]>
Commit | Line | Data |
---|---|---|
54936004 | 1 | /* |
5b6dd868 | 2 | * Virtual page mapping |
5fafdf24 | 3 | * |
54936004 FB |
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 | |
8167ee88 | 17 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. |
54936004 | 18 | */ |
14a48c1d | 19 | |
7b31bbc2 | 20 | #include "qemu/osdep.h" |
a8d25326 | 21 | #include "qemu-common.h" |
da34e65c | 22 | #include "qapi/error.h" |
54936004 | 23 | |
f348b6d1 | 24 | #include "qemu/cutils.h" |
6180a181 | 25 | #include "cpu.h" |
63c91552 | 26 | #include "exec/exec-all.h" |
51180423 | 27 | #include "exec/target_page.h" |
b67d9a52 | 28 | #include "tcg.h" |
741da0d3 | 29 | #include "hw/qdev-core.h" |
c7e002c5 | 30 | #include "hw/qdev-properties.h" |
4485bd26 | 31 | #if !defined(CONFIG_USER_ONLY) |
47c8ca53 | 32 | #include "hw/boards.h" |
33c11879 | 33 | #include "hw/xen/xen.h" |
4485bd26 | 34 | #endif |
9c17d615 | 35 | #include "sysemu/kvm.h" |
2ff3de68 | 36 | #include "sysemu/sysemu.h" |
14a48c1d | 37 | #include "sysemu/tcg.h" |
1de7afc9 PB |
38 | #include "qemu/timer.h" |
39 | #include "qemu/config-file.h" | |
75a34036 | 40 | #include "qemu/error-report.h" |
b6b71cb5 | 41 | #include "qemu/qemu-print.h" |
53a5960a | 42 | #if defined(CONFIG_USER_ONLY) |
a9c94277 | 43 | #include "qemu.h" |
432d268c | 44 | #else /* !CONFIG_USER_ONLY */ |
741da0d3 | 45 | #include "exec/memory.h" |
df43d49c | 46 | #include "exec/ioport.h" |
741da0d3 | 47 | #include "sysemu/dma.h" |
b58c5c2d | 48 | #include "sysemu/hostmem.h" |
79ca7a1b | 49 | #include "sysemu/hw_accel.h" |
741da0d3 | 50 | #include "exec/address-spaces.h" |
9c17d615 | 51 | #include "sysemu/xen-mapcache.h" |
0ab8ed18 | 52 | #include "trace-root.h" |
d3a5038c | 53 | |
e2fa71f5 | 54 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE |
e2fa71f5 DDAG |
55 | #include <linux/falloc.h> |
56 | #endif | |
57 | ||
53a5960a | 58 | #endif |
0dc3f44a | 59 | #include "qemu/rcu_queue.h" |
4840f10e | 60 | #include "qemu/main-loop.h" |
5b6dd868 | 61 | #include "translate-all.h" |
7615936e | 62 | #include "sysemu/replay.h" |
0cac1b66 | 63 | |
022c62cb | 64 | #include "exec/memory-internal.h" |
220c3ebd | 65 | #include "exec/ram_addr.h" |
508127e2 | 66 | #include "exec/log.h" |
67d95c15 | 67 | |
9dfeca7c BR |
68 | #include "migration/vmstate.h" |
69 | ||
b35ba30f | 70 | #include "qemu/range.h" |
794e8f30 MT |
71 | #ifndef _WIN32 |
72 | #include "qemu/mmap-alloc.h" | |
73 | #endif | |
b35ba30f | 74 | |
be9b23c4 PX |
75 | #include "monitor/monitor.h" |
76 | ||
db7b5426 | 77 | //#define DEBUG_SUBPAGE |
1196be37 | 78 | |
e2eef170 | 79 | #if !defined(CONFIG_USER_ONLY) |
0dc3f44a MD |
80 | /* ram_list is read under rcu_read_lock()/rcu_read_unlock(). Writes |
81 | * are protected by the ramlist lock. | |
82 | */ | |
0d53d9fe | 83 | RAMList ram_list = { .blocks = QLIST_HEAD_INITIALIZER(ram_list.blocks) }; |
62152b8a AK |
84 | |
85 | static MemoryRegion *system_memory; | |
309cb471 | 86 | static MemoryRegion *system_io; |
62152b8a | 87 | |
f6790af6 AK |
88 | AddressSpace address_space_io; |
89 | AddressSpace address_space_memory; | |
2673a5da | 90 | |
0844e007 | 91 | MemoryRegion io_mem_rom, io_mem_notdirty; |
acc9d80b | 92 | static MemoryRegion io_mem_unassigned; |
e2eef170 | 93 | #endif |
9fa3e853 | 94 | |
20bccb82 PM |
95 | #ifdef TARGET_PAGE_BITS_VARY |
96 | int target_page_bits; | |
97 | bool target_page_bits_decided; | |
98 | #endif | |
99 | ||
f481ee2d PB |
100 | CPUTailQ cpus = QTAILQ_HEAD_INITIALIZER(cpus); |
101 | ||
6a00d601 FB |
102 | /* current CPU in the current thread. It is only valid inside |
103 | cpu_exec() */ | |
f240eb6f | 104 | __thread CPUState *current_cpu; |
2e70f6ef | 105 | /* 0 = Do not count executed instructions. |
bf20dc07 | 106 | 1 = Precise instruction counting. |
2e70f6ef | 107 | 2 = Adaptive rate instruction counting. */ |
5708fc66 | 108 | int use_icount; |
6a00d601 | 109 | |
a0be0c58 YZ |
110 | uintptr_t qemu_host_page_size; |
111 | intptr_t qemu_host_page_mask; | |
a0be0c58 | 112 | |
20bccb82 PM |
113 | bool set_preferred_target_page_bits(int bits) |
114 | { | |
115 | /* The target page size is the lowest common denominator for all | |
116 | * the CPUs in the system, so we can only make it smaller, never | |
117 | * larger. And we can't make it smaller once we've committed to | |
118 | * a particular size. | |
119 | */ | |
120 | #ifdef TARGET_PAGE_BITS_VARY | |
121 | assert(bits >= TARGET_PAGE_BITS_MIN); | |
122 | if (target_page_bits == 0 || target_page_bits > bits) { | |
123 | if (target_page_bits_decided) { | |
124 | return false; | |
125 | } | |
126 | target_page_bits = bits; | |
127 | } | |
128 | #endif | |
129 | return true; | |
130 | } | |
131 | ||
e2eef170 | 132 | #if !defined(CONFIG_USER_ONLY) |
4346ae3e | 133 | |
20bccb82 PM |
134 | static void finalize_target_page_bits(void) |
135 | { | |
136 | #ifdef TARGET_PAGE_BITS_VARY | |
137 | if (target_page_bits == 0) { | |
138 | target_page_bits = TARGET_PAGE_BITS_MIN; | |
139 | } | |
140 | target_page_bits_decided = true; | |
141 | #endif | |
142 | } | |
143 | ||
1db8abb1 PB |
144 | typedef struct PhysPageEntry PhysPageEntry; |
145 | ||
146 | struct PhysPageEntry { | |
9736e55b | 147 | /* How many bits skip to next level (in units of L2_SIZE). 0 for a leaf. */ |
8b795765 | 148 | uint32_t skip : 6; |
9736e55b | 149 | /* index into phys_sections (!skip) or phys_map_nodes (skip) */ |
8b795765 | 150 | uint32_t ptr : 26; |
1db8abb1 PB |
151 | }; |
152 | ||
8b795765 MT |
153 | #define PHYS_MAP_NODE_NIL (((uint32_t)~0) >> 6) |
154 | ||
03f49957 | 155 | /* Size of the L2 (and L3, etc) page tables. */ |
57271d63 | 156 | #define ADDR_SPACE_BITS 64 |
03f49957 | 157 | |
026736ce | 158 | #define P_L2_BITS 9 |
03f49957 PB |
159 | #define P_L2_SIZE (1 << P_L2_BITS) |
160 | ||
161 | #define P_L2_LEVELS (((ADDR_SPACE_BITS - TARGET_PAGE_BITS - 1) / P_L2_BITS) + 1) | |
162 | ||
163 | typedef PhysPageEntry Node[P_L2_SIZE]; | |
0475d94f | 164 | |
53cb28cb | 165 | typedef struct PhysPageMap { |
79e2b9ae PB |
166 | struct rcu_head rcu; |
167 | ||
53cb28cb MA |
168 | unsigned sections_nb; |
169 | unsigned sections_nb_alloc; | |
170 | unsigned nodes_nb; | |
171 | unsigned nodes_nb_alloc; | |
172 | Node *nodes; | |
173 | MemoryRegionSection *sections; | |
174 | } PhysPageMap; | |
175 | ||
1db8abb1 | 176 | struct AddressSpaceDispatch { |
729633c2 | 177 | MemoryRegionSection *mru_section; |
1db8abb1 PB |
178 | /* This is a multi-level map on the physical address space. |
179 | * The bottom level has pointers to MemoryRegionSections. | |
180 | */ | |
181 | PhysPageEntry phys_map; | |
53cb28cb | 182 | PhysPageMap map; |
1db8abb1 PB |
183 | }; |
184 | ||
90260c6c JK |
185 | #define SUBPAGE_IDX(addr) ((addr) & ~TARGET_PAGE_MASK) |
186 | typedef struct subpage_t { | |
187 | MemoryRegion iomem; | |
16620684 | 188 | FlatView *fv; |
90260c6c | 189 | hwaddr base; |
2615fabd | 190 | uint16_t sub_section[]; |
90260c6c JK |
191 | } subpage_t; |
192 | ||
b41aac4f LPF |
193 | #define PHYS_SECTION_UNASSIGNED 0 |
194 | #define PHYS_SECTION_NOTDIRTY 1 | |
195 | #define PHYS_SECTION_ROM 2 | |
5312bd8b | 196 | |
e2eef170 | 197 | static void io_mem_init(void); |
62152b8a | 198 | static void memory_map_init(void); |
9458a9a1 | 199 | static void tcg_log_global_after_sync(MemoryListener *listener); |
09daed84 | 200 | static void tcg_commit(MemoryListener *listener); |
e2eef170 | 201 | |
32857f4d PM |
202 | /** |
203 | * CPUAddressSpace: all the information a CPU needs about an AddressSpace | |
204 | * @cpu: the CPU whose AddressSpace this is | |
205 | * @as: the AddressSpace itself | |
206 | * @memory_dispatch: its dispatch pointer (cached, RCU protected) | |
207 | * @tcg_as_listener: listener for tracking changes to the AddressSpace | |
208 | */ | |
209 | struct CPUAddressSpace { | |
210 | CPUState *cpu; | |
211 | AddressSpace *as; | |
212 | struct AddressSpaceDispatch *memory_dispatch; | |
213 | MemoryListener tcg_as_listener; | |
214 | }; | |
215 | ||
8deaf12c GH |
216 | struct DirtyBitmapSnapshot { |
217 | ram_addr_t start; | |
218 | ram_addr_t end; | |
219 | unsigned long dirty[]; | |
220 | }; | |
221 | ||
6658ffb8 | 222 | #endif |
fd6ce8f6 | 223 | |
6d9a1304 | 224 | #if !defined(CONFIG_USER_ONLY) |
d6f2ea22 | 225 | |
53cb28cb | 226 | static void phys_map_node_reserve(PhysPageMap *map, unsigned nodes) |
d6f2ea22 | 227 | { |
101420b8 | 228 | static unsigned alloc_hint = 16; |
53cb28cb | 229 | if (map->nodes_nb + nodes > map->nodes_nb_alloc) { |
c95cfd04 | 230 | map->nodes_nb_alloc = MAX(alloc_hint, map->nodes_nb + nodes); |
53cb28cb | 231 | map->nodes = g_renew(Node, map->nodes, map->nodes_nb_alloc); |
101420b8 | 232 | alloc_hint = map->nodes_nb_alloc; |
d6f2ea22 | 233 | } |
f7bf5461 AK |
234 | } |
235 | ||
db94604b | 236 | static uint32_t phys_map_node_alloc(PhysPageMap *map, bool leaf) |
f7bf5461 AK |
237 | { |
238 | unsigned i; | |
8b795765 | 239 | uint32_t ret; |
db94604b PB |
240 | PhysPageEntry e; |
241 | PhysPageEntry *p; | |
f7bf5461 | 242 | |
53cb28cb | 243 | ret = map->nodes_nb++; |
db94604b | 244 | p = map->nodes[ret]; |
f7bf5461 | 245 | assert(ret != PHYS_MAP_NODE_NIL); |
53cb28cb | 246 | assert(ret != map->nodes_nb_alloc); |
db94604b PB |
247 | |
248 | e.skip = leaf ? 0 : 1; | |
249 | e.ptr = leaf ? PHYS_SECTION_UNASSIGNED : PHYS_MAP_NODE_NIL; | |
03f49957 | 250 | for (i = 0; i < P_L2_SIZE; ++i) { |
db94604b | 251 | memcpy(&p[i], &e, sizeof(e)); |
d6f2ea22 | 252 | } |
f7bf5461 | 253 | return ret; |
d6f2ea22 AK |
254 | } |
255 | ||
53cb28cb | 256 | static void phys_page_set_level(PhysPageMap *map, PhysPageEntry *lp, |
56b15076 | 257 | hwaddr *index, uint64_t *nb, uint16_t leaf, |
2999097b | 258 | int level) |
f7bf5461 AK |
259 | { |
260 | PhysPageEntry *p; | |
03f49957 | 261 | hwaddr step = (hwaddr)1 << (level * P_L2_BITS); |
108c49b8 | 262 | |
9736e55b | 263 | if (lp->skip && lp->ptr == PHYS_MAP_NODE_NIL) { |
db94604b | 264 | lp->ptr = phys_map_node_alloc(map, level == 0); |
92e873b9 | 265 | } |
db94604b | 266 | p = map->nodes[lp->ptr]; |
03f49957 | 267 | lp = &p[(*index >> (level * P_L2_BITS)) & (P_L2_SIZE - 1)]; |
f7bf5461 | 268 | |
03f49957 | 269 | while (*nb && lp < &p[P_L2_SIZE]) { |
07f07b31 | 270 | if ((*index & (step - 1)) == 0 && *nb >= step) { |
9736e55b | 271 | lp->skip = 0; |
c19e8800 | 272 | lp->ptr = leaf; |
07f07b31 AK |
273 | *index += step; |
274 | *nb -= step; | |
2999097b | 275 | } else { |
53cb28cb | 276 | phys_page_set_level(map, lp, index, nb, leaf, level - 1); |
2999097b AK |
277 | } |
278 | ++lp; | |
f7bf5461 AK |
279 | } |
280 | } | |
281 | ||
ac1970fb | 282 | static void phys_page_set(AddressSpaceDispatch *d, |
56b15076 | 283 | hwaddr index, uint64_t nb, |
2999097b | 284 | uint16_t leaf) |
f7bf5461 | 285 | { |
2999097b | 286 | /* Wildly overreserve - it doesn't matter much. */ |
53cb28cb | 287 | phys_map_node_reserve(&d->map, 3 * P_L2_LEVELS); |
5cd2c5b6 | 288 | |
53cb28cb | 289 | phys_page_set_level(&d->map, &d->phys_map, &index, &nb, leaf, P_L2_LEVELS - 1); |
92e873b9 FB |
290 | } |
291 | ||
b35ba30f MT |
292 | /* Compact a non leaf page entry. Simply detect that the entry has a single child, |
293 | * and update our entry so we can skip it and go directly to the destination. | |
294 | */ | |
efee678d | 295 | static void phys_page_compact(PhysPageEntry *lp, Node *nodes) |
b35ba30f MT |
296 | { |
297 | unsigned valid_ptr = P_L2_SIZE; | |
298 | int valid = 0; | |
299 | PhysPageEntry *p; | |
300 | int i; | |
301 | ||
302 | if (lp->ptr == PHYS_MAP_NODE_NIL) { | |
303 | return; | |
304 | } | |
305 | ||
306 | p = nodes[lp->ptr]; | |
307 | for (i = 0; i < P_L2_SIZE; i++) { | |
308 | if (p[i].ptr == PHYS_MAP_NODE_NIL) { | |
309 | continue; | |
310 | } | |
311 | ||
312 | valid_ptr = i; | |
313 | valid++; | |
314 | if (p[i].skip) { | |
efee678d | 315 | phys_page_compact(&p[i], nodes); |
b35ba30f MT |
316 | } |
317 | } | |
318 | ||
319 | /* We can only compress if there's only one child. */ | |
320 | if (valid != 1) { | |
321 | return; | |
322 | } | |
323 | ||
324 | assert(valid_ptr < P_L2_SIZE); | |
325 | ||
326 | /* Don't compress if it won't fit in the # of bits we have. */ | |
526ca236 WY |
327 | if (P_L2_LEVELS >= (1 << 6) && |
328 | lp->skip + p[valid_ptr].skip >= (1 << 6)) { | |
b35ba30f MT |
329 | return; |
330 | } | |
331 | ||
332 | lp->ptr = p[valid_ptr].ptr; | |
333 | if (!p[valid_ptr].skip) { | |
334 | /* If our only child is a leaf, make this a leaf. */ | |
335 | /* By design, we should have made this node a leaf to begin with so we | |
336 | * should never reach here. | |
337 | * But since it's so simple to handle this, let's do it just in case we | |
338 | * change this rule. | |
339 | */ | |
340 | lp->skip = 0; | |
341 | } else { | |
342 | lp->skip += p[valid_ptr].skip; | |
343 | } | |
344 | } | |
345 | ||
8629d3fc | 346 | void address_space_dispatch_compact(AddressSpaceDispatch *d) |
b35ba30f | 347 | { |
b35ba30f | 348 | if (d->phys_map.skip) { |
efee678d | 349 | phys_page_compact(&d->phys_map, d->map.nodes); |
b35ba30f MT |
350 | } |
351 | } | |
352 | ||
29cb533d FZ |
353 | static inline bool section_covers_addr(const MemoryRegionSection *section, |
354 | hwaddr addr) | |
355 | { | |
356 | /* Memory topology clips a memory region to [0, 2^64); size.hi > 0 means | |
357 | * the section must cover the entire address space. | |
358 | */ | |
258dfaaa | 359 | return int128_gethi(section->size) || |
29cb533d | 360 | range_covers_byte(section->offset_within_address_space, |
258dfaaa | 361 | int128_getlo(section->size), addr); |
29cb533d FZ |
362 | } |
363 | ||
003a0cf2 | 364 | static MemoryRegionSection *phys_page_find(AddressSpaceDispatch *d, hwaddr addr) |
92e873b9 | 365 | { |
003a0cf2 PX |
366 | PhysPageEntry lp = d->phys_map, *p; |
367 | Node *nodes = d->map.nodes; | |
368 | MemoryRegionSection *sections = d->map.sections; | |
97115a8d | 369 | hwaddr index = addr >> TARGET_PAGE_BITS; |
31ab2b4a | 370 | int i; |
f1f6e3b8 | 371 | |
9736e55b | 372 | for (i = P_L2_LEVELS; lp.skip && (i -= lp.skip) >= 0;) { |
c19e8800 | 373 | if (lp.ptr == PHYS_MAP_NODE_NIL) { |
9affd6fc | 374 | return §ions[PHYS_SECTION_UNASSIGNED]; |
31ab2b4a | 375 | } |
9affd6fc | 376 | p = nodes[lp.ptr]; |
03f49957 | 377 | lp = p[(index >> (i * P_L2_BITS)) & (P_L2_SIZE - 1)]; |
5312bd8b | 378 | } |
b35ba30f | 379 | |
29cb533d | 380 | if (section_covers_addr(§ions[lp.ptr], addr)) { |
b35ba30f MT |
381 | return §ions[lp.ptr]; |
382 | } else { | |
383 | return §ions[PHYS_SECTION_UNASSIGNED]; | |
384 | } | |
f3705d53 AK |
385 | } |
386 | ||
79e2b9ae | 387 | /* Called from RCU critical section */ |
c7086b4a | 388 | static MemoryRegionSection *address_space_lookup_region(AddressSpaceDispatch *d, |
90260c6c JK |
389 | hwaddr addr, |
390 | bool resolve_subpage) | |
9f029603 | 391 | { |
729633c2 | 392 | MemoryRegionSection *section = atomic_read(&d->mru_section); |
90260c6c JK |
393 | subpage_t *subpage; |
394 | ||
07c114bb PB |
395 | if (!section || section == &d->map.sections[PHYS_SECTION_UNASSIGNED] || |
396 | !section_covers_addr(section, addr)) { | |
003a0cf2 | 397 | section = phys_page_find(d, addr); |
07c114bb | 398 | atomic_set(&d->mru_section, section); |
729633c2 | 399 | } |
90260c6c JK |
400 | if (resolve_subpage && section->mr->subpage) { |
401 | subpage = container_of(section->mr, subpage_t, iomem); | |
53cb28cb | 402 | section = &d->map.sections[subpage->sub_section[SUBPAGE_IDX(addr)]]; |
90260c6c JK |
403 | } |
404 | return section; | |
9f029603 JK |
405 | } |
406 | ||
79e2b9ae | 407 | /* Called from RCU critical section */ |
90260c6c | 408 | static MemoryRegionSection * |
c7086b4a | 409 | address_space_translate_internal(AddressSpaceDispatch *d, hwaddr addr, hwaddr *xlat, |
90260c6c | 410 | hwaddr *plen, bool resolve_subpage) |
149f54b5 PB |
411 | { |
412 | MemoryRegionSection *section; | |
965eb2fc | 413 | MemoryRegion *mr; |
a87f3954 | 414 | Int128 diff; |
149f54b5 | 415 | |
c7086b4a | 416 | section = address_space_lookup_region(d, addr, resolve_subpage); |
149f54b5 PB |
417 | /* Compute offset within MemoryRegionSection */ |
418 | addr -= section->offset_within_address_space; | |
419 | ||
420 | /* Compute offset within MemoryRegion */ | |
421 | *xlat = addr + section->offset_within_region; | |
422 | ||
965eb2fc | 423 | mr = section->mr; |
b242e0e0 PB |
424 | |
425 | /* MMIO registers can be expected to perform full-width accesses based only | |
426 | * on their address, without considering adjacent registers that could | |
427 | * decode to completely different MemoryRegions. When such registers | |
428 | * exist (e.g. I/O ports 0xcf8 and 0xcf9 on most PC chipsets), MMIO | |
429 | * regions overlap wildly. For this reason we cannot clamp the accesses | |
430 | * here. | |
431 | * | |
432 | * If the length is small (as is the case for address_space_ldl/stl), | |
433 | * everything works fine. If the incoming length is large, however, | |
434 | * the caller really has to do the clamping through memory_access_size. | |
435 | */ | |
965eb2fc | 436 | if (memory_region_is_ram(mr)) { |
e4a511f8 | 437 | diff = int128_sub(section->size, int128_make64(addr)); |
965eb2fc PB |
438 | *plen = int128_get64(int128_min(diff, int128_make64(*plen))); |
439 | } | |
149f54b5 PB |
440 | return section; |
441 | } | |
90260c6c | 442 | |
a411c84b PB |
443 | /** |
444 | * address_space_translate_iommu - translate an address through an IOMMU | |
445 | * memory region and then through the target address space. | |
446 | * | |
447 | * @iommu_mr: the IOMMU memory region that we start the translation from | |
448 | * @addr: the address to be translated through the MMU | |
449 | * @xlat: the translated address offset within the destination memory region. | |
450 | * It cannot be %NULL. | |
451 | * @plen_out: valid read/write length of the translated address. It | |
452 | * cannot be %NULL. | |
453 | * @page_mask_out: page mask for the translated address. This | |
454 | * should only be meaningful for IOMMU translated | |
455 | * addresses, since there may be huge pages that this bit | |
456 | * would tell. It can be %NULL if we don't care about it. | |
457 | * @is_write: whether the translation operation is for write | |
458 | * @is_mmio: whether this can be MMIO, set true if it can | |
459 | * @target_as: the address space targeted by the IOMMU | |
2f7b009c | 460 | * @attrs: transaction attributes |
a411c84b PB |
461 | * |
462 | * This function is called from RCU critical section. It is the common | |
463 | * part of flatview_do_translate and address_space_translate_cached. | |
464 | */ | |
465 | static MemoryRegionSection address_space_translate_iommu(IOMMUMemoryRegion *iommu_mr, | |
466 | hwaddr *xlat, | |
467 | hwaddr *plen_out, | |
468 | hwaddr *page_mask_out, | |
469 | bool is_write, | |
470 | bool is_mmio, | |
2f7b009c PM |
471 | AddressSpace **target_as, |
472 | MemTxAttrs attrs) | |
a411c84b PB |
473 | { |
474 | MemoryRegionSection *section; | |
475 | hwaddr page_mask = (hwaddr)-1; | |
476 | ||
477 | do { | |
478 | hwaddr addr = *xlat; | |
479 | IOMMUMemoryRegionClass *imrc = memory_region_get_iommu_class_nocheck(iommu_mr); | |
2c91bcf2 PM |
480 | int iommu_idx = 0; |
481 | IOMMUTLBEntry iotlb; | |
482 | ||
483 | if (imrc->attrs_to_index) { | |
484 | iommu_idx = imrc->attrs_to_index(iommu_mr, attrs); | |
485 | } | |
486 | ||
487 | iotlb = imrc->translate(iommu_mr, addr, is_write ? | |
488 | IOMMU_WO : IOMMU_RO, iommu_idx); | |
a411c84b PB |
489 | |
490 | if (!(iotlb.perm & (1 << is_write))) { | |
491 | goto unassigned; | |
492 | } | |
493 | ||
494 | addr = ((iotlb.translated_addr & ~iotlb.addr_mask) | |
495 | | (addr & iotlb.addr_mask)); | |
496 | page_mask &= iotlb.addr_mask; | |
497 | *plen_out = MIN(*plen_out, (addr | iotlb.addr_mask) - addr + 1); | |
498 | *target_as = iotlb.target_as; | |
499 | ||
500 | section = address_space_translate_internal( | |
501 | address_space_to_dispatch(iotlb.target_as), addr, xlat, | |
502 | plen_out, is_mmio); | |
503 | ||
504 | iommu_mr = memory_region_get_iommu(section->mr); | |
505 | } while (unlikely(iommu_mr)); | |
506 | ||
507 | if (page_mask_out) { | |
508 | *page_mask_out = page_mask; | |
509 | } | |
510 | return *section; | |
511 | ||
512 | unassigned: | |
513 | return (MemoryRegionSection) { .mr = &io_mem_unassigned }; | |
514 | } | |
515 | ||
d5e5fafd PX |
516 | /** |
517 | * flatview_do_translate - translate an address in FlatView | |
518 | * | |
519 | * @fv: the flat view that we want to translate on | |
520 | * @addr: the address to be translated in above address space | |
521 | * @xlat: the translated address offset within memory region. It | |
