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/>.
25 #include <sys/types.h>
28 #include <linux/mman.h>
29 #include <linux/unistd.h>
32 #include "qemu-common.h"
36 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
37 static __thread int mmap_lock_count;
41 if (mmap_lock_count++ == 0) {
42 pthread_mutex_lock(&mmap_mutex);
46 void mmap_unlock(void)
48 if (--mmap_lock_count == 0) {
49 pthread_mutex_unlock(&mmap_mutex);
53 /* Grab lock to make sure things are in a consistent state after fork(). */
54 void mmap_fork_start(void)
58 pthread_mutex_lock(&mmap_mutex);
61 void mmap_fork_end(int child)
64 pthread_mutex_init(&mmap_mutex, NULL);
66 pthread_mutex_unlock(&mmap_mutex);
69 /* NOTE: all the constants are the HOST ones, but addresses are target. */
70 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
72 abi_ulong end, host_start, host_end, addr;
76 printf("mprotect: start=0x" TARGET_ABI_FMT_lx
77 "len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c\n", start, len,
78 prot & PROT_READ ? 'r' : '-',
79 prot & PROT_WRITE ? 'w' : '-',
80 prot & PROT_EXEC ? 'x' : '-');
83 if ((start & ~TARGET_PAGE_MASK) != 0)
85 len = TARGET_PAGE_ALIGN(len);
89 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
94 host_start = start & qemu_host_page_mask;
95 host_end = HOST_PAGE_ALIGN(end);
96 if (start > host_start) {
97 /* handle host page containing start */
99 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
100 prot1 |= page_get_flags(addr);
102 if (host_end == host_start + qemu_host_page_size) {
103 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
104 prot1 |= page_get_flags(addr);
108 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
111 host_start += qemu_host_page_size;
113 if (end < host_end) {
115 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
116 prot1 |= page_get_flags(addr);
118 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
122 host_end -= qemu_host_page_size;
125 /* handle the pages in the middle */
126 if (host_start < host_end) {
127 ret = mprotect(g2h(host_start), host_end - host_start, prot);
131 page_set_flags(start, start + len, prot | PAGE_VALID);
139 /* map an incomplete host page */
140 static int mmap_frag(abi_ulong real_start,
141 abi_ulong start, abi_ulong end,
142 int prot, int flags, int fd, abi_ulong offset)
144 abi_ulong real_end, addr;
148 real_end = real_start + qemu_host_page_size;
149 host_start = g2h(real_start);
151 /* get the protection of the target pages outside the mapping */
153 for(addr = real_start; addr < real_end; addr++) {
154 if (addr < start || addr >= end)
155 prot1 |= page_get_flags(addr);
159 /* no page was there, so we allocate one */
160 void *p = mmap(host_start, qemu_host_page_size, prot,
161 flags | MAP_ANONYMOUS, -1, 0);
168 prot_new = prot | prot1;
169 if (!(flags & MAP_ANONYMOUS)) {
170 /* msync() won't work here, so we return an error if write is
171 possible while it is a shared mapping */
172 if ((flags & MAP_TYPE) == MAP_SHARED &&
176 /* adjust protection to be able to read */
177 if (!(prot1 & PROT_WRITE))
178 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
180 /* read the corresponding file data */
181 if (pread(fd, g2h(start), end - start, offset) == -1)
184 /* put final protection */
185 if (prot_new != (prot1 | PROT_WRITE))
186 mprotect(host_start, qemu_host_page_size, prot_new);
188 /* just update the protection */
189 if (prot_new != prot1) {
190 mprotect(host_start, qemu_host_page_size, prot_new);
196 #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64
197 # define TASK_UNMAPPED_BASE (1ul << 38)
198 #elif defined(__CYGWIN__)
199 /* Cygwin doesn't have a whole lot of address space. */
200 # define TASK_UNMAPPED_BASE 0x18000000
202 # define TASK_UNMAPPED_BASE 0x40000000
204 abi_ulong mmap_next_start = TASK_UNMAPPED_BASE;
206 unsigned long last_brk;
208 #ifdef CONFIG_USE_GUEST_BASE
209 /* Subroutine of mmap_find_vma, used when we have pre-allocated a chunk
210 of guest address space. */
211 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size)
218 if (size > RESERVED_VA) {
219 return (abi_ulong)-1;
222 size = HOST_PAGE_ALIGN(size);
223 end_addr = start + size;
224 if (end_addr > RESERVED_VA) {
225 end_addr = RESERVED_VA;
227 addr = end_addr - qemu_host_page_size;
230 if (addr > end_addr) {
232 return (abi_ulong)-1;
234 end_addr = RESERVED_VA;
235 addr = end_addr - qemu_host_page_size;
239 prot = page_get_flags(addr);
243 if (addr + size == end_addr) {
246 addr -= qemu_host_page_size;
249 if (start == mmap_next_start) {
250 mmap_next_start = addr;
258 * Find and reserve a free memory area of size 'size'. The search
260 * It must be called with mmap_lock() held.
261 * Return -1 if error.
263 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
269 /* If 'start' == 0, then a default start address is used. */
271 start = mmap_next_start;
273 start &= qemu_host_page_mask;
276 size = HOST_PAGE_ALIGN(size);
278 #ifdef CONFIG_USE_GUEST_BASE
280 return mmap_find_vma_reserved(start, size);
285 wrapped = repeat = 0;
288 for (;; prev = ptr) {
290 * Reserve needed memory area to avoid a race.
291 * It should be discarded using:
292 * - mmap() with MAP_FIXED flag
293 * - mremap() with MREMAP_FIXED flag
294 * - shmat() with SHM_REMAP flag
296 ptr = mmap(g2h(addr), size, PROT_NONE,
297 MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
299 /* ENOMEM, if host address space has no memory */
300 if (ptr == MAP_FAILED) {
301 return (abi_ulong)-1;
304 /* Count the number of sequential returns of the same address.
305 This is used to modify the search algorithm below. */
306 repeat = (ptr == prev ? repeat + 1 : 0);
308 if (h2g_valid(ptr + size - 1)) {
311 if ((addr & ~TARGET_PAGE_MASK) == 0) {
313 if (start == mmap_next_start && addr >= TASK_UNMAPPED_BASE) {
314 mmap_next_start = addr + size;
319 /* The address is not properly aligned for the target. */
322 /* Assume the result that the kernel gave us is the
323 first with enough free space, so start again at the
324 next higher target page. */
325 addr = TARGET_PAGE_ALIGN(addr);
328 /* Sometimes the kernel decides to perform the allocation
329 at the top end of memory instead. */
330 addr &= TARGET_PAGE_MASK;
333 /* Start over at low memory. */
337 /* Fail. This unaligned block must the last. */
342 /* Since the result the kernel gave didn't fit, start
343 again at low memory. If any repetition, fail. */
344 addr = (repeat ? -1 : 0);
347 /* Unmap and try again. */
350 /* ENOMEM if we checked the whole of the target address space. */
351 if (addr == (abi_ulong)-1) {
352 return (abi_ulong)-1;
353 } else if (addr == 0) {
355 return (abi_ulong)-1;
358 /* Don't actually use 0 when wrapping, instead indicate
359 that we'd truly like an allocation in low memory. */
360 addr = (mmap_min_addr > TARGET_PAGE_SIZE
361 ? TARGET_PAGE_ALIGN(mmap_min_addr)
363 } else if (wrapped && addr >= start) {
364 return (abi_ulong)-1;
369 /* NOTE: all the constants are the HOST ones */
370 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
371 int flags, int fd, abi_ulong offset)
373 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
378 printf("mmap: start=0x" TARGET_ABI_FMT_lx
379 " len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c flags=",
381 prot & PROT_READ ? 'r' : '-',
382 prot & PROT_WRITE ? 'w' : '-',
383 prot & PROT_EXEC ? 'x' : '-');
384 if (flags & MAP_FIXED)
385 printf("MAP_FIXED ");
386 if (flags & MAP_ANONYMOUS)
388 switch(flags & MAP_TYPE) {
390 printf("MAP_PRIVATE ");
393 printf("MAP_SHARED ");
396 printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
399 printf("fd=%d offset=" TARGET_ABI_FMT_lx "\n", fd, offset);
403 if (offset & ~TARGET_PAGE_MASK) {
408 len = TARGET_PAGE_ALIGN(len);
411 real_start = start & qemu_host_page_mask;
412 host_offset = offset & qemu_host_page_mask;
414 /* If the user is asking for the kernel to find a location, do that
415 before we truncate the length for mapping files below. */
416 if (!(flags & MAP_FIXED)) {
417 host_len = len + offset - host_offset;
418 host_len = HOST_PAGE_ALIGN(host_len);
419 start = mmap_find_vma(real_start, host_len);
420 if (start == (abi_ulong)-1) {
426 /* When mapping files into a memory area larger than the file, accesses
427 to pages beyond the file size will cause a SIGBUS.
