[PATCH V5 1/9] vhost: don't set uaddr for invalid address

[PATCH V5 1/9] vhost: don't set uaddr for invalid address

[Jason Wang]

We should not setup uaddr for the invalid address, otherwise we may
try to pin or prefetch mapping of wrong pages.

Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 0536f8526359..488380a581dc 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -2082,7 +2082,8 @@ static long vhost_vring_set_num_addr(struct vhost_dev *d,
 	}
 
 #if VHOST_ARCH_CAN_ACCEL_UACCESS
-	vhost_setup_vq_uaddr(vq);
+	if (r == 0)
+		vhost_setup_vq_uaddr(vq);
 
 	if (d->mm)
 		mmu_notifier_register(&d->mmu_notifier, d->mm);
-- 
2.18.1

[PATCH V5 2/9] vhost: validate MMU notifier registration

[Jason Wang]

The return value of mmu_notifier_register() is not checked in
vhost_vring_set_num_addr(). This will cause an out of sync between mm
and MMU notifier thus a double free. To solve this, introduce a
boolean flag to track whether MMU notifier is registered and only do
unregistering when it was true.

Reported-and-tested-by:
syzbot+e58112d71f77113ddb7b@syzkaller.appspotmail.com
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 19 +++++++++++++++----
 drivers/vhost/vhost.h |  1 +
 2 files changed, 16 insertions(+), 4 deletions(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 488380a581dc..17f6abea192e 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -629,6 +629,7 @@ void vhost_dev_init(struct vhost_dev *dev,
 	dev->iov_limit = iov_limit;
 	dev->weight = weight;
 	dev->byte_weight = byte_weight;
+	dev->has_notifier = false;
 	init_llist_head(&dev->work_list);
 	init_waitqueue_head(&dev->wait);
 	INIT_LIST_HEAD(&dev->read_list);
@@ -730,6 +731,7 @@ long vhost_dev_set_owner(struct vhost_dev *dev)
 	if (err)
 		goto err_mmu_notifier;
 #endif
+	dev->has_notifier = true;
 
 	return 0;
 
@@ -959,7 +961,11 @@ void vhost_dev_cleanup(struct vhost_dev *dev)
 	}
 	if (dev->mm) {
 #if VHOST_ARCH_CAN_ACCEL_UACCESS
-		mmu_notifier_unregister(&dev->mmu_notifier, dev->mm);
+		if (dev->has_notifier) {
+			mmu_notifier_unregister(&dev->mmu_notifier,
+						dev->mm);
+			dev->has_notifier = false;
+		}
 #endif
 		mmput(dev->mm);
 	}
@@ -2064,8 +2070,10 @@ static long vhost_vring_set_num_addr(struct vhost_dev *d,
 	/* Unregister MMU notifer to allow invalidation callback
 	 * can access vq->uaddrs[] without holding a lock.
 	 */
-	if (d->mm)
+	if (d->has_notifier) {
 		mmu_notifier_unregister(&d->mmu_notifier, d->mm);
+		d->has_notifier = false;
+	}
 
 	vhost_uninit_vq_maps(vq);
 #endif
@@ -2085,8 +2093,11 @@ static long vhost_vring_set_num_addr(struct vhost_dev *d,
 	if (r == 0)
 		vhost_setup_vq_uaddr(vq);
 
-	if (d->mm)
-		mmu_notifier_register(&d->mmu_notifier, d->mm);
+	if (d->mm) {
+		r = mmu_notifier_register(&d->mmu_notifier, d->mm);
+		if (!r)
+			d->has_notifier = true;
+	}
 #endif
 
 	mutex_unlock(&vq->mutex);
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index 42a8c2a13ab1..a9a2a93857d2 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -214,6 +214,7 @@ struct vhost_dev {
 	int iov_limit;
 	int weight;
 	int byte_weight;
+	bool has_notifier;
 };
 
 bool vhost_exceeds_weight(struct vhost_virtqueue *vq, int pkts, int total_len);
-- 
2.18.1

[PATCH V5 3/9] vhost: fix vhost map leak

[Jason Wang]

We don't free map during vhost_map_unprefetch(). This means it could
be leaked. Fixing by free the map.

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 4 +---
 1 file changed, 1 insertion(+), 3 deletions(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 17f6abea192e..2a3154976277 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -302,9 +302,7 @@ static void vhost_vq_meta_reset(struct vhost_dev *d)
 static void vhost_map_unprefetch(struct vhost_map *map)
 {
 	kfree(map->pages);
-	map->pages = NULL;
-	map->npages = 0;
-	map->addr = NULL;
+	kfree(map);
 }
 
 static void vhost_uninit_vq_maps(struct vhost_virtqueue *vq)
-- 
2.18.1

[PATCH V5 4/9] vhost: reset invalidate_count in vhost_set_vring_num_addr()

[Jason Wang]

The vhost_set_vring_num_addr() could be called in the middle of
invalidate_range_start() and invalidate_range_end(). If we don't reset
invalidate_count after the un-registering of MMU notifier, the
invalidate_cont will run out of sync (e.g never reach zero). This will
in fact disable the fast accessor path. Fixing by reset the count to
zero.

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 4 ++++
 1 file changed, 4 insertions(+)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 2a3154976277..2a7217c33668 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -2073,6 +2073,10 @@ static long vhost_vring_set_num_addr(struct vhost_dev *d,
 		d->has_notifier = false;
 	}
 
+	/* reset invalidate_count in case we are in the middle of
+	 * invalidate_start() and invalidate_end().
+	 */
+	vq->invalidate_count = 0;
 	vhost_uninit_vq_maps(vq);
 #endif
 
-- 
2.18.1

[PATCH V5 5/9] vhost: mark dirty pages during map uninit

[Jason Wang]

We don't mark dirty pages if the map was teared down outside MMU
notifier. This will lead untracked dirty pages. Fixing by marking
dirty pages during map uninit.

