Re: [PATCH] Use struct for time_value_t and define seconds as long_integer_t.

[Samuel Thibault]

Flavio Cruz, le sam. 17 déc. 2022 16:30:45 -0500, a ecrit:
> On Sat, Dec 17, 2022 at 03:24:52AM +0100, Samuel Thibault wrote:
> > Applied, thanks!
> > 
> > Flavio Cruz, le ven. 16 déc. 2022 21:06:25 -0500, a ecrit:
> > > On 64 bit kernels, seconds will be 64 bits long and won't suffer from
> > > the 2038 problem. We also add a new type rpc_time_value_t to handle the
> > > conversion between 32 bit userland and 64 bit kernel.
> > 
> > Yes, I was thinking that we'd just switch to 64bit time with the 64bit
> > kernel.
> > 
> > We have the same issue with the device_read/write operations, for which
> > we'd want to make the offset 64bit.
> 
> Can you expand a bit on this? Are you referring to recnum_t?

I was referring to whatever is used to express offsets on disks :)

Apparently device_read uses recnum_t indeed, we want to make sure we get
it to 64bits.

> I see that we also use mach_msg_type_number_t to pass the size but that is a
> similar problem as io_read.

Yes, and I doubt we really need to expose 4GiB reads/writes to drivers.

Samuel