2 * mmap support for qemu
4 * Copyright (c) 2003 Fabrice Bellard
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by
8 * the Free Software Foundation; either version 2 of the License, or
9 * (at your option) any later version.
11 * This program is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
14 * GNU General Public License for more details.
16 * You should have received a copy of the GNU General Public License
17 * along with this program; if not, see <http://www.gnu.org/licenses/>.
19 #include "qemu/osdep.h"
22 #include "qemu-common.h"
23 #include "translate-all.h"
27 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
28 static __thread int mmap_lock_count;
32 if (mmap_lock_count++ == 0) {
33 pthread_mutex_lock(&mmap_mutex);
37 void mmap_unlock(void)
39 if (--mmap_lock_count == 0) {
40 pthread_mutex_unlock(&mmap_mutex);
44 bool have_mmap_lock(void)
46 return mmap_lock_count > 0 ? true : false;
49 /* Grab lock to make sure things are in a consistent state after fork(). */
50 void mmap_fork_start(void)
54 pthread_mutex_lock(&mmap_mutex);
57 void mmap_fork_end(int child)
60 pthread_mutex_init(&mmap_mutex, NULL);
62 pthread_mutex_unlock(&mmap_mutex);
65 /* NOTE: all the constants are the HOST ones, but addresses are target. */
66 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
68 abi_ulong end, host_start, host_end, addr;
72 printf("mprotect: start=0x" TARGET_ABI_FMT_lx
73 "len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c\n", start, len,
74 prot & PROT_READ ? 'r' : '-',
75 prot & PROT_WRITE ? 'w' : '-',
76 prot & PROT_EXEC ? 'x' : '-');
79 if ((start & ~TARGET_PAGE_MASK) != 0)
80 return -TARGET_EINVAL;
81 len = TARGET_PAGE_ALIGN(len);
83 if (!guest_range_valid(start, len)) {
84 return -TARGET_ENOMEM;
86 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
91 host_start = start & qemu_host_page_mask;
92 host_end = HOST_PAGE_ALIGN(end);
93 if (start > host_start) {
94 /* handle host page containing start */
96 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
97 prot1 |= page_get_flags(addr);
99 if (host_end == host_start + qemu_host_page_size) {
100 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
101 prot1 |= page_get_flags(addr);
105 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
108 host_start += qemu_host_page_size;
110 if (end < host_end) {
112 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
113 prot1 |= page_get_flags(addr);
115 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
119 host_end -= qemu_host_page_size;
122 /* handle the pages in the middle */
123 if (host_start < host_end) {
124 ret = mprotect(g2h(host_start), host_end - host_start, prot);
128 page_set_flags(start, start + len, prot | PAGE_VALID);
136 /* map an incomplete host page */
137 static int mmap_frag(abi_ulong real_start,
138 abi_ulong start, abi_ulong end,
139 int prot, int flags, int fd, abi_ulong offset)
141 abi_ulong real_end, addr;
145 real_end = real_start + qemu_host_page_size;
146 host_start = g2h(real_start);
148 /* get the protection of the target pages outside the mapping */
150 for(addr = real_start; addr < real_end; addr++) {
151 if (addr < start || addr >= end)
152 prot1 |= page_get_flags(addr);
156 /* no page was there, so we allocate one */
157 void *p = mmap(host_start, qemu_host_page_size, prot,
158 flags | MAP_ANONYMOUS, -1, 0);
165 prot_new = prot | prot1;
166 if (!(flags & MAP_ANONYMOUS)) {
167 /* msync() won't work here, so we return an error if write is
168 possible while it is a shared mapping */
169 if ((flags & MAP_TYPE) == MAP_SHARED &&
173 /* adjust protection to be able to read */
174 if (!(prot1 & PROT_WRITE))
175 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
177 /* read the corresponding file data */
178 if (pread(fd, g2h(start), end - start, offset) == -1)
181 /* put final protection */
182 if (prot_new != (prot1 | PROT_WRITE))
183 mprotect(host_start, qemu_host_page_size, prot_new);
185 if (prot_new != prot1) {
186 mprotect(host_start, qemu_host_page_size, prot_new);
188 if (prot_new & PROT_WRITE) {
189 memset(g2h(start), 0, end - start);
195 #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64
196 # define TASK_UNMAPPED_BASE (1ul << 38)
198 # define TASK_UNMAPPED_BASE 0x40000000
200 abi_ulong mmap_next_start = TASK_UNMAPPED_BASE;
202 unsigned long last_brk;
204 /* Subroutine of mmap_find_vma, used when we have pre-allocated a chunk
205 of guest address space. */
206 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size)
213 if (size > reserved_va) {
214 return (abi_ulong)-1;
217 size = HOST_PAGE_ALIGN(size);
218 end_addr = start + size;
219 if (end_addr > reserved_va) {
220 end_addr = reserved_va;
222 addr = end_addr - qemu_host_page_size;
225 if (addr > end_addr) {
227 return (abi_ulong)-1;
229 end_addr = reserved_va;
230 addr = end_addr - qemu_host_page_size;
234 prot = page_get_flags(addr);
238 if (addr && addr + size == end_addr) {
241 addr -= qemu_host_page_size;
244 if (start == mmap_next_start) {
245 mmap_next_start = addr;
252 * Find and reserve a free memory area of size 'size'. The search
254 * It must be called with mmap_lock() held.
