Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Wed, 28 Dec 2022 02:57:38 +0100
From:   Ahelenia =?utf-8?Q?Ziemia=C5=84ska?= 
        <nabijaczleweli@nabijaczleweli.xyz>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Ingo Molnar <mingo@redhat.com>,
        Arnaldo Carvalho de Melo <acme@kernel.org>,
        Mark Rutland <mark.rutland@arm.com>,
        Alexander Shishkin <alexander.shishkin@linux.intel.com>,
        Jiri Olsa <jolsa@kernel.org>,
        Namhyung Kim <namhyung@kernel.org>,
        Nathan Chancellor <nathan@kernel.org>,
        Nick Desaulniers <ndesaulniers@google.com>,
        Tom Rix <trix@redhat.com>, linux-perf-users@vger.kernel.org,
        linux-kernel@vger.kernel.org, llvm@lists.linux.dev
Subject: [PATCH v2 1/4] perf python: make clang feature detection work
Message-ID: <ea47c9d024e10a0cf50ae7f049320d89b307b5c7.1672192591.git.nabijaczleweli@nabijaczleweli.xyz>
References: <44bb9a29bd7a0385c25ebfd69abd0a41ac2d753b.1672187710.git.nabijaczleweli@nabijaczleweli.xyz>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha512;
        protocol="application/pgp-signature"; boundary="tip2335yapfu46iq"
Content-Disposition: inline
In-Reply-To: <44bb9a29bd7a0385c25ebfd69abd0a41ac2d753b.1672187710.git.nabijaczleweli@nabijaczleweli.xyz>
User-Agent: NeoMutt/20220429
To:     unlisted-recipients:; (no To-header on input)
Precedence: bulk


--tip2335yapfu46iq
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

As it stands, it's /entirely/ broken: cc_options is pasted together
with the tested option /as a single argument/.

For example, for CC=3D'cc -O3', this will run (the equivalent of)
  cc "-O3 -mcet" ../build/feature/test-heello.c
  cc "-O3 -fcf-protection" ../build/feature/test-heello.c
  ..
which, obviously, doesn't work, and has /never/ worked. The commit
referenced in Fixes: is just the first one that uses this approach.

Instead of emulating shell tokenisation and turning tokens into words
badly (well, we barely do it, which is even more alarming. the only way
I can foresee this having ever worked for anyone is if it, very literally,
returned False for /every/ clang_has_option() invocation with
non-single-token $CC, which, I mean, sure, but it means that every clang
build like this has all hardening off), just... run these through the
shell, like they will be at point-of-use anyway.

As a demo, with CC=3D'/tmp/cc -O3' and /tmp/cc consisting of
  #!/bin/sh
  echo "$*" | grep -qFe '-mcet' && printf 'CC:"%s"\n' "$@" "" > /dev/tty
  exec cc "$@"
before:
    AR      /mnt/filling/store/nabijaczleweli/code/linux/tools/perf/libsubc=
md/libsubcmd.a
  CC:"-O3 -mcet"
  CC:"/mnt/filling/store/nabijaczleweli/code/linux/tools/build/feature/test=
-hello.c"
  CC:""
  CC:"-O3 -mcet"
  CC:"/mnt/filling/store/nabijaczleweli/code/linux/tools/build/feature/test=
-hello.c"
  CC:""
    LD      /mnt/filling/store/nabijaczleweli/code/linux/tools/perf/libbpf/=
staticobjs/libbpf-in.o
after:
    GEN     python/perf.cpython-39-x86_64-linux-gnu.so
  CC:"-O3"
  CC:"-mcet"
  CC:"/mnt/filling/store/nabijaczleweli/code/linux/tools/build/feature/test=
-hello.c"
  CC:""
  CC:"-O3"
  CC:"-mcet"
  CC:"/mnt/filling/store/nabijaczleweli/code/linux/tools/build/feature/test=
-hello.c"
  CC:""
    LD      /mnt/filling/store/nabijaczleweli/code/linux/tools/perf/libbpf/=
staticobjs/libbpf-in.o

Fixes: commit 3cad53a6f9cdb ("perf python: Account for multiple words in
 CC")
Signed-off-by: Ahelenia Ziemia=C5=84ska <nabijaczleweli@nabijaczleweli.xyz>
---
 tools/perf/util/setup.py | 14 ++------------
 1 file changed, 2 insertions(+), 12 deletions(-)

diff --git a/tools/perf/util/setup.py b/tools/perf/util/setup.py
index c294db713677..1cee26c63613 100644
--- a/tools/perf/util/setup.py
+++ b/tools/perf/util/setup.py
@@ -2,21 +2,11 @@ from os import getenv, path
 from subprocess import Popen, PIPE
 from re import sub
=20
-cc =3D getenv("CC")
-
-# Check if CC has options, as is the case in yocto, where it uses CC=3D"cc=
 --sysroot..."
-cc_tokens =3D cc.split()
-if len(cc_tokens) > 1:
-    cc =3D cc_tokens[0]
-    cc_options =3D " ".join([str(e) for e in cc_tokens[1:]]) + " "
-else:
-    cc_options =3D ""
-
-cc_is_clang =3D b"clang version" in Popen([cc, "-v"], stderr=3DPIPE).stder=
r.readline()
+cc_is_clang =3D b"clang version" in Popen(["/bin/sh", "-c", "$CC -v"], std=
err=3DPIPE).stderr.readline()
 src_feature_tests  =3D getenv('srctree') + '/tools/build/feature'
=20
 def clang_has_option(option):
-    cc_output =3D Popen([cc, cc_options + option, path.join(src_feature_te=
sts, "test-hello.c") ], stderr=3DPIPE).stderr.readlines()
+    cc_output =3D Popen(["/bin/sh", "-c", '$CC "$@"', "", option, path.joi=
n(src_feature_tests, "test-hello.c")], stderr=3DPIPE).stderr.readlines()
     return [o for o in cc_output if ((b"unknown argument" in o) or (b"is n=
ot supported" in o))] =3D=3D [ ]
=20
 if cc_is_clang:
--=20
2.30.2


--tip2335yapfu46iq
Content-Type: application/pgp-signature; name="signature.asc"

-----BEGIN PGP SIGNATURE-----
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=9Kw2
-----END PGP SIGNATURE-----

--tip2335yapfu46iq--