arch/um/kernel/tlb.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
   4  */
   5
   6 #include <linux/mm.h>
   7 #include <linux/module.h>
   8 #include <linux/sched/signal.h>
   9
  10 #include <asm/tlbflush.h>
  11 #include <as-layout.h>
  12 #include <mem_user.h>
  13 #include <os.h>
  14 #include <skas.h>
  15 #include <kern_util.h>
  16
  17 struct host_vm_change {
  18         struct host_vm_op {
  19                 enum { NONE, MMAP, MUNMAP, MPROTECT } type;
  20                 union {
  21                         struct {
  22                                 unsigned long addr;
  23                                 unsigned long len;
  24                                 unsigned int prot;
  25                                 int fd;
  26                                 __u64 offset;
  27                         } mmap;
  28                         struct {
  29                                 unsigned long addr;
  30                                 unsigned long len;
  31                         } munmap;
  32                         struct {
  33                                 unsigned long addr;
  34                                 unsigned long len;
  35                                 unsigned int prot;
  36                         } mprotect;
  37                 } u;
  38         } ops[1];
  39         int userspace;
  40         int index;
  41         struct mm_struct *mm;
  42         void *data;
  43         int force;
  44 };
  45
  46 #define INIT_HVC(mm, force, userspace) \
  47         ((struct host_vm_change) \
  48          { .ops         = { { .type = NONE } }, \
  49            .mm          = mm, \
  50            .data        = NULL, \
  51            .userspace   = userspace, \
  52            .index       = 0, \
  53            .force       = force })
  54
  55 static void report_enomem(void)
  56 {
  57         printk(KERN_ERR "UML ran out of memory on the host side! "
  58                         "This can happen due to a memory limitation or "
  59                         "vm.max_map_count has been reached.\n");
  60 }
  61
  62 static int do_ops(struct host_vm_change *hvc, int end,
  63                   int finished)
  64 {
  65         struct host_vm_op *op;
  66         int i, ret = 0;
  67
  68         for (i = 0; i < end && !ret; i++) {
  69                 op = &hvc->ops[i];
  70                 switch (op->type) {
  71                 case MMAP:
  72                         if (hvc->userspace)
  73                                 ret = map(&hvc->mm->context.id, op->u.mmap.addr,
  74                                           op->u.mmap.len, op->u.mmap.prot,
  75                                           op->u.mmap.fd,
  76                                           op->u.mmap.offset, finished,
  77                                           &hvc->data);
  78                         else
  79                                 map_memory(op->u.mmap.addr, op->u.mmap.offset,
  80                                            op->u.mmap.len, 1, 1, 1);
  81                         break;
  82                 case MUNMAP:
  83                         if (hvc->userspace)
  84                                 ret = unmap(&hvc->mm->context.id,
  85                                             op->u.munmap.addr,
  86                                             op->u.munmap.len, finished,
  87                                             &hvc->data);
  88                         else
  89                                 ret = os_unmap_memory(
  90                                         (void *) op->u.munmap.addr,
  91                                                       op->u.munmap.len);
  92
  93                         break;
  94                 case MPROTECT:
  95                         if (hvc->userspace)
  96                                 ret = protect(&hvc->mm->context.id,
  97                                               op->u.mprotect.addr,
  98                                               op->u.mprotect.len,
  99                                               op->u.mprotect.prot,
 100                                               finished, &hvc->data);
 101                         else
 102                                 ret = os_protect_memory(
 103                                         (void *) op->u.mprotect.addr,
 104                                                         op->u.mprotect.len,
 105                                                         1, 1, 1);
 106                         break;
 107                 default:
 108                         printk(KERN_ERR "Unknown op type %d in do_ops\n",
 109                                op->type);
 110                         BUG();
 111                         break;
 112                 }
 113         }
 114
 115         if (ret == -ENOMEM)
 116                 report_enomem();
 117
