RE: IOMMU/DMA map_resource support for peer-to-peer

RE: IOMMU/DMA map_resource support for peer-to-peer

[Goz, Ben]

Hi Will,
I'm interesting on your work on this patch set to support map_resource for peer to peer memory mapping, if you can please refer me to git repository includes your patches.
And I want to know please if the IOMMU should be configured with pass through or any other configuration, and if there is a general limitation or something I should know for bringing it up.

Regards,
Ben

RE: IOMMU/DMA map_resource support for peer-to-peer

[William Davis]

Hi Ben,

> I'm interesting on your work on this patch set to support map_resource for
> peer to peer memory mapping, if you can please refer me to git repository
> includes your patches.

The patches are not hosted in any externally-accessible repo. The last patch set was built on top of the linux-pci tree at commit bc0195aad0daa2ad5b0d76cce22b167bc3435590 (linux 4.2-rc2).
 
> And I want to know please if the IOMMU should be configured with pass
> through or any other configuration, and if there is a general limitation or
> something I should know for bringing it up.

My testing was mostly Intel-based, but it did not require passthrough mode; the goal of the patches was to get PCI P2P working on configurations where the IOMMU is configured for full translation. It should only be limited to what the chipset is capable of, in terms of PCI P2P topologies.

Thanks,
Will

--
nvpublic

RE: IOMMU/DMA map_resource support for peer-to-peer

[Goz, Ben]

Hi Will I can't find your commits on linux-pci git repository and the only tag that I can think that you meant to say is pci-v4.2-fixes-2 and the patches not found there.
Can you please refer me to the exact place I can find the patch set?

Regards,
Ben

-----Original Message-----
From: William Davis [mailto:wdavis@nvidia.com] 
Sent: Monday, December 21, 2015 6:46 PM
To: Goz, Ben
Cc: linux-pci@vger.kernel.org; Koenig, Christian; Bridgman, John; Terence Ripperda; John Hubbard
Subject: RE: IOMMU/DMA map_resource support for peer-to-peer

Hi Ben,

> I'm interesting on your work on this patch set to support map_resource 
> for peer to peer memory mapping, if you can please refer me to git 
> repository includes your patches.

The patches are not hosted in any externally-accessible repo. The last patch set was built on top of the linux-pci tree at commit bc0195aad0daa2ad5b0d76cce22b167bc3435590 (linux 4.2-rc2).
 
> And I want to know please if the IOMMU should be configured with pass 
> through or any other configuration, and if there is a general 
> limitation or something I should know for bringing it up.

My testing was mostly Intel-based, but it did not require passthrough mode; the goal of the patches was to get PCI P2P working on configurations where the IOMMU is configured for full translation. It should only be limited to what the chipset is capable of, in terms of PCI P2P topologies.

Thanks,
Will

--
nvpublic

RE: IOMMU/DMA map_resource support for peer-to-peer

[William Davis]

> Hi Will I can't find your commits on linux-pci git repository and the only
> tag that I can think that you meant to say is pci-v4.2-fixes-2 and the
> patches not found there.
> Can you please refer me to the exact place I can find the patch set?
> 

As I said below, the patches aren't hosted in any externally-accessible repo - that is, the only git repository to which the patches have been committed is a local development repo that is a clone of linux-pci.

Each of the patch set versions are only available via the mailing list archives, e.g., (this is the latest version)

http://www.spinics.net/lists/linux-pci/msg44560.html

You'll have to apply the patches manually.

> 
> Hi Ben,
> 
> > I'm interesting on your work on this patch set to support map_resource
> > for peer to peer memory mapping, if you can please refer me to git
> > repository includes your patches.
> 
> The patches are not hosted in any externally-accessible repo. The last
> patch set was built on top of the linux-pci tree at commit
> bc0195aad0daa2ad5b0d76cce22b167bc3435590 (linux 4.2-rc2).
> 
> > And I want to know please if the IOMMU should be configured with pass
> > through or any other configuration, and if there is a general
> > limitation or something I should know for bringing it up.
> 
> My testing was mostly Intel-based, but it did not require passthrough mode;
> the goal of the patches was to get PCI P2P working on configurations where
> the IOMMU is configured for full translation. It should only be limited to
> what the chipset is capable of, in terms of PCI P2P topologies.
> 
> Thanks,
> Will
> 
--
nvpublic