[PATCH 20/28] aarch64: Add glibc-hwcaps support

Szabolcs Nagy szabolcs.nagy@arm.com
Wed Oct 14 15:09:51 GMT 2020


The 10/14/2020 16:44, Florian Weimer wrote:
> * Szabolcs Nagy:
> 
> > The 10/14/2020 11:15, Adhemerval Zanella via Libc-alpha wrote:
> >> On 14/10/2020 11:08, Florian Weimer wrote:
> >> > * Adhemerval Zanella via Libc-alpha:
> >> >> On 01/10/2020 13:33, Florian Weimer via Libc-alpha wrote:
> >> >>> At this point, only the "atomics" subdirectory is available,
> >> >>> for libraries built using LSE atomics.
> >> >>
> >> >> LGTM, thanks.
> >> >>
> >> >> Reviewed-by: Adhemerval Zanella  <adhemerval.zanella@linaro.org>
> >> > 
> >> > Thanks.  I didn't want to skip AArch64, but it's really more of a stub
> >> > implementation:
> >> > 
> >> >>> +const char _dl_hwcaps_subdirs[] = "atomics";
> >> > 
> >> > Does this still make sense with GCC defaulting to -moutline-atomics for
> >> > ARMv8a?  That's what I wonder.
> >> > 
> >> > Eventually, we need to define some levels for AArch64, and I'm not sure
> >> > to what extent they would align with the official 8.X versions.  To me,
> >> > they look more like a bouquet you can choose from.
> >> 
> >> I am not sure how ARM maintainers would like to handle it, either by
> >> defining based on ARMv8.x revisions, a subsets of HWCAP capabilities,
> >> or by not defining anything.
> >> 
> >> For now I think the atomic makes sense because of -moutline-atomics,
> >> although it is essentially ARMv8.1.
> >
> > the atomics path logic was added in glibc 2.28 in
> > commit 397c54c1afa531242602fe3ac7bb47eff0e909f9
> > (see the reasoning in the commit message).
> >
> > since then this is public api that users may rely
> > on (although likely there are not many users..)
> >
> > so i dont want to remove it (and yes it's not
> > very helpful now that outline-atomics is the
> > default in gcc).
> 
> This patch adds glibc-hwcaps/atomics as a new subdirectory.  The whole
> series does not remove the legacy directories.
> 
> Based on what you wrote, I should drop this patch, right?

ah i missed that legacy is handled independently.

in that case, yes please drop this.


More information about the Libc-alpha mailing list