The target of this is getting at the point where the recovery path is
completely removed as we could count on pre-allocation once the
path compressed trie is implemented.
than 4GB. Specifically, the inlined version of 'ptoa' of the the 'int'
count of pages overflowed on 64-bit platforms. While here, change
vm_object_madvise() to accept two vm_pindex_t parameters (start and end)
rather than a (start, count) tuple to match other VM APIs as suggested
by alc@.
if the filesystem performed short write and we are skipping the page
due to this.
Propogate write error from the pager back to the callers of
vm_pageout_flush(). Report the failure to write a page from the
requested range as the FALSE return value from vm_object_page_clean(),
and propagate it back to msync(2) to return EIO to usermode.
While there, convert the clearobjflags variable in the
vm_object_page_clean() and arguments of the helper functions to
boolean.
PR: kern/165927
Reviewed by: alc
MFC after: 2 weeks
callers of vm_page_insert().
The default action for every caller is to unwind-back the operation
besides vm_page_rename() where this has proven to be impossible to do.
For that case, it just spins until the page is not available to be
allocated. However, due to vm_page_rename() to be mostly rare (and
having never hit this panic in the past) it is tought to be a very
seldom thing and not a possible performance factor.
The patch has been tested with an atomic counter returning NULL from
the zone allocator every 1/100000 allocations. Per-printf, I've verified
that a typical buildkernel could trigger this 30 times. The patch
survived to 2 hours of repeated buildkernel/world.
Several technical notes:
- The vm_page_insert() is moved, in several callers, closer to failure
points. This could be committed separately before vmcontention hits
the tree just to verify -CURRENT is happy with it.
- vm_page_rename() does not need to have the page lock in the callers
as it hide that as an implementation detail. Do the locking internally.
- now vm_page_insert() returns an int, with 0 meaning everything was ok,
thus KPI is broken by this patch.
generation change if requested mode is async. The object generation is
only changed when the object is marked as OBJ_MIGHTBEDIRTY. For async
mode it is enough to write each dirty page, not to make a guarantee that
all pages are cleared after the vm_object_page_clean() returned.
Diagnosed by: truckman
Tested by: flo
Reviewed by: alc, truckman
MFC after: 2 weeks
MS_SYNC flag. The system must guarantee that all writes are finished
before syscalls returned. Schedule the writes in async mode, which is
much faster and allows the clustering to occur. Wait for writes using
VOP_FSYNC(), since we are syncing the whole file mapping.
Potentially, the restriction to only apply the optimization can be
relaxed by not requiring that the mapping cover whole file, as it is
done by other OSes.
Reported and tested by: az
Reviewed by: alc
MFC after: 2 weeks
The SYSCTL_NODE macro defines a list that stores all child-elements of
that node. If there's no SYSCTL_DECL macro anywhere else, there's no
reason why it shouldn't be static.
madvise(2) except that it operates on a file descriptor instead of a
memory region. It is currently only supported on regular files.
Just as with madvise(2), the advice given to posix_fadvise(2) can be
divided into two types. The first type provide hints about data access
patterns and are used in the file read and write routines to modify the
I/O flags passed down to VOP_READ() and VOP_WRITE(). These modes are
thus filesystem independent. Note that to ease implementation (and
since this API is only advisory anyway), only a single non-normal
range is allowed per file descriptor.
The second type of hints are used to hint to the OS that data will or
will not be used. These hints are implemented via a new VOP_ADVISE().
A default implementation is provided which does nothing for the WILLNEED
request and attempts to move any clean pages to the cache page queue for
the DONTNEED request. This latter case required two other changes.
First, a new V_CLEANONLY flag was added to vinvalbuf(). This requests
vinvalbuf() to only flush clean buffers for the vnode from the buffer
cache and to not remove any backing pages from the vnode. This is
used to ensure clean pages are not wired into the buffer cache before
attempting to move them to the cache page queue. The second change adds
a new vm_object_page_cache() method. This method is somewhat similar to
vm_object_page_remove() except that instead of freeing each page in the
specified range, it attempts to move clean pages to the cache queue if
possible.
To preserve the ABI of struct file, the f_cdevpriv pointer is now reused
in a union to point to the currently active advice region if one is
present for regular files.
Reviewed by: jilles, kib, arch@
Approved by: re (kib)
MFC after: 1 month
tree.
Reclaim all the nodes related to the radix tree for a specified
vm_object when calling vm_object_terminate() via the newly added
interface vm_radix_reclaim_nodes().
The function is recursive, but we have a well-defined maximum depth,
thus the amount of necessary stack can be easilly calculated.
Reported by: alc
Discussed and reviewed by: jeff
Black nodes support standard active pages and red nodes support cached
pages. Red nodes may be removed without the object lock but will not
collapse unused tree nodes. Red nodes may not be directly inserted,
instead a new function is supplied to convert between black and red.
- Handle cached pages and active pages in the same loop in vm_object_split,
vm_object_backing_scan, and vm_object_terminate.
- Retire the splay page handling as the ifdefs are too difficult to
maintain.
- Slightly optimize the vm_radix_lookupn() function.
height and a pointer so that the update to the root is atomic. This
permits safe lookups in parallel with tree expansion. Shrinking the
space requirements is a small bonus.
the vm object pages splay.
