A safe SCHED_IDLE implementation [LWN.net]

A safe SCHED_IDLE implementation

Posted Jul 5, 2002 19:04 UTC (Fri) by xoddam (subscriber, #2322)
Parent article: A safe SCHED_IDLE implementation

(You seem to have ommitted <pre> on the post,
which is only really readable if you 'view source').

As discussed, surely it is still possible for a priority
inversion to occur if a SCHED_BATCH process holds a
shared resource in userspace? Processes holding kernel
resources may not be suspended indefinitely, but
deadlocks on locks which can be held while running in
userspace (such as pthreads mutexes or lock files)
will still be possible. I suppose the idea is that
that is a problem to be solved in userspace :-)

to post comments

A safe SCHED_IDLE implementation

Posted Jul 7, 2002 21:38 UTC (Sun) by Peter (guest, #1127) [Link]

As discussed, surely it is still possible for a priority inversion to occur if a SCHED_BATCH process holds a shared resource in userspace?

Priority inversion can be simulated by merely taking a resource and never releasing it. So if you can't trust your users not to DOS you with a priority inversion, you shouldn't give them permission to lock those shared resources in the first place.

As for unintentional priority inversion - yes, it's a userspace problem. (: