Re: [Qemu-devel] [PATCH v2 for-2.11 0/2] Improvements for the pxe tester

[Cornelia Huck]

On Fri, 11 Aug 2017 11:49:22 +0200
Cornelia Huck <address@hidden> wrote:

> On Fri, 11 Aug 2017 07:57:54 +0200
> Thomas Huth <address@hidden> wrote:
> 
> > The first patch improves the buffer handling in the pxe tester a
> > little bit by allocating a separate buffer on the heap for each
> > architecture. This also gets rid of the huge pre-initialized
> > array in the tester, shrinking the size of the executable by
> > half of a megabyte!
> > The second patch adds s390x support to the pxe tester. Starting
> > with QEMU 2.10, the guest firmware on s390x can now net-boot via
> > TFTP, too, so we can automatically test this code in the pxe tester.
> > 
> > v2: Adressed Cornelia's review feedback from the first version, e.g.:
> >  - Use g_malloc0() instead of g_malloc()
> >  - Use sizeof with parentheses
> > 
> > Thomas Huth (2):
> >   tests/boot-sector: Do not overwrite the x86 buffer on other
> >     architectures
> >   tests/pxe: Check virtio-net-ccw on s390x
> > 
> >  tests/Makefile.include |  1 +
> >  tests/boot-sector.c    | 61 
> > +++++++++++++++++++++++++++++++++++++-------------
> >  tests/pxe-test.c       |  7 ++++++
> >  3 files changed, 54 insertions(+), 15 deletions(-)
> >   
> 
> Haven't reviewed the updates yet, but tried it out (on my laptop and on
> a z/VM guest). Laptop went fine; on the z/VM guest, I forgot to turn on
> vm.allocate_pgste at first:
> 
> TEST: tests/pxe-test... (pid=56084)
>   /s390x/pxe/virtio-ccw:                                               
> ioctl(KVM_CREATE_VM) failed: 22 Invalid argument
> Host kernel setup problem detected. Please verify:
> - for kernels supporting the switch_amode or user_mode parameters, whether
>   user space is running in primary address space
> - for kernels supporting the vm.allocate_pgste sysctl, whether it is enabled
> qemu-system-s390x: failed to initialize KVM: Invalid argument
> qemu-system-s390x: Back to tcg accelerator
> OK
> PASS: tests/pxe-test
> 
> It does the correct fallback (the need for allocate_pgste will go away
> with future kernels), so this is fine. However, it prompted me to test
> with --disable-tcg:
> 
> TEST: tests/boot-serial-test... (pid=60499)
>   /s390x/boot-serial/s390-ccw-virtio:                                  
> qemu-system-s390x: -machine accel=tcg: No accelerator found
> socket_accept failed: Resource temporarily unavailable
> **
> ERROR:/root/git/qemu/tests/libqtest.c:212:qtest_init_without_qmp_handshake: 
> assertion failed: (s->fd >= 0 && s->qmp_fd >= 0)
> FAIL
> GTester: last random seed: R02S83c8a1860bbd8c2658647407c7070b7e
> (pid=60503)
> FAIL: tests/boot-serial-test
> 
> There are probably more tests that rely on tcg always being available,
> which is no longer true. (Also, the boot-serial test should probably
> check if qemu even started rather than run into in a timeout.)
> 
> [Your patches are fine, but I think we really need to consider tests
> and accelerators.]

A quick test on x86_64 with --disable-tcg showed no further problems
than boot-serial-test, so at least it's only the one for now (and it
might make sense to simply use accel=tcg:kvm for it).