]>
Commit | Line | Data |
---|---|---|
54936004 FB |
1 | /* |
2 | * mmap support for qemu | |
5fafdf24 | 3 | * |
54936004 FB |
4 | * Copyright (c) 2003 Fabrice Bellard |
5 | * | |
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. | |
10 | * | |
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. | |
15 | * | |
16 | * You should have received a copy of the GNU General Public License | |
8167ee88 | 17 | * along with this program; if not, see <http://www.gnu.org/licenses/>. |
54936004 | 18 | */ |
d39594e9 | 19 | #include "qemu/osdep.h" |
225a206c | 20 | #include <sys/shm.h> |
11d96056 | 21 | #include "trace.h" |
10d0d505 | 22 | #include "exec/log.h" |
54936004 | 23 | #include "qemu.h" |
3b249d26 | 24 | #include "user-internals.h" |
5423e6d3 | 25 | #include "user-mmap.h" |
8655b4c7 | 26 | #include "target_mman.h" |
044e95c8 | 27 | #include "qemu/interval-tree.h" |
54936004 | 28 | |
5a534314 PM |
29 | #ifdef TARGET_ARM |
30 | #include "target/arm/cpu-features.h" | |
31 | #endif | |
32 | ||
1e6eec8b | 33 | static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER; |
dfd3f85c | 34 | static __thread int mmap_lock_count; |
c8a706fe PB |
35 | |
36 | void mmap_lock(void) | |
37 | { | |
38 | if (mmap_lock_count++ == 0) { | |
39 | pthread_mutex_lock(&mmap_mutex); | |
40 | } | |
41 | } | |
42 | ||
43 | void mmap_unlock(void) | |
44 | { | |
990ef918 | 45 | assert(mmap_lock_count > 0); |
c8a706fe PB |
46 | if (--mmap_lock_count == 0) { |
47 | pthread_mutex_unlock(&mmap_mutex); | |
48 | } | |
49 | } | |
d5975363 | 50 | |
301e40ed AB |
51 | bool have_mmap_lock(void) |
52 | { | |
53 | return mmap_lock_count > 0 ? true : false; | |
54 | } | |
55 | ||
d5975363 PB |
56 | /* Grab lock to make sure things are in a consistent state after fork(). */ |
57 | void mmap_fork_start(void) | |
58 | { | |
59 | if (mmap_lock_count) | |
60 | abort(); | |
61 | pthread_mutex_lock(&mmap_mutex); | |
62 | } | |
63 | ||
64 | void mmap_fork_end(int child) | |
65 | { | |
2b730f79 | 66 | if (child) { |
d5975363 | 67 | pthread_mutex_init(&mmap_mutex, NULL); |
2b730f79 | 68 | } else { |
d5975363 | 69 | pthread_mutex_unlock(&mmap_mutex); |
2b730f79 | 70 | } |
d5975363 | 71 | } |
c8a706fe | 72 | |
044e95c8 RH |
73 | /* Protected by mmap_lock. */ |
74 | static IntervalTreeRoot shm_regions; | |
75 | ||
76 | static void shm_region_add(abi_ptr start, abi_ptr last) | |
77 | { | |
78 | IntervalTreeNode *i = g_new0(IntervalTreeNode, 1); | |
79 | ||
80 | i->start = start; | |
81 | i->last = last; | |
82 | interval_tree_insert(i, &shm_regions); | |
83 | } | |
84 | ||
85 | static abi_ptr shm_region_find(abi_ptr start) | |
86 | { | |
87 | IntervalTreeNode *i; | |
88 | ||
89 | for (i = interval_tree_iter_first(&shm_regions, start, start); i; | |
90 | i = interval_tree_iter_next(i, start, start)) { | |
91 | if (i->start == start) { | |
92 | return i->last; | |
93 | } | |
94 | } | |
95 | return 0; | |
96 | } | |
97 | ||
98 | static void shm_region_rm_complete(abi_ptr start, abi_ptr last) | |
99 | { | |
100 | IntervalTreeNode *i, *n; | |
101 | ||
102 | for (i = interval_tree_iter_first(&shm_regions, start, last); i; i = n) { | |
103 | n = interval_tree_iter_next(i, start, last); | |
104 | if (i->start >= start && i->last <= last) { | |
105 | interval_tree_remove(i, &shm_regions); | |
106 | g_free(i); | |
107 | } | |
108 | } | |
109 | } | |
110 | ||
9dba3ca5 RH |
111 | /* |
112 | * Validate target prot bitmask. | |
113 | * Return the prot bitmask for the host in *HOST_PROT. | |
114 | * Return 0 if the target prot bitmask is invalid, otherwise | |
115 | * the internal qemu page_flags (which will include PAGE_VALID). | |
116 | */ | |
0dd55812 | 117 | static int validate_prot_to_pageflags(int prot) |
9dba3ca5 RH |
118 | { |
119 | int valid = PROT_READ | PROT_WRITE | PROT_EXEC | TARGET_PROT_SEM; | |
120 | int page_flags = (prot & PAGE_BITS) | PAGE_VALID; | |
121 | ||
be5d6f48 | 122 | #ifdef TARGET_AARCH64 |
d109b46d | 123 | { |
be5d6f48 | 124 | ARMCPU *cpu = ARM_CPU(thread_cpu); |
d109b46d RH |
125 | |
126 | /* | |
127 | * The PROT_BTI bit is only accepted if the cpu supports the feature. | |
128 | * Since this is the unusual case, don't bother checking unless | |
129 | * the bit has been requested. If set and valid, record the bit | |
130 | * within QEMU's page_flags. | |
131 | */ | |
132 | if ((prot & TARGET_PROT_BTI) && cpu_isar_feature(aa64_bti, cpu)) { | |
be5d6f48 RH |
133 | valid |= TARGET_PROT_BTI; |
134 | page_flags |= PAGE_BTI; | |
135 | } | |
d109b46d RH |
136 | /* Similarly for the PROT_MTE bit. */ |
137 | if ((prot & TARGET_PROT_MTE) && cpu_isar_feature(aa64_mte, cpu)) { | |
138 | valid |= TARGET_PROT_MTE; | |
139 | page_flags |= PAGE_MTE; | |
140 | } | |
be5d6f48 | 141 | } |
4c184e70 HD |
142 | #elif defined(TARGET_HPPA) |
143 | valid |= PROT_GROWSDOWN | PROT_GROWSUP; | |
be5d6f48 RH |
144 | #endif |
145 | ||
9dba3ca5 RH |
146 | return prot & ~valid ? 0 : page_flags; |
147 | } | |
148 | ||
0dd55812 RH |
149 | /* |
150 | * For the host, we need not pass anything except read/write/exec. | |
151 | * While PROT_SEM is allowed by all hosts, it is also ignored, so | |
152 | * don't bother transforming guest bit to host bit. Any other | |
153 | * target-specific prot bits will not be understood by the host | |
154 | * and will need to be encoded into page_flags for qemu emulation. | |
155 | * | |
156 | * Pages that are executable by the guest will never be executed | |
157 | * by the host, but the host will need to be able to read them. | |
158 | */ | |
159 | static int target_to_host_prot(int prot) | |
160 | { | |
161 | return (prot & (PROT_READ | PROT_WRITE)) | | |
162 | (prot & PROT_EXEC ? PROT_READ : 0); | |
163 | } | |
164 | ||
