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 SYSTEM_RAM_ADDR_H 20 #define SYSTEM_RAM_ADDR_H 21 22 #include "system/xen.h" 23 #include "system/tcg.h" 24 #include "exec/cputlb.h" 25 #include "exec/ramlist.h" 26 #include "system/ramblock.h" 27 #include "system/memory.h" 28 #include "exec/target_page.h" 29 #include "qemu/rcu.h" 30 31 #include "exec/hwaddr.h" 32 #include "exec/cpu-common.h" 33 34 extern uint64_t total_dirty_pages; 35 36 /** 37 * clear_bmap_size: calculate clear bitmap size 38 * 39 * @pages: number of guest pages 40 * @shift: guest page number shift 41 * 42 * Returns: number of bits for the clear bitmap 43 */ 44 static inline long clear_bmap_size(uint64_t pages, uint8_t shift) 45 { 46 return DIV_ROUND_UP(pages, 1UL << shift); 47 } 48 49 /** 50 * clear_bmap_set: set clear bitmap for the page range. Must be with 51 * bitmap_mutex held. 52 * 53 * @rb: the ramblock to operate on 54 * @start: the start page number 55 * @size: number of pages to set in the bitmap 56 * 57 * Returns: None 58 */ 59 static inline void clear_bmap_set(RAMBlock *rb, uint64_t start, 60 uint64_t npages) 61 { 62 uint8_t shift = rb->clear_bmap_shift; 63 64 bitmap_set(rb->clear_bmap, start >> shift, clear_bmap_size(npages, shift)); 65 } 66 67 /** 68 * clear_bmap_test_and_clear: test clear bitmap for the page, clear if set. 69 * Must be with bitmap_mutex held. 70 * 71 * @rb: the ramblock to operate on 72 * @page: the page number to check 73 * 74 * Returns: true if the bit was set, false otherwise 75 */ 76 static inline bool clear_bmap_test_and_clear(RAMBlock *rb, uint64_t page) 77 { 78 uint8_t shift = rb->clear_bmap_shift; 79 80 return bitmap_test_and_clear(rb->clear_bmap, page >> shift, 1); 81 } 82 83 static inline bool offset_in_ramblock(RAMBlock *b, ram_addr_t offset) 84 { 85 return (b && b->host && offset < b->used_length) ? true : false; 86 } 87 88 static inline void *ramblock_ptr(RAMBlock *block, ram_addr_t offset) 89 { 90 assert(offset_in_ramblock(block, offset)); 91 return (char *)block->host + offset; 92 } 93 94 static inline unsigned long int ramblock_recv_bitmap_offset(void *host_addr, 95 RAMBlock *rb) 96 { 97 uint64_t host_addr_offset = 98 (uint64_t)(uintptr_t)(host_addr - (void *)rb->host); 99 return host_addr_offset >> TARGET_PAGE_BITS; 100 } 101 102 bool ramblock_is_pmem(RAMBlock *rb); 103 104 /** 105 * qemu_ram_alloc_from_file, 106 * qemu_ram_alloc_from_fd: Allocate a ram block from the specified backing 107 * file or device 108 * 109 * Parameters: 110 * @size: the size in bytes of the ram block 111 * @max_size: the maximum size of the block after resizing 112 * @mr: the memory region where the ram block is 113 * @resized: callback after calls to qemu_ram_resize 114 * @ram_flags: RamBlock flags. Supported flags: RAM_SHARED, RAM_PMEM, 115 * RAM_NORESERVE, RAM_PROTECTED, RAM_NAMED_FILE, RAM_READONLY, 116 * RAM_READONLY_FD, RAM_GUEST_MEMFD 117 * @mem_path or @fd: specify the backing file or device 118 * @offset: Offset into target file 119 * @grow: extend file if necessary (but an empty file is always extended). 120 * @errp: pointer to Error*, to store an error if it happens 121 * 122 * Return: 123 * On success, return a pointer to the ram block. 