Lock free ring buffers and SMP safety

Lock free ring buffers and SMP safety

[Matt Bucknall]

Hello,

Yes, this old chestnut. I've spent the best part of today trying to 
ascertain whether or not, in practical terms, a single-producer, 
single-consumer lock free ring-buffer can be written purely in C/C++ and 
be SMP safe on multiple architectures without memory barriers. 
Information I've found has been largely conflicting so I've given up 
trying to get a conclusive answer. Instead I'll just go with the flow.

What is the general consensus amongst ALSA application developers when 
it comes to implementing this kind of buffer? Is it possible to use 
memory barriers through some user space API call that I don't know about?

Thanks in advance for any comments/suggestions.
Matt.

Re: Lock free ring buffers and SMP safety

[Takashi Iwai]

At Sun, 18 Nov 2007 17:31:52 +0000,
Matt Bucknall wrote:
> 
> Hello,
> 
> Yes, this old chestnut. I've spent the best part of today trying to 
> ascertain whether or not, in practical terms, a single-producer, 
> single-consumer lock free ring-buffer can be written purely in C/C++ and 
> be SMP safe on multiple architectures without memory barriers. 
> Information I've found has been largely conflicting so I've given up 
> trying to get a conclusive answer. Instead I'll just go with the flow.
> 
> What is the general consensus amongst ALSA application developers when 
> it comes to implementing this kind of buffer? Is it possible to use 
> memory barriers through some user space API call that I don't know about?

Well, this question will be better answered on LAD, maybe.

AFAIK, the lock-free ringbuffer is possible in most architectures with
a help of memory barriers or atomic ops.  But these arch-specific
things are still headaches if you think of portability, of course.


Takashi