]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * Virtual page mapping | |
3 | * | |
4 | * Copyright (c) 2003 Fabrice Bellard | |
5 | * | |
6 | * This library is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU Lesser General Public | |
8 | * License as published by the Free Software Foundation; either | |
9 | * version 2 of the License, or (at your option) any later version. | |
10 | * | |
11 | * This library 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 GNU | |
14 | * Lesser General Public License for more details. | |
15 | * | |
16 | * You should have received a copy of the GNU Lesser General Public | |
17 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. | |
18 | */ | |
19 | #include "qemu/osdep.h" | |
20 | #include "qapi/error.h" | |
21 | #ifndef _WIN32 | |
22 | #endif | |
23 | ||
24 | #include "qemu/cutils.h" | |
25 | #include "cpu.h" | |
26 | #include "exec/exec-all.h" | |
27 | #include "tcg.h" | |
28 | #include "hw/qdev-core.h" | |
29 | #if !defined(CONFIG_USER_ONLY) | |
30 | #include "hw/boards.h" | |
31 | #include "hw/xen/xen.h" | |
32 | #endif | |
33 | #include "sysemu/kvm.h" | |
34 | #include "sysemu/sysemu.h" | |
35 | #include "qemu/timer.h" | |
36 | #include "qemu/config-file.h" | |
37 | #include "qemu/error-report.h" | |
38 | #if defined(CONFIG_USER_ONLY) | |
39 | #include "qemu.h" | |
40 | #else /* !CONFIG_USER_ONLY */ | |
41 | #include "hw/hw.h" | |
42 | #include "exec/memory.h" | |
43 | #include "exec/ioport.h" | |
44 | #include "sysemu/dma.h" | |
45 | #include "exec/address-spaces.h" | |
46 | #include "sysemu/xen-mapcache.h" | |
47 | #include "trace.h" | |
48 | #endif | |
49 | #include "exec/cpu-all.h" | |
50 | #include "qemu/rcu_queue.h" | |
51 | #include "qemu/main-loop.h" | |
52 | #include "translate-all.h" | |
53 | #include "sysemu/replay.h" | |
54 | ||
55 | #include "exec/memory-internal.h" | |
56 | #include "exec/ram_addr.h" | |
57 | #include "exec/log.h" | |
58 | ||
59 | #include "migration/vmstate.h" | |
60 | ||
61 | #include "qemu/range.h" | |
62 | #ifndef _WIN32 | |
63 | #include "qemu/mmap-alloc.h" | |
64 | #endif | |
65 | ||
66 | //#define DEBUG_SUBPAGE | |
67 | ||
68 | #if !defined(CONFIG_USER_ONLY) | |
69 | /* ram_list is read under rcu_read_lock()/rcu_read_unlock(). Writes | |
70 | * are protected by the ramlist lock. | |
71 | */ | |
72 | RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) }; | |
73 | ||
74 | static MemoryRegion *system_memory; | |
75 | static MemoryRegion *system_io; | |
76 | ||
77 | AddressSpace address_space_io; | |
78 | AddressSpace address_space_memory; | |
79 | ||
80 | MemoryRegion io_mem_rom, io_mem_notdirty; | |
81 | static MemoryRegion io_mem_unassigned; | |
82 | ||
83 | /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */ | |
84 | #define RAM_PREALLOC (1 << 0) | |
85 | ||
86 | /* RAM is mmap-ed with MAP_SHARED */ | |
87 | #define RAM_SHARED (1 << 1) | |
88 | ||
89 | /* Only a portion of RAM (used_length) is actually used, and migrated. | |
90 | * This used_length size can change across reboots. | |
91 | */ | |
92 | #define RAM_RESIZEABLE (1 << 2) | |
93 | ||
94 | #endif | |
95 | ||
96 | struct CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus); | |
97 | /* current CPU in the current thread. It is only valid inside | |
98 | cpu_exec() */ | |
99 | __thread CPUState *current_cpu; | |
100 | /* 0 = Do not count executed instructions. | |
101 | 1 = Precise instruction counting. | |
102 | 2 = Adaptive rate instruction counting. */ | |
103 | int use_icount; | |
104 | ||
105 | #if !defined(CONFIG_USER_ONLY) | |
106 | ||
107 | typedef struct PhysPageEntry PhysPageEntry; | |
108 | ||
109 | struct PhysPageEntry { | |
110 | /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */ | |
111 | uint32_t skip : 6; | |
112 | /* index into phys_sections (!skip) or phys_map_nodes (skip) */ | |
113 | uint32_t ptr : 26; | |
114 | }; | |
115 | ||
116 | #define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6) | |
117 | ||
118 | /* Size of the L2 (and L3, etc) page tables. */ | |
119 | #define ADDR_SPACE_BITS 64 | |
120 | ||
121 | #define P_L2_BITS 9 | |
122 | #define P_L2_SIZE (1 << P_L2_BITS) | |
123 | ||
124 | #define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1) | |
125 | ||
126 | typedef PhysPageEntry Node[P_L2_SIZE]; | |
127 | ||
128 | typedef struct PhysPageMap { | |
129 | struct rcu_head rcu; | |
130 | ||
131 | unsigned sections_nb; | |
132 | unsigned sections_nb_alloc; | |
133 | unsigned nodes_nb; | |
134 | unsigned nodes_nb_alloc; | |
135 | Node *nodes; | |
136 | MemoryRegionSection *sections; | |
137 | } PhysPageMap; | |
138 | ||
139 | struct AddressSpaceDispatch { | |
140 | struct rcu_head rcu; | |
141 | ||
142 | MemoryRegionSection *mru_section; | |
143 | /* This is a multi-level map on the physical address space. | |
144 | * The bottom level has pointers to MemoryRegionSections. | |
145 | */ | |
146 | PhysPageEntry phys_map; | |
147 | PhysPageMap map; | |
148 | AddressSpace *as; | |
149 | }; | |
150 | ||
151 | #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK) | |
152 | typedef struct subpage_t { | |
153 | MemoryRegion iomem; | |
154 | AddressSpace *as; | |
155 | hwaddr base; | |
156 | uint16_t sub_section[TARGET_PAGE_SIZE]; | |
157 | } subpage_t; | |
158 | ||
159 | #define PHYS_SECTION_UNASSIGNED 0 | |
160 | #define PHYS_SECTION_NOTDIRTY 1 | |
161 | #define PHYS_SECTION_ROM 2 | |
162 | #define PHYS_SECTION_WATCH 3 | |
163 | ||
164 | static void io_mem_init(void); | |
165 | static void memory_map_init(void); | |
166 | static void tcg_commit(MemoryListener *listener); | |
167 | ||
168 | static MemoryRegion io_mem_watch; | |
169 | ||
170 | /** | |
171 | * CPUAddressSpace: all the information a CPU needs about an AddressSpace | |
172 | * @cpu: the CPU whose AddressSpace this is | |
173 | * @as: the AddressSpace itself | |
174 | * @memory_dispatch: its dispatch pointer (cached, RCU protected) | |
175 | * @tcg_as_listener: listener for tracking changes to the AddressSpace | |
176 | */ | |
177 | struct CPUAddressSpace { | |
178 | CPUState *cpu; | |
179 | AddressSpace *as; | |
180 | struct AddressSpaceDispatch *memory_dispatch; | |
181 | MemoryListener tcg_as_listener; | |
182 | }; | |
183 | ||
184 | #endif | |
185 | ||
186 | #if !defined(CONFIG_USER_ONLY) | |
187 | ||
188 | static void phys_map_node_reserve(PhysPageMap *map, unsigned nodes) | |
189 | { | |
190 | static unsigned alloc_hint = 16; | |
191 | if (map->nodes_nb + nodes > map->nodes_nb_alloc) { | |
192 | map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, alloc_hint); | |
193 | map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, map->nodes_nb + nodes); | |
194 | map->nodes = g_renew(Node, map->nodes, map->nodes_nb_alloc); | |
195 | alloc_hint = map->nodes_nb_alloc; | |
196 | } | |
197 | } | |
198 | ||
199 | static uint32_t phys_map_node_alloc(PhysPageMap *map, bool leaf) | |
200 | { | |
201 | unsigned i; | |
202 | uint32_t ret; | |
203 | PhysPageEntry e; | |
204 | PhysPageEntry *p; | |
205 | ||
206 | ret = map->nodes_nb++; | |
207 | p = map->nodes[ret]; | |
208 | assert(ret != PHYS_MAP_NODE_NIL); | |
209 | assert(ret != map->nodes_nb_alloc); | |
210 | ||
211 | e.skip = leaf ? 0 : 1; | |
212 | e.ptr = leaf ? PHYS_SECTION_UNASSIGNED : PHYS_MAP_NODE_NIL; | |
213 | for (i = 0; i < P_L2_SIZE; ++i) { | |
214 | memcpy(&p[i], &e, sizeof(e)); | |
215 | } | |
216 | return ret; | |
217 | } | |
218 | ||
219 | static void phys_page_set_level(PhysPageMap *map, PhysPageEntry *lp, | |
220 | hwaddr *index, hwaddr *nb, uint16_t leaf, | |
221 | int level) | |
222 | { | |
223 | PhysPageEntry *p; | |
224 | hwaddr step = (hwaddr)1 << (level * P_L2_BITS); | |
225 | ||
226 | if (lp->skip && lp->ptr == PHYS_MAP_NODE_NIL) { | |
227 | lp->ptr = phys_map_node_alloc(map, level == 0); | |
228 | } | |
229 | p = map->nodes[lp->ptr]; | |
230 | lp = &p[(*index >> (level * P_L2_BITS)) & (P_L2_SIZE - 1)]; | |
231 | ||
232 | while (*nb && lp < &p[P_L2_SIZE]) { | |
233 | if ((*index & (step - 1)) == 0 && *nb >= step) { | |
234 | lp->skip = 0; | |
235 | lp->ptr = leaf; | |
236 | *index += step; | |
237 | *nb -= step; | |
238 | } else { | |
239 | phys_page_set_level(map, lp, index, nb, leaf, level - 1); | |
240 | } | |
241 | ++lp; | |
242 | } | |
243 | } | |
244 | ||
245 | static void phys_page_set(AddressSpaceDispatch *d, | |
246 | hwaddr index, hwaddr nb, | |
247 | uint16_t leaf) | |
248 | { | |
249 | /* Wildly overreserve - it doesn't matter much. */ | |
250 | phys_map_node_reserve(&d->map, 3 * P_L2_LEVELS); | |
251 | ||
252 | phys_page_set_level(&d->map, &d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1); | |
253 | } | |
254 | ||
255 | /* Compact a non leaf page entry. Simply detect that the entry has a single child, | |
256 | * and update our entry so we can skip it and go directly to the destination. | |
257 | */ | |
258 | static void phys_page_compact(PhysPageEntry *lp, Node *nodes, unsigned long *compacted) | |
259 | { | |
260 | unsigned valid_ptr = P_L2_SIZE; | |
261 | int valid = 0; | |
262 | PhysPageEntry *p; | |
263 | int i; | |
264 | ||
265 | if (lp->ptr == PHYS_MAP_NODE_NIL) { | |
266 | return; | |
267 | } | |
268 | ||
269 | p = nodes[lp->ptr]; | |
270 | for (i = 0; i < P_L2_SIZE; i++) { | |
271 | if (p[i].ptr == PHYS_MAP_NODE_NIL) { | |
272 | continue; | |
273 | } | |
274 | ||
275 | valid_ptr = i; | |
276 | valid++; | |
277 | if (p[i].skip) { | |
278 | phys_page_compact(&p[i], nodes, compacted); | |
279 | } | |
280 | } | |
281 | ||
282 | /* We can only compress if there's only one child. */ | |
283 | if (valid != 1) { | |
284 | return; | |
285 | } | |
286 | ||
287 | assert(valid_ptr < P_L2_SIZE); | |
288 | ||
289 | /* Don't compress if it won't fit in the # of bits we have. */ | |
290 | if (lp->skip + p[valid_ptr].skip >= (1 << 3)) { | |
291 | return; | |
292 | } | |
293 | ||
294 | lp->ptr = p[valid_ptr].ptr; | |
295 | if (!p[valid_ptr].skip) { | |
296 | /* If our only child is a leaf, make this a leaf. */ | |
297 | /* By design, we should have made this node a leaf to begin with so we | |
298 | * should never reach here. | |
299 | * But since it's so simple to handle this, let's do it just in case we | |
300 | * change this rule. | |
301 | */ | |
302 | lp->skip = 0; | |
303 | } else { | |
304 | lp->skip += p[valid_ptr].skip; | |
305 | } | |
306 | } | |
307 | ||
308 | static void phys_page_compact_all(AddressSpaceDispatch *d, int nodes_nb) | |
309 | { | |
310 | DECLARE_BITMAP(compacted, nodes_nb); | |
311 | ||
312 | if (d->phys_map.skip) { | |
313 | phys_page_compact(&d->phys_map, d->map.nodes, compacted); | |
314 | } | |
315 | } | |
316 | ||
317 | static inline bool section_covers_addr(const MemoryRegionSection *section, | |
318 | hwaddr addr) | |
319 | { | |
320 | /* Memory topology clips a memory region to [0, 2^64); size.hi > 0 means | |
321 | * the section must cover the entire address space. | |
322 | */ | |
323 | return section->size.hi || | |
324 | range_covers_byte(section->offset_within_address_space, | |
325 | section->size.lo, addr); | |
326 | } | |
327 | ||
328 | static MemoryRegionSection *phys_page_find(PhysPageEntry lp, hwaddr addr, | |
329 | Node *nodes, MemoryRegionSection *sections) | |
330 | { | |
331 | PhysPageEntry *p; | |
332 | hwaddr index = addr >> TARGET_PAGE_BITS; | |
333 | int i; | |
334 | ||
335 | for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) { | |
336 | if (lp.ptr == PHYS_MAP_NODE_NIL) { | |
337 | return §ions[PHYS_SECTION_UNASSIGNED]; | |
338 | } | |
339 | p = nodes[lp.ptr]; | |
340 | lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)]; | |
341 | } | |
342 | ||
343 | if (section_covers_addr(§ions[lp.ptr], addr)) { | |
344 | return §ions[lp.ptr]; | |
345 | } else { | |
346 | return §ions[PHYS_SECTION_UNASSIGNED]; | |
347 | } | |
348 | } | |
349 | ||
350 | bool memory_region_is_unassigned(MemoryRegion *mr) | |
351 | { | |
352 | return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device | |
353 | && mr != &io_mem_watch; | |
354 | } | |
355 | ||
356 | /* Called from RCU critical section */ | |
357 | static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d, | |
358 | hwaddr addr, | |
359 | bool resolve_subpage) | |
360 | { | |
361 | MemoryRegionSection *section = atomic_read(&d->mru_section); | |
362 | subpage_t *subpage; | |
363 | bool update; | |
364 | ||
365 | if (section && section != &d->map.sections[PHYS_SECTION_UNASSIGNED] && | |
366 | section_covers_addr(section, addr)) { | |
367 | update = false; | |
368 | } else { | |
369 | section = phys_page_find(d->phys_map, addr, d->map.nodes, | |
370 | d->map.sections); | |
371 | update = true; | |
372 | } | |
373 | if (resolve_subpage && section->mr->subpage) { | |
374 | subpage = container_of(section->mr, subpage_t, iomem); | |
375 | section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]]; | |
376 | } | |
377 | if (update) { | |
378 | atomic_set(&d->mru_section, section); | |
379 | } | |
380 | return section; | |
381 | } | |
382 | ||
383 | /* Called from RCU critical section */ | |
384 | static MemoryRegionSection * | |
385 | address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat, | |
386 | hwaddr *plen, bool resolve_subpage) | |
387 | { | |
388 | MemoryRegionSection *section; | |
389 | MemoryRegion *mr; | |
390 | Int128 diff; | |
391 | ||
392 | section = address_space_lookup_region(d, addr, resolve_subpage); | |
393 | /* Compute offset within MemoryRegionSection */ | |
394 | addr -= section->offset_within_address_space; | |
395 | ||
396 | /* Compute offset within MemoryRegion */ | |
397 | *xlat = addr + section->offset_within_region; | |
398 | ||
399 | mr = section->mr; | |
400 | ||
401 | /* MMIO registers can be expected to perform full-width accesses based only | |
402 | * on their address, without considering adjacent registers that could | |
403 | * decode to completely different MemoryRegions. When such registers | |
404 | * exist (e.g. I/O ports 0xcf8 and 0xcf9 on most PC chipsets), MMIO | |
405 | * regions overlap wildly. For this reason we cannot clamp the accesses | |
406 | * here. | |
407 | * | |
408 | * If the length is small (as is the case for address_space_ldl/stl), | |
409 | * everything works fine. If the incoming length is large, however, | |
410 | * the caller really has to do the clamping through memory_access_size. | |
411 | */ | |
412 | if (memory_region_is_ram(mr)) { | |
413 | diff = int128_sub(section->size, int128_make64(addr)); | |
414 | *plen = int128_get64(int128_min(diff, int128_make64(*plen))); | |
415 | } | |
416 | return section; | |
417 | } | |
418 | ||
419 | /* Called from RCU critical section */ | |
420 | MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr, | |
421 | hwaddr *xlat, hwaddr *plen, | |
422 | bool is_write) | |
423 | { | |
424 | IOMMUTLBEntry iotlb; | |
425 | MemoryRegionSection *section; | |
426 | MemoryRegion *mr; | |
427 | ||
428 | for (;;) { | |
429 | AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch); | |
430 | section = address_space_translate_internal(d, addr, &addr, plen, true); | |
431 | mr = section->mr; | |
432 | ||
433 | if (!mr->iommu_ops) { | |
434 | break; | |
435 | } | |
436 | ||
437 | iotlb = mr->iommu_ops->translate(mr, addr, is_write); | |
438 | addr = ((iotlb.translated_addr & ~iotlb.addr_mask) | |
439 | | (addr & iotlb.addr_mask)); | |
440 | *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1); | |
441 | if (!(iotlb.perm & (1 << is_write))) { | |
442 | mr = &io_mem_unassigned; | |
443 | break; | |
444 | } | |
445 | ||
446 | as = iotlb.target_as; | |
447 | } | |
448 | ||
449 | if (xen_enabled() && memory_access_is_direct(mr, is_write)) { | |
450 | hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr; | |
451 | *plen = MIN(page, *plen); | |
452 | } | |
453 | ||
454 | *xlat = addr; | |
455 | return mr; | |
456 | } | |
457 | ||
458 | /* Called from RCU critical section */ | |
459 | MemoryRegionSection * | |
460 | address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr, | |
461 | hwaddr *xlat, hwaddr *plen) | |
462 | { | |
463 | MemoryRegionSection *section; | |
464 | AddressSpaceDispatch *d = cpu->cpu_ases[asidx].memory_dispatch; | |
465 | ||
466 | section = address_space_translate_internal(d, addr, xlat, plen, false); | |
467 | ||
468 | assert(!section->mr->iommu_ops); | |
469 | return section; | |
470 | } | |
471 | #endif | |
472 | ||
473 | #if !defined(CONFIG_USER_ONLY) | |
474 | ||
475 | static int cpu_common_post_load(void *opaque, int version_id) | |
476 | { | |
477 | CPUState *cpu = opaque; | |
478 | ||
479 | /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the | |
480 | version_id is increased. */ | |
481 | cpu->interrupt_request &= ~0x01; | |
482 | tlb_flush(cpu, 1); | |
483 | ||
484 | return 0; | |
485 | } | |
486 | ||
487 | static int cpu_common_pre_load(void *opaque) | |
488 | { | |
489 | CPUState *cpu = opaque; | |
490 | ||
491 | cpu->exception_index = -1; | |
492 | ||
493 | return 0; | |
494 | } | |
495 | ||
496 | static bool cpu_common_exception_index_needed(void *opaque) | |
497 | { | |
498 | CPUState *cpu = opaque; | |
499 | ||
500 | return tcg_enabled() && cpu->exception_index != -1; | |
501 | } | |
502 | ||
503 | static const VMStateDescription vmstate_cpu_common_exception_index = { | |
504 | .name = "cpu_common/exception_index", | |
505 | .version_id = 1, | |
506 | .minimum_version_id = 1, | |
507 | .needed = cpu_common_exception_index_needed, | |
508 | .fields = (VMStateField[]) { | |
509 | VMSTATE_INT32(exception_index, CPUState), | |
510 | VMSTATE_END_OF_LIST() | |
511 | } | |
512 | }; | |
513 | ||
514 | static bool cpu_common_crash_occurred_needed(void *opaque) | |
515 | { | |
516 | CPUState *cpu = opaque; | |
517 | ||
518 | return cpu->crash_occurred; | |
519 | } | |
520 | ||
521 | static const VMStateDescription vmstate_cpu_common_crash_occurred = { | |
522 | .name = "cpu_common/crash_occurred", | |
523 | .version_id = 1, | |
524 | .minimum_version_id = 1, | |
525 | .needed = cpu_common_crash_occurred_needed, | |
526 | .fields = (VMStateField[]) { | |
527 | VMSTATE_BOOL(crash_occurred, CPUState), | |
528 | VMSTATE_END_OF_LIST() | |
529 | } | |
530 | }; | |
531 | ||
532 | const VMStateDescription vmstate_cpu_common = { | |
533 | .name = "cpu_common", | |
534 | .version_id = 1, | |
535 | .minimum_version_id = 1, | |
536 | .pre_load = cpu_common_pre_load, | |
537 | .post_load = cpu_common_post_load, | |
538 | .fields = (VMStateField[]) { | |
539 | VMSTATE_UINT32(halted, CPUState), | |
540 | VMSTATE_UINT32(interrupt_request, CPUState), | |
541 | VMSTATE_END_OF_LIST() | |
542 | }, | |
543 | .subsections = (const VMStateDescription*[]) { | |
544 | &vmstate_cpu_common_exception_index, | |
545 | &vmstate_cpu_common_crash_occurred, | |
546 | NULL | |
547 | } | |
548 | }; | |
549 | ||
550 | #endif | |
551 | ||
552 | CPUState *qemu_get_cpu(int index) | |
553 | { | |
554 | CPUState *cpu; | |
555 | ||
556 | CPU_FOREACH(cpu) { | |
557 | if (cpu->cpu_index == index) { | |
558 | return cpu; | |
559 | } | |
560 | } | |
561 | ||
562 | return NULL; | |
563 | } | |
564 | ||
565 | #if !defined(CONFIG_USER_ONLY) | |
566 | void cpu_address_space_init(CPUState *cpu, AddressSpace *as, int asidx) | |
567 | { | |
568 | CPUAddressSpace *newas; | |
569 | ||
570 | /* Target code should have set num_ases before calling us */ | |
571 | assert(asidx < cpu->num_ases); | |
572 | ||
573 | if (asidx == 0) { | |
574 | /* address space 0 gets the convenience alias */ | |
575 | cpu->as = as; | |
576 | } | |
577 | ||
578 | /* KVM cannot currently support multiple address spaces. */ | |
579 | assert(asidx == 0 || !kvm_enabled()); | |
580 | ||
581 | if (!cpu->cpu_ases) { | |
582 | cpu->cpu_ases = g_new0(CPUAddressSpace, cpu->num_ases); | |
583 | } | |
584 | ||
585 | newas = &cpu->cpu_ases[asidx]; | |
586 | newas->cpu = cpu; | |
587 | newas->as = as; | |
588 | if (tcg_enabled()) { | |
589 | newas->tcg_as_listener.commit = tcg_commit; | |
590 | memory_listener_register(&newas->tcg_as_listener, as); | |
591 | } | |
592 | } | |
593 | ||
594 | AddressSpace *cpu_get_address_space(CPUState *cpu, int asidx) | |
595 | { | |
596 | /* Return the AddressSpace corresponding to the specified index */ | |
597 | return cpu->cpu_ases[asidx].as; | |
598 | } | |
599 | #endif | |
600 | ||
601 | static bool cpu_index_auto_assigned; | |
602 | ||
603 | static int cpu_get_free_index(void) | |
604 | { | |
605 | CPUState *some_cpu; | |
606 | int cpu_index = 0; | |
607 | ||
608 | cpu_index_auto_assigned = true; | |
609 | CPU_FOREACH(some_cpu) { | |
610 | cpu_index++; | |
611 | } | |
612 | return cpu_index; | |
613 | } | |
614 | ||
615 | void cpu_exec_exit(CPUState *cpu) | |
616 | { | |
617 | CPUClass *cc = CPU_GET_CLASS(cpu); | |
618 | ||
619 | cpu_list_lock(); | |
620 | if (cpu->node.tqe_prev == NULL) { | |
621 | /* there is nothing to undo since cpu_exec_init() hasn't been called */ | |
622 | cpu_list_unlock(); | |
623 | return; | |
624 | } | |
625 | ||
626 | assert(!(cpu_index_auto_assigned && cpu != QTAILQ_LAST(&cpus, CPUTailQ))); | |
627 | ||
628 | QTAILQ_REMOVE(&cpus, cpu, node); | |
629 | cpu->node.tqe_prev = NULL; | |
630 | cpu->cpu_index = UNASSIGNED_CPU_INDEX; | |
631 | cpu_list_unlock(); | |
632 | ||
633 | if (cc->vmsd != NULL) { | |
634 | vmstate_unregister(NULL, cc->vmsd, cpu); | |
635 | } | |
636 | if (qdev_get_vmsd(DEVICE(cpu)) == NULL) { | |
637 | vmstate_unregister(NULL, &vmstate_cpu_common, cpu); | |
638 | } | |
639 | } | |
640 | ||
641 | void cpu_exec_init(CPUState *cpu, Error **errp) | |
642 | { | |
643 | CPUClass *cc ATTRIBUTE_UNUSED = CPU_GET_CLASS(cpu); | |
644 | Error *local_err ATTRIBUTE_UNUSED = NULL; | |
645 | ||
646 | cpu->as = NULL; | |
647 | cpu->num_ases = 0; | |
648 | ||
649 | #ifndef CONFIG_USER_ONLY | |
650 | cpu->thread_id = qemu_get_thread_id(); | |
651 | ||
652 | /* This is a softmmu CPU object, so create a property for it | |
653 | * so users can wire up its memory. (This can't go in qom/cpu.c | |
654 | * because that file is compiled only once for both user-mode | |
655 | * and system builds.) The default if no link is set up is to use | |
656 | * the system address space. | |
657 | */ | |
658 | object_property_add_link(OBJECT(cpu), "memory", TYPE_MEMORY_REGION, | |
659 | (Object **)&cpu->memory, | |
660 | qdev_prop_allow_set_link_before_realize, | |
661 | OBJ_PROP_LINK_UNREF_ON_RELEASE, | |
662 | &error_abort); | |
663 | cpu->memory = system_memory; | |
664 | object_ref(OBJECT(cpu->memory)); | |
665 | #endif | |
666 | ||
667 | cpu_list_lock(); | |
668 | if (cpu->cpu_index == UNASSIGNED_CPU_INDEX) { | |
669 | cpu->cpu_index = cpu_get_free_index(); | |
670 | assert(cpu->cpu_index != UNASSIGNED_CPU_INDEX); | |
671 | } else { | |
672 | assert(!cpu_index_auto_assigned); | |
673 | } | |
674 | QTAILQ_INSERT_TAIL(&cpus, cpu, node); | |
675 | cpu_list_unlock(); | |
676 | ||
677 | #ifndef CONFIG_USER_ONLY | |
678 | if (qdev_get_vmsd(DEVICE(cpu)) == NULL) { | |
679 | vmstate_register(NULL, cpu->cpu_index, &vmstate_cpu_common, cpu); | |
680 | } | |
681 | if (cc->vmsd != NULL) { | |
682 | vmstate_register(NULL, cpu->cpu_index, cc->vmsd, cpu); | |
683 | } | |
684 | #endif | |
685 | } | |
686 | ||
687 | #if defined(CONFIG_USER_ONLY) | |
688 | static void breakpoint_invalidate(CPUState *cpu, target_ulong pc) | |
689 | { | |
690 | tb_invalidate_phys_page_range(pc, pc + 1, 0); | |
691 | } | |
692 | #else | |
693 | static void breakpoint_invalidate(CPUState *cpu, target_ulong pc) | |
694 | { | |
695 | MemTxAttrs attrs; | |
696 | hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs); | |
697 | int asidx = cpu_asidx_from_attrs(cpu, attrs); | |
698 | if (phys != -1) { | |
699 | tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as, | |
700 | phys | (pc & ~TARGET_PAGE_MASK)); | |
701 | } | |
702 | } | |
703 | #endif | |
704 | ||
705 | #if defined(CONFIG_USER_ONLY) | |
706 | void cpu_watchpoint_remove_all(CPUState *cpu, int mask) | |
707 | ||
708 | { | |
709 | } | |
710 | ||
711 | int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len, | |
712 | int flags) | |
713 | { | |
714 | return -ENOSYS; | |
715 | } | |
716 | ||
717 | void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint) | |
718 | { | |
719 | } | |
720 | ||
721 | int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len, | |
722 | int flags, CPUWatchpoint **watchpoint) | |
723 | { | |
724 | return -ENOSYS; | |
725 | } | |
726 | #else | |
727 | /* Add a watchpoint. */ | |
728 | int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len, | |
729 | int flags, CPUWatchpoint **watchpoint) | |
730 | { | |
731 | CPUWatchpoint *wp; | |
732 | ||
733 | /* forbid ranges which are empty or run off the end of the address space */ | |
734 | if (len == 0 || (addr + len - 1) < addr) { | |
735 | error_report("tried to set invalid watchpoint at %" | |
736 | VADDR_PRIx ", len=%" VADDR_PRIu, addr, len); | |
737 | return -EINVAL; | |
738 | } | |
739 | wp = g_malloc(sizeof(*wp)); | |
740 | ||
741 | wp->vaddr = addr; | |
742 | wp->len = len; | |
743 | wp->flags = flags; | |
744 | ||
745 | /* keep all GDB-injected watchpoints in front */ | |
746 | if (flags & BP_GDB) { | |
747 | QTAILQ_INSERT_HEAD(&cpu->watchpoints, wp, entry); | |
748 | } else { | |
749 | QTAILQ_INSERT_TAIL(&cpu->watchpoints, wp, entry); | |
750 | } | |
751 | ||
752 | tlb_flush_page(cpu, addr); | |
753 | ||
754 | if (watchpoint) | |
755 | *watchpoint = wp; | |
756 | return 0; | |
757 | } | |
758 | ||
759 | /* Remove a specific watchpoint. */ | |
760 | int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len, | |
761 | int flags) | |
762 | { | |
763 | CPUWatchpoint *wp; | |
764 | ||
765 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { | |
766 | if (addr == wp->vaddr && len == wp->len | |
767 | && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) { | |
768 | cpu_watchpoint_remove_by_ref(cpu, wp); | |
769 | return 0; | |
770 | } | |
771 | } | |
772 | return -ENOENT; | |
773 | } | |
774 | ||
775 | /* Remove a specific watchpoint by reference. */ | |
776 | void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint) | |
777 | { | |
778 | QTAILQ_REMOVE(&cpu->watchpoints, watchpoint, entry); | |
779 | ||
780 | tlb_flush_page(cpu, watchpoint->vaddr); | |
781 | ||
782 | g_free(watchpoint); | |
783 | } | |
784 | ||
785 | /* Remove all matching watchpoints. */ | |
786 | void cpu_watchpoint_remove_all(CPUState *cpu, int mask) | |
787 | { | |
788 | CPUWatchpoint *wp, *next; | |
789 | ||
790 | QTAILQ_FOREACH_SAFE(wp, &cpu->watchpoints, entry, next) { | |
791 | if (wp->flags & mask) { | |
792 | cpu_watchpoint_remove_by_ref(cpu, wp); | |
793 | } | |
794 | } | |
795 | } | |
796 | ||
797 | /* Return true if this watchpoint address matches the specified | |
798 | * access (ie the address range covered by the watchpoint overlaps | |
799 | * partially or completely with the address range covered by the | |
800 | * access). | |
801 | */ | |
802 | static inline bool cpu_watchpoint_address_matches(CPUWatchpoint *wp, | |
803 | vaddr addr, | |
804 | vaddr len) | |
805 | { | |
806 | /* We know the lengths are non-zero, but a little caution is | |
807 | * required to avoid errors in the case where the range ends | |
808 | * exactly at the top of the address space and so addr + len | |
809 | * wraps round to zero. | |
810 | */ | |
811 | vaddr wpend = wp->vaddr + wp->len - 1; | |
812 | vaddr addrend = addr + len - 1; | |
813 | ||
814 | return !(addr > wpend || wp->vaddr > addrend); | |
815 | } | |
816 | ||
817 | #endif | |
818 | ||
819 | /* Add a breakpoint. */ | |
820 | int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags, | |
821 | CPUBreakpoint **breakpoint) | |
822 | { | |
823 | CPUBreakpoint *bp; | |
824 | ||
825 | bp = g_malloc(sizeof(*bp)); | |
826 | ||
827 | bp->pc = pc; | |
828 | bp->flags = flags; | |
829 | ||
830 | /* keep all GDB-injected breakpoints in front */ | |
831 | if (flags & BP_GDB) { | |
832 | QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry); | |
833 | } else { | |
834 | QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry); | |
835 | } | |
836 | ||
837 | breakpoint_invalidate(cpu, pc); | |
838 | ||
839 | if (breakpoint) { | |
840 | *breakpoint = bp; | |
841 | } | |
842 | return 0; | |
843 | } | |
844 | ||
845 | /* Remove a specific breakpoint. */ | |
846 | int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags) | |
847 | { | |
848 | CPUBreakpoint *bp; | |
849 | ||
850 | QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) { | |
851 | if (bp->pc == pc && bp->flags == flags) { | |
852 | cpu_breakpoint_remove_by_ref(cpu, bp); | |
853 | return 0; | |
854 | } | |
855 | } | |
856 | return -ENOENT; | |
857 | } | |
858 | ||
859 | /* Remove a specific breakpoint by reference. */ | |
860 | void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *breakpoint) | |
861 | { | |
862 | QTAILQ_REMOVE(&cpu->breakpoints, breakpoint, entry); | |
863 | ||
864 | breakpoint_invalidate(cpu, breakpoint->pc); | |
865 | ||
866 | g_free(breakpoint); | |
867 | } | |
868 | ||
869 | /* Remove all matching breakpoints. */ | |
870 | void cpu_breakpoint_remove_all(CPUState *cpu, int mask) | |
871 | { | |
872 | CPUBreakpoint *bp, *next; | |
873 | ||
874 | QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) { | |
875 | if (bp->flags & mask) { | |
876 | cpu_breakpoint_remove_by_ref(cpu, bp); | |
877 | } | |
878 | } | |
879 | } | |
880 | ||
881 | /* enable or disable single step mode. EXCP_DEBUG is returned by the | |
882 | CPU loop after each instruction */ | |
883 | void cpu_single_step(CPUState *cpu, int enabled) | |
884 | { | |
885 | if (cpu->singlestep_enabled != enabled) { | |
886 | cpu->singlestep_enabled = enabled; | |
887 | if (kvm_enabled()) { | |
888 | kvm_update_guest_debug(cpu, 0); | |
889 | } else { | |
890 | /* must flush all the translated code to avoid inconsistencies */ | |
891 | /* XXX: only flush what is necessary */ | |
892 | tb_flush(cpu); | |
893 | } | |
894 | } | |
895 | } | |
896 | ||
897 | void cpu_abort(CPUState *cpu, const char *fmt, ...) | |
898 | { | |
899 | va_list ap; | |
900 | va_list ap2; | |
901 | ||
902 | va_start(ap, fmt); | |
903 | va_copy(ap2, ap); | |
904 | fprintf(stderr, "qemu: fatal: "); | |
905 | vfprintf(stderr, fmt, ap); | |
906 | fprintf(stderr, "\n"); | |
907 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP); | |
908 | if (qemu_log_separate()) { | |
909 | qemu_log("qemu: fatal: "); | |
910 | qemu_log_vprintf(fmt, ap2); | |
911 | qemu_log("\n"); | |
912 | log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP); | |
913 | qemu_log_flush(); | |
914 | qemu_log_close(); | |
915 | } | |
916 | va_end(ap2); | |
917 | va_end(ap); | |
918 | replay_finish(); | |
919 | #if defined(CONFIG_USER_ONLY) | |
920 | { | |
921 | struct sigaction act; | |
922 | sigfillset(&act.sa_mask); | |
923 | act.sa_handler = SIG_DFL; | |
924 | sigaction(SIGABRT, &act, NULL); | |
925 | } | |
926 | #endif | |
927 | abort(); | |
928 | } | |
929 | ||
930 | #if !defined(CONFIG_USER_ONLY) | |
931 | /* Called from RCU critical section */ | |
932 | static RAMBlock *qemu_get_ram_block(ram_addr_t addr) | |
933 | { | |
934 | RAMBlock *block; | |
935 | ||
936 | block = atomic_rcu_read(&ram_list.mru_block); | |
937 | if (block && addr - block->offset < block->max_length) { | |
938 | return block; | |
939 | } | |
940 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
941 | if (addr - block->offset < block->max_length) { | |
942 | goto found; | |
943 | } | |
944 | } | |
945 | ||
946 | fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr); | |
947 | abort(); | |
948 | ||
949 | found: | |
950 | /* It is safe to write mru_block outside the iothread lock. This | |
951 | * is what happens: | |
952 | * | |
953 | * mru_block = xxx | |
954 | * rcu_read_unlock() | |
955 | * xxx removed from list | |
956 | * rcu_read_lock() | |
957 | * read mru_block | |
958 | * mru_block = NULL; | |
959 | * call_rcu(reclaim_ramblock, xxx); | |
960 | * rcu_read_unlock() | |
961 | * | |
962 | * atomic_rcu_set is not needed here. The block was already published | |
963 | * when it was placed into the list. Here we're just making an extra | |
964 | * copy of the pointer. | |
965 | */ | |
966 | ram_list.mru_block = block; | |
967 | return block; | |
968 | } | |
969 | ||
970 | static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length) | |
971 | { | |
972 | CPUState *cpu; | |
973 | ram_addr_t start1; | |
974 | RAMBlock *block; | |
975 | ram_addr_t end; | |
976 | ||
977 | end = TARGET_PAGE_ALIGN(start + length); | |
978 | start &= TARGET_PAGE_MASK; | |
979 | ||
980 | rcu_read_lock(); | |
981 | block = qemu_get_ram_block(start); | |
982 | assert(block == qemu_get_ram_block(end - 1)); | |
983 | start1 = (uintptr_t)ramblock_ptr(block, start - block->offset); | |
984 | CPU_FOREACH(cpu) { | |
985 | tlb_reset_dirty(cpu, start1, length); | |
986 | } | |
987 | rcu_read_unlock(); | |
988 | } | |
989 | ||
990 | /* Note: start and end must be within the same ram block. */ | |
991 | bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start, | |
992 | ram_addr_t length, | |
993 | unsigned client) | |
994 | { | |
995 | DirtyMemoryBlocks *blocks; | |
996 | unsigned long end, page; | |
997 | bool dirty = false; | |
998 | ||
999 | if (length == 0) { | |
1000 | return false; | |
1001 | } | |
1002 | ||
1003 | end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS; | |
1004 | page = start >> TARGET_PAGE_BITS; | |
1005 | ||
1006 | rcu_read_lock(); | |
1007 | ||
1008 | blocks = atomic_rcu_read(&ram_list.dirty_memory[client]); | |
1009 | ||
1010 | while (page < end) { | |
1011 | unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE; | |
1012 | unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE; | |
1013 | unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset); | |
1014 | ||
1015 | dirty |= bitmap_test_and_clear_atomic(blocks->blocks[idx], | |
1016 | offset, num); | |
1017 | page += num; | |
1018 | } | |
1019 | ||
1020 | rcu_read_unlock(); | |
1021 | ||
1022 | if (dirty && tcg_enabled()) { | |
1023 | tlb_reset_dirty_range_all(start, length); | |
1024 | } | |
1025 | ||
1026 | return dirty; | |
1027 | } | |
1028 | ||
1029 | /* Called from RCU critical section */ | |
1030 | hwaddr memory_region_section_get_iotlb(CPUState *cpu, | |
1031 | MemoryRegionSection *section, | |
1032 | target_ulong vaddr, | |
1033 | hwaddr paddr, hwaddr xlat, | |
1034 | int prot, | |
1035 | target_ulong *address) | |
1036 | { | |
1037 | hwaddr iotlb; | |
1038 | CPUWatchpoint *wp; | |
1039 | ||
1040 | if (memory_region_is_ram(section->mr)) { | |
1041 | /* Normal RAM. */ | |
1042 | iotlb = memory_region_get_ram_addr(section->mr) + xlat; | |
1043 | if (!section->readonly) { | |
1044 | iotlb |= PHYS_SECTION_NOTDIRTY; | |
1045 | } else { | |
1046 | iotlb |= PHYS_SECTION_ROM; | |
1047 | } | |
1048 | } else { | |
1049 | AddressSpaceDispatch *d; | |
1050 | ||
1051 | d = atomic_rcu_read(§ion->address_space->dispatch); | |
1052 | iotlb = section - d->map.sections; | |
1053 | iotlb += xlat; | |
1054 | } | |
1055 | ||
1056 | /* Make accesses to pages with watchpoints go via the | |
1057 | watchpoint trap routines. */ | |
1058 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { | |
1059 | if (cpu_watchpoint_address_matches(wp, vaddr, TARGET_PAGE_SIZE)) { | |
1060 | /* Avoid trapping reads of pages with a write breakpoint. */ | |
1061 | if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) { | |
1062 | iotlb = PHYS_SECTION_WATCH + paddr; | |
1063 | *address |= TLB_MMIO; | |
1064 | break; | |
1065 | } | |
1066 | } | |
1067 | } | |
1068 | ||
1069 | return iotlb; | |
1070 | } | |
1071 | #endif /* defined(CONFIG_USER_ONLY) */ | |
1072 | ||
1073 | #if !defined(CONFIG_USER_ONLY) | |
1074 | ||
1075 | static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end, | |
1076 | uint16_t section); | |
1077 | static subpage_t *subpage_init(AddressSpace *as, hwaddr base); | |
1078 | ||
1079 | static void *(*phys_mem_alloc)(size_t size, uint64_t *align) = | |
1080 | qemu_anon_ram_alloc; | |
1081 | ||
1082 | /* | |
1083 | * Set a custom physical guest memory alloator. | |
1084 | * Accelerators with unusual needs may need this. Hopefully, we can | |
1085 | * get rid of it eventually. | |
1086 | */ | |
1087 | void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align)) | |
1088 | { | |
1089 | phys_mem_alloc = alloc; | |
1090 | } | |
1091 | ||
1092 | static uint16_t phys_section_add(PhysPageMap *map, | |
1093 | MemoryRegionSection *section) | |
1094 | { | |
1095 | /* The physical section number is ORed with a page-aligned | |
1096 | * pointer to produce the iotlb entries. Thus it should | |
1097 | * never overflow into the page-aligned value. | |
1098 | */ | |
1099 | assert(map->sections_nb < TARGET_PAGE_SIZE); | |
1100 | ||
1101 | if (map->sections_nb == map->sections_nb_alloc) { | |
1102 | map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16); | |
1103 | map->sections = g_renew(MemoryRegionSection, map->sections, | |
1104 | map->sections_nb_alloc); | |
1105 | } | |
1106 | map->sections[map->sections_nb] = *section; | |
1107 | memory_region_ref(section->mr); | |
1108 | return map->sections_nb++; | |
1109 | } | |
1110 | ||
1111 | static void phys_section_destroy(MemoryRegion *mr) | |
1112 | { | |
1113 | bool have_sub_page = mr->subpage; | |
1114 | ||
1115 | memory_region_unref(mr); | |
1116 | ||
1117 | if (have_sub_page) { | |
1118 | subpage_t *subpage = container_of(mr, subpage_t, iomem); | |
1119 | object_unref(OBJECT(&subpage->iomem)); | |
1120 | g_free(subpage); | |
1121 | } | |
1122 | } | |
1123 | ||
1124 | static void phys_sections_free(PhysPageMap *map) | |
1125 | { | |
1126 | while (map->sections_nb > 0) { | |
1127 | MemoryRegionSection *section = &map->sections[--map->sections_nb]; | |
1128 | phys_section_destroy(section->mr); | |
1129 | } | |
1130 | g_free(map->sections); | |
1131 | g_free(map->nodes); | |
1132 | } | |
1133 | ||
1134 | static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section) | |
1135 | { | |
1136 | subpage_t *subpage; | |
1137 | hwaddr base = section->offset_within_address_space | |
1138 | & TARGET_PAGE_MASK; | |
1139 | MemoryRegionSection *existing = phys_page_find(d->phys_map, base, | |
1140 | d->map.nodes, d->map.sections); | |
1141 | MemoryRegionSection subsection = { | |
1142 | .offset_within_address_space = base, | |
1143 | .size = int128_make64(TARGET_PAGE_SIZE), | |
1144 | }; | |
1145 | hwaddr start, end; | |
1146 | ||
1147 | assert(existing->mr->subpage || existing->mr == &io_mem_unassigned); | |
1148 | ||
1149 | if (!(existing->mr->subpage)) { | |
1150 | subpage = subpage_init(d->as, base); | |
1151 | subsection.address_space = d->as; | |
1152 | subsection.mr = &subpage->iomem; | |
1153 | phys_page_set(d, base >> TARGET_PAGE_BITS, 1, | |
1154 | phys_section_add(&d->map, &subsection)); | |
1155 | } else { | |
1156 | subpage = container_of(existing->mr, subpage_t, iomem); | |
1157 | } | |
1158 | start = section->offset_within_address_space & ~TARGET_PAGE_MASK; | |
1159 | end = start + int128_get64(section->size) - 1; | |
1160 | subpage_register(subpage, start, end, | |
1161 | phys_section_add(&d->map, section)); | |
1162 | } | |
1163 | ||
1164 | ||
1165 | static void register_multipage(AddressSpaceDispatch *d, | |
1166 | MemoryRegionSection *section) | |
1167 | { | |
1168 | hwaddr start_addr = section->offset_within_address_space; | |
1169 | uint16_t section_index = phys_section_add(&d->map, section); | |
1170 | uint64_t num_pages = int128_get64(int128_rshift(section->size, | |
1171 | TARGET_PAGE_BITS)); | |
1172 | ||
1173 | assert(num_pages); | |
1174 | phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index); | |
1175 | } | |
1176 | ||
1177 | static void mem_add(MemoryListener *listener, MemoryRegionSection *section) | |
1178 | { | |
1179 | AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener); | |
1180 | AddressSpaceDispatch *d = as->next_dispatch; | |
1181 | MemoryRegionSection now = *section, remain = *section; | |
1182 | Int128 page_size = int128_make64(TARGET_PAGE_SIZE); | |
1183 | ||
1184 | if (now.offset_within_address_space & ~TARGET_PAGE_MASK) { | |
1185 | uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space) | |
1186 | - now.offset_within_address_space; | |
1187 | ||
1188 | now.size = int128_min(int128_make64(left), now.size); | |
1189 | register_subpage(d, &now); | |
1190 | } else { | |
1191 | now.size = int128_zero(); | |
1192 | } | |
1193 | while (int128_ne(remain.size, now.size)) { | |
1194 | remain.size = int128_sub(remain.size, now.size); | |
1195 | remain.offset_within_address_space += int128_get64(now.size); | |
1196 | remain.offset_within_region += int128_get64(now.size); | |
1197 | now = remain; | |
1198 | if (int128_lt(remain.size, page_size)) { | |
1199 | register_subpage(d, &now); | |
1200 | } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) { | |
1201 | now.size = page_size; | |
1202 | register_subpage(d, &now); | |
1203 | } else { | |
1204 | now.size = int128_and(now.size, int128_neg(page_size)); | |
1205 | register_multipage(d, &now); | |
1206 | } | |
1207 | } | |
1208 | } | |
1209 | ||
1210 | void qemu_flush_coalesced_mmio_buffer(void) | |
1211 | { | |
1212 | if (kvm_enabled()) | |
1213 | kvm_flush_coalesced_mmio_buffer(); | |
1214 | } | |
1215 | ||
1216 | void qemu_mutex_lock_ramlist(void) | |
1217 | { | |
1218 | qemu_mutex_lock(&ram_list.mutex); | |
1219 | } | |
1220 | ||
1221 | void qemu_mutex_unlock_ramlist(void) | |
1222 | { | |
1223 | qemu_mutex_unlock(&ram_list.mutex); | |
1224 | } | |
1225 | ||
1226 | #ifdef __linux__ | |
1227 | static void *file_ram_alloc(RAMBlock *block, | |
1228 | ram_addr_t memory, | |
1229 | const char *path, | |
1230 | Error **errp) | |
1231 | { | |
1232 | bool unlink_on_error = false; | |
1233 | char *filename; | |
1234 | char *sanitized_name; | |
1235 | char *c; | |
1236 | void *area = MAP_FAILED; | |
1237 | int fd = -1; | |
1238 | int64_t page_size; | |
1239 | ||
1240 | if (kvm_enabled() && !kvm_has_sync_mmu()) { | |
1241 | error_setg(errp, | |
1242 | "host lacks kvm mmu notifiers, -mem-path unsupported"); | |
1243 | return NULL; | |
1244 | } | |
1245 | ||
1246 | for (;;) { | |
1247 | fd = open(path, O_RDWR); | |
1248 | if (fd >= 0) { | |
1249 | /* @path names an existing file, use it */ | |
1250 | break; | |
1251 | } | |
1252 | if (errno == ENOENT) { | |
1253 | /* @path names a file that doesn't exist, create it */ | |
1254 | fd = open(path, O_RDWR | O_CREAT | O_EXCL, 0644); | |
1255 | if (fd >= 0) { | |
1256 | unlink_on_error = true; | |
1257 | break; | |
1258 | } | |
1259 | } else if (errno == EISDIR) { | |
1260 | /* @path names a directory, create a file there */ | |
1261 | /* Make name safe to use with mkstemp by replacing '/' with '_'. */ | |
1262 | sanitized_name = g_strdup(memory_region_name(block->mr)); | |
1263 | for (c = sanitized_name; *c != '\0'; c++) { | |
1264 | if (*c == '/') { | |
1265 | *c = '_'; | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path, | |
1270 | sanitized_name); | |
1271 | g_free(sanitized_name); | |
1272 | ||
1273 | fd = mkstemp(filename); | |
1274 | if (fd >= 0) { | |
1275 | unlink(filename); | |
1276 | g_free(filename); | |
1277 | break; | |
1278 | } | |
1279 | g_free(filename); | |
1280 | } | |
1281 | if (errno != EEXIST && errno != EINTR) { | |
1282 | error_setg_errno(errp, errno, | |
1283 | "can't open backing store %s for guest RAM", | |
1284 | path); | |
1285 | goto error; | |
1286 | } | |
1287 | /* | |
1288 | * Try again on EINTR and EEXIST. The latter happens when | |
1289 | * something else creates the file between our two open(). | |
1290 | */ | |
1291 | } | |
1292 | ||
1293 | page_size = qemu_fd_getpagesize(fd); | |
1294 | block->mr->align = MAX(page_size, QEMU_VMALLOC_ALIGN); | |
1295 | ||
1296 | if (memory < page_size) { | |
1297 | error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to " | |
1298 | "or larger than page size 0x%" PRIx64, | |
1299 | memory, page_size); | |
1300 | goto error; | |
1301 | } | |
1302 | ||
1303 | memory = ROUND_UP(memory, page_size); | |
1304 | ||
1305 | /* | |
1306 | * ftruncate is not supported by hugetlbfs in older | |
1307 | * hosts, so don't bother bailing out on errors. | |
1308 | * If anything goes wrong with it under other filesystems, | |
1309 | * mmap will fail. | |
1310 | */ | |
1311 | if (ftruncate(fd, memory)) { | |
1312 | perror("ftruncate"); | |
1313 | } | |
1314 | ||
1315 | area = qemu_ram_mmap(fd, memory, block->mr->align, | |
1316 | block->flags & RAM_SHARED); | |
1317 | if (area == MAP_FAILED) { | |
1318 | error_setg_errno(errp, errno, | |
1319 | "unable to map backing store for guest RAM"); | |
1320 | goto error; | |
1321 | } | |
1322 | ||
1323 | if (mem_prealloc) { | |
1324 | os_mem_prealloc(fd, area, memory, errp); | |
1325 | if (errp && *errp) { | |
1326 | goto error; | |
1327 | } | |
1328 | } | |
1329 | ||
1330 | block->fd = fd; | |
1331 | return area; | |
1332 | ||
1333 | error: | |
1334 | if (area != MAP_FAILED) { | |
1335 | qemu_ram_munmap(area, memory); | |
1336 | } | |
1337 | if (unlink_on_error) { | |
1338 | unlink(path); | |
1339 | } | |
1340 | if (fd != -1) { | |
1341 | close(fd); | |
1342 | } | |
1343 | return NULL; | |
1344 | } | |
1345 | #endif | |
1346 | ||
1347 | /* Called with the ramlist lock held. */ | |
1348 | static ram_addr_t find_ram_offset(ram_addr_t size) | |
1349 | { | |
1350 | RAMBlock *block, *next_block; | |
1351 | ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX; | |
1352 | ||
1353 | assert(size != 0); /* it would hand out same offset multiple times */ | |
1354 | ||
1355 | if (QLIST_EMPTY_RCU(&ram_list.blocks)) { | |
1356 | return 0; | |
1357 | } | |
1358 | ||
1359 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1360 | ram_addr_t end, next = RAM_ADDR_MAX; | |
1361 | ||
1362 | end = block->offset + block->max_length; | |
1363 | ||
1364 | QLIST_FOREACH_RCU(next_block, &ram_list.blocks, next) { | |
1365 | if (next_block->offset >= end) { | |
1366 | next = MIN(next, next_block->offset); | |
1367 | } | |
1368 | } | |
1369 | if (next - end >= size && next - end < mingap) { | |
1370 | offset = end; | |
1371 | mingap = next - end; | |
1372 | } | |
1373 | } | |
1374 | ||
1375 | if (offset == RAM_ADDR_MAX) { | |
1376 | fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n", | |
1377 | (uint64_t)size); | |
1378 | abort(); | |
1379 | } | |
1380 | ||
1381 | return offset; | |
1382 | } | |
1383 | ||
1384 | ram_addr_t last_ram_offset(void) | |
1385 | { | |
1386 | RAMBlock *block; | |
1387 | ram_addr_t last = 0; | |
1388 | ||
1389 | rcu_read_lock(); | |
1390 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1391 | last = MAX(last, block->offset + block->max_length); | |
1392 | } | |
1393 | rcu_read_unlock(); | |
1394 | return last; | |
1395 | } | |
1396 | ||
1397 | static void qemu_ram_setup_dump(void *addr, ram_addr_t size) | |
1398 | { | |
1399 | int ret; | |
1400 | ||
1401 | /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */ | |
1402 | if (!machine_dump_guest_core(current_machine)) { | |
1403 | ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP); | |
1404 | if (ret) { | |
1405 | perror("qemu_madvise"); | |
1406 | fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, " | |
1407 | "but dump_guest_core=off specified\n"); | |
1408 | } | |
1409 | } | |
1410 | } | |
1411 | ||
1412 | const char *qemu_ram_get_idstr(RAMBlock *rb) | |
1413 | { | |
1414 | return rb->idstr; | |
1415 | } | |
1416 | ||
1417 | /* Called with iothread lock held. */ | |
1418 | void qemu_ram_set_idstr(RAMBlock *new_block, const char *name, DeviceState *dev) | |
1419 | { | |
1420 | RAMBlock *block; | |
1421 | ||
1422 | assert(new_block); | |
1423 | assert(!new_block->idstr[0]); | |
1424 | ||
1425 | if (dev) { | |
1426 | char *id = qdev_get_dev_path(dev); | |
1427 | if (id) { | |
1428 | snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id); | |
1429 | g_free(id); | |
1430 | } | |
1431 | } | |
1432 | pstrcat(new_block->idstr, sizeof(new_block->idstr), name); | |
1433 | ||
1434 | rcu_read_lock(); | |
1435 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1436 | if (block != new_block && | |
1437 | !strcmp(block->idstr, new_block->idstr)) { | |
1438 | fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n", | |
1439 | new_block->idstr); | |
1440 | abort(); | |
1441 | } | |
1442 | } | |
1443 | rcu_read_unlock(); | |
1444 | } | |
1445 | ||
1446 | /* Called with iothread lock held. */ | |
1447 | void qemu_ram_unset_idstr(RAMBlock *block) | |
1448 | { | |
1449 | /* FIXME: arch_init.c assumes that this is not called throughout | |
1450 | * migration. Ignore the problem since hot-unplug during migration | |
1451 | * does not work anyway. | |
1452 | */ | |
1453 | if (block) { | |
1454 | memset(block->idstr, 0, sizeof(block->idstr)); | |
1455 | } | |
1456 | } | |
1457 | ||
1458 | static int memory_try_enable_merging(void *addr, size_t len) | |
1459 | { | |
1460 | if (!machine_mem_merge(current_machine)) { | |
1461 | /* disabled by the user */ | |
1462 | return 0; | |
1463 | } | |
1464 | ||
1465 | return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE); | |
1466 | } | |
1467 | ||
1468 | /* Only legal before guest might have detected the memory size: e.g. on | |
1469 | * incoming migration, or right after reset. | |
1470 | * | |
1471 | * As memory core doesn't know how is memory accessed, it is up to | |
1472 | * resize callback to update device state and/or add assertions to detect | |
1473 | * misuse, if necessary. | |
1474 | */ | |
1475 | int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp) | |
1476 | { | |
1477 | assert(block); | |
1478 | ||
1479 | newsize = HOST_PAGE_ALIGN(newsize); | |
1480 | ||
1481 | if (block->used_length == newsize) { | |
1482 | return 0; | |
1483 | } | |
1484 | ||
1485 | if (!(block->flags & RAM_RESIZEABLE)) { | |
1486 | error_setg_errno(errp, EINVAL, | |
1487 | "Length mismatch: %s: 0x" RAM_ADDR_FMT | |
1488 | " in != 0x" RAM_ADDR_FMT, block->idstr, | |
1489 | newsize, block->used_length); | |
1490 | return -EINVAL; | |
1491 | } | |
1492 | ||
1493 | if (block->max_length < newsize) { | |
1494 | error_setg_errno(errp, EINVAL, | |
1495 | "Length too large: %s: 0x" RAM_ADDR_FMT | |
1496 | " > 0x" RAM_ADDR_FMT, block->idstr, | |
1497 | newsize, block->max_length); | |
1498 | return -EINVAL; | |
1499 | } | |
1500 | ||
1501 | cpu_physical_memory_clear_dirty_range(block->offset, block->used_length); | |
1502 | block->used_length = newsize; | |
1503 | cpu_physical_memory_set_dirty_range(block->offset, block->used_length, | |
1504 | DIRTY_CLIENTS_ALL); | |
1505 | memory_region_set_size(block->mr, newsize); | |
1506 | if (block->resized) { | |
1507 | block->resized(block->idstr, newsize, block->host); | |
1508 | } | |
1509 | return 0; | |
1510 | } | |
1511 | ||
1512 | /* Called with ram_list.mutex held */ | |
1513 | static void dirty_memory_extend(ram_addr_t old_ram_size, | |
1514 | ram_addr_t new_ram_size) | |
1515 | { | |
1516 | ram_addr_t old_num_blocks = DIV_ROUND_UP(old_ram_size, | |
1517 | DIRTY_MEMORY_BLOCK_SIZE); | |
1518 | ram_addr_t new_num_blocks = DIV_ROUND_UP(new_ram_size, | |
1519 | DIRTY_MEMORY_BLOCK_SIZE); | |
1520 | int i; | |
1521 | ||
1522 | /* Only need to extend if block count increased */ | |
1523 | if (new_num_blocks <= old_num_blocks) { | |
1524 | return; | |
1525 | } | |
1526 | ||
1527 | for (i = 0; i < DIRTY_MEMORY_NUM; i++) { | |
1528 | DirtyMemoryBlocks *old_blocks; | |
1529 | DirtyMemoryBlocks *new_blocks; | |
1530 | int j; | |
1531 | ||
1532 | old_blocks = atomic_rcu_read(&ram_list.dirty_memory[i]); | |
1533 | new_blocks = g_malloc(sizeof(*new_blocks) + | |
1534 | sizeof(new_blocks->blocks[0]) * new_num_blocks); | |
1535 | ||
1536 | if (old_num_blocks) { | |
1537 | memcpy(new_blocks->blocks, old_blocks->blocks, | |
1538 | old_num_blocks * sizeof(old_blocks->blocks[0])); | |
1539 | } | |
1540 | ||
1541 | for (j = old_num_blocks; j < new_num_blocks; j++) { | |
1542 | new_blocks->blocks[j] = bitmap_new(DIRTY_MEMORY_BLOCK_SIZE); | |
1543 | } | |
1544 | ||
1545 | atomic_rcu_set(&ram_list.dirty_memory[i], new_blocks); | |
1546 | ||
1547 | if (old_blocks) { | |
1548 | g_free_rcu(old_blocks, rcu); | |
1549 | } | |
1550 | } | |
1551 | } | |
1552 | ||
1553 | static void ram_block_add(RAMBlock *new_block, Error **errp) | |
1554 | { | |
1555 | RAMBlock *block; | |
1556 | RAMBlock *last_block = NULL; | |
1557 | ram_addr_t old_ram_size, new_ram_size; | |
1558 | Error *err = NULL; | |
1559 | ||
1560 | old_ram_size = last_ram_offset() >> TARGET_PAGE_BITS; | |
1561 | ||
1562 | qemu_mutex_lock_ramlist(); | |
1563 | new_block->offset = find_ram_offset(new_block->max_length); | |
1564 | ||
1565 | if (!new_block->host) { | |
1566 | if (xen_enabled()) { | |
1567 | xen_ram_alloc(new_block->offset, new_block->max_length, | |
1568 | new_block->mr, &err); | |
1569 | if (err) { | |
1570 | error_propagate(errp, err); | |
1571 | qemu_mutex_unlock_ramlist(); | |
1572 | return; | |
1573 | } | |
1574 | } else { | |
1575 | new_block->host = phys_mem_alloc(new_block->max_length, | |
1576 | &new_block->mr->align); | |
1577 | if (!new_block->host) { | |
1578 | error_setg_errno(errp, errno, | |
1579 | "cannot set up guest memory '%s'", | |
1580 | memory_region_name(new_block->mr)); | |
1581 | qemu_mutex_unlock_ramlist(); | |
1582 | return; | |
1583 | } | |
1584 | memory_try_enable_merging(new_block->host, new_block->max_length); | |
1585 | } | |
1586 | } | |
1587 | ||
1588 | new_ram_size = MAX(old_ram_size, | |
1589 | (new_block->offset + new_block->max_length) >> TARGET_PAGE_BITS); | |
1590 | if (new_ram_size > old_ram_size) { | |
1591 | migration_bitmap_extend(old_ram_size, new_ram_size); | |
1592 | dirty_memory_extend(old_ram_size, new_ram_size); | |
1593 | } | |
1594 | /* Keep the list sorted from biggest to smallest block. Unlike QTAILQ, | |
1595 | * QLIST (which has an RCU-friendly variant) does not have insertion at | |
1596 | * tail, so save the last element in last_block. | |
1597 | */ | |
1598 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1599 | last_block = block; | |
1600 | if (block->max_length < new_block->max_length) { | |
1601 | break; | |
1602 | } | |
1603 | } | |
1604 | if (block) { | |
1605 | QLIST_INSERT_BEFORE_RCU(block, new_block, next); | |
1606 | } else if (last_block) { | |
1607 | QLIST_INSERT_AFTER_RCU(last_block, new_block, next); | |
1608 | } else { /* list is empty */ | |
1609 | QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next); | |
1610 | } | |
1611 | ram_list.mru_block = NULL; | |
1612 | ||
1613 | /* Write list before version */ | |
1614 | smp_wmb(); | |
1615 | ram_list.version++; | |
1616 | qemu_mutex_unlock_ramlist(); | |
1617 | ||
1618 | cpu_physical_memory_set_dirty_range(new_block->offset, | |
1619 | new_block->used_length, | |
1620 | DIRTY_CLIENTS_ALL); | |
1621 | ||
1622 | if (new_block->host) { | |
1623 | qemu_ram_setup_dump(new_block->host, new_block->max_length); | |
1624 | qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_HUGEPAGE); | |
1625 | qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_DONTFORK); | |
1626 | if (kvm_enabled()) { | |
1627 | kvm_setup_guest_memory(new_block->host, new_block->max_length); | |
1628 | } | |
1629 | } | |
1630 | } | |
1631 | ||
1632 | #ifdef __linux__ | |
1633 | RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr, | |
1634 | bool share, const char *mem_path, | |
1635 | Error **errp) | |
1636 | { | |
1637 | RAMBlock *new_block; | |
1638 | Error *local_err = NULL; | |
1639 | ||
1640 | if (xen_enabled()) { | |
1641 | error_setg(errp, "-mem-path not supported with Xen"); | |
1642 | return NULL; | |
1643 | } | |
1644 | ||
1645 | if (phys_mem_alloc != qemu_anon_ram_alloc) { | |
1646 | /* | |
1647 | * file_ram_alloc() needs to allocate just like | |
1648 | * phys_mem_alloc, but we haven't bothered to provide | |
1649 | * a hook there. | |
1650 | */ | |
1651 | error_setg(errp, | |
1652 | "-mem-path not supported with this accelerator"); | |
1653 | return NULL; | |
1654 | } | |
1655 | ||
1656 | size = HOST_PAGE_ALIGN(size); | |
1657 | new_block = g_malloc0(sizeof(*new_block)); | |
1658 | new_block->mr = mr; | |
1659 | new_block->used_length = size; | |
1660 | new_block->max_length = size; | |
1661 | new_block->flags = share ? RAM_SHARED : 0; | |
1662 | new_block->host = file_ram_alloc(new_block, size, | |
1663 | mem_path, errp); | |
1664 | if (!new_block->host) { | |
1665 | g_free(new_block); | |
1666 | return NULL; | |
1667 | } | |
1668 | ||
1669 | ram_block_add(new_block, &local_err); | |
1670 | if (local_err) { | |
1671 | g_free(new_block); | |
1672 | error_propagate(errp, local_err); | |
1673 | return NULL; | |
1674 | } | |
1675 | return new_block; | |
1676 | } | |
1677 | #endif | |
1678 | ||
1679 | static | |
1680 | RAMBlock *qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size, | |
1681 | void (*resized)(const char*, | |
1682 | uint64_t length, | |
1683 | void *host), | |
1684 | void *host, bool resizeable, | |
1685 | MemoryRegion *mr, Error **errp) | |
1686 | { | |
1687 | RAMBlock *new_block; | |
1688 | Error *local_err = NULL; | |
1689 | ||
1690 | size = HOST_PAGE_ALIGN(size); | |
1691 | max_size = HOST_PAGE_ALIGN(max_size); | |
1692 | new_block = g_malloc0(sizeof(*new_block)); | |
1693 | new_block->mr = mr; | |
1694 | new_block->resized = resized; | |
1695 | new_block->used_length = size; | |
1696 | new_block->max_length = max_size; | |
1697 | assert(max_size >= size); | |
1698 | new_block->fd = -1; | |
1699 | new_block->host = host; | |
1700 | if (host) { | |
1701 | new_block->flags |= RAM_PREALLOC; | |
1702 | } | |
1703 | if (resizeable) { | |
1704 | new_block->flags |= RAM_RESIZEABLE; | |
1705 | } | |
1706 | ram_block_add(new_block, &local_err); | |
1707 | if (local_err) { | |
1708 | g_free(new_block); | |
1709 | error_propagate(errp, local_err); | |
1710 | return NULL; | |
1711 | } | |
1712 | return new_block; | |
1713 | } | |
1714 | ||
1715 | RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host, | |
1716 | MemoryRegion *mr, Error **errp) | |
1717 | { | |
1718 | return qemu_ram_alloc_internal(size, size, NULL, host, false, mr, errp); | |
1719 | } | |
1720 | ||
1721 | RAMBlock *qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr, Error **errp) | |
1722 | { | |
1723 | return qemu_ram_alloc_internal(size, size, NULL, NULL, false, mr, errp); | |
1724 | } | |
1725 | ||
1726 | RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t maxsz, | |
1727 | void (*resized)(const char*, | |
1728 | uint64_t length, | |
1729 | void *host), | |
1730 | MemoryRegion *mr, Error **errp) | |
1731 | { | |
1732 | return qemu_ram_alloc_internal(size, maxsz, resized, NULL, true, mr, errp); | |
1733 | } | |
1734 | ||
1735 | static void reclaim_ramblock(RAMBlock *block) | |
1736 | { | |
1737 | if (block->flags & RAM_PREALLOC) { | |
1738 | ; | |
1739 | } else if (xen_enabled()) { | |
1740 | xen_invalidate_map_cache_entry(block->host); | |
1741 | #ifndef _WIN32 | |
1742 | } else if (block->fd >= 0) { | |
1743 | qemu_ram_munmap(block->host, block->max_length); | |
1744 | close(block->fd); | |
1745 | #endif | |
1746 | } else { | |
1747 | qemu_anon_ram_free(block->host, block->max_length); | |
1748 | } | |
1749 | g_free(block); | |
1750 | } | |
1751 | ||
1752 | void qemu_ram_free(RAMBlock *block) | |
1753 | { | |
1754 | if (!block) { | |
1755 | return; | |
1756 | } | |
1757 | ||
1758 | qemu_mutex_lock_ramlist(); | |
1759 | QLIST_REMOVE_RCU(block, next); | |
1760 | ram_list.mru_block = NULL; | |
1761 | /* Write list before version */ | |
1762 | smp_wmb(); | |
1763 | ram_list.version++; | |
1764 | call_rcu(block, reclaim_ramblock, rcu); | |
1765 | qemu_mutex_unlock_ramlist(); | |
1766 | } | |
1767 | ||
1768 | #ifndef _WIN32 | |
1769 | void qemu_ram_remap(ram_addr_t addr, ram_addr_t length) | |
1770 | { | |
1771 | RAMBlock *block; | |
1772 | ram_addr_t offset; | |
1773 | int flags; | |
1774 | void *area, *vaddr; | |
1775 | ||
1776 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1777 | offset = addr - block->offset; | |
1778 | if (offset < block->max_length) { | |
1779 | vaddr = ramblock_ptr(block, offset); | |
1780 | if (block->flags & RAM_PREALLOC) { | |
1781 | ; | |
1782 | } else if (xen_enabled()) { | |
1783 | abort(); | |
1784 | } else { | |
1785 | flags = MAP_FIXED; | |
1786 | if (block->fd >= 0) { | |
1787 | flags |= (block->flags & RAM_SHARED ? | |
1788 | MAP_SHARED : MAP_PRIVATE); | |
1789 | area = mmap(vaddr, length, PROT_READ | PROT_WRITE, | |
1790 | flags, block->fd, offset); | |
1791 | } else { | |
1792 | /* | |
1793 | * Remap needs to match alloc. Accelerators that | |
1794 | * set phys_mem_alloc never remap. If they did, | |
1795 | * we'd need a remap hook here. | |
1796 | */ | |
1797 | assert(phys_mem_alloc == qemu_anon_ram_alloc); | |
1798 | ||
1799 | flags |= MAP_PRIVATE | MAP_ANONYMOUS; | |
1800 | area = mmap(vaddr, length, PROT_READ | PROT_WRITE, | |
1801 | flags, -1, 0); | |
1802 | } | |
1803 | if (area != vaddr) { | |
1804 | fprintf(stderr, "Could not remap addr: " | |
1805 | RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n", | |
1806 | length, addr); | |
1807 | exit(1); | |
1808 | } | |
1809 | memory_try_enable_merging(vaddr, length); | |
1810 | qemu_ram_setup_dump(vaddr, length); | |
1811 | } | |
1812 | } | |
1813 | } | |
1814 | } | |
1815 | #endif /* !_WIN32 */ | |
1816 | ||
1817 | /* Return a host pointer to ram allocated with qemu_ram_alloc. | |
1818 | * This should not be used for general purpose DMA. Use address_space_map | |
1819 | * or address_space_rw instead. For local memory (e.g. video ram) that the | |
1820 | * device owns, use memory_region_get_ram_ptr. | |
1821 | * | |
1822 | * Called within RCU critical section. | |
1823 | */ | |
1824 | void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr) | |
1825 | { | |
1826 | RAMBlock *block = ram_block; | |
1827 | ||
1828 | if (block == NULL) { | |
1829 | block = qemu_get_ram_block(addr); | |
1830 | addr -= block->offset; | |
1831 | } | |
1832 | ||
1833 | if (xen_enabled() && block->host == NULL) { | |
1834 | /* We need to check if the requested address is in the RAM | |
1835 | * because we don't want to map the entire memory in QEMU. | |
1836 | * In that case just map until the end of the page. | |
1837 | */ | |
1838 | if (block->offset == 0) { | |
1839 | return xen_map_cache(addr, 0, 0); | |
1840 | } | |
1841 | ||
1842 | block->host = xen_map_cache(block->offset, block->max_length, 1); | |
1843 | } | |
1844 | return ramblock_ptr(block, addr); | |
1845 | } | |
1846 | ||
1847 | /* Return a host pointer to guest's ram. Similar to qemu_map_ram_ptr | |
1848 | * but takes a size argument. | |
1849 | * | |
1850 | * Called within RCU critical section. | |
1851 | */ | |
1852 | static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr, | |
1853 | hwaddr *size) | |
1854 | { | |
1855 | RAMBlock *block = ram_block; | |
1856 | if (*size == 0) { | |
1857 | return NULL; | |
1858 | } | |
1859 | ||
1860 | if (block == NULL) { | |
1861 | block = qemu_get_ram_block(addr); | |
1862 | addr -= block->offset; | |
1863 | } | |
1864 | *size = MIN(*size, block->max_length - addr); | |
1865 | ||
1866 | if (xen_enabled() && block->host == NULL) { | |
1867 | /* We need to check if the requested address is in the RAM | |
1868 | * because we don't want to map the entire memory in QEMU. | |
1869 | * In that case just map the requested area. | |
1870 | */ | |
1871 | if (block->offset == 0) { | |
1872 | return xen_map_cache(addr, *size, 1); | |
1873 | } | |
1874 | ||
1875 | block->host = xen_map_cache(block->offset, block->max_length, 1); | |
1876 | } | |
1877 | ||
1878 | return ramblock_ptr(block, addr); | |
1879 | } | |
1880 | ||
1881 | /* | |
1882 | * Translates a host ptr back to a RAMBlock, a ram_addr and an offset | |
1883 | * in that RAMBlock. | |
1884 | * | |
1885 | * ptr: Host pointer to look up | |
1886 | * round_offset: If true round the result offset down to a page boundary | |
1887 | * *ram_addr: set to result ram_addr | |
1888 | * *offset: set to result offset within the RAMBlock | |
1889 | * | |
1890 | * Returns: RAMBlock (or NULL if not found) | |
1891 | * | |
1892 | * By the time this function returns, the returned pointer is not protected | |
1893 | * by RCU anymore. If the caller is not within an RCU critical section and | |
1894 | * does not hold the iothread lock, it must have other means of protecting the | |
1895 | * pointer, such as a reference to the region that includes the incoming | |
1896 | * ram_addr_t. | |
1897 | */ | |
1898 | RAMBlock *qemu_ram_block_from_host(void *ptr, bool round_offset, | |
1899 | ram_addr_t *offset) | |
1900 | { | |
1901 | RAMBlock *block; | |
1902 | uint8_t *host = ptr; | |
1903 | ||
1904 | if (xen_enabled()) { | |
1905 | ram_addr_t ram_addr; | |
1906 | rcu_read_lock(); | |
1907 | ram_addr = xen_ram_addr_from_mapcache(ptr); | |
1908 | block = qemu_get_ram_block(ram_addr); | |
1909 | if (block) { | |
1910 | *offset = ram_addr - block->offset; | |
1911 | } | |
1912 | rcu_read_unlock(); | |
1913 | return block; | |
1914 | } | |
1915 | ||
1916 | rcu_read_lock(); | |
1917 | block = atomic_rcu_read(&ram_list.mru_block); | |
1918 | if (block && block->host && host - block->host < block->max_length) { | |
1919 | goto found; | |
1920 | } | |
1921 | ||
1922 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1923 | /* This case append when the block is not mapped. */ | |
1924 | if (block->host == NULL) { | |
1925 | continue; | |
1926 | } | |
1927 | if (host - block->host < block->max_length) { | |
1928 | goto found; | |
1929 | } | |
1930 | } | |
1931 | ||
1932 | rcu_read_unlock(); | |
1933 | return NULL; | |
1934 | ||
1935 | found: | |
1936 | *offset = (host - block->host); | |
1937 | if (round_offset) { | |
1938 | *offset &= TARGET_PAGE_MASK; | |
1939 | } | |
1940 | rcu_read_unlock(); | |
1941 | return block; | |
1942 | } | |
1943 | ||
1944 | /* | |
1945 | * Finds the named RAMBlock | |
1946 | * | |
1947 | * name: The name of RAMBlock to find | |
1948 | * | |
1949 | * Returns: RAMBlock (or NULL if not found) | |
1950 | */ | |
1951 | RAMBlock *qemu_ram_block_by_name(const char *name) | |
1952 | { | |
1953 | RAMBlock *block; | |
1954 | ||
1955 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1956 | if (!strcmp(name, block->idstr)) { | |
1957 | return block; | |
1958 | } | |
1959 | } | |
1960 | ||
1961 | return NULL; | |
1962 | } | |
1963 | ||
1964 | /* Some of the softmmu routines need to translate from a host pointer | |
1965 | (typically a TLB entry) back to a ram offset. */ | |
1966 | ram_addr_t qemu_ram_addr_from_host(void *ptr) | |
1967 | { | |
1968 | RAMBlock *block; | |
1969 | ram_addr_t offset; | |
1970 | ||
1971 | block = qemu_ram_block_from_host(ptr, false, &offset); | |
1972 | if (!block) { | |
1973 | return RAM_ADDR_INVALID; | |
1974 | } | |
1975 | ||
1976 | return block->offset + offset; | |
1977 | } | |
1978 | ||
1979 | /* Called within RCU critical section. */ | |
1980 | static void notdirty_mem_write(void *opaque, hwaddr ram_addr, | |
1981 | uint64_t val, unsigned size) | |
1982 | { | |
1983 | if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) { | |
1984 | tb_invalidate_phys_page_fast(ram_addr, size); | |
1985 | } | |
1986 | switch (size) { | |
1987 | case 1: | |
1988 | stb_p(qemu_map_ram_ptr(NULL, ram_addr), val); | |
1989 | break; | |
1990 | case 2: | |
1991 | stw_p(qemu_map_ram_ptr(NULL, ram_addr), val); | |
1992 | break; | |
1993 | case 4: | |
1994 | stl_p(qemu_map_ram_ptr(NULL, ram_addr), val); | |
1995 | break; | |
1996 | default: | |
1997 | abort(); | |
1998 | } | |
1999 | /* Set both VGA and migration bits for simplicity and to remove | |
2000 | * the notdirty callback faster. | |
2001 | */ | |
2002 | cpu_physical_memory_set_dirty_range(ram_addr, size, | |
2003 | DIRTY_CLIENTS_NOCODE); | |
2004 | /* we remove the notdirty callback only if the code has been | |
2005 | flushed */ | |
2006 | if (!cpu_physical_memory_is_clean(ram_addr)) { | |
2007 | tlb_set_dirty(current_cpu, current_cpu->mem_io_vaddr); | |
2008 | } | |
2009 | } | |
2010 | ||
2011 | static bool notdirty_mem_accepts(void *opaque, hwaddr addr, | |
2012 | unsigned size, bool is_write) | |
2013 | { | |
2014 | return is_write; | |
2015 | } | |
2016 | ||
2017 | static const MemoryRegionOps notdirty_mem_ops = { | |
2018 | .write = notdirty_mem_write, | |
2019 | .valid.accepts = notdirty_mem_accepts, | |
2020 | .endianness = DEVICE_NATIVE_ENDIAN, | |
2021 | }; | |
2022 | ||
2023 | /* Generate a debug exception if a watchpoint has been hit. */ | |
2024 | static void check_watchpoint(int offset, int len, MemTxAttrs attrs, int flags) | |
2025 | { | |
2026 | CPUState *cpu = current_cpu; | |
2027 | CPUClass *cc = CPU_GET_CLASS(cpu); | |
2028 | CPUArchState *env = cpu->env_ptr; | |
2029 | target_ulong pc, cs_base; | |
2030 | target_ulong vaddr; | |
2031 | CPUWatchpoint *wp; | |
2032 | uint32_t cpu_flags; | |
2033 | ||
2034 | if (cpu->watchpoint_hit) { | |
2035 | /* We re-entered the check after replacing the TB. Now raise | |
2036 | * the debug interrupt so that is will trigger after the | |
2037 | * current instruction. */ | |
2038 | cpu_interrupt(cpu, CPU_INTERRUPT_DEBUG); | |
2039 | return; | |
2040 | } | |
2041 | vaddr = (cpu->mem_io_vaddr & TARGET_PAGE_MASK) + offset; | |
2042 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { | |
2043 | if (cpu_watchpoint_address_matches(wp, vaddr, len) | |
2044 | && (wp->flags & flags)) { | |
2045 | if (flags == BP_MEM_READ) { | |
2046 | wp->flags |= BP_WATCHPOINT_HIT_READ; | |
2047 | } else { | |
2048 | wp->flags |= BP_WATCHPOINT_HIT_WRITE; | |
2049 | } | |
2050 | wp->hitaddr = vaddr; | |
2051 | wp->hitattrs = attrs; | |
2052 | if (!cpu->watchpoint_hit) { | |
2053 | if (wp->flags & BP_CPU && | |
2054 | !cc->debug_check_watchpoint(cpu, wp)) { | |
2055 | wp->flags &= ~BP_WATCHPOINT_HIT; | |
2056 | continue; | |
2057 | } | |
2058 | cpu->watchpoint_hit = wp; | |
2059 | tb_check_watchpoint(cpu); | |
2060 | if (wp->flags & BP_STOP_BEFORE_ACCESS) { | |
2061 | cpu->exception_index = EXCP_DEBUG; | |
2062 | cpu_loop_exit(cpu); | |
2063 | } else { | |
2064 | cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags); | |
2065 | tb_gen_code(cpu, pc, cs_base, cpu_flags, 1); | |
2066 | cpu_loop_exit_noexc(cpu); | |
2067 | } | |
2068 | } | |
2069 | } else { | |
2070 | wp->flags &= ~BP_WATCHPOINT_HIT; | |
2071 | } | |
2072 | } | |
2073 | } | |
2074 | ||
2075 | /* Watchpoint access routines. Watchpoints are inserted using TLB tricks, | |
2076 | so these check for a hit then pass through to the normal out-of-line | |
2077 | phys routines. */ | |
2078 | static MemTxResult watch_mem_read(void *opaque, hwaddr addr, uint64_t *pdata, | |
2079 | unsigned size, MemTxAttrs attrs) | |
2080 | { | |
2081 | MemTxResult res; | |
2082 | uint64_t data; | |
2083 | int asidx = cpu_asidx_from_attrs(current_cpu, attrs); | |
2084 | AddressSpace *as = current_cpu->cpu_ases[asidx].as; | |
2085 | ||
2086 | check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_READ); | |
2087 | switch (size) { | |
2088 | case 1: | |
2089 | data = address_space_ldub(as, addr, attrs, &res); | |
2090 | break; | |
2091 | case 2: | |
2092 | data = address_space_lduw(as, addr, attrs, &res); | |
2093 | break; | |
2094 | case 4: | |
2095 | data = address_space_ldl(as, addr, attrs, &res); | |
2096 | break; | |
2097 | default: abort(); | |
2098 | } | |
2099 | *pdata = data; | |
2100 | return res; | |
2101 | } | |
2102 | ||
2103 | static MemTxResult watch_mem_write(void *opaque, hwaddr addr, | |
2104 | uint64_t val, unsigned size, | |
2105 | MemTxAttrs attrs) | |
2106 | { | |
2107 | MemTxResult res; | |
2108 | int asidx = cpu_asidx_from_attrs(current_cpu, attrs); | |
2109 | AddressSpace *as = current_cpu->cpu_ases[asidx].as; | |
2110 | ||
2111 | check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_WRITE); | |
2112 | switch (size) { | |
2113 | case 1: | |
2114 | address_space_stb(as, addr, val, attrs, &res); | |
2115 | break; | |
2116 | case 2: | |
2117 | address_space_stw(as, addr, val, attrs, &res); | |
2118 | break; | |
2119 | case 4: | |
2120 | address_space_stl(as, addr, val, attrs, &res); | |
2121 | break; | |
2122 | default: abort(); | |
2123 | } | |
2124 | return res; | |
2125 | } | |
2126 | ||
2127 | static const MemoryRegionOps watch_mem_ops = { | |
2128 | .read_with_attrs = watch_mem_read, | |
2129 | .write_with_attrs = watch_mem_write, | |
2130 | .endianness = DEVICE_NATIVE_ENDIAN, | |
2131 | }; | |
2132 | ||
2133 | static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data, | |
2134 | unsigned len, MemTxAttrs attrs) | |
2135 | { | |
2136 | subpage_t *subpage = opaque; | |
2137 | uint8_t buf[8]; | |
2138 | MemTxResult res; | |
2139 | ||
2140 | #if defined(DEBUG_SUBPAGE) | |
2141 | printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__, | |
2142 | subpage, len, addr); | |
2143 | #endif | |
2144 | res = address_space_read(subpage->as, addr + subpage->base, | |
2145 | attrs, buf, len); | |
2146 | if (res) { | |
2147 | return res; | |
2148 | } | |
2149 | switch (len) { | |
2150 | case 1: | |
2151 | *data = ldub_p(buf); | |
2152 | return MEMTX_OK; | |
2153 | case 2: | |
2154 | *data = lduw_p(buf); | |
2155 | return MEMTX_OK; | |
2156 | case 4: | |
2157 | *data = ldl_p(buf); | |
2158 | return MEMTX_OK; | |
2159 | case 8: | |
2160 | *data = ldq_p(buf); | |
2161 | return MEMTX_OK; | |
2162 | default: | |
2163 | abort(); | |
2164 | } | |
2165 | } | |
2166 | ||
2167 | static MemTxResult subpage_write(void *opaque, hwaddr addr, | |
2168 | uint64_t value, unsigned len, MemTxAttrs attrs) | |
2169 | { | |
2170 | subpage_t *subpage = opaque; | |
2171 | uint8_t buf[8]; | |
2172 | ||
2173 | #if defined(DEBUG_SUBPAGE) | |
2174 | printf("%s: subpage %p len %u addr " TARGET_FMT_plx | |
2175 | " value %"PRIx64"\n", | |
2176 | __func__, subpage, len, addr, value); | |
2177 | #endif | |
2178 | switch (len) { | |
2179 | case 1: | |
2180 | stb_p(buf, value); | |
2181 | break; | |
2182 | case 2: | |
2183 | stw_p(buf, value); | |
2184 | break; | |
2185 | case 4: | |
2186 | stl_p(buf, value); | |
2187 | break; | |
2188 | case 8: | |
2189 | stq_p(buf, value); | |
2190 | break; | |
2191 | default: | |
2192 | abort(); | |
2193 | } | |
2194 | return address_space_write(subpage->as, addr + subpage->base, | |
2195 | attrs, buf, len); | |
2196 | } | |
2197 | ||
2198 | static bool subpage_accepts(void *opaque, hwaddr addr, | |
2199 | unsigned len, bool is_write) | |
2200 | { | |
2201 | subpage_t *subpage = opaque; | |
2202 | #if defined(DEBUG_SUBPAGE) | |
2203 | printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n", | |
2204 | __func__, subpage, is_write ? 'w' : 'r', len, addr); | |
2205 | #endif | |
2206 | ||
2207 | return address_space_access_valid(subpage->as, addr + subpage->base, | |
2208 | len, is_write); | |
2209 | } | |
2210 | ||
2211 | static const MemoryRegionOps subpage_ops = { | |
2212 | .read_with_attrs = subpage_read, | |
2213 | .write_with_attrs = subpage_write, | |
2214 | .impl.min_access_size = 1, | |
2215 | .impl.max_access_size = 8, | |
2216 | .valid.min_access_size = 1, | |
2217 | .valid.max_access_size = 8, | |
2218 | .valid.accepts = subpage_accepts, | |
2219 | .endianness = DEVICE_NATIVE_ENDIAN, | |
2220 | }; | |
2221 | ||
2222 | static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end, | |
2223 | uint16_t section) | |
2224 | { | |
2225 | int idx, eidx; | |
2226 | ||
2227 | if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE) | |
2228 | return -1; | |
2229 | idx = SUBPAGE_IDX(start); | |
2230 | eidx = SUBPAGE_IDX(end); | |
2231 | #if defined(DEBUG_SUBPAGE) | |
2232 | printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n", | |
2233 | __func__, mmio, start, end, idx, eidx, section); | |
2234 | #endif | |
2235 | for (; idx <= eidx; idx++) { | |
2236 | mmio->sub_section[idx] = section; | |
2237 | } | |
2238 | ||
2239 | return 0; | |
2240 | } | |
2241 | ||
2242 | static subpage_t *subpage_init(AddressSpace *as, hwaddr base) | |
2243 | { | |
2244 | subpage_t *mmio; | |
2245 | ||
2246 | mmio = g_malloc0(sizeof(subpage_t)); | |
2247 | ||
2248 | mmio->as = as; | |
2249 | mmio->base = base; | |
2250 | memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio, | |
2251 | NULL, TARGET_PAGE_SIZE); | |
2252 | mmio->iomem.subpage = true; | |
2253 | #if defined(DEBUG_SUBPAGE) | |
2254 | printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__, | |
2255 | mmio, base, TARGET_PAGE_SIZE); | |
2256 | #endif | |
2257 | subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED); | |
2258 | ||
2259 | return mmio; | |
2260 | } | |
2261 | ||
2262 | static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as, | |
2263 | MemoryRegion *mr) | |
2264 | { | |
2265 | assert(as); | |
2266 | MemoryRegionSection section = { | |
2267 | .address_space = as, | |
2268 | .mr = mr, | |
2269 | .offset_within_address_space = 0, | |
2270 | .offset_within_region = 0, | |
2271 | .size = int128_2_64(), | |
2272 | }; | |
2273 | ||
2274 | return phys_section_add(map, §ion); | |
2275 | } | |
2276 | ||
2277 | MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index, MemTxAttrs attrs) | |
2278 | { | |
2279 | int asidx = cpu_asidx_from_attrs(cpu, attrs); | |
2280 | CPUAddressSpace *cpuas = &cpu->cpu_ases[asidx]; | |
2281 | AddressSpaceDispatch *d = atomic_rcu_read(&cpuas->memory_dispatch); | |
2282 | MemoryRegionSection *sections = d->map.sections; | |
2283 | ||
2284 | return sections[index & ~TARGET_PAGE_MASK].mr; | |
2285 | } | |
2286 | ||
2287 | static void io_mem_init(void) | |
2288 | { | |
2289 | memory_region_init_io(&io_mem_rom, NULL, &unassigned_mem_ops, NULL, NULL, UINT64_MAX); | |
2290 | memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL, | |
2291 | NULL, UINT64_MAX); | |
2292 | memory_region_init_io(&io_mem_notdirty, NULL, ¬dirty_mem_ops, NULL, | |
2293 | NULL, UINT64_MAX); | |
2294 | memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL, | |
2295 | NULL, UINT64_MAX); | |
2296 | } | |
2297 | ||
2298 | static void mem_begin(MemoryListener *listener) | |
2299 | { | |
2300 | AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener); | |
2301 | AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1); | |
2302 | uint16_t n; | |
2303 | ||
2304 | n = dummy_section(&d->map, as, &io_mem_unassigned); | |
2305 | assert(n == PHYS_SECTION_UNASSIGNED); | |
2306 | n = dummy_section(&d->map, as, &io_mem_notdirty); | |
2307 | assert(n == PHYS_SECTION_NOTDIRTY); | |
2308 | n = dummy_section(&d->map, as, &io_mem_rom); | |
2309 | assert(n == PHYS_SECTION_ROM); | |
2310 | n = dummy_section(&d->map, as, &io_mem_watch); | |
2311 | assert(n == PHYS_SECTION_WATCH); | |
2312 | ||
2313 | d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 }; | |
2314 | d->as = as; | |
2315 | as->next_dispatch = d; | |
2316 | } | |
2317 | ||
2318 | static void address_space_dispatch_free(AddressSpaceDispatch *d) | |
2319 | { | |
2320 | phys_sections_free(&d->map); | |
2321 | g_free(d); | |
2322 | } | |
2323 | ||
2324 | static void mem_commit(MemoryListener *listener) | |
2325 | { | |
2326 | AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener); | |
2327 | AddressSpaceDispatch *cur = as->dispatch; | |
2328 | AddressSpaceDispatch *next = as->next_dispatch; | |
2329 | ||
2330 | phys_page_compact_all(next, next->map.nodes_nb); | |
2331 | ||
2332 | atomic_rcu_set(&as->dispatch, next); | |
2333 | if (cur) { | |
2334 | call_rcu(cur, address_space_dispatch_free, rcu); | |
2335 | } | |
2336 | } | |
2337 | ||
2338 | static void tcg_commit(MemoryListener *listener) | |
2339 | { | |
2340 | CPUAddressSpace *cpuas; | |
2341 | AddressSpaceDispatch *d; | |
2342 | ||
2343 | /* since each CPU stores ram addresses in its TLB cache, we must | |
2344 | reset the modified entries */ | |
2345 | cpuas = container_of(listener, CPUAddressSpace, tcg_as_listener); | |
2346 | cpu_reloading_memory_map(); | |
2347 | /* The CPU and TLB are protected by the iothread lock. | |
2348 | * We reload the dispatch pointer now because cpu_reloading_memory_map() | |
2349 | * may have split the RCU critical section. | |
2350 | */ | |
2351 | d = atomic_rcu_read(&cpuas->as->dispatch); | |
2352 | cpuas->memory_dispatch = d; | |
2353 | tlb_flush(cpuas->cpu, 1); | |
2354 | } | |
2355 | ||
2356 | void address_space_init_dispatch(AddressSpace *as) | |
2357 | { | |
2358 | as->dispatch = NULL; | |
2359 | as->dispatch_listener = (MemoryListener) { | |
2360 | .begin = mem_begin, | |
2361 | .commit = mem_commit, | |
2362 | .region_add = mem_add, | |
2363 | .region_nop = mem_add, | |
2364 | .priority = 0, | |
2365 | }; | |
2366 | memory_listener_register(&as->dispatch_listener, as); | |
2367 | } | |
2368 | ||
2369 | void address_space_unregister(AddressSpace *as) | |
2370 | { | |
2371 | memory_listener_unregister(&as->dispatch_listener); | |
2372 | } | |
2373 | ||
2374 | void address_space_destroy_dispatch(AddressSpace *as) | |
2375 | { | |
2376 | AddressSpaceDispatch *d = as->dispatch; | |
2377 | ||
2378 | atomic_rcu_set(&as->dispatch, NULL); | |
2379 | if (d) { | |
2380 | call_rcu(d, address_space_dispatch_free, rcu); | |
2381 | } | |
2382 | } | |
2383 | ||
2384 | static void memory_map_init(void) | |
2385 | { | |
2386 | system_memory = g_malloc(sizeof(*system_memory)); | |
2387 | ||
2388 | memory_region_init(system_memory, NULL, "system", UINT64_MAX); | |
2389 | address_space_init(&address_space_memory, system_memory, "memory"); | |
2390 | ||
2391 | system_io = g_malloc(sizeof(*system_io)); | |
2392 | memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io", | |
2393 | 65536); | |
2394 | address_space_init(&address_space_io, system_io, "I/O"); | |
2395 | } | |
2396 | ||
2397 | MemoryRegion *get_system_memory(void) | |
2398 | { | |
2399 | return system_memory; | |
2400 | } | |
2401 | ||
2402 | MemoryRegion *get_system_io(void) | |
2403 | { | |
2404 | return system_io; | |
2405 | } | |
2406 | ||
2407 | #endif /* !defined(CONFIG_USER_ONLY) */ | |
2408 | ||
2409 | /* physical memory access (slow version, mainly for debug) */ | |
2410 | #if defined(CONFIG_USER_ONLY) | |
2411 | int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr, | |
2412 | uint8_t *buf, int len, int is_write) | |
2413 | { | |
2414 | int l, flags; | |
2415 | target_ulong page; | |
2416 | void * p; | |
2417 | ||
2418 | while (len > 0) { | |
2419 | page = addr & TARGET_PAGE_MASK; | |
2420 | l = (page + TARGET_PAGE_SIZE) - addr; | |
2421 | if (l > len) | |
2422 | l = len; | |
2423 | flags = page_get_flags(page); | |
2424 | if (!(flags & PAGE_VALID)) | |
2425 | return -1; | |
2426 | if (is_write) { | |
2427 | if (!(flags & PAGE_WRITE)) | |
2428 | return -1; | |
2429 | /* XXX: this code should not depend on lock_user */ | |
2430 | if (!(p = lock_user(VERIFY_WRITE, addr, l, 0))) | |
2431 | return -1; | |
2432 | memcpy(p, buf, l); | |
2433 | unlock_user(p, addr, l); | |
2434 | } else { | |
2435 | if (!(flags & PAGE_READ)) | |
2436 | return -1; | |
2437 | /* XXX: this code should not depend on lock_user */ | |
2438 | if (!(p = lock_user(VERIFY_READ, addr, l, 1))) | |
2439 | return -1; | |
2440 | memcpy(buf, p, l); | |
2441 | unlock_user(p, addr, 0); | |
2442 | } | |
2443 | len -= l; | |
2444 | buf += l; | |
2445 | addr += l; | |
2446 | } | |
2447 | return 0; | |
2448 | } | |
2449 | ||
2450 | #else | |
2451 | ||
2452 | static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr, | |
2453 | hwaddr length) | |
2454 | { | |
2455 | uint8_t dirty_log_mask = memory_region_get_dirty_log_mask(mr); | |
2456 | addr += memory_region_get_ram_addr(mr); | |
2457 | ||
2458 | /* No early return if dirty_log_mask is or becomes 0, because | |
2459 | * cpu_physical_memory_set_dirty_range will still call | |
2460 | * xen_modified_memory. | |
2461 | */ | |
2462 | if (dirty_log_mask) { | |
2463 | dirty_log_mask = | |
2464 | cpu_physical_memory_range_includes_clean(addr, length, dirty_log_mask); | |
2465 | } | |
2466 | if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) { | |
2467 | tb_invalidate_phys_range(addr, addr + length); | |
2468 | dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE); | |
2469 | } | |
2470 | cpu_physical_memory_set_dirty_range(addr, length, dirty_log_mask); | |
2471 | } | |
2472 | ||
2473 | static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr) | |
2474 | { | |
2475 | unsigned access_size_max = mr->ops->valid.max_access_size; | |
2476 | ||
2477 | /* Regions are assumed to support 1-4 byte accesses unless | |
2478 | otherwise specified. */ | |
2479 | if (access_size_max == 0) { | |
2480 | access_size_max = 4; | |
2481 | } | |
2482 | ||
2483 | /* Bound the maximum access by the alignment of the address. */ | |
2484 | if (!mr->ops->impl.unaligned) { | |
2485 | unsigned align_size_max = addr & -addr; | |
2486 | if (align_size_max != 0 && align_size_max < access_size_max) { | |
2487 | access_size_max = align_size_max; | |
2488 | } | |
2489 | } | |
2490 | ||
2491 | /* Don't attempt accesses larger than the maximum. */ | |
2492 | if (l > access_size_max) { | |
2493 | l = access_size_max; | |
2494 | } | |
2495 | l = pow2floor(l); | |
2496 | ||
2497 | return l; | |
2498 | } | |
2499 | ||
2500 | static bool prepare_mmio_access(MemoryRegion *mr) | |
2501 | { | |
2502 | bool unlocked = !qemu_mutex_iothread_locked(); | |
2503 | bool release_lock = false; | |
2504 | ||
2505 | if (unlocked && mr->global_locking) { | |
2506 | qemu_mutex_lock_iothread(); | |
2507 | unlocked = false; | |
2508 | release_lock = true; | |
2509 | } | |
2510 | if (mr->flush_coalesced_mmio) { | |
2511 | if (unlocked) { | |
2512 | qemu_mutex_lock_iothread(); | |
2513 | } | |
2514 | qemu_flush_coalesced_mmio_buffer(); | |
2515 | if (unlocked) { | |
2516 | qemu_mutex_unlock_iothread(); | |
2517 | } | |
2518 | } | |
2519 | ||
2520 | return release_lock; | |
2521 | } | |
2522 | ||
2523 | /* Called within RCU critical section. */ | |
2524 | static MemTxResult address_space_write_continue(AddressSpace *as, hwaddr addr, | |
2525 | MemTxAttrs attrs, | |
2526 | const uint8_t *buf, | |
2527 | int len, hwaddr addr1, | |
2528 | hwaddr l, MemoryRegion *mr) | |
2529 | { | |
2530 | uint8_t *ptr; | |
2531 | uint64_t val; | |
2532 | MemTxResult result = MEMTX_OK; | |
2533 | bool release_lock = false; | |
2534 | ||
2535 | for (;;) { | |
2536 | if (!memory_access_is_direct(mr, true)) { | |
2537 | release_lock |= prepare_mmio_access(mr); | |
2538 | l = memory_access_size(mr, l, addr1); | |
2539 | /* XXX: could force current_cpu to NULL to avoid | |
2540 | potential bugs */ | |
2541 | switch (l) { | |
2542 | case 8: | |
2543 | /* 64 bit write access */ | |
2544 | val = ldq_p(buf); | |
2545 | result |= memory_region_dispatch_write(mr, addr1, val, 8, | |
2546 | attrs); | |
2547 | break; | |
2548 | case 4: | |
2549 | /* 32 bit write access */ | |
2550 | val = ldl_p(buf); | |
2551 | result |= memory_region_dispatch_write(mr, addr1, val, 4, | |
2552 | attrs); | |
2553 | break; | |
2554 | case 2: | |
2555 | /* 16 bit write access */ | |
2556 | val = lduw_p(buf); | |
2557 | result |= memory_region_dispatch_write(mr, addr1, val, 2, | |
2558 | attrs); | |
2559 | break; | |
2560 | case 1: | |
2561 | /* 8 bit write access */ | |
2562 | val = ldub_p(buf); | |
2563 | result |= memory_region_dispatch_write(mr, addr1, val, 1, | |
2564 | attrs); | |
2565 | break; | |
2566 | default: | |
2567 | abort(); | |
2568 | } | |
2569 | } else { | |
2570 | /* RAM case */ | |
2571 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
2572 | memcpy(ptr, buf, l); | |
2573 | invalidate_and_set_dirty(mr, addr1, l); | |
2574 | } | |
2575 | ||
2576 | if (release_lock) { | |
2577 | qemu_mutex_unlock_iothread(); | |
2578 | release_lock = false; | |
2579 | } | |
2580 | ||
2581 | len -= l; | |
2582 | buf += l; | |
2583 | addr += l; | |
2584 | ||
2585 | if (!len) { | |
2586 | break; | |
2587 | } | |
2588 | ||
2589 | l = len; | |
2590 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
2591 | } | |
2592 | ||
2593 | return result; | |
2594 | } | |
2595 | ||
2596 | MemTxResult address_space_write(AddressSpace *as, hwaddr addr, MemTxAttrs attrs, | |
2597 | const uint8_t *buf, int len) | |
2598 | { | |
2599 | hwaddr l; | |
2600 | hwaddr addr1; | |
2601 | MemoryRegion *mr; | |
2602 | MemTxResult result = MEMTX_OK; | |
2603 | ||
2604 | if (len > 0) { | |
2605 | rcu_read_lock(); | |
2606 | l = len; | |
2607 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
2608 | result = address_space_write_continue(as, addr, attrs, buf, len, | |
2609 | addr1, l, mr); | |
2610 | rcu_read_unlock(); | |
2611 | } | |
2612 | ||
2613 | return result; | |
2614 | } | |
2615 | ||
2616 | /* Called within RCU critical section. */ | |
2617 | MemTxResult address_space_read_continue(AddressSpace *as, hwaddr addr, | |
2618 | MemTxAttrs attrs, uint8_t *buf, | |
2619 | int len, hwaddr addr1, hwaddr l, | |
2620 | MemoryRegion *mr) | |
2621 | { | |
2622 | uint8_t *ptr; | |
2623 | uint64_t val; | |
2624 | MemTxResult result = MEMTX_OK; | |
2625 | bool release_lock = false; | |
2626 | ||
2627 | for (;;) { | |
2628 | if (!memory_access_is_direct(mr, false)) { | |
2629 | /* I/O case */ | |
2630 | release_lock |= prepare_mmio_access(mr); | |
2631 | l = memory_access_size(mr, l, addr1); | |
2632 | switch (l) { | |
2633 | case 8: | |
2634 | /* 64 bit read access */ | |
2635 | result |= memory_region_dispatch_read(mr, addr1, &val, 8, | |
2636 | attrs); | |
2637 | stq_p(buf, val); | |
2638 | break; | |
2639 | case 4: | |
2640 | /* 32 bit read access */ | |
2641 | result |= memory_region_dispatch_read(mr, addr1, &val, 4, | |
2642 | attrs); | |
2643 | stl_p(buf, val); | |
2644 | break; | |
2645 | case 2: | |
2646 | /* 16 bit read access */ | |
2647 | result |= memory_region_dispatch_read(mr, addr1, &val, 2, | |
2648 | attrs); | |
2649 | stw_p(buf, val); | |
2650 | break; | |
2651 | case 1: | |
2652 | /* 8 bit read access */ | |
2653 | result |= memory_region_dispatch_read(mr, addr1, &val, 1, | |
2654 | attrs); | |
2655 | stb_p(buf, val); | |
2656 | break; | |
2657 | default: | |
2658 | abort(); | |
2659 | } | |
2660 | } else { | |
2661 | /* RAM case */ | |
2662 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
2663 | memcpy(buf, ptr, l); | |
2664 | } | |
2665 | ||
2666 | if (release_lock) { | |
2667 | qemu_mutex_unlock_iothread(); | |
2668 | release_lock = false; | |
2669 | } | |
2670 | ||
2671 | len -= l; | |
2672 | buf += l; | |
2673 | addr += l; | |
2674 | ||
2675 | if (!len) { | |
2676 | break; | |
2677 | } | |
2678 | ||
2679 | l = len; | |
2680 | mr = address_space_translate(as, addr, &addr1, &l, false); | |
2681 | } | |
2682 | ||
2683 | return result; | |
2684 | } | |
2685 | ||
2686 | MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr, | |
2687 | MemTxAttrs attrs, uint8_t *buf, int len) | |
2688 | { | |
2689 | hwaddr l; | |
2690 | hwaddr addr1; | |
2691 | MemoryRegion *mr; | |
2692 | MemTxResult result = MEMTX_OK; | |
2693 | ||
2694 | if (len > 0) { | |
2695 | rcu_read_lock(); | |
2696 | l = len; | |
2697 | mr = address_space_translate(as, addr, &addr1, &l, false); | |
2698 | result = address_space_read_continue(as, addr, attrs, buf, len, | |
2699 | addr1, l, mr); | |
2700 | rcu_read_unlock(); | |
2701 | } | |
2702 | ||
2703 | return result; | |
2704 | } | |
2705 | ||
2706 | MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, MemTxAttrs attrs, | |
2707 | uint8_t *buf, int len, bool is_write) | |
2708 | { | |
2709 | if (is_write) { | |
2710 | return address_space_write(as, addr, attrs, (uint8_t *)buf, len); | |
2711 | } else { | |
2712 | return address_space_read(as, addr, attrs, (uint8_t *)buf, len); | |
2713 | } | |
2714 | } | |
2715 | ||
2716 | void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf, | |
2717 | int len, int is_write) | |
2718 | { | |
2719 | address_space_rw(&address_space_memory, addr, MEMTXATTRS_UNSPECIFIED, | |
2720 | buf, len, is_write); | |
2721 | } | |
2722 | ||
2723 | enum write_rom_type { | |
2724 | WRITE_DATA, | |
2725 | FLUSH_CACHE, | |
2726 | }; | |
2727 | ||
2728 | static inline void cpu_physical_memory_write_rom_internal(AddressSpace *as, | |
2729 | hwaddr addr, const uint8_t *buf, int len, enum write_rom_type type) | |
2730 | { | |
2731 | hwaddr l; | |
2732 | uint8_t *ptr; | |
2733 | hwaddr addr1; | |
2734 | MemoryRegion *mr; | |
2735 | ||
2736 | rcu_read_lock(); | |
2737 | while (len > 0) { | |
2738 | l = len; | |
2739 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
2740 | ||
2741 | if (!(memory_region_is_ram(mr) || | |
2742 | memory_region_is_romd(mr))) { | |
2743 | l = memory_access_size(mr, l, addr1); | |
2744 | } else { | |
2745 | /* ROM/RAM case */ | |
2746 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
2747 | switch (type) { | |
2748 | case WRITE_DATA: | |
2749 | memcpy(ptr, buf, l); | |
2750 | invalidate_and_set_dirty(mr, addr1, l); | |
2751 | break; | |
2752 | case FLUSH_CACHE: | |
2753 | flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l); | |
2754 | break; | |
2755 | } | |
2756 | } | |
2757 | len -= l; | |
2758 | buf += l; | |
2759 | addr += l; | |
2760 | } | |
2761 | rcu_read_unlock(); | |
2762 | } | |
2763 | ||
2764 | /* used for ROM loading : can write in RAM and ROM */ | |
2765 | void cpu_physical_memory_write_rom(AddressSpace *as, hwaddr addr, | |
2766 | const uint8_t *buf, int len) | |
2767 | { | |
2768 | cpu_physical_memory_write_rom_internal(as, addr, buf, len, WRITE_DATA); | |
2769 | } | |
2770 | ||
2771 | void cpu_flush_icache_range(hwaddr start, int len) | |
2772 | { | |
2773 | /* | |
2774 | * This function should do the same thing as an icache flush that was | |
2775 | * triggered from within the guest. For TCG we are always cache coherent, | |
2776 | * so there is no need to flush anything. For KVM / Xen we need to flush | |
2777 | * the host's instruction cache at least. | |
2778 | */ | |
2779 | if (tcg_enabled()) { | |
2780 | return; | |
2781 | } | |
2782 | ||
2783 | cpu_physical_memory_write_rom_internal(&address_space_memory, | |
2784 | start, NULL, len, FLUSH_CACHE); | |
2785 | } | |
2786 | ||
2787 | typedef struct { | |
2788 | MemoryRegion *mr; | |
2789 | void *buffer; | |
2790 | hwaddr addr; | |
2791 | hwaddr len; | |
2792 | bool in_use; | |
2793 | } BounceBuffer; | |
2794 | ||
2795 | static BounceBuffer bounce; | |
2796 | ||
2797 | typedef struct MapClient { | |
2798 | QEMUBH *bh; | |
2799 | QLIST_ENTRY(MapClient) link; | |
2800 | } MapClient; | |
2801 | ||
2802 | QemuMutex map_client_list_lock; | |
2803 | static QLIST_HEAD(map_client_list, MapClient) map_client_list | |
2804 | = QLIST_HEAD_INITIALIZER(map_client_list); | |
2805 | ||
2806 | static void cpu_unregister_map_client_do(MapClient *client) | |
2807 | { | |
2808 | QLIST_REMOVE(client, link); | |
2809 | g_free(client); | |
2810 | } | |
2811 | ||
2812 | static void cpu_notify_map_clients_locked(void) | |
2813 | { | |
2814 | MapClient *client; | |
2815 | ||
2816 | while (!QLIST_EMPTY(&map_client_list)) { | |
2817 | client = QLIST_FIRST(&map_client_list); | |
2818 | qemu_bh_schedule(client->bh); | |
2819 | cpu_unregister_map_client_do(client); | |
2820 | } | |
2821 | } | |
2822 | ||
2823 | void cpu_register_map_client(QEMUBH *bh) | |
2824 | { | |
2825 | MapClient *client = g_malloc(sizeof(*client)); | |
2826 | ||
2827 | qemu_mutex_lock(&map_client_list_lock); | |
2828 | client->bh = bh; | |
2829 | QLIST_INSERT_HEAD(&map_client_list, client, link); | |
2830 | if (!atomic_read(&bounce.in_use)) { | |
2831 | cpu_notify_map_clients_locked(); | |
2832 | } | |
2833 | qemu_mutex_unlock(&map_client_list_lock); | |
2834 | } | |
2835 | ||
2836 | void cpu_exec_init_all(void) | |
2837 | { | |
2838 | qemu_mutex_init(&ram_list.mutex); | |
2839 | io_mem_init(); | |
2840 | memory_map_init(); | |
2841 | qemu_mutex_init(&map_client_list_lock); | |
2842 | } | |
2843 | ||
2844 | void cpu_unregister_map_client(QEMUBH *bh) | |
2845 | { | |
2846 | MapClient *client; | |
2847 | ||
2848 | qemu_mutex_lock(&map_client_list_lock); | |
2849 | QLIST_FOREACH(client, &map_client_list, link) { | |
2850 | if (client->bh == bh) { | |
2851 | cpu_unregister_map_client_do(client); | |
2852 | break; | |
2853 | } | |
2854 | } | |
2855 | qemu_mutex_unlock(&map_client_list_lock); | |
2856 | } | |
2857 | ||
2858 | static void cpu_notify_map_clients(void) | |
2859 | { | |
2860 | qemu_mutex_lock(&map_client_list_lock); | |
2861 | cpu_notify_map_clients_locked(); | |
2862 | qemu_mutex_unlock(&map_client_list_lock); | |
2863 | } | |
2864 | ||
2865 | bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write) | |
2866 | { | |
2867 | MemoryRegion *mr; | |
2868 | hwaddr l, xlat; | |
2869 | ||
2870 | rcu_read_lock(); | |
2871 | while (len > 0) { | |
2872 | l = len; | |
2873 | mr = address_space_translate(as, addr, &xlat, &l, is_write); | |
2874 | if (!memory_access_is_direct(mr, is_write)) { | |
2875 | l = memory_access_size(mr, l, addr); | |
2876 | if (!memory_region_access_valid(mr, xlat, l, is_write)) { | |
2877 | return false; | |
2878 | } | |
2879 | } | |
2880 | ||
2881 | len -= l; | |
2882 | addr += l; | |
2883 | } | |
2884 | rcu_read_unlock(); | |
2885 | return true; | |
2886 | } | |
2887 | ||
2888 | /* Map a physical memory region into a host virtual address. | |
2889 | * May map a subset of the requested range, given by and returned in *plen. | |
2890 | * May return NULL if resources needed to perform the mapping are exhausted. | |
2891 | * Use only for reads OR writes - not for read-modify-write operations. | |
2892 | * Use cpu_register_map_client() to know when retrying the map operation is | |
2893 | * likely to succeed. | |
2894 | */ | |
2895 | void *address_space_map(AddressSpace *as, | |
2896 | hwaddr addr, | |
2897 | hwaddr *plen, | |
2898 | bool is_write) | |
2899 | { | |
2900 | hwaddr len = *plen; | |
2901 | hwaddr done = 0; | |
2902 | hwaddr l, xlat, base; | |
2903 | MemoryRegion *mr, *this_mr; | |
2904 | void *ptr; | |
2905 | ||
2906 | if (len == 0) { | |
2907 | return NULL; | |
2908 | } | |
2909 | ||
2910 | l = len; | |
2911 | rcu_read_lock(); | |
2912 | mr = address_space_translate(as, addr, &xlat, &l, is_write); | |
2913 | ||
2914 | if (!memory_access_is_direct(mr, is_write)) { | |
2915 | if (atomic_xchg(&bounce.in_use, true)) { | |
2916 | rcu_read_unlock(); | |
2917 | return NULL; | |
2918 | } | |
2919 | /* Avoid unbounded allocations */ | |
2920 | l = MIN(l, TARGET_PAGE_SIZE); | |
2921 | bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l); | |
2922 | bounce.addr = addr; | |
2923 | bounce.len = l; | |
2924 | ||
2925 | memory_region_ref(mr); | |
2926 | bounce.mr = mr; | |
2927 | if (!is_write) { | |
2928 | address_space_read(as, addr, MEMTXATTRS_UNSPECIFIED, | |
2929 | bounce.buffer, l); | |
2930 | } | |
2931 | ||
2932 | rcu_read_unlock(); | |
2933 | *plen = l; | |
2934 | return bounce.buffer; | |
2935 | } | |
2936 | ||
2937 | base = xlat; | |
2938 | ||
2939 | for (;;) { | |
2940 | len -= l; | |
2941 | addr += l; | |
2942 | done += l; | |
2943 | if (len == 0) { | |
2944 | break; | |
2945 | } | |
2946 | ||
2947 | l = len; | |
2948 | this_mr = address_space_translate(as, addr, &xlat, &l, is_write); | |
2949 | if (this_mr != mr || xlat != base + done) { | |
2950 | break; | |
2951 | } | |
2952 | } | |
2953 | ||
2954 | memory_region_ref(mr); | |
2955 | *plen = done; | |
2956 | ptr = qemu_ram_ptr_length(mr->ram_block, base, plen); | |
2957 | rcu_read_unlock(); | |
2958 | ||
2959 | return ptr; | |
2960 | } | |
2961 | ||
2962 | /* Unmaps a memory region previously mapped by address_space_map(). | |
2963 | * Will also mark the memory as dirty if is_write == 1. access_len gives | |
2964 | * the amount of memory that was actually read or written by the caller. | |
2965 | */ | |
2966 | void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len, | |
2967 | int is_write, hwaddr access_len) | |
2968 | { | |
2969 | if (buffer != bounce.buffer) { | |
2970 | MemoryRegion *mr; | |
2971 | ram_addr_t addr1; | |
2972 | ||
2973 | mr = memory_region_from_host(buffer, &addr1); | |
2974 | assert(mr != NULL); | |
2975 | if (is_write) { | |
2976 | invalidate_and_set_dirty(mr, addr1, access_len); | |
2977 | } | |
2978 | if (xen_enabled()) { | |
2979 | xen_invalidate_map_cache_entry(buffer); | |
2980 | } | |
2981 | memory_region_unref(mr); | |
2982 | return; | |
2983 | } | |
2984 | if (is_write) { | |
2985 | address_space_write(as, bounce.addr, MEMTXATTRS_UNSPECIFIED, | |
2986 | bounce.buffer, access_len); | |
2987 | } | |
2988 | qemu_vfree(bounce.buffer); | |
2989 | bounce.buffer = NULL; | |
2990 | memory_region_unref(bounce.mr); | |
2991 | atomic_mb_set(&bounce.in_use, false); | |
2992 | cpu_notify_map_clients(); | |
2993 | } | |
2994 | ||
2995 | void *cpu_physical_memory_map(hwaddr addr, | |
2996 | hwaddr *plen, | |
2997 | int is_write) | |
2998 | { | |
2999 | return address_space_map(&address_space_memory, addr, plen, is_write); | |
3000 | } | |
3001 | ||
3002 | void cpu_physical_memory_unmap(void *buffer, hwaddr len, | |
3003 | int is_write, hwaddr access_len) | |
3004 | { | |
3005 | return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len); | |
3006 | } | |
3007 | ||
3008 | /* warning: addr must be aligned */ | |
3009 | static inline uint32_t address_space_ldl_internal(AddressSpace *as, hwaddr addr, | |
3010 | MemTxAttrs attrs, | |
3011 | MemTxResult *result, | |
3012 | enum device_endian endian) | |
3013 | { | |
3014 | uint8_t *ptr; | |
3015 | uint64_t val; | |
3016 | MemoryRegion *mr; | |
3017 | hwaddr l = 4; | |
3018 | hwaddr addr1; | |
3019 | MemTxResult r; | |
3020 | bool release_lock = false; | |
3021 | ||
3022 | rcu_read_lock(); | |
3023 | mr = address_space_translate(as, addr, &addr1, &l, false); | |
3024 | if (l < 4 || !memory_access_is_direct(mr, false)) { | |
3025 | release_lock |= prepare_mmio_access(mr); | |
3026 | ||
3027 | /* I/O case */ | |
3028 | r = memory_region_dispatch_read(mr, addr1, &val, 4, attrs); | |
3029 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3030 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3031 | val = bswap32(val); | |
3032 | } | |
3033 | #else | |
3034 | if (endian == DEVICE_BIG_ENDIAN) { | |
3035 | val = bswap32(val); | |
3036 | } | |
3037 | #endif | |
3038 | } else { | |
3039 | /* RAM case */ | |
3040 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
3041 | switch (endian) { | |
3042 | case DEVICE_LITTLE_ENDIAN: | |
3043 | val = ldl_le_p(ptr); | |
3044 | break; | |
3045 | case DEVICE_BIG_ENDIAN: | |
3046 | val = ldl_be_p(ptr); | |
3047 | break; | |
3048 | default: | |
3049 | val = ldl_p(ptr); | |
3050 | break; | |
3051 | } | |
3052 | r = MEMTX_OK; | |
3053 | } | |
3054 | if (result) { | |
3055 | *result = r; | |
3056 | } | |
3057 | if (release_lock) { | |
3058 | qemu_mutex_unlock_iothread(); | |
3059 | } | |
3060 | rcu_read_unlock(); | |
3061 | return val; | |
3062 | } | |
3063 | ||
3064 | uint32_t address_space_ldl(AddressSpace *as, hwaddr addr, | |
3065 | MemTxAttrs attrs, MemTxResult *result) | |
3066 | { | |
3067 | return address_space_ldl_internal(as, addr, attrs, result, | |
3068 | DEVICE_NATIVE_ENDIAN); | |
3069 | } | |
3070 | ||
3071 | uint32_t address_space_ldl_le(AddressSpace *as, hwaddr addr, | |
3072 | MemTxAttrs attrs, MemTxResult *result) | |
3073 | { | |
3074 | return address_space_ldl_internal(as, addr, attrs, result, | |
3075 | DEVICE_LITTLE_ENDIAN); | |
3076 | } | |
3077 | ||
3078 | uint32_t address_space_ldl_be(AddressSpace *as, hwaddr addr, | |
3079 | MemTxAttrs attrs, MemTxResult *result) | |
3080 | { | |
3081 | return address_space_ldl_internal(as, addr, attrs, result, | |
3082 | DEVICE_BIG_ENDIAN); | |
3083 | } | |
3084 | ||
3085 | uint32_t ldl_phys(AddressSpace *as, hwaddr addr) | |
3086 | { | |
3087 | return address_space_ldl(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3088 | } | |
3089 | ||
3090 | uint32_t ldl_le_phys(AddressSpace *as, hwaddr addr) | |
3091 | { | |
3092 | return address_space_ldl_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3093 | } | |
3094 | ||
3095 | uint32_t ldl_be_phys(AddressSpace *as, hwaddr addr) | |
3096 | { | |
3097 | return address_space_ldl_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3098 | } | |
3099 | ||
3100 | /* warning: addr must be aligned */ | |
3101 | static inline uint64_t address_space_ldq_internal(AddressSpace *as, hwaddr addr, | |
3102 | MemTxAttrs attrs, | |
3103 | MemTxResult *result, | |
3104 | enum device_endian endian) | |
3105 | { | |
3106 | uint8_t *ptr; | |
3107 | uint64_t val; | |
3108 | MemoryRegion *mr; | |
3109 | hwaddr l = 8; | |
3110 | hwaddr addr1; | |
3111 | MemTxResult r; | |
3112 | bool release_lock = false; | |
3113 | ||
3114 | rcu_read_lock(); | |
3115 | mr = address_space_translate(as, addr, &addr1, &l, | |
3116 | false); | |
3117 | if (l < 8 || !memory_access_is_direct(mr, false)) { | |
3118 | release_lock |= prepare_mmio_access(mr); | |
3119 | ||
3120 | /* I/O case */ | |
3121 | r = memory_region_dispatch_read(mr, addr1, &val, 8, attrs); | |
3122 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3123 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3124 | val = bswap64(val); | |
3125 | } | |
3126 | #else | |
3127 | if (endian == DEVICE_BIG_ENDIAN) { | |
3128 | val = bswap64(val); | |
3129 | } | |
3130 | #endif | |
3131 | } else { | |
3132 | /* RAM case */ | |
3133 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
3134 | switch (endian) { | |
3135 | case DEVICE_LITTLE_ENDIAN: | |
3136 | val = ldq_le_p(ptr); | |
3137 | break; | |
3138 | case DEVICE_BIG_ENDIAN: | |
3139 | val = ldq_be_p(ptr); | |
3140 | break; | |
3141 | default: | |
3142 | val = ldq_p(ptr); | |
3143 | break; | |
3144 | } | |
3145 | r = MEMTX_OK; | |
3146 | } | |
3147 | if (result) { | |
3148 | *result = r; | |
3149 | } | |
3150 | if (release_lock) { | |
3151 | qemu_mutex_unlock_iothread(); | |
3152 | } | |
3153 | rcu_read_unlock(); | |
3154 | return val; | |
3155 | } | |
3156 | ||
3157 | uint64_t address_space_ldq(AddressSpace *as, hwaddr addr, | |
3158 | MemTxAttrs attrs, MemTxResult *result) | |
3159 | { | |
3160 | return address_space_ldq_internal(as, addr, attrs, result, | |
3161 | DEVICE_NATIVE_ENDIAN); | |
3162 | } | |
3163 | ||
3164 | uint64_t address_space_ldq_le(AddressSpace *as, hwaddr addr, | |
3165 | MemTxAttrs attrs, MemTxResult *result) | |
3166 | { | |
3167 | return address_space_ldq_internal(as, addr, attrs, result, | |
3168 | DEVICE_LITTLE_ENDIAN); | |
3169 | } | |
3170 | ||
3171 | uint64_t address_space_ldq_be(AddressSpace *as, hwaddr addr, | |
3172 | MemTxAttrs attrs, MemTxResult *result) | |
3173 | { | |
3174 | return address_space_ldq_internal(as, addr, attrs, result, | |
3175 | DEVICE_BIG_ENDIAN); | |
3176 | } | |
3177 | ||
3178 | uint64_t ldq_phys(AddressSpace *as, hwaddr addr) | |
3179 | { | |
3180 | return address_space_ldq(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3181 | } | |
3182 | ||
3183 | uint64_t ldq_le_phys(AddressSpace *as, hwaddr addr) | |
3184 | { | |
3185 | return address_space_ldq_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3186 | } | |
3187 | ||
3188 | uint64_t ldq_be_phys(AddressSpace *as, hwaddr addr) | |
3189 | { | |
3190 | return address_space_ldq_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3191 | } | |
3192 | ||
3193 | /* XXX: optimize */ | |
3194 | uint32_t address_space_ldub(AddressSpace *as, hwaddr addr, | |
3195 | MemTxAttrs attrs, MemTxResult *result) | |
3196 | { | |
3197 | uint8_t val; | |
3198 | MemTxResult r; | |
3199 | ||
3200 | r = address_space_rw(as, addr, attrs, &val, 1, 0); | |
3201 | if (result) { | |
3202 | *result = r; | |
3203 | } | |
3204 | return val; | |
3205 | } | |
3206 | ||
3207 | uint32_t ldub_phys(AddressSpace *as, hwaddr addr) | |
3208 | { | |
3209 | return address_space_ldub(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3210 | } | |
3211 | ||
3212 | /* warning: addr must be aligned */ | |
3213 | static inline uint32_t address_space_lduw_internal(AddressSpace *as, | |
3214 | hwaddr addr, | |
3215 | MemTxAttrs attrs, | |
3216 | MemTxResult *result, | |
3217 | enum device_endian endian) | |
3218 | { | |
3219 | uint8_t *ptr; | |
3220 | uint64_t val; | |
3221 | MemoryRegion *mr; | |
3222 | hwaddr l = 2; | |
3223 | hwaddr addr1; | |
3224 | MemTxResult r; | |
3225 | bool release_lock = false; | |
3226 | ||
3227 | rcu_read_lock(); | |
3228 | mr = address_space_translate(as, addr, &addr1, &l, | |
3229 | false); | |
3230 | if (l < 2 || !memory_access_is_direct(mr, false)) { | |
