Currently both the page lock and a page queue lock must be held in
order to enqueue, dequeue or requeue a page in a given page queue.
The queue locks are a scalability bottleneck in many workloads. This
change reduces page queue lock contention by batching queue operations.
To detangle the page and page queue locks, per-CPU batch queues are
used to reference pages with pending queue operations. The requested
operation is encoded in the page's aflags field with the page lock
held, after which the page is enqueued for a deferred batch operation.
Page queue scans are similarly optimized to minimize the amount of
work performed with a page queue lock held.
Reviewed by: kib, jeff (previous versions)
Tested by: pho
Sponsored by: Dell EMC Isilon
Differential Revision: https://reviews.freebsd.org/D14893
per-cpu alloc and free of pages. The cache is filled with as few trips
to the phys allocator as possible by the use of a new
vm_phys_alloc_npages() function which allocates as many as N pages.
This code was originally by markj with the import function rewritten by
me.
Reviewed by: markj, kib
Tested by: pho
Sponsored by: Netflix, Dell/EMC Isilon
Differential Revision: https://reviews.freebsd.org/D14905
It is possible to provide insane values for size in contigmalloc(9)
request, which usually not reaches the phys allocator due to failing
KVA allocation. But with the forthcoming 4/4 i386, where 32bit
architecture has almost 4G KVA, contigmalloc(1G) is not unreasonable
outright and KVA might be available sometimes.
Then, the calculation of pa_end could wrap around, depending on the
physical address, and the checks in vm_phys_alloc_seg_contig() would
pass while the iteration in the loop after the 'done' label goes out
of the vm_page_array bounds.
Fix it by detecting the wrap.
Reported and tested by: pho
Reviewed by: alc, markj
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
Differential revision: https://reviews.freebsd.org/D14767
global to per-domain state. Protect reservations with the free lock
from the domain that they belong to. Refactor to make vm domains more
of a first class object.
Reviewed by: markj, kib, gallatin
Tested by: pho
Sponsored by: Netflix, Dell/EMC Isilon
Differential Revision: https://reviews.freebsd.org/D14000
allocated with a tag to come from the specified domain if it meets the
other constraints provided by the tag. Automatically create a tag at
the root of each bus specifying the domain local to that bus if
available.
Reviewed by: jhb, kib
Tested by: pho
Sponsored by: Netflix, Dell/EMC Isilon
Differential Revision: https://reviews.freebsd.org/D13545
userspace to control NUMA policy administratively and programmatically.
Implement domainset based iterators in the page layer.
Remove the now legacy numa_* syscalls.
Cleanup some header polution created by having seq.h in proc.h.
Reviewed by: markj, kib
Discussed with: alc
Tested by: pho
Sponsored by: Netflix, Dell/EMC Isilon
Differential Revision: https://reviews.freebsd.org/D13403
Commit r326346 moved domain iterators from physical layer to vm_page one,
but it also removed translation of freelist to flind for
vm_page_alloc_freelist() call. Before it expects VM_FREELIST_ parameter,
but after it expect freelist index.
On small WiFi boxes with few megabytes of RAM, there is only one freelist
VM_FREELIST_LOWMEM (1) and there is no VM_FREELIST_DEFAULT(0) (see file
sys/mips/include/vmparam.h). It results in freelist 1 with flind 0.
At first, this commit renames flind to freelist in vm_page_alloc_freelist
to avoid misunderstanding about input parameters. Then on physical layer it
restores translation for correct handling of freelist parameter.
Reported by: landonf
Reviewed by: jeff
Differential Revision: https://reviews.freebsd.org/D13351
This gives a marginal improvement in the vm_page_array initialization
time. Also garbage-collect the now-unused vm_phys_paddr_to_segind().
Reviewed by: alc, kib
MFC after: 1 week
Differential Revision: https://reviews.freebsd.org/D13270
Mainly focus on files that use BSD 2-Clause license, however the tool I
was using misidentified many licenses so this was mostly a manual - error
prone - task.
The Software Package Data Exchange (SPDX) group provides a specification
to make it easier for automated tools to detect and summarize well known
opensource licenses. We are gradually adopting the specification, noting
that the tags are considered only advisory and do not, in any way,
superceed or replace the license texts.
No functional change intended.
We currently initialize the vm_page array in three passes: one to zero
the array, one to initialize the "order" field of each page (necessary
when inserting them into the vm_phys buddy allocator one-by-one), and
one to initialize the remaining non-zero fields and individually insert
each page into the allocator.
Merge the three passes into one following a suggestion from alc:
initialize vm_page fields in a single pass, and use vm_phys_free_contig()
to efficiently insert physical memory segments into the buddy allocator.
This reduces the initialization time to a third or a quarter of what it
was before on most systems that I tested.
Reviewed by: alc, kib
MFC after: 3 weeks
Differential Revision: https://reviews.freebsd.org/D12248
Idle page zeroing has been disabled by default on all architectures since
r170816 and has some bugs that make it seemingly unusable. Specifically,
the idle-priority pagezero thread exacerbates contention for the free page
lock, and yields the CPU without releasing it in non-preemptive kernels. The
pagezero thread also does not behave correctly when superpage reservations
are enabled: its target is a function of v_free_count, which includes
reserved-but-free pages, but it is only able to zero pages belonging to the
physical memory allocator.
Reviewed by: alc, imp, kib
Differential Revision: https://reviews.freebsd.org/D7714
rounddown2 tends to produce longer lines than the original code
and when the code has a high indentation level it was not really
advantageous to do the replacement.
This tries to strike a balance between readability using the macros
and flexibility of having the expressions, so not everything is
converted.
VM_NUMA_ALLOC is used to enable use of domain-aware memory allocation in
the virtual memory system. DEVICE_NUMA is used to enable affinity
reporting for devices such as bus_get_domain().
MAXMEMDOM must still be set to a value greater than for any NUMA support
to be effective. Note that 'cpuset -gd' always works if MAXMEMDOM is
enabled and the system supports NUMA.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D5782
than ascending order in vm_phys_alloc_contig() so that, for example, a
sequence of contigmalloc(low=0, high=4GB) calls doesn't exhaust the supply
of low physical memory resulting in a later contigmalloc(low=0, high=1MB)
failure.
Reported by: cy
Tested by: cy
Sponsored by: EMC / Isilon Storage Division
address and use this mechanism when:
1. kmem_alloc_{attr,contig}() can't find suitable free pages in the physical
memory allocator's free page lists. This replaces the long-standing
approach of scanning the inactive and inactive queues, converting clean
pages into PG_CACHED pages and laundering dirty pages. In contrast, the
new mechanism does not use PG_CACHED pages nor does it trigger a large
number of I/O operations.
2. on 32-bit MIPS processors, uma_small_alloc() and the pmap can't find
free pages in the physical memory allocator's free page lists that are
covered by the direct map. Tested by: adrian
3. ttm_bo_global_init() and ttm_vm_page_alloc_dma32() can't find suitable
free pages in the physical memory allocator's free page lists.
In the coming months, I expect that this new mechanism will be applied in
other places. For example, balloon drivers should use relocation to
minimize fragmentation of the guest physical address space.
Make vm_phys_alloc_contig() a little smarter (and more efficient in some
cases). Specifically, use vm_phys_segs[] earlier to avoid scanning free
page lists that can't possibly contain suitable pages.
Reviewed by: kib, markj
Glanced at: jhb
Discussed with: jeff
Sponsored by: EMC / Isilon Storage Division
Differential Revision: https://reviews.freebsd.org/D4444
This is based on work done by jeff@ and jhb@, as well as the numa.diff
patch that has been circulating when someone asks for first-touch NUMA
on -10 or -11.
* Introduce a simple set of VM policy and iterator types.
* tie the policy types into the vm_phys path for now, mirroring how
the initial first-touch allocation work was enabled.
* add syscalls to control changing thread and process defaults.
* add a global NUMA VM domain policy.
* implement a simple cascade policy order - if a thread policy exists, use it;
if a process policy exists, use it; use the default policy.
* processes inherit policies from their parent processes, threads inherit
policies from their parent threads.
* add a simple tool (numactl) to query and modify default thread/process
policities.
* add documentation for the new syscalls, for numa and for numactl.
* re-enable first touch NUMA again by default, as now policies can be
set in a variety of methods.
This is only relevant for very specific workloads.
This doesn't pretend to be a final NUMA solution.
The previous defaults in -HEAD (with MAXMEMDOM set) can be achieved by
'sysctl vm.default_policy=rr'.
This is only relevant if MAXMEMDOM is set to something other than 1.
Ie, if you're using GENERIC or a modified kernel with non-NUMA, then
this is a glorified no-op for you.
Thank you to Norse Corp for giving me access to rather large
(for FreeBSD!) NUMA machines in order to develop and verify this.
Thank you to Dell for providing me with dual socket sandybridge
and westmere v3 hardware to do NUMA development with.
Thank you to Scott Long at Netflix for providing me with access
to the two-socket, four-domain haswell v3 hardware.
Thank you to Peter Holm for running the stress testing suite
against the NUMA branch during various stages of development!
Tested:
* MIPS (regression testing; non-NUMA)
* i386 (regression testing; non-NUMA GENERIC)
* amd64 (regression testing; non-NUMA GENERIC)
* westmere, 2 socket (thankyou norse!)
* sandy bridge, 2 socket (thankyou dell!)
* ivy bridge, 2 socket (thankyou norse!)
* westmere-EX, 4 socket / 1TB RAM (thankyou norse!)
* haswell, 2 socket (thankyou norse!)
* haswell v3, 2 socket (thankyou dell)
* haswell v3, 2x18 core (thankyou scott long / netflix!)
* Peter Holm ran a stress test suite on this work and found one
issue, but has not been able to verify it (it doesn't look NUMA
related, and he only saw it once over many testing runs.)
* I've tested bhyve instances running in fixed NUMA domains and cpusets;
all seems to work correctly.
Verified:
* intel-pcm - pcm-numa.x and pcm-memory.x, whilst selecting different
NUMA policies for processes under test.
Review:
This was reviewed through phabricator (https://reviews.freebsd.org/D2559)
as well as privately and via emails to freebsd-arch@. The git history
with specific attributes is available at https://github.com/erikarn/freebsd/
in the NUMA branch (https://github.com/erikarn/freebsd/compare/local/adrian_numa_policy).
This has been reviewed by a number of people (stas, rpaulo, kib, ngie,
wblock) but not achieved a clear consensus. My hope is that with further
exposure and testing more functionality can be implemented and evaluated.
Notes:
* The VM doesn't handle unbalanced domains very well, and if you have an overly
unbalanced memory setup whilst under high memory pressure, VM page allocation
may fail leading to a kernel panic. This was a problem in the past, but it's
much more easily triggered now with these tools.
* This work only controls the path through vm_phys; it doesn't yet strongly/predictably
affect contigmalloc, KVA placement, UMA, etc. So, driver placement of memory
isn't really guaranteed in any way. That's next on my plate.
Sponsored by: Norse Corp, Inc.; Dell
a basic ACPI SLIT table parser.
For now this just exports the map via sysctl; it'll eventually be useful
to userland when there's more useful NUMA support in -HEAD.
* Add an optional mem_locality map;
* add a mapping function taking from/to domain and returning the
relative cost, or -1 if it's not available;
* Add a very basic SLIT parser to x86 ACPI.
Differential Revision: https://reviews.freebsd.org/D2460
Reviewed by: rpaulo, stas, jhb
Sponsored by: Norse Corp, Inc (hardware, coding); Dell (hardware)
managing pages from different address ranges. Generally speaking, this
feature is used to increase the likelihood that physical pages are
available that can meet special DMA requirements or can be accessed through
a limited-coverage direct mapping (e.g., MIPS). However, prior to this
change, the configuration of the free lists was static, i.e., it was
determined at compile time. Consequentally, free lists could be created
for address ranges that held no actual pages, for example, on 32-bit MIPS-
based systems with 512 MB or less of physical memory. This change makes
the creation of the free lists dynamic, i.e., it is based on the available
physical memory at boot time.
On 64-bit x86-based systems with 64 GB or more of physical memory, create
free lists for managing pages with physical addresses below 4 GB. This
change is to address reported problems with initializing devices that
require the allocation of physical pages below 4 GB on some systems with
128 GB or more of physical memory.
PR: 185727
Differential Revision: https://reviews.freebsd.org/D1274
Reviewed by: jhb, kib
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
on i386 PAE. Previously, VM_PHYSSEG_SPARSE could not be used on amd64 and
i386 because vm_page_startup() would not create vm_page structures for the
kernel page table pages allocated during pmap_bootstrap() but those vm_page
structures are needed when the kernel attempts to promote the corresponding
kernel virtual addresses to superpage mappings. To address this problem, a
new public function, vm_phys_add_seg(), is introduced and vm_phys_init() is
updated to reflect the creation of vm_phys_seg structures by calls to
vm_phys_add_seg().
Discussed with: Svatopluk Kraus
MFC after: 3 weeks
Sponsored by: EMC / Isilon Storage Division
With the current implementation of managed fictitious ranges when
also using VM_PHYSSEG_DENSE, a user could try to register a
fictitious range that starts inside of vm_page_array, but then
overrruns it (because the end of the fictitious range is greater than
vm_page_array_size + first_page). This would result in PHYS_TO_VM_PAGE
returning unallocated pages from past the end of vm_page_array. The
same could happen if a user tried to register a segment that starts
outside of vm_page_array but ends inside of it.
