Re: [PATCH v3 04/20] perf disasm: Add e_machine/e_flags to struct arch

[Ian Rogers <irogers@google.com>]

On Fri, Nov 8, 2024 at 10:38 AM Namhyung Kim <namhyung@kernel.org> wrote:
>
> On Fri, Nov 08, 2024 at 10:19:52AM -0800, Ian Rogers wrote:
> > On Fri, Nov 8, 2024 at 9:33 AM Namhyung Kim <namhyung@kernel.org> wrote:
> > >
> > > On Wed, Oct 16, 2024 at 05:25:04PM -0700, Ian Rogers wrote:
> > > > Currently functions like get_dwarf_regnum only work with the host
> > > > architecture. Carry the elf machine and flags in struct arch so that
> > > > in disassembly these can be used to allow cross platform disassembly.
> > > >
> > > > Signed-off-by: Ian Rogers <irogers@google.com>
> > > > ---
> > > >  tools/perf/arch/arc/annotate/instructions.c       | 2 ++
> > > >  tools/perf/arch/arm/annotate/instructions.c       | 2 ++
> > > >  tools/perf/arch/arm64/annotate/instructions.c     | 2 ++
> > > >  tools/perf/arch/csky/annotate/instructions.c      | 7 ++++++-
> > > >  tools/perf/arch/loongarch/annotate/instructions.c | 2 ++
> > > >  tools/perf/arch/mips/annotate/instructions.c      | 2 ++
> > > >  tools/perf/arch/powerpc/annotate/instructions.c   | 2 ++
> > > >  tools/perf/arch/riscv64/annotate/instructions.c   | 2 ++
> > > >  tools/perf/arch/s390/annotate/instructions.c      | 2 ++
> > > >  tools/perf/arch/sparc/annotate/instructions.c     | 2 ++
> > > >  tools/perf/arch/x86/annotate/instructions.c       | 3 ++-
> > > >  tools/perf/util/disasm.h                          | 4 ++++
> > > >  12 files changed, 30 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/tools/perf/arch/arc/annotate/instructions.c b/tools/perf/arch/arc/annotate/instructions.c
> > > > index 2f00e995c7e3..e5619770a1af 100644
> > > > --- a/tools/perf/arch/arc/annotate/instructions.c
> > > > +++ b/tools/perf/arch/arc/annotate/instructions.c
> > > > @@ -5,5 +5,7 @@ static int arc__annotate_init(struct arch *arch, char *cpuid __maybe_unused)
> > > >  {
> > > >       arch->initialized = true;
> > > >       arch->objdump.comment_char = ';';
> > > > +     arch->e_machine = EM_ARC;
> > > > +     arch->e_flags = 0;
> > > >       return 0;
> > > >  }
> > > > diff --git a/tools/perf/arch/arm/annotate/instructions.c b/tools/perf/arch/arm/annotate/instructions.c
> > > > index 2ff6cedeb9c5..cf91a43362b0 100644
> > > > --- a/tools/perf/arch/arm/annotate/instructions.c
> > > > +++ b/tools/perf/arch/arm/annotate/instructions.c
> > > > @@ -53,6 +53,8 @@ static int arm__annotate_init(struct arch *arch, char *cpuid __maybe_unused)
> > > >       arch->associate_instruction_ops   = arm__associate_instruction_ops;
> > > >       arch->objdump.comment_char        = ';';
> > > >       arch->objdump.skip_functions_char = '+';
> > > > +     arch->e_machine = EM_ARM;
> > > > +     arch->e_flags = 0;
> > > >       return 0;
> > > >
> > > >  out_free_call:
> > > > diff --git a/tools/perf/arch/arm64/annotate/instructions.c b/tools/perf/arch/arm64/annotate/instructions.c
> > > > index f86d9f4798bd..d465d093e7eb 100644
> > > > --- a/tools/perf/arch/arm64/annotate/instructions.c
> > > > +++ b/tools/perf/arch/arm64/annotate/instructions.c
> > > > @@ -113,6 +113,8 @@ static int arm64__annotate_init(struct arch *arch, char *cpuid __maybe_unused)
> > > >       arch->associate_instruction_ops   = arm64__associate_instruction_ops;
> > > >       arch->objdump.comment_char        = '/';
> > > >       arch->objdump.skip_functions_char = '+';
> > > > +     arch->e_machine = EM_AARCH64;
> > > > +     arch->e_flags = 0;
> > > >       return 0;
> > > >
> > > >  out_free_call:
> > > > diff --git a/tools/perf/arch/csky/annotate/instructions.c b/tools/perf/arch/csky/annotate/instructions.c
> > > > index 5337bfb7d5fc..14270311d215 100644
> > > > --- a/tools/perf/arch/csky/annotate/instructions.c
> > > > +++ b/tools/perf/arch/csky/annotate/instructions.c
> > > > @@ -43,6 +43,11 @@ static int csky__annotate_init(struct arch *arch, char *cpuid __maybe_unused)
> > > >       arch->initialized = true;
> > > >       arch->objdump.comment_char = '/';
> > > >       arch->associate_instruction_ops = csky__associate_ins_ops;
> > > > -
> > > > +     arch->e_machine = EM_CSKY;
> > > > +#if defined(__CSKYABIV2__)
> > > > +     arch->e_flags = EF_CSKY_ABIV2;
> > > > +#else
> > > > +     arch->e_flags = EF_CSKY_ABIV1;
> > > > +#endif
> > >
> > > By moving this into the general code, it should take care of old systems
> > > that doesn't have the macro.
> > >
> > >   In file included from util/disasm.c:109:
> > >   /linux/tools/perf/arch/csky/annotate/instructions.c: In function 'csky__annotate_init':
> > >   /linux/tools/perf/arch/csky/annotate/instructions.c:50:25: error: 'EF_CSKY_ABIV1' undeclared (first use in this function)
> > >      50 |         arch->e_flags = EF_CSKY_ABIV1;
> > >         |                         ^~~~~~~~~~~~~
> > >   /linux/tools/perf/arch/csky/annotate/instructions.c:50:25: note: each undeclared identifier is reported only once for each function it appears in
> >
> > EF_CSKY_ABIV1 is defined in elf.h and has been there at least 5 years in libelf:
> > https://sourceware.org/git/?p=elfutils.git;a=commit;f=libelf/elf.h;h=9c82942ae7355a3226c53a92c2c73b33193c5e33
> > I suspected the issue here is missing elf.h include, but the .c file
> > is included in tools/perf/util/disasm.c and that must have a
> > transitive dependency given other things are building. Do you want me
> > to send a patch making this conditional with extra #ifdefs or re-send
> > the series?
>
> Yeah, it's unfortunate but I think we can have a small incremental diff
> here to define them if it's not there.  Then I'll squash it to the
> patch.

Thanks, sgtm. Could you also explicitly include elf.h into disasm.c
rather than depending on the transitive #include? There's obviously
tech debt to not have .c files including each other to work around the
build system.

Ian

> >
> > > Also, I think __CSKYABIV2__ is defined only when the host is csky.  So
> > > it'll use ABI v1 on cross env.  I'm not sure if it's a problem.  We may
> > > need to save the ABI somewhere in the metadata later.
> >
> > Agreed. In general we should read e_machine and e_flags from the ELF
> > file, so I'm not sure new metadata is needed. This patch is trying to
> > lay groundwork for that.
>
> I understand that.  Yeah it should come from the binary.
>
> Thanks,
> Namhyung
>