522 | * cannot be @NULL. | |
523 | * @plen_out: valid read/write length of the translated address. It | |
524 | * can be @NULL when we don't care about it. | |
525 | * @page_mask_out: page mask for the translated address. This | |
526 | * should only be meaningful for IOMMU translated | |
527 | * addresses, since there may be huge pages that this bit | |
528 | * would tell. It can be @NULL if we don't care about it. | |
529 | * @is_write: whether the translation operation is for write | |
530 | * @is_mmio: whether this can be MMIO, set true if it can | |
ad2804d9 | 531 | * @target_as: the address space targeted by the IOMMU |
49e14aa8 | 532 | * @attrs: memory transaction attributes |
d5e5fafd PX |
533 | * |
534 | * This function is called from RCU critical section | |
535 | */ | |
16620684 AK |
536 | static MemoryRegionSection flatview_do_translate(FlatView *fv, |
537 | hwaddr addr, | |
538 | hwaddr *xlat, | |
d5e5fafd PX |
539 | hwaddr *plen_out, |
540 | hwaddr *page_mask_out, | |
16620684 AK |
541 | bool is_write, |
542 | bool is_mmio, | |
49e14aa8 PM |
543 | AddressSpace **target_as, |
544 | MemTxAttrs attrs) | |
052c8fa9 | 545 | { |
052c8fa9 | 546 | MemoryRegionSection *section; |
3df9d748 | 547 | IOMMUMemoryRegion *iommu_mr; |
d5e5fafd PX |
548 | hwaddr plen = (hwaddr)(-1); |
549 | ||
ad2804d9 PB |
550 | if (!plen_out) { |
551 | plen_out = &plen; | |
d5e5fafd | 552 | } |
052c8fa9 | 553 | |
a411c84b PB |
554 | section = address_space_translate_internal( |
555 | flatview_to_dispatch(fv), addr, xlat, | |
556 | plen_out, is_mmio); | |
052c8fa9 | 557 | |
a411c84b PB |
558 | iommu_mr = memory_region_get_iommu(section->mr); |
559 | if (unlikely(iommu_mr)) { | |
560 | return address_space_translate_iommu(iommu_mr, xlat, | |
561 | plen_out, page_mask_out, | |
562 | is_write, is_mmio, | |
2f7b009c | 563 | target_as, attrs); |
052c8fa9 | 564 | } |
d5e5fafd | 565 | if (page_mask_out) { |
a411c84b PB |
566 | /* Not behind an IOMMU, use default page size. */ |
567 | *page_mask_out = ~TARGET_PAGE_MASK; | |
d5e5fafd PX |
568 | } |
569 | ||
a764040c | 570 | return *section; |
052c8fa9 JW |
571 | } |
572 | ||
573 | /* Called from RCU critical section */ | |
a764040c | 574 | IOMMUTLBEntry address_space_get_iotlb_entry(AddressSpace *as, hwaddr addr, |
7446eb07 | 575 | bool is_write, MemTxAttrs attrs) |
90260c6c | 576 | { |
a764040c | 577 | MemoryRegionSection section; |
076a93d7 | 578 | hwaddr xlat, page_mask; |
30951157 | 579 | |
076a93d7 PX |
580 | /* |
581 | * This can never be MMIO, and we don't really care about plen, | |
582 | * but page mask. | |
583 | */ | |
584 | section = flatview_do_translate(address_space_to_flatview(as), addr, &xlat, | |
49e14aa8 PM |
585 | NULL, &page_mask, is_write, false, &as, |
586 | attrs); | |
30951157 | 587 | |
a764040c PX |
588 | /* Illegal translation */ |
589 | if (section.mr == &io_mem_unassigned) { | |
590 | goto iotlb_fail; | |
591 | } | |
30951157 | 592 | |
a764040c PX |
593 | /* Convert memory region offset into address space offset */ |
594 | xlat += section.offset_within_address_space - | |
595 | section.offset_within_region; | |
596 | ||
a764040c | 597 | return (IOMMUTLBEntry) { |
e76bb18f | 598 | .target_as = as, |
076a93d7 PX |
599 | .iova = addr & ~page_mask, |
600 | .translated_addr = xlat & ~page_mask, | |
601 | .addr_mask = page_mask, | |
a764040c PX |
602 | /* IOTLBs are for DMAs, and DMA only allows on RAMs. */ |
603 | .perm = IOMMU_RW, | |
604 | }; | |
605 | ||
606 | iotlb_fail: | |
607 | return (IOMMUTLBEntry) {0}; | |
608 | } | |
609 | ||
610 | /* Called from RCU critical section */ | |
16620684 | 611 | MemoryRegion *flatview_translate(FlatView *fv, hwaddr addr, hwaddr *xlat, |
efa99a2f PM |
612 | hwaddr *plen, bool is_write, |
613 | MemTxAttrs attrs) | |
a764040c PX |
614 | { |
615 | MemoryRegion *mr; | |
616 | MemoryRegionSection section; | |
16620684 | 617 | AddressSpace *as = NULL; |
a764040c PX |
618 | |
619 | /* This can be MMIO, so setup MMIO bit. */ | |
d5e5fafd | 620 | section = flatview_do_translate(fv, addr, xlat, plen, NULL, |
49e14aa8 | 621 | is_write, true, &as, attrs); |
a764040c PX |
622 | mr = section.mr; |
623 | ||
fe680d0d | 624 | if (xen_enabled() && memory_access_is_direct(mr, is_write)) { |
a87f3954 | 625 | hwaddr page = ((addr & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE) - addr; |
23820dbf | 626 | *plen = MIN(page, *plen); |
a87f3954 PB |
627 | } |
628 | ||
30951157 | 629 | return mr; |
90260c6c JK |
630 | } |
631 | ||
1f871c5e PM |
632 | typedef struct TCGIOMMUNotifier { |
633 | IOMMUNotifier n; | |
634 | MemoryRegion *mr; | |
635 | CPUState *cpu; | |
636 | int iommu_idx; | |
637 | bool active; | |
638 | } TCGIOMMUNotifier; | |
639 | ||
640 | static void tcg_iommu_unmap_notify(IOMMUNotifier *n, IOMMUTLBEntry *iotlb) | |
641 | { | |
642 | TCGIOMMUNotifier *notifier = container_of(n, TCGIOMMUNotifier, n); | |
643 | ||
644 | if (!notifier->active) { | |
645 | return; | |
646 | } | |
647 | tlb_flush(notifier->cpu); | |
648 | notifier->active = false; | |
649 | /* We leave the notifier struct on the list to avoid reallocating it later. | |
650 | * Generally the number of IOMMUs a CPU deals with will be small. | |
651 | * In any case we can't unregister the iommu notifier from a notify | |
652 | * callback. | |
653 | */ | |
654 | } | |
655 | ||
656 | static void tcg_register_iommu_notifier(CPUState *cpu, | |
657 | IOMMUMemoryRegion *iommu_mr, | |
658 | int iommu_idx) | |
659 | { | |
660 | /* Make sure this CPU has an IOMMU notifier registered for this | |
661 | * IOMMU/IOMMU index combination, so that we can flush its TLB | |
662 | * when the IOMMU tells us the mappings we've cached have changed. | |
663 | */ | |
664 | MemoryRegion *mr = MEMORY_REGION(iommu_mr); | |
665 | TCGIOMMUNotifier *notifier; | |
666 | int i; | |
667 | ||
668 | for (i = 0; i < cpu->iommu_notifiers->len; i++) { | |
5601be3b | 669 | notifier = g_array_index(cpu->iommu_notifiers, TCGIOMMUNotifier *, i); |
1f871c5e PM |
670 | if (notifier->mr == mr && notifier->iommu_idx == iommu_idx) { |
671 | break; | |
672 | } | |
673 | } | |
674 | if (i == cpu->iommu_notifiers->len) { | |
675 | /* Not found, add a new entry at the end of the array */ | |
676 | cpu->iommu_notifiers = g_array_set_size(cpu->iommu_notifiers, i + 1); | |
5601be3b PM |
677 | notifier = g_new0(TCGIOMMUNotifier, 1); |
678 | g_array_index(cpu->iommu_notifiers, TCGIOMMUNotifier *, i) = notifier; | |
1f871c5e PM |
679 | |
680 | notifier->mr = mr; | |
681 | notifier->iommu_idx = iommu_idx; | |
682 | notifier->cpu = cpu; | |
683 | /* Rather than trying to register interest in the specific part | |
684 | * of the iommu's address space that we've accessed and then | |
685 | * expand it later as subsequent accesses touch more of it, we | |
686 | * just register interest in the whole thing, on the assumption | |
687 | * that iommu reconfiguration will be rare. | |
688 | */ | |
689 | iommu_notifier_init(¬ifier->n, | |
690 | tcg_iommu_unmap_notify, | |
691 | IOMMU_NOTIFIER_UNMAP, | |
692 | 0, | |
693 | HWADDR_MAX, | |
694 | iommu_idx); | |
695 | memory_region_register_iommu_notifier(notifier->mr, ¬ifier->n); | |
696 | } | |
697 | ||
698 | if (!notifier->active) { | |
699 | notifier->active = true; | |
700 | } | |
701 | } | |
702 | ||
703 | static void tcg_iommu_free_notifier_list(CPUState *cpu) | |
704 | { | |
705 | /* Destroy the CPU's notifier list */ | |
706 | int i; | |
707 | TCGIOMMUNotifier *notifier; | |
708 | ||
709 | for (i = 0; i < cpu->iommu_notifiers->len; i++) { | |
5601be3b | 710 | notifier = g_array_index(cpu->iommu_notifiers, TCGIOMMUNotifier *, i); |
1f871c5e | 711 | memory_region_unregister_iommu_notifier(notifier->mr, ¬ifier->n); |
5601be3b | 712 | g_free(notifier); |
1f871c5e PM |
713 | } |
714 | g_array_free(cpu->iommu_notifiers, true); | |
715 | } | |
716 | ||
79e2b9ae | 717 | /* Called from RCU critical section */ |
90260c6c | 718 | MemoryRegionSection * |
d7898cda | 719 | address_space_translate_for_iotlb(CPUState *cpu, int asidx, hwaddr addr, |
1f871c5e PM |
720 | hwaddr *xlat, hwaddr *plen, |
721 | MemTxAttrs attrs, int *prot) | |
90260c6c | 722 | { |
30951157 | 723 | MemoryRegionSection *section; |
1f871c5e PM |
724 | IOMMUMemoryRegion *iommu_mr; |
725 | IOMMUMemoryRegionClass *imrc; | |
726 | IOMMUTLBEntry iotlb; | |
727 | int iommu_idx; | |
f35e44e7 | 728 | AddressSpaceDispatch *d = atomic_rcu_read(&cpu->cpu_ases[asidx].memory_dispatch); |
d7898cda | 729 | |
1f871c5e PM |
730 | for (;;) { |
731 | section = address_space_translate_internal(d, addr, &addr, plen, false); | |
732 | ||
733 | iommu_mr = memory_region_get_iommu(section->mr); | |
734 | if (!iommu_mr) { | |
735 | break; | |
736 | } | |
737 | ||
738 | imrc = memory_region_get_iommu_class_nocheck(iommu_mr); | |
739 | ||
740 | iommu_idx = imrc->attrs_to_index(iommu_mr, attrs); | |
741 | tcg_register_iommu_notifier(cpu, iommu_mr, iommu_idx); | |
742 | /* We need all the permissions, so pass IOMMU_NONE so the IOMMU | |
743 | * doesn't short-cut its translation table walk. | |
744 | */ | |
745 | iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, iommu_idx); | |
746 | addr = ((iotlb.translated_addr & ~iotlb.addr_mask) | |
747 | | (addr & iotlb.addr_mask)); | |
748 | /* Update the caller's prot bits to remove permissions the IOMMU | |
749 | * is giving us a failure response for. If we get down to no | |
750 | * permissions left at all we can give up now. | |
751 | */ | |
752 | if (!(iotlb.perm & IOMMU_RO)) { | |
753 | *prot &= ~(PAGE_READ | PAGE_EXEC); | |
754 | } | |
755 | if (!(iotlb.perm & IOMMU_WO)) { | |
756 | *prot &= ~PAGE_WRITE; | |
757 | } | |
758 | ||
759 | if (!*prot) { | |
760 | goto translate_fail; | |
761 | } | |
762 | ||
763 | d = flatview_to_dispatch(address_space_to_flatview(iotlb.target_as)); | |
764 | } | |
30951157 | 765 | |
3df9d748 | 766 | assert(!memory_region_is_iommu(section->mr)); |
1f871c5e | 767 | *xlat = addr; |
30951157 | 768 | return section; |
1f871c5e PM |
769 | |
770 | translate_fail: | |
771 | return &d->map.sections[PHYS_SECTION_UNASSIGNED]; | |
90260c6c | 772 | } |
5b6dd868 | 773 | #endif |
fd6ce8f6 | 774 | |
b170fce3 | 775 | #if !defined(CONFIG_USER_ONLY) |
5b6dd868 BS |
776 | |
777 | static int cpu_common_post_load(void *opaque, int version_id) | |
fd6ce8f6 | 778 | { |
259186a7 | 779 | CPUState *cpu = opaque; |
a513fe19 | 780 | |
5b6dd868 BS |
781 | /* 0x01 was CPU_INTERRUPT_EXIT. This line can be removed when the |
782 | version_id is increased. */ | |
259186a7 | 783 | cpu->interrupt_request &= ~0x01; |
d10eb08f | 784 | tlb_flush(cpu); |
5b6dd868 | 785 | |
15a356c4 PD |
786 | /* loadvm has just updated the content of RAM, bypassing the |
787 | * usual mechanisms that ensure we flush TBs for writes to | |
788 | * memory we've translated code from. So we must flush all TBs, | |
789 | * which will now be stale. | |
790 | */ | |
791 | tb_flush(cpu); | |
792 | ||
5b6dd868 | 793 | return 0; |
a513fe19 | 794 | } |
7501267e | 795 | |
6c3bff0e PD |
796 | static int cpu_common_pre_load(void *opaque) |
797 | { | |
798 | CPUState *cpu = opaque; | |
799 | ||
adee6424 | 800 | cpu->exception_index = -1; |
6c3bff0e PD |
801 | |
802 | return 0; | |
803 | } | |
804 | ||
805 | static bool cpu_common_exception_index_needed(void *opaque) | |
806 | { | |
807 | CPUState *cpu = opaque; | |
808 | ||
adee6424 | 809 | return tcg_enabled() && cpu->exception_index != -1; |
6c3bff0e PD |
810 | } |
811 | ||
812 | static const VMStateDescription vmstate_cpu_common_exception_index = { | |
813 | .name = "cpu_common/exception_index", | |
814 | .version_id = 1, | |
815 | .minimum_version_id = 1, | |
5cd8cada | 816 | .needed = cpu_common_exception_index_needed, |
6c3bff0e PD |
817 | .fields = (VMStateField[]) { |
818 | VMSTATE_INT32(exception_index, CPUState), | |
819 | VMSTATE_END_OF_LIST() | |
820 | } | |
821 | }; | |
822 | ||
bac05aa9 AS |
823 | static bool cpu_common_crash_occurred_needed(void *opaque) |
824 | { | |
825 | CPUState *cpu = opaque; | |
826 | ||
827 | return cpu->crash_occurred; | |
828 | } | |
829 | ||
830 | static const VMStateDescription vmstate_cpu_common_crash_occurred = { | |
831 | .name = "cpu_common/crash_occurred", | |
832 | .version_id = 1, | |
833 | .minimum_version_id = 1, | |
834 | .needed = cpu_common_crash_occurred_needed, | |
835 | .fields = (VMStateField[]) { | |
836 | VMSTATE_BOOL(crash_occurred, CPUState), | |
837 | VMSTATE_END_OF_LIST() | |
838 | } | |
839 | }; | |
840 | ||
1a1562f5 | 841 | const VMStateDescription vmstate_cpu_common = { |
5b6dd868 BS |
842 | .name = "cpu_common", |
843 | .version_id = 1, | |
844 | .minimum_version_id = 1, | |
6c3bff0e | 845 | .pre_load = cpu_common_pre_load, |
5b6dd868 | 846 | .post_load = cpu_common_post_load, |
35d08458 | 847 | .fields = (VMStateField[]) { |
259186a7 AF |
848 | VMSTATE_UINT32(halted, CPUState), |
849 | VMSTATE_UINT32(interrupt_request, CPUState), | |
5b6dd868 | 850 | VMSTATE_END_OF_LIST() |
6c3bff0e | 851 | }, |
5cd8cada JQ |
852 | .subsections = (const VMStateDescription*[]) { |
853 | &vmstate_cpu_common_exception_index, | |
bac05aa9 | 854 | &vmstate_cpu_common_crash_occurred, |
5cd8cada | 855 | NULL |
5b6dd868 BS |
856 | } |
857 | }; | |
1a1562f5 | 858 | |
5b6dd868 | 859 | #endif |
ea041c0e | 860 | |
38d8f5c8 | 861 | CPUState *qemu_get_cpu(int index) |
ea041c0e | 862 | { |
bdc44640 | 863 | CPUState *cpu; |
ea041c0e | 864 | |
bdc44640 | 865 | CPU_FOREACH(cpu) { |
55e5c285 | 866 | if (cpu->cpu_index == index) { |
bdc44640 | 867 | return cpu; |
55e5c285 | 868 | } |
ea041c0e | 869 | } |
5b6dd868 | 870 | |
bdc44640 | 871 | return NULL; |
ea041c0e FB |
872 | } |
873 | ||
09daed84 | 874 | #if !defined(CONFIG_USER_ONLY) |
80ceb07a PX |
875 | void cpu_address_space_init(CPUState *cpu, int asidx, |
876 | const char *prefix, MemoryRegion *mr) | |
09daed84 | 877 | { |
12ebc9a7 | 878 | CPUAddressSpace *newas; |
80ceb07a | 879 | AddressSpace *as = g_new0(AddressSpace, 1); |
87a621d8 | 880 | char *as_name; |
80ceb07a PX |
881 | |
882 | assert(mr); | |
87a621d8 PX |
883 | as_name = g_strdup_printf("%s-%d", prefix, cpu->cpu_index); |
884 | address_space_init(as, mr, as_name); | |
885 | g_free(as_name); | |
12ebc9a7 PM |
886 | |
887 | /* Target code should have set num_ases before calling us */ | |
888 | assert(asidx < cpu->num_ases); | |
889 | ||
56943e8c PM |
890 | if (asidx == 0) { |
891 | /* address space 0 gets the convenience alias */ | |
892 | cpu->as = as; | |
893 | } | |
894 | ||
12ebc9a7 PM |
895 | /* KVM cannot currently support multiple address spaces. */ |
896 | assert(asidx == 0 || !kvm_enabled()); | |
09daed84 | 897 | |
12ebc9a7 PM |
898 | if (!cpu->cpu_ases) { |
899 | cpu->cpu_ases = g_new0(CPUAddressSpace, cpu->num_ases); | |
09daed84 | 900 | } |
32857f4d | 901 | |
12ebc9a7 PM |
902 | newas = &cpu->cpu_ases[asidx]; |
903 | newas->cpu = cpu; | |
904 | newas->as = as; | |
56943e8c | 905 | if (tcg_enabled()) { |
9458a9a1 | 906 | newas->tcg_as_listener.log_global_after_sync = tcg_log_global_after_sync; |
12ebc9a7 PM |
907 | newas->tcg_as_listener.commit = tcg_commit; |
908 | memory_listener_register(&newas->tcg_as_listener, as); | |
56943e8c | 909 | } |
09daed84 | 910 | } |
651a5bc0 PM |
911 | |
912 | AddressSpace *cpu_get_address_space(CPUState *cpu, int asidx) | |
913 | { | |
914 | /* Return the AddressSpace corresponding to the specified index */ | |
915 | return cpu->cpu_ases[asidx].as; | |
916 | } | |
09daed84 EI |
917 | #endif |
918 | ||
7bbc124e | 919 | void cpu_exec_unrealizefn(CPUState *cpu) |
1c59eb39 | 920 | { |
9dfeca7c BR |
921 | CPUClass *cc = CPU_GET_CLASS(cpu); |
922 | ||
267f685b | 923 | cpu_list_remove(cpu); |
9dfeca7c BR |
924 | |
925 | if (cc->vmsd != NULL) { | |
926 | vmstate_unregister(NULL, cc->vmsd, cpu); | |
927 | } | |
928 | if (qdev_get_vmsd(DEVICE(cpu)) == NULL) { | |
929 | vmstate_unregister(NULL, &vmstate_cpu_common, cpu); | |
930 | } | |
1f871c5e PM |
931 | #ifndef CONFIG_USER_ONLY |
932 | tcg_iommu_free_notifier_list(cpu); | |
933 | #endif | |
1c59eb39 BR |
934 | } |
935 | ||
c7e002c5 FZ |
936 | Property cpu_common_props[] = { |
937 | #ifndef CONFIG_USER_ONLY | |
938 | /* Create a memory property for softmmu CPU object, | |
2e5b09fd | 939 | * so users can wire up its memory. (This can't go in hw/core/cpu.c |
c7e002c5 FZ |
940 | * because that file is compiled only once for both user-mode |
941 | * and system builds.) The default if no link is set up is to use | |
942 | * the system address space. | |
943 | */ | |
944 | DEFINE_PROP_LINK("memory", CPUState, memory, TYPE_MEMORY_REGION, | |
945 | MemoryRegion *), | |
946 | #endif | |
947 | DEFINE_PROP_END_OF_LIST(), | |
948 | }; | |
949 | ||
39e329e3 | 950 | void cpu_exec_initfn(CPUState *cpu) |
ea041c0e | 951 | { |
56943e8c | 952 | cpu->as = NULL; |
12ebc9a7 | 953 | cpu->num_ases = 0; |
56943e8c | 954 | |
291135b5 | 955 | #ifndef CONFIG_USER_ONLY |
291135b5 | 956 | cpu->thread_id = qemu_get_thread_id(); |
6731d864 PC |
957 | cpu->memory = system_memory; |
958 | object_ref(OBJECT(cpu->memory)); | |
291135b5 | 959 | #endif |
39e329e3 LV |
960 | } |
961 | ||
ce5b1bbf | 962 | void cpu_exec_realizefn(CPUState *cpu, Error **errp) |
39e329e3 | 963 | { |
55c3ceef | 964 | CPUClass *cc = CPU_GET_CLASS(cpu); |
2dda6354 | 965 | static bool tcg_target_initialized; |
291135b5 | 966 | |
267f685b | 967 | cpu_list_add(cpu); |
1bc7e522 | 968 | |
2dda6354 EC |
969 | if (tcg_enabled() && !tcg_target_initialized) { |
970 | tcg_target_initialized = true; | |
55c3ceef RH |
971 | cc->tcg_initialize(); |
972 | } | |
5005e253 | 973 | tlb_init(cpu); |
55c3ceef | 974 | |
1bc7e522 | 975 | #ifndef CONFIG_USER_ONLY |
e0d47944 | 976 | if (qdev_get_vmsd(DEVICE(cpu)) == NULL) { |
741da0d3 | 977 | vmstate_register(NULL, cpu->cpu_index, &vmstate_cpu_common, cpu); |
e0d47944 | 978 | } |
b170fce3 | 979 | if (cc->vmsd != NULL) { |
741da0d3 | 980 | vmstate_register(NULL, cpu->cpu_index, cc->vmsd, cpu); |
b170fce3 | 981 | } |
1f871c5e | 982 | |
5601be3b | 983 | cpu->iommu_notifiers = g_array_new(false, true, sizeof(TCGIOMMUNotifier *)); |
741da0d3 | 984 | #endif |
ea041c0e FB |
985 | } |
986 | ||
c1c8cfe5 | 987 | const char *parse_cpu_option(const char *cpu_option) |
2278b939 IM |
988 | { |
989 | ObjectClass *oc; | |
990 | CPUClass *cc; | |
991 | gchar **model_pieces; | |
992 | const char *cpu_type; | |
993 | ||
c1c8cfe5 | 994 | model_pieces = g_strsplit(cpu_option, ",", 2); |
5b863f3e EH |
995 | if (!model_pieces[0]) { |
996 | error_report("-cpu option cannot be empty"); | |
997 | exit(1); | |
998 | } | |
2278b939 IM |
999 | |
1000 | oc = cpu_class_by_name(CPU_RESOLVING_TYPE, model_pieces[0]); | |
1001 | if (oc == NULL) { | |
1002 | error_report("unable to find CPU model '%s'", model_pieces[0]); | |
1003 | g_strfreev(model_pieces); | |
1004 | exit(EXIT_FAILURE); | |
1005 | } | |
1006 | ||
1007 | cpu_type = object_class_get_name(oc); | |
1008 | cc = CPU_CLASS(oc); | |
1009 | cc->parse_features(cpu_type, model_pieces[1], &error_fatal); | |
1010 | g_strfreev(model_pieces); | |
1011 | return cpu_type; | |
1012 | } | |
1013 | ||
c40d4792 | 1014 | #if defined(CONFIG_USER_ONLY) |
8bca9a03 | 1015 | void tb_invalidate_phys_addr(target_ulong addr) |
1e7855a5 | 1016 | { |
406bc339 | 1017 | mmap_lock(); |
8bca9a03 | 1018 | tb_invalidate_phys_page_range(addr, addr + 1, 0); |
406bc339 PK |
1019 | mmap_unlock(); |
1020 | } | |
8bca9a03 PB |
1021 | |
1022 | static void breakpoint_invalidate(CPUState *cpu, target_ulong pc) | |
1023 | { | |
1024 | tb_invalidate_phys_addr(pc); | |
1025 | } | |
406bc339 | 1026 | #else |
8bca9a03 PB |
1027 | void tb_invalidate_phys_addr(AddressSpace *as, hwaddr addr, MemTxAttrs attrs) |
1028 | { | |
1029 | ram_addr_t ram_addr; | |
1030 | MemoryRegion *mr; | |
1031 | hwaddr l = 1; | |
1032 | ||
c40d4792 PB |
1033 | if (!tcg_enabled()) { |
1034 | return; | |
1035 | } | |
1036 | ||
8bca9a03 PB |
1037 | rcu_read_lock(); |
1038 | mr = address_space_translate(as, addr, &addr, &l, false, attrs); | |
1039 | if (!(memory_region_is_ram(mr) | |
1040 | || memory_region_is_romd(mr))) { | |
1041 | rcu_read_unlock(); | |
1042 | return; | |
1043 | } | |
1044 | ram_addr = memory_region_get_ram_addr(mr) + addr; | |
1045 | tb_invalidate_phys_page_range(ram_addr, ram_addr + 1, 0); | |
1046 | rcu_read_unlock(); | |
1047 | } | |
1048 | ||
406bc339 PK |
1049 | static void breakpoint_invalidate(CPUState *cpu, target_ulong pc) |
1050 | { | |
1051 | MemTxAttrs attrs; | |
1052 | hwaddr phys = cpu_get_phys_page_attrs_debug(cpu, pc, &attrs); | |
1053 | int asidx = cpu_asidx_from_attrs(cpu, attrs); | |
