payloads/libpayload/libc/malloc.c

   1 /*
   2  *
   3  * Copyright (C) 2008 Advanced Micro Devices, Inc.
   4  * Copyright (C) 2008-2010 coresystems GmbH
   5  *
   6  * Redistribution and use in source and binary forms, with or without
   7  * modification, are permitted provided that the following conditions
   8  * are met:
   9  * 1. Redistributions of source code must retain the above copyright
  10  *    notice, this list of conditions and the following disclaimer.
  11  * 2. Redistributions in binary form must reproduce the above copyright
  12  *    notice, this list of conditions and the following disclaimer in the
  13  *    documentation and/or other materials provided with the distribution.
  14  * 3. The name of the author may not be used to endorse or promote products
  15  *    derived from this software without specific prior written permission.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 /*
  31  * This is a classically weak malloc() implementation. We have a relatively
  32  * small and static heap, so we take the easy route with an O(N) loop
  33  * through the tree for every malloc() and free(). Obviously, this doesn't
  34  * scale past a few hundred KB (if that).
  35  *
  36  * We're also susceptible to the usual buffer overrun poisoning, though the
  37  * risk is within acceptable ranges for this implementation (don't overrun
  38  * your buffers, kids!).
  39  */
  40
  41 #define IN_MALLOC_C
  42 #include <libpayload.h>
  43 #include <stdint.h>
  44
  45 struct memory_type {
  46         void *start;
  47         void *end;
  48         struct align_region_t* align_regions;
  49 #if CONFIG(LP_DEBUG_MALLOC)
  50         int magic_initialized;
  51         size_t minimal_free;
  52         const char *name;
  53 #endif
  54 };
  55
  56 extern char _heap, _eheap;      /* Defined in the ldscript. */
  57
  58 static struct memory_type default_type =
  59         { (void *)&_heap, (void *)&_eheap, NULL
  60 #if CONFIG(LP_DEBUG_MALLOC)
  61         , 0, 0, "HEAP"
  62 #endif
  63         };
  64 static struct memory_type *const heap = &default_type;
  65 static struct memory_type *dma = &default_type;
  66
  67 typedef u64 hdrtype_t;
  68 #define HDRSIZE (sizeof(hdrtype_t))
  69
  70 #define SIZE_BITS ((HDRSIZE << 3) - 7)
  71 #define MAGIC     (((hdrtype_t)0x2a) << (SIZE_BITS + 1))
  72 #define FLAG_FREE (((hdrtype_t)0x01) << (SIZE_BITS + 0))
  73 #define MAX_SIZE  ((((hdrtype_t)0x01) << SIZE_BITS) - 1)
  74
  75 #define SIZE(_h) ((_h) & MAX_SIZE)
  76
  77 #define _HEADER(_s, _f) ((hdrtype_t) (MAGIC | (_f) | ((_s) & MAX_SIZE)))
  78
  79 #define FREE_BLOCK(_s) _HEADER(_s, FLAG_FREE)
  80 #define USED_BLOCK(_s) _HEADER(_s, 0)
  81
  82 #define IS_FREE(_h) (((_h) & (MAGIC | FLAG_FREE)) == (MAGIC | FLAG_FREE))
  83 #define HAS_MAGIC(_h) (((_h) & MAGIC) == MAGIC)
  84
  85 static int free_aligned(void* addr, struct memory_type *type);
  86 void print_malloc_map(void);
  87
  88 void init_dma_memory(void *start, u32 size)
  89 {
  90         if (dma_initialized()) {
  91                 printf("ERROR: %s called twice!\n", __func__);
  92                 return;
  93         }
  94
  95         /*
  96          * DMA memory might not be zeroed by coreboot on stage loading, so make
  97          * sure we clear the magic cookie from last boot.
