2005-01-07 02:29:27 +00:00
|
|
|
/*-
|
1994-05-24 10:09:53 +00:00
|
|
|
* Copyright (c) 1991, 1993
|
|
|
|
* The Regents of the University of California. All rights reserved.
|
1994-05-25 09:21:21 +00:00
|
|
|
* Copyright (c) 1994 John S. Dyson
|
|
|
|
* All rights reserved.
|
|
|
|
* Copyright (c) 1994 David Greenman
|
|
|
|
* 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.
|
|
|
|
* 3. All advertising materials mentioning features or use of this software
|
2000-03-27 20:41:17 +00:00
|
|
|
* must display the following acknowledgement:
|
1994-05-24 10:09:53 +00:00
|
|
|
* This product includes software developed by the University of
|
|
|
|
* California, Berkeley and its contributors.
|
|
|
|
* 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.
|
|
|
|
*
|
1994-08-02 07:55:43 +00:00
|
|
|
* from: @(#)vm_fault.c 8.4 (Berkeley) 1/12/94
|
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.
|
|
|
|
*/
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Page fault handling module.
|
|
|
|
*/
|
2003-06-11 23:50:51 +00:00
|
|
|
|
|
|
|
#include <sys/cdefs.h>
|
|
|
|
__FBSDID("$FreeBSD$");
|
|
|
|
|
2012-04-05 17:13:14 +00:00
|
|
|
#include "opt_ktrace.h"
|
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>
|
2001-05-22 00:56:25 +00:00
|
|
|
#include <sys/kernel.h>
|
2001-05-01 08:13:21 +00:00
|
|
|
#include <sys/lock.h>
|
1994-05-25 09:21:21 +00:00
|
|
|
#include <sys/proc.h>
|
|
|
|
#include <sys/resourcevar.h>
|
2013-03-09 02:32:23 +00:00
|
|
|
#include <sys/rwlock.h>
|
2001-05-19 01:28:09 +00:00
|
|
|
#include <sys/sysctl.h>
|
2001-05-22 00:56:25 +00:00
|
|
|
#include <sys/vmmeter.h>
|
|
|
|
#include <sys/vnode.h>
|
2012-04-05 17:13:14 +00:00
|
|
|
#ifdef KTRACE
|
|
|
|
#include <sys/ktrace.h>
|
|
|
|
#endif
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
#include <vm/vm.h>
|
1995-12-07 12:48:31 +00:00
|
|
|
#include <vm/vm_param.h>
|
|
|
|
#include <vm/pmap.h>
|
|
|
|
#include <vm/vm_map.h>
|
|
|
|
#include <vm/vm_object.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
#include <vm/vm_page.h>
|
|
|
|
#include <vm/vm_pageout.h>
|
1994-11-06 09:55:31 +00:00
|
|
|
#include <vm/vm_kern.h>
|
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
1995-07-13 08:48:48 +00:00
|
|
|
#include <vm/vm_pager.h>
|
1995-12-07 12:48:31 +00:00
|
|
|
#include <vm/vm_extern.h>
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2003-10-03 22:46:53 +00:00
|
|
|
#define PFBAK 4
|
|
|
|
#define PFFOR 4
|
|
|
|
|
2002-03-19 22:20:14 +00:00
|
|
|
static int vm_fault_additional_pages(vm_page_t, int, int, vm_page_t *, int *);
|
1994-05-25 09:21:21 +00:00
|
|
|
|
2012-05-10 15:16:42 +00:00
|
|
|
#define VM_FAULT_READ_BEHIND 8
|
|
|
|
#define VM_FAULT_READ_MAX (1 + VM_FAULT_READ_AHEAD_MAX)
|
|
|
|
#define VM_FAULT_NINCR (VM_FAULT_READ_MAX / VM_FAULT_READ_BEHIND)
|
|
|
|
#define VM_FAULT_SUM (VM_FAULT_NINCR * (VM_FAULT_NINCR + 1) / 2)
|
|
|
|
#define VM_FAULT_CACHE_BEHIND (VM_FAULT_READ_BEHIND * VM_FAULT_SUM)
|
1994-05-25 09:21:21 +00:00
|
|
|
|
1998-03-07 20:45:47 +00:00
|
|
|
struct faultstate {
|
|
|
|
vm_page_t m;
|
|
|
|
vm_object_t object;
|
|
|
|
vm_pindex_t pindex;
|
|
|
|
vm_page_t first_m;
|
|
|
|
vm_object_t first_object;
|
|
|
|
vm_pindex_t first_pindex;
|
|
|
|
vm_map_t map;
|
|
|
|
vm_map_entry_t entry;
|
2002-03-18 15:08:09 +00:00
|
|
|
int lookup_still_valid;
|
1998-03-07 20:45:47 +00:00
|
|
|
struct vnode *vp;
|
|
|
|
};
|
|
|
|
|
2012-05-10 15:16:42 +00:00
|
|
|
static void vm_fault_cache_behind(const struct faultstate *fs, int distance);
|
2014-02-02 20:21:53 +00:00
|
|
|
static void vm_fault_prefault(const struct faultstate *fs, vm_offset_t addra,
|
|
|
|
int faultcount, int reqpage);
|
2012-05-10 15:16:42 +00:00
|
|
|
|
2006-03-08 06:31:46 +00:00
|
|
|
static inline void
|
1998-03-07 20:45:47 +00:00
|
|
|
release_page(struct faultstate *fs)
|
|
|
|
{
|
2009-02-08 19:37:01 +00:00
|
|
|
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_xunbusy(fs->m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(fs->m);
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_page_deactivate(fs->m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs->m);
|
1998-03-07 20:45:47 +00:00
|
|
|
fs->m = NULL;
|
|
|
|
}
|
|
|
|
|
2006-03-08 06:31:46 +00:00
|
|
|
static inline void
|
1998-03-07 20:45:47 +00:00
|
|
|
unlock_map(struct faultstate *fs)
|
|
|
|
{
|
2009-02-08 19:37:01 +00:00
|
|
|
|
2002-03-18 15:08:09 +00:00
|
|
|
if (fs->lookup_still_valid) {
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_map_lookup_done(fs->map, fs->entry);
|
2002-03-18 15:08:09 +00:00
|
|
|
fs->lookup_still_valid = FALSE;
|
1998-03-07 20:45:47 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
static void
|
2004-12-24 19:31:54 +00:00
|
|
|
unlock_and_deallocate(struct faultstate *fs)
|
1998-03-07 20:45:47 +00:00
|
|
|
{
|
2003-06-22 21:35:41 +00:00
|
|
|
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_object_pip_wakeup(fs->object);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs->object);
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs->object != fs->first_object) {
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(fs->first_object);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(fs->first_m);
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_page_free(fs->first_m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs->first_m);
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_object_pip_wakeup(fs->first_object);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs->first_object);
|
1998-03-07 20:45:47 +00:00
|
|
|
fs->first_m = NULL;
|
|
|
|
}
|
2004-12-24 19:31:54 +00:00
|
|
|
vm_object_deallocate(fs->first_object);
|
1998-03-07 20:45:47 +00:00
|
|
|
unlock_map(fs);
|
|
|
|
if (fs->vp != NULL) {
|
2001-07-04 16:20:28 +00:00
|
|
|
vput(fs->vp);
|
1998-03-07 20:45:47 +00:00
|
|
|
fs->vp = NULL;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
1999-09-21 00:36:16 +00:00
|
|
|
/*
|
|
|
|
* TRYPAGER - used by vm_fault to calculate whether the pager for the
|
|
|
|
* current object *might* contain the page.
|
|
|
|
*
|
|
|
|
* default objects are zero-fill, there is no real pager.
|
|
|
|
*/
|
|
|
|
#define TRYPAGER (fs.object->type != OBJT_DEFAULT && \
|
2009-11-18 18:05:54 +00:00
|
|
|
((fault_flags & VM_FAULT_CHANGE_WIRING) == 0 || wired))
|
1999-09-21 00:36:16 +00:00
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* vm_fault:
|
|
|
|
*
|
2000-03-26 15:20:23 +00:00
|
|
|
* Handle a page fault occurring at the given address,
|
1994-05-24 10:09:53 +00:00
|
|
|
* requiring the given permissions, in the map specified.
|
|
|
|
* If successful, the page is inserted into the
|
|
|
|
* associated physical map.
|
|
|
|
*
|
|
|
|
* NOTE: the given address should be truncated to the
|
|
|
|
* proper page address.
|
|
|
|
*
|
|
|
|
* KERN_SUCCESS is returned if the page fault is handled; otherwise,
|
|
|
|
* a standard error specifying why the fault is fatal is returned.
|
|
|
|
*
|
|
|
|
* The map in question must be referenced, and remains so.
|
2001-07-04 16:20:28 +00:00
|
|
|
* Caller may hold no locks.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
|
|
|
int
|
2001-05-19 01:28:09 +00:00
|
|
|
vm_fault(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
|
2010-12-20 22:49:31 +00:00
|
|
|
int fault_flags)
|
|
|
|
{
|
2012-04-05 17:13:14 +00:00
|
|
|
struct thread *td;
|
|
|
|
int result;
|
2010-12-20 22:49:31 +00:00
|
|
|
|
2012-04-05 17:13:14 +00:00
|
|
|
td = curthread;
|
|
|
|
if ((td->td_pflags & TDP_NOFAULTING) != 0)
|
2011-07-09 15:21:10 +00:00
|
|
|
return (KERN_PROTECTION_FAILURE);
|
2012-04-05 17:13:14 +00:00
|
|
|
#ifdef KTRACE
|
|
|
|
if (map != kernel_map && KTRPOINT(td, KTR_FAULT))
|
|
|
|
ktrfault(vaddr, fault_type);
|
|
|
|
#endif
|
2013-08-05 08:55:35 +00:00
|
|
|
result = vm_fault_hold(map, trunc_page(vaddr), fault_type, fault_flags,
|
|
|
|
NULL);
|
2012-04-05 17:13:14 +00:00
|
|
|
#ifdef KTRACE
|
|
|
|
if (map != kernel_map && KTRPOINT(td, KTR_FAULTEND))
|
|
|
|
ktrfaultend(result);
|
|
|
|
#endif
|
|
|
|
return (result);
|
2010-12-20 22:49:31 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
int
|
2013-08-05 08:55:35 +00:00
|
|
|
vm_fault_hold(vm_map_t map, vm_offset_t vaddr, vm_prot_t fault_type,
|
2010-12-20 22:49:31 +00:00
|
|
|
int fault_flags, vm_page_t *m_hold)
|
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
|
|
|
vm_prot_t prot;
|
2012-05-10 15:16:42 +00:00
|
|
|
long ahead, behind;
|
|
|
|
int alloc_req, era, faultcount, nera, reqpage, result;
|
|
|
|
boolean_t growstack, is_first_object_locked, wired;
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
int map_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
|
|
|
vm_object_t next_object;
|
2012-05-10 15:16:42 +00:00
|
|
|
vm_page_t marray[VM_FAULT_READ_MAX];
|
1998-03-07 20:45:47 +00:00
|
|
|
int hardfault;
|
|
|
|
struct faultstate fs;
|
2009-02-08 20:23:46 +00:00
|
|
|
struct vnode *vp;
|
|
|
|
int locked, error;
|
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
|
|
|
|
1998-03-07 20:45:47 +00:00
|
|
|
hardfault = 0;
|
2002-04-18 03:28:27 +00:00
|
|
|
growstack = TRUE;
|
2007-06-04 21:38:48 +00:00
|
|
|
PCPU_INC(cnt.v_vm_faults);
|
2009-02-08 20:23:46 +00:00
|
|
|
fs.vp = NULL;
|
2012-05-10 15:16:42 +00:00
|
|
|
faultcount = reqpage = 0;
|
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
|
|
|
RetryFault:;
|
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
|
|
|
* Find the backing store object and offset into it to begin the
|
|
|
|
* search.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1999-09-21 00:36:16 +00:00
|
|
|
fs.map = map;
|
2002-04-19 04:20:31 +00:00
|
|
|
result = vm_map_lookup(&fs.map, vaddr, fault_type, &fs.entry,
|
|
|
|
&fs.first_object, &fs.first_pindex, &prot, &wired);
|
|
|
|
if (result != KERN_SUCCESS) {
|
2009-11-18 18:05:54 +00:00
|
|
|
if (growstack && result == KERN_INVALID_ADDRESS &&
|
|
|
|
map != kernel_map) {
|
|
|
|
result = vm_map_growstack(curproc, vaddr);
|
|
|
|
if (result != KERN_SUCCESS)
|
|
|
|
return (KERN_FAILURE);
|
|
|
|
growstack = FALSE;
|
|
|
|
goto RetryFault;
|
1998-01-17 09:17:02 +00:00
|
|
|
}
|
2009-11-18 18:05:54 +00:00
|
|
|
return (result);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1995-04-09 06:03:56 +00:00
|
|
|
|
1998-03-07 20:45:47 +00:00
|
|
|
map_generation = fs.map->timestamp;
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.entry->eflags & MAP_ENTRY_NOFAULT) {
|
1996-11-30 22:41:49 +00:00
|
|
|
panic("vm_fault: fault on nofault entry, addr: %lx",
|
1998-07-11 12:07:52 +00:00
|
|
|
(u_long)vaddr);
|
1996-11-30 22:41:49 +00:00
|
|
|
}
|
|
|
|
|
2013-07-11 05:58:28 +00:00
|
|
|
if (fs.entry->eflags & MAP_ENTRY_IN_TRANSITION &&
|
|
|
|
fs.entry->wiring_thread != curthread) {
|
|
|
|
vm_map_unlock_read(fs.map);
|
|
|
|
vm_map_lock(fs.map);
|
|
|
|
if (vm_map_lookup_entry(fs.map, vaddr, &fs.entry) &&
|
|
|
|
(fs.entry->eflags & MAP_ENTRY_IN_TRANSITION)) {
|
|
|
|
fs.entry->eflags |= MAP_ENTRY_NEEDS_WAKEUP;
|
|
|
|
vm_map_unlock_and_wait(fs.map, 0);
|
|
|
|
} else
|
|
|
|
vm_map_unlock(fs.map);
|
|
|
|
goto RetryFault;
|
|
|
|
}
|
|
|
|
|
Make our v_usecount vnode reference count work identically to the
original BSD code. The association between the vnode and the vm_object
no longer includes reference counts. The major difference is that
vm_object's are no longer freed gratuitiously from the vnode, and so
once an object is created for the vnode, it will last as long as the
vnode does.
