#include <vm/vm_map.h>
#include <vm/vm_pageout.h>
#include <vm/memory_object.h>
+#include <vm/vm_compressor_algorithms.h>
+#include <vm/vm_fault.h>
#include <mach/mach_host.h> /* for host_info() */
#include <kern/ledger.h>
+#include <kern/policy_internal.h>
-#include <default_pager/default_pager_alerts.h>
-#include <default_pager/default_pager_object_server.h>
+#include <i386/misc_protos.h>
#include <IOKit/IOHibernatePrivate.h>
* the boot-arg & device-tree code.
*/
-extern ipc_port_t min_pages_trigger_port;
-extern lck_mtx_t paging_segments_lock;
-#define PSL_LOCK() lck_mtx_lock(&paging_segments_lock)
-#define PSL_UNLOCK() lck_mtx_unlock(&paging_segments_lock)
-
-
int vm_compressor_mode = VM_PAGER_COMPRESSOR_WITH_SWAP;
int vm_scale = 16;
int vm_compressor_is_active = 0;
int vm_compression_limit = 0;
+int vm_compressor_available = 0;
-extern boolean_t vm_swap_up;
extern void vm_pageout_io_throttle(void);
#if CHECKSUM_THE_DATA || CHECKSUM_THE_SWAP || CHECKSUM_THE_COMPRESSED_DATA
extern unsigned int hash_string(char *cp, int len);
#endif
+#define UNPACK_C_SIZE(cs) ((cs->c_size == (PAGE_SIZE-1)) ? PAGE_SIZE : cs->c_size)
+#define PACK_C_SIZE(cs, size) (cs->c_size = ((size == PAGE_SIZE) ? PAGE_SIZE - 1 : size))
-struct c_slot {
- uint64_t c_offset:C_SEG_OFFSET_BITS,
- c_size:12,
- c_packed_ptr:36;
-#if CHECKSUM_THE_DATA
- unsigned int c_hash_data;
-#endif
-#if CHECKSUM_THE_COMPRESSED_DATA
- unsigned int c_hash_compressed_data;
-#endif
+struct c_sv_hash_entry {
+ union {
+ struct {
+ uint32_t c_sv_he_ref;
+ uint32_t c_sv_he_data;
+ } c_sv_he;
+ uint64_t c_sv_he_record;
+
+ } c_sv_he_un;
};
-#define UNPACK_C_SIZE(cs) ((cs->c_size == (PAGE_SIZE-1)) ? PAGE_SIZE : cs->c_size)
-#define PACK_C_SIZE(cs, size) (cs->c_size = ((size == PAGE_SIZE) ? PAGE_SIZE - 1 : size))
+#define he_ref c_sv_he_un.c_sv_he.c_sv_he_ref
+#define he_data c_sv_he_un.c_sv_he.c_sv_he_data
+#define he_record c_sv_he_un.c_sv_he_record
+
+#define C_SV_HASH_MAX_MISS 32
+#define C_SV_HASH_SIZE ((1 << 10))
+#define C_SV_HASH_MASK ((1 << 10) - 1)
+#define C_SV_CSEG_ID ((1 << 22) - 1)
struct c_slot_mapping {
uint32_t s_cseg:22, /* segment number + 1 */
s_cindx:10; /* index in the segment */
};
+#define C_SLOT_MAX_INDEX (1 << 10)
typedef struct c_slot_mapping *c_slot_mapping_t;
union c_segu {
c_segment_t c_seg;
- uint32_t c_segno;
+ uintptr_t c_segno;
};
-#define C_SLOT_PACK_PTR(ptr) (((uintptr_t)ptr - (uintptr_t) VM_MIN_KERNEL_AND_KEXT_ADDRESS) >> 2)
-#define C_SLOT_UNPACK_PTR(cslot) ((uintptr_t)(cslot->c_packed_ptr << 2) + (uintptr_t) VM_MIN_KERNEL_AND_KEXT_ADDRESS)
+#define C_SLOT_PACK_PTR(ptr) (((uintptr_t)ptr - (uintptr_t) KERNEL_PMAP_HEAP_RANGE_START) >> 2)
+#define C_SLOT_UNPACK_PTR(cslot) ((uintptr_t)(cslot->c_packed_ptr << 2) + (uintptr_t) KERNEL_PMAP_HEAP_RANGE_START)
uint32_t c_segment_count = 0;
+uint32_t c_segment_count_max = 0;
uint64_t c_generation_id = 0;
uint64_t c_generation_id_flush_barrier;
clock_sec_t hibernate_flushing_deadline = 0;
-#if TRACK_BAD_C_SEGMENTS
-queue_head_t c_bad_list_head;
-uint32_t c_bad_count = 0;
+#if RECORD_THE_COMPRESSED_DATA
+char *c_compressed_record_sbuf;
+char *c_compressed_record_ebuf;
+char *c_compressed_record_cptr;
#endif
+
queue_head_t c_age_list_head;
queue_head_t c_swapout_list_head;
queue_head_t c_swappedin_list_head;
queue_head_t c_swappedout_list_head;
queue_head_t c_swappedout_sparse_list_head;
+queue_head_t c_major_list_head;
+queue_head_t c_filling_list_head;
+queue_head_t c_bad_list_head;
uint32_t c_age_count = 0;
uint32_t c_swapout_count = 0;
uint32_t c_swappedin_count = 0;
uint32_t c_swappedout_count = 0;
uint32_t c_swappedout_sparse_count = 0;
+uint32_t c_major_count = 0;
+uint32_t c_filling_count = 0;
+uint32_t c_empty_count = 0;
+uint32_t c_bad_count = 0;
+
queue_head_t c_minor_list_head;
uint32_t c_minor_count = 0;
+int c_overage_swapped_count = 0;
+int c_overage_swapped_limit = 0;
+
+int c_seg_fixed_array_len;
union c_segu *c_segments;
+vm_offset_t c_buffers;
+vm_size_t c_buffers_size;
caddr_t c_segments_next_page;
boolean_t c_segments_busy;
uint32_t c_segments_available;
uint32_t c_segments_limit;
uint32_t c_segments_nearing_limit;
+
+uint32_t c_segment_svp_in_hash;
+uint32_t c_segment_svp_hash_succeeded;
+uint32_t c_segment_svp_hash_failed;
+uint32_t c_segment_svp_zero_compressions;
+uint32_t c_segment_svp_nonzero_compressions;
+uint32_t c_segment_svp_zero_decompressions;
+uint32_t c_segment_svp_nonzero_decompressions;
+
+uint32_t c_segment_noncompressible_pages;
+
uint32_t c_segment_pages_compressed;
uint32_t c_segment_pages_compressed_limit;
uint32_t c_segment_pages_compressed_nearing_limit;
lck_grp_attr_t vm_compressor_lck_grp_attr;
lck_attr_t vm_compressor_lck_attr;
lck_grp_t vm_compressor_lck_grp;
-
-
-#if __i386__ || __x86_64__
lck_mtx_t *c_list_lock;
-#else /* __i386__ || __x86_64__ */
-lck_spin_t *c_list_lock;
-#endif /* __i386__ || __x86_64__ */
-
lck_rw_t c_master_lock;
boolean_t decompressions_blocked = FALSE;
uint32_t compressor_cpus;
char *compressor_scratch_bufs;
-
+char *kdp_compressor_scratch_buf;
+char *kdp_compressor_decompressed_page;
+addr64_t kdp_compressor_decompressed_page_paddr;
+ppnum_t kdp_compressor_decompressed_page_ppnum;
clock_sec_t start_of_sample_period_sec = 0;
clock_nsec_t start_of_sample_period_nsec = 0;
#define DECOMPRESSION_SAMPLE_MAX_AGE (60 * 30)
+boolean_t vm_swapout_ripe_segments = FALSE;
+uint32_t vm_ripe_target_age = (60 * 60 * 48);
+
uint32_t swapout_target_age = 0;
uint32_t age_of_decompressions_during_sample_period[DECOMPRESSION_SAMPLE_MAX_AGE];
uint32_t overage_decompressions_during_sample_period = 0;
int64_t c_segment_input_bytes __attribute__((aligned(8))) = 0;
int64_t c_segment_compressed_bytes __attribute__((aligned(8))) = 0;
int64_t compressor_bytes_used __attribute__((aligned(8))) = 0;
-uint64_t compressor_kvspace_used __attribute__((aligned(8))) = 0;
-uint64_t compressor_kvwaste_limit = 0;
+
+
+struct c_sv_hash_entry c_segment_sv_hash_table[C_SV_HASH_SIZE] __attribute__ ((aligned (8)));
+
static boolean_t compressor_needs_to_swap(void);
static void vm_compressor_swap_trigger_thread(void);
static void vm_compressor_compact_and_swap(boolean_t);
static void vm_compressor_age_swapped_in_segments(boolean_t);
+static void vm_compressor_take_paging_space_action(void);
+
boolean_t vm_compressor_low_on_space(void);
void compute_swapout_target_age(void);
int c_seg_minor_compaction_and_unlock(c_segment_t, boolean_t);
int c_seg_do_minor_compaction_and_unlock(c_segment_t, boolean_t, boolean_t, boolean_t);
void c_seg_try_minor_compaction_and_unlock(c_segment_t c_seg);
-void c_seg_need_delayed_compaction(c_segment_t);
void c_seg_move_to_sparse_list(c_segment_t);
void c_seg_insert_into_q(queue_head_t *, c_segment_t);
-boolean_t c_seg_try_free(c_segment_t);
-void c_seg_free(c_segment_t);
-void c_seg_free_locked(c_segment_t);
-
-
uint64_t vm_available_memory(void);
uint64_t vm_compressor_pages_compressed(void);
-extern unsigned int dp_pages_free, dp_pages_reserve;
+/*
+ * indicate the need to do a major compaction if
+ * the overall set of in-use compression segments
+ * becomes sparse... on systems that support pressure
+ * driven swapping, this will also cause swapouts to
+ * be initiated.
+ */
+static inline boolean_t vm_compressor_needs_to_major_compact()
+{
+ uint32_t incore_seg_count;
+
+ incore_seg_count = c_segment_count - c_swappedout_count - c_swappedout_sparse_count;
+
+ if ((c_segment_count >= (c_segments_nearing_limit / 8)) &&
+ ((incore_seg_count * C_SEG_MAX_PAGES) - VM_PAGE_COMPRESSOR_COUNT) >
+ ((incore_seg_count / 8) * C_SEG_MAX_PAGES))
+ return (1);
+ return (0);
+}
+
uint64_t
vm_available_memory(void)
}
-boolean_t
-vm_compression_available(void)
-{
- if ( !(COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE))
- return (FALSE);
-
- if (c_segments_available >= c_segments_limit || c_segment_pages_compressed >= c_segment_pages_compressed_limit)
- return (FALSE);
-
- return (TRUE);
-}
-
-
boolean_t
vm_compressor_low_on_space(void)
{
if (task == kernel_task)
return (0);
- if (vm_compressor_mode == COMPRESSED_PAGER_IS_ACTIVE || vm_compressor_mode == DEFAULT_FREEZER_COMPRESSED_PAGER_IS_ACTIVE) {
+ if (VM_CONFIG_SWAP_IS_ACTIVE) {
if ((vm_compressor_low_on_space() || HARD_THROTTLE_LIMIT_REACHED()) &&
(unsigned int)pmap_compressed(task->map->pmap) > (c_segment_pages_compressed / 4))
return (1);
- } else {
- if (((dp_pages_free + dp_pages_reserve < 2000) && VM_DYNAMIC_PAGING_ENABLED(memory_manager_default)) &&
- get_task_resident_size(task) > (((AVAILABLE_NON_COMPRESSED_MEMORY) * PAGE_SIZE) / 5))
- return (1);
}
return (0);
}
+
+static uint32_t no_paging_space_action_in_progress = 0;
+extern void memorystatus_send_low_swap_note(void);
+
+static void
+vm_compressor_take_paging_space_action(void)
+{
+ if (no_paging_space_action_in_progress == 0) {
+
+ if (OSCompareAndSwap(0, 1, (UInt32 *)&no_paging_space_action_in_progress)) {
+
+ if (no_paging_space_action()) {
+ memorystatus_send_low_swap_note();
+ }
+
+ no_paging_space_action_in_progress = 0;
+ }
+ }
+}
+
+
void
vm_compressor_init_locks(void)
{
thread_wakeup((event_t)&decompressions_blocked);
}
+static inline void cslot_copy(c_slot_t cdst, c_slot_t csrc) {
+#if CHECKSUM_THE_DATA
+ cdst->c_hash_data = csrc->c_hash_data;
+#endif
+#if CHECKSUM_THE_COMPRESSED_DATA
+ cdst->c_hash_compressed_data = csrc->c_hash_compressed_data;
+#endif
+ cdst->c_size = csrc->c_size;
+ cdst->c_packed_ptr = csrc->c_packed_ptr;
+}
+vm_map_t compressor_map;
void
vm_compressor_init(void)
thread_t thread;
struct c_slot cs_dummy;
c_slot_t cs = &cs_dummy;
+ int c_segment_min_size;
+ int c_segment_padded_size;
+ kern_return_t retval = KERN_SUCCESS;
+ vm_offset_t start_addr = 0;
+ vm_size_t c_segments_arr_size = 0, compressor_submap_size = 0;
+#if RECORD_THE_COMPRESSED_DATA
+ vm_size_t c_compressed_record_sbuf_size = 0;
+#endif /* RECORD_THE_COMPRESSED_DATA */
/*
* ensure that any pointer that gets created from
* use PAGE_REPLACEMENT_ALLOWED to coordinate with the compressor.
*/
-#if __i386__ || __x86_64__
c_list_lock = lck_mtx_alloc_init(&vm_compressor_lck_grp, &vm_compressor_lck_attr);
-#else /* __i386__ || __x86_64__ */
- c_list_lock = lck_spin_alloc_init(&vm_compressor_lck_grp, &vm_compressor_lck_attr);
-#endif /* __i386__ || __x86_64__ */
-#if TRACK_BAD_C_SEGMENTS
queue_init(&c_bad_list_head);
-#endif
queue_init(&c_age_list_head);
queue_init(&c_minor_list_head);
+ queue_init(&c_major_list_head);
+ queue_init(&c_filling_list_head);
queue_init(&c_swapout_list_head);
queue_init(&c_swappedin_list_head);
queue_init(&c_swappedout_list_head);
queue_init(&c_swappedout_sparse_list_head);
- compressor_segment_zone = zinit(sizeof (struct c_segment),
- 128000 * sizeof (struct c_segment),
- 8192, "compressor_segment");
+ c_segment_min_size = sizeof(struct c_segment) + (C_SEG_SLOT_VAR_ARRAY_MIN_LEN * sizeof(struct c_slot));
+
+ for (c_segment_padded_size = 128; c_segment_padded_size < c_segment_min_size; c_segment_padded_size = c_segment_padded_size << 1);
+
+ compressor_segment_zone = zinit(c_segment_padded_size, 128000 * c_segment_padded_size, PAGE_SIZE, "compressor_segment");
zone_change(compressor_segment_zone, Z_CALLERACCT, FALSE);
zone_change(compressor_segment_zone, Z_NOENCRYPT, TRUE);
+ c_seg_fixed_array_len = (c_segment_padded_size - sizeof(struct c_segment)) / sizeof(struct c_slot);
c_free_segno_head = -1;
c_segments_available = 0;
c_segment_pages_compressed_nearing_limit = (c_segment_pages_compressed_limit * 98) / 100;
c_segments_nearing_limit = (c_segments_limit * 98) / 100;
- compressor_kvwaste_limit = (vm_map_max(kernel_map) - vm_map_min(kernel_map)) / 16;
-
c_segments_busy = FALSE;
- if (kernel_memory_allocate(kernel_map, (vm_offset_t *)(&c_segments), (sizeof(union c_segu) * c_segments_limit), 0, KMA_KOBJECT | KMA_VAONLY) != KERN_SUCCESS)
- panic("vm_compressor_init: kernel_memory_allocate failed\n");
+ /*
+ * Submap needs space for:
+ * - c_segments
+ * - c_buffers
+ * - swap reclaimations -- C_SEG_BUFSIZE
+ */
+ c_segments_arr_size = vm_map_round_page((sizeof(union c_segu) * c_segments_limit),VM_MAP_PAGE_MASK(kernel_map));
+ c_buffers_size = vm_map_round_page(((vm_size_t)C_SEG_ALLOCSIZE * (vm_size_t)c_segments_limit), VM_MAP_PAGE_MASK(kernel_map));
+
+ compressor_submap_size = c_segments_arr_size + c_buffers_size + C_SEG_BUFSIZE;
+
+#if RECORD_THE_COMPRESSED_DATA
+ c_compressed_record_sbuf_size = (vm_size_t)C_SEG_ALLOCSIZE + (PAGE_SIZE * 2);
+ compressor_submap_size += c_compressed_record_sbuf_size;
+#endif /* RECORD_THE_COMPRESSED_DATA */
+
+ retval = kmem_suballoc(kernel_map, &start_addr, compressor_submap_size,
+ FALSE, VM_FLAGS_ANYWHERE | VM_FLAGS_PERMANENT | VM_MAKE_TAG(0),
+ &compressor_map);
+
+ if (retval != KERN_SUCCESS)
+ panic("vm_compressor_init: kmem_suballoc failed");
+
+ if (kernel_memory_allocate(compressor_map, (vm_offset_t *)(&c_segments), (sizeof(union c_segu) * c_segments_limit), 0, KMA_KOBJECT | KMA_VAONLY | KMA_PERMANENT, VM_KERN_MEMORY_COMPRESSOR) != KERN_SUCCESS)
+ panic("vm_compressor_init: kernel_memory_allocate failed - c_segments\n");
+ if (kernel_memory_allocate(compressor_map, &c_buffers, c_buffers_size, 0, KMA_COMPRESSOR | KMA_VAONLY | KMA_PERMANENT, VM_KERN_MEMORY_COMPRESSOR) != KERN_SUCCESS)
+ panic("vm_compressor_init: kernel_memory_allocate failed - c_buffers\n");
c_segments_next_page = (caddr_t)c_segments;
+ vm_compressor_algorithm_init();
{
host_basic_info_data_t hinfo;
host_info((host_t)BSD_HOST, HOST_BASIC_INFO, (host_info_t)&hinfo, &count);
compressor_cpus = hinfo.max_cpus;
+ compressor_scratch_bufs = kalloc_tag(compressor_cpus * vm_compressor_get_decode_scratch_size(), VM_KERN_MEMORY_COMPRESSOR);
- compressor_scratch_bufs = kalloc(compressor_cpus * WKdm_SCRATCH_BUF_SIZE);
+ kdp_compressor_scratch_buf = kalloc_tag(vm_compressor_get_decode_scratch_size(), VM_KERN_MEMORY_COMPRESSOR);
+ kdp_compressor_decompressed_page = kalloc_tag(PAGE_SIZE, VM_KERN_MEMORY_COMPRESSOR);
+ kdp_compressor_decompressed_page_paddr = kvtophys((vm_offset_t)kdp_compressor_decompressed_page);
+ kdp_compressor_decompressed_page_ppnum = (ppnum_t) atop(kdp_compressor_decompressed_page_paddr);
}
+#if CONFIG_FREEZE
+ freezer_compressor_scratch_buf = kalloc_tag(vm_compressor_get_encode_scratch_size(), VM_KERN_MEMORY_COMPRESSOR);
+#endif
+
+#if RECORD_THE_COMPRESSED_DATA
+ if (kernel_memory_allocate(compressor_map, (vm_offset_t *)&c_compressed_record_sbuf, c_compressed_record_sbuf_size, 0, KMA_KOBJECT, VM_KERN_MEMORY_COMPRESSOR) != KERN_SUCCESS)
+ panic("vm_compressor_init: kernel_memory_allocate failed - c_compressed_record_sbuf\n");
+
+ c_compressed_record_cptr = c_compressed_record_sbuf;
+ c_compressed_record_ebuf = c_compressed_record_sbuf + c_compressed_record_sbuf_size;
+#endif
if (kernel_thread_start_priority((thread_continue_t)vm_compressor_swap_trigger_thread, NULL,
BASEPRI_PREEMPT - 1, &thread) != KERN_SUCCESS) {
panic("vm_compressor_swap_trigger_thread: create failed");
}
- thread->options |= TH_OPT_VMPRIV;
-
thread_deallocate(thread);
- assert(default_pager_init_flag == 0);
-
if (vm_pageout_internal_start() != KERN_SUCCESS) {
panic("vm_compressor_init: Failed to start the internal pageout thread.\n");
}
-
- if ((vm_compressor_mode == VM_PAGER_COMPRESSOR_WITH_SWAP) ||
- (vm_compressor_mode == VM_PAGER_FREEZER_COMPRESSOR_WITH_SWAP)) {
+ if (VM_CONFIG_SWAP_IS_PRESENT)
vm_compressor_swap_init();
- }
- if (COMPRESSED_PAGER_IS_ACTIVE || DEFAULT_FREEZER_COMPRESSED_PAGER_IS_SWAPBACKED)
+ if (VM_CONFIG_COMPRESSOR_IS_ACTIVE)
vm_compressor_is_active = 1;
#if CONFIG_FREEZE
memorystatus_freeze_enabled = TRUE;
#endif /* CONFIG_FREEZE */
- default_pager_init_flag = 1;
+ vm_compressor_available = 1;
vm_page_reactivate_all_throttled();
}
void
-c_seg_need_delayed_compaction(c_segment_t c_seg)
+c_seg_need_delayed_compaction(c_segment_t c_seg, boolean_t c_list_lock_held)
{
boolean_t clear_busy = FALSE;
- if ( !lck_mtx_try_lock_spin_always(c_list_lock)) {
- C_SEG_BUSY(c_seg);
+ if (c_list_lock_held == FALSE) {
+ if ( !lck_mtx_try_lock_spin_always(c_list_lock)) {
+ C_SEG_BUSY(c_seg);
- lck_mtx_unlock_always(&c_seg->c_lock);
- lck_mtx_lock_spin_always(c_list_lock);
- lck_mtx_lock_spin_always(&c_seg->c_lock);
+ lck_mtx_unlock_always(&c_seg->c_lock);
+ lck_mtx_lock_spin_always(c_list_lock);
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
- clear_busy = TRUE;
+ clear_busy = TRUE;
+ }
}
- if (!c_seg->c_on_minorcompact_q && !c_seg->c_ondisk && !c_seg->c_on_swapout_q) {
+ assert(c_seg->c_state != C_IS_FILLING);
+
+ if (!c_seg->c_on_minorcompact_q && !(C_SEG_IS_ONDISK(c_seg))) {
queue_enter(&c_minor_list_head, c_seg, c_segment_t, c_list);
c_seg->c_on_minorcompact_q = 1;
c_minor_count++;
}
- lck_mtx_unlock_always(c_list_lock);
+ if (c_list_lock_held == FALSE)
+ lck_mtx_unlock_always(c_list_lock);
if (clear_busy == TRUE)
C_SEG_WAKEUP_DONE(c_seg);
clear_busy = TRUE;
}
- assert(c_seg->c_ondisk);
- assert(c_seg->c_on_swappedout_q);
- assert(!c_seg->c_on_swappedout_sparse_q);
-
- queue_remove(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedout_q = 0;
- c_swappedout_count--;
-
- c_seg_insert_into_q(&c_swappedout_sparse_list_head, c_seg);
- c_seg->c_on_swappedout_sparse_q = 1;
- c_swappedout_sparse_count++;
+ c_seg_switch_state(c_seg, C_ON_SWAPPEDOUTSPARSE_Q, FALSE);
c_seg_moved_to_sparse_list++;
assert(c_seg->c_busy);
+ /*
+ * check for the case that can occur when we are not swapping
+ * and this segment has been major compacted in the past
+ * and moved to the majorcompact q to remove it from further
+ * consideration... if the occupancy falls too low we need
+ * to put it back on the age_q so that it will be considered
+ * in the next major compaction sweep... if we don't do this
+ * we will eventually run into the c_segments_limit
+ */
+ if (c_seg->c_state == C_ON_MAJORCOMPACT_Q && C_SEG_SHOULD_MAJORCOMPACT(c_seg)) {
+
+ c_seg_switch_state(c_seg, C_ON_AGE_Q, FALSE);
+ }
if (!c_seg->c_on_minorcompact_q) {
if (clear_busy == TRUE)
C_SEG_WAKEUP_DONE(c_seg);
}
+void
+c_seg_switch_state(c_segment_t c_seg, int new_state, boolean_t insert_head)
+{
+ int old_state = c_seg->c_state;
-int try_free_succeeded = 0;
-int try_free_failed = 0;
+#if __i386__ || __x86_64__
+ if (new_state != C_IS_FILLING)
+ LCK_MTX_ASSERT(&c_seg->c_lock, LCK_MTX_ASSERT_OWNED);
+ LCK_MTX_ASSERT(c_list_lock, LCK_MTX_ASSERT_OWNED);
+#endif
+ switch (old_state) {
-boolean_t
-c_seg_try_free(c_segment_t c_seg)
-{
- /*
- * c_seg is currently on the delayed minor compaction
- * or the spapped out sparse queue and we have c_seg locked...
- * if we can get the c_list_lock w/o blocking (if we blocked we
- * could deadlock because the lock order is c_list_lock then c_seg's lock)
- * we'll pull it from the appropriate queue and free it
- */
- if ( !lck_mtx_try_lock_spin_always(c_list_lock)) {
- /*
- * c_list_lock is held, we need to bail
- */
- try_free_failed++;
- return (FALSE);
- }
- if (c_seg->c_on_minorcompact_q) {
- queue_remove(&c_minor_list_head, c_seg, c_segment_t, c_list);
- c_seg->c_on_minorcompact_q = 0;
- c_minor_count--;
- } else {
- assert(c_seg->c_on_swappedout_sparse_q);
+ case C_IS_EMPTY:
+ assert(new_state == C_IS_FILLING || new_state == C_IS_FREE);
- /*
- * c_seg_free_locked will remove it from the swappedout sparse list
- */
+ c_empty_count--;
+ break;
+
+ case C_IS_FILLING:
+ assert(new_state == C_ON_AGE_Q || new_state == C_ON_SWAPOUT_Q);
+
+ queue_remove(&c_filling_list_head, c_seg, c_segment_t, c_age_list);
+ c_filling_count--;
+ break;
+
+ case C_ON_AGE_Q:
+ assert(new_state == C_ON_SWAPOUT_Q || new_state == C_ON_MAJORCOMPACT_Q ||
+ new_state == C_IS_FREE);
+
+ queue_remove(&c_age_list_head, c_seg, c_segment_t, c_age_list);
+ c_age_count--;
+ break;
+
+ case C_ON_SWAPPEDIN_Q:
+ assert(new_state == C_ON_AGE_Q || new_state == C_IS_FREE);
+
+ queue_remove(&c_swappedin_list_head, c_seg, c_segment_t, c_age_list);
+ c_swappedin_count--;
+ break;
+
+ case C_ON_SWAPOUT_Q:
+ assert(new_state == C_ON_SWAPPEDOUT_Q || new_state == C_ON_SWAPPEDOUTSPARSE_Q ||
+ new_state == C_ON_AGE_Q || new_state == C_IS_FREE || new_state == C_IS_EMPTY);
+
+ queue_remove(&c_swapout_list_head, c_seg, c_segment_t, c_age_list);
+ thread_wakeup((event_t)&compaction_swapper_running);
+ c_swapout_count--;
+ break;
+
+ case C_ON_SWAPPEDOUT_Q:
+ assert(new_state == C_ON_SWAPPEDIN_Q || new_state == C_ON_AGE_Q ||
+ new_state == C_ON_SWAPPEDOUTSPARSE_Q ||
+ new_state == C_ON_BAD_Q || new_state == C_IS_EMPTY || new_state == C_IS_FREE);
+
+ queue_remove(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
+ c_swappedout_count--;
+ break;
+
+ case C_ON_SWAPPEDOUTSPARSE_Q:
+ assert(new_state == C_ON_SWAPPEDIN_Q || new_state == C_ON_AGE_Q ||
+ new_state == C_ON_BAD_Q || new_state == C_IS_EMPTY || new_state == C_IS_FREE);
+
+ queue_remove(&c_swappedout_sparse_list_head, c_seg, c_segment_t, c_age_list);
+ c_swappedout_sparse_count--;
+ break;
+
+ case C_ON_MAJORCOMPACT_Q:
+ assert(new_state == C_ON_AGE_Q || new_state == C_IS_FREE);
+
+ queue_remove(&c_major_list_head, c_seg, c_segment_t, c_age_list);
+ c_major_count--;
+ break;
+
+ case C_ON_BAD_Q:
+ assert(new_state == C_IS_FREE);
+
+ queue_remove(&c_bad_list_head, c_seg, c_segment_t, c_age_list);
+ c_bad_count--;
+ break;
+
+ default:
+ panic("c_seg %p has bad c_state = %d\n", c_seg, old_state);
}
- if (!c_seg->c_busy_swapping)
- C_SEG_BUSY(c_seg);
- c_seg_free_locked(c_seg);
+ switch(new_state) {
+ case C_IS_FREE:
+ assert(old_state != C_IS_FILLING);
- try_free_succeeded++;
+ break;
- return (TRUE);
+ case C_IS_EMPTY:
+ assert(old_state == C_ON_SWAPOUT_Q || old_state == C_ON_SWAPPEDOUT_Q || old_state == C_ON_SWAPPEDOUTSPARSE_Q);
+
+ c_empty_count++;
+ break;
+
+ case C_IS_FILLING:
+ assert(old_state == C_IS_EMPTY);
+
+ queue_enter(&c_filling_list_head, c_seg, c_segment_t, c_age_list);
+ c_filling_count++;
+ break;
+
+ case C_ON_AGE_Q:
+ assert(old_state == C_IS_FILLING || old_state == C_ON_SWAPPEDIN_Q || old_state == C_ON_SWAPOUT_Q ||
+ old_state == C_ON_MAJORCOMPACT_Q || old_state == C_ON_SWAPPEDOUT_Q || old_state == C_ON_SWAPPEDOUTSPARSE_Q);
+
+ if (old_state == C_IS_FILLING)
+ queue_enter(&c_age_list_head, c_seg, c_segment_t, c_age_list);
+ else {
+ if (!queue_empty(&c_age_list_head)) {
+ c_segment_t c_first;
+
+ c_first = (c_segment_t)queue_first(&c_age_list_head);
+ c_seg->c_creation_ts = c_first->c_creation_ts;
+ }
+ queue_enter_first(&c_age_list_head, c_seg, c_segment_t, c_age_list);
+ }
+ c_age_count++;
+ break;
+
+ case C_ON_SWAPPEDIN_Q:
+ assert(c_seg->c_state == C_ON_SWAPPEDOUT_Q || c_seg->c_state == C_ON_SWAPPEDOUTSPARSE_Q);
+
+ if (insert_head == TRUE)
+ queue_enter_first(&c_swappedin_list_head, c_seg, c_segment_t, c_age_list);
+ else
+ queue_enter(&c_swappedin_list_head, c_seg, c_segment_t, c_age_list);
+ c_swappedin_count++;
+ break;
+
+ case C_ON_SWAPOUT_Q:
+ assert(old_state == C_ON_AGE_Q || old_state == C_IS_FILLING);
+
+ if (insert_head == TRUE)
+ queue_enter_first(&c_swapout_list_head, c_seg, c_segment_t, c_age_list);
+ else
+ queue_enter(&c_swapout_list_head, c_seg, c_segment_t, c_age_list);
+ c_swapout_count++;
+ break;
+
+ case C_ON_SWAPPEDOUT_Q:
+ assert(c_seg->c_state == C_ON_SWAPOUT_Q);
+
+ if (insert_head == TRUE)
+ queue_enter_first(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
+ else
+ queue_enter(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
+ c_swappedout_count++;
+ break;
+
+ case C_ON_SWAPPEDOUTSPARSE_Q:
+ assert(c_seg->c_state == C_ON_SWAPOUT_Q || c_seg->c_state == C_ON_SWAPPEDOUT_Q);
+
+ if (insert_head == TRUE)
+ queue_enter_first(&c_swappedout_sparse_list_head, c_seg, c_segment_t, c_age_list);
+ else
+ queue_enter(&c_swappedout_sparse_list_head, c_seg, c_segment_t, c_age_list);
+
+ c_swappedout_sparse_count++;
+ break;
+
+ case C_ON_MAJORCOMPACT_Q:
+ assert(c_seg->c_state == C_ON_AGE_Q);
+
+ if (insert_head == TRUE)
+ queue_enter_first(&c_major_list_head, c_seg, c_segment_t, c_age_list);
+ else
+ queue_enter(&c_major_list_head, c_seg, c_segment_t, c_age_list);
+ c_major_count++;
+ break;
+
+ case C_ON_BAD_Q:
+ assert(c_seg->c_state == C_ON_SWAPPEDOUT_Q || c_seg->c_state == C_ON_SWAPPEDOUTSPARSE_Q);
+
+ if (insert_head == TRUE)
+ queue_enter_first(&c_bad_list_head, c_seg, c_segment_t, c_age_list);
+ else
+ queue_enter(&c_bad_list_head, c_seg, c_segment_t, c_age_list);
+ c_bad_count++;
+ break;
+
+ default:
+ panic("c_seg %p requesting bad c_state = %d\n", c_seg, new_state);
+ }
+ c_seg->c_state = new_state;
}
+
void
c_seg_free(c_segment_t c_seg)
{
void
c_seg_free_locked(c_segment_t c_seg)
{
- int segno, i;
+ int segno;
int pages_populated = 0;
int32_t *c_buffer = NULL;
uint64_t c_swap_handle = 0;
+ assert(c_seg->c_busy);
assert(!c_seg->c_on_minorcompact_q);
+ assert(!c_seg->c_busy_swapping);
- if (c_seg->c_on_age_q) {
- queue_remove(&c_age_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_age_q = 0;
- c_age_count--;
- } else if (c_seg->c_on_swappedin_q) {
- queue_remove(&c_swappedin_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedin_q = 0;
- c_swappedin_count--;
- } else if (c_seg->c_on_swapout_q) {
- queue_remove(&c_swapout_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swapout_q = 0;
- c_swapout_count--;
- thread_wakeup((event_t)&compaction_swapper_running);
- } else if (c_seg->c_on_swappedout_q) {
- queue_remove(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedout_q = 0;
- c_swappedout_count--;
- } else if (c_seg->c_on_swappedout_sparse_q) {
- queue_remove(&c_swappedout_sparse_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedout_sparse_q = 0;
- c_swappedout_sparse_count--;
- }
-#if TRACK_BAD_C_SEGMENTS
- else if (c_seg->c_on_bad_q) {
- queue_remove(&c_bad_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_bad_q = 0;
- c_bad_count--;
- }
-#endif
- segno = c_seg->c_mysegno;
- c_segments[segno].c_segno = c_free_segno_head;
- c_free_segno_head = segno;
- c_segment_count--;
+ if (c_seg->c_overage_swap == TRUE) {
+ c_overage_swapped_count--;
+ c_seg->c_overage_swap = FALSE;
+ }
+ if ( !(C_SEG_IS_ONDISK(c_seg)))
+ c_buffer = c_seg->c_store.c_buffer;
+ else
+ c_swap_handle = c_seg->c_store.c_swap_handle;
- lck_mtx_unlock_always(c_list_lock);
+ c_seg_switch_state(c_seg, C_IS_FREE, FALSE);
- if (c_seg->c_wanted) {
- thread_wakeup((event_t) (c_seg));
- c_seg->c_wanted = 0;
- }
- if (c_seg->c_busy_swapping) {
- c_seg->c_must_free = 1;
+ lck_mtx_unlock_always(c_list_lock);
- lck_mtx_unlock_always(&c_seg->c_lock);
- return;
- }
- if (c_seg->c_ondisk == 0) {
+ if (c_buffer) {
pages_populated = (round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset))) / PAGE_SIZE;
-
- c_buffer = c_seg->c_store.c_buffer;
c_seg->c_store.c_buffer = NULL;
- } else {
- /*
- * Free swap space on disk.