1054 | if (phys != -1) { | |
1055 | /* Locks grabbed by tb_invalidate_phys_addr */ | |
1056 | tb_invalidate_phys_addr(cpu->cpu_ases[asidx].as, | |
c874dc4f | 1057 | phys | (pc & ~TARGET_PAGE_MASK), attrs); |
406bc339 | 1058 | } |
1e7855a5 | 1059 | } |
406bc339 | 1060 | #endif |
d720b93d | 1061 | |
74841f04 | 1062 | #ifndef CONFIG_USER_ONLY |
6658ffb8 | 1063 | /* Add a watchpoint. */ |
75a34036 | 1064 | int cpu_watchpoint_insert(CPUState *cpu, vaddr addr, vaddr len, |
a1d1bb31 | 1065 | int flags, CPUWatchpoint **watchpoint) |
6658ffb8 | 1066 | { |
c0ce998e | 1067 | CPUWatchpoint *wp; |
6658ffb8 | 1068 | |
05068c0d | 1069 | /* forbid ranges which are empty or run off the end of the address space */ |
07e2863d | 1070 | if (len == 0 || (addr + len - 1) < addr) { |
75a34036 AF |
1071 | error_report("tried to set invalid watchpoint at %" |
1072 | VADDR_PRIx ", len=%" VADDR_PRIu, addr, len); | |
b4051334 AL |
1073 | return -EINVAL; |
1074 | } | |
7267c094 | 1075 | wp = g_malloc(sizeof(*wp)); |
a1d1bb31 AL |
1076 | |
1077 | wp->vaddr = addr; | |
05068c0d | 1078 | wp->len = len; |
a1d1bb31 AL |
1079 | wp->flags = flags; |
1080 | ||
2dc9f411 | 1081 | /* keep all GDB-injected watchpoints in front */ |
ff4700b0 AF |
1082 | if (flags & BP_GDB) { |
1083 | QTAILQ_INSERT_HEAD(&cpu->watchpoints, wp, entry); | |
1084 | } else { | |
1085 | QTAILQ_INSERT_TAIL(&cpu->watchpoints, wp, entry); | |
1086 | } | |
6658ffb8 | 1087 | |
31b030d4 | 1088 | tlb_flush_page(cpu, addr); |
a1d1bb31 AL |
1089 | |
1090 | if (watchpoint) | |
1091 | *watchpoint = wp; | |
1092 | return 0; | |
6658ffb8 PB |
1093 | } |
1094 | ||
a1d1bb31 | 1095 | /* Remove a specific watchpoint. */ |
75a34036 | 1096 | int cpu_watchpoint_remove(CPUState *cpu, vaddr addr, vaddr len, |
a1d1bb31 | 1097 | int flags) |
6658ffb8 | 1098 | { |
a1d1bb31 | 1099 | CPUWatchpoint *wp; |
6658ffb8 | 1100 | |
ff4700b0 | 1101 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { |
05068c0d | 1102 | if (addr == wp->vaddr && len == wp->len |
6e140f28 | 1103 | && flags == (wp->flags & ~BP_WATCHPOINT_HIT)) { |
75a34036 | 1104 | cpu_watchpoint_remove_by_ref(cpu, wp); |
6658ffb8 PB |
1105 | return 0; |
1106 | } | |
1107 | } | |
a1d1bb31 | 1108 | return -ENOENT; |
6658ffb8 PB |
1109 | } |
1110 | ||
a1d1bb31 | 1111 | /* Remove a specific watchpoint by reference. */ |
75a34036 | 1112 | void cpu_watchpoint_remove_by_ref(CPUState *cpu, CPUWatchpoint *watchpoint) |
a1d1bb31 | 1113 | { |
ff4700b0 | 1114 | QTAILQ_REMOVE(&cpu->watchpoints, watchpoint, entry); |
7d03f82f | 1115 | |
31b030d4 | 1116 | tlb_flush_page(cpu, watchpoint->vaddr); |
a1d1bb31 | 1117 | |
7267c094 | 1118 | g_free(watchpoint); |
a1d1bb31 AL |
1119 | } |
1120 | ||
1121 | /* Remove all matching watchpoints. */ | |
75a34036 | 1122 | void cpu_watchpoint_remove_all(CPUState *cpu, int mask) |
a1d1bb31 | 1123 | { |
c0ce998e | 1124 | CPUWatchpoint *wp, *next; |
a1d1bb31 | 1125 | |
ff4700b0 | 1126 | QTAILQ_FOREACH_SAFE(wp, &cpu->watchpoints, entry, next) { |
75a34036 AF |
1127 | if (wp->flags & mask) { |
1128 | cpu_watchpoint_remove_by_ref(cpu, wp); | |
1129 | } | |
c0ce998e | 1130 | } |
7d03f82f | 1131 | } |
05068c0d PM |
1132 | |
1133 | /* Return true if this watchpoint address matches the specified | |
1134 | * access (ie the address range covered by the watchpoint overlaps | |
1135 | * partially or completely with the address range covered by the | |
1136 | * access). | |
1137 | */ | |
56ad8b00 RH |
1138 | static inline bool watchpoint_address_matches(CPUWatchpoint *wp, |
1139 | vaddr addr, vaddr len) | |
05068c0d PM |
1140 | { |
1141 | /* We know the lengths are non-zero, but a little caution is | |
1142 | * required to avoid errors in the case where the range ends | |
1143 | * exactly at the top of the address space and so addr + len | |
1144 | * wraps round to zero. | |
1145 | */ | |
1146 | vaddr wpend = wp->vaddr + wp->len - 1; | |
1147 | vaddr addrend = addr + len - 1; | |
1148 | ||
1149 | return !(addr > wpend || wp->vaddr > addrend); | |
1150 | } | |
1151 | ||
56ad8b00 RH |
1152 | /* Return flags for watchpoints that match addr + prot. */ |
1153 | int cpu_watchpoint_address_matches(CPUState *cpu, vaddr addr, vaddr len) | |
1154 | { | |
1155 | CPUWatchpoint *wp; | |
1156 | int ret = 0; | |
1157 | ||
1158 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { | |
1159 | if (watchpoint_address_matches(wp, addr, TARGET_PAGE_SIZE)) { | |
1160 | ret |= wp->flags; | |
1161 | } | |
1162 | } | |
1163 | return ret; | |
1164 | } | |
74841f04 | 1165 | #endif /* !CONFIG_USER_ONLY */ |
7d03f82f | 1166 | |
a1d1bb31 | 1167 | /* Add a breakpoint. */ |
b3310ab3 | 1168 | int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags, |
a1d1bb31 | 1169 | CPUBreakpoint **breakpoint) |
4c3a88a2 | 1170 | { |
c0ce998e | 1171 | CPUBreakpoint *bp; |
3b46e624 | 1172 | |
7267c094 | 1173 | bp = g_malloc(sizeof(*bp)); |
4c3a88a2 | 1174 | |
a1d1bb31 AL |
1175 | bp->pc = pc; |
1176 | bp->flags = flags; | |
1177 | ||
2dc9f411 | 1178 | /* keep all GDB-injected breakpoints in front */ |
00b941e5 | 1179 | if (flags & BP_GDB) { |
f0c3c505 | 1180 | QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry); |
00b941e5 | 1181 | } else { |
f0c3c505 | 1182 | QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry); |
00b941e5 | 1183 | } |
3b46e624 | 1184 | |
f0c3c505 | 1185 | breakpoint_invalidate(cpu, pc); |
a1d1bb31 | 1186 | |
00b941e5 | 1187 | if (breakpoint) { |
a1d1bb31 | 1188 | *breakpoint = bp; |
00b941e5 | 1189 | } |
4c3a88a2 | 1190 | return 0; |
4c3a88a2 FB |
1191 | } |
1192 | ||
a1d1bb31 | 1193 | /* Remove a specific breakpoint. */ |
b3310ab3 | 1194 | int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags) |
a1d1bb31 | 1195 | { |
a1d1bb31 AL |
1196 | CPUBreakpoint *bp; |
1197 | ||
f0c3c505 | 1198 | QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) { |
a1d1bb31 | 1199 | if (bp->pc == pc && bp->flags == flags) { |
b3310ab3 | 1200 | cpu_breakpoint_remove_by_ref(cpu, bp); |
a1d1bb31 AL |
1201 | return 0; |
1202 | } | |
7d03f82f | 1203 | } |
a1d1bb31 | 1204 | return -ENOENT; |
7d03f82f EI |
1205 | } |
1206 | ||
a1d1bb31 | 1207 | /* Remove a specific breakpoint by reference. */ |
b3310ab3 | 1208 | void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *breakpoint) |
4c3a88a2 | 1209 | { |
f0c3c505 AF |
1210 | QTAILQ_REMOVE(&cpu->breakpoints, breakpoint, entry); |
1211 | ||
1212 | breakpoint_invalidate(cpu, breakpoint->pc); | |
a1d1bb31 | 1213 | |
7267c094 | 1214 | g_free(breakpoint); |
a1d1bb31 AL |
1215 | } |
1216 | ||
1217 | /* Remove all matching breakpoints. */ | |
b3310ab3 | 1218 | void cpu_breakpoint_remove_all(CPUState *cpu, int mask) |
a1d1bb31 | 1219 | { |
c0ce998e | 1220 | CPUBreakpoint *bp, *next; |
a1d1bb31 | 1221 | |
f0c3c505 | 1222 | QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) { |
b3310ab3 AF |
1223 | if (bp->flags & mask) { |
1224 | cpu_breakpoint_remove_by_ref(cpu, bp); | |
1225 | } | |
c0ce998e | 1226 | } |
4c3a88a2 FB |
1227 | } |
1228 | ||
c33a346e FB |
1229 | /* enable or disable single step mode. EXCP_DEBUG is returned by the |
1230 | CPU loop after each instruction */ | |
3825b28f | 1231 | void cpu_single_step(CPUState *cpu, int enabled) |
c33a346e | 1232 | { |
ed2803da AF |
1233 | if (cpu->singlestep_enabled != enabled) { |
1234 | cpu->singlestep_enabled = enabled; | |
1235 | if (kvm_enabled()) { | |
38e478ec | 1236 | kvm_update_guest_debug(cpu, 0); |
ed2803da | 1237 | } else { |
ccbb4d44 | 1238 | /* must flush all the translated code to avoid inconsistencies */ |
e22a25c9 | 1239 | /* XXX: only flush what is necessary */ |
bbd77c18 | 1240 | tb_flush(cpu); |
e22a25c9 | 1241 | } |
c33a346e | 1242 | } |
c33a346e FB |
1243 | } |
1244 | ||
a47dddd7 | 1245 | void cpu_abort(CPUState *cpu, const char *fmt, ...) |
7501267e FB |
1246 | { |
1247 | va_list ap; | |
493ae1f0 | 1248 | va_list ap2; |
7501267e FB |
1249 | |
1250 | va_start(ap, fmt); | |
493ae1f0 | 1251 | va_copy(ap2, ap); |
7501267e FB |
1252 | fprintf(stderr, "qemu: fatal: "); |
1253 | vfprintf(stderr, fmt, ap); | |
1254 | fprintf(stderr, "\n"); | |
90c84c56 | 1255 | cpu_dump_state(cpu, stderr, CPU_DUMP_FPU | CPU_DUMP_CCOP); |
013a2942 | 1256 | if (qemu_log_separate()) { |
1ee73216 | 1257 | qemu_log_lock(); |
93fcfe39 AL |
1258 | qemu_log("qemu: fatal: "); |
1259 | qemu_log_vprintf(fmt, ap2); | |
1260 | qemu_log("\n"); | |
a0762859 | 1261 | log_cpu_state(cpu, CPU_DUMP_FPU | CPU_DUMP_CCOP); |
31b1a7b4 | 1262 | qemu_log_flush(); |
1ee73216 | 1263 | qemu_log_unlock(); |
93fcfe39 | 1264 | qemu_log_close(); |
924edcae | 1265 | } |
493ae1f0 | 1266 | va_end(ap2); |
f9373291 | 1267 | va_end(ap); |
7615936e | 1268 | replay_finish(); |
fd052bf6 RV |
1269 | #if defined(CONFIG_USER_ONLY) |
1270 | { | |
1271 | struct sigaction act; | |
1272 | sigfillset(&act.sa_mask); | |
1273 | act.sa_handler = SIG_DFL; | |
8347c185 | 1274 | act.sa_flags = 0; |
fd052bf6 RV |
1275 | sigaction(SIGABRT, &act, NULL); |
1276 | } | |
1277 | #endif | |
7501267e FB |
1278 | abort(); |
1279 | } | |
1280 | ||
0124311e | 1281 | #if !defined(CONFIG_USER_ONLY) |
0dc3f44a | 1282 | /* Called from RCU critical section */ |
041603fe PB |
1283 | static RAMBlock *qemu_get_ram_block(ram_addr_t addr) |
1284 | { | |
1285 | RAMBlock *block; | |
1286 | ||
43771539 | 1287 | block = atomic_rcu_read(&ram_list.mru_block); |
9b8424d5 | 1288 | if (block && addr - block->offset < block->max_length) { |
68851b98 | 1289 | return block; |
041603fe | 1290 | } |
99e15582 | 1291 | RAMBLOCK_FOREACH(block) { |
9b8424d5 | 1292 | if (addr - block->offset < block->max_length) { |
041603fe PB |
1293 | goto found; |
1294 | } | |
1295 | } | |
1296 | ||
1297 | fprintf(stderr, "Bad ram offset %" PRIx64 "\n", (uint64_t)addr); | |
1298 | abort(); | |
1299 | ||
1300 | found: | |
43771539 PB |
1301 | /* It is safe to write mru_block outside the iothread lock. This |
1302 | * is what happens: | |
1303 | * | |
1304 | * mru_block = xxx | |
1305 | * rcu_read_unlock() | |
1306 | * xxx removed from list | |
1307 | * rcu_read_lock() | |
1308 | * read mru_block | |
1309 | * mru_block = NULL; | |
1310 | * call_rcu(reclaim_ramblock, xxx); | |
1311 | * rcu_read_unlock() | |
1312 | * | |
1313 | * atomic_rcu_set is not needed here. The block was already published | |
1314 | * when it was placed into the list. Here we're just making an extra | |
1315 | * copy of the pointer. | |
1316 | */ | |
041603fe PB |
1317 | ram_list.mru_block = block; |
1318 | return block; | |
1319 | } | |
1320 | ||
a2f4d5be | 1321 | static void tlb_reset_dirty_range_all(ram_addr_t start, ram_addr_t length) |
d24981d3 | 1322 | { |
9a13565d | 1323 | CPUState *cpu; |
041603fe | 1324 | ram_addr_t start1; |
a2f4d5be JQ |
1325 | RAMBlock *block; |
1326 | ram_addr_t end; | |
1327 | ||
f28d0dfd | 1328 | assert(tcg_enabled()); |
a2f4d5be JQ |
1329 | end = TARGET_PAGE_ALIGN(start + length); |
1330 | start &= TARGET_PAGE_MASK; | |
d24981d3 | 1331 | |
0dc3f44a | 1332 | rcu_read_lock(); |
041603fe PB |
1333 | block = qemu_get_ram_block(start); |
1334 | assert(block == qemu_get_ram_block(end - 1)); | |
1240be24 | 1335 | start1 = (uintptr_t)ramblock_ptr(block, start - block->offset); |
9a13565d PC |
1336 | CPU_FOREACH(cpu) { |
1337 | tlb_reset_dirty(cpu, start1, length); | |
1338 | } | |
0dc3f44a | 1339 | rcu_read_unlock(); |
d24981d3 JQ |
1340 | } |
1341 | ||
5579c7f3 | 1342 | /* Note: start and end must be within the same ram block. */ |
03eebc9e SH |
1343 | bool cpu_physical_memory_test_and_clear_dirty(ram_addr_t start, |
1344 | ram_addr_t length, | |
1345 | unsigned client) | |
1ccde1cb | 1346 | { |
5b82b703 | 1347 | DirtyMemoryBlocks *blocks; |
03eebc9e | 1348 | unsigned long end, page; |
5b82b703 | 1349 | bool dirty = false; |
077874e0 PX |
1350 | RAMBlock *ramblock; |
1351 | uint64_t mr_offset, mr_size; | |
03eebc9e SH |
1352 | |
1353 | if (length == 0) { | |
1354 | return false; | |
1355 | } | |
f23db169 | 1356 | |
03eebc9e SH |
1357 | end = TARGET_PAGE_ALIGN(start + length) >> TARGET_PAGE_BITS; |
1358 | page = start >> TARGET_PAGE_BITS; | |
5b82b703 SH |
1359 | |
1360 | rcu_read_lock(); | |
1361 | ||
1362 | blocks = atomic_rcu_read(&ram_list.dirty_memory[client]); | |
077874e0 PX |
1363 | ramblock = qemu_get_ram_block(start); |
1364 | /* Range sanity check on the ramblock */ | |
1365 | assert(start >= ramblock->offset && | |
1366 | start + length <= ramblock->offset + ramblock->used_length); | |
5b82b703 SH |
1367 | |
1368 | while (page < end) { | |
1369 | unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE; | |
1370 | unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE; | |
1371 | unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset); | |
1372 | ||
1373 | dirty |= bitmap_test_and_clear_atomic(blocks->blocks[idx], | |
1374 | offset, num); | |
1375 | page += num; | |
1376 | } | |
1377 | ||
077874e0 PX |
1378 | mr_offset = (ram_addr_t)(page << TARGET_PAGE_BITS) - ramblock->offset; |
1379 | mr_size = (end - page) << TARGET_PAGE_BITS; | |
1380 | memory_region_clear_dirty_bitmap(ramblock->mr, mr_offset, mr_size); | |
1381 | ||
5b82b703 | 1382 | rcu_read_unlock(); |
03eebc9e SH |
1383 | |
1384 | if (dirty && tcg_enabled()) { | |
a2f4d5be | 1385 | tlb_reset_dirty_range_all(start, length); |
5579c7f3 | 1386 | } |
03eebc9e SH |
1387 | |
1388 | return dirty; | |
1ccde1cb FB |
1389 | } |
1390 | ||
8deaf12c | 1391 | DirtyBitmapSnapshot *cpu_physical_memory_snapshot_and_clear_dirty |
5dea4079 | 1392 | (MemoryRegion *mr, hwaddr offset, hwaddr length, unsigned client) |
8deaf12c GH |
1393 | { |
1394 | DirtyMemoryBlocks *blocks; | |
5dea4079 | 1395 | ram_addr_t start = memory_region_get_ram_addr(mr) + offset; |
8deaf12c GH |
1396 | unsigned long align = 1UL << (TARGET_PAGE_BITS + BITS_PER_LEVEL); |
1397 | ram_addr_t first = QEMU_ALIGN_DOWN(start, align); | |
1398 | ram_addr_t last = QEMU_ALIGN_UP(start + length, align); | |
1399 | DirtyBitmapSnapshot *snap; | |
1400 | unsigned long page, end, dest; | |
1401 | ||
1402 | snap = g_malloc0(sizeof(*snap) + | |
1403 | ((last - first) >> (TARGET_PAGE_BITS + 3))); | |
1404 | snap->start = first; | |
1405 | snap->end = last; | |
1406 | ||
1407 | page = first >> TARGET_PAGE_BITS; | |
1408 | end = last >> TARGET_PAGE_BITS; | |
1409 | dest = 0; | |
1410 | ||
1411 | rcu_read_lock(); | |
1412 | ||
1413 | blocks = atomic_rcu_read(&ram_list.dirty_memory[client]); | |
1414 | ||
1415 | while (page < end) { | |
1416 | unsigned long idx = page / DIRTY_MEMORY_BLOCK_SIZE; | |
1417 | unsigned long offset = page % DIRTY_MEMORY_BLOCK_SIZE; | |
1418 | unsigned long num = MIN(end - page, DIRTY_MEMORY_BLOCK_SIZE - offset); | |
1419 | ||
1420 | assert(QEMU_IS_ALIGNED(offset, (1 << BITS_PER_LEVEL))); | |
1421 | assert(QEMU_IS_ALIGNED(num, (1 << BITS_PER_LEVEL))); | |
1422 | offset >>= BITS_PER_LEVEL; | |
1423 | ||
1424 | bitmap_copy_and_clear_atomic(snap->dirty + dest, | |
1425 | blocks->blocks[idx] + offset, | |
1426 | num); | |
1427 | page += num; | |
1428 | dest += num >> BITS_PER_LEVEL; | |
1429 | } | |
1430 | ||
1431 | rcu_read_unlock(); | |
1432 | ||
1433 | if (tcg_enabled()) { | |
1434 | tlb_reset_dirty_range_all(start, length); | |
1435 | } | |
1436 | ||
077874e0 PX |
1437 | memory_region_clear_dirty_bitmap(mr, offset, length); |
1438 | ||
8deaf12c GH |
1439 | return snap; |
1440 | } | |
1441 | ||
1442 | bool cpu_physical_memory_snapshot_get_dirty(DirtyBitmapSnapshot *snap, | |
1443 | ram_addr_t start, | |
1444 | ram_addr_t length) | |
1445 | { | |
1446 | unsigned long page, end; | |
1447 | ||
1448 | assert(start >= snap->start); | |
1449 | assert(start + length <= snap->end); | |
1450 | ||
1451 | end = TARGET_PAGE_ALIGN(start + length - snap->start) >> TARGET_PAGE_BITS; | |
1452 | page = (start - snap->start) >> TARGET_PAGE_BITS; | |
1453 | ||
1454 | while (page < end) { | |
1455 | if (test_bit(page, snap->dirty)) { | |
1456 | return true; | |
1457 | } | |
1458 | page++; | |
1459 | } | |
1460 | return false; | |
1461 | } | |
1462 | ||
79e2b9ae | 1463 | /* Called from RCU critical section */ |
bb0e627a | 1464 | hwaddr memory_region_section_get_iotlb(CPUState *cpu, |
149f54b5 PB |
1465 | MemoryRegionSection *section, |
1466 | target_ulong vaddr, | |
1467 | hwaddr paddr, hwaddr xlat, | |
1468 | int prot, | |
1469 | target_ulong *address) | |
e5548617 | 1470 | { |
a8170e5e | 1471 | hwaddr iotlb; |
e5548617 | 1472 | |
cc5bea60 | 1473 | if (memory_region_is_ram(section->mr)) { |
e5548617 | 1474 | /* Normal RAM. */ |
e4e69794 | 1475 | iotlb = memory_region_get_ram_addr(section->mr) + xlat; |
e5548617 | 1476 | if (!section->readonly) { |
b41aac4f | 1477 | iotlb |= PHYS_SECTION_NOTDIRTY; |
e5548617 | 1478 | } else { |
b41aac4f | 1479 | iotlb |= PHYS_SECTION_ROM; |
e5548617 BS |
1480 | } |
1481 | } else { | |
0b8e2c10 PM |
1482 | AddressSpaceDispatch *d; |
1483 | ||
16620684 | 1484 | d = flatview_to_dispatch(section->fv); |
0b8e2c10 | 1485 | iotlb = section - d->map.sections; |
149f54b5 | 1486 | iotlb += xlat; |
e5548617 BS |
1487 | } |
1488 | ||
e5548617 BS |
1489 | return iotlb; |
1490 | } | |
9fa3e853 FB |
1491 | #endif /* defined(CONFIG_USER_ONLY) */ |
1492 | ||
e2eef170 | 1493 | #if !defined(CONFIG_USER_ONLY) |
8da3ff18 | 1494 | |
b797ab1a WY |
1495 | static int subpage_register(subpage_t *mmio, uint32_t start, uint32_t end, |
1496 | uint16_t section); | |
16620684 | 1497 | static subpage_t *subpage_init(FlatView *fv, hwaddr base); |
54688b1e | 1498 | |
06329cce | 1499 | static void *(*phys_mem_alloc)(size_t size, uint64_t *align, bool shared) = |
a2b257d6 | 1500 | qemu_anon_ram_alloc; |
91138037 MA |
1501 | |
1502 | /* | |
1503 | * Set a custom physical guest memory alloator. | |
1504 | * Accelerators with unusual needs may need this. Hopefully, we can | |
1505 | * get rid of it eventually. | |
1506 | */ | |
06329cce | 1507 | void phys_mem_set_alloc(void *(*alloc)(size_t, uint64_t *align, bool shared)) |
91138037 MA |
1508 | { |
1509 | phys_mem_alloc = alloc; | |
1510 | } | |
1511 | ||
53cb28cb MA |
1512 | static uint16_t phys_section_add(PhysPageMap *map, |
1513 | MemoryRegionSection *section) | |
5312bd8b | 1514 | { |
68f3f65b PB |
1515 | /* The physical section number is ORed with a page-aligned |
1516 | * pointer to produce the iotlb entries. Thus it should | |
1517 | * never overflow into the page-aligned value. | |
1518 | */ | |
53cb28cb | 1519 | assert(map->sections_nb < TARGET_PAGE_SIZE); |
68f3f65b | 1520 | |
53cb28cb MA |
1521 | if (map->sections_nb == map->sections_nb_alloc) { |
1522 | map->sections_nb_alloc = MAX(map->sections_nb_alloc * 2, 16); | |
1523 | map->sections = g_renew(MemoryRegionSection, map->sections, | |
1524 | map->sections_nb_alloc); | |
5312bd8b | 1525 | } |
53cb28cb | 1526 | map->sections[map->sections_nb] = *section; |
dfde4e6e | 1527 | memory_region_ref(section->mr); |
53cb28cb | 1528 | return map->sections_nb++; |
5312bd8b AK |
1529 | } |
1530 | ||
058bc4b5 PB |
1531 | static void phys_section_destroy(MemoryRegion *mr) |
1532 | { | |
55b4e80b DS |
1533 | bool have_sub_page = mr->subpage; |
1534 | ||
dfde4e6e PB |
1535 | memory_region_unref(mr); |
1536 | ||
55b4e80b | 1537 | if (have_sub_page) { |
058bc4b5 | 1538 | subpage_t *subpage = container_of(mr, subpage_t, iomem); |
b4fefef9 | 1539 | object_unref(OBJECT(&subpage->iomem)); |
058bc4b5 PB |
1540 | g_free(subpage); |
1541 | } | |
1542 | } | |
1543 | ||
6092666e | 1544 | static void phys_sections_free(PhysPageMap *map) |
5312bd8b | 1545 | { |
9affd6fc PB |
1546 | while (map->sections_nb > 0) { |
1547 | MemoryRegionSection *section = &map->sections[--map->sections_nb]; | |
058bc4b5 PB |
1548 | phys_section_destroy(section->mr); |
1549 | } | |
9affd6fc PB |
1550 | g_free(map->sections); |
1551 | g_free(map->nodes); | |
5312bd8b AK |
1552 | } |
1553 | ||
9950322a | 1554 | static void register_subpage(FlatView *fv, MemoryRegionSection *section) |
0f0cb164 | 1555 | { |
9950322a | 1556 | AddressSpaceDispatch *d = flatview_to_dispatch(fv); |
0f0cb164 | 1557 | subpage_t *subpage; |
a8170e5e | 1558 | hwaddr base = section->offset_within_address_space |
0f0cb164 | 1559 | & TARGET_PAGE_MASK; |
003a0cf2 | 1560 | MemoryRegionSection *existing = phys_page_find(d, base); |
0f0cb164 AK |
1561 | MemoryRegionSection subsection = { |
1562 | .offset_within_address_space = base, | |
052e87b0 | 1563 | .size = int128_make64(TARGET_PAGE_SIZE), |
0f0cb164 | 1564 | }; |
a8170e5e | 1565 | hwaddr start, end; |
0f0cb164 | 1566 | |
f3705d53 | 1567 | assert(existing->mr->subpage || existing->mr == &io_mem_unassigned); |
0f0cb164 | 1568 | |
f3705d53 | 1569 | if (!(existing->mr->subpage)) { |
16620684 AK |
