[PATCH v2 2/5] fbdev: Give atafb proper parent

[PATCH v2 2/5] fbdev: Give atafb proper parent

[Michael Schmitz]

From: Miro Kropacek <miro.kropacek@gmail.com>

The atafb fb device registers no parent, causing a missing symlink
(/sys/class/graphics/fb0/device). Xorg fbdevhw driver looks for that
symlink when scanning for devices, so add a parent node for atafb.

Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Cc: <linux-fbdev@vger.kernel.org>
Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
---
 drivers/video/fbdev/atafb.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/video/fbdev/atafb.c b/drivers/video/fbdev/atafb.c
index b8ed1c537293..3d540803dc2b 100644
--- a/drivers/video/fbdev/atafb.c
+++ b/drivers/video/fbdev/atafb.c
@@ -3101,6 +3101,11 @@ static int __init atafb_probe(struct platform_device *pdev)
 #endif /* ATAFB_EXT */
 
 //	strcpy(fb_info.mode->name, "Atari Builtin ");
+	/* Parent the fb device properly: without this fb0 registers as a
+	 * virtual sysfs device with no /sys/class/graphics/fb0/device link,
+	 * which makes Xorg's fbdevhw reject it ("No devices detected").
+	 */
+	fb_info.device = &pdev->dev;
 	fb_info.fbops = &atafb_ops;
 	// try to set default (detected; requested) var
 	do_fb_set_var(&atafb_predefined[default_par - 1], 1);
-- 
2.17.1

Re: [PATCH v2 2/5] fbdev: Give atafb proper parent

[Helge Deller]

On 8/16/26 09:37, Michael Schmitz wrote:
> From: Miro Kropacek <miro.kropacek@gmail.com>
> 
> The atafb fb device registers no parent, causing a missing symlink
> (/sys/class/graphics/fb0/device). Xorg fbdevhw driver looks for that
> symlink when scanning for devices, so add a parent node for atafb.
> 
> Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
> Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
> Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
> Cc: <linux-fbdev@vger.kernel.org>
> Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html

Did anything on the fbdev patches (#2-#4) changed between v1 and v2?

Helge
   

Re: [PATCH v2 2/5] fbdev: Give atafb proper parent

[Michael Schmitz]

Hi Helge,

no changes to the fbdev patches - just didn't know how else to avoid 
issues with b4.

Cheers,

	Michael


Am 17.08.2026 um 02:38 schrieb Helge Deller:
> On 8/16/26 09:37, Michael Schmitz wrote:
>> From: Miro Kropacek <miro.kropacek@gmail.com>
>>
>> The atafb fb device registers no parent, causing a missing symlink
>> (/sys/class/graphics/fb0/device). Xorg fbdevhw driver looks for that
>> symlink when scanning for devices, so add a parent node for atafb.
>>
>> Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
>> Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
>> Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
>> Cc: <linux-fbdev@vger.kernel.org>
>> Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
>
> Did anything on the fbdev patches (#2-#4) changed between v1 and v2?
>
> Helge
>

Re: [PATCH v2 2/5] fbdev: Give atafb proper parent

[Thomas Zimmermann]

Hi

Am 16.08.26 um 09:37 schrieb Michael Schmitz:
> From: Miro Kropacek <miro.kropacek@gmail.com>
>
> The atafb fb device registers no parent, causing a missing symlink
> (/sys/class/graphics/fb0/device). Xorg fbdevhw driver looks for that
> symlink when scanning for devices, so add a parent node for atafb.

Do you really have these devices to test with?

If so, could you also replace the hard coded fb_info with a call to 
framebuffer_alloc()?