- */
- c_swap_handle = c_seg->c_store.c_swap_handle;
+ } else
c_seg->c_store.c_swap_handle = (uint64_t)-1;
- }
+
lck_mtx_unlock_always(&c_seg->c_lock);
if (c_buffer) {
if (pages_populated)
- kernel_memory_depopulate(kernel_map, (vm_offset_t) c_buffer, pages_populated * PAGE_SIZE, KMA_COMPRESSOR);
+ kernel_memory_depopulate(compressor_map, (vm_offset_t) c_buffer, pages_populated * PAGE_SIZE, KMA_COMPRESSOR);
- kmem_free(kernel_map, (vm_offset_t) c_buffer, C_SEG_ALLOCSIZE);
- OSAddAtomic64(-C_SEG_ALLOCSIZE, &compressor_kvspace_used);
-
- } else if (c_swap_handle)
+ } else if (c_swap_handle) {
+ /*
+ * Free swap space on disk.
+ */
vm_swap_free(c_swap_handle);
+ }
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
+ C_SEG_WAKEUP_DONE(c_seg);
+ lck_mtx_unlock_always(&c_seg->c_lock);
+
+ segno = c_seg->c_mysegno;
+
+ lck_mtx_lock_spin_always(c_list_lock);
+ /*
+ * because the c_buffer is now associated with the segno,
+ * we can't put the segno back on the free list until
+ * after we have depopulated the c_buffer range, or
+ * we run the risk of depopulating a range that is
+ * now being used in one of the compressor heads
+ */
+ c_segments[segno].c_segno = c_free_segno_head;
+ c_free_segno_head = segno;
+ c_segment_count--;
+
+ lck_mtx_unlock_always(c_list_lock);
-#if __i386__ || __x86_64__
lck_mtx_destroy(&c_seg->c_lock, &vm_compressor_lck_grp);
-#else /* __i386__ || __x86_64__ */
- lck_spin_destroy(&c_seg->c_lock, &vm_compressor_lck_grp);
-#endif /* __i386__ || __x86_64__ */
- for (i = 0; i < C_SEG_SLOT_ARRAYS; i++) {
- if (c_seg->c_slots[i] == 0)
- break;
+ if (c_seg->c_slot_var_array_len)
+ kfree(c_seg->c_slot_var_array, sizeof(struct c_slot) * c_seg->c_slot_var_array_len);
- kfree((char *)c_seg->c_slots[i], sizeof(struct c_slot) * C_SEG_SLOT_ARRAY_SIZE);
- }
zfree(compressor_segment_zone, c_seg);
}
int i;
c_slot_t c_dst;
c_slot_t c_src;
- boolean_t need_unlock = TRUE;
assert(c_seg->c_busy);
c_seg_free(c_seg);
return (1);
}
+ lck_mtx_unlock_always(&c_seg->c_lock);
+
if (c_seg->c_firstemptyslot >= c_seg->c_nextslot || C_SEG_UNUSED_BYTES(c_seg) < PAGE_SIZE)
goto done;
+#if DEVELOPMENT || DEBUG
+ C_SEG_MAKE_WRITEABLE(c_seg);
+#endif
+
#if VALIDATE_C_SEGMENTS
c_seg->c_was_minor_compacted++;
#endif
if (c_size == 0)
continue;
- memcpy(&c_seg->c_store.c_buffer[c_offset], &c_seg->c_store.c_buffer[c_src->c_offset], c_size);
+ c_rounded_size = (c_size + C_SEG_OFFSET_ALIGNMENT_MASK) & ~C_SEG_OFFSET_ALIGNMENT_MASK;
-#if CHECKSUM_THE_DATA
- c_dst->c_hash_data = c_src->c_hash_data;
-#endif
-#if CHECKSUM_THE_COMPRESSED_DATA
- c_dst->c_hash_compressed_data = c_src->c_hash_compressed_data;
-#endif
- c_dst->c_size = c_src->c_size;
- c_dst->c_packed_ptr = c_src->c_packed_ptr;
+ memcpy(&c_seg->c_store.c_buffer[c_offset], &c_seg->c_store.c_buffer[c_src->c_offset], c_rounded_size);
+
+ cslot_copy(c_dst, c_src);
c_dst->c_offset = c_offset;
slot_ptr = (c_slot_mapping_t)C_SLOT_UNPACK_PTR(c_dst);
slot_ptr->s_cindx = c_indx;
- c_rounded_size = (c_size + C_SEG_OFFSET_ALIGNMENT_MASK) & ~C_SEG_OFFSET_ALIGNMENT_MASK;
-
c_offset += C_SEG_BYTES_TO_OFFSET(c_rounded_size);
PACK_C_SIZE(c_src, 0);
c_indx++;
#if VALIDATE_C_SEGMENTS
c_seg_validate(c_seg, TRUE);
#endif
-
if (old_populated_offset > c_seg->c_populated_offset) {
uint32_t gc_size;
int32_t *gc_ptr;
gc_size = C_SEG_OFFSET_TO_BYTES(old_populated_offset - c_seg->c_populated_offset);
gc_ptr = &c_seg->c_store.c_buffer[c_seg->c_populated_offset];
+ kernel_memory_depopulate(compressor_map, (vm_offset_t)gc_ptr, gc_size, KMA_COMPRESSOR);
+ }
+
+#if DEVELOPMENT || DEBUG
+ C_SEG_WRITE_PROTECT(c_seg);
+#endif
+
+done:
+ if (clear_busy == TRUE) {
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
+ C_SEG_WAKEUP_DONE(c_seg);
lck_mtx_unlock_always(&c_seg->c_lock);
+ }
+ return (0);
+}
- kernel_memory_depopulate(kernel_map, (vm_offset_t)gc_ptr, gc_size, KMA_COMPRESSOR);
- if (clear_busy == TRUE)
- lck_mtx_lock_spin_always(&c_seg->c_lock);
+static void
+c_seg_alloc_nextslot(c_segment_t c_seg)
+{
+ struct c_slot *old_slot_array = NULL;
+ struct c_slot *new_slot_array = NULL;
+ int newlen;
+ int oldlen;
+
+ if (c_seg->c_nextslot < c_seg_fixed_array_len)
+ return;
+
+ if ((c_seg->c_nextslot - c_seg_fixed_array_len) >= c_seg->c_slot_var_array_len) {
+
+ oldlen = c_seg->c_slot_var_array_len;
+ old_slot_array = c_seg->c_slot_var_array;
+
+ if (oldlen == 0)
+ newlen = C_SEG_SLOT_VAR_ARRAY_MIN_LEN;
else
- need_unlock = FALSE;
- }
-done:
- if (need_unlock == TRUE) {
- if (clear_busy == TRUE)
- C_SEG_WAKEUP_DONE(c_seg);
+ newlen = oldlen * 2;
+
+ new_slot_array = (struct c_slot *)kalloc(sizeof(struct c_slot) * newlen);
+
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
+
+ if (old_slot_array)
+ memcpy((char *)new_slot_array, (char *)old_slot_array, sizeof(struct c_slot) * oldlen);
+
+ c_seg->c_slot_var_array_len = newlen;
+ c_seg->c_slot_var_array = new_slot_array;
lck_mtx_unlock_always(&c_seg->c_lock);
+
+ if (old_slot_array)
+ kfree(old_slot_array, sizeof(struct c_slot) * oldlen);
}
- return (0);
}
uint64_t moved_bytes;
uint64_t wasted_space_in_swapouts;
uint64_t count_of_swapouts;
+ uint64_t count_of_freed_segs;
} c_seg_major_compact_stats;
c_seg_major_compact_stats.asked_permission++;
- if (c_seg_src->c_filling) {
- /*
- * we're at or near the head... don't compact
- */
- return (FALSE);
- }
if (c_seg_src->c_bytes_used >= C_MAJOR_COMPACTION_SIZE_APPROPRIATE &&
c_seg_dst->c_bytes_used >= C_MAJOR_COMPACTION_SIZE_APPROPRIATE)
return (FALSE);
- if (c_seg_dst->c_nextoffset >= C_SEG_OFF_LIMIT || c_seg_dst->c_nextslot >= C_SLOT_MAX) {
+ if (c_seg_dst->c_nextoffset >= C_SEG_OFF_LIMIT || c_seg_dst->c_nextslot >= C_SLOT_MAX_INDEX) {
/*
* destination segment is full... can't compact
*/
int i;
c_slot_t c_dst;
c_slot_t c_src;
- int slotarray;
boolean_t keep_compacting = TRUE;
/*
* from c_seg_src to c_seg_dst and update both c_segment's
* state w/o holding the master lock
*/
+#if DEVELOPMENT || DEBUG
+ C_SEG_MAKE_WRITEABLE(c_seg_dst);
+#endif
#if VALIDATE_C_SEGMENTS
c_seg_dst->c_was_major_compacted++;
}
if (C_SEG_OFFSET_TO_BYTES(c_seg_dst->c_populated_offset - c_seg_dst->c_nextoffset) < (unsigned) c_size) {
+ int size_to_populate;
+
/* doesn't fit */
- if ((C_SEG_OFFSET_TO_BYTES(c_seg_dst->c_populated_offset) == C_SEG_BUFSIZE)) {
+ size_to_populate = C_SEG_BUFSIZE - C_SEG_OFFSET_TO_BYTES(c_seg_dst->c_populated_offset);
+
+ if (size_to_populate == 0) {
/* can't fit */
keep_compacting = FALSE;
break;
}
- kernel_memory_populate(kernel_map,
+ if (size_to_populate > C_SEG_MAX_POPULATE_SIZE)
+ size_to_populate = C_SEG_MAX_POPULATE_SIZE;
+
+ kernel_memory_populate(compressor_map,
(vm_offset_t) &c_seg_dst->c_store.c_buffer[c_seg_dst->c_populated_offset],
- PAGE_SIZE,
- KMA_COMPRESSOR);
+ size_to_populate,
+ KMA_COMPRESSOR,
+ VM_KERN_MEMORY_COMPRESSOR);
- c_seg_dst->c_populated_offset += C_SEG_BYTES_TO_OFFSET(PAGE_SIZE);
+ c_seg_dst->c_populated_offset += C_SEG_BYTES_TO_OFFSET(size_to_populate);
assert(C_SEG_OFFSET_TO_BYTES(c_seg_dst->c_populated_offset) <= C_SEG_BUFSIZE);
}
+ c_seg_alloc_nextslot(c_seg_dst);
- slotarray = C_SEG_SLOTARRAY_FROM_INDEX(c_seg_dst, c_seg_dst->c_nextslot);
-
- if (c_seg_dst->c_slots[slotarray] == 0) {
- KERNEL_DEBUG(0xe0400008 | DBG_FUNC_START, 0, 0, 0, 0, 0);
- c_seg_dst->c_slots[slotarray] = (struct c_slot *)
- kalloc(sizeof(struct c_slot) *
- C_SEG_SLOT_ARRAY_SIZE);
- KERNEL_DEBUG(0xe0400008 | DBG_FUNC_END, 0, 0, 0, 0, 0);
- }
c_dst = C_SEG_SLOT_FROM_INDEX(c_seg_dst, c_seg_dst->c_nextslot);
memcpy(&c_seg_dst->c_store.c_buffer[c_seg_dst->c_nextoffset], &c_seg_src->c_store.c_buffer[c_src->c_offset], c_size);
c_seg_major_compact_stats.moved_slots++;
c_seg_major_compact_stats.moved_bytes += c_size;
-#if CHECKSUM_THE_DATA
- c_dst->c_hash_data = c_src->c_hash_data;
-#endif
-#if CHECKSUM_THE_COMPRESSED_DATA
- c_dst->c_hash_compressed_data = c_src->c_hash_compressed_data;
-#endif
- c_dst->c_size = c_src->c_size;
- c_dst->c_packed_ptr = c_src->c_packed_ptr;
+ cslot_copy(c_dst, c_src);
c_dst->c_offset = c_seg_dst->c_nextoffset;
if (c_seg_dst->c_firstemptyslot == c_seg_dst->c_nextslot)
c_seg_src->c_bytes_unused += c_rounded_size;
c_seg_src->c_firstemptyslot = 0;
- if (c_seg_dst->c_nextoffset >= C_SEG_OFF_LIMIT || c_seg_dst->c_nextslot >= C_SLOT_MAX) {
+ if (c_seg_dst->c_nextoffset >= C_SEG_OFF_LIMIT || c_seg_dst->c_nextslot >= C_SLOT_MAX_INDEX) {
/* dest segment is now full */
keep_compacting = FALSE;
break;
}
}
+#if DEVELOPMENT || DEBUG
+ C_SEG_WRITE_PROTECT(c_seg_dst);
+#endif
if (dst_slot < c_seg_dst->c_nextslot) {
PAGE_REPLACEMENT_ALLOWED(TRUE);
}
-int compaction_swapper_inited = 0;
int compaction_swapper_init_now = 0;
int compaction_swapper_running = 0;
+int compaction_swapper_awakened = 0;
int compaction_swapper_abort = 0;
{
boolean_t should_swap = FALSE;
- if (vm_swap_up == TRUE) {
+ if (vm_swapout_ripe_segments == TRUE && c_overage_swapped_count < c_overage_swapped_limit) {
+ c_segment_t c_seg;
+ clock_sec_t now;
+ clock_sec_t age;
+ clock_nsec_t nsec;
+
+ clock_get_system_nanotime(&now, &nsec);
+ age = 0;
+
+ lck_mtx_lock_spin_always(c_list_lock);
+
+ if ( !queue_empty(&c_age_list_head)) {
+ c_seg = (c_segment_t) queue_first(&c_age_list_head);
+
+ age = now - c_seg->c_creation_ts;
+ }
+ lck_mtx_unlock_always(c_list_lock);
+
+ if (age >= vm_ripe_target_age)
+ return (TRUE);
+ }
+ if (VM_CONFIG_SWAP_IS_ACTIVE) {
if (COMPRESSOR_NEEDS_TO_SWAP()) {
return (TRUE);
}
if (VM_PAGE_Q_THROTTLED(&vm_pageout_queue_external) && vm_page_anonymous_count < (vm_page_inactive_count / 20)) {
return (TRUE);
}
- if (vm_page_free_count < (vm_page_free_reserved - COMPRESSOR_FREE_RESERVED_LIMIT))
+ if (vm_page_free_count < (vm_page_free_reserved - (COMPRESSOR_FREE_RESERVED_LIMIT * 2)))
return (TRUE);
}
compute_swapout_target_age();
if (swapout_target_age)
should_swap = TRUE;
- if (vm_swap_up == FALSE) {
-
- if (should_swap) {
#if CONFIG_JETSAM
- if (vm_compressor_thrashing_detected == FALSE) {
- vm_compressor_thrashing_detected = TRUE;
+ if (should_swap || c_segment_pages_compressed > c_segment_pages_compressed_nearing_limit) {
+
+ if (vm_compressor_thrashing_detected == FALSE) {
+ vm_compressor_thrashing_detected = TRUE;
- if (swapout_target_age) {
- memorystatus_kill_on_VM_thrashing(TRUE /* async */);
- compressor_thrashing_induced_jetsam++;
- } else {
- memorystatus_kill_on_FC_thrashing(TRUE /* async */);
- filecache_thrashing_induced_jetsam++;
- }
- /*
- * let the jetsam take precedence over
- * any major compactions we might have
- * been able to do... otherwise we run
- * the risk of doing major compactions
- * on segments we're about to free up
- * due to the jetsam activity.
- */
- should_swap = FALSE;
+ if (swapout_target_age || c_segment_pages_compressed > c_segment_pages_compressed_nearing_limit) {
+ memorystatus_kill_on_VM_thrashing(TRUE /* async */);
+ compressor_thrashing_induced_jetsam++;
+ } else {
+ memorystatus_kill_on_FC_thrashing(TRUE /* async */);
+ filecache_thrashing_induced_jetsam++;
}
+ }
+ /*
+ * let the jetsam take precedence over
+ * any major compactions we might have
+ * been able to do... otherwise we run
+ * the risk of doing major compactions
+ * on segments we're about to free up
+ * due to the jetsam activity.
+ */
+ should_swap = FALSE;
+ }
+
#endif /* CONFIG_JETSAM */
- } else
- should_swap = COMPRESSOR_NEEDS_TO_MAJOR_COMPACT();
+
+ if (should_swap == FALSE) {
+ /*
+ * vm_compressor_needs_to_major_compact returns true only if we're
+ * about to run out of available compressor segments... in this
+ * case, we absolutely need to run a major compaction even if
+ * we've just kicked off a jetsam or we don't otherwise need to
+ * swap... terminating objects releases
+ * pages back to the uncompressed cache, but does not guarantee
+ * that we will free up even a single compression segment
+ */
+ should_swap = vm_compressor_needs_to_major_compact();
}
/*
* will cause the major compaction engine to
* run, but will not trigger any swapping...
* segments that have been major compacted
- * will be moved to the swapped_out_q
- * but will not have the c_ondisk flag set
+ * will be moved to the majorcompact queue
*/
return (should_swap);
}
#endif /* CONFIG_JETSAM */
uint32_t vm_wake_compactor_swapper_calls = 0;
+uint32_t vm_run_compactor_already_running = 0;
+uint32_t vm_run_compactor_empty_minor_q = 0;
+uint32_t vm_run_compactor_did_compact = 0;
+uint32_t vm_run_compactor_waited = 0;
void
-vm_wake_compactor_swapper(void)
+vm_run_compactor(void)
{
- boolean_t need_major_compaction = FALSE;
+ if (c_segment_count == 0)
+ return;
+
+ lck_mtx_lock_spin_always(c_list_lock);
+
+ if (c_minor_count == 0) {
+ vm_run_compactor_empty_minor_q++;
+
+ lck_mtx_unlock_always(c_list_lock);
+ return;
+ }
+ if (compaction_swapper_running) {
+
+ if (vm_restricted_to_single_processor == FALSE) {
+ vm_run_compactor_already_running++;
+
+ lck_mtx_unlock_always(c_list_lock);
+ return;
+ }
+ vm_run_compactor_waited++;
+
+ assert_wait((event_t)&compaction_swapper_running, THREAD_UNINT);
+
+ lck_mtx_unlock_always(c_list_lock);
+
+ thread_block(THREAD_CONTINUE_NULL);
- if (compaction_swapper_running)
return;
+ }
+ vm_run_compactor_did_compact++;
+
+ fastwake_warmup = FALSE;
+ compaction_swapper_running = 1;
+
+ vm_compressor_do_delayed_compactions(FALSE);
+
+ compaction_swapper_running = 0;
- if (c_minor_count == 0 && need_major_compaction == FALSE)
+ lck_mtx_unlock_always(c_list_lock);
+
+ thread_wakeup((event_t)&compaction_swapper_running);
+}
+
+
+void
+vm_wake_compactor_swapper(void)
+{
+ if (compaction_swapper_running || compaction_swapper_awakened || c_segment_count == 0)
return;
+ if (c_minor_count || vm_compressor_needs_to_major_compact()) {
+
+ lck_mtx_lock_spin_always(c_list_lock);
+
+ fastwake_warmup = FALSE;
+
+ if (compaction_swapper_running == 0 && compaction_swapper_awakened == 0) {
+
+ vm_wake_compactor_swapper_calls++;
+
+ compaction_swapper_awakened = 1;
+ thread_wakeup((event_t)&c_compressor_swap_trigger);
+ }
+ lck_mtx_unlock_always(c_list_lock);
+ }
+}
+
+
+void
+vm_consider_swapping()
+{
+ c_segment_t c_seg, c_seg_next;
+ clock_sec_t now;
+ clock_nsec_t nsec;
+
+ assert(VM_CONFIG_SWAP_IS_PRESENT);
+
lck_mtx_lock_spin_always(c_list_lock);
- fastwake_warmup = FALSE;
+ compaction_swapper_abort = 1;
- if (compaction_swapper_running == 0) {
- vm_wake_compactor_swapper_calls++;
+ while (compaction_swapper_running) {
+ assert_wait((event_t)&compaction_swapper_running, THREAD_UNINT);
- thread_wakeup((event_t)&c_compressor_swap_trigger);
+ lck_mtx_unlock_always(c_list_lock);
- compaction_swapper_running = 1;
+ thread_block(THREAD_CONTINUE_NULL);
+
+ lck_mtx_lock_spin_always(c_list_lock);
+ }
+ compaction_swapper_abort = 0;
+ compaction_swapper_running = 1;
+
+ vm_swapout_ripe_segments = TRUE;
+
+ if (!queue_empty(&c_major_list_head)) {
+
+ clock_get_system_nanotime(&now, &nsec);
+
+ c_seg = (c_segment_t)queue_first(&c_major_list_head);
+
+ while (!queue_end(&c_major_list_head, (queue_entry_t)c_seg)) {
+
+ if (c_overage_swapped_count >= c_overage_swapped_limit)
+ break;
+
+ c_seg_next = (c_segment_t) queue_next(&c_seg->c_age_list);
+
+ if ((now - c_seg->c_creation_ts) >= vm_ripe_target_age) {
+
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
+
+ c_seg_switch_state(c_seg, C_ON_AGE_Q, FALSE);
+
+ lck_mtx_unlock_always(&c_seg->c_lock);
+ }
+ c_seg = c_seg_next;
+ }
}
+ vm_compressor_compact_and_swap(FALSE);
+
+ compaction_swapper_running = 0;
+
+ vm_swapout_ripe_segments = FALSE;
+
lck_mtx_unlock_always(c_list_lock);
+
+ thread_wakeup((event_t)&compaction_swapper_running);
}
{
boolean_t need_wakeup = FALSE;
- if (compaction_swapper_running)
+ if (c_segment_count == 0)
+ return;
+
+ if (compaction_swapper_running || compaction_swapper_awakened)
return;
if (!compaction_swapper_inited && !compaction_swapper_init_now) {
fastwake_warmup = FALSE;
- if (compaction_swapper_running == 0) {
+ if (compaction_swapper_running == 0 && compaction_swapper_awakened == 0) {
memoryshot(VM_WAKEUP_COMPACTOR_SWAPPER, DBG_FUNC_NONE);
+ compaction_swapper_awakened = 1;
thread_wakeup((event_t)&c_compressor_swap_trigger);
-
- compaction_swapper_running = 1;
}
lck_mtx_unlock_always(c_list_lock);
}
boolean_t needs_to_swap = FALSE;
- lck_mtx_assert(c_list_lock, LCK_MTX_ASSERT_OWNED);
+ LCK_MTX_ASSERT(c_list_lock, LCK_MTX_ASSERT_OWNED);
while (!queue_empty(&c_minor_list_head) && needs_to_swap == FALSE) {
c_seg_do_minor_compaction_and_unlock(c_seg, TRUE, FALSE, TRUE);
- if (vm_swap_up == TRUE && (number_compacted++ > DELAYED_COMPACTIONS_PER_PASS)) {
+ if (VM_CONFIG_SWAP_IS_ACTIVE && (number_compacted++ > DELAYED_COMPACTIONS_PER_PASS)) {
if ((flush_all == TRUE || compressor_needs_to_swap() == TRUE) && c_swapout_count < C_SWAPOUT_LIMIT)
needs_to_swap = TRUE;
lck_mtx_lock_spin_always(&c_seg->c_lock);
- queue_remove(&c_swappedin_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedin_q = 0;
- c_swappedin_count--;
-
- c_seg_insert_into_q(&c_age_list_head, c_seg);
- c_seg->c_on_age_q = 1;
- c_age_count++;
+ c_seg_switch_state(c_seg, C_ON_AGE_Q, FALSE);
lck_mtx_unlock_always(&c_seg->c_lock);
}
}
+extern int vm_num_swap_files;
+extern int vm_num_pinned_swap_files;
+extern int vm_swappin_enabled;
+
+extern unsigned int vm_swapfile_total_segs_used;
+extern unsigned int vm_swapfile_total_segs_alloced;
+
+
void
vm_compressor_flush(void)
{
lck_mtx_unlock_always(c_list_lock);
+ thread_wakeup((event_t)&compaction_swapper_running);
+
clock_get_uptime(&endTime);
SUB_ABSOLUTETIME(&endTime, &startTime);
absolutetime_to_nanoseconds(endTime, &nsec);
- HIBLOG("vm_compressor_flush completed - took %qd msecs\n", nsec / 1000000ULL);
+ HIBLOG("vm_compressor_flush completed - took %qd msecs - vm_num_swap_files = %d, vm_num_pinned_swap_files = %d, vm_swappin_enabled = %d\n",
+ nsec / 1000000ULL, vm_num_swap_files, vm_num_pinned_swap_files, vm_swappin_enabled);
}
-extern void vm_swap_file_set_tuneables(void);
int compaction_swap_trigger_thread_awakened = 0;
-
static void
vm_compressor_swap_trigger_thread(void)
{
+ current_thread()->options |= TH_OPT_VMPRIV;
+
/*
* compaction_swapper_init_now is set when the first call to
* vm_consider_waking_compactor_swapper is made from
* be operating on the correct directory (in case the default
* of /var/vm/ is overridden by the dymanic_pager
*/
- if (compaction_swapper_init_now && !compaction_swapper_inited) {
- if (vm_compressor_mode == VM_PAGER_COMPRESSOR_WITH_SWAP)
- vm_swap_file_set_tuneables();
+ if (compaction_swapper_init_now) {
+ vm_compaction_swapper_do_init();
- compaction_swapper_inited = 1;
+ if (vm_restricted_to_single_processor == TRUE)
+ thread_vm_bind_group_add();
+
+ compaction_swapper_init_now = 0;
}
lck_mtx_lock_spin_always(c_list_lock);
compaction_swap_trigger_thread_awakened++;
+ compaction_swapper_awakened = 0;
- vm_compressor_compact_and_swap(FALSE);
+ if (compaction_swapper_running == 0) {
+
+ compaction_swapper_running = 1;
+
+ vm_compressor_compact_and_swap(FALSE);
+ compaction_swapper_running = 0;
+ }
assert_wait((event_t)&c_compressor_swap_trigger, THREAD_UNINT);
- compaction_swapper_running = 0;
- thread_wakeup((event_t)&compaction_swapper_running);
+ if (compaction_swapper_running == 0)
+ thread_wakeup((event_t)&compaction_swapper_running);
lck_mtx_unlock_always(c_list_lock);
}
-#define DELAY_TRIM_ON_WAKE_SECS 4
+#define DELAY_TRIM_ON_WAKE_SECS 25
void
vm_compressor_delay_trim(void)
return;
}
- if (compaction_swapper_running == 0) {
+ if (compaction_swapper_running == 0 && compaction_swapper_awakened == 0) {
fastwake_warmup = TRUE;
- compaction_swapper_running = 1;
+
+ compaction_swapper_awakened = 1;
thread_wakeup((event_t)&c_compressor_swap_trigger);
}
lck_mtx_unlock_always(c_list_lock);
void
do_fastwake_warmup(void)
{
- uint64_t my_thread_id;
c_segment_t c_seg = NULL;
AbsoluteTime startTime, endTime;
uint64_t nsec;
lck_mtx_unlock_always(c_list_lock);
- my_thread_id = current_thread()->thread_id;
- proc_set_task_policy_thread(kernel_task, my_thread_id,
- TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER2);
+ proc_set_thread_policy(current_thread(),
+ TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER2);
PAGE_REPLACEMENT_DISALLOWED(TRUE);
c_seg->c_generation_id > last_c_segment_to_warm_generation_id)
break;
+ if (vm_page_free_count < (AVAILABLE_MEMORY / 4))
+ break;
+
lck_mtx_lock_spin_always(&c_seg->c_lock);
lck_mtx_unlock_always(c_list_lock);
c_seg_wait_on_busy(c_seg);
PAGE_REPLACEMENT_DISALLOWED(TRUE);
} else {
- c_seg_swapin(c_seg, TRUE);
-
- lck_mtx_unlock_always(&c_seg->c_lock);
+ if (c_seg_swapin(c_seg, TRUE, FALSE) == 0)
+ lck_mtx_unlock_always(&c_seg->c_lock);
c_segment_warmup_count++;
PAGE_REPLACEMENT_DISALLOWED(FALSE);
PAGE_REPLACEMENT_DISALLOWED(FALSE);
- proc_set_task_policy_thread(kernel_task, my_thread_id,
- TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER0);
+ proc_set_thread_policy(current_thread(),
+ TASK_POLICY_INTERNAL, TASK_POLICY_IO, THROTTLE_LEVEL_COMPRESSOR_TIER0);
clock_get_uptime(&endTime);