1570 | subpage = subpage_init(fv, base); |
1571 | subsection.fv = fv; | |
0f0cb164 | 1572 | subsection.mr = &subpage->iomem; |
ac1970fb | 1573 | phys_page_set(d, base >> TARGET_PAGE_BITS, 1, |
53cb28cb | 1574 | phys_section_add(&d->map, &subsection)); |
0f0cb164 | 1575 | } else { |
f3705d53 | 1576 | subpage = container_of(existing->mr, subpage_t, iomem); |
0f0cb164 AK |
1577 | } |
1578 | start = section->offset_within_address_space & ~TARGET_PAGE_MASK; | |
052e87b0 | 1579 | end = start + int128_get64(section->size) - 1; |
53cb28cb MA |
1580 | subpage_register(subpage, start, end, |
1581 | phys_section_add(&d->map, section)); | |
0f0cb164 AK |
1582 | } |
1583 | ||
1584 | ||
9950322a | 1585 | static void register_multipage(FlatView *fv, |
052e87b0 | 1586 | MemoryRegionSection *section) |
33417e70 | 1587 | { |
9950322a | 1588 | AddressSpaceDispatch *d = flatview_to_dispatch(fv); |
a8170e5e | 1589 | hwaddr start_addr = section->offset_within_address_space; |
53cb28cb | 1590 | uint16_t section_index = phys_section_add(&d->map, section); |
052e87b0 PB |
1591 | uint64_t num_pages = int128_get64(int128_rshift(section->size, |
1592 | TARGET_PAGE_BITS)); | |
dd81124b | 1593 | |
733d5ef5 PB |
1594 | assert(num_pages); |
1595 | phys_page_set(d, start_addr >> TARGET_PAGE_BITS, num_pages, section_index); | |
33417e70 FB |
1596 | } |
1597 | ||
494d1997 WY |
1598 | /* |
1599 | * The range in *section* may look like this: | |
1600 | * | |
1601 | * |s|PPPPPPP|s| | |
1602 | * | |
1603 | * where s stands for subpage and P for page. | |
1604 | */ | |
8629d3fc | 1605 | void flatview_add_to_dispatch(FlatView *fv, MemoryRegionSection *section) |
0f0cb164 | 1606 | { |
494d1997 | 1607 | MemoryRegionSection remain = *section; |
052e87b0 | 1608 | Int128 page_size = int128_make64(TARGET_PAGE_SIZE); |
0f0cb164 | 1609 | |
494d1997 WY |
1610 | /* register first subpage */ |
1611 | if (remain.offset_within_address_space & ~TARGET_PAGE_MASK) { | |
1612 | uint64_t left = TARGET_PAGE_ALIGN(remain.offset_within_address_space) | |
1613 | - remain.offset_within_address_space; | |
733d5ef5 | 1614 | |
494d1997 | 1615 | MemoryRegionSection now = remain; |
052e87b0 | 1616 | now.size = int128_min(int128_make64(left), now.size); |
9950322a | 1617 | register_subpage(fv, &now); |
494d1997 WY |
1618 | if (int128_eq(remain.size, now.size)) { |
1619 | return; | |
1620 | } | |
052e87b0 PB |
1621 | remain.size = int128_sub(remain.size, now.size); |
1622 | remain.offset_within_address_space += int128_get64(now.size); | |
1623 | remain.offset_within_region += int128_get64(now.size); | |
494d1997 WY |
1624 | } |
1625 | ||
1626 | /* register whole pages */ | |
1627 | if (int128_ge(remain.size, page_size)) { | |
1628 | MemoryRegionSection now = remain; | |
1629 | now.size = int128_and(now.size, int128_neg(page_size)); | |
1630 | register_multipage(fv, &now); | |
1631 | if (int128_eq(remain.size, now.size)) { | |
1632 | return; | |
69b67646 | 1633 | } |
494d1997 WY |
1634 | remain.size = int128_sub(remain.size, now.size); |
1635 | remain.offset_within_address_space += int128_get64(now.size); | |
1636 | remain.offset_within_region += int128_get64(now.size); | |
0f0cb164 | 1637 | } |
494d1997 WY |
1638 | |
1639 | /* register last subpage */ | |
1640 | register_subpage(fv, &remain); | |
0f0cb164 AK |
1641 | } |
1642 | ||
62a2744c SY |
1643 | void qemu_flush_coalesced_mmio_buffer(void) |
1644 | { | |
1645 | if (kvm_enabled()) | |
1646 | kvm_flush_coalesced_mmio_buffer(); | |
1647 | } | |
1648 | ||
b2a8658e UD |
1649 | void qemu_mutex_lock_ramlist(void) |
1650 | { | |
1651 | qemu_mutex_lock(&ram_list.mutex); | |
1652 | } | |
1653 | ||
1654 | void qemu_mutex_unlock_ramlist(void) | |
1655 | { | |
1656 | qemu_mutex_unlock(&ram_list.mutex); | |
1657 | } | |
1658 | ||
be9b23c4 PX |
1659 | void ram_block_dump(Monitor *mon) |
1660 | { | |
1661 | RAMBlock *block; | |
1662 | char *psize; | |
1663 | ||
1664 | rcu_read_lock(); | |
1665 | monitor_printf(mon, "%24s %8s %18s %18s %18s\n", | |
1666 | "Block Name", "PSize", "Offset", "Used", "Total"); | |
1667 | RAMBLOCK_FOREACH(block) { | |
1668 | psize = size_to_str(block->page_size); | |
1669 | monitor_printf(mon, "%24s %8s 0x%016" PRIx64 " 0x%016" PRIx64 | |
1670 | " 0x%016" PRIx64 "\n", block->idstr, psize, | |
1671 | (uint64_t)block->offset, | |
1672 | (uint64_t)block->used_length, | |
1673 | (uint64_t)block->max_length); | |
1674 | g_free(psize); | |
1675 | } | |
1676 | rcu_read_unlock(); | |
1677 | } | |
1678 | ||
9c607668 AK |
1679 | #ifdef __linux__ |
1680 | /* | |
1681 | * FIXME TOCTTOU: this iterates over memory backends' mem-path, which | |
1682 | * may or may not name the same files / on the same filesystem now as | |
1683 | * when we actually open and map them. Iterate over the file | |
1684 | * descriptors instead, and use qemu_fd_getpagesize(). | |
1685 | */ | |
905b7ee4 | 1686 | static int find_min_backend_pagesize(Object *obj, void *opaque) |
9c607668 | 1687 | { |
9c607668 AK |
1688 | long *hpsize_min = opaque; |
1689 | ||
1690 | if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) { | |
7d5489e6 DG |
1691 | HostMemoryBackend *backend = MEMORY_BACKEND(obj); |
1692 | long hpsize = host_memory_backend_pagesize(backend); | |
2b108085 | 1693 | |
7d5489e6 | 1694 | if (host_memory_backend_is_mapped(backend) && (hpsize < *hpsize_min)) { |
0de6e2a3 | 1695 | *hpsize_min = hpsize; |
9c607668 AK |
1696 | } |
1697 | } | |
1698 | ||
1699 | return 0; | |
1700 | } | |
1701 | ||
905b7ee4 DH |
1702 | static int find_max_backend_pagesize(Object *obj, void *opaque) |
1703 | { | |
1704 | long *hpsize_max = opaque; | |
1705 | ||
1706 | if (object_dynamic_cast(obj, TYPE_MEMORY_BACKEND)) { | |
1707 | HostMemoryBackend *backend = MEMORY_BACKEND(obj); | |
1708 | long hpsize = host_memory_backend_pagesize(backend); | |
1709 | ||
1710 | if (host_memory_backend_is_mapped(backend) && (hpsize > *hpsize_max)) { | |
1711 | *hpsize_max = hpsize; | |
1712 | } | |
1713 | } | |
1714 | ||
1715 | return 0; | |
1716 | } | |
1717 | ||
1718 | /* | |
1719 | * TODO: We assume right now that all mapped host memory backends are | |
1720 | * used as RAM, however some might be used for different purposes. | |
1721 | */ | |
1722 | long qemu_minrampagesize(void) | |
9c607668 AK |
1723 | { |
1724 | long hpsize = LONG_MAX; | |
1725 | long mainrampagesize; | |
1726 | Object *memdev_root; | |
aa570207 | 1727 | MachineState *ms = MACHINE(qdev_get_machine()); |
9c607668 | 1728 | |
0de6e2a3 | 1729 | mainrampagesize = qemu_mempath_getpagesize(mem_path); |
9c607668 AK |
1730 | |
1731 | /* it's possible we have memory-backend objects with | |
1732 | * hugepage-backed RAM. these may get mapped into system | |
1733 | * address space via -numa parameters or memory hotplug | |
1734 | * hooks. we want to take these into account, but we | |
1735 | * also want to make sure these supported hugepage | |
1736 | * sizes are applicable across the entire range of memory | |
1737 | * we may boot from, so we take the min across all | |
1738 | * backends, and assume normal pages in cases where a | |
1739 | * backend isn't backed by hugepages. | |
1740 | */ | |
1741 | memdev_root = object_resolve_path("/objects", NULL); | |
1742 | if (memdev_root) { | |
905b7ee4 | 1743 | object_child_foreach(memdev_root, find_min_backend_pagesize, &hpsize); |
9c607668 AK |
1744 | } |
1745 | if (hpsize == LONG_MAX) { | |
1746 | /* No additional memory regions found ==> Report main RAM page size */ | |
1747 | return mainrampagesize; | |
1748 | } | |
1749 | ||
1750 | /* If NUMA is disabled or the NUMA nodes are not backed with a | |
1751 | * memory-backend, then there is at least one node using "normal" RAM, | |
1752 | * so if its page size is smaller we have got to report that size instead. | |
1753 | */ | |
1754 | if (hpsize > mainrampagesize && | |
aa570207 TX |
1755 | (ms->numa_state == NULL || |
1756 | ms->numa_state->num_nodes == 0 || | |
7e721e7b | 1757 | ms->numa_state->nodes[0].node_memdev == NULL)) { |
9c607668 AK |
1758 | static bool warned; |
1759 | if (!warned) { | |
1760 | error_report("Huge page support disabled (n/a for main memory)."); | |
1761 | warned = true; | |
1762 | } | |
1763 | return mainrampagesize; | |
1764 | } | |
1765 | ||
1766 | return hpsize; | |
1767 | } | |
905b7ee4 DH |
1768 | |
1769 | long qemu_maxrampagesize(void) | |
1770 | { | |
1771 | long pagesize = qemu_mempath_getpagesize(mem_path); | |
1772 | Object *memdev_root = object_resolve_path("/objects", NULL); | |
1773 | ||
1774 | if (memdev_root) { | |
1775 | object_child_foreach(memdev_root, find_max_backend_pagesize, | |
1776 | &pagesize); | |
1777 | } | |
1778 | return pagesize; | |
1779 | } | |
9c607668 | 1780 | #else |
905b7ee4 DH |
1781 | long qemu_minrampagesize(void) |
1782 | { | |
1783 | return getpagesize(); | |
1784 | } | |
1785 | long qemu_maxrampagesize(void) | |
9c607668 AK |
1786 | { |
1787 | return getpagesize(); | |
1788 | } | |
1789 | #endif | |
1790 | ||
d5dbde46 | 1791 | #ifdef CONFIG_POSIX |
d6af99c9 HZ |
1792 | static int64_t get_file_size(int fd) |
1793 | { | |
72d41eb4 SH |
1794 | int64_t size; |
1795 | #if defined(__linux__) | |
1796 | struct stat st; | |
1797 | ||
1798 | if (fstat(fd, &st) < 0) { | |
1799 | return -errno; | |
1800 | } | |
1801 | ||
1802 | /* Special handling for devdax character devices */ | |
1803 | if (S_ISCHR(st.st_mode)) { | |
1804 | g_autofree char *subsystem_path = NULL; | |
1805 | g_autofree char *subsystem = NULL; | |
1806 | ||
1807 | subsystem_path = g_strdup_printf("/sys/dev/char/%d:%d/subsystem", | |
1808 | major(st.st_rdev), minor(st.st_rdev)); | |
1809 | subsystem = g_file_read_link(subsystem_path, NULL); | |
1810 | ||
1811 | if (subsystem && g_str_has_suffix(subsystem, "/dax")) { | |
1812 | g_autofree char *size_path = NULL; | |
1813 | g_autofree char *size_str = NULL; | |
1814 | ||
1815 | size_path = g_strdup_printf("/sys/dev/char/%d:%d/size", | |
1816 | major(st.st_rdev), minor(st.st_rdev)); | |
1817 | ||
1818 | if (g_file_get_contents(size_path, &size_str, NULL, NULL)) { | |
1819 | return g_ascii_strtoll(size_str, NULL, 0); | |
1820 | } | |
1821 | } | |
1822 | } | |
1823 | #endif /* defined(__linux__) */ | |
1824 | ||
1825 | /* st.st_size may be zero for special files yet lseek(2) works */ | |
1826 | size = lseek(fd, 0, SEEK_END); | |
d6af99c9 HZ |
1827 | if (size < 0) { |
1828 | return -errno; | |
1829 | } | |
1830 | return size; | |
1831 | } | |
1832 | ||
8d37b030 MAL |
1833 | static int file_ram_open(const char *path, |
1834 | const char *region_name, | |
1835 | bool *created, | |
1836 | Error **errp) | |
c902760f MT |
1837 | { |
1838 | char *filename; | |
8ca761f6 PF |
1839 | char *sanitized_name; |
1840 | char *c; | |
5c3ece79 | 1841 | int fd = -1; |
c902760f | 1842 | |
8d37b030 | 1843 | *created = false; |
fd97fd44 MA |
1844 | for (;;) { |
1845 | fd = open(path, O_RDWR); | |
1846 | if (fd >= 0) { | |
1847 | /* @path names an existing file, use it */ | |
1848 | break; | |
8d31d6b6 | 1849 | } |
fd97fd44 MA |
1850 | if (errno == ENOENT) { |
1851 | /* @path names a file that doesn't exist, create it */ | |
1852 | fd = open(path, O_RDWR | O_CREAT | O_EXCL, 0644); | |
1853 | if (fd >= 0) { | |
8d37b030 | 1854 | *created = true; |
fd97fd44 MA |
1855 | break; |
1856 | } | |
1857 | } else if (errno == EISDIR) { | |
1858 | /* @path names a directory, create a file there */ | |
1859 | /* Make name safe to use with mkstemp by replacing '/' with '_'. */ | |
8d37b030 | 1860 | sanitized_name = g_strdup(region_name); |
fd97fd44 MA |
1861 | for (c = sanitized_name; *c != '\0'; c++) { |
1862 | if (*c == '/') { | |
1863 | *c = '_'; | |
1864 | } | |
1865 | } | |
8ca761f6 | 1866 | |
fd97fd44 MA |
1867 | filename = g_strdup_printf("%s/qemu_back_mem.%s.XXXXXX", path, |
1868 | sanitized_name); | |
1869 | g_free(sanitized_name); | |
8d31d6b6 | 1870 | |
fd97fd44 MA |
1871 | fd = mkstemp(filename); |
1872 | if (fd >= 0) { | |
1873 | unlink(filename); | |
1874 | g_free(filename); | |
1875 | break; | |
1876 | } | |
1877 | g_free(filename); | |
8d31d6b6 | 1878 | } |
fd97fd44 MA |
1879 | if (errno != EEXIST && errno != EINTR) { |
1880 | error_setg_errno(errp, errno, | |
1881 | "can't open backing store %s for guest RAM", | |
1882 | path); | |
8d37b030 | 1883 | return -1; |
fd97fd44 MA |
1884 | } |
1885 | /* | |
1886 | * Try again on EINTR and EEXIST. The latter happens when | |
1887 | * something else creates the file between our two open(). | |
1888 | */ | |
8d31d6b6 | 1889 | } |
c902760f | 1890 | |
8d37b030 MAL |
1891 | return fd; |
1892 | } | |
1893 | ||
1894 | static void *file_ram_alloc(RAMBlock *block, | |
1895 | ram_addr_t memory, | |
1896 | int fd, | |
1897 | bool truncate, | |
1898 | Error **errp) | |
1899 | { | |
5cc8767d | 1900 | MachineState *ms = MACHINE(qdev_get_machine()); |
8d37b030 MAL |
1901 | void *area; |
1902 | ||
863e9621 | 1903 | block->page_size = qemu_fd_getpagesize(fd); |
98376843 HZ |
1904 | if (block->mr->align % block->page_size) { |
1905 | error_setg(errp, "alignment 0x%" PRIx64 | |
1906 | " must be multiples of page size 0x%zx", | |
1907 | block->mr->align, block->page_size); | |
1908 | return NULL; | |
61362b71 DH |
1909 | } else if (block->mr->align && !is_power_of_2(block->mr->align)) { |
1910 | error_setg(errp, "alignment 0x%" PRIx64 | |
1911 | " must be a power of two", block->mr->align); | |
1912 | return NULL; | |
98376843 HZ |
1913 | } |
1914 | block->mr->align = MAX(block->page_size, block->mr->align); | |
8360668e HZ |
1915 | #if defined(__s390x__) |
1916 | if (kvm_enabled()) { | |
1917 | block->mr->align = MAX(block->mr->align, QEMU_VMALLOC_ALIGN); | |
1918 | } | |
1919 | #endif | |
fd97fd44 | 1920 | |
863e9621 | 1921 | if (memory < block->page_size) { |
fd97fd44 | 1922 | error_setg(errp, "memory size 0x" RAM_ADDR_FMT " must be equal to " |
863e9621 DDAG |
1923 | "or larger than page size 0x%zx", |
1924 | memory, block->page_size); | |
8d37b030 | 1925 | return NULL; |
1775f111 HZ |
1926 | } |
1927 | ||
863e9621 | 1928 | memory = ROUND_UP(memory, block->page_size); |
c902760f MT |
1929 | |
1930 | /* | |
1931 | * ftruncate is not supported by hugetlbfs in older | |
1932 | * hosts, so don't bother bailing out on errors. | |
1933 | * If anything goes wrong with it under other filesystems, | |
1934 | * mmap will fail. | |
d6af99c9 HZ |
1935 | * |
1936 | * Do not truncate the non-empty backend file to avoid corrupting | |
1937 | * the existing data in the file. Disabling shrinking is not | |
1938 | * enough. For example, the current vNVDIMM implementation stores | |
1939 | * the guest NVDIMM labels at the end of the backend file. If the | |
1940 | * backend file is later extended, QEMU will not be able to find | |
1941 | * those labels. Therefore, extending the non-empty backend file | |
1942 | * is disabled as well. | |
c902760f | 1943 | */ |
8d37b030 | 1944 | if (truncate && ftruncate(fd, memory)) { |
9742bf26 | 1945 | perror("ftruncate"); |
7f56e740 | 1946 | } |
c902760f | 1947 | |
d2f39add | 1948 | area = qemu_ram_mmap(fd, memory, block->mr->align, |
2ac0f162 | 1949 | block->flags & RAM_SHARED, block->flags & RAM_PMEM); |
c902760f | 1950 | if (area == MAP_FAILED) { |
7f56e740 | 1951 | error_setg_errno(errp, errno, |
fd97fd44 | 1952 | "unable to map backing store for guest RAM"); |
8d37b030 | 1953 | return NULL; |
c902760f | 1954 | } |
ef36fa14 MT |
1955 | |
1956 | if (mem_prealloc) { | |
5cc8767d | 1957 | os_mem_prealloc(fd, area, memory, ms->smp.cpus, errp); |
056b68af | 1958 | if (errp && *errp) { |
53adb9d4 | 1959 | qemu_ram_munmap(fd, area, memory); |
8d37b030 | 1960 | return NULL; |
056b68af | 1961 | } |
ef36fa14 MT |
1962 | } |
1963 | ||
04b16653 | 1964 | block->fd = fd; |
c902760f MT |
1965 | return area; |
1966 | } | |
1967 | #endif | |
1968 | ||
154cc9ea DDAG |
1969 | /* Allocate space within the ram_addr_t space that governs the |
1970 | * dirty bitmaps. | |
1971 | * Called with the ramlist lock held. | |
1972 | */ | |
d17b5288 | 1973 | static ram_addr_t find_ram_offset(ram_addr_t size) |
04b16653 AW |
1974 | { |
1975 | RAMBlock *block, *next_block; | |
3e837b2c | 1976 | ram_addr_t offset = RAM_ADDR_MAX, mingap = RAM_ADDR_MAX; |
04b16653 | 1977 | |
49cd9ac6 SH |
1978 | assert(size != 0); /* it would hand out same offset multiple times */ |
1979 | ||
0dc3f44a | 1980 | if (QLIST_EMPTY_RCU(&ram_list.blocks)) { |
04b16653 | 1981 | return 0; |
0d53d9fe | 1982 | } |
04b16653 | 1983 | |
99e15582 | 1984 | RAMBLOCK_FOREACH(block) { |
154cc9ea | 1985 | ram_addr_t candidate, next = RAM_ADDR_MAX; |
04b16653 | 1986 | |
801110ab DDAG |
1987 | /* Align blocks to start on a 'long' in the bitmap |
1988 | * which makes the bitmap sync'ing take the fast path. | |
1989 | */ | |
154cc9ea | 1990 | candidate = block->offset + block->max_length; |
801110ab | 1991 | candidate = ROUND_UP(candidate, BITS_PER_LONG << TARGET_PAGE_BITS); |
04b16653 | 1992 | |
154cc9ea DDAG |
1993 | /* Search for the closest following block |
1994 | * and find the gap. | |
1995 | */ | |
99e15582 | 1996 | RAMBLOCK_FOREACH(next_block) { |
154cc9ea | 1997 | if (next_block->offset >= candidate) { |
04b16653 AW |
1998 | next = MIN(next, next_block->offset); |
1999 | } | |
2000 | } | |
154cc9ea DDAG |
2001 | |
2002 | /* If it fits remember our place and remember the size | |
2003 | * of gap, but keep going so that we might find a smaller | |
2004 | * gap to fill so avoiding fragmentation. | |
2005 | */ | |
2006 | if (next - candidate >= size && next - candidate < mingap) { | |
2007 | offset = candidate; | |
2008 | mingap = next - candidate; | |
04b16653 | 2009 | } |
154cc9ea DDAG |
2010 | |
2011 | trace_find_ram_offset_loop(size, candidate, offset, next, mingap); | |
04b16653 | 2012 | } |
3e837b2c AW |
2013 | |
2014 | if (offset == RAM_ADDR_MAX) { | |
2015 | fprintf(stderr, "Failed to find gap of requested size: %" PRIu64 "\n", | |
2016 | (uint64_t)size); | |
2017 | abort(); | |
2018 | } | |
2019 | ||
154cc9ea DDAG |
2020 | trace_find_ram_offset(size, offset); |
2021 | ||
04b16653 AW |
2022 | return offset; |
2023 | } | |
2024 | ||
c136180c | 2025 | static unsigned long last_ram_page(void) |
d17b5288 AW |
2026 | { |
2027 | RAMBlock *block; | |
2028 | ram_addr_t last = 0; | |
2029 | ||
0dc3f44a | 2030 | rcu_read_lock(); |
99e15582 | 2031 | RAMBLOCK_FOREACH(block) { |
62be4e3a | 2032 | last = MAX(last, block->offset + block->max_length); |
0d53d9fe | 2033 | } |
0dc3f44a | 2034 | rcu_read_unlock(); |
b8c48993 | 2035 | return last >> TARGET_PAGE_BITS; |
d17b5288 AW |
2036 | } |
2037 | ||
ddb97f1d JB |
2038 | static void qemu_ram_setup_dump(void *addr, ram_addr_t size) |
2039 | { | |
2040 | int ret; | |
ddb97f1d JB |
2041 | |
2042 | /* Use MADV_DONTDUMP, if user doesn't want the guest memory in the core */ | |
47c8ca53 | 2043 | if (!machine_dump_guest_core(current_machine)) { |
ddb97f1d JB |
2044 | ret = qemu_madvise(addr, size, QEMU_MADV_DONTDUMP); |
2045 | if (ret) { | |
2046 | perror("qemu_madvise"); | |
2047 | fprintf(stderr, "madvise doesn't support MADV_DONTDUMP, " | |
2048 | "but dump_guest_core=off specified\n"); | |
2049 | } | |
2050 | } | |
2051 | } | |
2052 | ||
422148d3 DDAG |
2053 | const char *qemu_ram_get_idstr(RAMBlock *rb) |
2054 | { | |
2055 | return rb->idstr; | |
2056 | } | |
2057 | ||
754cb9c0 YK |
2058 | void *qemu_ram_get_host_addr(RAMBlock *rb) |
2059 | { | |
2060 | return rb->host; | |
2061 | } | |
2062 | ||
2063 | ram_addr_t qemu_ram_get_offset(RAMBlock *rb) | |
2064 | { | |
2065 | return rb->offset; | |
2066 | } | |
2067 | ||
2068 | ram_addr_t qemu_ram_get_used_length(RAMBlock *rb) | |
2069 | { | |
2070 | return rb->used_length; | |
2071 | } | |
2072 | ||
463a4ac2 DDAG |
2073 | bool qemu_ram_is_shared(RAMBlock *rb) |
2074 | { | |
2075 | return rb->flags & RAM_SHARED; | |
2076 | } | |
2077 | ||
2ce16640 DDAG |
2078 | /* Note: Only set at the start of postcopy */ |
2079 | bool qemu_ram_is_uf_zeroable(RAMBlock *rb) | |
2080 | { | |
2081 | return rb->flags & RAM_UF_ZEROPAGE; | |
2082 | } | |
2083 | ||
2084 | void qemu_ram_set_uf_zeroable(RAMBlock *rb) | |
2085 | { | |
2086 | rb->flags |= RAM_UF_ZEROPAGE; | |
2087 | } | |
2088 | ||
b895de50 CLG |
2089 | bool qemu_ram_is_migratable(RAMBlock *rb) |
2090 | { | |
2091 | return rb->flags & RAM_MIGRATABLE; | |
2092 | } | |
2093 | ||
2094 | void qemu_ram_set_migratable(RAMBlock *rb) | |
2095 | { | |
2096 | rb->flags |= RAM_MIGRATABLE; | |
2097 | } | |
2098 | ||
2099 | void qemu_ram_unset_migratable(RAMBlock *rb) | |
2100 | { | |
2101 | rb->flags &= ~RAM_MIGRATABLE; | |
2102 | } | |
2103 | ||
ae3a7047 | 2104 | /* Called with iothread lock held. */ |
