diff --git a/sys/vm/vm_kern.c b/sys/vm/vm_kern.c
index 22776e2196b0..6343fb66cfa3 100644
--- a/sys/vm/vm_kern.c
+++ b/sys/vm/vm_kern.c
@@ -1,1056 +1,1058 @@
/*-
* SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
*
* Copyright (c) 1991, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* The Mach Operating System project at Carnegie-Mellon University.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*
* Copyright (c) 1987, 1990 Carnegie-Mellon University.
* All rights reserved.
*
* Authors: Avadis Tevanian, Jr., Michael Wayne Young
*
* Permission to use, copy, modify and distribute this software and
* its documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie the
* rights to redistribute these changes.
*/
/*
* Kernel memory management.
*/
#include <sys/cdefs.h>
#include "opt_vm.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/asan.h>
#include <sys/domainset.h>
#include <sys/eventhandler.h>
#include <sys/kernel.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/msan.h>
#include <sys/proc.h>
#include <sys/rwlock.h>
#include <sys/smp.h>
#include <sys/sysctl.h>
#include <sys/vmem.h>
#include <sys/vmmeter.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/vm_domainset.h>
#include <vm/vm_kern.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
#include <vm/vm_pagequeue.h>
#include <vm/vm_phys.h>
#include <vm/vm_radix.h>
#include <vm/vm_extern.h>
#include <vm/uma.h>
struct vm_map kernel_map_store;
struct vm_map exec_map_store;
struct vm_map pipe_map_store;
const void *zero_region;
CTASSERT((ZERO_REGION_SIZE & PAGE_MASK) == 0);
/* NB: Used by kernel debuggers. */
const u_long vm_maxuser_address = VM_MAXUSER_ADDRESS;
u_int exec_map_entry_size;
u_int exec_map_entries;
SYSCTL_ULONG(_vm, OID_AUTO, min_kernel_address, CTLFLAG_RD,
SYSCTL_NULL_ULONG_PTR, VM_MIN_KERNEL_ADDRESS, "Min kernel address");
SYSCTL_ULONG(_vm, OID_AUTO, max_kernel_address, CTLFLAG_RD,
#if defined(__arm__)
&vm_max_kernel_address, 0,
#else
SYSCTL_NULL_ULONG_PTR, VM_MAX_KERNEL_ADDRESS,
#endif
"Max kernel address");
#if VM_NRESERVLEVEL > 1
#define KVA_QUANTUM_SHIFT (VM_LEVEL_1_ORDER + VM_LEVEL_0_ORDER + \
PAGE_SHIFT)
#elif VM_NRESERVLEVEL > 0
#define KVA_QUANTUM_SHIFT (VM_LEVEL_0_ORDER + PAGE_SHIFT)
#else
/* On non-superpage architectures we want large import sizes. */
#define KVA_QUANTUM_SHIFT (8 + PAGE_SHIFT)
#endif
#define KVA_QUANTUM (1ul << KVA_QUANTUM_SHIFT)
#define KVA_NUMA_IMPORT_QUANTUM (KVA_QUANTUM * 128)
extern void uma_startup2(void);
/*
* kva_alloc:
*
* Allocate a virtual address range with no underlying object and
* no initial mapping to physical memory. Any mapping from this
* range to physical memory must be explicitly created prior to
* its use, typically with pmap_qenter(). Any attempt to create
* a mapping on demand through vm_fault() will result in a panic.
*/
vm_offset_t
kva_alloc(vm_size_t size)
{
vm_offset_t addr;
TSENTER();
size = round_page(size);
if (vmem_xalloc(kernel_arena, size, 0, 0, 0, VMEM_ADDR_MIN,
VMEM_ADDR_MAX, M_BESTFIT | M_NOWAIT, &addr))
return (0);
TSEXIT();
return (addr);
}
/*
* kva_alloc_aligned:
*
* Allocate a virtual address range as in kva_alloc where the base
* address is aligned to align.
*/
vm_offset_t
kva_alloc_aligned(vm_size_t size, vm_size_t align)
{
vm_offset_t addr;
TSENTER();
size = round_page(size);
if (vmem_xalloc(kernel_arena, size, align, 0, 0, VMEM_ADDR_MIN,
VMEM_ADDR_MAX, M_BESTFIT | M_NOWAIT, &addr))
return (0);
TSEXIT();
return (addr);
}
/*
* kva_free:
*
* Release a region of kernel virtual memory allocated
* with kva_alloc, and return the physical pages
* associated with that region.
*
* This routine may not block on kernel maps.
*/
void
kva_free(vm_offset_t addr, vm_size_t size)
{
size = round_page(size);
vmem_xfree(kernel_arena, addr, size);
}
/*
* Update sanitizer shadow state to reflect a new allocation. Force inlining to
* help make KMSAN origin tracking more precise.
*/
static __always_inline void
kmem_alloc_san(vm_offset_t addr, vm_size_t size, vm_size_t asize, int flags)
{
if ((flags & M_ZERO) == 0) {
kmsan_mark((void *)addr, asize, KMSAN_STATE_UNINIT);
kmsan_orig((void *)addr, asize, KMSAN_TYPE_KMEM,
KMSAN_RET_ADDR);
} else {
kmsan_mark((void *)addr, asize, KMSAN_STATE_INITED);
}
kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
}
static vm_page_t
kmem_alloc_contig_pages(vm_object_t object, vm_pindex_t pindex, int domain,
int pflags, u_long npages, vm_paddr_t low, vm_paddr_t high,
u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
{
vm_page_t m;
int tries;
bool wait, reclaim;
VM_OBJECT_ASSERT_WLOCKED(object);
wait = (pflags & VM_ALLOC_WAITOK) != 0;
reclaim = (pflags & VM_ALLOC_NORECLAIM) == 0;
pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
pflags |= VM_ALLOC_NOWAIT;
for (tries = wait ? 3 : 1;; tries--) {
m = vm_page_alloc_contig_domain(object, pindex, domain, pflags,
npages, low, high, alignment, boundary, memattr);
if (m != NULL || tries == 0 || !reclaim)
break;
VM_OBJECT_WUNLOCK(object);
if (vm_page_reclaim_contig_domain(domain, pflags, npages,
low, high, alignment, boundary) == ENOMEM && wait)
vm_wait_domain(domain);
VM_OBJECT_WLOCK(object);
}
return (m);
}
/*
* Allocates a region from the kernel address map and physical pages
* within the specified address range to the kernel object. Creates a
* wired mapping from this region to these pages, and returns the
* region's starting virtual address. The allocated pages are not
* necessarily physically contiguous. If M_ZERO is specified through the
* given flags, then the pages are zeroed before they are mapped.
*/
static void *
kmem_alloc_attr_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
vm_paddr_t high, vm_memattr_t memattr)
{
vmem_t *vmem;
vm_object_t object;
vm_offset_t addr, i, offset;
vm_page_t m;
vm_size_t asize;
int pflags;
vm_prot_t prot;
object = kernel_object;
asize = round_page(size);
vmem = vm_dom[domain].vmd_kernel_arena;
if (vmem_alloc(vmem, asize, M_BESTFIT | flags, &addr))
return (0);
offset = addr - VM_MIN_KERNEL_ADDRESS;
pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
VM_OBJECT_WLOCK(object);
for (i = 0; i < asize; i += PAGE_SIZE) {
m = kmem_alloc_contig_pages(object, atop(offset + i),
domain, pflags, 1, low, high, PAGE_SIZE, 0, memattr);
if (m == NULL) {
VM_OBJECT_WUNLOCK(object);
kmem_unback(object, addr, i);
vmem_free(vmem, addr, asize);
return (0);
}
KASSERT(vm_page_domain(m) == domain,
("kmem_alloc_attr_domain: Domain mismatch %d != %d",
vm_page_domain(m), domain));
if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
vm_page_valid(m);
pmap_enter(kernel_pmap, addr + i, m, prot,
prot | PMAP_ENTER_WIRED, 0);
}
VM_OBJECT_WUNLOCK(object);
kmem_alloc_san(addr, size, asize, flags);
return ((void *)addr);
}
void *
kmem_alloc_attr(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
vm_memattr_t memattr)
{
return (kmem_alloc_attr_domainset(DOMAINSET_RR(), size, flags, low,
high, memattr));
}
void *
kmem_alloc_attr_domainset(struct domainset *ds, vm_size_t size, int flags,
vm_paddr_t low, vm_paddr_t high, vm_memattr_t memattr)
{
struct vm_domainset_iter di;
vm_page_t bounds[2];
void *addr;
int domain;
int start_segind;
start_segind = -1;
vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
do {
addr = kmem_alloc_attr_domain(domain, size, flags, low, high,
memattr);
if (addr != NULL)
break;
if (start_segind == -1)
start_segind = vm_phys_lookup_segind(low);
if (vm_phys_find_range(bounds, start_segind, domain,
atop(round_page(size)), low, high) == -1) {
vm_domainset_iter_ignore(&di, domain);
}
} while (vm_domainset_iter_policy(&di, &domain) == 0);
return (addr);
}
/*
* Allocates a region from the kernel address map and physically
* contiguous pages within the specified address range to the kernel
* object. Creates a wired mapping from this region to these pages, and
* returns the region's starting virtual address. If M_ZERO is specified
* through the given flags, then the pages are zeroed before they are
* mapped.
*/
static void *
kmem_alloc_contig_domain(int domain, vm_size_t size, int flags, vm_paddr_t low,
vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr)
{
vmem_t *vmem;
vm_object_t object;
vm_offset_t addr, offset, tmp;
vm_page_t end_m, m;
vm_size_t asize;
u_long npages;
int pflags;
object = kernel_object;
asize = round_page(size);
vmem = vm_dom[domain].vmd_kernel_arena;
if (vmem_alloc(vmem, asize, flags | M_BESTFIT, &addr))
return (NULL);
offset = addr - VM_MIN_KERNEL_ADDRESS;
pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
npages = atop(asize);
VM_OBJECT_WLOCK(object);
m = kmem_alloc_contig_pages(object, atop(offset), domain,
pflags, npages, low, high, alignment, boundary, memattr);
if (m == NULL) {
VM_OBJECT_WUNLOCK(object);
vmem_free(vmem, addr, asize);
return (NULL);
}
KASSERT(vm_page_domain(m) == domain,
("kmem_alloc_contig_domain: Domain mismatch %d != %d",
vm_page_domain(m), domain));
end_m = m + npages;
tmp = addr;
for (; m < end_m; m++) {
if ((flags & M_ZERO) && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
vm_page_valid(m);
pmap_enter(kernel_pmap, tmp, m, VM_PROT_RW,
VM_PROT_RW | PMAP_ENTER_WIRED, 0);
tmp += PAGE_SIZE;
}
VM_OBJECT_WUNLOCK(object);
kmem_alloc_san(addr, size, asize, flags);
return ((void *)addr);
}
void *
kmem_alloc_contig(vm_size_t size, int flags, vm_paddr_t low, vm_paddr_t high,
u_long alignment, vm_paddr_t boundary, vm_memattr_t memattr)
{
return (kmem_alloc_contig_domainset(DOMAINSET_RR(), size, flags, low,
high, alignment, boundary, memattr));
}
void *
kmem_alloc_contig_domainset(struct domainset *ds, vm_size_t size, int flags,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr)
{
struct vm_domainset_iter di;
vm_page_t bounds[2];
void *addr;
int domain;
int start_segind;
start_segind = -1;
vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
do {
addr = kmem_alloc_contig_domain(domain, size, flags, low, high,
alignment, boundary, memattr);
if (addr != NULL)
break;
if (start_segind == -1)
start_segind = vm_phys_lookup_segind(low);
if (vm_phys_find_range(bounds, start_segind, domain,
atop(round_page(size)), low, high) == -1) {
vm_domainset_iter_ignore(&di, domain);
}
} while (vm_domainset_iter_policy(&di, &domain) == 0);
return (addr);
}
/*
* kmem_subinit:
*
* Initializes a map to manage a subrange
* of the kernel virtual address space.
*
* Arguments are as follows:
*
* parent Map to take range from
* min, max Returned endpoints of map
* size Size of range to find
* superpage_align Request that min is superpage aligned
*/
void
kmem_subinit(vm_map_t map, vm_map_t parent, vm_offset_t *min, vm_offset_t *max,
vm_size_t size, bool superpage_align)
{
int ret;
size = round_page(size);
*min = vm_map_min(parent);
ret = vm_map_find(parent, NULL, 0, min, size, 0, superpage_align ?
VMFS_SUPER_SPACE : VMFS_ANY_SPACE, VM_PROT_ALL, VM_PROT_ALL,
MAP_ACC_NO_CHARGE);
if (ret != KERN_SUCCESS)
panic("kmem_subinit: bad status return of %d", ret);
*max = *min + size;
vm_map_init(map, vm_map_pmap(parent), *min, *max);
if (vm_map_submap(parent, *min, *max, map) != KERN_SUCCESS)
panic("kmem_subinit: unable to change range to submap");
}
/*
* kmem_malloc_domain:
*
* Allocate wired-down pages in the kernel's address space.
*/
static void *
kmem_malloc_domain(int domain, vm_size_t size, int flags)
{
vmem_t *arena;
vm_offset_t addr;
vm_size_t asize;
int rv;
if (__predict_true((flags & (M_EXEC | M_NEVERFREED)) == 0))
arena = vm_dom[domain].vmd_kernel_arena;
else if ((flags & M_EXEC) != 0)
arena = vm_dom[domain].vmd_kernel_rwx_arena;
else
arena = vm_dom[domain].vmd_kernel_nofree_arena;
asize = round_page(size);
if (vmem_alloc(arena, asize, flags | M_BESTFIT, &addr))
return (0);
rv = kmem_back_domain(domain, kernel_object, addr, asize, flags);
if (rv != KERN_SUCCESS) {
vmem_free(arena, addr, asize);
return (0);
}
kasan_mark((void *)addr, size, asize, KASAN_KMEM_REDZONE);
return ((void *)addr);
}
void *
kmem_malloc(vm_size_t size, int flags)
{
void * p;
TSENTER();
p = kmem_malloc_domainset(DOMAINSET_RR(), size, flags);
TSEXIT();
return (p);
}
void *
kmem_malloc_domainset(struct domainset *ds, vm_size_t size, int flags)
{
struct vm_domainset_iter di;
void *addr;
int domain;
vm_domainset_iter_policy_init(&di, ds, &domain, &flags);
do {
addr = kmem_malloc_domain(domain, size, flags);
if (addr != NULL)
break;
} while (vm_domainset_iter_policy(&di, &domain) == 0);
return (addr);
}
/*
* kmem_back_domain:
*
* Allocate physical pages from the specified domain for the specified
* virtual address range.
*/
int
kmem_back_domain(int domain, vm_object_t object, vm_offset_t addr,
vm_size_t size, int flags)
{
vm_offset_t offset, i;
vm_page_t m, mpred;
vm_prot_t prot;
int pflags;
KASSERT(object == kernel_object,
("kmem_back_domain: only supports kernel object."));
offset = addr - VM_MIN_KERNEL_ADDRESS;
pflags = malloc2vm_flags(flags) | VM_ALLOC_WIRED;
pflags &= ~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL);
if (flags & M_WAITOK)
pflags |= VM_ALLOC_WAITFAIL;
prot = (flags & M_EXEC) != 0 ? VM_PROT_ALL : VM_PROT_RW;
i = 0;
VM_OBJECT_WLOCK(object);
retry:
mpred = vm_radix_lookup_le(&object->rtree, atop(offset + i));
for (; i < size; i += PAGE_SIZE, mpred = m) {
m = vm_page_alloc_domain_after(object, atop(offset + i),
domain, pflags, mpred);
/*
* Ran out of space, free everything up and return. Don't need
* to lock page queues here as we know that the pages we got
* aren't on any queues.
*/
if (m == NULL) {
if ((flags & M_NOWAIT) == 0)
goto retry;
VM_OBJECT_WUNLOCK(object);
kmem_unback(object, addr, i);
return (KERN_NO_SPACE);
}
KASSERT(vm_page_domain(m) == domain,
("kmem_back_domain: Domain mismatch %d != %d",
vm_page_domain(m), domain));
if (flags & M_ZERO && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
KASSERT((m->oflags & VPO_UNMANAGED) != 0,
("kmem_malloc: page %p is managed", m));
vm_page_valid(m);
pmap_enter(kernel_pmap, addr + i, m, prot,
prot | PMAP_ENTER_WIRED, 0);
if (__predict_false((prot & VM_PROT_EXECUTE) != 0))
m->oflags |= VPO_KMEM_EXEC;
}
VM_OBJECT_WUNLOCK(object);
kmem_alloc_san(addr, size, size, flags);
return (KERN_SUCCESS);
}
/*
* kmem_back:
*
* Allocate physical pages for the specified virtual address range.
*/
int
kmem_back(vm_object_t object, vm_offset_t addr, vm_size_t size, int flags)
{
vm_offset_t end, next, start;
int domain, rv;
KASSERT(object == kernel_object,
("kmem_back: only supports kernel object."));
for (start = addr, end = addr + size; addr < end; addr = next) {
/*
* We must ensure that pages backing a given large virtual page
* all come from the same physical domain.
*/
if (vm_ndomains > 1) {
domain = (addr >> KVA_QUANTUM_SHIFT) % vm_ndomains;
while (VM_DOMAIN_EMPTY(domain))
domain++;
next = roundup2(addr + 1, KVA_QUANTUM);
if (next > end || next < start)
next = end;
} else {
domain = 0;
next = end;
}
rv = kmem_back_domain(domain, object, addr, next - addr, flags);
if (rv != KERN_SUCCESS) {
kmem_unback(object, start, addr - start);
break;
}
}
return (rv);
}
/*
* kmem_unback:
*
* Unmap and free the physical pages underlying the specified virtual
* address range.
*
* A physical page must exist within the specified object at each index
* that is being unmapped.
*/
static struct vmem *
_kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
{
+ struct pctrie_iter pages;
struct vmem *arena;
- vm_page_t m, next;
+ vm_page_t m;
vm_offset_t end, offset;
int domain;
KASSERT(object == kernel_object,
("kmem_unback: only supports kernel object."));
if (size == 0)
return (NULL);
pmap_remove(kernel_pmap, addr, addr + size);
offset = addr - VM_MIN_KERNEL_ADDRESS;
end = offset + size;
VM_OBJECT_WLOCK(object);
- m = vm_page_lookup(object, atop(offset));
+ vm_page_iter_init(&pages, object);
+ m = vm_page_iter_lookup(&pages, atop(offset));
domain = vm_page_domain(m);
if (__predict_true((m->oflags & VPO_KMEM_EXEC) == 0))
arena = vm_dom[domain].vmd_kernel_arena;
else
arena = vm_dom[domain].vmd_kernel_rwx_arena;
- for (; offset < end; offset += PAGE_SIZE, m = next) {
- next = vm_page_next(m);
+ for (; offset < end; offset += PAGE_SIZE,
+ m = vm_page_iter_lookup(&pages, atop(offset))) {
vm_page_xbusy_claim(m);
vm_page_unwire_noq(m);
- vm_page_free(m);
+ vm_page_iter_free(&pages);
}
VM_OBJECT_WUNLOCK(object);
return (arena);
}
void
kmem_unback(vm_object_t object, vm_offset_t addr, vm_size_t size)
{
(void)_kmem_unback(object, addr, size);
}
/*
* kmem_free:
*
* Free memory allocated with kmem_malloc. The size must match the
* original allocation.
*/
void
kmem_free(void *addr, vm_size_t size)
{
struct vmem *arena;
size = round_page(size);
kasan_mark(addr, size, size, 0);
arena = _kmem_unback(kernel_object, (uintptr_t)addr, size);
if (arena != NULL)
vmem_free(arena, (uintptr_t)addr, size);
}
/*
* kmap_alloc_wait:
*
* Allocates pageable memory from a sub-map of the kernel. If the submap
* has no room, the caller sleeps waiting for more memory in the submap.
*
* This routine may block.
*/
vm_offset_t
kmap_alloc_wait(vm_map_t map, vm_size_t size)
{
vm_offset_t addr;
size = round_page(size);
if (!swap_reserve(size))
return (0);
for (;;) {
/*
* To make this work for more than one map, use the map's lock
* to lock out sleepers/wakers.
*/
vm_map_lock(map);
addr = vm_map_findspace(map, vm_map_min(map), size);
if (addr + size <= vm_map_max(map))
break;
/* no space now; see if we can ever get space */
if (vm_map_max(map) - vm_map_min(map) < size) {
vm_map_unlock(map);
swap_release(size);
return (0);
}
map->needs_wakeup = TRUE;
vm_map_unlock_and_wait(map, 0);
}
vm_map_insert(map, NULL, 0, addr, addr + size, VM_PROT_RW, VM_PROT_RW,
MAP_ACC_CHARGED);
vm_map_unlock(map);
return (addr);
}
/*
* kmap_free_wakeup:
*
* Returns memory to a submap of the kernel, and wakes up any processes
* waiting for memory in that map.
*/
void
kmap_free_wakeup(vm_map_t map, vm_offset_t addr, vm_size_t size)
{
vm_map_lock(map);
(void) vm_map_delete(map, trunc_page(addr), round_page(addr + size));
if (map->needs_wakeup) {
map->needs_wakeup = FALSE;
vm_map_wakeup(map);
}
vm_map_unlock(map);
}
void
kmem_init_zero_region(void)
{
vm_offset_t addr, i;
vm_page_t m;
/*
* Map a single physical page of zeros to a larger virtual range.
* This requires less looping in places that want large amounts of
* zeros, while not using much more physical resources.
*/
addr = kva_alloc(ZERO_REGION_SIZE);
m = vm_page_alloc_noobj(VM_ALLOC_WIRED | VM_ALLOC_ZERO |
VM_ALLOC_NOFREE);
for (i = 0; i < ZERO_REGION_SIZE; i += PAGE_SIZE)
pmap_qenter(addr + i, &m, 1);
pmap_protect(kernel_pmap, addr, addr + ZERO_REGION_SIZE, VM_PROT_READ);
zero_region = (const void *)addr;
}
/*
* Import KVA from the kernel map into the kernel arena.
*/
static int
kva_import(void *unused, vmem_size_t size, int flags, vmem_addr_t *addrp)
{
vm_offset_t addr;
int result;
TSENTER();
KASSERT((size % KVA_QUANTUM) == 0,
("kva_import: Size %jd is not a multiple of %d",
(intmax_t)size, (int)KVA_QUANTUM));
addr = vm_map_min(kernel_map);
result = vm_map_find(kernel_map, NULL, 0, &addr, size, 0,
VMFS_SUPER_SPACE, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
if (result != KERN_SUCCESS) {
TSEXIT();
return (ENOMEM);
}
*addrp = addr;
TSEXIT();
return (0);
}
/*
* Import KVA from a parent arena into a per-domain arena. Imports must be
* KVA_QUANTUM-aligned and a multiple of KVA_QUANTUM in size.
*/
static int
kva_import_domain(void *arena, vmem_size_t size, int flags, vmem_addr_t *addrp)
{
KASSERT((size % KVA_QUANTUM) == 0,
("kva_import_domain: Size %jd is not a multiple of %d",
(intmax_t)size, (int)KVA_QUANTUM));
return (vmem_xalloc(arena, size, KVA_QUANTUM, 0, 0, VMEM_ADDR_MIN,
VMEM_ADDR_MAX, flags, addrp));
}
/*
* kmem_init:
*
* Create the kernel map; insert a mapping covering kernel text,
* data, bss, and all space allocated thus far (`boostrap' data). The
* new map will thus map the range between VM_MIN_KERNEL_ADDRESS and
* `start' as allocated, and the range between `start' and `end' as free.
* Create the kernel vmem arena and its per-domain children.
*/
void
kmem_init(vm_offset_t start, vm_offset_t end)
{
vm_size_t quantum;
int domain;
vm_map_init(kernel_map, kernel_pmap, VM_MIN_KERNEL_ADDRESS, end);
kernel_map->system_map = 1;
vm_map_lock(kernel_map);
/* N.B.: cannot use kgdb to debug, starting with this assignment ... */
(void)vm_map_insert(kernel_map, NULL, 0,
#ifdef __amd64__
KERNBASE,
#else
VM_MIN_KERNEL_ADDRESS,
#endif
start, VM_PROT_ALL, VM_PROT_ALL, MAP_NOFAULT);
/* ... and ending with the completion of the above `insert' */
#ifdef __amd64__
/*
* Mark KVA used for the page array as allocated. Other platforms
* that handle vm_page_array allocation can simply adjust virtual_avail
* instead.
*/
(void)vm_map_insert(kernel_map, NULL, 0, (vm_offset_t)vm_page_array,
(vm_offset_t)vm_page_array + round_2mpage(vm_page_array_size *
sizeof(struct vm_page)),
VM_PROT_RW, VM_PROT_RW, MAP_NOFAULT);
#endif
vm_map_unlock(kernel_map);
/*
* Use a large import quantum on NUMA systems. This helps minimize
* interleaving of superpages, reducing internal fragmentation within
* the per-domain arenas.
*/
if (vm_ndomains > 1 && PMAP_HAS_DMAP)
quantum = KVA_NUMA_IMPORT_QUANTUM;
else
quantum = KVA_QUANTUM;
/*
* Initialize the kernel_arena. This can grow on demand.
*/
vmem_init(kernel_arena, "kernel arena", 0, 0, PAGE_SIZE, 0, 0);
vmem_set_import(kernel_arena, kva_import, NULL, NULL, quantum);
for (domain = 0; domain < vm_ndomains; domain++) {
/*
* Initialize the per-domain arenas. These are used to color
* the KVA space in a way that ensures that virtual large pages
* are backed by memory from the same physical domain,
* maximizing the potential for superpage promotion.
*/
vm_dom[domain].vmd_kernel_arena = vmem_create(
"kernel arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
vmem_set_import(vm_dom[domain].vmd_kernel_arena,
kva_import_domain, NULL, kernel_arena, quantum);
/*
* In architectures with superpages, maintain separate arenas
* for allocations with permissions that differ from the
* "standard" read/write permissions used for kernel memory
* and pages that are never released, so as not to inhibit
* superpage promotion.
*
* Use the base import quantum since these arenas are rarely
* used.
*/
#if VM_NRESERVLEVEL > 0
vm_dom[domain].vmd_kernel_rwx_arena = vmem_create(
"kernel rwx arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
vm_dom[domain].vmd_kernel_nofree_arena = vmem_create(
"kernel NOFREE arena domain", 0, 0, PAGE_SIZE, 0, M_WAITOK);
vmem_set_import(vm_dom[domain].vmd_kernel_rwx_arena,
kva_import_domain, (vmem_release_t *)vmem_xfree,
kernel_arena, KVA_QUANTUM);
vmem_set_import(vm_dom[domain].vmd_kernel_nofree_arena,
kva_import_domain, (vmem_release_t *)vmem_xfree,
kernel_arena, KVA_QUANTUM);
#else
vm_dom[domain].vmd_kernel_rwx_arena =
vm_dom[domain].vmd_kernel_arena;
vm_dom[domain].vmd_kernel_nofree_arena =
vm_dom[domain].vmd_kernel_arena;
#endif
}
/*
* This must be the very first call so that the virtual address
* space used for early allocations is properly marked used in
* the map.
*/
uma_startup2();
}
/*
* kmem_bootstrap_free:
*
* Free pages backing preloaded data (e.g., kernel modules) to the
* system. Currently only supported on platforms that create a
* vm_phys segment for preloaded data.
*/
void
kmem_bootstrap_free(vm_offset_t start, vm_size_t size)
{
#if defined(__i386__) || defined(__amd64__)
struct vm_domain *vmd;
vm_offset_t end, va;
vm_paddr_t pa;
vm_page_t m;
end = trunc_page(start + size);
start = round_page(start);
#ifdef __amd64__
/*
* Preloaded files do not have execute permissions by default on amd64.
* Restore the default permissions to ensure that the direct map alias
* is updated.
*/
pmap_change_prot(start, end - start, VM_PROT_RW);
#endif
for (va = start; va < end; va += PAGE_SIZE) {
pa = pmap_kextract(va);
m = PHYS_TO_VM_PAGE(pa);
vmd = vm_pagequeue_domain(m);
vm_domain_free_lock(vmd);
vm_phys_free_pages(m, 0);
vm_domain_free_unlock(vmd);
vm_domain_freecnt_inc(vmd, 1);
vm_cnt.v_page_count++;
}
pmap_remove(kernel_pmap, start, end);
(void)vmem_add(kernel_arena, start, end - start, M_WAITOK);
#endif
}
#ifdef PMAP_WANT_ACTIVE_CPUS_NAIVE
void
pmap_active_cpus(pmap_t pmap, cpuset_t *res)
{
struct thread *td;
struct proc *p;
struct vmspace *vm;
int c;
CPU_ZERO(res);
CPU_FOREACH(c) {
td = cpuid_to_pcpu[c]->pc_curthread;
p = td->td_proc;
if (p == NULL)
continue;
vm = vmspace_acquire_ref(p);
if (vm == NULL)
continue;
if (pmap == vmspace_pmap(vm))
CPU_SET(c, res);
vmspace_free(vm);
}
}
#endif
/*
* Allow userspace to directly trigger the VM drain routine for testing
* purposes.
*/
static int
debug_vm_lowmem(SYSCTL_HANDLER_ARGS)
{
int error, i;
i = 0;
error = sysctl_handle_int(oidp, &i, 0, req);
if (error != 0)
return (error);
if ((i & ~(VM_LOW_KMEM | VM_LOW_PAGES)) != 0)
return (EINVAL);
if (i != 0)
EVENTHANDLER_INVOKE(vm_lowmem, i);
return (0);
}
SYSCTL_PROC(_debug, OID_AUTO, vm_lowmem,
CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_vm_lowmem, "I",
"set to trigger vm_lowmem event with given flags");
static int
debug_uma_reclaim(SYSCTL_HANDLER_ARGS)
{
int error, i;
i = 0;
error = sysctl_handle_int(oidp, &i, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
if (i != UMA_RECLAIM_TRIM && i != UMA_RECLAIM_DRAIN &&
i != UMA_RECLAIM_DRAIN_CPU)
return (EINVAL);
uma_reclaim(i);
return (0);
}
SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim,
CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0, debug_uma_reclaim, "I",
"set to generate request to reclaim uma caches");
static int
debug_uma_reclaim_domain(SYSCTL_HANDLER_ARGS)
{
int domain, error, request;
request = 0;
error = sysctl_handle_int(oidp, &request, 0, req);
if (error != 0 || req->newptr == NULL)
return (error);
domain = request >> 4;
request &= 0xf;
if (request != UMA_RECLAIM_TRIM && request != UMA_RECLAIM_DRAIN &&
request != UMA_RECLAIM_DRAIN_CPU)
return (EINVAL);
if (domain < 0 || domain >= vm_ndomains)
return (EINVAL);
uma_reclaim_domain(request, domain);
return (0);
}
SYSCTL_PROC(_debug, OID_AUTO, uma_reclaim_domain,
CTLTYPE_INT | CTLFLAG_MPSAFE | CTLFLAG_RW, 0, 0,
debug_uma_reclaim_domain, "I",
"");
diff --git a/sys/vm/vm_object.c b/sys/vm/vm_object.c
index e6324647e29e..21773318cea0 100644
--- a/sys/vm/vm_object.c
+++ b/sys/vm/vm_object.c
@@ -1,2808 +1,2808 @@
/*-
* SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
*
* Copyright (c) 1991, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* The Mach Operating System project at Carnegie-Mellon University.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*
* Copyright (c) 1987, 1990 Carnegie-Mellon University.
* All rights reserved.
*
* Authors: Avadis Tevanian, Jr., Michael Wayne Young
*
* Permission to use, copy, modify and distribute this software and
* its documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie the
* rights to redistribute these changes.
*/
/*
* Virtual memory object module.
*/
#include "opt_vm.h"
#include <sys/systm.h>
#include <sys/blockcount.h>
#include <sys/conf.h>
#include <sys/cpuset.h>
#include <sys/ipc.h>
#include <sys/jail.h>
#include <sys/limits.h>
#include <sys/lock.h>
#include <sys/mman.h>
#include <sys/mount.h>
#include <sys/kernel.h>
#include <sys/mutex.h>
#include <sys/pctrie.h>
#include <sys/proc.h>
#include <sys/refcount.h>
#include <sys/shm.h>
#include <sys/sx.h>
#include <sys/sysctl.h>
#include <sys/resourcevar.h>
#include <sys/refcount.h>
#include <sys/rwlock.h>
#include <sys/user.h>
#include <sys/vnode.h>
#include <sys/vmmeter.h>
#include <vm/vm.h>
#include <vm/vm_param.h>
#include <vm/pmap.h>
#include <vm/vm_map.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
#include <vm/vm_pager.h>
#include <vm/vm_phys.h>
#include <vm/vm_pagequeue.h>
#include <vm/swap_pager.h>
#include <vm/vm_kern.h>
#include <vm/vm_extern.h>
#include <vm/vm_radix.h>
#include <vm/vm_reserv.h>
#include <vm/uma.h>
static int old_msync;
SYSCTL_INT(_vm, OID_AUTO, old_msync, CTLFLAG_RW, &old_msync, 0,
"Use old (insecure) msync behavior");
static int vm_object_page_collect_flush(vm_object_t object, vm_page_t p,
int pagerflags, int flags, boolean_t *allclean,
boolean_t *eio);
static boolean_t vm_object_page_remove_write(vm_page_t p, int flags,
boolean_t *allclean);
static void vm_object_backing_remove(vm_object_t object);
/*
* Virtual memory objects maintain the actual data
* associated with allocated virtual memory. A given
* page of memory exists within exactly one object.
*
* An object is only deallocated when all "references"
* are given up. Only one "reference" to a given
* region of an object should be writeable.
*
* Associated with each object is a list of all resident
* memory pages belonging to that object; this list is
* maintained by the "vm_page" module, and locked by the object's
* lock.
*
* Each object also records a "pager" routine which is
* used to retrieve (and store) pages to the proper backing
* storage. In addition, objects may be backed by other
* objects from which they were virtual-copied.
*
* The only items within the object structure which are
* modified after time of creation are:
* reference count locked by object's lock
* pager routine locked by object's lock
*
*/
struct object_q vm_object_list;
struct mtx vm_object_list_mtx; /* lock for object list and count */
struct vm_object kernel_object_store;
static SYSCTL_NODE(_vm_stats, OID_AUTO, object, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
"VM object stats");
static COUNTER_U64_DEFINE_EARLY(object_collapses);
SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapses, CTLFLAG_RD,
&object_collapses,
"VM object collapses");
static COUNTER_U64_DEFINE_EARLY(object_bypasses);
SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, bypasses, CTLFLAG_RD,
&object_bypasses,
"VM object bypasses");
static COUNTER_U64_DEFINE_EARLY(object_collapse_waits);
SYSCTL_COUNTER_U64(_vm_stats_object, OID_AUTO, collapse_waits, CTLFLAG_RD,
&object_collapse_waits,
"Number of sleeps for collapse");
static uma_zone_t obj_zone;
static int vm_object_zinit(void *mem, int size, int flags);
#ifdef INVARIANTS
static void vm_object_zdtor(void *mem, int size, void *arg);
static void
vm_object_zdtor(void *mem, int size, void *arg)
{
vm_object_t object;
object = (vm_object_t)mem;
KASSERT(object->ref_count == 0,
("object %p ref_count = %d", object, object->ref_count));
KASSERT(TAILQ_EMPTY(&object->memq),
("object %p has resident pages in its memq", object));
KASSERT(vm_radix_is_empty(&object->rtree),
("object %p has resident pages in its trie", object));
#if VM_NRESERVLEVEL > 0
KASSERT(LIST_EMPTY(&object->rvq),
("object %p has reservations",
object));
#endif
KASSERT(!vm_object_busied(object),
("object %p busy = %d", object, blockcount_read(&object->busy)));
KASSERT(object->resident_page_count == 0,
("object %p resident_page_count = %d",
object, object->resident_page_count));
KASSERT(atomic_load_int(&object->shadow_count) == 0,
("object %p shadow_count = %d",
object, atomic_load_int(&object->shadow_count)));
KASSERT(object->type == OBJT_DEAD,
("object %p has non-dead type %d",
object, object->type));
KASSERT(object->charge == 0 && object->cred == NULL,
("object %p has non-zero charge %ju (%p)",
object, (uintmax_t)object->charge, object->cred));
}
#endif
static int
vm_object_zinit(void *mem, int size, int flags)
{
vm_object_t object;
object = (vm_object_t)mem;
rw_init_flags(&object->lock, "vmobject", RW_DUPOK | RW_NEW);
/* These are true for any object that has been freed */
object->type = OBJT_DEAD;
vm_radix_init(&object->rtree);
refcount_init(&object->ref_count, 0);
blockcount_init(&object->paging_in_progress);
blockcount_init(&object->busy);
object->resident_page_count = 0;
atomic_store_int(&object->shadow_count, 0);
object->flags = OBJ_DEAD;
mtx_lock(&vm_object_list_mtx);
TAILQ_INSERT_TAIL(&vm_object_list, object, object_list);
mtx_unlock(&vm_object_list_mtx);
return (0);
}
static void
_vm_object_allocate(objtype_t type, vm_pindex_t size, u_short flags,
vm_object_t object, void *handle)
{
TAILQ_INIT(&object->memq);
LIST_INIT(&object->shadow_head);
object->type = type;
object->flags = flags;
if ((flags & OBJ_SWAP) != 0) {
pctrie_init(&object->un_pager.swp.swp_blks);
object->un_pager.swp.writemappings = 0;
}
/*
* Ensure that swap_pager_swapoff() iteration over object_list
* sees up to date type and pctrie head if it observed
* non-dead object.
*/
atomic_thread_fence_rel();
object->pg_color = 0;
object->size = size;
object->domain.dr_policy = NULL;
object->generation = 1;
object->cleangeneration = 1;
refcount_init(&object->ref_count, 1);
object->memattr = VM_MEMATTR_DEFAULT;
object->cred = NULL;
object->charge = 0;
object->handle = handle;
object->backing_object = NULL;
object->backing_object_offset = (vm_ooffset_t) 0;
#if VM_NRESERVLEVEL > 0
LIST_INIT(&object->rvq);
#endif
umtx_shm_object_init(object);
}
/*
* vm_object_init:
*
* Initialize the VM objects module.
*/
void
vm_object_init(void)
{
TAILQ_INIT(&vm_object_list);
mtx_init(&vm_object_list_mtx, "vm object_list", NULL, MTX_DEF);
rw_init(&kernel_object->lock, "kernel vm object");
vm_radix_init(&kernel_object->rtree);
_vm_object_allocate(OBJT_PHYS, atop(VM_MAX_KERNEL_ADDRESS -
VM_MIN_KERNEL_ADDRESS), OBJ_UNMANAGED, kernel_object, NULL);
#if VM_NRESERVLEVEL > 0
kernel_object->flags |= OBJ_COLORED;
kernel_object->pg_color = (u_short)atop(VM_MIN_KERNEL_ADDRESS);
#endif
kernel_object->un_pager.phys.ops = &default_phys_pg_ops;
/*
* The lock portion of struct vm_object must be type stable due
* to vm_pageout_fallback_object_lock locking a vm object
* without holding any references to it.
*
* paging_in_progress is valid always. Lockless references to
* the objects may acquire pip and then check OBJ_DEAD.
*/
obj_zone = uma_zcreate("VM OBJECT", sizeof (struct vm_object), NULL,
#ifdef INVARIANTS
vm_object_zdtor,
#else
NULL,
#endif
vm_object_zinit, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
vm_radix_zinit();
}
void
vm_object_clear_flag(vm_object_t object, u_short bits)
{
VM_OBJECT_ASSERT_WLOCKED(object);
object->flags &= ~bits;
}
/*
* Sets the default memory attribute for the specified object. Pages
* that are allocated to this object are by default assigned this memory
* attribute.
*
* Presently, this function must be called before any pages are allocated
* to the object. In the future, this requirement may be relaxed for
* "default" and "swap" objects.
*/
int
vm_object_set_memattr(vm_object_t object, vm_memattr_t memattr)
{
VM_OBJECT_ASSERT_WLOCKED(object);
if (object->type == OBJT_DEAD)
return (KERN_INVALID_ARGUMENT);
if (!TAILQ_EMPTY(&object->memq))
return (KERN_FAILURE);
object->memattr = memattr;
return (KERN_SUCCESS);
}
void
vm_object_pip_add(vm_object_t object, short i)
{
if (i > 0)
blockcount_acquire(&object->paging_in_progress, i);
}
void
vm_object_pip_wakeup(vm_object_t object)
{
vm_object_pip_wakeupn(object, 1);
}
void
vm_object_pip_wakeupn(vm_object_t object, short i)
{
if (i > 0)
blockcount_release(&object->paging_in_progress, i);
}
/*
* Atomically drop the object lock and wait for pip to drain. This protects
* from sleep/wakeup races due to identity changes. The lock is not re-acquired
* on return.
*/
static void
vm_object_pip_sleep(vm_object_t object, const char *waitid)
{
(void)blockcount_sleep(&object->paging_in_progress, &object->lock,
waitid, PVM | PDROP);
}
void
vm_object_pip_wait(vm_object_t object, const char *waitid)
{
VM_OBJECT_ASSERT_WLOCKED(object);
blockcount_wait(&object->paging_in_progress, &object->lock, waitid,
PVM);
}
void
vm_object_pip_wait_unlocked(vm_object_t object, const char *waitid)
{
VM_OBJECT_ASSERT_UNLOCKED(object);
blockcount_wait(&object->paging_in_progress, NULL, waitid, PVM);
}
/*
* vm_object_allocate:
*
* Returns a new object with the given size.
