include/exec/ram_addr.h

   1 /*
   2  * Declarations for cpu physical memory functions
   3  *
   4  * Copyright 2011 Red Hat, Inc. and/or its affiliates
   5  *
   6  * Authors:
   7  *  Avi Kivity <avi@redhat.com>
   8  *
   9  * This work is licensed under the terms of the GNU GPL, version 2 or
  10  * later.  See the COPYING file in the top-level directory.
  11  *
  12  */
  13
  14 /*
  15  * This header is for use by exec.c and memory.c ONLY.  Do not include it.
  16  * The functions declared here will be removed soon.
  17  */
  18
  19 #ifndef RAM_ADDR_H
  20 #define RAM_ADDR_H
  21
  22 #ifndef CONFIG_USER_ONLY
  23 #include "hw/xen/xen.h"
  24 #include "exec/ramlist.h"
  25
  26 struct RAMBlock {
  27     struct rcu_head rcu;
  28     struct MemoryRegion *mr;
  29     uint8_t *host;
  30     uint8_t *colo_cache; /* For colo, VM's ram cache */
  31     ram_addr_t offset;
  32     ram_addr_t used_length;
  33     ram_addr_t max_length;
  34     void (*resized)(const char*, uint64_t length, void *host);
  35     uint32_t flags;
  36     /* Protected by iothread lock.  */
  37     char idstr[256];
  38     /* RCU-enabled, writes protected by the ramlist lock */
  39     QLIST_ENTRY(RAMBlock) next;
  40     QLIST_HEAD(, RAMBlockNotifier) ramblock_notifiers;
  41     int fd;
  42     size_t page_size;
  43     /* dirty bitmap used during migration */
  44     unsigned long *bmap;
  45     /* bitmap of pages that haven't been sent even once
  46      * only maintained and used in postcopy at the moment
  47      * where it's used to send the dirtymap at the start
  48      * of the postcopy phase
  49      */
  50     unsigned long *unsentmap;
  51     /* bitmap of already received pages in postcopy */
  52     unsigned long *receivedmap;
  53 };
  54
  55 static inline bool offset_in_ramblock(RAMBlock *b, ram_addr_t offset)
  56 {
  57     return (b && b->host && offset < b->used_length) ? true : false;
  58 }
  59
  60 static inline void *ramblock_ptr(RAMBlock *block, ram_addr_t offset)
  61 {
  62     assert(offset_in_ramblock(block, offset));
  63     return (char *)block->host + offset;
  64 }
  65
  66 static inline unsigned long int ramblock_recv_bitmap_offset(void *host_addr,
  67                                                             RAMBlock *rb)
  68 {
  69     uint64_t host_addr_offset =
  70             (uint64_t)(uintptr_t)(host_addr - (void *)rb->host);
  71     return host_addr_offset >> TARGET_PAGE_BITS;
  72 }
  73
  74 bool ramblock_is_pmem(RAMBlock *rb);
  75
  76 long qemu_getrampagesize(void);
  77
  78 /**
  79  * qemu_ram_alloc_from_file,
  80  * qemu_ram_alloc_from_fd:  Allocate a ram block from the specified backing
  81  *                          file or device
  82  *
  83  * Parameters:
  84  *  @size: the size in bytes of the ram block
  85  *  @mr: the memory region where the ram block is
  86  *  @ram_flags: specify the properties of the ram block, which can be one
  87  *              or bit-or of following values
  88  *              - RAM_SHARED: mmap the backing file or device with MAP_SHARED
  89  *              - RAM_PMEM: the backend @mem_path or @fd is persistent memory
  90  *              Other bits are ignored.
  91  *  @mem_path or @fd: specify the backing file or device
  92  *  @errp: pointer to Error*, to store an error if it happens
  93  *
  94  * Return:
  95  *  On success, return a pointer to the ram block.
  96  *  On failure, return NULL.