53a5960a | 165 | /* NOTE: all the constants are the HOST ones, but addresses are target. */ |
9dba3ca5 | 166 | int target_mprotect(abi_ulong start, abi_ulong len, int target_prot) |
54936004 | 167 | { |
621ac47d | 168 | int host_page_size = qemu_real_host_page_size(); |
7bdc1acc RH |
169 | abi_ulong starts[3]; |
170 | abi_ulong lens[3]; | |
171 | int prots[3]; | |
172 | abi_ulong host_start, host_last, last; | |
173 | int prot1, ret, page_flags, nranges; | |
54936004 | 174 | |
9dba3ca5 | 175 | trace_target_mprotect(start, len, target_prot); |
54936004 | 176 | |
9dba3ca5 | 177 | if ((start & ~TARGET_PAGE_MASK) != 0) { |
78cf3390 | 178 | return -TARGET_EINVAL; |
9dba3ca5 | 179 | } |
0dd55812 | 180 | page_flags = validate_prot_to_pageflags(target_prot); |
9dba3ca5 RH |
181 | if (!page_flags) { |
182 | return -TARGET_EINVAL; | |
183 | } | |
7bdc1acc RH |
184 | if (len == 0) { |
185 | return 0; | |
186 | } | |
54936004 | 187 | len = TARGET_PAGE_ALIGN(len); |
46b12f46 | 188 | if (!guest_range_valid_untagged(start, len)) { |
78cf3390 | 189 | return -TARGET_ENOMEM; |
ebf9a363 | 190 | } |
3b46e624 | 191 | |
7bdc1acc | 192 | last = start + len - 1; |
621ac47d | 193 | host_start = start & -host_page_size; |
b36b2b1d | 194 | host_last = ROUND_UP(last, host_page_size) - 1; |
7bdc1acc RH |
195 | nranges = 0; |
196 | ||
197 | mmap_lock(); | |
198 | ||
621ac47d | 199 | if (host_last - host_start < host_page_size) { |
7bdc1acc | 200 | /* Single host page contains all guest pages: sum the prot. */ |
0dd55812 | 201 | prot1 = target_prot; |
7bdc1acc RH |
202 | for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) { |
203 | prot1 |= page_get_flags(a); | |
54936004 | 204 | } |
7bdc1acc RH |
205 | for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) { |
206 | prot1 |= page_get_flags(a + 1); | |
d418c81e | 207 | } |
7bdc1acc | 208 | starts[nranges] = host_start; |
621ac47d | 209 | lens[nranges] = host_page_size; |
7bdc1acc RH |
210 | prots[nranges] = prot1; |
211 | nranges++; | |
212 | } else { | |
213 | if (host_start < start) { | |
214 | /* Host page contains more than one guest page: sum the prot. */ | |
215 | prot1 = target_prot; | |
216 | for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) { | |
217 | prot1 |= page_get_flags(a); | |
218 | } | |
219 | /* If the resulting sum differs, create a new range. */ | |
220 | if (prot1 != target_prot) { | |
221 | starts[nranges] = host_start; | |
621ac47d | 222 | lens[nranges] = host_page_size; |
7bdc1acc RH |
223 | prots[nranges] = prot1; |
224 | nranges++; | |
621ac47d | 225 | host_start += host_page_size; |
7bdc1acc | 226 | } |
9dba3ca5 | 227 | } |
7bdc1acc RH |
228 | |
229 | if (last < host_last) { | |
230 | /* Host page contains more than one guest page: sum the prot. */ | |
231 | prot1 = target_prot; | |
232 | for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) { | |
233 | prot1 |= page_get_flags(a + 1); | |
234 | } | |
235 | /* If the resulting sum differs, create a new range. */ | |
236 | if (prot1 != target_prot) { | |
621ac47d | 237 | host_last -= host_page_size; |
7bdc1acc | 238 | starts[nranges] = host_last + 1; |
621ac47d | 239 | lens[nranges] = host_page_size; |
7bdc1acc RH |
240 | prots[nranges] = prot1; |
241 | nranges++; | |
242 | } | |
54936004 | 243 | } |
7bdc1acc RH |
244 | |
245 | /* Create a range for the middle, if any remains. */ | |
246 | if (host_start < host_last) { | |
247 | starts[nranges] = host_start; | |
248 | lens[nranges] = host_last - host_start + 1; | |
249 | prots[nranges] = target_prot; | |
250 | nranges++; | |
9dba3ca5 | 251 | } |
54936004 | 252 | } |
3b46e624 | 253 | |
7bdc1acc RH |
254 | for (int i = 0; i < nranges; ++i) { |
255 | ret = mprotect(g2h_untagged(starts[i]), lens[i], | |
256 | target_to_host_prot(prots[i])); | |
9dba3ca5 | 257 | if (ret != 0) { |
c8a706fe | 258 | goto error; |
9dba3ca5 | 259 | } |
54936004 | 260 | } |
aa98e2d8 | 261 | |
7bdc1acc | 262 | page_set_flags(start, last, page_flags); |
aa98e2d8 IL |
263 | ret = 0; |
264 | ||
7bdc1acc | 265 | error: |
c8a706fe PB |
266 | mmap_unlock(); |
267 | return ret; | |
54936004 FB |
268 | } |
269 | ||
2952b642 RH |
270 | /* |
271 | * Perform munmap on behalf of the target, with host parameters. | |
272 | * If reserved_va, we must replace the memory reservation. | |
273 | */ | |
274 | static int do_munmap(void *addr, size_t len) | |
275 | { | |
276 | if (reserved_va) { | |
277 | void *ptr = mmap(addr, len, PROT_NONE, | |
278 | MAP_FIXED | MAP_ANONYMOUS | |
279 | | MAP_PRIVATE | MAP_NORESERVE, -1, 0); | |
280 | return ptr == addr ? 0 : -1; | |
281 | } | |
282 | return munmap(addr, len); | |
283 | } | |
284 | ||
54936004 | 285 | /* map an incomplete host page */ |
99982beb RH |
286 | static bool mmap_frag(abi_ulong real_start, abi_ulong start, abi_ulong last, |
287 | int prot, int flags, int fd, off_t offset) | |
54936004 | 288 | { |
621ac47d | 289 | int host_page_size = qemu_real_host_page_size(); |
99982beb | 290 | abi_ulong real_last; |
53a5960a | 291 | void *host_start; |
99982beb RH |
292 | int prot_old, prot_new; |
293 | int host_prot_old, host_prot_new; | |
54936004 | 294 | |
99982beb RH |
295 | if (!(flags & MAP_ANONYMOUS) |
296 | && (flags & MAP_TYPE) == MAP_SHARED | |
297 | && (prot & PROT_WRITE)) { | |
298 | /* | |
299 | * msync() won't work with the partial page, so we return an | |
300 | * error if write is possible while it is a shared mapping. | |
301 | */ | |
302 | errno = EINVAL; | |
303 | return false; | |
304 | } | |
305 | ||
621ac47d | 306 | real_last = real_start + host_page_size - 1; |
3e8f1628 | 307 | host_start = g2h_untagged(real_start); |
54936004 | 308 | |
99982beb RH |
309 | /* Get the protection of the target pages outside the mapping. */ |