SUB_ABSOLUTETIME(&endTime, &startTime);
{
c_segment_t c_seg, c_seg_next;
boolean_t keep_compacting;
+ clock_sec_t now;
+ clock_nsec_t nsec;
if (fastwake_warmup == TRUE) {
vm_compressor_age_swapped_in_segments(flush_all);
+ /*
+ * we only need to grab the timestamp once per
+ * invocation of this function since the
+ * timescale we're interested in is measured
+ * in days
+ */
+ clock_get_system_nanotime(&now, &nsec);
while (!queue_empty(&c_age_list_head) && compaction_swapper_abort == 0) {
if (hibernate_flushing == TRUE) {
clock_sec_t sec;
- clock_nsec_t nsec;
if (hibernate_should_abort()) {
HIBLOG("vm_compressor_flush - hibernate_should_abort returned TRUE\n");
HIBLOG("vm_compressor_flush - out of swap space\n");
break;
}
+ if (vm_swap_files_pinned() == FALSE) {
+ HIBLOG("vm_compressor_flush - unpinned swap files\n");
+ break;
+ }
+ if (hibernate_in_progress_with_pinned_swap == TRUE &&
+ (vm_swapfile_total_segs_alloced == vm_swapfile_total_segs_used)) {
+ HIBLOG("vm_compressor_flush - out of pinned swap space\n");
+ break;
+ }
clock_get_system_nanotime(&sec, &nsec);
if (sec > hibernate_flushing_deadline) {
needs_to_swap = compressor_needs_to_swap();
+ if (needs_to_swap == TRUE && vm_swap_low_on_space())
+ vm_compressor_take_paging_space_action();
+
lck_mtx_lock_spin_always(c_list_lock);
if (needs_to_swap == FALSE)
break;
c_seg = (c_segment_t) queue_first(&c_age_list_head);
- if (flush_all == TRUE && c_seg->c_generation_id > c_generation_id_flush_barrier)
- break;
+ assert(c_seg->c_state == C_ON_AGE_Q);
- if (c_seg->c_filling) {
- /*
- * we're at or near the head... no more work to do
- */
+ if (flush_all == TRUE && c_seg->c_generation_id > c_generation_id_flush_barrier)
break;
- }
+
lck_mtx_lock_spin_always(&c_seg->c_lock);
if (c_seg->c_busy) {
* found an empty c_segment and freed it
* so go grab the next guy in the queue
*/
+ c_seg_major_compact_stats.count_of_freed_segs++;
continue;
}
/*
if (queue_end(&c_age_list_head, (queue_entry_t)c_seg_next))
break;
+ assert(c_seg_next->c_state == C_ON_AGE_Q);
+
if (c_seg_major_compact_ok(c_seg, c_seg_next) == FALSE)
break;
* found an empty c_segment and freed it
* so we can't continue to use c_seg_next
*/
+ c_seg_major_compact_stats.count_of_freed_segs++;
continue;
}
* by passing TRUE, we ask for c_busy to be cleared
* and c_wanted to be taken care of
*/
- c_seg_minor_compaction_and_unlock(c_seg_next, TRUE);
+ if (c_seg_minor_compaction_and_unlock(c_seg_next, TRUE))
+ c_seg_major_compact_stats.count_of_freed_segs++;
PAGE_REPLACEMENT_DISALLOWED(FALSE);
} /* major compaction */
- c_seg_major_compact_stats.wasted_space_in_swapouts += C_SEG_BUFSIZE - c_seg->c_bytes_used;
- c_seg_major_compact_stats.count_of_swapouts++;
-
lck_mtx_lock_spin_always(&c_seg->c_lock);
assert(c_seg->c_busy);
- assert(c_seg->c_on_age_q);
assert(!c_seg->c_on_minorcompact_q);
- queue_remove(&c_age_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_age_q = 0;
- c_age_count--;
+ if (VM_CONFIG_SWAP_IS_ACTIVE) {
+ /*
+ * This mode of putting a generic c_seg on the swapout list is
+ * only supported when we have general swapping enabled
+ */
+ c_seg_switch_state(c_seg, C_ON_SWAPOUT_Q, FALSE);
+ } else {
+ if ((vm_swapout_ripe_segments == TRUE && c_overage_swapped_count < c_overage_swapped_limit)) {
- if (vm_swap_up == TRUE) {
- queue_enter(&c_swapout_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swapout_q = 1;
- c_swapout_count++;
+ assert(VM_CONFIG_SWAP_IS_PRESENT);
+ /*
+ * we are running compressor sweeps with swap-behind
+ * make sure the c_seg has aged enough before swapping it
+ * out...
+ */
+ if ((now - c_seg->c_creation_ts) >= vm_ripe_target_age) {
+ c_seg->c_overage_swap = TRUE;
+ c_overage_swapped_count++;
+ c_seg_switch_state(c_seg, C_ON_SWAPOUT_Q, FALSE);
+ }
+ }
+ }
+ if (c_seg->c_state == C_ON_AGE_Q) {
+ /*
+ * this c_seg didn't get moved to the swapout queue
+ * so we need to move it out of the way...
+ * we just did a major compaction on it so put it
+ * on that queue
+ */
+ c_seg_switch_state(c_seg, C_ON_MAJORCOMPACT_Q, FALSE);
} else {
- queue_enter(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedout_q = 1;
- c_swappedout_count++;
+ c_seg_major_compact_stats.wasted_space_in_swapouts += C_SEG_BUFSIZE - c_seg->c_bytes_used;
+ c_seg_major_compact_stats.count_of_swapouts++;
}
C_SEG_WAKEUP_DONE(c_seg);
}
-static uint32_t no_paging_space_action_in_progress = 0;
-extern void memorystatus_send_low_swap_note(void);
-
-
static c_segment_t
c_seg_allocate(c_segment_t *current_chead)
{
- clock_sec_t sec;
- clock_nsec_t nsec;
c_segment_t c_seg;
- int slotarray;
+ int min_needed;
+ int size_to_populate;
+
+ if (vm_compressor_low_on_space())
+ vm_compressor_take_paging_space_action();
if ( (c_seg = *current_chead) == NULL ) {
uint32_t c_segno;
- if (vm_compressor_low_on_space() || vm_swap_low_on_space()) {
-
- if (no_paging_space_action_in_progress == 0) {
-
- if (OSCompareAndSwap(0, 1, (UInt32 *)&no_paging_space_action_in_progress)) {
-
- if (no_paging_space_action()) {
- memorystatus_send_low_swap_note();
- }
-
- no_paging_space_action_in_progress = 0;
- }
- }
- }
- KERNEL_DEBUG(0xe0400004 | DBG_FUNC_START, 0, 0, 0, 0, 0);
-
lck_mtx_lock_spin_always(c_list_lock);
while (c_segments_busy == TRUE) {
lck_mtx_lock_spin_always(c_list_lock);
}
if (c_free_segno_head == (uint32_t)-1) {
+ uint32_t c_segments_available_new;
if (c_segments_available >= c_segments_limit || c_segment_pages_compressed >= c_segment_pages_compressed_limit) {
lck_mtx_unlock_always(c_list_lock);
- KERNEL_DEBUG(0xe0400004 | DBG_FUNC_END, 0, 0, 0, 1, 0);
return (NULL);
}
c_segments_busy = TRUE;
lck_mtx_unlock_always(c_list_lock);
- kernel_memory_populate(kernel_map, (vm_offset_t)c_segments_next_page, PAGE_SIZE, KMA_KOBJECT);
+ kernel_memory_populate(compressor_map, (vm_offset_t)c_segments_next_page,
+ PAGE_SIZE, KMA_KOBJECT, VM_KERN_MEMORY_COMPRESSOR);
c_segments_next_page += PAGE_SIZE;
- for (c_segno = c_segments_available + 1; c_segno < (c_segments_available + C_SEGMENTS_PER_PAGE); c_segno++)
+ c_segments_available_new = c_segments_available + C_SEGMENTS_PER_PAGE;
+
+ if (c_segments_available_new > c_segments_limit)
+ c_segments_available_new = c_segments_limit;
+
+ for (c_segno = c_segments_available + 1; c_segno < c_segments_available_new; c_segno++)
c_segments[c_segno - 1].c_segno = c_segno;
lck_mtx_lock_spin_always(c_list_lock);
c_segments[c_segno - 1].c_segno = c_free_segno_head;
c_free_segno_head = c_segments_available;
- c_segments_available += C_SEGMENTS_PER_PAGE;
+ c_segments_available = c_segments_available_new;
c_segments_busy = FALSE;
thread_wakeup((event_t) (&c_segments_busy));
}
c_segno = c_free_segno_head;
- c_free_segno_head = c_segments[c_segno].c_segno;
+ assert(c_segno >= 0 && c_segno < c_segments_limit);
- lck_mtx_unlock_always(c_list_lock);
+ c_free_segno_head = (uint32_t)c_segments[c_segno].c_segno;
- c_seg = (c_segment_t)zalloc(compressor_segment_zone);
- bzero((char *)c_seg, sizeof(struct c_segment));
-
- if (kernel_memory_allocate(kernel_map, (vm_offset_t *)(&c_seg->c_store.c_buffer), C_SEG_ALLOCSIZE, 0, KMA_COMPRESSOR | KMA_VAONLY) != KERN_SUCCESS) {
- zfree(compressor_segment_zone, c_seg);
-
- lck_mtx_lock_spin_always(c_list_lock);
-
- c_segments[c_segno].c_segno = c_free_segno_head;
- c_free_segno_head = c_segno;
+ /*
+ * do the rest of the bookkeeping now while we're still behind
+ * the list lock and grab our generation id now into a local
+ * so that we can install it once we have the c_seg allocated
+ */
+ c_segment_count++;
+ if (c_segment_count > c_segment_count_max)
+ c_segment_count_max = c_segment_count;
- lck_mtx_unlock_always(c_list_lock);
+ lck_mtx_unlock_always(c_list_lock);
- KERNEL_DEBUG(0xe0400004 | DBG_FUNC_END, 0, 0, 0, 2, 0);
+ c_seg = (c_segment_t)zalloc(compressor_segment_zone);
+ bzero((char *)c_seg, sizeof(struct c_segment));
- return (NULL);
- }
- OSAddAtomic64(C_SEG_ALLOCSIZE, &compressor_kvspace_used);
+ c_seg->c_store.c_buffer = (int32_t *)C_SEG_BUFFER_ADDRESS(c_segno);
-#if __i386__ || __x86_64__
lck_mtx_init(&c_seg->c_lock, &vm_compressor_lck_grp, &vm_compressor_lck_attr);
-#else /* __i386__ || __x86_64__ */
- lck_spin_init(&c_seg->c_lock, &vm_compressor_lck_grp, &vm_compressor_lck_attr);
-#endif /* __i386__ || __x86_64__ */
- kernel_memory_populate(kernel_map, (vm_offset_t)(c_seg->c_store.c_buffer), 3 * PAGE_SIZE, KMA_COMPRESSOR);
-
- c_seg->c_populated_offset = C_SEG_BYTES_TO_OFFSET(3 * PAGE_SIZE);
- c_seg->c_firstemptyslot = C_SLOT_MAX;
+ c_seg->c_state = C_IS_EMPTY;
+ c_seg->c_firstemptyslot = C_SLOT_MAX_INDEX;
c_seg->c_mysegno = c_segno;
- c_seg->c_filling = 1;
lck_mtx_lock_spin_always(c_list_lock);
-
- c_segment_count++;
+ c_empty_count++;
+ c_seg_switch_state(c_seg, C_IS_FILLING, FALSE);
c_segments[c_segno].c_seg = c_seg;
-
- c_seg->c_generation_id = c_generation_id++;
-
- queue_enter(&c_age_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_age_q = 1;
- c_age_count++;
-
+ assert(c_segments[c_segno].c_segno > c_segments_available);
lck_mtx_unlock_always(c_list_lock);
- clock_get_system_nanotime(&sec, &nsec);
- c_seg->c_creation_ts = (uint32_t)sec;
-
*current_chead = c_seg;
- KERNEL_DEBUG(0xe0400004 | DBG_FUNC_END, c_seg, 0, 0, 3, 0);
+#if DEVELOPMENT || DEBUG
+ C_SEG_MAKE_WRITEABLE(c_seg);
+#endif
+
}
- slotarray = C_SEG_SLOTARRAY_FROM_INDEX(c_seg, c_seg->c_nextslot);
+ c_seg_alloc_nextslot(c_seg);
+
+ size_to_populate = C_SEG_ALLOCSIZE - C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset);
+
+ if (size_to_populate) {
+
+ min_needed = PAGE_SIZE + (C_SEG_ALLOCSIZE - C_SEG_BUFSIZE);
- if (c_seg->c_slots[slotarray] == 0) {
- KERNEL_DEBUG(0xe0400008 | DBG_FUNC_START, 0, 0, 0, 0, 0);
+ if (C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset - c_seg->c_nextoffset) < (unsigned) min_needed) {
- c_seg->c_slots[slotarray] = (struct c_slot *)kalloc(sizeof(struct c_slot) * C_SEG_SLOT_ARRAY_SIZE);
+ if (size_to_populate > C_SEG_MAX_POPULATE_SIZE)
+ size_to_populate = C_SEG_MAX_POPULATE_SIZE;
- KERNEL_DEBUG(0xe0400008 | DBG_FUNC_END, 0, 0, 0, 0, 0);
+ kernel_memory_populate(compressor_map,
+ (vm_offset_t) &c_seg->c_store.c_buffer[c_seg->c_populated_offset],
+ size_to_populate,
+ KMA_COMPRESSOR,
+ VM_KERN_MEMORY_COMPRESSOR);
+ } else
+ size_to_populate = 0;
}
-
PAGE_REPLACEMENT_DISALLOWED(TRUE);
lck_mtx_lock_spin_always(&c_seg->c_lock);
+ if (size_to_populate)
