[Qemu-devel] [PATCH v5] Enable kvm emulated watchdog

[Bharat Bhushan]

Enable the KVM emulated watchdog if KVM supports (use the
capability enablement in watchdog handler). Also watchdog exit
(KVM_EXIT_WATCHDOG) handling is added.
Watchdog state machine is cleared whenever VM state changes to running.
This is to handle the cases like return from debug halt etc.

Signed-off-by: Bharat Bhushan <address@hidden>
---
 hw/ppc_booke.c       |   46 +++++++++++++++++++++++++++---
 target-ppc/kvm.c     |   74 ++++++++++++++++++++++++++++++++++++++++++++++++++
 target-ppc/kvm_ppc.h |   23 +++++++++++++++
 3 files changed, 138 insertions(+), 5 deletions(-)

diff --git a/hw/ppc_booke.c b/hw/ppc_booke.c
index 25a4e91..897d528 100644
--- a/hw/ppc_booke.c
+++ b/hw/ppc_booke.c
@@ -28,7 +28,7 @@
 #include "nvram.h"
 #include "qemu/log.h"
 #include "loader.h"
-
+#include "kvm_ppc.h"
 
 /* Timer Control Register */
 
@@ -211,6 +211,7 @@ void store_booke_tsr(CPUPPCState *env, target_ulong val)
     PowerPCCPU *cpu = ppc_env_get_cpu(env);
 
     env->spr[SPR_BOOKE_TSR] &= ~val;
+    kvmppc_clear_tsr_bits(cpu, val);
     booke_update_irq(cpu);
 }
 
@@ -222,6 +223,7 @@ void store_booke_tcr(CPUPPCState *env, target_ulong val)
 
     tb_env = env->tb_env;
     env->spr[SPR_BOOKE_TCR] = val;
+    kvmppc_set_tcr(cpu);
 
     booke_update_irq(cpu);
 
@@ -234,7 +236,6 @@ void store_booke_tcr(CPUPPCState *env, target_ulong val)
                              booke_get_wdt_target(env, tb_env),
                              &booke_timer->wdt_next,
                              booke_timer->wdt_timer);
-
 }
 
 static void ppc_booke_timer_reset_handle(void *opaque)
@@ -242,16 +243,39 @@ static void ppc_booke_timer_reset_handle(void *opaque)
     PowerPCCPU *cpu = opaque;
     CPUPPCState *env = &cpu->env;
 
-    env->spr[SPR_BOOKE_TSR] = 0;
-    env->spr[SPR_BOOKE_TCR] = 0;
+    store_booke_tcr(env, 0);
+    store_booke_tsr(env, -1);
+}
 
-    booke_update_irq(cpu);
+/*
+ * This function will be called whenever the CPU state changes.
+ * CPU states are defined "typedef enum RunState".
+ * Regarding timer, When CPU state changes to running after debug halt
+ * or similar cases which takes time then in between final watchdog
+ * expiry happenes. This will cause exit to QEMU and configured watchdog
+ * action will be taken. To avoid this we always clear the watchdog state when
+ * state changes to running.
+ */
+static void cpu_state_change_handler(void *opaque, int running, RunState state)
+{
+    PowerPCCPU *cpu = opaque;
+    CPUPPCState *env = &cpu->env;
+
+    if (!running) {
+        return;
+    }
+
+    /*
+     * Clear watchdog interrupt condition by clearing TSR.
+     */
+    store_booke_tsr(env, TSR_ENW | TSR_WIS | TSR_WRS_MASK);
 }
 
 void ppc_booke_timers_init(PowerPCCPU *cpu, uint32_t freq, uint32_t flags)
 {
     ppc_tb_t *tb_env;
     booke_timer_t *booke_timer;
+    int ret = 0;
 
     tb_env      = g_malloc0(sizeof(ppc_tb_t));
     booke_timer = g_malloc0(sizeof(booke_timer_t));
@@ -269,5 +293,17 @@ void ppc_booke_timers_init(PowerPCCPU *cpu, uint32_t freq, 
uint32_t flags)
     booke_timer->wdt_timer =
         qemu_new_timer_ns(vm_clock, &booke_wdt_cb, cpu);
 
+    ret = kvmppc_booke_watchdog_enable(cpu);
+
+    if (ret) {
+        /* TODO: Start the QEMU emulated watchdog if not running on KVM.
+         * Also start the QEMU emulated watchdog if KVM does not support
+         * emulated watchdog or somehow it is not enabled (supported but
+         * not enabled is though some bug and requires debugging :)).
+         */
+    }
+
+    qemu_add_vm_change_state_handler(cpu_state_change_handler, cpu);
+
     qemu_register_reset(ppc_booke_timer_reset_handle, cpu);
 }
diff --git a/target-ppc/kvm.c b/target-ppc/kvm.c
index 2c64c63..bf893a1 100644
--- a/target-ppc/kvm.c
+++ b/target-ppc/kvm.c
@@ -36,6 +36,7 @@
 #include "hw/sysbus.h"
 #include "hw/spapr.h"
 #include "hw/spapr_vio.h"
+#include "hw/watchdog.h"
 
 //#define DEBUG_KVM
 
@@ -61,6 +62,7 @@ static int cap_ppc_smt;
 static int cap_ppc_rma;
 static int cap_spapr_tce;
 static int cap_hior;
+static int cap_ppc_watchdog;
 
