Re: [PATCH] selftests/vDSO: Fix undefined CLONE_NEWTIME by including

[Purva Yeshi <purvayeshi550@gmail.com>]

On 30/01/25 12:32, Christophe Leroy wrote:
> 
> 
> Le 29/01/2025 à 19:17, Purva Yeshi a écrit :
>>
>> On 27/01/25 13:32, Christophe Leroy wrote:
>>>
>>>
>>> Le 26/01/2025 à 11:59, Purva Yeshi a écrit :
>>>> [Vous ne recevez pas souvent de courriers de purvayeshi550@gmail.com.
>>>> Découvrez pourquoi ceci est important à
>>>> https://aka.ms/LearnAboutSenderIdentification ]
>>>>
>>>> Fix the build failure caused by the undefined `CLONE_NEWTIME`.
>>>> Include the `linux/sched.h` header file where the function is 
>>>> defined to
>>>> ensure successful compilation of the selftests.
>>>
>>> This is supposed to be already fixed by commit 34d5b600172b ("selftests:
>>> vDSO: Explicitly include sched.h")
>>>
>>> Can you explain what is the exact problem still ? And why 
>>> linux/sched.h ?
>>
>> Yes, I noticed that sched.h is already included, but I still encountered
>> an "undeclared CLONE_NEWTIME" error during compilation.
> 
> Must be that your sched.h is not up-to-date I guess. On my side I have:
> 
> /usr/include/linux/sched.h:#define CLONE_NEWTIME    0x00000080    /* New 
> time namespace */
> /usr/include/bits/sched.h:#define CLONE_NEWTIME    0x00000080      /* 
> New time namespace */
> 
> And
> 
> /usr/include/sched.h:#include <bits/sched.h>
> 
> 
>>
>> Error I got:
>> CC       vdso_test_getrandom
>> vdso_test_getrandom.c: In function ‘kselftest’:
>> vdso_test_getrandom.c:257:29: error: ‘CLONE_NEWTIME’ undeclared (first
>> use in this function); did you mean ‘CLONE_NEWPID’?
>>    257 |         ksft_assert(unshare(CLONE_NEWTIME) == 0);
>>        |                             ^~~~~
>> vdso_test_getrandom.c:47:20: note: in definition of macro ‘ksft_assert’
>>     47 |         do { if (!(condition)) ksft_exit_fail_msg("Assertion
>> failed: %s\n", #condition); } while (0)
>>        |                    ^~~~~
>> vdso_test_getrandom.c:257:29: note: each undeclared identifier is
>> reported only once for each function it appears in
>>    257 |         ksft_assert(unshare(CLONE_NEWTIME) == 0);
>>        |                             ^~~~~
>> vdso_test_getrandom.c:47:20: note: in definition of macro ‘ksft_assert’
>>     47 |         do { if (!(condition)) ksft_exit_fail_msg("Assertion
>> failed: %s\n", #condition); } while (0)
>>        |                    ^~~~~
>> make[1]: * [../lib.mk:222:
>> /home/purva/linux/tools/testing/selftests/vDSO/vdso_test_getrandom] 
>> Error 1
>> make[1]: Leaving directory 
>> '/home/purva/linux/tools/testing/selftests/vDSO'
>>
>> I found that CLONE_NEWTIME is declared in both sched.h and
>> linux/sched.h. Since sched.h was already included, it was surprising
>> that the error persisted. Adding linux/sched.h as a header resolved the
>> issue, and the selftests compiled successfully after that.
> 
> Can you recheck that the sched.h that contains CLONE_NEWTIME is really 
> the one used by your compiler ?

I rechecked the header files on my system and got the following output:

purva@purva-IdeaPad-Gaming-3-15IHU6:~/linux$ grep -r "CLONE_NEWTIME" 
/usr/include/linux/sched.h /usr/include/bits/sched.h /usr/include/sched.h
/usr/include/linux/sched.h:#define CLONE_NEWTIME	0x00000080	/* New time 
namespace */

This shows that the definition of CLONE_NEWTIME exists only in 
/usr/include/linux/sched.h

> 
>>
>>>
>>> Did you properly build kernel headers before building selftests ?
>>
>> Yes, I ensured that I properly built the kernel headers before building
>> the selftests by following the documentation provided here
> 
> 
> At the end we should probably wonder if we want selftests to build with 
> old libc's that do not include latest defines. If we want to, then you 
> should probably replace sched.h by linux/sched.h . I'm not sure about 
> what to do really.

Since sched.h from libc may not define CLONE_NEWTIME on older systems, 
should we explicitly include linux/sched.h to ensure compatibility with 
older libc versions?

Thank you for your time and feedback!

Best regards,
Purva Yeshi