x86: vdso: sanitize asm helpers for vgetcpu.c

Message ID 20230207161349.309901-1-arnd@kernel.org
State New
Headers
Series x86: vdso: sanitize asm helpers for vgetcpu.c |

Commit Message

Arnd Bergmann Feb. 7, 2023, 4:13 p.m. UTC
  From: Arnd Bergmann <arnd@arndb.de>

The x86 vdso implementation includes a few include/linux/*.h kernel
headers outside of the include/vdso/*.h space.  This causes a new
warning when building with clang, after the vgetcpu code is added
to the vdso32 support on 64-bit kernels:

In file included from arch/x86/entry/vdso/vdso32/vgetcpu.c:2:
In file included from arch/x86/entry/vdso/vdso32/../vgetcpu.c:8:
In file included from include/linux/kernel.h:22:
In file included from include/linux/bitops.h:68:
In file included from arch/x86/include/asm/bitops.h:420:
arch/x86/include/asm/arch_hweight.h:49:15: error: invalid input size for constraint 'D'
                         : REG_IN (w));
                                   ^
In file included from arch/x86/entry/vdso/vdso32/vgetcpu.c:2:
In file included from arch/x86/entry/vdso/vdso32/../vgetcpu.c:8:
In file included from include/linux/kernel.h:25:
In file included from include/linux/math.h:6:
arch/x86/include/asm/div64.h:85:34: error: invalid output size for constraint '=a'
        asm ("mulq %2; divq %3" : "=a" (q)
                                        ^

Unlike gcc, clang checks inline asm constraints before dead code
elimination, which breaks both __arch_hweight64() and mul_u64_u64_div_u64()
when these are included with CONFIG_64BIT set but compiled with clang -m32,
even when there are no callers.

Change both affected headers to the check for 32-bit vs 64-bit is done
correctly for vdso32. It would be nice to also limit the included headers
further, to avoid subtle differences in the header contents, but that
requires a larger cleanup.

Fixes: 92d33063c081 ("x86/vdso: Provide getcpu for x86-32.")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 arch/x86/entry/vdso/vdso32/vgetcpu.c | 2 ++
 arch/x86/include/asm/arch_hweight.h  | 4 ++--
 arch/x86/include/asm/div64.h         | 2 +-
 3 files changed, 5 insertions(+), 3 deletions(-)
  

Comments

Borislav Petkov Feb. 7, 2023, 4:18 p.m. UTC | #1
On Tue, Feb 07, 2023 at 05:13:41PM +0100, Arnd Bergmann wrote:
> From: Arnd Bergmann <arnd@arndb.de>
> 
> The x86 vdso implementation includes a few include/linux/*.h kernel
> headers outside of the include/vdso/*.h space.  This causes a new
> warning when building with clang, after the vgetcpu code is added
> to the vdso32 support on 64-bit kernels:

...

https://lore.kernel.org/r/Y%2BIsCWQdXEr8d9Vy@linutronix.de
  

Patch

diff --git a/arch/x86/entry/vdso/vdso32/vgetcpu.c b/arch/x86/entry/vdso/vdso32/vgetcpu.c
index b777f84ffae9..9f8cf77c7077 100644
--- a/arch/x86/entry/vdso/vdso32/vgetcpu.c
+++ b/arch/x86/entry/vdso/vdso32/vgetcpu.c
@@ -1,2 +1,4 @@ 
 // SPDX-License-Identifier: GPL-2.0
+#define BUILD_VDSO32
+
 #include "../vgetcpu.c"
diff --git a/arch/x86/include/asm/arch_hweight.h b/arch/x86/include/asm/arch_hweight.h
index ba88edd0d58b..cca0941673e0 100644
--- a/arch/x86/include/asm/arch_hweight.h
+++ b/arch/x86/include/asm/arch_hweight.h
@@ -4,7 +4,7 @@ 
 
 #include <asm/cpufeatures.h>
 
-#ifdef CONFIG_64BIT
+#if defined(CONFIG_64BIT) && !defined(BUILD_VDSO32)
 #define REG_IN "D"
 #define REG_OUT "a"
 #else
@@ -33,7 +33,7 @@  static inline unsigned int __arch_hweight8(unsigned int w)
 	return __arch_hweight32(w & 0xff);
 }
 
-#ifdef CONFIG_X86_32
+#if defined(CONFIG_X86_32) || defined(BUILD_VDSO32)
 static inline unsigned long __arch_hweight64(__u64 w)
 {
 	return  __arch_hweight32((u32)w) +
diff --git a/arch/x86/include/asm/div64.h b/arch/x86/include/asm/div64.h
index b8f1dc0761e4..97bd076db0b2 100644
--- a/arch/x86/include/asm/div64.h
+++ b/arch/x86/include/asm/div64.h
@@ -2,7 +2,7 @@ 
 #ifndef _ASM_X86_DIV64_H
 #define _ASM_X86_DIV64_H
 
-#ifdef CONFIG_X86_32
+#if defined(CONFIG_X86_32) || defined(BUILD_VDSO32)
 
 #include <linux/types.h>
 #include <linux/log2.h>