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"
36 #if defined(CONFIG_USE_NPTL)
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 /* We aren't threadsafe to start with, so no need to worry about locking. */
75 void mmap_unlock(void)
80 /* NOTE: all the constants are the HOST ones, but addresses are target. */
81 int target_mprotect(abi_ulong start
, abi_ulong len
, int prot
)
83 abi_ulong end
, host_start
, host_end
, addr
;
87 printf("mprotect: start=0x" TARGET_ABI_FMT_lx
88 "len=0x" TARGET_ABI_FMT_lx
" prot=%c%c%c\n", start
, len
,
89 prot
& PROT_READ
? 'r' : '-',
90 prot
& PROT_WRITE
? 'w' : '-',
91 prot
& PROT_EXEC
? 'x' : '-');
94 if ((start
& ~TARGET_PAGE_MASK
) != 0)
96 len
= TARGET_PAGE_ALIGN(len
);
100 prot
&= PROT_READ
| PROT_WRITE
| PROT_EXEC
;
105 host_start
= start
& qemu_host_page_mask
;
106 host_end
= HOST_PAGE_ALIGN(end
);
107 if (start
> host_start
) {
108 /* handle host page containing start */
110 for(addr
= host_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
111 prot1
|= page_get_flags(addr
);
113 if (host_end
== host_start
+ qemu_host_page_size
) {
114 for(addr
= end
; addr
< host_end
; addr
+= TARGET_PAGE_SIZE
) {
115 prot1
|= page_get_flags(addr
);
119 ret
= mprotect(g2h(host_start
), qemu_host_page_size
, prot1
& PAGE_BITS
);
122 host_start
+= qemu_host_page_size
;
124 if (end
< host_end
) {
126 for(addr
= end
; addr
< host_end
; addr
+= TARGET_PAGE_SIZE
) {
127 prot1
|= page_get_flags(addr
);
129 ret
= mprotect(g2h(host_end
- qemu_host_page_size
), qemu_host_page_size
,
133 host_end
-= qemu_host_page_size
;
136 /* handle the pages in the middle */
137 if (host_start
< host_end
) {
138 ret
= mprotect(g2h(host_start
), host_end
- host_start
, prot
);
142 page_set_flags(start
, start
+ len
, prot
| PAGE_VALID
);
150 /* map an incomplete host page */
151 static int mmap_frag(abi_ulong real_start
,
152 abi_ulong start
, abi_ulong end
,
153 int prot
, int flags
, int fd
, abi_ulong offset
)
155 abi_ulong real_end
, addr
;
159 real_end
= real_start
+ qemu_host_page_size
;
160 host_start
= g2h(real_start
);
162 /* get the protection of the target pages outside the mapping */
164 for(addr
= real_start
; addr
< real_end
; addr
++) {
165 if (addr
< start
|| addr
>= end
)
166 prot1
|= page_get_flags(addr
);
170 /* no page was there, so we allocate one */
171 void *p
= mmap(host_start
, qemu_host_page_size
, prot
,
172 flags
| MAP_ANONYMOUS
, -1, 0);
179 prot_new
= prot
| prot1
;
180 if (!(flags
& MAP_ANONYMOUS
)) {
181 /* msync() won't work here, so we return an error if write is
182 possible while it is a shared mapping */
183 if ((flags
& MAP_TYPE
) == MAP_SHARED
&&
187 /* adjust protection to be able to read */
188 if (!(prot1
& PROT_WRITE
))
189 mprotect(host_start
, qemu_host_page_size
, prot1
| PROT_WRITE
);
191 /* read the corresponding file data */
192 if (pread(fd
, g2h(start
), end
- start
, offset
) == -1)
195 /* put final protection */
196 if (prot_new
!= (prot1
| PROT_WRITE
))
197 mprotect(host_start
, qemu_host_page_size
, prot_new
);
199 /* just update the protection */
200 if (prot_new
!= prot1
) {
201 mprotect(host_start
, qemu_host_page_size
, prot_new
);
207 #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64
208 # define TASK_UNMAPPED_BASE (1ul << 38)
209 #elif defined(__CYGWIN__)
210 /* Cygwin doesn't have a whole lot of address space. */
211 # define TASK_UNMAPPED_BASE 0x18000000
213 # define TASK_UNMAPPED_BASE 0x40000000
