[libvirt] [PATCH 4/4] qemu: don't be as insistent about adding dmi-to-pci-bridge or pci-bridge

Laine Stump laine at laine.org
Mon Apr 25 16:09:16 UTC 2016


On 04/25/2016 10:53 AM, Marcel Apfelbaum wrote:
> On 04/25/2016 05:28 PM, Laine Stump wrote:
>> On 04/23/2016 12:46 PM, Cole Robinson wrote:
>>> On 04/21/2016 02:48 PM, Laine Stump wrote:
>>>> Previously there was no way to have a Q35 domain that didn't have
>>>> these two controllers. This patch skips their creation as long as
>>>> there are some other kinds of pci controllers at index 1 and 2
>>>> (e.g. some pcie-root-port controllers).
>>>>
>>>> I'm hoping that soon we won't add them at all, plugging all devices
>>>> into auto-added pcie-*-port ports instead, but in the meantime this
>>>> makes it easier to experiment with alternative bus hierarchies.
>>>> ---
>>>>   src/qemu/qemu_domain.c | 18 +++++++++++-------
>>>>   1 file changed, 11 insertions(+), 7 deletions(-)
>>>>
>>>> diff --git a/src/qemu/qemu_domain.c b/src/qemu/qemu_domain.c
>>>> index 86b7d13..0b342e2 100644
>>>> --- a/src/qemu/qemu_domain.c
>>>> +++ b/src/qemu/qemu_domain.c
>>>> @@ -1581,14 +1581,18 @@ 
>>>> qemuDomainDefAddDefaultDevices(virDomainDefPtr def,
>>>> VIR_DOMAIN_CONTROLLER_MODEL_PCIE_ROOT)) {
>>>>               goto cleanup;
>>>>           }
>>>> -        if (virDomainDefMaybeAddController(
>>>> -               def, VIR_DOMAIN_CONTROLLER_TYPE_PCI, 1,
>>>> - VIR_DOMAIN_CONTROLLER_MODEL_DMI_TO_PCI_BRIDGE) < 0 ||
>>>> -            virDomainDefMaybeAddController(
>>>> -               def, VIR_DOMAIN_CONTROLLER_TYPE_PCI, 2,
>>>> -               VIR_DOMAIN_CONTROLLER_MODEL_PCI_BRIDGE) < 0) {
>>>> +        /* add a dmi-to-pci-bridge and a pci-bridge if there are 
>>>> no pci controllers
>>>> +         * other than the pcie-root. This is so that there will be 
>>>> hot-pluggable
>>>> +         * PCI slots available
>>>> +         */
>>>> +        if (virDomainControllerFind(def, 
>>>> VIR_DOMAIN_CONTROLLER_TYPE_PCI, 1) < 0 &&
>>>> +            !virDomainDefAddController(def, 
>>>> VIR_DOMAIN_CONTROLLER_TYPE_PCI, 1,
>>>> + VIR_DOMAIN_CONTROLLER_MODEL_DMI_TO_PCI_BRIDGE))
>>>> +            goto cleanup;
>>>> +        if (virDomainControllerFind(def, 
>>>> VIR_DOMAIN_CONTROLLER_TYPE_PCI, 2) < 0 &&
>>>> +            !virDomainDefAddController(def, 
>>>> VIR_DOMAIN_CONTROLLER_TYPE_PCI, 2,
>>>> + VIR_DOMAIN_CONTROLLER_MODEL_PCI_BRIDGE))
>>>>               goto cleanup;
>>>> -        }
>>>>       }
>>>>       if (addDefaultMemballoon && !def->memballoon) {
>>>>
>>> Sounds like another qemuxml2xml test case candidate
>>
>> ... and it turns out this patch doesn't go far enough to have any 
>> useful effect. The problem is that the code that automatically 
>> assigns PCI addresses does its best to always be sure that there is at
>> least one empty hot-pluggable "standard PCI" slot on the system, so 
>> even if we don't explicitly add any pci-bridge to a Q35 domain, the 
>> address-assignment code will automatically add one anyway; and
>> since the pci-bridge requires a standard PCI slot itself, we still 
>> need a dmi-to-pci-bridge anyway (which provides 32 non-hotpluggable 
>> standard PCI slots).
>>
>> I had been aiming to re-do how libvirt sets up the PCI controllers 
>> for Q35, eventually eliminating the current dmi-to-pci-bridge + 
>> pci-bridge in favor of a pure PCIe setup, using pcie-root-ports and
>> pcie-switch-(up|down)stream-ports, and this was going to be the first 
>> step towards that. But after discussing that idea more with Alex 
>> Williamson on Friday, I think we *shouldn't* do it, but just
>> leave the PCI controller setup essentially as it is (one possible 
>> change - allow connecting pci-bridge directly into a pcie-root port, 
>> thus eliminating the dmi-to-pci-bridge)
>>
>
> Connecting the pci-bridge directly into a pcie-root port is a little 
> odd, but I suppose it would work.
> Since I am late to the party, can you please explain why do you want 
> to eliminate  the dmi-to-pci controller ? Because it doesn't support 
> hotplug?