310 | prot_old = 0; | |
311 | for (abi_ulong a = real_start; a < start; a += TARGET_PAGE_SIZE) { | |
312 | prot_old |= page_get_flags(a); | |
313 | } | |
314 | for (abi_ulong a = real_last; a > last; a -= TARGET_PAGE_SIZE) { | |
315 | prot_old |= page_get_flags(a); | |
54936004 | 316 | } |
3b46e624 | 317 | |
99982beb RH |
318 | if (prot_old == 0) { |
319 | /* | |
320 | * Since !(prot_old & PAGE_VALID), there were no guest pages | |
321 | * outside of the fragment we need to map. Allocate a new host | |
322 | * page to cover, discarding whatever else may have been present. | |
323 | */ | |
621ac47d | 324 | void *p = mmap(host_start, host_page_size, |
0dd55812 | 325 | target_to_host_prot(prot), |
80210bcd | 326 | flags | MAP_ANONYMOUS, -1, 0); |
ddcdd8c4 AO |
327 | if (p != host_start) { |
328 | if (p != MAP_FAILED) { | |
621ac47d | 329 | munmap(p, host_page_size); |
ddcdd8c4 AO |
330 | errno = EEXIST; |
331 | } | |
99982beb | 332 | return false; |
2b730f79 | 333 | } |
99982beb | 334 | prot_old = prot; |
54936004 | 335 | } |
99982beb | 336 | prot_new = prot | prot_old; |
54936004 | 337 | |
99982beb RH |
338 | host_prot_old = target_to_host_prot(prot_old); |
339 | host_prot_new = target_to_host_prot(prot_new); | |
3b46e624 | 340 | |
99982beb RH |
341 | /* Adjust protection to be able to write. */ |
342 | if (!(host_prot_old & PROT_WRITE)) { | |
343 | host_prot_old |= PROT_WRITE; | |
621ac47d | 344 | mprotect(host_start, host_page_size, host_prot_old); |
99982beb | 345 | } |
3b46e624 | 346 | |
99982beb RH |
347 | /* Read or zero the new guest pages. */ |
348 | if (flags & MAP_ANONYMOUS) { | |
349 | memset(g2h_untagged(start), 0, last - start + 1); | |
54936004 | 350 | } else { |
99982beb RH |
351 | if (pread(fd, g2h_untagged(start), last - start + 1, offset) == -1) { |
352 | return false; | |
e6deac9c | 353 | } |
54936004 | 354 | } |
99982beb RH |
355 | |
356 | /* Put final protection */ | |
357 | if (host_prot_new != host_prot_old) { | |
621ac47d | 358 | mprotect(host_start, host_page_size, host_prot_new); |
99982beb RH |
359 | } |
360 | return true; | |
54936004 FB |
361 | } |
362 | ||
c8fb5cf9 | 363 | abi_ulong task_unmapped_base; |
da2b71fa | 364 | abi_ulong elf_et_dyn_base; |
c8fb5cf9 | 365 | abi_ulong mmap_next_start; |
a03e2d42 | 366 | |
2b730f79 RH |
367 | /* |
368 | * Subroutine of mmap_find_vma, used when we have pre-allocated | |
369 | * a chunk of guest address space. | |
370 | */ | |
30ab9ef2 RH |
371 | static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size, |
372 | abi_ulong align) | |
68a1c816 | 373 | { |
4c13048e | 374 | target_ulong ret; |
68a1c816 | 375 | |
4c13048e RH |
376 | ret = page_find_range_empty(start, reserved_va, size, align); |
377 | if (ret == -1 && start > mmap_min_addr) { | |
378 | /* Restart at the beginning of the address space. */ | |
379 | ret = page_find_range_empty(mmap_min_addr, start - 1, size, align); | |
68a1c816 PB |
380 | } |
381 | ||
4c13048e | 382 | return ret; |
68a1c816 PB |
383 | } |
384 | ||
fe3b4152 KS |
385 | /* |
386 | * Find and reserve a free memory area of size 'size'. The search | |
387 | * starts at 'start'. | |
388 | * It must be called with mmap_lock() held. | |
389 | * Return -1 if error. | |
390 | */ | |
30ab9ef2 | 391 | abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size, abi_ulong align) |
a03e2d42 | 392 | { |
621ac47d | 393 | int host_page_size = qemu_real_host_page_size(); |
14f24e14 | 394 | void *ptr, *prev; |
fe3b4152 | 395 | abi_ulong addr; |
14f24e14 | 396 | int wrapped, repeat; |
fe3b4152 | 397 | |
621ac47d | 398 | align = MAX(align, host_page_size); |
443b7505 | 399 | |
fe3b4152 | 400 | /* If 'start' == 0, then a default start address is used. */ |
14f24e14 | 401 | if (start == 0) { |
fe3b4152 | 402 | start = mmap_next_start; |
14f24e14 | 403 | } else { |
621ac47d | 404 | start &= -host_page_size; |
14f24e14 | 405 | } |
30ab9ef2 | 406 | start = ROUND_UP(start, align); |
b36b2b1d | 407 | size = ROUND_UP(size, host_page_size); |
fe3b4152 | 408 | |
b76f21a7 | 409 | if (reserved_va) { |
30ab9ef2 | 410 | return mmap_find_vma_reserved(start, size, align); |
68a1c816 PB |
411 | } |
412 | ||
a03e2d42 | 413 | addr = start; |
14f24e14 RH |
414 | wrapped = repeat = 0; |
415 | prev = 0; | |
fe3b4152 | 416 | |
14f24e14 | 417 | for (;; prev = ptr) { |
fe3b4152 KS |
418 | /* |
419 | * Reserve needed memory area to avoid a race. | |
420 | * It should be discarded using: | |
421 | * - mmap() with MAP_FIXED flag | |
422 | * - mremap() with MREMAP_FIXED flag | |
423 | * - shmat() with SHM_REMAP flag | |
424 | */ | |
3e8f1628 | 425 | ptr = mmap(g2h_untagged(addr), size, PROT_NONE, |
2b730f79 | 426 | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0); |
fe3b4152 KS |
427 | |
428 | /* ENOMEM, if host address space has no memory */ | |
14f24e14 | 429 | if (ptr == MAP_FAILED) { |
fe3b4152 | 430 | return (abi_ulong)-1; |
14f24e14 RH |
431 | } |
432 | ||
2b730f79 RH |
433 | /* |
434 | * Count the number of sequential returns of the same address. | |
435 | * This is used to modify the search algorithm below. | |
436 | */ | |
14f24e14 RH |
437 | repeat = (ptr == prev ? repeat + 1 : 0); |
438 | ||
439 | if (h2g_valid(ptr + size - 1)) { | |
440 | addr = h2g(ptr); | |
fe3b4152 | 441 | |
30ab9ef2 | 442 | if ((addr & (align - 1)) == 0) { |
14f24e14 | 443 | /* Success. */ |
c8fb5cf9 | 444 | if (start == mmap_next_start && addr >= task_unmapped_base) { |
14f24e14 RH |
445 | mmap_next_start = addr + size; |
446 | } | |
447 | return addr; | |
448 | } | |
fe3b4152 | 449 | |
14f24e14 RH |
450 | /* The address is not properly aligned for the target. */ |
451 | switch (repeat) { | |
452 | case 0: | |
2b730f79 RH |
