| 1 | /* |
| 2 | * CPU thread main loop - common bits for user and system mode emulation |
| 3 | * |
| 4 | * Copyright (c) 2003-2005 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 2.1 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | * This library is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 14 | * Lesser General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU Lesser General Public |
| 17 | * License along with this library; if not, see <http://www.gnu.org/licenses/>. |
| 18 | */ |
| 19 | |
| 20 | #include "qemu/osdep.h" |
| 21 | #include "qemu/main-loop.h" |
| 22 | #include "exec/cpu-common.h" |
| 23 | #include "hw/core/cpu.h" |
| 24 | #include "qemu/lockable.h" |
| 25 | #include "trace/trace-root.h" |
| 26 | |
| 27 | QemuMutex qemu_cpu_list_lock; |
| 28 | static QemuCond exclusive_cond; |
| 29 | static QemuCond exclusive_resume; |
| 30 | static QemuCond qemu_work_cond; |
| 31 | |
| 32 | /* >= 1 if a thread is inside start_exclusive/end_exclusive. Written |
| 33 | * under qemu_cpu_list_lock, read with atomic operations. |
| 34 | */ |
| 35 | static int pending_cpus; |
| 36 | |
| 37 | void qemu_init_cpu_list(void) |
| 38 | { |
| 39 | /* This is needed because qemu_init_cpu_list is also called by the |
| 40 | * child process in a fork. */ |
| 41 | pending_cpus = 0; |
| 42 | |
| 43 | qemu_mutex_init(&qemu_cpu_list_lock); |
| 44 | qemu_cond_init(&exclusive_cond); |
| 45 | qemu_cond_init(&exclusive_resume); |
| 46 | qemu_cond_init(&qemu_work_cond); |
| 47 | } |
| 48 | |
| 49 | void cpu_list_lock(void) |
| 50 | { |
| 51 | qemu_mutex_lock(&qemu_cpu_list_lock); |
| 52 | } |
| 53 | |
| 54 | void cpu_list_unlock(void) |
| 55 | { |
| 56 | qemu_mutex_unlock(&qemu_cpu_list_lock); |
| 57 | } |
| 58 | |
| 59 | |
| 60 | int cpu_get_free_index(void) |
| 61 | { |
| 62 | CPUState *some_cpu; |
| 63 | int max_cpu_index = 0; |
| 64 | |
| 65 | CPU_FOREACH(some_cpu) { |
| 66 | if (some_cpu->cpu_index >= max_cpu_index) { |
| 67 | max_cpu_index = some_cpu->cpu_index + 1; |
| 68 | } |
| 69 | } |
| 70 | return max_cpu_index; |
| 71 | } |
| 72 | |
| 73 | CPUTailQ cpus_queue = QTAILQ_HEAD_INITIALIZER(cpus_queue); |
| 74 | static unsigned int cpu_list_generation_id; |
| 75 | |
| 76 | unsigned int cpu_list_generation_id_get(void) |
| 77 | { |
| 78 | return cpu_list_generation_id; |
| 79 | } |
| 80 | |
| 81 | void cpu_list_add(CPUState *cpu) |
| 82 | { |
| 83 | static bool cpu_index_auto_assigned; |
| 84 | |
| 85 | QEMU_LOCK_GUARD(&qemu_cpu_list_lock); |
| 86 | if (cpu->cpu_index == UNASSIGNED_CPU_INDEX) { |
| 87 | cpu_index_auto_assigned = true; |
| 88 | cpu->cpu_index = cpu_get_free_index(); |
| 89 | assert(cpu->cpu_index != UNASSIGNED_CPU_INDEX); |
| 90 | } else { |
| 91 | assert(!cpu_index_auto_assigned); |
| 92 | } |
| 93 | QTAILQ_INSERT_TAIL_RCU(&cpus_queue, cpu, node); |
| 94 | cpu_list_generation_id++; |
| 95 | } |
| 96 | |
| 97 | void cpu_list_remove(CPUState *cpu) |
| 98 | { |
| 99 | QEMU_LOCK_GUARD(&qemu_cpu_list_lock); |
| 100 | if (!QTAILQ_IN_USE(cpu, node)) { |
| 101 | /* there is nothing to undo since cpu_exec_init() hasn't been called */ |
| 102 | return; |