*/
vm_object_t
vm_object_allocate(objtype_t type, vm_pindex_t size)
{
vm_object_t object;
u_short flags;
switch (type) {
case OBJT_DEAD:
panic("vm_object_allocate: can't create OBJT_DEAD");
case OBJT_SWAP:
flags = OBJ_COLORED | OBJ_SWAP;
break;
case OBJT_DEVICE:
case OBJT_SG:
flags = OBJ_FICTITIOUS | OBJ_UNMANAGED;
break;
case OBJT_MGTDEVICE:
flags = OBJ_FICTITIOUS;
break;
case OBJT_PHYS:
flags = OBJ_UNMANAGED;
break;
case OBJT_VNODE:
flags = 0;
break;
default:
panic("vm_object_allocate: type %d is undefined or dynamic",
type);
}
object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
_vm_object_allocate(type, size, flags, object, NULL);
return (object);
}
vm_object_t
vm_object_allocate_dyn(objtype_t dyntype, vm_pindex_t size, u_short flags)
{
vm_object_t object;
MPASS(dyntype >= OBJT_FIRST_DYN /* && dyntype < nitems(pagertab) */);
object = (vm_object_t)uma_zalloc(obj_zone, M_WAITOK);
_vm_object_allocate(dyntype, size, flags, object, NULL);
return (object);
}
/*
* vm_object_allocate_anon:
*
* Returns a new default object of the given size and marked as
* anonymous memory for special split/collapse handling. Color
* to be initialized by the caller.
*/
vm_object_t
vm_object_allocate_anon(vm_pindex_t size, vm_object_t backing_object,
struct ucred *cred, vm_size_t charge)
{
vm_object_t handle, object;
if (backing_object == NULL)
handle = NULL;
else if ((backing_object->flags & OBJ_ANON) != 0)
handle = backing_object->handle;
else
handle = backing_object;
object = uma_zalloc(obj_zone, M_WAITOK);
_vm_object_allocate(OBJT_SWAP, size,
OBJ_ANON | OBJ_ONEMAPPING | OBJ_SWAP, object, handle);
object->cred = cred;
object->charge = cred != NULL ? charge : 0;
return (object);
}
static void
vm_object_reference_vnode(vm_object_t object)
{
u_int old;
/*
* vnode objects need the lock for the first reference
* to serialize with vnode_object_deallocate().
*/
if (!refcount_acquire_if_gt(&object->ref_count, 0)) {
VM_OBJECT_RLOCK(object);
old = refcount_acquire(&object->ref_count);
if (object->type == OBJT_VNODE && old == 0)
vref(object->handle);
VM_OBJECT_RUNLOCK(object);
}
}
/*
* vm_object_reference:
*
* Acquires a reference to the given object.
*/
void
vm_object_reference(vm_object_t object)
{
if (object == NULL)
return;
if (object->type == OBJT_VNODE)
vm_object_reference_vnode(object);
else
refcount_acquire(&object->ref_count);
KASSERT((object->flags & OBJ_DEAD) == 0,
("vm_object_reference: Referenced dead object."));
}
/*
* vm_object_reference_locked:
*
* Gets another reference to the given object.
*
* The object must be locked.
*/
void
vm_object_reference_locked(vm_object_t object)
{
u_int old;
VM_OBJECT_ASSERT_LOCKED(object);
old = refcount_acquire(&object->ref_count);
if (object->type == OBJT_VNODE && old == 0)
vref(object->handle);
KASSERT((object->flags & OBJ_DEAD) == 0,
("vm_object_reference: Referenced dead object."));
}
/*
* Handle deallocating an object of type OBJT_VNODE.
*/
static void
vm_object_deallocate_vnode(vm_object_t object)
{
struct vnode *vp = (struct vnode *) object->handle;
bool last;
KASSERT(object->type == OBJT_VNODE,
("vm_object_deallocate_vnode: not a vnode object"));
KASSERT(vp != NULL, ("vm_object_deallocate_vnode: missing vp"));
/* Object lock to protect handle lookup. */
last = refcount_release(&object->ref_count);
VM_OBJECT_RUNLOCK(object);
if (!last)
return;
if (!umtx_shm_vnobj_persistent)
umtx_shm_object_terminated(object);
/* vrele may need the vnode lock. */
vrele(vp);
}
/*
* We dropped a reference on an object and discovered that it had a
* single remaining shadow. This is a sibling of the reference we
* dropped. Attempt to collapse the sibling and backing object.
*/
static vm_object_t
vm_object_deallocate_anon(vm_object_t backing_object)
{
vm_object_t object;
/* Fetch the final shadow. */
object = LIST_FIRST(&backing_object->shadow_head);
KASSERT(object != NULL &&
atomic_load_int(&backing_object->shadow_count) == 1,
("vm_object_anon_deallocate: ref_count: %d, shadow_count: %d",
backing_object->ref_count,
atomic_load_int(&backing_object->shadow_count)));
KASSERT((object->flags & OBJ_ANON) != 0,
("invalid shadow object %p", object));
if (!VM_OBJECT_TRYWLOCK(object)) {
/*
* Prevent object from disappearing since we do not have a
* reference.
*/
vm_object_pip_add(object, 1);
VM_OBJECT_WUNLOCK(backing_object);
VM_OBJECT_WLOCK(object);
vm_object_pip_wakeup(object);
} else
VM_OBJECT_WUNLOCK(backing_object);
/*
* Check for a collapse/terminate race with the last reference holder.
*/
if ((object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) != 0 ||
!refcount_acquire_if_not_zero(&object->ref_count)) {
VM_OBJECT_WUNLOCK(object);
return (NULL);
}
backing_object = object->backing_object;
if (backing_object != NULL && (backing_object->flags & OBJ_ANON) != 0)
vm_object_collapse(object);
VM_OBJECT_WUNLOCK(object);
return (object);
}
/*
* vm_object_deallocate:
*
* Release a reference to the specified object,
* gained either through a vm_object_allocate
* or a vm_object_reference call. When all references
* are gone, storage associated with this object
* may be relinquished.
*
* No object may be locked.
*/
void
vm_object_deallocate(vm_object_t object)
{
vm_object_t temp;
bool released;
while (object != NULL) {
/*
* If the reference count goes to 0 we start calling
* vm_object_terminate() on the object chain. A ref count
* of 1 may be a special case depending on the shadow count
* being 0 or 1. These cases require a write lock on the
* object.
*/
if ((object->flags & OBJ_ANON) == 0)
released = refcount_release_if_gt(&object->ref_count, 1);
else
released = refcount_release_if_gt(&object->ref_count, 2);
if (released)
return;
if (object->type == OBJT_VNODE) {
VM_OBJECT_RLOCK(object);
if (object->type == OBJT_VNODE) {
vm_object_deallocate_vnode(object);
return;
}
VM_OBJECT_RUNLOCK(object);
}
VM_OBJECT_WLOCK(object);
KASSERT(object->ref_count > 0,
("vm_object_deallocate: object deallocated too many times: %d",
object->type));
/*
* If this is not the final reference to an anonymous
* object we may need to collapse the shadow chain.
*/
if (!refcount_release(&object->ref_count)) {
if (object->ref_count > 1 ||
atomic_load_int(&object->shadow_count) == 0) {
if ((object->flags & OBJ_ANON) != 0 &&
object->ref_count == 1)
vm_object_set_flag(object,
OBJ_ONEMAPPING);
VM_OBJECT_WUNLOCK(object);
return;
}
/* Handle collapsing last ref on anonymous objects. */
object = vm_object_deallocate_anon(object);
continue;
}
/*
* Handle the final reference to an object. We restart
* the loop with the backing object to avoid recursion.
*/
umtx_shm_object_terminated(object);
temp = object->backing_object;
if (temp != NULL) {
KASSERT(object->type == OBJT_SWAP,
("shadowed tmpfs v_object 2 %p", object));
vm_object_backing_remove(object);
}
KASSERT((object->flags & OBJ_DEAD) == 0,
("vm_object_deallocate: Terminating dead object."));
vm_object_set_flag(object, OBJ_DEAD);
vm_object_terminate(object);
object = temp;
}
}
void
vm_object_destroy(vm_object_t object)
{
uma_zfree(obj_zone, object);
}
static void
vm_object_sub_shadow(vm_object_t object)
{
KASSERT(object->shadow_count >= 1,
("object %p sub_shadow count zero", object));
atomic_subtract_int(&object->shadow_count, 1);
}
static void
vm_object_backing_remove_locked(vm_object_t object)
{
vm_object_t backing_object;
backing_object = object->backing_object;
VM_OBJECT_ASSERT_WLOCKED(object);
VM_OBJECT_ASSERT_WLOCKED(backing_object);
KASSERT((object->flags & OBJ_COLLAPSING) == 0,
("vm_object_backing_remove: Removing collapsing object."));
vm_object_sub_shadow(backing_object);
if ((object->flags & OBJ_SHADOWLIST) != 0) {
LIST_REMOVE(object, shadow_list);
vm_object_clear_flag(object, OBJ_SHADOWLIST);
}
object->backing_object = NULL;
}
static void
vm_object_backing_remove(vm_object_t object)
{
vm_object_t backing_object;
VM_OBJECT_ASSERT_WLOCKED(object);
backing_object = object->backing_object;
if ((object->flags & OBJ_SHADOWLIST) != 0) {
VM_OBJECT_WLOCK(backing_object);
vm_object_backing_remove_locked(object);
VM_OBJECT_WUNLOCK(backing_object);
} else {
object->backing_object = NULL;
vm_object_sub_shadow(backing_object);
}
}
static void
vm_object_backing_insert_locked(vm_object_t object, vm_object_t backing_object)
{
VM_OBJECT_ASSERT_WLOCKED(object);
atomic_add_int(&backing_object->shadow_count, 1);
if ((backing_object->flags & OBJ_ANON) != 0) {
VM_OBJECT_ASSERT_WLOCKED(backing_object);
LIST_INSERT_HEAD(&backing_object->shadow_head, object,
shadow_list);
vm_object_set_flag(object, OBJ_SHADOWLIST);
}
object->backing_object = backing_object;
}
static void
vm_object_backing_insert(vm_object_t object, vm_object_t backing_object)
{
VM_OBJECT_ASSERT_WLOCKED(object);
if ((backing_object->flags & OBJ_ANON) != 0) {
VM_OBJECT_WLOCK(backing_object);
vm_object_backing_insert_locked(object, backing_object);
VM_OBJECT_WUNLOCK(backing_object);
} else {
object->backing_object = backing_object;
atomic_add_int(&backing_object->shadow_count, 1);
}
}
/*
* Insert an object into a backing_object's shadow list with an additional
* reference to the backing_object added.
*/
static void
vm_object_backing_insert_ref(vm_object_t object, vm_object_t backing_object)
{
VM_OBJECT_ASSERT_WLOCKED(object);
if ((backing_object->flags & OBJ_ANON) != 0) {
VM_OBJECT_WLOCK(backing_object);
KASSERT((backing_object->flags & OBJ_DEAD) == 0,
("shadowing dead anonymous object"));
vm_object_reference_locked(backing_object);
vm_object_backing_insert_locked(object, backing_object);
vm_object_clear_flag(backing_object, OBJ_ONEMAPPING);
VM_OBJECT_WUNLOCK(backing_object);
} else {
vm_object_reference(backing_object);
atomic_add_int(&backing_object->shadow_count, 1);
object->backing_object = backing_object;
}
}
/*
* Transfer a backing reference from backing_object to object.
*/
static void
vm_object_backing_transfer(vm_object_t object, vm_object_t backing_object)
{
vm_object_t new_backing_object;
/*
* Note that the reference to backing_object->backing_object
* moves from within backing_object to within object.
*/
vm_object_backing_remove_locked(object);
new_backing_object = backing_object->backing_object;
if (new_backing_object == NULL)
return;
if ((new_backing_object->flags & OBJ_ANON) != 0) {
VM_OBJECT_WLOCK(new_backing_object);
vm_object_backing_remove_locked(backing_object);
vm_object_backing_insert_locked(object, new_backing_object);
VM_OBJECT_WUNLOCK(new_backing_object);
} else {
/*
* shadow_count for new_backing_object is left
* unchanged, its reference provided by backing_object
* is replaced by object.
*/
object->backing_object = new_backing_object;
backing_object->backing_object = NULL;
}
}
/*
* Wait for a concurrent collapse to settle.
*/
static void
vm_object_collapse_wait(vm_object_t object)
{
VM_OBJECT_ASSERT_WLOCKED(object);
while ((object->flags & OBJ_COLLAPSING) != 0) {
vm_object_pip_wait(object, "vmcolwait");
counter_u64_add(object_collapse_waits, 1);
}
}
/*
* Waits for a backing object to clear a pending collapse and returns
* it locked if it is an ANON object.
*/
static vm_object_t
vm_object_backing_collapse_wait(vm_object_t object)
{
vm_object_t backing_object;
VM_OBJECT_ASSERT_WLOCKED(object);
for (;;) {
backing_object = object->backing_object;
if (backing_object == NULL ||
(backing_object->flags & OBJ_ANON) == 0)
return (NULL);
VM_OBJECT_WLOCK(backing_object);
if ((backing_object->flags & (OBJ_DEAD | OBJ_COLLAPSING)) == 0)
break;
VM_OBJECT_WUNLOCK(object);
vm_object_pip_sleep(backing_object, "vmbckwait");
counter_u64_add(object_collapse_waits, 1);
VM_OBJECT_WLOCK(object);
}
return (backing_object);
}
/*
* vm_object_terminate_single_page removes a pageable page from the object,
* and removes it from the paging queues and frees it, if it is not wired.
* It is invoked via callback from vm_object_terminate_pages.
*/
static void
vm_object_terminate_single_page(vm_page_t p, void *objectv)
{
vm_object_t object __diagused = objectv;
vm_page_assert_unbusied(p);
KASSERT(p->object == object &&
(p->ref_count & VPRC_OBJREF) != 0,
("%s: page %p is inconsistent", __func__, p));
p->object = NULL;
if (vm_page_drop(p, VPRC_OBJREF) == VPRC_OBJREF) {
KASSERT((object->flags & OBJ_UNMANAGED) != 0 ||
vm_page_astate_load(p).queue != PQ_NONE,
("%s: page %p does not belong to a queue", __func__, p));
VM_CNT_INC(v_pfree);
vm_page_free(p);
}
}
/*
* vm_object_terminate_pages removes any remaining pageable pages
* from the object and resets the object to an empty state.
*/
static void
vm_object_terminate_pages(vm_object_t object)
{
VM_OBJECT_ASSERT_WLOCKED(object);
/*
* If the object contained any pages, then reset it to an empty state.
* Rather than incrementally removing each page from the object, the
* page and object are reset to any empty state.
*/
if (object->resident_page_count == 0)
return;
vm_radix_reclaim_callback(&object->rtree,
vm_object_terminate_single_page, object);
TAILQ_INIT(&object->memq);
object->resident_page_count = 0;
if (object->type == OBJT_VNODE)
vdrop(object->handle);
}
/*
* vm_object_terminate actually destroys the specified object, freeing
* up all previously used resources.
*
* The object must be locked.
* This routine may block.
*/
void
vm_object_terminate(vm_object_t object)
{
VM_OBJECT_ASSERT_WLOCKED(object);
KASSERT((object->flags & OBJ_DEAD) != 0,
("terminating non-dead obj %p", object));
KASSERT((object->flags & OBJ_COLLAPSING) == 0,
("terminating collapsing obj %p", object));
KASSERT(object->backing_object == NULL,
("terminating shadow obj %p", object));
/*
* Wait for the pageout daemon and other current users to be
* done with the object. Note that new paging_in_progress
* users can come after this wait, but they must check
* OBJ_DEAD flag set (without unlocking the object), and avoid
* the object being terminated.
*/
vm_object_pip_wait(object, "objtrm");
KASSERT(object->ref_count == 0,
("vm_object_terminate: object with references, ref_count=%d",
object->ref_count));
if ((object->flags & OBJ_PG_DTOR) == 0)
vm_object_terminate_pages(object);
#if VM_NRESERVLEVEL > 0
if (__predict_false(!LIST_EMPTY(&object->rvq)))
vm_reserv_break_all(object);
#endif
KASSERT(object->cred == NULL || (object->flags & OBJ_SWAP) != 0,
("%s: non-swap obj %p has cred", __func__, object));
/*
* Let the pager know object is dead.
*/
vm_pager_deallocate(object);
VM_OBJECT_WUNLOCK(object);
vm_object_destroy(object);
}
/*
* Make the page read-only so that we can clear the object flags. However, if
* this is a nosync mmap then the object is likely to stay dirty so do not
* mess with the page and do not clear the object flags. Returns TRUE if the
* page should be flushed, and FALSE otherwise.
*/
static boolean_t
vm_object_page_remove_write(vm_page_t p, int flags, boolean_t *allclean)
{
vm_page_assert_busied(p);
/*
* If we have been asked to skip nosync pages and this is a
* nosync page, skip it. Note that the object flags were not
* cleared in this case so we do not have to set them.
*/
if ((flags & OBJPC_NOSYNC) != 0 && (p->a.flags & PGA_NOSYNC) != 0) {
*allclean = FALSE;
return (FALSE);
} else {
pmap_remove_write(p);
return (p->dirty != 0);
}
}
/*
* vm_object_page_clean
*
* Clean all dirty pages in the specified range of object. Leaves page
* on whatever queue it is currently on. If NOSYNC is set then do not
* write out pages with PGA_NOSYNC set (originally comes from MAP_NOSYNC),
* leaving the object dirty.
*
* For swap objects backing tmpfs regular files, do not flush anything,
* but remove write protection on the mapped pages to update mtime through
* mmaped writes.
*
* When stuffing pages asynchronously, allow clustering. XXX we need a
* synchronous clustering mode implementation.
*
* Odd semantics: if start == end, we clean everything.
*
* The object must be locked.
*
* Returns FALSE if some page from the range was not written, as
* reported by the pager, and TRUE otherwise.
*/
boolean_t
vm_object_page_clean(vm_object_t object, vm_ooffset_t start, vm_ooffset_t end,
int flags)
{
vm_page_t np, p;
vm_pindex_t pi, tend, tstart;
int curgeneration, n, pagerflags;
boolean_t eio, res, allclean;
VM_OBJECT_ASSERT_WLOCKED(object);
if (!vm_object_mightbedirty(object) || object->resident_page_count == 0)
return (TRUE);
pagerflags = (flags & (OBJPC_SYNC | OBJPC_INVAL)) != 0 ?
VM_PAGER_PUT_SYNC : VM_PAGER_CLUSTER_OK;
pagerflags |= (flags & OBJPC_INVAL) != 0 ? VM_PAGER_PUT_INVAL : 0;
tstart = OFF_TO_IDX(start);
tend = (end == 0) ? object->size : OFF_TO_IDX(end + PAGE_MASK);
allclean = tstart == 0 && tend >= object->size;
res = TRUE;
rescan:
curgeneration = object->generation;
for (p = vm_page_find_least(object, tstart); p != NULL; p = np) {
pi = p->pindex;
if (pi >= tend)
break;
np = TAILQ_NEXT(p, listq);
if (vm_page_none_valid(p))
continue;
if (vm_page_busy_acquire(p, VM_ALLOC_WAITFAIL) == 0) {
if (object->generation != curgeneration &&
(flags & OBJPC_SYNC) != 0)
goto rescan;
np = vm_page_find_least(object, pi);
continue;
}
if (!vm_object_page_remove_write(p, flags, &allclean)) {
vm_page_xunbusy(p);
continue;
}
if (object->type == OBJT_VNODE) {
n = vm_object_page_collect_flush(object, p, pagerflags,
flags, &allclean, &eio);
if (eio) {
res = FALSE;
allclean = FALSE;
}
if (object->generation != curgeneration &&
(flags & OBJPC_SYNC) != 0)
goto rescan;
/*
* If the VOP_PUTPAGES() did a truncated write, so
* that even the first page of the run is not fully
* written, vm_pageout_flush() returns 0 as the run
* length. Since the condition that caused truncated
* write may be permanent, e.g. exhausted free space,
* accepting n == 0 would cause an infinite loop.
*
* Forwarding the iterator leaves the unwritten page
* behind, but there is not much we can do there if
* filesystem refuses to write it.
*/
if (n == 0) {
n = 1;
allclean = FALSE;
}
} else {
n = 1;
vm_page_xunbusy(p);
}
np = vm_page_find_least(object, pi + n);
}
#if 0
VOP_FSYNC(vp, (pagerflags & VM_PAGER_PUT_SYNC) ? MNT_WAIT : 0);
#endif
/*
* Leave updating cleangeneration for tmpfs objects to tmpfs
* scan. It needs to update mtime, which happens for other
* filesystems during page writeouts.
*/
if (allclean && object->type == OBJT_VNODE)
object->cleangeneration = curgeneration;
return (res);
}
static int
vm_object_page_collect_flush(vm_object_t object, vm_page_t p, int pagerflags,
int flags, boolean_t *allclean, boolean_t *eio)
{
vm_page_t ma[2 * vm_pageout_page_count - 1], tp;
int base, count, runlen;
vm_page_lock_assert(p, MA_NOTOWNED);
vm_page_assert_xbusied(p);
VM_OBJECT_ASSERT_WLOCKED(object);
base = nitems(ma) / 2;
ma[base] = p;
for (count = 1, tp = p; count < vm_pageout_page_count; count++) {
tp = vm_page_next(tp);
if (tp == NULL || vm_page_tryxbusy(tp) == 0)
break;
if (!vm_object_page_remove_write(tp, flags, allclean)) {
vm_page_xunbusy(tp);
break;
}
ma[base + count] = tp;
}
for (tp = p; count < vm_pageout_page_count; count++) {
tp = vm_page_prev(tp);
if (tp == NULL || vm_page_tryxbusy(tp) == 0)
break;
if (!vm_object_page_remove_write(tp, flags, allclean)) {
vm_page_xunbusy(tp);
break;
}
ma[--base] = tp;
}
vm_pageout_flush(&ma[base], count, pagerflags, nitems(ma) / 2 - base,
&runlen, eio);
return (runlen);
}
/*
* Note that there is absolutely no sense in writing out
* anonymous objects, so we track down the vnode object
* to write out.
* We invalidate (remove) all pages from the address space
* for semantic correctness.
*
* If the backing object is a device object with unmanaged pages, then any
* mappings to the specified range of pages must be removed before this
* function is called.
*
* Note: certain anonymous maps, such as MAP_NOSYNC maps,
* may start out with a NULL object.
*/
boolean_t
vm_object_sync(vm_object_t object, vm_ooffset_t offset, vm_size_t size,
boolean_t syncio, boolean_t invalidate)
{
vm_object_t backing_object;
struct vnode *vp;
struct mount *mp;
int error, flags, fsync_after;
boolean_t res;
if (object == NULL)
return (TRUE);
res = TRUE;
error = 0;
VM_OBJECT_WLOCK(object);
while ((backing_object = object->backing_object) != NULL) {
VM_OBJECT_WLOCK(backing_object);
offset += object->backing_object_offset;
VM_OBJECT_WUNLOCK(object);
object = backing_object;
if (object->size < OFF_TO_IDX(offset + size))
size = IDX_TO_OFF(object->size) - offset;
}
/*
* Flush pages if writing is allowed, invalidate them
* if invalidation requested. Pages undergoing I/O
* will be ignored by vm_object_page_remove().
*
* We cannot lock the vnode and then wait for paging
* to complete without deadlocking against vm_fault.
* Instead we simply call vm_object_page_remove() and
* allow it to block internally on a page-by-page
* basis when it encounters pages undergoing async
* I/O.
*/
if (object->type == OBJT_VNODE &&
vm_object_mightbedirty(object) != 0 &&
((vp = object->handle)->v_vflag & VV_NOSYNC) == 0) {
VM_OBJECT_WUNLOCK(object);
(void)vn_start_write(vp, &mp, V_WAIT);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
if (syncio && !invalidate && offset == 0 &&
atop(size) == object->size) {
/*
* If syncing the whole mapping of the file,
* it is faster to schedule all the writes in
* async mode, also allowing the clustering,
* and then wait for i/o to complete.
*/
flags = 0;
fsync_after = TRUE;
} else {
flags = (syncio || invalidate) ? OBJPC_SYNC : 0;
flags |= invalidate ? (OBJPC_SYNC | OBJPC_INVAL) : 0;
fsync_after = FALSE;
}
VM_OBJECT_WLOCK(object);
res = vm_object_page_clean(object, offset, offset + size,
flags);
VM_OBJECT_WUNLOCK(object);
if (fsync_after) {
for (;;) {
error = VOP_FSYNC(vp, MNT_WAIT, curthread);
if (error != ERELOOKUP)
break;
/*
* Allow SU/bufdaemon to handle more
* dependencies in the meantime.
*/
VOP_UNLOCK(vp);
vn_finished_write(mp);
(void)vn_start_write(vp, &mp, V_WAIT);
vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
}
}
VOP_UNLOCK(vp);
vn_finished_write(mp);
if (error != 0)
res = FALSE;
VM_OBJECT_WLOCK(object);
}
if ((object->type == OBJT_VNODE ||
object->type == OBJT_DEVICE) && invalidate) {
if (object->type == OBJT_DEVICE)
/*
* The option OBJPR_NOTMAPPED must be passed here
* because vm_object_page_remove() cannot remove
* unmanaged mappings.
*/
flags = OBJPR_NOTMAPPED;
else if (old_msync)
flags = 0;
else
flags = OBJPR_CLEANONLY;
vm_object_page_remove(object, OFF_TO_IDX(offset),
OFF_TO_IDX(offset + size + PAGE_MASK), flags);
}
VM_OBJECT_WUNLOCK(object);
return (res);
}
/*
* Determine whether the given advice can be applied to the object. Advice is
* not applied to unmanaged pages since they never belong to page queues, and
* since MADV_FREE is destructive, it can apply only to anonymous pages that
* have been mapped at most once.
*/
static bool
vm_object_advice_applies(vm_object_t object, int advice)
{
if ((object->flags & OBJ_UNMANAGED) != 0)
return (false);
if (advice != MADV_FREE)
return (true);
return ((object->flags & (OBJ_ONEMAPPING | OBJ_ANON)) ==
(OBJ_ONEMAPPING | OBJ_ANON));
}
static void
vm_object_madvise_freespace(vm_object_t object, int advice, vm_pindex_t pindex,
vm_size_t size)
{
if (advice == MADV_FREE)
vm_pager_freespace(object, pindex, size);
}
/*
* vm_object_madvise:
*
* Implements the madvise function at the object/page level.
*
* MADV_WILLNEED (any object)
*
* Activate the specified pages if they are resident.
*
* MADV_DONTNEED (any object)
*
* Deactivate the specified pages if they are resident.
*
* MADV_FREE (OBJT_SWAP objects, OBJ_ONEMAPPING only)
*
* Deactivate and clean the specified pages if they are
* resident. This permits the process to reuse the pages
* without faulting or the kernel to reclaim the pages
* without I/O.
*/
void
vm_object_madvise(vm_object_t object, vm_pindex_t pindex, vm_pindex_t end,
int advice)
{
vm_pindex_t tpindex;
vm_object_t backing_object, tobject;
vm_page_t m, tm;
if (object == NULL)
return;
relookup:
VM_OBJECT_WLOCK(object);
if (!vm_object_advice_applies(object, advice)) {
VM_OBJECT_WUNLOCK(object);
return;
}
for (m = vm_page_find_least(object, pindex); pindex < end; pindex++) {
tobject = object;
/*
* If the next page isn't resident in the top-level object, we
* need to search the shadow chain. When applying MADV_FREE, we
* take care to release any swap space used to store
* non-resident pages.
*/
if (m == NULL || pindex < m->pindex) {
/*
* Optimize a common case: if the top-level object has
* no backing object, we can skip over the non-resident
* range in constant time.
*/
if (object->backing_object == NULL) {
tpindex = (m != NULL && m->pindex < end) ?
m->pindex : end;
vm_object_madvise_freespace(object, advice,
pindex, tpindex - pindex);
if ((pindex = tpindex) == end)
break;
goto next_page;
}
tpindex = pindex;
do {
vm_object_madvise_freespace(tobject, advice,
tpindex, 1);
/*
* Prepare to search the next object in the
* chain.
*/
backing_object = tobject->backing_object;
if (backing_object == NULL)
goto next_pindex;
VM_OBJECT_WLOCK(backing_object);
tpindex +=
OFF_TO_IDX(tobject->backing_object_offset);
if (tobject != object)
VM_OBJECT_WUNLOCK(tobject);
tobject = backing_object;
if (!vm_object_advice_applies(tobject, advice))
goto next_pindex;
} while ((tm = vm_page_lookup(tobject, tpindex)) ==
NULL);
} else {
next_page:
tm = m;
m = TAILQ_NEXT(m, listq);
}
/*
* If the page is not in a normal state, skip it. The page
* can not be invalidated while the object lock is held.
*/
if (!vm_page_all_valid(tm) || vm_page_wired(tm))
goto next_pindex;
KASSERT((tm->flags & PG_FICTITIOUS) == 0,
("vm_object_madvise: page %p is fictitious", tm));
KASSERT((tm->oflags & VPO_UNMANAGED) == 0,
("vm_object_madvise: page %p is not managed", tm));
if (vm_page_tryxbusy(tm) == 0) {
if (object != tobject)
VM_OBJECT_WUNLOCK(object);
if (advice == MADV_WILLNEED) {
/*
* Reference the page before unlocking and
* sleeping so that the page daemon is less
* likely to reclaim it.
*/
vm_page_aflag_set(tm, PGA_REFERENCED);
}
if (!vm_page_busy_sleep(tm, "madvpo", 0))
VM_OBJECT_WUNLOCK(tobject);
goto relookup;
}
vm_page_advise(tm, advice);
vm_page_xunbusy(tm);
vm_object_madvise_freespace(tobject, advice, tm->pindex, 1);
next_pindex:
if (tobject != object)
VM_OBJECT_WUNLOCK(tobject);
}
VM_OBJECT_WUNLOCK(object);
}
/*
* vm_object_shadow:
*
* Create a new object which is backed by the
* specified existing object range. The source
* object reference is deallocated.
*
* The new object and offset into that object
* are returned in the source parameters.
*/
void
vm_object_shadow(vm_object_t *object, vm_ooffset_t *offset, vm_size_t length,
struct ucred *cred, bool shared)
{
vm_object_t source;
vm_object_t result;
source = *object;
/*
* Don't create the new object if the old object isn't shared.
*
* If we hold the only reference we can guarantee that it won't
* increase while we have the map locked. Otherwise the race is
* harmless and we will end up with an extra shadow object that
* will be collapsed later.
*/
if (source != NULL && source->ref_count == 1 &&
(source->flags & OBJ_ANON) != 0)
return;
/*
* Allocate a new object with the given length.
*/
result = vm_object_allocate_anon(atop(length), source, cred, length);
/*
* Store the offset into the source object, and fix up the offset into
* the new object.
*/
result->backing_object_offset = *offset;
if (shared || source != NULL) {
VM_OBJECT_WLOCK(result);
/*
* The new object shadows the source object, adding a
* reference to it. Our caller changes his reference
* to point to the new object, removing a reference to
* the source object. Net result: no change of
* reference count, unless the caller needs to add one
* more reference due to forking a shared map entry.
*/
if (shared) {
vm_object_reference_locked(result);
vm_object_clear_flag(result, OBJ_ONEMAPPING);
}
/*
* Try to optimize the result object's page color when
* shadowing in order to maintain page coloring
* consistency in the combined shadowed object.
*/
if (source != NULL) {
vm_object_backing_insert(result, source);
result->domain = source->domain;
#if VM_NRESERVLEVEL > 0
vm_object_set_flag(result,
(source->flags & OBJ_COLORED));
result->pg_color = (source->pg_color +
OFF_TO_IDX(*offset)) & ((1 << (VM_NFREEORDER -
1)) - 1);
#endif
}
VM_OBJECT_WUNLOCK(result);
}
/*
* Return the new things
*/
*offset = 0;
*object = result;
}
/*
* vm_object_split:
*
* Split the pages in a map entry into a new object. This affords
* easier removal of unused pages, and keeps object inheritance from
* being a negative impact on memory usage.
*/
void
vm_object_split(vm_map_entry_t entry)
{
- vm_page_t m, m_next;
+ struct pctrie_iter pages;
+ vm_page_t m;
vm_object_t orig_object, new_object, backing_object;
- vm_pindex_t idx, offidxstart;
+ vm_pindex_t offidxstart;
vm_size_t size;
orig_object = entry->object.vm_object;
KASSERT((orig_object->flags & OBJ_ONEMAPPING) != 0,
("vm_object_split: Splitting object with multiple mappings."));
if ((orig_object->flags & OBJ_ANON) == 0)
return;
if (orig_object->ref_count <= 1)
return;
VM_OBJECT_WUNLOCK(orig_object);
offidxstart = OFF_TO_IDX(entry->offset);
size = atop(entry->end - entry->start);
new_object = vm_object_allocate_anon(size, orig_object,
orig_object->cred, ptoa(size));
/*
* We must wait for the orig_object to complete any in-progress
* collapse so that the swap blocks are stable below. The
* additional reference on backing_object by new object will
* prevent further collapse operations until split completes.
*/
VM_OBJECT_WLOCK(orig_object);
vm_object_collapse_wait(orig_object);
/*
* At this point, the new object is still private, so the order in
* which the original and new objects are locked does not matter.
*/
VM_OBJECT_WLOCK(new_object);
new_object->domain = orig_object->domain;
backing_object = orig_object->backing_object;
if (backing_object != NULL) {
vm_object_backing_insert_ref(new_object, backing_object);
new_object->backing_object_offset =
orig_object->backing_object_offset + entry->offset;
}
if (orig_object->cred != NULL) {
crhold(orig_object->cred);
KASSERT(orig_object->charge >= ptoa(size),
("orig_object->charge < 0"));
orig_object->charge -= ptoa(size);
}
/*
* Mark the split operation so that swap_pager_getpages() knows
* that the object is in transition.
*/
vm_object_set_flag(orig_object, OBJ_SPLIT);
-#ifdef INVARIANTS
- idx = 0;
-#endif
+ vm_page_iter_limit_init(&pages, orig_object, offidxstart + size);
retry:
- m = vm_page_find_least(orig_object, offidxstart);
- KASSERT(m == NULL || idx <= m->pindex - offidxstart,
- ("%s: object %p was repopulated", __func__, orig_object));
- for (; m != NULL && (idx = m->pindex - offidxstart) < size;
- m = m_next) {
- m_next = TAILQ_NEXT(m, listq);
-
+ pctrie_iter_reset(&pages);
+ for (m = vm_page_iter_lookup_ge(&pages, offidxstart); m != NULL;
+ m = vm_radix_iter_step(&pages)) {
/*
* We must wait for pending I/O to complete before we can
* rename the page.
*
* We do not have to VM_PROT_NONE the page as mappings should
* not be changed by this operation.
*/
if (vm_page_tryxbusy(m) == 0) {
VM_OBJECT_WUNLOCK(new_object);
if (vm_page_busy_sleep(m, "spltwt", 0))
VM_OBJECT_WLOCK(orig_object);
VM_OBJECT_WLOCK(new_object);
goto retry;
}
/*
* The page was left invalid. Likely placed there by
* an incomplete fault. Just remove and ignore.
*/
if (vm_page_none_valid(m)) {
- if (vm_page_remove(m))
+ if (vm_page_iter_remove(&pages))
vm_page_free(m);
continue;
}
/* vm_page_rename() will dirty the page. */
- if (vm_page_rename(m, new_object, idx)) {
+ if (vm_page_rename(&pages, new_object, m->pindex - offidxstart)) {
vm_page_xunbusy(m);
VM_OBJECT_WUNLOCK(new_object);
VM_OBJECT_WUNLOCK(orig_object);
vm_radix_wait();
VM_OBJECT_WLOCK(orig_object);
VM_OBJECT_WLOCK(new_object);
goto retry;
}
#if VM_NRESERVLEVEL > 0
/*
* If some of the reservation's allocated pages remain with
* the original object, then transferring the reservation to
* the new object is neither particularly beneficial nor
* particularly harmful as compared to leaving the reservation
* with the original object. If, however, all of the
* reservation's allocated pages are transferred to the new
* object, then transferring the reservation is typically
* beneficial. Determining which of these two cases applies
* would be more costly than unconditionally renaming the
* reservation.
*/
vm_reserv_rename(m, new_object, orig_object, offidxstart);
#endif
}
/*
* swap_pager_copy() can sleep, in which case the orig_object's
* and new_object's locks are released and reacquired.
*/
swap_pager_copy(orig_object, new_object, offidxstart, 0);
TAILQ_FOREACH(m, &new_object->memq, listq)
vm_page_xunbusy(m);
vm_object_clear_flag(orig_object, OBJ_SPLIT);
VM_OBJECT_WUNLOCK(orig_object);
VM_OBJECT_WUNLOCK(new_object);
entry->object.vm_object = new_object;
entry->offset = 0LL;
vm_object_deallocate(orig_object);
VM_OBJECT_WLOCK(new_object);
}
static vm_page_t
-vm_object_collapse_scan_wait(vm_object_t object, vm_page_t p)
+vm_object_collapse_scan_wait(struct pctrie_iter *pages, vm_object_t object,
+ vm_page_t p)
{
vm_object_t backing_object;
VM_OBJECT_ASSERT_WLOCKED(object);
backing_object = object->backing_object;
VM_OBJECT_ASSERT_WLOCKED(backing_object);
KASSERT(p == NULL || p->object == object || p->object == backing_object,
("invalid ownership %p %p %p", p, object, backing_object));
/* The page is only NULL when rename fails. */
if (p == NULL) {
VM_OBJECT_WUNLOCK(object);
VM_OBJECT_WUNLOCK(backing_object);
vm_radix_wait();
VM_OBJECT_WLOCK(object);
} else if (p->object == object) {
VM_OBJECT_WUNLOCK(backing_object);
if (vm_page_busy_sleep(p, "vmocol", 0))
VM_OBJECT_WLOCK(object);
} else {
VM_OBJECT_WUNLOCK(object);
if (!vm_page_busy_sleep(p, "vmocol", 0))
VM_OBJECT_WUNLOCK(backing_object);
VM_OBJECT_WLOCK(object);
}
VM_OBJECT_WLOCK(backing_object);
- return (TAILQ_FIRST(&backing_object->memq));
+ vm_page_iter_init(pages, backing_object);
+ return (vm_page_iter_lookup_ge(pages, 0));
}
static void
vm_object_collapse_scan(vm_object_t object)
{
+ struct pctrie_iter pages;
vm_object_t backing_object;
vm_page_t next, p, pp;
vm_pindex_t backing_offset_index, new_pindex;
VM_OBJECT_ASSERT_WLOCKED(object);
VM_OBJECT_ASSERT_WLOCKED(object->backing_object);
backing_object = object->backing_object;
backing_offset_index = OFF_TO_IDX(object->backing_object_offset);
/*
* Our scan
*/
- for (p = TAILQ_FIRST(&backing_object->memq); p != NULL; p = next) {
+ vm_page_iter_init(&pages, backing_object);
+ for (p = vm_page_iter_lookup_ge(&pages, 0); p != NULL; p = next) {
next = TAILQ_NEXT(p, listq);
new_pindex = p->pindex - backing_offset_index;
/*
* Check for busy page
*/
if (vm_page_tryxbusy(p) == 0) {
- next = vm_object_collapse_scan_wait(object, p);
+ next = vm_object_collapse_scan_wait(&pages, object, p);
continue;
}
KASSERT(object->backing_object == backing_object,
("vm_object_collapse_scan: backing object mismatch %p != %p",
object->backing_object, backing_object));
KASSERT(p->object == backing_object,
("vm_object_collapse_scan: object mismatch %p != %p",
p->object, backing_object));
if (p->pindex < backing_offset_index ||
new_pindex >= object->size) {
vm_pager_freespace(backing_object, p->pindex, 1);
KASSERT(!pmap_page_is_mapped(p),
("freeing mapped page %p", p));
- if (vm_page_remove(p))
+ if (vm_page_iter_remove(&pages))
vm_page_free(p);
+ next = vm_radix_iter_step(&pages);
continue;
}
if (!vm_page_all_valid(p)) {
KASSERT(!pmap_page_is_mapped(p),
("freeing mapped page %p", p));
- if (vm_page_remove(p))
+ if (vm_page_iter_remove(&pages))
vm_page_free(p);
+ next = vm_radix_iter_step(&pages);
continue;
}
pp = vm_page_lookup(object, new_pindex);
if (pp != NULL && vm_page_tryxbusy(pp) == 0) {
vm_page_xunbusy(p);
/*
* The page in the parent is busy and possibly not
* (yet) valid. Until its state is finalized by the
* busy bit owner, we can't tell whether it shadows the
* original page.
*/
- next = vm_object_collapse_scan_wait(object, pp);
+ next = vm_object_collapse_scan_wait(&pages, object, pp);
continue;
}
if (pp != NULL && vm_page_none_valid(pp)) {
/*
* The page was invalid in the parent. Likely placed
* there by an incomplete fault. Just remove and
* ignore. p can replace it.
*/
if (vm_page_remove(pp))
vm_page_free(pp);
pp = NULL;
}
if (pp != NULL || vm_pager_has_page(object, new_pindex, NULL,
NULL)) {
/*
* The page already exists in the parent OR swap exists
* for this location in the parent. Leave the parent's
* page alone. Destroy the original page from the
* backing object.
*/
vm_pager_freespace(backing_object, p->pindex, 1);
KASSERT(!pmap_page_is_mapped(p),
("freeing mapped page %p", p));
- if (vm_page_remove(p))
- vm_page_free(p);
if (pp != NULL)
vm_page_xunbusy(pp);
+ if (vm_page_iter_remove(&pages))
+ vm_page_free(p);
+ next = vm_radix_iter_step(&pages);
continue;
}
/*
* Page does not exist in parent, rename the page from the
* backing object to the main object.
*
* If the page was mapped to a process, it can remain mapped
* through the rename. vm_page_rename() will dirty the page.
*/
- if (vm_page_rename(p, object, new_pindex)) {
+ if (vm_page_rename(&pages, object, new_pindex)) {
vm_page_xunbusy(p);
- next = vm_object_collapse_scan_wait(object, NULL);
+ next = vm_object_collapse_scan_wait(&pages, object,
+ NULL);
continue;
}
/* Use the old pindex to free the right page. */
vm_pager_freespace(backing_object, new_pindex +
backing_offset_index, 1);
#if VM_NRESERVLEVEL > 0
/*
* Rename the reservation.
*/
vm_reserv_rename(p, object, backing_object,
backing_offset_index);
#endif
vm_page_xunbusy(p);
+ next = vm_radix_iter_step(&pages);
}
return;
}
/*
* vm_object_collapse:
*
* Collapse an object with the object backing it.
