Re: [PATCH v3] x86/entry: emit a symbol for register restoring thunk - Fāng-ruì Sòng

From: "Fāng-ruì Sòng" <maskray@google.com>
To: Borislav Petkov <bp@alien8.de>
Cc: Nick Desaulniers <ndesaulniers@google.com>,
	Andy Lutomirski <luto@kernel.org>,
	Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Arnd Bergmann <arnd@arndb.de>,
	Josh Poimboeuf <jpoimboe@redhat.com>, X86 ML <x86@kernel.org>,
	"H. Peter Anvin" <hpa@zytor.com>,
	Nathan Chancellor <natechancellor@gmail.com>,
	LKML <linux-kernel@vger.kernel.org>,
	clang-built-linux <clang-built-linux@googlegroups.com>
Subject: Re: [PATCH v3] x86/entry: emit a symbol for register restoring thunk
Date: Mon, 11 Jan 2021 16:41:52 -0800	[thread overview]
Message-ID: <CAFP8O3+uEE4Lity-asyFLN6_+8qRUD3hgcZVapXwk6EfmGM+DA@mail.gmail.com> (raw)
In-Reply-To: <20210112003839.GL25645@zn.tnic>

On Mon, Jan 11, 2021 at 4:38 PM Borislav Petkov <bp@alien8.de> wrote:
>
> On Mon, Jan 11, 2021 at 12:38:06PM -0800, Nick Desaulniers wrote:
> > Arnd found a randconfig that produces the warning:
> >
> > arch/x86/entry/thunk_64.o: warning: objtool: missing symbol for insn at
> > offset 0x3e
> >
> > when building with LLVM_IAS=1 (use Clang's integrated assembler). Josh
> > notes:
> >
> >   With the LLVM assembler stripping the .text section symbol, objtool
> >   has no way to reference this code when it generates ORC unwinder
> >   entries, because this code is outside of any ELF function.
> >
> > Fangrui notes that this optimization is helpful for reducing images size
> > when compiling with -ffunction-sections and -fdata-sections. I have
> > observerd on the order of tens of thousands of symbols for the kernel
> > images built with those flags. A patch has been authored against GNU
> > binutils to match this behavior, with a new flag
> > --generate-unused-section-symbols=[yes|no].
> >
> > We can omit the .L prefix on a label to emit an entry into the symbol
> > table for the label, with STB_LOCAL binding.  This enables objtool to
> > generate proper unwind info here with LLVM_IAS=1.
> >
> > Cc: Fangrui Song <maskray@google.com>
> > Link: https://github.com/ClangBuiltLinux/linux/issues/1209
> > Link: https://reviews.llvm.org/D93783
> > Link: https://sourceware.org/binutils/docs/as/Symbol-Names.html
> > Link: https://sourceware.org/pipermail/binutils/2020-December/114671.html
> > Reported-by: Arnd Bergmann <arnd@arndb.de>
> > Suggested-by: Josh Poimboeuf <jpoimboe@redhat.com>
> > Signed-off-by: Nick Desaulniers <ndesaulniers@google.com>
> > ---
> > Changes v2 -> v3:
> > * rework to use STB_LOCAL rather than STB_GLOBAL by dropping .L prefix,
> >   as per Josh.
>
> Ok so I read a bit around those links above...
>
> Are you trying to tell me here that we can't use .L-prefixed local
> labels anymore because, well, clang's assembler is way too overzealous
> when stripping symbols to save whopping KiBs of memory?!

To be fair: we cannot use .L-prefixed local because of the objtool limitation.
The LLVM integrated assembler behavior is a good one and binutils
global maintainers have agreed so H.J. went ahead and implemented it
for GNU as x86.

> Btw Josh made sense to me when asking for a flag or so to keep .text.
>
> And I see --generate-unused-section-symbols=[yes|no] for binutils.
>
> So why isn't there a patch using that switch on clang too instead of the
> kernel having to dance yet again for some tool?

--generate-unused-section-symbols=[yes|no] as an assembler option has
been rejected.

> :-\
>
> --
> Regards/Gruss,
>     Boris.
>
> https://people.kernel.org/tglx/notes-about-netiquette