[PATCH net v2 0/2] virtio_net: fix lock warning and unrecoverable state

[PATCH net v2 0/2] virtio_net: fix lock warning and unrecoverable state

[Heng Qi]

Patch 1 describes and fixes an issue where dim cannot return to
normal state in certain scenarios.

Patch 2 attempts to resolve lockdep's complaints that holding many
nested locks and when there is a maximum number of queues may also
be problematic, so try to remove the dim_lock.

Changelog
=======
v1->v2:
  Patch(2/2): rephase the commit log.

Heng Qi (2):
  virtio_net: fix possible dim status unrecoverable
  Revert "virtio_net: Add a lock for per queue RX coalesce"

 drivers/net/virtio_net.c | 55 ++++++++++------------------------------
 1 file changed, 13 insertions(+), 42 deletions(-)

-- 
2.32.0.3.g01195cf9f

[PATCH net v2 1/2] virtio_net: fix possible dim status unrecoverable

[Heng Qi]

When the dim worker is scheduled, if it no longer needs to issue
commands, dim may not be able to return to the working state later.

For example, the following single queue scenario:
  1. The dim worker of rxq0 is scheduled, and the dim status is
     changed to DIM_APPLY_NEW_PROFILE;
  2. dim is disabled or parameters have not been modified;
  3. virtnet_rx_dim_work exits directly;

Then, even if net_dim is invoked again, it cannot work because the
state is not restored to DIM_START_MEASURE.

Fixes: 6208799553a8 ("virtio-net: support rx netdim")
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
Reviewed-by: Jiri Pirko <jiri@nvidia.com>
---
 drivers/net/virtio_net.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 4e1a0fc0d555..1cad06cef230 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -4417,9 +4417,9 @@ static void virtnet_rx_dim_work(struct work_struct *work)
 		if (err)
 			pr_debug("%s: Failed to send dim parameters on rxq%d\n",
 				 dev->name, qnum);
-		dim->state = DIM_START_MEASURE;
 	}
 out:
+	dim->state = DIM_START_MEASURE;
 	mutex_unlock(&rq->dim_lock);
 }
 
-- 
2.32.0.3.g01195cf9f

[PATCH net v2 2/2] Revert "virtio_net: Add a lock for per queue RX coalesce"

[Heng Qi]

This reverts commit 4d4ac2ececd3c42a08dd32a6e3a4aaf25f7efe44.

When the following snippet is run, lockdep will report a deadlock[1].

  /* Acquire all queues dim_locks */
  for (i = 0; i < vi->max_queue_pairs; i++)
          mutex_lock(&vi->rq[i].dim_lock);

There's no deadlock here because the vq locks are always taken
in the same order, but lockdep can not figure it out, and we
can not make each lock a separate class because there can be more
than MAX_LOCKDEP_SUBCLASSES of vqs.

However, dropping the lock is harmless:
  1. If dim is enabled, modifications made by dim worker to coalescing
     params may cause the user's query results to be dirty data.
  2. In scenarios (a) and (b), a spurious dim worker is scheduled,
     but this can be handled correctly:
     (a)
       1. dim is on
       2. net_dim call schedules a worker
       3. dim is turning off
       4. The worker checks that dim is off and then exits after
          restoring dim's state.
       5. The worker will not be scheduled until the next time dim is on.

     (b)
       1. dim is on
       2. net_dim call schedules a worker
       3. The worker checks that dim is on and keeps going
       4. dim is turning off
       5. The worker successfully configure this parameter to the device.
       6. The worker will not be scheduled until the next time dim is on.

[1]
========================================================
WARNING: possible recursive locking detected
6.9.0-rc7+ #319 Not tainted
--------------------------------------------
ethtool/962 is trying to acquire lock:

but task is already holding lock:

other info that might help us debug this:
Possible unsafe locking scenario:

      CPU0
      ----
 lock(&vi->rq[i].dim_lock);
 lock(&vi->rq[i].dim_lock);

*** DEADLOCK ***

 May be due to missing lock nesting notation

3 locks held by ethtool/962:
 #0: ffffffff82dbaab0 (cb_lock){++++}-{3:3}, at: genl_rcv+0x19/0x40
 #1: ffffffff82dad0a8 (rtnl_mutex){+.+.}-{3:3}, at:
				ethnl_default_set_doit+0xbe/0x1e0

stack backtrace:
CPU: 6 PID: 962 Comm: ethtool Not tainted 6.9.0-rc7+ #319
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
	   rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
Call Trace:
 <TASK>
 dump_stack_lvl+0x79/0xb0
 check_deadlock+0x130/0x220
 __lock_acquire+0x861/0x990
 lock_acquire.part.0+0x72/0x1d0
 ? lock_acquire+0xf8/0x130
 __mutex_lock+0x71/0xd50
 virtnet_set_coalesce+0x151/0x190
 __ethnl_set_coalesce.isra.0+0x3f8/0x4d0
 ethnl_set_coalesce+0x34/0x90
 ethnl_default_set_doit+0xdd/0x1e0
 genl_family_rcv_msg_doit+0xdc/0x130
 genl_family_rcv_msg+0x154/0x230
 ? __pfx_ethnl_default_set_doit+0x10/0x10
 genl_rcv_msg+0x4b/0xa0
 ? __pfx_genl_rcv_msg+0x10/0x10
 netlink_rcv_skb+0x5a/0x110
 genl_rcv+0x28/0x40
 netlink_unicast+0x1af/0x280
 netlink_sendmsg+0x20e/0x460
 __sys_sendto+0x1fe/0x210
 ? find_held_lock+0x2b/0x80
 ? do_user_addr_fault+0x3a2/0x8a0
 ? __lock_release+0x5e/0x160
 ? do_user_addr_fault+0x3a2/0x8a0
 ? lock_release+0x72/0x140
 ? do_user_addr_fault+0x3a7/0x8a0
 __x64_sys_sendto+0x29/0x30
 do_syscall_64+0x78/0x180
 entry_SYSCALL_64_after_hwframe+0x76/0x7e

Fixes: 4d4ac2ececd3 ("virtio_net: Add a lock for per queue RX coalesce")
Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>
---
v1->v2: Rephase the commit log.

 drivers/net/virtio_net.c | 53 +++++++++-------------------------------
 1 file changed, 12 insertions(+), 41 deletions(-)

diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 1cad06cef230..e4a1dff2a64a 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -316,9 +316,6 @@ struct receive_queue {
 	/* Is dynamic interrupt moderation enabled? */
 	bool dim_enabled;
 
-	/* Used to protect dim_enabled and inter_coal */
-	struct mutex dim_lock;
-
 	/* Dynamic Interrupt Moderation */
 	struct dim dim;
 
@@ -2368,10 +2365,6 @@ static int virtnet_poll(struct napi_struct *napi, int budget)
 	/* Out of packets? */
 	if (received < budget) {
 		napi_complete = virtqueue_napi_complete(napi, rq->vq, received);
-		/* Intentionally not taking dim_lock here. This may result in a
-		 * spurious net_dim call. But if that happens virtnet_rx_dim_work
-		 * will not act on the scheduled work.
-		 */
 		if (napi_complete && rq->dim_enabled)
 			virtnet_rx_dim_update(vi, rq);
 	}
@@ -3247,11 +3240,9 @@ static int virtnet_set_ringparam(struct net_device *dev,
 				return err;
 
 			/* The reason is same as the transmit virtqueue reset */
-			mutex_lock(&vi->rq[i].dim_lock);
 			err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, i,
 							       vi->intr_coal_rx.max_usecs,
 							       vi->intr_coal_rx.max_packets);
-			mutex_unlock(&vi->rq[i].dim_lock);
 			if (err)
 				return err;
 		}
@@ -4257,7 +4248,6 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
 	struct virtio_net_ctrl_coal_rx *coal_rx __free(kfree) = NULL;
 	bool rx_ctrl_dim_on = !!ec->use_adaptive_rx_coalesce;
 	struct scatterlist sgs_rx;
-	int ret = 0;
 	int i;
 
 	if (rx_ctrl_dim_on && !virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL))
@@ -4267,22 +4257,16 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
 			       ec->rx_max_coalesced_frames != vi->intr_coal_rx.max_packets))
 		return -EINVAL;
 
