Re: [PATCH] KVM: arm64: nv: Set ISTATUS for emulated timers, If timer expired

[Eric Auger <eauger@redhat.com>]

Hi Ganapatrao, Marc,

On 12/19/24 10:30 AM, Ganapatrao Kulkarni wrote:
> 
> 
> On 09-12-2024 11:04 pm, Eric Auger wrote:
>> Hi Marc,
>>
>> On 12/9/24 18:30, Marc Zyngier wrote:
>>> On Mon, 09 Dec 2024 16:46:30 +0000,
>>> Eric Auger <eauger@redhat.com> wrote:
>>>>
>>>> Hi,
>>>>
>>>> On 12/9/24 16:39, Ganapatrao Kulkarni wrote:
>>>>>
>>>>>
>>>>> On 09-12-2024 06:50 pm, Marc Zyngier wrote:
>>>>>> On Mon, 09 Dec 2024 12:25:34 +0000,
>>>>>> Ganapatrao Kulkarni <gankulkarni@os.amperecomputing.com> wrote:
>>>>>>>>>
>>>>>>>>> During automated testing of Nested Virtualization using
>>>>>>>>> avocado-vt,
>>>>>>>>
>>>>>>>> Which is not merged upstream. So what branch are you using?
>>>>>>>> Based on
>>>>>>>> what kernel version? On what HW? With which virtualisation
>>>>>>>> features?
>>>>>>>>
>>>>>>>
>>>>>>> Testing is done on Ampere's AmpereOne platform using 6.10 based
>>>>>>> kernel
>>>>>>> with NV patches from your repo.
>>>>>>
>>>>>> Grmbl... *Which* patches? At least give me the SHA1 of the branch,
>>>>>> because I have no idea what you are running. And 6.10 is definitely
>>>>>> not something I care about. If you're using the NV patches, the
>>>>>> *minimum* you should run is 6.13-rc1, because that's what the current
>>>>>> code is based on.
>>>>>>
>>>>>
>>>>> I tried 6.13-rc1 based nv-next branch today, which failed to boot
>>>>> UEFI as L1. Yet to debug this.
>>>>
>>>> I confirm am stuck with the same issue with nv-next on AmpereOne.
>>>
>>> All I can say is that it Works For Me (TM) on M2 and Snapdragon, using
>>> kvmtool, a fresh EDK2 built by Ard, and with the guest running purely
>>> VHE (FEAT_E2H0 not implemented).
>>>
>>> I haven't tried QEMU, I don't have an AmpereOne. I have also spent way
>>> too much time on this recently. So I'm afraid people interested in
>>> making this particular combination work will have to debug it.
>>
>> yes I do agree. I am currently setting up kvmtool to double check and I
>> will debug the qemu issue.
> 
> I could boot Guest-hypervisor using QEMU with -bios and nv-next as L0
> with edk2 (QEMU_EFI.fd) built from upstream tree and fedora-41 as Guest-
> Hypervisor.
> 
> The issue was disappeared when upgraded from fc39 to fc41. Same was
> experienced with the kvmtool boot as well(KVMTOOL_EFI.fd).
I also confirm that using a mailine edk2 fixed the issues I faced
previously (fed/rhel L1 guest not booting).

I used Marc's nv-next branch and qemu rebase.

With a rhel L1 guest I can now boot buildroot, debian and rhel guest as
L2 (feat mainline edk2). For rhel, I tested different kinds of page size
combinations for L1/L2 (4k and 64kB) and it worked.

I tested on AmpereOne and Grace-Hopper systems

Eric
>