2005-01-07 02:29:27 +00:00
|
|
|
/*-
|
2001-07-04 23:27:09 +00:00
|
|
|
* Copyright (c) 1991 Regents of the University of California.
|
|
|
|
|
* All rights reserved.
|
|
|
|
|
*
|
|
|
|
|
* This code is derived from software contributed to Berkeley by
|
|
|
|
|
* The Mach Operating System project at Carnegie-Mellon University.
|
|
|
|
|
*
|
|
|
|
|
* Redistribution and use in source and binary forms, with or without
|
|
|
|
|
* modification, are permitted provided that the following conditions
|
|
|
|
|
* are met:
|
|
|
|
|
* 1. Redistributions of source code must retain the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer.
|
|
|
|
|
* 2. Redistributions in binary form must reproduce the above copyright
|
|
|
|
|
* notice, this list of conditions and the following disclaimer in the
|
|
|
|
|
* documentation and/or other materials provided with the distribution.
|
|
|
|
|
* 4. Neither the name of the University nor the names of its contributors
|
|
|
|
|
* may be used to endorse or promote products derived from this software
|
|
|
|
|
* without specific prior written permission.
|
|
|
|
|
*
|
|
|
|
|
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
|
|
|
|
|
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
|
|
|
|
|
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
|
|
|
|
|
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
|
|
|
|
|
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
|
|
|
|
|
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
|
|
|
|
|
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
|
|
|
|
|
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
|
|
|
|
|
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
|
|
|
|
|
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
|
|
|
|
|
* SUCH DAMAGE.
|
|
|
|
|
*
|
|
|
|
|
* from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
|
|
|
|
|
*/
|
|
|
|
|
|
2005-01-07 02:29:27 +00:00
|
|
|
/*-
|
2001-07-04 23:27:09 +00:00
|
|
|
* Copyright (c) 1987, 1990 Carnegie-Mellon University.
|
|
|
|
|
* All rights reserved.
|
|
|
|
|
*
|
|
|
|
|
* Authors: Avadis Tevanian, Jr., Michael Wayne Young
|
|
|
|
|
*
|
|
|
|
|
* Permission to use, copy, modify and distribute this software and
|
|
|
|
|
* its documentation is hereby granted, provided that both the copyright
|
|
|
|
|
* notice and this permission notice appear in all copies of the
|
|
|
|
|
* software, derivative works or modified versions, and any portions
|
|
|
|
|
* thereof, and that both notices appear in supporting documentation.
|
|
|
|
|
*
|
|
|
|
|
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
|
|
|
|
|
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
|
|
|
|
|
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
|
|
|
|
|
*
|
|
|
|
|
* Carnegie Mellon requests users of this software to return to
|
|
|
|
|
*
|
|
|
|
|
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
|
|
|
|
|
* School of Computer Science
|
|
|
|
|
* Carnegie Mellon University
|
|
|
|
|
* Pittsburgh PA 15213-3890
|
|
|
|
|
*
|
|
|
|
|
* any improvements or extensions that they make and grant Carnegie the
|
|
|
|
|
* rights to redistribute these changes.
|
|
|
|
|
*/
|
|
|
|
|
|
2003-06-11 23:50:51 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
|
2001-07-04 23:27:09 +00:00
|
|
|
#include <sys/param.h>
|
|
|
|
|
#include <sys/systm.h>
|
|
|
|
|
#include <sys/lock.h>
|
|
|
|
|
#include <sys/malloc.h>
|
|
|
|
|
#include <sys/mutex.h>
|
|
|
|
|
#include <sys/proc.h>
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
#include <sys/kernel.h>
|
|
|
|
|
#include <sys/linker_set.h>
|
|
|
|
|
#include <sys/sysctl.h>
|
2001-07-04 23:27:09 +00:00
|
|
|
#include <sys/vmmeter.h>
|
|
|
|
|
#include <sys/vnode.h>
|
|
|
|
|
|
|
|
|
|
#include <vm/vm.h>
|
|
|
|
|
#include <vm/vm_param.h>
|
|
|
|
|
#include <vm/vm_kern.h>
|
2001-10-13 04:23:37 +00:00
|
|
|
#include <vm/pmap.h>
|
|
|
|
|
#include <vm/vm_map.h>
|
2001-07-04 23:27:09 +00:00
|
|
|
#include <vm/vm_object.h>
|
|
|
|
|
#include <vm/vm_page.h>
|
|
|
|
|
#include <vm/vm_pageout.h>
|
|
|
|
|
#include <vm/vm_pager.h>
|
|
|
|
|
#include <vm/vm_extern.h>
|
|
|
|
|
|
2002-07-15 06:33:31 +00:00
|
|
|
static int
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
vm_contig_launder_page(vm_page_t m)
|
2002-07-15 06:33:31 +00:00
|
|
|
{
|
|
|
|
|
vm_object_t object;
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
vm_page_t m_tmp;
|
2003-10-18 21:09:21 +00:00
|
|
|
struct vnode *vp;
|
2002-07-15 06:33:31 +00:00
|
|
|
|
2004-10-24 19:32:19 +00:00
|
|
|
object = m->object;
|
|
|
|
|
if (!VM_OBJECT_TRYLOCK(object))
|
|
|
|
|
return (EAGAIN);
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
if (vm_page_sleep_if_busy(m, TRUE, "vpctw0")) {
|
2004-10-24 19:32:19 +00:00
|
|
|
VM_OBJECT_UNLOCK(object);
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
vm_page_lock_queues();
|
|
|
|
|
return (EBUSY);
|
|
|
|
|
}
|
|
|
|
|
vm_page_test_dirty(m);
|
|
|
|
|
if (m->dirty == 0 && m->hold_count == 0)
|
|
|
|
|
pmap_remove_all(m);
|
|
|
|
|
if (m->dirty) {
|
|
|
|
|
if (object->type == OBJT_VNODE) {
