2 * mmap support for qemu
4 * Copyright (c) 2003 - 2008 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"
23 static pthread_mutex_t mmap_mutex
= PTHREAD_MUTEX_INITIALIZER
;
24 static __thread
int mmap_lock_count
;
28 if (mmap_lock_count
++ == 0) {
29 pthread_mutex_lock(&mmap_mutex
);
33 void mmap_unlock(void)
35 assert(mmap_lock_count
> 0);
36 if (--mmap_lock_count
== 0) {
37 pthread_mutex_unlock(&mmap_mutex
);
41 bool have_mmap_lock(void)
43 return mmap_lock_count
> 0 ? true : false;
46 /* Grab lock to make sure things are in a consistent state after fork(). */
47 void mmap_fork_start(void)
51 pthread_mutex_lock(&mmap_mutex
);
54 void mmap_fork_end(int child
)
57 pthread_mutex_init(&mmap_mutex
, NULL
);
59 pthread_mutex_unlock(&mmap_mutex
);
62 /* NOTE: all the constants are the HOST ones, but addresses are target. */
63 int target_mprotect(abi_ulong start
, abi_ulong len
, int prot
)
65 abi_ulong end
, host_start
, host_end
, addr
;
68 qemu_log_mask(CPU_LOG_PAGE
, "mprotect: start=0x" TARGET_ABI_FMT_lx
69 " len=0x" TARGET_ABI_FMT_lx
" prot=%c%c%c\n", start
, len
,
70 prot
& PROT_READ
? 'r' : '-',
71 prot
& PROT_WRITE
? 'w' : '-',
72 prot
& PROT_EXEC
? 'x' : '-');
73 if ((start
& ~TARGET_PAGE_MASK
) != 0)
75 len
= TARGET_PAGE_ALIGN(len
);
79 prot
&= PROT_READ
| PROT_WRITE
| PROT_EXEC
;
84 host_start
= start
& qemu_host_page_mask
;
85 host_end
= HOST_PAGE_ALIGN(end
);
86 if (start
> host_start
) {
87 /* handle host page containing start */
89 for (addr
= host_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
90 prot1
|= page_get_flags(addr
);
92 if (host_end
== host_start
+ qemu_host_page_size
) {
93 for (addr
= end
; addr
< host_end
; addr
+= TARGET_PAGE_SIZE
) {
94 prot1
|= page_get_flags(addr
);
98 ret
= mprotect(g2h_untagged(host_start
),
99 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_untagged(host_end
- qemu_host_page_size
),
110 qemu_host_page_size
, prot1
& PAGE_BITS
);
113 host_end
-= qemu_host_page_size
;
116 /* handle the pages in the middle */
117 if (host_start
< host_end
) {
118 ret
= mprotect(g2h_untagged(host_start
), host_end
- host_start
, prot
);
122 page_set_flags(start
, start
+ len
- 1, prot
| PAGE_VALID
);
131 * map an incomplete host page
133 * mmap_frag can be called with a valid fd, if flags doesn't contain one of
134 * MAP_ANON, MAP_STACK, MAP_GUARD. If we need to map a page in those cases, we
135 * pass fd == -1. However, if flags contains MAP_GUARD then MAP_ANON cannot be
138 * * If fd is valid (not -1) we want to map the pages with MAP_ANON.
139 * * If flags contains MAP_GUARD we don't want to add MAP_ANON because it
140 * will be rejected. See kern_mmap's enforcing of constraints for MAP_GUARD
141 * in sys/vm/vm_mmap.c.
142 * * If flags contains MAP_ANON it doesn't matter if we add it or not.
143 * * If flags contains MAP_STACK, mmap adds MAP_ANON when called so doesn't
144 * matter if we add it or not either. See enforcing of constraints for
145 * MAP_STACK in kern_mmap.
147 * Don't add MAP_ANON for the flags that use fd == -1 without specifying the
148 * flags directly, with the assumption that future flags that require fd == -1
149 * will also not require MAP_ANON.
