pgtable.c revision 1b948d6caec4f28e3524244ca0f77c6ae8ddceef
1/* 2 * Copyright IBM Corp. 2007, 2011 3 * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com> 4 */ 5 6#include <linux/sched.h> 7#include <linux/kernel.h> 8#include <linux/errno.h> 9#include <linux/gfp.h> 10#include <linux/mm.h> 11#include <linux/swap.h> 12#include <linux/smp.h> 13#include <linux/highmem.h> 14#include <linux/pagemap.h> 15#include <linux/spinlock.h> 16#include <linux/module.h> 17#include <linux/quicklist.h> 18#include <linux/rcupdate.h> 19#include <linux/slab.h> 20#include <linux/swapops.h> 21 22#include <asm/pgtable.h> 23#include <asm/pgalloc.h> 24#include <asm/tlb.h> 25#include <asm/tlbflush.h> 26#include <asm/mmu_context.h> 27 28#ifndef CONFIG_64BIT 29#define ALLOC_ORDER 1 30#define FRAG_MASK 0x0f 31#else 32#define ALLOC_ORDER 2 33#define FRAG_MASK 0x03 34#endif 35 36 37unsigned long *crst_table_alloc(struct mm_struct *mm) 38{ 39 struct page *page = alloc_pages(GFP_KERNEL, ALLOC_ORDER); 40 41 if (!page) 42 return NULL; 43 return (unsigned long *) page_to_phys(page); 44} 45 46void crst_table_free(struct mm_struct *mm, unsigned long *table) 47{ 48 free_pages((unsigned long) table, ALLOC_ORDER); 49} 50 51#ifdef CONFIG_64BIT 52static void __crst_table_upgrade(void *arg) 53{ 54 struct mm_struct *mm = arg; 55 56 if (current->active_mm == mm) 57 update_user_asce(mm); 58 __tlb_flush_local(); 59} 60 61int crst_table_upgrade(struct mm_struct *mm, unsigned long limit) 62{ 63 unsigned long *table, *pgd; 64 unsigned long entry; 65 int flush; 66 67 BUG_ON(limit > (1UL << 53)); 68 flush = 0; 69repeat: 70 table = crst_table_alloc(mm); 71 if (!table) 72 return -ENOMEM; 73 spin_lock_bh(&mm->page_table_lock); 74 if (mm->context.asce_limit < limit) { 75 pgd = (unsigned long *) mm->pgd; 76 if (mm->context.asce_limit <= (1UL << 31)) { 77 entry = _REGION3_ENTRY_EMPTY; 78 mm->context.asce_limit = 1UL << 42; 79 mm->context.asce_bits = _ASCE_TABLE_LENGTH | 80 _ASCE_USER_BITS | 81 _ASCE_TYPE_REGION3; 82 } else { 83 entry = _REGION2_ENTRY_EMPTY; 84 mm->context.asce_limit = 1UL << 53; 85 mm->context.asce_bits = _ASCE_TABLE_LENGTH | 86 _ASCE_USER_BITS | 87 _ASCE_TYPE_REGION2; 88 } 89 crst_table_init(table, entry); 90 pgd_populate(mm, (pgd_t *) table, (pud_t *) pgd); 91 mm->pgd = (pgd_t *) table; 92 mm->task_size = mm->context.asce_limit; 93 table = NULL; 94 flush = 1; 95 } 96 spin_unlock_bh(&mm->page_table_lock); 97 if (table) 98 crst_table_free(mm, table); 99 if (mm->context.asce_limit < limit) 100 goto repeat; 101 if (flush) 102 on_each_cpu(__crst_table_upgrade, mm, 0); 103 return 0; 104} 105 106void crst_table_downgrade(struct mm_struct *mm, unsigned long limit) 107{ 108 pgd_t *pgd; 109 110 if (current->active_mm == mm) { 111 clear_user_asce(mm); 112 __tlb_flush_mm(mm); 113 } 114 while (mm->context.asce_limit > limit) { 115 pgd = mm->pgd; 116 switch (pgd_val(*pgd) & _REGION_ENTRY_TYPE_MASK) { 117 case _REGION_ENTRY_TYPE_R2: 118 mm->context.asce_limit = 1UL << 42; 119 mm->context.asce_bits = _ASCE_TABLE_LENGTH | 120 _ASCE_USER_BITS | 121 _ASCE_TYPE_REGION3; 122 break; 123 case _REGION_ENTRY_TYPE_R3: 124 mm->context.asce_limit = 1UL << 31; 125 mm->context.asce_bits = _ASCE_TABLE_LENGTH | 126 _ASCE_USER_BITS | 127 _ASCE_TYPE_SEGMENT; 128 break; 129 default: 130 BUG(); 131 } 132 mm->pgd = (pgd_t *) (pgd_val(*pgd) & _REGION_ENTRY_ORIGIN); 133 mm->task_size = mm->context.asce_limit; 134 crst_table_free(mm, (unsigned long *) pgd); 135 } 136 if (current->active_mm == mm) 137 update_user_asce(mm); 138} 139#endif 140 141#ifdef CONFIG_PGSTE 142 143/** 144 * gmap_alloc - allocate a guest address space 145 * @mm: pointer to the parent mm_struct 146 * 147 * Returns a guest address space structure. 148 */ 149struct gmap *gmap_alloc(struct mm_struct *mm) 150{ 151 struct gmap *gmap; 152 struct page *page; 153 unsigned long *table; 154 155 gmap = kzalloc(sizeof(struct gmap), GFP_KERNEL); 156 if (!gmap) 157 goto out; 158 INIT_LIST_HEAD(&gmap->crst_list); 159 gmap->mm = mm; 160 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER); 161 if (!page) 162 goto out_free; 163 list_add(&page->lru, &gmap->crst_list); 164 table = (unsigned long *) page_to_phys(page); 165 crst_table_init(table, _REGION1_ENTRY_EMPTY); 166 gmap->table = table; 167 gmap->asce = _ASCE_TYPE_REGION1 | _ASCE_TABLE_LENGTH | 168 _ASCE_USER_BITS | __pa(table); 169 list_add(&gmap->list, &mm->context.gmap_list); 170 return gmap; 171 172out_free: 173 kfree(gmap); 174out: 175 return NULL; 176} 177EXPORT_SYMBOL_GPL(gmap_alloc); 178 179static int gmap_unlink_segment(struct gmap *gmap, unsigned long *table) 180{ 181 struct gmap_pgtable *mp; 182 struct gmap_rmap *rmap; 183 struct page *page; 184 185 if (*table & _SEGMENT_ENTRY_INVALID) 186 return 0; 187 page = pfn_to_page(*table >> PAGE_SHIFT); 188 mp = (struct gmap_pgtable *) page->index; 189 list_for_each_entry(rmap, &mp->mapper, list) { 190 if (rmap->entry != table) 191 continue; 192 list_del(&rmap->list); 193 kfree(rmap); 194 break; 195 } 196 *table = mp->vmaddr | _SEGMENT_ENTRY_INVALID | _SEGMENT_ENTRY_PROTECT; 197 return 1; 198} 199 200static void gmap_flush_tlb(struct gmap *gmap) 201{ 202 if (MACHINE_HAS_IDTE) 203 __tlb_flush_asce(gmap->mm, (unsigned long) gmap->table | 204 _ASCE_TYPE_REGION1); 205 else 206 __tlb_flush_global(); 207} 208 209/** 210 * gmap_free - free a guest address space 211 * @gmap: pointer to the guest address space structure 212 */ 213void gmap_free(struct gmap *gmap) 214{ 215 struct page *page, *next; 216 unsigned long *table; 217 int i; 218 219 220 /* Flush tlb. */ 221 if (MACHINE_HAS_IDTE) 222 __tlb_flush_asce(gmap->mm, (unsigned long) gmap->table | 223 _ASCE_TYPE_REGION1); 224 else 225 __tlb_flush_global(); 226 227 /* Free all segment & region tables. */ 228 down_read(&gmap->mm->mmap_sem); 229 spin_lock(&gmap->mm->page_table_lock); 230 list_for_each_entry_safe(page, next, &gmap->crst_list, lru) { 231 table = (unsigned long *) page_to_phys(page); 232 if ((*table & _REGION_ENTRY_TYPE_MASK) == 0) 233 /* Remove gmap rmap structures for segment table. */ 234 for (i = 0; i < PTRS_PER_PMD; i++, table++) 235 gmap_unlink_segment(gmap, table); 236 __free_pages(page, ALLOC_ORDER); 237 } 238 spin_unlock(&gmap->mm->page_table_lock); 239 up_read(&gmap->mm->mmap_sem); 240 list_del(&gmap->list); 241 kfree(gmap); 242} 243EXPORT_SYMBOL_GPL(gmap_free); 244 245/** 246 * gmap_enable - switch primary space to the guest address space 247 * @gmap: pointer to the guest address space structure 248 */ 249void gmap_enable(struct gmap *gmap) 250{ 251 S390_lowcore.gmap = (unsigned long) gmap; 252} 253EXPORT_SYMBOL_GPL(gmap_enable); 254 255/** 256 * gmap_disable - switch back to the standard primary address space 257 * @gmap: pointer to the guest address space structure 258 */ 259void gmap_disable(struct gmap *gmap) 260{ 261 S390_lowcore.gmap = 0UL; 262} 263EXPORT_SYMBOL_GPL(gmap_disable); 264 265/* 266 * gmap_alloc_table is assumed to be called with mmap_sem held 267 */ 268static int gmap_alloc_table(struct gmap *gmap, 269 unsigned long *table, unsigned long init) 270 __releases(&gmap->mm->page_table_lock) 271 __acquires(&gmap->mm->page_table_lock) 272{ 273 struct page *page; 274 unsigned long *new; 275 276 /* since we dont free the gmap table until gmap_free we can unlock */ 277 spin_unlock(&gmap->mm->page_table_lock); 278 page = alloc_pages(GFP_KERNEL, ALLOC_ORDER); 279 spin_lock(&gmap->mm->page_table_lock); 280 if (!page) 281 return -ENOMEM; 282 new = (unsigned long *) page_to_phys(page); 283 crst_table_init(new, init); 284 if (*table & _REGION_ENTRY_INVALID) { 285 list_add(&page->lru, &gmap->crst_list); 286 *table = (unsigned long) new | _REGION_ENTRY_LENGTH | 287 (*table & _REGION_ENTRY_TYPE_MASK); 288 } else 289 __free_pages(page, ALLOC_ORDER); 290 return 0; 291} 292 293/** 294 * gmap_unmap_segment - unmap segment from the guest address space 295 * @gmap: pointer to the guest address space structure 296 * @addr: address in the guest address space 297 * @len: length of the memory area to unmap 298 * 299 * Returns 0 if the unmap succeeded, -EINVAL if not. 300 */ 301int gmap_unmap_segment(struct gmap *gmap, unsigned long to, unsigned long len) 302{ 303 unsigned long *table; 304 unsigned long off; 305 int flush; 306 307 if ((to | len) & (PMD_SIZE - 1)) 308 return -EINVAL; 309 if (len == 0 || to + len < to) 310 return -EINVAL; 311 312 flush = 0; 313 down_read(&gmap->mm->mmap_sem); 314 spin_lock(&gmap->mm->page_table_lock); 315 for (off = 0; off < len; off += PMD_SIZE) { 316 /* Walk the guest addr space page table */ 317 table = gmap->table + (((to + off) >> 53) & 0x7ff); 318 if (*table & _REGION_ENTRY_INVALID) 319 goto out; 320 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 321 table = table + (((to + off) >> 42) & 0x7ff); 322 if (*table & _REGION_ENTRY_INVALID) 323 goto out; 324 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 325 table = table + (((to + off) >> 31) & 0x7ff); 326 if (*table & _REGION_ENTRY_INVALID) 327 goto out; 328 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 329 table = table + (((to + off) >> 20) & 0x7ff); 330 331 /* Clear segment table entry in guest address space. */ 332 flush |= gmap_unlink_segment(gmap, table); 333 *table = _SEGMENT_ENTRY_INVALID; 334 } 335out: 336 spin_unlock(&gmap->mm->page_table_lock); 337 up_read(&gmap->mm->mmap_sem); 338 if (flush) 339 gmap_flush_tlb(gmap); 340 return 0; 341} 342EXPORT_SYMBOL_GPL(gmap_unmap_segment); 343 344/** 345 * gmap_mmap_segment - map a segment to the guest address space 346 * @gmap: pointer to the guest address space structure 347 * @from: source address in the parent address space 348 * @to: target address in the guest address space 349 * 350 * Returns 0 if the mmap succeeded, -EINVAL or -ENOMEM if not. 351 */ 352int gmap_map_segment(struct gmap *gmap, unsigned long from, 353 unsigned long to, unsigned long len) 354{ 355 unsigned long *table; 356 unsigned long off; 357 int flush; 358 359 if ((from | to | len) & (PMD_SIZE - 1)) 360 return -EINVAL; 361 if (len == 0 || from + len > TASK_MAX_SIZE || 362 from + len < from || to + len < to) 363 return -EINVAL; 364 365 flush = 0; 366 down_read(&gmap->mm->mmap_sem); 367 spin_lock(&gmap->mm->page_table_lock); 368 for (off = 0; off < len; off += PMD_SIZE) { 369 /* Walk the gmap address space page table */ 370 table = gmap->table + (((to + off) >> 53) & 0x7ff); 371 if ((*table & _REGION_ENTRY_INVALID) && 372 gmap_alloc_table(gmap, table, _REGION2_ENTRY_EMPTY)) 373 goto out_unmap; 374 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 375 table = table + (((to + off) >> 42) & 0x7ff); 376 if ((*table & _REGION_ENTRY_INVALID) && 377 gmap_alloc_table(gmap, table, _REGION3_ENTRY_EMPTY)) 378 goto out_unmap; 379 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 380 table = table + (((to + off) >> 31) & 0x7ff); 381 if ((*table & _REGION_ENTRY_INVALID) && 382 gmap_alloc_table(gmap, table, _SEGMENT_ENTRY_EMPTY)) 383 goto out_unmap; 384 table = (unsigned long *) (*table & _REGION_ENTRY_ORIGIN); 385 table = table + (((to + off) >> 20) & 0x7ff); 386 387 /* Store 'from' address in an invalid segment table entry. */ 388 flush |= gmap_unlink_segment(gmap, table); 389 *table = (from + off) | (_SEGMENT_ENTRY_INVALID | 390 _SEGMENT_ENTRY_PROTECT); 391 } 392 spin_unlock(&gmap->mm->page_table_lock); 393 up_read(&gmap->mm->mmap_sem); 394 if (flush) 395 gmap_flush_tlb(gmap); 396 return 0; 397 398out_unmap: 399 spin_unlock(&gmap->mm->page_table_lock); 400 up_read(&gmap->mm->mmap_sem); 401 gmap_unmap_segment(gmap, to, len); 402 return -ENOMEM; 403} 404EXPORT_SYMBOL_GPL(gmap_map_segment); 405 406static unsigned long *gmap_table_walk(unsigned long address, struct gmap *gmap) 407{ 408 unsigned long *table; 409 410 table = gmap->table + ((address >> 53) & 0x7ff); 411 if (unlikely(*table & _REGION_ENTRY_INVALID)) 412 return ERR_PTR(-EFAULT); 413 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 414 table = table + ((address >> 42) & 0x7ff); 415 if (unlikely(*table & _REGION_ENTRY_INVALID)) 416 return ERR_PTR(-EFAULT); 417 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 418 table = table + ((address >> 31) & 0x7ff); 419 if (unlikely(*table & _REGION_ENTRY_INVALID)) 420 return ERR_PTR(-EFAULT); 421 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 422 table = table + ((address >> 20) & 0x7ff); 423 return table; 424} 425 426/** 427 * __gmap_translate - translate a guest address to a user space address 428 * @address: guest address 429 * @gmap: pointer to guest mapping meta data structure 430 * 431 * Returns user space address which corresponds to the guest address or 432 * -EFAULT if no such mapping exists. 433 * This function does not establish potentially missing page table entries. 434 * The mmap_sem of the mm that belongs to the address space must be held 435 * when this function gets called. 436 */ 437unsigned long __gmap_translate(unsigned long address, struct gmap *gmap) 438{ 439 unsigned long *segment_ptr, vmaddr, segment; 440 struct gmap_pgtable *mp; 441 struct page *page; 442 443 current->thread.gmap_addr = address; 444 segment_ptr = gmap_table_walk(address, gmap); 445 if (IS_ERR(segment_ptr)) 446 return PTR_ERR(segment_ptr); 447 /* Convert the gmap address to an mm address. */ 448 segment = *segment_ptr; 449 if (!(segment & _SEGMENT_ENTRY_INVALID)) { 450 page = pfn_to_page(segment >> PAGE_SHIFT); 451 mp = (struct gmap_pgtable *) page->index; 452 return mp->vmaddr | (address & ~PMD_MASK); 453 } else if (segment & _SEGMENT_ENTRY_PROTECT) { 454 vmaddr = segment & _SEGMENT_ENTRY_ORIGIN; 455 return vmaddr | (address & ~PMD_MASK); 456 } 457 return -EFAULT; 458} 459EXPORT_SYMBOL_GPL(__gmap_translate); 460 461/** 462 * gmap_translate - translate a guest address to a user space address 463 * @address: guest address 464 * @gmap: pointer to guest mapping meta data structure 465 * 466 * Returns user space address which corresponds to the guest address or 467 * -EFAULT if no such mapping exists. 468 * This function does not establish potentially missing page table entries. 469 */ 470unsigned long gmap_translate(unsigned long address, struct gmap *gmap) 471{ 472 unsigned long rc; 473 474 down_read(&gmap->mm->mmap_sem); 475 rc = __gmap_translate(address, gmap); 476 up_read(&gmap->mm->mmap_sem); 477 return rc; 478} 479EXPORT_SYMBOL_GPL(gmap_translate); 480 481static int gmap_connect_pgtable(unsigned long address, unsigned long segment, 482 unsigned long *segment_ptr, struct gmap *gmap) 483{ 484 unsigned long vmaddr; 485 struct vm_area_struct *vma; 486 struct gmap_pgtable *mp; 487 struct gmap_rmap *rmap; 488 struct mm_struct *mm; 489 struct page *page; 490 pgd_t *pgd; 491 pud_t *pud; 492 pmd_t *pmd; 493 494 mm = gmap->mm; 495 vmaddr = segment & _SEGMENT_ENTRY_ORIGIN; 496 vma = find_vma(mm, vmaddr); 497 if (!vma || vma->vm_start > vmaddr) 498 return -EFAULT; 499 /* Walk the parent mm page table */ 500 pgd = pgd_offset(mm, vmaddr); 501 pud = pud_alloc(mm, pgd, vmaddr); 502 if (!pud) 503 return -ENOMEM; 504 pmd = pmd_alloc(mm, pud, vmaddr); 505 if (!pmd) 506 return -ENOMEM; 507 if (!pmd_present(*pmd) && 508 __pte_alloc(mm, vma, pmd, vmaddr)) 509 return -ENOMEM; 510 /* pmd now points to a valid segment table entry. */ 511 rmap = kmalloc(sizeof(*rmap), GFP_KERNEL|__GFP_REPEAT); 512 if (!rmap) 513 return -ENOMEM; 514 /* Link gmap segment table entry location to page table. */ 515 page = pmd_page(*pmd); 516 mp = (struct gmap_pgtable *) page->index; 517 rmap->gmap = gmap; 518 rmap->entry = segment_ptr; 519 rmap->vmaddr = address & PMD_MASK; 520 spin_lock(&mm->page_table_lock); 521 if (*segment_ptr == segment) { 522 list_add(&rmap->list, &mp->mapper); 523 /* Set gmap segment table entry to page table. */ 524 *segment_ptr = pmd_val(*pmd) & PAGE_MASK; 525 rmap = NULL; 526 } 527 spin_unlock(&mm->page_table_lock); 528 kfree(rmap); 529 return 0; 530} 531 532static void gmap_disconnect_pgtable(struct mm_struct *mm, unsigned long *table) 533{ 534 struct gmap_rmap *rmap, *next; 535 struct gmap_pgtable *mp; 536 struct page *page; 537 int flush; 538 539 flush = 