[U-Boot] [PATCH] video: sunxi: Enable cache for the framebuffer

[U-Boot] [PATCH] video: sunxi: Enable cache for the framebuffer

[Siarhei Siamashka]

Because the framebuffer resides in a hidden chopped off part of DRAM,
caching is not enabled for this area automatically when the MMU is
configured. So it needs to be set explicitly.

Additionally, the cfb console has a function to check whether the
framebuffer resides in DRAM or not. And the hidden sunxi framebuffer
needs special treatment again.

This fix resolves the sluggish scrolling problem.

Signed-off-by: Siarhei Siamashka <siarhei.siamashka@gmail.com>
---
 drivers/video/cfb_console.c   |  2 +-
 drivers/video/sunxi_display.c | 10 ++++++++++
 2 files changed, 11 insertions(+), 1 deletion(-)

diff --git a/drivers/video/cfb_console.c b/drivers/video/cfb_console.c
index a81affa..23b50df 100644
--- a/drivers/video/cfb_console.c
+++ b/drivers/video/cfb_console.c
@@ -2090,7 +2090,7 @@ static void *video_logo(void)
 }
 #endif
 
-static int cfb_fb_is_in_dram(void)
+__weak int cfb_fb_is_in_dram(void)
 {
 	bd_t *bd = gd->bd;
 #if defined(CONFIG_ARM) || defined(CONFIG_AVR32) || defined(COFNIG_NDS32) || \
diff --git a/drivers/video/sunxi_display.c b/drivers/video/sunxi_display.c
index f5f24fc..22a1319 100644
--- a/drivers/video/sunxi_display.c
+++ b/drivers/video/sunxi_display.c
@@ -1060,6 +1060,12 @@ static const char *sunxi_get_mon_desc(enum sunxi_monitor monitor)
 	return NULL; /* never reached */
 }
 
+/* The framebuffer is in DRAM, even though it is in the hidden part of it */
+int cfb_fb_is_in_dram(void)
+{
+	return 1;
+}
+
 void *video_hw_init(void)
 {
 	static GraphicDevice *graphic_device = &sunxi_display.graphic_device;
@@ -1079,6 +1085,10 @@ void *video_hw_init(void)
 	       CONFIG_SUNXI_FB_SIZE >> 10);
 	gd->fb_base = gd->ram_top;
 
+	mmu_set_region_dcache_behaviour(gd->fb_base,
+					CONFIG_SUNXI_FB_SIZE,
+					DCACHE_WRITEBACK);
+
 	video_get_ctfb_res_modes(RES_MODE_1024x768, 24, &mode,
 				 &sunxi_display.depth, &options);
 #ifdef CONFIG_VIDEO_HDMI
-- 
2.0.5

[U-Boot] [PATCH] video: sunxi: Enable cache for the framebuffer

[Ian Campbell]

On Sat, 2015-01-31 at 03:21 +0200, Siarhei Siamashka wrote:
> Because the framebuffer resides in a hidden chopped off part of DRAM,
> caching is not enabled for this area automatically when the MMU is
> configured. So it needs to be set explicitly.
> 
> Additionally, the cfb console has a function to check whether the
> framebuffer resides in DRAM or not. And the hidden sunxi framebuffer
> needs special treatment again.

I wonder if this (essentially the need to make cfb_fb_is_in_dram __weak)
suggests that we are not correctly doing this hiding? Or at least not
doing it the same as other platforms which steal DRAM for use as video
ram.

We seem to set pGD->frameAdrs (==VIDEO_FB_ADRS in cfb_console)
correctly, which is one half of cfb_fb_is_in_dram's input. So is the
real issue that we omit the fb from bd->bi_dram instead of including it
there and marking it reserved in another way?

I have a look around some of the other drivers for other platforms and I
must confess I couldn't figure out how this all fits together for
them...

In terms of the kernel interface I'd expect we want to use an FDT
memeserve for the region, i.e. call fdt_add_reservemap_entry at some
suitable place.

Anatolij, how are we supposed to be doing this?