When a vnode object reference count is incremented, then the underlying
vnode reference count is incremented also. The two "objects" are now
more intimately related, and so the interactions are now much less
complex.
When vnodes are now normally placed onto the free queue with an object still
attached. The rundown of the object happens at vnode rundown time, and
happens with exactly the same filesystem semantics of the original VFS
code. There is absolutely no need for vnode_pager_uncache and other
travesties like that anymore.
A side-effect of these changes is that SMP locking should be much simpler,
the I/O copyin/copyout optimizations work, NFS should be more ponderable,
and further work on layered filesystems should be less frustrating, because
of the totally coherent management of the vnode objects and vnodes.
Please be careful with your system while running this code, but I would
greatly appreciate feedback as soon a reasonably possible.
1998-01-06 05:26:17 +00:00
|
|
|
/*
|
|
|
|
* Make a reference to this object to prevent its disposal while we
|
|
|
|
* are messing with it. Once we have the reference, the map is free
|
|
|
|
* to be diddled. Since objects reference their shadows (and copies),
|
|
|
|
* they will stay around as well.
|
2001-11-09 21:34:45 +00:00
|
|
|
*
|
|
|
|
* Bump the paging-in-progress count to prevent size changes (e.g.
|
|
|
|
* truncation operations) during I/O. This must be done after
|
|
|
|
* obtaining the vnode lock in order to avoid possible deadlocks.
|
Make our v_usecount vnode reference count work identically to the
original BSD code. The association between the vnode and the vm_object
no longer includes reference counts. The major difference is that
vm_object's are no longer freed gratuitiously from the vnode, and so
once an object is created for the vnode, it will last as long as the
vnode does.
When a vnode object reference count is incremented, then the underlying
vnode reference count is incremented also. The two "objects" are now
more intimately related, and so the interactions are now much less
complex.
When vnodes are now normally placed onto the free queue with an object still
attached. The rundown of the object happens at vnode rundown time, and
happens with exactly the same filesystem semantics of the original VFS
code. There is absolutely no need for vnode_pager_uncache and other
travesties like that anymore.
A side-effect of these changes is that SMP locking should be much simpler,
the I/O copyin/copyout optimizations work, NFS should be more ponderable,
and further work on layered filesystems should be less frustrating, because
of the totally coherent management of the vnode objects and vnodes.
Please be careful with your system while running this code, but I would
greatly appreciate feedback as soon a reasonably possible.
1998-01-06 05:26:17 +00:00
|
|
|
*/
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(fs.first_object);
|
2003-12-26 23:33:37 +00:00
|
|
|
vm_object_reference_locked(fs.first_object);
|
1998-08-06 08:33:19 +00:00
|
|
|
vm_object_pip_add(fs.first_object, 1);
|
Make our v_usecount vnode reference count work identically to the
original BSD code. The association between the vnode and the vm_object
no longer includes reference counts. The major difference is that
vm_object's are no longer freed gratuitiously from the vnode, and so
once an object is created for the vnode, it will last as long as the
vnode does.
When a vnode object reference count is incremented, then the underlying
vnode reference count is incremented also. The two "objects" are now
more intimately related, and so the interactions are now much less
complex.
When vnodes are now normally placed onto the free queue with an object still
attached. The rundown of the object happens at vnode rundown time, and
happens with exactly the same filesystem semantics of the original VFS
code. There is absolutely no need for vnode_pager_uncache and other
travesties like that anymore.
A side-effect of these changes is that SMP locking should be much simpler,
the I/O copyin/copyout optimizations work, NFS should be more ponderable,
and further work on layered filesystems should be less frustrating, because
of the totally coherent management of the vnode objects and vnodes.
Please be careful with your system while running this code, but I would
greatly appreciate feedback as soon a reasonably possible.
1998-01-06 05:26:17 +00:00
|
|
|
|
2002-03-18 15:08:09 +00:00
|
|
|
fs.lookup_still_valid = TRUE;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
if (wired)
|
2009-11-27 22:08:29 +00:00
|
|
|
fault_type = prot | (fault_type & VM_PROT_COPY);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.first_m = NULL;
|
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
|
|
|
* Search for the page at object/offset.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.object = fs.first_object;
|
|
|
|
fs.pindex = fs.first_pindex;
|
1994-05-24 10:09:53 +00:00
|
|
|
while (TRUE) {
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* If the object is dead, we stop here
|
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.object->flags & OBJ_DEAD) {
|
|
|
|
unlock_and_deallocate(&fs);
|
1998-01-17 09:17:02 +00:00
|
|
|
return (KERN_PROTECTION_FAILURE);
|
|
|
|
}
|
1999-01-21 08:29:12 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* See if page is resident
|
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.m = vm_page_lookup(fs.object, fs.pindex);
|
|
|
|
if (fs.m != NULL) {
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
1999-01-21 08:29:12 +00:00
|
|
|
* Wait/Retry if the page is busy. We have to do this
|
2013-08-09 11:11:11 +00:00
|
|
|
* if the page is either exclusive or shared busy
|
|
|
|
* because the vm_pager may be using read busy for
|
|
|
|
* pageouts (and even pageins if it is the vnode
|
|
|
|
* pager), and we could end up trying to pagein and
|
|
|
|
* pageout the same page simultaneously.
|
1999-01-21 08:29:12 +00:00
|
|
|
*
|
|
|
|
* We can theoretically allow the busy case on a read
|
|
|
|
* fault if the page is marked valid, but since such
|
|
|
|
* pages are typically already pmap'd, putting that
|
|
|
|
* special case in might be more effort then it is
|
|
|
|
* worth. We cannot under any circumstances mess
|
2013-08-09 11:11:11 +00:00
|
|
|
* around with a shared busied page except, perhaps,
|
1999-01-21 08:29:12 +00:00
|
|
|
* to pmap it.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2013-08-09 11:11:11 +00:00
|
|
|
if (vm_page_busied(fs.m)) {
|
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
|
|
|
/*
|
|
|
|
* Reference the page before unlocking and
|
|
|
|
* sleeping so that the page daemon is less
|
|
|
|
* likely to reclaim it.
|
|
|
|
*/
|
2011-09-06 10:30:11 +00:00
|
|
|
vm_page_aflag_set(fs.m, PGA_REFERENCED);
|
2004-12-24 19:31:54 +00:00
|
|
|
if (fs.object != fs.first_object) {
|
2013-03-09 02:32:23 +00:00
|
|
|
if (!VM_OBJECT_TRYWLOCK(
|
2010-05-20 08:51:01 +00:00
|
|
|
fs.first_object)) {
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs.object);
|
|
|
|
VM_OBJECT_WLOCK(fs.first_object);
|
|
|
|
VM_OBJECT_WLOCK(fs.object);
|
2010-05-20 08:51:01 +00:00
|
|
|
}
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(fs.first_m);
|
2004-12-24 19:31:54 +00:00
|
|
|
vm_page_free(fs.first_m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.first_m);
|
2004-12-24 19:31:54 +00:00
|
|
|
vm_object_pip_wakeup(fs.first_object);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs.first_object);
|
2004-12-24 19:31:54 +00:00
|
|
|
fs.first_m = NULL;
|
|
|
|
}
|
|
|
|
unlock_map(&fs);
|
|
|
|
if (fs.m == vm_page_lookup(fs.object,
|
|
|
|
fs.pindex)) {
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_sleep_if_busy(fs.m, "vmpfw");
|
2004-12-24 19:31:54 +00:00
|
|
|
}
|
|
|
|
vm_object_pip_wakeup(fs.object);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs.object);
|
2007-06-04 21:38:48 +00:00
|
|
|
PCPU_INC(cnt.v_intrans);
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_object_deallocate(fs.first_object);
|
1994-05-24 10:09:53 +00:00
|
|
|
goto RetryFault;
|
|
|
|
}
|
2013-09-16 06:25:54 +00:00
|
|
|
vm_page_lock(fs.m);
|
2012-11-13 02:50:39 +00:00
|
|
|
vm_page_remque(fs.m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.m);
|
1999-01-23 06:00:27 +00:00
|
|
|
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* Mark page busy for other processes, and the
|
|
|
|
* pagedaemon. If it still isn't completely valid
|
|
|
|
* (readable), jump to readrest, else break-out ( we
|
|
|
|
* found the page ).
|
|
|
|
*/
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_xbusy(fs.m);
|
2011-01-15 19:21:28 +00:00
|
|
|
if (fs.m->valid != VM_PAGE_BITS_ALL)
|
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
|
|
|
goto readrest;
|
1994-05-24 10:09:53 +00:00
|
|
|
break;
|
|
|
|
}
|
1999-01-21 08:29:12 +00:00
|
|
|
|
|
|
|
/*
|
1999-09-21 00:36:16 +00:00
|
|
|
* Page is not resident, If this is the search termination
|
|
|
|
* or the pager might contain the page, allocate a new page.
|
1999-01-21 08:29:12 +00:00
|
|
|
*/
|
1999-09-21 00:36:16 +00:00
|
|
|
if (TRYPAGER || fs.object == fs.first_object) {
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.pindex >= fs.object->size) {
|
|
|
|
unlock_and_deallocate(&fs);
|
1995-05-18 02:59:26 +00:00
|
|
|
return (KERN_PROTECTION_FAILURE);
|
|
|
|
}
|
1995-09-24 19:47:58 +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
|
|
|
* Allocate a new page for this object/offset pair.
|
2010-04-06 10:43:01 +00:00
|
|
|
*
|
|
|
|
* Unlocked read of the p_flag is harmless. At
|
|
|
|
* worst, the P_KILLED might be not observed
|
|
|
|
* there, and allocation can fail, causing
|
|
|
|
* restart and new reading of the p_flag.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1999-09-21 00:36:16 +00:00
|
|
|
fs.m = NULL;
|
2010-04-06 10:43:01 +00:00
|
|
|
if (!vm_page_count_severe() || P_KILLED(curproc)) {
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
|
|
|
if ((fs.object->flags & OBJ_COLORED) == 0) {
|
|
|
|
fs.object->flags |= OBJ_COLORED;
|
|
|
|
fs.object->pg_color = atop(vaddr) -
|
|
|
|
fs.pindex;
|
|
|
|
}
|
|
|
|
#endif
|
2010-04-06 10:43:01 +00:00
|
|
|
alloc_req = P_KILLED(curproc) ?