+ c_seg->c_populated_offset += C_SEG_BYTES_TO_OFFSET(size_to_populate);
+
return (c_seg);
}
-
static void
c_current_seg_filled(c_segment_t c_seg, c_segment_t *current_chead)
{
uint32_t unused_bytes;
uint32_t offset_to_depopulate;
+ int new_state = C_ON_AGE_Q;
+ clock_sec_t sec;
+ clock_nsec_t nsec;
unused_bytes = trunc_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset - c_seg->c_nextoffset));
lck_mtx_unlock_always(&c_seg->c_lock);
kernel_memory_depopulate(
- kernel_map,
+ compressor_map,
(vm_offset_t) &c_seg->c_store.c_buffer[offset_to_depopulate],
unused_bytes,
KMA_COMPRESSOR);
c_seg->c_populated_offset = offset_to_depopulate;
}
- c_seg->c_filling = 0;
+ assert(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset) <= C_SEG_BUFSIZE);
- if (C_SEG_UNUSED_BYTES(c_seg) >= PAGE_SIZE)
- c_seg_need_delayed_compaction(c_seg);
+#if DEVELOPMENT || DEBUG
+ {
+ boolean_t c_seg_was_busy = FALSE;
+
+ if ( !c_seg->c_busy)
+ C_SEG_BUSY(c_seg);
+ else
+ c_seg_was_busy = TRUE;
lck_mtx_unlock_always(&c_seg->c_lock);
+ C_SEG_WRITE_PROTECT(c_seg);
+
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
+
+ if (c_seg_was_busy == FALSE)
+ C_SEG_WAKEUP_DONE(c_seg);
+ }
+#endif
+
+#if CONFIG_FREEZE
+ if (current_chead == (c_segment_t*)&freezer_chead &&
+ VM_CONFIG_SWAP_IS_PRESENT &&
+ VM_CONFIG_FREEZER_SWAP_IS_ACTIVE &&
+ c_freezer_swapout_count < VM_MAX_FREEZER_CSEG_SWAP_COUNT) {
+ new_state = C_ON_SWAPOUT_Q;
+ }
+#endif /* CONFIG_FREEZE */
+
+ clock_get_system_nanotime(&sec, &nsec);
+ c_seg->c_creation_ts = (uint32_t)sec;
+
+ lck_mtx_lock_spin_always(c_list_lock);
+
+#if CONFIG_FREEZE
+ if (c_seg->c_state == C_ON_SWAPOUT_Q)
+ c_freezer_swapout_count++;
+#endif /* CONFIG_FREEZE */
+
+ c_seg->c_generation_id = c_generation_id++;
+ c_seg_switch_state(c_seg, new_state, FALSE);
+
+ if (c_seg->c_state == C_ON_AGE_Q && C_SEG_UNUSED_BYTES(c_seg) >= PAGE_SIZE)
+ c_seg_need_delayed_compaction(c_seg, TRUE);
+
+ lck_mtx_unlock_always(c_list_lock);
+
+#if CONFIG_FREEZE
+ if (c_seg->c_state == C_ON_SWAPOUT_Q)
+ thread_wakeup((event_t)&c_swapout_list_head);
+#endif /* CONFIG_FREEZE */
+
*current_chead = NULL;
}
* returns with c_seg locked
*/
void
-c_seg_swapin_requeue(c_segment_t c_seg)
+c_seg_swapin_requeue(c_segment_t c_seg, boolean_t has_data, boolean_t minor_compact_ok, boolean_t age_on_swapin_q)
{
clock_sec_t sec;
clock_nsec_t nsec;
lck_mtx_lock_spin_always(c_list_lock);
lck_mtx_lock_spin_always(&c_seg->c_lock);
- if (c_seg->c_on_swappedout_q) {
- queue_remove(&c_swappedout_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedout_q = 0;
- c_swappedout_count--;
+ assert(c_seg->c_busy_swapping);
+ assert(c_seg->c_busy);
+
+ c_seg->c_busy_swapping = 0;
+
+ if (c_seg->c_overage_swap == TRUE) {
+ c_overage_swapped_count--;
+ c_seg->c_overage_swap = FALSE;
+ }
+ if (has_data == TRUE) {
+ if (age_on_swapin_q == TRUE)
+ c_seg_switch_state(c_seg, C_ON_SWAPPEDIN_Q, FALSE);
+ else
+ c_seg_switch_state(c_seg, C_ON_AGE_Q, FALSE);
+
+ if (minor_compact_ok == TRUE && !c_seg->c_on_minorcompact_q && C_SEG_UNUSED_BYTES(c_seg) >= PAGE_SIZE)
+ c_seg_need_delayed_compaction(c_seg, TRUE);
} else {
- assert(c_seg->c_on_swappedout_sparse_q);
+ c_seg->c_store.c_buffer = (int32_t*) NULL;
+ c_seg->c_populated_offset = C_SEG_BYTES_TO_OFFSET(0);
- queue_remove(&c_swappedout_sparse_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedout_sparse_q = 0;
- c_swappedout_sparse_count--;
- }
- if (c_seg->c_store.c_buffer) {
- queue_enter(&c_swappedin_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_swappedin_q = 1;
- c_swappedin_count++;
+ c_seg_switch_state(c_seg, C_ON_BAD_Q, FALSE);
}
-#if TRACK_BAD_C_SEGMENTS
- else {
- queue_enter(&c_bad_list_head, c_seg, c_segment_t, c_age_list);
- c_seg->c_on_bad_q = 1;
- c_bad_count++;
- }
-#endif
c_seg->c_swappedin_ts = (uint32_t)sec;
- c_seg->c_ondisk = 0;
- c_seg->c_was_swapped_in = 1;
lck_mtx_unlock_always(c_list_lock);
}
/*
- * c_seg has to be locked and is returned locked.
+ * c_seg has to be locked and is returned locked if the c_seg isn't freed
* PAGE_REPLACMENT_DISALLOWED has to be TRUE on entry and is returned TRUE
+ * c_seg_swapin returns 1 if the c_seg was freed, 0 otherwise
*/
-void
-c_seg_swapin(c_segment_t c_seg, boolean_t force_minor_compaction)
+int
+c_seg_swapin(c_segment_t c_seg, boolean_t force_minor_compaction, boolean_t age_on_swapin_q)
{
vm_offset_t addr = 0;
uint32_t io_size = 0;
uint64_t f_offset;
+ assert(C_SEG_IS_ONDISK(c_seg));
+
#if !CHECKSUM_THE_SWAP
- if (c_seg->c_ondisk)
- c_seg_trim_tail(c_seg);
+ c_seg_trim_tail(c_seg);
#endif
io_size = round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset));
f_offset = c_seg->c_store.c_swap_handle;
C_SEG_BUSY(c_seg);
+ c_seg->c_busy_swapping = 1;
+
+ /*
+ * This thread is likely going to block for I/O.
+ * Make sure it is ready to run when the I/O completes because
+ * it needs to clear the busy bit on the c_seg so that other
+ * waiting threads can make progress too. To do that, boost
+ * the rwlock_count so that the priority is boosted.
+ */
+ set_thread_rwlock_boost();
lck_mtx_unlock_always(&c_seg->c_lock);
-
- if (c_seg->c_ondisk) {
- PAGE_REPLACEMENT_DISALLOWED(FALSE);
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
- if (kernel_memory_allocate(kernel_map, &addr, C_SEG_ALLOCSIZE, 0, KMA_COMPRESSOR | KMA_VAONLY) != KERN_SUCCESS)
- panic("c_seg_swapin: kernel_memory_allocate failed\n");
+ addr = (vm_offset_t)C_SEG_BUFFER_ADDRESS(c_seg->c_mysegno);
+ c_seg->c_store.c_buffer = (int32_t*) addr;
- kernel_memory_populate(kernel_map, addr, io_size, KMA_COMPRESSOR);
+ kernel_memory_populate(compressor_map, addr, io_size, KMA_COMPRESSOR, VM_KERN_MEMORY_COMPRESSOR);
- if (vm_swap_get(addr, f_offset, io_size) != KERN_SUCCESS) {
- PAGE_REPLACEMENT_DISALLOWED(TRUE);
+ if (vm_swap_get(c_seg, f_offset, io_size) != KERN_SUCCESS) {
+ PAGE_REPLACEMENT_DISALLOWED(TRUE);
- kernel_memory_depopulate(kernel_map, addr, io_size, KMA_COMPRESSOR);
- kmem_free(kernel_map, addr, C_SEG_ALLOCSIZE);
+ c_seg->c_store.c_swap_handle = f_offset;
- c_seg->c_store.c_buffer = (int32_t*) NULL;
- c_seg->c_populated_offset = C_SEG_BYTES_TO_OFFSET(0);
- } else {
- c_seg->c_store.c_buffer = (int32_t*) addr;
+ kernel_memory_depopulate(compressor_map, addr, io_size, KMA_COMPRESSOR);
+
+ c_seg_swapin_requeue(c_seg, FALSE, TRUE, age_on_swapin_q);
+ } else {
+ c_seg->c_store.c_buffer = (int32_t*) addr;
#if ENCRYPTED_SWAP
- vm_swap_decrypt(c_seg);
+ vm_swap_decrypt(c_seg);
#endif /* ENCRYPTED_SWAP */
#if CHECKSUM_THE_SWAP
- if (c_seg->cseg_swap_size != io_size)
- panic("swapin size doesn't match swapout size");
+ if (c_seg->cseg_swap_size != io_size)
+ panic("swapin size doesn't match swapout size");
- if (c_seg->cseg_hash != hash_string((char*) c_seg->c_store.c_buffer, (int)io_size)) {
- panic("c_seg_swapin - Swap hash mismatch\n");
- }
+ if (c_seg->cseg_hash != hash_string((char*) c_seg->c_store.c_buffer, (int)io_size)) {
+ panic("c_seg_swapin - Swap hash mismatch\n");
+ }
#endif /* CHECKSUM_THE_SWAP */
- PAGE_REPLACEMENT_DISALLOWED(TRUE);
+ PAGE_REPLACEMENT_DISALLOWED(TRUE);
- if (force_minor_compaction == TRUE) {
- lck_mtx_lock_spin_always(&c_seg->c_lock);
-
- c_seg_minor_compaction_and_unlock(c_seg, FALSE);
+ c_seg_swapin_requeue(c_seg, TRUE, force_minor_compaction == TRUE ? FALSE : TRUE, age_on_swapin_q);
+
+ OSAddAtomic64(c_seg->c_bytes_used, &compressor_bytes_used);
+
+ if (force_minor_compaction == TRUE) {
+ if (c_seg_minor_compaction_and_unlock(c_seg, FALSE)) {
+ /*
+ * Drop the rwlock_count so that the thread priority
+ * is returned back to where it is supposed to be.
+ */
+ clear_thread_rwlock_boost();
+ return (1);
}
- OSAddAtomic64(c_seg->c_bytes_used, &compressor_bytes_used);
- OSAddAtomic64(C_SEG_ALLOCSIZE, &compressor_kvspace_used);
+
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
}
}
- c_seg_swapin_requeue(c_seg);
-
C_SEG_WAKEUP_DONE(c_seg);
+
+ /*
+ * Drop the rwlock_count so that the thread priority
+ * is returned back to where it is supposed to be.
+ */
+ clear_thread_rwlock_boost();
+
+ return (0);
+}
+
+
+static void
+c_segment_sv_hash_drop_ref(int hash_indx)
+{
+ struct c_sv_hash_entry o_sv_he, n_sv_he;
+
+ while (1) {
+
+ o_sv_he.he_record = c_segment_sv_hash_table[hash_indx].he_record;
+
+ n_sv_he.he_ref = o_sv_he.he_ref - 1;
+ n_sv_he.he_data = o_sv_he.he_data;
+
+ if (OSCompareAndSwap64((UInt64)o_sv_he.he_record, (UInt64)n_sv_he.he_record, (UInt64 *) &c_segment_sv_hash_table[hash_indx].he_record) == TRUE) {
+ if (n_sv_he.he_ref == 0)
+ OSAddAtomic(-1, &c_segment_svp_in_hash);
+ break;
+ }
+ }
}
+static int
+c_segment_sv_hash_insert(uint32_t data)
+{
+ int hash_sindx;
+ int misses;
+ struct c_sv_hash_entry o_sv_he, n_sv_he;
+ boolean_t got_ref = FALSE;
+
+ if (data == 0)
+ OSAddAtomic(1, &c_segment_svp_zero_compressions);
+ else
+ OSAddAtomic(1, &c_segment_svp_nonzero_compressions);
+
+ hash_sindx = data & C_SV_HASH_MASK;
+
+ for (misses = 0; misses < C_SV_HASH_MAX_MISS; misses++)
+ {
+ o_sv_he.he_record = c_segment_sv_hash_table[hash_sindx].he_record;
+
+ while (o_sv_he.he_data == data || o_sv_he.he_ref == 0) {
+ n_sv_he.he_ref = o_sv_he.he_ref + 1;
+ n_sv_he.he_data = data;
+
+ if (OSCompareAndSwap64((UInt64)o_sv_he.he_record, (UInt64)n_sv_he.he_record, (UInt64 *) &c_segment_sv_hash_table[hash_sindx].he_record) == TRUE) {
+ if (n_sv_he.he_ref == 1)
+ OSAddAtomic(1, &c_segment_svp_in_hash);
+ got_ref = TRUE;
+ break;
+ }
+ o_sv_he.he_record = c_segment_sv_hash_table[hash_sindx].he_record;
+ }
+ if (got_ref == TRUE)
+ break;
+ hash_sindx++;
+
+ if (hash_sindx == C_SV_HASH_SIZE)
+ hash_sindx = 0;
+ }
+ if (got_ref == FALSE)
+ return(-1);
+
+ return (hash_sindx);
+}
+
+
+#if RECORD_THE_COMPRESSED_DATA
+
+static void
+c_compressed_record_data(char *src, int c_size)
+{
+ if ((c_compressed_record_cptr + c_size + 4) >= c_compressed_record_ebuf)
+ panic("c_compressed_record_cptr >= c_compressed_record_ebuf");
+
+ *(int *)((void *)c_compressed_record_cptr) = c_size;
+
+ c_compressed_record_cptr += 4;
+
+ memcpy(c_compressed_record_cptr, src, c_size);
+ c_compressed_record_cptr += c_size;
+}
+#endif
+
+
static int
c_compress_page(char *src, c_slot_mapping_t slot_ptr, c_segment_t *current_chead, char *scratch_buf)
{
return (1);
/*
* returns with c_seg lock held
- * and PAGE_REPLACEMENT_DISALLOWED(TRUE)
+ * and PAGE_REPLACEMENT_DISALLOWED(TRUE)...
