per_pixel >> 3);
+	sc->fbsize = sc->info.line_length * sc->s.height;
 
 	if (offscreen) {
 		/* Don't fail if malloc fails, just continue rendering directly
Signature · 7 lines
1.8.3.2


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

PATCH v2 3/6 gui: convert the bmp renderer to respect line_length

In reply to Andre Heider
---
 lib/gui/bmp.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
lib/gui/bmp.c +4 -4
index 6bf8cd0..dcf3095 100644
@@ -78,8 +78,8 @@ static int bmp_renderer(struct screen *sc, struct surface *s, struct image *img)
 			image = (char *)bmp +
 					le32_to_cpu(bmp->header.data_offset);
 			image += (img->height - y - 1) * img->width * (bits_per_pixel >> 3);
-			adr = buf + ((y + starty) * sc->s.width + startx) *
-					(sc->info.bits_per_pixel >> 3);
+			adr = buf + (y + starty) * sc->info.line_length +
+					startx * (sc->info.bits_per_pixel >> 3);
 			for (x = 0; x < width; x++) {
 				int pixel;
 
@@ -100,8 +100,8 @@ static int bmp_renderer(struct screen *sc, struct surface *s, struct image *img)
 			image = (char *)bmp +
 					le32_to_cpu(bmp->header.data_offset);
 			image += (img->height - y - 1) * img->width * (bits_per_pixel >> 3);
-			adr = buf + ((y + starty) * sc->s.width + startx) *
-					(sc->info.bits_per_pixel >> 3);
+			adr = buf + (y + starty) * sc->info.line_length +
+					startx * (sc->info.bits_per_pixel >> 3);
 			for (x = 0; x < width; x++) {
 				char *pixel;
 
Signature · 7 lines
1.8.3.2


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

PATCH v2 4/6 ARM: bcm2835: add missing mbox overscan response field

In reply to Andre Heider

Add the missing "right" field to struct bcm2835_mbox_tag_overscan.

---
 arch/arm/mach-bcm2835/include/mach/mbox.h | 1 +
 1 file changed, 1 insertion(+)
arch/arm/mach-bcm2835/include/mach/mbox.h +1 -0
index 6c36389..fb8a9bf 100644
@@ -355,6 +355,7 @@ struct bcm2835_mbox_tag_overscan {
 			u32 top;
 			u32 bottom;
 			u32 left;
+			u32 right;
 		} resp;
 	} body;
 };
Signature · 7 lines
1.8.3.2


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

PATCH v2 5/6 video: add a BCM2835 framebuffer driver

In reply to Andre Heider

Use the mailbox driver to set up a framebuffer based on the firmware configuration.

---
 arch/arm/boards/raspberry-pi/rpi.c        |   1 +
 arch/arm/mach-bcm2835/include/mach/core.h |   5 ++
 drivers/video/Kconfig                     |   6 ++
 drivers/video/Makefile                    |   1 +
 drivers/video/bcm2835.c                   | 136 ++++++++++++++++++++++++++++++
 5 files changed, 149 insertions(+)
 create mode 100644 drivers/video/bcm2835.c
arch/arm/boards/raspberry-pi/rpi.c +1 -0
index 2717fee..96025ba 100644
@@ -140,6 +140,7 @@ static int rpi_env_init(void)
 static int rpi_devices_init(void)
 {
 	bcm2835_register_mci();
+	bcm2835_register_fb();
 	armlinux_set_architecture(MACH_TYPE_BCM2708);
 	armlinux_set_bootparams((void *)(0x00000100));
 	rpi_env_init();
arch/arm/mach-bcm2835/include/mach/core.h +5 -0
index a095db8..477ecb9 100644
@@ -27,4 +27,9 @@ static void inline bcm2835_register_mci(void)
 			IORESOURCE_MEM, NULL);
 }
 
+static void inline bcm2835_register_fb(void)
+{
+	add_generic_device("bcm2835_fb", 0, NULL, 0, 0, 0, NULL);
+}
+
 #endif
drivers/video/Kconfig +6 -0
index 0639d9c..b532e7d 100644
@@ -71,4 +71,10 @@ config DRIVER_VIDEO_PXA
 	  Add support for the frame buffer device found on the PXA270
 	  CPU.
 
