1 /*
2 * QEMU TCG Single Threaded vCPUs implementation
3 *
4 * Copyright (c) 2003-2008 Fabrice Bellard
5 * Copyright (c) 2014 Red Hat Inc.
6 *
7 * Permission is hereby granted, free of charge, to any person obtaining a copy
8 * of this software and associated documentation files (the "Software"), to deal
9 * in the Software without restriction, including without limitation the rights
10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
11 * copies of the Software, and to permit persons to whom the Software is
12 * furnished to do so, subject to the following conditions:
13 *
14 * The above copyright notice and this permission notice shall be included in
15 * all copies or substantial portions of the Software.
16 *
17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
23 * THE SOFTWARE.
24 */
25
26 #include "qemu/osdep.h"
27 #include "qemu/lockable.h"
28 #include "sysemu/tcg.h"
29 #include "sysemu/replay.h"
30 #include "sysemu/cpu-timers.h"
31 #include "qemu/main-loop.h"
32 #include "qemu/notify.h"
33 #include "qemu/guest-random.h"
34 #include "exec/exec-all.h"
35 #include "tcg/startup.h"
36 #include "tcg-accel-ops.h"
37 #include "tcg-accel-ops-rr.h"
38 #include "tcg-accel-ops-icount.h"
39
40 /* Kick all RR vCPUs */
rr_kick_vcpu_thread(CPUState * unused)41 void rr_kick_vcpu_thread(CPUState *unused)
42 {
43 CPUState *cpu;
44
45 CPU_FOREACH(cpu) {
46 cpu_exit(cpu);
47 };
48 }
49
50 /*
51 * TCG vCPU kick timer
52 *
53 * The kick timer is responsible for moving single threaded vCPU
54 * emulation on to the next vCPU. If more than one vCPU is running a
55 * timer event we force a cpu->exit so the next vCPU can get
56 * scheduled.
57 *
58 * The timer is removed if all vCPUs are idle and restarted again once
59 * idleness is complete.
60 */
61
62 static QEMUTimer *rr_kick_vcpu_timer;
63 static CPUState *rr_current_cpu;
64
rr_next_kick_time(void)65 static inline int64_t rr_next_kick_time(void)
66 {
67 return qemu_clock_get_ns(QEMU_CLOCK_VIRTUAL) + TCG_KICK_PERIOD;
68 }
69
70 /* Kick the currently round-robin scheduled vCPU to next */
rr_kick_next_cpu(void)71 static void rr_kick_next_cpu(void)
72 {
73 CPUState *cpu;
74 do {
75 cpu = qatomic_read(&rr_current_cpu);
76 if (cpu) {
77 cpu_exit(cpu);
78 }
79 /* Finish kicking this cpu before reading again. */
80 smp_mb();
81 } while (cpu != qatomic_read(&rr_current_cpu));
82 }
83
rr_kick_thread(void * opaque)84 static void rr_kick_thread(void *opaque)
85 {
86 timer_mod(rr_kick_vcpu_timer, rr_next_kick_time());
87 rr_kick_next_cpu();
88 }
89
rr_start_kick_timer(void)90 static void rr_start_kick_timer(void)
91 {
92 if (!rr_kick_vcpu_timer && CPU_NEXT(first_cpu)) {
93 rr_kick_vcpu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
94 rr_kick_thread, NULL);
95 }
96 if (rr_kick_vcpu_timer && !timer_pending(rr_kick_vcpu_timer)) {
97 timer_mod(rr_kick_vcpu_timer, rr_next_kick_time());
98 }
99 }
100
rr_stop_kick_timer(void)101 static void rr_stop_kick_timer(void)
102 {
103 if (rr_kick_vcpu_timer && timer_pending(rr_kick_vcpu_timer)) {
104 timer_del(rr_kick_vcpu_timer);
105 }
106 }
107
rr_wait_io_event(void)108 static void rr_wait_io_event(void)