429 For example, if mmaping a file of 100 bytes on a host with 4K pages
430 emulating a target with 8K pages, the target expects to be able to
431 access the first 8K. But the host will trap us on any access beyond
434 When emulating a target with a larger page-size than the hosts, we
435 may need to truncate file maps at EOF and add extra anonymous pages
436 up to the targets page boundary. */
438 if ((qemu_real_host_page_size < TARGET_PAGE_SIZE)
439 && !(flags & MAP_ANONYMOUS)) {
442 if (fstat (fd, &sb) == -1)
445 /* Are we trying to create a map beyond EOF?. */
446 if (offset + len > sb.st_size) {
447 /* If so, truncate the file map at eof aligned with
448 the hosts real pagesize. Additional anonymous maps
449 will be created beyond EOF. */
450 len = (sb.st_size - offset);
451 len += qemu_real_host_page_size - 1;
452 len &= ~(qemu_real_host_page_size - 1);
456 if (!(flags & MAP_FIXED)) {
457 unsigned long host_start;
460 host_len = len + offset - host_offset;
461 host_len = HOST_PAGE_ALIGN(host_len);
463 /* Note: we prefer to control the mapping address. It is
464 especially important if qemu_host_page_size >
465 qemu_real_host_page_size */
466 p = mmap(g2h(start), host_len, prot,
467 flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
470 /* update start so that it points to the file position at 'offset' */
471 host_start = (unsigned long)p;
472 if (!(flags & MAP_ANONYMOUS)) {
473 p = mmap(g2h(start), len, prot,
474 flags | MAP_FIXED, fd, host_offset);
475 if (p == MAP_FAILED) {
476 munmap(g2h(start), host_len);
479 host_start += offset - host_offset;
481 start = h2g(host_start);
483 if (start & ~TARGET_PAGE_MASK) {
488 real_end = HOST_PAGE_ALIGN(end);
491 * Test if requested memory area fits target address space
492 * It can fail only on 64-bit host with 32-bit target.
493 * On any other target/host host mmap() handles this error correctly.
495 if ((unsigned long)start + len - 1 > (abi_ulong) -1) {
500 /* worst case: we cannot map the file because the offset is not
501 aligned, so we read it */
502 if (!(flags & MAP_ANONYMOUS) &&
503 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
504 /* msync() won't work here, so we return an error if write is
505 possible while it is a shared mapping */
506 if ((flags & MAP_TYPE) == MAP_SHARED &&
507 (prot & PROT_WRITE)) {
511 retaddr = target_mmap(start, len, prot | PROT_WRITE,
512 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
516 if (pread(fd, g2h(start), len, offset) == -1)
518 if (!(prot & PROT_WRITE)) {
519 ret = target_mprotect(start, len, prot);
528 /* handle the start of the mapping */
529 if (start > real_start) {
530 if (real_end == real_start + qemu_host_page_size) {
531 /* one single host page */
532 ret = mmap_frag(real_start, start, end,
533 prot, flags, fd, offset);
538 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
539 prot, flags, fd, offset);
542 real_start += qemu_host_page_size;
544 /* handle the end of the mapping */
545 if (end < real_end) {
546 ret = mmap_frag(real_end - qemu_host_page_size,
547 real_end - qemu_host_page_size, real_end,
549 offset + real_end - qemu_host_page_size - start);
552 real_end -= qemu_host_page_size;
555 /* map the middle (easier) */
556 if (real_start < real_end) {
558 unsigned long offset1;
559 if (flags & MAP_ANONYMOUS)
562 offset1 = offset + real_start - start;
563 p = mmap(g2h(real_start), real_end - real_start,
564 prot, flags, fd, offset1);
570 page_set_flags(start, start + len, prot | PAGE_VALID);
573 printf("ret=0x" TARGET_ABI_FMT_lx "\n", start);
577 tb_invalidate_phys_range(start, start + len, 0);
585 static void mmap_reserve(abi_ulong start, abi_ulong size)
587 abi_ulong real_start;
