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
,
208 abi_ulong addr
, end_addr
, incr
= qemu_host_page_size
;
212 if (size
> reserved_va
) {
213 return (abi_ulong
)-1;
216 /* Note that start and size have already been aligned by mmap_find_vma. */
218 end_addr
= start
+ size
;
219 if (start
> reserved_va
- size
) {
220 /* Start at the top of the address space. */
221 end_addr
= ((reserved_va
- size
) & -align
) + size
;
225 /* Search downward from END_ADDR, checking to see if a page is in use. */
229 if (addr
> end_addr
) {
231 /* Failure. The entire address space has been searched. */
232 return (abi_ulong
)-1;
234 /* Re-start at the top of the address space. */
235 addr
= end_addr
= ((reserved_va
- size
) & -align
) + size
;
238 prot
= page_get_flags(addr
);
240 /* Page in use. Restart below this page. */
241 addr
= end_addr
= ((addr
- size
) & -align
) + size
;
242 } else if (addr
&& addr
+ size
== end_addr
) {
243 /* Success! All pages between ADDR and END_ADDR are free. */
244 if (start
== mmap_next_start
) {
245 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
, abi_ulong align
)
265 align
= MAX(align
, qemu_host_page_size
);
267 /* If 'start' == 0, then a default start address is used. */
269 start
= mmap_next_start
;
271 start
&= qemu_host_page_mask
;
273 start
= ROUND_UP(start
, align
);
275 size
= HOST_PAGE_ALIGN(size
);
278 return mmap_find_vma_reserved(start
, size
, align
);
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
& (align
- 1)) == 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
= ROUND_UP(addr
, align
);
325 /* Sometimes the kernel decides to perform the allocation
326 at the top end of memory instead. */
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
);
405 /* Also check for overflows... */
406 len
= TARGET_PAGE_ALIGN(len
);
412 if (offset
& ~TARGET_PAGE_MASK
) {
417 real_start
= start
& qemu_host_page_mask
;
418 host_offset
= offset
& qemu_host_page_mask
;
420 /* If the user is asking for the kernel to find a location, do that
421 before we truncate the length for mapping files below. */
422 if (!(flags
& MAP_FIXED
)) {
423 host_len
= len
+ offset
- host_offset
;
424 host_len
= HOST_PAGE_ALIGN(host_len
);
425 start
= mmap_find_vma(real_start
, host_len
, TARGET_PAGE_SIZE
);
426 if (start
== (abi_ulong
)-1) {
432 /* When mapping files into a memory area larger than the file, accesses
433 to pages beyond the file size will cause a SIGBUS.
435 For example, if mmaping a file of 100 bytes on a host with 4K pages
436 emulating a target with 8K pages, the target expects to be able to
437 access the first 8K. But the host will trap us on any access beyond
440 When emulating a target with a larger page-size than the hosts, we
441 may need to truncate file maps at EOF and add extra anonymous pages
442 up to the targets page boundary. */
444 if ((qemu_real_host_page_size
< qemu_host_page_size
) &&
445 !(flags
& MAP_ANONYMOUS
)) {
448 if (fstat (fd
, &sb
) == -1)
451 /* Are we trying to create a map beyond EOF?. */
452 if (offset
+ len
> sb
.st_size
) {
453 /* If so, truncate the file map at eof aligned with
454 the hosts real pagesize. Additional anonymous maps
455 will be created beyond EOF. */
456 len
= REAL_HOST_PAGE_ALIGN(sb
.st_size
- offset
);
460 if (!(flags
& MAP_FIXED
)) {
461 unsigned long host_start
;
464 host_len
= len
+ offset
- host_offset
;
465 host_len
= HOST_PAGE_ALIGN(host_len
);
467 /* Note: we prefer to control the mapping address. It is
468 especially important if qemu_host_page_size >
469 qemu_real_host_page_size */
470 p
= mmap(g2h(start
), host_len
, prot
,
471 flags
| MAP_FIXED
| MAP_ANONYMOUS
, -1, 0);
474 /* update start so that it points to the file position at 'offset' */
475 host_start
= (unsigned long)p
;
476 if (!(flags
& MAP_ANONYMOUS
)) {
477 p
= mmap(g2h(start
), len
, prot
,
478 flags
| MAP_FIXED
, fd
, host_offset
);
479 if (p
== MAP_FAILED
) {
480 munmap(g2h(start
), host_len
);
483 host_start
+= offset
- host_offset
;
485 start
= h2g(host_start
);
487 if (start
& ~TARGET_PAGE_MASK
) {
492 real_end
= HOST_PAGE_ALIGN(end
);
495 * Test if requested memory area fits target address space
496 * It can fail only on 64-bit host with 32-bit target.
