In very high throughput workloads, the inactive scan can become overwhelmed
as you have many cores producing pages and a single core freeing. Since
Mark's introduction of batched pagequeue operations, we can now run multiple
inactive threads working on independent batches.
To avoid confusing the pid and other control algorithms, I (Jeff) do this in
a mpi-like fan out and collect model that is driven from the primary page
daemon. It decides whether the shortfall can be overcome with a single
thread and if not dispatches multiple threads and waits for their results.
The heuristic is based on timing the pageout activity and averaging a
pages-per-second variable which is exponentially decayed. This is visible in
sysctl and may be interesting for other purposes.
I (Jeff) have verified that this does indeed double our paging throughput
when used with two threads. With four we tend to run into other contention
problems. For now I would like to commit this infrastructure with only a
single thread enabled.
The number of worker threads per domain can be controlled with the
'vm.pageout_threads_per_domain' tunable.
Submitted by: jeff (earlier version)
Discussed with: markj
Tested by: pho
Sponsored by: probably Netflix (based on contemporary commits)
Differential Revision: https://reviews.freebsd.org/D21629
Finish updating comments to reflect new locking protocols introduced
over the past year. In particular, vm_page_lock is now effectively
unused.
Reviewed by: kib
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D25868
Use atomic(9) to load the lock state. Some places were doing this
already, so it was inconsistent. In initialization code, the lock state
is still initialized with plain stores.
Reviewed by: alc, kib
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D25861
vm_page_xbusy_claim() could clobber the waiter bit. For its original
use, kernel memory pages, this was not a problem since nothing would
ever block on the busy lock for such pages. r363607 introduced a new
use where this could in principle be a problem.
Fix the problem by using atomic_cmpset to update the lock owner. Since
this macro is defined only for INVARIANTS kernels the extra overhead
doesn't seem prohibitive.
Reported by: vangyzen
Reviewed by: alc, kib, vangyzen
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D25859
that might be wired. If the page is wired then it cannot be freed now,
but the thread that eventually unwires it will free it at that point.
Reviewed by: markj, kib
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D25430
A concurrent unlocked lookup can wire the page after
vm_page_release_locked() releases the last wiring, in which case
vm_page_release_locked() must not free the page. Once the xbusy lock is
acquired, that, the object lock and the fact that the page is unmapped
ensure that the wire count cannot increase, so re-check for new wirings
after the page is xbusied.
Update the comment above vm_page_wired() to reflect the new
synchronization rules.
Reported by: glebius
Reviewed by: alc, jeff, kib
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D24592
lookup pages. These variants will fall back to their locked counterparts
if the page is not present.
Discussed with: kib, markj
Differential Revision: https://reviews.freebsd.org/D23449
be able to guarantee that they can be racquired without blocking.
Reviewed by: kib
Discussed with: markj
Differential Revision: https://reviews.freebsd.org/D23506
potential bugs that access freed pages as well as providing a path
towards lockless page lookup.
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D23444
Update vm_page_scan_contig() and vm_page_reclaim_run() to stop using
vm_page_change_lock(). It has no use after r356157. Remove
vm_page_change_lock() now that it has no users.
Remove an unncessary check for wirings in vm_page_scan_contig(), which
was previously checking twice. The check is racy until
vm_page_reclaim_run() ensures that the page is unmapped, so one check is
sufficient.
Reviewed by: jeff, kib (previous versions)
Tested by: pho (previous version)
Differential Revision: https://reviews.freebsd.org/D23279
The previous series of patches orphaned some vm_page functions, so
remove them.
Reviewed by: dougm, kib
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D22886
With the previous reviews, the page lock is no longer required in order
to perform queue operations on a page. It is also no longer needed in
the page queue scans. This change effectively eliminates remaining uses
of the page lock and also the false sharing caused by multiple pages
sharing a page lock.
Reviewed by: jeff
Tested by: pho
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D22885
This is in preparation for eliminating the use of the vm_page lock for
protecting queue state operations.
Introduce the vm_page_pqstate_commit_*() functions. These functions act
as helpers around vm_page_astate_fcmpset() and are specialized for
specific types of operations. vm_page_pqstate_commit() wraps these
functions.
Convert a number of routines to use these new helpers. Use
vm_page_release_toq() in vm_page_unwire() and vm_page_release() to
atomically release a wiring reference and release the page into a queue.
This has the side effect that vm_page_unwire() will leave the page in
the active queue if it is already present there.