>
> Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
> Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
> Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
> Cc: <linux-fbdev@vger.kernel.org>
> Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
> ---
>   drivers/video/fbdev/atafb.c | 5 +++++
>   1 file changed, 5 insertions(+)
>
> diff --git a/drivers/video/fbdev/atafb.c b/drivers/video/fbdev/atafb.c
> index b8ed1c537293..3d540803dc2b 100644
> --- a/drivers/video/fbdev/atafb.c
> +++ b/drivers/video/fbdev/atafb.c
> @@ -3101,6 +3101,11 @@ static int __init atafb_probe(struct platform_device *pdev)
>   #endif /* ATAFB_EXT */
>   
>   //	strcpy(fb_info.mode->name, "Atari Builtin ");
> +	/* Parent the fb device properly: without this fb0 registers as a
> +	 * virtual sysfs device with no /sys/class/graphics/fb0/device link,
> +	 * which makes Xorg's fbdevhw reject it ("No devices detected").
> +	 */
> +	fb_info.device = &pdev->dev;

Because this is supposed to happen in framebuffer_alloc() and keeping it 
there would be preferable.

Best regards
Thomas

>   	fb_info.fbops = &atafb_ops;
>   	// try to set default (detected; requested) var
>   	do_fb_set_var(&atafb_predefined[default_par - 1], 1);

-- 
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstr. 146, 90461 Nürnberg, Germany, www.suse.com
GF: Jochen Jaser, Andrew McDonald, (HRB 36809, AG Nürnberg)

Re: [PATCH v2 2/5] fbdev: Give atafb proper parent

[Michael Schmitz]

Hi Thomas,

Am 17.08.2026 um 18:34 schrieb Thomas Zimmermann:
> Hi
>
> Am 16.08.26 um 09:37 schrieb Michael Schmitz:
>> From: Miro Kropacek <miro.kropacek@gmail.com>
>>
>> The atafb fb device registers no parent, causing a missing symlink
>> (/sys/class/graphics/fb0/device). Xorg fbdevhw driver looks for that
>> symlink when scanning for devices, so add a parent node for atafb.
>
> Do you really have these devices to test with?

Not the particular hardware that patches 3 and 4 add support for, but 
the 'normal' Videl that is found in every Falcon, yes.

> If so, could you also replace the hard coded fb_info with a call to
> framebuffer_alloc()?

Can't see how that would work - fb_info isn't passed to all functions 
that need to use it, especially in the hardware detection phase (hence 
the use of a static fb_info struct)?

Cheers,

	Michael

>>
>> Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
>> Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
>> Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
>> Cc: <linux-fbdev@vger.kernel.org>
>> Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
>> ---
>>   drivers/video/fbdev/atafb.c | 5 +++++
>>   1 file changed, 5 insertions(+)
>>
>> diff --git a/drivers/video/fbdev/atafb.c b/drivers/video/fbdev/atafb.c
>> index b8ed1c537293..3d540803dc2b 100644
>> --- a/drivers/video/fbdev/atafb.c
>> +++ b/drivers/video/fbdev/atafb.c
>> @@ -3101,6 +3101,11 @@ static int __init atafb_probe(struct
>> platform_device *pdev)
>>   #endif /* ATAFB_EXT */
>>     //    strcpy(fb_info.mode->name, "Atari Builtin ");
>> +    /* Parent the fb device properly: without this fb0 registers as a
>> +     * virtual sysfs device with no /sys/class/graphics/fb0/device link,
>> +     * which makes Xorg's fbdevhw reject it ("No devices detected").
>> +     */
>> +    fb_info.device = &pdev->dev;
>
> Because this is supposed to happen in framebuffer_alloc() and keeping it
> there would be preferable.
>
> Best regards
> Thomas
>
>>       fb_info.fbops = &atafb_ops;
>>       // try to set default (detected; requested) var
>>       do_fb_set_var(&atafb_predefined[default_par - 1], 1);
>

Re: [PATCH v2 2/5] fbdev: Give atafb proper parent

[Thomas Zimmermann]

