Re: [PATCH v2] tools/memory-model: Add extra ordering for locks and remove it for ordinary release/acquire

From: Daniel Lustig <dlustig@nvidia.com>
To: Will Deacon <will.deacon@arm.com>, Peter Zijlstra <peterz@infradead.org>
Cc: Andrea Parri <andrea.parri@amarulasolutions.com>,
	Alan Stern <stern@rowland.harvard.edu>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	LKMM Maintainers -- Akira Yokosawa <akiyks@gmail.com>,
	Boqun Feng <boqun.feng@gmail.com>,
	David Howells <dhowells@redhat.com>,
	Jade Alglave <j.alglave@ucl.ac.uk>,
	Luc Maranget <luc.maranget@inria.fr>,
	Nicholas Piggin <npiggin@gmail.com>,
	Kernel development list <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v2] tools/memory-model: Add extra ordering for locks and remove it for ordinary release/acquire
Date: Thu, 12 Jul 2018 19:17:46 -0700	[thread overview]
Message-ID: <96aa1571-d9b9-9554-7017-d173e1a795e9@nvidia.com> (raw)
In-Reply-To: <20180712094510.GA23415@arm.com>

On 7/12/2018 2:45 AM, Will Deacon wrote:
> On Thu, Jul 12, 2018 at 11:34:32AM +0200, Peter Zijlstra wrote:
>> On Thu, Jul 12, 2018 at 09:40:40AM +0200, Peter Zijlstra wrote:
>>> And I think if we raise atomic*_acquire() to require TSO (but ideally
>>> raise it to RCsc) we're there.
>>
>> To clarify, just the RmW-acquire. Things like atomic_read_acquire() can
>> stay smp_load_acquire() and be RCpc.
> 
> I don't have strong opinions about strengthening RmW atomics to TSO, so
> if it helps to unblock Alan's patch (which doesn't go near this!) then I'll
> go with it. The important part is that we continue to allow roach motel
> into the RmW for other accesses in the non-fully-ordered cases.
> 
> Daniel -- your AMO instructions are cool with this, right? It's just the
> fence-based implementations that will need help?
> 
> Will
Right, let me pull this part out of the overly-long response I just gave
on the thread with Linus :)

if we pair AMOs with AMOs, we get RCsc, and everything is fine.  If we
start mixing in fences (mostly because we don't currently have native
load-acquire or store-release opcodes), then that's when all the rest of the
complexity comes in.

Dan