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 2a7217c33668..c12cdadb0855 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -305,6 +305,18 @@ static void vhost_map_unprefetch(struct vhost_map *map)
 	kfree(map);
 }
 
+static void vhost_set_map_dirty(struct vhost_virtqueue *vq,
+				struct vhost_map *map, int index)
+{
+	struct vhost_uaddr *uaddr = &vq->uaddrs[index];
+	int i;
+
+	if (uaddr->write) {
+		for (i = 0; i < map->npages; i++)
+			set_page_dirty(map->pages[i]);
+	}
+}
+
 static void vhost_uninit_vq_maps(struct vhost_virtqueue *vq)
 {
 	struct vhost_map *map[VHOST_NUM_ADDRS];
@@ -314,8 +326,10 @@ static void vhost_uninit_vq_maps(struct vhost_virtqueue *vq)
 	for (i = 0; i < VHOST_NUM_ADDRS; i++) {
 		map[i] = rcu_dereference_protected(vq->maps[i],
 				  lockdep_is_held(&vq->mmu_lock));
-		if (map[i])
+		if (map[i]) {
+			vhost_set_map_dirty(vq, map[i], i);
 			rcu_assign_pointer(vq->maps[i], NULL);
+		}
 	}
 	spin_unlock(&vq->mmu_lock);
 
@@ -353,7 +367,6 @@ static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
 {
 	struct vhost_uaddr *uaddr = &vq->uaddrs[index];
 	struct vhost_map *map;
-	int i;
 
 	if (!vhost_map_range_overlap(uaddr, start, end))
 		return;
@@ -364,10 +377,7 @@ static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
 	map = rcu_dereference_protected(vq->maps[index],
 					lockdep_is_held(&vq->mmu_lock));
 	if (map) {
-		if (uaddr->write) {
-			for (i = 0; i < map->npages; i++)
-				set_page_dirty(map->pages[i]);
-		}
+		vhost_set_map_dirty(vq, map, index);
 		rcu_assign_pointer(vq->maps[index], NULL);
 	}
 	spin_unlock(&vq->mmu_lock);
-- 
2.18.1

[PATCH V5 6/9] vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()

[Jason Wang]

There's no need for RCU synchronization in vhost_uninit_vq_maps()
since we've already serialized with readers (memory accessors). This
also avoid the possible userspace DOS through ioctl() because of the
possible high latency caused by synchronize_rcu().

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index c12cdadb0855..cfc11f9ed9c9 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -333,7 +333,9 @@ static void vhost_uninit_vq_maps(struct vhost_virtqueue *vq)
 	}
 	spin_unlock(&vq->mmu_lock);
 
-	synchronize_rcu();
+	/* No need for synchronize_rcu() or kfree_rcu() since we are
+	 * serialized with memory accessors (e.g vq mutex held).
+	 */
 
 	for (i = 0; i < VHOST_NUM_ADDRS; i++)
 		if (map[i])
-- 
2.18.1

[PATCH V5 7/9] vhost: do not use RCU to synchronize MMU notifier with worker

[Jason Wang]

We used to use RCU to synchronize MMU notifier with worker. This leads
calling synchronize_rcu() in invalidate_range_start(). But on a busy
system, there would be many factors that may slow down the
synchronize_rcu() which makes it unsuitable to be called in MMU
notifier. This path switch to use a simple spinlock to do the
synchronization.

Benchmark was done through testpmd + vhost_net + XDP_DROP on
tap. Compare to copy_{to|from}_user() path, on Sandy Bridge (without
SMAP support), 1.5% PPS improvement was measured; on Broadwell (with
SMAP and enabled), 14% PPS improvement was measured.

This means we are not as fast as what 7f466032dc9e did because the
spinlock overhead in the datapath. This needs to be addressed in the
future.

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 115 ++++++++++++++++++++++--------------------
 drivers/vhost/vhost.h |   5 +-
 2 files changed, 62 insertions(+), 58 deletions(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index cfc11f9ed9c9..29e8abe694f7 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -324,17 +324,16 @@ static void vhost_uninit_vq_maps(struct vhost_virtqueue *vq)
 
 	spin_lock(&vq->mmu_lock);
 	for (i = 0; i < VHOST_NUM_ADDRS; i++) {
-		map[i] = rcu_dereference_protected(vq->maps[i],
-				  lockdep_is_held(&vq->mmu_lock));
+		map[i] = vq->maps[i];
 		if (map[i]) {
 			vhost_set_map_dirty(vq, map[i], i);
-			rcu_assign_pointer(vq->maps[i], NULL);
+			vq->maps[i] = NULL;
 		}
 	}
 	spin_unlock(&vq->mmu_lock);
 
-	/* No need for synchronize_rcu() or kfree_rcu() since we are
-	 * serialized with memory accessors (e.g vq mutex held).
+	/* No need for synchronization since we are serialized with
+	 * memory accessors (e.g vq mutex held).
 	 */
 
 	for (i = 0; i < VHOST_NUM_ADDRS; i++)
@@ -362,6 +361,16 @@ static bool vhost_map_range_overlap(struct vhost_uaddr *uaddr,
 	return !(end < uaddr->uaddr || start > uaddr->uaddr - 1 + uaddr->size);
 }
 
+static void inline vhost_vq_access_map_begin(struct vhost_virtqueue *vq)
+{
+	spin_lock(&vq->mmu_lock);
+}
+
+static void inline vhost_vq_access_map_end(struct vhost_virtqueue *vq)
+{
+	spin_unlock(&vq->mmu_lock);
+}
+
 static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
 				      int index,
 				      unsigned long start,
@@ -376,16 +385,14 @@ static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
 	spin_lock(&vq->mmu_lock);
 	++vq->invalidate_count;
 
-	map = rcu_dereference_protected(vq->maps[index],
-					lockdep_is_held(&vq->mmu_lock));
+	map = vq->maps[index];
 	if (map) {
+		vq->maps[index] = NULL;
 		vhost_set_map_dirty(vq, map, index);
-		rcu_assign_pointer(vq->maps[index], NULL);
 	}
 	spin_unlock(&vq->mmu_lock);
 
 	if (map) {
-		synchronize_rcu();
 		vhost_map_unprefetch(map);
 	}
 }
@@ -457,7 +464,7 @@ static void vhost_init_maps(struct vhost_dev *dev)
 	for (i = 0; i < dev->nvqs; ++i) {
 		vq = dev->vqs[i];
 		for (j = 0; j < VHOST_NUM_ADDRS; j++)
-			RCU_INIT_POINTER(vq->maps[j], NULL);
+			vq->maps[j] = NULL;
 	}
 }
 #endif
@@ -921,7 +928,7 @@ static int vhost_map_prefetch(struct vhost_virtqueue *vq,
 	map->npages = npages;
 	map->pages = pages;
 
