Re: [Xen-devel] [Question] vcpu-set before or after xen_pause_requested

[Ian Jackson <Ian.Jackson@xxxxxxxxxxxxx>]

Liu, Jinsong writes ("[Xen-devel] [Question] vcpu-set before or after 
xen_pause_requested"):
> At xenstore.c --> xenstore_process_event(), where is right place to add 
> xenstore_process_vcpu_set_event():
> 1. add xenstore_process_vcpu_set_event() before xen_pause_requested check? or
> 2. add xenstore_process_vcpu_set_event() after xen_pause_requested check?
> That means, may vcpu-set take effect between xm save/resume command? seems 
> now pci hotplug and usb hotplug do so.

Is your vcpu-set done by writing to the xenstore key
   /local/domain/0/device-model/<domain>/command
?  If so it should be added as a new branch to the ifs in to
xenstore_process_dm_command_event, and you should make sure that when
your command is processed you correctly write a "processed" value to
that same xenstore key, just as usb-add does.  Because only one
command can be processed at once there is no possibility that a domain
which is supposed to be pausing could simultaneously be told to do a
hotplug (or vcpu-set).

Please also add corresponding code to xl and libxl to do the vcpu-set
operation.  We're hoping make xl and libxl the primary mechanism in
4.1.

If your vcpu-set is not done by writing to /command, how does it work
and what is the protocol ?  Or haven't you designed the protocol yet?
If you haven't I'd suggest using the existing /command mechanism.

The logic surrounding pci-{rem,ins} seems to be wrong to me because it
does not write a "processed" value.  So don't use that as an example.

Ian.

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