Re: [PATCH v3 00/20] Remove PERF_HAVE_DWARF_REGS

[Namhyung Kim <namhyung@kernel.org>]

On Fri, Nov 08, 2024 at 09:00:40AM +0900, Masami Hiramatsu wrote:
> Hi,
> 
> On Thu, 7 Nov 2024 11:29:34 -0800
> Namhyung Kim <namhyung@kernel.org> wrote:
> 
> > Hi all,
> > 
> > On Wed, Oct 16, 2024 at 05:25:00PM -0700, Ian Rogers wrote:
> > > These changes are on top of:
> > > https://lore.kernel.org/lkml/20241017001354.56973-1-irogers@google.com/
> > > 
> > > Prior to these patches PERF_HAVE_DWARF_REGS indicated the presence of
> > > dwarf-regs.c in the arch directory. dwarf-regs.c provided upto 4
> > > functions:
> > > 
> > > 1) regs_query_register_offset would translate a register name into a
> > > pt_regs offset and was used by BPF prologues. BPF prologues existed
> > > for BPF events and support for these was removed many releases ago.
> > > This code was dead and could be removed.
> > > 
> > > 2) get_arch_regstr duplicated get_dwarf_regstr and so it could be
> > > removed.  The case for csky was a little more complicated as the ABI
> > > controlled the string. The callers of get_dwarf_regstr were updated to
> > > also pass the ELF flags so that on csky the ABI appropriate table
> > > could be used. As the argument is only used on csky this a no-op for
> > > everything else.
> > > 
> > > 3) get_arch_regnum translated a register name back to a dwarf number
> > > and only existed on x86 where "al", "ax", "eax" and "rax" could all
> > > mean register 0. This code was moved to util with similar
> > > machine/flags logic to get_arch_regstr and for consistency with it.
> > > 
> > > 4) get_powerpc_regs a PowerPC specific function used by annotate that
> > > should really be in util.
> > > 
> > > 2 and 3 required the wiring through of the ELF machine and flags in
> > > callers to get_dwarf_regstr and get_dwarf_regnum. When these values
> > > weren't dependent on an ELF file a new EM_HOST and EF_HOST were added
> > > to give the host ELF machine and flags. These 2 #defines got rid of
> > > the existing separate arch files and #ifdefs.
> > > 
> > > v3: These files were separated from the rest of the v2 libdw clean up
> > >     in:
> > > https://lore.kernel.org/lkml/CAP-5=fVZH3L-6y_sxLwSmT8WyMXDMFnuqUksNULdQYJCPNBFYw@mail.gmail.com/
> > > 
> > > Ian Rogers (20):
> > >   perf bpf-prologue: Remove unused file
> > >   perf dwarf-regs: Remove PERF_HAVE_ARCH_REGS_QUERY_REGISTER_OFFSET
> > >   perf dwarf-regs: Add EM_HOST and EF_HOST defines
> > >   perf disasm: Add e_machine/e_flags to struct arch
> > >   perf dwarf-regs: Pass accurate disassembly machine to get_dwarf_regnum
> > >   perf dwarf-regs: Pass ELF flags to get_dwarf_regstr
> > >   perf dwarf-regs: Move x86 dwarf-regs out of arch
> > >   perf arm64: Remove dwarf-regs.c
> > >   perf arm: Remove dwarf-regs.c
> > >   perf dwarf-regs: Move csky dwarf-regs out of arch
> > >   perf loongarch: Remove dwarf-regs.c
> > >   perf mips: Remove dwarf-regs.c
> > >   perf dwarf-regs: Move powerpc dwarf-regs out of arch
> > >   perf riscv: Remove dwarf-regs.c and add dwarf-regs-table.h
> > >   perf s390: Remove dwarf-regs.c
> > >   perf sh: Remove dwarf-regs.c
> > >   perf sparc: Remove dwarf-regs.c
> > >   perf xtensa: Remove dwarf-regs.c
> > >   perf dwarf-regs: Remove get_arch_regstr code
> > >   perf build: Remove PERF_HAVE_DWARF_REGS
> > 
> > These look all good and I'm about to apply the series.
> > 
> > Masami, do you have any remaining concerns?  It'd be nice if you can
> > give your Reviewed-by.
> 
> I think it looks good to me.
> 
> Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Thanks for your review!
Namhyung