[RFC v3 05/23] sysdeps/timespec_get: Use clock_gettime64 if avaliable

Arnd Bergmann arnd@arndb.de
Wed Jul 17 08:54:00 GMT 2019


On Wed, Jul 17, 2019 at 10:41 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * Arnd Bergmann:
>
> > I understand that it would be much slower on that particular combination,
> > I just can't think of anyone who would run into this in practice outside of
> > validation testing that makes sure glibc does run this way.
>
> I think it will happen if you run a glibc which has been built against
> current kernel headers (and with this patch or something like it) on a
> host which has an older kernel.  Due to the way the patch has been
> written, it will also impact legacy applications which call the 32-bit
> functions, not just 32-bit applications which use the 64-bit interfaces.

Ah right, makes sense.

> I could be totally wrong about this, or our disagreement could be about
> the relevance of this scenario.  Not sure which.
>
> (I think supporting old host kernels is important to our users.)

No, I was just wrong and had only thought about applications calling
directly into clock_gettime() while using a 64-bit time_t, but not
the case of glibc-internal functions calling __clock_gettime64()
for convenience, which is something I even advocated for in one of
my other comments.

       Arnd



More information about the Libc-alpha mailing list