Am 17.08.26 um 09:49 schrieb Michael Schmitz:
> Hi Thomas,
>
> Am 17.08.2026 um 18:34 schrieb Thomas Zimmermann:
>> Hi
>>
>> Am 16.08.26 um 09:37 schrieb Michael Schmitz:
>>> From: Miro Kropacek <miro.kropacek@gmail.com>
>>>
>>> The atafb fb device registers no parent, causing a missing symlink
>>> (/sys/class/graphics/fb0/device). Xorg fbdevhw driver looks for that
>>> symlink when scanning for devices, so add a parent node for atafb.
>>
>> Do you really have these devices to test with?
>
> Not the particular hardware that patches 3 and 4 add support for, but 
> the 'normal' Videl that is found in every Falcon, yes.
>
>> If so, could you also replace the hard coded fb_info with a call to
>> framebuffer_alloc()?
>
> Can't see how that would work - fb_info isn't passed to all functions 
> that need to use it, especially in the hardware detection phase (hence 
> the use of a static fb_info struct)?

You'd have to modify the driver to pass the allocated instance, of 
course. It's not a trivial oneliner, but if this driver is still 
relevant, it might be worth improving it.

Best regards
Thomas

>
> Cheers,
>
>     Michael
>
>>>
>>> Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
>>> Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
>>> Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
>>> Cc: <linux-fbdev@vger.kernel.org>
>>> Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
>>> ---
>>>   drivers/video/fbdev/atafb.c | 5 +++++
>>>   1 file changed, 5 insertions(+)
>>>
>>> diff --git a/drivers/video/fbdev/atafb.c b/drivers/video/fbdev/atafb.c
>>> index b8ed1c537293..3d540803dc2b 100644
>>> --- a/drivers/video/fbdev/atafb.c
>>> +++ b/drivers/video/fbdev/atafb.c
>>> @@ -3101,6 +3101,11 @@ static int __init atafb_probe(struct
>>> platform_device *pdev)
>>>   #endif /* ATAFB_EXT */
>>>     //    strcpy(fb_info.mode->name, "Atari Builtin ");
>>> +    /* Parent the fb device properly: without this fb0 registers as a
>>> +     * virtual sysfs device with no /sys/class/graphics/fb0/device 
>>> link,
>>> +     * which makes Xorg's fbdevhw reject it ("No devices detected").
>>> +     */
>>> +    fb_info.device = &pdev->dev;
>>
>> Because this is supposed to happen in framebuffer_alloc() and keeping it
>> there would be preferable.
>>
>> Best regards
>> Thomas
>>
>>>       fb_info.fbops = &atafb_ops;
>>>       // try to set default (detected; requested) var
>>>       do_fb_set_var(&atafb_predefined[default_par - 1], 1);
>>
>

-- 
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstr. 146, 90461 Nürnberg, Germany, www.suse.com
GF: Jochen Jaser, Andrew McDonald, (HRB 36809, AG Nürnberg)

Re: [PATCH v2 2/5] fbdev: Give atafb proper parent

[Michael Schmitz]

Hi Thomas,

Am 17.08.2026 um 19:57 schrieb Thomas Zimmermann:
>>>> The atafb fb device registers no parent, causing a missing symlink
>>>> (/sys/class/graphics/fb0/device). Xorg fbdevhw driver looks for that
>>>> symlink when scanning for devices, so add a parent node for atafb.
>>>
>>> Do you really have these devices to test with?
>>
>> Not the particular hardware that patches 3 and 4 add support for, but
>> the 'normal' Videl that is found in every Falcon, yes.

N.B.: actual hardware won't be essential to test - an emulator such as 
ARAnyM will do fine.

>>
>>> If so, could you also replace the hard coded fb_info with a call to
>>> framebuffer_alloc()?
>>
>> Can't see how that would work - fb_info isn't passed to all functions
>> that need to use it, especially in the hardware detection phase (hence
>> the use of a static fb_info struct)?

On second thought, depending on how early in driver init fb_info is 
needed, I may get away with turning it into a struct *fb_info and using 
that where it can't easily be passed.

>
> You'd have to modify the driver to pass the allocated instance, of
> course. It's not a trivial oneliner, but if this driver is still
> relevant, it might be worth improving it.

Your point is taken - I'll take a look at what would be required to pass 
fb_info to all users. It's been nineteen years since rewrote atafb so 
getting my head around the call sequences may take awhile.

