RE: [Xen-devel] [PATCH] qemu-xen: fix cpu hotplug

["Liu, Jinsong" <jinsong.liu@xxxxxxxxx>]

Christoph Egger wrote:
> On Friday 03 September 2010 08:28:53 Liu, Jinsong wrote:
>> Ian Jackson wrote:
>>> Stefano Stabellini writes ("[Xen-devel] [PATCH] qemu-xen: fix cpu
>>> 
>>> hotplug"):
>>>> qemu-xen: fix cpu hotplug
>>>> 
>>>> The current xenstore watch path for a vcpu-set event is wrong and
>>>> is also wrong the code to parse it.  This patch fixes both of them:
>>> 
>>> Thanks.  So it seems you're saying it's completely broken in
>>> xen-unstable.
>> 
>> Stabellini, I read your attached patch, it's OK.
>> In fact, we firstly implemented xenstore watch by same scheme of your
>> patch, watching each cpu node status:
>> /local/domain/xx/cpu/yy/availability=offline (online)
>> 
>> However, we finally didn't use this scheme. We watch 'common' node
>> instead: /local/domain/xx/cpu in this way, only 1 watch point need.
>> Considering vcpu number may become more and more in the future (say,
>> more than 128), it's more simple and reasonable. (Watches can be set
>> at points in the hierarchy and an individual watch will be triggered
>> when anything at or below that point in the hierachy changes)
> 
> Does this scheme allow to say how many cores per cpu exist ?
> 
> When you run a Windows guest with a license for one cpu socket,
> then you can use 4 cores. But if one cpu is equal to one socket,
> then you can't use SMP for the Windows guest.
> 
> Christoph
> 

Seems this is another story?

Jinsong

> 
>> 
>>> I have CC'd a bunch of Intel folks who were doing some other
>>> work on cpu hotplug.  They were dealing with a race when multiple
>>> CPUs were added at once. 
>>> 
>>> So I think there must be some confusion.  Perhaps xl and xend have
>>> different ideas about what the xenstore syntax is for these
>>> operations ? 
>>> 
>>> Jinsong Liu et al: would you care to comment ?  Particularly about
>>> 
>>> this:
>>>> A xenstore vcpu hotplug command is of the following form:
>>>> path: /local/domain/DOMID/cpu/VCPU_NUMBER/availability
>>>> values: "online" or "offline"
>> 
>> Yes, I think there must be some confusion.
>> Currently 'xm vcpu-set' command works fine with both PV and HVM vcpu
>> hotplug. 
>> 
>> Stabellini/Jackson, would your please tell me what xl recently
>> happened for vcpu hotplug? is there any different ideas xl and xend
>> about xenstore syntax? (Each time 'xm vcpu-set' executed, xend will
>> write all xenstore cpu node status) 
>> 
>>> That doesn't seem to match up with what's in xend, which seems to
>>> write "cpu_avail" in the vm tree.
>>> 
>>> Thanks,
>>> Ian.


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