[PATCH v7 0/3] y2038: Linux: Introduce __clock_settime64 function

[Joseph Myers]

On Tue, 17 Sep 2019, Lukasz Majewski wrote:

> - New 32 bits glibc ports (like RISC-V 32) will get __TIMESIZE ==
>   64 (__WORDSIZE == 32) and no need to define the -D_TIME_BITS=64
>   during the compilation. They will just get 64 bit time API support
>   from the outset.

Yes, at least if such ports wish to use 64-bit time; I don't think we've 
really discussed if we want to *require* 64-bit time for future ports 
(e.g. the next revised resubmissions of the ARC and NDS32 ports).  
Certainly the work required right now for ARC or NDS32 to use 64-bit time 
would be significantly more than the work for RV32 (because they also 
support older kernel versions without the 64-bit-time syscalls, so all the 
Y2038 work for fallback at runtime to older syscalls becomes relevant), 
unless they decide on 5.1 or later as minimum kernel version.

> - Already supported 32 bits architectures (like armv7-a with __WORDSIZE
>   == 32) will keep __TIMESIZE == 32 and require -D_TIME_BITS=64 for
>   compilation. 

Yes.

>   After glibc sets the minimal supported kernel version to 5.1 and all
>   conversions for syscalls to support 64 bit time API are done the
>   __TIMESIZE will be set to 64 and -D_TIME_BITS=64 will not be required
>   anymore for compilation.

No.  __TIMESIZE means the size of time_t in the unsuffixed ABIs in glibc, 
not the _TIME_BITS-dependent size of time_t in the current compilation.  
We hope in future to make _TIME_BITS=64 the default and only API supported 
for new compilations (which is independent of what the minimum kernel 
version is), but __TIMESIZE would still be 32, because the unsuffixed ABIs 
would remain compatible with existing binaries using 32-bit time.

> Ok. So then we shall keep the condition:
> 
> #if __WORDSIZE == 64 \
>   || (defined __SYSCALL_WORDSIZE && __SYSCALL_WORDSIZE == 64)
> # define __timespec64 timespec
> #else

No.  __timespec64 should be defined to timespec whenever __TIMESIZE == 64.  
The timespec to which it is defined, in the public header, would gain 
padding.

The condition

#if __WORDSIZE == 64 \
  || (defined __SYSCALL_WORDSIZE && __SYSCALL_WORDSIZE == 64)

is correct as a condition for struct timespec (in the public header) *not* 
to have padding.

-- 
Joseph S. Myers
joseph@codesourcery.com