cpus: properly kick CPUs out of inner execution loop

Now that cpu_exit() actually kicks all accelerators, use it whenever
the message to another thread is processed in qemu_wait_io_event().

Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

cpu-common.c

@@ -137,7 +137,8 @@ static void queue_work_on_cpu(CPUState *cpu, struct qemu_work_item *wi)
     wi->done = false;
     qemu_mutex_unlock(&cpu->work_mutex);
 
-    qemu_cpu_kick(cpu);
+    /* exit the inner loop and reach qemu_wait_io_event_common().  */
+    cpu_exit(cpu);
 }
 
 void do_run_on_cpu(CPUState *cpu, run_on_cpu_func func, run_on_cpu_data data,

hw/ppc/ppc.c

@@ -190,6 +190,7 @@ static void ppc970_set_irq(void *opaque, int pin, int level)
             if (level) {
                 trace_ppc_irq_cpu("stop");
                 cs->halted = 1;
+                cpu_exit(cs);
             } else {
                 trace_ppc_irq_cpu("restart");
                 cs->halted = 0;
@@ -386,6 +387,7 @@ static void ppc40x_set_irq(void *opaque, int pin, int level)
             if (level) {
                 trace_ppc_irq_cpu("stop");
                 cs->halted = 1;
+                cpu_exit(cs);
             } else {
                 trace_ppc_irq_cpu("restart");
                 cs->halted = 0;

hw/ppc/spapr_hcall.c

@@ -509,8 +509,8 @@ static target_ulong h_cede(PowerPCCPU *cpu, SpaprMachineState *spapr,
     if (!cpu_has_work(cs)) {
         cs->halted = 1;
         cs->exception_index = EXCP_HLT;
-        qatomic_set(&cs->exit_request, true);
         ppc_maybe_interrupt(env);
+        cpu_exit(cs);
     }
 
     return H_SUCCESS;
@@ -531,8 +531,8 @@ static target_ulong h_confer_self(PowerPCCPU *cpu)
     }
     cs->halted = 1;
     cs->exception_index = EXCP_HALTED;
-    qatomic_set(&cs->exit_request, true);
     ppc_maybe_interrupt(&cpu->env);
+    cpu_exit(cs);
 
     return H_SUCCESS;
 }
@@ -624,8 +624,7 @@ static target_ulong h_confer(PowerPCCPU *cpu, SpaprMachineState *spapr,
     }
 
     cs->exception_index = EXCP_YIELD;
-    qatomic_set(&cs->exit_request, true);
-    cpu_loop_exit(cs);
+    cpu_exit(cs);
 
     return H_SUCCESS;
 }

hw/ppc/spapr_rtas.c

@@ -221,7 +221,7 @@ static void rtas_stop_self(PowerPCCPU *cpu, SpaprMachineState *spapr,
     cs->halted = 1;
     ppc_store_lpcr(cpu, env->spr[SPR_LPCR] & ~pcc->lpcr_pm);
     kvmppc_set_reg_ppc_online(cpu, 0);
-    qemu_cpu_kick(cs);
+    cpu_exit(cs);
 }
 
 static void rtas_ibm_suspend_me(PowerPCCPU *cpu, SpaprMachineState *spapr,

replay/replay-events.c

@@ -118,7 +118,8 @@ void replay_add_event(ReplayAsyncEventKind event_kind,
 
     g_assert(replay_mutex_locked());
     QTAILQ_INSERT_TAIL(&events_list, event, events);
-    qemu_cpu_kick(first_cpu);
+    /* Kick the TCG thread out of tcg_cpu_exec().  */
+    cpu_exit(first_cpu);
 }
 
 void replay_bh_schedule_event(QEMUBH *bh)

system/cpu-timers.c

@@ -246,14 +246,14 @@ void qemu_timer_notify_cb(void *opaque, QEMUClockType type)
 
     if (qemu_in_vcpu_thread()) {
         /*
-         * A CPU is currently running; kick it back out to the
+         * A CPU is currently running; send it out of the
          * tcg_cpu_exec() loop so it will recalculate its
          * icount deadline immediately.
          */
-        qemu_cpu_kick(current_cpu);
+        cpu_exit(current_cpu);
     } else if (first_cpu) {
         /*
-         * qemu_cpu_kick is not enough to kick a halted CPU out of
+         * cpu_exit() is not enough to kick a halted CPU out of
          * qemu_tcg_wait_io_event.  async_run_on_cpu, instead,
          * causes cpu_thread_is_idle to return false.  This way,
          * handle_icount_deadline can run.

system/cpus.c

@@ -604,7 +604,7 @@ void cpu_pause(CPUState *cpu)
         qemu_cpu_stop(cpu, true);
     } else {
         cpu->stop = true;
-        qemu_cpu_kick(cpu);
+        cpu_exit(cpu);
     }
 }
 
@@ -644,6 +644,7 @@ void pause_all_vcpus(void)
 
     while (!all_vcpus_paused()) {
         qemu_cond_wait(&qemu_pause_cond, &bql);
+        /* FIXME: is this needed? */
         CPU_FOREACH(cpu) {
             qemu_cpu_kick(cpu);
         }
@@ -672,7 +673,7 @@ void cpu_remove_sync(CPUState *cpu)
 {
     cpu->stop = true;
     cpu->unplug = true;
-    qemu_cpu_kick(cpu);
+    cpu_exit(cpu);
     bql_unlock();
     qemu_thread_join(cpu->thread);
     bql_lock();

target/arm/tcg/mte_helper.c

@@ -591,7 +591,7 @@ static void mte_async_check_fail(CPUARMState *env, uint64_t dirty_ptr,
      * which is rather sooner than "normal".  But the alternative
      * is waiting until the next syscall.
      */
-    qemu_cpu_kick(env_cpu(env));
+    cpu_exit(env_cpu(env));
 #endif
 }
 

target/i386/kvm/hyperv.c

@@ -81,7 +81,6 @@ int kvm_hv_handle_exit(X86CPU *cpu, struct kvm_hyperv_exit *exit)
          * necessary because memory hierarchy is being changed
          */
         async_safe_run_on_cpu(CPU(cpu), async_synic_update, RUN_ON_CPU_NULL);
-        cpu_exit(CPU(cpu));
 
         return EXCP_INTERRUPT;
     case KVM_EXIT_HYPERV_HCALL: {