This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: RFC: Check IFUNC_XXX to enable/disable ifunc feature
- From: "H.J. Lu" <hjl dot tools at gmail dot com>
- To: "Carlos O'Donell" <carlos at redhat dot com>
- Cc: GNU C Library <libc-alpha at sourceware dot org>, Siddhesh Poyarekar <sid at reserved-bit dot com>
- Date: Sun, 26 Jun 2016 15:08:05 -0700
- Subject: Re: RFC: Check IFUNC_XXX to enable/disable ifunc feature
- Authentication-results: sourceware.org; auth=none
- References: <CAMe9rOq1KakWO6bTb7LORhT1+3LRBoxtzV44NPstusAmy2+j_w at mail dot gmail dot com> <466674fe-b5b3-e254-2350-3aca10c25d81 at redhat dot com>
On Sun, Jun 26, 2016 at 11:55 AM, Carlos O'Donell <carlos@redhat.com> wrote:
> On 06/26/2016 10:50 AM, H.J. Lu wrote:
>> Hi,
>>
>> I'd like to check IFUNC_XXX environmental variable to enable/disable
>> the ifunc feature, XXX. That is
>>
>> 1. IFUNC_XXX=1 to enable XXX.
>> 2. IFUNC_XXX=0 to disable XXX.
>>
>> Any comments?
>
> How would it work? Does it just mean that the IFUNC selector bypasses
> the disabled choices?
It will override bits in dl_x86_cpu_features.
> Also it does not meet Tunables consensus:
> "Tunable namespace should be clearly defined"
> https://sourceware.org/glibc/wiki/TuningLibraryRuntimeBehavior?highlight=%28Tunables%29
>
> Should perhaps be "GLIBC_<VER>_IFUNC_<ARCH>_<PUBLIC_IFUNC_NAME>"?
They won't be version-dependent. How about
GLIBC_IFUNC_X86=sse2=0:....
or
GLIBC_IFUNC=sse2=0:....
since IFUNC is target dependent.
--
H.J.