d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
4a240f6ce7
freebsd-skq/sys/vm/vm_reserv.c

1003 lines
30 KiB
C
Raw Normal View History

Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*-
* Copyright (c) 2002-2006 Rice University
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
* Copyright (c) 2007-2011 Alan L. Cox <alc@cs.rice.edu>
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
* All rights reserved.
*
* This software was developed for the FreeBSD Project by Alan L. Cox,
* Olivier Crameri, Peter Druschel, Sitaram Iyer, and Juan Navarro.
*
* 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.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS 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 COPYRIGHT
* HOLDERS 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.
*/
/*
* Superpage reservation management module
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
*
* Any external functions defined by this module are only to be used by the
* virtual memory system.
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*/
#include <sys/cdefs.h>
__FBSDID("$FreeBSD$");
#include "opt_vm.h"
#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mutex.h>
#include <sys/queue.h>
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
#include <sys/rwlock.h>
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
#include <sys/sbuf.h>
#include <sys/sysctl.h>
#include <sys/systm.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_phys.h>
Sync back vmcontention branch into HEAD: Replace the per-object resident and cached pages splay tree with a path-compressed multi-digit radix trie. Along with this, switch also the x86-specific handling of idle page tables to using the radix trie. This change is supposed to do the following: - Allowing the acquisition of read locking for lookup operations of the resident/cached pages collections as the per-vm_page_t splay iterators are now removed. - Increase the scalability of the operations on the page collections. The radix trie does rely on the consumers locking to ensure atomicity of its operations. In order to avoid deadlocks the bisection nodes are pre-allocated in the UMA zone. This can be done safely because the algorithm needs at maximum one new node per insert which means the maximum number of the desired nodes is the number of available physical frames themselves. However, not all the times a new bisection node is really needed. The radix trie implements path-compression because UFS indirect blocks can lead to several objects with a very sparse trie, increasing the number of levels to usually scan. It also helps in the nodes pre-fetching by introducing the single node per-insert property. This code is not generalized (yet) because of the possible loss of performance by having much of the sizes in play configurable. However, efforts to make this code more general and then reusable in further different consumers might be really done. The only KPI change is the removal of the function vm_page_splay() which is now reaped. The only KBI change, instead, is the removal of the left/right iterators from struct vm_page, which are now reaped. Further technical notes broken into mealpieces can be retrieved from the svn branch: http://svn.freebsd.org/base/user/attilio/vmcontention/ Sponsored by: EMC / Isilon storage division In collaboration with: alc, jeff Tested by: flo, pho, jhb, davide Tested by: ian (arm) Tested by: andreast (powerpc)
2013-03-18 00:25:02 +00:00
#include <vm/vm_radix.h>
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
#include <vm/vm_reserv.h>
/*
* The reservation system supports the speculative allocation of large physical
* pages ("superpages"). Speculative allocation enables the fully-automatic
* utilization of superpages by the virtual memory system. In other words, no
* programmatic directives are required to use superpages.
*/
#if VM_NRESERVLEVEL > 0
/*
* The number of small pages that are contained in a level 0 reservation
*/
#define VM_LEVEL_0_NPAGES (1 << VM_LEVEL_0_ORDER)
/*
* The number of bits by which a physical address is shifted to obtain the
* reservation number
*/
#define VM_LEVEL_0_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT)
/*
* The size of a level 0 reservation in bytes
*/
#define VM_LEVEL_0_SIZE (1 << VM_LEVEL_0_SHIFT)
/*
* Computes the index of the small page underlying the given (object, pindex)
* within the reservation's array of small pages.
*/
#define VM_RESERV_INDEX(object, pindex) \
(((object)->pg_color + (pindex)) & (VM_LEVEL_0_NPAGES - 1))
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
/*
* The size of a population map entry
*/
typedef u_long popmap_t;
/*
* The number of bits in a population map entry
*/
#define NBPOPMAP (NBBY * sizeof(popmap_t))
/*
* The number of population map entries in a reservation
*/
#define NPOPMAP howmany(VM_LEVEL_0_NPAGES, NBPOPMAP)
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
* The reservation structure
*
* A reservation structure is constructed whenever a large physical page is
* speculatively allocated to an object. The reservation provides the small
* physical pages for the range [pindex, pindex + VM_LEVEL_0_NPAGES) of offsets
* within that object. The reservation's "popcnt" tracks the number of these
* small physical pages that are in use at any given time. When and if the
* reservation is not fully utilized, it appears in the queue of partially-
* populated reservations. The reservation always appears on the containing
* object's list of reservations.
*
* A partially-populated reservation can be broken and reclaimed at any time.
*/
struct vm_reserv {
TAILQ_ENTRY(vm_reserv) partpopq;
LIST_ENTRY(vm_reserv) objq;
vm_object_t object; /* containing object */
vm_pindex_t pindex; /* offset within object */
vm_page_t pages; /* first page of a superpage */
int popcnt; /* # of pages in use */
char inpartpopq;
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
popmap_t popmap[NPOPMAP]; /* bit vector of used pages */
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
};
/*
* The reservation array
*
* This array is analoguous in function to vm_page_array. It differs in the
* respect that it may contain a greater number of useful reservation
* structures than there are (physical) superpages. These "invalid"
* reservation structures exist to trade-off space for time in the
* implementation of vm_reserv_from_page(). Invalid reservation structures are
* distinguishable from "valid" reservation structures by inspecting the
* reservation's "pages" field. Invalid reservation structures have a NULL
* "pages" field.
*
* vm_reserv_from_page() maps a small (physical) page to an element of this
* array by computing a physical reservation number from the page's physical
* address. The physical reservation number is used as the array index.
*
* An "active" reservation is a valid reservation structure that has a non-NULL
* "object" field and a non-zero "popcnt" field. In other words, every active
* reservation belongs to a particular object. Moreover, every active
* reservation has an entry in the containing object's list of reservations.
*/
static vm_reserv_t vm_reserv_array;
/*
* The partially-populated reservation queue
*
* This queue enables the fast recovery of an unused cached or free small page
Previously, when vm_page_free_toq() was performed on a page belonging to a reservation, unless all of the reservation's pages were free, the reservation was moved to the head of the partially-populated reservations queue, where it would be the next reservation to be broken in case the free page queues were emptied. Now, instead, I am moving it to the tail. Very likely this reservation is in the process of being freed in its entirety, so placing it at the tail of the queue makes it more likely that the underlying physical memory will be returned to the free page queues as one contiguous chunk. If a reservation must be broken, it will, instead, be the longest unchanged reservation, which is arguably the reservation that is least likely to ever achieve promotion or be freed in its entirety. MFC after: 6 weeks
2009-04-11 09:09:00 +00:00
* from a partially-populated reservation. The reservation at the head of
* this queue is the least-recently-changed, partially-populated reservation.
