DMARC-Filter: OpenDMARC Filter v1.4.2 sourceware.org 990643858D20
To: Takashi Yano <takashi DOT yano AT nifty DOT ne DOT jp>, cygwin AT cygwin DOT com,
        Noel Grandin <noelgrandin AT gmail DOT com>
Subject: Re: [PATCH v2] Cygwin: pthread: Fix a race issue introduced by the
 commit 2c5433e5da82
Date: Thu, 30 May 2024 10:47:38 +0200
Message-ID: <1721092.SVj2RUA3Z2@nimes>
In-Reply-To: <0126a98a-3bdc-4303-a0d2-09f4c7009392@gmail.com>
References: <20240530050538 DOT 53724-1-takashi DOT yano AT nifty DOT ne DOT jp>
 <0126a98a-3bdc-4303-a0d2-09f4c7009392 AT gmail DOT com>
MIME-Version: 1.0
Precedence: list
From: Bruno Haible via Cygwin <cygwin AT cygwin DOT com>
Reply-To: Bruno Haible <bruno AT clisp DOT org>
Content-Type: text/plain; charset="utf-8"
Errors-To: cygwin-bounces+archive-cygwin=delorie DOT com AT cygwin DOT com
Sender: "Cygwin" <cygwin-bounces+archive-cygwin=delorie DOT com AT cygwin DOT com>
Content-Transfer-Encoding: 8bit

Noel Grandin wrote in cygwin-patches:
> Pardon my ignorance, but why not rather use the Windows SRWLock functionality?
> https://learn.microsoft.com/en-us/windows/win32/sync/slim-reader-writer--srw--locks
> 
> SRW locks are very fast, only require a single pointer-sized storage area, can be statically initialised, and do not 
> need to be destroyed, so there is no possibibilty of memory leakage.
> 
> Then the implementation simply becomes
> 
> int pthread::once (pthread_once_t *once_control, void (*init_routine) (void))
> {
>      AcquireSRWLockExclusive(once_control->lock);
>      if (!once_control->state)
>      {
>          init_routine()
>          once_control->state = 1;
>      }
>      ReleaseSRWLockExclusive(once_control->lock);
> }

SRW locks are spin-locks. Since they are only pointer-sized,
ReleaseSRWLockExclusive cannot notify other threads — unlike CRITICAL_SECTION.
Therefore, AcquireSRWLockExclusive must busy-loop when the lock is already
held.

But busy-looping is a bad practice for pthread_once. In the case,
for example, that the init_routine reads a multi-megabyte data
structure into memory and parses it, the other threads that wait
on the result of this operation would by busy-looping, eating
full CPU time.

Bruno




-- 
Problem reports:      https://cygwin.com/problems.html
FAQ:                  https://cygwin.com/faq/
Documentation:        https://cygwin.com/docs.html
Unsubscribe info:     https://cygwin.com/ml/#unsubscribe-simple