|
|
|
|
VM_ALLOC_SYSTEM : VM_ALLOC_NORMAL;
|
|
|
|
if (fs.object->type != OBJT_VNODE &&
|
|
|
|
fs.object->backing_object == NULL)
|
|
|
|
alloc_req |= VM_ALLOC_ZERO;
|
1999-09-21 00:36:16 +00:00
|
|
|
fs.m = vm_page_alloc(fs.object, fs.pindex,
|
2010-04-06 10:43:01 +00:00
|
|
|
alloc_req);
|
1999-09-21 00:36:16 +00:00
|
|
|
}
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.m == NULL) {
|
|
|
|
unlock_and_deallocate(&fs);
|
2002-02-19 18:34:02 +00:00
|
|
|
VM_WAITPFAULT;
|
1994-05-24 10:09:53 +00:00
|
|
|
goto RetryFault;
|
2009-06-07 19:38:26 +00:00
|
|
|
} else if (fs.m->valid == VM_PAGE_BITS_ALL)
|
2007-10-08 20:09:53 +00:00
|
|
|
break;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1998-01-17 09:17:02 +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
|
|
|
readrest:
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
1999-09-21 00:36:16 +00:00
|
|
|
* We have found a valid page or we have allocated a new page.
|
|
|
|
* The page thus may not be valid or may not be entirely
|
|
|
|
* valid.
|
|
|
|
*
|
|
|
|
* Attempt to fault-in the page if there is a chance that the
|
|
|
|
* pager has it, and potentially fault in additional pages
|
|
|
|
* at the same time.
|
1999-01-21 08:29:12 +00:00
|
|
|
*/
|
1999-09-21 00:36:16 +00:00
|
|
|
if (TRYPAGER) {
|
1994-05-24 10:09:53 +00:00
|
|
|
int rv;
|
1999-08-01 06:05:09 +00:00
|
|
|
u_char behavior = vm_map_entry_behavior(fs.entry);
|
1996-05-19 07:36:50 +00:00
|
|
|
|
2010-04-06 10:43:01 +00:00
|
|
|
if (behavior == MAP_ENTRY_BEHAV_RANDOM ||
|
|
|
|
P_KILLED(curproc)) {
|
2012-05-10 15:16:42 +00:00
|
|
|
behind = 0;
|
1996-05-19 07:36:50 +00:00
|
|
|
ahead = 0;
|
2012-05-10 15:16:42 +00:00
|
|
|
} else if (behavior == MAP_ENTRY_BEHAV_SEQUENTIAL) {
|
1996-05-19 07:36:50 +00:00
|
|
|
behind = 0;
|
2012-05-10 15:16:42 +00:00
|
|
|
ahead = atop(fs.entry->end - vaddr) - 1;
|
|
|
|
if (ahead > VM_FAULT_READ_AHEAD_MAX)
|
|
|
|
ahead = VM_FAULT_READ_AHEAD_MAX;
|
|
|
|
if (fs.pindex == fs.entry->next_read)
|
|
|
|
vm_fault_cache_behind(&fs,
|
|
|
|
VM_FAULT_READ_MAX);
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
} else {
|
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
|
|
|
/*
|
2012-05-10 15:16:42 +00:00
|
|
|
* If this is a sequential page fault, then
|
|
|
|
* arithmetically increase the number of pages
|
|
|
|
* in the read-ahead window. Otherwise, reset
|
|
|
|
* the read-ahead window to its smallest 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
|
|
|
*/
|
2012-05-10 15:16:42 +00:00
|
|
|
behind = atop(vaddr - fs.entry->start);
|
|
|
|
if (behind > VM_FAULT_READ_BEHIND)
|
|
|
|
behind = VM_FAULT_READ_BEHIND;
|
|
|
|
ahead = atop(fs.entry->end - vaddr) - 1;
|
|
|
|
era = fs.entry->read_ahead;
|
|
|
|
if (fs.pindex == fs.entry->next_read) {
|
|
|
|
nera = era + behind;
|
|
|
|
if (nera > VM_FAULT_READ_AHEAD_MAX)
|
|
|
|
nera = VM_FAULT_READ_AHEAD_MAX;
|
|
|
|
behind = 0;
|
|
|
|
if (ahead > nera)
|
|
|
|
ahead = nera;
|
|
|
|
if (era == VM_FAULT_READ_AHEAD_MAX)
|
|
|
|
vm_fault_cache_behind(&fs,
|
|
|
|
VM_FAULT_CACHE_BEHIND);
|
|
|
|
} else if (ahead > VM_FAULT_READ_AHEAD_MIN)
|
|
|
|
ahead = VM_FAULT_READ_AHEAD_MIN;
|
|
|
|
if (era != ahead)
|
|
|
|
fs.entry->read_ahead = ahead;
|
1996-05-19 07:36:50 +00:00
|
|
|
}
|
2009-02-08 20:23:46 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Call the pager to retrieve the data, if any, after
|
|
|
|
* releasing the lock on the map. We hold a ref on
|
2013-08-09 11:11:11 +00:00
|
|
|
* fs.object and the pages are exclusive busied.
|
2009-02-08 20:23:46 +00:00
|
|
|
*/
|
|
|
|
unlock_map(&fs);
|
|
|
|
|
|
|
|
if (fs.object->type == OBJT_VNODE) {
|
|
|
|
vp = fs.object->handle;
|
|
|
|
if (vp == fs.vp)
|
|
|
|
goto vnode_locked;
|
|
|
|
else if (fs.vp != NULL) {
|
|
|
|
vput(fs.vp);
|
|
|
|
fs.vp = NULL;
|
|
|
|
}
|
|
|
|
locked = VOP_ISLOCKED(vp);
|
|
|
|
|
|
|
|
if (locked != LK_EXCLUSIVE)
|
|
|
|
locked = LK_SHARED;
|
|
|
|
/* Do not sleep for vnode lock while fs.m is busy */
|
|
|
|
error = vget(vp, locked | LK_CANRECURSE |
|
|
|
|
LK_NOWAIT, curthread);
|
|
|
|
if (error != 0) {
|
|
|
|
vhold(vp);
|
|
|
|
release_page(&fs);
|
|
|
|
unlock_and_deallocate(&fs);
|
|
|
|
error = vget(vp, locked | LK_RETRY |
|
|
|
|
LK_CANRECURSE, curthread);
|
|
|
|
vdrop(vp);
|
|
|
|
fs.vp = vp;
|
|
|
|
KASSERT(error == 0,
|
|
|
|
("vm_fault: vget failed"));
|
|
|
|
goto RetryFault;
|
|
|
|
}
|
|
|
|
fs.vp = vp;
|
|
|
|
}
|
|
|
|
vnode_locked:
|
|
|
|
KASSERT(fs.vp == NULL || !fs.map->system_map,
|
|
|
|
("vm_fault: vnode-backed object mapped by system map"));
|
|
|
|
|
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 we find out if any other pages should be paged
|
|
|
|
* in at this time this routine checks to see if the
|
|
|
|
* pages surrounding this fault reside in the same
|
|
|
|
* object as the page for this fault. If they do,
|
|
|
|
* then they are faulted in also into the object. The
|
|
|
|
* array "marray" returned contains an array of
|
|
|
|
* vm_page_t structs where one of them is the
|
|
|
|
* vm_page_t passed to the routine. The reqpage
|
|
|
|
* return value is the index into the marray for the
|
|
|
|
* vm_page_t passed to the routine.
|
1999-01-21 08:29:12 +00:00
|
|
|
*
|
2013-08-09 11:11:11 +00:00
|
|
|
* fs.m plus the additional pages are exclusive busied.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1994-10-09 01:52:19 +00:00
|
|
|
faultcount = vm_fault_additional_pages(
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.m, behind, ahead, marray, &reqpage);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
rv = faultcount ?
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_pager_get_pages(fs.object, marray, faultcount,
|
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
1995-07-13 08:48:48 +00:00
|
|
|
reqpage) : VM_PAGER_FAIL;
|
1995-09-24 19:47:58 +00:00
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
if (rv == VM_PAGER_OK) {
|
1996-07-28 01:14:01 +00:00
|
|
|
/*
|
|
|
|
* Found the page. Leave it busy while we play
|
|
|
|
* with it.
|
|
|
|
*/
|
|
|
|
|
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
|
|
|
* Relookup in case pager changed page. Pager
|
|
|
|
* is responsible for disposition of old page
|
|
|
|
* if moved.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.m = vm_page_lookup(fs.object, fs.pindex);
|
2002-03-10 21:52:48 +00:00
|
|
|
if (!fs.m) {
|
1998-03-07 20:45:47 +00:00
|
|
|
unlock_and_deallocate(&fs);
|
1996-07-28 01:14:01 +00:00
|
|
|
goto RetryFault;
|
1995-04-09 06:03:56 +00:00
|
|
|
}
|
1995-05-30 08:16:23 +00:00
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
hardfault++;
|
1999-01-21 08:29:12 +00:00
|
|
|
break; /* break to PAGE HAS BEEN FOUND */
|
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
|
|
|
* Remove the bogus page (which does not exist at this
|
|
|
|
* object/offset); before doing so, we must get back
|
|
|
|
* our object lock to preserve our invariant.
|
1995-05-30 08:16:23 +00:00
|
|
|
*
|
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
1995-07-13 08:48:48 +00:00
|
|
|
* Also wake up any other process that may want to bring
|
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
|
|
|
* in this page.
|
1995-05-30 08:16:23 +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
|
|
|
* If this is the top-level object, we must leave the
|
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
1995-07-13 08:48:48 +00:00
|
|
|
* busy page to prevent another process from rushing
|
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
|
|
|
* past us, and inserting the page in that object at
|
|
|
|
* the same time that we are.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1994-11-06 09:55:31 +00:00
|
|
|
if (rv == VM_PAGER_ERROR)
|
1998-07-22 09:38:04 +00:00
|
|
|
printf("vm_fault: pager read error, pid %d (%s)\n",
|
|
|
|
curproc->p_pid, curproc->p_comm);
|
1994-05-25 09:21:21 +00:00
|
|
|
/*
|
1994-11-06 09:55:31 +00:00
|
|
|
* Data outside the range of the pager or an I/O error
|
1994-05-25 09:21:21 +00:00
|
|
|
*/
|
1994-11-06 09:55:31 +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
|
|
|
* XXX - the check for kernel_map is a kludge to work
|
|
|
|
* around having the machine panic on a kernel space
|
|
|
|
* fault w/ I/O error.
|
1994-11-06 09:55:31 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
if (((fs.map != kernel_map) && (rv == VM_PAGER_ERROR)) ||
|
1998-01-17 09:17:02 +00:00
|
|
|
(rv == VM_PAGER_BAD)) {
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(fs.m);
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_page_free(fs.m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.m);
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.m = NULL;
|
|
|
|
unlock_and_deallocate(&fs);
|
1994-11-06 09:55:31 +00:00
|
|
|
return ((rv == VM_PAGER_ERROR) ? KERN_FAILURE : KERN_PROTECTION_FAILURE);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.object != fs.first_object) {
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(fs.m);
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_page_free(fs.m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.m);
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.m = NULL;
|
1994-05-25 09:21:21 +00:00
|
|
|
/*
|
|
|
|
* XXX - we cannot just fall out at this
|
|
|
|
* point, m has been freed and is invalid!
|
|
|
|
*/
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
}
|
1999-09-21 00:36:16 +00:00
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
1999-01-21 08:29:12 +00:00
|
|
|
* We get here if the object has default pager (or unwiring)
|
|
|
|
* or the pager doesn't have the page.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.object == fs.first_object)
|
|
|
|
fs.first_m = fs.m;
|
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
|
|
|
* Move on to the next object. Lock the next object before
|
|
|
|
* unlocking the current one.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.pindex += OFF_TO_IDX(fs.object->backing_object_offset);
|
|
|
|
next_object = fs.object->backing_object;
|
1994-05-24 10:09:53 +00:00
|
|
|
if (next_object == NULL) {
|
|
|
|
/*
|
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
|
|
|
* If there's no object left, fill the page in the top
|
|
|
|
* object with zeros.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.object != fs.first_object) {
|
|
|
|
vm_object_pip_wakeup(fs.object);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs.object);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.object = fs.first_object;
|
|
|
|
fs.pindex = fs.first_pindex;
|
|
|
|
fs.m = fs.first_m;
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(fs.object);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.first_m = NULL;
|
1994-05-24 10:09:53 +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
|
|
|
/*
|
|
|
|
* Zero the page if necessary and mark it valid.
