How to query the number of vcpus?

How to query the number of vcpus?

[Kip Macy]

What is the right way to query how many vcpus a guest is using? As I
mentioned earlier it would seem to fall under the heading of
GETDOMAININFO, but that obviously doesn't work.

    -Kip

Re: How to query the number of vcpus?

[Keir Fraser]

SMP save/restore is work in progress so I expect this kind of vital 
info isn't exported via the control interfaces at all right now.

  -- Keir

On 29 Apr 2005, at 21:08, Kip Macy wrote:

> What is the right way to query how many vcpus a guest is using? As I
> mentioned earlier it would seem to fall under the heading of
> GETDOMAININFO, but that obviously doesn't work.
>
>     -Kip
>
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: How to query the number of vcpus?

[Ryan Harper]

* Kip Macy <kip.macy@gmail.com> [2005-04-29 15:09]:
> What is the right way to query how many vcpus a guest is using? As I
> mentioned earlier it would seem to fall under the heading of
> GETDOMAININFO, but that obviously doesn't work.

I posted a [1]patch a while back that surfaced that info in the
GETDOMAININFO call.  I'm still working more into that patch to support
a cpumap which designate which physical cpus a vcpu may use, the current
form of that patch is [2]here.

1.  http://lists.xensource.com/archives/html/xen-devel/2005-04/msg00412.html
2.  http://lists.xensource.com/archives/html/xen-devel/2005-04/msg00890.html

-- 
Ryan Harper
Software Engineer; Linux Technology Center
IBM Corp., Austin, Tx
(512) 838-9253   T/L: 678-9253
ryanh@us.ibm.com

Re: How to query the number of vcpus?

[Kip Macy]

Thanks. I need this for coredump and debugging on SMP domains. 

I didn't feel like querying successively higher exec_domains until a
call failed ;-)

               -Kip
     
On 4/29/05, Ryan Harper <ryanh@us.ibm.com> wrote:
> * Kip Macy <kip.macy@gmail.com> [2005-04-29 15:09]:
> > What is the right way to query how many vcpus a guest is using? As I
> > mentioned earlier it would seem to fall under the heading of
> > GETDOMAININFO, but that obviously doesn't work.
> 
> I posted a [1]patch a while back that surfaced that info in the
> GETDOMAININFO call.  I'm still working more into that patch to support
> a cpumap which designate which physical cpus a vcpu may use, the current
> form of that patch is [2]here.
> 
> 1.  http://lists.xensource.com/archives/html/xen-devel/2005-04/msg00412.html
> 2.  http://lists.xensource.com/archives/html/xen-devel/2005-04/msg00890.html
> 
> --
> Ryan Harper
> Software Engineer; Linux Technology Center
> IBM Corp., Austin, Tx
> (512) 838-9253   T/L: 678-9253
> ryanh@us.ibm.com
>

Re: How to query the number of vcpus?

[Keir Fraser]

> * Kip Macy <kip.macy@gmail.com> [2005-04-29 15:09]:
> > What is the right way to query how many vcpus a guest is using? As I
> > mentioned earlier it would seem to fall under the heading of
> > GETDOMAININFO, but that obviously doesn't work.
> 
> I posted a [1]patch a while back that surfaced that info in the
> GETDOMAININFO call.  I'm still working more into that patch to support
> a cpumap which designate which physical cpus a vcpu may use, the current
> form of that patch is [2]here.
> 
> 1.  http://lists.xensource.com/archives/html/xen-devel/2005-04/msg00412.html
> 2.  http://lists.xensource.com/archives/html/xen-devel/2005-04/msg00890.html

There's already a cpumask in Xen for physical CPUs (or will be -- it's
called cpuset right now, but that's going to change to fit the Linux
convention). 

It probably makes sense to rename cpumap to vcpumask, or at least to
vcpumap.

Also I'd change the appearances of the mask in the dom0 interface from:
 cpumap_t cpumap[MAX_VIRT_CPUS]
And:
 cpumap_t *cpumap
To:
 u8 vcpumask[MAX_VCPUID/8]

Having an array of '_t' is odd as each entry in the array is really
logically just a single bit. And I would like to separate the size of
the array at the admin interfaces from MAX_VIRT_CPUS: it would be nice
to be able to increase MAX_VIRT_CPUS without changing the admin
interface. That's why I'd add something like MAX_VCPUID (probably set
it to 256).

I cc'ed Christian because he is merging one of your patches at the
moment. It might be the one without the cpumap additions though.

 -- Keir

Re: How to query the number of vcpus?

[David Hopwood]

Keir Fraser wrote:
> Also I'd change the appearances of the mask in the dom0 interface from:
>  cpumap_t cpumap[MAX_VIRT_CPUS]
> And:
>  cpumap_t *cpumap
> To:
>  u8 vcpumask[MAX_VCPUID/8]

u8 vcpumask[MAX_VCPUID/8 + 1]

> Having an array of '_t' is odd as each entry in the array is really
> logically just a single bit. And I would like to separate the size of
> the array at the admin interfaces from MAX_VIRT_CPUS: it would be nice
> to be able to increase MAX_VIRT_CPUS without changing the admin
> interface. That's why I'd add something like MAX_VCPUID (probably set
> it to 256).