|
|
|
|
|
vm_page_unlock_queues();
|
|
|
|
|
vp = object->handle;
|
|
|
|
|
VM_OBJECT_UNLOCK(object);
|
|
|
|
|
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, curthread);
|
|
|
|
|
VM_OBJECT_LOCK(object);
|
|
|
|
|
vm_object_page_clean(object, 0, 0, OBJPC_SYNC);
|
|
|
|
|
VM_OBJECT_UNLOCK(object);
|
|
|
|
|
VOP_UNLOCK(vp, 0, curthread);
|
|
|
|
|
vm_page_lock_queues();
|
|
|
|
|
return (0);
|
|
|
|
|
} else if (object->type == OBJT_SWAP ||
|
|
|
|
|
object->type == OBJT_DEFAULT) {
|
|
|
|
|
m_tmp = m;
|
|
|
|
|
vm_pageout_flush(&m_tmp, 1, VM_PAGER_PUT_SYNC);
|
|
|
|
|
VM_OBJECT_UNLOCK(object);
|
|
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
} else if (m->hold_count == 0)
|
|
|
|
|
vm_page_cache(m);
|
2004-10-24 19:32:19 +00:00
|
|
|
VM_OBJECT_UNLOCK(object);
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
return (0);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
vm_contig_launder(int queue)
|
|
|
|
|
{
|
|
|
|
|
vm_page_t m, next;
|
|
|
|
|
int error;
|
|
|
|
|
|
2002-07-15 06:33:31 +00:00
|
|
|
for (m = TAILQ_FIRST(&vm_page_queues[queue].pl); m != NULL; m = next) {
|
|
|
|
|
next = TAILQ_NEXT(m, pageq);
|
|
|
|
|
KASSERT(m->queue == queue,
|
|
|
|
|
("vm_contig_launder: page %p's queue is not %d", m, queue));
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
error = vm_contig_launder_page(m);
|
|
|
|
|
if (error == 0)
|
2002-07-15 06:33:31 +00:00
|
|
|
return (TRUE);
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
if (error == EBUSY)
|
|
|
|
|
return (FALSE);
|
2002-07-15 06:33:31 +00:00
|
|
|
}
|
|
|
|
|
return (FALSE);
|
|
|
|
|
}
|
|
|
|
|
|
2001-07-04 23:27:09 +00:00
|
|
|
/*
|
|
|
|
|
* This interface is for merging with malloc() someday.
|
|
|
|
|
* Even if we never implement compaction so that contiguous allocation
|
|
|
|
|
* works after initialization time, malloc()'s data structures are good
|
|
|
|
|
* for statistics and for allocations of less than a page.
|
|
|
|
|
*/
|
2002-05-22 01:01:37 +00:00
|
|
|
static void *
|
2001-07-04 23:27:09 +00:00
|
|
|
contigmalloc1(
|
|
|
|
|
unsigned long size, /* should be size_t here and for malloc() */
|
|
|
|
|
struct malloc_type *type,
|
|
|
|
|
int flags,
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t low,
|
|
|
|
|
vm_paddr_t high,
|
2001-07-04 23:27:09 +00:00
|
|
|
unsigned long alignment,
|
|
|
|
|
unsigned long boundary,
|
|
|
|
|
vm_map_t map)
|
|
|
|
|
{
|
2004-01-10 21:17:53 +00:00
|
|
|
int i, start;
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t phys;
|
2003-07-05 20:51:22 +00:00
|
|
|
vm_object_t object;
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_offset_t addr, tmp_addr;
|
2003-07-02 01:47:47 +00:00
|
|
|
int pass, pqtype;
|
Make contigmalloc() more reliable:
1. Remove a race whereby contigmalloc() would deadlock against the
running processes in the system if they kept reinstantiating
the memory on the active and inactive page queues that it was
trying to flush out. The process doing the contigmalloc() would
sit in "swwrt" forever and the swap pager would be going at full
force, but never get anywhere. Instead of doing it until the
queues are empty, launder for as many iterations as there are
pages in the queue.
2. Do all laundering to swap synchronously; previously, the vnode
laundering was synchronous and the swap laundering not.
3. Increase the number of launder-or-allocate passes to three, from
two, while failing without bothering to do all the laundering on
the third pass if allocation was not possible. This effectively
gives exactly two chances to launder enough contiguous memory,
helpful with high memory churn where a lot of memory from one pass
to the next (and during a single laundering loop) becomes dirtied
again.
I can now reliably hot-plug hardware requiring a 256KB contigmalloc()
without having the kldload/cbb ithread sit around failing to make
progress, while running a busy X session. Previously, it took killing
X to get contigmalloc() to get further (that is, quiescing the system),
and even then contigmalloc() returned failure.
2004-06-15 01:02:00 +00:00
|
|
|
int inactl, actl, inactmax, actmax;
|
2001-07-04 23:27:09 +00:00
|
|
|
vm_page_t pga = vm_page_array;
|
|
|
|
|
|
|
|
|
|
size = round_page(size);
|
|
|
|
|
if (size == 0)
|
|
|
|
|
panic("contigmalloc1: size must not be 0");
|
|
|
|
|
if ((alignment & (alignment - 1)) != 0)
|
|
|
|
|
panic("contigmalloc1: alignment must be a power of 2");
|
|
|
|
|
if ((boundary & (boundary - 1)) != 0)
|
|
|
|
|
panic("contigmalloc1: boundary must be a power of 2");
|
|
|
|
|
|
|
|
|
|
start = 0;
|
Make contigmalloc() more reliable:
1. Remove a race whereby contigmalloc() would deadlock against the
running processes in the system if they kept reinstantiating
the memory on the active and inactive page queues that it was
trying to flush out. The process doing the contigmalloc() would
sit in "swwrt" forever and the swap pager would be going at full
force, but never get anywhere. Instead of doing it until the
queues are empty, launder for as many iterations as there are
pages in the queue.