  98          */
  99         *(hdrtype_t *)start = 0;
 100
 101         dma = malloc(sizeof(*dma));
 102         dma->start = start;
 103         dma->end = start + size;
 104         dma->align_regions = NULL;
 105
 106 #if CONFIG(LP_DEBUG_MALLOC)
 107         dma->minimal_free = 0;
 108         dma->magic_initialized = 0;
 109         dma->name = "DMA";
 110
 111         printf("Initialized cache-coherent DMA memory at [%p:%p]\n", start, start + size);
 112 #endif
 113 }
 114
 115 int dma_initialized()
 116 {
 117         return dma != heap;
 118 }
 119
 120 /* For boards that don't initialize DMA we assume all locations are coherent */
 121 int dma_coherent(const void *ptr)
 122 {
 123         return !dma_initialized() || (dma->start <= ptr && dma->end > ptr);
 124 }
 125
 126 /* Find free block of size >= len */
 127 static hdrtype_t volatile *find_free_block(int len, struct memory_type *type)
 128 {
 129         hdrtype_t header;
 130         hdrtype_t volatile *ptr = (hdrtype_t volatile *)type->start;
 131
 132         /* Align the size. */
 133         len = ALIGN_UP(len, HDRSIZE);
 134
 135         if (!len || len > MAX_SIZE)
 136                 return (void *)NULL;
 137
 138         /* Make sure the region is setup correctly. */
 139         if (!HAS_MAGIC(*ptr)) {
 140                 size_t size = (type->end - type->start) - HDRSIZE;
 141                 *ptr = FREE_BLOCK(size);
 142 #if CONFIG(LP_DEBUG_MALLOC)
 143                 type->magic_initialized = 1;
 144                 type->minimal_free = size;
 145 #endif
 146         }
 147
 148         /* Find some free space. */
 149         do {
 150                 header = *ptr;
 151                 int size = SIZE(header);
 152
 153                 if (!HAS_MAGIC(header) || size == 0) {
 154                         printf("memory allocator panic. (%s%s)\n",
 155                                !HAS_MAGIC(header) ? " no magic " : "",
 156                                    size == 0 ? " size=0 " : "");
 157                         halt();
 158                 }
 159
 160                 if ((header & FLAG_FREE) && len <= size)
 161                         return ptr;
 162
 163                 ptr = (hdrtype_t volatile *)((uintptr_t)ptr + HDRSIZE + size);
 164
 165         } while (ptr < (hdrtype_t *) type->end);
 166
 167         /* Nothing available. */
 168         return NULL;
 169 }
 170
 171 /* Mark the block with length 'len' as used */
 172 static void use_block(hdrtype_t volatile *ptr, int len)
 173 {
 174         /* Align the size. */
 175         len = ALIGN_UP(len, HDRSIZE);
 176
 177         hdrtype_t volatile *nptr = (hdrtype_t volatile *)
 178                 ((uintptr_t)ptr + HDRSIZE + len);
 179         int size = SIZE(*ptr);
 180         int nsize = size - (HDRSIZE + len);
 181
 182         /*
 183          * If there is still room in this block, then mark it as such otherwise
 184          * account the whole space for that block.
 185          */
 186         if (nsize > 0) {
 187                 /* Mark the block as used. */
 188                 *ptr = USED_BLOCK(len);
 189
 190                 /* Create a new free block. */
 191                 *nptr = FREE_BLOCK(nsize);
 192         } else {
 193                 /* Mark the block as used. */
 194                 *ptr = USED_BLOCK(size);
 195         }
 196 }
 197
 198 static void *alloc(int len, struct memory_type *type)
 199 {
 200         hdrtype_t volatile *ptr = find_free_block(len, type);
 201
 202         if (ptr == NULL)
 203                 return NULL;
 204
 205         use_block(ptr, len);
 206         return (void *)((uintptr_t)ptr + HDRSIZE);
 207 }
 208
 209 static void _consolidate(struct memory_type *type)
