Re: [Xen-ia64-devel] vcpu-pin bug to dom0 ?

[Masaki Kanno <kanno.masaki@xxxxxxxxxxxxxx>]

>Hi Yongkang,
>
>>Hi all,
>>
>>I find a strange issue that if I only assign 1 or 2 vcpus to dom0, I can 
>>not use vcpu-pin to pin the No. 0 vcpu. It will report: "Invalid argument".
>> But it works, if I let Xen0 see all vcpus when booting.
>>
>>For example, in IA32, set (dom0-cpus 1)
>>After Xen0 boot up, "xm vcpu-p 0 0 0" will see errors.
>>If setting (dom0-cpus 0), above command works.
>>
>>In IA64, set dom0_max_vcpus=2 (totally have 16 vcpus)
>>After Xen0 boot up, "xm vcpu-p 0 0 0" will see errors.
>>But "xm vcpu-p 0 1 0" works.
>>
>
>I think that you can solve this problem by applying the following 
>patch, and inputting "xm vcpu-pin 0 0 0" from two consoles at the 
>same time...  Need many retry :-)
>

Hi Yongkang,

Sorry, that is an expected behavior, my patch is far from perfect.
As Keir noted, we need a scheduler fix to correctly solve this.

Not apply my patch:
 +---------------+------------------------------+-----------+
 | Target domain | pCPU that vCPU processing    | Result    |
 |               | 'xm vcpu-pin' command works  |           |
 +---------------+------------------------------+-----------+
 | Domain-0      | == Target pCPU               | Error(22) |
 |               +------------------------------+-----------+
 |               | != Target pCPU               | Error(22) |
 +---------------+------------------------------+-----------+
 | Domain-U      |  -                           | OK        |
 +---------------+------------------------------+-----------+

Apply my patch:
 +---------------+------------------------------+-----------+
 | Target domain | pCPU that vCPU processing    | Result    |
 |               | 'xm vcpu-pin' command works  |           |
 +---------------+------------------------------+-----------+
 | Domain-0      | == Target pCPU               | OK        |
 |               +------------------------------+-----------+
 |               | != Target pCPU               | Error(16) |
 +---------------+------------------------------+-----------+
 | Domain-U      |  -                           | OK        |
 +---------------+------------------------------+-----------+


Best regards,
 Kan


>diff -r 5b9ff5e8653a xen/common/domctl.c
>--- a/xen/common/domctl.c       Sun Aug 27 06:56:01 2006 +0100
>+++ b/xen/common/domctl.c       Mon Aug 28 18:01:28 2006 +0900
>@@ -380,13 +380,6 @@ long do_domctl(XEN_GUEST_HANDLE(xen_domc
> 
>         if ( op->cmd == XEN_DOMCTL_setvcpuaffinity )
>         {
>-            if ( v == current )
>-            {
>-                ret = -EINVAL;
>-                put_domain(d);
>-                break;
>-            }
>-
>             xenctl_cpumap_to_cpumask(
>                 &new_affinity, &op->u.vcpuaffinity.cpumap);
>             ret = vcpu_set_affinity(v, &new_affinity);
>
>
>But, if Domain-0 has one virtual CPU, this problem cannot be solved 
>even if applying this patch. If you are using CREDIT scheduler, 
>'xm vcpu-pin 0 0 0' makes an error by the following line.
>
>
>static int
>csched_vcpu_set_affinity(struct vcpu *vc, cpumask_t *affinity)
>{
>    unsigned long flags;
>    int lcpu;
>
>    if ( vc == current )
>    {
>        /* No locking needed but also can't move on the spot... */
>        if ( !cpu_isset(vc->processor, *affinity) )
>            return -EBUSY;   <---- This!
>
>        vc->cpu_affinity = *affinity;
>    }
>
>
>Hi Keir,
>Do you have good ideas to solve this problem?
>
>
>Best regards,
> Kan
>
>>Best Regards,
>>Yongkang (Kangkang)
>>
>>_______________________________________________
>>Xen-ia64-devel mailing list
>>Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
>>http://lists.xensource.com/xen-ia64-devel
>
>
>
>_______________________________________________
>Xen-ia64-devel mailing list
>Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
>http://lists.xensource.com/xen-ia64-devel


_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel