#include <debug.h>
#include <libkern/OSAtomic.h>
#include <mach/clock_types.h>
#include <mach/vm_prot.h>
#include <mach/vm_statistics.h>
#include <mach/sdt.h>
#include <kern/counters.h>
#include <kern/sched_prim.h>
#include <kern/task.h>
#include <kern/thread.h>
#include <kern/zalloc.h>
#include <kern/xpr.h>
#include <vm/pmap.h>
#include <vm/vm_init.h>
#include <vm/vm_map.h>
#include <vm/vm_page.h>
#include <vm/vm_pageout.h>
#include <vm/vm_kern.h>
#include <kern/misc_protos.h>
#include <zone_debug.h>
#include <vm/cpm.h>
#include <ppc/mappings.h>
#include <pexpert/pexpert.h>
#include <vm/vm_protos.h>
#include <vm/memory_object.h>
#include <vm/vm_purgeable_internal.h>
#if CONFIG_EMBEDDED
#include <sys/kern_memorystatus.h>
#endif
int speculative_age_index = 0;
int speculative_steal_index = 0;
struct vm_speculative_age_q vm_page_queue_speculative[VM_PAGE_MAX_SPECULATIVE_AGE_Q + 1];
vm_offset_t virtual_space_start;
vm_offset_t virtual_space_end;
int vm_page_pages;
typedef struct {
vm_page_t pages;
#if MACH_PAGE_HASH_STATS
int cur_count;
int hi_count;
#endif
} vm_page_bucket_t;
vm_page_bucket_t *vm_page_buckets;
unsigned int vm_page_bucket_count = 0;
unsigned int vm_page_hash_mask;
unsigned int vm_page_hash_shift;
uint32_t vm_page_bucket_hash;
decl_simple_lock_data(,vm_page_bucket_lock)
#if MACH_PAGE_HASH_STATS
void
hash_debug(void)
{
int i;
int numbuckets = 0;
int highsum = 0;
int maxdepth = 0;
for (i = 0; i < vm_page_bucket_count; i++) {
if (vm_page_buckets[i].hi_count) {
numbuckets++;
highsum += vm_page_buckets[i].hi_count;
if (vm_page_buckets[i].hi_count > maxdepth)
maxdepth = vm_page_buckets[i].hi_count;
}
}
printf("Total number of buckets: %d\n", vm_page_bucket_count);
printf("Number used buckets: %d = %d%%\n",
numbuckets, 100*numbuckets/vm_page_bucket_count);
printf("Number unused buckets: %d = %d%%\n",
vm_page_bucket_count - numbuckets,
100*(vm_page_bucket_count-numbuckets)/vm_page_bucket_count);
printf("Sum of bucket max depth: %d\n", highsum);
printf("Average bucket depth: %d.%2d\n",
highsum/vm_page_bucket_count,
highsum%vm_page_bucket_count);
printf("Maximum bucket depth: %d\n", maxdepth);
}
#endif
vm_size_t page_size = PAGE_SIZE;
vm_size_t page_mask = PAGE_MASK;
int page_shift = PAGE_SHIFT;
struct vm_page vm_page_template;
vm_page_t vm_pages = VM_PAGE_NULL;
unsigned int vm_pages_count = 0;
unsigned int vm_colors;
unsigned int vm_color_mask;
unsigned int vm_cache_geometry_colors = 0;
queue_head_t vm_page_queue_free[MAX_COLORS];
vm_page_t vm_page_queue_fictitious;
unsigned int vm_page_free_wanted;
unsigned int vm_page_free_wanted_privileged;
unsigned int vm_page_free_count;
unsigned int vm_page_fictitious_count;
unsigned int vm_page_free_count_minimum;
zone_t vm_page_zone;
decl_mutex_data(,vm_page_alloc_lock)
unsigned int io_throttle_zero_fill;
vm_offset_t vm_page_fictitious_addr = (vm_offset_t) -1;
vm_offset_t vm_page_guard_addr = (vm_offset_t) -2;
queue_head_t vm_page_queue_active;
queue_head_t vm_page_queue_inactive;
queue_head_t vm_page_queue_zf;
unsigned int vm_page_active_count;
unsigned int vm_page_inactive_count;
unsigned int vm_page_throttled_count;
unsigned int vm_page_speculative_count;
unsigned int vm_page_wire_count;
unsigned int vm_page_gobble_count = 0;
unsigned int vm_page_wire_count_warning = 0;
unsigned int vm_page_gobble_count_warning = 0;
unsigned int vm_page_purgeable_count = 0;
uint64_t vm_page_purged_count = 0;
unsigned int vm_page_speculative_recreated = 0;
unsigned int vm_page_speculative_created = 0;
unsigned int vm_page_speculative_used = 0;
ppnum_t vm_lopage_poolstart = 0;
ppnum_t vm_lopage_poolend = 0;
int vm_lopage_poolsize = 0;
uint64_t max_valid_dma_address = 0xffffffffffffffffULL;
unsigned int vm_page_free_target = 0;
unsigned int vm_page_free_min = 0;
unsigned int vm_page_inactive_target = 0;
unsigned int vm_page_inactive_min = 0;
unsigned int vm_page_free_reserved = 0;
unsigned int vm_page_zfill_throttle_count = 0;
boolean_t vm_page_deactivate_hint = TRUE;
void
vm_set_page_size(void)
{
page_mask = page_size - 1;
if ((page_mask & page_size) != 0)
panic("vm_set_page_size: page size not a power of two");
for (page_shift = 0; ; page_shift++)
if ((1U << page_shift) == page_size)
break;
}
static void
vm_page_set_colors( void )
{
unsigned int n, override;
if ( PE_parse_boot_arg("colors", &override) )
n = override;
else if ( vm_cache_geometry_colors )
n = vm_cache_geometry_colors;
else n = DEFAULT_COLORS;
if ( n == 0 )
n = 1;
if ( n > MAX_COLORS )
n = MAX_COLORS;
if ( ( n & (n - 1)) !=0 )
panic("vm_page_set_colors");
vm_colors = n;
vm_color_mask = n - 1;
}
void
vm_page_bootstrap(
vm_offset_t *startp,
vm_offset_t *endp)
{
register vm_page_t m;
unsigned int i;
unsigned int log1;
unsigned int log2;
unsigned int size;
m = &vm_page_template;
m->object = VM_OBJECT_NULL;
m->offset = (vm_object_offset_t) -1;
m->wire_count = 0;
m->pageq.next = NULL;
m->pageq.prev = NULL;
m->listq.next = NULL;
m->listq.prev = NULL;
m->speculative = FALSE;
m->throttled = FALSE;
m->inactive = FALSE;
m->active = FALSE;
m->no_cache = FALSE;
m->laundry = FALSE;
m->free = FALSE;
m->pmapped = FALSE;
m->wpmapped = FALSE;
m->reference = FALSE;
m->pageout = FALSE;
m->dump_cleaning = FALSE;
m->list_req_pending = FALSE;
m->busy = TRUE;
m->wanted = FALSE;
m->tabled = FALSE;
m->fictitious = FALSE;
m->private = FALSE;
m->absent = FALSE;
m->error = FALSE;
m->dirty = FALSE;
m->cleaning = FALSE;
m->precious = FALSE;
m->clustered = FALSE;
m->unusual = FALSE;
m->restart = FALSE;
m->zero_fill = FALSE;
m->encrypted = FALSE;
m->encrypted_cleaning = FALSE;
m->deactivated = FALSE;
m->phys_page = 0;
mutex_init(&vm_page_queue_free_lock, 0);
mutex_init(&vm_page_queue_lock, 0);
mutex_init(&vm_purgeable_queue_lock, 0);
for (i = 0; i < PURGEABLE_Q_TYPE_MAX; i++) {
int group;
purgeable_queues[i].token_q_head = 0;
purgeable_queues[i].token_q_tail = 0;
for (group = 0; group < NUM_VOLATILE_GROUPS; group++)
queue_init(&purgeable_queues[i].objq[group]);
purgeable_queues[i].type = i;
purgeable_queues[i].new_pages = 0;
#if MACH_ASSERT
purgeable_queues[i].debug_count_tokens = 0;
purgeable_queues[i].debug_count_objects = 0;
#endif
};
for (i = 0; i < MAX_COLORS; i++ )
queue_init(&vm_page_queue_free[i]);
queue_init(&vm_lopage_queue_free);
vm_page_queue_fictitious = VM_PAGE_NULL;
queue_init(&vm_page_queue_active);
queue_init(&vm_page_queue_inactive);
queue_init(&vm_page_queue_throttled);
queue_init(&vm_page_queue_zf);
for ( i = 0; i <= VM_PAGE_MAX_SPECULATIVE_AGE_Q; i++ ) {
queue_init(&vm_page_queue_speculative[i].age_q);
vm_page_queue_speculative[i].age_ts.tv_sec = 0;
vm_page_queue_speculative[i].age_ts.tv_nsec = 0;
}
vm_page_free_wanted = 0;
vm_page_free_wanted_privileged = 0;
vm_page_set_colors();
vm_map_steal_memory();
zone_steal_memory();
simple_lock_init(&vm_page_bucket_lock, 0);
