Skip to content

RFC Migrate on fault for device pages

Mika Penttilä linux-mm lkml oe-kbuild-all 1 version In review

Get series with b4
b4 am [email protected]

Versions newest first

  1. v1 RFC Latest4 patches 17 repliesIn review

Patches in v1 No applies recorded yet

#Subject and filesSizeRepliesApplied
1/4mm: use current as mmu notifier's ownerlib/test_hmm.c mm/huge_memory.c mm/rmap.c+10 -512 replies
2/4mm: unified fault and migrate device page pathsinclude/linux/hmm.h include/linux/migrate.h mm/hmm.c +1 more+420 -193 replies
3/4mm:/migrate_device.c: remove migrate_vma_collect_*() functionsmm/migrate_device.c+1 -3110 replies
4/4mm: add new testcase for the migrate on fault caselib/test_hmm.c lib/test_hmm_uapi.h tools/testing/selftests/mm/hmm-tests.c+161 -90 replies

Reviews trailers per patch

No Reviewed-by, Acked-by or Tested-by trailers for v1 yet.

Read the v1 thread

Cover letter [RFC PATCH 0/4] Migrate on fault for device pages

As of this writing, the way device page faulting and migration works is not optimal, if you want to do both fault handling and migration at once.

Being able to migrate not present pages (or pages mapped with incorrect permissions, eg. COW) to the GPU requires doing either of the following sequences:

1. hmm_range_fault() - fault in non-present pages with correct permissions,etc.
2. migrate_vma_*() - migrate the pages

Or:

1. migrate_vma_*() - migrate present pages
2. If non-present pages detected by migrate_vma_*():
a) call hmm_range_fault() to fault pages in
b) call migrate_vma_*() again to migrate now present pages

The problem with the first sequence is that you always have to do two page walks even when most of the time the pages are present or zero page mappings so the common case takes a performance hit.

The second sequence is better for the common case, but far worse if pages aren't present because now you have to walk the page tables three times (once to find the page is not present, once so hmm_range_fault() can find a non-present page to fault in and once again to setup the migration). It also tricky to code correctly.

We should be able to walk the page table once, faulting pages in as required and replacing them with migration entries if requested.

These patches add a new flag to HMM APIs, HMM_PFN_REQ_MIGRATE, which tells to prepare for migration also during fault handling. Also, for the migrate_vma_setup() call paths, a flag, MIGRATE_VMA_FAULT, is added to tell to add fault handling to migrate. The original idea came from Alistair.

These patches are based on 6.16 mainline, and they pass the HMM selftests. The support for THP pages from Balbir should not be to hard to integrate into this, and that
is something I am looking at.

Cc David Hildenbrand <[email protected]>
Cc Jason Gunthorpe <[email protected]>
Cc Leon Romanovsky <[email protected]>
Cc Alistair Popple <[email protected]>
Cc Balbir Singh <[email protected]>

Mika Penttilä (4):
mm: use current as mmu notifier's owner
mm: unified fault and migrate device page paths mm:/migrate_device.c: remove migrate_vma_collect_*() functions
mm: add new testcase for the migrate on fault case

 include/linux/hmm.h                    |  10 +-
 include/linux/migrate.h                |   6 +-
 lib/test_hmm.c                         | 105 ++++++-
 lib/test_hmm_uapi.h                    |  17 +-
 mm/hmm.c                               | 351 +++++++++++++++++++++-
 mm/huge_memory.c                       |   6 +-
 mm/migrate_device.c                    | 384 +++++--------------------
 mm/rmap.c                              |   4 +-
 tools/testing/selftests/mm/hmm-tests.c |  53 ++++
 9 files changed, 592 insertions(+), 344 deletions(-)
Signature · 1 line
2.50.0