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 static void smaps_account(struct mem_size_stats *mss, struct page *page, 451 unsigned long size, bool young, bool dirty) 452 { 453 int mapcount; 454 455 if (PageAnon(page)) 456 mss->anonymous += size; 457 458 mss->resident += size; 459 /* Accumulate the size in pages that have been accessed. */ 460 if (young || PageReferenced(page)) 461 mss->referenced += size; 462 mapcount = page_mapcount(page); 463 if (mapcount >= 2) { 464 u64 pss_delta; 465 466 if (dirty || PageDirty(page)) 467 mss->shared_dirty += size; 468 else 469 mss->shared_clean += size; 470 pss_delta = (u64)size << PSS_SHIFT; 471 do_div(pss_delta, mapcount); 472 mss->pss += pss_delta; 473 } else { 474 if (dirty || PageDirty(page)) 475 mss->private_dirty += size; 476 else 477 mss->private_clean += size; 478 mss->pss += (u64)size << PSS_SHIFT; 479 } 480 } 481 482 static void smaps_pte_entry(pte_t *pte, unsigned long addr, 483 struct mm_walk *walk) 484 { 485 struct mem_size_stats *mss = walk->private; 486 struct vm_area_struct *vma = mss->vma; 487 pgoff_t pgoff = linear_page_index(vma, addr); 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 } else if (pte_file(*pte)) { 500 if (pte_to_pgoff(*pte) != pgoff) 501 mss->nonlinear += PAGE_SIZE; 502 } 503 504 if (!page) 505 return; 506 507 if (page->index != pgoff) 508 mss->nonlinear += PAGE_SIZE; 509 510 smaps_account(mss, page, PAGE_SIZE, pte_young(*pte), pte_dirty(*pte)); 511 } 512 513 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 514 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 515 struct mm_walk *walk) 516 { 517 struct mem_size_stats *mss = walk->private; 518 struct vm_area_struct *vma = mss->vma; 519 struct page *page; 520 521 /* FOLL_DUMP will return -EFAULT on huge zero page */ 522 page = follow_trans_huge_pmd(vma, addr, pmd, FOLL_DUMP); 523 if (IS_ERR_OR_NULL(page)) 524 return; 525 mss->anonymous_thp += HPAGE_PMD_SIZE; 526 smaps_account(mss, page, HPAGE_PMD_SIZE, 527 pmd_young(*pmd), pmd_dirty(*pmd)); 528 } 529 #else 530 static void smaps_pmd_entry(pmd_t *pmd, unsigned long addr, 531 struct mm_walk *walk) 532 { 533 } 534 #endif 535 536 static int smaps_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 537 struct mm_walk *walk) 538 { 539 struct mem_size_stats *mss = walk->private; 540 struct vm_area_struct *vma = mss->vma; 541 pte_t *pte; 542 spinlock_t *ptl; 543 544 if (pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 545 smaps_pmd_entry(pmd, addr, walk); 546 spin_unlock(ptl); 547 return 0; 548 } 549 550 if (pmd_trans_unstable(pmd)) 551 return 0; 552 /* 553 * The mmap_sem held all the way back in m_start() is what 554 * keeps khugepaged out of here and from collapsing things 555 * in here. 556 */ 557 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 558 for (; addr != end; pte++, addr += PAGE_SIZE) 559 smaps_pte_entry(pte, addr, walk); 560 pte_unmap_unlock(pte - 1, ptl); 561 cond_resched(); 562 return 0; 563 } 564 565 static void show_smap_vma_flags(struct seq_file *m, struct vm_area_struct *vma) 566 { 567 /* 568 * Don't forget to update Documentation/ on changes. 569 */ 570 static const char mnemonics[BITS_PER_LONG][2] = { 571 /* 572 * In case if we meet a flag we don't know about. 573 */ 574 [0 ... (BITS_PER_LONG-1)] = "??", 575 576 [ilog2(VM_READ)] = "rd", 577 [ilog2(VM_WRITE)] = "wr", 578 [ilog2(VM_EXEC)] = "ex", 579 [ilog2(VM_SHARED)] = "sh", 580 [ilog2(VM_MAYREAD)] = "mr", 581 [ilog2(VM_MAYWRITE)] = "mw", 582 [ilog2(VM_MAYEXEC)] = "me", 583 [ilog2(VM_MAYSHARE)] = "ms", 584 [ilog2(VM_GROWSDOWN)] = "gd", 585 [ilog2(VM_PFNMAP)] = "pf", 586 [ilog2(VM_DENYWRITE)] = "dw", 587 #ifdef CONFIG_X86_INTEL_MPX 588 [ilog2(VM_MPX)] = "mp", 589 #endif 590 [ilog2(VM_LOCKED)] = "lo", 591 [ilog2(VM_IO)] = "io", 592 [ilog2(VM_SEQ_READ)] = "sr", 593 [ilog2(VM_RAND_READ)] = "rr", 594 [ilog2(VM_DONTCOPY)] = "dc", 595 [ilog2(VM_DONTEXPAND)] = "de", 596 [ilog2(VM_ACCOUNT)] = "ac", 597 [ilog2(VM_NORESERVE)] = "nr", 598 [ilog2(VM_HUGETLB)] = "ht", 599 [ilog2(VM_NONLINEAR)] = "nl", 600 [ilog2(VM_ARCH_1)] = "ar", 601 [ilog2(VM_DONTDUMP)] = "dd", 602 #ifdef CONFIG_MEM_SOFT_DIRTY 603 [ilog2(VM_SOFTDIRTY)] = "sd", 604 #endif 605 [ilog2(VM_MIXEDMAP)] = "mm", 606 [ilog2(VM_HUGEPAGE)] = "hg", 607 [ilog2(VM_NOHUGEPAGE)] = "nh", 608 [ilog2(VM_MERGEABLE)] = "mg", 609 }; 610 size_t i; 611 612 seq_puts(m, "VmFlags: "); 613 for (i = 0; i < BITS_PER_LONG; i++) { 614 if (vma->vm_flags & (1UL << i)) { 615 seq_printf(m, "%c%c ", 616 mnemonics[i][0], mnemonics[i][1]); 617 } 618 } 619 seq_putc(m, '\n'); 620 } 621 622 static int show_smap(struct seq_file *m, void *v, int is_pid) 623 { 624 struct vm_area_struct *vma = v; 625 struct mem_size_stats mss; 626 struct mm_walk smaps_walk = { 627 .pmd_entry = smaps_pte_range, 628 .mm = vma->vm_mm, 629 .private = &mss, 630 }; 631 632 memset(&mss, 0, sizeof mss); 633 mss.vma = vma; 634 /* mmap_sem is held in m_start */ 635 if (vma->vm_mm && !is_vm_hugetlb_page(vma)) 636 walk_page_range(vma->vm_start, vma->vm_end, &smaps_walk); 637 638 show_map_vma(m, vma, is_pid); 639 640 seq_printf(m, 641 "Size: %8lu kB\n" 642 "Rss: %8lu kB\n" 643 "Pss: %8lu kB\n" 644 "Shared_Clean: %8lu kB\n" 645 "Shared_Dirty: %8lu kB\n" 646 "Private_Clean: %8lu kB\n" 647 "Private_Dirty: %8lu kB\n" 648 "Referenced: %8lu kB\n" 649 "Anonymous: %8lu kB\n" 650 "AnonHugePages: %8lu kB\n" 651 "Swap: %8lu kB\n" 652 "KernelPageSize: %8lu kB\n" 653 "MMUPageSize: %8lu kB\n" 654 "Locked: %8lu kB\n", 655 (vma->vm_end - vma->vm_start) >> 10, 656 mss.resident >> 10, 657 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)), 658 mss.shared_clean >> 10, 659 mss.shared_dirty >> 10, 660 mss.private_clean >> 10, 661 mss.private_dirty >> 10, 662 mss.referenced >> 10, 663 mss.anonymous >> 10, 664 mss.anonymous_thp >> 10, 665 mss.swap >> 10, 666 vma_kernel_pagesize(vma) >> 10, 667 vma_mmu_pagesize(vma) >> 10, 668 (vma->vm_flags & VM_LOCKED) ? 669 (unsigned long)(mss.pss >> (10 + PSS_SHIFT)) : 0); 670 671 if (vma->vm_flags & VM_NONLINEAR) 672 seq_printf(m, "Nonlinear: %8lu kB\n", 673 mss.nonlinear >> 10); 674 675 show_smap_vma_flags(m, vma); 676 m_cache_vma(m, vma); 677 return 0; 678 } 679 680 static int show_pid_smap(struct seq_file *m, void *v) 681 { 682 return show_smap(m, v, 1); 683 } 684 685 static int show_tid_smap(struct seq_file *m, void *v) 686 { 687 return show_smap(m, v, 0); 688 } 689 690 static const struct seq_operations proc_pid_smaps_op = { 691 .start = m_start, 692 .next = m_next, 693 .stop = m_stop, 694 .show = show_pid_smap 695 }; 696 697 static const struct seq_operations proc_tid_smaps_op = { 698 .start = m_start, 699 .next = m_next, 700 .stop = m_stop, 701 .show = show_tid_smap 702 }; 703 704 static int pid_smaps_open(struct inode *inode, struct file *file) 705 { 706 return do_maps_open(inode, file, &proc_pid_smaps_op); 707 } 708 709 static int tid_smaps_open(struct inode *inode, struct file *file) 710 { 711 return do_maps_open(inode, file, &proc_tid_smaps_op); 712 } 713 714 const struct file_operations proc_pid_smaps_operations = { 715 .open = pid_smaps_open, 716 .read = seq_read, 717 .llseek = seq_lseek, 718 .release = proc_map_release, 719 }; 720 721 const struct file_operations proc_tid_smaps_operations = { 722 .open = tid_smaps_open, 723 .read = seq_read, 724 .llseek = seq_lseek, 725 .release = proc_map_release, 726 }; 727 728 /* 729 * We do not want to have constant page-shift bits sitting in 730 * pagemap entries and are about to reuse them some time soon. 731 * 732 * Here's the "migration strategy": 733 * 1. when the system boots these bits remain what they are, 734 * but a warning about future change is printed in log; 735 * 2. once anyone clears soft-dirty bits via clear_refs file, 736 * these flag is set to denote, that user is aware of the 737 * new API and those page-shift bits change their meaning. 738 * The respective warning is printed in dmesg; 739 * 3. In a couple of releases we will remove all the mentions 740 * of page-shift in pagemap entries. 741 */ 742 743 static bool soft_dirty_cleared __read_mostly; 744 745 enum clear_refs_types { 746 CLEAR_REFS_ALL = 1, 747 CLEAR_REFS_ANON, 748 CLEAR_REFS_MAPPED, 749 CLEAR_REFS_SOFT_DIRTY, 750 CLEAR_REFS_LAST, 751 }; 752 753 struct clear_refs_private { 754 struct vm_area_struct *vma; 755 enum clear_refs_types type; 756 }; 757 758 static inline void clear_soft_dirty(struct vm_area_struct *vma, 759 unsigned long addr, pte_t *pte) 760 { 761 #ifdef CONFIG_MEM_SOFT_DIRTY 762 /* 763 * The soft-dirty tracker uses #PF-s to catch writes 764 * to pages, so write-protect the pte as well. See the 765 * Documentation/vm/soft-dirty.txt for full description 766 * of how soft-dirty works. 767 */ 768 pte_t ptent = *pte; 769 770 if (pte_present(ptent)) { 771 ptent = pte_wrprotect(ptent); 772 ptent = pte_clear_flags(ptent, _PAGE_SOFT_DIRTY); 773 } else if (is_swap_pte(ptent)) { 774 ptent = pte_swp_clear_soft_dirty(ptent); 775 } else if (pte_file(ptent)) { 776 ptent = pte_file_clear_soft_dirty(ptent); 777 } 778 779 set_pte_at(vma->vm_mm, addr, pte, ptent); 780 #endif 781 } 782 783 static int clear_refs_pte_range(pmd_t *pmd, unsigned long addr, 784 unsigned long end, struct mm_walk *walk) 785 { 786 struct clear_refs_private *cp = walk->private; 787 struct vm_area_struct *vma = cp->vma; 788 pte_t *pte, ptent; 789 spinlock_t *ptl; 790 struct page *page; 791 792 split_huge_page_pmd(vma, addr, pmd); 793 if (pmd_trans_unstable(pmd)) 794 return 0; 795 796 pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); 797 for (; addr != end; pte++, addr += PAGE_SIZE) { 798 ptent = *pte; 799 800 if (cp->type == CLEAR_REFS_SOFT_DIRTY) { 801 clear_soft_dirty(vma, addr, pte); 802 continue; 803 } 804 805 if (!pte_present(ptent)) 806 continue; 807 808 page = vm_normal_page(vma, addr, ptent); 809 if (!page) 810 continue; 811 812 /* Clear accessed and referenced bits. */ 813 ptep_test_and_clear_young(vma, addr, pte); 814 ClearPageReferenced(page); 815 } 816 pte_unmap_unlock(pte - 1, ptl); 817 cond_resched(); 818 return 0; 819 } 820 821 static ssize_t clear_refs_write(struct file *file, const char __user *buf, 822 size_t count, loff_t *ppos) 823 { 824 struct task_struct *task; 825 char buffer[PROC_NUMBUF]; 826 struct mm_struct *mm; 827 struct vm_area_struct *vma; 828 enum clear_refs_types type; 829 int itype; 830 int rv; 831 832 memset(buffer, 0, sizeof(buffer)); 833 if (count > sizeof(buffer) - 1) 834 count = sizeof(buffer) - 1; 835 if (copy_from_user(buffer, buf, count)) 836 return -EFAULT; 837 rv = kstrtoint(strstrip(buffer), 10, &itype); 838 if (rv < 0) 839 return rv; 840 type = (enum clear_refs_types)itype; 841 if (type < CLEAR_REFS_ALL || type >= CLEAR_REFS_LAST) 842 return -EINVAL; 843 844 if (type == CLEAR_REFS_SOFT_DIRTY) { 845 soft_dirty_cleared = true; 846 pr_warn_once("The pagemap bits 55-60 has changed their meaning!" 847 " See the linux/Documentation/vm/pagemap.txt for " 848 "details.\n"); 849 } 850 851 task = get_proc_task(file_inode(file)); 852 if (!task) 853 return -ESRCH; 854 mm = get_task_mm(task); 855 if (mm) { 856 struct clear_refs_private cp = { 857 .type = type, 858 }; 859 struct mm_walk clear_refs_walk = { 860 .pmd_entry = clear_refs_pte_range, 861 .mm = mm, 862 .private = &cp, 863 }; 864 down_read(&mm->mmap_sem); 865 if (type == CLEAR_REFS_SOFT_DIRTY) { 866 for (vma = mm->mmap; vma; vma = vma->vm_next) { 867 if (!(vma->vm_flags & VM_SOFTDIRTY)) 868 continue; 869 up_read(&mm->mmap_sem); 870 down_write(&mm->mmap_sem); 871 for (vma = mm->mmap; vma; vma = vma->vm_next) { 872 vma->vm_flags &= ~VM_SOFTDIRTY; 873 vma_set_page_prot(vma); 874 } 875 downgrade_write(&mm->mmap_sem); 876 break; 877 } 878 mmu_notifier_invalidate_range_start(mm, 0, -1); 879 } 880 for (vma = mm->mmap; vma; vma = vma->vm_next) { 881 cp.vma = vma; 882 if (is_vm_hugetlb_page(vma)) 883 continue; 884 /* 885 * Writing 1 to /proc/pid/clear_refs affects all pages. 886 * 887 * Writing 2 to /proc/pid/clear_refs only affects 888 * Anonymous pages. 889 * 890 * Writing 3 to /proc/pid/clear_refs only affects file 891 * mapped pages. 892 * 893 * Writing 4 to /proc/pid/clear_refs affects all pages. 894 */ 895 if (type == CLEAR_REFS_ANON && vma->vm_file) 896 continue; 897 if (type == CLEAR_REFS_MAPPED && !vma->vm_file) 898 continue; 899 walk_page_range(vma->vm_start, vma->vm_end, 900 &clear_refs_walk); 901 } 902 if (type == CLEAR_REFS_SOFT_DIRTY) 903 mmu_notifier_invalidate_range_end(mm, 0, -1); 904 flush_tlb_mm(mm); 905 up_read(&mm->mmap_sem); 906 mmput(mm); 907 } 908 put_task_struct(task); 909 910 return count; 911 } 912 913 const struct file_operations proc_clear_refs_operations = { 914 .write = clear_refs_write, 915 .llseek = noop_llseek, 916 }; 917 918 typedef struct { 919 u64 pme; 920 } pagemap_entry_t; 921 922 struct pagemapread { 923 int pos, len; /* units: PM_ENTRY_BYTES, not bytes */ 924 pagemap_entry_t *buffer; 925 bool v2; 926 }; 927 928 #define PAGEMAP_WALK_SIZE (PMD_SIZE) 929 #define PAGEMAP_WALK_MASK (PMD_MASK) 930 931 #define PM_ENTRY_BYTES sizeof(pagemap_entry_t) 932 #define PM_STATUS_BITS 3 933 #define PM_STATUS_OFFSET (64 - PM_STATUS_BITS) 934 #define PM_STATUS_MASK (((1LL << PM_STATUS_BITS) - 1) << PM_STATUS_OFFSET) 935 #define PM_STATUS(nr) (((nr) << PM_STATUS_OFFSET) & PM_STATUS_MASK) 936 #define PM_PSHIFT_BITS 6 937 #define PM_PSHIFT_OFFSET (PM_STATUS_OFFSET - PM_PSHIFT_BITS) 938 #define PM_PSHIFT_MASK (((1LL << PM_PSHIFT_BITS) - 1) << PM_PSHIFT_OFFSET) 939 #define __PM_PSHIFT(x) (((u64) (x) << PM_PSHIFT_OFFSET) & PM_PSHIFT_MASK) 940 #define PM_PFRAME_MASK ((1LL << PM_PSHIFT_OFFSET) - 1) 941 #define PM_PFRAME(x) ((x) & PM_PFRAME_MASK) 942 /* in "new" pagemap pshift bits are occupied with more status bits */ 943 #define PM_STATUS2(v2, x) (__PM_PSHIFT(v2 ? x : PAGE_SHIFT)) 944 945 #define __PM_SOFT_DIRTY (1LL) 946 #define PM_PRESENT PM_STATUS(4LL) 947 #define PM_SWAP PM_STATUS(2LL) 948 #define PM_FILE PM_STATUS(1LL) 949 #define PM_NOT_PRESENT(v2) PM_STATUS2(v2, 0) 950 #define PM_END_OF_BUFFER 1 951 952 static inline pagemap_entry_t make_pme(u64 val) 953 { 954 return (pagemap_entry_t) { .pme = val }; 955 } 956 957 static int add_to_pagemap(unsigned long addr, pagemap_entry_t *pme, 958 struct pagemapread *pm) 959 { 960 pm->buffer[pm->pos++] = *pme; 961 if (pm->pos >= pm->len) 962 return PM_END_OF_BUFFER; 963 return 0; 964 } 965 966 static int pagemap_pte_hole(unsigned long start, unsigned long end, 967 struct mm_walk *walk) 968 { 969 struct pagemapread *pm = walk->private; 970 unsigned long addr = start; 971 int err = 0; 972 973 while (addr < end) { 974 struct vm_area_struct *vma = find_vma(walk->mm, addr); 975 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 976 /* End of address space hole, which we mark as non-present. */ 977 unsigned long hole_end; 978 979 if (vma) 980 hole_end = min(end, vma->vm_start); 981 else 982 hole_end = end; 983 984 for (; addr < hole_end; addr += PAGE_SIZE) { 985 err = add_to_pagemap(addr, &pme, pm); 986 if (err) 987 goto out; 988 } 989 990 if (!vma) 991 break; 992 993 /* Addresses in the VMA. */ 994 if (vma->vm_flags & VM_SOFTDIRTY) 995 pme.pme |= PM_STATUS2(pm->v2, __PM_SOFT_DIRTY); 996 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 997 err = add_to_pagemap(addr, &pme, pm); 998 if (err) 999 goto out; 1000 } 1001 } 1002 out: 1003 return err; 1004 } 1005 1006 static void pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1007 struct vm_area_struct *vma, unsigned long addr, pte_t pte) 1008 { 1009 u64 frame, flags; 1010 struct page *page = NULL; 1011 int flags2 = 0; 1012 1013 if (pte_present(pte)) { 1014 frame = pte_pfn(pte); 1015 flags = PM_PRESENT; 1016 page = vm_normal_page(vma, addr, pte); 1017 if (pte_soft_dirty(pte)) 1018 flags2 |= __PM_SOFT_DIRTY; 1019 } else if (is_swap_pte(pte)) { 1020 swp_entry_t entry; 1021 if (pte_swp_soft_dirty(pte)) 1022 flags2 |= __PM_SOFT_DIRTY; 1023 entry = pte_to_swp_entry(pte); 1024 frame = swp_type(entry) | 1025 (swp_offset(entry) << MAX_SWAPFILES_SHIFT); 1026 flags = PM_SWAP; 1027 if (is_migration_entry(entry)) 1028 page = migration_entry_to_page(entry); 1029 } else { 1030 if (vma->vm_flags & VM_SOFTDIRTY) 1031 flags2 |= __PM_SOFT_DIRTY; 1032 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, flags2)); 1033 return; 1034 } 1035 1036 if (page && !PageAnon(page)) 1037 flags |= PM_FILE; 1038 if ((vma->vm_flags & VM_SOFTDIRTY)) 1039 flags2 |= __PM_SOFT_DIRTY; 1040 1041 *pme = make_pme(PM_PFRAME(frame) | PM_STATUS2(pm->v2, flags2) | flags); 1042 } 1043 1044 #ifdef CONFIG_TRANSPARENT_HUGEPAGE 1045 static void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1046 pmd_t pmd, int offset, int pmd_flags2) 1047 { 1048 /* 1049 * Currently pmd for thp is always present because thp can not be 1050 * swapped-out, migrated, or HWPOISONed (split in such cases instead.) 