2. Do all laundering to swap synchronously; previously, the vnode
laundering was synchronous and the swap laundering not.
3. Increase the number of launder-or-allocate passes to three, from
two, while failing without bothering to do all the laundering on
the third pass if allocation was not possible. This effectively
gives exactly two chances to launder enough contiguous memory,
helpful with high memory churn where a lot of memory from one pass
to the next (and during a single laundering loop) becomes dirtied
again.
I can now reliably hot-plug hardware requiring a 256KB contigmalloc()
without having the kldload/cbb ithread sit around failing to make
progress, while running a busy X session. Previously, it took killing
X to get contigmalloc() to get further (that is, quiescing the system),
and even then contigmalloc() returned failure.
2004-06-15 01:02:00 +00:00
|
|
|
for (pass = 2; pass >= 0; pass--) {
|
2002-08-04 07:07:34 +00:00
|
|
|
vm_page_lock_queues();
|
2004-03-02 08:25:58 +00:00
|
|
|
again0:
|
2004-01-08 20:48:26 +00:00
|
|
|
mtx_lock_spin(&vm_page_queue_free_mtx);
|
2001-07-04 23:27:09 +00:00
|
|
|
again:
|
|
|
|
|
/*
|
2003-07-02 01:47:47 +00:00
|
|
|
* Find first page in array that is free, within range,
|
|
|
|
|
* aligned, and such that the boundary won't be crossed.
|
2001-07-04 23:27:09 +00:00
|
|
|
*/
|
|
|
|
|
for (i = start; i < cnt.v_page_count; i++) {
|
|
|
|
|
phys = VM_PAGE_TO_PHYS(&pga[i]);
|
|
|
|
|
pqtype = pga[i].queue - pga[i].pc;
|
|
|
|
|
if (((pqtype == PQ_FREE) || (pqtype == PQ_CACHE)) &&
|
|
|
|
|
(phys >= low) && (phys < high) &&
|
|
|
|
|
((phys & (alignment - 1)) == 0) &&
|
|
|
|
|
(((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0))
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* If the above failed or we will exceed the upper bound, fail.
|
|
|
|
|
*/
|
|
|
|
|
if ((i == cnt.v_page_count) ||
|
|
|
|
|
((VM_PAGE_TO_PHYS(&pga[i]) + size) > high)) {
|
2004-01-08 20:48:26 +00:00
|
|
|
mtx_unlock_spin(&vm_page_queue_free_mtx);
|
Make contigmalloc() more reliable:
1. Remove a race whereby contigmalloc() would deadlock against the
running processes in the system if they kept reinstantiating
the memory on the active and inactive page queues that it was
trying to flush out. The process doing the contigmalloc() would
sit in "swwrt" forever and the swap pager would be going at full
force, but never get anywhere. Instead of doing it until the
queues are empty, launder for as many iterations as there are
pages in the queue.
2. Do all laundering to swap synchronously; previously, the vnode
laundering was synchronous and the swap laundering not.
3. Increase the number of launder-or-allocate passes to three, from
two, while failing without bothering to do all the laundering on
the third pass if allocation was not possible. This effectively
gives exactly two chances to launder enough contiguous memory,
helpful with high memory churn where a lot of memory from one pass
to the next (and during a single laundering loop) becomes dirtied
again.
I can now reliably hot-plug hardware requiring a 256KB contigmalloc()
without having the kldload/cbb ithread sit around failing to make
progress, while running a busy X session. Previously, it took killing
X to get contigmalloc() to get further (that is, quiescing the system),
and even then contigmalloc() returned failure.
2004-06-15 01:02:00 +00:00
|
|
|
/*
|
|
|
|
|
* Instead of racing to empty the inactive/active
|
|
|
|
|
* queues, give up, even with more left to free,
|
|
|
|
|
* if we try more than the initial amount of pages.
|
|
|
|
|
*
|
|
|
|
|
* There's no point attempting this on the last pass.
|
|
|
|
|
*/
|
|
|
|
|
if (pass > 0) {
|
|
|
|
|
inactl = actl = 0;
|
|
|
|
|
inactmax = vm_page_queues[PQ_INACTIVE].lcnt;
|
|
|
|
|
actmax = vm_page_queues[PQ_ACTIVE].lcnt;
|
2001-07-04 23:27:09 +00:00
|
|
|
again1:
|
Make contigmalloc() more reliable:
1. Remove a race whereby contigmalloc() would deadlock against the
running processes in the system if they kept reinstantiating
the memory on the active and inactive page queues that it was
trying to flush out. The process doing the contigmalloc() would
sit in "swwrt" forever and the swap pager would be going at full
force, but never get anywhere. Instead of doing it until the
queues are empty, launder for as many iterations as there are
pages in the queue.
2. Do all laundering to swap synchronously; previously, the vnode
laundering was synchronous and the swap laundering not.
3. Increase the number of launder-or-allocate passes to three, from
two, while failing without bothering to do all the laundering on
the third pass if allocation was not possible. This effectively
gives exactly two chances to launder enough contiguous memory,
helpful with high memory churn where a lot of memory from one pass
to the next (and during a single laundering loop) becomes dirtied
again.
