Re: [PATCH] bpf: Tweak BPF jump table optimizations for objtool compatibility

From: Alexei Starovoitov
Date: Wed May 06 2020 - 12:45:18 EST


On Wed, May 6, 2020 at 8:53 AM Josh Poimboeuf <jpoimboe@xxxxxxxxxx> wrote:
>
> On Tue, May 05, 2020 at 04:59:39PM -0700, Alexei Starovoitov wrote:
> > As far as workaround I prefer the following:
> > From 94bbc27c5a70d78846a5cb675df4cf8732883564 Mon Sep 17 00:00:00 2001
> > From: Alexei Starovoitov <ast@xxxxxxxxxx>
> > Date: Tue, 5 May 2020 16:52:41 -0700
> > Subject: [PATCH] bpf,objtool: tweak interpreter compilation flags to help objtool
> >
> > tbd
> >
> > Fixes: 3193c0836f20 ("bpf: Disable GCC -fgcse optimization for ___bpf_prog_run()")
> > Reported-by: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
> > Reported-by: Arnd Bergmann <arnd@xxxxxxxx>
> > Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx>
> > ---
> > include/linux/compiler-gcc.h | 2 +-
> > 1 file changed, 1 insertion(+), 1 deletion(-)
> >
> > diff --git a/include/linux/compiler-gcc.h b/include/linux/compiler-gcc.h
> > index d7ee4c6bad48..05104c3cc033 100644
> > --- a/include/linux/compiler-gcc.h
> > +++ b/include/linux/compiler-gcc.h
> > @@ -171,4 +171,4 @@
> > #define __diag_GCC_8(s)
> > #endif
> >
> > -#define __no_fgcse __attribute__((optimize("-fno-gcse")))
> > +#define __no_fgcse __attribute__((optimize("-fno-gcse,-fno-omit-frame-pointer")))
> > --
> > 2.23.0
> >
> > I've tested it with gcc 8,9,10 and clang 11 with FP=y and with ORC=y.
> > All works.
> > I think it's safer to go with frame pointers even for ORC=y considering
> > all the pain this issue had caused. Even if objtool gets confused again
> > in the future __bpf_prog_run() will have frame pointers and kernel stack
> > unwinding can fall back from ORC to FP for that frame.
> > wdyt?
>
> It seems dangerous to me. The GCC manual recommends against it.

The manual can says that it's broken. That won't stop the world from using it.
Just google projects that are using it. For example: qt, lz4, unreal engine, etc
Telling compiler to disable gcse via flag is a guaranteed way to avoid
that optimization that breaks objtool whereas messing with C code is nothing
but guess work. gcc can still do gcse.