593 real_start = start & qemu_host_page_mask;
594 real_end = HOST_PAGE_ALIGN(start + size);
596 if (start > real_start) {
597 /* handle host page containing start */
599 for (addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
600 prot |= page_get_flags(addr);
602 if (real_end == real_start + qemu_host_page_size) {
603 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
604 prot |= page_get_flags(addr);
609 real_start += qemu_host_page_size;
611 if (end < real_end) {
613 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
614 prot |= page_get_flags(addr);
617 real_end -= qemu_host_page_size;
619 if (real_start != real_end) {
620 mmap(g2h(real_start), real_end - real_start, PROT_NONE,
621 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE,
626 int target_munmap(abi_ulong start, abi_ulong len)
628 abi_ulong end, real_start, real_end, addr;
632 printf("munmap: start=0x" TARGET_ABI_FMT_lx " len=0x"
633 TARGET_ABI_FMT_lx "\n",
636 if (start & ~TARGET_PAGE_MASK)
638 len = TARGET_PAGE_ALIGN(len);
643 real_start = start & qemu_host_page_mask;
644 real_end = HOST_PAGE_ALIGN(end);
646 if (start > real_start) {
647 /* handle host page containing start */
649 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
650 prot |= page_get_flags(addr);
652 if (real_end == real_start + qemu_host_page_size) {
653 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
654 prot |= page_get_flags(addr);
659 real_start += qemu_host_page_size;
661 if (end < real_end) {
663 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
664 prot |= page_get_flags(addr);
667 real_end -= qemu_host_page_size;
671 /* unmap what we can */
672 if (real_start < real_end) {
674 mmap_reserve(real_start, real_end - real_start);
676 ret = munmap(g2h(real_start), real_end - real_start);
681 page_set_flags(start, start + len, 0);
682 tb_invalidate_phys_range(start, start + len, 0);
688 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
689 abi_ulong new_size, unsigned long flags,
697 if (flags & MREMAP_FIXED) {
698 host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
703 if (RESERVED_VA && host_addr != MAP_FAILED) {
704 /* If new and old addresses overlap then the above mremap will
705 already have failed with EINVAL. */
706 mmap_reserve(old_addr, old_size);
708 } else if (flags & MREMAP_MAYMOVE) {
709 abi_ulong mmap_start;
711 mmap_start = mmap_find_vma(0, new_size);
713 if (mmap_start == -1) {
715 host_addr = MAP_FAILED;
717 host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
719 flags | MREMAP_FIXED,
722 mmap_reserve(old_addr, old_size);
727 if (RESERVED_VA && old_size < new_size) {
729 for (addr = old_addr + old_size;
730 addr < old_addr + new_size;
732 prot |= page_get_flags(addr);
736 host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
737 if (host_addr != MAP_FAILED && RESERVED_VA && old_size > new_size) {
738 mmap_reserve(old_addr + old_size, new_size - old_size);
742 host_addr = MAP_FAILED;
744 /* Check if address fits target address space */
745 if ((unsigned long)host_addr + new_size > (abi_ulong)-1) {
746 /* Revert mremap() changes */
747 host_addr = mremap(g2h(old_addr), new_size, old_size, flags);
749 host_addr = MAP_FAILED;
753 if (host_addr == MAP_FAILED) {
756 new_addr = h2g(host_addr);
757 prot = page_get_flags(old_addr);
758 page_set_flags(old_addr, old_addr + old_size, 0);
759 page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
761 tb_invalidate_phys_range(new_addr, new_addr + new_size, 0);
766 int target_msync(abi_ulong start, abi_ulong len, int flags)
770 if (start & ~TARGET_PAGE_MASK)
772 len = TARGET_PAGE_ALIGN(len);
779 start &= qemu_host_page_mask;
780 return msync(g2h(start), end - start, flags);