Re: [PATCH v3 0/4] KVM: x86: hyper-v: Conditionally allow SynIC with APICv/AVIC

[Paolo Bonzini <pbonzini@redhat.com>]

On 14/06/21 11:51, Maxim Levitsky wrote:
> On Mon, 2021-06-14 at 09:40 +0200, Vitaly Kuznetsov wrote:
>> Maxim Levitsky <mlevitsk@redhat.com> writes:
>>
>>> On Wed, 2021-06-09 at 17:09 +0200, Vitaly Kuznetsov wrote:
>>>> Changes since v2:
>>>> - First two patches got merged, rebase.
>>>> - Use 'enable_apicv = avic = ...' in PATCH1 [Paolo]
>>>> - Collect R-b tags for PATCH2 [Sean, Max]
>>>> - Use hv_apicv_update_work() to get out of SRCU lock [Max]
>>>> - "KVM: x86: Check for pending interrupts when APICv is getting disabled"
>>>>    added.
>>>>
>>>> Original description:
>>>>
>>>> APICV_INHIBIT_REASON_HYPERV is currently unconditionally forced upon
>>>> SynIC activation as SynIC's AutoEOI is incompatible with APICv/AVIC. It is,
>>>> however, possible to track whether the feature was actually used by the
>>>> guest and only inhibit APICv/AVIC when needed.
>>>>
>>>> The series can be tested with the followin hack:
>>>>
>>>> diff --git a/arch/x86/kvm/cpuid.c b/arch/x86/kvm/cpuid.c
>>>> index 9a48f138832d..65a9974f80d9 100644
>>>> --- a/arch/x86/kvm/cpuid.c
>>>> +++ b/arch/x86/kvm/cpuid.c
>>>> @@ -147,6 +147,13 @@ void kvm_update_cpuid_runtime(struct kvm_vcpu *vcpu)
>>>>                                             vcpu->arch.ia32_misc_enable_msr &
>>>>                                             MSR_IA32_MISC_ENABLE_MWAIT);
>>>>          }
>>>> +
>>>> +       /* Dirty hack: force HV_DEPRECATING_AEOI_RECOMMENDED. Not to be merged! */
>>>> +       best = kvm_find_cpuid_entry(vcpu, HYPERV_CPUID_ENLIGHTMENT_INFO, 0);
>>>> +       if (best) {
>>>> +               best->eax &= ~HV_X64_APIC_ACCESS_RECOMMENDED;
>>>> +               best->eax |= HV_DEPRECATING_AEOI_RECOMMENDED;
>>>> +       }
>>>>   }
>>>>   EXPORT_SYMBOL_GPL(kvm_update_cpuid_runtime);
>>>>   
>>>> Vitaly Kuznetsov (4):
>>>>    KVM: x86: Use common 'enable_apicv' variable for both APICv and AVIC
>>>>    KVM: x86: Drop vendor specific functions for APICv/AVIC enablement
>>>>    KVM: x86: Check for pending interrupts when APICv is getting disabled
>>>>    KVM: x86: hyper-v: Deactivate APICv only when AutoEOI feature is in
>>>>      use
>>>>
>>>>   arch/x86/include/asm/kvm_host.h |  9 +++++-
>>>>   arch/x86/kvm/hyperv.c           | 51 +++++++++++++++++++++++++++++----
>>>>   arch/x86/kvm/svm/avic.c         | 14 ++++-----
>>>>   arch/x86/kvm/svm/svm.c          | 22 ++++++++------
>>>>   arch/x86/kvm/svm/svm.h          |  2 --
>>>>   arch/x86/kvm/vmx/capabilities.h |  1 -
>>>>   arch/x86/kvm/vmx/vmx.c          |  2 --
>>>>   arch/x86/kvm/x86.c              | 18 ++++++++++--
>>>>   8 files changed, 86 insertions(+), 33 deletions(-)
>>>>
>>>
>>> Hi!
>>>
>>> I hate to say it, but at least one of my VMs doesn't boot amymore
>>> with avic=1, after the recent updates. I'll bisect this soon,
>>> but this is likely related to this series.
>>>
>>> I will also review this series very soon.
>>>
>>> When the VM fails, it hangs on the OVMF screen and I see this
>>> in qemu logs:
>>>
>>> KVM: injection failed, MSI lost (Operation not permitted)
>>> KVM: injection failed, MSI lost (Operation not permitted)
>>> KVM: injection failed, MSI lost (Operation not permitted)
>>> KVM: injection failed, MSI lost (Operation not permitted)
>>> KVM: injection failed, MSI lost (Operation not permitted)
>>> KVM: injection failed, MSI lost (Operation not permitted)
>>>
>>
>> -EPERM?? Interesting... strace(1) may come handy...
> 
> 
> Hi Vitaly!
>   
> I spent all yesterday debugging this and I found out what is going on:
> (spoiler alert: hacks are bad)
> 
> The call to kvm_request_apicv_update was moved to a delayed work which is fine at first glance
> but turns out that we both don't notice that kvm doesn't allow to update the guest
> memory map from non vcpu thread which is what kvm_request_apicv_update does
> on AVIC.
>   
> The memslot update is to switch between regular r/w mapped dummy page
> which is not really used but doesn't hurt to be there, and between paging entry with
> reserved bits, used for MMIO, which AVIC sadly needs because it is written in the
> spec that AVIC's MMIO despite being redirected to the avic_vapic_bar, still needs a valid
> R/W mapping in the NPT, whose physical address is ignored.
> 
> So, in avic_update_access_page we have this nice hack:
>   
> if ((kvm->arch.apic_access_page_done == activate) ||
> 	    (kvm->mm != current->mm))
> 		goto out;
>   
> So instead of crashing this function just does nothing.
> So AVIC MMIO is still mapped R/W to a dummy page, but the AVIC itself
> is disabled on all vCPUs by kvm_request_apicv_update (with
> KVM_REQ_APICV_UPDATE request)
> 
> So now all guest APIC writes just disappear to that dummy
> page, and we have a guest that seems to run but can't really
> continue.
> 
> The -EPERM in the error message I reported, is just -1, returned by
> KVM_SIGNAL_MSI which is likely result of gross missmatch between
> state of the KVM's APIC registers and that dummy page which contains
> whatever the guest wrote there and what the guest thinks
> the APIC registers are.
> 
> I am curently thinking on how to do the whole thing with
> KVM's requests, I'll try to prepare a patch today.

I'll drop the last two patches in the series.

Paolo