X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/b0d623f7f2ae71ed96e60569f61f9a9a27016e80..13f56ec4e58bf8687e2a68032c093c0213dd519b:/osfmk/kern/zalloc.c diff --git a/osfmk/kern/zalloc.c b/osfmk/kern/zalloc.c index d1d08bf96..9d1afa5d6 100644 --- a/osfmk/kern/zalloc.c +++ b/osfmk/kern/zalloc.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2009 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -71,6 +71,7 @@ #include #include #include +#include #include #include @@ -94,15 +95,9 @@ #include #include +#include #include -#if defined(__ppc__) -/* for fake zone stat routines */ -#include -#include -#endif - - /* * Zone Corruption Debugging * @@ -114,7 +109,7 @@ * each other when re-using the zone element, to detect modifications. * (3) poison the freed memory by overwriting it with 0xdeadbeef. * - * The first two checks are farily light weight and are enabled by specifying "-zc" + * The first two checks are fairly light weight and are enabled by specifying "-zc" * in the boot-args. If you want more aggressive checking for use-after-free bugs * and you don't mind the additional overhead, then turn on poisoning by adding * "-zp" to the boot-args in addition to "-zc". If you specify -zp without -zc, @@ -125,6 +120,48 @@ boolean_t check_freed_element = FALSE; /* enabled by -zc in boot-args */ boolean_t zfree_clear = FALSE; /* enabled by -zp in boot-args */ +/* + * 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 struct fake_zone_info fake_zones[] = { + { + .name = "kernel_stacks", + .init = stack_fake_zone_init, + .query = stack_fake_zone_info, + }, +#if defined(__i386__) || defined (__x86_64__) + { + .name = "page_tables", + .init = pt_fake_zone_init, + .query = pt_fake_zone_info, + }, +#endif /* i386 */ + { + .name = "kalloc.large", + .init = kalloc_fake_zone_init, + .query = kalloc_fake_zone_info, + }, +}; +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) + +/* + * Allocation helper macros + */ #define is_kernel_data_addr(a) (!(a) || ((a) >= vm_min_kernel_address && !((a) & 0x3))) #define ADD_TO_ZONE(zone, element) \ @@ -159,13 +196,14 @@ MACRO_BEGIN \ if (zfree_clear) { \ unsigned int ii; \ for (ii = sizeof(vm_offset_t) / sizeof(uint32_t); \ - ii < zone->elem_size/sizeof(uint32_t) - sizeof(vm_offset_t) / sizeof(uint32_t); \ + ii < (zone)->elem_size/sizeof(uint32_t) - sizeof(vm_offset_t) / sizeof(uint32_t); \ ii++) \ if (((uint32_t *)(ret))[ii] != (uint32_t)0xdeadbeef) \ panic("a freed zone element has been modified");\ } \ } \ (zone)->count++; \ + (zone)->sum_count++; \ (zone)->free_elements = *((vm_offset_t *)(ret)); \ } \ MACRO_END @@ -177,27 +215,47 @@ MACRO_END #endif /* ZONE_DEBUG */ /* - * Support for garbage collection of unused zone pages: + * 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 { - struct zone_page_table_entry *link; - short alloc_count; - short collect_count; + 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, - int value); + vm_size_t size); void zone_page_alloc( vm_offset_t addr, vm_size_t size); void zone_page_free_element( - struct zone_page_table_entry **free_pages, + zone_page_index_t *free_page_list, vm_offset_t addr, vm_size_t size); @@ -229,9 +287,12 @@ 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; @@ -265,22 +326,23 @@ MACRO_END #define lock_try_zone(zone) lck_mtx_try_lock_spin(&zone->lock) -kern_return_t zget_space( - vm_offset_t size, - vm_offset_t *result); - -decl_simple_lock_data(,zget_space_lock) -vm_offset_t zalloc_next_space; -vm_offset_t zalloc_end_of_space; -vm_size_t zalloc_wasted_space; - /* * Garbage collection map information */ -struct zone_page_table_entry * zone_page_table; +#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; vm_offset_t zone_map_min_address; vm_offset_t zone_map_max_address; 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 @@ -303,10 +365,6 @@ lck_mtx_ext_t zone_lck_ext; ((vm_offset_t)(zone_virtual_addr((vm_map_address_t)addr)) + size -1) < zone_map_max_address) #endif -#define ZONE_PAGE_USED 0 -#define ZONE_PAGE_UNUSED -1 - - /* * Protects first_zone, last_zone, num_zones, * and the next_zone field of zones. @@ -319,8 +377,7 @@ unsigned int num_zones; boolean_t zone_gc_allowed = TRUE; boolean_t zone_gc_forced = FALSE; boolean_t panic_include_zprint = FALSE; -unsigned zone_gc_last_tick = 0; -unsigned zone_gc_max_rate = 0; /* in ticks */ +boolean_t zone_gc_allowed_by_time_throttle = TRUE; /* * Zone leak debugging code @@ -365,15 +422,13 @@ static char zone_name_to_log[MAX_ZONE_NAME] = ""; /* the zone name we're logging * 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. - * - * MAX_DEPTH configures how deep of a stack trace is taken on each zalloc in the zone of interrest. 15 - * levels is usually enough to get past all the layers of code in kalloc and IOKit and see who the actual - * caller is up above these lower levels. */ - +#if defined(__LP64__) +#define ZRECORDS_MAX 16000 /* 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 */ -#define MAX_DEPTH 15 /* number of levels of the stack trace to record */ /* * Each record in the log contains a pointer to the zone element it refers to, a "time" number that allows @@ -387,7 +442,7 @@ struct zrecord { void *z_element; /* the element that was zalloc'ed of zfree'ed */ uint32_t z_opcode:1, /* whether it was a zalloc or zfree */ z_time:31; /* time index when operation was done */ - void *z_pc[MAX_DEPTH]; /* stack trace of caller */ + void *z_pc[MAX_ZTRACE_DEPTH]; /* stack trace of caller */ }; /* @@ -457,7 +512,526 @@ log_this_zone(const char *zonename, const char *logname) extern boolean_t zlog_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 z_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 z_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 */ +#define ZLEAK_ALLOCATION_MAP_NUM 16384 +#define ZLEAK_TRACE_MAP_NUM 8192 + +uint32_t zleak_alloc_buckets = ZLEAK_ALLOCATION_MAP_NUM; +uint32_t zleak_trace_buckets = 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_mtx_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", &z_sample_factor, sizeof(z_sample_factor))) { + printf("Zone leak factor override:%u\n", z_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!"); + } + } + + /* 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!"); + } + } + + /* 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_mtx_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 + * 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_mtx_lock_spin(&zleak_lock); + if (zleak_state & (ZLEAK_STATE_ACTIVE | ZLEAK_STATE_ACTIVATING | ZLEAK_STATE_FAILED)) { + lck_mtx_unlock(&zleak_lock); + return KERN_SUCCESS; + } + + zleak_state |= ZLEAK_STATE_ACTIVATING; + lck_mtx_unlock(&zleak_lock); + + /* Allocate and