Re: [gcv v3 06/35] scheduler: Replace __get_cpu_var uses

[Steven Rostedt]

On Thu, 29 Aug 2013 18:15:43 +0000
Christoph Lameter <cl@xxxxxxxxx> wrote:

> On Thu, 29 Aug 2013, Steven Rostedt wrote:
> 
> Its not really an atomic operation in the classic sense.

It doesn't need to be atomic, it could mean it is used within atomic
locations. Basically, "can't be interrupted here". I just said
"something like", it didn't even need to be that.

> 
> this_cpu_no_preempt_check_read ?

I would make it much shorter. You could use "raw_this_cpu_read()",
which usually means "no checks here". Or, "this_cpu_read_nopreempt()".

> 
> The problem that I have is also that a kernel with preemption is not
> something that see anywhere these days. Looks more like an academic
> exercise? Does this really matter? All the distro I see use

Um, my paycheck depends on PREEMPT_RT working. And there's a lot of
interest in real PREEMPT by audio folks. It's no more an
academic exercise than people wanting really low kernel latency.

> PREEMPT_VOLUNTARY. Performance degradation is significant if massive
> amounts of checks and preempt disable/enable points are added to the
> kernel.

They are usually disabled for production systems. But we run a bunch of
tests with the debug checks enabled, which catch bugs before we ship a
kernel for a production system.

> 
> Do we agree that it is necessary and useful to add another variant of
> this_cpu ops for this? The concern of having too many variants is no
> longer there? Adding another variant is not that difficult just code
> intensive.

How many places use the this_cpu_*() without preemption disabled? I
wouldn't think there's many. I never complained about another variant,
so you need to ask those that have. The tough question for me is what
that variant name should be ;-)

-- Steve
--
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/