109 {
110 CPUState *cpu;
111
112 while (all_cpu_threads_idle()) {
113 rr_stop_kick_timer();
114 qemu_cond_wait_bql(first_cpu->halt_cond);
115 }
116
117 rr_start_kick_timer();
118
119 CPU_FOREACH(cpu) {
120 qemu_wait_io_event_common(cpu);
121 }
122 }
123
124 /*
125 * Destroy any remaining vCPUs which have been unplugged and have
126 * finished running
127 */
rr_deal_with_unplugged_cpus(void)128 static void rr_deal_with_unplugged_cpus(void)
129 {
130 CPUState *cpu;
131
132 CPU_FOREACH(cpu) {
133 if (cpu->unplug && !cpu_can_run(cpu)) {
134 tcg_cpu_destroy(cpu);
135 break;
136 }
137 }
138 }
139
rr_force_rcu(Notifier * notify,void * data)140 static void rr_force_rcu(Notifier *notify, void *data)
141 {
142 rr_kick_next_cpu();
143 }
144
145 /*
146 * Calculate the number of CPUs that we will process in a single iteration of
147 * the main CPU thread loop so that we can fairly distribute the instruction
148 * count across CPUs.
149 *
150 * The CPU count is cached based on the CPU list generation ID to avoid
151 * iterating the list every time.
152 */
rr_cpu_count(void)153 static int rr_cpu_count(void)
154 {
155 static unsigned int last_gen_id = ~0;
156 static int cpu_count;
157 CPUState *cpu;
158
159 QEMU_LOCK_GUARD(&qemu_cpu_list_lock);
160
161 if (cpu_list_generation_id_get() != last_gen_id) {
162 cpu_count = 0;
163 CPU_FOREACH(cpu) {
164 ++cpu_count;
165 }
166 last_gen_id = cpu_list_generation_id_get();
167 }
168
169 return cpu_count;
170 }
171
172 /*
173 * In the single-threaded case each vCPU is simulated in turn. If
174 * there is more than a single vCPU we create a simple timer to kick
175 * the vCPU and ensure we don't get stuck in a tight loop in one vCPU.
176 * This is done explicitly rather than relying on side-effects
177 * elsewhere.
178 */
179
rr_cpu_thread_fn(void * arg)180 static void *rr_cpu_thread_fn(void *arg)
181 {
182 Notifier force_rcu;
183 CPUState *cpu = arg;
184
185 assert(tcg_enabled());
186 rcu_register_thread();
187 force_rcu.notify = rr_force_rcu;
188 rcu_add_force_rcu_notifier(&force_rcu);
189 tcg_register_thread();
190
191 bql_lock();
192 qemu_thread_get_self(cpu->thread);
193
194 cpu->thread_id = qemu_get_thread_id();
195 cpu->neg.can_do_io = true;
196 cpu_thread_signal_created(cpu);
197 qemu_guest_random_seed_thread_part2(cpu->random_seed);
198
199 /* wait for initial kick-off after machine start */
200 while (first_cpu->stopped) {
201 qemu_cond_wait_bql(first_cpu->halt_cond);
202
203 /* process any pending work */
204 CPU_FOREACH(cpu) {
205 current_cpu = cpu;
206 qemu_wait_io_event_common(cpu);
207 }
208 }
209
210 rr_start_kick_timer();
211
212 cpu = first_cpu;
213
214 /* process any pending work */
215 cpu->exit_request = 1;
216
217 while (1) {
218 /* Only used for icount_enabled() */
219 int64_t cpu_budget = 0;
220
221 bql_unlock();
222 replay_mutex_lock();
223 bql_lock();
224
225 if (icount_enabled()) {
226 int cpu_count = rr_cpu_count();
227
228 /* Account partial waits to QEMU_CLOCK_VIRTUAL. */
229 icount_account_warp_timer();
230 /*
231 * Run the timers here. This is much more efficient than
232 * waking up the I/O thread and waiting for completion.