1051 * This if-check is just to prepare for future implementation. 1052 */ 1053 if (pmd_present(pmd)) 1054 *pme = make_pme(PM_PFRAME(pmd_pfn(pmd) + offset) 1055 | PM_STATUS2(pm->v2, pmd_flags2) | PM_PRESENT); 1056 else 1057 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | PM_STATUS2(pm->v2, pmd_flags2)); 1058 } 1059 #else 1060 static inline void thp_pmd_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1061 pmd_t pmd, int offset, int pmd_flags2) 1062 { 1063 } 1064 #endif 1065 1066 static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, 1067 struct mm_walk *walk) 1068 { 1069 struct vm_area_struct *vma; 1070 struct pagemapread *pm = walk->private; 1071 spinlock_t *ptl; 1072 pte_t *pte; 1073 int err = 0; 1074 1075 /* find the first VMA at or above 'addr' */ 1076 vma = find_vma(walk->mm, addr); 1077 if (vma && pmd_trans_huge_lock(pmd, vma, &ptl) == 1) { 1078 int pmd_flags2; 1079 1080 if ((vma->vm_flags & VM_SOFTDIRTY) || pmd_soft_dirty(*pmd)) 1081 pmd_flags2 = __PM_SOFT_DIRTY; 1082 else 1083 pmd_flags2 = 0; 1084 1085 for (; addr != end; addr += PAGE_SIZE) { 1086 unsigned long offset; 1087 pagemap_entry_t pme; 1088 1089 offset = (addr & ~PAGEMAP_WALK_MASK) >> 1090 PAGE_SHIFT; 1091 thp_pmd_to_pagemap_entry(&pme, pm, *pmd, offset, pmd_flags2); 1092 err = add_to_pagemap(addr, &pme, pm); 1093 if (err) 1094 break; 1095 } 1096 spin_unlock(ptl); 1097 return err; 1098 } 1099 1100 if (pmd_trans_unstable(pmd)) 1101 return 0; 1102 1103 while (1) { 1104 /* End of address space hole, which we mark as non-present. */ 1105 unsigned long hole_end; 1106 1107 if (vma) 1108 hole_end = min(end, vma->vm_start); 1109 else 1110 hole_end = end; 1111 1112 for (; addr < hole_end; addr += PAGE_SIZE) { 1113 pagemap_entry_t pme = make_pme(PM_NOT_PRESENT(pm->v2)); 1114 1115 err = add_to_pagemap(addr, &pme, pm); 1116 if (err) 1117 return err; 1118 } 1119 1120 if (!vma || vma->vm_start >= end) 1121 break; 1122 /* 1123 * We can't possibly be in a hugetlb VMA. In general, 1124 * for a mm_walk with a pmd_entry and a hugetlb_entry, 1125 * the pmd_entry can only be called on addresses in a 1126 * hugetlb if the walk starts in a non-hugetlb VMA and 1127 * spans a hugepage VMA. Since pagemap_read walks are 1128 * PMD-sized and PMD-aligned, this will never be true. 1129 */ 1130 BUG_ON(is_vm_hugetlb_page(vma)); 1131 1132 /* Addresses in the VMA. */ 1133 for (; addr < min(end, vma->vm_end); addr += PAGE_SIZE) { 1134 pagemap_entry_t pme; 1135 pte = pte_offset_map(pmd, addr); 1136 pte_to_pagemap_entry(&pme, pm, vma, addr, *pte); 1137 pte_unmap(pte); 1138 err = add_to_pagemap(addr, &pme, pm); 1139 if (err) 1140 return err; 1141 } 1142 1143 if (addr == end) 1144 break; 1145 1146 vma = find_vma(walk->mm, addr); 1147 } 1148 1149 cond_resched(); 1150 1151 return err; 1152 } 1153 1154 #ifdef CONFIG_HUGETLB_PAGE 1155 static void huge_pte_to_pagemap_entry(pagemap_entry_t *pme, struct pagemapread *pm, 1156 pte_t pte, int offset, int flags2) 1157 { 1158 if (pte_present(pte)) 1159 *pme = make_pme(PM_PFRAME(pte_pfn(pte) + offset) | 1160 PM_STATUS2(pm->v2, flags2) | 1161 PM_PRESENT); 1162 else 1163 *pme = make_pme(PM_NOT_PRESENT(pm->v2) | 1164 PM_STATUS2(pm->v2, flags2)); 1165 } 1166 1167 /* This function walks within one hugetlb entry in the single call */ 1168 static int pagemap_hugetlb_range(pte_t *pte, unsigned long hmask, 1169 unsigned long addr, unsigned long end, 1170 struct mm_walk *walk) 1171 { 1172 struct pagemapread *pm = walk->private; 1173 struct vm_area_struct *vma; 1174 int err = 0; 1175 int flags2; 1176 pagemap_entry_t pme; 1177 1178 vma = find_vma(walk->mm, addr); 1179 WARN_ON_ONCE(!vma); 1180 1181 if (vma && (vma->vm_flags & VM_SOFTDIRTY)) 1182 flags2 = __PM_SOFT_DIRTY; 1183 else 1184 flags2 = 0; 1185 1186 for (; addr != end; addr += PAGE_SIZE) { 1187 int offset = (addr & ~hmask) >> PAGE_SHIFT; 1188 huge_pte_to_pagemap_entry(&pme, pm, *pte, offset, flags2); 1189 err = add_to_pagemap(addr, &pme, pm); 1190 if (err) 1191 return err; 1192 } 1193 1194 cond_resched(); 1195 1196 return err; 1197 } 1198 #endif /* HUGETLB_PAGE */ 1199 1200 /* 1201 * /proc/pid/pagemap - an array mapping virtual pages to pfns 1202 * 1203 * For each page in the address space, this file contains one 64-bit entry 1204 * consisting of the following: 1205 * 1206 * Bits 0-54 page frame number (PFN) if present 1207 * Bits 0-4 swap type if swapped 1208 * Bits 5-54 swap offset if swapped 1209 * Bits 55-60 page shift (page size = 1<<page shift) 1210 * Bit 61 page is file-page or shared-anon 1211 * Bit 62 page swapped 1212 * Bit 63 page present 1213 * 1214 * If the page is not present but in swap, then the PFN contains an 1215 * encoding of the swap file number and the page's offset into the 1216 * swap. Unmapped pages return a null PFN. This allows determining 1217 * precisely which pages are mapped (or in swap) and comparing mapped 1218 * pages between processes. 1219 * 1220 * Efficient users of this interface will use /proc/pid/maps to 1221 * determine which areas of memory are actually mapped and llseek to 1222 * skip over unmapped regions. 1223 */ 1224 static ssize_t pagemap_read(struct file *file, char __user *buf, 1225 size_t count, loff_t *ppos) 1226 { 1227 struct task_struct *task = get_proc_task(file_inode(file)); 1228 struct mm_struct *mm; 1229 struct pagemapread pm; 1230 int ret = -ESRCH; 1231 struct mm_walk pagemap_walk = {}; 1232 unsigned long src; 1233 unsigned long svpfn; 1234 unsigned long start_vaddr; 1235 unsigned long end_vaddr; 1236 int copied = 0; 1237 1238 if (!task) 1239 goto out; 1240 1241 ret = -EINVAL; 1242 /* file position must be aligned */ 1243 if ((*ppos % PM_ENTRY_BYTES) || (count % PM_ENTRY_BYTES)) 1244 goto out_task; 1245 1246 ret = 0; 1247 if (!count) 1248 goto out_task; 1249 1250 pm.v2 = soft_dirty_cleared; 1251 pm.len = (PAGEMAP_WALK_SIZE >> PAGE_SHIFT); 1252 pm.buffer = kmalloc(pm.len * PM_ENTRY_BYTES, GFP_TEMPORARY); 1253 ret = -ENOMEM; 1254 if (!pm.buffer) 1255 goto out_task; 1256 1257 mm = mm_access(task, PTRACE_MODE_READ); 1258 ret = PTR_ERR(mm); 1259 if (!mm || IS_ERR(mm)) 1260 goto out_free; 1261 1262 pagemap_walk.pmd_entry = pagemap_pte_range; 1263 pagemap_walk.pte_hole = pagemap_pte_hole; 1264 #ifdef CONFIG_HUGETLB_PAGE 1265 pagemap_walk.hugetlb_entry = pagemap_hugetlb_range; 1266 #endif 1267 pagemap_walk.mm = mm; 1268 pagemap_walk.private = ± 1269 1270 src = *ppos; 1271 svpfn = src / PM_ENTRY_BYTES; 1272 start_vaddr = svpfn << PAGE_SHIFT; 1273 end_vaddr = TASK_SIZE_OF(task); 1274 1275 /* watch out for wraparound */ 1276 if (svpfn > TASK_SIZE_OF(task) >> PAGE_SHIFT) 1277 start_vaddr = end_vaddr; 1278 1279 /* 1280 * The odds are that this will stop walking way 1281 * before end_vaddr, because the length of the 1282 * user buffer is tracked in "pm", and the walk 1283 * will stop when we hit the end of the buffer. 1284 */ 1285 ret = 0; 1286 while (count && (start_vaddr < end_vaddr)) { 1287 int len; 1288 unsigned long end; 1289 1290 pm.pos = 0; 1291 end = (start_vaddr + PAGEMAP_WALK_SIZE) & PAGEMAP_WALK_MASK; 1292 /* overflow ? */ 1293 if (end < start_vaddr || end > end_vaddr) 1294 end = end_vaddr; 1295 down_read(&mm->mmap_sem); 1296 ret = walk_page_range(start_vaddr, end, &pagemap_walk); 1297 up_read(&mm->mmap_sem); 1298 start_vaddr = end; 1299 1300 len = min(count, PM_ENTRY_BYTES * pm.pos); 1301 if (copy_to_user(buf, pm.buffer, len)) { 1302 ret = -EFAULT; 1303 goto out_mm; 1304 } 1305 copied += len; 1306 buf += len; 1307 count -= len; 1308 } 1309 *ppos += copied; 1310 if (!ret || ret == PM_END_OF_BUFFER) 1311 ret = copied; 1312 1313 out_mm: 1314 mmput(mm); 1315 out_free: 1316 kfree(pm.buffer); 1317 out_task: 1318 put_task_struct(task); 1319 out: 1320 return ret; 1321 } 1322 1323 static int pagemap_open(struct inode *inode, struct file *file) 1324 { 1325 pr_warn_once("Bits 55-60 of /proc/PID/pagemap entries are about " 1326 "to stop being page-shift some time soon. See the " 1327 "linux/Documentation/vm/pagemap.txt for details.