Re: [PATCH] objtool: drop redundant flags generation

[Josh Poimboeuf]

On Mon, Mar 06, 2017 at 06:00:25PM +0100, Nicholas Mc Guire wrote:
> The generator was emitting quite a few duplicate flags which was making
> doublebitand.cocci nervous. This awk hack resolves the duplicate issue.
> 
> Signed-off-by: Nicholas Mc Guire <der.herr@xxxxxxx>
> ---
> 
> The coccinelle complaints emitted was about 230 findings total:
> ./arch/x86/lib/inat-tables.c:214:10-20: duplicated argument to & or |
> ./arch/x86/lib/inat-tables.c:214:23-33: duplicated argument to & or |
> ./arch/x86/lib/inat-tables.c:218:10-20: duplicated argument to & or |
> ./arch/x86/lib/inat-tables.c:218:23-33: duplicated argument to & or |
> ....
> ./tools/objtool/arch/x86/insn/inat-tables.c:214:10-20: duplicated argument to & or |
> ./tools/objtool/arch/x86/insn/inat-tables.c:214:23-33: duplicated argument to & or |
> ./tools/objtool/arch/x86/insn/inat-tables.c:218:10-20: duplicated argument to & or |
> ./tools/objtool/arch/x86/insn/inat-tables.c:218:23-33: duplicated argument to & or |
> ...
> spatch --sp-file scripts/coccinelle/tests/doublebitand.cocci inat-tables.c -D report
> will give you the full list - all are caused by duplicates in the generated
> output by the add_flags function in the two instances of gen-insn-attr-x86.awk.
> 
> Q: The two copies of gen-insn-attr-x86.awk are identical and its not actually clear
>    why this duplication is needed ? Further the maintainers list emitted for the
>    two files differ.
> 
> Patch was checked by manual review of the diff between the initial file and the
> regenerated file after the below patch was applied.
> Second verification was by make tools/objtool and comparing the generated binaries
> in tools/objtool/arch/x86/decode.o with diff.
> 
> Patch is against 4.11-rc1 (localversion-next is next-20170306)
> 
>  arch/x86/tools/gen-insn-attr-x86.awk              | 12 ++++++++++--
>  tools/objtool/arch/x86/insn/gen-insn-attr-x86.awk | 12 ++++++++++--

There's actually a third copy of the decoder in:

  tools/perf/util/intel-pt-decoder/

Yes, the duplication is a pain, but it's part of an effort to keep
'tools/*' source independent from kernel code.

Maybe we can at least combine the objtool and perf versions someday.

-- 
Josh