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 /* Grab lock to make sure things are in a consistent state after fork(). */
45 void mmap_fork_start(void)
49 pthread_mutex_lock(&mmap_mutex);
52 void mmap_fork_end(int child)
55 pthread_mutex_init(&mmap_mutex, NULL);
57 pthread_mutex_unlock(&mmap_mutex);
60 /* NOTE: all the constants are the HOST ones, but addresses are target. */
61 int target_mprotect(abi_ulong start, abi_ulong len, int prot)
63 abi_ulong end, host_start, host_end, addr;
67 printf("mprotect: start=0x" TARGET_ABI_FMT_lx
68 "len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c\n", start, len,
69 prot & PROT_READ ? 'r' : '-',
70 prot & PROT_WRITE ? 'w' : '-',
71 prot & PROT_EXEC ? 'x' : '-');
74 if ((start & ~TARGET_PAGE_MASK) != 0)
76 len = TARGET_PAGE_ALIGN(len);
80 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
85 host_start = start & qemu_host_page_mask;
86 host_end = HOST_PAGE_ALIGN(end);
87 if (start > host_start) {
88 /* handle host page containing start */
90 for(addr = host_start; addr < start; addr += TARGET_PAGE_SIZE) {
91 prot1 |= page_get_flags(addr);
93 if (host_end == host_start + qemu_host_page_size) {
94 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
95 prot1 |= page_get_flags(addr);
99 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
102 host_start += qemu_host_page_size;
104 if (end < host_end) {
106 for(addr = end; addr < host_end; addr += TARGET_PAGE_SIZE) {
107 prot1 |= page_get_flags(addr);
109 ret = mprotect(g2h(host_end - qemu_host_page_size), qemu_host_page_size,
113 host_end -= qemu_host_page_size;
116 /* handle the pages in the middle */
117 if (host_start < host_end) {
118 ret = mprotect(g2h(host_start), host_end - host_start, prot);
122 page_set_flags(start, start + len, prot | PAGE_VALID);
130 /* map an incomplete host page */
131 static int mmap_frag(abi_ulong real_start,
132 abi_ulong start, abi_ulong end,
133 int prot, int flags, int fd, abi_ulong offset)
135 abi_ulong real_end, addr;
139 real_end = real_start + qemu_host_page_size;
140 host_start = g2h(real_start);
142 /* get the protection of the target pages outside the mapping */
144 for(addr = real_start; addr < real_end; addr++) {
145 if (addr < start || addr >= end)
146 prot1 |= page_get_flags(addr);
150 /* no page was there, so we allocate one */
151 void *p = mmap(host_start, qemu_host_page_size, prot,
152 flags | MAP_ANONYMOUS, -1, 0);
159 prot_new = prot | prot1;
160 if (!(flags & MAP_ANONYMOUS)) {
161 /* msync() won't work here, so we return an error if write is
162 possible while it is a shared mapping */
163 if ((flags & MAP_TYPE) == MAP_SHARED &&
167 /* adjust protection to be able to read */
168 if (!(prot1 & PROT_WRITE))
169 mprotect(host_start, qemu_host_page_size, prot1 | PROT_WRITE);
171 /* read the corresponding file data */
172 if (pread(fd, g2h(start), end - start, offset) == -1)
175 /* put final protection */
176 if (prot_new != (prot1 | PROT_WRITE))
177 mprotect(host_start, qemu_host_page_size, prot_new);
179 if (prot_new != prot1) {
180 mprotect(host_start, qemu_host_page_size, prot_new);
182 if (prot_new & PROT_WRITE) {
183 memset(g2h(start), 0, end - start);
189 #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64
190 # define TASK_UNMAPPED_BASE (1ul << 38)
191 #elif defined(__CYGWIN__)
192 /* Cygwin doesn't have a whole lot of address space. */
193 # define TASK_UNMAPPED_BASE 0x18000000
195 # define TASK_UNMAPPED_BASE 0x40000000
197 abi_ulong mmap_next_start = TASK_UNMAPPED_BASE;
199 unsigned long last_brk;
201 /* Subroutine of mmap_find_vma, used when we have pre-allocated a chunk
202 of guest address space. */
203 static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size)
210 if (size > reserved_va) {
211 return (abi_ulong)-1;
214 size = HOST_PAGE_ALIGN(size);
215 end_addr = start + size;
216 if (end_addr > reserved_va) {
217 end_addr = reserved_va;
219 addr = end_addr - qemu_host_page_size;
222 if (addr > end_addr) {
224 return (abi_ulong)-1;
226 end_addr = reserved_va;
227 addr = end_addr - qemu_host_page_size;
231 prot = page_get_flags(addr);
235 if (addr + size == end_addr) {
238 addr -= qemu_host_page_size;
241 if (start == mmap_next_start) {
242 mmap_next_start = addr;
249 * Find and reserve a free memory area of size 'size'. The search
251 * It must be called with mmap_lock() held.
