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/signal.h>
10 #include <asm/pgtable.h>
11 #include <asm/tlbflush.h>
12 #include <as-layout.h>
16 #include <kern_util.h>
18 struct host_vm_change {
20 enum { NONE, MMAP, MUNMAP, MPROTECT } type;
47 #define INIT_HVC(mm, force, userspace) \
48 ((struct host_vm_change) \
49 { .ops = { { .type = NONE } }, \
52 .userspace = userspace, \
56 static void report_enomem(void)
58 printk(KERN_ERR "UML ran out of memory on the host side! "
59 "This can happen due to a memory limitation or "
60 "vm.max_map_count has been reached.\n");
63 static int do_ops(struct host_vm_change *hvc, int end,
66 struct host_vm_op *op;
69 for (i = 0; i < end && !ret; i++) {
74 ret = map(&hvc->mm->context.id, op->u.mmap.addr,
75 op->u.mmap.len, op->u.mmap.prot,
77 op->u.mmap.offset, finished,
80 map_memory(op->u.mmap.addr, op->u.mmap.offset,
81 op->u.mmap.len, 1, 1, 1);
85 ret = unmap(&hvc->mm->context.id,
87 op->u.munmap.len, finished,
90 ret = os_unmap_memory(
91 (void *) op->u.munmap.addr,
97 ret = protect(&hvc->mm->context.id,
101 finished, &hvc->data);
103 ret = os_protect_memory(
104 (void *) op->u.mprotect.addr,
109 printk(KERN_ERR "Unknown op type %d in do_ops\n",
122 static int add_mmap(unsigned long virt, unsigned long phys, unsigned long len,
123 unsigned int prot, struct host_vm_change *hvc)
126 struct host_vm_op *last;
127 int fd = -1, ret = 0;
130 fd = phys_mapping(phys, &offset);
133 if (hvc->index != 0) {
134 last = &hvc->ops[hvc->index - 1];
135 if ((last->type == MMAP) &&
136 (last->u.mmap.addr + last->u.mmap.len == virt) &&
137 (last->u.mmap.prot == prot) && (last->u.mmap.fd == fd) &&
138 (last->u.mmap.offset + last->u.mmap.len == offset)) {
139 last->u.mmap.len += len;
144 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
145 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
149 hvc->ops[hvc->index++] = ((struct host_vm_op)
151 .u = { .mmap = { .addr = virt,
160 static int add_munmap(unsigned long addr, unsigned long len,
161 struct host_vm_change *hvc)
163 struct host_vm_op *last;
166 if ((addr >= STUB_START) && (addr < STUB_END))
169 if (hvc->index != 0) {
170 last = &hvc->ops[hvc->index - 1];
171 if ((last->type == MUNMAP) &&
172 (last->u.munmap.addr + last->u.mmap.len == addr)) {
173 last->u.munmap.len += len;
178 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
179 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
183 hvc->ops[hvc->index++] = ((struct host_vm_op)
185 .u = { .munmap = { .addr = addr,
190 static int add_mprotect(unsigned long addr, unsigned long len,
191 unsigned int prot, struct host_vm_change *hvc)
193 struct host_vm_op *last;
196 if (hvc->index != 0) {
197 last = &hvc->ops[hvc->index - 1];
198 if ((last->type == MPROTECT) &&
199 (last->u.mprotect.addr + last->u.mprotect.len == addr) &&
200 (last->u.mprotect.prot == prot)) {
201 last->u.mprotect.len += len;
206 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
207 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
211 hvc->ops[hvc->index++] = ((struct host_vm_op)
213 .u = { .mprotect = { .addr = addr,
219 #define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
221 static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
223 struct host_vm_change *hvc)
226 int r, w, x, prot, ret = 0;
228 pte = pte_offset_kernel(pmd, addr);
230 if ((addr >= STUB_START) && (addr < STUB_END))
236 if (!pte_young(*pte)) {
239 } else if (!pte_dirty(*pte))
242 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
243 (x ? UM_PROT_EXEC : 0));
244 if (hvc->force || pte_newpage(*pte)) {
245 if (pte_present(*pte)) {
246 if (pte_newpage(*pte))
247 ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
248 PAGE_SIZE, prot, hvc);
250 ret = add_munmap(addr, PAGE_SIZE, hvc);
251 } else if (pte_newprot(*pte))
252 ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
253 *pte = pte_mkuptodate(*pte);
254 } while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
258 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
260 struct host_vm_change *hvc)
266 pmd = pmd_offset(pud, addr);
268 next = pmd_addr_end(addr, end);
269 if (!pmd_present(*pmd)) {
270 if (hvc->force || pmd_newpage(*pmd)) {
271 ret = add_munmap(addr, next - addr, hvc);
272 pmd_mkuptodate(*pmd);
275 else ret = update_pte_range(pmd, addr, next, hvc);
276 } while (pmd++, addr = next, ((addr < end) && !ret));
280 static inline int update_pud_range(pgd_t *pgd, unsigned long addr,
282 struct host_vm_change *hvc)
288 pud = pud_offset(pgd, addr);
290 next = pud_addr_end(addr, end);
291 if (!pud_present(*pud)) {
292 if (hvc->force || pud_newpage(*pud)) {
293 ret = add_munmap(addr, next - addr, hvc);
294 pud_mkuptodate(*pud);
297 else ret = update_pmd_range(pud, addr, next, hvc);
298 } while (pud++, addr = next, ((addr < end) && !ret));
302 void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
303 unsigned long end_addr, int force)
306 struct host_vm_change hvc;
307 unsigned long addr = start_addr, next;
308 int ret = 0, userspace = 1;
310 hvc = INIT_HVC(mm, force, userspace);
311 pgd = pgd_offset(mm, addr);
313 next = pgd_addr_end(addr, end_addr);
314 if (!pgd_present(*pgd)) {
315 if (force || pgd_newpage(*pgd)) {
316 ret = add_munmap(addr, next - addr, &hvc);
317 pgd_mkuptodate(*pgd);
320 else ret = update_pud_range(pgd, addr, next, &hvc);
321 } while (pgd++, addr = next, ((addr < end_addr) && !ret));
324 ret = do_ops(&hvc, hvc.index, 1);
326 /* This is not an else because ret is modified above */
328 printk(KERN_ERR "fix_range_common: failed, killing current "
329 "process: %d\n", task_tgid_vnr(current));
330 /* We are under mmap_sem, release it such that current can terminate */
331 up_write(¤t->mm->mmap_sem);
333 do_signal(¤t->thread.regs);
337 static int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
339 struct mm_struct *mm;
344 unsigned long addr, last;
345 int updated = 0, err = 0, force = 0, userspace = 0;
346 struct host_vm_change hvc;
349 hvc = INIT_HVC(mm, force, userspace);
350 for (addr = start; addr < end;) {
351 pgd = pgd_offset(mm, addr);
352 if (!pgd_present(*pgd)) {
353 last = ADD_ROUND(addr, PGDIR_SIZE);
356 if (pgd_newpage(*pgd)) {
358 err = add_munmap(addr, last - addr, &hvc);
360 panic("munmap failed, errno = %d\n",
367 pud = pud_offset(pgd, addr);
368 if (!pud_present(*pud)) {
369 last = ADD_ROUND(addr, PUD_SIZE);
372 if (pud_newpage(*pud)) {
374 err = add_munmap(addr, last - addr, &hvc);
376 panic("munmap failed, errno = %d\n",
383 pmd = pmd_offset(pud, addr);
384 if (!pmd_present(*pmd)) {
385 last = ADD_ROUND(addr, PMD_SIZE);
388 if (pmd_newpage(*pmd)) {
390 err = add_munmap(addr, last - addr, &hvc);
392 panic("munmap failed, errno = %d\n",
399 pte = pte_offset_kernel(pmd, addr);
400 if (!pte_present(*pte) || pte_newpage(*pte)) {
402 err = add_munmap(addr, PAGE_SIZE, &hvc);
