1 #include <linux/mm.h> 2 #include <linux/vmacache.h> 3 #include <linux/hugetlb.h> 4 #include <linux/huge_mm.h> 5 #include <linux/mount.h> 6 #include <linux/seq_file.h> 7 #include <linux/highmem.h> 8 #include <linux/ptrace.h> 9 #include <linux/slab.h> 10 #include <linux/pagemap.h> 11 #include <linux/mempolicy.h> 12 #include <linux/rmap.h> 13 #include <linux/swap.h> 14 #include <linux/swapops.h> 15 #include <linux/mmu_notifier.h> 16 17 #include <asm/elf.h> 18 #include <asm/uaccess.h> 19 #include <asm/tlbflush.h> 20 #include "internal.h" 21 22 void task_mem(struct seq_file *m, struct mm_struct *mm) 23 { 24 unsigned long data, text, lib, swap; 25 unsigned long hiwater_vm, total_vm, hiwater_rss, total_rss; 26 27 /* 28 * Note: to minimize their overhead, mm maintains hiwater_vm and 29 * hiwater_rss only when about to *lower* total_vm or rss. Any 30 * collector of these hiwater stats must therefore get total_vm 31 * and rss too, which will usually be the higher. Barriers? not 32 * worth the effort, such snapshots can always be inconsistent. 33 */ 34 hiwater_vm = total_vm = mm->total_vm; 35 if (hiwater_vm < mm->hiwater_vm) 36 hiwater_vm = mm->hiwater_vm; 37 hiwater_rss = total_rss = get_mm_rss(mm); 38 if (hiwater_rss < mm->hiwater_rss) 39 hiwater_rss = mm->hiwater_rss; 40 41 data = mm->total_vm - mm->shared_vm - mm->stack_vm; 42 text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) >> 10; 43 lib = (mm->exec_vm << (PAGE_SHIFT-10)) - text; 44 swap = get_mm_counter(mm, MM_SWAPENTS); 45 seq_printf(m, 46 "VmPeak:\t%8lu kB\n" 47 "VmSize:\t%8lu kB\n" 48 "VmLck:\t%8lu kB\n" 49 "VmPin:\t%8lu kB\n" 50 "VmHWM:\t%8lu kB\n" 51 "VmRSS:\t%8lu kB\n" 52 "VmData:\t%8lu kB\n" 53 "VmStk:\t%8lu kB\n" 54 "VmExe:\t%8lu kB\n" 55 "VmLib:\t%8lu kB\n" 56 "VmPTE:\t%8lu kB\n" 57 "VmSwap:\t%8lu kB\n", 58 hiwater_vm << (PAGE_SHIFT-10), 59 total_vm << (PAGE_SHIFT-10), 60 mm->locked_vm << (PAGE_SHIFT-10), 61 mm->pinned_vm << (PAGE_SHIFT-10), 62 hiwater_rss << (PAGE_SHIFT-10), 63 total_rss << (PAGE_SHIFT-10), 64 data << (PAGE_SHIFT-10), 65 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 66 (PTRS_PER_PTE * sizeof(pte_t) * 67 atomic_long_read(&mm->nr_ptes)) >> 10, 68 swap << (PAGE_SHIFT-10)); 69 } 70 71 unsigned long task_vsize(struct mm_struct *mm) 72 { 73 return PAGE_SIZE * mm->total_vm; 74 } 75 76 unsigned long task_statm(struct mm_struct *mm, 77 unsigned long *shared, unsigned long *text, 78 unsigned long *data, unsigned long *resident) 79 { 80 *shared = get_mm_counter(mm, MM_FILEPAGES); 81 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 82 >> PAGE_SHIFT; 83 *data = mm->total_vm - mm->shared_vm; 84 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 85 return mm->total_vm; 86 } 87 88 #ifdef CONFIG_NUMA 89 /* 90 * Save get_task_policy() for show_numa_map(). 91 */ 92 static void hold_task_mempolicy(struct proc_maps_private *priv) 93 { 94 struct task_struct *task = priv->task; 95 96 task_lock(task); 97 priv->task_mempolicy = get_task_policy(task); 98 mpol_get(priv->task_mempolicy); 99 task_unlock(task); 100 } 101 static void release_task_mempolicy(struct proc_maps_private *priv) 102 { 103 mpol_put(priv->task_mempolicy); 104 } 105 #else 106 static void hold_task_mempolicy(struct proc_maps_private *priv) 107 { 108 } 109 static void release_task_mempolicy(struct proc_maps_private *priv) 110 { 111 } 112 #endif 113 114 static void vma_stop(struct proc_maps_private *priv) 115 { 116 struct mm_struct *mm = priv->mm; 117 118 release_task_mempolicy(priv); 119 up_read(&mm->mmap_sem); 120 mmput(mm); 121 } 122 123 static struct vm_area_struct * 124 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma) 125 { 126 if (vma == priv->tail_vma) 127 return NULL; 128 return vma->vm_next ?: priv->tail_vma; 129 } 130 131 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma) 132 { 133 if (m->count < m->size) /* vma is copied successfully */ 134 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL; 135 } 136 137 static void *m_start(struct seq_file *m, loff_t *ppos) 138 { 139 struct proc_maps_private *priv = m->private; 140 unsigned long last_addr = m->version; 141 struct mm_struct *mm; 142 struct vm_area_struct *vma; 143 unsigned int pos = *ppos; 144 145 /* See m_cache_vma(). Zero at the start or after lseek. */ 146 if (last_addr == -1UL) 147 return NULL; 148 149 priv->task = get_proc_task(priv->inode); 150 if (!priv->task) 151 return ERR_PTR(-ESRCH); 152 153 mm = priv->mm; 154 if (!mm || !atomic_inc_not_zero(&mm->mm_users)) 155 return NULL; 156 157 down_read(&mm->mmap_sem); 158 hold_task_mempolicy(priv); 159 priv->tail_vma = get_gate_vma(mm); 160 161 if (last_addr) { 162 vma = find_vma(mm, last_addr); 163 if (vma && (vma = m_next_vma(priv, vma))) 164 return vma; 165 } 166 167 m->version = 0; 168 if (pos < mm->map_count) { 169 for (vma = mm->mmap; pos; pos--) { 170 m->version = vma->vm_start; 171 vma = vma->vm_next; 172 } 173 return vma; 174 } 175 176 /* we do not bother to update m->version in this case */ 177 if (pos == mm->map_count && priv->tail_vma) 178 return priv->tail_vma; 179 180 vma_stop(priv); 181 return NULL; 182 } 183 184 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 185 { 186 struct proc_maps_private *priv = m->private; 187 struct vm_area_struct *next; 188 189 (*pos)++; 190 next = m_next_vma(priv, v); 191 if (!next) 192 vma_stop(priv); 193 return next; 194 } 195 196 static void m_stop(struct seq_file *m, void *v) 197 { 198 struct proc_maps_private *priv = m->private; 199 200 if (!IS_ERR_OR_NULL(v)) 201 vma_stop(priv); 202 if (priv->task) { 203 put_task_struct(priv->task); 204 priv->task = NULL; 205 } 206 } 207 208 static int proc_maps_open(struct inode *inode, struct file *file, 209 const struct seq_operations *ops, int psize) 210 { 211 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 212 213 if (!priv) 214 return -ENOMEM; 215 216 priv->inode = inode; 217 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 218 if (IS_ERR(priv->mm)) { 219 int err = PTR_ERR(priv->mm); 220 221 seq_release_private(inode, file); 222 return err; 223 } 224 225 return 0; 226 } 227 228 static int proc_map_release(struct inode *inode, struct file *file) 229 { 230 struct seq_file *seq = file->private_data; 231 struct proc_maps_private *priv = seq->private; 232 233 if (priv->mm) 234 mmdrop(priv->mm); 235 236 return seq_release_private(inode, file); 237 } 238 239 static int do_maps_open(struct inode *inode, struct file *file, 240 const struct seq_operations *ops) 241 { 242 return proc_maps_open(inode, file, ops, 243 sizeof(struct proc_maps_private)); 244 } 245 246 static pid_t pid_of_stack(struct proc_maps_private *priv, 247 struct vm_area_struct *vma, bool is_pid) 248 { 249 struct inode *inode = priv->inode; 250 struct task_struct *task; 251 pid_t ret = 0; 252 253 rcu_read_lock(); 254 task = pid_task(proc_pid(inode), PIDTYPE_PID); 255 if (task) { 256 task = task_of_stack(task, vma, is_pid); 257 if (task) 258 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info); 259 } 260 rcu_read_unlock(); 261 262 return ret; 263 } 264 265 static void 266 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) 267 { 268 struct mm_struct *mm = vma->vm_mm; 269 struct file *file = vma->vm_file; 270 struct proc_maps_private *priv = m->private; 271 vm_flags_t flags = vma->vm_flags; 272 unsigned long ino = 0; 273 unsigned long long pgoff = 0; 274 unsigned long start, end; 275 dev_t dev = 0; 276 const char *name = NULL; 277 278 if (file) { 279 struct inode *inode = file_inode(vma->vm_file); 280 dev = inode->i_sb->s_dev; 281 ino = inode->i_ino; 282 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 283 } 284 285 /* We don't show the stack guard page in /proc/maps */ 286 start = vma->vm_start; 287 if (stack_guard_page_start(vma, start)) 288 start += PAGE_SIZE; 289 end = vma->vm_end; 290 if (stack_guard_page_end(vma, end)) 291 end -= PAGE_SIZE; 292 293 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 294 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ", 295 start, 296 end, 297 flags & VM_READ ? 'r' : '-', 298 flags & VM_WRITE ? 'w' : '-', 299 flags & VM_EXEC ? 'x' : '-', 300 flags & VM_MAYSHARE ? 's' : 'p', 301 pgoff, 302 MAJOR(dev), MINOR(dev), ino); 303 304 /* 305 * Print the dentry name for named mappings, and a 306 * special [heap] marker for the heap: 307 */ 308 if (file) { 309 seq_pad(m, ' '); 310 seq_path(m, &file->f_path, "\n"); 311 goto done; 312 } 313 314 if (vma->vm_ops && vma->vm_ops->name) { 315 name = vma->vm_ops->name(vma); 316 if (name) 317 goto done; 318 } 319 320 name = arch_vma_name(vma); 321 if (!name) { 322 pid_t tid; 323 324 if (!mm) { 325 name = "[vdso]"; 326 goto done; 327 } 328 329 if (vma->vm_start <= mm->brk && 330 vma->vm_end >= mm->start_brk) { 331 name = "[heap]"; 332 goto done; 333 } 334 335 tid = pid_of_stack(priv, vma, is_pid); 336 if (tid != 0) { 337 /* 338 * Thread stack in /proc/PID/task/TID/maps or 339 * the main process stack. 340 */ 341 if (!is_pid || (vma->vm_start <= mm->start_stack && 342 vma->vm_end >= mm->start_stack)) { 343 name = "[stack]"; 344 } else { 345 /* Thread stack in /proc/PID/maps */ 346 seq_pad(m, ' '); 347 seq_printf(m, "[stack:%d]", tid); 348 } 349 } 350 } 351 352 done: 353 if (name) { 354 seq_pad(m, ' '); 355 seq_puts(m, name); 356 } 357 seq_putc(m, '\n'); 358 } 359 360 static int show_map(struct seq_file *m, void *v, int is_pid) 361 { 362 show_map_vma(m, v, is_pid); 363 m_cache_vma(m, v); 364 return 0; 365 } 366 367 static int show_pid_map(struct seq_file *m, void *v) 368 { 369 return show_map(m, v, 1); 370 } 371 372 static int show_tid_map(struct seq_file *m, void *v) 373 { 374 return show_map(m, v, 0); 375 } 376 377 static const struct seq_operations proc_pid_maps_op = { 378 .start = m_start, 379 .next = m_next, 380 .stop = m_stop, 381 .show = show_pid_map 382 }; 383 384 static const struct seq_operations proc_tid_maps_op = { 385 .start = m_start, 386 .next = m_next, 387 .stop = m_stop, 388 .show = show_tid_map 389 }; 390 391 static int pid_maps_open(struct inode *inode, struct file *file) 392 { 393 return do_maps_open(inode, file, &proc_pid_maps_op); 394 } 395 396 static int tid_maps_open(struct inode *inode, struct file *file) 397 { 398 return do_maps_open(inode, file, &proc_tid_maps_op); 399 } 400 401 const struct file_operations proc_pid_maps_operations = { 402 .open = pid_maps_open, 403 .read = seq_read, 404 .llseek = seq_lseek, 405 .release = proc_map_release, 406 }; 407 408 const struct file_operations proc_tid_maps_operations = { 409 .open = tid_maps_open, 410 .read = seq_read, 411 .llseek = seq_lseek, 412 .release = proc_map_release, 413 }; 414 415 /* 416 * Proportional Set Size(PSS): my share of RSS. 417 * 418 * PSS of a process is the count of pages it has in memory, where each 419 * page is divided by the number of processes sharing it. So if a 420 * process has 1000 pages all to itself, and 1000 shared with one other 421 * process, its PSS will be 1500. 