On Fri, Mar 14, 2025, at 12:55, Will Deacon wrote:
> On Fri, Mar 14, 2025 at 08:09:39AM +0100, Thomas Huth wrote:
>> __ASSEMBLY__ is only defined by the Makefile of the kernel, so
>> this is not really useful for uapi headers (unless the userspace
>> Makefile defines it, too). Let's switch to __ASSEMBLER__ which
>> gets set automatically by the compiler when compiling assembly
>> code.
>>
>> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
>> Cc: Will Deacon <will@xxxxxxxxxx>
>> Signed-off-by: Thomas Huth <thuth@xxxxxxxxxx>
>> ---
>> arch/arm64/include/uapi/asm/kvm.h | 2 +-
>> arch/arm64/include/uapi/asm/ptrace.h | 4 ++--
>> arch/arm64/include/uapi/asm/sigcontext.h | 4 ++--
>> 3 files changed, 5 insertions(+), 5 deletions(-)
>
> Is there a risk of breaking userspace with this? I wonder if it would
> be more conservative to do something like:
>
> #if !defined(__ASSEMBLY__) && !defined(__ASSEMBLER__)
>
> so that if somebody is doing '#define __ASSEMBLY__' then they get the
> same behaviour as today.
>
> Or maybe we don't care?
I think the main risk we would have is user applications relying
on the __ASSEMBLER__ checks in new kernel headers and not defining
__ASSEMBLY__. This would result in the application not building
against old kernel headers that only check against __ASSEMBLY__.
Checking for both in the kernel headers does not solve this
problem, and I think we can still decide that we don't care:
in the worst case, an application using the headers from assembly
will have to get fixed later when it needs to be built against
old headers.
I checked that old gcc versions pass __ASSEMBLY__ at least as
far back as gcc-2.95, and it should be completely safe to
assume that no older gcc versions would be used on kernel
headers, and they probably would choke on c99 features like
'long long'. I would also assume that any other compiler that
may be used to include kernel headers has to have enough
gcc compatibility to define all the common macros.
Arnd
