Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Mon, 6 Apr 2020 13:04:38 +0200
From:   Peter Zijlstra <peterz@infradead.org>
To:     Fangrui Song <maskray@google.com>
Cc:     x86@kernel.org, linux-kernel@vger.kernel.org, rostedt@goodmis.org,
        mhiramat@kernel.org, bristot@redhat.com, jbaron@akamai.com,
        torvalds@linux-foundation.org, tglx@linutronix.de,
        mingo@kernel.org, namit@vmware.com, hpa@zytor.com, luto@kernel.org,
        ard.biesheuvel@linaro.org, jpoimboe@redhat.com,
        clang-built-linux@googlegroups.com, hjl.tools@gmail.com
Subject: Re: [RESEND][PATCH v3 09/17] x86/static_call: Add out-of-line static
 call implementation
Message-ID: <20200406110438.GJ20730@hirez.programming.kicks-ass.net>
References: <20200324135603.483964896@infradead.org>
 <20200324142245.819003994@infradead.org>
 <20200406010859.bcfouhukcgmg2on7@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200406010859.bcfouhukcgmg2on7@google.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sun, Apr 05, 2020 at 06:08:59PM -0700, Fangrui Song wrote:
> On 2020-03-24, Peter Zijlstra wrote:

> > +#define ARCH_DEFINE_STATIC_CALL_TRAMP(name, func)			\
> > +	asm(".pushsection .text, \"ax\"				\n"	\
> > +	    ".align 4						\n"	\
> > +	    ".globl " STATIC_CALL_TRAMP_STR(name) "		\n"	\
> > +	    STATIC_CALL_TRAMP_STR(name) ":			\n"	\
> > +	    "	jmp.d32 " #func "				\n"	\
> > +	    ".type " STATIC_CALL_TRAMP_STR(name) ", @function	\n"	\
> > +	    ".size " STATIC_CALL_TRAMP_STR(name) ", . - " STATIC_CALL_TRAMP_STR(name) " \n" \
> > +	    ".popsection					\n")
> > +
> > +#endif /* _ASM_STATIC_CALL_H */
> 
> Hi Peter,
> 
> Coming here from https://github.com/ClangBuiltLinux/linux/issues/974
> 
> jmp.d32 is not recognized by clang integrated assembler.
> The syntax appears to be very rarely used. According to Debian Code Search,
> u-boot is the only project using this syntax.

*groan*... I was going to use it in more places :-/

> In March 2017, gas added the pseudo prefix {disp32}
> https://sourceware.org/git/?p=binutils-gdb.git;a=commit;h=86fa6981e7487e2c2df4337aa75ed2d93c32eaf2
> which generalizes jmp.d32  ({disp32} jmp foo)

That's all well and cute, but I can't use that because its too new. Not
until we raise the minimum gcc/bintils version to something that
includes that.

> I wonder whether the instruction jmp.d32 can be replaced with the trick in
> arch/x86/include/asm/jump_label.h for clang portability.
> 
> % grep -A2 'jmp.d32' arch/x86/include/asm/jump_label.h
>         /* Equivalent to "jmp.d32 \target" */
>         .byte           0xe9
>         .long           \target - .Lstatic_jump_after_\@

Sure, but I was hoping to move away from that since all assemblers
should now support jmp.d32. Except of course, you have to go ruin
things.

The thing is, jmp.d32 reads so much nicer than the above crap.

Also, I still need a meta instruction like:

	nopjmp $label

what works just like 'jmp' but instead emits either a nop2 or a nop5.
I tried various hacks to get GAS to emit that, but no luck :/

The only up-side from that new syntax is that I suppose we can go write:

  {disp8} push \vec

without gas shitting itself and emitting a 5 byte push just because..
Except of course we can't, for the same reason I can't go around and
use:

  {disp32} jmp

in the above code.