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