x86/percpu: Fix "multiple identical address spaces specified for type" clang warning
Subsystems (2)
PATCH 1/2 x86/percpu: Fix "multiple identical address spaces specified for type" clang warning
The clang build with named address spaces enabled currently fails with:
error: multiple identical address spaces specified for type [-Werror,-Wduplicate-decl-specifier]
The warning is emitted when accessing const_pcpu_hot structure, which is already declared in __seg_gs named address space.
Use specialized accessor for __raw_cpu_read_const() instead, avoiding redeclaring __seg_gs named address space via __raw_cpu_read().
--- arch/x86/include/asm/percpu.h | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)
arch/x86/include/asm/percpu.h +4 -1
@@ -160,7 +160,10 @@ do { \ *(qual __my_cpu_type(pcp) *)__my_cpu_ptr(&(pcp)) = (val); \ } while (0) -#define __raw_cpu_read_const (pcp) __raw_cpu_read(, , pcp) +#define __raw_cpu_read_const (pcp) \ +({ \ + *(typeof(pcp) *)(__force uintptr_t)(&(pcp)); \ +})#else /* !CONFIG_USE_X86_SEG_SUPPORT: */
Signature · 1 line
2.42.0
PATCH 2/2 x86/percpu: *DO NOT MERGE* Enable named address spaces for clang
Enable named address spaces feature for clang.
However, someone from clang side should qualify the compiler before this feature is enabled for real.
--- arch/x86/Kconfig | 1 - 1 file changed, 1 deletion(-)
arch/x86/Kconfig +0 -1
@@ -2433,7 +2433,6 @@ endmenu config CC_HAS_NAMED_AS def_bool $(success,echo 'int __seg_fs fs; int __seg_gs gs;' | $(CC) -x c - -S -o /dev/null) - depends on CC_IS_GCC config CC_HAS_NAMED_AS_FIXED_SANITIZERS def_bool CC_IS_GCC && GCC_VERSION >= 130300
Signature · 1 line
2.42.0
Quoted 11 lines · Uros Bizjak
The clang build with named address spaces enabled currently fails with:
error: multiple identical address spaces specified for type [-Werror,-Wduplicate-decl-specifier]
The warning is emitted when accessing
const_pcpu_hotstructure, which is already declared in__seg_gsnamed address space.Use specialized accessor for
__raw_cpu_read_const()instead, avoiding redeclaring__seg_gsnamed address space via__raw_cpu_read().Signed-off-by Uros Bizjak <[email protected]>
Thanks, this resolves the warning for me when patch 2 is applied.
However, as both of us have noticed [1][2], there is a problem in LLVM's x86 backend with the address space casting that the kernel does, so patch 2 should not be merged.
[1]: https://github.com/ClangBuiltLinux/linux/issues/2013
[2]: https://github.com/llvm/llvm-project/issues/93449
Quoted 32 lines
Cc Thomas Gleixner <[email protected]>Cc Ingo Molnar <[email protected]>Cc Borislav Petkov <[email protected]>Cc Dave Hansen <[email protected]>Cc "H. Peter Anvin" <[email protected]>Cc Nathan Chancellor <[email protected]>Cc Nick Desaulniers <[email protected]>Cc Bill Wendling <[email protected]>Cc Justin Stitt <[email protected]>--- arch/x86/include/asm/percpu.h | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)arch/x86/include/asm/percpu.h +4
-1@@ -160,7 +160,10 @@ do { \ *(qual __my_cpu_type(pcp) *)__my_cpu_ptr(&(pcp)) = (val); \ } while (0) -#define __raw_cpu_read_const (pcp) __raw_cpu_read(, , pcp) +#define __raw_cpu_read_const (pcp) \ +({ \ + *(typeof(pcp) *)(__force uintptr_t)(&(pcp)); \ +})#else /* !CONFIG_USE_X86_SEG_SUPPORT: */ Signature · 1 line
2.42.0
Quoted 29 lines · Uros Bizjak
The clang build with named address spaces enabled currently fails with:
error: multiple identical address spaces specified for type [-Werror,-Wduplicate-decl-specifier]
The warning is emitted when accessing
const_pcpu_hotstructure, which is already declared in__seg_gsnamed address space.Use specialized accessor for
__raw_cpu_read_const()instead, avoiding redeclaring__seg_gsnamed address space via__raw_cpu_read().Signed-off-by Uros Bizjak <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Ingo Molnar <[email protected]>Cc Borislav Petkov <[email protected]>Cc Dave Hansen <[email protected]>Cc "H. Peter Anvin" <[email protected]>Cc Nathan Chancellor <[email protected]>Cc Nick Desaulniers <[email protected]>Cc Bill Wendling <[email protected]>Cc Justin Stitt <[email protected]>--- arch/x86/include/asm/percpu.h | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)arch/x86/include/asm/percpu.h +0
-0
arch/x86/include/asm/percpu.h do { \ · line 166
Quoted 8 lines
arch/x86/include/asm/percpu.h +4
-1@@ -160,7 +160,10 @@ do { \ *(qual __my_cpu_type(pcp) *)__my_cpu_ptr(&(pcp)) = (val); \ } while (0) -#define __raw_cpu_read_const (pcp) __raw_cpu_read(, , pcp) +#define __raw_cpu_read_const (pcp) \ +({ \ + *(typeof(pcp) *)(__force uintptr_t)(&(pcp)); \ +})
So shouldn't this be carried together with the patch that enables named address spaces on Clang? Is it possible that this patch might not be needed in a Clang version where named address spaces work as expected?
Thanks,
Ingo
Quoted 36 lines · Ingo Molnar
Quoted 32 lines · Uros Bizjak
The clang build with named address spaces enabled currently fails with:
error: multiple identical address spaces specified for type [-Werror,-Wduplicate-decl-specifier]
The warning is emitted when accessing
const_pcpu_hotstructure, which is already declared in__seg_gsnamed address space.Use specialized accessor for
__raw_cpu_read_const()instead, avoiding redeclaring__seg_gsnamed address space via__raw_cpu_read().Signed-off-by Uros Bizjak <[email protected]>Cc Thomas Gleixner <[email protected]>Cc Ingo Molnar <[email protected]>Cc Borislav Petkov <[email protected]>Cc Dave Hansen <[email protected]>Cc "H. Peter Anvin" <[email protected]>Cc Nathan Chancellor <[email protected]>Cc Nick Desaulniers <[email protected]>Cc Bill Wendling <[email protected]>Cc Justin Stitt <[email protected]>--- arch/x86/include/asm/percpu.h | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-)arch/x86/include/asm/percpu.h +0
-0@@ -160,7 +160,10 @@ do { \ *(qual __my_cpu_type(pcp) *)__my_cpu_ptr(&(pcp)) = (val); \ }while (0 )
Quoted 9 lines
arch/x86/include/asm/percpu.h do { \ · line 166
Quoted 5 lines
arch/x86/include/asm/percpu.h +4
-1-#define __raw_cpu_read_const (pcp) __raw_cpu_read(, , pcp) +#define __raw_cpu_read_const (pcp) \ +({ \ + *(typeof(pcp) *)(__force uintptr_t)(&(pcp)); \ +})So shouldn't this be carried together with the patch that enables named address spaces on Clang? Is it possible that this patch might not be needed in a Clang version where named address spaces work as expected?
No, as diagnosed by Clang, there are indeed two identical specifiers, so the patch also applies as a general patch.
Regarding Clang, before enabling the feature, issue #93449 [1] (internal compiler error) has to be fixed.
[1] https://github.com/llvm/llvm-project/issues/93449
Uros.