|
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
if ((fs.m->flags & PG_ZERO) == 0) {
|
2002-08-25 00:22:31 +00:00
|
|
|
pmap_zero_page(fs.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
|
|
|
} else {
|
2007-06-04 21:38:48 +00:00
|
|
|
PCPU_INC(cnt.v_ozfod);
|
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
|
|
|
}
|
2007-06-04 21:38:48 +00:00
|
|
|
PCPU_INC(cnt.v_zfod);
|
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
|
|
|
fs.m->valid = VM_PAGE_BITS_ALL;
|
2014-02-09 01:59:52 +00:00
|
|
|
/* Don't try to prefault neighboring pages. */
|
|
|
|
faultcount = 1;
|
1999-01-21 08:29:12 +00:00
|
|
|
break; /* break to PAGE HAS BEEN FOUND */
|
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
|
|
|
} else {
|
2003-06-22 05:36:53 +00:00
|
|
|
KASSERT(fs.object != next_object,
|
|
|
|
("object loop %p", next_object));
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(next_object);
|
2003-06-22 05:36:53 +00:00
|
|
|
vm_object_pip_add(next_object, 1);
|
|
|
|
if (fs.object != fs.first_object)
|
|
|
|
vm_object_pip_wakeup(fs.object);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs.object);
|
2003-06-22 05:36:53 +00:00
|
|
|
fs.object = next_object;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
}
|
1999-01-21 08:29:12 +00:00
|
|
|
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_assert_xbusied(fs.m);
|
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
|
|
|
* PAGE HAS BEEN FOUND. [Loop invariant still holds -- the object lock
|
|
|
|
* is held.]
|
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
|
|
|
* If the page is being written, but isn't already owned by the
|
|
|
|
* top-level object, we have to copy it into a new page owned by the
|
|
|
|
* top-level object.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
if (fs.object != fs.first_object) {
|
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
|
|
|
/*
|
|
|
|
* We only really need to copy if we want to write it.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2009-11-26 05:16:07 +00:00
|
|
|
if ((fault_type & (VM_PROT_COPY | VM_PROT_WRITE)) != 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
|
|
|
/*
|
1999-01-21 08:29:12 +00:00
|
|
|
* This allows pages to be virtually copied from a
|
|
|
|
* backing_object into the first_object, where the
|
|
|
|
* backing object has no other refs to it, and cannot
|
|
|
|
* gain any more refs. Instead of a bcopy, we just
|
|
|
|
* move the page from the backing object to the
|
|
|
|
* first object. Note that we must mark the page
|
|
|
|
* dirty in the first object so that it will go out
|
|
|
|
* to swap when needed.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2003-06-21 06:31:42 +00:00
|
|
|
is_first_object_locked = FALSE;
|
2003-06-20 04:20:36 +00:00
|
|
|
if (
|
1996-03-02 02:54:24 +00:00
|
|
|
/*
|
|
|
|
* Only one shadow object
|
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
(fs.object->shadow_count == 1) &&
|
1996-03-02 02:54:24 +00:00
|
|
|
/*
|
|
|
|
* No COW refs, except us
|
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
(fs.object->ref_count == 1) &&
|
1996-03-02 02:54:24 +00:00
|
|
|
/*
|
2000-03-27 20:41:17 +00:00
|
|
|
* No one else can look this object up
|
1996-03-02 02:54:24 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
(fs.object->handle == NULL) &&
|
1996-03-02 02:54:24 +00:00
|
|
|
/*
|
|
|
|
* No other ways to look the object up
|
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
((fs.object->type == OBJT_DEFAULT) ||
|
|
|
|
(fs.object->type == OBJT_SWAP)) &&
|
2013-03-09 02:32:23 +00:00
|
|
|
(is_first_object_locked = VM_OBJECT_TRYWLOCK(fs.first_object)) &&
|
1996-03-02 02:54:24 +00:00
|
|
|
/*
|
|
|
|
* We don't chase down the shadow chain
|
|
|
|
*/
|
2003-06-20 04:20:36 +00:00
|
|
|
fs.object == fs.first_object->backing_object) {
|
1996-03-02 02:54:24 +00:00
|
|
|
/*
|
|
|
|
* get rid of the unnecessary page
|
|
|
|
*/
|
2010-05-06 18:58:32 +00:00
|
|
|
vm_page_lock(fs.first_m);
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_page_free(fs.first_m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.first_m);
|
1996-03-02 02:54:24 +00:00
|
|
|
/*
|
1999-01-21 08:29:12 +00:00
|
|
|
* grab the page and put it into the
|
|
|
|
* process'es object. The page is
|
|
|
|
* automatically made dirty.
|
1996-03-02 02:54:24 +00:00
|
|
|
*/
|
2013-08-09 11:28:55 +00:00
|
|
|
if (vm_page_rename(fs.m, fs.first_object,
|
|
|
|
fs.first_pindex)) {
|
|
|
|
unlock_and_deallocate(&fs);
|
|
|
|
goto RetryFault;
|
|
|
|
}
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_xbusy(fs.m);
|
2003-06-22 00:00:11 +00:00
|
|
|
fs.first_m = fs.m;
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.m = NULL;
|
2007-06-04 21:38:48 +00:00
|
|
|
PCPU_INC(cnt.v_cow_optim);
|
1996-03-02 02:54:24 +00:00
|
|
|
} else {
|
|
|
|
/*
|
|
|
|
* Oh, well, lets copy it.
|
|
|
|
*/
|
2003-10-08 05:35:12 +00:00
|
|
|
pmap_copy_page(fs.m, fs.first_m);
|
|
|
|
fs.first_m->valid = VM_PAGE_BITS_ALL;
|
2009-11-27 22:08:29 +00:00
|
|
|
if (wired && (fault_flags &
|
|
|
|
VM_FAULT_CHANGE_WIRING) == 0) {
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(fs.first_m);
|
2009-11-27 22:08:29 +00:00
|
|
|
vm_page_wire(fs.first_m);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.first_m);
|
|
|
|
|
|
|
|
vm_page_lock(fs.m);
|
2009-11-27 22:08:29 +00:00
|
|
|
vm_page_unwire(fs.m, FALSE);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.m);
|
2009-11-27 22:08:29 +00:00
|
|
|
}
|
1998-03-07 20:45:47 +00:00
|
|
|
/*
|
|
|
|
* We no longer need the old page or object.
|
|
|
|
*/
|
|
|
|
release_page(&fs);
|
1996-03-02 02:54:24 +00:00
|
|
|
}
|
1999-01-21 08:29:12 +00:00
|
|
|
/*
|
|
|
|
* fs.object != fs.first_object due to above
|
|
|
|
* conditional
|
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_object_pip_wakeup(fs.object);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs.object);
|
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
|
|
|
* Only use the new page below...
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
1998-03-07 20:45:47 +00:00
|
|
|
fs.object = fs.first_object;
|
|
|
|
fs.pindex = fs.first_pindex;
|
2003-06-22 00:00:11 +00:00
|
|
|
fs.m = fs.first_m;
|
2003-06-22 21:35:41 +00:00
|
|
|
if (!is_first_object_locked)
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(fs.object);
|
2007-06-04 21:38:48 +00:00
|
|
|
PCPU_INC(cnt.v_cow_faults);
|
2012-05-23 18:10:54 +00:00
|
|
|
curthread->td_cow++;
|
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
|
|
|
} else {
|
|
|
|
prot &= ~VM_PROT_WRITE;
|
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
|
|
|
* We must verify that the maps have not changed since our last
|
|
|
|
* lookup.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2004-08-12 20:14:49 +00:00
|
|
|
if (!fs.lookup_still_valid) {
|
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_object_t retry_object;
|
1995-12-11 04:58:34 +00:00
|
|
|
vm_pindex_t retry_pindex;
|
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_prot_t retry_prot;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2004-08-12 20:14:49 +00:00
|
|
|
if (!vm_map_trylock_read(fs.map)) {
|
2001-03-14 06:48:53 +00:00
|
|
|
release_page(&fs);
|
|
|
|
unlock_and_deallocate(&fs);
|
|
|
|
goto RetryFault;
|
|
|
|
}
|
2004-08-12 20:14:49 +00:00
|
|
|
fs.lookup_still_valid = TRUE;
|
|
|
|
if (fs.map->timestamp != map_generation) {
|
|
|
|
result = vm_map_lookup_locked(&fs.map, vaddr, fault_type,
|
|
|
|
&fs.entry, &retry_object, &retry_pindex, &retry_prot, &wired);
|
1999-02-17 09:08:29 +00:00
|
|
|
|
2004-08-12 20:14:49 +00:00
|
|
|
/*
|
2006-02-02 21:55:38 +00:00
|
|
|
* If we don't need the page any longer, put it on the inactive
|
2004-08-12 20:14:49 +00:00
|
|
|
* list (the easiest thing to do here). If no one needs it,
|
|
|
|
* pageout will grab it eventually.
|
|
|
|
*/
|
|
|
|
if (result != KERN_SUCCESS) {
|
|
|
|
release_page(&fs);
|
|
|
|
unlock_and_deallocate(&fs);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2004-08-12 20:14:49 +00:00
|
|
|
/*
|
|
|
|
* If retry of map lookup would have blocked then
|
|
|
|
* retry fault from start.
|
|
|
|
*/
|
|
|
|
if (result == KERN_FAILURE)
|
|
|
|
goto RetryFault;
|
|
|
|
return (result);
|
|
|
|
}
|
|
|
|
if ((retry_object != fs.first_object) ||
|
|
|
|
(retry_pindex != fs.first_pindex)) {
|
|
|
|
release_page(&fs);
|
|
|
|
unlock_and_deallocate(&fs);
|
|
|
|
goto RetryFault;
|
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2004-08-12 20:14:49 +00:00
|
|
|
/*
|
|
|
|
* Check whether the protection has changed or the object has
|
|
|
|
* been copied while we left the map unlocked. Changing from
|
|
|
|
* read to write permission is OK - we leave the page
|
|
|
|
* write-protected, and catch the write fault. Changing from
|
|
|
|
* write to read permission means that we can't mark the page
|
|
|
|
* write-enabled after all.
|
|
|
|
*/
|
|
|
|
prot &= retry_prot;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
}
|
2009-02-08 20:23:46 +00:00
|
|
|
/*
|
2009-02-25 07:52:53 +00:00
|
|
|
* If the page was filled by a pager, update the map entry's
|
|
|
|
* last read offset. Since the pager does not return the
|
|
|
|
* actual set of pages that it read, this update is based on
|
|
|
|
* the requested set. Typically, the requested and actual
|
|
|
|
* sets are the same.
|
2009-02-08 20:23:46 +00:00
|
|
|
*
|
|
|
|
* XXX The following assignment modifies the map
|
|
|
|
* without holding a write lock on it.
|
|
|
|
*/
|
2009-02-25 07:52:53 +00:00
|
|
|
if (hardfault)
|
2012-05-10 15:16:42 +00:00
|
|
|
fs.entry->next_read = fs.pindex + faultcount - reqpage;
|
2009-02-08 20:23:46 +00:00
|
|
|
|
2010-12-20 22:49:31 +00:00
|
|
|
if ((prot & VM_PROT_WRITE) != 0 ||
|
|
|
|
(fault_flags & VM_FAULT_DIRTY) != 0) {
|
2006-08-13 00:11:09 +00:00
|
|
|
vm_object_set_writeable_dirty(fs.object);
|
1999-12-12 03:19:33 +00:00
|
|
|
|
1995-03-27 02:41:00 +00:00
|
|
|
/*
|
2006-08-13 00:11:09 +00:00
|
|
|
* If this is a NOSYNC mmap we do not want to set VPO_NOSYNC
|
1999-12-12 03:19:33 +00:00
|
|
|
* if the page is already dirty to prevent data written with
|
|
|
|
* the expectation of being synced from not being synced.
|
|
|
|
* Likewise if this entry does not request NOSYNC then make
|
|
|
|
* sure the page isn't marked NOSYNC. Applications sharing
|
|
|
|
* data should use the same flags to avoid ping ponging.