255 (if it's the maximum ID rather than the number of IDs).

-- 
David Hopwood <david.nospam.hopwood@blueyonder.co.uk>

Re: How to query the number of vcpus?

[Keir Fraser]

>> 1.   
>> http://lists.xensource.com/archives/html/xen-devel/2005-04/ 
>> msg00412.html
>> 2.   
>> http://lists.xensource.com/archives/html/xen-devel/2005-04/ 
>> msg00890.html
>
> There's already a cpumask in Xen for physical CPUs (or will be -- it's
> called cpuset right now, but that's going to change to fit the Linux
> convention).
>
> It probably makes sense to rename cpumap to vcpumask, or at least to
> vcpumap.

Just ignore my previous email -- I completely misunderstood that the  
cpumaps you added really are physical cpu maps and not virtual cpu  
maps.

cpumap_t is a reasonable name as at least it won't conflict with type  
names internal to Xen.

However, why pass a pointer to a cpumap_t into PINCPU? And should we  
cater now for systems with too many cpus to be represented by a long?

  -- Keir

RE: How to query the number of vcpus?

[Ian Pratt]

> cpumap_t is a reasonable name as at least it won't conflict 
> with type names internal to Xen.
> 
> However, why pass a pointer to a cpumap_t into PINCPU? And 
> should we cater now for systems with too many cpus to be 
> represented by a long?

I think so -- might as well get the interface right. I was thinking
something like the fd mask that get's passed to select, where you
sepcify the highest set bit.

Ian

Re: How to query the number of vcpus?

[Keir Fraser]

On 30 Apr 2005, at 19:03, Ian Pratt wrote:

>> cpumap_t is a reasonable name as at least it won't conflict
>> with type names internal to Xen.
>>
>> However, why pass a pointer to a cpumap_t into PINCPU? And
>> should we cater now for systems with too many cpus to be
>> represented by a long?
>
> I think so -- might as well get the interface right. I was thinking
> something like the fd mask that get's passed to select, where you
> sepcify the highest set bit.

In some cases Xen returns a cpumap to the guest (in getdomaininfo). In 
that case the fd mask trick won't really work as the guest needs to 
know in advance the biggetst map that Xen may return. Anyway, a map of 
256 bits is going to give us a fair amount of headroom I think.

  -- Keir

RE: How to query the number of vcpus?

[Ian Pratt]

 
> > I think so -- might as well get the interface right. I was thinking 
> > something like the fd mask that get's passed to select, where you 
> > sepcify the highest set bit.
> 
> In some cases Xen returns a cpumap to the guest (in 
> getdomaininfo). In that case the fd mask trick won't really 
> work as the guest needs to know in advance the biggetst map 
> that Xen may return. Anyway, a map of
> 256 bits is going to give us a fair amount of headroom I think.

Why do we need to ever return a cpumap to the guest? Being able to read
CPU pining information out isn't very useful, and both CPUs and VCPUs
can be enumerated in the normal manner. Is there something I've
forgotten?

Ian

Re: How to query the number of vcpus?

[Ryan Harper]

* Ian Pratt <m+Ian.Pratt@cl.cam.ac.uk> [2005-05-02 09:07]:
>  
> > > I think so -- might as well get the interface right. I was thinking 
> > > something like the fd mask that get's passed to select, where you 
> > > sepcify the highest set bit.
> > 
> > In some cases Xen returns a cpumap to the guest (in 
> > getdomaininfo). In that case the fd mask trick won't really 
> > work as the guest needs to know in advance the biggetst map 
> > that Xen may return. Anyway, a map of
> > 256 bits is going to give us a fair amount of headroom I think.
> 
> Why do we need to ever return a cpumap to the guest? Being able to read
> CPU pining information out isn't very useful, and both CPUs and VCPUs
> can be enumerated in the normal manner. Is there something I've
> forgotten?

If I were writing a load-balancing app I'd like to know where a domains'
vcpus are pinnable.  You had previously requested the cpumap to support
setting a list of physical cpus on which a vcpu can run.  It seems odd
(to me) to allow the setting of a variable, but to not be able to see what
the current value is.  Such an interface requires me to always send a
pin operation to ensure the cpumap is set how I want since I cannot
confirm the current value.

-- 
Ryan Harper
Software Engineer; Linux Technology Center
IBM Corp., Austin, Tx
(512) 838-9253   T/L: 678-9253
ryanh@us.ibm.com

Re: How to query the number of vcpus?

[Keir Fraser]

On 2 May 2005, at 16:21, Ryan Harper wrote:

> If I were writing a load-balancing app I'd like to know where a 
> domains'
> vcpus are pinnable.  You had previously requested the cpumap to support
> setting a list of physical cpus on which a vcpu can run.  It seems odd
> (to me) to allow the setting of a variable, but to not be able to see 
> what
> the current value is.  Such an interface requires me to always send a
> pin operation to ensure the cpumap is set how I want since I cannot
> confirm the current value.

Makes sense to me. cpumap_t should be bigger than a long then, to allow 
for future growth.

  -- Keir