-	rcu_assign_pointer(vq->maps[index], map);
+	vq->maps[index] = map;
 	/* No need for a synchronize_rcu(). This function should be
 	 * called by dev->worker so we are serialized with all
 	 * readers.
@@ -1216,18 +1223,18 @@ static inline int vhost_put_avail_event(struct vhost_virtqueue *vq)
 	struct vring_used *used;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
+		map = vq->maps[VHOST_ADDR_USED];
 		if (likely(map)) {
 			used = map->addr;
 			*((__virtio16 *)&used->ring[vq->num]) =
 				cpu_to_vhost16(vq, vq->avail_idx);
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1245,18 +1252,18 @@ static inline int vhost_put_used(struct vhost_virtqueue *vq,
 	size_t size;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
+		map = vq->maps[VHOST_ADDR_USED];
 		if (likely(map)) {
 			used = map->addr;
 			size = count * sizeof(*head);
 			memcpy(used->ring + idx, head, size);
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1272,17 +1279,17 @@ static inline int vhost_put_used_flags(struct vhost_virtqueue *vq)
 	struct vring_used *used;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
+		map = vq->maps[VHOST_ADDR_USED];
 		if (likely(map)) {
 			used = map->addr;
 			used->flags = cpu_to_vhost16(vq, vq->used_flags);
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1298,17 +1305,17 @@ static inline int vhost_put_used_idx(struct vhost_virtqueue *vq)
 	struct vring_used *used;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
+		map = vq->maps[VHOST_ADDR_USED];
 		if (likely(map)) {
 			used = map->addr;
 			used->idx = cpu_to_vhost16(vq, vq->last_used_idx);
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1362,17 +1369,17 @@ static inline int vhost_get_avail_idx(struct vhost_virtqueue *vq,
 	struct vring_avail *avail;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
+		map = vq->maps[VHOST_ADDR_AVAIL];
 		if (likely(map)) {
 			avail = map->addr;
 			*idx = avail->idx;
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1387,17 +1394,17 @@ static inline int vhost_get_avail_head(struct vhost_virtqueue *vq,
 	struct vring_avail *avail;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
+		map = vq->maps[VHOST_ADDR_AVAIL];
 		if (likely(map)) {
 			avail = map->addr;
 			*head = avail->ring[idx & (vq->num - 1)];
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1413,17 +1420,17 @@ static inline int vhost_get_avail_flags(struct vhost_virtqueue *vq,
 	struct vring_avail *avail;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
+		map = vq->maps[VHOST_ADDR_AVAIL];
 		if (likely(map)) {
 			avail = map->addr;
 			*flags = avail->flags;
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1438,15 +1445,15 @@ static inline int vhost_get_used_event(struct vhost_virtqueue *vq,
 	struct vring_avail *avail;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
-		map = rcu_dereference(vq->maps[VHOST_ADDR_AVAIL]);
+		vhost_vq_access_map_begin(vq);
+		map = vq->maps[VHOST_ADDR_AVAIL];
 		if (likely(map)) {
 			avail = map->addr;
 			*event = (__virtio16)avail->ring[vq->num];
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1461,17 +1468,17 @@ static inline int vhost_get_used_idx(struct vhost_virtqueue *vq,
 	struct vring_used *used;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_USED]);
+		map = vq->maps[VHOST_ADDR_USED];
 		if (likely(map)) {
 			used = map->addr;
 			*idx = used->idx;
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1486,17 +1493,17 @@ static inline int vhost_get_desc(struct vhost_virtqueue *vq,
 	struct vring_desc *d;
 
 	if (!vq->iotlb) {
-		rcu_read_lock();
+		vhost_vq_access_map_begin(vq);
 
-		map = rcu_dereference(vq->maps[VHOST_ADDR_DESC]);
+		map = vq->maps[VHOST_ADDR_DESC];
 		if (likely(map)) {
 			d = map->addr;
 			*desc = *(d + idx);
-			rcu_read_unlock();
+			vhost_vq_access_map_end(vq);
 			return 0;
 		}
 
-		rcu_read_unlock();
+		vhost_vq_access_map_end(vq);
 	}
 #endif
 
@@ -1843,13 +1850,11 @@ static bool iotlb_access_ok(struct vhost_virtqueue *vq,
 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 static void vhost_vq_map_prefetch(struct vhost_virtqueue *vq)
 {
-	struct vhost_map __rcu *map;
+	struct vhost_map *map;
 	int i;
 
 	for (i = 0; i < VHOST_NUM_ADDRS; i++) {
-		rcu_read_lock();
-		map = rcu_dereference(vq->maps[i]);
-		rcu_read_unlock();
+		map = vq->maps[i];
 		if (unlikely(!map))
 			vhost_map_prefetch(vq, i);
 	}
diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h
index a9a2a93857d2..983d06e62f12 100644
--- a/drivers/vhost/vhost.h
+++ b/drivers/vhost/vhost.h
@@ -115,10 +115,9 @@ struct vhost_virtqueue {
 #if VHOST_ARCH_CAN_ACCEL_UACCESS
 	/* Read by memory accessors, modified by meta data
 	 * prefetching, MMU notifier and vring ioctl().
-	 * Synchonrized through mmu_lock (writers) and RCU (writers
-	 * and readers).
+	 * Synchonrized through mmu_lock.
 	 */
-	struct vhost_map __rcu *maps[VHOST_NUM_ADDRS];
+	struct vhost_map *maps[VHOST_NUM_ADDRS];
 	/* Read by MMU notifier, modified by vring ioctl(),
 	 * synchronized through MMU notifier
 	 * registering/unregistering.
-- 
2.18.1

[PATCH V5 8/9] vhost: correctly set dirty pages in MMU notifiers callback

[Jason Wang]

We need make sure there's no reference on the map before trying to
mark set dirty pages.

Reported-by: Michael S. Tsirkin <mst@redhat.com>
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 29e8abe694f7..d8863aaaf0f6 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -386,13 +386,12 @@ static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
 	++vq->invalidate_count;
 
 	map = vq->maps[index];
-	if (map) {
+	if (map)
 		vq->maps[index] = NULL;
-		vhost_set_map_dirty(vq, map, index);
-	}
 	spin_unlock(&vq->mmu_lock);
 
 	if (map) {
+		vhost_set_map_dirty(vq, map, index);
 		vhost_map_unprefetch(map);
 	}
 }
-- 
2.18.1

[PATCH V5 9/9] vhost: do not return -EAGAIN for non blocking invalidation too early

[Jason Wang]

Instead of returning -EAGAIN unconditionally, we'd better do that only
we're sure the range is overlapped with the metadata area.