fa53a0e5 | 2105 | void qemu_ram_set_idstr(RAMBlock *new_block, const char *name, DeviceState *dev) |
20cfe881 | 2106 | { |
fa53a0e5 | 2107 | RAMBlock *block; |
20cfe881 | 2108 | |
c5705a77 AK |
2109 | assert(new_block); |
2110 | assert(!new_block->idstr[0]); | |
84b89d78 | 2111 | |
09e5ab63 AL |
2112 | if (dev) { |
2113 | char *id = qdev_get_dev_path(dev); | |
84b89d78 CM |
2114 | if (id) { |
2115 | snprintf(new_block->idstr, sizeof(new_block->idstr), "%s/", id); | |
7267c094 | 2116 | g_free(id); |
84b89d78 CM |
2117 | } |
2118 | } | |
2119 | pstrcat(new_block->idstr, sizeof(new_block->idstr), name); | |
2120 | ||
ab0a9956 | 2121 | rcu_read_lock(); |
99e15582 | 2122 | RAMBLOCK_FOREACH(block) { |
fa53a0e5 GA |
2123 | if (block != new_block && |
2124 | !strcmp(block->idstr, new_block->idstr)) { | |
84b89d78 CM |
2125 | fprintf(stderr, "RAMBlock \"%s\" already registered, abort!\n", |
2126 | new_block->idstr); | |
2127 | abort(); | |
2128 | } | |
2129 | } | |
0dc3f44a | 2130 | rcu_read_unlock(); |
c5705a77 AK |
2131 | } |
2132 | ||
ae3a7047 | 2133 | /* Called with iothread lock held. */ |
fa53a0e5 | 2134 | void qemu_ram_unset_idstr(RAMBlock *block) |
20cfe881 | 2135 | { |
ae3a7047 MD |
2136 | /* FIXME: arch_init.c assumes that this is not called throughout |
2137 | * migration. Ignore the problem since hot-unplug during migration | |
2138 | * does not work anyway. | |
2139 | */ | |
20cfe881 HT |
2140 | if (block) { |
2141 | memset(block->idstr, 0, sizeof(block->idstr)); | |
2142 | } | |
2143 | } | |
2144 | ||
863e9621 DDAG |
2145 | size_t qemu_ram_pagesize(RAMBlock *rb) |
2146 | { | |
2147 | return rb->page_size; | |
2148 | } | |
2149 | ||
67f11b5c DDAG |
2150 | /* Returns the largest size of page in use */ |
2151 | size_t qemu_ram_pagesize_largest(void) | |
2152 | { | |
2153 | RAMBlock *block; | |
2154 | size_t largest = 0; | |
2155 | ||
99e15582 | 2156 | RAMBLOCK_FOREACH(block) { |
67f11b5c DDAG |
2157 | largest = MAX(largest, qemu_ram_pagesize(block)); |
2158 | } | |
2159 | ||
2160 | return largest; | |
2161 | } | |
2162 | ||
8490fc78 LC |
2163 | static int memory_try_enable_merging(void *addr, size_t len) |
2164 | { | |
75cc7f01 | 2165 | if (!machine_mem_merge(current_machine)) { |
8490fc78 LC |
2166 | /* disabled by the user */ |
2167 | return 0; | |
2168 | } | |
2169 | ||
2170 | return qemu_madvise(addr, len, QEMU_MADV_MERGEABLE); | |
2171 | } | |
2172 | ||
62be4e3a MT |
2173 | /* Only legal before guest might have detected the memory size: e.g. on |
2174 | * incoming migration, or right after reset. | |
2175 | * | |
2176 | * As memory core doesn't know how is memory accessed, it is up to | |
2177 | * resize callback to update device state and/or add assertions to detect | |
2178 | * misuse, if necessary. | |
2179 | */ | |
fa53a0e5 | 2180 | int qemu_ram_resize(RAMBlock *block, ram_addr_t newsize, Error **errp) |
62be4e3a | 2181 | { |
62be4e3a MT |
2182 | assert(block); |
2183 | ||
4ed023ce | 2184 | newsize = HOST_PAGE_ALIGN(newsize); |
129ddaf3 | 2185 | |
62be4e3a MT |
2186 | if (block->used_length == newsize) { |
2187 | return 0; | |
2188 | } | |
2189 | ||
2190 | if (!(block->flags & RAM_RESIZEABLE)) { | |
2191 | error_setg_errno(errp, EINVAL, | |
2192 | "Length mismatch: %s: 0x" RAM_ADDR_FMT | |
2193 | " in != 0x" RAM_ADDR_FMT, block->idstr, | |
2194 | newsize, block->used_length); | |
2195 | return -EINVAL; | |
2196 | } | |
2197 | ||
2198 | if (block->max_length < newsize) { | |
2199 | error_setg_errno(errp, EINVAL, | |
2200 | "Length too large: %s: 0x" RAM_ADDR_FMT | |
2201 | " > 0x" RAM_ADDR_FMT, block->idstr, | |
2202 | newsize, block->max_length); | |
2203 | return -EINVAL; | |
2204 | } | |
2205 | ||
2206 | cpu_physical_memory_clear_dirty_range(block->offset, block->used_length); | |
2207 | block->used_length = newsize; | |
58d2707e PB |
2208 | cpu_physical_memory_set_dirty_range(block->offset, block->used_length, |
2209 | DIRTY_CLIENTS_ALL); | |
62be4e3a MT |
2210 | memory_region_set_size(block->mr, newsize); |
2211 | if (block->resized) { | |
2212 | block->resized(block->idstr, newsize, block->host); | |
2213 | } | |
2214 | return 0; | |
2215 | } | |
2216 | ||
5b82b703 SH |
2217 | /* Called with ram_list.mutex held */ |
2218 | static void dirty_memory_extend(ram_addr_t old_ram_size, | |
2219 | ram_addr_t new_ram_size) | |
2220 | { | |
2221 | ram_addr_t old_num_blocks = DIV_ROUND_UP(old_ram_size, | |
2222 | DIRTY_MEMORY_BLOCK_SIZE); | |
2223 | ram_addr_t new_num_blocks = DIV_ROUND_UP(new_ram_size, | |
2224 | DIRTY_MEMORY_BLOCK_SIZE); | |
2225 | int i; | |
2226 | ||
2227 | /* Only need to extend if block count increased */ | |
2228 | if (new_num_blocks <= old_num_blocks) { | |
2229 | return; | |
2230 | } | |
2231 | ||
2232 | for (i = 0; i < DIRTY_MEMORY_NUM; i++) { | |
2233 | DirtyMemoryBlocks *old_blocks; | |
2234 | DirtyMemoryBlocks *new_blocks; | |
2235 | int j; | |
2236 | ||
2237 | old_blocks = atomic_rcu_read(&ram_list.dirty_memory[i]); | |
2238 | new_blocks = g_malloc(sizeof(*new_blocks) + | |
2239 | sizeof(new_blocks->blocks[0]) * new_num_blocks); | |
2240 | ||
2241 | if (old_num_blocks) { | |
2242 | memcpy(new_blocks->blocks, old_blocks->blocks, | |
2243 | old_num_blocks * sizeof(old_blocks->blocks[0])); | |
2244 | } | |
2245 | ||
2246 | for (j = old_num_blocks; j < new_num_blocks; j++) { | |
2247 | new_blocks->blocks[j] = bitmap_new(DIRTY_MEMORY_BLOCK_SIZE); | |
2248 | } | |
2249 | ||
2250 | atomic_rcu_set(&ram_list.dirty_memory[i], new_blocks); | |
2251 | ||
2252 | if (old_blocks) { | |
2253 | g_free_rcu(old_blocks, rcu); | |
2254 | } | |
2255 | } | |
2256 | } | |
2257 | ||
06329cce | 2258 | static void ram_block_add(RAMBlock *new_block, Error **errp, bool shared) |
c5705a77 | 2259 | { |
e1c57ab8 | 2260 | RAMBlock *block; |
0d53d9fe | 2261 | RAMBlock *last_block = NULL; |
2152f5ca | 2262 | ram_addr_t old_ram_size, new_ram_size; |
37aa7a0e | 2263 | Error *err = NULL; |
2152f5ca | 2264 | |
b8c48993 | 2265 | old_ram_size = last_ram_page(); |
c5705a77 | 2266 | |
b2a8658e | 2267 | qemu_mutex_lock_ramlist(); |
9b8424d5 | 2268 | new_block->offset = find_ram_offset(new_block->max_length); |
e1c57ab8 PB |
2269 | |
2270 | if (!new_block->host) { | |
2271 | if (xen_enabled()) { | |
9b8424d5 | 2272 | xen_ram_alloc(new_block->offset, new_block->max_length, |
37aa7a0e MA |
2273 | new_block->mr, &err); |
2274 | if (err) { | |
2275 | error_propagate(errp, err); | |
2276 | qemu_mutex_unlock_ramlist(); | |
39c350ee | 2277 | return; |
37aa7a0e | 2278 | } |
e1c57ab8 | 2279 | } else { |
9b8424d5 | 2280 | new_block->host = phys_mem_alloc(new_block->max_length, |
06329cce | 2281 | &new_block->mr->align, shared); |
39228250 | 2282 | if (!new_block->host) { |
ef701d7b HT |
2283 | error_setg_errno(errp, errno, |
2284 | "cannot set up guest memory '%s'", | |
2285 | memory_region_name(new_block->mr)); | |
2286 | qemu_mutex_unlock_ramlist(); | |
39c350ee | 2287 | return; |
39228250 | 2288 | } |
9b8424d5 | 2289 | memory_try_enable_merging(new_block->host, new_block->max_length); |
6977dfe6 | 2290 | } |
c902760f | 2291 | } |
94a6b54f | 2292 | |
dd631697 LZ |
2293 | new_ram_size = MAX(old_ram_size, |
2294 | (new_block->offset + new_block->max_length) >> TARGET_PAGE_BITS); | |
2295 | if (new_ram_size > old_ram_size) { | |
5b82b703 | 2296 | dirty_memory_extend(old_ram_size, new_ram_size); |
dd631697 | 2297 | } |
0d53d9fe MD |
2298 | /* Keep the list sorted from biggest to smallest block. Unlike QTAILQ, |
2299 | * QLIST (which has an RCU-friendly variant) does not have insertion at | |
2300 | * tail, so save the last element in last_block. | |
2301 | */ | |
99e15582 | 2302 | RAMBLOCK_FOREACH(block) { |
0d53d9fe | 2303 | last_block = block; |
9b8424d5 | 2304 | if (block->max_length < new_block->max_length) { |
abb26d63 PB |
2305 | break; |
2306 | } | |
2307 | } | |
2308 | if (block) { | |
0dc3f44a | 2309 | QLIST_INSERT_BEFORE_RCU(block, new_block, next); |
0d53d9fe | 2310 | } else if (last_block) { |
0dc3f44a | 2311 | QLIST_INSERT_AFTER_RCU(last_block, new_block, next); |
0d53d9fe | 2312 | } else { /* list is empty */ |
0dc3f44a | 2313 | QLIST_INSERT_HEAD_RCU(&ram_list.blocks, new_block, next); |
abb26d63 | 2314 | } |
0d6d3c87 | 2315 | ram_list.mru_block = NULL; |
94a6b54f | 2316 | |
0dc3f44a MD |
2317 | /* Write list before version */ |
2318 | smp_wmb(); | |
f798b07f | 2319 | ram_list.version++; |
b2a8658e | 2320 | qemu_mutex_unlock_ramlist(); |
f798b07f | 2321 | |
9b8424d5 | 2322 | cpu_physical_memory_set_dirty_range(new_block->offset, |
58d2707e PB |
2323 | new_block->used_length, |
2324 | DIRTY_CLIENTS_ALL); | |
94a6b54f | 2325 | |
a904c911 PB |
2326 | if (new_block->host) { |
2327 | qemu_ram_setup_dump(new_block->host, new_block->max_length); | |
2328 | qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_HUGEPAGE); | |
c2cd627d | 2329 | /* MADV_DONTFORK is also needed by KVM in absence of synchronous MMU */ |
a904c911 | 2330 | qemu_madvise(new_block->host, new_block->max_length, QEMU_MADV_DONTFORK); |
0987d735 | 2331 | ram_block_notify_add(new_block->host, new_block->max_length); |
e1c57ab8 | 2332 | } |
94a6b54f | 2333 | } |
e9a1ab19 | 2334 | |
d5dbde46 | 2335 | #ifdef CONFIG_POSIX |
38b3362d | 2336 | RAMBlock *qemu_ram_alloc_from_fd(ram_addr_t size, MemoryRegion *mr, |
cbfc0171 | 2337 | uint32_t ram_flags, int fd, |
38b3362d | 2338 | Error **errp) |
e1c57ab8 PB |
2339 | { |
2340 | RAMBlock *new_block; | |
ef701d7b | 2341 | Error *local_err = NULL; |
8d37b030 | 2342 | int64_t file_size; |
e1c57ab8 | 2343 | |
a4de8552 JH |
2344 | /* Just support these ram flags by now. */ |
2345 | assert((ram_flags & ~(RAM_SHARED | RAM_PMEM)) == 0); | |
2346 | ||
e1c57ab8 | 2347 | if (xen_enabled()) { |
7f56e740 | 2348 | error_setg(errp, "-mem-path not supported with Xen"); |
528f46af | 2349 | return NULL; |
e1c57ab8 PB |
2350 | } |
2351 | ||
e45e7ae2 MAL |
2352 | if (kvm_enabled() && !kvm_has_sync_mmu()) { |
2353 | error_setg(errp, | |
2354 | "host lacks kvm mmu notifiers, -mem-path unsupported"); | |
2355 | return NULL; | |
2356 | } | |
2357 | ||
e1c57ab8 PB |
2358 | if (phys_mem_alloc != qemu_anon_ram_alloc) { |
2359 | /* | |
2360 | * file_ram_alloc() needs to allocate just like | |
2361 | * phys_mem_alloc, but we haven't bothered to provide | |
2362 | * a hook there. | |
2363 | */ | |
7f56e740 PB |
2364 | error_setg(errp, |
2365 | "-mem-path not supported with this accelerator"); | |
528f46af | 2366 | return NULL; |
e1c57ab8 PB |
2367 | } |
2368 | ||
4ed023ce | 2369 | size = HOST_PAGE_ALIGN(size); |
8d37b030 MAL |
2370 | file_size = get_file_size(fd); |
2371 | if (file_size > 0 && file_size < size) { | |
2372 | error_setg(errp, "backing store %s size 0x%" PRIx64 | |
2373 | " does not match 'size' option 0x" RAM_ADDR_FMT, | |
2374 | mem_path, file_size, size); | |
8d37b030 MAL |
2375 | return NULL; |
2376 | } | |
2377 | ||
e1c57ab8 PB |
2378 | new_block = g_malloc0(sizeof(*new_block)); |
2379 | new_block->mr = mr; | |
9b8424d5 MT |
2380 | new_block->used_length = size; |
2381 | new_block->max_length = size; | |
cbfc0171 | 2382 | new_block->flags = ram_flags; |
8d37b030 | 2383 | new_block->host = file_ram_alloc(new_block, size, fd, !file_size, errp); |
7f56e740 PB |
2384 | if (!new_block->host) { |
2385 | g_free(new_block); | |
528f46af | 2386 | return NULL; |
7f56e740 PB |
2387 | } |
2388 | ||
cbfc0171 | 2389 | ram_block_add(new_block, &local_err, ram_flags & RAM_SHARED); |
ef701d7b HT |
2390 | if (local_err) { |
2391 | g_free(new_block); | |
2392 | error_propagate(errp, local_err); | |
528f46af | 2393 | return NULL; |
ef701d7b | 2394 | } |
528f46af | 2395 | return new_block; |
38b3362d MAL |
2396 | |
2397 | } | |
2398 | ||
2399 | ||
2400 | RAMBlock *qemu_ram_alloc_from_file(ram_addr_t size, MemoryRegion *mr, | |
cbfc0171 | 2401 | uint32_t ram_flags, const char *mem_path, |
38b3362d MAL |
2402 | Error **errp) |
2403 | { | |
2404 | int fd; | |
2405 | bool created; | |
2406 | RAMBlock *block; | |
2407 | ||
2408 | fd = file_ram_open(mem_path, memory_region_name(mr), &created, errp); | |
2409 | if (fd < 0) { | |
2410 | return NULL; | |
2411 | } | |
2412 | ||
cbfc0171 | 2413 | block = qemu_ram_alloc_from_fd(size, mr, ram_flags, fd, errp); |
38b3362d MAL |
2414 | if (!block) { |
2415 | if (created) { | |
2416 | unlink(mem_path); | |
2417 | } | |
2418 | close(fd); | |
2419 | return NULL; | |
2420 | } | |
2421 | ||
2422 | return block; | |
e1c57ab8 | 2423 | } |
0b183fc8 | 2424 | #endif |
e1c57ab8 | 2425 | |
62be4e3a | 2426 | static |
528f46af FZ |
2427 | RAMBlock *qemu_ram_alloc_internal(ram_addr_t size, ram_addr_t max_size, |
2428 | void (*resized)(const char*, | |
2429 | uint64_t length, | |
2430 | void *host), | |
06329cce | 2431 | void *host, bool resizeable, bool share, |
528f46af | 2432 | MemoryRegion *mr, Error **errp) |
e1c57ab8 PB |
2433 | { |
2434 | RAMBlock *new_block; | |
ef701d7b | 2435 | Error *local_err = NULL; |
e1c57ab8 | 2436 | |
4ed023ce DDAG |
2437 | size = HOST_PAGE_ALIGN(size); |
2438 | max_size = HOST_PAGE_ALIGN(max_size); | |
e1c57ab8 PB |
2439 | new_block = g_malloc0(sizeof(*new_block)); |
2440 | new_block->mr = mr; | |
62be4e3a | 2441 | new_block->resized = resized; |
9b8424d5 MT |
2442 | new_block->used_length = size; |
2443 | new_block->max_length = max_size; | |
62be4e3a | 2444 | assert(max_size >= size); |
e1c57ab8 | 2445 | new_block->fd = -1; |
863e9621 | 2446 | new_block->page_size = getpagesize(); |
e1c57ab8 PB |
2447 | new_block->host = host; |
2448 | if (host) { | |
7bd4f430 | 2449 | new_block->flags |= RAM_PREALLOC; |
e1c57ab8 | 2450 | } |
62be4e3a MT |
2451 | if (resizeable) { |
2452 | new_block->flags |= RAM_RESIZEABLE; | |
2453 | } | |
06329cce | 2454 | ram_block_add(new_block, &local_err, share); |
ef701d7b HT |
2455 | if (local_err) { |
2456 | g_free(new_block); | |
2457 | error_propagate(errp, local_err); | |
528f46af | 2458 | return NULL; |
ef701d7b | 2459 | } |
528f46af | 2460 | return new_block; |
e1c57ab8 PB |
2461 | } |
2462 | ||
528f46af | 2463 | RAMBlock *qemu_ram_alloc_from_ptr(ram_addr_t size, void *host, |
62be4e3a MT |
2464 | MemoryRegion *mr, Error **errp) |
2465 | { | |
06329cce MA |
2466 | return qemu_ram_alloc_internal(size, size, NULL, host, false, |
2467 | false, mr, errp); | |
62be4e3a MT |
2468 | } |
2469 | ||
06329cce MA |
2470 | RAMBlock *qemu_ram_alloc(ram_addr_t size, bool share, |
2471 | MemoryRegion *mr, Error **errp) | |
6977dfe6 | 2472 | { |
06329cce MA |
2473 | return qemu_ram_alloc_internal(size, size, NULL, NULL, false, |
2474 | share, mr, errp); | |
62be4e3a MT |
2475 | } |
2476 | ||
528f46af | 2477 | RAMBlock *qemu_ram_alloc_resizeable(ram_addr_t size, ram_addr_t maxsz, |
62be4e3a MT |
2478 | void (*resized)(const char*, |
2479 | uint64_t length, | |
2480 | void *host), | |
2481 | MemoryRegion *mr, Error **errp) | |
2482 | { | |
06329cce MA |
2483 | return qemu_ram_alloc_internal(size, maxsz, resized, NULL, true, |
2484 | false, mr, errp); | |
6977dfe6 YT |
2485 | } |
2486 | ||
43771539 PB |
2487 | static void reclaim_ramblock(RAMBlock *block) |
2488 | { | |
2489 | if (block->flags & RAM_PREALLOC) { | |
2490 | ; | |
2491 | } else if (xen_enabled()) { | |
2492 | xen_invalidate_map_cache_entry(block->host); | |
2493 | #ifndef _WIN32 | |
2494 | } else if (block->fd >= 0) { | |
53adb9d4 | 2495 | qemu_ram_munmap(block->fd, block->host, block->max_length); |
43771539 PB |
2496 | close(block->fd); |
2497 | #endif | |
2498 | } else { | |
2499 | qemu_anon_ram_free(block->host, block->max_length); | |
2500 | } | |
2501 | g_free(block); | |
2502 | } | |
2503 | ||
f1060c55 | 2504 | void qemu_ram_free(RAMBlock *block) |
e9a1ab19 | 2505 | { |
85bc2a15 MAL |
2506 | if (!block) { |
2507 | return; | |
2508 | } | |
2509 | ||
0987d735 PB |
2510 | if (block->host) { |
2511 | ram_block_notify_remove(block->host, block->max_length); | |
2512 | } | |
2513 | ||
b2a8658e | 2514 | qemu_mutex_lock_ramlist(); |
f1060c55 FZ |
2515 | QLIST_REMOVE_RCU(block, next); |
2516 | ram_list.mru_block = NULL; | |
2517 | /* Write list before version */ | |
2518 | smp_wmb(); | |
2519 | ram_list.version++; | |
2520 | call_rcu(block, reclaim_ramblock, rcu); | |
b2a8658e | 2521 | qemu_mutex_unlock_ramlist(); |
e9a1ab19 FB |
2522 | } |
2523 | ||
cd19cfa2 HY |
2524 | #ifndef _WIN32 |
2525 | void qemu_ram_remap(ram_addr_t addr, ram_addr_t length) | |
2526 | { | |
2527 | RAMBlock *block; | |
2528 | ram_addr_t offset; | |
2529 | int flags; | |
2530 | void *area, *vaddr; | |
2531 | ||
99e15582 | 2532 | RAMBLOCK_FOREACH(block) { |
cd19cfa2 | 2533 | offset = addr - block->offset; |
9b8424d5 | 2534 | if (offset < block->max_length) { |
1240be24 | 2535 | vaddr = ramblock_ptr(block, offset); |
7bd4f430 | 2536 | if (block->flags & RAM_PREALLOC) { |
cd19cfa2 | 2537 | ; |
dfeaf2ab MA |
2538 | } else if (xen_enabled()) { |
2539 | abort(); | |
cd19cfa2 HY |
2540 | } else { |
2541 | flags = MAP_FIXED; | |
3435f395 | 2542 | if (block->fd >= 0) { |
dbcb8981 PB |
2543 | flags |= (block->flags & RAM_SHARED ? |
2544 | MAP_SHARED : MAP_PRIVATE); | |
3435f395 MA |
2545 | area = mmap(vaddr, length, PROT_READ | PROT_WRITE, |
2546 | flags, block->fd, offset); | |
cd19cfa2 | 2547 | } else { |
2eb9fbaa MA |
2548 | /* |
2549 | * Remap needs to match alloc. Accelerators that | |
2550 | * set phys_mem_alloc never remap. If they did, | |
2551 | * we'd need a remap hook here. | |
2552 | */ | |
2553 | assert(phys_mem_alloc == qemu_anon_ram_alloc); | |
2554 | ||
cd19cfa2 HY |
2555 | flags |= MAP_PRIVATE | MAP_ANONYMOUS; |
2556 | area = mmap(vaddr, length, PROT_READ | PROT_WRITE, | |
2557 | flags, -1, 0); | |
cd19cfa2 HY |
2558 | } |
2559 | if (area != vaddr) { | |
493d89bf AF |
2560 | error_report("Could not remap addr: " |
2561 | RAM_ADDR_FMT "@" RAM_ADDR_FMT "", | |
2562 | length, addr); | |
cd19cfa2 HY |
2563 | exit(1); |
2564 | } | |
8490fc78 | 2565 | memory_try_enable_merging(vaddr, length); |
ddb97f1d | 2566 | qemu_ram_setup_dump(vaddr, length); |
cd19cfa2 | 2567 | } |
cd19cfa2 HY |
2568 | } |
2569 | } | |
2570 | } | |
2571 | #endif /* !_WIN32 */ | |
2572 | ||
1b5ec234 | 2573 | /* Return a host pointer to ram allocated with qemu_ram_alloc. |
ae3a7047 MD |
2574 | * This should not be used for general purpose DMA. Use address_space_map |
2575 | * or address_space_rw instead. For local memory (e.g. video ram) that the | |
2576 | * device owns, use memory_region_get_ram_ptr. | |
0dc3f44a | 2577 | * |
49b24afc | 2578 | * Called within RCU critical section. |
1b5ec234 | 2579 | */ |
0878d0e1 | 2580 | void *qemu_map_ram_ptr(RAMBlock *ram_block, ram_addr_t addr) |
1b5ec234 | 2581 | { |
3655cb9c GA |
2582 | RAMBlock *block = ram_block; |
2583 | ||
2584 | if (block == NULL) { | |
2585 | block = qemu_get_ram_block(addr); | |
0878d0e1 | 2586 | addr -= block->offset; |
3655cb9c | 2587 | } |
ae3a7047 MD |
2588 | |
2589 | if (xen_enabled() && block->host == NULL) { | |
0d6d3c87 PB |
2590 | /* We need to check if the requested address is in the RAM |
2591 | * because we don't want to map the entire memory in QEMU. | |
2592 | * In that case just map until the end of the page. | |
2593 | */ | |
2594 | if (block->offset == 0) { | |
1ff7c598 | 2595 | return xen_map_cache(addr, 0, 0, false); |
0d6d3c87 | 2596 | } |
ae3a7047 | 2597 | |
1ff7c598 | 2598 | block->host = xen_map_cache(block->offset, block->max_length, 1, false); |
0d6d3c87 | 2599 | } |
0878d0e1 | 2600 | return ramblock_ptr(block, addr); |
dc828ca1 PB |
2601 | } |
2602 | ||
0878d0e1 | 2603 | /* Return a host pointer to guest's ram. Similar to qemu_map_ram_ptr |
ae3a7047 | 2604 | * but takes a size argument. |
0dc3f44a | 2605 | * |
e81bcda5 | 2606 | * Called within RCU critical section. |
ae3a7047 | 2607 | */ |
3655cb9c | 2608 | static void *qemu_ram_ptr_length(RAMBlock *ram_block, ram_addr_t addr, |
f5aa69bd | 2609 | hwaddr *size, bool lock) |
38bee5dc | 2610 | { |
3655cb9c | 2611 | RAMBlock *block = ram_block; |
8ab934f9 SS |
2612 | if (*size == 0) { |
2613 | return NULL; | |
2614 | } | |
e81bcda5 | 2615 | |
3655cb9c GA |
2616 | if (block == NULL) { |
2617 | block = qemu_get_ram_block(addr); | |
0878d0e1 | 2618 | addr -= block->offset; |
3655cb9c | 2619 | } |
0878d0e1 | 2620 | *size = MIN(*size, block->max_length - addr); |
e81bcda5 PB |
2621 | |
2622 | if (xen_enabled() && block->host == NULL) { | |
2623 | /* We need to check if the requested address is in the RAM | |
2624 | * because we don't want to map the entire memory in QEMU. | |
2625 | * In that case just map the requested area. | |
2626 | */ | |
2627 | if (block->offset == 0) { | |
f5aa69bd | 2628 | return xen_map_cache(addr, *size, lock, lock); |
