]>
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" |
54936004 | 27 | |
1e6eec8b | 28 | static pthread_mutex_t mmap_mutex = PTHREAD_MUTEX_INITIALIZER; |
dfd3f85c | 29 | static __thread int mmap_lock_count; |
c8a706fe | 30 | |
225a206c RH |
31 | #define N_SHM_REGIONS 32 |
32 | ||
33 | static struct shm_region { | |
34 | abi_ulong start; | |
35 | abi_ulong size; | |
36 | bool in_use; | |
37 | } shm_regions[N_SHM_REGIONS]; | |
38 | ||
c8a706fe PB |
39 | void mmap_lock(void) |
40 | { | |
41 | if (mmap_lock_count++ == 0) { | |
42 | pthread_mutex_lock(&mmap_mutex); | |
43 | } | |
44 | } | |
45 | ||
46 | void mmap_unlock(void) | |
47 | { | |
990ef918 | 48 | assert(mmap_lock_count > 0); |
c8a706fe PB |
49 | if (--mmap_lock_count == 0) { |
50 | pthread_mutex_unlock(&mmap_mutex); | |
51 | } | |
52 | } | |
d5975363 | 53 | |
301e40ed AB |
54 | bool have_mmap_lock(void) |
55 | { | |
56 | return mmap_lock_count > 0 ? true : false; | |
57 | } | |
58 | ||
d5975363 PB |
59 | /* Grab lock to make sure things are in a consistent state after fork(). */ |
60 | void mmap_fork_start(void) | |
61 | { | |
62 | if (mmap_lock_count) | |
63 | abort(); | |
64 | pthread_mutex_lock(&mmap_mutex); | |
65 | } | |
66 | ||
67 | void mmap_fork_end(int child) | |
68 | { | |
2b730f79 | 69 | if (child) { |
d5975363 | 70 | pthread_mutex_init(&mmap_mutex, NULL); |
2b730f79 | 71 | } else { |
d5975363 | 72 | pthread_mutex_unlock(&mmap_mutex); |
2b730f79 | 73 | } |
d5975363 | 74 | } |
c8a706fe | 75 | |
9dba3ca5 RH |
76 | /* |
77 | * Validate target prot bitmask. | |
78 | * Return the prot bitmask for the host in *HOST_PROT. | |
79 | * Return 0 if the target prot bitmask is invalid, otherwise | |
80 | * the internal qemu page_flags (which will include PAGE_VALID). | |
81 | */ | |
0dd55812 | 82 | static int validate_prot_to_pageflags(int prot) |
9dba3ca5 RH |
83 | { |
84 | int valid = PROT_READ | PROT_WRITE | PROT_EXEC | TARGET_PROT_SEM; | |
85 | int page_flags = (prot & PAGE_BITS) | PAGE_VALID; | |
86 | ||
be5d6f48 | 87 | #ifdef TARGET_AARCH64 |
d109b46d | 88 | { |
be5d6f48 | 89 | ARMCPU *cpu = ARM_CPU(thread_cpu); |
d109b46d RH |
90 | |
91 | /* | |
92 | * The PROT_BTI bit is only accepted if the cpu supports the feature. | |
93 | * Since this is the unusual case, don't bother checking unless | |
94 | * the bit has been requested. If set and valid, record the bit | |
95 | * within QEMU's page_flags. | |
96 | */ | |
97 | if ((prot & TARGET_PROT_BTI) && cpu_isar_feature(aa64_bti, cpu)) { | |
be5d6f48 RH |
98 | valid |= TARGET_PROT_BTI; |
99 | page_flags |= PAGE_BTI; | |
100 | } | |
d109b46d RH |
101 | /* Similarly for the PROT_MTE bit. */ |
102 | if ((prot & TARGET_PROT_MTE) && cpu_isar_feature(aa64_mte, cpu)) { | |
103 | valid |= TARGET_PROT_MTE; | |
104 | page_flags |= PAGE_MTE; | |
105 | } | |
be5d6f48 | 106 | } |
4c184e70 HD |
107 | #elif defined(TARGET_HPPA) |
108 | valid |= PROT_GROWSDOWN | PROT_GROWSUP; | |
be5d6f48 RH |
109 | #endif |
110 | ||
9dba3ca5 RH |
111 | return prot & ~valid ? 0 : page_flags; |
112 | } | |
113 | ||
0dd55812 RH |
114 | /* |
115 | * For the host, we need not pass anything except read/write/exec. | |
116 | * While PROT_SEM is allowed by all hosts, it is also ignored, so | |
117 | * don't bother transforming guest bit to host bit. Any other | |
118 | * target-specific prot bits will not be understood by the host | |
119 | * and will need to be encoded into page_flags for qemu emulation. | |
120 | * | |
121 | * Pages that are executable by the guest will never be executed | |
122 | * by the host, but the host will need to be able to read them. | |
123 | */ | |
124 | static int target_to_host_prot(int prot) | |
125 | { | |
126 | return (prot & (PROT_READ | PROT_WRITE)) | | |
127 | (prot & PROT_EXEC ? PROT_READ : 0); | |
128 | } | |
129 | ||
53a5960a | 130 | /* NOTE: all the constants are the HOST ones, but addresses are target. */ |
9dba3ca5 | 131 | int target_mprotect(abi_ulong start, abi_ulong len, int target_prot) |
54936004 | 132 | { |
7bdc1acc RH |
133 | abi_ulong starts[3]; |
134 | abi_ulong lens[3]; | |
135 | int prots[3]; | |
136 | abi_ulong host_start, host_last, last; | |
137 | int prot1, ret, page_flags, nranges; | |
54936004 | 138 | |
9dba3ca5 | 139 | trace_target_mprotect(start, len, target_prot); |
54936004 | 140 | |
9dba3ca5 | 141 | if ((start & ~TARGET_PAGE_MASK) != 0) { |
78cf3390 | 142 | return -TARGET_EINVAL; |
9dba3ca5 | 143 | } |
0dd55812 | 144 | page_flags = validate_prot_to_pageflags(target_prot); |
9dba3ca5 RH |
145 | if (!page_flags) { |
146 | return -TARGET_EINVAL; | |
147 | } | |
7bdc1acc RH |
148 | if (len == 0) { |
149 | return 0; | |
150 | } | |
54936004 | 151 | len = TARGET_PAGE_ALIGN(len); |
46b12f46 | 152 | if (!guest_range_valid_untagged(start, len)) { |
78cf3390 | 153 | return -TARGET_ENOMEM; |
ebf9a363 | 154 | } |
3b46e624 | 155 | |
7bdc1acc | 156 | last = start + len - 1; |
83fb7adf | 157 | host_start = start & qemu_host_page_mask; |
7bdc1acc RH |
158 | host_last = HOST_PAGE_ALIGN(last) - 1; |
159 | nranges = 0; | |
160 | ||
161 | mmap_lock(); | |
162 | ||
163 | if (host_last - host_start < qemu_host_page_size) { | |
164 | /* Single host page contains all guest pages: sum the prot. */ | |
