Re: kernel bug using 2.6.23-rc9

[David Daney <ddaney@avtrex.com>]

Ralf Baechle wrote:
> On Thu, Oct 04, 2007 at 02:49:13PM +0200, Giuseppe Sacco wrote:
>
> > Hi, while testing the latest kernel on SGI O2 I got may kernel bugs like
> > this:
> >
> > Kernel bug detected[#9]:
> > Cpu 0
> > $ 0   : 0000000000000000 ffffffff9001fce0 0000000000000001 0000000000000f18
> > $ 4   : 980000000111b4b8 000000007f955f18 ffffffff80400000 0000000000003fff
> > $ 8   : 00000000000050f1 000000007f955f18 9800000006073d68 9800000006073d60
> > $12   : 0000000000000010 ffffffff80000008 ffffffff80091680 0000000000000000
> > $16   : 980000000111b4b8 980000000768ddd0 000000000000000e 000000007f955f18
> > $20   : 9800000000581908 9800000007898080 9800000006073d68 9800000006073d60
> > $24   : 0000000000478284 000000002ac30580                                  
> > $28   : 9800000006070000 9800000006073cd0 0000000000000000 ffffffff8001b390
> > Hi    : 000000000000f2d3
> > Lo    : 00000000000050f1
> > epc   : ffffffff8001c800 kmap_coherent+0x10/0x118     Tainted: G      D
> > ra    : ffffffff8001b390 __flush_anon_page+0x70/0x90
> Very interesting.  Can you describe me your setup or maybe even come up
> with a test case for this?
Perhaps: 'cat /proc/self/cmdline'

As we were hacking O_DIRECT support into 2.6.15, I found what seems like 
a very similar situation.
It seems that all users of get_user_pages() *except* 
/proc/*/[cmdline|environ] call get_user_pages() with addresses aligned 
on page boundaries.  I am not sure if that is the problem here, but it 
seems similar.
David Daney.