 210 {
 211         void *ptr = type->start;
 212
 213         while (ptr < type->end) {
 214                 void *nptr;
 215                 hdrtype_t hdr = *((hdrtype_t *) ptr);
 216                 unsigned int size = 0;
 217
 218                 if (!IS_FREE(hdr)) {
 219                         ptr += HDRSIZE + SIZE(hdr);
 220                         continue;
 221                 }
 222
 223                 size = SIZE(hdr);
 224                 nptr = ptr + HDRSIZE + SIZE(hdr);
 225
 226                 while (nptr < type->end) {
 227                         hdrtype_t nhdr = *((hdrtype_t *) nptr);
 228
 229                         if (!(IS_FREE(nhdr)))
 230                                 break;
 231
 232                         size += SIZE(nhdr) + HDRSIZE;
 233
 234                         *((hdrtype_t *) nptr) = 0;
 235
 236                         nptr += (HDRSIZE + SIZE(nhdr));
 237                 }
 238
 239                 *((hdrtype_t *) ptr) = FREE_BLOCK(size);
 240                 ptr = nptr;
 241         }
 242 }
 243
 244 void free(void *ptr)
 245 {
 246         hdrtype_t hdr;
 247         struct memory_type *type = heap;
 248
 249         /* No action occurs on NULL. */
 250         if (ptr == NULL)
 251                 return;
 252
 253         /* Sanity check. */
 254         if (ptr < type->start || ptr >= type->end) {
 255                 type = dma;
 256                 if (ptr < type->start || ptr >= type->end)
 257                         return;
 258         }
 259
 260         if (free_aligned(ptr, type)) return;
 261
 262         ptr -= HDRSIZE;
 263         hdr = *((hdrtype_t *) ptr);
 264
 265         /* Not our header (we're probably poisoned). */
 266         if (!HAS_MAGIC(hdr))
 267                 return;
 268
 269         /* Double free. */
 270         if (hdr & FLAG_FREE)
 271                 return;
 272
 273         *((hdrtype_t *) ptr) = FREE_BLOCK(SIZE(hdr));
 274         _consolidate(type);
 275 }
 276
 277 void *malloc(size_t size)
 278 {
 279         return alloc(size, heap);
 280 }
 281
 282 void *dma_malloc(size_t size)
 283 {
 284         return alloc(size, dma);
 285 }
 286
 287 void *calloc(size_t nmemb, size_t size)
 288 {
 289         size_t total = nmemb * size;
 290         void *ptr = alloc(total, heap);
 291
 292         if (ptr)
 293                 memset(ptr, 0, total);
 294
 295         return ptr;
 296 }
 297
 298 void *realloc(void *ptr, size_t size)
 299 {
 300         void *ret, *pptr;
 301         hdrtype_t volatile *block;
 302         unsigned int osize;
 303         struct memory_type *type = heap;
 304
 305         if (ptr == NULL)
 306                 return alloc(size, type);
 307
 308         pptr = ptr - HDRSIZE;
 309
 310         if (!HAS_MAGIC(*((hdrtype_t *) pptr)))
 311                 return NULL;
 312
 313         if (ptr < type->start || ptr >= type->end)
 314                 type = dma;
 315
 316         /* Get the original size of the block. */
 317         osize = SIZE(*((hdrtype_t *) pptr));
 318
 319         /*
 320          * Free the memory to update the tables - this won't touch the actual
 321          * memory, so we can still use it for the copy after we have
 322          * reallocated the new space.
 323          */
 324         free(ptr);
 325
 326         block = find_free_block(size, type);
 327         if (block == NULL)
 328                 return NULL;
 329
 330         ret = (void *)((uintptr_t)block + HDRSIZE);
 331
 332         /*
 333          * If ret == ptr, then no copy is needed. Otherwise, move the memory to
 334          * the new location, which might be before the old one and overlap since
 335          * the free() above includes a _consolidate().