| 103 | } |
| 104 | |
| 105 | QTAILQ_REMOVE_RCU(&cpus_queue, cpu, node); |
| 106 | cpu->cpu_index = UNASSIGNED_CPU_INDEX; |
| 107 | cpu_list_generation_id++; |
| 108 | } |
| 109 | |
| 110 | CPUState *qemu_get_cpu(int index) |
| 111 | { |
| 112 | CPUState *cpu; |
| 113 | |
| 114 | CPU_FOREACH(cpu) { |
| 115 | if (cpu->cpu_index == index) { |
| 116 | return cpu; |
| 117 | } |
| 118 | } |
| 119 | |
| 120 | return NULL; |
| 121 | } |
| 122 | |
| 123 | /* current CPU in the current thread. It is only valid inside cpu_exec() */ |
| 124 | __thread CPUState *current_cpu; |
| 125 | |
| 126 | struct qemu_work_item { |
| 127 | QSIMPLEQ_ENTRY(qemu_work_item) node; |
| 128 | run_on_cpu_func func; |
| 129 | run_on_cpu_data data; |
| 130 | bool free, exclusive, done; |
| 131 | }; |
| 132 | |
| 133 | static void queue_work_on_cpu(CPUState *cpu, struct qemu_work_item *wi) |
| 134 | { |
| 135 | qemu_mutex_lock(&cpu->work_mutex); |
| 136 | QSIMPLEQ_INSERT_TAIL(&cpu->work_list, wi, node); |
| 137 | wi->done = false; |
| 138 | qemu_mutex_unlock(&cpu->work_mutex); |
| 139 | |
| 140 | /* exit the inner loop and reach qemu_process_cpu_events_common(). */ |
| 141 | cpu_exit(cpu); |
| 142 | } |
| 143 | |
| 144 | void do_run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data, |
| 145 | QemuMutex *mutex) |
| 146 | { |
| 147 | struct qemu_work_item wi; |
| 148 | |
| 149 | if (qemu_cpu_is_self(cpu)) { |
| 150 | func(cpu, data); |
| 151 | return; |
| 152 | } |
| 153 | |
| 154 | wi.func = func; |
| 155 | wi.data = data; |
| 156 | wi.done = false; |
| 157 | wi.free = false; |
| 158 | wi.exclusive = false; |
| 159 | |
| 160 | queue_work_on_cpu(cpu, &wi); |
| 161 | while (!qatomic_load_acquire(&wi.done)) { |
| 162 | CPUState *self_cpu = current_cpu; |
| 163 | |
| 164 | qemu_cond_wait(&qemu_work_cond, mutex); |
| 165 | current_cpu = self_cpu; |
| 166 | } |
| 167 | } |
| 168 | |
| 169 | void async_run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data) |
| 170 | { |
| 171 | struct qemu_work_item *wi; |
| 172 | |
| 173 | wi = g_new0(struct qemu_work_item, 1); |
| 174 | wi->func = func; |
| 175 | wi->data = data; |
| 176 | wi->free = true; |
| 177 | |
| 178 | queue_work_on_cpu(cpu, wi); |
| 179 | } |
| 180 | |
| 181 | /* Wait for pending exclusive operations to complete. The CPU list lock |
| 182 | must be held. */ |
| 183 | static inline void exclusive_idle(void) |
| 184 | { |
| 185 | while (pending_cpus) { |
| 186 | qemu_cond_wait(&exclusive_resume, &qemu_cpu_list_lock); |
| 187 | } |
| 188 | } |
| 189 | |
| 190 | /* Start an exclusive operation. |
| 191 | Must only be called from outside cpu_exec. */ |
| 192 | void start_exclusive(void) |
| 193 | { |
| 194 | CPUState *other_cpu; |
| 195 | int running_cpus; |
| 196 | |
| 197 | /* Ensure we are not running, or start_exclusive will be blocked. */ |
| 198 | g_assert(!current_cpu->running); |
| 199 | |
| 200 | if (current_cpu->exclusive_context_count) { |
| 201 | current_cpu->exclusive_context_count++; |
| 202 | return; |
| 203 | } |
| 204 | |
| 205 | qemu_mutex_lock(&qemu_cpu_list_lock); |
| 206 | exclusive_idle(); |
| 207 | |
