From: "Zeng, Oak" <oak.zeng@intel.com>
To: "Christian König" <christian.koenig@amd.com>,
"Thomas Hellström" <thomas.hellstrom@linux.intel.com>,
"Daniel Vetter" <daniel@ffwll.ch>
Cc: "Brost, Matthew" <matthew.brost@intel.com>,
Dave Airlie <airlied@redhat.com>,
Felix Kuehling <felix.kuehling@amd.com>,
"Welty, Brian" <brian.welty@intel.com>,
"dri-devel@lists.freedesktop.org"
<dri-devel@lists.freedesktop.org>,
"Ghimiray, Himal Prasad" <himal.prasad.ghimiray@intel.com>,
"Bommu, Krishnaiah" <krishnaiah.bommu@intel.com>,
"Gupta, saurabhg" <saurabhg.gupta@intel.com>,
"Vishwanathapura,
Niranjana" <niranjana.vishwanathapura@intel.com>,
"intel-xe@lists.freedesktop.org" <intel-xe@lists.freedesktop.org>,
Danilo Krummrich <dakr@redhat.com>
Subject: RE: Making drm_gpuvm work across gpu devices
Date: Sat, 27 Jan 2024 02:21:59 +0000 [thread overview]
Message-ID: <SA1PR11MB69919C19A16C007A6B01FC9E92782@SA1PR11MB6991.namprd11.prod.outlook.com> (raw)
In-Reply-To: <851e3ddb-6c74-4ec5-8d05-df12b34eb2b4@amd.com>
[-- Attachment #1: Type: text/plain, Size: 8130 bytes --]
Regarding the idea of expanding userptr to support migration, we explored this idea long time ago. It provides similar functions of the system allocator but its interface is not as convenient as system allocator. Besides the shared virtual address space, another benefit of a system allocator is, you can offload cpu program to gpu easier, you don’t need to call driver specific API (such as register_userptr and vm_bind in this case) for memory allocation.
We also scoped the implementation. It turned out to be big, and not as beautiful as hmm. Why we gave up this approach.
From: Christian König <christian.koenig@amd.com>
Sent: Friday, January 26, 2024 7:52 AM
To: Thomas Hellström <thomas.hellstrom@linux.intel.com>; Daniel Vetter <daniel@ffwll.ch>
Cc: Brost, Matthew <matthew.brost@intel.com>; Felix Kuehling <felix.kuehling@amd.com>; Welty, Brian <brian.welty@intel.com>; Ghimiray, Himal Prasad <himal.prasad.ghimiray@intel.com>; Zeng, Oak <oak.zeng@intel.com>; Gupta, saurabhg <saurabhg.gupta@intel.com>; Danilo Krummrich <dakr@redhat.com>; dri-devel@lists.freedesktop.org; Bommu, Krishnaiah <krishnaiah.bommu@intel.com>; Dave Airlie <airlied@redhat.com>; Vishwanathapura, Niranjana <niranjana.vishwanathapura@intel.com>; intel-xe@lists.freedesktop.org
Subject: Re: Making drm_gpuvm work across gpu devices
Am 26.01.24 um 09:21 schrieb Thomas Hellström:
Hi, all
On Thu, 2024-01-25 at 19:32 +0100, Daniel Vetter wrote:
On Wed, Jan 24, 2024 at 09:33:12AM +0100, Christian König wrote:
Am 23.01.24 um 20:37 schrieb Zeng, Oak:
[SNIP]
Yes most API are per device based.
One exception I know is actually the kfd SVM API. If you look at
the svm_ioctl function, it is per-process based. Each kfd_process
represent a process across N gpu devices.
Yeah and that was a big mistake in my opinion. We should really not
do that
ever again.
Need to say, kfd SVM represent a shared virtual address space
across CPU and all GPU devices on the system. This is by the
definition of SVM (shared virtual memory). This is very different
from our legacy gpu *device* driver which works for only one
device (i.e., if you want one device to access another device's
memory, you will have to use dma-buf export/import etc).
Exactly that thinking is what we have currently found as blocker
for a
virtualization projects. Having SVM as device independent feature
which
somehow ties to the process address space turned out to be an
extremely bad
idea.
The background is that this only works for some use cases but not
all of
them.
What's working much better is to just have a mirror functionality
which says
that a range A..B of the process address space is mapped into a
range C..D
of the GPU address space.
Those ranges can then be used to implement the SVM feature required
for
higher level APIs and not something you need at the UAPI or even
inside the
low level kernel memory management.
