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, ptes, pmds; 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 ptes = PTRS_PER_PTE * sizeof(pte_t) * atomic_long_read(&mm->nr_ptes); 46 pmds = PTRS_PER_PMD * sizeof(pmd_t) * mm_nr_pmds(mm); 47 seq_printf(m, 48 "VmPeak:\t%8lu kB\n" 49 "VmSize:\t%8lu kB\n" 50 "VmLck:\t%8lu kB\n" 51 "VmPin:\t%8lu kB\n" 52 "VmHWM:\t%8lu kB\n" 53 "VmRSS:\t%8lu kB\n" 54 "VmData:\t%8lu kB\n" 55 "VmStk:\t%8lu kB\n" 56 "VmExe:\t%8lu kB\n" 57 "VmLib:\t%8lu kB\n" 58 "VmPTE:\t%8lu kB\n" 59 "VmPMD:\t%8lu kB\n" 60 "VmSwap:\t%8lu kB\n", 61 hiwater_vm << (PAGE_SHIFT-10), 62 total_vm << (PAGE_SHIFT-10), 63 mm->locked_vm << (PAGE_SHIFT-10), 64 mm->pinned_vm << (PAGE_SHIFT-10), 65 hiwater_rss << (PAGE_SHIFT-10), 66 total_rss << (PAGE_SHIFT-10), 67 data << (PAGE_SHIFT-10), 68 mm->stack_vm << (PAGE_SHIFT-10), text, lib, 69 ptes >> 10, 70 pmds >> 10, 71 swap << (PAGE_SHIFT-10)); 72 } 73 74 unsigned long task_vsize(struct mm_struct *mm) 75 { 76 return PAGE_SIZE * mm->total_vm; 77 } 78 79 unsigned long task_statm(struct mm_struct *mm, 80 unsigned long *shared, unsigned long *text, 81 unsigned long *data, unsigned long *resident) 82 { 83 *shared = get_mm_counter(mm, MM_FILEPAGES); 84 *text = (PAGE_ALIGN(mm->end_code) - (mm->start_code & PAGE_MASK)) 85 >> PAGE_SHIFT; 86 *data = mm->total_vm - mm->shared_vm; 87 *resident = *shared + get_mm_counter(mm, MM_ANONPAGES); 88 return mm->total_vm; 89 } 90 91 #ifdef CONFIG_NUMA 92 /* 93 * Save get_task_policy() for show_numa_map(). 94 */ 95 static void hold_task_mempolicy(struct proc_maps_private *priv) 96 { 97 struct task_struct *task = priv->task; 98 99 task_lock(task); 100 priv->task_mempolicy = get_task_policy(task); 101 mpol_get(priv->task_mempolicy); 102 task_unlock(task); 103 } 104 static void release_task_mempolicy(struct proc_maps_private *priv) 105 { 106 mpol_put(priv->task_mempolicy); 107 } 108 #else 109 static void hold_task_mempolicy(struct proc_maps_private *priv) 110 { 111 } 112 static void release_task_mempolicy(struct proc_maps_private *priv) 113 { 114 } 115 #endif 116 117 static void vma_stop(struct proc_maps_private *priv) 118 { 119 struct mm_struct *mm = priv->mm; 120 121 release_task_mempolicy(priv); 122 up_read(&mm->mmap_sem); 123 mmput(mm); 124 } 125 126 static struct vm_area_struct * 127 m_next_vma(struct proc_maps_private *priv, struct vm_area_struct *vma) 128 { 129 if (vma == priv->tail_vma) 130 return NULL; 131 return vma->vm_next ?: priv->tail_vma; 132 } 133 134 static void m_cache_vma(struct seq_file *m, struct vm_area_struct *vma) 135 { 136 if (m->count < m->size) /* vma is copied successfully */ 137 m->version = m_next_vma(m->private, vma) ? vma->vm_start : -1UL; 138 } 139 140 static void *m_start(struct seq_file *m, loff_t *ppos) 141 { 142 struct proc_maps_private *priv = m->private; 143 unsigned long last_addr = m->version; 144 struct mm_struct *mm; 145 struct vm_area_struct *vma; 146 unsigned int pos = *ppos; 147 148 /* See m_cache_vma(). Zero at the start or after lseek. */ 149 if (last_addr == -1UL) 150 return NULL; 151 152 priv->task = get_proc_task(priv->inode); 153 if (!priv->task) 154 return ERR_PTR(-ESRCH); 155 156 mm = priv->mm; 157 if (!mm || !atomic_inc_not_zero(&mm->mm_users)) 158 return NULL; 159 160 down_read(&mm->mmap_sem); 161 hold_task_mempolicy(priv); 162 priv->tail_vma = get_gate_vma(mm); 163 164 if (last_addr) { 165 vma = find_vma(mm, last_addr); 166 if (vma && (vma = m_next_vma(priv, vma))) 167 return vma; 168 } 169 170 m->version = 0; 171 if (pos < mm->map_count) { 172 for (vma = mm->mmap; pos; pos--) { 173 m->version = vma->vm_start; 174 vma = vma->vm_next; 175 } 176 return vma; 177 } 178 179 /* we do not bother to update m->version in this case */ 180 if (pos == mm->map_count && priv->tail_vma) 181 return priv->tail_vma; 182 183 vma_stop(priv); 184 return NULL; 185 } 186 187 static void *m_next(struct seq_file *m, void *v, loff_t *pos) 188 { 189 struct proc_maps_private *priv = m->private; 190 struct vm_area_struct *next; 191 192 (*pos)++; 193 next = m_next_vma(priv, v); 194 if (!next) 195 vma_stop(priv); 196 return next; 197 } 198 199 static void m_stop(struct seq_file *m, void *v) 200 { 201 struct proc_maps_private *priv = m->private; 202 203 if (!IS_ERR_OR_NULL(v)) 204 vma_stop(priv); 205 if (priv->task) { 206 put_task_struct(priv->task); 207 priv->task = NULL; 208 } 209 } 210 211 static int proc_maps_open(struct inode *inode, struct file *file, 212 const struct seq_operations *ops, int psize) 213 { 214 struct proc_maps_private *priv = __seq_open_private(file, ops, psize); 215 216 if (!priv) 217 return -ENOMEM; 218 219 priv->inode = inode; 220 priv->mm = proc_mem_open(inode, PTRACE_MODE_READ); 221 if (IS_ERR(priv->mm)) { 222 int err = PTR_ERR(priv->mm); 223 224 seq_release_private(inode, file); 225 return err; 226 } 227 228 return 0; 229 } 230 231 static int proc_map_release(struct inode *inode, struct file *file) 232 { 233 struct seq_file *seq = file->private_data; 234 struct proc_maps_private *priv = seq->private; 235 236 if (priv->mm) 237 mmdrop(priv->mm); 238 239 return seq_release_private(inode, file); 240 } 241 242 static int do_maps_open(struct inode *inode, struct file *file, 243 const struct seq_operations *ops) 244 { 245 return proc_maps_open(inode, file, ops, 246 sizeof(struct proc_maps_private)); 247 } 248 249 static pid_t pid_of_stack(struct proc_maps_private *priv, 250 struct vm_area_struct *vma, bool is_pid) 251 { 252 struct inode *inode = priv->inode; 253 struct task_struct *task; 254 pid_t ret = 0; 255 256 rcu_read_lock(); 257 task = pid_task(proc_pid(inode), PIDTYPE_PID); 258 if (task) { 259 task = task_of_stack(task, vma, is_pid); 260 if (task) 261 ret = task_pid_nr_ns(task, inode->i_sb->s_fs_info); 262 } 263 rcu_read_unlock(); 264 265 return ret; 266 } 267 268 static void 269 show_map_vma(struct seq_file *m, struct vm_area_struct *vma, int is_pid) 270 { 271 struct mm_struct *mm = vma->vm_mm; 272 struct file *file = vma->vm_file; 273 struct proc_maps_private *priv = m->private; 274 vm_flags_t flags = vma->vm_flags; 275 unsigned long ino = 0; 276 unsigned long long pgoff = 0; 277 unsigned long start, end; 278 dev_t dev = 0; 279 const char *name = NULL; 280 281 if (file) { 282 struct inode *inode = file_inode(vma->vm_file); 283 dev = inode->i_sb->s_dev; 284 ino = inode->i_ino; 285 pgoff = ((loff_t)vma->vm_pgoff) << PAGE_SHIFT; 286 } 287 288 /* We don't show the stack guard page in /proc/maps */ 289 start = vma->vm_start; 290 if (stack_guard_page_start(vma, start)) 291 start += PAGE_SIZE; 292 end = vma->vm_end; 293 if (stack_guard_page_end(vma, end)) 294 end -= PAGE_SIZE; 295 296 seq_setwidth(m, 25 + sizeof(void *) * 6 - 1); 297 seq_printf(m, "%08lx-%08lx %c%c%c%c %08llx %02x:%02x %lu ", 298 start, 299 end, 300 flags & VM_READ ? 'r' : '-', 301 flags & VM_WRITE ? 'w' : '-', 302 flags & VM_EXEC ? 'x' : '-', 303 flags & VM_MAYSHARE ? 's' : 'p', 304 pgoff, 305 MAJOR(dev), MINOR(dev), ino); 306 307 /* 308 * Print the dentry name for named mappings, and a 309 * special [heap] marker for the heap: 310 */ 311 if (file) { 312 seq_pad(m, ' '); 313 seq_path(m, &file->f_path, "\n"); 314 goto done; 315 } 316 317 if (vma->vm_ops && vma->vm_ops->name) { 318 name = vma->vm_ops->name(vma); 319 if (name) 320 goto done; 321 } 322 323 name = arch_vma_name(vma); 324 if (!name) { 325 pid_t tid; 326 327 if (!mm) { 328 name = "[vdso]"; 329 goto done; 330 } 331 332 if (vma->vm_start <= mm->brk && 333 vma->vm_end >= mm->start_brk) { 334 name = "[heap]"; 335 goto done; 336 } 337 338 tid = pid_of_stack(priv, vma, is_pid); 339 if (tid != 0) { 340 /* 341 * Thread stack in /proc/PID/task/TID/maps or 342 * the main process stack. 343 */ 344 if (!is_pid || (vma->vm_start <= mm->start_stack && 345 vma->vm_end >= mm->start_stack)) { 346 name = "[stack]"; 347 } else { 348 /* Thread stack in /proc/PID/maps */ 349 seq_pad(m, ' '); 350 seq_printf(m, "[stack:%d]", tid); 351 } 352 } 353 } 354 355 done: 356 if (name) { 357 seq_pad(m, ' '); 358 seq_puts(m, name); 359 } 360 seq_putc(m, '\n'); 361 } 362 363 static int show_map(struct seq_file *m, void *v, int is_pid) 364 { 365 show_map_vma(m, v, is_pid); 366 m_cache_vma(m, v); 367 return 0; 368 } 369 370 static int show_pid_map(struct seq_file *m, void *v) 371 { 372 return show_map(m, v, 1); 373 } 374 375 static int show_tid_map(struct seq_file *m, void *v) 376 { 377 return show_map(m, v, 0); 378 } 379 380 static const struct seq_operations proc_pid_maps_op = { 381 .start = m_start, 382 .next = m_next, 383 .stop = m_stop, 384 .show = show_pid_map 385 }; 386 387 static const struct seq_operations proc_tid_maps_op = { 388 .start = m_start, 389 .next = m_next, 390 .stop = m_stop, 391 .show = show_tid_map 392 }; 393 394 static int pid_maps_open(struct inode *inode, struct file *file) 395 { 396 return do_maps_open(inode, file, &proc_pid_maps_op); 397 } 398 399 static int tid_maps_open(struct inode *inode, struct file *file) 400 { 401 return do_maps_open(inode, file, &proc_tid_maps_op); 402 } 403 404 const struct file_operations proc_pid_maps_operations = { 405 .open = pid_maps_open, 406 .read = seq_read, 407 .llseek = seq_lseek, 408 .release = proc_map_release, 409 }; 410 411 const struct file_operations proc_tid_maps_operations = { 412 .open = tid_maps_open, 413 .read = seq_read, 414 .llseek = seq_lseek, 415 .release = proc_map_release, 416 }; 417 418 /* 419 * Proportional Set Size(PSS): my share of RSS. 420 * 421 * PSS of a process is the count of pages it has in memory, where each 422 * page is divided by the number of processes sharing it. So if a 423 * process has 1000 pages all to itself, and 1000 shared with one other 424 * process, its PSS will be 1500. 425 * 426 * To keep (accumulated) division errors low, we adopt a 64bit 427 * fixed-point pss counter to minimize division errors. So (pss >> 428 * PSS_SHIFT) would be the real byte count. 429 * 430 * A shift of 12 before division means (assuming 4K page size): 431 * - 1M 3-user-pages add up to 8KB errors; 432 * - supports mapcount up to 2^24, or 16M; 433 * - supports PSS up to 2^52 bytes, or 4PB. 434 */ 435 #define PSS_SHIFT 12 436 437 #ifdef CONFIG_PROC_PAGE_MONITOR 438 struct mem_size_stats { 439 unsigned long resident; 440 unsigned long shared_clean; 441 unsigned long shared_dirty; 442 unsigned long private_clean; 443 unsigned long private_dirty; 444 unsigned long referenced; 445 unsigned long anonymous; 446 unsigned long anonymous_thp; 447 unsigned long swap; 448 u64 pss; 449 }; 450 451 static void smaps_account(struct mem_size_stats *mss, struct page *page, 452 unsigned long size, bool young, bool dirty) 453 { 454 int mapcount; 455 456 if (PageAnon(page)) 457 mss->anonymous += size; 458 459 mss->resident += size; 460 /* Accumulate the size in pages that have been accessed. */ 461 if (young || PageReferenced(page)) 462 mss->referenced += size; 463 mapcount = page_mapcount(page); 464 if (mapcount >= 2) { 465 u64 pss_delta; 466 467 if (dirty || PageDirty(page)) 468 mss->shared_dirty += size; 469 else 470 mss->shared_clean += size; 471 pss_delta = (u64)size << PSS_SHIFT; 472 do_div(pss_delta, mapcount); 473 mss->pss += pss_delta; 474 } else { 475 if (dirty || PageDirty(page)) 476 mss->private_dirty += size; 477 else 478 mss->private_clean += size; 479 mss->pss += (u64)size << PSS_SHIFT; 480 } 481 } 482 483 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 484 struct mm_walk *walk) 485 { 486 struct mem_size_stats *mss = walk->private; 487 struct vm_area_struct *vma = walk->vma; 488 struct page *page = NULL; 489 490 if (pte_present(*pte)) { 491 page = vm_normal_page(vma, addr, *pte); 492 } else if (is_swap_pte(*pte)) { 493 swp_entry_t swpent = pte_to_swp_entry(*pte); 494 495 if (!non_swap_entry(swpent)) 496 mss->swap += PAGE_SIZE; 497 else if (is_migration_entry(swpent)) 498 page = migration_entry_to_page(swpent); 499 } 500 501 if (!page) 502 return; 503 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte)); 504 } 505 506 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 507 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 508 struct mm_walk *walk) 509 { 510 struct mem_size_stats *mss = walk->private; 511 struct vm_area_struct *vma = walk->vma; 512 struct page *page; 513 514 /* FOLL_DUMP will return -EFAULT on huge zero page */ 515 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 516 if (IS_ERR_OR_NULL(page)) 517 return; 518 mss->anonymous_thp += HPAGE_PMD_SIZE; 519 smaps_account(mss, page, HPAGE_PMD_SIZE, 520 pmd_young(*pmd), pmd_dirty(*pmd)); 521 } 522 #else 523 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 524 struct mm_walk *walk) 525 { 526 } 527 #endif 528 529 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 530 struct mm_walk *walk) 531 { 532 struct vm_area_struct *vma = walk->vma; 533 pte_t *pte; 534 spinlock_t *ptl; 535 536 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 537 smaps_pmd_entry(pmd, addr, walk); 538 spin_unlock(ptl); 539 return 0; 540 } 541 542 if (pmd_trans_unstable(pmd)) 543 return 0; 544 /* 545 * The mmap_sem held all the way back in m_start() is what 546 * keeps khugepaged out of here and from collapsing things 547 * in here. 548 */ 549 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 550 for (; addr != end; pte++, addr += PAGE_SIZE) 551 smaps_pte_entry(pte, addr, walk); 552 pte_unmap_unlock(pte - 1, ptl); 553 cond_resched(); 554 return 0; 555 } 556 557 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 558 { 559 /* 560 * Don't forget to update Documentation/ on changes. 561 */ 562 static const char mnemonics[BITS_PER_LONG][2] = { 563 /* 564 * In case if we meet a flag we don't know about. 565 */ 566 [0 ... (BITS_PER_LONG-1)] = "??", 567 568 [ilog2(VM_READ)] = "rd", 569 [ilog2(VM_WRITE)] = "wr", 570 [ilog2(VM_EXEC)] = "ex", 571 [ilog2(VM_SHARED)] = "sh", 572 [ilog2(VM_MAYREAD)] = "mr", 573 [ilog2(VM_MAYWRITE)] = "mw", 574 [ilog2(VM_MAYEXEC)] = "me", 575 [ilog2(VM_MAYSHARE)] = "ms", 576 [ilog2(VM_GROWSDOWN)] = "gd", 577 [ilog2(VM_PFNMAP)] = "pf", 578 [ilog2(VM_DENYWRITE)] = "dw", 579 #ifdef CONFIG_X86_INTEL_MPX 580 [ilog2(VM_MPX)] = "mp", 581 #endif 582 [ilog2(VM_LOCKED)] = "lo", 583 [ilog2(VM_IO)] = "io", 584 [ilog2(VM_SEQ_READ)] = "sr", 585 [ilog2(VM_RAND_READ)] = "rr", 586 [ilog2(VM_DONTCOPY)] = "dc", 587 [ilog2(VM_DONTEXPAND)] = "de", 588 [ilog2(VM_ACCOUNT)] = "ac", 589 [ilog2(VM_NORESERVE)] = "nr", 590 [ilog2(VM_HUGETLB)] = "ht", 591 [ilog2(VM_ARCH_1)] = "ar", 592 [ilog2(VM_DONTDUMP)] = "dd", 593 #ifdef CONFIG_MEM_SOFT_DIRTY 594 [ilog2(VM_SOFTDIRTY)] = "sd", 595 #endif 596 [ilog2(VM_MIXEDMAP)] = "mm", 597 [ilog2(VM_HUGEPAGE)] = "hg", 598 [ilog2(VM_NOHUGEPAGE)] = "nh", 599 [ilog2(VM_MERGEABLE)] = "mg", 600 }; 601 size_t i; 602 603 seq_puts(m, "VmFlags: "); 604 for (i = 0; i < BITS_PER_LONG; i++) { 605 if (vma->vm_flags & (1UL << i)) { 606 seq_printf(m, "%c%c ", 607 mnemonics[i][0], mnemonics[i][1]); 608 } 609 } 610 seq_putc(m, '\n'); 611 } 612 613 static int show_smap(struct seq_file *m, void *v, int is_pid) 614 { 615 struct vm_area_struct *vma = v; 616 struct mem_size_stats mss; 617 struct mm_walk smaps_walk = { 618 .pmd_entry = smaps_pte_range, 619 .mm = vma->vm_mm, 620 .private = &mss, 621 }; 622 623 memset(&mss, 0, sizeof mss); 624 /* mmap_sem is held in m_start */ 625 walk_page_vma(vma, &smaps_walk); 626 627 show_map_vma(m, vma, is_pid); 628 629 seq_printf(m, 630 "Size: %8lu kB\n" 631 "Rss: %8lu kB\n" 632 "Pss: %8lu kB\n" 633 "Shared_Clean: %8lu kB\n" 634 "Shared_Dirty: %8lu kB\n" 635 "Private_Clean: %8lu kB\n" 636 "Private_Dirty: %8lu kB\n" 637 "Referenced: %8lu kB\n" 638 "Anonymous: %8lu kB\n" 639 "AnonHugePages: %8lu kB\n" 640 "Swap: %8lu kB\n" 641 "KernelPageSize: %8lu kB\n" 642 "MMUPageSize: %8lu kB\n" 643 "Locked: %8lu kB\n", 644 (vma->vm_end - vma->vm_start) >> 10, 645 mss.resident >> 10, 646 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 647 mss.shared_clean >> 10, 648 mss.shared_dirty >> 10, 649 mss.private_clean >> 10, 650 mss.private_dirty >> 10, 651 mss.referenced >> 10, 652 mss.anonymous >> 10, 653 mss.anonymous_thp >> 10, 654 mss.swap >> 10, 655 vma_kernel_pagesize(vma) >> 10, 656 vma_mmu_pagesize(vma) >> 10, 657 (vma->vm_flags & VM_LOCKED) ? 658 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 659 660 show_smap_vma_flags(m, vma); 661 m_cache_vma(m, vma); 662 return 0; 663 } 664 665 static int show_pid_smap(struct seq_file *m, void *v) 666 { 667 return show_smap(m, v, 1); 668 } 669 670 static int show_tid_smap(struct seq_file *m, void *v) 671 { 672 return show_smap(m, v, 0); 673 } 674 675 static const struct seq_operations proc_pid_smaps_op = { 676 .start = m_start, 677 .next = m_next, 678 .stop = m_stop, 679 .show = show_pid_smap 680 }; 681 682 static const struct seq_operations proc_tid_smaps_op = { 683 .start = m_start, 684 .next = m_next, 685 .stop = m_stop, 686 .show = show_tid_smap 687 }; 688 689 static int pid_smaps_open(struct inode *inode, struct file *file) 690 { 691 return do_maps_open(inode, file, &proc_pid_smaps_op); 692 } 693 694 static int tid_smaps_open(struct inode *inode, struct file *file) 695 { 696 return do_maps_open(inode, file, &proc_tid_smaps_op); 697 } 698 699 const struct file_operations proc_pid_smaps_operations = { 700 .open = pid_smaps_open, 701 .read = seq_read, 702 .llseek = seq_lseek, 703 .release = proc_map_release, 704 }; 705 706 const struct file_operations proc_tid_smaps_operations = { 707 .open = tid_smaps_open, 708 .read = seq_read, 709 .llseek = seq_lseek, 710 .release = proc_map_release, 711 }; 712 713 /* 714 * We do not want to have constant page-shift bits sitting in 715 * pagemap entries and are about to reuse them some time soon. 716 * 717 * Here's the "migration strategy": 718 * 1. when the system boots these bits remain what they are, 719 * but a warning about future change is printed in log; 720 * 2. once anyone clears soft-dirty bits via clear_refs file, 721 * these flag is set to denote, that user is aware of the 722 * new API and those page-shift bits change their meaning. 723 * The respective warning is printed in dmesg; 724 * 3. In a couple of releases we will remove all the mentions 725 * of page-shift in pagemap entries. 726 */ 727 728 static bool soft_dirty_cleared __read_mostly; 729 730 enum clear_refs_types { 731 CLEAR_REFS_ALL = 1, 732 CLEAR_REFS_ANON, 733 CLEAR_REFS_MAPPED, 734 CLEAR_REFS_SOFT_DIRTY, 735 CLEAR_REFS_LAST, 736 }; 737 738 struct clear_refs_private { 739 struct vm_area_struct *vma; 740 enum clear_refs_types type; 741 }; 742 743 static inline void clear_soft_dirty(struct vm_area_struct *vma, 744 unsigned long addr, pte_t *pte) 745 { 746 #ifdef CONFIG_MEM_SOFT_DIRTY 747 /* 748 * The soft-dirty tracker uses #PF-s to catch writes 749 * to pages, so write-protect the pte as well. See the 750 * Documentation/vm/soft-dirty.txt for full description 751 * of how soft-dirty works. 752 */ 753 pte_t ptent = *pte; 754 755 if (pte_present(ptent)) { 756 ptent = pte_wrprotect(ptent); 757 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY); 758 } else if (is_swap_pte(ptent)) { 759 ptent = pte_swp_clear_soft_dirty(ptent); 760 } 761 762 set_pte_at(vma->vm_mm, addr, pte, ptent); 763 #endif 764 } 765 766 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 767 unsigned long end, struct mm_walk *walk) 768 { 769 struct clear_refs_private *cp = walk->private; 770 struct vm_area_struct *vma = cp->vma; 771 pte_t *pte, ptent; 772 spinlock_t *ptl; 773 struct page *page; 774 775 split_huge_page_pmd(vma, addr, pmd); 776 if (pmd_trans_unstable(pmd)) 777 return 0; 778 779 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 780 for (; addr != end; pte++, addr += PAGE_SIZE) { 781 ptent = *pte; 782 783 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 784 clear_soft_dirty(vma, addr, pte); 785 continue; 786 } 787 788 if (!pte_present(ptent)) 789 continue; 790 791 page = vm_normal_page(vma, addr, ptent); 792 if (!page) 793 continue; 794 795 /* Clear accessed and referenced bits. */ 796 ptep_test_and_clear_young(vma, addr, pte); 797 ClearPageReferenced(page); 798 } 799 pte_unmap_unlock(pte - 1, ptl); 800 cond_resched(); 801 return 0; 802 } 803 804 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 805 size_t count, loff_t *ppos) 806 { 807 struct task_struct *task; 808 char buffer[PROC_NUMBUF]; 809 struct mm_struct *mm; 810 struct vm_area_struct *vma; 811 enum clear_refs_types type; 812 int itype; 813 int rv; 814 815 memset(buffer, 0, sizeof(buffer)); 816 if (count > sizeof(buffer) - 1) 817 count = sizeof(buffer) - 1; 818 if (copy_from_user(buffer, buf, count)) 819 return -EFAULT; 820 rv = kstrtoint(strstrip(buffer), 10, &itype); 821 if (rv < 0) 822 return rv; 823 type = (enum clear_refs_types)itype; 824 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 825 return -EINVAL; 826 827 if (type == CLEAR_REFS_SOFT_DIRTY) { 828 soft_dirty_cleared = true; 829 pr_warn_once("The pagemap bits 55-60 has changed their meaning!" 830 " See the linux/Documentation/vm/pagemap.txt for " 831 "details.\n"); 832 } 833 834 task = get_proc_task(file_inode(file)); 835 if (!task) 836 return -ESRCH; 837 mm = get_task_mm(task); 838 if (mm) { 839 struct clear_refs_private cp = { 840 .type = type, 841 }; 842 struct mm_walk clear_refs_walk = { 843 .pmd_entry = clear_refs_pte_range, 844 .mm = mm, 845 .private = &cp, 846 }; 847 down_read(&mm->mmap_sem); 848 if (type == CLEAR_REFS_SOFT_DIRTY) { 849 for (vma = mm->mmap; vma; vma = vma->vm_next) { 850 if (!(vma->vm_flags & VM_SOFTDIRTY)) 851 continue; 852 up_read(&mm->mmap_sem); 853 down_write(&mm->mmap_sem); 854 for (vma = mm->mmap; vma; vma = vma->vm_next) { 855 vma->vm_flags &= ~VM_SOFTDIRTY; 856 vma_set_page_prot(vma); 857 } 858 downgrade_write(&mm->mmap_sem); 859 break; 860 } 861 mmu_notifier_invalidate_range_start(mm, 0, -1); 862 } 863 for (vma = mm->mmap; vma; vma = vma->vm_next) { 864 cp.vma = vma; 865 if (is_vm_hugetlb_page(vma)) 866 continue; 867 /* 868 * Writing 1 to /proc/pid/clear_refs affects all pages. 869 * 870 * Writing 2 to /proc/pid/clear_refs only affects 871 * Anonymous pages. 872 * 873 * Writing 3 to /proc/pid/clear_refs only affects file 874 * mapped pages. 875 * 876 * Writing 4 to /proc/pid/clear_refs affects all pages. 877 */ 878 if (type == CLEAR_REFS_ANON && vma->vm_file) 879 continue; 880 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 881 continue; 882 walk_page_range(vma->vm_start, vma->vm_end, 883 &clear_refs_walk); 884 } 885 if (type == CLEAR_REFS_SOFT_DIRTY) 886 mmu_notifier_invalidate_range_end(mm, 0, -1); 887 flush_tlb_mm(mm); 888 up_read(&mm->mmap_sem); 889 mmput(mm); 890 } 891 put_task_struct(task); 892 893 return count; 894 } 895 896 const struct file_operations proc_clear_refs_operations = { 897 .write = clear_refs_write, 898 .llseek = noop_llseek, 899 }; 900 901 typedef struct { 902 u64 pme; 903 } pagemap_entry_t; 904 905 struct pagemapread { 906 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 907 pagemap_entry_t *buffer; 908 bool v2; 909 }; 910 911 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 912 #define PAGEMAP_WALK_MASK (PMD_MASK) 913 914 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 915 #define PM_STATUS_BITS 3 916 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 917 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 918 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 919 #define PM_PSHIFT_BITS 6 920 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 921 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 922 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 923 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 924 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 925 /* in "new" pagemap pshift bits are occupied with more status bits */ 926 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT)) 927 928 #define __PM_SOFT_DIRTY (1LL) 929 #define PM_PRESENT PM_STATUS(4LL) 930 #define PM_SWAP PM_STATUS(2LL) 931 #define PM_FILE PM_STATUS(1LL) 932 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0) 933 #define PM_END_OF_BUFFER 1 934 935 static inline pagemap_entry_t make_pme(u64 val) 936 { 937 return (pagemap_entry_t) { .pme = val }; 938 } 939 940 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 941 struct pagemapread *pm) 942 { 943 pm->buffer[pm->pos++] = *pme; 944 if (pm->pos >= pm->len) 945 return PM_END_OF_BUFFER; 946 return 0; 947 } 948 949 static int pagemap_pte_hole(unsigned long start, unsigned long end, 950 struct mm_walk *walk) 951 { 952 struct pagemapread *pm = walk->private; 953 unsigned long addr = start; 954 int err = 0; 955 956 while (addr < end) { 957 struct vm_area_struct *vma = find_vma(walk->mm, addr); 958 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 959 /* End of address space hole, which we mark as non-present. */ 960 unsigned long hole_end; 961 962 if (vma) 963 hole_end = min(end, vma->vm_start); 964 else 965 hole_end = end; 966 967 for (; addr < hole_end; addr += PAGE_SIZE) { 968 err = add_to_pagemap(addr, &pme, pm); 969 if (err) 970 goto out; 971 } 972 973 if (!vma) 974 break; 975 976 /* Addresses in the VMA. */ 977 if (vma->vm_flags & VM_SOFTDIRTY) 978 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY); 979 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 980 err = add_to_pagemap(addr, &pme, pm); 981 if (err) 982 goto out; 983 } 984 } 985 out: 986 return err; 987 } 988 989 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 990 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 991 { 992 u64 frame, flags; 993 struct page *page = NULL; 994 int flags2 = 0; 995 996 if (pte_present(pte)) { 997 frame = pte_pfn(pte); 998 flags = PM_PRESENT; 999 page = vm_normal_page(vma, addr, pte); 1000 if (pte_soft_dirty(pte)) 1001 flags2 |= __PM_SOFT_DIRTY; 1002 } else if (is_swap_pte(pte)) { 1003 swp_entry_t entry; 1004 if (pte_swp_soft_dirty(pte)) 1005 flags2 |= __PM_SOFT_DIRTY; 1006 entry = pte_to_swp_entry(pte); 1007 frame = swp_type(entry) | 1008 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1009 flags = PM_SWAP; 1010 if (is_migration_entry(entry)) 1011 page = migration_entry_to_page(entry); 1012 } else { 1013 if (vma->vm_flags & VM_SOFTDIRTY) 1014 flags2 |= __PM_SOFT_DIRTY; 1015 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2)); 1016 return; 1017 } 1018 1019 if (page && !PageAnon(page)) 1020 flags |= PM_FILE; 1021 if ((vma->vm_flags & VM_SOFTDIRTY)) 1022 flags2 |= __PM_SOFT_DIRTY; 1023 1024 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags); 1025 } 1026 1027 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1028 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1029 pmd_t pmd, int offset, int pmd_flags2) 1030 { 1031 /* 1032 * Currently pmd for thp is always present because thp can not be 1033 * swapped-out, migrated, or HWPOISONed (split in such cases instead.) 1034 * This if-check is just to prepare for future implementation. 1035 */ 1036 if (pmd_present(pmd)) 1037 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset) 1038 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT); 1039 else 1040 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2)); 1041 } 1042 #else 1043 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1044 pmd_t pmd, int offset, int pmd_flags2) 1045 { 1046 } 1047 #endif 1048 1049 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 1050 struct mm_walk *walk) 1051 { 1052 struct vm_area_struct *vma; 1053 struct pagemapread *pm = walk->private; 1054 spinlock_t *ptl; 1055 pte_t *pte, *orig_pte; 1056 int err = 0; 1057 1058 /* find the first VMA at or above 'addr' */ 1059 vma = find_vma(walk->mm, addr); 1060 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 1061 int pmd_flags2; 1062 1063 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd)) 1064 pmd_flags2 = __PM_SOFT_DIRTY; 1065 else 1066 pmd_flags2 = 0; 1067 1068 for (; addr != end; addr += PAGE_SIZE) { 1069 unsigned long offset; 1070 pagemap_entry_t pme; 1071 1072 offset = (addr & ~PAGEMAP_WALK_MASK) >> 1073 PAGE_SHIFT; 1074 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2); 1075 err = add_to_pagemap(addr, &pme, pm); 1076 if (err) 1077 break; 1078 } 1079 spin_unlock(ptl); 1080 return err; 1081 } 1082 1083 if (pmd_trans_unstable(pmd)) 1084 return 0; 1085 1086 while (1) { 1087 /* End of address space hole, which we mark as non-present. */ 1088 unsigned long hole_end; 1089 1090 if (vma) 1091 hole_end = min(end, vma->vm_start); 1092 else 1093 hole_end = end; 1094 1095 for (; addr < hole_end; addr += PAGE_SIZE) { 1096 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 1097 1098 err = add_to_pagemap(addr, &pme, pm); 1099 if (err) 1100 return err; 1101 } 1102 1103 if (!vma || vma->vm_start >= end) 1104 break; 1105 /* 1106 * We can't possibly be in a hugetlb VMA. In general, 1107 * for a mm_walk with a pmd_entry and a hugetlb_entry, 1108 * the pmd_entry can only be called on addresses in a 1109 * hugetlb if the walk starts in a non-hugetlb VMA and 1110 * spans a hugepage VMA. Since pagemap_read walks are 1111 * PMD-sized and PMD-aligned, this will never be true. 1112 */ 1113 BUG_ON(is_vm_hugetlb_page(vma)); 1114 1115 /* Addresses in the VMA. */ 1116 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1117 for (; addr < min(end, vma->vm_end); pte++, addr += PAGE_SIZE) { 1118 pagemap_entry_t pme; 1119 1120 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte); 1121 err = add_to_pagemap(addr, &pme, pm); 1122 if (err) 1123 break; 1124 } 1125 pte_unmap_unlock(orig_pte, ptl); 1126 1127 if (err) 1128 return err; 1129 1130 if (addr == end) 1131 break; 1132 1133 vma = find_vma(walk->mm, addr); 1134 } 1135 1136 cond_resched(); 1137 1138 return err; 1139 } 1140 1141 #ifdef CONFIG_HUGETLB_PAGE 1142 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1143 pte_t pte, int offset, int flags2) 1144 { 1145 if (pte_present(pte)) 1146 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) | 1147 PM_STATUS2(pm->v2, flags2) | 1148 PM_PRESENT); 1149 else 1150 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | 1151 PM_STATUS2(pm->v2, flags2)); 1152 } 1153 1154 /* This function walks within one hugetlb entry in the single call */ 1155 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 1156 unsigned long addr, unsigned long end, 1157 struct mm_walk *walk) 1158 { 1159 struct pagemapread *pm = walk->private; 1160 struct vm_area_struct *vma; 1161 int err = 0; 1162 int flags2; 1163 pagemap_entry_t pme; 1164 1165 vma = find_vma(walk->mm, addr); 1166 WARN_ON_ONCE(!vma); 1167 1168 if (vma && (vma->vm_flags & VM_SOFTDIRTY)) 1169 flags2 = __PM_SOFT_DIRTY; 1170 else 1171 flags2 = 0; 1172 1173 for (; addr != end; addr += PAGE_SIZE) { 1174 int offset = (addr & ~hmask) >> PAGE_SHIFT; 1175 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2); 1176 err = add_to_pagemap(addr, &pme, pm); 1177 if (err) 1178 return err; 1179 } 1180 1181 cond_resched(); 1182 1183 return err; 1184 } 1185 #endif /* HUGETLB_PAGE */ 1186 1187 /* 1188 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1189 * 1190 * For each page in the address space, this file contains one 64-bit entry 1191 * consisting of the following: 1192 * 1193 * Bits 0-54 page frame number (PFN) if present 1194 * Bits 0-4 swap type if swapped 1195 * Bits 5-54 swap offset if swapped 1196 * Bits 55-60 page shift (page size = 1<<page shift) 1197 * Bit 61 page is file-page or shared-anon 1198 * Bit 62 page swapped 1199 * Bit 63 page present 1200 * 1201 * If the page is not present but in swap, then the PFN contains an 1202 * encoding of the swap file number and the page's offset into the 1203 * swap. Unmapped pages return a null PFN. This allows determining 1204 * precisely which pages are mapped (or in swap) and comparing mapped 1205 * pages between processes. 