On Thu, 2004-09-09 at 17:12 +0000, Martim Carbone wrote: Carbone! Hello! > Why can't the kernel be preempted when locks are being held inside it? > > If a thread executing in kernel space enters a critical region and > gets preempted, the lock will protect the critical region from other > threads, so I don't see what is the problem. This would work for semaphores (which in Linux are sleeping locks) but not for spinlocks (which are busy-wait locks). So, in fact, you can preempt while a semaphore is held but not while a spin lock is held. The reason is because you can deadlock (acquire lock A, preempt, new thread acquires lock A). It goes further than this, too, in that code holding a spin locks believes itself to be atomic. If it suddenly was scheduled away for awhile, that might annoy whatever it is doing. It would also not help lock contention. This might beg the question "well, why not make spinlocks into sleeping locks?" This is totally possible, if interrupt handlers were threaded. Right now they are not threaded, so they cannot sleep. Hope this helps, Robert Love -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
