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