atomicity

[taman]

When I want to execute a little code snippet absolute atomically on it's current CPU, which instructions do I have to call before? Do I need local_irq_disable() *and* preempt_disable() or is one of them sufficient? I.e. is one a subset of the other or where is the difference between these calls?

[bloggins666]

You will find some API functions called: atomic_read(), atomic_set(), atomic_add() and so forth. You may find that these functions will cover your "code snippet" and do the job.

As a developer's note, I usually have the source code of the kernel all indexed through some IDE (in my case cscope, I guess I'm showing my age.) available to me so that I can check out the code of the API that I'm using. It comes in handy for questions like what you have asked. When I looked at "preempt_disable()" I got the following:

#define preempt_disable() \
do { \
inc_preempt_count(); \
barrier(); \
} while (0)

When I searched for "local_irq_disable()", they did not seem to be related at all.

Hope this helps.

Cheers!!

[taman]

Hmm, atomic_read/set/add are just usable to make very small pieces of code atomically - like reading, setting or adding variables. But what I have to do is to make a code snippet of say 10 or 20 lines atomically.

Thanks for the hint with cscope. Usually I use online source browsers like lxr for that. Anyhow, I looked at the functions, but I still wonder, if I need one or both of them. To be save I am just using both of them in the moment