* Pages in the backing object are moved into the
* parent, and the backing object is deallocated.
*/
void
vm_object_collapse(vm_object_t object)
{
vm_object_t backing_object, new_backing_object;
VM_OBJECT_ASSERT_WLOCKED(object);
while (TRUE) {
KASSERT((object->flags & (OBJ_DEAD | OBJ_ANON)) == OBJ_ANON,
("collapsing invalid object"));
/*
* Wait for the backing_object to finish any pending
* collapse so that the caller sees the shortest possible
* shadow chain.
*/
backing_object = vm_object_backing_collapse_wait(object);
if (backing_object == NULL)
return;
KASSERT(object->ref_count > 0 &&
object->ref_count > atomic_load_int(&object->shadow_count),
("collapse with invalid ref %d or shadow %d count.",
object->ref_count, atomic_load_int(&object->shadow_count)));
KASSERT((backing_object->flags &
(OBJ_COLLAPSING | OBJ_DEAD)) == 0,
("vm_object_collapse: Backing object already collapsing."));
KASSERT((object->flags & (OBJ_COLLAPSING | OBJ_DEAD)) == 0,
("vm_object_collapse: object is already collapsing."));
/*
* We know that we can either collapse the backing object if
* the parent is the only reference to it, or (perhaps) have
* the parent bypass the object if the parent happens to shadow
* all the resident pages in the entire backing object.
*/
if (backing_object->ref_count == 1) {
KASSERT(atomic_load_int(&backing_object->shadow_count)
== 1,
("vm_object_collapse: shadow_count: %d",
atomic_load_int(&backing_object->shadow_count)));
vm_object_pip_add(object, 1);
vm_object_set_flag(object, OBJ_COLLAPSING);
vm_object_pip_add(backing_object, 1);
vm_object_set_flag(backing_object, OBJ_DEAD);
/*
* If there is exactly one reference to the backing
* object, we can collapse it into the parent.
*/
vm_object_collapse_scan(object);
/*
* Move the pager from backing_object to object.
*
* swap_pager_copy() can sleep, in which case the
* backing_object's and object's locks are released and
* reacquired.
*/
swap_pager_copy(backing_object, object,
OFF_TO_IDX(object->backing_object_offset), TRUE);
/*
* Object now shadows whatever backing_object did.
*/
vm_object_clear_flag(object, OBJ_COLLAPSING);
vm_object_backing_transfer(object, backing_object);
object->backing_object_offset +=
backing_object->backing_object_offset;
VM_OBJECT_WUNLOCK(object);
vm_object_pip_wakeup(object);
/*
* Discard backing_object.
*
* Since the backing object has no pages, no pager left,
* and no object references within it, all that is
* necessary is to dispose of it.
*/
KASSERT(backing_object->ref_count == 1, (
"backing_object %p was somehow re-referenced during collapse!",
backing_object));
vm_object_pip_wakeup(backing_object);
(void)refcount_release(&backing_object->ref_count);
umtx_shm_object_terminated(backing_object);
vm_object_terminate(backing_object);
counter_u64_add(object_collapses, 1);
VM_OBJECT_WLOCK(object);
} else {
/*
* If we do not entirely shadow the backing object,
* there is nothing we can do so we give up.
*
* The object lock and backing_object lock must not
* be dropped during this sequence.
*/
if (!swap_pager_scan_all_shadowed(object)) {
VM_OBJECT_WUNLOCK(backing_object);
break;
}
/*
* Make the parent shadow the next object in the
* chain. Deallocating backing_object will not remove
* it, since its reference count is at least 2.
*/
vm_object_backing_remove_locked(object);
new_backing_object = backing_object->backing_object;
if (new_backing_object != NULL) {
vm_object_backing_insert_ref(object,
new_backing_object);
object->backing_object_offset +=
backing_object->backing_object_offset;
}
/*
* Drop the reference count on backing_object. Since
* its ref_count was at least 2, it will not vanish.
*/
(void)refcount_release(&backing_object->ref_count);
KASSERT(backing_object->ref_count >= 1, (
"backing_object %p was somehow dereferenced during collapse!",
backing_object));
VM_OBJECT_WUNLOCK(backing_object);
counter_u64_add(object_bypasses, 1);
}
/*
* Try again with this object's new backing object.
*/
}
}
/*
* vm_object_page_remove:
*
* For the given object, either frees or invalidates each of the
* specified pages. In general, a page is freed. However, if a page is
* wired for any reason other than the existence of a managed, wired
* mapping, then it may be invalidated but not removed from the object.
* Pages are specified by the given range ["start", "end") and the option
* OBJPR_CLEANONLY. As a special case, if "end" is zero, then the range
* extends from "start" to the end of the object. If the option
* OBJPR_CLEANONLY is specified, then only the non-dirty pages within the
* specified range are affected. If the option OBJPR_NOTMAPPED is
* specified, then the pages within the specified range must have no
* mappings. Otherwise, if this option is not specified, any mappings to
* the specified pages are removed before the pages are freed or
* invalidated.
*
* In general, this operation should only be performed on objects that
* contain managed pages. There are, however, two exceptions. First, it
* is performed on the kernel and kmem objects by vm_map_entry_delete().
* Second, it is used by msync(..., MS_INVALIDATE) to invalidate device-
* backed pages. In both of these cases, the option OBJPR_CLEANONLY must
* not be specified and the option OBJPR_NOTMAPPED must be specified.
*
* The object must be locked.
*/
void
vm_object_page_remove(vm_object_t object, vm_pindex_t start, vm_pindex_t end,
int options)
{
- vm_page_t p, next;
+ struct pctrie_iter pages;
+ vm_page_t p;
VM_OBJECT_ASSERT_WLOCKED(object);
KASSERT((object->flags & OBJ_UNMANAGED) == 0 ||
(options & (OBJPR_CLEANONLY | OBJPR_NOTMAPPED)) == OBJPR_NOTMAPPED,
("vm_object_page_remove: illegal options for object %p", object));
if (object->resident_page_count == 0)
return;
vm_object_pip_add(object, 1);
+ vm_page_iter_limit_init(&pages, object, end);
again:
- p = vm_page_find_least(object, start);
-
- /*
- * Here, the variable "p" is either (1) the page with the least pindex
- * greater than or equal to the parameter "start" or (2) NULL.
- */
- for (; p != NULL && (p->pindex < end || end == 0); p = next) {
- next = TAILQ_NEXT(p, listq);
-
+ pctrie_iter_reset(&pages);
+ for (p = vm_page_iter_lookup_ge(&pages, start); p != NULL;
+ p = vm_radix_iter_step(&pages)) {
/*
* Skip invalid pages if asked to do so. Try to avoid acquiring
* the busy lock, as some consumers rely on this to avoid
* deadlocks.
*
* A thread may concurrently transition the page from invalid to
* valid using only the busy lock, so the result of this check
* is immediately stale. It is up to consumers to handle this,
* for instance by ensuring that all invalid->valid transitions
* happen with a mutex held, as may be possible for a
* filesystem.
*/
if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p))
continue;
/*
* If the page is wired for any reason besides the existence
* of managed, wired mappings, then it cannot be freed. For
* example, fictitious pages, which represent device memory,
* are inherently wired and cannot be freed. They can,
* however, be invalidated if the option OBJPR_CLEANONLY is
* not specified.
*/
if (vm_page_tryxbusy(p) == 0) {
if (vm_page_busy_sleep(p, "vmopar", 0))
VM_OBJECT_WLOCK(object);
goto again;
}
if ((options & OBJPR_VALIDONLY) != 0 && vm_page_none_valid(p)) {
vm_page_xunbusy(p);
continue;
}
if (vm_page_wired(p)) {
wired:
if ((options & OBJPR_NOTMAPPED) == 0 &&
object->ref_count != 0)
pmap_remove_all(p);
if ((options & OBJPR_CLEANONLY) == 0) {
vm_page_invalid(p);
vm_page_undirty(p);
}
vm_page_xunbusy(p);
continue;
}
KASSERT((p->flags & PG_FICTITIOUS) == 0,
("vm_object_page_remove: page %p is fictitious", p));
if ((options & OBJPR_CLEANONLY) != 0 &&
!vm_page_none_valid(p)) {
if ((options & OBJPR_NOTMAPPED) == 0 &&
object->ref_count != 0 &&
!vm_page_try_remove_write(p))
goto wired;
if (p->dirty != 0) {
vm_page_xunbusy(p);
continue;
}
}
if ((options & OBJPR_NOTMAPPED) == 0 &&
object->ref_count != 0 && !vm_page_try_remove_all(p))
goto wired;
- vm_page_free(p);
+ vm_page_iter_free(&pages);
}
vm_object_pip_wakeup(object);
vm_pager_freespace(object, start, (end == 0 ? object->size : end) -
start);
}
/*
* vm_object_page_noreuse:
*
* For the given object, attempt to move the specified pages to
* the head of the inactive queue. This bypasses regular LRU
* operation and allows the pages to be reused quickly under memory
* pressure. If a page is wired for any reason, then it will not
* be queued. Pages are specified by the range ["start", "end").
* As a special case, if "end" is zero, then the range extends from
* "start" to the end of the object.
*
* This operation should only be performed on objects that
* contain non-fictitious, managed pages.
*
* The object must be locked.
*/
void
vm_object_page_noreuse(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
{
vm_page_t p, next;
VM_OBJECT_ASSERT_LOCKED(object);
KASSERT((object->flags & (OBJ_FICTITIOUS | OBJ_UNMANAGED)) == 0,
("vm_object_page_noreuse: illegal object %p", object));
if (object->resident_page_count == 0)
return;
p = vm_page_find_least(object, start);
/*
* Here, the variable "p" is either (1) the page with the least pindex
* greater than or equal to the parameter "start" or (2) NULL.
*/
for (; p != NULL && (p->pindex < end || end == 0); p = next) {
next = TAILQ_NEXT(p, listq);
vm_page_deactivate_noreuse(p);
}
}
/*
* Populate the specified range of the object with valid pages. Returns
* TRUE if the range is successfully populated and FALSE otherwise.
*
* Note: This function should be optimized to pass a larger array of
* pages to vm_pager_get_pages() before it is applied to a non-
* OBJT_DEVICE object.
*
* The object must be locked.
*/
boolean_t
vm_object_populate(vm_object_t object, vm_pindex_t start, vm_pindex_t end)
{
vm_page_t m;
vm_pindex_t pindex;
int rv;
VM_OBJECT_ASSERT_WLOCKED(object);
for (pindex = start; pindex < end; pindex++) {
rv = vm_page_grab_valid(&m, object, pindex, VM_ALLOC_NORMAL);
if (rv != VM_PAGER_OK)
break;
/*
* Keep "m" busy because a subsequent iteration may unlock
* the object.
*/
}
if (pindex > start) {
m = vm_page_lookup(object, start);
while (m != NULL && m->pindex < pindex) {
vm_page_xunbusy(m);
m = TAILQ_NEXT(m, listq);
}
}
return (pindex == end);
}
/*
* Routine: vm_object_coalesce
* Function: Coalesces two objects backing up adjoining
* regions of memory into a single object.
*
* returns TRUE if objects were combined.
*
* NOTE: Only works at the moment if the second object is NULL -
* if it's not, which object do we lock first?
*
* Parameters:
* prev_object First object to coalesce
* prev_offset Offset into prev_object
* prev_size Size of reference to prev_object
* next_size Size of reference to the second object
* reserved Indicator that extension region has
* swap accounted for
*
* Conditions:
* The object must *not* be locked.
*/
boolean_t
vm_object_coalesce(vm_object_t prev_object, vm_ooffset_t prev_offset,
vm_size_t prev_size, vm_size_t next_size, boolean_t reserved)
{
vm_pindex_t next_pindex;
if (prev_object == NULL)
return (TRUE);
if ((prev_object->flags & OBJ_ANON) == 0)
return (FALSE);
VM_OBJECT_WLOCK(prev_object);
/*
* Try to collapse the object first.
*/
vm_object_collapse(prev_object);
/*
* Can't coalesce if: . more than one reference . paged out . shadows
* another object . has a copy elsewhere (any of which mean that the
* pages not mapped to prev_entry may be in use anyway)
*/
if (prev_object->backing_object != NULL) {
VM_OBJECT_WUNLOCK(prev_object);
return (FALSE);
}
prev_size >>= PAGE_SHIFT;
next_size >>= PAGE_SHIFT;
next_pindex = OFF_TO_IDX(prev_offset) + prev_size;
if (prev_object->ref_count > 1 &&
prev_object->size != next_pindex &&
(prev_object->flags & OBJ_ONEMAPPING) == 0) {
VM_OBJECT_WUNLOCK(prev_object);
return (FALSE);
}
/*
* Account for the charge.
*/
if (prev_object->cred != NULL) {
/*
* If prev_object was charged, then this mapping,
* although not charged now, may become writable
* later. Non-NULL cred in the object would prevent
* swap reservation during enabling of the write
* access, so reserve swap now. Failed reservation
* cause allocation of the separate object for the map
* entry, and swap reservation for this entry is
* managed in appropriate time.
*/
if (!reserved && !swap_reserve_by_cred(ptoa(next_size),
prev_object->cred)) {
VM_OBJECT_WUNLOCK(prev_object);
return (FALSE);
}
prev_object->charge += ptoa(next_size);
}
/*
* Remove any pages that may still be in the object from a previous
* deallocation.
*/
if (next_pindex < prev_object->size) {
vm_object_page_remove(prev_object, next_pindex, next_pindex +
next_size, 0);
#if 0
if (prev_object->cred != NULL) {
KASSERT(prev_object->charge >=
ptoa(prev_object->size - next_pindex),
("object %p overcharged 1 %jx %jx", prev_object,
(uintmax_t)next_pindex, (uintmax_t)next_size));
prev_object->charge -= ptoa(prev_object->size -
next_pindex);
}
#endif
}
/*
* Extend the object if necessary.
*/
if (next_pindex + next_size > prev_object->size)
prev_object->size = next_pindex + next_size;
VM_OBJECT_WUNLOCK(prev_object);
return (TRUE);
}
void
vm_object_set_writeable_dirty_(vm_object_t object)
{
atomic_add_int(&object->generation, 1);
}
bool
vm_object_mightbedirty_(vm_object_t object)
{
return (object->generation != object->cleangeneration);
}
/*
* vm_object_unwire:
*
* For each page offset within the specified range of the given object,
* find the highest-level page in the shadow chain and unwire it. A page
* must exist at every page offset, and the highest-level page must be
* wired.
*/
void
vm_object_unwire(vm_object_t object, vm_ooffset_t offset, vm_size_t length,
uint8_t queue)
{
vm_object_t tobject, t1object;
vm_page_t m, tm;
vm_pindex_t end_pindex, pindex, tpindex;
int depth, locked_depth;
KASSERT((offset & PAGE_MASK) == 0,
("vm_object_unwire: offset is not page aligned"));
KASSERT((length & PAGE_MASK) == 0,
("vm_object_unwire: length is not a multiple of PAGE_SIZE"));
/* The wired count of a fictitious page never changes. */
if ((object->flags & OBJ_FICTITIOUS) != 0)
return;
pindex = OFF_TO_IDX(offset);
end_pindex = pindex + atop(length);
again:
locked_depth = 1;
VM_OBJECT_RLOCK(object);
m = vm_page_find_least(object, pindex);
while (pindex < end_pindex) {
if (m == NULL || pindex < m->pindex) {
/*
* The first object in the shadow chain doesn't
* contain a page at the current index. Therefore,
* the page must exist in a backing object.
*/
tobject = object;
tpindex = pindex;
depth = 0;
do {
tpindex +=
OFF_TO_IDX(tobject->backing_object_offset);
tobject = tobject->backing_object;
KASSERT(tobject != NULL,
("vm_object_unwire: missing page"));
if ((tobject->flags & OBJ_FICTITIOUS) != 0)
goto next_page;
depth++;
if (depth == locked_depth) {
locked_depth++;
VM_OBJECT_RLOCK(tobject);
}
} while ((tm = vm_page_lookup(tobject, tpindex)) ==
NULL);
} else {
tm = m;
m = TAILQ_NEXT(m, listq);
}
if (vm_page_trysbusy(tm) == 0) {
for (tobject = object; locked_depth >= 1;
locked_depth--) {
t1object = tobject->backing_object;
if (tm->object != tobject)
VM_OBJECT_RUNLOCK(tobject);
tobject = t1object;
}
tobject = tm->object;
if (!vm_page_busy_sleep(tm, "unwbo",
VM_ALLOC_IGN_SBUSY))
VM_OBJECT_RUNLOCK(tobject);
goto again;
}
vm_page_unwire(tm, queue);
vm_page_sunbusy(tm);
next_page:
pindex++;
}
/* Release the accumulated object locks. */
for (tobject = object; locked_depth >= 1; locked_depth--) {
t1object = tobject->backing_object;
VM_OBJECT_RUNLOCK(tobject);
tobject = t1object;
}
}
/*
* Return the vnode for the given object, or NULL if none exists.
* For tmpfs objects, the function may return NULL if there is
* no vnode allocated at the time of the call.
*/
struct vnode *
vm_object_vnode(vm_object_t object)
{
struct vnode *vp;
VM_OBJECT_ASSERT_LOCKED(object);
vm_pager_getvp(object, &vp, NULL);
return (vp);
}
/*
* Busy the vm object. This prevents new pages belonging to the object from
* becoming busy. Existing pages persist as busy. Callers are responsible
* for checking page state before proceeding.
*/
void
vm_object_busy(vm_object_t obj)
{
VM_OBJECT_ASSERT_LOCKED(obj);
blockcount_acquire(&obj->busy, 1);
/* The fence is required to order loads of page busy. */
atomic_thread_fence_acq_rel();
}
void
vm_object_unbusy(vm_object_t obj)
{
blockcount_release(&obj->busy, 1);
}
void
vm_object_busy_wait(vm_object_t obj, const char *wmesg)
{
VM_OBJECT_ASSERT_UNLOCKED(obj);
(void)blockcount_sleep(&obj->busy, NULL, wmesg, PVM);
}
/*
* This function aims to determine if the object is mapped,
* specifically, if it is referenced by a vm_map_entry. Because
* objects occasionally acquire transient references that do not
* represent a mapping, the method used here is inexact. However, it
* has very low overhead and is good enough for the advisory
* vm.vmtotal sysctl.
*/
bool
vm_object_is_active(vm_object_t obj)
{
return (obj->ref_count > atomic_load_int(&obj->shadow_count));
}
static int
vm_object_list_handler(struct sysctl_req *req, bool swap_only)
{
struct kinfo_vmobject *kvo;
char *fullpath, *freepath;
struct vnode *vp;
struct vattr va;
vm_object_t obj;
vm_page_t m;
struct cdev *cdev;
struct cdevsw *csw;
u_long sp;
int count, error, ref;
key_t key;
unsigned short seq;
bool want_path;
if (req->oldptr == NULL) {
/*
* If an old buffer has not been provided, generate an
* estimate of the space needed for a subsequent call.
*/
mtx_lock(&vm_object_list_mtx);
count = 0;
TAILQ_FOREACH(obj, &vm_object_list, object_list) {
if (obj->type == OBJT_DEAD)
continue;
count++;
}
mtx_unlock(&vm_object_list_mtx);
return (SYSCTL_OUT(req, NULL, sizeof(struct kinfo_vmobject) *
count * 11 / 10));
}
want_path = !(swap_only || jailed(curthread->td_ucred));
kvo = malloc(sizeof(*kvo), M_TEMP, M_WAITOK | M_ZERO);
error = 0;
/*
* VM objects are type stable and are never removed from the
* list once added. This allows us to safely read obj->object_list
* after reacquiring the VM object lock.
*/
mtx_lock(&vm_object_list_mtx);
TAILQ_FOREACH(obj, &vm_object_list, object_list) {
if (obj->type == OBJT_DEAD ||
(swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0))
continue;
VM_OBJECT_RLOCK(obj);
if (obj->type == OBJT_DEAD ||
(swap_only && (obj->flags & (OBJ_ANON | OBJ_SWAP)) == 0)) {
VM_OBJECT_RUNLOCK(obj);
continue;
}
mtx_unlock(&vm_object_list_mtx);
kvo->kvo_size = ptoa(obj->size);
kvo->kvo_resident = obj->resident_page_count;
kvo->kvo_ref_count = obj->ref_count;
kvo->kvo_shadow_count = atomic_load_int(&obj->shadow_count);
kvo->kvo_memattr = obj->memattr;
kvo->kvo_active = 0;
kvo->kvo_inactive = 0;
kvo->kvo_flags = 0;
if (!swap_only) {
TAILQ_FOREACH(m, &obj->memq, listq) {
/*
* A page may belong to the object but be
* dequeued and set to PQ_NONE while the
* object lock is not held. This makes the
* reads of m->queue below racy, and we do not
* count pages set to PQ_NONE. However, this
* sysctl is only meant to give an
* approximation of the system anyway.
*/
if (vm_page_active(m))
kvo->kvo_active++;
else if (vm_page_inactive(m))
kvo->kvo_inactive++;
else if (vm_page_in_laundry(m))
kvo->kvo_laundry++;
}
}
kvo->kvo_vn_fileid = 0;
kvo->kvo_vn_fsid = 0;
kvo->kvo_vn_fsid_freebsd11 = 0;
freepath = NULL;
fullpath = "";
vp = NULL;
kvo->kvo_type = vm_object_kvme_type(obj, want_path ? &vp :
NULL);
if (vp != NULL) {
vref(vp);
} else if ((obj->flags & OBJ_ANON) != 0) {
MPASS(kvo->kvo_type == KVME_TYPE_SWAP);
kvo->kvo_me = (uintptr_t)obj;
/* tmpfs objs are reported as vnodes */
kvo->kvo_backing_obj = (uintptr_t)obj->backing_object;
sp = swap_pager_swapped_pages(obj);
kvo->kvo_swapped = sp > UINT32_MAX ? UINT32_MAX : sp;
}
if ((obj->type == OBJT_DEVICE || obj->type == OBJT_MGTDEVICE) &&
(obj->flags & OBJ_CDEVH) != 0) {
cdev = obj->un_pager.devp.handle;
if (cdev != NULL) {
csw = dev_refthread(cdev, &ref);
if (csw != NULL) {
strlcpy(kvo->kvo_path, cdev->si_name,
sizeof(kvo->kvo_path));
dev_relthread(cdev, ref);
}
}
}
VM_OBJECT_RUNLOCK(obj);
if ((obj->flags & OBJ_SYSVSHM) != 0) {
kvo->kvo_flags |= KVMO_FLAG_SYSVSHM;
shmobjinfo(obj, &key, &seq);
kvo->kvo_vn_fileid = key;
kvo->kvo_vn_fsid_freebsd11 = seq;
}
if ((obj->flags & OBJ_POSIXSHM) != 0) {
kvo->kvo_flags |= KVMO_FLAG_POSIXSHM;
shm_get_path(obj, kvo->kvo_path,
sizeof(kvo->kvo_path));
}
if (vp != NULL) {
vn_fullpath(vp, &fullpath, &freepath);
vn_lock(vp, LK_SHARED | LK_RETRY);
if (VOP_GETATTR(vp, &va, curthread->td_ucred) == 0) {
kvo->kvo_vn_fileid = va.va_fileid;
kvo->kvo_vn_fsid = va.va_fsid;
kvo->kvo_vn_fsid_freebsd11 = va.va_fsid;
/* truncate */
}
vput(vp);
strlcpy(kvo->kvo_path, fullpath, sizeof(kvo->kvo_path));
free(freepath, M_TEMP);
}
/* Pack record size down */
kvo->kvo_structsize = offsetof(struct kinfo_vmobject, kvo_path)
+ strlen(kvo->kvo_path) + 1;
kvo->kvo_structsize = roundup(kvo->kvo_structsize,
sizeof(uint64_t));
error = SYSCTL_OUT(req, kvo, kvo->kvo_structsize);
maybe_yield();
mtx_lock(&vm_object_list_mtx);
if (error)
break;
}
mtx_unlock(&vm_object_list_mtx);
free(kvo, M_TEMP);
return (error);
}
static int
sysctl_vm_object_list(SYSCTL_HANDLER_ARGS)
{
return (vm_object_list_handler(req, false));
}
SYSCTL_PROC(_vm, OID_AUTO, objects, CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP |
CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_object_list, "S,kinfo_vmobject",
"List of VM objects");
static int
sysctl_vm_object_list_swap(SYSCTL_HANDLER_ARGS)
{
return (vm_object_list_handler(req, true));
}
/*
* This sysctl returns list of the anonymous or swap objects. Intent
* is to provide stripped optimized list useful to analyze swap use.
* Since technically non-swap (default) objects participate in the
* shadow chains, and are converted to swap type as needed by swap
* pager, we must report them.
*/
SYSCTL_PROC(_vm, OID_AUTO, swap_objects,
CTLTYPE_STRUCT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, NULL, 0,
sysctl_vm_object_list_swap, "S,kinfo_vmobject",
"List of swap VM objects");
#include "opt_ddb.h"
#ifdef DDB
#include <sys/kernel.h>
#include <sys/cons.h>
#include <ddb/ddb.h>
static int
_vm_object_in_map(vm_map_t map, vm_object_t object, vm_map_entry_t entry)
{
vm_map_t tmpm;
vm_map_entry_t tmpe;
vm_object_t obj;
if (map == 0)
return 0;
if (entry == 0) {
VM_MAP_ENTRY_FOREACH(tmpe, map) {
if (_vm_object_in_map(map, object, tmpe)) {
return 1;
}
}
} else if (entry->eflags & MAP_ENTRY_IS_SUB_MAP) {
tmpm = entry->object.sub_map;
VM_MAP_ENTRY_FOREACH(tmpe, tmpm) {
if (_vm_object_in_map(tmpm, object, tmpe)) {
return 1;
}
}
} else if ((obj = entry->object.vm_object) != NULL) {
for (; obj; obj = obj->backing_object)
if (obj == object) {
return 1;
}
}
return 0;
}
static int
vm_object_in_map(vm_object_t object)
{
struct proc *p;
/* sx_slock(&allproc_lock); */
FOREACH_PROC_IN_SYSTEM(p) {
if (!p->p_vmspace /* || (p->p_flag & (P_SYSTEM|P_WEXIT)) */)
continue;
if (_vm_object_in_map(&p->p_vmspace->vm_map, object, 0)) {
/* sx_sunlock(&allproc_lock); */
return 1;
}
}
/* sx_sunlock(&allproc_lock); */
if (_vm_object_in_map(kernel_map, object, 0))
return 1;
return 0;
}
DB_SHOW_COMMAND_FLAGS(vmochk, vm_object_check, DB_CMD_MEMSAFE)
{
vm_object_t object;
/*
* make sure that internal objs are in a map somewhere
* and none have zero ref counts.
*/
TAILQ_FOREACH(object, &vm_object_list, object_list) {
if ((object->flags & OBJ_ANON) != 0) {
if (object->ref_count == 0) {
db_printf("vmochk: internal obj has zero ref count: %ld\n",
(long)object->size);
}
if (!vm_object_in_map(object)) {
db_printf(
"vmochk: internal obj is not in a map: "
"ref: %d, size: %lu: 0x%lx, backing_object: %p\n",
object->ref_count, (u_long)object->size,
(u_long)object->size,
(void *)object->backing_object);
}
}
if (db_pager_quit)
return;
}
}
/*
* vm_object_print: [ debug ]
*/
DB_SHOW_COMMAND(object, vm_object_print_static)
{
/* XXX convert args. */
vm_object_t object = (vm_object_t)addr;
boolean_t full = have_addr;
vm_page_t p;
/* XXX count is an (unused) arg. Avoid shadowing it. */
#define count was_count
int count;
if (object == NULL)
return;
db_iprintf(
"Object %p: type=%d, size=0x%jx, res=%d, ref=%d, flags=0x%x ruid %d charge %jx\n",
object, (int)object->type, (uintmax_t)object->size,
object->resident_page_count, object->ref_count, object->flags,
object->cred ? object->cred->cr_ruid : -1, (uintmax_t)object->charge);
db_iprintf(" sref=%d, backing_object(%d)=(%p)+0x%jx\n",
atomic_load_int(&object->shadow_count),
object->backing_object ? object->backing_object->ref_count : 0,
object->backing_object, (uintmax_t)object->backing_object_offset);
if (!full)
return;
db_indent += 2;
count = 0;
TAILQ_FOREACH(p, &object->memq, listq) {
if (count == 0)
db_iprintf("memory:=");
else if (count == 6) {
db_printf("\n");
db_iprintf(" ...");
count = 0;
} else
db_printf(",");
count++;
db_printf("(off=0x%jx,page=0x%jx)",
(uintmax_t)p->pindex, (uintmax_t)VM_PAGE_TO_PHYS(p));
if (db_pager_quit)
break;
}
if (count != 0)
db_printf("\n");
db_indent -= 2;
}
/* XXX. */
#undef count
/* XXX need this non-static entry for calling from vm_map_print. */
void
vm_object_print(
/* db_expr_t */ long addr,
boolean_t have_addr,
/* db_expr_t */ long count,
char *modif)
{
vm_object_print_static(addr, have_addr, count, modif);
}
DB_SHOW_COMMAND_FLAGS(vmopag, vm_object_print_pages, DB_CMD_MEMSAFE)
{
vm_object_t object;
vm_pindex_t fidx;
vm_paddr_t pa;
vm_page_t m, prev_m;
int rcount;
TAILQ_FOREACH(object, &vm_object_list, object_list) {
db_printf("new object: %p\n", (void *)object);
if (db_pager_quit)
return;
rcount = 0;
fidx = 0;
pa = -1;
TAILQ_FOREACH(m, &object->memq, listq) {
if ((prev_m = TAILQ_PREV(m, pglist, listq)) != NULL &&
prev_m->pindex + 1 != m->pindex) {
if (rcount) {
db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
(long)fidx, rcount, (long)pa);
if (db_pager_quit)
return;
rcount = 0;
}
}
if (rcount &&
(VM_PAGE_TO_PHYS(m) == pa + rcount * PAGE_SIZE)) {
++rcount;
continue;
}
if (rcount) {
db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
(long)fidx, rcount, (long)pa);
if (db_pager_quit)
return;
}
fidx = m->pindex;
pa = VM_PAGE_TO_PHYS(m);
rcount = 1;
}
if (rcount) {
db_printf(" index(%ld)run(%d)pa(0x%lx)\n",
(long)fidx, rcount, (long)pa);
if (db_pager_quit)
return;
}
}
}
#endif /* DDB */
diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c
index 0b9b55337b52..7d093579e35d 100644
--- a/sys/vm/vm_page.c
+++ b/sys/vm/vm_page.c
@@ -1,5882 +1,5948 @@
/*-
* SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
*
* Copyright (c) 1991 Regents of the University of California.
* All rights reserved.
* Copyright (c) 1998 Matthew Dillon. All Rights Reserved.
*
* This code is derived from software contributed to Berkeley by
* The Mach Operating System project at Carnegie-Mellon University.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
/*-
* Copyright (c) 1987, 1990 Carnegie-Mellon University.
* All rights reserved.
*
* Authors: Avadis Tevanian, Jr., Michael Wayne Young
*
* Permission to use, copy, modify and distribute this software and
* its documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie the
* rights to redistribute these changes.
*/
/*
* Resident memory management module.
*/
#include <sys/cdefs.h>
#include "opt_vm.h"
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/counter.h>
#include <sys/domainset.h>
#include <sys/kernel.h>
#include <sys/limits.h>
#include <sys/linker.h>
#include <sys/lock.h>
#include <sys/malloc.h>
#include <sys/mman.h>
#include <sys/msgbuf.h>
#include <sys/mutex.h>
#include <sys/proc.h>
#include <sys/rwlock.h>
#include <sys/sleepqueue.h>
#include <sys/sbuf.h>
#include <sys/sched.h>
#include <sys/smp.h>
#include <sys/sysctl.h>
#include <sys/vmmeter.h>
#include <sys/vnode.h>
#include <vm/vm.h>
#include <vm/pmap.h>
#include <vm/vm_param.h>
#include <vm/vm_domainset.h>
#include <vm/vm_kern.h>
#include <vm/vm_map.h>
#include <vm/vm_object.h>
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
#include <vm/vm_phys.h>
#include <vm/vm_pagequeue.h>
#include <vm/vm_pager.h>
#include <vm/vm_radix.h>
#include <vm/vm_reserv.h>
#include <vm/vm_extern.h>
#include <vm/vm_dumpset.h>
#include <vm/uma.h>
#include <vm/uma_int.h>
#include <machine/md_var.h>
struct vm_domain vm_dom[MAXMEMDOM];
DPCPU_DEFINE_STATIC(struct vm_batchqueue, pqbatch[MAXMEMDOM][PQ_COUNT]);
struct mtx_padalign __exclusive_cache_line pa_lock[PA_LOCK_COUNT];
struct mtx_padalign __exclusive_cache_line vm_domainset_lock;
/* The following fields are protected by the domainset lock. */
domainset_t __exclusive_cache_line vm_min_domains;
domainset_t __exclusive_cache_line vm_severe_domains;
static int vm_min_waiters;
static int vm_severe_waiters;
static int vm_pageproc_waiters;
static SYSCTL_NODE(_vm_stats, OID_AUTO, page, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
"VM page statistics");
static COUNTER_U64_DEFINE_EARLY(pqstate_commit_retries);
SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, pqstate_commit_retries,
CTLFLAG_RD, &pqstate_commit_retries,
"Number of failed per-page atomic queue state updates");
static COUNTER_U64_DEFINE_EARLY(queue_ops);
SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, queue_ops,
CTLFLAG_RD, &queue_ops,
"Number of batched queue operations");
static COUNTER_U64_DEFINE_EARLY(queue_nops);
SYSCTL_COUNTER_U64(_vm_stats_page, OID_AUTO, queue_nops,
CTLFLAG_RD, &queue_nops,
"Number of batched queue operations with no effects");
/*
* bogus page -- for I/O to/from partially complete buffers,
* or for paging into sparsely invalid regions.
*/
vm_page_t bogus_page;
vm_page_t vm_page_array;
long vm_page_array_size;
long first_page;
struct bitset *vm_page_dump;
long vm_page_dump_pages;
static TAILQ_HEAD(, vm_page) blacklist_head;
static int sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS);
SYSCTL_PROC(_vm, OID_AUTO, page_blacklist, CTLTYPE_STRING | CTLFLAG_RD |
CTLFLAG_MPSAFE, NULL, 0, sysctl_vm_page_blacklist, "A", "Blacklist pages");
static uma_zone_t fakepg_zone;
static vm_page_t vm_page_alloc_after(vm_object_t object, vm_pindex_t pindex,
int req, vm_page_t mpred);
static void vm_page_alloc_check(vm_page_t m);
static vm_page_t vm_page_alloc_nofree_domain(int domain, int req);
static bool _vm_page_busy_sleep(vm_object_t obj, vm_page_t m,
vm_pindex_t pindex, const char *wmesg, int allocflags, bool locked);
static void vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits);
static void vm_page_enqueue(vm_page_t m, uint8_t queue);
-static bool vm_page_free_prep(vm_page_t m);
+static bool vm_page_free_prep(vm_page_t m, bool do_remove);
static void vm_page_free_toq(vm_page_t m);
+static void vm_page_free_toq_impl(vm_page_t m, bool do_remove);
static void vm_page_init(void *dummy);
static int vm_page_insert_after(vm_page_t m, vm_object_t object,
vm_pindex_t pindex, vm_page_t mpred);
static void vm_page_insert_radixdone(vm_page_t m, vm_object_t object,
vm_page_t mpred);
static void vm_page_mvqueue(vm_page_t m, const uint8_t queue,
const uint16_t nflag);
static int vm_page_reclaim_run(int req_class, int domain, u_long npages,
vm_page_t m_run, vm_paddr_t high);
static void vm_page_release_toq(vm_page_t m, uint8_t nqueue, bool noreuse);
static int vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object,
int req);
static int vm_page_zone_import(void *arg, void **store, int cnt, int domain,
int flags);
static void vm_page_zone_release(void *arg, void **store, int cnt);
SYSINIT(vm_page, SI_SUB_VM, SI_ORDER_SECOND, vm_page_init, NULL);
static void
vm_page_init(void *dummy)
{
fakepg_zone = uma_zcreate("fakepg", sizeof(struct vm_page), NULL, NULL,
NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
bogus_page = vm_page_alloc_noobj(VM_ALLOC_WIRED | VM_ALLOC_NOFREE);
}
static int pgcache_zone_max_pcpu;
SYSCTL_INT(_vm, OID_AUTO, pgcache_zone_max_pcpu,
CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &pgcache_zone_max_pcpu, 0,
"Per-CPU page cache size");
/*
* The cache page zone is initialized later since we need to be able to allocate
* pages before UMA is fully initialized.
*/
static void
vm_page_init_cache_zones(void *dummy __unused)
{
struct vm_domain *vmd;
struct vm_pgcache *pgcache;
int cache, domain, maxcache, pool;
TUNABLE_INT_FETCH("vm.pgcache_zone_max_pcpu", &pgcache_zone_max_pcpu);
maxcache = pgcache_zone_max_pcpu * mp_ncpus;
for (domain = 0; domain < vm_ndomains; domain++) {
vmd = VM_DOMAIN(domain);
for (pool = 0; pool < VM_NFREEPOOL; pool++) {
pgcache = &vmd->vmd_pgcache[pool];
pgcache->domain = domain;
pgcache->pool = pool;
pgcache->zone = uma_zcache_create("vm pgcache",
PAGE_SIZE, NULL, NULL, NULL, NULL,
vm_page_zone_import, vm_page_zone_release, pgcache,
UMA_ZONE_VM);
/*
* Limit each pool's zone to 0.1% of the pages in the
* domain.
*/
cache = maxcache != 0 ? maxcache :
vmd->vmd_page_count / 1000;
uma_zone_set_maxcache(pgcache->zone, cache);
}
}
}
SYSINIT(vm_page2, SI_SUB_VM_CONF, SI_ORDER_ANY, vm_page_init_cache_zones, NULL);
/* Make sure that u_long is at least 64 bits when PAGE_SIZE is 32K. */
#if PAGE_SIZE == 32768
#ifdef CTASSERT
CTASSERT(sizeof(u_long) >= 8);
#endif
#endif
/*
* vm_set_page_size:
*
* Sets the page size, perhaps based upon the memory
* size. Must be called before any use of page-size
* dependent functions.
*/
void
vm_set_page_size(void)
{
if (vm_cnt.v_page_size == 0)
vm_cnt.v_page_size = PAGE_SIZE;
if (((vm_cnt.v_page_size - 1) & vm_cnt.v_page_size) != 0)
panic("vm_set_page_size: page size not a power of two");
}
/*
* vm_page_blacklist_next:
*
* Find the next entry in the provided string of blacklist
* addresses. Entries are separated by space, comma, or newline.
* If an invalid integer is encountered then the rest of the
* string is skipped. Updates the list pointer to the next
* character, or NULL if the string is exhausted or invalid.
*/
static vm_paddr_t
vm_page_blacklist_next(char **list, char *end)
{
vm_paddr_t bad;
char *cp, *pos;
if (list == NULL || *list == NULL)
return (0);
if (**list =='\0') {
*list = NULL;
return (0);
}
/*
* If there's no end pointer then the buffer is coming from
* the kenv and we know it's null-terminated.
*/
if (end == NULL)
end = *list + strlen(*list);
/* Ensure that strtoq() won't walk off the end */
if (*end != '\0') {
if (*end == '\n' || *end == ' ' || *end == ',')
*end = '\0';
else {
printf("Blacklist not terminated, skipping\n");
*list = NULL;
return (0);
}
}
for (pos = *list; *pos != '\0'; pos = cp) {
bad = strtoq(pos, &cp, 0);
if (*cp == '\0' || *cp == ' ' || *cp == ',' || *cp == '\n') {
if (bad == 0) {
if (++cp < end)
continue;
else
break;
}
} else
break;
if (*cp == '\0' || ++cp >= end)
*list = NULL;
else
*list = cp;
return (trunc_page(bad));
}
printf("Garbage in RAM blacklist, skipping\n");
*list = NULL;
return (0);
}
bool
vm_page_blacklist_add(vm_paddr_t pa, bool verbose)
{
struct vm_domain *vmd;
vm_page_t m;
bool found;
m = vm_phys_paddr_to_vm_page(pa);
if (m == NULL)
return (true); /* page does not exist, no failure */
vmd = VM_DOMAIN(vm_phys_domain(pa));
vm_domain_free_lock(vmd);
found = vm_phys_unfree_page(pa);
vm_domain_free_unlock(vmd);
if (found) {
vm_domain_freecnt_inc(vmd, -1);
TAILQ_INSERT_TAIL(&blacklist_head, m, listq);
if (verbose)
printf("Skipping page with pa 0x%jx\n", (uintmax_t)pa);
}
return (found);
}
/*
* vm_page_blacklist_check:
*
* Iterate through the provided string of blacklist addresses, pulling
* each entry out of the physical allocator free list and putting it
* onto a list for reporting via the vm.page_blacklist sysctl.
*/
static void
vm_page_blacklist_check(char *list, char *end)
{
vm_paddr_t pa;
char *next;
next = list;
while (next != NULL) {
if ((pa = vm_page_blacklist_next(&next, end)) == 0)
continue;
vm_page_blacklist_add(pa, bootverbose);
}
}
/*
* vm_page_blacklist_load:
*
* Search for a special module named "ram_blacklist". It'll be a
* plain text file provided by the user via the loader directive
* of the same name.
*/
static void
vm_page_blacklist_load(char **list, char **end)
{
void *mod;
u_char *ptr;
u_int len;
mod = NULL;
ptr = NULL;
mod = preload_search_by_type("ram_blacklist");
if (mod != NULL) {
ptr = preload_fetch_addr(mod);
len = preload_fetch_size(mod);
}
*list = ptr;
if (ptr != NULL)
*end = ptr + len;
else
*end = NULL;
return;
}
static int
sysctl_vm_page_blacklist(SYSCTL_HANDLER_ARGS)
{
vm_page_t m;
struct sbuf sbuf;
int error, first;
first = 1;
error = sysctl_wire_old_buffer(req, 0);
if (error != 0)
return (error);
sbuf_new_for_sysctl(&sbuf, NULL, 128, req);
TAILQ_FOREACH(m, &blacklist_head, listq) {
sbuf_printf(&sbuf, "%s%#jx", first ? "" : ",",
(uintmax_t)m->phys_addr);
first = 0;
}
error = sbuf_finish(&sbuf);
sbuf_delete(&sbuf);
return (error);
}
/*
* Initialize a dummy page for use in scans of the specified paging queue.