 118         return ret;
 119 }
 120
 121 static int add_mmap(unsigned long virt, unsigned long phys, unsigned long len,
 122                     unsigned int prot, struct host_vm_change *hvc)
 123 {
 124         __u64 offset;
 125         struct host_vm_op *last;
 126         int fd = -1, ret = 0;
 127
 128         if (virt + len > STUB_START && virt < STUB_END)
 129                 return -EINVAL;
 130
 131         if (hvc->userspace)
 132                 fd = phys_mapping(phys, &offset);
 133         else
 134                 offset = phys;
 135         if (hvc->index != 0) {
 136                 last = &hvc->ops[hvc->index - 1];
 137                 if ((last->type == MMAP) &&
 138                    (last->u.mmap.addr + last->u.mmap.len == virt) &&
 139                    (last->u.mmap.prot == prot) && (last->u.mmap.fd == fd) &&
 140                    (last->u.mmap.offset + last->u.mmap.len == offset)) {
 141                         last->u.mmap.len += len;
 142                         return 0;
 143                 }
 144         }
 145
 146         if (hvc->index == ARRAY_SIZE(hvc->ops)) {
 147                 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
 148                 hvc->index = 0;
 149         }
 150
 151         hvc->ops[hvc->index++] = ((struct host_vm_op)
 152                                   { .type       = MMAP,
 153                                     .u = { .mmap = { .addr      = virt,
 154                                                      .len       = len,
 155                                                      .prot      = prot,
 156                                                      .fd        = fd,
 157                                                      .offset    = offset }
 158                            } });
 159         return ret;
 160 }
 161
 162 static int add_munmap(unsigned long addr, unsigned long len,
 163                       struct host_vm_change *hvc)
 164 {
 165         struct host_vm_op *last;
 166         int ret = 0;
 167
 168         if (addr + len > STUB_START && addr < STUB_END)
 169                 return -EINVAL;
 170
 171         if (hvc->index != 0) {
 172                 last = &hvc->ops[hvc->index - 1];
 173                 if ((last->type == MUNMAP) &&
 174                    (last->u.munmap.addr + last->u.mmap.len == addr)) {
 175                         last->u.munmap.len += len;
 176                         return 0;
 177                 }
 178         }
 179
 180         if (hvc->index == ARRAY_SIZE(hvc->ops)) {
 181                 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
 182                 hvc->index = 0;
 183         }
 184
 185         hvc->ops[hvc->index++] = ((struct host_vm_op)
 186                                   { .type       = MUNMAP,
 187                                     .u = { .munmap = { .addr    = addr,
 188                                                        .len     = len } } });
 189         return ret;
 190 }
 191
 192 static int add_mprotect(unsigned long addr, unsigned long len,
 193                         unsigned int prot, struct host_vm_change *hvc)
 194 {
 195         struct host_vm_op *last;
 196         int ret = 0;
 197
 198         if (addr + len > STUB_START && addr < STUB_END)
 199                 return -EINVAL;
 200
 201         if (hvc->index != 0) {
 202                 last = &hvc->ops[hvc->index - 1];
 203                 if ((last->type == MPROTECT) &&
 204                    (last->u.mprotect.addr + last->u.mprotect.len == addr) &&
 205                    (last->u.mprotect.prot == prot)) {
 206                         last->u.mprotect.len += len;
 207                         return 0;
 208                 }
 209         }
 210
 211         if (hvc->index == ARRAY_SIZE(hvc->ops)) {
 212                 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
 213                 hvc->index = 0;
 214         }
 215
 216         hvc->ops[hvc->index++] = ((struct host_vm_op)
 217                                   { .type       = MPROTECT,
 218                                     .u = { .mprotect = { .addr  = addr,
 219                                                          .len   = len,
 220                                                          .prot  = prot } } });
 221         return ret;
 222 }
 223
 224 #define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
 225
 226 static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
 227                                    unsigned long end,
 228                                    struct host_vm_change *hvc)
 229 {
 230         pte_t *pte;
 231         int r, w, x, prot, ret = 0;
 232
 233         pte = pte_offset_kernel(pmd, addr);
 234         do {
 235                 if ((addr >= STUB_START) && (addr < STUB_END))
 236                         continue;
 237
 238                 r = pte_read(*pte);