497 * On any other target/host host mmap() handles this error correctly.
499 if (!guest_range_valid(start
, len
)) {
504 /* worst case: we cannot map the file because the offset is not
505 aligned, so we read it */
506 if (!(flags
& MAP_ANONYMOUS
) &&
507 (offset
& ~qemu_host_page_mask
) != (start
& ~qemu_host_page_mask
)) {
508 /* msync() won't work here, so we return an error if write is
509 possible while it is a shared mapping */
510 if ((flags
& MAP_TYPE
) == MAP_SHARED
&&
511 (prot
& PROT_WRITE
)) {
515 retaddr
= target_mmap(start
, len
, prot
| PROT_WRITE
,
516 MAP_FIXED
| MAP_PRIVATE
| MAP_ANONYMOUS
,
520 if (pread(fd
, g2h(start
), len
, offset
) == -1)
522 if (!(prot
& PROT_WRITE
)) {
523 ret
= target_mprotect(start
, len
, prot
);
529 /* handle the start of the mapping */
530 if (start
> real_start
) {
531 if (real_end
== real_start
+ qemu_host_page_size
) {
532 /* one single host page */
533 ret
= mmap_frag(real_start
, start
, end
,
534 prot
, flags
, fd
, offset
);
539 ret
= mmap_frag(real_start
, start
, real_start
+ qemu_host_page_size
,
540 prot
, flags
, fd
, offset
);
543 real_start
+= qemu_host_page_size
;
545 /* handle the end of the mapping */
546 if (end
< real_end
) {
547 ret
= mmap_frag(real_end
- qemu_host_page_size
,
548 real_end
- qemu_host_page_size
, end
,
550 offset
+ real_end
- qemu_host_page_size
- start
);
553 real_end
-= qemu_host_page_size
;
556 /* map the middle (easier) */
557 if (real_start
< real_end
) {
559 unsigned long offset1
;
560 if (flags
& MAP_ANONYMOUS
)
563 offset1
= offset
+ real_start
- start
;
564 p
= mmap(g2h(real_start
), real_end
- real_start
,
565 prot
, flags
, fd
, offset1
);
571 page_set_flags(start
, start
+ len
, prot
| PAGE_VALID
);
574 printf("ret=0x" TARGET_ABI_FMT_lx
"\n", start
);
578 tb_invalidate_phys_range(start
, start
+ len
);
586 static void mmap_reserve(abi_ulong start
, abi_ulong size
)
588 abi_ulong real_start
;
594 real_start
= start
& qemu_host_page_mask
;
595 real_end
= HOST_PAGE_ALIGN(start
+ size
);
597 if (start
> real_start
) {
598 /* handle host page containing start */
600 for (addr
= real_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
601 prot
|= page_get_flags(addr
);
603 if (real_end
== real_start
+ qemu_host_page_size
) {
604 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
605 prot
|= page_get_flags(addr
);
610 real_start
+= qemu_host_page_size
;
612 if (end
< real_end
) {
614 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
615 prot
|= page_get_flags(addr
);
618 real_end
-= qemu_host_page_size
;
620 if (real_start
!= real_end
) {
621 mmap(g2h(real_start
), real_end
- real_start
, PROT_NONE
,
622 MAP_FIXED
| MAP_ANONYMOUS
| MAP_PRIVATE
| MAP_NORESERVE
,
627 int target_munmap(abi_ulong start
, abi_ulong len
)
629 abi_ulong end
, real_start
, real_end
, addr
;
633 printf("munmap: start=0x" TARGET_ABI_FMT_lx
" len=0x"
634 TARGET_ABI_FMT_lx
"\n",
637 if (start