if (vm_page_bucket_count == 0) {
unsigned int npages = pmap_free_pages();
vm_page_bucket_count = 1;
while (vm_page_bucket_count < npages)
vm_page_bucket_count <<= 1;
}
vm_page_hash_mask = vm_page_bucket_count - 1;
size = vm_page_bucket_count;
for (log1 = 0; size > 1; log1++)
size /= 2;
size = sizeof(struct vm_object);
for (log2 = 0; size > 1; log2++)
size /= 2;
vm_page_hash_shift = log1/2 - log2 + 1;
vm_page_bucket_hash = 1 << ((log1 + 1) >> 1);
vm_page_bucket_hash |= 1 << ((log1 + 1) >> 2);
vm_page_bucket_hash |= 1;
if (vm_page_hash_mask & vm_page_bucket_count)
printf("vm_page_bootstrap: WARNING -- strange page hash\n");
vm_page_buckets = (vm_page_bucket_t *)
pmap_steal_memory(vm_page_bucket_count *
sizeof(vm_page_bucket_t));
for (i = 0; i < vm_page_bucket_count; i++) {
register vm_page_bucket_t *bucket = &vm_page_buckets[i];
bucket->pages = VM_PAGE_NULL;
#if MACH_PAGE_HASH_STATS
bucket->cur_count = 0;
bucket->hi_count = 0;
#endif
}
pmap_startup(&virtual_space_start, &virtual_space_end);
virtual_space_start = round_page(virtual_space_start);
virtual_space_end = trunc_page(virtual_space_end);
*startp = virtual_space_start;
*endp = virtual_space_end;
vm_page_wire_count = atop_64(max_mem) - vm_page_free_count;
vm_page_free_count_minimum = vm_page_free_count;
printf("vm_page_bootstrap: %d free pages and %d wired pages\n",
vm_page_free_count, vm_page_wire_count);
simple_lock_init(&vm_paging_lock, 0);
}
#ifndef MACHINE_PAGES
void *
pmap_steal_memory(
vm_size_t size)
{
vm_offset_t addr, vaddr;
ppnum_t phys_page;
size = (size + sizeof (void *) - 1) &~ (sizeof (void *) - 1);
if (virtual_space_start == virtual_space_end) {
pmap_virtual_space(&virtual_space_start, &virtual_space_end);
virtual_space_start = round_page(virtual_space_start);
virtual_space_end = trunc_page(virtual_space_end);
}
addr = virtual_space_start;
virtual_space_start += size;
kprintf("pmap_steal_memory: %08X - %08X; size=%08X\n", addr, virtual_space_start, size);
for (vaddr = round_page(addr);
vaddr < addr + size;
vaddr += PAGE_SIZE) {
if (!pmap_next_page(&phys_page))
panic("pmap_steal_memory");
pmap_enter(kernel_pmap, vaddr, phys_page,
VM_PROT_READ|VM_PROT_WRITE,
VM_WIMG_USE_DEFAULT, FALSE);
vm_page_wire_count++;
}
return (void *) addr;
}
void
pmap_startup(
vm_offset_t *startp,
vm_offset_t *endp)
{
unsigned int i, npages, pages_initialized, fill, fillval;
ppnum_t phys_page;
addr64_t tmpaddr;
unsigned int num_of_lopages = 0;
unsigned int last_index;
tmpaddr = (addr64_t)pmap_free_pages() * (addr64_t)PAGE_SIZE;
tmpaddr = tmpaddr + (addr64_t)(round_page_32(virtual_space_start) - virtual_space_start);
npages = (unsigned int)(tmpaddr / (addr64_t)(PAGE_SIZE + sizeof(*vm_pages)));
vm_pages = (vm_page_t) pmap_steal_memory(npages * sizeof *vm_pages);
for (i = 0, pages_initialized = 0; i < npages; i++) {
if (!pmap_next_page(&phys_page))
break;
vm_page_init(&vm_pages[i], phys_page);
vm_page_pages++;
pages_initialized++;
}
vm_pages_count = pages_initialized;
fill = 0;
if (PE_parse_boot_arg("fill", &fillval)) fill = 1;
for (i = 0, num_of_lopages = vm_lopage_poolsize; num_of_lopages && i < pages_initialized; num_of_lopages--, i++) {
vm_page_t m;
m = &vm_pages[i];
if (m->phys_page >= (1 << (32 - PAGE_SHIFT)))
panic("couldn't reserve the lopage pool: not enough lo pages\n");
if (m->phys_page < vm_lopage_poolend)
panic("couldn't reserve the lopage pool: page list out of order\n");
vm_lopage_poolend = m->phys_page;
if (vm_lopage_poolstart == 0)
vm_lopage_poolstart = m->phys_page;
else {
if (m->phys_page < vm_lopage_poolstart)
panic("couldn't reserve the lopage pool: page list out of order\n");
}
if (fill)
fillPage(m->phys_page, fillval);
vm_page_release(m);
}
last_index = i;
if (2 == vm_himemory_mode) {
for (i = last_index + 1; i <= pages_initialized; i++) {
if(fill) fillPage(vm_pages[i - 1].phys_page, fillval);
vm_page_release(&vm_pages[i - 1]);
}
}
else
for (i = pages_initialized; i > last_index; i--) {
if(fill) fillPage(vm_pages[i - 1].phys_page, fillval);
vm_page_release(&vm_pages[i - 1]);
}
#if 0
{
vm_page_t xx, xxo, xxl;
int i, j, k, l;
j = 0;
xxl = 0;
for( i = 0; i < vm_colors; i++ ) {
queue_iterate(&vm_page_queue_free[i],
xx,
vm_page_t,
pageq) {
j++;
if(j > vm_page_free_count) {
panic("pmap_startup: too many pages, xx = %08X, xxl = %08X\n", xx, xxl);
}
l = vm_page_free_count - j;
k = 0;
if(((j - 1) & 0xFFFF) == 0) kprintf("checking number %d of %d\n", j, vm_page_free_count);
for(xxo = xx->pageq.next; xxo != &vm_page_queue_free[i]; xxo = xxo->pageq.next) {
k++;
if(k > l) panic("pmap_startup: too many in secondary check %d %d\n", k, l);
if((xx->phys_page & 0xFFFFFFFF) == (xxo->phys_page & 0xFFFFFFFF)) {
panic("pmap_startup: duplicate physaddr, xx = %08X, xxo = %08X\n", xx, xxo);
}
}
xxl = xx;
}
}
if(j != vm_page_free_count) {
panic("pmap_startup: vm_page_free_count does not match, calc = %d, vm_page_free_count = %08X\n", j, vm_page_free_count);
}
}
#endif
virtual_space_start = round_page_32(virtual_space_start);
*startp = virtual_space_start;
*endp = virtual_space_end;
}
#endif
void
vm_page_module_init(void)
{
vm_page_zone = zinit((vm_size_t) sizeof(struct vm_page),
0, PAGE_SIZE, "vm pages");
#if ZONE_DEBUG
zone_debug_disable(vm_page_zone);
#endif
zone_change(vm_page_zone, Z_EXPAND, FALSE);
zone_change(vm_page_zone, Z_EXHAUST, TRUE);
zone_change(vm_page_zone, Z_FOREIGN, TRUE);
vm_page_zone->count += vm_page_pages;
vm_page_zone->cur_size += vm_page_pages * vm_page_zone->elem_size;
mutex_init(&vm_page_alloc_lock, 0);
}
void
vm_page_create(
ppnum_t start,
ppnum_t end)
{
ppnum_t phys_page;
vm_page_t m;
for (phys_page = start;
phys_page < end;
phys_page++) {
while ((m = (vm_page_t) vm_page_grab_fictitious())
== VM_PAGE_NULL)
vm_page_more_fictitious();
vm_page_init(m, phys_page);
vm_page_pages++;
vm_page_release(m);
}
}
#define vm_page_hash(object, offset) (\
( (natural_t)((uint32_t)object * vm_page_bucket_hash) + ((uint32_t)atop_64(offset) ^ vm_page_bucket_hash))\
& vm_page_hash_mask)
void
vm_page_insert(
vm_page_t mem,
vm_object_t object,
vm_object_offset_t offset)
{
vm_page_insert_internal(mem, object, offset, FALSE);
}
void
vm_page_insert_internal(
vm_page_t mem,
vm_object_t object,
vm_object_offset_t offset,
boolean_t queues_lock_held)
{
register vm_page_bucket_t *bucket;
XPR(XPR_VM_PAGE,
"vm_page_insert, object 0x%X offset 0x%X page 0x%X\n",
(integer_t)object, (integer_t)offset, (integer_t)mem, 0,0);
VM_PAGE_CHECK(mem);
if (object == vm_submap_object) {
panic("vm_page_insert(vm_submap_object,0x%llx)\n", offset);
}
vm_object_lock_assert_exclusive(object);
#if DEBUG
if (mem->tabled || mem->object != VM_OBJECT_NULL)
panic("vm_page_insert: page %p for (obj=%p,off=0x%llx) "
"already in (obj=%p,off=0x%llx)",
mem, object, offset, mem->object, mem->offset);
#endif
assert(!object->internal || offset < object->size);