+config DRIVER_VIDEO_BCM2835
+	bool "BCM2835 framebuffer driver"
+	depends on ARCH_BCM2835
+	help
+	  Add support for the BCM2835/VideoCore frame buffer device.
+
 endif
drivers/video/Makefile +1 -0
index 67169d1..244feab 100644
@@ -9,3 +9,4 @@ obj-$(CONFIG_DRIVER_VIDEO_S3C24XX) += s3c24xx.o
 obj-$(CONFIG_DRIVER_VIDEO_PXA) += pxa.o
 obj-$(CONFIG_DRIVER_VIDEO_SDL) += sdl.o
 obj-$(CONFIG_DRIVER_VIDEO_OMAP) += omap.o
+obj-$(CONFIG_DRIVER_VIDEO_BCM2835) += bcm2835.o
drivers/video/bcm2835.c new file +136 -0

136 lines not on this page · raw message

Signature · 7 lines
1.8.3.2


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

PATCH v2 6/6 video: set up the kernel's simple framebuffer driver

In reply to Andre Heider

Add support to configure the active framebuffer for the kernel through device tree.

---
 drivers/video/Kconfig    |   8 +++
 drivers/video/Makefile   |   1 +
 drivers/video/simplefb.c | 171 +++++++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 180 insertions(+)
 create mode 100644 drivers/video/simplefb.c
drivers/video/Kconfig +8 -0
index b532e7d..ccfaba8 100644
@@ -77,4 +77,12 @@ config DRIVER_VIDEO_BCM2835
 	help
 	  Add support for the BCM2835/VideoCore frame buffer device.
 
+config DRIVER_VIDEO_SIMPLEFB
+	bool "Simple framebuffer support"
+	depends on OFTREE
+	depends on DRIVER_VIDEO_BCM2835
+	help
+	  Add support for setting up the kernel's simple framebuffer driver
+	  based on the active barebox framebuffer.
+
 endif
drivers/video/Makefile +1 -0
index 244feab..31edfca 100644
@@ -10,3 +10,4 @@ obj-$(CONFIG_DRIVER_VIDEO_PXA) += pxa.o
 obj-$(CONFIG_DRIVER_VIDEO_SDL) += sdl.o
 obj-$(CONFIG_DRIVER_VIDEO_OMAP) += omap.o
 obj-$(CONFIG_DRIVER_VIDEO_BCM2835) += bcm2835.o
+obj-$(CONFIG_DRIVER_VIDEO_SIMPLEFB) += simplefb.o
drivers/video/simplefb.c new file +171 -0

171 lines not on this page · raw message

Signature · 7 lines
1.8.3.2


_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#12 lore raw
In reply to Andre Heider

Hi Andre,

Quoted 15 lines · Andre Heider

This is v2 of my patch set posted on 10/24.
It adds framebuffer support for the RPi and provides an additional config knob to set up the kernel's simplefb driver. The active barebox framebuffer is basis for that feature, so that other platforms can make use of it.

This simplefb option depends on the bcm2835 fb driver since drivers need to be aware of simplefb: framebuffers need to stay configured and cannot be teared down.

The RPi case is rather simple in this regard. The allocated framebuffer lies within the memory range of the VideoCore of the ARM/VC memory split. Hence, there is no memory range that needs to be reserved as far as the ARM side of things is concerned.
This likely needs to be added once other fb driver want to use simplefb.

Quoted 5 lines

Changes since v1:
* renamed "pitch" to "line_length" to match the kernel
* the bcm2835 framebuffer driver is now standalone
* configuring simplefb for the kernel is based on the active barebox framebuffer

I applied this series without the simplefb patch.

