]> git.proxmox.com Git - mirror_qemu.git/blobdiff - cpus.c
minikconf: do not include variables from MINIKCONF_ARGS in config-all-devices.mak
[mirror_qemu.git] / cpus.c
diff --git a/cpus.c b/cpus.c
index 719788320f94c4df910400717df647db59865036..1af51b73dd251f81c4e44be177373187c1c99ef0 100644 (file)
--- a/cpus.c
+++ b/cpus.c
@@ -23,6 +23,7 @@
  */
 
 #include "qemu/osdep.h"
+#include "qemu-common.h"
 #include "qemu/config-file.h"
 #include "cpu.h"
 #include "monitor/monitor.h"
@@ -31,7 +32,9 @@
 #include "qapi/qapi-events-run-state.h"
 #include "qapi/qmp/qerror.h"
 #include "qemu/error-report.h"
+#include "qemu/qemu-print.h"
 #include "sysemu/sysemu.h"
+#include "sysemu/tcg.h"
 #include "sysemu/block-backend.h"
 #include "exec/gdbstub.h"
 #include "sysemu/dma.h"
@@ -49,6 +52,7 @@
 #include "qemu/option.h"
 #include "qemu/bitmap.h"
 #include "qemu/seqlock.h"
+#include "qemu/guest-random.h"
 #include "tcg.h"
 #include "hw/nmi.h"
 #include "sysemu/replay.h"
@@ -211,12 +215,12 @@ void qemu_tcg_configure(QemuOpts *opts, Error **errp)
                 error_setg(errp, "No MTTCG when icount is enabled");
             } else {
 #ifndef TARGET_SUPPORTS_MTTCG
-                error_report("Guest not yet converted to MTTCG - "
-                             "you may get unexpected results");
+                warn_report("Guest not yet converted to MTTCG - "
+                            "you may get unexpected results");
 #endif
                 if (!check_tcg_memory_orders_compatible()) {
-                    error_report("Guest expects a stronger memory ordering "
-                                 "than the host provides");
+                    warn_report("Guest expects a stronger memory ordering "
+                                "than the host provides");
                     error_printf("This may cause strange/hard to debug errors\n");
                 }
                 mttcg_enabled = true;
@@ -237,7 +241,8 @@ void qemu_tcg_configure(QemuOpts *opts, Error **errp)
  */
 static int64_t cpu_get_icount_executed(CPUState *cpu)
 {
-    return cpu->icount_budget - (cpu->icount_decr.u16.low + cpu->icount_extra);
+    return (cpu->icount_budget -
+            (cpu_neg(cpu)->icount_decr.u16.low + cpu->icount_extra));
 }
 
 /*
@@ -245,21 +250,27 @@ static int64_t cpu_get_icount_executed(CPUState *cpu)
  * account executed instructions. This is done by the TCG vCPU
  * thread so the main-loop can see time has moved forward.
  */
-void cpu_update_icount(CPUState *cpu)
+static void cpu_update_icount_locked(CPUState *cpu)
 {
     int64_t executed = cpu_get_icount_executed(cpu);
     cpu->icount_budget -= executed;
 
-#ifndef CONFIG_ATOMIC64
+    atomic_set_i64(&timers_state.qemu_icount,
+                   timers_state.qemu_icount + executed);
+}
+
+/*
+ * Update the global shared timer_state.qemu_icount to take into
+ * account executed instructions. This is done by the TCG vCPU
+ * thread so the main-loop can see time has moved forward.
+ */
+void cpu_update_icount(CPUState *cpu)
+{
     seqlock_write_lock(&timers_state.vm_clock_seqlock,
                        &timers_state.vm_clock_lock);
-#endif
-    atomic_set__nocheck(&timers_state.qemu_icount,
-                        timers_state.qemu_icount + executed);
-#ifndef CONFIG_ATOMIC64
+    cpu_update_icount_locked(cpu);
     seqlock_write_unlock(&timers_state.vm_clock_seqlock,
                          &timers_state.vm_clock_lock);
-#endif
 }
 
 static int64_t cpu_get_icount_raw_locked(void)
@@ -272,16 +283,17 @@ static int64_t cpu_get_icount_raw_locked(void)
             exit(1);
         }
         /* Take into account what has run */
-        cpu_update_icount(cpu);
+        cpu_update_icount_locked(cpu);
     }
-    /* The read is protected by the seqlock, so __nocheck is okay.  */
-    return atomic_read__nocheck(&timers_state.qemu_icount);
+    /* The read is protected by the seqlock, but needs atomic64 to avoid UB */
+    return atomic_read_i64(&timers_state.qemu_icount);
 }
 
 static int64_t cpu_get_icount_locked(void)
 {
     int64_t icount = cpu_get_icount_raw_locked();
-    return atomic_read__nocheck(&timers_state.qemu_icount_bias) + cpu_icount_to_ns(icount);
+    return atomic_read_i64(&timers_state.qemu_icount_bias) +
+        cpu_icount_to_ns(icount);
 }
 
 int64_t cpu_get_icount_raw(void)
@@ -454,9 +466,9 @@ static void icount_adjust(void)
                    timers_state.icount_time_shift + 1);
     }
     last_delta = delta;
-    atomic_set__nocheck(&timers_state.qemu_icount_bias,
-                        cur_icount - (timers_state.qemu_icount
-                                      << timers_state.icount_time_shift));
+    atomic_set_i64(&timers_state.qemu_icount_bias,
+                   cur_icount - (timers_state.qemu_icount
+                                 << timers_state.icount_time_shift));
     seqlock_write_unlock(&timers_state.vm_clock_seqlock,
                          &timers_state.vm_clock_lock);
 }
@@ -502,8 +514,8 @@ static void icount_warp_rt(void)
     seqlock_write_lock(&timers_state.vm_clock_seqlock,
                        &timers_state.vm_clock_lock);
     if (runstate_is_running()) {
-        int64_t clock = REPLAY_CLOCK(REPLAY_CLOCK_VIRTUAL_RT,
-                                     cpu_get_clock_locked());
+        int64_t clock = REPLAY_CLOCK_LOCKED(REPLAY_CLOCK_VIRTUAL_RT,
+                                            cpu_get_clock_locked());
         int64_t warp_delta;
 
         warp_delta = clock - timers_state.vm_clock_warp_start;
@@ -516,8 +528,8 @@ static void icount_warp_rt(void)
             int64_t delta = clock - cur_icount;
             warp_delta = MIN(warp_delta, delta);
         }
-        atomic_set__nocheck(&timers_state.qemu_icount_bias,
-                            timers_state.qemu_icount_bias + warp_delta);
+        atomic_set_i64(&timers_state.qemu_icount_bias,
+                       timers_state.qemu_icount_bias + warp_delta);
     }
     timers_state.vm_clock_warp_start = -1;
     seqlock_write_unlock(&timers_state.vm_clock_seqlock,
@@ -548,8 +560,8 @@ void qtest_clock_warp(int64_t dest)
 
         seqlock_write_lock(&timers_state.vm_clock_seqlock,
                            &timers_state.vm_clock_lock);
-        atomic_set__nocheck(&timers_state.qemu_icount_bias,
-                            timers_state.qemu_icount_bias + warp);
+        atomic_set_i64(&timers_state.qemu_icount_bias,
+                       timers_state.qemu_icount_bias + warp);
         seqlock_write_unlock(&timers_state.vm_clock_seqlock,
                              &timers_state.vm_clock_lock);
 
@@ -576,18 +588,29 @@ void qemu_start_warp_timer(void)
         return;
     }
 
-    /* warp clock deterministically in record/replay mode */
-    if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_START)) {
-        return;
-    }
+    if (replay_mode != REPLAY_MODE_PLAY) {
+        if (!all_cpu_threads_idle()) {
+            return;
+        }
 
-    if (!all_cpu_threads_idle()) {
-        return;
-    }
+        if (qtest_enabled()) {
+            /* When testing, qtest commands advance icount.  */
+            return;
+        }
 