 336          */
 337         if (ret != ptr)
 338                 memmove(ret, ptr, osize > size ? size : osize);
 339
 340         /* Mark the block as used. */
 341         use_block(block, size);
 342
 343         return ret;
 344 }
 345
 346 struct align_region_t
 347 {
 348         /* If alignment is 0 then the region represents a large region which
 349          * has no metadata for tracking subelements. */
 350         int alignment;
 351         /* start in memory, and size in bytes */
 352         void* start;
 353         int size;
 354         /* layout within a region:
 355           - num_elements bytes, 0: free, 1: used, 2: used, combines with next
 356           - padding to alignment
 357           - data section
 358           - waste space
 359
 360           start_data points to the start of the data section
 361         */
 362         void* start_data;
 363         /* number of free blocks sized "alignment" */
 364         int free;
 365         struct align_region_t *next;
 366 };
 367
 368 static inline int region_is_large(const struct align_region_t *r)
 369 {
 370         return r->alignment == 0;
 371 }
 372
 373 static inline int addr_in_region(const struct align_region_t *r, void *addr)
 374 {
 375         return ((addr >= r->start_data) && (addr < r->start_data + r->size));
 376 }
 377
 378 /* num_elements == 0 indicates a large aligned region instead of a smaller
 379  * region comprised of alignment-sized chunks. */
 380 static struct align_region_t *allocate_region(int alignment, int num_elements,
 381                                         size_t size, struct memory_type *type)
 382 {
 383         struct align_region_t *r;
 384         size_t extra_space;
 385
 386 #if CONFIG(LP_DEBUG_MALLOC)
 387         printf("%s(old align_regions=%p, alignment=%u, num_elements=%u, size=%zu)\n",
 388                 __func__, type->align_regions, alignment, num_elements, size);
 389 #endif
 390
 391         r = malloc(sizeof(*r));
 392
 393         if (r == NULL)
 394                 return NULL;
 395
 396         memset(r, 0, sizeof(*r));
 397
 398         if (num_elements != 0) {
 399                 r->alignment = alignment;
 400                 r->size = num_elements * alignment;
 401                 r->free = num_elements;
 402                 /* Allocate enough memory for alignment requirements and
 403                  * metadata for each chunk. */
 404                 extra_space = num_elements;
 405         } else {
 406                 /* Large aligned allocation. Set alignment = 0. */
 407                 r->alignment = 0;
 408                 r->size = size;
 409                 extra_space = 0;
 410         }
 411
 412         r->start = alloc(r->size + alignment + extra_space, type);
 413
 414         if (r->start == NULL) {
 415                 free(r);
 416                 return NULL;
 417         }
 418
 419         r->start_data = (void *)ALIGN_UP((uintptr_t)r->start + extra_space,
 420                                         alignment);
 421
 422         /* Clear any (if requested) metadata. */
 423         memset(r->start, 0, extra_space);
 424
 425         /* Link the region with the rest. */
 426         r->next = type->align_regions;
 427         type->align_regions = r;
 428
 429         return r;
 430 }
 431
 432 static void try_free_region(struct align_region_t **prev_link)
 433 {
 434         struct align_region_t *r = *prev_link;
 435
 436         /* All large regions are immediately free-able. Non-large regions
 437          * need to be checked for the fully freed state. */
 438         if (!region_is_large(r)) {
 439                 if (r->free != r->size / r->alignment)
 440                         return;
 441         }
 442
 443         /* Unlink region from link list. */
 444         *prev_link = r->next;
 445
 446         /* Free the data and metadata. */
 447         free(r->start);
 448         free(r);
 449 }
 450
 451 static int free_aligned(void* addr, struct memory_type *type)
 452 {
 453         struct align_region_t **prev_link = &type->align_regions;
 454
 455         while (*prev_link != NULL)
 456         {
 457                 if (!addr_in_region(*prev_link, addr)) {
 458                         prev_link = &((*prev_link)->next);
 459                         continue;
 460                 }
 461
 462                 if (region_is_large(*prev_link)) {
 463                         try_free_region(prev_link);
 464                         return 1;
 465                 }
 466
 467                 int i = (addr-(*prev_link)->start_data)/(*prev_link)->alignment;
 468                 u8 *meta = (*prev_link)->start;
 469                 while (meta[i] == 2)
 470                 {
 471                         meta[i++] = 0;
 472                         (*prev_link)->free++;
 473                 }
 474                 meta[i] = 0;
 475                 (*prev_link)->free++;
 476                 try_free_region(prev_link);
 477                 return 1;
 478         }
 479         return 0;
 480 }
 481