|
1995-03-27 02:41:00 +00:00
|
|
|
*/
|
2001-02-28 04:26:43 +00:00
|
|
|
if (fs.entry->eflags & MAP_ENTRY_NOSYNC) {
|
|
|
|
if (fs.m->dirty == 0)
|
2006-08-13 00:11:09 +00:00
|
|
|
fs.m->oflags |= VPO_NOSYNC;
|
2001-02-28 04:26:43 +00:00
|
|
|
} else {
|
2006-08-13 00:11:09 +00:00
|
|
|
fs.m->oflags &= ~VPO_NOSYNC;
|
2001-02-28 04:26:43 +00:00
|
|
|
}
|
2009-11-27 20:24:11 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* If the fault is a write, we know that this page is being
|
|
|
|
* written NOW so dirty it explicitly to save on
|
|
|
|
* pmap_is_modified() calls later.
|
|
|
|
*
|
|
|
|
* Also tell the backing pager, if any, that it should remove
|
|
|
|
* any swap backing since the page is now dirty.
|
|
|
|
*/
|
2010-12-20 22:49:31 +00:00
|
|
|
if (((fault_type & VM_PROT_WRITE) != 0 &&
|
|
|
|
(fault_flags & VM_FAULT_CHANGE_WIRING) == 0) ||
|
|
|
|
(fault_flags & VM_FAULT_DIRTY) != 0) {
|
1999-01-24 06:04:52 +00:00
|
|
|
vm_page_dirty(fs.m);
|
1999-01-21 08:29:12 +00:00
|
|
|
vm_pager_page_unswapped(fs.m);
|
1995-03-27 02:41:00 +00:00
|
|
|
}
|
|
|
|
}
|
1995-04-09 06:03:56 +00:00
|
|
|
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_assert_xbusied(fs.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
|
|
|
/*
|
2009-04-26 20:54:57 +00:00
|
|
|
* Page must be completely valid or it is not fit to
|
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
|
|
|
* map into user space. vm_pager_get_pages() ensures this.
|
|
|
|
*/
|
2009-04-26 20:54:57 +00:00
|
|
|
KASSERT(fs.m->valid == VM_PAGE_BITS_ALL,
|
|
|
|
("vm_fault: page %p partially invalid", fs.m));
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(fs.object);
|
2003-10-04 21:35:48 +00:00
|
|
|
|
2004-08-09 18:46:39 +00:00
|
|
|
/*
|
|
|
|
* Put this page into the physical map. We had to do the unlock above
|
|
|
|
* because pmap_enter() may sleep. We don't put the page
|
|
|
|
* back on the active queue until later so that the pageout daemon
|
|
|
|
* won't find it (yet).
|
|
|
|
*/
|
2008-01-03 07:34:34 +00:00
|
|
|
pmap_enter(fs.map->pmap, vaddr, fault_type, fs.m, prot, wired);
|
2014-02-09 01:59:52 +00:00
|
|
|
if (faultcount != 1 && (fault_flags & VM_FAULT_CHANGE_WIRING) == 0 &&
|
|
|
|
wired == 0)
|
2014-02-02 20:21:53 +00:00
|
|
|
vm_fault_prefault(&fs, vaddr, faultcount, reqpage);
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(fs.object);
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(fs.m);
|
1996-06-01 20:50:57 +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
|
|
|
* If the page is not wired down, then put it where the pageout daemon
|
|
|
|
* can find it.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2009-11-18 18:05:54 +00:00
|
|
|
if (fault_flags & VM_FAULT_CHANGE_WIRING) {
|
1994-05-24 10:09:53 +00:00
|
|
|
if (wired)
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_page_wire(fs.m);
|
1994-05-24 10:09:53 +00:00
|
|
|
else
|
1998-10-28 13:37:02 +00:00
|
|
|
vm_page_unwire(fs.m, 1);
|
2010-05-07 15:49:43 +00:00
|
|
|
} else
|
1998-03-07 20:45:47 +00:00
|
|
|
vm_page_activate(fs.m);
|
2010-12-20 22:49:31 +00:00
|
|
|
if (m_hold != NULL) {
|
|
|
|
*m_hold = fs.m;
|
2013-08-05 08:55:35 +00:00
|
|
|
vm_page_hold(fs.m);
|
2010-12-20 22:49:31 +00:00
|
|
|
}
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_unlock(fs.m);
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_xunbusy(fs.m);
|
2003-04-22 20:01:56 +00:00
|
|
|
|
2004-08-09 06:01:46 +00:00
|
|
|
/*
|
|
|
|
* Unlock everything, and return
|
|
|
|
*/
|
|
|
|
unlock_and_deallocate(&fs);
|
2013-01-28 12:54:53 +00:00
|
|
|
if (hardfault) {
|
|
|
|
PCPU_INC(cnt.v_io_faults);
|
2007-06-01 01:12:45 +00:00
|
|
|
curthread->td_ru.ru_majflt++;
|
2013-01-28 12:54:53 +00:00
|
|
|
} else
|
2007-06-01 01:12:45 +00:00
|
|
|
curthread->td_ru.ru_minflt++;
|
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
|
|
|
return (KERN_SUCCESS);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
2012-05-10 15:16:42 +00:00
|
|
|
/*
|
|
|
|
* Speed up the reclamation of up to "distance" pages that precede the
|
|
|
|
* faulting pindex within the first object of the shadow chain.
|
|
|
|
*/
|
|
|
|
static void
|
|
|
|
vm_fault_cache_behind(const struct faultstate *fs, int distance)
|
|
|
|
{
|
|
|
|
vm_object_t first_object, object;
|
|
|
|
vm_page_t m, m_prev;
|
|
|
|
vm_pindex_t pindex;
|
|
|
|
|
|
|
|
object = fs->object;
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_ASSERT_WLOCKED(object);
|
2012-05-10 15:16:42 +00:00
|
|
|
first_object = fs->first_object;
|
|
|
|
if (first_object != object) {
|
2013-03-09 02:32:23 +00:00
|
|
|
if (!VM_OBJECT_TRYWLOCK(first_object)) {
|
|
|
|
VM_OBJECT_WUNLOCK(object);
|
|
|
|
VM_OBJECT_WLOCK(first_object);
|
|
|
|
VM_OBJECT_WLOCK(object);
|
2012-05-10 15:16:42 +00:00
|
|
|
}
|
|
|
|
}
|
In the past four years, we've added two new vm object types. Each time,
similar changes had to be made in various places throughout the machine-
independent virtual memory layer to support the new vm object type.
However, in most of these places, it's actually not the type of the vm
object that matters to us but instead certain attributes of its pages.
For example, OBJT_DEVICE, OBJT_MGTDEVICE, and OBJT_SG objects contain
fictitious pages. In other words, in most of these places, we were
testing the vm object's type to determine if it contained fictitious (or
unmanaged) pages.
To both simplify the code in these places and make the addition of future
vm object types easier, this change introduces two new vm object flags
that describe attributes of the vm object's pages, specifically, whether
they are fictitious or unmanaged.
Reviewed and tested by: kib
2012-12-09 00:32:38 +00:00
|
|
|
/* Neither fictitious nor unmanaged pages can be cached. */
|
|
|
|
if ((first_object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0) {
|
2012-05-10 15:16:42 +00:00
|
|
|
if (fs->first_pindex < distance)
|
|
|
|
pindex = 0;
|
|
|
|
else
|
|
|
|
pindex = fs->first_pindex - distance;
|
|
|
|
if (pindex < OFF_TO_IDX(fs->entry->offset))
|
|
|
|
pindex = OFF_TO_IDX(fs->entry->offset);
|
|
|
|
m = first_object != object ? fs->first_m : fs->m;
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_assert_xbusied(m);
|
2012-05-10 15:16:42 +00:00
|
|
|
m_prev = vm_page_prev(m);
|
|
|
|
while ((m = m_prev) != NULL && m->pindex >= pindex &&
|
|
|
|
m->valid == VM_PAGE_BITS_ALL) {
|
|
|
|
m_prev = vm_page_prev(m);
|
2013-08-09 11:11:11 +00:00
|
|
|
if (vm_page_busied(m))
|
2012-05-10 15:16:42 +00:00
|
|
|
continue;
|
|
|
|
vm_page_lock(m);
|
|
|
|
if (m->hold_count == 0 && m->wire_count == 0) {
|
|
|
|
pmap_remove_all(m);
|
|
|
|
vm_page_aflag_clear(m, PGA_REFERENCED);
|
|
|
|
if (m->dirty != 0)
|
|
|
|
vm_page_deactivate(m);
|
|
|
|
else
|
|
|
|
vm_page_cache(m);
|
|
|
|
}
|
|
|
|
vm_page_unlock(m);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (first_object != object)
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(first_object);
|
2012-05-10 15:16:42 +00:00
|
|
|
}
|
|
|
|
|
2003-10-03 22:46:53 +00:00
|
|
|
/*
|
|
|
|
* vm_fault_prefault provides a quick way of clustering
|
|
|
|
* pagefaults into a processes address space. It is a "cousin"
|
|
|
|
* of vm_map_pmap_enter, except it runs at page fault time instead
|
|
|
|
* of mmap time.
|
|
|
|
*/
|
|
|
|
static void
|
2014-02-02 20:21:53 +00:00
|
|
|
vm_fault_prefault(const struct faultstate *fs, vm_offset_t addra,
|
|
|
|
int faultcount, int reqpage)
|
2003-10-03 22:46:53 +00:00
|
|
|
{
|
2014-02-02 20:21:53 +00:00
|
|
|
pmap_t pmap;
|
|
|
|
vm_map_entry_t entry;
|
|
|
|
vm_object_t backing_object, lobject;
|
2003-10-03 22:46:53 +00:00
|
|
|
vm_offset_t addr, starta;
|
|
|
|
vm_pindex_t pindex;
|
2006-06-15 01:01:06 +00:00
|
|
|
vm_page_t m;
|
2014-02-02 20:21:53 +00:00
|
|
|
int backward, forward, i;
|
2003-10-03 22:46:53 +00:00
|
|
|
|
2014-02-02 20:21:53 +00:00
|
|
|
pmap = fs->map->pmap;
|
2004-10-17 20:29:28 +00:00
|
|
|
if (pmap != vmspace_pmap(curthread->td_proc->p_vmspace))
|
2003-10-03 22:46:53 +00:00
|
|
|
return;
|
|
|
|
|
2014-02-02 20:21:53 +00:00
|
|
|
if (faultcount > 0) {
|
|
|
|
backward = reqpage;
|
|
|
|
forward = faultcount - reqpage - 1;
|
|
|
|
} else {
|
|
|
|
backward = PFBAK;
|
|
|
|
forward = PFFOR;
|
|
|
|
}
|
|
|
|
entry = fs->entry;
|
2003-10-03 22:46:53 +00:00
|
|
|
|
2014-02-02 20:21:53 +00:00
|
|
|
starta = addra - backward * PAGE_SIZE;
|
2003-10-03 22:46:53 +00:00
|
|
|
if (starta < entry->start) {
|
|
|
|
starta = entry->start;
|
|
|
|
} else if (starta > addra) {
|
|
|
|
starta = 0;
|
|
|
|
}
|
|
|
|
|
2014-02-02 20:21:53 +00:00
|
|
|
/*
|
|
|
|
* Generate the sequence of virtual addresses that are candidates for
|
|
|
|
* prefaulting in an outward spiral from the faulting virtual address,
|
|
|
|
* "addra". Specifically, the sequence is "addra - PAGE_SIZE", "addra
|
|
|
|
* + PAGE_SIZE", "addra - 2 * PAGE_SIZE", "addra + 2 * PAGE_SIZE", ...
|
|
|
|
* If the candidate address doesn't have a backing physical page, then
|
|
|
|
* the loop immediately terminates.