422 * 423 * To keep (accumulated) division errors low, we adopt a 64bit 424 * fixed-point pss counter to minimize division errors. So (pss >> 425 * PSS_SHIFT) would be the real byte count. 426 * 427 * A shift of 12 before division means (assuming 4K page size): 428 * - 1M 3-user-pages add up to 8KB errors; 429 * - supports mapcount up to 2^24, or 16M; 430 * - supports PSS up to 2^52 bytes, or 4PB. 431 */ 432 #define PSS_SHIFT 12 433 434 #ifdef CONFIG_PROC_PAGE_MONITOR 435 struct mem_size_stats { 436 struct vm_area_struct *vma; 437 unsigned long resident; 438 unsigned long shared_clean; 439 unsigned long shared_dirty; 440 unsigned long private_clean; 441 unsigned long private_dirty; 442 unsigned long referenced; 443 unsigned long anonymous; 444 unsigned long anonymous_thp; 445 unsigned long swap; 446 unsigned long nonlinear; 447 u64 pss; 448 }; 449 450 451 static void smaps_pte_entry(pte_t ptent, unsigned long addr, 452 unsigned long ptent_size, struct mm_walk *walk) 453 { 454 struct mem_size_stats *mss = walk->private; 455 struct vm_area_struct *vma = mss->vma; 456 pgoff_t pgoff = linear_page_index(vma, addr); 457 struct page *page = NULL; 458 int mapcount; 459 460 if (pte_present(ptent)) { 461 page = vm_normal_page(vma, addr, ptent); 462 } else if (is_swap_pte(ptent)) { 463 swp_entry_t swpent = pte_to_swp_entry(ptent); 464 465 if (!non_swap_entry(swpent)) 466 mss->swap += ptent_size; 467 else if (is_migration_entry(swpent)) 468 page = migration_entry_to_page(swpent); 469 } else if (pte_file(ptent)) { 470 if (pte_to_pgoff(ptent) != pgoff) 471 mss->nonlinear += ptent_size; 472 } 473 474 if (!page) 475 return; 476 477 if (PageAnon(page)) 478 mss->anonymous += ptent_size; 479 480 if (page->index != pgoff) 481 mss->nonlinear += ptent_size; 482 483 mss->resident += ptent_size; 484 /* Accumulate the size in pages that have been accessed. */ 485 if (pte_young(ptent) || PageReferenced(page)) 486 mss->referenced += ptent_size; 487 mapcount = page_mapcount(page); 488 if (mapcount >= 2) { 489 if (pte_dirty(ptent) || PageDirty(page)) 490 mss->shared_dirty += ptent_size; 491 else 492 mss->shared_clean += ptent_size; 493 mss->pss += (ptent_size << PSS_SHIFT) / mapcount; 494 } else { 495 if (pte_dirty(ptent) || PageDirty(page)) 496 mss->private_dirty += ptent_size; 497 else 498 mss->private_clean += ptent_size; 499 mss->pss += (ptent_size << PSS_SHIFT); 500 } 501 } 502 503 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 504 struct mm_walk *walk) 505 { 506 struct mem_size_stats *mss = walk->private; 507 struct vm_area_struct *vma = mss->vma; 508 pte_t *pte; 509 spinlock_t *ptl; 510 511 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 512 smaps_pte_entry(*(pte_t *)pmd, addr, HPAGE_PMD_SIZE, walk); 513 spin_unlock(ptl); 514 mss->anonymous_thp += HPAGE_PMD_SIZE; 515 return 0; 516 } 517 518 if (pmd_trans_unstable(pmd)) 519 return 0; 520 /* 521 * The mmap_sem held all the way back in m_start() is what 522 * keeps khugepaged out of here and from collapsing things 523 * in here. 524 */ 525 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 526 for (; addr != end; pte++, addr += PAGE_SIZE) 527 smaps_pte_entry(*pte, addr, PAGE_SIZE, walk); 528 pte_unmap_unlock(pte - 1, ptl); 529 cond_resched(); 530 return 0; 531 } 532 533 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 534 { 535 /* 536 * Don't forget to update Documentation/ on changes. 537 */ 538 static const char mnemonics[BITS_PER_LONG][2] = { 539 /* 540 * In case if we meet a flag we don't know about. 541 */ 542 [0 ... (BITS_PER_LONG-1)] = "??", 543 544 [ilog2(VM_READ)] = "rd", 545 [ilog2(VM_WRITE)] = "wr", 546 [ilog2(VM_EXEC)] = "ex", 547 [ilog2(VM_SHARED)] = "sh", 548 [ilog2(VM_MAYREAD)] = "mr", 549 [ilog2(VM_MAYWRITE)] = "mw", 550 [ilog2(VM_MAYEXEC)] = "me", 551 [ilog2(VM_MAYSHARE)] = "ms", 552 [ilog2(VM_GROWSDOWN)] = "gd", 553 [ilog2(VM_PFNMAP)] = "pf", 554 [ilog2(VM_DENYWRITE)] = "dw", 555 [ilog2(VM_LOCKED)] = "lo", 556 [ilog2(VM_IO)] = "io", 557 [ilog2(VM_SEQ_READ)] = "sr", 558 [ilog2(VM_RAND_READ)] = "rr", 559 [ilog2(VM_DONTCOPY)] = "dc", 560 [ilog2(VM_DONTEXPAND)] = "de", 561 [ilog2(VM_ACCOUNT)] = "ac", 562 [ilog2(VM_NORESERVE)] = "nr", 563 [ilog2(VM_HUGETLB)] = "ht", 564 [ilog2(VM_NONLINEAR)] = "nl", 565 [ilog2(VM_ARCH_1)] = "ar", 566 [ilog2(VM_DONTDUMP)] = "dd", 567 #ifdef CONFIG_MEM_SOFT_DIRTY 568 [ilog2(VM_SOFTDIRTY)] = "sd", 569 #endif 570 [ilog2(VM_MIXEDMAP)] = "mm", 571 [ilog2(VM_HUGEPAGE)] = "hg", 572 [ilog2(VM_NOHUGEPAGE)] = "nh", 573 [ilog2(VM_MERGEABLE)] = "mg", 574 }; 575 size_t i; 576 577 seq_puts(m, "VmFlags: "); 578 for (i = 0; i < BITS_PER_LONG; i++) { 579 if (vma->vm_flags & (1UL << i)) { 580 seq_printf(m, "%c%c ", 581 mnemonics[i][0], mnemonics[i][1]); 582 } 583 } 584 seq_putc(m, '\n'); 585 } 586 587 static int show_smap(struct seq_file *m, void *v, int is_pid) 588 { 589 struct vm_area_struct *vma = v; 590 struct mem_size_stats mss; 591 struct mm_walk smaps_walk = { 592 .pmd_entry = smaps_pte_range, 593 .mm = vma->vm_mm, 594 .private = &mss, 595 }; 596 597 memset(&mss, 0, sizeof mss); 598 mss.vma = vma; 599 /* mmap_sem is held in m_start */ 600 if (vma->vm_mm && !is_vm_hugetlb_page(vma)) 601 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); 602 603 show_map_vma(m, vma, is_pid); 604 605 seq_printf(m, 606 "Size: %8lu kB\n" 607 "Rss: %8lu kB\n" 608 "Pss: %8lu kB\n" 609 "Shared_Clean: %8lu kB\n" 610 "Shared_Dirty: %8lu kB\n" 611 "Private_Clean: %8lu kB\n" 612 "Private_Dirty: %8lu kB\n" 613 "Referenced: %8lu kB\n" 614 "Anonymous: %8lu kB\n" 615 "AnonHugePages: %8lu kB\n" 616 "Swap: %8lu kB\n" 617 "KernelPageSize: %8lu kB\n" 618 "MMUPageSize: %8lu kB\n" 619 "Locked: %8lu kB\n", 620 (vma->vm_end - vma->vm_start) >> 10, 621 mss.resident >> 10, 622 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 623 mss.shared_clean >> 10, 624 mss.shared_dirty >> 10, 625 mss.private_clean >> 10, 626 mss.private_dirty >> 10, 627 mss.referenced >> 10, 628 mss.anonymous >> 10, 629 mss.anonymous_thp >> 10, 630 mss.swap >> 10, 631 vma_kernel_pagesize(vma) >> 10, 632 vma_mmu_pagesize(vma) >> 10, 633 (vma->vm_flags & VM_LOCKED) ? 634 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 635 636 if (vma->vm_flags & VM_NONLINEAR) 637 seq_printf(m, "Nonlinear: %8lu kB\n", 638 mss.nonlinear >> 10); 639 640 show_smap_vma_flags(m, vma); 641 m_cache_vma(m, vma); 642 return 0; 643 } 644 645 static int show_pid_smap(struct seq_file *m, void *v) 646 { 647 return show_smap(m, v, 1); 648 } 649 650 static int show_tid_smap(struct seq_file *m, void *v) 651 { 652 return show_smap(m, v, 0); 653 } 654 655 static const struct seq_operations proc_pid_smaps_op = { 656 .start = m_start, 657 .next = m_next, 658 .stop = m_stop, 659 .show = show_pid_smap 660 }; 661 662 static const struct seq_operations proc_tid_smaps_op = { 663 .start = m_start, 664 .next = m_next, 665 .stop = m_stop, 666 .show = show_tid_smap 667 }; 668 669 static int pid_smaps_open(struct inode *inode, struct file *file) 670 { 671 return do_maps_open(inode, file, &proc_pid_smaps_op); 672 } 673 674 static int tid_smaps_open(struct inode *inode, struct file *file) 675 { 676 return do_maps_open(inode, file, &proc_tid_smaps_op); 677 } 678 679 const struct file_operations proc_pid_smaps_operations = { 680 .open = pid_smaps_open, 681 .read = seq_read, 682 .llseek = seq_lseek, 683 .release = proc_map_release, 684 }; 685 686 const struct file_operations proc_tid_smaps_operations = { 687 .open = tid_smaps_open, 688 .read = seq_read, 689 .llseek = seq_lseek, 690 .release = proc_map_release, 691 }; 692 693 /* 694 * We do not want to have constant page-shift bits sitting in 695 * pagemap entries and are about to reuse them some time soon. 696 * 697 * Here's the "migration strategy": 698 * 1. when the system boots these bits remain what they are, 699 * but a warning about future change is printed in log; 700 * 2. once anyone clears soft-dirty bits via clear_refs file, 701 * these flag is set to denote, that user is aware of the 702 * new API and those page-shift bits change their meaning. 703 * The respective warning is printed in dmesg; 704 * 3. In a couple of releases we will remove all the mentions 705 * of page-shift in pagemap entries. 706 */ 707 708 static bool soft_dirty_cleared __read_mostly; 709 710 enum clear_refs_types { 711 CLEAR_REFS_ALL = 1, 712 CLEAR_REFS_ANON, 713 CLEAR_REFS_MAPPED, 714 CLEAR_REFS_SOFT_DIRTY, 715 CLEAR_REFS_LAST, 716 }; 717 718 struct clear_refs_private { 719 struct vm_area_struct *vma; 720 enum clear_refs_types type; 721 }; 722 723 static inline void clear_soft_dirty(struct vm_area_struct *vma, 724 unsigned long addr, pte_t *pte) 725 { 726 #ifdef CONFIG_MEM_SOFT_DIRTY 727 /* 728 * The soft-dirty tracker uses #PF-s to catch writes 729 * to pages, so write-protect the pte as well. See the 730 * Documentation/vm/soft-dirty.txt for full description 731 * of how soft-dirty works. 732 */ 733 pte_t ptent = *pte; 734 735 if (pte_present(ptent)) { 736 ptent = pte_wrprotect(ptent); 737 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY); 738 } else if (is_swap_pte(ptent)) { 739 ptent = pte_swp_clear_soft_dirty(ptent); 740 } else if (pte_file(ptent)) { 741 ptent = pte_file_clear_soft_dirty(ptent); 742 } 743 744 set_pte_at(vma->vm_mm, addr, pte, ptent); 745 #endif 746 } 747 748 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 749 unsigned long end, struct mm_walk *walk) 750 { 751 struct clear_refs_private *cp = walk->private; 752 struct vm_area_struct *vma = cp->vma; 753 pte_t *pte, ptent; 754 spinlock_t *ptl; 755 struct page *page; 756 757 split_huge_page_pmd(vma, addr, pmd); 758 if (pmd_trans_unstable(pmd)) 759 return 0; 760 761 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 762 for (; addr != end; pte++, addr += PAGE_SIZE) { 763 ptent = *pte; 764 765 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 766 clear_soft_dirty(vma, addr, pte); 767 continue; 768 } 769 770 if (!pte_present(ptent)) 771 continue; 772 773 page = vm_normal_page(vma, addr, ptent); 774 if (!page) 775 continue; 776 777 /* Clear accessed and referenced bits. */ 778 ptep_test_and_clear_young(vma, addr, pte); 779 ClearPageReferenced(page); 780 } 781 pte_unmap_unlock(pte - 1, ptl); 782 cond_resched(); 783 return 0; 784 } 785 786 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 787 size_t count, loff_t *ppos) 788 { 789 struct task_struct *task; 790 char buffer[PROC_NUMBUF]; 791 struct mm_struct *mm; 792 struct vm_area_struct *vma; 793 enum clear_refs_types type; 794 int itype; 795 int rv; 796 797 memset(buffer, 0, sizeof(buffer)); 798 if (count > sizeof(buffer) - 1) 799 count = sizeof(buffer) - 1; 800 if (copy_from_user(buffer, buf, count)) 801 return -EFAULT; 802 rv = kstrtoint(strstrip(buffer), 10, &itype); 803 if (rv < 0) 804 return rv; 805 type = (enum clear_refs_types)itype; 806 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 807 return -EINVAL; 808 809 if (type == CLEAR_REFS_SOFT_DIRTY) { 810 soft_dirty_cleared = true; 811 pr_warn_once("The pagemap bits 55-60 has changed their meaning!" 