2005-01-07 02:29:27 +00:00
|
|
|
/*-
|
1994-05-25 09:21:21 +00:00
|
|
|
* Copyright (c) 1991 Regents of the University of California.
|
|
|
|
* All rights reserved.
|
2008-03-18 06:52:15 +00:00
|
|
|
* Copyright (c) 1998 Matthew Dillon. All Rights Reserved.
|
1994-05-24 10:09:53 +00:00
|
|
|
*
|
|
|
|
* 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.
|
|
|
|
*
|
1994-05-25 09:21:21 +00:00
|
|
|
* from: @(#)vm_page.c 7.4 (Berkeley) 5/7/91
|
|
|
|
*/
|
|
|
|
|
2005-01-07 02:29:27 +00:00
|
|
|
/*-
|
1994-05-24 10:09:53 +00:00
|
|
|
* Copyright (c) 1987, 1990 Carnegie-Mellon University.
|
|
|
|
* All rights reserved.
|
|
|
|
*
|
|
|
|
* Authors: Avadis Tevanian, Jr., Michael Wayne Young
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*
|
1994-05-24 10:09:53 +00:00
|
|
|
* 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.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*
|
|
|
|
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
|
|
|
|
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
|
1994-05-24 10:09:53 +00:00
|
|
|
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*
|
1994-05-24 10:09:53 +00:00
|
|
|
* 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.
|
|
|
|
*/
|
|
|
|
|
2001-07-04 23:27:09 +00:00
|
|
|
/*
|
|
|
|
* GENERAL RULES ON VM_PAGE MANIPULATION
|
|
|
|
*
|
|
|
|
* - a pageq mutex is required when adding or removing a page from a
|
|
|
|
* page queue (vm_page_queue[]), regardless of other mutexes or the
|
|
|
|
* busy state of a page.
|
|
|
|
*
|
|
|
|
* - a hash chain mutex is required when associating or disassociating
|
|
|
|
* a page from the VM PAGE CACHE hash table (vm_page_buckets),
|
|
|
|
* regardless of other mutexes or the busy state of a page.
|
|
|
|
*
|
|
|
|
* - either a hash chain mutex OR a busied page is required in order
|
|
|
|
* to modify the page flags. A hash chain mutex must be obtained in
|
|
|
|
* order to busy a page. A page's flags cannot be modified by a
|
|
|
|
* hash chain mutex if the page is marked busy.
|
|
|
|
*
|
|
|
|
* - The object memq mutex is held when inserting or removing
|
|
|
|
* pages from an object (vm_page_insert() or vm_page_remove()). This
|
|
|
|
* is different from the object's main mutex.
|
|
|
|
*
|
|
|
|
* Generally speaking, you have to be aware of side effects when running
|
|
|
|
* vm_page ops. A vm_page_lookup() will return with the hash chain
|
|
|
|
* locked, whether it was able to lookup the page or not. vm_page_free(),
|
|
|
|
* vm_page_cache(), vm_page_activate(), and a number of other routines
|
|
|
|
* will release the hash chain mutex for you. Intermediate manipulation
|
|
|
|
* routines such as vm_page_flag_set() expect the hash chain to be held
|
|
|
|
* on entry and the hash chain will remain held on return.
|
|
|
|
*
|
|
|
|
* pageq scanning can only occur with the pageq in question locked.
|
|
|
|
* We have a known bottleneck with the active queue, but the cache
|
|
|
|
* and free queues are actually arrays already.
|
|
|
|
*/
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* Resident memory management module.
|
|
|
|
*/
|
|
|
|
|
2003-06-11 23:50:51 +00:00
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
2007-12-29 19:53:04 +00:00
|
|
|
#include "opt_vm.h"
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <sys/param.h>
|
|
|
|
#include <sys/systm.h>
|
2001-05-01 08:13:21 +00:00
|
|
|
#include <sys/lock.h>
|
2005-08-12 12:24:19 +00:00
|
|
|
#include <sys/kernel.h>
|
2009-01-03 13:24:08 +00:00
|
|
|
#include <sys/limits.h>
|
1996-01-27 00:13:33 +00:00
|
|
|
#include <sys/malloc.h>
|
2001-05-22 07:01:11 +00:00
|
|
|
#include <sys/mutex.h>
|
1994-05-25 09:21:21 +00:00
|
|
|
#include <sys/proc.h>
|
2005-08-12 12:24:19 +00:00
|
|
|
#include <sys/sysctl.h>
|
1995-12-07 12:48:31 +00:00
|
|
|
#include <sys/vmmeter.h>
|
1997-12-29 00:25:11 +00:00
|
|
|
#include <sys/vnode.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
#include <vm/vm.h>
|
2010-04-30 00:46:43 +00:00
|
|
|
#include <vm/pmap.h>
|
1995-12-07 12:48:31 +00:00
|
|
|
#include <vm/vm_param.h>
|
1995-03-16 18:17:34 +00:00
|
|
|
#include <vm/vm_kern.h>
|
1995-12-07 12:48:31 +00:00
|
|
|
#include <vm/vm_object.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <vm/vm_page.h>
|
|
|
|
#include <vm/vm_pageout.h>
|
1999-01-21 08:29:12 +00:00
|
|
|
#include <vm/vm_pager.h>
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
#include <vm/vm_phys.h>
|
2007-12-29 19:53:04 +00:00
|
|
|
#include <vm/vm_reserv.h>
|
1995-12-07 12:48:31 +00:00
|
|
|
#include <vm/vm_extern.h>
|
2002-03-19 09:11:49 +00:00
|
|
|
#include <vm/uma.h>
|
|
|
|
#include <vm/uma_int.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
|
Introduce minidumps. Full physical memory crash dumps are still available
via the debug.minidump sysctl and tunable.
Traditional dumps store all physical memory. This was once a good thing
when machines had a maximum of 64M of ram and 1GB of kvm. These days,
machines often have many gigabytes of ram and a smaller amount of kvm.
libkvm+kgdb don't have a way to access physical ram that is not mapped
into kvm at the time of the crash dump, so the extra ram being dumped
is mostly wasted.
Minidumps invert the process. Instead of dumping physical memory in
in order to guarantee that all of kvm's backing is dumped, minidumps
instead dump only memory that is actively mapped into kvm.
amd64 has a direct map region that things like UMA use. Obviously we
cannot dump all of the direct map region because that is effectively
an old style all-physical-memory dump. Instead, introduce a bitmap
and two helper routines (dump_add_page(pa) and dump_drop_page(pa)) that
allow certain critical direct map pages to be included in the dump.
uma_machdep.c's allocator is the intended consumer.
Dumps are a custom format. At the very beginning of the file is a header,
then a copy of the message buffer, then the bitmap of pages present in
the dump, then the final level of the kvm page table trees (2MB mappings
are expanded into a 4K page mappings), then the sparse physical pages
according to the bitmap. libkvm can now conveniently access the kvm
page table entries.
Booting my test 8GB machine, forcing it into ddb and forcing a dump
leads to a 48MB minidump. While this is a best case, I expect minidumps
to be in the 100MB-500MB range. Obviously, never larger than physical
memory of course.
minidumps are on by default. It would want be necessary to turn them off
if it was necessary to debug corrupt kernel page table management as that
would mess up minidumps as well.
Both minidumps and regular dumps are supported on the same machine.
2006-04-21 04:24:50 +00:00
|
|
|
#include <machine/md_var.h>
|
|
|
|
|
2010-04-30 00:46:43 +00:00
|
|
|
#if defined(__amd64__) || defined (__i386__)
|
|
|
|
extern struct sysctl_oid_list sysctl__vm_pmap_children;
|
|
|
|
#else
|
|
|
|
SYSCTL_NODE(_vm, OID_AUTO, pmap, CTLFLAG_RD, 0, "VM/pmap parameters");
|
|
|
|
#endif
|
|
|
|
|
|
|
|
static uint64_t pmap_tryrelock_calls;
|
|
|
|
SYSCTL_QUAD(_vm_pmap, OID_AUTO, tryrelock_calls, CTLFLAG_RD,
|
|
|
|
&pmap_tryrelock_calls, 0, "Number of tryrelock calls");
|
|
|
|
|
|
|
|
static int pmap_tryrelock_restart;
|
|
|
|
SYSCTL_INT(_vm_pmap, OID_AUTO, tryrelock_restart, CTLFLAG_RD,
|
|
|
|
&pmap_tryrelock_restart, 0, "Number of tryrelock restarts");
|
|
|
|
|
|
|
|
static int pmap_tryrelock_race;
|
|
|
|
SYSCTL_INT(_vm_pmap, OID_AUTO, tryrelock_race, CTLFLAG_RD,
|
|
|
|
&pmap_tryrelock_race, 0, "Number of tryrelock pmap race cases");
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* Associated with page of user-allocatable memory is a
|
|
|
|
* page structure.
|
|
|
|
*/
|
1996-01-19 04:00:31 +00:00
|
|
|
|
2008-03-18 06:52:15 +00:00
|
|
|
struct vpgqueues vm_page_queues[PQ_COUNT];
|
2009-10-04 18:53:10 +00:00
|
|
|
struct vpglocks vm_page_queue_lock;
|
|
|
|
struct vpglocks vm_page_queue_free_lock;
|
2002-07-04 22:07:37 +00:00
|
|
|
|
2010-04-30 00:46:43 +00:00
|
|
|
struct vpglocks pa_lock[PA_LOCK_COUNT] __aligned(CACHE_LINE_SIZE);
|
|
|
|
|
1996-11-17 02:38:31 +00:00
|
|
|
vm_page_t vm_page_array = 0;
|
2000-05-21 12:50:18 +00:00
|
|
|
int vm_page_array_size = 0;
|
1996-11-17 02:38:31 +00:00
|
|
|
long first_page = 0;
|
|
|
|
int vm_page_zero_count = 0;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2005-08-12 12:24:19 +00:00
|
|
|
static int boot_pages = UMA_BOOT_PAGES;
|
|
|
|
TUNABLE_INT("vm.boot_pages", &boot_pages);
|
|
|
|
SYSCTL_INT(_vm, OID_AUTO, boot_pages, CTLFLAG_RD, &boot_pages, 0,
|
|
|
|
"number of pages allocated for bootstrapping the VM system");
|
|
|
|
|
2008-03-18 06:52:15 +00:00
|
|
|
static void vm_page_enqueue(int queue, vm_page_t m);
|
|
|
|
|
2008-09-26 18:44:40 +00:00
|
|
|
/* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
|
|
|
|
#if PAGE_SIZE == 32768
|
|
|
|
#ifdef CTASSERT
|
|
|
|
CTASSERT(sizeof(u_long) >= 8);
|
|
|
|
#endif
|
|
|
|
#endif
|
|
|
|
|
2010-04-30 00:46:43 +00:00
|
|
|
/*
|
|
|
|
* Try to acquire a physical address lock while a pmap is locked. If we
|
|
|
|
* fail to trylock we unlock and lock the pmap directly and cache the
|
|
|
|
* locked pa in *locked. The caller should then restart their loop in case
|
|
|
|
* the virtual to physical mapping has changed.
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
vm_page_pa_tryrelock(pmap_t pmap, vm_paddr_t pa, vm_paddr_t *locked)
|
|
|
|
{
|
|
|
|
vm_paddr_t lockpa;
|
|
|
|
uint32_t gen_count;
|
|
|
|
|
|
|
|
gen_count = pmap->pm_gen_count;
|
|
|
|
atomic_add_long((volatile long *)&pmap_tryrelock_calls, 1);
|
|
|
|
lockpa = *locked;
|
|
|
|
*locked = pa;
|
|
|
|
if (lockpa) {
|
|
|
|
PA_LOCK_ASSERT(lockpa, MA_OWNED);
|
|
|
|
if (PA_LOCKPTR(pa) == PA_LOCKPTR(lockpa))
|
|
|
|
return (0);
|
|
|
|
PA_UNLOCK(lockpa);
|
|
|
|
}
|
|
|
|
if (PA_TRYLOCK(pa))
|
|
|
|
return (0);
|
|
|
|
PMAP_UNLOCK(pmap);
|
|
|
|
atomic_add_int((volatile int *)&pmap_tryrelock_restart, 1);
|
|
|
|
PA_LOCK(pa);
|
|
|
|
PMAP_LOCK(pmap);
|
|
|
|
|
|
|
|
if (pmap->pm_gen_count != gen_count + 1) {
|
|
|
|
pmap->pm_retries++;
|
|
|
|
atomic_add_int((volatile int *)&pmap_tryrelock_race, 1);
|
|
|
|
return (EAGAIN);
|
|
|
|
}
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* vm_set_page_size:
|
|
|
|
*
|
|
|
|
* Sets the page size, perhaps based upon the memory
|
|
|
|
* size. Must be called before any use of page-size
|
|
|
|
* dependent functions.
|
|
|
|
*/
|
1995-05-30 08:16:23 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_set_page_size(void)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2007-05-31 22:52:15 +00:00
|
|
|
if (cnt.v_page_size == 0)
|
|
|
|
cnt.v_page_size = PAGE_SIZE;
|
|
|
|
if (((cnt.v_page_size - 1) & cnt.v_page_size) != 0)
|
1994-05-24 10:09:53 +00:00
|
|
|
panic("vm_set_page_size: page size not a power of two");
|
|
|
|
}
|
|
|
|
|
2006-06-23 16:44:24 +00:00
|
|
|
/*
|
|
|
|
* vm_page_blacklist_lookup:
|
|
|
|
*
|
|
|
|
* See if a physical address in this page has been listed
|
|
|
|
* in the blacklist tunable. Entries in the tunable are
|
|
|
|
* separated by spaces or commas. If an invalid integer is
|
|
|
|
* encountered then the rest of the string is skipped.
|
|
|
|
*/
|
|
|
|
static int
|
|
|
|
vm_page_blacklist_lookup(char *list, vm_paddr_t pa)
|
|
|
|
{
|
|
|
|
vm_paddr_t bad;
|
|
|
|
char *cp, *pos;
|
|
|
|
|
|
|
|
for (pos = list; *pos != '\0'; pos = cp) {
|
|
|
|
bad = strtoq(pos, &cp, 0);
|
|
|
|
if (*cp != '\0') {
|
|
|
|
if (*cp == ' ' || *cp == ',') {
|
|
|
|
cp++;
|
|
|
|
if (cp == pos)
|
|
|
|
continue;
|
|
|
|
} else
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
if (pa == trunc_page(bad))
|
|
|
|
return (1);
|
|
|
|
}
|
|
|
|
return (0);
|
|
|
|
}
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* vm_page_startup:
|
|
|
|
*
|
|
|
|
* Initializes the resident memory module.
|
|
|
|
*
|
|
|
|
* Allocates memory for the page cells, and
|
|
|
|
* for the object/offset-to-page hash table headers.
|
|
|
|
* Each page cell is initialized and placed on the free list.
|
|
|
|
*/
|
1994-05-25 09:21:21 +00:00
|
|
|
vm_offset_t
|
2004-04-04 23:33:36 +00:00
|
|
|
vm_page_startup(vm_offset_t vaddr)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2001-07-04 19:00:13 +00:00
|
|
|
vm_offset_t mapped;
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t page_range;
|
|
|
|
vm_paddr_t new_end;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
int i;
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t pa;
|
1994-05-25 09:21:21 +00:00
|
|
|
int nblocks;
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t last_pa;
|
2006-06-23 16:44:24 +00:00
|
|
|
char *list;
|
1994-05-25 09:21:21 +00:00
|
|
|
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
/* the biggest memory array is the second group of pages */
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t end;
|
|
|
|
vm_paddr_t biggestsize;
|
2006-12-08 08:44:47 +00:00
|
|
|
vm_paddr_t low_water, high_water;
|
2003-03-25 00:07:06 +00:00
|
|
|
int biggestone;
|
1994-05-25 09:21:21 +00:00
|
|
|
|
|
|
|
biggestsize = 0;
|
|
|
|
biggestone = 0;
|
|
|
|
nblocks = 0;
|
|
|
|
vaddr = round_page(vaddr);
|
|
|
|
|
|
|
|
for (i = 0; phys_avail[i + 1]; i += 2) {
|
|
|
|
phys_avail[i] = round_page(phys_avail[i]);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
phys_avail[i + 1] = trunc_page(phys_avail[i + 1]);
|
1994-05-25 09:21:21 +00:00
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2006-12-08 08:44:47 +00:00
|
|
|
low_water = phys_avail[0];
|
|
|
|
high_water = phys_avail[1];
|
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
for (i = 0; phys_avail[i + 1]; i += 2) {
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t size = phys_avail[i + 1] - phys_avail[i];
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
if (size > biggestsize) {
|
|
|
|
biggestone = i;
|
|
|
|
biggestsize = size;
|
|
|
|
}
|
2006-12-08 08:44:47 +00:00
|
|
|
if (phys_avail[i] < low_water)
|
|
|
|
low_water = phys_avail[i];
|
|
|
|
if (phys_avail[i + 1] > high_water)
|
|
|
|
high_water = phys_avail[i + 1];
|
1994-05-25 09:21:21 +00:00
|
|
|
++nblocks;
|
|
|
|
}
|
|
|
|
|
2008-08-17 23:40:29 +00:00
|
|
|
#ifdef XEN
|
|
|
|
low_water = 0;
|
|
|
|
#endif
|
|
|
|
|
2001-03-01 19:21:24 +00:00
|
|
|
end = phys_avail[biggestone+1];
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2002-07-04 22:07:37 +00:00
|
|
|
/*
|
2004-03-02 08:25:58 +00:00
|
|
|
* Initialize the locks.
|
2002-07-04 22:07:37 +00:00
|
|
|
*/
|
2004-07-29 18:56:31 +00:00
|
|
|
mtx_init(&vm_page_queue_mtx, "vm page queue mutex", NULL, MTX_DEF |
|
|
|
|
MTX_RECURSE);
|
2002-07-04 22:07:37 +00:00
|
|
|
mtx_init(&vm_page_queue_free_mtx, "vm page queue free mutex", NULL,
|
2007-02-05 06:02:55 +00:00
|
|
|
MTX_DEF);
|
2002-07-04 22:07:37 +00:00
|
|
|
|
2010-04-30 00:46:43 +00:00
|
|
|
/* Setup page locks. */
|
|
|
|
for (i = 0; i < PA_LOCK_COUNT; i++)
|
|
|
|
mtx_init(&pa_lock[i].data, "page lock", NULL,
|
|
|
|
MTX_DEF | MTX_RECURSE | MTX_DUPOK);
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
2007-12-02 07:43:42 +00:00
|
|
|
* Initialize the queue headers for the hold queue, the active queue,
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* and the inactive queue.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2008-03-18 06:52:15 +00:00
|
|
|
for (i = 0; i < PQ_COUNT; i++)
|
|
|
|
TAILQ_INIT(&vm_page_queues[i].pl);
|
|
|
|
vm_page_queues[PQ_INACTIVE].cnt = &cnt.v_inactive_count;
|
|
|
|
vm_page_queues[PQ_ACTIVE].cnt = &cnt.v_active_count;
|
|
|
|
vm_page_queues[PQ_HOLD].cnt = &cnt.v_active_count;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2002-03-19 09:11:49 +00:00
|
|
|
/*
|
2002-11-03 22:20:42 +00:00
|
|
|
* Allocate memory for use when boot strapping the kernel memory
|
|
|
|
* allocator.
|
2002-03-19 09:11:49 +00:00
|
|
|
*/
|
2005-08-12 12:24:19 +00:00
|
|
|
new_end = end - (boot_pages * UMA_SLAB_SIZE);
|
2002-03-19 09:11:49 +00:00
|
|
|
new_end = trunc_page(new_end);
|
|
|
|
mapped = pmap_map(&vaddr, new_end, end,
|
|
|
|
VM_PROT_READ | VM_PROT_WRITE);
|
2005-10-08 21:03:54 +00:00
|
|
|
bzero((void *)mapped, end - new_end);
|
|
|
|
uma_startup((void *)mapped, boot_pages);
|
2002-03-19 09:11:49 +00:00
|
|
|
|
2008-11-06 16:20:27 +00:00
|
|
|
#if defined(__amd64__) || defined(__i386__) || defined(__arm__)
|
Introduce minidumps. Full physical memory crash dumps are still available
via the debug.minidump sysctl and tunable.