38bee5dc SS |
2629 | } |
2630 | ||
f5aa69bd | 2631 | block->host = xen_map_cache(block->offset, block->max_length, 1, lock); |
38bee5dc | 2632 | } |
e81bcda5 | 2633 | |
0878d0e1 | 2634 | return ramblock_ptr(block, addr); |
38bee5dc SS |
2635 | } |
2636 | ||
f90bb71b DDAG |
2637 | /* Return the offset of a hostpointer within a ramblock */ |
2638 | ram_addr_t qemu_ram_block_host_offset(RAMBlock *rb, void *host) | |
2639 | { | |
2640 | ram_addr_t res = (uint8_t *)host - (uint8_t *)rb->host; | |
2641 | assert((uintptr_t)host >= (uintptr_t)rb->host); | |
2642 | assert(res < rb->max_length); | |
2643 | ||
2644 | return res; | |
2645 | } | |
2646 | ||
422148d3 DDAG |
2647 | /* |
2648 | * Translates a host ptr back to a RAMBlock, a ram_addr and an offset | |
2649 | * in that RAMBlock. | |
2650 | * | |
2651 | * ptr: Host pointer to look up | |
2652 | * round_offset: If true round the result offset down to a page boundary | |
2653 | * *ram_addr: set to result ram_addr | |
2654 | * *offset: set to result offset within the RAMBlock | |
2655 | * | |
2656 | * Returns: RAMBlock (or NULL if not found) | |
ae3a7047 MD |
2657 | * |
2658 | * By the time this function returns, the returned pointer is not protected | |
2659 | * by RCU anymore. If the caller is not within an RCU critical section and | |
2660 | * does not hold the iothread lock, it must have other means of protecting the | |
2661 | * pointer, such as a reference to the region that includes the incoming | |
2662 | * ram_addr_t. | |
2663 | */ | |
422148d3 | 2664 | RAMBlock *qemu_ram_block_from_host(void *ptr, bool round_offset, |
422148d3 | 2665 | ram_addr_t *offset) |
5579c7f3 | 2666 | { |
94a6b54f PB |
2667 | RAMBlock *block; |
2668 | uint8_t *host = ptr; | |
2669 | ||
868bb33f | 2670 | if (xen_enabled()) { |
f615f396 | 2671 | ram_addr_t ram_addr; |
0dc3f44a | 2672 | rcu_read_lock(); |
f615f396 PB |
2673 | ram_addr = xen_ram_addr_from_mapcache(ptr); |
2674 | block = qemu_get_ram_block(ram_addr); | |
422148d3 | 2675 | if (block) { |
d6b6aec4 | 2676 | *offset = ram_addr - block->offset; |
422148d3 | 2677 | } |
0dc3f44a | 2678 | rcu_read_unlock(); |
422148d3 | 2679 | return block; |
712c2b41 SS |
2680 | } |
2681 | ||
0dc3f44a MD |
2682 | rcu_read_lock(); |
2683 | block = atomic_rcu_read(&ram_list.mru_block); | |
9b8424d5 | 2684 | if (block && block->host && host - block->host < block->max_length) { |
23887b79 PB |
2685 | goto found; |
2686 | } | |
2687 | ||
99e15582 | 2688 | RAMBLOCK_FOREACH(block) { |
432d268c JN |
2689 | /* This case append when the block is not mapped. */ |
2690 | if (block->host == NULL) { | |
2691 | continue; | |
2692 | } | |
9b8424d5 | 2693 | if (host - block->host < block->max_length) { |
23887b79 | 2694 | goto found; |
f471a17e | 2695 | } |
94a6b54f | 2696 | } |
432d268c | 2697 | |
0dc3f44a | 2698 | rcu_read_unlock(); |
1b5ec234 | 2699 | return NULL; |
23887b79 PB |
2700 | |
2701 | found: | |
422148d3 DDAG |
2702 | *offset = (host - block->host); |
2703 | if (round_offset) { | |
2704 | *offset &= TARGET_PAGE_MASK; | |
2705 | } | |
0dc3f44a | 2706 | rcu_read_unlock(); |
422148d3 DDAG |
2707 | return block; |
2708 | } | |
2709 | ||
e3dd7493 DDAG |
2710 | /* |
2711 | * Finds the named RAMBlock | |
2712 | * | |
2713 | * name: The name of RAMBlock to find | |
2714 | * | |
2715 | * Returns: RAMBlock (or NULL if not found) | |
2716 | */ | |
2717 | RAMBlock *qemu_ram_block_by_name(const char *name) | |
2718 | { | |
2719 | RAMBlock *block; | |
2720 | ||
99e15582 | 2721 | RAMBLOCK_FOREACH(block) { |
e3dd7493 DDAG |
2722 | if (!strcmp(name, block->idstr)) { |
2723 | return block; | |
2724 | } | |
2725 | } | |
2726 | ||
2727 | return NULL; | |
2728 | } | |
2729 | ||
422148d3 DDAG |
2730 | /* Some of the softmmu routines need to translate from a host pointer |
2731 | (typically a TLB entry) back to a ram offset. */ | |
07bdaa41 | 2732 | ram_addr_t qemu_ram_addr_from_host(void *ptr) |
422148d3 DDAG |
2733 | { |
2734 | RAMBlock *block; | |
f615f396 | 2735 | ram_addr_t offset; |
422148d3 | 2736 | |
f615f396 | 2737 | block = qemu_ram_block_from_host(ptr, false, &offset); |
422148d3 | 2738 | if (!block) { |
07bdaa41 | 2739 | return RAM_ADDR_INVALID; |
422148d3 DDAG |
2740 | } |
2741 | ||
07bdaa41 | 2742 | return block->offset + offset; |
e890261f | 2743 | } |
f471a17e | 2744 | |
27266271 PM |
2745 | /* Called within RCU critical section. */ |
2746 | void memory_notdirty_write_prepare(NotDirtyInfo *ndi, | |
2747 | CPUState *cpu, | |
2748 | vaddr mem_vaddr, | |
2749 | ram_addr_t ram_addr, | |
2750 | unsigned size) | |
2751 | { | |
2752 | ndi->cpu = cpu; | |
2753 | ndi->ram_addr = ram_addr; | |
2754 | ndi->mem_vaddr = mem_vaddr; | |
2755 | ndi->size = size; | |
0ac20318 | 2756 | ndi->pages = NULL; |
ba051fb5 | 2757 | |
5aa1ef71 | 2758 | assert(tcg_enabled()); |
52159192 | 2759 | if (!cpu_physical_memory_get_dirty_flag(ram_addr, DIRTY_MEMORY_CODE)) { |
0ac20318 EC |
2760 | ndi->pages = page_collection_lock(ram_addr, ram_addr + size); |
2761 | tb_invalidate_phys_page_fast(ndi->pages, ram_addr, size); | |
3a7d929e | 2762 | } |
27266271 PM |
2763 | } |
2764 | ||
2765 | /* Called within RCU critical section. */ | |
2766 | void memory_notdirty_write_complete(NotDirtyInfo *ndi) | |
2767 | { | |
0ac20318 | 2768 | if (ndi->pages) { |
f28d0dfd | 2769 | assert(tcg_enabled()); |
0ac20318 EC |
2770 | page_collection_unlock(ndi->pages); |
2771 | ndi->pages = NULL; | |
27266271 PM |
2772 | } |
2773 | ||
2774 | /* Set both VGA and migration bits for simplicity and to remove | |
2775 | * the notdirty callback faster. | |
2776 | */ | |
2777 | cpu_physical_memory_set_dirty_range(ndi->ram_addr, ndi->size, | |
2778 | DIRTY_CLIENTS_NOCODE); | |
2779 | /* we remove the notdirty callback only if the code has been | |
2780 | flushed */ | |
2781 | if (!cpu_physical_memory_is_clean(ndi->ram_addr)) { | |
2782 | tlb_set_dirty(ndi->cpu, ndi->mem_vaddr); | |
2783 | } | |
2784 | } | |
2785 | ||
2786 | /* Called within RCU critical section. */ | |
2787 | static void notdirty_mem_write(void *opaque, hwaddr ram_addr, | |
2788 | uint64_t val, unsigned size) | |
2789 | { | |
2790 | NotDirtyInfo ndi; | |
2791 | ||
2792 | memory_notdirty_write_prepare(&ndi, current_cpu, current_cpu->mem_io_vaddr, | |
2793 | ram_addr, size); | |
2794 | ||
6d3ede54 | 2795 | stn_p(qemu_map_ram_ptr(NULL, ram_addr), size, val); |
27266271 | 2796 | memory_notdirty_write_complete(&ndi); |
9fa3e853 FB |
2797 | } |
2798 | ||
b018ddf6 | 2799 | static bool notdirty_mem_accepts(void *opaque, hwaddr addr, |
8372d383 PM |
2800 | unsigned size, bool is_write, |
2801 | MemTxAttrs attrs) | |
b018ddf6 PB |
2802 | { |
2803 | return is_write; | |
2804 | } | |
2805 | ||
0e0df1e2 | 2806 | static const MemoryRegionOps notdirty_mem_ops = { |
0e0df1e2 | 2807 | .write = notdirty_mem_write, |
b018ddf6 | 2808 | .valid.accepts = notdirty_mem_accepts, |
0e0df1e2 | 2809 | .endianness = DEVICE_NATIVE_ENDIAN, |
ad52878f AB |
2810 | .valid = { |
2811 | .min_access_size = 1, | |
2812 | .max_access_size = 8, | |
2813 | .unaligned = false, | |
2814 | }, | |
2815 | .impl = { | |
2816 | .min_access_size = 1, | |
2817 | .max_access_size = 8, | |
2818 | .unaligned = false, | |
2819 | }, | |
1ccde1cb FB |
2820 | }; |
2821 | ||
0f459d16 | 2822 | /* Generate a debug exception if a watchpoint has been hit. */ |
0026348b DH |
2823 | void cpu_check_watchpoint(CPUState *cpu, vaddr addr, vaddr len, |
2824 | MemTxAttrs attrs, int flags, uintptr_t ra) | |
0f459d16 | 2825 | { |
568496c0 | 2826 | CPUClass *cc = CPU_GET_CLASS(cpu); |
a1d1bb31 | 2827 | CPUWatchpoint *wp; |
0f459d16 | 2828 | |
5aa1ef71 | 2829 | assert(tcg_enabled()); |
ff4700b0 | 2830 | if (cpu->watchpoint_hit) { |
50b107c5 RH |
2831 | /* |
2832 | * We re-entered the check after replacing the TB. | |
2833 | * Now raise the debug interrupt so that it will | |
2834 | * trigger after the current instruction. | |
2835 | */ | |
2836 | qemu_mutex_lock_iothread(); | |
93afeade | 2837 | cpu_interrupt(cpu, CPU_INTERRUPT_DEBUG); |
50b107c5 | 2838 | qemu_mutex_unlock_iothread(); |
06d55cc1 AL |
2839 | return; |
2840 | } | |
0026348b DH |
2841 | |
2842 | addr = cc->adjust_watchpoint_address(cpu, addr, len); | |
ff4700b0 | 2843 | QTAILQ_FOREACH(wp, &cpu->watchpoints, entry) { |
56ad8b00 | 2844 | if (watchpoint_address_matches(wp, addr, len) |
05068c0d | 2845 | && (wp->flags & flags)) { |
08225676 PM |
2846 | if (flags == BP_MEM_READ) { |
2847 | wp->flags |= BP_WATCHPOINT_HIT_READ; | |
2848 | } else { | |
2849 | wp->flags |= BP_WATCHPOINT_HIT_WRITE; | |
2850 | } | |
0026348b | 2851 | wp->hitaddr = MAX(addr, wp->vaddr); |
66b9b43c | 2852 | wp->hitattrs = attrs; |
ff4700b0 | 2853 | if (!cpu->watchpoint_hit) { |
568496c0 SF |
2854 | if (wp->flags & BP_CPU && |
2855 | !cc->debug_check_watchpoint(cpu, wp)) { | |
2856 | wp->flags &= ~BP_WATCHPOINT_HIT; | |
2857 | continue; | |
2858 | } | |
ff4700b0 | 2859 | cpu->watchpoint_hit = wp; |
a5e99826 | 2860 | |
0ac20318 | 2861 | mmap_lock(); |
239c51a5 | 2862 | tb_check_watchpoint(cpu); |
6e140f28 | 2863 | if (wp->flags & BP_STOP_BEFORE_ACCESS) { |
27103424 | 2864 | cpu->exception_index = EXCP_DEBUG; |
0ac20318 | 2865 | mmap_unlock(); |
0026348b | 2866 | cpu_loop_exit_restore(cpu, ra); |
6e140f28 | 2867 | } else { |
9b990ee5 RH |
2868 | /* Force execution of one insn next time. */ |
2869 | cpu->cflags_next_tb = 1 | curr_cflags(); | |
0ac20318 | 2870 | mmap_unlock(); |
0026348b DH |
2871 | if (ra) { |
2872 | cpu_restore_state(cpu, ra, true); | |
2873 | } | |
6886b980 | 2874 | cpu_loop_exit_noexc(cpu); |
6e140f28 | 2875 | } |
06d55cc1 | 2876 | } |
6e140f28 AL |
2877 | } else { |
2878 | wp->flags &= ~BP_WATCHPOINT_HIT; | |
0f459d16 PB |
2879 | } |
2880 | } | |
2881 | } | |
2882 | ||
b2a44fca | 2883 | static MemTxResult flatview_read(FlatView *fv, hwaddr addr, |
0c249ff7 | 2884 | MemTxAttrs attrs, uint8_t *buf, hwaddr len); |
16620684 | 2885 | static MemTxResult flatview_write(FlatView *fv, hwaddr addr, MemTxAttrs attrs, |
0c249ff7 LZ |
2886 | const uint8_t *buf, hwaddr len); |
2887 | static bool flatview_access_valid(FlatView *fv, hwaddr addr, hwaddr len, | |
eace72b7 | 2888 | bool is_write, MemTxAttrs attrs); |
16620684 | 2889 | |
f25a49e0 PM |
2890 | static MemTxResult subpage_read(void *opaque, hwaddr addr, uint64_t *data, |
2891 | unsigned len, MemTxAttrs attrs) | |
db7b5426 | 2892 | { |
acc9d80b | 2893 | subpage_t *subpage = opaque; |
ff6cff75 | 2894 | uint8_t buf[8]; |
5c9eb028 | 2895 | MemTxResult res; |
791af8c8 | 2896 | |
db7b5426 | 2897 | #if defined(DEBUG_SUBPAGE) |
016e9d62 | 2898 | printf("%s: subpage %p len %u addr " TARGET_FMT_plx "\n", __func__, |
acc9d80b | 2899 | subpage, len, addr); |
db7b5426 | 2900 | #endif |
16620684 | 2901 | res = flatview_read(subpage->fv, addr + subpage->base, attrs, buf, len); |
5c9eb028 PM |
2902 | if (res) { |
2903 | return res; | |
f25a49e0 | 2904 | } |
6d3ede54 PM |
2905 | *data = ldn_p(buf, len); |
2906 | return MEMTX_OK; | |
db7b5426 BS |
2907 | } |
2908 | ||
f25a49e0 PM |
2909 | static MemTxResult subpage_write(void *opaque, hwaddr addr, |
2910 | uint64_t value, unsigned len, MemTxAttrs attrs) | |
db7b5426 | 2911 | { |
acc9d80b | 2912 | subpage_t *subpage = opaque; |
ff6cff75 | 2913 | uint8_t buf[8]; |
acc9d80b | 2914 | |
db7b5426 | 2915 | #if defined(DEBUG_SUBPAGE) |
016e9d62 | 2916 | printf("%s: subpage %p len %u addr " TARGET_FMT_plx |
acc9d80b JK |
2917 | " value %"PRIx64"\n", |
2918 | __func__, subpage, len, addr, value); | |
db7b5426 | 2919 | #endif |
6d3ede54 | 2920 | stn_p(buf, len, value); |
16620684 | 2921 | return flatview_write(subpage->fv, addr + subpage->base, attrs, buf, len); |
db7b5426 BS |
2922 | } |
2923 | ||
c353e4cc | 2924 | static bool subpage_accepts(void *opaque, hwaddr addr, |
8372d383 PM |
2925 | unsigned len, bool is_write, |
2926 | MemTxAttrs attrs) | |
c353e4cc | 2927 | { |
acc9d80b | 2928 | subpage_t *subpage = opaque; |
c353e4cc | 2929 | #if defined(DEBUG_SUBPAGE) |
016e9d62 | 2930 | printf("%s: subpage %p %c len %u addr " TARGET_FMT_plx "\n", |
acc9d80b | 2931 | __func__, subpage, is_write ? 'w' : 'r', len, addr); |
c353e4cc PB |
2932 | #endif |
2933 | ||
16620684 | 2934 | return flatview_access_valid(subpage->fv, addr + subpage->base, |
eace72b7 | 2935 | len, is_write, attrs); |
c353e4cc PB |
2936 | } |
2937 | ||
70c68e44 | 2938 | static const MemoryRegionOps subpage_ops = { |
f25a49e0 PM |
2939 | .read_with_attrs = subpage_read, |
2940 | .write_with_attrs = subpage_write, | |
ff6cff75 PB |
2941 | .impl.min_access_size = 1, |
2942 | .impl.max_access_size = 8, | |
2943 | .valid.min_access_size = 1, | |
2944 | .valid.max_access_size = 8, | |
c353e4cc | 2945 | .valid.accepts = subpage_accepts, |
70c68e44 | 2946 | .endianness = DEVICE_NATIVE_ENDIAN, |
db7b5426 BS |
2947 | }; |
2948 | ||
b797ab1a WY |
2949 | static int subpage_register(subpage_t *mmio, uint32_t start, uint32_t end, |
2950 | uint16_t section) | |
db7b5426 BS |
2951 | { |
2952 | int idx, eidx; | |
2953 | ||
2954 | if (start >= TARGET_PAGE_SIZE || end >= TARGET_PAGE_SIZE) | |
2955 | return -1; | |
2956 | idx = SUBPAGE_IDX(start); | |
2957 | eidx = SUBPAGE_IDX(end); | |
2958 | #if defined(DEBUG_SUBPAGE) | |
016e9d62 AK |
2959 | printf("%s: %p start %08x end %08x idx %08x eidx %08x section %d\n", |
2960 | __func__, mmio, start, end, idx, eidx, section); | |
db7b5426 | 2961 | #endif |
db7b5426 | 2962 | for (; idx <= eidx; idx++) { |
5312bd8b | 2963 | mmio->sub_section[idx] = section; |
db7b5426 BS |
2964 | } |
2965 | ||
2966 | return 0; | |
2967 | } | |
2968 | ||
16620684 | 2969 | static subpage_t *subpage_init(FlatView *fv, hwaddr base) |
db7b5426 | 2970 | { |
c227f099 | 2971 | subpage_t *mmio; |
db7b5426 | 2972 | |
b797ab1a | 2973 | /* mmio->sub_section is set to PHYS_SECTION_UNASSIGNED with g_malloc0 */ |
2615fabd | 2974 | mmio = g_malloc0(sizeof(subpage_t) + TARGET_PAGE_SIZE * sizeof(uint16_t)); |
16620684 | 2975 | mmio->fv = fv; |
1eec614b | 2976 | mmio->base = base; |
2c9b15ca | 2977 | memory_region_init_io(&mmio->iomem, NULL, &subpage_ops, mmio, |
b4fefef9 | 2978 | NULL, TARGET_PAGE_SIZE); |
b3b00c78 | 2979 | mmio->iomem.subpage = true; |
db7b5426 | 2980 | #if defined(DEBUG_SUBPAGE) |
016e9d62 AK |
2981 | printf("%s: %p base " TARGET_FMT_plx " len %08x\n", __func__, |
2982 | mmio, base, TARGET_PAGE_SIZE); | |
db7b5426 | 2983 | #endif |
db7b5426 BS |
2984 | |
2985 | return mmio; | |
2986 | } | |
2987 | ||
16620684 | 2988 | static uint16_t dummy_section(PhysPageMap *map, FlatView *fv, MemoryRegion *mr) |
5312bd8b | 2989 | { |
16620684 | 2990 | assert(fv); |
5312bd8b | 2991 | MemoryRegionSection section = { |
16620684 | 2992 | .fv = fv, |
5312bd8b AK |
2993 | .mr = mr, |
2994 | .offset_within_address_space = 0, | |
2995 | .offset_within_region = 0, | |
052e87b0 | 2996 | .size = int128_2_64(), |
5312bd8b AK |
2997 | }; |
2998 | ||
53cb28cb | 2999 | return phys_section_add(map, §ion); |
5312bd8b AK |
3000 | } |
3001 | ||
8af36743 PM |
3002 | static void readonly_mem_write(void *opaque, hwaddr addr, |
3003 | uint64_t val, unsigned size) | |
3004 | { | |
3005 | /* Ignore any write to ROM. */ | |
3006 | } | |
3007 | ||
3008 | static bool readonly_mem_accepts(void *opaque, hwaddr addr, | |
8372d383 PM |
3009 | unsigned size, bool is_write, |
3010 | MemTxAttrs attrs) | |
8af36743 PM |
3011 | { |
3012 | return is_write; | |
3013 | } | |
3014 | ||
3015 | /* This will only be used for writes, because reads are special cased | |
3016 | * to directly access the underlying host ram. | |
3017 | */ | |
3018 | static const MemoryRegionOps readonly_mem_ops = { | |
3019 | .write = readonly_mem_write, | |
3020 | .valid.accepts = readonly_mem_accepts, | |
3021 | .endianness = DEVICE_NATIVE_ENDIAN, | |
3022 | .valid = { | |
3023 | .min_access_size = 1, | |
3024 | .max_access_size = 8, | |
3025 | .unaligned = false, | |
3026 | }, | |
3027 | .impl = { | |
3028 | .min_access_size = 1, | |
3029 | .max_access_size = 8, | |
3030 | .unaligned = false, | |
3031 | }, | |
3032 | }; | |
3033 | ||
2d54f194 PM |
3034 | MemoryRegionSection *iotlb_to_section(CPUState *cpu, |
3035 | hwaddr index, MemTxAttrs attrs) | |
aa102231 | 3036 | { |
a54c87b6 PM |
3037 | int asidx = cpu_asidx_from_attrs(cpu, attrs); |
3038 | CPUAddressSpace *cpuas = &cpu->cpu_ases[asidx]; | |
32857f4d | 3039 | AddressSpaceDispatch *d = atomic_rcu_read(&cpuas->memory_dispatch); |
79e2b9ae | 3040 | MemoryRegionSection *sections = d->map.sections; |
9d82b5a7 | 3041 | |
2d54f194 | 3042 | return §ions[index & ~TARGET_PAGE_MASK]; |
aa102231 AK |
3043 | } |
3044 | ||
e9179ce1 AK |
3045 | static void io_mem_init(void) |
3046 | { | |
8af36743 PM |
3047 | memory_region_init_io(&io_mem_rom, NULL, &readonly_mem_ops, |
3048 | NULL, NULL, UINT64_MAX); | |
2c9b15ca | 3049 | memory_region_init_io(&io_mem_unassigned, NULL, &unassigned_mem_ops, NULL, |
1f6245e5 | 3050 | NULL, UINT64_MAX); |
8d04fb55 JK |
3051 | |
3052 | /* io_mem_notdirty calls tb_invalidate_phys_page_fast, | |
3053 | * which can be called without the iothread mutex. | |
3054 | */ | |
2c9b15ca | 3055 | memory_region_init_io(&io_mem_notdirty, NULL, ¬dirty_mem_ops, NULL, |
1f6245e5 | 3056 | NULL, UINT64_MAX); |
8d04fb55 | 3057 | memory_region_clear_global_locking(&io_mem_notdirty); |
e9179ce1 AK |
3058 | } |
3059 | ||
8629d3fc | 3060 | AddressSpaceDispatch *address_space_dispatch_new(FlatView *fv) |
00752703 | 3061 | { |
53cb28cb MA |
3062 | AddressSpaceDispatch *d = g_new0(AddressSpaceDispatch, 1); |
3063 | uint16_t n; | |
3064 | ||
16620684 | 3065 | n = dummy_section(&d->map, fv, &io_mem_unassigned); |
53cb28cb | 3066 | assert(n == PHYS_SECTION_UNASSIGNED); |
16620684 | 3067 | n = dummy_section(&d->map, fv, &io_mem_notdirty); |
53cb28cb | 3068 | assert(n == PHYS_SECTION_NOTDIRTY); |
16620684 | 3069 | n = dummy_section(&d->map, fv, &io_mem_rom); |
53cb28cb | 3070 | assert(n == PHYS_SECTION_ROM); |
00752703 | 3071 | |
9736e55b | 3072 | d->phys_map = (PhysPageEntry) { .ptr = PHYS_MAP_NODE_NIL, .skip = 1 }; |
66a6df1d AK |
3073 | |
3074 | return d; | |
00752703 PB |
3075 | } |
3076 | ||
66a6df1d | 3077 | void address_space_dispatch_free(AddressSpaceDispatch *d) |
79e2b9ae PB |
3078 | { |
3079 | phys_sections_free(&d->map); | |
3080 | g_free(d); | |
3081 | } | |
3082 | ||
9458a9a1 PB |
3083 | static void do_nothing(CPUState *cpu, run_on_cpu_data d) |
3084 | { | |
3085 | } | |
3086 | ||
3087 | static void tcg_log_global_after_sync(MemoryListener *listener) | |
3088 | { | |
3089 | CPUAddressSpace *cpuas; | |
3090 | ||
3091 | /* Wait for the CPU to end the current TB. This avoids the following | |
3092 | * incorrect race: | |
3093 | * | |
3094 | * vCPU migration | |
3095 | * ---------------------- ------------------------- | |
3096 | * TLB check -> slow path | |
3097 | * notdirty_mem_write | |
3098 | * write to RAM | |
3099 | * mark dirty | |
3100 | * clear dirty flag | |
3101 | * TLB check -> fast path | |
3102 | * read memory | |
3103 | * write to RAM | |
3104 | * | |
3105 | * by pushing the migration thread's memory read after the vCPU thread has | |
3106 | * written the memory. | |
3107 | */ | |
3108 | cpuas = container_of(listener, CPUAddressSpace, tcg_as_listener); | |
3109 | run_on_cpu(cpuas->cpu, do_nothing, RUN_ON_CPU_NULL); | |
3110 | } | |
3111 | ||
1d71148e | 3112 | static void tcg_commit(MemoryListener *listener) |
50c1e149 | 3113 | { |
32857f4d PM |
3114 | CPUAddressSpace *cpuas; |
3115 | AddressSpaceDispatch *d; | |
117712c3 | 3116 | |
f28d0dfd | 3117 | assert(tcg_enabled()); |
117712c3 AK |
3118 | /* since each CPU stores ram addresses in its TLB cache, we must |
3119 | reset the modified entries */ | |
32857f4d PM |
3120 | cpuas = container_of(listener, CPUAddressSpace, tcg_as_listener); |
3121 | cpu_reloading_memory_map(); | |
3122 | /* The CPU and TLB are protected by the iothread lock. | |
3123 | * We reload the dispatch pointer now because cpu_reloading_memory_map() | |
3124 | * may have split the RCU critical section. | |
3125 | */ | |
66a6df1d | 3126 | d = address_space_to_dispatch(cpuas->as); |
f35e44e7 | 3127 | atomic_rcu_set(&cpuas->memory_dispatch, d); |
d10eb08f | 3128 | tlb_flush(cpuas->cpu); |
50c1e149 AK |
3129 | } |
3130 | ||
62152b8a AK |
3131 | static void memory_map_init(void) |
3132 | { | |
7267c094 | 3133 | system_memory = g_malloc(sizeof(*system_memory)); |
03f49957 | 3134 | |
57271d63 | 3135 | memory_region_init(system_memory, NULL, "system", UINT64_MAX); |
7dca8043 | 3136 | address_space_init(&address_space_memory, system_memory, "memory"); |
309cb471 | 3137 | |
7267c094 | 3138 | system_io = g_malloc(sizeof(*system_io)); |
3bb28b72 JK |
