> On 13 Aug 2020, at 18:28, Julien Grall <jul...@xen.org> wrote:
>
> Hi Bertrand,
>
> On 31/07/2020 14:16, Bertrand Marquis wrote:
>>> On 30 Jul 2020, at 22:50, Julien Grall <jul...@xen.org> wrote:
>>> On 30/07/2020 11:24, Bertrand Marquis wrote:
>>>> At the moment on Arm, a Linux guest running with KTPI enabled will
>>>> cause the following error when a context switch happens in user mode:
>>>> (XEN) p2m.c:1890: d1v0: Failed to walk page-table va 0xffffff837ebe0cd0
>>>> The error is caused by the virtual address for the runstate area
>>>> registered by the guest only being accessible when the guest is running
>>>> in kernel space when KPTI is enabled.
>>>> To solve this issue, this patch is doing the translation from virtual
>>>> address to physical address during the hypercall and mapping the
>>>> required pages using vmap. This is removing the conversion from virtual
>>>> to physical address during the context switch which is solving the
>>>> problem with KPTI.
>>>
>>> To echo what Jan said on the previous version, this is a change in a stable
>>> ABI and therefore may break existing guest. FAOD, I agree in principle with
>>> the idea. However, we want to explain why breaking the ABI is the *only*
>>> viable solution.
>>>
>>> From my understanding, it is not possible to fix without an ABI breakage
>>> because the hypervisor doesn't know when the guest will switch back from
>>> userspace to kernel space. The risk is the information provided by the
>>> runstate wouldn't contain accurate information and could affect how the
>>> guest handle stolen time.
>>>
>>> Additionally there are a few issues with the current interface:
>>> 1) It is assuming the virtual address cannot be re-used by the userspace.
>>> Thanksfully Linux have a split address space. But this may change with KPTI
>>> in place.
>>> 2) When update the page-tables, the guest has to go through an invalid
>>> mapping. So the translation may fail at any point.
>>>
>>> IOW, the existing interface can lead to random memory corruption and
>>> inacurracy of the stolen time.
>> I agree but i am not sure what you want me to do here.
>> Should i add more details in the commit message ?
>>>
>>>> This is done only on arm architecture, the behaviour on x86 is not
>>>> modified by this patch and the address conversion is done as before
>>>> during each context switch.
>>>> This is introducing several limitations in comparison to the previous
>>>> behaviour (on arm only):
>>>> - if the guest is remapping the area at a different physical address Xen
>>>> will continue to update the area at the previous physical address. As
>>>> the area is in kernel space and usually defined as a global variable this
>>>> is something which is believed not to happen. If this is required by a
>>>> guest, it will have to call the hypercall with the new area (even if it
>>>> is at the same virtual address).
>>>> - the area needs to be mapped during the hypercall. For the same reasons
>>>> as for the previous case, even if the area is registered for a different
>>>> vcpu. It is believed that registering an area using a virtual address
>>>> unmapped is not something done.
>>>
>>> This is not clear whether the virtual address refer to the current vCPU or
>>> the vCPU you register the runstate for. From the past discussion, I think
>>> you refer to the former. It would be good to clarify.
>> Ok i will try to clarify.
>>>
>>> Additionally, all the new restrictions should be documented in the public
>>> interface. So an OS developper can find the differences between the
>>> architectures.
>>>
>>> To answer Jan's concern, we certainly don't know all the guest OSes
>>> existing, however we also need to balance the benefit for a large majority
>>> of the users.
>>>
>>> From previous discussion, the current approach was deemed to be acceptable
>>> on Arm and, AFAICT, also x86 (see [1]).
>>>
>>> TBH, I would rather see the approach to be common. For that, we would an
>>> agreement from Andrew and Jan in the approach here. Meanwhile, I think this
>>> is the best approach to address the concern from Arm users.
>> From this I get that you want me to document the specific behaviour on Arm
>> on the public header describing the hypercall, right ?
>
> Yes please. The public header is usually where an OS developper will look for
> details. Although, at the moment, the documentation is not very great as you
> often have to dig in Xen code to understand how it is meant to work :(. But
> we are trying to improve that.
Ok i will add some comments in the header.
Cheers
Bertrand
>
> Cheers,
>
> --
> Julien Grall