-	/* Acquire all queues dim_locks */
-	for (i = 0; i < vi->max_queue_pairs; i++)
-		mutex_lock(&vi->rq[i].dim_lock);
-
 	if (rx_ctrl_dim_on && !vi->rx_dim_enabled) {
 		vi->rx_dim_enabled = true;
 		for (i = 0; i < vi->max_queue_pairs; i++)
 			vi->rq[i].dim_enabled = true;
-		goto unlock;
+		return 0;
 	}
 
 	coal_rx = kzalloc(sizeof(*coal_rx), GFP_KERNEL);
-	if (!coal_rx) {
-		ret = -ENOMEM;
-		goto unlock;
-	}
+	if (!coal_rx)
+		return -ENOMEM;
 
 	if (!rx_ctrl_dim_on && vi->rx_dim_enabled) {
 		vi->rx_dim_enabled = false;
@@ -4300,10 +4284,8 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
 				  VIRTIO_NET_CTRL_NOTF_COAL_RX_SET,
-				  &sgs_rx)) {
-		ret = -EINVAL;
-		goto unlock;
-	}
+				  &sgs_rx))
+		return -EINVAL;
 
 	vi->intr_coal_rx.max_usecs = ec->rx_coalesce_usecs;
 	vi->intr_coal_rx.max_packets = ec->rx_max_coalesced_frames;
@@ -4311,11 +4293,8 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
 		vi->rq[i].intr_coal.max_usecs = ec->rx_coalesce_usecs;
 		vi->rq[i].intr_coal.max_packets = ec->rx_max_coalesced_frames;
 	}
-unlock:
-	for (i = vi->max_queue_pairs - 1; i >= 0; i--)
-		mutex_unlock(&vi->rq[i].dim_lock);
 
-	return ret;
+	return 0;
 }
 
 static int virtnet_send_notf_coal_cmds(struct virtnet_info *vi,
@@ -4339,24 +4318,19 @@ static int virtnet_send_rx_notf_coal_vq_cmds(struct virtnet_info *vi,
 					     u16 queue)
 {
 	bool rx_ctrl_dim_on = !!ec->use_adaptive_rx_coalesce;
+	bool cur_rx_dim = vi->rq[queue].dim_enabled;
 	u32 max_usecs, max_packets;
-	bool cur_rx_dim;
 	int err;
 
-	mutex_lock(&vi->rq[queue].dim_lock);
-	cur_rx_dim = vi->rq[queue].dim_enabled;
 	max_usecs = vi->rq[queue].intr_coal.max_usecs;
 	max_packets = vi->rq[queue].intr_coal.max_packets;
 
 	if (rx_ctrl_dim_on && (ec->rx_coalesce_usecs != max_usecs ||
-			       ec->rx_max_coalesced_frames != max_packets)) {
-		mutex_unlock(&vi->rq[queue].dim_lock);
+			       ec->rx_max_coalesced_frames != max_packets))
 		return -EINVAL;
-	}
 
 	if (rx_ctrl_dim_on && !cur_rx_dim) {
 		vi->rq[queue].dim_enabled = true;
-		mutex_unlock(&vi->rq[queue].dim_lock);
 		return 0;
 	}
 
@@ -4369,8 +4343,10 @@ static int virtnet_send_rx_notf_coal_vq_cmds(struct virtnet_info *vi,
 	err = virtnet_send_rx_ctrl_coal_vq_cmd(vi, queue,
 					       ec->rx_coalesce_usecs,
 					       ec->rx_max_coalesced_frames);
-	mutex_unlock(&vi->rq[queue].dim_lock);
-	return err;
+	if (err)
+		return err;
+
+	return 0;
 }
 
 static int virtnet_send_notf_coal_vq_cmds(struct virtnet_info *vi,
@@ -4404,7 +4380,6 @@ static void virtnet_rx_dim_work(struct work_struct *work)
 
 	qnum = rq - vi->rq;
 