3139 | memory_region_init_io(system_io, NULL, &unassigned_io_ops, NULL, "io", |
3140 | 65536); | |
7dca8043 | 3141 | address_space_init(&address_space_io, system_io, "I/O"); |
62152b8a AK |
3142 | } |
3143 | ||
3144 | MemoryRegion *get_system_memory(void) | |
3145 | { | |
3146 | return system_memory; | |
3147 | } | |
3148 | ||
309cb471 AK |
3149 | MemoryRegion *get_system_io(void) |
3150 | { | |
3151 | return system_io; | |
3152 | } | |
3153 | ||
e2eef170 PB |
3154 | #endif /* !defined(CONFIG_USER_ONLY) */ |
3155 | ||
13eb76e0 FB |
3156 | /* physical memory access (slow version, mainly for debug) */ |
3157 | #if defined(CONFIG_USER_ONLY) | |
f17ec444 | 3158 | int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr, |
0c249ff7 | 3159 | uint8_t *buf, target_ulong len, int is_write) |
13eb76e0 | 3160 | { |
0c249ff7 LZ |
3161 | int flags; |
3162 | target_ulong l, page; | |
53a5960a | 3163 | void * p; |
13eb76e0 FB |
3164 | |
3165 | while (len > 0) { | |
3166 | page = addr & TARGET_PAGE_MASK; | |
3167 | l = (page + TARGET_PAGE_SIZE) - addr; | |
3168 | if (l > len) | |
3169 | l = len; | |
3170 | flags = page_get_flags(page); | |
3171 | if (!(flags & PAGE_VALID)) | |
a68fe89c | 3172 | return -1; |
13eb76e0 FB |
3173 | if (is_write) { |
3174 | if (!(flags & PAGE_WRITE)) | |
a68fe89c | 3175 | return -1; |
579a97f7 | 3176 | /* XXX: this code should not depend on lock_user */ |
72fb7daa | 3177 | if (!(p = lock_user(VERIFY_WRITE, addr, l, 0))) |
a68fe89c | 3178 | return -1; |
72fb7daa AJ |
3179 | memcpy(p, buf, l); |
3180 | unlock_user(p, addr, l); | |
13eb76e0 FB |
3181 | } else { |
3182 | if (!(flags & PAGE_READ)) | |
a68fe89c | 3183 | return -1; |
579a97f7 | 3184 | /* XXX: this code should not depend on lock_user */ |
72fb7daa | 3185 | if (!(p = lock_user(VERIFY_READ, addr, l, 1))) |
a68fe89c | 3186 | return -1; |
72fb7daa | 3187 | memcpy(buf, p, l); |
5b257578 | 3188 | unlock_user(p, addr, 0); |
13eb76e0 FB |
3189 | } |
3190 | len -= l; | |
3191 | buf += l; | |
3192 | addr += l; | |
3193 | } | |
a68fe89c | 3194 | return 0; |
13eb76e0 | 3195 | } |
8df1cd07 | 3196 | |
13eb76e0 | 3197 | #else |
51d7a9eb | 3198 | |
845b6214 | 3199 | static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr, |
a8170e5e | 3200 | hwaddr length) |
51d7a9eb | 3201 | { |
e87f7778 | 3202 | uint8_t dirty_log_mask = memory_region_get_dirty_log_mask(mr); |
0878d0e1 PB |
3203 | addr += memory_region_get_ram_addr(mr); |
3204 | ||
e87f7778 PB |
3205 | /* No early return if dirty_log_mask is or becomes 0, because |
3206 | * cpu_physical_memory_set_dirty_range will still call | |
3207 | * xen_modified_memory. | |
3208 | */ | |
3209 | if (dirty_log_mask) { | |
3210 | dirty_log_mask = | |
3211 | cpu_physical_memory_range_includes_clean(addr, length, dirty_log_mask); | |
3212 | } | |
3213 | if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) { | |
5aa1ef71 | 3214 | assert(tcg_enabled()); |
e87f7778 PB |
3215 | tb_invalidate_phys_range(addr, addr + length); |
3216 | dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE); | |
51d7a9eb | 3217 | } |
e87f7778 | 3218 | cpu_physical_memory_set_dirty_range(addr, length, dirty_log_mask); |
51d7a9eb AP |
3219 | } |
3220 | ||
047be4ed SH |
3221 | void memory_region_flush_rom_device(MemoryRegion *mr, hwaddr addr, hwaddr size) |
3222 | { | |
3223 | /* | |
3224 | * In principle this function would work on other memory region types too, | |
3225 | * but the ROM device use case is the only one where this operation is | |
3226 | * necessary. Other memory regions should use the | |
3227 | * address_space_read/write() APIs. | |
3228 | */ | |
3229 | assert(memory_region_is_romd(mr)); | |
3230 | ||
3231 | invalidate_and_set_dirty(mr, addr, size); | |
3232 | } | |
3233 | ||
23326164 | 3234 | static int memory_access_size(MemoryRegion *mr, unsigned l, hwaddr addr) |
82f2563f | 3235 | { |
e1622f4b | 3236 | unsigned access_size_max = mr->ops->valid.max_access_size; |
23326164 RH |
3237 | |
3238 | /* Regions are assumed to support 1-4 byte accesses unless | |
3239 | otherwise specified. */ | |
23326164 RH |
3240 | if (access_size_max == 0) { |
3241 | access_size_max = 4; | |
3242 | } | |
3243 | ||
3244 | /* Bound the maximum access by the alignment of the address. */ | |
3245 | if (!mr->ops->impl.unaligned) { | |
3246 | unsigned align_size_max = addr & -addr; | |
3247 | if (align_size_max != 0 && align_size_max < access_size_max) { | |
3248 | access_size_max = align_size_max; | |
3249 | } | |
82f2563f | 3250 | } |
23326164 RH |
3251 | |
3252 | /* Don't attempt accesses larger than the maximum. */ | |
3253 | if (l > access_size_max) { | |
3254 | l = access_size_max; | |
82f2563f | 3255 | } |
6554f5c0 | 3256 | l = pow2floor(l); |
23326164 RH |
3257 | |
3258 | return l; | |
82f2563f PB |
3259 | } |
3260 | ||
4840f10e | 3261 | static bool prepare_mmio_access(MemoryRegion *mr) |
125b3806 | 3262 | { |
4840f10e JK |
3263 | bool unlocked = !qemu_mutex_iothread_locked(); |
3264 | bool release_lock = false; | |
3265 | ||
3266 | if (unlocked && mr->global_locking) { | |
3267 | qemu_mutex_lock_iothread(); | |
3268 | unlocked = false; | |
3269 | release_lock = true; | |
3270 | } | |
125b3806 | 3271 | if (mr->flush_coalesced_mmio) { |
4840f10e JK |
3272 | if (unlocked) { |
3273 | qemu_mutex_lock_iothread(); | |
3274 | } | |
125b3806 | 3275 | qemu_flush_coalesced_mmio_buffer(); |
4840f10e JK |
3276 | if (unlocked) { |
3277 | qemu_mutex_unlock_iothread(); | |
3278 | } | |
125b3806 | 3279 | } |
4840f10e JK |
3280 | |
3281 | return release_lock; | |
125b3806 PB |
3282 | } |
3283 | ||
a203ac70 | 3284 | /* Called within RCU critical section. */ |
16620684 AK |
3285 | static MemTxResult flatview_write_continue(FlatView *fv, hwaddr addr, |
3286 | MemTxAttrs attrs, | |
3287 | const uint8_t *buf, | |
0c249ff7 | 3288 | hwaddr len, hwaddr addr1, |
16620684 | 3289 | hwaddr l, MemoryRegion *mr) |
13eb76e0 | 3290 | { |
13eb76e0 | 3291 | uint8_t *ptr; |
791af8c8 | 3292 | uint64_t val; |
3b643495 | 3293 | MemTxResult result = MEMTX_OK; |
4840f10e | 3294 | bool release_lock = false; |
3b46e624 | 3295 | |
a203ac70 | 3296 | for (;;) { |
eb7eeb88 PB |
3297 | if (!memory_access_is_direct(mr, true)) { |
3298 | release_lock |= prepare_mmio_access(mr); | |
3299 | l = memory_access_size(mr, l, addr1); | |
3300 | /* XXX: could force current_cpu to NULL to avoid | |
3301 | potential bugs */ | |
9bf825bf | 3302 | val = ldn_he_p(buf, l); |
3d9e7c3e | 3303 | result |= memory_region_dispatch_write(mr, addr1, val, |
9bf825bf | 3304 | size_memop(l), attrs); |
13eb76e0 | 3305 | } else { |
eb7eeb88 | 3306 | /* RAM case */ |
f5aa69bd | 3307 | ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l, false); |
eb7eeb88 PB |
3308 | memcpy(ptr, buf, l); |
3309 | invalidate_and_set_dirty(mr, addr1, l); | |
13eb76e0 | 3310 | } |
4840f10e JK |
3311 | |
3312 | if (release_lock) { | |
3313 | qemu_mutex_unlock_iothread(); | |
3314 | release_lock = false; | |
3315 | } | |
3316 | ||
13eb76e0 FB |
3317 | len -= l; |
3318 | buf += l; | |
3319 | addr += l; | |
a203ac70 PB |
3320 | |
3321 | if (!len) { | |
3322 | break; | |
3323 | } | |
3324 | ||
3325 | l = len; | |
efa99a2f | 3326 | mr = flatview_translate(fv, addr, &addr1, &l, true, attrs); |
13eb76e0 | 3327 | } |
fd8aaa76 | 3328 | |
3b643495 | 3329 | return result; |
13eb76e0 | 3330 | } |
8df1cd07 | 3331 | |
4c6ebbb3 | 3332 | /* Called from RCU critical section. */ |
16620684 | 3333 | static MemTxResult flatview_write(FlatView *fv, hwaddr addr, MemTxAttrs attrs, |
0c249ff7 | 3334 | const uint8_t *buf, hwaddr len) |
ac1970fb | 3335 | { |
eb7eeb88 | 3336 | hwaddr l; |
eb7eeb88 PB |
3337 | hwaddr addr1; |
3338 | MemoryRegion *mr; | |
3339 | MemTxResult result = MEMTX_OK; | |
eb7eeb88 | 3340 | |
4c6ebbb3 | 3341 | l = len; |
efa99a2f | 3342 | mr = flatview_translate(fv, addr, &addr1, &l, true, attrs); |
4c6ebbb3 PB |
3343 | result = flatview_write_continue(fv, addr, attrs, buf, len, |
3344 | addr1, l, mr); | |
a203ac70 PB |
3345 | |
3346 | return result; | |
3347 | } | |
3348 | ||
3349 | /* Called within RCU critical section. */ | |
16620684 AK |
3350 | MemTxResult flatview_read_continue(FlatView *fv, hwaddr addr, |
3351 | MemTxAttrs attrs, uint8_t *buf, | |
0c249ff7 | 3352 | hwaddr len, hwaddr addr1, hwaddr l, |
16620684 | 3353 | MemoryRegion *mr) |
a203ac70 PB |
3354 | { |
3355 | uint8_t *ptr; | |
3356 | uint64_t val; | |
3357 | MemTxResult result = MEMTX_OK; | |
3358 | bool release_lock = false; | |
eb7eeb88 | 3359 | |
a203ac70 | 3360 | for (;;) { |
eb7eeb88 PB |
3361 | if (!memory_access_is_direct(mr, false)) { |
3362 | /* I/O case */ | |
3363 | release_lock |= prepare_mmio_access(mr); | |
3364 | l = memory_access_size(mr, l, addr1); | |
3d9e7c3e | 3365 | result |= memory_region_dispatch_read(mr, addr1, &val, |
9bf825bf TN |
3366 | size_memop(l), attrs); |
3367 | stn_he_p(buf, l, val); | |
eb7eeb88 PB |
3368 | } else { |
3369 | /* RAM case */ | |
f5aa69bd | 3370 | ptr = qemu_ram_ptr_length(mr->ram_block, addr1, &l, false); |
eb7eeb88 PB |
3371 | memcpy(buf, ptr, l); |
3372 | } | |
3373 | ||
3374 | if (release_lock) { | |
3375 | qemu_mutex_unlock_iothread(); | |
3376 | release_lock = false; | |
3377 | } | |
3378 | ||
3379 | len -= l; | |
3380 | buf += l; | |
3381 | addr += l; | |
a203ac70 PB |
3382 | |
3383 | if (!len) { | |
3384 | break; | |
3385 | } | |
3386 | ||
3387 | l = len; | |
efa99a2f | 3388 | mr = flatview_translate(fv, addr, &addr1, &l, false, attrs); |
a203ac70 PB |
3389 | } |
3390 | ||
3391 | return result; | |
3392 | } | |
3393 | ||
b2a44fca PB |
3394 | /* Called from RCU critical section. */ |
3395 | static MemTxResult flatview_read(FlatView *fv, hwaddr addr, | |
0c249ff7 | 3396 | MemTxAttrs attrs, uint8_t *buf, hwaddr len) |
a203ac70 PB |
3397 | { |
3398 | hwaddr l; | |
3399 | hwaddr addr1; | |
3400 | MemoryRegion *mr; | |
eb7eeb88 | 3401 | |
b2a44fca | 3402 | l = len; |
efa99a2f | 3403 | mr = flatview_translate(fv, addr, &addr1, &l, false, attrs); |
b2a44fca PB |
3404 | return flatview_read_continue(fv, addr, attrs, buf, len, |
3405 | addr1, l, mr); | |
ac1970fb AK |
3406 | } |
3407 | ||
b2a44fca | 3408 | MemTxResult address_space_read_full(AddressSpace *as, hwaddr addr, |
0c249ff7 | 3409 | MemTxAttrs attrs, uint8_t *buf, hwaddr len) |
b2a44fca PB |
3410 | { |
3411 | MemTxResult result = MEMTX_OK; | |
3412 | FlatView *fv; | |
3413 | ||
3414 | if (len > 0) { | |
3415 | rcu_read_lock(); | |
3416 | fv = address_space_to_flatview(as); | |
3417 | result = flatview_read(fv, addr, attrs, buf, len); | |
3418 | rcu_read_unlock(); | |
3419 | } | |
3420 | ||
3421 | return result; | |
3422 | } | |
3423 | ||
4c6ebbb3 PB |
3424 | MemTxResult address_space_write(AddressSpace *as, hwaddr addr, |
3425 | MemTxAttrs attrs, | |
0c249ff7 | 3426 | const uint8_t *buf, hwaddr len) |
4c6ebbb3 PB |
3427 | { |
3428 | MemTxResult result = MEMTX_OK; | |
3429 | FlatView *fv; | |
3430 | ||
3431 | if (len > 0) { | |
3432 | rcu_read_lock(); | |
3433 | fv = address_space_to_flatview(as); | |
3434 | result = flatview_write(fv, addr, attrs, buf, len); | |
3435 | rcu_read_unlock(); | |
3436 | } | |
3437 | ||
3438 | return result; | |
3439 | } | |
3440 | ||
db84fd97 | 3441 | MemTxResult address_space_rw(AddressSpace *as, hwaddr addr, MemTxAttrs attrs, |
0c249ff7 | 3442 | uint8_t *buf, hwaddr len, bool is_write) |
db84fd97 PB |
3443 | { |
3444 | if (is_write) { | |
3445 | return address_space_write(as, addr, attrs, buf, len); | |
3446 | } else { | |
3447 | return address_space_read_full(as, addr, attrs, buf, len); | |
3448 | } | |
3449 | } | |
3450 | ||
a8170e5e | 3451 | void cpu_physical_memory_rw(hwaddr addr, uint8_t *buf, |
0c249ff7 | 3452 | hwaddr len, int is_write) |
ac1970fb | 3453 | { |
5c9eb028 PM |
3454 | address_space_rw(&address_space_memory, addr, MEMTXATTRS_UNSPECIFIED, |
3455 | buf, len, is_write); | |
ac1970fb AK |
3456 | } |
3457 | ||
582b55a9 AG |
3458 | enum write_rom_type { |
3459 | WRITE_DATA, | |
3460 | FLUSH_CACHE, | |
3461 | }; | |
3462 | ||
75693e14 PM |
3463 | static inline MemTxResult address_space_write_rom_internal(AddressSpace *as, |
3464 | hwaddr addr, | |
3465 | MemTxAttrs attrs, | |
3466 | const uint8_t *buf, | |
0c249ff7 | 3467 | hwaddr len, |
75693e14 | 3468 | enum write_rom_type type) |
d0ecd2aa | 3469 | { |
149f54b5 | 3470 | hwaddr l; |
d0ecd2aa | 3471 | uint8_t *ptr; |
149f54b5 | 3472 | hwaddr addr1; |
5c8a00ce | 3473 | MemoryRegion *mr; |
3b46e624 | 3474 | |
41063e1e | 3475 | rcu_read_lock(); |
d0ecd2aa | 3476 | while (len > 0) { |
149f54b5 | 3477 | l = len; |
75693e14 | 3478 | mr = address_space_translate(as, addr, &addr1, &l, true, attrs); |
3b46e624 | 3479 | |
5c8a00ce PB |
3480 | if (!(memory_region_is_ram(mr) || |
3481 | memory_region_is_romd(mr))) { | |
b242e0e0 | 3482 | l = memory_access_size(mr, l, addr1); |
d0ecd2aa | 3483 | } else { |
d0ecd2aa | 3484 | /* ROM/RAM case */ |
0878d0e1 | 3485 | ptr = qemu_map_ram_ptr(mr->ram_block, addr1); |
582b55a9 AG |
3486 | switch (type) { |
3487 | case WRITE_DATA: | |
3488 | memcpy(ptr, buf, l); | |
845b6214 | 3489 | invalidate_and_set_dirty(mr, addr1, l); |
582b55a9 AG |
3490 | break; |
3491 | case FLUSH_CACHE: | |
3492 | flush_icache_range((uintptr_t)ptr, (uintptr_t)ptr + l); | |
3493 | break; | |
3494 | } | |
d0ecd2aa FB |
3495 | } |
3496 | len -= l; | |
3497 | buf += l; | |
3498 | addr += l; | |
3499 | } | |
41063e1e | 3500 | rcu_read_unlock(); |
75693e14 | 3501 | return MEMTX_OK; |
d0ecd2aa FB |
3502 | } |
3503 | ||
582b55a9 | 3504 | /* used for ROM loading : can write in RAM and ROM */ |
3c8133f9 PM |
3505 | MemTxResult address_space_write_rom(AddressSpace *as, hwaddr addr, |
3506 | MemTxAttrs attrs, | |
0c249ff7 | 3507 | const uint8_t *buf, hwaddr len) |
582b55a9 | 3508 | { |
3c8133f9 PM |
3509 | return address_space_write_rom_internal(as, addr, attrs, |
3510 | buf, len, WRITE_DATA); | |
582b55a9 AG |
3511 | } |
3512 | ||
0c249ff7 | 3513 | void cpu_flush_icache_range(hwaddr start, hwaddr len) |
582b55a9 AG |
3514 | { |
3515 | /* | |
3516 | * This function should do the same thing as an icache flush that was | |
3517 | * triggered from within the guest. For TCG we are always cache coherent, | |
3518 | * so there is no need to flush anything. For KVM / Xen we need to flush | |
3519 | * the host's instruction cache at least. | |
3520 | */ | |
3521 | if (tcg_enabled()) { | |
3522 | return; | |
3523 | } | |
3524 | ||
75693e14 PM |
3525 | address_space_write_rom_internal(&address_space_memory, |
3526 | start, MEMTXATTRS_UNSPECIFIED, | |
3527 | NULL, len, FLUSH_CACHE); | |
582b55a9 AG |
3528 | } |
3529 | ||
6d16c2f8 | 3530 | typedef struct { |
d3e71559 | 3531 | MemoryRegion *mr; |
6d16c2f8 | 3532 | void *buffer; |
a8170e5e AK |
3533 | hwaddr addr; |
3534 | hwaddr len; | |
c2cba0ff | 3535 | bool in_use; |
6d16c2f8 AL |
3536 | } BounceBuffer; |
3537 | ||
3538 | static BounceBuffer bounce; | |
3539 | ||
ba223c29 | 3540 | typedef struct MapClient { |
e95205e1 | 3541 | QEMUBH *bh; |
72cf2d4f | 3542 | QLIST_ENTRY(MapClient) link; |
ba223c29 AL |
3543 | } MapClient; |
3544 | ||
38e047b5 | 3545 | QemuMutex map_client_list_lock; |
b58deb34 | 3546 | static QLIST_HEAD(, MapClient) map_client_list |
72cf2d4f | 3547 | = QLIST_HEAD_INITIALIZER(map_client_list); |
ba223c29 | 3548 | |
e95205e1 FZ |
3549 | static void cpu_unregister_map_client_do(MapClient *client) |
3550 | { | |
3551 | QLIST_REMOVE(client, link); | |
3552 | g_free(client); | |
3553 | } | |
3554 | ||
33b6c2ed FZ |
3555 | static void cpu_notify_map_clients_locked(void) |
3556 | { | |
3557 | MapClient *client; | |
3558 | ||
3559 | while (!QLIST_EMPTY(&map_client_list)) { | |
3560 | client = QLIST_FIRST(&map_client_list); | |
e95205e1 FZ |
3561 | qemu_bh_schedule(client->bh); |
3562 | cpu_unregister_map_client_do(client); | |
33b6c2ed FZ |
3563 | } |
3564 | } | |
3565 | ||
e95205e1 | 3566 | void cpu_register_map_client(QEMUBH *bh) |
ba223c29 | 3567 | { |
7267c094 | 3568 | MapClient *client = g_malloc(sizeof(*client)); |
ba223c29 | 3569 | |
38e047b5 | 3570 | qemu_mutex_lock(&map_client_list_lock); |
e95205e1 | 3571 | client->bh = bh; |
72cf2d4f | 3572 | QLIST_INSERT_HEAD(&map_client_list, client, link); |
33b6c2ed FZ |
3573 | if (!atomic_read(&bounce.in_use)) { |
3574 | cpu_notify_map_clients_locked(); | |
3575 | } | |
38e047b5 | 3576 | qemu_mutex_unlock(&map_client_list_lock); |
ba223c29 AL |
3577 | } |
3578 | ||
38e047b5 | 3579 | void cpu_exec_init_all(void) |
ba223c29 | 3580 | { |
38e047b5 | 3581 | qemu_mutex_init(&ram_list.mutex); |
20bccb82 PM |
3582 | /* The data structures we set up here depend on knowing the page size, |
3583 | * so no more changes can be made after this point. | |
3584 | * In an ideal world, nothing we did before we had finished the | |
3585 | * machine setup would care about the target page size, and we could | |
3586 | * do this much later, rather than requiring board models to state | |
3587 | * up front what their requirements are. | |
3588 | */ | |
3589 | finalize_target_page_bits(); | |
38e047b5 | 3590 | io_mem_init(); |
680a4783 | 3591 | memory_map_init(); |
38e047b5 | 3592 | qemu_mutex_init(&map_client_list_lock); |
ba223c29 AL |
3593 | } |
3594 | ||
e95205e1 | 3595 | void cpu_unregister_map_client(QEMUBH *bh) |
ba223c29 AL |
3596 | { |
3597 | MapClient *client; | |
3598 | ||
e95205e1 FZ |
3599 | qemu_mutex_lock(&map_client_list_lock); |
3600 | QLIST_FOREACH(client, &map_client_list, link) { | |
3601 | if (client->bh == bh) { | |
3602 | cpu_unregister_map_client_do(client); | |
3603 | break; | |
3604 | } | |
ba223c29 | 3605 | } |
e95205e1 | 3606 | qemu_mutex_unlock(&map_client_list_lock); |
ba223c29 AL |
3607 | } |
3608 | ||
3609 | static void cpu_notify_map_clients(void) | |
3610 | { | |
38e047b5 | 3611 | qemu_mutex_lock(&map_client_list_lock); |
33b6c2ed | 3612 | cpu_notify_map_clients_locked(); |
38e047b5 | 3613 | qemu_mutex_unlock(&map_client_list_lock); |
ba223c29 AL |
3614 | } |
3615 | ||
0c249ff7 | 3616 | static bool flatview_access_valid(FlatView *fv, hwaddr addr, hwaddr len, |
eace72b7 | 3617 | bool is_write, MemTxAttrs attrs) |
51644ab7 | 3618 | { |
5c8a00ce | 3619 | MemoryRegion *mr; |
51644ab7 PB |
3620 | hwaddr l, xlat; |
3621 | ||
3622 | while (len > 0) { | |
3623 | l = len; | |
efa99a2f | 3624 | mr = flatview_translate(fv, addr, &xlat, &l, is_write, attrs); |
5c8a00ce PB |
3625 | if (!memory_access_is_direct(mr, is_write)) { |
3626 | l = memory_access_size(mr, l, addr); | |
eace72b7 | 3627 | if (!memory_region_access_valid(mr, xlat, l, is_write, attrs)) { |
51644ab7 PB |
3628 | return false; |
3629 | } | |
3630 | } | |
3631 | ||
3632 | len -= l; | |
3633 | addr += l; | |
3634 | } | |
3635 | return true; | |
3636 | } | |
3637 | ||
16620684 | 3638 | bool address_space_access_valid(AddressSpace *as, hwaddr addr, |
0c249ff7 | 3639 | hwaddr len, bool is_write, |
fddffa42 | 3640 | MemTxAttrs attrs) |
16620684 | 3641 | { |
11e732a5 PB |
3642 | FlatView *fv; |
3643 | bool result; | |
3644 | ||
3645 | rcu_read_lock(); | |
3646 | fv = address_space_to_flatview(as); | |
eace72b7 | 3647 | result = flatview_access_valid(fv, addr, len, is_write, attrs); |
11e732a5 PB |
3648 | rcu_read_unlock(); |
3649 | return result; | |
16620684 AK |
3650 | } |
3651 | ||
715c31ec | 3652 | static hwaddr |
16620684 | 3653 | flatview_extend_translation(FlatView *fv, hwaddr addr, |
53d0790d PM |
3654 | hwaddr target_len, |
3655 | MemoryRegion *mr, hwaddr base, hwaddr len, | |
3656 | bool is_write, MemTxAttrs attrs) | |
715c31ec PB |
3657 | { |
3658 | hwaddr done = 0; | |
3659 | hwaddr xlat; | |
3660 | MemoryRegion *this_mr; | |
3661 | ||
3662 | for (;;) { | |
3663 | target_len -= len; | |
3664 | addr += len; | |
3665 | done += len; | |
3666 | if (target_len == 0) { | |
3667 | return done; | |
3668 | } | |
3669 | ||
3670 | len = target_len; | |
16620684 | 3671 | this_mr = flatview_translate(fv, addr, &xlat, |
efa99a2f | 3672 | &len, is_write, attrs); |
715c31ec PB |
3673 | if (this_mr != mr || xlat != base + done) { |
3674 | return done; | |
3675 | } | |
3676 | } | |
3677 | } | |
3678 | ||
6d16c2f8 AL |
3679 | /* Map a physical memory region into a host virtual address. |
3680 | * May map a subset of the requested range, given by and returned in *plen. | |
3681 | * May return NULL if resources needed to perform the mapping are exhausted. | |
3682 | * Use only for reads OR writes - not for read-modify-write operations. | |
ba223c29 AL |
3683 | * Use cpu_register_map_client() to know when retrying the map operation is |
3684 | * likely to succeed. | |
6d16c2f8 | 3685 | */ |
ac1970fb | 3686 | void *address_space_map(AddressSpace *as, |
