]>
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 | #ifndef _WIN32 | |
21 | #include <sys/mman.h> | |
22 | #endif | |
23 | ||
24 | #include "qemu-common.h" | |
25 | #include "cpu.h" | |
26 | #include "tcg.h" | |
27 | #include "hw/hw.h" | |
28 | #if !defined(CONFIG_USER_ONLY) | |
29 | #include "hw/boards.h" | |
30 | #endif | |
31 | #include "hw/qdev.h" | |
32 | #include "sysemu/kvm.h" | |
33 | #include "sysemu/sysemu.h" | |
34 | #include "hw/xen/xen.h" | |
35 | #include "qemu/timer.h" | |
36 | #include "qemu/config-file.h" | |
37 | #include "qemu/error-report.h" | |
38 | #include "exec/memory.h" | |
39 | #include "sysemu/dma.h" | |
40 | #include "exec/address-spaces.h" | |
41 | #if defined(CONFIG_USER_ONLY) | |
42 | #include <qemu.h> | |
43 | #else /* !CONFIG_USER_ONLY */ | |
44 | #include "sysemu/xen-mapcache.h" | |
45 | #include "trace.h" | |
46 | #endif | |
47 | #include "exec/cpu-all.h" | |
48 | #include "qemu/rcu_queue.h" | |
49 | #include "qemu/main-loop.h" | |
50 | #include "translate-all.h" | |
51 | #include "sysemu/replay.h" | |
52 | ||
53 | #include "exec/memory-internal.h" | |
54 | #include "exec/ram_addr.h" | |
55 | #include "exec/log.h" | |
56 | ||
57 | #include "qemu/range.h" | |
58 | #ifndef _WIN32 | |
59 | #include "qemu/mmap-alloc.h" | |
60 | #endif | |
61 | ||
62 | //#define DEBUG_SUBPAGE | |
63 | ||
64 | #if !defined(CONFIG_USER_ONLY) | |
65 | /* ram_list is read under rcu_read_lock()/rcu_read_unlock(). Writes | |
66 | * are protected by the ramlist lock. | |
67 | */ | |
68 | RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) }; | |
69 | ||
70 | static MemoryRegion *system_memory; | |
71 | static MemoryRegion *system_io; | |
72 | ||
73 | AddressSpace address_space_io; | |
74 | AddressSpace address_space_memory; | |
75 | ||
76 | MemoryRegion io_mem_rom, io_mem_notdirty; | |
77 | static MemoryRegion io_mem_unassigned; | |
78 | ||
79 | /* RAM is pre-allocated and passed into qemu_ram_alloc_from_ptr */ | |
80 | #define RAM_PREALLOC (1 << 0) | |
81 | ||
82 | /* RAM is mmap-ed with MAP_SHARED */ | |
83 | #define RAM_SHARED (1 << 1) | |
84 | ||
85 | /* Only a portion of RAM (used_length) is actually used, and migrated. | |
86 | * This used_length size can change across reboots. | |
87 | */ | |
88 | #define RAM_RESIZEABLE (1 << 2) | |
89 | ||
90 | #endif | |
91 | ||
92 | struct CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus); | |
93 | /* current CPU in the current thread. It is only valid inside | |
94 | cpu_exec() */ | |
95 | __thread CPUState *current_cpu; | |
96 | /* 0 = Do not count executed instructions. | |
97 | 1 = Precise instruction counting. | |
98 | 2 = Adaptive rate instruction counting. */ | |
99 | int use_icount; | |
100 | ||
101 | #if !defined(CONFIG_USER_ONLY) | |
102 | ||
103 | typedef struct PhysPageEntry PhysPageEntry; | |
104 | ||
105 | struct PhysPageEntry { | |
106 | /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */ | |
107 | uint32_t skip : 6; | |
108 | /* index into phys_sections (!skip) or phys_map_nodes (skip) */ | |
109 | uint32_t ptr : 26; | |
110 | }; | |
111 | ||
112 | #define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6) | |
113 | ||
114 | /* Size of the L2 (and L3, etc) page tables. */ | |
115 | #define ADDR_SPACE_BITS 64 | |
116 | ||
117 | #define P_L2_BITS 9 | |
118 | #define P_L2_SIZE (1 << P_L2_BITS) | |
119 | ||
120 | #define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1) | |
121 | ||
122 | typedef PhysPageEntry Node[P_L2_SIZE]; | |
123 | ||
124 | typedef struct PhysPageMap { | |
125 | struct rcu_head rcu; | |
126 | ||
127 | unsigned sections_nb; | |
128 | unsigned sections_nb_alloc; | |
129 | unsigned nodes_nb; | |
130 | unsigned nodes_nb_alloc; | |
131 | Node *nodes; | |
132 | MemoryRegionSection *sections; | |
133 | } PhysPageMap; | |
134 | ||
135 | struct AddressSpaceDispatch { | |
136 | struct rcu_head rcu; | |
137 | ||
138 | MemoryRegionSection *mru_section; | |
139 | /* This is a multi-level map on the physical address space. | |
140 | * The bottom level has pointers to MemoryRegionSections. | |
141 | */ | |
142 | PhysPageEntry phys_map; | |
143 | PhysPageMap map; | |
144 | AddressSpace *as; | |
145 | }; | |
146 | ||
147 | #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK) | |
148 | typedef struct subpage_t { | |
149 | MemoryRegion iomem; | |
150 | AddressSpace *as; | |
151 | hwaddr base; | |
152 | uint16_t sub_section[TARGET_PAGE_SIZE]; | |
153 | } subpage_t; | |
154 | ||
155 | #define PHYS_SECTION_UNASSIGNED 0 | |
156 | #define PHYS_SECTION_NOTDIRTY 1 | |
157 | #define PHYS_SECTION_ROM 2 | |
158 | #define PHYS_SECTION_WATCH 3 | |
159 | ||
160 | static void io_mem_init(void); | |
161 | static void memory_map_init(void); | |
162 | static void tcg_commit(MemoryListener *listener); | |
163 | ||
164 | static MemoryRegion io_mem_watch; | |
165 | ||
166 | /** | |
167 | * CPUAddressSpace: all the information a CPU needs about an AddressSpace | |
168 | * @cpu: the CPU whose AddressSpace this is | |
169 | * @as: the AddressSpace itself | |
170 | * @memory_dispatch: its dispatch pointer (cached, RCU protected) | |
171 | * @tcg_as_listener: listener for tracking changes to the AddressSpace | |
172 | */ | |
173 | struct CPUAddressSpace { | |
174 | CPUState *cpu; | |
175 | AddressSpace *as; | |
176 | struct AddressSpaceDispatch *memory_dispatch; | |
177 | MemoryListener tcg_as_listener; | |
178 | }; | |
179 | ||
180 | #endif | |
181 | ||
182 | #if !defined(CONFIG_USER_ONLY) | |
183 | ||
184 | static void phys_map_node_reserve(PhysPageMap *map, unsigned nodes) | |
185 | { | |
186 | if (map->nodes_nb + nodes > map->nodes_nb_alloc) { | |
187 | map->nodes_nb_alloc = MAX(map->nodes_nb_alloc * 2, 16); | |
188 | map->nodes_nb_alloc = MAX(map->nodes_nb_alloc, map->nodes_nb + nodes); | |
189 | map->nodes = g_renew(Node, map->nodes, map->nodes_nb_alloc); | |
190 | } | |
191 | } | |
192 | ||
193 | static uint32_t phys_map_node_alloc(PhysPageMap *map, bool leaf) | |
194 | { | |
195 | unsigned i; | |
196 | uint32_t ret; | |
197 | PhysPageEntry e; | |
198 | PhysPageEntry *p; | |
199 | ||
200 | ret = map->nodes_nb++; | |
201 | p = map->nodes[ret]; | |
202 | assert(ret != PHYS_MAP_NODE_NIL); | |
203 | assert(ret != map->nodes_nb_alloc); | |
204 | ||
205 | e.skip = leaf ? 0 : 1; | |
206 | e.ptr = leaf ? PHYS_SECTION_UNASSIGNED : PHYS_MAP_NODE_NIL; | |
207 | for (i = 0; i < P_L2_SIZE; ++i) { | |
208 | memcpy(&p[i], &e, sizeof(e)); | |
209 | } | |
210 | return ret; | |
211 | } | |
212 | ||
213 | static void phys_page_set_level(PhysPageMap *map, PhysPageEntry *lp, | |
214 | hwaddr *index, hwaddr *nb, uint16_t leaf, | |
215 | int level) | |
216 | { | |
217 | PhysPageEntry *p; | |
218 | hwaddr step = (hwaddr)1 << (level * P_L2_BITS); | |
219 | ||
220 | if (lp->skip && lp->ptr == PHYS_MAP_NODE_NIL) { | |
221 | lp->ptr = phys_map_node_alloc(map, level == 0); | |
222 | } | |
223 | p = map->nodes[lp->ptr]; | |
224 | lp = &p[(*index >> (level * P_L2_BITS)) & (P_L2_SIZE - 1)]; | |
225 | ||
226 | while (*nb && lp < &p[P_L2_SIZE]) { | |
227 | if ((*index & (step - 1)) == 0 && *nb >= step) { | |
228 | lp->skip = 0; | |
229 | lp->ptr = leaf; | |
230 | *index += step; | |
231 | *nb -= step; | |
232 | } else { | |
233 | phys_page_set_level(map, lp, index, nb, leaf, level - 1); | |
234 | } | |
235 | ++lp; | |
236 | } | |
237 | } | |
238 | ||
239 | static void phys_page_set(AddressSpaceDispatch *d, | |
240 | hwaddr index, hwaddr nb, | |
241 | uint16_t leaf) | |
242 | { | |
243 | /* Wildly overreserve - it doesn't matter much. */ | |
244 | phys_map_node_reserve(&d->map, 3 * P_L2_LEVELS); | |
245 | ||
246 | phys_page_set_level(&d->map, &d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1); | |
247 | } | |
248 | ||
249 | /* Compact a non leaf page entry. Simply detect that the entry has a single child, | |
250 | * and update our entry so we can skip it and go directly to the destination. | |
251 | */ | |
252 | static void phys_page_compact(PhysPageEntry *lp, Node *nodes, unsigned long *compacted) | |
253 | { | |
254 | unsigned valid_ptr = P_L2_SIZE; | |
255 | int valid = 0; | |
256 | PhysPageEntry *p; | |
257 | int i; | |
258 | ||
259 | if (lp->ptr == PHYS_MAP_NODE_NIL) { | |
260 | return; | |
261 | } | |
262 | ||
263 | p = nodes[lp->ptr]; | |
264 | for (i = 0; i < P_L2_SIZE; i++) { | |
265 | if (p[i].ptr == PHYS_MAP_NODE_NIL) { | |
266 | continue; | |
267 | } | |
268 | ||
269 | valid_ptr = i; | |
270 | valid++; | |
271 | if (p[i].skip) { | |
272 | phys_page_compact(&p[i], nodes, compacted); | |
273 | } | |
274 | } | |
275 | ||
276 | /* We can only compress if there's only one child. */ | |
277 | if (valid != 1) { | |
278 | return; | |
279 | } | |
280 | ||
281 | assert(valid_ptr < P_L2_SIZE); | |
282 | ||
283 | /* Don't compress if it won't fit in the # of bits we have. */ | |
284 | if (lp->skip + p[valid_ptr].skip >= (1 << 3)) { | |
285 | return; | |
286 | } | |
287 | ||
288 | lp->ptr = p[valid_ptr].ptr; | |
289 | if (!p[valid_ptr].skip) { | |
290 | /* If our only child is a leaf, make this a leaf. */ | |
291 | /* By design, we should have made this node a leaf to begin with so we | |
292 | * should never reach here. | |
293 | * But since it's so simple to handle this, let's do it just in case we | |
294 | * change this rule. | |
295 | */ | |
296 | lp->skip = 0; | |
297 | } else { | |
298 | lp->skip += p[valid_ptr].skip; | |
299 | } | |
300 | } | |
301 | ||
302 | static void phys_page_compact_all(AddressSpaceDispatch *d, int nodes_nb) | |
303 | { | |
304 | DECLARE_BITMAP(compacted, nodes_nb); | |
305 | ||
306 | if (d->phys_map.skip) { | |
307 | phys_page_compact(&d->phys_map, d->map.nodes, compacted); | |
308 | } | |
309 | } | |
310 | ||
311 | static inline bool section_covers_addr(const MemoryRegionSection *section, | |
312 | hwaddr addr) | |
313 | { | |
314 | /* Memory topology clips a memory region to [0, 2^64); size.hi > 0 means | |
315 | * the section must cover the entire address space. | |
316 | */ | |
317 | return section->size.hi || | |
318 | range_covers_byte(section->offset_within_address_space, | |
319 | section->size.lo, addr); | |
320 | } | |
321 | ||
322 | static MemoryRegionSection *phys_page_find(PhysPageEntry lp, hwaddr addr, | |
323 | Node *nodes, MemoryRegionSection *sections) | |
324 | { | |
325 | PhysPageEntry *p; | |
326 | hwaddr index = addr >> TARGET_PAGE_BITS; | |
327 | int i; | |
328 | ||
329 | for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) { | |
330 | if (lp.ptr == PHYS_MAP_NODE_NIL) { | |
331 | return §ions[PHYS_SECTION_UNASSIGNED]; | |
332 | } | |
333 | p = nodes[lp.ptr]; | |
334 | lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)]; | |
335 | } | |
336 | ||
337 | if (section_covers_addr(§ions[lp.ptr], addr)) { | |
338 | return §ions[lp.ptr]; | |
339 | } else { | |
340 | return §ions[PHYS_SECTION_UNASSIGNED]; | |
341 | } | |
342 | } | |
343 | ||
344 | bool memory_region_is_unassigned(MemoryRegion *mr) | |
345 | { | |
346 | return mr != &io_mem_rom && mr != &io_mem_notdirty && !mr->rom_device | |
347 | && mr != &io_mem_watch; | |
348 | } | |
349 | ||
350 | /* Called from RCU critical section */ | |
351 | static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d, | |
352 | hwaddr addr, | |
353 | bool resolve_subpage) | |
354 | { | |
355 | MemoryRegionSection *section = atomic_read(&d->mru_section); | |
356 | subpage_t *subpage; | |
357 | bool update; | |
358 | ||
359 | if (section && section != &d->map.sections[PHYS_SECTION_UNASSIGNED] && | |
360 | section_covers_addr(section, addr)) { | |
361 | update = false; | |
362 | } else { | |
363 | section = phys_page_find(d->phys_map, addr, d->map.nodes, | |
364 | d->map.sections); | |
365 | update = true; | |
366 | } | |
367 | if (resolve_subpage && section->mr->subpage) { | |
368 | subpage = container_of(section->mr, subpage_t, iomem); | |
369 | section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]]; | |
370 | } | |
371 | if (update) { | |
372 | atomic_set(&d->mru_section, section); | |
373 | } | |
374 | return section; | |
375 | } | |
376 | ||
377 | /* Called from RCU critical section */ | |
378 | static MemoryRegionSection * | |
379 | address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat, | |
380 | hwaddr *plen, bool resolve_subpage) | |
381 | { | |
382 | MemoryRegionSection *section; | |
383 | MemoryRegion *mr; | |
384 | Int128 diff; | |
385 | ||
386 | section = address_space_lookup_region(d, addr, resolve_subpage); | |
387 | /* Compute offset within MemoryRegionSection */ | |
388 | addr -= section->offset_within_address_space; | |
389 | ||
390 | /* Compute offset within MemoryRegion */ | |
391 | *xlat = addr + section->offset_within_region; | |
392 | ||
393 | mr = section->mr; | |
394 | ||
395 | /* MMIO registers can be expected to perform full-width accesses based only | |
396 | * on their address, without considering adjacent registers that could | |
397 | * decode to completely different MemoryRegions. When such registers | |
398 | * exist (e.g. I/O ports 0xcf8 and 0xcf9 on most PC chipsets), MMIO | |
399 | * regions overlap wildly. For this reason we cannot clamp the accesses | |
400 | * here. | |
401 | * | |
402 | * If the length is small (as is the case for address_space_ldl/stl), | |
403 | * everything works fine. If the incoming length is large, however, | |
404 | * the caller really has to do the clamping through memory_access_size. | |
405 | */ | |
406 | if (memory_region_is_ram(mr)) { | |
407 | diff = int128_sub(section->size, int128_make64(addr)); | |
408 | *plen = int128_get64(int128_min(diff, int128_make64(*plen))); | |
409 | } | |
410 | return section; | |
411 | } | |
412 | ||
413 | /* Called from RCU critical section */ | |
414 | MemoryRegion *address_space_translate(AddressSpace *as, hwaddr addr, | |
415 | hwaddr *xlat, hwaddr *plen, | |
416 | bool is_write) | |
417 | { | |
418 | IOMMUTLBEntry iotlb; | |
419 | MemoryRegionSection *section; | |
420 | MemoryRegion *mr; | |
421 | ||
422 | for (;;) { | |
423 | AddressSpaceDispatch *d = atomic_rcu_read(&as->dispatch); | |
424 | section = address_space_translate_internal(d, addr, &addr, plen, true); | |
425 | mr = section->mr; | |
426 | ||
427 | if (!mr->iommu_ops) { | |
428 | break; | |
429 | } | |
430 | ||
431 | iotlb = mr->iommu_ops->translate(mr, addr, is_write); | |
432 | addr = ((iotlb.translated_addr & ~iotlb.addr_mask) | |
433 | | (addr & iotlb.addr_mask)); | |
434 | *plen = MIN(*plen, (addr | iotlb.addr_mask) - addr + 1); | |
435 | if (!(iotlb.perm & (1 << is_write))) { | |
436 | mr = &io_mem_unassigned; | |
437 | break; | |
438 | } | |
439 | ||
440 | as = iotlb.target_as; | |
441 | } | |
442 | ||
443 | if (xen_enabled() && memory_access_is_direct(mr, is_write)) { | |
444 | hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr; | |
445 | *plen = MIN(page, *plen); | |
446 | } | |
447 | ||
448 | *xlat = addr; | |
449 | return mr; | |
450 | } | |
451 | ||
452 | /* Called from RCU critical section */ | |
453 | MemoryRegionSection * | |
454 | address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr, | |
455 | hwaddr *xlat, hwaddr *plen) | |
456 | { | |
457 | MemoryRegionSection *section; | |
458 | AddressSpaceDispatch *d = cpu->cpu_ases[asidx].memory_dispatch; | |
459 | ||
460 | section = address_space_translate_internal(d, addr, xlat, plen, false); | |
461 | ||
462 | assert(!section->mr->iommu_ops); | |
463 | return section; | |
464 | } | |
465 | #endif | |
466 | ||
467 | #if !defined(CONFIG_USER_ONLY) | |
468 | ||
469 | static int cpu_common_post_load(void *opaque, int version_id) | |
470 | { | |
471 | CPUState *cpu = opaque; | |