assert(object->pageout == mem->pageout);
assert(vm_page_lookup(object, offset) == VM_PAGE_NULL);
mem->object = object;
mem->offset = offset;
bucket = &vm_page_buckets[vm_page_hash(object, offset)];
simple_lock(&vm_page_bucket_lock);
mem->next = bucket->pages;
bucket->pages = mem;
#if MACH_PAGE_HASH_STATS
if (++bucket->cur_count > bucket->hi_count)
bucket->hi_count = bucket->cur_count;
#endif
simple_unlock(&vm_page_bucket_lock);
VM_PAGE_INSERT(mem, object);
mem->tabled = TRUE;
object->resident_page_count++;
if (object->purgable == VM_PURGABLE_VOLATILE ||
object->purgable == VM_PURGABLE_EMPTY) {
if (queues_lock_held == FALSE)
vm_page_lockspin_queues();
vm_page_purgeable_count++;
if (queues_lock_held == FALSE)
vm_page_unlock_queues();
}
}
void
vm_page_replace(
register vm_page_t mem,
register vm_object_t object,
register vm_object_offset_t offset)
{
vm_page_bucket_t *bucket;
vm_page_t found_m = VM_PAGE_NULL;
VM_PAGE_CHECK(mem);
vm_object_lock_assert_exclusive(object);
#if DEBUG
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
if (mem->tabled || mem->object != VM_OBJECT_NULL)
panic("vm_page_replace: page %p for (obj=%p,off=0x%llx) "
"already in (obj=%p,off=0x%llx)",
mem, object, offset, mem->object, mem->offset);
#endif
mem->object = object;
mem->offset = offset;
bucket = &vm_page_buckets[vm_page_hash(object, offset)];
simple_lock(&vm_page_bucket_lock);
if (bucket->pages) {
vm_page_t *mp = &bucket->pages;
register vm_page_t m = *mp;
do {
if (m->object == object && m->offset == offset) {
*mp = m->next;
found_m = m;
break;
}
mp = &m->next;
} while ((m = *mp));
mem->next = bucket->pages;
} else {
mem->next = VM_PAGE_NULL;
}
bucket->pages = mem;
simple_unlock(&vm_page_bucket_lock);
if (found_m) {
VM_PAGE_REMOVE(found_m);
found_m->tabled = FALSE;
found_m->object = VM_OBJECT_NULL;
found_m->offset = (vm_object_offset_t) -1;
object->resident_page_count--;
if (object->purgable == VM_PURGABLE_VOLATILE ||
object->purgable == VM_PURGABLE_EMPTY) {
assert(vm_page_purgeable_count > 0);
vm_page_purgeable_count--;
}
vm_page_free(found_m);
}
VM_PAGE_INSERT(mem, object);
mem->tabled = TRUE;
object->resident_page_count++;
if (object->purgable == VM_PURGABLE_VOLATILE ||
object->purgable == VM_PURGABLE_EMPTY) {
vm_page_purgeable_count++;
}
}
void
vm_page_remove(
register vm_page_t mem)
{
register vm_page_bucket_t *bucket;
register vm_page_t this;
XPR(XPR_VM_PAGE,
"vm_page_remove, object 0x%X offset 0x%X page 0x%X\n",
(integer_t)mem->object, (integer_t)mem->offset,
(integer_t)mem, 0,0);
#if DEBUG
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
vm_object_lock_assert_exclusive(mem->object);
assert(mem->tabled);
assert(!mem->cleaning);
VM_PAGE_CHECK(mem);
bucket = &vm_page_buckets[vm_page_hash(mem->object, mem->offset)];
simple_lock(&vm_page_bucket_lock);
if ((this = bucket->pages) == mem) {
bucket->pages = mem->next;
} else {
register vm_page_t *prev;
for (prev = &this->next;
(this = *prev) != mem;
prev = &this->next)
continue;
*prev = this->next;
}
#if MACH_PAGE_HASH_STATS
bucket->cur_count--;
#endif
simple_unlock(&vm_page_bucket_lock);
VM_PAGE_REMOVE(mem);
mem->object->resident_page_count--;
if (mem->object->purgable == VM_PURGABLE_VOLATILE ||
mem->object->purgable == VM_PURGABLE_EMPTY) {
assert(vm_page_purgeable_count > 0);
vm_page_purgeable_count--;
}
mem->tabled = FALSE;
mem->object = VM_OBJECT_NULL;
mem->offset = (vm_object_offset_t) -1;
}
unsigned long vm_page_lookup_hint = 0;
unsigned long vm_page_lookup_hint_next = 0;
unsigned long vm_page_lookup_hint_prev = 0;
unsigned long vm_page_lookup_hint_miss = 0;
unsigned long vm_page_lookup_bucket_NULL = 0;
unsigned long vm_page_lookup_miss = 0;
vm_page_t
vm_page_lookup(
register vm_object_t object,
register vm_object_offset_t offset)
{
register vm_page_t mem;
register vm_page_bucket_t *bucket;
queue_entry_t qe;
vm_object_lock_assert_held(object);
mem = object->memq_hint;
if (mem != VM_PAGE_NULL) {
assert(mem->object == object);
if (mem->offset == offset) {
vm_page_lookup_hint++;
return mem;
}
qe = queue_next(&mem->listq);
if (! queue_end(&object->memq, qe)) {
vm_page_t next_page;
next_page = (vm_page_t) qe;
assert(next_page->object == object);
if (next_page->offset == offset) {
vm_page_lookup_hint_next++;
object->memq_hint = next_page;
return next_page;
}
}
qe = queue_prev(&mem->listq);
if (! queue_end(&object->memq, qe)) {
vm_page_t prev_page;
prev_page = (vm_page_t) qe;
assert(prev_page->object == object);
if (prev_page->offset == offset) {
vm_page_lookup_hint_prev++;
object->memq_hint = prev_page;
return prev_page;
}
}
}
bucket = &vm_page_buckets[vm_page_hash(object, offset)];
if (bucket->pages == VM_PAGE_NULL) {
vm_page_lookup_bucket_NULL++;
return (VM_PAGE_NULL);
}
simple_lock(&vm_page_bucket_lock);
for (mem = bucket->pages; mem != VM_PAGE_NULL; mem = mem->next) {
VM_PAGE_CHECK(mem);
if ((mem->object == object) && (mem->offset == offset))
break;
}
simple_unlock(&vm_page_bucket_lock);
if (mem != VM_PAGE_NULL) {
if (object->memq_hint != VM_PAGE_NULL) {
vm_page_lookup_hint_miss++;
}
assert(mem->object == object);
object->memq_hint = mem;
} else
vm_page_lookup_miss++;
return(mem);
}
void
vm_page_rename(
register vm_page_t mem,
register vm_object_t new_object,
vm_object_offset_t new_offset,
boolean_t encrypted_ok)
{
assert(mem->object != new_object);
if (!encrypted_ok && mem->encrypted) {
panic("vm_page_rename: page %p is encrypted\n", mem);
}
XPR(XPR_VM_PAGE,
"vm_page_rename, new object 0x%X, offset 0x%X page 0x%X\n",
(integer_t)new_object, (integer_t)new_offset,
(integer_t)mem, 0,0);
vm_page_lockspin_queues();
vm_page_remove(mem);
vm_page_insert(mem, new_object, new_offset);
vm_page_unlock_queues();
}
void
vm_page_init(
vm_page_t mem,
ppnum_t phys_page)
{
assert(phys_page);
*mem = vm_page_template;
mem->phys_page = phys_page;
}
int c_vm_page_grab_fictitious = 0;
int c_vm_page_release_fictitious = 0;
int c_vm_page_more_fictitious = 0;
extern vm_page_t vm_page_grab_fictitious_common(vm_offset_t phys_addr);
vm_page_t
vm_page_grab_fictitious_common(
vm_offset_t phys_addr)
{
register vm_page_t m;
m = (vm_page_t)zget(vm_page_zone);
if (m) {
vm_page_init(m, phys_addr);
m->fictitious = TRUE;
}
c_vm_page_grab_fictitious++;
return m;
}
vm_page_t
vm_page_grab_fictitious(void)
{
return vm_page_grab_fictitious_common(vm_page_fictitious_addr);
}
vm_page_t
vm_page_grab_guard(void)
{
return vm_page_grab_fictitious_common(vm_page_guard_addr);
}
void
vm_page_release_fictitious(
register vm_page_t m)
{
assert(!m->free);
assert(m->busy);
assert(m->fictitious);
assert(m->phys_page == vm_page_fictitious_addr ||
m->phys_page == vm_page_guard_addr);
c_vm_page_release_fictitious++;
#if DEBUG
if (m->free)
panic("vm_page_release_fictitious");
#endif
m->free = TRUE;
zfree(vm_page_zone, m);
}
void vm_page_more_fictitious(void)
{
register vm_page_t m;
vm_offset_t addr;
