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;
Reply-To: Sean Christopherson <seanjc@google.com>
Date:   Wed,  2 Feb 2022 00:49:41 +0000
In-Reply-To: <20220202004945.2540433-1-seanjc@google.com>
Message-Id: <20220202004945.2540433-2-seanjc@google.com>
Mime-Version: 1.0
References: <20220202004945.2540433-1-seanjc@google.com>
Subject: [PATCH v2 1/5] Kconfig: Add option for asm goto w/ tied outputs to
 workaround clang-13 bug
From:   Sean Christopherson <seanjc@google.com>
To:     Paolo Bonzini <pbonzini@redhat.com>,
        Nathan Chancellor <nathan@kernel.org>,
        Nick Desaulniers <ndesaulniers@google.com>
Cc:     Sean Christopherson <seanjc@google.com>,
        Vitaly Kuznetsov <vkuznets@redhat.com>,
        Wanpeng Li <wanpengli@tencent.com>,
        Jim Mattson <jmattson@google.com>,
        Joerg Roedel <joro@8bytes.org>, kvm@vger.kernel.org,
        llvm@lists.linux.dev, linux-kernel@vger.kernel.org,
        Peter Zijlstra <peterz@infradead.org>,
        Tadeusz Struk <tadeusz.struk@linaro.org>,
        syzbot+6cde2282daa792c49ab8@syzkaller.appspotmail.com
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Add a config option to guard (future) usage of asm_volatile_goto() that
includes "tied outputs", i.e. "+" constraints that specify both an input
and output parameter.  clang-13 has a bug[1] that causes compilation of
such inline asm to fail, and KVM wants to use a "+m" constraint to
implement a uaccess form of CMPXCHG[2].  E.g. the test code fails with

  <stdin>:1:29: error: invalid operand in inline asm: '.long (${1:l}) - .'
  int foo(int *x) { asm goto (".long (%l[bar]) - .\n": "+m"(*x) ::: bar); return *x; bar: return 0; }
                            ^
  <stdin>:1:29: error: unknown token in expression
  <inline asm>:1:9: note: instantiated into assembly here
          .long () - .
                 ^
  2 errors generated.

on clang-13, but passes on gcc (with appropriate asm goto support).  The
bug is fixed in clang-14, but won't be backported to clang-13 as the
changes are too invasive/risky.

gcc also had a similar bug[3], fixed in gcc-11, where gcc failed to
account for its behavior of assigning two numbers to tied outputs (one
for input, one for output) when evaluating symbolic references.

[1] https://github.com/ClangBuiltLinux/linux/issues/1512
[2] https://lore.kernel.org/all/YfMruK8%2F1izZ2VHS@google.com
[3] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=98096

Suggested-by: Nick Desaulniers <ndesaulniers@google.com>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <seanjc@google.com>
---
 init/Kconfig | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/init/Kconfig b/init/Kconfig
index e9119bf54b1f..4a7a569706c5 100644
--- a/init/Kconfig
+++ b/init/Kconfig
@@ -77,6 +77,11 @@ config CC_HAS_ASM_GOTO_OUTPUT
 	depends on CC_HAS_ASM_GOTO
 	def_bool $(success,echo 'int foo(int x) { asm goto ("": "=r"(x) ::: bar); return x; bar: return 0; }' | $(CC) -x c - -c -o /dev/null)
 
+config CC_HAS_ASM_GOTO_TIED_OUTPUT
+	depends on CC_HAS_ASM_GOTO_OUTPUT
+	# Detect buggy gcc and clang, fixed in gcc-11 clang-14.
+	def_bool $(success,echo 'int foo(int *x) { asm goto (".long (%l[bar]) - .\n": "+m"(*x) ::: bar); return *x; bar: return 0; }' | $CC -x c - -c -o /dev/null)
+
 config TOOLS_SUPPORT_RELR
 	def_bool $(success,env "CC=$(CC)" "LD=$(LD)" "NM=$(NM)" "OBJCOPY=$(OBJCOPY)" $(srctree)/scripts/tools-support-relr.sh)
 
-- 
2.35.0.rc2.247.g8bbb082509-goog