2019-01-09 02:03:46

by Andrea Arcangeli

Subject: [PATCH 0/1] mm/hugetlb.c: teach follow_hugetlb_page() to handle FOLL_NOWAIT

Hello,

this fixes a regression that resurfaced in hugetlbfs code after we
fixed it for the core VM a few months ago. This is only reproducible
doing postcopy live migration of KVM (the only user of FOLL_NOWAIT) if
backed by hugetlbfs memory. It's unrelated to userfaultfd, but
userfaultfd reproduces it easily because it's an heavy user of
VM_FAULT_RETRY retvals.

Thanks,
Andrea

Andrea Arcangeli (1):
mm/hugetlb.c: teach follow_hugetlb_page() to handle FOLL_NOWAIT

mm/hugetlb.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

2019-01-09 02:03:46

by Andrea Arcangeli

Subject: [PATCH 1/1] mm/hugetlb.c: teach follow_hugetlb_page() to handle FOLL_NOWAIT

hugetlb needs the same fix as faultin_nopage (which was applied in
96312e61282ae3f6537a562625706498cbc75594) or KVM hangs because it
thinks the mmap_sem was already released by hugetlb_fault() if it
returned VM_FAULT_RETRY, but it wasn't in the FOLL_NOWAIT case.

Fixes: ce53053ce378 ("kvm: switch get_user_page_nowait() to get_user_pages_unlocked()")
Signed-off-by: Andrea Arcangeli <[email protected]>
Tested-by: "Dr. David Alan Gilbert" <[email protected]>
Reported-by: "Dr. David Alan Gilbert" <[email protected]>
---
mm/hugetlb.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index e37efd5d8318..b3622d7888c8 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -4301,7 +4301,8 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
break;
}
if (ret & VM_FAULT_RETRY) {
- if (nonblocking)
+ if (nonblocking &&
+ !(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
*nonblocking = 0;
*nr_pages = 0;
/*

2019-01-09 02:57:31

by Mike Kravetz

Subject: Re: [PATCH 1/1] mm/hugetlb.c: teach follow_hugetlb_page() to handle FOLL_NOWAIT

On 1/8/19 6:02 PM, Andrea Arcangeli wrote:
> hugetlb needs the same fix as faultin_nopage (which was applied in
> 96312e61282ae3f6537a562625706498cbc75594) or KVM hangs because it
> thinks the mmap_sem was already released by hugetlb_fault() if it
> returned VM_FAULT_RETRY, but it wasn't in the FOLL_NOWAIT case.
>
> Fixes: ce53053ce378 ("kvm: switch get_user_page_nowait() to get_user_pages_unlocked()")
> Signed-off-by: Andrea Arcangeli <[email protected]>
> Tested-by: "Dr. David Alan Gilbert" <[email protected]>
> Reported-by: "Dr. David Alan Gilbert" <[email protected]>

Thanks for fixing this.

Reviewed-by: Mike Kravetz <[email protected]>
--
Mike Kravetz

> ---
> mm/hugetlb.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index e37efd5d8318..b3622d7888c8 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -4301,7 +4301,8 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> break;
> }
> if (ret & VM_FAULT_RETRY) {
> - if (nonblocking)
> + if (nonblocking &&
> + !(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
> *nonblocking = 0;
> *nr_pages = 0;
> /*
>

2019-01-09 05:06:46

by Peter Xu

Subject: Re: [PATCH 1/1] mm/hugetlb.c: teach follow_hugetlb_page() to handle FOLL_NOWAIT

On Tue, Jan 08, 2019 at 09:02:03PM -0500, Andrea Arcangeli wrote:
> hugetlb needs the same fix as faultin_nopage (which was applied in
> 96312e61282ae3f6537a562625706498cbc75594) or KVM hangs because it
> thinks the mmap_sem was already released by hugetlb_fault() if it
> returned VM_FAULT_RETRY, but it wasn't in the FOLL_NOWAIT case.
>
> Fixes: ce53053ce378 ("kvm: switch get_user_page_nowait() to get_user_pages_unlocked()")
> Signed-off-by: Andrea Arcangeli <[email protected]>
> Tested-by: "Dr. David Alan Gilbert" <[email protected]>
> Reported-by: "Dr. David Alan Gilbert" <[email protected]>

FWIW:

Reviewed-by: Peter Xu <[email protected]>

> ---
> mm/hugetlb.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index e37efd5d8318..b3622d7888c8 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -4301,7 +4301,8 @@ long follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma,
> break;
> }
> if (ret & VM_FAULT_RETRY) {
> - if (nonblocking)
> + if (nonblocking &&
> + !(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
> *nonblocking = 0;
> *nr_pages = 0;
> /*

Regards,

--
Peter Xu