Date:	Mon, 12 Oct 2009 18:37:17 +0300
From:	Török Edwin <edwin@...mav.net>
To:	Peter Zijlstra <peterz@...radead.org>
CC:	Ingo Molnar <mingo@...e.hu>,
	Linux Kernel <linux-kernel@...r.kernel.org>,
	aCaB <acab@...mav.net>, David Howells <dhowells@...hat.com>,
	Nick Piggin <npiggin@...e.de>,
	Linus Torvalds <torvalds@...ux-foundation.org>,
	Thomas Gleixner <tglx@...utronix.de>
Subject: Re: Mutex vs semaphores scheduler bug

On 2009-10-12 17:53, Peter Zijlstra wrote:
> On Sat, 2009-10-10 at 17:57 +0300, Török Edwin wrote:
>> If a semaphore (such as mmap_sem) is heavily congested, then using a
>> userspace mutex makes the program faster.
>>
>> For example using a mutex around *anonymous* mmaps, speeds it up
>> significantly (~80% on this microbenchmark,
>> ~15% on real applications). Such workarounds shouldn't  be necessary for
>> userspace applications, the kernel should
>> by default use the most efficient implementation for locks.
> 
> Should, yes, does, no.
> 
>> However when using a mutex the number of context switches is SMALLER by
>> 40-60%.
> 
> That matches the problem, see below.
> 
>> I think its a bug in the scheduler, it scheduler the mutex case much
>> better. 
> 
> It's not, the scheduler doesn't know about mutexes/futexes/rwsems.
> 
>> Maybe because userspace also spins a bit before actually calling
>> futex().
> 
> Nope, if we would ever spin, it would be in the kernel after calling
> FUTEX_LOCK (which currently doesn't exist). glibc shouldn't do any
> spinning on its own (if it does, I have yet another reason to try and
> supplant the glibc futex code).

I think it doesn't by default, I was mislead by the huge number of cases
in pthread_mutex_lock.c. The default one does this:

__lll_lock_wait:
	cfi_startproc
	pushq	%r10
	cfi_adjust_cfa_offset(8)
	pushq	%rdx
	cfi_adjust_cfa_offset(8)
	cfi_offset(%r10, -16)
	cfi_offset(%rdx, -24)
	xorq	%r10, %r10	/* No timeout.  */
	movl	$2, %edx
	LOAD_FUTEX_WAIT (%esi)

	cmpl	%edx, %eax	/* NB:	 %edx == 2 */
	jne	2f

1:	movl	$SYS_futex, %eax
	syscall

2:	movl	%edx, %eax
	xchgl	%eax, (%rdi)	/* NB:	 lock is implied */

	testl	%eax, %eax
	jnz	1b

	popq	%rdx
	cfi_adjust_cfa_offset(-8)
	cfi_restore(%rdx)
	popq	%r10
	cfi_adjust_cfa_offset(-8)
	cfi_restore(%r10)
	retq


> 
>> I think its important to optimize the mmap_sem semaphore
> 
> It is.
> 
> The problem appears to be that rwsem doesn't allow lock-stealing

OK, sorry for mistaking lack of lock-stealing with scheduler bug.

>, and
> very strictly maintains FIFO order on contention. This results in extra
> schedules and reduced performance as you noticed.
> 
> What happens is that when we release a contended rwsem we assign it to
> the next waiter, if before that waiter gets ran, another (running) tasks
> comes along and tries to acquire the lock, that gets put to sleep, even
> though it could possibly get to acquire it (and the woken waiter would
> detect failure and go back to sleep).

The reason I initially thought it was a scheduler bug is that it seemed
it has something to do with wakeups, and threads are sleeping for too
long waiting for the lock.
But I think the scheduler can't give preference to tasks which would be
able to acquire a semaphore they were sleeping on, because that'd throw
 fair scheduling off-balance, right?

> 
> So what I think we need to do is have a look at all this lib/rwsem.c
> slowpath code and hack in lock stealing.
> 
> 

Best regards,
--Edwin
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists