Re: 2.4.x oops with X

[Marcelo Tosatti <marcelo.tosatti@cyclades.com>]

On Fri, Jan 07, 2005 at 07:13:50PM +0100, Andreas Hartmann wrote:
> Hello Marcelo,
> 
> Marcelo Tosatti schrieb:
> > On Fri, Jan 07, 2005 at 10:03:11AM +0100, Andreas Hartmann wrote:
> >> Hello!

<snip>

> >> Jan  6 15:30:03 athlon kernel: kernel BUG at memory.c:535!
> >> Jan  6 15:30:03 athlon kernel: invalid operand: 0000
> >> Jan  6 15:30:03 athlon kernel: CPU:    0
> >> Jan  6 15:30:03 athlon kernel: EIP:    0010:[<c0137e93>]    Not tainted
> >> Using defaults from ksymoops -t elf32-i386 -a i386
> >> Jan  6 15:30:03 athlon kernel: EFLAGS: 00010282
> >> Jan  6 15:30:03 athlon kernel: eax: 00000045   ebx: 000a0000   ecx:
> >> d2226000   edx: db897f7c
> >> Jan  6 15:30:03 athlon kernel: esi: ffffffff   edi: c54106c0   ebp:
> >> 00000001   esp: d2227c04
> >> Jan  6 15:30:03 athlon kernel: ds: 0018   es: 0018   ss: 0018
> >> Jan  6 15:30:03 athlon kernel: Process X (pid: 21100, stackpage=d2227000)
> >> Jan  6 15:30:03 athlon kernel: Stack: c0246860 c54106c0 000800ff 00000000
> >> 00002cb0 00000010 00000000 c54106c0
> >> Jan  6 15:30:03 athlon kernel:        000a0000 000a03d4 d2226000 c0166b9c
> >> d2226000 db673b40 000a0000 00000001
> >> Jan  6 15:30:04 athlon kernel:        00000000 00000001 d2227c6c d2227c70
> >> 00002cb0 00000003 01388000 0070c000
> >> Jan  6 15:30:04 athlon kernel: Call Trace: [<c0166b9c>]  [<c019023d>]
> >> [<c018f587>]  [<c0148612>]  [<c0152d90>]  [<c0116546>]  [<c0106e84>]
> >> [<c01223a1>]  [<c0107ef0>]  [<c012281f>]  [<c0107ef0>]  [<c010705c>]
> >> Jan  6 15:30:04 athlon kernel: Code: 0f 0b 17 02 97 65 24 c0 be f2 ff ff
> >> ff eb b2 ff 41 14 eb 86
> >> 
> >> >>EIP; c0137e93 <get_user_pages+163/200>   <=====
> >> Trace; c0166b9c <elf_core_dump+7ec/975>
> >> Trace; c019023d <do_journal_end+bd/b60>
> >> Trace; c018f587 <journal_end+27/30>
> >> Trace; c0148612 <do_truncate+72/a0>
> >> Trace; c0152d90 <do_coredump+170/177>
> >> Trace; c0116546 <schedule+236/3e0>
> >> Trace; c0106e84 <do_signal+214/2c0>
> >> Trace; c01223a1 <deliver_signal+31/70>
> >> Trace; c0107ef0 <do_general_protection+0/a0>
> >> Trace; c012281f <force_sig+1f/30>
> >> Trace; c0107ef0 <do_general_protection+0/a0>
> >> Trace; c010705c <signal_return+14/18>
> >> Code;  c0137e93 <get_user_pages+163/200>
> >> 00000000 <_EIP>:
> >> Code;  c0137e93 <get_user_pages+163/200>   <=====
> >>    0:   0f 0b                     ud2a      <=====
> >> Code;  c0137e95 <get_user_pages+165/200>
> >>    2:   17                        pop    %ss
> >> Code;  c0137e96 <get_user_pages+166/200>
> >>    3:   02 97 65 24 c0 be         add    0xbec02465(%edi),%dl
> >> Code;  c0137e9c <get_user_pages+16c/200>
> >>    9:   f2 ff                     repnz (bad)
> >> Code;  c0137e9e <get_user_pages+16e/200>
> >>    b:   ff                        (bad)
> >> Code;  c0137e9f <get_user_pages+16f/200>
> >>    c:   ff eb                     ljmp   *<internal disassembler error>
> >> Code;  c0137ea1 <get_user_pages+171/200>
> >>    e:   b2 ff                     mov    $0xff,%dl
> >> Code;  c0137ea3 <get_user_pages+173/200>
> >>   10:   41                        inc    %ecx
> >> Code;  c0137ea4 <get_user_pages+174/200>
> >>   11:   14 eb                     adc    $0xeb,%al
> >> Code;  c0137ea6 <get_user_pages+176/200>
> >>   13:   86 00                     xchg   %al,(%eax)
> > 
> > We added a BUG() call in get_user_pages() to catch VM_IO flagged vma's 
> > (virtual memory areas) with PageReserved pages.
> > 
> > Can you disable AGP and run X ? 
> 
> 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.

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;