Re: [RFC][PATCH 0/4] Redesign cpu_hotplug locking.

[Nick Piggin]

Andrew Morton wrote:
> On Sun, 27 Aug 2006 12:41:16 +0530
> Dipankar Sarma <dipankar@xxxxxxxxxx> wrote:

> > Is this too difficult for people to follow ?
>
>
> Apparently.  What's happening is that lock_cpu_hotplug() is seen as some
> amazing thing which will prevent an *event* from occurring.

It prevents the event from occurring as much as a lock taken in the
prepare notifier does, right? Or am I misunderstanding something?

> There's an old saying "lock data, not code".  What data is being locked
> here?  It's the subsystem's per-cpu resources which we want to lock.  We
> shouldn't consider the lock as being some way of preventing an event from
> happening.

I agree. Where possible these things should be very simple either with
the per-cpu macros, or using local locking (versus one's notifier).

I think there can be some valid use of the hotplug lock when working
with cpumasks rather than individual CPUs (or if you simply want to
prevent a cpu from going down while programming hardware) and having a
new lock and new notifier is too heavyweight. Or do we want to move
away from the hotplug lock completely?

Hmm, so I don't know if I like the idea of a reentrant rwmutex for the
hotplug lock so it can be sprinkled everywhere... call it a refcount
or not it smells slightly BKLish (looks easy but it could be a
nightmare to audit).

--
SUSE Labs, Novell Inc.
Send instant messages to your online friends http://au.messenger.yahoo.com 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/