gen/stack_logging.c

   1 /*
   2  * Copyright (c) 1999 Apple Computer, Inc. All rights reserved.
   3  *
   4  * @APPLE_LICENSE_HEADER_START@
   5  *
   6  * This file contains Original Code and/or Modifications of Original Code
   7  * as defined in and that are subject to the Apple Public Source License
   8  * Version 2.0 (the 'License'). You may not use this file except in
   9  * compliance with the License. Please obtain a copy of the License at
  10  * http://www.opensource.apple.com/apsl/ and read it before using this
  11  * file.
  12  *
  13  * The Original Code and all software distributed under the License are
  14  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  15  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
  16  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  18  * Please see the License for the specific language governing rights and
  19  * limitations under the License.
  20  *
  21  * @APPLE_LICENSE_HEADER_END@
  22  */
  23
  24 /*      Bertrand from vmutils -> CF -> System */
  25
  26 #import "stack_logging.h"
  27 #import "malloc_printf.h"
  28
  29 #import <libc.h>
  30 #import <pthread.h>
  31 #import <mach/mach.h>
  32 #include <mach/vm_statistics.h>
  33 #import <malloc/malloc.h>
  34 #import <stdlib.h>
  35
  36 extern void spin_lock(int *);
  37 extern void spin_unlock(int *);
  38 extern void thread_stack_pcs(vm_address_t *, unsigned, unsigned *);
  39
  40 static inline void *allocate_pages(unsigned) __attribute__((always_inline));
  41 static inline void *allocate_pages(unsigned bytes) {
  42     void *address;
  43     if (vm_allocate(mach_task_self(), (vm_address_t *)&address, bytes,
  44                     VM_MAKE_TAG(VM_MEMORY_ANALYSIS_TOOL)| TRUE)) {
  45         malloc_printf("*** out of memory while stack logging\n");
  46         abort();
  47     }
  48     return (void *)address;
  49 }
  50
  51 static inline void deallocate_pages(void *, unsigned) __attribute__((always_inline));
  52 static inline void deallocate_pages(void *ptr, unsigned bytes) {
  53     vm_deallocate(mach_task_self(), (vm_address_t)ptr, bytes);
  54 }
  55
  56 static inline void copy_pages(const void *, void *, unsigned) __attribute__((always_inline));
  57 static inline void copy_pages(const void *source, void *dest, unsigned bytes) {
  58     if (vm_copy(mach_task_self(), (vm_address_t)source, bytes, (vm_address_t)dest)) memmove(dest, source, bytes);
  59 }
  60
  61 /***************        Uniquing stack          ***********/
  62
  63 #define MAX_COLLIDE     8
  64
  65 #define MAX_NUM_PC      512
  66
  67 static int enter_pair_in_table(unsigned *table, unsigned numPages, unsigned *uniquedParent, unsigned thisPC) {
  68     // uniquedParent is in-out; return 1 is collisions max not exceeded
  69     unsigned    base = numPages * vm_page_size / (sizeof(int)*2*2);
  70     unsigned    hash = base + (((*uniquedParent) << 4) ^ (thisPC >> 2)) % (base - 1); // modulo odd number for hashing
  71     unsigned    collisions = MAX_COLLIDE;
  72     while (collisions--) {
  73         unsigned        *head = table + hash*2;
  74         if (! head[0] && !head[1]) {
  75             /* end of chain; store this entry! */
  76             /* Note that we need to test for both head[0] and head[1] as (0, -1) is a valid entry */
  77             head[0] = thisPC;
  78             head[1] = *uniquedParent;
  79             *uniquedParent = hash;
  80             return 1;
  81         }
  82         if ((head[0] == thisPC) && (head[1] == *uniquedParent)) {
  83             /* we found the proper entry, the value for the pair is the entry offset */
  84             *uniquedParent = hash;
  85             return 1;
  86         }
  87         hash++;
  88         if (hash == base*2) hash = base;
  89     }
  90     return 0;
