Re: [RFC 07/13] sched: Reduce stack size requirements in kernel/sched.c

[Mike Travis <travis@sgi.com>]

Peter Zijlstra wrote:
> On Sun, 2008-09-07 at 04:00 -0700, Andrew Morton wrote:
>> On Sun, 07 Sep 2008 12:24:47 +0200 Peter Zijlstra <peterz@infradead.org> wrote:
>>
>>> get_online_cpus() can sleep, but you just disabled preemption with those
>>> get_cpumask_var() horribles!
>> make cpu_hotplug.refcount an atomic_t.
> 
> A much easier fix is just re-ordering those operations and do the
> get_online_cpus() before disabling preemption. But it does indicate this
> patch series isn't carefully constructed.

Yes, it's mostly a hunt for comments on my part... ;-)
> 
>>> Couldn't be arsed to look through the rest, but I really hate this
>>> cpumask_ptr() stuff that relies on disabling preemption.
>> that's harder to fix ;)
> 
> Looking at more patches than just the sched one convinced me more that
> this approach isn't a good one. It seems to make code much more
> fragile. 
> 
> See patch 9, there it was needed to draw out the callgraph in order to
> map stuff to these global variables - we're adding global dependencies
> to code that didn't have any, increasing complexity.

Again, yes, as I got farther into that one, it became clear that having
static cpumask_t temps over too large a range was ending up very messy.

Thanks,
Mike