215 static abi_ulong mmap_next_start
= TASK_UNMAPPED_BASE
;
217 unsigned long last_brk
;
219 /* Subroutine of mmap_find_vma, used when we have pre-allocated a chunk
220 of guest address space. */
221 static abi_ulong
mmap_find_vma_reserved(abi_ulong start
, abi_ulong size
)
228 if (size
> RESERVED_VA
) {
229 return (abi_ulong
)-1;
233 for (addr
= start
; last_addr
+ size
!= addr
; addr
+= qemu_host_page_size
) {
234 if (last_addr
+ size
>= RESERVED_VA
235 || (abi_ulong
)(last_addr
+ size
) < last_addr
) {
237 return (abi_ulong
)-1;
239 last_addr
= qemu_host_page_size
;
244 prot
= page_get_flags(addr
);
246 last_addr
= addr
+ qemu_host_page_size
;
249 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 truely 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
;
368 unsigned long host_start
;
373 printf("mmap: start=0x" TARGET_ABI_FMT_lx
374 " len=0x" TARGET_ABI_FMT_lx
" prot=%c%c%c flags=",
376 prot
& PROT_READ
? 'r' : '-',
377 prot
& PROT_WRITE
? 'w' : '-',
378 prot
& PROT_EXEC
? 'x' : '-');
379 if (flags
& MAP_FIXED
)
380 printf("MAP_FIXED ");
381 if (flags
& MAP_ANONYMOUS
)
383 switch(flags
& MAP_TYPE
) {
385 printf("MAP_PRIVATE ");
388 printf("MAP_SHARED ");
391 printf("[MAP_TYPE=0x%x] ", flags
& MAP_TYPE
);
394 printf("fd=%d offset=" TARGET_ABI_FMT_lx
"\n", fd
, offset
);
398 if (offset
& ~TARGET_PAGE_MASK
) {
403 len
= TARGET_PAGE_ALIGN(len
);
406 real_start
= start
& qemu_host_page_mask
;
408 /* When mapping files into a memory area larger than the file, accesses
409 to pages beyond the file size will cause a SIGBUS.
411 For example, if mmaping a file of 100 bytes on a host with 4K pages
412 emulating a target with 8K pages, the target expects to be able to
413 access the first 8K. But the host will trap us on any access beyond
416 When emulating a target with a larger page-size than the hosts, we
417 may need to truncate file maps at EOF and add extra anonymous pages
418 up to the targets page boundary. */
420 if ((qemu_real_host_page_size
< TARGET_PAGE_SIZE
)
421 && !(flags
& MAP_ANONYMOUS
)) {
424 if (fstat (fd
, &sb
) == -1)
427 /* Are we trying to create a map beyond EOF?. */
428 if (offset
+ len
> sb
.st_size
) {
429 /* If so, truncate the file map at eof aligned with
430 the hosts real pagesize. Additional anonymous maps
431 will be created beyond EOF. */
432 len
= (sb
.st_size
- offset
);
433 len
+= qemu_real_host_page_size
- 1;
434 len
&= ~(qemu_real_host_page_size
- 1);
438 if (!(flags
& MAP_FIXED
)) {
439 abi_ulong mmap_start
;
441 host_offset
= offset
& qemu_host_page_mask
;
442 host_len
= len
+ offset
- host_offset
;
443 host_len
= HOST_PAGE_ALIGN(host_len
);
444 mmap_start
= mmap_find_vma(real_start
, host_len
);
445 if (mmap_start
== (abi_ulong
)-1) {
449 /* Note: we prefer to control the mapping address. It is
450 especially important if qemu_host_page_size >
451 qemu_real_host_page_size */
452 p
= mmap(g2h(mmap_start
),
453 host_len
, prot
, flags
| MAP_FIXED
| MAP_ANONYMOUS
, -1, 0);
456 /* update start so that it points to the file position at 'offset' */
457 host_start
= (unsigned long)p
;
458 if (!(flags
& MAP_ANONYMOUS
)) {
459 p
= mmap(g2h(mmap_start
), len
, prot
,
460 flags
| MAP_FIXED
, fd
, host_offset
);
461 host_start
+= offset
- host_offset
;
463 start
= h2g(host_start
);
465 if (start
& ~TARGET_PAGE_MASK
) {
470 real_end
= HOST_PAGE_ALIGN(end
);
473 * Test if requested memory area fits target address space
474 * It can fail only on 64-bit host with 32-bit target.
