When configuring TC-MQPRIO offload, only turn off netdev carrier and
don't bring physical link down in hardware. Otherwise, when the
physical link is brought up again after configuration, it gets
re-trained and stalls ongoing traffic.
Also, when firmware is no longer accessible or crashed, avoid sending
FLOWC and waiting for reply that will never come.
Fix following hung_task_timeout_secs trace seen in these cases.
INFO: task tc:20807 blocked for more than 122 seconds.
      Tainted: G S                5.13.0-rc3+ #122
"echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
task:tc   state:D stack:14768 pid:20807 ppid: 19366 flags:0x00000000
Call Trace:
 __schedule+0x27b/0x6a0
 schedule+0x37/0xa0
 schedule_preempt_disabled+0x5/0x10
 __mutex_lock.isra.14+0x2a0/0x4a0
 ? netlink_lookup+0x120/0x1a0
 ? rtnl_fill_ifinfo+0x10f0/0x10f0
 __netlink_dump_start+0x70/0x250
 rtnetlink_rcv_msg+0x28b/0x380
 ? rtnl_fill_ifinfo+0x10f0/0x10f0
 ? rtnl_calcit.isra.42+0x120/0x120
 netlink_rcv_skb+0x4b/0xf0
 netlink_unicast+0x1a0/0x280
 netlink_sendmsg+0x216/0x440
 sock_sendmsg+0x56/0x60
 __sys_sendto+0xe9/0x150
 ? handle_mm_fault+0x6d/0x1b0
 ? do_user_addr_fault+0x1c5/0x620
 __x64_sys_sendto+0x1f/0x30
 do_syscall_64+0x3c/0x80
 entry_SYSCALL_64_after_hwframe+0x44/0xae
RIP: 0033:0x7f7f73218321
RSP: 002b:
00007ffd19626208 EFLAGS: 
00000246 ORIG_RAX: 
000000000000002c
RAX: 
ffffffffffffffda RBX: 
000055b7c0a8b240 RCX: 
00007f7f73218321
RDX: 
0000000000000028 RSI: 
00007ffd19626210 RDI: 
0000000000000003
RBP: 
000055b7c08680ff R08: 
0000000000000000 R09: 
0000000000000000
R10: 
0000000000000000 R11: 
0000000000000246 R12: 
000055b7c085f5f6
R13: 
000055b7c085f60a R14: 
00007ffd19636470 R15: 
00007ffd196262a0
Fixes: b1396c2bd675 ("cxgb4: parse and configure TC-MQPRIO offload")
Signed-off-by: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
                          bool persistent, u8 *smt_idx);
 int cxgb4_get_msix_idx_from_bmap(struct adapter *adap);
 void cxgb4_free_msix_idx_in_bmap(struct adapter *adap, u32 msix_idx);
-int cxgb_open(struct net_device *dev);
-int cxgb_close(struct net_device *dev);
 void cxgb4_enable_rx(struct adapter *adap, struct sge_rspq *q);
 void cxgb4_quiesce_rx(struct sge_rspq *q);
 int cxgb4_port_mirror_alloc(struct net_device *dev);
 
 /*
  * net_device operations
  */
-int cxgb_open(struct net_device *dev)
+static int cxgb_open(struct net_device *dev)
 {
        struct port_info *pi = netdev_priv(dev);
        struct adapter *adapter = pi->adapter;
        return err;
 }
 
-int cxgb_close(struct net_device *dev)
+static int cxgb_close(struct net_device *dev)
 {
        struct port_info *pi = netdev_priv(dev);
        struct adapter *adapter = pi->adapter;
 
         * down before configuring tc params.
         */
        if (netif_running(dev)) {
-               cxgb_close(dev);
+               netif_tx_stop_all_queues(dev);
+               netif_carrier_off(dev);
                needs_bring_up = true;
        }
 
        }
 
 out:
-       if (needs_bring_up)
-               cxgb_open(dev);
+       if (needs_bring_up) {
+               netif_tx_start_all_queues(dev);
+               netif_carrier_on(dev);
+       }
 
        mutex_unlock(&adap->tc_mqprio->mqprio_mutex);
        return ret;
 
        if (!eosw_txq)
                return -ENOMEM;
 
+       if (!(adap->flags & CXGB4_FW_OK)) {
+               /* Don't stall caller when access to FW is lost */
+               complete(&eosw_txq->completion);
+               return -EIO;
+       }
+
        skb = alloc_skb(len, GFP_KERNEL);
        if (!skb)
                return -ENOMEM;