| 208 | /* Make all other cpus stop executing. */ |
| 209 | qatomic_set(&pending_cpus, 1); |
| 210 | |
| 211 | /* Write pending_cpus before reading other_cpu->running. */ |
| 212 | smp_mb(); |
| 213 | running_cpus = 0; |
| 214 | CPU_FOREACH(other_cpu) { |
| 215 | if (qatomic_read(&other_cpu->running)) { |
| 216 | other_cpu->has_waiter = true; |
| 217 | running_cpus++; |
| 218 | qemu_cpu_kick(other_cpu); |
| 219 | } |
| 220 | } |
| 221 | |
| 222 | qatomic_set(&pending_cpus, running_cpus + 1); |
| 223 | while (pending_cpus > 1) { |
| 224 | qemu_cond_wait(&exclusive_cond, &qemu_cpu_list_lock); |
| 225 | } |
| 226 | |
| 227 | /* Can release mutex, no one will enter another exclusive |
| 228 | * section until end_exclusive resets pending_cpus to 0. |
| 229 | */ |
| 230 | qemu_mutex_unlock(&qemu_cpu_list_lock); |
| 231 | |
| 232 | current_cpu->exclusive_context_count = 1; |
| 233 | } |
| 234 | |
| 235 | /* Finish an exclusive operation. */ |
| 236 | void end_exclusive(void) |
| 237 | { |
| 238 | current_cpu->exclusive_context_count--; |
| 239 | if (current_cpu->exclusive_context_count) { |
| 240 | return; |
| 241 | } |
| 242 | |
| 243 | qemu_mutex_lock(&qemu_cpu_list_lock); |
| 244 | qatomic_set(&pending_cpus, 0); |
| 245 | qemu_cond_broadcast(&exclusive_resume); |
| 246 | qemu_mutex_unlock(&qemu_cpu_list_lock); |
| 247 | } |
| 248 | |
| 249 | /* Wait for exclusive ops to finish, and begin cpu execution. */ |
| 250 | void cpu_exec_start(CPUState *cpu) |
| 251 | { |
| 252 | trace_cpu_exec_start(cpu->cpu_index); |
| 253 | |
| 254 | qatomic_set(&cpu->running, true); |
| 255 | |
| 256 | /* Write cpu->running before reading pending_cpus. */ |
| 257 | smp_mb(); |
| 258 | |
| 259 | /* 1. start_exclusive saw cpu->running == true and pending_cpus >= 1. |
| 260 | * After taking the lock we'll see cpu->has_waiter == true and run---not |
| 261 | * for long because start_exclusive kicked us. cpu_exec_end will |
| 262 | * decrement pending_cpus and signal the waiter. |
| 263 | * |
| 264 | * 2. start_exclusive saw cpu->running == false but pending_cpus >= 1. |
| 265 | * This includes the case when an exclusive item is running now. |
| 266 | * Then we'll see cpu->has_waiter == false and wait for the item to |
| 267 | * complete. |
| 268 | * |
| 269 | * 3. pending_cpus == 0. Then start_exclusive is definitely going to |
| 270 | * see cpu->running == true, and it will kick the CPU. |
| 271 | */ |
| 272 | if (unlikely(qatomic_read(&pending_cpus))) { |
| 273 | QEMU_LOCK_GUARD(&qemu_cpu_list_lock); |
| 274 | if (!cpu->has_waiter) { |
| 275 | /* Not counted in pending_cpus, let the exclusive item |
| 276 | * run. Since we have the lock, just set cpu->running to true |
| 277 | * while holding it; no need to check pending_cpus again. |
| 278 | */ |
| 279 | qatomic_set(&cpu->running, false); |
| 280 | exclusive_idle(); |
| 281 | /* Now pending_cpus is zero. */ |
| 282 | qatomic_set(&cpu->running, true); |
| 283 | } else { |
| 284 | /* Counted in pending_cpus, go ahead and release the |
| 285 | * waiter at cpu_exec_end. |
| 286 | */ |
| 287 | } |
| 288 | } |
| 289 | } |
| 290 | |
