Re: [RFC PATCH v2 1/2] rseq: x86: implement abort-at-ip extension

From: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
To: Christian Brauner <christian.brauner@ubuntu.com>
Cc: Christian Brauner <brauner@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	linux-kernel <linux-kernel@vger.kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	paulmck <paulmck@kernel.org>, Boqun Feng <boqun.feng@gmail.com>,
	"H. Peter Anvin" <hpa@zytor.com>, Paul Turner <pjt@google.com>,
	linux-api <linux-api@vger.kernel.org>,
	Florian Weimer <fw@deneb.enyo.de>,
	David Laight <David.Laight@ACULAB.COM>,
	carlos <carlos@redhat.com>
Subject: Re: [RFC PATCH v2 1/2] rseq: x86: implement abort-at-ip extension
Date: Wed, 12 Jan 2022 09:47:29 -0500 (EST)	[thread overview]
Message-ID: <1475639366.24565.1641998849957.JavaMail.zimbra@efficios.com> (raw)
In-Reply-To: <20220112084617.32bjjo774n7vvyct@wittgenstein>

----- On Jan 12, 2022, at 3:46 AM, Christian Brauner christian.brauner@ubuntu.com wrote:

> On Tue, Jan 11, 2022 at 12:43:05PM -0500, Mathieu Desnoyers wrote:
[...]
>> >> + *
>> >> + * rseq critical sections defined with the RSEQ_CS_FLAG_ABORT_AT_IP flag
>> >> + * have the following behavior on abort: when the stack grows down: the
>> >> + * stack pointer is decremented to skip the redzone, and decremented of
>> >> + * the pointer size.  The aborted address (abort-at-ip) is stored at
>> >> + * this stack pointer location.  The user-space abort handler needs to
>> >> + * pop the abort-at-ip address from the stack, and add the redzone size
>> >> + * to the stack pointer.
>> >> + *
>> >> + * TODO: describe stack grows up.
>> > 
>> > Is this intentional or did you forget? :)
>> 
>> Since I did not implement abort-at-ip on stack-grows-up architectures, I felt
>> it would be too early to describe the algorithm. I can simply remove the TODO
>> altogether and we'll take care of it when we get there ? If I had to try to
>> wordsmith it, it would look like e.g.:
>> 
>>  *                                    [...] When the stack grows up: the
>>  * stack pointer is incremented to skip the redzone, and incremented of
>>  * the pointer size.  The aborted address (abort-at-ip) is stored immediately
>>  * under this stack pointer location.  The user-space abort handler needs to
>>  * pop the abort-at-ip address from the stack, and subtract the redzone size
>>  * from the stack pointer.
>> 
>> [ Please let me know if I got somehow confused in my understanding of stack
>> grows
>> up architectures. ]
>> 
>> I'm also unsure whether any of the stack grows up architecture have redzones ?
> 
> I don't think so? From when I last touched that piece of arch code when
> massaging copy_thread() I only remember parisc as having an upwards
> growing stack.
> 
>> From a quick grep for redzone in Linux arch/, only openrisc, powerpc64 and
> > x86-64 appear to have redzones.

I figured it was kind of silly to special-case arch-agnostic comments for stack
grows up/down, how about the following instead ?

 * rseq critical sections defined with the RSEQ_CS_FLAG_ABORT_AT_IP flag
 * have the following behavior on abort: the stack pointer is adjusted to
 * skip over the redzone [*], and the aborted address (abort-at-ip) is pushed
 * at this stack pointer location.  The user-space abort handler needs to pop
 * the abort-at-ip address from the stack, and adjust the stack pointer to skip
 * back over the redzone.
 *
 * [*] The openrisc, powerpc64 and x86-64 architectures define a "redzone" as a
 *     stack area beyond the stack pointer which can be used by the compiler
 *     to store local variables in leaf functions.

Thanks,

Mathieu

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