| 1 |
On Thu, Mar 8, 2018 at 4:14 PM, Taiidan@×××.com <Taiidan@×××.com> wrote: |
| 2 |
> On 03/07/2018 09:02 PM, R0b0t1 wrote: |
| 3 |
> |
| 4 |
>> On Wed, Mar 7, 2018 at 7:52 PM, Taiidan@×××.com <Taiidan@×××.com> wrote: |
| 5 |
>>> |
| 6 |
>>> I bought a LSI-9211-8i / SAS 2008 controller which reports support for |
| 7 |
>>> SR-IOV in lspci and I am wondering how I can use it. |
| 8 |
>>> |
| 9 |
>>> There is no info on the internet about this not even for their newer |
| 10 |
>>> controllers where there is a lot of advertising about SR-IOV. |
| 11 |
>>> |
| 12 |
>>> The idea is that you can assign a RAID array, individual hard drive, etc |
| 13 |
>>> to |
| 14 |
>>> a VF which is then assigned to a VM via IOMMU providing better almost |
| 15 |
>>> native |
| 16 |
>>> performance vs emulated disks. |
| 17 |
>>> |
| 18 |
>>> Thanks! |
| 19 |
>> |
| 20 |
>> If it supports SR-IOV you can pass it to a guest with VFIO. If it did |
| 21 |
>> not support SR-IOV it would not support VFIO. |
| 22 |
> |
| 23 |
> I know - my question is how do I create the virtual functions and assign the |
| 24 |
> drives to them instead of simply attaching the entire controller? |
| 25 |
> |
| 26 |
> According to LSI's press release you could have for instance 5 different |
| 27 |
> RAID's assigned to 5 different VM's via virtual functions - not simply all |
| 28 |
> of them assign to one VM via assigning the controller like a non SR-IOV |
| 29 |
> device |
| 30 |
> |
| 31 |
|
| 32 |
https://wiki.archlinux.org/index.php/PCI_passthrough_via_OVMF |
| 33 |
https://wiki.installgentoo.com/index.php/PCI_passthrough |
| 34 |
https://www.kernel.org/doc/Documentation/vfio.txt |
| 35 |
|
| 36 |
The one sticking point is that you need to figure out the layout of |
| 37 |
your PCIe lanes to share multiple devices without conflicts. |
| 38 |
|
| 39 |
Cheers, |
| 40 |
R0b0t1 |
Archives