Hi, I''m running xen-unstable c/s 19603 with a single 2.6.18.8-xen kernel image used for both dom0 and domUs. I''m experiencing a time freeze when I restore a domU checkpoint file on another physical host. Basically, both date (referring to /etc/localtime) and gettimeofday() (issuing a gettimeofday syscall) repeatedly report unchanging values for tens of seconds: debian:/var/tmp# ./timer time: sec=1265844232, usec=728054 debian:/var/tmp# ./timer time: sec=1265844232, usec=728054 debian:/var/tmp# ./timer time: sec=1265844232, usec=728054 debian:/var/tmp# date Wed Feb 10 23:23:52 UTC 2010 debian:/var/tmp# date Wed Feb 10 23:23:52 UTC 2010 debian:/var/tmp# date Wed Feb 10 23:23:52 UTC 2010 The timer (TSC??) springs back to life after 20-30 seconds. Hardware: Sun Fire X2250, 2 socket, quad-core = total of 8 execution threads. Processor: Intel Xeon E5472 @ 3GHz Arch: x86_64 I''ve seen some discussion about TSC skew, and tried setting clocksource to acpi instead of the default hpet - didn''t help. I also tried echoing "1" to /proc/sys/xen/independent_wallclock to no avail. Finally, no luck debugging with xen gdb, because setting a breakpoint in do_gettimeofday is futile - it fires non-stop. Does anybody have any suggestions? In my case, it is not just a TSC skew - the clock stalls for quite an extended period of time, while the restored VM is otherwise operational and responds to all sorts of commands unless they execute anything that translates into a nanosleep syscall. The latter, of course, won''t return until the clock starts going again. Thanks, Alex. _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Is behaviour different if you put a line ''tsc_mode=2'' in your domain config file as passed to ''xm create''? -- Keir On 10/02/2010 23:51, "Alexey Tumanov" <atumanov@gmail.com> wrote:> Hi, > > I''m running xen-unstable c/s 19603 with a single 2.6.18.8-xen kernel image > used for both dom0 and domUs. I''m experiencing a time freeze when I restore a > domU checkpoint file on another physical host. Basically, both date (referring > to /etc/localtime) and gettimeofday() (issuing a gettimeofday syscall) > repeatedly report unchanging values for tens of seconds: > debian:/var/tmp# ./timer > time: sec=1265844232, usec=728054 > debian:/var/tmp# ./timer > time: sec=1265844232, usec=728054 > debian:/var/tmp# ./timer > time: sec=1265844232, usec=728054 > debian:/var/tmp# date > Wed Feb 10 23:23:52 UTC 2010 > debian:/var/tmp# date > Wed Feb 10 23:23:52 UTC 2010 > debian:/var/tmp# date > Wed Feb 10 23:23:52 UTC 2010 > > The timer (TSC??) springs back to life after 20-30 seconds. > Hardware: Sun Fire X2250, 2 socket, quad-core = total of 8 execution threads. > Processor: Intel Xeon E5472 @ 3GHz > Arch: x86_64 > > I''ve seen some discussion about TSC skew, and tried setting clocksource to > acpi instead of the default hpet - didn''t help. I also tried echoing "1" to > /proc/sys/xen/independent_wallclock to no avail. Finally, no luck debugging > with xen gdb, because setting a breakpoint in do_gettimeofday is futile - it > fires non-stop. > > Does anybody have any suggestions? In my case, it is not just a TSC skew - the > clock stalls for quite an extended period of time, while the restored VM is > otherwise operational and responds to all sorts of commands unless they > execute anything that translates into a nanosleep syscall. The latter, of > course, won''t return until the clock starts going again. > > Thanks, > Alex. >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
same behavior. But, according to this doc: http://lists.xensource.com/archives/html/xen-changelog/2009-12/msg00035.html tsc_mode is a new feature in Xen-4? Xen-unstable c/s 19603 is roughly at 3.4.0 level. Thanks, Alex. On 10 February 2010 19:09, Keir Fraser <keir.fraser@eu.citrix.com> wrote:> Is behaviour different if you put a line ''tsc_mode=2'' in your domain config > file as passed to ''xm create''? > > -- Keir > > On 10/02/2010 23:51, "Alexey Tumanov" <atumanov@gmail.com> wrote: > > > Hi, > > > > I''m running xen-unstable c/s 19603 with a single 2.6.18.8-xen kernel > image > > used for both dom0 and domUs. I''m experiencing a time freeze when I > restore