[PATCH 1/2] KVM: Add KVM_EXIT_RESET to user space API header

[Alexander Graf]

On 10.12.2013, at 17:07, Anup Patel <anup at brainfault.org> wrote:

> On Tue, Dec 10, 2013 at 9:19 PM, Alexander Graf <agraf at suse.de> wrote:
>> 
>> On 10.12.2013, at 05:23, Anup Patel <anup at brainfault.org> wrote:
>> 
>>> On Tue, Dec 10, 2013 at 7:43 AM, Alexander Graf <agraf at suse.de> wrote:
>>>> 
>>>> On 25.11.2013, at 16:49, Anup Patel <anup.patel at linaro.org> wrote:
>>>> 
>>>>> Currently, we don't have an exit reason for VM reset emulation
>>>>> in user space hence this patch adds exit reason KVM_EXIT_RESET
>>>>> for this purpose.
>>>>> 
>>>>> This newly added KVM_EXIT_RESET will be used by KVM arm/arm64
>>>>> in-kernel PSCI support to reset VMs.
>>>>> 
>>>>> Signed-off-by: Anup Patel <anup.patel at linaro.org>
>>>>> Signed-off-by: Pranavkumar Sawargaonkar <pranavkumar at linaro.org>
>>>>> ---
>>>>> include/uapi/linux/kvm.h |    1 +
>>>>> 1 file changed, 1 insertion(+)
>>>>> 
>>>>> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
>>>>> index 902f124..64a04cc 100644
>>>>> --- a/include/uapi/linux/kvm.h
>>>>> +++ b/include/uapi/linux/kvm.h
>>>>> @@ -171,6 +171,7 @@ struct kvm_pit_config {
>>>>> #define KVM_EXIT_WATCHDOG         21
>>>>> #define KVM_EXIT_S390_TSCH        22
>>>>> #define KVM_EXIT_EPR              23
>>>>> +#define KVM_EXIT_RESET            24
>>>> 
>>>> I have to admit that I'm not particularly happy with the exit name. It's not obvious from the name under which circumstances it gets triggered. Does it get triggered when a core level reset happens? Does it get triggered when a system level reset happened? When the guest requests one?
>>> 
>>> The KVM_EXIT_RESET gets triggered when system level reset is
>>> initiated by VCPU. For arm/arm64, this is through SYSTEM_RESET
>>> PSCI call. In KVM x86 SVM/VMX, we have KVM_EXIT_SHUTDOWN
>>> being used for system shutdown which we have re-used for arm/arm64.
>> 
>> Yeah, that name already did mislead you once :).
>> 
>> KVM_EXIT_SHUTDOWN happens on
>> 
>>  * triple fault
>>  * CPU internal severe problems
>> 
>> the latter is defined as:
>> 
>> In contrast, an error that cannot be contained and is of such severity that it has compromised the continued operation of a processor core requires immediate action to terminate system processing and may result in a hardware-enforced shutdown. In the shutdown state, the execution of instructions by that processor core is halted. See Section 8.2.9 “#DF—Double-Fault Exception (Vector 8)” on page 220 for a description of the shutdown processor state.
>> 
>> Triple faults are used commonly in 286 code to switch from PG to real mode. So they _have_ to be emulated as core reset. Otherwise you break old guests.
>> 
>> However, the scope of this exit is definitely vcpu wide. What you are looking for is a system wide notification. Commonly this happens through MMIO, but I can see why you wouldn't want that with PSCI interpreted in the kernel. That's why I asked you to create a completely new one to not add up the the confusion.
> 
> Thanks for the info on the x86 part.
> 
> I think all this info should have been part of KVM api documentation
> for KVM_EXIT_SHUTDOWN.
> 
>> 
>>> 
>>>> 
>>>> I know what it does, but I find the name too generic for what it is. What you're really doing is introduce a new communication channel in parallel to MMIO / PIO / HCALL which is only used for system level reset / shutdown today.
>>>> 
>>>> Can we treat it as such? Could you please make this a common exit number that's called something like
>>>> 
>>>> KVM_EXIT_SYSTEM_EVENT
>>>> 
>>>> with a parameter that can either be TRIGGER_SHUTDOWN or TRIGGER_RESET.
>>>> 
>>>> That way it's obvious what's going on and people don't get confused.
>>> 
>>> I don't foresee any system level operations other than SHUTDOWN
>>> and RESET to be handled from KVM in-kernel code but I might be
>>> wrong.
>> 
>> The good but about the EXIT_SYSTEM_EVENT is that it's immediately obvious that we're not talking about a vcpu local event. But I'm open to better names.
>> 
>>> May be we can rename KVM_EXIT_SHUTDOWN and KVM_EXIT_RESET
>>> to KVM_EXIT_SYSTEM_SHUTDOWN and KVM_EXIT_SYSTEM_RESET ??
>> 
>> You definitely can not rename KVM_EXIT_SHUTDOWN. It's part of the KVM API. In fact, I think it's a bad idea to even reuse the name as it clearly works on vcpu level.
> 
> Sure, it makes sense to avoid use of KVM_EXIT_SHUTDOWN
> in arm/arm64.
> 
> How about adding exit reasons KVM_EXIT_SYSTEM_RESET
> and KVM_EXIT_SYSTEM_SHUTDOWN ?
> 
> These exit reasons will be used for arm/arm64 but can also be
> used by other architectures if they want.

I don't think we'll be able to reuse anything for other archs. For hcall PPC for example, we want to keep the hcall number scheme as the same between guest <-> kvm and kvm <-> qemu. That makes the overall logic easier, as the hcall number space is already properly standardized.


Alex