& ~TARGET_PAGE_MASK
)
638 return -TARGET_EINVAL
;
639 len
= TARGET_PAGE_ALIGN(len
);
640 if (len
== 0 || !guest_range_valid(start
, len
)) {
641 return -TARGET_EINVAL
;
646 real_start
= start
& qemu_host_page_mask
;
647 real_end
= HOST_PAGE_ALIGN(end
);
649 if (start
> real_start
) {
650 /* handle host page containing start */
652 for(addr
= real_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
653 prot
|= page_get_flags(addr
);
655 if (real_end
== real_start
+ qemu_host_page_size
) {
656 for(addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
657 prot
|= page_get_flags(addr
);
662 real_start
+= qemu_host_page_size
;
664 if (end
< real_end
) {
666 for(addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
667 prot
|= page_get_flags(addr
);
670 real_end
-= qemu_host_page_size
;
674 /* unmap what we can */
675 if (real_start
< real_end
) {
677 mmap_reserve(real_start
, real_end
- real_start
);
679 ret
= munmap(g2h(real_start
), real_end
- real_start
);
684 page_set_flags(start
, start
+ len
, 0);
685 tb_invalidate_phys_range(start
, start
+ len
);
691 abi_long
target_mremap(abi_ulong old_addr
, abi_ulong old_size
,
692 abi_ulong new_size
, unsigned long flags
,
698 if (!guest_range_valid(old_addr
, old_size
) ||
699 ((flags
& MREMAP_FIXED
) &&
700 !guest_range_valid(new_addr
, new_size
))) {
707 if (flags
& MREMAP_FIXED
) {
708 host_addr
= mremap(g2h(old_addr
), old_size
, new_size
,
709 flags
, g2h(new_addr
));
711 if (reserved_va
&& host_addr
!= MAP_FAILED
) {
712 /* If new and old addresses overlap then the above mremap will
713 already have failed with EINVAL. */
714 mmap_reserve(old_addr
, old_size
);
716 } else if (flags
& MREMAP_MAYMOVE
) {
717 abi_ulong mmap_start
;
719 mmap_start
= mmap_find_vma(0, new_size
, TARGET_PAGE_SIZE
);
721 if (mmap_start
== -1) {
723 host_addr
= MAP_FAILED
;
725 host_addr
= mremap(g2h(old_addr
), old_size
, new_size
,
726 flags
| MREMAP_FIXED
, g2h(mmap_start
));
728 mmap_reserve(old_addr
, old_size
);
733 if (reserved_va
&& old_size
< new_size
) {
735 for (addr
= old_addr
+ old_size
;
736 addr
< old_addr
+ new_size
;
738 prot
|= page_get_flags(addr
);
742 host_addr
= mremap(g2h(old_addr
), old_size
, new_size
, flags
);
743 if (host_addr
!= MAP_FAILED
&& reserved_va
&& old_size
> new_size
) {
744 mmap_reserve(old_addr
+ old_size
, new_size
- old_size
);
748 host_addr
= MAP_FAILED
;
750 /* Check if address fits target address space */
751 if ((unsigned long)host_addr
+ new_size
> (abi_ulong
)-1) {
752 /* Revert mremap() changes */
753 host_addr
= mremap(g2h(old_addr
), new_size
, old_size
, flags
);
755 host_addr
= MAP_FAILED
;
759 if (host_addr
== MAP_FAILED
) {
762 new_addr
= h2g(host_addr
);
763 prot
= page_get_flags(old_addr
);
764 page_set_flags(old_addr
, old_addr
+ old_size
, 0);
765 page_set_flags(new_addr
, new_addr
+ new_size
, prot
| PAGE_VALID
);
767 tb_invalidate_phys_range(new_addr
, new_addr
+ new_size
);