On 01/11/2017 04:13 AM, Andrea Bolognani wrote:
On Tue, 2017-01-10 at 15:51 -0500, Laine Stump wrote:
ARGH! ARGH to you too :)
(Talk like a Pirate Day isn't until Sept 19.)
In my final rebase before pushing, I pulled in Andrea's patches that switch aarch64/virt to using PCI by default, and the test case for that resulted in a make check failure:
564) QEMU XML-2-ARGV aarch64-virtio-pci-default ... libvirt: QEMU Driver error : unsupported configuration: 'multifunction=on' is not supported with this QEMU binary FAILED
Is it really true that the aarch64 qemu doesn't support multifunction devices? If so, that really needs to be fixed. In the meantime, this means I still can't push my patches, because doing so will break aarch64. I think the only problem is that current aarch64 test cases don't enable the QEMU_CAPS_PCI_MULTIFUNCTION capability.
Yes. That's correct. I don't know what I was (or wasn't) thinking. It's all fixed now, and ready to push.
I will try this on actual hardware now and get back to you, but really that's all you should have to do in order not to break the test suite with your changes.
I'll try to come up with a patch to conditionalize AGGREGATE_SLOT on support for multifunction (which I suppose I should have done to begin with, but I wouldn't have expected that a platform that supports PCIe doesn't support multifunction devices :-/) That's a good idea in any case, yes.
Yeah, I'm planning on doing that in a cleanup patch. I wrote one last night but it was somehow broken and I was too tired to figure out why, so I'll save it for later.
However, realistically speaking, I think your initial intuition that all platforms that support PCIe also support multifunction still holds as far as real-world scenarios are concerned.
-- Andrea Bolognani / Red Hat / Virtualization
-- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list