-	mutex_lock(&rq->dim_lock);
 	if (!rq->dim_enabled)
 		goto out;
 
@@ -4420,7 +4395,6 @@ static void virtnet_rx_dim_work(struct work_struct *work)
 	}
 out:
 	dim->state = DIM_START_MEASURE;
-	mutex_unlock(&rq->dim_lock);
 }
 
 static int virtnet_coal_params_supported(struct ethtool_coalesce *ec)
@@ -4558,13 +4532,11 @@ static int virtnet_get_per_queue_coalesce(struct net_device *dev,
 		return -EINVAL;
 
 	if (virtio_has_feature(vi->vdev, VIRTIO_NET_F_VQ_NOTF_COAL)) {
-		mutex_lock(&vi->rq[queue].dim_lock);
 		ec->rx_coalesce_usecs = vi->rq[queue].intr_coal.max_usecs;
 		ec->tx_coalesce_usecs = vi->sq[queue].intr_coal.max_usecs;
 		ec->tx_max_coalesced_frames = vi->sq[queue].intr_coal.max_packets;
 		ec->rx_max_coalesced_frames = vi->rq[queue].intr_coal.max_packets;
 		ec->use_adaptive_rx_coalesce = vi->rq[queue].dim_enabled;
-		mutex_unlock(&vi->rq[queue].dim_lock);
 	} else {
 		ec->rx_max_coalesced_frames = 1;
 
@@ -5396,7 +5368,6 @@ static int virtnet_alloc_queues(struct virtnet_info *vi)
 
 		u64_stats_init(&vi->rq[i].stats.syncp);
 		u64_stats_init(&vi->sq[i].stats.syncp);
-		mutex_init(&vi->rq[i].dim_lock);
 	}
 
 	return 0;
-- 
2.32.0.3.g01195cf9f

Re: [PATCH net v2 2/2] Revert "virtio_net: Add a lock for per queue RX coalesce"

[Jiri Pirko]