TODO:
- Handle differently the negative keys for having smaller depth
index nodes (negative keys caming from indirect blocks)
- Fix the get_node() by having support for a low reserved objects
directly from UMA
- Implement the lookup_le and re-enable VM_NRESERVELEVEL = 1
- Try to rework the superpage splay of idle pages and the cache splay
for every vm object in order to regain space on vm_page structure
- Verify performance and improve them (likely by having consumers to deal
with several ranges of pages manually?)
Obtained from: jeff, Mayur Shardul (GSoC 2009)
flags field. Updates to the atomic flags are performed using the atomic
ops on the containing word, do not require any vm lock to be held, and
are non-blocking. The vm_page_aflag_set(9) and vm_page_aflag_clear(9)
functions are provided to modify afalgs.
Document the changes to flags field to only require the page lock.
Introduce vm_page_reference(9) function to provide a stable KPI and
KBI for filesystems like tmpfs and zfs which need to mark a page as
referenced.
Reviewed by: alc, attilio
Tested by: marius, flo (sparc64); andreast (powerpc, powerpc64)
Approved by: re (bz)
to VPO_UNMANAGED (and also making the flag protected by the vm object
lock, instead of vm page queue lock).
- Mark the fake pages with both PG_FICTITIOUS (as it is now) and
VPO_UNMANAGED. As a consequence, pmap code now can use use just
VPO_UNMANAGED to decide whether the page is unmanaged.
Reviewed by: alc
Tested by: pho (x86, previous version), marius (sparc64),
marcel (arm, ia64, powerpc), ray (mips)
Sponsored by: The FreeBSD Foundation
Approved by: re (bz)
option to vm_object_page_remove() asserts that the specified range of pages
is not mapped, or more precisely that none of these pages have any managed
mappings. Thus, vm_object_page_remove() need not call pmap_remove_all() on
the pages.
This change not only saves time by eliminating pointless calls to
pmap_remove_all(), but it also eliminates an inconsistency in the use of
pmap_remove_all() versus related functions, like pmap_remove_write(). It
eliminates harmless but pointless calls to pmap_remove_all() that were being
performed on PG_UNMANAGED pages.
Update all of the existing assertions on pmap_remove_all() to reflect this
change.
Reviewed by: kib
VM_PAGER_AGAIN to VM_PAGER_ERROR for the uwritten pages. Return
VM_PAGER_AGAIN for the partially written page. Always forward at least
one page in the loop of vm_object_page_clean().
VM_PAGER_ERROR causes the page reactivation and does not clear the
page dirty state, so the write is not lost.
The change fixes an infinite loop in vm_object_page_clean() when the
filesystem returns permanent errors for some page writes.
Reported and tested by: gavin
Reviewed by: alc, rmacklem
MFC after: 1 week
Hold the vnode around the region where object lock is dropped, until
vnode lock is acquired.
Do not drop the vnode reference for a case when the object was
deallocated during unlock. Note that in this case, VV_TEXT is cleared
by vnode_pager_dealloc().
Reported and tested by: pho
Reviewed by: alc
MFC after: 3 days
incorrectly calling vm_object_page_clean(). They are passing the length of
the range rather than the ending offset of the range.
Perform the OFF_TO_IDX() conversion in vm_object_page_clean() rather than the
callers.
Reviewed by: kib
MFC after: 3 weeks
consumer of the flag, and it used the flag because OBJ_MIGHTBEDIRTY
was cleared early in vm_object_page_clean, before the cleaning pass
was done. This is no longer true after r216799.
Moreover, since OBJ_CLEANING is a flag, and not the counter, it could
be reset too prematurely when parallel vm_object_page_clean() are
performed.
Reviewed by: alc (as a part of the bigger patch)
MFC after: 1 month (after r216799 is merged)
vm_object_set_writeable_dirty().
Fix an issue where restart of the scan in vm_object_page_clean() did
not removed write permissions for newly added pages or, if the mapping
for some already scanned page changed to writeable due to fault.
Merge the two loops in vm_object_page_clean(), doing the remove of
write permission and cleaning in the same loop. The restart of the
loop then correctly downgrade writeable mappings.
Fix an issue where a second caller to msync() might actually return
before the first caller had actually completed flushing the
pages. Clear the OBJ_MIGHTBEDIRTY flag after the cleaning loop, not
before.
Calls to pmap_is_modified() are not needed after pmap_remove_write()
there.
Proposed, reviewed and tested by: alc
MFC after: 1 week
in "struct vm_object". This is required to make it possible to account
for per-jail swap usage.
Reviewed by: kib@
Tested by: pho@
Sponsored by: FreeBSD Foundation
backing storage. Such pages might be then reused, racing with the
assert in vm_object_page_collect_flush() that verified that dirty
pages from the run (most likely, pages with VM_PAGER_AGAIN status) are
write-protected still. In fact, the page indexes for the pages that
were removed from the object page list should be ignored by
vm_object_page_clean().
Return the length of successfully written run from vm_pageout_flush(),
that is, the count of pages between requested page and first page
after requested with status VM_PAGER_AGAIN. Supply the requested page
index in the array to vm_pageout_flush(). Use the returned run length
to forward the index of next page to clean in vm_object_page_clean().
