Re: [dpdk-dev] Query : Does Binding with vfio-pci is supported inside qemu-kvm guest/vm instance.?

[Maxime Coquelin <maxime.coquelin@redhat.com>]

Hi John,

On 4/3/19 9:49 AM, John Sucaet wrote:
> Thanks, Anatoly.
> 
> Maxime, could you enlighten me a bit? I basically would like to know 
> whether I should be able to make my 32-bit dpdk application work with 
> virtio-pci-net and vfio-pci (with a 64-bit kernel), or if I should do 
> the effort to port to 64-bit (which I would like to avoid for now).

I think that it should work, but that's not something I have tried.
I will try to reproduce it this week to get a precise idea of what is
going wrong.

Thanks for reporting the issue,
Maxime

> Thank you
> John
> 
> On 04/02/2019 03:38 PM, Burakov, Anatoly wrote:
>> On 02-Apr-19 11:38 AM, John Sucaet wrote:
>>> Hi Anatoly,
>>>
>>> As you said: There's no reason to use igb_uio, ever!
>>
>> That was partly tongue in cheek, but point taken :)
>>
>>> I would like to ask whether vfio-pci with or without vIOMMU 
>>> should/could work for virtio-pci net devices in the case of a 32-bit 
>>> dpdk application, on a 64-bit kernel (4.9) inside a guest VM 
>>> (qemu-2.10.2-1.fc27)?
>>>
>>> I tried both a 64-bit and a 32-bit version of the same application, 
>>> but only in the case of the 64-bit application, the port was found by 
>>> eal. The 32-bit application gave errors like:
>>>
>>> EAL: pci_map_resource(): cannot mmap(16, 0xf4a01000, 0x4000, 0x0): 
>>> Invalid argument (0xffffffff)
>>> EAL: Failed to map pci BAR4
>>> EAL:   0000:00:02.0 mapping BAR4 failed: Invalid argument
>>> EAL: Can't write to PCI bar (0) : offset (12)
>>> EAL: Can't read from PCI bar (0) : offset (12)
>>> EAL: Can't read from PCI bar (0) : offset (12)
>>> EAL: Can't write to PCI bar (0) : offset (12)
>>> EAL: Can't read from PCI bar (0) : offset (12)
>>> EAL: Can't write to PCI bar (0) : offset (12)
>>> EAL: Can't read from PCI bar (0) : offset (0)
>>> EAL: Can't write to PCI bar (0) : offset (4)
>>> EAL: Can't write to PCI bar (0) : offset (14)
>>> EAL: Can't read from PCI bar (0) : offset (14)
>>> EAL: Can't read from PCI bar (0) : offset (1a)
>>> EAL: Can't read from PCI bar (0) : offset (1c)
>>> EAL: Can't write to PCI bar (0) : offset (e)
>>> EAL: Can't read from PCI bar (0) : offset (c)
>>> virtio_init_queue(): virtqueue size is not powerof 2
>>> EAL: Requested device 0000:00:02.0 cannot be used
>>>
>>> Maybe you have an idea what went wrong here?
>>>
>>> By preference, I would like to continue to use the 32-bit application 
>>> which worked fine with the igb_uio driver.
>>
>> Unfortunately, i am not very familiar with virtio and wouldn't know 
>> whether it's supposed to work under these conditions. Perhaps Maxime 
>> would be of more help here (CC'd).
>>
>>>
>>>
>>> Thanks
>>>
>>> John
>>>
>>>
>>> On 03/12/2019 11:57 AM, Burakov, Anatoly wrote:
>>>> On 12-Mar-19 10:20 AM, Bruce Richardson wrote:
>>>>> On Tue, Mar 12, 2019 at 05:54:39PM +0800, Jason Wang wrote:
>>>>>>
>>>>>> On 2019/3/12 下午5:42, Thanneeru Srinivasulu wrote:
>>>>>>> Thanks Bruce..
>>>>>>>
>>>>>>> On Tue, Mar 12, 2019 at 3:08 PM Bruce Richardson
>>>>>>> <bruce.richardson@intel.com> wrote:
>>>>>>>> On Tue, Mar 12, 2019 at 10:57:55AM +0530, Thanneeru Srinivasulu 
>>>>>>>> wrote:
>>>>>>>>> Hi Everyone.
>>>>>>>>>
>>>>>>>>> I did attached pice  to Guest VM using vfio-pci with qemu 
>>>>>>>>> command, and then
>>>>>>>>> tried binding the pcie bdf with vfio-pci, observing binding 
>>>>>>>>> failure with
>>>>>>>>> vfio-pci.
>>>>>>>>>
>>>>>>>>> Where as when tryied with igb_uio, everything works fine.
>>>>>>>>>
>>>>>>>>> Does Binding with vfio-pci is supported inside VM/guest?
>>>>>>>>>
>>>>>>>> vfio support requires the presence of an IOMMU, and you 
>>>>>>>> generally don't
>>>>>>>> have an IOMMU available in a VM.
>>>>>>>>
>>>>>>>> /Bruce
>>>>>>
>>>>>>
>>>>>> Actually, Qemu support vIOMMU + VFIO in guest[1], all you need is 
>>>>>> to add a
>>>>>> intel IOMMU and enabling caching mode.
>>>>>>
>>>>>> Thanks
>>>>>>
>>>>>>
>>>>>> [1]
>>>>>>
>>>>>> https://www.lfasiallc.com/wp-content/uploads/2017/11/Device-Assignment-with-Nested-Guests-and-DPDK_Peter-Xu.pdf 
>>>>>>
>>>>>>
>>>>> Thanks for the info.
>>>>>
>>>>> /Bruce
>>>>>
>>>>
>>>> One more thing: even without vIOMMU, VFIO has no-IOMMU mode which 
>>>> can be enabled (for a recent-enough kernel). This will make VFIO 
>>>> work even in cases where the guest doesn't have IOMMU emulation. 
>>>> See? There's no reason to use igb_uio, ever! :D
>>>>
>>>
>>>
>>
>>
>