Convert the page queue scans to use the new helpers. Simplify
vm_pageout_reinsert_inactive(), which requeues pages that were found to
be busy during an inactive queue scan, to avoid duplicating the work of
vm_pqbatch_process_page(). In particular, if PGA_REQUEUE or
PGA_REQUEUE_HEAD is set, let that be handled during batch processing.
Reviewed by: jeff
Tested by: pho
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D22770
Differential Revision: https://reviews.freebsd.org/D22771
Differential Revision: https://reviews.freebsd.org/D22772
Differential Revision: https://reviews.freebsd.org/D22773
Differential Revision: https://reviews.freebsd.org/D22776
removed from objects including calls to free. Pages must not be xbusy
when freed and not on an object. Strengthen assertions to match these
expectations. In practice very little code had to change busy handling
to meet these rules but we can now make stronger guarantees to busy
holders and avoid conditionally dropping busy in free.
Refine vm_page_remove() and vm_page_replace() semantics now that we have
stronger guarantees about busy state. This removes redundant and
potentially problematic code that has proliferated.
Discussed with: markj
Reviewed by: kib
Differential Revision: https://reviews.freebsd.org/D22822
an exclusive object lock.
Previously swap space was freed on a best effort basis when a page that
had valid swap was dirtied, thus invalidating the swap copy. This may be
done inconsistently and requires the object lock which is not always
convenient.
Instead, track when swap space is present. The first dirty is responsible
for deleting space or setting PGA_SWAP_FREE which will trigger background
scans to free the swap space.
Simplify the locking in vm_fault_dirty() now that we can reliably identify
the first dirty.
Discussed with: alc, kib, markj
Differential Revision: https://reviews.freebsd.org/D22654
Introduce primitives vm_page_astate_load() and vm_page_astate_fcmpset()
to operate on the 32-bit per-page atomic state. Modify
vm_page_pqstate_fcmpset() to use them. No functional change intended.
Introduce PGA_QUEUE_OP_MASK, a subset of PGA_QUEUE_STATE_MASK that only
includes queue operation flags. This will be used in subsequent
patches.
Reviewed by: alc, jeff, kib
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D22753
This is a 32-bit structure embedded in each vm_page, consisting mostly
of page queue state. The use of a structure makes it easy to store a
snapshot of a page's queue state in a stack variable and use cmpset
loops to update that state without requiring the page lock.
This change merely adds the structure and updates references to atomic
state fields. No functional change intended.
Reviewed by: alc, jeff, kib
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D22650
union members in vm_page.h to store the zone and slab. Remove some nearby
dead code.
Reviewed by: markj
Differential Revision: https://reviews.freebsd.org/D22564
Record as much bits from curthread into busy_lock as fits. Low bits
for struct thread * representation are zero due to struct and zone
alignment, and they leave space for busy flags (perhaps except
statically allocated thread0). Upper bits are not very interesting
for assert, and in most practical situations recorded value should
allow to manually identify the owner with certainity.
Assert that unbusy is performed by the owner, except few places where
unbusy is done in io completion handler. For this case, add
_unchecked variants of asserts and unbusy primitives.
Reviewed by: markj (previous version)
Tested by: pho
Sponsored by: The FreeBSD Foundation
Differential revision: https://reviews.freebsd.org/D22298
We are now out of aflags bits, whereas the "flags" field only makes use
of five of its sixteen bits, so narrow "flags" to eight bits. I have no
intention of adding a new aflag in the near future, but would like to
combine the aflags, queue and act_count fields into a single atomically
updated word. This will allow vm_page_pqstate_cmpset() to become much
simpler and is a step towards eliminating the use of the page lock array
in updating per-page queue state.
The change modifies the layout of struct vm_page, so bump
__FreeBSD_version.
Reviewed by: alc, dougm, jeff, kib
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D22397
After r352110 the page lock no longer protects a page's identity, so
there is no purpose in locking the page in pmap_mincore(). Instead,
if vm.mincore_mapped is set to the non-default value of 0, re-lookup
the page after acquiring its object lock, which holds the page's
identity stable.
The change removes the last callers of vm_page_pa_tryrelock(), so
remove it.
Reviewed by: kib
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21823
Atomics are used for page busy and valid state when the shared busy is
held. The details of the locking protocol and valid and dirty
synchronization are in the updated vm_page.h comments.
Reviewed by: kib, markj
Tested by: pho
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D21594
busy acquires while held.
