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"
33 #include "translate-all.h"
37 static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER;
38 static __thread int mmap_lock_count;
42 if (mmap_lock_count++ == 0) {
43 pthread_mutex_lock(&mmap_mutex);
47 void mmap_unlock(void)
49 if (--mmap_lock_count == 0) {
50 pthread_mutex_unlock(&mmap_mutex);
54 /* Grab lock to make sure things are in a consistent state after fork(). */
55 void mmap_fork_start(void)
59 pthread_mutex_lock(&mmap_mutex);
62 void mmap_fork_end(int child)
65 pthread_mutex_init(&mmap_mutex, NULL);
67 pthread_mutex_unlock(&mmap_mutex);
70 /* NOTE: all the constants are the HOST ones, but addresses are target. */
71 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
73 abi_ulong end, host_start, host_end, addr;
77 printf("mprotect: start=0x" TARGET_ABI_FMT_lx
78 "len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c\n", start, len,
79 prot & PROT_READ ? 'r' : '-',
80 prot & PROT_WRITE ? 'w' : '-',
81 prot & PROT_EXEC ? 'x' : '-');
84 if ((start & ~TARGET_PAGE_MASK) != 0)
86 len = TARGET_PAGE_ALIGN(len);
90 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
95 host_start = start & qemu_host_page_mask;
96 host_end = HOST_PAGE_ALIGN(end);
97 if (start > host_start) {
98 /* handle host page containing start */
100 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
101 prot1 |= page_get_flags(addr);
103 if (host_end == host_start + qemu_host_page_size) {
104 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
105 prot1 |= page_get_flags(addr);
109 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
112 host_start += qemu_host_page_size;
114 if (end < host_end) {
116 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
117 prot1 |= page_get_flags(addr);
119 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
123 host_end -= qemu_host_page_size;
126 /* handle the pages in the middle */
127 if (host_start < host_end) {
128 ret = mprotect(g2h(host_start), host_end - host_start, prot);
132 page_set_flags(start, start + len, prot | PAGE_VALID);
140 /* map an incomplete host page */
141 static int mmap_frag(abi_ulong real_start,
142 abi_ulong start, abi_ulong end,
143 int prot, int flags, int fd, abi_ulong offset)
145 abi_ulong real_end, addr;
149 real_end = real_start + qemu_host_page_size;
150 host_start = g2h(real_start);
152 /* get the protection of the target pages outside the mapping */
154 for(addr = real_start; addr < real_end; addr++) {
155 if (addr < start || addr >= end)
156 prot1 |= page_get_flags(addr);
160 /* no page was there, so we allocate one */
161 void *p = mmap(host_start, qemu_host_page_size, prot,
162 flags | MAP_ANONYMOUS, -1, 0);
169 prot_new = prot | prot1;
170 if (!(flags & MAP_ANONYMOUS)) {
171 /* msync() won't work here, so we return an error if write is
172 possible while it is a shared mapping */
173 if ((flags & MAP_TYPE) == MAP_SHARED &&
177 /* adjust protection to be able to read */
178 if (!(prot1 & PROT_WRITE))
179 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
181 /* read the corresponding file data */
182 if (pread(fd, g2h(start), end - start, offset) == -1)
185 /* put final protection */
186 if (prot_new != (prot1 | PROT_WRITE))
187 mprotect(host_start, qemu_host_page_size, prot_new);
189 /* just update the protection */
190 if (prot_new != prot1) {
191 mprotect(host_start, qemu_host_page_size, prot_new);
197 #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64
198 # define TASK_UNMAPPED_BASE (1ul << 38)
199 #elif defined(__CYGWIN__)
200 /* Cygwin doesn't have a whole lot of address space. */
201 # define TASK_UNMAPPED_BASE 0x18000000
203 # define TASK_UNMAPPED_BASE 0x40000000
205 abi_ulong mmap_next_start = TASK_UNMAPPED_BASE;
207 unsigned long last_brk;
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;
257 * Find and reserve a free memory area of size 'size'. The search
259 * It must be called with mmap_lock() held.