I guess just because it's an extra device that we use only for the 
purpose of connecting the pci-bridge. But definitely the entire reason 
for putting in a pci-bridge is to have non-Express slots that support 
hotplug.

> I have some bad news in that direction. Even the pci-bridge (devices 
> behind it) does not support hotplug on Q35, it is on my todo list.

I now remember somebody telling me that within the last several weeks 
(possibly you?)

So if hotplug doesn't work for pci-bridge slots on Q35, then we 
currently wouldn't be losing anything if we just used the 
dmi-to-pci-bridge slots directly (in either case, hotplug won't work).

Alex had suggested maybe the dmi-to-pci-bridge could be enhanced to 
support hotplug, or possibly a similar but generic controller could be 
added that supported hotplug. What is the difficulty of that vs. fixing 
hotplug support on pci-bridge? (getting both would be best, of course). 
Would it be better to make the slots of the current dmi-to-pci-bridge 
(i82801b11-bridge) hotpluggable? Or to create a new device?

>
>> What prompted the idea to change to pure PCIe controllers was that 
>> qemu allows any PCI device to be plugged into a PCIe slot and it 
>> apparently functions just fine, and there had been occasional
>> complaints that plugging everything into a pci-bridge was somehow 
>> problematic (at one point someone suggested that virtio-net didn't 
>> work correctly on ARM if plugged into a non-Express slot, but I
>> think that was later disproved).
>>
>> The problem with PCIe-only controllers, as Alex pointed out, is that 
>> when a non-Express device is plugged into a PCIe slot, the guest OS 
>> will see an apparently PCIe device that has no PCIe
>> capabilities, and while so far this hasn't caused any problem, there 
>> is no guarantee that it won't - PCIe devices are supposed to have 
>> PCIe capabilities. Since the only emulated qemu device that does
>> this is the NEC UHCI USB3 controller, it sees that we'll need to keep 
>> setting up the pci-bridge and plugging all the rest of the devices in 
>> there.
>
> Actually all virtio devices are now PCI Express devices if virtio-1 is 
> enabled. The QEMU command-line is -device 
> virtio-<dev>-pci,disable-modern=false.

So that defaults to true? How does this relate to the experimental 
x-disable-pcie setting? Are they the same thing? What happens if you set 
this and then plug it into a non-Express slot? Does setting this flag 
change anything else for the device that would, e.g., require a 
different guest driver or other changes to the qemu commandline?

It sounds like I can use this - just check for disable-modern on each 
virtio device when getting qemu capabilities, then prefer a PCIe slot 
for any unaddressed device that has it.


> The only requirement is that the virtio device would not be connected 
> directly to pcie.0 root bus, but to a pcie root port or switch.

So they have to be connected to a *-port even if you don't care about 
hotplug? The entire deal with the ports on the root complex are 
confusing to me - Alex had said the other day that it *does* happen that 
non-Express devices end up connected directly to the root complex in 
real hardware (although probably this violates the spec), so it may be 
reasonable to expect guest OSes to deal with that (that, and dgilbert's 
success doing it, are what gave me the courage to suggest plugging 
pci-bridge directly into a root complex port).


>
>>
>> Still, it would be nice to allow *those who really want to* to have a 
>> "pure PCIe" controller setup. We could do that if we did two things:
>
> Sure, pure PCIe setup would be nice.
>
>>
>> 1) modify the PCI address auto-assignment code to not insist on 
>> always having at least one hot-pluggable standard PCI slot available.
>>
>> and one of the following:
>>
>> 2a) don't always add a dmi-to-pci-bridge, but instead only do so *if 
>> necessary* in order to plug in a pci-bridge (which would only be 
>> added if necessary).
>>
>>   or
>>
>> 2b) permit auto-assigning a pci-bridge to plug into a port of 
>> pcie-root, thus eliminating the need for dmi-to-pci-bridge completely.
>>
>> Does anyone have opinions about (1) or (2b)?
>
> I personally don't like 2b because would not work in 'real' world, so 
> 1 and 2a seems fine to me.
>
> Bottom line, maybe we can tackle the dmi-to-pci bridge to cover your 
> requirements.
>

I guess you mean "make the dmi-to-pci-bridge slots hotpluggable" by 
this, right?




More information about the libvir-list mailing list