Cheers,

	Michael

>
> Best regards
> Thomas
>
>>
>> Cheers,
>>
>>     Michael
>>
>>>>
>>>> Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
>>>> Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
>>>> Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
>>>> Cc: <linux-fbdev@vger.kernel.org>
>>>> Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
>>>> ---
>>>>   drivers/video/fbdev/atafb.c | 5 +++++
>>>>   1 file changed, 5 insertions(+)
>>>>
>>>> diff --git a/drivers/video/fbdev/atafb.c b/drivers/video/fbdev/atafb.c
>>>> index b8ed1c537293..3d540803dc2b 100644
>>>> --- a/drivers/video/fbdev/atafb.c
>>>> +++ b/drivers/video/fbdev/atafb.c
>>>> @@ -3101,6 +3101,11 @@ static int __init atafb_probe(struct
>>>> platform_device *pdev)
>>>>   #endif /* ATAFB_EXT */
>>>>     //    strcpy(fb_info.mode->name, "Atari Builtin ");
>>>> +    /* Parent the fb device properly: without this fb0 registers as a
>>>> +     * virtual sysfs device with no /sys/class/graphics/fb0/device
>>>> link,
>>>> +     * which makes Xorg's fbdevhw reject it ("No devices detected").
>>>> +     */
>>>> +    fb_info.device = &pdev->dev;
>>>
>>> Because this is supposed to happen in framebuffer_alloc() and keeping it
>>> there would be preferable.
>>>
>>> Best regards
>>> Thomas
>>>
>>>>       fb_info.fbops = &atafb_ops;
>>>>       // try to set default (detected; requested) var
>>>>       do_fb_set_var(&atafb_predefined[default_par - 1], 1);
>>>
>>
>

[PATCH v2 3/5] fbdev: Add support for further video bit depths on atafb:external

[Michael Schmitz]

From: Miro Kropacek <miro.kropacek@gmail.com>

Supervidel offers additional video bit depths (8-bit chunky, 16-bit
RGB565 (also on the originaln Videl) and ARGB888.

Add code to support these bit depths.

Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Cc: <linux-fbdev@vger.kernel.org>
Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
---
 drivers/video/fbdev/atafb.c | 89 ++++++++++++++++++++++++++++++++++---
 1 file changed, 82 insertions(+), 7 deletions(-)

diff --git a/drivers/video/fbdev/atafb.c b/drivers/video/fbdev/atafb.c
index 3d540803dc2b..50853d7a08e3 100644
--- a/drivers/video/fbdev/atafb.c
+++ b/drivers/video/fbdev/atafb.c
@@ -2081,8 +2081,12 @@ static int ext_encode_fix(struct fb_fix_screeninfo *fix, struct atafb_par *par)
 			 external_pmode == FB_TYPE_PACKED_PIXELS) ?
 				FB_VISUAL_MONO10 : FB_VISUAL_MONO01;
 	} else {
-		/* Use STATIC if we don't know how to access color registers */
-		int visual = external_vgaiobase ?
+		/* Use STATIC if we don't know how to access color registers;
+		 * SuperVidel 8bpp chunky (fb in SV RAM) uses the Falcon palette
+		 */
+		int visual = (external_vgaiobase ||
+			      (external_depth == 8 &&
+			       external_addr >= 0xa0000000)) ?
 					 FB_VISUAL_PSEUDOCOLOR :
 					 FB_VISUAL_STATIC_PSEUDOCOLOR;
 		switch (external_pmode) {
@@ -2159,6 +2163,35 @@ static int ext_encode_var(struct fb_var_screeninfo *var, struct atafb_par *par)
 	var->transp.offset = 0;
 	var->transp.length = 0;
 	var->transp.msb_right = 0;
+	if (external_pmode == -1 && external_depth == 16) {
+		/* RGB565 truecolor (e.g. SuperVidel native mode) */
+		var->red.offset = 11;
+		var->red.length = 5;
+		var->green.offset = 5;
+		var->green.length = 6;
+		var->blue.offset = 0;
+		var->blue.length = 5;
+	} else if (external_pmode == -1 && external_depth == 32) {
+		/* ARGB8888 truecolor (e.g. SuperVidel native mode) */
+		var->red.offset = 16;
+		var->red.length = 8;
+		var->green.offset = 8;
+		var->green.length = 8;
+		var->blue.offset = 0;
+		var->blue.length = 8;
+		var->transp.offset = 24;
+		var->transp.length = 8;
+	} else if (external_pmode == FB_TYPE_PACKED_PIXELS &&
+		   external_depth == 8 && external_addr >= 0xa0000000) {
+		/* SuperVidel 8bpp chunky: palette has 8 bits per channel.
+		 * Without this, fb_get_color_depth() sees length 0 and
+		 * fbcon falls back to its 2-color palette — the console
+		 * text (color 7) stays black on black.
+		 */
+		var->red.length = 8;
+		var->green.length = 8;
+		var->blue.length = 8;
+	}
 	var->yres_virtual = var->yres;
 	var->xoffset = 0;
 	var->yoffset = 0;
@@ -2193,6 +2226,38 @@ static int ext_setcolreg(unsigned int regno, unsigned int red,
 {
 	unsigned char colmask = (1 << external_bitspercol) - 1;
 