812 " See the linux/Documentation/vm/pagemap.txt for " 813 "details.\n"); 814 } 815 816 task = get_proc_task(file_inode(file)); 817 if (!task) 818 return -ESRCH; 819 mm = get_task_mm(task); 820 if (mm) { 821 struct clear_refs_private cp = { 822 .type = type, 823 }; 824 struct mm_walk clear_refs_walk = { 825 .pmd_entry = clear_refs_pte_range, 826 .mm = mm, 827 .private = &cp, 828 }; 829 down_read(&mm->mmap_sem); 830 if (type == CLEAR_REFS_SOFT_DIRTY) 831 mmu_notifier_invalidate_range_start(mm, 0, -1); 832 for (vma = mm->mmap; vma; vma = vma->vm_next) { 833 cp.vma = vma; 834 if (is_vm_hugetlb_page(vma)) 835 continue; 836 /* 837 * Writing 1 to /proc/pid/clear_refs affects all pages. 838 * 839 * Writing 2 to /proc/pid/clear_refs only affects 840 * Anonymous pages. 841 * 842 * Writing 3 to /proc/pid/clear_refs only affects file 843 * mapped pages. 844 * 845 * Writing 4 to /proc/pid/clear_refs affects all pages. 846 */ 847 if (type == CLEAR_REFS_ANON && vma->vm_file) 848 continue; 849 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 850 continue; 851 if (type == CLEAR_REFS_SOFT_DIRTY) { 852 if (vma->vm_flags & VM_SOFTDIRTY) 853 vma->vm_flags &= ~VM_SOFTDIRTY; 854 } 855 walk_page_range(vma->vm_start, vma->vm_end, 856 &clear_refs_walk); 857 } 858 if (type == CLEAR_REFS_SOFT_DIRTY) 859 mmu_notifier_invalidate_range_end(mm, 0, -1); 860 flush_tlb_mm(mm); 861 up_read(&mm->mmap_sem); 862 mmput(mm); 863 } 864 put_task_struct(task); 865 866 return count; 867 } 868 869 const struct file_operations proc_clear_refs_operations = { 870 .write = clear_refs_write, 871 .llseek = noop_llseek, 872 }; 873 874 typedef struct { 875 u64 pme; 876 } pagemap_entry_t; 877 878 struct pagemapread { 879 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 880 pagemap_entry_t *buffer; 881 bool v2; 882 }; 883 884 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 885 #define PAGEMAP_WALK_MASK (PMD_MASK) 886 887 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 888 #define PM_STATUS_BITS 3 889 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 890 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 891 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 892 #define PM_PSHIFT_BITS 6 893 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 894 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 895 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 896 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 897 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 898 /* in "new" pagemap pshift bits are occupied with more status bits */ 899 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT)) 900 901 #define __PM_SOFT_DIRTY (1LL) 902 #define PM_PRESENT PM_STATUS(4LL) 903 #define PM_SWAP PM_STATUS(2LL) 904 #define PM_FILE PM_STATUS(1LL) 905 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0) 906 #define PM_END_OF_BUFFER 1 907 908 static inline pagemap_entry_t make_pme(u64 val) 909 { 910 return (pagemap_entry_t) { .pme = val }; 911 } 912 913 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 914 struct pagemapread *pm) 915 { 916 pm->buffer[pm->pos++] = *pme; 917 if (pm->pos >= pm->len) 918 return PM_END_OF_BUFFER; 919 return 0; 920 } 921 922 static int pagemap_pte_hole(unsigned long start, unsigned long end, 923 struct mm_walk *walk) 924 { 925 struct pagemapread *pm = walk->private; 926 unsigned long addr = start; 927 int err = 0; 928 929 while (addr < end) { 930 struct vm_area_struct *vma = find_vma(walk->mm, addr); 931 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 932 /* End of address space hole, which we mark as non-present. */ 933 unsigned long hole_end; 934 935 if (vma) 936 hole_end = min(end, vma->vm_start); 937 else 938 hole_end = end; 939 940 for (; addr < hole_end; addr += PAGE_SIZE) { 941 err = add_to_pagemap(addr, &pme, pm); 942 if (err) 943 goto out; 944 } 945 946 if (!vma) 947 break; 948 949 /* Addresses in the VMA. */ 950 if (vma->vm_flags & VM_SOFTDIRTY) 951 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY); 952 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 953 err = add_to_pagemap(addr, &pme, pm); 954 if (err) 955 goto out; 956 } 957 } 958 out: 959 return err; 960 } 961 962 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 963 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 964 { 965 u64 frame, flags; 966 struct page *page = NULL; 967 int flags2 = 0; 968 969 if (pte_present(pte)) { 970 frame = pte_pfn(pte); 971 flags = PM_PRESENT; 972 page = vm_normal_page(vma, addr, pte); 973 if (pte_soft_dirty(pte)) 974 flags2 |= __PM_SOFT_DIRTY; 975 } else if (is_swap_pte(pte)) { 976 swp_entry_t entry; 977 if (pte_swp_soft_dirty(pte)) 978 flags2 |= __PM_SOFT_DIRTY; 979 entry = pte_to_swp_entry(pte); 980 frame = swp_type(entry) | 981 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 982 flags = PM_SWAP; 983 if (is_migration_entry(entry)) 984 page = migration_entry_to_page(entry); 985 } else { 986 if (vma->vm_flags & VM_SOFTDIRTY) 987 flags2 |= __PM_SOFT_DIRTY; 988 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2)); 989 return; 990 } 991 992 if (page && !PageAnon(page)) 993 flags |= PM_FILE; 994 if ((vma->vm_flags & VM_SOFTDIRTY)) 995 flags2 |= __PM_SOFT_DIRTY; 996 997 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags); 998 } 999 1000 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1001 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1002 pmd_t pmd, int offset, int pmd_flags2) 1003 { 1004 /* 1005 * Currently pmd for thp is always present because thp can not be 1006 * swapped-out, migrated, or HWPOISONed (split in such cases instead.) 1007 * This if-check is just to prepare for future implementation. 1008 */ 1009 if (pmd_present(pmd)) 1010 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset) 1011 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT); 1012 else 1013 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2)); 1014 } 1015 #else 1016 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1017 pmd_t pmd, int offset, int pmd_flags2) 1018 { 1019 } 1020 #endif 1021 1022 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 1023 struct mm_walk *walk) 1024 { 1025 struct vm_area_struct *vma; 1026 struct pagemapread *pm = walk->private; 1027 spinlock_t *ptl; 1028 pte_t *pte; 1029 int err = 0; 1030 1031 /* find the first VMA at or above 'addr' */ 1032 vma = find_vma(walk->mm, addr); 1033 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 1034 int pmd_flags2; 1035 1036 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd)) 1037 pmd_flags2 = __PM_SOFT_DIRTY; 1038 else 1039 pmd_flags2 = 0; 1040 1041 for (; addr != end; addr += PAGE_SIZE) { 1042 unsigned long offset; 1043 pagemap_entry_t pme; 1044 1045 offset = (addr & ~PAGEMAP_WALK_MASK) >> 1046 PAGE_SHIFT; 1047 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2); 1048 err = add_to_pagemap(addr, &pme, pm); 1049 if (err) 1050 break; 1051 } 1052 spin_unlock(ptl); 1053 return err; 1054 } 1055 1056 if (pmd_trans_unstable(pmd)) 1057 return 0; 1058 1059 while (1) { 1060 /* End of address space hole, which we mark as non-present. */ 1061 unsigned long hole_end; 1062 1063 if (vma) 1064 hole_end = min(end, vma->vm_start); 1065 else 1066 hole_end = end; 1067 1068 for (; addr < hole_end; addr += PAGE_SIZE) { 1069 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 1070 1071 err = add_to_pagemap(addr, &pme, pm); 1072 if (err) 1073 return err; 1074 } 1075 1076 if (!vma || vma->vm_start >= end) 1077 break; 1078 /* 1079 * We can't possibly be in a hugetlb VMA. In general, 1080 * for a mm_walk with a pmd_entry and a hugetlb_entry, 1081 * the pmd_entry can only be called on addresses in a 1082 * hugetlb if the walk starts in a non-hugetlb VMA and 1083 * spans a hugepage VMA. Since pagemap_read walks are 1084 * PMD-sized and PMD-aligned, this will never be true. 1085 */ 1086 BUG_ON(is_vm_hugetlb_page(vma)); 1087 1088 /* Addresses in the VMA. */ 1089 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1090 pagemap_entry_t pme; 1091 pte = pte_offset_map(pmd, addr); 1092 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte); 1093 pte_unmap(pte); 1094 err = add_to_pagemap(addr, &pme, pm); 1095 if (err) 1096 return err; 1097 } 1098 1099 if (addr == end) 1100 break; 1101 1102 vma = find_vma(walk->mm, addr); 1103 } 1104 1105 cond_resched(); 1106 1107 return err; 1108 } 1109 1110 #ifdef CONFIG_HUGETLB_PAGE 1111 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1112 pte_t pte, int offset, int flags2) 1113 { 1114 if (pte_present(pte)) 1115 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) | 1116 PM_STATUS2(pm->v2, flags2) | 1117 PM_PRESENT); 1118 else 1119 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | 1120 PM_STATUS2(pm->v2, flags2)); 1121 } 1122 1123 /* This function walks within one hugetlb entry in the single call */ 1124 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 1125 unsigned long addr, unsigned long end, 1126 struct mm_walk *walk) 1127 { 1128 struct pagemapread *pm = walk->private; 1129 struct vm_area_struct *vma; 1130 int err = 0; 1131 int flags2; 1132 pagemap_entry_t pme; 1133 1134 vma = find_vma(walk->mm, addr); 1135 WARN_ON_ONCE(!vma); 1136 1137 if (vma && (vma->vm_flags & VM_SOFTDIRTY)) 1138 flags2 = __PM_SOFT_DIRTY; 1139 else 1140 flags2 = 0; 1141 1142 for (; addr != end; addr += PAGE_SIZE) { 1143 int offset = (addr & ~hmask) >> PAGE_SHIFT; 1144 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2); 1145 err = add_to_pagemap(addr, &pme, pm); 1146 if (err) 1147 return err; 1148 } 1149 1150 cond_resched(); 1151 1152 return err; 1153 } 1154 #endif /* HUGETLB_PAGE */ 1155 1156 /* 1157 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1158 * 1159 * For each page in the address space, this file contains one 64-bit entry 1160 * consisting of the following: 1161 * 1162 * Bits 0-54 page frame number (PFN) if present 1163 * Bits 0-4 swap type if swapped 1164 * Bits 5-54 swap offset if swapped 1165 * Bits 55-60 page shift (page size = 1<<page shift) 1166 * Bit 61 page is file-page or shared-anon 1167 * Bit 62 page swapped 1168 * Bit 63 page present 1169 * 1170 * If the page is not present but in swap, then the PFN contains an 1171 * encoding of the swap file number and the page's offset into the 1172 * swap. Unmapped pages return a null PFN. This allows determining 1173 * precisely which pages are mapped (or in swap) and comparing mapped 1174 * pages between processes. 