kern_return_t retval;
int i;
c_vm_page_more_fictitious++;
mutex_lock(&vm_page_alloc_lock);
if (zone_free_count(vm_page_zone) > 5) {
mutex_unlock(&vm_page_alloc_lock);
return;
}
retval = kernel_memory_allocate(zone_map,
&addr, PAGE_SIZE, VM_PROT_ALL,
KMA_KOBJECT|KMA_NOPAGEWAIT);
if (retval != KERN_SUCCESS) {
mutex_unlock(&vm_page_alloc_lock);
vm_page_wait(THREAD_UNINT);
return;
}
m = (vm_page_t)addr;
for (i = PAGE_SIZE/sizeof(struct vm_page); i > 0; i--) {
vm_page_init(m, vm_page_fictitious_addr);
m->fictitious = TRUE;
m++;
}
zcram(vm_page_zone, (void *) addr, PAGE_SIZE);
mutex_unlock(&vm_page_alloc_lock);
}
int
vm_pool_low(void)
{
return( vm_page_free_count < vm_page_free_reserved );
}
int vm_himemory_mode = 0;
unsigned int vm_lopage_free_count = 0;
unsigned int vm_lopage_max_count = 0;
queue_head_t vm_lopage_queue_free;
vm_page_t
vm_page_grablo(void)
{
register vm_page_t mem;
unsigned int vm_lopage_alloc_count;
if (vm_lopage_poolsize == 0)
return (vm_page_grab());
mutex_lock(&vm_page_queue_free_lock);
if (! queue_empty(&vm_lopage_queue_free)) {
queue_remove_first(&vm_lopage_queue_free,
mem,
vm_page_t,
pageq);
assert(mem->free);
assert(mem->busy);
assert(!mem->pmapped);
assert(!mem->wpmapped);
mem->pageq.next = NULL;
mem->pageq.prev = NULL;
mem->free = FALSE;
vm_lopage_free_count--;
vm_lopage_alloc_count = (vm_lopage_poolend - vm_lopage_poolstart) - vm_lopage_free_count;
if (vm_lopage_alloc_count > vm_lopage_max_count)
vm_lopage_max_count = vm_lopage_alloc_count;
} else {
mem = VM_PAGE_NULL;
}
mutex_unlock(&vm_page_queue_free_lock);
return (mem);
}
#define COLOR_GROUPS_TO_STEAL 4
vm_page_t
vm_page_grab( void )
{
vm_page_t mem;
disable_preemption();
if ((mem = PROCESSOR_DATA(current_processor(), free_pages))) {
return_page_from_cpu_list:
PROCESSOR_DATA(current_processor(), page_grab_count) += 1;
PROCESSOR_DATA(current_processor(), free_pages) = mem->pageq.next;
mem->pageq.next = NULL;
enable_preemption();
assert(mem->listq.next == NULL && mem->listq.prev == NULL);
assert(mem->tabled == FALSE);
assert(mem->object == VM_OBJECT_NULL);
assert(!mem->laundry);
assert(!mem->free);
assert(pmap_verify_free(mem->phys_page));
assert(mem->busy);
assert(!mem->encrypted);
assert(!mem->pmapped);
assert(!mem->wpmapped);
return mem;
}
enable_preemption();
mutex_lock(&vm_page_queue_free_lock);
if (vm_page_wire_count_warning > 0
&& vm_page_wire_count >= vm_page_wire_count_warning) {
printf("mk: vm_page_grab(): high wired page count of %d\n",
vm_page_wire_count);
assert(vm_page_wire_count < vm_page_wire_count_warning);
}
if (vm_page_gobble_count_warning > 0
&& vm_page_gobble_count >= vm_page_gobble_count_warning) {
printf("mk: vm_page_grab(): high gobbled page count of %d\n",
vm_page_gobble_count);
assert(vm_page_gobble_count < vm_page_gobble_count_warning);
}
if ((vm_page_free_count < vm_page_free_reserved) &&
!(current_thread()->options & TH_OPT_VMPRIV)) {
mutex_unlock(&vm_page_queue_free_lock);
mem = VM_PAGE_NULL;
}
else {
vm_page_t head;
vm_page_t tail;
unsigned int pages_to_steal;
unsigned int color;
while ( vm_page_free_count == 0 ) {
mutex_unlock(&vm_page_queue_free_lock);
VM_PAGE_WAIT();
mutex_lock(&vm_page_queue_free_lock);
}
disable_preemption();
if ((mem = PROCESSOR_DATA(current_processor(), free_pages))) {
mutex_unlock(&vm_page_queue_free_lock);
goto return_page_from_cpu_list;
}
if (vm_page_free_count <= vm_page_free_reserved)
pages_to_steal = 1;
else {
pages_to_steal = COLOR_GROUPS_TO_STEAL * vm_colors;
if (pages_to_steal > (vm_page_free_count - vm_page_free_reserved))
pages_to_steal = (vm_page_free_count - vm_page_free_reserved);
}
color = PROCESSOR_DATA(current_processor(), start_color);
head = tail = NULL;
while (pages_to_steal--) {
if (--vm_page_free_count < vm_page_free_count_minimum)
vm_page_free_count_minimum = vm_page_free_count;
while (queue_empty(&vm_page_queue_free[color]))
color = (color + 1) & vm_color_mask;
queue_remove_first(&vm_page_queue_free[color],
mem,
vm_page_t,
pageq);
mem->pageq.next = NULL;
mem->pageq.prev = NULL;
color = (color + 1) & vm_color_mask;
if (head == NULL)
head = mem;
else
tail->pageq.next = (queue_t)mem;
tail = mem;
mem->pageq.prev = NULL;
assert(mem->listq.next == NULL && mem->listq.prev == NULL);
assert(mem->tabled == FALSE);
assert(mem->object == VM_OBJECT_NULL);
assert(!mem->laundry);
assert(mem->free);
mem->free = FALSE;
assert(pmap_verify_free(mem->phys_page));
assert(mem->busy);
assert(!mem->free);
assert(!mem->encrypted);
assert(!mem->pmapped);
assert(!mem->wpmapped);
}
PROCESSOR_DATA(current_processor(), free_pages) = head->pageq.next;
PROCESSOR_DATA(current_processor(), start_color) = color;
PROCESSOR_DATA(current_processor(), page_grab_count) += 1;
mem = head;
mem->pageq.next = NULL;
mutex_unlock(&vm_page_queue_free_lock);
enable_preemption();
}
if ((vm_page_free_count < vm_page_free_min) ||
((vm_page_free_count < vm_page_free_target) &&
((vm_page_inactive_count + vm_page_speculative_count) < vm_page_inactive_min)))
thread_wakeup((event_t) &vm_page_free_wanted);
#if CONFIG_EMBEDDED
{
int percent_avail;
percent_avail =
(vm_page_active_count + vm_page_inactive_count +
vm_page_speculative_count + vm_page_free_count +
(IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 /
atop_64(max_mem);
if (percent_avail <= (kern_memorystatus_level - 5)) {
kern_memorystatus_level = percent_avail;
thread_wakeup((event_t)&kern_memorystatus_wakeup);
}
}
#endif
return mem;
}
void
vm_page_release(
register vm_page_t mem)
{
unsigned int color;
#if 0
unsigned int pindex;
phys_entry *physent;
physent = mapping_phys_lookup(mem->phys_page, &pindex);
if(physent->ppLink & ppN) {
panic("vm_page_release: already released - %08X %08X\n", mem, mem->phys_page);
}
physent->ppLink = physent->ppLink | ppN;
#endif
assert(!mem->private && !mem->fictitious);
mutex_lock(&vm_page_queue_free_lock);
#if DEBUG
if (mem->free)
panic("vm_page_release");
#endif
mem->free = TRUE;
assert(mem->busy);
assert(!mem->laundry);
assert(mem->object == VM_OBJECT_NULL);
assert(mem->pageq.next == NULL &&
mem->pageq.prev == NULL);
assert(mem->listq.next == NULL &&
mem->listq.prev == NULL);
if (mem->phys_page <= vm_lopage_poolend && mem->phys_page >= vm_lopage_poolstart) {
queue_enter_first(&vm_lopage_queue_free,
mem,
vm_page_t,
pageq);
vm_lopage_free_count++;
} else {
color = mem->phys_page & vm_color_mask;
queue_enter_first(&vm_page_queue_free[color],
mem,
vm_page_t,
pageq);
vm_page_free_count++;
if ((vm_page_free_wanted_privileged > 0) && vm_page_free_count) {
vm_page_free_wanted_privileged--;
thread_wakeup_one((event_t) &vm_page_free_wanted_privileged);
} else if ((vm_page_free_wanted > 0) &&
(vm_page_free_count >= vm_page_free_reserved)) {
vm_page_free_wanted--;
thread_wakeup_one((event_t) &vm_page_free_count);
}
}
mutex_unlock(&vm_page_queue_free_lock);
#if CONFIG_EMBEDDED
{
int percent_avail;
percent_avail =
(vm_page_active_count + vm_page_inactive_count +
vm_page_speculative_count + vm_page_free_count +
(IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 /
atop_64(max_mem);
if (percent_avail >= (kern_memorystatus_level + 5)) {
kern_memorystatus_level = percent_avail;
thread_wakeup((event_t)&kern_memorystatus_wakeup);
}
}
#endif
}
boolean_t
vm_page_wait(
int interruptible )
{
kern_return_t wait_result;
int need_wakeup = 0;
int is_privileged = current_thread()->options & TH_OPT_VMPRIV;
mutex_lock(&vm_page_queue_free_lock);
if (is_privileged && vm_page_free_count) {
mutex_unlock(&vm_page_queue_free_lock);
return TRUE;
}
if (vm_page_free_count < vm_page_free_target) {
if (is_privileged) {
if (vm_page_free_wanted_privileged++ == 0)
need_wakeup = 1;
wait_result = assert_wait((event_t)&vm_page_free_wanted_privileged, interruptible);
} else {
if (vm_page_free_wanted++ == 0)
need_wakeup = 1;
wait_result = assert_wait((event_t)&vm_page_free_count, interruptible);
}
mutex_unlock(&vm_page_queue_free_lock);
counter(c_vm_page_wait_block++);
if (need_wakeup)
thread_wakeup((event_t)&vm_page_free_wanted);
if (wait_result == THREAD_WAITING)
wait_result = thread_block(THREAD_CONTINUE_NULL);
return(wait_result == THREAD_AWAKENED);
} else {
mutex_unlock(&vm_page_queue_free_lock);
return TRUE;
}
}
vm_page_t
vm_page_alloc(
vm_object_t object,
vm_object_offset_t offset)
{
register vm_page_t mem;
vm_object_lock_assert_exclusive(object);
mem = vm_page_grab();
if (mem == VM_PAGE_NULL)
return VM_PAGE_NULL;
vm_page_insert(mem, object, offset);
return(mem);
}
vm_page_t
vm_page_alloclo(
vm_object_t object,
vm_object_offset_t offset)
{
register vm_page_t mem;
vm_object_lock_assert_exclusive(object);
mem = vm_page_grablo();
if (mem == VM_PAGE_NULL)
return VM_PAGE_NULL;
vm_page_insert(mem, object, offset);
return(mem);
}
vm_page_t
vm_page_alloc_guard(
vm_object_t object,
vm_object_offset_t offset)
{
register vm_page_t mem;
vm_object_lock_assert_exclusive(object);
mem = vm_page_grab_guard();
if (mem == VM_PAGE_NULL)
return VM_PAGE_NULL;
vm_page_insert(mem, object, offset);
return(mem);
}
counter(unsigned int c_laundry_pages_freed = 0;)
boolean_t vm_page_free_verify = TRUE;
void
vm_page_free_prepare(
register vm_page_t mem)
{
VM_PAGE_CHECK(mem);
assert(!mem->free);
assert(!mem->cleaning);
assert(!mem->pageout);
#if DEBUG
if (vm_page_free_verify && !mem->fictitious && !mem->private) {
assert(pmap_verify_free(mem->phys_page));
}
if (mem->object)
vm_object_lock_assert_exclusive(mem->object);
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
if (mem->free)
panic("vm_page_free: freeing page on free list\n");
#endif
if (mem->laundry) {
vm_pageout_throttle_up(mem);
counter(++c_laundry_pages_freed);
}
if (mem->tabled)
vm_page_remove(mem);
VM_PAGE_QUEUES_REMOVE(mem);
if (mem->wire_count) {
if (!mem->private && !mem->fictitious)
vm_page_wire_count--;
mem->wire_count = 0;
assert(!mem->gobbled);
} else if (mem->gobbled) {
if (!mem->private && !mem->fictitious)
vm_page_wire_count--;
vm_page_gobble_count--;
}
mem->gobbled = FALSE;
PAGE_WAKEUP(mem);
mem->busy = TRUE;
mem->absent = FALSE;
mem->error = FALSE;
mem->dirty = FALSE;
mem->precious = FALSE;
mem->reference = FALSE;
mem->encrypted = FALSE;
mem->encrypted_cleaning = FALSE;
mem->deactivated = FALSE;
mem->pmapped = FALSE;
mem->wpmapped = FALSE;
if (mem->private) {
mem->private = FALSE;
mem->fictitious = TRUE;
mem->phys_page = vm_page_fictitious_addr;
}
if (!mem->fictitious) {
if (mem->zero_fill == TRUE) {
mem->zero_fill = FALSE;
OSAddAtomic(-1, (SInt32 *)&vm_zf_count);
}
vm_page_init(mem, mem->phys_page);
}
}
void
vm_page_free(
vm_page_t mem)
{
vm_page_free_prepare(mem);
if (mem->fictitious) {
vm_page_release_fictitious(mem);
} else {
vm_page_release(mem);
}
}
void
vm_page_free_list(
vm_page_t mem)
{
vm_page_t nxt;
int pg_count = 0;
int color;
int inuse_list_head = -1;
queue_head_t free_list[MAX_COLORS];
int inuse[MAX_COLORS];
for (color = 0; color < (signed) vm_colors; color++) {
queue_init(&free_list[color]);
}
#if DEBUG
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
while (mem) {
#if DEBUG
if (mem->tabled || mem->object)
panic("vm_page_free_list: freeing tabled page\n");
if (mem->inactive || mem->active || mem->throttled || mem->free)
panic("vm_page_free_list: freeing page on list\n");
if (vm_page_free_verify && !mem->fictitious && !mem->private) {
assert(pmap_verify_free(mem->phys_page));
}
#endif
assert(mem->pageq.prev == NULL);
assert(mem->busy);
assert(!mem->free);
nxt = (vm_page_t)(mem->pageq.next);
if (!mem->fictitious) {
mem->free = TRUE;
color = mem->phys_page & vm_color_mask;
if (queue_empty(&free_list[color])) {
inuse[color] = inuse_list_head;
inuse_list_head = color;
}
queue_enter_first(&free_list[color],
mem,
vm_page_t,
pageq);
pg_count++;
} else {
assert(mem->phys_page == vm_page_fictitious_addr ||
mem->phys_page == vm_page_guard_addr);
vm_page_release_fictitious(mem);
}
mem = nxt;
}
if (pg_count) {
unsigned int avail_free_count;
mutex_lock(&vm_page_queue_free_lock);
color = inuse_list_head;
while( color != -1 ) {
vm_page_t first, last;
vm_page_t first_free;
first = (vm_page_t) queue_first(&free_list[color]);
last = (vm_page_t) queue_last(&free_list[color]);
first_free = (vm_page_t) queue_first(&vm_page_queue_free[color]);
if (queue_empty(&vm_page_queue_free[color])) {
queue_last(&vm_page_queue_free[color]) =
(queue_entry_t) last;
} else {
queue_prev(&first_free->pageq) =
(queue_entry_t) last;
}
queue_first(&vm_page_queue_free[color]) =
(queue_entry_t) first;
queue_prev(&first->pageq) =
(queue_entry_t) &vm_page_queue_free[color];
queue_next(&last->pageq) =
(queue_entry_t) first_free;
color = inuse[color];
}
vm_page_free_count += pg_count;
avail_free_count = vm_page_free_count;
while ((vm_page_free_wanted_privileged > 0) && avail_free_count) {
vm_page_free_wanted_privileged--;
avail_free_count--;
thread_wakeup_one((event_t) &vm_page_free_wanted_privileged);
}
if ((vm_page_free_wanted > 0) &&
(avail_free_count >= vm_page_free_reserved)) {
unsigned int available_pages;
if (avail_free_count >= vm_page_free_reserved) {
available_pages = (avail_free_count - vm_page_free_reserved);
} else {
available_pages = 0;
}
if (available_pages >= vm_page_free_wanted) {
vm_page_free_wanted = 0;
thread_wakeup((event_t) &vm_page_free_count);
} else {
while (available_pages--) {
vm_page_free_wanted--;
thread_wakeup_one((event_t) &vm_page_free_count);
}
}
}
mutex_unlock(&vm_page_queue_free_lock);
#if CONFIG_EMBEDDED
{
int percent_avail;
percent_avail =
(vm_page_active_count + vm_page_inactive_count +
vm_page_speculative_count + vm_page_free_count +
(IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 /
atop_64(max_mem);
if (percent_avail >= (kern_memorystatus_level + 5)) {
kern_memorystatus_level = percent_avail;
thread_wakeup((event_t)&kern_memorystatus_wakeup);
}
}
#endif
}
}
void
vm_page_wire(
register vm_page_t mem)
{
VM_PAGE_CHECK(mem);
#if DEBUG
if (mem->object)
vm_object_lock_assert_exclusive(mem->object);
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
if (mem->wire_count == 0) {
VM_PAGE_QUEUES_REMOVE(mem);
if (!mem->private && !mem->fictitious && !mem->gobbled)
vm_page_wire_count++;
if (mem->gobbled)
vm_page_gobble_count--;
mem->gobbled = FALSE;
if (mem->zero_fill == TRUE) {
mem->zero_fill = FALSE;
OSAddAtomic(-1, (SInt32 *)&vm_zf_count);
}
}
assert(!mem->gobbled);
mem->wire_count++;
}
void
vm_page_gobble(
register vm_page_t mem)
{
vm_page_lockspin_queues();
VM_PAGE_CHECK(mem);
assert(!mem->gobbled);
assert(mem->wire_count == 0);
if (!mem->gobbled && mem->wire_count == 0) {
if (!mem->private && !mem->fictitious)
vm_page_wire_count++;
}
vm_page_gobble_count++;
mem->gobbled = TRUE;
vm_page_unlock_queues();
}
void
vm_page_unwire(
register vm_page_t mem)
{
VM_PAGE_CHECK(mem);
assert(mem->wire_count > 0);
#if DEBUG
if (mem->object)
vm_object_lock_assert_exclusive(mem->object);
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
if (--mem->wire_count == 0) {
assert(!mem->private && !mem->fictitious);
vm_page_wire_count--;
assert(!mem->laundry);
assert(mem->object != kernel_object);
assert(mem->pageq.next == NULL && mem->pageq.prev == NULL);
if (!IP_VALID(memory_manager_default) &&
mem->dirty && mem->object->internal &&
(mem->object->purgable == VM_PURGABLE_DENY ||
mem->object->purgable == VM_PURGABLE_NONVOLATILE ||
mem->object->purgable == VM_PURGABLE_VOLATILE)) {
queue_enter(&vm_page_queue_throttled, mem, vm_page_t, pageq);
vm_page_throttled_count++;
mem->throttled = TRUE;
} else {
queue_enter(&vm_page_queue_active, mem, vm_page_t, pageq);
vm_page_active_count++;
mem->active = TRUE;
}
mem->reference = TRUE;
}
}
void
vm_page_deactivate(
register vm_page_t m)
{
boolean_t rapid_age = FALSE;
VM_PAGE_CHECK(m);
assert(m->object != kernel_object);
assert(m->phys_page != vm_page_guard_addr);
#if DEBUG
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
if (m->gobbled) {
assert(m->wire_count == 0);
if (!m->private && !m->fictitious)
vm_page_wire_count--;
vm_page_gobble_count--;
m->gobbled = FALSE;
}
if (m->private || (m->wire_count != 0))
return;
if (m->active && m->deactivated == TRUE) {
if (!pmap_is_referenced(m->phys_page))
rapid_age = TRUE;
}
if (rapid_age == FALSE && !m->fictitious && !m->absent)
pmap_clear_reference(m->phys_page);
m->reference = FALSE;
m->deactivated = FALSE;
m->no_cache = FALSE;
if (!m->inactive) {
VM_PAGE_QUEUES_REMOVE(m);
assert(!m->laundry);
assert(m->pageq.next == NULL && m->pageq.prev == NULL);
if (!IP_VALID(memory_manager_default) &&
m->dirty && m->object->internal &&
(m->object->purgable == VM_PURGABLE_DENY ||
m->object->purgable == VM_PURGABLE_NONVOLATILE ||
m->object->purgable == VM_PURGABLE_VOLATILE )) {
queue_enter(&vm_page_queue_throttled, m, vm_page_t, pageq);
m->throttled = TRUE;
vm_page_throttled_count++;
} else {
if (rapid_age == TRUE ||
(!m->fictitious && m->object->named && m->object->ref_count == 1)) {
vm_page_speculate(m, FALSE);
vm_page_speculative_recreated++;
return;
} else {
if (m->zero_fill) {
queue_enter(&vm_page_queue_zf, m, vm_page_t, pageq);
vm_zf_queue_count++;
} else {
queue_enter(&vm_page_queue_inactive, m, vm_page_t, pageq);
}
}
m->inactive = TRUE;
if (!m->fictitious) {
vm_page_inactive_count++;
token_new_pagecount++;
}
}
}
}
void
vm_page_activate(
register vm_page_t m)
{
VM_PAGE_CHECK(m);
#ifdef FIXME_4778297
assert(m->object != kernel_object);
#endif
assert(m->phys_page != vm_page_guard_addr);
#if DEBUG
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
if (m->gobbled) {
assert(m->wire_count == 0);
if (!m->private && !m->fictitious)
vm_page_wire_count--;
vm_page_gobble_count--;
m->gobbled = FALSE;
}
if (m->private)
return;
#if DEBUG
if (m->active)
panic("vm_page_activate: already active");
#endif
if (m->speculative) {
DTRACE_VM2(pgrec, int, 1, (uint64_t *), NULL);
DTRACE_VM2(pgfrec, int, 1, (uint64_t *), NULL);
}
VM_PAGE_QUEUES_REMOVE(m);
if (m->wire_count == 0) {
assert(!m->laundry);
assert(m->pageq.next == NULL && m->pageq.prev == NULL);
if (!IP_VALID(memory_manager_default) &&
!m->fictitious && m->dirty && m->object->internal &&
(m->object->purgable == VM_PURGABLE_DENY ||
m->object->purgable == VM_PURGABLE_NONVOLATILE ||
m->object->purgable == VM_PURGABLE_VOLATILE )) {
queue_enter(&vm_page_queue_throttled, m, vm_page_t, pageq);
m->throttled = TRUE;
vm_page_throttled_count++;
} else {
queue_enter(&vm_page_queue_active, m, vm_page_t, pageq);
m->active = TRUE;
if (!m->fictitious)
vm_page_active_count++;
}
m->reference = TRUE;
m->no_cache = FALSE;
}
}
void
vm_page_speculate(
vm_page_t m,
boolean_t new)
{
struct vm_speculative_age_q *aq;
VM_PAGE_CHECK(m);
assert(m->object != kernel_object);
assert(!m->speculative && !m->active && !m->inactive && !m->throttled);
assert(m->phys_page != vm_page_guard_addr);
assert(m->pageq.next == NULL && m->pageq.prev == NULL);
#if DEBUG
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
if (m->wire_count == 0) {
mach_timespec_t ts;
clock_get_system_nanotime(&ts.tv_sec, (unsigned *)&ts.tv_nsec);
if (vm_page_speculative_count == 0) {
speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
aq = &vm_page_queue_speculative[speculative_age_index];
aq->age_ts.tv_sec = VM_PAGE_SPECULATIVE_Q_AGE_MS / 1000;
aq->age_ts.tv_nsec = (VM_PAGE_SPECULATIVE_Q_AGE_MS % 1000) * 1000 * NSEC_PER_USEC;
ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
} else {
aq = &vm_page_queue_speculative[speculative_age_index];
if (CMP_MACH_TIMESPEC(&ts, &aq->age_ts) >= 0) {
speculative_age_index++;
if (speculative_age_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q)
speculative_age_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
if (speculative_age_index == speculative_steal_index) {
speculative_steal_index = speculative_age_index + 1;
if (speculative_steal_index > VM_PAGE_MAX_SPECULATIVE_AGE_Q)
speculative_steal_index = VM_PAGE_MIN_SPECULATIVE_AGE_Q;
}
aq = &vm_page_queue_speculative[speculative_age_index];
if (!queue_empty(&aq->age_q))
vm_page_speculate_ageit(aq);
aq->age_ts.tv_sec = VM_PAGE_SPECULATIVE_Q_AGE_MS / 1000;
aq->age_ts.tv_nsec = (VM_PAGE_SPECULATIVE_Q_AGE_MS % 1000) * 1000 * NSEC_PER_USEC;
ADD_MACH_TIMESPEC(&aq->age_ts, &ts);
}
}
enqueue_tail(&aq->age_q, &m->pageq);
m->speculative = TRUE;
vm_page_speculative_count++;
if (new == TRUE) {
m->object->pages_created++;
vm_page_speculative_created++;
}
}
}
void
vm_page_speculate_ageit(struct vm_speculative_age_q *aq)
{
struct vm_speculative_age_q *sq;
vm_page_t t;
sq = &vm_page_queue_speculative[VM_PAGE_SPECULATIVE_AGED_Q];
if (queue_empty(&sq->age_q)) {
sq->age_q.next = aq->age_q.next;
sq->age_q.prev = aq->age_q.prev;
t = (vm_page_t)sq->age_q.next;
t->pageq.prev = &sq->age_q;
t = (vm_page_t)sq->age_q.prev;
t->pageq.next = &sq->age_q;
} else {
t = (vm_page_t)sq->age_q.prev;
t->pageq.next = aq->age_q.next;
t = (vm_page_t)aq->age_q.next;
t->pageq.prev = sq->age_q.prev;
t = (vm_page_t)aq->age_q.prev;
t->pageq.next = &sq->age_q;
sq->age_q.prev = aq->age_q.prev;
}
queue_init(&aq->age_q);
}
void
vm_page_lru(
vm_page_t m)
{
VM_PAGE_CHECK(m);
assert(m->object != kernel_object);
assert(m->phys_page != vm_page_guard_addr);
#if DEBUG
_mutex_assert(&vm_page_queue_lock, MA_OWNED);
#endif
if (m->active || m->reference)
return;
if (m->private || (m->wire_count != 0))
return;
m->no_cache = FALSE;
VM_PAGE_QUEUES_REMOVE(m);
assert(!m->laundry);
assert(m->pageq.next == NULL && m->pageq.prev == NULL);
queue_enter(&vm_page_queue_inactive, m, vm_page_t, pageq);
m->inactive = TRUE;
vm_page_inactive_count++;
token_new_pagecount++;
}
void
vm_page_part_zero_fill(
vm_page_t m,
vm_offset_t m_pa,
vm_size_t len)
{
vm_page_t tmp;
VM_PAGE_CHECK(m);
#ifdef PMAP_ZERO_PART_PAGE_IMPLEMENTED
pmap_zero_part_page(m->phys_page, m_pa, len);
#else
while (1) {
tmp = vm_page_grab();
if (tmp == VM_PAGE_NULL) {
vm_page_wait(THREAD_UNINT);
continue;
}
break;
}
vm_page_zero_fill(tmp);
if(m_pa != 0) {
vm_page_part_copy(m, 0, tmp, 0, m_pa);
}
if((m_pa + len) < PAGE_SIZE) {
vm_page_part_copy(m, m_pa + len, tmp,
m_pa + len, PAGE_SIZE - (m_pa + len));
}
vm_page_copy(tmp,m);
vm_page_lock_queues();
vm_page_free(tmp);
vm_page_unlock_queues();
#endif
}
void
vm_page_zero_fill(
vm_page_t m)
{
XPR(XPR_VM_PAGE,
"vm_page_zero_fill, object 0x%X offset 0x%X page 0x%X\n",
(integer_t)m->object, (integer_t)m->offset, (integer_t)m, 0,0);
VM_PAGE_CHECK(m);
pmap_zero_page(m->phys_page);
}
void
vm_page_part_copy(
vm_page_t src_m,
vm_offset_t src_pa,
vm_page_t dst_m,
vm_offset_t dst_pa,
vm_size_t len)
{
VM_PAGE_CHECK(src_m);
VM_PAGE_CHECK(dst_m);
pmap_copy_part_page(src_m->phys_page, src_pa,
dst_m->phys_page, dst_pa, len);
}
int vm_page_copy_cs_validations = 0;
int vm_page_copy_cs_tainted = 0;
void
vm_page_copy(
vm_page_t src_m,
vm_page_t dest_m)
{
XPR(XPR_VM_PAGE,
"vm_page_copy, object 0x%X offset 0x%X to object 0x%X offset 0x%X\n",
(integer_t)src_m->object, src_m->offset,
(integer_t)dest_m->object, dest_m->offset,
0);
VM_PAGE_CHECK(src_m);
VM_PAGE_CHECK(dest_m);
if (src_m->encrypted) {
panic("vm_page_copy: source page %p is encrypted\n", src_m);
}
dest_m->encrypted = FALSE;
if (src_m->object != VM_OBJECT_NULL &&
src_m->object->code_signed) {
vm_page_copy_cs_validations++;
vm_page_validate_cs(src_m);
}
dest_m->cs_validated = src_m->cs_validated;
dest_m->cs_tainted = src_m->cs_tainted;
if (dest_m->cs_tainted) {
assert(dest_m->cs_validated);
vm_page_copy_cs_tainted++;
}
pmap_copy_page(src_m->phys_page, dest_m->phys_page);
}
#if MACH_ASSERT
static int
vm_page_verify_contiguous(
vm_page_t pages,
unsigned int npages)
{
register vm_page_t m;
unsigned int page_count;
vm_offset_t prev_addr;
prev_addr = pages->phys_page;
page_count = 1;
for (m = NEXT_PAGE(pages); m != VM_PAGE_NULL; m = NEXT_PAGE(m)) {
if (m->phys_page != prev_addr + 1) {
printf("m %p prev_addr 0x%x, current addr 0x%x\n",
m, prev_addr, m->phys_page);
printf("pages %p page_count %d\n", pages, page_count);
panic("vm_page_verify_contiguous: not contiguous!");
}
prev_addr = m->phys_page;
++page_count;
}
if (page_count != npages) {
printf("pages %p actual count 0x%x but requested 0x%x\n",
pages, page_count, npages);
panic("vm_page_verify_contiguous: count error");
}
return 1;
}
#endif
#if MACH_ASSERT
static void
vm_page_verify_free_lists( void )
{
unsigned int color, npages;
vm_page_t m;
vm_page_t prev_m;
npages = 0;
mutex_lock(&vm_page_queue_free_lock);
for( color = 0; color < vm_colors; color++ ) {
prev_m = (vm_page_t) &vm_page_queue_free[color];
queue_iterate(&vm_page_queue_free[color],
m,
vm_page_t,
pageq) {
if ((vm_page_t) m->pageq.prev != prev_m)
panic("vm_page_verify_free_lists: corrupted prev ptr");
if ( ! m->free )
panic("vm_page_verify_free_lists: not free");
if ( ! m->busy )
panic("vm_page_verify_free_lists: not busy");
if ( (m->phys_page & vm_color_mask) != color)
panic("vm_page_verify_free_lists: wrong color");
++npages;
prev_m = m;
}
}
if (npages != vm_page_free_count)
panic("vm_page_verify_free_lists: npages %u free_count %d",
npages, vm_page_free_count);
mutex_unlock(&vm_page_queue_free_lock);
}
#endif
#define MAX_CONSIDERED_BEFORE_YIELD 1000
#define RESET_STATE_OF_RUN() \
MACRO_BEGIN \
prevcontaddr = -2; \
free_considered = 0; \
substitute_needed = 0; \
npages = 0; \
MACRO_END
static vm_page_t
vm_page_find_contiguous(
unsigned int contig_pages,
ppnum_t max_pnum,
boolean_t wire)
{
vm_page_t m = NULL;
ppnum_t prevcontaddr;
unsigned int npages, considered;
unsigned int page_idx, start_idx;
int free_considered, free_available;
int substitute_needed;
#if MACH_ASSERT
uint32_t tv_start_sec, tv_start_usec, tv_end_sec, tv_end_usec;
int yielded = 0;
int dumped_run = 0;
int stolen_pages = 0;
#endif
if (contig_pages == 0)
return VM_PAGE_NULL;
#if MACH_ASSERT
vm_page_verify_free_lists();
clock_get_system_microtime(&tv_start_sec, &tv_start_usec);
#endif
vm_page_lock_queues();
mutex_lock(&vm_page_queue_free_lock);
RESET_STATE_OF_RUN();
considered = 0;
free_available = vm_page_free_count - vm_page_free_reserved;
for (page_idx = 0, start_idx = 0;
npages < contig_pages && page_idx < vm_pages_count;
page_idx++) {
retry:
m = &vm_pages[page_idx];
if (max_pnum && m->phys_page > max_pnum) {
break;
}
if (m->phys_page <= vm_lopage_poolend &&
m->phys_page >= vm_lopage_poolstart) {
RESET_STATE_OF_RUN();
} else if (m->wire_count || m->gobbled ||
m->encrypted || m->encrypted_cleaning || m->cs_validated || m->cs_tainted ||
m->error || m->absent || m->pageout_queue || m->laundry || m->wanted || m->precious ||
m->cleaning || m->overwriting || m->restart || m->unusual || m->list_req_pending) {
RESET_STATE_OF_RUN();
} else if (!m->free && !m->active && !m->inactive && !m->speculative && !m->throttled) {
RESET_STATE_OF_RUN();
} else if (!m->free && (!m->tabled || m->busy)) {
RESET_STATE_OF_RUN();
} else {
if (m->phys_page != prevcontaddr + 1) {
npages = 1;
start_idx = page_idx;
} else {
npages++;
}
prevcontaddr = m->phys_page;
if (m->pmapped || m->dirty)
substitute_needed++;
if (m->free) {
free_considered++;
}
if ((free_considered + substitute_needed) > free_available) {
RESET_STATE_OF_RUN();
if (free_available && considered <= MAX_CONSIDERED_BEFORE_YIELD) {
considered++;
goto retry;
}
}
}
if (considered > MAX_CONSIDERED_BEFORE_YIELD && npages <= 1) {
mutex_unlock(&vm_page_queue_free_lock);
vm_page_unlock_queues();
mutex_pause(0);
vm_page_lock_queues();
mutex_lock(&vm_page_queue_free_lock);
RESET_STATE_OF_RUN();
free_available = vm_page_free_count - vm_page_free_reserved;
considered = 0;
#if MACH_ASSERT
yielded++;
#endif
goto retry;
}
considered++;
}
m = VM_PAGE_NULL;
if (npages != contig_pages)
mutex_unlock(&vm_page_queue_free_lock);
else {
vm_page_t m1;
vm_page_t m2;
unsigned int cur_idx;
unsigned int tmp_start_idx;
vm_object_t locked_object = VM_OBJECT_NULL;
boolean_t abort_run = FALSE;
tmp_start_idx = start_idx;
while (start_idx < page_idx) {
m1 = &vm_pages[start_idx++];
if (m1->free) {
unsigned int color;
color = m1->phys_page & vm_color_mask;
queue_remove(&vm_page_queue_free[color],
m1,
vm_page_t,
pageq);
vm_page_free_count--;
}
}
if (vm_page_free_count < vm_page_free_count_minimum)
vm_page_free_count_minimum = vm_page_free_count;
mutex_unlock(&vm_page_queue_free_lock);
start_idx = tmp_start_idx;
cur_idx = page_idx - 1;
while (start_idx++ < page_idx) {
m1 = &vm_pages[cur_idx--];
if (m1->free) {
assert(m1->free);
assert(m1->busy);
assert(!m1->wanted);
assert(!m1->laundry);
m1->free = FALSE;
} else {
vm_object_t object;
if (abort_run == TRUE)
continue;
object = m1->object;
if (object != locked_object) {
if (locked_object) {
vm_object_unlock(locked_object);
locked_object = VM_OBJECT_NULL;
}
if (vm_object_lock_try(object))
locked_object = object;
}
if (locked_object == VM_OBJECT_NULL ||
(m1->wire_count || m1->gobbled ||
m1->encrypted || m1->encrypted_cleaning || m1->cs_validated || m1->cs_tainted ||
m1->error || m1->absent || m1->pageout_queue || m1->laundry || m1->wanted || m1->precious ||
m1->cleaning || m1->overwriting || m1->restart || m1->unusual || m1->list_req_pending || m1->busy)) {
if (locked_object) {
vm_object_unlock(locked_object);
locked_object = VM_OBJECT_NULL;
}
tmp_start_idx = cur_idx;
abort_run = TRUE;
continue;
}
if (m1->pmapped || m1->dirty) {
int refmod;
vm_object_offset_t offset;
m2 = vm_page_grab();
if (m2 == VM_PAGE_NULL) {
if (locked_object) {
vm_object_unlock(locked_object);
locked_object = VM_OBJECT_NULL;
}
tmp_start_idx = cur_idx;
abort_run = TRUE;
continue;
}
if (m1->pmapped)
refmod = pmap_disconnect(m1->phys_page);
else
refmod = 0;
vm_page_copy(m1, m2);
m2->reference = m1->reference;
m2->dirty = m1->dirty;
if (refmod & VM_MEM_REFERENCED)
m2->reference = TRUE;
if (refmod & VM_MEM_MODIFIED)
m2->dirty = TRUE;
offset = m1->offset;
vm_page_free_prepare(m1);
pmap_clear_refmod(m2->phys_page, VM_MEM_MODIFIED | VM_MEM_REFERENCED);
vm_page_insert_internal(m2, locked_object, offset, TRUE);
if (m2->reference)
vm_page_activate(m2);
else
vm_page_deactivate(m2);
PAGE_WAKEUP_DONE(m2);
} else {
vm_page_free_prepare(m1);
}
#if MACH_ASSERT
stolen_pages++;
#endif
}
m1->pageq.next = (queue_entry_t) m;
m1->pageq.prev = NULL;
m = m1;
}
if (locked_object) {
vm_object_unlock(locked_object);
locked_object = VM_OBJECT_NULL;
}
if (abort_run == TRUE) {
if (m != VM_PAGE_NULL) {
vm_page_free_list(m);
}
#if MACH_ASSERT
dumped_run++;
#endif
page_idx = tmp_start_idx + 2;
if (page_idx >= vm_pages_count)
goto done_scanning;
mutex_lock(&vm_page_queue_free_lock);
RESET_STATE_OF_RUN();
free_available = vm_page_free_count - vm_page_free_reserved;
goto retry;
}
for (m1 = m; m1 != VM_PAGE_NULL; m1 = NEXT_PAGE(m1)) {
if (wire == TRUE)
m1->wire_count++;
else
m1->gobbled = TRUE;
}
if (wire == FALSE)
vm_page_gobble_count += npages;
vm_page_wire_count += npages;
assert(vm_page_verify_contiguous(m, npages));
}
done_scanning:
vm_page_unlock_queues();
#if MACH_ASSERT
clock_get_system_microtime(&tv_end_sec, &tv_end_usec);
tv_end_sec -= tv_start_sec;
if (tv_end_usec < tv_start_usec) {
tv_end_sec--;
tv_end_usec += 1000000;
}
tv_end_usec -= tv_start_usec;
if (tv_end_usec >= 1000000) {
tv_end_sec++;
tv_end_sec -= 1000000;
}
printf("vm_find_page_contiguous(num=%d,low=%d): found %d pages in %d.%06ds... scanned %d pages... yielded %d times... dumped run %d times... stole %d pages\n",
contig_pages, max_pnum, npages, tv_end_sec, tv_end_usec, page_idx, yielded, dumped_run, stolen_pages);
vm_page_verify_free_lists();
#endif
return m;
}
kern_return_t
cpm_allocate(
vm_size_t size,
vm_page_t *list,
ppnum_t max_pnum,
boolean_t wire)
{
vm_page_t pages;
unsigned int npages;
if (size % page_size != 0)
return KERN_INVALID_ARGUMENT;
npages = size / page_size;
pages = vm_page_find_contiguous(npages, max_pnum, wire);
if (pages == VM_PAGE_NULL)
return KERN_NO_SPACE;
if ((vm_page_free_count < vm_page_free_min) ||
((vm_page_free_count < vm_page_free_target) &&
((vm_page_inactive_count + vm_page_speculative_count) < vm_page_inactive_min)))
thread_wakeup((event_t) &vm_page_free_wanted);
#if CONFIG_EMBEDDED
{
int percent_avail;
percent_avail =
(vm_page_active_count + vm_page_inactive_count +
vm_page_speculative_count + vm_page_free_count +
(IP_VALID(memory_manager_default)?0:vm_page_purgeable_count) ) * 100 /
atop_64(max_mem);
if (percent_avail <= (kern_memorystatus_level - 5)) {
kern_memorystatus_level = percent_avail;
thread_wakeup((event_t)&kern_memorystatus_wakeup);
}
}
#endif
assert(vm_page_verify_contiguous(pages, npages));
*list = pages;
return KERN_SUCCESS;
}
#include <mach_vm_debug.h>
#if MACH_VM_DEBUG
#include <mach_debug/hash_info.h>
#include <vm/vm_debug.h>
unsigned int
vm_page_info(
hash_info_bucket_t *info,
unsigned int count)
{
unsigned int i;
if (vm_page_bucket_count < count)
count = vm_page_bucket_count;
for (i = 0; i < count; i++) {
vm_page_bucket_t *bucket = &vm_page_buckets[i];
unsigned int bucket_count = 0;
vm_page_t m;
simple_lock(&vm_page_bucket_lock);
for (m = bucket->pages; m != VM_PAGE_NULL; m = m->next)
bucket_count++;
simple_unlock(&vm_page_bucket_lock);
info[i].hib_count = bucket_count;
}
return vm_page_bucket_count;
}
#endif
#include <mach_kdb.h>
#if MACH_KDB
#include <ddb/db_output.h>
#include <vm/vm_print.h>
#define printf kdbprintf
void
vm_page_print(
db_addr_t db_addr)
{
vm_page_t p;
p = (vm_page_t) (long) db_addr;
iprintf("page 0x%x\n", p);
db_indent += 2;
iprintf("object=0x%x", p->object);
printf(", offset=0x%x", p->offset);
printf(", wire_count=%d", p->wire_count);
iprintf("%sinactive, %sactive, %sthrottled, %sgobbled, %slaundry, %sfree, %sref, %sencrypted\n",
(p->inactive ? "" : "!"),
(p->active ? "" : "!"),
(p->throttled ? "" : "!"),
(p->gobbled ? "" : "!"),
(p->laundry ? "" : "!"),
(p->free ? "" : "!"),
(p->reference ? "" : "!"),
(p->encrypted ? "" : "!"));
iprintf("%sbusy, %swanted, %stabled, %sfictitious, %sprivate, %sprecious\n",
(p->busy ? "" : "!"),
(p->wanted ? "" : "!"),
(p->tabled ? "" : "!"),
(p->fictitious ? "" : "!"),
(p->private ? "" : "!"),
(p->precious ? "" : "!"));
iprintf("%sabsent, %serror, %sdirty, %scleaning, %spageout, %sclustered\n",
(p->absent ? "" : "!"),
(p->error ? "" : "!"),
(p->dirty ? "" : "!"),
(p->cleaning ? "" : "!"),
(p->pageout ? "" : "!"),
(p->clustered ? "" : "!"));
iprintf("%soverwriting, %srestart, %sunusual\n",
(p->overwriting ? "" : "!"),
(p->restart ? "" : "!"),
(p->unusual ? "" : "!"));
iprintf("phys_page=0x%x", p->phys_page);
db_indent -= 2;
}
#endif