[libvirt] [PATCH] hostdev: skip ACS check when using VFIO for device assignment

[Laine Stump]

On 08/26/2015 04:13 AM, Martin Kletzander wrote:
> On Wed, Aug 26, 2015 at 02:07:32AM -0400, Laine Stump wrote:
>> The ACS checks are meaningless when using the more modern VFIO driver
>> for device assignment since VFIO has its own more complete and exact
>> checks, but I didn't realize that when I added support for VFIO. This
>> patch eliminates the ACS check when preparing PCI devices for
>> assignment if VFIO is being used.
>>
>> This resolves:
>>
>>  https://bugzilla.redhat.com/show_bug.cgi?id=1256486
>
> I don't remember how it's possible as neatly as I managed to do
> somehow, but I believe git has good support for 'Key: Value' on one
> line so that you can then easily see, for example, patches that have
> Key == Resolves etc., so I'm using Resolves: link-to-bz, but as I
> said, I forgot how to use that formatting neatly (it was almost like
> one parameter only).  So this is just an FYI, no need to change it.

Interesting, I didn't know about that.

>
>> ---
>> src/util/virhostdev.c | 5 ++---
>> 1 file changed, 2 insertions(+), 3 deletions(-)
>>
>> diff --git a/src/util/virhostdev.c b/src/util/virhostdev.c
>> index 529753c..0bc5120 100644
>> --- a/src/util/virhostdev.c
>> +++ b/src/util/virhostdev.c
>> @@ -1,6 +1,6 @@
>> /* virhostdev.c: hostdev management
>>  *
>> - * Copyright (C) 2006-2007, 2009-2013 Red Hat, Inc.
>> + * Copyright (C) 2006-2007, 2009-2015 Red Hat, Inc.
>>  * Copyright (C) 2006 Daniel P. Berrange
>>  * Copyright (C) 2014 SUSE LINUX Products GmbH, Nuernberg, Germany.
>>  *
>> @@ -230,7 +230,6 @@
>> virHostdevGetPCIHostDeviceList(virDomainHostdevDefPtr *hostdevs, int
>> nhostdevs)
>>             virObjectUnref(list);
>>             return NULL;
>>         }
>> -
>>         if (virPCIDeviceListAdd(list, dev) < 0) {
>>             virPCIDeviceFree(dev);
>>             virObjectUnref(list);
>> @@ -579,7 +578,7 @@ virHostdevPreparePCIDevices(virHostdevManagerPtr
>> hostdev_mgr,
>>         struct virHostdevIsPCINodeDeviceUsedData data = {hostdev_mgr,
>> dom_name,
>>                                                          usesVfio};
>>
>> -        if (!virPCIDeviceIsAssignable(dev, strict_acs_check)) {
>> +        if (!usesVfio && !virPCIDeviceIsAssignable(dev,
>> strict_acs_check)) {
>
> I suppose we could unset the strict check in case of VFIO as that
> would work too, but was not meant for this job.

Just in case someone were to be incredibly intelligent and try to use
legacy assignment and vfio in the same system, we can't just disable the
global strict_acs check. Besides, even if the strict ACS check is
disabled, we still go poking around in sysfs (calling
virPCIDeviceIsBehindSwitchLackingACS()), which we really *don't* want to
do, as that is causing problems for people attempting to do vfio device
assignment from a session-mode libvirt (which should otherwise work as
long as the device is pre-bound to vfio-pci and all the device node
permissions are set). Who wants to do vfio device assignment from
session-mode libvirt? gnome-boxes of course! They want to do GPU
passthrough.

>
> ACK as-is.

Thanks!