Date: Thu, 11 Jan 2018 22:23:48 +0100
From: Willy Tarreau <w@1wt.eu>
To: Alan Cox <gnomes@lxorguk.ukuu.org.uk>
Cc: Josh Poimboeuf <jpoimboe@redhat.com>,
        Olivier Galibert <galibert@pobox.com>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        Borislav Petkov <bp@alien8.de>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Alexei Starovoitov <alexei.starovoitov@gmail.com>,
        Andy Lutomirski <luto@kernel.org>,
        Peter Zijlstra <peterz@infradead.org>,
        LKML <linux-kernel@vger.kernel.org>, X86 ML <x86@kernel.org>,
        Brian Gerst <brgerst@gmail.com>,
        Ingo Molnar <mingo@kernel.org>,
        Thomas Gleixner <tglx@linutronix.de>,
        "H. Peter Anvin" <hpa@zytor.com>,
        Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Kees Cook <keescook@chromium.org>
Subject: Re: [RFC PATCH v2 6/6] x86/entry/pti: don't switch PGD on when
 pti_disable is set
Message-ID: <20180111212348.GA15528@1wt.eu>
References: <20180111182147.masunghp5km6igjq@ast-mbp.dhcp.thefacebook.com>
 <20180111183207.dah7imbuvuhvrrk6@treble>
 <a8b043b5-84a9-af92-ffcc-4b9c68f6917d@linux.intel.com>
 <CA+55aFyOnk1rd+iC_gvXpjzsv5JtJ_YXkFMukrn2WnZUS9mVmw@mail.gmail.com>
 <4e32af93-f632-ae17-eed4-c7023c1b9cc5@linux.intel.com>
 <20180111190700.y4hultq3awxuw6fj@pd.tnic>
 <6144aefc-767f-ff36-b424-f63f4bd4c579@linux.intel.com>
 <CAMFK0gvZ2fPTFFtTeW8A0_gmEovnHnozjxmQ8niHGkxVnpgFow@mail.gmail.com>
 <20180111192601.cck3eq26vorybudk@treble>
 <20180111193431.7e2d7f63@alans-desktop>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180111193431.7e2d7f63@alans-desktop>
User-Agent: Mutt/1.6.1 (2016-04-27)
Sender: linux-kernel-owner@vger.kernel.org

On Thu, Jan 11, 2018 at 07:34:31PM +0000, Alan Cox wrote:
> On Thu, 11 Jan 2018 13:26:01 -0600
> Josh Poimboeuf <jpoimboe@redhat.com> wrote:
> 
> > On Thu, Jan 11, 2018 at 08:19:35PM +0100, Olivier Galibert wrote:
> > > Wouldn't the time taken by an easy syscall like getuid be a clear indicator?  
> > 
> > I think only if you had a baseline measurement to compare against.
> 
> getuid can also be cached by a smart libc. getppid() is what micro
> benchmarks seem to favour.

I initially tried getpid() which I found to be cached by glibc, but I
switched to write(-1, "a", 1) in the example in the cover letter and
the test for 3 million runs roughly climbs from 200 ms to 900 ms under
kvm (with PCID this time, I didn't retest without).

Willy