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 <[email protected]>
   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 "cpu.h"
  24 #include "sysemu/xen.h"
  25 #include "sysemu/tcg.h"
  26 #include "exec/ramlist.h"
  27 #include "exec/ramblock.h"
  28
  29 /**
  30  * clear_bmap_size: calculate clear bitmap size
  31  *
  32  * @pages: number of guest pages
  33  * @shift: guest page number shift
  34  *
  35  * Returns: number of bits for the clear bitmap
  36  */
  37 static inline long clear_bmap_size(uint64_t pages, uint8_t shift)
  38 {
  39     return DIV_ROUND_UP(pages, 1UL << shift);
  40 }
  41
  42 /**
  43  * clear_bmap_set: set clear bitmap for the page range
  44  *
  45  * @rb: the ramblock to operate on
  46  * @start: the start page number
  47  * @size: number of pages to set in the bitmap
  48  *
  49  * Returns: None
  50  */
  51 static inline void clear_bmap_set(RAMBlock *rb, uint64_t start,
  52                                   uint64_t npages)
  53 {
  54     uint8_t shift = rb->clear_bmap_shift;
  55
  56     bitmap_set_atomic(rb->clear_bmap, start >> shift,
  57                       clear_bmap_size(npages, shift));
  58 }
  59
  60 /**
  61  * clear_bmap_test_and_clear: test clear bitmap for the page, clear if set
  62  *
  63  * @rb: the ramblock to operate on
  64  * @page: the page number to check
  65  *
  66  * Returns: true if the bit was set, false otherwise
  67  */
  68 static inline bool clear_bmap_test_and_clear(RAMBlock *rb, uint64_t page)
  69 {
  70     uint8_t shift = rb->clear_bmap_shift;
  71
  72     return bitmap_test_and_clear_atomic(rb->clear_bmap, page >> shift, 1);
  73 }
  74
  75 static inline bool offset_in_ramblock(RAMBlock *b, ram_addr_t offset)
  76 {
  77     return (b && b->host && offset < b->used_length) ? true : false;
  78 }
  79
  80 static inline void *ramblock_ptr(RAMBlock *block, ram_addr_t offset)
  81 {
  82     assert(offset_in_ramblock(block, offset));
  83     return (char *)block->host + offset;
  84 }
  85
  86 static inline unsigned long int ramblock_recv_bitmap_offset(void *host_addr,
  87                                                             RAMBlock *rb)
  88 {
  89     uint64_t host_addr_offset =
  90             (uint64_t)(uintptr_t)(host_addr - (void *)rb->host);
  91     return host_addr_offset >> TARGET_PAGE_BITS;
  92 }
  93
  94 bool ramblock_is_pmem(RAMBlock *rb);
  95
  96 long qemu_minrampagesize(void);
  97 long qemu_maxrampagesize(void);
  98
  99 /**
 100  * qemu_ram_alloc_from_file,
 101  * qemu_ram_alloc_from_fd:  Allocate a ram block from the specified backing
 102  *                          file or device
 103  *
 104  * Parameters:
 105  *  @size: the size in bytes of the ram block
 106  *  @mr: the memory region where the ram block is
 107  *  @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM.
 108  *  @mem_path or @fd: specify the backing file or device
 109  *  @readonly: true to open @path for reading, false for read/write.
 110  *  @errp: pointer to Error*, to store an error if it happens
 111  *
 112  * Return:
 113  *  On success, return a pointer to the ram block.
 114  *  On failure, return NULL.