In order to fix this, allow vm_phys_fictitious_{reg/unreg}_range to
use a set of pages from vm_page_array, and allocate the rest.
Sponsored by: Citrix Systems R&D
Reviewed by: kib, alc
vm/vm_phys.c:
- Allow registering/unregistering fictitious ranges that overrun
vm_page_array.
The number of vm fictitious regions was limited to 8 by default, but
Xen will make heavy usage of those kind of regions in order to map
memory from foreign domains, so instead of increasing the default
number, change the implementation to use a red-black tree to track vm
fictitious ranges.
The public interface remains the same.
Sponsored by: Citrix Systems R&D
Reviewed by: kib, alc
Approved by: gibbs
vm/vm_phys.c:
- Replace the vm fictitious static array with a red-black tree.
- Use a rwlock instead of a mutex, since now we also need to take the
lock in vm_phys_fictitious_to_vm_page, and it can be shared.
into the area backed by vm_page_array wrongly compared end with
vm_page_array_size. It should be adjusted by first_page index to be
correct.
Also, the corner and incorrect case of the requested range extending
after the end of the vm_page_array was incorrectly handled by
allocating the segment.
Fix the comparision for the end of range and return EINVAL if the end
extends beyond vm_page_array.
Discussed with: royger
Sponsored by: The FreeBSD Foundation
MFC after: 1 week
To reduce the diff struct pcu.cnt field was not renamed, so
PCPU_OP(cnt.field) is still used. pc_cnt and pcpu are also used in
kvm(3) and vmstat(8). The goal was to not affect externally used KPI.
Bump __FreeBSD_version_ in case some out-of-tree module/code relies on the
the global cnt variable.
Exp-run revealed no ports using it directly.
No objection from: arch@
Sponsored by: EMC / Isilon Storage Division
no longer any need for the page's PG_CACHED and PG_FREE flags to be set and
cleared while the free page queues lock is held. Thus, vm_page_alloc(),
vm_page_alloc_contig(), and vm_page_alloc_freelist() can wait until after
the free page queues lock is released to clear the page's flags. Moreover,
the PG_FREE flag can be retired. Now that the reservation system no longer
uses it, its only uses are in a few assertions. Eliminating these
assertions is no real loss. Other assertions catch the same types of
misbehavior, like doubly freeing a page (see r260032) or dirtying a free
page (free pages are invalid and only valid pages can be dirtied).
Eliminate an unneeded variable from vm_page_alloc_contig().
Sponsored by: EMC / Isilon Storage Division
additional information, when the page is guaranteed to not belong to a
paging queue. Usually, this results in a lot of type casts which make
reasoning about the code correctness harder.
Sometimes m->object is used instead of pageq, which could cause real
and confusing bugs if non-NULL m->object is leaked. See r141955 and
r253140 for examples.
Change the pageq member into a union containing explicitly-typed
members. Use them instead of type-punning or abusing m->object in x86
pmaps, uma and vm_page_alloc_contig().
Requested and reviewed by: alc
Sponsored by: The FreeBSD Foundation
Unify the 2 concept into a real, minimal, sxlock where the shared
acquisition represent the soft busy and the exclusive acquisition
represent the hard busy.
The old VPO_WANTED mechanism becames the hard-path for this new lock
and it becomes per-page rather than per-object.
The vm_object lock becames an interlock for this functionality:
it can be held in both read or write mode.
However, if the vm_object lock is held in read mode while acquiring
or releasing the busy state, the thread owner cannot make any
assumption on the busy state unless it is also busying it.
Also:
- Add a new flag to directly shared busy pages while vm_page_alloc
and vm_page_grab are being executed. This will be very helpful
once these functions happen under a read object lock.
- Move the swapping sleep into its own per-object flag
The KPI is heavilly changed this is why the version is bumped.
It is very likely that some VM ports users will need to change
their own code.
Sponsored by: EMC / Isilon storage division
Discussed with: alc
Reviewed by: jeff, kib
Tested by: gavin, bapt (older version)
Tested by: pho, scottl
into threads each processing queue in a single domain. The structure
of the pagedaemons and queues is kept intact, most of the changes come
from the need for code to find an owning page queue for given page,
calculated from the segment containing the page.
The tie between NUMA domain and pagedaemon thread/pagequeue split is
rather arbitrary, the multithreaded daemon could be allowed for the
single-domain machines, or one domain might be split into several page
domains, to further increase concurrency.