124 * On failure, return NULL. 125 */ 126 typedef void (*qemu_ram_resize_cb)(const char *, uint64_t length, void *host); 127 128 RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr, 129 uint32_t ram_flags, const char *mem_path, 130 off_t offset, Error **errp); 131 RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, ram_addr_t max_size, 132 qemu_ram_resize_cb resized, MemoryRegion *mr, 133 uint32_t ram_flags, int fd, off_t offset, 134 bool grow, 135 Error **errp); 136 137 RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host, 138 MemoryRegion *mr, Error **errp); 139 RAMBlock *qemu_ram_alloc(ram_addr_t size, uint32_t ram_flags, MemoryRegion *mr, 140 Error **errp); 141 RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t max_size, 142 qemu_ram_resize_cb resized, 143 MemoryRegion *mr, Error **errp); 144 void qemu_ram_free(RAMBlock *block); 145 146 int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp); 147 148 void qemu_ram_msync(RAMBlock *block, ram_addr_t start, ram_addr_t length); 149 150 /* Clear whole block of mem */ 151 static inline void qemu_ram_block_writeback(RAMBlock *block) 152 { 153 qemu_ram_msync(block, 0, block->used_length); 154 } 155 156 #define DIRTY_CLIENTS_ALL ((1 << DIRTY_MEMORY_NUM) - 1) 157 #define DIRTY_CLIENTS_NOCODE (DIRTY_CLIENTS_ALL & ~(1 << DIRTY_MEMORY_CODE)) 158 159 static inline bool cpu_physical_memory_get_dirty(ram_addr_t start, 160 ram_addr_t length, 161 unsigned client) 162 { 163 DirtyMemoryBlocks *blocks; 164 unsigned long end, page; 165 unsigned long idx, offset, base; 166 bool dirty = false; 167 168 assert(client < DIRTY_MEMORY_NUM); 169 170 end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS; 171 page = start >> TARGET_PAGE_BITS; 172 173 WITH_RCU_READ_LOCK_GUARD() { 174 blocks = qatomic_rcu_read(&ram_list.dirty_memory[client]); 175 176 idx = page / DIRTY_MEMORY_BLOCK_SIZE; 177 offset = page % DIRTY_MEMORY_BLOCK_SIZE; 178 base = page - offset; 179 while (page < end) { 180 unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE); 181 unsigned long num = next - base; 182 unsigned long found = find_next_bit(blocks->blocks[idx], 183 num, offset); 184 if (found < num) { 185 dirty = true; 186 break; 187 } 188 189 page = next; 190 idx++; 191 offset = 0; 192 base += DIRTY_MEMORY_BLOCK_SIZE; 193 } 194 } 195 196 return dirty; 197 } 198 199 static inline bool cpu_physical_memory_all_dirty(ram_addr_t start, 200 ram_addr_t length, 201 unsigned client) 202 { 203 DirtyMemoryBlocks *blocks; 204 unsigned long end, page; 205 unsigned long idx, offset, base; 206 bool dirty = true; 207 208 assert(client < DIRTY_MEMORY_NUM); 209 210 end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS; 211 page = start >> TARGET_PAGE_BITS; 212 213 RCU_READ_LOCK_GUARD(); 214 215 blocks = qatomic_rcu_read(&ram_list.dirty_memory[client]); 216 217 idx = page / DIRTY_MEMORY_BLOCK_SIZE; 218 offset = page % DIRTY_MEMORY_BLOCK_SIZE; 219 base = page - offset; 220 while (page < end) { 221 unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE); 222 unsigned long num = next - base; 223 unsigned long found = find_next_zero_bit(blocks->blocks[idx], num, offset); 224 if (found < num) { 225 dirty = false; 