[libvirt] [RFC PATCH] support multifunction PCI device

Wed May 11 08:05:42 UTC 2011

At 05/11/2011 03:55 PM, Daniel P. Berrange Write:
> On Wed, May 11, 2011 at 10:30:43AM +0800, Wen Congyang wrote:
>> At 05/10/2011 06:00 PM, Daniel P. Berrange Write:
>>> On Tue, May 10, 2011 at 02:00:17PM +0800, Wen Congyang wrote:
>>>> We want to use more than 200+ device. Libvirt does not use multi
>>>> function PCI device and PCI-to-PCI bridge. So we can not use more
>>>> than 200+ device if it's a PCI device or it's controller is a PCI
>>>> device.
>>>
>>> IMHO using multi function support is a dead end, because it
>>> makes hotplug completely unusable and is not even possible
>>> for most PCI devices. The way we want to raise the device
>>> limit is
>>>
>>>   a. Supporting multiple PCI domains (multiple PCI root complexes)
>>
>> IIUC, qemu do not support multiple PCI domains now.
> 
> There are people who are working on making that supported in KVM
> real soon.

Good news.

> 
>>>   b. Adding PCI bridges
>>
>> I do not know whether qemu supports it. I find hw/pci_bridge.c in qemu's source,
>> so qemu may support it. I will confirm it.
> 
> I'm told this is already supported, but I've not tested it myself.

I'm still reading the qemu's code. Do you know the driver's name(For example:
the driver for scsi controller in qemu is lsi.)

> 
>>>   c. A virtio-scsi controller to allow > 1 virtio disk per PCI device
>>
>> Hmm, do you mean this:
>> -device virtio-blk-pci,ports=4,drive0=hda,drive1=hdb,drive2=hdc,...
> 
> No it is actually going to be a new device type, and I'm expecting
> it will work very like the LSI scsi does in terms of device setup.
> eg create a controller device, and then create devices for each drive.
> 
>   -device virtio-scsi-pci,id=vscsi1
>   -drive file=/some/disk,id=vscsi1.1
>   -device virtio-blk-pci,bus=vscsi1,drive=vscsi1.1
>   -drive file=/some/disk,id=vscsi1.2
>   -device virtio-blk-pci,bus=vscsi1,drive=vscsi1.2
>   -drive file=/some/disk,id=vscsi1.2
>   -device virtio-blk-pci,bus=vscsi1,drive=vscsi1.2
> 
>>> any of those will dramatically increase the number of devices
>>> we can use without the horrible hotplug problems that multifunction
>>> introduces.
>>>
>>>> This patch adds the support of multi function PCI device. It
>>>> does not support to hot plug/unplug multi function PCI device.
>>>>
>>>> TODO:
>>>> 1. support to hot plug multi function PCI device
>>>>    We only can hot plug one device at one time. I think we should
>>>>    introduce another command 'virsh attach-devices XXX' to support
>>>>    hot plug more than one device at one time?
>>>
>>> Since you can't do practical hotplug of multifunction devices at the
>>> QEMU layer at all, there's nothing useful we can do at libvirt either.
>>
>> We can hotplug multifunction PCI devices like this:
>> 1. Before hot pluging,
>> # lspci
>> 00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02)
>> 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II]
>> 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II]
>> 00:01.2 USB Controller: Intel Corporation 82371SB PIIX3 USB [Natoma/Triton II] (rev 01)
>> 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 03)
>> 00:02.0 VGA compatible controller: Cirrus Logic GD 5446
>> 00:03.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8139/8139C/8139C+ (rev 20)
>> 00:04.0 RAM memory: Red Hat, Inc Virtio memory balloon
>> 00:05.0 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>>
>> 2. hot plug multi function PCI devices:
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi2,bus=pci.0,addr=0x06.0x07'
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi3,bus=pci.0,addr=0x06.0x06'
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi4,bus=pci.0,addr=0x06.0x05'
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi5,bus=pci.0,addr=0x06.0x04'
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi6,bus=pci.0,addr=0x06.0x03'
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi7,bus=pci.0,addr=0x06.0x02'
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi8,bus=pci.0,addr=0x06.0x01'
>> # virsh qemu-monitor-command vm1 --hmp 'device_add lsi,id=scsi9,multifunction=on,bus=pci.0,addr=0x06.0x00'
> 
> Hmm, that's kinda wierd & i'm suprised it works, particularly for LSI
> since I thought guest drivers would need support for multifunction too.

I only test LSI. If we can hotplug the device by single PCI device, I think it can work
fine for multi function.

> 
>>
>> 3. After hot plug multi function PCI devices:
>> # lspci
>> 00:00.0 Host bridge: Intel Corporation 440FX - 82441FX PMC [Natoma] (rev 02)
>> 00:01.0 ISA bridge: Intel Corporation 82371SB PIIX3 ISA [Natoma/Triton II]
>> 00:01.1 IDE interface: Intel Corporation 82371SB PIIX3 IDE [Natoma/Triton II]
>> 00:01.2 USB Controller: Intel Corporation 82371SB PIIX3 USB [Natoma/Triton II] (rev 01)
>> 00:01.3 Bridge: Intel Corporation 82371AB/EB/MB PIIX4 ACPI (rev 03)
>> 00:02.0 VGA compatible controller: Cirrus Logic GD 5446
>> 00:03.0 Ethernet controller: Realtek Semiconductor Co., Ltd. RTL-8139/8139C/8139C+ (rev 20)
>> 00:04.0 RAM memory: Red Hat, Inc Virtio memory balloon
>> 00:05.0 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.0 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.1 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.2 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.3 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.4 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.5 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.6 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>> 00:06.7 SCSI storage controller: LSI Logic / Symbios Logic 53c895a
>>
>>
>>>
>>>>
>>>> 2. support to hot unplug multi function PCI device
>>>>    hot unpluging multi function PCI device meas that all the other
>>>>    devices on the same slot will be hot unpluged. So we should do
>>>>    some cleanup and remove the other devices too. If the other
>>>>    device on the same slot does not support hot unpluged, or it is a
>>>>    a controller and some other devices still use this controller,
>>>>    I think we should refuse to hot unplug this mutlti function PCI
>>>>    device.
>>>
>>> IMHO these kind of restrictions will make life really unpleasant
>>> for applications and are a reason we should *not* support the
>>> multifunction code. Instead we should focus on one of the other
>>> 3 options I mention above.
>>
>> Yes, there's too many restrictions about hot plug/unplug multifunction
>> PCI devices.
> 
> 
> Regards,
> Daniel