When you talk about migrating memory to a device you also do this
on a per
device basis and *not* tied to the process address space. If you
then get
crappy performance because userspace gave contradicting information
where to
migrate memory then that's a bug in userspace and not something the
kernel
should try to prevent somehow.
[SNIP]
I think if you start using the same drm_gpuvm for multiple
devices you
will sooner or later start to run into the same mess we have
seen with
KFD, where we moved more and more functionality from the KFD to
the DRM
render node because we found that a lot of the stuff simply
doesn't work
correctly with a single object to maintain the state.
As I understand it, KFD is designed to work across devices. A
single pseudo /dev/kfd device represent all hardware gpu devices.
That is why during kfd open, many pdd (process device data) is
created, each for one hardware device for this process.
Yes, I'm perfectly aware of that. And I can only repeat myself that
I see
this design as a rather extreme failure. And I think it's one of
the reasons
why NVidia is so dominant with Cuda.
This whole approach KFD takes was designed with the idea of
extending the
CPU process into the GPUs, but this idea only works for a few use
cases and
is not something we should apply to drivers in general.
A very good example are virtualization use cases where you end up
with CPU
address != GPU address because the VAs are actually coming from the
guest VM
and not the host process.
SVM is a high level concept of OpenCL, Cuda, ROCm etc.. This should
not have
any influence on the design of the kernel UAPI.
If you want to do something similar as KFD for Xe I think you need
to get
explicit permission to do this from Dave and Daniel and maybe even
Linus.
I think the one and only one exception where an SVM uapi like in kfd
makes
sense, is if the _hardware_ itself, not the software stack defined
semantics that you've happened to build on top of that hw, enforces a
1:1
mapping with the cpu process address space.
Which means your hardware is using PASID, IOMMU based translation,
PCI-ATS
(address translation services) or whatever your hw calls it and has
_no_
device-side pagetables on top. Which from what I've seen all devices
with
device-memory have, simply because they need some place to store
whether
that memory is currently in device memory or should be translated
using
PASID. Currently there's no gpu that works with PASID only, but there
are
some on-cpu-die accelerator things that do work like that.
Maybe in the future there will be some accelerators that are fully
cpu
cache coherent (including atomics) with something like CXL, and the
on-device memory is managed as normal system memory with struct page
as
ZONE_DEVICE and accelerator va -> physical address translation is
only
done with PASID ... but for now I haven't seen that, definitely not
in
upstream drivers.
And the moment you have some per-device pagetables or per-device
memory
management of some sort (like using gpuva mgr) then I'm 100% agreeing
with
Christian that the kfd SVM model is too strict and not a great idea.
Cheers, Sima
I'm trying to digest all the comments here, The end goal is to be able
to support something similar to this here:
https://developer.nvidia.com/blog/simplifying-gpu-application-development-with-heterogeneous-memory-management/
Christian, If I understand you correctly, you're strongly suggesting
not to try to manage a common virtual address space across different
devices in the kernel, but merely providing building blocks to do so,
like for example a generalized userptr with migration support using
HMM; That way each "mirror" of the CPU mm would be per device and
inserted into the gpu_vm just like any other gpu_vma, and user-space
would dictate the A..B -> C..D mapping by choosing the GPU_VA for the
vma.
Exactly that, yes.
Sima, it sounds like you're suggesting to shy away from hmm and not
even attempt to support this except if it can be done using IOMMU sva
on selected hardware?
I think that comment goes more into the direction of: If you have ATS/ATC/PRI capable hardware which exposes the functionality to make memory reads and writes directly into the address space of the CPU then yes an SVM only interface is ok because the hardware can't do anything else. But as long as you have something like GPUVM then please don't restrict yourself.
Which I totally agree on as well. The ATS/ATC/PRI combination doesn't allow using separate page tables device and CPU and so also not separate VAs.
This was one of the reasons why we stopped using this approach for AMD GPUs.
Regards,
Christian.
Could you clarify a bit?
