+ vm_offset_t old_head;
+ struct zone_page_metadata *page_meta;
+
+ vm_offset_t *primary = (vm_offset_t *) element;
+ vm_offset_t *backup = get_backup_ptr(zone->elem_size, primary);
+
+ if (zone->use_page_list) {
+ page_meta = get_zone_page_metadata((struct zone_free_element *)element);
+ assert(page_meta->zone == zone);
+ old_head = (vm_offset_t)page_meta->elements;
+ } else {
+ old_head = (vm_offset_t)zone->free_elements;
+ }
+
+#if MACH_ASSERT
+ if (__improbable(!is_sane_zone_element(zone, old_head)))
+ panic("zfree: invalid head pointer %p for freelist of zone %s\n",
+ (void *) old_head, zone->zone_name);
+#endif
+
+ if (__improbable(!is_sane_zone_element(zone, element)))
+ panic("zfree: freeing invalid pointer %p to zone %s\n",
+ (void *) element, zone->zone_name);
+
+ /*
+ * Always write a redundant next pointer
+ * So that it is more difficult to forge, xor it with a random cookie
+ * A poisoned element is indicated by using zp_poisoned_cookie
+ * instead of zp_nopoison_cookie
+ */
+
+ *backup = old_head ^ (poison ? zp_poisoned_cookie : zp_nopoison_cookie);
+
+ /* Insert this element at the head of the free list */
+ *primary = old_head;
+ if (zone->use_page_list) {
+ page_meta->elements = (struct zone_free_element *)element;
+ page_meta->free_count++;
+ if (zone->allows_foreign && !from_zone_map(element, zone->elem_size)) {
+ if (page_meta->free_count == 1) {
+ /* first foreign element freed on page, move from all_used */
+ remqueue((queue_entry_t)page_meta);
+ enqueue_tail(&zone->pages.any_free_foreign, (queue_entry_t)page_meta);
+ } else {
+ /* no other list transitions */
+ }
+ } else if (page_meta->free_count == page_meta->alloc_count) {
+ /* whether the page was on the intermediate or all_used, queue, move it to free */
+ remqueue((queue_entry_t)page_meta);
+ enqueue_tail(&zone->pages.all_free, (queue_entry_t)page_meta);
+ } else if (page_meta->free_count == 1) {
+ /* first free element on page, move from all_used */
+ remqueue((queue_entry_t)page_meta);
+ enqueue_tail(&zone->pages.intermediate, (queue_entry_t)page_meta);
+ }
+ } else {
+ zone->free_elements = (struct zone_free_element *)element;
+ }
+ zone->count--;
+ zone->countfree++;
+}
+
+
+/*
+ * Removes an element from the zone's free list, returning 0 if the free list is empty.
+ * Verifies that the next-pointer and backup next-pointer are intact,
+ * and verifies that a poisoned element hasn't been modified.
+ */
+static inline vm_offset_t
+try_alloc_from_zone(zone_t zone,
+ boolean_t* check_poison)
+{
+ vm_offset_t element;
+ struct zone_page_metadata *page_meta;
+
+ *check_poison = FALSE;
+
+ /* if zone is empty, bail */
+ if (zone->use_page_list) {
+ if (zone->allows_foreign && !queue_empty(&zone->pages.any_free_foreign))
+ page_meta = (struct zone_page_metadata *)queue_first(&zone->pages.any_free_foreign);
+ else if (!queue_empty(&zone->pages.intermediate))
+ page_meta = (struct zone_page_metadata *)queue_first(&zone->pages.intermediate);
+ else if (!queue_empty(&zone->pages.all_free))
+ page_meta = (struct zone_page_metadata *)queue_first(&zone->pages.all_free);
+ else {
+ return 0;
+ }
+
+ /* Check if page_meta passes is_sane_zone_element */
+ if (__improbable(!is_sane_zone_page_metadata(zone, (vm_offset_t)page_meta)))
+ panic("zalloc: invalid metadata structure %p for freelist of zone %s\n",
+ (void *) page_meta, zone->zone_name);
+ assert(page_meta->zone == zone);
+ element = (vm_offset_t)page_meta->elements;
+ } else {
+ if (zone->free_elements == NULL)
+ return 0;
+
+ element = (vm_offset_t)zone->free_elements;
+ }
+
+#if MACH_ASSERT
+ if (__improbable(!is_sane_zone_element(zone, element)))
+ panic("zfree: invalid head pointer %p for freelist of zone %s\n",
+ (void *) element, zone->zone_name);
+#endif
+
+ vm_offset_t *primary = (vm_offset_t *) element;
+ vm_offset_t *backup = get_backup_ptr(zone->elem_size, primary);
+
+ vm_offset_t next_element = *primary;
+ vm_offset_t next_element_backup = *backup;
+
+ /*
+ * backup_ptr_mismatch_panic will determine what next_element
+ * should have been, and print it appropriately
+ */
+ if (__improbable(!is_sane_zone_element(zone, next_element)))
+ backup_ptr_mismatch_panic(zone, element, next_element, next_element_backup);
+
+ /* Check the backup pointer for the regular cookie */
+ if (__improbable(next_element != (next_element_backup ^ zp_nopoison_cookie))) {
+
+ /* Check for the poisoned cookie instead */
+ if (__improbable(next_element != (next_element_backup ^ zp_poisoned_cookie)))
+ /* Neither cookie is valid, corruption has occurred */
+ backup_ptr_mismatch_panic(zone, element, next_element, next_element_backup);
+
+ /*
+ * Element was marked as poisoned, so check its integrity before using it.
+ */
+ *check_poison = TRUE;
+ }
+
+ if (zone->use_page_list) {
+
+ /* Make sure the page_meta is at the correct offset from the start of page */
+ if (__improbable(page_meta != get_zone_page_metadata((struct zone_free_element *)element)))
+ panic("zalloc: metadata located at incorrect location on page of zone %s\n",
+ zone->zone_name);
+
+ /* Make sure next_element belongs to the same page as page_meta */
+ if (next_element) {
+ if (__improbable(page_meta != get_zone_page_metadata((struct zone_free_element *)next_element)))
+ panic("zalloc: next element pointer %p for element %p points to invalid element for zone %s\n",
+ (void *)next_element, (void *)element, zone->zone_name);
+ }
+ }
+
+ /* Remove this element from the free list */
+ if (zone->use_page_list) {
+
+ page_meta->elements = (struct zone_free_element *)next_element;
+ page_meta->free_count--;
+
+ if (zone->allows_foreign && !from_zone_map(element, zone->elem_size)) {
+ if (page_meta->free_count == 0) {
+ /* move to all used */
+ remqueue((queue_entry_t)page_meta);
+ enqueue_tail(&zone->pages.all_used, (queue_entry_t)page_meta);
+ } else {
+ /* no other list transitions */
+ }
+ } else if (page_meta->free_count == 0) {
+ /* remove from intermediate or free, move to all_used */
+ remqueue((queue_entry_t)page_meta);
+ enqueue_tail(&zone->pages.all_used, (queue_entry_t)page_meta);
+ } else if (page_meta->alloc_count == page_meta->free_count + 1) {
+ /* remove from free, move to intermediate */
+ remqueue((queue_entry_t)page_meta);
+ enqueue_tail(&zone->pages.intermediate, (queue_entry_t)page_meta);
+ }
+ } else {
+ zone->free_elements = (struct zone_free_element *)next_element;
+ }
+ zone->countfree--;
+ zone->count++;
+ zone->sum_count++;
+
+ return element;
+}
+
+
+/*
+ * End of zone poisoning
+ */
+
+/*
+ * Fake zones for things that want to report via zprint but are not actually zones.
