Return to Implementing Threads

From: Evan Jones <ejones@uwaterloo.ca>
Date: Mon Oct 21, 2002  17:56:02 Europe/Zurich
To: [...]
Subject: Re: About locking in multi-threaded environments

On Sunday, Oct 20, 2002, at 10:18 Europe/Zurich, [...] wrote:
> I've just come along with
> "Implementing a Thread Library on Linux",
> and I want you to reconsider your text about no need to use
> locking when using a user level thread model/library on
> a single processor. Although you alert of the consecuences of
> such 'technique', I am strongly against such idea because it is
> simply bad style, error-prone programming.

This is very true: Threading is a complicated and frequently error 
prone programming task. From a strictly "academic" discussion about 
threading, what I have said is technically true: There is no REAL 
concurrency, and there is also no REAL pre-emption.

This means that locking is not necessary to avoid conflicts when using 
shared resources, because as a programmer you have explicit control 
over when threads execute and when control switches between threads. 
However, you are absolutely correct: If you do not take care when 
adding thread_yield() calls, you can write software that will 
malfunction. However, if you are not careful with calls to lock and 
unlock the semaphores or whatever locking primitive you are using, 
similar problems can arise.


Still, I will add a note to that paragraph of my document to attempt to 
make my statements more clear.

Let me know if you have any further questions or comments,

Evan