+	if (external_pmode == -1 && external_depth == 16) {
+		/* truecolor: only the pseudo palette for fbcon is needed */
+		if (regno > 15)
+			return 1;
+		((u32 *)info->pseudo_palette)[regno] = (red & 0xf800) |
+						       ((green & 0xfc00) >> 5) |
+						       ((blue & 0xf800) >> 11);
+		return 0;
+	}
+	if (external_pmode == -1 && external_depth == 32) {
+		/* ARGB8888, alpha forced opaque */
+		if (regno > 15)
+			return 1;
+		((u32 *)info->pseudo_palette)[regno] = 0xff000000 |
+						       ((red & 0xff00) << 8) |
+						       (green & 0xff00) |
+						       ((blue & 0xff00) >> 8);
+		return 0;
+	}
+	if (external_pmode == FB_TYPE_PACKED_PIXELS && external_depth == 8 &&
+	    external_addr >= 0xa0000000) {
+		/* SuperVidel native 8bpp chunky scans out via the Falcon
+		 * palette registers, honoring all 8 bits per channel
+		 */
+		if (regno > 255)
+			return 1;
+		f030_col[regno] = ((red & 0xff00) << 16) |
+				  ((green & 0xff00) << 8) |
+				  ((blue & 0xff00) >> 8);
+		return 0;
+	}
+
 	if (!external_vgaiobase)
 		return 1;
 
@@ -2422,7 +2487,9 @@ static void atafb_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
 		return;
 
 #ifdef ATAFB_FALCON
-	if (info->var.bits_per_pixel == 16) {
+	/* chunky modes (Falcon hicolor, external packed/truecolor) */
+	if (info->fix.type == FB_TYPE_PACKED_PIXELS &&
+	    info->var.bits_per_pixel > 1) {
 		cfb_fillrect(info, rect);
 		return;
 	}
@@ -2463,7 +2530,9 @@ static void atafb_copyarea(struct fb_info *info, const struct fb_copyarea *area)
 	int rev_copy = 0;
 
 #ifdef ATAFB_FALCON
-	if (info->var.bits_per_pixel == 16) {
+	/* chunky modes (Falcon hicolor, external packed/truecolor) */
+	if (info->fix.type == FB_TYPE_PACKED_PIXELS &&
+	    info->var.bits_per_pixel > 1) {
 		cfb_copyarea(info, area);
 		return;
 	}
@@ -2517,7 +2586,9 @@ static void atafb_imageblit(struct fb_info *info, const struct fb_image *image)
 	u32 dx, dy, width, height, pitch;
 