\n"); 1328 return 0; 1329 } 1330 1331 const struct file_operations proc_pagemap_operations = { 1332 .llseek = mem_lseek, /* borrow this */ 1333 .read = pagemap_read, 1334 .open = pagemap_open, 1335 }; 1336 #endif /* CONFIG_PROC_PAGE_MONITOR */ 1337 1338 #ifdef CONFIG_NUMA 1339 1340 struct numa_maps { 1341 struct vm_area_struct *vma; 1342 unsigned long pages; 1343 unsigned long anon; 1344 unsigned long active; 1345 unsigned long writeback; 1346 unsigned long mapcount_max; 1347 unsigned long dirty; 1348 unsigned long swapcache; 1349 unsigned long node[MAX_NUMNODES]; 1350 }; 1351 1352 struct numa_maps_private { 1353 struct proc_maps_private proc_maps; 1354 struct numa_maps md; 1355 }; 1356 1357 static void gather_stats(struct page *page, struct numa_maps *md, int pte_dirty, 1358 unsigned long nr_pages) 1359 { 1360 int count = page_mapcount(page); 1361 1362 md->pages += nr_pages; 1363 if (pte_dirty || PageDirty(page)) 1364 md->dirty += nr_pages; 1365 1366 if (PageSwapCache(page)) 1367 md->swapcache += nr_pages; 1368 1369 if (PageActive(page) || PageUnevictable(page)) 1370 md->active += nr_pages; 1371 1372 if (PageWriteback(page)) 1373 md->writeback += nr_pages; 1374 1375 if (PageAnon(page)) 1376 md->anon += nr_pages; 1377 1378 if (count > md->mapcount_max) 1379 md->mapcount_max = count; 1380 1381 md->node[page_to_nid(page)] += nr_pages; 1382 } 1383 1384 static struct page *can_gather_numa_stats(pte_t pte, struct vm_area_struct *vma, 1385 unsigned long addr) 1386 { 1387 struct page *page; 1388 int nid; 1389 1390 if (!pte_present(pte)) 1391 return NULL; 1392 1393 page = vm_normal_page(vma, addr, pte); 1394 if (!page) 1395 return NULL; 1396 1397 if (PageReserved(page)) 1398 return NULL; 1399 1400 nid = page_to_nid(page); 1401 if (!node_isset(nid, node_states[N_MEMORY])) 1402 return NULL; 1403 1404 return page; 1405 } 1406 1407 static int gather_pte_stats(pmd_t *pmd, unsigned long addr, 1408 unsigned long end, struct mm_walk *walk) 1409 { 1410 struct numa_maps *md; 1411 spinlock_t *ptl; 1412 pte_t *orig_pte; 1413 pte_t *pte; 1414 1415 md = walk->private; 1416 1417 if (pmd_trans_huge_lock(pmd, md->vma, &ptl) == 1) { 1418 pte_t huge_pte = *(pte_t *)pmd; 1419 struct page *page; 1420 1421 page = can_gather_numa_stats(huge_pte, md->vma, addr); 1422 if (page) 1423 gather_stats(page, md, pte_dirty(huge_pte), 1424 HPAGE_PMD_SIZE/PAGE_SIZE); 1425 spin_unlock(ptl); 1426 return 0; 1427 } 1428 1429 if (pmd_trans_unstable(pmd)) 1430 return 0; 1431 orig_pte = pte = pte_offset_map_lock(walk->mm, pmd, addr, &ptl); 1432 do { 1433 struct page *page = can_gather_numa_stats(*pte, md->vma, addr); 1434 if (!page) 1435 continue; 1436 gather_stats(page, md, pte_dirty(*pte), 1); 1437 1438 } while (pte++, addr += PAGE_SIZE, addr != end); 1439 pte_unmap_unlock(orig_pte, ptl); 1440 return 0; 1441 } 1442 #ifdef CONFIG_HUGETLB_PAGE 1443 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1444 unsigned long addr, unsigned long end, struct mm_walk *walk) 1445 { 1446 struct numa_maps *md; 1447 struct page *page; 1448 1449 if (!pte_present(*pte)) 1450 return 0; 1451 1452 page = pte_page(*pte); 1453 if (!page) 1454 return 0; 1455 1456 md = walk->private; 1457 gather_stats(page, md, pte_dirty(*pte), 1); 1458 return 0; 1459 } 1460 1461 #else 1462 static int gather_hugetbl_stats(pte_t *pte, unsigned long hmask, 1463 unsigned long addr, unsigned long end, struct mm_walk *walk) 1464 { 1465 return 0; 1466 } 1467 #endif 1468 1469 /* 1470 * Display pages allocated per node and memory policy via /proc. 1471 */ 1472 static int show_numa_map(struct seq_file *m, void *v, int is_pid) 1473 { 1474 struct numa_maps_private *numa_priv = m->private; 1475 struct proc_maps_private *proc_priv = &numa_priv->proc_maps; 1476 struct vm_area_struct *vma = v; 1477 struct numa_maps *md = &numa_priv->md; 1478 struct file *file = vma->vm_file; 1479 struct mm_struct *mm = vma->vm_mm; 1480 struct mm_walk walk = {}; 1481 struct mempolicy *pol; 1482 char buffer[64]; 1483 int nid; 1484 1485 if (!mm) 1486 return 0; 1487 1488 /* Ensure we start with an empty set of numa_maps statistics. */ 1489 memset(md, 0, sizeof(*md)); 1490 1491 md->vma = vma; 1492 1493 walk.hugetlb_entry = gather_hugetbl_stats; 1494 walk.pmd_entry = gather_pte_stats; 1495 walk.private = md; 1496 walk.mm = mm; 1497 1498 pol = __get_vma_policy(vma, vma->vm_start); 1499 if (pol) { 1500 mpol_to_str(buffer, sizeof(buffer), pol); 1501 mpol_cond_put(pol); 1502 } else { 1503 mpol_to_str(buffer, sizeof(buffer), proc_priv->task_mempolicy); 1504 } 1505 1506 seq_printf(m, "%08lx %s", vma->vm_start, buffer); 1507 1508 if (file) { 1509 seq_puts(m, " file="); 1510 seq_path(m, &file->f_path, "\n\t= "); 1511 } else if (vma->vm_start <= mm->brk && vma->vm_end >= mm->start_brk) { 1512 seq_puts(m, " heap"); 1513 } else { 1514 pid_t tid = pid_of_stack(proc_priv, vma, is_pid); 1515 if (tid != 0) { 1516 /* 1517 * Thread stack in /proc/PID/task/TID/maps or 1518 * the main process stack. 1519 */ 1520 if (!is_pid || (vma->vm_start <= mm->start_stack && 1521 vma->vm_end >= mm->start_stack)) 1522 seq_puts(m, " stack"); 1523 else 1524 seq_printf(m, " stack:%d", tid); 1525 } 1526 } 1527 1528 if (is_vm_hugetlb_page(vma)) 1529 seq_puts(m, " huge"); 1530 1531 walk_page_range(vma->vm_start, vma->vm_end, &walk); 1532 1533 if (!md->pages) 1534 goto out; 1535 1536 if (md->anon) 1537 seq_printf(m, " anon=%lu", md->anon); 1538 1539 if (md->dirty) 1540 seq_printf(m, " dirty=%lu", md->dirty); 1541 1542 if (md->pages != md->anon && md->pages != md->dirty) 1543 seq_printf(m, " mapped=%lu", md->pages); 1544 1545 if (md->mapcount_max > 1) 1546 seq_printf(m, " mapmax=%lu", md->mapcount_max); 1547 1548 if (md->swapcache) 1549 seq_printf(m, " swapcache=%lu", md->swapcache); 1550 1551 if (md->active < md->pages && !is_vm_hugetlb_page(vma)) 1552 seq_printf(m, " active=%lu", md->active); 1553 1554 if (md->writeback) 1555 seq_printf(m, " writeback=%lu", md->writeback); 1556 1557 for_each_node_state(nid, N_MEMORY) 1558 if (md->node[nid]) 1559 seq_printf(m, " N%d=%lu", nid, md->node[nid]); 1560 out: 1561 seq_putc(m, '\n'); 1562 m_cache_vma(m, vma); 1563 return 0; 1564 } 1565 1566 static int show_pid_numa_map(struct seq_file *m, void *v) 1567 { 1568 return show_numa_map(m, v, 1); 1569 } 1570 1571 static int show_tid_numa_map(struct seq_file *m, void *v) 1572 { 1573 return show_numa_map(m, v, 0); 1574 } 1575 1576 static const struct seq_operations proc_pid_numa_maps_op = { 1577 .start = m_start, 1578 .next = m_next, 1579 .stop = m_stop, 1580 .show = show_pid_numa_map, 1581 }; 1582 1583 static const struct seq_operations proc_tid_numa_maps_op = { 1584 .start = m_start, 1585 .next = m_next, 1586 .stop = m_stop, 1587 .show = show_tid_numa_map, 1588 }; 1589 1590 static int numa_maps_open(struct inode *inode, struct file *file, 1591 const struct seq_operations *ops) 1592 { 1593 return proc_maps_open(inode, file, ops, 1594 sizeof(struct numa_maps_private)); 1595 } 1596 1597 static int pid_numa_maps_open(struct inode *inode, struct file *file) 1598 { 1599 return numa_maps_open(inode, file, &proc_pid_numa_maps_op); 1600 } 1601 1602 static int tid_numa_maps_open(struct inode *inode, struct file *file) 1603 { 1604 return numa_maps_open(inode, file, &proc_tid_numa_maps_op); 1605 } 1606 1607 const struct file_operations proc_pid_numa_maps_operations = { 1608 .open = pid_numa_maps_open, 1609 .read = seq_read, 1610 .llseek = seq_lseek, 1611 .release = proc_map_release, 1612 }; 1613 1614 const struct file_operations proc_tid_numa_maps_operations = { 1615 .open = tid_numa_maps_open, 1616 .read = seq_read, 1617 .llseek = seq_lseek, 1618 .release = proc_map_release, 1619 }; 1620 #endif /* CONFIG_NUMA */ 1621