There is a memory barrier in followed case. When set the port down,
hclgevf_set_timmer will set DOWN in state. Meanwhile, the service task has
different behaviour based on whether the state is DOWN. Thus, to make sure
service task see DOWN, use smp_mb__after_atomic after calling set_bit().
CPU0 CPU1
========================== ===================================
hclgevf_set_timer_task() hclgevf_periodic_service_task()
set_bit(DOWN,state) test_bit(DOWN,state)
pf also has this issue.
Fixes:
ff200099d271 ("net: hns3: remove unnecessary work in hclgevf_main")
Fixes:
1c6dfe6fc6f7 ("net: hns3: remove mailbox and reset work in hclge_main")
Signed-off-by: Peiyang Wang <wangpeiyang1@huawei.com>
Signed-off-by: Jijie Shao <shaojijie@huawei.com>
Reviewed-by: Simon Horman <horms@kernel.org>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
/* Set the DOWN flag here to disable link updating */
set_bit(HCLGE_STATE_DOWN, &hdev->state);
- /* flush memory to make sure DOWN is seen by service task */
- smp_mb__before_atomic();
+ smp_mb__after_atomic(); /* flush memory to make sure DOWN is seen by service task */
hclge_flush_link_update(hdev);
}
}
} else {
set_bit(HCLGEVF_STATE_DOWN, &hdev->state);
- /* flush memory to make sure DOWN is seen by service task */
- smp_mb__before_atomic();
+ smp_mb__after_atomic(); /* flush memory to make sure DOWN is seen by service task */
hclgevf_flush_link_update(hdev);
}
}