Multithreaded locking for mmap().
[qemu/mini2440.git] / linux-user / mmap.c
blobc0821386d92bc624cf251f53bd9f8d4babde9404
1 /*
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, write to the Free Software
18 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
20 #include <stdlib.h>
21 #include <stdio.h>
22 #include <stdarg.h>
23 #include <string.h>
24 #include <unistd.h>
25 #include <errno.h>
26 #include <sys/mman.h>
28 #include "qemu.h"
30 //#define DEBUG_MMAP
32 #if defined(USE_NPTL)
33 pthread_mutex_t mmap_mutex;
34 static int __thread mmap_lock_count;
36 void mmap_lock(void)
38 if (mmap_lock_count++ == 0) {
39 pthread_mutex_lock(&mmap_mutex);
43 void mmap_unlock(void)
45 if (--mmap_lock_count == 0) {
46 pthread_mutex_unlock(&mmap_mutex);
49 #else
50 /* We aren't threadsafe to start with, so no need to worry about locking. */
51 void mmap_lock(void)
55 void mmap_unlock(void)
58 #endif
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;
64 int prot1, ret;
66 #ifdef DEBUG_MMAP
67 printf("mprotect: start=0x" TARGET_FMT_lx
68 "len=0x" TARGET_FMT_lx " prot=%c%c%c\n", start, len,
69 prot & PROT_READ ? 'r' : '-',
70 prot & PROT_WRITE ? 'w' : '-',
71 prot & PROT_EXEC ? 'x' : '-');
72 #endif
74 if ((start & ~TARGET_PAGE_MASK) != 0)
75 return -EINVAL;
76 len = TARGET_PAGE_ALIGN(len);
77 end = start + len;
78 if (end < start)
79 return -EINVAL;
80 prot &= PROT_READ | PROT_WRITE | PROT_EXEC;
81 if (len == 0)
82 return 0;
84 mmap_lock();
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 */
89 prot1 = prot;
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);
97 end = host_end;
99 ret = mprotect(g2h(host_start), qemu_host_page_size, prot1 & PAGE_BITS);
100 if (ret != 0)
101 goto error;
102 host_start += qemu_host_page_size;
104 if (end < host_end) {
105 prot1 = prot;
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,
110 prot1 & PAGE_BITS);
111 if (ret != 0)
112 goto error;
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);
119 if (ret != 0)
120 goto error;
122 page_set_flags(start, start + len, prot | PAGE_VALID);
123 mmap_unlock();
124 return 0;
125 error:
126 mmap_unlock();
127 return ret;
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;
136 void *host_start;
137 int prot1, prot_new;
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 */
143 prot1 = 0;
144 for(addr = real_start; addr < real_end; addr++) {
145 if (addr < start || addr >= end)
146 prot1 |= page_get_flags(addr);
149 if (prot1 == 0) {
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);
153 if (p == MAP_FAILED)
154 return -1;
155 prot1 = prot;
157 prot1 &= PAGE_BITS;
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 &&
164 (prot & PROT_WRITE))
165 return -EINVAL;
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 pread(fd, g2h(start), end - start, offset);
174 /* put final protection */
175 if (prot_new != (prot1 | PROT_WRITE))
176 mprotect(host_start, qemu_host_page_size, prot_new);
177 } else {
178 /* just update the protection */
179 if (prot_new != prot1) {
180 mprotect(host_start, qemu_host_page_size, prot_new);
183 return 0;
186 #if defined(__CYGWIN__)
187 /* Cygwin doesn't have a whole lot of address space. */
188 static abi_ulong mmap_next_start = 0x18000000;
189 #else
190 static abi_ulong mmap_next_start = 0x40000000;
191 #endif
193 unsigned long last_brk;
195 /* find a free memory area of size 'size'. The search starts at
196 'start'. If 'start' == 0, then a default start address is used.
197 Return -1 if error.
199 /* page_init() marks pages used by the host as reserved to be sure not
200 to use them. */
201 static abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size)
203 abi_ulong addr, addr1, addr_start;
204 int prot;
205 unsigned long new_brk;
207 new_brk = (unsigned long)sbrk(0);
208 if (last_brk && last_brk < new_brk && last_brk == (target_ulong)last_brk) {
209 /* This is a hack to catch the host allocating memory with brk().