3231 | release_lock |= prepare_mmio_access(mr); | |
3232 | ||
3233 | /* I/O case */ | |
3234 | r = memory_region_dispatch_read(mr, addr1, &val, 2, attrs); | |
3235 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3236 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3237 | val = bswap16(val); | |
3238 | } | |
3239 | #else | |
3240 | if (endian == DEVICE_BIG_ENDIAN) { | |
3241 | val = bswap16(val); | |
3242 | } | |
3243 | #endif | |
3244 | } else { | |
3245 | /* RAM case */ | |
3246 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
3247 | switch (endian) { | |
3248 | case DEVICE_LITTLE_ENDIAN: | |
3249 | val = lduw_le_p(ptr); | |
3250 | break; | |
3251 | case DEVICE_BIG_ENDIAN: | |
3252 | val = lduw_be_p(ptr); | |
3253 | break; | |
3254 | default: | |
3255 | val = lduw_p(ptr); | |
3256 | break; | |
3257 | } | |
3258 | r = MEMTX_OK; | |
3259 | } | |
3260 | if (result) { | |
3261 | *result = r; | |
3262 | } | |
3263 | if (release_lock) { | |
3264 | qemu_mutex_unlock_iothread(); | |
3265 | } | |
3266 | rcu_read_unlock(); | |
3267 | return val; | |
3268 | } | |
3269 | ||
3270 | uint32_t address_space_lduw(AddressSpace *as, hwaddr addr, | |
3271 | MemTxAttrs attrs, MemTxResult *result) | |
3272 | { | |
3273 | return address_space_lduw_internal(as, addr, attrs, result, | |
3274 | DEVICE_NATIVE_ENDIAN); | |
3275 | } | |
3276 | ||
3277 | uint32_t address_space_lduw_le(AddressSpace *as, hwaddr addr, | |
3278 | MemTxAttrs attrs, MemTxResult *result) | |
3279 | { | |
3280 | return address_space_lduw_internal(as, addr, attrs, result, | |
3281 | DEVICE_LITTLE_ENDIAN); | |
3282 | } | |
3283 | ||
3284 | uint32_t address_space_lduw_be(AddressSpace *as, hwaddr addr, | |
3285 | MemTxAttrs attrs, MemTxResult *result) | |
3286 | { | |
3287 | return address_space_lduw_internal(as, addr, attrs, result, | |
3288 | DEVICE_BIG_ENDIAN); | |
3289 | } | |
3290 | ||
3291 | uint32_t lduw_phys(AddressSpace *as, hwaddr addr) | |
3292 | { | |
3293 | return address_space_lduw(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3294 | } | |
3295 | ||
3296 | uint32_t lduw_le_phys(AddressSpace *as, hwaddr addr) | |
3297 | { | |
3298 | return address_space_lduw_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3299 | } | |
3300 | ||
3301 | uint32_t lduw_be_phys(AddressSpace *as, hwaddr addr) | |
3302 | { | |
3303 | return address_space_lduw_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3304 | } | |
3305 | ||
3306 | /* warning: addr must be aligned. The ram page is not masked as dirty | |
3307 | and the code inside is not invalidated. It is useful if the dirty | |
3308 | bits are used to track modified PTEs */ | |
3309 | void address_space_stl_notdirty(AddressSpace *as, hwaddr addr, uint32_t val, | |
3310 | MemTxAttrs attrs, MemTxResult *result) | |
3311 | { | |
3312 | uint8_t *ptr; | |
3313 | MemoryRegion *mr; | |
3314 | hwaddr l = 4; | |
3315 | hwaddr addr1; | |
3316 | MemTxResult r; | |
3317 | uint8_t dirty_log_mask; | |
3318 | bool release_lock = false; | |
3319 | ||
3320 | rcu_read_lock(); | |
3321 | mr = address_space_translate(as, addr, &addr1, &l, | |
3322 | true); | |
3323 | if (l < 4 || !memory_access_is_direct(mr, true)) { | |
3324 | release_lock |= prepare_mmio_access(mr); | |
3325 | ||
3326 | r = memory_region_dispatch_write(mr, addr1, val, 4, attrs); | |
3327 | } else { | |
3328 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
3329 | stl_p(ptr, val); | |
3330 | ||
3331 | dirty_log_mask = memory_region_get_dirty_log_mask(mr); | |
3332 | dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE); | |
3333 | cpu_physical_memory_set_dirty_range(memory_region_get_ram_addr(mr) + addr, | |
3334 | 4, dirty_log_mask); | |
3335 | r = MEMTX_OK; | |
3336 | } | |
3337 | if (result) { | |
3338 | *result = r; | |
3339 | } | |
3340 | if (release_lock) { | |
3341 | qemu_mutex_unlock_iothread(); | |
3342 | } | |
3343 | rcu_read_unlock(); | |
3344 | } | |
3345 | ||
3346 | void stl_phys_notdirty(AddressSpace *as, hwaddr addr, uint32_t val) | |
3347 | { | |
3348 | address_space_stl_notdirty(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3349 | } | |
3350 | ||
3351 | /* warning: addr must be aligned */ | |
3352 | static inline void address_space_stl_internal(AddressSpace *as, | |
3353 | hwaddr addr, uint32_t val, | |
3354 | MemTxAttrs attrs, | |
3355 | MemTxResult *result, | |
3356 | enum device_endian endian) | |
3357 | { | |
3358 | uint8_t *ptr; | |
3359 | MemoryRegion *mr; | |
3360 | hwaddr l = 4; | |
3361 | hwaddr addr1; | |
3362 | MemTxResult r; | |
3363 | bool release_lock = false; | |
3364 | ||
3365 | rcu_read_lock(); | |
3366 | mr = address_space_translate(as, addr, &addr1, &l, | |
3367 | true); | |
3368 | if (l < 4 || !memory_access_is_direct(mr, true)) { | |
3369 | release_lock |= prepare_mmio_access(mr); | |
3370 | ||
3371 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3372 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3373 | val = bswap32(val); | |
3374 | } | |
3375 | #else | |
3376 | if (endian == DEVICE_BIG_ENDIAN) { | |
3377 | val = bswap32(val); | |
3378 | } | |
3379 | #endif | |
3380 | r = memory_region_dispatch_write(mr, addr1, val, 4, attrs); | |
3381 | } else { | |
3382 | /* RAM case */ | |
3383 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
3384 | switch (endian) { | |
3385 | case DEVICE_LITTLE_ENDIAN: | |
3386 | stl_le_p(ptr, val); | |
3387 | break; | |
3388 | case DEVICE_BIG_ENDIAN: | |
3389 | stl_be_p(ptr, val); | |
3390 | break; | |
3391 | default: | |
3392 | stl_p(ptr, val); | |
3393 | break; | |
3394 | } | |
3395 | invalidate_and_set_dirty(mr, addr1, 4); | |
3396 | r = MEMTX_OK; | |
3397 | } | |
3398 | if (result) { | |
3399 | *result = r; | |
3400 | } | |
3401 | if (release_lock) { | |
3402 | qemu_mutex_unlock_iothread(); | |
3403 | } | |
3404 | rcu_read_unlock(); | |
3405 | } | |
3406 | ||
3407 | void address_space_stl(AddressSpace *as, hwaddr addr, uint32_t val, | |
3408 | MemTxAttrs attrs, MemTxResult *result) | |
3409 | { | |
3410 | address_space_stl_internal(as, addr, val, attrs, result, | |
3411 | DEVICE_NATIVE_ENDIAN); | |
3412 | } | |
3413 | ||
3414 | void address_space_stl_le(AddressSpace *as, hwaddr addr, uint32_t val, | |
3415 | MemTxAttrs attrs, MemTxResult *result) | |
3416 | { | |
3417 | address_space_stl_internal(as, addr, val, attrs, result, | |
3418 | DEVICE_LITTLE_ENDIAN); | |
3419 | } | |
3420 | ||
3421 | void address_space_stl_be(AddressSpace *as, hwaddr addr, uint32_t val, | |
3422 | MemTxAttrs attrs, MemTxResult *result) | |
3423 | { | |
3424 | address_space_stl_internal(as, addr, val, attrs, result, | |
3425 | DEVICE_BIG_ENDIAN); | |
3426 | } | |
3427 | ||
3428 | void stl_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3429 | { | |
3430 | address_space_stl(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3431 | } | |
3432 | ||
3433 | void stl_le_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3434 | { | |
3435 | address_space_stl_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3436 | } | |
3437 | ||
3438 | void stl_be_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3439 | { | |
3440 | address_space_stl_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3441 | } | |
3442 | ||
3443 | /* XXX: optimize */ | |
3444 | void address_space_stb(AddressSpace *as, hwaddr addr, uint32_t val, | |
3445 | MemTxAttrs attrs, MemTxResult *result) | |
3446 | { | |
3447 | uint8_t v = val; | |
3448 | MemTxResult r; | |
3449 | ||
3450 | r = address_space_rw(as, addr, attrs, &v, 1, 1); | |
3451 | if (result) { | |
3452 | *result = r; | |
3453 | } | |
3454 | } | |
3455 | ||
3456 | void stb_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3457 | { | |
3458 | address_space_stb(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3459 | } | |
3460 | ||
3461 | /* warning: addr must be aligned */ | |
3462 | static inline void address_space_stw_internal(AddressSpace *as, | |
3463 | hwaddr addr, uint32_t val, | |
3464 | MemTxAttrs attrs, | |
3465 | MemTxResult *result, | |
3466 | enum device_endian endian) | |
3467 | { | |
3468 | uint8_t *ptr; | |
3469 | MemoryRegion *mr; | |
3470 | hwaddr l = 2; | |
3471 | hwaddr addr1; | |
3472 | MemTxResult r; | |
3473 | bool release_lock = false; | |
3474 | ||
3475 | rcu_read_lock(); | |
3476 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
3477 | if (l < 2 || !memory_access_is_direct(mr, true)) { | |
3478 | release_lock |= prepare_mmio_access(mr); | |
3479 | ||
3480 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3481 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3482 | val = bswap16(val); | |
3483 | } | |
3484 | #else | |
3485 | if (endian == DEVICE_BIG_ENDIAN) { | |
3486 | val = bswap16(val); | |
3487 | } | |
3488 | #endif | |
3489 | r = memory_region_dispatch_write(mr, addr1, val, 2, attrs); | |
3490 | } else { | |
3491 | /* RAM case */ | |
3492 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); | |
3493 | switch (endian) { | |
3494 | case DEVICE_LITTLE_ENDIAN: | |
3495 | stw_le_p(ptr, val); | |
3496 | break; | |
3497 | case DEVICE_BIG_ENDIAN: | |
3498 | stw_be_p(ptr, val); | |
3499 | break; | |
3500 | default: | |
3501 | stw_p(ptr, val); | |
3502 | break; | |
3503 | } | |
3504 | invalidate_and_set_dirty(mr, addr1, 2); | |
3505 | r = MEMTX_OK; | |
3506 | } | |
3507 | if (result) { | |
3508 | *result = r; | |
3509 | } | |
3510 | if (release_lock) { | |
3511 | qemu_mutex_unlock_iothread(); | |
3512 | } | |
3513 | rcu_read_unlock(); | |
3514 | } | |
3515 | ||
3516 | void address_space_stw(AddressSpace *as, hwaddr addr, uint32_t val, | |
3517 | MemTxAttrs attrs, MemTxResult *result) | |
3518 | { | |
3519 | address_space_stw_internal(as, addr, val, attrs, result, | |
3520 | DEVICE_NATIVE_ENDIAN); | |
3521 | } | |
3522 | ||
3523 | void address_space_stw_le(AddressSpace *as, hwaddr addr, uint32_t val, | |
3524 | MemTxAttrs attrs, MemTxResult *result) | |
3525 | { | |
3526 | address_space_stw_internal(as, addr, val, attrs, result, | |
3527 | DEVICE_LITTLE_ENDIAN); | |
3528 | } | |
3529 | ||
3530 | void address_space_stw_be(AddressSpace *as, hwaddr addr, uint32_t val, | |
3531 | MemTxAttrs attrs, MemTxResult *result) | |
3532 | { | |
3533 | address_space_stw_internal(as, addr, val, attrs, result, | |
3534 | DEVICE_BIG_ENDIAN); | |
3535 | } | |
3536 | ||
3537 | void stw_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3538 | { | |
3539 | address_space_stw(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3540 | } | |
3541 | ||
3542 | void stw_le_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3543 | { | |
3544 | address_space_stw_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3545 | } | |
3546 | ||
3547 | void stw_be_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3548 | { | |
3549 | address_space_stw_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3550 | } | |
3551 | ||
3552 | /* XXX: optimize */ | |
3553 | void address_space_stq(AddressSpace *as, hwaddr addr, uint64_t val, | |
3554 | MemTxAttrs attrs, MemTxResult *result) | |
3555 | { | |
3556 | MemTxResult r; | |
3557 | val = tswap64(val); | |
3558 | r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1); | |
3559 | if (result) { | |
3560 | *result = r; | |
3561 | } | |
3562 | } | |
3563 | ||
3564 | void address_space_stq_le(AddressSpace *as, hwaddr addr, uint64_t val, | |
3565 | MemTxAttrs attrs, MemTxResult *result) | |
3566 | { | |
3567 | MemTxResult r; | |
3568 | val = cpu_to_le64(val); | |
3569 | r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1); | |
3570 | if (result) { | |
3571 | *result = r; | |
3572 | } | |
3573 | } | |
3574 | void address_space_stq_be(AddressSpace *as, hwaddr addr, uint64_t val, | |
3575 | MemTxAttrs attrs, MemTxResult *result) | |
3576 | { | |
3577 | MemTxResult r; | |
3578 | val = cpu_to_be64(val); | |
3579 | r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1); | |
3580 | if (result) { | |
3581 | *result = r; | |
3582 | } | |
3583 | } | |
3584 | ||
3585 | void stq_phys(AddressSpace *as, hwaddr addr, uint64_t val) | |
3586 | { | |
3587 | address_space_stq(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3588 | } | |
3589 | ||
3590 | void stq_le_phys(AddressSpace *as, hwaddr addr, uint64_t val) | |
3591 | { | |
3592 | address_space_stq_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3593 | } | |
3594 | ||
3595 | void stq_be_phys(AddressSpace *as, hwaddr addr, uint64_t val) | |
3596 | { | |
3597 | address_space_stq_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3598 | } | |
3599 | ||
3600 | /* virtual memory access for debug (includes writing to ROM) */ | |
3601 | int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr, | |
3602 | uint8_t *buf, int len, int is_write) | |
3603 | { | |
3604 | int l; | |
3605 | hwaddr phys_addr; | |
3606 | target_ulong page; | |
3607 | ||
3608 | while (len > 0) { | |
3609 | int asidx; | |
3610 | MemTxAttrs attrs; | |
3611 | ||
3612 | page = addr & TARGET_PAGE_MASK; | |
3613 | phys_addr = cpu_get_phys_page_attrs_debug(cpu, page, &attrs); | |
3614 | asidx = cpu_asidx_from_attrs(cpu, attrs); | |
3615 | /* if no physical page mapped, return an error */ | |
3616 | if (phys_addr == -1) | |
3617 | return -1; | |
3618 | l = (page + TARGET_PAGE_SIZE) - addr; | |
3619 | if (l > len) | |
3620 | l = len; | |
3621 | phys_addr += (addr & ~TARGET_PAGE_MASK); | |
3622 | if (is_write) { | |
3623 | cpu_physical_memory_write_rom(cpu->cpu_ases[asidx].as, | |
3624 | phys_addr, buf, l); | |
3625 | } else { | |
3626 | address_space_rw(cpu->cpu_ases[asidx].as, phys_addr, | |
3627 | MEMTXATTRS_UNSPECIFIED, | |
3628 | buf, l, 0); | |
3629 | } | |
3630 | len -= l; | |
3631 | buf += l; | |
3632 | addr += l; | |
3633 | } | |
3634 | return 0; | |
3635 | } | |
3636 | ||
3637 | /* | |
3638 | * Allows code that needs to deal with migration bitmaps etc to still be built | |
3639 | * target independent. | |
3640 | */ | |
3641 | size_t qemu_target_page_bits(void) | |
3642 | { | |
3643 | return TARGET_PAGE_BITS; | |
3644 | } | |
3645 | ||
3646 | #endif | |
3647 | ||
3648 | /* | |
3649 | * A helper function for the _utterly broken_ virtio device model to find out if | |
3650 | * it's running on a big endian machine. Don't do this at home kids! | |
3651 | */ | |
3652 | bool target_words_bigendian(void); | |
3653 | bool target_words_bigendian(void) | |
3654 | { | |
3655 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3656 | return true; | |
3657 | #else | |
3658 | return false; | |
3659 | #endif | |
3660 | } | |
3661 | ||
3662 | #ifndef CONFIG_USER_ONLY | |
3663 | bool cpu_physical_memory_is_io(hwaddr phys_addr) | |
3664 | { | |
3665 | MemoryRegion*mr; | |
3666 | hwaddr l = 1; | |
3667 | bool res; | |
3668 | ||
3669 | rcu_read_lock(); | |
3670 | mr = address_space_translate(&address_space_memory, | |
3671 | phys_addr, &phys_addr, &l, false); | |
3672 | ||
3673 | res = !(memory_region_is_ram(mr) || memory_region_is_romd(mr)); | |
3674 | rcu_read_unlock(); | |
3675 | return res; | |
3676 | } | |
3677 | ||
3678 | int qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque) | |
3679 | { | |
3680 | RAMBlock *block; | |
3681 | int ret = 0; | |
3682 | ||
3683 | rcu_read_lock(); | |
3684 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
3685 | ret = func(block->idstr, block->host, block->offset, | |
3686 | block->used_length, opaque); | |
3687 | if (ret) { | |
3688 | break; | |
3689 | } | |
3690 | } | |
3691 | rcu_read_unlock(); | |
3692 | return ret; | |
3693 | } | |
3694 | #endif |