 482 static void *alloc_aligned(size_t align, size_t size, struct memory_type *type)
 483 {
 484         /* Define a large request to be 1024 bytes for either alignment or
 485          * size of allocation. */
 486         const size_t large_request = 1024;
 487
 488         if (size == 0) return 0;
 489         if (type->align_regions == 0) {
 490                 type->align_regions = malloc(sizeof(struct align_region_t));
 491                 if (type->align_regions == NULL)
 492                         return NULL;
 493                 memset(type->align_regions, 0, sizeof(struct align_region_t));
 494         }
 495         struct align_region_t *reg = type->align_regions;
 496
 497         if (size >= large_request || align >= large_request) {
 498                 reg = allocate_region(align, 0, size, type);
 499                 if (reg == NULL)
 500                         return NULL;
 501                 return reg->start_data;
 502         }
 503
 504 look_further:
 505         while (reg != 0)
 506         {
 507                 if ((reg->alignment == align) && (reg->free >= (size + align - 1)/align))
 508                 {
 509 #if CONFIG(LP_DEBUG_MALLOC)
 510                         printf("  found memalign region. %u free, %zu required\n", reg->free, (size + align - 1)/align);
 511 #endif
 512                         break;
 513                 }
 514                 reg = reg->next;
 515         }
 516         if (reg == 0)
 517         {
 518 #if CONFIG(LP_DEBUG_MALLOC)
 519                 printf("  need to allocate a new memalign region\n");
 520 #endif
 521                 /* get align regions */
 522                 reg = allocate_region(align, large_request/align, size, type);
 523 #if CONFIG(LP_DEBUG_MALLOC)
 524                 printf("  ... returned %p\n", reg);
 525 #endif
 526         }
 527         if (reg == 0) {
 528                 /* Nothing available. */
 529                 return (void *)NULL;
 530         }
 531
 532         int i, count = 0, target = (size+align-1)/align;
 533         for (i = 0; i < (reg->size/align); i++)
 534         {
 535                 if (((u8*)reg->start)[i] == 0)
 536                 {
 537                         count++;
 538                         if (count == target) {
 539                                 count = i+1-count;
 540                                 for (i=0; i<target-1; i++)
 541                                 {
 542                                         ((u8*)reg->start)[count+i]=2;
 543                                 }
 544                                 ((u8*)reg->start)[count+target-1]=1;
 545                                 reg->free -= target;
 546                                 return reg->start_data+(align*count);
 547                         }
 548                 } else {
 549                         count = 0;
 550                 }
 551         }
 552         /* The free space in this region is fragmented,
 553            so we will move on and try the next one: */
 554         reg = reg->next;
 555         goto look_further; // end condition is once a new region is allocated - it always has enough space
 556 }
 557
 558 void *memalign(size_t align, size_t size)
 559 {
 560         return alloc_aligned(align, size, heap);
 561 }
 562
 563 void *dma_memalign(size_t align, size_t size)
 564 {
 565         return alloc_aligned(align, size, dma);
 566 }
 567
 568 /* This is for debugging purposes. */
 569 #if CONFIG(LP_DEBUG_MALLOC)
 570 void print_malloc_map(void)
 571 {
 572         struct memory_type *type = heap;
 573         void *ptr;
 574         int free_memory;
 575
 576 again:
 577         ptr = type->start;
 578         free_memory = 0;
 579
 580         while (ptr < type->end) {
 581                 hdrtype_t hdr = *((hdrtype_t *) ptr);
 582
 583                 if (!HAS_MAGIC(hdr)) {
 584                         if (type->magic_initialized)
 585                                 printf("%s: Poisoned magic - we're toast\n", type->name);
 586                         else
 587                                 printf("%s: No magic yet - going to initialize\n", type->name);
 588                         break;
 589                 }
 590
 591                 /* FIXME: Verify the size of the block. */
 592
 593                 printf("%s %x: %s (%llx bytes)\n", type->name,
 594                        (unsigned int)(ptr - type->start),
 595                        hdr & FLAG_FREE ? "FREE" : "USED", SIZE(hdr));
 596
 597                 if (hdr & FLAG_FREE)
 598                         free_memory += SIZE(hdr);
 599
 600                 ptr += HDRSIZE + SIZE(hdr);
 601         }
 602
 603         if (free_memory && (type->minimal_free > free_memory))
 604                 type->minimal_free = free_memory;
 605         printf("%s: Maximum memory consumption: %zu bytes\n", type->name,
 606                 (type->end - type->start) - HDRSIZE - type->minimal_free);
 607
 608         if (type != dma) {
 609                 type = dma;
 610                 goto again;
 611         }
 612 }
 613 #endif