226 break; 227 } 228 229 page = next; 230 idx++; 231 offset = 0; 232 base += DIRTY_MEMORY_BLOCK_SIZE; 233 } 234 235 return dirty; 236 } 237 238 static inline bool cpu_physical_memory_get_dirty_flag(ram_addr_t addr, 239 unsigned client) 240 { 241 return cpu_physical_memory_get_dirty(addr, 1, client); 242 } 243 244 static inline bool cpu_physical_memory_is_clean(ram_addr_t addr) 245 { 246 bool vga = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_VGA); 247 bool code = cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_CODE); 248 bool migration = 249 cpu_physical_memory_get_dirty_flag(addr, DIRTY_MEMORY_MIGRATION); 250 return !(vga && code && migration); 251 } 252 253 static inline uint8_t cpu_physical_memory_range_includes_clean(ram_addr_t start, 254 ram_addr_t length, 255 uint8_t mask) 256 { 257 uint8_t ret = 0; 258 259 if (mask & (1 << DIRTY_MEMORY_VGA) && 260 !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_VGA)) { 261 ret |= (1 << DIRTY_MEMORY_VGA); 262 } 263 if (mask & (1 << DIRTY_MEMORY_CODE) && 264 !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_CODE)) { 265 ret |= (1 << DIRTY_MEMORY_CODE); 266 } 267 if (mask & (1 << DIRTY_MEMORY_MIGRATION) && 268 !cpu_physical_memory_all_dirty(start, length, DIRTY_MEMORY_MIGRATION)) { 269 ret |= (1 << DIRTY_MEMORY_MIGRATION); 270 } 271 return ret; 272 } 273 274 static inline void cpu_physical_memory_set_dirty_flag(ram_addr_t addr, 275 unsigned client) 276 { 277 unsigned long page, idx, offset; 278 DirtyMemoryBlocks *blocks; 279 280 assert(client < DIRTY_MEMORY_NUM); 281 282 page = addr >> TARGET_PAGE_BITS; 283 idx = page / DIRTY_MEMORY_BLOCK_SIZE; 284 offset = page % DIRTY_MEMORY_BLOCK_SIZE; 285 286 RCU_READ_LOCK_GUARD(); 287 288 blocks = qatomic_rcu_read(&ram_list.dirty_memory[client]); 289 290 set_bit_atomic(offset, blocks->blocks[idx]); 291 } 292 293 static inline void cpu_physical_memory_set_dirty_range(ram_addr_t start, 294 ram_addr_t length, 295 uint8_t mask) 296 { 297 DirtyMemoryBlocks *blocks[DIRTY_MEMORY_NUM]; 298 unsigned long end, page; 299 unsigned long idx, offset, base; 300 int i; 301 302 if (!mask && !xen_enabled()) { 303 return; 304 } 305 306 end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS; 307 page = start >> TARGET_PAGE_BITS; 308 309 WITH_RCU_READ_LOCK_GUARD() { 310 for (i = 0; i < DIRTY_MEMORY_NUM; i++) { 311 blocks[i] = qatomic_rcu_read(&ram_list.dirty_memory[i]); 312 } 313 314 idx = page / DIRTY_MEMORY_BLOCK_SIZE; 315 offset = page % DIRTY_MEMORY_BLOCK_SIZE; 316 base = page - offset; 317 while (page < end) { 318 unsigned long next = MIN(end, base + DIRTY_MEMORY_BLOCK_SIZE); 319 320 if (likely(mask & (1 << DIRTY_MEMORY_MIGRATION))) { 321 bitmap_set_atomic(blocks[DIRTY_MEMORY_MIGRATION]->blocks[idx], 322 offset, next - page); 323 } 324 if (unlikely(mask & (1 << DIRTY_MEMORY_VGA))) { 325 bitmap_set_atomic(blocks[DIRTY_MEMORY_VGA]->blocks[idx], 326 offset, next - page); 327 } 328 if (unlikely(mask & (1 << DIRTY_MEMORY_CODE))) { 329 bitmap_set_atomic(blocks[DIRTY_MEMORY_CODE]->blocks[idx], 330 offset, next - page); 331 } 332 333 page = next; 334 idx++; 335 offset = 0; 336 base += DIRTY_MEMORY_BLOCK_SIZE; 337 } 338 } 339 340 if (xen_enabled()) { 341 xen_hvm_modified_memory(start, length); 342 } 343 } 344 345 #if !defined(_WIN32) 346 347 /* 348 * Contrary to cpu_physical_memory_sync_dirty_bitmap() this function returns 349 * the number of dirty pages in @bitmap passed as argument. On the other hand, 350 * cpu_physical_memory_sync_dirty_bitmap() returns newly dirtied pages that 351 * weren't set in the global migration bitmap. 