f4a0e407eee4d66e9fb8b62db6a3e36110d2854a
[linux-2.6.git] / 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 "asm/pgtable.h"
8 #include "asm/tlbflush.h"
9 #include "as-layout.h"
10 #include "mem_user.h"
11 #include "os.h"
12 #include "skas.h"
13 #include "tlb.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                 r = pte_read(*pte);
187                 w = pte_write(*pte);
188                 x = pte_exec(*pte);
189                 if (!pte_young(*pte)) {
190                         r = 0;
191                         w = 0;
192                 } else if (!pte_dirty(*pte)) {
193                         w = 0;
194                 }
195                 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
196                         (x ? UM_PROT_EXEC : 0));
197                 if (hvc->force || pte_newpage(*pte)) {
198                         if (pte_present(*pte))
199                                 ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
200                                                PAGE_SIZE, prot, hvc);
201                         else ret = add_munmap(addr, PAGE_SIZE, hvc);
202                 }
203                 else if (pte_newprot(*pte))
204                         ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
205                 *pte = pte_mkuptodate(*pte);
206         } while (pte++, addr += PAGE_SIZE, ((addr != end) && !ret));
207         return ret;
208 }
209
210 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
211                                    unsigned long end,
212                                    struct host_vm_change *hvc)
213 {
214         pmd_t *pmd;
215         unsigned long next;
216         int ret = 0;
217
218         pmd = pmd_offset(pud, addr);
219         do {
220                 next = pmd_addr_end(addr, end);
221                 if (!pmd_present(*pmd)) {
222                         if (hvc->force || pmd_newpage(*pmd)) {
223                                 ret = add_munmap(addr, next - addr, hvc);
224                                 pmd_mkuptodate(*pmd);
225                         }
226                 }
227                 else ret = update_pte_range(pmd, addr, next, hvc);
228         } while (pmd++, addr = next, ((addr != end) && !ret));
229         return ret;
230 }
231
232 static inline int update_pud_range(pgd_t *pgd, unsigned long addr,
233                                    unsigned long end,
234                                    struct host_vm_change *hvc)
235 {
236         pud_t *pud;
237         unsigned long next;
238         int ret = 0;
239
240         pud = pud_offset(pgd, addr);
241         do {
242                 next = pud_addr_end(addr, end);
243                 if (!pud_present(*pud)) {
244                         if (hvc->force || pud_newpage(*pud)) {
245                                 ret = add_munmap(addr, next - addr, hvc);
246                                 pud_mkuptodate(*pud);
247                         }
248                 }
249                 else ret = update_pmd_range(pud, addr, next, hvc);
250         } while (pud++, addr = next, ((addr != end) && !ret));
251         return ret;
252 }
253
254 void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
255                       unsigned long end_addr, int force)
256 {
257         pgd_t *pgd;
258         struct host_vm_change hvc;
259         unsigned long addr = start_addr, next;
260         int ret = 0;
261
262         hvc = INIT_HVC(mm, force);
263         pgd = pgd_offset(mm, addr);
264         do {
265                 next = pgd_addr_end(addr, end_addr);
266                 if (!pgd_present(*pgd)) {
267                         if (force || pgd_newpage(*pgd)) {
268                                 ret = add_munmap(addr, next - addr, &hvc);
269                                 pgd_mkuptodate(*pgd);
270                         }
271                 }
272                 else ret = update_pud_range(pgd, addr, next, &hvc);
273         } while (pgd++, addr = next, ((addr != end_addr) && !ret));
274
275         if (!ret)
276                 ret = do_ops(&hvc, hvc.index, 1);
277
278         /* This is not an else because ret is modified above */
279         if (ret) {
280                 printk(KERN_ERR "fix_range_common: failed, killing current "
281                        "process\n");
282                 force_sig(SIGKILL, current);
283         }
284 }
285
286 int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
287 {
288         struct mm_struct *mm;
289         pgd_t *pgd;
290         pud_t *pud;
291         pmd_t *pmd;
292         pte_t *pte;
293         unsigned long addr, last;
294         int updated = 0, err;
295
296         mm = &init_mm;
297         for (addr = start; addr < end;) {
298                 pgd = pgd_offset(mm, addr);
299                 if (!pgd_present(*pgd)) {
300                         last = ADD_ROUND(addr, PGDIR_SIZE);
301                         if (last > end)
302                                 last = end;
303                         if (pgd_newpage(*pgd)) {
304                                 updated = 1;
305                                 err = os_unmap_memory((void *) addr,
306                                                       last - addr);
307                                 if (err < 0)
308                                         panic("munmap failed, errno = %d\n",
309                                               -err);
310                         }
311                         addr = last;
312                         continue;
313                 }
314
315                 pud = pud_offset(pgd, addr);
316                 if (!pud_present(*pud)) {
317                         last = ADD_ROUND(addr, PUD_SIZE);
318                         if (last > end)
319                                 last = end;
320                         if (pud_newpage(*pud)) {
321                                 updated = 1;
322                                 err = os_unmap_memory((void *) addr,
323                                                       last - addr);
324                                 if (err < 0)
