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)
81 len = TARGET_PAGE_ALIGN(len);
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)
196 #elif defined(__CYGWIN__)
197 /* Cygwin doesn't have a whole lot of address space. */
198 # define TASK_UNMAPPED_BASE 0x18000000
200 # define TASK_UNMAPPED_BASE 0x40000000
202 abi_ulong mmap_next_start = TASK_UNMAPPED_BASE;
204 unsigned long last_brk;
206 /* Subroutine of mmap_find_vma, used when we have pre-allocated a chunk
207 of guest address space. */
208 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size)
215 if (size > reserved_va) {
216 return (abi_ulong)-1;
219 size = HOST_PAGE_ALIGN(size);
220 end_addr = start + size;
221 if (end_addr > reserved_va) {
222 end_addr = reserved_va;
224 addr = end_addr - qemu_host_page_size;
227 if (addr > end_addr) {
229 return (abi_ulong)-1;
231 end_addr = reserved_va;
232 addr = end_addr - qemu_host_page_size;
236 prot = page_get_flags(addr);
240 if (addr + size == end_addr) {
243 addr -= qemu_host_page_size;
246 if (start == mmap_next_start) {
247 mmap_next_start = addr;
254 * Find and reserve a free memory area of size 'size'. The search
256 * It must be called with mmap_lock() held.
257 * Return -1 if error.
259 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
265 /* If 'start' == 0, then a default start address is used. */
267 start = mmap_next_start;
269 start &= qemu_host_page_mask;
272 size = HOST_PAGE_ALIGN(size);
275 return mmap_find_vma_reserved(start, size);
279 wrapped = repeat = 0;
282 for (;; prev = ptr) {
284 * Reserve needed memory area to avoid a race.
285 * It should be discarded using:
286 * - mmap() with MAP_FIXED flag
287 * - mremap() with MREMAP_FIXED flag
288 * - shmat() with SHM_REMAP flag
290 ptr = mmap(g2h(addr), size, PROT_NONE,
291 MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
293 /* ENOMEM, if host address space has no memory */
294 if (ptr == MAP_FAILED) {
295 return (abi_ulong)-1;
298 /* Count the number of sequential returns of the same address.
299 This is used to modify the search algorithm below. */
300 repeat = (ptr == prev ? repeat + 1 : 0);
302 if (h2g_valid(ptr + size - 1)) {
305 if ((addr & ~TARGET_PAGE_MASK) == 0) {
307 if (start == mmap_next_start && addr >= TASK_UNMAPPED_BASE) {
308 mmap_next_start = addr + size;
313 /* The address is not properly aligned for the target. */
316 /* Assume the result that the kernel gave us is the
317 first with enough free space, so start again at the
318 next higher target page. */
319 addr = TARGET_PAGE_ALIGN(addr);
322 /* Sometimes the kernel decides to perform the allocation
323 at the top end of memory instead. */
324 addr &= TARGET_PAGE_MASK;
327 /* Start over at low memory. */
331 /* Fail. This unaligned block must the last. */
336 /* Since the result the kernel gave didn't fit, start
337 again at low memory. If any repetition, fail. */
338 addr = (repeat ? -1 : 0);
341 /* Unmap and try again. */
344 /* ENOMEM if we checked the whole of the target address space. */
345 if (addr == (abi_ulong)-1) {
346 return (abi_ulong)-1;
347 } else if (addr == 0) {
349 return (abi_ulong)-1;
352 /* Don't actually use 0 when wrapping, instead indicate
353 that we'd truly like an allocation in low memory. */
354 addr = (mmap_min_addr > TARGET_PAGE_SIZE
355 ? TARGET_PAGE_ALIGN(mmap_min_addr)
357 } else if (wrapped && addr >= start) {
358 return (abi_ulong)-1;
363 /* NOTE: all the constants are the HOST ones */
364 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
365 int flags, int fd, abi_ulong offset)
367 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
372 printf("mmap: start=0x" TARGET_ABI_FMT_lx
373 " len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c flags=",
375 prot & PROT_READ ? 'r' : '-',
376 prot & PROT_WRITE ? 'w' : '-',
377 prot & PROT_EXEC ? 'x' : '-');
378 if (flags & MAP_FIXED)
379 printf("MAP_FIXED ");
380 if (flags & MAP_ANONYMOUS)
382 switch(flags & MAP_TYPE) {
384 printf("MAP_PRIVATE ");
387 printf("MAP_SHARED ");
390 printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
393 printf("fd=%d offset=" TARGET_ABI_FMT_lx "\n", fd, offset);
397 if (offset & ~TARGET_PAGE_MASK) {
402 len = TARGET_PAGE_ALIGN(len);
405 real_start = start & qemu_host_page_mask;
406 host_offset = offset & qemu_host_page_mask;
408 /* If the user is asking for the kernel to find a location, do that
409 before we truncate the length for mapping files below. */
410 if (!(flags & MAP_FIXED)) {
411 host_len = len + offset - host_offset;
412 host_len = HOST_PAGE_ALIGN(host_len);
413 start = mmap_find_vma(real_start, host_len);
414 if (start == (abi_ulong)-1) {
420 /* When mapping files into a memory area larger than the file, accesses
421 to pages beyond the file size will cause a SIGBUS.