  91 }
  92
  93 unsigned stack_logging_get_unique_stack(unsigned **table, unsigned *table_num_pages, unsigned *stack_entries, unsigned count, unsigned num_hot_to_skip) {
  94     unsigned    uniquedParent = (unsigned)-1;
  95     // we skip the warmest entries that are an artefact of the code
  96     while (num_hot_to_skip--) {
  97         if (count > 0) { stack_entries++; count--; }
  98     }
  99     while (count--) {
 100         unsigned        thisPC = stack_entries[count];
 101         while (!enter_pair_in_table(*table, *table_num_pages, &uniquedParent, thisPC)) {
 102             unsigned    *newTable;
 103             unsigned    oldBytes = (*table_num_pages) * vm_page_size;
 104             newTable = allocate_pages(oldBytes*2);
 105             copy_pages(*table, newTable, oldBytes);
 106             deallocate_pages(*table, oldBytes);
 107             *table_num_pages *= 2;
 108             *table = newTable;
 109         }
 110     }
 111     return uniquedParent;
 112 }
 113
 114 /***************        Logging stack and arguments             ***********/
 115
 116 stack_logging_record_list_t *stack_logging_the_record_list = NULL;
 117
 118 int stack_logging_enable_logging = 0;
 119
 120 int stack_logging_dontcompact = 0;
 121
 122 static int stack_logging_spin_lock = 0;
 123
 124 static stack_logging_record_list_t *GrowLogRecords(stack_logging_record_list_t *records, unsigned desiredNumRecords) {
 125     stack_logging_record_list_t *new_records;
 126     unsigned    old_size = records->overall_num_bytes;
 127     if (desiredNumRecords*sizeof(stack_logging_record_t)+sizeof(stack_logging_record_list_t) < records->overall_num_bytes) return records;
 128     records->overall_num_bytes += records->overall_num_bytes + vm_page_size; // in order to always get an even number of pages
 129     new_records = allocate_pages(records->overall_num_bytes);
 130     copy_pages(records, new_records, old_size);
 131     deallocate_pages(records, old_size);
 132     return new_records;
 133 }
 134
 135 static void prepare_to_log_stack(void) {
 136     if (!stack_logging_the_record_list) {
 137         unsigned        totalSize = 4 * vm_page_size;
 138         stack_logging_the_record_list = allocate_pages(totalSize);
 139         memset(stack_logging_the_record_list, 0, sizeof(stack_logging_record_list_t));
 140         stack_logging_the_record_list->overall_num_bytes = totalSize;
 141         stack_logging_the_record_list->uniquing_table_num_pages = 128;
 142         stack_logging_the_record_list->uniquing_table = allocate_pages(stack_logging_the_record_list->uniquing_table_num_pages * vm_page_size);
 143     }
 144 }
 145
 146 void stack_logging_log_stack(unsigned type, unsigned arg1, unsigned arg2, unsigned arg3, unsigned result, unsigned num_hot_to_skip) {
 147     stack_logging_record_t      *rec;
 148     if (!stack_logging_enable_logging) return;
 149     // printf("stack_logging_log_stack 0x%x 0x%x 0x%x 0x%x -> 0x%x\n", type, arg1, arg2, arg3, result);
 150     if (type & stack_logging_flag_zone) {
 151         // just process it now and be done with it!
 152         arg1 = arg2; arg2 = arg3; arg3 = 0; type &= ~stack_logging_flag_zone;
 153     }
 154     if (type & stack_logging_flag_calloc) {
 155         // just process it now and be done with it!
 156         arg1 *= arg2; arg2 = arg3; arg3 = 0; type &= ~stack_logging_flag_calloc;
 157     }
 158     if (type & stack_logging_flag_object) {
 159         unsigned        *class = (unsigned *)arg1;
 160         arg1 = arg2 + class[5]; // corresponds to the instance_size field
 161         arg2 = 0; arg3 = 0; type = stack_logging_type_alloc;
 162     }
 163     if (type & stack_logging_flag_cleared) {
 164         type &= ~stack_logging_flag_cleared;
 165     }
 166     if (type & stack_logging_flag_handle) {
