Re: [RFC PATCH v9 for 4.15 01/14] Restartable sequences system call

From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
To: Ben Maurer <bmaurer@fb.com>
Cc: Andi Kleen <andi@firstfloor.org>, carlos <carlos@redhat.com>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	"Paul E. McKenney" <paulmck@linux.vnet.ibm.com>,
	David Goldblatt <davidgoldblatt@fb.com>, Qi Wang <qiwang@fb.com>,
	Boqun Feng <boqun.feng@gmail.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Paul Turner <pjt@google.com>, Andrew Hunter <ahh@google.com>,
	Andy Lutomirski <luto@amacapital.net>,
	Dave Watson <davejwatson@fb.com>,
	Josh Triplett <josh@joshtriplett.org>,
	Will Deacon <will.deacon@arm.com>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Chris Lameter <cl@linux.com>, Ingo Molnar <mingo@redhat.com>,
	"H. Peter Anvin" <hpa@zytor.com>, rostedt <rostedt@goodmis.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Russell King <linux@arm.linux.org.uk>,
	Catalin Marinas <catalin.marinas@arm.com>,
	Michael Kerrisk <mtk.manpages@gmail.com>,
	Alexander Viro <viro@zeniv.linux.org.uk>,
	linux-api <linux-api@vger.kernel.org>
Subject: Re: [RFC PATCH v9 for 4.15 01/14] Restartable sequences system call
Date: Tue, 17 Oct 2017 17:48:04 +0000 (UTC)	[thread overview]
Message-ID: <1630038054.43152.1508262484464.JavaMail.zimbra@efficios.com> (raw)
In-Reply-To: <CY4PR15MB16886FD43FB48592F3F5892FCF4C0@CY4PR15MB1688.namprd15.prod.outlook.com>

----- On Oct 17, 2017, at 12:41 PM, Ben Maurer bmaurer@fb.com wrote:

>> I have a use-case for keeping the reference counting in place though. It's
>> use of rseq in signal handlers.
> 
> Would this be solved by saying that the rseq api will return an error if you
> register and there's already a block registered. In this case the signal
> handler would register the rseq abi area just as the non-signal code is trying
> to do the same. The non-signal code would see this error code and realize that
> its job had been done for it and then go on it's way.

Yes, that should work, as long as we return a specific error code, e.g. -EBUSY,
to tell the caller that rseq has actually been registered.

> 
> It would be unsafe for signal handler code to *unregister* the area, but I don't
> think that's necessary.

Right.

> 
> Basically we'd support a refcount of either 0 or 1, but nothing else.

Yep, I'll try this out.

> 
> If a signal handler registers the ABI area, how will it ensure the ABI is
> cleaned up at thread exit? I can't imagine pthread_key is signal safe.

You have a very good point there. This highlights a signal-safety issue
I have in liburcu-bp when used by lttng-ust. pthread_setspecific is
indeed not listed as being signal-safe: it can perform on-demand memory
allocation when a second level array is needed.

I'll have to scratch my head a bit to fix this one.

Thanks!

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com