Reported by: avg
Reviewed by: alc
MFC after: 1 week
object page list. The only use of object generation count now is a
restart of the scan in vm_object_page_clean(), which makes sense to do
on the page addition. Page removals do not affect the dirtiness of the
object, as well as manipulations with the shadow chain.
Suggested and reviewed by: alc
MFC after: 1 week
is ordered by page index. This greatly simplifies the implementation,
since we no longer need to mark the pages with VPO_CLEANCHK to denote
the progress. It is enough to remember the current position by index
before dropping the object lock.
Remove VPO_CLEANCHK and VM_PAGER_IGNORE_CLEANCHK as unused.
Garbage-collect vm.msync_flush_flags sysctl.
Suggested and reviewed by: alc
Tested by: pho
independent code. Move this code into mincore(), and eliminate the
page queues lock from pmap_mincore().
Push down the page queues lock into pmap_clear_modify(),
pmap_clear_reference(), and pmap_is_modified(). Assert that these
functions are never passed an unmanaged page.
Eliminate an inaccurate comment from powerpc/powerpc/mmu_if.m:
Contrary to what the comment says, pmap_mincore() is not simply an
optimization. Without a complete pmap_mincore() implementation,
mincore() cannot return either MINCORE_MODIFIED or MINCORE_REFERENCED
because only the pmap can provide this information.
Eliminate the page queues lock from vfs_setdirty_locked_object(),
vm_pageout_clean(), vm_object_page_collect_flush(), and
vm_object_page_clean(). Generally speaking, these are all accesses
to the page's dirty field, which are synchronized by the containing
vm object's lock.
Reduce the scope of the page queues lock in vm_object_madvise() and
vm_page_dontneed().
Reviewed by: kib (an earlier version)
vm_page_try_to_free(). Consequently, push down the page queues lock into
pmap_enter_quick(), pmap_page_wired_mapped(), pmap_remove_all(), and
pmap_remove_write().
Push down the page queues lock into Xen's pmap_page_is_mapped(). (I
overlooked the Xen pmap in r207702.)
Switch to a per-processor counter for the total number of pages cached.
managed pages that didn't already have that lock held. (Freeing an
unmanaged page, such as the various pmaps use, doesn't require the page
lock.)
This allows a change in vm_page_remove()'s locking requirements. It now
expects the page lock to be held instead of the page queues lock.
Consequently, the page queues lock is no longer required at all by callers
to vm_page_rename().
Discussed with: kib
architecture from page queue lock to a hashed array of page locks
(based on a patch by Jeff Roberson), I've implemented page lock
support in the MI code and have only moved vm_page's hold_count
out from under page queue mutex to page lock. This changes
pmap_extract_and_hold on all pmaps.
Supported by: Bitgravity Inc.
Discussed with: alc, jeffr, and kib
or unmanaged before acquiring the page queues lock. Neither of these
tests require that lock. Moreover, a better way of testing if the page
is unmanaged is to test the type of vm object. This avoids a pointless
vm_page_lookup().
MFC after: 3 weeks
that page only makes sense if the advice is MADV_WILLNEED. In that case,
the intention is to activate the page, so discouraging the page daemon
from reclaiming the page makes sense. In contrast, in the other cases,
MADV_DONTNEED and MADV_FREE, it makes no sense whatsoever to discourage
the page daemon from reclaiming the page by setting PG_REFERENCED.
Wrap a nearby line.
Discussed with: kib
MFC after: 3 weeks
sleeping on that page is nonsensical. Doing so reduces the likelihood
that the page daemon will reclaim the page before the thread waiting in
vm_object_backing_scan() is reawakened. However, it does not guarantee
that the page is not reclaimed, so vm_object_backing_scan() restarts
after reawakening. More importantly, this muddles the meaning of
PG_REFERENCED. There is no reason to believe that the caller of
vm_object_backing_scan() is going to use (i.e., access) the contents of
the page. There is especially no reason to believe that an access is
more likely because vm_object_backing_scan() had to sleep on the page.
Discussed with: kib
MFC after: 3 weeks
flag. Besides providing the redundand information, need to update both
vnode and object flags causes more acquisition of vnode interlock.
OBJ_MIGHTBEDIRTY is only checked for vnode-backed vm objects.
Remove VI_OBJDIRTY and make sure that OBJ_MIGHTBEDIRTY is set only for
vnode-backed vm objects.
Suggested and reviewed by: alc
Tested by: pho
MFC after: 3 weeks
a device pager (OBJT_DEVICE) object in that it uses fictitious pages to
provide aliases to other memory addresses. The primary difference is that
it uses an sglist(9) to determine the physical addresses for a given offset
into the object instead of invoking the d_mmap() method in a device driver.
Reviewed by: alc
Approved by: re (kensmith)
MFC after: 2 weeks
dependent memory attributes:
Rename vm_cache_mode_t to vm_memattr_t. The new name reflects the
fact that there are machine-dependent memory attributes that have
nothing to do with controlling the cache's behavior.
Introduce vm_object_set_memattr() for setting the default memory
attributes that will be given to an object's pages.
Introduce and use pmap_page_{get,set}_memattr() for getting and
setting a page's machine-dependent memory attributes. Add full
support for these functions on amd64 and i386 and stubs for them on
the other architectures. The function pmap_page_set_memattr() is also
responsible for any other machine-dependent aspects of changing a
page's memory attributes, such as flushing the cache or updating the
direct map. The uses include kmem_alloc_contig(), vm_page_alloc(),
and the device pager:
kmem_alloc_contig() can now be used to allocate kernel memory with
non-default memory attributes on amd64 and i386.
vm_page_alloc() and the device pager will set the memory attributes
for the real or fictitious page according to the object's default
memory attributes.
Update the various pmap functions on amd64 and i386 that map pages to
incorporate each page's memory attributes in the mapping.
Notes: (1) Inherent to this design are safety features that prevent
the specification of inconsistent memory attributes by different
mappings on amd64 and i386. In addition, the device pager provides a
warning when a device driver creates a fictitious page with memory
attributes that are inconsistent with the real page that the
fictitious page is an alias for. (2) Storing the machine-dependent
memory attributes for amd64 and i386 as a dedicated "int" in "struct
md_page" represents a compromise between space efficiency and the ease
of MFCing these changes to RELENG_7.
In collaboration with: jhb
Approved by: re (kib)
valid mask. Consequently, there is no need to perform a bit-wise and of
the page's dirty and valid masks in order to determine which parts of a
page are dirty and valid.
Eliminate an unnecessary #include.
rlimit RLIMIT_SWAP that limits the amount of swap that may be reserved
for the uid.
The accounting information (charge) is associated with either map entry,
or vm object backing the entry, assuming the object is the first one
in the shadow chain and entry does not require COW. Charge is moved
from entry to object on allocation of the object, e.g. during the mmap,
assuming the object is allocated, or on the first page fault on the
entry. It moves back to the entry on forks due to COW setup.
The per-entry granularity of accounting makes the charge process fair
for processes that change uid during lifetime, and decrements charge
for proper uid when region is unmapped.
The interface of vm_pager_allocate(9) is extended by adding struct ucred *,
that is used to charge appropriate uid when allocation if performed by
kernel, e.g. md(4).
Several syscalls, among them is fork(2), may now return ENOMEM when
global or per-uid limits are enforced.
In collaboration with: pho
Reviewed by: alc
Approved by: re (kensmith)
memory with 4MB pages was added to pmap_object_init_pt(). This code
assumes that the pages of a OBJT_DEVICE object are always physically
contiguous. Unfortunately, this is not always the case. For example,
jhb@ informs me that the recently introduced /dev/ksyms driver creates
a OBJT_DEVICE object that violates this assumption. Thus, this
revision modifies pmap_object_init_pt() to abort the mapping if the
OBJT_DEVICE object's pages are not physically contiguous. This
revision also changes some inconsistent if not buggy behavior. For
example, the i386 version aborts if the first 4MB virtual page that
would be mapped is already valid. However, it incorrectly replaces
any subsequent 4MB virtual page mappings that it encounters,
potentially leaking a page table page. The amd64 version has a bug of
my own creation. It potentially busies the wrong page and always an
insufficent number of pages if it blocks allocating a page table page.
To my knowledge, there have been no reports of these bugs, hence,
their persistance. I suspect that the existing restrictions that
pmap_object_init_pt() placed on the OBJT_DEVICE objects that it would
choose to map, for example, that the first page must be aligned on a 2
or 4MB physical boundary and that the size of the mapping must be a
multiple of the large page size, were enough to avoid triggering the
bug for drivers like ksyms. However, one side effect of testing the
OBJT_DEVICE object's pages for physical contiguity is that a dubious
difference between pmap_object_init_pt() and the standard path for
mapping devices pages, i.e., vm_fault(), has been eliminated.
Previously, pmap_object_init_pt() would only instantiate the first
PG_FICTITOUS page being mapped because it never examined the rest.
Now, however, pmap_object_init_pt() uses the new function
vm_object_populate() to instantiate them all (in order to support
testing their physical contiguity). These pages need to be
instantiated for the mechanism that I have prototyped for
automatically maintaining the consistency of the PAT settings across
multiple mappings, particularly, amd64's direct mapping, to work.
(Translation: This change is also being made to support jhb@'s work on
the Nvidia feature requests.)
Discussed with: jhb@
vmopag" implementation. The vm_page_lookup() code modifies splay tree
of the object pages, and asserts that object lock is taken. First issue
could cause kernel data corruption, and second one instantly panics the
INVARIANTS-enabled kernel.
Take the advantage of the fact that object->memq is ordered by page index,
and iterate over memq to calculate the runs.
While there, make the code slightly more style-compliant by moving
variables declarations to the right place.
Discussed with: jhb, alc
Reviewed by: alc
MFC after: 2 weeks
In particular following functions KPI results modified:
- bufobj_invalbuf()
- bufsync()
and BO_SYNC() "virtual method" of the buffer objects set.
Main consumers of bufobj functions are affected by this change too and,
in particular, functions which changed their KPI are:
- vinvalbuf()
- g_vfs_close()
Due to the KPI breakage, __FreeBSD_version will be bumped in a later
commit.
As a side note, please consider just temporary the 'curthread' argument
passing to VOP_SYNC() (in bufsync()) as it will be axed out ASAP
Reviewed by: kib
Tested by: Giovanni Trematerra <giovanni dot trematerra at gmail dot com>
Directory IO without a VM object will store data in 'malloced' buffers
severely limiting caching of the data. Without this change VM objects for
directories are only created on an open() of the directory.
TODO: Inline test if VM object already exists to avoid locking/function call
overhead.
Tested by: kris@
Reviewed by: jeff@
Reported by: David Filo
vm_object_reference(). This is intended to get rid of vget()
consumers who don't wish to acquire a lock. This is functionally
the same as calling vref(). vm_object_reference_locked() already
uses vref.
Discussed with: alc
pmap_remove_all() must not be called on fictitious pages. To date,
fictitious pages have been allocated from zeroed memory, effectively
hiding this problem because the fictitious pages appear to have an empty
pv list. Submitted by: Kostik Belousov
Rewrite the comments describing vm_object_page_remove() to better
describe what it does. Add an assertion. Reviewed by: Kostik Belousov
MFC after: 1 week
only anonymous default (OBJT_DEFAULT) and swap (OBJT_SWAP) objects should
ever have OBJ_ONEMAPPING set. However, vm_object_deallocate() was
setting it on device (OBJT_DEVICE) objects. As a result,
vm_object_page_remove() could be called on a device object and if that
occurred pmap_remove_all() would be called on the device object's pages.
However, a device object's pages are fictitious, and fictitious pages do
not have an initialized pv list (struct md_page).
To date, fictitious pages have been allocated from zeroed memory,
effectively hiding this problem. Now, however, the conversion of rotting
diagnostics to invariants in the amd64 and i386 pmaps has revealed the
problem. Specifically, assertion failures have occurred during the
initialization phase of the X server on some hardware.
MFC after: 1 week
Discussed with: Kostik Belousov
Reported by: Michiel Boland
conjuction with 'thread' argument passing which is always curthread.
Remove the unuseful extra-argument and pass explicitly curthread to lower
layer functions, when necessary.
KPI results broken by this change, which should affect several ports, so
version bumping and manpage update will be further committed.
Tested by: kris, pho, Diego Sardina <siarodx at gmail dot com>
Remove this argument and pass curthread directly to underlying
VOP_LOCK1() VFS method. This modify makes the code cleaner and in
particular remove an annoying dependence helping next lockmgr() cleanup.
KPI results, obviously, changed.
Manpage and FreeBSD_version will be updated through further commits.
As a side note, would be valuable to say that next commits will address
a similar cleanup about VFS methods, in particular vop_lock1 and
vop_unlock.
Tested by: Diego Sardina <siarodx at gmail dot com>,
Andrea Di Pasquale <whyx dot it at gmail dot com>
machine-independent support for superpages. (The earlier part was
the rewrite of the physical memory allocator.) The remainder of the
code required for superpages support is machine-dependent and will
be added to the various pmap implementations at a later date.
Initially, I am only supporting one large page size per architecture.
Moreover, I am only enabling the reservation system on amd64. (In
an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0
in amd64/include/vmparam.h or your kernel configuration file.)
First, a file is mmap(2)ed and then mlock(2)ed. Later, it is truncated.
Under "normal" circumstances, i.e., when the file is not mlock(2)ed, the
pages beyond the EOF are unmapped and freed. However, when the file is
mlock(2)ed, the pages beyond the EOF are unmapped but not freed because
they have a non-zero wire count. This can be a mistake. Specifically,
it is a mistake if the sole reason why the pages are wired is because of
wired, managed mappings. Previously, unmapping the pages destroys these
wired, managed mappings, but does not reduce the pages' wire count.
Consequently, when the file is unmapped, the pages are not unwired
because the wired mapping has been destroyed. Moreover, when the vm
object is finally destroyed, the pages are leaked because they are still
wired. The fix is to reduce the pages' wired count by the number of
wired, managed mappings destroyed. To do this, I introduce a new pmap
function pmap_page_wired_mappings() that returns the number of managed
mappings to the given physical page that are wired, and I use this
function in vm_object_page_remove().
Reviewed by: tegge
MFC after: 6 weeks
cache: vm_object_page_remove() should convert any cached pages that
fall with the specified range to free pages. Otherwise, there could
be a problem if a file is first truncated and then regrown.
Specifically, some old data from prior to the truncation might reappear.
Generalize vm_page_cache_free() to support the conversion of either a
subset or the entirety of an object's cached pages.
Reported by: tegge
Reviewed by: tegge
Approved by: re (kensmith)
ways:
(1) Cached pages are no longer kept in the object's resident page
splay tree and memq. Instead, they are kept in a separate per-object
splay tree of cached pages. However, access to this new per-object
splay tree is synchronized by the _free_ page queues lock, not to be
confused with the heavily contended page queues lock. Consequently, a
cached page can be reclaimed by vm_page_alloc(9) without acquiring the
object's lock or the page queues lock.
This solves a problem independently reported by tegge@ and Isilon.
Specifically, they observed the page daemon consuming a great deal of
CPU time because of pages bouncing back and forth between the cache
queue (PQ_CACHE) and the inactive queue (PQ_INACTIVE). The source of
this problem turned out to be a deadlock avoidance strategy employed
when selecting a cached page to reclaim in vm_page_select_cache().
However, the root cause was really that reclaiming a cached page
required the acquisition of an object lock while the page queues lock
was already held. Thus, this change addresses the problem at its
root, by eliminating the need to acquire the object's lock.
Moreover, keeping cached pages in the object's primary splay tree and
memq was, in effect, optimizing for the uncommon case. Cached pages
are reclaimed far, far more often than they are reactivated. Instead,
this change makes reclamation cheaper, especially in terms of
synchronization overhead, and reactivation more expensive, because
reactivated pages will have to be reentered into the object's primary
splay tree and memq.
(2) Cached pages are now stored alongside free pages in the physical
memory allocator's buddy queues, increasing the likelihood that large
allocations of contiguous physical memory (i.e., superpages) will
succeed.
Finally, as a result of this change long-standing restrictions on when
and where a cached page can be reclaimed and returned by
vm_page_alloc(9) are eliminated. Specifically, calls to
vm_page_alloc(9) specifying VM_ALLOC_INTERRUPT can now reclaim and
return a formerly cached page. Consequently, a call to malloc(9)
specifying M_NOWAIT is less likely to fail.
Discussed with: many over the course of the summer, including jeff@,
Justin Husted @ Isilon, peter@, tegge@
Tested by: an earlier version by kris@
Approved by: re (kensmith)
This allocator uses a binary buddy system with a twist. First and
foremost, this allocator is required to support the implementation of
superpages. As a side effect, it enables a more robust implementation
of contigmalloc(9). Moreover, this reimplementation of
contigmalloc(9) eliminates the acquisition of Giant by
contigmalloc(..., M_NOWAIT, ...).
The twist is that this allocator tries to reduce the number of TLB
misses incurred by accesses through a direct map to small, UMA-managed
objects and page table pages. Roughly speaking, the physical pages
that are allocated for such purposes are clustered together in the
physical address space. The performance benefits vary. In the most
extreme case, a uniprocessor kernel running on an Opteron, I measured
an 18% reduction in system time during a buildworld.
This allocator does not implement page coloring. The reason is that
superpages have much the same effect. The contiguous physical memory
allocation necessary for a superpage is inherently colored.
Finally, the one caveat is that this allocator does not effectively
support prezeroed pages. I hope this is temporary. On i386, this is
a slight pessimization. However, on amd64, the beneficial effects of
the direct-map optimization outweigh the ill effects. I speculate
that this is true in general of machines with a direct map.
Approved by: re
In particular:
- Add an explicative table for locking of struct vmmeter members
- Apply new rules for some of those members
- Remove some unuseful comments
Heavily reviewed by: alc, bde, jeff
Approved by: jeff (mentor)
Now, we assume no more sched_lock protection for some of them and use the
distribuited loads method for vmmeter (distribuited through CPUs).
Reviewed by: alc, bde
Approved by: jeff (mentor)
Probabilly, a general approach is not the better solution here, so we should
solve the sched_lock protection problems separately.
Requested by: alc
Approved by: jeff (mentor)
vmcnts. This can be used to abstract away pcpu details but also changes
to use atomics for all counters now. This means sched lock is no longer
responsible for protecting counts in the switch routines.
Contributed by: Attilio Rao <attilio@FreeBSD.org>
causing a crash.
Suppose that we have two objects, obj and backing_obj, where
backing_obj is obj's backing object. Further, suppose that
backing_obj has a reference count of two. One being the reference
held by obj and the other by a map entry. Now, suppose that the map
entry is deallocated and its reference removed by
vm_object_deallocate(). vm_object_deallocate() recognizes that the
only remaining reference is from a shadow object, obj, and calls
vm_object_collapse() on obj. vm_object_collapse() executes
if (backing_object->ref_count == 1) {
/*
* If there is exactly one reference to the backing
* object, we can collapse it into the parent.
*/
vm_object_backing_scan(object, OBSC_COLLAPSE_WAIT);
vm_object_backing_scan(OBSC_COLLAPSE_WAIT) executes
if (op & OBSC_COLLAPSE_WAIT) {
vm_object_set_flag(backing_object, OBJ_DEAD);
}
Finally, suppose that either vm_object_backing_scan() or
vm_object_collapse() sleeps releasing its locks. At this instant,
another thread executes vm_object_split(). It crashes in
vm_object_reference_locked() on the assertion that the object is not
dead. If, however, assertions are not enabled, it crashes much later,
after the object has been recycled, in vm_object_deallocate() because
the shadow count and shadow list are inconsistent.
Reviewed by: tegge
Reported by: jhb
MFC after: 1 week
immediately flag any page that is allocated to a OBJT_PHYS object as
unmanaged in vm_page_alloc() rather than waiting for a later call to
vm_page_unmanage(). This allows for the elimination of some uses of
the page queues lock.
Change the type of the kernel and kmem objects from OBJT_DEFAULT to
OBJT_PHYS. This allows us to take advantage of the above change to
simplify the allocation of unmanaged pages in kmem_alloc() and
kmem_malloc().
Remove vm_page_unmanage(). It is no longer used.
page queues-synchronized flag. Reduce the scope of the page queues lock in
vm_fault() accordingly.
Move vm_fault()'s call to vm_object_set_writeable_dirty() outside of the
scope of the page queues lock. Reviewed by: tegge
Additionally, eliminate an unnecessary dereference in computing the
argument that is passed to vm_object_set_writeable_dirty().
synchronized by the lock on the object containing the page.
Transition PG_WANTED and PG_SWAPINPROG to use the new field,
eliminating the need for holding the page queues lock when setting
or clearing these flags. Rename PG_WANTED and PG_SWAPINPROG to
VPO_WANTED and VPO_SWAPINPROG, respectively.
Eliminate the assertion that the page queues lock is held in
vm_page_io_finish().
Eliminate the acquisition and release of the page queues lock
around calls to vm_page_io_finish() in kern_sendfile() and
vfs_unbusy_pages().
Originally, I had adopted sparc64's name, pmap_clear_write(), for the
function that is now pmap_remove_write(). However, this function is more
like pmap_remove_all() than like pmap_clear_modify() or
pmap_clear_reference(), hence, the name change.
The higher-level rationale behind this change is described in
src/sys/amd64/amd64/pmap.c revision 1.567. The short version is that I'm
trying to clean up and fix our support for execute access.
Reviewed by: marcel@ (ia64)
pointer: When vm_object_deallocate() sleeps because of a non-zero
paging in progress count on either object or object's shadow,
vm_object_deallocate() must ensure that object is still the shadow's
backing object when it reawakens. In fact, object may have been
deallocated while vm_object_deallocate() slept. If so, reacquiring
the lock on object can lead to a deadlock.
Submitted by: ups@
MFC after: 3 weeks
be called without any vnode locks held. Remove calls to vn_start_write() and
vn_finished_write() in vnode_pager_putpages() and add these calls before the
vnode lock is obtained to most of the callers that don't already have them.
object that requires Giant in vm_object_deallocate(). This is somewhat
hairy in that if we can't obtain Giant directly, we have to drop the
object lock, then lock Giant, then relock the object lock and verify that
we still need Giant. If we don't (because the object changed to OBJT_DEAD
for example), then we drop Giant before continuing.
Reviewed by: alc
Tested by: kris
the resident page count matches the object size. We know it fully backs
its parent in this case.
Reviewed by: acl, tegge
Sponsored by: Isilon Systems, Inc.
- provide an interface (macros) to the page coloring part of the VM system,
this allows to try different coloring algorithms without the need to
touch every file [1]
- make the page queue tuning values readable: sysctl vm.stats.pagequeue
- autotuning of the page coloring values based upon the cache size instead
of options in the kernel config (disabling of the page coloring as a
kernel option is still possible)
MD changes:
- detection of the cache size: only IA32 and AMD64 (untested) contains
cache size detection code, every other arch just comes with a dummy
function (this results in the use of default values like it was the
case without the autotuning of the page coloring)
- print some more info on Intel CPU's (like we do on AMD and Transmeta
CPU's)
Note to AMD owners (IA32 and AMD64): please run "sysctl vm.stats.pagequeue"
and report if the cache* values are zero (= bug in the cache detection code)
or not.
Based upon work by: Chad David <davidc@acns.ab.ca> [1]
Reviewed by: alc, arch (in 2004)
Discussed with: alc, Chad David, arch (in 2004)
vm_object_backing_scan() was not written to handle. Specifically, a wired
page within a backing object that is shadowed by a page within the shadow
object. Handle this state by removing the wired page from the backing
object. The wired page will be freed by socow_iodone().
Stop masking errors: If a page is being freed by vm_object_backing_scan(),
assert that it is no longer mapped rather than quietly destroying any
mappings.
Tested by: Harald Schmalzbauer
due to the vm object being locked.
When a process writes large amounts of data to a file, the vm object associated
with that file can contain most of the physical pages on the machine. If the
process is preempted while holding the lock on the vm object, pagedaemon would
be able to move very few pages from PQ_INACTIVE to PQ_CACHE or from PQ_ACTIVE
to PQ_INACTIVE, resulting in unlimited cleaning of dirty pages belonging to
other vm objects.
Temporarily unlock the page queues lock while locking vm objects to avoid lock
order violation. Detect and handle relevant page queue changes.
This change depends on both the lock portion of struct vm_object and normal
struct vm_page being type stable.
Reviewed by: alc
underlying vnode requires Giant.
- In vm_fault only acquire Giant if the underlying object has NEEDSGIANT
set.
- In vm_object_shadow inherit the NEEDSGIANT flag from the backing object.
queues lock in vm_object_backing_scan(). Updates to the page's PG_BUSY
flag and busy field are synchronized by the containing object's lock.
Testing the page's hold_count and wire_count in vm_object_backing_scan()'s
OBSC_COLLAPSE_NOWAIT case is unnecessary. There is no reason why the held
or wired pages cannot be migrated to the shadow object.
Reviewed by: tegge
- Use VFS_LOCK_GIANT() rather than directly acquiring giant in places
where giant is only held because vfs requires it.
