On Tue, Jun 04, 2013 at 09:46:14PM +0800, Osier Yang wrote:
On 04/06/13 21:13, Daniel P. Berrange wrote:
On Tue, Jun 04, 2013 at 06:59:02AM -0600, Eric Blake wrote:
On 06/04/2013 05:49 AM, Osier Yang wrote:
https://bugzilla.redhat.com/show_bug.cgi?id=964177
Though both libvirt and QEMU's document say RTC_CHANGE returns the offset from the host UTC, qemu actually returns the offset from the specified date instead when specific date is privided s/privided/provided/
(-rtc base=$date).
It's not safe for qemu to fix it in code, it worked like that for 3 years, changing it now may break other QEMU use cases. What qemu tries to do is to fix the document:
http://lists.gnu.org/archive/html/qemu-devel/2013-05/msg04782.html
And in libvirt side, instead of reply on the qemu, this covert s/covert/convert/
the offset returned from qemu to the offset from host UTC, by:
/* * a: the offset from qemu RTC_CHANGE event * b: The specified date (-rtc base=$date) * c: the host date when libvirt gets the RTC_CHANGE event * offset: What libvirt will report */
offset = a + (b - c);
The specified date (-rtc base=$date) is recorded in clock's def as an internal only member (may be useful to exposed outside?). --- src/conf/domain_conf.h | 3 +++ src/qemu/qemu_command.c | 3 +++ src/qemu/qemu_process.c | 12 ++++++++++++ 3 files changed, 18 insertions(+) Incomplete. You need to track the start time across libvirtd restarts (in internal XML) for this to reliably work for an event received after a restart; you also have to cope with a libvirtd restart not finding the field in internal XML (because the libvirtd restart was due to upgrading libvirt in the meantime).
diff --git a/src/conf/domain_conf.h b/src/conf/domain_conf.h index 3a71d6c..3947a56 100644 --- a/src/conf/domain_conf.h +++ b/src/conf/domain_conf.h @@ -1767,6 +1767,9 @@ struct _virDomainClockDef { struct { long long adjustment; int basis; + + /* Store the start time of guest process, internaly only */ Spelling; either 'internal' or 'internally'
+ time_t starttime; } variable; /* Timezone name, when diff --git a/src/qemu/qemu_command.c b/src/qemu/qemu_command.c index c4a162a..9254525 100644 --- a/src/qemu/qemu_command.c +++ b/src/qemu/qemu_command.c @@ -5518,6 +5518,9 @@ qemuBuildClockArgStr(virDomainClockDefPtr def) now += def->data.variable.adjustment; gmtime_r(&now, &nowbits); + /* Store the starttime of qemu process */ + def->data.variable.starttime = now; Is there anything we can read out of /proc/nnn for the qemu process that would give us a more accurate start time? In fact, why not use virProcessGetStartTime()? And if virProcessGetStartTime is reliable across libvirtd restarts, then you might not need to store a time_t starttime in _virDomainClockDef. It isn't the start time of the QEMU process that we care about here. The offset is relative to the timestamp specified via the -clock command line arg. So using QEMU procss startup would be wrong.
hm, I don't see libvirt uses "-clock" option. And the offset is from the specified of "-rtc base=$date" with my testing.
Sorry, yes, i meant -rtc, not -clock Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|