Re: 2.4.x oops with X

[Dave Jones <davej@redhat.com>]

On Fri, Jan 07, 2005 at 03:01:52PM -0200, Marcelo Tosatti wrote:

 > > Well, what do you mean with "disable AGP"? I can't disable it in the BIOS.
 > > I disabled DRI in the XF86Config-file. agpgart and r128 haven't been
 > > loaded (they are built as modules). The behaviour of the X-starting
 > > doesn't change and it's always the same: 
 > 
 > I meant not loading the agpgart/r128 modules, but it seems they are loaded 
 > on demand and X actually can't work without them.

If he disabled DRI in the X config file, they shouldn't be getting
loaded, so I'm curious why agp is in the picture..

 > Well the problem is the core dumping code (elf_core_dump function) is trying to write
 > your ATI card memory to disk, which is wrong. 
 > 
 > agp's mmap() method does not mark the memory region it creates as VM_IO to 
 > indicate its a device memory mapped region, and it should AFAICS.
 > 
 > The following corrects the situation and should stop the BUG() from happening, 
 > however the SIGSEGV which X is receiving seems to be a different thing.
 > 
 > Please try the following patch
 > 
 > Davej?
 > 
 > 
 > --- linux-2.4.28/drivers/char/agp/agpgart_fe.c.orig	2005-01-07 16:42:24.732957320 -0200
 > +++ linux-2.4.28/drivers/char/agp/agpgart_fe.c	2005-01-07 16:42:30.329106576 -0200
 > @@ -651,6 +651,7 @@
 >  			unlock_kernel();
 >  			return -EAGAIN;
 >  		}
 > +		vma->vm_flags |= VM_IO;
 >  		AGP_UNLOCK();
 >  		unlock_kernel();
 >  		return 0;
 > @@ -667,6 +668,7 @@
 >  			unlock_kernel();
 >  			return -EAGAIN;
 >  		}
 > +		vma->vm_flags |= VM_IO;
 >  		AGP_UNLOCK();
 >  		unlock_kernel();
 >  		return 0;

Looks ok on a first glance.

		Dave