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"
26 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
27 static __thread int mmap_lock_count;
31 if (mmap_lock_count++ == 0) {
32 pthread_mutex_lock(&mmap_mutex);
36 void mmap_unlock(void)
38 if (--mmap_lock_count == 0) {
39 pthread_mutex_unlock(&mmap_mutex);
43 bool have_mmap_lock(void)
45 return mmap_lock_count > 0 ? true : false;
48 /* Grab lock to make sure things are in a consistent state after fork(). */
49 void mmap_fork_start(void)
53 pthread_mutex_lock(&mmap_mutex);
56 void mmap_fork_end(int child)
59 pthread_mutex_init(&mmap_mutex, NULL);
61 pthread_mutex_unlock(&mmap_mutex);
64 /* NOTE: all the constants are the HOST ones, but addresses are target. */
65 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
67 abi_ulong end, host_start, host_end, addr;
71 printf("mprotect: start=0x" TARGET_ABI_FMT_lx
72 "len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c\n", start, len,
73 prot & PROT_READ ? 'r' : '-',
74 prot & PROT_WRITE ? 'w' : '-',
75 prot & PROT_EXEC ? 'x' : '-');
78 if ((start & ~TARGET_PAGE_MASK) != 0)
79 return -TARGET_EINVAL;
80 len = TARGET_PAGE_ALIGN(len);
82 if (!guest_range_valid(start, len)) {
83 return -TARGET_ENOMEM;
85 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
90 host_start = start & qemu_host_page_mask;
91 host_end = HOST_PAGE_ALIGN(end);
92 if (start > host_start) {
93 /* handle host page containing start */
95 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
96 prot1 |= page_get_flags(addr);
98 if (host_end == host_start + qemu_host_page_size) {
99 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
100 prot1 |= page_get_flags(addr);
104 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
107 host_start += qemu_host_page_size;
109 if (end < host_end) {
111 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
112 prot1 |= page_get_flags(addr);
114 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
118 host_end -= qemu_host_page_size;
121 /* handle the pages in the middle */
122 if (host_start < host_end) {
123 ret = mprotect(g2h(host_start), host_end - host_start, prot);
127 page_set_flags(start, start + len, prot | PAGE_VALID);
135 /* map an incomplete host page */
136 static int mmap_frag(abi_ulong real_start,
137 abi_ulong start, abi_ulong end,
138 int prot, int flags, int fd, abi_ulong offset)
140 abi_ulong real_end, addr;
144 real_end = real_start + qemu_host_page_size;
145 host_start = g2h(real_start);
147 /* get the protection of the target pages outside the mapping */
149 for(addr = real_start; addr < real_end; addr++) {
150 if (addr < start || addr >= end)
151 prot1 |= page_get_flags(addr);
155 /* no page was there, so we allocate one */
156 void *p = mmap(host_start, qemu_host_page_size, prot,
157 flags | MAP_ANONYMOUS, -1, 0);
164 prot_new = prot | prot1;
165 if (!(flags & MAP_ANONYMOUS)) {
166 /* msync() won't work here, so we return an error if write is
167 possible while it is a shared mapping */
168 if ((flags & MAP_TYPE) == MAP_SHARED &&
172 /* adjust protection to be able to read */
173 if (!(prot1 & PROT_WRITE))
174 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
176 /* read the corresponding file data */
177 if (pread(fd, g2h(start), end - start, offset) == -1)
180 /* put final protection */
181 if (prot_new != (prot1 | PROT_WRITE))
182 mprotect(host_start, qemu_host_page_size, prot_new);
184 if (prot_new != prot1) {
185 mprotect(host_start, qemu_host_page_size, prot_new);
187 if (prot_new & PROT_WRITE) {
188 memset(g2h(start), 0, end - start);
194 #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64
195 # define TASK_UNMAPPED_BASE (1ul << 38)
197 # define TASK_UNMAPPED_BASE 0x40000000
199 abi_ulong mmap_next_start = TASK_UNMAPPED_BASE;
201 unsigned long last_brk;
203 /* Subroutine of mmap_find_vma, used when we have pre-allocated a chunk
204 of guest address space. */
205 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size)
212 if (size > reserved_va) {
213 return (abi_ulong)-1;
216 size = HOST_PAGE_ALIGN(size);
217 end_addr = start + size;
218 if (end_addr > reserved_va) {
219 end_addr = reserved_va;
221 addr = end_addr - qemu_host_page_size;
224 if (addr > end_addr) {
226 return (abi_ulong)-1;
228 end_addr = reserved_va;
229 addr = end_addr - qemu_host_page_size;
233 prot = page_get_flags(addr);
237 if (addr && addr + size == end_addr) {
240 addr -= qemu_host_page_size;
243 if (start == mmap_next_start) {
244 mmap_next_start = addr;
251 * Find and reserve a free memory area of size 'size'. The search
253 * It must be called with mmap_lock() held.