404 panic("munmap failed, errno = %d\n",
406 if (pte_present(*pte))
407 err = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
410 else if (pte_newprot(*pte)) {
412 err = add_mprotect(addr, PAGE_SIZE, 0, &hvc);
417 err = do_ops(&hvc, hvc.index, 1);
420 panic("flush_tlb_kernel failed, errno = %d\n", err);
424 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
430 struct mm_struct *mm = vma->vm_mm;
432 int r, w, x, prot, err = 0;
435 address &= PAGE_MASK;
436 pgd = pgd_offset(mm, address);
437 if (!pgd_present(*pgd))
440 pud = pud_offset(pgd, address);
441 if (!pud_present(*pud))
444 pmd = pmd_offset(pud, address);
445 if (!pmd_present(*pmd))
448 pte = pte_offset_kernel(pmd, address);
453 if (!pte_young(*pte)) {
456 } else if (!pte_dirty(*pte)) {
460 mm_id = &mm->context.id;
461 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
462 (x ? UM_PROT_EXEC : 0));
463 if (pte_newpage(*pte)) {
464 if (pte_present(*pte)) {
465 unsigned long long offset;
468 fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
469 err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
472 else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
474 else if (pte_newprot(*pte))
475 err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
484 *pte = pte_mkuptodate(*pte);
489 printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
493 pgd_t *pgd_offset_proc(struct mm_struct *mm, unsigned long address)
495 return pgd_offset(mm, address);
498 pud_t *pud_offset_proc(pgd_t *pgd, unsigned long address)
500 return pud_offset(pgd, address);
503 pmd_t *pmd_offset_proc(pud_t *pud, unsigned long address)
505 return pmd_offset(pud, address);
508 pte_t *pte_offset_proc(pmd_t *pmd, unsigned long address)
510 return pte_offset_kernel(pmd, address);
513 pte_t *addr_pte(struct task_struct *task, unsigned long addr)
515 pgd_t *pgd = pgd_offset(task->mm, addr);
516 pud_t *pud = pud_offset(pgd, addr);
517 pmd_t *pmd = pmd_offset(pud, addr);
519 return pte_offset_map(pmd, addr);
522 void flush_tlb_all(void)
525 * Don't bother flushing if this address space is about to be
528 if (atomic_read(¤t->mm->mm_users) == 0)
531 flush_tlb_mm(current->mm);
534 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
536 flush_tlb_kernel_range_common(start, end);
539 void flush_tlb_kernel_vm(void)
541 flush_tlb_kernel_range_common(start_vm, end_vm);
544 void __flush_tlb_one(unsigned long addr)
546 flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
549 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
550 unsigned long end_addr, int force)
553 * Don't bother flushing if this address space is about to be
556 if (atomic_read(&mm->mm_users) == 0)
559 fix_range_common(mm, start_addr, end_addr, force);
562 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
565 if (vma->vm_mm == NULL)
566 flush_tlb_kernel_range_common(start, end);
567 else fix_range(vma->vm_mm, start, end, 0);
569 EXPORT_SYMBOL(flush_tlb_range);
571 void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
574 fix_range(mm, start, end, 0);
577 void flush_tlb_mm(struct mm_struct *mm)
579 struct vm_area_struct *vma = mm->mmap;
581 while (vma != NULL) {
582 fix_range(mm, vma->vm_start, vma->vm_end, 0);
587 void force_flush_all(void)
589 struct mm_struct *mm = current->mm;
590 struct vm_area_struct *vma = mm->mmap;
592 while (vma != NULL) {
593 fix_range(mm, vma->vm_start, vma->vm_end, 1);