Re: [patch] cpufreq: mark cpufreq_tsc() as core_initcall_sync

[Alan Stern]

On Thu, 16 Nov 2006, Linus Torvalds wrote:

> 
> 
> On Thu, 16 Nov 2006, Alan Stern wrote:
> > On Thu, 16 Nov 2006, Linus Torvalds wrote:
> > >
> > > Paul, it would be _really_ nice to have some way to just initialize 
> > > that SRCU thing statically. This kind of crud is just crazy.
> > 
> > I looked into this back when SRCU was first added.  It's essentially 
> > impossible to do it, because the per-cpu memory allocation & usage APIs 
> > are completely different for the static and the dynamic cases.
> 
> I don't think that's how you'd want to do it.
> 
> There's no way to do an initialization of a percpu allocation statically. 
> That's pretty obvious.

Hmmm...  What about DEFINE_PER_CPU in include/asm-generic/percpu.h 
combined with setup_per_cpu_areas() in init/main.c?  So long as you want 
all the CPUs to start with the same initial values, it should work.

> What I'd suggest instead, is to make the allocation dynamic, and make it 
> inside the srcu functions (kind of like I did now, but I did it at a 
> higher level).
> 
> Doing it at the high level was trivial right now, but we may well end up 
> hitting this problem again if people start using SRCU more. Right now I 
> suspect the cpufreq notifier is the only thing that uses SRCU, and it 
> already showed this problem with SRCU initializers.
> 
> So I was more thinking about moving my "one special case high level hack" 
> down lower, down to the SRCU level, so that we'll never see _more_ of 
> those horrible hacks. We'll still have the hacky thing, but at least it 
> will be limited to a single place - the SRCU code itself.

Another possible approach (but equally disgusting) is to use this static 
allocation approach, and have the SRCU structure include both a static and 
a dynamic percpu pointer together with a flag indicating which should be 
used.

Alan Stern

-
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/