Re: [Xen-devel] when timer go back in dom0 save and restore or migrate, PV domain hung

[Keir Fraser <keir.fraser@xxxxxxxxxxxxx>]

hrtimers add wall_to_monotonic to xtime to get a timesource that doesn't (or shouldn't!) warp.

 -- Keir

On 26/11/08 14:20, "Tian, Kevin" <kevin.tian@xxxxxxxxx> wrote:

how about hrtimers? one mode is CLOCK_REALTIME, which uses getnstimeofday as expiration. Once system time is changed either in local or new machine, that expiration can't be adjusted. but i'm not sure whether it still makes sense to try hrtimers in a guest.

Thanks
Kevin

 

---

From: Keir Fraser  [mailto:keir.fraser@xxxxxxxxxxxxx]
Sent: Wednesday, November 26,  2008 10:11 PM
To: Tian, Kevin; 'James Song';  xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: Re: [Xen-devel] when timer go  back in dom0 save and restore or migrate, PV domain hung

 
The problem hasn't been fully explained, but I can say  that PV guests expect system time to jump across s/r and deal with that. For  example, Linux doesn't use Xen system time internally, but uses its progress  to periodically update jiffies, which does not warp across s/r.

We have  had problems corrupting wc_sec/wc_nsec in xc_domain_restore.c, but that was  fixed some time ago.

 -- Keir

On 26/11/08 14:00, "Tian,  Kevin" <kevin.tian@xxxxxxxxx> wrote:

 

This is not a s/r or lm specific issue. For example, system time  can be changed even when pv guest is running. Your patch only hacks restore  point once, and wc_sec can still be changed later when system time is  changed on-the-fly again.

IIRC, pv guest can catch up wall clock change in timer interrupt,  and time_resume will sync internal processed system time with new system  time after restored. But I'm not sure whether it's enough. Actually the more  interesting is the uptime difference. For example, timer with expiration  calculated on previous system time may wait nearly infinite if uptime among  two boxes vary a lot. But I think such issue should have been considered  already, e.g. some user tool assistance. I think Keir can comment better  here.

BTW, do you happen to know what exactly dom0 hangs on? In some  busy loop to catch up time, or long delay to some critical timer  expiration?

Thanks,
Kevin

 

 
 

---

From:  xen-devel-bounces@xxxxxxxxxxxxxxxxxxx  [mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx]  On Behalf Of James  Song
Sent: Tuesday, November 25,  2008 4:02 PM
To:   xen-devel@xxxxxxxxxxxxxxxxxxx
Subject: [Xen-devel] when  timer go  back in dom0 save and restore or migrate, PV domain  hung

 
Hi,
   I  find PV domin hung, When we take those steps   
         1, save PV  domain  
         2,  change system time of  PV domain back  
         3, restore   a PV domain
        or   
         1, migrate  a PV domain  from Machine A to Machine  B
         2, the system   time of Machine B is slower than Machine A.
   the  problem is  wc_sec will be change when system-time chanaged in dom0  or restore in a  slower-system-time machine, but when restoring, xen  don't restore the wc_sec  of share_info from xenstore and use native  one. So guest os will hang.  
this patch will work for this  issue.

 Thanks
 -- Song  Wei

diff -r  a5ed0dbc829f tools/libxc/xc_domain_restore.c
---   a/tools/libxc/xc_domain_restore.c    Tue Nov 18  14:34:14 2008  +0800
+++ b/tools/libxc/xc_domain_restore.c     Fri Nov 21  17:34:15 2008 +0800
@@ -328,6  +328,16  @@
 
     /* For info  only  */
     nr_pfns = 0;
+      //jsong@xxxxxxxxxx, james song
+     memset(&domctl, 0,  sizeof(domctl));
+     domctl.domain =  dom;
+     domctl.cmd    =   XEN_DOMCTL_restoredomain;
+    frc =  do_domctl(xc_handle,  &domctl);
+    if ( frc  != 0 )
+     {
+              ERROR("Unable  to set flag of  restore.");
+              goto  out;
+     }
 
     if (   read_exact(io_fd, &p2m_size, sizeof(unsigned long))   )
     {
@@ -1120,6 +1130,8   @@
 
     /* restore saved  vcpu_info and arch  specific info  */
     MEMCPY_FIELD(new_shared_info,   old_shared_info, vcpu_info);
+      MEMCPY_FIELD(new_shared_info, old_shared_info,   wc_nsec);
+    MEMCPY_FIELD(new_shared_info,   old_shared_info,  wc_sec);
      MEMCPY_FIELD(new_shared_info,  old_shared_info,   arch);
 