  97  */
  98 RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
  99                                    uint32_t ram_flags, const char *mem_path,
 100                                    Error **errp);
 101 RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, MemoryRegion *mr,
 102                                  uint32_t ram_flags, int fd,
 103                                  Error **errp);
 104
 105 RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
 106                                   MemoryRegion *mr, Error **errp);
 107 RAMBlock *qemu_ram_alloc(ram_addr_t size, bool share, MemoryRegion *mr,
 108                          Error **errp);
 109 RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t max_size,
 110                                     void (*resized)(const char*,
 111                                                     uint64_t length,
 112                                                     void *host),
 113                                     MemoryRegion *mr, Error **errp);
 114 void qemu_ram_free(RAMBlock *block);
 115
 116 int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp);
 117
 118 #define DIRTY_CLIENTS_ALL     ((1 << DIRTY_MEMORY_NUM) - 1)
 119 #define DIRTY_CLIENTS_NOCODE  (DIRTY_CLIENTS_ALL & ~(1 << DIRTY_MEMORY_CODE))
 120
 121 void tb_invalidate_phys_range(ram_addr_t start, ram_addr_t end);
 122
 123 static inline bool cpu_physical_memory_get_dirty(ram_addr_t start,
 124                                                  ram_addr_t length,
 125                                                  unsigned client)
 126 {
 127     DirtyMemoryBlocks *blocks;
 128     unsigned long end, page;
 129     unsigned long idx, offset, base;
 130     bool dirty = false;
 131
 132     assert(client < DIRTY_MEMORY_NUM);
 133
 134     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
 135     page = start >> TARGET_PAGE_BITS;
 136
 137     rcu_read_lock();
 138
 139     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
 140
 141     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 142     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 143     base = page - offset;
 144     while (page < end) {
 145         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
 146         unsigned long num = next - base;
 147         unsigned long found = find_next_bit(blocks->blocks[idx], num, offset);
 148         if (found < num) {
 149             dirty = true;
 150             break;
 151         }
 152
 153         page = next;
 154         idx++;
 155         offset = 0;
 156         base += DIRTY_MEMORY_BLOCK_SIZE;
 157     }
 158
 159     rcu_read_unlock();
 160
 161     return dirty;
 162 }
 163
 164 static inline bool cpu_physical_memory_all_dirty(ram_addr_t start,
 165                                                  ram_addr_t length,
 166                                                  unsigned client)
 167 {
 168     DirtyMemoryBlocks *blocks;
 169     unsigned long end, page;
 170     unsigned long idx, offset, base;
 171     bool dirty = true;
 172
 173     assert(client < DIRTY_MEMORY_NUM);
 174
 175     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
 176     page = start >> TARGET_PAGE_BITS;
 177
 178     rcu_read_lock();
 179
 180     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
 181
 182     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 183     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 184     base = page - offset;
 185     while (page < end) {
 186         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
 187         unsigned long num = next - base;
 188         unsigned long found = find_next_zero_bit(blocks->blocks[idx], num, offset);
 189         if (found < num) {
 190             dirty = false;
 191             break;
 192         }
 193
 194         page = next;
 195         idx++;
 196         offset = 0;
 197         base += DIRTY_MEMORY_BLOCK_SIZE;
 198     }
 199
 200     rcu_read_unlock();
 201
 202     return dirty;
 203 }
 204
 205 static inline bool cpu_physical_memory_get_dirty_flag(ram_addr_t addr,
 206                                                       unsigned client)
 207 {
 208     return cpu_physical_memory_get_dirty(addr, 1, client);
 209 }
 210
 211 static inline bool cpu_physical_memory_is_clean(ram_addr_t addr)
 212 {
 213     bool vga = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_VGA);
 214     bool code = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_CODE);
 215     bool migration =
 216         cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_MIGRATION);
 217     return !(vga && code && migration);
 218 }
 219
 220 static inline uint8_t cpu_physical_memory_range_includes_clean(ram_addr_t start,
 221                                                                ram_addr_t length,
 222                                                                uint8_t mask)
 223 {
 224     uint8_t ret = 0;
 225
 226     if (mask & (1 << DIRTY_MEMORY_VGA) &&
 227         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_VGA)) {
 228         ret |= (1 << DIRTY_MEMORY_VGA);
 229     }
 230     if (mask & (1 << DIRTY_MEMORY_CODE) &&
 231         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_CODE)) {