Thu, May 23, 2024 at 09:46:51AM CEST, hengqi@linux.alibaba.com wrote:
>This reverts commit 4d4ac2ececd3c42a08dd32a6e3a4aaf25f7efe44.
>
>When the following snippet is run, lockdep will report a deadlock[1].
>
>  /* Acquire all queues dim_locks */
>  for (i = 0; i < vi->max_queue_pairs; i++)
>          mutex_lock(&vi->rq[i].dim_lock);
>
>There's no deadlock here because the vq locks are always taken
>in the same order, but lockdep can not figure it out, and we
>can not make each lock a separate class because there can be more
>than MAX_LOCKDEP_SUBCLASSES of vqs.
>
>However, dropping the lock is harmless:
>  1. If dim is enabled, modifications made by dim worker to coalescing
>     params may cause the user's query results to be dirty data.
>  2. In scenarios (a) and (b), a spurious dim worker is scheduled,
>     but this can be handled correctly:
>     (a)
>       1. dim is on
>       2. net_dim call schedules a worker
>       3. dim is turning off
>       4. The worker checks that dim is off and then exits after
>          restoring dim's state.
>       5. The worker will not be scheduled until the next time dim is on.
>
>     (b)
>       1. dim is on
>       2. net_dim call schedules a worker
>       3. The worker checks that dim is on and keeps going
>       4. dim is turning off
>       5. The worker successfully configure this parameter to the device.
>       6. The worker will not be scheduled until the next time dim is on.
>
>[1]
>========================================================
>WARNING: possible recursive locking detected
>6.9.0-rc7+ #319 Not tainted
>--------------------------------------------
>ethtool/962 is trying to acquire lock:
>
>but task is already holding lock:
>
>other info that might help us debug this:
>Possible unsafe locking scenario:
>
>      CPU0
>      ----
> lock(&vi->rq[i].dim_lock);
> lock(&vi->rq[i].dim_lock);
>
>*** DEADLOCK ***
>
> May be due to missing lock nesting notation
>
>3 locks held by ethtool/962:
> #0: ffffffff82dbaab0 (cb_lock){++++}-{3:3}, at: genl_rcv+0x19/0x40
> #1: ffffffff82dad0a8 (rtnl_mutex){+.+.}-{3:3}, at:
>				ethnl_default_set_doit+0xbe/0x1e0
>
>stack backtrace:
>CPU: 6 PID: 962 Comm: ethtool Not tainted 6.9.0-rc7+ #319
>Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
>	   rel-1.16.0-0-gd239552ce722-prebuilt.qemu.org 04/01/2014
>Call Trace:
> <TASK>
> dump_stack_lvl+0x79/0xb0
> check_deadlock+0x130/0x220
> __lock_acquire+0x861/0x990
> lock_acquire.part.0+0x72/0x1d0
> ? lock_acquire+0xf8/0x130
> __mutex_lock+0x71/0xd50
> virtnet_set_coalesce+0x151/0x190
> __ethnl_set_coalesce.isra.0+0x3f8/0x4d0
> ethnl_set_coalesce+0x34/0x90
> ethnl_default_set_doit+0xdd/0x1e0
> genl_family_rcv_msg_doit+0xdc/0x130
> genl_family_rcv_msg+0x154/0x230
> ? __pfx_ethnl_default_set_doit+0x10/0x10
> genl_rcv_msg+0x4b/0xa0
> ? __pfx_genl_rcv_msg+0x10/0x10
> netlink_rcv_skb+0x5a/0x110
> genl_rcv+0x28/0x40
> netlink_unicast+0x1af/0x280
> netlink_sendmsg+0x20e/0x460
> __sys_sendto+0x1fe/0x210
> ? find_held_lock+0x2b/0x80
> ? do_user_addr_fault+0x3a2/0x8a0
> ? __lock_release+0x5e/0x160
> ? do_user_addr_fault+0x3a2/0x8a0
> ? lock_release+0x72/0x140
> ? do_user_addr_fault+0x3a7/0x8a0
> __x64_sys_sendto+0x29/0x30
> do_syscall_64+0x78/0x180
> entry_SYSCALL_64_after_hwframe+0x76/0x7e
>
>Fixes: 4d4ac2ececd3 ("virtio_net: Add a lock for per queue RX coalesce")
>Signed-off-by: Heng Qi <hengqi@linux.alibaba.com>

Reviewed-by: Jiri Pirko <jiri@nvidia.com>

Re: [PATCH net v2 2/2] Revert "virtio_net: Add a lock for per queue RX coalesce"

[Paolo Abeni]

On Thu, 2024-05-23 at 15:46 +0800, Heng Qi wrote:
> This reverts commit 4d4ac2ececd3c42a08dd32a6e3a4aaf25f7efe44.
> 
> When the following snippet is run, lockdep will report a deadlock[1].
> 
>   /* Acquire all queues dim_locks */
>   for (i = 0; i < vi->max_queue_pairs; i++)
>           mutex_lock(&vi->rq[i].dim_lock);
> 
> There's no deadlock here because the vq locks are always taken
> in the same order, but lockdep can not figure it out, and we
> can not make each lock a separate class because there can be more
> than MAX_LOCKDEP_SUBCLASSES of vqs.
> 
> However, dropping the lock is harmless:
>   1. If dim is enabled, modifications made by dim worker to coalescing
>      params may cause the user's query results to be dirty data.

It looks like the above can confuse the user-space/admin?

Have you considered instead re-factoring
virtnet_send_rx_notf_coal_cmds() to avoid acquiring all the mutex in
sequence?

Thanks!