+ */
+struct fake_zone_info {
+ const char* name;
+ void (*init)(int);
+ void (*query)(int *,
+ vm_size_t *, vm_size_t *, vm_size_t *, vm_size_t *,
+ uint64_t *, int *, int *, int *);
+};
+
+static const struct fake_zone_info fake_zones[] = {
+};
+static const unsigned int num_fake_zones =
+ sizeof (fake_zones) / sizeof (fake_zones[0]);
+
+/*
+ * Zone info options
+ */
+boolean_t zinfo_per_task = FALSE; /* enabled by -zinfop in boot-args */
+#define ZINFO_SLOTS 200 /* for now */
+#define ZONES_MAX (ZINFO_SLOTS - num_fake_zones - 1)
+
+/*
+ * Support for garbage collection of unused zone pages
+ *
+ * The kernel virtually allocates the "zone map" submap of the kernel
+ * map. When an individual zone needs more storage, memory is allocated
+ * out of the zone map, and the two-level "zone_page_table" is
+ * on-demand expanded so that it has entries for those pages.
+ * zone_page_init()/zone_page_alloc() initialize "alloc_count"
+ * to the number of zone elements that occupy the zone page (which may
+ * be a minimum of 1, including if a zone element spans multiple
+ * pages).
+ *
+ * Asynchronously, the zone_gc() logic attempts to walk zone free
+ * lists to see if all the elements on a zone page are free. If
+ * "collect_count" (which it increments during the scan) matches
+ * "alloc_count", the zone page is a candidate for collection and the
+ * physical page is returned to the VM system. During this process, the
+ * first word of the zone page is re-used to maintain a linked list of
+ * to-be-collected zone pages.
+ */
+typedef uint32_t zone_page_index_t;
+#define ZONE_PAGE_INDEX_INVALID ((zone_page_index_t)0xFFFFFFFFU)
+
+struct zone_page_table_entry {
+ volatile uint16_t alloc_count;
+ volatile uint16_t collect_count;
+};
+
+#define ZONE_PAGE_USED 0
+#define ZONE_PAGE_UNUSED 0xffff
+
+/* Forwards */
+void zone_page_init(
+ vm_offset_t addr,
+ vm_size_t size);
+
+void zone_page_alloc(
+ vm_offset_t addr,
+ vm_size_t size);
+
+void zone_page_free_element(
+ zone_page_index_t *free_page_head,
+ zone_page_index_t *free_page_tail,
+ vm_offset_t addr,
+ vm_size_t size);
+
+void zone_page_collect(
+ vm_offset_t addr,
+ vm_size_t size);
+
+boolean_t zone_page_collectable(
+ vm_offset_t addr,
+ vm_size_t size);
+
+void zone_page_keep(
+ vm_offset_t addr,
+ vm_size_t size);
+
+void zone_display_zprint(void);
+
+zone_t zone_find_largest(void);
+
+/*
+ * Async allocation of zones
+ * This mechanism allows for bootstrapping an empty zone which is setup with
+ * non-blocking flags. The first call to zalloc_noblock() will kick off a thread_call
+ * to zalloc_async. We perform a zalloc() (which may block) and then an immediate free.
+ * This will prime the zone for the next use.
+ *
+ * Currently the thread_callout function (zalloc_async) will loop through all zones
+ * looking for any zone with async_pending set and do the work for it.
+ *
+ * NOTE: If the calling thread for zalloc_noblock is lower priority than thread_call,
+ * then zalloc_noblock to an empty zone may succeed.
+ */
+void zalloc_async(
+ thread_call_param_t p0,
+ thread_call_param_t p1);
+
+static thread_call_data_t call_async_alloc;
+
+vm_map_t zone_map = VM_MAP_NULL;
+
+zone_t zone_zone = ZONE_NULL; /* the zone containing other zones */
+
+zone_t zinfo_zone = ZONE_NULL; /* zone of per-task zone info */
+
+/*
+ * The VM system gives us an initial chunk of memory.
+ * It has to be big enough to allocate the zone_zone
+ * all the way through the pmap zone.
+ */
+
+vm_offset_t zdata;
+vm_size_t zdata_size;
+/*
+ * Align elements that use the zone page list to 32 byte boundaries.
+ */
+#define ZONE_ELEMENT_ALIGNMENT 32
+
+#define zone_wakeup(zone) thread_wakeup((event_t)(zone))
+#define zone_sleep(zone) \
+ (void) lck_mtx_sleep(&(zone)->lock, LCK_SLEEP_SPIN, (event_t)(zone), THREAD_UNINT);
+
+/*
+ * The zone_locks_grp allows for collecting lock statistics.
+ * All locks are associated to this group in zinit.
+ * Look at tools/lockstat for debugging lock contention.
+ */
+
+lck_grp_t zone_locks_grp;
+lck_grp_attr_t zone_locks_grp_attr;
+
+#define lock_zone_init(zone) \
+MACRO_BEGIN \
+ lck_attr_setdefault(&(zone)->lock_attr); \
+ lck_mtx_init_ext(&(zone)->lock, &(zone)->lock_ext, \
+ &zone_locks_grp, &(zone)->lock_attr); \
+MACRO_END
+
+#define lock_try_zone(zone) lck_mtx_try_lock_spin(&zone->lock)
+
+/*
+ * Garbage collection map information
+ */
+#define ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE (32)
+struct zone_page_table_entry * volatile zone_page_table[ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE];
+vm_size_t zone_page_table_used_size;
+unsigned int zone_pages;
+unsigned int zone_page_table_second_level_size; /* power of 2 */
+unsigned int zone_page_table_second_level_shift_amount;
+
+#define zone_page_table_first_level_slot(x) ((x) >> zone_page_table_second_level_shift_amount)
+#define zone_page_table_second_level_slot(x) ((x) & (zone_page_table_second_level_size - 1))
+
+void zone_page_table_expand(zone_page_index_t pindex);
+struct zone_page_table_entry *zone_page_table_lookup(zone_page_index_t pindex);
+
+/*
+ * Exclude more than one concurrent garbage collection
+ */
+decl_lck_mtx_data(, zone_gc_lock)
+
+lck_attr_t zone_gc_lck_attr;
+lck_grp_t zone_gc_lck_grp;
+lck_grp_attr_t zone_gc_lck_grp_attr;
+lck_mtx_ext_t zone_gc_lck_ext;
+
+/*
+ * Protects first_zone, last_zone, num_zones,
+ * and the next_zone field of zones.
