[vfio-users] RocketRaid 1144c / AsMedia 1042a

[Alex Williamson]

On Tue, Jul 5, 2016 at 7:45 PM, Nick Rodgers <dotbatman at nerdfortress.com>
wrote:

> Thanks for the response, Alex. Quirking the reset fixed the device (Sonnet
> Allegro Pro) between reboots!
>

Great, should we add that quirk upstream?  Looking at pciids, I assume you
just added:

DECLARE_PCI_FIXUP_HEADER(0x1b73, 0x1100, quirk_no_bus_reset);

To the quirks.c file, please confirm.


> So far they seem stable in my VMs, but dmesg is generating these messages
> when I first start the VM:
>
> vfio-pci 0000:04:00.0: irq bypass producer (token ffff880371a19c80)
> registration fails: -22 (several times)
> kvm: zapping shadow pages for mmio generation wraparound
> kvm: zapping shadow pages for mmio generation wraparound
> vfio-pci 0000:05:00.0: irq bypass producer (token ffff8804441cd000)
> registration fails: -22 (several times)
>
> Are these messages I should be concerned about? Otherwise my testing looks
> good. I've gotten 100+mb/sec transfers on two VMs concurrently using
> different controllers.
>

No, the wraparound is just spam from the kernel, the irq bypass thing is
because you're on AMD and it doesn't get as much testing as Intel, fixed by:

http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=b52f3ed02221252d8ee2c7d756e76fad4a5e84f6
http://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=14717e2031862d9aa2512b24a7df42cf68a977ec


> With regards to the Renesas devices on the Startech board, they were all
> bound to vfio-pci. Should I try pci-stub instead?
>

The suggestion with pci-stub is simply to use it to prevent xhci-hcd from
claiming the device on the host, so the VM can see it in a "pristine"
state.  You'll of course need to bind it to vfio-pci prior to assigning to
the guest.  Try it and see if it helps.  Thanks,

Alex
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/vfio-users/attachments/20160713/50ee771f/attachment.htm>