On Wed, 5 Jun 2002, chas williams wrote: > In message <[EMAIL PROTECTED] > du>,Derrick J Brashear writes: > >going to sleep with the GLOCK isn't likely to result in fewer "strange > >side effects". > > is this case it is though. splitting the afs_getevent() into a > getevent and addevent isnt going to solve the problem because no > one went back through the code and called add_event before any > sleeps/wakeups. since add_event is called before the sleep and > needs to use alloc to get something actually sleep (we cant put > it on the wait queue until we have it) on it gunlocks and possibly > allows a wakeup to get by.
so why not add addevent to all the systypes and call it everywhere, always. > i really think its a bad idea for alloc to unlock/lock the glock > especially in this case since isnt an easy work around for the > above situation. trading one set of side effects for another isn't really a win, you just piss off a different group of users _______________________________________________ OpenAFS-devel mailing list [EMAIL PROTECTED] https://lists.openafs.org/mailman/listinfo/openafs-devel
