On Fri, Aug 19, 2022 at 10:42 AM Masahiro Yamada <masahiroy@xxxxxxxxxx> wrote:
>
> On Sat, Aug 20, 2022 at 2:28 AM Nick Desaulniers
> <ndesaulniers@xxxxxxxxxx> wrote:
> >
> > On Fri, Aug 19, 2022 at 10:16 AM Alexandre Belloni
> > <alexandre.belloni@xxxxxxxxxxx> wrote:
> > >
> > > On 19/08/2022 10:00:53-0700, Nick Desaulniers wrote:
> > > > GCC has supported asm goto since 4.5, and Clang has since version 9.0.0.
> > > > The minimum supported versions of these tools for the build according to
> > > > Documentation/process/changes.rst are 5.1 and 11.0.0 respectively.
> > > >
> > > > Remove the feature detection script, Kconfig option, and clean up some
> > > > fallback code that is no longer supported.
> > > >
> > > > The removed script was also testing for a GCC specific bug that was
> > > > fixed in the 4.7 release.
> > > >
> > > > The script was also not portable; users of Dash shell reported errors
> > > > when it was invoked.
> > > >
> > >
> > > To be clear, the script was portable, what is not working with dash is
> > > the current detection of CC_HAS_ASM_GOTO_TIED_OUTPUT. I'll try the other
> > > suggestion from Masahiro.
> >
> > Ah, that was his point about echo; that makes more sense.
> >
> > Unless a v2 is required, perhaps Masahiro would be kind enough to drop
> > this sentence from the commit message when applying?
>
> I can if there is nothing else to fix.
>
>
> And, Alexandre's Reported-by is irrelevant.
>
> As for the subject prefix "Kconfig:",
> I prefer something else, for example,
> "asm goto:" or "jump label:".
>
> I want to reserve "Kconfig:" for changes in scripts/kconfig/.
>
>
> >
> > >
> > > > --- a/arch/x86/include/asm/cpufeature.h
> > > > +++ b/arch/x86/include/asm/cpufeature.h
> > > > @@ -155,11 +155,11 @@ extern void clear_cpu_cap(struct cpuinfo_x86 *c, unsigned int bit);
> > > >
> > > > #define setup_force_cpu_bug(bit) setup_force_cpu_cap(bit)
> > > >
> > > > -#if defined(__clang__) && !defined(CONFIG_CC_HAS_ASM_GOTO)
> > > > +#if defined(__clang__) && __clang_major__ < 9
> > >
> > > Shouldn't we simply mandates clang >= 9 and drop the whole section? This
> > > is what you do later on.
> >
> > I considered it, but I don't think it would be safe to do so in this
> > header. If you look at the comment block below it, it mentions that
> > these kernel headers are being sucked into UAPI headers that are used
> > outside of the kernel builds, such as when building eBPF programs. So
> > we don't know what userspace tools might be consuming these headers.
> > The original intent of the guard was to not break eBPF compilation
> > with older clang releases, so I've retained. that functionality.
> >
> > + Alexei to review
> > (author of
> > commit b1ae32dbab50 ("x86/cpufeature: Guard asm_volatile_goto usage
> > for BPF compilation")
>
>
> I am not sure at this point.
> Wait for the input from Alexei.
It's probably best to remove this old workaround.
bpftrace still has code that does:
cflags.push_back("-D__BPF_TRACING__");
and bpftrace sort-of works with llvm 6+,
but they need to move to a newer clang for tons of other reasons
including lots of fixes in the clang bpf backend that were
added over the years.
So just nuke this old hack.