Thanks,
Thomas
[-- Attachment #2: Type: text/html, Size: 15387 bytes --]
next prev parent reply other threads:[~2024-01-27 2:22 UTC|newest]
Thread overview: 123+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-01-17 22:12 [PATCH 00/23] XeKmd basic SVM support Oak Zeng
2024-01-17 22:12 ` [PATCH 01/23] drm/xe/svm: Add SVM document Oak Zeng
2024-01-17 22:12 ` [PATCH 02/23] drm/xe/svm: Add svm key data structures Oak Zeng
2024-01-17 22:12 ` [PATCH 03/23] drm/xe/svm: create xe svm during vm creation Oak Zeng
2024-01-17 22:12 ` [PATCH 04/23] drm/xe/svm: Trace svm creation Oak Zeng
2024-01-17 22:12 ` [PATCH 05/23] drm/xe/svm: add helper to retrieve svm range from address Oak Zeng
2024-01-17 22:12 ` [PATCH 06/23] drm/xe/svm: Introduce a helper to build sg table from hmm range Oak Zeng
2024-04-05 0:39 ` Jason Gunthorpe
2024-04-05 3:33 ` Zeng, Oak
2024-04-05 12:37 ` Jason Gunthorpe
2024-04-05 16:42 ` Zeng, Oak
2024-04-05 18:02 ` Jason Gunthorpe
2024-04-09 16:45 ` Zeng, Oak
2024-04-09 17:24 ` Jason Gunthorpe
2024-04-23 21:17 ` Zeng, Oak
2024-04-24 2:31 ` Matthew Brost
2024-04-24 13:57 ` Jason Gunthorpe
2024-04-24 16:35 ` Matthew Brost
2024-04-24 16:44 ` Jason Gunthorpe
2024-04-24 16:56 ` Matthew Brost
2024-04-24 17:48 ` Jason Gunthorpe
2024-04-24 13:48 ` Jason Gunthorpe
2024-04-24 23:59 ` Zeng, Oak
2024-04-25 1:05 ` Jason Gunthorpe
2024-04-26 9:55 ` Thomas Hellström
2024-04-26 12:00 ` Jason Gunthorpe
2024-04-26 14:49 ` Thomas Hellström
2024-04-26 16:35 ` Jason Gunthorpe
2024-04-29 8:25 ` Thomas Hellström
2024-04-30 17:30 ` Jason Gunthorpe
2024-04-30 18:57 ` Daniel Vetter
2024-05-01 0:09 ` Jason Gunthorpe
2024-05-02 8:04 ` Daniel Vetter
2024-05-02 9:11 ` Thomas Hellström
2024-05-02 12:46 ` Jason Gunthorpe
2024-05-02 15:01 ` Thomas Hellström
2024-05-02 19:25 ` Zeng, Oak
2024-05-03 13:37 ` Jason Gunthorpe
2024-05-03 14:43 ` Zeng, Oak
2024-05-03 16:28 ` Jason Gunthorpe
2024-05-03 20:29 ` Zeng, Oak
2024-05-04 1:03 ` Dave Airlie
2024-05-06 13:04 ` Daniel Vetter
2024-05-06 23:50 ` Matthew Brost
2024-05-07 11:56 ` Jason Gunthorpe
2024-05-06 13:33 ` Jason Gunthorpe
2024-04-09 17:33 ` Matthew Brost
2024-01-17 22:12 ` [PATCH 07/23] drm/xe/svm: Add helper for binding hmm range to gpu Oak Zeng
2024-01-17 22:12 ` [PATCH 08/23] drm/xe/svm: Add helper to invalidate svm range from GPU Oak Zeng
2024-01-17 22:12 ` [PATCH 09/23] drm/xe/svm: Remap and provide memmap backing for GPU vram Oak Zeng
2024-01-17 22:12 ` [PATCH 10/23] drm/xe/svm: Introduce svm migration function Oak Zeng
2024-01-17 22:12 ` [PATCH 11/23] drm/xe/svm: implement functions to allocate and free device memory Oak Zeng
2024-01-17 22:12 ` [PATCH 12/23] drm/xe/svm: Trace buddy block allocation and free Oak Zeng
2024-01-17 22:12 ` [PATCH 13/23] drm/xe/svm: Handle CPU page fault Oak Zeng
2024-01-17 22:12 ` [PATCH 14/23] drm/xe/svm: trace svm range migration Oak Zeng
2024-01-17 22:12 ` [PATCH 15/23] drm/xe/svm: Implement functions to register and unregister mmu notifier Oak Zeng
2024-01-17 22:12 ` [PATCH 16/23] drm/xe/svm: Implement the mmu notifier range invalidate callback Oak Zeng
2024-01-17 22:12 ` [PATCH 17/23] drm/xe/svm: clean up svm range during process exit Oak Zeng