255 * Return -1 if error.
257 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
263 /* If 'start' == 0, then a default start address is used. */
265 start = mmap_next_start;
267 start &= qemu_host_page_mask;
270 size = HOST_PAGE_ALIGN(size);
273 return mmap_find_vma_reserved(start, size);
277 wrapped = repeat = 0;
280 for (;; prev = ptr) {
282 * Reserve needed memory area to avoid a race.
283 * It should be discarded using:
284 * - mmap() with MAP_FIXED flag
285 * - mremap() with MREMAP_FIXED flag
286 * - shmat() with SHM_REMAP flag
288 ptr = mmap(g2h(addr), size, PROT_NONE,
289 MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
291 /* ENOMEM, if host address space has no memory */
292 if (ptr == MAP_FAILED) {
293 return (abi_ulong)-1;
296 /* Count the number of sequential returns of the same address.
297 This is used to modify the search algorithm below. */
298 repeat = (ptr == prev ? repeat + 1 : 0);
300 if (h2g_valid(ptr + size - 1)) {
303 if ((addr & ~TARGET_PAGE_MASK) == 0) {
305 if (start == mmap_next_start && addr >= TASK_UNMAPPED_BASE) {
306 mmap_next_start = addr + size;
311 /* The address is not properly aligned for the target. */
314 /* Assume the result that the kernel gave us is the
315 first with enough free space, so start again at the
316 next higher target page. */
317 addr = TARGET_PAGE_ALIGN(addr);
320 /* Sometimes the kernel decides to perform the allocation
321 at the top end of memory instead. */
322 addr &= TARGET_PAGE_MASK;
325 /* Start over at low memory. */
329 /* Fail. This unaligned block must the last. */
334 /* Since the result the kernel gave didn't fit, start
335 again at low memory. If any repetition, fail. */
336 addr = (repeat ? -1 : 0);
339 /* Unmap and try again. */
342 /* ENOMEM if we checked the whole of the target address space. */
343 if (addr == (abi_ulong)-1) {
344 return (abi_ulong)-1;
345 } else if (addr == 0) {
347 return (abi_ulong)-1;
350 /* Don't actually use 0 when wrapping, instead indicate
351 that we'd truly like an allocation in low memory. */
352 addr = (mmap_min_addr > TARGET_PAGE_SIZE
353 ? TARGET_PAGE_ALIGN(mmap_min_addr)
355 } else if (wrapped && addr >= start) {
356 return (abi_ulong)-1;
361 /* NOTE: all the constants are the HOST ones */
362 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
363 int flags, int fd, abi_ulong offset)
365 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
370 printf("mmap: start=0x" TARGET_ABI_FMT_lx
371 " len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c flags=",
373 prot & PROT_READ ? 'r' : '-',
374 prot & PROT_WRITE ? 'w' : '-',
375 prot & PROT_EXEC ? 'x' : '-');
376 if (flags & MAP_FIXED)
377 printf("MAP_FIXED ");
378 if (flags & MAP_ANONYMOUS)
380 switch(flags & MAP_TYPE) {
382 printf("MAP_PRIVATE ");
385 printf("MAP_SHARED ");
388 printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
391 printf("fd=%d offset=" TARGET_ABI_FMT_lx "\n", fd, offset);
395 if (offset & ~TARGET_PAGE_MASK) {
400 len = TARGET_PAGE_ALIGN(len);
403 real_start = start & qemu_host_page_mask;
404 host_offset = offset & qemu_host_page_mask;
406 /* If the user is asking for the kernel to find a location, do that
407 before we truncate the length for mapping files below. */
408 if (!(flags & MAP_FIXED)) {
409 host_len = len + offset - host_offset;
410 host_len = HOST_PAGE_ALIGN(host_len);
411 start = mmap_find_vma(real_start, host_len);
412 if (start == (abi_ulong)-1) {
418 /* When mapping files into a memory area larger than the file, accesses
419 to pages beyond the file size will cause a SIGBUS.