+ * c_nextslot has been allocated and
+ * c_store.c_buffer populated
*/
+ assert(c_seg->c_state == C_IS_FILLING);
+
cs = C_SEG_SLOT_FROM_INDEX(c_seg, c_seg->c_nextslot);
cs->c_packed_ptr = C_SLOT_PACK_PTR(slot_ptr);
if (max_csize > PAGE_SIZE)
max_csize = PAGE_SIZE;
- if (C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset -
- c_seg->c_nextoffset)
- < (unsigned) max_csize + PAGE_SIZE &&
- (C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset)
- < C_SEG_ALLOCSIZE)) {
- lck_mtx_unlock_always(&c_seg->c_lock);
-
- kernel_memory_populate(kernel_map,
- (vm_offset_t) &c_seg->c_store.c_buffer[c_seg->c_populated_offset],
- PAGE_SIZE,
- KMA_COMPRESSOR);
-
- lck_mtx_lock_spin_always(&c_seg->c_lock);
-
- c_seg->c_populated_offset += C_SEG_BYTES_TO_OFFSET(PAGE_SIZE);
- }
-
#if CHECKSUM_THE_DATA
cs->c_hash_data = hash_string(src, PAGE_SIZE);
#endif
- c_size = WKdm_compress_new((WK_word *)(uintptr_t)src, (WK_word *)(uintptr_t)&c_seg->c_store.c_buffer[cs->c_offset],
+ if (vm_compressor_algorithm() != VM_COMPRESSOR_DEFAULT_CODEC) {
+ } else {
+ c_size = WKdm_compress_new((const WK_word *)(uintptr_t)src, (WK_word *)(uintptr_t)&c_seg->c_store.c_buffer[cs->c_offset],
(WK_word *)(uintptr_t)scratch_buf, max_csize - 4);
+ }
assert(c_size <= (max_csize - 4) && c_size >= -1);
if (c_size == -1) {
if (max_csize < PAGE_SIZE) {
c_current_seg_filled(c_seg, current_chead);
+ assert(*current_chead == NULL);
- PAGE_REPLACEMENT_DISALLOWED(FALSE);
+ lck_mtx_unlock_always(&c_seg->c_lock);
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
goto retry;
}
c_size = PAGE_SIZE;
memcpy(&c_seg->c_store.c_buffer[cs->c_offset], src, c_size);
+
+ OSAddAtomic(1, &c_segment_noncompressible_pages);
+
+ } else if (c_size == 0) {
+ int hash_index;
+
+ /*
+ * special case - this is a page completely full of a single 32 bit value
+ */
+ hash_index = c_segment_sv_hash_insert(*(uint32_t *)(uintptr_t)src);
+
+ if (hash_index != -1) {
+ slot_ptr->s_cindx = hash_index;
+ slot_ptr->s_cseg = C_SV_CSEG_ID;
+
+ OSAddAtomic(1, &c_segment_svp_hash_succeeded);
+#if RECORD_THE_COMPRESSED_DATA
+ c_compressed_record_data(src, 4);
+#endif
+ goto sv_compression;
+ }
+ c_size = 4;
+
+ memcpy(&c_seg->c_store.c_buffer[cs->c_offset], src, c_size);
+
+ OSAddAtomic(1, &c_segment_svp_hash_failed);
}
+
+#if RECORD_THE_COMPRESSED_DATA
+ c_compressed_record_data((char *)&c_seg->c_store.c_buffer[cs->c_offset], c_size);
+#endif
+
#if CHECKSUM_THE_COMPRESSED_DATA
cs->c_hash_compressed_data = hash_string((char *)&c_seg->c_store.c_buffer[cs->c_offset], c_size);
#endif
/* <csegno=0,indx=0> would mean "empty slot", so use csegno+1 */
slot_ptr->s_cseg = c_seg->c_mysegno + 1;
- if (c_seg->c_nextoffset >= C_SEG_OFF_LIMIT || c_seg->c_nextslot >= C_SLOT_MAX)
+sv_compression:
+ if (c_seg->c_nextoffset >= C_SEG_OFF_LIMIT || c_seg->c_nextslot >= C_SLOT_MAX_INDEX) {
c_current_seg_filled(c_seg, current_chead);
- else
- lck_mtx_unlock_always(&c_seg->c_lock);
+ assert(*current_chead == NULL);
+ }
+ lck_mtx_unlock_always(&c_seg->c_lock);
PAGE_REPLACEMENT_DISALLOWED(FALSE);
- OSAddAtomic64(c_rounded_size, &compressor_bytes_used);
+#if RECORD_THE_COMPRESSED_DATA
+ if ((c_compressed_record_cptr - c_compressed_record_sbuf) >= C_SEG_ALLOCSIZE) {
+ c_compressed_record_write(c_compressed_record_sbuf, (int)(c_compressed_record_cptr - c_compressed_record_sbuf));
+ c_compressed_record_cptr = c_compressed_record_sbuf;
+ }
+#endif
+ if (c_size) {
+ OSAddAtomic64(c_size, &c_segment_compressed_bytes);
+ OSAddAtomic64(c_rounded_size, &compressor_bytes_used);
+ }
OSAddAtomic64(PAGE_SIZE, &c_segment_input_bytes);
- OSAddAtomic64(c_size, &c_segment_compressed_bytes);
OSAddAtomic(1, &c_segment_pages_compressed);
OSAddAtomic(1, &sample_period_compression_count);
return (0);
}
+static inline void sv_decompress(int32_t *ddst, int32_t pattern) {
+#if __x86_64__
+ memset_word(ddst, pattern, PAGE_SIZE / sizeof(int32_t));
+#else
+ size_t i;
+
+ /* Unroll the pattern fill loop 4x to encourage the
+ * compiler to emit NEON stores, cf.
+ * <rdar://problem/25839866> Loop autovectorization
+ * anomalies.
+ * We use separate loops for each PAGE_SIZE
+ * to allow the autovectorizer to engage, as PAGE_SIZE
+ * is currently not a constant.
+ */
+
+ if (PAGE_SIZE == 4096) {
+ for (i = 0; i < (4096U / sizeof(int32_t)); i += 4) {
+ *ddst++ = pattern;
+ *ddst++ = pattern;
+ *ddst++ = pattern;
+ *ddst++ = pattern;
+ }
+ } else {
+ assert(PAGE_SIZE == 16384);
+ for (i = 0; i < (int)(16384U / sizeof(int32_t)); i += 4) {
+ *ddst++ = pattern;
+ *ddst++ = pattern;
+ *ddst++ = pattern;
+ *ddst++ = pattern;
+ }
+ }
+#endif
+}
static int
c_decompress_page(char *dst, volatile c_slot_mapping_t slot_ptr, int flags, int *zeroslot)
{
c_slot_t cs;
c_segment_t c_seg;
+ uint32_t c_segno;
int c_indx;
int c_rounded_size;
uint32_t c_size;
int retval = 0;
- boolean_t c_seg_has_data = TRUE;
- boolean_t c_seg_swappedin = FALSE;
boolean_t need_unlock = TRUE;
boolean_t consider_defragmenting = FALSE;
+ boolean_t kdp_mode = FALSE;
+
+ if (__improbable(flags & C_KDP)) {
+ if (not_in_kdp) {
+ panic("C_KDP passed to decompress page from outside of debugger context");
+ }
+
+ assert((flags & C_KEEP) == C_KEEP);
+ assert((flags & C_DONT_BLOCK) == C_DONT_BLOCK);
+
+ if ((flags & (C_DONT_BLOCK | C_KEEP)) != (C_DONT_BLOCK | C_KEEP)) {
+ return (-2);
+ }
+
+ kdp_mode = TRUE;
+ *zeroslot = 0;
+ }
ReTry:
- PAGE_REPLACEMENT_DISALLOWED(TRUE);
+ if (__probable(!kdp_mode)) {
+ PAGE_REPLACEMENT_DISALLOWED(TRUE);
+ } else {
+ if (kdp_lck_rw_lock_is_acquired_exclusive(&c_master_lock)) {
+ return (-2);
+ }
+ }
#if HIBERNATION
/*
if (dst && decompressions_blocked == TRUE) {
if (flags & C_DONT_BLOCK) {
- PAGE_REPLACEMENT_DISALLOWED(FALSE);
+ if (__probable(!kdp_mode)) {
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
+ }
*zeroslot = 0;
return (-2);
}
#endif
/* s_cseg is actually "segno+1" */
- c_seg = c_segments[slot_ptr->s_cseg - 1].c_seg;
+ c_segno = slot_ptr->s_cseg - 1;
- lck_mtx_lock_spin_always(&c_seg->c_lock);
+ if (__improbable(c_segno >= c_segments_available))
+ panic("c_decompress_page: c_segno %d >= c_segments_available %d, slot_ptr(%p), slot_data(%x)",
+ c_segno, c_segments_available, slot_ptr, *(int *)((void *)slot_ptr));
- if (flags & C_DONT_BLOCK) {
- if (c_seg->c_busy || (c_seg->c_ondisk && dst)) {
+ if (__improbable(c_segments[c_segno].c_segno < c_segments_available))
+ panic("c_decompress_page: c_segno %d is free, slot_ptr(%p), slot_data(%x)",
+ c_segno, slot_ptr, *(int *)((void *)slot_ptr));
- retval = -2;
+ c_seg = c_segments[c_segno].c_seg;
+
+ if (__probable(!kdp_mode)) {
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
+ } else {
+ if (kdp_lck_mtx_lock_spin_is_acquired(&c_seg->c_lock)) {
+ return (-2);
+ }
+ }
+
+ assert(c_seg->c_state != C_IS_EMPTY && c_seg->c_state != C_IS_FREE);
+
+ if (dst == NULL && c_seg->c_busy_swapping) {
+ assert(c_seg->c_busy);
+
+ goto bypass_busy_check;
+ }
+ if (flags & C_DONT_BLOCK) {
+ if (c_seg->c_busy || (C_SEG_IS_ONDISK(c_seg) && dst)) {
*zeroslot = 0;
+ retval = -2;
goto done;
}
}
goto ReTry;
}
+bypass_busy_check:
+
c_indx = slot_ptr->s_cindx;
+ if (__improbable(c_indx >= c_seg->c_nextslot))
+ panic("c_decompress_page: c_indx %d >= c_nextslot %d, c_seg(%p), slot_ptr(%p), slot_data(%x)",
+ c_indx, c_seg->c_nextslot, c_seg, slot_ptr, *(int *)((void *)slot_ptr));
+
cs = C_SEG_SLOT_FROM_INDEX(c_seg, c_indx);
c_size = UNPACK_C_SIZE(cs);
+ if (__improbable(c_size == 0))
+ panic("c_decompress_page: c_size == 0, c_seg(%p), slot_ptr(%p), slot_data(%x)",
+ c_seg, slot_ptr, *(int *)((void *)slot_ptr));
+
c_rounded_size = (c_size + C_SEG_OFFSET_ALIGNMENT_MASK) & ~C_SEG_OFFSET_ALIGNMENT_MASK;
if (dst) {
clock_sec_t cur_ts_sec;
clock_nsec_t cur_ts_nsec;
- if (c_seg->c_on_swappedout_q || c_seg->c_on_swappedout_sparse_q) {
- if (c_seg->c_ondisk)
- c_seg_swappedin = TRUE;
- c_seg_swapin(c_seg, FALSE);
+ if (C_SEG_IS_ONDISK(c_seg)) {
+ assert(kdp_mode == FALSE);
+ retval = c_seg_swapin(c_seg, FALSE, TRUE);
+ assert(retval == 0);
+
+ retval = 1;
}
- if (c_seg->c_store.c_buffer == NULL) {
- c_seg_has_data = FALSE;
+ if (c_seg->c_state == C_ON_BAD_Q) {
+ assert(c_seg->c_store.c_buffer == NULL);
+
+ retval = -1;
goto c_seg_invalid_data;
}
#if CHECKSUM_THE_COMPRESSED_DATA
if (cs->c_hash_compressed_data != hash_string((char *)&c_seg->c_store.c_buffer[cs->c_offset], c_size))
- panic("compressed data doesn't match original");
+ panic("compressed data doesn't match original hash: 0x%x, seg: %p, offset: %d, c_size: %d", cs->c_hash_compressed_data, c_seg, cs->c_offset, c_size);
#endif
if (c_rounded_size == PAGE_SIZE) {
/*
* page wasn't compressible... just copy it out
*/
memcpy(dst, &c_seg->c_store.c_buffer[cs->c_offset], PAGE_SIZE);
+ } else if (c_size == 4) {
+ int32_t data;
+ int32_t *dptr;
+
+ /*
+ * page was populated with a single value
+ * that didn't fit into our fast hash
+ * so we packed it in as a single non-compressed value
+ * that we need to populate the page with
+ */
+ dptr = (int32_t *)(uintptr_t)dst;
+ data = *(int32_t *)(&c_seg->c_store.c_buffer[cs->c_offset]);
+ sv_decompress(dptr, data);
} else {
uint32_t my_cpu_no;
char *scratch_buf;
- /*
- * we're behind the c_seg lock held in spin mode
- * which means pre-emption is disabled... therefore
- * the following sequence is atomic and safe
- */
- my_cpu_no = cpu_number();
+ if (__probable(!kdp_mode)) {
+ /*
+ * we're behind the c_seg lock held in spin mode
+ * which means pre-emption is disabled... therefore
+ * the following sequence is atomic and safe
+ */
+ my_cpu_no = cpu_number();
- assert(my_cpu_no < compressor_cpus);
+ assert(my_cpu_no < compressor_cpus);
+
+ scratch_buf = &compressor_scratch_bufs[my_cpu_no * vm_compressor_get_decode_scratch_size()];
+ } else {
+ scratch_buf = kdp_compressor_scratch_buf;
+ }
- scratch_buf = &compressor_scratch_bufs[my_cpu_no * WKdm_SCRATCH_BUF_SIZE];
+ if (vm_compressor_algorithm() != VM_COMPRESSOR_DEFAULT_CODEC) {
+ } else {
WKdm_decompress_new((WK_word *)(uintptr_t)&c_seg->c_store.c_buffer[cs->c_offset],
(WK_word *)(uintptr_t)dst, (WK_word *)(uintptr_t)scratch_buf, c_size);
+ }
}
#if CHECKSUM_THE_DATA
if (cs->c_hash_data != hash_string(dst, PAGE_SIZE))
- panic("decompressed data doesn't match original");
+ panic("decompressed data doesn't match original cs: %p, hash: %d, offset: %d, c_size: %d", cs, cs->c_hash_data, cs->c_offset, c_size);
+
#endif
- if (!c_seg->c_was_swapped_in) {
+ if (c_seg->c_swappedin_ts == 0 && !kdp_mode) {
clock_get_system_nanotime(&cur_ts_sec, &cur_ts_nsec);
age_of_cseg = (uint32_t)cur_ts_sec - c_seg->c_creation_ts;
-
if (age_of_cseg < DECOMPRESSION_SAMPLE_MAX_AGE)
OSAddAtomic(1, &age_of_decompressions_during_sample_period[age_of_cseg]);
else
OSAddAtomic(1, &sample_period_decompression_count);
}
- } else {
- if (c_seg->c_store.c_buffer == NULL)
- c_seg_has_data = FALSE;
}
c_seg_invalid_data:
- if (c_seg_has_data == TRUE) {
- if (c_seg_swappedin == TRUE)
- retval = 1;
- else
- retval = 0;
- } else
- retval = -1;
-
if (flags & C_KEEP) {
*zeroslot = 0;
goto done;
}
+ assert(kdp_mode == FALSE);
+
c_seg->c_bytes_unused += c_rounded_size;
c_seg->c_bytes_used -= c_rounded_size;
PACK_C_SIZE(cs, 0);
OSAddAtomic(-1, &c_segment_pages_compressed);
- if (c_seg_has_data == TRUE && !c_seg->c_ondisk) {
+ if (c_seg->c_state != C_ON_BAD_Q && !(C_SEG_IS_ONDISK(c_seg))) {
/*
- * c_ondisk == TRUE can occur when we're doing a
+ * C_SEG_IS_ONDISK == TRUE can occur when we're doing a
* free of a compressed page (i.e. dst == NULL)
*/
OSAddAtomic64(-c_rounded_size, &compressor_bytes_used);
}
- if (!c_seg->c_filling) {
+ if (c_seg->c_busy_swapping) {
+ /*
+ * bypass case for c_busy_swapping...