472 | ||
473 | /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the | |
474 | version_id is increased. */ | |
475 | cpu->interrupt_request &= ~0x01; | |
476 | tlb_flush(cpu, 1); | |
477 | ||
478 | return 0; | |
479 | } | |
480 | ||
481 | static int cpu_common_pre_load(void *opaque) | |
482 | { | |
483 | CPUState *cpu = opaque; | |
484 | ||
485 | cpu->exception_index = -1; | |
486 | ||
487 | return 0; | |
488 | } | |
489 | ||
490 | static bool cpu_common_exception_index_needed(void *opaque) | |
491 | { | |
492 | CPUState *cpu = opaque; | |
493 | ||
494 | return tcg_enabled() && cpu->exception_index != -1; | |
495 | } | |
496 | ||
497 | static const VMStateDescription vmstate_cpu_common_exception_index = { | |
498 | .name = "cpu_common/exception_index", | |
499 | .version_id = 1, | |
500 | .minimum_version_id = 1, | |
501 | .needed = cpu_common_exception_index_needed, | |
502 | .fields = (VMStateField[]) { | |
503 | VMSTATE_INT32(exception_index, CPUState), | |
504 | VMSTATE_END_OF_LIST() | |
505 | } | |
506 | }; | |
507 | ||
508 | static bool cpu_common_crash_occurred_needed(void *opaque) | |
509 | { | |
510 | CPUState *cpu = opaque; | |
511 | ||
512 | return cpu->crash_occurred; | |
513 | } | |
514 | ||
515 | static const VMStateDescription vmstate_cpu_common_crash_occurred = { | |
516 | .name = "cpu_common/crash_occurred", | |
517 | .version_id = 1, | |
518 | .minimum_version_id = 1, | |
519 | .needed = cpu_common_crash_occurred_needed, | |
520 | .fields = (VMStateField[]) { | |
521 | VMSTATE_BOOL(crash_occurred, CPUState), | |
522 | VMSTATE_END_OF_LIST() | |
523 | } | |
524 | }; | |
525 | ||
526 | const VMStateDescription vmstate_cpu_common = { | |
527 | .name = "cpu_common", | |
528 | .version_id = 1, | |
529 | .minimum_version_id = 1, | |
530 | .pre_load = cpu_common_pre_load, | |
531 | .post_load = cpu_common_post_load, | |
532 | .fields = (VMStateField[]) { | |
533 | VMSTATE_UINT32(halted, CPUState), | |
534 | VMSTATE_UINT32(interrupt_request, CPUState), | |
535 | VMSTATE_END_OF_LIST() | |
536 | }, | |
537 | .subsections = (const VMStateDescription*[]) { | |
538 | &vmstate_cpu_common_exception_index, | |
539 | &vmstate_cpu_common_crash_occurred, | |
540 | NULL | |
541 | } | |
542 | }; | |
543 | ||
544 | #endif | |
545 | ||
546 | CPUState *qemu_get_cpu(int index) | |
547 | { | |
548 | CPUState *cpu; | |
549 | ||
550 | CPU_FOREACH(cpu) { | |
551 | if (cpu->cpu_index == index) { | |
552 | return cpu; | |
553 | } | |
554 | } | |
555 | ||
556 | return NULL; | |
557 | } | |
558 | ||
559 | #if !defined(CONFIG_USER_ONLY) | |
560 | void cpu_address_space_init(CPUState *cpu, AddressSpace *as, int asidx) | |
561 | { | |
562 | CPUAddressSpace *newas; | |
563 | ||
564 | /* Target code should have set num_ases before calling us */ | |
565 | assert(asidx < cpu->num_ases); | |
566 | ||
567 | if (asidx == 0) { | |
568 | /* address space 0 gets the convenience alias */ | |
569 | cpu->as = as; | |
570 | } | |
571 | ||
572 | /* KVM cannot currently support multiple address spaces. */ | |
573 | assert(asidx == 0 || !kvm_enabled()); | |
574 | ||
575 | if (!cpu->cpu_ases) { | |
576 | cpu->cpu_ases = g_new0(CPUAddressSpace, cpu->num_ases); | |
577 | } | |
578 | ||
579 | newas = &cpu->cpu_ases[asidx]; | |
580 | newas->cpu = cpu; | |
581 | newas->as = as; | |
582 | if (tcg_enabled()) { | |
583 | newas->tcg_as_listener.commit = tcg_commit; | |
584 | memory_listener_register(&newas->tcg_as_listener, as); | |
585 | } | |
586 | } | |
587 | ||
588 | AddressSpace *cpu_get_address_space(CPUState *cpu, int asidx) | |
589 | { | |
590 | /* Return the AddressSpace corresponding to the specified index */ | |
591 | return cpu->cpu_ases[asidx].as; | |
592 | } | |
593 | #endif | |
594 | ||
595 | #ifndef CONFIG_USER_ONLY | |
596 | static DECLARE_BITMAP(cpu_index_map, MAX_CPUMASK_BITS); | |
597 | ||
598 | static int cpu_get_free_index(Error **errp) | |
599 | { | |
600 | int cpu = find_first_zero_bit(cpu_index_map, MAX_CPUMASK_BITS); | |
601 | ||
602 | if (cpu >= MAX_CPUMASK_BITS) { | |
603 | error_setg(errp, "Trying to use more CPUs than max of %d", | |
604 | MAX_CPUMASK_BITS); | |
605 | return -1; | |
606 | } | |
607 | ||
608 | bitmap_set(cpu_index_map, cpu, 1); | |
609 | return cpu; | |
610 | } | |
611 | ||
612 | void cpu_exec_exit(CPUState *cpu) | |
613 | { | |
614 | if (cpu->cpu_index == -1) { | |
615 | /* cpu_index was never allocated by this @cpu or was already freed. */ | |
616 | return; | |
617 | } | |
618 | ||
619 | bitmap_clear(cpu_index_map, cpu->cpu_index, 1); | |
620 | cpu->cpu_index = -1; | |
621 | } | |
622 | #else | |
623 | ||
624 | static int cpu_get_free_index(Error **errp) | |
625 | { | |
626 | CPUState *some_cpu; | |
627 | int cpu_index = 0; | |
628 | ||
629 | CPU_FOREACH(some_cpu) { | |
630 | cpu_index++; | |
631 | } | |
632 | return cpu_index; | |
633 | } | |
634 | ||
635 | void cpu_exec_exit(CPUState *cpu) | |
636 | { | |
637 | } | |
638 | #endif | |
639 | ||
640 | void cpu_exec_init(CPUState *cpu, Error **errp) | |
641 | { | |
642 | CPUClass *cc = CPU_GET_CLASS(cpu); | |
643 | int cpu_index; | |
644 | Error *local_err = 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 | #if defined(CONFIG_USER_ONLY) | |
668 | cpu_list_lock(); | |
669 | #endif | |
670 | cpu_index = cpu->cpu_index = cpu_get_free_index(&local_err); | |
671 | if (local_err) { | |
672 | error_propagate(errp, local_err); | |
673 | #if defined(CONFIG_USER_ONLY) | |
674 | cpu_list_unlock(); | |
675 | #endif | |
676 | return; | |
677 | } | |
678 | QTAILQ_INSERT_TAIL(&cpus, cpu, node); | |
679 | #if defined(CONFIG_USER_ONLY) | |
680 | cpu_list_unlock(); | |
681 | #endif | |
682 | if (qdev_get_vmsd(DEVICE(cpu)) == NULL) { | |
683 | vmstate_register(NULL, cpu_index, &vmstate_cpu_common, cpu); | |
684 | } | |
685 | if (cc->vmsd != NULL) { | |
686 | vmstate_register(NULL, cpu_index, cc->vmsd, cpu); | |
687 | } | |
688 | } | |
689 | ||
690 | #if defined(CONFIG_USER_ONLY) | |
691 | static void breakpoint_invalidate(CPUState *cpu, target_ulong pc) | |
692 | { | |
693 | tb_invalidate_phys_page_range(pc, pc + 1, 0); | |
694 | } | |
695 | #else | |
696 | static void breakpoint_invalidate(CPUState *cpu, target_ulong pc) | |
697 | { | |
698 | MemTxAttrs attrs; | |
699 | hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs); | |
700 | int asidx = cpu_asidx_from_attrs(cpu, attrs); | |
701 | if (phys != -1) { | |
702 | tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as, | |
703 | phys | (pc & ~TARGET_PAGE_MASK)); | |
704 | } | |
705 | } | |
706 | #endif | |
707 | ||
708 | #if defined(CONFIG_USER_ONLY) | |
709 | void cpu_watchpoint_remove_all(CPUState *cpu, int mask) | |
710 | ||
711 | { | |
712 | } | |
713 | ||
714 | int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len, | |
715 | int flags) | |
716 | { | |
717 | return -ENOSYS; | |
718 | } | |
719 | ||
720 | void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint) | |
721 | { | |
722 | } | |
723 | ||
724 | int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len, | |
725 | int flags, CPUWatchpoint **watchpoint) | |
726 | { | |
727 | return -ENOSYS; | |
728 | } | |
729 | #else | |
730 | /* Add a watchpoint. */ | |
731 | int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len, | |
732 | int flags, CPUWatchpoint **watchpoint) | |
733 | { | |
734 | CPUWatchpoint *wp; | |
735 | ||
736 | /* forbid ranges which are empty or run off the end of the address space */ | |
737 | if (len == 0 || (addr + len - 1) < addr) { | |
738 | error_report("tried to set invalid watchpoint at %" | |
739 | VADDR_PRIx ", len=%" VADDR_PRIu, addr, len); | |
740 | return -EINVAL; | |
741 | } | |
742 | wp = g_malloc(sizeof(*wp)); | |
743 | ||
744 | wp->vaddr = addr; | |
745 | wp->len = len; | |
746 | wp->flags = flags; | |
747 | ||
748 | /* keep all GDB-injected watchpoints in front */ | |
749 | if (flags & BP_GDB) { | |
750 | QTAILQ_INSERT_HEAD(&cpu->watchpoints, wp, entry); | |
751 | } else { | |
752 | QTAILQ_INSERT_TAIL(&cpu->watchpoints, wp, entry); | |
753 | } | |
754 | ||
755 | tlb_flush_page(cpu, addr); | |
756 | ||
757 | if (watchpoint) | |
758 | *watchpoint = wp; | |
759 | return 0; | |
760 | } | |
761 | ||
762 | /* Remove a specific watchpoint. */ | |
763 | int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len, | |
764 | int flags) | |
765 | { | |
766 | CPUWatchpoint *wp; | |
767 | ||
768 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { | |
769 | if (addr == wp->vaddr && len == wp->len | |
770 | && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) { | |
771 | cpu_watchpoint_remove_by_ref(cpu, wp); | |
772 | return 0; | |
773 | } | |
774 | } | |
775 | return -ENOENT; | |
776 | } | |
777 | ||
778 | /* Remove a specific watchpoint by reference. */ | |
779 | void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint) | |
780 | { | |
781 | QTAILQ_REMOVE(&cpu->watchpoints, watchpoint, entry); | |
782 | ||
783 | tlb_flush_page(cpu, watchpoint->vaddr); | |
784 | ||
785 | g_free(watchpoint); | |
786 | } | |
787 | ||
788 | /* Remove all matching watchpoints. */ | |
789 | void cpu_watchpoint_remove_all(CPUState *cpu, int mask) | |
790 | { | |
791 | CPUWatchpoint *wp, *next; | |
792 | ||
793 | QTAILQ_FOREACH_SAFE(wp, &cpu->watchpoints, entry, next) { | |
794 | if (wp->flags & mask) { | |
795 | cpu_watchpoint_remove_by_ref(cpu, wp); | |
796 | } | |
797 | } | |
798 | } | |
799 | ||
800 | /* Return true if this watchpoint address matches the specified | |
801 | * access (ie the address range covered by the watchpoint overlaps | |
802 | * partially or completely with the address range covered by the | |
803 | * access). | |
804 | */ | |
805 | static inline bool cpu_watchpoint_address_matches(CPUWatchpoint *wp, | |
806 | vaddr addr, | |
807 | vaddr len) | |
808 | { | |
809 | /* We know the lengths are non-zero, but a little caution is | |
810 | * required to avoid errors in the case where the range ends | |
811 | * exactly at the top of the address space and so addr + len | |
812 | * wraps round to zero. | |
813 | */ | |
814 | vaddr wpend = wp->vaddr + wp->len - 1; | |
815 | vaddr addrend = addr + len - 1; | |
816 | ||
817 | return !(addr > wpend || wp->vaddr > addrend); | |
818 | } | |
819 | ||
820 | #endif | |
821 | ||
822 | /* Add a breakpoint. */ | |
823 | int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags, | |
824 | CPUBreakpoint **breakpoint) | |
825 | { | |
826 | CPUBreakpoint *bp; | |
827 | ||
828 | bp = g_malloc(sizeof(*bp)); | |
829 | ||
830 | bp->pc = pc; | |
831 | bp->flags = flags; | |
832 | ||
833 | /* keep all GDB-injected breakpoints in front */ | |
834 | if (flags & BP_GDB) { | |
835 | QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry); | |
836 | } else { | |
837 | QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry); | |
838 | } | |
839 | ||
840 | breakpoint_invalidate(cpu, pc); | |
841 | ||
842 | if (breakpoint) { | |
843 | *breakpoint = bp; | |
844 | } | |
845 | return 0; | |
846 | } | |
847 | ||
848 | /* Remove a specific breakpoint. */ | |
849 | int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags) | |
850 | { | |
851 | CPUBreakpoint *bp; | |
852 | ||
853 | QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) { | |
854 | if (bp->pc == pc && bp->flags == flags) { | |
855 | cpu_breakpoint_remove_by_ref(cpu, bp); | |
856 | return 0; | |
857 | } | |
858 | } | |
859 | return -ENOENT; | |
860 | } | |
861 | ||
862 | /* Remove a specific breakpoint by reference. */ | |
863 | void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *breakpoint) | |
864 | { | |
865 | QTAILQ_REMOVE(&cpu->breakpoints, breakpoint, entry); | |
866 | ||
867 | breakpoint_invalidate(cpu, breakpoint->pc); | |
868 | ||
869 | g_free(breakpoint); | |
870 | } | |
871 | ||
872 | /* Remove all matching breakpoints. */ | |
873 | void cpu_breakpoint_remove_all(CPUState *cpu, int mask) | |
874 | { | |
875 | CPUBreakpoint *bp, *next; | |
876 | ||
877 | QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) { | |
878 | if (bp->flags & mask) { | |
879 | cpu_breakpoint_remove_by_ref(cpu, bp); | |
880 | } | |
881 | } | |
882 | } | |
883 | ||
884 | /* enable or disable single step mode. EXCP_DEBUG is returned by the | |
885 | CPU loop after each instruction */ | |
886 | void cpu_single_step(CPUState *cpu, int enabled) | |
887 | { | |
888 | if (cpu->singlestep_enabled != enabled) { | |
889 | cpu->singlestep_enabled = enabled; | |
890 | if (kvm_enabled()) { | |
891 | kvm_update_guest_debug(cpu, 0); | |
892 | } else { | |
893 | /* must flush all the translated code to avoid inconsistencies */ | |
894 | /* XXX: only flush what is necessary */ | |
895 | tb_flush(cpu); | |
896 | } | |
897 | } | |
898 | } | |
899 | ||
900 | void cpu_abort(CPUState *cpu, const char *fmt, ...) | |
901 | { | |
902 | va_list ap; | |
903 | va_list ap2; | |
904 | ||
905 | va_start(ap, fmt); | |
906 | va_copy(ap2, ap); | |
907 | fprintf(stderr, "qemu: fatal: "); | |
908 | vfprintf(stderr, fmt, ap); | |
909 | fprintf(stderr, "\n"); | |
910 | cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU | CPU_DUMP_CCOP); | |
911 | if (qemu_log_separate()) { | |
912 | qemu_log("qemu: fatal: "); | |
913 | qemu_log_vprintf(fmt, ap2); | |
914 | qemu_log("\n"); | |
915 | log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP); | |
916 | qemu_log_flush(); | |
917 | qemu_log_close(); | |
918 | } | |
919 | va_end(ap2); | |
920 | va_end(ap); | |
921 | replay_finish(); | |
922 | #if defined(CONFIG_USER_ONLY) | |
923 | { | |
924 | struct sigaction act; | |
925 | sigfillset(&act.sa_mask); | |
926 | act.sa_handler = SIG_DFL; | |
927 | sigaction(SIGABRT, &act, NULL); | |
928 | } | |
929 | #endif | |
930 | abort(); | |
931 | } | |
932 | ||
933 | #if !defined(CONFIG_USER_ONLY) | |
934 | /* Called from RCU critical section */ | |
935 | static RAMBlock *qemu_get_ram_block(ram_addr_t addr) | |
936 | { | |
937 | RAMBlock *block; | |
938 | ||
939 | block = atomic_rcu_read(&ram_list.mru_block); | |
940 | if (block && addr - block->offset < block->max_length) { | |
941 | return block; | |
942 | } | |
943 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
944 | if (addr - block->offset < block->max_length) { | |
945 | goto found; | |
946 | } | |
947 | } | |
948 | ||
949 | fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr); | |
950 | abort(); | |
951 | ||
952 | found: | |
953 | /* It is safe to write mru_block outside the iothread lock. This | |
954 | * is what happens: | |
955 | * | |
956 | * mru_block = xxx | |
957 | * rcu_read_unlock() | |
958 | * xxx removed from list | |
959 | * rcu_read_lock() | |
960 | * read mru_block | |
961 | * mru_block = NULL; | |
962 | * call_rcu(reclaim_ramblock, xxx); | |
963 | * rcu_read_unlock() | |
964 | * | |
965 | * atomic_rcu_set is not needed here. The block was already published | |
966 | * when it was placed into the list. Here we're just making an extra | |
967 | * copy of the pointer. | |
968 | */ | |
969 | ram_list.mru_block = block; | |
970 | return block; | |
971 | } | |
972 | ||
973 | static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length) | |
974 | { | |
975 | CPUState *cpu; | |
976 | ram_addr_t start1; | |
977 | RAMBlock *block; | |
978 | ram_addr_t end; | |
979 | ||
980 | end = TARGET_PAGE_ALIGN(start + length); | |
981 | start &= TARGET_PAGE_MASK; | |
982 | ||
983 | rcu_read_lock(); | |
984 | block = qemu_get_ram_block(start); | |
985 | assert(block == qemu_get_ram_block(end - 1)); | |
986 | start1 = (uintptr_t)ramblock_ptr(block, start - block->offset); | |
987 | CPU_FOREACH(cpu) { | |
988 | tlb_reset_dirty(cpu, start1, length); | |
989 | } | |