 #ifdef ATAFB_FALCON
-	if (info->var.bits_per_pixel == 16) {
+	/* chunky modes (Falcon hicolor, external packed/truecolor) */
+	if (info->fix.type == FB_TYPE_PACKED_PIXELS &&
+	    info->var.bits_per_pixel > 1) {
 		cfb_imageblit(info, image);
 		return;
 	}
@@ -2753,7 +2824,7 @@ static void __init atafb_setup_ext(char *spec)
 		return;
 	depth = simple_strtoul(p, NULL, 10);
 	if (depth != 1 && depth != 2 && depth != 4 && depth != 8 &&
-	    depth != 16 && depth != 24)
+	    depth != 16 && depth != 24 && depth != 32)
 		return;
 
 	p = strsep(&spec, ";");
@@ -3131,7 +3202,11 @@ static int __init atafb_probe(struct platform_device *pdev)
 
 	atafb_set_disp(&fb_info);
 
-	fb_alloc_cmap(&(fb_info.cmap), 1 << fb_info.var.bits_per_pixel, 0);
+	/* truecolor visuals only need the 16-entry console palette; this
+	 * also avoids 1 << 32 overflowing at 32bpp
+	 */
+	fb_alloc_cmap(&(fb_info.cmap), fb_info.var.bits_per_pixel > 8 ?
+		      16 : 1 << fb_info.var.bits_per_pixel, 0);
 
 
 	dev_info(&pdev->dev, "Determined %dx%d, depth %d\n", fb_info.var.xres,
-- 
2.17.1

[PATCH v2 4/5] fbdev: Add support for SuperVidel's SuperBlitter

[Michael Schmitz]

From: Miro Kropacek <miro.kropacek@gmail.com>

The SuperVidel graphics FPGA includes a hardware blitter (bit block
transfer engine) operating within the SuperVidel DDR2 video RAM.

Two versions of this blitter exist. Later versions (>= 9) of the
SuperVidel firmware support an asynchronous command FIFO, older
versions must be polled for command completion.

Add hardware-accelerated copyarea, fillrect and imageblit fb
operations (falling back to the non-accelerated versions for
anything that exceeds blitter capabilities).

Signed-off-by: Miro Kropacek <miro.kropacek@gmail.com>
Reviewed-by: Michael Schmitz <schmitzmic@gmail.com>
Signed-off-by: Michael Schmitz <schmitzmic@gmail.com>
Cc: <linux-fbdev@vger.kernel.org>
Link: https://lists.debian.org/debian-68k/2026/08/msg00000.html
---
 drivers/video/fbdev/atafb.c | 197 ++++++++++++++++++++++++++++++++++++
 1 file changed, 197 insertions(+)

diff --git a/drivers/video/fbdev/atafb.c b/drivers/video/fbdev/atafb.c
index 50853d7a08e3..dc2d4154e1b6 100644
--- a/drivers/video/fbdev/atafb.c
+++ b/drivers/video/fbdev/atafb.c
@@ -2304,6 +2304,185 @@ static int ext_detect(void)
 	return 1;
 }
 
