lkml.org 
[lkml]   [2018]   [Mar]   [23]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
Subject[RFC PATCH 0/3] mmu_notifier contextual information
Date
From: Jérôme Glisse <jglisse@redhat.com>

This patchset are the improvements to mmu_notifier i wish to discuss
at next LSF/MM. I am sending now to give time to people to look at
them and think about them.

git://people.freedesktop.org/~glisse/linux mmu-notifier-rfc
https://cgit.freedesktop.org/~glisse/linux/log/?h=mmu-notifier-rfc

First patch just use a struct for invalidate_range_start/end arguments
this make the other 2 patches easier and smaller.

The idea is to provide more information to mmu_notifier listener on
the context of each invalidation. When a range is invalidated this
can be for various reasons (munmap, protection change, OOM, ...). If
listener can distinguish between those it can take better action.

For instance if device driver allocate structure to track a range of
virtual address prior to this patch it always have to assume that it
has to free those on each mmu_notifieir callback (having to assume it
is a munmap) and reallocate those latter when the device try to do
something with that range again.

OOM is also an interesting case, recently a patchset was added to
avoid OOM on a mm if a blocking mmu_notifier listener have been
registered [1]. This can be improve by adding a new OOM event type and
having listener take special path on those. All mmu_notifier i know
can easily have a special path for OOM that do not block (beside
taking a short lived, across driver, spinlock). If mmu_notifier usage
grows (from a point of view of more process using devices that rely on
them) then we should also make sure OOM can do its bidding.


The last part of the patchset is to allow more concurrency between a
range being invalidated and someone wanting to look at CPU page table
for a different range of address. I don't have any benchmark for those
but i expect this will be common with HMM and mirror once we can run
real workload. It can also replace lot of custom and weird counting
of active mmu_notifier done listener side (KVM, ODP, ...) with some-
thing cleaner.


I have try to leverage all this in KVM but it did not seems to give any
significant performance improvements (KVM patches at [2]). Tested with
the host kernel using this patchset and KVM patches, and running thing
like kernel compilation in the guest. Maybe it is not the kind of work-
load that can benefit from this.


[1] http://lkml.iu.edu/hypermail/linux/kernel/1712.1/02108.html
[2] https://cgit.freedesktop.org/~glisse/linux/log/?h=mmu-notifier-rfc-kvm

Cc: David Rientjes <rientjes@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Joerg Roedel <joro@8bytes.org>
Cc: Christian König <christian.koenig@amd.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Leon Romanovsky <leonro@mellanox.com>
Cc: Artemy Kovalyov <artemyko@mellanox.com>
Cc: Evgeny Baskakov <ebaskakov@nvidia.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mark Hairgrove <mhairgrove@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Mike Marciniszyn <mike.marciniszyn@intel.com>
Cc: Dennis Dalessandro <dennis.dalessandro@intel.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: Sudeep Dutt <sudeep.dutt@intel.com>
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Cc: Dimitri Sivanich <sivanich@sgi.com>

Jérôme Glisse (3):
mm/mmu_notifier: use struct for invalidate_range_start/end parameters
mm/mmu_notifier: provide context information about range invalidation
mm/mmu_notifier: keep track of ranges being invalidated

drivers/gpu/drm/amd/amdgpu/amdgpu_mn.c | 17 ++---
drivers/gpu/drm/i915/i915_gem_userptr.c | 13 ++--
drivers/gpu/drm/radeon/radeon_mn.c | 11 +--
drivers/infiniband/core/umem_odp.c | 16 ++--
drivers/infiniband/hw/hfi1/mmu_rb.c | 12 ++-
drivers/misc/mic/scif/scif_dma.c | 10 +--
drivers/misc/sgi-gru/grutlbpurge.c | 13 ++--
drivers/xen/gntdev.c | 7 +-
fs/dax.c | 8 +-
fs/proc/task_mmu.c | 8 +-
include/linux/mm.h | 3 +-
include/linux/mmu_notifier.h | 129 ++++++++++++++++++++++++++------
kernel/events/uprobes.c | 11 +--
mm/hmm.c | 15 ++--
mm/huge_memory.c | 69 +++++++++--------
mm/hugetlb.c | 47 ++++++------
mm/khugepaged.c | 12 +--
mm/ksm.c | 24 +++---
mm/madvise.c | 21 +++---
mm/memory.c | 97 +++++++++++++-----------
mm/migrate.c | 47 ++++++------
mm/mmu_notifier.c | 44 +++++++++--
mm/mprotect.c | 14 ++--
mm/mremap.c | 12 +--
mm/oom_kill.c | 19 +++--
mm/rmap.c | 22 ++++--
virt/kvm/kvm_main.c | 12 +--
27 files changed, 420 insertions(+), 293 deletions(-)

--
2.14.3

\
 
 \ /
  Last update: 2018-03-23 18:20    [W:0.055 / U:12.132 seconds]
©2003-2018 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site