This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


[Xen-devel] [PATCH] Fix a NO_IDLE_HZ timer bug

To: Linux Kernel Mailing List <linux-kernel@xxxxxxxxxxxxxxx>, Andrew Morton <akpm@xxxxxxxx>, schwidefsky@xxxxxxxxxx, george@xxxxxxxxxx, Xen-devel <xen-devel@xxxxxxxxxxxxxxxxxxx>
Subject: [Xen-devel] [PATCH] Fix a NO_IDLE_HZ timer bug
From: Zachary Amsden <zach@xxxxxxxxxx>
Date: Thu, 18 May 2006 13:48:46 -0700
Delivery-date: Thu, 18 May 2006 13:54:34 -0700
Envelope-to: www-data@xxxxxxxxxxxxxxxxxx
List-help: <mailto:xen-devel-request@lists.xensource.com?subject=help>
List-id: Xen developer discussion <xen-devel.lists.xensource.com>
List-post: <mailto:xen-devel@lists.xensource.com>
List-subscribe: <http://lists.xensource.com/cgi-bin/mailman/listinfo/xen-devel>, <mailto:xen-devel-request@lists.xensource.com?subject=subscribe>
List-unsubscribe: <http://lists.xensource.com/cgi-bin/mailman/listinfo/xen-devel>, <mailto:xen-devel-request@lists.xensource.com?subject=unsubscribe>
Sender: xen-devel-bounces@xxxxxxxxxxxxxxxxxxx
User-agent: Thunderbird (X11/20060420)
This bug affects at least s390, Xen, and occurred during testing of NO_IDLE_HZ with our VMI patches. The bug is rather subtle and rare. Jiffies can be advanced by other CPUs, but if the current CPU has not processed timer softirqs in a while, the timer wheel can be behind jiffies, causing an overflow when comparing the next timer wheel expiration with the default high resolution timeout (no timeout), which is relative to jiffies.

It also looks like s390 has another bug. When compiling the 32-bit kernel, doesn't this computation overflow:

arch/s390/kernel/time.c, stop_hz_timer:274

        * This cpu is going really idle. Set up the clock comparator
        * for the next event.
       next = next_timer_interrupt();
       do {
               seq = read_seqbegin_irqsave(&xtime_lock, flags);
               timer = (__u64)(next - jiffies) + jiffies_64;
       } while (read_seqretry_irqrestore(&xtime_lock, seq, flags));

Since jiffies can advance between next_timer_interrupt and the read under xtime lock, next-jiffies could be negative. I would think you want to cast that to signed long instead of __u64, but I'm not totally qualified to talk about s390.

Under certain timing conditions, a race during boot occurs where timer ticks
are being processed on remote CPUs.  The remote timer ticks can increment
jiffies, and if this happens during a window when a timeout is very close to
expiring but a local tick has not yet been delivered, you can end up with

1) No softirq pending
2) A local timer wheel which is not synced to jiffies
3) No high resolution timer active
4) A local timer which is supposed to fire before the current jiffies value.

In this circumstance, the comparison in next_timer_interrupt overflows, because
the base of the comparison for high resolution timers is jiffies, but for the
softirq timer wheel, it is relative the the current base of the wheel

Signed-off-by: Zachary Amsden <zach@xxxxxxxxxx>

Index: linux-2.6.17-rc/kernel/timer.c
--- linux-2.6.17-rc.orig/kernel/timer.c 2006-05-18 13:32:22.000000000 -0700
+++ linux-2.6.17-rc/kernel/timer.c      2006-05-18 13:34:59.000000000 -0700
@@ -541,6 +541,22 @@ found:
+       /*
+        * It can happen that other CPUs service timer IRQs and increment
+        * jiffies, but we have not yet got a local timer tick to process
+        * the timer wheels.  In that case, the expiry time can be before
+        * jiffies, but since the high-resolution timer here is relative to
+        * jiffies, the default expression when high-resolution timers are
+        * not active,
+        *
+        *   time_before(MAX_JIFFY_OFFSET + jiffies, expires)
+        *
+        * would falsely evaluate to true.  If that is the case, just
+        * return jiffies so that we can immediately fire the local timer
+        */
+       if (time_before(expires, jiffies))
+               return jiffies;
        if (time_before(hr_expires, expires))
                return hr_expires;
Xen-devel mailing list
<Prev in Thread] Current Thread [Next in Thread>