 239                 w = pte_write(*pte);
 240                 x = pte_exec(*pte);
 241                 if (!pte_young(*pte)) {
 242                         r = 0;
 243                         w = 0;
 244                 } else if (!pte_dirty(*pte))
 245                         w = 0;
 246
 247                 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
 248                         (x ? UM_PROT_EXEC : 0));
 249                 if (hvc->force || pte_newpage(*pte)) {
 250                         if (pte_present(*pte)) {
 251                                 if (pte_newpage(*pte))
 252                                         ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
 253                                                        PAGE_SIZE, prot, hvc);
 254                         } else
 255                                 ret = add_munmap(addr, PAGE_SIZE, hvc);
 256                 } else if (pte_newprot(*pte))
 257                         ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
 258                 *pte = pte_mkuptodate(*pte);
 259         } while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
 260         return ret;
 261 }
 262
 263 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
 264                                    unsigned long end,
 265                                    struct host_vm_change *hvc)
 266 {
 267         pmd_t *pmd;
 268         unsigned long next;
 269         int ret = 0;
 270
 271         pmd = pmd_offset(pud, addr);
 272         do {
 273                 next = pmd_addr_end(addr, end);
 274                 if (!pmd_present(*pmd)) {
 275                         if (hvc->force || pmd_newpage(*pmd)) {
 276                                 ret = add_munmap(addr, next - addr, hvc);
 277                                 pmd_mkuptodate(*pmd);
 278                         }
 279                 }
 280                 else ret = update_pte_range(pmd, addr, next, hvc);
 281         } while (pmd++, addr = next, ((addr < end) && !ret));
 282         return ret;
 283 }
 284
 285 static inline int update_pud_range(p4d_t *p4d, unsigned long addr,
 286                                    unsigned long end,
 287                                    struct host_vm_change *hvc)
 288 {
 289         pud_t *pud;
 290         unsigned long next;
 291         int ret = 0;
 292
 293         pud = pud_offset(p4d, addr);
 294         do {
 295                 next = pud_addr_end(addr, end);
 296                 if (!pud_present(*pud)) {
 297                         if (hvc->force || pud_newpage(*pud)) {
 298                                 ret = add_munmap(addr, next - addr, hvc);
 299                                 pud_mkuptodate(*pud);
 300                         }
 301                 }
 302                 else ret = update_pmd_range(pud, addr, next, hvc);
 303         } while (pud++, addr = next, ((addr < end) && !ret));
 304         return ret;
 305 }
 306
 307 static inline int update_p4d_range(pgd_t *pgd, unsigned long addr,
 308                                    unsigned long end,
 309                                    struct host_vm_change *hvc)
 310 {
 311         p4d_t *p4d;
 312         unsigned long next;
 313         int ret = 0;
 314
 315         p4d = p4d_offset(pgd, addr);
 316         do {
 317                 next = p4d_addr_end(addr, end);
 318                 if (!p4d_present(*p4d)) {
 319                         if (hvc->force || p4d_newpage(*p4d)) {
 320                                 ret = add_munmap(addr, next - addr, hvc);
 321                                 p4d_mkuptodate(*p4d);
 322                         }
 323                 } else
 324                         ret = update_pud_range(p4d, addr, next, hvc);
 325         } while (p4d++, addr = next, ((addr < end) && !ret));
 326         return ret;
 327 }
 328
 329 void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
 330                       unsigned long end_addr, int force)
 331 {
 332         pgd_t *pgd;
 333         struct host_vm_change hvc;
 334         unsigned long addr = start_addr, next;
 335         int ret = 0, userspace = 1;
 336
 337         hvc = INIT_HVC(mm, force, userspace);
 338         pgd = pgd_offset(mm, addr);
 339         do {
 340                 next = pgd_addr_end(addr, end_addr);
 341                 if (!pgd_present(*pgd)) {
 342                         if (force || pgd_newpage(*pgd)) {
 343                                 ret = add_munmap(addr, next - addr, &hvc);
 344                                 pgd_mkuptodate(*pgd);
 345                         }
 346                 } else
 347                         ret = update_p4d_range(pgd, addr, next, &hvc);
 348         } while (pgd++, addr = next, ((addr < end_addr) && !ret));
 349
 350         if (!ret)
 351                 ret = do_ops(&hvc, hvc.index, 1);
 352
