This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [Patch] Use VDSO interface for gettimeofday on aarch64
- From: Siddhesh Poyarekar <siddhesh at gotplt dot org>
- To: Andrew Pinski <pinskia at gmail dot com>
- Cc: sellcey at cavium dot com, libc-alpha <libc-alpha at sourceware dot org>
- Date: Fri, 11 May 2018 11:21:12 +0530
- Subject: Re: [Patch] Use VDSO interface for gettimeofday on aarch64
- References: <1525975253.28825.227.camel@cavium.com> <61da438e-3345-9ecb-d981-807ed9f67e63@gotplt.org> <CA+=Sn1marqZc1hg1_YtaZiZWSEZSkzs1U_ei8KUA3xkET8t7aQ@mail.gmail.com>
On 05/11/2018 10:13 AM, Andrew Pinski wrote:
That does not measure the difference here. We were already using the
VDSO before; just differently. Before we had a PLT that calls into
the libc and would do then an check on the VDSO and then (another)
indirect call (to the VDSO). After we just have an PLT (indirect)
call into the VDSO directly. Basically saving the check and an extra
indirect call (the indirect call can be costly if not predicted
correctly).
Ah OK, got it.
Walking the ifuncs won't benchmark the difference that is gotten here.
You will see the biggest benefit from Apache Bench and running Apache
as the web server (maybe any other webserver which calls gettimeofday
a lot as you are producing a date for each socket connection). And
yes this is considered a standard benchmark when it comes to CDN.
OK, thanks. Looks good to me too then. The benchmark would still be
nice to have (let me know if y'all don't intend to write it so that I
can put it in my list of things to do), but not a blocker for this patch.
Siddhesh