1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
7 #include <linux/module.h>
8 #include <linux/sched/signal.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;
46 #define INIT_HVC(mm, force, userspace) \
47 ((struct host_vm_change) \
48 { .ops = { { .type = NONE } }, \
51 .userspace = userspace, \
55 static void report_enomem(void)
57 printk(KERN_ERR "UML ran out of memory on the host side! "
58 "This can happen due to a memory limitation or "
59 "vm.max_map_count has been reached.\n");
62 static int do_ops(struct host_vm_change *hvc, int end,
65 struct host_vm_op *op;
68 for (i = 0; i < end && !ret; i++) {
73 ret = map(&hvc->mm->context.id, op->u.mmap.addr,
74 op->u.mmap.len, op->u.mmap.prot,
76 op->u.mmap.offset, finished,
79 map_memory(op->u.mmap.addr, op->u.mmap.offset,
80 op->u.mmap.len, 1, 1, 1);
84 ret = unmap(&hvc->mm->context.id,
86 op->u.munmap.len, finished,
89 ret = os_unmap_memory(
90 (void *) op->u.munmap.addr,
96 ret = protect(&hvc->mm->context.id,
100 finished, &hvc->data);
102 ret = os_protect_memory(
103 (void *) op->u.mprotect.addr,
108 printk(KERN_ERR "Unknown op type %d in do_ops\n",
121 static int add_mmap(unsigned long virt, unsigned long phys, unsigned long len,
122 unsigned int prot, struct host_vm_change *hvc)
125 struct host_vm_op *last;
126 int fd = -1, ret = 0;
129 fd = phys_mapping(phys, &offset);
132 if (hvc->index != 0) {
133 last = &hvc->ops[hvc->index - 1];
134 if ((last->type == MMAP) &&
135 (last->u.mmap.addr + last->u.mmap.len == virt) &&
136 (last->u.mmap.prot == prot) && (last->u.mmap.fd == fd) &&
137 (last->u.mmap.offset + last->u.mmap.len == offset)) {
138 last->u.mmap.len += len;
143 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
144 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
148 hvc->ops[hvc->index++] = ((struct host_vm_op)
150 .u = { .mmap = { .addr = virt,
159 static int add_munmap(unsigned long addr, unsigned long len,
160 struct host_vm_change *hvc)
162 struct host_vm_op *last;
165 if ((addr >= STUB_START) && (addr < STUB_END))
168 if (hvc->index != 0) {
169 last = &hvc->ops[hvc->index - 1];
170 if ((last->type == MUNMAP) &&
171 (last->u.munmap.addr + last->u.mmap.len == addr)) {
172 last->u.munmap.len += len;
177 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
178 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
182 hvc->ops[hvc->index++] = ((struct host_vm_op)
184 .u = { .munmap = { .addr = addr,
189 static int add_mprotect(unsigned long addr, unsigned long len,
190 unsigned int prot, struct host_vm_change *hvc)
192 struct host_vm_op *last;
195 if (hvc->index != 0) {
196 last = &hvc->ops[hvc->index - 1];
197 if ((last->type == MPROTECT) &&
198 (last->u.mprotect.addr + last->u.mprotect.len == addr) &&
199 (last->u.mprotect.prot == prot)) {
200 last->u.mprotect.len += len;
205 if (hvc->index == ARRAY_SIZE(hvc->ops)) {
206 ret = do_ops(hvc, ARRAY_SIZE(hvc->ops), 0);
210 hvc->ops[hvc->index++] = ((struct host_vm_op)
212 .u = { .mprotect = { .addr = addr,
218 #define ADD_ROUND(n, inc) (((n) + (inc)) & ~((inc) - 1))
220 static inline int update_pte_range(pmd_t *pmd, unsigned long addr,
222 struct host_vm_change *hvc)
225 int r, w, x, prot, ret = 0;
227 pte = pte_offset_kernel(pmd, addr);
229 if ((addr >= STUB_START) && (addr < STUB_END))
235 if (!pte_young(*pte)) {
238 } else if (!pte_dirty(*pte))
241 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
242 (x ? UM_PROT_EXEC : 0));
243 if (hvc->force || pte_newpage(*pte)) {
244 if (pte_present(*pte)) {
245 if (pte_newpage(*pte))
246 ret = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
247 PAGE_SIZE, prot, hvc);
249 ret = add_munmap(addr, PAGE_SIZE, hvc);
250 } else if (pte_newprot(*pte))
251 ret = add_mprotect(addr, PAGE_SIZE, prot, hvc);
252 *pte = pte_mkuptodate(*pte);
253 } while (pte++, addr += PAGE_SIZE, ((addr < end) && !ret));
257 static inline int update_pmd_range(pud_t *pud, unsigned long addr,
259 struct host_vm_change *hvc)
265 pmd = pmd_offset(pud, addr);
267 next = pmd_addr_end(addr, end);
268 if (!pmd_present(*pmd)) {
269 if (hvc->force || pmd_newpage(*pmd)) {
270 ret = add_munmap(addr, next - addr, hvc);
271 pmd_mkuptodate(*pmd);
274 else ret = update_pte_range(pmd, addr, next, hvc);
275 } while (pmd++, addr = next, ((addr < end) && !ret));
279 static inline int update_pud_range(p4d_t *p4d, unsigned long addr,
281 struct host_vm_change *hvc)
287 pud = pud_offset(p4d, addr);
289 next = pud_addr_end(addr, end);
290 if (!pud_present(*pud)) {
291 if (hvc->force || pud_newpage(*pud)) {
292 ret = add_munmap(addr, next - addr, hvc);
293 pud_mkuptodate(*pud);
296 else ret = update_pmd_range(pud, addr, next, hvc);
297 } while (pud++, addr = next, ((addr < end) && !ret));
301 static inline int update_p4d_range(pgd_t *pgd, unsigned long addr,
303 struct host_vm_change *hvc)
309 p4d = p4d_offset(pgd, addr);
311 next = p4d_addr_end(addr, end);
312 if (!p4d_present(*p4d)) {
313 if (hvc->force || p4d_newpage(*p4d)) {
314 ret = add_munmap(addr, next - addr, hvc);
315 p4d_mkuptodate(*p4d);
318 ret = update_pud_range(p4d, addr, next, hvc);
319 } while (p4d++, addr = next, ((addr < end) && !ret));
323 void fix_range_common(struct mm_struct *mm, unsigned long start_addr,
324 unsigned long end_addr, int force)
327 struct host_vm_change hvc;
328 unsigned long addr = start_addr, next;
329 int ret = 0, userspace = 1;
331 hvc = INIT_HVC(mm, force, userspace);
332 pgd = pgd_offset(mm, addr);
334 next = pgd_addr_end(addr, end_addr);
335 if (!pgd_present(*pgd)) {
336 if (force || pgd_newpage(*pgd)) {
337 ret = add_munmap(addr, next - addr, &hvc);
338 pgd_mkuptodate(*pgd);
341 ret = update_p4d_range(pgd, addr, next, &hvc);
342 } while (pgd++, addr = next, ((addr < end_addr) && !ret));
345 ret = do_ops(&hvc, hvc.index, 1);
347 /* This is not an else because ret is modified above */
349 printk(KERN_ERR "fix_range_common: failed, killing current "
350 "process: %d\n", task_tgid_vnr(current));
351 /* We are under mmap_sem, release it such that current can terminate */
352 mmap_write_unlock(current->mm);
354 do_signal(¤t->thread.regs);
358 static int flush_tlb_kernel_range_common(unsigned long start, unsigned long end)
360 struct mm_struct *mm;
366 unsigned long addr, last;
367 int updated = 0, err = 0, force = 0, userspace = 0;
368 struct host_vm_change hvc;
371 hvc = INIT_HVC(mm, force, userspace);
372 for (addr = start; addr < end;) {
373 pgd = pgd_offset(mm, addr);
374 if (!pgd_present(*pgd)) {
375 last = ADD_ROUND(addr, PGDIR_SIZE);
378 if (pgd_newpage(*pgd)) {
380 err = add_munmap(addr, last - addr, &hvc);
382 panic("munmap failed, errno = %d\n",
389 p4d = p4d_offset(pgd, addr);
390 if (!p4d_present(*p4d)) {
391 last = ADD_ROUND(addr, P4D_SIZE);
394 if (p4d_newpage(*p4d)) {
396 err = add_munmap(addr, last - addr, &hvc);
398 panic("munmap failed, errno = %d\n",
