There are two aspects to the sequential access optimization: (1) read ahead
of pages that are expected to be accessed in the near future and (2) unmap
and cache behind of pages that are not expected to be accessed again. This
revision changes both aspects.
The read ahead optimization is now more effective. It starts with the same
initial read window as before, but arithmetically grows the window on
sequential page faults. This can yield increased read bandwidth. For
example, on one of my machines, a program using mmap() to read a file that
is several times larger than the machine's physical memory takes about 17%
less time to complete.
The unmap and cache behind optimization is now more selectively applied.
The read ahead window must grow to its maximum size before unmap and cache
behind is performed. This significantly reduces the number of times that
pages are unmapped and cached only to be reactivated a short time later.
The unmap and cache behind optimization now clears each page's referenced
flag. Previously, in the case of dirty pages, if the containing file was
still mapped at the time that the page daemon examined the dirty pages,
they would be reactivated.
From a stylistic standpoint, this revision also cleanly separates the
implementation of the read ahead and unmap/cache behind optimizations.
Glanced at: kib
MFC after: 2 weeks
v_writecount. Keep the amount of the virtual address space used by
the mappings in the new vm_object un_pager.vnp.writemappings
counter. The vnode v_writecount is incremented when writemappings gets
non-zero value, and decremented when writemappings is returned to
zero.
Writeable shared vnode-backed mappings are accounted for in vm_mmap(),
and vm_map_insert() is instructed to set MAP_ENTRY_VN_WRITECNT flag on
the created map entry. During deferred map entry deallocation,
vm_map_process_deferred() checks for MAP_ENTRY_VN_WRITECOUNT and
decrements writemappings for the vm object.
Now, the writeable mount cannot be demoted to read-only while
writeable shared mappings of the vnodes from the mount point
exist. Also, execve(2) fails for such files with ETXTBUSY, as it
should be.
Noted by: tegge
Reviewed by: tegge (long time ago, early version), alc
Tested by: pho
MFC after: 3 weeks
for a shared mapping and marking the entry for inheritance.
Other thread might execute vmspace_fork() in between (e.g. by fork(2)),
resulting in the mapping becoming private.
Noted and reviewed by: alc
MFC after: 1 week
condition in proc_rwmem() and to (2) simplify the implementation of the
cxgb driver's vm_fault_hold_user_pages(). Specifically, in proc_rwmem()
the requested read or write could fail because the targeted page could be
reclaimed between the calls to vm_fault() and vm_page_hold().
In collaboration with: kib@
MFC after: 6 weeks
vmspace_fork and vm_map_wire that would lead to "vm_fault_copy_wired: page
missing" panics. While faulting in pages for a map entry that is being
wired down, mark the containing map as busy. In vmspace_fork wait until the
map is unbusy, before we try to copy the entries.
Reviewed by: kib
MFC after: 5 days
Sponsored by: Isilon Systems, Inc.
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
its value as a loop invariant. Currently this is a no-op because
'atomic_cmpset_int()' clobbers all memory on current architectures.
- Use atomic_fetchadd_int() instead of an atomic_cmpset_int() loop to drop
a reference in vmspace_free().
Reviewed by: alc
MFC after: 1 month
vm_map_unlock_nodefer() part of the synchronization interface for maps.
Add comments to vm_map_unlock_and_wait() and vm_map_wakeup() describing
how they should be used. In particular, describe the deferred deallocations
issue with vm_map_unlock_and_wait().
Redo the implementation of vm_map_unlock_and_wait() so that it passes
along the caller's file and line information, just like the other map
locking primitives.
Reviewed by: kib
X-MFC after: r212824
on map unlock to the lock downgrade and later read unlock operation.
System map entries cannot be backed by OBJT_VNODE objects, no need to
defer deallocation for them. Map entries from user maps do not require
the owner map for deallocation, and can be accumulated in the
thread-local list for freeing when a user map is unlocked.
Move the collection of entries for deferred reclamation into
vm_map_delete(). Create helper vm_map_process_deferred(), that is
called from locations where processing is feasible. Do not process
deferred entries in vm_map_unlock_and_wait() since map_sleep_mtx is
held.