990 | rcu_read_unlock(); | |
991 | } | |
992 | ||
993 | /* Note: start and end must be within the same ram block. */ | |
994 | bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start, | |
995 | ram_addr_t length, | |
996 | unsigned client) | |
997 | { | |
998 | DirtyMemoryBlocks *blocks; | |
999 | unsigned long end, page; | |
1000 | bool dirty = false; | |
1001 | ||
1002 | if (length == 0) { | |
1003 | return false; | |
1004 | } | |
1005 | ||
1006 | end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS; | |
1007 | page = start >> TARGET_PAGE_BITS; | |
1008 | ||
1009 | rcu_read_lock(); | |
1010 | ||
1011 | blocks = atomic_rcu_read(&ram_list.dirty_memory[client]); | |
1012 | ||
1013 | while (page < end) { | |
1014 | unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE; | |
1015 | unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE; | |
1016 | unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset); | |
1017 | ||
1018 | dirty |= bitmap_test_and_clear_atomic(blocks->blocks[idx], | |
1019 | offset, num); | |
1020 | page += num; | |
1021 | } | |
1022 | ||
1023 | rcu_read_unlock(); | |
1024 | ||
1025 | if (dirty && tcg_enabled()) { | |
1026 | tlb_reset_dirty_range_all(start, length); | |
1027 | } | |
1028 | ||
1029 | return dirty; | |
1030 | } | |
1031 | ||
1032 | /* Called from RCU critical section */ | |
1033 | hwaddr memory_region_section_get_iotlb(CPUState *cpu, | |
1034 | MemoryRegionSection *section, | |
1035 | target_ulong vaddr, | |
1036 | hwaddr paddr, hwaddr xlat, | |
1037 | int prot, | |
1038 | target_ulong *address) | |
1039 | { | |
1040 | hwaddr iotlb; | |
1041 | CPUWatchpoint *wp; | |
1042 | ||
1043 | if (memory_region_is_ram(section->mr)) { | |
1044 | /* Normal RAM. */ | |
1045 | iotlb = (memory_region_get_ram_addr(section->mr) & TARGET_PAGE_MASK) | |
1046 | + xlat; | |
1047 | if (!section->readonly) { | |
1048 | iotlb |= PHYS_SECTION_NOTDIRTY; | |
1049 | } else { | |
1050 | iotlb |= PHYS_SECTION_ROM; | |
1051 | } | |
1052 | } else { | |
1053 | AddressSpaceDispatch *d; | |
1054 | ||
1055 | d = atomic_rcu_read(§ion->address_space->dispatch); | |
1056 | iotlb = section - d->map.sections; | |
1057 | iotlb += xlat; | |
1058 | } | |
1059 | ||
1060 | /* Make accesses to pages with watchpoints go via the | |
1061 | watchpoint trap routines. */ | |
1062 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { | |
1063 | if (cpu_watchpoint_address_matches(wp, vaddr, TARGET_PAGE_SIZE)) { | |
1064 | /* Avoid trapping reads of pages with a write breakpoint. */ | |
1065 | if ((prot & PAGE_WRITE) || (wp->flags & BP_MEM_READ)) { | |
1066 | iotlb = PHYS_SECTION_WATCH + paddr; | |
1067 | *address |= TLB_MMIO; | |
1068 | break; | |
1069 | } | |
1070 | } | |
1071 | } | |
1072 | ||
1073 | return iotlb; | |
1074 | } | |
1075 | #endif /* defined(CONFIG_USER_ONLY) */ | |
1076 | ||
1077 | #if !defined(CONFIG_USER_ONLY) | |
1078 | ||
1079 | static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end, | |
1080 | uint16_t section); | |
1081 | static subpage_t *subpage_init(AddressSpace *as, hwaddr base); | |
1082 | ||
1083 | static void *(*phys_mem_alloc)(size_t size, uint64_t *align) = | |
1084 | qemu_anon_ram_alloc; | |
1085 | ||
1086 | /* | |
1087 | * Set a custom physical guest memory alloator. | |
1088 | * Accelerators with unusual needs may need this. Hopefully, we can | |
1089 | * get rid of it eventually. | |
1090 | */ | |
1091 | void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align)) | |
1092 | { | |
1093 | phys_mem_alloc = alloc; | |
1094 | } | |
1095 | ||
1096 | static uint16_t phys_section_add(PhysPageMap *map, | |
1097 | MemoryRegionSection *section) | |
1098 | { | |
1099 | /* The physical section number is ORed with a page-aligned | |
1100 | * pointer to produce the iotlb entries. Thus it should | |
1101 | * never overflow into the page-aligned value. | |
1102 | */ | |
1103 | assert(map->sections_nb < TARGET_PAGE_SIZE); | |
1104 | ||
1105 | if (map->sections_nb == map->sections_nb_alloc) { | |
1106 | map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16); | |
1107 | map->sections = g_renew(MemoryRegionSection, map->sections, | |
1108 | map->sections_nb_alloc); | |
1109 | } | |
1110 | map->sections[map->sections_nb] = *section; | |
1111 | memory_region_ref(section->mr); | |
1112 | return map->sections_nb++; | |
1113 | } | |
1114 | ||
1115 | static void phys_section_destroy(MemoryRegion *mr) | |
1116 | { | |
1117 | bool have_sub_page = mr->subpage; | |
1118 | ||
1119 | memory_region_unref(mr); | |
1120 | ||
1121 | if (have_sub_page) { | |
1122 | subpage_t *subpage = container_of(mr, subpage_t, iomem); | |
1123 | object_unref(OBJECT(&subpage->iomem)); | |
1124 | g_free(subpage); | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | static void phys_sections_free(PhysPageMap *map) | |
1129 | { | |
1130 | while (map->sections_nb > 0) { | |
1131 | MemoryRegionSection *section = &map->sections[--map->sections_nb]; | |
1132 | phys_section_destroy(section->mr); | |
1133 | } | |
1134 | g_free(map->sections); | |
1135 | g_free(map->nodes); | |
1136 | } | |
1137 | ||
1138 | static void register_subpage(AddressSpaceDispatch *d, MemoryRegionSection *section) | |
1139 | { | |
1140 | subpage_t *subpage; | |
1141 | hwaddr base = section->offset_within_address_space | |
1142 | & TARGET_PAGE_MASK; | |
1143 | MemoryRegionSection *existing = phys_page_find(d->phys_map, base, | |
1144 | d->map.nodes, d->map.sections); | |
1145 | MemoryRegionSection subsection = { | |
1146 | .offset_within_address_space = base, | |
1147 | .size = int128_make64(TARGET_PAGE_SIZE), | |
1148 | }; | |
1149 | hwaddr start, end; | |
1150 | ||
1151 | assert(existing->mr->subpage || existing->mr == &io_mem_unassigned); | |
1152 | ||
1153 | if (!(existing->mr->subpage)) { | |
1154 | subpage = subpage_init(d->as, base); | |
1155 | subsection.address_space = d->as; | |
1156 | subsection.mr = &subpage->iomem; | |
1157 | phys_page_set(d, base >> TARGET_PAGE_BITS, 1, | |
1158 | phys_section_add(&d->map, &subsection)); | |
1159 | } else { | |
1160 | subpage = container_of(existing->mr, subpage_t, iomem); | |
1161 | } | |
1162 | start = section->offset_within_address_space & ~TARGET_PAGE_MASK; | |
1163 | end = start + int128_get64(section->size) - 1; | |
1164 | subpage_register(subpage, start, end, | |
1165 | phys_section_add(&d->map, section)); | |
1166 | } | |
1167 | ||
1168 | ||
1169 | static void register_multipage(AddressSpaceDispatch *d, | |
1170 | MemoryRegionSection *section) | |
1171 | { | |
1172 | hwaddr start_addr = section->offset_within_address_space; | |
1173 | uint16_t section_index = phys_section_add(&d->map, section); | |
1174 | uint64_t num_pages = int128_get64(int128_rshift(section->size, | |
1175 | TARGET_PAGE_BITS)); | |
1176 | ||
1177 | assert(num_pages); | |
1178 | phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index); | |
1179 | } | |
1180 | ||
1181 | static void mem_add(MemoryListener *listener, MemoryRegionSection *section) | |
1182 | { | |
1183 | AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener); | |
1184 | AddressSpaceDispatch *d = as->next_dispatch; | |
1185 | MemoryRegionSection now = *section, remain = *section; | |
1186 | Int128 page_size = int128_make64(TARGET_PAGE_SIZE); | |
1187 | ||
1188 | if (now.offset_within_address_space & ~TARGET_PAGE_MASK) { | |
1189 | uint64_t left = TARGET_PAGE_ALIGN(now.offset_within_address_space) | |
1190 | - now.offset_within_address_space; | |
1191 | ||
1192 | now.size = int128_min(int128_make64(left), now.size); | |
1193 | register_subpage(d, &now); | |
1194 | } else { | |
1195 | now.size = int128_zero(); | |
1196 | } | |
1197 | while (int128_ne(remain.size, now.size)) { | |
1198 | remain.size = int128_sub(remain.size, now.size); | |
1199 | remain.offset_within_address_space += int128_get64(now.size); | |
1200 | remain.offset_within_region += int128_get64(now.size); | |
1201 | now = remain; | |
1202 | if (int128_lt(remain.size, page_size)) { | |
1203 | register_subpage(d, &now); | |
1204 | } else if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) { | |
1205 | now.size = page_size; | |
1206 | register_subpage(d, &now); | |
1207 | } else { | |
1208 | now.size = int128_and(now.size, int128_neg(page_size)); | |
1209 | register_multipage(d, &now); | |
1210 | } | |
1211 | } | |
1212 | } | |
1213 | ||
1214 | void qemu_flush_coalesced_mmio_buffer(void) | |
1215 | { | |
1216 | if (kvm_enabled()) | |
1217 | kvm_flush_coalesced_mmio_buffer(); | |
1218 | } | |
1219 | ||
1220 | void qemu_mutex_lock_ramlist(void) | |
1221 | { | |
1222 | qemu_mutex_lock(&ram_list.mutex); | |
1223 | } | |
1224 | ||
1225 | void qemu_mutex_unlock_ramlist(void) | |
1226 | { | |
1227 | qemu_mutex_unlock(&ram_list.mutex); | |
1228 | } | |
1229 | ||
1230 | #ifdef __linux__ | |
1231 | ||
1232 | #include <sys/vfs.h> | |
1233 | ||
1234 | #define HUGETLBFS_MAGIC 0x958458f6 | |
1235 | ||
1236 | static long gethugepagesize(int fd) | |
1237 | { | |
1238 | struct statfs fs; | |
1239 | int ret; | |
1240 | ||
1241 | do { | |
1242 | ret = fstatfs(fd, &fs); | |
1243 | } while (ret != 0 && errno == EINTR); | |
1244 | ||
1245 | if (ret != 0) { | |
1246 | return -1; | |
1247 | } | |
1248 | ||
1249 | return fs.f_bsize; | |
1250 | } | |
1251 | ||
1252 | static void *file_ram_alloc(RAMBlock *block, | |
1253 | ram_addr_t memory, | |
1254 | const char *path, | |
1255 | Error **errp) | |
1256 | { | |
1257 | bool unlink_on_error = false; | |
1258 | char *filename; | |
1259 | char *sanitized_name; | |
1260 | char *c; | |
1261 | void *area; | |
1262 | int fd; | |
1263 | int64_t hpagesize; | |
1264 | ||
1265 | if (kvm_enabled() && !kvm_has_sync_mmu()) { | |
1266 | error_setg(errp, | |
1267 | "host lacks kvm mmu notifiers, -mem-path unsupported"); | |
1268 | return NULL; | |
1269 | } | |
1270 | ||
1271 | for (;;) { | |
1272 | fd = open(path, O_RDWR); | |
1273 | if (fd >= 0) { | |
1274 | /* @path names an existing file, use it */ | |
1275 | break; | |
1276 | } | |
1277 | if (errno == ENOENT) { | |
1278 | /* @path names a file that doesn't exist, create it */ | |
1279 | fd = open(path, O_RDWR | O_CREAT | O_EXCL, 0644); | |
1280 | if (fd >= 0) { | |
1281 | unlink_on_error = true; | |
1282 | break; | |
1283 | } | |
1284 | } else if (errno == EISDIR) { | |
1285 | /* @path names a directory, create a file there */ | |
1286 | /* Make name safe to use with mkstemp by replacing '/' with '_'. */ | |
1287 | sanitized_name = g_strdup(memory_region_name(block->mr)); | |
1288 | for (c = sanitized_name; *c != '\0'; c++) { | |
1289 | if (*c == '/') { | |
1290 | *c = '_'; | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path, | |
1295 | sanitized_name); | |
1296 | g_free(sanitized_name); | |
1297 | ||
1298 | fd = mkstemp(filename); | |
1299 | if (fd >= 0) { | |
1300 | unlink(filename); | |
1301 | g_free(filename); | |
1302 | break; | |
1303 | } | |
1304 | g_free(filename); | |
1305 | } | |
1306 | if (errno != EEXIST && errno != EINTR) { | |
1307 | error_setg_errno(errp, errno, | |
1308 | "can't open backing store %s for guest RAM", | |
1309 | path); | |
1310 | goto error; | |
1311 | } | |
1312 | /* | |
1313 | * Try again on EINTR and EEXIST. The latter happens when | |
1314 | * something else creates the file between our two open(). | |
1315 | */ | |
1316 | } | |
1317 | ||
1318 | hpagesize = gethugepagesize(fd); | |
1319 | if (hpagesize < 0) { | |
1320 | error_setg_errno(errp, errno, "can't get page size for %s", | |
1321 | path); | |
1322 | goto error; | |
1323 | } | |
1324 | block->mr->align = hpagesize; | |
1325 | ||
1326 | if (memory < hpagesize) { | |
1327 | error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to " | |
1328 | "or larger than page size 0x%" PRIx64, | |
1329 | memory, hpagesize); | |
1330 | goto error; | |
1331 | } | |
1332 | ||
1333 | memory = ROUND_UP(memory, hpagesize); | |
1334 | ||
1335 | /* | |
1336 | * ftruncate is not supported by hugetlbfs in older | |
1337 | * hosts, so don't bother bailing out on errors. | |
1338 | * If anything goes wrong with it under other filesystems, | |
1339 | * mmap will fail. | |
1340 | */ | |
1341 | if (ftruncate(fd, memory)) { | |
1342 | perror("ftruncate"); | |
1343 | } | |
1344 | ||
1345 | area = qemu_ram_mmap(fd, memory, hpagesize, block->flags & RAM_SHARED); | |
1346 | if (area == MAP_FAILED) { | |
1347 | error_setg_errno(errp, errno, | |
1348 | "unable to map backing store for guest RAM"); | |
1349 | close(fd); | |
1350 | goto error; | |
1351 | } | |
1352 | ||
1353 | if (mem_prealloc) { | |
1354 | os_mem_prealloc(fd, area, memory); | |
1355 | } | |
1356 | ||
1357 | block->fd = fd; | |
1358 | return area; | |
1359 | ||
1360 | error: | |
1361 | if (unlink_on_error) { | |
1362 | unlink(path); | |
1363 | } | |
1364 | close(fd); | |
1365 | return NULL; | |
1366 | } | |
1367 | #endif | |
1368 | ||
1369 | /* Called with the ramlist lock held. */ | |
1370 | static ram_addr_t find_ram_offset(ram_addr_t size) | |
1371 | { | |
1372 | RAMBlock *block, *next_block; | |
1373 | ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX; | |
1374 | ||
1375 | assert(size != 0); /* it would hand out same offset multiple times */ | |
1376 | ||
1377 | if (QLIST_EMPTY_RCU(&ram_list.blocks)) { | |
1378 | return 0; | |
1379 | } | |
1380 | ||
1381 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1382 | ram_addr_t end, next = RAM_ADDR_MAX; | |
1383 | ||
1384 | end = block->offset + block->max_length; | |
1385 | ||
1386 | QLIST_FOREACH_RCU(next_block, &ram_list.blocks, next) { | |
1387 | if (next_block->offset >= end) { | |
1388 | next = MIN(next, next_block->offset); | |
1389 | } | |
1390 | } | |
1391 | if (next - end >= size && next - end < mingap) { | |
1392 | offset = end; | |
1393 | mingap = next - end; | |
1394 | } | |
1395 | } | |
1396 | ||
1397 | if (offset == RAM_ADDR_MAX) { | |
1398 | fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n", | |
1399 | (uint64_t)size); | |
1400 | abort(); | |
1401 | } | |
1402 | ||
1403 | return offset; | |
1404 | } | |
1405 | ||
1406 | ram_addr_t last_ram_offset(void) | |
1407 | { | |
1408 | RAMBlock *block; | |
1409 | ram_addr_t last = 0; | |
1410 | ||
1411 | rcu_read_lock(); | |
1412 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1413 | last = MAX(last, block->offset + block->max_length); | |
1414 | } | |
1415 | rcu_read_unlock(); | |
1416 | return last; | |
1417 | } | |
1418 | ||
1419 | static void qemu_ram_setup_dump(void *addr, ram_addr_t size) | |
1420 | { | |
1421 | int ret; | |
1422 | ||
1423 | /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */ | |
1424 | if (!machine_dump_guest_core(current_machine)) { | |
1425 | ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP); | |
1426 | if (ret) { | |
1427 | perror("qemu_madvise"); | |
1428 | fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, " | |
1429 | "but dump_guest_core=off specified\n"); | |
1430 | } | |
1431 | } | |
1432 | } | |
1433 | ||
1434 | /* Called within an RCU critical section, or while the ramlist lock | |
1435 | * is held. | |
1436 | */ | |
1437 | static RAMBlock *find_ram_block(ram_addr_t addr) | |
1438 | { | |
1439 | RAMBlock *block; | |
1440 | ||
1441 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1442 | if (block->offset == addr) { | |
1443 | return block; | |
1444 | } | |
1445 | } | |
1446 | ||
1447 | return NULL; | |
1448 | } | |
1449 | ||
1450 | const char *qemu_ram_get_idstr(RAMBlock *rb) | |
1451 | { | |
1452 | return rb->idstr; | |
1453 | } | |
1454 | ||
1455 | /* Called with iothread lock held. */ | |
1456 | void qemu_ram_set_idstr(ram_addr_t addr, const char *name, DeviceState *dev) | |
1457 | { | |