For the simplefb I'd like to use the following patch instead. It's your patch with some adjustments:

- add a register_simplefb device parameter to make it configurable whether a simplefb node should be created or not.
- register the OF fixup in register_framebuffer(). This way we always have a simplefb node once a framebuffer is registered without getting spurious warnings when there is no framebuffer present.
- check if the framebuffer is actually enabled. If it's not, we shouldn't register a simplefb node.

Are you fine with this and if yes, could you give the patch a test? It depends on the add-context-pointer-to-of_register_fixup patch I just posted. Just try the current -next branch, it should contain all you need for testing.

One oddity I just realized is that my patch will try and create a /framebuffer node multiple times if multiple framebuffers exist. I would have to fix this.

Sascha

8<----------------------------------------------------------

From 97607e85cdc44824e9617c74325bc9bc6405c383 Mon Sep 17 00:00:00 2001
From: Andre Heider <[email protected]>
Date: Tue, 5 Nov 2013 00:01:04 +0100
Subject: [PATCH] video: set up the kernel's simple framebuffer driver

Add support to configure the active framebuffer for the kernel through device tree.

---
 drivers/video/Kconfig    |   7 ++
 drivers/video/Makefile   |   1 +
 drivers/video/fb.c       |   7 ++
 drivers/video/simplefb.c | 167 +++++++++++++++++++++++++++++++++++++++++++++++
 include/fb.h             |   7 ++
 5 files changed, 189 insertions(+)
 create mode 100644 drivers/video/simplefb.c
drivers/video/Kconfig +7 -0
index b532e7d..5539266 100644
@@ -77,4 +77,11 @@ config DRIVER_VIDEO_BCM2835
 	help
 	  Add support for the BCM2835/VideoCore frame buffer device.
 
+config DRIVER_VIDEO_SIMPLEFB
+	bool "Simple framebuffer support"
+	depends on OFTREE
+	help
+	  Add support for setting up the kernel's simple framebuffer driver
+	  based on the active barebox framebuffer.
+
 endif
drivers/video/Makefile +1 -0
index 244feab..31edfca 100644
@@ -10,3 +10,4 @@ obj-$(CONFIG_DRIVER_VIDEO_PXA) += pxa.o
 obj-$(CONFIG_DRIVER_VIDEO_SDL) += sdl.o
 obj-$(CONFIG_DRIVER_VIDEO_OMAP) += omap.o
 obj-$(CONFIG_DRIVER_VIDEO_BCM2835) += bcm2835.o
+obj-$(CONFIG_DRIVER_VIDEO_SIMPLEFB) += simplefb.o
drivers/video/fb.c +7 -0
index 0159994..4263027 100644
@@ -163,6 +163,13 @@ int register_framebuffer(struct fb_info *info)
 	if (ret)
 		goto err_unregister;
 
+	if (IS_ENABLED(CONFIG_DRIVER_VIDEO_SIMPLEFB)) {
+		ret = fb_register_simplefb(info);
+		if (ret)
+			dev_err(&info->dev, "failed to register simplefb: %s\n",
+					strerror(-ret));
+	}
+
 	return 0;
 
 err_unregister:
drivers/video/simplefb.c new file +167 -0

167 lines not on this page · raw message

include/fb.h +7 -0
index cf9ffa0..98d5a03 100644
@@ -105,6 +105,9 @@ struct fb_info {
 
 	int enabled;
 	int p_enable;
+	int register_simplefb;		/* If true a simplefb device node will
+					 * be created.
+					 */
 };
 
 int register_framebuffer(struct fb_info *info);
@@ -115,5 +118,9 @@ int register_framebuffer(struct fb_info *info);
 
 extern struct bus_type fb_bus;
 
+/* fb internal functions */
+
+int fb_register_simplefb(struct fb_info *info);
+
 #endif /* __FB_H */
 