423 For example, if mmaping a file of 100 bytes on a host with 4K pages
424 emulating a target with 8K pages, the target expects to be able to
425 access the first 8K. But the host will trap us on any access beyond
428 When emulating a target with a larger page-size than the hosts, we
429 may need to truncate file maps at EOF and add extra anonymous pages
430 up to the targets page boundary. */
432 if ((qemu_real_host_page_size < TARGET_PAGE_SIZE)
433 && !(flags & MAP_ANONYMOUS)) {
436 if (fstat (fd, &sb) == -1)
439 /* Are we trying to create a map beyond EOF?. */
440 if (offset + len > sb.st_size) {
441 /* If so, truncate the file map at eof aligned with
442 the hosts real pagesize. Additional anonymous maps
443 will be created beyond EOF. */
444 len = REAL_HOST_PAGE_ALIGN(sb.st_size - offset);
448 if (!(flags & MAP_FIXED)) {
449 unsigned long host_start;
452 host_len = len + offset - host_offset;
453 host_len = HOST_PAGE_ALIGN(host_len);
455 /* Note: we prefer to control the mapping address. It is
456 especially important if qemu_host_page_size >
457 qemu_real_host_page_size */
458 p = mmap(g2h(start), host_len, prot,
459 flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
462 /* update start so that it points to the file position at 'offset' */
463 host_start = (unsigned long)p;
464 if (!(flags & MAP_ANONYMOUS)) {
465 p = mmap(g2h(start), len, prot,
466 flags | MAP_FIXED, fd, host_offset);
467 if (p == MAP_FAILED) {
468 munmap(g2h(start), host_len);
471 host_start += offset - host_offset;
473 start = h2g(host_start);
475 if (start & ~TARGET_PAGE_MASK) {
480 real_end = HOST_PAGE_ALIGN(end);
483 * Test if requested memory area fits target address space
484 * It can fail only on 64-bit host with 32-bit target.
485 * On any other target/host host mmap() handles this error correctly.