Reported-by: Jason Gunthorpe <jgg@ziepe.ca>
Fixes: 7f466032dc9e ("vhost: access vq metadata through kernel virtual address")
Signed-off-by: Jason Wang <jasowang@redhat.com>
---
 drivers/vhost/vhost.c | 32 +++++++++++++++++++-------------
 1 file changed, 19 insertions(+), 13 deletions(-)

diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index d8863aaaf0f6..f98155f28f02 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -371,16 +371,19 @@ static void inline vhost_vq_access_map_end(struct vhost_virtqueue *vq)
 	spin_unlock(&vq->mmu_lock);
 }
 
-static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
-				      int index,
-				      unsigned long start,
-				      unsigned long end)
+static int vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
+				     int index,
+				     unsigned long start,
+				     unsigned long end,
+				     bool blockable)
 {
 	struct vhost_uaddr *uaddr = &vq->uaddrs[index];
 	struct vhost_map *map;
 
 	if (!vhost_map_range_overlap(uaddr, start, end))
-		return;
+		return 0;
+	else if (!blockable)
+		return -EAGAIN;
 
 	spin_lock(&vq->mmu_lock);
 	++vq->invalidate_count;
@@ -394,6 +397,8 @@ static void vhost_invalidate_vq_start(struct vhost_virtqueue *vq,
 		vhost_set_map_dirty(vq, map, index);
 		vhost_map_unprefetch(map);
 	}
+
+	return 0;
 }
 
 static void vhost_invalidate_vq_end(struct vhost_virtqueue *vq,
@@ -414,18 +419,19 @@ static int vhost_invalidate_range_start(struct mmu_notifier *mn,
 {
 	struct vhost_dev *dev = container_of(mn, struct vhost_dev,
 					     mmu_notifier);
-	int i, j;
-
-	if (!mmu_notifier_range_blockable(range))
-		return -EAGAIN;
+	bool blockable = mmu_notifier_range_blockable(range);
+	int i, j, ret;
 
 	for (i = 0; i < dev->nvqs; i++) {
 		struct vhost_virtqueue *vq = dev->vqs[i];
 
-		for (j = 0; j < VHOST_NUM_ADDRS; j++)
-			vhost_invalidate_vq_start(vq, j,
-						  range->start,
-						  range->end);
+		for (j = 0; j < VHOST_NUM_ADDRS; j++) {
+			ret = vhost_invalidate_vq_start(vq, j,
+							range->start,
+							range->end, blockable);
+			if (ret)
+				return ret;
+		}
 	}
 
 	return 0;
-- 
2.18.1

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Michael S. Tsirkin]

On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
> Hi all:
> 
> This series try to fix several issues introduced by meta data
> accelreation series. Please review.
> 
> Changes from V4:
> - switch to use spinlock synchronize MMU notifier with accessors
> 
> Changes from V3:
> - remove the unnecessary patch
> 
> Changes from V2:
> - use seqlck helper to synchronize MMU notifier with vhost worker
> 
> Changes from V1:
> - try not use RCU to syncrhonize MMU notifier with vhost worker
> - set dirty pages after no readers
> - return -EAGAIN only when we find the range is overlapped with
>   metadata
> 
> Jason Wang (9):
>   vhost: don't set uaddr for invalid address
>   vhost: validate MMU notifier registration
>   vhost: fix vhost map leak
>   vhost: reset invalidate_count in vhost_set_vring_num_addr()
>   vhost: mark dirty pages during map uninit
>   vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
>   vhost: do not use RCU to synchronize MMU notifier with worker
>   vhost: correctly set dirty pages in MMU notifiers callback
>   vhost: do not return -EAGAIN for non blocking invalidation too early
> 
>  drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
>  drivers/vhost/vhost.h |   6 +-
>  2 files changed, 122 insertions(+), 86 deletions(-)

This generally looks more solid.

But this amounts to a significant overhaul of the code.

At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
for this release, and then re-apply a corrected version
for the next one?


> -- 
> 2.18.1

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Wang]

On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
> On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
>> Hi all:
>>
>> This series try to fix several issues introduced by meta data
>> accelreation series. Please review.
>>
>> Changes from V4:
>> - switch to use spinlock synchronize MMU notifier with accessors
>>
>> Changes from V3:
>> - remove the unnecessary patch
>>
>> Changes from V2:
>> - use seqlck helper to synchronize MMU notifier with vhost worker
>>
>> Changes from V1:
>> - try not use RCU to syncrhonize MMU notifier with vhost worker
>> - set dirty pages after no readers
>> - return -EAGAIN only when we find the range is overlapped with
>>    metadata
>>
>> Jason Wang (9):
>>    vhost: don't set uaddr for invalid address
>>    vhost: validate MMU notifier registration
>>    vhost: fix vhost map leak
>>    vhost: reset invalidate_count in vhost_set_vring_num_addr()
>>    vhost: mark dirty pages during map uninit
>>    vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
>>    vhost: do not use RCU to synchronize MMU notifier with worker
>>    vhost: correctly set dirty pages in MMU notifiers callback
>>    vhost: do not return -EAGAIN for non blocking invalidation too early
>>
>>   drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
>>   drivers/vhost/vhost.h |   6 +-
>>   2 files changed, 122 insertions(+), 86 deletions(-)
> This generally looks more solid.
>
> But this amounts to a significant overhaul of the code.
>
> At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
> for this release, and then re-apply a corrected version
> for the next one?


If possible, consider we've actually disabled the feature. How about 
just queued those patches for next release?

Thanks


>
>> -- 
>> 2.18.1
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[David Miller]

From: Jason Wang <jasowang@redhat.com>
Date: Mon, 12 Aug 2019 10:44:51 +0800

> On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
>> At this point how about we revert
>> 7f466032dc9e5a61217f22ea34b2df932786bbfc
>> for this release, and then re-apply a corrected version
>> for the next one?
> 
> If possible, consider we've actually disabled the feature. How about
> just queued those patches for next release?

I'm tossing this series while you and Michael decide how to move forward.

