kernel-hardening - Re: Re: [RFC v4 PATCH 00/13] HARDENED

Follow @Openwall on Twitter for new release announcements and other news

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20161115180226.GA8828@leverpostej>
Date: Tue, 15 Nov 2016 18:02:26 +0000
From: Mark Rutland <mark.rutland@....com>
To: Peter Zijlstra <peterz@...radead.org>
Cc: kernel-hardening@...ts.openwall.com, Will Deacon <will.deacon@....com>,
	Greg KH <gregkh@...uxfoundation.org>,
	David Windsor <dave@...gbits.org>,
	Elena Reshetova <elena.reshetova@...el.com>,
	Arnd Bergmann <arnd@...db.de>, Thomas Gleixner <tglx@...utronix.de>,
	Ingo Molnar <mingo@...hat.com>,
	"H. Peter Anvin" <h.peter.anvin@...el.com>
Subject: Re: Re: [RFC v4 PATCH 00/13] HARDENED_ATOMIC

On Tue, Nov 15, 2016 at 12:21:49PM +0100, Peter Zijlstra wrote:
> On Tue, Nov 15, 2016 at 10:36:45AM +0000, Mark Rutland wrote:
> > On Fri, Nov 11, 2016 at 09:05:00PM +0100, Peter Zijlstra wrote:
> > > On Fri, Nov 11, 2016 at 06:31:18PM +0000, Mark Rutland wrote:
> > > > On Fri, Nov 11, 2016 at 09:43:00AM -0800, Kees Cook wrote:
> > > > > (And now Greg went missing from the reply? Re-added...)
> > > > > 
> > > > > On Thu, Nov 10, 2016 at 11:50 PM, David Windsor <dave@...gbits.org> wrote:
> > > > > > On Thu, Nov 10, 2016 at 6:38 PM, Greg KH <gregkh@...uxfoundation.org> wrote:
> > > > > >> On Thu, Nov 10, 2016 at 03:15:44PM -0800, Kees Cook wrote:
> > > > 
> > > > > As far as refcount_t is concerned, I worry using cmpxchg will be too
> > > > > costly, but it's worth benchmarking.
> > > > 
> > > > If that does turn out to be a problem, we could allow architectures to
> > > > provide their own implementations of the API, with a generic fallback
> > > > otherwise, as we do for other features.
> > > 
> > > Note that only LL/SC archs can do somewhat better. x86/s390/sparc64 etc
> > > al must use cmpxchg, there's just no other way to get an actual atomic
> > > inc/dec with over/under-flow detection.
> > 
> > Sure; arm64 will also fall into that bucket with the ARMv8.1 atomics.
> 
> arm64v8.1 still has ll/sc surely ;-)

Yes, they survive. ;)

I'd simply assumed we'd use the ARMv8.1 atomics for refcount_*() since
we use those for the whole atomic_*() suite (when available). 

> Or are you saying a cmpxchg loop is better for arm64v8.1 than an ll/sc
> ?

I'm not sure either way; that'll have to fall to benchmarking, I guess.

Thanks,
Mark.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.