+ */
+decl_simple_lock_data(, all_zones_lock)
+zone_t first_zone;
+zone_t *last_zone;
+unsigned int num_zones;
+
+boolean_t zone_gc_allowed = TRUE;
+boolean_t zone_gc_forced = FALSE;
+boolean_t panic_include_zprint = FALSE;
+boolean_t zone_gc_allowed_by_time_throttle = TRUE;
+
+vm_offset_t panic_kext_memory_info = 0;
+vm_size_t panic_kext_memory_size = 0;
+
+#define ZALLOC_DEBUG_ZONEGC 0x00000001
+#define ZALLOC_DEBUG_ZCRAM 0x00000002
+uint32_t zalloc_debug = 0;
+
+/*
+ * Zone leak debugging code
+ *
+ * When enabled, this code keeps a log to track allocations to a particular zone that have not
+ * yet been freed. Examining this log will reveal the source of a zone leak. The log is allocated
+ * only when logging is enabled, so there is no effect on the system when it's turned off. Logging is
+ * off by default.
+ *
+ * Enable the logging via the boot-args. Add the parameter "zlog=<zone>" to boot-args where <zone>
+ * is the name of the zone you wish to log.
+ *
+ * This code only tracks one zone, so you need to identify which one is leaking first.
+ * Generally, you'll know you have a leak when you get a "zalloc retry failed 3" panic from the zone
+ * garbage collector. Note that the zone name printed in the panic message is not necessarily the one
+ * containing the leak. So do a zprint from gdb and locate the zone with the bloated size. This
+ * is most likely the problem zone, so set zlog in boot-args to this zone name, reboot and re-run the test. The
+ * next time it panics with this message, examine the log using the kgmacros zstack, findoldest and countpcs.
+ * See the help in the kgmacros for usage info.
+ *
+ *
+ * Zone corruption logging
+ *
+ * Logging can also be used to help identify the source of a zone corruption. First, identify the zone
+ * that is being corrupted, then add "-zc zlog=<zone name>" to the boot-args. When -zc is used in conjunction
+ * with zlog, it changes the logging style to track both allocations and frees to the zone. So when the
+ * corruption is detected, examining the log will show you the stack traces of the callers who last allocated
+ * and freed any particular element in the zone. Use the findelem kgmacro with the address of the element that's been
+ * corrupted to examine its history. This should lead to the source of the corruption.
+ */
+
+static int log_records; /* size of the log, expressed in number of records */
+
+#define MAX_ZONE_NAME 32 /* max length of a zone name we can take from the boot-args */
+
+static char zone_name_to_log[MAX_ZONE_NAME] = ""; /* the zone name we're logging, if any */
+
+/* Log allocations and frees to help debug a zone element corruption */
+boolean_t corruption_debug_flag = FALSE; /* enabled by "-zc" boot-arg */
+
+/*
+ * The number of records in the log is configurable via the zrecs parameter in boot-args. Set this to
+ * the number of records you want in the log. For example, "zrecs=1000" sets it to 1000 records. Note
+ * that the larger the size of the log, the slower the system will run due to linear searching in the log,
+ * but one doesn't generally care about performance when tracking down a leak. The log is capped at 8000
+ * records since going much larger than this tends to make the system unresponsive and unbootable on small
+ * memory configurations. The default value is 4000 records.
+ */
+
+#if defined(__LP64__)
+#define ZRECORDS_MAX 128000 /* Max records allowed in the log */
+#else
+#define ZRECORDS_MAX 8000 /* Max records allowed in the log */
+#endif
+#define ZRECORDS_DEFAULT 4000 /* default records in log if zrecs is not specificed in boot-args */
+
+/*
+ * Each record in the log contains a pointer to the zone element it refers to,
+ * and a small array to hold the pc's from the stack trace. A
+ * record is added to the log each time a zalloc() is done in the zone_of_interest. For leak debugging,
+ * the record is cleared when a zfree() is done. For corruption debugging, the log tracks both allocs and frees.
+ * If the log fills, old records are replaced as if it were a circular buffer.
+ */
+
+
+/*
+ * Opcodes for the btlog operation field:
+ */
+
+#define ZOP_ALLOC 1
+#define ZOP_FREE 0
+
+/*
+ * The allocation log and all the related variables are protected by the zone lock for the zone_of_interest
+ */
+static btlog_t *zlog_btlog; /* the log itself, dynamically allocated when logging is enabled */
+static zone_t zone_of_interest = NULL; /* the zone being watched; corresponds to zone_name_to_log */
+
+/*
+ * Decide if we want to log this zone by doing a string compare between a zone name and the name
+ * of the zone to log. Return true if the strings are equal, false otherwise. Because it's not
+ * possible to include spaces in strings passed in via the boot-args, a period in the logname will
+ * match a space in the zone name.
+ */
+
+static int
+log_this_zone(const char *zonename, const char *logname)
+{
+ int len;
+ const char *zc = zonename;
+ const char *lc = logname;
+
+ /*
+ * Compare the strings. We bound the compare by MAX_ZONE_NAME.
+ */
+
+ for (len = 1; len <= MAX_ZONE_NAME; zc++, lc++, len++) {
+
+ /*
+ * If the current characters don't match, check for a space in
+ * in the zone name and a corresponding period in the log name.
+ * If that's not there, then the strings don't match.
+ */
+
+ if (*zc != *lc && !(*zc == ' ' && *lc == '.'))
+ break;
+
+ /*
+ * The strings are equal so far. If we're at the end, then it's a match.
+ */
+
+ if (*zc == '\0')
+ return TRUE;
+ }
+
+ return FALSE;
+}
+
+
+/*
+ * Test if we want to log this zalloc/zfree event. We log if this is the zone we're interested in and
+ * the buffer for the records has been allocated.
+ */
+
+#define DO_LOGGING(z) (zlog_btlog && (z) == zone_of_interest)
+
+extern boolean_t kmem_alloc_ready;
+
+#if CONFIG_ZLEAKS
+#pragma mark -
+#pragma mark Zone Leak Detection
+
+/*
+ * The zone leak detector, abbreviated 'zleak', keeps track of a subset of the currently outstanding
+ * allocations made by the zone allocator. Every zleak_sample_factor allocations in each zone, we capture a
+ * backtrace. Every free, we examine the table and determine if the allocation was being tracked,
+ * and stop tracking it if it was being tracked.
+ *
+ * We track the allocations in the zallocations hash table, which stores the address that was returned from
+ * the zone allocator. Each stored entry in the zallocations table points to an entry in the ztraces table, which
+ * stores the backtrace associated with that allocation. This provides uniquing for the relatively large
+ * backtraces - we don't store them more than once.