Paolo

Re: [PATCH net v2 2/2] Revert "virtio_net: Add a lock for per queue RX coalesce"

[Heng Qi]

On Mon, 27 May 2024 12:42:43 +0200, Paolo Abeni <pabeni@redhat.com> wrote:
> On Thu, 2024-05-23 at 15:46 +0800, Heng Qi wrote:
> > This reverts commit 4d4ac2ececd3c42a08dd32a6e3a4aaf25f7efe44.
> > 
> > When the following snippet is run, lockdep will report a deadlock[1].
> > 
> >   /* Acquire all queues dim_locks */
> >   for (i = 0; i < vi->max_queue_pairs; i++)
> >           mutex_lock(&vi->rq[i].dim_lock);
> > 
> > There's no deadlock here because the vq locks are always taken
> > in the same order, but lockdep can not figure it out, and we
> > can not make each lock a separate class because there can be more
> > than MAX_LOCKDEP_SUBCLASSES of vqs.
> > 
> > However, dropping the lock is harmless:
> >   1. If dim is enabled, modifications made by dim worker to coalescing
> >      params may cause the user's query results to be dirty data.
> 
> It looks like the above can confuse the user-space/admin?

Maybe, but we don't seem to guarantee this --
the global query interface (.get_coalesce) cannot 
guarantee correct results when the DIM and .get_per_queue_coalesce are present:

1. DIM has been around for a long time (it will modify the per-queue parameters),
   but many nics only have interfaces for querying global parameters.
2. Some nics provide the .get_per_queue_coalesce interface, it is not
   synchronized with DIM.

So I think this is acceptable.

> 
> Have you considered instead re-factoring
> virtnet_send_rx_notf_coal_cmds() to avoid acquiring all the mutex in
> sequence?

Perhaps it is a way to not traverse and update the parameters of each queue
in the global settings interface.

Thanks.

> 
> Thanks!
> 
> Paolo
> 

Re: [PATCH net v2 2/2] Revert "virtio_net: Add a lock for per queue RX coalesce"

[Paolo Abeni]

On Tue, 2024-05-28 at 11:06 +0800, Heng Qi wrote:
> On Mon, 27 May 2024 12:42:43 +0200, Paolo Abeni <pabeni@redhat.com> wrote:
> > On Thu, 2024-05-23 at 15:46 +0800, Heng Qi wrote:
> > > This reverts commit 4d4ac2ececd3c42a08dd32a6e3a4aaf25f7efe44.
> > > 
> > > When the following snippet is run, lockdep will report a deadlock[1].
> > > 
> > >   /* Acquire all queues dim_locks */
> > >   for (i = 0; i < vi->max_queue_pairs; i++)
> > >           mutex_lock(&vi->rq[i].dim_lock);
> > > 
> > > There's no deadlock here because the vq locks are always taken
> > > in the same order, but lockdep can not figure it out, and we
> > > can not make each lock a separate class because there can be more
> > > than MAX_LOCKDEP_SUBCLASSES of vqs.
> > > 
> > > However, dropping the lock is harmless:
> > >   1. If dim is enabled, modifications made by dim worker to coalescing
> > >      params may cause the user's query results to be dirty data.
> > 
> > It looks like the above can confuse the user-space/admin?
> 
> Maybe, but we don't seem to guarantee this --
> the global query interface (.get_coalesce) cannot 
> guarantee correct results when the DIM and .get_per_queue_coalesce are present:
> 
> 1. DIM has been around for a long time (it will modify the per-queue parameters),
>    but many nics only have interfaces for querying global parameters.
> 2. Some nics provide the .get_per_queue_coalesce interface, it is not
>    synchronized with DIM.
> 
> So I think this is acceptable.

Yes, the above sounds acceptable to me.

> > Have you considered instead re-factoring
> > virtnet_send_rx_notf_coal_cmds() to avoid acquiring all the mutex in
> > sequence?
> 
> Perhaps it is a way to not traverse and update the parameters of each queue
> in the global settings interface.

I'm wondering if something as dumb as the following would suffice? Not
even compile-tested.
---
diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
index 4a802c0ea2cb..d844f4c89152 100644
--- a/drivers/net/virtio_net.c
+++ b/drivers/net/virtio_net.c
@@ -4267,27 +4267,27 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
 			       ec->rx_max_coalesced_frames != vi->intr_coal_rx.max_packets))
 		return -EINVAL;
 