> 
> This fix resolves the sluggish scrolling problem.
> 
> Signed-off-by: Siarhei Siamashka <siarhei.siamashka@gmail.com>
> ---
>  drivers/video/cfb_console.c   |  2 +-
>  drivers/video/sunxi_display.c | 10 ++++++++++
>  2 files changed, 11 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/video/cfb_console.c b/drivers/video/cfb_console.c
> index a81affa..23b50df 100644
> --- a/drivers/video/cfb_console.c
> +++ b/drivers/video/cfb_console.c
> @@ -2090,7 +2090,7 @@ static void *video_logo(void)
>  }
>  #endif
>  
> -static int cfb_fb_is_in_dram(void)
> +__weak int cfb_fb_is_in_dram(void)
>  {
>  	bd_t *bd = gd->bd;
>  #if defined(CONFIG_ARM) || defined(CONFIG_AVR32) || defined(COFNIG_NDS32) || \
> diff --git a/drivers/video/sunxi_display.c b/drivers/video/sunxi_display.c
> index f5f24fc..22a1319 100644
> --- a/drivers/video/sunxi_display.c
> +++ b/drivers/video/sunxi_display.c
> @@ -1060,6 +1060,12 @@ static const char *sunxi_get_mon_desc(enum sunxi_monitor monitor)
>  	return NULL; /* never reached */
>  }
>  
> +/* The framebuffer is in DRAM, even though it is in the hidden part of it */
> +int cfb_fb_is_in_dram(void)
> +{
> +	return 1;
> +}
> +
>  void *video_hw_init(void)
>  {
>  	static GraphicDevice *graphic_device = &sunxi_display.graphic_device;
> @@ -1079,6 +1085,10 @@ void *video_hw_init(void)
>  	       CONFIG_SUNXI_FB_SIZE >> 10);
>  	gd->fb_base = gd->ram_top;
>  
> +	mmu_set_region_dcache_behaviour(gd->fb_base,
> +					CONFIG_SUNXI_FB_SIZE,
> +					DCACHE_WRITEBACK);
> +
>  	video_get_ctfb_res_modes(RES_MODE_1024x768, 24, &mode,
>  				 &sunxi_display.depth, &options);
>  #ifdef CONFIG_VIDEO_HDMI

[U-Boot] [PATCH] video: sunxi: Enable cache for the framebuffer

[Siarhei Siamashka]

On Sat, 31 Jan 2015 10:34:36 +0000
Ian Campbell <ijc@hellion.org.uk> wrote:

> On Sat, 2015-01-31 at 03:21 +0200, Siarhei Siamashka wrote:
> > Because the framebuffer resides in a hidden chopped off part of DRAM,
> > caching is not enabled for this area automatically when the MMU is
> > configured. So it needs to be set explicitly.
> > 
> > Additionally, the cfb console has a function to check whether the
> > framebuffer resides in DRAM or not. And the hidden sunxi framebuffer
> > needs special treatment again.
> 
> I wonder if this (essentially the need to make cfb_fb_is_in_dram __weak)
> suggests that we are not correctly doing this hiding? Or at least not
> doing it the same as other platforms which steal DRAM for use as video
> ram.
> 
> We seem to set pGD->frameAdrs (==VIDEO_FB_ADRS in cfb_console)
> correctly, which is one half of cfb_fb_is_in_dram's input. So is the
> real issue that we omit the fb from bd->bi_dram instead of including it
> there and marking it reserved in another way?

Since sunxi is using the following feature for the framebuffer
reservation, we are indeed not doing it correctly (or at least
doing something that does not match the current documentation):

- CONFIG_SYS_MEM_TOP_HIDE (PPC only):
                If CONFIG_SYS_MEM_TOP_HIDE is defined in the board config header,
                this specified memory area will get subtracted from the top
                (end) of RAM and won't get "touched" at all by U-Boot. By
                fixing up gd->ram_size the Linux kernel should gets passed
                the now "corrected" memory size and won't touch it either.
                This should work for arch/ppc and arch/powerpc. Only Linux
                board ports in arch/powerpc with bootwrapper support that
                recalculate the memory size from the SDRAM controller setup
                will have to get fixed in Linux additionally.

                This option can be used as a workaround for the 440EPx/GRx
                CHIP 11 errata where the last 256 bytes in SDRAM shouldn't
                be touched.

                WARNING: Please make sure that this value is a multiple of
                the Linux page size (normally 4k). If this is not the case,
                then the end address of the Linux memory will be located at a
                non page size aligned address and this could cause major
                problems.

Trying to to add the sunxi framebuffer into more u-boot structures is
going to violate the promise, stating that "this specified memory area
won't get touched at all by U-Boot". And by doing so, violate it even
more than it is already violated by sunxi_display and cfb_console.