| 291 | /* Mark cpu as not executing, and release pending exclusive ops. */ |
| 292 | void cpu_exec_end(CPUState *cpu) |
| 293 | { |
| 294 | qatomic_set(&cpu->running, false); |
| 295 | |
| 296 | /* Write cpu->running before reading pending_cpus. */ |
| 297 | smp_mb(); |
| 298 | |
| 299 | /* 1. start_exclusive saw cpu->running == true. Then it will increment |
| 300 | * pending_cpus and wait for exclusive_cond. After taking the lock |
| 301 | * we'll see cpu->has_waiter == true. |
| 302 | * |
| 303 | * 2. start_exclusive saw cpu->running == false but here pending_cpus >= 1. |
| 304 | * This includes the case when an exclusive item started after setting |
| 305 | * cpu->running to false and before we read pending_cpus. Then we'll see |
| 306 | * cpu->has_waiter == false and not touch pending_cpus. The next call to |
| 307 | * cpu_exec_start will run exclusive_idle if still necessary, thus waiting |
| 308 | * for the item to complete. |
| 309 | * |
| 310 | * 3. pending_cpus == 0. Then start_exclusive is definitely going to |
| 311 | * see cpu->running == false, and it can ignore this CPU until the |
| 312 | * next cpu_exec_start. |
| 313 | */ |
| 314 | if (unlikely(qatomic_read(&pending_cpus))) { |
| 315 | QEMU_LOCK_GUARD(&qemu_cpu_list_lock); |
| 316 | if (cpu->has_waiter) { |
| 317 | cpu->has_waiter = false; |
| 318 | qatomic_set(&pending_cpus, pending_cpus - 1); |
| 319 | if (pending_cpus == 1) { |
| 320 | qemu_cond_signal(&exclusive_cond); |
| 321 | } |
| 322 | } |
| 323 | } |
| 324 | trace_cpu_exec_end(cpu->cpu_index); |
| 325 | } |
| 326 | |
| 327 | void async_safe_run_on_cpu(CPUState *cpu, run_on_cpu_func func, |
| 328 | run_on_cpu_data data) |
| 329 | { |
| 330 | struct qemu_work_item *wi; |
| 331 | |
| 332 | wi = g_new0(struct qemu_work_item, 1); |
| 333 | wi->func = func; |
| 334 | wi->data = data; |
| 335 | wi->free = true; |
| 336 | wi->exclusive = true; |
| 337 | |
| 338 | queue_work_on_cpu(cpu, wi); |
| 339 | } |
| 340 | |
| 341 | void free_queued_cpu_work(CPUState *cpu) |
| 342 | { |
| 343 | while (!QSIMPLEQ_EMPTY(&cpu->work_list)) { |
| 344 | struct qemu_work_item *wi = QSIMPLEQ_FIRST(&cpu->work_list); |
| 345 | QSIMPLEQ_REMOVE_HEAD(&cpu->work_list, node); |
| 346 | if (wi->free) { |
| 347 | g_free(wi); |
| 348 | } |
| 349 | } |
| 350 | } |
| 351 | |
| 352 | void process_queued_cpu_work(CPUState *cpu) |
| 353 | { |
| 354 | struct qemu_work_item *wi; |
| 355 | |
| 356 | qemu_mutex_lock(&cpu->work_mutex); |
| 357 | if (QSIMPLEQ_EMPTY(&cpu->work_list)) { |
| 358 | qemu_mutex_unlock(&cpu->work_mutex); |
| 359 | return; |
| 360 | } |
| 361 | while (!QSIMPLEQ_EMPTY(&cpu->work_list)) { |
| 362 | wi = QSIMPLEQ_FIRST(&cpu->work_list); |
| 363 | QSIMPLEQ_REMOVE_HEAD(&cpu->work_list, node); |
| 364 | qemu_mutex_unlock(&cpu->work_mutex); |
| 365 | if (wi->exclusive) { |
| 366 | /* Running work items outside the BQL avoids the following deadlock: |
| 367 | * 1) start_exclusive() is called with the BQL taken while another |
| 368 | * CPU is running; 2) cpu_exec in the other CPU tries to takes the |
| 369 | * BQL, so it goes to sleep; start_exclusive() is sleeping too, so |