Sponsored By: Isilon Systems, Inc.
and BBO is BO's backing object. Now, suppose that O and BO are being
collapsed. Furthermore, suppose that BO has been marked dead
(OBJ_DEAD) by vm_object_backing_scan() and that either
vm_object_backing_scan() has been forced to sleep due to encountering
a busy page or vm_object_collapse() has been forced to sleep due to
memory allocation in the swap pager. If vm_object_deallocate() is
then called on BBO and BO is BBO's only shadow object,
vm_object_deallocate() will collapse BO and BBO. In doing so, it adds
a necessary temporary reference to BO. If this collapse also sleeps
and the prior collapse resumes first, the temporary reference will
cause vm_object_collapse to panic with the message "backing_object %p
was somehow re-referenced during collapse!"
Resolve this race by changing vm_object_deallocate() such that it
doesn't collapse BO and BBO if BO is marked dead. Once O and BO are
collapsed, vm_object_collapse() will attempt to collapse O and BBO.
So, vm_object_deallocate() on BBO need do nothing.
Reported by: Peter Holm on 20050107
URL: http://www.holm.cc/stress/log/cons102.html
In collaboration with: tegge@
Candidate for RELENG_4 and RELENG_5
MFC after: 2 weeks
I'm not sure why a credential was added to these in the first place, it is
not used anywhere and it doesn't make much sense:
The credentials for syncing a file (ability to write to the
file) should be checked at the system call level.
Credentials for syncing one or more filesystems ("none")
should be checked at the system call level as well.
If the filesystem implementation needs a particular credential
to carry out the syncing it would logically have to the
cached mount credential, or a credential cached along with
any delayed write data.
Discussed with: rwatson
because this call is only needed to wake threads that slept when they
discovered a dead object connected to a vnode. To eliminate unnecessary
calls to wakeup() by vnode_pager_dealloc(), introduce a new flag,
OBJ_DISCONNECTWNT.
Reviewed by: tegge@
need for most calls to vm_page_busy(). Specifically, most calls to
vm_page_busy() occur immediately prior to a call to vm_page_remove().
In such cases, the containing vm object is locked across both calls.
Consequently, the setting of the vm page's PG_BUSY flag is not even
visible to other threads that are following the synchronization
protocol.
This change (1) eliminates the calls to vm_page_busy() that
immediately precede a call to vm_page_remove() or functions, such as
vm_page_free() and vm_page_rename(), that call it and (2) relaxes the
requirement in vm_page_remove() that the vm page's PG_BUSY flag is
set. Now, the vm page's PG_BUSY flag is set only when the vm object
lock is released while the vm page is still in transition. Typically,
this is when it is undergoing I/O.
so that they know whether the allocation is supposed to be able to sleep
or not.
* Allow uma_zone constructors and initialation functions to return either
success or error. Almost all of the ones in the tree currently return
success unconditionally, but mbuf is a notable exception: the packet
zone constructor wants to be able to fail if it cannot suballocate an
mbuf cluster, and the mbuf allocators want to be able to fail in general
in a MAC kernel if the MAC mbuf initializer fails. This fixes the
panics people are seeing when they run out of memory for mbuf clusters.
* Allow debug.nosleepwithlocks on WITNESS to be disabled, without changing
the default.
Both bmilekic and jeff have reviewed the changes made to make failable
zone allocations work.
vm/vm_object.c revision 1.88) and vm_object_sync() (originating in
vm/vm_map.c revision 1.36): When descending a chain of backing objects,
both use the wrong object's backing offset. Consequently, both may
operate on the wrong pages.
Quoting Matt, "This could be responsible for all of the sporatic madvise
oddness that has been reported over the years."
Reviewed by: Matt Dillon
kmem_alloc_pageable(). The difference between these is that an errant
memory access to the zone will be detected sooner with
kmem_alloc_nofault().
The following changes serve to eliminate the following lock-order
reversal reported by witness:
1st 0xc1a3c084 vm object (vm object) @ vm/swap_pager.c:1311
2nd 0xc07acb00 swap_pager swhash (swap_pager swhash) @ vm/swap_pager.c:1797
3rd 0xc1804bdc vm object (vm object) @ vm/uma_core.c:931
There is no potential deadlock in this case. However, witness is unable
to recognize this because vm objects used by UMA have the same type as
ordinary vm objects. To remedy this, we make the following changes:
- Add a mutex type argument to VM_OBJECT_LOCK_INIT().
- Use the mutex type argument to assign distinct types to special
vm objects such as the kernel object, kmem object, and UMA objects.
- Define a static swap zone object for use by UMA. (Only static
objects are assigned a special mutex type.)
vm objects shadowing source in vm_object_shadow(). This closes a race where
vm_object_collapse() could be called with a partially uninitialized object
argument causing symptoms that looked like hardware problems, e.g. signal 6,
10, 11 or a /bin/sh busy-waiting for a nonexistant child process.
that msync(2) is its only caller.
- Migrate the parts of the old vm_map_clean() that examined the internals
of a vm object to a new function vm_object_sync() that is implemented in
vm_object.c. At the same, introduce the necessary vm object locking so
that vm_map_sync() and vm_object_sync() can be called without Giant.
Reviewed by: tegge