0; 540 spin_lock(&mm->page_table_lock); 541 page = pfn_to_page(__pa(table) >> PAGE_SHIFT); 542 mp = (struct gmap_pgtable *) page->index; 543 list_for_each_entry_safe(rmap, next, &mp->mapper, list) { 544 *rmap->entry = mp->vmaddr | (_SEGMENT_ENTRY_INVALID | 545 _SEGMENT_ENTRY_PROTECT); 546 list_del(&rmap->list); 547 kfree(rmap); 548 flush = 1; 549 } 550 spin_unlock(&mm->page_table_lock); 551 if (flush) 552 __tlb_flush_global(); 553} 554 555/* 556 * this function is assumed to be called with mmap_sem held 557 */ 558unsigned long __gmap_fault(unsigned long address, struct gmap *gmap) 559{ 560 unsigned long *segment_ptr, segment; 561 struct gmap_pgtable *mp; 562 struct page *page; 563 int rc; 564 565 current->thread.gmap_addr = address; 566 segment_ptr = gmap_table_walk(address, gmap); 567 if (IS_ERR(segment_ptr)) 568 return -EFAULT; 569 /* Convert the gmap address to an mm address. */ 570 while (1) { 571 segment = *segment_ptr; 572 if (!(segment & _SEGMENT_ENTRY_INVALID)) { 573 /* Page table is present */ 574 page = pfn_to_page(segment >> PAGE_SHIFT); 575 mp = (struct gmap_pgtable *) page->index; 576 return mp->vmaddr | (address & ~PMD_MASK); 577 } 578 if (!(segment & _SEGMENT_ENTRY_PROTECT)) 579 /* Nothing mapped in the gmap address space. */ 580 break; 581 rc = gmap_connect_pgtable(address, segment, segment_ptr, gmap); 582 if (rc) 583 return rc; 584 } 585 return -EFAULT; 586} 587 588unsigned long gmap_fault(unsigned long address, struct gmap *gmap) 589{ 590 unsigned long rc; 591 592 down_read(&gmap->mm->mmap_sem); 593 rc = __gmap_fault(address, gmap); 594 up_read(&gmap->mm->mmap_sem); 595 596 return rc; 597} 598EXPORT_SYMBOL_GPL(gmap_fault); 599 600static void gmap_zap_swap_entry(swp_entry_t entry, struct mm_struct *mm) 601{ 602 if (!non_swap_entry(entry)) 603 dec_mm_counter(mm, MM_SWAPENTS); 604 else if (is_migration_entry(entry)) { 605 struct page *page = migration_entry_to_page(entry); 606 607 if (PageAnon(page)) 608 dec_mm_counter(mm, MM_ANONPAGES); 609 else 610 dec_mm_counter(mm, MM_FILEPAGES); 611 } 612 free_swap_and_cache(entry); 613} 614 615/** 616 * The mm->mmap_sem lock must be held 617 */ 618static void gmap_zap_unused(struct mm_struct *mm, unsigned long address) 619{ 620 unsigned long ptev, pgstev; 621 spinlock_t *ptl; 622 pgste_t pgste; 623 pte_t *ptep, pte; 624 625 ptep = get_locked_pte(mm, address, &ptl); 626 if (unlikely(!ptep)) 627 return; 628 pte = *ptep; 629 if (!pte_swap(pte)) 630 goto out_pte; 631 /* Zap unused and logically-zero pages */ 632 pgste = pgste_get_lock(ptep); 633 pgstev = pgste_val(pgste); 634 ptev = pte_val(pte); 635 if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) || 636 ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) { 637 gmap_zap_swap_entry(pte_to_swp_entry(pte), mm); 638 pte_clear(mm, address, ptep); 639 } 640 pgste_set_unlock(ptep, pgste); 641out_pte: 642 pte_unmap_unlock(*ptep, ptl); 643} 644 645/* 646 * this function is assumed to be called with mmap_sem held 647 */ 648void __gmap_zap(unsigned long address, struct gmap *gmap) 649{ 650 unsigned long *table, *segment_ptr; 651 unsigned long segment, pgstev, ptev; 652 struct gmap_pgtable *mp; 653 struct page *page; 654 655 segment_ptr = gmap_table_walk(address, gmap); 656 if (IS_ERR(segment_ptr)) 657 return; 658 segment = *segment_ptr; 659 if (segment & _SEGMENT_ENTRY_INVALID) 660 return; 661 page = pfn_to_page(segment >> PAGE_SHIFT); 662 mp = (struct gmap_pgtable *) page->index; 663 address = mp->vmaddr | (address & ~PMD_MASK); 664 /* Page table is present */ 665 table = (unsigned long *)(segment & _SEGMENT_ENTRY_ORIGIN); 666 table = table + ((address >> 12) & 0xff); 667 pgstev = table[PTRS_PER_PTE]; 668 ptev = table[0]; 669 /* quick check, checked again with locks held */ 670 if (((pgstev & _PGSTE_GPS_USAGE_MASK) == _PGSTE_GPS_USAGE_UNUSED) || 671 ((pgstev & _PGSTE_GPS_ZERO) && (ptev & _PAGE_INVALID))) 672 gmap_zap_unused(gmap->mm, address); 673} 674EXPORT_SYMBOL_GPL(__gmap_zap); 675 676void gmap_discard(unsigned long from, unsigned long to, struct gmap *gmap) 677{ 678 679 unsigned long *table, address, size; 680 struct vm_area_struct *vma; 681 struct gmap_pgtable *mp; 682 struct page *page; 683 684 down_read(&gmap->mm->mmap_sem); 685 address = from; 686 while (address < to) { 687 /* Walk the gmap address space page table */ 688 table = gmap->table + ((address >> 53) & 0x7ff); 689 if (unlikely(*table & _REGION_ENTRY_INVALID)) { 690 address = (address + PMD_SIZE) & PMD_MASK; 691 continue; 692 } 693 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 694 table = table + ((address >> 42) & 0x7ff); 695 if (unlikely(*table & _REGION_ENTRY_INVALID)) { 696 address = (address + PMD_SIZE) & PMD_MASK; 697 continue; 698 } 699 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 700 table = table + ((address >> 31) & 0x7ff); 701 if (unlikely(*table & _REGION_ENTRY_INVALID)) { 702 address = (address + PMD_SIZE) & PMD_MASK; 703 continue; 704 } 705 table = (unsigned long *)(*table & _REGION_ENTRY_ORIGIN); 706 table = table + ((address >> 20) & 0x7ff); 707 if (unlikely(*table & _SEGMENT_ENTRY_INVALID)) { 708 address = (address + PMD_SIZE) & PMD_MASK; 709 continue; 710 } 711 page = pfn_to_page(*table >> PAGE_SHIFT); 712 mp = (struct gmap_pgtable *) page->index; 713 vma = find_vma(gmap->mm, mp->vmaddr); 714 size = min(to - address, PMD_SIZE - (address & ~PMD_MASK)); 715 zap_page_range(vma, mp->vmaddr | (address & ~PMD_MASK), 716 size, NULL); 717 address = (address + PMD_SIZE) & PMD_MASK; 718 } 719 up_read(&gmap->mm->mmap_sem); 720} 721EXPORT_SYMBOL_GPL(gmap_discard); 722 723static LIST_HEAD(gmap_notifier_list); 724static DEFINE_SPINLOCK(gmap_notifier_lock); 725 726/** 727 * gmap_register_ipte_notifier - register a pte invalidation callback 728 * @nb: pointer to the gmap notifier block 729 */ 730void gmap_register_ipte_notifier(struct gmap_notifier *nb) 731{ 732 spin_lock(&gmap_notifier_lock); 733 list_add(&nb->list, &gmap_notifier_list); 734 spin_unlock(&gmap_notifier_lock); 735} 736EXPORT_SYMBOL_GPL(gmap_register_ipte_notifier); 737 738/** 739 * gmap_unregister_ipte_notifier - remove a pte invalidation callback 740 * @nb: pointer to the gmap notifier block 741 */ 742void gmap_unregister_ipte_notifier(struct gmap_notifier *nb) 743{ 744 spin_lock(&gmap_notifier_lock); 745 list_del_init(&nb->list); 746 spin_unlock(&gmap_notifier_lock); 747} 748EXPORT_SYMBOL_GPL(gmap_unregister_ipte_notifier); 749 750/** 751 * gmap_ipte_notify - mark a range of ptes for invalidation notification 752 * @gmap: pointer to guest mapping meta data structure 753 * @start: virtual address in the guest address space 754 * @len: size of area 755 * 756 * Returns 0 if for each page in the given range a gmap mapping exists and 757 * the invalidation notification could be set. If the gmap mapping is missing 758 * for one or more pages -EFAULT is returned. If no memory could be allocated 759 * -ENOMEM is returned. This function establishes missing page table entries. 760 */ 761int gmap_ipte_notify(struct gmap *gmap, unsigned long start, unsigned long len) 762{ 763 unsigned long addr; 764 spinlock_t *ptl; 765 pte_t *ptep, entry; 766 pgste_t pgste; 767 int rc = 0; 768 769 if ((start & ~PAGE_MASK) || (len & ~PAGE_MASK)) 770 return -EINVAL; 771 down_read(&gmap->mm->mmap_sem); 772 while (len) { 773 /* Convert gmap address and connect the page tables */ 774 addr = __gmap_fault(start, gmap); 775 if (IS_ERR_VALUE(addr)) { 776 rc = addr; 777 break; 778 } 779 /* Get the page mapped */ 780 if (fixup_user_fault(current, gmap->mm, addr, FAULT_FLAG_WRITE)) { 781 rc = -EFAULT; 782 break; 783 } 784 /* Walk the process page table, lock and get pte pointer */ 785 ptep = get_locked_pte(gmap->mm, addr, &ptl); 786 if (unlikely(!ptep)) 787 continue; 788 /* Set notification bit in the pgste of the pte */ 789 entry = *ptep; 790 if ((pte_val(entry) & (_PAGE_INVALID | _PAGE_PROTECT)) == 0) { 791 pgste = pgste_get_lock(ptep); 792 pgste_val(pgste) |= PGSTE_IN_BIT; 793 pgste_set_unlock(ptep, pgste); 794 start += PAGE_SIZE; 795 len -= PAGE_SIZE; 796 } 797 spin_unlock(ptl); 798 } 799 up_read(&gmap->mm->mmap_sem); 800 return rc; 801} 802EXPORT_SYMBOL_GPL(gmap_ipte_notify); 803 804/** 805 * gmap_do_ipte_notify - call all invalidation callbacks for a specific pte. 806 * @mm: pointer to the process mm_struct 807 * @pte: pointer to the page table entry 808 * 809 * This function is assumed to be called with the page table lock held 810 * for the pte to notify. 811 */ 812void gmap_do_ipte_notify(struct mm_struct *mm, pte_t *pte) 813{ 814 unsigned long segment_offset; 815 struct gmap_notifier *nb; 816 struct gmap_pgtable *mp; 817 struct gmap_rmap *rmap; 818 struct page *page; 819 820 segment_offset = ((unsigned long) pte) & (255 * sizeof(pte_t)); 821 segment_offset = segment_offset * (4096 / sizeof(pte_t)); 822 page = pfn_to_page(__pa(pte) >> PAGE_SHIFT); 823 mp = (struct gmap_pgtable *) page->index; 824 spin_lock(&gmap_notifier_lock); 825 list_for_each_entry(rmap, &mp->mapper, list) { 826 list_for_each_entry(nb, &gmap_notifier_list, list) 827 nb->notifier_call(rmap->gmap, 828 rmap->vmaddr + segment_offset); 829 } 830 spin_unlock(&gmap_notifier_lock); 831} 832 833static inline int page_table_with_pgste(struct page *page) 834{ 835 return atomic_read(&page->_mapcount) == 0; 836} 837 838static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm, 839 unsigned long vmaddr) 840{ 841 struct page *page; 842 unsigned long *table; 843 struct gmap_pgtable *mp; 844 845 page = alloc_page(GFP_KERNEL|__GFP_REPEAT); 846 if (!page) 847 return NULL; 848 mp = kmalloc(sizeof(*mp), GFP_KERNEL|__GFP_REPEAT); 849 if (!mp) { 850 __free_page(page); 851 return NULL; 852 } 853 if (!pgtable_page_ctor(page)) { 854 kfree(mp); 855 __free_page(page); 856 return NULL; 857 } 858 mp->vmaddr = vmaddr & PMD_MASK; 859 INIT_LIST_HEAD(&mp->mapper); 860 page->index = (unsigned long) mp; 861 atomic_set(&page->_mapcount, 0); 862 table = (unsigned long *) page_to_phys(page); 863 clear_table(table, _PAGE_INVALID, PAGE_SIZE/2); 864 clear_table(table + PTRS_PER_PTE, PGSTE_HR_BIT | PGSTE_HC_BIT, 865 PAGE_SIZE/2); 866 return table; 867} 868 869static inline void page_table_free_pgste(unsigned long *table) 870{ 871 struct page *page; 872 struct gmap_pgtable *mp; 873 874 page = pfn_to_page(__pa(table) >> PAGE_SHIFT); 875 mp = (struct gmap_pgtable *) page->index; 876 BUG_ON(!list_empty(&mp->mapper)); 877 pgtable_page_dtor(page); 878 atomic_set(&page->_mapcount, -1); 879 kfree(mp); 880 __free_page(page); 881} 882 883static inline unsigned long page_table_reset_pte(struct mm_struct *mm, 884 pmd_t *pmd, unsigned long addr, unsigned long end) 885{ 886 pte_t *start_pte, *pte; 887 spinlock_t *ptl; 888 pgste_t