 /* XXX We have a race condition where we actually have a level triggered
  *     interrupt, but the infrastructure can't expose that yet, so the guest
@@ -90,6 +92,7 @@ int kvm_arch_init(KVMState *s)
     cap_ppc_rma = kvm_check_extension(s, KVM_CAP_PPC_RMA);
     cap_spapr_tce = kvm_check_extension(s, KVM_CAP_SPAPR_TCE);
     cap_hior = kvm_check_extension(s, KVM_CAP_PPC_HIOR);
+    cap_ppc_watchdog = kvm_check_extension(s, KVM_CAP_PPC_BOOKE_WATCHDOG);
 
     if (!cap_interrupt_level) {
         fprintf(stderr, "KVM: Couldn't find level irq capability. Expect the "
@@ -856,6 +859,12 @@ int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
         run->epr.epr = ldl_phys(env->mpic_iack);
         ret = 0;
         break;
+    case KVM_EXIT_WATCHDOG:
+        dprintf("handle watchdog expiry\n");
+        watchdog_perform_action();
+        ret = 0;
+        break;
+
     default:
         fprintf(stderr, "KVM: unknown exit reason %d\n", run->exit_reason);
         ret = -1;
@@ -865,6 +874,71 @@ int kvm_arch_handle_exit(CPUState *cs, struct kvm_run *run)
     return ret;
 }
 
+int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
+{
+    CPUState *cs = CPU(cpu);
+    uint32_t bits = tsr_bits;
+    struct kvm_one_reg reg = {
+        .id = KVM_REG_PPC_OR_TSR,
+        .addr = (uintptr_t) &bits,
+    };
+
+    return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+}
+
+int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits)
+{
+
+    CPUState *cs = CPU(cpu);
+    uint32_t bits = tsr_bits;
+    struct kvm_one_reg reg = {
+        .id = KVM_REG_PPC_CLEAR_TSR,
+        .addr = (uintptr_t) &bits,
+    };
+
+    return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+}
+
+int kvmppc_set_tcr(PowerPCCPU *cpu)
+{
+    CPUState *cs = CPU(cpu);
+    CPUPPCState *env = &cpu->env;
+    uint32_t tcr = env->spr[SPR_BOOKE_TCR];
+
+    struct kvm_one_reg reg = {
+        .id = KVM_REG_PPC_TCR,
+        .addr = (uintptr_t) &tcr,
+    };
+
+    return kvm_vcpu_ioctl(cs, KVM_SET_ONE_REG, &reg);
+}
+
+int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
+{
+    CPUState *cs = CPU(cpu);
+    struct kvm_enable_cap encap = {};
+    int ret;
+
+    if (!kvm_enabled()) {
+        return -1;
+    }
+
+    if (!cap_ppc_watchdog) {
+        printf("warning: KVM does not support watchdog");
+        return -1;
+    }
+
+    encap.cap = KVM_CAP_PPC_BOOKE_WATCHDOG;
+    ret = kvm_vcpu_ioctl(cs, KVM_ENABLE_CAP, &encap);
+    if (ret < 0) {
+        fprintf(stderr, "%s: couldn't enable KVM_CAP_PPC_BOOKE_WATCHDOG: %s\n",
+                __func__, strerror(-ret));
+        return ret;
+    }
+
+    return ret;
+}
+
 static int read_cpuinfo(const char *field, char *value, int len)
 {
     FILE *f;
diff --git a/target-ppc/kvm_ppc.h b/target-ppc/kvm_ppc.h
index c30b006..4b4cd00 100644
--- a/target-ppc/kvm_ppc.h
+++ b/target-ppc/kvm_ppc.h
@@ -27,6 +27,10 @@ int kvmppc_set_interrupt(PowerPCCPU *cpu, int irq, int 
level);
 void kvmppc_set_papr(PowerPCCPU *cpu);
 void kvmppc_set_mpic_proxy(PowerPCCPU *cpu, int mpic_proxy);
 int kvmppc_smt_threads(void);
+int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits);
+int kvmppc_or_tsr_bits(PowerPCCPU *cpu, uint32_t tsr_bits);
+int kvmppc_set_tcr(PowerPCCPU *cpu);
+int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu);
 #ifndef CONFIG_USER_ONLY
 off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem);
 void *kvmppc_create_spapr_tce(uint32_t liobn, uint32_t window_size, int *pfd);
@@ -91,6 +95,25 @@ static inline int kvmppc_smt_threads(void)
     return 1;
 }
 
+int kvmppc_or_tsr_bits(PowerPCCPU *cpu, u32 tsr_bits)
+{
+    return 0;
+}
+int kvmppc_clear_tsr_bits(PowerPCCPU *cpu, u32 tsr_bits)
+{
+    return 0;
+}
+
+int kvmppc_set_tcr(PowerPCCPU *cpu)
+{
+    return 0;
+}
+
+static inline int kvmppc_booke_watchdog_enable(PowerPCCPU *cpu)
+{
+    return -1;
+}
+
 #ifndef CONFIG_USER_ONLY
 static inline off_t kvmppc_alloc_rma(const char *name, MemoryRegion *sysmem)
 {
-- 
1.7.0.4