1458 | RAMBlock *new_block, *block; | |
1459 | ||
1460 | rcu_read_lock(); | |
1461 | new_block = find_ram_block(addr); | |
1462 | assert(new_block); | |
1463 | assert(!new_block->idstr[0]); | |
1464 | ||
1465 | if (dev) { | |
1466 | char *id = qdev_get_dev_path(dev); | |
1467 | if (id) { | |
1468 | snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id); | |
1469 | g_free(id); | |
1470 | } | |
1471 | } | |
1472 | pstrcat(new_block->idstr, sizeof(new_block->idstr), name); | |
1473 | ||
1474 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1475 | if (block != new_block && !strcmp(block->idstr, new_block->idstr)) { | |
1476 | fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n", | |
1477 | new_block->idstr); | |
1478 | abort(); | |
1479 | } | |
1480 | } | |
1481 | rcu_read_unlock(); | |
1482 | } | |
1483 | ||
1484 | /* Called with iothread lock held. */ | |
1485 | void qemu_ram_unset_idstr(ram_addr_t addr) | |
1486 | { | |
1487 | RAMBlock *block; | |
1488 | ||
1489 | /* FIXME: arch_init.c assumes that this is not called throughout | |
1490 | * migration. Ignore the problem since hot-unplug during migration | |
1491 | * does not work anyway. | |
1492 | */ | |
1493 | ||
1494 | rcu_read_lock(); | |
1495 | block = find_ram_block(addr); | |
1496 | if (block) { | |
1497 | memset(block->idstr, 0, sizeof(block->idstr)); | |
1498 | } | |
1499 | rcu_read_unlock(); | |
1500 | } | |
1501 | ||
1502 | static int memory_try_enable_merging(void *addr, size_t len) | |
1503 | { | |
1504 | if (!machine_mem_merge(current_machine)) { | |
1505 | /* disabled by the user */ | |
1506 | return 0; | |
1507 | } | |
1508 | ||
1509 | return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE); | |
1510 | } | |
1511 | ||
1512 | /* Only legal before guest might have detected the memory size: e.g. on | |
1513 | * incoming migration, or right after reset. | |
1514 | * | |
1515 | * As memory core doesn't know how is memory accessed, it is up to | |
1516 | * resize callback to update device state and/or add assertions to detect | |
1517 | * misuse, if necessary. | |
1518 | */ | |
1519 | int qemu_ram_resize(ram_addr_t base, ram_addr_t newsize, Error **errp) | |
1520 | { | |
1521 | RAMBlock *block = find_ram_block(base); | |
1522 | ||
1523 | assert(block); | |
1524 | ||
1525 | newsize = HOST_PAGE_ALIGN(newsize); | |
1526 | ||
1527 | if (block->used_length == newsize) { | |
1528 | return 0; | |
1529 | } | |
1530 | ||
1531 | if (!(block->flags & RAM_RESIZEABLE)) { | |
1532 | error_setg_errno(errp, EINVAL, | |
1533 | "Length mismatch: %s: 0x" RAM_ADDR_FMT | |
1534 | " in != 0x" RAM_ADDR_FMT, block->idstr, | |
1535 | newsize, block->used_length); | |
1536 | return -EINVAL; | |
1537 | } | |
1538 | ||
1539 | if (block->max_length < newsize) { | |
1540 | error_setg_errno(errp, EINVAL, | |
1541 | "Length too large: %s: 0x" RAM_ADDR_FMT | |
1542 | " > 0x" RAM_ADDR_FMT, block->idstr, | |
1543 | newsize, block->max_length); | |
1544 | return -EINVAL; | |
1545 | } | |
1546 | ||
1547 | cpu_physical_memory_clear_dirty_range(block->offset, block->used_length); | |
1548 | block->used_length = newsize; | |
1549 | cpu_physical_memory_set_dirty_range(block->offset, block->used_length, | |
1550 | DIRTY_CLIENTS_ALL); | |
1551 | memory_region_set_size(block->mr, newsize); | |
1552 | if (block->resized) { | |
1553 | block->resized(block->idstr, newsize, block->host); | |
1554 | } | |
1555 | return 0; | |
1556 | } | |
1557 | ||
1558 | /* Called with ram_list.mutex held */ | |
1559 | static void dirty_memory_extend(ram_addr_t old_ram_size, | |
1560 | ram_addr_t new_ram_size) | |
1561 | { | |
1562 | ram_addr_t old_num_blocks = DIV_ROUND_UP(old_ram_size, | |
1563 | DIRTY_MEMORY_BLOCK_SIZE); | |
1564 | ram_addr_t new_num_blocks = DIV_ROUND_UP(new_ram_size, | |
1565 | DIRTY_MEMORY_BLOCK_SIZE); | |
1566 | int i; | |
1567 | ||
1568 | /* Only need to extend if block count increased */ | |
1569 | if (new_num_blocks <= old_num_blocks) { | |
1570 | return; | |
1571 | } | |
1572 | ||
1573 | for (i = 0; i < DIRTY_MEMORY_NUM; i++) { | |
1574 | DirtyMemoryBlocks *old_blocks; | |
1575 | DirtyMemoryBlocks *new_blocks; | |
1576 | int j; | |
1577 | ||
1578 | old_blocks = atomic_rcu_read(&ram_list.dirty_memory[i]); | |
1579 | new_blocks = g_malloc(sizeof(*new_blocks) + | |
1580 | sizeof(new_blocks->blocks[0]) * new_num_blocks); | |
1581 | ||
1582 | if (old_num_blocks) { | |
1583 | memcpy(new_blocks->blocks, old_blocks->blocks, | |
1584 | old_num_blocks * sizeof(old_blocks->blocks[0])); | |
1585 | } | |
1586 | ||
1587 | for (j = old_num_blocks; j < new_num_blocks; j++) { | |
1588 | new_blocks->blocks[j] = bitmap_new(DIRTY_MEMORY_BLOCK_SIZE); | |
1589 | } | |
1590 | ||
1591 | atomic_rcu_set(&ram_list.dirty_memory[i], new_blocks); | |
1592 | ||
1593 | if (old_blocks) { | |
1594 | g_free_rcu(old_blocks, rcu); | |
1595 | } | |
1596 | } | |
1597 | } | |
1598 | ||
1599 | static void ram_block_add(RAMBlock *new_block, Error **errp) | |
1600 | { | |
1601 | RAMBlock *block; | |
1602 | RAMBlock *last_block = NULL; | |
1603 | ram_addr_t old_ram_size, new_ram_size; | |
1604 | Error *err = NULL; | |
1605 | ||
1606 | old_ram_size = last_ram_offset() >> TARGET_PAGE_BITS; | |
1607 | ||
1608 | qemu_mutex_lock_ramlist(); | |
1609 | new_block->offset = find_ram_offset(new_block->max_length); | |
1610 | ||
1611 | if (!new_block->host) { | |
1612 | if (xen_enabled()) { | |
1613 | xen_ram_alloc(new_block->offset, new_block->max_length, | |
1614 | new_block->mr, &err); | |
1615 | if (err) { | |
1616 | error_propagate(errp, err); | |
1617 | qemu_mutex_unlock_ramlist(); | |
1618 | } | |
1619 | } else { | |
1620 | new_block->host = phys_mem_alloc(new_block->max_length, | |
1621 | &new_block->mr->align); | |
1622 | if (!new_block->host) { | |
1623 | error_setg_errno(errp, errno, | |
1624 | "cannot set up guest memory '%s'", | |
1625 | memory_region_name(new_block->mr)); | |
1626 | qemu_mutex_unlock_ramlist(); | |
1627 | } | |
1628 | memory_try_enable_merging(new_block->host, new_block->max_length); | |
1629 | } | |
1630 | } | |
1631 | ||
1632 | new_ram_size = MAX(old_ram_size, | |
1633 | (new_block->offset + new_block->max_length) >> TARGET_PAGE_BITS); | |
1634 | if (new_ram_size > old_ram_size) { | |
1635 | migration_bitmap_extend(old_ram_size, new_ram_size); | |
1636 | dirty_memory_extend(old_ram_size, new_ram_size); | |
1637 | } | |
1638 | /* Keep the list sorted from biggest to smallest block. Unlike QTAILQ, | |
1639 | * QLIST (which has an RCU-friendly variant) does not have insertion at | |
1640 | * tail, so save the last element in last_block. | |
1641 | */ | |
1642 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1643 | last_block = block; | |
1644 | if (block->max_length < new_block->max_length) { | |
1645 | break; | |
1646 | } | |
1647 | } | |
1648 | if (block) { | |
1649 | QLIST_INSERT_BEFORE_RCU(block, new_block, next); | |
1650 | } else if (last_block) { | |
1651 | QLIST_INSERT_AFTER_RCU(last_block, new_block, next); | |
1652 | } else { /* list is empty */ | |
1653 | QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next); | |
1654 | } | |
1655 | ram_list.mru_block = NULL; | |
1656 | ||
1657 | /* Write list before version */ | |
1658 | smp_wmb(); | |
1659 | ram_list.version++; | |
1660 | qemu_mutex_unlock_ramlist(); | |
1661 | ||
1662 | cpu_physical_memory_set_dirty_range(new_block->offset, | |
1663 | new_block->used_length, | |
1664 | DIRTY_CLIENTS_ALL); | |
1665 | ||
1666 | if (new_block->host) { | |
1667 | qemu_ram_setup_dump(new_block->host, new_block->max_length); | |
1668 | qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_HUGEPAGE); | |
1669 | qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_DONTFORK); | |
1670 | if (kvm_enabled()) { | |
1671 | kvm_setup_guest_memory(new_block->host, new_block->max_length); | |
1672 | } | |
1673 | } | |
1674 | } | |
1675 | ||
1676 | #ifdef __linux__ | |
1677 | RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr, | |
1678 | bool share, const char *mem_path, | |
1679 | Error **errp) | |
1680 | { | |
1681 | RAMBlock *new_block; | |
1682 | Error *local_err = NULL; | |
1683 | ||
1684 | if (xen_enabled()) { | |
1685 | error_setg(errp, "-mem-path not supported with Xen"); | |
1686 | return NULL; | |
1687 | } | |
1688 | ||
1689 | if (phys_mem_alloc != qemu_anon_ram_alloc) { | |
1690 | /* | |
1691 | * file_ram_alloc() needs to allocate just like | |
1692 | * phys_mem_alloc, but we haven't bothered to provide | |
1693 | * a hook there. | |
1694 | */ | |
1695 | error_setg(errp, | |
1696 | "-mem-path not supported with this accelerator"); | |
1697 | return NULL; | |
1698 | } | |
1699 | ||
1700 | size = HOST_PAGE_ALIGN(size); | |
1701 | new_block = g_malloc0(sizeof(*new_block)); | |
1702 | new_block->mr = mr; | |
1703 | new_block->used_length = size; | |
1704 | new_block->max_length = size; | |
1705 | new_block->flags = share ? RAM_SHARED : 0; | |
1706 | new_block->host = file_ram_alloc(new_block, size, | |
1707 | mem_path, errp); | |
1708 | if (!new_block->host) { | |
1709 | g_free(new_block); | |
1710 | return NULL; | |
1711 | } | |
1712 | ||
1713 | ram_block_add(new_block, &local_err); | |
1714 | if (local_err) { | |
1715 | g_free(new_block); | |
1716 | error_propagate(errp, local_err); | |
1717 | return NULL; | |
1718 | } | |
1719 | return new_block; | |
1720 | } | |
1721 | #endif | |
1722 | ||
1723 | static | |
1724 | RAMBlock *qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size, | |
1725 | void (*resized)(const char*, | |
1726 | uint64_t length, | |
1727 | void *host), | |
1728 | void *host, bool resizeable, | |
1729 | MemoryRegion *mr, Error **errp) | |
1730 | { | |
1731 | RAMBlock *new_block; | |
1732 | Error *local_err = NULL; | |
1733 | ||
1734 | size = HOST_PAGE_ALIGN(size); | |
1735 | max_size = HOST_PAGE_ALIGN(max_size); | |
1736 | new_block = g_malloc0(sizeof(*new_block)); | |
1737 | new_block->mr = mr; | |
1738 | new_block->resized = resized; | |
1739 | new_block->used_length = size; | |
1740 | new_block->max_length = max_size; | |
1741 | assert(max_size >= size); | |
1742 | new_block->fd = -1; | |
1743 | new_block->host = host; | |
1744 | if (host) { | |
1745 | new_block->flags |= RAM_PREALLOC; | |
1746 | } | |
1747 | if (resizeable) { | |
1748 | new_block->flags |= RAM_RESIZEABLE; | |
1749 | } | |
1750 | ram_block_add(new_block, &local_err); | |
1751 | if (local_err) { | |
1752 | g_free(new_block); | |
1753 | error_propagate(errp, local_err); | |
1754 | return NULL; | |
1755 | } | |
1756 | return new_block; | |
1757 | } | |
1758 | ||
1759 | RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host, | |
1760 | MemoryRegion *mr, Error **errp) | |
1761 | { | |
1762 | return qemu_ram_alloc_internal(size, size, NULL, host, false, mr, errp); | |
1763 | } | |
1764 | ||
1765 | RAMBlock *qemu_ram_alloc(ram_addr_t size, MemoryRegion *mr, Error **errp) | |
1766 | { | |
1767 | return qemu_ram_alloc_internal(size, size, NULL, NULL, false, mr, errp); | |
1768 | } | |
1769 | ||
1770 | RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t maxsz, | |
1771 | void (*resized)(const char*, | |
1772 | uint64_t length, | |
1773 | void *host), | |
1774 | MemoryRegion *mr, Error **errp) | |
1775 | { | |
1776 | return qemu_ram_alloc_internal(size, maxsz, resized, NULL, true, mr, errp); | |
1777 | } | |
1778 | ||
1779 | static void reclaim_ramblock(RAMBlock *block) | |
1780 | { | |
1781 | if (block->flags & RAM_PREALLOC) { | |
1782 | ; | |
1783 | } else if (xen_enabled()) { | |
1784 | xen_invalidate_map_cache_entry(block->host); | |
1785 | #ifndef _WIN32 | |
1786 | } else if (block->fd >= 0) { | |
1787 | qemu_ram_munmap(block->host, block->max_length); | |
1788 | close(block->fd); | |
1789 | #endif | |
1790 | } else { | |
1791 | qemu_anon_ram_free(block->host, block->max_length); | |
1792 | } | |
1793 | g_free(block); | |
1794 | } | |
1795 | ||
1796 | void qemu_ram_free(RAMBlock *block) | |
1797 | { | |
1798 | qemu_mutex_lock_ramlist(); | |
1799 | QLIST_REMOVE_RCU(block, next); | |
1800 | ram_list.mru_block = NULL; | |
1801 | /* Write list before version */ | |
1802 | smp_wmb(); | |
1803 | ram_list.version++; | |
1804 | call_rcu(block, reclaim_ramblock, rcu); | |
1805 | qemu_mutex_unlock_ramlist(); | |
1806 | } | |
1807 | ||
1808 | #ifndef _WIN32 | |
1809 | void qemu_ram_remap(ram_addr_t addr, ram_addr_t length) | |
1810 | { | |
1811 | RAMBlock *block; | |
1812 | ram_addr_t offset; | |
1813 | int flags; | |
1814 | void *area, *vaddr; | |
1815 | ||
1816 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1817 | offset = addr - block->offset; | |
1818 | if (offset < block->max_length) { | |
1819 | vaddr = ramblock_ptr(block, offset); | |
1820 | if (block->flags & RAM_PREALLOC) { | |
1821 | ; | |
1822 | } else if (xen_enabled()) { | |
1823 | abort(); | |
1824 | } else { | |
1825 | flags = MAP_FIXED; | |
1826 | if (block->fd >= 0) { | |
1827 | flags |= (block->flags & RAM_SHARED ? | |
1828 | MAP_SHARED : MAP_PRIVATE); | |
1829 | area = mmap(vaddr, length, PROT_READ | PROT_WRITE, | |
1830 | flags, block->fd, offset); | |
1831 | } else { | |
1832 | /* | |
1833 | * Remap needs to match alloc. Accelerators that | |
1834 | * set phys_mem_alloc never remap. If they did, | |
1835 | * we'd need a remap hook here. | |
1836 | */ | |
1837 | assert(phys_mem_alloc == qemu_anon_ram_alloc); | |
1838 | ||
1839 | flags |= MAP_PRIVATE | MAP_ANONYMOUS; | |
1840 | area = mmap(vaddr, length, PROT_READ | PROT_WRITE, | |
1841 | flags, -1, 0); | |
1842 | } | |
1843 | if (area != vaddr) { | |
1844 | fprintf(stderr, "Could not remap addr: " | |
1845 | RAM_ADDR_FMT "@" RAM_ADDR_FMT "\n", | |
1846 | length, addr); | |
1847 | exit(1); | |
1848 | } | |
1849 | memory_try_enable_merging(vaddr, length); | |
1850 | qemu_ram_setup_dump(vaddr, length); | |
1851 | } | |
1852 | } | |
1853 | } | |
1854 | } | |
1855 | #endif /* !_WIN32 */ | |
1856 | ||
1857 | int qemu_get_ram_fd(ram_addr_t addr) | |
1858 | { | |
1859 | RAMBlock *block; | |
1860 | int fd; | |
1861 | ||
1862 | rcu_read_lock(); | |
1863 | block = qemu_get_ram_block(addr); | |
1864 | fd = block->fd; | |
1865 | rcu_read_unlock(); | |
1866 | return fd; | |
1867 | } | |
1868 | ||
1869 | void qemu_set_ram_fd(ram_addr_t addr, int fd) | |
1870 | { | |
1871 | RAMBlock *block; | |
1872 | ||
1873 | rcu_read_lock(); | |
1874 | block = qemu_get_ram_block(addr); | |
1875 | block->fd = fd; | |
1876 | rcu_read_unlock(); | |
1877 | } | |
1878 | ||
1879 | void *qemu_get_ram_block_host_ptr(ram_addr_t addr) | |
1880 | { | |
1881 | RAMBlock *block; | |
1882 | void *ptr; | |
1883 | ||
1884 | rcu_read_lock(); | |
1885 | block = qemu_get_ram_block(addr); | |
1886 | ptr = ramblock_ptr(block, 0); | |
1887 | rcu_read_unlock(); | |
1888 | return ptr; | |
1889 | } | |
1890 | ||
1891 | /* Return a host pointer to ram allocated with qemu_ram_alloc. | |
1892 | * This should not be used for general purpose DMA. Use address_space_map | |
1893 | * or address_space_rw instead. For local memory (e.g. video ram) that the | |
1894 | * device owns, use memory_region_get_ram_ptr. | |
1895 | * | |
1896 | * Called within RCU critical section. | |
1897 | */ | |
1898 | void *qemu_get_ram_ptr(RAMBlock *ram_block, ram_addr_t addr) | |
1899 | { | |
1900 | RAMBlock *block = ram_block; | |
1901 | ||
1902 | if (block == NULL) { | |
1903 | block = qemu_get_ram_block(addr); | |
1904 | } | |
1905 | ||
1906 | if (xen_enabled() && block->host == NULL) { | |
1907 | /* We need to check if the requested address is in the RAM | |
1908 | * because we don't want to map the entire memory in QEMU. | |
1909 | * In that case just map until the end of the page. | |