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Michael S. Tsirkin]

On Mon, Aug 12, 2019 at 10:44:51AM +0800, Jason Wang wrote:
> 
> On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
> > On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
> > > Hi all:
> > > 
> > > This series try to fix several issues introduced by meta data
> > > accelreation series. Please review.
> > > 
> > > Changes from V4:
> > > - switch to use spinlock synchronize MMU notifier with accessors
> > > 
> > > Changes from V3:
> > > - remove the unnecessary patch
> > > 
> > > Changes from V2:
> > > - use seqlck helper to synchronize MMU notifier with vhost worker
> > > 
> > > Changes from V1:
> > > - try not use RCU to syncrhonize MMU notifier with vhost worker
> > > - set dirty pages after no readers
> > > - return -EAGAIN only when we find the range is overlapped with
> > >    metadata
> > > 
> > > Jason Wang (9):
> > >    vhost: don't set uaddr for invalid address
> > >    vhost: validate MMU notifier registration
> > >    vhost: fix vhost map leak
> > >    vhost: reset invalidate_count in vhost_set_vring_num_addr()
> > >    vhost: mark dirty pages during map uninit
> > >    vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
> > >    vhost: do not use RCU to synchronize MMU notifier with worker
> > >    vhost: correctly set dirty pages in MMU notifiers callback
> > >    vhost: do not return -EAGAIN for non blocking invalidation too early
> > > 
> > >   drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
> > >   drivers/vhost/vhost.h |   6 +-
> > >   2 files changed, 122 insertions(+), 86 deletions(-)
> > This generally looks more solid.
> > 
> > But this amounts to a significant overhaul of the code.
> > 
> > At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
> > for this release, and then re-apply a corrected version
> > for the next one?
> 
> 
> If possible, consider we've actually disabled the feature. How about just
> queued those patches for next release?
> 
> Thanks

Sorry if I was unclear. My idea is that
1. I revert the disabled code
2. You send a patch readding it with all the fixes squashed
3. Maybe optimizations on top right away?
4. We queue *that* for next and see what happens.

And the advantage over the patchy approach is that the current patches
are hard to review. E.g.  it's not reasonable to ask RCU guys to review
the whole of vhost for RCU usage but it's much more reasonable to ask
about a specific patch.


-- 
MST
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Gunthorpe]

On Mon, Aug 12, 2019 at 05:49:08AM -0400, Michael S. Tsirkin wrote:
> On Mon, Aug 12, 2019 at 10:44:51AM +0800, Jason Wang wrote:
> > 
> > On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
> > > On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
> > > > Hi all:
> > > > 
> > > > This series try to fix several issues introduced by meta data
> > > > accelreation series. Please review.
> > > > 
> > > > Changes from V4:
> > > > - switch to use spinlock synchronize MMU notifier with accessors
> > > > 
> > > > Changes from V3:
> > > > - remove the unnecessary patch
> > > > 
> > > > Changes from V2:
> > > > - use seqlck helper to synchronize MMU notifier with vhost worker
> > > > 
> > > > Changes from V1:
> > > > - try not use RCU to syncrhonize MMU notifier with vhost worker
> > > > - set dirty pages after no readers
> > > > - return -EAGAIN only when we find the range is overlapped with
> > > >    metadata
> > > > 
> > > > Jason Wang (9):
> > > >    vhost: don't set uaddr for invalid address
> > > >    vhost: validate MMU notifier registration
> > > >    vhost: fix vhost map leak
> > > >    vhost: reset invalidate_count in vhost_set_vring_num_addr()
> > > >    vhost: mark dirty pages during map uninit
> > > >    vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
> > > >    vhost: do not use RCU to synchronize MMU notifier with worker
> > > >    vhost: correctly set dirty pages in MMU notifiers callback
> > > >    vhost: do not return -EAGAIN for non blocking invalidation too early
> > > > 
> > > >   drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
> > > >   drivers/vhost/vhost.h |   6 +-
> > > >   2 files changed, 122 insertions(+), 86 deletions(-)
> > > This generally looks more solid.
> > > 
> > > But this amounts to a significant overhaul of the code.
> > > 
> > > At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
> > > for this release, and then re-apply a corrected version
> > > for the next one?
> > 
> > 
> > If possible, consider we've actually disabled the feature. How about just
> > queued those patches for next release?
> > 
> > Thanks
> 
> Sorry if I was unclear. My idea is that
> 1. I revert the disabled code
> 2. You send a patch readding it with all the fixes squashed
> 3. Maybe optimizations on top right away?
> 4. We queue *that* for next and see what happens.
> 
> And the advantage over the patchy approach is that the current patches
> are hard to review. E.g.  it's not reasonable to ask RCU guys to review
> the whole of vhost for RCU usage but it's much more reasonable to ask
> about a specific patch.

I think there are other problems here too, I don't like that the use
of mmu notifiers is so different from every other driver, or that GUP
is called under spinlock.

So I favor the revert and try again approach as well. It is hard to
get a clear picture with these endless bug fix patches

Jason
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Wang]

On 2019/8/12 下午5:49, Michael S. Tsirkin wrote:
> On Mon, Aug 12, 2019 at 10:44:51AM +0800, Jason Wang wrote:
>> On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
>>> On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
>>>> Hi all:
>>>>
>>>> This series try to fix several issues introduced by meta data
>>>> accelreation series. Please review.
>>>>
>>>> Changes from V4:
>>>> - switch to use spinlock synchronize MMU notifier with accessors
>>>>
>>>> Changes from V3:
>>>> - remove the unnecessary patch
>>>>
>>>> Changes from V2:
>>>> - use seqlck helper to synchronize MMU notifier with vhost worker
>>>>
>>>> Changes from V1:
>>>> - try not use RCU to syncrhonize MMU notifier with vhost worker
>>>> - set dirty pages after no readers
>>>> - return -EAGAIN only when we find the range is overlapped with
>>>>     metadata
>>>>
>>>> Jason Wang (9):
>>>>     vhost: don't set uaddr for invalid address
>>>>     vhost: validate MMU notifier registration
>>>>     vhost: fix vhost map leak
>>>>     vhost: reset invalidate_count in vhost_set_vring_num_addr()
>>>>     vhost: mark dirty pages during map uninit
>>>>     vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
>>>>     vhost: do not use RCU to synchronize MMU notifier with worker
>>>>     vhost: correctly set dirty pages in MMU notifiers callback
>>>>     vhost: do not return -EAGAIN for non blocking invalidation too early
>>>>
>>>>    drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
>>>>    drivers/vhost/vhost.h |   6 +-
>>>>    2 files changed, 122 insertions(+), 86 deletions(-)
>>> This generally looks more solid.
>>>
>>> But this amounts to a significant overhaul of the code.
>>>
>>> At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
>>> for this release, and then re-apply a corrected version
>>> for the next one?
>>
>> If possible, consider we've actually disabled the feature. How about just
>> queued those patches for next release?
>>
>> Thanks
> Sorry if I was unclear. My idea is that
> 1. I revert the disabled code
> 2. You send a patch readding it with all the fixes squashed
> 3. Maybe optimizations on top right away?
> 4. We queue *that* for next and see what happens.
>
> And the advantage over the patchy approach is that the current patches
> are hard to review. E.g.  it's not reasonable to ask RCU guys to review
> the whole of vhost for RCU usage but it's much more reasonable to ask
> about a specific patch.