+ *
+ * Data collection begins when the zone map is 50% full, and only occurs for zones that are taking up
+ * a large amount of virtual space.
+ */
+#define ZLEAK_STATE_ENABLED 0x01 /* Zone leak monitoring should be turned on if zone_map fills up. */
+#define ZLEAK_STATE_ACTIVE 0x02 /* We are actively collecting traces. */
+#define ZLEAK_STATE_ACTIVATING 0x04 /* Some thread is doing setup; others should move along. */
+#define ZLEAK_STATE_FAILED 0x08 /* Attempt to allocate tables failed. We will not try again. */
+uint32_t zleak_state = 0; /* State of collection, as above */
+
+boolean_t panic_include_ztrace = FALSE; /* Enable zleak logging on panic */
+vm_size_t zleak_global_tracking_threshold; /* Size of zone map at which to start collecting data */
+vm_size_t zleak_per_zone_tracking_threshold; /* Size a zone will have before we will collect data on it */
+unsigned int zleak_sample_factor = 1000; /* Allocations per sample attempt */
+
+/*
+ * Counters for allocation statistics.
+ */
+
+/* Times two active records want to occupy the same spot */
+unsigned int z_alloc_collisions = 0;
+unsigned int z_trace_collisions = 0;
+
+/* Times a new record lands on a spot previously occupied by a freed allocation */
+unsigned int z_alloc_overwrites = 0;
+unsigned int z_trace_overwrites = 0;
+
+/* Times a new alloc or trace is put into the hash table */
+unsigned int z_alloc_recorded = 0;
+unsigned int z_trace_recorded = 0;
+
+/* Times zleak_log returned false due to not being able to acquire the lock */
+unsigned int z_total_conflicts = 0;
+
+
+#pragma mark struct zallocation
+/*
+ * Structure for keeping track of an allocation
+ * An allocation bucket is in use if its element is not NULL
+ */
+struct zallocation {
+ uintptr_t za_element; /* the element that was zalloc'ed or zfree'ed, NULL if bucket unused */
+ vm_size_t za_size; /* how much memory did this allocation take up? */
+ uint32_t za_trace_index; /* index into ztraces for backtrace associated with allocation */
+ /* TODO: #if this out */
+ uint32_t za_hit_count; /* for determining effectiveness of hash function */
+};
+
+/* Size must be a power of two for the zhash to be able to just mask off bits instead of mod */
+uint32_t zleak_alloc_buckets = CONFIG_ZLEAK_ALLOCATION_MAP_NUM;
+uint32_t zleak_trace_buckets = CONFIG_ZLEAK_TRACE_MAP_NUM;
+
+vm_size_t zleak_max_zonemap_size;
+
+/* Hashmaps of allocations and their corresponding traces */
+static struct zallocation* zallocations;
+static struct ztrace* ztraces;
+
+/* not static so that panic can see this, see kern/debug.c */
+struct ztrace* top_ztrace;
+
+/* Lock to protect zallocations, ztraces, and top_ztrace from concurrent modification. */
+static lck_spin_t zleak_lock;
+static lck_attr_t zleak_lock_attr;
+static lck_grp_t zleak_lock_grp;
+static lck_grp_attr_t zleak_lock_grp_attr;
+
+/*
+ * Initializes the zone leak monitor. Called from zone_init()
+ */
+static void
+zleak_init(vm_size_t max_zonemap_size)
+{
+ char scratch_buf[16];
+ boolean_t zleak_enable_flag = FALSE;
+
+ zleak_max_zonemap_size = max_zonemap_size;
+ zleak_global_tracking_threshold = max_zonemap_size / 2;
+ zleak_per_zone_tracking_threshold = zleak_global_tracking_threshold / 8;
+
+ /* -zleakoff (flag to disable zone leak monitor) */
+ if (PE_parse_boot_argn("-zleakoff", scratch_buf, sizeof(scratch_buf))) {
+ zleak_enable_flag = FALSE;
+ printf("zone leak detection disabled\n");
+ } else {
+ zleak_enable_flag = TRUE;
+ printf("zone leak detection enabled\n");
+ }
+
+ /* zfactor=XXXX (override how often to sample the zone allocator) */
+ if (PE_parse_boot_argn("zfactor", &zleak_sample_factor, sizeof(zleak_sample_factor))) {
+ printf("Zone leak factor override: %u\n", zleak_sample_factor);
+ }
+
+ /* zleak-allocs=XXXX (override number of buckets in zallocations) */
+ if (PE_parse_boot_argn("zleak-allocs", &zleak_alloc_buckets, sizeof(zleak_alloc_buckets))) {
+ printf("Zone leak alloc buckets override: %u\n", zleak_alloc_buckets);
+ /* uses 'is power of 2' trick: (0x01000 & 0x00FFF == 0) */
+ if (zleak_alloc_buckets == 0 || (zleak_alloc_buckets & (zleak_alloc_buckets-1))) {
+ printf("Override isn't a power of two, bad things might happen!\n");
+ }
+ }
+
+ /* zleak-traces=XXXX (override number of buckets in ztraces) */
+ if (PE_parse_boot_argn("zleak-traces", &zleak_trace_buckets, sizeof(zleak_trace_buckets))) {
+ printf("Zone leak trace buckets override: %u\n", zleak_trace_buckets);
+ /* uses 'is power of 2' trick: (0x01000 & 0x00FFF == 0) */
+ if (zleak_trace_buckets == 0 || (zleak_trace_buckets & (zleak_trace_buckets-1))) {
+ printf("Override isn't a power of two, bad things might happen!\n");
+ }
+ }
+
+ /* allocate the zleak_lock */
+ lck_grp_attr_setdefault(&zleak_lock_grp_attr);
+ lck_grp_init(&zleak_lock_grp, "zleak_lock", &zleak_lock_grp_attr);
+ lck_attr_setdefault(&zleak_lock_attr);
+ lck_spin_init(&zleak_lock, &zleak_lock_grp, &zleak_lock_attr);
+
+ if (zleak_enable_flag) {
+ zleak_state = ZLEAK_STATE_ENABLED;
+ }
+}
+
+#if CONFIG_ZLEAKS
+
+/*
+ * Support for kern.zleak.active sysctl - a simplified
+ * version of the zleak_state variable.