This allows code that would need to acquire and release a very large number
of page busy locks to use the old mechanism where busy is only checked and
not held. This comes at the cost of false positives but never false
negatives which the single consumer, vm_fault_soft_fast(), handles.
Reviewed by: kib
Tested by: pho
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D21592
This is the first in a series of patches that promotes the page busy field
to a first class lock that no longer requires the object lock for
consistency.
Reviewed by: kib, markj
Tested by: pho
Sponsored by: Netflix, Intel
Differential Revision: https://reviews.freebsd.org/D21548
Convert all remaining references to that field to "ref_count" and update
comments accordingly. No functional change intended.
Reviewed by: alc, kib
Sponsored by: Intel, Netflix
Differential Revision: https://reviews.freebsd.org/D21768
VPRC_BLOCKED is a refcount flag used to indicate that a thread is
tearing down mappings of a page. When set, it causes attempts to wire a
page via a pmap lookup to fail. It should never represent the last
reference to a page, so assert this.
Suggested by: kib
Reviewed by: alc, kib
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21639
This function loaded the page's queue index before setting PGA_DEQUEUE.
In this window the page daemon may have deactivated the page, updating
its queue index. Make the operation atomic using vm_page_pqstate_cmpset();
the page daemon will not modify the page once it observes that PGA_DEQUEUE
is set.
Reported and tested by: pho
Reviewed by: alc, kib
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21639
- Remove a dead variable from the amd64 pmap_extract_and_hold().
- Fix grammar in the vm_page_wire man page.
Reported by: alc
Reviewed by: alc, kib
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21639
- VM_ALLOC_NOCREAT will grab without creating a page.
- vm_page_grab_valid() will grab and page in if necessary.
- vm_page_busy_acquire() automates some busy acquire loops.
Discussed with: alc, kib, markj
Tested by: pho (part of larger branch)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21546
races with page busy state. The object lock is still used as an interlock
to ensure that the identity stays valid. Most callers should use
vm_page_sleep_if_busy() to handle the locking particulars.
Reviewed by: alc, kib, markj
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21255
There are several mechanisms by which a vm_page reference is held,
preventing the page from being freed back to the page allocator. In
particular, holding the page's object lock is sufficient to prevent the
page from being freed; holding the busy lock or a wiring is sufficent as
well. These references are protected by the page lock, which must
therefore be acquired for many per-page operations. This results in
false sharing since the page locks are external to the vm_page
structures themselves and each lock protects multiple structures.
Transition to using an atomically updated per-page reference counter.
The object's reference is counted using a flag bit in the counter. A
second flag bit is used to atomically block new references via
pmap_extract_and_hold() while removing managed mappings of a page.
Thus, the reference count of a page is guaranteed not to increase if the
page is unbusied, unmapped, and the object's write lock is held. As
a consequence of this, the page lock no longer protects a page's
identity; operations which move pages between objects are now
synchronized solely by the objects' locks.
The vm_page_wire() and vm_page_unwire() KPIs are changed. The former
requires that either the object lock or the busy lock is held. The
latter no longer has a return value and may free the page if it releases
the last reference to that page. vm_page_unwire_noq() behaves the same
as before; the caller is responsible for checking its return value and
freeing or enqueuing the page as appropriate. vm_page_wire_mapped() is
introduced for use in pmap_extract_and_hold(). It fails if the page is
concurrently being unmapped, typically triggering a fallback to the
fault handler. vm_page_wire() no longer requires the page lock and
vm_page_unwire() now internally acquires the page lock when releasing
the last wiring of a page (since the page lock still protects a page's
queue state). In particular, synchronization details are no longer
leaked into the caller.
The change excises the page lock from several frequently executed code
paths. In particular, vm_object_terminate() no longer bounces between
page locks as it releases an object's pages, and direct I/O and
sendfile(SF_NOCACHE) completions no longer require the page lock. In
these latter cases we now get linear scalability in the common scenario
where different threads are operating on different files.
__FreeBSD_version is bumped. The DRM ports have been updated to
accomodate the KPI changes.
Reviewed by: jeff (earlier version)
Tested by: gallatin (earlier version), pho
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20486
Queue operations on a page use the page lock when updating the page to
reflect the desired queue state, and the page queue lock when physically
enqueuing or dequeuing a page. Multiple pages share a given page lock,
but queue state is per-page; this false sharing results in heavy lock
contention.
