On Wed, 25 Aug 2021 at 03:36, John Hubbard [off-list ref] wrote:

On 8/24/21 10:32 AM, Jason Gunthorpe wrote:
...

quoted

quoted

quoted

And yes at least for the amdgpu driver we migrate the memory to host
memory as soon as it is pinned and I would expect that other GPU drivers
do something similar.

Well...for many topologies, migrating to host memory will result in a
dramatically slower p2p setup. For that reason, some GPU drivers may
want to allow pinning of video memory in some situations.

Ideally, you've got modern ODP devices and you don't even need to pin.
But if not, and you still hope to do high performance p2p between a GPU
and a non-ODP Infiniband device, then you would need to leave the pinned
memory in vidmem.

So I think we don't want to rule out that behavior, right? Or is the
thinking more like, "you're lucky that this old non-ODP setup works at
all, and we'll make it work by routing through host/cpu memory, but it
will be slow"?

I think it depends on the user, if the user creates memory which is
permanently located on the GPU then it should be pinnable in this way
without force migration. But if the memory is inherently migratable
then it just cannot be pinned in the GPU at all as we can't
indefinately block migration from happening eg if the CPU touches it
later or something.

OK. I just want to avoid creating any API-level assumptions that dma_buf_pin()
necessarily implies or requires migrating to host memory.

I'm not sure we should be allowing dma_buf_pin at all on
non-migratable memory, what's to stop someone just pinning all the
VRAM and making the GPU unuseable?

I understand not considering more than a single user in these
situations is enterprise thinking, but I do worry about pinning is
always fine type of thinking when things are shared or multi-user.

My impression from this is we've designed hardware that didn't
consider the problem, and now to let us use that hardware in horrible
ways we should just allow it to pin all the things.

Dave.