Reviewed by: alc, rstone (previous versions)
Tested by: pho
MFC after: 2 weeks
address space for an address as aligned by the new pmap_align_tlb()
function, which is for constraints imposed by the TLB. [1]
o) Add a kmem_alloc_nofault_space() function, which acts like
kmem_alloc_nofault() but allows the caller to specify which find-space
option to use. [1]
o) Use kmem_alloc_nofault_space() with VMFS_TLB_ALIGNED_SPACE to allocate the
kernel stack address on MIPS. [1]
o) Make pmap_align_tlb() on MIPS align addresses so that they do not start on
an odd boundary within the TLB, so that they are suitable for insertion as
wired entries and do not have to share a TLB entry with another mapping,
assuming they are appropriately-sized.
o) Eliminate md_realstack now that the kstack will be appropriately-aligned on
MIPS.
o) Increase the number of guard pages to 2 so that we retain the proper
alignment of the kstack address.
Reviewed by: [1] alc
X-MFC-after: Making sure alc has not come up with a better interface.
pages.
(Note: Claims made in the comments about the handling of breakpoints in
wired pages have been false for roughly a decade. This and another bug
involving breakpoints will be fixed in coming changes.)
Reviewed by: kib
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)
the mappings without any of read and execution rights, in particular,
the PROT_NONE entries. This makes mlockall(2) work for the process
address space that has such mappings.
Since protection mode of the entry may change between setting
MAP_ENTRY_IN_TRANSITION and final pass over the region that records
the wire status of the entries, allocate new map entry flag
MAP_ENTRY_WIRE_SKIPPED to mark the skipped PROT_NONE entries.
Reported and tested by: Hans Ottevanger <fbsdhackers beasties demon nl>
Reviewed by: alc
MFC after: 3 weeks
function, done in r188334. Instead, collect the entries that shall be
freed, in the deferred_freelist member of the map. Automatically purge
the deferred freelist when map is unlocked.
Tested by: pho
Reviewed by: alc
hold the map lock there, and might need the vnode lock for OBJT_VNODE
objects. Postpone object deallocation until caller of vm_map_delete()
drops the map lock. Link the map entries to be freed into the freelist,
that is released by the new helper function vm_map_entry_free_freelist().
Reviewed by: tegge, alc
Tested by: pho
support for VMFS_ALIGNED_SPACE, which requests the allocation of an
address range best suited to superpages. The old options TRUE and FALSE
are mapped to VMFS_ANY_SPACE and VMFS_NO_SPACE, so that there is no
immediate need to update all of vm_map_find(9)'s callers.
While I'm here, correct a misstatement about vm_map_find(9)'s return
values in the man page.
structure. This allows per-CPU variations of struct pmap on a
single architecture without affecting the machine-independent
fields. As such, the PMAP variations don't affect the ABI. They
become part of it.
For this, introduce vm_map_fixed() that does that for MAP_FIXED case.
Dropping the lock allowed for parallel thread to occupy the freed space.
Reported by: Tijl Coosemans <tijl ulyssis org>
Reviewed by: alc
Approved by: re (kensmith)
MFC after: 2 weeks
vmspace_exitfree() and vmspace_free() which could result in the same
vmspace being freed twice.
Factor out part of exit1() into new function vmspace_exit(). Attach
to vmspace0 to allow old vmspace to be freed earlier.
Add new function, vmspace_acquire_ref(), for obtaining a vmspace
reference for a vmspace belonging to another process. Avoid changing
vmspace refcount from 0 to 1 since that could also lead to the same
vmspace being freed twice.
Change vmtotal() and swapout_procs() to use vmspace_acquire_ref().
Reviewed by: alc
algorithm built into the map entry splay tree. This replaces the
first_free hint in struct vm_map with two fields in vm_map_entry:
adj_free, the amount of free space following a map entry, and
max_free, the maximum amount of free space in the entry's subtree.
These fields make it possible to find a first-fit free region of a
given size in one pass down the tree, so O(log n) amortized using
splay trees.
This significantly reduces the overhead in vm_map_findspace() for
applications that mmap() many hundreds or thousands of regions, and
has a negligible slowdown (0.1%) on buildworld. See, for example, the
discussion of a micro-benchmark titled "Some mmap observations
compared to Linux 2.6/OpenBSD" on -hackers in late October 2003.