-    if (qtest_enabled()) {
-        /* When testing, qtest commands advance icount.  */
-        return;
+        replay_checkpoint(CHECKPOINT_CLOCK_WARP_START);
+    } else {
+        /* warp clock deterministically in record/replay mode */
+        if (!replay_checkpoint(CHECKPOINT_CLOCK_WARP_START)) {
+            /* vCPU is sleeping and warp can't be started.
+               It is probably a race condition: notification sent
+               to vCPU was processed in advance and vCPU went to sleep.
+               Therefore we have to wake it up for doing someting. */
+            if (replay_has_checkpoint()) {
+                qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
+            }
+            return;
+        }
     }
 
     /* We want to use the earliest deadline from ALL vm_clocks */
@@ -620,8 +643,8 @@ void qemu_start_warp_timer(void)
              */
             seqlock_write_lock(&timers_state.vm_clock_seqlock,
                                &timers_state.vm_clock_lock);
-            atomic_set__nocheck(&timers_state.qemu_icount_bias,
-                                timers_state.qemu_icount_bias + deadline);
+            atomic_set_i64(&timers_state.qemu_icount_bias,
+                           timers_state.qemu_icount_bias + deadline);
             seqlock_write_unlock(&timers_state.vm_clock_seqlock,
                                  &timers_state.vm_clock_lock);
             qemu_clock_notify(QEMU_CLOCK_VIRTUAL);
@@ -823,6 +846,7 @@ int cpu_throttle_get_percentage(void)
 void cpu_ticks_init(void)
 {
     seqlock_init(&timers_state.vm_clock_seqlock);
+    qemu_spin_init(&timers_state.vm_clock_lock);
     vmstate_register(NULL, 0, &vmstate_timers, &timers_state);
     throttle_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL_RT,
                                            cpu_throttle_timer_tick, NULL);
@@ -964,6 +988,8 @@ static void start_tcg_kick_timer(void)
     if (!tcg_kick_vcpu_timer && CPU_NEXT(first_cpu)) {
         tcg_kick_vcpu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL,
                                            kick_tcg_thread, NULL);
+    }
+    if (tcg_kick_vcpu_timer && !timer_pending(tcg_kick_vcpu_timer)) {
         timer_mod(tcg_kick_vcpu_timer, qemu_tcg_next_kick());
     }
 }
@@ -971,9 +997,8 @@ static void start_tcg_kick_timer(void)
 static void stop_tcg_kick_timer(void)
 {
     assert(!mttcg_enabled);
-    if (tcg_kick_vcpu_timer) {
+    if (tcg_kick_vcpu_timer && timer_pending(tcg_kick_vcpu_timer)) {
         timer_del(tcg_kick_vcpu_timer);
-        tcg_kick_vcpu_timer = NULL;
     }
 }
 
@@ -989,7 +1014,7 @@ void hw_error(const char *fmt, ...)
     fprintf(stderr, "\n");
     CPU_FOREACH(cpu) {
         fprintf(stderr, "CPU #%d:\n", cpu->cpu_index);
-        cpu_dump_state(cpu, stderr, fprintf, CPU_DUMP_FPU);
+        cpu_dump_state(cpu, stderr, CPU_DUMP_FPU);
     }
     va_end(ap);
     abort();
@@ -1200,16 +1225,20 @@ static void qemu_wait_io_event_common(CPUState *cpu)
     process_queued_cpu_work(cpu);
 }
 
-static void qemu_tcg_rr_wait_io_event(CPUState *cpu)
+static void qemu_tcg_rr_wait_io_event(void)
 {
+    CPUState *cpu;
+
     while (all_cpu_threads_idle()) {
         stop_tcg_kick_timer();
-        qemu_cond_wait(cpu->halt_cond, &qemu_global_mutex);
+        qemu_cond_wait(first_cpu->halt_cond, &qemu_global_mutex);
     }
 
     start_tcg_kick_timer();
 