Ok. Then I agree to revert.

Thanks

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Wang]

On 2019/8/12 下午9:02, Jason Gunthorpe wrote:
> On Mon, Aug 12, 2019 at 05:49:08AM -0400, Michael S. Tsirkin wrote:
>> On Mon, Aug 12, 2019 at 10:44:51AM +0800, Jason Wang wrote:
>>> On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
>>>> On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
>>>>> Hi all:
>>>>>
>>>>> This series try to fix several issues introduced by meta data
>>>>> accelreation series. Please review.
>>>>>
>>>>> Changes from V4:
>>>>> - switch to use spinlock synchronize MMU notifier with accessors
>>>>>
>>>>> Changes from V3:
>>>>> - remove the unnecessary patch
>>>>>
>>>>> Changes from V2:
>>>>> - use seqlck helper to synchronize MMU notifier with vhost worker
>>>>>
>>>>> Changes from V1:
>>>>> - try not use RCU to syncrhonize MMU notifier with vhost worker
>>>>> - set dirty pages after no readers
>>>>> - return -EAGAIN only when we find the range is overlapped with
>>>>>     metadata
>>>>>
>>>>> Jason Wang (9):
>>>>>     vhost: don't set uaddr for invalid address
>>>>>     vhost: validate MMU notifier registration
>>>>>     vhost: fix vhost map leak
>>>>>     vhost: reset invalidate_count in vhost_set_vring_num_addr()
>>>>>     vhost: mark dirty pages during map uninit
>>>>>     vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
>>>>>     vhost: do not use RCU to synchronize MMU notifier with worker
>>>>>     vhost: correctly set dirty pages in MMU notifiers callback
>>>>>     vhost: do not return -EAGAIN for non blocking invalidation too early
>>>>>
>>>>>    drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
>>>>>    drivers/vhost/vhost.h |   6 +-
>>>>>    2 files changed, 122 insertions(+), 86 deletions(-)
>>>> This generally looks more solid.
>>>>
>>>> But this amounts to a significant overhaul of the code.
>>>>
>>>> At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
>>>> for this release, and then re-apply a corrected version
>>>> for the next one?
>>>
>>> If possible, consider we've actually disabled the feature. How about just
>>> queued those patches for next release?
>>>
>>> Thanks
>> Sorry if I was unclear. My idea is that
>> 1. I revert the disabled code
>> 2. You send a patch readding it with all the fixes squashed
>> 3. Maybe optimizations on top right away?
>> 4. We queue *that* for next and see what happens.
>>
>> And the advantage over the patchy approach is that the current patches
>> are hard to review. E.g.  it's not reasonable to ask RCU guys to review
>> the whole of vhost for RCU usage but it's much more reasonable to ask
>> about a specific patch.
> I think there are other problems here too, I don't like that the use
> of mmu notifiers is so different from every other driver, or that GUP
> is called under spinlock.


What kind of issues do you see? Spinlock is to synchronize GUP with MMU 
notifier in this series.

Btw, back to the original question. May I know why synchronize_rcu() is 
not suitable? Consider:

- MMU notifier are allowed to sleep
- MMU notifier could be preempted

If you mean something that prevents RCU grace period from running. I'm 
afraid MMU notifier is not the only victim.  But it should be no more 
worse than some one is holding a lock for very long time. If the only 
concern is the preemption of vhost kthread, I can switch to use 
rcu_read_lock_bh() instead.

Thanks


>
> So I favor the revert and try again approach as well. It is hard to
> get a clear picture with these endless bug fix patches
>
> Jason


Ok.

Thanks

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Gunthorpe]

On Tue, Aug 13, 2019 at 04:31:07PM +0800, Jason Wang wrote:

> What kind of issues do you see? Spinlock is to synchronize GUP with MMU
> notifier in this series.

A GUP that can't sleep can't pagefault which makes it a really weird
pattern

> Btw, back to the original question. May I know why synchronize_rcu() is not
> suitable? Consider:

We already went over this. You'd need to determine it doesn't somehow
deadlock the mm on reclaim paths. Maybe it is OK, the rcq_gq_wq is
marked WQ_MEM_RECLAIM at least..

I also think Michael was concerned about the latency spikes a long RCU
delay would cause.

Jason

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Christoph Hellwig]

On Tue, Aug 13, 2019 at 08:57:07AM -0300, Jason Gunthorpe wrote:
> On Tue, Aug 13, 2019 at 04:31:07PM +0800, Jason Wang wrote:
> 
> > What kind of issues do you see? Spinlock is to synchronize GUP with MMU
> > notifier in this series.
> 
> A GUP that can't sleep can't pagefault which makes it a really weird
> pattern

get_user_pages/get_user_pages_fast must not be called under a spinlock.
We have the somewhat misnamed __get_user_page_fast that just does a
lookup for existing pages and never faults for a few places that need
to do that lookup from contexts where we can't sleep.

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Wang]

On 2019/8/14 上午12:41, Christoph Hellwig wrote:
> On Tue, Aug 13, 2019 at 08:57:07AM -0300, Jason Gunthorpe wrote:
>> On Tue, Aug 13, 2019 at 04:31:07PM +0800, Jason Wang wrote:
>>
>>> What kind of issues do you see? Spinlock is to synchronize GUP with MMU
>>> notifier in this series.
>> A GUP that can't sleep can't pagefault which makes it a really weird
>> pattern
> get_user_pages/get_user_pages_fast must not be called under a spinlock.
> We have the somewhat misnamed __get_user_page_fast that just does a
> lookup for existing pages and never faults for a few places that need
> to do that lookup from contexts where we can't sleep.


Yes, I do use __get_user_pages_fast() in the code.

Thanks

_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Wang]

On 2019/8/13 下午7:57, Jason Gunthorpe wrote:
> On Tue, Aug 13, 2019 at 04:31:07PM +0800, Jason Wang wrote:
>
>> What kind of issues do you see? Spinlock is to synchronize GUP with MMU
>> notifier in this series.
> A GUP that can't sleep can't pagefault which makes it a really weird
> pattern


My understanding is __get_user_pages_fast() assumes caller can fail or 
have fallback. And we have graceful fallback to copy_{to|from}_user().