487 if ((unsigned long)start + len - 1 > (abi_ulong) -1) {
492 /* worst case: we cannot map the file because the offset is not
493 aligned, so we read it */
494 if (!(flags & MAP_ANONYMOUS) &&
495 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
496 /* msync() won't work here, so we return an error if write is
497 possible while it is a shared mapping */
498 if ((flags & MAP_TYPE) == MAP_SHARED &&
499 (prot & PROT_WRITE)) {
503 retaddr = target_mmap(start, len, prot | PROT_WRITE,
504 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
508 if (pread(fd, g2h(start), len, offset) == -1)
510 if (!(prot & PROT_WRITE)) {
511 ret = target_mprotect(start, len, prot);
517 /* handle the start of the mapping */
518 if (start > real_start) {
519 if (real_end == real_start + qemu_host_page_size) {
520 /* one single host page */
521 ret = mmap_frag(real_start, start, end,
522 prot, flags, fd, offset);
527 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
528 prot, flags, fd, offset);
531 real_start += qemu_host_page_size;
533 /* handle the end of the mapping */
534 if (end < real_end) {
535 ret = mmap_frag(real_end - qemu_host_page_size,
536 real_end - qemu_host_page_size, end,
538 offset + real_end - qemu_host_page_size - start);
541 real_end -= qemu_host_page_size;
544 /* map the middle (easier) */
545 if (real_start < real_end) {
547 unsigned long offset1;
548 if (flags & MAP_ANONYMOUS)
551 offset1 = offset + real_start - start;
552 p = mmap(g2h(real_start), real_end - real_start,
553 prot, flags, fd, offset1);
559 page_set_flags(start, start + len, prot | PAGE_VALID);
562 printf("ret=0x" TARGET_ABI_FMT_lx "\n", start);
566 tb_invalidate_phys_range(start, start + len);
574 static void mmap_reserve(abi_ulong start, abi_ulong size)
576 abi_ulong real_start;
582 real_start = start & qemu_host_page_mask;
583 real_end = HOST_PAGE_ALIGN(start + size);
585 if (start > real_start) {
586 /* handle host page containing start */
588 for (addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
589 prot |= page_get_flags(addr);
591 if (real_end == real_start + qemu_host_page_size) {
592 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
593 prot |= page_get_flags(addr);
598 real_start += qemu_host_page_size;
600 if (end < real_end) {
602 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
603 prot |= page_get_flags(addr);
606 real_end -= qemu_host_page_size;
608 if (real_start != real_end) {
609 mmap(g2h(real_start), real_end - real_start, PROT_NONE,
610 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE,
615 int target_munmap(abi_ulong start, abi_ulong len)
617 abi_ulong end, real_start, real_end, addr;
621 printf("munmap: start=0x" TARGET_ABI_FMT_lx " len=0x"
622 TARGET_ABI_FMT_lx "\n",
625 if (start & ~TARGET_PAGE_MASK)
627 len = TARGET_PAGE_ALIGN(len);
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,
686 if (flags & MREMAP_FIXED) {
687 host_addr = mremap(g2h(old_addr), old_size, new_size,
688 flags, g2h(new_addr));
690 if (reserved_va && host_addr != MAP_FAILED) {
691 /* If new and old addresses overlap then the above mremap will
692 already have failed with EINVAL. */
693 mmap_reserve(old_addr, old_size);
695 } else if (flags & MREMAP_MAYMOVE) {
696 abi_ulong mmap_start;
698 mmap_start = mmap_find_vma(0, new_size);
700 if (mmap_start == -1) {
702 host_addr = MAP_FAILED;
704 host_addr = mremap(g2h(old_addr), old_size, new_size,
705 flags | MREMAP_FIXED, g2h(mmap_start));
707 mmap_reserve(old_addr, old_size);
712 if (reserved_va && old_size < new_size) {
714 for (addr = old_addr + old_size;
715 addr < old_addr + new_size;
717 prot |= page_get_flags(addr);
721 host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
722 if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
723 mmap_reserve(old_addr + old_size, new_size - old_size);
727 host_addr = MAP_FAILED;
729 /* Check if address fits target address space */
730 if ((unsigned long)host_addr + new_size > (abi_ulong)-1) {
731 /* Revert mremap() changes */
732 host_addr = mremap(g2h(old_addr), new_size, old_size, flags);
734 host_addr = MAP_FAILED;
738 if (host_addr == MAP_FAILED) {
741 new_addr = h2g(host_addr);
742 prot = page_get_flags(old_addr);
743 page_set_flags(old_addr, old_addr + old_size, 0);
744 page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
746 tb_invalidate_phys_range(new_addr, new_addr + new_size);
751 int target_msync(abi_ulong start, abi_ulong len, int flags)
755 if (start & ~TARGET_PAGE_MASK)
757 len = TARGET_PAGE_ALIGN(len);
764 start &= qemu_host_page_mask;
765 return msync(g2h(start), end - start, flags);