Re: ppc64le kunit test failure: guest_state_buffer_test

[Vaibhav Jain <vaibhav@linux.ibm.com>]

Hi Eric,

I just successfully tested this kunit test on kernel tag next-20260602
and qemu master HEAD == 405c32d2b18a("Merge tag 'pull-tpm-2026-06-01-1'
of https://github.com/stefanberger/qemu-tpm into staging") [1]

If you are using Qemu >= v10.0.0 then the kunit test failure you
reported is due to Qemu not enabling 'cap-nested-papr' capability which
is needed to enable KVM-HV Apiv2 support in Qemu.

Without 'cap-nested-papr' Qemu doesnt register the H_GUEST_GET_STATE
Hcall which this kunit test relies on. Hence it returns -2 (H_FUNCTION)
for this unsupported HCall. The failure of kunit 'test_gs_hostwide_msg'
is expected as the underlying hypervisor doesnt have support nested-papr
APIv2 capability.

To fix this please enable qemu-system-ppc64 machine's capability
cap-nested-papr to enable nested-papr APIv2 support.

Below is the test log showing successful 'guest_state_buffer_test' kunit
test passing :

$ ./qemu-system-ppc64 --version
QEMU emulator version 11.0.50 (v11.0.0-1621-g405c32d2b1)
Copyright (c) 2003-2026 Fabrice Bellard and the QEMU Project developers

# kunit test is specific to nested-papr APIv2 so qemu cap-nested-papr needs to be enabled
$ ./qemu-system-ppc64 -display none -nographic -kernel ~/linux/vmlinux
-machine pseries,cap-nested-papr=true

<snip>
Booting from memory...                                       
OF stdout device is: /vdevice/vty@71000000                                                                                       
Preparing to boot Linux version 7.1.0-rc6-next-20260602 (vaibhav@*********) (gcc (GCC) 16.1.1 20260515 (Red Hat 16.1.1-2), GNU ld version 2.46-3.fc44) #2 SMP PREEMPT Wed Jun  3 04:11:38 CDT 2026
Detected machine type: 0000000000000101

<snip>
[    4.335850][    T1]     KTAP version 1
[    4.335896][    T1]     # Subtest: guest_state_buffer_test
[    4.335946][    T1]     # module: test_guest_state_buffer
[    4.335970][    T1]     1..7
[    4.337296][    T1]     ok 1 test_creating_buffer
[    4.338998][    T1]     ok 2 test_adding_element
[    4.341996][    T1]     ok 3 test_gs_bitmap
[    4.343406][    T1]     ok 4 test_gs_parsing
[    4.345607][    T1]     ok 5 test_gs_msg
[    4.347247][    T1]     ok 6 test_gs_hostwide_msg
[    4.348012][  T131]     # test_gs_hostwide_counters: Guest Heap Size=0 bytes
[    4.348183][  T131]     # test_gs_hostwide_counters: Guest Heap Size Max=0 bytes
[    4.348350][  T131]     # test_gs_hostwide_counters: Guest Page-table Size=0 bytes
[    4.348653][  T131]     # test_gs_hostwide_counters: Guest Page-table Size Max=0 bytes
[    4.348813][  T131]     # test_gs_hostwide_counters: Guest Page-table Reclaim Size=0 bytes
[    4.349354][    T1]     ok 7 test_gs_hostwide_counters
[    4.349569][    T1] # guest_state_buffer_test: pass:7 fail:0 skip:0 total:7
[    4.349635][    T1] # Totals: pass:7 fail:0 skip:0 total:7
[    4.349708][    T1] ok 4 guest_state_buffer_test


Can you try adding 'cap-nested-papr=true' to the Qemu machine you are
using and see if the problem resolves for you. If it persists can you
please share the Qemu command line you are using.

[1] https://gitlab.com/qemu-project/qemu/-/commit/405c32d2b18a683ba36301351af75125d9afda08


Eric Biggers <ebiggers@kernel.org> writes:

> On Wed, Jun 03, 2026 at 01:03:09PM +0530, Vaibhav Jain wrote:
>> Hi Eric,
>> 
>> Thanks for trying and reporting this. This kunit test depends on
>> availablility of Qemu commit 5f7d861e("spapr: nested: Add support for
>> reporting Hostwide state counter ") [1] that was merged in v10.0.0.
>> 
>> Since you havent mentioned the qemu version used I assume its a version
>> < v10.0.0 . With the qemu patch available you should see this test
>> passing with results similar to as originally described in original
>> cover letter of the patch series at [2] that introduced this kunit test.
>> 
>> [1] https://gitlab.com/qemu-project/qemu/-/commit/5f7d861e65d90e0446b8f22a0bc859a5d8058ea6
>> 
>> [2] https://lore.kernel.org/all/20250416162740.93143-1-vaibhav@linux.ibm.com/
>
> Nope, it fails even on the master branch of QEMU.
>
> - Eric

-- 
Cheers
~ Vaibhav