[RFC PATCH 3/4] linux: consistently use time64 syscalls if available

Adhemerval Zanella Netto adhemerval.zanella@linaro.org
Wed Nov 11 13:56:09 GMT 2026



On 11/11/26 10:04, Arnd Bergmann wrote:
> On Wed, Nov 11, 2026, at 13:09, Adhemerval Zanella Netto wrote:
>> On 10/11/26 11:51, Arnd Bergmann wrote:
>>> From: Arnd Bergmann <arnd@arndb.de>
> 
>>> --- a/sysdeps/unix/sysv/linux/poll.c
>>> +++ b/sysdeps/unix/sysv/linux/poll.c
>>> @@ -27,6 +27,18 @@ __poll (struct pollfd *fds, nfds_t nfds, int timeout)
>>>  {
>>>  #ifdef __NR_poll
>>>    return SYSCALL_CANCEL (poll, fds, nfds, timeout);
>>> +#elif defined(__ASSUME_TIME64_SYSCALLS) && defined(__NR_ppoll_time64)
>>> +  struct __timespec64 timeout_ts;
>>> +  struct __timespec64 *timeout_ts_p = NULL;
>>> +
>>> +  if (timeout >= 0)
>>> +    {
>>> +      timeout_ts.tv_sec = timeout / 1000;
>>> +      timeout_ts.tv_nsec = (timeout % 1000) * 1000000;
>>> +      timeout_ts_p = &timeout_ts;
>>> +    }
>>> +
>>> +  return SYSCALL_CANCEL (ppoll_time64, fds, nfds, timeout_ts_p, NULL, 0);
>>
>> The could avoid duplicating the timeout conversion by using the usual:
>>
>>   #ifndef __NR_ppoll_time64
>>   # define __NR_ppoll_time64 __NR_ppoll
>>   #endif
>>
>> It is fine either way.
> 
> I don't think that would work here, since we need both cases
> depending on the expected kernel ABI version.
> 
>  - on __ASSUME_TIME64_SYSCALLS (building for new kernel versions)
>    we want __NR_ppoll_time64 and need to convert to __timespec64
>  - when building for older 32-bit kernels, the #ifndef is false
>    because both __NR_ppoll_time64 and __NR_ppoll are defined,
>    but we want to use __NR_ppoll with the time32 timespec.
> 
>> Missing ')'. More importantly, tst-clone3.c and i386/tst-bz21269.c are 
>> regular tests
>> and do not include <kernel-features.h>, so __ASSUME_TIME64_SYSCALLS is 
>> never defined there.
> 
> Is it ok to include that header here it then, or does this need to
> try call both syscalls?

I think we can just mimic what the internal futex code itself does. I think we can
eventually phase this out with my futex wrapper proposal.

> 
>> The clone3 needs Linux 5.3, so when tst-clone3* reach futex_wait, futex_time64
>> always exists.
> 
> Ok
> 
>    Arnd



More information about the Libc-alpha mailing list