1910 | */ | |
1911 | if (block->offset == 0) { | |
1912 | return xen_map_cache(addr, 0, 0); | |
1913 | } | |
1914 | ||
1915 | block->host = xen_map_cache(block->offset, block->max_length, 1); | |
1916 | } | |
1917 | return ramblock_ptr(block, addr - block->offset); | |
1918 | } | |
1919 | ||
1920 | /* Return a host pointer to guest's ram. Similar to qemu_get_ram_ptr | |
1921 | * but takes a size argument. | |
1922 | * | |
1923 | * Called within RCU critical section. | |
1924 | */ | |
1925 | static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr, | |
1926 | hwaddr *size) | |
1927 | { | |
1928 | RAMBlock *block = ram_block; | |
1929 | ram_addr_t offset_inside_block; | |
1930 | if (*size == 0) { | |
1931 | return NULL; | |
1932 | } | |
1933 | ||
1934 | if (block == NULL) { | |
1935 | block = qemu_get_ram_block(addr); | |
1936 | } | |
1937 | offset_inside_block = addr - block->offset; | |
1938 | *size = MIN(*size, block->max_length - offset_inside_block); | |
1939 | ||
1940 | if (xen_enabled() && block->host == NULL) { | |
1941 | /* We need to check if the requested address is in the RAM | |
1942 | * because we don't want to map the entire memory in QEMU. | |
1943 | * In that case just map the requested area. | |
1944 | */ | |
1945 | if (block->offset == 0) { | |
1946 | return xen_map_cache(addr, *size, 1); | |
1947 | } | |
1948 | ||
1949 | block->host = xen_map_cache(block->offset, block->max_length, 1); | |
1950 | } | |
1951 | ||
1952 | return ramblock_ptr(block, offset_inside_block); | |
1953 | } | |
1954 | ||
1955 | /* | |
1956 | * Translates a host ptr back to a RAMBlock, a ram_addr and an offset | |
1957 | * in that RAMBlock. | |
1958 | * | |
1959 | * ptr: Host pointer to look up | |
1960 | * round_offset: If true round the result offset down to a page boundary | |
1961 | * *ram_addr: set to result ram_addr | |
1962 | * *offset: set to result offset within the RAMBlock | |
1963 | * | |
1964 | * Returns: RAMBlock (or NULL if not found) | |
1965 | * | |
1966 | * By the time this function returns, the returned pointer is not protected | |
1967 | * by RCU anymore. If the caller is not within an RCU critical section and | |
1968 | * does not hold the iothread lock, it must have other means of protecting the | |
1969 | * pointer, such as a reference to the region that includes the incoming | |
1970 | * ram_addr_t. | |
1971 | */ | |
1972 | RAMBlock *qemu_ram_block_from_host(void *ptr, bool round_offset, | |
1973 | ram_addr_t *ram_addr, | |
1974 | ram_addr_t *offset) | |
1975 | { | |
1976 | RAMBlock *block; | |
1977 | uint8_t *host = ptr; | |
1978 | ||
1979 | if (xen_enabled()) { | |
1980 | rcu_read_lock(); | |
1981 | *ram_addr = xen_ram_addr_from_mapcache(ptr); | |
1982 | block = qemu_get_ram_block(*ram_addr); | |
1983 | if (block) { | |
1984 | *offset = (host - block->host); | |
1985 | } | |
1986 | rcu_read_unlock(); | |
1987 | return block; | |
1988 | } | |
1989 | ||
1990 | rcu_read_lock(); | |
1991 | block = atomic_rcu_read(&ram_list.mru_block); | |
1992 | if (block && block->host && host - block->host < block->max_length) { | |
1993 | goto found; | |
1994 | } | |
1995 | ||
1996 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
1997 | /* This case append when the block is not mapped. */ | |
1998 | if (block->host == NULL) { | |
1999 | continue; | |
2000 | } | |
2001 | if (host - block->host < block->max_length) { | |
2002 | goto found; | |
2003 | } | |
2004 | } | |
2005 | ||
2006 | rcu_read_unlock(); | |
2007 | return NULL; | |
2008 | ||
2009 | found: | |
2010 | *offset = (host - block->host); | |
2011 | if (round_offset) { | |
2012 | *offset &= TARGET_PAGE_MASK; | |
2013 | } | |
2014 | *ram_addr = block->offset + *offset; | |
2015 | rcu_read_unlock(); | |
2016 | return block; | |
2017 | } | |
2018 | ||
2019 | /* | |
2020 | * Finds the named RAMBlock | |
2021 | * | |
2022 | * name: The name of RAMBlock to find | |
2023 | * | |
2024 | * Returns: RAMBlock (or NULL if not found) | |
2025 | */ | |
2026 | RAMBlock *qemu_ram_block_by_name(const char *name) | |
2027 | { | |
2028 | RAMBlock *block; | |
2029 | ||
2030 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
2031 | if (!strcmp(name, block->idstr)) { | |
2032 | return block; | |
2033 | } | |
2034 | } | |
2035 | ||
2036 | return NULL; | |
2037 | } | |
2038 | ||
2039 | /* Some of the softmmu routines need to translate from a host pointer | |
2040 | (typically a TLB entry) back to a ram offset. */ | |
2041 | MemoryRegion *qemu_ram_addr_from_host(void *ptr, ram_addr_t *ram_addr) | |
2042 | { | |
2043 | RAMBlock *block; | |
2044 | ram_addr_t offset; /* Not used */ | |
2045 | ||
2046 | block = qemu_ram_block_from_host(ptr, false, ram_addr, &offset); | |
2047 | ||
2048 | if (!block) { | |
2049 | return NULL; | |
2050 | } | |
2051 | ||
2052 | return block->mr; | |
2053 | } | |
2054 | ||
2055 | /* Called within RCU critical section. */ | |
2056 | static void notdirty_mem_write(void *opaque, hwaddr ram_addr, | |
2057 | uint64_t val, unsigned size) | |
2058 | { | |
2059 | if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) { | |
2060 | tb_invalidate_phys_page_fast(ram_addr, size); | |
2061 | } | |
2062 | switch (size) { | |
2063 | case 1: | |
2064 | stb_p(qemu_get_ram_ptr(NULL, ram_addr), val); | |
2065 | break; | |
2066 | case 2: | |
2067 | stw_p(qemu_get_ram_ptr(NULL, ram_addr), val); | |
2068 | break; | |
2069 | case 4: | |
2070 | stl_p(qemu_get_ram_ptr(NULL, ram_addr), val); | |
2071 | break; | |
2072 | default: | |
2073 | abort(); | |
2074 | } | |
2075 | /* Set both VGA and migration bits for simplicity and to remove | |
2076 | * the notdirty callback faster. | |
2077 | */ | |
2078 | cpu_physical_memory_set_dirty_range(ram_addr, size, | |
2079 | DIRTY_CLIENTS_NOCODE); | |
2080 | /* we remove the notdirty callback only if the code has been | |
2081 | flushed */ | |
2082 | if (!cpu_physical_memory_is_clean(ram_addr)) { | |
2083 | tlb_set_dirty(current_cpu, current_cpu->mem_io_vaddr); | |
2084 | } | |
2085 | } | |
2086 | ||
2087 | static bool notdirty_mem_accepts(void *opaque, hwaddr addr, | |
2088 | unsigned size, bool is_write) | |
2089 | { | |
2090 | return is_write; | |
2091 | } | |
2092 | ||
2093 | static const MemoryRegionOps notdirty_mem_ops = { | |
2094 | .write = notdirty_mem_write, | |
2095 | .valid.accepts = notdirty_mem_accepts, | |
2096 | .endianness = DEVICE_NATIVE_ENDIAN, | |
2097 | }; | |
2098 | ||
2099 | /* Generate a debug exception if a watchpoint has been hit. */ | |
2100 | static void check_watchpoint(int offset, int len, MemTxAttrs attrs, int flags) | |
2101 | { | |
2102 | CPUState *cpu = current_cpu; | |
2103 | CPUClass *cc = CPU_GET_CLASS(cpu); | |
2104 | CPUArchState *env = cpu->env_ptr; | |
2105 | target_ulong pc, cs_base; | |
2106 | target_ulong vaddr; | |
2107 | CPUWatchpoint *wp; | |
2108 | int cpu_flags; | |
2109 | ||
2110 | if (cpu->watchpoint_hit) { | |
2111 | /* We re-entered the check after replacing the TB. Now raise | |
2112 | * the debug interrupt so that is will trigger after the | |
2113 | * current instruction. */ | |
2114 | cpu_interrupt(cpu, CPU_INTERRUPT_DEBUG); | |
2115 | return; | |
2116 | } | |
2117 | vaddr = (cpu->mem_io_vaddr & TARGET_PAGE_MASK) + offset; | |
2118 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { | |
2119 | if (cpu_watchpoint_address_matches(wp, vaddr, len) | |
2120 | && (wp->flags & flags)) { | |
2121 | if (flags == BP_MEM_READ) { | |
2122 | wp->flags |= BP_WATCHPOINT_HIT_READ; | |
2123 | } else { | |
2124 | wp->flags |= BP_WATCHPOINT_HIT_WRITE; | |
2125 | } | |
2126 | wp->hitaddr = vaddr; | |
2127 | wp->hitattrs = attrs; | |
2128 | if (!cpu->watchpoint_hit) { | |
2129 | if (wp->flags & BP_CPU && | |
2130 | !cc->debug_check_watchpoint(cpu, wp)) { | |
2131 | wp->flags &= ~BP_WATCHPOINT_HIT; | |
2132 | continue; | |
2133 | } | |
2134 | cpu->watchpoint_hit = wp; | |
2135 | tb_check_watchpoint(cpu); | |
2136 | if (wp->flags & BP_STOP_BEFORE_ACCESS) { | |
2137 | cpu->exception_index = EXCP_DEBUG; | |
2138 | cpu_loop_exit(cpu); | |
2139 | } else { | |
2140 | cpu_get_tb_cpu_state(env, &pc, &cs_base, &cpu_flags); | |
2141 | tb_gen_code(cpu, pc, cs_base, cpu_flags, 1); | |
2142 | cpu_resume_from_signal(cpu, NULL); | |
2143 | } | |
2144 | } | |
2145 | } else { | |
2146 | wp->flags &= ~BP_WATCHPOINT_HIT; | |
2147 | } | |
2148 | } | |
2149 | } | |
2150 | ||
2151 | /* Watchpoint access routines. Watchpoints are inserted using TLB tricks, | |
2152 | so these check for a hit then pass through to the normal out-of-line | |
2153 | phys routines. */ | |
2154 | static MemTxResult watch_mem_read(void *opaque, hwaddr addr, uint64_t *pdata, | |
2155 | unsigned size, MemTxAttrs attrs) | |
2156 | { | |
2157 | MemTxResult res; | |
2158 | uint64_t data; | |
2159 | int asidx = cpu_asidx_from_attrs(current_cpu, attrs); | |
2160 | AddressSpace *as = current_cpu->cpu_ases[asidx].as; | |
2161 | ||
2162 | check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_READ); | |
2163 | switch (size) { | |
2164 | case 1: | |
2165 | data = address_space_ldub(as, addr, attrs, &res); | |
2166 | break; | |
2167 | case 2: | |
2168 | data = address_space_lduw(as, addr, attrs, &res); | |
2169 | break; | |
2170 | case 4: | |
2171 | data = address_space_ldl(as, addr, attrs, &res); | |
2172 | break; | |
2173 | default: abort(); | |
2174 | } | |
2175 | *pdata = data; | |
2176 | return res; | |
2177 | } | |
2178 | ||
2179 | static MemTxResult watch_mem_write(void *opaque, hwaddr addr, | |
2180 | uint64_t val, unsigned size, | |
2181 | MemTxAttrs attrs) | |
2182 | { | |
2183 | MemTxResult res; | |
2184 | int asidx = cpu_asidx_from_attrs(current_cpu, attrs); | |
2185 | AddressSpace *as = current_cpu->cpu_ases[asidx].as; | |
2186 | ||
2187 | check_watchpoint(addr & ~TARGET_PAGE_MASK, size, attrs, BP_MEM_WRITE); | |
2188 | switch (size) { | |
2189 | case 1: | |
2190 | address_space_stb(as, addr, val, attrs, &res); | |
2191 | break; | |
2192 | case 2: | |
2193 | address_space_stw(as, addr, val, attrs, &res); | |
2194 | break; | |
2195 | case 4: | |
2196 | address_space_stl(as, addr, val, attrs, &res); | |
2197 | break; | |
2198 | default: abort(); | |
2199 | } | |
2200 | return res; | |
2201 | } | |
2202 | ||
2203 | static const MemoryRegionOps watch_mem_ops = { | |
2204 | .read_with_attrs = watch_mem_read, | |
2205 | .write_with_attrs = watch_mem_write, | |
2206 | .endianness = DEVICE_NATIVE_ENDIAN, | |
2207 | }; | |
2208 | ||
2209 | static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data, | |
2210 | unsigned len, MemTxAttrs attrs) | |
2211 | { | |
2212 | subpage_t *subpage = opaque; | |
2213 | uint8_t buf[8]; | |
2214 | MemTxResult res; | |
2215 | ||
2216 | #if defined(DEBUG_SUBPAGE) | |
2217 | printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__, | |
2218 | subpage, len, addr); | |
2219 | #endif | |
2220 | res = address_space_read(subpage->as, addr + subpage->base, | |
2221 | attrs, buf, len); | |
2222 | if (res) { | |
2223 | return res; | |
2224 | } | |
2225 | switch (len) { | |
2226 | case 1: | |
2227 | *data = ldub_p(buf); | |
2228 | return MEMTX_OK; | |
2229 | case 2: | |
2230 | *data = lduw_p(buf); | |
2231 | return MEMTX_OK; | |
2232 | case 4: | |
2233 | *data = ldl_p(buf); | |
2234 | return MEMTX_OK; | |
2235 | case 8: | |
2236 | *data = ldq_p(buf); | |
2237 | return MEMTX_OK; | |
2238 | default: | |
2239 | abort(); | |
2240 | } | |
2241 | } | |
2242 | ||
2243 | static MemTxResult subpage_write(void *opaque, hwaddr addr, | |
2244 | uint64_t value, unsigned len, MemTxAttrs attrs) | |
2245 | { | |
2246 | subpage_t *subpage = opaque; | |
2247 | uint8_t buf[8]; | |
2248 | ||
2249 | #if defined(DEBUG_SUBPAGE) | |
2250 | printf("%s: subpage %p len %u addr " TARGET_FMT_plx | |
2251 | " value %"PRIx64"\n", | |
2252 | __func__, subpage, len, addr, value); | |
2253 | #endif | |
2254 | switch (len) { | |
2255 | case 1: | |
2256 | stb_p(buf, value); | |
2257 | break; | |
2258 | case 2: | |
2259 | stw_p(buf, value); | |
2260 | break; | |
2261 | case 4: | |
2262 | stl_p(buf, value); | |
2263 | break; | |
2264 | case 8: | |
2265 | stq_p(buf, value); | |
2266 | break; | |
2267 | default: | |
2268 | abort(); | |
2269 | } | |
2270 | return address_space_write(subpage->as, addr + subpage->base, | |
2271 | attrs, buf, len); | |
2272 | } | |
2273 | ||
2274 | static bool subpage_accepts(void *opaque, hwaddr addr, | |
2275 | unsigned len, bool is_write) | |
2276 | { | |
2277 | subpage_t *subpage = opaque; | |
2278 | #if defined(DEBUG_SUBPAGE) | |
2279 | printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n", | |
2280 | __func__, subpage, is_write ? 'w' : 'r', len, addr); | |
2281 | #endif | |
2282 | ||
2283 | return address_space_access_valid(subpage->as, addr + subpage->base, | |
2284 | len, is_write); | |
2285 | } | |
2286 | ||
2287 | static const MemoryRegionOps subpage_ops = { | |
2288 | .read_with_attrs = subpage_read, | |
2289 | .write_with_attrs = subpage_write, | |
2290 | .impl.min_access_size = 1, | |
2291 | .impl.max_access_size = 8, | |
2292 | .valid.min_access_size = 1, | |
2293 | .valid.max_access_size = 8, | |
2294 | .valid.accepts = subpage_accepts, | |
2295 | .endianness = DEVICE_NATIVE_ENDIAN, | |
2296 | }; | |
2297 | ||
2298 | static int subpage_register (subpage_t *mmio, uint32_t start, uint32_t end, | |
2299 | uint16_t section) | |
2300 | { | |
2301 | int idx, eidx; | |
2302 | ||
2303 | if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE) | |
2304 | return -1; | |
2305 | idx = SUBPAGE_IDX(start); | |
2306 | eidx = SUBPAGE_IDX(end); | |
2307 | #if defined(DEBUG_SUBPAGE) | |
2308 | printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n", | |
2309 | __func__, mmio, start, end, idx, eidx, section); | |
2310 | #endif | |
2311 | for (; idx <= eidx; idx++) { | |
2312 | mmio->sub_section[idx] = section; | |
2313 | } | |
2314 | ||
2315 | return 0; | |
2316 | } | |
2317 | ||
2318 | static subpage_t *subpage_init(AddressSpace *as, hwaddr base) | |
2319 | { | |
2320 | subpage_t *mmio; | |
2321 | ||
2322 | mmio = g_malloc0(sizeof(subpage_t)); | |
2323 | ||
2324 | mmio->as = as; | |
2325 | mmio->base = base; | |
2326 | memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio, | |
2327 | NULL, TARGET_PAGE_SIZE); | |
2328 | mmio->iomem.subpage = true; | |
2329 | #if defined(DEBUG_SUBPAGE) | |
2330 | printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__, | |
2331 | mmio, base, TARGET_PAGE_SIZE); | |
2332 | #endif | |
2333 | subpage_register(mmio, 0, TARGET_PAGE_SIZE-1, PHYS_SECTION_UNASSIGNED); | |
2334 | ||
2335 | return mmio; | |
2336 | } | |
2337 | ||
2338 | static uint16_t dummy_section(PhysPageMap *map, AddressSpace *as, | |
2339 | MemoryRegion *mr) | |
2340 | { | |
2341 | assert(as); | |
2342 | MemoryRegionSection section = { | |
2343 | .address_space = as, | |
2344 | .mr = mr, | |
2345 | .offset_within_address_space = 0, | |
2346 | .offset_within_region = 0, | |
2347 | .size = int128_2_64(), | |
2348 | }; | |
2349 | ||
2350 | return phys_section_add(map, §ion); | |
2351 | } | |
2352 | ||
2353 | MemoryRegion *iotlb_to_region(CPUState *cpu, hwaddr index, MemTxAttrs attrs) | |
2354 | { | |
2355 | int asidx = cpu_asidx_from_attrs(cpu, attrs); | |
2356 | CPUAddressSpace *cpuas = &cpu->cpu_ases[asidx]; | |
2357 | AddressSpaceDispatch *d = atomic_rcu_read(&cpuas->memory_dispatch); | |
2358 | MemoryRegionSection *sections = d->map.sections; | |
2359 | ||
2360 | return sections[index & ~TARGET_PAGE_MASK].mr; | |
2361 | } | |
2362 | ||
2363 | static void io_mem_init(void) | |
2364 | { | |
2365 | memory_region_init_io(&io_mem_rom, NULL, &unassigned_mem_ops, NULL, NULL, UINT64_MAX); | |
