framebuffer for au1000 based board.

Hello, I'm writing framebuffer (800x600@16bpp) for au1000 based board for latest linux-2.6.13 mips kernel. video memory is located at address 0xbe00_0000.
The problem is that I cannot correctly mmap video memory to userspace.

mmap was taken from au1500 lcd framebuffer driver(code follows)

start, fbi->fix.smem_start, off are all equal to 0xbe00_0000 (video memory start)

mmaping from userspace is working ok, but writing to mmaped memory gives no result on display.

code like memset(0xbe00_0000,0,FB_MEM_SIZE) from kernel space is working as expected.

from userspace running test on mmaped fb gives no visible results.

cat /proc/156/maps

00400000-00407000 r-xp 00000000 08:01 27785      /root/fbtest
00446000-00447000 rw-p 00006000 08:01 27785      /root/fbtest
00447000-00449000 rwxp 00447000 00:00 0          [heap]
2aaa8000-2ab93000 rw-s be000000 08:01 17858      /dev/fb0
7fb12000-7fb27000 rwxp 7fb12000 00:00 0          [stack]

Also I have problems with ioremapping:
//info->screen_base = ioremap(AU1000VLFB_BASE_PHYS,
AU1000VLFB_FB_LEN); // <-doesn't work when do 'dd if=/dev/zero of=/dev/fb0 count=2048'
info->screen_base = AU1000VLFB_BASE_PHYS; // <- this one clears screen
when 'dd if=/dev/zero of=/dev/fb0 count=2048'

ioremap gives me 0xc000_0000 from 0xbe00_0000. I'm not sure if this correct.

Can't resolve problem myself.
Thank you in advance!
Best Regards, Maxim Moroz.
---------------------------------------

/* fb_mmap
* Map video memory in user space. We don't use the generic fb_mmap method mainly
* to allow the use of the TLB streaming flag (CCA=6)
*/
int au1000fb_fb_mmap(struct fb_info *fbi, struct file *file, struct vm_area_struct *vma) {
unsigned int len;
unsigned long start=0, off;

if (vma->vm_pgoff > (~0UL >> PAGE_SHIFT)) {
return -EINVAL;
}

start = fbi->fix.smem_start & PAGE_MASK;
len = PAGE_ALIGN((start & ~PAGE_MASK) + fbi->fix.smem_len);

off = vma->vm_pgoff << PAGE_SHIFT;

if ((vma->vm_end - vma->vm_start + off) > len) {
return -EINVAL;
}

off += start;
vma->vm_pgoff = off >> PAGE_SHIFT;

vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); pgprot_val(vma->vm_page_prot) |= (6 << 9); //CCA=6

vma->vm_flags |= VM_IO;

printk("%lu %lu %lu %lu %lu\n",

         fbi->fix.smem_start,
         start,
         off,
         vma->vm_start,
         vma->vm_end - vma->vm_start
         );

if (remap_pfn_range(vma, vma->vm_start, off >> PAGE_SHIFT,

       vma->vm_end - vma->vm_start,
       vma->vm_page_prot)) {

return -EAGAIN;
}

return 0;
}

What happens if you just use /dev/mem and that address?

Bad choice. Neither the Au1000 nor the Au1500 have internal frame buffers, so in both cases these are going to be board specific devices. Since the Au1500 has a PCI bridge, I suspect that driver is designed to work with the 36-bit address from the Au1500 core. You are going to have to write a custom mmap() function that does the proper mapping for your board design. Also, the Au1000 had some challenging bus interface issues to things like graphics controllers and you have to choose the proper memory controller configuration and hardware design to even have a chance at this working.

Good Luck, you may have lots of work ahead of you very specific to your board design :-)

	-- Dan

Dan Malek пишет:

The choice is not bad. au1000 also has 36 bit addresses.
I was toooooo bad using address 0xbe00_0000.
changed to 0x1e00_0000 and got working video.
Thanks all who made such perfect port to au1x00 processors! Really pleased to work with them... Heh.. Tried blackfin some time ago. that was horrible :-D

He-he, board is almost working already.

Best Regards,
Maxim Moroz.

The Au1500 doesn't have an internal video controller so I'm not sure which fb driver you based your work on.

Ah, so you just had the physical address setup incorrectly.

Enjoy :)

Pete