325                                         panic("munmap failed, errno = %d\n",
326                                               -err);
327                         }
328                         addr = last;
329                         continue;
330                 }
331
332                 pmd = pmd_offset(pud, addr);
333                 if (!pmd_present(*pmd)) {
334                         last = ADD_ROUND(addr, PMD_SIZE);
335                         if (last > end)
336                                 last = end;
337                         if (pmd_newpage(*pmd)) {
338                                 updated = 1;
339                                 err = os_unmap_memory((void *) addr,
340                                                       last - addr);
341                                 if (err < 0)
342                                         panic("munmap failed, errno = %d\n",
343                                               -err);
344                         }
345                         addr = last;
346                         continue;
347                 }
348
349                 pte = pte_offset_kernel(pmd, addr);
350                 if (!pte_present(*pte) || pte_newpage(*pte)) {
351                         updated = 1;
352                         err = os_unmap_memory((void *) addr,
353                                               PAGE_SIZE);
354                         if (err < 0)
355                                 panic("munmap failed, errno = %d\n",
356                                       -err);
357                         if (pte_present(*pte))
358                                 map_memory(addr,
359                                            pte_val(*pte) & PAGE_MASK,
360                                            PAGE_SIZE, 1, 1, 1);
361                 }
362                 else if (pte_newprot(*pte)) {
363                         updated = 1;
364                         os_protect_memory((void *) addr, PAGE_SIZE, 1, 1, 1);
365                 }
366                 addr += PAGE_SIZE;
367         }
368         return updated;
369 }
370
371 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
372 {
373         pgd_t *pgd;
374         pud_t *pud;
375         pmd_t *pmd;
376         pte_t *pte;
377         struct mm_struct *mm = vma->vm_mm;
378         void *flush = NULL;
379         int r, w, x, prot, err = 0;
380         struct mm_id *mm_id;
381
382         address &= PAGE_MASK;
383         pgd = pgd_offset(mm, address);
384         if (!pgd_present(*pgd))
385                 goto kill;
386
387         pud = pud_offset(pgd, address);
388         if (!pud_present(*pud))
389                 goto kill;
390
391         pmd = pmd_offset(pud, address);
392         if (!pmd_present(*pmd))
393                 goto kill;
394
395         pte = pte_offset_kernel(pmd, address);
396
397         r = pte_read(*pte);
398         w = pte_write(*pte);
399         x = pte_exec(*pte);
400         if (!pte_young(*pte)) {
401                 r = 0;
402                 w = 0;
403         } else if (!pte_dirty(*pte)) {
404                 w = 0;
405         }
406
407         mm_id = &mm->context.id;
408         prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
409                 (x ? UM_PROT_EXEC : 0));
410         if (pte_newpage(*pte)) {
411                 if (pte_present(*pte)) {
412                         unsigned long long offset;
413                         int fd;
414
415                         fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
416                         err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
417                                   1, &flush);
418                 }
419                 else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
420         }
421         else if (pte_newprot(*pte))
422                 err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
423
424         if (err)
425                 goto kill;
426
427         *pte = pte_mkuptodate(*pte);
428
429         return;
430
431 kill:
432         printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
433         force_sig(SIGKILL, current);
434 }
435
436 pgd_t *pgd_offset_proc(struct mm_struct *mm, unsigned long address)
437 {
438         return pgd_offset(mm, address);
439 }
440
441 pud_t *pud_offset_proc(pgd_t *pgd, unsigned long address)
442 {
443         return pud_offset(pgd, address);
444 }
445
446 pmd_t *pmd_offset_proc(pud_t *pud, unsigned long address)
447 {
448         return pmd_offset(pud, address);
449 }
450
451 pte_t *pte_offset_proc(pmd_t *pmd, unsigned long address)
452 {
453         return pte_offset_kernel(pmd, address);
454 }
455
456 pte_t *addr_pte(struct task_struct *task, unsigned long addr)
457 {
458         pgd_t *pgd = pgd_offset(task->mm, addr);
459         pud_t *pud = pud_offset(pgd, addr);
460         pmd_t *pmd = pmd_offset(pud, addr);
461
462         return pte_offset_map(pmd, addr);
463 }
464
465 void flush_tlb_all(void)
466 {
467         flush_tlb_mm(current->mm);
468 }
469
470 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
471 {
472         flush_tlb_kernel_range_common(start, end);
473 }
474
475 void flush_tlb_kernel_vm(void)
476 {
477         flush_tlb_kernel_range_common(start_vm, end_vm);
478 }
479
480 void __flush_tlb_one(unsigned long addr)
481 {
482         flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
483 }
484
485 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
486                       unsigned long end_addr, int force)
487 {
488         if (!proc_mm && (end_addr > STUB_START))
489                 end_addr = STUB_START;
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(struct mm_struct *mm)
503 {
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         end = proc_mm ? task_size : STUB_START;
514         fix_range(mm, 0, end, 0);
515 }
516
517 void force_flush_all(void)
518 {
519         struct mm_struct *mm = current->mm;
520         struct vm_area_struct *vma = mm->mmap;
521
522         while (vma != NULL) {
523                 fix_range(mm, vma->vm_start, vma->vm_end, 1);
524                 vma = vma->vm_next;
525         }
526 }