This patchset makes it possible for MGLRU to consult secondary MMUs
while doing aging, not just during eviction. This allows for more
accurate reclaim decisions, which is especially important for proactive
reclaim.
This series makes the necessary MMU notifier changes to MGLRU and then
includes optimizations on top of that. This series also now includes
changes to access_tracking_perf_test to verify that aging works properly
for pages that are mainly used by KVM.
access_tracking_perf_test also has a mode (-p) to check performance of
MGLRU aging while the VM is faulting memory in. Here are some results:
Testing MGLRU aging while vCPUs are faulting in memory on x86 with the
TDP MMU. THPs disabled.
The test results varied a decent amount from run to run. I did my best to
take representative averages, but nonetheless, the big picture is the
important part.
Main takeaways:
- With the optimizations, the workload is much less impacted
by the presence of aging.
- With the optimizations, MGLRU is able to do aging much more
quickly, especially at 8+ vCPUs on my machine.
./access_tracking_perf_test -p -l -b 1G -v $N_VCPUS # 1G per vCPU
num_vcpus vcpu wall time aging avg pass time
1 (no aging) 0.878822016 n/a
1 (no opt) 0.938250568 0.008236007
1 (opt) 0.912270190 0.007314582
2 (no aging) 0.984959659 n/a
2 (no opt) 1.057880728 0.017989741
2 (opt) 1.037735641 0.013996319
4 (no aging) 1.264881581 n/a
4 (no opt) 1.318849182 0.056164918
4 (opt) 1.314653465 0.029311993
8 (no aging) 1.473883699 n/a
8 (no opt) 1.589441079 0.227419586s
8 (opt) 1.498439592 0.063857740s
16 (no aging) 2.048766096 n/a
16 (no opt) 2.399335597 1.247142841s
16 (opt) 2.000914001 0.121628689s
32 (no aging) 3.316256321 n/a
32 (no opt) 3.955417018 4.347290433
32 (opt) 3.355274507 0.250886289
64 (no aging) 6.498958516 n/a
64 (no opt) 7.127533884 9.815592054
64 (opt) 6.442582168 1.392907010
112 (no aging) 8.498029491 n/a
112 (no opt) 10.21372495 13.47381656
112 (opt) 8.896963554 2.292223850
Previous versions of this series included logic in MGLRU and KVM to
support batching the updates to secondary page tables. This version
removes this logic, as it was complex and not necessary to enable
proactive reclaim. This optimization, as well as the additional
optimizations for arm64 and powerpc, can be done in a later series.
Changes since v3[1]:
- Vastly simplified the series (thanks David). Removed mmu notifier
batching logic entirely.
- Cleaned up how locking is done for mmu_notifier_test/clear_young
(thanks David).
- Look-around is now only done when there are no secondary MMUs
subscribed to MMU notifiers.
- CONFIG_LRU_GEN_WALKS_SECONDARY_MMU has been added.
- Fixed the lockless implementation of kvm_{test,}age_gfn for x86
(thanks David).
- Added MGLRU functional and performance tests to
access_tracking_perf_test (thanks Axel).
- In v3, an mm would be completely ignored (for aging) if there was a
secondary MMU but support for secondary MMU walking was missing. Now,
missing secondary MMU walking support simply skips the notifier
calls (except for eviction).
- Added a sanity check for that range->lockless and range->on_lock are
never both provided for the memslot walk.
For the changes from v2[2] to v3, see v3[1].
This series applies cleanly to mm/mm-unstable and kvm/queue.
[1]: https://lore.kernel.org/linux-mm/20240401232946.1837665-1-jthoughton@xxxxxxxxxx/
[2]: https://lore.kernel.org/kvmarm/20230526234435.662652-1-yuzhao@xxxxxxxxxx/
James Houghton (7):
mm/Kconfig: Add LRU_GEN_WALKS_SECONDARY_MMU
mm: multi-gen LRU: Have secondary MMUs participate in aging
KVM: Add lockless memslot walk to KVM
KVM: Move MMU lock acquisition for test/clear_young to architecture
KVM: x86: Relax locking for kvm_test_age_gfn and kvm_age_gfn
KVM: arm64: Relax locking for kvm_test_age_gfn and kvm_age_gfn
KVM: selftests: Add multi-gen LRU aging to access_tracking_perf_test
Documentation/admin-guide/mm/multigen_lru.rst | 6 +-
arch/arm64/kvm/hyp/pgtable.c | 9 +-
arch/arm64/kvm/mmu.c | 30 +-
arch/loongarch/kvm/mmu.c | 20 +-
arch/mips/kvm/mmu.c | 21 +-
arch/powerpc/kvm/book3s.c | 14 +-
arch/riscv/kvm/mmu.c | 26 +-
arch/x86/include/asm/kvm_host.h | 1 +
arch/x86/kvm/mmu/mmu.c | 10 +-
arch/x86/kvm/mmu/tdp_iter.h | 27 +-
arch/x86/kvm/mmu/tdp_mmu.c | 67 ++-
include/linux/kvm_host.h | 1 +
include/linux/mmzone.h | 6 +-
mm/Kconfig | 8 +
mm/rmap.c | 9 +-
mm/vmscan.c | 144 +++++--
tools/testing/selftests/kvm/Makefile | 1 +
.../selftests/kvm/access_tracking_perf_test.c | 365 ++++++++++++++--
.../selftests/kvm/include/lru_gen_util.h | 55 +++
.../testing/selftests/kvm/lib/lru_gen_util.c | 391 ++++++++++++++++++
virt/kvm/kvm_main.c | 38 +-
21 files changed, 1104 insertions(+), 145 deletions(-)
create mode 100644 tools/testing/selftests/kvm/include/lru_gen_util.h
create mode 100644 tools/testing/selftests/kvm/lib/lru_gen_util.c
base-commit: e0cce98fe279b64f4a7d81b7f5c3a23d80b92fbc
--
2.45.1.288.g0e0cd299f1-goog