Traditional dumps store all physical memory. This was once a good thing
when machines had a maximum of 64M of ram and 1GB of kvm. These days,
machines often have many gigabytes of ram and a smaller amount of kvm.
libkvm+kgdb don't have a way to access physical ram that is not mapped
into kvm at the time of the crash dump, so the extra ram being dumped
is mostly wasted.
Minidumps invert the process. Instead of dumping physical memory in
in order to guarantee that all of kvm's backing is dumped, minidumps
instead dump only memory that is actively mapped into kvm.
amd64 has a direct map region that things like UMA use. Obviously we
cannot dump all of the direct map region because that is effectively
an old style all-physical-memory dump. Instead, introduce a bitmap
and two helper routines (dump_add_page(pa) and dump_drop_page(pa)) that
allow certain critical direct map pages to be included in the dump.
uma_machdep.c's allocator is the intended consumer.
Dumps are a custom format. At the very beginning of the file is a header,
then a copy of the message buffer, then the bitmap of pages present in
the dump, then the final level of the kvm page table trees (2MB mappings
are expanded into a 4K page mappings), then the sparse physical pages
according to the bitmap. libkvm can now conveniently access the kvm
page table entries.
Booting my test 8GB machine, forcing it into ddb and forcing a dump
leads to a 48MB minidump. While this is a best case, I expect minidumps
to be in the 100MB-500MB range. Obviously, never larger than physical
memory of course.
minidumps are on by default. It would want be necessary to turn them off
if it was necessary to debug corrupt kernel page table management as that
would mess up minidumps as well.
Both minidumps and regular dumps are supported on the same machine.
2006-04-21 04:24:50 +00:00
|
|
|
/*
|
|
|
|
* Allocate a bitmap to indicate that a random physical page
|
|
|
|
* needs to be included in a minidump.
|
|
|
|
*
|
|
|
|
* The amd64 port needs this to indicate which direct map pages
|
|
|
|
* need to be dumped, via calls to dump_add_page()/dump_drop_page().
|
|
|
|
*
|
|
|
|
* However, i386 still needs this workspace internally within the
|
|
|
|
* minidump code. In theory, they are not needed on i386, but are
|
|
|
|
* included should the sf_buf code decide to use them.
|
|
|
|
*/
|
|
|
|
page_range = phys_avail[(nblocks - 1) * 2 + 1] / PAGE_SIZE;
|
|
|
|
vm_page_dump_size = round_page(roundup2(page_range, NBBY) / NBBY);
|
|
|
|
new_end -= vm_page_dump_size;
|
|
|
|
vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
|
|
|
|
new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
|
|
|
|
bzero((void *)vm_page_dump, vm_page_dump_size);
|
|
|
|
#endif
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* Compute the number of pages of memory that will be available for
|
|
|
|
* use (taking into account the overhead of a page structure per
|
|
|
|
* page).
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2006-12-08 08:44:47 +00:00
|
|
|
first_page = low_water / PAGE_SIZE;
|
2007-05-05 19:50:28 +00:00
|
|
|
#ifdef VM_PHYSSEG_SPARSE
|
|
|
|
page_range = 0;
|
|
|
|
for (i = 0; phys_avail[i + 1] != 0; i += 2)
|
|
|
|
page_range += atop(phys_avail[i + 1] - phys_avail[i]);
|
|
|
|
#elif defined(VM_PHYSSEG_DENSE)
|
2006-12-08 08:44:47 +00:00
|
|
|
page_range = high_water / PAGE_SIZE - first_page;
|
2007-05-05 19:50:28 +00:00
|
|
|
#else
|
|
|
|
#error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
|
|
|
|
#endif
|
2001-03-01 19:21:24 +00:00
|
|
|
end = new_end;
|
2001-03-07 05:29:21 +00:00
|
|
|
|
2003-12-22 02:04:08 +00:00
|
|
|
/*
|
|
|
|
* Reserve an unmapped guard page to trap access to vm_page_array[-1].
|
|
|
|
*/
|
|
|
|
vaddr += PAGE_SIZE;
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* Initialize the mem entry structures now, and put them in the free
|
|
|
|
* queue.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2001-03-01 19:21:24 +00:00
|
|
|
new_end = trunc_page(end - page_range * sizeof(struct vm_page));
|
2001-03-07 05:29:21 +00:00
|
|
|
mapped = pmap_map(&vaddr, new_end, end,
|
2001-03-01 19:21:24 +00:00
|
|
|
VM_PROT_READ | VM_PROT_WRITE);
|
2001-03-07 05:29:21 +00:00
|
|
|
vm_page_array = (vm_page_t) mapped;
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
|
|
|
/*
|
|
|
|
* Allocate memory for the reservation management system's data
|
|
|
|
* structures.
|
|
|
|
*/
|
|
|
|
new_end = vm_reserv_startup(&vaddr, new_end, high_water);
|
|
|
|
#endif
|
2006-05-31 22:55:23 +00:00
|
|
|
#ifdef __amd64__
|
|
|
|
/*
|
|
|
|
* pmap_map on amd64 comes out of the direct-map, not kvm like i386,
|
|
|
|
* so the pages must be tracked for a crashdump to include this data.
|
|
|
|
* This includes the vm_page_array and the early UMA bootstrap pages.
|
|
|
|
*/
|
|
|
|
for (pa = new_end; pa < phys_avail[biggestone + 1]; pa += PAGE_SIZE)
|
|
|
|
dump_add_page(pa);
|
|
|
|
#endif
|
2003-03-17 03:16:00 +00:00
|
|
|
phys_avail[biggestone + 1] = new_end;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2006-11-08 19:11:54 +00:00
|
|
|
/*
|
|
|
|
* Clear all of the page structures
|
|
|
|
*/
|
|
|
|
bzero((caddr_t) vm_page_array, page_range * sizeof(struct vm_page));
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
for (i = 0; i < page_range; i++)
|
|
|
|
vm_page_array[i].order = VM_NFREEORDER;
|
2006-11-08 19:11:54 +00:00
|
|
|
vm_page_array_size = page_range;
|
|
|
|
|
1999-03-19 05:21:03 +00:00
|
|
|
/*
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
* Initialize the physical memory allocator.
|
|
|
|
*/
|
|
|
|
vm_phys_init();
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Add every available physical page that is not blacklisted to
|
|
|
|
* the free lists.
|
1999-03-19 05:21:03 +00:00
|
|
|
*/
|
2007-05-31 22:52:15 +00:00
|
|
|
cnt.v_page_count = 0;
|
|
|
|
cnt.v_free_count = 0;
|
2006-06-23 16:44:24 +00:00
|
|
|
list = getenv("vm.blacklist");
|
2006-11-08 18:43:47 +00:00
|
|
|
for (i = 0; phys_avail[i + 1] != 0; i += 2) {
|
2001-03-01 19:21:24 +00:00
|
|
|
pa = phys_avail[i];
|
2003-03-17 03:16:00 +00:00
|
|
|
last_pa = phys_avail[i + 1];
|
2006-11-08 18:43:47 +00:00
|
|
|
while (pa < last_pa) {
|
2006-06-23 16:44:24 +00:00
|
|
|
if (list != NULL &&
|
|
|
|
vm_page_blacklist_lookup(list, pa))
|
|
|
|
printf("Skipping page with pa 0x%jx\n",
|
|
|
|
(uintmax_t)pa);
|
|
|
|
else
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
vm_phys_add_page(pa);
|
1994-05-25 09:21:21 +00:00
|
|
|
pa += PAGE_SIZE;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
}
|
2006-06-23 16:44:24 +00:00
|
|
|
freeenv(list);
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
|
|
|
/*
|
|
|
|
* Initialize the reservation management system.
|
|
|
|
*/
|
|
|
|
vm_reserv_init();
|
|
|
|
#endif
|
2001-03-07 05:29:21 +00:00
|
|
|
return (vaddr);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
2001-07-04 20:15:18 +00:00
|
|
|
void
|
|
|
|
vm_page_flag_set(vm_page_t m, unsigned short bits)
|
|
|
|
{
|
2002-12-19 07:23:46 +00:00
|
|
|
|
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2001-07-31 04:09:52 +00:00
|
|
|
m->flags |= bits;
|
2001-07-04 20:15:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
vm_page_flag_clear(vm_page_t m, unsigned short bits)
|
|
|
|
{
|
2002-12-28 22:49:37 +00:00
|
|
|
|
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2001-07-31 04:09:52 +00:00
|
|
|
m->flags &= ~bits;
|
2001-07-04 20:15:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
vm_page_busy(vm_page_t m)
|
|
|
|
{
|
2004-10-24 23:53:47 +00:00
|
|
|
|
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
2006-10-22 04:28:14 +00:00
|
|
|
KASSERT((m->oflags & VPO_BUSY) == 0,
|
2001-07-04 20:15:18 +00:00
|
|
|
("vm_page_busy: page already busy!!!"));
|
2006-10-22 04:28:14 +00:00
|
|
|
m->oflags |= VPO_BUSY;
|
2001-07-04 20:15:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_flash:
|
|
|
|
*
|
|
|
|
* wakeup anyone waiting for the page.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_flash(vm_page_t m)
|
|
|
|
{
|
2004-10-25 19:52:44 +00:00
|
|
|
|
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
2006-08-09 17:43:27 +00:00
|
|
|
if (m->oflags & VPO_WANTED) {
|
|
|
|
m->oflags &= ~VPO_WANTED;
|
2001-07-04 20:15:18 +00:00
|
|
|
wakeup(m);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_wakeup:
|
|
|
|
*
|
2006-10-22 04:28:14 +00:00
|
|
|
* clear the VPO_BUSY flag and wakeup anyone waiting for the
|
2001-07-04 20:15:18 +00:00
|
|
|
* page.
|
|
|
|
*
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_wakeup(vm_page_t m)
|
|
|
|
{
|
2004-10-24 23:53:47 +00:00
|
|
|
|
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
2006-10-22 04:28:14 +00:00
|
|
|
KASSERT(m->oflags & VPO_BUSY, ("vm_page_wakeup: page not busy!!!"));
|
|
|
|
m->oflags &= ~VPO_BUSY;
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_flash(m);
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
vm_page_io_start(vm_page_t m)
|
|
|
|
{
|
2002-07-31 07:27:08 +00:00
|
|
|
|
2004-10-17 22:33:40 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
2001-07-31 04:09:52 +00:00
|
|
|
m->busy++;
|
2001-07-04 20:15:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
vm_page_io_finish(vm_page_t m)
|
|
|
|
{
|
2002-08-01 17:57:42 +00:00
|
|
|
|
2004-10-17 22:33:40 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
2001-07-31 04:09:52 +00:00
|
|
|
m->busy--;
|
2001-07-04 20:15:18 +00:00
|
|
|
if (m->busy == 0)
|
|
|
|
vm_page_flash(m);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Keep page from being freed by the page daemon
|
|
|
|
* much of the same effect as wiring, except much lower
|
|
|
|
* overhead and should be used only for *very* temporary
|
|
|
|
* holding ("wiring").
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_hold(vm_page_t mem)
|
|
|
|
{
|
2003-01-20 09:24:03 +00:00
|
|
|
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock_assert(mem, MA_OWNED);
|
2001-07-04 20:15:18 +00:00
|
|
|
mem->hold_count++;
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
vm_page_unhold(vm_page_t mem)
|
|
|
|
{
|
2002-12-15 00:06:02 +00:00
|
|
|
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock_assert(mem, MA_OWNED);
|
2001-07-04 20:15:18 +00:00
|
|
|
--mem->hold_count;
|
|
|
|
KASSERT(mem->hold_count >= 0, ("vm_page_unhold: hold count < 0!!!"));
|
2010-05-06 16:39:43 +00:00
|
|
|
if (mem->hold_count == 0 && VM_PAGE_INQUEUE2(mem, PQ_HOLD))
|
2002-02-19 23:19:30 +00:00
|
|
|
vm_page_free_toq(mem);
|
2001-07-04 20:15:18 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_free:
|
|
|
|
*
|
2007-02-17 19:37:00 +00:00
|
|
|
* Free a page.
|
2001-07-04 20:15:18 +00:00
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_free(vm_page_t m)
|
|
|
|
{
|
2007-02-18 05:54:42 +00:00
|
|
|
|
|
|
|
m->flags &= ~PG_ZERO;
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_free_toq(m);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_free_zero:
|
|
|
|
*
|
|
|
|
* Free a page to the zerod-pages queue
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_free_zero(vm_page_t m)
|
|
|
|
{
|
2007-02-18 05:54:42 +00:00
|
|
|
|
|
|
|
m->flags |= PG_ZERO;
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_free_toq(m);
|
|
|
|
}
|
|
|
|
|
2002-07-29 19:41:22 +00:00
|
|
|
/*
|
2006-08-27 19:50:13 +00:00
|
|
|
* vm_page_sleep:
|
2002-07-29 19:41:22 +00:00
|
|
|
*
|
It makes no sense for vm_page_sleep_if_busy()'s helper, vm_page_sleep(),
to unconditionally set PG_REFERENCED on a page before sleeping. In many
cases, it's perfectly ok for the page to disappear, i.e., be reclaimed by
the page daemon, before the caller to vm_page_sleep() is reawakened.
Instead, we now explicitly set PG_REFERENCED in those cases where having
the page persist until the caller is awakened is clearly desirable. Note,
however, that setting PG_REFERENCED on the page is still only a hint,
and not a guarantee that the page should persist.
2010-05-02 17:33:46 +00:00
|
|
|
* Sleep and release the page and page queues locks.
|
2006-08-27 19:50:13 +00:00
|
|
|
*
|
|
|
|
* The object containing the given page must be locked.
|
2002-07-29 19:41:22 +00:00
|
|
|
*/
|
2006-08-27 19:50:13 +00:00
|
|
|
void
|
|
|
|
vm_page_sleep(vm_page_t m, const char *msg)
|
2002-07-29 19:41:22 +00:00
|
|
|
{
|
|
|
|
|
2004-12-26 21:46:44 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
It makes no sense for vm_page_sleep_if_busy()'s helper, vm_page_sleep(),
to unconditionally set PG_REFERENCED on a page before sleeping. In many
cases, it's perfectly ok for the page to disappear, i.e., be reclaimed by
the page daemon, before the caller to vm_page_sleep() is reawakened.
Instead, we now explicitly set PG_REFERENCED in those cases where having
the page persist until the caller is awakened is clearly desirable. Note,
however, that setting PG_REFERENCED on the page is still only a hint,
and not a guarantee that the page should persist.
2010-05-02 17:33:46 +00:00
|
|
|
if (mtx_owned(&vm_page_queue_mtx))
|
|
|
|
vm_page_unlock_queues();
|
|
|
|
if (mtx_owned(vm_page_lockptr(m)))
|
|
|
|
vm_page_unlock(m);
|
2006-08-27 19:50:13 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* It's possible that while we sleep, the page will get
|
|
|
|
* unbusied and freed. If we are holding the object
|
|
|
|
* lock, we will assume we hold a reference to the object
|
|
|
|
* such that even if m->object changes, we can re-lock
|
|
|
|
* it.
|
|
|
|
*/
|
|
|
|
m->oflags |= VPO_WANTED;
|
|
|
|
msleep(m, VM_OBJECT_MTX(m->object), PVM, msg, 0);
|
2002-07-29 19:41:22 +00:00
|
|
|
}
|
|
|
|
|
2001-07-04 20:15:18 +00:00
|
|
|
/*
|
|
|
|
* vm_page_dirty:
|
|
|
|
*
|
|
|
|
* make page all dirty
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_dirty(vm_page_t m)
|
|
|
|
{
|
2009-05-30 22:06:58 +00:00
|
|
|
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
KASSERT((m->flags & PG_CACHED) == 0,
|
2001-07-04 20:15:18 +00:00
|
|
|
("vm_page_dirty: page in cache!"));
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
KASSERT(!VM_PAGE_IS_FREE(m),
|
2003-02-24 17:30:45 +00:00
|
|
|
("vm_page_dirty: page is free!"));
|
2009-05-30 22:06:58 +00:00
|
|
|
KASSERT(m->valid == VM_PAGE_BITS_ALL,
|
|
|
|
("vm_page_dirty: page is invalid!"));
|
2001-07-04 20:15:18 +00:00
|
|
|
m->dirty = VM_PAGE_BITS_ALL;
|
|
|
|
}
|
|
|
|
|
2002-10-18 17:24:30 +00:00
|
|
|
/*
|
|
|
|
* vm_page_splay:
|
|
|
|
*
|
|
|
|
* Implements Sleator and Tarjan's top-down splay algorithm. Returns
|
|
|
|
* the vm_page containing the given pindex. If, however, that
|
|
|
|
* pindex is not found in the vm_object, returns a vm_page that is
|
|
|
|
* adjacent to the pindex, coming before or after it.
|
|
|
|
*/
|
2002-11-04 19:21:39 +00:00
|
|
|
vm_page_t
|
2002-10-18 17:24:30 +00:00
|
|
|
vm_page_splay(vm_pindex_t pindex, vm_page_t root)
|
|
|
|
{
|
|
|
|
struct vm_page dummy;
|
|
|
|
vm_page_t lefttreemax, righttreemin, y;
|
|
|
|
|
|
|
|
if (root == NULL)
|
|
|
|
return (root);
|
|
|
|
lefttreemax = righttreemin = &dummy;
|
|
|
|
for (;; root = y) {
|
|
|
|
if (pindex < root->pindex) {
|
|
|
|
if ((y = root->left) == NULL)
|
|
|
|
break;
|
|
|
|
if (pindex < y->pindex) {
|
|
|
|
/* Rotate right. */
|
|
|
|
root->left = y->right;
|
|
|
|
y->right = root;
|
|
|
|
root = y;
|
|
|
|
if ((y = root->left) == NULL)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/* Link into the new root's right tree. */
|
|
|
|
righttreemin->left = root;
|
|
|
|
righttreemin = root;
|
|
|
|
} else if (pindex > root->pindex) {
|
|
|
|
if ((y = root->right) == NULL)
|
|
|
|
break;
|
|
|
|
if (pindex > y->pindex) {
|
|
|
|
/* Rotate left. */
|
|
|
|
root->right = y->left;
|
|
|
|
y->left = root;
|
|
|
|
root = y;
|
|
|
|
if ((y = root->right) == NULL)
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/* Link into the new root's left tree. */
|
|
|
|
lefttreemax->right = root;
|
|
|
|
lefttreemax = root;
|
|
|
|
} else
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
/* Assemble the new root. */
|
|
|
|
lefttreemax->right = root->left;
|
|
|
|
righttreemin->left = root->right;
|
|
|
|
root->left = dummy.right;
|
|
|
|
root->right = dummy.left;
|
|
|
|
return (root);
|
|
|
|
}
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* vm_page_insert: [ internal use only ]
|
|
|
|
*
|
1998-12-23 01:52:47 +00:00
|
|
|
* Inserts the given mem entry into the object and object list.
|
|
|
|
*
|
|
|
|
* The pagetables are not updated but will presumably fault the page
|
|
|
|
* in if necessary, or if a kernel page the caller will at some point
|
|
|
|
* enter the page into the kernel's pmap. We are not allowed to block
|
|
|
|
* here so we *can't* do this anyway.
|
1994-05-24 10:09:53 +00:00
|
|
|
*
|
2004-06-19 04:19:47 +00:00
|
|
|
* The object and page must be locked.
