Re: [RFC Patch v3 18/18] x86/hvm: Always set pending event injection when loading VMC[BS] state.

[Aravind Gopalakrishnan <aravind.gopalakrishnan@amd.com>]

On 9/5/2014 4:11 AM, Wen Congyang wrote:
> In colo mode, secondary vm is running, so VM_ENTRY_INTR_INFO may
> valid before restoring vmcs. If there is no pending event after
> restoring vm, we should clear it.
>
> Signed-off-by: Wen Congyang <wency@cn.fujitsu.com>
>
> Also clear pending software exceptions.
> Copy the fix to SVM as well.
>
> Signed-off-by: Tim Deegan <tim@xen.org>
>
> vmx part:
> Cc: Jun Nakajima <jun.nakajima@intel.com>
> Acked-by: Kevin Tian <kevin.tian@intel.com>
>
> svm part:
> Cc: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@amd.com>
> Cc: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> ---
>   xen/arch/x86/hvm/svm/svm.c | 15 ++++++++-------
>   xen/arch/x86/hvm/vmx/vmx.c | 25 ++++++++++++-------------
>   2 files changed, 20 insertions(+), 20 deletions(-)
>
> diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
> index b5188e6..053e511 100644
> --- a/xen/arch/x86/hvm/svm/svm.c
> +++ b/xen/arch/x86/hvm/svm/svm.c
> @@ -321,16 +321,17 @@ static int svm_vmcb_restore(struct vcpu *v, struct hvm_hw_cpu *c)
>           vmcb_set_h_cr3(vmcb, pagetable_get_paddr(p2m_get_pagetable(p2m)));
>       }
>   
> -    if ( c->pending_valid )
> +    if ( c->pending_valid &&
> +         hvm_event_needs_reinjection(c->pending_type, c->pending_vector) )
>       {
>           gdprintk(XENLOG_INFO, "Re-injecting %#"PRIx32", %#"PRIx32"\n",
>                    c->pending_event, c->error_code);
> -
> -        if ( hvm_event_needs_reinjection(c->pending_type, c->pending_vector) )
> -        {
> -            vmcb->eventinj.bytes = c->pending_event;
> -            vmcb->eventinj.fields.errorcode = c->error_code;
> -        }
> +        vmcb->eventinj.bytes = c->pending_event;
> +        vmcb->eventinj.fields.errorcode = c->error_code;
> +    }
> +    else
> +    {
> +        vmcb->eventinj.bytes = 0;
>       }
>   
>       vmcb->cleanbits.bytes = 0;
>

Hi,
You mention that this 'fix' is just copied for svm. But you have not 
seen the problem of "VM_ENTRY_INTR_INFO may be valid" (whose svm 
equivalent is "vmcb->eventinj.bytes is valid").
My concern is that we should test colo mode for svm first, since, if the 
problem is never really seen on svm, then fix may not be _necessary_

At this point, my problems are with test setups. I can help testing 
scenarios, but as Wen had mentioned, 'colo testing' might be the way to 
test.
So, if I can get some pointers to how I can reproduce the issue at hand, 
then it would be very helpful.

(Tim had mentioned we could try to simulate it by running a guest that 
takes lot of faults and save-restore another guest over it. However,
I am not having much luck following this route. I got a hvm guest to 
continuously take SW exceptions on all vcpus and tried to save-restore.
I can't see vmcb->eventinj.bytes containing any valid info.)

Thanks,
-Aravind.