252 * Return -1 if error.
254 abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
260 /* If 'start' == 0, then a default start address is used. */
262 start = mmap_next_start;
264 start &= qemu_host_page_mask;
267 size = HOST_PAGE_ALIGN(size);
270 return mmap_find_vma_reserved(start, size);
274 wrapped = repeat = 0;
277 for (;; prev = ptr) {
279 * Reserve needed memory area to avoid a race.
280 * It should be discarded using:
281 * - mmap() with MAP_FIXED flag
282 * - mremap() with MREMAP_FIXED flag
283 * - shmat() with SHM_REMAP flag
285 ptr = mmap(g2h(addr), size, PROT_NONE,
286 MAP_ANONYMOUS|MAP_PRIVATE|MAP_NORESERVE, -1, 0);
288 /* ENOMEM, if host address space has no memory */
289 if (ptr == MAP_FAILED) {
290 return (abi_ulong)-1;
293 /* Count the number of sequential returns of the same address.
294 This is used to modify the search algorithm below. */
295 repeat = (ptr == prev ? repeat + 1 : 0);
297 if (h2g_valid(ptr + size - 1)) {
300 if ((addr & ~TARGET_PAGE_MASK) == 0) {
302 if (start == mmap_next_start && addr >= TASK_UNMAPPED_BASE) {
303 mmap_next_start = addr + size;
308 /* The address is not properly aligned for the target. */
311 /* Assume the result that the kernel gave us is the
312 first with enough free space, so start again at the
313 next higher target page. */
314 addr = TARGET_PAGE_ALIGN(addr);
317 /* Sometimes the kernel decides to perform the allocation
318 at the top end of memory instead. */
319 addr &= TARGET_PAGE_MASK;
322 /* Start over at low memory. */
326 /* Fail. This unaligned block must the last. */
331 /* Since the result the kernel gave didn't fit, start
332 again at low memory. If any repetition, fail. */
333 addr = (repeat ? -1 : 0);
336 /* Unmap and try again. */
339 /* ENOMEM if we checked the whole of the target address space. */
340 if (addr == (abi_ulong)-1) {
341 return (abi_ulong)-1;
342 } else if (addr == 0) {
344 return (abi_ulong)-1;
347 /* Don't actually use 0 when wrapping, instead indicate
348 that we'd truly like an allocation in low memory. */
349 addr = (mmap_min_addr > TARGET_PAGE_SIZE
350 ? TARGET_PAGE_ALIGN(mmap_min_addr)
352 } else if (wrapped && addr >= start) {
353 return (abi_ulong)-1;
358 /* NOTE: all the constants are the HOST ones */
359 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
360 int flags, int fd, abi_ulong offset)
362 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
367 printf("mmap: start=0x" TARGET_ABI_FMT_lx
368 " len=0x" TARGET_ABI_FMT_lx " prot=%c%c%c flags=",
370 prot & PROT_READ ? 'r' : '-',
371 prot & PROT_WRITE ? 'w' : '-',
372 prot & PROT_EXEC ? 'x' : '-');
373 if (flags & MAP_FIXED)
374 printf("MAP_FIXED ");
375 if (flags & MAP_ANONYMOUS)
377 switch(flags & MAP_TYPE) {
379 printf("MAP_PRIVATE ");
382 printf("MAP_SHARED ");
385 printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
388 printf("fd=%d offset=" TARGET_ABI_FMT_lx "\n", fd, offset);
392 if (offset & ~TARGET_PAGE_MASK) {
397 len = TARGET_PAGE_ALIGN(len);
400 real_start = start & qemu_host_page_mask;
401 host_offset = offset & qemu_host_page_mask;
403 /* If the user is asking for the kernel to find a location, do that
404 before we truncate the length for mapping files below. */
405 if (!(flags & MAP_FIXED)) {
406 host_len = len + offset - host_offset;
407 host_len = HOST_PAGE_ALIGN(host_len);
408 start = mmap_find_vma(real_start, host_len);
409 if (start == (abi_ulong)-1) {
415 /* When mapping files into a memory area larger than the file, accesses
416 to pages beyond the file size will cause a SIGBUS.
