Re: [RFC] New locking primitive for 2.5

[Mark Frazer <mark@somanetworks.com>]

Christoph Hellwig <hch@ns.caldera.de> [02/02/07 14:41]:
> What about the following instead:
> 
> 	combi_lock(struct combilock *x, int spin);
> 	combi_unlock(struct combilock *x);

how about
        combi_lock (struct combilock *x, int canblock, int shouldblock);

Where the should block flag is copied to the mutex once it's held by
the caller to indicate to new threads grabbing the lock how long the
lock will be held for.

For locks that are held for some short duration tasks and some long
duration tasks, the holder should indicate how long the lock will be held.
I'd consider that an improvement over the "spin for a while then block"
idea.

Interrupt handlers can't block, so we need a flag to never block.

-mark