OpenBSD adopted this approach in March 2002, and NetBSD added it in
November 2003, both with Red-Black trees.
Submitted by: Mark W. Krentel
to avoid later changes before pmap_enter() and vm_fault_prefault()
has completed.
Simplify deadlock avoidance by not blocking on vm map relookup.
In collaboration with: alc
* Allow no-fault wiring/unwiring to succeed for consistency;
however, the wired count remains at zero, so it's a special case.
* Fix issues inside vm_map_wire() and vm_map_unwire() where the
exact state of user wiring (one or zero) and system wiring
(zero or more) could be confused; for example, system unwiring
could succeed in removing a user wire, instead of being an
error.
* Require all mappings to be unwired before they are deleted.
When VM space is still wired upon deletion, it will be waited
upon for the following unwire. This makes vslock(9) work
rather than allowing kernel-locked memory to be deleted
out from underneath of its consumer as it would before.
maps. We always acquire the sx lock exclusively here, but we can't
use a mutex because we want to be able to sleep while holding the
lock. This is completely equivalent to what we were doing with the
lockmgr(9) locks before.
Approved by: alc
vmspace to the new vmspace in vmspace_exec() is mostly wasted effort. With
one exception, vm_swrss, the copied fields are immediately overwritten.
Instead, initialize these fields to zero in vmspace_alloc(), eliminating a
bcopy() from vmspace_exec() and a bzero() from vmspace_fork().
would actually map the file with read access enabled. According to
http://www.opengroup.org/onlinepubs/007904975/functions/mmap.html this is
an error. Similarly, an madvise(..., MADV_WILLNEED) would enable read
access on a virtual address range that was PROT_NONE.
The solution implemented herein is (1) to pass a vm_prot_t to
vm_map_pmap_enter() describing the allowed access and (2) to make
vm_map_pmap_enter() responsible for understanding the limitations of
pmap_enter_quick().
Submitted by: "Mark W. Krentel" <krentel@dreamscape.com>
PR: kern/64573
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
use the ability on ia64 to map the register stack. The orientation of
the stack (i.e. its grow direction) is passed to vm_map_stack() in the
overloaded cow argument. Since the grow direction is represented by
bits, it is possible and allowed to create bi-directional stacks.
This is not an advertised feature, more of a side-effect.
Fix a bug in vm_map_growstack() that's specific to rstacks and which
we could only find by having the ability to create rstacks: when
the mapped stack ends at the faulting address, we have not actually
mapped the faulting address. we need to include or cover the faulting
address.
Note that at this time mmap(2) has not been extended to allow the
creation of rstacks by processes. If such a need arises, this can
be done.
Tested on: alpha, i386, ia64, sparc64
growable (stack) entries that not only grow down, but also grow up.
Have vm_map_growstack() take these flags into account when growing
an entry.
This is the first step in adding support for upward growable stacks.
It is a required feature on ia64 to support the register stack (or
rstack as I like to call it -- it also means reverse stack). We do
not currently create rstacks, so the upward growing is not exercised
and the change should be a functional no-op.
Reviewed by: alc
- All those diffs to syscalls.master for each architecture *are*
necessary. This needed clarification; the stub code generation for
mlockall() was disabled, which would prevent applications from
linking to this API (suggested by mux)
- Giant has been quoshed. It is no longer held by the code, as
the required locking has been pushed down within vm_map.c.
- Callers must specify VM_MAP_WIRE_HOLESOK or VM_MAP_WIRE_NOHOLES
to express their intention explicitly.
- Inspected at the vmstat, top and vm pager sysctl stats level.
Paging-in activity is occurring correctly, using a test harness.
- The RES size for a process may appear to be greater than its SIZE.
This is believed to be due to mappings of the same shared library
page being wired twice. Further exploration is needed.
- Believed to back out of allocations and locks correctly
(tested with WITNESS, MUTEX_PROFILING, INVARIANTS and DIAGNOSTIC).
PR: kern/43426, standards/54223
Reviewed by: jake, alc
Approved by: jake (mentor)
MFC after: 2 weeks
process to kill, don't block on a map lock while holding the
process lock. Instead, skip processes whose map locks are held
and find something else to kill.
- Add vm_map_trylock_read() to support the above.
Reviewed by: alc, mike (mentor)