>
>> Btw, back to the original question. May I know why synchronize_rcu() is not
>> suitable? Consider:
> We already went over this. You'd need to determine it doesn't somehow
> deadlock the mm on reclaim paths. Maybe it is OK, the rcq_gq_wq is
> marked WQ_MEM_RECLAIM at least..


Yes, will take a look at this.


>
> I also think Michael was concerned about the latency spikes a long RCU
> delay would cause.


I don't think it's a real problem consider MMU notifier could be 
preempted or blocked.

Thanks


>
> Jason
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Gunthorpe]

On Thu, Aug 15, 2019 at 11:26:46AM +0800, Jason Wang wrote:
> 
> On 2019/8/13 下午7:57, Jason Gunthorpe wrote:
> > On Tue, Aug 13, 2019 at 04:31:07PM +0800, Jason Wang wrote:
> > 
> > > What kind of issues do you see? Spinlock is to synchronize GUP with MMU
> > > notifier in this series.
> > A GUP that can't sleep can't pagefault which makes it a really weird
> > pattern
> 
> 
> My understanding is __get_user_pages_fast() assumes caller can fail or have
> fallback. And we have graceful fallback to copy_{to|from}_user().

My point is that if you can fall back to copy_user then it is weird to
call the special non-sleeping GUP under a spinlock.

AFAIK the only reason this is done is because of the way the notifier
is being locked...

Jason
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Michael S. Tsirkin]

On Tue, Aug 13, 2019 at 04:12:49PM +0800, Jason Wang wrote:
> 
> On 2019/8/12 下午5:49, Michael S. Tsirkin wrote:
> > On Mon, Aug 12, 2019 at 10:44:51AM +0800, Jason Wang wrote:
> > > On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
> > > > On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
> > > > > Hi all:
> > > > > 
> > > > > This series try to fix several issues introduced by meta data
> > > > > accelreation series. Please review.
> > > > > 
> > > > > Changes from V4:
> > > > > - switch to use spinlock synchronize MMU notifier with accessors
> > > > > 
> > > > > Changes from V3:
> > > > > - remove the unnecessary patch
> > > > > 
> > > > > Changes from V2:
> > > > > - use seqlck helper to synchronize MMU notifier with vhost worker
> > > > > 
> > > > > Changes from V1:
> > > > > - try not use RCU to syncrhonize MMU notifier with vhost worker
> > > > > - set dirty pages after no readers
> > > > > - return -EAGAIN only when we find the range is overlapped with
> > > > >     metadata
> > > > > 
> > > > > Jason Wang (9):
> > > > >     vhost: don't set uaddr for invalid address
> > > > >     vhost: validate MMU notifier registration
> > > > >     vhost: fix vhost map leak
> > > > >     vhost: reset invalidate_count in vhost_set_vring_num_addr()
> > > > >     vhost: mark dirty pages during map uninit
> > > > >     vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
> > > > >     vhost: do not use RCU to synchronize MMU notifier with worker
> > > > >     vhost: correctly set dirty pages in MMU notifiers callback
> > > > >     vhost: do not return -EAGAIN for non blocking invalidation too early
> > > > > 
> > > > >    drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
> > > > >    drivers/vhost/vhost.h |   6 +-
> > > > >    2 files changed, 122 insertions(+), 86 deletions(-)
> > > > This generally looks more solid.
> > > > 
> > > > But this amounts to a significant overhaul of the code.
> > > > 
> > > > At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
> > > > for this release, and then re-apply a corrected version
> > > > for the next one?
> > > 
> > > If possible, consider we've actually disabled the feature. How about just
> > > queued those patches for next release?
> > > 
> > > Thanks
> > Sorry if I was unclear. My idea is that
> > 1. I revert the disabled code
> > 2. You send a patch readding it with all the fixes squashed
> > 3. Maybe optimizations on top right away?
> > 4. We queue *that* for next and see what happens.
> > 
> > And the advantage over the patchy approach is that the current patches
> > are hard to review. E.g.  it's not reasonable to ask RCU guys to review
> > the whole of vhost for RCU usage but it's much more reasonable to ask
> > about a specific patch.
> 
> 
> Ok. Then I agree to revert.
> 
> Thanks

Great, so please send the following:
- revert
- squashed and fixed patch

-- 
MST
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Wang]

On 2019/8/20 上午5:08, Michael S. Tsirkin wrote:
> On Tue, Aug 13, 2019 at 04:12:49PM +0800, Jason Wang wrote:
>> On 2019/8/12 下午5:49, Michael S. Tsirkin wrote:
>>> On Mon, Aug 12, 2019 at 10:44:51AM +0800, Jason Wang wrote:
>>>> On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
>>>>> On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
>>>>>> Hi all:
>>>>>>
>>>>>> This series try to fix several issues introduced by meta data
>>>>>> accelreation series. Please review.
>>>>>>
>>>>>> Changes from V4:
>>>>>> - switch to use spinlock synchronize MMU notifier with accessors
>>>>>>
>>>>>> Changes from V3:
>>>>>> - remove the unnecessary patch
>>>>>>
>>>>>> Changes from V2:
>>>>>> - use seqlck helper to synchronize MMU notifier with vhost worker
>>>>>>
>>>>>> Changes from V1:
>>>>>> - try not use RCU to syncrhonize MMU notifier with vhost worker
>>>>>> - set dirty pages after no readers
>>>>>> - return -EAGAIN only when we find the range is overlapped with
>>>>>>      metadata
>>>>>>
>>>>>> Jason Wang (9):
>>>>>>      vhost: don't set uaddr for invalid address
>>>>>>      vhost: validate MMU notifier registration
>>>>>>      vhost: fix vhost map leak
>>>>>>      vhost: reset invalidate_count in vhost_set_vring_num_addr()
>>>>>>      vhost: mark dirty pages during map uninit
>>>>>>      vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
>>>>>>      vhost: do not use RCU to synchronize MMU notifier with worker
>>>>>>      vhost: correctly set dirty pages in MMU notifiers callback
>>>>>>      vhost: do not return -EAGAIN for non blocking invalidation too early
>>>>>>
>>>>>>     drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
>>>>>>     drivers/vhost/vhost.h |   6 +-
>>>>>>     2 files changed, 122 insertions(+), 86 deletions(-)
>>>>> This generally looks more solid.
>>>>>
>>>>> But this amounts to a significant overhaul of the code.
>>>>>
>>>>> At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
>>>>> for this release, and then re-apply a corrected version
>>>>> for the next one?
>>>> If possible, consider we've actually disabled the feature. How about just
>>>> queued those patches for next release?
>>>>
>>>> Thanks
>>> Sorry if I was unclear. My idea is that
>>> 1. I revert the disabled code
>>> 2. You send a patch readding it with all the fixes squashed
>>> 3. Maybe optimizations on top right away?
>>> 4. We queue *that* for next and see what happens.
>>>
>>> And the advantage over the patchy approach is that the current patches
>>> are hard to review. E.g.  it's not reasonable to ask RCU guys to review
>>> the whole of vhost for RCU usage but it's much more reasonable to ask
>>> about a specific patch.
>>
>> Ok. Then I agree to revert.
>>
>> Thanks
> Great, so please send the following:
> - revert
> - squashed and fixed patch


Just to confirm, do you want me to send a single series or two?