pgste; 889 890 start_pte = pte_offset_map_lock(mm, pmd, addr, &ptl); 891 pte = start_pte; 892 do { 893 pgste = pgste_get_lock(pte); 894 pgste_val(pgste) &= ~_PGSTE_GPS_USAGE_MASK; 895 pgste_set_unlock(pte, pgste); 896 } while (pte++, addr += PAGE_SIZE, addr != end); 897 pte_unmap_unlock(start_pte, ptl); 898 899 return addr; 900} 901 902static inline unsigned long page_table_reset_pmd(struct mm_struct *mm, 903 pud_t *pud, unsigned long addr, unsigned long end) 904{ 905 unsigned long next; 906 pmd_t *pmd; 907 908 pmd = pmd_offset(pud, addr); 909 do { 910 next = pmd_addr_end(addr, end); 911 if (pmd_none_or_clear_bad(pmd)) 912 continue; 913 next = page_table_reset_pte(mm, pmd, addr, next); 914 } while (pmd++, addr = next, addr != end); 915 916 return addr; 917} 918 919static inline unsigned long page_table_reset_pud(struct mm_struct *mm, 920 pgd_t *pgd, unsigned long addr, unsigned long end) 921{ 922 unsigned long next; 923 pud_t *pud; 924 925 pud = pud_offset(pgd, addr); 926 do { 927 next = pud_addr_end(addr, end); 928 if (pud_none_or_clear_bad(pud)) 929 continue; 930 next = page_table_reset_pmd(mm, pud, addr, next); 931 } while (pud++, addr = next, addr != end); 932 933 return addr; 934} 935 936void page_table_reset_pgste(struct mm_struct *mm, 937 unsigned long start, unsigned long end) 938{ 939 unsigned long addr, next; 940 pgd_t *pgd; 941 942 addr = start; 943 down_read(&mm->mmap_sem); 944 pgd = pgd_offset(mm, addr); 945 do { 946 next = pgd_addr_end(addr, end); 947 if (pgd_none_or_clear_bad(pgd)) 948 continue; 949 next = page_table_reset_pud(mm, pgd, addr, next); 950 } while (pgd++, addr = next, addr != end); 951 up_read(&mm->mmap_sem); 952} 953EXPORT_SYMBOL(page_table_reset_pgste); 954 955int set_guest_storage_key(struct mm_struct *mm, unsigned long addr, 956 unsigned long key, bool nq) 957{ 958 spinlock_t *ptl; 959 pgste_t old, new; 960 pte_t *ptep; 961 962 down_read(&mm->mmap_sem); 963 ptep = get_locked_pte(current->mm, addr, &ptl); 964 if (unlikely(!ptep)) { 965 up_read(&mm->mmap_sem); 966 return -EFAULT; 967 } 968 969 new = old = pgste_get_lock(ptep); 970 pgste_val(new) &= ~(PGSTE_GR_BIT | PGSTE_GC_BIT | 971 PGSTE_ACC_BITS | PGSTE_FP_BIT); 972 pgste_val(new) |= (key & (_PAGE_CHANGED | _PAGE_REFERENCED)) << 48; 973 pgste_val(new) |= (key & (_PAGE_ACC_BITS | _PAGE_FP_BIT)) << 56; 974 if (!(pte_val(*ptep) & _PAGE_INVALID)) { 975 unsigned long address, bits, skey; 976 977 address = pte_val(*ptep) & PAGE_MASK; 978 skey = (unsigned long) page_get_storage_key(address); 979 bits = skey & (_PAGE_CHANGED | _PAGE_REFERENCED); 980 skey = key & (_PAGE_ACC_BITS | _PAGE_FP_BIT); 981 /* Set storage key ACC and FP */ 982 page_set_storage_key(address, skey, !nq); 983 /* Merge host changed & referenced into pgste */ 984 pgste_val(new) |= bits << 52; 985 } 986 /* changing the guest storage key is considered a change of the page */ 987 if ((pgste_val(new) ^ pgste_val(old)) & 988 (PGSTE_ACC_BITS | PGSTE_FP_BIT | PGSTE_GR_BIT | PGSTE_GC_BIT)) 989 pgste_val(new) |= PGSTE_HC_BIT; 990 991 pgste_set_unlock(ptep, new); 992 pte_unmap_unlock(*ptep, ptl); 993 up_read(&mm->mmap_sem); 994 return 0; 995} 996EXPORT_SYMBOL(set_guest_storage_key); 997 998#else /* CONFIG_PGSTE */ 999 1000static inline int page_table_with_pgste(struct page *page) 1001{ 1002 return 0; 1003} 1004 1005static inline unsigned long *page_table_alloc_pgste(struct mm_struct *mm, 1006 unsigned long vmaddr) 1007{ 1008 return NULL; 1009} 1010 1011static inline void page_table_free_pgste(unsigned long *table) 1012{ 1013} 1014 1015static inline void gmap_disconnect_pgtable(struct mm_struct *mm, 1016 unsigned long *table) 1017{ 1018} 1019 1020#endif /* CONFIG_PGSTE */ 1021 1022static inline unsigned int atomic_xor_bits(atomic_t *v, unsigned int bits) 1023{ 1024 unsigned int old, new; 1025 1026 do { 1027 old = atomic_read(v); 1028 new = old ^ bits; 1029 } while (atomic_cmpxchg(v, old, new) != old); 1030 return new; 1031} 1032 1033/* 1034 * page table entry allocation/free routines. 1035 */ 1036unsigned long *page_table_alloc(struct mm_struct *mm, unsigned long vmaddr) 1037{ 1038 unsigned long *uninitialized_var(table); 1039 struct page *uninitialized_var(page); 1040 unsigned int mask, bit; 1041 1042 if (mm_has_pgste(mm)) 1043 return page_table_alloc_pgste(mm, vmaddr); 1044 /* Allocate fragments of a 4K page as 1K/2K page table */ 1045 spin_lock_bh(&mm->context.list_lock); 1046 mask = FRAG_MASK; 1047 if (!list_empty(&mm->context.pgtable_list)) { 1048 page = list_first_entry(&mm->context.pgtable_list, 1049 struct page, lru); 1050 table = (unsigned long *) page_to_phys(page); 1051 mask = atomic_read(&page->_mapcount); 1052 mask = mask | (mask >> 4); 1053 } 1054 if ((mask & FRAG_MASK) == FRAG_MASK) { 1055 spin_unlock_bh(&mm->context.list_lock); 1056 page = alloc_page(GFP_KERNEL|__GFP_REPEAT); 1057 if (!page) 1058 return NULL; 1059 if (!pgtable_page_ctor(page)) { 1060 __free_page(page); 1061 return NULL; 1062 } 1063 atomic_set(&page->_mapcount, 1); 1064 table = (unsigned long *) page_to_phys(page); 1065 clear_table(table, _PAGE_INVALID, PAGE_SIZE); 1066 spin_lock_bh(&mm->context.list_lock); 1067 list_add(&page->lru, &mm->context.pgtable_list); 1068 } else { 1069 for (bit = 1; mask & bit; bit <<= 1) 1070 table += PTRS_PER_PTE; 1071 mask = atomic_xor_bits(&page->_mapcount, bit); 1072 if ((mask & FRAG_MASK) == FRAG_MASK) 1073 list_del(&page->lru); 1074 } 1075 spin_unlock_bh(&mm->context.list_lock); 1076 return table; 1077} 1078 1079void page_table_free(struct mm_struct *mm, unsigned long *table) 1080{ 1081 struct page *page; 1082 unsigned int bit, mask; 1083 1084 page = pfn_to_page(__pa(table) >> PAGE_SHIFT); 1085 if (page_table_with_pgste(page)) { 1086 gmap_disconnect_pgtable(mm, table); 1087 return page_table_free_pgste(table); 1088 } 1089 /* Free 1K/2K page table fragment of a 4K page */ 1090 bit = 1 << ((__pa(table) & ~PAGE_MASK)/(PTRS_PER_PTE*sizeof(pte_t))); 1091 spin_lock_bh(&mm->context.list_lock); 1092 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK) 1093 list_del(&page->lru); 1094 mask = atomic_xor_bits(&page->_mapcount, bit); 1095 if (mask & FRAG_MASK) 1096 list_add(&page->lru, &mm->context.pgtable_list); 1097 spin_unlock_bh(&mm->context.list_lock); 1098 if (mask == 0) { 1099 pgtable_page_dtor(page); 1100 atomic_set(&page->_mapcount, -1); 1101 __free_page(page); 1102 } 1103} 1104 1105static void __page_table_free_rcu(void *table, unsigned bit) 1106{ 1107 struct page *page; 1108 1109 if (bit == FRAG_MASK) 1110 return page_table_free_pgste(table); 1111 /* Free 1K/2K page table fragment of a 4K page */ 1112 page = pfn_to_page(__pa(table) >> PAGE_SHIFT); 1113 if (atomic_xor_bits(&page->_mapcount, bit) == 0) { 1114 pgtable_page_dtor(page); 1115 atomic_set(&page->_mapcount, -1); 1116 __free_page(page); 1117 } 1118} 1119 1120void page_table_free_rcu(struct mmu_gather *tlb, unsigned long *table) 1121{ 1122 struct mm_struct *mm; 1123 struct page *page; 1124 unsigned int bit, mask; 1125 1126 mm = tlb->mm; 1127 page = pfn_to_page(__pa(table) >> PAGE_SHIFT); 1128 if (page_table_with_pgste(page)) { 1129 gmap_disconnect_pgtable(mm, table); 1130 table = (unsigned long *) (__pa(table) | FRAG_MASK); 1131 tlb_remove_table(tlb, table); 1132 return; 1133 } 1134 bit = 1 << ((__pa(table) & ~PAGE_MASK) / (PTRS_PER_PTE*sizeof(pte_t))); 1135 spin_lock_bh(&mm->context.list_lock); 1136 if ((atomic_read(&page->_mapcount) & FRAG_MASK) != FRAG_MASK) 1137 list_del(&page->lru); 1138 mask = atomic_xor_bits(&page->_mapcount, bit | (bit << 4)); 1139 if (mask & FRAG_MASK) 1140 list_add_tail(&page->lru, &mm->context.pgtable_list); 1141 spin_unlock_bh(&mm->context.list_lock); 1142 table = (unsigned long *) (__pa(table) | (bit << 4)); 1143 tlb_remove_table(tlb, table); 1144} 1145 1146static void __tlb_remove_table(void *_table) 1147{ 1148 const unsigned long mask = (FRAG_MASK << 4) | FRAG_MASK; 1149 void *table = (void *)((unsigned long) _table & ~mask); 1150 unsigned type = (unsigned long) _table & mask; 1151 1152 if (type) 1153 __page_table_free_rcu(table, type); 1154 else 1155 free_pages((unsigned long) table, ALLOC_ORDER); 1156} 1157 1158static void tlb_remove_table_smp_sync(void *arg) 1159{ 1160 /* Simply deliver the interrupt */ 1161} 1162 1163static void tlb_remove_table_one(void *table) 1164{ 1165 /* 1166 * This isn't an RCU grace period and hence the page-tables cannot be 1167 * assumed to be actually RCU-freed. 1168 * 1169 * It is however sufficient for software page-table walkers that rely 1170 * on IRQ disabling. See the comment near struct mmu_table_batch. 1171 */ 1172 smp_call_function(tlb_remove_table_smp_sync, NULL, 1); 1173 __tlb_remove_table(table); 1174} 1175 1176static void tlb_remove_table_rcu(struct rcu_head *head) 1177{ 1178 struct mmu_table_batch *batch; 1179 int i; 1180 1181 batch = container_of(head, struct mmu_table_batch, rcu); 1182 1183 for (i = 0; i < batch->nr; i++) 1184 __tlb_remove_table(batch->tables[i]); 1185 1186 free_page((unsigned long)batch); 1187} 1188 1189void tlb_table_flush(struct mmu_gather *tlb) 1190{ 1191 struct mmu_table_batch **batch = &tlb->batch; 1192 1193 if (*batch) { 1194 call_rcu_sched(&(*batch)->rcu, tlb_remove_table_rcu); 1195 *batch = NULL; 1196 } 1197} 1198 1199void tlb_remove_table(struct mmu_gather *tlb, void *table) 1200{ 1201 struct mmu_table_batch **batch = &tlb->batch; 1202 1203 tlb->mm->context.flush_mm = 1; 1204 if (*batch == NULL) { 1205 *batch = (struct mmu_table_batch *) 1206 __get_free_page(GFP_NOWAIT | __GFP_NOWARN); 1207 if (*batch == NULL) { 1208 __tlb_flush_mm_lazy(tlb->mm); 1209 tlb_remove_table_one(table); 1210 return; 1211 } 1212 (*batch)->nr = 0; 1213 } 1214 (*batch)->tables[(*batch)->nr++] = table; 1215 if ((*batch)->nr == MAX_TABLE_BATCH) 1216 tlb_flush_mmu(tlb); 1217} 1218 1219#ifdef CONFIG_TRANSPARENT_HUGEPAGE 1220static inline void thp_split_vma(struct vm_area_struct *vma) 1221{ 1222 unsigned long addr; 1223 1224 for (addr = vma->vm_start; addr < vma->vm_end; addr += PAGE_SIZE) 1225 follow_page(vma, addr, FOLL_SPLIT); 1226} 1227 1228static inline void thp_split_mm(struct mm_struct *mm) 1229{ 1230 struct vm_area_struct *vma; 1231 1232 for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) { 1233 thp_split_vma(vma); 1234 vma->vm_flags &= ~VM_HUGEPAGE; 1235 vma->vm_flags |= VM_NOHUGEPAGE; 1236 } 1237 mm->def_flags |= VM_NOHUGEPAGE; 1238} 1239#else 1240static inline void thp_split_mm(struct mm_struct *mm) 1241{ 1242} 1243#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1244 1245static unsigned long page_table_realloc_pmd(struct mmu_gather *tlb, 1246 struct mm_struct *mm, pud_t *pud, 1247 unsigned long addr, unsigned long end) 1248{ 1249 unsigned long next, *table, *new; 1250 