+/* ------------------- SuperVidel SuperBlitter ---------------------- */
+
+/*
+ * Hardware blitter in the SuperVidel FPGA, operating within SV DDR2 RAM.
+ * FW revision >= 9 provides a command FIFO (async operation); older
+ * revisions are programmed directly with busy-polling.
+ */
+#define SVBLIT_REGS_PHYS	0x80010000
+#define SVBLIT_SRC1		0x58	/* bits 26:0 */
+#define SVBLIT_SRC2		0x5c
+#define SVBLIT_DST		0x60
+#define SVBLIT_COUNT		0x64	/* bytes per line - 1 */
+#define SVBLIT_SRC1_OFFSET	0x68	/* line start to next line start */
+#define SVBLIT_SRC2_OFFSET	0x6c
+#define SVBLIT_DST_OFFSET	0x70
+#define SVBLIT_MASK_AND_LINES	0x74	/* bits 11:0: number of lines */
+#define SVBLIT_CONTROL		0x78	/* bit 0: busy/start, bits 4:1: mode */
+#define SVBLIT_VERSION		0x7c	/* bits 9:0: FW revision */
+#define SVBLIT_FIFO		0x80	/* wr: data; rd: bit 0 empty, bit 1 full */
+
+/*
+ * SuperBlitter bug: Instead of declared 2048 bytes, 2032 is the real maximum.
+ */
+#define SVBLIT_MAX_SPAN	2032
+
+static void __iomem *svblit_regs;
+static int svblit_fw;
+
+static inline u32 svblit_rd(unsigned int reg)
+{
+	return __raw_readl(svblit_regs + reg);
+}
+
+static inline void svblit_wr(unsigned int reg, u32 val)
+{
+	__raw_writel(val, svblit_regs + reg);
+}
+
+/* wait until all queued blits have finished */
+static void svblit_wait(void)
+{
+	if (svblit_fw >= 9)
+		/* FIFO empty flag = fewer than 9 longwords queued */
+		while (!(svblit_rd(SVBLIT_FIFO) & 1))
+			cpu_relax();
+	while (svblit_rd(SVBLIT_CONTROL) & 1)
+		cpu_relax();
+}
+
+/*
+ * FW >= 9 queues commands through the 512-longword FIFO: a command is
+ * 9 longwords (registers 0x58..0x78 in order), executed whenever >= 9
+ * words are queued and the blitter is idle. The full flag rises at
+ * >= 500 queued words, so below it there is always room for a whole
+ * command — one flag check per command prevents overflow (dropped
+ * words would desync the 9-word framing until an SV reinit, which is
+ * exactly what overflowing did before this guard existed). Older FW
+ * is programmed directly with busy-polling.
+ *
+ * The line byte count field is 11 bits but see SVBLIT_MAX_SPAN.
+ */
+static void svblit_copy(u32 src, u32 dst, u32 nbytes, u32 src_offset,
+			u32 dst_offset, u32 lines)
+{
+	while (nbytes) {
+		u32 chunk = min(nbytes, SVBLIT_MAX_SPAN);
+
+		if (svblit_fw >= 9) {
+			while (svblit_rd(SVBLIT_FIFO) & 2)
+				cpu_relax();
+			svblit_wr(SVBLIT_FIFO, src);
+			svblit_wr(SVBLIT_FIFO, 0);
+			svblit_wr(SVBLIT_FIFO, dst);
+			svblit_wr(SVBLIT_FIFO, chunk - 1);
+			svblit_wr(SVBLIT_FIFO, src_offset);
+			svblit_wr(SVBLIT_FIFO, 0);
+			svblit_wr(SVBLIT_FIFO, dst_offset);
+			svblit_wr(SVBLIT_FIFO, lines);
+			svblit_wr(SVBLIT_FIFO, 0x01);
+		} else {
+			while (svblit_rd(SVBLIT_CONTROL) & 1)
+				cpu_relax();
+			svblit_wr(SVBLIT_SRC1, src);
+			svblit_wr(SVBLIT_SRC2, 0);
+			svblit_wr(SVBLIT_DST, dst);
+			svblit_wr(SVBLIT_COUNT, chunk - 1);
+			svblit_wr(SVBLIT_SRC1_OFFSET, src_offset);
+			svblit_wr(SVBLIT_SRC2_OFFSET, 0);
+			svblit_wr(SVBLIT_DST_OFFSET, dst_offset);
+			svblit_wr(SVBLIT_MASK_AND_LINES, lines);
+			svblit_wr(SVBLIT_CONTROL, 0x01);
+		}
+
+		src += chunk;
+		dst += chunk;
+		nbytes -= chunk;
+	}
+}
+
+static int svblit_sync(struct fb_info *info)
+{
+	svblit_wait();
+	return 0;
+}
+
+static void svblit_copyarea(struct fb_info *info,
+			    const struct fb_copyarea *area)
+{
+	u32 bytespp = info->var.bits_per_pixel / 8;
+	u32 pitch = info->fix.line_length;
+
+	/*
+	 * The blitter walks lines in ascending order, so overlapping
+	 * moves down/right would read already overwritten data. Those
+	 * are rare for fbcon (scrolling backwards); leave them and
+	 * oversized areas to the CPU.
+	 */
+	if (area->height > 4095 ||
+	    area->dy > area->sy ||
+	    (area->dy == area->sy && area->dx > area->sx)) {
+		svblit_wait();
+		cfb_copyarea(info, area);
+		return;
+	}
+
+	svblit_copy(external_addr + area->sy * pitch + area->sx * bytespp,
+		    external_addr + area->dy * pitch + area->dx * bytespp,
+		    area->width * bytespp, pitch, pitch, area->height);
+	/* async: every CPU access to the fb goes through svblit_wait() */
+}
+
+static void svblit_fillrect(struct fb_info *info,
+			    const struct fb_fillrect *rect)
+{
+	u32 bytespp = info->var.bits_per_pixel / 8;
+	u32 pitch = info->fix.line_length;
+	u8 *line;
+	u32 pix;
+
+	svblit_wait();		/* the CPU is about to touch the fb */
+
+	if (rect->rop != ROP_COPY || rect->height <= 1 ||
+	    rect->height > 4096) {
+		cfb_fillrect(info, rect);
+		return;
+	}
+
+	pix = (info->fix.visual == FB_VISUAL_TRUECOLOR) ?
+		((u32 *)info->pseudo_palette)[rect->color] : rect->color;
+
+	/* draw the first line with the CPU ... */
+	line = (u8 *)info->screen_base + rect->dy * pitch +
+	       rect->dx * bytespp;
+	switch (bytespp) {
+	case 1:
+		memset(line, pix, rect->width);
+		break;
+	case 2:
+		memset16((u16 *)line, pix, rect->width);
+		break;
+	default:
+		memset32((u32 *)line, pix, rect->width);
+		break;
+	}
+
+	/* ... and let the blitter replicate it into the other lines */
+	svblit_copy(external_addr + rect->dy * pitch + rect->dx * bytespp,
+		    external_addr + (rect->dy + 1) * pitch +
+		    rect->dx * bytespp,
+		    rect->width * bytespp, 0, pitch, rect->height - 1);
+}
+
+static void svblit_imageblit(struct fb_info *info,
+			     const struct fb_image *image)
+{
+	svblit_wait();		/* CPU rendering must not race queued blits */
+	cfb_imageblit(info, image);
+}
+
 #endif /* ATAFB_EXT */
 