421 For example, if mmaping a file of 100 bytes on a host with 4K pages
422 emulating a target with 8K pages, the target expects to be able to
423 access the first 8K. But the host will trap us on any access beyond
426 When emulating a target with a larger page-size than the hosts, we
427 may need to truncate file maps at EOF and add extra anonymous pages
428 up to the targets page boundary. */
430 if ((qemu_real_host_page_size < qemu_host_page_size) &&
431 !(flags & MAP_ANONYMOUS)) {
434 if (fstat (fd, &sb) == -1)
437 /* Are we trying to create a map beyond EOF?. */
438 if (offset + len > sb.st_size) {
439 /* If so, truncate the file map at eof aligned with
440 the hosts real pagesize. Additional anonymous maps
441 will be created beyond EOF. */
442 len = REAL_HOST_PAGE_ALIGN(sb.st_size - offset);
446 if (!(flags & MAP_FIXED)) {
447 unsigned long host_start;
450 host_len = len + offset - host_offset;
451 host_len = HOST_PAGE_ALIGN(host_len);
453 /* Note: we prefer to control the mapping address. It is
454 especially important if qemu_host_page_size >
455 qemu_real_host_page_size */
456 p = mmap(g2h(start), host_len, prot,
457 flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
460 /* update start so that it points to the file position at 'offset' */
461 host_start = (unsigned long)p;
462 if (!(flags & MAP_ANONYMOUS)) {
463 p = mmap(g2h(start), len, prot,
464 flags | MAP_FIXED, fd, host_offset);
465 if (p == MAP_FAILED) {
466 munmap(g2h(start), host_len);
469 host_start += offset - host_offset;
471 start = h2g(host_start);
473 if (start & ~TARGET_PAGE_MASK) {
478 real_end = HOST_PAGE_ALIGN(end);
481 * Test if requested memory area fits target address space
482 * It can fail only on 64-bit host with 32-bit target.
483 * On any other target/host host mmap() handles this error correctly.
485 if (!guest_range_valid(start, len)) {
490 /* worst case: we cannot map the file because the offset is not
491 aligned, so we read it */
492 if (!(flags & MAP_ANONYMOUS) &&
493 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
494 /* msync() won't work here, so we return an error if write is
495 possible while it is a shared mapping */
496 if ((flags & MAP_TYPE) == MAP_SHARED &&
497 (prot & PROT_WRITE)) {
501 retaddr = target_mmap(start, len, prot | PROT_WRITE,
502 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
506 if (pread(fd, g2h(start), len, offset) == -1)
508 if (!(prot & PROT_WRITE)) {
509 ret = target_mprotect(start, len, prot);
515 /* handle the start of the mapping */
516 if (start > real_start) {
517 if (real_end == real_start + qemu_host_page_size) {
518 /* one single host page */
519 ret = mmap_frag(real_start, start, end,
520 prot, flags, fd, offset);
525 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
526 prot, flags, fd, offset);
529 real_start += qemu_host_page_size;
531 /* handle the end of the mapping */
532 if (end < real_end) {
533 ret = mmap_frag(real_end - qemu_host_page_size,
534 real_end - qemu_host_page_size, end,
536 offset + real_end - qemu_host_page_size - start);
539 real_end -= qemu_host_page_size;
542 /* map the middle (easier) */
543 if (real_start < real_end) {
545 unsigned long offset1;
546 if (flags & MAP_ANONYMOUS)
549 offset1 = offset + real_start - start;
550 p = mmap(g2h(real_start), real_end - real_start,
551 prot, flags, fd, offset1);
557 page_set_flags(start, start + len, prot | PAGE_VALID);
560 printf("ret=0x" TARGET_ABI_FMT_lx "\n", start);
564 tb_invalidate_phys_range(start, start + len);
572 static void mmap_reserve(abi_ulong start, abi_ulong size)
574 abi_ulong real_start;
580 real_start = start & qemu_host_page_mask;