Signature · 13 lines
1.8.4.rc3


-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

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

Hi Sascha,

drivers/video/fb.c int register_framebuffer(struct fb_info *info) · line 170

Quoted 10 lines · Sascha Hauer

Hi Andre,

drivers/video/fb.c +5 -0
@@ -163,6 +163,13 @@ int register_framebuffer(struct fb_info *info)
 	if (ret)
 		goto err_unregister;
 
+	if (IS_ENABLED(CONFIG_DRIVER_VIDEO_SIMPLEFB)) {
+		ret = fb_register_simplefb(info);
+		if (ret)
+			dev_err(&info->dev, "failed to register simplefb: %s\n",
+					strerror(-ret));

maybe we should return ret here?

drivers/video/simplefb.c

Quoted 6 lines
drivers/video/fb.c +2 -0
+	}
+
 	return 0;
 
 err_unregister:
drivers/video/simplefb.c +0 -0

- Alex

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

#14 lore raw
Quoted 8 lines · Alexander Aring

Hi Sascha,

Quoted 5 lines · Sascha Hauer

Hi Andre,

drivers/video/fb.c +0 -0
@@ -163,6 +163,13 @@ int register_framebuffer(struct fb_info *info)
 	if (ret)
 		goto err_unregister;
 
Quoted 6 lines

drivers/video/fb.c int register_framebuffer(struct fb_info *info) · line 170

Quoted 5 lines
drivers/video/fb.c +5 -0
+	if (IS_ENABLED(CONFIG_DRIVER_VIDEO_SIMPLEFB)) {
+		ret = fb_register_simplefb(info);
+		if (ret)
+			dev_err(&info->dev, "failed to register simplefb: %s\n",
+					strerror(-ret));

maybe we should return ret here?

I think it's fine to continue without simplefb here. In the end the framebuffer works, just without simplefb.

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#15 lore raw
Quoted 16 lines · Sascha Hauer
Quoted 8 lines · Alexander Aring

Hi Sascha,

Quoted 5 lines · Sascha Hauer

Hi Andre,

drivers/video/fb.c +0 -0
@@ -163,6 +163,13 @@ int register_framebuffer(struct fb_info *info)
 	if (ret)
 		goto err_unregister;
 
Quoted 6 lines

drivers/video/fb.c int register_framebuffer(struct fb_info *info) · line 170

Quoted 5 lines
drivers/video/fb.c +5 -0
+	if (IS_ENABLED(CONFIG_DRIVER_VIDEO_SIMPLEFB)) {
+		ret = fb_register_simplefb(info);
+		if (ret)
+			dev_err(&info->dev, "failed to register simplefb: %s\n",
+					strerror(-ret));

maybe we should return ret here?

Quoted 3 lines

I think it's fine to continue without simplefb here. In the end the framebuffer works, just without simplefb.

ok, thanks.

- Alex

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

#16 lore raw
In reply to Sascha Hauer

Hi Sascha,

Quoted 18 lines · Sascha Hauer

Hi Andre,

Quoted 15 lines · Andre Heider

This is v2 of my patch set posted on 10/24.
It adds framebuffer support for the RPi and provides an additional config knob to set up the kernel's simplefb driver. The active barebox framebuffer is basis for that feature, so that other platforms can make use of it.

This simplefb option depends on the bcm2835 fb driver since drivers need to be aware of simplefb: framebuffers need to stay configured and cannot be teared down.

The RPi case is rather simple in this regard. The allocated framebuffer lies within the memory range of the VideoCore of the ARM/VC memory split. Hence, there is no memory range that needs to be reserved as far as the ARM side of things is concerned.
This likely needs to be added once other fb driver want to use simplefb.

Quoted 7 lines
Quoted 5 lines

Changes since v1:
* renamed "pitch" to "line_length" to match the kernel
* the bcm2835 framebuffer driver is now standalone
* configuring simplefb for the kernel is based on the active barebox framebuffer

I applied this series without the simplefb patch.

thanks!

