2 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
3 * Licensed under the GPL
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>
15 #include <kern_util.h>
17 struct host_vm_change {
19 enum { NONE, MMAP, MUNMAP, MPROTECT } type;
45 #define INIT_HVC(mm, force) \
46 ((struct host_vm_change) \
47 { .ops = { { .type = NONE } }, \
48 .id = &mm->context.id, \
53 static int do_ops(struct host_vm_change *hvc, int end,
56 struct host_vm_op *op;
59 for (i = 0; i < end && !ret; i++) {
63 ret = map(hvc->id, op->u.mmap.addr, op->u.mmap.len,
64 op->u.mmap.prot, op->u.mmap.fd,
65 op->u.mmap.offset, finished, &hvc->data);
68 ret = unmap(hvc->id, op->u.munmap.addr,
69 op->u.munmap.len, finished, &hvc->data);
72 ret = protect(hvc->id, op->u.mprotect.addr,
73 op->u.mprotect.len, op->u.mprotect.prot,
74 finished, &hvc->data);
77 printk(KERN_ERR "Unknown op type %d in do_ops\n",
87 static int add_mmap(unsigned long virt, unsigned long phys, unsigned long len,
88 unsigned int prot, struct host_vm_change *hvc)
91 struct host_vm_op *last;
94 fd = phys_mapping(phys, &offset);
95 if (hvc->index != 0) {
96 last = &hvc->ops[hvc->index - 1];
97 if ((last->type == MMAP) &&
98 (last->u.mmap.addr + last->u.mmap.len == virt) &&
99 (last->u.mmap.prot == prot) && (last->u.mmap.fd == fd) &&
100 (last->u.mmap.offset + last->u.mmap.len == offset)) {
101 last->u.mmap.len += len;
106 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
107 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
111 hvc->ops[hvc->index++] = ((struct host_vm_op)
113 .u = { .mmap = { .addr = virt,
122 static int add_munmap(unsigned long addr, unsigned long len,
123 struct host_vm_change *hvc)
125 struct host_vm_op *last;
128 if ((addr >= STUB_START) && (addr < STUB_END))
131 if (hvc->index != 0) {
132 last = &hvc->ops[hvc->index - 1];
133 if ((last->type == MUNMAP) &&
134 (last->u.munmap.addr + last->u.mmap.len == addr)) {
135 last->u.munmap.len += len;
140 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
141 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
145 hvc->ops[hvc->index++] = ((struct host_vm_op)
147 .u = { .munmap = { .addr = addr,
152 static int add_mprotect(unsigned long addr, unsigned long len,
153 unsigned int prot, struct host_vm_change *hvc)
155 struct host_vm_op *last;
158 if (hvc->index != 0) {
159 last = &hvc->ops[hvc->index - 1];
160 if ((last->type == MPROTECT) &&
161 (last->u.mprotect.addr + last->u.mprotect.len == addr) &&
162 (last->u.mprotect.prot == prot)) {
163 last->u.mprotect.len += len;
168 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
169 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
173 hvc->ops[hvc->index++] = ((struct host_vm_op)
175 .u = { .mprotect = { .addr = addr,
181 #define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
183 static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
185 struct host_vm_change *hvc)
188 int r, w, x, prot, ret = 0;
190 pte = pte_offset_kernel(pmd, addr);
192 if ((addr >= STUB_START) && (addr < STUB_END))
198 if (!pte_young(*pte)) {
201 } else if (!pte_dirty(*pte))
204 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
205 (x ? UM_PROT_EXEC : 0));
206 if (hvc->force || pte_newpage(*pte)) {
207 if (pte_present(*pte))
208 ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
209 PAGE_SIZE, prot, hvc);
211 ret = add_munmap(addr, PAGE_SIZE, hvc);
212 } else if (pte_newprot(*pte))
213 ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
214 *pte = pte_mkuptodate(*pte);
215 } while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
219 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
221 struct host_vm_change *hvc)
227 pmd = pmd_offset(pud, addr);
229 next = pmd_addr_end(addr, end);
230 if (!pmd_present(*pmd)) {
231 if (hvc->force || pmd_newpage(*pmd)) {
232 ret = add_munmap(addr, next - addr, hvc);
233 pmd_mkuptodate(*pmd);
236 else ret = update_pte_range(pmd, addr, next, hvc);
237 } while (pmd++, addr = next, ((addr < end) && !ret));
241 static inline int update_pud_range(pgd_t *pgd, unsigned long addr,
243 struct host_vm_change *hvc)
249 pud = pud_offset(pgd, addr);
251 next = pud_addr_end(addr, end);
252 if (!pud_present(*pud)) {
253 if (hvc->force || pud_newpage(*pud)) {
254 ret = add_munmap(addr, next - addr, hvc);
255 pud_mkuptodate(*pud);
258 else ret = update_pmd_range(pud, addr, next, hvc);
259 } while (pud++, addr = next, ((addr < end) && !ret));
263 void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
264 unsigned long end_addr, int force)
267 struct host_vm_change hvc;
268 unsigned long addr = start_addr, next;
271 hvc = INIT_HVC(mm, force);
272 pgd = pgd_offset(mm, addr);
274 next = pgd_addr_end(addr, end_addr);
275 if (!pgd_present(*pgd)) {
276 if (force || pgd_newpage(*pgd)) {
