The reason for this is that we can not afford to waist CPU cycle for a full time slice of a process. So in case T1 acquired the spinlock and then prempted, now whenever T2 will be resuduled before T1 again reschedules, processor cycles will be waisted for whole time slice, in that cases T2 will just spin on the lock for the whole of its time slice, waiting for the lock to be released (this is the case of uni processor). In SMP the the situation might not be this much critical, as T1 can be rescheduled on processor 2 at the same time when T2 is getting executed at processor 1. I think Robert said right that critical section should not be prempted if we have acquired a spinlock, but it can be premptem if it has acquired a semaphore, as semaphore is a sleeping lock and does not put the other processes on spining, waiting for that lock to be released, rather those waiting processes are put to sleep on waitqueues. Regards, Gaurav -----Original Message----- From: kernelnewbies-bounce@xxxxxxxxxxxx [mailto:kernelnewbies-bounce@xxxxxxxxxxxx] On Behalf Of Robert Love Sent: Friday, September 10, 2004 9:06 AM To: martim@xxxxxxxxxxxxxxxxx Cc: kernelnewbies@xxxxxxxxxxxx Subject: Re: About kernel preemption On Fri, 2004-09-10 at 00:00 -0300, Martim Carbone wrote: > Is it really a deadlock? Let's assume the following scenario: > > - Thread 1 acquires lock A (spinlock) and gets preempted > - Thread 2 tries to acquire lock A and spins > - At some point in the future, the system timer interrupts thread 2, > determines that its timeslice has finished and sets need_resched. Some > other process it scheduled > - After X reschedules, thread 1 is once again scheduled and releases lock A. > - After Y reschedules, thread 2 is once again scheduled and, this > time, lock A is free, so it acquires the lock and executes the > critical region. Are you serious? I suppose in this example, assuming that T2 is always suspended and that T1 is always rescheduled, you do not have a deadlock, but an indefinite postponement, although this is not much more acceptable than a straight up deadlock. Robert Love -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/ -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