2366 | memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL, | |
2367 | NULL, UINT64_MAX); | |
2368 | memory_region_init_io(&io_mem_notdirty, NULL, ¬dirty_mem_ops, NULL, | |
2369 | NULL, UINT64_MAX); | |
2370 | memory_region_init_io(&io_mem_watch, NULL, &watch_mem_ops, NULL, | |
2371 | NULL, UINT64_MAX); | |
2372 | } | |
2373 | ||
2374 | static void mem_begin(MemoryListener *listener) | |
2375 | { | |
2376 | AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener); | |
2377 | AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1); | |
2378 | uint16_t n; | |
2379 | ||
2380 | n = dummy_section(&d->map, as, &io_mem_unassigned); | |
2381 | assert(n == PHYS_SECTION_UNASSIGNED); | |
2382 | n = dummy_section(&d->map, as, &io_mem_notdirty); | |
2383 | assert(n == PHYS_SECTION_NOTDIRTY); | |
2384 | n = dummy_section(&d->map, as, &io_mem_rom); | |
2385 | assert(n == PHYS_SECTION_ROM); | |
2386 | n = dummy_section(&d->map, as, &io_mem_watch); | |
2387 | assert(n == PHYS_SECTION_WATCH); | |
2388 | ||
2389 | d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 }; | |
2390 | d->as = as; | |
2391 | as->next_dispatch = d; | |
2392 | } | |
2393 | ||
2394 | static void address_space_dispatch_free(AddressSpaceDispatch *d) | |
2395 | { | |
2396 | phys_sections_free(&d->map); | |
2397 | g_free(d); | |
2398 | } | |
2399 | ||
2400 | static void mem_commit(MemoryListener *listener) | |
2401 | { | |
2402 | AddressSpace *as = container_of(listener, AddressSpace, dispatch_listener); | |
2403 | AddressSpaceDispatch *cur = as->dispatch; | |
2404 | AddressSpaceDispatch *next = as->next_dispatch; | |
2405 | ||
2406 | phys_page_compact_all(next, next->map.nodes_nb); | |
2407 | ||
2408 | atomic_rcu_set(&as->dispatch, next); | |
2409 | if (cur) { | |
2410 | call_rcu(cur, address_space_dispatch_free, rcu); | |
2411 | } | |
2412 | } | |
2413 | ||
2414 | static void tcg_commit(MemoryListener *listener) | |
2415 | { | |
2416 | CPUAddressSpace *cpuas; | |
2417 | AddressSpaceDispatch *d; | |
2418 | ||
2419 | /* since each CPU stores ram addresses in its TLB cache, we must | |
2420 | reset the modified entries */ | |
2421 | cpuas = container_of(listener, CPUAddressSpace, tcg_as_listener); | |
2422 | cpu_reloading_memory_map(); | |
2423 | /* The CPU and TLB are protected by the iothread lock. | |
2424 | * We reload the dispatch pointer now because cpu_reloading_memory_map() | |
2425 | * may have split the RCU critical section. | |
2426 | */ | |
2427 | d = atomic_rcu_read(&cpuas->as->dispatch); | |
2428 | cpuas->memory_dispatch = d; | |
2429 | tlb_flush(cpuas->cpu, 1); | |
2430 | } | |
2431 | ||
2432 | void address_space_init_dispatch(AddressSpace *as) | |
2433 | { | |
2434 | as->dispatch = NULL; | |
2435 | as->dispatch_listener = (MemoryListener) { | |
2436 | .begin = mem_begin, | |
2437 | .commit = mem_commit, | |
2438 | .region_add = mem_add, | |
2439 | .region_nop = mem_add, | |
2440 | .priority = 0, | |
2441 | }; | |
2442 | memory_listener_register(&as->dispatch_listener, as); | |
2443 | } | |
2444 | ||
2445 | void address_space_unregister(AddressSpace *as) | |
2446 | { | |
2447 | memory_listener_unregister(&as->dispatch_listener); | |
2448 | } | |
2449 | ||
2450 | void address_space_destroy_dispatch(AddressSpace *as) | |
2451 | { | |
2452 | AddressSpaceDispatch *d = as->dispatch; | |
2453 | ||
2454 | atomic_rcu_set(&as->dispatch, NULL); | |
2455 | if (d) { | |
2456 | call_rcu(d, address_space_dispatch_free, rcu); | |
2457 | } | |
2458 | } | |
2459 | ||
2460 | static void memory_map_init(void) | |
2461 | { | |
2462 | system_memory = g_malloc(sizeof(*system_memory)); | |
2463 | ||
2464 | memory_region_init(system_memory, NULL, "system", UINT64_MAX); | |
2465 | address_space_init(&address_space_memory, system_memory, "memory"); | |
2466 | ||
2467 | system_io = g_malloc(sizeof(*system_io)); | |
2468 | memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io", | |
2469 | 65536); | |
2470 | address_space_init(&address_space_io, system_io, "I/O"); | |
2471 | } | |
2472 | ||
2473 | MemoryRegion *get_system_memory(void) | |
2474 | { | |
2475 | return system_memory; | |
2476 | } | |
2477 | ||
2478 | MemoryRegion *get_system_io(void) | |
2479 | { | |
2480 | return system_io; | |
2481 | } | |
2482 | ||
2483 | #endif /* !defined(CONFIG_USER_ONLY) */ | |
2484 | ||
2485 | /* physical memory access (slow version, mainly for debug) */ | |
2486 | #if defined(CONFIG_USER_ONLY) | |
2487 | int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr, | |
2488 | uint8_t *buf, int len, int is_write) | |
2489 | { | |
2490 | int l, flags; | |
2491 | target_ulong page; | |
2492 | void * p; | |
2493 | ||
2494 | while (len > 0) { | |
2495 | page = addr & TARGET_PAGE_MASK; | |
2496 | l = (page + TARGET_PAGE_SIZE) - addr; | |
2497 | if (l > len) | |
2498 | l = len; | |
2499 | flags = page_get_flags(page); | |
2500 | if (!(flags & PAGE_VALID)) | |
2501 | return -1; | |
2502 | if (is_write) { | |
2503 | if (!(flags & PAGE_WRITE)) | |
2504 | return -1; | |
2505 | /* XXX: this code should not depend on lock_user */ | |
2506 | if (!(p = lock_user(VERIFY_WRITE, addr, l, 0))) | |
2507 | return -1; | |
2508 | memcpy(p, buf, l); | |
2509 | unlock_user(p, addr, l); | |
2510 | } else { | |
2511 | if (!(flags & PAGE_READ)) | |
2512 | return -1; | |
2513 | /* XXX: this code should not depend on lock_user */ | |
2514 | if (!(p = lock_user(VERIFY_READ, addr, l, 1))) | |
2515 | return -1; | |
2516 | memcpy(buf, p, l); | |
2517 | unlock_user(p, addr, 0); | |
2518 | } | |
2519 | len -= l; | |
2520 | buf += l; | |
2521 | addr += l; | |
2522 | } | |
2523 | return 0; | |
2524 | } | |
2525 | ||
2526 | #else | |
2527 | ||
2528 | static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr, | |
2529 | hwaddr length) | |
2530 | { | |
2531 | uint8_t dirty_log_mask = memory_region_get_dirty_log_mask(mr); | |
2532 | /* No early return if dirty_log_mask is or becomes 0, because | |
2533 | * cpu_physical_memory_set_dirty_range will still call | |
2534 | * xen_modified_memory. | |
2535 | */ | |
2536 | if (dirty_log_mask) { | |
2537 | dirty_log_mask = | |
2538 | cpu_physical_memory_range_includes_clean(addr, length, dirty_log_mask); | |
2539 | } | |
2540 | if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) { | |
2541 | tb_invalidate_phys_range(addr, addr + length); | |
2542 | dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE); | |
2543 | } | |
2544 | cpu_physical_memory_set_dirty_range(addr, length, dirty_log_mask); | |
2545 | } | |
2546 | ||
2547 | static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr) | |
2548 | { | |
2549 | unsigned access_size_max = mr->ops->valid.max_access_size; | |
2550 | ||
2551 | /* Regions are assumed to support 1-4 byte accesses unless | |
2552 | otherwise specified. */ | |
2553 | if (access_size_max == 0) { | |
2554 | access_size_max = 4; | |
2555 | } | |
2556 | ||
2557 | /* Bound the maximum access by the alignment of the address. */ | |
2558 | if (!mr->ops->impl.unaligned) { | |
2559 | unsigned align_size_max = addr & -addr; | |
2560 | if (align_size_max != 0 && align_size_max < access_size_max) { | |
2561 | access_size_max = align_size_max; | |
2562 | } | |
2563 | } | |
2564 | ||
2565 | /* Don't attempt accesses larger than the maximum. */ | |
2566 | if (l > access_size_max) { | |
2567 | l = access_size_max; | |
2568 | } | |
2569 | l = pow2floor(l); | |
2570 | ||
2571 | return l; | |
2572 | } | |
2573 | ||
2574 | static bool prepare_mmio_access(MemoryRegion *mr) | |
2575 | { | |
2576 | bool unlocked = !qemu_mutex_iothread_locked(); | |
2577 | bool release_lock = false; | |
2578 | ||
2579 | if (unlocked && mr->global_locking) { | |
2580 | qemu_mutex_lock_iothread(); | |
2581 | unlocked = false; | |
2582 | release_lock = true; | |
2583 | } | |
2584 | if (mr->flush_coalesced_mmio) { | |
2585 | if (unlocked) { | |
2586 | qemu_mutex_lock_iothread(); | |
2587 | } | |
2588 | qemu_flush_coalesced_mmio_buffer(); | |
2589 | if (unlocked) { | |
2590 | qemu_mutex_unlock_iothread(); | |
2591 | } | |
2592 | } | |
2593 | ||
2594 | return release_lock; | |
2595 | } | |
2596 | ||
2597 | /* Called within RCU critical section. */ | |
2598 | static MemTxResult address_space_write_continue(AddressSpace *as, hwaddr addr, | |
2599 | MemTxAttrs attrs, | |
2600 | const uint8_t *buf, | |
2601 | int len, hwaddr addr1, | |
2602 | hwaddr l, MemoryRegion *mr) | |
2603 | { | |
2604 | uint8_t *ptr; | |
2605 | uint64_t val; | |
2606 | MemTxResult result = MEMTX_OK; | |
2607 | bool release_lock = false; | |
2608 | ||
2609 | for (;;) { | |
2610 | if (!memory_access_is_direct(mr, true)) { | |
2611 | release_lock |= prepare_mmio_access(mr); | |
2612 | l = memory_access_size(mr, l, addr1); | |
2613 | /* XXX: could force current_cpu to NULL to avoid | |
2614 | potential bugs */ | |
2615 | switch (l) { | |
2616 | case 8: | |
2617 | /* 64 bit write access */ | |
2618 | val = ldq_p(buf); | |
2619 | result |= memory_region_dispatch_write(mr, addr1, val, 8, | |
2620 | attrs); | |
2621 | break; | |
2622 | case 4: | |
2623 | /* 32 bit write access */ | |
2624 | val = ldl_p(buf); | |
2625 | result |= memory_region_dispatch_write(mr, addr1, val, 4, | |
2626 | attrs); | |
2627 | break; | |
2628 | case 2: | |
2629 | /* 16 bit write access */ | |
2630 | val = lduw_p(buf); | |
2631 | result |= memory_region_dispatch_write(mr, addr1, val, 2, | |
2632 | attrs); | |
2633 | break; | |
2634 | case 1: | |
2635 | /* 8 bit write access */ | |
2636 | val = ldub_p(buf); | |
2637 | result |= memory_region_dispatch_write(mr, addr1, val, 1, | |
2638 | attrs); | |
2639 | break; | |
2640 | default: | |
2641 | abort(); | |
2642 | } | |
2643 | } else { | |
2644 | addr1 += memory_region_get_ram_addr(mr); | |
2645 | /* RAM case */ | |
2646 | ptr = qemu_get_ram_ptr(mr->ram_block, addr1); | |
2647 | memcpy(ptr, buf, l); | |
2648 | invalidate_and_set_dirty(mr, addr1, l); | |
2649 | } | |
2650 | ||
2651 | if (release_lock) { | |
2652 | qemu_mutex_unlock_iothread(); | |
2653 | release_lock = false; | |
2654 | } | |
2655 | ||
2656 | len -= l; | |
2657 | buf += l; | |
2658 | addr += l; | |
2659 | ||
2660 | if (!len) { | |
2661 | break; | |
2662 | } | |
2663 | ||
2664 | l = len; | |
2665 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
2666 | } | |
2667 | ||
2668 | return result; | |
2669 | } | |
2670 | ||
2671 | MemTxResult address_space_write(AddressSpace *as, hwaddr addr, MemTxAttrs attrs, | |
2672 | const uint8_t *buf, int len) | |
2673 | { | |
2674 | hwaddr l; | |
2675 | hwaddr addr1; | |
2676 | MemoryRegion *mr; | |
2677 | MemTxResult result = MEMTX_OK; | |
2678 | ||
2679 | if (len > 0) { | |
2680 | rcu_read_lock(); | |
2681 | l = len; | |
2682 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
2683 | result = address_space_write_continue(as, addr, attrs, buf, len, | |
2684 | addr1, l, mr); | |
2685 | rcu_read_unlock(); | |
2686 | } | |
2687 | ||
2688 | return result; | |
2689 | } | |
2690 | ||
2691 | /* Called within RCU critical section. */ | |
2692 | MemTxResult address_space_read_continue(AddressSpace *as, hwaddr addr, | |
2693 | MemTxAttrs attrs, uint8_t *buf, | |
2694 | int len, hwaddr addr1, hwaddr l, | |
2695 | MemoryRegion *mr) | |
2696 | { | |
2697 | uint8_t *ptr; | |
2698 | uint64_t val; | |
2699 | MemTxResult result = MEMTX_OK; | |
2700 | bool release_lock = false; | |
2701 | ||
2702 | for (;;) { | |
2703 | if (!memory_access_is_direct(mr, false)) { | |
2704 | /* I/O case */ | |
2705 | release_lock |= prepare_mmio_access(mr); | |
2706 | l = memory_access_size(mr, l, addr1); | |
2707 | switch (l) { | |
2708 | case 8: | |
2709 | /* 64 bit read access */ | |
2710 | result |= memory_region_dispatch_read(mr, addr1, &val, 8, | |
2711 | attrs); | |
2712 | stq_p(buf, val); | |
2713 | break; | |
2714 | case 4: | |
2715 | /* 32 bit read access */ | |
2716 | result |= memory_region_dispatch_read(mr, addr1, &val, 4, | |
2717 | attrs); | |
2718 | stl_p(buf, val); | |
2719 | break; | |
2720 | case 2: | |
2721 | /* 16 bit read access */ | |
2722 | result |= memory_region_dispatch_read(mr, addr1, &val, 2, | |
2723 | attrs); | |
2724 | stw_p(buf, val); | |
2725 | break; | |
2726 | case 1: | |
2727 | /* 8 bit read access */ | |
2728 | result |= memory_region_dispatch_read(mr, addr1, &val, 1, | |
2729 | attrs); | |
2730 | stb_p(buf, val); | |
2731 | break; | |
2732 | default: | |
2733 | abort(); | |
2734 | } | |
2735 | } else { | |
2736 | /* RAM case */ | |
2737 | ptr = qemu_get_ram_ptr(mr->ram_block, | |
2738 | memory_region_get_ram_addr(mr) + addr1); | |
2739 | memcpy(buf, ptr, l); | |
2740 | } | |
2741 | ||
2742 | if (release_lock) { | |
2743 | qemu_mutex_unlock_iothread(); | |
2744 | release_lock = false; | |
2745 | } | |
2746 | ||
2747 | len -= l; | |
2748 | buf += l; | |
2749 | addr += l; | |
2750 | ||
2751 | if (!len) { | |
2752 | break; | |
2753 | } | |
2754 | ||
2755 | l = len; | |
2756 | mr = address_space_translate(as, addr, &addr1, &l, false); | |
2757 | } | |
2758 | ||
2759 | return result; | |
2760 | } | |
2761 | ||
2762 | MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr, | |
2763 | MemTxAttrs attrs, uint8_t *buf, int len) | |
2764 | { | |
2765 | hwaddr l; | |
2766 | hwaddr addr1; | |
2767 | MemoryRegion *mr; | |
2768 | MemTxResult result = MEMTX_OK; | |
2769 | ||
2770 | if (len > 0) { | |
2771 | rcu_read_lock(); | |
2772 | l = len; | |
2773 | mr = address_space_translate(as, addr, &addr1, &l, false); | |
2774 | result = address_space_read_continue(as, addr, attrs, buf, len, | |
2775 | addr1, l, mr); | |
2776 | rcu_read_unlock(); | |
2777 | } | |
2778 | ||
2779 | return result; | |
2780 | } | |
2781 | ||
2782 | MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, MemTxAttrs attrs, | |
2783 | uint8_t *buf, int len, bool is_write) | |
2784 | { | |
2785 | if (is_write) { | |
2786 | return address_space_write(as, addr, attrs, (uint8_t *)buf, len); | |
2787 | } else { | |
2788 | return address_space_read(as, addr, attrs, (uint8_t *)buf, len); | |
2789 | } | |
2790 | } | |
2791 | ||
2792 | void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf, | |
2793 | int len, int is_write) | |
2794 | { | |
2795 | address_space_rw(&address_space_memory, addr, MEMTXATTRS_UNSPECIFIED, | |
2796 | buf, len, is_write); | |
2797 | } | |
2798 | ||
2799 | enum write_rom_type { | |
2800 | WRITE_DATA, | |
2801 | FLUSH_CACHE, | |
2802 | }; | |
2803 | ||
2804 | static inline void cpu_physical_memory_write_rom_internal(AddressSpace *as, | |
2805 | hwaddr addr, const uint8_t *buf, int len, enum write_rom_type type) | |
2806 | { | |
2807 | hwaddr l; | |
2808 | uint8_t *ptr; | |
2809 | hwaddr addr1; | |
2810 | MemoryRegion *mr; | |
2811 | ||
2812 | rcu_read_lock(); | |
2813 | while (len > 0) { | |
2814 | l = len; | |
2815 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
2816 | ||
2817 | if (!(memory_region_is_ram(mr) || | |
2818 | memory_region_is_romd(mr))) { | |
2819 | l = memory_access_size(mr, l, addr1); | |
2820 | } else { | |
2821 | addr1 += memory_region_get_ram_addr(mr); | |
2822 | /* ROM/RAM case */ | |
2823 | ptr = qemu_get_ram_ptr(mr->ram_block, addr1); | |
2824 | switch (type) { | |
2825 | case WRITE_DATA: | |
2826 | memcpy(ptr, buf, l); | |
2827 | invalidate_and_set_dirty(mr, addr1, l); | |
2828 | break; | |
2829 | case FLUSH_CACHE: | |
2830 | flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l); | |
2831 | break; | |
2832 | } | |
2833 | } | |
2834 | len -= l; | |
2835 | buf += l; | |
2836 | addr += l; | |
2837 | } | |
2838 | rcu_read_unlock(); | |
2839 | } | |
2840 | ||
2841 | /* used for ROM loading : can write in RAM and ROM */ | |