352 */ 353 static inline 354 uint64_t cpu_physical_memory_set_dirty_lebitmap(unsigned long *bitmap, 355 ram_addr_t start, 356 ram_addr_t pages) 357 { 358 unsigned long i, j; 359 unsigned long page_number, c, nbits; 360 hwaddr addr; 361 ram_addr_t ram_addr; 362 uint64_t num_dirty = 0; 363 unsigned long len = (pages + HOST_LONG_BITS - 1) / HOST_LONG_BITS; 364 unsigned long hpratio = qemu_real_host_page_size() / TARGET_PAGE_SIZE; 365 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); 366 367 /* start address is aligned at the start of a word? */ 368 if ((((page * BITS_PER_LONG) << TARGET_PAGE_BITS) == start) && 369 (hpratio == 1)) { 370 unsigned long **blocks[DIRTY_MEMORY_NUM]; 371 unsigned long idx; 372 unsigned long offset; 373 long k; 374 long nr = BITS_TO_LONGS(pages); 375 376 idx = (start >> TARGET_PAGE_BITS) / DIRTY_MEMORY_BLOCK_SIZE; 377 offset = BIT_WORD((start >> TARGET_PAGE_BITS) % 378 DIRTY_MEMORY_BLOCK_SIZE); 379 380 WITH_RCU_READ_LOCK_GUARD() { 381 for (i = 0; i < DIRTY_MEMORY_NUM; i++) { 382 blocks[i] = 383 qatomic_rcu_read(&ram_list.dirty_memory[i])->blocks; 384 } 385 386 for (k = 0; k < nr; k++) { 387 if (bitmap[k]) { 388 unsigned long temp = leul_to_cpu(bitmap[k]); 389 390 nbits = ctpopl(temp); 391 qatomic_or(&blocks[DIRTY_MEMORY_VGA][idx][offset], temp); 392 393 if (global_dirty_tracking) { 394 qatomic_or( 395 &blocks[DIRTY_MEMORY_MIGRATION][idx][offset], 396 temp); 397 if (unlikely( 398 global_dirty_tracking & GLOBAL_DIRTY_DIRTY_RATE)) { 399 total_dirty_pages += nbits; 400 } 401 } 402 403 num_dirty += nbits; 404 405 if (tcg_enabled()) { 406 qatomic_or(&blocks[DIRTY_MEMORY_CODE][idx][offset], 407 temp); 408 } 409 } 410 411 if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) { 412 offset = 0; 413 idx++; 414 } 415 } 416 } 417 418 if (xen_enabled()) { 419 xen_hvm_modified_memory(start, pages << TARGET_PAGE_BITS); 420 } 421 } else { 422 uint8_t clients = tcg_enabled() ? DIRTY_CLIENTS_ALL : DIRTY_CLIENTS_NOCODE; 423 424 if (!global_dirty_tracking) { 425 clients &= ~(1 << DIRTY_MEMORY_MIGRATION); 426 } 427 428 /* 429 * bitmap-traveling is faster than memory-traveling (for addr...) 430 * especially when most of the memory is not dirty. 431 */ 432 for (i = 0; i < len; i++) { 433 if (bitmap[i] != 0) { 434 c = leul_to_cpu(bitmap[i]); 435 nbits = ctpopl(c); 436 if (unlikely(global_dirty_tracking & GLOBAL_DIRTY_DIRTY_RATE)) { 437 total_dirty_pages += nbits; 438 } 439 num_dirty += nbits; 440 do { 441 j = ctzl(c); 442 c &= ~(1ul << j); 443 page_number = (i * HOST_LONG_BITS + j) * hpratio; 444 addr = page_number * TARGET_PAGE_SIZE; 445 ram_addr = start + addr; 446 cpu_physical_memory_set_dirty_range(ram_addr, 447 TARGET_PAGE_SIZE * hpratio, clients); 448 } while (c != 0); 449 } 450 } 451 } 452 453 return num_dirty; 454 } 455 #endif /* not _WIN32 */ 456 457 static inline void cpu_physical_memory_dirty_bits_cleared(ram_addr_t start, 458 ram_addr_t length) 459 { 460 if (tcg_enabled()) { 461 tlb_reset_dirty_range_all(start, length); 462 } 463 464 } 465 bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start, 466 ram_addr_t length, 467 unsigned client); 468 469 DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty 470 (MemoryRegion *mr, hwaddr offset, hwaddr length, unsigned client); 471 472 bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap, 473 ram_addr_t start, 474 ram_addr_t length); 475 476 static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start, 477 ram_addr_t length) 478 { 479 cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_MIGRATION); 480 cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_VGA); 481 cpu_physical_memory_test_and_clear_dirty(start, length, DIRTY_MEMORY_CODE); 482 } 483 484 485 /* Called with RCU critical section */ 486 static inline 487 uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, 488 ram_addr_t start, 489 ram_addr_t length) 490 { 491 ram_addr_t addr; 492 unsigned long word = BIT_WORD((start + rb->offset) >> TARGET_PAGE_BITS); 493 uint64_t num_dirty = 0; 494 unsigned long *dest = rb->bmap; 495 496 /* start address and length is aligned at the start of a word? */ 497 if (((word * BITS_PER_LONG) << TARGET_PAGE_BITS) == 498 (start + rb->offset) && 499 !(length & ((BITS_PER_LONG << TARGET_PAGE_BITS) - 1))) { 500 int k; 501 int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); 502 unsigned long * const *src; 503 unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE; 504 unsigned long offset = BIT_WORD((word * BITS_PER_LONG) % 505 DIRTY_MEMORY_BLOCK_SIZE); 506 unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); 507 508 src = qatomic_rcu_read( 509 &ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION])->blocks; 510 511 for (k = page; k < page + nr; k++) { 512 if (src[idx][offset]) { 513 unsigned long bits = qatomic_xchg(&src[idx][offset], 0); 514 unsigned long new_dirty; 515 new_dirty = ~dest[k]; 516 dest[k] |= bits; 517 new_dirty &= bits; 518 num_dirty += ctpopl(new_dirty); 519 } 520 521 if (++offset >= BITS_TO_LONGS(DIRTY_MEMORY_BLOCK_SIZE)) { 522 offset = 0; 523 idx++; 524 } 525 } 526 if (num_dirty) { 527 cpu_physical_memory_dirty_bits_cleared(start, length); 528 } 529 530 if (rb->clear_bmap) { 531 /* 532 * Postpone the dirty bitmap clear to the point before we 533 * really send the pages, also we will split the clear 534 * dirty procedure into smaller chunks. 535 */ 536 clear_bmap_set(rb, start >> TARGET_PAGE_BITS, 537 length >> TARGET_PAGE_BITS); 538 } else { 539 /* Slow path - still do that in a huge chunk */ 540 memory_region_clear_dirty_bitmap(rb->mr, start, length); 541 } 542 } else { 543 ram_addr_t offset = rb->offset; 544 545 for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { 546 if (cpu_physical_memory_test_and_clear_dirty( 547 start + addr + offset, 548 TARGET_PAGE_SIZE, 549 DIRTY_MEMORY_MIGRATION)) { 550 long k = (start + addr) >> TARGET_PAGE_BITS; 551 if (!test_and_set_bit(k, dest)) { 552 num_dirty++; 553 } 554 } 555 } 556 } 557 558 return num_dirty; 559 } 560 561 #endif 562