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1996-09-08 20:44:49 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2002-10-18 17:24:30 +00:00
|
|
|
vm_page_t root;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2003-11-01 04:54:23 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
|
1998-10-21 14:46:42 +00:00
|
|
|
if (m->object != NULL)
|
2004-01-08 20:48:26 +00:00
|
|
|
panic("vm_page_insert: page already inserted");
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
/*
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* Record the object/offset pair in this page
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1996-01-19 04:00:31 +00:00
|
|
|
m->object = object;
|
|
|
|
m->pindex = pindex;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
/*
|
2002-10-18 17:24:30 +00:00
|
|
|
* Now link into the object's ordered list of backed pages.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2002-11-09 08:27:42 +00:00
|
|
|
root = object->root;
|
2002-10-18 17:24:30 +00:00
|
|
|
if (root == NULL) {
|
|
|
|
m->left = NULL;
|
|
|
|
m->right = NULL;
|
|
|
|
TAILQ_INSERT_TAIL(&object->memq, m, listq);
|
|
|
|
} else {
|
2002-11-09 08:27:42 +00:00
|
|
|
root = vm_page_splay(pindex, root);
|
|
|
|
if (pindex < root->pindex) {
|
|
|
|
m->left = root->left;
|
|
|
|
m->right = root;
|
|
|
|
root->left = NULL;
|
|
|
|
TAILQ_INSERT_BEFORE(root, m, listq);
|
2004-01-08 20:48:26 +00:00
|
|
|
} else if (pindex == root->pindex)
|
|
|
|
panic("vm_page_insert: offset already allocated");
|
|
|
|
else {
|
2002-11-09 08:27:42 +00:00
|
|
|
m->right = root->right;
|
|
|
|
m->left = root;
|
|
|
|
root->right = NULL;
|
|
|
|
TAILQ_INSERT_AFTER(&object->memq, root, m, listq);
|
|
|
|
}
|
2002-10-18 17:24:30 +00:00
|
|
|
}
|
|
|
|
object->root = m;
|
1999-07-31 04:19:49 +00:00
|
|
|
object->generation++;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
/*
|
1999-01-21 08:29:12 +00:00
|
|
|
* show that the object has one more resident page.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
|
|
|
object->resident_page_count++;
|
2005-03-15 14:14:09 +00:00
|
|
|
/*
|
|
|
|
* Hold the vnode until the last page is released.
|
|
|
|
*/
|
|
|
|
if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
|
|
|
|
vhold((struct vnode *)object->handle);
|
1999-02-24 21:26:26 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Since we are inserting a new and possibly dirty page,
|
2006-07-21 06:40:29 +00:00
|
|
|
* update the object's OBJ_MIGHTBEDIRTY flag.
|
1999-02-24 21:26:26 +00:00
|
|
|
*/
|
|
|
|
if (m->flags & PG_WRITEABLE)
|
2001-10-26 00:08:05 +00:00
|
|
|
vm_object_set_writeable_dirty(object);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
1999-01-21 08:29:12 +00:00
|
|
|
* vm_page_remove:
|
1994-05-24 10:09:53 +00:00
|
|
|
* NOTE: used by device pager as well -wfj
|
|
|
|
*
|
|
|
|
* Removes the given mem entry from the object/offset-page
|
1999-01-21 08:29:12 +00:00
|
|
|
* table and the object page list, but do not invalidate/terminate
|
|
|
|
* the backing store.
|
1994-05-24 10:09:53 +00:00
|
|
|
*
|
2004-06-19 04:19:47 +00:00
|
|
|
* The object and page must be locked.
|
1999-01-21 08:29:12 +00:00
|
|
|
* The underlying pmap entry (if any) is NOT removed here.
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1999-02-15 06:52:14 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_remove(vm_page_t m)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
1998-02-05 03:32:49 +00:00
|
|
|
vm_object_t object;
|
2002-10-18 17:24:30 +00:00
|
|
|
vm_page_t root;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2010-05-03 07:00:50 +00:00
|
|
|
if ((m->flags & PG_UNMANAGED) == 0)
|
2010-05-05 18:16:06 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2004-11-03 20:17:31 +00:00
|
|
|
if ((object = m->object) == NULL)
|
1999-02-15 06:52:14 +00:00
|
|
|
return;
|
2004-11-03 20:17:31 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
|
2006-10-22 04:28:14 +00:00
|
|
|
if (m->oflags & VPO_BUSY) {
|
|
|
|
m->oflags &= ~VPO_BUSY;
|
2004-11-03 20:17:31 +00:00
|
|
|
vm_page_flash(m);
|
1998-01-31 11:56:53 +00:00
|
|
|
}
|
1999-01-21 08:29:12 +00:00
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* Now remove from the object's list of backed pages.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2002-10-18 17:24:30 +00:00
|
|
|
if (m != object->root)
|
|
|
|
vm_page_splay(m->pindex, object->root);
|
|
|
|
if (m->left == NULL)
|
|
|
|
root = m->right;
|
|
|
|
else {
|
|
|
|
root = vm_page_splay(m->pindex, m->left);
|
|
|
|
root->right = m->right;
|
|
|
|
}
|
|
|
|
object->root = root;
|
1998-02-05 03:32:49 +00:00
|
|
|
TAILQ_REMOVE(&object->memq, m, listq);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
/*
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
* And show that the object has one fewer resident page.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-02-05 03:32:49 +00:00
|
|
|
object->resident_page_count--;
|
|
|
|
object->generation++;
|
2005-03-15 14:14:09 +00:00
|
|
|
/*
|
|
|
|
* The vnode may now be recycled.
|
|
|
|
*/
|
|
|
|
if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
|
|
|
|
vdrop((struct vnode *)object->handle);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1998-10-21 14:46:42 +00:00
|
|
|
m->object = NULL;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_lookup:
|
|
|
|
*
|
|
|
|
* Returns the page associated with the object/offset
|
|
|
|
* pair specified; if none is found, NULL is returned.
|
|
|
|
*
|
2002-10-18 17:24:30 +00:00
|
|
|
* The object must be locked.
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
1999-01-21 08:29:12 +00:00
|
|
|
* This is a critical path routine
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1995-05-30 08:16:23 +00:00
|
|
|
vm_page_t
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2001-07-04 19:00:13 +00:00
|
|
|
vm_page_t m;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2003-08-28 23:23:04 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
|
2003-12-31 19:02:01 +00:00
|
|
|
if ((m = object->root) != NULL && m->pindex != pindex) {
|
|
|
|
m = vm_page_splay(pindex, m);
|
|
|
|
if ((object->root = m)->pindex != pindex)
|
|
|
|
m = NULL;
|
|
|
|
}
|
2002-04-30 21:24:47 +00:00
|
|
|
return (m);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_rename:
|
|
|
|
*
|
|
|
|
* Move the given memory entry from its
|
|
|
|
* current object to the specified target object/offset.
|
|
|
|
*
|
|
|
|
* The object must be locked.
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
|
|
|
*
|
1999-01-21 08:29:12 +00:00
|
|
|
* Note: swap associated with the page must be invalidated by the move. We
|
|
|
|
* have to do this for several reasons: (1) we aren't freeing the
|
|
|
|
* page, (2) we are dirtying the page, (3) the VM system is probably
|
|
|
|
* moving the page from object A to B, and will then later move
|
|
|
|
* the backing store from A to B and we can't have a conflict.
|
|
|
|
*
|
|
|
|
* Note: we *always* dirty the page. It is necessary both for the
|
|
|
|
* fact that we moved it, and because we may be invalidating
|
1999-01-24 06:00:31 +00:00
|
|
|
* swap. If the page is on the cache, we have to deactivate it
|
|
|
|
* or vm_page_dirty() will panic. Dirty pages are not allowed
|
|
|
|
* on the cache.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1995-05-30 08:16:23 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
1996-01-19 04:00:31 +00:00
|
|
|
vm_page_remove(m);
|
|
|
|
vm_page_insert(m, new_object, new_pindex);
|
1999-01-24 06:00:31 +00:00
|
|
|
vm_page_dirty(m);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
|
|
|
/*
|
2007-09-27 04:21:59 +00:00
|
|
|
* Convert all of the given object's cached pages that have a
|
|
|
|
* pindex within the given range into free pages. If the value
|
|
|
|
* zero is given for "end", then the range's upper bound is
|
|
|
|
* infinity. If the given object is backed by a vnode and it
|
|
|
|
* transitions from having one or more cached pages to none, the
|
|
|
|
* vnode's hold count is reduced.
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
*/
|
|
|
|
void
|
2007-09-27 04:21:59 +00:00
|
|
|
vm_page_cache_free(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
{
|
2007-09-27 04:21:59 +00:00
|
|
|
vm_page_t m, m_next;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
boolean_t empty;
|
|
|
|
|
|
|
|
mtx_lock(&vm_page_queue_free_mtx);
|
2007-09-27 04:21:59 +00:00
|
|
|
if (__predict_false(object->cache == NULL)) {
|
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
m = object->cache = vm_page_splay(start, object->cache);
|
|
|
|
if (m->pindex < start) {
|
|
|
|
if (m->right == NULL)
|
|
|
|
m = NULL;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
else {
|
2007-09-27 04:21:59 +00:00
|
|
|
m_next = vm_page_splay(start, m->right);
|
|
|
|
m_next->left = m;
|
|
|
|
m->right = NULL;
|
|
|
|
m = object->cache = m_next;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* At this point, "m" is either (1) a reference to the page
|
|
|
|
* with the least pindex that is greater than or equal to
|
|
|
|
* "start" or (2) NULL.
|
|
|
|
*/
|
|
|
|
for (; m != NULL && (m->pindex < end || end == 0); m = m_next) {
|
|
|
|
/*
|
|
|
|
* Find "m"'s successor and remove "m" from the
|
|
|
|
* object's cache.
|
|
|
|
*/
|
|
|
|
if (m->right == NULL) {
|
|
|
|
object->cache = m->left;
|
|
|
|
m_next = NULL;
|
|
|
|
} else {
|
|
|
|
m_next = vm_page_splay(start, m->right);
|
|
|
|
m_next->left = m->left;
|
|
|
|
object->cache = m_next;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
}
|
2007-09-27 04:21:59 +00:00
|
|
|
/* Convert "m" to a free page. */
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
m->object = NULL;
|
|
|
|
m->valid = 0;
|
|
|
|
/* Clear PG_CACHED and set PG_FREE. */
|
|
|
|
m->flags ^= PG_CACHED | PG_FREE;
|
|
|
|
KASSERT((m->flags & (PG_CACHED | PG_FREE)) == PG_FREE,
|
|
|
|
("vm_page_cache_free: page %p has inconsistent flags", m));
|
|
|
|
cnt.v_cache_count--;
|
|
|
|
cnt.v_free_count++;
|
|
|
|
}
|
2007-09-27 04:21:59 +00:00
|
|
|
empty = object->cache == NULL;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
2007-09-27 04:21:59 +00:00
|
|
|
if (object->type == OBJT_VNODE && empty)
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
vdrop(object->handle);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Returns the cached page that is associated with the given
|
|
|
|
* object and offset. If, however, none exists, returns NULL.
|
1998-12-23 01:52:47 +00:00
|
|
|
*
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
* The free page queue must be locked.
|
|
|
|
*/
|
|
|
|
static inline vm_page_t
|
|
|
|
vm_page_cache_lookup(vm_object_t object, vm_pindex_t pindex)
|
|
|
|
{
|
|
|
|
vm_page_t m;
|
|
|
|
|
|
|
|
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
|
|
|
|
if ((m = object->cache) != NULL && m->pindex != pindex) {
|
|
|
|
m = vm_page_splay(pindex, m);
|
|
|
|
if ((object->cache = m)->pindex != pindex)
|
|
|
|
m = NULL;
|
|
|
|
}
|
|
|
|
return (m);
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Remove the given cached page from its containing object's
|
|
|
|
* collection of cached pages.
|
1998-12-23 01:52:47 +00:00
|
|
|
*
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
* The free page queue must be locked.
|
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
|
|
|
*/
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
void
|
|
|
|
vm_page_cache_remove(vm_page_t m)
|
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
|
|
|
{
|
2005-01-07 05:02:19 +00:00
|
|
|
vm_object_t object;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
vm_page_t root;
|
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
|
|
|
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
|
|
|
|
KASSERT((m->flags & PG_CACHED) != 0,
|
|
|
|
("vm_page_cache_remove: page %p is not cached", m));
|
|
|
|
object = m->object;
|
|
|
|
if (m != object->cache) {
|
|
|
|
root = vm_page_splay(m->pindex, object->cache);
|
|
|
|
KASSERT(root == m,
|
|
|
|
("vm_page_cache_remove: page %p is not cached in object %p",
|
|
|
|
m, object));
|
|
|
|
}
|
|
|
|
if (m->left == NULL)
|
|
|
|
root = m->right;
|
|
|
|
else if (m->right == NULL)
|
|
|
|
root = m->left;
|
|
|
|
else {
|
|
|
|
root = vm_page_splay(m->pindex, m->left);
|
|
|
|
root->right = m->right;
|
|
|
|
}
|
|
|
|
object->cache = root;
|
|
|
|
m->object = NULL;
|
|
|
|
cnt.v_cache_count--;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Transfer all of the cached pages with offset greater than or
|
|
|
|
* equal to 'offidxstart' from the original object's cache to the
|
2007-10-27 00:09:30 +00:00
|
|
|
* new object's cache. However, any cached pages with offset
|
|
|
|
* greater than or equal to the new object's size are kept in the
|
|
|
|
* original object. Initially, the new object's cache must be
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
* empty. Offset 'offidxstart' in the original object must
|
|
|
|
* correspond to offset zero in the new object.
|
|
|
|
*
|
|
|
|
* The new object must be locked.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_cache_transfer(vm_object_t orig_object, vm_pindex_t offidxstart,
|
|
|
|
vm_object_t new_object)
|
|
|
|
{
|
|
|
|
vm_page_t m, m_next;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Insertion into an object's collection of cached pages
|
|
|
|
* requires the object to be locked. In contrast, removal does
|
|
|
|
* not.
|
|
|
|
*/
|
|
|
|
VM_OBJECT_LOCK_ASSERT(new_object, MA_OWNED);
|
|
|
|
KASSERT(new_object->cache == NULL,
|
|
|
|
("vm_page_cache_transfer: object %p has cached pages",
|
|
|
|
new_object));
|
|
|
|
mtx_lock(&vm_page_queue_free_mtx);
|
|
|
|
if ((m = orig_object->cache) != NULL) {
|
|
|
|
/*
|
|
|
|
* Transfer all of the pages with offset greater than or
|
|
|
|
* equal to 'offidxstart' from the original object's
|
|
|
|
* cache to the new object's cache.
|
|
|
|
*/
|
|
|
|
m = vm_page_splay(offidxstart, m);
|
|
|
|
if (m->pindex < offidxstart) {
|
|
|
|
orig_object->cache = m;
|
|
|
|
new_object->cache = m->right;
|
|
|
|
m->right = NULL;
|
|
|
|
} else {
|
|
|
|
orig_object->cache = m->left;
|
|
|
|
new_object->cache = m;
|
|
|
|
m->left = NULL;
|
|
|
|
}
|
|
|
|
while ((m = new_object->cache) != NULL) {
|
2007-10-27 00:09:30 +00:00
|
|
|
if ((m->pindex - offidxstart) >= new_object->size) {
|
|
|
|
/*
|
|
|
|
* Return all of the cached pages with
|
|
|
|
* offset greater than or equal to the
|
|
|
|
* new object's size to the original
|
|
|
|
* object's cache.
|
|
|
|
*/
|
|
|
|
new_object->cache = m->left;
|
|
|
|
m->left = orig_object->cache;
|
|
|
|
orig_object->cache = m;
|
|
|
|
break;
|
|
|
|
}
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
m_next = vm_page_splay(m->pindex, m->right);
|
2007-10-27 00:09:30 +00:00
|
|
|
/* Update the page's object and offset. */
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
m->object = new_object;
|
|
|
|
m->pindex -= offidxstart;
|
|
|
|
if (m_next == NULL)
|
|
|
|
break;
|
|
|
|
m->right = NULL;
|
|
|
|
m_next->left = m;
|
|
|
|
new_object->cache = m_next;
|
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
|
|
|
}
|
2007-10-27 00:09:30 +00:00
|
|
|
KASSERT(new_object->cache == NULL ||
|
|
|
|
new_object->type == OBJT_SWAP,
|
|
|
|
("vm_page_cache_transfer: object %p's type is incompatible"
|
|
|
|
" with cached pages", new_object));
|
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
|
|
|
}
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
This mega-commit is meant to fix numerous interrelated problems. There
has been some bitrot and incorrect assumptions in the vfs_bio code. These
problems have manifest themselves worse on NFS type filesystems, but can
still affect local filesystems under certain circumstances. Most of
the problems have involved mmap consistancy, and as a side-effect broke
the vfs.ioopt code. This code might have been committed seperately, but
almost everything is interrelated.