2842 | void cpu_physical_memory_write_rom(AddressSpace *as, hwaddr addr, | |
2843 | const uint8_t *buf, int len) | |
2844 | { | |
2845 | cpu_physical_memory_write_rom_internal(as, addr, buf, len, WRITE_DATA); | |
2846 | } | |
2847 | ||
2848 | void cpu_flush_icache_range(hwaddr start, int len) | |
2849 | { | |
2850 | /* | |
2851 | * This function should do the same thing as an icache flush that was | |
2852 | * triggered from within the guest. For TCG we are always cache coherent, | |
2853 | * so there is no need to flush anything. For KVM / Xen we need to flush | |
2854 | * the host's instruction cache at least. | |
2855 | */ | |
2856 | if (tcg_enabled()) { | |
2857 | return; | |
2858 | } | |
2859 | ||
2860 | cpu_physical_memory_write_rom_internal(&address_space_memory, | |
2861 | start, NULL, len, FLUSH_CACHE); | |
2862 | } | |
2863 | ||
2864 | typedef struct { | |
2865 | MemoryRegion *mr; | |
2866 | void *buffer; | |
2867 | hwaddr addr; | |
2868 | hwaddr len; | |
2869 | bool in_use; | |
2870 | } BounceBuffer; | |
2871 | ||
2872 | static BounceBuffer bounce; | |
2873 | ||
2874 | typedef struct MapClient { | |
2875 | QEMUBH *bh; | |
2876 | QLIST_ENTRY(MapClient) link; | |
2877 | } MapClient; | |
2878 | ||
2879 | QemuMutex map_client_list_lock; | |
2880 | static QLIST_HEAD(map_client_list, MapClient) map_client_list | |
2881 | = QLIST_HEAD_INITIALIZER(map_client_list); | |
2882 | ||
2883 | static void cpu_unregister_map_client_do(MapClient *client) | |
2884 | { | |
2885 | QLIST_REMOVE(client, link); | |
2886 | g_free(client); | |
2887 | } | |
2888 | ||
2889 | static void cpu_notify_map_clients_locked(void) | |
2890 | { | |
2891 | MapClient *client; | |
2892 | ||
2893 | while (!QLIST_EMPTY(&map_client_list)) { | |
2894 | client = QLIST_FIRST(&map_client_list); | |
2895 | qemu_bh_schedule(client->bh); | |
2896 | cpu_unregister_map_client_do(client); | |
2897 | } | |
2898 | } | |
2899 | ||
2900 | void cpu_register_map_client(QEMUBH *bh) | |
2901 | { | |
2902 | MapClient *client = g_malloc(sizeof(*client)); | |
2903 | ||
2904 | qemu_mutex_lock(&map_client_list_lock); | |
2905 | client->bh = bh; | |
2906 | QLIST_INSERT_HEAD(&map_client_list, client, link); | |
2907 | if (!atomic_read(&bounce.in_use)) { | |
2908 | cpu_notify_map_clients_locked(); | |
2909 | } | |
2910 | qemu_mutex_unlock(&map_client_list_lock); | |
2911 | } | |
2912 | ||
2913 | void cpu_exec_init_all(void) | |
2914 | { | |
2915 | qemu_mutex_init(&ram_list.mutex); | |
2916 | io_mem_init(); | |
2917 | memory_map_init(); | |
2918 | qemu_mutex_init(&map_client_list_lock); | |
2919 | } | |
2920 | ||
2921 | void cpu_unregister_map_client(QEMUBH *bh) | |
2922 | { | |
2923 | MapClient *client; | |
2924 | ||
2925 | qemu_mutex_lock(&map_client_list_lock); | |
2926 | QLIST_FOREACH(client, &map_client_list, link) { | |
2927 | if (client->bh == bh) { | |
2928 | cpu_unregister_map_client_do(client); | |
2929 | break; | |
2930 | } | |
2931 | } | |
2932 | qemu_mutex_unlock(&map_client_list_lock); | |
2933 | } | |
2934 | ||
2935 | static void cpu_notify_map_clients(void) | |
2936 | { | |
2937 | qemu_mutex_lock(&map_client_list_lock); | |
2938 | cpu_notify_map_clients_locked(); | |
2939 | qemu_mutex_unlock(&map_client_list_lock); | |
2940 | } | |
2941 | ||
2942 | bool address_space_access_valid(AddressSpace *as, hwaddr addr, int len, bool is_write) | |
2943 | { | |
2944 | MemoryRegion *mr; | |
2945 | hwaddr l, xlat; | |
2946 | ||
2947 | rcu_read_lock(); | |
2948 | while (len > 0) { | |
2949 | l = len; | |
2950 | mr = address_space_translate(as, addr, &xlat, &l, is_write); | |
2951 | if (!memory_access_is_direct(mr, is_write)) { | |
2952 | l = memory_access_size(mr, l, addr); | |
2953 | if (!memory_region_access_valid(mr, xlat, l, is_write)) { | |
2954 | return false; | |
2955 | } | |
2956 | } | |
2957 | ||
2958 | len -= l; | |
2959 | addr += l; | |
2960 | } | |
2961 | rcu_read_unlock(); | |
2962 | return true; | |
2963 | } | |
2964 | ||
2965 | /* Map a physical memory region into a host virtual address. | |
2966 | * May map a subset of the requested range, given by and returned in *plen. | |
2967 | * May return NULL if resources needed to perform the mapping are exhausted. | |
2968 | * Use only for reads OR writes - not for read-modify-write operations. | |
2969 | * Use cpu_register_map_client() to know when retrying the map operation is | |
2970 | * likely to succeed. | |
2971 | */ | |
2972 | void *address_space_map(AddressSpace *as, | |
2973 | hwaddr addr, | |
2974 | hwaddr *plen, | |
2975 | bool is_write) | |
2976 | { | |
2977 | hwaddr len = *plen; | |
2978 | hwaddr done = 0; | |
2979 | hwaddr l, xlat, base; | |
2980 | MemoryRegion *mr, *this_mr; | |
2981 | ram_addr_t raddr; | |
2982 | void *ptr; | |
2983 | ||
2984 | if (len == 0) { | |
2985 | return NULL; | |
2986 | } | |
2987 | ||
2988 | l = len; | |
2989 | rcu_read_lock(); | |
2990 | mr = address_space_translate(as, addr, &xlat, &l, is_write); | |
2991 | ||
2992 | if (!memory_access_is_direct(mr, is_write)) { | |
2993 | if (atomic_xchg(&bounce.in_use, true)) { | |
2994 | rcu_read_unlock(); | |
2995 | return NULL; | |
2996 | } | |
2997 | /* Avoid unbounded allocations */ | |
2998 | l = MIN(l, TARGET_PAGE_SIZE); | |
2999 | bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l); | |
3000 | bounce.addr = addr; | |
3001 | bounce.len = l; | |
3002 | ||
3003 | memory_region_ref(mr); | |
3004 | bounce.mr = mr; | |
3005 | if (!is_write) { | |
3006 | address_space_read(as, addr, MEMTXATTRS_UNSPECIFIED, | |
3007 | bounce.buffer, l); | |
3008 | } | |
3009 | ||
3010 | rcu_read_unlock(); | |
3011 | *plen = l; | |
3012 | return bounce.buffer; | |
3013 | } | |
3014 | ||
3015 | base = xlat; | |
3016 | raddr = memory_region_get_ram_addr(mr); | |
3017 | ||
3018 | for (;;) { | |
3019 | len -= l; | |
3020 | addr += l; | |
3021 | done += l; | |
3022 | if (len == 0) { | |
3023 | break; | |
3024 | } | |
3025 | ||
3026 | l = len; | |
3027 | this_mr = address_space_translate(as, addr, &xlat, &l, is_write); | |
3028 | if (this_mr != mr || xlat != base + done) { | |
3029 | break; | |
3030 | } | |
3031 | } | |
3032 | ||
3033 | memory_region_ref(mr); | |
3034 | *plen = done; | |
3035 | ptr = qemu_ram_ptr_length(mr->ram_block, raddr + base, plen); | |
3036 | rcu_read_unlock(); | |
3037 | ||
3038 | return ptr; | |
3039 | } | |
3040 | ||
3041 | /* Unmaps a memory region previously mapped by address_space_map(). | |
3042 | * Will also mark the memory as dirty if is_write == 1. access_len gives | |
3043 | * the amount of memory that was actually read or written by the caller. | |
3044 | */ | |
3045 | void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len, | |
3046 | int is_write, hwaddr access_len) | |
3047 | { | |
3048 | if (buffer != bounce.buffer) { | |
3049 | MemoryRegion *mr; | |
3050 | ram_addr_t addr1; | |
3051 | ||
3052 | mr = qemu_ram_addr_from_host(buffer, &addr1); | |
3053 | assert(mr != NULL); | |
3054 | if (is_write) { | |
3055 | invalidate_and_set_dirty(mr, addr1, access_len); | |
3056 | } | |
3057 | if (xen_enabled()) { | |
3058 | xen_invalidate_map_cache_entry(buffer); | |
3059 | } | |
3060 | memory_region_unref(mr); | |
3061 | return; | |
3062 | } | |
3063 | if (is_write) { | |
3064 | address_space_write(as, bounce.addr, MEMTXATTRS_UNSPECIFIED, | |
3065 | bounce.buffer, access_len); | |
3066 | } | |
3067 | qemu_vfree(bounce.buffer); | |
3068 | bounce.buffer = NULL; | |
3069 | memory_region_unref(bounce.mr); | |
3070 | atomic_mb_set(&bounce.in_use, false); | |
3071 | cpu_notify_map_clients(); | |
3072 | } | |
3073 | ||
3074 | void *cpu_physical_memory_map(hwaddr addr, | |
3075 | hwaddr *plen, | |
3076 | int is_write) | |
3077 | { | |
3078 | return address_space_map(&address_space_memory, addr, plen, is_write); | |
3079 | } | |
3080 | ||
3081 | void cpu_physical_memory_unmap(void *buffer, hwaddr len, | |
3082 | int is_write, hwaddr access_len) | |
3083 | { | |
3084 | return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len); | |
3085 | } | |
3086 | ||
3087 | /* warning: addr must be aligned */ | |
3088 | static inline uint32_t address_space_ldl_internal(AddressSpace *as, hwaddr addr, | |
3089 | MemTxAttrs attrs, | |
3090 | MemTxResult *result, | |
3091 | enum device_endian endian) | |
3092 | { | |
3093 | uint8_t *ptr; | |
3094 | uint64_t val; | |
3095 | MemoryRegion *mr; | |
3096 | hwaddr l = 4; | |
3097 | hwaddr addr1; | |
3098 | MemTxResult r; | |
3099 | bool release_lock = false; | |
3100 | ||
3101 | rcu_read_lock(); | |
3102 | mr = address_space_translate(as, addr, &addr1, &l, false); | |
3103 | if (l < 4 || !memory_access_is_direct(mr, false)) { | |
3104 | release_lock |= prepare_mmio_access(mr); | |
3105 | ||
3106 | /* I/O case */ | |
3107 | r = memory_region_dispatch_read(mr, addr1, &val, 4, attrs); | |
3108 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3109 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3110 | val = bswap32(val); | |
3111 | } | |
3112 | #else | |
3113 | if (endian == DEVICE_BIG_ENDIAN) { | |
3114 | val = bswap32(val); | |
3115 | } | |
3116 | #endif | |
3117 | } else { | |
3118 | /* RAM case */ | |
3119 | ptr = qemu_get_ram_ptr(mr->ram_block, | |
3120 | (memory_region_get_ram_addr(mr) | |
3121 | & TARGET_PAGE_MASK) | |
3122 | + addr1); | |
3123 | switch (endian) { | |
3124 | case DEVICE_LITTLE_ENDIAN: | |
3125 | val = ldl_le_p(ptr); | |
3126 | break; | |
3127 | case DEVICE_BIG_ENDIAN: | |
3128 | val = ldl_be_p(ptr); | |
3129 | break; | |
3130 | default: | |
3131 | val = ldl_p(ptr); | |
3132 | break; | |
3133 | } | |
3134 | r = MEMTX_OK; | |
3135 | } | |
3136 | if (result) { | |
3137 | *result = r; | |
3138 | } | |
3139 | if (release_lock) { | |
3140 | qemu_mutex_unlock_iothread(); | |
3141 | } | |
3142 | rcu_read_unlock(); | |
3143 | return val; | |
3144 | } | |
3145 | ||
3146 | uint32_t address_space_ldl(AddressSpace *as, hwaddr addr, | |
3147 | MemTxAttrs attrs, MemTxResult *result) | |
3148 | { | |
3149 | return address_space_ldl_internal(as, addr, attrs, result, | |
3150 | DEVICE_NATIVE_ENDIAN); | |
3151 | } | |
3152 | ||
3153 | uint32_t address_space_ldl_le(AddressSpace *as, hwaddr addr, | |
3154 | MemTxAttrs attrs, MemTxResult *result) | |
3155 | { | |
3156 | return address_space_ldl_internal(as, addr, attrs, result, | |
3157 | DEVICE_LITTLE_ENDIAN); | |
3158 | } | |
3159 | ||
3160 | uint32_t address_space_ldl_be(AddressSpace *as, hwaddr addr, | |
3161 | MemTxAttrs attrs, MemTxResult *result) | |
3162 | { | |
3163 | return address_space_ldl_internal(as, addr, attrs, result, | |
3164 | DEVICE_BIG_ENDIAN); | |
3165 | } | |
3166 | ||
3167 | uint32_t ldl_phys(AddressSpace *as, hwaddr addr) | |
3168 | { | |
3169 | return address_space_ldl(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3170 | } | |
3171 | ||
3172 | uint32_t ldl_le_phys(AddressSpace *as, hwaddr addr) | |
3173 | { | |
3174 | return address_space_ldl_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3175 | } | |
3176 | ||
3177 | uint32_t ldl_be_phys(AddressSpace *as, hwaddr addr) | |
3178 | { | |
3179 | return address_space_ldl_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3180 | } | |
3181 | ||
3182 | /* warning: addr must be aligned */ | |
3183 | static inline uint64_t address_space_ldq_internal(AddressSpace *as, hwaddr addr, | |
3184 | MemTxAttrs attrs, | |
3185 | MemTxResult *result, | |
3186 | enum device_endian endian) | |
3187 | { | |
3188 | uint8_t *ptr; | |
3189 | uint64_t val; | |
3190 | MemoryRegion *mr; | |
3191 | hwaddr l = 8; | |
3192 | hwaddr addr1; | |
3193 | MemTxResult r; | |
3194 | bool release_lock = false; | |
3195 | ||
3196 | rcu_read_lock(); | |
3197 | mr = address_space_translate(as, addr, &addr1, &l, | |
3198 | false); | |
3199 | if (l < 8 || !memory_access_is_direct(mr, false)) { | |
3200 | release_lock |= prepare_mmio_access(mr); | |
3201 | ||
3202 | /* I/O case */ | |
3203 | r = memory_region_dispatch_read(mr, addr1, &val, 8, attrs); | |
3204 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3205 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3206 | val = bswap64(val); | |
3207 | } | |
3208 | #else | |
3209 | if (endian == DEVICE_BIG_ENDIAN) { | |
3210 | val = bswap64(val); | |
3211 | } | |
3212 | #endif | |
3213 | } else { | |
3214 | /* RAM case */ | |
3215 | ptr = qemu_get_ram_ptr(mr->ram_block, | |
3216 | (memory_region_get_ram_addr(mr) | |
3217 | & TARGET_PAGE_MASK) | |
3218 | + addr1); | |
3219 | switch (endian) { | |
3220 | case DEVICE_LITTLE_ENDIAN: | |
3221 | val = ldq_le_p(ptr); | |
3222 | break; | |
3223 | case DEVICE_BIG_ENDIAN: | |
3224 | val = ldq_be_p(ptr); | |
3225 | break; | |
3226 | default: | |
3227 | val = ldq_p(ptr); | |
3228 | break; | |
3229 | } | |
3230 | r = MEMTX_OK; | |
3231 | } | |
3232 | if (result) { | |
3233 | *result = r; | |
3234 | } | |
3235 | if (release_lock) { | |
3236 | qemu_mutex_unlock_iothread(); | |
3237 | } | |
3238 | rcu_read_unlock(); | |
3239 | return val; | |
3240 | } | |
3241 | ||
3242 | uint64_t address_space_ldq(AddressSpace *as, hwaddr addr, | |
3243 | MemTxAttrs attrs, MemTxResult *result) | |
3244 | { | |
3245 | return address_space_ldq_internal(as, addr, attrs, result, | |
3246 | DEVICE_NATIVE_ENDIAN); | |
3247 | } | |
3248 | ||
3249 | uint64_t address_space_ldq_le(AddressSpace *as, hwaddr addr, | |
3250 | MemTxAttrs attrs, MemTxResult *result) | |
3251 | { | |
3252 | return address_space_ldq_internal(as, addr, attrs, result, | |
3253 | DEVICE_LITTLE_ENDIAN); | |
3254 | } | |
3255 | ||
3256 | uint64_t address_space_ldq_be(AddressSpace *as, hwaddr addr, | |
3257 | MemTxAttrs attrs, MemTxResult *result) | |
3258 | { | |
3259 | return address_space_ldq_internal(as, addr, attrs, result, | |
3260 | DEVICE_BIG_ENDIAN); | |
3261 | } | |
3262 | ||
3263 | uint64_t ldq_phys(AddressSpace *as, hwaddr addr) | |
3264 | { | |
3265 | return address_space_ldq(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3266 | } | |
3267 | ||
3268 | uint64_t ldq_le_phys(AddressSpace *as, hwaddr addr) | |
3269 | { | |
3270 | return address_space_ldq_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3271 | } | |
3272 | ||
3273 | uint64_t ldq_be_phys(AddressSpace *as, hwaddr addr) | |
3274 | { | |
3275 | return address_space_ldq_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3276 | } | |
3277 | ||
3278 | /* XXX: optimize */ | |
3279 | uint32_t address_space_ldub(AddressSpace *as, hwaddr addr, | |
3280 | MemTxAttrs attrs, MemTxResult *result) | |
3281 | { | |
3282 | uint8_t val; | |
3283 | MemTxResult r; | |
3284 | ||
3285 | r = address_space_rw(as, addr, attrs, &val, 1, 0); | |
3286 | if (result) { | |
3287 | *result = r; | |
3288 | } | |
3289 | return val; | |
3290 | } | |
3291 | ||
3292 | uint32_t ldub_phys(AddressSpace *as, hwaddr addr) | |
3293 | { | |
3294 | return address_space_ldub(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3295 | } | |
3296 | ||
3297 | /* warning: addr must be aligned */ | |
3298 | static inline uint32_t address_space_lduw_internal(AddressSpace *as, | |
3299 | hwaddr addr, | |
3300 | MemTxAttrs attrs, | |
3301 | MemTxResult *result, | |
3302 | enum device_endian endian) | |
3303 | { | |
3304 | uint8_t *ptr; | |
3305 | uint64_t val; | |
3306 | MemoryRegion *mr; | |
3307 | hwaddr l = 2; | |
3308 | hwaddr addr1; | |
3309 | MemTxResult r; | |
3310 | bool release_lock = false; | |
3311 | ||
3312 | rcu_read_lock(); | |
3313 | mr = address_space_translate(as, addr, &addr1, &l, | |
3314 | false); | |
3315 | if (l < 2 || !memory_access_is_direct(mr, false)) { | |
3316 | release_lock |= prepare_mmio_access(mr); | |
3317 | ||
3318 | /* I/O case */ | |
