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;
46 #define INIT_HVC(mm, force) \
47 ((struct host_vm_change) \
48 { .ops = { { .type = NONE } }, \
49 .id = &mm->context.id, \
54 static void report_enomem(void)
56 printk(KERN_ERR "UML ran out of memory on the host side! "
57 "This can happen due to a memory limitation or "
58 "vm.max_map_count has been reached.\n");
61 static int do_ops(struct host_vm_change *hvc, int end,
64 struct host_vm_op *op;
67 for (i = 0; i < end && !ret; i++) {
71 ret = map(hvc->id, op->u.mmap.addr, op->u.mmap.len,
72 op->u.mmap.prot, op->u.mmap.fd,
73 op->u.mmap.offset, finished, &hvc->data);
76 ret = unmap(hvc->id, op->u.munmap.addr,
77 op->u.munmap.len, finished, &hvc->data);
80 ret = protect(hvc->id, op->u.mprotect.addr,
81 op->u.mprotect.len, op->u.mprotect.prot,
82 finished, &hvc->data);
85 printk(KERN_ERR "Unknown op type %d in do_ops\n",
98 static int add_mmap(unsigned long virt, unsigned long phys, unsigned long len,
99 unsigned int prot, struct host_vm_change *hvc)
102 struct host_vm_op *last;
105 fd = phys_mapping(phys, &offset);
106 if (hvc->index != 0) {
107 last = &hvc->ops[hvc->index - 1];
108 if ((last->type == MMAP) &&
109 (last->u.mmap.addr + last->u.mmap.len == virt) &&
110 (last->u.mmap.prot == prot) && (last->u.mmap.fd == fd) &&
111 (last->u.mmap.offset + last->u.mmap.len == offset)) {
112 last->u.mmap.len += len;
117 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
118 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
122 hvc->ops[hvc->index++] = ((struct host_vm_op)
124 .u = { .mmap = { .addr = virt,
133 static int add_munmap(unsigned long addr, unsigned long len,
134 struct host_vm_change *hvc)
136 struct host_vm_op *last;
139 if ((addr >= STUB_START) && (addr < STUB_END))
142 if (hvc->index != 0) {
143 last = &hvc->ops[hvc->index - 1];
144 if ((last->type == MUNMAP) &&
145 (last->u.munmap.addr + last->u.mmap.len == addr)) {
146 last->u.munmap.len += len;
151 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
152 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
156 hvc->ops[hvc->index++] = ((struct host_vm_op)
158 .u = { .munmap = { .addr = addr,
163 static int add_mprotect(unsigned long addr, unsigned long len,
164 unsigned int prot, struct host_vm_change *hvc)
166 struct host_vm_op *last;
169 if (hvc->index != 0) {
170 last = &hvc->ops[hvc->index - 1];
171 if ((last->type == MPROTECT) &&
172 (last->u.mprotect.addr + last->u.mprotect.len == addr) &&
173 (last->u.mprotect.prot == prot)) {
174 last->u.mprotect.len += len;
179 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
180 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
184 hvc->ops[hvc->index++] = ((struct host_vm_op)
186 .u = { .mprotect = { .addr = addr,
192 #define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
194 static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
196 struct host_vm_change *hvc)
199 int r, w, x, prot, ret = 0;
201 pte = pte_offset_kernel(pmd, addr);
203 if ((addr >= STUB_START) && (addr < STUB_END))
209 if (!pte_young(*pte)) {
212 } else if (!pte_dirty(*pte))
215 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
216 (x ? UM_PROT_EXEC : 0));
217 if (hvc->force || pte_newpage(*pte)) {
218 if (pte_present(*pte))
219 ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
220 PAGE_SIZE, prot, hvc);
222 ret = add_munmap(addr, PAGE_SIZE, hvc);
223 } else if (pte_newprot(*pte))
224 ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
225 *pte = pte_mkuptodate(*pte);
226 } while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
230 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
232 struct host_vm_change *hvc)
238 pmd = pmd_offset(pud, addr);
240 next = pmd_addr_end(addr, end);
241 if (!pmd_present(*pmd)) {
242 if (hvc->force || pmd_newpage(*pmd)) {
243 ret = add_munmap(addr, next - addr, hvc);
244 pmd_mkuptodate(*pmd);
247 else ret = update_pte_range(pmd, addr, next, hvc);
248 } while (pmd++, addr = next, ((addr < end) && !ret));
252 static inline int update_pud_range(pgd_t *pgd, unsigned long addr,
254 struct host_vm_change *hvc)
260 pud = pud_offset(pgd, addr);
262 next = pud_addr_end(addr, end);
263 if (!pud_present(*pud)) {
264 if (hvc->force || pud_newpage(*pud)) {
265 ret = add_munmap(addr, next - addr, hvc);
266 pud_mkuptodate(*pud);
269 else ret = update_pmd_range(pud, addr, next, hvc);
270 } while (pud++, addr = next, ((addr < end) && !ret));
274 void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
275 unsigned long end_addr, int force)
278 struct host_vm_change hvc;
279 unsigned long addr = start_addr, next;
282 hvc = INIT_HVC(mm, force);
283 pgd = pgd_offset(mm, addr);
285 next = pgd_addr_end(addr, end_addr);
286 if (!pgd_present(*pgd)) {
