Re: [RFC PATCH 06/13] m68k: Drop ARCH_USES_GETTIMEOFFSET

From: Michael Schmitz <schmitzmic@gmail.com>
To: Finn Thain <fthain@telegraphics.com.au>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>,
	Arnd Bergmann <arnd@arndb.de>,
	Stephen N Chivers <schivers@csc.com.au>,
	Thomas Gleixner <tglx@linutronix.de>,
	Daniel Lezcano <daniel.lezcano@linaro.org>,
	John Stultz <john.stultz@linaro.org>,
	linux-m68k <linux-m68k@lists.linux-m68k.org>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Philip Blundell <philb@gnu.org>,
	Joshua Thompson <funaho@jurai.org>
Subject: Re: [RFC PATCH 06/13] m68k: Drop ARCH_USES_GETTIMEOFFSET
Date: Tue, 13 Nov 2018 21:24:30 +1300	[thread overview]
Message-ID: <0054c37f-ef34-e892-ff1d-02968ab529fa@gmail.com> (raw)
In-Reply-To: <alpine.LNX.2.21.1811131700010.366@nippy.intranet>

Hi Finn,

Am 13.11.2018 um 19:15 schrieb Finn Thain:
> On Tue, 13 Nov 2018, Michael Schmitz wrote:
>
>>
>>> (It appears that a QEMU-emulated Mac does not benefit from having a
>>> clocksource that's more accurate than the 'jiffies' clocksource, in
>>> spite of "clocksource: Switched to clocksource via1".)
>>
>> With the current code, kernel log timestamps have 10 ms resolution on
>> Atari. Time resolution of times reported by initcall_debug is roughly 40
>> us. I'd expect that changes with falling back to jiffies only. Might be
>> worth a try on QEMU Mac.
>
> The initcall debug output shows the same precision as my earlier tests
> with userland. The VIA timer only gets updated when QEMU wants to update
> it, which seems to be about every 9765 us. Hence, using the 'jiffies'
> clocksource would be effectively no regression for a virtual Mac.

Running a recent kernel under ARAnyM shows 40 ns resolution so the Atari 
hardware emulation is a little more complete.

Using that for initial tests, I can confirm that timer resolution is 
reduced to 10ms after patch 6, and gets restored to 40ns after applying 
the full series (once clocksource_init runs, that is - the first part of 
the boot is at 10ms resolution only, a regression compared to with use 
of arch_gettimeoffset).

Unfortunately, I can't log in at the console with all patches applied. I 
immediately get the 'login timeout exceeded' message. Weird...

Cheers,

	Michael