 353         /* This is not an else because ret is modified above */
 354         if (ret) {
 355                 struct mm_id *mm_idp = &current->mm->context.id;
 356
 357                 printk(KERN_ERR "fix_range_common: failed, killing current "
 358                        "process: %d\n", task_tgid_vnr(current));
 359                 mm_idp->kill = 1;
 360         }
 361 }
 362
 363 static int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
 364 {
 365         struct mm_struct *mm;
 366         pgd_t *pgd;
 367         p4d_t *p4d;
 368         pud_t *pud;
 369         pmd_t *pmd;
 370         pte_t *pte;
 371         unsigned long addr, last;
 372         int updated = 0, err = 0, force = 0, userspace = 0;
 373         struct host_vm_change hvc;
 374
 375         mm = &init_mm;
 376         hvc = INIT_HVC(mm, force, userspace);
 377         for (addr = start; addr < end;) {
 378                 pgd = pgd_offset(mm, addr);
 379                 if (!pgd_present(*pgd)) {
 380                         last = ADD_ROUND(addr, PGDIR_SIZE);
 381                         if (last > end)
 382                                 last = end;
 383                         if (pgd_newpage(*pgd)) {
 384                                 updated = 1;
 385                                 err = add_munmap(addr, last - addr, &hvc);
 386                                 if (err < 0)
 387                                         panic("munmap failed, errno = %d\n",
 388                                               -err);
 389                         }
 390                         addr = last;
 391                         continue;
 392                 }
 393
 394                 p4d = p4d_offset(pgd, addr);
 395                 if (!p4d_present(*p4d)) {
 396                         last = ADD_ROUND(addr, P4D_SIZE);
 397                         if (last > end)
 398                                 last = end;
 399                         if (p4d_newpage(*p4d)) {
 400                                 updated = 1;
 401                                 err = add_munmap(addr, last - addr, &hvc);
 402                                 if (err < 0)
 403                                         panic("munmap failed, errno = %d\n",
 404                                               -err);
 405                         }
 406                         addr = last;
 407                         continue;
 408                 }
 409
 410                 pud = pud_offset(p4d, addr);
 411                 if (!pud_present(*pud)) {
 412                         last = ADD_ROUND(addr, PUD_SIZE);
 413                         if (last > end)
 414                                 last = end;
 415                         if (pud_newpage(*pud)) {
 416                                 updated = 1;
 417                                 err = add_munmap(addr, last - addr, &hvc);
 418                                 if (err < 0)
 419                                         panic("munmap failed, errno = %d\n",
 420                                               -err);
 421                         }
 422                         addr = last;
 423                         continue;
 424                 }
 425
 426                 pmd = pmd_offset(pud, addr);
 427                 if (!pmd_present(*pmd)) {
 428                         last = ADD_ROUND(addr, PMD_SIZE);
 429                         if (last > end)
 430                                 last = end;
 431                         if (pmd_newpage(*pmd)) {
 432                                 updated = 1;
 433                                 err = add_munmap(addr, last - addr, &hvc);
 434                                 if (err < 0)
 435                                         panic("munmap failed, errno = %d\n",
 436                                               -err);
 437                         }
 438                         addr = last;
 439                         continue;
 440                 }
 441
 442                 pte = pte_offset_kernel(pmd, addr);
 443                 if (!pte_present(*pte) || pte_newpage(*pte)) {
 444                         updated = 1;
 445                         err = add_munmap(addr, PAGE_SIZE, &hvc);
 446                         if (err < 0)
 447                                 panic("munmap failed, errno = %d\n",
 448                                       -err);
 449                         if (pte_present(*pte))
 450                                 err = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
 451                                                PAGE_SIZE, 0, &hvc);
 452                 }
 453                 else if (pte_newprot(*pte)) {
 454                         updated = 1;
 455                         err = add_mprotect(addr, PAGE_SIZE, 0, &hvc);
 456                 }
 457                 addr += PAGE_SIZE;
 458         }
 459         if (!err)