475 * On any other target/host host mmap() handles this error correctly.
477 if ((unsigned long)start
+ len
- 1 > (abi_ulong
) -1) {
482 /* worst case: we cannot map the file because the offset is not
483 aligned, so we read it */
484 if (!(flags
& MAP_ANONYMOUS
) &&
485 (offset
& ~qemu_host_page_mask
) != (start
& ~qemu_host_page_mask
)) {
486 /* msync() won't work here, so we return an error if write is
487 possible while it is a shared mapping */
488 if ((flags
& MAP_TYPE
) == MAP_SHARED
&&
489 (prot
& PROT_WRITE
)) {
493 retaddr
= target_mmap(start
, len
, prot
| PROT_WRITE
,
494 MAP_FIXED
| MAP_PRIVATE
| MAP_ANONYMOUS
,
498 if (pread(fd
, g2h(start
), len
, offset
) == -1)
500 if (!(prot
& PROT_WRITE
)) {
501 ret
= target_mprotect(start
, len
, prot
);
510 /* handle the start of the mapping */
511 if (start
> real_start
) {
512 if (real_end
== real_start
+ qemu_host_page_size
) {
513 /* one single host page */
514 ret
= mmap_frag(real_start
, start
, end
,
515 prot
, flags
, fd
, offset
);
520 ret
= mmap_frag(real_start
, start
, real_start
+ qemu_host_page_size
,
521 prot
, flags
, fd
, offset
);
524 real_start
+= qemu_host_page_size
;
526 /* handle the end of the mapping */
527 if (end
< real_end
) {
528 ret
= mmap_frag(real_end
- qemu_host_page_size
,
529 real_end
- qemu_host_page_size
, real_end
,
531 offset
+ real_end
- qemu_host_page_size
- start
);
534 real_end
-= qemu_host_page_size
;
537 /* map the middle (easier) */
538 if (real_start
< real_end
) {
540 unsigned long offset1
;
541 if (flags
& MAP_ANONYMOUS
)
544 offset1
= offset
+ real_start
- start
;
545 p
= mmap(g2h(real_start
), real_end
- real_start
,
546 prot
, flags
, fd
, offset1
);
552 page_set_flags(start
, start
+ len
, prot
| PAGE_VALID
);
555 printf("ret=0x" TARGET_ABI_FMT_lx
"\n", start
);
566 static void mmap_reserve(abi_ulong start
, abi_ulong size
)
568 abi_ulong real_start
;
574 real_start
= start
& qemu_host_page_mask
;
575 real_end
= HOST_PAGE_ALIGN(start
+ size
);
577 if (start
> real_start
) {
578 /* handle host page containing start */
580 for (addr
= real_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
581 prot
|= page_get_flags(addr
);
583 if (real_end
== real_start
+ qemu_host_page_size
) {
584 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
585 prot
|= page_get_flags(addr
);
590 real_start
+= qemu_host_page_size
;
592 if (end
< real_end
) {
594 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
595 prot
|= page_get_flags(addr
);
598 real_end
-= qemu_host_page_size
;
600 if (real_start
!= real_end
) {
601 mmap(g2h(real_start
), real_end
- real_start
, PROT_NONE
,
602 MAP_FIXED
| MAP_ANONYMOUS
| MAP_PRIVATE
| MAP_NORESERVE
,
607 int target_munmap(abi_ulong start
, abi_ulong len
)
609 abi_ulong end
, real_start
, real_end
, addr
;
613 printf("munmap: start=0x" TARGET_ABI_FMT_lx
" len=0x"
614 TARGET_ABI_FMT_lx
"\n",
617 if (start
& ~TARGET_PAGE_MASK
)
619 len
= TARGET_PAGE_ALIGN(len
);
624 real_start
= start
& qemu_host_page_mask
;
625 real_end
= HOST_PAGE_ALIGN(end
);
627 if (start
> real_start
) {
628 /* handle host page containing start */
630 for(addr
= real_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
631 prot
|= page_get_flags(addr
);
633 if (real_end
== real_start
+ qemu_host_page_size
) {
634 for(addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
635 prot
|= page_get_flags(addr
);
640 real_start
+= qemu_host_page_size
;
642 if (end
< real_end
) {
644 for(addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
645 prot
|= page_get_flags(addr
);
648 real_end
-= qemu_host_page_size
;
652 /* unmap what we can */
653 if (real_start
< real_end
) {
655 mmap_reserve(real_start
, real_end
- real_start
);
657 ret
= munmap(g2h(real_start
), real_end
- real_start
);
662 page_set_flags(start
, start
+ len
, 0);
667 abi_long
target_mremap(abi_ulong old_addr
, abi_ulong old_size
,
668 abi_ulong new_size
, unsigned long flags
,
676 if (flags
& MREMAP_FIXED
) {
677 host_addr
= (void *) syscall(__NR_mremap
, g2h(old_addr
),
682 if (RESERVED_VA
&& host_addr
!= MAP_FAILED
) {
683 /* If new and old addresses overlap then the above mremap will
684 already have failed with EINVAL. */
685 mmap_reserve(old_addr
, old_size
);
687 } else if (flags
& MREMAP_MAYMOVE
) {
688 abi_ulong mmap_start
;
690 mmap_start
= mmap_find_vma(0, new_size
);
692 if (mmap_start
== -1) {
694 host_addr
= MAP_FAILED
;
696 host_addr
= (void *) syscall(__NR_mremap
, g2h(old_addr
),
698 flags
| MREMAP_FIXED
,
700 mmap_reserve(old_addr
, old_size
);
704 if (RESERVED_VA
&& old_size
< new_size
) {
706 for (addr
= old_addr
+ old_size
;
707 addr
< old_addr
+ new_size
;
709 prot
|= page_get_flags(addr
);
713 host_addr
= mremap(g2h(old_addr
), old_size
, new_size
, flags
);
714 if (host_addr
!= MAP_FAILED
&& RESERVED_VA
&& old_size
> new_size
) {
715 mmap_reserve(old_addr
+ old_size
, new_size
- old_size
);
719 host_addr
= MAP_FAILED
;
721 /* Check if address fits target address space */
722 if ((unsigned long)host_addr
+ new_size
> (abi_ulong
)-1) {
723 /* Revert mremap() changes */
724 host_addr
= mremap(g2h(old_addr
), new_size
, old_size
, flags
);
726 host_addr
= MAP_FAILED
;
730 if (host_addr
== MAP_FAILED
) {
733 new_addr
= h2g(host_addr
);
734 prot
= page_get_flags(old_addr
);
735 page_set_flags(old_addr
, old_addr
+ old_size
, 0);
736 page_set_flags(new_addr
, new_addr
+ new_size
, prot
| PAGE_VALID
);
742 int target_msync(abi_ulong start
, abi_ulong len
, int flags
)
746 if (start
& ~TARGET_PAGE_MASK
)
748 len
= TARGET_PAGE_ALIGN(len
);
755 start
&= qemu_host_page_mask
;
756 return msync(g2h(start
), end
- start
, flags
);