1) Allow (pmap_object_init_pt) prefaulting of buffer-busy pages that
are fully valid.
2) Rather than deactivating erroneously read initial (header) pages in
kern_exec, we now free them.
3) Fix the rundown of non-VMIO buffers that are in an inconsistent
(missing vp) state.
4) Fix the disassociation of pages from buffers in brelse. The previous
code had rotted and was faulty in a couple of important circumstances.
5) Remove a gratuitious buffer wakeup in vfs_vmio_release.
6) Remove a crufty and currently unused cluster mechanism for VBLK
files in vfs_bio_awrite. When the code is functional, I'll add back
a cleaner version.
7) The page busy count wakeups assocated with the buffer cache usage were
incorrectly cleaned up in a previous commit by me. Revert to the
original, correct version, but with a cleaner implementation.
8) The cluster read code now tries to keep data associated with buffers
more aggressively (without breaking the heuristics) when it is presumed
that the read data (buffers) will be soon needed.
9) Change to filesystem lockmgr locks so that they use LK_NOPAUSE. The
delay loop waiting is not useful for filesystem locks, due to the
length of the time intervals.
10) Correct and clean-up spec_getpages.
11) Implement a fully functional nfs_getpages, nfs_putpages.
12) Fix nfs_write so that modifications are coherent with the NFS data on
the server disk (at least as well as NFS seems to allow.)
13) Properly support MS_INVALIDATE on NFS.
14) Properly pass down MS_INVALIDATE to lower levels of the VM code from
vm_map_clean.
15) Better support the notion of pages being busy but valid, so that
fewer in-transit waits occur. (use p->busy more for pageouts instead
of PG_BUSY.) Since the page is fully valid, it is still usable for
reads.
16) It is possible (in error) for cached pages to be busy. Make the
page allocation code handle that case correctly. (It should probably
be a printf or panic, but I want the system to handle coding errors
robustly. I'll probably add a printf.)
17) Correct the design and usage of vm_page_sleep. It didn't handle
consistancy problems very well, so make the design a little less
lofty. After vm_page_sleep, if it ever blocked, it is still important
to relookup the page (if the object generation count changed), and
verify it's status (always.)
18) In vm_pageout.c, vm_pageout_clean had rotted, so clean that up.
19) Push the page busy for writes and VM_PROT_READ into vm_pageout_flush.
20) Fix vm_pager_put_pages and it's descendents to support an int flag
instead of a boolean, so that we can pass down the invalidate bit.
1998-03-07 21:37:31 +00:00
|
|
|
}
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* vm_page_alloc:
|
|
|
|
*
|
|
|
|
* Allocate and return a memory cell associated
|
|
|
|
* with this VM object/offset pair.
|
|
|
|
*
|
1995-03-01 23:30:04 +00:00
|
|
|
* page_req classes:
|
|
|
|
* VM_ALLOC_NORMAL normal process request
|
|
|
|
* VM_ALLOC_SYSTEM system *really* needs a page
|
|
|
|
* VM_ALLOC_INTERRUPT interrupt time request
|
1995-09-03 20:40:43 +00:00
|
|
|
* VM_ALLOC_ZERO zero page
|
2010-02-27 17:09:28 +00:00
|
|
|
* VM_ALLOC_WIRED wire the allocated page
|
|
|
|
* VM_ALLOC_NOOBJ page is not associated with a vm object
|
|
|
|
* VM_ALLOC_NOBUSY do not set the page busy
|
|
|
|
* VM_ALLOC_IFNOTCACHED return NULL, do not reactivate if the page
|
|
|
|
* is cached
|
1995-01-24 10:14:09 +00:00
|
|
|
*
|
2010-02-27 17:09:28 +00:00
|
|
|
* This routine may not sleep.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1994-05-25 09:21:21 +00:00
|
|
|
vm_page_t
|
2002-07-18 04:08:10 +00:00
|
|
|
vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
struct vnode *vp = NULL;
|
|
|
|
vm_object_t m_object;
|
|
|
|
vm_page_t m;
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
int flags, page_req;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2003-01-11 20:07:17 +00:00
|
|
|
page_req = req & VM_ALLOC_CLASS_MASK;
|
2004-12-31 19:50:45 +00:00
|
|
|
KASSERT(curthread->td_intr_nesting_level == 0 ||
|
|
|
|
page_req == VM_ALLOC_INTERRUPT,
|
|
|
|
("vm_page_alloc(NORMAL|SYSTEM) in interrupt context"));
|
2003-01-11 20:07:17 +00:00
|
|
|
|
2002-11-01 00:59:03 +00:00
|
|
|
if ((req & VM_ALLOC_NOOBJ) == 0) {
|
|
|
|
KASSERT(object != NULL,
|
|
|
|
("vm_page_alloc: NULL object."));
|
2003-08-20 20:24:29 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
}
|
2002-11-01 00:59:03 +00:00
|
|
|
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* The pager is allowed to eat deeper into the free page list.
|
|
|
|
*/
|
1995-03-01 23:30:04 +00:00
|
|
|
if ((curproc == pageproc) && (page_req != VM_ALLOC_INTERRUPT)) {
|
|
|
|
page_req = VM_ALLOC_SYSTEM;
|
|
|
|
};
|
1995-05-30 08:16:23 +00:00
|
|
|
|
2007-02-05 06:02:55 +00:00
|
|
|
mtx_lock(&vm_page_queue_free_mtx);
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
if (cnt.v_free_count + cnt.v_cache_count > cnt.v_free_reserved ||
|
1999-02-15 06:52:14 +00:00
|
|
|
(page_req == VM_ALLOC_SYSTEM &&
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
cnt.v_free_count + cnt.v_cache_count > cnt.v_interrupt_free_min) ||
|
|
|
|
(page_req == VM_ALLOC_INTERRUPT &&
|
|
|
|
cnt.v_free_count + cnt.v_cache_count > 0)) {
|
1999-02-15 06:52:14 +00:00
|
|
|
/*
|
2003-01-08 19:58:42 +00:00
|
|
|
* Allocate from the free queue if the number of free pages
|
|
|
|
* exceeds the minimum for the request class.
|
1999-02-15 06:52:14 +00:00
|
|
|
*/
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
if (object != NULL &&
|
|
|
|
(m = vm_page_cache_lookup(object, pindex)) != NULL) {
|
|
|
|
if ((req & VM_ALLOC_IFNOTCACHED) != 0) {
|
2007-02-05 06:02:55 +00:00
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
2006-02-15 22:29:53 +00:00
|
|
|
return (NULL);
|
|
|
|
}
|
2007-12-20 22:45:54 +00:00
|
|
|
if (vm_phys_unfree_page(m))
|
|
|
|
vm_phys_set_pool(VM_FREEPOOL_DEFAULT, m, 0);
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
|
|
|
else if (!vm_reserv_reactivate_page(m))
|
|
|
|
#else
|
2007-12-20 22:45:54 +00:00
|
|
|
else
|
2007-12-29 19:53:04 +00:00
|
|
|
#endif
|
2007-12-20 22:45:54 +00:00
|
|
|
panic("vm_page_alloc: cache page %p is missing"
|
|
|
|
" from the free queue", m);
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
} else if ((req & VM_ALLOC_IFCACHED) != 0) {
|
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
|
|
|
return (NULL);
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
2008-05-15 18:52:31 +00:00
|
|
|
} else if (object == NULL || object->type == OBJT_DEVICE ||
|
2010-04-05 06:23:31 +00:00
|
|
|
object->type == OBJT_SG ||
|
2007-12-29 19:53:04 +00:00
|
|
|
(object->flags & OBJ_COLORED) == 0 ||
|
|
|
|
(m = vm_reserv_alloc_page(object, pindex)) == NULL) {
|
|
|
|
#else
|
|
|
|
} else {
|
|
|
|
#endif
|
2007-07-14 21:21:17 +00:00
|
|
|
m = vm_phys_alloc_pages(object != NULL ?
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT, 0);
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
2008-04-06 18:09:28 +00:00
|
|
|
if (m == NULL && vm_reserv_reclaim_inactive()) {
|
2007-12-29 19:53:04 +00:00
|
|
|
m = vm_phys_alloc_pages(object != NULL ?
|
|
|
|
VM_FREEPOOL_DEFAULT : VM_FREEPOOL_DIRECT,
|
|
|
|
0);
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
}
|
1999-02-15 06:52:14 +00:00
|
|
|
} else {
|
|
|
|
/*
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
* Not allocatable, give up.
|
1999-02-15 06:52:14 +00:00
|
|
|
*/
|
2007-02-05 06:02:55 +00:00
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
2003-01-14 06:57:03 +00:00
|
|
|
atomic_add_int(&vm_pageout_deficit, 1);
|
1999-02-15 06:52:14 +00:00
|
|
|
pagedaemon_wakeup();
|
|
|
|
return (NULL);
|
1995-03-01 23:30:04 +00:00
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
1999-02-15 06:52:14 +00:00
|
|
|
/*
|
|
|
|
* At this point we had better have found a good page.
|
|
|
|
*/
|
|
|
|
|
2009-06-21 00:21:33 +00:00
|
|
|
KASSERT(m != NULL, ("vm_page_alloc: missing page"));
|
2009-07-12 23:31:20 +00:00
|
|
|
KASSERT(m->queue == PQ_NONE,
|
|
|
|
("vm_page_alloc: page %p has unexpected queue %d", m, m->queue));
|
2009-06-21 00:21:33 +00:00
|
|
|
KASSERT(m->wire_count == 0, ("vm_page_alloc: page %p is wired", m));
|
|
|
|
KASSERT(m->hold_count == 0, ("vm_page_alloc: page %p is held", m));
|
|
|
|
KASSERT(m->busy == 0, ("vm_page_alloc: page %p is busy", m));
|
|
|
|
KASSERT(m->dirty == 0, ("vm_page_alloc: page %p is dirty", m));
|
2009-07-12 23:31:20 +00:00
|
|
|
KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
|
|
|
|
("vm_page_alloc: page %p has unexpected memattr %d", m,
|
|
|
|
pmap_page_get_memattr(m)));
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
if ((m->flags & PG_CACHED) != 0) {
|
|
|
|
KASSERT(m->valid != 0,
|
|
|
|
("vm_page_alloc: cached page %p is invalid", m));
|
|
|
|
if (m->object == object && m->pindex == pindex)
|
|
|
|
cnt.v_reactivated++;
|
|
|
|
else
|
|
|
|
m->valid = 0;
|
|
|
|
m_object = m->object;
|
|
|
|
vm_page_cache_remove(m);
|
|
|
|
if (m_object->type == OBJT_VNODE && m_object->cache == NULL)
|
|
|
|
vp = m_object->handle;
|
|
|
|
} else {
|
|
|
|
KASSERT(VM_PAGE_IS_FREE(m),
|
|
|
|
("vm_page_alloc: page %p is not free", m));
|
|
|
|
KASSERT(m->valid == 0,
|
|
|
|
("vm_page_alloc: free page %p is valid", m));
|
|
|
|
cnt.v_free_count--;
|
|
|
|
}
|
1999-01-21 08:29:12 +00:00
|
|
|
|
1999-02-15 06:52:14 +00:00
|
|
|
/*
|
|
|
|
* Initialize structure. Only the PG_ZERO flag is inherited.
|
|
|
|
*/
|
2006-10-22 04:28:14 +00:00
|
|
|
flags = 0;
|
1999-02-08 00:37:36 +00:00
|
|
|
if (m->flags & PG_ZERO) {
|
1999-01-21 08:29:12 +00:00
|
|
|
vm_page_zero_count--;
|
2003-01-12 23:32:46 +00:00
|
|
|
if (req & VM_ALLOC_ZERO)
|
2006-10-22 04:28:14 +00:00
|
|
|
flags = PG_ZERO;
|
1996-01-19 04:00:31 +00:00
|
|
|
}
|
2007-10-07 18:03:03 +00:00
|
|
|
if (object == NULL || object->type == OBJT_PHYS)
|
2007-02-25 06:14:58 +00:00
|
|
|
flags |= PG_UNMANAGED;
|
2003-01-12 23:32:46 +00:00
|
|
|
m->flags = flags;
|
2006-10-22 04:28:14 +00:00
|
|
|
if (req & (VM_ALLOC_NOBUSY | VM_ALLOC_NOOBJ))
|
|
|
|
m->oflags = 0;
|
|
|
|
else
|
|
|
|
m->oflags = VPO_BUSY;
|
2002-07-18 04:08:10 +00:00
|
|
|
if (req & VM_ALLOC_WIRED) {
|
2007-05-31 22:52:15 +00:00
|
|
|
atomic_add_int(&cnt.v_wire_count, 1);
|
2002-07-18 04:08:10 +00:00
|
|
|
m->wire_count = 1;
|
2009-06-21 00:21:33 +00:00
|
|
|
}
|
This commit does a couple of things:
Re-enables the RSS limiting, and the routine is now tail-recursive,
making it much more safe (eliminates the possiblity of kernel stack
overflow.) Also, the RSS limiting is a little more intelligent about
finding the likely objects that are pushing the process over the limit.
Added some sysctls that help with VM system tuning.
New sysctl features:
1) Enable/disable lru pageout algorithm.
vm.pageout_algorithm = 0, default algorithm that works
well, especially using X windows and heavy
memory loading. Can have adverse effects,
sometimes slowing down program loading.
vm.pageout_algorithm = 1, close to true LRU. Works much
better than clock, etc. Does not work as well as
the default algorithm in general. Certain memory
"malloc" type benchmarks work a little better with
this setting.
Please give me feedback on the performance results
associated with these.
2) Enable/disable swapping.
vm.swapping_enabled = 1, default.
vm.swapping_enabled = 0, useful for cases where swapping
degrades performance.
The config option "NO_SWAPPING" is still operative, and
takes precedence over the sysctl. If "NO_SWAPPING" is
specified, the sysctl still exists, but "vm.swapping_enabled"
is hard-wired to "0".
Each of these can be changed "on the fly."
1996-06-26 05:39:27 +00:00
|
|
|
m->act_count = 0;
|
2007-02-05 06:02:55 +00:00
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
1995-01-10 09:19:52 +00:00
|
|
|
|
2009-07-12 23:31:20 +00:00
|
|
|
if (object != NULL) {
|
2009-07-18 01:50:05 +00:00
|
|
|
/* Ignore device objects; the pager sets "memattr" for them. */
|
|
|
|
if (object->memattr != VM_MEMATTR_DEFAULT &&
|
2009-07-24 13:50:29 +00:00
|
|
|
object->type != OBJT_DEVICE && object->type != OBJT_SG)
|
2009-07-12 23:31:20 +00:00
|
|
|
pmap_page_set_memattr(m, object->memattr);
|
2002-11-01 00:59:03 +00:00
|
|
|
vm_page_insert(m, object, pindex);
|
2009-07-12 23:31:20 +00:00
|
|
|
} else
|
2003-09-22 00:56:13 +00:00
|
|
|
m->pindex = pindex;
|
1995-01-10 09:19:52 +00:00
|
|
|
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
/*
|
|
|
|
* The following call to vdrop() must come after the above call
|
|
|
|
* to vm_page_insert() in case both affect the same object and
|
|
|
|
* vnode. Otherwise, the affected vnode's hold count could
|
|
|
|
* temporarily become zero.
|
|
|
|
*/
|
|
|
|
if (vp != NULL)
|
|
|
|
vdrop(vp);
|
|
|
|
|
1995-03-01 23:30:04 +00:00
|
|
|
/*
|
|
|
|
* Don't wakeup too often - wakeup the pageout daemon when
|
|
|
|
* we would be nearly out of memory.
|
|
|
|
*/
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
if (vm_paging_needed())
|
1995-03-01 23:30:04 +00:00
|
|
|
pagedaemon_wakeup();
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1996-01-19 04:00:31 +00:00
|
|
|
return (m);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
1998-12-23 01:52:47 +00:00
|
|
|
/*
|
|
|
|
* vm_wait: (also see VM_WAIT macro)
|
|
|
|
*
|
|
|
|
* Block until free pages are available for allocation
|
2002-02-19 18:34:02 +00:00
|
|
|
* - Called in various places before memory allocations.
|
1998-12-23 01:52:47 +00:00
|
|
|
*/
|
1996-11-28 23:15:07 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_wait(void)
|
1996-11-28 23:15:07 +00:00
|
|
|
{
|
|
|
|
|
2007-02-07 06:37:30 +00:00
|
|
|
mtx_lock(&vm_page_queue_free_mtx);
|
1996-11-28 23:15:07 +00:00
|
|
|
if (curproc == pageproc) {
|
|
|
|
vm_pageout_pages_needed = 1;
|
2007-02-07 06:37:30 +00:00
|
|
|
msleep(&vm_pageout_pages_needed, &vm_page_queue_free_mtx,
|
2003-02-01 21:18:16 +00:00
|
|
|
PDROP | PSWP, "VMWait", 0);
|
1996-11-28 23:15:07 +00:00
|
|
|
} else {
|
|
|
|
if (!vm_pages_needed) {
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
vm_pages_needed = 1;
|
1996-11-28 23:15:07 +00:00
|
|
|
wakeup(&vm_pages_needed);
|
|
|
|
}
|
2007-05-31 22:52:15 +00:00
|
|
|
msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PVM,
|
|
|
|
"vmwait", 0);
|
1996-11-28 23:15:07 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2002-02-19 18:34:02 +00:00
|
|
|
/*
|
|
|
|
* vm_waitpfault: (also see VM_WAITPFAULT macro)
|
|
|
|
*
|
|
|
|
* Block until free pages are available for allocation
|
|
|
|
* - Called only in vm_fault so that processes page faulting
|
|
|
|
* can be easily tracked.
|
2002-02-19 18:50:49 +00:00
|
|
|
* - Sleeps at a lower priority than vm_wait() so that vm_wait()ing
|
|
|
|
* processes will be able to grab memory first. Do not change
|
|
|
|
* this balance without careful testing first.
|
2002-02-19 18:34:02 +00:00
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_waitpfault(void)
|
|
|
|
{
|
|
|
|
|
2007-02-07 06:37:30 +00:00
|
|
|
mtx_lock(&vm_page_queue_free_mtx);
|
2002-02-19 18:34:02 +00:00
|
|
|
if (!vm_pages_needed) {
|
|
|
|
vm_pages_needed = 1;
|
|
|
|
wakeup(&vm_pages_needed);
|
|
|
|
}
|
2007-05-31 22:52:15 +00:00
|
|
|
msleep(&cnt.v_free_count, &vm_page_queue_free_mtx, PDROP | PUSER,
|
2003-02-01 21:18:16 +00:00
|
|
|
"pfault", 0);
|
2002-02-19 18:34:02 +00:00
|
|
|
}
|
|
|
|
|
2008-03-18 06:52:15 +00:00
|
|
|
/*
|
2008-03-19 20:24:35 +00:00
|
|
|
* vm_page_requeue:
|
2008-03-18 06:52:15 +00:00
|
|
|
*
|
|
|
|
* If the given page is contained within a page queue, move it to the tail
|
|
|
|
* of that queue.
|
|
|
|
*
|
|
|
|
* The page queues must be locked.
|
|
|
|
*/
|
|
|
|
void
|
2008-03-19 20:24:35 +00:00
|
|
|
vm_page_requeue(vm_page_t m)
|
2008-03-18 06:52:15 +00:00
|
|
|
{
|
|
|
|
int queue = VM_PAGE_GETQUEUE(m);
|
|
|
|
struct vpgqueues *vpq;
|
|
|
|
|
|
|
|
if (queue != PQ_NONE) {
|
|
|
|
vpq = &vm_page_queues[queue];
|
|
|
|
TAILQ_REMOVE(&vpq->pl, m, pageq);
|
|
|
|
TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_pageq_remove:
|
|
|
|
*
|
|
|
|
* Remove a page from its queue.
|
|
|
|
*
|
|
|
|
* The queue containing the given page must be locked.
|
|
|
|
* This routine may not block.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_pageq_remove(vm_page_t m)
|
|
|
|
{
|
|
|
|
int queue = VM_PAGE_GETQUEUE(m);
|
|
|
|
struct vpgqueues *pq;
|
|
|
|
|
|
|
|
if (queue != PQ_NONE) {
|
|
|
|
VM_PAGE_SETQUEUE2(m, PQ_NONE);
|
|
|
|
pq = &vm_page_queues[queue];
|
|
|
|
TAILQ_REMOVE(&pq->pl, m, pageq);
|
|
|
|
(*pq->cnt)--;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_enqueue:
|
|
|
|
*
|
|
|
|
* Add the given page to the specified queue.
|
|
|
|
*
|
|
|
|
* The page queues must be locked.