418 For example, if mmaping a file of 100 bytes on a host with 4K pages
419 emulating a target with 8K pages, the target expects to be able to
420 access the first 8K. But the host will trap us on any access beyond
423 When emulating a target with a larger page-size than the hosts, we
424 may need to truncate file maps at EOF and add extra anonymous pages
425 up to the targets page boundary. */
427 if ((qemu_real_host_page_size < TARGET_PAGE_SIZE)
428 && !(flags & MAP_ANONYMOUS)) {
431 if (fstat (fd, &sb) == -1)
434 /* Are we trying to create a map beyond EOF?. */
435 if (offset + len > sb.st_size) {
436 /* If so, truncate the file map at eof aligned with
437 the hosts real pagesize. Additional anonymous maps
438 will be created beyond EOF. */
439 len = REAL_HOST_PAGE_ALIGN(sb.st_size - offset);
443 if (!(flags & MAP_FIXED)) {
444 unsigned long host_start;
447 host_len = len + offset - host_offset;
448 host_len = HOST_PAGE_ALIGN(host_len);
450 /* Note: we prefer to control the mapping address. It is
451 especially important if qemu_host_page_size >
452 qemu_real_host_page_size */
453 p = mmap(g2h(start), host_len, prot,
454 flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0);
457 /* update start so that it points to the file position at 'offset' */
458 host_start = (unsigned long)p;
459 if (!(flags & MAP_ANONYMOUS)) {
460 p = mmap(g2h(start), len, prot,
461 flags | MAP_FIXED, fd, host_offset);
462 if (p == MAP_FAILED) {
463 munmap(g2h(start), host_len);
466 host_start += offset - host_offset;
468 start = h2g(host_start);
470 if (start & ~TARGET_PAGE_MASK) {
475 real_end = HOST_PAGE_ALIGN(end);
478 * Test if requested memory area fits target address space
479 * It can fail only on 64-bit host with 32-bit target.
480 * On any other target/host host mmap() handles this error correctly.
482 if ((unsigned long)start + len - 1 > (abi_ulong) -1) {
487 /* worst case: we cannot map the file because the offset is not
488 aligned, so we read it */
489 if (!(flags & MAP_ANONYMOUS) &&
490 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
491 /* msync() won't work here, so we return an error if write is
492 possible while it is a shared mapping */
493 if ((flags & MAP_TYPE) == MAP_SHARED &&
494 (prot & PROT_WRITE)) {
498 retaddr = target_mmap(start, len, prot | PROT_WRITE,
499 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
503 if (pread(fd, g2h(start), len, offset) == -1)
505 if (!(prot & PROT_WRITE)) {
506 ret = target_mprotect(start, len, prot);
512 /* handle the start of the mapping */
513 if (start > real_start) {
514 if (real_end == real_start + qemu_host_page_size) {
515 /* one single host page */
516 ret = mmap_frag(real_start, start, end,
517 prot, flags, fd, offset);
522 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
523 prot, flags, fd, offset);
526 real_start += qemu_host_page_size;
528 /* handle the end of the mapping */
529 if (end < real_end) {
530 ret = mmap_frag(real_end - qemu_host_page_size,
531 real_end - qemu_host_page_size, end,
533 offset + real_end - qemu_host_page_size - start);
536 real_end -= qemu_host_page_size;
539 /* map the middle (easier) */
540 if (real_start < real_end) {
542 unsigned long offset1;
543 if (flags & MAP_ANONYMOUS)
546 offset1 = offset + real_start - start;
547 p = mmap(g2h(real_start), real_end - real_start,
548 prot, flags, fd, offset1);
554 page_set_flags(start, start + len, prot | PAGE_VALID);
557 printf("ret=0x" TARGET_ABI_FMT_lx "\n", start);
561 tb_invalidate_phys_range(start, start + len);
569 static void mmap_reserve(abi_ulong start, abi_ulong size)
571 abi_ulong real_start;
577 real_start = start & qemu_host_page_mask;
