On 4/2/22 15:47, Michael Straube wrote:
Hi all, smatch reported a sleeping in atomic context. rtw_set_802_11_disassociate() <- disables preempt -> _rtw_pwr_wakeup() -> ips_leave() rtw_set_802_11_disassociate() takes a spinlock and ips_leave() uses a mutex. I'm fairly new to the locking stuff, but as far as I know this is not a false positive since mutex can sleep, but that's not allowed under a spinlock. What is the best way to handle this? I'm not sure if converting the mutex to a spinlock (including all the other places where the mutex is used) is the right thing to do?
In drivers/net/wireless/realtek/rtlwifi, we had a similar problem. There it was handled by putting the lps_enter() and lps_leave() operations in a separate workqueue. In this case, the routines were rtl_lps_enter() and rtl_lps_leave(). Each of them sets a variable to indicate whether enter_ps is true or false, and schedules the workqueue. In the workqueue's callback routine, the routines to start/stop ps mode are called. The code is in drivers/net/wireless/realtek/rtlwifi/ps.c.
This solution is only one of many, and there may be a better one. Larry