a8170e5e AK |
3687 | hwaddr addr, |
3688 | hwaddr *plen, | |
f26404fb PM |
3689 | bool is_write, |
3690 | MemTxAttrs attrs) | |
6d16c2f8 | 3691 | { |
a8170e5e | 3692 | hwaddr len = *plen; |
715c31ec PB |
3693 | hwaddr l, xlat; |
3694 | MemoryRegion *mr; | |
e81bcda5 | 3695 | void *ptr; |
ad0c60fa | 3696 | FlatView *fv; |
6d16c2f8 | 3697 | |
e3127ae0 PB |
3698 | if (len == 0) { |
3699 | return NULL; | |
3700 | } | |
38bee5dc | 3701 | |
e3127ae0 | 3702 | l = len; |
41063e1e | 3703 | rcu_read_lock(); |
ad0c60fa | 3704 | fv = address_space_to_flatview(as); |
efa99a2f | 3705 | mr = flatview_translate(fv, addr, &xlat, &l, is_write, attrs); |
41063e1e | 3706 | |
e3127ae0 | 3707 | if (!memory_access_is_direct(mr, is_write)) { |
c2cba0ff | 3708 | if (atomic_xchg(&bounce.in_use, true)) { |
41063e1e | 3709 | rcu_read_unlock(); |
e3127ae0 | 3710 | return NULL; |
6d16c2f8 | 3711 | } |
e85d9db5 KW |
3712 | /* Avoid unbounded allocations */ |
3713 | l = MIN(l, TARGET_PAGE_SIZE); | |
3714 | bounce.buffer = qemu_memalign(TARGET_PAGE_SIZE, l); | |
e3127ae0 PB |
3715 | bounce.addr = addr; |
3716 | bounce.len = l; | |
d3e71559 PB |
3717 | |
3718 | memory_region_ref(mr); | |
3719 | bounce.mr = mr; | |
e3127ae0 | 3720 | if (!is_write) { |
16620684 | 3721 | flatview_read(fv, addr, MEMTXATTRS_UNSPECIFIED, |
5c9eb028 | 3722 | bounce.buffer, l); |
8ab934f9 | 3723 | } |
6d16c2f8 | 3724 | |
41063e1e | 3725 | rcu_read_unlock(); |
e3127ae0 PB |
3726 | *plen = l; |
3727 | return bounce.buffer; | |
3728 | } | |
3729 | ||
e3127ae0 | 3730 | |
d3e71559 | 3731 | memory_region_ref(mr); |
16620684 | 3732 | *plen = flatview_extend_translation(fv, addr, len, mr, xlat, |
53d0790d | 3733 | l, is_write, attrs); |
f5aa69bd | 3734 | ptr = qemu_ram_ptr_length(mr->ram_block, xlat, plen, true); |
e81bcda5 PB |
3735 | rcu_read_unlock(); |
3736 | ||
3737 | return ptr; | |
6d16c2f8 AL |
3738 | } |
3739 | ||
ac1970fb | 3740 | /* Unmaps a memory region previously mapped by address_space_map(). |
6d16c2f8 AL |
3741 | * Will also mark the memory as dirty if is_write == 1. access_len gives |
3742 | * the amount of memory that was actually read or written by the caller. | |
3743 | */ | |
a8170e5e AK |
3744 | void address_space_unmap(AddressSpace *as, void *buffer, hwaddr len, |
3745 | int is_write, hwaddr access_len) | |
6d16c2f8 AL |
3746 | { |
3747 | if (buffer != bounce.buffer) { | |
d3e71559 PB |
3748 | MemoryRegion *mr; |
3749 | ram_addr_t addr1; | |
3750 | ||
07bdaa41 | 3751 | mr = memory_region_from_host(buffer, &addr1); |
d3e71559 | 3752 | assert(mr != NULL); |
6d16c2f8 | 3753 | if (is_write) { |
845b6214 | 3754 | invalidate_and_set_dirty(mr, addr1, access_len); |
6d16c2f8 | 3755 | } |
868bb33f | 3756 | if (xen_enabled()) { |
e41d7c69 | 3757 | xen_invalidate_map_cache_entry(buffer); |
050a0ddf | 3758 | } |
d3e71559 | 3759 | memory_region_unref(mr); |
6d16c2f8 AL |
3760 | return; |
3761 | } | |
3762 | if (is_write) { | |
5c9eb028 PM |
3763 | address_space_write(as, bounce.addr, MEMTXATTRS_UNSPECIFIED, |
3764 | bounce.buffer, access_len); | |
6d16c2f8 | 3765 | } |
f8a83245 | 3766 | qemu_vfree(bounce.buffer); |
6d16c2f8 | 3767 | bounce.buffer = NULL; |
d3e71559 | 3768 | memory_region_unref(bounce.mr); |
c2cba0ff | 3769 | atomic_mb_set(&bounce.in_use, false); |
ba223c29 | 3770 | cpu_notify_map_clients(); |
6d16c2f8 | 3771 | } |
d0ecd2aa | 3772 | |
a8170e5e AK |
3773 | void *cpu_physical_memory_map(hwaddr addr, |
3774 | hwaddr *plen, | |
ac1970fb AK |
3775 | int is_write) |
3776 | { | |
f26404fb PM |
3777 | return address_space_map(&address_space_memory, addr, plen, is_write, |
3778 | MEMTXATTRS_UNSPECIFIED); | |
ac1970fb AK |
3779 | } |
3780 | ||
a8170e5e AK |
3781 | void cpu_physical_memory_unmap(void *buffer, hwaddr len, |
3782 | int is_write, hwaddr access_len) | |
ac1970fb AK |
3783 | { |
3784 | return address_space_unmap(&address_space_memory, buffer, len, is_write, access_len); | |
3785 | } | |
3786 | ||
0ce265ff PB |
3787 | #define ARG1_DECL AddressSpace *as |
3788 | #define ARG1 as | |
3789 | #define SUFFIX | |
3790 | #define TRANSLATE(...) address_space_translate(as, __VA_ARGS__) | |
0ce265ff PB |
3791 | #define RCU_READ_LOCK(...) rcu_read_lock() |
3792 | #define RCU_READ_UNLOCK(...) rcu_read_unlock() | |
3793 | #include "memory_ldst.inc.c" | |
1e78bcc1 | 3794 | |
1f4e496e PB |
3795 | int64_t address_space_cache_init(MemoryRegionCache *cache, |
3796 | AddressSpace *as, | |
3797 | hwaddr addr, | |
3798 | hwaddr len, | |
3799 | bool is_write) | |
3800 | { | |
48564041 PB |
3801 | AddressSpaceDispatch *d; |
3802 | hwaddr l; | |
3803 | MemoryRegion *mr; | |
3804 | ||
3805 | assert(len > 0); | |
3806 | ||
3807 | l = len; | |
3808 | cache->fv = address_space_get_flatview(as); | |
3809 | d = flatview_to_dispatch(cache->fv); | |
3810 | cache->mrs = *address_space_translate_internal(d, addr, &cache->xlat, &l, true); | |
3811 | ||
3812 | mr = cache->mrs.mr; | |
3813 | memory_region_ref(mr); | |
3814 | if (memory_access_is_direct(mr, is_write)) { | |
53d0790d PM |
3815 | /* We don't care about the memory attributes here as we're only |
3816 | * doing this if we found actual RAM, which behaves the same | |
3817 | * regardless of attributes; so UNSPECIFIED is fine. | |
3818 | */ | |
48564041 | 3819 | l = flatview_extend_translation(cache->fv, addr, len, mr, |
53d0790d PM |
3820 | cache->xlat, l, is_write, |
3821 | MEMTXATTRS_UNSPECIFIED); | |
48564041 PB |
3822 | cache->ptr = qemu_ram_ptr_length(mr->ram_block, cache->xlat, &l, true); |
3823 | } else { | |
3824 | cache->ptr = NULL; | |
3825 | } | |
3826 | ||
3827 | cache->len = l; | |
3828 | cache->is_write = is_write; | |
3829 | return l; | |
1f4e496e PB |
3830 | } |
3831 | ||
3832 | void address_space_cache_invalidate(MemoryRegionCache *cache, | |
3833 | hwaddr addr, | |
3834 | hwaddr access_len) | |
3835 | { | |
48564041 PB |
3836 | assert(cache->is_write); |
3837 | if (likely(cache->ptr)) { | |
3838 | invalidate_and_set_dirty(cache->mrs.mr, addr + cache->xlat, access_len); | |
3839 | } | |
1f4e496e PB |
3840 | } |
3841 | ||
3842 | void address_space_cache_destroy(MemoryRegionCache *cache) | |
3843 | { | |
48564041 PB |
3844 | if (!cache->mrs.mr) { |
3845 | return; | |
3846 | } | |
3847 | ||
3848 | if (xen_enabled()) { | |
3849 | xen_invalidate_map_cache_entry(cache->ptr); | |
3850 | } | |
3851 | memory_region_unref(cache->mrs.mr); | |
3852 | flatview_unref(cache->fv); | |
3853 | cache->mrs.mr = NULL; | |
3854 | cache->fv = NULL; | |
3855 | } | |
3856 | ||
3857 | /* Called from RCU critical section. This function has the same | |
3858 | * semantics as address_space_translate, but it only works on a | |
3859 | * predefined range of a MemoryRegion that was mapped with | |
3860 | * address_space_cache_init. | |
3861 | */ | |
3862 | static inline MemoryRegion *address_space_translate_cached( | |
3863 | MemoryRegionCache *cache, hwaddr addr, hwaddr *xlat, | |
bc6b1cec | 3864 | hwaddr *plen, bool is_write, MemTxAttrs attrs) |
48564041 PB |
3865 | { |
3866 | MemoryRegionSection section; | |
3867 | MemoryRegion *mr; | |
3868 | IOMMUMemoryRegion *iommu_mr; | |
3869 | AddressSpace *target_as; | |
3870 | ||
3871 | assert(!cache->ptr); | |
3872 | *xlat = addr + cache->xlat; | |
3873 | ||
3874 | mr = cache->mrs.mr; | |
3875 | iommu_mr = memory_region_get_iommu(mr); | |
3876 | if (!iommu_mr) { | |
3877 | /* MMIO region. */ | |
3878 | return mr; | |
3879 | } | |
3880 | ||
3881 | section = address_space_translate_iommu(iommu_mr, xlat, plen, | |
3882 | NULL, is_write, true, | |
2f7b009c | 3883 | &target_as, attrs); |
48564041 PB |
3884 | return section.mr; |
3885 | } | |
3886 | ||
3887 | /* Called from RCU critical section. address_space_read_cached uses this | |
3888 | * out of line function when the target is an MMIO or IOMMU region. | |
3889 | */ | |
3890 | void | |
3891 | address_space_read_cached_slow(MemoryRegionCache *cache, hwaddr addr, | |
0c249ff7 | 3892 | void *buf, hwaddr len) |
48564041 PB |
3893 | { |
3894 | hwaddr addr1, l; | |
3895 | MemoryRegion *mr; | |
3896 | ||
3897 | l = len; | |
bc6b1cec PM |
3898 | mr = address_space_translate_cached(cache, addr, &addr1, &l, false, |
3899 | MEMTXATTRS_UNSPECIFIED); | |
48564041 PB |
3900 | flatview_read_continue(cache->fv, |
3901 | addr, MEMTXATTRS_UNSPECIFIED, buf, len, | |
3902 | addr1, l, mr); | |
3903 | } | |
3904 | ||
3905 | /* Called from RCU critical section. address_space_write_cached uses this | |
3906 | * out of line function when the target is an MMIO or IOMMU region. | |
3907 | */ | |
3908 | void | |
3909 | address_space_write_cached_slow(MemoryRegionCache *cache, hwaddr addr, | |
0c249ff7 | 3910 | const void *buf, hwaddr len) |
48564041 PB |
3911 | { |
3912 | hwaddr addr1, l; | |
3913 | MemoryRegion *mr; | |
3914 | ||
3915 | l = len; | |
bc6b1cec PM |
3916 | mr = address_space_translate_cached(cache, addr, &addr1, &l, true, |
3917 | MEMTXATTRS_UNSPECIFIED); | |
48564041 PB |
3918 | flatview_write_continue(cache->fv, |
3919 | addr, MEMTXATTRS_UNSPECIFIED, buf, len, | |
3920 | addr1, l, mr); | |
1f4e496e PB |
3921 | } |
3922 | ||
3923 | #define ARG1_DECL MemoryRegionCache *cache | |
3924 | #define ARG1 cache | |
48564041 PB |
3925 | #define SUFFIX _cached_slow |
3926 | #define TRANSLATE(...) address_space_translate_cached(cache, __VA_ARGS__) | |
48564041 PB |
3927 | #define RCU_READ_LOCK() ((void)0) |
3928 | #define RCU_READ_UNLOCK() ((void)0) | |
1f4e496e PB |
3929 | #include "memory_ldst.inc.c" |
3930 | ||
5e2972fd | 3931 | /* virtual memory access for debug (includes writing to ROM) */ |
f17ec444 | 3932 | int cpu_memory_rw_debug(CPUState *cpu, target_ulong addr, |
0c249ff7 | 3933 | uint8_t *buf, target_ulong len, int is_write) |
13eb76e0 | 3934 | { |
a8170e5e | 3935 | hwaddr phys_addr; |
0c249ff7 | 3936 | target_ulong l, page; |
13eb76e0 | 3937 | |
79ca7a1b | 3938 | cpu_synchronize_state(cpu); |
13eb76e0 | 3939 | while (len > 0) { |
5232e4c7 PM |
3940 | int asidx; |
3941 | MemTxAttrs attrs; | |
3942 | ||
13eb76e0 | 3943 | page = addr & TARGET_PAGE_MASK; |
5232e4c7 PM |
3944 | phys_addr = cpu_get_phys_page_attrs_debug(cpu, page, &attrs); |
3945 | asidx = cpu_asidx_from_attrs(cpu, attrs); | |
13eb76e0 FB |
3946 | /* if no physical page mapped, return an error */ |
3947 | if (phys_addr == -1) | |
3948 | return -1; | |
3949 | l = (page + TARGET_PAGE_SIZE) - addr; | |
3950 | if (l > len) | |
3951 | l = len; | |
5e2972fd | 3952 | phys_addr += (addr & ~TARGET_PAGE_MASK); |
2e38847b | 3953 | if (is_write) { |
3c8133f9 | 3954 | address_space_write_rom(cpu->cpu_ases[asidx].as, phys_addr, |
ea7a5330 | 3955 | attrs, buf, l); |
2e38847b | 3956 | } else { |
5232e4c7 | 3957 | address_space_rw(cpu->cpu_ases[asidx].as, phys_addr, |
ea7a5330 | 3958 | attrs, buf, l, 0); |
2e38847b | 3959 | } |
13eb76e0 FB |
3960 | len -= l; |
3961 | buf += l; | |
3962 | addr += l; | |
3963 | } | |
3964 | return 0; | |
3965 | } | |
038629a6 DDAG |
3966 | |
3967 | /* | |
3968 | * Allows code that needs to deal with migration bitmaps etc to still be built | |
3969 | * target independent. | |
3970 | */ | |
20afaed9 | 3971 | size_t qemu_target_page_size(void) |
038629a6 | 3972 | { |
20afaed9 | 3973 | return TARGET_PAGE_SIZE; |
038629a6 DDAG |
3974 | } |
3975 | ||
46d702b1 JQ |
3976 | int qemu_target_page_bits(void) |
3977 | { | |
3978 | return TARGET_PAGE_BITS; | |
3979 | } | |
3980 | ||
3981 | int qemu_target_page_bits_min(void) | |
3982 | { | |
3983 | return TARGET_PAGE_BITS_MIN; | |
3984 | } | |
a68fe89c | 3985 | #endif |
13eb76e0 | 3986 | |
98ed8ecf | 3987 | bool target_words_bigendian(void) |
8e4a424b BS |
3988 | { |
3989 | #if defined(TARGET_WORDS_BIGENDIAN) | |
3990 | return true; | |
3991 | #else | |
3992 | return false; | |
3993 | #endif | |
3994 | } | |
3995 | ||
76f35538 | 3996 | #ifndef CONFIG_USER_ONLY |
a8170e5e | 3997 | bool cpu_physical_memory_is_io(hwaddr phys_addr) |
76f35538 | 3998 | { |
5c8a00ce | 3999 | MemoryRegion*mr; |
149f54b5 | 4000 | hwaddr l = 1; |
41063e1e | 4001 | bool res; |
76f35538 | 4002 | |
41063e1e | 4003 | rcu_read_lock(); |
5c8a00ce | 4004 | mr = address_space_translate(&address_space_memory, |
bc6b1cec PM |
4005 | phys_addr, &phys_addr, &l, false, |
4006 | MEMTXATTRS_UNSPECIFIED); | |
76f35538 | 4007 | |
41063e1e PB |
4008 | res = !(memory_region_is_ram(mr) || memory_region_is_romd(mr)); |
4009 | rcu_read_unlock(); | |
4010 | return res; | |
76f35538 | 4011 | } |
bd2fa51f | 4012 | |
e3807054 | 4013 | int qemu_ram_foreach_block(RAMBlockIterFunc func, void *opaque) |
bd2fa51f MH |
4014 | { |
4015 | RAMBlock *block; | |
e3807054 | 4016 | int ret = 0; |
bd2fa51f | 4017 | |
0dc3f44a | 4018 | rcu_read_lock(); |
99e15582 | 4019 | RAMBLOCK_FOREACH(block) { |
754cb9c0 | 4020 | ret = func(block, opaque); |
e3807054 DDAG |
4021 | if (ret) { |
4022 | break; | |
4023 | } | |
bd2fa51f | 4024 | } |
0dc3f44a | 4025 | rcu_read_unlock(); |
e3807054 | 4026 | return ret; |
bd2fa51f | 4027 | } |
d3a5038c DDAG |
4028 | |
4029 | /* | |
4030 | * Unmap pages of memory from start to start+length such that | |
4031 | * they a) read as 0, b) Trigger whatever fault mechanism | |
4032 | * the OS provides for postcopy. | |
4033 | * The pages must be unmapped by the end of the function. | |
4034 | * Returns: 0 on success, none-0 on failure | |
4035 | * | |
4036 | */ | |
4037 | int ram_block_discard_range(RAMBlock *rb, uint64_t start, size_t length) | |
4038 | { | |
4039 | int ret = -1; | |
4040 | ||
4041 | uint8_t *host_startaddr = rb->host + start; | |
4042 | ||
4043 | if ((uintptr_t)host_startaddr & (rb->page_size - 1)) { | |
4044 | error_report("ram_block_discard_range: Unaligned start address: %p", | |
4045 | host_startaddr); | |
4046 | goto err; | |
4047 | } | |
4048 | ||
4049 | if ((start + length) <= rb->used_length) { | |
db144f70 | 4050 | bool need_madvise, need_fallocate; |
d3a5038c DDAG |
4051 | uint8_t *host_endaddr = host_startaddr + length; |
4052 | if ((uintptr_t)host_endaddr & (rb->page_size - 1)) { | |
4053 | error_report("ram_block_discard_range: Unaligned end address: %p", | |
4054 | host_endaddr); | |
4055 | goto err; | |
4056 | } | |
4057 | ||
4058 | errno = ENOTSUP; /* If we are missing MADVISE etc */ | |
4059 | ||
db144f70 DDAG |
4060 | /* The logic here is messy; |
4061 | * madvise DONTNEED fails for hugepages | |
4062 | * fallocate works on hugepages and shmem | |
4063 | */ | |
4064 | need_madvise = (rb->page_size == qemu_host_page_size); | |
4065 | need_fallocate = rb->fd != -1; | |
4066 | if (need_fallocate) { | |
4067 | /* For a file, this causes the area of the file to be zero'd | |
4068 | * if read, and for hugetlbfs also causes it to be unmapped | |
4069 | * so a userfault will trigger. | |
e2fa71f5 DDAG |
4070 | */ |
4071 | #ifdef CONFIG_FALLOCATE_PUNCH_HOLE | |
4072 | ret = fallocate(rb->fd, FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE, | |
4073 | start, length); | |
db144f70 DDAG |
4074 | if (ret) { |
4075 | ret = -errno; | |
4076 | error_report("ram_block_discard_range: Failed to fallocate " | |
4077 | "%s:%" PRIx64 " +%zx (%d)", | |
4078 | rb->idstr, start, length, ret); | |
4079 | goto err; | |
4080 | } | |
4081 | #else | |
4082 | ret = -ENOSYS; | |
4083 | error_report("ram_block_discard_range: fallocate not available/file" | |
4084 | "%s:%" PRIx64 " +%zx (%d)", | |
4085 | rb->idstr, start, length, ret); | |
4086 | goto err; | |
e2fa71f5 DDAG |
4087 | #endif |
4088 | } | |
db144f70 DDAG |
4089 | if (need_madvise) { |
4090 | /* For normal RAM this causes it to be unmapped, | |
4091 | * for shared memory it causes the local mapping to disappear | |
4092 | * and to fall back on the file contents (which we just | |
4093 | * fallocate'd away). | |
4094 | */ | |
4095 | #if defined(CONFIG_MADVISE) | |
4096 | ret = madvise(host_startaddr, length, MADV_DONTNEED); | |
4097 | if (ret) { | |
4098 | ret = -errno; | |
4099 | error_report("ram_block_discard_range: Failed to discard range " | |
4100 | "%s:%" PRIx64 " +%zx (%d)", | |
4101 | rb->idstr, start, length, ret); | |
4102 | goto err; | |
4103 | } | |
4104 | #else | |
4105 | ret = -ENOSYS; | |
4106 | error_report("ram_block_discard_range: MADVISE not available" | |
d3a5038c DDAG |
4107 | "%s:%" PRIx64 " +%zx (%d)", |
4108 | rb->idstr, start, length, ret); | |
db144f70 DDAG |
4109 | goto err; |
4110 | #endif | |
d3a5038c | 4111 | } |
db144f70 DDAG |
4112 | trace_ram_block_discard_range(rb->idstr, host_startaddr, length, |
4113 | need_madvise, need_fallocate, ret); | |
d3a5038c DDAG |
4114 | } else { |
4115 | error_report("ram_block_discard_range: Overrun block '%s' (%" PRIu64 | |
4116 | "/%zx/" RAM_ADDR_FMT")", | |
4117 | rb->idstr, start, length, rb->used_length); | |
4118 | } | |
4119 | ||
4120 | err: | |
4121 | return ret; | |
4122 | } | |
4123 | ||
a4de8552 JH |
4124 | bool ramblock_is_pmem(RAMBlock *rb) |
4125 | { | |
4126 | return rb->flags & RAM_PMEM; | |
4127 | } | |
4128 | ||
ec3f8c99 | 4129 | #endif |
a0be0c58 YZ |
4130 | |
4131 | void page_size_init(void) | |
4132 | { | |
4133 | /* NOTE: we can always suppose that qemu_host_page_size >= | |
4134 | TARGET_PAGE_SIZE */ | |
a0be0c58 YZ |
4135 | if (qemu_host_page_size == 0) { |
4136 | qemu_host_page_size = qemu_real_host_page_size; | |
4137 | } | |
4138 | if (qemu_host_page_size < TARGET_PAGE_SIZE) { | |
4139 | qemu_host_page_size = TARGET_PAGE_SIZE; | |
4140 | } | |
4141 | qemu_host_page_mask = -(intptr_t)qemu_host_page_size; | |
4142 | } | |
5e8fd947 AK |
4143 | |
4144 | #if !defined(CONFIG_USER_ONLY) | |
4145 | ||
b6b71cb5 | 4146 | static void mtree_print_phys_entries(int start, int end, int skip, int ptr) |
5e8fd947 AK |
4147 | { |
4148 | if (start == end - 1) { | |
b6b71cb5 | 4149 | qemu_printf("\t%3d ", start); |
5e8fd947 | 4150 | } else { |
b6b71cb5 | 4151 | qemu_printf("\t%3d..%-3d ", start, end - 1); |
5e8fd947 | 4152 | } |
b6b71cb5 | 4153 | qemu_printf(" skip=%d ", skip); |
5e8fd947 | 4154 | if (ptr == PHYS_MAP_NODE_NIL) { |
b6b71cb5 | 4155 | qemu_printf(" ptr=NIL"); |
5e8fd947 | 4156 | } else if (!skip) { |
b6b71cb5 | 4157 | qemu_printf(" ptr=#%d", ptr); |
5e8fd947 | 4158 | } else { |
b6b71cb5 | 4159 | qemu_printf(" ptr=[%d]", ptr); |
5e8fd947 | 4160 | } |
b6b71cb5 | 4161 | qemu_printf("\n"); |
5e8fd947 AK |
4162 | } |
4163 | ||
4164 | #define MR_SIZE(size) (int128_nz(size) ? (hwaddr)int128_get64( \ | |
4165 | int128_sub((size), int128_one())) : 0) | |
4166 | ||
b6b71cb5 | 4167 | void mtree_print_dispatch(AddressSpaceDispatch *d, MemoryRegion *root) |
5e8fd947 AK |
4168 | { |
4169 | int i; | |
4170 | ||
b6b71cb5 MA |
4171 | qemu_printf(" Dispatch\n"); |
4172 | qemu_printf(" Physical sections\n"); | |
5e8fd947 AK |
4173 | |
4174 | for (i = 0; i < d->map.sections_nb; ++i) { | |
4175 | MemoryRegionSection *s = d->map.sections + i; | |
4176 | const char *names[] = { " [unassigned]", " [not dirty]", | |
4177 | " [ROM]", " [watch]" }; | |
4178 | ||
b6b71cb5 MA |
4179 | qemu_printf(" #%d @" TARGET_FMT_plx ".." TARGET_FMT_plx |
4180 | " %s%s%s%s%s", | |
5e8fd947 AK |
4181 | i, |
4182 | s->offset_within_address_space, | |
4183 | s->offset_within_address_space + MR_SIZE(s->mr->size), | |
4184 | s->mr->name ? s->mr->name : "(noname)", | |
4185 | i < ARRAY_SIZE(names) ? names[i] : "", | |
4186 | s->mr == root ? " [ROOT]" : "", | |
4187 | s == d->mru_section ? " [MRU]" : "", | |
4188 | s->mr->is_iommu ? " [iommu]" : ""); | |
4189 | ||
4190 | if (s->mr->alias) { | |
b6b71cb5 | 4191 | qemu_printf(" alias=%s", s->mr->alias->name ? |
5e8fd947 AK |
4192 | s->mr->alias->name : "noname"); |
4193 | } | |
b6b71cb5 | 4194 | qemu_printf("\n"); |
5e8fd947 AK |
4195 | } |
4196 | ||
b6b71cb5 | 4197 | qemu_printf(" Nodes (%d bits per level, %d levels) ptr=[%d] skip=%d\n", |
5e8fd947 AK |
4198 | P_L2_BITS, P_L2_LEVELS, d->phys_map.ptr, d->phys_map.skip); |
4199 | for (i = 0; i < d->map.nodes_nb; ++i) { | |
4200 | int j, jprev; | |
4201 | PhysPageEntry prev; | |
4202 | Node *n = d->map.nodes + i; | |
4203 | ||
b6b71cb5 | 4204 | qemu_printf(" [%d]\n", i); |
5e8fd947 AK |
4205 | |
4206 | for (j = 0, jprev = 0, prev = *n[0]; j < ARRAY_SIZE(*n); ++j) { | |
4207 | PhysPageEntry *pe = *n + j; | |
4208 | ||
4209 | if (pe->ptr == prev.ptr && pe->skip == prev.skip) { | |
4210 | continue; | |
4211 | } | |
4212 | ||
b6b71cb5 | 4213 | mtree_print_phys_entries(jprev, j, prev.skip, prev.ptr); |
5e8fd947 AK |
4214 | |
4215 | jprev = j; | |
4216 | prev = *pe; | |
4217 | } | |
4218 | ||
4219 | if (jprev != ARRAY_SIZE(*n)) { | |
b6b71cb5 | 4220 | mtree_print_phys_entries(jprev, j, prev.skip, prev.ptr); |
5e8fd947 AK |
4221 | } |
4222 | } | |
4223 | } | |
4224 | ||
4225 | #endif |