 460                 err = do_ops(&hvc, hvc.index, 1);
 461
 462         if (err < 0)
 463                 panic("flush_tlb_kernel failed, errno = %d\n", err);
 464         return updated;
 465 }
 466
 467 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
 468 {
 469         pgd_t *pgd;
 470         p4d_t *p4d;
 471         pud_t *pud;
 472         pmd_t *pmd;
 473         pte_t *pte;
 474         struct mm_struct *mm = vma->vm_mm;
 475         void *flush = NULL;
 476         int r, w, x, prot, err = 0;
 477         struct mm_id *mm_id;
 478
 479         address &= PAGE_MASK;
 480
 481         if (address >= STUB_START && address < STUB_END)
 482                 goto kill;
 483
 484         pgd = pgd_offset(mm, address);
 485         if (!pgd_present(*pgd))
 486                 goto kill;
 487
 488         p4d = p4d_offset(pgd, address);
 489         if (!p4d_present(*p4d))
 490                 goto kill;
 491
 492         pud = pud_offset(p4d, address);
 493         if (!pud_present(*pud))
 494                 goto kill;
 495
 496         pmd = pmd_offset(pud, address);
 497         if (!pmd_present(*pmd))
 498                 goto kill;
 499
 500         pte = pte_offset_kernel(pmd, address);
 501
 502         r = pte_read(*pte);
 503         w = pte_write(*pte);
 504         x = pte_exec(*pte);
 505         if (!pte_young(*pte)) {
 506                 r = 0;
 507                 w = 0;
 508         } else if (!pte_dirty(*pte)) {
 509                 w = 0;
 510         }
 511
 512         mm_id = &mm->context.id;
 513         prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
 514                 (x ? UM_PROT_EXEC : 0));
 515         if (pte_newpage(*pte)) {
 516                 if (pte_present(*pte)) {
 517                         unsigned long long offset;
 518                         int fd;
 519
 520                         fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
 521                         err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
 522                                   1, &flush);
 523                 }
 524                 else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
 525         }
 526         else if (pte_newprot(*pte))
 527                 err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
 528
 529         if (err) {
 530                 if (err == -ENOMEM)
 531                         report_enomem();
 532
 533                 goto kill;
 534         }
 535
 536         *pte = pte_mkuptodate(*pte);
 537
 538         return;
 539
 540 kill:
 541         printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
 542         force_sig(SIGKILL);
 543 }
 544
 545 void flush_tlb_all(void)
 546 {
 547         /*
 548          * Don't bother flushing if this address space is about to be
 549          * destroyed.
 550          */
 551         if (atomic_read(&current->mm->mm_users) == 0)
 552                 return;
 553
 554         flush_tlb_mm(current->mm);
 555 }
 556
 557 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 558 {
 559         flush_tlb_kernel_range_common(start, end);
 560 }
 561
 562 void flush_tlb_kernel_vm(void)
 563 {
 564         flush_tlb_kernel_range_common(start_vm, end_vm);
 565 }
 566
 567 void __flush_tlb_one(unsigned long addr)
 568 {
 569         flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
 570 }
 571
 572 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
 573                       unsigned long end_addr, int force)
 574 {
 575         /*
 576          * Don't bother flushing if this address space is about to be
 577          * destroyed.
 578          */
 579         if (atomic_read(&mm->mm_users) == 0)
 580                 return;
 581
 582         fix_range_common(mm, start_addr, end_addr, force);
 583 }
 584
 585 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 586                      unsigned long end)
 587 {
 588         if (vma->vm_mm == NULL)
 589                 flush_tlb_kernel_range_common(start, end);
 590         else fix_range(vma->vm_mm, start, end, 0);
 591 }
 592 EXPORT_SYMBOL(flush_tlb_range);
 593
 594 void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
 595                         unsigned long end)
 596 {
 597         fix_range(mm, start, end, 0);
 598 }
 599
 600 void flush_tlb_mm(struct mm_struct *mm)
 601 {
 602         struct vm_area_struct *vma = mm->mmap;
 603
 604         while (vma != NULL) {
 605                 fix_range(mm, vma->vm_start, vma->vm_end, 0);
 606                 vma = vma->vm_next;
 607         }
 608 }
 609
 610 void force_flush_all(void)
 611 {
 612         struct mm_struct *mm = current->mm;
 613         struct vm_area_struct *vma = mm->mmap;
 614
 615         while (vma != NULL) {
 616                 fix_range(mm, vma->vm_start, vma->vm_end, 1);
 617                 vma = vma->vm_next;
 618         }
 619 }