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