Skip to contentSkip to conversation

vfio: missing patch in linux 3.6

DISCUSSION Andreas Hartmann 3 participants 7 messages Oct 4, 2012 – Oct 14, 2012

Subsystems (2)
lore
Share & tools
lore
#1 lore raw
Quoted 8 lines · Andreas Hartmann

Hello Alex,

I just tested vfio as part of linux 3.6 and detected, that it doesn't work because of the following missing patch:

http://news.gmane.org/find-root.php?group=gmane.linux.kernel.pci&article=16422

Could you please get it applied for the next stable release?

Hi Andreas,

This patch needs to go through Bjorn's PCI tree, but as noted in the comments I have some outstanding questions that need to be answered, probably by Joerg. Thanks,

Alex

#2 lore raw
Quoted 10 lines · Alex Williamson
Quoted 9 lines · Andreas Hartmann

Hello Alex,

I just tested vfio as part of linux 3.6 and detected, that it doesn't work because of the following missing patch:

http://news.gmane.org/find-root.php?group=gmane.linux.kernel.pci&article=16422

Could you please get it applied for the next stable release?

Quoted 6 lines

Hi Andreas,

This patch needs to go through Bjorn's PCI tree, but as noted in the comments I have some outstanding questions that need to be answered, probably by Joerg. Thanks,

Sorry Alex,
I missed your additional questions because this patch just works pretty fine here with kernel 3.4.x.

But you're right, your questions should be resolved before freeing it to the wild.

@Joerg:
Please, liberate me from compiling and patching each kernel myself to get it working and give all the other users out there a chance to use this pretty code and the IOMMU support of AMD's hardware :-) by answering Alex' questions.

Thanks,
kind regards,
Andreas

#3 lore raw
In reply to Alex Williamson
Quoted 8 lines · Alex Williamson
Quoted 7 lines · Andreas Hartmann

Hello Alex,

I just tested vfio as part of linux 3.6 and detected, that it doesn't work because of the following missing patch:

http://news.gmane.org/find-root.php?group=gmane.linux.kernel.pci&article=16422

Quoted 7 lines
Quoted 1 line

Could you please get it applied for the next stable release?

Hi Andreas,

This patch needs to go through Bjorn's PCI tree, but as noted in the comments I have some outstanding questions that need to be answered, probably by Joerg. Thanks,

What are the open questions? I have confirmation from the hardware people that the south bridges are peer-2-peer safe. Other questions?

Joerg

Signature · 5 lines
AMD Operating System Research Center

Advanced Micro Devices GmbH Einsteinring 24 85609 Dornach
General Managers: Alberto Bozzo
Registration: Dornach, Landkr. Muenchen; Registerger. Muenchen, HRB Nr. 43632
#4 lore raw
Quoted 11 lines · Joerg
Quoted 10 lines · Alex Williamson
Quoted 8 lines · Andreas Hartmann

Hello Alex,

I just tested vfio as part of linux 3.6 and detected, that it doesn't work because of the following missing patch:

http://news.gmane.org/find-root.php?group=gmane.linux.kernel.pci&article=16422

Could you please get it applied for the next stable release?

Quoted 8 lines
Quoted 5 lines

Hi Andreas,

This patch needs to go through Bjorn's PCI tree, but as noted in the comments I have some outstanding questions that need to be answered, probably by Joerg. Thanks,

What are the open questions? I have confirmation from the hardware people that the south bridges are peer-2-peer safe. Other questions?

Hi Joerg,

There are a couple questions in the link above. Since the devices don't expose a PCIe capability, we probably need to add a check to look at the upstream device and verify we're not on a legacy bus where ACS can't be enforced. Then there's the general question of whether the confirmation of no peer-to-peer applies to every case where we might see this device (some of them seem to have history that pre-dates this specific package implementation) or do we need to try to identify specific package properties in addition to just a device ID? Thanks,

Alex

#5 lore raw
Quoted 4 lines · Alex Williamson

There are a couple questions in the link above. Since the devices don't expose a PCIe capability, we probably need to add a check to look at the upstream device and verify we're not on a legacy bus where ACS can't be enforced.

You can certainly do that, but for existing hardware (with an AMD IOMMU) this check would never be true. These devices are always on a seperate south-bridge chip which is connected to the north-bridge via PCIe. So the topology where you find these devices is always the same in an AMD IOMMU system.

Quoted 5 lines

Then there's the general question of whether the confirmation of no peer-to-peer applies to every case where we might see this device (some of them seem to have history that pre-dates this specific package implementation) or do we need to try to identify specific package properties in addition to just a device ID?

The device ids are kept the same to maintain compatability with older software. Besides that, my statement about the peer-2-peer safety is true for all south-bridges that you can find in an AMD IOMMU capable system.

Regards,

Joerg

Signature · 5 lines
AMD Operating System Research Center

Advanced Micro Devices GmbH Einsteinring 24 85609 Dornach
General Managers: Alberto Bozzo
Registration: Dornach, Landkr. Muenchen; Registerger. Muenchen, HRB Nr. 43632
#6 lore raw
Quoted 18 lines · Joerg
Quoted 4 lines · Alex Williamson

There are a couple questions in the link above. Since the devices don't expose a PCIe capability, we probably need to add a check to look at the upstream device and verify we're not on a legacy bus where ACS can't be enforced.

You can certainly do that, but for existing hardware (with an AMD IOMMU) this check would never be true. These devices are always on a seperate south-bridge chip which is connected to the north-bridge via PCIe. So the topology where you find these devices is always the same in an AMD IOMMU system.

Quoted 5 lines

Then there's the general question of whether the confirmation of no peer-to-peer applies to every case where we might see this device (some of them seem to have history that pre-dates this specific package implementation) or do we need to try to identify specific package properties in addition to just a device ID?

Quoted 4 lines

The device ids are kept the same to maintain compatability with older software. Besides that, my statement about the peer-2-peer safety is true for all south-bridges that you can find in an AMD IOMMU capable system.

http://news.gmane.org/find-root.php?group=gmane.linux.kernel.pci&article=16422

I think, Joerg wrote clearly that there is no problem any more to apply this patch. I run it since a long time meanwhile without any problem. Please, why aren't these patches applied?

Thanks,
kind regards,
Andreas Hartmann