[PATCH v7 5/6] mm: handle swap page faults under per-VMA lock

[Suren Baghdasaryan <surenb@google.com>]

When page fault is handled under per-VMA lock protection, all swap page
faults are retried with mmap_lock because folio_lock_or_retry has to drop
and reacquire mmap_lock if folio could not be immediately locked.
Follow the same pattern as mmap_lock to drop per-VMA lock when waiting
for folio and retrying once folio is available.
With this obstacle removed, enable do_swap_page to operate under
per-VMA lock protection. Drivers implementing ops->migrate_to_ram might
still rely on mmap_lock, therefore we have to fall back to mmap_lock in
that particular case.
Note that the only time do_swap_page calls synchronous swap_readpage
is when SWP_SYNCHRONOUS_IO is set, which is only set for
QUEUE_FLAG_SYNCHRONOUS devices: brd, zram and nvdimms (both btt and
pmem). Therefore we don't sleep in this path, and there's no need to
drop the mmap or per-VMA lock.

Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Tested-by: Alistair Popple <apopple@nvidia.com>
Reviewed-by: Alistair Popple <apopple@nvidia.com>
Acked-by: Peter Xu <peterx@redhat.com>
---
 include/linux/mm.h | 13 +++++++++++++
 mm/filemap.c       | 17 ++++++++---------
 mm/memory.c        | 16 ++++++++++------
 3 files changed, 31 insertions(+), 15 deletions(-)

diff --git a/include/linux/mm.h b/include/linux/mm.h
index 39aa409e84d5..54ab11214f4f 100644
--- a/include/linux/mm.h
+++ b/include/linux/mm.h
@@ -720,6 +720,14 @@ static inline void vma_mark_detached(struct vm_area_struct *vma, bool detached)
 	vma->detached = detached;
 }
 
+static inline void release_fault_lock(struct vm_fault *vmf)
+{
+	if (vmf->flags & FAULT_FLAG_VMA_LOCK)
+		vma_end_read(vmf->vma);
+	else
+		mmap_read_unlock(vmf->vma->vm_mm);
+}
+
 struct vm_area_struct *lock_vma_under_rcu(struct mm_struct *mm,
 					  unsigned long address);
 
@@ -735,6 +743,11 @@ static inline void vma_assert_write_locked(struct vm_area_struct *vma) {}
 static inline void vma_mark_detached(struct vm_area_struct *vma,
 				     bool detached) {}
 
+static inline void release_fault_lock(struct vm_fault *vmf)
+{
+	mmap_read_unlock(vmf->vma->vm_mm);
+}
+
 #endif /* CONFIG_PER_VMA_LOCK */
 
 /*
diff --git a/mm/filemap.c b/mm/filemap.c
index 5da5ad6f7f4c..5ac1b7beea2a 100644
--- a/mm/filemap.c
+++ b/mm/filemap.c
@@ -1671,27 +1671,26 @@ static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
  * Return values:
  * 0 - folio is locked.
  * non-zero - folio is not locked.
- *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
- *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
- *     which case mmap_lock is still held.
+ *     mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
+ *     vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
+ *     FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
  *
  * If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
- * with the folio locked and the mmap_lock unperturbed.
+ * with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
  */
 vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
 {
-	struct mm_struct *mm = vmf->vma->vm_mm;
 	unsigned int flags = vmf->flags;
 
 	if (fault_flag_allow_retry_first(flags)) {
 		/*
-		 * CAUTION! In this case, mmap_lock is not released
-		 * even though return VM_FAULT_RETRY.
+		 * CAUTION! In this case, mmap_lock/per-VMA lock is not
+		 * released even though returning VM_FAULT_RETRY.
 		 */
 		if (flags & FAULT_FLAG_RETRY_NOWAIT)
 			return VM_FAULT_RETRY;
 
-		mmap_read_unlock(mm);
+		release_fault_lock(vmf);
 		if (flags & FAULT_FLAG_KILLABLE)
 			folio_wait_locked_killable(folio);
 		else
@@ -1703,7 +1702,7 @@ vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
 
 		ret = __folio_lock_killable(folio);
 		if (ret) {
-			mmap_read_unlock(mm);
+			release_fault_lock(vmf);
 			return VM_FAULT_RETRY;
 		}
 	} else {
diff --git a/mm/memory.c b/mm/memory.c
index 4ae3f046f593..bb0f68a73b0c 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -3729,12 +3729,6 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 	if (!pte_unmap_same(vmf))
 		goto out;
 
-	if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
-		ret = VM_FAULT_RETRY;
-		vma_end_read(vma);
-		goto out;
-	}
-
 	entry = pte_to_swp_entry(vmf->orig_pte);
 	if (unlikely(non_swap_entry(entry))) {
 		if (is_migration_entry(entry)) {
@@ -3744,6 +3738,16 @@ vm_fault_t do_swap_page(struct vm_fault *vmf)
 			vmf->page = pfn_swap_entry_to_page(entry);
 			ret = remove_device_exclusive_entry(vmf);
 		} else if (is_device_private_entry(entry)) {
+			if (vmf->flags & FAULT_FLAG_VMA_LOCK) {
+				/*
+				 * migrate_to_ram is not yet ready to operate
+				 * under VMA lock.
+				 */
+				vma_end_read(vma);
+				ret = VM_FAULT_RETRY;
+				goto out;
+			}
+
 			vmf->page = pfn_swap_entry_to_page(entry);
 			vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd,
 					vmf->address, &vmf->ptl);
-- 
2.41.0.255.g8b1d071c50-goog