1175 * 1176 * Efficient users of this interface will use /proc/pid/maps to 1177 * determine which areas of memory are actually mapped and llseek to 1178 * skip over unmapped regions. 1179 */ 1180 static ssize_t pagemap_read(struct file *file, char __user *buf, 1181 size_t count, loff_t *ppos) 1182 { 1183 struct task_struct *task = get_proc_task(file_inode(file)); 1184 struct mm_struct *mm; 1185 struct pagemapread pm; 1186 int ret = -ESRCH; 1187 struct mm_walk pagemap_walk = {}; 1188 unsigned long src; 1189 unsigned long svpfn; 1190 unsigned long start_vaddr; 1191 unsigned long end_vaddr; 1192 int copied = 0; 1193 1194 if (!task) 1195 goto out; 1196 1197 ret = -EINVAL; 1198 /* file position must be aligned */ 1199 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1200 goto out_task; 1201 1202 ret = 0; 1203 if (!count) 1204 goto out_task; 1205 1206 pm.v2 = soft_dirty_cleared; 1207 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1208 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); 1209 ret = -ENOMEM; 1210 if (!pm.buffer) 1211 goto out_task; 1212 1213 mm = mm_access(task, PTRACE_MODE_READ); 1214 ret = PTR_ERR(mm); 1215 if (!mm || IS_ERR(mm)) 1216 goto out_free; 1217 1218 pagemap_walk.pmd_entry = pagemap_pte_range; 1219 pagemap_walk.pte_hole = pagemap_pte_hole; 1220 #ifdef CONFIG_HUGETLB_PAGE 1221 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1222 #endif 1223 pagemap_walk.mm = mm; 1224 pagemap_walk.private = ± 1225 1226 src = *ppos; 1227 svpfn = src / PM_ENTRY_BYTES; 1228 start_vaddr = svpfn << PAGE_SHIFT; 1229 end_vaddr = TASK_SIZE_OF(task); 1230 1231 /* watch out for wraparound */ 1232 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 1233 start_vaddr = end_vaddr; 1234 1235 /* 1236 * The odds are that this will stop walking way 1237 * before end_vaddr, because the length of the 1238 * user buffer is tracked in "pm", and the walk 1239 * will stop when we hit the end of the buffer. 1240 */ 1241 ret = 0; 1242 while (count && (start_vaddr < end_vaddr)) { 1243 int len; 1244 unsigned long end; 1245 1246 pm.pos = 0; 1247 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1248 /* overflow ? */ 1249 if (end < start_vaddr || end > end_vaddr) 1250 end = end_vaddr; 1251 down_read(&mm->mmap_sem); 1252 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1253 up_read(&mm->mmap_sem); 1254 start_vaddr = end; 1255 1256 len = min(count, PM_ENTRY_BYTES * pm.pos); 1257 if (copy_to_user(buf, pm.buffer, len)) { 1258 ret = -EFAULT; 1259 goto out_mm; 1260 } 1261 copied += len; 1262 buf += len; 1263 count -= len; 1264 } 1265 *ppos += copied; 1266 if (!ret || ret == PM_END_OF_BUFFER) 1267 ret = copied; 1268 1269 out_mm: 1270 mmput(mm); 1271 out_free: 1272 kfree(pm.buffer); 1273 out_task: 1274 put_task_struct(task); 1275 out: 1276 return ret; 1277 } 1278 1279 static int pagemap_open(struct inode *inode, struct file *file) 1280 { 1281 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about " 1282 "to stop being page-shift some time soon. See the " 1283 "linux/Documentation/vm/pagemap.txt for details.\n"); 1284 return 0; 1285 } 1286 1287 const struct file_operations proc_pagemap_operations = { 1288 .llseek = mem_lseek, /* borrow this */ 1289 .read = pagemap_read, 1290 .open = pagemap_open, 1291 }; 1292 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1293 1294 #ifdef CONFIG_NUMA 1295 1296 struct numa_maps { 1297 struct vm_area_struct *vma; 1298 unsigned long pages; 1299 unsigned long anon; 1300 unsigned long active; 1301 unsigned long writeback; 1302 unsigned long mapcount_max; 1303 unsigned long dirty; 1304 unsigned long swapcache; 1305 unsigned long node[MAX_NUMNODES]; 1306 }; 1307 1308 struct numa_maps_private { 1309 struct proc_maps_private proc_maps; 1310 struct numa_maps md; 1311 }; 1312 1313 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1314 unsigned long nr_pages) 1315 { 1316 int count = page_mapcount(page); 1317 1318 md->pages += nr_pages; 1319 if (pte_dirty || PageDirty(page)) 1320 md->dirty += nr_pages; 1321 1322 if (PageSwapCache(page)) 1323 md->swapcache += nr_pages; 1324 1325 if (PageActive(page) || PageUnevictable(page)) 1326 md->active += nr_pages; 1327 1328 if (PageWriteback(page)) 1329 md->writeback += nr_pages; 1330 1331 if (PageAnon(page)) 1332 md->anon += nr_pages; 1333 1334 if (count > md->mapcount_max) 1335 md->mapcount_max = count; 1336 1337 md->node[page_to_nid(page)] += nr_pages; 1338 } 1339 1340 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1341 unsigned long addr) 1342 { 1343 struct page *page; 1344 int nid; 1345 1346 if (!pte_present(pte)) 1347 return NULL; 1348 1349 page = vm_normal_page(vma, addr, pte); 1350 if (!page) 1351 return NULL; 1352 1353 if (PageReserved(page)) 1354 return NULL; 1355 1356 nid = page_to_nid(page); 1357 if (!node_isset(nid, node_states[N_MEMORY])) 1358 return NULL; 1359 1360 return page; 1361 } 1362 1363 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1364 unsigned long end, struct mm_walk *walk) 1365 { 1366 struct numa_maps *md; 1367 spinlock_t *ptl; 1368 pte_t *orig_pte; 1369 pte_t *pte; 1370 1371 md = walk->private; 1372 1373 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) { 1374 pte_t huge_pte = *(pte_t *)pmd; 1375 struct page *page; 