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_untagged(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. See also above. */
171 void *p
= mmap(host_start
, qemu_host_page_size
, prot
,
172 flags
| ((fd
!= -1) ? MAP_ANON
: 0), -1, 0);
179 prot_new
= prot
| prot1
;
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
& TARGET_BSD_MAP_FLAGMASK
) == 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_untagged(start
), end
- start
, offset
) == -1) {
196 /* put final protection */
197 if (prot_new
!= (prot1
| PROT_WRITE
))
198 mprotect(host_start
, qemu_host_page_size
, prot_new
);
200 if (prot_new
!= prot1
) {
201 mprotect(host_start
, qemu_host_page_size
, prot_new
);
203 if (prot_new
& PROT_WRITE
) {
204 memset(g2h_untagged(start
), 0, end
- start
);
210 #if HOST_LONG_BITS == 64 && TARGET_ABI_BITS == 64
211 # define TASK_UNMAPPED_BASE (1ul << 38)
213 # define TASK_UNMAPPED_BASE 0x40000000
215 abi_ulong mmap_next_start
= TASK_UNMAPPED_BASE
;
217 unsigned long last_brk
;
220 * Subroutine of mmap_find_vma, used when we have pre-allocated a chunk of guest
223 static abi_ulong
mmap_find_vma_reserved(abi_ulong start
, abi_ulong size
,
228 ret
= page_find_range_empty(start
, reserved_va
, size
, alignment
);
229 if (ret
== -1 && start
> TARGET_PAGE_SIZE
) {
230 /* Restart at the beginning of the address space. */
231 ret
= page_find_range_empty(TARGET_PAGE_SIZE
, start
- 1,
239 * Find and reserve a free memory area of size 'size'. The search
241 * It must be called with mmap_lock() held.
242 * Return -1 if error.
244 static abi_ulong
mmap_find_vma_aligned(abi_ulong start
, abi_ulong size
,
252 /* If 'start' == 0, then a default start address is used. */
254 start
= mmap_next_start
;
256 start
&= qemu_host_page_mask
;
259 size
= HOST_PAGE_ALIGN(size
);
262 return mmap_find_vma_reserved(start
, size
,
263 (alignment
!= 0 ? 1 << alignment
: 0));
267 wrapped
= repeat
= 0;
269 flags
= MAP_ANON
| MAP_PRIVATE
;
270 if (alignment
!= 0) {
271 flags
|= MAP_ALIGNED(alignment
);
274 for (;; prev
= ptr
) {
276 * Reserve needed memory area to avoid a race.
277 * It should be discarded using:
278 * - mmap() with MAP_FIXED flag
279 * - mremap() with MREMAP_FIXED flag
280 * - shmat() with SHM_REMAP flag
282 ptr
= mmap(g2h_untagged(addr
), size
, PROT_NONE
,
285 /* ENOMEM, if host address space has no memory */
286 if (ptr
== MAP_FAILED
) {
287 return (abi_ulong
)-1;
291 * Count the number of sequential returns of the same address.
292 * This is used to modify the search algorithm below.
294 repeat
= (ptr
== prev
? repeat
+ 1 : 0);
296 if (h2g_valid(ptr
+ size
- 1)) {
299 if ((addr
& ~TARGET_PAGE_MASK
) == 0) {
301 if (start
== mmap_next_start
&& addr
>= TASK_UNMAPPED_BASE
) {
302 mmap_next_start
= addr
+ size
;
307 /* 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.
315 addr
= TARGET_PAGE_ALIGN(addr
);
319 * Sometimes the kernel decides to perform the allocation
320 * at the top end of memory instead.
322 addr
&= TARGET_PAGE_MASK
;
325 /* Start over at low memory. */
329 /* Fail. This unaligned block must the last. */
335 * Since the result the kernel gave didn't fit, start
336 * 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;
353 * Don't actually use 0 when wrapping, instead indicate
354 * that we'd truly like an allocation in low memory.
356 addr
= TARGET_PAGE_SIZE
;
357 } else if (wrapped
&& addr
>= start
) {
358 return (abi_ulong
)-1;
363 abi_ulong
mmap_find_vma(abi_ulong start
, abi_ulong size
)
365 return mmap_find_vma_aligned(start
, size
, 0);
368 /* NOTE: all the constants are the HOST ones */
369 abi_long
target_mmap(abi_ulong start
, abi_ulong len
, int prot
,
370 int flags
, int fd
, off_t offset
)
372 abi_ulong ret
, end
, real_start
, real_end
, retaddr
, host_offset
, host_len
;
375 if (qemu_loglevel_mask(CPU_LOG_PAGE
)) {
376 qemu_log("mmap: start=0x" TARGET_ABI_FMT_lx
