Re: Locks used in the FAT file system are non-atomic and in fact, don't work on SMP systems

Alan Cox (alan@lxorguk.ukuu.org.uk)
Thu, 26 Aug 1999 20:17:08 +0100 (BST)

> I code reviewed sleep_on and wake_up in the linux kernel. I wrote the SMP
> kernel for NetWare 4, and discovered that unless you are holding the lock
> over the context switch, there is always a very small window between
> something going to sleep and waking up where A) you can end up not waking up
> anyone or B) two guys can get through at the same time. The logic doesn't

Im confused. What exactly do you think wake_up and sleep_on guarantee

wake_up wakes everyone on the queue at some arbitary instant
sleep_on makes no guarantee about what occurs if you allow a wakeup while
you are in sleep_on, until you are asleep.

That is why code tends to use add_wait_queue/remove_wait_queue.

> post operation will allow more than one guy to wake up in such a scenario.
> It has to do with how Intel's pipelines work and assumptions about cached
> data. MESI really is "messy" to deal with in some cases on SMP.

The wait queue locks are store barriers because the lock will always have
executed a lock btsl prior to being taken. The unlock likewise has done
a lock btrl

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/