Using __kuser_cmpxchg

Thu Nov 15 19:13:00 GMT 2007

Andrew Haley wrote:
> Nicolas Pitre writes:
> > On 2007年11月15日, Andrew Haley wrote:
> >
> > > Andrew Haley writes:
> > > > Paul Brook writes:
> > > > > For recent kernels there is a kernel helper you should use.
> > > > > See __kernel_cmpxchg in linux/arch/arm/kernel/entry-armv.S
> > > > >
> > > > > Any kernel capable of running on armv6/v7 hardware should have this helper.
> > > > > So should all targets on targets that define __ARM_EABI__
> > > > >
> > > > > So I suggest using the kernel helper for EABI and v6/v7 libraries, everyone
> > > > > else will have to make do with the crappy old sequence.
> > > >
> > > > OK, I'll investigate this as soon as I get some time.
> > >
> > > It seems to be impossible to use this kernel helper because
> > >
> > > * - A failure might be transient, i.e. it is possible, although unlikely,
> > > * that "failure" be returned even if *ptr == oldval.
> > >
> > > If this were to happen, the logic used to block and unblock threads
> > > wouldn't work.
> >
> > Why?
>> These are threads in the Java sense, not in the POSIX sense.
>> Each thread has a permit. 
>> "A call to park will return immediately if the permit is available,
> consuming it in the process; otherwise it may block. A call to unpark
> makes the permit available, if it was not already available. (Unlike
> with Semaphores though, permits do not accumulate. There is at most
> one.)
>> "Methods park and unpark provide efficient means of blocking and
> unblocking threads that do not encounter the problems that cause the
> deprecated methods Thread.suspend and Thread.resume to be unusable for
> such purposes: Races between one thread invoking park and another
> thread trying to unpark it will preserve liveness, due to the permit."
>> http://java.sun.com/javase/6/docs/api/java/util/concurrent/locks/LockSupport.html
>> Java threads can be in three states, PARKED, RUNNING, and PERMIT.
> When unpark() is called, we do this:
>> /* If this thread is in state RUNNING, give it a permit and return
> immediately. */
> if (compare_and_swap
> (ptr, Thread::THREAD_PARK_RUNNING, Thread::THREAD_PARK_PERMIT))
> return;
>> /* If this thread is parked, put it into state RUNNING and send it a
> signal. */
> if (compare_and_swap
> (ptr, Thread::THREAD_PARK_PARKED, Thread::THREAD_PARK_RUNNING))
> {
> ...
>> To get back to ARM's __kuser_cmpxchg: if this second compare_and_swap
> fails spuriously we won't unblock the thread.

Am I missing something here?
Is it really impossible to implement compare_and_swap on these ARM 
platforms?
David Daney