* In principle, this function only needs to set the flag PG_MARKER.
* Nonetheless, it write busies the page as a safety precaution.
*/
void
vm_page_init_marker(vm_page_t marker, int queue, uint16_t aflags)
{
bzero(marker, sizeof(*marker));
marker->flags = PG_MARKER;
marker->a.flags = aflags;
marker->busy_lock = VPB_CURTHREAD_EXCLUSIVE;
marker->a.queue = queue;
}
static void
vm_page_domain_init(int domain)
{
struct vm_domain *vmd;
struct vm_pagequeue *pq;
int i;
vmd = VM_DOMAIN(domain);
bzero(vmd, sizeof(*vmd));
*__DECONST(const char **, &vmd->vmd_pagequeues[PQ_INACTIVE].pq_name) =
"vm inactive pagequeue";
*__DECONST(const char **, &vmd->vmd_pagequeues[PQ_ACTIVE].pq_name) =
"vm active pagequeue";
*__DECONST(const char **, &vmd->vmd_pagequeues[PQ_LAUNDRY].pq_name) =
"vm laundry pagequeue";
*__DECONST(const char **,
&vmd->vmd_pagequeues[PQ_UNSWAPPABLE].pq_name) =
"vm unswappable pagequeue";
vmd->vmd_domain = domain;
vmd->vmd_page_count = 0;
vmd->vmd_free_count = 0;
vmd->vmd_segs = 0;
vmd->vmd_oom = FALSE;
for (i = 0; i < PQ_COUNT; i++) {
pq = &vmd->vmd_pagequeues[i];
TAILQ_INIT(&pq->pq_pl);
mtx_init(&pq->pq_mutex, pq->pq_name, "vm pagequeue",
MTX_DEF | MTX_DUPOK);
pq->pq_pdpages = 0;
vm_page_init_marker(&vmd->vmd_markers[i], i, 0);
}
mtx_init(&vmd->vmd_free_mtx, "vm page free queue", NULL, MTX_DEF);
mtx_init(&vmd->vmd_pageout_mtx, "vm pageout lock", NULL, MTX_DEF);
snprintf(vmd->vmd_name, sizeof(vmd->vmd_name), "%d", domain);
/*
* inacthead is used to provide FIFO ordering for LRU-bypassing
* insertions.
*/
vm_page_init_marker(&vmd->vmd_inacthead, PQ_INACTIVE, PGA_ENQUEUED);
TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_INACTIVE].pq_pl,
&vmd->vmd_inacthead, plinks.q);
/*
* The clock pages are used to implement active queue scanning without
* requeues. Scans start at clock[0], which is advanced after the scan
* ends. When the two clock hands meet, they are reset and scanning
* resumes from the head of the queue.
*/
vm_page_init_marker(&vmd->vmd_clock[0], PQ_ACTIVE, PGA_ENQUEUED);
vm_page_init_marker(&vmd->vmd_clock[1], PQ_ACTIVE, PGA_ENQUEUED);
TAILQ_INSERT_HEAD(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl,
&vmd->vmd_clock[0], plinks.q);
TAILQ_INSERT_TAIL(&vmd->vmd_pagequeues[PQ_ACTIVE].pq_pl,
&vmd->vmd_clock[1], plinks.q);
}
/*
* Initialize a physical page in preparation for adding it to the free
* lists.
*/
void
vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind, int pool)
{
m->object = NULL;
m->ref_count = 0;
m->busy_lock = VPB_FREED;
m->flags = m->a.flags = 0;
m->phys_addr = pa;
m->a.queue = PQ_NONE;
m->psind = 0;
m->segind = segind;
m->order = VM_NFREEORDER;
m->pool = pool;
m->valid = m->dirty = 0;
pmap_page_init(m);
}
#ifndef PMAP_HAS_PAGE_ARRAY
static vm_paddr_t
vm_page_array_alloc(vm_offset_t *vaddr, vm_paddr_t end, vm_paddr_t page_range)
{
vm_paddr_t new_end;
/*
* Reserve an unmapped guard page to trap access to vm_page_array[-1].
* However, because this page is allocated from KVM, out-of-bounds
* accesses using the direct map will not be trapped.
*/
*vaddr += PAGE_SIZE;
/*
* Allocate physical memory for the page structures, and map it.
*/
new_end = trunc_page(end - page_range * sizeof(struct vm_page));
vm_page_array = (vm_page_t)pmap_map(vaddr, new_end, end,
VM_PROT_READ | VM_PROT_WRITE);
vm_page_array_size = page_range;
return (new_end);
}
#endif
/*
* vm_page_startup:
*
* Initializes the resident memory module. Allocates physical memory for
* bootstrapping UMA and some data structures that are used to manage
* physical pages. Initializes these structures, and populates the free
* page queues.
*/
vm_offset_t
vm_page_startup(vm_offset_t vaddr)
{
struct vm_phys_seg *seg;
struct vm_domain *vmd;
vm_page_t m;
char *list, *listend;
vm_paddr_t end, high_avail, low_avail, new_end, size;
vm_paddr_t page_range __unused;
vm_paddr_t last_pa, pa, startp, endp;
u_long pagecount;
#if MINIDUMP_PAGE_TRACKING
u_long vm_page_dump_size;
#endif
int biggestone, i, segind;
#ifdef WITNESS
vm_offset_t mapped;
int witness_size;
#endif
#if defined(__i386__) && defined(VM_PHYSSEG_DENSE)
long ii;
#endif
#ifdef VM_FREEPOOL_LAZYINIT
int lazyinit;
#endif
vaddr = round_page(vaddr);
vm_phys_early_startup();
biggestone = vm_phys_avail_largest();
end = phys_avail[biggestone+1];
/*
* Initialize the page and queue locks.
*/
mtx_init(&vm_domainset_lock, "vm domainset lock", NULL, MTX_DEF);
for (i = 0; i < PA_LOCK_COUNT; i++)
mtx_init(&pa_lock[i], "vm page", NULL, MTX_DEF);
for (i = 0; i < vm_ndomains; i++)
vm_page_domain_init(i);
new_end = end;
#ifdef WITNESS
witness_size = round_page(witness_startup_count());
new_end -= witness_size;
mapped = pmap_map(&vaddr, new_end, new_end + witness_size,
VM_PROT_READ | VM_PROT_WRITE);
bzero((void *)mapped, witness_size);
witness_startup((void *)mapped);
#endif
#if MINIDUMP_PAGE_TRACKING
/*
* Allocate a bitmap to indicate that a random physical page
* needs to be included in a minidump.
*
* The amd64 port needs this to indicate which direct map pages
* need to be dumped, via calls to dump_add_page()/dump_drop_page().
*
* However, i386 still needs this workspace internally within the
* minidump code. In theory, they are not needed on i386, but are
* included should the sf_buf code decide to use them.
*/
last_pa = 0;
vm_page_dump_pages = 0;
for (i = 0; dump_avail[i + 1] != 0; i += 2) {
vm_page_dump_pages += howmany(dump_avail[i + 1], PAGE_SIZE) -
dump_avail[i] / PAGE_SIZE;
if (dump_avail[i + 1] > last_pa)
last_pa = dump_avail[i + 1];
}
vm_page_dump_size = round_page(BITSET_SIZE(vm_page_dump_pages));
new_end -= vm_page_dump_size;
vm_page_dump = (void *)(uintptr_t)pmap_map(&vaddr, new_end,
new_end + vm_page_dump_size, VM_PROT_READ | VM_PROT_WRITE);
bzero((void *)vm_page_dump, vm_page_dump_size);
#if MINIDUMP_STARTUP_PAGE_TRACKING
/*
* Include the UMA bootstrap pages, witness pages and vm_page_dump
* in a crash dump. When pmap_map() uses the direct map, they are
* not automatically included.
*/
for (pa = new_end; pa < end; pa += PAGE_SIZE)
dump_add_page(pa);
#endif
#else
(void)last_pa;
#endif
phys_avail[biggestone + 1] = new_end;
#ifdef __amd64__
/*
* Request that the physical pages underlying the message buffer be
* included in a crash dump. Since the message buffer is accessed
* through the direct map, they are not automatically included.
*/
pa = DMAP_TO_PHYS((vm_offset_t)msgbufp->msg_ptr);
last_pa = pa + round_page(msgbufsize);
while (pa < last_pa) {
dump_add_page(pa);
pa += PAGE_SIZE;
}
#endif
/*
* Compute the number of pages of memory that will be available for
* use, taking into account the overhead of a page structure per page.
* In other words, solve
* "available physical memory" - round_page(page_range *
* sizeof(struct vm_page)) = page_range * PAGE_SIZE
* for page_range.
*/
low_avail = phys_avail[0];
high_avail = phys_avail[1];
for (i = 0; i < vm_phys_nsegs; i++) {
if (vm_phys_segs[i].start < low_avail)
low_avail = vm_phys_segs[i].start;
if (vm_phys_segs[i].end > high_avail)
high_avail = vm_phys_segs[i].end;
}
/* Skip the first chunk. It is already accounted for. */
for (i = 2; phys_avail[i + 1] != 0; i += 2) {
if (phys_avail[i] < low_avail)
low_avail = phys_avail[i];
if (phys_avail[i + 1] > high_avail)
high_avail = phys_avail[i + 1];
}
first_page = low_avail / PAGE_SIZE;
#ifdef VM_PHYSSEG_SPARSE
size = 0;
for (i = 0; i < vm_phys_nsegs; i++)
size += vm_phys_segs[i].end - vm_phys_segs[i].start;
for (i = 0; phys_avail[i + 1] != 0; i += 2)
size += phys_avail[i + 1] - phys_avail[i];
#elif defined(VM_PHYSSEG_DENSE)
size = high_avail - low_avail;
#else
#error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
#endif
#ifdef PMAP_HAS_PAGE_ARRAY
pmap_page_array_startup(size / PAGE_SIZE);
biggestone = vm_phys_avail_largest();
end = new_end = phys_avail[biggestone + 1];
#else
#ifdef VM_PHYSSEG_DENSE
/*
* In the VM_PHYSSEG_DENSE case, the number of pages can account for
* the overhead of a page structure per page only if vm_page_array is
* allocated from the last physical memory chunk. Otherwise, we must
* allocate page structures representing the physical memory
* underlying vm_page_array, even though they will not be used.
*/
if (new_end != high_avail)
page_range = size / PAGE_SIZE;
else
#endif
{
page_range = size / (PAGE_SIZE + sizeof(struct vm_page));
/*
* If the partial bytes remaining are large enough for
* a page (PAGE_SIZE) without a corresponding
* 'struct vm_page', then new_end will contain an
* extra page after subtracting the length of the VM
* page array. Compensate by subtracting an extra
* page from new_end.
*/
if (size % (PAGE_SIZE + sizeof(struct vm_page)) >= PAGE_SIZE) {
if (new_end == high_avail)
high_avail -= PAGE_SIZE;
new_end -= PAGE_SIZE;
}
}
end = new_end;
new_end = vm_page_array_alloc(&vaddr, end, page_range);
#endif
#if VM_NRESERVLEVEL > 0
/*
* Allocate physical memory for the reservation management system's
* data structures, and map it.
*/
new_end = vm_reserv_startup(&vaddr, new_end);
#endif
#if MINIDUMP_PAGE_TRACKING && MINIDUMP_STARTUP_PAGE_TRACKING
/*
* Include vm_page_array and vm_reserv_array in a crash dump.
*/
for (pa = new_end; pa < end; pa += PAGE_SIZE)
dump_add_page(pa);
#endif
phys_avail[biggestone + 1] = new_end;
/*
* Add physical memory segments corresponding to the available
* physical pages.
*/
for (i = 0; phys_avail[i + 1] != 0; i += 2)
if (vm_phys_avail_size(i) != 0)
vm_phys_add_seg(phys_avail[i], phys_avail[i + 1]);
/*
* Initialize the physical memory allocator.
*/
vm_phys_init();
#ifdef VM_FREEPOOL_LAZYINIT
lazyinit = 1;
TUNABLE_INT_FETCH("debug.vm.lazy_page_init", &lazyinit);
#endif
/*
* Initialize the page structures and add every available page to the
* physical memory allocator's free lists.
*/
#if defined(__i386__) && defined(VM_PHYSSEG_DENSE)
for (ii = 0; ii < vm_page_array_size; ii++) {
m = &vm_page_array[ii];
vm_page_init_page(m, (first_page + ii) << PAGE_SHIFT, 0,
VM_FREEPOOL_DEFAULT);
m->flags = PG_FICTITIOUS;
}
#endif
vm_cnt.v_page_count = 0;
for (segind = 0; segind < vm_phys_nsegs; segind++) {
seg = &vm_phys_segs[segind];
/*
* If lazy vm_page initialization is not enabled, simply
* initialize all of the pages in the segment. Otherwise, we
* only initialize:
* 1. Pages not covered by phys_avail[], since they might be
* freed to the allocator at some future point, e.g., by
* kmem_bootstrap_free().
* 2. The first page of each run of free pages handed to the
* vm_phys allocator, which in turn defers initialization
* of pages until they are needed.
* This avoids blocking the boot process for long periods, which
* may be relevant for VMs (which ought to boot as quickly as
* possible) and/or systems with large amounts of physical
* memory.
*/
#ifdef VM_FREEPOOL_LAZYINIT
if (lazyinit) {
startp = seg->start;
for (i = 0; phys_avail[i + 1] != 0; i += 2) {
if (startp >= seg->end)
break;
if (phys_avail[i + 1] < startp)
continue;
if (phys_avail[i] <= startp) {
startp = phys_avail[i + 1];
continue;
}
m = vm_phys_seg_paddr_to_vm_page(seg, startp);
for (endp = MIN(phys_avail[i], seg->end);
startp < endp; startp += PAGE_SIZE, m++) {
vm_page_init_page(m, startp, segind,
VM_FREEPOOL_DEFAULT);
}
}
} else
#endif
for (m = seg->first_page, pa = seg->start;
pa < seg->end; m++, pa += PAGE_SIZE) {
vm_page_init_page(m, pa, segind,
VM_FREEPOOL_DEFAULT);
}
/*
* Add the segment's pages that are covered by one of
* phys_avail's ranges to the free lists.
*/
for (i = 0; phys_avail[i + 1] != 0; i += 2) {
if (seg->end <= phys_avail[i] ||
seg->start >= phys_avail[i + 1])
continue;
startp = MAX(seg->start, phys_avail[i]);
endp = MIN(seg->end, phys_avail[i + 1]);
pagecount = (u_long)atop(endp - startp);
if (pagecount == 0)
continue;
m = vm_phys_seg_paddr_to_vm_page(seg, startp);
#ifdef VM_FREEPOOL_LAZYINIT
if (lazyinit) {
vm_page_init_page(m, startp, segind,
VM_FREEPOOL_LAZYINIT);
}
#endif
vmd = VM_DOMAIN(seg->domain);
vm_domain_free_lock(vmd);
vm_phys_enqueue_contig(m, pagecount);
vm_domain_free_unlock(vmd);
vm_domain_freecnt_inc(vmd, pagecount);
vm_cnt.v_page_count += (u_int)pagecount;
vmd->vmd_page_count += (u_int)pagecount;
vmd->vmd_segs |= 1UL << segind;
}
}
/*
* Remove blacklisted pages from the physical memory allocator.
*/
TAILQ_INIT(&blacklist_head);
vm_page_blacklist_load(&list, &listend);
vm_page_blacklist_check(list, listend);
list = kern_getenv("vm.blacklist");
vm_page_blacklist_check(list, NULL);
freeenv(list);
#if VM_NRESERVLEVEL > 0
/*
* Initialize the reservation management system.
*/
vm_reserv_init();
#endif
return (vaddr);
}
void
vm_page_reference(vm_page_t m)
{
vm_page_aflag_set(m, PGA_REFERENCED);
}
/*
* vm_page_trybusy
*
* Helper routine for grab functions to trylock busy.
*
* Returns true on success and false on failure.
*/
static bool
vm_page_trybusy(vm_page_t m, int allocflags)
{
if ((allocflags & (VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY)) != 0)
return (vm_page_trysbusy(m));
else
return (vm_page_tryxbusy(m));
}
/*
* vm_page_tryacquire
*
* Helper routine for grab functions to trylock busy and wire.
*
* Returns true on success and false on failure.
*/
static inline bool
vm_page_tryacquire(vm_page_t m, int allocflags)
{
bool locked;
locked = vm_page_trybusy(m, allocflags);
if (locked && (allocflags & VM_ALLOC_WIRED) != 0)
vm_page_wire(m);
return (locked);
}
/*
* vm_page_busy_acquire:
*
* Acquire the busy lock as described by VM_ALLOC_* flags. Will loop
* and drop the object lock if necessary.
*/
bool
vm_page_busy_acquire(vm_page_t m, int allocflags)
{
vm_object_t obj;
bool locked;
/*
* The page-specific object must be cached because page
* identity can change during the sleep, causing the
* re-lock of a different object.
* It is assumed that a reference to the object is already
* held by the callers.
*/
obj = atomic_load_ptr(&m->object);
for (;;) {
if (vm_page_tryacquire(m, allocflags))
return (true);
if ((allocflags & VM_ALLOC_NOWAIT) != 0)
return (false);
if (obj != NULL)
locked = VM_OBJECT_WOWNED(obj);
else
locked = false;
MPASS(locked || vm_page_wired(m));
if (_vm_page_busy_sleep(obj, m, m->pindex, "vmpba", allocflags,
locked) && locked)
VM_OBJECT_WLOCK(obj);
if ((allocflags & VM_ALLOC_WAITFAIL) != 0)
return (false);
KASSERT(m->object == obj || m->object == NULL,
("vm_page_busy_acquire: page %p does not belong to %p",
m, obj));
}
}
/*
* vm_page_busy_downgrade:
*
* Downgrade an exclusive busy page into a single shared busy page.
*/
void
vm_page_busy_downgrade(vm_page_t m)
{
u_int x;
vm_page_assert_xbusied(m);
x = vm_page_busy_fetch(m);
for (;;) {
if (atomic_fcmpset_rel_int(&m->busy_lock,
&x, VPB_SHARERS_WORD(1)))
break;
}
if ((x & VPB_BIT_WAITERS) != 0)
wakeup(m);
}
/*
*
* vm_page_busy_tryupgrade:
*
* Attempt to upgrade a single shared busy into an exclusive busy.
*/
int
vm_page_busy_tryupgrade(vm_page_t m)
{
u_int ce, x;
vm_page_assert_sbusied(m);
x = vm_page_busy_fetch(m);
ce = VPB_CURTHREAD_EXCLUSIVE;
for (;;) {
if (VPB_SHARERS(x) > 1)
return (0);
KASSERT((x & ~VPB_BIT_WAITERS) == VPB_SHARERS_WORD(1),
("vm_page_busy_tryupgrade: invalid lock state"));
if (!atomic_fcmpset_acq_int(&m->busy_lock, &x,
ce | (x & VPB_BIT_WAITERS)))
continue;
return (1);
}
}
/*
* vm_page_sbusied:
*
* Return a positive value if the page is shared busied, 0 otherwise.
*/
int
vm_page_sbusied(vm_page_t m)
{
u_int x;
x = vm_page_busy_fetch(m);
return ((x & VPB_BIT_SHARED) != 0 && x != VPB_UNBUSIED);
}
/*
* vm_page_sunbusy:
*
* Shared unbusy a page.
*/
void
vm_page_sunbusy(vm_page_t m)
{
u_int x;
vm_page_assert_sbusied(m);
x = vm_page_busy_fetch(m);
for (;;) {
KASSERT(x != VPB_FREED,
("vm_page_sunbusy: Unlocking freed page."));
if (VPB_SHARERS(x) > 1) {
if (atomic_fcmpset_int(&m->busy_lock, &x,
x - VPB_ONE_SHARER))
break;
continue;
}
KASSERT((x & ~VPB_BIT_WAITERS) == VPB_SHARERS_WORD(1),
("vm_page_sunbusy: invalid lock state"));
if (!atomic_fcmpset_rel_int(&m->busy_lock, &x, VPB_UNBUSIED))
continue;
if ((x & VPB_BIT_WAITERS) == 0)
break;
wakeup(m);
break;
}
}
/*
* vm_page_busy_sleep:
*
* Sleep if the page is busy, using the page pointer as wchan.
* This is used to implement the hard-path of the busying mechanism.
*
* If VM_ALLOC_IGN_SBUSY is specified in allocflags, the function
* will not sleep if the page is shared-busy.
*
* The object lock must be held on entry.
*
* Returns true if it slept and dropped the object lock, or false
* if there was no sleep and the lock is still held.
*/
bool
vm_page_busy_sleep(vm_page_t m, const char *wmesg, int allocflags)
{
vm_object_t obj;
obj = m->object;
VM_OBJECT_ASSERT_LOCKED(obj);
return (_vm_page_busy_sleep(obj, m, m->pindex, wmesg, allocflags,
true));
}
/*
* vm_page_busy_sleep_unlocked:
*
* Sleep if the page is busy, using the page pointer as wchan.
* This is used to implement the hard-path of busying mechanism.
*
* If VM_ALLOC_IGN_SBUSY is specified in allocflags, the function
* will not sleep if the page is shared-busy.
*
* The object lock must not be held on entry. The operation will
* return if the page changes identity.
*/
void
vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m, vm_pindex_t pindex,
const char *wmesg, int allocflags)
{
VM_OBJECT_ASSERT_UNLOCKED(obj);
(void)_vm_page_busy_sleep(obj, m, pindex, wmesg, allocflags, false);
}
/*
* _vm_page_busy_sleep:
*
* Internal busy sleep function. Verifies the page identity and
* lockstate against parameters. Returns true if it sleeps and
* false otherwise.
*
* allocflags uses VM_ALLOC_* flags to specify the lock required.
*
* If locked is true the lock will be dropped for any true returns
* and held for any false returns.
*/
static bool
_vm_page_busy_sleep(vm_object_t obj, vm_page_t m, vm_pindex_t pindex,
const char *wmesg, int allocflags, bool locked)
{
bool xsleep;
u_int x;
/*
* If the object is busy we must wait for that to drain to zero
* before trying the page again.
*/
if (obj != NULL && vm_object_busied(obj)) {
if (locked)
VM_OBJECT_DROP(obj);
vm_object_busy_wait(obj, wmesg);
return (true);
}
if (!vm_page_busied(m))
return (false);
xsleep = (allocflags & (VM_ALLOC_SBUSY | VM_ALLOC_IGN_SBUSY)) != 0;
sleepq_lock(m);
x = vm_page_busy_fetch(m);
do {
/*
* If the page changes objects or becomes unlocked we can
* simply return.
*/
if (x == VPB_UNBUSIED ||
(xsleep && (x & VPB_BIT_SHARED) != 0) ||
m->object != obj || m->pindex != pindex) {
sleepq_release(m);
return (false);
}
if ((x & VPB_BIT_WAITERS) != 0)
break;
} while (!atomic_fcmpset_int(&m->busy_lock, &x, x | VPB_BIT_WAITERS));
if (locked)
VM_OBJECT_DROP(obj);
DROP_GIANT();
sleepq_add(m, NULL, wmesg, 0, 0);
sleepq_wait(m, PVM);
PICKUP_GIANT();
return (true);
}
/*
* vm_page_trysbusy:
*
* Try to shared busy a page.
* If the operation succeeds 1 is returned otherwise 0.
* The operation never sleeps.
*/
int
vm_page_trysbusy(vm_page_t m)
{
vm_object_t obj;
u_int x;
obj = m->object;
x = vm_page_busy_fetch(m);
for (;;) {
if ((x & VPB_BIT_SHARED) == 0)
return (0);
/*
* Reduce the window for transient busies that will trigger
* false negatives in vm_page_ps_test().
*/
if (obj != NULL && vm_object_busied(obj))
return (0);
if (atomic_fcmpset_acq_int(&m->busy_lock, &x,
x + VPB_ONE_SHARER))
break;
}
/* Refetch the object now that we're guaranteed that it is stable. */
obj = m->object;
if (obj != NULL && vm_object_busied(obj)) {
vm_page_sunbusy(m);
return (0);
}
return (1);
}
/*
* vm_page_tryxbusy:
*
* Try to exclusive busy a page.
* If the operation succeeds 1 is returned otherwise 0.
* The operation never sleeps.
*/
int
vm_page_tryxbusy(vm_page_t m)
{
vm_object_t obj;
if (atomic_cmpset_acq_int(&m->busy_lock, VPB_UNBUSIED,
VPB_CURTHREAD_EXCLUSIVE) == 0)
return (0);
obj = m->object;
if (obj != NULL && vm_object_busied(obj)) {
vm_page_xunbusy(m);
return (0);
}
return (1);
}
static void
vm_page_xunbusy_hard_tail(vm_page_t m)
{
atomic_store_rel_int(&m->busy_lock, VPB_UNBUSIED);
/* Wake the waiter. */
wakeup(m);
}
/*
* vm_page_xunbusy_hard:
*
* Called when unbusy has failed because there is a waiter.
*/
void
vm_page_xunbusy_hard(vm_page_t m)
{
vm_page_assert_xbusied(m);
vm_page_xunbusy_hard_tail(m);
}
void
vm_page_xunbusy_hard_unchecked(vm_page_t m)
{
vm_page_assert_xbusied_unchecked(m);
vm_page_xunbusy_hard_tail(m);
}
static void
vm_page_busy_free(vm_page_t m)
{
u_int x;
atomic_thread_fence_rel();
x = atomic_swap_int(&m->busy_lock, VPB_FREED);
if ((x & VPB_BIT_WAITERS) != 0)
wakeup(m);
}
/*
* vm_page_unhold_pages:
*
* Unhold each of the pages that is referenced by the given array.
*/
void
vm_page_unhold_pages(vm_page_t *ma, int count)
{
for (; count != 0; count--) {
vm_page_unwire(*ma, PQ_ACTIVE);
ma++;
}
}
vm_page_t
PHYS_TO_VM_PAGE(vm_paddr_t pa)
{
vm_page_t m;
#ifdef VM_PHYSSEG_SPARSE
m = vm_phys_paddr_to_vm_page(pa);
if (m == NULL)
m = vm_phys_fictitious_to_vm_page(pa);
return (m);
#elif defined(VM_PHYSSEG_DENSE)
long pi;
pi = atop(pa);
if (pi >= first_page && (pi - first_page) < vm_page_array_size) {
m = &vm_page_array[pi - first_page];
return (m);
}
return (vm_phys_fictitious_to_vm_page(pa));
#else
#error "Either VM_PHYSSEG_DENSE or VM_PHYSSEG_SPARSE must be defined."
#endif
}
/*
* vm_page_getfake:
*
* Create a fictitious page with the specified physical address and
* memory attribute. The memory attribute is the only the machine-
* dependent aspect of a fictitious page that must be initialized.
*/
vm_page_t
vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr)
{
vm_page_t m;
m = uma_zalloc(fakepg_zone, M_WAITOK | M_ZERO);
vm_page_initfake(m, paddr, memattr);
return (m);
}
void
vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
{
if ((m->flags & PG_FICTITIOUS) != 0) {
/*
* The page's memattr might have changed since the
* previous initialization. Update the pmap to the
* new memattr.
*/
goto memattr;
}
m->phys_addr = paddr;
m->a.queue = PQ_NONE;
/* Fictitious pages don't use "segind". */
m->flags = PG_FICTITIOUS;
/* Fictitious pages don't use "order" or "pool". */
m->oflags = VPO_UNMANAGED;
m->busy_lock = VPB_CURTHREAD_EXCLUSIVE;
/* Fictitious pages are unevictable. */
m->ref_count = 1;
pmap_page_init(m);
memattr:
pmap_page_set_memattr(m, memattr);
}
/*
* vm_page_putfake:
*
* Release a fictitious page.
*/
void
vm_page_putfake(vm_page_t m)
{
KASSERT((m->oflags & VPO_UNMANAGED) != 0, ("managed %p", m));
KASSERT((m->flags & PG_FICTITIOUS) != 0,
("vm_page_putfake: bad page %p", m));
vm_page_assert_xbusied(m);
vm_page_busy_free(m);
uma_zfree(fakepg_zone, m);
}
/*
* vm_page_updatefake:
*
* Update the given fictitious page to the specified physical address and
* memory attribute.
*/
void
vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr)
{
KASSERT((m->flags & PG_FICTITIOUS) != 0,
("vm_page_updatefake: bad page %p", m));
m->phys_addr = paddr;
pmap_page_set_memattr(m, memattr);
}
/*
* vm_page_free:
*
* Free a page.
*/
void
vm_page_free(vm_page_t m)
{
m->flags &= ~PG_ZERO;
vm_page_free_toq(m);
}
+/*
+ * vm_page_iter_free:
+ *
+ * Free the current page, as identified by iterator.
+ */
+void
+vm_page_iter_free(struct pctrie_iter *pages)
+{
+ vm_page_t m;
+
+ m = vm_radix_iter_page(pages);
+ vm_radix_iter_remove(pages);
+ m->flags &= ~PG_ZERO;
+ vm_page_free_toq_impl(m, false);
+}
+
/*
* vm_page_free_zero:
*
* Free a page to the zerod-pages queue
*/
void
vm_page_free_zero(vm_page_t m)
{
m->flags |= PG_ZERO;
vm_page_free_toq(m);
}
/*
* Unbusy and handle the page queueing for a page from a getpages request that
* was optionally read ahead or behind.
*/
void
vm_page_readahead_finish(vm_page_t m)
{
/* We shouldn't put invalid pages on queues. */
KASSERT(!vm_page_none_valid(m), ("%s: %p is invalid", __func__, m));
/*
* Since the page is not the actually needed one, whether it should
* be activated or deactivated is not obvious. Empirical results
* have shown that deactivating the page is usually the best choice,
* unless the page is wanted by another thread.
*/
if ((vm_page_busy_fetch(m) & VPB_BIT_WAITERS) != 0)
vm_page_activate(m);
else
vm_page_deactivate(m);
vm_page_xunbusy_unchecked(m);
}
/*
* Destroy the identity of an invalid page and free it if possible.
* This is intended to be used when reading a page from backing store fails.
*/
void
vm_page_free_invalid(vm_page_t m)
{
KASSERT(vm_page_none_valid(m), ("page %p is valid", m));
KASSERT(!pmap_page_is_mapped(m), ("page %p is mapped", m));
KASSERT(m->object != NULL, ("page %p has no object", m));
VM_OBJECT_ASSERT_WLOCKED(m->object);
/*
* We may be attempting to free the page as part of the handling for an
* I/O error, in which case the page was xbusied by a different thread.
*/
vm_page_xbusy_claim(m);
/*
* If someone has wired this page while the object lock
* was not held, then the thread that unwires is responsible
* for freeing the page. Otherwise just free the page now.
* The wire count of this unmapped page cannot change while
* we have the page xbusy and the page's object wlocked.
*/
if (vm_page_remove(m))
vm_page_free(m);
}
/*
* vm_page_dirty_KBI: [ internal use only ]
*
* Set all bits in the page's dirty field.
*
* The object containing the specified page must be locked if the
* call is made from the machine-independent layer.
*
* See vm_page_clear_dirty_mask().
*
* This function should only be called by vm_page_dirty().
*/
void
vm_page_dirty_KBI(vm_page_t m)
{
/* Refer to this operation by its public name. */
KASSERT(vm_page_all_valid(m), ("vm_page_dirty: page is invalid!"));
m->dirty = VM_PAGE_BITS_ALL;
}
/*
* Insert the given page into the given object at the given pindex. mpred is
* used for memq linkage. From vm_page_insert, lookup is true, mpred is
* initially NULL, and this procedure looks it up. From vm_page_insert_after
* and vm_page_iter_insert, lookup is false and mpred is known to the caller
* to be valid, and may be NULL if this will be the page with the lowest
* pindex.
*
* The procedure is marked __always_inline to suggest to the compiler to
* eliminate the lookup parameter and the associated alternate branch.
*/
static __always_inline int
vm_page_insert_lookup(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
struct pctrie_iter *pages, bool iter, vm_page_t mpred, bool lookup)
{
int error;
VM_OBJECT_ASSERT_WLOCKED(object);
KASSERT(m->object == NULL,
("vm_page_insert: page %p already inserted", m));
/*
* Record the object/offset pair in this page.
*/
m->object = object;
m->pindex = pindex;
m->ref_count |= VPRC_OBJREF;
/*
* Add this page to the object's radix tree, and look up mpred if
* needed.
*/
if (iter) {
KASSERT(!lookup, ("%s: cannot lookup mpred", __func__));
error = vm_radix_iter_insert(pages, m);
} else if (lookup)
error = vm_radix_insert_lookup_lt(&object->rtree, m, &mpred);
else
error = vm_radix_insert(&object->rtree, m);
if (__predict_false(error != 0)) {
m->object = NULL;
m->pindex = 0;
m->ref_count &= ~VPRC_OBJREF;
return (1);
}
/*
* Now link into the object's ordered list of backed pages.
*/
vm_page_insert_radixdone(m, object, mpred);
vm_pager_page_inserted(object, m);
return (0);
}
/*
* vm_page_insert: [ internal use only ]
*
* Inserts the given mem entry into the object and object list.
*
* The object must be locked.
*/
int
vm_page_insert(vm_page_t m, vm_object_t object, vm_pindex_t pindex)
{
return (vm_page_insert_lookup(m, object, pindex, NULL, false, NULL,
true));
}
/*
* vm_page_insert_after:
*
* Inserts the page "m" into the specified object at offset "pindex".
*
* The page "mpred" must immediately precede the offset "pindex" within
* the specified object.
*
* The object must be locked.
*/
static int
vm_page_insert_after(vm_page_t m, vm_object_t object, vm_pindex_t pindex,
vm_page_t mpred)
{
return (vm_page_insert_lookup(m, object, pindex, NULL, false, mpred,
false));
}
/*
* vm_page_iter_insert:
*
* Tries to insert the page "m" into the specified object at offset
* "pindex" using the iterator "pages". Returns 0 if the insertion was
* successful.
*
* The page "mpred" must immediately precede the offset "pindex" within
* the specified object.
*
* The object must be locked.
*/
static int
vm_page_iter_insert(struct pctrie_iter *pages, vm_page_t m, vm_object_t object,
vm_pindex_t pindex, vm_page_t mpred)
{
return (vm_page_insert_lookup(m, object, pindex, pages, true, mpred,
false));
}
/*
* vm_page_insert_radixdone:
*
* Complete page "m" insertion into the specified object after the
* radix trie hooking.
*
* The page "mpred" must precede the offset "m->pindex" within the
* specified object.
*
* The object must be locked.
*/
static void
vm_page_insert_radixdone(vm_page_t m, vm_object_t object, vm_page_t mpred)
{
VM_OBJECT_ASSERT_WLOCKED(object);
KASSERT(object != NULL && m->object == object,
("vm_page_insert_radixdone: page %p has inconsistent object", m));
KASSERT((m->ref_count & VPRC_OBJREF) != 0,
("vm_page_insert_radixdone: page %p is missing object ref", m));
if (mpred != NULL) {
KASSERT(mpred->object == object,
("vm_page_insert_radixdone: object doesn't contain mpred"));
KASSERT(mpred->pindex < m->pindex,
("vm_page_insert_radixdone: mpred doesn't precede pindex"));
KASSERT(TAILQ_NEXT(mpred, listq) == NULL ||
m->pindex < TAILQ_NEXT(mpred, listq)->pindex,
("vm_page_insert_radixdone: pindex doesn't precede msucc"));
} else {
KASSERT(TAILQ_EMPTY(&object->memq) ||
m->pindex < TAILQ_FIRST(&object->memq)->pindex,
("vm_page_insert_radixdone: no mpred but not first page"));
}
if (mpred != NULL)
TAILQ_INSERT_AFTER(&object->memq, mpred, m, listq);
else
TAILQ_INSERT_HEAD(&object->memq, m, listq);
/*
* Show that the object has one more resident page.
*/
object->resident_page_count++;
/*
* Hold the vnode until the last page is released.
*/
if (object->resident_page_count == 1 && object->type == OBJT_VNODE)
vhold(object->handle);
/*
* Since we are inserting a new and possibly dirty page,
* update the object's generation count.
*/
if (pmap_page_is_write_mapped(m))
vm_object_set_writeable_dirty(object);
}
/*
- * Do the work to remove a page from its object. The caller is responsible for
- * updating the page's fields to reflect this removal.
+ * vm_page_remove_radixdone
+ *
+ * Complete page "m" removal from the specified object after the radix trie
+ * unhooking.
+ *
+ * The caller is responsible for updating the page's fields to reflect this
+ * removal.
*/
static void
-vm_page_object_remove(vm_page_t m)
+vm_page_remove_radixdone(vm_page_t m)
{
vm_object_t object;
- vm_page_t mrem __diagused;
vm_page_assert_xbusied(m);
object = m->object;
VM_OBJECT_ASSERT_WLOCKED(object);
KASSERT((m->ref_count & VPRC_OBJREF) != 0,
("page %p is missing its object ref", m));
/* Deferred free of swap space. */
if ((m->a.flags & PGA_SWAP_FREE) != 0)
vm_pager_page_unswapped(m);
vm_pager_page_removed(object, m);
-
m->object = NULL;
- mrem = vm_radix_remove(&object->rtree, m->pindex);
- KASSERT(mrem == m, ("removed page %p, expected page %p", mrem, m));
/*
* Now remove from the object's list of backed pages.
*/
TAILQ_REMOVE(&object->memq, m, listq);
/*
* And show that the object has one fewer resident page.
*/
object->resident_page_count--;
/*
* The vnode may now be recycled.
*/
if (object->resident_page_count == 0 && object->type == OBJT_VNODE)
vdrop(object->handle);
}
/*
* vm_page_remove:
*
* Removes the specified page from its containing object, but does not
* invalidate any backing storage. Returns true if the object's reference
* was the last reference to the page, and false otherwise.
*
* The object must be locked and the page must be exclusively busied.
* The exclusive busy will be released on return. If this is not the
* final ref and the caller does not hold a wire reference it may not
* continue to access the page.
*/
bool
vm_page_remove(vm_page_t m)
{
bool dropped;
dropped = vm_page_remove_xbusy(m);
vm_page_xunbusy(m);
return (dropped);
}
+/*
+ * vm_page_iter_remove:
+ *
+ * Remove the current page, as identified by iterator, and remove it from the
+ * radix tree.
+ */
+bool
+vm_page_iter_remove(struct pctrie_iter *pages)
+{
+ vm_page_t m;
+ bool dropped;
+
+ m = vm_radix_iter_page(pages);
+ vm_radix_iter_remove(pages);
+ vm_page_remove_radixdone(m);
+ dropped = (vm_page_drop(m, VPRC_OBJREF) == VPRC_OBJREF);
+ vm_page_xunbusy(m);
+
+ return (dropped);
+}
+
+/*
+ * vm_page_radix_remove
+ *
+ * Removes the specified page from the radix tree.
+ */
+static void
+vm_page_radix_remove(vm_page_t m)
+{
+ vm_page_t mrem __diagused;
+
+ mrem = vm_radix_remove(&m->object->rtree, m->pindex);
+ KASSERT(mrem == m,
+ ("removed page %p, expected page %p", mrem, m));
+}
+
/*
* vm_page_remove_xbusy
*
* Removes the page but leaves the xbusy held. Returns true if this
* removed the final ref and false otherwise.
*/
bool
vm_page_remove_xbusy(vm_page_t m)
{
- vm_page_object_remove(m);
+ vm_page_radix_remove(m);
+ vm_page_remove_radixdone(m);
return (vm_page_drop(m, VPRC_OBJREF) == VPRC_OBJREF);
}
/*
* vm_page_lookup:
*
* Returns the page associated with the object/offset
* pair specified; if none is found, NULL is returned.
*
* The object must be locked.
*/
vm_page_t
vm_page_lookup(vm_object_t object, vm_pindex_t pindex)
{
VM_OBJECT_ASSERT_LOCKED(object);
return (vm_radix_lookup(&object->rtree, pindex));
}
/*
* vm_page_iter_init:
*
* Initialize iterator for vm pages.
*/
void
vm_page_iter_init(struct pctrie_iter *pages, vm_object_t object)
{
vm_radix_iter_init(pages, &object->rtree);
}
/*
* vm_page_iter_init:
*
* Initialize iterator for vm pages.
*/
void
vm_page_iter_limit_init(struct pctrie_iter *pages, vm_object_t object,
vm_pindex_t limit)
{
vm_radix_iter_limit_init(pages, &object->rtree, limit);
}
/*
* vm_page_iter_lookup:
*
* Returns the page associated with the object/offset pair specified, and
* stores the path to its position; if none is found, NULL is returned.
*
* The iter pctrie must be locked.
*/
vm_page_t
vm_page_iter_lookup(struct pctrie_iter *pages, vm_pindex_t pindex)
{
return (vm_radix_iter_lookup(pages, pindex));
}
/*
* vm_page_lookup_unlocked:
*
* Returns the page associated with the object/offset pair specified;
* if none is found, NULL is returned. The page may be no longer be
* present in the object at the time that this function returns. Only
* useful for opportunistic checks such as inmem().
*/
vm_page_t
vm_page_lookup_unlocked(vm_object_t object, vm_pindex_t pindex)
{
return (vm_radix_lookup_unlocked(&object->rtree, pindex));
}
/*
* vm_page_relookup:
*
* Returns a page that must already have been busied by
* the caller. Used for bogus page replacement.
*/
vm_page_t
vm_page_relookup(vm_object_t object, vm_pindex_t pindex)
{
vm_page_t m;
m = vm_page_lookup_unlocked(object, pindex);
KASSERT(m != NULL && (vm_page_busied(m) || vm_page_wired(m)) &&
m->object == object && m->pindex == pindex,
("vm_page_relookup: Invalid page %p", m));
return (m);
}
/*
* This should only be used by lockless functions for releasing transient
* incorrect acquires. The page may have been freed after we acquired a
* busy lock. In this case busy_lock == VPB_FREED and we have nothing
* further to do.
*/
static void
vm_page_busy_release(vm_page_t m)
{
u_int x;
x = vm_page_busy_fetch(m);
for (;;) {
if (x == VPB_FREED)
break;
if ((x & VPB_BIT_SHARED) != 0 && VPB_SHARERS(x) > 1) {
if (atomic_fcmpset_int(&m->busy_lock, &x,
x - VPB_ONE_SHARER))
break;
continue;
}
KASSERT((x & VPB_BIT_SHARED) != 0 ||
(x & ~VPB_BIT_WAITERS) == VPB_CURTHREAD_EXCLUSIVE,
("vm_page_busy_release: %p xbusy not owned.", m));
if (!atomic_fcmpset_rel_int(&m->busy_lock, &x, VPB_UNBUSIED))
continue;
if ((x & VPB_BIT_WAITERS) != 0)
wakeup(m);
break;
}
}
/*
* vm_page_find_least:
*
* Returns the page associated with the object with least pindex
* greater than or equal to the parameter pindex, or NULL.
*
* The object must be locked.
*/
vm_page_t
vm_page_find_least(vm_object_t object, vm_pindex_t pindex)
{
vm_page_t m;
VM_OBJECT_ASSERT_LOCKED(object);
if ((m = TAILQ_FIRST(&object->memq)) != NULL && m->pindex < pindex)
m = vm_radix_lookup_ge(&object->rtree, pindex);
return (m);
}
/*
* vm_page_iter_lookup_ge:
*
* Returns the page associated with the object with least pindex
* greater than or equal to the parameter pindex, or NULL. Initializes the
* iterator to point to that page.
*
* The iter pctrie must be locked.
*/
vm_page_t
vm_page_iter_lookup_ge(struct pctrie_iter *pages, vm_pindex_t pindex)
{
return (vm_radix_iter_lookup_ge(pages, pindex));
}
/*
* Returns the given page's successor (by pindex) within the object if it is
* resident; if none is found, NULL is returned.
*
* The object must be locked.
*/
vm_page_t
vm_page_next(vm_page_t m)
{
vm_page_t next;
VM_OBJECT_ASSERT_LOCKED(m->object);
if ((next = TAILQ_NEXT(m, listq)) != NULL) {
MPASS(next->object == m->object);
if (next->pindex != m->pindex + 1)
next = NULL;
}
return (next);
}
/*
* Returns the given page's predecessor (by pindex) within the object if it is
* resident; if none is found, NULL is returned.
*
* The object must be locked.
*/
vm_page_t
vm_page_prev(vm_page_t m)
{
vm_page_t prev;
VM_OBJECT_ASSERT_LOCKED(m->object);
if ((prev = TAILQ_PREV(m, pglist, listq)) != NULL) {
MPASS(prev->object == m->object);
if (prev->pindex != m->pindex - 1)
prev = NULL;
}
return (prev);
}
/*
* Uses the page mnew as a replacement for an existing page at index
* pindex which must be already present in the object.
*
* Both pages must be exclusively busied on enter. The old page is
* unbusied on exit.
*
* A return value of true means mold is now free. If this is not the
* final ref and the caller does not hold a wire reference it may not
* continue to access the page.
*/
static bool
vm_page_replace_hold(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
vm_page_t mold)
{
vm_page_t mret __diagused;
bool dropped;
VM_OBJECT_ASSERT_WLOCKED(object);
vm_page_assert_xbusied(mold);
KASSERT(mnew->object == NULL && (mnew->ref_count & VPRC_OBJREF) == 0,
("vm_page_replace: page %p already in object", mnew));
/*
* This function mostly follows vm_page_insert() and
* vm_page_remove() without the radix, object count and vnode
* dance. Double check such functions for more comments.
*/
mnew->object = object;
mnew->pindex = pindex;
atomic_set_int(&mnew->ref_count, VPRC_OBJREF);
mret = vm_radix_replace(&object->rtree, mnew);
KASSERT(mret == mold,
("invalid page replacement, mold=%p, mret=%p", mold, mret));
KASSERT((mold->oflags & VPO_UNMANAGED) ==
(mnew->oflags & VPO_UNMANAGED),
("vm_page_replace: mismatched VPO_UNMANAGED"));
/* Keep the resident page list in sorted order. */
TAILQ_INSERT_AFTER(&object->memq, mold, mnew, listq);
TAILQ_REMOVE(&object->memq, mold, listq);
mold->object = NULL;
/*
* The object's resident_page_count does not change because we have
* swapped one page for another, but the generation count should
* change if the page is dirty.
*/
if (pmap_page_is_write_mapped(mnew))
vm_object_set_writeable_dirty(object);
dropped = vm_page_drop(mold, VPRC_OBJREF) == VPRC_OBJREF;
vm_page_xunbusy(mold);
return (dropped);
}
void
vm_page_replace(vm_page_t mnew, vm_object_t object, vm_pindex_t pindex,
vm_page_t mold)
{
vm_page_assert_xbusied(mnew);
if (vm_page_replace_hold(mnew, object, pindex, mold))
vm_page_free(mold);
}
/*
* vm_page_rename:
*
- * Move the given memory entry from its
- * current object to the specified target object/offset.
+ * Move the current page, as identified by iterator, from its current
+ * object to the specified target object/offset.
*
* Note: swap associated with the page must be invalidated by the move. We
* have to do this for several reasons: (1) we aren't freeing the
* page, (2) we are dirtying the page, (3) the VM system is probably
* moving the page from object A to B, and will then later move
* the backing store from A to B and we can't have a conflict.
*
* Note: we *always* dirty the page. It is necessary both for the
* fact that we moved it, and because we may be invalidating
* swap.
*
* The objects must be locked.
*/
int
-vm_page_rename(vm_page_t m, vm_object_t new_object, vm_pindex_t new_pindex)
+vm_page_rename(struct pctrie_iter *pages,
+ vm_object_t new_object, vm_pindex_t new_pindex)
{
- vm_page_t mpred;
+ vm_page_t m, mpred;
vm_pindex_t opidx;
VM_OBJECT_ASSERT_WLOCKED(new_object);
+ m = vm_radix_iter_page(pages);
KASSERT(m->ref_count != 0, ("vm_page_rename: page %p has no refs", m));
/*
* Create a custom version of vm_page_insert() which does not depend
* by m_prev and can cheat on the implementation aspects of the
* function.
*/
opidx = m->pindex;
m->pindex = new_pindex;
if (vm_radix_insert_lookup_lt(&new_object->rtree, m, &mpred) != 0) {
m->pindex = opidx;
return (1);
}
/*
* The operation cannot fail anymore. The removal must happen before
* the listq iterator is tainted.
*/
m->pindex = opidx;
- vm_page_object_remove(m);
+ vm_radix_iter_remove(pages);
+ vm_page_remove_radixdone(m);
/* Return back to the new pindex to complete vm_page_insert(). */
m->pindex = new_pindex;
m->object = new_object;
vm_page_insert_radixdone(m, new_object, mpred);
vm_page_dirty(m);
vm_pager_page_inserted(new_object, m);
return (0);
}
/*
* vm_page_mpred:
*
* Return the greatest page of the object with index <= pindex,
* or NULL, if there is none. Assumes object lock is held.
*/
vm_page_t
vm_page_mpred(vm_object_t object, vm_pindex_t pindex)
{
return (vm_radix_lookup_le(&object->rtree, pindex));
}
/*
* vm_page_alloc:
*
* Allocate and return a page that is associated with the specified
* object and offset pair. By default, this page is exclusive busied.
*
* The caller must always specify an allocation class.
*
* allocation classes:
* VM_ALLOC_NORMAL normal process request
* VM_ALLOC_SYSTEM system *really* needs a page
* VM_ALLOC_INTERRUPT interrupt time request
*
* optional allocation flags:
* VM_ALLOC_COUNT(number) the number of additional pages that the caller
* intends to allocate
* VM_ALLOC_NOBUSY do not exclusive busy the page
* VM_ALLOC_NODUMP do not include the page in a kernel core dump
* VM_ALLOC_SBUSY shared busy the allocated page
* VM_ALLOC_WIRED wire the allocated page
* VM_ALLOC_ZERO prefer a zeroed page
*/
vm_page_t
vm_page_alloc(vm_object_t object, vm_pindex_t pindex, int req)
{
return (vm_page_alloc_after(object, pindex, req,
vm_page_mpred(object, pindex)));
}
/*
* Allocate a page in the specified object with the given page index. To
* optimize insertion of the page into the object, the caller must also specify
* the resident page in the object with largest index smaller than the given
* page index, or NULL if no such page exists.
*/
static vm_page_t
vm_page_alloc_after(vm_object_t object, vm_pindex_t pindex,
int req, vm_page_t mpred)
{
struct vm_domainset_iter di;
vm_page_t m;
int domain;
vm_domainset_iter_page_init(&di, object, pindex, &domain, &req);
do {
m = vm_page_alloc_domain_after(object, pindex, domain, req,
mpred);
if (m != NULL)
break;
} while (vm_domainset_iter_page(&di, object, &domain) == 0);
return (m);
}
/*
* Returns true if the number of free pages exceeds the minimum
* for the request class and false otherwise.
*/
static int
_vm_domain_allocate(struct vm_domain *vmd, int req_class, int npages)
{
u_int limit, old, new;
if (req_class == VM_ALLOC_INTERRUPT)
limit = 0;
else if (req_class == VM_ALLOC_SYSTEM)
limit = vmd->vmd_interrupt_free_min;
else
limit = vmd->vmd_free_reserved;
/*
* Attempt to reserve the pages. Fail if we're below the limit.
*/
limit += npages;
old = atomic_load_int(&vmd->vmd_free_count);
do {
if (old < limit)
return (0);
new = old - npages;
} while (atomic_fcmpset_int(&vmd->vmd_free_count, &old, new) == 0);
/* Wake the page daemon if we've crossed the threshold. */
if (vm_paging_needed(vmd, new) && !vm_paging_needed(vmd, old))
pagedaemon_wakeup(vmd->vmd_domain);
/* Only update bitsets on transitions. */
if ((old >= vmd->vmd_free_min && new < vmd->vmd_free_min) ||
(old >= vmd->vmd_free_severe && new < vmd->vmd_free_severe))
vm_domain_set(vmd);
return (1);
}
int
vm_domain_allocate(struct vm_domain *vmd, int req, int npages)
{
int req_class;
/*
* The page daemon is allowed to dig deeper into the free page list.
*/
req_class = req & VM_ALLOC_CLASS_MASK;
if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
req_class = VM_ALLOC_SYSTEM;
return (_vm_domain_allocate(vmd, req_class, npages));
}
vm_page_t
vm_page_alloc_domain_after(vm_object_t object, vm_pindex_t pindex, int domain,
int req, vm_page_t mpred)
{
struct vm_domain *vmd;
vm_page_t m;
int flags;
#define VPA_FLAGS (VM_ALLOC_CLASS_MASK | VM_ALLOC_WAITFAIL | \
VM_ALLOC_NOWAIT | VM_ALLOC_NOBUSY | \
VM_ALLOC_SBUSY | VM_ALLOC_WIRED | \
VM_ALLOC_NODUMP | VM_ALLOC_ZERO | \
VM_ALLOC_NOFREE | VM_ALLOC_COUNT_MASK)
KASSERT((req & ~VPA_FLAGS) == 0,
("invalid request %#x", req));
KASSERT(((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
("invalid request %#x", req));
KASSERT(mpred == NULL || mpred->pindex < pindex,
("mpred %p doesn't precede pindex 0x%jx", mpred,
(uintmax_t)pindex));
VM_OBJECT_ASSERT_WLOCKED(object);
flags = 0;
m = NULL;
if (!vm_pager_can_alloc_page(object, pindex))
return (NULL);
again:
if (__predict_false((req & VM_ALLOC_NOFREE) != 0)) {
m = vm_page_alloc_nofree_domain(domain, req);
if (m != NULL)
goto found;
}
#if VM_NRESERVLEVEL > 0
/*
* Can we allocate the page from a reservation?
*/
if (vm_object_reserv(object) &&
(m = vm_reserv_alloc_page(object, pindex, domain, req, mpred)) !=
NULL) {
goto found;
}
#endif
vmd = VM_DOMAIN(domain);
if (vmd->vmd_pgcache[VM_FREEPOOL_DEFAULT].zone != NULL) {
m = uma_zalloc(vmd->vmd_pgcache[VM_FREEPOOL_DEFAULT].zone,
M_NOWAIT | M_NOVM);
if (m != NULL) {
flags |= PG_PCPU_CACHE;
goto found;
}
}
if (vm_domain_allocate(vmd, req, 1)) {
/*
* If not, allocate it from the free page queues.
*/
vm_domain_free_lock(vmd);
m = vm_phys_alloc_pages(domain, VM_FREEPOOL_DEFAULT, 0);
vm_domain_free_unlock(vmd);
if (m == NULL) {
vm_domain_freecnt_inc(vmd, 1);
#if VM_NRESERVLEVEL > 0
if (vm_reserv_reclaim_inactive(domain))
goto again;
#endif
}
}
if (m == NULL) {
/*
* Not allocatable, give up.
*/
if (vm_domain_alloc_fail(vmd, object, req))
goto again;
return (NULL);
}
/*
* At this point we had better have found a good page.
*/
found:
vm_page_dequeue(m);
vm_page_alloc_check(m);
/*
* Initialize the page. Only the PG_ZERO flag is inherited.
*/
flags |= m->flags & PG_ZERO;
if ((req & VM_ALLOC_NODUMP) != 0)
flags |= PG_NODUMP;
if ((req & VM_ALLOC_NOFREE) != 0)
flags |= PG_NOFREE;
m->flags = flags;
m->a.flags = 0;
m->oflags = (object->flags & OBJ_UNMANAGED) != 0 ? VPO_UNMANAGED : 0;
if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
m->busy_lock = VPB_CURTHREAD_EXCLUSIVE;
else if ((req & VM_ALLOC_SBUSY) != 0)
m->busy_lock = VPB_SHARERS_WORD(1);
else
m->busy_lock = VPB_UNBUSIED;
if (req & VM_ALLOC_WIRED) {
vm_wire_add(1);
m->ref_count = 1;
}
m->a.act_count = 0;
if (vm_page_insert_after(m, object, pindex, mpred)) {
if (req & VM_ALLOC_WIRED) {
vm_wire_sub(1);
m->ref_count = 0;
}
KASSERT(m->object == NULL, ("page %p has object", m));
m->oflags = VPO_UNMANAGED;
m->busy_lock = VPB_UNBUSIED;
/* Don't change PG_ZERO. */
vm_page_free_toq(m);
if (req & VM_ALLOC_WAITFAIL) {
VM_OBJECT_WUNLOCK(object);
vm_radix_wait();
VM_OBJECT_WLOCK(object);
}
return (NULL);
}
/* Ignore device objects; the pager sets "memattr" for them. */
if (object->memattr != VM_MEMATTR_DEFAULT &&
(object->flags & OBJ_FICTITIOUS) == 0)
pmap_page_set_memattr(m, object->memattr);
return (m);
}
/*
* vm_page_alloc_contig:
*
* Allocate a contiguous set of physical pages of the given size "npages"
* from the free lists. All of the physical pages must be at or above
* the given physical address "low" and below the given physical address
* "high". The given value "alignment" determines the alignment of the
* first physical page in the set. If the given value "boundary" is
* non-zero, then the set of physical pages cannot cross any physical
* address boundary that is a multiple of that value. Both "alignment"
* and "boundary" must be a power of two.
*
* If the specified memory attribute, "memattr", is VM_MEMATTR_DEFAULT,
* then the memory attribute setting for the physical pages is configured
* to the object's memory attribute setting. Otherwise, the memory
* attribute setting for the physical pages is configured to "memattr",
* overriding the object's memory attribute setting. However, if the
* object's memory attribute setting is not VM_MEMATTR_DEFAULT, then the
* memory attribute setting for the physical pages cannot be configured
* to VM_MEMATTR_DEFAULT.
*
* The specified object may not contain fictitious pages.
*
* The caller must always specify an allocation class.
*
* allocation classes:
* VM_ALLOC_NORMAL normal process request
* VM_ALLOC_SYSTEM system *really* needs a page
* VM_ALLOC_INTERRUPT interrupt time request
*
* optional allocation flags:
* VM_ALLOC_NOBUSY do not exclusive busy the page
* VM_ALLOC_NODUMP do not include the page in a kernel core dump
* VM_ALLOC_SBUSY shared busy the allocated page
* VM_ALLOC_WIRED wire the allocated page
* VM_ALLOC_ZERO prefer a zeroed page
*/
vm_page_t
vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
vm_paddr_t boundary, vm_memattr_t memattr)
{
struct vm_domainset_iter di;
vm_page_t bounds[2];
vm_page_t m;
int domain;
int start_segind;
start_segind = -1;
vm_domainset_iter_page_init(&di, object, pindex, &domain, &req);
do {
m = vm_page_alloc_contig_domain(object, pindex, domain, req,
npages, low, high, alignment, boundary, memattr);
if (m != NULL)
break;
if (start_segind == -1)
start_segind = vm_phys_lookup_segind(low);
if (vm_phys_find_range(bounds, start_segind, domain,
npages, low, high) == -1) {
vm_domainset_iter_ignore(&di, domain);
}
} while (vm_domainset_iter_page(&di, object, &domain) == 0);
return (m);
}
static vm_page_t
vm_page_find_contig_domain(int domain, int req, u_long npages, vm_paddr_t low,
vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
{
struct vm_domain *vmd;
vm_page_t m_ret;
/*
* Can we allocate the pages without the number of free pages falling
* below the lower bound for the allocation class?
*/
vmd = VM_DOMAIN(domain);
if (!vm_domain_allocate(vmd, req, npages))
return (NULL);
/*
* Try to allocate the pages from the free page queues.
*/
vm_domain_free_lock(vmd);
m_ret = vm_phys_alloc_contig(domain, npages, low, high,
alignment, boundary);
vm_domain_free_unlock(vmd);
if (m_ret != NULL)
return (m_ret);
#if VM_NRESERVLEVEL > 0
/*
* Try to break a reservation to allocate the pages.
*/
if ((req & VM_ALLOC_NORECLAIM) == 0) {
m_ret = vm_reserv_reclaim_contig(domain, npages, low,
high, alignment, boundary);
if (m_ret != NULL)
return (m_ret);
}
#endif
vm_domain_freecnt_inc(vmd, npages);
return (NULL);
}
vm_page_t
vm_page_alloc_contig_domain(vm_object_t object, vm_pindex_t pindex, int domain,
int req, u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
vm_paddr_t boundary, vm_memattr_t memattr)
{
struct pctrie_iter pages;
vm_page_t m, m_ret, mpred;
u_int busy_lock, flags, oflags;
#define VPAC_FLAGS (VPA_FLAGS | VM_ALLOC_NORECLAIM)
KASSERT((req & ~VPAC_FLAGS) == 0,
("invalid request %#x", req));
KASSERT(((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
("invalid request %#x", req));
KASSERT((req & (VM_ALLOC_WAITOK | VM_ALLOC_NORECLAIM)) !=
(VM_ALLOC_WAITOK | VM_ALLOC_NORECLAIM),
("invalid request %#x", req));
VM_OBJECT_ASSERT_WLOCKED(object);
KASSERT((object->flags & OBJ_FICTITIOUS) == 0,
("vm_page_alloc_contig: object %p has fictitious pages",
object));
KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
vm_page_iter_init(&pages, object);
mpred = vm_radix_iter_lookup_le(&pages, pindex);
KASSERT(mpred == NULL || mpred->pindex != pindex,
("vm_page_alloc_contig: pindex already allocated"));
for (;;) {
#if VM_NRESERVLEVEL > 0
/*
* Can we allocate the pages from a reservation?
*/
if (vm_object_reserv(object) &&
(m_ret = vm_reserv_alloc_contig(object, pindex, domain, req,
mpred, npages, low, high, alignment, boundary)) != NULL) {
break;
}
#endif
if ((m_ret = vm_page_find_contig_domain(domain, req, npages,
low, high, alignment, boundary)) != NULL)
break;
if (!vm_domain_alloc_fail(VM_DOMAIN(domain), object, req))
return (NULL);
}
/*
* Initialize the pages. Only the PG_ZERO flag is inherited.
*/
flags = PG_ZERO;
if ((req & VM_ALLOC_NODUMP) != 0)
flags |= PG_NODUMP;
oflags = (object->flags & OBJ_UNMANAGED) != 0 ? VPO_UNMANAGED : 0;
if ((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) == 0)
busy_lock = VPB_CURTHREAD_EXCLUSIVE;
else if ((req & VM_ALLOC_SBUSY) != 0)
busy_lock = VPB_SHARERS_WORD(1);
else
busy_lock = VPB_UNBUSIED;
if ((req & VM_ALLOC_WIRED) != 0)
vm_wire_add(npages);
if (object->memattr != VM_MEMATTR_DEFAULT &&
memattr == VM_MEMATTR_DEFAULT)
memattr = object->memattr;
for (m = m_ret; m < &m_ret[npages]; m++) {
vm_page_dequeue(m);
vm_page_alloc_check(m);
m->a.flags = 0;
m->flags = (m->flags | PG_NODUMP) & flags;
m->busy_lock = busy_lock;
if ((req & VM_ALLOC_WIRED) != 0)
m->ref_count = 1;
m->a.act_count = 0;
m->oflags = oflags;
if (vm_page_iter_insert(&pages, m, object, pindex, mpred)) {
if ((req & VM_ALLOC_WIRED) != 0)
vm_wire_sub(npages);
KASSERT(m->object == NULL,
("page %p has object", m));
mpred = m;
for (m = m_ret; m < &m_ret[npages]; m++) {
if (m <= mpred &&
(req & VM_ALLOC_WIRED) != 0)
m->ref_count = 0;
m->oflags = VPO_UNMANAGED;
m->busy_lock = VPB_UNBUSIED;
/* Don't change PG_ZERO. */
vm_page_free_toq(m);
}
if (req & VM_ALLOC_WAITFAIL) {
VM_OBJECT_WUNLOCK(object);
vm_radix_wait();
VM_OBJECT_WLOCK(object);
}
return (NULL);
}
mpred = m;
if (memattr != VM_MEMATTR_DEFAULT)
pmap_page_set_memattr(m, memattr);
pindex++;
}
return (m_ret);
}
/*
* Allocate a physical page that is not intended to be inserted into a VM
* object.
*/
vm_page_t
vm_page_alloc_noobj_domain(int domain, int req)
{
struct vm_domain *vmd;
vm_page_t m;
int flags;
#define VPAN_FLAGS (VM_ALLOC_CLASS_MASK | VM_ALLOC_WAITFAIL | \
VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | \
VM_ALLOC_NOBUSY | VM_ALLOC_WIRED | \
VM_ALLOC_NODUMP | VM_ALLOC_ZERO | \
VM_ALLOC_NOFREE | VM_ALLOC_COUNT_MASK)
KASSERT((req & ~VPAN_FLAGS) == 0,
("invalid request %#x", req));
flags = ((req & VM_ALLOC_NODUMP) != 0 ? PG_NODUMP : 0) |
((req & VM_ALLOC_NOFREE) != 0 ? PG_NOFREE : 0);
vmd = VM_DOMAIN(domain);
again:
if (__predict_false((req & VM_ALLOC_NOFREE) != 0)) {
m = vm_page_alloc_nofree_domain(domain, req);
if (m != NULL)
goto found;
}
if (vmd->vmd_pgcache[VM_FREEPOOL_DIRECT].zone != NULL) {
m = uma_zalloc(vmd->vmd_pgcache[VM_FREEPOOL_DIRECT].zone,
M_NOWAIT | M_NOVM);
if (m != NULL) {
flags |= PG_PCPU_CACHE;
goto found;
}
}
if (vm_domain_allocate(vmd, req, 1)) {
vm_domain_free_lock(vmd);
m = vm_phys_alloc_pages(domain, VM_FREEPOOL_DIRECT, 0);
vm_domain_free_unlock(vmd);
if (m == NULL) {
vm_domain_freecnt_inc(vmd, 1);
#if VM_NRESERVLEVEL > 0
if (vm_reserv_reclaim_inactive(domain))
goto again;
#endif
}
}
if (m == NULL) {
if (vm_domain_alloc_fail(vmd, NULL, req))
goto again;
return (NULL);
}
found:
vm_page_dequeue(m);
vm_page_alloc_check(m);
/*
* Consumers should not rely on a useful default pindex value.
*/
m->pindex = 0xdeadc0dedeadc0de;
m->flags = (m->flags & PG_ZERO) | flags;
m->a.flags = 0;
m->oflags = VPO_UNMANAGED;
m->busy_lock = VPB_UNBUSIED;
if ((req & VM_ALLOC_WIRED) != 0) {
vm_wire_add(1);
m->ref_count = 1;
}
if ((req & VM_ALLOC_ZERO) != 0 && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
return (m);
}
#if VM_NRESERVLEVEL > 1
#define VM_NOFREE_IMPORT_ORDER (VM_LEVEL_1_ORDER + VM_LEVEL_0_ORDER)
#elif VM_NRESERVLEVEL > 0
#define VM_NOFREE_IMPORT_ORDER VM_LEVEL_0_ORDER
#else
#define VM_NOFREE_IMPORT_ORDER 8
#endif
/*
* Allocate a single NOFREE page.
*
* This routine hands out NOFREE pages from higher-order
* physical memory blocks in order to reduce memory fragmentation.
* When a NOFREE for a given domain chunk is used up,
* the routine will try to fetch a new one from the freelists
* and discard the old one.
*/
static vm_page_t
vm_page_alloc_nofree_domain(int domain, int req)
{
vm_page_t m;
struct vm_domain *vmd;
struct vm_nofreeq *nqp;
KASSERT((req & VM_ALLOC_NOFREE) != 0, ("invalid request %#x", req));
vmd = VM_DOMAIN(domain);
nqp = &vmd->vmd_nofreeq;
vm_domain_free_lock(vmd);
if (nqp->offs >= (1 << VM_NOFREE_IMPORT_ORDER) || nqp->ma == NULL) {
if (!vm_domain_allocate(vmd, req,
1 << VM_NOFREE_IMPORT_ORDER)) {
vm_domain_free_unlock(vmd);
return (NULL);
}
nqp->ma = vm_phys_alloc_pages(domain, VM_FREEPOOL_DEFAULT,
VM_NOFREE_IMPORT_ORDER);
if (nqp->ma == NULL) {
vm_domain_freecnt_inc(vmd, 1 << VM_NOFREE_IMPORT_ORDER);
vm_domain_free_unlock(vmd);
return (NULL);
}
nqp->offs = 0;
}
m = &nqp->ma[nqp->offs++];
vm_domain_free_unlock(vmd);
VM_CNT_ADD(v_nofree_count, 1);
return (m);
}
vm_page_t
vm_page_alloc_noobj(int req)
{
struct vm_domainset_iter di;
vm_page_t m;
int domain;
vm_domainset_iter_page_init(&di, NULL, 0, &domain, &req);
do {
m = vm_page_alloc_noobj_domain(domain, req);
if (m != NULL)
break;
} while (vm_domainset_iter_page(&di, NULL, &domain) == 0);
return (m);
}
vm_page_t
vm_page_alloc_noobj_contig(int req, u_long npages, vm_paddr_t low,
vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr)
{
struct vm_domainset_iter di;
vm_page_t m;
int domain;
vm_domainset_iter_page_init(&di, NULL, 0, &domain, &req);
do {
m = vm_page_alloc_noobj_contig_domain(domain, req, npages, low,
high, alignment, boundary, memattr);
if (m != NULL)
break;
} while (vm_domainset_iter_page(&di, NULL, &domain) == 0);
return (m);
}
vm_page_t
vm_page_alloc_noobj_contig_domain(int domain, int req, u_long npages,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr)
{
vm_page_t m, m_ret;
u_int flags;
#define VPANC_FLAGS (VPAN_FLAGS | VM_ALLOC_NORECLAIM)
KASSERT((req & ~VPANC_FLAGS) == 0,
("invalid request %#x", req));
KASSERT((req & (VM_ALLOC_WAITOK | VM_ALLOC_NORECLAIM)) !=
(VM_ALLOC_WAITOK | VM_ALLOC_NORECLAIM),
("invalid request %#x", req));
KASSERT(((req & (VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)) !=
(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY)),
("invalid request %#x", req));
KASSERT(npages > 0, ("vm_page_alloc_contig: npages is zero"));
while ((m_ret = vm_page_find_contig_domain(domain, req, npages,
low, high, alignment, boundary)) == NULL) {
if (!vm_domain_alloc_fail(VM_DOMAIN(domain), NULL, req))
return (NULL);
}
/*
* Initialize the pages. Only the PG_ZERO flag is inherited.
*/
flags = PG_ZERO;
if ((req & VM_ALLOC_NODUMP) != 0)
flags |= PG_NODUMP;
if ((req & VM_ALLOC_WIRED) != 0)
vm_wire_add(npages);
for (m = m_ret; m < &m_ret[npages]; m++) {
vm_page_dequeue(m);
vm_page_alloc_check(m);
/*
* Consumers should not rely on a useful default pindex value.
*/
m->pindex = 0xdeadc0dedeadc0de;
m->a.flags = 0;
m->flags = (m->flags | PG_NODUMP) & flags;
m->busy_lock = VPB_UNBUSIED;
if ((req & VM_ALLOC_WIRED) != 0)
m->ref_count = 1;
m->a.act_count = 0;
m->oflags = VPO_UNMANAGED;
/*
* Zero the page before updating any mappings since the page is
* not yet shared with any devices which might require the
* non-default memory attribute. pmap_page_set_memattr()
* flushes data caches before returning.
*/
if ((req & VM_ALLOC_ZERO) != 0 && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
if (memattr != VM_MEMATTR_DEFAULT)
pmap_page_set_memattr(m, memattr);
}
return (m_ret);
}
/*
* Check a page that has been freshly dequeued from a freelist.
*/
static void
vm_page_alloc_check(vm_page_t m)
{
KASSERT(m->object == NULL, ("page %p has object", m));
KASSERT(m->a.queue == PQ_NONE &&
(m->a.flags & PGA_QUEUE_STATE_MASK) == 0,
("page %p has unexpected queue %d, flags %#x",
m, m->a.queue, (m->a.flags & PGA_QUEUE_STATE_MASK)));
KASSERT(m->ref_count == 0, ("page %p has references", m));
KASSERT(vm_page_busy_freed(m), ("page %p is not freed", m));
KASSERT(m->dirty == 0, ("page %p is dirty", m));
KASSERT(pmap_page_get_memattr(m) == VM_MEMATTR_DEFAULT,
("page %p has unexpected memattr %d",
m, pmap_page_get_memattr(m)));
KASSERT(vm_page_none_valid(m), ("free page %p is valid", m));
pmap_vm_page_alloc_check(m);
}
static int
vm_page_zone_import(void *arg, void **store, int cnt, int domain, int flags)
{
struct vm_domain *vmd;
struct vm_pgcache *pgcache;
int i;
pgcache = arg;
vmd = VM_DOMAIN(pgcache->domain);
/*
* The page daemon should avoid creating extra memory pressure since its
* main purpose is to replenish the store of free pages.
*/
if (vmd->vmd_severeset || curproc == pageproc ||
!_vm_domain_allocate(vmd, VM_ALLOC_NORMAL, cnt))
return (0);
domain = vmd->vmd_domain;
vm_domain_free_lock(vmd);
i = vm_phys_alloc_npages(domain, pgcache->pool, cnt,
(vm_page_t *)store);
vm_domain_free_unlock(vmd);
if (cnt != i)
vm_domain_freecnt_inc(vmd, cnt - i);
return (i);
}
static void
vm_page_zone_release(void *arg, void **store, int cnt)
{
struct vm_domain *vmd;
struct vm_pgcache *pgcache;
vm_page_t m;
int i;
pgcache = arg;
vmd = VM_DOMAIN(pgcache->domain);
vm_domain_free_lock(vmd);
for (i = 0; i < cnt; i++) {
m = (vm_page_t)store[i];
vm_phys_free_pages(m, 0);
}
vm_domain_free_unlock(vmd);
vm_domain_freecnt_inc(vmd, cnt);
}
#define VPSC_ANY 0 /* No restrictions. */
#define VPSC_NORESERV 1 /* Skip reservations; implies VPSC_NOSUPER. */
#define VPSC_NOSUPER 2 /* Skip superpages. */
/*
* vm_page_scan_contig:
*
* Scan vm_page_array[] between the specified entries "m_start" and
* "m_end" for a run of contiguous physical pages that satisfy the
* specified conditions, and return the lowest page in the run. The
* specified "alignment" determines the alignment of the lowest physical
* page in the run. If the specified "boundary" is non-zero, then the
* run of physical pages cannot span a physical address that is a
* multiple of "boundary".
*
* "m_end" is never dereferenced, so it need not point to a vm_page
* structure within vm_page_array[].
*
* "npages" must be greater than zero. "m_start" and "m_end" must not
* span a hole (or discontiguity) in the physical address space. Both
* "alignment" and "boundary" must be a power of two.
*/
static vm_page_t
vm_page_scan_contig(u_long npages, vm_page_t m_start, vm_page_t m_end,
u_long alignment, vm_paddr_t boundary, int options)
{
vm_object_t object;
vm_paddr_t pa;
vm_page_t m, m_run;
#if VM_NRESERVLEVEL > 0
int level;
#endif
int m_inc, order, run_ext, run_len;
KASSERT(npages > 0, ("npages is 0"));
KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
m_run = NULL;
run_len = 0;
for (m = m_start; m < m_end && run_len < npages; m += m_inc) {
KASSERT((m->flags & PG_MARKER) == 0,
("page %p is PG_MARKER", m));
KASSERT((m->flags & PG_FICTITIOUS) == 0 || m->ref_count >= 1,
("fictitious page %p has invalid ref count", m));
/*
* If the current page would be the start of a run, check its
* physical address against the end, alignment, and boundary
* conditions. If it doesn't satisfy these conditions, either
* terminate the scan or advance to the next page that
* satisfies the failed condition.
*/
if (run_len == 0) {
KASSERT(m_run == NULL, ("m_run != NULL"));
if (m + npages > m_end)
break;
pa = VM_PAGE_TO_PHYS(m);
if (!vm_addr_align_ok(pa, alignment)) {
m_inc = atop(roundup2(pa, alignment) - pa);
continue;
}
if (!vm_addr_bound_ok(pa, ptoa(npages), boundary)) {
m_inc = atop(roundup2(pa, boundary) - pa);
continue;
}
} else
KASSERT(m_run != NULL, ("m_run == NULL"));
retry:
m_inc = 1;
if (vm_page_wired(m))
run_ext = 0;
#if VM_NRESERVLEVEL > 0
else if ((level = vm_reserv_level(m)) >= 0 &&
(options & VPSC_NORESERV) != 0) {
run_ext = 0;
/* Advance to the end of the reservation. */
pa = VM_PAGE_TO_PHYS(m);
m_inc = atop(roundup2(pa + 1, vm_reserv_size(level)) -
pa);
}
#endif
else if ((object = atomic_load_ptr(&m->object)) != NULL) {
/*
* The page is considered eligible for relocation if
* and only if it could be laundered or reclaimed by
* the page daemon.
*/
VM_OBJECT_RLOCK(object);
if (object != m->object) {
VM_OBJECT_RUNLOCK(object);
goto retry;
}
/* Don't care: PG_NODUMP, PG_ZERO. */
if ((object->flags & OBJ_SWAP) == 0 &&
object->type != OBJT_VNODE) {
run_ext = 0;
#if VM_NRESERVLEVEL > 0
} else if ((options & VPSC_NOSUPER) != 0 &&
(level = vm_reserv_level_iffullpop(m)) >= 0) {
run_ext = 0;
/* Advance to the end of the superpage. */
pa = VM_PAGE_TO_PHYS(m);
m_inc = atop(roundup2(pa + 1,
vm_reserv_size(level)) - pa);
#endif
} else if (object->memattr == VM_MEMATTR_DEFAULT &&
vm_page_queue(m) != PQ_NONE && !vm_page_busied(m)) {
/*
* The page is allocated but eligible for
* relocation. Extend the current run by one
* page.
*/
KASSERT(pmap_page_get_memattr(m) ==
VM_MEMATTR_DEFAULT,
("page %p has an unexpected memattr", m));
KASSERT((m->oflags & (VPO_SWAPINPROG |
VPO_SWAPSLEEP | VPO_UNMANAGED)) == 0,
("page %p has unexpected oflags", m));
/* Don't care: PGA_NOSYNC. */
run_ext = 1;
} else
run_ext = 0;
VM_OBJECT_RUNLOCK(object);
#if VM_NRESERVLEVEL > 0
} else if (level >= 0) {
/*
* The page is reserved but not yet allocated. In
* other words, it is still free. Extend the current
* run by one page.
*/
run_ext = 1;
#endif
} else if ((order = m->order) < VM_NFREEORDER) {
/*
* The page is enqueued in the physical memory
* allocator's free page queues. Moreover, it is the
* first page in a power-of-two-sized run of
* contiguous free pages. Add these pages to the end
* of the current run, and jump ahead.
*/
run_ext = 1 << order;
m_inc = 1 << order;
} else {
/*
* Skip the page for one of the following reasons: (1)
* It is enqueued in the physical memory allocator's
* free page queues. However, it is not the first
* page in a run of contiguous free pages. (This case
* rarely occurs because the scan is performed in
* ascending order.) (2) It is not reserved, and it is
* transitioning from free to allocated. (Conversely,
* the transition from allocated to free for managed
* pages is blocked by the page busy lock.) (3) It is
* allocated but not contained by an object and not
* wired, e.g., allocated by Xen's balloon driver.
*/
run_ext = 0;
}
/*
* Extend or reset the current run of pages.
*/
if (run_ext > 0) {
if (run_len == 0)
m_run = m;
run_len += run_ext;
} else {
if (run_len > 0) {
m_run = NULL;
run_len = 0;
}
}
}
if (run_len >= npages)
return (m_run);
return (NULL);
}
/*
* vm_page_reclaim_run:
*
* Try to relocate each of the allocated virtual pages within the
* specified run of physical pages to a new physical address. Free the
* physical pages underlying the relocated virtual pages. A virtual page
* is relocatable if and only if it could be laundered or reclaimed by
* the page daemon. Whenever possible, a virtual page is relocated to a
* physical address above "high".
*
* Returns 0 if every physical page within the run was already free or
* just freed by a successful relocation. Otherwise, returns a non-zero
* value indicating why the last attempt to relocate a virtual page was
* unsuccessful.
*
* "req_class" must be an allocation class.
*/
static int
vm_page_reclaim_run(int req_class, int domain, u_long npages, vm_page_t m_run,
vm_paddr_t high)
{
struct vm_domain *vmd;
struct spglist free;
vm_object_t object;
vm_paddr_t pa;
vm_page_t m, m_end, m_new;
int error, order, req;
KASSERT((req_class & VM_ALLOC_CLASS_MASK) == req_class,
("req_class is not an allocation class"));
SLIST_INIT(&free);
error = 0;
m = m_run;
m_end = m_run + npages;
for (; error == 0 && m < m_end; m++) {
KASSERT((m->flags & (PG_FICTITIOUS | PG_MARKER)) == 0,
("page %p is PG_FICTITIOUS or PG_MARKER", m));
/*
* Racily check for wirings. Races are handled once the object
* lock is held and the page is unmapped.
*/
if (vm_page_wired(m))
error = EBUSY;
else if ((object = atomic_load_ptr(&m->object)) != NULL) {
/*
* The page is relocated if and only if it could be
* laundered or reclaimed by the page daemon.
*/
VM_OBJECT_WLOCK(object);
/* Don't care: PG_NODUMP, PG_ZERO. */
if (m->object != object ||
((object->flags & OBJ_SWAP) == 0 &&
object->type != OBJT_VNODE))
error = EINVAL;
else if (object->memattr != VM_MEMATTR_DEFAULT)
error = EINVAL;
else if (vm_page_queue(m) != PQ_NONE &&
vm_page_tryxbusy(m) != 0) {
if (vm_page_wired(m)) {
vm_page_xunbusy(m);
error = EBUSY;
goto unlock;
}
KASSERT(pmap_page_get_memattr(m) ==
VM_MEMATTR_DEFAULT,
("page %p has an unexpected memattr", m));
KASSERT(m->oflags == 0,
("page %p has unexpected oflags", m));
/* Don't care: PGA_NOSYNC. */
if (!vm_page_none_valid(m)) {
/*
* First, try to allocate a new page
* that is above "high". Failing
* that, try to allocate a new page
* that is below "m_run". Allocate
* the new page between the end of
* "m_run" and "high" only as a last
* resort.
*/
req = req_class;
if ((m->flags & PG_NODUMP) != 0)
req |= VM_ALLOC_NODUMP;
if (trunc_page(high) !=
~(vm_paddr_t)PAGE_MASK) {
m_new =
vm_page_alloc_noobj_contig(
req, 1, round_page(high),
~(vm_paddr_t)0, PAGE_SIZE,
0, VM_MEMATTR_DEFAULT);
} else
m_new = NULL;
if (m_new == NULL) {
pa = VM_PAGE_TO_PHYS(m_run);
m_new =
vm_page_alloc_noobj_contig(
req, 1, 0, pa - 1,
PAGE_SIZE, 0,
VM_MEMATTR_DEFAULT);
}
if (m_new == NULL) {
pa += ptoa(npages);
m_new =
vm_page_alloc_noobj_contig(
req, 1, pa, high, PAGE_SIZE,
0, VM_MEMATTR_DEFAULT);
}
if (m_new == NULL) {
vm_page_xunbusy(m);
error = ENOMEM;
goto unlock;
}
/*
* Unmap the page and check for new
* wirings that may have been acquired
* through a pmap lookup.
*/
if (object->ref_count != 0 &&
!vm_page_try_remove_all(m)) {
vm_page_xunbusy(m);
vm_page_free(m_new);
error = EBUSY;
goto unlock;
}
/*
* Replace "m" with the new page. For
* vm_page_replace(), "m" must be busy
* and dequeued. Finally, change "m"
* as if vm_page_free() was called.
*/
m_new->a.flags = m->a.flags &
~PGA_QUEUE_STATE_MASK;
KASSERT(m_new->oflags == VPO_UNMANAGED,
("page %p is managed", m_new));
m_new->oflags = 0;
pmap_copy_page(m, m_new);
m_new->valid = m->valid;
m_new->dirty = m->dirty;
m->flags &= ~PG_ZERO;
vm_page_dequeue(m);
if (vm_page_replace_hold(m_new, object,
m->pindex, m) &&
- vm_page_free_prep(m))
+ vm_page_free_prep(m, true))
SLIST_INSERT_HEAD(&free, m,
plinks.s.ss);
/*
* The new page must be deactivated
* before the object is unlocked.
*/
vm_page_deactivate(m_new);
} else {
m->flags &= ~PG_ZERO;
vm_page_dequeue(m);
- if (vm_page_free_prep(m))
+ if (vm_page_free_prep(m, true))
SLIST_INSERT_HEAD(&free, m,
plinks.s.ss);
KASSERT(m->dirty == 0,
("page %p is dirty", m));
}
} else
error = EBUSY;
unlock:
VM_OBJECT_WUNLOCK(object);
} else {
MPASS(vm_page_domain(m) == domain);
vmd = VM_DOMAIN(domain);
vm_domain_free_lock(vmd);
order = m->order;
if (order < VM_NFREEORDER) {
/*
* The page is enqueued in the physical memory
* allocator's free page queues. Moreover, it
* is the first page in a power-of-two-sized
* run of contiguous free pages. Jump ahead
* to the last page within that run, and
* continue from there.
*/
m += (1 << order) - 1;
}
#if VM_NRESERVLEVEL > 0
else if (vm_reserv_is_page_free(m))
order = 0;
#endif
vm_domain_free_unlock(vmd);
if (order == VM_NFREEORDER)
error = EINVAL;
}
}
if ((m = SLIST_FIRST(&free)) != NULL) {
int cnt;
vmd = VM_DOMAIN(domain);
cnt = 0;
vm_domain_free_lock(vmd);
do {
MPASS(vm_page_domain(m) == domain);
SLIST_REMOVE_HEAD(&free, plinks.s.ss);
vm_phys_free_pages(m, 0);
cnt++;
} while ((m = SLIST_FIRST(&free)) != NULL);
vm_domain_free_unlock(vmd);
vm_domain_freecnt_inc(vmd, cnt);
}
return (error);
}
#define NRUNS 16
#define RUN_INDEX(count, nruns) ((count) % (nruns))
#define MIN_RECLAIM 8
/*
* vm_page_reclaim_contig:
*
* Reclaim allocated, contiguous physical memory satisfying the specified
* conditions by relocating the virtual pages using that physical memory.
* Returns 0 if reclamation is successful, ERANGE if the specified domain
* can't possibly satisfy the reclamation request, or ENOMEM if not
* currently able to reclaim the requested number of pages. Since
* relocation requires the allocation of physical pages, reclamation may
* fail with ENOMEM due to a shortage of free pages. When reclamation
* fails in this manner, callers are expected to perform vm_wait() before
* retrying a failed allocation operation, e.g., vm_page_alloc_contig().
*
* The caller must always specify an allocation class through "req".
*
* allocation classes:
* VM_ALLOC_NORMAL normal process request
* VM_ALLOC_SYSTEM system *really* needs a page
* VM_ALLOC_INTERRUPT interrupt time request
*
* The optional allocation flags are ignored.
*
* "npages" must be greater than zero. Both "alignment" and "boundary"
* must be a power of two.
*/
int
vm_page_reclaim_contig_domain_ext(int domain, int req, u_long npages,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
int desired_runs)
{
struct vm_domain *vmd;
vm_page_t bounds[2], m_run, _m_runs[NRUNS], *m_runs;
u_long count, minalign, reclaimed;
int error, i, min_reclaim, nruns, options, req_class;
int segind, start_segind;
int ret;
KASSERT(npages > 0, ("npages is 0"));
KASSERT(powerof2(alignment), ("alignment is not a power of 2"));
KASSERT(powerof2(boundary), ("boundary is not a power of 2"));
ret = ENOMEM;
/*
* If the caller wants to reclaim multiple runs, try to allocate
* space to store the runs. If that fails, fall back to the old
* behavior of just reclaiming MIN_RECLAIM pages.
*/
if (desired_runs > 1)
m_runs = malloc((NRUNS + desired_runs) * sizeof(*m_runs),
M_TEMP, M_NOWAIT);
else
m_runs = NULL;
if (m_runs == NULL) {
m_runs = _m_runs;
nruns = NRUNS;
} else {
nruns = NRUNS + desired_runs - 1;
}
min_reclaim = MAX(desired_runs * npages, MIN_RECLAIM);
/*
* The caller will attempt an allocation after some runs have been
* reclaimed and added to the vm_phys buddy lists. Due to limitations
* of vm_phys_alloc_contig(), round up the requested length to the next
* power of two or maximum chunk size, and ensure that each run is
* suitably aligned.
*/
minalign = 1ul << imin(flsl(npages - 1), VM_NFREEORDER - 1);
npages = roundup2(npages, minalign);
if (alignment < ptoa(minalign))
alignment = ptoa(minalign);
/*
* The page daemon is allowed to dig deeper into the free page list.
*/
req_class = req & VM_ALLOC_CLASS_MASK;
if (curproc == pageproc && req_class != VM_ALLOC_INTERRUPT)
req_class = VM_ALLOC_SYSTEM;
start_segind = vm_phys_lookup_segind(low);
/*
* Return if the number of free pages cannot satisfy the requested
* allocation.
*/
vmd = VM_DOMAIN(domain);
count = vmd->vmd_free_count;
if (count < npages + vmd->vmd_free_reserved || (count < npages +
vmd->vmd_interrupt_free_min && req_class == VM_ALLOC_SYSTEM) ||
(count < npages && req_class == VM_ALLOC_INTERRUPT))
goto done;
/*
* Scan up to three times, relaxing the restrictions ("options") on
* the reclamation of reservations and superpages each time.
*/
for (options = VPSC_NORESERV;;) {
bool phys_range_exists = false;
/*
* Find the highest runs that satisfy the given constraints
* and restrictions, and record them in "m_runs".
*/
count = 0;
segind = start_segind;
while ((segind = vm_phys_find_range(bounds, segind, domain,
npages, low, high)) != -1) {
phys_range_exists = true;
while ((m_run = vm_page_scan_contig(npages, bounds[0],
bounds[1], alignment, boundary, options))) {
bounds[0] = m_run + npages;
m_runs[RUN_INDEX(count, nruns)] = m_run;
count++;
}
segind++;
}
if (!phys_range_exists) {
ret = ERANGE;
goto done;
}
/*
* Reclaim the highest runs in LIFO (descending) order until
* the number of reclaimed pages, "reclaimed", is at least
* "min_reclaim". Reset "reclaimed" each time because each
* reclamation is idempotent, and runs will (likely) recur
* from one scan to the next as restrictions are relaxed.
*/
reclaimed = 0;
for (i = 0; count > 0 && i < nruns; i++) {
count--;
m_run = m_runs[RUN_INDEX(count, nruns)];
error = vm_page_reclaim_run(req_class, domain, npages,
m_run, high);
if (error == 0) {
reclaimed += npages;
if (reclaimed >= min_reclaim) {
ret = 0;
goto done;
}
}
}
/*
* Either relax the restrictions on the next scan or return if
* the last scan had no restrictions.
*/
if (options == VPSC_NORESERV)
options = VPSC_NOSUPER;
else if (options == VPSC_NOSUPER)
options = VPSC_ANY;
else if (options == VPSC_ANY) {
if (reclaimed != 0)
ret = 0;
goto done;
}
}
done:
if (m_runs != _m_runs)
free(m_runs, M_TEMP);
return (ret);
}
int
vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary)
{
return (vm_page_reclaim_contig_domain_ext(domain, req, npages, low, high,
alignment, boundary, 1));
}
int
vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low, vm_paddr_t high,
u_long alignment, vm_paddr_t boundary)
{
struct vm_domainset_iter di;
int domain, ret, status;
ret = ERANGE;
vm_domainset_iter_page_init(&di, NULL, 0, &domain, &req);
do {
status = vm_page_reclaim_contig_domain(domain, req, npages, low,
high, alignment, boundary);
if (status == 0)
return (0);
else if (status == ERANGE)
vm_domainset_iter_ignore(&di, domain);
else {
KASSERT(status == ENOMEM, ("Unrecognized error %d "
"from vm_page_reclaim_contig_domain()", status));
ret = ENOMEM;
}
} while (vm_domainset_iter_page(&di, NULL, &domain) == 0);
return (ret);
}
/*
* Set the domain in the appropriate page level domainset.
*/
void
vm_domain_set(struct vm_domain *vmd)
{
mtx_lock(&vm_domainset_lock);
if (!vmd->vmd_minset && vm_paging_min(vmd)) {
vmd->vmd_minset = 1;
DOMAINSET_SET(vmd->vmd_domain, &vm_min_domains);
}
if (!vmd->vmd_severeset && vm_paging_severe(vmd)) {
vmd->vmd_severeset = 1;
DOMAINSET_SET(vmd->vmd_domain, &vm_severe_domains);
}
mtx_unlock(&vm_domainset_lock);
}
/*
* Clear the domain from the appropriate page level domainset.
*/
void
vm_domain_clear(struct vm_domain *vmd)
{
mtx_lock(&vm_domainset_lock);
if (vmd->vmd_minset && !vm_paging_min(vmd)) {
vmd->vmd_minset = 0;
DOMAINSET_CLR(vmd->vmd_domain, &vm_min_domains);
if (vm_min_waiters != 0) {
vm_min_waiters = 0;
wakeup(&vm_min_domains);
}
}
if (vmd->vmd_severeset && !vm_paging_severe(vmd)) {
vmd->vmd_severeset = 0;
DOMAINSET_CLR(vmd->vmd_domain, &vm_severe_domains);
if (vm_severe_waiters != 0) {
vm_severe_waiters = 0;
wakeup(&vm_severe_domains);
}
}
/*
* If pageout daemon needs pages, then tell it that there are
* some free.
*/
if (vmd->vmd_pageout_pages_needed &&
vmd->vmd_free_count >= vmd->vmd_pageout_free_min) {
wakeup(&vmd->vmd_pageout_pages_needed);
vmd->vmd_pageout_pages_needed = 0;
}
/* See comments in vm_wait_doms(). */
if (vm_pageproc_waiters) {
vm_pageproc_waiters = 0;
wakeup(&vm_pageproc_waiters);
}
mtx_unlock(&vm_domainset_lock);
}
/*
* Wait for free pages to exceed the min threshold globally.
*/
void
vm_wait_min(void)
{
mtx_lock(&vm_domainset_lock);
while (vm_page_count_min()) {
vm_min_waiters++;
msleep(&vm_min_domains, &vm_domainset_lock, PVM, "vmwait", 0);
}
mtx_unlock(&vm_domainset_lock);
}
/*
* Wait for free pages to exceed the severe threshold globally.
*/
void
vm_wait_severe(void)
{
mtx_lock(&vm_domainset_lock);
while (vm_page_count_severe()) {
vm_severe_waiters++;
msleep(&vm_severe_domains, &vm_domainset_lock, PVM,
"vmwait", 0);
}
mtx_unlock(&vm_domainset_lock);
}
u_int
vm_wait_count(void)
{
return (vm_severe_waiters + vm_min_waiters + vm_pageproc_waiters);
}
int
vm_wait_doms(const domainset_t *wdoms, int mflags)
{
int error;
error = 0;
/*
* We use racey wakeup synchronization to avoid expensive global
* locking for the pageproc when sleeping with a non-specific vm_wait.
* To handle this, we only sleep for one tick in this instance. It
* is expected that most allocations for the pageproc will come from
* kmem or vm_page_grab* which will use the more specific and
* race-free vm_wait_domain().
*/
if (curproc == pageproc) {
mtx_lock(&vm_domainset_lock);
vm_pageproc_waiters++;
error = msleep(&vm_pageproc_waiters, &vm_domainset_lock,
PVM | PDROP | mflags, "pageprocwait", 1);
} else {
/*
* XXX Ideally we would wait only until the allocation could
* be satisfied. This condition can cause new allocators to
* consume all freed pages while old allocators wait.
*/
mtx_lock(&vm_domainset_lock);
if (vm_page_count_min_set(wdoms)) {
if (pageproc == NULL)
panic("vm_wait in early boot");
vm_min_waiters++;
error = msleep(&vm_min_domains, &vm_domainset_lock,
PVM | PDROP | mflags, "vmwait", 0);
} else
mtx_unlock(&vm_domainset_lock);
}
return (error);
}
/*
* vm_wait_domain:
*
* Sleep until free pages are available for allocation.
* - Called in various places after failed memory allocations.
*/
void
vm_wait_domain(int domain)
{
struct vm_domain *vmd;
domainset_t wdom;
vmd = VM_DOMAIN(domain);
vm_domain_free_assert_unlocked(vmd);
if (curproc == pageproc) {
mtx_lock(&vm_domainset_lock);
if (vmd->vmd_free_count < vmd->vmd_pageout_free_min) {
vmd->vmd_pageout_pages_needed = 1;
msleep(&vmd->vmd_pageout_pages_needed,
&vm_domainset_lock, PDROP | PSWP, "VMWait", 0);
} else
mtx_unlock(&vm_domainset_lock);
} else {
DOMAINSET_ZERO(&wdom);
DOMAINSET_SET(vmd->vmd_domain, &wdom);
vm_wait_doms(&wdom, 0);
}
}
static int
vm_wait_flags(vm_object_t obj, int mflags)
{
struct domainset *d;
d = NULL;
/*
* Carefully fetch pointers only once: the struct domainset
* itself is ummutable but the pointer might change.
*/
if (obj != NULL)
d = obj->domain.dr_policy;
if (d == NULL)
d = curthread->td_domain.dr_policy;
return (vm_wait_doms(&d->ds_mask, mflags));
}
/*
* vm_wait:
*
* Sleep until free pages are available for allocation in the
* affinity domains of the obj. If obj is NULL, the domain set
* for the calling thread is used.
* Called in various places after failed memory allocations.
*/
void
vm_wait(vm_object_t obj)
{
(void)vm_wait_flags(obj, 0);
}
int
vm_wait_intr(vm_object_t obj)
{
return (vm_wait_flags(obj, PCATCH));
}
/*
* vm_domain_alloc_fail:
*
* Called when a page allocation function fails. Informs the
* pagedaemon and performs the requested wait. Requires the
* domain_free and object lock on entry. Returns with the
* object lock held and free lock released. Returns an error when
* retry is necessary.
*
*/
static int
vm_domain_alloc_fail(struct vm_domain *vmd, vm_object_t object, int req)
{
vm_domain_free_assert_unlocked(vmd);
atomic_add_int(&vmd->vmd_pageout_deficit,
max((u_int)req >> VM_ALLOC_COUNT_SHIFT, 1));
if (req & (VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL)) {
if (object != NULL)
VM_OBJECT_WUNLOCK(object);
vm_wait_domain(vmd->vmd_domain);
if (object != NULL)
VM_OBJECT_WLOCK(object);
if (req & VM_ALLOC_WAITOK)
return (EAGAIN);
}
return (0);
}
/*
* vm_waitpfault:
*
* Sleep until free pages are available for allocation.
* - Called only in vm_fault so that processes page faulting
* can be easily tracked.
* - Sleeps at a lower priority than vm_wait() so that vm_wait()ing
* processes will be able to grab memory first. Do not change
* this balance without careful testing first.
*/
void
vm_waitpfault(struct domainset *dset, int timo)
{
/*
* XXX Ideally we would wait only until the allocation could
* be satisfied. This condition can cause new allocators to
* consume all freed pages while old allocators wait.
*/
mtx_lock(&vm_domainset_lock);
if (vm_page_count_min_set(&dset->ds_mask)) {
vm_min_waiters++;
msleep(&vm_min_domains, &vm_domainset_lock, PUSER | PDROP,
"pfault", timo);
} else
mtx_unlock(&vm_domainset_lock);
}
static struct vm_pagequeue *
_vm_page_pagequeue(vm_page_t m, uint8_t queue)
{
return (&vm_pagequeue_domain(m)->vmd_pagequeues[queue]);
}
#ifdef INVARIANTS
static struct vm_pagequeue *
vm_page_pagequeue(vm_page_t m)
{
return (_vm_page_pagequeue(m, vm_page_astate_load(m).queue));
}
#endif
static __always_inline bool
vm_page_pqstate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new)
{
vm_page_astate_t tmp;
tmp = *old;
do {
if (__predict_true(vm_page_astate_fcmpset(m, old, new)))
return (true);
counter_u64_add(pqstate_commit_retries, 1);
} while (old->_bits == tmp._bits);
return (false);
}
/*
* Do the work of committing a queue state update that moves the page out of
* its current queue.
*/
static bool
_vm_page_pqstate_commit_dequeue(struct vm_pagequeue *pq, vm_page_t m,
vm_page_astate_t *old, vm_page_astate_t new)
{
vm_page_t next;
vm_pagequeue_assert_locked(pq);
KASSERT(vm_page_pagequeue(m) == pq,
("%s: queue %p does not match page %p", __func__, pq, m));
KASSERT(old->queue != PQ_NONE && new.queue != old->queue,
("%s: invalid queue indices %d %d",
__func__, old->queue, new.queue));
/*
* Once the queue index of the page changes there is nothing
* synchronizing with further updates to the page's physical
* queue state. Therefore we must speculatively remove the page
* from the queue now and be prepared to roll back if the queue
* state update fails. If the page is not physically enqueued then
* we just update its queue index.
*/
if ((old->flags & PGA_ENQUEUED) != 0) {
new.flags &= ~PGA_ENQUEUED;
next = TAILQ_NEXT(m, plinks.q);
TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
vm_pagequeue_cnt_dec(pq);
if (!vm_page_pqstate_fcmpset(m, old, new)) {
if (next == NULL)
TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
else
TAILQ_INSERT_BEFORE(next, m, plinks.q);
vm_pagequeue_cnt_inc(pq);
return (false);
} else {
return (true);
}
} else {
return (vm_page_pqstate_fcmpset(m, old, new));
}
}
static bool
vm_page_pqstate_commit_dequeue(vm_page_t m, vm_page_astate_t *old,
vm_page_astate_t new)
{
struct vm_pagequeue *pq;
vm_page_astate_t as;
bool ret;
pq = _vm_page_pagequeue(m, old->queue);
/*
* The queue field and PGA_ENQUEUED flag are stable only so long as the
* corresponding page queue lock is held.
*/
vm_pagequeue_lock(pq);
as = vm_page_astate_load(m);
if (__predict_false(as._bits != old->_bits)) {
*old = as;
ret = false;
} else {
ret = _vm_page_pqstate_commit_dequeue(pq, m, old, new);
}
vm_pagequeue_unlock(pq);
return (ret);
}
/*
* Commit a queue state update that enqueues or requeues a page.
*/
static bool
_vm_page_pqstate_commit_requeue(struct vm_pagequeue *pq, vm_page_t m,
vm_page_astate_t *old, vm_page_astate_t new)
{
struct vm_domain *vmd;
vm_pagequeue_assert_locked(pq);
KASSERT(old->queue != PQ_NONE && new.queue == old->queue,
("%s: invalid queue indices %d %d",
__func__, old->queue, new.queue));
new.flags |= PGA_ENQUEUED;
if (!vm_page_pqstate_fcmpset(m, old, new))
return (false);
if ((old->flags & PGA_ENQUEUED) != 0)
TAILQ_REMOVE(&pq->pq_pl, m, plinks.q);
else
vm_pagequeue_cnt_inc(pq);
/*
* Give PGA_REQUEUE_HEAD precedence over PGA_REQUEUE. In particular, if
* both flags are set in close succession, only PGA_REQUEUE_HEAD will be
* applied, even if it was set first.
*/
if ((old->flags & PGA_REQUEUE_HEAD) != 0) {
vmd = vm_pagequeue_domain(m);
KASSERT(pq == &vmd->vmd_pagequeues[PQ_INACTIVE],
("%s: invalid page queue for page %p", __func__, m));
TAILQ_INSERT_BEFORE(&vmd->vmd_inacthead, m, plinks.q);
} else {
TAILQ_INSERT_TAIL(&pq->pq_pl, m, plinks.q);
}
return (true);
}
/*
* Commit a queue state update that encodes a request for a deferred queue
* operation.
*/
static bool
vm_page_pqstate_commit_request(vm_page_t m, vm_page_astate_t *old,
vm_page_astate_t new)
{
KASSERT(old->queue == new.queue || new.queue != PQ_NONE,
("%s: invalid state, queue %d flags %x",
__func__, new.queue, new.flags));
if (old->_bits != new._bits &&
!vm_page_pqstate_fcmpset(m, old, new))
return (false);
vm_page_pqbatch_submit(m, new.queue);
return (true);
}
/*
* A generic queue state update function. This handles more cases than the
* specialized functions above.
*/
bool
vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new)
{
if (old->_bits == new._bits)
return (true);
if (old->queue != PQ_NONE && new.queue != old->queue) {
if (!vm_page_pqstate_commit_dequeue(m, old, new))
return (false);
if (new.queue != PQ_NONE)
vm_page_pqbatch_submit(m, new.queue);
} else {
if (!vm_page_pqstate_fcmpset(m, old, new))
return (false);
if (new.queue != PQ_NONE &&
((new.flags & ~old->flags) & PGA_QUEUE_OP_MASK) != 0)
vm_page_pqbatch_submit(m, new.queue);
}
return (true);
}
/*
* Apply deferred queue state updates to a page.
*/
static inline void
vm_pqbatch_process_page(struct vm_pagequeue *pq, vm_page_t m, uint8_t queue)
{
vm_page_astate_t new, old;
CRITICAL_ASSERT(curthread);
vm_pagequeue_assert_locked(pq);
KASSERT(queue < PQ_COUNT,
("%s: invalid queue index %d", __func__, queue));
KASSERT(pq == _vm_page_pagequeue(m, queue),
("%s: page %p does not belong to queue %p", __func__, m, pq));
for (old = vm_page_astate_load(m);;) {
if (__predict_false(old.queue != queue ||
(old.flags & PGA_QUEUE_OP_MASK) == 0)) {
counter_u64_add(queue_nops, 1);
break;
}
KASSERT((m->oflags & VPO_UNMANAGED) == 0,
("%s: page %p is unmanaged", __func__, m));
new = old;
if ((old.flags & PGA_DEQUEUE) != 0) {
new.flags &= ~PGA_QUEUE_OP_MASK;
new.queue = PQ_NONE;
if (__predict_true(_vm_page_pqstate_commit_dequeue(pq,
m, &old, new))) {
counter_u64_add(queue_ops, 1);
break;
}
} else {
new.flags &= ~(PGA_REQUEUE | PGA_REQUEUE_HEAD);
if (__predict_true(_vm_page_pqstate_commit_requeue(pq,
m, &old, new))) {
counter_u64_add(queue_ops, 1);
break;
}
}
}
}
static void
vm_pqbatch_process(struct vm_pagequeue *pq, struct vm_batchqueue *bq,
uint8_t queue)
{
int i;
for (i = 0; i < bq->bq_cnt; i++)
vm_pqbatch_process_page(pq, bq->bq_pa[i], queue);
vm_batchqueue_init(bq);
}
/*
* vm_page_pqbatch_submit: [ internal use only ]
*
* Enqueue a page in the specified page queue's batched work queue.
* The caller must have encoded the requested operation in the page
* structure's a.flags field.
*/
void
vm_page_pqbatch_submit(vm_page_t m, uint8_t queue)
{
struct vm_batchqueue *bq;
struct vm_pagequeue *pq;
int domain, slots_remaining;
KASSERT(queue < PQ_COUNT, ("invalid queue %d", queue));
domain = vm_page_domain(m);
critical_enter();
bq = DPCPU_PTR(pqbatch[domain][queue]);
slots_remaining = vm_batchqueue_insert(bq, m);
if (slots_remaining > (VM_BATCHQUEUE_SIZE >> 1)) {
/* keep building the bq */
critical_exit();
return;
} else if (slots_remaining > 0 ) {
/* Try to process the bq if we can get the lock */
pq = &VM_DOMAIN(domain)->vmd_pagequeues[queue];
if (vm_pagequeue_trylock(pq)) {
vm_pqbatch_process(pq, bq, queue);
vm_pagequeue_unlock(pq);
}
critical_exit();
return;
}
critical_exit();
/* if we make it here, the bq is full so wait for the lock */
pq = &VM_DOMAIN(domain)->vmd_pagequeues[queue];
vm_pagequeue_lock(pq);
critical_enter();
bq = DPCPU_PTR(pqbatch[domain][queue]);
vm_pqbatch_process(pq, bq, queue);
vm_pqbatch_process_page(pq, m, queue);
vm_pagequeue_unlock(pq);
critical_exit();
}
/*
* vm_page_pqbatch_drain: [ internal use only ]
*
* Force all per-CPU page queue batch queues to be drained. This is
* intended for use in severe memory shortages, to ensure that pages
* do not remain stuck in the batch queues.
*/
void
vm_page_pqbatch_drain(void)
{
struct thread *td;
struct vm_domain *vmd;
struct vm_pagequeue *pq;
int cpu, domain, queue;
td = curthread;
CPU_FOREACH(cpu) {
thread_lock(td);
sched_bind(td, cpu);
thread_unlock(td);
for (domain = 0; domain < vm_ndomains; domain++) {
vmd = VM_DOMAIN(domain);
for (queue = 0; queue < PQ_COUNT; queue++) {
pq = &vmd->vmd_pagequeues[queue];
vm_pagequeue_lock(pq);
critical_enter();
vm_pqbatch_process(pq,
DPCPU_PTR(pqbatch[domain][queue]), queue);
critical_exit();
vm_pagequeue_unlock(pq);
}
}
}
thread_lock(td);
sched_unbind(td);
thread_unlock(td);
}
/*
* vm_page_dequeue_deferred: [ internal use only ]
*
* Request removal of the given page from its current page
* queue. Physical removal from the queue may be deferred
* indefinitely.
*/
void
vm_page_dequeue_deferred(vm_page_t m)
{
vm_page_astate_t new, old;
old = vm_page_astate_load(m);
do {
if (old.queue == PQ_NONE) {
KASSERT((old.flags & PGA_QUEUE_STATE_MASK) == 0,
("%s: page %p has unexpected queue state",
__func__, m));
break;
}
new = old;
new.flags |= PGA_DEQUEUE;
} while (!vm_page_pqstate_commit_request(m, &old, new));
}
/*
* vm_page_dequeue:
*
* Remove the page from whichever page queue it's in, if any, before
* returning.
*/
void
vm_page_dequeue(vm_page_t m)
{
vm_page_astate_t new, old;
old = vm_page_astate_load(m);
do {
if (old.queue == PQ_NONE) {
KASSERT((old.flags & PGA_QUEUE_STATE_MASK) == 0,
("%s: page %p has unexpected queue state",
__func__, m));
break;
}
new = old;
new.flags &= ~PGA_QUEUE_OP_MASK;
new.queue = PQ_NONE;
} while (!vm_page_pqstate_commit_dequeue(m, &old, new));
}
/*
* Schedule the given page for insertion into the specified page queue.
* Physical insertion of the page may be deferred indefinitely.
*/
static void
vm_page_enqueue(vm_page_t m, uint8_t queue)
{
KASSERT(m->a.queue == PQ_NONE &&
(m->a.flags & PGA_QUEUE_STATE_MASK) == 0,
("%s: page %p is already enqueued", __func__, m));
KASSERT(m->ref_count > 0,
("%s: page %p does not carry any references", __func__, m));
m->a.queue = queue;
if ((m->a.flags & PGA_REQUEUE) == 0)
vm_page_aflag_set(m, PGA_REQUEUE);
vm_page_pqbatch_submit(m, queue);
}
/*
* vm_page_free_prep:
*
* Prepares the given page to be put on the free list,
* disassociating it from any VM object. The caller may return
* the page to the free list only if this function returns true.
*
* The object, if it exists, must be locked, and then the page must
* be xbusy. Otherwise the page must be not busied. A managed
* page must be unmapped.
*/
static bool
-vm_page_free_prep(vm_page_t m)
+vm_page_free_prep(vm_page_t m, bool do_remove)
{
/*
* Synchronize with threads that have dropped a reference to this
* page.
*/
atomic_thread_fence_acq();
#if defined(DIAGNOSTIC) && defined(PHYS_TO_DMAP)
if (PMAP_HAS_DMAP && (m->flags & PG_ZERO) != 0) {
uint64_t *p;
int i;
p = (uint64_t *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(m));
for (i = 0; i < PAGE_SIZE / sizeof(uint64_t); i++, p++)
KASSERT(*p == 0, ("vm_page_free_prep %p PG_ZERO %d %jx",
m, i, (uintmax_t)*p));
}
#endif
KASSERT((m->flags & PG_NOFREE) == 0,
("%s: attempting to free a PG_NOFREE page", __func__));
if ((m->oflags & VPO_UNMANAGED) == 0) {
KASSERT(!pmap_page_is_mapped(m),
("vm_page_free_prep: freeing mapped page %p", m));
KASSERT((m->a.flags & (PGA_EXECUTABLE | PGA_WRITEABLE)) == 0,
("vm_page_free_prep: mapping flags set in page %p", m));
} else {
KASSERT(m->a.queue == PQ_NONE,
("vm_page_free_prep: unmanaged page %p is queued", m));
}
VM_CNT_INC(v_tfree);
if (m->object != NULL) {
KASSERT(((m->oflags & VPO_UNMANAGED) != 0) ==
((m->object->flags & OBJ_UNMANAGED) != 0),
("vm_page_free_prep: managed flag mismatch for page %p",
m));
vm_page_assert_xbusied(m);
/*
* The object reference can be released without an atomic
* operation.
*/
KASSERT((m->flags & PG_FICTITIOUS) != 0 ||
m->ref_count == VPRC_OBJREF,
("vm_page_free_prep: page %p has unexpected ref_count %u",
m, m->ref_count));
- vm_page_object_remove(m);
+ if (do_remove)
+ vm_page_radix_remove(m);
+ vm_page_remove_radixdone(m);
m->ref_count -= VPRC_OBJREF;
} else
vm_page_assert_unbusied(m);
vm_page_busy_free(m);
/*
* If fictitious remove object association and
* return.
*/
if ((m->flags & PG_FICTITIOUS) != 0) {
KASSERT(m->ref_count == 1,
("fictitious page %p is referenced", m));
KASSERT(m->a.queue == PQ_NONE,
("fictitious page %p is queued", m));
return (false);
}
/*
* Pages need not be dequeued before they are returned to the physical
* memory allocator, but they must at least be marked for a deferred
* dequeue.
*/
if ((m->oflags & VPO_UNMANAGED) == 0)
vm_page_dequeue_deferred(m);
m->valid = 0;
vm_page_undirty(m);
if (m->ref_count != 0)
panic("vm_page_free_prep: page %p has references", m);
/*
* Restore the default memory attribute to the page.
*/
if (pmap_page_get_memattr(m) != VM_MEMATTR_DEFAULT)
pmap_page_set_memattr(m, VM_MEMATTR_DEFAULT);
#if VM_NRESERVLEVEL > 0
/*
* Determine whether the page belongs to a reservation. If the page was
* allocated from a per-CPU cache, it cannot belong to a reservation, so
* as an optimization, we avoid the check in that case.
*/
if ((m->flags & PG_PCPU_CACHE) == 0 && vm_reserv_free_page(m))
return (false);
#endif
return (true);
}
-/*
- * vm_page_free_toq:
- *
- * Returns the given page to the free list, disassociating it
- * from any VM object.
- *
- * The object must be locked. The page must be exclusively busied if it
- * belongs to an object.
- */
static void
-vm_page_free_toq(vm_page_t m)
+vm_page_free_toq_impl(vm_page_t m, bool do_remove)
{
struct vm_domain *vmd;
uma_zone_t zone;
- if (!vm_page_free_prep(m))
+ if (!vm_page_free_prep(m, do_remove))
return;
vmd = vm_pagequeue_domain(m);
zone = vmd->vmd_pgcache[m->pool].zone;
if ((m->flags & PG_PCPU_CACHE) != 0 && zone != NULL) {
uma_zfree(zone, m);
return;
}
vm_domain_free_lock(vmd);
vm_phys_free_pages(m, 0);
vm_domain_free_unlock(vmd);
vm_domain_freecnt_inc(vmd, 1);
}
+/*
+ * vm_page_free_toq:
+ *
+ * Returns the given page to the free list, disassociating it
+ * from any VM object.
+ *
+ * The object must be locked. The page must be exclusively busied if it
+ * belongs to an object.
+ */
+static void
+vm_page_free_toq(vm_page_t m)
+{
+ vm_page_free_toq_impl(m, true);
+}
+
/*
* vm_page_free_pages_toq:
*
* Returns a list of pages to the free list, disassociating it
* from any VM object. In other words, this is equivalent to
* calling vm_page_free_toq() for each page of a list of VM objects.
*/
int
vm_page_free_pages_toq(struct spglist *free, bool update_wire_count)
{
vm_page_t m;
int count;
if (SLIST_EMPTY(free))
return (0);
count = 0;
while ((m = SLIST_FIRST(free)) != NULL) {
count++;
SLIST_REMOVE_HEAD(free, plinks.s.ss);
vm_page_free_toq(m);
}
if (update_wire_count)
vm_wire_sub(count);
return (count);
}
/*
* Mark this page as wired down. For managed pages, this prevents reclamation
* by the page daemon, or when the containing object, if any, is destroyed.
*/
void
vm_page_wire(vm_page_t m)
{
u_int old;
#ifdef INVARIANTS
if (m->object != NULL && !vm_page_busied(m) &&
!vm_object_busied(m->object))
VM_OBJECT_ASSERT_LOCKED(m->object);
#endif
KASSERT((m->flags & PG_FICTITIOUS) == 0 ||
VPRC_WIRE_COUNT(m->ref_count) >= 1,
("vm_page_wire: fictitious page %p has zero wirings", m));
old = atomic_fetchadd_int(&m->ref_count, 1);
KASSERT(VPRC_WIRE_COUNT(old) != VPRC_WIRE_COUNT_MAX,
("vm_page_wire: counter overflow for page %p", m));
if (VPRC_WIRE_COUNT(old) == 0) {
if ((m->oflags & VPO_UNMANAGED) == 0)
vm_page_aflag_set(m, PGA_DEQUEUE);
vm_wire_add(1);
}
}
/*
* Attempt to wire a mapped page following a pmap lookup of that page.
* This may fail if a thread is concurrently tearing down mappings of the page.
* The transient failure is acceptable because it translates to the
* failure of the caller pmap_extract_and_hold(), which should be then
* followed by the vm_fault() fallback, see e.g. vm_fault_quick_hold_pages().
*/
bool
vm_page_wire_mapped(vm_page_t m)
{
u_int old;
old = atomic_load_int(&m->ref_count);
do {
KASSERT(old > 0,
("vm_page_wire_mapped: wiring unreferenced page %p", m));
if ((old & VPRC_BLOCKED) != 0)
return (false);
} while (!atomic_fcmpset_int(&m->ref_count, &old, old + 1));
if (VPRC_WIRE_COUNT(old) == 0) {
if ((m->oflags & VPO_UNMANAGED) == 0)
vm_page_aflag_set(m, PGA_DEQUEUE);
vm_wire_add(1);
}
return (true);
}
/*
* Release a wiring reference to a managed page. If the page still belongs to
* an object, update its position in the page queues to reflect the reference.
* If the wiring was the last reference to the page, free the page.
*/
static void
vm_page_unwire_managed(vm_page_t m, uint8_t nqueue, bool noreuse)
{
u_int old;
KASSERT((m->oflags & VPO_UNMANAGED) == 0,
("%s: page %p is unmanaged", __func__, m));
/*
* Update LRU state before releasing the wiring reference.
* Use a release store when updating the reference count to
* synchronize with vm_page_free_prep().
*/
old = atomic_load_int(&m->ref_count);
do {
u_int count;
KASSERT(VPRC_WIRE_COUNT(old) > 0,
("vm_page_unwire: wire count underflow for page %p", m));
count = old & ~VPRC_BLOCKED;
if (count > VPRC_OBJREF + 1) {
/*
* The page has at least one other wiring reference. An
* earlier iteration of this loop may have called
* vm_page_release_toq() and cleared PGA_DEQUEUE, so
* re-set it if necessary.
*/
if ((vm_page_astate_load(m).flags & PGA_DEQUEUE) == 0)
vm_page_aflag_set(m, PGA_DEQUEUE);
} else if (count == VPRC_OBJREF + 1) {
/*
* This is the last wiring. Clear PGA_DEQUEUE and
* update the page's queue state to reflect the
* reference. If the page does not belong to an object
* (i.e., the VPRC_OBJREF bit is clear), we only need to
* clear leftover queue state.
*/
vm_page_release_toq(m, nqueue, noreuse);
} else if (count == 1) {
vm_page_aflag_clear(m, PGA_DEQUEUE);
}
} while (!atomic_fcmpset_rel_int(&m->ref_count, &old, old - 1));
if (VPRC_WIRE_COUNT(old) == 1) {
vm_wire_sub(1);
if (old == 1)
vm_page_free(m);
}
}
/*
* Release one wiring of the specified page, potentially allowing it to be
* paged out.
*
* Only managed pages belonging to an object can be paged out. If the number
* of wirings transitions to zero and the page is eligible for page out, then
* the page is added to the specified paging queue. If the released wiring
* represented the last reference to the page, the page is freed.
*/
void
vm_page_unwire(vm_page_t m, uint8_t nqueue)
{
KASSERT(nqueue < PQ_COUNT,
("vm_page_unwire: invalid queue %u request for page %p",
nqueue, m));
if ((m->oflags & VPO_UNMANAGED) != 0) {
if (vm_page_unwire_noq(m) && m->ref_count == 0)
vm_page_free(m);
return;
}
vm_page_unwire_managed(m, nqueue, false);
}
/*
* Unwire a page without (re-)inserting it into a page queue. It is up
* to the caller to enqueue, requeue, or free the page as appropriate.
* In most cases involving managed pages, vm_page_unwire() should be used
* instead.
*/
bool
vm_page_unwire_noq(vm_page_t m)
{
u_int old;
old = vm_page_drop(m, 1);
KASSERT(VPRC_WIRE_COUNT(old) != 0,
("%s: counter underflow for page %p", __func__, m));
KASSERT((m->flags & PG_FICTITIOUS) == 0 || VPRC_WIRE_COUNT(old) > 1,
("%s: missing ref on fictitious page %p", __func__, m));
if (VPRC_WIRE_COUNT(old) > 1)
return (false);
if ((m->oflags & VPO_UNMANAGED) == 0)
vm_page_aflag_clear(m, PGA_DEQUEUE);
vm_wire_sub(1);
return (true);
}
/*
* Ensure that the page ends up in the specified page queue. If the page is
* active or being moved to the active queue, ensure that its act_count is
* at least ACT_INIT but do not otherwise mess with it.
*/
static __always_inline void
vm_page_mvqueue(vm_page_t m, const uint8_t nqueue, const uint16_t nflag)
{
vm_page_astate_t old, new;
KASSERT(m->ref_count > 0,
("%s: page %p does not carry any references", __func__, m));
KASSERT(nflag == PGA_REQUEUE || nflag == PGA_REQUEUE_HEAD,
("%s: invalid flags %x", __func__, nflag));
if ((m->oflags & VPO_UNMANAGED) != 0 || vm_page_wired(m))
return;
old = vm_page_astate_load(m);
do {
if ((old.flags & PGA_DEQUEUE) != 0)
break;
new = old;
new.flags &= ~PGA_QUEUE_OP_MASK;
if (nqueue == PQ_ACTIVE)
new.act_count = max(old.act_count, ACT_INIT);
if (old.queue == nqueue) {
/*
* There is no need to requeue pages already in the
* active queue.
*/
if (nqueue != PQ_ACTIVE ||
(old.flags & PGA_ENQUEUED) == 0)
new.flags |= nflag;
} else {
new.flags |= nflag;
new.queue = nqueue;
}
} while (!vm_page_pqstate_commit(m, &old, new));
}
/*
* Put the specified page on the active list (if appropriate).
*/
void
vm_page_activate(vm_page_t m)
{
vm_page_mvqueue(m, PQ_ACTIVE, PGA_REQUEUE);
}
/*
* Move the specified page to the tail of the inactive queue, or requeue
* the page if it is already in the inactive queue.
*/
void
vm_page_deactivate(vm_page_t m)
{
vm_page_mvqueue(m, PQ_INACTIVE, PGA_REQUEUE);
}
void
vm_page_deactivate_noreuse(vm_page_t m)
{
vm_page_mvqueue(m, PQ_INACTIVE, PGA_REQUEUE_HEAD);
}
/*
* Put a page in the laundry, or requeue it if it is already there.
*/
void
vm_page_launder(vm_page_t m)
{
vm_page_mvqueue(m, PQ_LAUNDRY, PGA_REQUEUE);
}
/*
* Put a page in the PQ_UNSWAPPABLE holding queue.
*/
void
vm_page_unswappable(vm_page_t m)
{
VM_OBJECT_ASSERT_LOCKED(m->object);
KASSERT((m->oflags & VPO_UNMANAGED) == 0,
("page %p already unswappable", m));
vm_page_dequeue(m);
vm_page_enqueue(m, PQ_UNSWAPPABLE);
}
/*
* Release a page back to the page queues in preparation for unwiring.
*/
static void
vm_page_release_toq(vm_page_t m, uint8_t nqueue, const bool noreuse)
{
vm_page_astate_t old, new;
uint16_t nflag;
/*
* Use a check of the valid bits to determine whether we should
* accelerate reclamation of the page. The object lock might not be
* held here, in which case the check is racy. At worst we will either
* accelerate reclamation of a valid page and violate LRU, or
* unnecessarily defer reclamation of an invalid page.
*
* If we were asked to not cache the page, place it near the head of the
* inactive queue so that is reclaimed sooner.
*/
if (noreuse || vm_page_none_valid(m)) {
nqueue = PQ_INACTIVE;
nflag = PGA_REQUEUE_HEAD;
} else {
nflag = PGA_REQUEUE;
}
old = vm_page_astate_load(m);
do {
new = old;
/*
* If the page is already in the active queue and we are not
* trying to accelerate reclamation, simply mark it as
* referenced and avoid any queue operations.
*/
new.flags &= ~PGA_QUEUE_OP_MASK;
if (nflag != PGA_REQUEUE_HEAD && old.queue == PQ_ACTIVE &&
(old.flags & PGA_ENQUEUED) != 0)
new.flags |= PGA_REFERENCED;
else {
new.flags |= nflag;
new.queue = nqueue;
}
} while (!vm_page_pqstate_commit(m, &old, new));
}
/*
* Unwire a page and either attempt to free it or re-add it to the page queues.
*/
void
vm_page_release(vm_page_t m, int flags)
{
vm_object_t object;
KASSERT((m->oflags & VPO_UNMANAGED) == 0,
("vm_page_release: page %p is unmanaged", m));
if ((flags & VPR_TRYFREE) != 0) {
for (;;) {
object = atomic_load_ptr(&m->object);
if (object == NULL)
break;
/* Depends on type-stability. */
if (vm_page_busied(m) || !VM_OBJECT_TRYWLOCK(object))
break;
if (object == m->object) {
vm_page_release_locked(m, flags);
VM_OBJECT_WUNLOCK(object);
return;
}
VM_OBJECT_WUNLOCK(object);
}
}
vm_page_unwire_managed(m, PQ_INACTIVE, flags != 0);
}
/* See vm_page_release(). */
void
vm_page_release_locked(vm_page_t m, int flags)
{
VM_OBJECT_ASSERT_WLOCKED(m->object);
KASSERT((m->oflags & VPO_UNMANAGED) == 0,
("vm_page_release_locked: page %p is unmanaged", m));
if (vm_page_unwire_noq(m)) {
if ((flags & VPR_TRYFREE) != 0 &&
(m->object->ref_count == 0 || !pmap_page_is_mapped(m)) &&
m->dirty == 0 && vm_page_tryxbusy(m)) {
/*
* An unlocked lookup may have wired the page before the
* busy lock was acquired, in which case the page must
* not be freed.
*/
if (__predict_true(!vm_page_wired(m))) {
vm_page_free(m);
return;
}
vm_page_xunbusy(m);
} else {
vm_page_release_toq(m, PQ_INACTIVE, flags != 0);
}
}
}
static bool
vm_page_try_blocked_op(vm_page_t m, void (*op)(vm_page_t))
{
u_int old;
KASSERT(m->object != NULL && (m->oflags & VPO_UNMANAGED) == 0,
("vm_page_try_blocked_op: page %p has no object", m));
KASSERT(vm_page_busied(m),
("vm_page_try_blocked_op: page %p is not busy", m));
VM_OBJECT_ASSERT_LOCKED(m->object);
old = atomic_load_int(&m->ref_count);
do {
KASSERT(old != 0,
("vm_page_try_blocked_op: page %p has no references", m));
KASSERT((old & VPRC_BLOCKED) == 0,
("vm_page_try_blocked_op: page %p blocks wirings", m));
if (VPRC_WIRE_COUNT(old) != 0)
return (false);
} while (!atomic_fcmpset_int(&m->ref_count, &old, old | VPRC_BLOCKED));
(op)(m);
/*
* If the object is read-locked, new wirings may be created via an
* object lookup.
*/
old = vm_page_drop(m, VPRC_BLOCKED);
KASSERT(!VM_OBJECT_WOWNED(m->object) ||
old == (VPRC_BLOCKED | VPRC_OBJREF),
("vm_page_try_blocked_op: unexpected refcount value %u for %p",
old, m));
return (true);
}
/*
* Atomically check for wirings and remove all mappings of the page.
*/
bool
vm_page_try_remove_all(vm_page_t m)
{
return (vm_page_try_blocked_op(m, pmap_remove_all));
}
/*
* Atomically check for wirings and remove all writeable mappings of the page.
*/
bool
vm_page_try_remove_write(vm_page_t m)
{
return (vm_page_try_blocked_op(m, pmap_remove_write));
}
/*
* vm_page_advise
*
* Apply the specified advice to the given page.
*/
void
vm_page_advise(vm_page_t m, int advice)
{
VM_OBJECT_ASSERT_WLOCKED(m->object);
vm_page_assert_xbusied(m);
if (advice == MADV_FREE)
/*
* Mark the page clean. This will allow the page to be freed
* without first paging it out. MADV_FREE pages are often
* quickly reused by malloc(3), so we do not do anything that
* would result in a page fault on a later access.
*/
vm_page_undirty(m);
else if (advice != MADV_DONTNEED) {
if (advice == MADV_WILLNEED)
vm_page_activate(m);
return;
}
if (advice != MADV_FREE && m->dirty == 0 && pmap_is_modified(m))
vm_page_dirty(m);
/*
* Clear any references to the page. Otherwise, the page daemon will
* immediately reactivate the page.
*/
vm_page_aflag_clear(m, PGA_REFERENCED);
/*
* Place clean pages near the head of the inactive queue rather than
* the tail, thus defeating the queue's LRU operation and ensuring that
* the page will be reused quickly. Dirty pages not already in the
* laundry are moved there.
*/
if (m->dirty == 0)
vm_page_deactivate_noreuse(m);
else if (!vm_page_in_laundry(m))
vm_page_launder(m);
}
/*
* vm_page_grab_release
*
* Helper routine for grab functions to release busy on return.
*/
static inline void
vm_page_grab_release(vm_page_t m, int allocflags)
{
if ((allocflags & VM_ALLOC_NOBUSY) != 0) {
if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0)
vm_page_sunbusy(m);
else
vm_page_xunbusy(m);
}
}
/*
* vm_page_grab_sleep
*
* Sleep for busy according to VM_ALLOC_ parameters. Returns true
* if the caller should retry and false otherwise.
*
* If the object is locked on entry the object will be unlocked with
* false returns and still locked but possibly having been dropped
* with true returns.
*/
static bool
vm_page_grab_sleep(vm_object_t object, vm_page_t m, vm_pindex_t pindex,
const char *wmesg, int allocflags, bool locked)
{
if ((allocflags & VM_ALLOC_NOWAIT) != 0)
return (false);
/*
* Reference the page before unlocking and sleeping so that
* the page daemon is less likely to reclaim it.
*/
if (locked && (allocflags & VM_ALLOC_NOCREAT) == 0)
vm_page_reference(m);
if (_vm_page_busy_sleep(object, m, pindex, wmesg, allocflags, locked) &&
locked)
VM_OBJECT_WLOCK(object);
if ((allocflags & VM_ALLOC_WAITFAIL) != 0)
return (false);
return (true);
}
/*
* Assert that the grab flags are valid.
*/
static inline void
vm_page_grab_check(int allocflags)
{
KASSERT((allocflags & VM_ALLOC_NOBUSY) == 0 ||
(allocflags & VM_ALLOC_WIRED) != 0,
("vm_page_grab*: the pages must be busied or wired"));
KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
(allocflags & VM_ALLOC_IGN_SBUSY) != 0,
("vm_page_grab*: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
}
/*
* Calculate the page allocation flags for grab.
*/
static inline int
vm_page_grab_pflags(int allocflags)
{
int pflags;
pflags = allocflags &
~(VM_ALLOC_NOWAIT | VM_ALLOC_WAITOK | VM_ALLOC_WAITFAIL |
VM_ALLOC_NOBUSY | VM_ALLOC_IGN_SBUSY);
if ((allocflags & VM_ALLOC_NOWAIT) == 0)
pflags |= VM_ALLOC_WAITFAIL;
if ((allocflags & VM_ALLOC_IGN_SBUSY) != 0)
pflags |= VM_ALLOC_SBUSY;
return (pflags);
}
/*
* Grab a page, waiting until we are waken up due to the page
* changing state. We keep on waiting, if the page continues
* to be in the object. If the page doesn't exist, first allocate it
* and then conditionally zero it.
*
* This routine may sleep.
*
* The object must be locked on entry. The lock will, however, be released
* and reacquired if the routine sleeps.
*/
vm_page_t
vm_page_grab(vm_object_t object, vm_pindex_t pindex, int allocflags)
{
vm_page_t m;
VM_OBJECT_ASSERT_WLOCKED(object);
vm_page_grab_check(allocflags);
retrylookup:
if ((m = vm_page_lookup(object, pindex)) != NULL) {
if (!vm_page_tryacquire(m, allocflags)) {
if (vm_page_grab_sleep(object, m, pindex, "pgrbwt",
allocflags, true))
goto retrylookup;
return (NULL);
}
goto out;
}
if ((allocflags & VM_ALLOC_NOCREAT) != 0)
return (NULL);
m = vm_page_alloc(object, pindex, vm_page_grab_pflags(allocflags));
if (m == NULL) {
if ((allocflags & (VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL)) != 0)
return (NULL);
goto retrylookup;
}
if (allocflags & VM_ALLOC_ZERO && (m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
out:
vm_page_grab_release(m, allocflags);
return (m);
}
/*
* Attempt to validate a page, locklessly acquiring it if necessary, given a
* (object, pindex) tuple and either an invalided page or NULL. The resulting
* page will be validated against the identity tuple, and busied or wired as
* requested. A NULL page returned guarantees that the page was not in radix at
* the time of the call but callers must perform higher level synchronization or
* retry the operation under a lock if they require an atomic answer. This is
* the only lock free validation routine, other routines can depend on the
* resulting page state.
*
* The return value PAGE_NOT_ACQUIRED indicates that the operation failed due to
* caller flags.
*/
#define PAGE_NOT_ACQUIRED ((vm_page_t)1)
static vm_page_t
vm_page_acquire_unlocked(vm_object_t object, vm_pindex_t pindex, vm_page_t m,
int allocflags)
{
if (m == NULL)
m = vm_page_lookup_unlocked(object, pindex);
for (; m != NULL; m = vm_page_lookup_unlocked(object, pindex)) {
if (vm_page_trybusy(m, allocflags)) {
if (m->object == object && m->pindex == pindex) {
if ((allocflags & VM_ALLOC_WIRED) != 0)
vm_page_wire(m);
vm_page_grab_release(m, allocflags);
break;
}
/* relookup. */
vm_page_busy_release(m);
cpu_spinwait();
continue;
}
if (!vm_page_grab_sleep(object, m, pindex, "pgnslp",
allocflags, false))
return (PAGE_NOT_ACQUIRED);
}
return (m);
}
/*
* Try to locklessly grab a page and fall back to the object lock if NOCREAT
* is not set.
*/
vm_page_t
vm_page_grab_unlocked(vm_object_t object, vm_pindex_t pindex, int allocflags)
{
vm_page_t m;
vm_page_grab_check(allocflags);
m = vm_page_acquire_unlocked(object, pindex, NULL, allocflags);
if (m == PAGE_NOT_ACQUIRED)
return (NULL);
if (m != NULL)
return (m);
/*
* The radix lockless lookup should never return a false negative
* errors. If the user specifies NOCREAT they are guaranteed there
* was no page present at the instant of the call. A NOCREAT caller
* must handle create races gracefully.
*/
if ((allocflags & VM_ALLOC_NOCREAT) != 0)
return (NULL);
VM_OBJECT_WLOCK(object);
m = vm_page_grab(object, pindex, allocflags);
VM_OBJECT_WUNLOCK(object);
return (m);
}
/*
* Grab a page and make it valid, paging in if necessary. Pages missing from
* their pager are zero filled and validated. If a VM_ALLOC_COUNT is supplied
* and the page is not valid as many as VM_INITIAL_PAGEIN pages can be brought
* in simultaneously. Additional pages will be left on a paging queue but
* will neither be wired nor busy regardless of allocflags.
*/
int
vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex, int allocflags)
{
vm_page_t m;
vm_page_t ma[VM_INITIAL_PAGEIN];
int after, i, pflags, rv;
KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
(allocflags & VM_ALLOC_IGN_SBUSY) != 0,
("vm_page_grab_valid: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY mismatch"));
KASSERT((allocflags &
(VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL | VM_ALLOC_ZERO)) == 0,
("vm_page_grab_valid: Invalid flags 0x%X", allocflags));
VM_OBJECT_ASSERT_WLOCKED(object);
pflags = allocflags & ~(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY |
VM_ALLOC_WIRED | VM_ALLOC_IGN_SBUSY);
pflags |= VM_ALLOC_WAITFAIL;
retrylookup:
if ((m = vm_page_lookup(object, pindex)) != NULL) {
/*
* If the page is fully valid it can only become invalid
* with the object lock held. If it is not valid it can
* become valid with the busy lock held. Therefore, we
* may unnecessarily lock the exclusive busy here if we
* race with I/O completion not using the object lock.
* However, we will not end up with an invalid page and a
* shared lock.
*/
if (!vm_page_trybusy(m,
vm_page_all_valid(m) ? allocflags : 0)) {
(void)vm_page_grab_sleep(object, m, pindex, "pgrbwt",
allocflags, true);
goto retrylookup;
}
if (vm_page_all_valid(m))
goto out;
if ((allocflags & VM_ALLOC_NOCREAT) != 0) {
vm_page_busy_release(m);
*mp = NULL;
return (VM_PAGER_FAIL);
}
} else if ((allocflags & VM_ALLOC_NOCREAT) != 0) {
*mp = NULL;
return (VM_PAGER_FAIL);
} else if ((m = vm_page_alloc(object, pindex, pflags)) == NULL) {
if (!vm_pager_can_alloc_page(object, pindex)) {
*mp = NULL;
return (VM_PAGER_AGAIN);
}
goto retrylookup;
}
vm_page_assert_xbusied(m);
if (vm_pager_has_page(object, pindex, NULL, &after)) {
after = MIN(after, VM_INITIAL_PAGEIN);
after = MIN(after, allocflags >> VM_ALLOC_COUNT_SHIFT);
after = MAX(after, 1);
ma[0] = m;
for (i = 1; i < after; i++) {
if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) {
if (vm_page_any_valid(ma[i]) ||
!vm_page_tryxbusy(ma[i]))
break;
} else {
ma[i] = vm_page_alloc(object, m->pindex + i,
VM_ALLOC_NORMAL);
if (ma[i] == NULL)
break;
}
}
after = i;
vm_object_pip_add(object, after);
VM_OBJECT_WUNLOCK(object);
rv = vm_pager_get_pages(object, ma, after, NULL, NULL);
VM_OBJECT_WLOCK(object);
vm_object_pip_wakeupn(object, after);
/* Pager may have replaced a page. */
m = ma[0];
if (rv != VM_PAGER_OK) {
for (i = 0; i < after; i++) {
if (!vm_page_wired(ma[i]))
vm_page_free(ma[i]);
else
vm_page_xunbusy(ma[i]);
}
*mp = NULL;
return (rv);
}
for (i = 1; i < after; i++)
vm_page_readahead_finish(ma[i]);
MPASS(vm_page_all_valid(m));
} else {
vm_page_zero_invalid(m, TRUE);
}
out:
if ((allocflags & VM_ALLOC_WIRED) != 0)
vm_page_wire(m);
if ((allocflags & VM_ALLOC_SBUSY) != 0 && vm_page_xbusied(m))
vm_page_busy_downgrade(m);
else if ((allocflags & VM_ALLOC_NOBUSY) != 0)
vm_page_busy_release(m);
*mp = m;
return (VM_PAGER_OK);
}
/*
* Locklessly grab a valid page. If the page is not valid or not yet
* allocated this will fall back to the object lock method.
*/
int
vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object,
vm_pindex_t pindex, int allocflags)
{
vm_page_t m;
int flags;
int error;
KASSERT((allocflags & VM_ALLOC_SBUSY) == 0 ||
(allocflags & VM_ALLOC_IGN_SBUSY) != 0,
("vm_page_grab_valid_unlocked: VM_ALLOC_SBUSY/VM_ALLOC_IGN_SBUSY "
"mismatch"));
KASSERT((allocflags &
(VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL | VM_ALLOC_ZERO)) == 0,
("vm_page_grab_valid_unlocked: Invalid flags 0x%X", allocflags));
/*
* Attempt a lockless lookup and busy. We need at least an sbusy
* before we can inspect the valid field and return a wired page.
*/
flags = allocflags & ~(VM_ALLOC_NOBUSY | VM_ALLOC_WIRED);
vm_page_grab_check(flags);
m = vm_page_acquire_unlocked(object, pindex, NULL, flags);
if (m == PAGE_NOT_ACQUIRED)
return (VM_PAGER_FAIL);
if (m != NULL) {
if (vm_page_all_valid(m)) {
if ((allocflags & VM_ALLOC_WIRED) != 0)
vm_page_wire(m);
vm_page_grab_release(m, allocflags);
*mp = m;
return (VM_PAGER_OK);
}
vm_page_busy_release(m);
}
if ((allocflags & VM_ALLOC_NOCREAT) != 0) {
*mp = NULL;
return (VM_PAGER_FAIL);
}
VM_OBJECT_WLOCK(object);
error = vm_page_grab_valid(mp, object, pindex, allocflags);
VM_OBJECT_WUNLOCK(object);
return (error);
}
/*
* Return the specified range of pages from the given object. For each
* page offset within the range, if a page already exists within the object
* at that offset and it is busy, then wait for it to change state. If,
* instead, the page doesn't exist, then allocate it.
*
* The caller must always specify an allocation class.
*
* allocation classes:
* VM_ALLOC_NORMAL normal process request
* VM_ALLOC_SYSTEM system *really* needs the pages
*
* The caller must always specify that the pages are to be busied and/or
* wired.
*
* optional allocation flags:
* VM_ALLOC_IGN_SBUSY do not sleep on soft busy pages
* VM_ALLOC_NOBUSY do not exclusive busy the page
* VM_ALLOC_NOWAIT do not sleep
* VM_ALLOC_SBUSY set page to sbusy state
* VM_ALLOC_WIRED wire the pages
* VM_ALLOC_ZERO zero and validate any invalid pages
*
* If VM_ALLOC_NOWAIT is not specified, this routine may sleep. Otherwise, it
* may return a partial prefix of the requested range.
*/
int
vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
vm_page_t *ma, int count)
{
vm_page_t m, mpred;
int pflags;
int i;
VM_OBJECT_ASSERT_WLOCKED(object);
KASSERT(((u_int)allocflags >> VM_ALLOC_COUNT_SHIFT) == 0,
("vm_page_grap_pages: VM_ALLOC_COUNT() is not allowed"));
KASSERT(count > 0,
("vm_page_grab_pages: invalid page count %d", count));
vm_page_grab_check(allocflags);
pflags = vm_page_grab_pflags(allocflags);
i = 0;
retrylookup:
m = vm_page_mpred(object, pindex + i);
if (m == NULL || m->pindex != pindex + i) {
mpred = m;
m = NULL;
} else
mpred = TAILQ_PREV(m, pglist, listq);
for (; i < count; i++) {
if (m != NULL) {
if (!vm_page_tryacquire(m, allocflags)) {
if (vm_page_grab_sleep(object, m, pindex + i,
"grbmaw", allocflags, true))
goto retrylookup;
break;
}
} else {
if ((allocflags & VM_ALLOC_NOCREAT) != 0)
break;
m = vm_page_alloc_after(object, pindex + i,
pflags | VM_ALLOC_COUNT(count - i), mpred);
if (m == NULL) {
if ((allocflags & (VM_ALLOC_NOWAIT |
VM_ALLOC_WAITFAIL)) != 0)
break;
goto retrylookup;
}
}
if (vm_page_none_valid(m) &&
(allocflags & VM_ALLOC_ZERO) != 0) {
if ((m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
vm_page_valid(m);
}
vm_page_grab_release(m, allocflags);
ma[i] = mpred = m;
m = vm_page_next(m);
}
return (i);
}
/*
* Unlocked variant of vm_page_grab_pages(). This accepts the same flags
* and will fall back to the locked variant to handle allocation.
*/
int
vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex,
int allocflags, vm_page_t *ma, int count)
{
vm_page_t m;
int flags;
int i;
KASSERT(count > 0,
("vm_page_grab_pages_unlocked: invalid page count %d", count));
vm_page_grab_check(allocflags);
/*
* Modify flags for lockless acquire to hold the page until we
* set it valid if necessary.
*/
flags = allocflags & ~VM_ALLOC_NOBUSY;
vm_page_grab_check(flags);
m = NULL;
for (i = 0; i < count; i++, pindex++) {
/*
* We may see a false NULL here because the previous page has
* been removed or just inserted and the list is loaded without
* barriers. Switch to radix to verify.
*/
if (m == NULL || QMD_IS_TRASHED(m) || m->pindex != pindex ||
atomic_load_ptr(&m->object) != object) {
/*
* This guarantees the result is instantaneously
* correct.
*/
m = NULL;
}
m = vm_page_acquire_unlocked(object, pindex, m, flags);
if (m == PAGE_NOT_ACQUIRED)
return (i);
if (m == NULL)
break;
if ((flags & VM_ALLOC_ZERO) != 0 && vm_page_none_valid(m)) {
if ((m->flags & PG_ZERO) == 0)
pmap_zero_page(m);
vm_page_valid(m);
}
/* m will still be wired or busy according to flags. */
vm_page_grab_release(m, allocflags);
ma[i] = m;
m = TAILQ_NEXT(m, listq);
}
if (i == count || (allocflags & VM_ALLOC_NOCREAT) != 0)
return (i);
count -= i;
VM_OBJECT_WLOCK(object);
i += vm_page_grab_pages(object, pindex, allocflags, &ma[i], count);
VM_OBJECT_WUNLOCK(object);
return (i);
}
/*
* Mapping function for valid or dirty bits in a page.
*
* Inputs are required to range within a page.
*/
vm_page_bits_t
vm_page_bits(int base, int size)
{
int first_bit;
int last_bit;
KASSERT(
base + size <= PAGE_SIZE,
("vm_page_bits: illegal base/size %d/%d", base, size)
);
if (size == 0) /* handle degenerate case */
return (0);
first_bit = base >> DEV_BSHIFT;
last_bit = (base + size - 1) >> DEV_BSHIFT;
return (((vm_page_bits_t)2 << last_bit) -
((vm_page_bits_t)1 << first_bit));
}
void
vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set)
{
#if PAGE_SIZE == 32768
atomic_set_64((uint64_t *)bits, set);
#elif PAGE_SIZE == 16384
atomic_set_32((uint32_t *)bits, set);
#elif (PAGE_SIZE == 8192) && defined(atomic_set_16)
atomic_set_16((uint16_t *)bits, set);
#elif (PAGE_SIZE == 4096) && defined(atomic_set_8)
atomic_set_8((uint8_t *)bits, set);
#else /* PAGE_SIZE <= 8192 */
uintptr_t addr;
int shift;
addr = (uintptr_t)bits;
/*
* Use a trick to perform a 32-bit atomic on the
* containing aligned word, to not depend on the existence
* of atomic_{set, clear}_{8, 16}.
*/
shift = addr & (sizeof(uint32_t) - 1);
#if BYTE_ORDER == BIG_ENDIAN
shift = (sizeof(uint32_t) - sizeof(vm_page_bits_t) - shift) * NBBY;
#else
shift *= NBBY;
#endif
addr &= ~(sizeof(uint32_t) - 1);
atomic_set_32((uint32_t *)addr, set << shift);
#endif /* PAGE_SIZE */
}
static inline void
vm_page_bits_clear(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t clear)
{
#if PAGE_SIZE == 32768
atomic_clear_64((uint64_t *)bits, clear);
#elif PAGE_SIZE == 16384
atomic_clear_32((uint32_t *)bits, clear);
#elif (PAGE_SIZE == 8192) && defined(atomic_clear_16)
atomic_clear_16((uint16_t *)bits, clear);
#elif (PAGE_SIZE == 4096) && defined(atomic_clear_8)
atomic_clear_8((uint8_t *)bits, clear);
#else /* PAGE_SIZE <= 8192 */
uintptr_t addr;
int shift;
addr = (uintptr_t)bits;
/*
* Use a trick to perform a 32-bit atomic on the
* containing aligned word, to not depend on the existence
* of atomic_{set, clear}_{8, 16}.
*/
shift = addr & (sizeof(uint32_t) - 1);
#if BYTE_ORDER == BIG_ENDIAN
shift = (sizeof(uint32_t) - sizeof(vm_page_bits_t) - shift) * NBBY;
#else
shift *= NBBY;
#endif
addr &= ~(sizeof(uint32_t) - 1);
atomic_clear_32((uint32_t *)addr, clear << shift);
#endif /* PAGE_SIZE */
}
static inline vm_page_bits_t
vm_page_bits_swap(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t newbits)
{
#if PAGE_SIZE == 32768
uint64_t old;
old = *bits;
while (atomic_fcmpset_64(bits, &old, newbits) == 0);
return (old);
#elif PAGE_SIZE == 16384
uint32_t old;
old = *bits;
while (atomic_fcmpset_32(bits, &old, newbits) == 0);
return (old);
#elif (PAGE_SIZE == 8192) && defined(atomic_fcmpset_16)
uint16_t old;
old = *bits;
while (atomic_fcmpset_16(bits, &old, newbits) == 0);
return (old);
#elif (PAGE_SIZE == 4096) && defined(atomic_fcmpset_8)
uint8_t old;
old = *bits;
while (atomic_fcmpset_8(bits, &old, newbits) == 0);
return (old);
#else /* PAGE_SIZE <= 4096*/
uintptr_t addr;
uint32_t old, new, mask;
int shift;
addr = (uintptr_t)bits;
/*
* Use a trick to perform a 32-bit atomic on the
* containing aligned word, to not depend on the existence
* of atomic_{set, swap, clear}_{8, 16}.
*/
shift = addr & (sizeof(uint32_t) - 1);
#if BYTE_ORDER == BIG_ENDIAN
shift = (sizeof(uint32_t) - sizeof(vm_page_bits_t) - shift) * NBBY;
#else
shift *= NBBY;
#endif
addr &= ~(sizeof(uint32_t) - 1);
mask = VM_PAGE_BITS_ALL << shift;
old = *bits;
do {
new = old & ~mask;
new |= newbits << shift;
} while (atomic_fcmpset_32((uint32_t *)addr, &old, new) == 0);
return (old >> shift);
#endif /* PAGE_SIZE */
}
/*
* vm_page_set_valid_range:
*
* Sets portions of a page valid. The arguments are expected
* to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
* of any partial chunks touched by the range. The invalid portion of
* such chunks will be zeroed.
*
* (base + size) must be less then or equal to PAGE_SIZE.
*/
void
vm_page_set_valid_range(vm_page_t m, int base, int size)
{
int endoff, frag;
vm_page_bits_t pagebits;
vm_page_assert_busied(m);
if (size == 0) /* handle degenerate case */
return;
/*
* If the base is not DEV_BSIZE aligned and the valid
* bit is clear, we have to zero out a portion of the
* first block.
*/
if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
(m->valid & (1 << (base >> DEV_BSHIFT))) == 0)
pmap_zero_page_area(m, frag, base - frag);
/*
* If the ending offset is not DEV_BSIZE aligned and the
* valid bit is clear, we have to zero out a portion of
* the last block.
*/
endoff = base + size;
if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
(m->valid & (1 << (endoff >> DEV_BSHIFT))) == 0)
pmap_zero_page_area(m, endoff,
DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
/*
* Assert that no previously invalid block that is now being validated
* is already dirty.
*/
KASSERT((~m->valid & vm_page_bits(base, size) & m->dirty) == 0,
("vm_page_set_valid_range: page %p is dirty", m));
/*
* Set valid bits inclusive of any overlap.
*/
pagebits = vm_page_bits(base, size);
if (vm_page_xbusied(m))
m->valid |= pagebits;
else
vm_page_bits_set(m, &m->valid, pagebits);
}
/*
* Set the page dirty bits and free the invalid swap space if
* present. Returns the previous dirty bits.
*/
vm_page_bits_t
vm_page_set_dirty(vm_page_t m)
{
vm_page_bits_t old;
VM_PAGE_OBJECT_BUSY_ASSERT(m);
if (vm_page_xbusied(m) && !pmap_page_is_write_mapped(m)) {
old = m->dirty;
m->dirty = VM_PAGE_BITS_ALL;
} else
old = vm_page_bits_swap(m, &m->dirty, VM_PAGE_BITS_ALL);
if (old == 0 && (m->a.flags & PGA_SWAP_SPACE) != 0)
vm_pager_page_unswapped(m);
return (old);
}
/*
* Clear the given bits from the specified page's dirty field.
*/
static __inline void
vm_page_clear_dirty_mask(vm_page_t m, vm_page_bits_t pagebits)
{
vm_page_assert_busied(m);
/*
* If the page is xbusied and not write mapped we are the
* only thread that can modify dirty bits. Otherwise, The pmap
* layer can call vm_page_dirty() without holding a distinguished
* lock. The combination of page busy and atomic operations
* suffice to guarantee consistency of the page dirty field.
*/
if (vm_page_xbusied(m) && !pmap_page_is_write_mapped(m))
m->dirty &= ~pagebits;
else
vm_page_bits_clear(m, &m->dirty, pagebits);
}
/*
* vm_page_set_validclean:
*
* Sets portions of a page valid and clean. The arguments are expected
* to be DEV_BSIZE aligned but if they aren't the bitmap is inclusive
* of any partial chunks touched by the range. The invalid portion of
* such chunks will be zero'd.
*
* (base + size) must be less then or equal to PAGE_SIZE.
*/
void
vm_page_set_validclean(vm_page_t m, int base, int size)
{
vm_page_bits_t oldvalid, pagebits;
int endoff, frag;
vm_page_assert_busied(m);
if (size == 0) /* handle degenerate case */
return;
/*
* If the base is not DEV_BSIZE aligned and the valid
* bit is clear, we have to zero out a portion of the
* first block.
*/
if ((frag = rounddown2(base, DEV_BSIZE)) != base &&
(m->valid & ((vm_page_bits_t)1 << (base >> DEV_BSHIFT))) == 0)
pmap_zero_page_area(m, frag, base - frag);
/*
* If the ending offset is not DEV_BSIZE aligned and the
* valid bit is clear, we have to zero out a portion of
* the last block.
*/
endoff = base + size;
if ((frag = rounddown2(endoff, DEV_BSIZE)) != endoff &&
(m->valid & ((vm_page_bits_t)1 << (endoff >> DEV_BSHIFT))) == 0)
pmap_zero_page_area(m, endoff,
DEV_BSIZE - (endoff & (DEV_BSIZE - 1)));
/*
* Set valid, clear dirty bits. If validating the entire
* page we can safely clear the pmap modify bit. We also
* use this opportunity to clear the PGA_NOSYNC flag. If a process
* takes a write fault on a MAP_NOSYNC memory area the flag will
* be set again.
*
* We set valid bits inclusive of any overlap, but we can only
* clear dirty bits for DEV_BSIZE chunks that are fully within
* the range.
*/
oldvalid = m->valid;
pagebits = vm_page_bits(base, size);
if (vm_page_xbusied(m))
m->valid |= pagebits;
else
vm_page_bits_set(m, &m->valid, pagebits);
#if 0 /* NOT YET */
if ((frag = base & (DEV_BSIZE - 1)) != 0) {
frag = DEV_BSIZE - frag;
base += frag;
size -= frag;
if (size < 0)
size = 0;
}
pagebits = vm_page_bits(base, size & (DEV_BSIZE - 1));
#endif
if (base == 0 && size == PAGE_SIZE) {
/*
* The page can only be modified within the pmap if it is
* mapped, and it can only be mapped if it was previously
* fully valid.
*/
if (oldvalid == VM_PAGE_BITS_ALL)
/*
* Perform the pmap_clear_modify() first. Otherwise,
* a concurrent pmap operation, such as
* pmap_protect(), could clear a modification in the
* pmap and set the dirty field on the page before
* pmap_clear_modify() had begun and after the dirty
* field was cleared here.
*/
pmap_clear_modify(m);
m->dirty = 0;
vm_page_aflag_clear(m, PGA_NOSYNC);
} else if (oldvalid != VM_PAGE_BITS_ALL && vm_page_xbusied(m))
m->dirty &= ~pagebits;
else
vm_page_clear_dirty_mask(m, pagebits);
}
void
vm_page_clear_dirty(vm_page_t m, int base, int size)
{
vm_page_clear_dirty_mask(m, vm_page_bits(base, size));
}
/*
* vm_page_set_invalid:
*
* Invalidates DEV_BSIZE'd chunks within a page. Both the
* valid and dirty bits for the effected areas are cleared.
*/
void
vm_page_set_invalid(vm_page_t m, int base, int size)
{
vm_page_bits_t bits;
vm_object_t object;
/*
* The object lock is required so that pages can't be mapped
* read-only while we're in the process of invalidating them.
*/
object = m->object;
VM_OBJECT_ASSERT_WLOCKED(object);
vm_page_assert_busied(m);
if (object->type == OBJT_VNODE && base == 0 && IDX_TO_OFF(m->pindex) +
size >= object->un_pager.vnp.vnp_size)
bits = VM_PAGE_BITS_ALL;
else
bits = vm_page_bits(base, size);
if (object->ref_count != 0 && vm_page_all_valid(m) && bits != 0)
pmap_remove_all(m);
KASSERT((bits == 0 && vm_page_all_valid(m)) ||
!pmap_page_is_mapped(m),
("vm_page_set_invalid: page %p is mapped", m));
if (vm_page_xbusied(m)) {
m->valid &= ~bits;
m->dirty &= ~bits;
} else {
vm_page_bits_clear(m, &m->valid, bits);
vm_page_bits_clear(m, &m->dirty, bits);
}
}
/*
* vm_page_invalid:
*
* Invalidates the entire page. The page must be busy, unmapped, and
* the enclosing object must be locked. The object locks protects
* against concurrent read-only pmap enter which is done without
* busy.
*/
void
vm_page_invalid(vm_page_t m)
{
vm_page_assert_busied(m);
VM_OBJECT_ASSERT_WLOCKED(m->object);
MPASS(!pmap_page_is_mapped(m));
if (vm_page_xbusied(m))
m->valid = 0;
else
vm_page_bits_clear(m, &m->valid, VM_PAGE_BITS_ALL);
}
/*
* vm_page_zero_invalid()
*
* The kernel assumes that the invalid portions of a page contain
* garbage, but such pages can be mapped into memory by user code.
* When this occurs, we must zero out the non-valid portions of the
* page so user code sees what it expects.
*
* Pages are most often semi-valid when the end of a file is mapped
* into memory and the file's size is not page aligned.
*/
void
vm_page_zero_invalid(vm_page_t m, boolean_t setvalid)
{
int b;
int i;
/*
* Scan the valid bits looking for invalid sections that
* must be zeroed. Invalid sub-DEV_BSIZE'd areas ( where the
* valid bit may be set ) have already been zeroed by
* vm_page_set_validclean().
*/
for (b = i = 0; i <= PAGE_SIZE / DEV_BSIZE; ++i) {
if (i == (PAGE_SIZE / DEV_BSIZE) ||
(m->valid & ((vm_page_bits_t)1 << i))) {
if (i > b) {
pmap_zero_page_area(m,
b << DEV_BSHIFT, (i - b) << DEV_BSHIFT);
}
b = i + 1;
}
}
/*
* setvalid is TRUE when we can safely set the zero'd areas
* as being valid. We can do this if there are no cache consistency
* issues. e.g. it is ok to do with UFS, but not ok to do with NFS.
*/
if (setvalid)
vm_page_valid(m);
}
/*
* vm_page_is_valid:
*
* Is (partial) page valid? Note that the case where size == 0
* will return FALSE in the degenerate case where the page is
* entirely invalid, and TRUE otherwise.
*
* Some callers envoke this routine without the busy lock held and
* handle races via higher level locks. Typical callers should
* hold a busy lock to prevent invalidation.
*/
int
vm_page_is_valid(vm_page_t m, int base, int size)
{
vm_page_bits_t bits;
bits = vm_page_bits(base, size);
return (vm_page_any_valid(m) && (m->valid & bits) == bits);
}
/*
* Returns true if all of the specified predicates are true for the entire
* (super)page and false otherwise.
*/
bool
vm_page_ps_test(vm_page_t m, int psind, int flags, vm_page_t skip_m)
{
vm_object_t object;
int i, npages;
object = m->object;
if (skip_m != NULL && skip_m->object != object)
return (false);
VM_OBJECT_ASSERT_LOCKED(object);
KASSERT(psind <= m->psind,
("psind %d > psind %d of m %p", psind, m->psind, m));
npages = atop(pagesizes[psind]);
/*
* The physically contiguous pages that make up a superpage, i.e., a
* page with a page size index ("psind") greater than zero, will
* occupy adjacent entries in vm_page_array[].
*/
for (i = 0; i < npages; i++) {
/* Always test object consistency, including "skip_m". */
if (m[i].object != object)
return (false);
if (&m[i] == skip_m)
continue;
if ((flags & PS_NONE_BUSY) != 0 && vm_page_busied(&m[i]))
return (false);
if ((flags & PS_ALL_DIRTY) != 0) {
/*
* Calling vm_page_test_dirty() or pmap_is_modified()
* might stop this case from spuriously returning
* "false". However, that would require a write lock
* on the object containing "m[i]".
*/
if (m[i].dirty != VM_PAGE_BITS_ALL)
return (false);
}
if ((flags & PS_ALL_VALID) != 0 &&
m[i].valid != VM_PAGE_BITS_ALL)
return (false);
}
return (true);
}
/*
* Set the page's dirty bits if the page is modified.
*/
void
vm_page_test_dirty(vm_page_t m)
{
vm_page_assert_busied(m);
if (m->dirty != VM_PAGE_BITS_ALL && pmap_is_modified(m))
vm_page_dirty(m);
}
void
vm_page_valid(vm_page_t m)
{
vm_page_assert_busied(m);
if (vm_page_xbusied(m))
m->valid = VM_PAGE_BITS_ALL;
else
vm_page_bits_set(m, &m->valid, VM_PAGE_BITS_ALL);
}
void
vm_page_lock_KBI(vm_page_t m, const char *file, int line)
{
mtx_lock_flags_(vm_page_lockptr(m), 0, file, line);
}
void
vm_page_unlock_KBI(vm_page_t m, const char *file, int line)
{
mtx_unlock_flags_(vm_page_lockptr(m), 0, file, line);
}
int
vm_page_trylock_KBI(vm_page_t m, const char *file, int line)
{
return (mtx_trylock_flags_(vm_page_lockptr(m), 0, file, line));
}
#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
void
vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line)
{
vm_page_lock_assert_KBI(m, MA_OWNED, file, line);
}
void
vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line)
{
mtx_assert_(vm_page_lockptr(m), a, file, line);
}
#endif
#ifdef INVARIANTS
void
vm_page_object_busy_assert(vm_page_t m)
{
/*
* Certain of the page's fields may only be modified by the
* holder of a page or object busy.
*/
if (m->object != NULL && !vm_page_busied(m))
VM_OBJECT_ASSERT_BUSY(m->object);
}
void
vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits)
{
if ((bits & PGA_WRITEABLE) == 0)
return;
/*
* The PGA_WRITEABLE flag can only be set if the page is
* managed, is exclusively busied or the object is locked.
* Currently, this flag is only set by pmap_enter().
*/
KASSERT((m->oflags & VPO_UNMANAGED) == 0,
("PGA_WRITEABLE on unmanaged page"));
if (!vm_page_xbusied(m))
VM_OBJECT_ASSERT_BUSY(m->object);
}
#endif
#include "opt_ddb.h"
#ifdef DDB
#include <sys/kernel.h>
#include <ddb/ddb.h>
DB_SHOW_COMMAND_FLAGS(page, vm_page_print_page_info, DB_CMD_MEMSAFE)
{
db_printf("vm_cnt.v_free_count: %d\n", vm_free_count());
db_printf("vm_cnt.v_inactive_count: %d\n", vm_inactive_count());
db_printf("vm_cnt.v_active_count: %d\n", vm_active_count());
db_printf("vm_cnt.v_laundry_count: %d\n", vm_laundry_count());
db_printf("vm_cnt.v_wire_count: %d\n", vm_wire_count());
db_printf("vm_cnt.v_free_reserved: %d\n", vm_cnt.v_free_reserved);
db_printf("vm_cnt.v_free_min: %d\n", vm_cnt.v_free_min);
db_printf("vm_cnt.v_free_target: %d\n", vm_cnt.v_free_target);
db_printf("vm_cnt.v_inactive_target: %d\n", vm_cnt.v_inactive_target);
}
DB_SHOW_COMMAND_FLAGS(pageq, vm_page_print_pageq_info, DB_CMD_MEMSAFE)
{
int dom;
db_printf("pq_free %d\n", vm_free_count());
for (dom = 0; dom < vm_ndomains; dom++) {
db_printf(
"dom %d page_cnt %d free %d pq_act %d pq_inact %d pq_laund %d pq_unsw %d\n",
dom,
vm_dom[dom].vmd_page_count,
vm_dom[dom].vmd_free_count,
vm_dom[dom].vmd_pagequeues[PQ_ACTIVE].pq_cnt,
vm_dom[dom].vmd_pagequeues[PQ_INACTIVE].pq_cnt,
vm_dom[dom].vmd_pagequeues[PQ_LAUNDRY].pq_cnt,
vm_dom[dom].vmd_pagequeues[PQ_UNSWAPPABLE].pq_cnt);
}
}
DB_SHOW_COMMAND(pginfo, vm_page_print_pginfo)
{
vm_page_t m;
boolean_t phys, virt;
if (!have_addr) {
db_printf("show pginfo addr\n");
return;
}
phys = strchr(modif, 'p') != NULL;
virt = strchr(modif, 'v') != NULL;
if (virt)
m = PHYS_TO_VM_PAGE(pmap_kextract(addr));
else if (phys)
m = PHYS_TO_VM_PAGE(addr);
else
m = (vm_page_t)addr;
db_printf(
"page %p obj %p pidx 0x%jx phys 0x%jx q %d ref 0x%x\n"
" af 0x%x of 0x%x f 0x%x act %d busy %x valid 0x%x dirty 0x%x\n",
m, m->object, (uintmax_t)m->pindex, (uintmax_t)m->phys_addr,
m->a.queue, m->ref_count, m->a.flags, m->oflags,
m->flags, m->a.act_count, m->busy_lock, m->valid, m->dirty);
}
#endif /* DDB */
diff --git a/sys/vm/vm_page.h b/sys/vm/vm_page.h
index 893608bcacf1..613896e77dd9 100644
--- a/sys/vm/vm_page.h
+++ b/sys/vm/vm_page.h
@@ -1,1040 +1,1042 @@
/*-
* SPDX-License-Identifier: (BSD-3-Clause AND MIT-CMU)
*
* Copyright (c) 1991, 1993
* The Regents of the University of California. All rights reserved.
*
* This code is derived from software contributed to Berkeley by
* The Mach Operating System project at Carnegie-Mellon University.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in the
* documentation and/or other materials provided with the distribution.
* 3. Neither the name of the University nor the names of its contributors
* may be used to endorse or promote products derived from this software
* without specific prior written permission.
*
* THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*
*
* Copyright (c) 1987, 1990 Carnegie-Mellon University.
* All rights reserved.
*
* Authors: Avadis Tevanian, Jr., Michael Wayne Young
*
* Permission to use, copy, modify and distribute this software and
* its documentation is hereby granted, provided that both the copyright
* notice and this permission notice appear in all copies of the
* software, derivative works or modified versions, and any portions
* thereof, and that both notices appear in supporting documentation.
*
* CARNEGIE MELLON ALLOWS FREE USE OF THIS SOFTWARE IN ITS "AS IS"
* CONDITION. CARNEGIE MELLON DISCLAIMS ANY LIABILITY OF ANY KIND
* FOR ANY DAMAGES WHATSOEVER RESULTING FROM THE USE OF THIS SOFTWARE.
*
* Carnegie Mellon requests users of this software to return to
*
* Software Distribution Coordinator or Software.Distribution@CS.CMU.EDU
* School of Computer Science
* Carnegie Mellon University
* Pittsburgh PA 15213-3890
*
* any improvements or extensions that they make and grant Carnegie the
* rights to redistribute these changes.
*/
/*
* Resident memory system definitions.
*/
#ifndef _VM_PAGE_
#define _VM_PAGE_
#include <vm/pmap.h>
#include <vm/_vm_phys.h>
/*
* Management of resident (logical) pages.
*
* A small structure is kept for each resident
* page, indexed by page number. Each structure
* is an element of several collections:
*
* A radix tree used to quickly
* perform object/offset lookups
*
* A list of all pages for a given object,
* so they can be quickly deactivated at
* time of deallocation.
*
* An ordered list of pages due for pageout.
*
* In addition, the structure contains the object
* and offset to which this page belongs (for pageout),
* and sundry status bits.
*
* In general, operations on this structure's mutable fields are
* synchronized using either one of or a combination of locks. If a
* field is annotated with two of these locks then holding either is
* sufficient for read access but both are required for write access.
* The queue lock for a page depends on the value of its queue field and is
* described in detail below.
*
* The following annotations are possible:
* (A) the field must be accessed using atomic(9) and may require
* additional synchronization.
* (B) the page busy lock.
* (C) the field is immutable.
* (F) the per-domain lock for the free queues.
* (M) Machine dependent, defined by pmap layer.
* (O) the object that the page belongs to.
* (Q) the page's queue lock.
*
* The busy lock is an embedded reader-writer lock that protects the
* page's contents and identity (i.e., its <object, pindex> tuple) as
* well as certain valid/dirty modifications. To avoid bloating the
* the page structure, the busy lock lacks some of the features available
* the kernel's general-purpose synchronization primitives. As a result,
* busy lock ordering rules are not verified, lock recursion is not
* detected, and an attempt to xbusy a busy page or sbusy an xbusy page
* results will trigger a panic rather than causing the thread to block.
* vm_page_sleep_if_busy() can be used to sleep until the page's busy
* state changes, after which the caller must re-lookup the page and
* re-evaluate its state. vm_page_busy_acquire() will block until
* the lock is acquired.
*
* The valid field is protected by the page busy lock (B) and object
* lock (O). Transitions from invalid to valid are generally done
* via I/O or zero filling and do not require the object lock.
* These must be protected with the busy lock to prevent page-in or
* creation races. Page invalidation generally happens as a result
* of truncate or msync. When invalidated, pages must not be present
* in pmap and must hold the object lock to prevent concurrent
* speculative read-only mappings that do not require busy. I/O
* routines may check for validity without a lock if they are prepared
* to handle invalidation races with higher level locks (vnode) or are
* unconcerned with races so long as they hold a reference to prevent
* recycling. When a valid bit is set while holding a shared busy
* lock (A) atomic operations are used to protect against concurrent
* modification.
*
* In contrast, the synchronization of accesses to the page's
* dirty field is a mix of machine dependent (M) and busy (B). In
* the machine-independent layer, the page busy must be held to
* operate on the field. However, the pmap layer is permitted to
* set all bits within the field without holding that lock. If the
* underlying architecture does not support atomic read-modify-write
* operations on the field's type, then the machine-independent
* layer uses a 32-bit atomic on the aligned 32-bit word that
* contains the dirty field. In the machine-independent layer,
* the implementation of read-modify-write operations on the
* field is encapsulated in vm_page_clear_dirty_mask(). An
* exclusive busy lock combined with pmap_remove_{write/all}() is the
* only way to ensure a page can not become dirty. I/O generally
* removes the page from pmap to ensure exclusive access and atomic
* writes.
*
* The ref_count field tracks references to the page. References that
* prevent the page from being reclaimable are called wirings and are
* counted in the low bits of ref_count. The containing object's
* reference, if one exists, is counted using the VPRC_OBJREF bit in the
* ref_count field. Additionally, the VPRC_BLOCKED bit is used to
* atomically check for wirings and prevent new wirings via
* pmap_extract_and_hold(). When a page belongs to an object, it may be
* wired only when the object is locked, or the page is busy, or by
* pmap_extract_and_hold(). As a result, if the object is locked and the
* page is not busy (or is exclusively busied by the current thread), and
* the page is unmapped, its wire count will not increase. The ref_count
* field is updated using atomic operations in most cases, except when it
* is known that no other references to the page exist, such as in the page
* allocator. A page may be present in the page queues, or even actively
* scanned by the page daemon, without an explicitly counted referenced.
* The page daemon must therefore handle the possibility of a concurrent
* free of the page.
*
* The queue state of a page consists of the queue and act_count fields of
* its atomically updated state, and the subset of atomic flags specified
* by PGA_QUEUE_STATE_MASK. The queue field contains the page's page queue
* index, or PQ_NONE if it does not belong to a page queue. To modify the
* queue field, the page queue lock corresponding to the old value must be
* held, unless that value is PQ_NONE, in which case the queue index must
* be updated using an atomic RMW operation. There is one exception to
* this rule: the page daemon may transition the queue field from
* PQ_INACTIVE to PQ_NONE immediately prior to freeing the page during an
* inactive queue scan. At that point the page is already dequeued and no
* other references to that vm_page structure can exist. The PGA_ENQUEUED
* flag, when set, indicates that the page structure is physically inserted
* into the queue corresponding to the page's queue index, and may only be
* set or cleared with the corresponding page queue lock held.
*
* To avoid contention on page queue locks, page queue operations (enqueue,
* dequeue, requeue) are batched using fixed-size per-CPU queues. A
* deferred operation is requested by setting one of the flags in
* PGA_QUEUE_OP_MASK and inserting an entry into a batch queue. When a
* queue is full, an attempt to insert a new entry will lock the page
* queues and trigger processing of the pending entries. The
* type-stability of vm_page structures is crucial to this scheme since the
* processing of entries in a given batch queue may be deferred
* indefinitely. In particular, a page may be freed with pending batch
* queue entries. The page queue operation flags must be set using atomic
* RWM operations.
*/
#if PAGE_SIZE == 4096
#define VM_PAGE_BITS_ALL 0xffu
typedef uint8_t vm_page_bits_t;
#elif PAGE_SIZE == 8192
#define VM_PAGE_BITS_ALL 0xffffu
typedef uint16_t vm_page_bits_t;
#elif PAGE_SIZE == 16384
#define VM_PAGE_BITS_ALL 0xffffffffu
typedef uint32_t vm_page_bits_t;
#elif PAGE_SIZE == 32768
#define VM_PAGE_BITS_ALL 0xfffffffffffffffflu
typedef uint64_t vm_page_bits_t;
#endif
typedef union vm_page_astate {
struct {
uint16_t flags;
uint8_t queue;
uint8_t act_count;
};
uint32_t _bits;
} vm_page_astate_t;
struct vm_page {
union {
TAILQ_ENTRY(vm_page) q; /* page queue or free list (Q) */
struct {
SLIST_ENTRY(vm_page) ss; /* private slists */
} s;
struct {
u_long p;
u_long v;
} memguard;
struct {
void *slab;
void *zone;
} uma;
} plinks;
TAILQ_ENTRY(vm_page) listq; /* pages in same object (O) */
vm_object_t object; /* which object am I in (O) */
vm_pindex_t pindex; /* offset into object (O,P) */
vm_paddr_t phys_addr; /* physical address of page (C) */
struct md_page md; /* machine dependent stuff */
u_int ref_count; /* page references (A) */
u_int busy_lock; /* busy owners lock (A) */
union vm_page_astate a; /* state accessed atomically (A) */
uint8_t order; /* index of the buddy queue (F) */
uint8_t pool; /* vm_phys freepool index (F) */
uint8_t flags; /* page PG_* flags (P) */
uint8_t oflags; /* page VPO_* flags (O) */
int8_t psind; /* pagesizes[] index (O) */
int8_t segind; /* vm_phys segment index (C) */
/* NOTE that these must support one bit per DEV_BSIZE in a page */
/* so, on normal X86 kernels, they must be at least 8 bits wide */
vm_page_bits_t valid; /* valid DEV_BSIZE chunk map (O,B) */
vm_page_bits_t dirty; /* dirty DEV_BSIZE chunk map (M,B) */
};
/*
* Special bits used in the ref_count field.
*
* ref_count is normally used to count wirings that prevent the page from being
* reclaimed, but also supports several special types of references that do not
* prevent reclamation. Accesses to the ref_count field must be atomic unless
* the page is unallocated.
*
* VPRC_OBJREF is the reference held by the containing object. It can set or
* cleared only when the corresponding object's write lock is held.
*
* VPRC_BLOCKED is used to atomically block wirings via pmap lookups while
* attempting to tear down all mappings of a given page. The page busy lock and
* object write lock must both be held in order to set or clear this bit.
*/
#define VPRC_BLOCKED 0x40000000u /* mappings are being removed */
#define VPRC_OBJREF 0x80000000u /* object reference, cleared with (O) */
#define VPRC_WIRE_COUNT(c) ((c) & ~(VPRC_BLOCKED | VPRC_OBJREF))
#define VPRC_WIRE_COUNT_MAX (~(VPRC_BLOCKED | VPRC_OBJREF))
/*
* Page flags stored in oflags:
*
* Access to these page flags is synchronized by the lock on the object
* containing the page (O).
*
* Note: VPO_UNMANAGED (used by OBJT_DEVICE, OBJT_PHYS and OBJT_SG)
* indicates that the page is not under PV management but
* otherwise should be treated as a normal page. Pages not
* under PV management cannot be paged out via the
* object/vm_page_t because there is no knowledge of their pte
* mappings, and such pages are also not on any PQ queue.
*
*/
#define VPO_KMEM_EXEC 0x01 /* kmem mapping allows execution */
#define VPO_SWAPSLEEP 0x02 /* waiting for swap to finish */
#define VPO_UNMANAGED 0x04 /* no PV management for page */
#define VPO_SWAPINPROG 0x08 /* swap I/O in progress on page */
/*
* Busy page implementation details.
* The algorithm is taken mostly by rwlock(9) and sx(9) locks implementation,
* even if the support for owner identity is removed because of size
* constraints. Checks on lock recursion are then not possible, while the
* lock assertions effectiveness is someway reduced.
*/
#define VPB_BIT_SHARED 0x01
#define VPB_BIT_EXCLUSIVE 0x02
#define VPB_BIT_WAITERS 0x04
#define VPB_BIT_FLAGMASK \
(VPB_BIT_SHARED | VPB_BIT_EXCLUSIVE | VPB_BIT_WAITERS)
#define VPB_SHARERS_SHIFT 3
#define VPB_SHARERS(x) \
(((x) & ~VPB_BIT_FLAGMASK) >> VPB_SHARERS_SHIFT)
#define VPB_SHARERS_WORD(x) ((x) << VPB_SHARERS_SHIFT | VPB_BIT_SHARED)
#define VPB_ONE_SHARER (1 << VPB_SHARERS_SHIFT)
#define VPB_SINGLE_EXCLUSIVE VPB_BIT_EXCLUSIVE
#ifdef INVARIANTS
#define VPB_CURTHREAD_EXCLUSIVE \
(VPB_BIT_EXCLUSIVE | ((u_int)(uintptr_t)curthread & ~VPB_BIT_FLAGMASK))
#else
#define VPB_CURTHREAD_EXCLUSIVE VPB_SINGLE_EXCLUSIVE
#endif
#define VPB_UNBUSIED VPB_SHARERS_WORD(0)
/* Freed lock blocks both shared and exclusive. */
#define VPB_FREED (0xffffffff - VPB_BIT_SHARED)
#define PQ_NONE 255
#define PQ_INACTIVE 0
#define PQ_ACTIVE 1
#define PQ_LAUNDRY 2
#define PQ_UNSWAPPABLE 3
#define PQ_COUNT 4
#ifndef VM_PAGE_HAVE_PGLIST
TAILQ_HEAD(pglist, vm_page);
#define VM_PAGE_HAVE_PGLIST
#endif
SLIST_HEAD(spglist, vm_page);
#ifdef _KERNEL
extern vm_page_t bogus_page;
#endif /* _KERNEL */
extern struct mtx_padalign pa_lock[];
#if defined(__arm__)
#define PDRSHIFT PDR_SHIFT
#elif !defined(PDRSHIFT)
#define PDRSHIFT 21
#endif
#define pa_index(pa) ((pa) >> PDRSHIFT)
#define PA_LOCKPTR(pa) ((struct mtx *)(&pa_lock[pa_index(pa) % PA_LOCK_COUNT]))
#define PA_LOCKOBJPTR(pa) ((struct lock_object *)PA_LOCKPTR((pa)))
#define PA_LOCK(pa) mtx_lock(PA_LOCKPTR(pa))
#define PA_TRYLOCK(pa) mtx_trylock(PA_LOCKPTR(pa))
#define PA_UNLOCK(pa) mtx_unlock(PA_LOCKPTR(pa))
#define PA_UNLOCK_COND(pa) \
do { \
if ((pa) != 0) { \
PA_UNLOCK((pa)); \
(pa) = 0; \
} \
} while (0)
#define PA_LOCK_ASSERT(pa, a) mtx_assert(PA_LOCKPTR(pa), (a))
#if defined(KLD_MODULE) && !defined(KLD_TIED)
#define vm_page_lock(m) vm_page_lock_KBI((m), LOCK_FILE, LOCK_LINE)
#define vm_page_unlock(m) vm_page_unlock_KBI((m), LOCK_FILE, LOCK_LINE)
#define vm_page_trylock(m) vm_page_trylock_KBI((m), LOCK_FILE, LOCK_LINE)
#else /* !KLD_MODULE */
#define vm_page_lockptr(m) (PA_LOCKPTR(VM_PAGE_TO_PHYS((m))))
#define vm_page_lock(m) mtx_lock(vm_page_lockptr((m)))
#define vm_page_unlock(m) mtx_unlock(vm_page_lockptr((m)))
#define vm_page_trylock(m) mtx_trylock(vm_page_lockptr((m)))
#endif
#if defined(INVARIANTS)
#define vm_page_assert_locked(m) \
vm_page_assert_locked_KBI((m), __FILE__, __LINE__)
#define vm_page_lock_assert(m, a) \
vm_page_lock_assert_KBI((m), (a), __FILE__, __LINE__)
#else
#define vm_page_assert_locked(m)
#define vm_page_lock_assert(m, a)
#endif
/*
* The vm_page's aflags are updated using atomic operations. To set or clear
* these flags, the functions vm_page_aflag_set() and vm_page_aflag_clear()
* must be used. Neither these flags nor these functions are part of the KBI.
*
* PGA_REFERENCED may be cleared only if the page is locked. It is set by
* both the MI and MD VM layers. However, kernel loadable modules should not
* directly set this flag. They should call vm_page_reference() instead.
*
* PGA_WRITEABLE is set exclusively on managed pages by pmap_enter().
* When it does so, the object must be locked, or the page must be
* exclusive busied. The MI VM layer must never access this flag
* directly. Instead, it should call pmap_page_is_write_mapped().
*
* PGA_EXECUTABLE may be set by pmap routines, and indicates that a page has
* at least one executable mapping. It is not consumed by the MI VM layer.
*
* PGA_NOSYNC must be set and cleared with the page busy lock held.
*
* PGA_ENQUEUED is set and cleared when a page is inserted into or removed
* from a page queue, respectively. It determines whether the plinks.q field
* of the page is valid. To set or clear this flag, page's "queue" field must
* be a valid queue index, and the corresponding page queue lock must be held.
*
* PGA_DEQUEUE is set when the page is scheduled to be dequeued from a page
* queue, and cleared when the dequeue request is processed. A page may
* have PGA_DEQUEUE set and PGA_ENQUEUED cleared, for instance if a dequeue
* is requested after the page is scheduled to be enqueued but before it is
* actually inserted into the page queue.
*
* PGA_REQUEUE is set when the page is scheduled to be enqueued or requeued
* in its page queue.
*
* PGA_REQUEUE_HEAD is a special flag for enqueuing pages near the head of
* the inactive queue, thus bypassing LRU.
*
* The PGA_DEQUEUE, PGA_REQUEUE and PGA_REQUEUE_HEAD flags must be set using an
* atomic RMW operation to ensure that the "queue" field is a valid queue index,
* and the corresponding page queue lock must be held when clearing any of the
* flags.
*
* PGA_SWAP_FREE is used to defer freeing swap space to the pageout daemon
* when the context that dirties the page does not have the object write lock
* held.
*/
#define PGA_WRITEABLE 0x0001 /* page may be mapped writeable */
#define PGA_REFERENCED 0x0002 /* page has been referenced */
#define PGA_EXECUTABLE 0x0004 /* page may be mapped executable */
#define PGA_ENQUEUED 0x0008 /* page is enqueued in a page queue */
#define PGA_DEQUEUE 0x0010 /* page is due to be dequeued */
#define PGA_REQUEUE 0x0020 /* page is due to be requeued */
#define PGA_REQUEUE_HEAD 0x0040 /* page requeue should bypass LRU */
#define PGA_NOSYNC 0x0080 /* do not collect for syncer */
#define PGA_SWAP_FREE 0x0100 /* page with swap space was dirtied */
#define PGA_SWAP_SPACE 0x0200 /* page has allocated swap space */
#define PGA_QUEUE_OP_MASK (PGA_DEQUEUE | PGA_REQUEUE | PGA_REQUEUE_HEAD)
#define PGA_QUEUE_STATE_MASK (PGA_ENQUEUED | PGA_QUEUE_OP_MASK)
/*
* Page flags. Updates to these flags are not synchronized, and thus they must
* be set during page allocation or free to avoid races.
*
* The PG_PCPU_CACHE flag is set at allocation time if the page was
* allocated from a per-CPU cache. It is cleared the next time that the
* page is allocated from the physical memory allocator.
*/
#define PG_PCPU_CACHE 0x01 /* was allocated from per-CPU caches */
#define PG_FICTITIOUS 0x02 /* physical page doesn't exist */
#define PG_ZERO 0x04 /* page is zeroed */
#define PG_MARKER 0x08 /* special queue marker page */
#define PG_NODUMP 0x10 /* don't include this page in a dump */
#define PG_NOFREE 0x20 /* page should never be freed. */
/*
* Misc constants.
*/
#define ACT_DECLINE 1
#define ACT_ADVANCE 3
#define ACT_INIT 5
#define ACT_MAX 64
#ifdef _KERNEL
#include <sys/kassert.h>
#include <machine/atomic.h>
struct pctrie_iter;
/*
* Each pageable resident page falls into one of five lists:
*
* free
* Available for allocation now.
*
* inactive
* Low activity, candidates for reclamation.
* This list is approximately LRU ordered.
*
* laundry
* This is the list of pages that should be
* paged out next.
*
* unswappable
* Dirty anonymous pages that cannot be paged
* out because no swap device is configured.
*
* active
* Pages that are "active", i.e., they have been
* recently referenced.
*
*/
extern vm_page_t vm_page_array; /* First resident page in table */
extern long vm_page_array_size; /* number of vm_page_t's */
extern long first_page; /* first physical page number */
#define VM_PAGE_TO_PHYS(entry) ((entry)->phys_addr)
/*
* PHYS_TO_VM_PAGE() returns the vm_page_t object that represents a memory
* page to which the given physical address belongs. The correct vm_page_t
* object is returned for addresses that are not page-aligned.
*/
vm_page_t PHYS_TO_VM_PAGE(vm_paddr_t pa);
/*
* Page allocation parameters for vm_page for the functions
* vm_page_alloc(), vm_page_grab(), vm_page_alloc_contig() and
* vm_page_alloc_freelist(). Some functions support only a subset
* of the flags, and ignore others, see the flags legend.
*
* The meaning of VM_ALLOC_ZERO differs slightly between the vm_page_alloc*()
* and the vm_page_grab*() functions. See these functions for details.
*
* Bits 0 - 1 define class.
* Bits 2 - 15 dedicated for flags.
* Legend:
* (a) - vm_page_alloc() supports the flag.
* (c) - vm_page_alloc_contig() supports the flag.
* (g) - vm_page_grab() supports the flag.
* (n) - vm_page_alloc_noobj() and vm_page_alloc_freelist() support the flag.
* (p) - vm_page_grab_pages() supports the flag.
* Bits above 15 define the count of additional pages that the caller
* intends to allocate.
*/
#define VM_ALLOC_NORMAL 0
#define VM_ALLOC_INTERRUPT 1
#define VM_ALLOC_SYSTEM 2
#define VM_ALLOC_CLASS_MASK 3
#define VM_ALLOC_WAITOK 0x0008 /* (acn) Sleep and retry */
#define VM_ALLOC_WAITFAIL 0x0010 /* (acn) Sleep and return error */
#define VM_ALLOC_WIRED 0x0020 /* (acgnp) Allocate a wired page */
#define VM_ALLOC_ZERO 0x0040 /* (acgnp) Allocate a zeroed page */
#define VM_ALLOC_NORECLAIM 0x0080 /* (c) Do not reclaim after failure */
#define VM_ALLOC_NOFREE 0x0100 /* (an) Page will never be released */
#define VM_ALLOC_NOBUSY 0x0200 /* (acgp) Do not excl busy the page */
#define VM_ALLOC_NOCREAT 0x0400 /* (gp) Don't create a page */
#define VM_ALLOC_AVAIL1 0x0800
#define VM_ALLOC_IGN_SBUSY 0x1000 /* (gp) Ignore shared busy flag */
#define VM_ALLOC_NODUMP 0x2000 /* (ag) don't include in dump */
#define VM_ALLOC_SBUSY 0x4000 /* (acgp) Shared busy the page */
#define VM_ALLOC_NOWAIT 0x8000 /* (acgnp) Do not sleep */
#define VM_ALLOC_COUNT_MAX 0xffff
#define VM_ALLOC_COUNT_SHIFT 16
#define VM_ALLOC_COUNT_MASK (VM_ALLOC_COUNT(VM_ALLOC_COUNT_MAX))
#define VM_ALLOC_COUNT(count) ({ \
KASSERT((count) <= VM_ALLOC_COUNT_MAX, \
("%s: invalid VM_ALLOC_COUNT value", __func__)); \
(count) << VM_ALLOC_COUNT_SHIFT; \
})
#ifdef M_NOWAIT
static inline int
malloc2vm_flags(int malloc_flags)
{
int pflags;
KASSERT((malloc_flags & M_USE_RESERVE) == 0 ||
(malloc_flags & M_NOWAIT) != 0,
("M_USE_RESERVE requires M_NOWAIT"));
pflags = (malloc_flags & M_USE_RESERVE) != 0 ? VM_ALLOC_INTERRUPT :
VM_ALLOC_SYSTEM;
if ((malloc_flags & M_ZERO) != 0)
pflags |= VM_ALLOC_ZERO;
if ((malloc_flags & M_NODUMP) != 0)
pflags |= VM_ALLOC_NODUMP;
if ((malloc_flags & M_NOWAIT))
pflags |= VM_ALLOC_NOWAIT;
if ((malloc_flags & M_WAITOK))
pflags |= VM_ALLOC_WAITOK;
if ((malloc_flags & M_NORECLAIM))
pflags |= VM_ALLOC_NORECLAIM;
if ((malloc_flags & M_NEVERFREED))
pflags |= VM_ALLOC_NOFREE;
return (pflags);
}
#endif
/*
* Predicates supported by vm_page_ps_test():
*
* PS_ALL_DIRTY is true only if the entire (super)page is dirty.
* However, it can be spuriously false when the (super)page has become
* dirty in the pmap but that information has not been propagated to the
* machine-independent layer.
*/
#define PS_ALL_DIRTY 0x1
#define PS_ALL_VALID 0x2
#define PS_NONE_BUSY 0x4
bool vm_page_busy_acquire(vm_page_t m, int allocflags);
void vm_page_busy_downgrade(vm_page_t m);
int vm_page_busy_tryupgrade(vm_page_t m);
bool vm_page_busy_sleep(vm_page_t m, const char *msg, int allocflags);
void vm_page_busy_sleep_unlocked(vm_object_t obj, vm_page_t m,
vm_pindex_t pindex, const char *wmesg, int allocflags);
void vm_page_free(vm_page_t m);
+void vm_page_iter_free(struct pctrie_iter *);
void vm_page_free_zero(vm_page_t m);
void vm_page_activate (vm_page_t);
void vm_page_advise(vm_page_t m, int advice);
vm_page_t vm_page_mpred(vm_object_t, vm_pindex_t);
vm_page_t vm_page_alloc(vm_object_t, vm_pindex_t, int);
vm_page_t vm_page_alloc_domain_after(vm_object_t, vm_pindex_t, int, int,
vm_page_t);
vm_page_t vm_page_alloc_contig(vm_object_t object, vm_pindex_t pindex, int req,
u_long npages, vm_paddr_t low, vm_paddr_t high, u_long alignment,
vm_paddr_t boundary, vm_memattr_t memattr);
vm_page_t vm_page_alloc_contig_domain(vm_object_t object,
vm_pindex_t pindex, int domain, int req, u_long npages, vm_paddr_t low,
vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr);
vm_page_t vm_page_alloc_noobj(int);
vm_page_t vm_page_alloc_noobj_domain(int, int);
vm_page_t vm_page_alloc_noobj_contig(int req, u_long npages, vm_paddr_t low,
vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr);
vm_page_t vm_page_alloc_noobj_contig_domain(int domain, int req, u_long npages,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
vm_memattr_t memattr);
void vm_page_bits_set(vm_page_t m, vm_page_bits_t *bits, vm_page_bits_t set);
bool vm_page_blacklist_add(vm_paddr_t pa, bool verbose);
vm_page_t vm_page_grab(vm_object_t, vm_pindex_t, int);
vm_page_t vm_page_grab_unlocked(vm_object_t, vm_pindex_t, int);
int vm_page_grab_pages(vm_object_t object, vm_pindex_t pindex, int allocflags,
vm_page_t *ma, int count);
int vm_page_grab_pages_unlocked(vm_object_t object, vm_pindex_t pindex,
int allocflags, vm_page_t *ma, int count);
int vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex,
int allocflags);
int vm_page_grab_valid_unlocked(vm_page_t *mp, vm_object_t object,
vm_pindex_t pindex, int allocflags);
void vm_page_deactivate(vm_page_t);
void vm_page_deactivate_noreuse(vm_page_t);
void vm_page_dequeue(vm_page_t m);
void vm_page_dequeue_deferred(vm_page_t m);
vm_page_t vm_page_find_least(vm_object_t, vm_pindex_t);
vm_page_t vm_page_iter_lookup_ge(struct pctrie_iter *, vm_pindex_t);
void vm_page_free_invalid(vm_page_t);
vm_page_t vm_page_getfake(vm_paddr_t paddr, vm_memattr_t memattr);
void vm_page_initfake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
void vm_page_init_marker(vm_page_t marker, int queue, uint16_t aflags);
void vm_page_init_page(vm_page_t m, vm_paddr_t pa, int segind, int pool);
int vm_page_insert (vm_page_t, vm_object_t, vm_pindex_t);
void vm_page_invalid(vm_page_t m);
void vm_page_launder(vm_page_t m);
vm_page_t vm_page_lookup(vm_object_t, vm_pindex_t);
void vm_page_iter_init(struct pctrie_iter *, vm_object_t);
void vm_page_iter_limit_init(struct pctrie_iter *, vm_object_t, vm_pindex_t);
vm_page_t vm_page_iter_lookup(struct pctrie_iter *, vm_pindex_t);
vm_page_t vm_page_lookup_unlocked(vm_object_t, vm_pindex_t);
vm_page_t vm_page_next(vm_page_t m);
void vm_page_pqbatch_drain(void);
void vm_page_pqbatch_submit(vm_page_t m, uint8_t queue);
bool vm_page_pqstate_commit(vm_page_t m, vm_page_astate_t *old,
vm_page_astate_t new);
vm_page_t vm_page_prev(vm_page_t m);
bool vm_page_ps_test(vm_page_t m, int psind, int flags, vm_page_t skip_m);
void vm_page_putfake(vm_page_t m);
void vm_page_readahead_finish(vm_page_t m);
int vm_page_reclaim_contig(int req, u_long npages, vm_paddr_t low,
vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
int vm_page_reclaim_contig_domain(int domain, int req, u_long npages,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary);
int vm_page_reclaim_contig_domain_ext(int domain, int req, u_long npages,
vm_paddr_t low, vm_paddr_t high, u_long alignment, vm_paddr_t boundary,
int desired_runs);
void vm_page_reference(vm_page_t m);
#define VPR_TRYFREE 0x01
#define VPR_NOREUSE 0x02
void vm_page_release(vm_page_t m, int flags);
void vm_page_release_locked(vm_page_t m, int flags);
vm_page_t vm_page_relookup(vm_object_t, vm_pindex_t);
bool vm_page_remove(vm_page_t);
+bool vm_page_iter_remove(struct pctrie_iter *);
bool vm_page_remove_xbusy(vm_page_t);
-int vm_page_rename(vm_page_t, vm_object_t, vm_pindex_t);
+int vm_page_rename(struct pctrie_iter *, vm_object_t, vm_pindex_t);
void vm_page_replace(vm_page_t mnew, vm_object_t object,
vm_pindex_t pindex, vm_page_t mold);
int vm_page_sbusied(vm_page_t m);
vm_page_bits_t vm_page_set_dirty(vm_page_t m);
void vm_page_set_valid_range(vm_page_t m, int base, int size);
vm_offset_t vm_page_startup(vm_offset_t vaddr);
void vm_page_sunbusy(vm_page_t m);
bool vm_page_try_remove_all(vm_page_t m);
bool vm_page_try_remove_write(vm_page_t m);
int vm_page_trysbusy(vm_page_t m);
int vm_page_tryxbusy(vm_page_t m);
void vm_page_unhold_pages(vm_page_t *ma, int count);
void vm_page_unswappable(vm_page_t m);
void vm_page_unwire(vm_page_t m, uint8_t queue);
bool vm_page_unwire_noq(vm_page_t m);
void vm_page_updatefake(vm_page_t m, vm_paddr_t paddr, vm_memattr_t memattr);
void vm_page_wire(vm_page_t);
bool vm_page_wire_mapped(vm_page_t m);
void vm_page_xunbusy_hard(vm_page_t m);
void vm_page_xunbusy_hard_unchecked(vm_page_t m);
void vm_page_set_validclean (vm_page_t, int, int);
void vm_page_clear_dirty(vm_page_t, int, int);
void vm_page_set_invalid(vm_page_t, int, int);
void vm_page_valid(vm_page_t m);
int vm_page_is_valid(vm_page_t, int, int);
void vm_page_test_dirty(vm_page_t);
vm_page_bits_t vm_page_bits(int base, int size);
void vm_page_zero_invalid(vm_page_t m, boolean_t setvalid);
int vm_page_free_pages_toq(struct spglist *free, bool update_wire_count);
void vm_page_dirty_KBI(vm_page_t m);
void vm_page_lock_KBI(vm_page_t m, const char *file, int line);
void vm_page_unlock_KBI(vm_page_t m, const char *file, int line);
int vm_page_trylock_KBI(vm_page_t m, const char *file, int line);
#if defined(INVARIANTS) || defined(INVARIANT_SUPPORT)
void vm_page_assert_locked_KBI(vm_page_t m, const char *file, int line);
void vm_page_lock_assert_KBI(vm_page_t m, int a, const char *file, int line);
#endif
#define vm_page_busy_fetch(m) atomic_load_int(&(m)->busy_lock)
#define vm_page_assert_busied(m) \
KASSERT(vm_page_busied(m), \
("vm_page_assert_busied: page %p not busy @ %s:%d", \
(m), __FILE__, __LINE__))
#define vm_page_assert_sbusied(m) \
KASSERT(vm_page_sbusied(m), \
("vm_page_assert_sbusied: page %p not shared busy @ %s:%d", \
(m), __FILE__, __LINE__))
#define vm_page_assert_unbusied(m) \
KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) != \
VPB_CURTHREAD_EXCLUSIVE, \
("vm_page_assert_unbusied: page %p busy_lock %#x owned" \
" by me (%p) @ %s:%d", \
(m), (m)->busy_lock, curthread, __FILE__, __LINE__)); \
#define vm_page_assert_xbusied_unchecked(m) do { \
KASSERT(vm_page_xbusied(m), \
("vm_page_assert_xbusied: page %p not exclusive busy @ %s:%d", \
(m), __FILE__, __LINE__)); \
} while (0)
#define vm_page_assert_xbusied(m) do { \
vm_page_assert_xbusied_unchecked(m); \
KASSERT((vm_page_busy_fetch(m) & ~VPB_BIT_WAITERS) == \
VPB_CURTHREAD_EXCLUSIVE, \
("vm_page_assert_xbusied: page %p busy_lock %#x not owned" \
" by me (%p) @ %s:%d", \
(m), (m)->busy_lock, curthread, __FILE__, __LINE__)); \
} while (0)
#define vm_page_busied(m) \
(vm_page_busy_fetch(m) != VPB_UNBUSIED)
#define vm_page_xbusied(m) \
((vm_page_busy_fetch(m) & VPB_SINGLE_EXCLUSIVE) != 0)
#define vm_page_busy_freed(m) \
(vm_page_busy_fetch(m) == VPB_FREED)
/* Note: page m's lock must not be owned by the caller. */
#define vm_page_xunbusy(m) do { \
if (!atomic_cmpset_rel_int(&(m)->busy_lock, \
VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \
vm_page_xunbusy_hard(m); \
} while (0)
#define vm_page_xunbusy_unchecked(m) do { \
if (!atomic_cmpset_rel_int(&(m)->busy_lock, \
VPB_CURTHREAD_EXCLUSIVE, VPB_UNBUSIED)) \
vm_page_xunbusy_hard_unchecked(m); \
} while (0)
#ifdef INVARIANTS
void vm_page_object_busy_assert(vm_page_t m);
#define VM_PAGE_OBJECT_BUSY_ASSERT(m) vm_page_object_busy_assert(m)
void vm_page_assert_pga_writeable(vm_page_t m, uint16_t bits);
#define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) \
vm_page_assert_pga_writeable(m, bits)
/*
* Claim ownership of a page's xbusy state. In non-INVARIANTS kernels this
* operation is a no-op since ownership is not tracked. In particular
* this macro does not provide any synchronization with the previous owner.
*/
#define vm_page_xbusy_claim(m) do { \
u_int _busy_lock; \
\
vm_page_assert_xbusied_unchecked((m)); \
do { \
_busy_lock = vm_page_busy_fetch(m); \
} while (!atomic_cmpset_int(&(m)->busy_lock, _busy_lock, \
(_busy_lock & VPB_BIT_FLAGMASK) | VPB_CURTHREAD_EXCLUSIVE)); \
} while (0)
#else
#define VM_PAGE_OBJECT_BUSY_ASSERT(m) (void)0
#define VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits) (void)0
#define vm_page_xbusy_claim(m)
#endif
#if BYTE_ORDER == BIG_ENDIAN
#define VM_PAGE_AFLAG_SHIFT 16
#else
#define VM_PAGE_AFLAG_SHIFT 0
#endif
/*
* Load a snapshot of a page's 32-bit atomic state.
*/
static inline vm_page_astate_t
vm_page_astate_load(vm_page_t m)
{
vm_page_astate_t a;
a._bits = atomic_load_32(&m->a._bits);
return (a);
}
/*
* Atomically compare and set a page's atomic state.
*/
static inline bool
vm_page_astate_fcmpset(vm_page_t m, vm_page_astate_t *old, vm_page_astate_t new)
{
KASSERT(new.queue == PQ_INACTIVE || (new.flags & PGA_REQUEUE_HEAD) == 0,
("%s: invalid head requeue request for page %p", __func__, m));
KASSERT((new.flags & PGA_ENQUEUED) == 0 || new.queue != PQ_NONE,
("%s: setting PGA_ENQUEUED with PQ_NONE in page %p", __func__, m));
KASSERT(new._bits != old->_bits,
("%s: bits are unchanged", __func__));
return (atomic_fcmpset_32(&m->a._bits, &old->_bits, new._bits) != 0);
}
/*
* Clear the given bits in the specified page.
*/
static inline void
vm_page_aflag_clear(vm_page_t m, uint16_t bits)
{
uint32_t *addr, val;
/*
* Access the whole 32-bit word containing the aflags field with an
* atomic update. Parallel non-atomic updates to the other fields
* within this word are handled properly by the atomic update.
*/
addr = (void *)&m->a;
val = bits << VM_PAGE_AFLAG_SHIFT;
atomic_clear_32(addr, val);
}
/*
* Set the given bits in the specified page.
*/
static inline void
vm_page_aflag_set(vm_page_t m, uint16_t bits)
{
uint32_t *addr, val;
VM_PAGE_ASSERT_PGA_WRITEABLE(m, bits);
/*
* Access the whole 32-bit word containing the aflags field with an
* atomic update. Parallel non-atomic updates to the other fields
* within this word are handled properly by the atomic update.
*/
addr = (void *)&m->a;
val = bits << VM_PAGE_AFLAG_SHIFT;
atomic_set_32(addr, val);
}
/*
* vm_page_dirty:
*
* Set all bits in the page's dirty field.
*
* The object containing the specified page must be locked if the
* call is made from the machine-independent layer.
*
* See vm_page_clear_dirty_mask().
*/
static __inline void
vm_page_dirty(vm_page_t m)
{
/* Use vm_page_dirty_KBI() under INVARIANTS to save memory. */
#if (defined(KLD_MODULE) && !defined(KLD_TIED)) || defined(INVARIANTS)
vm_page_dirty_KBI(m);
#else
m->dirty = VM_PAGE_BITS_ALL;
#endif
}
/*
* vm_page_undirty:
*
* Set page to not be dirty. Note: does not clear pmap modify bits
*/
static __inline void
vm_page_undirty(vm_page_t m)
{
VM_PAGE_OBJECT_BUSY_ASSERT(m);
m->dirty = 0;
}
static inline uint8_t
_vm_page_queue(vm_page_astate_t as)
{
if ((as.flags & PGA_DEQUEUE) != 0)
return (PQ_NONE);
return (as.queue);
}
/*
* vm_page_queue:
*
* Return the index of the queue containing m.
*/
static inline uint8_t
vm_page_queue(vm_page_t m)
{
return (_vm_page_queue(vm_page_astate_load(m)));
}
static inline bool
vm_page_active(vm_page_t m)
{
return (vm_page_queue(m) == PQ_ACTIVE);
}
static inline bool
vm_page_inactive(vm_page_t m)
{
return (vm_page_queue(m) == PQ_INACTIVE);
}
static inline bool
vm_page_in_laundry(vm_page_t m)
{
uint8_t queue;
queue = vm_page_queue(m);
return (queue == PQ_LAUNDRY || queue == PQ_UNSWAPPABLE);
}
static inline void
vm_page_clearref(vm_page_t m)
{
u_int r;
r = m->ref_count;
while (atomic_fcmpset_int(&m->ref_count, &r, r & (VPRC_BLOCKED |
VPRC_OBJREF)) == 0)
;
}
/*
* vm_page_drop:
*
* Release a reference to a page and return the old reference count.
*/
static inline u_int
vm_page_drop(vm_page_t m, u_int val)
{
u_int old;
/*
* Synchronize with vm_page_free_prep(): ensure that all updates to the
* page structure are visible before it is freed.
*/
atomic_thread_fence_rel();
old = atomic_fetchadd_int(&m->ref_count, -val);
KASSERT(old != VPRC_BLOCKED,
("vm_page_drop: page %p has an invalid refcount value", m));
return (old);
}
/*
* vm_page_wired:
*
* Perform a racy check to determine whether a reference prevents the page
* from being reclaimable. If the page's object is locked, and the page is
* unmapped and exclusively busied by the current thread, no new wirings
* may be created.
*/
static inline bool
vm_page_wired(vm_page_t m)
{
return (VPRC_WIRE_COUNT(m->ref_count) > 0);
}
static inline bool
vm_page_all_valid(vm_page_t m)
{
return (m->valid == VM_PAGE_BITS_ALL);
}
static inline bool
vm_page_any_valid(vm_page_t m)
{
return (m->valid != 0);
}
static inline bool
vm_page_none_valid(vm_page_t m)
{
return (m->valid == 0);
}
static inline int
vm_page_domain(vm_page_t m __numa_used)
{
#ifdef NUMA
int domn, segind;
segind = m->segind;
KASSERT(segind < vm_phys_nsegs, ("segind %d m %p", segind, m));
domn = vm_phys_segs[segind].domain;
KASSERT(domn >= 0 && domn < vm_ndomains, ("domain %d m %p", domn, m));
return (domn);
#else
return (0);
#endif
}
#endif /* _KERNEL */
#endif /* !_VM_PAGE_ */