Re: [PATCH v2] samples/bpf: Add a workaround for asm_inline

From: Song Liu
Date: Wed Oct 02 2019 - 17:05:39 EST




> On Oct 2, 2019, at 1:22 PM, Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote:
>
> On Wed, Oct 2, 2019 at 12:17 PM KP Singh <kpsingh@xxxxxxxxxxxx> wrote:
>>
>> From: KP Singh <kpsingh@xxxxxxxxxx>
>>
>> This was added in:
>>
>> commit eb111869301e ("compiler-types.h: add asm_inline definition")
>>
>> and breaks samples/bpf as clang does not support asm __inline.
>>
>> Co-developed-by: Florent Revest <revest@xxxxxxxxxx>
>> Signed-off-by: Florent Revest <revest@xxxxxxxxxx>
>> Signed-off-by: KP Singh <kpsingh@xxxxxxxxxx>
>> ---
>>
>> Changes since v1:
>>
>> - Dropped the rename from asm_workaround.h to asm_goto_workaround.h
>> - Dropped the fix for task_fd_query_user.c as it is updated in
>> https://lore.kernel.org/bpf/20191001112249.27341-1-bjorn.topel@xxxxxxxxx/
>>
>> samples/bpf/asm_goto_workaround.h | 13 ++++++++++++-
>> 1 file changed, 12 insertions(+), 1 deletion(-)
>>
>> diff --git a/samples/bpf/asm_goto_workaround.h b/samples/bpf/asm_goto_workaround.h
>> index 7409722727ca..7048bb3594d6 100644
>> --- a/samples/bpf/asm_goto_workaround.h
>> +++ b/samples/bpf/asm_goto_workaround.h
>> @@ -3,7 +3,8 @@
>> #ifndef __ASM_GOTO_WORKAROUND_H
>> #define __ASM_GOTO_WORKAROUND_H
>>
>> -/* this will bring in asm_volatile_goto macro definition
>> +/*
>> + * This will bring in asm_volatile_goto and asm_inline macro definitions
>> * if enabled by compiler and config options.
>> */
>> #include <linux/types.h>
>> @@ -13,5 +14,15 @@
>> #define asm_volatile_goto(x...) asm volatile("invalid use of asm_volatile_goto")
>> #endif
>>
>> +/*
>> + * asm_inline is defined as asm __inline in "include/linux/compiler_types.h"
>> + * if supported by the kernel's CC (i.e CONFIG_CC_HAS_ASM_INLINE) which is not
>> + * supported by CLANG.
>> + */
>> +#ifdef asm_inline
>> +#undef asm_inline
>> +#define asm_inline asm
>> +#endif
>
> Would it be better to just #undef CONFIG_CC_HAS_ASM_INLINE for BPF programs?

I guess that is still useful when gcc fully support BPF?

Thanks,
Song