254 * Return -1 if error.
256 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
262 /* If 'start' == 0, then a default start address is used. */
264 start = mmap_next_start;
266 start &= qemu_host_page_mask;
269 size = HOST_PAGE_ALIGN(size);
272 return mmap_find_vma_reserved(start, size);
276 wrapped = repeat = 0;
279 for (;; prev = ptr) {
281 * Reserve needed memory area to avoid a race.
282 * It should be discarded using:
283 * - mmap() with MAP_FIXED flag
284 * - mremap() with MREMAP_FIXED flag
285 * - shmat() with SHM_REMAP flag
287 ptr = mmap(g2h(addr), size, PROT_NONE,
288 MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
290 /* ENOMEM, if host address space has no memory */
291 if (ptr == MAP_FAILED) {
292 return (abi_ulong)-1;
295 /* Count the number of sequential returns of the same address.
296 This is used to modify the search algorithm below. */
297 repeat = (ptr == prev ? repeat + 1 : 0);
299 if (h2g_valid(ptr + size - 1)) {
302 if ((addr & ~TARGET_PAGE_MASK) == 0) {
304 if (start == mmap_next_start && addr >= TASK_UNMAPPED_BASE) {
305 mmap_next_start = addr + size;
310 /* The address is not properly aligned for the target. */
313 /* Assume the result that the kernel gave us is the
314 first with enough free space, so start again at the
315 next higher target page. */
316 addr = TARGET_PAGE_ALIGN(addr);
319 /* Sometimes the kernel decides to perform the allocation
320 at the top end of memory instead. */
321 addr &= TARGET_PAGE_MASK;
324 /* Start over at low memory. */
328 /* Fail. This unaligned block must the last. */
333 /* Since the result the kernel gave didn't fit, start
334 again at low memory. If any repetition, fail. */
335 addr = (repeat ? -1 : 0);
338 /* Unmap and try again. */
341 /* ENOMEM if we checked the whole of the target address space. */
342 if (addr == (abi_ulong)-1) {
343 return (abi_ulong)-1;
344 } else if (addr == 0) {
346 return (abi_ulong)-1;
349 /* Don't actually use 0 when wrapping, instead indicate
350 that we'd truly like an allocation in low memory. */
351 addr = (mmap_min_addr > TARGET_PAGE_SIZE
352 ? TARGET_PAGE_ALIGN(mmap_min_addr)
354 } else if (wrapped && addr >= start) {
355 return (abi_ulong)-1;
360 /* NOTE: all the constants are the HOST ones */
361 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
362 int flags, int fd, abi_ulong offset)
364 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
369 printf("mmap: start=0x" TARGET_ABI_FMT_lx
370 " len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c flags=",
372 prot & PROT_READ ? 'r' : '-',
373 prot & PROT_WRITE ? 'w' : '-',
374 prot & PROT_EXEC ? 'x' : '-');
375 if (flags & MAP_FIXED)
376 printf("MAP_FIXED ");
377 if (flags & MAP_ANONYMOUS)
379 switch(flags & MAP_TYPE) {
381 printf("MAP_PRIVATE ");
384 printf("MAP_SHARED ");
387 printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
390 printf("fd=%d offset=" TARGET_ABI_FMT_lx "\n", fd, offset);
394 if (offset & ~TARGET_PAGE_MASK) {
399 len = TARGET_PAGE_ALIGN(len);
402 real_start = start & qemu_host_page_mask;
403 host_offset = offset & qemu_host_page_mask;
405 /* If the user is asking for the kernel to find a location, do that
406 before we truncate the length for mapping files below. */
407 if (!(flags & MAP_FIXED)) {
408 host_len = len + offset - host_offset;
409 host_len = HOST_PAGE_ALIGN(host_len);
410 start = mmap_find_vma(real_start, host_len);
411 if (start == (abi_ulong)-1) {
417 /* When mapping files into a memory area larger than the file, accesses
418 to pages beyond the file size will cause a SIGBUS.
