linux-xfs.vger.kernel.org archive mirror
 help / color / mirror / Atom feed
From: Felix Kuehling <felix.kuehling@amd.com>
To: Christoph Hellwig <hch@lst.de>
Cc: "Sierra Guiza, Alejandro (Alex)" <alex.sierra@amd.com>,
	akpm@linux-foundation.org, linux-mm@kvack.org,
	rcampbell@nvidia.com, linux-ext4@vger.kernel.org,
	linux-xfs@vger.kernel.org, amd-gfx@lists.freedesktop.org,
	dri-devel@lists.freedesktop.org, jgg@nvidia.com,
	jglisse@redhat.com
Subject: Re: [PATCH v1 03/14] mm: add iomem vma selection for memory migration
Date: Wed, 1 Sep 2021 11:40:43 -0400	[thread overview]
Message-ID: <11d64457-9d61-f82d-6c98-d68762dce85d@amd.com> (raw)
In-Reply-To: <20210901082925.GA21961@lst.de>


Am 2021-09-01 um 4:29 a.m. schrieb Christoph Hellwig:
> On Mon, Aug 30, 2021 at 01:04:43PM -0400, Felix Kuehling wrote:
>>>> driver code is not really involved in updating the CPU mappings. Maybe
>>>> it's something we need to do in the migration helpers.
>>> It looks like I'm totally misunderstanding what you are adding here
>>> then.  Why do we need any special treatment at all for memory that
>>> has normal struct pages and is part of the direct kernel map?
>> The pages are like normal memory for purposes of mapping them in CPU
>> page tables and for coherent access from the CPU.
> That's the user page tables.  What about the kernel direct map?
> If there is a normal kernel struct page backing there really should
> be no need for the pgmap.

I'm not sure. The physical address ranges are in the UEFI system address
map as special-purpose memory. Does Linux create the struct pages and
kernel direct map for that without a pgmap call? I didn't see that last
time I went digging through that code.


>
>> From an application
>> perspective, we want file-backed and anonymous mappings to be able to
>> use DEVICE_PUBLIC pages with coherent CPU access. The goal is to
>> optimize performance for GPU heavy workloads while minimizing the need
>> to migrate data back-and-forth between system memory and device memory.
> I don't really understand that part.  file backed pages are always
> allocated by the file system using the pagecache helpers, that is
> using the page allocator.  Anonymouns memory also always comes from
> the page allocator.

I'm coming at this from my experience with DEVICE_PRIVATE. Both
anonymous and file-backed pages should be migrateable to DEVICE_PRIVATE
memory by the migrate_vma_* helpers for more efficient access by our
GPU. (*) It's part of the basic premise of HMM as I understand it. I
would expect the same thing to work for DEVICE_PUBLIC memory.

(*) I believe migrating file-backed pages to DEVICE_PRIVATE doesn't
currently work, but that's something I'm hoping to fix at some point.


>
>> The pages are special in two ways:
>>
>>  1. The memory is managed not by the Linux buddy allocator, but by the
>>     GPU driver's TTM memory manager
> Why?

It's a system architectural decision based on the access latency to the
memory and the expected use cases that we do not want the GPU driver and
the Linux buddy allocator and VM subsystem competing for the same device
memory.


>
>>  2. We want to migrate data in response to GPU page faults and
>>     application hints using the migrate_vma helpers
> Why? 

Device memory has much higher bandwidth and much lower latency than
regular system memory for the GPU to access. It's essential for enabling
good GPU application performance. Page-based memory migration enables
good performance with more intuitive programming models such as
managed/unified memory in HIP or unified shared memory in OpenMP. We do
this on our discrete GPUs with DEVICE_PRIVATE memory.

I see DEVICE_PUBLIC as an improved version of DEVICE_PRIVATE that allows
the CPU to map the device memory coherently to minimize the need for
migrations when CPU and GPU access the same memory concurrently or
alternatingly. But we're not going as far as putting that memory
entirely under the management of the Linux memory manager and VM
subsystem. Our (and HPE's) system architects decided that this memory is
not suitable to be used like regular NUMA system memory by the Linux
memory manager.

