Re: [PATCH 10/22] HWPOISON: check and isolate corrupted free pages v2

[Wu Fengguang <fengguang.wu@intel.com>]

On Tue, Jun 16, 2009 at 07:29:45PM +0800, Hugh Dickins wrote:
> On Tue, 16 Jun 2009, Wu Fengguang wrote:
> > 
> > Right.  Then the original __ClearPageBuddy() call in bad_page() is
> > questionable, I guess this line was there just for the sake of safety
> > (ie. the buddy allocator itself goes wrong):
> > 
> > sound-2.6/mm/page_alloc.c
> > 
> >         @@ -269,7 +269,6 @@ static void bad_page(struct page *page)
> >                 dump_stack();
> >          out:
> >                 /* Leave bad fields for debug, except PageBuddy could make trouble */
> > ===>            __ClearPageBuddy(page);
> >                 add_taint(TAINT_BAD_PAGE);
> >          }
> 
> I didn't put that in for the case of the buddy allocator going wrong
> (not sure if there could be such a case - I don't mean that the buddy
> allocator is provably perfect! but how would it get here if it were
> wrong?).  No, I put that in for the case when the flag bits in struct
> page have themselves got corrupted somehow, and hence we arrive at
> bad_page(): most of the bits are best left as they are, to provide
> maximum debug info; but leaving PageBuddy set there might conceivably
> allow this corrupted struct page to get paired up with its buddy later,
> and so freed for reuse, when we're trying to make sure it's never reused.

Hugh, thank you for the detailed explanations!  You are always informative :)

Thanks,
Fengguang