0dd55812 | 165 | prot1 = target_prot; |
7bdc1acc RH |
166 | for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) { |
167 | prot1 |= page_get_flags(a); | |
54936004 | 168 | } |
7bdc1acc RH |
169 | for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) { |
170 | prot1 |= page_get_flags(a + 1); | |
d418c81e | 171 | } |
7bdc1acc RH |
172 | starts[nranges] = host_start; |
173 | lens[nranges] = qemu_host_page_size; | |
174 | prots[nranges] = prot1; | |
175 | nranges++; | |
176 | } else { | |
177 | if (host_start < start) { | |
178 | /* Host page contains more than one guest page: sum the prot. */ | |
179 | prot1 = target_prot; | |
180 | for (abi_ulong a = host_start; a < start; a += TARGET_PAGE_SIZE) { | |
181 | prot1 |= page_get_flags(a); | |
182 | } | |
183 | /* If the resulting sum differs, create a new range. */ | |
184 | if (prot1 != target_prot) { | |
185 | starts[nranges] = host_start; | |
186 | lens[nranges] = qemu_host_page_size; | |
187 | prots[nranges] = prot1; | |
188 | nranges++; | |
189 | host_start += qemu_host_page_size; | |
190 | } | |
9dba3ca5 | 191 | } |
7bdc1acc RH |
192 | |
193 | if (last < host_last) { | |
194 | /* Host page contains more than one guest page: sum the prot. */ | |
195 | prot1 = target_prot; | |
196 | for (abi_ulong a = last; a < host_last; a += TARGET_PAGE_SIZE) { | |
197 | prot1 |= page_get_flags(a + 1); | |
198 | } | |
199 | /* If the resulting sum differs, create a new range. */ | |
200 | if (prot1 != target_prot) { | |
201 | host_last -= qemu_host_page_size; | |
202 | starts[nranges] = host_last + 1; | |
203 | lens[nranges] = qemu_host_page_size; | |
204 | prots[nranges] = prot1; | |
205 | nranges++; | |
206 | } | |
54936004 | 207 | } |
7bdc1acc RH |
208 | |
209 | /* Create a range for the middle, if any remains. */ | |
210 | if (host_start < host_last) { | |
211 | starts[nranges] = host_start; | |
212 | lens[nranges] = host_last - host_start + 1; | |
213 | prots[nranges] = target_prot; | |
214 | nranges++; | |
9dba3ca5 | 215 | } |
54936004 | 216 | } |
3b46e624 | 217 | |
7bdc1acc RH |
218 | for (int i = 0; i < nranges; ++i) { |
219 | ret = mprotect(g2h_untagged(starts[i]), lens[i], | |
220 | target_to_host_prot(prots[i])); | |
9dba3ca5 | 221 | if (ret != 0) { |
c8a706fe | 222 | goto error; |
9dba3ca5 | 223 | } |
54936004 | 224 | } |
aa98e2d8 | 225 | |
7bdc1acc | 226 | page_set_flags(start, last, page_flags); |
aa98e2d8 IL |
227 | ret = 0; |
228 | ||
7bdc1acc | 229 | error: |
c8a706fe PB |
230 | mmap_unlock(); |
231 | return ret; | |
54936004 FB |
232 | } |
233 | ||
234 | /* map an incomplete host page */ | |
99982beb RH |
235 | static bool mmap_frag(abi_ulong real_start, abi_ulong start, abi_ulong last, |
236 | int prot, int flags, int fd, off_t offset) | |
54936004 | 237 | { |
99982beb | 238 | abi_ulong real_last; |
53a5960a | 239 | void *host_start; |
99982beb RH |
240 | int prot_old, prot_new; |
241 | int host_prot_old, host_prot_new; | |
54936004 | 242 | |
99982beb RH |
243 | if (!(flags & MAP_ANONYMOUS) |
244 | && (flags & MAP_TYPE) == MAP_SHARED | |
245 | && (prot & PROT_WRITE)) { | |
246 | /* | |
247 | * msync() won't work with the partial page, so we return an | |
248 | * error if write is possible while it is a shared mapping. | |
249 | */ | |
250 | errno = EINVAL; | |
251 | return false; | |
252 | } | |
253 | ||
254 | real_last = real_start + qemu_host_page_size - 1; | |
3e8f1628 | 255 | host_start = g2h_untagged(real_start); |
54936004 | 256 | |
99982beb RH |
257 | /* Get the protection of the target pages outside the mapping. */ |
258 | prot_old = 0; | |
259 | for (abi_ulong a = real_start; a < start; a += TARGET_PAGE_SIZE) { | |
260 | prot_old |= page_get_flags(a); | |
261 | } | |
262 | for (abi_ulong a = real_last; a > last; a -= TARGET_PAGE_SIZE) { | |
263 | prot_old |= page_get_flags(a); | |
54936004 | 264 | } |
3b46e624 | 265 | |
99982beb RH |
266 | if (prot_old == 0) { |
267 | /* | |
268 | * Since !(prot_old & PAGE_VALID), there were no guest pages | |
269 | * outside of the fragment we need to map. Allocate a new host | |
270 | * page to cover, discarding whatever else may have been present. | |
271 | */ | |
0dd55812 RH |
272 | void *p = mmap(host_start, qemu_host_page_size, |
273 | target_to_host_prot(prot), | |
80210bcd | 274 | flags | MAP_ANONYMOUS, -1, 0); |
ddcdd8c4 AO |
275 | if (p != host_start) { |
276 | if (p != MAP_FAILED) { | |
277 | munmap(p, qemu_host_page_size); | |
278 | errno = EEXIST; | |
279 | } | |
99982beb | 280 | return false; |
2b730f79 | 281 | } |
99982beb | 282 | prot_old = prot; |
54936004 | 283 | } |
99982beb | 284 | prot_new = prot | prot_old; |
54936004 | 285 | |
99982beb RH |
286 | host_prot_old = target_to_host_prot(prot_old); |
287 | host_prot_new = target_to_host_prot(prot_new); | |
3b46e624 | 288 | |
99982beb RH |
289 | /* Adjust protection to be able to write. */ |
290 | if (!(host_prot_old & PROT_WRITE)) { | |
291 | host_prot_old |= PROT_WRITE; | |
292 | mprotect(host_start, qemu_host_page_size, host_prot_old); | |
293 | } | |
3b46e624 | 294 | |
99982beb RH |
295 | /* Read or zero the new guest pages. */ |
296 | if (flags & MAP_ANONYMOUS) { | |
297 | memset(g2h_untagged(start), 0, last - start + 1); | |
54936004 | 298 | } else { |
