Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 14 Jun 2018 17:55:48 +0800
From:   Feng Tang <feng.tang@intel.com>
To:     Thomas Gleixner <tglx@linutronix.de>
Cc:     Petr Mladek <pmladek@suse.com>, Ingo Molnar <mingo@kernel.org>,
        "H . Peter Anvin" <hpa@zytor.com>,
        Alan Cox <gnomes@lxorguk.ukuu.org.uk>,
        Peter Zijlstra <peterz@infradead.org>,
        linux-kernel@vger.kernel.org, alek.du@intel.com,
        len.brown@intel.com, feng.tang@intel.com
Subject: Re: [RFC 1/2] printk: Enable platform to provide a early boot clock
Message-ID: <20180614095548.5lbnpm3db2bu2gov@shbuild888>
References: <1527672059-6225-1-git-send-email-feng.tang@intel.com>
 <20180614083851.m6euwwl3xkdtjakd@shbuild888>
 <alpine.DEB.2.21.1806141107260.1581@nanos.tec.linutronix.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <alpine.DEB.2.21.1806141107260.1581@nanos.tec.linutronix.de>
User-Agent: NeoMutt/20170609 (1.8.3)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi Thomas,

On Thu, Jun 14, 2018 at 11:08:04AM +0200, Thomas Gleixner wrote:
> On Thu, 14 Jun 2018, Feng Tang wrote:
> > On Wed, May 30, 2018 at 05:20:58PM +0800, Feng Tang wrote:
> > > Currently printk timestamp mostly come from the sched_clock which
> > > depends on the clock setup, so there are many kernel logs started
> > > with "[    0.000000]   " before the clock is calibrated.
> > > 
> > > This patch will provide an debug option for specific platform to
> > > provide a early boot time clock, so that we can have time info in
> > > kernel log much earlier, which can show the time info for the early
> > > kernel boot, and make boottime tuning/optimization easier (boot time
> > > is critical for phone/tablet and embedded devices).
> > > 
> > > Capable platform only need to setup the "boot_printk_clock_fn"
> > > which could return time in nano seconds.
> > > 
> > > Together with a TSC patch on x86 system, we have easily captured
> > > some early boottime killer like unwind_init() which takes about
> > > 300ms in boot phase.
> > 
> > Hi Petr and all,
> > 
> > As the 2/2 tsc related patch is still under review/discussion, can
> > we consider taking this first? As this may benefit other archs.
> > For example, Intel Curie platform has an always-on 32KHz osc clock,
> > which is accurate but low frequency, and it could be used as
> > early printk timestamp until the high-resolution timer is initialized
> > and used as sched_clock. Don't know if ARM or other platforms
> > have similar use case.
> 
> Can we please _NOT_ add half sorted core stuff in a hurry?

Got it.

The status for the 2/2 tsc patch is Peter suggested to bring the tsc_init()
earlier, with that this 1/2 is not needed for x86. I followed Peter's
suggestion and worked out a hack version with which the tsc inits fine
and printk timestamp works much earlier (detail is in the 2/2 thread).

But there is one blocker, the native_sched_clock()(tsc.c) use one staic key
__use_tsc and sched_clock_cpu()(clock.c) use another "__sched_clock_stable",
which cannot be used so early, one problem is the static_branch_enable()
will use pageing related code while the paging is not setup ready yet.
I haven't got idea to solve this first problem regarding static key.

Thanks,
Feng