-	/* Acquire all queues dim_locks */
-	for (i = 0; i < vi->max_queue_pairs; i++)
-		mutex_lock(&vi->rq[i].dim_lock);
-
 	if (rx_ctrl_dim_on && !vi->rx_dim_enabled) {
 		vi->rx_dim_enabled = true;
-		for (i = 0; i < vi->max_queue_pairs; i++)
+		for (i = 0; i < vi->max_queue_pairs; i++) {
+			mutex_lock(&vi->rq[i].dim_lock);
 			vi->rq[i].dim_enabled = true;
-		goto unlock;
+			mutex_unlock(&vi->rq[i].dim_lock);
+		}
+		return 0;
 	}
 
 	coal_rx = kzalloc(sizeof(*coal_rx), GFP_KERNEL);
-	if (!coal_rx) {
-		ret = -ENOMEM;
-		goto unlock;
-	}
+	if (!coal_rx)
+		return -ENOMEM;
 
 	if (!rx_ctrl_dim_on && vi->rx_dim_enabled) {
 		vi->rx_dim_enabled = false;
-		for (i = 0; i < vi->max_queue_pairs; i++)
+		for (i = 0; i < vi->max_queue_pairs; i++) {
+			mutex_lock(&vi->rq[i].dim_lock);
 			vi->rq[i].dim_enabled = false;
+			mutex_unlock(&vi->rq[i].dim_lock);
+		}
 	}
 
 	/* Since the per-queue coalescing params can be set,
@@ -4300,21 +4300,17 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
 
 	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
 				  VIRTIO_NET_CTRL_NOTF_COAL_RX_SET,
-				  &sgs_rx)) {
-		ret = -EINVAL;
-		goto unlock;
-	}
+				  &sgs_rx))
+		return -EINVAL;
 
 	vi->intr_coal_rx.max_usecs = ec->rx_coalesce_usecs;
 	vi->intr_coal_rx.max_packets = ec->rx_max_coalesced_frames;
 	for (i = 0; i < vi->max_queue_pairs; i++) {
+		mutex_lock(&vi->rq[i].dim_lock);
 		vi->rq[i].intr_coal.max_usecs = ec->rx_coalesce_usecs;
 		vi->rq[i].intr_coal.max_packets = ec->rx_max_coalesced_frames;
-	}
-unlock:
-	for (i = vi->max_queue_pairs - 1; i >= 0; i--)
 		mutex_unlock(&vi->rq[i].dim_lock);
-
+	}
 	return ret;
 }
---

Otherwise I think you need to add {READ,WRITE}_ONCE annotations while
touching the dim fields to avoid data races.

Thanks,

Paolo

Re: [PATCH net v2 2/2] Revert "virtio_net: Add a lock for per queue RX coalesce"

[Heng Qi]

在 2024/5/28 下午6:04, Paolo Abeni 写道:
> On Tue, 2024-05-28 at 11:06 +0800, Heng Qi wrote:
>> On Mon, 27 May 2024 12:42:43 +0200, Paolo Abeni <pabeni@redhat.com> wrote:
>>> On Thu, 2024-05-23 at 15:46 +0800, Heng Qi wrote:
>>>> This reverts commit 4d4ac2ececd3c42a08dd32a6e3a4aaf25f7efe44.
>>>>
>>>> When the following snippet is run, lockdep will report a deadlock[1].
>>>>
>>>>    /* Acquire all queues dim_locks */
>>>>    for (i = 0; i < vi->max_queue_pairs; i++)
>>>>            mutex_lock(&vi->rq[i].dim_lock);
>>>>
>>>> There's no deadlock here because the vq locks are always taken
>>>> in the same order, but lockdep can not figure it out, and we
>>>> can not make each lock a separate class because there can be more
>>>> than MAX_LOCKDEP_SUBCLASSES of vqs.
>>>>
>>>> However, dropping the lock is harmless:
>>>>    1. If dim is enabled, modifications made by dim worker to coalescing
>>>>       params may cause the user's query results to be dirty data.
>>> It looks like the above can confuse the user-space/admin?
>> Maybe, but we don't seem to guarantee this --
>> the global query interface (.get_coalesce) cannot
>> guarantee correct results when the DIM and .get_per_queue_coalesce are present:
>>
>> 1. DIM has been around for a long time (it will modify the per-queue parameters),
>>     but many nics only have interfaces for querying global parameters.
>> 2. Some nics provide the .get_per_queue_coalesce interface, it is not
>>     synchronized with DIM.
>>
>> So I think this is acceptable.
> Yes, the above sounds acceptable to me.
>
>>> Have you considered instead re-factoring
>>> virtnet_send_rx_notf_coal_cmds() to avoid acquiring all the mutex in
>>> sequence?
>> Perhaps it is a way to not traverse and update the parameters of each queue
>> in the global settings interface.
> I'm wondering if something as dumb as the following would suffice? Not
> even compile-tested.