99982beb RH |
299 | if (pread(fd, g2h_untagged(start), last - start + 1, offset) == -1) { |
300 | return false; | |
e6deac9c | 301 | } |
54936004 | 302 | } |
99982beb RH |
303 | |
304 | /* Put final protection */ | |
305 | if (host_prot_new != host_prot_old) { | |
306 | mprotect(host_start, qemu_host_page_size, host_prot_new); | |
307 | } | |
308 | return true; | |
54936004 FB |
309 | } |
310 | ||
c8fb5cf9 | 311 | abi_ulong task_unmapped_base; |
da2b71fa | 312 | abi_ulong elf_et_dyn_base; |
c8fb5cf9 | 313 | abi_ulong mmap_next_start; |
a03e2d42 | 314 | |
2b730f79 RH |
315 | /* |
316 | * Subroutine of mmap_find_vma, used when we have pre-allocated | |
317 | * a chunk of guest address space. | |
318 | */ | |
30ab9ef2 RH |
319 | static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size, |
320 | abi_ulong align) | |
68a1c816 | 321 | { |
4c13048e | 322 | target_ulong ret; |
68a1c816 | 323 | |
4c13048e RH |
324 | ret = page_find_range_empty(start, reserved_va, size, align); |
325 | if (ret == -1 && start > mmap_min_addr) { | |
326 | /* Restart at the beginning of the address space. */ | |
327 | ret = page_find_range_empty(mmap_min_addr, start - 1, size, align); | |
68a1c816 PB |
328 | } |
329 | ||
4c13048e | 330 | return ret; |
68a1c816 PB |
331 | } |
332 | ||
fe3b4152 KS |
333 | /* |
334 | * Find and reserve a free memory area of size 'size'. The search | |
335 | * starts at 'start'. | |
336 | * It must be called with mmap_lock() held. | |
337 | * Return -1 if error. | |
338 | */ | |
30ab9ef2 | 339 | abi_ulong mmap_find_vma(abi_ulong start, abi_ulong size, abi_ulong align) |
a03e2d42 | 340 | { |
14f24e14 | 341 | void *ptr, *prev; |
fe3b4152 | 342 | abi_ulong addr; |
14f24e14 | 343 | int wrapped, repeat; |
fe3b4152 | 344 | |
443b7505 RH |
345 | align = MAX(align, qemu_host_page_size); |
346 | ||
fe3b4152 | 347 | /* If 'start' == 0, then a default start address is used. */ |
14f24e14 | 348 | if (start == 0) { |
fe3b4152 | 349 | start = mmap_next_start; |
14f24e14 RH |
350 | } else { |
351 | start &= qemu_host_page_mask; | |
352 | } | |
30ab9ef2 | 353 | start = ROUND_UP(start, align); |
14f24e14 RH |
354 | |
355 | size = HOST_PAGE_ALIGN(size); | |
fe3b4152 | 356 | |
b76f21a7 | 357 | if (reserved_va) { |
30ab9ef2 | 358 | return mmap_find_vma_reserved(start, size, align); |
68a1c816 PB |
359 | } |
360 | ||
a03e2d42 | 361 | addr = start; |
14f24e14 RH |
362 | wrapped = repeat = 0; |
363 | prev = 0; | |
fe3b4152 | 364 | |
14f24e14 | 365 | for (;; prev = ptr) { |
fe3b4152 KS |
366 | /* |
367 | * Reserve needed memory area to avoid a race. | |
368 | * It should be discarded using: | |
369 | * - mmap() with MAP_FIXED flag | |
370 | * - mremap() with MREMAP_FIXED flag | |
371 | * - shmat() with SHM_REMAP flag | |
372 | */ | |
3e8f1628 | 373 | ptr = mmap(g2h_untagged(addr), size, PROT_NONE, |
2b730f79 | 374 | MAP_ANONYMOUS | MAP_PRIVATE | MAP_NORESERVE, -1, 0); |
fe3b4152 KS |
375 | |
376 | /* ENOMEM, if host address space has no memory */ | |
14f24e14 | 377 | if (ptr == MAP_FAILED) { |
fe3b4152 | 378 | return (abi_ulong)-1; |
14f24e14 RH |
379 | } |
380 | ||
2b730f79 RH |
381 | /* |
382 | * Count the number of sequential returns of the same address. | |
383 | * This is used to modify the search algorithm below. | |
384 | */ | |
14f24e14 RH |
385 | repeat = (ptr == prev ? repeat + 1 : 0); |
386 | ||
387 | if (h2g_valid(ptr + size - 1)) { | |
388 | addr = h2g(ptr); | |
fe3b4152 | 389 | |
30ab9ef2 | 390 | if ((addr & (align - 1)) == 0) { |
14f24e14 | 391 | /* Success. */ |
c8fb5cf9 | 392 | if (start == mmap_next_start && addr >= task_unmapped_base) { |
14f24e14 RH |
393 | mmap_next_start = addr + size; |
394 | } | |
395 | return addr; | |
396 | } | |
fe3b4152 | 397 | |
14f24e14 RH |
398 | /* The address is not properly aligned for the target. */ |
399 | switch (repeat) { | |
400 | case 0: | |
2b730f79 RH |
401 | /* |
402 | * Assume the result that the kernel gave us is the | |
403 | * first with enough free space, so start again at the | |
404 | * next higher target page. | |
405 | */ | |
30ab9ef2 | 406 | addr = ROUND_UP(addr, align); |
14f24e14 RH |
407 | break; |
408 | case 1: | |
2b730f79 RH |
409 | /* |
410 | * Sometimes the kernel decides to perform the allocation | |
411 | * at the top end of memory instead. | |
412 | */ | |
30ab9ef2 | 413 | addr &= -align; |
14f24e14 RH |
414 | break; |
415 | case 2: | |
416 | /* Start over at low memory. */ | |
417 | addr = 0; | |
418 | break; | |
419 | default: | |
420 | /* Fail. This unaligned block must the last. */ | |
421 | addr = -1; | |
422 | break; | |
423 | } | |
424 | } else { | |
2b730f79 RH |
425 | /* |
426 | * Since the result the kernel gave didn't fit, start | |
427 | * again at low memory. If any repetition, fail. | |
428 | */ | |
14f24e14 RH |
429 | addr = (repeat ? -1 : 0); |
430 | } | |
431 | ||
432 | /* Unmap and try again. */ | |
fe3b4152 | 433 | munmap(ptr, size); |
fe3b4152 | 434 | |
14f24e14 | 435 | /* ENOMEM if we checked the whole of the target address space. */ |
d0b3e4f5 | 436 | if (addr == (abi_ulong)-1) { |
a03e2d42 | 437 | return (abi_ulong)-1; |
14f24e14 RH |
438 | } else if (addr == 0) { |
439 | if (wrapped) { | |
440 | return (abi_ulong)-1; | |
