Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Wed, 15 Jul 2020 10:20:16 -0400 (EDT)
From:   Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
To:     Christian Brauner <christian.brauner@ubuntu.com>
Cc:     Florian Weimer <fweimer@redhat.com>, carlos <carlos@redhat.com>,
        Peter Zijlstra <peterz@infradead.org>,
        linux-kernel <linux-kernel@vger.kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        paulmck <paulmck@linux.ibm.com>,
        Boqun Feng <boqun.feng@gmail.com>,
        "H. Peter Anvin" <hpa@zytor.com>, Paul Turner <pjt@google.com>,
        linux-api <linux-api@vger.kernel.org>
Message-ID: <1765484769.14232.1594822816264.JavaMail.zimbra@efficios.com>
In-Reply-To: <20200715135554.4q6aamx4iqd4mnqh@wittgenstein>
References: <20200714030348.6214-1-mathieu.desnoyers@efficios.com> <2452161.11491.1594732791558.JavaMail.zimbra@efficios.com> <71f08b3a-56f5-0e0f-53b0-cc680f7e8181@redhat.com> <2053637148.14136.1594818777608.JavaMail.zimbra@efficios.com> <87y2nk29rp.fsf@oldenburg2.str.redhat.com> <882700738.14181.1594819884049.JavaMail.zimbra@efficios.com> <87mu4028uk.fsf@oldenburg2.str.redhat.com> <20200715135554.4q6aamx4iqd4mnqh@wittgenstein>
Subject: Re: [RFC PATCH 2/4] rseq: Allow extending struct rseq
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
Thread-Topic: rseq: Allow extending struct rseq
Thread-Index: dcnj3oHgdUUwDl0MIo+e+CqqnEuWyQ==
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

----- On Jul 15, 2020, at 9:55 AM, Christian Brauner christian.brauner@ubuntu.com wrote:

> On Wed, Jul 15, 2020 at 03:42:11PM +0200, Florian Weimer wrote:
>> * Mathieu Desnoyers:
>> 
>> > So indeed it could be done today without upgrading the toolchains by
>> > writing custom assembler for each architecture to get the thread's
>> > struct rseq. AFAIU the ABI to access the thread pointer is fixed for
>> > each architecture, right ?
>> 
>> Yes, determining the thread pointer and access initial-exec TLS
>> variables is baked into the ABI.
>> 
>> > How would this allow early-rseq-adopter libraries to interact with
>> > glibc ?
>> 
>> Under all extension proposals I've seen so far, early adopters are
>> essentially incompatible with glibc rseq registration.  I don't think
>> you can have it both ways.
> 
> Who are the early adopters? And if we aren't being compatible with them
> under the extensible schemes proposed we should be able to achieve
> compatibility with non-early adopters, right? Which I guess is more
> important. (I still struggle to make sense what qualifies as an early
> adopter/what the difference to a non-early adopter is.)

Early adopter libraries and applications are meant to be able to use rseq
without requiring upgrade of the entire environment to a newer glibc.

I maintain early adopter projects (liburcu, lttng-ust) which postpone using
rseq outside of prototype branches until we agree on an ABI to share
__rseq_abi between glibc and early adopter libraries. The last thing I
want is for those projects to break when an end-user upgrades their
glibc. tcmalloc is another early adopter which have less strict
compatibility requirements: they are OK with breaking changes requiring
upgrading and rebuilding tcmalloc.

Indeed, until we cast in stone the layout of struct rseq as exposed by
glibc, I think we have some freedom in our definition of "early adopter",
because pretty much every relevant open source project which want to use
rseq is waiting on glibc to define that ABI, to use rseq either as an
early-adopter or through a dependency on newer glibc.

Thanks,

Mathieu


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