 115  */
 116 RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr,
 117                                    uint32_t ram_flags, const char *mem_path,
 118                                    bool readonly, Error **errp);
 119 RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, MemoryRegion *mr,
 120                                  uint32_t ram_flags, int fd, off_t offset,
 121                                  bool readonly, Error **errp);
 122
 123 RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host,
 124                                   MemoryRegion *mr, Error **errp);
 125 RAMBlock *qemu_ram_alloc(ram_addr_t size, uint32_t ram_flags, MemoryRegion *mr,
 126                          Error **errp);
 127 RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t max_size,
 128                                     void (*resized)(const char*,
 129                                                     uint64_t length,
 130                                                     void *host),
 131                                     MemoryRegion *mr, Error **errp);
 132 void qemu_ram_free(RAMBlock *block);
 133
 134 int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp);
 135
 136 void qemu_ram_msync(RAMBlock *block, ram_addr_t start, ram_addr_t length);
 137
 138 /* Clear whole block of mem */
 139 static inline void qemu_ram_block_writeback(RAMBlock *block)
 140 {
 141     qemu_ram_msync(block, 0, block->used_length);
 142 }
 143
 144 #define DIRTY_CLIENTS_ALL     ((1 << DIRTY_MEMORY_NUM) - 1)
 145 #define DIRTY_CLIENTS_NOCODE  (DIRTY_CLIENTS_ALL & ~(1 << DIRTY_MEMORY_CODE))
 146
 147 void tb_invalidate_phys_range(ram_addr_t start, ram_addr_t end);
 148
 149 static inline bool cpu_physical_memory_get_dirty(ram_addr_t start,
 150                                                  ram_addr_t length,
 151                                                  unsigned client)
 152 {
 153     DirtyMemoryBlocks *blocks;
 154     unsigned long end, page;
 155     unsigned long idx, offset, base;
 156     bool dirty = false;
 157
 158     assert(client < DIRTY_MEMORY_NUM);
 159
 160     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
 161     page = start >> TARGET_PAGE_BITS;
 162
 163     WITH_RCU_READ_LOCK_GUARD() {
 164         blocks = qatomic_rcu_read(&ram_list.dirty_memory[client]);
 165
 166         idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 167         offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 168         base = page - offset;
 169         while (page < end) {
 170             unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
 171             unsigned long num = next - base;
 172             unsigned long found = find_next_bit(blocks->blocks[idx],
 173                                                 num, offset);
 174             if (found < num) {
 175                 dirty = true;
 176                 break;
 177             }
 178
 179             page = next;
 180             idx++;
 181             offset = 0;
 182             base += DIRTY_MEMORY_BLOCK_SIZE;
 183         }
 184     }
 185
 186     return dirty;
 187 }
 188
 189 static inline bool cpu_physical_memory_all_dirty(ram_addr_t start,
 190                                                  ram_addr_t length,
 191                                                  unsigned client)
 192 {
 193     DirtyMemoryBlocks *blocks;
 194     unsigned long end, page;
 195     unsigned long idx, offset, base;
 196     bool dirty = true;
 197
 198     assert(client < DIRTY_MEMORY_NUM);
 199
 200     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
 201     page = start >> TARGET_PAGE_BITS;
 202
 203     RCU_READ_LOCK_GUARD();
 204
 205     blocks = qatomic_rcu_read(&ram_list.dirty_memory[client]);
 206
 207     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 208     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 209     base = page - offset;
 210     while (page < end) {
 211         unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
 212         unsigned long num = next - base;
 213         unsigned long found = find_next_zero_bit(blocks->blocks[idx], num, offset);
 214         if (found < num) {
 215             dirty = false;
 216             break;
 217         }
 218
 219         page = next;
 220         idx++;
 221         offset = 0;
 222         base += DIRTY_MEMORY_BLOCK_SIZE;
 223     }
 224
 225     return dirty;
 226 }
 227
 228 static inline bool cpu_physical_memory_get_dirty_flag(ram_addr_t addr,
 229                                                       unsigned client)
 230 {
 231     return cpu_physical_memory_get_dirty(addr, 1, client);
 232 }
 233
 234 static inline bool cpu_physical_memory_is_clean(ram_addr_t addr)
 235 {
 236     bool vga = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_VGA);
 237     bool code = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_CODE);
 238     bool migration =
 239         cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_MIGRATION);
 240     return !(vga && code && migration);
 241 }
 242
 243 static inline uint8_t cpu_physical_memory_range_includes_clean(ram_addr_t start,
 244                                                                ram_addr_t length,
 245                                                                uint8_t mask)
 246 {
 247     uint8_t ret = 0;
 248
 249     if (mask & (1 << DIRTY_MEMORY_VGA) &&
 250         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_VGA)) {
 251         ret |= (1 << DIRTY_MEMORY_VGA);
 252     }
 253     if (mask & (1 << DIRTY_MEMORY_CODE) &&
 254         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_CODE)) {
