This is the mail archive of the libc-alpha@sourceware.org mailing list for the glibc project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [PATCH] Add ifunc memcpy and memmove for aarch64


On Tue, Mar 14, 2017 at 11:45 AM, Szabolcs Nagy <szabolcs.nagy@arm.com> wrote:
> On 01/03/17 18:48, Steve Ellcey wrote:
>> On Thu, 2017-02-23 at 22:23 +0530, Siddhesh Poyarekar wrote:
>>> On Thursday 23 February 2017 10:12 PM, Steve Ellcey wrote:
>>>>
>>>> Just to be clear, the earlier patch would be this one, right?
>>>>
>>>> https://sourceware.org/ml/libc-alpha/2017-02/msg00175.html
>>> Yes.  It does everything necessary to get multiarch enabled and working
>>> correctly for kernels that support it.  The hwcap_mask feature is
>>> something that can be added on top and I can commit to doing that within
>>> this release.
>>>
>>> Siddhesh
>>
>> Ping.  Does anyone have any comments or objections to this patch
>> that enables IFUNC on aarch64?
>
> the patch looks ok, with HWCAP in bits/hwcap.h instead of
>
> +/* Needed here until this gets into kernel sources.  */
> +#ifndef HWCAP_CPUID
> +# define HWCAP_CPUID            (1 << 11)
> +#endif
>
> the hwcap value is not yet in linux v4.10, but already
> allocated, if we are committed to this value then i
> think it's better to only have it in one place.
> you may need to include bis/hwcap.h in some files.
>
> i was not sure if we should wait for this to be in a
> linux release, but i guess the value won't change now.

It is in 4.11rc1 though.  Is that ok enough?

>
> an unrelated issue i was wondering about is how the
> upcoming ilp32 ifunc resolver will receive the hwcap
> argument: will it only see 32bit (unsigned long hwcap)
> or 64 bits (with different ifunc resolver prototype)
> and if there is something in this area that needs to
> be changed before ifuncs are used.

Right now, the hwcap is not using the full 64bit.  But the kernel will
have to split it into hwcap and hwcap2 anyways when we hit that point.
Most likely we should have the ifunc resolver take an uint64_t now
rather than latter.

Thanks,
Andrew Pinski

>


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]