377 " len=0x" TARGET_ABI_FMT_lx
" prot=%c%c%c flags=",
379 prot
& PROT_READ
? 'r' : '-',
380 prot
& PROT_WRITE
? 'w' : '-',
381 prot
& PROT_EXEC
? 'x' : '-');
382 if (flags
& MAP_ALIGNMENT_MASK
) {
383 qemu_log("MAP_ALIGNED(%u) ",
384 (flags
& MAP_ALIGNMENT_MASK
) >> MAP_ALIGNMENT_SHIFT
);
386 if (flags
& MAP_GUARD
) {
387 qemu_log("MAP_GUARD ");
389 if (flags
& MAP_FIXED
) {
390 qemu_log("MAP_FIXED ");
392 if (flags
& MAP_ANON
) {
393 qemu_log("MAP_ANON ");
395 if (flags
& MAP_EXCL
) {
396 qemu_log("MAP_EXCL ");
398 if (flags
& MAP_PRIVATE
) {
399 qemu_log("MAP_PRIVATE ");
401 if (flags
& MAP_SHARED
) {
402 qemu_log("MAP_SHARED ");
404 if (flags
& MAP_NOCORE
) {
405 qemu_log("MAP_NOCORE ");
407 if (flags
& MAP_STACK
) {
408 qemu_log("MAP_STACK ");
410 qemu_log("fd=%d offset=0x%lx\n", fd
, offset
);
413 if ((flags
& MAP_ANON
) && fd
!= -1) {
417 if (flags
& MAP_STACK
) {
418 if ((fd
!= -1) || ((prot
& (PROT_READ
| PROT_WRITE
)) !=
419 (PROT_READ
| PROT_WRITE
))) {
424 if ((flags
& MAP_GUARD
) && (prot
!= PROT_NONE
|| fd
!= -1 ||
425 offset
!= 0 || (flags
& (MAP_SHARED
| MAP_PRIVATE
|
426 /* MAP_PREFAULT | */ /* MAP_PREFAULT not in mman.h */
427 MAP_PREFAULT_READ
| MAP_ANON
| MAP_STACK
)) != 0)) {
432 if (offset
& ~TARGET_PAGE_MASK
) {
442 /* Check for overflows */
443 len
= TARGET_PAGE_ALIGN(len
);
449 real_start
= start
& qemu_host_page_mask
;
450 host_offset
= offset
& qemu_host_page_mask
;
453 * If the user is asking for the kernel to find a location, do that
454 * before we truncate the length for mapping files below.
456 if (!(flags
& MAP_FIXED
)) {
457 host_len
= len
+ offset
- host_offset
;
458 host_len
= HOST_PAGE_ALIGN(host_len
);
459 if ((flags
& MAP_ALIGNMENT_MASK
) != 0)
460 start
= mmap_find_vma_aligned(real_start
, host_len
,
461 (flags
& MAP_ALIGNMENT_MASK
) >> MAP_ALIGNMENT_SHIFT
);
463 start
= mmap_find_vma(real_start
, host_len
);
464 if (start
== (abi_ulong
)-1) {
471 * When mapping files into a memory area larger than the file, accesses
472 * to pages beyond the file size will cause a SIGBUS.
474 * For example, if mmaping a file of 100 bytes on a host with 4K pages
475 * emulating a target with 8K pages, the target expects to be able to
476 * access the first 8K. But the host will trap us on any access beyond
479 * When emulating a target with a larger page-size than the hosts, we
480 * may need to truncate file maps at EOF and add extra anonymous pages
481 * up to the targets page boundary.
484 if ((qemu_real_host_page_size() < qemu_host_page_size
) && fd
!= -1) {
487 if (fstat(fd
, &sb
) == -1) {
491 /* Are we trying to create a map beyond EOF?. */
492 if (offset
+ len
> sb
.st_size
) {
494 * If so, truncate the file map at eof aligned with
495 * the hosts real pagesize. Additional anonymous maps
496 * will be created beyond EOF.
498 len
= REAL_HOST_PAGE_ALIGN(sb
.st_size
- offset
);
502 if (!(flags
& MAP_FIXED
)) {
503 unsigned long host_start
;
506 host_len
= len
+ offset
- host_offset
;
507 host_len
= HOST_PAGE_ALIGN(host_len
);
510 * Note: we prefer to control the mapping address. It is
511 * especially important if qemu_host_page_size >
512 * qemu_real_host_page_size
514 p
= mmap(g2h_untagged(start
), host_len
, prot
,
515 flags
| MAP_FIXED
| ((fd
!= -1) ? MAP_ANON
: 0), -1, 0);
518 /* update start so that it points to the file position at 'offset' */
519 host_start
= (unsigned long)p
;
521 p
= mmap(g2h_untagged(start
), len
, prot
,
522 flags
| MAP_FIXED
, fd
, host_offset
);
523 if (p
== MAP_FAILED
) {
524 munmap(g2h_untagged(start
), host_len
);
527 host_start
+= offset
- host_offset
;
529 start
= h2g(host_start
);
531 if (start
& ~TARGET_PAGE_MASK
) {
536 real_end
= HOST_PAGE_ALIGN(end
);
539 * Test if requested memory area fits target address space
540 * It can fail only on 64-bit host with 32-bit target.
541 * On any other target/host host mmap() handles this error correctly.
543 if (!guest_range_valid_untagged(start
, len
)) {
549 * worst case: we cannot map the file because the offset is not
550 * aligned, so we read it
553 (offset
& ~qemu_host_page_mask
) != (start
& ~qemu_host_page_mask
)) {
555 * msync() won't work here, so we return an error if write is
556 * possible while it is a shared mapping
558 if ((flags
& TARGET_BSD_MAP_FLAGMASK
) == MAP_SHARED
&&
559 (prot
& PROT_WRITE
)) {
563 retaddr
= target_mmap(start
, len
, prot
| PROT_WRITE
,
564 MAP_FIXED
| MAP_PRIVATE
| MAP_ANON
,
568 if (pread(fd
, g2h_untagged(start
), len
, offset
) == -1) {
571 if (!(prot
& PROT_WRITE
)) {
572 ret
= target_mprotect(start
, len
, prot
);
578 /* Reject the mapping if any page within the range is mapped */
579 if ((flags
& MAP_EXCL
) && !page_check_range_empty(start
, end
- 1)) {
584 /* handle the start of the mapping */
585 if (start
> real_start
) {
586 if (real_end
== real_start
+ qemu_host_page_size
) {
587 /* one single host page */
588 ret
= mmap_frag(real_start
, start
, end
,
589 prot
, flags
, fd
, offset
);
594 ret
= mmap_frag(real_start
, start
, real_start
+ qemu_host_page_size
,
595 prot
, flags
, fd
, offset
);
598 real_start
+= qemu_host_page_size
;
600 /* handle the end of the mapping */
601 if (end
< real_end
) {
602 ret
= mmap_frag(real_end
- qemu_host_page_size
,
603 real_end
- qemu_host_page_size
, end
,
605 offset
+ real_end
- qemu_host_page_size
- start
);
608 real_end
-= qemu_host_page_size
;
611 /* map the middle (easier) */
612 if (real_start
< real_end
) {
614 unsigned long offset1
;
615 if (flags
& MAP_ANON
)
618 offset1
= offset
+ real_start
- start
;
619 p
= mmap(g2h_untagged(real_start
), real_end
- real_start
,
620 prot
, flags
, fd
, offset1
);
626 page_set_flags(start
, start
+ len
- 1, prot
| PAGE_VALID
);
629 printf("ret=0x" TARGET_ABI_FMT_lx
"\n", start
);
640 static void mmap_reserve(abi_ulong start
, abi_ulong size
)
642 abi_ulong real_start
;
648 real_start
= start
& qemu_host_page_mask
;
649 real_end
= HOST_PAGE_ALIGN(start
+ size
);
651 if (start
> real_start
) {
652 /* handle host page containing start */
654 for (addr
= real_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
655 prot
|= page_get_flags(addr
);
657 if (real_end
== real_start
+ qemu_host_page_size
) {
658 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
659 prot
|= page_get_flags(addr
);
664 real_start
+= qemu_host_page_size
;
667 if (end
< real_end
) {
669 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
670 prot
|= page_get_flags(addr
);
673 real_end
-= qemu_host_page_size
;
676 if (real_start
!= real_end
) {
677 mmap(g2h_untagged(real_start
), real_end
- real_start
, PROT_NONE
,
678 MAP_FIXED
| MAP_ANON
| MAP_PRIVATE
, -1, 0);
682 int target_munmap(abi_ulong start
, abi_ulong len
)
684 abi_ulong end
, real_start
, real_end
, addr
;
688 printf("munmap: start=0x" TARGET_ABI_FMT_lx
" len=0x"
689 TARGET_ABI_FMT_lx
"\n",
692 if (start
& ~TARGET_PAGE_MASK
)
694 len
= TARGET_PAGE_ALIGN(len
);
699 real_start
= start
& qemu_host_page_mask
;
700 real_end
= HOST_PAGE_ALIGN(end
);
702 if (start
> real_start
) {
703 /* handle host page containing start */
705 for (addr
= real_start
; addr
< start
; addr
+= TARGET_PAGE_SIZE
) {
706 prot
|= page_get_flags(addr
);
708 if (real_end
== real_start
+ qemu_host_page_size
) {
709 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
710 prot
|= page_get_flags(addr
);
715 real_start
+= qemu_host_page_size
;
717 if (end
< real_end
) {
719 for (addr
= end
; addr
< real_end
; addr
+= TARGET_PAGE_SIZE
) {
720 prot
|= page_get_flags(addr
);
723 real_end
-= qemu_host_page_size
;
727 /* unmap what we can */
728 if (real_start
< real_end
) {
730 mmap_reserve(real_start
, real_end
- real_start
);
732 ret
= munmap(g2h_untagged(real_start
), real_end
- real_start
);
737 page_set_flags(start
, start
+ len
- 1, 0);
743 int target_msync(abi_ulong start
, abi_ulong len
, int flags
)
747 if (start
& ~TARGET_PAGE_MASK
)
749 len
= TARGET_PAGE_ALIGN(len
);
756 start
&= qemu_host_page_mask
;
757 return msync(g2h_untagged(start
), end
- start
, flags
);