Re: [RFC PATCH 4/4] x86/asm: Use ASM_CALL() macro for inline asm statements with call instructions

From: Dmitry Vyukov <dvyukov@google.com>
To: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>,
	"x86@kernel.org" <x86@kernel.org>,
	LKML <linux-kernel@vger.kernel.org>,
	Ingo Molnar <mingo@kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	"H. Peter Anvin" <hpa@zytor.com>,
	Andy Lutomirski <luto@kernel.org>,
	Linus Torvalds <torvalds@linux-foundation.org>,
	Alexander Potapenko <glider@google.com>,
	Matthias Kaehlcke <mka@chromium.org>,
	Arnd Bergmann <arnd@arndb.de>
Subject: Re: [RFC PATCH 4/4] x86/asm: Use ASM_CALL() macro for inline asm statements with call instructions
Date: Thu, 31 Aug 2017 17:36:45 +0200	[thread overview]
Message-ID: <CACT4Y+YgoyqnhrKpyfW4HKv_Y1KNUL-yPJCd-s52zCArSep-Lg@mail.gmail.com> (raw)
In-Reply-To: <20170831152153.dre6gjdqcucmi5oc@treble>

On Thu, Aug 31, 2017 at 5:21 PM, Josh Poimboeuf <jpoimboe@redhat.com> wrote:
> On Thu, Aug 31, 2017 at 04:50:41PM +0200, Peter Zijlstra wrote:
>> On Thu, Aug 31, 2017 at 09:11:20AM -0500, Josh Poimboeuf wrote:
>> > Inline asm statements which have call instructions can be problematic.
>> > GCC doesn't know about the call instructions, so in some cases it can
>> > insert the asm before setting up the frame pointer.  This can result in
>> > bad stack traces when unwinding from the called function.
>> >
>> > Previously we worked around this issue by listing the stack pointer as
>> > an input/output constraint for the inline asm.  That works for GCC, but
>> > unfortunately it doesn't work for Clang.  In fact, it causes Clang to
>> > corrupt the stack pointer.
>>
>> Sounds like it ought to get fixed regardless and then it might as well
>> do the right thing ;-)
>
> There was some disagreement about what the "right thing" is because it's
> an undocumented and unintuitive interface.
>
> And I use the term "interface" loosely.  It was apparently a side effect
> which was mentioned to me on the GCC mailing list.

Yes, as far as I understand, there is just no defined semantics for
this. Passing sp as is when asm block asks to pass in sp looks like a
perfectly reasonable thing to do (also faster code). We could use
something like asm("..." ::: "frame"), but we don't have this in
compilers.