-    qemu_wait_io_event_common(cpu);
+    CPU_FOREACH(cpu) {
+        qemu_wait_io_event_common(cpu);
+    }
 }
 
 static void qemu_wait_io_event(CPUState *cpu)
@@ -1251,6 +1280,7 @@ static void *qemu_kvm_cpu_thread_fn(void *arg)
     /* signal CPU creation */
     cpu->created = true;
     qemu_cond_signal(&qemu_cpu_cond);
+    qemu_guest_random_seed_thread_part2(cpu->random_seed);
 
     do {
         if (cpu_can_run(cpu)) {
@@ -1294,6 +1324,7 @@ static void *qemu_dummy_cpu_thread_fn(void *arg)
     /* signal CPU creation */
     cpu->created = true;
     qemu_cond_signal(&qemu_cpu_cond);
+    qemu_guest_random_seed_thread_part2(cpu->random_seed);
 
     do {
         qemu_mutex_unlock_iothread();
@@ -1309,6 +1340,7 @@ static void *qemu_dummy_cpu_thread_fn(void *arg)
         qemu_wait_io_event(cpu);
     } while (!cpu->unplug);
 
+    qemu_mutex_unlock_iothread();
     rcu_unregister_thread();
     return NULL;
 #endif
@@ -1360,12 +1392,12 @@ static void prepare_icount_for_run(CPUState *cpu)
          * each vCPU execution. However u16.high can be raised
          * asynchronously by cpu_exit/cpu_interrupt/tcg_handle_interrupt
          */
-        g_assert(cpu->icount_decr.u16.low == 0);
+        g_assert(cpu_neg(cpu)->icount_decr.u16.low == 0);
         g_assert(cpu->icount_extra == 0);
 
         cpu->icount_budget = tcg_get_icount_limit();
         insns_left = MIN(0xffff, cpu->icount_budget);
-        cpu->icount_decr.u16.low = insns_left;
+        cpu_neg(cpu)->icount_decr.u16.low = insns_left;
         cpu->icount_extra = cpu->icount_budget - insns_left;
 
         replay_mutex_lock();
@@ -1379,7 +1411,7 @@ static void process_icount_data(CPUState *cpu)
         cpu_update_icount(cpu);
 
         /* Reset the counters */
-        cpu->icount_decr.u16.low = 0;
+        cpu_neg(cpu)->icount_decr.u16.low = 0;
         cpu->icount_extra = 0;
         cpu->icount_budget = 0;
 
@@ -1405,7 +1437,8 @@ static int tcg_cpu_exec(CPUState *cpu)
     ret = cpu_exec(cpu);
     cpu_exec_end(cpu);
 #ifdef CONFIG_PROFILER
-    tcg_time += profile_getclock() - ti;
+    atomic_set(&tcg_ctx->prof.cpu_exec_time,
+               tcg_ctx->prof.cpu_exec_time + profile_getclock() - ti);
 #endif
     return ret;
 }
@@ -1451,6 +1484,7 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg)
     cpu->created = true;
     cpu->can_do_io = 1;
     qemu_cond_signal(&qemu_cpu_cond);
+    qemu_guest_random_seed_thread_part2(cpu->random_seed);
 
     /* wait for initial kick-off after machine start */
     while (first_cpu->stopped) {
@@ -1533,7 +1567,15 @@ static void *qemu_tcg_rr_cpu_thread_fn(void *arg)
             atomic_mb_set(&cpu->exit_request, 0);
         }
 
-        qemu_tcg_rr_wait_io_event(cpu ? cpu : first_cpu);
+        if (use_icount && all_cpu_threads_idle()) {
+            /*
+             * When all cpus are sleeping (e.g in WFI), to avoid a deadlock
+             * in the main_loop, wake it up in order to start the warp timer.
+             */
+            qemu_notify_event();
+        }
+
+        qemu_tcg_rr_wait_io_event();
         deal_with_unplugged_cpus();
     }
 