Thanks


_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Michael S. Tsirkin]

On Tue, Aug 20, 2019 at 10:29:32AM +0800, Jason Wang wrote:
> 
> On 2019/8/20 上午5:08, Michael S. Tsirkin wrote:
> > On Tue, Aug 13, 2019 at 04:12:49PM +0800, Jason Wang wrote:
> > > On 2019/8/12 下午5:49, Michael S. Tsirkin wrote:
> > > > On Mon, Aug 12, 2019 at 10:44:51AM +0800, Jason Wang wrote:
> > > > > On 2019/8/11 上午1:52, Michael S. Tsirkin wrote:
> > > > > > On Fri, Aug 09, 2019 at 01:48:42AM -0400, Jason Wang wrote:
> > > > > > > Hi all:
> > > > > > > 
> > > > > > > This series try to fix several issues introduced by meta data
> > > > > > > accelreation series. Please review.
> > > > > > > 
> > > > > > > Changes from V4:
> > > > > > > - switch to use spinlock synchronize MMU notifier with accessors
> > > > > > > 
> > > > > > > Changes from V3:
> > > > > > > - remove the unnecessary patch
> > > > > > > 
> > > > > > > Changes from V2:
> > > > > > > - use seqlck helper to synchronize MMU notifier with vhost worker
> > > > > > > 
> > > > > > > Changes from V1:
> > > > > > > - try not use RCU to syncrhonize MMU notifier with vhost worker
> > > > > > > - set dirty pages after no readers
> > > > > > > - return -EAGAIN only when we find the range is overlapped with
> > > > > > >      metadata
> > > > > > > 
> > > > > > > Jason Wang (9):
> > > > > > >      vhost: don't set uaddr for invalid address
> > > > > > >      vhost: validate MMU notifier registration
> > > > > > >      vhost: fix vhost map leak
> > > > > > >      vhost: reset invalidate_count in vhost_set_vring_num_addr()
> > > > > > >      vhost: mark dirty pages during map uninit
> > > > > > >      vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
> > > > > > >      vhost: do not use RCU to synchronize MMU notifier with worker
> > > > > > >      vhost: correctly set dirty pages in MMU notifiers callback
> > > > > > >      vhost: do not return -EAGAIN for non blocking invalidation too early
> > > > > > > 
> > > > > > >     drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
> > > > > > >     drivers/vhost/vhost.h |   6 +-
> > > > > > >     2 files changed, 122 insertions(+), 86 deletions(-)
> > > > > > This generally looks more solid.
> > > > > > 
> > > > > > But this amounts to a significant overhaul of the code.
> > > > > > 
> > > > > > At this point how about we revert 7f466032dc9e5a61217f22ea34b2df932786bbfc
> > > > > > for this release, and then re-apply a corrected version
> > > > > > for the next one?
> > > > > If possible, consider we've actually disabled the feature. How about just
> > > > > queued those patches for next release?
> > > > > 
> > > > > Thanks
> > > > Sorry if I was unclear. My idea is that
> > > > 1. I revert the disabled code
> > > > 2. You send a patch readding it with all the fixes squashed
> > > > 3. Maybe optimizations on top right away?
> > > > 4. We queue *that* for next and see what happens.
> > > > 
> > > > And the advantage over the patchy approach is that the current patches
> > > > are hard to review. E.g.  it's not reasonable to ask RCU guys to review
> > > > the whole of vhost for RCU usage but it's much more reasonable to ask
> > > > about a specific patch.
> > > 
> > > Ok. Then I agree to revert.
> > > 
> > > Thanks
> > Great, so please send the following:
> > - revert
> > - squashed and fixed patch
> 
> 
> Just to confirm, do you want me to send a single series or two?
> 
> Thanks
> 

One is fine.

-- 
MST
_______________________________________________
Virtualization mailing list
Virtualization@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/virtualization

Re: [PATCH V5 0/9] Fixes for vhost metadata acceleration

[Michael S. Tsirkin]

On Thu, Aug 08, 2019 at 08:54:54PM +0800, Jason Wang wrote:
> I don't have any objection to convert  to spinlock() but just want to
> know if any case that the above smp_mb() + counter looks good to you?

So how about we try this:
- revert the original patch for this release
- new safe patch with a spinlock for the next release
- whatever improvements we can come up with on top

Thoughts?

Because I think this needs much more scrutiny than we can
give an incremental patch.

-- 
MST

[PATCH V5 0/9] Fixes for vhost metadata acceleration

[Jason Wang]

Hi all:

This series try to fix several issues introduced by meta data
accelreation series. Please review.

Changes from V4:
- switch to use spinlock synchronize MMU notifier with accessors

Changes from V3:
- remove the unnecessary patch

Changes from V2:
- use seqlck helper to synchronize MMU notifier with vhost worker

Changes from V1:
- try not use RCU to syncrhonize MMU notifier with vhost worker
- set dirty pages after no readers
- return -EAGAIN only when we find the range is overlapped with
  metadata

Jason Wang (9):
  vhost: don't set uaddr for invalid address
  vhost: validate MMU notifier registration
  vhost: fix vhost map leak
  vhost: reset invalidate_count in vhost_set_vring_num_addr()
  vhost: mark dirty pages during map uninit
  vhost: don't do synchronize_rcu() in vhost_uninit_vq_maps()
  vhost: do not use RCU to synchronize MMU notifier with worker
  vhost: correctly set dirty pages in MMU notifiers callback
  vhost: do not return -EAGAIN for non blocking invalidation too early

 drivers/vhost/vhost.c | 202 +++++++++++++++++++++++++-----------------
 drivers/vhost/vhost.h |   6 +-
 2 files changed, 122 insertions(+), 86 deletions(-)

-- 
2.18.1