a > > domU checkpoint file on another physical host. Basically, both date > (referring > > to /etc/localtime) and gettimeofday() (issuing a gettimeofday syscall) > > repeatedly report unchanging values for tens of seconds: > > debian:/var/tmp# ./timer > > time: sec=1265844232, usec=728054 > > debian:/var/tmp# ./timer > > time: sec=1265844232, usec=728054 > > debian:/var/tmp# ./timer > > time: sec=1265844232, usec=728054 > > debian:/var/tmp# date > > Wed Feb 10 23:23:52 UTC 2010 > > debian:/var/tmp# date > > Wed Feb 10 23:23:52 UTC 2010 > > debian:/var/tmp# date > > Wed Feb 10 23:23:52 UTC 2010 > > > > The timer (TSC??) springs back to life after 20-30 seconds. > > Hardware: Sun Fire X2250, 2 socket, quad-core = total of 8 execution > threads. > > Processor: Intel Xeon E5472 @ 3GHz > > Arch: x86_64 > > > > I''ve seen some discussion about TSC skew, and tried setting clocksource > to > > acpi instead of the default hpet - didn''t help. I also tried echoing "1" > to > > /proc/sys/xen/independent_wallclock to no avail. Finally, no luck > debugging > > with xen gdb, because setting a breakpoint in do_gettimeofday is futile - > it > > fires non-stop. > > > > Does anybody have any suggestions? In my case, it is not just a TSC skew > - the > > clock stalls for quite an extended period of time, while the restored VM > is > > otherwise operational and responds to all sorts of commands unless they > > execute anything that translates into a nanosleep syscall. The latter, of > > course, won''t return until the clock starts going again. > > > > Thanks, > > Alex. > > > > >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
> Is behaviour different if you put a line ''tsc_mode=2'' in your domain > config file as passed to ''xm create''?Keir -- C/s 19603 I think predates all of the tsc work, though the problem might be related to timer_mode. Alexey -- Why such an old changeset? There''s been a LOT of work on time since then. If you are using a released product by a vendor with this changeset, you might want to check with that vendor. If not, and you aren''t able to update to a newer Xen, or if you update and it doesn''t fix the problem, please reply again. Dan> -----Original Message----- > From: Keir Fraser [mailto:keir.fraser@eu.citrix.com] > Sent: Wednesday, February 10, 2010 5:09 PM > To: Alexey Tumanov; xen-devel@lists.xensource.com > Subject: Re: [Xen-devel] time freeze on save/restore, x86_64 > > Is behaviour different if you put a line ''tsc_mode=2'' in your domain > config > file as passed to ''xm create''? > > -- Keir > > On 10/02/2010 23:51, "Alexey Tumanov" <atumanov@gmail.com> wrote: > > > Hi, > > > > I''m running xen-unstable c/s 19603 with a single 2.6.18.8-xen kernel > image > > used for both dom0 and domUs. I''m experiencing a time freeze when I > restore a > > domU checkpoint file on another physical host. Basically, both date > (referring > > to /etc/localtime) and gettimeofday() (issuing a gettimeofday > syscall) > > repeatedly report unchanging values for tens of seconds: > > debian:/var/tmp# ./timer > > time: sec=1265844232, usec=728054 > > debian:/var/tmp# ./timer > > time: sec=1265844232, usec=728054 > > debian:/var/tmp# ./timer > > time: sec=1265844232, usec=728054 > > debian:/var/tmp# date > > Wed Feb 10 23:23:52 UTC 2010 > > debian:/var/tmp# date > > Wed Feb 10 23:23:52 UTC 2010 > > debian:/var/tmp# date > > Wed Feb 10 23:23:52 UTC 2010 > > > > The timer (TSC??) springs back to life after 20-30 seconds. > > Hardware: Sun Fire X2250, 2 socket, quad-core = total of 8 execution > threads. > > Processor: Intel Xeon E5472 @ 3GHz > > Arch: x86_64 > > > > I''ve seen some discussion about TSC skew, and tried setting > clocksource to > > acpi instead of the default hpet - didn''t help. I also tried echoing > "1" to > > /proc/sys/xen/independent_wallclock to no avail. Finally, no luck > debugging > > with xen gdb, because setting a breakpoint in do_gettimeofday is > futile - it > > fires non-stop. > > > > Does anybody have any suggestions? In my case, it is not just a TSC > skew - the > > clock stalls for quite an extended period of time, while the restored > VM is > > otherwise operational and responds to all sorts of commands unless > they > > execute anything that translates into a nanosleep syscall. The > latter, of > > course, won''t return until the clock starts going again. > > > > Thanks, > > Alex. > > > > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