     /* clear any  pending events and  the selector */
diff -r a5ed0dbc829f  xen/arch/x86/time.c
---  a/xen/arch/x86/time.c     Tue Nov 18 14:34:14 2008 +0800
+++   b/xen/arch/x86/time.c    Fri Nov 21 17:34:15 2008  +0800
@@  -689,7 +689,6  @@
      wmb();
     (*version)++;
 }
-
 void   update_vcpu_system_time(struct vcpu  *v)
 {
      struct cpu_time        *t;
@@ -703,7 +702,6   @@
 
     if (  u->tsc_timestamp ==  t->local_tsc_stamp  )
          return;
-
      version_update_begin(&u->version);
 
      u->tsc_timestamp      = t->local_tsc_stamp;
@@  -713,14  +711,19  @@
 
      version_update_end(&u->version);
 }
-
 void   update_domain_wallclock_time(struct domain   *d)
 {
      spin_lock(&wc_lock);
+     if(d->after_restore  )
+     {
+          d->after_restore =  0;
+        goto  out;  //jsong@xxxxxxxxxx
+     }
      version_update_begin(&shared_info(d,   wc_version));
     shared_info(d,  wc_sec)  =  wc_sec +  d->time_offset_seconds;
     shared_info(d,   wc_nsec) =  wc_nsec;
      version_update_end(&shared_info(d,   wc_version));
+out:
      spin_unlock(&wc_lock);
 }
 
@@  -751,7 +754,6  @@
     u64  x;
     u32 y,  _wc_sec,  _wc_nsec;
     struct domain   *d;
-
     x = (secs * 1000000000ULL)  + (u64)nsecs -  system_time_base;
     y  = do_div(x,  1000000000);
 
@@ -1050,7 +1052,6  @@
 struct tm   wallclock_time(void)
 {
     uint64_t   seconds;
-
     if ( !wc_sec   )
         return  (struct tm) { 0  };
 
diff -r a5ed0dbc829f  xen/common/domctl.c
---  a/xen/common/domctl.c     Tue Nov 18 14:34:14 2008 +0800
+++   b/xen/common/domctl.c    Fri Nov 21 17:34:15 2008  +0800
@@  -24,7 +24,6 @@
 #include  <asm/current.h>
 #include   <public/domctl.h>
 #include   <xsm/xsm.h>
-
 extern long   arch_do_domctl(
     struct xen_domctl  *op,  XEN_GUEST_HANDLE(xen_domctl_t) u_domctl);
 
@@  -315,6 +314,16   @@
         ret =   0;
     }
      break;
+    case XEN_DOMCTL_restoredomain:
+    {
+         struct domain  *d;
+         if ( (d =   rcu_lock_domain_by_id(op->domain)) == NULL  )
+              break;
+          
+         d->after_restore =   1;
+          rcu_unlock_domain(d);
+          break;
+    }
 
     case   XEN_DOMCTL_createdomain:
     {
diff  -r a5ed0dbc829f  xen/include/public/domctl.h
---   a/xen/include/public/domctl.h    Tue Nov 18 14:34:14  2008  +0800
+++ b/xen/include/public/domctl.h     Fri Nov 21  17:34:15 2008 +0800
@@ -61,6 +61,7  @@
 #define  XEN_DOMCTL_destroydomain       2
 #define   XEN_DOMCTL_pausedomain          3
 #define  XEN_DOMCTL_unpausedomain       4
+#define  XEN_DOMCTL_restoredomain        51
 #define  XEN_DOMCTL_resumedomain        27
 
 #define   XEN_DOMCTL_getdomaininfo      5
diff -r   a5ed0dbc829f xen/include/xen/sched.h
---   a/xen/include/xen/sched.h    Tue Nov 18 14:34:14 2008   +0800
+++ b/xen/include/xen/sched.h    Fri Nov 21  17:34:15  2008 +0800
@@ -231,6 +231,7  @@
      * cause a  deadlock.  Acquirers don't spin waiting; they   preempt.
      */
      spinlock_t  hypercall_deadlock_mutex;
+    int after_restore;   //jsong@xxxxxxxxxx
 };
 
 struct   domain_setup_info
---------------------------------------------------------------------------------------------
 Thanks
--Song   wei

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel