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 u64 pss; 447 }; 448 449 static void smaps_account(struct mem_size_stats *mss, struct page *page, 450 unsigned long size, bool young, bool dirty) 451 { 452 int mapcount; 453 454 if (PageAnon(page)) 455 mss->anonymous += size; 456 457 mss->resident += size; 458 /* Accumulate the size in pages that have been accessed. */ 459 if (young || PageReferenced(page)) 460 mss->referenced += size; 461 mapcount = page_mapcount(page); 462 if (mapcount >= 2) { 463 u64 pss_delta; 464 465 if (dirty || PageDirty(page)) 466 mss->shared_dirty += size; 467 else 468 mss->shared_clean += size; 469 pss_delta = (u64)size << PSS_SHIFT; 470 do_div(pss_delta, mapcount); 471 mss->pss += pss_delta; 472 } else { 473 if (dirty || PageDirty(page)) 474 mss->private_dirty += size; 475 else 476 mss->private_clean += size; 477 mss->pss += (u64)size << PSS_SHIFT; 478 } 479 } 480 481 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 482 struct mm_walk *walk) 483 { 484 struct mem_size_stats *mss = walk->private; 485 struct vm_area_struct *vma = mss->vma; 486 struct page *page = NULL; 487 488 if (pte_present(*pte)) { 489 page = vm_normal_page(vma, addr, *pte); 490 } else if (is_swap_pte(*pte)) { 491 swp_entry_t swpent = pte_to_swp_entry(*pte); 492 493 if (!non_swap_entry(swpent)) 494 mss->swap += PAGE_SIZE; 495 else if (is_migration_entry(swpent)) 496 page = migration_entry_to_page(swpent); 497 } 498 499 if (!page) 500 return; 501 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte)); 502 } 503 504 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 505 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 506 struct mm_walk *walk) 507 { 508 struct mem_size_stats *mss = walk->private; 509 struct vm_area_struct *vma = mss->vma; 510 struct page *page; 511 512 /* FOLL_DUMP will return -EFAULT on huge zero page */ 513 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 514 if (IS_ERR_OR_NULL(page)) 515 return; 516 mss->anonymous_thp += HPAGE_PMD_SIZE; 517 smaps_account(mss, page, HPAGE_PMD_SIZE, 518 pmd_young(*pmd), pmd_dirty(*pmd)); 519 } 520 #else 521 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 522 struct mm_walk *walk) 523 { 524 } 525 #endif 526 527 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 528 struct mm_walk *walk) 529 { 530 struct mem_size_stats *mss = walk->private; 531 struct vm_area_struct *vma = mss->vma; 532 pte_t *pte; 533 spinlock_t *ptl; 534 535 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 536 smaps_pmd_entry(pmd, addr, walk); 537 spin_unlock(ptl); 538 return 0; 539 } 540 541 if (pmd_trans_unstable(pmd)) 542 return 0; 543 /* 544 * The mmap_sem held all the way back in m_start() is what 545 * keeps khugepaged out of here and from collapsing things 546 * in here. 547 */ 548 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 549 for (; addr != end; pte++, addr += PAGE_SIZE) 550 smaps_pte_entry(pte, addr, walk); 551 pte_unmap_unlock(pte - 1, ptl); 552 cond_resched(); 553 return 0; 554 } 555 556 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 557 { 558 /* 559 * Don't forget to update Documentation/ on changes. 560 */ 561 static const char mnemonics[BITS_PER_LONG][2] = { 562 /* 563 * In case if we meet a flag we don't know about. 564 */ 565 [0 ... (BITS_PER_LONG-1)] = "??", 566 567 [ilog2(VM_READ)] = "rd", 568 [ilog2(VM_WRITE)] = "wr", 569 [ilog2(VM_EXEC)] = "ex", 570 [ilog2(VM_SHARED)] = "sh", 571 [ilog2(VM_MAYREAD)] = "mr", 572 [ilog2(VM_MAYWRITE)] = "mw", 573 [ilog2(VM_MAYEXEC)] = "me", 574 [ilog2(VM_MAYSHARE)] = "ms", 575 [ilog2(VM_GROWSDOWN)] = "gd", 576 [ilog2(VM_PFNMAP)] = "pf", 577 [ilog2(VM_DENYWRITE)] = "dw", 578 #ifdef CONFIG_X86_INTEL_MPX 579 [ilog2(VM_MPX)] = "mp", 580 #endif 581 [ilog2(VM_LOCKED)] = "lo", 582 [ilog2(VM_IO)] = "io", 583 [ilog2(VM_SEQ_READ)] = "sr", 584 [ilog2(VM_RAND_READ)] = "rr", 585 [ilog2(VM_DONTCOPY)] = "dc", 586 [ilog2(VM_DONTEXPAND)] = "de", 587 [ilog2(VM_ACCOUNT)] = "ac", 588 [ilog2(VM_NORESERVE)] = "nr", 589 [ilog2(VM_HUGETLB)] = "ht", 590 [ilog2(VM_ARCH_1)] = "ar", 591 [ilog2(VM_DONTDUMP)] = "dd", 592 #ifdef CONFIG_MEM_SOFT_DIRTY 593 [ilog2(VM_SOFTDIRTY)] = "sd", 594 #endif 595 [ilog2(VM_MIXEDMAP)] = "mm", 596 [ilog2(VM_HUGEPAGE)] = "hg", 597 [ilog2(VM_NOHUGEPAGE)] = "nh", 598 [ilog2(VM_MERGEABLE)] = "mg", 599 }; 600 size_t i; 601 602 seq_puts(m, "VmFlags: "); 603 for (i = 0; i < BITS_PER_LONG; i++) { 604 if (vma->vm_flags & (1UL << i)) { 605 seq_printf(m, "%c%c ", 606 mnemonics[i][0], mnemonics[i][1]); 607 } 608 } 609 seq_putc(m, '\n'); 610 } 611 612 static int show_smap(struct seq_file *m, void *v, int is_pid) 613 { 614 struct vm_area_struct *vma = v; 615 struct mem_size_stats mss; 616 struct mm_walk smaps_walk = { 617 .pmd_entry = smaps_pte_range, 618 .mm = vma->vm_mm, 619 .private = &mss, 620 }; 621 622 memset(&mss, 0, sizeof mss); 623 mss.vma = vma; 624 /* mmap_sem is held in m_start */ 625 if (vma->vm_mm && !is_vm_hugetlb_page(vma)) 626 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); 627 628 show_map_vma(m, vma, is_pid); 629 630 seq_printf(m, 631 "Size: %8lu kB\n" 632 "Rss: %8lu kB\n" 633 "Pss: %8lu kB\n" 634 "Shared_Clean: %8lu kB\n" 635 "Shared_Dirty: %8lu kB\n" 636 "Private_Clean: %8lu kB\n" 637 "Private_Dirty: %8lu kB\n" 638 "Referenced: %8lu kB\n" 639 "Anonymous: %8lu kB\n" 640 "AnonHugePages: %8lu kB\n" 641 "Swap: %8lu kB\n" 642 "KernelPageSize: %8lu kB\n" 643 "MMUPageSize: %8lu kB\n" 644 "Locked: %8lu kB\n", 645 (vma->vm_end - vma->vm_start) >> 10, 646 mss.resident >> 10, 647 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 648 mss.shared_clean >> 10, 649 mss.shared_dirty >> 10, 650 mss.private_clean >> 10, 651 mss.private_dirty >> 10, 652 mss.referenced >> 10, 653 mss.anonymous >> 10, 654 mss.anonymous_thp >> 10, 655 mss.swap >> 10, 656 vma_kernel_pagesize(vma) >> 10, 657 vma_mmu_pagesize(vma) >> 10, 658 (vma->vm_flags & VM_LOCKED) ? 659 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 660 661 show_smap_vma_flags(m, vma); 662 m_cache_vma(m, vma); 663 return 0; 664 } 665 666 static int show_pid_smap(struct seq_file *m, void *v) 667 { 668 return show_smap(m, v, 1); 669 } 670 671 static int show_tid_smap(struct seq_file *m, void *v) 672 { 673 return show_smap(m, v, 0); 674 } 675 676 static const struct seq_operations proc_pid_smaps_op = { 677 .start = m_start, 678 .next = m_next, 679 .stop = m_stop, 680 .show = show_pid_smap 681 }; 682 683 static const struct seq_operations proc_tid_smaps_op = { 684 .start = m_start, 685 .next = m_next, 686 .stop = m_stop, 687 .show = show_tid_smap 688 }; 689 690 static int pid_smaps_open(struct inode *inode, struct file *file) 691 { 692 return do_maps_open(inode, file, &proc_pid_smaps_op); 693 } 694 695 static int tid_smaps_open(struct inode *inode, struct file *file) 696 { 697 return do_maps_open(inode, file, &proc_tid_smaps_op); 698 } 699 700 const struct file_operations proc_pid_smaps_operations = { 701 .open = pid_smaps_open, 702 .read = seq_read, 703 .llseek = seq_lseek, 704 .release = proc_map_release, 705 }; 706 707 const struct file_operations proc_tid_smaps_operations = { 708 .open = tid_smaps_open, 709 .read = seq_read, 710 .llseek = seq_lseek, 711 .release = proc_map_release, 712 }; 713 714 /* 715 * We do not want to have constant page-shift bits sitting in 716 * pagemap entries and are about to reuse them some time soon. 717 * 718 * Here's the "migration strategy": 719 * 1. when the system boots these bits remain what they are, 720 * but a warning about future change is printed in log; 721 * 2. once anyone clears soft-dirty bits via clear_refs file, 722 * these flag is set to denote, that user is aware of the 723 * new API and those page-shift bits change their meaning. 724 * The respective warning is printed in dmesg; 725 * 3. In a couple of releases we will remove all the mentions 726 * of page-shift in pagemap entries. 727 */ 728 729 static bool soft_dirty_cleared __read_mostly; 730 731 enum clear_refs_types { 732 CLEAR_REFS_ALL = 1, 733 CLEAR_REFS_ANON, 734 CLEAR_REFS_MAPPED, 735 CLEAR_REFS_SOFT_DIRTY, 736 CLEAR_REFS_LAST, 737 }; 738 739 struct clear_refs_private { 740 struct vm_area_struct *vma; 741 enum clear_refs_types type; 742 }; 743 744 static inline void clear_soft_dirty(struct vm_area_struct *vma, 745 unsigned long addr, pte_t *pte) 746 { 747 #ifdef CONFIG_MEM_SOFT_DIRTY 748 /* 749 * The soft-dirty tracker uses #PF-s to catch writes 750 * to pages, so write-protect the pte as well. See the 751 * Documentation/vm/soft-dirty.txt for full description 752 * of how soft-dirty works. 753 */ 754 pte_t ptent = *pte; 755 756 if (pte_present(ptent)) { 757 ptent = pte_wrprotect(ptent); 758 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY); 759 } else if (is_swap_pte(ptent)) { 760 ptent = pte_swp_clear_soft_dirty(ptent); 761 } 762 763 set_pte_at(vma->vm_mm, addr, pte, ptent); 764 #endif 765 } 766 767 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 768 unsigned long end, struct mm_walk *walk) 769 { 770 struct clear_refs_private *cp = walk->private; 771 struct vm_area_struct *vma = cp->vma; 772 pte_t *pte, ptent; 773 spinlock_t *ptl; 774 struct page *page; 775 776 split_huge_page_pmd(vma, addr, pmd); 777 if (pmd_trans_unstable(pmd)) 778 return 0; 779 780 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 781 for (; addr != end; pte++, addr += PAGE_SIZE) { 782 ptent = *pte; 783 784 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 785 clear_soft_dirty(vma, addr, pte); 786 continue; 787 } 788 789 if (!pte_present(ptent)) 790 continue; 791 792 page = vm_normal_page(vma, addr, ptent); 793 if (!page) 794 continue; 795 796 /* Clear accessed and referenced bits. */ 797 ptep_test_and_clear_young(vma, addr, pte); 798 ClearPageReferenced(page); 799 } 800 pte_unmap_unlock(pte - 1, ptl); 801 cond_resched(); 802 return 0; 803 } 804 805 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 806 size_t count, loff_t *ppos) 807 { 808 struct task_struct *task; 809 char buffer[PROC_NUMBUF]; 810 struct mm_struct *mm; 811 struct vm_area_struct *vma; 812 enum clear_refs_types type; 813 int itype; 814 int rv; 815 816 memset(buffer, 0, sizeof(buffer)); 817 if (count > sizeof(buffer) - 1) 818 count = sizeof(buffer) - 1; 819 if (copy_from_user(buffer, buf, count)) 820 return -EFAULT; 821 rv = kstrtoint(strstrip(buffer), 10, &itype); 822 if (rv < 0) 823 return rv; 824 type = (enum clear_refs_types)itype; 825 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 826 return -EINVAL; 827 828 if (type == CLEAR_REFS_SOFT_DIRTY) { 829 soft_dirty_cleared = true; 830 pr_warn_once("The pagemap bits 55-60 has changed their meaning!" 831 " See the linux/Documentation/vm/pagemap.txt for " 832 "details.\n"); 833 } 834 835 task = get_proc_task(file_inode(file)); 836 if (!task) 837 return -ESRCH; 838 mm = get_task_mm(task); 839 if (mm) { 840 struct clear_refs_private cp = { 841 .type = type, 842 }; 843 struct mm_walk clear_refs_walk = { 844 .pmd_entry = clear_refs_pte_range, 845 .mm = mm, 846 .private = &cp, 847 }; 848 down_read(&mm->mmap_sem); 849 if (type == CLEAR_REFS_SOFT_DIRTY) { 850 for (vma = mm->mmap; vma; vma = vma->vm_next) { 851 if (!(vma->vm_flags & VM_SOFTDIRTY)) 852 continue; 853 up_read(&mm->mmap_sem); 854 down_write(&mm->mmap_sem); 855 for (vma = mm->mmap; vma; vma = vma->vm_next) { 856 vma->vm_flags &= ~VM_SOFTDIRTY; 857 vma_set_page_prot(vma); 858 } 859 downgrade_write(&mm->mmap_sem); 860 break; 861 } 862 mmu_notifier_invalidate_range_start(mm, 0, -1); 863 } 864 for (vma = mm->mmap; vma; vma = vma->vm_next) { 865 cp.vma = vma; 866 if (is_vm_hugetlb_page(vma)) 867 continue; 868 /* 869 * Writing 1 to /proc/pid/clear_refs affects all pages. 870 * 871 * Writing 2 to /proc/pid/clear_refs only affects 872 * Anonymous pages. 873 * 874 * Writing 3 to /proc/pid/clear_refs only affects file 875 * mapped pages. 876 * 877 * Writing 4 to /proc/pid/clear_refs affects all pages. 878 */ 879 if (type == CLEAR_REFS_ANON && vma->vm_file) 880 continue; 881 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 882 continue; 883 walk_page_range(vma->vm_start, vma->vm_end, 884 &clear_refs_walk); 885 } 886 if (type == CLEAR_REFS_SOFT_DIRTY) 887 mmu_notifier_invalidate_range_end(mm, 0, -1); 888 flush_tlb_mm(mm); 889 up_read(&mm->mmap_sem); 890 mmput(mm); 891 } 892 put_task_struct(task); 893 894 return count; 895 } 896 897 const struct file_operations proc_clear_refs_operations = { 898 .write = clear_refs_write, 899 .llseek = noop_llseek, 900 }; 901 902 typedef struct { 903 u64 pme; 904 } pagemap_entry_t; 905 906 struct pagemapread { 907 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 908 pagemap_entry_t *buffer; 909 bool v2; 910 }; 911 912 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 913 #define PAGEMAP_WALK_MASK (PMD_MASK) 914 915 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 916 #define PM_STATUS_BITS 3 917 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 918 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 919 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 920 #define PM_PSHIFT_BITS 6 921 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 922 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 923 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 924 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 925 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 926 /* in "new" pagemap pshift bits are occupied with more status bits */ 927 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT)) 928 929 #define __PM_SOFT_DIRTY (1LL) 930 #define PM_PRESENT PM_STATUS(4LL) 931 #define PM_SWAP PM_STATUS(2LL) 932 #define PM_FILE PM_STATUS(1LL) 933 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0) 934 #define PM_END_OF_BUFFER 1 935 936 static inline pagemap_entry_t make_pme(u64 val) 937 { 938 return (pagemap_entry_t) { .pme = val }; 939 } 940 941 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 942 struct pagemapread *pm) 943 { 944 pm->buffer[pm->pos++] = *pme; 945 if (pm->pos >= pm->len) 946 return PM_END_OF_BUFFER; 947 return 0; 948 } 949 950 static int pagemap_pte_hole(unsigned long start, unsigned long end, 951 struct mm_walk *walk) 952 { 953 struct pagemapread *pm = walk->private; 954 unsigned long addr = start; 955 int err = 0; 956 957 while (addr < end) { 958 struct vm_area_struct *vma = find_vma(walk->mm, addr); 959 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 960 /* End of address space hole, which we mark as non-present. */ 961 unsigned long hole_end; 962 963 if (vma) 964 hole_end = min(end, vma->vm_start); 965 else 966 hole_end = end; 967 968 for (; addr < hole_end; addr += PAGE_SIZE) { 969 err = add_to_pagemap(addr, &pme, pm); 970 if (err) 971 goto out; 972 } 973 974 if (!vma) 975 break; 976 977 /* Addresses in the VMA. */ 978 if (vma->vm_flags & VM_SOFTDIRTY) 979 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY); 980 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 981 err = add_to_pagemap(addr, &pme, pm); 982 if (err) 983 goto out; 984 } 985 } 986 out: 987 return err; 988 } 989 990 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 991 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 992 { 993 u64 frame, flags; 994 struct page *page = NULL; 995 int flags2 = 0; 996 997 if (pte_present(pte)) { 998 frame = pte_pfn(pte); 999 flags = PM_PRESENT; 1000 page = vm_normal_page(vma, addr, pte); 1001 if (pte_soft_dirty(pte)) 1002 flags2 |= __PM_SOFT_DIRTY; 1003 } else if (is_swap_pte(pte)) { 1004 swp_entry_t entry; 1005 if (pte_swp_soft_dirty(pte)) 1006 flags2 |= __PM_SOFT_DIRTY; 1007 entry = pte_to_swp_entry(pte); 1008 frame = swp_type(entry) | 1009 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1010 flags = PM_SWAP; 1011 if (is_migration_entry(entry)) 1012 page = migration_entry_to_page(entry); 1013 } else { 1014 if (vma->vm_flags & VM_SOFTDIRTY) 1015 flags2 |= __PM_SOFT_DIRTY; 1016 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2)); 1017 return; 1018 } 1019 1020 if (page && !PageAnon(page)) 1021 flags |= PM_FILE; 1022 if ((vma->vm_flags & VM_SOFTDIRTY)) 1023 flags2 |= __PM_SOFT_DIRTY; 1024 1025 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags); 1026 } 1027 1028 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1029 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1030 pmd_t pmd, int offset, int pmd_flags2) 1031 { 1032 /* 1033 * Currently pmd for thp is always present because thp can not be 1034 * swapped-out, migrated, or HWPOISONed (split in such cases instead.) 1035 * This if-check is just to prepare for future implementation. 1036 */ 1037 if (pmd_present(pmd)) 1038 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset) 1039 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT); 1040 else 1041 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2)); 1042 } 1043 #else 1044 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1045 pmd_t pmd, int offset, int pmd_flags2) 1046 { 1047 } 1048 #endif 1049 1050 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 1051 struct mm_walk *walk) 1052 { 1053 struct vm_area_struct *vma; 1054 struct pagemapread *pm = walk->private; 1055 spinlock_t *ptl; 1056 pte_t *pte; 1057 int err = 0; 1058 1059 /* find the first VMA at or above 'addr' */ 1060 vma = find_vma(walk->mm, addr); 1061 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 1062 int pmd_flags2; 1063 1064 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd)) 1065 pmd_flags2 = __PM_SOFT_DIRTY; 1066 else 1067 pmd_flags2 = 0; 1068 1069 for (; addr != end; addr += PAGE_SIZE) { 1070 unsigned long offset; 1071 pagemap_entry_t pme; 1072 1073 offset = (addr & ~PAGEMAP_WALK_MASK) >> 1074 PAGE_SHIFT; 1075 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2); 1076 err = add_to_pagemap(addr, &pme, pm); 1077 if (err) 1078 break; 1079 } 1080 spin_unlock(ptl); 1081 return err; 1082 } 1083 1084 if (pmd_trans_unstable(pmd)) 1085 return 0; 1086 1087 while (1) { 1088 /* End of address space hole, which we mark as non-present. */ 1089 unsigned long hole_end; 1090 1091 if (vma) 1092 hole_end = min(end, vma->vm_start); 1093 else 1094 hole_end = end; 1095 1096 for (; addr < hole_end; addr += PAGE_SIZE) { 1097 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 1098 1099 err = add_to_pagemap(addr, &pme, pm); 1100 if (err) 1101 return err; 1102 } 1103 1104 if (!vma || vma->vm_start >= end) 1105 break; 1106 /* 1107 * We can't possibly be in a hugetlb VMA. In general, 1108 * for a mm_walk with a pmd_entry and a hugetlb_entry, 1109 * the pmd_entry can only be called on addresses in a 1110 * hugetlb if the walk starts in a non-hugetlb VMA and 1111 * spans a hugepage VMA. Since pagemap_read walks are 1112 * PMD-sized and PMD-aligned, this will never be true. 1113 */ 1114 BUG_ON(is_vm_hugetlb_page(vma)); 1115 1116 /* Addresses in the VMA. */ 1117 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1118 pagemap_entry_t pme; 1119 pte = pte_offset_map(pmd, addr); 1120 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte); 1121 pte_unmap(pte); 1122 err = add_to_pagemap(addr, &pme, pm); 1123 if (err) 1124 return err; 1125 } 1126 1127 if (addr == end) 1128 break; 1129 1130 vma = find_vma(walk->mm, addr); 1131 } 1132 1133 cond_resched(); 1134 1135 return err; 1136 } 1137 1138 #ifdef CONFIG_HUGETLB_PAGE 1139 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1140 pte_t pte, int offset, int flags2) 1141 { 1142 if (pte_present(pte)) 1143 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) | 1144 PM_STATUS2(pm->v2, flags2) | 1145 PM_PRESENT); 1146 else 1147 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | 1148 PM_STATUS2(pm->v2, flags2)); 1149 } 1150 1151 /* This function walks within one hugetlb entry in the single call */ 1152 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 1153 unsigned long addr, unsigned long end, 1154 struct mm_walk *walk) 1155 { 1156 struct pagemapread *pm = walk->private; 1157 struct vm_area_struct *vma; 1158 int err = 0; 1159 int flags2; 1160 pagemap_entry_t pme; 1161 1162 vma = find_vma(walk->mm, addr); 1163 WARN_ON_ONCE(!vma); 1164 1165 if (vma && (vma->vm_flags & VM_SOFTDIRTY)) 1166 flags2 = __PM_SOFT_DIRTY; 1167 else 1168 flags2 = 0; 1169 1170 for (; addr != end; addr += PAGE_SIZE) { 1171 int offset = (addr & ~hmask) >> PAGE_SHIFT; 1172 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2); 1173 err = add_to_pagemap(addr, &pme, pm); 1174 if (err) 1175 return err; 1176 } 1177 1178 cond_resched(); 1179 1180 return err; 1181 } 1182 #endif /* HUGETLB_PAGE */ 1183 1184 /* 1185 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1186 * 1187 * For each page in the address space, this file contains one 64-bit entry 1188 * consisting of the following: 1189 * 1190 * Bits 0-54 page frame number (PFN) if present 1191 * Bits 0-4 swap type if swapped 1192 * Bits 5-54 swap offset if swapped 1193 * Bits 55-60 page shift (page size = 1<<page shift) 1194 * Bit 61 page is file-page or shared-anon 1195 * Bit 62 page swapped 1196 * Bit 63 page present 1197 * 1198 * If the page is not present but in swap, then the PFN contains an 1199 * encoding of the swap file number and the page's offset into the 1200 * swap. Unmapped pages return a null PFN. This allows determining 1201 * precisely which pages are mapped (or in swap) and comparing mapped 1202 * pages between processes. 1203 * 1204 * Efficient users of this interface will use /proc/pid/maps to 1205 * determine which areas of memory are actually mapped and llseek to 1206 * skip over unmapped regions. 1207 */ 1208 static ssize_t pagemap_read(struct file *file, char __user *buf, 1209 size_t count, loff_t *ppos) 1210 { 1211 struct task_struct *task = get_proc_task(file_inode(file)); 1212 struct mm_struct *mm; 1213 struct pagemapread pm; 1214 int ret = -ESRCH; 1215 struct mm_walk pagemap_walk = {}; 1216 unsigned long src; 1217 unsigned long svpfn; 1218 unsigned long start_vaddr; 1219 unsigned long end_vaddr; 1220 int copied = 0; 1221 1222 if (!task) 1223 goto out; 1224 1225 ret = -EINVAL; 1226 /* file position must be aligned */ 1227 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1228 goto out_task; 1229 1230 ret = 0; 1231 if (!count) 1232 goto out_task; 1233 1234 pm.v2 = soft_dirty_cleared; 1235 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1236 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); 1237 ret = -ENOMEM; 1238 if (!pm.buffer) 1239 goto out_task; 1240 1241 mm = mm_access(task, PTRACE_MODE_READ); 1242 ret = PTR_ERR(mm); 1243 if (!mm || IS_ERR(mm)) 1244 goto out_free; 1245 1246 pagemap_walk.pmd_entry = pagemap_pte_range; 1247 pagemap_walk.pte_hole = pagemap_pte_hole; 1248 #ifdef CONFIG_HUGETLB_PAGE 1249 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1250 #endif 1251 pagemap_walk.mm = mm; 1252 pagemap_walk.private = ± 1253 1254 src = *ppos; 1255 svpfn = src / PM_ENTRY_BYTES; 1256 start_vaddr = svpfn << PAGE_SHIFT; 1257 end_vaddr = TASK_SIZE_OF(task); 1258 1259 /* watch out for wraparound */ 1260 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 1261 start_vaddr = end_vaddr; 1262 1263 /* 1264 * The odds are that this will stop walking way 1265 * before end_vaddr, because the length of the 1266 * user buffer is tracked in "pm", and the walk 1267 * will stop when we hit the end of the buffer. 1268 */ 1269 ret = 0; 1270 while (count && (start_vaddr < end_vaddr)) { 1271 int len; 1272 unsigned long end; 1273 1274 pm.pos = 0; 1275 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1276 /* overflow ? */ 1277 if (end < start_vaddr || end > end_vaddr) 1278 end = end_vaddr; 1279 down_read(&mm->mmap_sem); 1280 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1281 up_read(&mm->mmap_sem); 1282 start_vaddr = end; 1283 1284 len = min(count, PM_ENTRY_BYTES * pm.pos); 1285 if (copy_to_user(buf, pm.buffer, len)) { 1286 ret = -EFAULT; 1287 goto out_mm; 1288 } 1289 copied += len; 1290 buf += len; 1291 count -= len; 1292 } 1293 *ppos += copied; 1294 if (!ret || ret == PM_END_OF_BUFFER) 1295 ret = copied; 1296 1297 out_mm: 1298 mmput(mm); 1299 out_free: 1300 kfree(pm.buffer); 1301 out_task: 1302 put_task_struct(task); 1303 out: 1304 return ret; 1305 } 1306 1307 static int pagemap_open(struct inode *inode, struct file *file) 1308 { 1309 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about " 1310 "to stop being page-shift some time soon. See the " 1311 "linux/Documentation/vm/pagemap.txt for details.\n"); 1312 return 0; 1313 } 1314 1315 const struct file_operations proc_pagemap_operations = { 1316 .llseek = mem_lseek, /* borrow this */ 1317 .read = pagemap_read, 1318 .open = pagemap_open, 1319 }; 1320 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1321 1322 #ifdef CONFIG_NUMA 1323 1324 struct numa_maps { 1325 struct vm_area_struct *vma; 1326 unsigned long pages; 1327 unsigned long anon; 1328 unsigned long active; 1329 unsigned long writeback; 1330 unsigned long mapcount_max; 1331 unsigned long dirty; 1332 unsigned long swapcache; 1333 unsigned long node[MAX_NUMNODES]; 1334 }; 1335 1336 struct numa_maps_private { 1337 struct proc_maps_private proc_maps; 1338 struct numa_maps md; 1339 }; 1340 1341 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1342 unsigned long nr_pages) 1343 { 1344 int count = page_mapcount(page); 1345 1346 md->pages += nr_pages; 1347 if (pte_dirty || PageDirty(page)) 1348 md->dirty += nr_pages; 1349 1350 if (PageSwapCache(page)) 1351 md->swapcache += nr_pages; 1352 1353 if (PageActive(page) || PageUnevictable(page)) 1354 md->active += nr_pages; 1355 1356 if (PageWriteback(page)) 1357 md->writeback += nr_pages; 1358 1359 if (PageAnon(page)) 1360 md->anon += nr_pages; 1361 1362 if (count > md->mapcount_max) 1363 md->mapcount_max = count; 1364 1365 md->node[page_to_nid(page)] += nr_pages; 1366 } 1367 1368 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1369 unsigned long addr) 1370 { 1371 struct page *page; 1372 int nid; 1373 1374 if (!pte_present(pte)) 1375 return NULL; 1376 1377 page = vm_normal_page(vma, addr, pte); 1378 if (!page) 1379 return NULL; 1380 1381 if (PageReserved(page)) 1382 return NULL; 1383 1384 nid = page_to_nid(page); 1385 if (!node_isset(nid, node_states[N_MEMORY])) 1386 return NULL; 1387 1388 return page; 1389 } 1390 1391 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1392 unsigned long end, struct mm_walk *walk) 1393 { 1394 struct numa_maps *md; 1395 spinlock_t *ptl; 1396 pte_t *orig_pte; 1397 pte_t *pte; 1398 1399 md = walk->private; 1400 1401 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) { 1402 pte_t huge_pte = *(pte_t *)pmd; 1403 struct page *page; 1404 1405 page = can_gather_numa_stats(huge_pte, md->vma, addr); 1406 if (page) 1407 gather_stats(page, md, pte_dirty(huge_pte), 1408 HPAGE_PMD_SIZE/PAGE_SIZE); 1409 spin_unlock(ptl); 1410 return 0; 1411 } 1412 1413 if (pmd_trans_unstable(pmd)) 1414 return 0; 1415 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1416 do { 1417 struct page *page = can_gather_numa_stats(*pte, md->vma, addr); 1418 if (!page) 1419 continue; 1420 gather_stats(page, md, pte_dirty(*pte), 1); 1421 1422 } while (pte++, addr += PAGE_SIZE, addr != end); 1423 pte_unmap_unlock(orig_pte, ptl); 1424 return 0; 1425 } 1426 #ifdef CONFIG_HUGETLB_PAGE 1427 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1428 unsigned long addr, unsigned long end, struct mm_walk *walk) 1429 { 1430 struct numa_maps *md; 1431 struct page *page; 1432 1433 if (!pte_present(*pte)) 1434 return 0; 1435 1436 page = pte_page(*pte); 1437 if (!page) 1438 return 0; 1439 1440 md = walk->private; 1441 gather_stats(page, md, pte_dirty(*pte), 1); 1442 return 0; 1443 } 1444 1445 #else 1446 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1447 unsigned long addr, unsigned long end, struct mm_walk *walk) 1448 { 1449 return 0; 1450 } 1451 #endif 1452 1453 /* 1454 * Display pages allocated per node and memory policy via /proc. 1455 */ 1456 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1457 { 1458 struct