2024-01-17 22:12 ` [PATCH 18/23] drm/xe/svm: Move a few structures to xe_gt.h Oak Zeng
2024-01-17 22:12 ` [PATCH 19/23] drm/xe/svm: migrate svm range to vram Oak Zeng
2024-01-17 22:12 ` [PATCH 20/23] drm/xe/svm: Populate svm range Oak Zeng
2024-01-17 22:12 ` [PATCH 21/23] drm/xe/svm: GPU page fault support Oak Zeng
2024-01-23 2:06 ` Welty, Brian
2024-01-23 3:09 ` Zeng, Oak
2024-01-23 3:21 ` Making drm_gpuvm work across gpu devices Zeng, Oak
2024-01-23 11:13 ` Christian König
2024-01-23 19:37 ` Zeng, Oak
2024-01-23 20:17 ` Felix Kuehling
2024-01-25 1:39 ` Zeng, Oak
2024-01-23 23:56 ` Danilo Krummrich
2024-01-24 3:57 ` Zeng, Oak
2024-01-24 4:14 ` Zeng, Oak
2024-01-24 6:48 ` Christian König
2024-01-25 22:13 ` Danilo Krummrich
2024-01-24 8:33 ` Christian König
2024-01-25 1:17 ` Zeng, Oak
2024-01-25 1:25 ` David Airlie
2024-01-25 5:25 ` Zeng, Oak
2024-01-26 10:09 ` Christian König
2024-01-26 20:13 ` Zeng, Oak
2024-01-29 10:10 ` Christian König
2024-01-29 20:09 ` Zeng, Oak
2024-01-25 11:00 ` 回复：Making " 周春明(日月)
2024-01-25 17:00 ` Zeng, Oak
2024-01-25 17:15 ` Making " Felix Kuehling
2024-01-25 18:37 ` Zeng, Oak
2024-01-26 13:23 ` Christian König
2024-01-25 16:42 ` Zeng, Oak
2024-01-25 18:32 ` Daniel Vetter
2024-01-25 21:02 ` Zeng, Oak
2024-01-26 8:21 ` Thomas Hellström
2024-01-26 12:52 ` Christian König
2024-01-27 2:21 ` Zeng, Oak [this message]
2024-01-29 10:19 ` Christian König
2024-01-30 0:21 ` Zeng, Oak
2024-01-30 8:39 ` Christian König
2024-01-30 22:29 ` Zeng, Oak
2024-01-30 23:12 ` David Airlie
2024-01-31 9:15 ` Daniel Vetter
2024-01-31 20:17 ` Zeng, Oak
2024-01-31 20:59 ` Zeng, Oak
2024-02-01 8:52 ` Christian König
2024-02-29 18:22 ` Zeng, Oak
2024-03-08 4:43 ` Zeng, Oak
2024-03-08 10:07 ` Christian König
2024-01-30 8:43 ` Thomas Hellström
2024-01-29 15:03 ` Felix Kuehling
2024-01-29 15:33 ` Christian König
2024-01-29 16:24 ` Felix Kuehling
2024-01-29 16:28 ` Christian König
2024-01-29 17:52 ` Felix Kuehling
2024-01-29 19:03 ` Christian König
2024-01-29 20:24 ` Felix Kuehling
2024-02-23 20:12 ` Zeng, Oak
2024-02-27 6:54 ` Christian König
2024-02-27 15:58 ` Zeng, Oak
2024-02-28 19:51 ` Zeng, Oak
2024-02-29 9:41 ` Christian König
2024-02-29 16:05 ` Zeng, Oak
2024-02-29 17:12 ` Thomas Hellström
2024-03-01 7:01 ` Christian König
2024-01-17 22:12 ` [PATCH 22/23] drm/xe/svm: Add DRM_XE_SVM kernel config entry Oak Zeng
2024-01-17 22:12 ` [PATCH 23/23] drm/xe/svm: Add svm memory hints interface Oak Zeng
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=SA1PR11MB69919C19A16C007A6B01FC9E92782@SA1PR11MB6991.namprd11.prod.outlook.com \
--to=oak.zeng@intel.com \
--cc=airlied@redhat.com \
--cc=brian.welty@intel.com \
--cc=christian.koenig@amd.com \
--cc=dakr@redhat.com \
--cc=daniel@ffwll.ch \
--cc=dri-devel@lists.freedesktop.org \
--cc=felix.kuehling@amd.com \
--cc=himal.prasad.ghimiray@intel.com \
--cc=intel-xe@lists.freedesktop.org \
--cc=krishnaiah.bommu@intel.com \
--cc=matthew.brost@intel.com \
--cc=niranjana.vishwanathapura@intel.com \
--cc=saurabhg.gupta@intel.com \
--cc=thomas.hellstrom@linux.intel.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).