453 | /* |
454 | * Assume the result that the kernel gave us is the | |
455 | * first with enough free space, so start again at the | |
456 | * next higher target page. | |
457 | */ | |
30ab9ef2 | 458 | addr = ROUND_UP(addr, align); |
14f24e14 RH |
459 | break; |
460 | case 1: | |
2b730f79 RH |
461 | /* |
462 | * Sometimes the kernel decides to perform the allocation | |
463 | * at the top end of memory instead. | |
464 | */ | |
30ab9ef2 | 465 | addr &= -align; |
14f24e14 RH |
466 | break; |
467 | case 2: | |
468 | /* Start over at low memory. */ | |
469 | addr = 0; | |
470 | break; | |
471 | default: | |
472 | /* Fail. This unaligned block must the last. */ | |
473 | addr = -1; | |
474 | break; | |
475 | } | |
476 | } else { | |
2b730f79 RH |
477 | /* |
478 | * Since the result the kernel gave didn't fit, start | |
479 | * again at low memory. If any repetition, fail. | |
480 | */ | |
14f24e14 RH |
481 | addr = (repeat ? -1 : 0); |
482 | } | |
483 | ||
484 | /* Unmap and try again. */ | |
fe3b4152 | 485 | munmap(ptr, size); |
fe3b4152 | 486 | |
14f24e14 | 487 | /* ENOMEM if we checked the whole of the target address space. */ |
d0b3e4f5 | 488 | if (addr == (abi_ulong)-1) { |
a03e2d42 | 489 | return (abi_ulong)-1; |
14f24e14 RH |
490 | } else if (addr == 0) { |
491 | if (wrapped) { | |
492 | return (abi_ulong)-1; | |
493 | } | |
494 | wrapped = 1; | |
2b730f79 RH |
495 | /* |
496 | * Don't actually use 0 when wrapping, instead indicate | |
497 | * that we'd truly like an allocation in low memory. | |
498 | */ | |
14f24e14 RH |
499 | addr = (mmap_min_addr > TARGET_PAGE_SIZE |
500 | ? TARGET_PAGE_ALIGN(mmap_min_addr) | |
501 | : TARGET_PAGE_SIZE); | |
502 | } else if (wrapped && addr >= start) { | |
503 | return (abi_ulong)-1; | |
504 | } | |
a03e2d42 | 505 | } |
a03e2d42 FB |
506 | } |
507 | ||
6ecc2557 RH |
508 | /* |
509 | * Record a successful mmap within the user-exec interval tree. | |
510 | */ | |
511 | static abi_long mmap_end(abi_ulong start, abi_ulong last, | |
512 | abi_ulong passthrough_start, | |
513 | abi_ulong passthrough_last, | |
514 | int flags, int page_flags) | |
515 | { | |
516 | if (flags & MAP_ANONYMOUS) { | |
517 | page_flags |= PAGE_ANON; | |
518 | } | |
519 | page_flags |= PAGE_RESET; | |
520 | if (passthrough_start > passthrough_last) { | |
521 | page_set_flags(start, last, page_flags); | |
522 | } else { | |
523 | if (start < passthrough_start) { | |
524 | page_set_flags(start, passthrough_start - 1, page_flags); | |
525 | } | |
526 | page_set_flags(passthrough_start, passthrough_last, | |
527 | page_flags | PAGE_PASSTHROUGH); | |
528 | if (passthrough_last < last) { | |
529 | page_set_flags(passthrough_last + 1, last, page_flags); | |
530 | } | |
531 | } | |
532 | shm_region_rm_complete(start, last); | |
533 | trace_target_mmap_complete(start); | |
534 | if (qemu_loglevel_mask(CPU_LOG_PAGE)) { | |
535 | FILE *f = qemu_log_trylock(); | |
536 | if (f) { | |
537 | fprintf(f, "page layout changed following mmap\n"); | |
538 | page_dump(f); | |
539 | qemu_log_unlock(f); | |
540 | } | |
541 | } | |
542 | return start; | |
543 | } | |
544 | ||
d558c395 | 545 | static abi_long target_mmap__locked(abi_ulong start, abi_ulong len, |
e8cec51b | 546 | int target_prot, int flags, int page_flags, |
d558c395 | 547 | int fd, off_t offset) |
54936004 | 548 | { |
621ac47d | 549 | int host_page_size = qemu_real_host_page_size(); |
f9cd8f5e RH |
550 | abi_ulong ret, last, real_start, real_last, retaddr, host_len; |
551 | abi_ulong passthrough_start = -1, passthrough_last = 0; | |
55baec0f | 552 | off_t host_offset; |
54936004 | 553 | |
621ac47d RH |
554 | real_start = start & -host_page_size; |
555 | host_offset = offset & -host_page_size; | |
a5e7ee46 | 556 | |
2b730f79 | 557 | /* |
ad87d26e RH |
558 | * For reserved_va, we are in full control of the allocation. |
559 | * Find a suitable hole and convert to MAP_FIXED. | |
2b730f79 | 560 | */ |
ad87d26e | 561 | if (reserved_va && !(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) { |
a5e7ee46 | 562 | host_len = len + offset - host_offset; |
ad87d26e RH |
563 | start = mmap_find_vma(real_start, host_len, |
564 | MAX(host_page_size, TARGET_PAGE_SIZE)); | |
a5e7ee46 RH |
565 | if (start == (abi_ulong)-1) { |
566 | errno = ENOMEM; | |
d558c395 | 567 | return -1; |
a5e7ee46 | 568 | } |
ad87d26e RH |
569 | start += offset - host_offset; |
570 | flags |= MAP_FIXED; | |
a5e7ee46 | 571 | } |
54936004 | 572 | |
2b730f79 RH |
573 | /* |
574 | * When mapping files into a memory area larger than the file, accesses | |
575 | * to pages beyond the file size will cause a SIGBUS. | |
576 | * | |
577 | * For example, if mmaping a file of 100 bytes on a host with 4K pages | |
578 | * emulating a target with 8K pages, the target expects to be able to | |
579 | * access the first 8K. But the host will trap us on any access beyond | |
580 | * 4K. | |
581 | * | |
582 | * When emulating a target with a larger page-size than the hosts, we | |
583 | * may need to truncate file maps at EOF and add extra anonymous pages | |
584 | * up to the targets page boundary. | |
585 | */ | |
621ac47d | 586 | if (host_page_size < TARGET_PAGE_SIZE && !(flags & MAP_ANONYMOUS)) { |
35f2fd04 | 587 | struct stat sb; |
54c5a2ae | 588 | |
2b730f79 | 589 | if (fstat(fd, &sb) == -1) { |
d558c395 | 590 | return -1; |
2b730f79 | 591 | } |
54c5a2ae | 592 | |
2b730f79 RH |
593 | /* Are we trying to create a map beyond EOF?. */ |
594 | if (offset + len > sb.st_size) { | |
595 | /* | |
596 | * If so, truncate the file map at eof aligned with | |
597 | * the hosts real pagesize. Additional anonymous maps | |
598 | * will be created beyond EOF. | |
599 | */ | |