420 For example, if mmaping a file of 100 bytes on a host with 4K pages
421 emulating a target with 8K pages, the target expects to be able to
422 access the first 8K. But the host will trap us on any access beyond
425 When emulating a target with a larger page-size than the hosts, we
426 may need to truncate file maps at EOF and add extra anonymous pages
427 up to the targets page boundary. */
429 if ((qemu_real_host_page_size < qemu_host_page_size) &&
430 !(flags & MAP_ANONYMOUS)) {
433 if (fstat (fd, &sb) == -1)
436 /* Are we trying to create a map beyond EOF?. */
437 if (offset + len > sb.st_size) {
438 /* If so, truncate the file map at eof aligned with
439 the hosts real pagesize. Additional anonymous maps
440 will be created beyond EOF. */
441 len = REAL_HOST_PAGE_ALIGN(sb.st_size - offset);
445 if (!(flags & MAP_FIXED)) {
446 unsigned long host_start;
449 host_len = len + offset - host_offset;
450 host_len = HOST_PAGE_ALIGN(host_len);
452 /* Note: we prefer to control the mapping address. It is
453 especially important if qemu_host_page_size >
454 qemu_real_host_page_size */
455 p = mmap(g2h(start), host_len, prot,
456 flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
459 /* update start so that it points to the file position at 'offset' */
460 host_start = (unsigned long)p;
461 if (!(flags & MAP_ANONYMOUS)) {
462 p = mmap(g2h(start), len, prot,
463 flags | MAP_FIXED, fd, host_offset);
464 if (p == MAP_FAILED) {
465 munmap(g2h(start), host_len);
468 host_start += offset - host_offset;
470 start = h2g(host_start);
472 if (start & ~TARGET_PAGE_MASK) {
477 real_end = HOST_PAGE_ALIGN(end);
480 * Test if requested memory area fits target address space
481 * It can fail only on 64-bit host with 32-bit target.
482 * On any other target/host host mmap() handles this error correctly.
484 if (!guest_range_valid(start, len)) {
489 /* worst case: we cannot map the file because the offset is not
490 aligned, so we read it */
491 if (!(flags & MAP_ANONYMOUS) &&
492 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
493 /* msync() won't work here, so we return an error if write is
494 possible while it is a shared mapping */
495 if ((flags & MAP_TYPE) == MAP_SHARED &&
496 (prot & PROT_WRITE)) {
500 retaddr = target_mmap(start, len, prot | PROT_WRITE,
501 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
505 if (pread(fd, g2h(start), len, offset) == -1)
507 if (!(prot & PROT_WRITE)) {
508 ret = target_mprotect(start, len, prot);
514 /* handle the start of the mapping */
515 if (start > real_start) {
516 if (real_end == real_start + qemu_host_page_size) {
517 /* one single host page */
518 ret = mmap_frag(real_start, start, end,
519 prot, flags, fd, offset);
524 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
525 prot, flags, fd, offset);
528 real_start += qemu_host_page_size;
530 /* handle the end of the mapping */
531 if (end < real_end) {
532 ret = mmap_frag(real_end - qemu_host_page_size,
533 real_end - qemu_host_page_size, end,
535 offset + real_end - qemu_host_page_size - start);
538 real_end -= qemu_host_page_size;
541 /* map the middle (easier) */
542 if (real_start < real_end) {
544 unsigned long offset1;
545 if (flags & MAP_ANONYMOUS)
548 offset1 = offset + real_start - start;
549 p = mmap(g2h(real_start), real_end - real_start,
550 prot, flags, fd, offset1);
556 page_set_flags(start, start + len, prot | PAGE_VALID);
559 printf("ret=0x" TARGET_ABI_FMT_lx "\n", start);
563 tb_invalidate_phys_range(start, start + len);
571 static void mmap_reserve(abi_ulong start, abi_ulong size)
573 abi_ulong real_start;
579 real_start = start & qemu_host_page_mask;