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*
* Access to this queue is synchronized by the free page queue lock.
*/
static TAILQ_HEAD(, vm_reserv) vm_rvq_partpop =
TAILQ_HEAD_INITIALIZER(vm_rvq_partpop);
static SYSCTL_NODE(_vm, OID_AUTO, reserv, CTLFLAG_RD, 0, "Reservation Info");
static long vm_reserv_broken;
SYSCTL_LONG(_vm_reserv, OID_AUTO, broken, CTLFLAG_RD,
&vm_reserv_broken, 0, "Cumulative number of broken reservations");
static long vm_reserv_freed;
SYSCTL_LONG(_vm_reserv, OID_AUTO, freed, CTLFLAG_RD,
&vm_reserv_freed, 0, "Cumulative number of freed reservations");
static int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS);
SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
sysctl_vm_reserv_partpopq, "A", "Partially-populated reservation queues");
static long vm_reserv_reclaimed;
SYSCTL_LONG(_vm_reserv, OID_AUTO, reclaimed, CTLFLAG_RD,
&vm_reserv_reclaimed, 0, "Cumulative number of reclaimed reservations");
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
static void vm_reserv_break(vm_reserv_t rv, vm_page_t m);
static void vm_reserv_depopulate(vm_reserv_t rv, int index);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
static vm_reserv_t vm_reserv_from_page(vm_page_t m);
static boolean_t vm_reserv_has_pindex(vm_reserv_t rv,
vm_pindex_t pindex);
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
static void vm_reserv_populate(vm_reserv_t rv, int index);
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
static void vm_reserv_reclaim(vm_reserv_t rv);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
* Describes the current state of the partially-populated reservation queue.
*/
static int
sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS)
{
struct sbuf sbuf;
vm_reserv_t rv;
int counter, error, level, unused_pages;
Explicitly wire the user buffer rather than doing it implicitly in sbuf_new_for_sysctl(9). This allows using an sbuf with a SYSCTL_OUT drain for extremely large amounts of data where the caller knows that appropriate references are held, and sleeping is not an issue. Inspired by: rwatson
2011-01-27 00:34:12 +00:00
error = sysctl_wire_old_buffer(req, 0);
if (error != 0)
return (error);
Re-add r212370 now that the LOR in powerpc64 has been resolved: Add a drain function for struct sysctl_req, and use it for a variety of handlers, some of which had to do awkward things to get a large enough SBUF_FIXEDLEN buffer. Note that some sysctl handlers were explicitly outputting a trailing NUL byte. This behaviour was preserved, though it should not be necessary. Reviewed by: phk (original patch)
2010-09-16 16:13:12 +00:00
sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
sbuf_printf(&sbuf, "\nLEVEL SIZE NUMBER\n\n");
for (level = -1; level <= VM_NRESERVLEVEL - 2; level++) {
counter = 0;
unused_pages = 0;
mtx_lock(&vm_page_queue_free_mtx);
TAILQ_FOREACH(rv, &vm_rvq_partpop/*[level]*/, partpopq) {
counter++;
unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt;
}
mtx_unlock(&vm_page_queue_free_mtx);
Correct some format strings used by sysctls. MFC after: 1 week
2010-10-30 18:00:53 +00:00
sbuf_printf(&sbuf, "%5d: %6dK, %6d\n", level,
Enable reservation-based physical memory allocation. Even without the creation of large page mappings in the pmap, it can provide modest performance benefits. In particular, for a "buildworld" on a 2x 1GHz Ultrasparc IIIi it reduced the wall clock time by 2.2% and the system time by 12.6%. Tested by: marius@
2010-11-10 17:57:34 +00:00
unused_pages * ((int)PAGE_SIZE / 1024), counter);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
}
Re-add r212370 now that the LOR in powerpc64 has been resolved: Add a drain function for struct sysctl_req, and use it for a variety of handlers, some of which had to do awkward things to get a large enough SBUF_FIXEDLEN buffer. Note that some sysctl handlers were explicitly outputting a trailing NUL byte. This behaviour was preserved, though it should not be necessary. Reviewed by: phk (original patch)
2010-09-16 16:13:12 +00:00
error = sbuf_finish(&sbuf);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
sbuf_delete(&sbuf);
return (error);
}
/*
* Reduces the given reservation's population count. If the population count
* becomes zero, the reservation is destroyed. Additionally, moves the
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
* reservation to the tail of the partially-populated reservation queue if the
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
* population count is non-zero.
*
* The free page queue lock must be held.
*/
static void
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_depopulate(vm_reserv_t rv, int index)
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
{
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
KASSERT(rv->object != NULL,
("vm_reserv_depopulate: reserv %p is free", rv));
KASSERT(rv->popcnt > 0,
("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv));
if (rv->inpartpopq) {
TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
rv->inpartpopq = FALSE;
}
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
clrbit(rv->popmap, index);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv->popcnt--;
if (rv->popcnt == 0) {
LIST_REMOVE(rv, objq);
rv->object = NULL;
vm_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER);
vm_reserv_freed++;
} else {
rv->inpartpopq = TRUE;
Previously, when vm_page_free_toq() was performed on a page belonging to a reservation, unless all of the reservation's pages were free, the reservation was moved to the head of the partially-populated reservations queue, where it would be the next reservation to be broken in case the free page queues were emptied. Now, instead, I am moving it to the tail. Very likely this reservation is in the process of being freed in its entirety, so placing it at the tail of the queue makes it more likely that the underlying physical memory will be returned to the free page queues as one contiguous chunk. If a reservation must be broken, it will, instead, be the longest unchanged reservation, which is arguably the reservation that is least likely to ever achieve promotion or be freed in its entirety. MFC after: 6 weeks
2009-04-11 09:09:00 +00:00
TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
}
}
/*
* Returns the reservation to which the given page might belong.
*/
static __inline vm_reserv_t
vm_reserv_from_page(vm_page_t m)
{
return (&vm_reserv_array[VM_PAGE_TO_PHYS(m) >> VM_LEVEL_0_SHIFT]);
}
/*
* Returns TRUE if the given reservation contains the given page index and
* FALSE otherwise.
*/
static __inline boolean_t
vm_reserv_has_pindex(vm_reserv_t rv, vm_pindex_t pindex)
{
return (((pindex - rv->pindex) & ~(VM_LEVEL_0_NPAGES - 1)) == 0);
}
/*
* Increases the given reservation's population count. Moves the reservation
* to the tail of the partially-populated reservation queue.
*
* The free page queue must be locked.