1206 * 1207 * Efficient users of this interface will use /proc/pid/maps to 1208 * determine which areas of memory are actually mapped and llseek to 1209 * skip over unmapped regions. 1210 */ 1211 static ssize_t pagemap_read(struct file *file, char __user *buf, 1212 size_t count, loff_t *ppos) 1213 { 1214 struct task_struct *task = get_proc_task(file_inode(file)); 1215 struct mm_struct *mm; 1216 struct pagemapread pm; 1217 int ret = -ESRCH; 1218 struct mm_walk pagemap_walk = {}; 1219 unsigned long src; 1220 unsigned long svpfn; 1221 unsigned long start_vaddr; 1222 unsigned long end_vaddr; 1223 int copied = 0; 1224 1225 if (!task) 1226 goto out; 1227 1228 ret = -EINVAL; 1229 /* file position must be aligned */ 1230 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1231 goto out_task; 1232 1233 ret = 0; 1234 if (!count) 1235 goto out_task; 1236 1237 pm.v2 = soft_dirty_cleared; 1238 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1239 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); 1240 ret = -ENOMEM; 1241 if (!pm.buffer) 1242 goto out_task; 1243 1244 mm = mm_access(task, PTRACE_MODE_READ); 1245 ret = PTR_ERR(mm); 1246 if (!mm || IS_ERR(mm)) 1247 goto out_free; 1248 1249 pagemap_walk.pmd_entry = pagemap_pte_range; 1250 pagemap_walk.pte_hole = pagemap_pte_hole; 1251 #ifdef CONFIG_HUGETLB_PAGE 1252 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1253 #endif 1254 pagemap_walk.mm = mm; 1255 pagemap_walk.private = ± 1256 1257 src = *ppos; 1258 svpfn = src / PM_ENTRY_BYTES; 1259 start_vaddr = svpfn << PAGE_SHIFT; 1260 end_vaddr = TASK_SIZE_OF(task); 1261 1262 /* watch out for wraparound */ 1263 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 1264 start_vaddr = end_vaddr; 1265 1266 /* 1267 * The odds are that this will stop walking way 1268 * before end_vaddr, because the length of the 1269 * user buffer is tracked in "pm", and the walk 1270 * will stop when we hit the end of the buffer. 1271 */ 1272 ret = 0; 1273 while (count && (start_vaddr < end_vaddr)) { 1274 int len; 1275 unsigned long end; 1276 1277 pm.pos = 0; 1278 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1279 /* overflow ? */ 1280 if (end < start_vaddr || end > end_vaddr) 1281 end = end_vaddr; 1282 down_read(&mm->mmap_sem); 1283 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1284 up_read(&mm->mmap_sem); 1285 start_vaddr = end; 1286 1287 len = min(count, PM_ENTRY_BYTES * pm.pos); 1288 if (copy_to_user(buf, pm.buffer, len)) { 1289 ret = -EFAULT; 1290 goto out_mm; 1291 } 1292 copied += len; 1293 buf += len; 1294 count -= len; 1295 } 1296 *ppos += copied; 1297 if (!ret || ret == PM_END_OF_BUFFER) 1298 ret = copied; 1299 1300 out_mm: 1301 mmput(mm); 1302 out_free: 1303 kfree(pm.buffer); 1304 out_task: 1305 put_task_struct(task); 1306 out: 1307 return ret; 1308 } 1309 1310 static int pagemap_open(struct inode *inode, struct file *file) 1311 { 1312 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about " 1313 "to stop being page-shift some time soon. See the " 1314 "linux/Documentation/vm/pagemap.txt for details.\n"); 1315 return 0; 1316 } 1317 1318 const struct file_operations proc_pagemap_operations = { 1319 .llseek = mem_lseek, /* borrow this */ 1320 .read = pagemap_read, 1321 .open = pagemap_open, 1322 }; 1323 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1324 1325 #ifdef CONFIG_NUMA 1326 1327 struct numa_maps { 1328 struct vm_area_struct *vma; 1329 unsigned long pages; 1330 unsigned long anon; 1331 unsigned long active; 1332 unsigned long writeback; 1333 unsigned long mapcount_max; 1334 unsigned long dirty; 1335 unsigned long swapcache; 1336 unsigned long node[MAX_NUMNODES]; 1337 }; 1338 1339 struct numa_maps_private { 1340 struct proc_maps_private proc_maps; 1341 struct numa_maps md; 1342 }; 1343 1344 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1345 unsigned long nr_pages) 1346 { 1347 int count = page_mapcount(page); 1348 1349 md->pages += nr_pages; 1350 if (pte_dirty || PageDirty(page)) 1351 md->dirty += nr_pages; 1352 1353 if (PageSwapCache(page)) 1354 md->swapcache += nr_pages; 1355 1356 if (PageActive(page) || PageUnevictable(page)) 1357 md->active += nr_pages; 1358 1359 if (PageWriteback(page)) 1360 md->writeback += nr_pages; 1361 1362 if (PageAnon(page)) 1363 md->anon += nr_pages; 1364 1365 if (count > md->mapcount_max) 1366 md->mapcount_max = count; 1367 1368 md->node[page_to_nid(page)] += nr_pages; 1369 } 1370 1371 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1372 unsigned long addr) 1373 { 1374 struct page *page; 1375 int nid; 1376 1377 if (!pte_present(pte)) 1378 return NULL; 1379 1380 page = vm_normal_page(vma, addr, pte); 1381 if (!page) 1382 return NULL; 1383 1384 if (PageReserved(page)) 1385 return NULL; 1386 1387 nid = page_to_nid(page); 1388 if (!node_isset(nid, node_states[N_MEMORY])) 1389 return NULL; 1390 1391 return page; 1392 } 1393 1394 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1395 unsigned long end, struct mm_walk *walk) 1396 { 1397 struct numa_maps *md; 1398 spinlock_t *ptl; 1399 pte_t *orig_pte; 1400 pte_t *pte; 1401 1402 md = walk->private; 1403 1404 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) { 1405 pte_t huge_pte = *(pte_t *)pmd; 1406 struct page *page; 1407 1408 page = can_gather_numa_stats(huge_pte, md->vma, addr); 1409 if (page) 1410 