This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [RFC v2 08/20] sysdeps/wait: Use waitid if avaliable
- From: Joseph Myers <joseph at codesourcery dot com>
- To: Florian Weimer <fweimer at redhat dot com>
- Cc: Andreas Schwab <schwab at suse dot de>, Arnd Bergmann <arnd at arndb dot de>, Zack Weinberg <zackw at panix dot com>, Alistair Francis <alistair dot francis at wdc dot com>, GNU C Library <libc-alpha at sourceware dot org>
- Date: Thu, 25 Jul 2019 15:49:09 +0000
- Subject: Re: [RFC v2 08/20] sysdeps/wait: Use waitid if avaliable
- References: <cover.1561421042.git.alistair.francis@wdc.com> <CAKCAbMhyUiEZcfg-BNL_sqUAFmtEWCEgoyUwMGOdB_jLbvKJ6Q@mail.gmail.com> <87ftnx6i0m.fsf@oldenburg2.str.redhat.com> <CAK8P3a39tr-qbG4X8SnM=3M6VfV+LzOvE6pVBqByN0YcPkf42w@mail.gmail.com> <877e9950hj.fsf@oldenburg2.str.redhat.com> <CAK8P3a0P1f=922peKELqanZqSY3MAdNiUK70UvZaxy75pS-2Qw@mail.gmail.com> <CAK8P3a2hj24voTemN54gBKay+cKU3vd_dr=KVacWjrO5foH3Bg@mail.gmail.com> <87a7e522wb.fsf@oldenburg2.str.redhat.com> <CAK8P3a3b32CNaXfR1w8cJuGS1H7qPTKLZc0BDxf-oW20i7aXvw@mail.gmail.com> <87r27hzrjd.fsf@oldenburg2.str.redhat.com> <CAK8P3a2HC_Ki0tf9B_A3Tm5w9zqrVLO8JG8u_ohc8aoSEY5okw@mail.gmail.com> <87ef3hzqjw.fsf@oldenburg2.str.redhat.com> <CAK8P3a2QW7WNuK0_zu44_fp3qWh7iL2V6Mae+zx+iLjUP7CitQ@mail.gmail.com> <8736jwjqkq.fsf@oldenburg2.str.redhat.com> <mvm36jwux93.fsf@suse.de> <878st8u3qp.fsf@oldenburg2.str.redhat.com> <mvmlfx8emc2.fsf@suse.de> <87bly4snx5.fsf@oldenburg2.str.redhat.com>
On Mon, 8 Jul 2019, Florian Weimer wrote:
> > I'm talking about legacy 32-bit architectures.
>
> Ah. Yes, I one would hope that.
>
> But there has been talk of a kernel option to remove it even there, and
> a suggestion that glibc would try the 64-bit time_t system call first,
> to see if it is available, and prefer that.
>
> I think this would be quite … wrong.
glibc should generally use the 64-bit time_t system call first on legacy
32-bit architectures (and especially in cases such as futex where there is
a large amount of logic in glibc at a higher level in the syscall, which
it is particularly important to avoid duplicating for different time_t
choices), not because of systems with the 32-bit syscall removed where it
was previously part of the kernel ABI expected by glibc (those are not
something reasonably supportable), but to avoid duplication and code bloat
at both the source and binary level and to have a single consistent design
(32-bit functions wrap the corresponding 64-bit ones, in all cases where
the function is defined in C rather than through syscalls.list) rather
than attempting to define what is or is not a sufficiently trivial
function implementation that duplication is not a problem.
--
Joseph S. Myers
joseph@codesourcery.com