|
|
|
|
*/
|
|
|
|
for (i = 0; i < 2 * imax(backward, forward); i++) {
|
|
|
|
addr = addra + ((i >> 1) + 1) * ((i & 1) == 0 ? -PAGE_SIZE :
|
|
|
|
PAGE_SIZE);
|
|
|
|
if (addr > addra + forward * PAGE_SIZE)
|
2003-10-03 22:46:53 +00:00
|
|
|
addr = 0;
|
|
|
|
|
|
|
|
if (addr < starta || addr >= entry->end)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (!pmap_is_prefaultable(pmap, addr))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
pindex = ((addr - entry->start) + entry->offset) >> PAGE_SHIFT;
|
2014-02-02 20:21:53 +00:00
|
|
|
lobject = entry->object.vm_object;
|
2013-05-17 19:02:36 +00:00
|
|
|
VM_OBJECT_RLOCK(lobject);
|
2003-10-03 22:46:53 +00:00
|
|
|
while ((m = vm_page_lookup(lobject, pindex)) == NULL &&
|
|
|
|
lobject->type == OBJT_DEFAULT &&
|
|
|
|
(backing_object = lobject->backing_object) != NULL) {
|
2009-10-25 17:30:50 +00:00
|
|
|
KASSERT((lobject->backing_object_offset & PAGE_MASK) ==
|
|
|
|
0, ("vm_fault_prefault: unaligned object offset"));
|
2003-10-03 22:46:53 +00:00
|
|
|
pindex += lobject->backing_object_offset >> PAGE_SHIFT;
|
2013-05-17 19:02:36 +00:00
|
|
|
VM_OBJECT_RLOCK(backing_object);
|
|
|
|
VM_OBJECT_RUNLOCK(lobject);
|
2003-10-03 22:46:53 +00:00
|
|
|
lobject = backing_object;
|
|
|
|
}
|
2003-10-04 21:35:48 +00:00
|
|
|
if (m == NULL) {
|
2013-05-17 19:02:36 +00:00
|
|
|
VM_OBJECT_RUNLOCK(lobject);
|
2003-10-03 22:46:53 +00:00
|
|
|
break;
|
2003-10-04 21:35:48 +00:00
|
|
|
}
|
2009-06-07 19:38:26 +00:00
|
|
|
if (m->valid == VM_PAGE_BITS_ALL &&
|
2010-05-08 20:34:01 +00:00
|
|
|
(m->flags & PG_FICTITIOUS) == 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
|
|
|
pmap_enter_quick(pmap, addr, m, entry->protection);
|
2013-05-17 19:02:36 +00:00
|
|
|
VM_OBJECT_RUNLOCK(lobject);
|
2003-10-03 22:46:53 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2010-12-25 21:26:56 +00:00
|
|
|
/*
|
|
|
|
* Hold each of the physical pages that are mapped by the specified range of
|
|
|
|
* virtual addresses, ["addr", "addr" + "len"), if those mappings are valid
|
|
|
|
* and allow the specified types of access, "prot". If all of the implied
|
|
|
|
* pages are successfully held, then the number of held pages is returned
|
|
|
|
* together with pointers to those pages in the array "ma". However, if any
|
|
|
|
* of the pages cannot be held, -1 is returned.
|
|
|
|
*/
|
|
|
|
int
|
|
|
|
vm_fault_quick_hold_pages(vm_map_t map, vm_offset_t addr, vm_size_t len,
|
|
|
|
vm_prot_t prot, vm_page_t *ma, int max_count)
|
|
|
|
{
|
|
|
|
vm_offset_t end, va;
|
|
|
|
vm_page_t *mp;
|
2013-11-20 08:45:26 +00:00
|
|
|
int count;
|
2010-12-25 21:26:56 +00:00
|
|
|
boolean_t pmap_failed;
|
|
|
|
|
2011-03-25 16:38:10 +00:00
|
|
|
if (len == 0)
|
|
|
|
return (0);
|
2013-11-20 08:45:26 +00:00
|
|
|
end = round_page(addr + len);
|
2010-12-25 21:26:56 +00:00
|
|
|
addr = trunc_page(addr);
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Check for illegal addresses.
|
|
|
|
*/
|
|
|
|
if (addr < vm_map_min(map) || addr > end || end > vm_map_max(map))
|
|
|
|
return (-1);
|
|
|
|
|
2013-11-20 08:45:26 +00:00
|
|
|
if (atop(end - addr) > max_count)
|
2010-12-25 21:26:56 +00:00
|
|
|
panic("vm_fault_quick_hold_pages: count > max_count");
|
2013-11-20 08:45:26 +00:00
|
|
|
count = atop(end - addr);
|
2010-12-25 21:26:56 +00:00
|
|
|
|
|
|
|
/*
|
|
|
|
* Most likely, the physical pages are resident in the pmap, so it is
|
|
|
|
* faster to try pmap_extract_and_hold() first.
|
|
|
|
*/
|
|
|
|
pmap_failed = FALSE;
|
|
|
|
for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE) {
|
|
|
|
*mp = pmap_extract_and_hold(map->pmap, va, prot);
|
|
|
|
if (*mp == NULL)
|
|
|
|
pmap_failed = TRUE;
|
|
|
|
else if ((prot & VM_PROT_WRITE) != 0 &&
|
2010-12-28 20:02:30 +00:00
|
|
|
(*mp)->dirty != VM_PAGE_BITS_ALL) {
|
2010-12-25 21:26:56 +00:00
|
|
|
/*
|
|
|
|
* Explicitly dirty the physical page. Otherwise, the
|
|
|
|
* caller's changes may go unnoticed because they are
|
|
|
|
* performed through an unmanaged mapping or by a DMA
|
|
|
|
* operation.
|
2011-06-19 19:13:24 +00:00
|
|
|
*
|
2011-09-28 14:57:50 +00:00
|
|
|
* The object lock is not held here.
|
|
|
|
* See vm_page_clear_dirty_mask().
|
2010-12-25 21:26:56 +00:00
|
|
|
*/
|
2011-06-19 19:13:24 +00:00
|
|
|
vm_page_dirty(*mp);
|
2010-12-25 21:26:56 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
if (pmap_failed) {
|
|
|
|
/*
|
|
|
|
* One or more pages could not be held by the pmap. Either no
|
|
|
|
* page was mapped at the specified virtual address or that
|
|
|
|
* mapping had insufficient permissions. Attempt to fault in
|
|
|
|
* and hold these pages.
|
|
|
|
*/
|
|
|
|
for (mp = ma, va = addr; va < end; mp++, va += PAGE_SIZE)
|
2013-08-05 08:55:35 +00:00
|
|
|
if (*mp == NULL && vm_fault_hold(map, va, prot,
|
2010-12-25 21:26:56 +00:00
|
|
|
VM_FAULT_NORMAL, mp) != KERN_SUCCESS)
|
|
|
|
goto error;
|
|
|
|
}
|
|
|
|
return (count);
|
|
|
|
error:
|
|
|
|
for (mp = ma; mp < ma + count; mp++)
|
|
|
|
if (*mp != NULL) {
|
|
|
|
vm_page_lock(*mp);
|
|
|
|
vm_page_unhold(*mp);
|
|
|
|
vm_page_unlock(*mp);
|
|
|
|
}
|
|
|
|
return (-1);
|
|
|
|
}
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
/*
|
|
|
|
* vm_fault_wire:
|
|
|
|
*
|
|
|
|
* Wire down a range of virtual addresses in a map.
|
|
|
|
*/
|
|
|
|
int
|
2004-05-22 04:53:51 +00:00
|
|
|
vm_fault_wire(vm_map_t map, vm_offset_t start, vm_offset_t end,
|
2009-11-18 18:05:54 +00:00
|
|
|
boolean_t fictitious)
|
1996-12-14 17:54:17 +00:00
|
|
|
{
|
2001-07-04 19:00:13 +00:00
|
|
|
vm_offset_t va;
|
1996-12-14 17:54:17 +00:00
|
|
|
int rv;
|
|
|
|
|
|
|
|
/*
|
|
|
|
* We simulate a fault to get the page and enter it in the physical
|
2002-07-24 19:47:56 +00:00
|
|
|
* map. For user wiring, we only ask for read access on currently
|
|
|
|
* read-only sections.
|
1996-12-14 17:54:17 +00:00
|
|
|
*/
|
|
|
|
for (va = start; va < end; va += PAGE_SIZE) {
|
2009-11-18 18:05:54 +00:00
|
|
|
rv = vm_fault(map, va, VM_PROT_NONE, VM_FAULT_CHANGE_WIRING);
|
1994-05-24 10:09:53 +00:00
|
|
|
if (rv) {
|
|
|
|
if (va != start)
|
2004-05-22 04:53:51 +00:00
|
|
|
vm_fault_unwire(map, start, va, fictitious);
|
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 (rv);
|
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
|
|
|
return (KERN_SUCCESS);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* vm_fault_unwire:
|
|
|
|
*
|
|
|
|
* Unwire a range of virtual addresses in a map.
|
|
|
|
*/
|
1994-05-25 09:21:21 +00:00
|
|
|
void
|
2004-05-22 04:53:51 +00:00
|
|
|
vm_fault_unwire(vm_map_t map, vm_offset_t start, vm_offset_t end,
|
|
|
|
boolean_t fictitious)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2003-03-25 00:07:06 +00:00
|
|
|
vm_paddr_t pa;
|
|
|
|
vm_offset_t va;
|
2010-05-05 03:45:46 +00:00
|
|
|
vm_page_t m;
|
2001-07-04 19:00:13 +00:00
|
|
|
pmap_t pmap;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
pmap = vm_map_pmap(map);
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
* Since the pages are wired down, we must be able to get their
|
|
|
|
* mappings from the physical map system.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
|
|
|
for (va = start; va < end; va += PAGE_SIZE) {
|
|
|
|
pa = pmap_extract(pmap, va);
|
2003-03-25 00:07:06 +00:00
|
|
|
if (pa != 0) {
|
1996-05-18 03:38:05 +00:00
|
|
|
pmap_change_wiring(pmap, va, FALSE);
|
2004-05-22 04:53:51 +00:00
|
|
|
if (!fictitious) {
|
2010-05-05 03:45:46 +00:00
|
|
|
m = PHYS_TO_VM_PAGE(pa);
|
|
|
|
vm_page_lock(m);
|
|
|
|
vm_page_unwire(m, TRUE);
|
|
|
|
vm_page_unlock(m);
|
2004-05-22 04:53:51 +00:00
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/*
|
|
|
|
* Routine:
|
|
|
|
* vm_fault_copy_entry
|
|
|
|
* Function:
|
2009-10-27 10:15:58 +00:00
|
|
|
* Create new shadow object backing dst_entry with private copy of
|
|
|
|
* all underlying pages. When src_entry is equal to dst_entry,
|
|
|
|
* function implements COW for wired-down map entry. Otherwise,
|
|
|
|
* it forks wired entry into dst_map.
|
1994-05-24 10:09:53 +00:00
|
|
|
*
|
|
|
|
* In/out conditions:
|
|
|
|
* The source and destination maps must be locked for write.
|
|
|
|
* The source map entry must be wired down (or be a sharing map
|
|
|
|
* entry corresponding to a main map entry that is wired down).
|
|
|
|
*/
|
1994-05-25 09:21:21 +00:00
|
|
|
void
|
2009-07-03 22:17:37 +00:00
|
|
|
vm_fault_copy_entry(vm_map_t dst_map, vm_map_t src_map,
|
|
|
|
vm_map_entry_t dst_entry, vm_map_entry_t src_entry,
|
|
|
|
vm_ooffset_t *fork_charge)
|
1994-05-24 10:09:53 +00:00
|
|
|
{
|
2009-10-27 10:15:58 +00:00
|
|
|
vm_object_t backing_object, dst_object, object, src_object;
|
2009-10-26 00:01:52 +00:00
|
|
|
vm_pindex_t dst_pindex, pindex, src_pindex;
|
2009-10-27 10:15:58 +00:00
|
|
|
vm_prot_t access, prot;
|
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_offset_t vaddr;
|
|
|
|
vm_page_t dst_m;
|
|
|
|
vm_page_t src_m;
|
2009-10-27 10:15:58 +00:00
|
|
|
boolean_t src_readonly, upgrade;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
#ifdef lint
|
|
|
|
src_map++;
|
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
|
|
|
#endif /* lint */
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2009-10-27 10:15:58 +00:00
|
|
|
upgrade = src_entry == dst_entry;
|
|
|
|
|
1994-05-24 10:09:53 +00:00
|
|
|
src_object = src_entry->object.vm_object;
|
2009-10-26 00:01:52 +00:00
|
|
|
src_pindex = OFF_TO_IDX(src_entry->offset);
|
|
|
|
src_readonly = (src_entry->protection & VM_PROT_WRITE) == 0;
|
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
|
|
|
* Create the top-level object for the destination entry. (Doesn't
|
|
|
|
* actually shadow anything - we copy the pages directly.)