233 */
234 icount_handle_deadline();
235
236 cpu_budget = icount_percpu_budget(cpu_count);
237 }
238
239 replay_mutex_unlock();
240
241 if (!cpu) {
242 cpu = first_cpu;
243 }
244
245 while (cpu && cpu_work_list_empty(cpu) && !cpu->exit_request) {
246 /* Store rr_current_cpu before evaluating cpu_can_run(). */
247 qatomic_set_mb(&rr_current_cpu, cpu);
248
249 current_cpu = cpu;
250
251 qemu_clock_enable(QEMU_CLOCK_VIRTUAL,
252 (cpu->singlestep_enabled & SSTEP_NOTIMER) == 0);
253
254 if (cpu_can_run(cpu)) {
255 int r;
256
257 bql_unlock();
258 if (icount_enabled()) {
259 icount_prepare_for_run(cpu, cpu_budget);
260 }
261 r = tcg_cpu_exec(cpu);
262 if (icount_enabled()) {
263 icount_process_data(cpu);
264 }
265 bql_lock();
266
267 if (r == EXCP_DEBUG) {
268 cpu_handle_guest_debug(cpu);
269 break;
270 } else if (r == EXCP_ATOMIC) {
271 bql_unlock();
272 cpu_exec_step_atomic(cpu);
273 bql_lock();
274 break;
275 }
276 } else if (cpu->stop) {
277 if (cpu->unplug) {
278 cpu = CPU_NEXT(cpu);
279 }
280 break;
281 }
282
283 cpu = CPU_NEXT(cpu);
284 } /* while (cpu && !cpu->exit_request).. */
285
286 /* Does not need a memory barrier because a spurious wakeup is okay. */
287 qatomic_set(&rr_current_cpu, NULL);
288
289 if (cpu && cpu->exit_request) {
290 qatomic_set_mb(&cpu->exit_request, 0);
291 }
292
293 if (icount_enabled() && all_cpu_threads_idle()) {
294 /*
295 * When all cpus are sleeping (e.g in WFI), to avoid a deadlock
296 * in the main_loop, wake it up in order to start the warp timer.
297 */
298 qemu_notify_event();
299 }
300
301 rr_wait_io_event();
302 rr_deal_with_unplugged_cpus();
303 }
304
305 rcu_remove_force_rcu_notifier(&force_rcu);
306 rcu_unregister_thread();
307 return NULL;
308 }
309
rr_start_vcpu_thread(CPUState * cpu)310 void rr_start_vcpu_thread(CPUState *cpu)
311 {
312 char thread_name[VCPU_THREAD_NAME_SIZE];
313 static QemuCond *single_tcg_halt_cond;
314 static QemuThread *single_tcg_cpu_thread;
315
316 g_assert(tcg_enabled());
317 tcg_cpu_init_cflags(cpu, false);
318
319 if (!single_tcg_cpu_thread) {
320 single_tcg_halt_cond = cpu->halt_cond;
321 single_tcg_cpu_thread = cpu->thread;
322
323 /* share a single thread for all cpus with TCG */
324 snprintf(thread_name, VCPU_THREAD_NAME_SIZE, "ALL CPUs/TCG");
325 qemu_thread_create(cpu->thread, thread_name,
326 rr_cpu_thread_fn,
327 cpu, QEMU_THREAD_JOINABLE);
328 } else {
329 /* we share the thread, dump spare data */
330 g_free(cpu->thread);
331 qemu_cond_destroy(cpu->halt_cond);
332 g_free(cpu->halt_cond);
333 cpu->thread = single_tcg_cpu_thread;
334 cpu->halt_cond = single_tcg_halt_cond;
335
336 /* copy the stuff done at start of rr_cpu_thread_fn */
337 cpu->thread_id = first_cpu->thread_id;
338 cpu->neg.can_do_io = 1;
339 cpu->created = true;
340 }
341 }
342