*/
static void
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_populate(vm_reserv_t rv, int index)
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
{
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
KASSERT(rv->object != NULL,
("vm_reserv_populate: reserv %p is free", rv));
KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES,
("vm_reserv_populate: reserv %p is already full", rv));
if (rv->inpartpopq) {
TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
rv->inpartpopq = FALSE;
}
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
setbit(rv->popmap, index);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv->popcnt++;
if (rv->popcnt < VM_LEVEL_0_NPAGES) {
rv->inpartpopq = TRUE;
TAILQ_INSERT_TAIL(&vm_rvq_partpop, rv, partpopq);
}
}
/*
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
* Allocates a contiguous set of physical pages of the given size "npages"
* from an existing or newly-created reservation. All of the physical pages
* must be at or above the given physical address "low" and below the given
* physical address "high". The given value "alignment" determines the
* alignment of the first physical page in the set. If the given value
* "boundary" is non-zero, then the set of physical pages cannot cross any
* physical address boundary that is a multiple of that value. Both
* "alignment" and "boundary" must be a power of two.
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*
* The object and free page queue must be locked.
*/
vm_page_t
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
vm_reserv_alloc_contig(vm_object_t object, vm_pindex_t pindex, u_long npages,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
{
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
vm_paddr_t pa, size;
vm_page_t m, m_ret, mpred, msucc;
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
vm_pindex_t first, leftcap, rightcap;
vm_reserv_t rv;
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
u_long allocpages, maxpages, minpages;
int i, index, n;
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(object);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
KASSERT(npages != 0, ("vm_reserv_alloc_contig: npages is 0"));
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
* Is a reservation fundamentally impossible?
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*/
if (pindex < VM_RESERV_INDEX(object, pindex) ||
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
pindex + npages > object->size)
return (NULL);
/*
* All reservations of a particular size have the same alignment.
* Assuming that the first page is allocated from a reservation, the
* least significant bits of its physical address can be determined
* from its offset from the beginning of the reservation and the size
* of the reservation.
*
* Could the specified index within a reservation of the smallest
* possible size satisfy the alignment and boundary requirements?
*/
pa = VM_RESERV_INDEX(object, pindex) << PAGE_SHIFT;
if ((pa & (alignment - 1)) != 0)
return (NULL);
size = npages << PAGE_SHIFT;
if (((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
return (NULL);
/*
* Look for an existing reservation.
*/
Sync back vmcontention branch into HEAD: Replace the per-object resident and cached pages splay tree with a path-compressed multi-digit radix trie. Along with this, switch also the x86-specific handling of idle page tables to using the radix trie. This change is supposed to do the following: - Allowing the acquisition of read locking for lookup operations of the resident/cached pages collections as the per-vm_page_t splay iterators are now removed. - Increase the scalability of the operations on the page collections. The radix trie does rely on the consumers locking to ensure atomicity of its operations. In order to avoid deadlocks the bisection nodes are pre-allocated in the UMA zone. This can be done safely because the algorithm needs at maximum one new node per insert which means the maximum number of the desired nodes is the number of available physical frames themselves. However, not all the times a new bisection node is really needed. The radix trie implements path-compression because UFS indirect blocks can lead to several objects with a very sparse trie, increasing the number of levels to usually scan. It also helps in the nodes pre-fetching by introducing the single node per-insert property. This code is not generalized (yet) because of the possible loss of performance by having much of the sizes in play configurable. However, efforts to make this code more general and then reusable in further different consumers might be really done. The only KPI change is the removal of the function vm_page_splay() which is now reaped. The only KBI change, instead, is the removal of the left/right iterators from struct vm_page, which are now reaped. Further technical notes broken into mealpieces can be retrieved from the svn branch: http://svn.freebsd.org/base/user/attilio/vmcontention/ Sponsored by: EMC / Isilon storage division In collaboration with: alc, jeff Tested by: flo, pho, jhb, davide Tested by: ian (arm) Tested by: andreast (powerpc)
2013-03-18 00:25:02 +00:00
mpred = vm_radix_lookup_le(&object->rtree, pindex);
if (mpred != NULL) {
KASSERT(mpred->pindex < pindex,
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
("vm_reserv_alloc_contig: pindex already allocated"));
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv = vm_reserv_from_page(mpred);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
goto found;
Sync back vmcontention branch into HEAD: Replace the per-object resident and cached pages splay tree with a path-compressed multi-digit radix trie. Along with this, switch also the x86-specific handling of idle page tables to using the radix trie. This change is supposed to do the following: - Allowing the acquisition of read locking for lookup operations of the resident/cached pages collections as the per-vm_page_t splay iterators are now removed. - Increase the scalability of the operations on the page collections. The radix trie does rely on the consumers locking to ensure atomicity of its operations. In order to avoid deadlocks the bisection nodes are pre-allocated in the UMA zone. This can be done safely because the algorithm needs at maximum one new node per insert which means the maximum number of the desired nodes is the number of available physical frames themselves. However, not all the times a new bisection node is really needed. The radix trie implements path-compression because UFS indirect blocks can lead to several objects with a very sparse trie, increasing the number of levels to usually scan. It also helps in the nodes pre-fetching by introducing the single node per-insert property. This code is not generalized (yet) because of the possible loss of performance by having much of the sizes in play configurable. However, efforts to make this code more general and then reusable in further different consumers might be really done. The only KPI change is the removal of the function vm_page_splay() which is now reaped. The only KBI change, instead, is the removal of the left/right iterators from struct vm_page, which are now reaped. Further technical notes broken into mealpieces can be retrieved from the svn branch: http://svn.freebsd.org/base/user/attilio/vmcontention/ Sponsored by: EMC / Isilon storage division In collaboration with: alc, jeff Tested by: flo, pho, jhb, davide Tested by: ian (arm) Tested by: andreast (powerpc)
2013-03-18 00:25:02 +00:00
msucc = TAILQ_NEXT(mpred, listq);
} else
msucc = TAILQ_FIRST(&object->memq);
if (msucc != NULL) {
KASSERT(msucc->pindex > pindex,
("vm_reserv_alloc_page: pindex already allocated"));
rv = vm_reserv_from_page(msucc);
if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
goto found;
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
}
/*
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
* Could at least one reservation fit between the first index to the
* left that can be used and the first index to the right that cannot
* be used?
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*/
first = pindex - VM_RESERV_INDEX(object, pindex);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
if (mpred != NULL) {
if ((rv = vm_reserv_from_page(mpred))->object != object)
leftcap = mpred->pindex + 1;
else
leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
if (leftcap > first)
return (NULL);
}
minpages = VM_RESERV_INDEX(object, pindex) + npages;
maxpages = roundup2(minpages, VM_LEVEL_0_NPAGES);
allocpages = maxpages;
if (msucc != NULL) {
if ((rv = vm_reserv_from_page(msucc))->object != object)
rightcap = msucc->pindex;
else
rightcap = rv->pindex;
if (first + maxpages > rightcap) {
if (maxpages == VM_LEVEL_0_NPAGES)
return (NULL);
allocpages = minpages;
}
}
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
* Would the last new reservation extend past the end of the object?
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*/
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
if (first + maxpages > object->size) {
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
* Don't allocate the last new reservation if the object is a
* vnode or backed by another object that is a vnode.
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*/
if (object->type == OBJT_VNODE ||
(object->backing_object != NULL &&
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
object->backing_object->type == OBJT_VNODE)) {
if (maxpages == VM_LEVEL_0_NPAGES)
return (NULL);
allocpages = minpages;
}
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/* Speculate that the object may grow. */
}
/*
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
* Allocate and populate the new reservations. The alignment and
* boundary specified for this allocation may be different from the
* alignment and boundary specified for the requested pages. For
* instance, the specified index may not be the first page within the
* first new reservation.
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
*/
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
m = vm_phys_alloc_contig(allocpages, low, high, ulmax(alignment,
VM_LEVEL_0_SIZE), boundary > VM_LEVEL_0_SIZE ? boundary : 0);
if (m == NULL)
return (NULL);
m_ret = NULL;
index = VM_RESERV_INDEX(object, pindex);
do {
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv = vm_reserv_from_page(m);
KASSERT(rv->pages == m,
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
("vm_reserv_alloc_contig: reserv %p's pages is corrupted",
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv));
KASSERT(rv->object == NULL,
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
("vm_reserv_alloc_contig: reserv %p isn't free", rv));
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
LIST_INSERT_HEAD(&object->rvq, rv, objq);
rv->object = object;
rv->pindex = first;
KASSERT(rv->popcnt == 0,
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
("vm_reserv_alloc_contig: reserv %p's popcnt is corrupted",
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv));
KASSERT(!rv->inpartpopq,
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
("vm_reserv_alloc_contig: reserv %p's inpartpopq is TRUE",
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv));
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
for (i = 0; i < NPOPMAP; i++)
KASSERT(rv->popmap[i] == 0,
("vm_reserv_alloc_contig: reserv %p's popmap is corrupted",
rv));
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
n = ulmin(VM_LEVEL_0_NPAGES - index, npages);
for (i = 0; i < n; i++)
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_populate(rv, index + i);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
npages -= n;
if (m_ret == NULL) {
m_ret = &rv->pages[index];
index = 0;
}
m += VM_LEVEL_0_NPAGES;
first += VM_LEVEL_0_NPAGES;
allocpages -= VM_LEVEL_0_NPAGES;
Correct an off-by-one error in vm_reserv_alloc_contig() that resulted in the last reservation of a multi-reservation allocation not being initialized.
2012-07-15 21:46:19 +00:00
} while (allocpages > 0);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
return (m_ret);
/*
* Found a matching reservation.
*/
found:
index = VM_RESERV_INDEX(object, pindex);
/* Does the allocation fit within the reservation? */
if (index + npages > VM_LEVEL_0_NPAGES)
return (NULL);
m = &rv->pages[index];
pa = VM_PAGE_TO_PHYS(m);
if (pa < low || pa + size > high || (pa & (alignment - 1)) != 0 ||
((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0)
return (NULL);
/* Handle vm_page_rename(m, new_object, ...). */
for (i = 0; i < npages; i++)
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
if (isset(rv->popmap, index + i))
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
return (NULL);
for (i = 0; i < npages; i++)
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_populate(rv, index + i);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
return (m);
}
/*
* Allocates a page from an existing or newly-created reservation.
*
Refactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and vm_page_insert() so that (1) vm_radix_lookup_le() is never called while the free page queues lock is held and (2) vm_radix_lookup_le() is called at most once. This change reduces the average time that the free page queues lock is held by vm_page_alloc() as well as vm_page_alloc()'s average overall running time. Sponsored by: EMC / Isilon Storage Division
2013-05-12 16:50:18 +00:00
* The page "mpred" must immediately precede the offset "pindex" within the
* specified object.
*
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
* The object and free page queue must be locked.
*/
vm_page_t
Refactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and vm_page_insert() so that (1) vm_radix_lookup_le() is never called while the free page queues lock is held and (2) vm_radix_lookup_le() is called at most once. This change reduces the average time that the free page queues lock is held by vm_page_alloc() as well as vm_page_alloc()'s average overall running time. Sponsored by: EMC / Isilon Storage Division
2013-05-12 16:50:18 +00:00
vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex, vm_page_t mpred)
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
{
Refactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and vm_page_insert() so that (1) vm_radix_lookup_le() is never called while the free page queues lock is held and (2) vm_radix_lookup_le() is called at most once. This change reduces the average time that the free page queues lock is held by vm_page_alloc() as well as vm_page_alloc()'s average overall running time. Sponsored by: EMC / Isilon Storage Division
2013-05-12 16:50:18 +00:00
vm_page_t m, msucc;
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
vm_pindex_t first, leftcap, rightcap;
vm_reserv_t rv;
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
int i, index;
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(object);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
/*
* Is a reservation fundamentally impossible?
*/
if (pindex < VM_RESERV_INDEX(object, pindex) ||
pindex >= object->size)
return (NULL);
/*
* Look for an existing reservation.
*/
Sync back vmcontention branch into HEAD: Replace the per-object resident and cached pages splay tree with a path-compressed multi-digit radix trie. Along with this, switch also the x86-specific handling of idle page tables to using the radix trie. This change is supposed to do the following: - Allowing the acquisition of read locking for lookup operations of the resident/cached pages collections as the per-vm_page_t splay iterators are now removed. - Increase the scalability of the operations on the page collections. The radix trie does rely on the consumers locking to ensure atomicity of its operations. In order to avoid deadlocks the bisection nodes are pre-allocated in the UMA zone. This can be done safely because the algorithm needs at maximum one new node per insert which means the maximum number of the desired nodes is the number of available physical frames themselves. However, not all the times a new bisection node is really needed. The radix trie implements path-compression because UFS indirect blocks can lead to several objects with a very sparse trie, increasing the number of levels to usually scan. It also helps in the nodes pre-fetching by introducing the single node per-insert property. This code is not generalized (yet) because of the possible loss of performance by having much of the sizes in play configurable. However, efforts to make this code more general and then reusable in further different consumers might be really done. The only KPI change is the removal of the function vm_page_splay() which is now reaped. The only KBI change, instead, is the removal of the left/right iterators from struct vm_page, which are now reaped. Further technical notes broken into mealpieces can be retrieved from the svn branch: http://svn.freebsd.org/base/user/attilio/vmcontention/ Sponsored by: EMC / Isilon storage division In collaboration with: alc, jeff Tested by: flo, pho, jhb, davide Tested by: ian (arm) Tested by: andreast (powerpc)
2013-03-18 00:25:02 +00:00
if (mpred != NULL) {
PG_SLAB no longer serves a useful purpose, since m->object is no longer abused to store pointer to slab. Remove it. Reviewed by: alc Sponsored by: The FreeBSD Foundation Approved by: re (hrs)
2013-09-17 07:35:26 +00:00
KASSERT(mpred->object == object,
Refactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and vm_page_insert() so that (1) vm_radix_lookup_le() is never called while the free page queues lock is held and (2) vm_radix_lookup_le() is called at most once. This change reduces the average time that the free page queues lock is held by vm_page_alloc() as well as vm_page_alloc()'s average overall running time. Sponsored by: EMC / Isilon Storage Division
2013-05-12 16:50:18 +00:00
("vm_reserv_alloc_page: object doesn't contain mpred"));
Sync back vmcontention branch into HEAD: Replace the per-object resident and cached pages splay tree with a path-compressed multi-digit radix trie. Along with this, switch also the x86-specific handling of idle page tables to using the radix trie. This change is supposed to do the following: - Allowing the acquisition of read locking for lookup operations of the resident/cached pages collections as the per-vm_page_t splay iterators are now removed. - Increase the scalability of the operations on the page collections. The radix trie does rely on the consumers locking to ensure atomicity of its operations. In order to avoid deadlocks the bisection nodes are pre-allocated in the UMA zone. This can be done safely because the algorithm needs at maximum one new node per insert which means the maximum number of the desired nodes is the number of available physical frames themselves. However, not all the times a new bisection node is really needed. The radix trie implements path-compression because UFS indirect blocks can lead to several objects with a very sparse trie, increasing the number of levels to usually scan. It also helps in the nodes pre-fetching by introducing the single node per-insert property. This code is not generalized (yet) because of the possible loss of performance by having much of the sizes in play configurable. However, efforts to make this code more general and then reusable in further different consumers might be really done. The only KPI change is the removal of the function vm_page_splay() which is now reaped. The only KBI change, instead, is the removal of the left/right iterators from struct vm_page, which are now reaped. Further technical notes broken into mealpieces can be retrieved from the svn branch: http://svn.freebsd.org/base/user/attilio/vmcontention/ Sponsored by: EMC / Isilon storage division In collaboration with: alc, jeff Tested by: flo, pho, jhb, davide Tested by: ian (arm) Tested by: andreast (powerpc)
2013-03-18 00:25:02 +00:00
KASSERT(mpred->pindex < pindex,
Refactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and vm_page_insert() so that (1) vm_radix_lookup_le() is never called while the free page queues lock is held and (2) vm_radix_lookup_le() is called at most once. This change reduces the average time that the free page queues lock is held by vm_page_alloc() as well as vm_page_alloc()'s average overall running time. Sponsored by: EMC / Isilon Storage Division
2013-05-12 16:50:18 +00:00
("vm_reserv_alloc_page: mpred doesn't precede pindex"));
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
rv = vm_reserv_from_page(mpred);
if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
goto found;
Sync back vmcontention branch into HEAD: Replace the per-object resident and cached pages splay tree with a path-compressed multi-digit radix trie. Along with this, switch also the x86-specific handling of idle page tables to using the radix trie. This change is supposed to do the following: - Allowing the acquisition of read locking for lookup operations of the resident/cached pages collections as the per-vm_page_t splay iterators are now removed. - Increase the scalability of the operations on the page collections. The radix trie does rely on the consumers locking to ensure atomicity of its operations. In order to avoid deadlocks the bisection nodes are pre-allocated in the UMA zone. This can be done safely because the algorithm needs at maximum one new node per insert which means the maximum number of the desired nodes is the number of available physical frames themselves. However, not all the times a new bisection node is really needed. The radix trie implements path-compression because UFS indirect blocks can lead to several objects with a very sparse trie, increasing the number of levels to usually scan. It also helps in the nodes pre-fetching by introducing the single node per-insert property. This code is not generalized (yet) because of the possible loss of performance by having much of the sizes in play configurable. However, efforts to make this code more general and then reusable in further different consumers might be really done. The only KPI change is the removal of the function vm_page_splay() which is now reaped. The only KBI change, instead, is the removal of the left/right iterators from struct vm_page, which are now reaped. Further technical notes broken into mealpieces can be retrieved from the svn branch: http://svn.freebsd.org/base/user/attilio/vmcontention/ Sponsored by: EMC / Isilon storage division In collaboration with: alc, jeff Tested by: flo, pho, jhb, davide Tested by: ian (arm) Tested by: andreast (powerpc)
2013-03-18 00:25:02 +00:00
msucc = TAILQ_NEXT(mpred, listq);
} else
msucc = TAILQ_FIRST(&object->memq);
if (msucc != NULL) {
KASSERT(msucc->pindex > pindex,
Refactor vm_page_alloc()'s interactions with vm_reserv_alloc_page() and vm_page_insert() so that (1) vm_radix_lookup_le() is never called while the free page queues lock is held and (2) vm_radix_lookup_le() is called at most once. This change reduces the average time that the free page queues lock is held by vm_page_alloc() as well as vm_page_alloc()'s average overall running time. Sponsored by: EMC / Isilon Storage Division
2013-05-12 16:50:18 +00:00
("vm_reserv_alloc_page: msucc doesn't succeed pindex"));
Sync back vmcontention branch into HEAD: Replace the per-object resident and cached pages splay tree with a path-compressed multi-digit radix trie. Along with this, switch also the x86-specific handling of idle page tables to using the radix trie. This change is supposed to do the following: - Allowing the acquisition of read locking for lookup operations of the resident/cached pages collections as the per-vm_page_t splay iterators are now removed. - Increase the scalability of the operations on the page collections. The radix trie does rely on the consumers locking to ensure atomicity of its operations. In order to avoid deadlocks the bisection nodes are pre-allocated in the UMA zone. This can be done safely because the algorithm needs at maximum one new node per insert which means the maximum number of the desired nodes is the number of available physical frames themselves. However, not all the times a new bisection node is really needed. The radix trie implements path-compression because UFS indirect blocks can lead to several objects with a very sparse trie, increasing the number of levels to usually scan. It also helps in the nodes pre-fetching by introducing the single node per-insert property. This code is not generalized (yet) because of the possible loss of performance by having much of the sizes in play configurable. However, efforts to make this code more general and then reusable in further different consumers might be really done. The only KPI change is the removal of the function vm_page_splay() which is now reaped. The only KBI change, instead, is the removal of the left/right iterators from struct vm_page, which are now reaped. Further technical notes broken into mealpieces can be retrieved from the svn branch: http://svn.freebsd.org/base/user/attilio/vmcontention/ Sponsored by: EMC / Isilon storage division In collaboration with: alc, jeff Tested by: flo, pho, jhb, davide Tested by: ian (arm) Tested by: andreast (powerpc)
2013-03-18 00:25:02 +00:00
rv = vm_reserv_from_page(msucc);
if (rv->object == object && vm_reserv_has_pindex(rv, pindex))
goto found;
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
}
/*
* Could a reservation fit between the first index to the left that
* can be used and the first index to the right that cannot be used?
*/
first = pindex - VM_RESERV_INDEX(object, pindex);
if (mpred != NULL) {
if ((rv = vm_reserv_from_page(mpred))->object != object)
leftcap = mpred->pindex + 1;
else
leftcap = rv->pindex + VM_LEVEL_0_NPAGES;
if (leftcap > first)
return (NULL);
}
if (msucc != NULL) {
if ((rv = vm_reserv_from_page(msucc))->object != object)
rightcap = msucc->pindex;
else
rightcap = rv->pindex;
if (first + VM_LEVEL_0_NPAGES > rightcap)
return (NULL);
}
/*
* Would a new reservation extend past the end of the object?
*/
if (first + VM_LEVEL_0_NPAGES > object->size) {
/*
* Don't allocate a new reservation if the object is a vnode or
* backed by another object that is a vnode.
*/
if (object->type == OBJT_VNODE ||
(object->backing_object != NULL &&
object->backing_object->type == OBJT_VNODE))
return (NULL);
/* Speculate that the object may grow. */
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
}
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
/*
* Allocate and populate the new reservation.
*/
m = vm_phys_alloc_pages(VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER);
if (m == NULL)
return (NULL);
rv = vm_reserv_from_page(m);
KASSERT(rv->pages == m,
("vm_reserv_alloc_page: reserv %p's pages is corrupted", rv));
KASSERT(rv->object == NULL,
("vm_reserv_alloc_page: reserv %p isn't free", rv));
LIST_INSERT_HEAD(&object->rvq, rv, objq);
rv->object = object;
rv->pindex = first;
KASSERT(rv->popcnt == 0,
("vm_reserv_alloc_page: reserv %p's popcnt is corrupted", rv));
KASSERT(!rv->inpartpopq,
("vm_reserv_alloc_page: reserv %p's inpartpopq is TRUE", rv));
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
for (i = 0; i < NPOPMAP; i++)
KASSERT(rv->popmap[i] == 0,
("vm_reserv_alloc_page: reserv %p's popmap is corrupted",
rv));
index = VM_RESERV_INDEX(object, pindex);
vm_reserv_populate(rv, index);
return (&rv->pages[index]);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
/*
* Found a matching reservation.
*/
found:
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
index = VM_RESERV_INDEX(object, pindex);
m = &rv->pages[index];
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
/* Handle vm_page_rename(m, new_object, ...). */
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
if (isset(rv->popmap, index))
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
return (NULL);
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_populate(rv, index);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
return (m);
}
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
/*
* Breaks the given reservation. Except for the specified cached or free
* page, all cached and free pages in the reservation are returned to the
* physical memory allocator. The reservation's population count and map are
* reset to their initial state.
*
* The given reservation must not be in the partially-populated reservation
* queue. The free page queue lock must be held.
*/
static void
vm_reserv_break(vm_reserv_t rv, vm_page_t m)
{
int begin_zeroes, hi, i, lo;
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
KASSERT(rv->object != NULL,
("vm_reserv_break: reserv %p is free", rv));
KASSERT(!rv->inpartpopq,
("vm_reserv_break: reserv %p's inpartpopq is TRUE", rv));
LIST_REMOVE(rv, objq);
rv->object = NULL;
if (m != NULL) {
/*
* Since the reservation is being broken, there is no harm in
* abusing the population map to stop "m" from being returned
* to the physical memory allocator.
*/
i = m - rv->pages;
KASSERT(isclr(rv->popmap, i),
("vm_reserv_break: reserv %p's popmap is corrupted", rv));
setbit(rv->popmap, i);
rv->popcnt++;
}
i = hi = 0;
do {
/* Find the next 0 bit. Any previous 0 bits are < "hi". */
lo = ffsl(~(((1UL << hi) - 1) | rv->popmap[i]));
if (lo == 0) {
/* Redundantly clears bits < "hi". */
rv->popmap[i] = 0;
rv->popcnt -= NBPOPMAP - hi;
while (++i < NPOPMAP) {
lo = ffsl(~rv->popmap[i]);
if (lo == 0) {
rv->popmap[i] = 0;
rv->popcnt -= NBPOPMAP;
} else
break;
}
if (i == NPOPMAP)
break;
hi = 0;
}
KASSERT(lo > 0, ("vm_reserv_break: lo is %d", lo));
/* Convert from ffsl() to ordinary bit numbering. */
lo--;
if (lo > 0) {
/* Redundantly clears bits < "hi". */
rv->popmap[i] &= ~((1UL << lo) - 1);
rv->popcnt -= lo - hi;
}
begin_zeroes = NBPOPMAP * i + lo;
/* Find the next 1 bit. */
do
hi = ffsl(rv->popmap[i]);
while (hi == 0 && ++i < NPOPMAP);
if (i != NPOPMAP)
/* Convert from ffsl() to ordinary bit numbering. */
hi--;
vm_phys_free_contig(&rv->pages[begin_zeroes], NBPOPMAP * i +
hi - begin_zeroes);
} while (i < NPOPMAP);
KASSERT(rv->popcnt == 0,
("vm_reserv_break: reserv %p's popcnt is corrupted", rv));
vm_reserv_broken++;
}
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
* Breaks all reservations belonging to the given object.
*/
void
vm_reserv_break_all(vm_object_t object)
{
vm_reserv_t rv;
mtx_lock(&vm_page_queue_free_mtx);
while ((rv = LIST_FIRST(&object->rvq)) != NULL) {
KASSERT(rv->object == object,
("vm_reserv_break_all: reserv %p is corrupted", rv));
if (rv->inpartpopq) {
TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
rv->inpartpopq = FALSE;
}
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_break(rv, NULL);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
}
mtx_unlock(&vm_page_queue_free_mtx);
}
/*
* Frees the given page if it belongs to a reservation. Returns TRUE if the
* page is freed and FALSE otherwise.
*
* The free page queue lock must be held.
*/
boolean_t
vm_reserv_free_page(vm_page_t m)
{
vm_reserv_t rv;
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
rv = vm_reserv_from_page(m);
If a page belonging a reservation is cached, then mark the reservation so that it will be freed to the cache pool rather than the default pool. Otherwise, the cached pages within the reservation may be recycled sooner than necessary. Reported by: Andrey Zonov
2012-04-08 17:00:46 +00:00
if (rv->object == NULL)
return (FALSE);
if ((m->flags & PG_CACHED) != 0 && m->pool != VM_FREEPOOL_CACHE)
vm_phys_set_pool(VM_FREEPOOL_CACHE, rv->pages,
VM_LEVEL_0_ORDER);
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_depopulate(rv, m - rv->pages);
If a page belonging a reservation is cached, then mark the reservation so that it will be freed to the cache pool rather than the default pool. Otherwise, the cached pages within the reservation may be recycled sooner than necessary. Reported by: Andrey Zonov
2012-04-08 17:00:46 +00:00
return (TRUE);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
}
/*
* Initializes the reservation management system. Specifically, initializes
* the reservation array.
*
* Requires that vm_page_array and first_page are initialized!
*/
void
vm_reserv_init(void)
{
vm_paddr_t paddr;
int i;
/*
* Initialize the reservation array. Specifically, initialize the
* "pages" field for every element that has an underlying superpage.
*/
for (i = 0; phys_avail[i + 1] != 0; i += 2) {
paddr = roundup2(phys_avail[i], VM_LEVEL_0_SIZE);
while (paddr + VM_LEVEL_0_SIZE <= phys_avail[i + 1]) {
vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT].pages =
PHYS_TO_VM_PAGE(paddr);
paddr += VM_LEVEL_0_SIZE;
}
}
}
/*
* Returns a reservation level if the given page belongs to a fully-populated
* reservation and -1 otherwise.
*/
int
vm_reserv_level_iffullpop(vm_page_t m)
{
vm_reserv_t rv;
rv = vm_reserv_from_page(m);
return (rv->popcnt == VM_LEVEL_0_NPAGES ? 0 : -1);
}
/*
* Prepare for the reactivation of a cached page.
*
* First, suppose that the given page "m" was allocated individually, i.e., not
* as part of a reservation, and cached. Then, suppose a reservation
* containing "m" is allocated by the same object. Although "m" and the
* reservation belong to the same object, "m"'s pindex may not match the
* reservation's.
*
* The free page queue must be locked.
*/
boolean_t
vm_reserv_reactivate_page(vm_page_t m)
{
vm_reserv_t rv;
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
int index;
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
rv = vm_reserv_from_page(m);
if (rv->object == NULL)
return (FALSE);
KASSERT((m->flags & PG_CACHED) != 0,
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
("vm_reserv_reactivate_page: page %p is not cached", m));
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
if (m->object == rv->object &&
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
m->pindex - rv->pindex == (index = VM_RESERV_INDEX(m->object,
m->pindex)))
vm_reserv_populate(rv, index);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
else {
KASSERT(rv->inpartpopq,
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
("vm_reserv_reactivate_page: reserv %p's inpartpopq is FALSE",
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv));
TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
rv->inpartpopq = FALSE;
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
/* Don't release "m" to the physical memory allocator. */
vm_reserv_break(rv, m);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
}
return (TRUE);
}
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
/*
* Breaks the given partially-populated reservation, releasing its cached and
* free pages to the physical memory allocator.
*
* The free page queue lock must be held.
*/
static void
vm_reserv_reclaim(vm_reserv_t rv)
{
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
KASSERT(rv->inpartpopq,
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
("vm_reserv_reclaim: reserv %p's inpartpopq is FALSE", rv));
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
TAILQ_REMOVE(&vm_rvq_partpop, rv, partpopq);
rv->inpartpopq = FALSE;
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_reserv_break(rv, NULL);
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
vm_reserv_reclaimed++;
}
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
* Breaks the reservation at the head of the partially-populated reservation
* queue, releasing its cached and free pages to the physical memory
* allocator. Returns TRUE if a reservation is broken and FALSE otherwise.
*
* The free page queue lock must be held.
*/
boolean_t
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
vm_reserv_reclaim_inactive(void)
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
{
vm_reserv_t rv;
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
if ((rv = TAILQ_FIRST(&vm_rvq_partpop)) != NULL) {
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
vm_reserv_reclaim(rv);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
return (TRUE);
}
return (FALSE);
}
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
/*
* Searches the partially-populated reservation queue for the least recently
* active reservation with unused pages, i.e., cached or free, that satisfy the
* given request for contiguous physical memory. If a satisfactory reservation
* is found, it is broken. Returns TRUE if a reservation is broken and FALSE
* otherwise.
*
* The free page queue lock must be held.
*/
boolean_t
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
vm_reserv_reclaim_contig(u_long npages, vm_paddr_t low, vm_paddr_t high,
Eliminate vm_phys_bootstrap_alloc(). It was a failed attempt at eliminating duplicated code in the various pmap implementations. Micro-optimize vm_phys_free_pages(). Introduce vm_phys_free_contig(). It is fast routine for freeing an arbitrary number of physically contiguous pages. In particular, it doesn't require the number of pages to be a power of two. Use "u_long" instead of "unsigned long". Bruce Evans (bde@) has convinced me that the "boundary" parameters to kmem_alloc_contig(), vm_phys_alloc_contig(), and vm_reserv_reclaim_contig() should be of type "vm_paddr_t" and not "u_long". Make this change.