+ * let the swapin/swapout paths deal with putting
+ * the c_seg on the minor compaction queue if needed
+ */
+ assert(c_seg->c_busy);
+ goto done;
+ }
+ assert(!c_seg->c_busy);
+
+ if (c_seg->c_state != C_IS_FILLING) {
if (c_seg->c_bytes_used == 0) {
- if (!c_seg->c_ondisk) {
+ if ( !(C_SEG_IS_ONDISK(c_seg))) {
int pages_populated;
pages_populated = (round_page_32(C_SEG_OFFSET_TO_BYTES(c_seg->c_populated_offset))) / PAGE_SIZE;
c_seg->c_populated_offset = C_SEG_BYTES_TO_OFFSET(0);
if (pages_populated) {
+
+ assert(c_seg->c_state != C_ON_BAD_Q);
assert(c_seg->c_store.c_buffer != NULL);
C_SEG_BUSY(c_seg);
lck_mtx_unlock_always(&c_seg->c_lock);
- kernel_memory_depopulate(kernel_map, (vm_offset_t) c_seg->c_store.c_buffer, pages_populated * PAGE_SIZE, KMA_COMPRESSOR);
+ kernel_memory_depopulate(compressor_map, (vm_offset_t) c_seg->c_store.c_buffer, pages_populated * PAGE_SIZE, KMA_COMPRESSOR);
lck_mtx_lock_spin_always(&c_seg->c_lock);
C_SEG_WAKEUP_DONE(c_seg);
}
- if (!c_seg->c_on_minorcompact_q && !c_seg->c_on_swapout_q)
- c_seg_need_delayed_compaction(c_seg);
- } else
- assert(c_seg->c_on_swappedout_sparse_q);
+ if (!c_seg->c_on_minorcompact_q && c_seg->c_state != C_ON_SWAPOUT_Q)
+ c_seg_need_delayed_compaction(c_seg, FALSE);
+ } else {
+ if (c_seg->c_state != C_ON_SWAPPEDOUTSPARSE_Q) {
+ c_seg_move_to_sparse_list(c_seg);
+ consider_defragmenting = TRUE;
+ }
+ }
} else if (c_seg->c_on_minorcompact_q) {
- if (C_SEG_INCORE_IS_SPARSE(c_seg)) {
+ assert(c_seg->c_state != C_ON_BAD_Q);
+
+ if (C_SEG_SHOULD_MINORCOMPACT(c_seg)) {
c_seg_try_minor_compaction_and_unlock(c_seg);
need_unlock = FALSE;
}
- } else if (!c_seg->c_ondisk) {
+ } else if ( !(C_SEG_IS_ONDISK(c_seg))) {
- if (c_seg_has_data == TRUE && !c_seg->c_on_swapout_q && C_SEG_UNUSED_BYTES(c_seg) >= PAGE_SIZE) {
- c_seg_need_delayed_compaction(c_seg);
+ if (c_seg->c_state != C_ON_BAD_Q && c_seg->c_state != C_ON_SWAPOUT_Q && C_SEG_UNUSED_BYTES(c_seg) >= PAGE_SIZE) {
+ c_seg_need_delayed_compaction(c_seg, FALSE);
}
- } else if (!c_seg->c_on_swappedout_sparse_q && C_SEG_ONDISK_IS_SPARSE(c_seg)) {
+ } else if (c_seg->c_state != C_ON_SWAPPEDOUTSPARSE_Q && C_SEG_ONDISK_IS_SPARSE(c_seg)) {
c_seg_move_to_sparse_list(c_seg);
consider_defragmenting = TRUE;
}
}
done:
+ if (__improbable(kdp_mode)) {
+ return retval;
+ }
+
if (need_unlock == TRUE)
lck_mtx_unlock_always(&c_seg->c_lock);
int
vm_compressor_get(ppnum_t pn, int *slot, int flags)
{
+ c_slot_mapping_t slot_ptr;
char *dst;
int zeroslot = 1;
int retval;
#else
#error "unsupported architecture"
#endif
+ slot_ptr = (c_slot_mapping_t)slot;
+
+ if (slot_ptr->s_cseg == C_SV_CSEG_ID) {
+ int32_t data;
+ int32_t *dptr;
+
+ /*
+ * page was populated with a single value
+ * that found a home in our hash table
+ * grab that value from the hash and populate the page
+ * that we need to populate the page with
+ */
+ dptr = (int32_t *)(uintptr_t)dst;
+ data = c_segment_sv_hash_table[slot_ptr->s_cindx].he_data;
+#if __x86_64__
+ memset_word(dptr, data, PAGE_SIZE / sizeof(int32_t));
+#else
+ {
+ int i;
+
+ for (i = 0; i < (int)(PAGE_SIZE / sizeof(int32_t)); i++)
+ *dptr++ = data;
+ }
+#endif
+ c_segment_sv_hash_drop_ref(slot_ptr->s_cindx);
- retval = c_decompress_page(dst, (c_slot_mapping_t)slot, flags, &zeroslot);
+ if ( !(flags & C_KEEP)) {
+ OSAddAtomic(-1, &c_segment_pages_compressed);
+ *slot = 0;
+ }
+ if (data)
+ OSAddAtomic(1, &c_segment_svp_nonzero_decompressions);
+ else
+ OSAddAtomic(1, &c_segment_svp_zero_decompressions);
+
+ return (0);
+ }
+
+ retval = c_decompress_page(dst, slot_ptr, flags, &zeroslot);
/*
* zeroslot will be set to 0 by c_decompress_page if (flags & C_KEEP)
- * or (flags & C_DONT_BLOCK) and we found 'c_busy' or 'c_ondisk' set
+ * or (flags & C_DONT_BLOCK) and we found 'c_busy' or 'C_SEG_IS_ONDISK' to be TRUE
*/
if (zeroslot) {
*slot = 0;
* returns 0 if we successfully decompressed a page from a segment already in memory
* returns 1 if we had to first swap in the segment, before successfully decompressing the page
* returns -1 if we encountered an error swapping in the segment - decompression failed
- * returns -2 if (flags & C_DONT_BLOCK) and we found 'c_busy' or 'c_ondisk' set
+ * returns -2 if (flags & C_DONT_BLOCK) and we found 'c_busy' or 'C_SEG_IS_ONDISK' to be true
*/
return (retval);
}
int
vm_compressor_free(int *slot, int flags)
{
+ c_slot_mapping_t slot_ptr;
int zeroslot = 1;
int retval;
assert(flags == 0 || flags == C_DONT_BLOCK);
- retval = c_decompress_page(NULL, (c_slot_mapping_t)slot, flags, &zeroslot);
+ slot_ptr = (c_slot_mapping_t)slot;
+
+ if (slot_ptr->s_cseg == C_SV_CSEG_ID) {
+
+ c_segment_sv_hash_drop_ref(slot_ptr->s_cindx);
+ OSAddAtomic(-1, &c_segment_pages_compressed);
+
+ *slot = 0;
+ return (0);
+ }
+ retval = c_decompress_page(NULL, slot_ptr, flags, &zeroslot);
/*
* returns 0 if we successfully freed the specified compressed page
* returns -2 if (flags & C_DONT_BLOCK) and we found 'c_busy' set
if (retval == 0)
*slot = 0;
+ else
+ assert(retval == -2);
return (retval);
}
int c_indx;
c_slot_t cs;
- dst_slot = (c_slot_mapping_t) dst_slot_p;
src_slot = (c_slot_mapping_t) src_slot_p;
+ if (src_slot->s_cseg == C_SV_CSEG_ID) {
+ *dst_slot_p = *src_slot_p;
+ *src_slot_p = 0;
+ return;
+ }
+ dst_slot = (c_slot_mapping_t) dst_slot_p;
Retry:
PAGE_REPLACEMENT_DISALLOWED(TRUE);
/* get segment for src_slot */
/* lock segment */
lck_mtx_lock_spin_always(&c_seg->c_lock);
/* wait if it's busy */
- if (c_seg->c_busy) {
+ if (c_seg->c_busy && !c_seg->c_busy_swapping) {
PAGE_REPLACEMENT_DISALLOWED(FALSE);
c_seg_wait_on_busy(c_seg);
goto Retry;
lck_mtx_unlock_always(&c_seg->c_lock);
PAGE_REPLACEMENT_DISALLOWED(FALSE);
}
+
+#if CONFIG_FREEZE
+
+int freezer_finished_filling = 0;
+
+void
+vm_compressor_finished_filling(
+ void **current_chead)
+{
+ c_segment_t c_seg;
+
+ if ((c_seg = *(c_segment_t *)current_chead) == NULL)
+ return;
+
+ assert(c_seg->c_state == C_IS_FILLING);
+
+ lck_mtx_lock_spin_always(&c_seg->c_lock);
+
+ c_current_seg_filled(c_seg, (c_segment_t *)current_chead);
+
+ lck_mtx_unlock_always(&c_seg->c_lock);
+
+ freezer_finished_filling++;
+}
+
+
+/*
+ * This routine is used to transfer the compressed chunks from
+ * the c_seg/cindx pointed to by slot_p into a new c_seg headed
+ * by the current_chead and a new cindx within that c_seg.
+ *
+ * Currently, this routine is only used by the "freezer backed by
+ * compressor with swap" mode to create a series of c_segs that
+ * only contain compressed data belonging to one task. So, we
+ * move a task's previously compressed data into a set of new
+ * c_segs which will also hold the task's yet to be compressed data.
+ */
+
+kern_return_t
+vm_compressor_relocate(
+ void **current_chead,
+ int *slot_p)
+{
+ c_slot_mapping_t slot_ptr;
+ c_slot_mapping_t src_slot;
+ uint32_t c_rounded_size;
+ uint32_t c_size;
+ uint16_t dst_slot;
+ c_slot_t c_dst;
+ c_slot_t c_src;
+ int c_indx;
+ c_segment_t c_seg_dst = NULL;
+ c_segment_t c_seg_src = NULL;
+ kern_return_t kr = KERN_SUCCESS;
+
+
+ src_slot = (c_slot_mapping_t) slot_p;
+
+ if (src_slot->s_cseg == C_SV_CSEG_ID) {
+ /*
+ * no need to relocate... this is a page full of a single
+ * value which is hashed to a single entry not contained
+ * in a c_segment_t
+ */
+ return (kr);
+ }
+
+Relookup_dst:
+ c_seg_dst = c_seg_allocate((c_segment_t *)current_chead);
+ /*
+ * returns with c_seg lock held
+ * and PAGE_REPLACEMENT_DISALLOWED(TRUE)...
+ * c_nextslot has been allocated and
+ * c_store.c_buffer populated
+ */
+ if (c_seg_dst == NULL) {
+ /*
+ * Out of compression segments?
+ */
+ kr = KERN_RESOURCE_SHORTAGE;
+ goto out;
+ }
+
+ assert(c_seg_dst->c_busy == 0);
+
+ C_SEG_BUSY(c_seg_dst);
+
+ dst_slot = c_seg_dst->c_nextslot;
+
+ lck_mtx_unlock_always(&c_seg_dst->c_lock);
+
+Relookup_src:
+ c_seg_src = c_segments[src_slot->s_cseg - 1].c_seg;
+
+ assert(c_seg_dst != c_seg_src);
+
+ lck_mtx_lock_spin_always(&c_seg_src->c_lock);
+
+ if (C_SEG_IS_ONDISK(c_seg_src)) {
+
+ /*
+ * A "thaw" can mark a process as eligible for
+ * another freeze cycle without bringing any of
+ * its swapped out c_segs back from disk (because
+ * that is done on-demand).
+ *
+ * If the src c_seg we find for our pre-compressed
+ * data is already on-disk, then we are dealing
+ * with an app's data that is already packed and
+ * swapped out. Don't do anything.
+ */
+
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
+
+ lck_mtx_unlock_always(&c_seg_src->c_lock);
+
+ c_seg_src = NULL;
+
+ goto out;
+ }
+
+ if (c_seg_src->c_busy) {
+
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
+ c_seg_wait_on_busy(c_seg_src);
+
+ c_seg_src = NULL;
+
+ PAGE_REPLACEMENT_DISALLOWED(TRUE);
+
+ goto Relookup_src;
+ }
+
+ C_SEG_BUSY(c_seg_src);
+
+ lck_mtx_unlock_always(&c_seg_src->c_lock);
+
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
+
+ /* find the c_slot */
+ c_indx = src_slot->s_cindx;
+
+ c_src = C_SEG_SLOT_FROM_INDEX(c_seg_src, c_indx);
+
+ c_size = UNPACK_C_SIZE(c_src);
+
+ assert(c_size);
+
+ if (c_size > (uint32_t)(C_SEG_BUFSIZE - C_SEG_OFFSET_TO_BYTES((int32_t)c_seg_dst->c_nextoffset))) {
+ /*
+ * This segment is full. We need a new one.
+ */
+
+ PAGE_REPLACEMENT_DISALLOWED(TRUE);
+
+ lck_mtx_lock_spin_always(&c_seg_src->c_lock);
+ C_SEG_WAKEUP_DONE(c_seg_src);
+ lck_mtx_unlock_always(&c_seg_src->c_lock);
+
+ c_seg_src = NULL;
+
+ lck_mtx_lock_spin_always(&c_seg_dst->c_lock);
+
+ assert(c_seg_dst->c_busy);
+ assert(c_seg_dst->c_state == C_IS_FILLING);
+ assert(!c_seg_dst->c_on_minorcompact_q);
+
+ c_current_seg_filled(c_seg_dst, (c_segment_t *)current_chead);
+ assert(*current_chead == NULL);
+
+ C_SEG_WAKEUP_DONE(c_seg_dst);
+
+ lck_mtx_unlock_always(&c_seg_dst->c_lock);
+
+ c_seg_dst = NULL;
+
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
+
+ goto Relookup_dst;
+ }
+
+ c_dst = C_SEG_SLOT_FROM_INDEX(c_seg_dst, c_seg_dst->c_nextslot);
+
+ memcpy(&c_seg_dst->c_store.c_buffer[c_seg_dst->c_nextoffset], &c_seg_src->c_store.c_buffer[c_src->c_offset], c_size);
+
+ c_rounded_size = (c_size + C_SEG_OFFSET_ALIGNMENT_MASK) & ~C_SEG_OFFSET_ALIGNMENT_MASK;
+
+ cslot_copy(c_dst, c_src);
+ c_dst->c_offset = c_seg_dst->c_nextoffset;
+
+ if (c_seg_dst->c_firstemptyslot == c_seg_dst->c_nextslot)
+ c_seg_dst->c_firstemptyslot++;
+
+ c_seg_dst->c_nextslot++;
+ c_seg_dst->c_bytes_used += c_rounded_size;
+ c_seg_dst->c_nextoffset += C_SEG_BYTES_TO_OFFSET(c_rounded_size);
+
+
+ PACK_C_SIZE(c_src, 0);
+
+ c_seg_src->c_bytes_used -= c_rounded_size;
+ c_seg_src->c_bytes_unused += c_rounded_size;
+
+ if (c_indx < c_seg_src->c_firstemptyslot) {
+ c_seg_src->c_firstemptyslot = c_indx;
+ }
+
+ c_dst = C_SEG_SLOT_FROM_INDEX(c_seg_dst, dst_slot);
+
+ PAGE_REPLACEMENT_ALLOWED(TRUE);
+ slot_ptr = (c_slot_mapping_t)C_SLOT_UNPACK_PTR(c_dst);
+ /* <csegno=0,indx=0> would mean "empty slot", so use csegno+1 */
+ slot_ptr->s_cseg = c_seg_dst->c_mysegno + 1;
+ slot_ptr->s_cindx = dst_slot;
+
+ PAGE_REPLACEMENT_ALLOWED(FALSE);
+
+out:
+ if (c_seg_src) {
+
+ lck_mtx_lock_spin_always(&c_seg_src->c_lock);
+
+ C_SEG_WAKEUP_DONE(c_seg_src);
+
+ if (c_seg_src->c_bytes_used == 0 && c_seg_src->c_state != C_IS_FILLING) {
+ if (!c_seg_src->c_on_minorcompact_q)
+ c_seg_need_delayed_compaction(c_seg_src, FALSE);
+ }
+
+ lck_mtx_unlock_always(&c_seg_src->c_lock);
+ }
+
+ if (c_seg_dst) {
+
+ PAGE_REPLACEMENT_DISALLOWED(TRUE);
+
+ lck_mtx_lock_spin_always(&c_seg_dst->c_lock);
+
+ if (c_seg_dst->c_nextoffset >= C_SEG_OFF_LIMIT || c_seg_dst->c_nextslot >= C_SLOT_MAX_INDEX) {
+ /*
+ * Nearing or exceeded maximum slot and offset capacity.
+ */
+ assert(c_seg_dst->c_busy);
+ assert(c_seg_dst->c_state == C_IS_FILLING);
+ assert(!c_seg_dst->c_on_minorcompact_q);
+
+ c_current_seg_filled(c_seg_dst, (c_segment_t *)current_chead);
+ assert(*current_chead == NULL);
+ }
+
+ C_SEG_WAKEUP_DONE(c_seg_dst);
+
+ lck_mtx_unlock_always(&c_seg_dst->c_lock);
+
+ c_seg_dst = NULL;
+
+ PAGE_REPLACEMENT_DISALLOWED(FALSE);
+ }
+
+ return kr;
+}
+#endif /* CONFIG_FREEZE */
projects / apple / xnu.git / blobdiff