Quoted 12 lines

For the simplefb I'd like to use the following patch instead. It's your patch with some adjustments:

- add a register_simplefb device parameter to make it configurable whether a simplefb node should be created or not.
- register the OF fixup in register_framebuffer(). This way we always have a simplefb node once a framebuffer is registered without getting spurious warnings when there is no framebuffer present.
- check if the framebuffer is actually enabled. If it's not, we shouldn't register a simplefb node.

Quoted 4 lines

Are you fine with this and if yes, could you give the patch a test? It depends on the add-context-pointer-to-of_register_fixup patch I just posted. Just try the current -next branch, it should contain all you need for testing.

Neat, that looks nicer, so no objection here ;)

And yes, it works if I set fb0.enable=1 and fb0.register_simplefb=1, both are 0 per default.

Now I wonder who/where those are supposed to get set? Is that a job for the runtime environment?

I didn't really care about the enabled parameter until now, there isn't anything to do in the fb_enable callback for this fb driver and splash doesn't seem to care either.

Regards,
Andre

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

#17 lore raw

Hi Sascha,

Quoted 5 lines · Andre Heider

And yes, it works if I set fb0.enable=1 and fb0.register_simplefb=1, both are 0 per default.

Now I wonder who/where those are supposed to get set? Is that a job for the runtime environment?

if possible I'd like to make the fb work out of the box. Setting these two parameters to default to "1" feels wrong (and I didn't find other drivers to do so). Any suggestions?

Thanks,
Andre

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

#18 lore raw
Quoted 6 lines · Andre Heider

Hi Sascha,

Quoted 3 lines · Andre Heider

And yes, it works if I set fb0.enable=1 and fb0.register_simplefb=1, both are 0 per default.

Quoted 6 lines
Quoted 2 lines

Now I wonder who/where those are supposed to get set? Is that a job for the runtime environment?

if possible I'd like to make the fb work out of the box. Setting these two parameters to default to "1" feels wrong (and I didn't find other drivers to do so). Any suggestions?

Normally the policy is that a framebuffer is not enabled during registration. This makes sense because normally the framebuffer doesn't contain a useful picture, so users should set a picture and enable the framebuffer afterwards to avoid flickering.

In your case the situation is different though. The display hardware is already enabled when you enter barebox, right? In this case I suggest that you set info->p_enable to true before calling register_framebuffer from the bcm2835 driver. This should be the right thing to do since it only reflects the hardware status.

Also I think you can set info->register_simplefb to true aswell in the bcm2835 driver since on that platform this is probably a sane default.

Sascha

Signature · 9 lines
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |

_______________________________________________
barebox mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/barebox
#19 lore raw
Quoted 14 lines · Sascha Hauer
Quoted 6 lines · Andre Heider

Hi Sascha,

Quoted 3 lines · Andre Heider

And yes, it works if I set fb0.enable=1 and fb0.register_simplefb=1, both are 0 per default.

Quoted 6 lines
Quoted 2 lines

Now I wonder who/where those are supposed to get set? Is that a job for the runtime environment?

if possible I'd like to make the fb work out of the box. Setting these two parameters to default to "1" feels wrong (and I didn't find other drivers to do so). Any suggestions?

Quoted 7 lines

Normally the policy is that a framebuffer is not enabled during registration. This makes sense because normally the framebuffer doesn't contain a useful picture, so users should set a picture and enable the framebuffer afterwards to avoid flickering.

In your case the situation is different though. The display hardware is already enabled when you enter barebox, right?

Right.

Quoted 7 lines

In this case I suggest
that you set info->p_enable to true before calling register_framebuffer from the bcm2835 driver. This should be the right thing to do since it only reflects the hardware status.

Also I think you can set info->register_simplefb to true aswell in the bcm2835 driver since on that platform this is probably a sane default.

Ok, easy enough ;)

Thanks,
Andre

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