[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <1456391333.3709.8.camel@gmail.com>
Date: Thu, 25 Feb 2016 10:08:53 +0100
From: Mike Galbraith <umgwanakikbuti@...il.com>
To: Ingo Molnar <mingo@...nel.org>
Cc: Andy Lutomirski <luto@...capital.net>,
"H. Peter Anvin" <hpa@...or.com>,
Peter Zijlstra <peterz@...radead.org>,
Brian Gerst <brgerst@...il.com>,
Thomas Gleixner <tglx@...utronix.de>,
Linus Torvalds <torvalds@...ux-foundation.org>,
Andrew Lutomirski <luto@...nel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
Borislav Petkov <bp@...en8.de>,
Denys Vlasenko <dvlasenk@...hat.com>,
"linux-tip-commits@...r.kernel.org"
<linux-tip-commits@...r.kernel.org>,
Frédéric Weisbecker <fweisbec@...il.com>
Subject: Re: [tip:x86/urgent] x86/entry/32: Add an ASM_CLAC to
entry_SYSENTER_32
On Thu, 2016-02-25 at 09:40 +0100, Ingo Molnar wrote:
> * Mike Galbraith <umgwanakikbuti@...il.com> wrote:
>
> > On Thu, 2016-02-25 at 09:14 +0100, Ingo Molnar wrote:
> >
> > > But but ... 'context tracking' is not really something that a regular distro
> > > kernel cares about much - it's a nohz-full special AFAICS.
>
> Let me qualify that: with the timer code maintenance hat on I really love all nohz
> variants (the deeper the better), but now I have my x86 maintainer hat on, and as
> such I'm really annoyed at those nohz folks adding overhead to the syscall hot
> path! ;-)
>
> > (psst.. distros are shipping it)
>
> Yeah, indeed, Fedora does - but AFAICS:
>
> fomalhaut:~> grep NO_HZ /boot/config-4.1.13-100.fc21.x86_64
> CONFIG_NO_HZ_COMMON=y
> # CONFIG_NO_HZ_IDLE is not set
> CONFIG_NO_HZ_FULL=y
> # CONFIG_NO_HZ_FULL_ALL is not set
> # CONFIG_NO_HZ_FULL_SYSIDLE is not set
> CONFIG_NO_HZ=y
> CONFIG_RCU_FAST_NO_HZ=y
>
> ... which won't result in actual full-nohz CPUs unless you boot it with a special
> boot parameter, right?
Yeah, you have to manually enable it unless you (in a suicidal moment)
enable CONFIG_NO_HZ_FULL_ALL.
> What is the easiest way to query which/how many CPUs are in nohz-full mode and do
> context tracking? I somehow thought /proc/timer_* had that info, but that does not
> appear to be the case.
/sys/devices/system/cpu/nohz_full
-Mike
Powered by blists - more mailing lists