1376 1377 page = can_gather_numa_stats(huge_pte, md->vma, addr); 1378 if (page) 1379 gather_stats(page, md, pte_dirty(huge_pte), 1380 HPAGE_PMD_SIZE/PAGE_SIZE); 1381 spin_unlock(ptl); 1382 return 0; 1383 } 1384 1385 if (pmd_trans_unstable(pmd)) 1386 return 0; 1387 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1388 do { 1389 struct page *page = can_gather_numa_stats(*pte, md->vma, addr); 1390 if (!page) 1391 continue; 1392 gather_stats(page, md, pte_dirty(*pte), 1); 1393 1394 } while (pte++, addr += PAGE_SIZE, addr != end); 1395 pte_unmap_unlock(orig_pte, ptl); 1396 return 0; 1397 } 1398 #ifdef CONFIG_HUGETLB_PAGE 1399 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1400 unsigned long addr, unsigned long end, struct mm_walk *walk) 1401 { 1402 struct numa_maps *md; 1403 struct page *page; 1404 1405 if (!pte_present(*pte)) 1406 return 0; 1407 1408 page = pte_page(*pte); 1409 if (!page) 1410 return 0; 1411 1412 md = walk->private; 1413 gather_stats(page, md, pte_dirty(*pte), 1); 1414 return 0; 1415 } 1416 1417 #else 1418 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1419 unsigned long addr, unsigned long end, struct mm_walk *walk) 1420 { 1421 return 0; 1422 } 1423 #endif 1424 1425 /* 1426 * Display pages allocated per node and memory policy via /proc. 1427 */ 1428 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1429 { 1430 struct numa_maps_private *numa_priv = m->private; 1431 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1432 struct vm_area_struct *vma = v; 1433 struct numa_maps *md = &numa_priv->md; 1434 struct file *file = vma->vm_file; 1435 struct mm_struct *mm = vma->vm_mm; 1436 struct mm_walk walk = {}; 1437 struct mempolicy *pol; 1438 char buffer[64]; 1439 int nid; 1440 1441 if (!mm) 1442 return 0; 1443 1444 /* Ensure we start with an empty set of numa_maps statistics. */ 1445 memset(md, 0, sizeof(*md)); 1446 1447 md->vma = vma; 1448 1449 walk.hugetlb_entry = gather_hugetbl_stats; 1450 walk.pmd_entry = gather_pte_stats; 1451 walk.private = md; 1452 walk.mm = mm; 1453 1454 pol = __get_vma_policy(vma, vma->vm_start); 1455 if (pol) { 1456 mpol_to_str(buffer, sizeof(buffer), pol); 1457 mpol_cond_put(pol); 1458 } else { 1459 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1460 } 1461 1462 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1463 1464 if (file) { 1465 seq_puts(m, " file="); 1466 seq_path(m, &file->f_path, "\n\t= "); 1467 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1468 seq_puts(m, " heap"); 1469 } else { 1470 pid_t tid = pid_of_stack(proc_priv, vma, is_pid); 1471 if (tid != 0) { 1472 /* 1473 * Thread stack in /proc/PID/task/TID/maps or 1474 * the main process stack. 1475 */ 1476 if (!is_pid || (vma->vm_start <= mm->start_stack && 1477 vma->vm_end >= mm->start_stack)) 1478 seq_puts(m, " stack"); 1479 else 1480 seq_printf(m, " stack:%d", tid); 1481 } 1482 } 1483 1484 if (is_vm_hugetlb_page(vma)) 1485 seq_puts(m, " huge"); 1486 1487 walk_page_range(vma->vm_start, vma->vm_end, &walk); 1488 1489 if (!md->pages) 1490 goto out; 1491 1492 if (md->anon) 1493 seq_printf(m, " anon=%lu", md->anon); 1494 1495 if (md->dirty) 1496 seq_printf(m, " dirty=%lu", md->dirty); 1497 1498 if (md->pages != md->anon && md->pages != md->dirty) 1499 seq_printf(m, " mapped=%lu", md->pages); 1500 1501 if (md->mapcount_max > 1) 1502 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1503 1504 if (md->swapcache) 1505 seq_printf(m, " swapcache=%lu", md->swapcache); 1506 1507 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1508 seq_printf(m, " active=%lu", md->active); 1509 1510 if (md->writeback) 1511 seq_printf(m, " writeback=%lu", md->writeback); 1512 1513 for_each_node_state(nid, N_MEMORY) 1514 if (md->node[nid]) 1515 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1516 out: 1517 seq_putc(m, '\n'); 1518 m_cache_vma(m, vma); 1519 return 0; 1520 } 1521 1522 static int show_pid_numa_map(struct seq_file *m, void *v) 1523 { 1524 return show_numa_map(m, v, 1); 1525 } 1526 1527 static int show_tid_numa_map(struct seq_file *m, void *v) 1528 { 1529 return show_numa_map(m, v, 0); 1530 } 1531 1532 static const struct seq_operations proc_pid_numa_maps_op = { 1533 .start = m_start, 1534 .next = m_next, 1535 .stop = m_stop, 1536 .show = show_pid_numa_map, 1537 }; 1538 1539 static const struct seq_operations proc_tid_numa_maps_op = { 1540 .start = m_start, 1541 .next = m_next, 1542 .stop = m_stop, 1543 .show = show_tid_numa_map, 1544 }; 1545 1546 static int numa_maps_open(struct inode *inode, struct file *file, 1547 const struct seq_operations *ops) 1548 { 1549 return proc_maps_open(inode, file, ops, 1550 sizeof(struct numa_maps_private)); 1551 } 1552 1553 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1554 { 1555 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1556 } 1557 1558 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1559 { 1560 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1561 } 1562 1563 const struct file_operations proc_pid_numa_maps_operations = { 1564 .open = pid_numa_maps_open, 1565 .read = seq_read, 1566 .llseek = seq_lseek, 1567 .release = proc_map_release, 1568 }; 1569 1570 const struct file_operations proc_tid_numa_maps_operations = { 1571 .open = tid_numa_maps_open, 1572 .read = seq_read, 1573 .llseek = seq_lseek, 1574 .release = proc_map_release, 1575 }; 1576 #endif /* CONFIG_NUMA */ 1577