LKML Archive on lore.kernel.org
 help / color / Atom feed
From: Russell King - ARM Linux <linux@armlinux.org.uk>
To: Finn Thain <fthain@telegraphics.com.au>,
	John Stultz <john.stultz@linaro.org>
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>,
	linux-m68k@lists.linux-m68k.org, linux-kernel@vger.kernel.org,
	linux-arm-kernel@lists.infradead.org
Subject: Re: [RFC PATCH 01/13] arm: Fix mutual exclusion in arch_gettimeoffset
Date: Wed, 14 Nov 2018 12:05:00 +0000
Message-ID: <20181114120500.GE6567@n2100.armlinux.org.uk> (raw)
In-Reply-To: <d1559db8eddb5b06747b152ae69b63c3aa66ed09.1541995959.git.fthain@telegraphics.com.au>

On Mon, Nov 12, 2018 at 03:12:39PM +1100, Finn Thain wrote:
> Implementations of arch_gettimeoffset are generally not re-entrant
> and assume that interrupts have been disabled. Unfortunately this
> pre-condition got broken in v2.6.32.

To me, it looks way worse than what you think.

The original code sequence before conversion to arch_gettimeoffset()
was:

1. lock (inc. disabling IRQs)
2. read gettimeoffset correction
3. add offset to current xtime
4. unlock

This means there is no chance for a timer interrupt to happen between
reading the current offset and adding it to the current kernel time.
If the timer does roll over, then the interrupt will remain pending,
so the whole "read offset and add to xtime" is one atomic block and
we can use the pending interrupt to indicate whether the timer has
rolled over and apply the appropriate offset correction.

With the arch_gettimeoffset() code, that changes:

1. read clocksource (which will be jiffies)
2. compute clocksource delta
3. increment nanoseconds
4. add gettimeoffset correction

If we receive a timer interrupt part-way through this, for example,
between 2 and 3, then jiffies will increment (via do_timer()) and the
interrupt will be cleared.  This means that the check in
ioc_timer_gettimeoffset() for a pending interrupt, for example, will
be false, and we will not know that an interrupt has happened.

So, unless I'm missing something, commit 5cfc8ee0bb51 well and truely
broke the accuracy of timekeeping on these platforms, and will result
in timekeeping spontaneously jumping backwards.  I had been wondering
why NTP had become less stable on EBSA110, but never investigated.

However, that still does not justify blowing away this functionality
without replacement, as Finn has been proposing.

-- 
RMK's Patch system: http://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up

