ibmvnic: Ensure login failure recovery is safe from other resets

If a login request fails, the recovery process should be protected
against parallel resets. It is a known issue that freeing and
registering CRQ's in quick succession can result in a failover CRQ from
the VIOS. Processing a failover during login recovery is dangerous for
two reasons:
 1. This will result in two parallel initialization processes, this can
 cause serious issues during login.
 2. It is possible that the failover CRQ is received but never executed.
 We get notified of a pending failover through a transport event CRQ.
 The reset is not performed until a INIT CRQ request is received.
 Previously, if CRQ init fails during login recovery, then the ibmvnic
 irq is freed and the login process returned error. If failover_pending
 is true (a transport event was received), then the ibmvnic device
 would never be able to process the reset since it cannot receive the
 CRQ_INIT request due to the irq being freed. This leaved the device
 in a inoperable state.

Therefore, the login failure recovery process must be hardened against
these possible issues. Possible failovers (due to quick CRQ free and
init) must be avoided and any issues during re-initialization should be
dealt with instead of being propagated up the stack. This logic is
similar to that of ibmvnic_probe().

Fixes: dff515a ("ibmvnic: Harden device login requests")
Signed-off-by: Nick Child <[email protected]>
Reviewed-by: Simon Horman <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

Author: Nick Child

drivers/net/ethernet/ibm/ibmvnic.c

@@ -116,6 +116,7 @@ static void ibmvnic_tx_scrq_clean_buffer(struct ibmvnic_adapter *adapter,
 static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 				struct ibmvnic_long_term_buff *ltb);
 static void ibmvnic_disable_irqs(struct ibmvnic_adapter *adapter);
+static void flush_reset_queue(struct ibmvnic_adapter *adapter);
 
 struct ibmvnic_stat {
 	char name[ETH_GSTRING_LEN];
@@ -1507,8 +1508,8 @@ static const char *adapter_state_to_string(enum vnic_state state)
 
 static int ibmvnic_login(struct net_device *netdev)
 {
+	unsigned long flags, timeout = msecs_to_jiffies(20000);
 	struct ibmvnic_adapter *adapter = netdev_priv(netdev);
-	unsigned long timeout = msecs_to_jiffies(20000);
 	int retry_count = 0;
 	int retries = 10;
 	bool retry;
@@ -1573,6 +1574,7 @@ static int ibmvnic_login(struct net_device *netdev)
 					    "SCRQ irq initialization failed\n");
 				return rc;
 			}
+		/* Default/timeout error handling, reset and start fresh */
 		} else if (adapter->init_done_rc) {
 			netdev_warn(netdev, "Adapter login failed, init_done_rc = %d\n",
 				    adapter->init_done_rc);
@@ -1588,29 +1590,53 @@ static int ibmvnic_login(struct net_device *netdev)
 				    "Freeing and re-registering CRQs before attempting to login again\n");
 			retry = true;
 			adapter->init_done_rc = 0;
-			retry_count++;
 			release_sub_crqs(adapter, true);
-			reinit_init_done(adapter);
-			release_crq_queue(adapter);
-			/* If we don't sleep here then we risk an unnecessary
-			 * failover event from the VIOS. This is a known VIOS
-			 * issue caused by a vnic device freeing and registering
-			 * a CRQ too quickly.
+			/* Much of this is similar logic as ibmvnic_probe(),
+			 * we are essentially re-initializing communication
+			 * with the server. We really should not run any
+			 * resets/failovers here because this is already a form
+			 * of reset and we do not want parallel resets occurring
 			 */
-			msleep(1500);
-			rc = init_crq_queue(adapter);
-			if (rc) {
-				netdev_err(netdev, "login recovery: init CRQ failed %d\n",
-					   rc);
-				return -EIO;
-			}
+			do {
+				reinit_init_done(adapter);
+				/* Clear any failovers we got in the previous
+				 * pass since we are re-initializing the CRQ
+				 */
+				adapter->failover_pending = false;
+				release_crq_queue(adapter);
+				/* If we don't sleep here then we risk an
+				 * unnecessary failover event from the VIOS.
+				 * This is a known VIOS issue caused by a vnic
+				 * device freeing and registering a CRQ too
+				 * quickly.
+				 */
+				msleep(1500);
+				/* Avoid any resets, since we are currently
+				 * resetting.
+				 */
+				spin_lock_irqsave(&adapter->rwi_lock, flags);
+				flush_reset_queue(adapter);
+				spin_unlock_irqrestore(&adapter->rwi_lock,
+						       flags);
+
+				rc = init_crq_queue(adapter);
+				if (rc) {
+					netdev_err(netdev, "login recovery: init CRQ failed %d\n",
+						   rc);
+					return -EIO;
+				}
 
-			rc = ibmvnic_reset_init(adapter, false);
-			if (rc) {
-				netdev_err(netdev, "login recovery: Reset init failed %d\n",
-					   rc);
-				return -EIO;
-			}
+				rc = ibmvnic_reset_init(adapter, false);
+				if (rc)
+					netdev_err(netdev, "login recovery: Reset init failed %d\n",
+						   rc);
+				/* IBMVNIC_CRQ_INIT will return EAGAIN if it
+				 * fails, since ibmvnic_reset_init will free
+				 * irq's in failure, we won't be able to receive
+				 * new CRQs so we need to keep trying. probe()
+				 * handles this similarly.
+				 */
+			} while (rc == -EAGAIN && retry_count++ < retries);
 		}
 	} while (retry);