numa_maps_private *numa_priv = m->private; 1459 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1460 struct vm_area_struct *vma = v; 1461 struct numa_maps *md = &numa_priv->md; 1462 struct file *file = vma->vm_file; 1463 struct mm_struct *mm = vma->vm_mm; 1464 struct mm_walk walk = {}; 1465 struct mempolicy *pol; 1466 char buffer[64]; 1467 int nid; 1468 1469 if (!mm) 1470 return 0; 1471 1472 /* Ensure we start with an empty set of numa_maps statistics. */ 1473 memset(md, 0, sizeof(*md)); 1474 1475 md->vma = vma; 1476 1477 walk.hugetlb_entry = gather_hugetbl_stats; 1478 walk.pmd_entry = gather_pte_stats; 1479 walk.private = md; 1480 walk.mm = mm; 1481 1482 pol = __get_vma_policy(vma, vma->vm_start); 1483 if (pol) { 1484 mpol_to_str(buffer, sizeof(buffer), pol); 1485 mpol_cond_put(pol); 1486 } else { 1487 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1488 } 1489 1490 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1491 1492 if (file) { 1493 seq_puts(m, " file="); 1494 seq_path(m, &file->f_path, "\n\t= "); 1495 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1496 seq_puts(m, " heap"); 1497 } else { 1498 pid_t tid = pid_of_stack(proc_priv, vma, is_pid); 1499 if (tid != 0) { 1500 /* 1501 * Thread stack in /proc/PID/task/TID/maps or 1502 * the main process stack. 1503 */ 1504 if (!is_pid || (vma->vm_start <= mm->start_stack && 1505 vma->vm_end >= mm->start_stack)) 1506 seq_puts(m, " stack"); 1507 else 1508 seq_printf(m, " stack:%d", tid); 1509 } 1510 } 1511 1512 if (is_vm_hugetlb_page(vma)) 1513 seq_puts(m, " huge"); 1514 1515 walk_page_range(vma->vm_start, vma->vm_end, &walk); 1516 1517 if (!md->pages) 1518 goto out; 1519 1520 if (md->anon) 1521 seq_printf(m, " anon=%lu", md->anon); 1522 1523 if (md->dirty) 1524 seq_printf(m, " dirty=%lu", md->dirty); 1525 1526 if (md->pages != md->anon && md->pages != md->dirty) 1527 seq_printf(m, " mapped=%lu", md->pages); 1528 1529 if (md->mapcount_max > 1) 1530 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1531 1532 if (md->swapcache) 1533 seq_printf(m, " swapcache=%lu", md->swapcache); 1534 1535 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1536 seq_printf(m, " active=%lu", md->active); 1537 1538 if (md->writeback) 1539 seq_printf(m, " writeback=%lu", md->writeback); 1540 1541 for_each_node_state(nid, N_MEMORY) 1542 if (md->node[nid]) 1543 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1544 out: 1545 seq_putc(m, '\n'); 1546 m_cache_vma(m, vma); 1547 return 0; 1548 } 1549 1550 static int show_pid_numa_map(struct seq_file *m, void *v) 1551 { 1552 return show_numa_map(m, v, 1); 1553 } 1554 1555 static int show_tid_numa_map(struct seq_file *m, void *v) 1556 { 1557 return show_numa_map(m, v, 0); 1558 } 1559 1560 static const struct seq_operations proc_pid_numa_maps_op = { 1561 .start = m_start, 1562 .next = m_next, 1563 .stop = m_stop, 1564 .show = show_pid_numa_map, 1565 }; 1566 1567 static const struct seq_operations proc_tid_numa_maps_op = { 1568 .start = m_start, 1569 .next = m_next, 1570 .stop = m_stop, 1571 .show = show_tid_numa_map, 1572 }; 1573 1574 static int numa_maps_open(struct inode *inode, struct file *file, 1575 const struct seq_operations *ops) 1576 { 1577 return proc_maps_open(inode, file, ops, 1578 sizeof(struct numa_maps_private)); 1579 } 1580 1581 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1582 { 1583 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1584 } 1585 1586 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1587 { 1588 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1589 } 1590 1591 const struct file_operations proc_pid_numa_maps_operations = { 1592 .open = pid_numa_maps_open, 1593 .read = seq_read, 1594 .llseek = seq_lseek, 1595 .release = proc_map_release, 1596 }; 1597 1598 const struct file_operations proc_tid_numa_maps_operations = { 1599 .open = tid_numa_maps_open, 1600 .read = seq_read, 1601 .llseek = seq_lseek, 1602 .release = proc_map_release, 1603 }; 1604 #endif /* CONFIG_NUMA */ 1605