Re: [PATCH v5 00/23] ILP32 for ARM64

[Arnd Bergmann <arnd@arndb.de>]

On Thursday 01 October 2015 22:15:20 Yury Norov wrote:

> Regarding time_t, it, of course, doesn't takes much time to make it
> 32-bit, but I think 64 bit is better because of Y2038. X32 and mips
> n32 has time_t 64-bit (and ppc, not sure), and that's OK for them.

I'm pretty sure that n32 has 32-bit time_t, and we know that it still
causes real-world problems on x32: socket timestamps, v4l, alsa and
other subsystems all have bugs in this area that are hard to fix.

> That's OK for BSD as well. The objection may come from users of ABI,
> complaining portability problems, but I found no such complains in
> public discussions.
>
> Nevertheless, as I told, I do not see any problem to rework time_t.
> But some arguments supporting this decision are appreciated.
> 
> The downside of 32 bit time_t is that we still face Y2038 problem,
> but that's the other story fixing it.

The main reason for 32-bit time_t is compatibility with existing
ioctls (also getsockopts and some others), and having a sane way
for fixing them. We cannot change compat_time_t to be 64-bit
without breaking arm32 compat mode, and we can't use the native
64-bit ioctl implementation on ARM64/ILP32 because that breaks
all interfaces that pass 'long' or a pointer.

This means drivers that currently pass a time_t (or timeval, timespec
etc) need to not only have a compat_ioctl handler to convert it,
they also need to check whether which of the two compat modes they
are talking to. This is a mess to add (I know, because I'm working
on this for y2038 compliance for normal 32-bit mode), and making
the two behave differently makes it even harder to get right for
all cases.

> __kernel_long_t is the same. Now it's 64 bits length. Compatibility
> may suffer, but, again, there're no complains, and in long run it
> looks better.

__kernel_long_t isn't actually used that much, and rarely used in
places where it matters. The idea was to be able to reuse the
native syscalls rather than the compat syscall calls, but that
comes with the downside of defining the ABI in a way that is
incompatible with all other 32-bit user space.

Having a 64-bit __kernel_off_t is similar to the 64-bit time_t:
a good idea in principle, but it breaks device drivers that
expect user space to pass 32-bit arguments. For any interface
that really needs 64-bit data, we have to fix it for all
32-bit architectures, and we're better off avoiding special
cases.

	Arnd