]> git.ipfire.org Git - thirdparty/kernel/stable.git/commitdiff
ibmvnic: Return error code on TX scrq flush fail
authorNick Child <nnac123@linux.ibm.com>
Tue, 16 Apr 2024 16:41:28 +0000 (11:41 -0500)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Fri, 7 Mar 2025 15:56:34 +0000 (16:56 +0100)
[ Upstream commit 5cb431dcf8048572e9ffc6c30cdbd8832cbe502d ]

In ibmvnic_xmit() if ibmvnic_tx_scrq_flush() returns H_CLOSED then
it will inform upper level networking functions to disable tx
queues. H_CLOSED signals that the connection with the vnic server is
down and a transport event is expected to recover the device.

Previously, ibmvnic_tx_scrq_flush() was hard-coded to return success.
Therefore, the queues would remain active until ibmvnic_cleanup() is
called within do_reset().

The problem is that do_reset() depends on the RTNL lock. If several
ibmvnic devices are resetting then there can be a long wait time until
the last device can grab the lock. During this time the tx/rx queues
still appear active to upper level functions.

FYI, we do make a call to netif_carrier_off() outside the RTNL lock but
its calls to dev_deactivate() are also dependent on the RTNL lock.

As a result, large amounts of retransmissions were observed in a short
period of time, eventually leading to ETIMEOUT. This was specifically
seen with HNV devices, likely because of even more RTNL dependencies.

Therefore, ensure the return code of ibmvnic_tx_scrq_flush() is
propagated to the xmit function to allow for an earlier (and lock-less)
response to a transport event.

Signed-off-by: Nick Child <nnac123@linux.ibm.com>
Link: https://lore.kernel.org/r/20240416164128.387920-1-nnac123@linux.ibm.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Stable-dep-of: bdf5d13aa05e ("ibmvnic: Don't reference skb after sending to VIOS")
Signed-off-by: Sasha Levin <sashal@kernel.org>
drivers/net/ethernet/ibm/ibmvnic.c

index 6d17738c1c536fc0eafea27d700a2d70589a268a..7fe1fefef9934b54ff13d926525d2f39f06cc58e 100644 (file)
@@ -2181,7 +2181,7 @@ static int ibmvnic_tx_scrq_flush(struct ibmvnic_adapter *adapter,
                ibmvnic_tx_scrq_clean_buffer(adapter, tx_scrq);
        else
                ind_bufp->index = 0;
-       return 0;
+       return rc;
 }
 
 static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
@@ -2234,7 +2234,9 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
                tx_dropped++;
                tx_send_failed++;
                ret = NETDEV_TX_OK;
-               ibmvnic_tx_scrq_flush(adapter, tx_scrq);
+               lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
+               if (lpar_rc != H_SUCCESS)
+                       goto tx_err;
                goto out;
        }
 
@@ -2249,8 +2251,10 @@ static netdev_tx_t ibmvnic_xmit(struct sk_buff *skb, struct net_device *netdev)
                dev_kfree_skb_any(skb);
                tx_send_failed++;
                tx_dropped++;
-               ibmvnic_tx_scrq_flush(adapter, tx_scrq);
                ret = NETDEV_TX_OK;
+               lpar_rc = ibmvnic_tx_scrq_flush(adapter, tx_scrq);
+               if (lpar_rc != H_SUCCESS)
+                       goto tx_err;
                goto out;
        }