260 * Return -1 if error.
262 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
268 /* If 'start' == 0, then a default start address is used. */
270 start = mmap_next_start;
272 start &= qemu_host_page_mask;
275 size = HOST_PAGE_ALIGN(size);
278 return mmap_find_vma_reserved(start, size);
282 wrapped = repeat = 0;
285 for (;; prev = ptr) {
287 * Reserve needed memory area to avoid a race.
288 * It should be discarded using:
289 * - mmap() with MAP_FIXED flag
290 * - mremap() with MREMAP_FIXED flag
291 * - shmat() with SHM_REMAP flag
293 ptr = mmap(g2h(addr), size, PROT_NONE,
294 MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
296 /* ENOMEM, if host address space has no memory */
297 if (ptr == MAP_FAILED) {
298 return (abi_ulong)-1;
301 /* Count the number of sequential returns of the same address.
302 This is used to modify the search algorithm below. */
303 repeat = (ptr == prev ? repeat + 1 : 0);
305 if (h2g_valid(ptr + size - 1)) {
308 if ((addr & ~TARGET_PAGE_MASK) == 0) {
310 if (start == mmap_next_start && addr >= TASK_UNMAPPED_BASE) {
311 mmap_next_start = addr + size;
316 /* The address is not properly aligned for the target. */
319 /* Assume the result that the kernel gave us is the
320 first with enough free space, so start again at the
321 next higher target page. */
322 addr = TARGET_PAGE_ALIGN(addr);
325 /* Sometimes the kernel decides to perform the allocation
326 at the top end of memory instead. */
327 addr &= TARGET_PAGE_MASK;
330 /* Start over at low memory. */
334 /* Fail. This unaligned block must the last. */
339 /* Since the result the kernel gave didn't fit, start
340 again at low memory. If any repetition, fail. */
341 addr = (repeat ? -1 : 0);
344 /* Unmap and try again. */
347 /* ENOMEM if we checked the whole of the target address space. */
348 if (addr == (abi_ulong)-1) {
349 return (abi_ulong)-1;
350 } else if (addr == 0) {
352 return (abi_ulong)-1;
355 /* Don't actually use 0 when wrapping, instead indicate
356 that we'd truly like an allocation in low memory. */
357 addr = (mmap_min_addr > TARGET_PAGE_SIZE
358 ? TARGET_PAGE_ALIGN(mmap_min_addr)
360 } else if (wrapped && addr >= start) {
361 return (abi_ulong)-1;
366 /* NOTE: all the constants are the HOST ones */
367 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
368 int flags, int fd, abi_ulong offset)
370 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
375 printf("mmap: start=0x" TARGET_ABI_FMT_lx
376 " len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c flags=",
378 prot & PROT_READ ? 'r' : '-',
379 prot & PROT_WRITE ? 'w' : '-',
380 prot & PROT_EXEC ? 'x' : '-');
381 if (flags & MAP_FIXED)
382 printf("MAP_FIXED ");
383 if (flags & MAP_ANONYMOUS)
385 switch(flags & MAP_TYPE) {
387 printf("MAP_PRIVATE ");
390 printf("MAP_SHARED ");
393 printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
396 printf("fd=%d offset=" TARGET_ABI_FMT_lx "\n", fd, offset);
400 if (offset & ~TARGET_PAGE_MASK) {
405 len = TARGET_PAGE_ALIGN(len);
408 real_start = start & qemu_host_page_mask;
409 host_offset = offset & qemu_host_page_mask;
411 /* If the user is asking for the kernel to find a location, do that
412 before we truncate the length for mapping files below. */
413 if (!(flags & MAP_FIXED)) {
414 host_len = len + offset - host_offset;
415 host_len = HOST_PAGE_ALIGN(host_len);
416 start = mmap_find_vma(real_start, host_len);
417 if (start == (abi_ulong)-1) {
423 /* When mapping files into a memory area larger than the file, accesses
424 to pages beyond the file size will cause a SIGBUS.
