FOLL_NOWAIT and get_user_pages_unlocked

Hello,

KVM is hanging on postcopy live migration.

David tracked it down to commit ce53053ce378c21e7ffc45241fd67d6ee79daa2b and the problem is pretty obvious then.

Either we teach get_user_pages_locked/unlocked to handle FOLL_NOWAIT (so faultin_nopage works right even when the nonblocking pointer is not NULL) or we need to revert part of commit ce53053ce378c21e7ffc45241fd67d6ee79daa2b and keep using FOLL_NOWAIT only as parameter to get_user_pages (which won't ever set nonblocking pointer to non-NULL). I suppose the former approach is preferred to be more robust.

Thanks,
Andrea

Andrea Arcangeli (1):
mm: gup: teach get_user_pages_unlocked to handle FOLL_NOWAIT

 mm/gup.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

KVM is hanging during postcopy live migration with userfaultfd because get_user_pages_unlocked is not capable to handle FOLL_NOWAIT.

Earlier FOLL_NOWAIT was only ever passed to get_user_pages.

Specifically faultin_page (the callee of get_user_pages_unlocked caller) doesn't know that if FAULT_FLAG_RETRY_NOWAIT was set in the page fault flags, when VM_FAULT_RETRY is returned, the mmap_sem wasn't actually released (even if nonblocking is not NULL). So it sets *nonblocking to zero and the caller won't release the mmap_sem thinking it was already released, but it wasn't because of FOLL_NOWAIT.

---
 mm/gup.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)

index 1b46e6e74881..6afae32571ca 100644
@@ -516,7 +516,7 @@ static int faultin_page(struct task_struct *tsk, struct vm_area_struct *vma,
 	}
 
 	if (ret & VM_FAULT_RETRY) {
-		if (nonblocking)
+		if (nonblocking && !(fault_flags & FAULT_FLAG_RETRY_NOWAIT))
 			*nonblocking = 0;
 		return -EBUSY;
 	}

@@ -890,7 +890,10 @@ static __always_inline long __get_user_pages_locked(struct task_struct *tsk,
 				break;
 		}
 		if (*locked) {
-			/* VM_FAULT_RETRY didn't trigger */
+			/*
+			 * VM_FAULT_RETRY didn't trigger or it was a
+			 * FOLL_NOWAIT.
+			 */
 			if (!pages_done)
 				pages_done = ret;
 			break;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

I added

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>