 255         ret |= (1 << DIRTY_MEMORY_CODE);
 256     }
 257     if (mask & (1 << DIRTY_MEMORY_MIGRATION) &&
 258         !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_MIGRATION)) {
 259         ret |= (1 << DIRTY_MEMORY_MIGRATION);
 260     }
 261     return ret;
 262 }
 263
 264 static inline void cpu_physical_memory_set_dirty_flag(ram_addr_t addr,
 265                                                       unsigned client)
 266 {
 267     unsigned long page, idx, offset;
 268     DirtyMemoryBlocks *blocks;
 269
 270     assert(client < DIRTY_MEMORY_NUM);
 271
 272     page = addr >> TARGET_PAGE_BITS;
 273     idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 274     offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 275
 276     RCU_READ_LOCK_GUARD();
 277
 278     blocks = qatomic_rcu_read(&ram_list.dirty_memory[client]);
 279
 280     set_bit_atomic(offset, blocks->blocks[idx]);
 281 }
 282
 283 static inline void cpu_physical_memory_set_dirty_range(ram_addr_t start,
 284                                                        ram_addr_t length,
 285                                                        uint8_t mask)
 286 {
 287     DirtyMemoryBlocks *blocks[DIRTY_MEMORY_NUM];
 288     unsigned long end, page;
 289     unsigned long idx, offset, base;
 290     int i;
 291
 292     if (!mask && !xen_enabled()) {
 293         return;
 294     }
 295
 296     end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS;
 297     page = start >> TARGET_PAGE_BITS;
 298
 299     WITH_RCU_READ_LOCK_GUARD() {
 300         for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
 301             blocks[i] = qatomic_rcu_read(&ram_list.dirty_memory[i]);
 302         }
 303
 304         idx = page / DIRTY_MEMORY_BLOCK_SIZE;
 305         offset = page % DIRTY_MEMORY_BLOCK_SIZE;
 306         base = page - offset;
 307         while (page < end) {
 308             unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE);
 309
 310             if (likely(mask & (1 << DIRTY_MEMORY_MIGRATION))) {
 311                 bitmap_set_atomic(blocks[DIRTY_MEMORY_MIGRATION]->blocks[idx],
 312                                   offset, next - page);
 313             }
 314             if (unlikely(mask & (1 << DIRTY_MEMORY_VGA))) {
 315                 bitmap_set_atomic(blocks[DIRTY_MEMORY_VGA]->blocks[idx],
 316                                   offset, next - page);
 317             }
 318             if (unlikely(mask & (1 << DIRTY_MEMORY_CODE))) {
 319                 bitmap_set_atomic(blocks[DIRTY_MEMORY_CODE]->blocks[idx],
 320                                   offset, next - page);
 321             }
 322
 323             page = next;
 324             idx++;
 325             offset = 0;
 326             base += DIRTY_MEMORY_BLOCK_SIZE;
 327         }
 328     }
 329
 330     xen_hvm_modified_memory(start, length);
 331 }
 332
 333 #if !defined(_WIN32)
 334 static inline void cpu_physical_memory_set_dirty_lebitmap(unsigned long *bitmap,
 335                                                           ram_addr_t start,
 336                                                           ram_addr_t pages)
 337 {
 338     unsigned long i, j;
 339     unsigned long page_number, c;
 340     hwaddr addr;
 341     ram_addr_t ram_addr;
 342     unsigned long len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS;
 343     unsigned long hpratio = qemu_real_host_page_size / TARGET_PAGE_SIZE;
 344     unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
 345
 346     /* start address is aligned at the start of a word? */
 347     if ((((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) &&
 348         (hpratio == 1)) {
 349         unsigned long **blocks[DIRTY_MEMORY_NUM];
 350         unsigned long idx;
 351         unsigned long offset;
 352         long k;
 353         long nr = BITS_TO_LONGS(pages);
 354
 355         idx = (start >> TARGET_PAGE_BITS) / DIRTY_MEMORY_BLOCK_SIZE;
 356         offset = BIT_WORD((start >> TARGET_PAGE_BITS) %
 357                           DIRTY_MEMORY_BLOCK_SIZE);
 358
 359         WITH_RCU_READ_LOCK_GUARD() {
 360             for (i = 0; i < DIRTY_MEMORY_NUM; i++) {
 361                 blocks[i] =
 362                     qatomic_rcu_read(&ram_list.dirty_memory[i])->blocks;
 363             }
 364
 365             for (k = 0; k < nr; k++) {
 366                 if (bitmap[k]) {
 367                     unsigned long temp = leul_to_cpu(bitmap[k]);
 368
 369                     qatomic_or(&blocks[DIRTY_MEMORY_VGA][idx][offset], temp);
 370
 371                     if (global_dirty_log) {
 372                         qatomic_or(
 373                                 &blocks[DIRTY_MEMORY_MIGRATION][idx][offset],
 374                                 temp);
 375                     }
 376
 377                     if (tcg_enabled()) {
 378                         qatomic_or(&blocks[DIRTY_MEMORY_CODE][idx][offset],
 379                                    temp);
 380                     }
 381                 }
 382
 383                 if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
 384                     offset = 0;
 385                     idx++;
 386                 }
 387             }
 388         }
 389
 390         xen_hvm_modified_memory(start, pages << TARGET_PAGE_BITS);
 391     } else {
 392         uint8_t clients = tcg_enabled() ? DIRTY_CLIENTS_ALL : DIRTY_CLIENTS_NOCODE;
 393
 394         if (!global_dirty_log) {
 395             clients &= ~(1 << DIRTY_MEMORY_MIGRATION);
 396         }
 397
 398         /*
 399          * bitmap-traveling is faster than memory-traveling (for addr...)