I can now reliably hot-plug hardware requiring a 256KB contigmalloc()
without having the kldload/cbb ithread sit around failing to make
progress, while running a busy X session. Previously, it took killing
X to get contigmalloc() to get further (that is, quiescing the system),
and even then contigmalloc() returned failure.
2004-06-15 01:02:00 +00:00
|
|
|
if (inactl < inactmax &&
|
|
|
|
|
vm_contig_launder(PQ_INACTIVE)) {
|
|
|
|
|
inactl++;
|
|
|
|
|
goto again1;
|
|
|
|
|
}
|
|
|
|
|
if (actl < actmax &&
|
|
|
|
|
vm_contig_launder(PQ_ACTIVE)) {
|
|
|
|
|
actl++;
|
|
|
|
|
goto again1;
|
|
|
|
|
}
|
|
|
|
|
}
|
2002-08-04 07:07:34 +00:00
|
|
|
vm_page_unlock_queues();
|
2001-07-04 23:27:09 +00:00
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
start = i;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Check successive pages for contiguous and free.
|
|
|
|
|
*/
|
|
|
|
|
for (i = start + 1; i < (start + size / PAGE_SIZE); i++) {
|
|
|
|
|
pqtype = pga[i].queue - pga[i].pc;
|
|
|
|
|
if ((VM_PAGE_TO_PHYS(&pga[i]) !=
|
|
|
|
|
(VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)) ||
|
|
|
|
|
((pqtype != PQ_FREE) && (pqtype != PQ_CACHE))) {
|
|
|
|
|
start++;
|
|
|
|
|
goto again;
|
|
|
|
|
}
|
|
|
|
|
}
|
2004-03-02 08:25:58 +00:00
|
|
|
mtx_unlock_spin(&vm_page_queue_free_mtx);
|
2001-07-04 23:27:09 +00:00
|
|
|
for (i = start; i < (start + size / PAGE_SIZE); i++) {
|
|
|
|
|
vm_page_t m = &pga[i];
|
|
|
|
|
|
2002-03-03 22:36:14 +00:00
|
|
|
if ((m->queue - m->pc) == PQ_CACHE) {
|
Try to close a potential, but serious race in our VM subsystem.
Historically, our contigmalloc1() and contigmalloc2() assumes
that a page in PQ_CACHE can be unconditionally reused by busying
and freeing it. Unfortunatelly, when object happens to be not
NULL, the code will set m->object to NULL and disregard the fact
that the page is actually in the VM page bucket, resulting in
page bucket hash table corruption and finally, a filesystem
corruption, or a 'page not in hash' panic.
This commit has borrowed the idea taken from DragonFlyBSD's fix
to the VM fix by Matthew Dillon[1]. This version of patch will
do the following checks:
- When scanning pages in PQ_CACHE, check hold_count and
skip over pages that are held temporarily.
- For pages in PQ_CACHE and selected as candidate of being
freed, check if it is busy at that time.
Note: It seems that this is might be unrelated to kern/72539.
Obtained from: DragonFlyBSD, sys/vm/vm_contig.c,v 1.11 and 1.12 [1]
Reminded by: Matt Dillon
Reworked by: alc
MFC After: 1 week
2004-11-24 18:56:13 +00:00
|
|
|
if (m->hold_count != 0) {
|
|
|
|
|
start++;
|
|
|
|
|
goto again0;
|
|
|
|
|
}
|
2003-07-05 20:51:22 +00:00
|
|
|
object = m->object;
|
|
|
|
|
if (!VM_OBJECT_TRYLOCK(object)) {
|
|
|
|
|
start++;
|
2004-03-02 08:25:58 +00:00
|
|
|
goto again0;
|
2003-07-05 20:51:22 +00:00
|
|
|
}
|
Try to close a potential, but serious race in our VM subsystem.
Historically, our contigmalloc1() and contigmalloc2() assumes
that a page in PQ_CACHE can be unconditionally reused by busying
and freeing it. Unfortunatelly, when object happens to be not
NULL, the code will set m->object to NULL and disregard the fact
that the page is actually in the VM page bucket, resulting in
page bucket hash table corruption and finally, a filesystem
corruption, or a 'page not in hash' panic.
This commit has borrowed the idea taken from DragonFlyBSD's fix
to the VM fix by Matthew Dillon[1]. This version of patch will
do the following checks:
- When scanning pages in PQ_CACHE, check hold_count and
skip over pages that are held temporarily.
- For pages in PQ_CACHE and selected as candidate of being
freed, check if it is busy at that time.
Note: It seems that this is might be unrelated to kern/72539.
Obtained from: DragonFlyBSD, sys/vm/vm_contig.c,v 1.11 and 1.12 [1]
Reminded by: Matt Dillon
Reworked by: alc
MFC After: 1 week
2004-11-24 18:56:13 +00:00
|
|
|
if ((m->flags & PG_BUSY) || m->busy != 0) {
|
|
|
|
|
VM_OBJECT_UNLOCK(object);
|
|
|
|
|
start++;
|
|
|
|
|
goto again0;
|
|
|
|
|
}
|
2001-07-04 23:27:09 +00:00
|
|
|
vm_page_free(m);
|
2003-07-05 20:51:22 +00:00
|
|
|
VM_OBJECT_UNLOCK(object);
|
2001-07-04 23:27:09 +00:00
|
|
|
}
|
2004-01-08 20:48:26 +00:00
|
|
|
}
|
2004-03-02 08:25:58 +00:00
|
|
|
mtx_lock_spin(&vm_page_queue_free_mtx);
|
|
|
|
|
for (i = start; i < (start + size / PAGE_SIZE); i++) {
|
|
|
|
|
pqtype = pga[i].queue - pga[i].pc;
|
2004-03-05 04:46:32 +00:00
|
|
|
if (pqtype != PQ_FREE) {
|
2004-03-02 08:25:58 +00:00
|
|
|
start++;
|
|
|
|
|
goto again;
|
|
|
|
|
}
|
|
|
|
|
}
|
2004-01-08 20:48:26 +00:00
|
|
|
for (i = start; i < (start + size / PAGE_SIZE); i++) {
|
|
|
|
|
vm_page_t m = &pga[i];
|
2002-03-03 22:36:14 +00:00
|
|
|
vm_pageq_remove_nowakeup(m);
|
2001-07-04 23:27:09 +00:00
|
|
|
m->valid = VM_PAGE_BITS_ALL;
|
2001-10-17 17:34:34 +00:00
|
|
|
if (m->flags & PG_ZERO)
|
|
|
|
|
vm_page_zero_count--;
|
2003-07-25 21:02:25 +00:00
|
|
|
/* Don't clear the PG_ZERO flag, we'll need it later. */
|
2004-01-10 21:17:53 +00:00
|
|
|
m->flags = PG_UNMANAGED | (m->flags & PG_ZERO);
|
2003-07-02 01:47:47 +00:00
|
|
|
KASSERT(m->dirty == 0,
|
|
|
|
|
("contigmalloc1: page %p was dirty", m));
|
2001-07-04 23:27:09 +00:00
|
|
|
m->wire_count = 0;
|
|
|
|
|
m->busy = 0;
|
|
|
|
|
}
|
2004-01-08 20:48:26 +00:00
|
|
|
mtx_unlock_spin(&vm_page_queue_free_mtx);
|
2002-07-12 03:17:22 +00:00
|
|
|
vm_page_unlock_queues();
|
2001-07-04 23:27:09 +00:00
|
|
|
/*
|
|
|
|
|
* We've found a contiguous chunk that meets are requirements.
|
2003-07-02 01:47:47 +00:00
|
|
|
* Allocate kernel VM, unfree and assign the physical pages to
|
|
|
|
|
* it and return kernel VM pointer.
|
2001-07-04 23:27:09 +00:00
|
|
|
*/
|
2001-10-13 04:23:37 +00:00
|
|
|
vm_map_lock(map);
|
|
|
|
|
if (vm_map_findspace(map, vm_map_min(map), size, &addr) !=
|
|
|
|
|
KERN_SUCCESS) {
|
2001-07-04 23:27:09 +00:00
|
|
|
/*
|
|
|
|
|
* XXX We almost never run out of kernel virtual
|
|
|
|
|
* space, so we don't make the allocated memory
|
|
|
|
|
* above available.
|
|
|
|
|
*/
|
2001-10-13 04:23:37 +00:00
|
|
|
vm_map_unlock(map);
|
2001-07-04 23:27:09 +00:00
|
|
|
return (NULL);
|
|
|
|
|
}
|
2001-10-13 04:23:37 +00:00
|
|
|
vm_object_reference(kernel_object);
|
|
|
|
|
vm_map_insert(map, kernel_object, addr - VM_MIN_KERNEL_ADDRESS,
|
|
|
|
|
addr, addr + size, VM_PROT_ALL, VM_PROT_ALL, 0);
|
|
|
|
|
vm_map_unlock(map);
|
2001-07-04 23:27:09 +00:00
|
|
|
|
2001-10-13 04:23:37 +00:00
|
|
|
tmp_addr = addr;
|
2003-04-14 00:36:53 +00:00
|
|
|
VM_OBJECT_LOCK(kernel_object);
|
2001-07-04 23:27:09 +00:00
|
|
|
for (i = start; i < (start + size / PAGE_SIZE); i++) {
|
|
|
|
|
vm_page_t m = &pga[i];
|
|
|
|
|
vm_page_insert(m, kernel_object,
|
|
|
|
|
OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS));
|
2003-07-25 21:02:25 +00:00
|
|
|
if ((flags & M_ZERO) && !(m->flags & PG_ZERO))
|
2003-07-27 10:41:33 +00:00
|
|
|
pmap_zero_page(m);
|
2001-07-04 23:27:09 +00:00
|
|
|
tmp_addr += PAGE_SIZE;
|
|
|
|
|
}
|
2003-04-14 00:36:53 +00:00
|
|
|
VM_OBJECT_UNLOCK(kernel_object);
|
2003-08-11 07:14:08 +00:00
|
|
|
vm_map_wire(map, addr, addr + size,
|
|
|
|
|
VM_MAP_WIRE_SYSTEM|VM_MAP_WIRE_NOHOLES);
|
2001-07-04 23:27:09 +00:00
|
|
|
|
|
|
|
|
return ((void *)addr);
|
|
|
|
|
}
|
2003-07-02 01:47:47 +00:00
|
|
|
return (NULL);
|
2001-07-04 23:27:09 +00:00
|
|
|
}
|
|
|
|
|
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
static void
|
|
|
|
|
vm_page_release_contigl(vm_page_t m, vm_pindex_t count)
|
|
|
|
|
{
|
|
|
|
|
while (count--) {
|
|
|
|
|
vm_page_free_toq(m);
|
|
|
|
|
m++;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
|
vm_page_release_contig(vm_page_t m, vm_pindex_t count)
|
|
|
|
|
{
|
|
|
|
|
vm_page_lock_queues();
|
|
|
|
|
vm_page_release_contigl(m, count);
|
|
|
|
|
vm_page_unlock_queues();
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static int
|
|
|
|
|
vm_contig_unqueue_free(vm_page_t m)
|
|
|
|
|
{
|
|
|
|
|
int error = 0;
|
|
|
|
|
|
|
|
|
|
mtx_lock_spin(&vm_page_queue_free_mtx);
|
|
|
|
|
if ((m->queue - m->pc) == PQ_FREE)
|
|
|
|
|
vm_pageq_remove_nowakeup(m);
|
|
|
|
|
else
|
|
|
|
|
error = EAGAIN;
|
|
|
|
|
mtx_unlock_spin(&vm_page_queue_free_mtx);
|
|
|
|
|
if (error)
|
|
|
|
|
return (error);
|
|
|
|
|
m->valid = VM_PAGE_BITS_ALL;
|
|
|
|
|
if (m->flags & PG_ZERO)
|
|
|
|
|
vm_page_zero_count--;
|
|
|
|
|
/* Don't clear the PG_ZERO flag; we'll need it later. */
|
|
|
|
|
m->flags = PG_UNMANAGED | (m->flags & PG_ZERO);
|
|
|
|
|
KASSERT(m->dirty == 0,
|
|
|
|
|
("contigmalloc2: page %p was dirty", m));
|
|
|
|
|
m->wire_count = 0;
|
|
|
|
|
m->busy = 0;
|
|
|
|
|
return (error);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
vm_page_t
|
|
|
|
|
vm_page_alloc_contig(vm_pindex_t npages, vm_paddr_t low, vm_paddr_t high,
|
|
|
|
|
vm_offset_t alignment, vm_offset_t boundary)
|
|
|
|
|
{
|
|
|
|
|
vm_object_t object;
|
|
|
|
|
vm_offset_t size;
|
|
|
|
|
vm_paddr_t phys;
|
|
|
|
|
vm_page_t pga = vm_page_array;
|
|
|
|
|
int i, pass, pqtype, start;
|
|
|
|
|
|
|
|
|
|
size = npages << PAGE_SHIFT;
|
|
|
|
|
if (size == 0)
|
|
|
|
|
panic("vm_page_alloc_contig: size must not be 0");
|
|
|
|
|
if ((alignment & (alignment - 1)) != 0)
|
|
|
|
|
panic("vm_page_alloc_contig: alignment must be a power of 2");
|
|
|
|
|
if ((boundary & (boundary - 1)) != 0)
|
|
|
|
|
panic("vm_page_alloc_contig: boundary must be a power of 2");
|
|
|
|
|
|
|
|
|
|
for (pass = 0; pass < 2; pass++) {
|
|
|
|
|
start = vm_page_array_size;
|
|
|
|
|
vm_page_lock_queues();
|
|
|
|
|
retry:
|
|
|
|
|
start--;
|
|
|
|
|
/*
|
|
|
|
|
* Find last page in array that is free, within range,
|
|
|
|
|
* aligned, and such that the boundary won't be crossed.
|
|
|
|
|
*/
|
|
|
|
|
for (i = start; i >= 0; i--) {
|
|
|
|
|
phys = VM_PAGE_TO_PHYS(&pga[i]);
|
|
|
|
|
pqtype = pga[i].queue - pga[i].pc;
|
|
|
|
|
if (pass == 0) {
|
|
|
|
|
if (pqtype != PQ_FREE && pqtype != PQ_CACHE)
|
|
|
|
|
continue;
|
|
|
|
|
} else if (pqtype != PQ_FREE && pqtype != PQ_CACHE &&
|
|
|
|
|
pga[i].queue != PQ_ACTIVE &&
|
|
|
|
|
pga[i].queue != PQ_INACTIVE)
|
|
|
|
|
continue;
|
|
|
|
|
if (phys >= low && phys + size <= high &&
|
|
|
|
|
((phys & (alignment - 1)) == 0) &&
|
|
|
|
|
((phys ^ (phys + size - 1)) & ~(boundary - 1)) == 0)
|
|
|
|
|
break;
|
|
|
|
|
}
|
|
|
|
|
/* There are no candidates at all. */
|
|
|
|
|
if (i == -1) {
|
|
|
|
|
vm_page_unlock_queues();
|
|
|
|
|
continue;
|
|
|
|
|
}
|
|
|
|
|
start = i;
|
|
|
|
|
/*
|
|
|
|
|
* Check successive pages for contiguous and free.
|
|
|
|
|
*/
|
|
|
|
|
for (i = start + 1; i < start + npages; i++) {
|
|
|
|
|
pqtype = pga[i].queue - pga[i].pc;
|
|
|
|
|
if (VM_PAGE_TO_PHYS(&pga[i]) !=
|
|
|
|
|
VM_PAGE_TO_PHYS(&pga[i - 1]) + PAGE_SIZE)
|
|
|
|
|
goto retry;
|
|
|
|
|
if (pass == 0) {
|
|
|
|
|
if (pqtype != PQ_FREE && pqtype != PQ_CACHE)
|
|
|
|
|
goto retry;
|
|
|
|
|
} else if (pqtype != PQ_FREE && pqtype != PQ_CACHE &&
|
|
|
|
|
pga[i].queue != PQ_ACTIVE &&
|
|
|
|
|
pga[i].queue != PQ_INACTIVE)
|
|
|
|
|
goto retry;
|
|
|
|
|
}
|
|
|
|
|
for (i = start; i < start + npages; i++) {
|
|
|
|
|
vm_page_t m = &pga[i];
|
|
|
|
|
|
|
|
|
|
retry_page:
|
|
|
|
|
pqtype = m->queue - m->pc;
|
|
|
|
|
if (pass != 0 && pqtype != PQ_FREE &&
|
|
|
|
|
pqtype != PQ_CACHE) {
|
|
|
|
|
switch (m->queue) {
|
|
|
|
|
case PQ_ACTIVE:
|
|
|
|
|
case PQ_INACTIVE:
|
|
|
|
|
if (vm_contig_launder_page(m) != 0)
|
|
|
|
|
goto cleanup_freed;
|
|
|
|
|
pqtype = m->queue - m->pc;
|
|
|
|
|
if (pqtype == PQ_FREE ||
|
|
|
|
|
pqtype == PQ_CACHE)
|
|
|
|
|
break;
|
|
|
|
|
default:
|
|
|
|
|
cleanup_freed:
|
|
|
|
|
vm_page_release_contigl(&pga[start],
|
|
|
|
|
i - start);
|
|
|
|
|
goto retry;
|
|
|
|
|
}
|
|
|
|
|
}
|
|
|
|
|
if (pqtype == PQ_CACHE) {
|
Try to close a potential, but serious race in our VM subsystem.
Historically, our contigmalloc1() and contigmalloc2() assumes
that a page in PQ_CACHE can be unconditionally reused by busying
and freeing it. Unfortunatelly, when object happens to be not
NULL, the code will set m->object to NULL and disregard the fact
that the page is actually in the VM page bucket, resulting in
page bucket hash table corruption and finally, a filesystem
corruption, or a 'page not in hash' panic.
This commit has borrowed the idea taken from DragonFlyBSD's fix
to the VM fix by Matthew Dillon[1]. This version of patch will
do the following checks:
- When scanning pages in PQ_CACHE, check hold_count and
skip over pages that are held temporarily.
- For pages in PQ_CACHE and selected as candidate of being
freed, check if it is busy at that time.
Note: It seems that this is might be unrelated to kern/72539.
Obtained from: DragonFlyBSD, sys/vm/vm_contig.c,v 1.11 and 1.12 [1]
Reminded by: Matt Dillon
Reworked by: alc
MFC After: 1 week
2004-11-24 18:56:13 +00:00
|
|
|
if (m->hold_count != 0)
|
|
|
|
|
goto retry;
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
object = m->object;
|
|
|
|
|
if (!VM_OBJECT_TRYLOCK(object))
|
|
|
|
|
goto retry;
|
Try to close a potential, but serious race in our VM subsystem.
Historically, our contigmalloc1() and contigmalloc2() assumes
that a page in PQ_CACHE can be unconditionally reused by busying
and freeing it. Unfortunatelly, when object happens to be not
NULL, the code will set m->object to NULL and disregard the fact
that the page is actually in the VM page bucket, resulting in
page bucket hash table corruption and finally, a filesystem
corruption, or a 'page not in hash' panic.
This commit has borrowed the idea taken from DragonFlyBSD's fix
to the VM fix by Matthew Dillon[1]. This version of patch will
do the following checks:
- When scanning pages in PQ_CACHE, check hold_count and
skip over pages that are held temporarily.
- For pages in PQ_CACHE and selected as candidate of being
freed, check if it is busy at that time.
Note: It seems that this is might be unrelated to kern/72539.
Obtained from: DragonFlyBSD, sys/vm/vm_contig.c,v 1.11 and 1.12 [1]
Reminded by: Matt Dillon
Reworked by: alc
MFC After: 1 week
2004-11-24 18:56:13 +00:00
|
|
|
if ((m->flags & PG_BUSY) || m->busy != 0) {
|
|
|
|
|
VM_OBJECT_UNLOCK(object);
|
|
|
|
|
goto retry;
|
|
|
|
|
}
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
vm_page_free(m);
|
|
|
|
|
VM_OBJECT_UNLOCK(object);
|
|
|
|
|
}
|
|
|
|
|
/*
|
|
|
|
|
* There is no good API for freeing a page
|
|
|
|
|
* directly to PQ_NONE on our behalf, so spin.
|
|
|
|
|
*/
|
|
|
|
|
if (vm_contig_unqueue_free(m) != 0)
|
|
|
|
|
goto retry_page;
|
|
|
|
|
}
|
|
|
|
|
vm_page_unlock_queues();
|
|
|
|
|
/*
|
|
|
|
|
* We've found a contiguous chunk that meets are requirements.
|
|
|
|
|
*/
|
|
|
|
|
return (&pga[start]);
|
|
|
|
|
}
|
|
|
|
|
return (NULL);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
static void *
|
|
|
|
|
contigmalloc2(vm_page_t m, vm_pindex_t npages, int flags)
|
|
|
|
|
{
|
|
|
|
|
vm_object_t object = kernel_object;
|
|
|
|
|
vm_map_t map = kernel_map;
|
|
|
|
|
vm_offset_t addr, tmp_addr;
|
|
|
|
|
vm_pindex_t i;
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
|
* Allocate kernel VM, unfree and assign the physical pages to
|
|
|
|
|
* it and return kernel VM pointer.