+ */
+int
+get_zleak_state(void)
+{
+ if (zleak_state & ZLEAK_STATE_FAILED)
+ return (-1);
+ if (zleak_state & ZLEAK_STATE_ACTIVE)
+ return (1);
+ return (0);
+}
+
+#endif
+
+
+kern_return_t
+zleak_activate(void)
+{
+ kern_return_t retval;
+ vm_size_t z_alloc_size = zleak_alloc_buckets * sizeof(struct zallocation);
+ vm_size_t z_trace_size = zleak_trace_buckets * sizeof(struct ztrace);
+ void *allocations_ptr = NULL;
+ void *traces_ptr = NULL;
+
+ /* Only one thread attempts to activate at a time */
+ if (zleak_state & (ZLEAK_STATE_ACTIVE | ZLEAK_STATE_ACTIVATING | ZLEAK_STATE_FAILED)) {
+ return KERN_SUCCESS;
+ }
+
+ /* Indicate that we're doing the setup */
+ lck_spin_lock(&zleak_lock);
+ if (zleak_state & (ZLEAK_STATE_ACTIVE | ZLEAK_STATE_ACTIVATING | ZLEAK_STATE_FAILED)) {
+ lck_spin_unlock(&zleak_lock);
+ return KERN_SUCCESS;
+ }
+
+ zleak_state |= ZLEAK_STATE_ACTIVATING;
+ lck_spin_unlock(&zleak_lock);
+
+ /* Allocate and zero tables */
+ retval = kmem_alloc_kobject(kernel_map, (vm_offset_t*)&allocations_ptr, z_alloc_size, VM_KERN_MEMORY_OSFMK);
+ if (retval != KERN_SUCCESS) {
+ goto fail;
+ }
+
+ retval = kmem_alloc_kobject(kernel_map, (vm_offset_t*)&traces_ptr, z_trace_size, VM_KERN_MEMORY_OSFMK);
+ if (retval != KERN_SUCCESS) {
+ goto fail;
+ }
+
+ bzero(allocations_ptr, z_alloc_size);
+ bzero(traces_ptr, z_trace_size);
+
+ /* Everything's set. Install tables, mark active. */
+ zallocations = allocations_ptr;
+ ztraces = traces_ptr;
+
+ /*
+ * Initialize the top_ztrace to the first entry in ztraces,
+ * so we don't have to check for null in zleak_log
+ */
+ top_ztrace = &ztraces[0];
+
+ /*
+ * Note that we do need a barrier between installing
+ * the tables and setting the active flag, because the zfree()
+ * path accesses the table without a lock if we're active.
+ */
+ lck_spin_lock(&zleak_lock);
+ zleak_state |= ZLEAK_STATE_ACTIVE;
+ zleak_state &= ~ZLEAK_STATE_ACTIVATING;
+ lck_spin_unlock(&zleak_lock);
+
+ return 0;
+
+fail:
+ /*
+ * If we fail to allocate memory, don't further tax
+ * the system by trying again.
+ */
+ lck_spin_lock(&zleak_lock);
+ zleak_state |= ZLEAK_STATE_FAILED;
+ zleak_state &= ~ZLEAK_STATE_ACTIVATING;
+ lck_spin_unlock(&zleak_lock);
+
+ if (allocations_ptr != NULL) {
+ kmem_free(kernel_map, (vm_offset_t)allocations_ptr, z_alloc_size);
+ }
+
+ if (traces_ptr != NULL) {
+ kmem_free(kernel_map, (vm_offset_t)traces_ptr, z_trace_size);
+ }
+
+ return retval;
+}
+
+/*
+ * TODO: What about allocations that never get deallocated,
+ * especially ones with unique backtraces? Should we wait to record
+ * until after boot has completed?
+ * (How many persistent zallocs are there?)
+ */
+
+/*
+ * This function records the allocation in the allocations table,
+ * and stores the associated backtrace in the traces table
+ * (or just increments the refcount if the trace is already recorded)
+ * If the allocation slot is in use, the old allocation is replaced with the new allocation, and
+ * the associated trace's refcount is decremented.
+ * If the trace slot is in use, it returns.
+ * The refcount is incremented by the amount of memory the allocation consumes.
+ * The return value indicates whether to try again next time.
+ */
+static boolean_t
+zleak_log(uintptr_t* bt,
+ uintptr_t addr,
+ uint32_t depth,
+ vm_size_t allocation_size)
+{
+ /* Quit if there's someone else modifying the hash tables */
+ if (!lck_spin_try_lock(&zleak_lock)) {
+ z_total_conflicts++;
+ return FALSE;
+ }
+
+ struct zallocation* allocation = &zallocations[hashaddr(addr, zleak_alloc_buckets)];
+
+ uint32_t trace_index = hashbacktrace(bt, depth, zleak_trace_buckets);
+ struct ztrace* trace = &ztraces[trace_index];
+
+ allocation->za_hit_count++;
+ trace->zt_hit_count++;
+
+ /*
+ * If the allocation bucket we want to be in is occupied, and if the occupier
+ * has the same trace as us, just bail.
+ */
+ if (allocation->za_element != (uintptr_t) 0 && trace_index == allocation->za_trace_index) {
+ z_alloc_collisions++;
+
+ lck_spin_unlock(&zleak_lock);
+ return TRUE;
+ }
+
+ /* STEP 1: Store the backtrace in the traces array. */
+ /* A size of zero indicates that the trace bucket is free. */
+
+ if (trace->zt_size > 0 && bcmp(trace->zt_stack, bt, (depth * sizeof(uintptr_t))) != 0 ) {
+ /*
+ * Different unique trace with same hash!
+ * Just bail - if we're trying to record the leaker, hopefully the other trace will be deallocated
+ * and get out of the way for later chances
+ */
+ trace->zt_collisions++;
+ z_trace_collisions++;
+
+ lck_spin_unlock(&zleak_lock);
+ return TRUE;
+ } else if (trace->zt_size > 0) {
+ /* Same trace, already added, so increment refcount */
+ trace->zt_size += allocation_size;
+ } else {
+ /* Found an unused trace bucket, record the trace here! */
+ if (trace->zt_depth != 0) /* if this slot was previously used but not currently in use */
+ z_trace_overwrites++;
+
+ z_trace_recorded++;
+ trace->zt_size = allocation_size;
+ memcpy(trace->zt_stack, bt, (depth * sizeof(uintptr_t)) );
+
+ trace->zt_depth = depth;
+ trace->zt_collisions = 0;
+ }
+
+ /* STEP 2: Store the allocation record in the allocations array. */
+
+ if (allocation->za_element != (uintptr_t) 0) {
+ /*
+ * Straight up replace any allocation record that was there. We don't want to do the work
+ * to preserve the allocation entries that were there, because we only record a subset of the
+ * allocations anyways.
+ */
+
+ z_alloc_collisions++;
+
+ struct ztrace* associated_trace = &ztraces[allocation->za_trace_index];
+ /* Knock off old allocation's size, not the new allocation */
+ associated_trace->zt_size -= allocation->za_size;
+ } else if (allocation->za_trace_index != 0) {
+ /* Slot previously used but not currently in use */
+ z_alloc_overwrites++;
+ }
+
+ allocation->za_element = addr;
+ allocation->za_trace_index = trace_index;
+ allocation->za_size = allocation_size;
+
+ z_alloc_recorded++;
+
+ if (top_ztrace->zt_size < trace->zt_size)
+ top_ztrace = trace;
+
+ lck_spin_unlock(&zleak_lock);
+ return TRUE;
+}
+
+/*
+ * Free the allocation record and release the stacktrace.
+ * This should be as fast as possible because it will be called for every free.
+ */
+static void
+zleak_free(uintptr_t addr,
+ vm_size_t allocation_size)
+{
+ if (addr == (uintptr_t) 0)
+ return;
+
+ struct zallocation* allocation = &zallocations[hashaddr(addr, zleak_alloc_buckets)];
+
+ /* Double-checked locking: check to find out if we're interested, lock, check to make
+ * sure it hasn't changed, then modify it, and release the lock.
+ */
+
+ if (allocation->za_element == addr && allocation->za_trace_index < zleak_trace_buckets) {
+ /* if the allocation was the one, grab the lock, check again, then delete it */
+ lck_spin_lock(&zleak_lock);
+
+ if (allocation->za_element == addr && allocation->za_trace_index < zleak_trace_buckets) {
+ struct ztrace *trace;
+
+ /* allocation_size had better match what was passed into zleak_log - otherwise someone is freeing into the wrong zone! */
+ if (allocation->za_size != allocation_size) {
+ panic("Freeing as size %lu memory that was allocated with size %lu\n",
+ (uintptr_t)allocation_size, (uintptr_t)allocation->za_size);
+ }
+
+ trace = &ztraces[allocation->za_trace_index];
+
+ /* size of 0 indicates trace bucket is unused */
+ if (trace->zt_size > 0) {
+ trace->zt_size -= allocation_size;
+ }
+
+ /* A NULL element means the allocation bucket is unused */
+ allocation->za_element = 0;
+ }
+ lck_spin_unlock(&zleak_lock);
+ }
+}
+
+#endif /* CONFIG_ZLEAKS */
+
+/* These functions outside of CONFIG_ZLEAKS because they are also used in
+ * mbuf.c for mbuf leak-detection. This is why they lack the z_ prefix.
+ */
+
+/*
+ * This function captures a backtrace from the current stack and
+ * returns the number of frames captured, limited by max_frames.
+ * It's fast because it does no checking to make sure there isn't bad data.
+ * Since it's only called from threads that we're going to keep executing,
+ * if there's bad data we were going to die eventually.
+ * If this function is inlined, it doesn't record the frame of the function it's inside.
+ * (because there's no stack frame!)
+ */
+
+uint32_t
+fastbacktrace(uintptr_t* bt, uint32_t max_frames)
+{
+ uintptr_t* frameptr = NULL, *frameptr_next = NULL;
+ uintptr_t retaddr = 0;
+ uint32_t frame_index = 0, frames = 0;
+ uintptr_t kstackb, kstackt;
+ thread_t cthread = current_thread();
+
+ if (__improbable(cthread == NULL))
+ return 0;
+
+ kstackb = cthread->kernel_stack;
+ kstackt = kstackb + kernel_stack_size;
+ /* Load stack frame pointer (EBP on x86) into frameptr */
+ frameptr = __builtin_frame_address(0);
+ if (((uintptr_t)frameptr > kstackt) || ((uintptr_t)frameptr < kstackb))
+ frameptr = NULL;
+
+ while (frameptr != NULL && frame_index < max_frames ) {
+ /* Next frame pointer is pointed to by the previous one */
+ frameptr_next = (uintptr_t*) *frameptr;
+
+ /* Bail if we see a zero in the stack frame, that means we've reached the top of the stack */
+ /* That also means the return address is worthless, so don't record it */
+ if (frameptr_next == NULL)
+ break;
+ /* Verify thread stack bounds */
+ if (((uintptr_t)frameptr_next > kstackt) || ((uintptr_t)frameptr_next < kstackb))
+ break;
+ /* Pull return address from one spot above the frame pointer */
+ retaddr = *(frameptr + 1);
+
+ /* Store it in the backtrace array */
+ bt[frame_index++] = retaddr;
+
+ frameptr = frameptr_next;
+ }
+
+ /* Save the number of frames captured for return value */
+ frames = frame_index;
+
+ /* Fill in the rest of the backtrace with zeros */
+ while (frame_index < max_frames)
+ bt[frame_index++] = 0;
+
+ return frames;
+}
+
+/* "Thomas Wang's 32/64 bit mix functions." http://www.concentric.net/~Ttwang/tech/inthash.htm */
+uintptr_t
+hash_mix(uintptr_t x)
+{
+#ifndef __LP64__
+ x += ~(x << 15);
+ x ^= (x >> 10);
+ x += (x << 3 );
+ x ^= (x >> 6 );
+ x += ~(x << 11);
+ x ^= (x >> 16);
+#else
+ x += ~(x << 32);
+ x ^= (x >> 22);
+ x += ~(x << 13);
+ x ^= (x >> 8 );
+ x += (x << 3 );
+ x ^= (x >> 15);
+ x += ~(x << 27);
+ x ^= (x >> 31);
+#endif
+ return x;
+}
+
+uint32_t
+hashbacktrace(uintptr_t* bt, uint32_t depth, uint32_t max_size)
+{
+
+ uintptr_t hash = 0;
+ uintptr_t mask = max_size - 1;
+
+ while (depth) {
+ hash += bt[--depth];
+ }
+
+ hash = hash_mix(hash) & mask;
+
+ assert(hash < max_size);
+
+ return (uint32_t) hash;
+}
+
+/*
+ * TODO: Determine how well distributed this is
+ * max_size must be a power of 2. i.e 0x10000 because 0x10000-1 is 0x0FFFF which is a great bitmask
+ */
+uint32_t
+hashaddr(uintptr_t pt, uint32_t max_size)
+{
+ uintptr_t hash = 0;
+ uintptr_t mask = max_size - 1;
+
+ hash = hash_mix(pt) & mask;
+
+ assert(hash < max_size);
+
+ return (uint32_t) hash;
+}
+
+/* End of all leak-detection code */
+#pragma mark -
+
+/*
+ * zinit initializes a new zone. The zone data structures themselves
+ * are stored in a zone, which is initially a static structure that
+ * is initialized by zone_init.
+ */
+zone_t
+zinit(
+ vm_size_t size, /* the size of an element */
+ vm_size_t max, /* maximum memory to use */
+ vm_size_t alloc, /* allocation size */
+ const char *name) /* a name for the zone */
+{
+ zone_t z;
+ boolean_t use_page_list = FALSE;
+
+ if (zone_zone == ZONE_NULL) {
+
+ z = (struct zone *)zdata;
+ /* special handling in zcram() because the first element is being used */
+ } else
+ z = (zone_t) zalloc(zone_zone);
+
+ if (z == ZONE_NULL)
+ return(ZONE_NULL);
+
+ /* Zone elements must fit both a next pointer and a backup pointer */
+ vm_size_t minimum_element_size = sizeof(vm_offset_t) * 2;
+ if (size < minimum_element_size)
+ size = minimum_element_size;
+
+ /*
+ * Round element size to a multiple of sizeof(pointer)
+ * This also enforces that allocations will be aligned on pointer boundaries
+ */
+ size = ((size-1) + sizeof(vm_offset_t)) -
+ ((size-1) % sizeof(vm_offset_t));
+
+ if (alloc == 0)
+ alloc = PAGE_SIZE;
+
+ alloc = round_page(alloc);
+ max = round_page(max);
+
+ /*
+ * we look for an allocation size with less than 1% waste
+ * up to 5 pages in size...
+ * otherwise, we look for an allocation size with least fragmentation
+ * in the range of 1 - 5 pages
+ * This size will be used unless
+ * the user suggestion is larger AND has less fragmentation
+ */
+#if ZONE_ALIAS_ADDR
+ /* Favor PAGE_SIZE allocations unless we waste >10% space */
+ if ((size < PAGE_SIZE) && (PAGE_SIZE % size <= PAGE_SIZE / 10))
+ alloc = PAGE_SIZE;
+ else
+#endif
+#if defined(__LP64__)
+ if (((alloc % size) != 0) || (alloc > PAGE_SIZE * 8))
+#endif
+ {
+ vm_size_t best, waste; unsigned int i;
+ best = PAGE_SIZE;
+ waste = best % size;
+
+ for (i = 1; i <= 5; i++) {
+ vm_size_t tsize, twaste;
+
+ tsize = i * PAGE_SIZE;
+
+ if ((tsize % size) < (tsize / 100)) {
+ alloc = tsize;
+ goto use_this_allocation;
+ }
+ twaste = tsize % size;
+ if (twaste < waste)
+ best = tsize, waste = twaste;
+ }
+ if (alloc <= best || (alloc % size >= waste))
+ alloc = best;
+ }
+use_this_allocation:
+ if (max && (max < alloc))
+ max = alloc;
+
+ /*
+ * Opt into page list tracking if we can reliably map an allocation
+ * to its page_metadata, and if the wastage in the tail of
+ * the allocation is not too large
+ */
+
+ /* zone_zone can't use page metadata since the page metadata will overwrite zone metadata */
+ if (alloc == PAGE_SIZE && zone_zone != ZONE_NULL) {
+ vm_offset_t first_element_offset;
+ size_t zone_page_metadata_size = sizeof(struct zone_page_metadata);
+
+ if (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT == 0) {
+ first_element_offset = zone_page_metadata_size;
+ } else {
+ first_element_offset = zone_page_metadata_size + (ZONE_ELEMENT_ALIGNMENT - (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT));
+ }
+
+ if (((PAGE_SIZE - first_element_offset) % size) <= PAGE_SIZE / 100) {
+ use_page_list = TRUE;
+ }
+ }
+
+ z->free_elements = NULL;
+ queue_init(&z->pages.any_free_foreign);
+ queue_init(&z->pages.all_free);
+ queue_init(&z->pages.intermediate);
+ queue_init(&z->pages.all_used);
+ z->cur_size = 0;
+ z->page_count = 0;
+ z->max_size = max;
+ z->elem_size = size;
+ z->alloc_size = alloc;
+ z->zone_name = name;
+ z->count = 0;
+ z->countfree = 0;
+ z->sum_count = 0LL;
+ z->doing_alloc_without_vm_priv = FALSE;
+ z->doing_alloc_with_vm_priv = FALSE;
+ z->doing_gc = FALSE;
+ z->exhaustible = FALSE;
+ z->collectable = TRUE;
+ z->allows_foreign = FALSE;
+ z->expandable = TRUE;
+ z->waiting = FALSE;
+ z->async_pending = FALSE;
+ z->caller_acct = TRUE;
+ z->noencrypt = FALSE;
+ z->no_callout = FALSE;
+ z->async_prio_refill = FALSE;
+ z->gzalloc_exempt = FALSE;
+ z->alignment_required = FALSE;
+ z->use_page_list = use_page_list;
+ z->prio_refill_watermark = 0;
+ z->zone_replenish_thread = NULL;
+ z->zp_count = 0;
+#if CONFIG_ZLEAKS
+ z->zleak_capture = 0;
+ z->zleak_on = FALSE;
+#endif /* CONFIG_ZLEAKS */
+
+#if ZONE_DEBUG
+ z->active_zones.next = z->active_zones.prev = NULL;
+ zone_debug_enable(z);
+#endif /* ZONE_DEBUG */
+ lock_zone_init(z);
+
+ /*
+ * Add the zone to the all-zones list.
+ * If we are tracking zone info per task, and we have
+ * already used all the available stat slots, then keep
+ * using the overflow zone slot.
+ */
+ z->next_zone = ZONE_NULL;
+ simple_lock(&all_zones_lock);
+ *last_zone = z;
+ last_zone = &z->next_zone;
+ z->index = num_zones;
+ if (zinfo_per_task) {
+ if (num_zones > ZONES_MAX)
+ z->index = ZONES_MAX;
+ }
+ num_zones++;
+ simple_unlock(&all_zones_lock);
+
+ /*
+ * Check if we should be logging this zone. If so, remember the zone pointer.
+ */
+ if (log_this_zone(z->zone_name, zone_name_to_log)) {
+ zone_of_interest = z;
+ }
+
+ /*
+ * If we want to log a zone, see if we need to allocate buffer space for the log. Some vm related zones are
+ * zinit'ed before we can do a kmem_alloc, so we have to defer allocation in that case. kmem_alloc_ready is set to
+ * TRUE once enough of the VM system is up and running to allow a kmem_alloc to work. If we want to log one
+ * of the VM related zones that's set up early on, we will skip allocation of the log until zinit is called again
+ * later on some other zone. So note we may be allocating a buffer to log a zone other than the one being initialized
+ * right now.
+ */
+ if (zone_of_interest != NULL && zlog_btlog == NULL && kmem_alloc_ready) {
+ zlog_btlog = btlog_create(log_records, MAX_ZTRACE_DEPTH, NULL, NULL, NULL);
+ if (zlog_btlog) {
+ printf("zone: logging started for zone %s\n", zone_of_interest->zone_name);
+ } else {
+ printf("zone: couldn't allocate memory for zrecords, turning off zleak logging\n");
+ zone_of_interest = NULL;
+ }
+ }
+#if CONFIG_GZALLOC
+ gzalloc_zone_init(z);
+#endif
+ return(z);
+}
+unsigned zone_replenish_loops, zone_replenish_wakeups, zone_replenish_wakeups_initiated, zone_replenish_throttle_count;
+
+static void zone_replenish_thread(zone_t);
+
+/* High priority VM privileged thread used to asynchronously refill a designated
+ * zone, such as the reserved VM map entry zone.
+ */
+static void zone_replenish_thread(zone_t z) {
+ vm_size_t free_size;
+ current_thread()->options |= TH_OPT_VMPRIV;
+
+ for (;;) {
+ lock_zone(z);
+ assert(z->prio_refill_watermark != 0);
+ while ((free_size = (z->cur_size - (z->count * z->elem_size))) < (z->prio_refill_watermark * z->elem_size)) {
+ assert(z->doing_alloc_without_vm_priv == FALSE);
+ assert(z->doing_alloc_with_vm_priv == FALSE);
+ assert(z->async_prio_refill == TRUE);
+
+ unlock_zone(z);
+ int zflags = KMA_KOBJECT|KMA_NOPAGEWAIT;
+ vm_offset_t space, alloc_size;
+ kern_return_t kr;
+
+ if (vm_pool_low())
+ alloc_size = round_page(z->elem_size);
+ else
+ alloc_size = z->alloc_size;
+
+ if (z->noencrypt)
+ zflags |= KMA_NOENCRYPT;
+
+ kr = kernel_memory_allocate(zone_map, &space, alloc_size, 0, zflags, VM_KERN_MEMORY_ZONE);
+
+ if (kr == KERN_SUCCESS) {
+#if ZONE_ALIAS_ADDR
+ if (alloc_size == PAGE_SIZE)
+ space = zone_alias_addr(space);
+#endif
+ zcram(z, space, alloc_size);
+ } else if (kr == KERN_RESOURCE_SHORTAGE) {
+ VM_PAGE_WAIT();
+ } else if (kr == KERN_NO_SPACE) {
+ kr = kernel_memory_allocate(kernel_map, &space, alloc_size, 0, zflags, VM_KERN_MEMORY_ZONE);
+ if (kr == KERN_SUCCESS) {
+#if ZONE_ALIAS_ADDR
+ if (alloc_size == PAGE_SIZE)
+ space = zone_alias_addr(space);
+#endif
+ zcram(z, space, alloc_size);
+ } else {
+ assert_wait_timeout(&z->zone_replenish_thread, THREAD_UNINT, 1, 100 * NSEC_PER_USEC);
+ thread_block(THREAD_CONTINUE_NULL);
+ }
+ }
+
+ lock_zone(z);
+ zone_replenish_loops++;
+ }
+
+ unlock_zone(z);
+ /* Signal any potential throttled consumers, terminating
+ * their timer-bounded waits.
+ */
+ thread_wakeup(z);
+
+ assert_wait(&z->zone_replenish_thread, THREAD_UNINT);
+ thread_block(THREAD_CONTINUE_NULL);
+ zone_replenish_wakeups++;
+ }
+}
+
+void
+zone_prio_refill_configure(zone_t z, vm_size_t low_water_mark) {
+ z->prio_refill_watermark = low_water_mark;
+
+ z->async_prio_refill = TRUE;
+ OSMemoryBarrier();
+ kern_return_t tres = kernel_thread_start_priority((thread_continue_t)zone_replenish_thread, z, MAXPRI_KERNEL, &z->zone_replenish_thread);
+
+ if (tres != KERN_SUCCESS) {
+ panic("zone_prio_refill_configure, thread create: 0x%x", tres);
+ }
+
+ thread_deallocate(z->zone_replenish_thread);
+}
+
+/*
+ * Cram the given memory into the specified zone. Update the zone page count accordingly.
+ */
+void
+zcram(
+ zone_t zone,
+ vm_offset_t newmem,
+ vm_size_t size)
+{
+ vm_size_t elem_size;
+ boolean_t from_zm = FALSE;
+
+ /* Basic sanity checks */
+ assert(zone != ZONE_NULL && newmem != (vm_offset_t)0);
+ assert(!zone->collectable || zone->allows_foreign
+ || (from_zone_map(newmem, size)));
+
+ elem_size = zone->elem_size;
+
+ if (from_zone_map(newmem, size))
+ from_zm = TRUE;
+
+ if (zalloc_debug & ZALLOC_DEBUG_ZCRAM)
+ kprintf("zcram(%p[%s], 0x%lx%s, 0x%lx)\n", zone, zone->zone_name,
+ (unsigned long)newmem, from_zm ? "" : "[F]", (unsigned long)size);
+
+ if (from_zm && !zone->use_page_list)
+ zone_page_init(newmem, size);
+
+ ZONE_PAGE_COUNT_INCR(zone, (size / PAGE_SIZE));
+
+ lock_zone(zone);
+
+ if (zone->use_page_list) {
+ struct zone_page_metadata *page_metadata;
+ size_t zone_page_metadata_size = sizeof(struct zone_page_metadata);
+
+ assert((newmem & PAGE_MASK) == 0);
+ assert((size & PAGE_MASK) == 0);
+ for (; size > 0; newmem += PAGE_SIZE, size -= PAGE_SIZE) {
+
+ vm_size_t pos_in_page;
+ page_metadata = (struct zone_page_metadata *)(newmem);
+
+ page_metadata->pages.next = NULL;
+ page_metadata->pages.prev = NULL;
+ page_metadata->elements = NULL;
+ page_metadata->zone = zone;
+ page_metadata->alloc_count = 0;
+ page_metadata->free_count = 0;
+
+ enqueue_tail(&zone->pages.all_used, (queue_entry_t)page_metadata);
+
+ vm_offset_t first_element_offset;
+ if (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT == 0){
+ first_element_offset = zone_page_metadata_size;
+ } else {
+ first_element_offset = zone_page_metadata_size + (ZONE_ELEMENT_ALIGNMENT - (zone_page_metadata_size % ZONE_ELEMENT_ALIGNMENT));
+ }
+
+ for (pos_in_page = first_element_offset; (newmem + pos_in_page + elem_size) < (vm_offset_t)(newmem + PAGE_SIZE); pos_in_page += elem_size) {
+ page_metadata->alloc_count++;
+ zone->count++; /* compensate for free_to_zone */
+ free_to_zone(zone, newmem + pos_in_page, FALSE);
+ zone->cur_size += elem_size;
+ }
+ }
+ } else {
+ while (size >= elem_size) {
+ zone->count++; /* compensate for free_to_zone */
+ if (newmem == (vm_offset_t)zone) {
+ /* Don't free zone_zone zone */
+ } else {
+ free_to_zone(zone, newmem, FALSE);
+ }
+ if (from_zm)
+ zone_page_alloc(newmem, elem_size);
+ size -= elem_size;
+ newmem += elem_size;
+ zone->cur_size += elem_size;
+ }
+ }
+ unlock_zone(zone);
+}
+
+
+/*
+ * Steal memory for the zone package. Called from
+ * vm_page_bootstrap().
+ */
+void
+zone_steal_memory(void)
+{
+#if CONFIG_GZALLOC
+ gzalloc_configure();
+#endif
+ /* Request enough early memory to get to the pmap zone */
+ zdata_size = 12 * sizeof(struct zone);
+ zdata_size = round_page(zdata_size);
+ zdata = (vm_offset_t)pmap_steal_memory(zdata_size);
projects / apple / xnu.git / blobdiff