@@ -1552,11 +1594,11 @@ static void *qemu_hax_cpu_thread_fn(void *arg)
 
     cpu->thread_id = qemu_get_thread_id();
     cpu->created = true;
-    cpu->halted = 0;
     current_cpu = cpu;
 
     hax_init_vcpu(cpu);
     qemu_cond_signal(&qemu_cpu_cond);
+    qemu_guest_random_seed_thread_part2(cpu->random_seed);
 
     do {
         if (cpu_can_run(cpu)) {
@@ -1596,6 +1638,7 @@ static void *qemu_hvf_cpu_thread_fn(void *arg)
     /* signal CPU creation */
     cpu->created = true;
     qemu_cond_signal(&qemu_cpu_cond);
+    qemu_guest_random_seed_thread_part2(cpu->random_seed);
 
     do {
         if (cpu_can_run(cpu)) {
@@ -1636,6 +1679,7 @@ static void *qemu_whpx_cpu_thread_fn(void *arg)
     /* signal CPU creation */
     cpu->created = true;
     qemu_cond_signal(&qemu_cpu_cond);
+    qemu_guest_random_seed_thread_part2(cpu->random_seed);
 
     do {
         if (cpu_can_run(cpu)) {
@@ -1689,6 +1733,7 @@ static void *qemu_tcg_cpu_thread_fn(void *arg)
     cpu->can_do_io = 1;
     current_cpu = cpu;
     qemu_cond_signal(&qemu_cpu_cond);
+    qemu_guest_random_seed_thread_part2(cpu->random_seed);
 
     /* process any pending work */
     cpu->exit_request = 1;
@@ -1745,7 +1790,7 @@ static void qemu_cpu_kick_thread(CPUState *cpu)
     }
     cpu->thread_kicked = true;
     err = pthread_kill(cpu->thread->thread, SIG_IPI);
-    if (err) {
+    if (err && err != ESRCH) {
         fprintf(stderr, "qemu:%s: %s", __func__, strerror(err));
         exit(1);
     }
@@ -2036,6 +2081,7 @@ void qemu_init_vcpu(CPUState *cpu)
     cpu->nr_cores = smp_cores;
     cpu->nr_threads = smp_threads;
     cpu->stopped = true;
+    cpu->random_seed = qemu_guest_random_seed_thread_part1();
 
     if (!cpu->as) {
         /* If the target cpu hasn't set up any address spaces itself,
@@ -2067,7 +2113,8 @@ void qemu_init_vcpu(CPUState *cpu)
 void cpu_stop_current(void)
 {
     if (current_cpu) {
-        qemu_cpu_stop(current_cpu, true);
+        current_cpu->stop = true;
+        cpu_exit(current_cpu);
     }
 }
 
@@ -2145,11 +2192,11 @@ int vm_stop_force_state(RunState state)
     }
 }
 
-void list_cpus(FILE *f, fprintf_function cpu_fprintf, const char *optarg)
+void list_cpus(const char *optarg)
 {
     /* XXX: implement xxx_cpu_list for targets that still miss it */
 #if defined(cpu_list)
-    cpu_list(f, cpu_fprintf);
+    cpu_list();
 #endif
 }
 
@@ -2419,19 +2466,21 @@ void qmp_inject_nmi(Error **errp)
     nmi_monitor_handle(monitor_get_cpu_index(), errp);
 }
 
-void dump_drift_info(FILE *f, fprintf_function cpu_fprintf)
+void dump_drift_info(void)
 {
     if (!use_icount) {
         return;
     }
 
-    cpu_fprintf(f, "Host - Guest clock  %"PRIi64" ms\n",
+    qemu_printf("Host - Guest clock  %"PRIi64" ms\n",
                 (cpu_get_clock() - cpu_get_icount())/SCALE_MS);
     if (icount_align_option) {
-        cpu_fprintf(f, "Max guest delay     %"PRIi64" ms\n", -max_delay/SCALE_MS);
-        cpu_fprintf(f, "Max guest advance   %"PRIi64" ms\n", max_advance/SCALE_MS);
+        qemu_printf("Max guest delay     %"PRIi64" ms\n",
+                    -max_delay / SCALE_MS);
+        qemu_printf("Max guest advance   %"PRIi64" ms\n",
+                    max_advance / SCALE_MS);
     } else {
-        cpu_fprintf(f, "Max guest delay     NA\n");
-        cpu_fprintf(f, "Max guest advance   NA\n");
+        qemu_printf("Max guest delay     NA\n");
+        qemu_printf("Max guest advance   NA\n");
     }
 }