578 real_end = HOST_PAGE_ALIGN(start + size);
580 if (start > real_start) {
581 /* handle host page containing start */
583 for (addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
584 prot |= page_get_flags(addr);
586 if (real_end == real_start + qemu_host_page_size) {
587 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
588 prot |= page_get_flags(addr);
593 real_start += qemu_host_page_size;
595 if (end < real_end) {
597 for (addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
598 prot |= page_get_flags(addr);
601 real_end -= qemu_host_page_size;
603 if (real_start != real_end) {
604 mmap(g2h(real_start), real_end - real_start, PROT_NONE,
605 MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE,
610 int target_munmap(abi_ulong start, abi_ulong len)
612 abi_ulong end, real_start, real_end, addr;
616 printf("munmap: start=0x" TARGET_ABI_FMT_lx " len=0x"
617 TARGET_ABI_FMT_lx "\n",
620 if (start & ~TARGET_PAGE_MASK)
622 len = TARGET_PAGE_ALIGN(len);
627 real_start = start & qemu_host_page_mask;
628 real_end = HOST_PAGE_ALIGN(end);
630 if (start > real_start) {
631 /* handle host page containing start */
633 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
634 prot |= page_get_flags(addr);
636 if (real_end == real_start + qemu_host_page_size) {
637 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
638 prot |= page_get_flags(addr);
643 real_start += qemu_host_page_size;
645 if (end < real_end) {
647 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
648 prot |= page_get_flags(addr);
651 real_end -= qemu_host_page_size;
655 /* unmap what we can */
656 if (real_start < real_end) {
658 mmap_reserve(real_start, real_end - real_start);
660 ret = munmap(g2h(real_start), real_end - real_start);
665 page_set_flags(start, start + len, 0);
666 tb_invalidate_phys_range(start, start + len);
672 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
673 abi_ulong new_size, unsigned long flags,
681 if (flags & MREMAP_FIXED) {
682 host_addr = mremap(g2h(old_addr), old_size, new_size,
683 flags, g2h(new_addr));
685 if (reserved_va && host_addr != MAP_FAILED) {
686 /* If new and old addresses overlap then the above mremap will
687 already have failed with EINVAL. */
688 mmap_reserve(old_addr, old_size);
690 } else if (flags & MREMAP_MAYMOVE) {
691 abi_ulong mmap_start;
693 mmap_start = mmap_find_vma(0, new_size);
695 if (mmap_start == -1) {
697 host_addr = MAP_FAILED;
699 host_addr = mremap(g2h(old_addr), old_size, new_size,
700 flags | MREMAP_FIXED, g2h(mmap_start));
702 mmap_reserve(old_addr, old_size);
707 if (reserved_va && old_size < new_size) {
709 for (addr = old_addr + old_size;
710 addr < old_addr + new_size;
712 prot |= page_get_flags(addr);
716 host_addr = mremap(g2h(old_addr), old_size, new_size, flags);
717 if (host_addr != MAP_FAILED && reserved_va && old_size > new_size) {
718 mmap_reserve(old_addr + old_size, new_size - old_size);
722 host_addr = MAP_FAILED;
724 /* Check if address fits target address space */
725 if ((unsigned long)host_addr + new_size > (abi_ulong)-1) {
726 /* Revert mremap() changes */
727 host_addr = mremap(g2h(old_addr), new_size, old_size, flags);
729 host_addr = MAP_FAILED;
733 if (host_addr == MAP_FAILED) {
736 new_addr = h2g(host_addr);
737 prot = page_get_flags(old_addr);
738 page_set_flags(old_addr, old_addr + old_size, 0);
739 page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
741 tb_invalidate_phys_range(new_addr, new_addr + new_size);
746 int target_msync(abi_ulong start, abi_ulong len, int flags)
750 if (start & ~TARGET_PAGE_MASK)
752 len = TARGET_PAGE_ALIGN(len);
759 start &= qemu_host_page_mask;
760 return msync(g2h(start), end - start, flags);