On Tue, Jul 16, 2013 at 11:17:23AM +0530, Aneesh Kumar K.V wrote:
> Joonsoo Kim <iamjoonsoo.kim@xxxxxxx> writes:
>
> > On Mon, Jul 15, 2013 at 08:41:12PM +0530, Aneesh Kumar K.V wrote:
> >> Joonsoo Kim <iamjoonsoo.kim@xxxxxxx> writes:
> >>
> >> > If we map the region with MAP_NORESERVE and MAP_SHARED,
> >> > we can skip to check reserve counting and eventually we cannot be ensured
> >> > to allocate a huge page in fault time.
> >> > With following example code, you can easily find this situation.
> >> >
> >> > Assume 2MB, nr_hugepages = 100
> >> >
> >> > fd = hugetlbfs_unlinked_fd();
> >> > if (fd < 0)
> >> > return 1;
> >> >
> >> > size = 200 * MB;
> >> > flag = MAP_SHARED;
> >> > p = mmap(NULL, size, PROT_READ|PROT_WRITE, flag, fd, 0);
> >> > if (p == MAP_FAILED) {
> >> > fprintf(stderr, "mmap() failed: %s\n", strerror(errno));
> >> > return -1;
> >> > }
> >> >
> >> > size = 2 * MB;
> >> > flag = MAP_ANONYMOUS | MAP_SHARED | MAP_HUGETLB | MAP_NORESERVE;
> >> > p = mmap(NULL, size, PROT_READ|PROT_WRITE, flag, -1, 0);
> >> > if (p == MAP_FAILED) {
> >> > fprintf(stderr, "mmap() failed: %s\n", strerror(errno));
> >> > }
> >> > p[0] = '0';
> >> > sleep(10);
> >> >
> >> > During executing sleep(10), run 'cat /proc/meminfo' on another process.
> >> > You'll find a mentioned problem.
> >> >
> >> > Solution is simple. We should check VM_NORESERVE in vma_has_reserves().
> >> > This prevent to use a pre-allocated huge page if free count is under
> >> > the reserve count.
> >>
> >> You have a problem with this patch, which i guess you are fixing in
> >> patch 9. Consider two process
> >>
> >> a) MAP_SHARED on fd
> >> b) MAP_SHARED | MAP_NORESERVE on fd
> >>
> >> We should allow the (b) to access the page even if VM_NORESERVE is set
> >> and we are out of reserve space .
> >
> > I can't get your point.
> > Please elaborate more on this.
>
>
> One process mmap with MAP_SHARED and another one with MAP_SHARED | MAP_NORESERVE
> Now the first process will result in reserving the pages from the hugtlb
> pool. Now if the second process try to dequeue huge page and we don't
> have free space we will fail because
>
> vma_has_reservers will now return zero because VM_NORESERVE is set
> and we can have (h->free_huge_pages - h->resv_huge_pages) == 0;
I think that this behavior is correct, because a user who mapped with
VM_NORESERVE should not think their allocation always succeed. With patch 9,
he can be ensured to succeed, but I think it is side-effect.
> The below hunk in your patch 9 handles that
>
> + if (vma->vm_flags & VM_NORESERVE) {
> + /*
> + * This address is already reserved by other process(chg == 0),
> + * so, we should decreament reserved count. Without
> + * decreamenting, reserve count is remained after releasing
> + * inode, because this allocated page will go into page cache
> + * and is regarded as coming from reserved pool in releasing
> + * step. Currently, we don't have any other solution to deal
> + * with this situation properly, so add work-around here.
> + */
> + if (vma->vm_flags & VM_MAYSHARE && chg == 0)
> + return 1;
> + else
> + return 0;
> + }
>
> so may be both of these should be folded ?
I think that these patches should not be folded, because these handle
two separate issues. Reserve count mismatch issue mentioned in patch 9
is not introduced by patch 7.
Thanks.
>
> -aneesh
>
> --
> 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>
--
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>