Re: [boost] [lockfree] review

Subject: Re: [boost] [lockfree] review
From: Tim Blechmann (tim_at_[hidden])
Date: 2011-08-01 08:50:06

[quick post, have some more in my outbox, but i'm at a conference this week,
where they cut imap/smtp ports]

>> Anyway, what's the "wrong order?"  If one thread is depending on the
>> order of things that happen in another thread without both threads using
>> the synchronization necessary to ensure that order, it's a race
>> condition (a.k.a. a bug), right?
>
> All of the classic lock-free algorithms, including those that are  
> the basis for the synchronization primitives we should usually use,  
> have data races.  At a low level it is sometimes necessary to use a  
> variable for synchronization.

the fifo code for example contains code like:

atomic<> head_, tail_;

head = head_.load();
tail = tail_.load();

head2 = head_.load();

if (head == head2)
    do something

the correctness of this algorithm depends on the order of the loads: load head
before tail and load tail before head2. without a memory model, both  
the compiler
and the CPU are allowed to reorder loads (or head2 could simply use  
the value of
head without actually reloading it)

cheers, tim

---------------------------------------------------------------
diese nachricht wurde ueber klingt.org gesendet
this message was sent through klingt.org's webmail.