 167         if (stack_logging_type_alloc) {
 168             if (!result) return;
 169             stack_logging_log_stack(stack_logging_type_alloc, 0, 0, 0, result, num_hot_to_skip+1);
 170             stack_logging_log_stack(stack_logging_type_alloc, arg1, 0, 0, *((int *)result), num_hot_to_skip+1);
 171             return;
 172         }
 173         if (stack_logging_type_dealloc) {
 174             if (!arg1) return;
 175             stack_logging_log_stack(stack_logging_type_dealloc, *((int *)arg1), 0, 0, 0, num_hot_to_skip+1);
 176             stack_logging_log_stack(stack_logging_type_dealloc, arg1, 0, 0, 0, num_hot_to_skip+1);
 177             return;
 178         }
 179         fprintf(stderr, "*** Unknown logging type: 0x%x\n", type);
 180     }
 181     if (type == stack_logging_flag_set_handle_size) {
 182         if (!arg1) return;
 183         // Thanks to a horrible hack, arg3 contains the prvious handle value
 184         if (arg3 == *((int *)arg1)) return;
 185         stack_logging_log_stack(stack_logging_type_dealloc, arg3, 0, 0, 0, num_hot_to_skip+1);
 186         stack_logging_log_stack(stack_logging_type_alloc, arg2, 0, 0, *((int *)arg1), num_hot_to_skip+1);
 187         return;
 188     }
 189     if (type == (stack_logging_type_dealloc|stack_logging_type_alloc)) {
 190         if (arg1 == result) return; // realloc had no effect, skipping
 191         if (!arg1) {
 192             // realloc(NULL, size) same as malloc(size)
 193             type = stack_logging_type_alloc; arg1 = arg2; arg2 = arg3; arg3 = 0;
 194         } else {
 195             // realloc(arg1, arg2) -> result is same as free(arg1); malloc(arg2) -> result
 196             stack_logging_log_stack(stack_logging_type_dealloc, arg1, 0, 0, 0, num_hot_to_skip+1);
 197             stack_logging_log_stack(stack_logging_type_alloc, arg2, 0, 0, result, num_hot_to_skip+1);
 198             return;
 199         }
 200     }
 201     if (type == stack_logging_type_dealloc) {
 202         // simple free
 203         if (!arg1) return; // free(nil)
 204     }
 205     prepare_to_log_stack();
 206     spin_lock(&stack_logging_spin_lock);
 207     stack_logging_the_record_list = GrowLogRecords(stack_logging_the_record_list, stack_logging_the_record_list->num_records + 1);
 208     rec = stack_logging_the_record_list->records + stack_logging_the_record_list->num_records;
 209     // We take care of the common case of alloc-dealloc
 210     if (!stack_logging_dontcompact && stack_logging_the_record_list->num_records && (type == stack_logging_type_dealloc) && arg1 && ((rec-1)->type == stack_logging_type_alloc) && (arg1 == STACK_LOGGING_DISGUISE((rec-1)->address))) {
 211         stack_logging_the_record_list->num_records--;
 212         // printf("Erased previous record in alloc-dealloc sequence\n");
 213     } else {
 214         unsigned        stack_entries[MAX_NUM_PC];
 215         unsigned        count = 0;
 216         rec->type = type;
 217         if (type == stack_logging_type_dealloc) {
 218             rec->argument = 0;
 219             rec->address = STACK_LOGGING_DISGUISE(arg1); // we disguise the address
 220         } else if (type == stack_logging_type_alloc) {
 221             rec->argument = arg1;
 222             rec->address = STACK_LOGGING_DISGUISE(result); // we disguise the address
 223         } else {
 224             rec->argument = arg2;
 225             rec->address = STACK_LOGGING_DISGUISE(arg1); // we disguise the address
 226         }
 227         // printf("Before getting samples  0x%x 0x%x 0x%x 0x%x -> 0x%x\n", type, arg1, arg2, arg3, result);
 228         thread_stack_pcs(stack_entries, MAX_NUM_PC - 1, &count);
 229         // We put at the bottom of the stack a marker that denotes the thread (+1 for good measure...)