This alleviates the problem, and l would like to repost this fix.

Thanks.
> ---
> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c
> index 4a802c0ea2cb..d844f4c89152 100644
> --- a/drivers/net/virtio_net.c
> +++ b/drivers/net/virtio_net.c
> @@ -4267,27 +4267,27 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
>   			       ec->rx_max_coalesced_frames != vi->intr_coal_rx.max_packets))
>   		return -EINVAL;
>   
> -	/* Acquire all queues dim_locks */
> -	for (i = 0; i < vi->max_queue_pairs; i++)
> -		mutex_lock(&vi->rq[i].dim_lock);
> -
>   	if (rx_ctrl_dim_on && !vi->rx_dim_enabled) {
>   		vi->rx_dim_enabled = true;
> -		for (i = 0; i < vi->max_queue_pairs; i++)
> +		for (i = 0; i < vi->max_queue_pairs; i++) {
> +			mutex_lock(&vi->rq[i].dim_lock);
>   			vi->rq[i].dim_enabled = true;
> -		goto unlock;
> +			mutex_unlock(&vi->rq[i].dim_lock);
> +		}
> +		return 0;
>   	}
>   
>   	coal_rx = kzalloc(sizeof(*coal_rx), GFP_KERNEL);
> -	if (!coal_rx) {
> -		ret = -ENOMEM;
> -		goto unlock;
> -	}
> +	if (!coal_rx)
> +		return -ENOMEM;
>   
>   	if (!rx_ctrl_dim_on && vi->rx_dim_enabled) {
>   		vi->rx_dim_enabled = false;
> -		for (i = 0; i < vi->max_queue_pairs; i++)
> +		for (i = 0; i < vi->max_queue_pairs; i++) {
> +			mutex_lock(&vi->rq[i].dim_lock);
>   			vi->rq[i].dim_enabled = false;
> +			mutex_unlock(&vi->rq[i].dim_lock);
> +		}
>   	}
>   
>   	/* Since the per-queue coalescing params can be set,
> @@ -4300,21 +4300,17 @@ static int virtnet_send_rx_notf_coal_cmds(struct virtnet_info *vi,
>   
>   	if (!virtnet_send_command(vi, VIRTIO_NET_CTRL_NOTF_COAL,
>   				  VIRTIO_NET_CTRL_NOTF_COAL_RX_SET,
> -				  &sgs_rx)) {
> -		ret = -EINVAL;
> -		goto unlock;
> -	}
> +				  &sgs_rx))
> +		return -EINVAL;
>   
>   	vi->intr_coal_rx.max_usecs = ec->rx_coalesce_usecs;
>   	vi->intr_coal_rx.max_packets = ec->rx_max_coalesced_frames;
>   	for (i = 0; i < vi->max_queue_pairs; i++) {
> +		mutex_lock(&vi->rq[i].dim_lock);
>   		vi->rq[i].intr_coal.max_usecs = ec->rx_coalesce_usecs;
>   		vi->rq[i].intr_coal.max_packets = ec->rx_max_coalesced_frames;
> -	}
> -unlock:
> -	for (i = vi->max_queue_pairs - 1; i >= 0; i--)
>   		mutex_unlock(&vi->rq[i].dim_lock);
> -
> +	}
>   	return ret;
>   }
> ---
>
> Otherwise I think you need to add {READ,WRITE}_ONCE annotations while
> touching the dim fields to avoid data races.
>
> Thanks,
>
> Paolo
>
>
>
>