 232         ret |= (1 << DIRTY_MEMORY_CODE);
 233     }
 234     if (mask & (1 << DIRTY_MEMORY_MIGRATION) &&
 235         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_MIGRATION)) {
 236         ret |= (1 << DIRTY_MEMORY_MIGRATION);
 237     }
 238     return ret;
 239 }
 240
 241 static inline void cpu_physical_memory_set_dirty_flag(ram_addr_t addr,
 242                                                       unsigned client)
 243 {
 244     unsigned long page, idx, offset;
 245     DirtyMemoryBlocks *blocks;
 246
 247     assert(client < DIRTY_MEMORY_NUM);
 248
 249     page = addr >> TARGET_PAGE_BITS;
 250     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 251     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 252
 253     rcu_read_lock();
 254
 255     blocks = atomic_rcu_read(&ram_list.dirty_memory[client]);
 256
 257     set_bit_atomic(offset, blocks->blocks[idx]);
 258
 259     rcu_read_unlock();
 260 }
 261
 262 static inline void cpu_physical_memory_set_dirty_range(ram_addr_t start,
 263                                                        ram_addr_t length,
 264                                                        uint8_t mask)
 265 {
 266     DirtyMemoryBlocks *blocks[DIRTY_MEMORY_NUM];
 267     unsigned long end, page;
 268     unsigned long idx, offset, base;
 269     int i;
 270
 271     if (!mask && !xen_enabled()) {
 272         return;
 273     }
 274
 275     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
 276     page = start >> TARGET_PAGE_BITS;
 277
 278     rcu_read_lock();
 279
 280     for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
 281         blocks[i] = atomic_rcu_read(&ram_list.dirty_memory[i]);
 282     }
 283
 284     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 285     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 286     base = page - offset;
 287     while (page < end) {
 288         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
 289
 290         if (likely(mask & (1 << DIRTY_MEMORY_MIGRATION))) {
 291             bitmap_set_atomic(blocks[DIRTY_MEMORY_MIGRATION]->blocks[idx],
 292                               offset, next - page);
 293         }
 294         if (unlikely(mask & (1 << DIRTY_MEMORY_VGA))) {
 295             bitmap_set_atomic(blocks[DIRTY_MEMORY_VGA]->blocks[idx],
 296                               offset, next - page);
 297         }
 298         if (unlikely(mask & (1 << DIRTY_MEMORY_CODE))) {
 299             bitmap_set_atomic(blocks[DIRTY_MEMORY_CODE]->blocks[idx],
 300                               offset, next - page);
 301         }
 302
 303         page = next;
 304         idx++;
 305         offset = 0;
 306         base += DIRTY_MEMORY_BLOCK_SIZE;
 307     }
 308
 309     rcu_read_unlock();
 310
 311     xen_hvm_modified_memory(start, length);
 312 }
 313
 314 #if !defined(_WIN32)
 315 static inline void cpu_physical_memory_set_dirty_lebitmap(unsigned long *bitmap,
 316                                                           ram_addr_t start,
 317                                                           ram_addr_t pages)
 318 {
 319     unsigned long i, j;
 320     unsigned long page_number, c;
 321     hwaddr addr;
 322     ram_addr_t ram_addr;
 323     unsigned long len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
 324     unsigned long hpratio = getpagesize() / TARGET_PAGE_SIZE;
 325     unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
 326
 327     /* start address is aligned at the start of a word? */
 328     if ((((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) &&
 329         (hpratio == 1)) {
 330         unsigned long **blocks[DIRTY_MEMORY_NUM];
 331         unsigned long idx;
 332         unsigned long offset;
 333         long k;
 334         long nr = BITS_TO_LONGS(pages);
 335
 336         idx = (start >> TARGET_PAGE_BITS) / DIRTY_MEMORY_BLOCK_SIZE;
 337         offset = BIT_WORD((start >> TARGET_PAGE_BITS) %
 338                           DIRTY_MEMORY_BLOCK_SIZE);
 339
 340         rcu_read_lock();
 341
 342         for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
 343             blocks[i] = atomic_rcu_read(&ram_list.dirty_memory[i])->blocks;
 344         }
 345
 346         for (k = 0; k < nr; k++) {
 347             if (bitmap[k]) {
 348                 unsigned long temp = leul_to_cpu(bitmap[k]);
 349
 350                 atomic_or(&blocks[DIRTY_MEMORY_MIGRATION][idx][offset], temp);
 351                 atomic_or(&blocks[DIRTY_MEMORY_VGA][idx][offset], temp);
 352                 if (tcg_enabled()) {
 353                     atomic_or(&blocks[DIRTY_MEMORY_CODE][idx][offset], temp);
 354                 }
 355             }
 356
 357             if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
 358                 offset = 0;
 359                 idx++;
 360             }
 361         }
 362
 363         rcu_read_unlock();
 364
 365         xen_hvm_modified_memory(start, pages << TARGET_PAGE_BITS);
 366     } else {
 367         uint8_t clients = tcg_enabled() ? DIRTY_CLIENTS_ALL : DIRTY_CLIENTS_NOCODE;
 368         /*
 369          * bitmap-traveling is faster than memory-traveling (for addr...)