287 if (force || pgd_newpage(*pgd)) {
288 ret = add_munmap(addr, next - addr, &hvc);
289 pgd_mkuptodate(*pgd);
292 else ret = update_pud_range(pgd, addr, next, &hvc);
293 } while (pgd++, addr = next, ((addr < end_addr) && !ret));
296 ret = do_ops(&hvc, hvc.index, 1);
298 /* This is not an else because ret is modified above */
300 printk(KERN_ERR "fix_range_common: failed, killing current "
301 "process: %d\n", task_tgid_vnr(current));
302 /* We are under mmap_sem, release it such that current can terminate */
303 up_write(¤t->mm->mmap_sem);
304 force_sig(SIGKILL, current);
305 do_signal(¤t->thread.regs);
309 static int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
311 struct mm_struct *mm;
316 unsigned long addr, last;
317 int updated = 0, err;
320 for (addr = start; addr < end;) {
321 pgd = pgd_offset(mm, addr);
322 if (!pgd_present(*pgd)) {
323 last = ADD_ROUND(addr, PGDIR_SIZE);
326 if (pgd_newpage(*pgd)) {
328 err = os_unmap_memory((void *) addr,
331 panic("munmap failed, errno = %d\n",
338 pud = pud_offset(pgd, addr);
339 if (!pud_present(*pud)) {
340 last = ADD_ROUND(addr, PUD_SIZE);
343 if (pud_newpage(*pud)) {
345 err = os_unmap_memory((void *) addr,
348 panic("munmap failed, errno = %d\n",
355 pmd = pmd_offset(pud, addr);
356 if (!pmd_present(*pmd)) {
357 last = ADD_ROUND(addr, PMD_SIZE);
360 if (pmd_newpage(*pmd)) {
362 err = os_unmap_memory((void *) addr,
365 panic("munmap failed, errno = %d\n",
372 pte = pte_offset_kernel(pmd, addr);
373 if (!pte_present(*pte) || pte_newpage(*pte)) {
375 err = os_unmap_memory((void *) addr,
378 panic("munmap failed, errno = %d\n",
380 if (pte_present(*pte))
382 pte_val(*pte) & PAGE_MASK,
385 else if (pte_newprot(*pte)) {
387 os_protect_memory((void *) addr, PAGE_SIZE, 1, 1, 1);
394 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
400 struct mm_struct *mm = vma->vm_mm;
402 int r, w, x, prot, err = 0;
405 address &= PAGE_MASK;
406 pgd = pgd_offset(mm, address);
407 if (!pgd_present(*pgd))
410 pud = pud_offset(pgd, address);
411 if (!pud_present(*pud))
414 pmd = pmd_offset(pud, address);
415 if (!pmd_present(*pmd))
418 pte = pte_offset_kernel(pmd, address);
423 if (!pte_young(*pte)) {
426 } else if (!pte_dirty(*pte)) {
430 mm_id = &mm->context.id;
431 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
432 (x ? UM_PROT_EXEC : 0));
433 if (pte_newpage(*pte)) {
434 if (pte_present(*pte)) {
435 unsigned long long offset;
438 fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
439 err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
442 else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
444 else if (pte_newprot(*pte))
445 err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
454 *pte = pte_mkuptodate(*pte);
459 printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
460 force_sig(SIGKILL, current);
463 pgd_t *pgd_offset_proc(struct mm_struct *mm, unsigned long address)
465 return pgd_offset(mm, address);
468 pud_t *pud_offset_proc(pgd_t *pgd, unsigned long address)
470 return pud_offset(pgd, address);
473 pmd_t *pmd_offset_proc(pud_t *pud, unsigned long address)
475 return pmd_offset(pud, address);
478 pte_t *pte_offset_proc(pmd_t *pmd, unsigned long address)
480 return pte_offset_kernel(pmd, address);
483 pte_t *addr_pte(struct task_struct *task, unsigned long addr)
485 pgd_t *pgd = pgd_offset(task->mm, addr);
486 pud_t *pud = pud_offset(pgd, addr);
487 pmd_t *pmd = pmd_offset(pud, addr);
489 return pte_offset_map(pmd, addr);
492 void flush_tlb_all(void)
494 flush_tlb_mm(current->mm);
497 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
499 flush_tlb_kernel_range_common(start, end);
502 void flush_tlb_kernel_vm(void)
504 flush_tlb_kernel_range_common(start_vm, end_vm);
507 void __flush_tlb_one(unsigned long addr)
509 flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
512 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
513 unsigned long end_addr, int force)
515 fix_range_common(mm, start_addr, end_addr, force);
518 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
521 if (vma->vm_mm == NULL)
522 flush_tlb_kernel_range_common(start, end);
523 else fix_range(vma->vm_mm, start, end, 0);
525 EXPORT_SYMBOL(flush_tlb_range);
527 void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
531 * Don't bother flushing if this address space is about to be
534 if (atomic_read(&mm->mm_users) == 0)
537 fix_range(mm, start, end, 0);
540 void flush_tlb_mm(struct mm_struct *mm)
542 struct vm_area_struct *vma = mm->mmap;
544 while (vma != NULL) {
545 fix_range(mm, vma->vm_start, vma->vm_end, 0);
550 void force_flush_all(void)
552 struct mm_struct *mm = current->mm;
553 struct vm_area_struct *vma = mm->mmap;
555 while (vma != NULL) {
556 fix_range(mm, vma->vm_start, vma->vm_end, 1);