|
|
|
|
|
*/
|
|
|
|
|
vm_map_lock(map);
|
|
|
|
|
if (vm_map_findspace(map, vm_map_min(map), npages << PAGE_SHIFT, &addr)
|
|
|
|
|
!= KERN_SUCCESS) {
|
|
|
|
|
vm_map_unlock(map);
|
|
|
|
|
return (NULL);
|
|
|
|
|
}
|
|
|
|
|
vm_object_reference(object);
|
|
|
|
|
vm_map_insert(map, object, addr - VM_MIN_KERNEL_ADDRESS,
|
|
|
|
|
addr, addr + (npages << PAGE_SHIFT), VM_PROT_ALL, VM_PROT_ALL, 0);
|
|
|
|
|
vm_map_unlock(map);
|
|
|
|
|
tmp_addr = addr;
|
|
|
|
|
VM_OBJECT_LOCK(object);
|
|
|
|
|
for (i = 0; i < npages; i++) {
|
|
|
|
|
vm_page_insert(&m[i], object,
|
|
|
|
|
OFF_TO_IDX(tmp_addr - VM_MIN_KERNEL_ADDRESS));
|
|
|
|
|
if ((flags & M_ZERO) && !(m->flags & PG_ZERO))
|
|
|
|
|
pmap_zero_page(&m[i]);
|
|
|
|
|
tmp_addr += PAGE_SIZE;
|
|
|
|
|
}
|
|
|
|
|
VM_OBJECT_UNLOCK(object);
|
|
|
|
|
vm_map_wire(map, addr, addr + (npages << PAGE_SHIFT),
|
|
|
|
|
VM_MAP_WIRE_SYSTEM | VM_MAP_WIRE_NOHOLES);
|
|
|
|
|
return ((void *)addr);
|
|
|
|
|
}
|
|
|
|
|
|
2004-08-05 21:54:11 +00:00
|
|
|
static int vm_old_contigmalloc = 0;
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
SYSCTL_INT(_vm, OID_AUTO, old_contigmalloc,
|
|
|
|
|
CTLFLAG_RW, &vm_old_contigmalloc, 0, "Use the old contigmalloc algorithm");
|
|
|
|
|
TUNABLE_INT("vm.old_contigmalloc", &vm_old_contigmalloc);
|
|
|
|
|
|
2001-07-04 23:27:09 +00:00
|
|
|
void *
|
|
|
|
|
contigmalloc(
|
|
|
|
|
unsigned long size, /* should be size_t here and for malloc() */
|
|
|
|
|
struct malloc_type *type,
|
|
|
|
|
int flags,
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t low,
|
|
|
|
|
vm_paddr_t high,
|
2001-07-04 23:27:09 +00:00
|
|
|
unsigned long alignment,
|
|
|
|
|
unsigned long boundary)
|
|
|
|
|
{
|
|
|
|
|
void * ret;
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
vm_page_t pages;
|
|
|
|
|
vm_pindex_t npgs;
|
2001-07-04 23:27:09 +00:00
|
|
|
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
npgs = round_page(size) >> PAGE_SHIFT;
|
2003-07-26 21:48:46 +00:00
|
|
|
mtx_lock(&Giant);
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
if (vm_old_contigmalloc) {
|
|
|
|
|
ret = contigmalloc1(size, type, flags, low, high, alignment,
|
|
|
|
|
boundary, kernel_map);
|
|
|
|
|
} else {
|
|
|
|
|
pages = vm_page_alloc_contig(npgs, low, high,
|
|
|
|
|
alignment, boundary);
|
|
|
|
|
if (pages == NULL) {
|
|
|
|
|
ret = NULL;
|
|
|
|
|
} else {
|
|
|
|
|
ret = contigmalloc2(pages, npgs, flags);
|
|
|
|
|
if (ret == NULL)
|
|
|
|
|
vm_page_release_contig(pages, npgs);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
}
|
2003-07-26 21:48:46 +00:00
|
|
|
mtx_unlock(&Giant);
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
malloc_type_allocated(type, ret == NULL ? 0 : npgs << PAGE_SHIFT);
|
2001-07-04 23:27:09 +00:00
|
|
|
return (ret);
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
void
|
|
|
|
|
contigfree(void *addr, unsigned long size, struct malloc_type *type)
|
|
|
|
|
{
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
vm_pindex_t npgs;
|
2004-03-13 07:09:15 +00:00
|
|
|
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
npgs = round_page(size) >> PAGE_SHIFT;
|
2001-07-04 23:27:09 +00:00
|
|
|
kmem_free(kernel_map, (vm_offset_t)addr, size);
|
Reimplement contigmalloc(9) with an algorithm which stands a greatly-
improved chance of working despite pressure from running programs.
Instead of trying to throw a bunch of pages out to swap and hope for
the best, only a range that can potentially fulfill contigmalloc(9)'s
request will have its contents paged out (potentially, not forcibly)
at a time.
The new contigmalloc operation still operates in three passes, but it
could potentially be tuned to more or less. The first pass only looks
at pages in the cache and free pages, so they would be thrown out
without having to block. If this is not enough, the subsequent passes
page out any unwired memory. To combat memory pressure refragmenting
the section of memory being laundered, each page is removed from the
systems' free memory queue once it has been freed so that blocking
later doesn't cause the memory laundered so far to get reallocated.
The page-out operations are now blocking, as it would make little sense
to try to push out a page, then get its status immediately afterward
to remove it from the available free pages queue, if it's unlikely to
have been freed. Another change is that if KVA allocation fails, the
allocated memory segment will be freed and not leaked.
There is a sysctl/tunable, defaulting to on, which causes the old
contigmalloc() algorithm to be used. Nonetheless, I have been using
vm.old_contigmalloc=0 for over a month. It is safe to switch at
run-time to see the difference it makes.
A new interface has been used which does not require mapping the
allocated pages into KVA: vm_page.h functions vm_page_alloc_contig()
and vm_page_release_contig(). These are what vm.old_contigmalloc=0
uses internally, so the sysctl/tunable does not affect their operation.
When using the contigmalloc(9) and contigfree(9) interfaces, memory
is now tracked with malloc(9) stats. Several functions have been
exported from kern_malloc.c to allow other subsystems to use these
statistics, as well. This invalidates the BUGS section of the
contigmalloc(9) manpage.
2004-07-19 06:21:27 +00:00
|
|
|
malloc_type_freed(type, npgs << PAGE_SHIFT);
|
2001-07-04 23:27:09 +00:00
|
|
|
}
|