277 ret = add_munmap(addr, next - addr, &hvc);
278 pgd_mkuptodate(*pgd);
281 else ret = update_pud_range(pgd, addr, next, &hvc);
282 } while (pgd++, addr = next, ((addr < end_addr) && !ret));
285 ret = do_ops(&hvc, hvc.index, 1);
287 /* This is not an else because ret is modified above */
289 printk(KERN_ERR "fix_range_common: failed, killing current "
290 "process: %d\n", task_tgid_vnr(current));
291 /* We are under mmap_sem, release it such that current can terminate */
292 up_write(¤t->mm->mmap_sem);
293 force_sig(SIGKILL, current);
294 do_signal(¤t->thread.regs);
298 static int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
300 struct mm_struct *mm;
305 unsigned long addr, last;
306 int updated = 0, err;
309 for (addr = start; addr < end;) {
310 pgd = pgd_offset(mm, addr);
311 if (!pgd_present(*pgd)) {
312 last = ADD_ROUND(addr, PGDIR_SIZE);
315 if (pgd_newpage(*pgd)) {
317 err = os_unmap_memory((void *) addr,
320 panic("munmap failed, errno = %d\n",
327 pud = pud_offset(pgd, addr);
328 if (!pud_present(*pud)) {
329 last = ADD_ROUND(addr, PUD_SIZE);
332 if (pud_newpage(*pud)) {
334 err = os_unmap_memory((void *) addr,
337 panic("munmap failed, errno = %d\n",
344 pmd = pmd_offset(pud, addr);
345 if (!pmd_present(*pmd)) {
346 last = ADD_ROUND(addr, PMD_SIZE);
349 if (pmd_newpage(*pmd)) {
351 err = os_unmap_memory((void *) addr,
354 panic("munmap failed, errno = %d\n",
361 pte = pte_offset_kernel(pmd, addr);
362 if (!pte_present(*pte) || pte_newpage(*pte)) {
364 err = os_unmap_memory((void *) addr,
367 panic("munmap failed, errno = %d\n",
369 if (pte_present(*pte))
371 pte_val(*pte) & PAGE_MASK,
374 else if (pte_newprot(*pte)) {
376 os_protect_memory((void *) addr, PAGE_SIZE, 1, 1, 1);
383 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
389 struct mm_struct *mm = vma->vm_mm;
391 int r, w, x, prot, err = 0;
394 address &= PAGE_MASK;
395 pgd = pgd_offset(mm, address);
396 if (!pgd_present(*pgd))
399 pud = pud_offset(pgd, address);
400 if (!pud_present(*pud))
403 pmd = pmd_offset(pud, address);
404 if (!pmd_present(*pmd))
407 pte = pte_offset_kernel(pmd, address);
412 if (!pte_young(*pte)) {
415 } else if (!pte_dirty(*pte)) {
419 mm_id = &mm->context.id;
420 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
421 (x ? UM_PROT_EXEC : 0));
422 if (pte_newpage(*pte)) {
423 if (pte_present(*pte)) {
424 unsigned long long offset;
427 fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
428 err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
431 else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
433 else if (pte_newprot(*pte))
434 err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
439 *pte = pte_mkuptodate(*pte);
444 printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
445 force_sig(SIGKILL, current);
448 pgd_t *pgd_offset_proc(struct mm_struct *mm, unsigned long address)
450 return pgd_offset(mm, address);
453 pud_t *pud_offset_proc(pgd_t *pgd, unsigned long address)
455 return pud_offset(pgd, address);
458 pmd_t *pmd_offset_proc(pud_t *pud, unsigned long address)
460 return pmd_offset(pud, address);
463 pte_t *pte_offset_proc(pmd_t *pmd, unsigned long address)
465 return pte_offset_kernel(pmd, address);
468 pte_t *addr_pte(struct task_struct *task, unsigned long addr)
470 pgd_t *pgd = pgd_offset(task->mm, addr);
471 pud_t *pud = pud_offset(pgd, addr);
472 pmd_t *pmd = pmd_offset(pud, addr);
474 return pte_offset_map(pmd, addr);
477 void flush_tlb_all(void)
479 flush_tlb_mm(current->mm);
482 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
484 flush_tlb_kernel_range_common(start, end);
487 void flush_tlb_kernel_vm(void)
489 flush_tlb_kernel_range_common(start_vm, end_vm);
492 void __flush_tlb_one(unsigned long addr)
494 flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
497 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
498 unsigned long end_addr, int force)
500 fix_range_common(mm, start_addr, end_addr, force);
503 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
506 if (vma->vm_mm == NULL)
507 flush_tlb_kernel_range_common(start, end);
508 else fix_range(vma->vm_mm, start, end, 0);
510 EXPORT_SYMBOL(flush_tlb_range);
512 void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
516 * Don't bother flushing if this address space is about to be
519 if (atomic_read(&mm->mm_users) == 0)
522 fix_range(mm, start, end, 0);
525 void flush_tlb_mm(struct mm_struct *mm)
527 struct vm_area_struct *vma = mm->mmap;
529 while (vma != NULL) {
530 fix_range(mm, vma->vm_start, vma->vm_end, 0);
535 void force_flush_all(void)
537 struct mm_struct *mm = current->mm;
538 struct vm_area_struct *vma = mm->mmap;
540 while (vma != NULL) {
541 fix_range(mm, vma->vm_start, vma->vm_end, 1);