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
vm_page_enqueue(int queue, vm_page_t m)
|
|
|
|
{
|
|
|
|
struct vpgqueues *vpq;
|
|
|
|
|
|
|
|
vpq = &vm_page_queues[queue];
|
|
|
|
VM_PAGE_SETQUEUE2(m, queue);
|
|
|
|
TAILQ_INSERT_TAIL(&vpq->pl, m, pageq);
|
|
|
|
++*vpq->cnt;
|
|
|
|
}
|
|
|
|
|
1996-01-27 00:13:33 +00:00
|
|
|
/*
|
1996-05-18 03:38:05 +00:00
|
|
|
* vm_page_activate:
|
|
|
|
*
|
|
|
|
* Put the specified page on the active list (if appropriate).
|
1999-09-17 04:56:40 +00:00
|
|
|
* Ensure that act_count is at least ACT_INIT but do not otherwise
|
|
|
|
* mess with it.
|
1996-05-18 03:38:05 +00:00
|
|
|
*
|
|
|
|
* The page queues must be locked.
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
1996-01-27 00:13:33 +00:00
|
|
|
*/
|
1996-05-18 03:38:05 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_activate(vm_page_t m)
|
1994-09-27 18:00:29 +00:00
|
|
|
{
|
|
|
|
|
2002-08-11 00:21:40 +00:00
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2010-05-04 05:55:19 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2005-12-31 14:39:20 +00:00
|
|
|
if (VM_PAGE_GETKNOWNQUEUE2(m) != PQ_ACTIVE) {
|
2001-07-04 23:27:09 +00:00
|
|
|
vm_pageq_remove(m);
|
2000-05-29 22:40:54 +00:00
|
|
|
if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
|
1998-01-31 11:56:53 +00:00
|
|
|
if (m->act_count < ACT_INIT)
|
|
|
|
m->act_count = ACT_INIT;
|
2008-03-18 06:52:15 +00:00
|
|
|
vm_page_enqueue(PQ_ACTIVE, m);
|
1998-01-31 11:56:53 +00:00
|
|
|
}
|
|
|
|
} else {
|
This commit does a couple of things:
Re-enables the RSS limiting, and the routine is now tail-recursive,
making it much more safe (eliminates the possiblity of kernel stack
overflow.) Also, the RSS limiting is a little more intelligent about
finding the likely objects that are pushing the process over the limit.
Added some sysctls that help with VM system tuning.
New sysctl features:
1) Enable/disable lru pageout algorithm.
vm.pageout_algorithm = 0, default algorithm that works
well, especially using X windows and heavy
memory loading. Can have adverse effects,
sometimes slowing down program loading.
vm.pageout_algorithm = 1, close to true LRU. Works much
better than clock, etc. Does not work as well as
the default algorithm in general. Certain memory
"malloc" type benchmarks work a little better with
this setting.
Please give me feedback on the performance results
associated with these.
2) Enable/disable swapping.
vm.swapping_enabled = 1, default.
vm.swapping_enabled = 0, useful for cases where swapping
degrades performance.
The config option "NO_SWAPPING" is still operative, and
takes precedence over the sysctl. If "NO_SWAPPING" is
specified, the sysctl still exists, but "vm.swapping_enabled"
is hard-wired to "0".
Each of these can be changed "on the fly."
1996-06-26 05:39:27 +00:00
|
|
|
if (m->act_count < ACT_INIT)
|
|
|
|
m->act_count = ACT_INIT;
|
1994-09-27 18:00:29 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
1996-06-05 03:31:49 +00:00
|
|
|
/*
|
1999-02-15 06:52:14 +00:00
|
|
|
* vm_page_free_wakeup:
|
|
|
|
*
|
|
|
|
* Helper routine for vm_page_free_toq() and vm_page_cache(). This
|
|
|
|
* routine is called when a page has been added to the cache or free
|
|
|
|
* queues.
|
1998-12-23 01:52:47 +00:00
|
|
|
*
|
2004-06-19 04:19:47 +00:00
|
|
|
* The page queues must be locked.
|
1999-02-15 06:52:14 +00:00
|
|
|
* This routine may not block.
|
1996-06-05 03:31:49 +00:00
|
|
|
*/
|
2006-03-08 06:31:46 +00:00
|
|
|
static inline void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_free_wakeup(void)
|
1996-06-05 03:31:49 +00:00
|
|
|
{
|
2003-02-01 21:18:16 +00:00
|
|
|
|
2007-02-07 06:37:30 +00:00
|
|
|
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* if pageout daemon needs pages, then tell it that there are
|
|
|
|
* some free.
|
|
|
|
*/
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
if (vm_pageout_pages_needed &&
|
2007-05-31 22:52:15 +00:00
|
|
|
cnt.v_cache_count + cnt.v_free_count >= cnt.v_pageout_free_min) {
|
1999-01-21 08:29:12 +00:00
|
|
|
wakeup(&vm_pageout_pages_needed);
|
|
|
|
vm_pageout_pages_needed = 0;
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* wakeup processes that are waiting on memory if we hit a
|
|
|
|
* high water mark. And wakeup scheduler process if we have
|
|
|
|
* lots of memory. this process will swapin processes.
|
|
|
|
*/
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
if (vm_pages_needed && !vm_page_count_min()) {
|
1999-01-21 08:29:12 +00:00
|
|
|
vm_pages_needed = 0;
|
2007-05-31 22:52:15 +00:00
|
|
|
wakeup(&cnt.v_free_count);
|
1999-01-21 08:29:12 +00:00
|
|
|
}
|
|
|
|
}
|
1998-01-17 09:17:02 +00:00
|
|
|
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* vm_page_free_toq:
|
|
|
|
*
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
* Returns the given page to the free list,
|
1999-01-21 08:29:12 +00:00
|
|
|
* disassociating it with any VM object.
|
|
|
|
*
|
|
|
|
* Object and page must be locked prior to entry.
|
|
|
|
* This routine may not block.
|
|
|
|
*/
|
|
|
|
|
|
|
|
void
|
1999-02-08 00:37:36 +00:00
|
|
|
vm_page_free_toq(vm_page_t m)
|
1999-01-21 08:29:12 +00:00
|
|
|
{
|
|
|
|
|
2010-05-06 16:39:43 +00:00
|
|
|
if ((m->flags & PG_UNMANAGED) == 0) {
|
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
|
|
|
KASSERT(!pmap_page_is_mapped(m),
|
|
|
|
("vm_page_free_toq: freeing mapped page %p", m));
|
|
|
|
}
|
2007-06-04 21:45:18 +00:00
|
|
|
PCPU_INC(cnt.v_tfree);
|
1998-01-17 09:17:02 +00:00
|
|
|
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
if (m->busy || VM_PAGE_IS_FREE(m)) {
|
1998-07-11 07:46:16 +00:00
|
|
|
printf(
|
2006-10-22 04:28:14 +00:00
|
|
|
"vm_page_free: pindex(%lu), busy(%d), VPO_BUSY(%d), hold(%d)\n",
|
|
|
|
(u_long)m->pindex, m->busy, (m->oflags & VPO_BUSY) ? 1 : 0,
|
1998-07-11 07:46:16 +00:00
|
|
|
m->hold_count);
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
if (VM_PAGE_IS_FREE(m))
|
1996-06-05 03:31:49 +00:00
|
|
|
panic("vm_page_free: freeing free page");
|
|
|
|
else
|
|
|
|
panic("vm_page_free: freeing busy page");
|
|
|
|
}
|
|
|
|
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* unqueue, then remove page. Note that we cannot destroy
|
|
|
|
* the page here because we do not want to call the pager's
|
|
|
|
* callback routine until after we've put the page on the
|
|
|
|
* appropriate free queue.
|
|
|
|
*/
|
2010-05-06 16:39:43 +00:00
|
|
|
if (VM_PAGE_GETQUEUE(m) != PQ_NONE) {
|
|
|
|
vm_page_lock_queues();
|
|
|
|
vm_pageq_remove(m);
|
|
|
|
vm_page_unlock_queues();
|
|
|
|
}
|
1998-01-31 11:56:53 +00:00
|
|
|
vm_page_remove(m);
|
|
|
|
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* If fictitious remove object association and
|
|
|
|
* return, otherwise delay object association removal.
|
|
|
|
*/
|
1996-06-05 03:31:49 +00:00
|
|
|
if ((m->flags & PG_FICTITIOUS) != 0) {
|
1999-01-21 08:29:12 +00:00
|
|
|
return;
|
1996-06-05 03:31:49 +00:00
|
|
|
}
|
1998-01-31 11:56:53 +00:00
|
|
|
|
1998-03-01 04:18:54 +00:00
|
|
|
m->valid = 0;
|
1999-08-17 05:08:39 +00:00
|
|
|
vm_page_undirty(m);
|
1998-03-01 04:18:54 +00:00
|
|
|
|
1996-06-05 03:31:49 +00:00
|
|
|
if (m->wire_count != 0) {
|
|
|
|
if (m->wire_count > 1) {
|
1999-07-01 19:53:43 +00:00
|
|
|
panic("vm_page_free: invalid wire count (%d), pindex: 0x%lx",
|
|
|
|
m->wire_count, (long)m->pindex);
|
1996-06-05 03:31:49 +00:00
|
|
|
}
|
2004-06-19 04:19:47 +00:00
|
|
|
panic("vm_page_free: freeing wired page");
|
1996-06-05 03:31:49 +00:00
|
|
|
}
|
2002-02-19 23:19:30 +00:00
|
|
|
if (m->hold_count != 0) {
|
|
|
|
m->flags &= ~PG_ZERO;
|
2010-05-06 16:39:43 +00:00
|
|
|
vm_page_lock_queues();
|
2008-03-18 06:52:15 +00:00
|
|
|
vm_page_enqueue(PQ_HOLD, m);
|
2010-05-06 16:39:43 +00:00
|
|
|
vm_page_unlock_queues();
|
1999-02-08 00:37:36 +00:00
|
|
|
} else {
|
2009-07-12 23:31:20 +00:00
|
|
|
/*
|
|
|
|
* Restore the default memory attribute to the page.
|
|
|
|
*/
|
|
|
|
if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
|
|
|
|
pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Insert the page into the physical memory allocator's
|
|
|
|
* cache/free page queues.
|
|
|
|
*/
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
mtx_lock(&vm_page_queue_free_mtx);
|
2008-01-02 04:43:47 +00:00
|
|
|
m->flags |= PG_FREE;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
cnt.v_free_count++;
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
|
|
|
if (!vm_reserv_free_page(m))
|
|
|
|
#else
|
|
|
|
if (TRUE)
|
|
|
|
#endif
|
|
|
|
vm_phys_free_pages(m, 0);
|
2007-12-11 21:20:34 +00:00
|
|
|
if ((m->flags & PG_ZERO) != 0)
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
++vm_page_zero_count;
|
2007-12-11 21:20:34 +00:00
|
|
|
else
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
vm_page_zero_idle_wakeup();
|
|
|
|
vm_page_free_wakeup();
|
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
1996-06-05 03:31:49 +00:00
|
|
|
}
|
|
|
|
}
|
1994-05-25 09:21:21 +00:00
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* vm_page_wire:
|
|
|
|
*
|
|
|
|
* Mark this page as wired down by yet
|
|
|
|
* another map, removing it from paging queues
|
|
|
|
* as necessary.
|
|
|
|
*
|
2010-05-05 03:45:46 +00:00
|
|
|
* The page must be locked.
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1995-05-30 08:16:23 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_wire(vm_page_t m)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
|
|
|
|
2000-05-29 22:40:54 +00:00
|
|
|
/*
|
|
|
|
* Only bump the wire statistics if the page is not already wired,
|
|
|
|
* and only unqueue the page if it is on some queue (if it is unmanaged
|
|
|
|
* it is already off the queues).
|
|
|
|
*/
|
2010-05-03 17:55:32 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2004-05-22 04:53:51 +00:00
|
|
|
if (m->flags & PG_FICTITIOUS)
|
|
|
|
return;
|
1996-01-19 04:00:31 +00:00
|
|
|
if (m->wire_count == 0) {
|
2010-05-04 15:55:41 +00:00
|
|
|
if ((m->flags & PG_UNMANAGED) == 0) {
|
|
|
|
vm_page_lock_queues();
|
2001-07-04 23:27:09 +00:00
|
|
|
vm_pageq_remove(m);
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_unlock_queues();
|
|
|
|
}
|
2007-05-31 22:52:15 +00:00
|
|
|
atomic_add_int(&cnt.v_wire_count, 1);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1996-01-19 04:00:31 +00:00
|
|
|
m->wire_count++;
|
2001-08-22 04:01:56 +00:00
|
|
|
KASSERT(m->wire_count != 0, ("vm_page_wire: wire_count overflow m=%p", m));
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_unwire:
|
|
|
|
*
|
|
|
|
* Release one wiring of this page, potentially
|
|
|
|
* enabling it to be paged again.
|
|
|
|
*
|
1999-01-21 08:29:12 +00:00
|
|
|
* Many pages placed on the inactive queue should actually go
|
|
|
|
* into the cache, but it is difficult to figure out which. What
|
|
|
|
* we do instead, if the inactive target is well met, is to put
|
|
|
|
* clean pages at the head of the inactive queue instead of the tail.
|
|
|
|
* This will cause them to be moved to the cache more quickly and
|
|
|
|
* if not actively re-referenced, freed more quickly. If we just
|
|
|
|
* stick these pages at the end of the inactive queue, heavy filesystem
|
|
|
|
* meta-data accesses can cause an unnecessary paging load on memory bound
|
|
|
|
* processes. This optimization causes one-time-use metadata to be
|
|
|
|
* reused more quickly.
|
|
|
|
*
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
* BUT, if we are in a low-memory situation we have no choice but to
|
|
|
|
* put clean pages on the cache queue.
|
|
|
|
*
|
2000-03-26 15:20:23 +00:00
|
|
|
* A number of routines use vm_page_unwire() to guarantee that the page
|
1999-01-24 06:00:31 +00:00
|
|
|
* will go into either the inactive or active queues, and will NEVER
|
|
|
|
* be placed in the cache - for example, just after dirtying a page.
|
|
|
|
* dirty pages in the cache are not allowed.
|
|
|
|
*
|
2010-05-05 03:45:46 +00:00
|
|
|
* The page must be locked.
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1995-05-30 08:16:23 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_unwire(vm_page_t m, int activate)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2010-05-05 03:45:46 +00:00
|
|
|
if ((m->flags & PG_UNMANAGED) == 0)
|
2010-05-04 05:55:19 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2004-05-22 04:53:51 +00:00
|
|
|
if (m->flags & PG_FICTITIOUS)
|
|
|
|
return;
|
1998-01-31 11:56:53 +00:00
|
|
|
if (m->wire_count > 0) {
|
1996-01-19 04:00:31 +00:00
|
|
|
m->wire_count--;
|
1998-01-31 11:56:53 +00:00
|
|
|
if (m->wire_count == 0) {
|
2007-05-31 22:52:15 +00:00
|
|
|
atomic_subtract_int(&cnt.v_wire_count, 1);
|
2010-05-05 03:45:46 +00:00
|
|
|
if ((m->flags & PG_UNMANAGED) != 0)
|
|
|
|
return;
|
|
|
|
vm_page_lock_queues();
|
|
|
|
if (activate)
|
2008-03-18 06:52:15 +00:00
|
|
|
vm_page_enqueue(PQ_ACTIVE, m);
|
2002-03-04 18:55:26 +00:00
|
|
|
else {
|
2000-12-26 19:41:38 +00:00
|
|
|
vm_page_flag_clear(m, PG_WINATCFLS);
|
2008-03-18 06:52:15 +00:00
|
|
|
vm_page_enqueue(PQ_INACTIVE, m);
|
1998-10-28 13:37:02 +00:00
|
|
|
}
|
2010-05-05 03:45:46 +00:00
|
|
|
vm_page_unlock_queues();
|
1998-01-31 11:56:53 +00:00
|
|
|
}
|
|
|
|
} else {
|
2004-06-19 04:19:47 +00:00
|
|
|
panic("vm_page_unwire: invalid wire count: %d", m->wire_count);
|
1995-03-25 08:47:35 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
/*
|
2010-05-07 04:14:07 +00:00
|
|
|
* Move the specified page to the inactive queue.
|
1998-12-23 01:52:47 +00:00
|
|
|
*
|
1999-09-17 04:56:40 +00:00
|
|
|
* Normally athead is 0 resulting in LRU operation. athead is set
|
|
|
|
* to 1 if we want this page to be 'as if it were placed in the cache',
|
|
|
|
* except without unmapping it from the process address space.
|
|
|
|
*
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
1994-05-25 09:21:21 +00:00
|
|
|
*/
|
2006-03-08 06:31:46 +00:00
|
|
|
static inline void
|
1999-09-17 04:56:40 +00:00
|
|
|
_vm_page_deactivate(vm_page_t m, int athead)
|
1994-05-25 09:21:21 +00:00
|
|
|
{
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2010-05-04 05:55:19 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2004-06-19 04:19:47 +00:00
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
/*
|
1998-10-28 13:41:43 +00:00
|
|
|
* Ignore if already inactive.
|
1994-05-25 09:21:21 +00:00
|
|
|
*/
|
2005-12-31 14:39:20 +00:00
|
|
|
if (VM_PAGE_INQUEUE2(m, PQ_INACTIVE))
|
1996-01-19 04:00:31 +00:00
|
|
|
return;
|
2000-05-29 22:40:54 +00:00
|
|
|
if (m->wire_count == 0 && (m->flags & PG_UNMANAGED) == 0) {
|
2010-05-07 04:14:07 +00:00
|
|
|
vm_page_lock_queues();
|
2000-12-26 19:41:38 +00:00
|
|
|
vm_page_flag_clear(m, PG_WINATCFLS);
|
2001-07-04 23:27:09 +00:00
|
|
|
vm_pageq_remove(m);
|
1999-09-17 04:56:40 +00:00
|
|
|
if (athead)
|
1999-10-30 07:37:14 +00:00
|
|
|
TAILQ_INSERT_HEAD(&vm_page_queues[PQ_INACTIVE].pl, m, pageq);
|
1999-09-17 04:56:40 +00:00
|
|
|
else
|
1999-10-30 07:37:14 +00:00
|
|
|
TAILQ_INSERT_TAIL(&vm_page_queues[PQ_INACTIVE].pl, m, pageq);
|
2005-12-31 14:39:20 +00:00
|
|
|
VM_PAGE_SETQUEUE2(m, PQ_INACTIVE);
|
2007-05-31 22:52:15 +00:00
|
|
|
cnt.v_inactive_count++;
|
2010-05-07 04:14:07 +00:00
|
|
|
vm_page_unlock_queues();
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
1994-05-25 09:21:21 +00:00
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2010-05-07 04:14:07 +00:00
|
|
|
/*
|
|
|
|
* Move the specified page to the inactive queue.
|
|
|
|
*
|
|
|
|
* The page must be locked.
|
|
|
|
*/
|
1999-09-17 04:56:40 +00:00
|
|
|
void
|
|
|
|
vm_page_deactivate(vm_page_t m)
|
|
|
|
{
|
2010-05-07 04:14:07 +00:00
|
|
|
|
|
|
|
_vm_page_deactivate(m, 0);
|
1999-09-17 04:56:40 +00:00
|
|
|
}
|
|
|
|
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
/*
|
|
|
|
* vm_page_try_to_cache:
|
|
|
|
*
|
|
|
|
* Returns 0 on failure, 1 on success
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
vm_page_try_to_cache(vm_page_t m)
|
|
|
|
{
|
2001-05-19 01:28:09 +00:00
|
|
|
|
2002-07-20 20:58:46 +00:00
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2010-05-02 23:33:10 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2004-10-28 05:26:21 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
if (m->dirty || m->hold_count || m->busy || m->wire_count ||
|
2006-10-22 04:28:14 +00:00
|
|
|
(m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) {
|
2002-03-10 21:52:48 +00:00
|
|
|
return (0);
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
}
|
2004-02-14 08:54:37 +00:00
|
|
|
pmap_remove_all(m);
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
if (m->dirty)
|
2002-03-10 21:52:48 +00:00
|
|
|
return (0);
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
vm_page_cache(m);
|
2002-03-10 21:52:48 +00:00
|
|
|
return (1);
|
Implement a low-memory deadlock solution.
Removed most of the hacks that were trying to deal with low-memory
situations prior to now.
The new code is based on the concept that I/O must be able to function in
a low memory situation. All major modules related to I/O (except
networking) have been adjusted to allow allocation out of the system
reserve memory pool. These modules now detect a low memory situation but
rather then block they instead continue to operate, then return resources
to the memory pool instead of cache them or leave them wired.
Code has been added to stall in a low-memory situation prior to a vnode
being locked.
Thus situations where a process blocks in a low-memory condition while
holding a locked vnode have been reduced to near nothing. Not only will
I/O continue to operate, but many prior deadlock conditions simply no
longer exist.
Implement a number of VFS/BIO fixes
(found by Ian): in biodone(), bogus-page replacement code, the loop
was not properly incrementing loop variables prior to a continue
statement. We do not believe this code can be hit anyway but we
aren't taking any chances. We'll turn the whole section into a
panic (as it already is in brelse()) after the release is rolled.
In biodone(), the foff calculation was incorrectly
clamped to the iosize, causing the wrong foff to be calculated
for pages in the case of an I/O error or biodone() called without
initiating I/O. The problem always caused a panic before. Now it
doesn't. The problem is mainly an issue with NFS.
Fixed casts for ~PAGE_MASK. This code worked properly before only
because the calculations use signed arithmatic. Better to properly
extend PAGE_MASK first before inverting it for the 64 bit masking
op.
In brelse(), the bogus_page fixup code was improperly throwing
away the original contents of 'm' when it did the j-loop to
fix the bogus pages. The result was that it would potentially
invalidate parts of the *WRONG* page(!), leading to corruption.
There may still be cases where a background bitmap write is
being duplicated, causing potential corruption. We have identified
a potentially serious bug related to this but the fix is still TBD.
So instead this patch contains a KASSERT to detect the problem
and panic the machine rather then continue to corrupt the filesystem.
The problem does not occur very often.. it is very hard to
reproduce, and it may or may not be the cause of the corruption
people have reported.
Review by: (VFS/BIO: mckusick, Ian Dowse <iedowse@maths.tcd.ie>)
Testing by: (VM/Deadlock) Paul Saab <ps@yahoo-inc.com>
2000-11-18 23:06:26 +00:00
|
|
|
}
|
|
|
|
|
2001-05-24 07:22:27 +00:00
|
|
|
/*
|
|
|
|
* vm_page_try_to_free()
|
|
|
|
*
|
|
|
|
* Attempt to free the page. If we cannot free it, we do nothing.
|
|
|
|
* 1 is returned on success, 0 on failure.
|
|
|
|
*/
|
|
|
|
int
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_try_to_free(vm_page_t m)
|
2001-05-24 07:22:27 +00:00
|
|
|
{
|
2002-07-20 20:12:57 +00:00
|
|
|
|
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2010-05-02 23:33:10 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2003-06-19 01:50:14 +00:00
|
|
|
if (m->object != NULL)
|
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
2001-05-24 07:22:27 +00:00
|
|
|
if (m->dirty || m->hold_count || m->busy || m->wire_count ||
|
2006-10-22 04:28:14 +00:00
|
|
|
(m->oflags & VPO_BUSY) || (m->flags & PG_UNMANAGED)) {
|
2002-03-10 21:52:48 +00:00
|
|
|
return (0);
|
2001-05-24 07:22:27 +00:00
|
|
|
}
|
2004-02-19 07:43:55 +00:00
|
|
|
pmap_remove_all(m);
|
2001-05-24 07:22:27 +00:00
|
|
|
if (m->dirty)
|
2002-03-10 21:52:48 +00:00
|
|
|
return (0);
|
2001-05-24 07:22:27 +00:00
|
|
|
vm_page_free(m);
|
2002-03-10 21:52:48 +00:00
|
|
|
return (1);
|
2001-05-24 07:22:27 +00:00
|
|
|
}
|
|
|
|
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
/*
|
|
|
|
* vm_page_cache
|
|
|
|
*
|
1999-01-21 08:29:12 +00:00
|
|
|
* Put the specified page onto the page cache queue (if appropriate).
|
|
|
|
*
|
1998-12-23 01:52:47 +00:00
|
|
|
* This routine may not block.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*/
|
1995-05-30 08:16:23 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_cache(vm_page_t m)
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
{
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
vm_object_t object;
|
|
|
|
vm_page_t root;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2002-07-20 19:34:21 +00:00
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2010-05-02 23:33:10 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
object = m->object;
|
|
|
|
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
|
2006-10-22 04:28:14 +00:00
|
|
|
if ((m->flags & PG_UNMANAGED) || (m->oflags & VPO_BUSY) || m->busy ||
|
2003-08-23 20:29:29 +00:00
|
|
|
m->hold_count || m->wire_count) {
|
2007-06-16 21:07:51 +00:00
|
|
|
panic("vm_page_cache: attempting to cache busy page");
|
1996-03-28 04:53:28 +00:00
|
|
|
}
|
2007-10-08 18:01:38 +00:00
|
|
|
pmap_remove_all(m);
|
|
|
|
if (m->dirty != 0)
|
|
|
|
panic("vm_page_cache: page %p is dirty", m);
|
2007-11-05 10:25:12 +00:00
|
|
|
if (m->valid == 0 || object->type == OBJT_DEFAULT ||
|
|
|
|
(object->type == OBJT_SWAP &&
|
|
|
|
!vm_pager_has_page(object, m->pindex, NULL, NULL))) {
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
/*
|
|
|
|
* Hypothesis: A cache-elgible page belonging to a
|
2007-11-05 10:25:12 +00:00
|
|
|
* default object or swap object but without a backing
|
|
|
|
* store must be zero filled.
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
*/
|
|
|
|
vm_page_free(m);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
return;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
}
|
|
|
|
KASSERT((m->flags & PG_CACHED) == 0,
|
|
|
|
("vm_page_cache: page %p is already cached", m));
|
2007-07-27 20:01:22 +00:00
|
|
|
cnt.v_tcached++;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
/*
|
|
|
|
* Remove the page from the paging queues.
|
|
|
|
*/
|
|
|
|
vm_pageq_remove(m);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Remove the page from the object's collection of resident
|
|
|
|
* pages.
|
|
|
|
*/
|
|
|
|
if (m != object->root)
|
|
|
|
vm_page_splay(m->pindex, object->root);
|
|
|
|
if (m->left == NULL)
|
|
|
|
root = m->right;
|
|
|
|
else {
|
|
|
|
root = vm_page_splay(m->pindex, m->left);
|
|
|
|
root->right = m->right;
|
|
|
|
}
|
|
|
|
object->root = root;
|
|
|
|
TAILQ_REMOVE(&object->memq, m, listq);
|
|
|
|
object->resident_page_count--;
|
|
|
|
object->generation++;
|
|
|
|
|
2009-07-12 23:31:20 +00:00
|
|
|
/*
|
|
|
|
* Restore the default memory attribute to the page.
|
|
|
|
*/
|
|
|
|
if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
|
|
|
|
pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
|
|
|
|
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
/*
|
|
|
|
* Insert the page into the object's collection of cached pages
|
|
|
|
* and the physical memory allocator's cache/free page queues.
|
|
|
|
*/
|
|
|
|
vm_page_flag_clear(m, PG_ZERO);
|
2007-02-07 06:37:30 +00:00
|
|
|
mtx_lock(&vm_page_queue_free_mtx);
|
2008-01-02 04:43:47 +00:00
|
|
|
m->flags |= PG_CACHED;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
cnt.v_cache_count++;
|
|
|
|
root = object->cache;
|
|
|
|
if (root == NULL) {
|
|
|
|
m->left = NULL;
|
|
|
|
m->right = NULL;
|
|
|
|
} else {
|
|
|
|
root = vm_page_splay(m->pindex, root);
|
|
|
|
if (m->pindex < root->pindex) {
|
|
|
|
m->left = root->left;
|
|
|
|
m->right = root;
|
|
|
|
root->left = NULL;
|
|
|
|
} else if (__predict_false(m->pindex == root->pindex))
|
|
|
|
panic("vm_page_cache: offset already cached");
|
|
|
|
else {
|
|
|
|
m->right = root->right;
|
|
|
|
m->left = root;
|
|
|
|
root->right = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
object->cache = m;
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
|
|
|
if (!vm_reserv_free_page(m)) {
|
|
|
|
#else
|
|
|
|
if (TRUE) {
|
|
|
|
#endif
|
|
|
|
vm_phys_set_pool(VM_FREEPOOL_CACHE, m, 0);
|
|
|
|
vm_phys_free_pages(m, 0);
|
|
|
|
}
|
1996-06-16 20:37:31 +00:00
|
|
|
vm_page_free_wakeup();
|
2007-02-07 06:37:30 +00:00
|
|
|
mtx_unlock(&vm_page_queue_free_mtx);
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Increment the vnode's hold count if this is the object's only
|
|
|
|
* cached page. Decrement the vnode's hold count if this was
|
|
|
|
* the object's only resident page.
|
|
|
|
*/
|
|
|
|
if (object->type == OBJT_VNODE) {
|
|
|
|
if (root == NULL && object->resident_page_count != 0)
|
|
|
|
vhold(object->handle);
|
|
|
|
else if (root != NULL && object->resident_page_count == 0)
|
|
|
|
vdrop(object->handle);
|
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
|
|
|
|
1999-09-17 04:56:40 +00:00
|
|
|
/*
|
|
|
|
* vm_page_dontneed
|
|
|
|
*
|
|
|
|
* Cache, deactivate, or do nothing as appropriate. This routine
|
|
|
|
* is typically used by madvise() MADV_DONTNEED.
|
|
|
|
*
|
|
|
|
* Generally speaking we want to move the page into the cache so
|
|
|
|
* it gets reused quickly. However, this can result in a silly syndrome
|
|
|
|
* due to the page recycling too quickly. Small objects will not be
|
|
|
|
* fully cached. On the otherhand, if we move the page to the inactive
|
|
|
|
* queue we wind up with a problem whereby very large objects
|
|
|
|
* unnecessarily blow away our inactive and cache queues.
|
|
|
|
*
|
|
|
|
* The solution is to move the pages based on a fixed weighting. We
|
|
|
|
* either leave them alone, deactivate them, or move them to the cache,
|
|
|
|
* where moving them to the cache has the highest weighting.
|
|
|
|
* By forcing some pages into other queues we eventually force the
|
|
|
|
* system to balance the queues, potentially recovering other unrelated
|
|
|
|
* space from active. The idea is to not force this to happen too
|
|
|
|
* often.
|
|
|
|
*/
|
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_dontneed(vm_page_t m)
|
1999-09-17 04:56:40 +00:00
|
|
|
{
|
|
|
|
static int dnweight;
|
|
|
|
int dnw;
|
|
|
|
int head;
|
|
|
|
|
2002-07-23 04:39:48 +00:00
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2010-05-04 05:55:19 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
1999-09-17 04:56:40 +00:00
|
|
|
dnw = ++dnweight;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* occassionally leave the page alone
|
|
|
|
*/
|
|
|
|
if ((dnw & 0x01F0) == 0 ||
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
VM_PAGE_INQUEUE2(m, PQ_INACTIVE)) {
|
1999-09-17 04:56:40 +00:00
|
|
|
if (m->act_count >= ACT_INIT)
|
|
|
|
--m->act_count;
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
Essentially, neither madvise(..., MADV_DONTNEED) nor madvise(..., MADV_FREE)
work. (Moreover, I don't believe that they have ever worked as intended.)
The explanation is fairly simple. Both MADV_DONTNEED and MADV_FREE perform
vm_page_dontneed() on each page within the range given to madvise(). This
function moves the page to the inactive queue. Specifically, if the page is
clean, it is moved to the head of the inactive queue where it is first in
line for processing by the page daemon. On the other hand, if it is dirty,
it is placed at the tail. Let's further examine the case in which the page
is clean. Recall that the page is at the head of the line for processing by
the page daemon. The expectation of vm_page_dontneed()'s author was that
the page would be transferred from the inactive queue to the cache queue by
the page daemon. (Once the page is in the cache queue, it is, in effect,
free, that is, it can be reallocated to a new vm object by vm_page_alloc()
if it isn't reactivated quickly enough by a user of the old vm object.) The
trouble is that nowhere in the execution of either MADV_DONTNEED or
MADV_FREE is either the machine-independent reference flag (PG_REFERENCED)
or the reference bit in any page table entry (PTE) mapping the page cleared.
Consequently, the immediate reaction of the page daemon is to reactivate the
page because it is referenced. In effect, the madvise() was for naught.
The case in which the page was dirty is not too different. Instead of being
laundered, the page is reactivated.
Note: The essential difference between MADV_DONTNEED and MADV_FREE is
that MADV_FREE clears a page's dirty field. So, MADV_FREE is always
executing the clean case above.
This revision changes vm_page_dontneed() to clear both the machine-
independent reference flag (PG_REFERENCED) and the reference bit in all PTEs
mapping the page.
MFC after: 6 weeks
2008-06-06 18:38:43 +00:00
|
|
|
/*
|
|
|
|
* Clear any references to the page. Otherwise, the page daemon will
|
|
|
|
* immediately reactivate the page.
|
|
|
|
*/
|
|
|
|
vm_page_flag_clear(m, PG_REFERENCED);
|
|
|
|
pmap_clear_reference(m);
|
|
|
|
|
2004-02-19 07:43:55 +00:00
|
|
|
if (m->dirty == 0 && pmap_is_modified(m))
|
|
|
|
vm_page_dirty(m);
|
1999-09-17 04:56:40 +00:00
|
|
|
|
|
|
|
if (m->dirty || (dnw & 0x0070) == 0) {
|
|
|
|
/*
|
|
|
|
* Deactivate the page 3 times out of 32.
|
|
|
|
*/
|
|
|
|
head = 0;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Cache the page 28 times out of every 32. Note that
|
|
|
|
* the page is deactivated instead of cached, but placed
|
|
|
|
* at the head of the queue instead of the tail.
|
|
|
|
*/
|
|
|
|
head = 1;
|
|
|
|
}
|
|
|
|
_vm_page_deactivate(m, head);
|
|
|
|
}
|
|
|
|
|
1998-02-05 03:32:49 +00:00
|
|
|
/*
|
|
|
|
* Grab a page, waiting until we are waken up due to the page
|
|
|
|
* changing state. We keep on waiting, if the page continues
|
2004-04-24 21:36:23 +00:00
|
|
|
* to be in the object. If the page doesn't exist, first allocate it
|
|
|
|
* and then conditionally zero it.
|
1998-12-23 01:52:47 +00:00
|
|
|
*
|
|
|
|
* This routine may block.
|
1998-02-05 03:32:49 +00:00
|
|
|
*/
|
|
|
|
vm_page_t
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
|
1998-02-05 03:32:49 +00:00
|
|
|
{
|
|
|
|
vm_page_t m;
|
|
|
|
|
2003-08-21 20:59:07 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
|
1998-02-05 03:32:49 +00:00
|
|
|
retrylookup:
|
|
|
|
if ((m = vm_page_lookup(object, pindex)) != NULL) {
|
It makes no sense for vm_page_sleep_if_busy()'s helper, vm_page_sleep(),
to unconditionally set PG_REFERENCED on a page before sleeping. In many
cases, it's perfectly ok for the page to disappear, i.e., be reclaimed by
the page daemon, before the caller to vm_page_sleep() is reawakened.
Instead, we now explicitly set PG_REFERENCED in those cases where having
the page persist until the caller is awakened is clearly desirable. Note,
however, that setting PG_REFERENCED on the page is still only a hint,
and not a guarantee that the page should persist.
2010-05-02 17:33:46 +00:00
|
|
|
if ((m->oflags & VPO_BUSY) != 0 || m->busy != 0) {
|
|
|
|
if ((allocflags & VM_ALLOC_RETRY) != 0) {
|
|
|
|
/*
|
|
|
|
* Reference the page before unlocking and
|
|
|
|
* sleeping so that the page daemon is less
|
|
|
|
* likely to reclaim it.
|
|
|
|
*/
|
|
|
|
vm_page_lock_queues();
|
|
|
|
vm_page_flag_set(m, PG_REFERENCED);
|
|
|
|
}
|
|
|
|
vm_page_sleep(m, "pgrbwt");
|
2003-12-31 01:44:45 +00:00
|
|
|
if ((allocflags & VM_ALLOC_RETRY) == 0)
|
|
|
|
return (NULL);
|
1998-02-05 03:32:49 +00:00
|
|
|
goto retrylookup;
|
|
|
|
} else {
|
2006-10-22 21:18:48 +00:00
|
|
|
if ((allocflags & VM_ALLOC_WIRED) != 0) {
|
2010-05-03 17:55:32 +00:00
|
|
|
vm_page_lock(m);
|
2002-07-28 23:46:19 +00:00
|
|
|
vm_page_wire(m);
|
2010-05-03 17:55:32 +00:00
|
|
|
vm_page_unlock(m);
|
2006-10-22 21:18:48 +00:00
|
|
|
}
|
2004-10-24 06:15:36 +00:00
|
|
|
if ((allocflags & VM_ALLOC_NOBUSY) == 0)
|
|
|
|
vm_page_busy(m);
|
2004-04-24 21:36:23 +00:00
|
|
|
return (m);
|
1998-02-05 03:32:49 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
m = vm_page_alloc(object, pindex, allocflags & ~VM_ALLOC_RETRY);
|
|
|
|
if (m == NULL) {
|
2003-08-21 20:59:07 +00:00
|
|
|
VM_OBJECT_UNLOCK(object);
|
1998-02-05 03:32:49 +00:00
|
|
|
VM_WAIT;
|
2003-08-21 20:59:07 +00:00
|
|
|
VM_OBJECT_LOCK(object);
|
1998-02-05 03:32:49 +00:00
|
|
|
if ((allocflags & VM_ALLOC_RETRY) == 0)
|
2004-04-24 21:36:23 +00:00
|
|
|
return (NULL);
|
1998-02-05 03:32:49 +00:00
|
|
|
goto retrylookup;
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
} else if (m->valid != 0)
|
|
|
|
return (m);
|
2004-04-24 20:53:55 +00:00
|
|
|
if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
|
|
|
|
pmap_zero_page(m);
|
2004-04-24 21:36:23 +00:00
|
|
|
return (m);
|
1998-02-05 03:32:49 +00:00
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
|
|
|
/*
|
1999-04-05 19:38:30 +00:00
|
|
|
* Mapping function for valid bits or for dirty bits in
|
1998-12-23 01:52:47 +00:00
|
|
|
* a page. May not block.
|
1999-04-05 19:38:30 +00:00
|
|
|
*
|
|
|
|
* Inputs are required to range within a page.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*/
|
2007-06-17 04:19:05 +00:00
|
|
|
int
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
vm_page_bits(int base, int size)
|
|
|
|
{
|
1999-04-05 19:38:30 +00:00
|
|
|
int first_bit;
|
|
|
|
int last_bit;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
1999-04-05 19:38:30 +00:00
|
|
|
KASSERT(
|
|
|
|
base + size <= PAGE_SIZE,
|
|
|
|
("vm_page_bits: illegal base/size %d/%d", base, size)
|
|
|
|
);
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
|
1999-04-05 19:38:30 +00:00
|
|
|
if (size == 0) /* handle degenerate case */
|
2002-03-10 21:52:48 +00:00
|
|
|
return (0);
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
|
1999-04-05 19:38:30 +00:00
|
|
|
first_bit = base >> DEV_BSHIFT;
|
|
|
|
last_bit = (base + size - 1) >> DEV_BSHIFT;
|
|
|
|
|
|
|
|
return ((2 << last_bit) - (1 << first_bit));
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
|
|
|
|
2009-05-13 05:39:39 +00:00
|
|
|
/*
|
|
|
|
* vm_page_set_valid:
|
|
|
|
*
|
|
|
|
* Sets portions of a page valid. The arguments are expected
|
|
|
|
* to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
|
|
|
|
* of any partial chunks touched by the range. The invalid portion of
|
|
|
|
* such chunks will be zeroed.
|
|
|
|
*
|
|
|
|
* (base + size) must be less then or equal to PAGE_SIZE.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_set_valid(vm_page_t m, int base, int size)
|
|
|
|
{
|
|
|
|
int endoff, frag;
|
|
|
|
|
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
|
|
|
if (size == 0) /* handle degenerate case */
|
|
|
|
return;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the base is not DEV_BSIZE aligned and the valid
|
|
|
|
* bit is clear, we have to zero out a portion of the
|
|
|
|
* first block.
|
|
|
|
*/
|
|
|
|
if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
|
|
|
|
(m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
|
|
|
|
pmap_zero_page_area(m, frag, base - frag);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the ending offset is not DEV_BSIZE aligned and the
|
|
|
|
* valid bit is clear, we have to zero out a portion of
|
|
|
|
* the last block.
|
|
|
|
*/
|
|
|
|
endoff = base + size;
|
|
|
|
if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
|
|
|
|
(m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
|
|
|
|
pmap_zero_page_area(m, endoff,
|
|
|
|
DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
|
|
|
|
|
2009-05-30 22:06:58 +00:00
|
|
|
/*
|
|
|
|
* Assert that no previously invalid block that is now being validated
|
|
|
|
* is already dirty.
|
|
|
|
*/
|
|
|
|
KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
|
|
|
|
("vm_page_set_valid: page %p is dirty", m));
|
|
|
|
|
2009-05-13 05:39:39 +00:00
|
|
|
/*
|
|
|
|
* Set valid bits inclusive of any overlap.
|
|
|
|
*/
|
|
|
|
m->valid |= vm_page_bits(base, size);
|
|
|
|
}
|
|
|
|
|
1995-09-03 19:57:25 +00:00
|
|
|
/*
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
* vm_page_set_validclean:
|
1999-04-05 19:38:30 +00:00
|
|
|
*
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
* Sets portions of a page valid and clean. The arguments are expected
|
|
|
|
* to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
|
|
|
|
* of any partial chunks touched by the range. The invalid portion of
|
|
|
|
* such chunks will be zero'd.
|
1999-04-05 19:38:30 +00:00
|
|
|
*
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
* This routine may not block.
|
|
|
|
*
|
|
|
|
* (base + size) must be less then or equal to PAGE_SIZE.
|
1995-09-03 19:57:25 +00:00
|
|
|
*/
|
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_set_validclean(vm_page_t m, int base, int size)
|
1995-09-03 19:57:25 +00:00
|
|
|
{
|
1999-04-05 19:38:30 +00:00
|
|
|
int pagebits;
|
|
|
|
int frag;
|
|
|
|
int endoff;
|
|
|
|
|
2002-11-23 19:10:31 +00:00
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2003-10-21 19:36:51 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
1999-04-05 19:38:30 +00:00
|
|
|
if (size == 0) /* handle degenerate case */
|
|
|
|
return;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* If the base is not DEV_BSIZE aligned and the valid
|
|
|
|
* bit is clear, we have to zero out a portion of the
|
|
|
|
* first block.
|
|
|
|
*/
|
|
|
|
if ((frag = base & ~(DEV_BSIZE - 1)) != base &&
|
2002-04-15 16:00:03 +00:00
|
|
|
(m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
|
|
|
|
pmap_zero_page_area(m, frag, base - frag);
|
1999-04-05 19:38:30 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If the ending offset is not DEV_BSIZE aligned and the
|
|
|
|
* valid bit is clear, we have to zero out a portion of
|
|
|
|
* the last block.
|
|
|
|
*/
|
|
|
|
endoff = base + size;
|
|
|
|
if ((frag = endoff & ~(DEV_BSIZE - 1)) != endoff &&
|
2002-04-15 16:00:03 +00:00
|
|
|
(m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
|
|
|
|
pmap_zero_page_area(m, endoff,
|
|
|
|
DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
|
1999-04-05 19:38:30 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Set valid, clear dirty bits. If validating the entire
|
1999-12-12 03:19:33 +00:00
|
|
|
* page we can safely clear the pmap modify bit. We also
|
2006-08-13 00:11:09 +00:00
|
|
|
* use this opportunity to clear the VPO_NOSYNC flag. If a process
|
1999-12-12 03:19:33 +00:00
|
|
|
* takes a write fault on a MAP_NOSYNC memory area the flag will
|
|
|
|
* be set again.
|
2001-12-14 01:16:57 +00:00
|
|
|
*
|
|
|
|
* We set valid bits inclusive of any overlap, but we can only
|
|
|
|
* clear dirty bits for DEV_BSIZE chunks that are fully within
|
|
|
|
* the range.
|
1999-04-05 19:38:30 +00:00
|
|
|
*/
|
|
|
|
pagebits = vm_page_bits(base, size);
|
1995-09-03 19:57:25 +00:00
|
|
|
m->valid |= pagebits;
|
2001-12-14 01:16:57 +00:00
|
|
|
#if 0 /* NOT YET */
|
|
|
|
if ((frag = base & (DEV_BSIZE - 1)) != 0) {
|
|
|
|
frag = DEV_BSIZE - frag;
|
|
|
|
base += frag;
|
|
|
|
size -= frag;
|
|
|
|
if (size < 0)
|
|
|
|
size = 0;
|
|
|
|
}
|
|
|
|
pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
|
|
|
|
#endif
|
1995-09-03 19:57:25 +00:00
|
|
|
m->dirty &= ~pagebits;
|
1999-12-12 03:19:33 +00:00
|
|
|
if (base == 0 && size == PAGE_SIZE) {
|
2000-05-21 12:50:18 +00:00
|
|
|
pmap_clear_modify(m);
|
2006-08-13 00:11:09 +00:00
|
|
|
m->oflags &= ~VPO_NOSYNC;
|
1999-12-12 03:19:33 +00:00
|
|
|
}
|
1995-09-03 19:57:25 +00:00
|
|
|
}
|
|
|
|
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_clear_dirty(vm_page_t m, int base, int size)
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
{
|
2003-08-23 18:11:53 +00:00
|
|
|
|
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
m->dirty &= ~vm_page_bits(base, size);
|
|
|
|
}
|
|
|
|
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
/*
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
* vm_page_set_invalid:
|
|
|
|
*
|
|
|
|
* Invalidates DEV_BSIZE'd chunks within a page. Both the
|
|
|
|
* valid and dirty bits for the effected areas are cleared.
|
|
|
|
*
|
|
|
|
* May not block.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*/
|
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_set_invalid(vm_page_t m, int base, int size)
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
{
|
|
|
|
int bits;
|
|
|
|
|
2003-10-05 06:58:07 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
bits = vm_page_bits(base, size);
|
2003-08-23 18:11:53 +00:00
|
|
|
mtx_assert(&vm_page_queue_mtx, MA_OWNED);
|
2006-01-24 07:21:38 +00:00
|
|
|
if (m->valid == VM_PAGE_BITS_ALL && bits != 0)
|
|
|
|
pmap_remove_all(m);
|
The VFS/BIO subsystem contained a number of hacks in order to optimize
piecemeal, middle-of-file writes for NFS. These hacks have caused no
end of trouble, especially when combined with mmap(). I've removed
them. Instead, NFS will issue a read-before-write to fully
instantiate the struct buf containing the write. NFS does, however,
optimize piecemeal appends to files. For most common file operations,
you will not notice the difference. The sole remaining fragment in
the VFS/BIO system is b_dirtyoff/end, which NFS uses to avoid cache
coherency issues with read-merge-write style operations. NFS also
optimizes the write-covers-entire-buffer case by avoiding the
read-before-write. There is quite a bit of room for further
optimization in these areas.
The VM system marks pages fully-valid (AKA vm_page_t->valid =
VM_PAGE_BITS_ALL) in several places, most noteably in vm_fault. This
is not correct operation. The vm_pager_get_pages() code is now
responsible for marking VM pages all-valid. A number of VM helper
routines have been added to aid in zeroing-out the invalid portions of
a VM page prior to the page being marked all-valid. This operation is
necessary to properly support mmap(). The zeroing occurs most often
when dealing with file-EOF situations. Several bugs have been fixed
in the NFS subsystem, including bits handling file and directory EOF
situations and buf->b_flags consistancy issues relating to clearing
B_ERROR & B_INVAL, and handling B_DONE.
getblk() and allocbuf() have been rewritten. B_CACHE operation is now
formally defined in comments and more straightforward in
implementation. B_CACHE for VMIO buffers is based on the validity of
the backing store. B_CACHE for non-VMIO buffers is based simply on
whether the buffer is B_INVAL or not (B_CACHE set if B_INVAL clear,
and vise-versa). biodone() is now responsible for setting B_CACHE
when a successful read completes. B_CACHE is also set when a bdwrite()
is initiated and when a bwrite() is initiated. VFS VOP_BWRITE
routines (there are only two - nfs_bwrite() and bwrite()) are now
expected to set B_CACHE. This means that bowrite() and bawrite() also
set B_CACHE indirectly.
There are a number of places in the code which were previously using
buf->b_bufsize (which is DEV_BSIZE aligned) when they should have
been using buf->b_bcount. These have been fixed. getblk() now clears
B_DONE on return because the rest of the system is so bad about
dealing with B_DONE.
Major fixes to NFS/TCP have been made. A server-side bug could cause
requests to be lost by the server due to nfs_realign() overwriting
other rpc's in the same TCP mbuf chain. The server's kernel must be
recompiled to get the benefit of the fixes.
Submitted by: Matthew Dillon <dillon@apollo.backplane.com>
1999-05-02 23:57:16 +00:00
|
|
|
m->valid &= ~bits;
|
|
|
|
m->dirty &= ~bits;
|
Some VM improvements, including elimination of alot of Sig-11
problems. Tor Egge and others have helped with various VM bugs
lately, but don't blame him -- blame me!!!
pmap.c:
1) Create an object for kernel page table allocations. This
fixes a bogus allocation method previously used for such, by
grabbing pages from the kernel object, using bogus pindexes.
(This was a code cleanup, and perhaps a minor system stability
issue.)
pmap.c:
2) Pre-set the modify and accessed bits when prudent. This will
decrease bus traffic under certain circumstances.
vfs_bio.c, vfs_cluster.c:
3) Rather than calculating the beginning virtual byte offset
multiple times, stick the offset into the buffer header, so
that the calculated offset can be reused. (Long long multiplies
are often expensive, and this is a probably unmeasurable performance
improvement, and code cleanup.)
vfs_bio.c:
4) Handle write recursion more intelligently (but not perfectly) so
that it is less likely to cause a system panic, and is also
much more robust.
vfs_bio.c:
5) getblk incorrectly wrote out blocks that are incorrectly sized.
The problem is fixed, and writes blocks out ONLY when B_DELWRI
is true.
vfs_bio.c:
6) Check that already constituted buffers have fully valid pages. If
not, then make sure that the B_CACHE bit is not set. (This was
a major source of Sig-11 type problems.)
vfs_bio.c:
7) Fix a potential system deadlock due to an incorrectly specified
sleep priority while waiting for a buffer write operation. The
change that I made opens the system up to serious problems, and
we need to examine the issue of process sleep priorities.
vfs_cluster.c, vfs_bio.c:
8) Make clustered reads work more correctly (and more completely)
when buffers are already constituted, but not fully valid.
(This was another system reliability issue.)
vfs_subr.c, ffs_inode.c:
9) Create a vtruncbuf function, which is used by filesystems that
can truncate files. The vinvalbuf forced a file sync type operation,
while vtruncbuf only invalidates the buffers past the new end of file,
and also invalidates the appropriate pages. (This was a system reliabiliy
and performance issue.)
10) Modify FFS to use vtruncbuf.
vm_object.c:
11) Make the object rundown mechanism for OBJT_VNODE type objects work
more correctly. Included in that fix, create pager entries for
the OBJT_DEAD pager type, so that paging requests that might slip
in during race conditions are properly handled. (This was a system
reliability issue.)
vm_page.c:
12) Make some of the page validation routines be a little less picky
about arguments passed to them. Also, support page invalidation
change the object generation count so that we handle generation
counts a little more robustly.
vm_pageout.c:
13) Further reduce pageout daemon activity when the system doesn't
need help from it. There should be no additional performance
decrease even when the pageout daemon is running. (This was
a significant performance issue.)
vnode_pager.c:
14) Teach the vnode pager to handle race conditions during vnode
deallocations.
1998-03-16 01:56:03 +00:00
|
|
|
m->object->generation++;
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
1999-04-05 19:38:30 +00:00
|
|
|
* vm_page_zero_invalid()
|
|
|
|
*
|
|
|
|
* The kernel assumes that the invalid portions of a page contain
|
|
|
|
* garbage, but such pages can be mapped into memory by user code.
|
|
|
|
* When this occurs, we must zero out the non-valid portions of the
|
|
|
|
* page so user code sees what it expects.
|
|
|
|
*
|
|
|
|
* Pages are most often semi-valid when the end of a file is mapped
|
|
|
|
* into memory and the file's size is not page aligned.
|
|
|
|
*/
|
|
|
|
void
|
|
|
|
vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
|
|
|
|
{
|
|
|
|
int b;
|
|
|
|
int i;
|
|
|
|
|
2003-10-04 21:56:27 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
1999-04-05 19:38:30 +00:00
|
|
|
/*
|
|
|
|
* Scan the valid bits looking for invalid sections that
|
|
|
|
* must be zerod. Invalid sub-DEV_BSIZE'd areas ( where the
|
|
|
|
* valid bit may be set ) have already been zerod by
|
|
|
|
* vm_page_set_validclean().
|
|
|
|
*/
|
|
|
|
for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
|
|
|
|
if (i == (PAGE_SIZE / DEV_BSIZE) ||
|
|
|
|
(m->valid & (1 << i))
|
|
|
|
) {
|
|
|
|
if (i > b) {
|
2002-04-15 16:00:03 +00:00
|
|
|
pmap_zero_page_area(m,
|
|
|
|
b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
|
1999-04-05 19:38:30 +00:00
|
|
|
}
|
|
|
|
b = i + 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* setvalid is TRUE when we can safely set the zero'd areas
|
|
|
|
* as being valid. We can do this if there are no cache consistancy
|
|
|
|
* issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
|
|
|
|
*/
|
|
|
|
if (setvalid)
|
|
|
|
m->valid = VM_PAGE_BITS_ALL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_page_is_valid:
|
|
|
|
*
|
|
|
|
* Is (partial) page valid? Note that the case where size == 0
|
|
|
|
* will return FALSE in the degenerate case where the page is
|
|
|
|
* entirely invalid, and TRUE otherwise.
|
|
|
|
*
|
|
|
|
* May not block.
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
*/
|
|
|
|
int
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_is_valid(vm_page_t m, int base, int size)
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
{
|
1995-04-09 06:03:56 +00:00
|
|
|
int bits = vm_page_bits(base, size);
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
|
2003-10-04 19:23:29 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(m->object, MA_OWNED);
|
1995-04-09 06:03:56 +00:00
|
|
|
if (m->valid && ((m->valid & bits) == bits))
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
return 1;
|
|
|
|
else
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
1998-12-23 01:52:47 +00:00
|
|
|
/*
|
|
|
|
* update dirty bits from pmap/mmu. May not block.
|
|
|
|
*/
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
void
|
2001-07-04 20:15:18 +00:00
|
|
|
vm_page_test_dirty(vm_page_t m)
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
{
|
2000-05-21 12:50:18 +00:00
|
|
|
if ((m->dirty != VM_PAGE_BITS_ALL) && pmap_is_modified(m)) {
|
1999-01-24 06:00:31 +00:00
|
|
|
vm_page_dirty(m);
|
1996-07-30 03:08:57 +00:00
|
|
|
}
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
}
|
|
|
|
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
int so_zerocp_fullpage = 0;
|
|
|
|
|
Correct a problem in the ZERO_COPY_SOCKETS option, specifically, in
vm_page_cowfault(). Initially, if vm_page_cowfault() sleeps, the given
page is wired, preventing it from being recycled. However, when
transmission of the page completes, the page is unwired and returned to
the page queues. At that point, the page is not in any special state
that prevents it from being recycled. Consequently, vm_page_cowfault()
should verify that the page is still held by the same vm object before
retrying the replacement of the page. Note: The containing object is,
however, safe from being recycled by virtue of having a non-zero
paging-in-progress count.
While I'm here, add some assertions and comments.
Approved by: re (rwatson)
MFC After: 3 weeks
2007-07-10 18:41:34 +00:00
|
|
|
/*
|
|
|
|
* Replace the given page with a copy. The copied page assumes
|
|
|
|
* the portion of the given page's "wire_count" that is not the
|
|
|
|
* responsibility of this copy-on-write mechanism.
|
|
|
|
*
|
|
|
|
* The object containing the given page must have a non-zero
|
|
|
|
* paging-in-progress count and be locked.
|
|
|
|
*/
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
void
|
|
|
|
vm_page_cowfault(vm_page_t m)
|
|
|
|
{
|
|
|
|
vm_page_t mnew;
|
|
|
|
vm_object_t object;
|
|
|
|
vm_pindex_t pindex;
|
|
|
|
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
object = m->object;
|
Correct a problem in the ZERO_COPY_SOCKETS option, specifically, in
vm_page_cowfault(). Initially, if vm_page_cowfault() sleeps, the given
page is wired, preventing it from being recycled. However, when
transmission of the page completes, the page is unwired and returned to
the page queues. At that point, the page is not in any special state
that prevents it from being recycled. Consequently, vm_page_cowfault()
should verify that the page is still held by the same vm object before
retrying the replacement of the page. Note: The containing object is,
however, safe from being recycled by virtue of having a non-zero
paging-in-progress count.
While I'm here, add some assertions and comments.
Approved by: re (rwatson)
MFC After: 3 weeks
2007-07-10 18:41:34 +00:00
|
|
|
VM_OBJECT_LOCK_ASSERT(object, MA_OWNED);
|
|
|
|
KASSERT(object->paging_in_progress != 0,
|
|
|
|
("vm_page_cowfault: object %p's paging-in-progress count is zero.",
|
|
|
|
object));
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
pindex = m->pindex;
|
|
|
|
|
|
|
|
retry_alloc:
|
2005-11-08 06:33:21 +00:00
|
|
|
pmap_remove_all(m);
|
2010-05-05 18:16:06 +00:00
|
|
|
vm_page_unlock_queues();
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
vm_page_remove(m);
|
2006-08-10 04:48:29 +00:00
|
|
|
mnew = vm_page_alloc(object, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY);
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
if (mnew == NULL) {
|
|
|
|
vm_page_insert(m, object, pindex);
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_unlock(m);
|
2003-06-22 21:35:41 +00:00
|
|
|
VM_OBJECT_UNLOCK(object);
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
VM_WAIT;
|
2003-06-22 21:35:41 +00:00
|
|
|
VM_OBJECT_LOCK(object);
|
Correct a problem in the ZERO_COPY_SOCKETS option, specifically, in
vm_page_cowfault(). Initially, if vm_page_cowfault() sleeps, the given
page is wired, preventing it from being recycled. However, when
transmission of the page completes, the page is unwired and returned to
the page queues. At that point, the page is not in any special state
that prevents it from being recycled. Consequently, vm_page_cowfault()
should verify that the page is still held by the same vm object before
retrying the replacement of the page. Note: The containing object is,
however, safe from being recycled by virtue of having a non-zero
paging-in-progress count.
While I'm here, add some assertions and comments.
Approved by: re (rwatson)
MFC After: 3 weeks
2007-07-10 18:41:34 +00:00
|
|
|
if (m == vm_page_lookup(object, pindex)) {
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_lock(m);
|
Correct a problem in the ZERO_COPY_SOCKETS option, specifically, in
vm_page_cowfault(). Initially, if vm_page_cowfault() sleeps, the given
page is wired, preventing it from being recycled. However, when
transmission of the page completes, the page is unwired and returned to
the page queues. At that point, the page is not in any special state
that prevents it from being recycled. Consequently, vm_page_cowfault()
should verify that the page is still held by the same vm object before
retrying the replacement of the page. Note: The containing object is,
however, safe from being recycled by virtue of having a non-zero
paging-in-progress count.
While I'm here, add some assertions and comments.
Approved by: re (rwatson)
MFC After: 3 weeks
2007-07-10 18:41:34 +00:00
|
|
|
vm_page_lock_queues();
|
|
|
|
goto retry_alloc;
|
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Page disappeared during the wait.
|
|
|
|
*/
|
|
|
|
return;
|
|
|
|
}
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
if (m->cow == 0) {
|
|
|
|
/*
|
|
|
|
* check to see if we raced with an xmit complete when
|
|
|
|
* waiting to allocate a page. If so, put things back
|
|
|
|
* the way they were
|
|
|
|
*/
|
2010-05-05 18:16:06 +00:00
|
|
|
vm_page_unlock(m);
|
|
|
|
vm_page_lock(mnew);
|
|
|
|
vm_page_lock_queues();
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
vm_page_free(mnew);
|
2010-05-05 18:16:06 +00:00
|
|
|
vm_page_unlock_queues();
|
|
|
|
vm_page_unlock(mnew);
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
vm_page_insert(m, object, pindex);
|
|
|
|
} else { /* clear COW & copy page */
|
2003-10-08 05:35:12 +00:00
|
|
|
if (!so_zerocp_fullpage)
|
|
|
|
pmap_copy_page(m, mnew);
|
|
|
|
mnew->valid = VM_PAGE_BITS_ALL;
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
vm_page_dirty(mnew);
|
2005-11-01 04:30:21 +00:00
|
|
|
mnew->wire_count = m->wire_count - m->cow;
|
|
|
|
m->wire_count = m->cow;
|
2010-05-05 18:16:06 +00:00
|
|
|
vm_page_unlock(m);
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
vm_page_cowclear(vm_page_t m)
|
|
|
|
{
|
|
|
|
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
if (m->cow) {
|
2002-09-02 04:04:12 +00:00
|
|
|
m->cow--;
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
/*
|
|
|
|
* let vm_fault add back write permission lazily
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
/*
|
|
|
|
* sf_buf_free() will free the page, so we needn't do it here
|
|
|
|
*/
|
|
|
|
}
|
|
|
|
|
2009-01-03 13:24:08 +00:00
|
|
|
int
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
vm_page_cowsetup(vm_page_t m)
|
|
|
|
{
|
2002-09-02 04:04:12 +00:00
|
|
|
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_lock_assert(m, MA_OWNED);
|
2009-01-03 13:24:08 +00:00
|
|
|
if (m->cow == USHRT_MAX - 1)
|
|
|
|
return (EBUSY);
|
2002-09-02 04:04:12 +00:00
|
|
|
m->cow++;
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_lock_queues();
|
2006-08-01 19:06:06 +00:00
|
|
|
pmap_remove_write(m);
|
2010-05-04 15:55:41 +00:00
|
|
|
vm_page_unlock_queues();
|
2009-01-03 13:24:08 +00:00
|
|
|
return (0);
|
At long last, commit the zero copy sockets code.
MAKEDEV: Add MAKEDEV glue for the ti(4) device nodes.
ti.4: Update the ti(4) man page to include information on the
TI_JUMBO_HDRSPLIT and TI_PRIVATE_JUMBOS kernel options,
and also include information about the new character
device interface and the associated ioctls.
man9/Makefile: Add jumbo.9 and zero_copy.9 man pages and associated
links.
jumbo.9: New man page describing the jumbo buffer allocator
interface and operation.
zero_copy.9: New man page describing the general characteristics of
the zero copy send and receive code, and what an
application author should do to take advantage of the
zero copy functionality.
NOTES: Add entries for ZERO_COPY_SOCKETS, TI_PRIVATE_JUMBOS,
TI_JUMBO_HDRSPLIT, MSIZE, and MCLSHIFT.
conf/files: Add uipc_jumbo.c and uipc_cow.c.
conf/options: Add the 5 options mentioned above.
kern_subr.c: Receive side zero copy implementation. This takes
"disposable" pages attached to an mbuf, gives them to
a user process, and then recycles the user's page.
This is only active when ZERO_COPY_SOCKETS is turned on
and the kern.ipc.zero_copy.receive sysctl variable is
set to 1.
uipc_cow.c: Send side zero copy functions. Takes a page written
by the user and maps it copy on write and assigns it
kernel virtual address space. Removes copy on write
mapping once the buffer has been freed by the network
stack.
uipc_jumbo.c: Jumbo disposable page allocator code. This allocates
(optionally) disposable pages for network drivers that
want to give the user the option of doing zero copy
receive.
uipc_socket.c: Add kern.ipc.zero_copy.{send,receive} sysctls that are
enabled if ZERO_COPY_SOCKETS is turned on.
Add zero copy send support to sosend() -- pages get
mapped into the kernel instead of getting copied if
they meet size and alignment restrictions.
uipc_syscalls.c:Un-staticize some of the sf* functions so that they
can be used elsewhere. (uipc_cow.c)
if_media.c: In the SIOCGIFMEDIA ioctl in ifmedia_ioctl(), avoid
calling malloc() with M_WAITOK. Return an error if
the M_NOWAIT malloc fails.
The ti(4) driver and the wi(4) driver, at least, call
this with a mutex held. This causes witness warnings
for 'ifconfig -a' with a wi(4) or ti(4) board in the
system. (I've only verified for ti(4)).
ip_output.c: Fragment large datagrams so that each segment contains
a multiple of PAGE_SIZE amount of data plus headers.
This allows the receiver to potentially do page
flipping on receives.
if_ti.c: Add zero copy receive support to the ti(4) driver. If
TI_PRIVATE_JUMBOS is not defined, it now uses the
jumbo(9) buffer allocator for jumbo receive buffers.
Add a new character device interface for the ti(4)
driver for the new debugging interface. This allows
(a patched version of) gdb to talk to the Tigon board
and debug the firmware. There are also a few additional
debugging ioctls available through this interface.
Add header splitting support to the ti(4) driver.
Tweak some of the default interrupt coalescing
parameters to more useful defaults.
Add hooks for supporting transmit flow control, but
leave it turned off with a comment describing why it
is turned off.
if_tireg.h: Change the firmware rev to 12.4.11, since we're really
at 12.4.11 plus fixes from 12.4.13.
Add defines needed for debugging.
Remove the ti_stats structure, it is now defined in
sys/tiio.h.
ti_fw.h: 12.4.11 firmware.
ti_fw2.h: 12.4.11 firmware, plus selected fixes from 12.4.13,
and my header splitting patches. Revision 12.4.13
doesn't handle 10/100 negotiation properly. (This
firmware is the same as what was in the tree previously,
with the addition of header splitting support.)
sys/jumbo.h: Jumbo buffer allocator interface.
sys/mbuf.h: Add a new external mbuf type, EXT_DISPOSABLE, to
indicate that the payload buffer can be thrown away /
flipped to a userland process.
socketvar.h: Add prototype for socow_setup.
tiio.h: ioctl interface to the character portion of the ti(4)
driver, plus associated structure/type definitions.
uio.h: Change prototype for uiomoveco() so that we'll know
whether the source page is disposable.
ufs_readwrite.c:Update for new prototype of uiomoveco().
vm_fault.c: In vm_fault(), check to see whether we need to do a page
based copy on write fault.
vm_object.c: Add a new function, vm_object_allocate_wait(). This
does the same thing that vm_object allocate does, except
that it gives the caller the opportunity to specify whether
it should wait on the uma_zalloc() of the object structre.
This allows vm objects to be allocated while holding a
mutex. (Without generating WITNESS warnings.)
vm_object_allocate() is implemented as a call to
vm_object_allocate_wait() with the malloc flag set to
M_WAITOK.
vm_object.h: Add prototype for vm_object_allocate_wait().
vm_page.c: Add page-based copy on write setup, clear and fault
routines.
vm_page.h: Add page based COW function prototypes and variable in
the vm_page structure.
Many thanks to Drew Gallatin, who wrote the zero copy send and receive
code, and to all the other folks who have tested and reviewed this code
over the years.
2002-06-26 03:37:47 +00:00
|
|
|
}
|
|
|
|
|
1996-09-14 11:54:59 +00:00
|
|
|
#include "opt_ddb.h"
|
1995-04-16 09:59:16 +00:00
|
|
|
#ifdef DDB
|
1996-09-14 11:54:59 +00:00
|
|
|
#include <sys/kernel.h>
|
|
|
|
|
|
|
|
#include <ddb/ddb.h>
|
|
|
|
|
|
|
|
DB_SHOW_COMMAND(page, vm_page_print_page_info)
|
These changes embody the support of the fully coherent merged VM buffer cache,
much higher filesystem I/O performance, and much better paging performance. It
represents the culmination of over 6 months of R&D.
The majority of the merged VM/cache work is by John Dyson.
The following highlights the most significant changes. Additionally, there are
(mostly minor) changes to the various filesystem modules (nfs, msdosfs, etc) to
support the new VM/buffer scheme.
vfs_bio.c:
Significant rewrite of most of vfs_bio to support the merged VM buffer cache
scheme. The scheme is almost fully compatible with the old filesystem
interface. Significant improvement in the number of opportunities for write
clustering.
vfs_cluster.c, vfs_subr.c
Upgrade and performance enhancements in vfs layer code to support merged
VM/buffer cache. Fixup of vfs_cluster to eliminate the bogus pagemove stuff.
vm_object.c:
Yet more improvements in the collapse code. Elimination of some windows that
can cause list corruption.
vm_pageout.c:
Fixed it, it really works better now. Somehow in 2.0, some "enhancements"
broke the code. This code has been reworked from the ground-up.
vm_fault.c, vm_page.c, pmap.c, vm_object.c
Support for small-block filesystems with merged VM/buffer cache scheme.
pmap.c vm_map.c
Dynamic kernel VM size, now we dont have to pre-allocate excessive numbers of
kernel PTs.
vm_glue.c
Much simpler and more effective swapping code. No more gratuitous swapping.
proc.h
Fixed the problem that the p_lock flag was not being cleared on a fork.
swap_pager.c, vnode_pager.c
Removal of old vfs_bio cruft to support the past pseudo-coherency. Now the
code doesn't need it anymore.
machdep.c
Changes to better support the parameter values for the merged VM/buffer cache
scheme.
machdep.c, kern_exec.c, vm_glue.c
Implemented a seperate submap for temporary exec string space and another one
to contain process upages. This eliminates all map fragmentation problems
that previously existed.
ffs_inode.c, ufs_inode.c, ufs_readwrite.c
Changes for merged VM/buffer cache. Add "bypass" support for sneaking in on
busy buffers.
Submitted by: John Dyson and David Greenman
1995-01-09 16:06:02 +00:00
|
|
|
{
|
2007-05-31 22:52:15 +00:00
|
|
|
db_printf("cnt.v_free_count: %d\n", cnt.v_free_count);
|
|
|
|
db_printf("cnt.v_cache_count: %d\n", cnt.v_cache_count);
|
|
|
|
db_printf("cnt.v_inactive_count: %d\n", cnt.v_inactive_count);
|
|
|
|
db_printf("cnt.v_active_count: %d\n", cnt.v_active_count);
|
|
|
|
db_printf("cnt.v_wire_count: %d\n", cnt.v_wire_count);
|
|
|
|
db_printf("cnt.v_free_reserved: %d\n", cnt.v_free_reserved);
|
|
|
|
db_printf("cnt.v_free_min: %d\n", cnt.v_free_min);
|
|
|
|
db_printf("cnt.v_free_target: %d\n", cnt.v_free_target);
|
|
|
|
db_printf("cnt.v_cache_min: %d\n", cnt.v_cache_min);
|
|
|
|
db_printf("cnt.v_inactive_target: %d\n", cnt.v_inactive_target);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1996-09-08 20:44:49 +00:00
|
|
|
|
1996-09-14 11:54:59 +00:00
|
|
|
DB_SHOW_COMMAND(pageq, vm_page_print_pageq_info)
|
1996-09-08 20:44:49 +00:00
|
|
|
{
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
|
1996-09-14 11:54:59 +00:00
|
|
|
db_printf("PQ_FREE:");
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
db_printf(" %d", cnt.v_free_count);
|
1996-09-14 11:54:59 +00:00
|
|
|
db_printf("\n");
|
1996-09-08 20:44:49 +00:00
|
|
|
|
1996-09-14 11:54:59 +00:00
|
|
|
db_printf("PQ_CACHE:");
|
Change the management of cached pages (PQ_CACHE) in two fundamental
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)
2007-09-25 06:25:06 +00:00
|
|
|
db_printf(" %d", cnt.v_cache_count);
|
1996-09-14 11:54:59 +00:00
|
|
|
db_printf("\n");
|
1996-09-08 20:44:49 +00:00
|
|
|
|
1996-09-14 11:54:59 +00:00
|
|
|
db_printf("PQ_ACTIVE: %d, PQ_INACTIVE: %d\n",
|
Enable the new physical memory allocator.
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
2007-06-16 04:57:06 +00:00
|
|
|
*vm_page_queues[PQ_ACTIVE].cnt,
|
|
|
|
*vm_page_queues[PQ_INACTIVE].cnt);
|
1996-09-08 20:44:49 +00:00
|
|
|
}
|
1996-09-14 11:54:59 +00:00
|
|
|
#endif /* DDB */
|