Re: [Linux-fbdev-devel] request_mem_region problem ...

[Antonino D. <ad...@po...>]

On Thu, 2002-10-31 at 16:10, Sven Luther wrote:
> On Thu, Oct 31, 2002 at 08:34:05AM +0800, Antonino Daplas wrote:
[...]
> 
> I already solutioned this problem.
> 
> BTW, i don't understand how the tdfxfb does not have this problem, as it
> is from there that i tried to sole this.
> 
> The problem was that i did :
> 
>         /* Find the mmio register area */
> 	pm3_fix.mmio_start = pci_resource_start(pdev, 0);
> 	pm3_fix.mmio_len = pci_resource_len(pdev, 0);
> 
> And later :
> 
> 	if (!request_mem_region(pci_resource_start(pdev, 0), pci_resource_len(pdev, 0), ...
> 
> Changing this to :
> 
> 	if (!request_mem_region(pm3_fix.mmio_start, pm3_fix.mmio_len, ...
> 
> Solved the problem, so it seems that the pci_resource_ functions can be
> called only one time. Is this true ?
> 
Strange.  pci_resource_* are just macros, so they should be usable as
long as the pci_dev pointer is valid.

> Now, i seem to have everything fine, but it still hangs in
> register_framebuffer, a bit of searching has it that it is when
> register_framebuffer calls take_over_console. I didn't have time for
> looking into it more yesterday.
> 
> Still the screen is black, and i see a small (underscore) blinking
> cursor at the bottom left of the screen. The machine was dead though :
> 
> Oct 30 23:15:53 iliana kernel:  printing eip:
> Oct 30 23:15:53 iliana kernel: c023affc
> Oct 30 23:15:53 iliana kernel: Oops: 0000
> Oct 30 23:15:53 iliana kernel: pm3fb nls_iso8859-15 nls_cp437 sym53c8xx
> Oct 30 23:15:53 iliana kernel: CPU:    0
> Oct 30 23:15:53 iliana kernel: EIP:    0060:[fbcon_cursor+108/480] Not tainted
> Oct 30 23:15:53 iliana kernel: EFLAGS: 00010246
> Oct 30 23:15:53 iliana kernel: eax: 00000000   ebx: 00000000   ecx: 00000000   edx: c1143240
> Oct 30 23:15:53 iliana kernel: esi: c042d280   edi: 00000002   ebp: 00000018   esp: c4dcbc10
> Oct 30 23:15:53 iliana kernel: ds: 0068   es: 0068   ss: 0068
> Oct 30 23:15:53 iliana kernel: Process modprobe (pid: 495, threadinfo=c4dca000 task=c6bec040)
> Oct 30 23:15:53 iliana kernel: Stack: 00000000 c04156c0 00000000 00000000 c01f21f8 c1143240 00000002 00000000
> Oct 30 23:15:53 iliana kernel:        c04156c0 c01f5435 00000000 c039ef20 c03dfe9f 0000339f 000033cf c0414a30
> Oct 30 23:15:53 iliana kernel:        00002adf c01275e1 00000000 c0119439 c039ef20 c03dfe9f 00000030 000033cf
> 
Running ksymoops on that might help.  The source can be found at
ftp.kernel.org.  

If you define FBCON_HAS_ACCEL, then fb_imageblit, fb_copyarea and
fb_fillrect in info->fbops must not be null.  Otherwise, you'll always
get an oops. If you don't define FBCON_HAS_ACCEL, you'll end up with
fbcon_dummy, which is totally safe but will not show anything at all. 
This is all assuming you use gen_set_var.  

If you did not define fb_set_var at all, then info->disp.dispsw must
point to valid console drawing hooks before you register the
framebuffer.  (Actually, all fields in struct display must be valid
before you register the framebuffer)

Your oops tracing points to fbcon_cursor as the culprit.  If
dispsw->cursor is NULL, then you must have a valid dispsw->revc.
  
> How, well, maybe i should wait for the new patches from James to be
> accepted into the linus tree.
> 
> Notice, maybe this is due to me not having defined yet any of the
> check_var, set_par, blank, ... functions, but my idea, for the howto
> afterward, was to write first the most minimal working fbdev.
> 
> BTW, should i work on james BK tree directly ? Is there an howto on how
> to do that ? Or could it be possible to get a patch against 2.5.44 ?
> 
Try Documentation/BK-usage.  You will need a very fat pipe, then for a
start, 

bk clone bk://fbdev.bkbits.net/fbdev-2.5

> 
> when i try to use the cfb_xxx functions, it does not work. i copied the
> neofb line in the Makefile so that pm3fb should have added the 3 cfbxxx
> objects, but it does not work. When loading the cfbxxx.o object files as
> modules, insmod complains about lack of version in them, so i guess they
> are not modules. Still, they don't get linked with the pm3fb.o (and
> neither with the neofb.o when i tried it), and when loading pm3fb.o, i
> get a message about missing symbols cfb_xxx.
> 
> So, does the cfb_xxx generic accel stuff get linked correctly when
> building as modules or am i missing something ?
> 
You're correct, the neofb line in the Makefile is for static building
only.  The cfb_xxx will not compile as modules, yet.  If your driver has
its own Makefile, just add cfb_xxx.o objects to directly link to them.

Tony