441 | } | |
442 | wrapped = 1; | |
2b730f79 RH |
443 | /* |
444 | * Don't actually use 0 when wrapping, instead indicate | |
445 | * that we'd truly like an allocation in low memory. | |
446 | */ | |
14f24e14 RH |
447 | addr = (mmap_min_addr > TARGET_PAGE_SIZE |
448 | ? TARGET_PAGE_ALIGN(mmap_min_addr) | |
449 | : TARGET_PAGE_SIZE); | |
450 | } else if (wrapped && addr >= start) { | |
451 | return (abi_ulong)-1; | |
452 | } | |
a03e2d42 | 453 | } |
a03e2d42 FB |
454 | } |
455 | ||
54936004 | 456 | /* NOTE: all the constants are the HOST ones */ |
9dba3ca5 | 457 | abi_long target_mmap(abi_ulong start, abi_ulong len, int target_prot, |
55baec0f | 458 | int flags, int fd, off_t offset) |
54936004 | 459 | { |
f9cd8f5e RH |
460 | abi_ulong ret, last, real_start, real_last, retaddr, host_len; |
461 | abi_ulong passthrough_start = -1, passthrough_last = 0; | |
0dd55812 | 462 | int page_flags; |
55baec0f | 463 | off_t host_offset; |
54936004 | 464 | |
c8a706fe | 465 | mmap_lock(); |
9dba3ca5 | 466 | trace_target_mmap(start, len, target_prot, flags, fd, offset); |
54936004 | 467 | |
38138fab | 468 | if (!len) { |
e89f07d3 | 469 | errno = EINVAL; |
c8a706fe | 470 | goto fail; |
e89f07d3 | 471 | } |
54936004 | 472 | |
0dd55812 | 473 | page_flags = validate_prot_to_pageflags(target_prot); |
9dba3ca5 RH |
474 | if (!page_flags) { |
475 | errno = EINVAL; | |
476 | goto fail; | |
477 | } | |
478 | ||
38138fab | 479 | /* Also check for overflows... */ |
54936004 | 480 | len = TARGET_PAGE_ALIGN(len); |
38138fab AB |
481 | if (!len) { |
482 | errno = ENOMEM; | |
483 | goto fail; | |
484 | } | |
485 | ||
486 | if (offset & ~TARGET_PAGE_MASK) { | |
487 | errno = EINVAL; | |
488 | goto fail; | |
489 | } | |
490 | ||
228168cb RH |
491 | /* |
492 | * If we're mapping shared memory, ensure we generate code for parallel | |
493 | * execution and flush old translations. This will work up to the level | |
494 | * supported by the host -- anything that requires EXCP_ATOMIC will not | |
495 | * be atomic with respect to an external process. | |
496 | */ | |
497 | if (flags & MAP_SHARED) { | |
498 | CPUState *cpu = thread_cpu; | |
499 | if (!(cpu->tcg_cflags & CF_PARALLEL)) { | |
500 | cpu->tcg_cflags |= CF_PARALLEL; | |
501 | tb_flush(cpu); | |
502 | } | |
503 | } | |
504 | ||
53a5960a | 505 | real_start = start & qemu_host_page_mask; |
a5e7ee46 RH |
506 | host_offset = offset & qemu_host_page_mask; |
507 | ||
2b730f79 RH |
508 | /* |
509 | * If the user is asking for the kernel to find a location, do that | |
510 | * before we truncate the length for mapping files below. | |
511 | */ | |
03798605 | 512 | if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) { |
a5e7ee46 RH |
513 | host_len = len + offset - host_offset; |
514 | host_len = HOST_PAGE_ALIGN(host_len); | |
30ab9ef2 | 515 | start = mmap_find_vma(real_start, host_len, TARGET_PAGE_SIZE); |
a5e7ee46 RH |
516 | if (start == (abi_ulong)-1) { |
517 | errno = ENOMEM; | |
518 | goto fail; | |
519 | } | |
520 | } | |
54936004 | 521 | |
2b730f79 RH |
522 | /* |
523 | * When mapping files into a memory area larger than the file, accesses | |
524 | * to pages beyond the file size will cause a SIGBUS. | |
525 | * | |
526 | * For example, if mmaping a file of 100 bytes on a host with 4K pages | |
527 | * emulating a target with 8K pages, the target expects to be able to | |
528 | * access the first 8K. But the host will trap us on any access beyond | |
529 | * 4K. | |
530 | * | |
531 | * When emulating a target with a larger page-size than the hosts, we | |
532 | * may need to truncate file maps at EOF and add extra anonymous pages | |
533 | * up to the targets page boundary. | |
534 | */ | |
8e3b0cbb | 535 | if ((qemu_real_host_page_size() < qemu_host_page_size) && |
35f2fd04 MAL |
536 | !(flags & MAP_ANONYMOUS)) { |
537 | struct stat sb; | |
54c5a2ae | 538 | |
2b730f79 RH |
539 | if (fstat(fd, &sb) == -1) { |
540 | goto fail; | |
541 | } | |
54c5a2ae | 542 | |
2b730f79 RH |
543 | /* Are we trying to create a map beyond EOF?. */ |
544 | if (offset + len > sb.st_size) { | |
545 | /* | |
546 | * If so, truncate the file map at eof aligned with | |
547 | * the hosts real pagesize. Additional anonymous maps | |
548 | * will be created beyond EOF. | |
549 | */ | |
550 | len = REAL_HOST_PAGE_ALIGN(sb.st_size - offset); | |
551 | } | |
54c5a2ae EI |
552 | } |
553 | ||
03798605 | 554 | if (!(flags & (MAP_FIXED | MAP_FIXED_NOREPLACE))) { |
55baec0f | 555 | uintptr_t host_start; |
0dd55812 | 556 | int host_prot; |
a03e2d42 | 557 | void *p; |
a5e7ee46 | 558 | |
a03e2d42 FB |
559 | host_len = len + offset - host_offset; |
560 | host_len = HOST_PAGE_ALIGN(host_len); | |
0dd55812 | 561 | host_prot = target_to_host_prot(target_prot); |
a5e7ee46 | 562 | |
2b730f79 RH |
563 | /* |
564 | * Note: we prefer to control the mapping address. It is | |
565 | * especially important if qemu_host_page_size > | |
566 | * qemu_real_host_page_size. | |
567 | */ | |
3e8f1628 | 568 | p = mmap(g2h_untagged(start), host_len, host_prot, |
a5e7ee46 | 569 | flags | MAP_FIXED | MAP_ANONYMOUS, -1, 0); |
9dba3ca5 | 570 | if (p == MAP_FAILED) { |
c8a706fe | 571 | goto fail; |
9dba3ca5 | 572 | } |