426 For example, if mmaping a file of 100 bytes on a host with 4K pages
427 emulating a target with 8K pages, the target expects to be able to
428 access the first 8K. But the host will trap us on any access beyond
431 When emulating a target with a larger page-size than the hosts, we
432 may need to truncate file maps at EOF and add extra anonymous pages
433 up to the targets page boundary. */
435 if ((qemu_real_host_page_size < TARGET_PAGE_SIZE)
436 && !(flags & MAP_ANONYMOUS)) {
439 if (fstat (fd, &sb) == -1)
442 /* Are we trying to create a map beyond EOF?. */
443 if (offset + len > sb.st_size) {
444 /* If so, truncate the file map at eof aligned with
445 the hosts real pagesize. Additional anonymous maps
446 will be created beyond EOF. */
447 len = (sb.st_size - offset);
448 len += qemu_real_host_page_size - 1;
449 len &= ~(qemu_real_host_page_size - 1);
453 if (!(flags & MAP_FIXED)) {
454 unsigned long host_start;
457 host_len = len + offset - host_offset;
458 host_len = HOST_PAGE_ALIGN(host_len);
460 /* Note: we prefer to control the mapping address. It is
461 especially important if qemu_host_page_size >
462 qemu_real_host_page_size */
463 p = mmap(g2h(start), host_len, prot,
464 flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
467 /* update start so that it points to the file position at 'offset' */
468 host_start = (unsigned long)p;
469 if (!(flags & MAP_ANONYMOUS)) {
470 p = mmap(g2h(start), len, prot,
471 flags | MAP_FIXED, fd, host_offset);
472 if (p == MAP_FAILED) {
473 munmap(g2h(start), host_len);
476 host_start += offset - host_offset;
478 start = h2g(host_start);
480 if (start & ~TARGET_PAGE_MASK) {
485 real_end = HOST_PAGE_ALIGN(end);
488 * Test if requested memory area fits target address space
489 * It can fail only on 64-bit host with 32-bit target.
490 * On any other target/host host mmap() handles this error correctly.
492 if ((unsigned long)start + len - 1 > (abi_ulong) -1) {
497 /* worst case: we cannot map the file because the offset is not
498 aligned, so we read it */
499 if (!(flags & MAP_ANONYMOUS) &&
500 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
501 /* msync() won't work here, so we return an error if write is
502 possible while it is a shared mapping */
503 if ((flags & MAP_TYPE) == MAP_SHARED &&
504 (prot & PROT_WRITE)) {
508 retaddr = target_mmap(start, len, prot | PROT_WRITE,
509 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
513 if (pread(fd, g2h(start), len, offset) == -1)
515 if (!(prot & PROT_WRITE)) {
516 ret = target_mprotect(start, len, prot);
525 /* handle the start of the mapping */
526 if (start > real_start) {
527 if (real_end == real_start + qemu_host_page_size) {
528 /* one single host page */
529 ret = mmap_frag(real_start, start, end,
530 prot, flags, fd, offset);
535 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
536 prot, flags, fd, offset);
539 real_start += qemu_host_page_size;
541 /* handle the end of the mapping */
542 if (end < real_end) {
543 ret = mmap_frag(real_end - qemu_host_page_size,
544 real_end - qemu_host_page_size, real_end,
546 offset + real_end - qemu_host_page_size - start);
549 real_end -= qemu_host_page_size;
552 /* map the middle (easier) */
553 if (real_start < real_end) {
555 unsigned long offset1;
556 if (flags & MAP_ANONYMOUS)
559 offset1 = offset + real_start - start;
560 p = mmap(g2h(real_start), real_end - real_start,
561 prot, flags, fd, offset1);
567 page_set_flags(start, start + len, prot | PAGE_VALID);
570 printf("ret=0x" TARGET_ABI_FMT_lx "\n", start);
574 tb_invalidate_phys_range(start, start + len);
582 static void mmap_reserve(abi_ulong start, abi_ulong size)
584 abi_ulong real_start;
590 real_start = start & qemu_host_page_mask;
591 real_end = HOST_PAGE_ALIGN(start + size);
593 if (start > real_start) {
594 /* handle host page containing start */
596 for (addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
597 prot |= page_get_flags(addr);
599 if (real_end == real_start + qemu_host_page_size) {
600 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
601 prot |= page_get_flags(addr);
606 real_start += qemu_host_page_size;
608 if (end < real_end) {
610 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
611 prot |= page_get_flags(addr);
614 real_end -= qemu_host_page_size;
616 if (real_start != real_end) {
617 mmap(g2h(real_start), real_end - real_start, PROT_NONE,
618 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE,
623 int target_munmap(abi_ulong start, abi_ulong len)
625 abi_ulong end, real_start, real_end, addr;
629 printf("munmap: start=0x" TARGET_ABI_FMT_lx " len=0x"
630 TARGET_ABI_FMT_lx "\n",
633 if (start & ~TARGET_PAGE_MASK)
635 len = TARGET_PAGE_ALIGN(len);
640 real_start = start & qemu_host_page_mask;
641 real_end = HOST_PAGE_ALIGN(end);
643 if (start > real_start) {
644 /* handle host page containing start */
646 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
647 prot |= page_get_flags(addr);
649 if (real_end == real_start + qemu_host_page_size) {
650 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
651 prot |= page_get_flags(addr);
656 real_start += qemu_host_page_size;
658 if (end < real_end) {
660 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
661 prot |= page_get_flags(addr);
664 real_end -= qemu_host_page_size;
668 /* unmap what we can */
669 if (real_start < real_end) {
671 mmap_reserve(real_start, real_end - real_start);
673 ret = munmap(g2h(real_start), real_end - real_start);
678 page_set_flags(start, start + len, 0);
679 tb_invalidate_phys_range(start, start + len);
685 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
686 abi_ulong new_size, unsigned long flags,
694 if (flags & MREMAP_FIXED) {
695 host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
700 if (reserved_va && host_addr != MAP_FAILED) {
701 /* If new and old addresses overlap then the above mremap will
702 already have failed with EINVAL. */
703 mmap_reserve(old_addr, old_size);
705 } else if (flags & MREMAP_MAYMOVE) {
706 abi_ulong mmap_start;
708 mmap_start = mmap_find_vma(0, new_size);
710 if (mmap_start == -1) {
712 host_addr = MAP_FAILED;
714 host_addr = (void *) syscall(__NR_mremap, g2h(old_addr),
716 flags | MREMAP_FIXED,
719 mmap_reserve(old_addr, old_size);
724 if (reserved_va && old_size < new_size) {
726 for (addr = old_addr + old_size;
727 addr < old_addr + new_size;
729 prot |= page_get_flags(addr);
733 host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
734 if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
735 mmap_reserve(old_addr + old_size, new_size - old_size);
739 host_addr = MAP_FAILED;
741 /* Check if address fits target address space */
742 if ((unsigned long)host_addr + new_size > (abi_ulong)-1) {
743 /* Revert mremap() changes */
744 host_addr = mremap(g2h(old_addr), new_size, old_size, flags);
746 host_addr = MAP_FAILED;
750 if (host_addr == MAP_FAILED) {
753 new_addr = h2g(host_addr);
754 prot = page_get_flags(old_addr);
755 page_set_flags(old_addr, old_addr + old_size, 0);
756 page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
758 tb_invalidate_phys_range(new_addr, new_addr + new_size);
763 int target_msync(abi_ulong start, abi_ulong len, int flags)
767 if (start & ~TARGET_PAGE_MASK)
769 len = TARGET_PAGE_ALIGN(len);
776 start &= qemu_host_page_mask;
777 return msync(g2h(start), end - start, flags);