gather_stats(page, md, pte_dirty(huge_pte), 1411 HPAGE_PMD_SIZE/PAGE_SIZE); 1412 spin_unlock(ptl); 1413 return 0; 1414 } 1415 1416 if (pmd_trans_unstable(pmd)) 1417 return 0; 1418 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1419 do { 1420 struct page *page = can_gather_numa_stats(*pte, md->vma, addr); 1421 if (!page) 1422 continue; 1423 gather_stats(page, md, pte_dirty(*pte), 1); 1424 1425 } while (pte++, addr += PAGE_SIZE, addr != end); 1426 pte_unmap_unlock(orig_pte, ptl); 1427 return 0; 1428 } 1429 #ifdef CONFIG_HUGETLB_PAGE 1430 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1431 unsigned long addr, unsigned long end, struct mm_walk *walk) 1432 { 1433 struct numa_maps *md; 1434 struct page *page; 1435 1436 if (!pte_present(*pte)) 1437 return 0; 1438 1439 page = pte_page(*pte); 1440 if (!page) 1441 return 0; 1442 1443 md = walk->private; 1444 gather_stats(page, md, pte_dirty(*pte), 1); 1445 return 0; 1446 } 1447 1448 #else 1449 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1450 unsigned long addr, unsigned long end, struct mm_walk *walk) 1451 { 1452 return 0; 1453 } 1454 #endif 1455 1456 /* 1457 * Display pages allocated per node and memory policy via /proc. 1458 */ 1459 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1460 { 1461 struct numa_maps_private *numa_priv = m->private; 1462 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1463 struct vm_area_struct *vma = v; 1464 struct numa_maps *md = &numa_priv->md; 1465 struct file *file = vma->vm_file; 1466 struct mm_struct *mm = vma->vm_mm; 1467 struct mm_walk walk = {}; 1468 struct mempolicy *pol; 1469 char buffer[64]; 1470 int nid; 1471 1472 if (!mm) 1473 return 0; 1474 1475 /* Ensure we start with an empty set of numa_maps statistics. */ 1476 memset(md, 0, sizeof(*md)); 1477 1478 md->vma = vma; 1479 1480 walk.hugetlb_entry = gather_hugetbl_stats; 1481 walk.pmd_entry = gather_pte_stats; 1482 walk.private = md; 1483 walk.mm = mm; 1484 1485 pol = __get_vma_policy(vma, vma->vm_start); 1486 if (pol) { 1487 mpol_to_str(buffer, sizeof(buffer), pol); 1488 mpol_cond_put(pol); 1489 } else { 1490 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1491 } 1492 1493 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1494 1495 if (file) { 1496 seq_puts(m, " file="); 1497 seq_path(m, &file->f_path, "\n\t= "); 1498 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1499 seq_puts(m, " heap"); 1500 } else { 1501 pid_t tid = pid_of_stack(proc_priv, vma, is_pid); 1502 if (tid != 0) { 1503 /* 1504 * Thread stack in /proc/PID/task/TID/maps or 1505 * the main process stack. 1506 */ 1507 if (!is_pid || (vma->vm_start <= mm->start_stack && 1508 vma->vm_end >= mm->start_stack)) 1509 seq_puts(m, " stack"); 1510 else 1511 seq_printf(m, " stack:%d", tid); 1512 } 1513 } 1514 1515 if (is_vm_hugetlb_page(vma)) 1516 seq_puts(m, " huge"); 1517 1518 walk_page_range(vma->vm_start, vma->vm_end, &walk); 1519 1520 if (!md->pages) 1521 goto out; 1522 1523 if (md->anon) 1524 seq_printf(m, " anon=%lu", md->anon); 1525 1526 if (md->dirty) 1527 seq_printf(m, " dirty=%lu", md->dirty); 1528 1529 if (md->pages != md->anon && md->pages != md->dirty) 1530 seq_printf(m, " mapped=%lu", md->pages); 1531 1532 if (md->mapcount_max > 1) 1533 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1534 1535 if (md->swapcache) 1536 seq_printf(m, " swapcache=%lu", md->swapcache); 1537 1538 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1539 seq_printf(m, " active=%lu", md->active); 1540 1541 if (md->writeback) 1542 seq_printf(m, " writeback=%lu", md->writeback); 1543 1544 for_each_node_state(nid, N_MEMORY) 1545 if (md->node[nid]) 1546 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1547 out: 1548 seq_putc(m, '\n'); 1549 m_cache_vma(m, vma); 1550 return 0; 1551 } 1552 1553 static int show_pid_numa_map(struct seq_file *m, void *v) 1554 { 1555 return show_numa_map(m, v, 1); 1556 } 1557 1558 static int show_tid_numa_map(struct seq_file *m, void *v) 1559 { 1560 return show_numa_map(m, v, 0); 1561 } 1562 1563 static const struct seq_operations proc_pid_numa_maps_op = { 1564 .start = m_start, 1565 .next = m_next, 1566 .stop = m_stop, 1567 .show = show_pid_numa_map, 1568 }; 1569 1570 static const struct seq_operations proc_tid_numa_maps_op = { 1571 .start = m_start, 1572 .next = m_next, 1573 .stop = m_stop, 1574 .show = show_tid_numa_map, 1575 }; 1576 1577 static int numa_maps_open(struct inode *inode, struct file *file, 1578 const struct seq_operations *ops) 1579 { 1580 return proc_maps_open(inode, file, ops, 1581 sizeof(struct numa_maps_private)); 1582 } 1583 1584 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1585 { 1586 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1587 } 1588 1589 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1590 { 1591 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1592 } 1593 1594 const struct file_operations proc_pid_numa_maps_operations = { 1595 .open = pid_numa_maps_open, 1596 .read = seq_read, 1597 .llseek = seq_lseek, 1598 .release = proc_map_release, 1599 }; 1600 1601 const struct file_operations proc_tid_numa_maps_operations = { 1602 .open = tid_numa_maps_open, 1603 .read = seq_read, 1604 .llseek = seq_lseek, 1605 .release = proc_map_release, 1606 }; 1607 #endif /* CONFIG_NUMA */ 1608