580 real_end = HOST_PAGE_ALIGN(start + size);
582 if (start > real_start) {
583 /* handle host page containing start */
585 for (addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
586 prot |= page_get_flags(addr);
588 if (real_end == real_start + qemu_host_page_size) {
589 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
590 prot |= page_get_flags(addr);
595 real_start += qemu_host_page_size;
597 if (end < real_end) {
599 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
600 prot |= page_get_flags(addr);
603 real_end -= qemu_host_page_size;
605 if (real_start != real_end) {
606 mmap(g2h(real_start), real_end - real_start, PROT_NONE,
607 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE,
612 int target_munmap(abi_ulong start, abi_ulong len)
614 abi_ulong end, real_start, real_end, addr;
618 printf("munmap: start=0x" TARGET_ABI_FMT_lx " len=0x"
619 TARGET_ABI_FMT_lx "\n",
622 if (start & ~TARGET_PAGE_MASK)
623 return -TARGET_EINVAL;
624 len = TARGET_PAGE_ALIGN(len);
625 if (len == 0 || !guest_range_valid(start, len)) {
626 return -TARGET_EINVAL;
631 real_start = start & qemu_host_page_mask;
632 real_end = HOST_PAGE_ALIGN(end);
634 if (start > real_start) {
635 /* handle host page containing start */
637 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
638 prot |= page_get_flags(addr);
640 if (real_end == real_start + qemu_host_page_size) {
641 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
642 prot |= page_get_flags(addr);
647 real_start += qemu_host_page_size;
649 if (end < real_end) {
651 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
652 prot |= page_get_flags(addr);
655 real_end -= qemu_host_page_size;
659 /* unmap what we can */
660 if (real_start < real_end) {
662 mmap_reserve(real_start, real_end - real_start);
664 ret = munmap(g2h(real_start), real_end - real_start);
669 page_set_flags(start, start + len, 0);
670 tb_invalidate_phys_range(start, start + len);
676 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
677 abi_ulong new_size, unsigned long flags,
683 if (!guest_range_valid(old_addr, old_size) ||
684 ((flags & MREMAP_FIXED) &&
685 !guest_range_valid(new_addr, new_size))) {
692 if (flags & MREMAP_FIXED) {
693 host_addr = mremap(g2h(old_addr), old_size, new_size,
694 flags, g2h(new_addr));
696 if (reserved_va && host_addr != MAP_FAILED) {
697 /* If new and old addresses overlap then the above mremap will
698 already have failed with EINVAL. */
699 mmap_reserve(old_addr, old_size);
701 } else if (flags & MREMAP_MAYMOVE) {
702 abi_ulong mmap_start;
704 mmap_start = mmap_find_vma(0, new_size);
706 if (mmap_start == -1) {
708 host_addr = MAP_FAILED;
710 host_addr = mremap(g2h(old_addr), old_size, new_size,
711 flags | MREMAP_FIXED, g2h(mmap_start));
713 mmap_reserve(old_addr, old_size);
718 if (reserved_va && old_size < new_size) {
720 for (addr = old_addr + old_size;
721 addr < old_addr + new_size;
723 prot |= page_get_flags(addr);
727 host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
728 if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
729 mmap_reserve(old_addr + old_size, new_size - old_size);
733 host_addr = MAP_FAILED;
735 /* Check if address fits target address space */
736 if ((unsigned long)host_addr + new_size > (abi_ulong)-1) {
737 /* Revert mremap() changes */
738 host_addr = mremap(g2h(old_addr), new_size, old_size, flags);
740 host_addr = MAP_FAILED;
744 if (host_addr == MAP_FAILED) {
747 new_addr = h2g(host_addr);
748 prot = page_get_flags(old_addr);
749 page_set_flags(old_addr, old_addr + old_size, 0);
750 page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
752 tb_invalidate_phys_range(new_addr, new_addr + new_size);