Re: [PATCH v3 01/11] PCI/P2PDMA: Support peer-to-peer memory

[Jason Gunthorpe <jgg-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>]

On Mon, Mar 26, 2018 at 11:30:38AM -0600, Logan Gunthorpe wrote:
> 
> 
> On 26/03/18 10:41 AM, Jason Gunthorpe wrote:
> > On Mon, Mar 26, 2018 at 12:11:38PM +0100, Jonathan Cameron wrote:
> >> On Tue, 13 Mar 2018 10:43:55 -0600
> >> Logan Gunthorpe <logang-OTvnGxWRz7hWk0Htik3J/w@public.gmane.org> wrote:
> >>
> >>> On 12/03/18 09:28 PM, Sinan Kaya wrote:
> >>>> On 3/12/2018 3:35 PM, Logan Gunthorpe wrote:
> >>>> Regarding the switch business, It is amazing how much trouble you went into
> >>>> limit this functionality into very specific hardware.
> >>>>
> >>>> I thought that we reached to an agreement that code would not impose
> >>>> any limits on what user wants.
> >>>>
> >>>> What happened to all the emails we exchanged?  
> >>>
> >>> It turns out that root ports that support P2P are far less common than 
> >>> anyone thought. So it will likely have to be a white list.
> >>
> >> This came as a bit of a surprise to our PCIe architect.
> > 
> > I don't think it is a hardware problem.
> 
> The latest and greatest Power9 CPUs still explicitly do not support
> this.

I think this is another case of the HW can do it but the SW support is
missing. IOMMU configuration and maybe firmware too, for instance.

If I recall I saw a presentation that Coral was expected to use P2P
between the network and GPU.

> And, if I recall correctly, the ARM64 device we played with did
> not either -- but I suspect that will differ depending on vendor.

Wouldn't surprise me at all to see broken implementations in
ARM64.. But even there it needs IOMMU enablement to work at all if I
recall.

Bascially, this is probably not a HW problem that needs a HW bit, but
a OS/firmware problem to do all the enablement..

Jason