]> pilppa.com Git - linux-2.6-omap-h63xx.git/commit
[PATCH] X86_64 monotonic_clock goes backwards
authorDimitri Sivanich <sivanich@sgi.com>
Tue, 26 Sep 2006 08:52:34 +0000 (10:52 +0200)
committerAndi Kleen <andi@basil.nowhere.org>
Tue, 26 Sep 2006 08:52:34 +0000 (10:52 +0200)
commitcbf9b4bb76c9ce53b7fdde0dffcd000951b5f0d4
tree0942698b5aef01d2b89fe04111f7ba40b28865f0
parentd28c4393a7bf558538e9def269c1caeab6ec056f
[PATCH] X86_64 monotonic_clock goes backwards

I've noticed some erratic behavior while testing the X86_64 version
of monotonic_clock().

While spinning in a loop reading monotonic clock values (pinned to a
single cpu) I noticed that the difference between subsequent values
occasionally went negative (time going backwards).

I found that in the following code:
                this_offset = get_cycles_sync();
                /* FIXME: 1000 or 1000000? */
-->             offset = (this_offset - last_offset)*1000 / cpu_khz;
        }
        return base + offset;

the offset sometimes turns out to be 0, even though
this_offset > last_offset.

+Added fix From: Toyo Abe <toyoa@mvista.com>

The x86_64-mm-monotonic-clock.patch in 2.6.18-rc4-mm2 made a change to
the updating of monotonic_base. It now uses cycles_2_ns().

I suggest that a set_cyc2ns_scale() should be done prior to the setup_irq().
Because cycles_2_ns() can be called from the timer ISR right after the irq0
is enabled.

Signed-off-by: Toyo Abe <toyoa@mvista.com>
Signed-off-by: Dimitri Sivanich <sivanich@sgi.com>
Signed-off-by: Andi Kleen <ak@suse.de>
arch/x86_64/kernel/time.c