Right now, each pagedaemon thread tries to reach the global target,
precalculated at the start of the pass. This is not optimal, since it
could cause excessive page deactivation and freeing. The code should
be changed to re-check the global page deficit state in the loop after
some number of iterations.
The pagedaemons reach the quorum before starting the OOM, since one
thread inability to meet the target is normal for split queues. Only
when all pagedaemons fail to produce enough reusable pages, OOM is
started by single selected thread.
Launder is modified to take into account the segments layout with
regard to the region for which cleaning is performed.
Based on the preliminary patch by jeff, sponsored by EMC / Isilon
Storage Division.
Reviewed by: alc
Tested by: pho
Sponsored by: The FreeBSD Foundation
reset by pmap_page_init() right after being initialized in vm_page_initfake().
The statement above is with reference to the amd64 implementation of
pmap_page_init().
Fix this by calling 'pmap_page_init()' in 'vm_page_initfake()' before changing
the 'memattr'.
Reviewed by: kib
MFC after: 2 weeks
freelist.
o Split the pool of free pages queues really by domain and not rely on
definition of VM_RAW_NFREELIST.
o For MAXMEMDOM > 1, wrap the RR allocation logic into a specific
function that is called when calculating the allocation domain.
The RR counter is kept, currently, per-thread.
In the future it is expected that such function evolves in a real
policy decision referee, based on specific informations retrieved by
per-thread and per-vm_object attributes.
o Add the concept of "probed domains" under the form of vm_ndomains.
It is responsibility for every architecture willing to support multiple
memory domains to correctly probe vm_ndomains along with mem_affinity
segments attributes. Those two values are supposed to remain always
consistent.
Please also note that vm_ndomains and td_dom_rr_idx are both int
because segments already store domains as int. Ideally u_int would
have much more sense. Probabilly this should be cleaned up in the
future.
o Apply RR domain selection also to vm_phys_zero_pages_idle().
Sponsored by: EMC / Isilon storage division
Partly obtained from: jeff
Reviewed by: alc
Tested by: jeff
order to match the MAXCPU concept. The change should also be useful
for consolidation and consistency.
Sponsored by: EMC / Isilon storage division
Obtained from: jeff
Reviewed by: alc
- vm_phys_alloc_freelist_pages() can be called by vm_page_alloc_freelist()
to allocate a page from a specific freelist. In the NUMA case it did not
properly map the public VM_FREELIST_* constants to the correct backing
freelists, nor did it try all NUMA domains for allocations from
VM_FREELIST_DEFAULT.
- vm_phys_alloc_pages() did not pin the thread and each call to
vm_phys_alloc_freelist_pages() fetched the current domain to choose
which freelist to use. If a thread migrated domains during the loop
in vm_phys_alloc_pages() it could skip one of the freelists. If the
other freelists were out of memory then it is possible that
vm_phys_alloc_pages() would fail to allocate a page even though pages
were available resulting in a panic in vm_page_alloc().
Reviewed by: alc
MFC after: 1 week
for allocation of fictitious pages, for which PHYS_TO_VM_PAGE()
returns proper fictitious vm_page_t. The range should be de-registered
after consumer stopped using it.
De-inline the PHYS_TO_VM_PAGE() since it now carries code to iterate
over registered ranges.
A hash container might be developed instead of range registration
interface, and fake pages could be put automatically into the hash,
were PHYS_TO_VM_PAGE() could look them up later. This should be
considered before the MFC of the commit is done.
Sponsored by: The FreeBSD Foundation
Reviewed by: alc
MFC after: 1 month
yielding a new public interface, vm_page_alloc_contig(). This new function
addresses some of the limitations of the current interfaces, contigmalloc()
and kmem_alloc_contig(). For example, the physically contiguous memory that
is allocated with those interfaces can only be allocated to the kernel vm
object and must be mapped into the kernel virtual address space. It also
provides functionality that vm_phys_alloc_contig() doesn't, such as wiring
the returned pages. Moreover, unlike that function, it respects the low
water marks on the paging queues and wakes up the page daemon when
necessary. That said, at present, this new function can't be applied to all
types of vm objects. However, that restriction will be eliminated in the
coming weeks.
From a design standpoint, this change also addresses an inconsistency
between vm_phys_alloc_contig() and the other vm_phys_alloc*() functions.
Specifically, vm_phys_alloc_contig() manipulated vm_page fields that other
functions in vm/vm_phys.c didn't. Moreover, vm_phys_alloc_contig() knew
about vnodes and reservations. Now, vm_page_alloc_contig() is responsible
for these things.
Reviewed by: kib
Discussed with: jhb