Regards,
  Felix



  reply	other threads:[~2021-09-01 15:40 UTC|newest]

Thread overview: 42+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-08-25  3:48 [PATCH v1 00/14] Add MEMORY_DEVICE_PUBLIC for CPU-accessible coherent device memory Alex Sierra
2021-08-25  3:48 ` [PATCH v1 01/14] ext4/xfs: add page refcount helper Alex Sierra
2021-08-25  7:35   ` Christoph Hellwig
2021-08-25 15:09   ` Theodore Ts'o
2021-08-25 15:49   ` Darrick J. Wong
2021-08-25  3:48 ` [PATCH v1 02/14] mm: remove extra ZONE_DEVICE struct page refcount Alex Sierra
2021-08-25  7:35   ` Christoph Hellwig
2021-08-25 11:15   ` Vlastimil Babka
2021-08-25 17:49     ` Ralph Campbell
2021-08-27 11:26       ` Vlastimil Babka
2021-08-25  3:48 ` [PATCH v1 03/14] mm: add iomem vma selection for memory migration Alex Sierra
2021-08-25  7:40   ` Christoph Hellwig
2021-08-25  7:46   ` Christoph Hellwig
2021-08-25 18:24     ` Sierra Guiza, Alejandro (Alex)
2021-08-26 22:27       ` Felix Kuehling
2021-08-30  8:28         ` Christoph Hellwig
2021-08-30 17:04           ` Felix Kuehling
2021-09-01  8:29             ` Christoph Hellwig
2021-09-01 15:40               ` Felix Kuehling [this message]
2021-09-01 22:03                 ` Dave Chinner
2021-09-01 23:07                   ` Felix Kuehling
2021-09-02  1:14                     ` Dave Chinner
2021-09-09  4:55                       ` Felix Kuehling
2021-09-02  8:18                 ` Christoph Hellwig
2021-09-02 18:07                   ` Dan Williams
2021-09-09  4:02                   ` Felix Kuehling
2021-08-25 14:24   ` Felix Kuehling
2021-08-25  3:48 ` [PATCH v1 04/14] mm: add zone device public type memory support Alex Sierra
2021-08-25  3:48 ` [PATCH v1 05/14] drm/amdkfd: ref count init for device pages Alex Sierra
2021-08-25 14:34   ` Felix Kuehling
2021-08-25  3:48 ` [PATCH v1 06/14] drm/amdkfd: add SPM support for SVM Alex Sierra
2021-08-25 14:45   ` Felix Kuehling
2021-08-25  3:48 ` [PATCH v1 07/14] drm/amdkfd: public type as sys mem on migration to ram Alex Sierra
2021-08-25  3:48 ` [PATCH v1 08/14] mm: add public type support to migrate_vma helpers Alex Sierra
2021-08-25  7:47   ` Christoph Hellwig
2021-08-25  3:48 ` [PATCH v1 09/14] mm: call pgmap->ops->page_free for DEVICE_PUBLIC pages Alex Sierra
2021-08-25  7:46   ` Christoph Hellwig
2021-08-25  3:48 ` [PATCH v1 10/14] lib: test_hmm add ioctl to get zone device type Alex Sierra
2021-08-25  3:48 ` [PATCH v1 11/14] lib: test_hmm add module param for " Alex Sierra
2021-08-25  3:48 ` [PATCH v1 12/14] lib: add support for device public type in test_hmm Alex Sierra
2021-08-25  3:48 ` [PATCH v1 13/14] tools: update hmm-test to support device public type Alex Sierra
2021-08-25  3:48 ` [PATCH v1 14/14] tools: update test_hmm script to support SP config Alex Sierra

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=11d64457-9d61-f82d-6c98-d68762dce85d@amd.com \
    --to=felix.kuehling@amd.com \
    --cc=akpm@linux-foundation.org \
    --cc=alex.sierra@amd.com \
    --cc=amd-gfx@lists.freedesktop.org \
    --cc=dri-devel@lists.freedesktop.org \
    --cc=hch@lst.de \
    --cc=jgg@nvidia.com \
    --cc=jglisse@redhat.com \
    --cc=linux-ext4@vger.kernel.org \
    --cc=linux-mm@kvack.org \
    --cc=linux-xfs@vger.kernel.org \
    --cc=rcampbell@nvidia.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 NNTP newsgroup(s).