arch/um/kernel/tlb.c

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