On Fri, Jan 06, 2017 at 11:56:30AM -0500, Serguei Sagalovitch wrote:
On 2017-01-05 08:58 PM, Jerome Glisse wrote:
> On Thu, Jan 05, 2017 at 05:30:34PM -0700, Jason Gunthorpe wrote:
> > On Thu, Jan 05, 2017 at 06:23:52PM -0500, Jerome Glisse wrote:
> > > > I still don't understand what you driving at - you've said in
> > > > cases a user VMA exists.
> > > In the former case no, there is no VMA directly but if you want one than
> > > a device can provide one. But such VMA is useless as CPU access is not
> > > expected.
> > I disagree it is useless, the VMA is going to be necessary to support
> > upcoming things like CAPI, you need it to support O_DIRECT from the
> > filesystem, DPDK, etc. This is why I am opposed to any model that is
> > not VMA based for setting up RDMA - that is shorted sighted and does
> > not seem to reflect where the industry is going.
> > So focus on having VMA backed by actual physical memory that covers
> > your GPU objects and ask how do we wire up the '__user *' to the DMA
> > API in the best way so the DMA API still has enough information to
> > setup IOMMUs and whatnot.
> I am talking about 2 different thing. Existing hardware and API where you
> _do not_ have a vma and you do not need one. This is just existing stuff.
I do not understand why you assume that existing API doesn't need one.
I would say that a lot of __existing__ user level API and their support in
kernel (especially outside of graphics domain) assumes that we have vma and
deal with __user * pointers.
Well i am thinking to GPUDirect here. Some of GPUDirect use case do not have
vma (struct vm_area_struct) associated with them they directly apply to GPU
object that aren't expose to CPU. Yes some use case have vma for share buffer.
In the open source driver it is true that we have vma most often than not.
> Some close driver provide a functionality on top of this design.
> is do we want to do the same ? If yes and you insist on having a vma we
> could provide one but this is does not apply and is useless for where we
> are going with new hardware.
> With new hardware you just use malloc or mmap to allocate memory and then
> you use it directly with the device. Device driver can migrate any part of
> the process address space to device memory. In this scheme you have your
> usual VMAs but there is nothing special about them.
Assuming that the whole device memory is CPU accessible and it looks
like the direction where we are going:
- You forgot about use case when we want or need to allocate memory
directly on device (why we need to migrate anything if not needed?).
- We may want to use CPU to access such memory on device to avoid
any unnecessary migration back.
- We may have more device memory than the system one.
E.g. if you have 12 GPUs w/64GB each it will already give us ~0.7 TB
not mentioning NVDIMM cards which could also be used as memory
storage for other device access.
- We also may want/need to share GPU memory between different
Here i am talking about platform where GPU memory is not accessible at
all by the CPU (because of PCIe restriction, think CPU atomic operation
on IO memory).
So i really distinguish between CAPI/CCIX and PCIe. Some platform will
have CAPI/CCIX other wont. HMM apply mostly to the latter. Some of HMM
functionalities are still usefull with CAPI/CCIX.
Note that HMM do support allocation on GPU first. In current design this
can happen when GPU is the first to access an unpopulated virtual address.
For platform where GPU memory is accessible plan is either something
like CDM (Coherent Device Memory) or rely on ZONE_DEVICE. So all GPU
memory have struct page and those are like ordinary pages. CDM still
wants some restrictions like avoiding CPU allocation to happen on GPU
when there is memory pressure ... For all intent and purposes this
will work transparently in respect to RDMA because we assume on those
system that the RDMA is CAPI/CCIX and that it can peer to other device.
> Now when you try to do get_user_page() on any page that is
> device it will fails because we do not allow any device memory to be pin.
> There is various reasons for that and they are not going away in any hw
> in the planing (so for next few years).
> Still we do want to support peer to peer mapping. Plan is to only do so
> with ODP capable hardware. Still we need to solve the IOMMU issue and
> it needs special handling inside the RDMA device. The way it works is
> that RDMA ask for a GPU page, GPU check if it has place inside its PCI
> bar to map this page for the device, this can fail. If it succeed then
> you need the IOMMU to let the RDMA device access the GPU PCI bar.
> So here we have 2 orthogonal problem. First one is how to make 2 drivers
> talks to each other to setup mapping to allow peer to peer But I would assume and
> about IOMMU.
I think that there is the third problem: A lot of existing user level API
(MPI, IB Verbs, file i/o, etc.) deal with pointers to the buffers.
Potentially it would be ideally to support use cases when those buffers are
located in device memory avoiding any unnecessary migration /
Currently a lot of infrastructure in kernel assumes that this is the user
pointer and call "get_user_pages" to get s/g. What is your opinion
how it should be changed to deal with cases when "buffer" is in
For HMM plan is to restrict to ODP and either to replace ODP with HMM or change
ODP to not use get_user_pages_remote() but directly fetch informations from
CPU page table. Everything else stay as it is. I posted patchset to replace
ODP with HMM in the past.
For the older kind of API (GPUDirect on yesterday hardware) it uses special
userspace API. If we don't care about supporting those i don't mind much but
some people see benefit in not having to deal with vma (struct vm_area_struct).