On Wed, Apr 24, 2013 at 04:20:50PM +0800, Jianguo Wu wrote:
...
> Hi Naoya,
>
> I think the -EINVAL is returned from hugetlb_get_unmapped_area(),
> for the two testcases:
> 1) $ ./mmappu $((5 * 2 * 1024 * 1024 - 4096)) //len1 = 0x9ff000
> 2) $ ./mmappu $((5 * 2 * 1024 * 1024 - 4095)) //len2 = 0x9ff001
>
> In do_mmap_pgoff(), after "len = PAGE_ALIGN(len);", len1 = 0x9ff000,
> len2 = 0xa00000, so len2 will pass "if (len & ~huge_page_mask(h))" check in
> hugetlb_get_unmapped_area(), and len1 will return -EINVAL. As follow:
>
> do_mmap_pgoff()
> {
> ...
> /* Careful about overflows.. */
> len = PAGE_ALIGN(len);
> ...
> get_unmapped_area()
> -->hugetlb_get_unmapped_area()
> {
> ...
> if (len & ~huge_page_mask(h))
> return -EINVAL;
> ...
> }
> }
You are right, Jianguo. Thanks you.
I totally missed the point.
>
> do we need to align len to hugepage size if it's hugetlbfs mmap? something like below:
>
> ---
> mm/mmap.c | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/mm/mmap.c b/mm/mmap.c
> index 0db0de1..bd42be24 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -1188,7 +1188,10 @@ unsigned long do_mmap_pgoff(struct file *file, unsigned long addr,
> addr = round_hint_to_min(addr);
>
> /* Careful about overflows.. */
> - len = PAGE_ALIGN(len);
> + if (file && is_file_hugepages(file))
> + len = ALIGN(len, huge_page_size(hstate_file(file)));
> + else
> + len = PAGE_ALIGN(len);
> if (!len)
> return -ENOMEM;
>
> --
I like putting this alignment code in if (flags & MAP_HUGETLB) branch in
SYSCALL_DEFINE6(mmap_pgoff) as Johannes pointed out in another subthread,
because it adds no impact on mmap calls with !MAP_HUGETLB.
Thanks,
Naoya Horiguchi
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>