zero tables */ + retval = kmem_alloc_kobject(kernel_map, (vm_offset_t*)&allocations_ptr, z_alloc_size); + if (retval != KERN_SUCCESS) { + goto fail; + } + + retval = kmem_alloc_kobject(kernel_map, (vm_offset_t*)&traces_ptr, z_trace_size); + 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_mtx_lock_spin(&zleak_lock); + zleak_state |= ZLEAK_STATE_ACTIVE; + zleak_state &= ~ZLEAK_STATE_ACTIVATING; + lck_mtx_unlock(&zleak_lock); + + return 0; + +fail: + /* + * If we fail to allocate memory, don't further tax + * the system by trying again. + */ + lck_mtx_lock_spin(&zleak_lock); + zleak_state |= ZLEAK_STATE_FAILED; + zleak_state &= ~ZLEAK_STATE_ACTIVATING; + lck_mtx_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_mtx_try_lock_spin(&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_mtx_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_mtx_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_mtx_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_mtx_lock_spin(&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_mtx_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. + * This seems to work for x86 and X86_64. + * ARMTODO: Test it on ARM, I think it will work but I can't test it. If it works, remove the ifdef. + * 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) +{ +#if defined(__x86_64__) || defined(__i386__) + uintptr_t* frameptr = NULL, *frameptr_next = NULL; + uintptr_t retaddr = 0; + uint32_t frame_index = 0, frames = 0; + uintptr_t kstackb, kstackt; + + kstackb = current_thread()->kernel_stack; + kstackt = kstackb + kernel_stack_size; + /* Load stack frame pointer (EBP on x86) into frameptr */ + frameptr = __builtin_frame_address(0); + + 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; +#else + return OSBacktrace((void*)bt, max_frames); +#endif +} + +/* "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 @@ -473,9 +1047,10 @@ zinit( zone_t z; if (zone_zone == ZONE_NULL) { - if (zget_space(sizeof(struct zone), (vm_offset_t *)&z) - != KERN_SUCCESS) - return(ZONE_NULL); + + z = (struct zone *)zdata; + zdata += sizeof(*z); + zdata_size -= sizeof(*z); } else z = (zone_t) zalloc(zone_zone); if (z == ZONE_NULL) @@ -505,7 +1080,11 @@ zinit( alloc = PAGE_SIZE; else #endif - { vm_size_t best, waste; unsigned int i; +#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; @@ -536,6 +1115,7 @@ use_this_allocation: z->alloc_size = alloc; z->zone_name = name; z->count = 0; + z->sum_count = 0LL; z->doing_alloc = FALSE; z->doing_gc = FALSE; z->exhaustible = FALSE; @@ -544,6 +1124,18 @@ use_this_allocation: 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->prio_refill_watermark = 0; + z->zone_replenish_thread = NULL; +#if CONFIG_ZLEAKS + z->num_allocs = 0; + z->num_frees = 0; + z->zleak_capture = 0; + z->zleak_on = FALSE; +#endif /* CONFIG_ZLEAKS */ #if ZONE_DEBUG z->active_zones.next = z->active_zones.prev = NULL; @@ -553,13 +1145,20 @@ use_this_allocation: /* * 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; thread_call_setup(&z->call_async_alloc, zalloc_async, z); 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); @@ -599,18 +1198,98 @@ use_this_allocation: return(z); } +unsigned zone_replenish_loops, zone_replenish_wakeups, zone_replenish_wakeups_initiated; + +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 == 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); + + 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); + 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); + 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. */ void zcram( - register zone_t zone, - void *newaddr, + zone_t zone, + vm_offset_t newmem, vm_size_t size) { - register vm_size_t elem_size; - vm_offset_t newmem = (vm_offset_t) newaddr; + vm_size_t elem_size; + boolean_t from_zm = FALSE; /* Basic sanity checks */ assert(zone != ZONE_NULL && newmem != (vm_offset_t)0); @@ -619,10 +1298,16 @@ zcram( elem_size = zone->elem_size; + if (from_zone_map(newmem, size)) + from_zm = TRUE; + + if (from_zm) + zone_page_init(newmem, size); + lock_zone(zone); while (size >= elem_size) { ADD_TO_ZONE(zone, newmem); - if (from_zone_map(newmem, elem_size)) + if (from_zm) zone_page_alloc(newmem, elem_size); zone->count++; /* compensate for ADD_TO_ZONE */ size -= elem_size; @@ -632,90 +1317,6 @@ zcram( unlock_zone(zone); } -/* - * Contiguous space allocator for non-paged zones. Allocates "size" amount - * of memory from zone_map. - */ - -kern_return_t -zget_space( - vm_offset_t size, - vm_offset_t *result) -{ - vm_offset_t new_space = 0; - vm_size_t space_to_add = 0; - - simple_lock(&zget_space_lock); - while ((zalloc_next_space + size) > zalloc_end_of_space) { - /* - * Add at least one page to allocation area. - */ - - space_to_add = round_page(size); - - if (new_space == 0) { - kern_return_t retval; - /* - * Memory cannot be wired down while holding - * any locks that the pageout daemon might - * need to free up pages. [Making the zget_space - * lock a complex lock does not help in this - * regard.] - * - * Unlock and allocate memory. Because several - * threads might try to do this at once, don't - * use the memory before checking for available - * space again. - */ - - simple_unlock(&zget_space_lock); - - retval = kernel_memory_allocate(zone_map, &new_space, - space_to_add, 0, KMA_KOBJECT|KMA_NOPAGEWAIT); - if (retval != KERN_SUCCESS) - return(retval); -#if ZONE_ALIAS_ADDR - if (space_to_add == PAGE_SIZE) - new_space = zone_alias_addr(new_space); -#endif - zone_page_init(new_space, space_to_add, - ZONE_PAGE_USED); - simple_lock(&zget_space_lock); - continue; - } - - - /* - * Memory was allocated in a previous iteration. - * - * Check whether the new region is contiguous - * with the old one. - */ - - if (new_space != zalloc_end_of_space) { - /* - * Throw away the remainder of the - * old space, and start a new one. - */ - zalloc_wasted_space += - zalloc_end_of_space - zalloc_next_space; - zalloc_next_space = new_space; - } - - zalloc_end_of_space = new_space + space_to_add; - - new_space = 0; - } - *result = zalloc_next_space; - zalloc_next_space += size; - simple_unlock(&zget_space_lock); - - if (new_space != 0) - kmem_free(zone_map, new_space, space_to_add); - - return(KERN_SUCCESS); -} - /* * Steal memory for the zone package. Called from @@ -724,8 +1325,9 @@ zget_space( void zone_steal_memory(void) { - zdata_size = round_page(128*sizeof(struct zone)); - zdata = (vm_offset_t)((char *)pmap_steal_memory(zdata_size) - (char *)0); + /* Request enough early memory to get to the pmap zone */ + zdata_size = 12 * sizeof(struct zone); + zdata = (vm_offset_t)pmap_steal_memory(round_page(zdata_size)); } @@ -756,7 +1358,7 @@ zfill( return 