On 11/02/2010 00:20, "Alexey Tumanov" <atumanov@gmail.com> wrote:> same behavior. But, according to this doc: > http://lists.xensource.com/archives/html/xen-changelog/2009-12/msg00035.html > tsc_mode is a new feature in Xen-4? Xen-unstable c/s 19603 is roughly at 3.4.0 > level.If you can reproduce the problem with tip of xen-3.4-testing or xen-unstable then that''s more interesting. -- Keir _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
we''re maintaining a patch queue on top of this specific c/set, and rebasing the patch queue carries engineering overhead and testing. But if there''s a small number of changesets that fixes the problem, perhaps we could backport them? Alex. On 10 February 2010 19:22, Dan Magenheimer <dan.magenheimer@oracle.com>wrote:> > Is behaviour different if you put a line ''tsc_mode=2'' in your domain > > config file as passed to ''xm create''? > > Keir -- > > C/s 19603 I think predates all of the tsc work, though the problem > might be related to timer_mode. > > Alexey -- > > Why such an old changeset? There''s been a LOT of work on time since then. > If you are using a released product by a vendor with this changeset, > you might want to check with that vendor. If not, and you aren''t > able to update to a newer Xen, or if you update and it doesn''t > fix the problem, please reply again. > > Dan > > > -----Original Message----- > > From: Keir Fraser [mailto:keir.fraser@eu.citrix.com] > > Sent: Wednesday, February 10, 2010 5:09 PM > > To: Alexey Tumanov; xen-devel@lists.xensource.com > > Subject: Re: [Xen-devel] time freeze on save/restore, x86_64 > > > > Is behaviour different if you put a line ''tsc_mode=2'' in your domain > > config > > file as passed to ''xm create''? > > > > -- Keir > > > > On 10/02/2010 23:51, "Alexey Tumanov" <atumanov@gmail.com> wrote: > > > > > Hi, > > > > > > I''m running xen-unstable c/s 19603 with a single 2.6.18.8-xen kernel > > image > > > used for both dom0 and domUs. I''m experiencing a time freeze when I > > restore a > > > domU checkpoint file on another physical host. Basically, both date > > (referring > > > to /etc/localtime) and gettimeofday() (issuing a gettimeofday > > syscall) > > > repeatedly report unchanging values for tens of seconds: > > > debian:/var/tmp# ./timer > > > time: sec=1265844232, usec=728054 > > > debian:/var/tmp# ./timer > > > time: sec=1265844232, usec=728054 > > > debian:/var/tmp# ./timer > > > time: sec=1265844232, usec=728054 > > > debian:/var/tmp# date > > > Wed Feb 10 23:23:52 UTC 2010 > > > debian:/var/tmp# date > > > Wed Feb 10 23:23:52 UTC 2010 > > > debian:/var/tmp# date > > > Wed Feb 10 23:23:52 UTC 2010 > > > > > > The timer (TSC??) springs back to life after 20-30 seconds. > > > Hardware: Sun Fire X2250, 2 socket, quad-core = total of 8 execution > > threads. > > > Processor: Intel Xeon E5472 @ 3GHz > > > Arch: x86_64 > > > > > > I''ve seen some discussion about TSC skew, and tried setting > > clocksource to > > > acpi instead of the default hpet - didn''t help. I also tried echoing > > "1" to > > > /proc/sys/xen/independent_wallclock to no avail. Finally, no luck > > debugging > > > with xen gdb, because setting a breakpoint in do_gettimeofday is > > futile - it > > > fires non-stop. > > > > > > Does anybody have any suggestions? In my case, it is not just a TSC > > skew - the > > > clock stalls for quite an extended period of time, while the restored > > VM is > > > otherwise operational and responds to all sorts of commands unless > > they > > > execute anything that translates into a nanosleep syscall. The > > latter, of > > > course, won''t return until the clock starts going again. > > > > > > Thanks, > > > Alex. > > > > > > > > > > > _______________________________________________ > > Xen-devel mailing list > > Xen-devel@lists.xensource.com > > http://lists.xensource.com/xen-devel >_______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel