d/freebsd-skq
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
84f210952c
freebsd-skq/sys/vm/vm_reserv.c

1160 lines
34 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
/*-
sys: general adoption of SPDX licensing ID tags. Mainly focus on files that use BSD 2-Clause license, however the tool I was using misidentified many licenses so this was mostly a manual - error prone - task. The Software Package Data Exchange (SPDX) group provides a specification to make it easier for automated tools to detect and summarize well known opensource licenses. We are gradually adopting the specification, noting that the tags are considered only advisory and do not, in any way, superceed or replace the license texts. No functional change intended.
2017-11-27 15:23:17 +00:00
* SPDX-License-Identifier: BSD-2-Clause-FreeBSD
*
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>
All these files need sys/vmmeter.h, but now they got it implicitly included via sys/pcpu.h.
2017-04-17 17:07:00 +00:00
#include <sys/vmmeter.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.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
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* pages ("superpages"). Speculative allocation enables the fully automatic
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
* 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)
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
/*
* Clear a bit in the population map.
*/
static __inline void
popmap_clear(popmap_t popmap[], int i)
{
popmap[i / NBPOPMAP] &= ~(1UL << (i % NBPOPMAP));
}
/*
* Set a bit in the population map.
*/
static __inline void
popmap_set(popmap_t popmap[], int i)
{
popmap[i / NBPOPMAP] |= 1UL << (i % NBPOPMAP);
}
/*
* Is a bit in the population map clear?
*/
static __inline boolean_t
popmap_is_clear(popmap_t popmap[], int i)
{
return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) == 0);
}
/*
* Is a bit in the population map set?
*/
static __inline boolean_t
popmap_is_set(popmap_t popmap[], int i)
{
return ((popmap[i / NBPOPMAP] & (1UL << (i % NBPOPMAP))) != 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
/*
* 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
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* reservation is not fully utilized, it appears in the queue of partially
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
* populated reservations. The reservation always appears on the containing
* object's list of reservations.
*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* A partially populated reservation can be broken and reclaimed at any time.
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
*/
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 */
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
int domain; /* NUMA domain */
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
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;
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* The partially populated reservation queue
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
*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* This queue enables the fast recovery of an unused free small page 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.
*/
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
static TAILQ_HEAD(, vm_reserv) vm_rvq_partpop[MAXMEMDOM];
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 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");
Introduce a sysctl for reporting the number of fully populated reservations.
2015-08-06 21:27:50 +00:00
static int sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS);
SYSCTL_PROC(_vm_reserv, OID_AUTO, fullpop, CTLTYPE_INT | CTLFLAG_RD, NULL, 0,
sysctl_vm_reserv_fullpop, "I", "Current number of full reservations");
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 int sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS);
SYSCTL_OID(_vm_reserv, OID_AUTO, partpopq, CTLTYPE_STRING | CTLFLAG_RD, NULL, 0,
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
sysctl_vm_reserv_partpopq, "A", "Partially populated reservation queues");
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 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
Introduce a sysctl for reporting the number of fully populated reservations.
2015-08-06 21:27:50 +00:00
/*
* Returns the current number of full reservations.
*
* Since the number of full reservations is computed without acquiring the
* free page queue lock, the returned value may be inexact.
*/
static int
sysctl_vm_reserv_fullpop(SYSCTL_HANDLER_ARGS)
{
vm_paddr_t paddr;
struct vm_phys_seg *seg;
vm_reserv_t rv;
int fullpop, segind;
fullpop = 0;
for (segind = 0; segind < vm_phys_nsegs; segind++) {
seg = &vm_phys_segs[segind];
paddr = roundup2(seg->start, VM_LEVEL_0_SIZE);
while (paddr + VM_LEVEL_0_SIZE <= seg->end) {
rv = &vm_reserv_array[paddr >> VM_LEVEL_0_SHIFT];
fullpop += rv->popcnt == VM_LEVEL_0_NPAGES;
paddr += VM_LEVEL_0_SIZE;
}
}
return (sysctl_handle_int(oidp, &fullpop, 0, 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
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* Describes the current state of the partially populated reservation queue.
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 int
sysctl_vm_reserv_partpopq(SYSCTL_HANDLER_ARGS)
{
struct sbuf sbuf;
vm_reserv_t rv;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
int counter, error, domain, level, unused_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
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);
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
sbuf_printf(&sbuf, "\nDOMAIN LEVEL SIZE NUMBER\n\n");
for (domain = 0; domain < vm_ndomains; domain++) {
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[domain], partpopq) {
counter++;
unused_pages += VM_LEVEL_0_NPAGES - rv->popcnt;
}
mtx_unlock(&vm_page_queue_free_mtx);
sbuf_printf(&sbuf, "%6d, %7d, %6dK, %6d\n",
domain, level,
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
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +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));
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
KASSERT(popmap_is_set(rv->popmap, index),
Add "popmap" assertions: The page being freed isn't already free, and the page being allocated isn't already allocated. Sponsored by: EMC / Isilon Storage Division
2013-12-29 04:54:52 +00:00
("vm_reserv_depopulate: reserv %p's popmap[%d] is clear", 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
KASSERT(rv->popcnt > 0,
("vm_reserv_depopulate: reserv %p's popcnt is corrupted", rv));
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
("vm_reserv_depopulate: reserv %p's domain is corrupted %d",
rv, rv->domain));
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 (rv->inpartpopq) {
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], 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
rv->inpartpopq = FALSE;
Add a page size field to struct vm_page. Increase the page size field when a partially populated reservation becomes fully populated, and decrease this field when a fully populated reservation becomes partially populated. Use this field to simplify the implementation of pmap_enter_object() on amd64, arm, and i386. On all architectures where we support superpages, the cost of creating a superpage mapping is roughly the same as creating a base page mapping. For example, both kinds of mappings entail the creation of a single PTE and PV entry. With this in mind, use the page size field to make the implementation of vm_map_pmap_enter(..., MAP_PREFAULT_PARTIAL) a little smarter. Previously, if MAP_PREFAULT_PARTIAL was specified to vm_map_pmap_enter(), that function would only map base pages. Now, it will create up to 96 base page or superpage mappings. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2014-06-07 17:12:26 +00:00
} else {
KASSERT(rv->pages->psind == 1,
("vm_reserv_depopulate: reserv %p is already demoted",
rv));
rv->pages->psind = 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
}
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
popmap_clear(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;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
rv->domain = -1;
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_phys_free_pages(rv->pages, VM_LEVEL_0_ORDER);
vm_reserv_freed++;
} else {
rv->inpartpopq = TRUE;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
TAILQ_INSERT_TAIL(&vm_rvq_partpop[rv->domain], 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
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* to the tail of the partially populated reservation queue.
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 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));
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
KASSERT(popmap_is_clear(rv->popmap, index),
Add "popmap" assertions: The page being freed isn't already free, and the page being allocated isn't already allocated. Sponsored by: EMC / Isilon Storage Division
2013-12-29 04:54:52 +00:00
("vm_reserv_populate: reserv %p's popmap[%d] is set", 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
KASSERT(rv->popcnt < VM_LEVEL_0_NPAGES,
("vm_reserv_populate: reserv %p is already full", rv));
Add a page size field to struct vm_page. Increase the page size field when a partially populated reservation becomes fully populated, and decrease this field when a fully populated reservation becomes partially populated. Use this field to simplify the implementation of pmap_enter_object() on amd64, arm, and i386. On all architectures where we support superpages, the cost of creating a superpage mapping is roughly the same as creating a base page mapping. For example, both kinds of mappings entail the creation of a single PTE and PV entry. With this in mind, use the page size field to make the implementation of vm_map_pmap_enter(..., MAP_PREFAULT_PARTIAL) a little smarter. Previously, if MAP_PREFAULT_PARTIAL was specified to vm_map_pmap_enter(), that function would only map base pages. Now, it will create up to 96 base page or superpage mappings. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2014-06-07 17:12:26 +00:00
KASSERT(rv->pages->psind == 0,
("vm_reserv_populate: reserv %p is already promoted", rv));
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
("vm_reserv_populate: reserv %p's domain is corrupted %d",
rv, rv->domain));
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 (rv->inpartpopq) {
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], 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
rv->inpartpopq = FALSE;
}
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
popmap_set(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;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
TAILQ_INSERT_TAIL(&vm_rvq_partpop[rv->domain], rv, partpopq);
Add a page size field to struct vm_page. Increase the page size field when a partially populated reservation becomes fully populated, and decrease this field when a fully populated reservation becomes partially populated. Use this field to simplify the implementation of pmap_enter_object() on amd64, arm, and i386. On all architectures where we support superpages, the cost of creating a superpage mapping is roughly the same as creating a base page mapping. For example, both kinds of mappings entail the creation of a single PTE and PV entry. With this in mind, use the page size field to make the implementation of vm_map_pmap_enter(..., MAP_PREFAULT_PARTIAL) a little smarter. Previously, if MAP_PREFAULT_PARTIAL was specified to vm_map_pmap_enter(), that function would only map base pages. Now, it will create up to 96 base page or superpage mappings. Reviewed by: kib Sponsored by: EMC / Isilon Storage Division
2014-06-07 17:12:26 +00:00
} else
rv->pages->psind = 1;
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
* Allocates a contiguous set of physical pages of the given size "npages"
Fix a boundary case error in vm_reserv_alloc_contig(): If a reservation isn't being allocated for the last of the requested pages, because a reservation won't fit in the gap between allocated pages, then the reservation structure shouldn't be initialized. While I'm here, improve the nearby comments. Reported by: jeff, pho MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2014-09-10 05:52:30 +00:00
* from existing or newly created reservations. All of the physical pages
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
* 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
*
Relax the object type restrictions on vm_page_alloc_contig(). Specifically, add support for object types that were previously prohibited because they could contain PG_CACHED pages. Roughly halve the number of radix trie operations performed by vm_page_alloc_contig() using the same approach that is employed by vm_page_alloc(). Also, eliminate the radix trie lookup performed with the free page queues lock held. Tidy up the handling of radix trie insert failures in vm_page_alloc() and vm_page_alloc_contig(). Reviewed by: kib, markj Tested by: pho Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8878
2016-12-28 18:32:13 +00:00
* The page "mpred" must immediately precede the offset "pindex" within the
* specified 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
* The object and free page queue must be locked.
*/
vm_page_t
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
vm_reserv_alloc_contig(vm_object_t object, vm_pindex_t pindex, int domain,
u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
vm_paddr_t boundary, vm_page_t mpred)
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;
Relax the object type restrictions on vm_page_alloc_contig(). Specifically, add support for object types that were previously prohibited because they could contain PG_CACHED pages. Roughly halve the number of radix trie operations performed by vm_page_alloc_contig() using the same approach that is employed by vm_page_alloc(). Also, eliminate the radix trie lookup performed with the free page queues lock held. Tidy up the handling of radix trie insert failures in vm_page_alloc() and vm_page_alloc_contig(). Reviewed by: kib, markj Tested by: pho Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8878
2016-12-28 18:32:13 +00:00
vm_page_t m, m_ret, 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
if (mpred != NULL) {
Relax the object type restrictions on vm_page_alloc_contig(). Specifically, add support for object types that were previously prohibited because they could contain PG_CACHED pages. Roughly halve the number of radix trie operations performed by vm_page_alloc_contig() using the same approach that is employed by vm_page_alloc(). Also, eliminate the radix trie lookup performed with the free page queues lock held. Tidy up the handling of radix trie insert failures in vm_page_alloc() and vm_page_alloc_contig(). Reviewed by: kib, markj Tested by: pho Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8878
2016-12-28 18:32:13 +00:00
KASSERT(mpred->object == object,
("vm_reserv_alloc_contig: 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,
Relax the object type restrictions on vm_page_alloc_contig(). Specifically, add support for object types that were previously prohibited because they could contain PG_CACHED pages. Roughly halve the number of radix trie operations performed by vm_page_alloc_contig() using the same approach that is employed by vm_page_alloc(). Also, eliminate the radix trie lookup performed with the free page queues lock held. Tidy up the handling of radix trie insert failures in vm_page_alloc() and vm_page_alloc_contig(). Reviewed by: kib, markj Tested by: pho Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8878
2016-12-28 18:32:13 +00:00
("vm_reserv_alloc_contig: mpred doesn't precede pindex"));
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,
Relax the object type restrictions on vm_page_alloc_contig(). Specifically, add support for object types that were previously prohibited because they could contain PG_CACHED pages. Roughly halve the number of radix trie operations performed by vm_page_alloc_contig() using the same approach that is employed by vm_page_alloc(). Also, eliminate the radix trie lookup performed with the free page queues lock held. Tidy up the handling of radix trie insert failures in vm_page_alloc() and vm_page_alloc_contig(). Reviewed by: kib, markj Tested by: pho Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8878
2016-12-28 18:32:13 +00:00
("vm_reserv_alloc_contig: 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;
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
Fix a boundary case error in vm_reserv_alloc_contig(): If a reservation isn't being allocated for the last of the requested pages, because a reservation won't fit in the gap between allocated pages, then the reservation structure shouldn't be initialized. While I'm here, improve the nearby comments. Reported by: jeff, pho MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2014-09-10 05:52:30 +00:00
* left that can be used ("leftcap") and the first index to the right
* that cannot be used ("rightcap")?
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);
Fix a boundary case error in vm_reserv_alloc_contig(): If a reservation isn't being allocated for the last of the requested pages, because a reservation won't fit in the gap between allocated pages, then the reservation structure shouldn't be initialized. While I'm here, improve the nearby comments. Reported by: jeff, pho MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2014-09-10 05:52:30 +00:00
/*
* At least one reservation will fit between "leftcap"
* and "rightcap". However, a reservation for the
* last of the requested pages will not fit. Reduce
* the size of the upcoming allocation accordingly.
*/
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
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. */
}
/*
Fix a boundary case error in vm_reserv_alloc_contig(): If a reservation isn't being allocated for the last of the requested pages, because a reservation won't fit in the gap between allocated pages, then the reservation structure shouldn't be initialized. While I'm here, improve the nearby comments. Reported by: jeff, pho MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2014-09-10 05:52:30 +00:00
* Allocate the physical pages. 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
*/
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
m = vm_phys_alloc_contig(domain, allocpages, low, high, ulmax(alignment,
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_LEVEL_0_SIZE), boundary > VM_LEVEL_0_SIZE ? boundary : 0);
if (m == NULL)
return (NULL);
Fix a boundary case error in vm_reserv_alloc_contig(): If a reservation isn't being allocated for the last of the requested pages, because a reservation won't fit in the gap between allocated pages, then the reservation structure shouldn't be initialized. While I'm here, improve the nearby comments. Reported by: jeff, pho MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2014-09-10 05:52:30 +00:00
/*
* The allocated physical pages always begin at a reservation
* boundary, but they do not always end at a reservation boundary.
* Initialize every reservation that is completely covered by the
* allocated physical pages.
*/
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_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;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
rv->domain = vm_phys_domidx(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
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;
Fix a boundary case error in vm_reserv_alloc_contig(): If a reservation isn't being allocated for the last of the requested pages, because a reservation won't fit in the gap between allocated pages, then the reservation structure shouldn't be initialized. While I'm here, improve the nearby comments. Reported by: jeff, pho MFC after: 1 week Sponsored by: EMC / Isilon Storage Division
2014-09-10 05:52:30 +00:00
} while (allocpages >= VM_LEVEL_0_NPAGES);
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++)
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
if (popmap_is_set(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);
}
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* Allocates a page from an existing or newly created reservation.
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
* 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
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
vm_reserv_alloc_page(vm_object_t object, vm_pindex_t pindex, int domain,
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.
*/
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
m = vm_phys_alloc_pages(domain, VM_FREEPOOL_DEFAULT, VM_LEVEL_0_ORDER);
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 (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;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
rv->domain = vm_phys_domidx(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
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, ...). */
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
if (popmap_is_set(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
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* Breaks the given reservation. Except for the specified free page, all 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.
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
*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* The given reservation must not be in the partially populated reservation
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
* 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;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
rv->domain = -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
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;
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
KASSERT(popmap_is_clear(rv->popmap, 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_break: reserv %p's popmap is corrupted", rv));
Correct a bug in the management of the population map on big-endian machines. Specifically, there was a mismatch between how the routine allocation and deallocation operations accessed the population map and how the aggressively optimized reservation-breaking operation accessed it. So, problems only occurred when reservations were broken. This change makes the routine operations access the population map in the same way as the reservation breaking operation. This bug was introduced in r259999. PR: 187080 Tested by: jmg (on an "armeb" machine) Sponsored by: EMC / Isilon Storage Division
2014-06-11 16:11:12 +00:00
popmap_set(rv->popmap, 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
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) {
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], 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
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);
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;
By the time that vm_reserv_init() runs, vm_phys_segs[] is initialized. Use it instead of phys_avail[]. Discussed with: Svatopluk Kraus
2014-11-22 17:46:30 +00:00
struct vm_phys_seg *seg;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
int i, segind;
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
/*
* Initialize the reservation array. Specifically, initialize the
* "pages" field for every element that has an underlying superpage.
*/
By the time that vm_reserv_init() runs, vm_phys_segs[] is initialized. Use it instead of phys_avail[]. Discussed with: Svatopluk Kraus
2014-11-22 17:46:30 +00:00
for (segind = 0; segind < vm_phys_nsegs; segind++) {
seg = &vm_phys_segs[segind];
paddr = roundup2(seg->start, VM_LEVEL_0_SIZE);
while (paddr + VM_LEVEL_0_SIZE <= seg->end) {
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_array[paddr >> VM_LEVEL_0_SHIFT].pages =
PHYS_TO_VM_PAGE(paddr);
paddr += VM_LEVEL_0_SIZE;
}
}
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
for (i = 0; i < MAXMEMDOM; i++)
TAILQ_INIT(&vm_rvq_partpop[i]);
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 a new mechanism for relocating virtual pages to a new physical address and use this mechanism when: 1. kmem_alloc_{attr,contig}() can't find suitable free pages in the physical memory allocator's free page lists. This replaces the long-standing approach of scanning the inactive and inactive queues, converting clean pages into PG_CACHED pages and laundering dirty pages. In contrast, the new mechanism does not use PG_CACHED pages nor does it trigger a large number of I/O operations. 2. on 32-bit MIPS processors, uma_small_alloc() and the pmap can't find free pages in the physical memory allocator's free page lists that are covered by the direct map. Tested by: adrian 3. ttm_bo_global_init() and ttm_vm_page_alloc_dma32() can't find suitable free pages in the physical memory allocator's free page lists. In the coming months, I expect that this new mechanism will be applied in other places. For example, balloon drivers should use relocation to minimize fragmentation of the guest physical address space. Make vm_phys_alloc_contig() a little smarter (and more efficient in some cases). Specifically, use vm_phys_segs[] earlier to avoid scanning free page lists that can't possibly contain suitable pages. Reviewed by: kib, markj Glanced at: jhb Discussed with: jeff Sponsored by: EMC / Isilon Storage Division Differential Revision: https://reviews.freebsd.org/D4444
2015-12-19 18:42:50 +00:00
/*
* Returns true if the given page belongs to a reservation and that page is
* free. Otherwise, returns false.
*/
bool
vm_reserv_is_page_free(vm_page_t m)
{
vm_reserv_t rv;
mtx_assert(&vm_page_queue_free_mtx, MA_OWNED);
rv = vm_reserv_from_page(m);
if (rv->object == NULL)
return (false);
return (popmap_is_clear(rv->popmap, m - rv->pages));
}
/*
* If the given page belongs to a reservation, returns the level of that
* reservation. Otherwise, returns -1.
*/
int
vm_reserv_level(vm_page_t m)
{
vm_reserv_t rv;
rv = vm_reserv_from_page(m);
return (rv->object != NULL ? 0 : -1);
}
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
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* Returns a reservation level if the given page belongs to a fully populated
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
* 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);
}
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
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* Breaks the given partially populated reservation, releasing its free pages
* to the physical memory allocator.
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
*
* 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));
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
KASSERT(rv->domain >= 0 && rv->domain < vm_ndomains,
("vm_reserv_reclaim: reserv %p's domain is corrupted %d",
rv, rv->domain));
TAILQ_REMOVE(&vm_rvq_partpop[rv->domain], rv, partpopq);
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
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
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* Breaks the reservation at the head of the partially populated reservation
* queue, releasing its free pages to the physical memory allocator. Returns
* TRUE if a reservation is broken and FALSE otherwise.
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 free page queue lock must be held.
*/
boolean_t
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
vm_reserv_reclaim_inactive(int domain)
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);
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
if ((rv = TAILQ_FIRST(&vm_rvq_partpop[domain])) != 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
/*
Eliminate every mention of PG_CACHED pages from the comments in the machine- independent layer of the virtual memory system. Update some of the nearby comments to eliminate redundancy and improve clarity. In vm/vm_reserv.c, do not use hyphens after adverbs ending in -ly per The Chicago Manual of Style. Update the comment in vm/vm_page.h defining the four types of page queues to reflect the elimination of PG_CACHED pages and the introduction of the laundry queue. Reviewed by: kib, markj Sponsored by: Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D8752
2016-12-12 17:47:09 +00:00
* Searches the partially populated reservation queue for the least recently
* changed reservation with free pages 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.
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
*
* The free page queue lock must be held.
*/
boolean_t
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
vm_reserv_reclaim_contig(int domain, u_long npages, vm_paddr_t low,
vm_paddr_t high, 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;
Correct an error in vm_reserv_reclaim_contig(). In the highly unusual case that the reservation contained "low", the starting position in the popmap for the free page search was incorrectly calculated. The most likely (and visible) symptom of this error was the assertion failure, "vm_reserv_reclaim_contig: pa is too low".
2015-11-26 19:12:18 +00:00
int hi, i, lo, low_index, 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;
Move domain iterators into the page layer where domain selection should take place. This makes the majority of the phys layer explicitly domain specific. Reviewed by: markj, kib (some objections) Discussed with: alc Tested by: pho Sponsored by: Netflix & Dell EMC Isilon Differential Revision: https://reviews.freebsd.org/D13014
2017-11-28 23:18:35 +00:00
TAILQ_FOREACH(rv, &vm_rvq_partpop[domain], partpopq) {
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
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". */
Correct an error in vm_reserv_reclaim_contig(). In the highly unusual case that the reservation contained "low", the starting position in the popmap for the free page search was incorrectly calculated. The most likely (and visible) symptom of this error was the assertion failure, "vm_reserv_reclaim_contig: pa is too low".
2015-11-26 19:12:18 +00:00
low_index = (low + PAGE_MASK - pa) >> PAGE_SHIFT;
i = low_index / NBPOPMAP;
hi = low_index % NBPOPMAP;
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
} 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) {
Correct an off-by-one error in vm_reserv_reclaim_contig() that results in an infinite loop. Submitted by: Svatopluk Kraus MFC after: 1 week
2015-04-11 22:57:13 +00:00
/*
* The current page doesn't meet the alignment
* and/or boundary requirements. Continue
* searching this reservation until the rest
* of its free pages are either excluded or
* exhausted.
*/
hi = lo + 1;
if (hi >= NBPOPMAP) {
hi = 0;
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
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);
}
}
Introduce a new mechanism for relocating virtual pages to a new physical address and use this mechanism when: 1. kmem_alloc_{attr,contig}() can't find suitable free pages in the physical memory allocator's free page lists. This replaces the long-standing approach of scanning the inactive and inactive queues, converting clean pages into PG_CACHED pages and laundering dirty pages. In contrast, the new mechanism does not use PG_CACHED pages nor does it trigger a large number of I/O operations. 2. on 32-bit MIPS processors, uma_small_alloc() and the pmap can't find free pages in the physical memory allocator's free page lists that are covered by the direct map. Tested by: adrian 3. ttm_bo_global_init() and ttm_vm_page_alloc_dma32() can't find suitable free pages in the physical memory allocator's free page lists. In the coming months, I expect that this new mechanism will be applied in other places. For example, balloon drivers should use relocation to minimize fragmentation of the guest physical address space. Make vm_phys_alloc_contig() a little smarter (and more efficient in some cases). Specifically, use vm_phys_segs[] earlier to avoid scanning free page lists that can't possibly contain suitable pages. Reviewed by: kib, markj Glanced at: jhb Discussed with: jeff Sponsored by: EMC / Isilon Storage Division Differential Revision: https://reviews.freebsd.org/D4444
2015-12-19 18:42:50 +00:00
/*
* Returns the size (in bytes) of a reservation of the specified level.
*/
int
vm_reserv_size(int level)
{
switch (level) {
case 0:
return (VM_LEVEL_0_SIZE);
case -1:
return (PAGE_SIZE);
default:
return (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
/*
* 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);
}
Utilize pmap_enter(..., psind=1) in vm_fault_soft_fast() on amd64. (The Differential Revision discusses the benefits of this change.) Add a function, vm_reserv_to_superpage(), that returns the superpage containing the specified base page. Reviewed by: kib, markj Tested by: pho MFC after: 10 days Differential Revision: https://reviews.freebsd.org/D11556
2017-07-23 16:28:13 +00:00
/*
* Returns the superpage containing the given page.
*/
vm_page_t
vm_reserv_to_superpage(vm_page_t m)
{
vm_reserv_t rv;
VM_OBJECT_ASSERT_LOCKED(m->object);
rv = vm_reserv_from_page(m);
return (rv->object == m->object && rv->popcnt == VM_LEVEL_0_NPAGES ?
rv->pages : 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
#endif /* VM_NRESERVLEVEL > 0 */
Reference in New Issue Copy Permalink