Thread (67 messages) 67 messages, 10 authors, 2018-04-13

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

From: Bjorn Helgaas <helgaas@kernel.org>
Date: 2018-03-14 02:56:49
Also in: linux-nvme, linux-pci, linux-rdma, lkml, nvdimm

On Tue, Mar 13, 2018 at 05:21:20PM -0600, Logan Gunthorpe wrote:
On 13/03/18 05:08 PM, Bjorn Helgaas wrote:
quoted
On Tue, Mar 13, 2018 at 10:31:55PM +0000, Stephen  Bates wrote:
If it *is* necessary because Root Ports and devices below them behave
differently than in conventional PCI, I think you should include a
reference to the relevant section of the spec and check directly for a
Root Port.  I would prefer that over trying to exclude Root Ports by
looking for two upstream bridges.
Well we've established that we don't want to allow root ports.
I assume you want to exclude Root Ports because of multi-function
devices and the "route to self" error.  I was hoping for a reference
to that so I could learn more about it.

While I was looking for it, I found sec 6.12.1.2 (PCIe r4.0), "ACS
Functions in SR-IOV Capable and Multi-Function Devices", which seems
relevant.  It talks about "peer-to-peer Requests (between Functions of
the device)".  Thay says to me that multi-function devices can DMA
between themselves.
But we need to, at a minimum, do two pci_upstream_bridge() calls...

Remember, we need to check that a number of devices are behind the same
switch. So we need to find a _common_ upstream port for several devices.
I agree that peers need to have a common upstream bridge.  I think
you're saying peers need to have *two* common upstream bridges.  If I
understand correctly, requiring two common bridges is a way to ensure
that peers directly below Root Ports don't try to DMA to each other.

So I guess the first order of business is to nail down whether peers
below a Root Port are prohibited from DMAing to each other.  My
assumption, based on 6.12.1.2 and the fact that I haven't yet found
a prohibition, is that they can.
Excluding the multifunction device case (which I don't think is
applicable for reasons we've discussed before), this will *never* be the
first upstream port for a given device.
If you're looking for a common upstream bridge, you don't need to make
assumptions about whether the hierarchy is conventional PCI or PCIe or
how many levels are in the hierarchy.

You already have upstream_bridges_match(), which takes two pci_devs.
I think it should walk up the PCI hierarchy from the first device,
checking whether the bridge at each level is also a parent of the
second device.

Bjorn
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help