[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

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



Might this be a pv_ops bug in newer Linux kernels? I don’t really get what you’re describing though.

 -- Keir

On 27/11/08 10:21, "James Song" <jsong@xxxxxxxxxx> wrote:

Hi,
     Ok, now two machine A and B. the system-time of A is ahead of B. So wc_sec of A is also bigger than B. When PV dom in A migrate to B, we haven't upate that PV dom's wc_sec to equal with B. Ok, now we see pv dom's kernel:
    xen_sched_clock() in arch/86/xen/time.c andxen_clocksource_read()  arch/x86/kernel/time_32-xen.c
  you will find if state_entry_time of its's vcpu, because the state_entry_time is initalized in machine A. this time it more big than "now" of machine B. So no schedule, no system-update in Guest os.
I don't whether did I describe it clearly.

>>> "Tian, Kevin"  08/11/27 PM 9:18 >>>
there's a clock_was_set called for each settimeofday. In latest kernel, clock_was_set will adjust CLOCK_REALTIME queue accordingly, while in 2.6.18 it's defined as a nop. That says, current domU would be unable to handle wallclock change, but newer kernel with pvops could.  ---- yes, it works for FV, but for a modified PV domain, mybe not.

for the issue reported in original thread, I agree that James should dig into the hang and explain the exact reason first.

Thanks
Kevin


 

From: Keir Fraser  [mailto:keir.fraser@xxxxxxxxxxxxx]
Sent: Wednesday, November 26,  2008 10:58 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

 
So what happens if someone changes wallclock using  'date'? That's basically kind of what will appear to happen when s/r  occurs.

 -- Keir

On 26/11/08 14:32, "Tian, Kevin"  <kevin.tian@xxxxxxxxx> wrote:

 
hrtimer supports two timer bases: CLOCK_MONOTONIC and  CLOCK_REALTIME. wall_to_monotonic is only added in former case, and for  latter instead TOD is used directly per my reading. I did a quick search,  and it looks that futex and ntp are using CLOCK_REALTIME. Also there's one  vsyscall gate which can pass CLOCK_REALTIME from caller  too.

Thanks,
Kevin

 

 
 

From: Keir Fraser  [mailto:keir.fraser@xxxxxxxxxxxxx] <mailto:keir.fraser@xxxxxxxxxxxxx%5D>   
Sent: Wednesday, November 26,  2008 10:26 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

 
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] <mailto:keir.fraser@xxxxxxxxxxxxx%5D>    
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] <mailto:xen-devel-bounces@xxxxxxxxxxxxxxxxxxx%5D>     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

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.