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
NOTE: libkvm, w, ps, 'top', and any other utility which depends on struct
proc or any VM system structure will have to be rebuilt!!!
Much needed overhaul of the VM system. Included in this first round of
changes:
1) Improved pager interfaces: init, alloc, dealloc, getpages, putpages,
haspage, and sync operations are supported. The haspage interface now
provides information about clusterability. All pager routines now take
struct vm_object's instead of "pagers".
2) Improved data structures. In the previous paradigm, there is constant
confusion caused by pagers being both a data structure ("allocate a
pager") and a collection of routines. The idea of a pager structure has
escentially been eliminated. Objects now have types, and this type is
used to index the appropriate pager. In most cases, items in the pager
structure were duplicated in the object data structure and thus were
unnecessary. In the few cases that remained, a un_pager structure union
was created in the object to contain these items.
3) Because of the cleanup of #1 & #2, a lot of unnecessary layering can now
be removed. For instance, vm_object_enter(), vm_object_lookup(),
vm_object_remove(), and the associated object hash list were some of the
things that were removed.
4) simple_lock's removed. Discussion with several people reveals that the
SMP locking primitives used in the VM system aren't likely the mechanism
that we'll be adopting. Even if it were, the locking that was in the code
was very inadequate and would have to be mostly re-done anyway. The
locking in a uni-processor kernel was a no-op but went a long way toward
making the code difficult to read and debug.
5) Places that attempted to kludge-up the fact that we don't have kernel
thread support have been fixed to reflect the reality that we are really
dealing with processes, not threads. The VM system didn't have complete
thread support, so the comments and mis-named routines were just wrong.
We now use tsleep and wakeup directly in the lock routines, for instance.
6) Where appropriate, the pagers have been improved, especially in the
pager_alloc routines. Most of the pager_allocs have been rewritten and
are now faster and easier to maintain.
7) The pagedaemon pageout clustering algorithm has been rewritten and
now tries harder to output an even number of pages before and after
the requested page. This is sort of the reverse of the ideal pagein
algorithm and should provide better overall performance.
8) Unnecessary (incorrect) casts to caddr_t in calls to tsleep & wakeup
have been removed. Some other unnecessary casts have also been removed.
9) Some almost useless debugging code removed.
10) Terminology of shadow objects vs. backing objects straightened out.
The fact that the vm_object data structure escentially had this
backwards really confused things. The use of "shadow" and "backing
object" throughout the code is now internally consistent and correct
in the Mach terminology.
11) Several minor bug fixes, including one in the vm daemon that caused
0 RSS objects to not get purged as intended.
12) A "default pager" has now been created which cleans up the transition
of objects to the "swap" type. The previous checks throughout the code
for swp->pg_data != NULL were really ugly. This change also provides
the rudiments for future backing of "anonymous" memory by something
other than the swap pager (via the vnode pager, for example), and it
allows the decision about which of these pagers to use to be made
dynamically (although will need some additional decision code to do
this, of course).
13) (dyson) MAP_COPY has been deprecated and the corresponding "copy
object" code has been removed. MAP_COPY was undocumented and non-
standard. It was furthermore broken in several ways which caused its
behavior to degrade to MAP_PRIVATE. Binaries that use MAP_COPY will
continue to work correctly, but via the slightly different semantics
of MAP_PRIVATE.
14) (dyson) Sharing maps have been removed. It's marginal usefulness in a
threads design can be worked around in other ways. Both #12 and #13
were done to simplify the code and improve readability and maintain-
ability. (As were most all of these changes)
TODO:
1) Rewrite most of the vnode pager to use VOP_GETPAGES/PUTPAGES. Doing
this will reduce the vnode pager to a mere fraction of its current size.
2) Rewrite vm_fault and the swap/vnode pagers to use the clustering
information provided by the new haspage pager interface. This will
substantially reduce the overhead by eliminating a large number of
VOP_BMAP() calls. The VOP_BMAP() filesystem interface should be
improved to provide both a "behind" and "ahead" indication of
contiguousness.
3) Implement the extended features of pager_haspage in swap_pager_haspage().
It currently just says 0 pages ahead/behind.
4) Re-implement the swap device (swstrategy) in a more elegant way, perhaps
via a much more general mechanism that could also be used for disk
striping of regular filesystems.
5) Do something to improve the architecture of vm_object_collapse(). The
fact that it makes calls into the swap pager and knows too much about
how the swap pager operates really bothers me. It also doesn't allow
for collapsing of non-swap pager objects ("unnamed" objects backed by
other pagers).
1995-07-13 08:48:48 +00:00
|
|
|
dst_object = vm_object_allocate(OBJT_DEFAULT,
|
2005-09-07 01:42:30 +00:00
|
|
|
OFF_TO_IDX(dst_entry->end - dst_entry->start));
|
2007-12-29 19:53:04 +00:00
|
|
|
#if VM_NRESERVLEVEL > 0
|
|
|
|
dst_object->flags |= OBJ_COLORED;
|
|
|
|
dst_object->pg_color = atop(dst_entry->start);
|
|
|
|
#endif
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(dst_object);
|
2009-10-27 10:15:58 +00:00
|
|
|
KASSERT(upgrade || dst_entry->object.vm_object == NULL,
|
2009-07-03 22:17:37 +00:00
|
|
|
("vm_fault_copy_entry: vm_object not NULL"));
|
1994-05-24 10:09:53 +00:00
|
|
|
dst_entry->object.vm_object = dst_object;
|
|
|
|
dst_entry->offset = 0;
|
2009-07-03 22:17:37 +00:00
|
|
|
dst_object->charge = dst_entry->end - dst_entry->start;
|
2009-10-27 10:15:58 +00:00
|
|
|
if (fork_charge != NULL) {
|
2010-12-02 17:37:16 +00:00
|
|
|
KASSERT(dst_entry->cred == NULL,
|
2009-10-27 10:15:58 +00:00
|
|
|
("vm_fault_copy_entry: leaked swp charge"));
|
2010-12-02 17:37:16 +00:00
|
|
|
dst_object->cred = curthread->td_ucred;
|
|
|
|
crhold(dst_object->cred);
|
2009-10-27 10:15:58 +00:00
|
|
|
*fork_charge += dst_object->charge;
|
|
|
|
} else {
|
2010-12-02 17:37:16 +00:00
|
|
|
dst_object->cred = dst_entry->cred;
|
|
|
|
dst_entry->cred = NULL;
|
2009-10-27 10:15:58 +00:00
|
|
|
}
|
2009-10-31 17:39:56 +00:00
|
|
|
access = prot = dst_entry->protection;
|
2009-10-27 10:15:58 +00:00
|
|
|
/*
|
|
|
|
* If not an upgrade, then enter the mappings in the pmap as
|
|
|
|
* read and/or execute accesses. Otherwise, enter them as
|
|
|
|
* write accesses.
|
|
|
|
*
|
|
|
|
* A writeable large page mapping is only created if all of
|
|
|
|
* the constituent small page mappings are modified. Marking
|
|
|
|
* PTEs as modified on inception allows promotion to happen
|
|
|
|
* without taking potentially large number of soft faults.
|
|
|
|
*/
|
|
|
|
if (!upgrade)
|
|
|
|
access &= ~VM_PROT_WRITE;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
/*
|
2012-10-24 18:32:37 +00:00
|
|
|
* Loop through all of the virtual pages within the entry's
|
|
|
|
* range, copying each page from the source object to the
|
|
|
|
* destination object. Since the source is wired, those pages
|
|
|
|
* must exist. In contrast, the destination is pageable.
|
|
|
|
* Since the destination object does share any backing storage
|
|
|
|
* with the source object, all of its pages must be dirtied,
|
|
|
|
* regardless of whether they can be written.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2009-10-26 00:01:52 +00:00
|
|
|
for (vaddr = dst_entry->start, dst_pindex = 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
|
|
|
vaddr < dst_entry->end;
|
2009-10-26 00:01:52 +00:00
|
|
|
vaddr += PAGE_SIZE, dst_pindex++) {
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
/*
|
2009-10-26 00:01:52 +00:00
|
|
|
* Allocate a page in the destination object.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
|
|
|
do {
|
2009-10-26 00:01:52 +00:00
|
|
|
dst_m = vm_page_alloc(dst_object, dst_pindex,
|
|
|
|
VM_ALLOC_NORMAL);
|
1994-05-24 10:09:53 +00:00
|
|
|
if (dst_m == NULL) {
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(dst_object);
|
1994-05-24 10:09:53 +00:00
|
|
|
VM_WAIT;
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(dst_object);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
|
|
|
} while (dst_m == NULL);
|
|
|
|
|
|
|
|
/*
|
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
|
|
|
* Find the page in the source object, and copy it in.
|
|
|
|
* (Because the source is wired down, the page will be in
|
|
|
|
* memory.)
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2013-05-22 15:11:00 +00:00
|
|
|
VM_OBJECT_RLOCK(src_object);
|
2003-10-15 08:00:45 +00:00
|
|
|
object = src_object;
|
2009-10-26 00:01:52 +00:00
|
|
|
pindex = src_pindex + dst_pindex;
|
|
|
|
while ((src_m = vm_page_lookup(object, pindex)) == NULL &&
|
|
|
|
src_readonly &&
|
2003-10-15 08:00:45 +00:00
|
|
|
(backing_object = object->backing_object) != NULL) {
|
|
|
|
/*
|
|
|
|
* Allow fallback to backing objects if we are reading.
|
|
|
|
*/
|
2013-05-22 15:11:00 +00:00
|
|
|
VM_OBJECT_RLOCK(backing_object);
|
2003-10-15 08:00:45 +00:00
|
|
|
pindex += OFF_TO_IDX(object->backing_object_offset);
|
2013-05-22 15:11:00 +00:00
|
|
|
VM_OBJECT_RUNLOCK(object);
|
2003-10-15 08:00:45 +00:00
|
|
|
object = backing_object;
|
|
|
|
}
|
1994-05-24 10:09:53 +00:00
|
|
|
if (src_m == NULL)
|
|
|
|
panic("vm_fault_copy_wired: page missing");
|
2003-10-08 05:35:12 +00:00
|
|
|
pmap_copy_page(src_m, dst_m);
|
2013-05-22 15:11:00 +00:00
|
|
|
VM_OBJECT_RUNLOCK(object);
|
2003-10-08 05:35:12 +00:00
|
|
|
dst_m->valid = VM_PAGE_BITS_ALL;
|
2012-10-24 18:21:59 +00:00
|
|
|
dst_m->dirty = VM_PAGE_BITS_ALL;
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(dst_object);
|
1994-05-24 10:09:53 +00:00
|
|
|
|
|
|
|
/*
|
2009-10-27 10:15:58 +00:00
|
|
|
* Enter it in the pmap. If a wired, copy-on-write
|
|
|
|
* mapping is being replaced by a write-enabled
|
|
|
|
* mapping, then wire that new mapping.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2009-10-27 10:15:58 +00:00
|
|
|
pmap_enter(dst_map->pmap, vaddr, access, dst_m, prot, upgrade);
|
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
|
|
|
* Mark it no longer busy, and put it on the active list.
|
1994-05-24 10:09:53 +00:00
|
|
|
*/
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WLOCK(dst_object);
|
2010-04-30 00:46:43 +00:00
|
|
|
|
2009-10-27 10:15:58 +00:00
|
|
|
if (upgrade) {
|
2010-04-30 00:46:43 +00:00
|
|
|
vm_page_lock(src_m);
|
2009-10-27 10:15:58 +00:00
|
|
|
vm_page_unwire(src_m, 0);
|
2010-04-30 16:20:14 +00:00
|
|
|
vm_page_unlock(src_m);
|
2010-04-30 00:46:43 +00:00
|
|
|
|
|
|
|
vm_page_lock(dst_m);
|
2009-10-27 10:15:58 +00:00
|
|
|
vm_page_wire(dst_m);
|
2010-04-30 16:20:14 +00:00
|
|
|
vm_page_unlock(dst_m);
|
2010-04-30 00:46:43 +00:00
|
|
|
} else {
|
|
|
|
vm_page_lock(dst_m);
|
2009-10-27 10:15:58 +00:00
|
|
|
vm_page_activate(dst_m);
|
2010-04-30 16:20:14 +00:00
|
|
|
vm_page_unlock(dst_m);
|
2010-04-30 00:46:43 +00:00
|
|
|
}
|
2013-08-09 11:11:11 +00:00
|
|
|
vm_page_xunbusy(dst_m);
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_WUNLOCK(dst_object);
|
2009-10-27 10:15:58 +00:00
|
|
|
if (upgrade) {
|
|
|
|
dst_entry->eflags &= ~(MAP_ENTRY_COW | MAP_ENTRY_NEEDS_COPY);
|
|
|
|
vm_object_deallocate(src_object);
|
|
|
|
}
|
1994-05-25 09:21:21 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/*
|
|
|
|
* This routine checks around the requested page for other pages that
|
1995-09-24 19:47:58 +00:00
|
|
|
* might be able to be faulted in. This routine brackets the viable
|
|
|
|
* pages for the pages to be paged in.
|
1994-05-25 09:21:21 +00:00
|
|
|
*
|
|
|
|
* Inputs:
|
1995-09-24 19:47:58 +00:00
|
|
|
* m, rbehind, rahead
|
1994-05-25 09:21:21 +00:00
|
|
|
*
|
|
|
|
* Outputs:
|
|
|
|
* marray (array of vm_page_t), reqpage (index of requested page)
|
|
|
|
*
|
|
|
|
* Return value:
|
|
|
|
* number of pages in marray
|
|
|
|
*/
|
1998-02-09 06:11:36 +00:00
|
|
|
static int
|
1995-09-24 19:47:58 +00:00
|
|
|
vm_fault_additional_pages(m, rbehind, rahead, marray, reqpage)
|
1994-05-25 09:21:21 +00:00
|
|
|
vm_page_t m;
|
|
|
|
int rbehind;
|
1995-09-24 19:47:58 +00:00
|
|
|
int rahead;
|
1994-05-25 09:21:21 +00:00
|
|
|
vm_page_t *marray;
|
|
|
|
int *reqpage;
|
|
|
|
{
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
int i,j;
|
1994-05-25 09:21:21 +00:00
|
|
|
vm_object_t object;
|
1995-12-11 04:58:34 +00:00
|
|
|
vm_pindex_t pindex, startpindex, endpindex, tpindex;
|
1994-05-25 09:21:21 +00:00
|
|
|
vm_page_t rtm;
|
1995-09-04 04:44:26 +00:00
|
|
|
int cbehind, cahead;
|
1994-05-25 09:21:21 +00:00
|
|
|
|
2013-03-09 02:32:23 +00:00
|
|
|
VM_OBJECT_ASSERT_WLOCKED(m->object);
|
2001-05-19 01:28:09 +00:00
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
object = m->object;
|
1995-12-11 04:58:34 +00:00
|
|
|
pindex = m->pindex;
|
2007-04-05 20:49:46 +00:00
|
|
|
cbehind = cahead = 0;
|
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
/*
|
|
|
|
* if the requested page is not available, then give up now
|
|
|
|
*/
|
1999-01-21 08:29:12 +00:00
|
|
|
if (!vm_pager_has_page(object, pindex, &cbehind, &cahead)) {
|
1994-05-25 09:21:21 +00:00
|
|
|
return 0;
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
}
|
1994-05-25 09:21:21 +00:00
|
|
|
|
1995-09-24 19:47:58 +00:00
|
|
|
if ((cbehind == 0) && (cahead == 0)) {
|
|
|
|
*reqpage = 0;
|
|
|
|
marray[0] = m;
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (rahead > cahead) {
|
|
|
|
rahead = cahead;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (rbehind > cbehind) {
|
|
|
|
rbehind = cbehind;
|
1995-09-04 04:44:26 +00:00
|
|
|
}
|
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
/*
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
* scan backward for the read behind pages -- in memory
|
1994-05-25 09:21:21 +00:00
|
|
|
*/
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
if (pindex > 0) {
|
|
|
|
if (rbehind > pindex) {
|
1995-12-11 04:58:34 +00:00
|
|
|
rbehind = pindex;
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
startpindex = 0;
|
|
|
|
} else {
|
|
|
|
startpindex = pindex - rbehind;
|
|
|
|
}
|
|
|
|
|
2006-05-13 20:05:44 +00:00
|
|
|
if ((rtm = TAILQ_PREV(m, pglist, listq)) != NULL &&
|
|
|
|
rtm->pindex >= startpindex)
|
|
|
|
startpindex = rtm->pindex + 1;
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
|
2007-07-20 06:55:11 +00:00
|
|
|
/* tpindex is unsigned; beware of numeric underflow. */
|
|
|
|
for (i = 0, tpindex = pindex - 1; tpindex >= startpindex &&
|
|
|
|
tpindex < pindex; i++, tpindex--) {
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +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
|
|
|
rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL |
|
|
|
|
VM_ALLOC_IFNOTCACHED);
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
if (rtm == NULL) {
|
2007-07-20 06:55:11 +00:00
|
|
|
/*
|
|
|
|
* Shift the allocated pages to the
|
|
|
|
* beginning of the array.
|
|
|
|
*/
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
for (j = 0; j < i; j++) {
|
2007-07-20 06:55:11 +00:00
|
|
|
marray[j] = marray[j + tpindex + 1 -
|
|
|
|
startpindex];
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
}
|
2007-07-20 06:55:11 +00:00
|
|
|
break;
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
}
|
|
|
|
|
2007-07-20 06:55:11 +00:00
|
|
|
marray[tpindex - startpindex] = rtm;
|
1994-05-25 09:21:21 +00:00
|
|
|
}
|
1994-11-13 22:48:55 +00:00
|
|
|
} else {
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
startpindex = 0;
|
|
|
|
i = 0;
|
1994-05-25 09:21:21 +00:00
|
|
|
}
|
|
|
|
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
marray[i] = m;
|
|
|
|
/* page offset of the required page */
|
|
|
|
*reqpage = i;
|
|
|
|
|
|
|
|
tpindex = pindex + 1;
|
|
|
|
i++;
|
|
|
|
|
1994-05-25 09:21:21 +00:00
|
|
|
/*
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
* scan forward for the read ahead pages
|
1994-05-25 09:21:21 +00:00
|
|
|
*/
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
endpindex = tpindex + rahead;
|
2006-05-13 20:05:44 +00:00
|
|
|
if ((rtm = TAILQ_NEXT(m, listq)) != NULL && rtm->pindex < endpindex)
|
|
|
|
endpindex = rtm->pindex;
|
1995-12-11 04:58:34 +00:00
|
|
|
if (endpindex > object->size)
|
|
|
|
endpindex = object->size;
|
1994-05-24 10:09:53 +00:00
|
|
|
|
2002-03-10 21:52:48 +00:00
|
|
|
for (; tpindex < endpindex; i++, tpindex++) {
|
1994-05-25 09:21:21 +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
|
|
|
rtm = vm_page_alloc(object, tpindex, VM_ALLOC_NORMAL |
|
|
|
|
VM_ALLOC_IFNOTCACHED);
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
if (rtm == NULL) {
|
|
|
|
break;
|
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
|
|
|
}
|
1995-09-04 04:44:26 +00:00
|
|
|
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
marray[i] = rtm;
|
1994-05-25 09:21:21 +00:00
|
|
|
}
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
|
2007-07-06 21:25:21 +00:00
|
|
|
/* return number of pages */
|
VM level code cleanups.
1) Start using TSM.
Struct procs continue to point to upages structure, after being freed.
Struct vmspace continues to point to pte object and kva space for kstack.
u_map is now superfluous.
2) vm_map's don't need to be reference counted. They always exist either
in the kernel or in a vmspace. The vmspaces are managed by reference
counts.
3) Remove the "wired" vm_map nonsense.
4) No need to keep a cache of kernel stack kva's.
5) Get rid of strange looking ++var, and change to var++.
6) Change more data structures to use our "zone" allocator. Added
struct proc, struct vmspace and struct vnode. This saves a significant
amount of kva space and physical memory. Additionally, this enables
TSM for the zone managed memory.
7) Keep ioopt disabled for now.
8) Remove the now bogus "single use" map concept.
9) Use generation counts or id's for data structures residing in TSM, where
it allows us to avoid unneeded restart overhead during traversals, where
blocking might occur.
10) Account better for memory deficits, so the pageout daemon will be able
to make enough memory available (experimental.)
11) Fix some vnode locking problems. (From Tor, I think.)
12) Add a check in ufs_lookup, to avoid lots of unneeded calls to bcmp.
(experimental.)
13) Significantly shrink, cleanup, and make slightly faster the vm_fault.c
code. Use generation counts, get rid of unneded collpase operations,
and clean up the cluster code.
14) Make vm_zone more suitable for TSM.
This commit is partially as a result of discussions and contributions from
other people, including DG, Tor Egge, PHK, and probably others that I
have forgotten to attribute (so let me know, if I forgot.)
This is not the infamous, final cleanup of the vnode stuff, but a necessary
step. Vnode mgmt should be correct, but things might still change, and
there is still some missing stuff (like ioopt, and physical backing of
non-merged cache files, debugging of layering concepts.)
1998-01-22 17:30:44 +00:00
|
|
|
return i;
|
1994-05-24 10:09:53 +00:00
|
|
|
}
|
2011-07-09 15:21:10 +00:00
|
|
|
|
Handle spurious page faults that may occur in no-fault sections of the
kernel.
When access restrictions are added to a page table entry, we flush the
corresponding virtual address mapping from the TLB. In contrast, when
access restrictions are removed from a page table entry, we do not
flush the virtual address mapping from the TLB. This is exactly as
recommended in AMD's documentation. In effect, when access
restrictions are removed from a page table entry, AMD's MMUs will
transparently refresh a stale TLB entry. In short, this saves us from
having to perform potentially costly TLB flushes. In contrast,
Intel's MMUs are allowed to generate a spurious page fault based upon
the stale TLB entry. Usually, such spurious page faults are handled
by vm_fault() without incident. However, when we are executing
no-fault sections of the kernel, we are not allowed to execute
vm_fault(). This change introduces special-case handling for spurious
page faults that occur in no-fault sections of the kernel.
In collaboration with: kib
Tested by: gibbs (an earlier version)
I would also like to acknowledge Hiroki Sato's assistance in
diagnosing this problem.
MFC after: 1 week
2012-03-22 04:52:51 +00:00
|
|
|
/*
|
|
|
|
* Block entry into the machine-independent layer's page fault handler by
|
|
|
|
* the calling thread. Subsequent calls to vm_fault() by that thread will
|
|
|
|
* return KERN_PROTECTION_FAILURE. Enable machine-dependent handling of
|
|
|
|
* spurious page faults.
|
|
|
|
*/
|
2011-07-09 15:21:10 +00:00
|
|
|
int
|
|
|
|
vm_fault_disable_pagefaults(void)
|
|
|
|
{
|
|
|
|
|
Handle spurious page faults that may occur in no-fault sections of the
kernel.
When access restrictions are added to a page table entry, we flush the
corresponding virtual address mapping from the TLB. In contrast, when
access restrictions are removed from a page table entry, we do not
flush the virtual address mapping from the TLB. This is exactly as
recommended in AMD's documentation. In effect, when access
restrictions are removed from a page table entry, AMD's MMUs will
transparently refresh a stale TLB entry. In short, this saves us from
having to perform potentially costly TLB flushes. In contrast,
Intel's MMUs are allowed to generate a spurious page fault based upon
the stale TLB entry. Usually, such spurious page faults are handled
by vm_fault() without incident. However, when we are executing
no-fault sections of the kernel, we are not allowed to execute
vm_fault(). This change introduces special-case handling for spurious
page faults that occur in no-fault sections of the kernel.
In collaboration with: kib
Tested by: gibbs (an earlier version)
I would also like to acknowledge Hiroki Sato's assistance in
diagnosing this problem.
MFC after: 1 week
2012-03-22 04:52:51 +00:00
|
|
|
return (curthread_pflags_set(TDP_NOFAULTING | TDP_RESETSPUR));
|
2011-07-09 15:21:10 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void
|
|
|
|
vm_fault_enable_pagefaults(int save)
|
|
|
|
{
|
|
|
|
|
|
|
|
curthread_pflags_restore(save);
|
|
|
|
}
|