 370          * especially when most of the memory is not dirty.
 371          */
 372         for (i = 0; i < len; i++) {
 373             if (bitmap[i] != 0) {
 374                 c = leul_to_cpu(bitmap[i]);
 375                 do {
 376                     j = ctzl(c);
 377                     c &= ~(1ul << j);
 378                     page_number = (i * HOST_LONG_BITS + j) * hpratio;
 379                     addr = page_number * TARGET_PAGE_SIZE;
 380                     ram_addr = start + addr;
 381                     cpu_physical_memory_set_dirty_range(ram_addr,
 382                                        TARGET_PAGE_SIZE * hpratio, clients);
 383                 } while (c != 0);
 384             }
 385         }
 386     }
 387 }
 388 #endif /* not _WIN32 */
 389
 390 bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start,
 391                                               ram_addr_t length,
 392                                               unsigned client);
 393
 394 DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty
 395     (ram_addr_t start, ram_addr_t length, unsigned client);
 396
 397 bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap,
 398                                             ram_addr_t start,
 399                                             ram_addr_t length);
 400
 401 static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start,
 402                                                          ram_addr_t length)
 403 {
 404     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_MIGRATION);
 405     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_VGA);
 406     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_CODE);
 407 }
 408
 409
 410 static inline
 411 uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb,
 412                                                ram_addr_t start,
 413                                                ram_addr_t length,
 414                                                uint64_t *real_dirty_pages)
 415 {
 416     ram_addr_t addr;
 417     unsigned long word = BIT_WORD((start + rb->offset) >> TARGET_PAGE_BITS);
 418     uint64_t num_dirty = 0;
 419     unsigned long *dest = rb->bmap;
 420
 421     /* start address and length is aligned at the start of a word? */
 422     if (((word * BITS_PER_LONG) << TARGET_PAGE_BITS) ==
 423          (start + rb->offset) &&
 424         !(length & ((BITS_PER_LONG << TARGET_PAGE_BITS) - 1))) {
 425         int k;
 426         int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
 427         unsigned long * const *src;
 428         unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE;
 429         unsigned long offset = BIT_WORD((word * BITS_PER_LONG) %
 430                                         DIRTY_MEMORY_BLOCK_SIZE);
 431         unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
 432
 433         rcu_read_lock();
 434
 435         src = atomic_rcu_read(
 436                 &ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION])->blocks;
 437
 438         for (k = page; k < page + nr; k++) {
 439             if (src[idx][offset]) {
 440                 unsigned long bits = atomic_xchg(&src[idx][offset], 0);
 441                 unsigned long new_dirty;
 442                 *real_dirty_pages += ctpopl(bits);
 443                 new_dirty = ~dest[k];
 444                 dest[k] |= bits;
 445                 new_dirty &= bits;
 446                 num_dirty += ctpopl(new_dirty);
 447             }
 448
 449             if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
 450                 offset = 0;
 451                 idx++;
 452             }
 453         }
 454
 455         rcu_read_unlock();
 456     } else {
 457         ram_addr_t offset = rb->offset;
 458
 459         for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
 460             if (cpu_physical_memory_test_and_clear_dirty(
 461                         start + addr + offset,
 462                         TARGET_PAGE_SIZE,
 463                         DIRTY_MEMORY_MIGRATION)) {
 464                 *real_dirty_pages += 1;
 465                 long k = (start + addr) >> TARGET_PAGE_BITS;
 466                 if (!test_and_set_bit(k, dest)) {
 467                     num_dirty++;
 468                 }
 469             }
 470         }
 471     }
 472
 473     return num_dirty;
 474 }
 475 #endif
 476 #endif