 400          * especially when most of the memory is not dirty.
 401          */
 402         for (i = 0; i < len; i++) {
 403             if (bitmap[i] != 0) {
 404                 c = leul_to_cpu(bitmap[i]);
 405                 do {
 406                     j = ctzl(c);
 407                     c &= ~(1ul << j);
 408                     page_number = (i * HOST_LONG_BITS + j) * hpratio;
 409                     addr = page_number * TARGET_PAGE_SIZE;
 410                     ram_addr = start + addr;
 411                     cpu_physical_memory_set_dirty_range(ram_addr,
 412                                        TARGET_PAGE_SIZE * hpratio, clients);
 413                 } while (c != 0);
 414             }
 415         }
 416     }
 417 }
 418 #endif /* not _WIN32 */
 419
 420 bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start,
 421                                               ram_addr_t length,
 422                                               unsigned client);
 423
 424 DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty
 425     (MemoryRegion *mr, hwaddr offset, hwaddr length, unsigned client);
 426
 427 bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap,
 428                                             ram_addr_t start,
 429                                             ram_addr_t length);
 430
 431 static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start,
 432                                                          ram_addr_t length)
 433 {
 434     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_MIGRATION);
 435     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_VGA);
 436     cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_CODE);
 437 }
 438
 439
 440 /* Called with RCU critical section */
 441 static inline
 442 uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb,
 443                                                ram_addr_t start,
 444                                                ram_addr_t length)
 445 {
 446     ram_addr_t addr;
 447     unsigned long word = BIT_WORD((start + rb->offset) >> TARGET_PAGE_BITS);
 448     uint64_t num_dirty = 0;
 449     unsigned long *dest = rb->bmap;
 450
 451     /* start address and length is aligned at the start of a word? */
 452     if (((word * BITS_PER_LONG) << TARGET_PAGE_BITS) ==
 453          (start + rb->offset) &&
 454         !(length & ((BITS_PER_LONG << TARGET_PAGE_BITS) - 1))) {
 455         int k;
 456         int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS);
 457         unsigned long * const *src;
 458         unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE;
 459         unsigned long offset = BIT_WORD((word * BITS_PER_LONG) %
 460                                         DIRTY_MEMORY_BLOCK_SIZE);
 461         unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS);
 462
 463         src = qatomic_rcu_read(
 464                 &ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION])->blocks;
 465
 466         for (k = page; k < page + nr; k++) {
 467             if (src[idx][offset]) {
 468                 unsigned long bits = qatomic_xchg(&src[idx][offset], 0);
 469                 unsigned long new_dirty;
 470                 new_dirty = ~dest[k];
 471                 dest[k] |= bits;
 472                 new_dirty &= bits;
 473                 num_dirty += ctpopl(new_dirty);
 474             }
 475
 476             if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) {
 477                 offset = 0;
 478                 idx++;
 479             }
 480         }
 481
 482         if (rb->clear_bmap) {
 483             /*
 484              * Postpone the dirty bitmap clear to the point before we
 485              * really send the pages, also we will split the clear
 486              * dirty procedure into smaller chunks.
 487              */
 488             clear_bmap_set(rb, start >> TARGET_PAGE_BITS,
 489                            length >> TARGET_PAGE_BITS);
 490         } else {
 491             /* Slow path - still do that in a huge chunk */
 492             memory_region_clear_dirty_bitmap(rb->mr, start, length);
 493         }
 494     } else {
 495         ram_addr_t offset = rb->offset;
 496
 497         for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) {
 498             if (cpu_physical_memory_test_and_clear_dirty(
 499                         start + addr + offset,
 500                         TARGET_PAGE_SIZE,
 501                         DIRTY_MEMORY_MIGRATION)) {
 502                 long k = (start + addr) >> TARGET_PAGE_BITS;
 503                 if (!test_and_set_bit(k, dest)) {
 504                     num_dirty++;
 505                 }
 506             }
 507         }
 508     }
 509
 510     return num_dirty;
 511 }
 512 #endif
 513 #endif