Experimental Physics and
Industrial Control System

Michael Davidsaver <[email protected]> · Mon, 13 May 2013 10:57:25 -0400

On 05/11/2013 11:30 PM, Andrew Johnson wrote:

I've just been reading an interesting LWN article about a mutex design that allows random locking order of multiple mutexes without danger of deadlocks. This might be a useful approach for the multi-channel additions.

http://lwn.net/Articles/548909/

More details at the link in the first Conclusions paragraph.

This is similar to the first algorithm I came up with. The part I was

missing was the idea of "reservation ids" to avoid live lock. For epics

lock sets this would need to be extended to include some idea of

priority as well.

Because this didn't occur to me I came up with the idea of maintaining

the locks in a global order. This does guarantee forward progress, but

is more complex then the scheme described in the LWN article. global

ordering does avoids the need to make explicit decisions based on thread

priority though.

Definitely food for thought :)

Experimental Physics and Industrial Control System

Experimental Physics and
Industrial Control System