210 If it uses mmap then we loose.
211 FIXME: We really want to avoid the host allocating memory in
212 the first place, and maybe leave some slack to avoid switching
213 to mmap. */
214 page_set_flags(last_brk & TARGET_PAGE_MASK,
215 TARGET_PAGE_ALIGN(new_brk),
216 PAGE_RESERVED);
218 last_brk = new_brk;
220 size = HOST_PAGE_ALIGN(size);
221 start = start & qemu_host_page_mask;
222 addr = start;
223 if (addr == 0)
224 addr = mmap_next_start;
225 addr_start = addr;
226 for(;;) {
227 prot = 0;
228 for(addr1 = addr; addr1 < (addr + size); addr1 += TARGET_PAGE_SIZE) {
229 prot |= page_get_flags(addr1);
231 if (prot == 0)
232 break;
233 addr += qemu_host_page_size;
234 /* we found nothing */
235 if (addr == addr_start)
236 return (abi_ulong)-1;
238 if (start == 0)
239 mmap_next_start = addr + size;
240 return addr;
243 /* NOTE: all the constants are the HOST ones */
244 abi_long target_mmap(abi_ulong start, abi_ulong len, int prot,
245 int flags, int fd, abi_ulong offset)
247 abi_ulong ret, end, real_start, real_end, retaddr, host_offset, host_len;
248 unsigned long host_start;
250 mmap_lock();
251 #ifdef DEBUG_MMAP
253 printf("mmap: start=0x" TARGET_FMT_lx
254 " len=0x" TARGET_FMT_lx " prot=%c%c%c flags=",
255 start, len,
256 prot & PROT_READ ? 'r' : '-',
257 prot & PROT_WRITE ? 'w' : '-',
258 prot & PROT_EXEC ? 'x' : '-');
259 if (flags & MAP_FIXED)
260 printf("MAP_FIXED ");
261 if (flags & MAP_ANONYMOUS)
262 printf("MAP_ANON ");
263 switch(flags & MAP_TYPE) {
264 case MAP_PRIVATE:
265 printf("MAP_PRIVATE ");
266 break;
267 case MAP_SHARED:
268 printf("MAP_SHARED ");
269 break;
270 default:
271 printf("[MAP_TYPE=0x%x] ", flags & MAP_TYPE);
272 break;
274 printf("fd=%d offset=" TARGET_FMT_lx "\n", fd, offset);
276 #endif
278 if (offset & ~TARGET_PAGE_MASK) {
279 errno = EINVAL;
280 goto fail;
283 len = TARGET_PAGE_ALIGN(len);
284 if (len == 0)
285 goto the_end;
286 real_start = start & qemu_host_page_mask;
288 if (!(flags & MAP_FIXED)) {
289 abi_ulong mmap_start;
290 void *p;
291 host_offset = offset & qemu_host_page_mask;
292 host_len = len + offset - host_offset;
293 host_len = HOST_PAGE_ALIGN(host_len);
294 mmap_start = mmap_find_vma(real_start, host_len);
295 if (mmap_start == (abi_ulong)-1) {
296 errno = ENOMEM;
297 goto fail;
299 /* Note: we prefer to control the mapping address. It is
300 especially important if qemu_host_page_size >
301 qemu_real_host_page_size */
302 p = mmap(g2h(mmap_start),
303 host_len, prot, flags | MAP_FIXED, fd, host_offset);
304 if (p == MAP_FAILED)
305 goto fail;
306 /* update start so that it points to the file position at 'offset' */
307 host_start = (unsigned long)p;
308 if (!(flags & MAP_ANONYMOUS))
309 host_start += offset - host_offset;
310 start = h2g(host_start);
311 } else {
312 int flg;
313 target_ulong addr;
315 if (start & ~TARGET_PAGE_MASK) {
316 errno = EINVAL;
317 goto fail;
319 end = start + len;
320 real_end = HOST_PAGE_ALIGN(end);
322 for(addr = real_start; addr < real_end; addr += TARGET_PAGE_SIZE) {
323 flg = page_get_flags(addr);
324 if (flg & PAGE_RESERVED) {
325 errno = ENXIO;
326 goto fail;
330 /* worst case: we cannot map the file because the offset is not
331 aligned, so we read it */
332 if (!(flags & MAP_ANONYMOUS) &&
333 (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) {
334 /* msync() won't work here, so we return an error if write is
335 possible while it is a shared mapping */
336 if ((flags & MAP_TYPE) == MAP_SHARED &&
337 (prot & PROT_WRITE)) {
338 errno = EINVAL;
339 goto fail;
341 retaddr = target_mmap(start, len, prot | PROT_WRITE,
342 MAP_FIXED | MAP_PRIVATE | MAP_ANONYMOUS,
343 -1, 0);
344 if (retaddr == -1)
345 goto fail;
346 pread(fd, g2h(start), len, offset);
347 if (!(prot & PROT_WRITE)) {
348 ret = target_mprotect(start, len, prot);
349 if (ret != 0) {
350 start = ret;
351 goto the_end;
354 goto the_end;
357 /* handle the start of the mapping */
358 if (start > real_start) {
359 if (real_end == real_start + qemu_host_page_size) {
360 /* one single host page */
361 ret = mmap_frag(real_start, start, end,
362 prot, flags, fd, offset);
363 if (ret == -1)
364 goto fail;
365 goto the_end1;
367 ret = mmap_frag(real_start, start, real_start + qemu_host_page_size,
368 prot, flags, fd, offset);
369 if (ret == -1)
370 goto fail;
371 real_start += qemu_host_page_size;
373 /* handle the end of the mapping */
374 if (end < real_end) {
375 ret = mmap_frag(real_end - qemu_host_page_size,
376 real_end - qemu_host_page_size, real_end,
377 prot, flags, fd,
378 offset + real_end - qemu_host_page_size - start);
379 if (ret == -1)
380 goto fail;
381 real_end -= qemu_host_page_size;
384 /* map the middle (easier) */
385 if (real_start < real_end) {
386 void *p;
387 unsigned long offset1;
388 if (flags & MAP_ANONYMOUS)
389 offset1 = 0;
390 else
391 offset1 = offset + real_start - start;
392 p = mmap(g2h(real_start), real_end - real_start,
393 prot, flags, fd, offset1);
394 if (p == MAP_FAILED)
395 goto fail;
398 the_end1:
399 page_set_flags(start, start + len, prot | PAGE_VALID);
400 the_end:
401 #ifdef DEBUG_MMAP
402 printf("ret=0x" TARGET_FMT_lx "\n", start);
403 page_dump(stdout);
404 printf("\n");
405 #endif
406 mmap_unlock();
407 return start;
408 fail:
409 mmap_unlock();
410 return -1;
413 int target_munmap(abi_ulong start, abi_ulong len)
415 abi_ulong end, real_start, real_end, addr;
416 int prot, ret;
418 #ifdef DEBUG_MMAP
419 printf("munmap: start=0x%lx len=0x%lx\n", start, len);
420 #endif
421 if (start & ~TARGET_PAGE_MASK)
422 return -EINVAL;
423 len = TARGET_PAGE_ALIGN(len);
424 if (len == 0)
425 return -EINVAL;
426 mmap_lock();
427 end = start + len;
428 real_start = start & qemu_host_page_mask;
429 real_end = HOST_PAGE_ALIGN(end);
431 if (start > real_start) {
432 /* handle host page containing start */
433 prot = 0;
434 for(addr = real_start; addr < start; addr += TARGET_PAGE_SIZE) {
435 prot |= page_get_flags(addr);
437 if (real_end == real_start + qemu_host_page_size) {
438 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
439 prot |= page_get_flags(addr);
441 end = real_end;
443 if (prot != 0)
444 real_start += qemu_host_page_size;
446 if (end < real_end) {
447 prot = 0;
448 for(addr = end; addr < real_end; addr += TARGET_PAGE_SIZE) {
449 prot |= page_get_flags(addr);
451 if (prot != 0)
452 real_end -= qemu_host_page_size;
455 ret = 0;
456 /* unmap what we can */
457 if (real_start < real_end) {
458 ret = munmap(g2h(real_start), real_end - real_start);
461 if (ret == 0)
462 page_set_flags(start, start + len, 0);
463 mmap_unlock();
464 return ret;
467 /* XXX: currently, we only handle MAP_ANONYMOUS and not MAP_FIXED
468 blocks which have been allocated starting on a host page */
469 abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size,
470 abi_ulong new_size, unsigned long flags,
471 abi_ulong new_addr)
473 int prot;
474 unsigned long host_addr;
476 mmap_lock();
477 /* XXX: use 5 args syscall */
478 host_addr = (long)mremap(g2h(old_addr), old_size, new_size, flags);
479 if (host_addr == -1) {
480 new_addr = -1;
481 } else {
482 new_addr = h2g(host_addr);
483 prot = page_get_flags(old_addr);
484 page_set_flags(old_addr, old_addr + old_size, 0);
485 page_set_flags(new_addr, new_addr + new_size, prot | PAGE_VALID);
487 mmap_unlock();
488 return new_addr;
491 int target_msync(abi_ulong start, abi_ulong len, int flags)
493 abi_ulong end;
495 if (start & ~TARGET_PAGE_MASK)
496 return -EINVAL;
497 len = TARGET_PAGE_ALIGN(len);
498 end = start + len;
499 if (end < start)
500 return -EINVAL;
501 if (end == start)
502 return 0;
504 start &= qemu_host_page_mask;
505 return msync(g2h(start), end - start, flags);