581 real_end = HOST_PAGE_ALIGN(start + size);
583 if (start > real_start) {
584 /* handle host page containing start */
586 for (addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
587 prot |= page_get_flags(addr);
589 if (real_end == real_start + qemu_host_page_size) {
590 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
591 prot |= page_get_flags(addr);
596 real_start += qemu_host_page_size;
598 if (end < real_end) {
600 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
601 prot |= page_get_flags(addr);
604 real_end -= qemu_host_page_size;
606 if (real_start != real_end) {
607 mmap(g2h(real_start), real_end - real_start, PROT_NONE,
608 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE,
613 int target_munmap(abi_ulong start, abi_ulong len)
615 abi_ulong end, real_start, real_end, addr;
619 printf("munmap: start=0x" TARGET_ABI_FMT_lx " len=0x"
620 TARGET_ABI_FMT_lx "\n",
623 if (start & ~TARGET_PAGE_MASK)
624 return -TARGET_EINVAL;
625 len = TARGET_PAGE_ALIGN(len);
626 if (len == 0 || !guest_range_valid(start, len)) {
627 return -TARGET_EINVAL;
632 real_start = start & qemu_host_page_mask;
633 real_end = HOST_PAGE_ALIGN(end);
635 if (start > real_start) {
636 /* handle host page containing start */
638 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
639 prot |= page_get_flags(addr);
641 if (real_end == real_start + qemu_host_page_size) {
642 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
643 prot |= page_get_flags(addr);
648 real_start += qemu_host_page_size;
650 if (end < real_end) {
652 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
653 prot |= page_get_flags(addr);
656 real_end -= qemu_host_page_size;
660 /* unmap what we can */
661 if (real_start < real_end) {
663 mmap_reserve(real_start, real_end - real_start);
665 ret = munmap(g2h(real_start), real_end - real_start);
670 page_set_flags(start, start + len, 0);
671 tb_invalidate_phys_range(start, start + len);
677 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
678 abi_ulong new_size, unsigned long flags,
684 if (!guest_range_valid(old_addr, old_size) ||
685 ((flags & MREMAP_FIXED) &&
686 !guest_range_valid(new_addr, new_size))) {
693 if (flags & MREMAP_FIXED) {
694 host_addr = mremap(g2h(old_addr), old_size, new_size,
695 flags, g2h(new_addr));
697 if (reserved_va && host_addr != MAP_FAILED) {
698 /* If new and old addresses overlap then the above mremap will
699 already have failed with EINVAL. */
700 mmap_reserve(old_addr, old_size);
702 } else if (flags & MREMAP_MAYMOVE) {
703 abi_ulong mmap_start;
705 mmap_start = mmap_find_vma(0, new_size);
707 if (mmap_start == -1) {
709 host_addr = MAP_FAILED;
711 host_addr = mremap(g2h(old_addr), old_size, new_size,
712 flags | MREMAP_FIXED, g2h(mmap_start));
714 mmap_reserve(old_addr, old_size);
719 if (reserved_va && old_size < new_size) {
721 for (addr = old_addr + old_size;
722 addr < old_addr + new_size;
724 prot |= page_get_flags(addr);
728 host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
729 if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
730 mmap_reserve(old_addr + old_size, new_size - old_size);
734 host_addr = MAP_FAILED;
736 /* Check if address fits target address space */
737 if ((unsigned long)host_addr + new_size > (abi_ulong)-1) {
738 /* Revert mremap() changes */
739 host_addr = mremap(g2h(old_addr), new_size, old_size, flags);
741 host_addr = MAP_FAILED;
745 if (host_addr == MAP_FAILED) {
748 new_addr = h2g(host_addr);
749 prot = page_get_flags(old_addr);
750 page_set_flags(old_addr, old_addr + old_size, 0);
751 page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
753 tb_invalidate_phys_range(new_addr, new_addr + new_size);