On Thu, 2008-05-08 at 18:19 +0200, Hans Rosenfeld wrote:
quoted text
> On Thu, May 08, 2008 at 08:51:22AM -0700, Dave Hansen wrote:
> > Is there anything in your dmesg?
> 
> mm/memory.c:127: bad pmd ffff810076801040(80000000720000e7).
> 
> > There was a discussion on LKML in the last couple of days about
> > pmd_bad() triggering on huge pages.  Perhaps we're clearing the mapping
> > with the pmd_none_or_clear_bad(), and *THAT* is leaking the page.
> 
> That makes sense. I remember that explicitly munmapping the huge page
> would still work, but it doesn't. I don't quite remember what I did back
> then to test this, but I probably made some mistake there that led me to
> some false conclusions.

I can't see how it would possibly work with the code that we have today,
so I guess it was just a false assumption.

static inline int pmd_none_or_clear_bad(pmd_t *pmd)
{
        if (pmd_none(*pmd))
                return 1;
        if (unlikely(pmd_bad(*pmd))) {
                pmd_clear_bad(pmd);
                return 1;
        }
        return 0;
}

void pmd_clear_bad(pmd_t *pmd)
{
        pmd_ERROR(*pmd);
        pmd_clear(pmd);
}

That pmd_clear() will simply zero out the pmd and leak the page.

Sounds like Linus had the right idea:

quoted text
> I'd much rather have pdm_bad() etc fixed up instead, so that they do a 
> more proper test (not thinking that a PSE page is bad, since it clearly 
> isn't). And then, make them dependent on DEBUG_VM, because doing the 
> proper test will be more expensive.

-- Dave

--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/