3319 | r = memory_region_dispatch_read(mr, addr1, &val, 2, attrs); | |
3320 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3321 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3322 | val = bswap16(val); | |
3323 | } | |
3324 | #else | |
3325 | if (endian == DEVICE_BIG_ENDIAN) { | |
3326 | val = bswap16(val); | |
3327 | } | |
3328 | #endif | |
3329 | } else { | |
3330 | /* RAM case */ | |
3331 | ptr = qemu_get_ram_ptr(mr->ram_block, | |
3332 | (memory_region_get_ram_addr(mr) | |
3333 | & TARGET_PAGE_MASK) | |
3334 | + addr1); | |
3335 | switch (endian) { | |
3336 | case DEVICE_LITTLE_ENDIAN: | |
3337 | val = lduw_le_p(ptr); | |
3338 | break; | |
3339 | case DEVICE_BIG_ENDIAN: | |
3340 | val = lduw_be_p(ptr); | |
3341 | break; | |
3342 | default: | |
3343 | val = lduw_p(ptr); | |
3344 | break; | |
3345 | } | |
3346 | r = MEMTX_OK; | |
3347 | } | |
3348 | if (result) { | |
3349 | *result = r; | |
3350 | } | |
3351 | if (release_lock) { | |
3352 | qemu_mutex_unlock_iothread(); | |
3353 | } | |
3354 | rcu_read_unlock(); | |
3355 | return val; | |
3356 | } | |
3357 | ||
3358 | uint32_t address_space_lduw(AddressSpace *as, hwaddr addr, | |
3359 | MemTxAttrs attrs, MemTxResult *result) | |
3360 | { | |
3361 | return address_space_lduw_internal(as, addr, attrs, result, | |
3362 | DEVICE_NATIVE_ENDIAN); | |
3363 | } | |
3364 | ||
3365 | uint32_t address_space_lduw_le(AddressSpace *as, hwaddr addr, | |
3366 | MemTxAttrs attrs, MemTxResult *result) | |
3367 | { | |
3368 | return address_space_lduw_internal(as, addr, attrs, result, | |
3369 | DEVICE_LITTLE_ENDIAN); | |
3370 | } | |
3371 | ||
3372 | uint32_t address_space_lduw_be(AddressSpace *as, hwaddr addr, | |
3373 | MemTxAttrs attrs, MemTxResult *result) | |
3374 | { | |
3375 | return address_space_lduw_internal(as, addr, attrs, result, | |
3376 | DEVICE_BIG_ENDIAN); | |
3377 | } | |
3378 | ||
3379 | uint32_t lduw_phys(AddressSpace *as, hwaddr addr) | |
3380 | { | |
3381 | return address_space_lduw(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3382 | } | |
3383 | ||
3384 | uint32_t lduw_le_phys(AddressSpace *as, hwaddr addr) | |
3385 | { | |
3386 | return address_space_lduw_le(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3387 | } | |
3388 | ||
3389 | uint32_t lduw_be_phys(AddressSpace *as, hwaddr addr) | |
3390 | { | |
3391 | return address_space_lduw_be(as, addr, MEMTXATTRS_UNSPECIFIED, NULL); | |
3392 | } | |
3393 | ||
3394 | /* warning: addr must be aligned. The ram page is not masked as dirty | |
3395 | and the code inside is not invalidated. It is useful if the dirty | |
3396 | bits are used to track modified PTEs */ | |
3397 | void address_space_stl_notdirty(AddressSpace *as, hwaddr addr, uint32_t val, | |
3398 | MemTxAttrs attrs, MemTxResult *result) | |
3399 | { | |
3400 | uint8_t *ptr; | |
3401 | MemoryRegion *mr; | |
3402 | hwaddr l = 4; | |
3403 | hwaddr addr1; | |
3404 | MemTxResult r; | |
3405 | uint8_t dirty_log_mask; | |
3406 | bool release_lock = false; | |
3407 | ||
3408 | rcu_read_lock(); | |
3409 | mr = address_space_translate(as, addr, &addr1, &l, | |
3410 | true); | |
3411 | if (l < 4 || !memory_access_is_direct(mr, true)) { | |
3412 | release_lock |= prepare_mmio_access(mr); | |
3413 | ||
3414 | r = memory_region_dispatch_write(mr, addr1, val, 4, attrs); | |
3415 | } else { | |
3416 | addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK; | |
3417 | ptr = qemu_get_ram_ptr(mr->ram_block, addr1); | |
3418 | stl_p(ptr, val); | |
3419 | ||
3420 | dirty_log_mask = memory_region_get_dirty_log_mask(mr); | |
3421 | dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE); | |
3422 | cpu_physical_memory_set_dirty_range(addr1, 4, dirty_log_mask); | |
3423 | r = MEMTX_OK; | |
3424 | } | |
3425 | if (result) { | |
3426 | *result = r; | |
3427 | } | |
3428 | if (release_lock) { | |
3429 | qemu_mutex_unlock_iothread(); | |
3430 | } | |
3431 | rcu_read_unlock(); | |
3432 | } | |
3433 | ||
3434 | void stl_phys_notdirty(AddressSpace *as, hwaddr addr, uint32_t val) | |
3435 | { | |
3436 | address_space_stl_notdirty(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3437 | } | |
3438 | ||
3439 | /* warning: addr must be aligned */ | |
3440 | static inline void address_space_stl_internal(AddressSpace *as, | |
3441 | hwaddr addr, uint32_t val, | |
3442 | MemTxAttrs attrs, | |
3443 | MemTxResult *result, | |
3444 | enum device_endian endian) | |
3445 | { | |
3446 | uint8_t *ptr; | |
3447 | MemoryRegion *mr; | |
3448 | hwaddr l = 4; | |
3449 | hwaddr addr1; | |
3450 | MemTxResult r; | |
3451 | bool release_lock = false; | |
3452 | ||
3453 | rcu_read_lock(); | |
3454 | mr = address_space_translate(as, addr, &addr1, &l, | |
3455 | true); | |
3456 | if (l < 4 || !memory_access_is_direct(mr, true)) { | |
3457 | release_lock |= prepare_mmio_access(mr); | |
3458 | ||
3459 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3460 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3461 | val = bswap32(val); | |
3462 | } | |
3463 | #else | |
3464 | if (endian == DEVICE_BIG_ENDIAN) { | |
3465 | val = bswap32(val); | |
3466 | } | |
3467 | #endif | |
3468 | r = memory_region_dispatch_write(mr, addr1, val, 4, attrs); | |
3469 | } else { | |
3470 | /* RAM case */ | |
3471 | addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK; | |
3472 | ptr = qemu_get_ram_ptr(mr->ram_block, addr1); | |
3473 | switch (endian) { | |
3474 | case DEVICE_LITTLE_ENDIAN: | |
3475 | stl_le_p(ptr, val); | |
3476 | break; | |
3477 | case DEVICE_BIG_ENDIAN: | |
3478 | stl_be_p(ptr, val); | |
3479 | break; | |
3480 | default: | |
3481 | stl_p(ptr, val); | |
3482 | break; | |
3483 | } | |
3484 | invalidate_and_set_dirty(mr, addr1, 4); | |
3485 | r = MEMTX_OK; | |
3486 | } | |
3487 | if (result) { | |
3488 | *result = r; | |
3489 | } | |
3490 | if (release_lock) { | |
3491 | qemu_mutex_unlock_iothread(); | |
3492 | } | |
3493 | rcu_read_unlock(); | |
3494 | } | |
3495 | ||
3496 | void address_space_stl(AddressSpace *as, hwaddr addr, uint32_t val, | |
3497 | MemTxAttrs attrs, MemTxResult *result) | |
3498 | { | |
3499 | address_space_stl_internal(as, addr, val, attrs, result, | |
3500 | DEVICE_NATIVE_ENDIAN); | |
3501 | } | |
3502 | ||
3503 | void address_space_stl_le(AddressSpace *as, hwaddr addr, uint32_t val, | |
3504 | MemTxAttrs attrs, MemTxResult *result) | |
3505 | { | |
3506 | address_space_stl_internal(as, addr, val, attrs, result, | |
3507 | DEVICE_LITTLE_ENDIAN); | |
3508 | } | |
3509 | ||
3510 | void address_space_stl_be(AddressSpace *as, hwaddr addr, uint32_t val, | |
3511 | MemTxAttrs attrs, MemTxResult *result) | |
3512 | { | |
3513 | address_space_stl_internal(as, addr, val, attrs, result, | |
3514 | DEVICE_BIG_ENDIAN); | |
3515 | } | |
3516 | ||
3517 | void stl_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3518 | { | |
3519 | address_space_stl(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3520 | } | |
3521 | ||
3522 | void stl_le_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3523 | { | |
3524 | address_space_stl_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3525 | } | |
3526 | ||
3527 | void stl_be_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3528 | { | |
3529 | address_space_stl_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3530 | } | |
3531 | ||
3532 | /* XXX: optimize */ | |
3533 | void address_space_stb(AddressSpace *as, hwaddr addr, uint32_t val, | |
3534 | MemTxAttrs attrs, MemTxResult *result) | |
3535 | { | |
3536 | uint8_t v = val; | |
3537 | MemTxResult r; | |
3538 | ||
3539 | r = address_space_rw(as, addr, attrs, &v, 1, 1); | |
3540 | if (result) { | |
3541 | *result = r; | |
3542 | } | |
3543 | } | |
3544 | ||
3545 | void stb_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3546 | { | |
3547 | address_space_stb(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3548 | } | |
3549 | ||
3550 | /* warning: addr must be aligned */ | |
3551 | static inline void address_space_stw_internal(AddressSpace *as, | |
3552 | hwaddr addr, uint32_t val, | |
3553 | MemTxAttrs attrs, | |
3554 | MemTxResult *result, | |
3555 | enum device_endian endian) | |
3556 | { | |
3557 | uint8_t *ptr; | |
3558 | MemoryRegion *mr; | |
3559 | hwaddr l = 2; | |
3560 | hwaddr addr1; | |
3561 | MemTxResult r; | |
3562 | bool release_lock = false; | |
3563 | ||
3564 | rcu_read_lock(); | |
3565 | mr = address_space_translate(as, addr, &addr1, &l, true); | |
3566 | if (l < 2 || !memory_access_is_direct(mr, true)) { | |
3567 | release_lock |= prepare_mmio_access(mr); | |
3568 | ||
3569 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3570 | if (endian == DEVICE_LITTLE_ENDIAN) { | |
3571 | val = bswap16(val); | |
3572 | } | |
3573 | #else | |
3574 | if (endian == DEVICE_BIG_ENDIAN) { | |
3575 | val = bswap16(val); | |
3576 | } | |
3577 | #endif | |
3578 | r = memory_region_dispatch_write(mr, addr1, val, 2, attrs); | |
3579 | } else { | |
3580 | /* RAM case */ | |
3581 | addr1 += memory_region_get_ram_addr(mr) & TARGET_PAGE_MASK; | |
3582 | ptr = qemu_get_ram_ptr(mr->ram_block, addr1); | |
3583 | switch (endian) { | |
3584 | case DEVICE_LITTLE_ENDIAN: | |
3585 | stw_le_p(ptr, val); | |
3586 | break; | |
3587 | case DEVICE_BIG_ENDIAN: | |
3588 | stw_be_p(ptr, val); | |
3589 | break; | |
3590 | default: | |
3591 | stw_p(ptr, val); | |
3592 | break; | |
3593 | } | |
3594 | invalidate_and_set_dirty(mr, addr1, 2); | |
3595 | r = MEMTX_OK; | |
3596 | } | |
3597 | if (result) { | |
3598 | *result = r; | |
3599 | } | |
3600 | if (release_lock) { | |
3601 | qemu_mutex_unlock_iothread(); | |
3602 | } | |
3603 | rcu_read_unlock(); | |
3604 | } | |
3605 | ||
3606 | void address_space_stw(AddressSpace *as, hwaddr addr, uint32_t val, | |
3607 | MemTxAttrs attrs, MemTxResult *result) | |
3608 | { | |
3609 | address_space_stw_internal(as, addr, val, attrs, result, | |
3610 | DEVICE_NATIVE_ENDIAN); | |
3611 | } | |
3612 | ||
3613 | void address_space_stw_le(AddressSpace *as, hwaddr addr, uint32_t val, | |
3614 | MemTxAttrs attrs, MemTxResult *result) | |
3615 | { | |
3616 | address_space_stw_internal(as, addr, val, attrs, result, | |
3617 | DEVICE_LITTLE_ENDIAN); | |
3618 | } | |
3619 | ||
3620 | void address_space_stw_be(AddressSpace *as, hwaddr addr, uint32_t val, | |
3621 | MemTxAttrs attrs, MemTxResult *result) | |
3622 | { | |
3623 | address_space_stw_internal(as, addr, val, attrs, result, | |
3624 | DEVICE_BIG_ENDIAN); | |
3625 | } | |
3626 | ||
3627 | void stw_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3628 | { | |
3629 | address_space_stw(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3630 | } | |
3631 | ||
3632 | void stw_le_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3633 | { | |
3634 | address_space_stw_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3635 | } | |
3636 | ||
3637 | void stw_be_phys(AddressSpace *as, hwaddr addr, uint32_t val) | |
3638 | { | |
3639 | address_space_stw_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3640 | } | |
3641 | ||
3642 | /* XXX: optimize */ | |
3643 | void address_space_stq(AddressSpace *as, hwaddr addr, uint64_t val, | |
3644 | MemTxAttrs attrs, MemTxResult *result) | |
3645 | { | |
3646 | MemTxResult r; | |
3647 | val = tswap64(val); | |
3648 | r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1); | |
3649 | if (result) { | |
3650 | *result = r; | |
3651 | } | |
3652 | } | |
3653 | ||
3654 | void address_space_stq_le(AddressSpace *as, hwaddr addr, uint64_t val, | |
3655 | MemTxAttrs attrs, MemTxResult *result) | |
3656 | { | |
3657 | MemTxResult r; | |
3658 | val = cpu_to_le64(val); | |
3659 | r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1); | |
3660 | if (result) { | |
3661 | *result = r; | |
3662 | } | |
3663 | } | |
3664 | void address_space_stq_be(AddressSpace *as, hwaddr addr, uint64_t val, | |
3665 | MemTxAttrs attrs, MemTxResult *result) | |
3666 | { | |
3667 | MemTxResult r; | |
3668 | val = cpu_to_be64(val); | |
3669 | r = address_space_rw(as, addr, attrs, (void *) &val, 8, 1); | |
3670 | if (result) { | |
3671 | *result = r; | |
3672 | } | |
3673 | } | |
3674 | ||
3675 | void stq_phys(AddressSpace *as, hwaddr addr, uint64_t val) | |
3676 | { | |
3677 | address_space_stq(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3678 | } | |
3679 | ||
3680 | void stq_le_phys(AddressSpace *as, hwaddr addr, uint64_t val) | |
3681 | { | |
3682 | address_space_stq_le(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3683 | } | |
3684 | ||
3685 | void stq_be_phys(AddressSpace *as, hwaddr addr, uint64_t val) | |
3686 | { | |
3687 | address_space_stq_be(as, addr, val, MEMTXATTRS_UNSPECIFIED, NULL); | |
3688 | } | |
3689 | ||
3690 | /* virtual memory access for debug (includes writing to ROM) */ | |
3691 | int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr, | |
3692 | uint8_t *buf, int len, int is_write) | |
3693 | { | |
3694 | int l; | |
3695 | hwaddr phys_addr; | |
3696 | target_ulong page; | |
3697 | ||
3698 | while (len > 0) { | |
3699 | int asidx; | |
3700 | MemTxAttrs attrs; | |
3701 | ||
3702 | page = addr & TARGET_PAGE_MASK; | |
3703 | phys_addr = cpu_get_phys_page_attrs_debug(cpu, page, &attrs); | |
3704 | asidx = cpu_asidx_from_attrs(cpu, attrs); | |
3705 | /* if no physical page mapped, return an error */ | |
3706 | if (phys_addr == -1) | |
3707 | return -1; | |
3708 | l = (page + TARGET_PAGE_SIZE) - addr; | |
3709 | if (l > len) | |
3710 | l = len; | |
3711 | phys_addr += (addr & ~TARGET_PAGE_MASK); | |
3712 | if (is_write) { | |
3713 | cpu_physical_memory_write_rom(cpu->cpu_ases[asidx].as, | |
3714 | phys_addr, buf, l); | |
3715 | } else { | |
3716 | address_space_rw(cpu->cpu_ases[asidx].as, phys_addr, | |
3717 | MEMTXATTRS_UNSPECIFIED, | |
3718 | buf, l, 0); | |
3719 | } | |
3720 | len -= l; | |
3721 | buf += l; | |
3722 | addr += l; | |
3723 | } | |
3724 | return 0; | |
3725 | } | |
3726 | ||
3727 | /* | |
3728 | * Allows code that needs to deal with migration bitmaps etc to still be built | |
3729 | * target independent. | |
3730 | */ | |
3731 | size_t qemu_target_page_bits(void) | |
3732 | { | |
3733 | return TARGET_PAGE_BITS; | |
3734 | } | |
3735 | ||
3736 | #endif | |
3737 | ||
3738 | /* | |
3739 | * A helper function for the _utterly broken_ virtio device model to find out if | |
3740 | * it's running on a big endian machine. Don't do this at home kids! | |
3741 | */ | |
3742 | bool target_words_bigendian(void); | |
3743 | bool target_words_bigendian(void) | |
3744 | { | |
3745 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3746 | return true; | |
3747 | #else | |
3748 | return false; | |
3749 | #endif | |
3750 | } | |
3751 | ||
3752 | #ifndef CONFIG_USER_ONLY | |
3753 | bool cpu_physical_memory_is_io(hwaddr phys_addr) | |
3754 | { | |
3755 | MemoryRegion*mr; | |
3756 | hwaddr l = 1; | |
3757 | bool res; | |
3758 | ||
3759 | rcu_read_lock(); | |
3760 | mr = address_space_translate(&address_space_memory, | |
3761 | phys_addr, &phys_addr, &l, false); | |
3762 | ||
3763 | res = !(memory_region_is_ram(mr) || memory_region_is_romd(mr)); | |
3764 | rcu_read_unlock(); | |
3765 | return res; | |
3766 | } | |
3767 | ||
3768 | int qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque) | |
3769 | { | |
3770 | RAMBlock *block; | |
3771 | int ret = 0; | |
3772 | ||
3773 | rcu_read_lock(); | |
3774 | QLIST_FOREACH_RCU(block, &ram_list.blocks, next) { | |
3775 | ret = func(block->idstr, block->host, block->offset, | |
3776 | block->used_length, opaque); | |
3777 | if (ret) { | |
3778 | break; | |
3779 | } | |
3780 | } | |
3781 | rcu_read_unlock(); | |
3782 | return ret; | |
3783 | } | |
3784 | #endif |