Re: [BUG] Page table corruption from commit 9542ada803198e6eba29d3289abb39ea82047b92.

[Suresh Siddha <suresh.b.siddha@intel.com>]

On Thu, Jan 22, 2009 at 01:23:34PM -0800, Jeff Mahoney wrote:
> Ingo Molnar wrote:
> > Could you try the current -tip tree please?
> > 
> >   http://people.redhat.com/mingo/tip.git/README
> > 
> > There's a number of fixes in there that i think will fix your crash. In 
> > particular:
> > 
> >   9597134: x86: fix PTE corruption issue while mapping RAM using /dev/mem
> 
> Yep. That change fixed the corruption, but I still get the other
> warning. It originates with mainline commit
> 5899329b19100c0b82dc78e9b21ed8b920c9ffb3.

> WARNING: at arch/x86/mm/pat.c:637 reserve_pfn_range+0x5b/0x26d()

We are aware of this warning issue. We will soon have a fix for this.
For now, this is benign (essentially mmap of /dev/mem is doing
the reserve_memtype() twice, one through phys_mem_access_prot_allowed()
and reserve_pfn_range(). We are planning to remove this double
reserve, which should fix the warning).


> I get another warning at
> the bisect site, but it seems secondary. It doesn't occur in tip, though.
> 
> WARNING: at arch/x86/mm/pageattr.c:560 __change_page_attr_set_clr+0x10b/0xc0c() 
> CPA: called for zero pte. vaddr = ffff8800f7ff0000 cpa->vaddr = ffff8800f7ff0000

This one is fixed in tip by commit a1e46212a410793d575718818e81ddc442a65283
	x86: fix page attribute corruption with cpa()

thanks,
suresh