RE: [patch] futex-2.5.40-B5

Seth, Rohit (rohit.seth@intel.com)
Fri, 4 Oct 2002 17:53:07 -0700


Oleg, your solution is not generic for mm/memory.c If any code wants to get
the page_struct mapping the hugetlb page then that code should branch out to
arch specific hugetlb code to get the correct page_struct pointer. More on
the following lines in arch/..../mm/hugetlbpage.c file:

follow_lightweight_hugetlb_page(mm, start)
ptep = huge_pte_offset(mm, start);
pte = *ptep;
page = pte_page(pte);

page += ((start & ~HPAGE_MASK) >> PAGE_SHIFT);
return page;

> -----Original Message-----
> From: Oleg Nesterov [mailto:oleg@tv-sign.ru]
> Sent: Friday, October 04, 2002 5:00 PM
> To: linux-kernel@vger.kernel.org
> Cc: Ingo Molnar; Rohit Seth
> Subject: Re: [patch] futex-2.5.40-B5
>
>
> Hello.
>
> Ingo Molnar wrote:
> > the new lookup code first does a lightweight
> follow_page(), then if no
> > page is present we do the get_user_pages() thing.
>
> What if futex placed in VM_HUGETLB area?
> Then follow_page() return garbage.
>
> I beleive in i386 case it can be fixed something like this:
>
> --- mm/memory.c.orig Sat Oct 5 01:08:54 2002
> +++ mm/memory.c Sat Oct 5 03:31:28 2002
> @@ -480,6 +480,17 @@ follow_page(struct mm_struct *mm, unsign
> if (pmd_none(*pmd) || pmd_bad(*pmd))
> goto out;
>
> +#ifdef CONFIG_HUGETLB_PAGE
> + if (pmd_large(pmd)) {
> + ptep = (pte_t *) pmd;
> +
> + if (write && !pte_write(*ptep))
> + return NULL;
> +
> + return pte_page(*ptep) + ((address &
> ~HPAGE_MASK) >> PAGE_SHIFT);
> + }
> +#endif
> +
> ptep = pte_offset_map(pmd, address);
> if (!ptep)
> goto out;
>
>
> Then follow_hugetlb_page() hook can be killed.
>
> Oleg.
>
-
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/