a03e2d42 | 573 | /* update start so that it points to the file position at 'offset' */ |
55baec0f | 574 | host_start = (uintptr_t)p; |
54c5a2ae | 575 | if (!(flags & MAP_ANONYMOUS)) { |
3e8f1628 | 576 | p = mmap(g2h_untagged(start), len, host_prot, |
54c5a2ae | 577 | flags | MAP_FIXED, fd, host_offset); |
8384274e | 578 | if (p == MAP_FAILED) { |
3e8f1628 | 579 | munmap(g2h_untagged(start), host_len); |
8384274e JB |
580 | goto fail; |
581 | } | |
a03e2d42 | 582 | host_start += offset - host_offset; |
54c5a2ae | 583 | } |
a03e2d42 | 584 | start = h2g(host_start); |
f9cd8f5e | 585 | last = start + len - 1; |
f93b7695 | 586 | passthrough_start = start; |
f9cd8f5e | 587 | passthrough_last = last; |
a03e2d42 FB |
588 | } else { |
589 | if (start & ~TARGET_PAGE_MASK) { | |
e89f07d3 | 590 | errno = EINVAL; |
c8a706fe | 591 | goto fail; |
e89f07d3 | 592 | } |
f9cd8f5e RH |
593 | last = start + len - 1; |
594 | real_last = HOST_PAGE_ALIGN(last) - 1; | |
7ab240ad | 595 | |
7d37435b PB |
596 | /* |
597 | * Test if requested memory area fits target address space | |
598 | * It can fail only on 64-bit host with 32-bit target. | |
599 | * On any other target/host host mmap() handles this error correctly. | |
600 | */ | |
f9cd8f5e | 601 | if (last < start || !guest_range_valid_untagged(start, len)) { |
ebf9a363 | 602 | errno = ENOMEM; |
45bc1f52 AJ |
603 | goto fail; |
604 | } | |
605 | ||
c3dd50da AO |
606 | if (flags & MAP_FIXED_NOREPLACE) { |
607 | /* Validate that the chosen range is empty. */ | |
608 | if (!page_check_range_empty(start, last)) { | |
609 | errno = EEXIST; | |
610 | goto fail; | |
611 | } | |
612 | ||
613 | /* | |
614 | * With reserved_va, the entire address space is mmaped in the | |
615 | * host to ensure it isn't accidentally used for something else. | |
616 | * We have just checked that the guest address is not mapped | |
617 | * within the guest, but need to replace the host reservation. | |
618 | * | |
619 | * Without reserved_va, despite the guest address check above, | |
620 | * keep MAP_FIXED_NOREPLACE so that the guest does not overwrite | |
621 | * any host address mappings. | |
622 | */ | |
623 | if (reserved_va) { | |
624 | flags = (flags & ~MAP_FIXED_NOREPLACE) | MAP_FIXED; | |
625 | } | |
03798605 RH |
626 | } |
627 | ||
2b730f79 RH |
628 | /* |
629 | * worst case: we cannot map the file because the offset is not | |
630 | * aligned, so we read it | |
631 | */ | |
a03e2d42 FB |
632 | if (!(flags & MAP_ANONYMOUS) && |
633 | (offset & ~qemu_host_page_mask) != (start & ~qemu_host_page_mask)) { | |
2b730f79 RH |
634 | /* |
635 | * msync() won't work here, so we return an error if write is | |
636 | * possible while it is a shared mapping | |
637 | */ | |
0dd55812 RH |
638 | if ((flags & MAP_TYPE) == MAP_SHARED |
639 | && (target_prot & PROT_WRITE)) { | |
a03e2d42 | 640 | errno = EINVAL; |
c8a706fe | 641 | goto fail; |
a03e2d42 | 642 | } |
9dba3ca5 | 643 | retaddr = target_mmap(start, len, target_prot | PROT_WRITE, |
03798605 RH |
644 | (flags & (MAP_FIXED | MAP_FIXED_NOREPLACE)) |
645 | | MAP_PRIVATE | MAP_ANONYMOUS, | |
a03e2d42 | 646 | -1, 0); |
2b730f79 | 647 | if (retaddr == -1) { |
c8a706fe | 648 | goto fail; |
2b730f79 RH |
649 | } |
650 | if (pread(fd, g2h_untagged(start), len, offset) == -1) { | |
fb7e378c | 651 | goto fail; |
2b730f79 | 652 | } |
0dd55812 | 653 | if (!(target_prot & PROT_WRITE)) { |
9dba3ca5 | 654 | ret = target_mprotect(start, len, target_prot); |
86abac06 | 655 | assert(ret == 0); |
a03e2d42 FB |
656 | } |
657 | goto the_end; | |
54936004 | 658 | } |
2b730f79 | 659 | |
a03e2d42 FB |
660 | /* handle the start of the mapping */ |
661 | if (start > real_start) { | |
f9cd8f5e | 662 | if (real_last == real_start + qemu_host_page_size - 1) { |
a03e2d42 | 663 | /* one single host page */ |
f9cd8f5e | 664 | if (!mmap_frag(real_start, start, last, |
99982beb | 665 | target_prot, flags, fd, offset)) { |
c8a706fe | 666 | goto fail; |
2b730f79 | 667 | } |
a03e2d42 FB |
668 | goto the_end1; |
669 | } | |
99982beb RH |
670 | if (!mmap_frag(real_start, start, |
671 | real_start + qemu_host_page_size - 1, | |
672 | target_prot, flags, fd, offset)) { | |
c8a706fe | 673 | goto fail; |
2b730f79 | 674 | } |
a03e2d42 FB |
675 | real_start += qemu_host_page_size; |
676 | } | |
677 | /* handle the end of the mapping */ | |
f9cd8f5e RH |
678 | if (last < real_last) { |
679 | abi_ulong real_page = real_last - qemu_host_page_size + 1; | |
680 | if (!mmap_frag(real_page, real_page, last, | |
99982beb | 681 | target_prot, flags, fd, |
f9cd8f5e | 682 | offset + real_page - start)) { |
c8a706fe | 683 | goto fail; |
2b730f79 | 684 | } |
f9cd8f5e | 685 | real_last -= qemu_host_page_size; |
54936004 | 686 | } |
3b46e624 | 687 | |
a03e2d42 | 688 | /* map the middle (easier) */ |
f9cd8f5e | 689 | if (real_start < real_last) { |
ddcdd8c4 | 690 | void *p, *want_p; |
55baec0f | 691 | off_t offset1; |
ddcdd8c4 | 692 | size_t len1; |
55baec0f | 693 | |
2b730f79 | 694 | if (flags & MAP_ANONYMOUS) { |
a03e2d42 | 695 | offset1 = 0; |
2b730f79 | 696 | } else { |
a03e2d42 | 697 | offset1 = offset + real_start - start; |
2b730f79 | 698 | } |
ddcdd8c4 AO |
699 | len1 = real_last - real_start + 1; |
700 | want_p = g2h_untagged(real_start); | |
701 | ||
702 | p = mmap(want_p, len1, target_to_host_prot(target_prot), | |
703 | flags, fd, offset1); | |
704 | if (p != want_p) { | |
705 | if (p != MAP_FAILED) { | |
706 | munmap(p, len1); | |
707 | errno = EEXIST; | |
708 | } | |
c8a706fe | 709 | goto fail; |
2b730f79 | 710 | } |
f93b7695 | 711 | passthrough_start = real_start; |
f9cd8f5e | 712 | passthrough_last = real_last; |
a03e2d42 | 713 | } |
54936004 FB |
714 | } |
715 | the_end1: | |
26bab757 RH |
716 | if (flags & MAP_ANONYMOUS) { |
717 | page_flags |= PAGE_ANON; | |
718 | } | |
d9c58585 | 719 | page_flags |= PAGE_RESET; |
f9cd8f5e RH |
720 | if (passthrough_start > passthrough_last) { |
721 | page_set_flags(start, last, page_flags); | |
f93b7695 IL |
722 | } else { |
723 | if (start < passthrough_start) { | |
49840a4a | 724 | page_set_flags(start, passthrough_start - 1, page_flags); |
f93b7695 | 725 | } |
f9cd8f5e | 726 | page_set_flags(passthrough_start, passthrough_last, |
f93b7695 | 727 | page_flags | PAGE_PASSTHROUGH); |
f9cd8f5e RH |
728 | if (passthrough_last < last) { |
729 | page_set_flags(passthrough_last + 1, last, page_flags); | |
f93b7695 IL |
730 | } |
731 | } | |
54936004 | 732 | the_end: |
d0e165ae | 733 | trace_target_mmap_complete(start); |
10d0d505 | 734 | if (qemu_loglevel_mask(CPU_LOG_PAGE)) { |
93756fdc RH |
735 | FILE *f = qemu_log_trylock(); |
736 | if (f) { | |
737 | fprintf(f, "page layout changed following mmap\n"); | |
738 | page_dump(f); | |
739 | qemu_log_unlock(f); | |
740 | } | |
10d0d505 | 741 | } |
c8a706fe | 742 | mmap_unlock(); |
54936004 | 743 | return start; |
c8a706fe PB |
744 | fail: |
745 | mmap_unlock(); | |
746 | return -1; | |
54936004 FB |
747 | } |
748 | ||
558a4411 | 749 | static void mmap_reserve_or_unmap(abi_ulong start, abi_ulong len) |
68a1c816 PB |
750 | { |
751 | abi_ulong real_start; | |
260561d8 RH |
752 | abi_ulong real_last; |
753 | abi_ulong real_len; | |
754 | abi_ulong last; | |
755 | abi_ulong a; | |
558a4411 | 756 | void *host_start; |
68a1c816 PB |
757 | int prot; |
758 | ||
260561d8 | 759 | last = start + len - 1; |
68a1c816 | 760 | real_start = start & qemu_host_page_mask; |
260561d8 RH |
761 | real_last = HOST_PAGE_ALIGN(last) - 1; |
762 | ||
763 | /* | |
764 | * If guest pages remain on the first or last host pages, | |
765 | * adjust the deallocation to retain those guest pages. | |
766 | * The single page special case is required for the last page, | |
767 | * lest real_start overflow to zero. | |
768 | */ | |
769 | if (real_last - real_start < qemu_host_page_size) { | |
68a1c816 | 770 | prot = 0; |
260561d8 RH |
771 | for (a = real_start; a < start; a += TARGET_PAGE_SIZE) { |
772 | prot |= page_get_flags(a); | |
68a1c816 | 773 | } |
260561d8 RH |
774 | for (a = last; a < real_last; a += TARGET_PAGE_SIZE) { |
775 | prot |= page_get_flags(a + 1); | |
776 | } | |
777 | if (prot != 0) { | |
778 | return; | |
779 | } | |
780 | } else { | |
781 | for (prot = 0, a = real_start; a < start; a += TARGET_PAGE_SIZE) { | |
782 | prot |= page_get_flags(a); | |
68a1c816 | 783 | } |
2b730f79 | 784 | if (prot != 0) { |
68a1c816 | 785 | real_start += qemu_host_page_size; |
2b730f79 | 786 | } |
260561d8 RH |
787 | |
788 | for (prot = 0, a = last; a < real_last; a += TARGET_PAGE_SIZE) { | |
789 | prot |= page_get_flags(a + 1); | |
68a1c816 | 790 | } |
2b730f79 | 791 | if (prot != 0) { |
260561d8 RH |
792 | real_last -= qemu_host_page_size; |
793 | } | |
794 | ||
795 | if (real_last < real_start) { | |
796 | return; | |
2b730f79 | 797 | } |
68a1c816 | 798 | } |
260561d8 RH |
799 | |
800 | real_len = real_last - real_start + 1; | |
801 | host_start = g2h_untagged(real_start); | |
802 | ||
558a4411 RH |
803 | if (reserved_va) { |
804 | void *ptr = mmap(host_start, real_len, PROT_NONE, | |
805 | MAP_FIXED | MAP_ANONYMOUS | |
806 | | MAP_PRIVATE | MAP_NORESERVE, -1, 0); | |
807 | assert(ptr == host_start); | |
808 | } else { | |
809 | int ret = munmap(host_start, real_len); | |
810 | assert(ret == 0); | |
811 | } | |
68a1c816 PB |
812 | } |
813 | ||
992f48a0 | 814 | int target_munmap(abi_ulong start, abi_ulong len) |
54936004 | 815 | { |
b7b18d26 AB |
816 | trace_target_munmap(start, len); |
817 | ||
2b730f79 | 818 | if (start & ~TARGET_PAGE_MASK) { |
78cf3390 | 819 | return -TARGET_EINVAL; |
2b730f79 | 820 | } |
54936004 | 821 | len = TARGET_PAGE_ALIGN(len); |
46b12f46 | 822 | if (len == 0 || !guest_range_valid_untagged(start, len)) { |
78cf3390 | 823 | return -TARGET_EINVAL; |
ebf9a363 MF |
824 | } |
825 | ||
c8a706fe | 826 | mmap_lock(); |
d7b0c5d0 RH |
827 | mmap_reserve_or_unmap(start, len); |
828 | page_set_flags(start, start + len - 1, 0); | |
c8a706fe | 829 | mmap_unlock(); |
d7b0c5d0 RH |
830 | |
831 | return 0; | |
54936004 FB |
832 | } |
833 | ||
992f48a0 BS |
834 | abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size, |
835 | abi_ulong new_size, unsigned long flags, | |
836 | abi_ulong new_addr) | |
54936004 FB |
837 | { |
838 | int prot; | |
f19412a2 | 839 | void *host_addr; |
54936004 | 840 | |
46b12f46 | 841 | if (!guest_range_valid_untagged(old_addr, old_size) || |
ebf9a363 | 842 | ((flags & MREMAP_FIXED) && |
46b12f46 | 843 | !guest_range_valid_untagged(new_addr, new_size)) || |
ccc5ccc1 | 844 | ((flags & MREMAP_MAYMOVE) == 0 && |
46b12f46 | 845 | !guest_range_valid_untagged(old_addr, new_size))) { |
ebf9a363 MF |
846 | errno = ENOMEM; |
847 | return -1; | |
848 | } | |
849 | ||
c8a706fe | 850 | mmap_lock(); |
f19412a2 | 851 | |
68a1c816 | 852 | if (flags & MREMAP_FIXED) { |
3e8f1628 RH |
853 | host_addr = mremap(g2h_untagged(old_addr), old_size, new_size, |
854 | flags, g2h_untagged(new_addr)); | |
68a1c816 | 855 | |
b76f21a7 | 856 | if (reserved_va && host_addr != MAP_FAILED) { |
2b730f79 RH |
857 | /* |
858 | * If new and old addresses overlap then the above mremap will | |
859 | * already have failed with EINVAL. | |
860 | */ | |
558a4411 | 861 | mmap_reserve_or_unmap(old_addr, old_size); |
68a1c816 PB |
862 | } |
863 | } else if (flags & MREMAP_MAYMOVE) { | |
f19412a2 AJ |
864 | abi_ulong mmap_start; |
865 | ||
30ab9ef2 | 866 | mmap_start = mmap_find_vma(0, new_size, TARGET_PAGE_SIZE); |
f19412a2 AJ |
867 | |
868 | if (mmap_start == -1) { | |
869 | errno = ENOMEM; | |
870 | host_addr = MAP_FAILED; | |
68a1c816 | 871 | } else { |
3e8f1628 RH |
872 | host_addr = mremap(g2h_untagged(old_addr), old_size, new_size, |
873 | flags | MREMAP_FIXED, | |
874 | g2h_untagged(mmap_start)); | |
b76f21a7 | 875 | if (reserved_va) { |
558a4411 | 876 | mmap_reserve_or_unmap(old_addr, old_size); |
c65ffe6d | 877 | } |
68a1c816 | 878 | } |
3af72a4d | 879 | } else { |
68a1c816 | 880 | int prot = 0; |
b76f21a7 | 881 | if (reserved_va && old_size < new_size) { |
68a1c816 PB |
882 | abi_ulong addr; |
883 | for (addr = old_addr + old_size; | |
884 | addr < old_addr + new_size; | |
885 | addr++) { | |
886 | prot |= page_get_flags(addr); | |
887 | } | |
888 | } | |
889 | if (prot == 0) { | |
3e8f1628 RH |
890 | host_addr = mremap(g2h_untagged(old_addr), |
891 | old_size, new_size, flags); | |
56d19084 TK |
892 | |
893 | if (host_addr != MAP_FAILED) { | |
894 | /* Check if address fits target address space */ | |
46b12f46 | 895 | if (!guest_range_valid_untagged(h2g(host_addr), new_size)) { |
56d19084 | 896 | /* Revert mremap() changes */ |
3e8f1628 RH |
897 | host_addr = mremap(g2h_untagged(old_addr), |
898 | new_size, old_size, flags); | |
56d19084 TK |
899 | errno = ENOMEM; |
900 | host_addr = MAP_FAILED; | |
901 | } else if (reserved_va && old_size > new_size) { | |
558a4411 RH |
902 | mmap_reserve_or_unmap(old_addr + old_size, |
903 | old_size - new_size); | |
56d19084 | 904 | } |
68a1c816 PB |
905 | } |
906 | } else { | |
907 | errno = ENOMEM; | |
908 | host_addr = MAP_FAILED; | |
909 | } | |
f19412a2 AJ |
910 | } |
911 | ||
912 | if (host_addr == MAP_FAILED) { | |
c8a706fe PB |
913 | new_addr = -1; |
914 | } else { | |
915 | new_addr = h2g(host_addr); | |
916 | prot = page_get_flags(old_addr); | |
49840a4a RH |
917 | page_set_flags(old_addr, old_addr + old_size - 1, 0); |
918 | page_set_flags(new_addr, new_addr + new_size - 1, | |
d9c58585 | 919 | prot | PAGE_VALID | PAGE_RESET); |
c8a706fe PB |
920 | } |
921 | mmap_unlock(); | |
54936004 FB |
922 | return new_addr; |
923 | } | |
892a4f6a | 924 | |
892a4f6a IL |
925 | abi_long target_madvise(abi_ulong start, abi_ulong len_in, int advice) |
926 | { | |
e230ec09 | 927 | abi_ulong len; |
892a4f6a IL |
928 | int ret = 0; |
929 | ||
930 | if (start & ~TARGET_PAGE_MASK) { | |
931 | return -TARGET_EINVAL; | |
932 | } | |
e230ec09 | 933 | if (len_in == 0) { |
892a4f6a IL |
934 | return 0; |
935 | } | |
e230ec09 RH |
936 | len = TARGET_PAGE_ALIGN(len_in); |
937 | if (len == 0 || !guest_range_valid_untagged(start, len)) { | |
892a4f6a IL |
938 | return -TARGET_EINVAL; |
939 | } | |
940 | ||
4530deb1 HD |
941 | /* Translate for some architectures which have different MADV_xxx values */ |
942 | switch (advice) { | |
943 | case TARGET_MADV_DONTNEED: /* alpha */ | |
944 | advice = MADV_DONTNEED; | |
945 | break; | |
946 | case TARGET_MADV_WIPEONFORK: /* parisc */ | |
947 | advice = MADV_WIPEONFORK; | |
948 | break; | |
949 | case TARGET_MADV_KEEPONFORK: /* parisc */ | |
950 | advice = MADV_KEEPONFORK; | |
951 | break; | |
952 | /* we do not care about the other MADV_xxx values yet */ | |
953 | } | |
954 | ||
892a4f6a | 955 | /* |
4530deb1 HD |
956 | * Most advice values are hints, so ignoring and returning success is ok. |
957 | * | |
958 | * However, some advice values such as MADV_DONTNEED, MADV_WIPEONFORK and | |
959 | * MADV_KEEPONFORK are not hints and need to be emulated. | |
892a4f6a | 960 | * |
4530deb1 HD |
961 | * A straight passthrough for those may not be safe because qemu sometimes |
962 | * turns private file-backed mappings into anonymous mappings. | |
ecb796db RH |
963 | * If all guest pages have PAGE_PASSTHROUGH set, mappings have the |
964 | * same semantics for the host as for the guest. | |
892a4f6a | 965 | * |
4530deb1 HD |
966 | * We pass through MADV_WIPEONFORK and MADV_KEEPONFORK if possible and |
967 | * return failure if not. | |
968 | * | |
969 | * MADV_DONTNEED is passed through as well, if possible. | |
970 | * If passthrough isn't possible, we nevertheless (wrongly!) return | |
971 | * success, which is broken but some userspace programs fail to work | |
972 | * otherwise. Completely implementing such emulation is quite complicated | |
973 | * though. | |
892a4f6a IL |
974 | */ |
975 | mmap_lock(); | |
4530deb1 HD |
976 | switch (advice) { |
977 | case MADV_WIPEONFORK: | |
978 | case MADV_KEEPONFORK: | |
979 | ret = -EINVAL; | |
980 | /* fall through */ | |
981 | case MADV_DONTNEED: | |
ecb796db | 982 | if (page_check_range(start, len, PAGE_PASSTHROUGH)) { |
4530deb1 HD |
983 | ret = get_errno(madvise(g2h_untagged(start), len, advice)); |
984 | if ((advice == MADV_DONTNEED) && (ret == 0)) { | |
10310cbd | 985 | page_reset_target_data(start, start + len - 1); |
4530deb1 | 986 | } |
dbbf8975 | 987 | } |
892a4f6a IL |
988 | } |
989 | mmap_unlock(); | |
990 | ||
991 | return ret; | |
992 | } | |
225a206c RH |
993 | |
994 | #ifndef TARGET_FORCE_SHMLBA | |
995 | /* | |
996 | * For most architectures, SHMLBA is the same as the page size; | |
997 | * some architectures have larger values, in which case they should | |
998 | * define TARGET_FORCE_SHMLBA and provide a target_shmlba() function. | |
999 | * This corresponds to the kernel arch code defining __ARCH_FORCE_SHMLBA | |
1000 | * and defining its own value for SHMLBA. | |
1001 | * | |
1002 | * The kernel also permits SHMLBA to be set by the architecture to a | |
1003 | * value larger than the page size without setting __ARCH_FORCE_SHMLBA; | |
1004 | * this means that addresses are rounded to the large size if | |
1005 | * SHM_RND is set but addresses not aligned to that size are not rejected | |
1006 | * as long as they are at least page-aligned. Since the only architecture | |
1007 | * which uses this is ia64 this code doesn't provide for that oddity. | |
1008 | */ | |
1009 | static inline abi_ulong target_shmlba(CPUArchState *cpu_env) | |
1010 | { | |
1011 | return TARGET_PAGE_SIZE; | |
1012 | } | |
1013 | #endif | |
1014 | ||
1015 | abi_ulong target_shmat(CPUArchState *cpu_env, int shmid, | |
1016 | abi_ulong shmaddr, int shmflg) | |
1017 | { | |
1018 | CPUState *cpu = env_cpu(cpu_env); | |
1019 | abi_ulong raddr; | |
225a206c | 1020 | struct shmid_ds shm_info; |
69fa2708 | 1021 | int ret; |
225a206c RH |
1022 | abi_ulong shmlba; |
1023 | ||
1024 | /* shmat pointers are always untagged */ | |
1025 | ||
1026 | /* find out the length of the shared memory segment */ | |
1027 | ret = get_errno(shmctl(shmid, IPC_STAT, &shm_info)); | |
1028 | if (is_error(ret)) { | |
1029 | /* can't get length, bail out */ | |
1030 | return ret; | |
1031 | } | |
1032 | ||
1033 | shmlba = target_shmlba(cpu_env); | |
1034 | ||
1035 | if (shmaddr & (shmlba - 1)) { | |
1036 | if (shmflg & SHM_RND) { | |
1037 | shmaddr &= ~(shmlba - 1); | |
1038 | } else { | |
1039 | return -TARGET_EINVAL; | |
1040 | } | |
1041 | } | |
1042 | if (!guest_range_valid_untagged(shmaddr, shm_info.shm_segsz)) { | |
1043 | return -TARGET_EINVAL; | |
1044 | } | |
1045 | ||
69fa2708 RH |
1046 | WITH_MMAP_LOCK_GUARD() { |
1047 | void *host_raddr; | |
1048 | ||
1049 | if (shmaddr) { | |
1050 | host_raddr = shmat(shmid, (void *)g2h_untagged(shmaddr), shmflg); | |
1051 | } else { | |
1052 | abi_ulong mmap_start; | |
1053 | ||
1054 | /* In order to use the host shmat, we need to honor host SHMLBA. */ | |
1055 | mmap_start = mmap_find_vma(0, shm_info.shm_segsz, | |
1056 | MAX(SHMLBA, shmlba)); | |
1057 | ||
1058 | if (mmap_start == -1) { | |
1059 | return -TARGET_ENOMEM; | |
1060 | } | |
1061 | host_raddr = shmat(shmid, g2h_untagged(mmap_start), | |
1062 | shmflg | SHM_REMAP); | |
1063 | } | |
1064 | ||
1065 | if (host_raddr == (void *)-1) { | |
1066 | return get_errno(-1); | |
1067 | } | |
1068 | raddr = h2g(host_raddr); | |
1069 | ||
1070 | page_set_flags(raddr, raddr + shm_info.shm_segsz - 1, | |
1071 | PAGE_VALID | PAGE_RESET | PAGE_READ | | |
1072 | (shmflg & SHM_RDONLY ? 0 : PAGE_WRITE)); | |
1073 | ||
1074 | for (int i = 0; i < N_SHM_REGIONS; i++) { | |
1075 | if (!shm_regions[i].in_use) { | |
1076 | shm_regions[i].in_use = true; | |
1077 | shm_regions[i].start = raddr; | |
1078 | shm_regions[i].size = shm_info.shm_segsz; | |
1079 | break; | |
1080 | } | |
1081 | } | |
1082 | } | |
225a206c RH |
1083 | |
1084 | /* | |
1085 | * We're mapping shared memory, so ensure we generate code for parallel | |
1086 | * execution and flush old translations. This will work up to the level | |
1087 | * supported by the host -- anything that requires EXCP_ATOMIC will not | |
1088 | * be atomic with respect to an external process. | |
1089 | */ | |
1090 | if (!(cpu->tcg_cflags & CF_PARALLEL)) { | |
1091 | cpu->tcg_cflags |= CF_PARALLEL; | |
1092 | tb_flush(cpu); | |
1093 | } | |
1094 | ||
225a206c RH |
1095 | return raddr; |
1096 | } | |
1097 | ||
1098 | abi_long target_shmdt(abi_ulong shmaddr) | |
1099 | { | |
225a206c RH |
1100 | abi_long rv; |
1101 | ||
1102 | /* shmdt pointers are always untagged */ | |
1103 | ||
69fa2708 RH |
1104 | WITH_MMAP_LOCK_GUARD() { |
1105 | for (int i = 0; i < N_SHM_REGIONS; ++i) { | |
1106 | if (shm_regions[i].in_use && shm_regions[i].start == shmaddr) { | |
1107 | shm_regions[i].in_use = false; | |
1108 | page_set_flags(shmaddr, shmaddr + shm_regions[i].size - 1, 0); | |
1109 | break; | |
1110 | } | |
225a206c | 1111 | } |
69fa2708 | 1112 | rv = get_errno(shmdt(g2h_untagged(shmaddr))); |
225a206c | 1113 | } |
225a206c RH |
1114 | return rv; |
1115 | } |