 /* ------ This is the same for most hardware types -------- */
@@ -3168,6 +3347,24 @@ static int __init atafb_probe(struct platform_device *pdev)
 		phys_screen_base = external_addr;
 		screen_len = external_len & PAGE_MASK;
 		memset (screen_base, 0, external_len);
+
+		/* framebuffer in SV RAM: enable the SuperBlitter */
+		if (external_addr >= 0xa0000000) {
+			svblit_regs = ioremap(SVBLIT_REGS_PHYS, 0x100);
+			if (svblit_regs) {
+				svblit_fw = svblit_rd(SVBLIT_VERSION) & 0x1ff;
+				atafb_ops.fb_fillrect = svblit_fillrect;
+				atafb_ops.fb_copyarea = svblit_copyarea;
+				atafb_ops.fb_imageblit = svblit_imageblit;
+				atafb_ops.fb_sync = svblit_sync;
+				fb_info.flags |= FBINFO_HWACCEL_COPYAREA |
+						 FBINFO_HWACCEL_FILLRECT;
+				dev_info(&pdev->dev,
+					 "SuperBlitter enabled, FW revision %d (%s)\n",
+					 svblit_fw, svblit_fw >= 9 ?
+					 "async FIFO" : "sync");
+			}
+		}
 	}
 #endif /* ATAFB_EXT */
 
-- 
2.17.1