0; zone_change(zone, Z_FOREIGN, TRUE); - zcram(zone, (void *)memory, size); + zcram(zone, memory, size); nalloc = (int)(size / zone->elem_size); assert(nalloc >= nelem); @@ -771,10 +1373,26 @@ zfill( void zone_bootstrap(void) { - vm_size_t zone_zone_size; - vm_offset_t zone_zone_space; char temp_buf[16]; +#if 6094439 + /* enable zone checks by default, to try and catch offenders... */ +#if 0 + /* 7968354: turn "-zc" back off */ + check_freed_element = TRUE; + /* 7995202: turn "-zp" back off */ + zfree_clear = TRUE; +#endif + + /* ... but allow them to be turned off explicitely */ + if (PE_parse_boot_argn("-no_zc", temp_buf, sizeof (temp_buf))) { + check_freed_element = FALSE; + } + if (PE_parse_boot_argn("-no_zp", temp_buf, sizeof (temp_buf))) { + zfree_clear = FALSE; + } +#endif + /* see if we want freed zone element checking and/or poisoning */ if (PE_parse_boot_argn("-zc", temp_buf, sizeof (temp_buf))) { check_freed_element = TRUE; @@ -784,6 +1402,10 @@ zone_bootstrap(void) zfree_clear = TRUE; } + if (PE_parse_boot_argn("-zinfop", temp_buf, sizeof (temp_buf))) { + zinfo_per_task = TRUE; + } + /* * Check for and set up zone leak detection if requested via boot-args. We recognized two * boot-args: @@ -817,21 +1439,50 @@ zone_bootstrap(void) last_zone = &first_zone; num_zones = 0; - simple_lock_init(&zget_space_lock, 0); - zalloc_next_space = zdata; - zalloc_end_of_space = zdata + zdata_size; - zalloc_wasted_space = 0; - /* assertion: nobody else called zinit before us */ assert(zone_zone == ZONE_NULL); zone_zone = zinit(sizeof(struct zone), 128 * sizeof(struct zone), sizeof(struct zone), "zones"); zone_change(zone_zone, Z_COLLECT, FALSE); - zone_zone_size = zalloc_end_of_space - zalloc_next_space; - zget_space(zone_zone_size, &zone_zone_space); - zcram(zone_zone, (void *)zone_zone_space, zone_zone_size); + zone_change(zone_zone, Z_CALLERACCT, FALSE); + zone_change(zone_zone, Z_NOENCRYPT, TRUE); + + zcram(zone_zone, zdata, zdata_size); + + /* initialize fake zones and zone info if tracking by task */ + if (zinfo_per_task) { + vm_size_t zisize = sizeof(zinfo_usage_store_t) * ZINFO_SLOTS; + unsigned int i; + + for (i = 0; i < num_fake_zones; i++) + fake_zones[i].init(ZINFO_SLOTS - num_fake_zones + i); + zinfo_zone = zinit(zisize, zisize * CONFIG_TASK_MAX, + zisize, "per task zinfo"); + zone_change(zinfo_zone, Z_CALLERACCT, FALSE); + } } +void +zinfo_task_init(task_t task) +{ + if (zinfo_per_task) { + task->tkm_zinfo = zalloc(zinfo_zone); + memset(task->tkm_zinfo, 0, sizeof(zinfo_usage_store_t) * ZINFO_SLOTS); + } else { + task->tkm_zinfo = NULL; + } +} + +void +zinfo_task_free(task_t task) +{ + assert(task != kernel_task); + if (task->tkm_zinfo != NULL) { + zfree(zinfo_zone, task->tkm_zinfo); + task->tkm_zinfo = NULL; + } +} + void zone_init( vm_size_t max_zonemap_size) @@ -839,7 +1490,6 @@ zone_init( kern_return_t retval; vm_offset_t zone_min; vm_offset_t zone_max; - vm_size_t zone_table_size; retval = kmem_suballoc(kernel_map, &zone_min, max_zonemap_size, FALSE, VM_FLAGS_ANYWHERE | VM_FLAGS_PERMANENT, @@ -851,26 +1501,105 @@ zone_init( /* * Setup garbage collection information: */ - zone_table_size = atop_kernel(zone_max - zone_min) * - sizeof(struct zone_page_table_entry); - if (kmem_alloc_kobject(zone_map, (vm_offset_t *) &zone_page_table, - zone_table_size) != KERN_SUCCESS) - panic("zone_init"); - zone_min = (vm_offset_t)zone_page_table + round_page(zone_table_size); - zone_pages = (unsigned int)atop_kernel(zone_max - zone_min); zone_map_min_address = zone_min; zone_map_max_address = zone_max; + + zone_pages = (unsigned int)atop_kernel(zone_max - zone_min); + zone_page_table_used_size = sizeof(zone_page_table); + + zone_page_table_second_level_size = 1; + zone_page_table_second_level_shift_amount = 0; + + /* + * Find the power of 2 for the second level that allows + * the first level to fit in ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE + * slots. + */ + while ((zone_page_table_first_level_slot(zone_pages-1)) >= ZONE_PAGE_TABLE_FIRST_LEVEL_SIZE) { + zone_page_table_second_level_size <<= 1; + zone_page_table_second_level_shift_amount++; + } lck_grp_attr_setdefault(&zone_lck_grp_attr); lck_grp_init(&zone_lck_grp, "zones", &zone_lck_grp_attr); lck_attr_setdefault(&zone_lck_attr); lck_mtx_init_ext(&zone_gc_lock, &zone_lck_ext, &zone_lck_grp, &zone_lck_attr); - zone_page_init(zone_min, zone_max - zone_min, ZONE_PAGE_UNUSED); +#if CONFIG_ZLEAKS + /* + * Initialize the zone leak monitor + */ + zleak_init(max_zonemap_size); +#endif /* CONFIG_ZLEAKS */ +} + +void +zone_page_table_expand(zone_page_index_t pindex) +{ + unsigned int first_index; + struct zone_page_table_entry * volatile * first_level_ptr; + + assert(pindex < zone_pages); + + first_index = zone_page_table_first_level_slot(pindex); + first_level_ptr = &zone_page_table[first_index]; + + if (*first_level_ptr == NULL) { + /* + * We were able to verify the old first-level slot + * had NULL, so attempt to populate it. + */ + + vm_offset_t second_level_array = 0; + vm_size_t second_level_size = round_page(zone_page_table_second_level_size * sizeof(struct zone_page_table_entry)); + zone_page_index_t i; + struct zone_page_table_entry *entry_array; + + if (kmem_alloc_kobject(zone_map, &second_level_array, + second_level_size) != KERN_SUCCESS) { + panic("zone_page_table_expand"); + } + + /* + * zone_gc() may scan the "zone_page_table" directly, + * so make sure any slots have a valid unused state. + */ + entry_array = (struct zone_page_table_entry *)second_level_array; + for (i=0; i < zone_page_table_second_level_size; i++) { + entry_array[i].alloc_count = ZONE_PAGE_UNUSED; + entry_array[i].collect_count = 0; + } + + if (OSCompareAndSwapPtr(NULL, entry_array, first_level_ptr)) { + /* Old slot was NULL, replaced with expanded level */ + OSAddAtomicLong(second_level_size, &zone_page_table_used_size); + } else { + /* Old slot was not NULL, someone else expanded first */ + kmem_free(zone_map, second_level_array, second_level_size); + } + } else { + /* Old slot was not NULL, already been expanded */ + } +} + +struct zone_page_table_entry * +zone_page_table_lookup(zone_page_index_t pindex) +{ + unsigned int first_index = zone_page_table_first_level_slot(pindex); + struct zone_page_table_entry *second_level = zone_page_table[first_index]; + + if (second_level) { + return &second_level[zone_page_table_second_level_slot(pindex)]; + } + + return NULL; } extern volatile SInt32 kfree_nop_count; +#pragma mark - +#pragma mark zalloc_canblock + /* * zalloc returns an element from the specified zone. */ @@ -881,27 +1610,48 @@ zalloc_canblock( { vm_offset_t addr; kern_return_t retval; - void *bt[MAX_DEPTH]; /* only used if zone logging is