Re: [RFC][PATCH v3] slub: Keep page and object in sync in slab_alloc_node()

[Steven Rostedt <rostedt@goodmis.org>]

On Fri, 2013-01-18 at 11:09 -0800, Eric Dumazet wrote:
> On Fri, 2013-01-18 at 18:40 +0000, Christoph Lameter wrote:
> 
> > The fetching of the tid is the only critical thing here. If the tid is
> > retrieved from the right cpu then the cmpxchg will fail if any changes
> > occured to freelist or the page variable.
> > 
> > The tid can be retrieved without disabling preemption through
> > this_cpu_read().
> 
> Strictly speaking, this_cpu_read() _does_ disable preemption.

I was thinking the same thing.

> 
> Of course, on x86, this_cpu_read() uses __this_cpu_read()


Looking at using this_cpu_read() on tid, I can't see what would break.
We still need the check for page being NULL in node_match(). I'm still
uneasy about it. Just because we can't see how it might break doesn't
mean that it wont. As we have already found a bit of bugs in the current
code. I'd feel more comfortable with the explicit preempt_disable(). And
this is coming from an -rt guy that tries to avoid preempt_disable().
But you're (Christoph) the maintainer.

I guess if you use this_cpu_read() you don't need that barrier anymore.

-- Steve


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>