[PATCH v2] selftests/powerpc: Fix pkey syscall redefinitions

[PATCH v2] selftests/powerpc: Fix pkey syscall redefinitions

[Sandipan Das]

On some distros, there are conflicts w.r.t to redefinition
of pkey syscall numbers which cause build failures. This
fixes them.

Reported-by: Sachin Sant <sachinp@linux.vnet.ibm.com>
Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
---
Previous versions can be found at:
v1: https://lore.kernel.org/linuxppc-dev/20200803074043.466809-1-sandipan@linux.ibm.com/

Changes in v2:
- Fix incorrect commit message.

---
 tools/testing/selftests/powerpc/include/pkeys.h | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/tools/testing/selftests/powerpc/include/pkeys.h b/tools/testing/selftests/powerpc/include/pkeys.h
index 6ba95039a034..26eef5c1f8ea 100644
--- a/tools/testing/selftests/powerpc/include/pkeys.h
+++ b/tools/testing/selftests/powerpc/include/pkeys.h
@@ -31,8 +31,13 @@
 
 #define SI_PKEY_OFFSET	0x20
 
+#undef SYS_pkey_mprotect
 #define SYS_pkey_mprotect	386
+
+#undef SYS_pkey_alloc
 #define SYS_pkey_alloc		384
+
+#undef SYS_pkey_free
 #define SYS_pkey_free		385
 
 #define PKEY_BITS_PER_PKEY	2
-- 
2.25.1

Re: [PATCH v2] selftests/powerpc: Fix pkey syscall redefinitions

[Michael Ellerman]

Sandipan Das <sandipan@linux.ibm.com> writes:
> On some distros, there are conflicts w.r.t to redefinition
> of pkey syscall numbers which cause build failures. This
> fixes them.
>
> Reported-by: Sachin Sant <sachinp@linux.vnet.ibm.com>
> Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
> ---
> Previous versions can be found at:
> v1: https://lore.kernel.org/linuxppc-dev/20200803074043.466809-1-sandipan@linux.ibm.com/
>
> Changes in v2:
> - Fix incorrect commit message.
>
> ---
>  tools/testing/selftests/powerpc/include/pkeys.h | 5 +++++
>  1 file changed, 5 insertions(+)
>
> diff --git a/tools/testing/selftests/powerpc/include/pkeys.h b/tools/testing/selftests/powerpc/include/pkeys.h
> index 6ba95039a034..26eef5c1f8ea 100644
> --- a/tools/testing/selftests/powerpc/include/pkeys.h
> +++ b/tools/testing/selftests/powerpc/include/pkeys.h
> @@ -31,8 +31,13 @@
>  
>  #define SI_PKEY_OFFSET	0x20
>  
> +#undef SYS_pkey_mprotect
>  #define SYS_pkey_mprotect	386

We shouldn't undef them.

They should obviously never change, but if the system headers already
have a definition then we should use that, so I think it should be:

#ifndef SYS_pkey_mprotect
#define SYS_pkey_mprotect	386
#endif

cheers

Re: [PATCH v2] selftests/powerpc: Fix pkey syscall redefinitions

[Sandipan Das]

On 03/08/20 4:34 pm, Michael Ellerman wrote:
> Sandipan Das <sandipan@linux.ibm.com> writes:
>> On some distros, there are conflicts w.r.t to redefinition
>> of pkey syscall numbers which cause build failures. This
>> fixes them.
>>
>> Reported-by: Sachin Sant <sachinp@linux.vnet.ibm.com>
>> Signed-off-by: Sandipan Das <sandipan@linux.ibm.com>
>> ---
>> Previous versions can be found at:
>> v1: https://lore.kernel.org/linuxppc-dev/20200803074043.466809-1-sandipan@linux.ibm.com/
>>
>> Changes in v2:
>> - Fix incorrect commit message.
>>
>> ---
>>  tools/testing/selftests/powerpc/include/pkeys.h | 5 +++++
>>  1 file changed, 5 insertions(+)
>>
>> diff --git a/tools/testing/selftests/powerpc/include/pkeys.h b/tools/testing/selftests/powerpc/include/pkeys.h
>> index 6ba95039a034..26eef5c1f8ea 100644
>> --- a/tools/testing/selftests/powerpc/include/pkeys.h
>> +++ b/tools/testing/selftests/powerpc/include/pkeys.h
>> @@ -31,8 +31,13 @@
>>  
>>  #define SI_PKEY_OFFSET	0x20
>>  
>> +#undef SYS_pkey_mprotect
>>  #define SYS_pkey_mprotect	386
> 
> We shouldn't undef them.
> 
> They should obviously never change, but if the system headers already
> have a definition then we should use that, so I think it should be:
> 
> #ifndef SYS_pkey_mprotect
> #define SYS_pkey_mprotect	386
> #endif
> 

Agreed. This had me confused.

$ grep -nr "#define __NR_pkey_" /usr/include/
/usr/include/asm-generic/unistd.h:767:#define __NR_pkey_mprotect 288
/usr/include/asm-generic/unistd.h:769:#define __NR_pkey_alloc 289
/usr/include/asm-generic/unistd.h:771:#define __NR_pkey_free 290
/usr/include/powerpc64le-linux-gnu/asm/unistd_32.h:374:#define __NR_pkey_alloc  384
/usr/include/powerpc64le-linux-gnu/asm/unistd_32.h:375:#define __NR_pkey_free   385
/usr/include/powerpc64le-linux-gnu/asm/unistd_32.h:376:#define __NR_pkey_mprotect       386
/usr/include/powerpc64le-linux-gnu/asm/unistd_64.h:365:#define __NR_pkey_alloc  384
/usr/include/powerpc64le-linux-gnu/asm/unistd_64.h:366:#define __NR_pkey_free   385
/usr/include/powerpc64le-linux-gnu/asm/unistd_64.h:367:#define __NR_pkey_mprotect       386
...

But it looks like including unistd.h from a C program picks the
right values.


- Sandipan

RE: [PATCH v2] selftests/powerpc: Fix pkey syscall redefinitions

[David Laight]

> > +#undef SYS_pkey_mprotect
> >  #define SYS_pkey_mprotect	386
> 
> We shouldn't undef them.
> 
> They should obviously never change, but if the system headers already
> have a definition then we should use that, so I think it should be:
> 
> #ifndef SYS_pkey_mprotect
> #define SYS_pkey_mprotect	386
> #endif

If the definitions are identical the compiler won't complain.
So you probably actually want a matching definition so that,
provided at least one compile picks up both headers, you know
that the definitions actually match.

	David

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)