Thanks for the tips. I did not pay attention to optimization. I will send the patch with the fix. Thanks Nohee -----Original Message----- From: Jiri Slaby [mailto:jirislaby@xxxxxxxxx] Sent: Wednesday, October 13, 2010 6:21 AM To: Nohee Ko Cc: greg@xxxxxxxxx; devel@xxxxxxxxxxxxxxxxxxxxxx; joe@xxxxxxxxxxx; Henry Ptasinski; Brett Rudley; Venkat Rao; grundler@xxxxxxxxxx; jason@xxxxxxxxxxxxxx Subject: Re: [PATCH] staging:brcm80211:bug fix- rmmod hang problem On 10/12/2010 10:33 PM, nohee ko wrote: > --- a/drivers/staging/brcm80211/include/linuxver.h > +++ b/drivers/staging/brcm80211/include/linuxver.h > @@ -35,14 +35,11 @@ > #undef IP_TOS > #include <asm/io.h> > > -#define KILL_PROC(nr, sig) \ > +#define KILL_PROC(pid, sig) \ > do { \ > struct task_struct *tsk; \ > - struct pid *pid; \ > - pid = find_get_pid((pid_t)nr); \ > - tsk = pid_task(pid, PIDTYPE_PID); \ > - if (tsk) \ > - send_sig(sig, tsk, 1); \ > + tsk = pid_task(find_vpid(pid), PIDTYPE_PID); \ > + if (tsk) \ > + send_sig(sig, tsk, 1); \ This is broken. You have to hold rcu_read_lock() here otherwise you are unsafe and can blow up. Actually you don't need that crap at all. You have already task_struct, so you can easily call send_sig directly. And instead of these weird plays with signals and abuse of semaphores, you can use completion, can't you? regards, -- js _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/devel
