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