git/list[1] front-page[2] threads[3] people[4] search[5] about
 

Re: [RFT PATCH 1/2] win32: optimize condition variable implementation

From
Johannes Sixt <j.sixt@viscovery.net>
Date
Jun 8, 2010, 16:16 UTC
Message-ID
<4C0E6CC2.1080605@viscovery.net>
In-Reply-To
<1275917892-16437-2-git-send-email-bonzini@gnu.org>
Am 07.06.2010 15:38, schrieb Paolo Bonzini:
Show 39 quoted lines
>   int pthread_cond_wait(pthread_cond_t *cond, CRITICAL_SECTION *mutex)
>   {
> -	int last_waiter;
> +	int num_waiters;
>
> -	EnterCriticalSection(&cond->waiters_lock);
> +	/*
> +	 * This access is protected under the mutex.
> +	 */
>   	cond->waiters++;
> -	LeaveCriticalSection(&cond->waiters_lock);
>
>   	/*
>   	 * Unlock external mutex and wait for signal.
> @@ -105,17 +104,17 @@ int pthread_cond_wait(pthread_cond_t *cond, CRITICAL_SECTION *mutex)
>   	WaitForSingleObject(cond->sema, INFINITE);
>
>   	/*
> -	 * Decrease waiters count. If we are the last waiter, then we must
> +	 * Decrease waiters count.  The mutex prevents concurrent increments,
> +	 * so doing this decrement atomically is enough.
> +	 */
> +	num_waiters = InterlockedDecrement(&cond->waiters);
> +
> +	/* If we are the last waiter, then we must
>   	 * notify the broadcasting thread that it can continue.
>   	 * But if we continued due to cond_signal, we do not have to do that
>   	 * because the signaling thread knows that only one waiter continued.
>   	 */
> -	EnterCriticalSection(&cond->waiters_lock);
> -	cond->waiters--;
> -	last_waiter = cond->was_broadcast&&  cond->waiters == 0;
> -	LeaveCriticalSection(&cond->waiters_lock);
> -
> -	if (last_waiter) {
> +	if (num_waiters == 0&&  cond->was_broadcast) {
>   		/*
>   		 * cond_broadcast was issued while mutex was held. This means
>   		 * that all other waiters have continued, but are contending

This is not correct. While it is not possible that two threads increment waiters at the same time due to the external mutex, it is still possible that on thread increments, and a different one decrements. You lost all provisions to avoid that.

Furthermore, waiters_lock not only protects waiters, but also the combined state of waiters and was_broadcast. You break this protection. See also here:

Show 22 quoted lines
> @@ -168,12 +168,18 @@ int pthread_cond_signal(pthread_cond_t *cond)
>    */
>   int pthread_cond_broadcast(pthread_cond_t *cond)
>   {
> -	EnterCriticalSection(&cond->waiters_lock);
> +	/*
> +	 * As in pthread_cond_signal, access to cond->waiters and
> +	 * cond->was_broadcast is locked via the external mutex.
> +	 */
>
>   	if ((cond->was_broadcast = cond->waiters>  0)) {
> +		BOOLEAN result;
>   		/* wake up all waiters */
> -		ReleaseSemaphore(cond->sema, cond->waiters, NULL);
> -		LeaveCriticalSection(&cond->waiters_lock);
> +		result = ReleaseSemaphore(cond->sema, cond->waiters, NULL);
> +		if (!result)
> +			return err_win_to_posix(GetLastError());
> +
>   		/*
>   		 * At this point all waiters continue. Each one takes its
>   		 * slice of the semaphor. Now it's our turn to wait: Since
-- Hannes
Previous: Paolo BonziniNext: Paolo Bonzini
Message 3 of 9 in “win32: optimize emulation of condition variables”
  1. 0/2 win32: optimize emulation of condition variablesPaolo Bonzini, Jun 7, 2010
  2. 1/2 win32: optimize condition variable implementationPaolo Bonzini, Jun 7, 2010
  3. Johannes SixtJun 8, 2010
  4. Paolo BonziniJun 8, 2010
  5. 2/2 win32: optimize pthread_cond_broadcastPaolo Bonzini, Jun 7, 2010
  6. Johannes SixtJun 8, 2010
  7. Paolo BonziniJun 8, 2010
  8. Johannes SixtJun 8, 2010
  9. 3/2 fix race in win32 pthread_cond_signal causing spurious wakeupsPaolo Bonzini, Jun 13, 2010

Read the whole thread, see it on lore, or plain text.

$ cat FOOTERMessages come from the public archive at lore.kernel.org/git, fetched every hour. The front page is chosen and written each morning by an AI editor and can be wrong; the threads themselves are the record. About and API. For agents: an MCP server at https://gitlist.dev/mcp, and any thread, story or person page as Markdown by adding .md to its URL (or sending Accept: text/markdown). Details in /llms.txt.