enabled */ + uintptr_t zbt[MAX_ZTRACE_DEPTH]; /* used in zone leak logging and zone leak detection */ int numsaved = 0; - int i; + int i; + boolean_t zone_replenish_wakeup = FALSE; + +#if CONFIG_ZLEAKS + uint32_t zleak_tracedepth = 0; /* log this allocation if nonzero */ +#endif /* CONFIG_ZLEAKS */ assert(zone != ZONE_NULL); + + lock_zone(zone); /* * If zone logging is turned on and this is the zone we're tracking, grab a backtrace. */ - + if (DO_LOGGING(zone)) - numsaved = OSBacktrace(&bt[0], MAX_DEPTH); - - lock_zone(zone); + numsaved = OSBacktrace((void*) zbt, MAX_ZTRACE_DEPTH); + +#if CONFIG_ZLEAKS + /* + * Zone leak detection: capture a backtrace every z_sample_factor + * allocations in this zone. + */ + if (zone->zleak_on && (zone->zleak_capture++ % z_sample_factor == 0)) { + zone->zleak_capture = 1; + + /* Avoid backtracing twice if zone logging is on */ + if (numsaved == 0 ) + zleak_tracedepth = fastbacktrace(zbt, MAX_ZTRACE_DEPTH); + else + zleak_tracedepth = numsaved; + } +#endif /* CONFIG_ZLEAKS */ REMOVE_FROM_ZONE(zone, addr, vm_offset_t); - while ((addr == 0) && canblock && (zone->doing_gc)) { - zone->waiting = TRUE; - zone_sleep(zone); - REMOVE_FROM_ZONE(zone, addr, vm_offset_t); + if (zone->async_prio_refill && + ((zone->cur_size - (zone->count * zone->elem_size)) < (zone->prio_refill_watermark * zone->elem_size))) { + zone_replenish_wakeup = TRUE; + zone_replenish_wakeups_initiated++; } while ((addr == 0) && canblock) { @@ -915,8 +1665,20 @@ zalloc_canblock( */ zone->waiting = TRUE; zone_sleep(zone); - } - else { + } else if (zone->doing_gc) { + /* zone_gc() is running. Since we need an element + * from the free list that is currently being + * collected, set the waiting bit and try to + * interrupt the GC process, and try again + * when we obtain the lock. + */ + zone->waiting = TRUE; + zone_sleep(zone); + } else { + vm_offset_t space; + vm_size_t alloc_size; + int retry = 0; + if ((zone->cur_size + zone->elem_size) > zone->max_size) { if (zone->exhaustible) @@ -942,103 +1704,104 @@ zalloc_canblock( zone->doing_alloc = TRUE; unlock_zone(zone); - if (zone->collectable) { - vm_offset_t space; - vm_size_t alloc_size; - int retry = 0; - - for (;;) { - - if (vm_pool_low() || retry >= 1) - alloc_size = - round_page(zone->elem_size); - else - alloc_size = zone->alloc_size; - - retval = kernel_memory_allocate(zone_map, - &space, alloc_size, 0, - KMA_KOBJECT|KMA_NOPAGEWAIT); - if (retval == KERN_SUCCESS) { + for (;;) { + int zflags = KMA_KOBJECT|KMA_NOPAGEWAIT; + + if (vm_pool_low() || retry >= 1) + alloc_size = + round_page(zone->elem_size); + else + alloc_size = zone->alloc_size; + + if (zone->noencrypt) + zflags |= KMA_NOENCRYPT; + + retval = kernel_memory_allocate(zone_map, &space, alloc_size, 0, zflags); + if (retval == KERN_SUCCESS) { #if ZONE_ALIAS_ADDR - if (alloc_size == PAGE_SIZE) - space = zone_alias_addr(space); + if (alloc_size == PAGE_SIZE) + space = zone_alias_addr(space); #endif - zone_page_init(space, alloc_size, - ZONE_PAGE_USED); - zcram(zone, (void *)space, alloc_size); - - break; - } else if (retval != KERN_RESOURCE_SHORTAGE) { - retry++; - - if (retry == 2) { - zone_gc(); - printf("zalloc did gc\n"); - zone_display_zprint(); - } - if (retry == 3) { - panic_include_zprint = TRUE; - panic("zalloc: \"%s\" (%d elements) retry fail %d, kfree_nop_count: %d", zone->zone_name, zone->count, retval, (int)kfree_nop_count); - } - } else { - break; - } - } - lock_zone(zone); - zone->doing_alloc = FALSE; - if (zone->waiting) { - zone->waiting = FALSE; - zone_wakeup(zone); - } - REMOVE_FROM_ZONE(zone, addr, vm_offset_t); - if (addr == 0 && - retval == KERN_RESOURCE_SHORTAGE) { - unlock_zone(zone); - VM_PAGE_WAIT(); - lock_zone(zone); - } - } else { - vm_offset_t space; - retval = zget_space(zone->elem_size, &space); - - lock_zone(zone); - zone->doing_alloc = FALSE; - if (zone->waiting) { - zone->waiting = FALSE; - thread_wakeup((event_t)zone); - } - if (retval == KERN_SUCCESS) { - zone->count++; - zone->cur_size += zone->elem_size; -#if ZONE_DEBUG - if (zone_debug_enabled(zone)) { - enqueue_tail(&zone->active_zones, (queue_entry_t)space); +#if CONFIG_ZLEAKS + if ((zleak_state & (ZLEAK_STATE_ENABLED | ZLEAK_STATE_ACTIVE)) == ZLEAK_STATE_ENABLED) { + if (zone_map->size >= zleak_global_tracking_threshold) { + kern_return_t kr; + + kr = zleak_activate(); + if (kr != KERN_SUCCESS) { + printf("Failed to activate live zone leak debugging (%d).\n", kr); + } + } } -#endif - unlock_zone(zone); - zone_page_alloc(space, zone->elem_size); -#if ZONE_DEBUG - if (zone_debug_enabled(zone)) - space += ZONE_DEBUG_OFFSET; -#endif - addr = space; - goto success; - } - if (retval == KERN_RESOURCE_SHORTAGE) { - unlock_zone(zone); - VM_PAGE_WAIT(); - lock_zone(zone); + if ((zleak_state & ZLEAK_STATE_ACTIVE) && !(zone->zleak_on)) { + if (zone->cur_size > zleak_per_zone_tracking_threshold) { + zone->zleak_on = TRUE; + } + } +#endif /* CONFIG_ZLEAKS */ + + zcram(zone, space, alloc_size); + + break; + } else if (retval != KERN_RESOURCE_SHORTAGE) { + retry++; + + if (retry == 2) { + zone_gc(); + printf("zalloc did gc\n"); + zone_display_zprint(); + } + if (retry == 3) { + panic_include_zprint = TRUE; +#if CONFIG_ZLEAKS + if ((zleak_state & ZLEAK_STATE_ACTIVE)) { + panic_include_ztrace = TRUE; + } +#endif /* CONFIG_ZLEAKS */ + /* TODO: Change this to something more descriptive, perhaps + * 'zone_map exhausted' only if we get retval 3 (KERN_NO_SPACE). + */ + panic("zalloc: \"%s\" (%d elements) retry fail %d, kfree_nop_count: %d", zone->zone_name, zone->count, retval, (int)kfree_nop_count); + } } else { - panic("zalloc: \"%s\" (%d elements) zget_space returned %d", zone->zone_name, zone->count, retval); + break; } } + lock_zone(zone); + zone->doing_alloc = FALSE; + if (zone->waiting) { + zone->waiting = FALSE; + zone_wakeup(zone); + } + REMOVE_FROM_ZONE(zone, addr, vm_offset_t); + if (addr == 0 && + retval == KERN_RESOURCE_SHORTAGE) { + unlock_zone(zone); + + VM_PAGE_WAIT(); + lock_zone(zone); + } } if (addr == 0) REMOVE_FROM_ZONE(zone, addr, vm_offset_t); } +#if CONFIG_ZLEAKS + /* Zone leak detection: + * If we're sampling this allocation, add it to the zleaks hash table. + */ + if (addr && zleak_tracedepth > 0) { + /* Sampling can fail if another sample is happening at the same time in a different zone. */ + if (!zleak_log(zbt, addr, zleak_tracedepth, zone->elem_size)) { + /* If it failed, roll back the counter so we sample the next allocation instead. */ + zone->zleak_capture = z_sample_factor; + } + } +#endif /* CONFIG_ZLEAKS */ + + /* * See if we should be logging allocations in this zone. Logging is rarely done except when a leak is * suspected, so this code rarely executes. We