> I have a look around some of the other drivers for other platforms and I
> must confess I couldn't figure out how this all fits together for
> them...
> 
> In terms of the kernel interface I'd expect we want to use an FDT
> memeserve for the region, i.e. call fdt_add_reservemap_entry at some
> suitable place.
>
> Anatolij, how are we supposed to be doing this?
> 
> > 
> > This fix resolves the sluggish scrolling problem.
> > 
> > Signed-off-by: Siarhei Siamashka <siarhei.siamashka@gmail.com>
> > ---
> >  drivers/video/cfb_console.c   |  2 +-
> >  drivers/video/sunxi_display.c | 10 ++++++++++
> >  2 files changed, 11 insertions(+), 1 deletion(-)
> > 
> > diff --git a/drivers/video/cfb_console.c b/drivers/video/cfb_console.c
> > index a81affa..23b50df 100644
> > --- a/drivers/video/cfb_console.c
> > +++ b/drivers/video/cfb_console.c
> > @@ -2090,7 +2090,7 @@ static void *video_logo(void)
> >  }
> >  #endif
> >  
> > -static int cfb_fb_is_in_dram(void)
> > +__weak int cfb_fb_is_in_dram(void)
> >  {
> >  	bd_t *bd = gd->bd;
> >  #if defined(CONFIG_ARM) || defined(CONFIG_AVR32) || defined(COFNIG_NDS32) || \
> > diff --git a/drivers/video/sunxi_display.c b/drivers/video/sunxi_display.c
> > index f5f24fc..22a1319 100644
> > --- a/drivers/video/sunxi_display.c
> > +++ b/drivers/video/sunxi_display.c
> > @@ -1060,6 +1060,12 @@ static const char *sunxi_get_mon_desc(enum sunxi_monitor monitor)
> >  	return NULL; /* never reached */
> >  }
> >  
> > +/* The framebuffer is in DRAM, even though it is in the hidden part of it */
> > +int cfb_fb_is_in_dram(void)
> > +{
> > +	return 1;
> > +}
> > +
> >  void *video_hw_init(void)
> >  {
> >  	static GraphicDevice *graphic_device = &sunxi_display.graphic_device;
> > @@ -1079,6 +1085,10 @@ void *video_hw_init(void)
> >  	       CONFIG_SUNXI_FB_SIZE >> 10);
> >  	gd->fb_base = gd->ram_top;
> >  
> > +	mmu_set_region_dcache_behaviour(gd->fb_base,
> > +					CONFIG_SUNXI_FB_SIZE,
> > +					DCACHE_WRITEBACK);
> > +
> >  	video_get_ctfb_res_modes(RES_MODE_1024x768, 24, &mode,
> >  				 &sunxi_display.depth, &options);
> >  #ifdef CONFIG_VIDEO_HDMI
> 
> 



-- 
Best regards,
Siarhei Siamashka

[U-Boot] [PATCH] video: sunxi: Enable cache for the framebuffer

[Hans de Goede]

Hi,

On 31-01-15 02:21, Siarhei Siamashka wrote:
> Because the framebuffer resides in a hidden chopped off part of DRAM,
> caching is not enabled for this area automatically when the MMU is
> configured. So it needs to be set explicitly.
>
> Additionally, the cfb console has a function to check whether the
> framebuffer resides in DRAM or not. And the hidden sunxi framebuffer
> needs special treatment again.
>
> This fix resolves the sluggish scrolling problem.
>
> Signed-off-by: Siarhei Siamashka <siarhei.siamashka@gmail.com>

Thanks for figuring this out!

As already discussed it seems that the problem is that using
CONFIG_SYS_MEM_TOP_HIDE seems to be the wrong thing to do.

So I've written a patch with results in the same behavior in a slightly
different way, with as added bonus that we can give back unused memory
to the kernel when booting.

I'll Cc you on the patch-set when I send it.

Regards,

Hans

> ---
>   drivers/video/cfb_console.c   |  2 +-
>   drivers/video/sunxi_display.c | 10 ++++++++++
>   2 files changed, 11 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/video/cfb_console.c b/drivers/video/cfb_console.c
> index a81affa..23b50df 100644
> --- a/drivers/video/cfb_console.c
> +++ b/drivers/video/cfb_console.c
> @@ -2090,7 +2090,7 @@ static void *video_logo(void)
>   }
>   #endif
>
> -static int cfb_fb_is_in_dram(void)
> +__weak int cfb_fb_is_in_dram(void)
>   {
>   	bd_t *bd = gd->bd;
>   #if defined(CONFIG_ARM) || defined(CONFIG_AVR32) || defined(COFNIG_NDS32) || \
> diff --git a/drivers/video/sunxi_display.c b/drivers/video/sunxi_display.c
> index f5f24fc..22a1319 100644
> --- a/drivers/video/sunxi_display.c
> +++ b/drivers/video/sunxi_display.c
> @@ -1060,6 +1060,12 @@ static const char *sunxi_get_mon_desc(enum sunxi_monitor monitor)
>   	return NULL; /* never reached */
>   }
>
> +/* The framebuffer is in DRAM, even though it is in the hidden part of it */
> +int cfb_fb_is_in_dram(void)
> +{
> +	return 1;
> +}
> +
>   void *video_hw_init(void)
>   {
>   	static GraphicDevice *graphic_device = &sunxi_display.graphic_device;
> @@ -1079,6 +1085,10 @@ void *video_hw_init(void)
>   	       CONFIG_SUNXI_FB_SIZE >> 10);
>   	gd->fb_base = gd->ram_top;
>
> +	mmu_set_region_dcache_behaviour(gd->fb_base,
> +					CONFIG_SUNXI_FB_SIZE,
> +					DCACHE_WRITEBACK);
> +
>   	video_get_ctfb_res_modes(RES_MODE_1024x768, 24, &mode,
>   				 &sunxi_display.depth, &options);
>   #ifdef CONFIG_VIDEO_HDMI
>