e56922ab | 600 | len = ROUND_UP(sb.st_size - offset, host_page_size); |
2b730f79 | 601 | } |
54c5a2ae EI |
602 | } |
603 | ||
03798605 | 604 | if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) { |
55baec0f | 605 | uintptr_t host_start; |
0dd55812 | 606 | int host_prot; |
a03e2d42 | 607 | void *p; |
a5e7ee46 | 608 | |
a03e2d42 | 609 | host_len = len + offset - host_offset; |
b36b2b1d | 610 | host_len = ROUND_UP(host_len, host_page_size); |
0dd55812 | 611 | host_prot = target_to_host_prot(target_prot); |
a5e7ee46 | 612 | |
621ac47d | 613 | /* Note: we prefer to control the mapping address. */ |
3e8f1628 | 614 | p = mmap(g2h_untagged(start), host_len, host_prot, |
a5e7ee46 | 615 | flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0); |
9dba3ca5 | 616 | if (p == MAP_FAILED) { |
d558c395 | 617 | return -1; |
9dba3ca5 | 618 | } |
a03e2d42 | 619 | /* update start so that it points to the file position at 'offset' */ |
55baec0f | 620 | host_start = (uintptr_t)p; |
54c5a2ae | 621 | if (!(flags & MAP_ANONYMOUS)) { |
3e8f1628 | 622 | p = mmap(g2h_untagged(start), len, host_prot, |
54c5a2ae | 623 | flags | MAP_FIXED, fd, host_offset); |
8384274e | 624 | if (p == MAP_FAILED) { |
3e8f1628 | 625 | munmap(g2h_untagged(start), host_len); |
d558c395 | 626 | return -1; |
8384274e | 627 | } |
a03e2d42 | 628 | host_start += offset - host_offset; |
54c5a2ae | 629 | } |
a03e2d42 | 630 | start = h2g(host_start); |
f9cd8f5e | 631 | last = start + len - 1; |
f93b7695 | 632 | passthrough_start = start; |
f9cd8f5e | 633 | passthrough_last = last; |
a03e2d42 | 634 | } else { |
f9cd8f5e | 635 | last = start + len - 1; |
b36b2b1d | 636 | real_last = ROUND_UP(last, host_page_size) - 1; |
7ab240ad | 637 | |
c3dd50da AO |
638 | if (flags & MAP_FIXED_NOREPLACE) { |
639 | /* Validate that the chosen range is empty. */ | |
640 | if (!page_check_range_empty(start, last)) { | |
641 | errno = EEXIST; | |
d558c395 | 642 | return -1; |
c3dd50da AO |
643 | } |
644 | ||
645 | /* | |
646 | * With reserved_va, the entire address space is mmaped in the | |
647 | * host to ensure it isn't accidentally used for something else. | |
648 | * We have just checked that the guest address is not mapped | |
649 | * within the guest, but need to replace the host reservation. | |
650 | * | |
651 | * Without reserved_va, despite the guest address check above, | |
652 | * keep MAP_FIXED_NOREPLACE so that the guest does not overwrite | |
653 | * any host address mappings. | |
654 | */ | |
655 | if (reserved_va) { | |
656 | flags = (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED; | |
657 | } | |
03798605 RH |
658 | } |
659 | ||
2b730f79 RH |
660 | /* |
661 | * worst case: we cannot map the file because the offset is not | |
662 | * aligned, so we read it | |
663 | */ | |
a03e2d42 | 664 | if (!(flags & MAP_ANONYMOUS) && |
621ac47d | 665 | (offset & (host_page_size - 1)) != (start & (host_page_size - 1))) { |
2b730f79 RH |
666 | /* |
667 | * msync() won't work here, so we return an error if write is | |
668 | * possible while it is a shared mapping | |
669 | */ | |
0dd55812 RH |
670 | if ((flags & MAP_TYPE) == MAP_SHARED |
671 | && (target_prot & PROT_WRITE)) { | |
a03e2d42 | 672 | errno = EINVAL; |
d558c395 | 673 | return -1; |
a03e2d42 | 674 | } |
9dba3ca5 | 675 | retaddr = target_mmap(start, len, target_prot | PROT_WRITE, |
03798605 RH |
676 | (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) |
677 | | MAP_PRIVATE | MAP_ANONYMOUS, | |
a03e2d42 | 678 | -1, 0); |
2b730f79 | 679 | if (retaddr == -1) { |
d558c395 | 680 | return -1; |
2b730f79 RH |
681 | } |
682 | if (pread(fd, g2h_untagged(start), len, offset) == -1) { | |
d558c395 | 683 | return -1; |
2b730f79 | 684 | } |
0dd55812 | 685 | if (!(target_prot & PROT_WRITE)) { |
9dba3ca5 | 686 | ret = target_mprotect(start, len, target_prot); |
86abac06 | 687 | assert(ret == 0); |
a03e2d42 | 688 | } |
6ecc2557 | 689 | return mmap_end(start, last, -1, 0, flags, page_flags); |
54936004 | 690 | } |
2b730f79 | 691 | |
a03e2d42 FB |
692 | /* handle the start of the mapping */ |
693 | if (start > real_start) { | |
621ac47d | 694 | if (real_last == real_start + host_page_size - 1) { |
a03e2d42 | 695 | /* one single host page */ |
f9cd8f5e | 696 | if (!mmap_frag(real_start, start, last, |
99982beb | 697 | target_prot, flags, fd, offset)) { |
d558c395 | 698 | return -1; |
2b730f79 | 699 | } |
6ecc2557 | 700 | return mmap_end(start, last, -1, 0, flags, page_flags); |
a03e2d42 | 701 | } |
99982beb | 702 | if (!mmap_frag(real_start, start, |
621ac47d | 703 | real_start + host_page_size - 1, |
99982beb | 704 | target_prot, flags, fd, offset)) { |
d558c395 | 705 | return -1; |
2b730f79 | 706 | } |
621ac47d | 707 | real_start += host_page_size; |
a03e2d42 FB |
708 | } |
709 | /* handle the end of the mapping */ | |
f9cd8f5e | 710 | if (last < real_last) { |
621ac47d | 711 | abi_ulong real_page = real_last - host_page_size + 1; |
f9cd8f5e | 712 | if (!mmap_frag(real_page, real_page, last, |
99982beb | 713 | target_prot, flags, fd, |
f9cd8f5e | 714 | offset + real_page - start)) { |
d558c395 | 715 | return -1; |
2b730f79 | 716 | } |
621ac47d | 717 | real_last -= host_page_size; |
54936004 | 718 | } |
3b46e624 | 719 | |
a03e2d42 | 720 | /* map the middle (easier) */ |
f9cd8f5e | 721 | if (real_start < real_last) { |
ddcdd8c4 | 722 | void *p, *want_p; |
55baec0f | 723 | off_t offset1; |
ddcdd8c4 | 724 | size_t len1; |
55baec0f | 725 | |
2b730f79 | 726 | if (flags & MAP_ANONYMOUS) { |
a03e2d42 | 727 | offset1 = 0; |
2b730f79 | 728 | } else { |
a03e2d42 | 729 | offset1 = offset + real_start - start; |
2b730f79 | 730 | } |
ddcdd8c4 AO |
731 | len1 = real_last - real_start + 1; |
732 | want_p = g2h_untagged(real_start); | |
733 | ||
734 | p = mmap(want_p, len1, target_to_host_prot(target_prot), | |
735 | flags, fd, offset1); | |
736 | if (p != want_p) { | |
737 | if (p != MAP_FAILED) { | |
738 | munmap(p, len1); | |
739 | errno = EEXIST; | |
740 | } | |
d558c395 | 741 | return -1; |
2b730f79 | 742 | } |
f93b7695 | 743 | passthrough_start = real_start; |
f9cd8f5e | 744 | passthrough_last = real_last; |
a03e2d42 | 745 | } |
54936004 | 746 | } |
6ecc2557 RH |
747 | return mmap_end(start, last, passthrough_start, passthrough_last, |
748 | flags, page_flags); | |
d558c395 RH |
749 | } |
750 | ||
751 | /* NOTE: all the constants are the HOST ones */ | |
752 | abi_long target_mmap(abi_ulong start, abi_ulong len, int target_prot, | |
753 | int flags, int fd, off_t offset) | |
754 | { | |
755 | abi_long ret; | |
e8cec51b | 756 | int page_flags; |
d558c395 RH |
757 | |
758 | trace_target_mmap(start, len, target_prot, flags, fd, offset); | |
e8cec51b RH |
759 | |
760 | if (!len) { | |
761 | errno = EINVAL; | |
762 | return -1; | |
763 | } | |
764 | ||
765 | page_flags = validate_prot_to_pageflags(target_prot); | |
766 | if (!page_flags) { | |
767 | errno = EINVAL; | |
768 | return -1; | |
769 | } | |
770 | ||
771 | /* Also check for overflows... */ | |
772 | len = TARGET_PAGE_ALIGN(len); | |
773 | if (!len || len != (size_t)len) { | |
774 | errno = ENOMEM; | |
775 | return -1; | |
776 | } | |
777 | ||
778 | if (offset & ~TARGET_PAGE_MASK) { | |
779 | errno = EINVAL; | |
780 | return -1; | |
781 | } | |
782 | if (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) { | |
783 | if (start & ~TARGET_PAGE_MASK) { | |
784 | errno = EINVAL; | |
785 | return -1; | |
786 | } | |
787 | if (!guest_range_valid_untagged(start, len)) { | |
788 | errno = ENOMEM; | |
789 | return -1; | |
790 | } | |
791 | } | |
792 | ||
d558c395 RH |
793 | mmap_lock(); |
794 | ||
e8cec51b RH |
795 | ret = target_mmap__locked(start, len, target_prot, flags, |
796 | page_flags, fd, offset); | |
d558c395 | 797 | |
c8a706fe | 798 | mmap_unlock(); |
e8cec51b RH |
799 | |
800 | /* | |
801 | * If we're mapping shared memory, ensure we generate code for parallel | |
802 | * execution and flush old translations. This will work up to the level | |
803 | * supported by the host -- anything that requires EXCP_ATOMIC will not | |
804 | * be atomic with respect to an external process. | |
805 | */ | |
806 | if (ret != -1 && (flags & MAP_TYPE) != MAP_PRIVATE) { | |
807 | CPUState *cpu = thread_cpu; | |
808 | if (!(cpu->tcg_cflags & CF_PARALLEL)) { | |
809 | cpu->tcg_cflags |= CF_PARALLEL; | |
810 | tb_flush(cpu); | |
811 | } | |
812 | } | |
813 | ||
d558c395 | 814 | return ret; |
54936004 FB |
815 | } |
816 | ||
912ff698 | 817 | static int mmap_reserve_or_unmap(abi_ulong start, abi_ulong len) |
68a1c816 | 818 | { |
621ac47d | 819 | int host_page_size = qemu_real_host_page_size(); |
68a1c816 | 820 | abi_ulong real_start; |
260561d8 RH |
821 | abi_ulong real_last; |
822 | abi_ulong real_len; | |
823 | abi_ulong last; | |
824 | abi_ulong a; | |
558a4411 | 825 | void *host_start; |
68a1c816 PB |
826 | int prot; |
827 | ||
260561d8 | 828 | last = start + len - 1; |
621ac47d | 829 | real_start = start & -host_page_size; |
b36b2b1d | 830 | real_last = ROUND_UP(last, host_page_size) - 1; |
260561d8 RH |
831 | |
832 | /* | |
833 | * If guest pages remain on the first or last host pages, | |
834 | * adjust the deallocation to retain those guest pages. | |
835 | * The single page special case is required for the last page, | |
836 | * lest real_start overflow to zero. | |
837 | */ | |
621ac47d | 838 | if (real_last - real_start < host_page_size) { |
68a1c816 | 839 | prot = 0; |
260561d8 RH |
840 | for (a = real_start; a < start; a += TARGET_PAGE_SIZE) { |
841 | prot |= page_get_flags(a); | |
68a1c816 | 842 | } |
260561d8 RH |
843 | for (a = last; a < real_last; a += TARGET_PAGE_SIZE) { |
844 | prot |= page_get_flags(a + 1); | |
845 | } | |
846 | if (prot != 0) { | |
912ff698 | 847 | return 0; |
260561d8 RH |
848 | } |
849 | } else { | |
850 | for (prot = 0, a = real_start; a < start; a += TARGET_PAGE_SIZE) { | |
851 | prot |= page_get_flags(a); | |
68a1c816 | 852 | } |
2b730f79 | 853 | if (prot != 0) { |
621ac47d | 854 | real_start += host_page_size; |
2b730f79 | 855 | } |
260561d8 RH |
856 | |
857 | for (prot = 0, a = last; a < real_last; a += TARGET_PAGE_SIZE) { | |
858 | prot |= page_get_flags(a + 1); | |
68a1c816 | 859 | } |
2b730f79 | 860 | if (prot != 0) { |
621ac47d | 861 | real_last -= host_page_size; |
260561d8 RH |
862 | } |
863 | ||
864 | if (real_last < real_start) { | |
912ff698 | 865 | return 0; |
2b730f79 | 866 | } |
68a1c816 | 867 | } |
260561d8 RH |
868 | |
869 | real_len = real_last - real_start + 1; | |
870 | host_start = g2h_untagged(real_start); | |
871 | ||
2952b642 | 872 | return do_munmap(host_start, real_len); |
68a1c816 PB |
873 | } |
874 | ||
992f48a0 | 875 | int target_munmap(abi_ulong start, abi_ulong len) |
54936004 | 876 | { |
912ff698 RH |
877 | int ret; |
878 | ||
b7b18d26 AB |
879 | trace_target_munmap(start, len); |
880 | ||
2b730f79 | 881 | if (start & ~TARGET_PAGE_MASK) { |
912ff698 RH |
882 | errno = EINVAL; |
883 | return -1; | |
2b730f79 | 884 | } |
54936004 | 885 | len = TARGET_PAGE_ALIGN(len); |
46b12f46 | 886 | if (len == 0 || !guest_range_valid_untagged(start, len)) { |
912ff698 RH |
887 | errno = EINVAL; |
888 | return -1; | |
ebf9a363 MF |
889 | } |
890 | ||
c8a706fe | 891 | mmap_lock(); |
912ff698 RH |
892 | ret = mmap_reserve_or_unmap(start, len); |
893 | if (likely(ret == 0)) { | |
894 | page_set_flags(start, start + len - 1, 0); | |
895 | shm_region_rm_complete(start, start + len - 1); | |
896 | } | |
c8a706fe | 897 | mmap_unlock(); |
d7b0c5d0 | 898 | |
912ff698 | 899 | return ret; |
54936004 FB |
900 | } |
901 | ||
992f48a0 BS |
902 | abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size, |
903 | abi_ulong new_size, unsigned long flags, | |
904 | abi_ulong new_addr) | |
54936004 FB |
905 | { |
906 | int prot; | |
f19412a2 | 907 | void *host_addr; |
54936004 | 908 | |
46b12f46 | 909 | if (!guest_range_valid_untagged(old_addr, old_size) || |
ebf9a363 | 910 | ((flags & MREMAP_FIXED) && |
46b12f46 | 911 | !guest_range_valid_untagged(new_addr, new_size)) || |
ccc5ccc1 | 912 | ((flags & MREMAP_MAYMOVE) == 0 && |
46b12f46 | 913 | !guest_range_valid_untagged(old_addr, new_size))) { |
ebf9a363 MF |
914 | errno = ENOMEM; |
915 | return -1; | |
916 | } | |
917 | ||
c8a706fe | 918 | mmap_lock(); |
f19412a2 | 919 | |
68a1c816 | 920 | if (flags & MREMAP_FIXED) { |
3e8f1628 RH |
921 | host_addr = mremap(g2h_untagged(old_addr), old_size, new_size, |
922 | flags, g2h_untagged(new_addr)); | |
68a1c816 | 923 | |
b76f21a7 | 924 | if (reserved_va && host_addr != MAP_FAILED) { |
2b730f79 RH |
925 | /* |
926 | * If new and old addresses overlap then the above mremap will | |
927 | * already have failed with EINVAL. | |
928 | */ | |
558a4411 | 929 | mmap_reserve_or_unmap(old_addr, old_size); |
68a1c816 PB |
930 | } |
931 | } else if (flags & MREMAP_MAYMOVE) { | |
f19412a2 AJ |
932 | abi_ulong mmap_start; |
933 | ||
30ab9ef2 | 934 | mmap_start = mmap_find_vma(0, new_size, TARGET_PAGE_SIZE); |
f19412a2 AJ |
935 | |
936 | if (mmap_start == -1) { | |
937 | errno = ENOMEM; | |
938 | host_addr = MAP_FAILED; | |
68a1c816 | 939 | } else { |
3e8f1628 RH |
940 | host_addr = mremap(g2h_untagged(old_addr), old_size, new_size, |
941 | flags | MREMAP_FIXED, | |
942 | g2h_untagged(mmap_start)); | |
b76f21a7 | 943 | if (reserved_va) { |
558a4411 | 944 | mmap_reserve_or_unmap(old_addr, old_size); |
c65ffe6d | 945 | } |
68a1c816 | 946 | } |
3af72a4d | 947 | } else { |
ea800033 | 948 | int page_flags = 0; |
b76f21a7 | 949 | if (reserved_va && old_size < new_size) { |
68a1c816 PB |
950 | abi_ulong addr; |
951 | for (addr = old_addr + old_size; | |
952 | addr < old_addr + new_size; | |
953 | addr++) { | |
ea800033 | 954 | page_flags |= page_get_flags(addr); |
68a1c816 PB |
955 | } |
956 | } | |
ea800033 | 957 | if (page_flags == 0) { |
3e8f1628 RH |
958 | host_addr = mremap(g2h_untagged(old_addr), |
959 | old_size, new_size, flags); | |
56d19084 TK |
960 | |
961 | if (host_addr != MAP_FAILED) { | |
962 | /* Check if address fits target address space */ | |
46b12f46 | 963 | if (!guest_range_valid_untagged(h2g(host_addr), new_size)) { |
56d19084 | 964 | /* Revert mremap() changes */ |
3e8f1628 RH |
965 | host_addr = mremap(g2h_untagged(old_addr), |
966 | new_size, old_size, flags); | |
56d19084 TK |
967 | errno = ENOMEM; |
968 | host_addr = MAP_FAILED; | |
969 | } else if (reserved_va && old_size > new_size) { | |
558a4411 RH |
970 | mmap_reserve_or_unmap(old_addr + old_size, |
971 | old_size - new_size); | |
56d19084 | 972 | } |
68a1c816 PB |
973 | } |
974 | } else { | |
975 | errno = ENOMEM; | |
976 | host_addr = MAP_FAILED; | |
977 | } | |
f19412a2 AJ |
978 | } |
979 | ||
980 | if (host_addr == MAP_FAILED) { | |
c8a706fe PB |
981 | new_addr = -1; |
982 | } else { | |
983 | new_addr = h2g(host_addr); | |
984 | prot = page_get_flags(old_addr); | |
49840a4a | 985 | page_set_flags(old_addr, old_addr + old_size - 1, 0); |
044e95c8 | 986 | shm_region_rm_complete(old_addr, old_addr + old_size - 1); |
49840a4a | 987 | page_set_flags(new_addr, new_addr + new_size - 1, |
d9c58585 | 988 | prot | PAGE_VALID | PAGE_RESET); |
044e95c8 | 989 | shm_region_rm_complete(new_addr, new_addr + new_size - 1); |
c8a706fe PB |
990 | } |
991 | mmap_unlock(); | |
54936004 FB |
992 | return new_addr; |
993 | } | |
892a4f6a | 994 | |
892a4f6a IL |
995 | abi_long target_madvise(abi_ulong start, abi_ulong len_in, int advice) |
996 | { | |
e230ec09 | 997 | abi_ulong len; |
892a4f6a IL |
998 | int ret = 0; |
999 | ||
1000 | if (start & ~TARGET_PAGE_MASK) { | |
1001 | return -TARGET_EINVAL; | |
1002 | } | |
e230ec09 | 1003 | if (len_in == 0) { |
892a4f6a IL |
1004 | return 0; |
1005 | } | |
e230ec09 RH |
1006 | len = TARGET_PAGE_ALIGN(len_in); |
1007 | if (len == 0 || !guest_range_valid_untagged(start, len)) { | |
892a4f6a IL |
1008 | return -TARGET_EINVAL; |
1009 | } | |
1010 | ||
4530deb1 HD |
1011 | /* Translate for some architectures which have different MADV_xxx values */ |
1012 | switch (advice) { | |
1013 | case TARGET_MADV_DONTNEED: /* alpha */ | |
1014 | advice = MADV_DONTNEED; | |
1015 | break; | |
1016 | case TARGET_MADV_WIPEONFORK: /* parisc */ | |
1017 | advice = MADV_WIPEONFORK; | |
1018 | break; | |
1019 | case TARGET_MADV_KEEPONFORK: /* parisc */ | |
1020 | advice = MADV_KEEPONFORK; | |
1021 | break; | |
1022 | /* we do not care about the other MADV_xxx values yet */ | |
1023 | } | |
1024 | ||
892a4f6a | 1025 | /* |
4530deb1 HD |
1026 | * Most advice values are hints, so ignoring and returning success is ok. |
1027 | * | |
1028 | * However, some advice values such as MADV_DONTNEED, MADV_WIPEONFORK and | |
1029 | * MADV_KEEPONFORK are not hints and need to be emulated. | |
892a4f6a | 1030 | * |
4530deb1 HD |
1031 | * A straight passthrough for those may not be safe because qemu sometimes |
1032 | * turns private file-backed mappings into anonymous mappings. | |
ecb796db RH |
1033 | * If all guest pages have PAGE_PASSTHROUGH set, mappings have the |
1034 | * same semantics for the host as for the guest. | |
892a4f6a | 1035 | * |
4530deb1 HD |
1036 | * We pass through MADV_WIPEONFORK and MADV_KEEPONFORK if possible and |
1037 | * return failure if not. | |
1038 | * | |
1039 | * MADV_DONTNEED is passed through as well, if possible. | |
1040 | * If passthrough isn't possible, we nevertheless (wrongly!) return | |
1041 | * success, which is broken but some userspace programs fail to work | |
1042 | * otherwise. Completely implementing such emulation is quite complicated | |
1043 | * though. | |
892a4f6a IL |
1044 | */ |
1045 | mmap_lock(); | |
4530deb1 HD |
1046 | switch (advice) { |
1047 | case MADV_WIPEONFORK: | |
1048 | case MADV_KEEPONFORK: | |
1049 | ret = -EINVAL; | |
1050 | /* fall through */ | |
1051 | case MADV_DONTNEED: | |
ecb796db | 1052 | if (page_check_range(start, len, PAGE_PASSTHROUGH)) { |
4530deb1 HD |
1053 | ret = get_errno(madvise(g2h_untagged(start), len, advice)); |
1054 | if ((advice == MADV_DONTNEED) && (ret == 0)) { | |
10310cbd | 1055 | page_reset_target_data(start, start + len - 1); |
4530deb1 | 1056 | } |
dbbf8975 | 1057 | } |
892a4f6a IL |
1058 | } |
1059 | mmap_unlock(); | |
1060 | ||
1061 | return ret; | |
1062 | } | |
225a206c RH |
1063 | |
1064 | #ifndef TARGET_FORCE_SHMLBA | |
1065 | /* | |
1066 | * For most architectures, SHMLBA is the same as the page size; | |
1067 | * some architectures have larger values, in which case they should | |
1068 | * define TARGET_FORCE_SHMLBA and provide a target_shmlba() function. | |
1069 | * This corresponds to the kernel arch code defining __ARCH_FORCE_SHMLBA | |
1070 | * and defining its own value for SHMLBA. | |
1071 | * | |
1072 | * The kernel also permits SHMLBA to be set by the architecture to a | |
1073 | * value larger than the page size without setting __ARCH_FORCE_SHMLBA; | |
1074 | * this means that addresses are rounded to the large size if | |
1075 | * SHM_RND is set but addresses not aligned to that size are not rejected | |
1076 | * as long as they are at least page-aligned. Since the only architecture | |
1077 | * which uses this is ia64 this code doesn't provide for that oddity. | |
1078 | */ | |
1079 | static inline abi_ulong target_shmlba(CPUArchState *cpu_env) | |
1080 | { | |
1081 | return TARGET_PAGE_SIZE; | |
1082 | } | |
1083 | #endif | |
1084 | ||
1085 | abi_ulong target_shmat(CPUArchState *cpu_env, int shmid, | |
1086 | abi_ulong shmaddr, int shmflg) | |
1087 | { | |
1088 | CPUState *cpu = env_cpu(cpu_env); | |
1089 | abi_ulong raddr; | |
225a206c | 1090 | struct shmid_ds shm_info; |
69fa2708 | 1091 | int ret; |
225a206c RH |
1092 | abi_ulong shmlba; |
1093 | ||
1094 | /* shmat pointers are always untagged */ | |
1095 | ||
1096 | /* find out the length of the shared memory segment */ | |
1097 | ret = get_errno(shmctl(shmid, IPC_STAT, &shm_info)); | |
1098 | if (is_error(ret)) { | |
1099 | /* can't get length, bail out */ | |
1100 | return ret; | |
1101 | } | |
1102 | ||
1103 | shmlba = target_shmlba(cpu_env); | |
1104 | ||
1105 | if (shmaddr & (shmlba - 1)) { | |
1106 | if (shmflg & SHM_RND) { | |
1107 | shmaddr &= ~(shmlba - 1); | |
1108 | } else { | |
1109 | return -TARGET_EINVAL; | |
1110 | } | |
1111 | } | |
1112 | if (!guest_range_valid_untagged(shmaddr, shm_info.shm_segsz)) { | |
1113 | return -TARGET_EINVAL; | |
1114 | } | |
1115 | ||
69fa2708 RH |
1116 | WITH_MMAP_LOCK_GUARD() { |
1117 | void *host_raddr; | |
044e95c8 | 1118 | abi_ulong last; |
69fa2708 RH |
1119 | |
1120 | if (shmaddr) { | |
1121 | host_raddr = shmat(shmid, (void *)g2h_untagged(shmaddr), shmflg); | |
1122 | } else { | |
1123 | abi_ulong mmap_start; | |
1124 | ||
1125 | /* In order to use the host shmat, we need to honor host SHMLBA. */ | |
1126 | mmap_start = mmap_find_vma(0, shm_info.shm_segsz, | |
1127 | MAX(SHMLBA, shmlba)); | |
1128 | ||
1129 | if (mmap_start == -1) { | |
1130 | return -TARGET_ENOMEM; | |
1131 | } | |
1132 | host_raddr = shmat(shmid, g2h_untagged(mmap_start), | |
1133 | shmflg | SHM_REMAP); | |
1134 | } | |
1135 | ||
1136 | if (host_raddr == (void *)-1) { | |
1137 | return get_errno(-1); | |
1138 | } | |
1139 | raddr = h2g(host_raddr); | |
044e95c8 | 1140 | last = raddr + shm_info.shm_segsz - 1; |
69fa2708 | 1141 | |
044e95c8 | 1142 | page_set_flags(raddr, last, |
69fa2708 RH |
1143 | PAGE_VALID | PAGE_RESET | PAGE_READ | |
1144 | (shmflg & SHM_RDONLY ? 0 : PAGE_WRITE)); | |
1145 | ||
044e95c8 RH |
1146 | shm_region_rm_complete(raddr, last); |
1147 | shm_region_add(raddr, last); | |
69fa2708 | 1148 | } |
225a206c RH |
1149 | |
1150 | /* | |
1151 | * We're mapping shared memory, so ensure we generate code for parallel | |
1152 | * execution and flush old translations. This will work up to the level | |
1153 | * supported by the host -- anything that requires EXCP_ATOMIC will not | |
1154 | * be atomic with respect to an external process. | |
1155 | */ | |
1156 | if (!(cpu->tcg_cflags & CF_PARALLEL)) { | |
1157 | cpu->tcg_cflags |= CF_PARALLEL; | |
1158 | tb_flush(cpu); | |
1159 | } | |
1160 | ||
225a206c RH |
1161 | return raddr; |
1162 | } | |
1163 | ||
1164 | abi_long target_shmdt(abi_ulong shmaddr) | |
1165 | { | |
225a206c RH |
1166 | abi_long rv; |
1167 | ||
1168 | /* shmdt pointers are always untagged */ | |
1169 | ||
69fa2708 | 1170 | WITH_MMAP_LOCK_GUARD() { |
044e95c8 RH |
1171 | abi_ulong last = shm_region_find(shmaddr); |
1172 | if (last == 0) { | |
ceda5688 RH |
1173 | return -TARGET_EINVAL; |
1174 | } | |
1175 | ||
69fa2708 | 1176 | rv = get_errno(shmdt(g2h_untagged(shmaddr))); |
ceda5688 | 1177 | if (rv == 0) { |
044e95c8 | 1178 | abi_ulong size = last - shmaddr + 1; |
ceda5688 | 1179 | |
044e95c8 RH |
1180 | page_set_flags(shmaddr, last, 0); |
1181 | shm_region_rm_complete(shmaddr, last); | |
ceda5688 RH |
1182 | mmap_reserve_or_unmap(shmaddr, size); |
1183 | } | |
225a206c | 1184 | } |
225a206c RH |
1185 | return rv; |
1186 | } |