We observed two panics involving races with igb_reset_task.
The first panic is caused by this race condition:
kworker reboot -f
igb_reset_task
igb_reinit_locked
igb_down
napi_synchronize
__igb_shutdown
igb_clear_interrupt_scheme
igb_free_q_vectors
igb_free_q_vector
adapter->q_vector[v_idx] = NULL;
napi_disable
Panics trying to access
adapter->q_vector[v_idx].napi_state
The second panic (a divide error) is caused by this race:
kworker reboot -f tx packet
igb_reset_task
__igb_shutdown
rtnl_lock()
...
igb_clear_interrupt_scheme
igb_free_q_vectors
adapter->num_tx_queues = 0
...
rtnl_unlock()
rtnl_lock()
igb_reinit_locked
igb_down
igb_up
netif_tx_start_all_queues
dev_hard_start_xmit
igb_xmit_frame
igb_tx_queue_mapping
Panics on
r_idx % adapter->num_tx_queues
This commit applies to igb_reset_task the same changes that
were applied to ixgbe in commit
2f90b8657ec9 ("ixgbe: this patch
adds support for DCB to the kernel and ixgbe driver"),
commit
8f4c5c9fb87a ("ixgbe: reinit_locked() should be called with
rtnl_lock") and commit
88adce4ea8f9 ("ixgbe: fix possible race in
reset subtask").
Signed-off-by: Francesco Ruggeri <fruggeri@arista.com>
Tested-by: Aaron Brown <aaron.f.brown@intel.com>
Signed-off-by: Tony Nguyen <anthony.l.nguyen@intel.com>
struct igb_adapter *adapter;
adapter = container_of(work, struct igb_adapter, reset_task);
+ rtnl_lock();
+ /* If we're already down or resetting, just bail */
+ if (test_bit(__IGB_DOWN, &adapter->state) ||
+ test_bit(__IGB_RESETTING, &adapter->state)) {
+ rtnl_unlock();
+ return;
+ }
+
igb_dump(adapter);
netdev_err(adapter->netdev, "Reset adapter\n");
igb_reinit_locked(adapter);
+ rtnl_unlock();
}
/**