fix save or snapshot at kernel loading time with HVM vm
PATCH 1/1 fix save or snapshot at kernel loading time with HVM vm
If doing save or snapshot at the time of kernel loading of a HVN vm, event_channel haven't been initialized, so that make this saving with xenstore. a string named "suspend" not "suspended" will be sent to xend from checkpoint thread.
But in the shutdown function with parameter "suspend" think this request
should shutdown this hvm guest, but obviously, this is not the expectation of save or snapshot.
-James (Song Wei)
diff -r baccadfd9418 tools/python/xen/xend/XendDomainInfo.py
--- a/tools/python/xen/xend/XendDomainInfo.py Fri May 14 08:05:05 2010
+0100
+++ b/tools/python/xen/xend/XendDomainInfo.py Mon May 17 13:56:15 2010
+0800
@@ -530,7 +530,7 @@
self.storeDom("control/shutdown", reason)
# HVM domain shuts itself down only if it has PV drivers
- if self.info.is_hvm():
+ if self.info.is_hvm() and reason != 'suspend':
hvm_pvdrv = xc.hvm_get_param(self.domid,HVM_PARAM_CALLBACK_IRQ)
hvm_s_state = xc.hvm_get_param(self.domid,HVM_PARAM_ACPI_S_STATE)
if not hvm_pvdrv or hvm_s_state != 0:
Signature · 2 lines
View this message in context: http://old.nabble.com/-PATCH--fix-save-or-snapshot-at-kernel-loading-time-with-HVM-vm-tp28579732p28579732.html Sent from the Xen - Dev mailing list archive at Nabble.com.
Quoted 8 lines · James (song wei)
If doing save or snapshot at the time of kernel loading of a HVN vm,
event_channelhaven't been initialized, so that make this saving with xenstore. a string named "suspend" not "suspended" will be sent to xend from checkpoint thread.
But in the shutdown function with parameter "suspend" think this request
should shutdown this hvm guest, but obviously, this is not the expectation of save or snapshot.
I don't really understand this patch comment, and the solution just looks like a poor hack which noone in future will be able to fathom.
-- Keir
Quoted 19 lines
-James (Song Wei)
Signed-off-by James (Song Wei) <[email protected]>diff -r
baccadfd9418tools/python/xen/xend/XendDomainInfo.py
---a/tools/python/xen/xend/XendDomainInfo.pyFri May 14 08:05:05 2010
+0100
+++b/tools/python/xen/xend/XendDomainInfo.pyMon May 17 13:56:15 2010
+0800(quoted diff) +1
-1@@ -530,7 +530,7 @@ self.storeDom("control/shutdown", reason) # HVM domain shuts itself down only if it has PV drivers - if self.info.is_hvm(): + if self.info.is_hvm() and reason != 'suspend': hvm_pvdrv = xc.hvm_get_param(self.domid, HVM_PARAM_CALLBACK_IRQ) hvm_s_state = xc.hvm_get_param(self.domid,
HVM_PARAM_ACPI_S_STATE)if not hvm_pvdrv or hvm_s_state != 0:
Quoted 56 lines · Keir Fraser-3 wrote:
Quoted 12 lines · James (song wei)
If doing save or snapshot at the time of kernel loading of a HVN vm,
event_channelhaven't been initialized, so that make this saving with xenstore. a string named "suspend" not "suspended" will be sent to xend from checkpoint thread.
But in the shutdown function with parameter "suspend" think this request
should shutdown this hvm guest, but obviously, this is not the expectation
of save or snapshot.Quoted 2 lines
I don't really understand this patch comment, and the solution just looks like a poor hack which noone in future will be able to fathom.
Keir,
Sorry for not descripting this issue clearly. When I do save at the
beginning of HVM guest booting(first stage - loading kernel), I found guest os destoryed after saving completely. So I see the code and get this flow:
xc_save()( here, call this function, when loading the kernel at the time of guest booting. Event channel haven't been initialized,so that go tocompat_suspend()path in suspend()(save.c) -->> then, notify the thread named saveInputHandler in xend with a string "suspend". so that dominfo.shutdown is invoked.
So, VM shutdown after do saving operator.-James (Song Wei)
Quoted 17 lines
Signed-off-by James (Song Wei) <[email protected]>diff -r
baccadfd9418tools/python/xen/xend/XendDomainInfo.py
---a/tools/python/xen/xend/XendDomainInfo.pyFri May 14 08:05:05 2010
+0100
+++b/tools/python/xen/xend/XendDomainInfo.pyMon May 17 13:56:15 2010
+0800(quoted diff) +1
-1@@ -530,7 +530,7 @@ self.storeDom("control/shutdown", reason) # HVM domain shuts itself down only if it has PV drivers - if self.info.is_hvm(): + if self.info.is_hvm() and reason != 'suspend': hvm_pvdrv = xc.hvm_get_param(self.domid, HVM_PARAM_CALLBACK_IRQ) hvm_s_state = xc.hvm_get_param(self.domid,
HVM_PARAM_ACPI_S_STATE)Quoted 1 line
if not hvm_pvdrv or hvm_s_state != 0:_______________________________________________
Xen-devel mailing list
[email protected]
http://lists.xensource.com/xen-devel
Signature · 2 lines
View this message in context: http://old.nabble.com/-PATCH--fix-save-or-snapshot-at-kernel-loading-time-with-HVM-vm-tp28579732p28604661.html Sent from the Xen - Dev mailing list archive at Nabble.com.
Quoted 3 lines · James (song wei)
Quoted 3 lines
Sorry for not descripting this issue clearly. When I do save at the
beginning of HVM guest booting(first stage - loading kernel), I found guest os destoryed after saving completely.
Well, that's expected isn't it? Normally, saving a guest should save its state to a file and destroy the guest, so that the saved state can be safely resumed later.
Quoted 6 lines
Quoted 6 lines
So I see the code and get this flow:
xc_save()( here, call this function, when loading the kernel at the time of guest booting. Event channel haven't been initialized,so that go tocompat_suspend()path in suspend()(save.c) -->> then, notify the thread named saveInputHandler in xend with a string "suspend". so that dominfo.shutdown is invoked.
That's required. We need the guest to shutdown with state SHUTDOWN_suspend, so that we can save a consistent snapshot of its state. If you skip the shutdown then you break domain save in the absence of PV drivers. It will be broken because the guest will continue to run while xc_save does its work -- hence it will not save a consistent snapshot of the guest state.
-- Keir
Quoted 1 line
Quoted 1 line
So, VM shutdown after do saving operator.