struct page *page; 1251 pmd_t *pmd; 1252 1253 pmd = pmd_offset(pud, addr); 1254 do { 1255 next = pmd_addr_end(addr, end); 1256again: 1257 if (pmd_none_or_clear_bad(pmd)) 1258 continue; 1259 table = (unsigned long *) pmd_deref(*pmd); 1260 page = pfn_to_page(__pa(table) >> PAGE_SHIFT); 1261 if (page_table_with_pgste(page)) 1262 continue; 1263 /* Allocate new page table with pgstes */ 1264 new = page_table_alloc_pgste(mm, addr); 1265 if (!new) 1266 return -ENOMEM; 1267 1268 spin_lock(&mm->page_table_lock); 1269 if (likely((unsigned long *) pmd_deref(*pmd) == table)) { 1270 /* Nuke pmd entry pointing to the "short" page table */ 1271 pmdp_flush_lazy(mm, addr, pmd); 1272 pmd_clear(pmd); 1273 /* Copy ptes from old table to new table */ 1274 memcpy(new, table, PAGE_SIZE/2); 1275 clear_table(table, _PAGE_INVALID, PAGE_SIZE/2); 1276 /* Establish new table */ 1277 pmd_populate(mm, pmd, (pte_t *) new); 1278 /* Free old table with rcu, there might be a walker! */ 1279 page_table_free_rcu(tlb, table); 1280 new = NULL; 1281 } 1282 spin_unlock(&mm->page_table_lock); 1283 if (new) { 1284 page_table_free_pgste(new); 1285 goto again; 1286 } 1287 } while (pmd++, addr = next, addr != end); 1288 1289 return addr; 1290} 1291 1292static unsigned long page_table_realloc_pud(struct mmu_gather *tlb, 1293 struct mm_struct *mm, pgd_t *pgd, 1294 unsigned long addr, unsigned long end) 1295{ 1296 unsigned long next; 1297 pud_t *pud; 1298 1299 pud = pud_offset(pgd, addr); 1300 do { 1301 next = pud_addr_end(addr, end); 1302 if (pud_none_or_clear_bad(pud)) 1303 continue; 1304 next = page_table_realloc_pmd(tlb, mm, pud, addr, next); 1305 if (unlikely(IS_ERR_VALUE(next))) 1306 return next; 1307 } while (pud++, addr = next, addr != end); 1308 1309 return addr; 1310} 1311 1312static unsigned long page_table_realloc(struct mmu_gather *tlb, struct mm_struct *mm, 1313 unsigned long addr, unsigned long end) 1314{ 1315 unsigned long next; 1316 pgd_t *pgd; 1317 1318 pgd = pgd_offset(mm, addr); 1319 do { 1320 next = pgd_addr_end(addr, end); 1321 if (pgd_none_or_clear_bad(pgd)) 1322 continue; 1323 next = page_table_realloc_pud(tlb, mm, pgd, addr, next); 1324 if (unlikely(IS_ERR_VALUE(next))) 1325 return next; 1326 } while (pgd++, addr = next, addr != end); 1327 1328 return 0; 1329} 1330 1331/* 1332 * switch on pgstes for its userspace process (for kvm) 1333 */ 1334int s390_enable_sie(void) 1335{ 1336 struct task_struct *tsk = current; 1337 struct mm_struct *mm = tsk->mm; 1338 struct mmu_gather tlb; 1339 1340 /* Do we have pgstes? if yes, we are done */ 1341 if (mm_has_pgste(tsk->mm)) 1342 return 0; 1343 1344 down_write(&mm->mmap_sem); 1345 /* split thp mappings and disable thp for future mappings */ 1346 thp_split_mm(mm); 1347 /* Reallocate the page tables with pgstes */ 1348 tlb_gather_mmu(&tlb, mm, 0, TASK_SIZE); 1349 if (!page_table_realloc(&tlb, mm, 0, TASK_SIZE)) 1350 mm->context.has_pgste = 1; 1351 tlb_finish_mmu(&tlb, 0, TASK_SIZE); 1352 up_write(&mm->mmap_sem); 1353 return mm->context.has_pgste ? 0 : -ENOMEM; 1354} 1355EXPORT_SYMBOL_GPL(s390_enable_sie); 1356 1357#ifdef CONFIG_TRANSPARENT_HUGEPAGE 1358int pmdp_clear_flush_young(struct vm_area_struct *vma, unsigned long address, 1359 pmd_t *pmdp) 1360{ 1361 VM_BUG_ON(address & ~HPAGE_PMD_MASK); 1362 /* No need to flush TLB 1363 * On s390 reference bits are in storage key and never in TLB */ 1364 return pmdp_test_and_clear_young(vma, address, pmdp); 1365} 1366 1367int pmdp_set_access_flags(struct vm_area_struct *vma, 1368 unsigned long address, pmd_t *pmdp, 1369 pmd_t entry, int dirty) 1370{ 1371 VM_BUG_ON(address & ~HPAGE_PMD_MASK); 1372 1373 if (pmd_same(*pmdp, entry)) 1374 return 0; 1375 pmdp_invalidate(vma, address, pmdp); 1376 set_pmd_at(vma->vm_mm, address, pmdp, entry); 1377 return 1; 1378} 1379 1380static void pmdp_splitting_flush_sync(void *arg) 1381{ 1382 /* Simply deliver the interrupt */ 1383} 1384 1385void pmdp_splitting_flush(struct vm_area_struct *vma, unsigned long address, 1386 pmd_t *pmdp) 1387{ 1388 VM_BUG_ON(address & ~HPAGE_PMD_MASK); 1389 if (!test_and_set_bit(_SEGMENT_ENTRY_SPLIT_BIT, 1390 (unsigned long *) pmdp)) { 1391 /* need to serialize against gup-fast (IRQ disabled) */ 1392 smp_call_function(pmdp_splitting_flush_sync, NULL, 1); 1393 } 1394} 1395 1396void pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp, 1397 pgtable_t pgtable) 1398{ 1399 struct list_head *lh = (struct list_head *) pgtable; 1400 1401 assert_spin_locked(pmd_lockptr(mm, pmdp)); 1402 1403 /* FIFO */ 1404 if (!pmd_huge_pte(mm, pmdp)) 1405 INIT_LIST_HEAD(lh); 1406 else 1407 list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp)); 1408 pmd_huge_pte(mm, pmdp) = pgtable; 1409} 1410 1411pgtable_t pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp) 1412{ 1413 struct list_head *lh; 1414 pgtable_t pgtable; 1415 pte_t *ptep; 1416 1417 assert_spin_locked(pmd_lockptr(mm, pmdp)); 1418 1419 /* FIFO */ 1420 pgtable = pmd_huge_pte(mm, pmdp); 1421 lh = (struct list_head *) pgtable; 1422 if (list_empty(lh)) 1423 pmd_huge_pte(mm, pmdp) = NULL; 1424 else { 1425 pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next; 1426 list_del(lh); 1427 } 1428 ptep = (pte_t *) pgtable; 1429 pte_val(*ptep) = _PAGE_INVALID; 1430 ptep++; 1431 pte_val(*ptep) = _PAGE_INVALID; 1432 return pgtable; 1433} 1434#endif /* CONFIG_TRANSPARENT_HUGEPAGE */ 1435