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