405 pud = pud_offset(p4d, addr);
406 if (!pud_present(*pud)) {
407 last = ADD_ROUND(addr, PUD_SIZE);
410 if (pud_newpage(*pud)) {
412 err = add_munmap(addr, last - addr, &hvc);
414 panic("munmap failed, errno = %d\n",
421 pmd = pmd_offset(pud, addr);
422 if (!pmd_present(*pmd)) {
423 last = ADD_ROUND(addr, PMD_SIZE);
426 if (pmd_newpage(*pmd)) {
428 err = add_munmap(addr, last - addr, &hvc);
430 panic("munmap failed, errno = %d\n",
437 pte = pte_offset_kernel(pmd, addr);
438 if (!pte_present(*pte) || pte_newpage(*pte)) {
440 err = add_munmap(addr, PAGE_SIZE, &hvc);
442 panic("munmap failed, errno = %d\n",
444 if (pte_present(*pte))
445 err = add_mmap(addr, pte_val(*pte) & PAGE_MASK,
448 else if (pte_newprot(*pte)) {
450 err = add_mprotect(addr, PAGE_SIZE, 0, &hvc);
455 err = do_ops(&hvc, hvc.index, 1);
458 panic("flush_tlb_kernel failed, errno = %d\n", err);
462 void flush_tlb_page(struct vm_area_struct *vma, unsigned long address)
469 struct mm_struct *mm = vma->vm_mm;
471 int r, w, x, prot, err = 0;
474 address &= PAGE_MASK;
475 pgd = pgd_offset(mm, address);
476 if (!pgd_present(*pgd))
479 p4d = p4d_offset(pgd, address);
480 if (!p4d_present(*p4d))
483 pud = pud_offset(p4d, address);
484 if (!pud_present(*pud))
487 pmd = pmd_offset(pud, address);
488 if (!pmd_present(*pmd))
491 pte = pte_offset_kernel(pmd, address);
496 if (!pte_young(*pte)) {
499 } else if (!pte_dirty(*pte)) {
503 mm_id = &mm->context.id;
504 prot = ((r ? UM_PROT_READ : 0) | (w ? UM_PROT_WRITE : 0) |
505 (x ? UM_PROT_EXEC : 0));
506 if (pte_newpage(*pte)) {
507 if (pte_present(*pte)) {
508 unsigned long long offset;
511 fd = phys_mapping(pte_val(*pte) & PAGE_MASK, &offset);
512 err = map(mm_id, address, PAGE_SIZE, prot, fd, offset,
515 else err = unmap(mm_id, address, PAGE_SIZE, 1, &flush);
517 else if (pte_newprot(*pte))
518 err = protect(mm_id, address, PAGE_SIZE, prot, 1, &flush);
527 *pte = pte_mkuptodate(*pte);
532 printk(KERN_ERR "Failed to flush page for address 0x%lx\n", address);
536 void flush_tlb_all(void)
539 * Don't bother flushing if this address space is about to be
542 if (atomic_read(¤t->mm->mm_users) == 0)
545 flush_tlb_mm(current->mm);
548 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
550 flush_tlb_kernel_range_common(start, end);
553 void flush_tlb_kernel_vm(void)
555 flush_tlb_kernel_range_common(start_vm, end_vm);
558 void __flush_tlb_one(unsigned long addr)
560 flush_tlb_kernel_range_common(addr, addr + PAGE_SIZE);
563 static void fix_range(struct mm_struct *mm, unsigned long start_addr,
564 unsigned long end_addr, int force)
567 * Don't bother flushing if this address space is about to be
570 if (atomic_read(&mm->mm_users) == 0)
573 fix_range_common(mm, start_addr, end_addr, force);
576 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
579 if (vma->vm_mm == NULL)
580 flush_tlb_kernel_range_common(start, end);
581 else fix_range(vma->vm_mm, start, end, 0);
583 EXPORT_SYMBOL(flush_tlb_range);
585 void flush_tlb_mm_range(struct mm_struct *mm, unsigned long start,
588 fix_range(mm, start, end, 0);
591 void flush_tlb_mm(struct mm_struct *mm)
593 struct vm_area_struct *vma = mm->mmap;
595 while (vma != NULL) {
596 fix_range(mm, vma->vm_start, vma->vm_end, 0);
601 void force_flush_all(void)
603 struct mm_struct *mm = current->mm;
604 struct vm_area_struct *vma = mm->mmap;
606 while (vma != NULL) {
607 fix_range(mm, vma->vm_start, vma->vm_end, 1);