need to do this code while still holding the zone lock @@ -1098,9 +1861,9 @@ empty_slot: zrecords[zcurrent].z_opcode = ZOP_ALLOC; for (i = 0; i < numsaved; i++) - zrecords[zcurrent].z_pc[i] = bt[i]; + zrecords[zcurrent].z_pc[i] = (void*) zbt[i]; - for (; i < MAX_DEPTH; i++) + for (; i < MAX_ZTRACE_DEPTH; i++) zrecords[zcurrent].z_pc[i] = 0; zcurrent++; @@ -1109,7 +1872,7 @@ empty_slot: zcurrent = 0; } - if ((addr == 0) && !canblock && (zone->async_pending == FALSE) && (zone->exhaustible == FALSE) && (!vm_pool_low())) { + if ((addr == 0) && !canblock && (zone->async_pending == FALSE) && (zone->no_callout == FALSE) && (zone->exhaustible == FALSE) && (!vm_pool_low())) { zone->async_pending = TRUE; unlock_zone(zone); thread_call_enter(&zone->call_async_alloc); @@ -1123,12 +1886,33 @@ empty_slot: addr += ZONE_DEBUG_OFFSET; } #endif + +#if CONFIG_ZLEAKS + if (addr != 0) { + zone->num_allocs++; + } +#endif /* CONFIG_ZLEAKS */ unlock_zone(zone); -success: + if (zone_replenish_wakeup) + thread_wakeup(&zone->zone_replenish_thread); + TRACE_MACHLEAKS(ZALLOC_CODE, ZALLOC_CODE_2, zone->elem_size, addr); + if (addr) { + thread_t thr = current_thread(); + task_t task; + zinfo_usage_t zinfo; + + if (zone->caller_acct) + thr->tkm_private.alloc += zone->elem_size; + else + thr->tkm_shared.alloc += zone->elem_size; + + if ((task = thr->task) != NULL && (zinfo = task->tkm_zinfo) != NULL) + OSAddAtomic64(zone->elem_size, (int64_t *)&zinfo[zone->index].alloc); + } return((void *)addr); } @@ -1168,17 +1952,36 @@ zalloc_async( * * This form should be used when you can not block (like when * processing an interrupt). + * + * XXX: It seems like only vm_page_grab_fictitious_common uses this, and its + * friend vm_page_more_fictitious can block, so it doesn't seem like + * this is used for interrupts any more.... */ void * zget( register zone_t zone) { register vm_offset_t addr; + +#if CONFIG_ZLEAKS + uintptr_t zbt[MAX_ZTRACE_DEPTH]; /* used for zone leak detection */ + uint32_t zleak_tracedepth = 0; /* log this allocation if nonzero */ +#endif /* CONFIG_ZLEAKS */ assert( zone != ZONE_NULL ); if (!lock_try_zone(zone)) return NULL; + +#if CONFIG_ZLEAKS + /* + * Zone leak detection: capture a backtrace + */ + if (zone->zleak_on && (zone->zleak_capture++ % z_sample_factor == 0)) { + zone->zleak_capture = 1; + zleak_tracedepth = fastbacktrace(zbt, MAX_ZTRACE_DEPTH); + } +#endif /* CONFIG_ZLEAKS */ REMOVE_FROM_ZONE(zone, addr, vm_offset_t); #if ZONE_DEBUG @@ -1187,6 +1990,24 @@ zget( addr += ZONE_DEBUG_OFFSET; } #endif /* ZONE_DEBUG */ + +#if CONFIG_ZLEAKS + /* + * Zone leak detection: record the allocation + */ + if (zone->zleak_on && zleak_tracedepth > 0 && addr) { + /* Sampling can fail if another sample is happening at the same time in a different zone. */ + if (!zleak_log(zbt, addr, zleak_tracedepth, zone->elem_size)) { + /* If it failed, roll back the counter so we sample the next allocation instead. */ + zone->zleak_capture = z_sample_factor; + } + } + + if (addr != 0) { + zone->num_allocs++; + } +#endif /* CONFIG_ZLEAKS */ + unlock_zone(zone); return((void *) addr); @@ -1205,7 +2026,7 @@ zfree( void *addr) { vm_offset_t elem = (vm_offset_t) addr; - void *bt[MAX_DEPTH]; /* only used if zone logging is enable via boot-args */ + void *zbt[MAX_ZTRACE_DEPTH]; /* only used if zone logging is enabled via boot-args */ int numsaved = 0; assert(zone != ZONE_NULL); @@ -1215,7 +2036,7 @@ zfree( */ if (DO_LOGGING(zone)) - numsaved = OSBacktrace(&bt[0], MAX_DEPTH); + numsaved = OSBacktrace(&zbt[0], MAX_ZTRACE_DEPTH); #if MACH_ASSERT /* Basic sanity checks */ @@ -1263,9 +2084,9 @@ zfree( zrecords[zcurrent].z_opcode = ZOP_FREE; for (i = 0; i < numsaved; i++) - zrecords[zcurrent].z_pc[i] = bt[i]; + zrecords[zcurrent].z_pc[i] = zbt[i]; - for (; i < MAX_DEPTH; i++) + for (; i < MAX_ZTRACE_DEPTH; i++) zrecords[zcurrent].z_pc[i] = 0; zcurrent++; @@ -1310,7 +2131,7 @@ zfree( if (elem != (vm_offset_t)tmp_elem) panic("zfree()ing element from wrong zone"); } - remqueue(&zone->active_zones, (queue_t) elem); + remqueue((queue_t) elem); } #endif /* ZONE_DEBUG */ if (zone_check) { @@ -1329,7 +2150,19 @@ zfree( if (zone->count < 0) panic("zfree: count < 0!"); #endif + +#if CONFIG_ZLEAKS + zone->num_frees++; + + /* + * Zone leak detection: un-track the allocation + */ + if (zone->zleak_on) { + zleak_free(elem, zone->elem_size); + } +#endif /* CONFIG_ZLEAKS */ + /* * If elements have one or more pages, and memory is low, * request to run the garbage collection in the zone the next @@ -1340,6 +2173,20 @@ zfree( zone_gc_forced = TRUE; } unlock_zone(zone); + + { + thread_t thr = current_thread(); + task_t task; + zinfo_usage_t zinfo; + + if (zone->caller_acct) + thr->tkm_private.free += zone->elem_size; + else + thr->tkm_shared.free += zone->elem_size; + if ((task = thr->task) != NULL && (zinfo = task->tkm_zinfo) != NULL) + OSAddAtomic64(zone->elem_size, + (int64_t *)&zinfo[zone->index].free); + } } @@ -1356,6 +2203,9 @@ zone_change( assert( value == TRUE || value == FALSE ); switch(item){ + case Z_NOENCRYPT: + zone->noencrypt = value; + break; case Z_EXHAUST: zone->exhaustible = value; break; @@ -1368,6 +2218,12 @@ zone_change( case Z_FOREIGN: zone->allows_foreign = value; break; + case Z_CALLERACCT: + zone->caller_acct = value; + break; + case Z_NOCALLOUT: + zone->no_callout = value; + break; #if MACH_ASSERT default: panic("Zone_change: Wrong Item Type!"); @@ -1411,8 +2267,7 @@ zprealloc( if (size != 0) { if (kmem_alloc_kobject(zone_map, &addr, size) != KERN_SUCCESS) panic("zprealloc"); - zone_page_init(addr, size, ZONE_PAGE_USED); - zcram(zone, (void *)addr, size); + zcram(zone, addr, size); } } @@ -1426,7 +2281,7 @@ zone_page_collectable( vm_size_t size) { struct zone_page_table_entry *zp; - natural_t i, j; + zone_page_index_t i, j; #if ZONE_ALIAS_ADDR addr = zone_virtual_addr(addr); @@ -1436,12 +2291,14 @@ zone_page_collectable( panic("zone_page_collectable"); #endif - i = (natural_t)atop_kernel(addr-zone_map_min_address); - j = (natural_t)atop_kernel((addr+size-1) - zone_map_min_address); + i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address); + j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address); - for (zp = zone_page_table + i; i <= j; zp++, i++) + for (; i <= j; i++) { + zp = zone_page_table_lookup(i); if (zp->collect_count == zp->alloc_count) return (TRUE); + } return (FALSE); } @@ -1452,7 +2309,7 @@ zone_page_keep( vm_size_t size) { struct zone_page_table_entry *zp; - natural_t i, j; + zone_page_index_t i, j; #if ZONE_ALIAS_ADDR addr = zone_virtual_addr(addr); @@ -1462,11 +2319,13 @@ zone_page_keep( panic("zone_page_keep"); #endif - i = (natural_t)atop_kernel(addr-zone_map_min_address); - j = (natural_t)atop_kernel((addr+size-1) - zone_map_min_address); + i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address); + j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address); - for (zp = zone_page_table + i; i <= j; zp++, i++) + for (; i <= j; i++) { + zp = zone_page_table_lookup(i); zp->collect_count = 0; + } } void @@ -1475,7 +2334,7 @@ zone_page_collect( vm_size_t size) { struct zone_page_table_entry *zp; - natural_t i, j; + zone_page_index_t i, j; #if ZONE_ALIAS_ADDR addr = zone_virtual_addr(addr); @@ -1485,21 +2344,22 @@ zone_page_collect( panic("zone_page_collect"); #endif - i = (natural_t)atop_kernel(addr-zone_map_min_address); - j = (natural_t)atop_kernel((addr+size-1) - zone_map_min_address); + i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address); + j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address); - for (zp = zone_page_table + i; i <= j; zp++, i++) + for (; i <= j; i++) { + zp = zone_page_table_lookup(i); ++zp->collect_count; + } } void zone_page_init( vm_offset_t addr, - vm_size_t size, - int value) + vm_size_t size) { struct zone_page_table_entry *zp; - natural_t i, j; + zone_page_index_t i, j; #if ZONE_ALIAS_ADDR addr = zone_virtual_addr(addr); @@ -1509,11 +2369,16 @@ zone_page_init( panic("zone_page_init"); #endif - i = (natural_t)atop_kernel(addr-zone_map_min_address); - j = (natural_t)atop_kernel((addr+size-1) - zone_map_min_address); + i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address); + j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address); + + for (; i <= j; i++) { + /* make sure entry exists before marking unused */ + zone_page_table_expand(i); - for (zp = zone_page_table + i; i <= j; zp++, i++) { - zp->alloc_count = value; + zp = zone_page_table_lookup(i); + assert(zp); + zp->alloc_count = ZONE_PAGE_UNUSED; zp->collect_count = 0; } } @@ -1524,7 +2389,7 @@ zone_page_alloc( vm_size_t size) { struct zone_page_table_entry *zp; - natural_t i, j; + zone_page_index_t i, j; #if ZONE_ALIAS_ADDR addr = zone_virtual_addr(addr); @@ -1534,29 +2399,32 @@ zone_page_alloc( panic("zone_page_alloc"); #endif - i = (natural_t)atop_kernel(addr-zone_map_min_address); - j = (natural_t)atop_kernel((addr+size-1) - zone_map_min_address); + i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address); + j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address); + + for (; i <= j; i++) { + zp = zone_page_table_lookup(i); + assert(zp); - for (zp = zone_page_table + i; i <= j; zp++, i++) { /* - * Set alloc_count to (ZONE_PAGE_USED + 1) if + * Set alloc_count to ZONE_PAGE_USED if * it was previously set to ZONE_PAGE_UNUSED. */ if (zp->alloc_count == ZONE_PAGE_UNUSED) - zp->alloc_count = 1; - else - ++zp->alloc_count; + zp->alloc_count = ZONE_PAGE_USED; + + ++zp->alloc_count; } } void zone_page_free_element( - struct zone_page_table_entry **free_pages, + zone_page_index_t *free_page_list, vm_offset_t addr, vm_size_t size) { struct zone_page_table_entry *zp; - natural_t i, j; + zone_page_index_t i, j; #if ZONE_ALIAS_ADDR addr = zone_virtual_addr(addr); @@ -1566,18 +2434,28 @@ zone_page_free_element( panic("zone_page_free_element"); #endif - i = (natural_t)atop_kernel(addr-zone_map_min_address); - j = (natural_t)atop_kernel((addr+size-1) - zone_map_min_address); + i = (zone_page_index_t)atop_kernel(addr-zone_map_min_address); + j = (zone_page_index_t)atop_kernel((addr+size-1) - zone_map_min_address); + + for (; i <= j; i++) { + zp = zone_page_table_lookup(i); - for (zp = zone_page_table + i; i <= j; zp++, i++) { if (zp->collect_count > 0) --zp->collect_count; if (--zp->alloc_count == 0) { + vm_address_t free_page_address; + zp->alloc_count = ZONE_PAGE_UNUSED; zp->collect_count = 0; - zp->link = *free_pages; - *free_pages = zp; + + /* + * This element was the last one on this page, re-use the page's + * storage for a page freelist + */ + free_page_address = zone_map_min_address + PAGE_SIZE * ((vm_size_t)i); + *(zone_page_index_t *)free_page_address = *free_page_list; + *free_page_list = i; } } } @@ -1640,7 +2518,7 @@ zone_gc(void) unsigned int max_zones; zone_t z; unsigned int i; - struct zone_page_table_entry *zp, *zone_free_pages; + zone_page_index_t zone_free_page_head; lck_mtx_lock(&zone_gc_lock); @@ -1650,11 +2528,15 @@ zone_gc(void) simple_unlock(&all_zones_lock); #if MACH_ASSERT - for (i = 0; i < zone_pages; i++) - assert(zone_page_table[i].collect_count == 0); + for (i = 0; i < zone_pages; i++) { + struct zone_page_table_entry *zp; + + zp = zone_page_table_lookup(i); + assert(!zp || (zp->collect_count == 0)); + } #endif /* MACH_ASSERT */ - zone_free_pages = NULL; + zone_free_page_head = ZONE_PAGE_INDEX_INVALID; for (i = 0; i < max_zones; i++, z = z->next_zone) { unsigned int n, m; @@ -1735,6 +2617,7 @@ zone_gc(void) if (++n >= 50) { if (z->waiting == TRUE) { + /* z->waiting checked without lock held, rechecked below after locking */ lock_zone(z); if (keep != NULL) { @@ -1775,6 +2658,11 @@ zone_gc(void) ADD_LIST_TO_ZONE(z, keep, tail); + if (z->waiting) { + z->waiting = FALSE; + zone_wakeup(z); + } + unlock_zone(z); } @@ -1790,11 +2678,20 @@ zone_gc(void) n = 0; tail = keep = NULL; while (elt != NULL) { if (zone_page_collectable((vm_offset_t)elt, elt_size)) { + struct zone_free_element *next_elt = elt->next; + size_freed += elt_size; - zone_page_free_element(&zone_free_pages, + + /* + * If this is the last allocation on the page(s), + * we may use their storage to maintain the linked + * list of free-able pages. So store elt->next because + * "elt" may be scribbled over. + */ + zone_page_free_element(&zone_free_page_head, (vm_offset_t)elt, elt_size); - elt = elt->next; + elt = next_elt; ++zgc_stats.elems_freed; } @@ -1869,13 +2766,17 @@ zone_gc(void) * Reclaim the pages we are freeing. */ - while ((zp = zone_free_pages) != NULL) { - zone_free_pages = zp->link; + while (zone_free_page_head != ZONE_PAGE_INDEX_INVALID) { + zone_page_index_t zind = zone_free_page_head; + vm_address_t free_page_address; #if ZONE_ALIAS_ADDR - z = zone_virtual_addr((vm_map_address_t)z); + z = (zone_t)zone_virtual_addr((vm_map_address_t)z); #endif - kmem_free(zone_map, zone_map_min_address + PAGE_SIZE * - (zp - zone_page_table), PAGE_SIZE); + /* Use the first word of the page about to be freed to find the next free page */ + free_page_address = zone_map_min_address + PAGE_SIZE * ((vm_size_t)zind); + zone_free_page_head = *(zone_page_index_t *)free_page_address; + + kmem_free(zone_map, free_page_address, PAGE_SIZE); ++zgc_stats.pgs_freed; } @@ -1891,57 +2792,334 @@ zone_gc(void) void consider_zone_gc(boolean_t force) { - /* - * By default, don't attempt zone GC more frequently - * than once / 1 minutes. - */ - - if (zone_gc_max_rate == 0) - zone_gc_max_rate = (60 << SCHED_TICK_SHIFT) + 1; if (zone_gc_allowed && - ((sched_tick > (zone_gc_last_tick + zone_gc_max_rate)) || + (zone_gc_allowed_by_time_throttle || zone_gc_forced || force)) { zone_gc_forced = FALSE; - zone_gc_last_tick = sched_tick; + zone_gc_allowed_by_time_throttle = FALSE; /* reset periodically */ zone_gc(); } } -struct fake_zone_info { - const char* name; - void (*func)(int *, vm_size_t *, vm_size_t *, vm_size_t *, vm_size_t *, - int *, int *); -}; +/* + * By default, don't attempt zone GC more frequently + * than once / 1 minutes. + */ +void +compute_zone_gc_throttle(void *arg __unused) +{ + zone_gc_allowed_by_time_throttle = TRUE; +} -static struct fake_zone_info fake_zones[] = { - { - .name = "kernel_stacks", - .func = stack_fake_zone_info, - }, -#ifdef ppc - { - .name = "save_areas", - .func = save_fake_zone_info, - }, - { - .name = "pmap_mappings", - .func = mapping_fake_zone_info, - }, -#endif /* ppc */ -#if defined(__i386__) || defined (__x86_64__) - { - .name = "page_tables", - .func = pt_fake_zone_info, - }, -#endif /* i386 */ - { - .name = "kalloc.large", - .func = kalloc_fake_zone_info, - }, -}; +kern_return_t +task_zone_info( + task_t task, + mach_zone_name_array_t *namesp, + mach_msg_type_number_t *namesCntp, + task_zone_info_array_t *infop, + mach_msg_type_number_t *infoCntp) +{ + mach_zone_name_t *names; + vm_offset_t names_addr; + vm_size_t names_size; + task_zone_info_t *info; + vm_offset_t info_addr; + vm_size_t info_size; + unsigned int max_zones, i; + zone_t z; + mach_zone_name_t *zn; + task_zone_info_t *zi; + kern_return_t kr; + + vm_size_t used; + vm_map_copy_t copy; + + + if (task == TASK_NULL) + return KERN_INVALID_TASK; + + /* + * We assume that zones aren't freed once allocated. + * We won't pick up any zones that are allocated later. + */ + + simple_lock(&all_zones_lock); + max_zones = (unsigned int)(num_zones + num_fake_zones); + z = first_zone; + simple_unlock(&all_zones_lock); + + names_size = round_page(max_zones * sizeof *names); + kr = kmem_alloc_pageable(ipc_kernel_map, + &names_addr, names_size); + if (kr != KERN_SUCCESS) + return kr; + names = (mach_zone_name_t *) names_addr; + + info_size = round_page(max_zones * sizeof *info); + kr = kmem_alloc_pageable(ipc_kernel_map, + &info_addr, info_size); + if (kr != KERN_SUCCESS) { + kmem_free(ipc_kernel_map, + names_addr, names_size); + return kr; + } + + info = (task_zone_info_t *) info_addr; + + zn = &names[0]; + zi = &info[0]; + + for (i = 0; i < max_zones - num_fake_zones; i++) { + struct zone zcopy; + + assert(z != ZONE_NULL); + + lock_zone(z); + zcopy = *z; + unlock_zone(z); + + simple_lock(&all_zones_lock); + z = z->next_zone; + simple_unlock(&all_zones_lock); + + /* assuming here the name data is static */ + (void) strncpy(zn->mzn_name, zcopy.zone_name, + sizeof zn->mzn_name); + zn->mzn_name[sizeof zn->mzn_name - 1] = '\0'; + + zi->tzi_count = (uint64_t)zcopy.count; + zi->tzi_cur_size = (uint64_t)zcopy.cur_size; + zi->tzi_max_size = (uint64_t)zcopy.max_size; + zi->tzi_elem_size = (uint64_t)zcopy.elem_size; + zi->tzi_alloc_size = (uint64_t)zcopy.alloc_size; + zi->tzi_sum_size = zcopy.sum_count * zcopy.elem_size; + zi->tzi_exhaustible = (uint64_t)zcopy.exhaustible; + zi->tzi_collectable = (uint64_t)zcopy.collectable; + zi->tzi_caller_acct = (uint64_t)zcopy.caller_acct; + if (task->tkm_zinfo != NULL) { + zi->tzi_task_alloc = task->tkm_zinfo[zcopy.index].alloc; + zi->tzi_task_free = task->tkm_zinfo[zcopy.index].free; + } else { + zi->tzi_task_alloc = 0; + zi->tzi_task_free = 0; + } + zn++; + zi++; + } + + /* + * loop through the fake zones and fill them using the specialized + * functions + */ + for (i = 0; i < num_fake_zones; i++) { + int count, collectable, exhaustible, caller_acct, index; + vm_size_t cur_size, max_size, elem_size, alloc_size; + uint64_t sum_size; + + strncpy(zn->mzn_name, fake_zones[i].name, sizeof zn->mzn_name); + zn->mzn_name[sizeof zn->mzn_name - 1] = '\0'; + fake_zones[i].query(&count, &cur_size, + &max_size, &elem_size, + &alloc_size, &sum_size, + &collectable, &exhaustible, &caller_acct); + zi->tzi_count = (uint64_t)count; + zi->tzi_cur_size = (uint64_t)cur_size; + zi->tzi_max_size = (uint64_t)max_size; + zi->tzi_elem_size = (uint64_t)elem_size; + zi->tzi_alloc_size = (uint64_t)alloc_size; + zi->tzi_sum_size = sum_size; + zi->tzi_collectable = (uint64_t)collectable; + zi->tzi_exhaustible = (uint64_t)exhaustible; + zi->tzi_caller_acct = (uint64_t)caller_acct; + if (task->tkm_zinfo != NULL) { + index = ZINFO_SLOTS - num_fake_zones + i; + zi->tzi_task_alloc = task->tkm_zinfo[index].alloc; + zi->tzi_task_free = task->tkm_zinfo[index].free; + } else { + zi->tzi_task_alloc = 0; + zi->tzi_task_free = 0; + } + zn++; + zi++; + } + + used = max_zones * sizeof *names; + if (used != names_size) + bzero((char *) (names_addr + used), names_size - used); + + kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)names_addr, + (vm_map_size_t)names_size, TRUE, ©); + assert(kr == KERN_SUCCESS); + + *namesp = (mach_zone_name_t *) copy; + *namesCntp = max_zones; + + used = max_zones * sizeof *info; + + if (used != info_size) + bzero((char *) (info_addr + used), info_size - used); + + kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)info_addr, + (vm_map_size_t)info_size, TRUE, ©); + assert(kr == KERN_SUCCESS); + + *infop = (task_zone_info_t *) copy; + *infoCntp = max_zones; + + return KERN_SUCCESS; +} + +kern_return_t +mach_zone_info( + host_t host, + mach_zone_name_array_t *namesp, + mach_msg_type_number_t *namesCntp, + mach_zone_info_array_t *infop, + mach_msg_type_number_t *infoCntp) +{ + mach_zone_name_t *names; + vm_offset_t names_addr; + vm_size_t names_size; + mach_zone_info_t *info; + vm_offset_t info_addr; + vm_size_t info_size; + unsigned int max_zones, i; + zone_t z; + mach_zone_name_t *zn; + mach_zone_info_t *zi; + kern_return_t kr; + + vm_size_t used; + vm_map_copy_t copy; + + + if (host == HOST_NULL) + return KERN_INVALID_HOST; + + num_fake_zones = sizeof fake_zones / sizeof fake_zones[0]; + + /* + * We assume that zones aren't freed once allocated. + * We won't pick up any zones that are allocated later. + */ + + simple_lock(&all_zones_lock); + max_zones = (unsigned int)(num_zones + num_fake_zones); + z = first_zone; + simple_unlock(&all_zones_lock); + + names_size = round_page(max_zones * sizeof *names); + kr = kmem_alloc_pageable(ipc_kernel_map, + &names_addr, names_size); + if (kr != KERN_SUCCESS) + return kr; + names = (mach_zone_name_t *) names_addr; + + info_size = round_page(max_zones * sizeof *info); + kr = kmem_alloc_pageable(ipc_kernel_map, + &info_addr, info_size); + if (kr != KERN_SUCCESS) { + kmem_free(ipc_kernel_map, + names_addr, names_size); + return kr; + } + + info = (mach_zone_info_t *) info_addr; + + zn = &names[0]; + zi = &info[0]; + + for (i = 0; i < max_zones - num_fake_zones; i++) { + struct zone zcopy; + + assert(z != ZONE_NULL); + + lock_zone(z); + zcopy = *z; + unlock_zone(z); + + simple_lock(&all_zones_lock); + z = z->next_zone; + simple_unlock(&all_zones_lock); + + /* assuming here the name data is static */ + (void) strncpy(zn->mzn_name, zcopy.zone_name, + sizeof zn->mzn_name); + zn->mzn_name[sizeof zn->mzn_name - 1] = '\0'; + + zi->mzi_count = (uint64_t)zcopy.count; + zi->mzi_cur_size = (uint64_t)zcopy.cur_size; + zi->mzi_max_size = (uint64_t)zcopy.max_size; + zi->mzi_elem_size = (uint64_t)zcopy.elem_size; + zi->mzi_alloc_size = (uint64_t)zcopy.alloc_size; + zi->mzi_sum_size = zcopy.sum_count * zcopy.elem_size; + zi->mzi_exhaustible = (uint64_t)zcopy.exhaustible; + zi->mzi_collectable = (uint64_t)zcopy.collectable; + zn++; + zi++; + } + + /* + * loop through the fake zones and fill them using the specialized + * functions + */ + for (i = 0; i < num_fake_zones; i++) { + int count, collectable, exhaustible, caller_acct; + vm_size_t cur_size, max_size, elem_size, alloc_size; + uint64_t sum_size; + + strncpy(zn->mzn_name, fake_zones[i].name, sizeof zn->mzn_name); + zn->mzn_name[sizeof zn->mzn_name - 1] = '\0'; + fake_zones[i].query(&count, &cur_size, + &max_size, &elem_size, + &alloc_size, &sum_size, + &collectable, &exhaustible, &caller_acct); + zi->mzi_count = (uint64_t)count; + zi->mzi_cur_size = (uint64_t)cur_size; + zi->mzi_max_size = (uint64_t)max_size; + zi->mzi_elem_size = (uint64_t)elem_size; + zi->mzi_alloc_size = (uint64_t)alloc_size; + zi->mzi_sum_size = sum_size; + zi->mzi_collectable = (uint64_t)collectable; + zi->mzi_exhaustible = (uint64_t)exhaustible; + + zn++; + zi++; + } + + used = max_zones * sizeof *names; + if (used != names_size) + bzero((char *) (names_addr + used), names_size - used); + + kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)names_addr, + (vm_map_size_t)names_size, TRUE, ©); + assert(kr == KERN_SUCCESS); + + *namesp = (mach_zone_name_t *) copy; + *namesCntp = max_zones; + + used = max_zones * sizeof *info; + + if (used != info_size) + bzero((char *) (info_addr + used), info_size - used); + + kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)info_addr, + (vm_map_size_t)info_size, TRUE, ©); + assert(kr == KERN_SUCCESS); + + *infop = (mach_zone_info_t *) copy; + *infoCntp = max_zones; + + return KERN_SUCCESS; +} + +/* + * host_zone_info - LEGACY user interface for Mach zone information + * Should use mach_zone_info() instead! + */ kern_return_t host_zone_info( host_t host, @@ -1961,7 +3139,9 @@ host_zone_info( zone_name_t *zn; zone_info_t *zi; kern_return_t kr; - size_t num_fake_zones; + + vm_size_t used; + vm_map_copy_t copy; if (host == HOST_NULL) @@ -1987,40 +3167,28 @@ host_zone_info( z = first_zone; simple_unlock(&all_zones_lock); - if (max_zones <= *namesCntp) { - /* use in-line memory */ - names_size = *namesCntp * sizeof *names; - names = *namesp; - } else { - names_size = round_page(max_zones * sizeof *names); - kr = kmem_alloc_pageable(ipc_kernel_map, - &names_addr, names_size); - if (kr != KERN_SUCCESS) - return kr; - names = (zone_name_t *) names_addr; - } - - if (max_zones <= *infoCntp) { - /* use in-line memory */ - info_size = *infoCntp * sizeof *info; - info = *infop; - } else { - info_size = round_page(max_zones * sizeof *info); - kr = kmem_alloc_pageable(ipc_kernel_map, - &info_addr, info_size); - if (kr != KERN_SUCCESS) { - if (names != *namesp) - kmem_free(ipc_kernel_map, - names_addr, names_size); - return kr; - } - - info = (zone_info_t *) info_addr; + names_size = round_page(max_zones * sizeof *names); + kr = kmem_alloc_pageable(ipc_kernel_map, + &names_addr, names_size); + if (kr != KERN_SUCCESS) + return kr; + names = (zone_name_t *) names_addr; + + info_size = round_page(max_zones * sizeof *info); + kr = kmem_alloc_pageable(ipc_kernel_map, + &info_addr, info_size); + if (kr != KERN_SUCCESS) { + kmem_free(ipc_kernel_map, + names_addr, names_size); + return kr; } + + info = (zone_info_t *) info_addr; + zn = &names[0]; zi = &info[0]; - for (i = 0; i < num_zones; i++) { + for (i = 0; i < max_zones - num_fake_zones; i++) { struct zone zcopy; assert(z != ZONE_NULL); @@ -2055,57 +3223,49 @@ host_zone_info( * functions */ for (i = 0; i < num_fake_zones; i++) { + int caller_acct; + uint64_t sum_space; strncpy(zn->zn_name, fake_zones[i].name, sizeof zn->zn_name); zn->zn_name[sizeof zn->zn_name - 1] = '\0'; - fake_zones[i].func(&zi->zi_count, &zi->zi_cur_size, - &zi->zi_max_size, &zi->zi_elem_size, - &zi->zi_alloc_size, &zi->zi_collectable, - &zi->zi_exhaustible); + fake_zones[i].query(&zi->zi_count, &zi->zi_cur_size, + &zi->zi_max_size, &zi->zi_elem_size, + &zi->zi_alloc_size, &sum_space, + &zi->zi_collectable, &zi->zi_exhaustible, &caller_acct); zn++; zi++; } - if (names != *namesp) { - vm_size_t used; - vm_map_copy_t copy; - - used = max_zones * sizeof *names; - - if (used != names_size) - bzero((char *) (names_addr + used), names_size - used); + used = max_zones * sizeof *names; + if (used != names_size) + bzero((char *) (names_addr + used), names_size - used); - kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)names_addr, - (vm_map_size_t)names_size, TRUE, ©); - assert(kr == KERN_SUCCESS); + kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)names_addr, + (vm_map_size_t)names_size, TRUE, ©); + assert(kr == KERN_SUCCESS); - *namesp = (zone_name_t *) copy; - } + *namesp = (zone_name_t *) copy; *namesCntp = max_zones; - if (info != *infop) { - vm_size_t used; - vm_map_copy_t copy; - - used = max_zones * sizeof *info; + used = max_zones * sizeof *info; + if (used != info_size) + bzero((char *) (info_addr + used), info_size - used); - if (used != info_size) - bzero((char *) (info_addr + used), info_size - used); + kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)info_addr, + (vm_map_size_t)info_size, TRUE, ©); + assert(kr == KERN_SUCCESS); - kr = vm_map_copyin(ipc_kernel_map, (vm_map_address_t)info_addr, - (vm_map_size_t)info_size, TRUE, ©); - assert(kr == KERN_SUCCESS); - - *infop = (zone_info_t *) copy; - } + *infop = (zone_info_t *) copy; *infoCntp = max_zones; return KERN_SUCCESS; } extern unsigned int stack_total; +extern unsigned long long stack_allocs; #if defined(__i386__) || defined (__x86_64__) extern unsigned int inuse_ptepages_count; +extern long long alloc_ptepages_count; #endif void zone_display_zprint() @@ -2177,6 +3337,8 @@ db_print_zone( db_printf("C"); if (zcopy.expandable) db_printf("X"); + if (zcopy.caller_acct) + db_printf("A"); db_printf("\n"); }