Take a small step towards the use of atomic_cmpset to synchronize
updates to per-page queue state by introducing vm_page_pqstate_cmpset()
and using it in the page daemon. In the longer term the plan is to stop
using the page lock to protect page identity and rely only on the object
and page busy locks. However, since the page daemon avoids acquiring
the object lock except when necessary, some synchronization with a
concurrent free of the page is required. vm_page_pqstate_cmpset() can
be used to ensure that queue state updates are successful only if the
page is not scheduled for a dequeue, which is sufficient for the page
daemon.
Add vm_page_swapqueue(), which moves a page from one queue to another
using vm_page_pqstate_cmpset(). Use it in the active queue scan, which
does not use the object lock. Modify vm_page_dequeue_deferred() to
use vm_page_pqstate_cmpset() as well.
Reviewed by: kib
Discussed with: jeff
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21257
It will become useful for the page daemon to be able to directly create
a batch queue entry for a page, and without modifying the page
structure. Rename vm_pqbatch_submit_page() to vm_page_pqbatch_submit()
to keep the namespace consistent. No functional change intended.
Reviewed by: alc, kib
MFC after: 1 week
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21369
- Add a vm_pagequeue_remove() function to physically remove a page
from its queue and update the queue length.
- Remove vm_page_pagequeue_lockptr() and let vm_page_pagequeue()
return NULL for dequeued pages.
- Avoid unnecessarily reloading the queue index if vm_page_dequeue()
loses a race with a concurrent queue operation.
- Correct an always-true assertion: vm_page_dequeue() may be called
from the page allocator with the page unlocked. The assertion
m->order == VM_NFREEORDER simply tests whether the page has been
removed from the vm_phys free lists; instead, check whether the
page belongs to an object.
Reviewed by: kib
MFC after: 1 week
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D21341
Both of these functions atomically unwire a page, optionally attempt
to free the page, and enqueue or requeue the page. Add functions
vm_page_release() and vm_page_release_locked() to perform the same task.
The latter must be called with the page's object lock held.
As a side effect of this refactoring, the buffer cache will no longer
attempt to free mapped pages when completing direct I/O. This is
consistent with the handling of pages by sendfile(SF_NOCACHE).
Reviewed by: alc, kib
MFC after: 2 weeks
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20986
The hold_count and wire_count fields of struct vm_page are separate
reference counters with similar semantics. The remaining essential
differences are that holds are not counted as a reference with respect
to LRU, and holds have an implicit free-on-last unhold semantic whereas
vm_page_unwire() callers must explicitly determine whether to free the
page once the last reference to the page is released.
This change removes the KPIs which directly manipulate hold_count.
Functions such as vm_fault_quick_hold_pages() now return wired pages
instead. Since r328977 the overhead of maintaining LRU for wired pages
is lower, and in many cases vm_fault_quick_hold_pages() callers would
swap holds for wirings on the returned pages anyway, so with this change
we remove a number of page lock acquisitions.
No functional change is intended. __FreeBSD_version is bumped.
Reviewed by: alc, kib
Discussed with: jeff
Discussed with: jhb, np (cxgbe)
Tested by: pho (previous version)
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D19247
Some workloads benefit from having a per-CPU cache for
VM_FREEPOOL_DIRECT pages.
Reviewed by: dougm, kib
Discussed with: alc, jeff
MFC after: 2 weeks
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20858
Only free pages to the cache when they were allocated from that cache.
This mitigates rapid fragmentation of physical memory seen during
poudriere's dependency calculation phase. In particular, pages
belonging to broken reservations are no longer freed to the per-CPU
cache, so they get a chance to coalesce with freed pages during the
break. Otherwise, the optimized CoW handler may create object
chains in which multiple objects contain pages from the same
reservation, and the order in which we do object termination means
that the reservation is broken before all of those pages are freed,
so some of them end up in the per-CPU cache and thus permanently
fragment physical memory.
The flag may also be useful for eliding calls to vm_reserv_free_page(),
thus avoiding memory accesses for data that is likely not present
in the CPU caches.
Reviewed by: alc
Discussed with: jeff
MFC after: 1 week
Sponsored by: The FreeBSD Foundation
Differential Revision: https://reviews.freebsd.org/D20763
Use it to indicate whether the page may be safely freed following
its removal from the object. Also change vm_page_remove() to assume
that the page's object pointer is non-NULL, and have callers perform
this check instead.
This is a step towards an implementation of an atomic reference counter
for each physical page structure.
Reviewed by: alc, dougm, kib
MFC after: 1 week
Sponsored by: Netflix
Differential Revision: https://reviews.freebsd.org/D20758