2011-10-30 05:06:14 +00:00
u_long alignment, vm_paddr_t boundary)
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
{
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
vm_paddr_t pa, size;
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
vm_reserv_t rv;
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
int hi, i, lo, next_free;
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
if (npages > VM_LEVEL_0_NPAGES - 1)
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
return (FALSE);
Introduce vm_reserv_alloc_contig() and teach vm_page_alloc_contig() how to use superpage reservations. So, for the first time, kernel virtual memory that is allocated by contigmalloc(), kmem_alloc_attr(), and kmem_alloc_contig() can be promoted to superpages. In fact, even a series of small contigmalloc() allocations may collectively result in a promoted superpage. Eliminate some duplication of code in vm_reserv_alloc_page(). Change the type of vm_reserv_reclaim_contig()'s first parameter in order that it be consistent with other vm_*_contig() functions. Tested by: marius (sparc64)
2011-12-05 18:29:25 +00:00
size = npages << PAGE_SHIFT;
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
TAILQ_FOREACH(rv, &vm_rvq_partpop, partpopq) {
pa = VM_PAGE_TO_PHYS(&rv->pages[VM_LEVEL_0_NPAGES - 1]);
if (pa + PAGE_SIZE - size < low) {
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
/* This entire reservation is too low; go to next. */
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
continue;
}
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
pa = VM_PAGE_TO_PHYS(&rv->pages[0]);
if (pa + size > high) {
/* This entire reservation is too high; go to next. */
continue;
}
if (pa < low) {
/* Start the search for free pages at "low". */
i = (low - pa) / NBPOPMAP;
hi = (low - pa) % NBPOPMAP;
} else
i = hi = 0;
do {
/* Find the next free page. */
lo = ffsl(~(((1UL << hi) - 1) | rv->popmap[i]));
while (lo == 0 && ++i < NPOPMAP)
lo = ffsl(~rv->popmap[i]);
if (i == NPOPMAP)
break;
/* Convert from ffsl() to ordinary bit numbering. */
lo--;
next_free = NBPOPMAP * i + lo;
pa = VM_PAGE_TO_PHYS(&rv->pages[next_free]);
KASSERT(pa >= low,
("vm_reserv_reclaim_contig: pa is too low"));
if (pa + size > high) {
/* The rest of this reservation is too high. */
break;
} else if ((pa & (alignment - 1)) != 0 ||
((pa ^ (pa + size - 1)) & ~(boundary - 1)) != 0) {
/* Continue with this reservation. */
hi = lo;
continue;
}
/* Find the next used page. */
hi = ffsl(rv->popmap[i] & ~((1UL << lo) - 1));
while (hi == 0 && ++i < NPOPMAP) {
if ((NBPOPMAP * i - next_free) * PAGE_SIZE >=
size) {
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
vm_reserv_reclaim(rv);
return (TRUE);
}
MFp4 alc_popmap Change the way that reservations keep track of which pages are in use. Instead of using the page's PG_CACHED and PG_FREE flags, maintain a bit vector within the reservation. This approach has a couple benefits. First, it makes breaking reservations much cheaper because there are fewer cache misses to identify the unused pages. Second, it is a pre- requisite for supporting two or more reservation sizes.
2013-12-28 04:28:35 +00:00
hi = ffsl(rv->popmap[i]);
}
/* Convert from ffsl() to ordinary bit numbering. */
if (i != NPOPMAP)
hi--;
if ((NBPOPMAP * i + hi - next_free) * PAGE_SIZE >=
size) {
vm_reserv_reclaim(rv);
return (TRUE);
}
} while (i < NPOPMAP);
Introduce vm_reserv_reclaim_contig(). This function is used by contigmalloc(9) as a last resort to steal pages from an inactive, partially-used superpage reservation. Rename vm_reserv_reclaim() to vm_reserv_reclaim_inactive() and refactor it so that a separate subroutine is responsible for breaking the selected reservation. This subroutine is also used by vm_reserv_reclaim_contig().
2008-04-06 18:09:28 +00:00
}
return (FALSE);
}
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
/*
* Transfers the reservation underlying the given page to a new object.
*
* The object must be locked.
*/
void
vm_reserv_rename(vm_page_t m, vm_object_t new_object, vm_object_t old_object,
vm_pindex_t old_object_offset)
{
vm_reserv_t rv;
Switch the vm_object mutex to be a rwlock. This will enable in the future further optimizations where the vm_object lock will be held in read mode most of the time the page cache resident pool of pages are accessed for reading purposes. The change is mostly mechanical but few notes are reported: * The KPI changes as follow: - VM_OBJECT_LOCK() -> VM_OBJECT_WLOCK() - VM_OBJECT_TRYLOCK() -> VM_OBJECT_TRYWLOCK() - VM_OBJECT_UNLOCK() -> VM_OBJECT_WUNLOCK() - VM_OBJECT_LOCK_ASSERT(MA_OWNED) -> VM_OBJECT_ASSERT_WLOCKED() (in order to avoid visibility of implementation details) - The read-mode operations are added: VM_OBJECT_RLOCK(), VM_OBJECT_TRYRLOCK(), VM_OBJECT_RUNLOCK(), VM_OBJECT_ASSERT_RLOCKED(), VM_OBJECT_ASSERT_LOCKED() * The vm/vm_pager.h namespace pollution avoidance (forcing requiring sys/mutex.h in consumers directly to cater its inlining functions using VM_OBJECT_LOCK()) imposes that all the vm/vm_pager.h consumers now must include also sys/rwlock.h. * zfs requires a quite convoluted fix to include FreeBSD rwlocks into the compat layer because the name clash between FreeBSD and solaris versions must be avoided. At this purpose zfs redefines the vm_object locking functions directly, isolating the FreeBSD components in specific compat stubs. The KPI results heavilly broken by this commit. Thirdy part ports must be updated accordingly (I can think off-hand of VirtualBox, for example). Sponsored by: EMC / Isilon storage division Reviewed by: jeff Reviewed by: pjd (ZFS specific review) Discussed with: alc Tested by: pho
2013-03-09 02:32:23 +00:00
VM_OBJECT_ASSERT_WLOCKED(new_object);
Add the superpage reservation system. This is "part 2 of 2" of the machine-independent support for superpages. (The earlier part was the rewrite of the physical memory allocator.) The remainder of the code required for superpages support is machine-dependent and will be added to the various pmap implementations at a later date. Initially, I am only supporting one large page size per architecture. Moreover, I am only enabling the reservation system on amd64. (In an emergency, it can be disabled by setting VM_NRESERVLEVELS to 0 in amd64/include/vmparam.h or your kernel configuration file.)
2007-12-29 19:53:04 +00:00
rv = vm_reserv_from_page(m);
if (rv->object == old_object) {
mtx_lock(&vm_page_queue_free_mtx);
if (rv->object == old_object) {
LIST_REMOVE(rv, objq);
LIST_INSERT_HEAD(&new_object->rvq, rv, objq);
rv->object = new_object;
rv->pindex -= old_object_offset;
}
mtx_unlock(&vm_page_queue_free_mtx);
}
}
/*
* Allocates the virtual and physical memory required by the reservation
* management system's data structures, in particular, the reservation array.
*/
vm_paddr_t
vm_reserv_startup(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t high_water)
{
vm_paddr_t new_end;
size_t size;
/*
* Calculate the size (in bytes) of the reservation array. Round up
* from "high_water" because every small page is mapped to an element
* in the reservation array based on its physical address. Thus, the
* number of elements in the reservation array can be greater than the
* number of superpages.
*/
size = howmany(high_water, VM_LEVEL_0_SIZE) * sizeof(struct vm_reserv);
/*
* Allocate and map the physical memory for the reservation array. The
* next available virtual address is returned by reference.
*/
new_end = end - round_page(size);
vm_reserv_array = (void *)(uintptr_t)pmap_map(vaddr, new_end, end,
VM_PROT_READ | VM_PROT_WRITE);
bzero(vm_reserv_array, size);
/*
* Return the next available physical address.
*/
return (new_end);
}
#endif /* VM_NRESERVLEVEL > 0 */
Reference in New Issue Copy Permalink