 230         stack_entries[count++] = (int)pthread_self() + 1;
 231         /* now let's unique the sample */
 232         // printf("Uniquing 0x%x 0x%x 0x%x 0x%x -> 0x%x\n", type, arg1, arg2, arg3, result);
 233         rec->uniqued_stack = stack_logging_get_unique_stack(&stack_logging_the_record_list->uniquing_table, &stack_logging_the_record_list->uniquing_table_num_pages, stack_entries, count, num_hot_to_skip+2); // we additionally skip the warmest 2 entries that are an artefact of the code
 234         stack_logging_the_record_list->num_records++;
 235     }
 236     spin_unlock(&stack_logging_spin_lock);
 237 }
 238
 239 static kern_return_t default_reader(task_t task, vm_address_t address, vm_size_t size, void **ptr) {
 240     *ptr = (void *)address;
 241     return 0;
 242 }
 243
 244 static kern_return_t get_remote_records(task_t task, memory_reader_t reader, stack_logging_record_list_t **records) {
 245     // sets records
 246     vm_address_t        *remote_records_address_ref;
 247     kern_return_t       err;
 248     *records = NULL;
 249     err = reader(task, (vm_address_t)&stack_logging_the_record_list, sizeof(vm_address_t), (void **)&remote_records_address_ref);
 250     if (err) return err;
 251     if (!*remote_records_address_ref) {
 252         // printf("stack_logging: no stack record\n");
 253         return 0;
 254     }
 255     // printf("stack_logging: stack records at %p\n", (void *)(*remote_records_address_ref));
 256     // printf("stack_logging: reading %d bytes\n", sizeof(stack_logging_record_list_t));
 257     err = reader(task, *remote_records_address_ref, sizeof(stack_logging_record_list_t), (void **)records); // get the list head
 258     if (err) return err;
 259     // printf("stack_logging: overall num bytes = %d\n", records->overall_num_bytes);
 260     return reader(task, *remote_records_address_ref, (*records)->overall_num_bytes, (void **)records);
 261 }
 262
 263 kern_return_t stack_logging_get_frames(task_t task, memory_reader_t reader, vm_address_t address, vm_address_t *stack_frames_buffer, unsigned max_stack_frames, unsigned *num_frames) {
 264     stack_logging_record_list_t *records;
 265     kern_return_t       err;
 266     unsigned            index;
 267     unsigned            disguised = STACK_LOGGING_DISGUISE(address);
 268     if (!reader) reader = default_reader;
 269     *num_frames = 0;
 270     err = get_remote_records(task, reader, &records);
 271     if (err || !records) return err;
 272     // printf("stack_logging: %d records\n", records->num_records);
 273     index = 0;
 274     while (index < records->num_records) {
 275         stack_logging_record_t  *record = records->records + index;
 276         if (record->address == disguised) {
 277             return stack_logging_frames_for_uniqued_stack(task, reader, record->uniqued_stack, stack_frames_buffer, max_stack_frames, num_frames);
 278         }
 279         index++;
 280     }
 281     fprintf(stderr, "*** stack_logging: no record found for 0x%x\n", address);
 282     return 0;
 283 }
 284
 285 kern_return_t stack_logging_enumerate_records(task_t task, memory_reader_t reader, vm_address_t address, void enumerator(stack_logging_record_t, void *), void *context) {
 286     stack_logging_record_list_t *records;
 287     kern_return_t       err;
 288     unsigned            index;
 289     unsigned            disguised = STACK_LOGGING_DISGUISE(address);
 290     if (!reader) reader = default_reader;
 291     err = get_remote_records(task, reader, &records);
 292     if (err || !records) return err;
 293     // printf("stack_logging: %d records\n", records->num_records);
 294     index = 0;
 295     while (index < records->num_records) {
 296         stack_logging_record_t  *record = records->records + index;
 297         if (!address || (record->address == disguised)) enumerator(*record, context);
 298         index++;
 299     }
 300     return 0;
 301 }
 302
 303 kern_return_t stack_logging_frames_for_uniqued_stack(task_t task, memory_reader_t reader, unsigned uniqued_stack, vm_address_t *stack_frames_buffer, unsigned max_stack_frames, unsigned *num_frames) {
 304     stack_logging_record_list_t *records;
 305     unsigned            *uniquing_table;
 306     kern_return_t       err;
 307     if (!reader) reader = default_reader;
 308     *num_frames = 0;
 309     err = get_remote_records(task, reader, &records);
 310     if (err || !records) return err;
 311     err = reader(task, (vm_address_t)records->uniquing_table, records->uniquing_table_num_pages * vm_page_size, (void **)&uniquing_table);
 312     if (err) return err;
 313     while (max_stack_frames && (uniqued_stack != -1)) {
 314         unsigned        thisPC;
 315         if ((uniqued_stack * 2 + 1) * sizeof(unsigned) >= records->uniquing_table_num_pages * vm_page_size) {
 316             fprintf(stderr, "*** stack_logging: Invalid uniqued stack 0x%x", uniqued_stack);
 317             break;
 318         }
 319         thisPC = uniquing_table[uniqued_stack * 2];
 320         uniqued_stack = uniquing_table[uniqued_stack * 2 + 1];
 321         if (!thisPC && !uniqued_stack) {
 322             // Invalid entry
 323             fprintf(stderr, "*** stack_logging: Invalid entry 0x%x", thisPC);
 324             break;
 325         }
 326         stack_frames_buffer[0] = thisPC;
 327         stack_frames_buffer++;
 328         (*num_frames)++;
 329         max_stack_frames--;
 330     }
 331     return 0;
 332 }