  parent reply index

Thread overview: 53+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2018-11-12  4:12 [RFC PATCH 00/13] m68k: Drop arch_gettimeoffset and adopt clocksource API Finn Thain
2018-11-12  4:12 ` [RFC PATCH 08/13] m68k: atari: Convert to " Finn Thain
2018-11-12  4:12 ` [RFC PATCH 04/13] m68k: mac: Clean up unused timer definitions Finn Thain
2018-11-12  4:12 ` [RFC PATCH 10/13] m68k: hp300: Convert to clocksource API Finn Thain
2018-11-12  4:12 ` [RFC PATCH 11/13] m68k: mac: " Finn Thain
2018-11-12  4:12 ` [RFC PATCH 01/13] arm: Fix mutual exclusion in arch_gettimeoffset Finn Thain
2018-11-12  8:34   ` Christoph Hellwig
2018-11-13  3:39     ` Finn Thain
2018-11-13  9:20       ` Russell King - ARM Linux
2018-11-13 21:55         ` Finn Thain
2018-11-13 23:43           ` Russell King - ARM Linux
2018-11-14  1:35             ` Michael Schmitz
2018-11-14  7:58               ` Russell King - ARM Linux
2018-11-15  1:34                 ` Michael Schmitz
2018-11-14  3:17             ` Finn Thain
2018-11-14 14:16               ` Russell King - ARM Linux
2018-11-14 14:58                 ` Geert Uytterhoeven
2018-11-14 18:11                   ` Russell King - ARM Linux
2018-11-15  4:12                 ` Finn Thain
2018-11-16 17:47                   ` Russell King - ARM Linux
2018-11-16 22:49                     ` Finn Thain
2018-11-17  3:00                       ` Michael Schmitz
2018-11-14 12:05   ` Russell King - ARM Linux [this message]
2018-11-12  4:12 ` [RFC PATCH 02/13] m68k: " Finn Thain
2018-11-12  4:12 ` [RFC PATCH 07/13] m68k: amiga: Convert to clocksource API Finn Thain
2018-11-12  9:02   ` Geert Uytterhoeven
2018-11-12  9:21     ` Finn Thain
2018-11-12  4:12 ` [RFC PATCH 09/13] m68k: bvme6000: " Finn Thain
2018-11-12  4:12 ` [RFC PATCH 06/13] m68k: Drop ARCH_USES_GETTIMEOFFSET Finn Thain
2018-11-12  9:00   ` Geert Uytterhoeven
2018-11-12  9:06     ` Finn Thain
2018-11-13  2:29       ` Michael Schmitz
2018-11-13  3:14         ` Finn Thain
2018-11-13  4:50           ` Michael Schmitz
2018-11-13  6:15             ` Finn Thain
2018-11-13  8:24               ` Michael Schmitz
2018-11-13 22:11                 ` Finn Thain
2018-11-14  1:08                   ` Michael Schmitz
2018-11-14  2:58                     ` Michael Schmitz
2018-11-14 23:54                       ` Michael Schmitz
2018-11-15  4:37                         ` Finn Thain
2018-11-15  6:35                         ` Michael Schmitz
2018-11-16  0:04                           ` Finn Thain
2018-11-12  4:12 ` [RFC PATCH 13/13] m68k: mvme16x: Convert to clocksource API Finn Thain
2018-11-12 18:37   ` Thomas Gleixner
2018-11-13  0:11     ` Finn Thain
2018-11-13 22:04       ` Finn Thain
2018-11-13 22:10         ` [RFC PATCH 13/13] m68k: mvme16x: Convert to clocksource APIy Thomas Gleixner
2018-11-13 22:33           ` Finn Thain
2018-11-18 17:12             ` Thomas Gleixner
2018-11-12  4:12 ` [RFC PATCH 03/13] m68k: mac: Fix VIA timer counter accesses Finn Thain
2018-11-12  4:12 ` [RFC PATCH 12/13] m68k: mvme147: Convert to clocksource API Finn Thain
2018-11-12  4:12 ` [RFC PATCH 05/13] m68k: apollo, q40, sun3, sun3x: Remove arch_gettimeoffset implementations Finn Thain

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20181114120500.GE6567@n2100.armlinux.org.uk \
    --to=linux@armlinux.org.uk \
    --cc=arnd@arndb.de \
    --cc=daniel.lezcano@linaro.org \
    --cc=fthain@telegraphics.com.au \
    --cc=geert@linux-m68k.org \
    --cc=john.stultz@linaro.org \
    --cc=linux-arm-kernel@lists.infradead.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux-m68k@lists.linux-m68k.org \
    --cc=schivers@csc.com.au \
    --cc=tglx@linutronix.de \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link

LKML Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/lkml/0 lkml/git/0.git
	git clone --mirror https://lore.kernel.org/lkml/1 lkml/git/1.git
	git clone --mirror https://lore.kernel.org/lkml/2 lkml/git/2.git
	git clone --mirror https://lore.kernel.org/lkml/3 lkml/git/3.git
	git clone --mirror https://lore.kernel.org/lkml/4 lkml/git/4.git
	git clone --mirror https://lore.kernel.org/lkml/5 lkml/git/5.git
	git clone --mirror https://lore.kernel.org/lkml/6 lkml/git/6.git
	git clone --mirror https://lore.kernel.org/lkml/7 lkml/git/7.git
	git clone --mirror https://lore.kernel.org/lkml/8 lkml/git/8.git
	git clone --mirror https://lore.kernel.org/lkml/9 lkml/git/9.git
	git clone --mirror https://lore.kernel.org/lkml/10 lkml/git/10.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 lkml lkml/ https://lore.kernel.org/lkml \
		linux-kernel@vger.kernel.org
	public-inbox-index lkml

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.kernel.vger.linux-kernel


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git