| 370 | * neither CPU can proceed. |
| 371 | */ |
| 372 | bql_unlock(); |
| 373 | start_exclusive(); |
| 374 | wi->func(cpu, wi->data); |
| 375 | end_exclusive(); |
| 376 | bql_lock(); |
| 377 | } else { |
| 378 | wi->func(cpu, wi->data); |
| 379 | } |
| 380 | qemu_mutex_lock(&cpu->work_mutex); |
| 381 | if (wi->free) { |
| 382 | g_free(wi); |
| 383 | } else { |
| 384 | qatomic_store_release(&wi->done, true); |
| 385 | } |
| 386 | } |
| 387 | qemu_mutex_unlock(&cpu->work_mutex); |
| 388 | qemu_cond_broadcast(&qemu_work_cond); |
| 389 | } |
| 390 | |
| 391 | /* Return true if PC matches an installed breakpoint. */ |
| 392 | bool cpu_breakpoint_test(CPUState *cpu, vaddr pc, int mask) |
| 393 | { |
| 394 | CPUBreakpoint *bp; |
| 395 | |
| 396 | if (unlikely(!QTAILQ_EMPTY(&cpu->breakpoints))) { |
| 397 | QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) { |
| 398 | if (bp->pc == pc && (bp->flags & mask)) { |
| 399 | return true; |
| 400 | } |
| 401 | } |
| 402 | } |
| 403 | return false; |
| 404 | } |
| 405 | |
| 406 | /* Add a breakpoint. */ |
| 407 | int cpu_breakpoint_insert(CPUState *cpu, vaddr pc, int flags, |
| 408 | CPUBreakpoint **breakpoint) |
| 409 | { |
| 410 | CPUBreakpoint *bp; |
| 411 | |
| 412 | if (cpu->cc->gdb_adjust_breakpoint) { |
| 413 | pc = cpu->cc->gdb_adjust_breakpoint(cpu, pc); |
| 414 | } |
| 415 | |
| 416 | bp = g_malloc(sizeof(*bp)); |
| 417 | |
| 418 | bp->pc = pc; |
| 419 | bp->flags = flags; |
| 420 | |
| 421 | /* keep all GDB-injected breakpoints in front */ |
| 422 | if (flags & BP_GDB) { |
| 423 | QTAILQ_INSERT_HEAD(&cpu->breakpoints, bp, entry); |
| 424 | } else { |
| 425 | QTAILQ_INSERT_TAIL(&cpu->breakpoints, bp, entry); |
| 426 | } |
| 427 | |
| 428 | if (breakpoint) { |
| 429 | *breakpoint = bp; |
| 430 | } |
| 431 | |
| 432 | trace_breakpoint_insert(cpu->cpu_index, pc, flags); |
| 433 | return 0; |
| 434 | } |
| 435 | |
| 436 | /* Remove a specific breakpoint. */ |
| 437 | int cpu_breakpoint_remove(CPUState *cpu, vaddr pc, int flags) |
| 438 | { |
| 439 | CPUBreakpoint *bp; |
| 440 | |
| 441 | if (cpu->cc->gdb_adjust_breakpoint) { |
| 442 | pc = cpu->cc->gdb_adjust_breakpoint(cpu, pc); |
| 443 | } |
| 444 | |
| 445 | QTAILQ_FOREACH(bp, &cpu->breakpoints, entry) { |
| 446 | if (bp->pc == pc && bp->flags == flags) { |
| 447 | cpu_breakpoint_remove_by_ref(cpu, bp); |
| 448 | return 0; |
| 449 | } |
| 450 | } |
| 451 | return -ENOENT; |
| 452 | } |
| 453 | |
| 454 | /* Remove a specific breakpoint by reference. */ |
| 455 | void cpu_breakpoint_remove_by_ref(CPUState *cpu, CPUBreakpoint *bp) |
| 456 | { |
| 457 | QTAILQ_REMOVE(&cpu->breakpoints, bp, entry); |
| 458 | |
| 459 | trace_breakpoint_remove(cpu->cpu_index, bp->pc, bp->flags); |
| 460 | g_free(bp); |
| 461 | } |
| 462 | |
| 463 | /* Remove all matching breakpoints. */ |
| 464 | void cpu_breakpoint_remove_all(CPUState *cpu, int mask) |
| 465 | { |
| 466 | CPUBreakpoint *bp, *next; |
| 467 | |
| 468 | QTAILQ_FOREACH_SAFE(bp, &cpu->breakpoints, entry, next) { |
| 469 | if (bp->flags & mask) { |
| 470 | cpu_breakpoint_remove_by_ref(cpu, bp); |
| 471 | } |
| 472 | } |
| 473 | } |