s390: Fix cpu shutdown for KVM

qemu/kvm on s390 currently hangs on panic (doesnt exit on disabled wait) and also shuts down on cpu hot-unplug (SIGP stop). This patch tries to fix these simple cases.
On s390 a shutdown is the state of all CPUs being either stopped or disabled (for interrupts) waiting. We have to track this number to call the shutdown sequence accordingly. This patch implements the counting and shutdown handling for the kvm path in qemu.

Signed-off-by Christian Borntraeger<[email protected]>

---
  hw/s390-virtio.c   |    4 ++++
  target-s390x/kvm.c |   29 +++++++++++++++++++++++++----
  2 files changed, 29 insertions(+), 4 deletions(-)

Index: b/hw/s390-virtio.c
===================================================================

hw/s390-virtio.c +2 -0

+
  /* PC hardware initialisation */
  static void s390_init(ram_addr_t my_ram_size,
                        const char *boot_device,

@@ -189,6 +192,7 @@ static void s390_init(ram_addr_t my_ram_
 
      env->halted = 0;
      env->exception_index = 0;
+    s390_running_cpus = 1;
 
      if (kernel_filename) {
          kernel_size = load_image(kernel_filename, qemu_get_ram_ptr(0));

Index: b/target-s390x/kvm.c
===================================================================

target-s390x/kvm.c +25 -4

@@ -63,6 +63,14 @@
  #define SCLP_CMDW_READ_SCP_INFO         0x00020001
  #define SCLP_CMDW_READ_SCP_INFO_FORCED  0x00120001
 
+/*
+ * The number of running CPUs. On s390 a shutdown is the state of all CPUs
+ * being either stopped or disabled (for interrupts) waiting. We have to
+ * track this number to call the shutdown sequence accordingly. This
+ * number is modified either on startup or while holding the big qemu lock.
+ */
+int s390_running_cpus;
+
  const KVMCapabilityInfo kvm_arch_required_capabilities[] = {
      KVM_CAP_LAST_INFO
  };

@@ -185,6 +193,12 @@ void kvm_s390_interrupt_internal(CPUStat
          return;
      }
 
+    /*
+     * We can only deliver interrupts to (interrupt) enabled CPUs.
+     * We dont modify s390_running_cpus here, since CPUs in enabled wait
+     * will wait inside the kernel (no exit). Therefore, the targeted
+     * CPUs was neither disabled waiting or stopped for qemu.
+     */
      env->halted = 0;
      env->exception_index = -1;
      qemu_cpu_kick(env);

@@ -301,6 +315,7 @@ static int s390_cpu_restart(CPUState *en
      kvm_s390_interrupt(env, KVM_S390_RESTART, 0);
      env->halted = 0;
      env->exception_index = -1;
+    s390_running_cpus++;
      qemu_cpu_kick(env);
      dprintf("DONE: SIGP cpu restart: %p\n", env);
      return 0;

@@ -425,16 +440,24 @@ static int handle_intercept(CPUState *en
              r = handle_instruction(env, run);
              break;
          case ICPT_WAITPSW:
-            /* XXX What to do on system shutdown? */
+            if (--s390_running_cpus == 0) {
+                qemu_system_shutdown_request();
+            }
              env->halted = 1;
              env->exception_index = EXCP_HLT;
+            r = EXCP_HALTED;
              break;
          case ICPT_SOFT_INTERCEPT:
              fprintf(stderr, "KVM unimplemented icpt SOFT\n");
              exit(1);
              break;
          case ICPT_CPU_STOP:
-            qemu_system_shutdown_request();
+            if (--s390_running_cpus == 0) {
+                qemu_system_shutdown_request();
+            }
+            env->halted = 1;
+            env->exception_index = EXCP_HLT;
+            r = EXCP_HALTED;
              break;
          case ICPT_IO:
              fprintf(stderr, "KVM unimplemented icpt IO\n");

@@ -468,8 +491,6 @@ int kvm_arch_handle_exit(CPUState *env,
 
      if (ret == 0) {
          ret = EXCP_INTERRUPT;
-    } else if (ret>  0) {
-        ret = 0;
      }
      return ret;
  }

+     * We can only deliver interrupts to (interrupt) enabled CPUs.
+     * We dont call kvm_s390_add_running_cpu here, since CPUs in enabled wait
+     * will wait inside the kernel (no exit). Therefore, the targeted
+     * CPUs was neither disabled waiting or stopped for qemu.

target-s390x/kvm.c +0 -0

@@ -425,16 +432,24 @@ static int handle_intercept(CPUState *en
             r = handle_instruction(env, run);
             break;

        case ICPT_WAITPSW:
-            /* XXX What to do on system shutdown? */
+            if (kvm_s390_del_running_cpu(env) == 0) {
+                qemu_system_shutdown_request();
+            }
            env->halted = 1;
            env->exception_index = EXCP_HLT;
+            r = EXCP_HALTED;
            break;
        case ICPT_SOFT_INTERCEPT:
            fprintf(stderr, "KVM unimplemented icpt SOFT\n"); exit(1);
            break;
        case ICPT_CPU_STOP:
-            qemu_system_shutdown_request();
+            if (kvm_s390_del_running_cpu(env) == 0) {

+                qemu_system_shutdown_request();
+            }
+            env->halted = 1;
+            env->exception_index = EXCP_HLT;
+            r = EXCP_HALTED;
            break;
        case ICPT_IO:
            fprintf(stderr, "KVM unimplemented icpt IO\n");

target-s390x/kvm.c void kvm_s390_interrupt_internal(CPUStat · line 192

Quoted 4 lines · Christian Borntraeger

target-s390x/kvm.c +4 -0

+     * We can only deliver interrupts to (interrupt) enabled CPUs.
+     * We dont call kvm_s390_add_running_cpu here, since CPUs in enabled wait
+     * will wait inside the kernel (no exit). Therefore, the targeted
+     * CPUs was neither disabled waiting or stopped for qemu.

Grammar nits, since I'm commenting anyway:
"don't"
"neither disabled waiting nor stopped"

This makes the ICPT_CPU_STOP and ICPT_WAITPSW cases identical, right? You should just fold them together.

Something like the following?

s390: Fix cpu shutdown for KVM

On s390 a shutdown is the state of all CPUs being either stopped or disabled (for interrupts) waiting. We have to track this number to call the shutdown sequence accordingly. This patch implements the counting and shutdown handling for the kvm path in qemu.

Signed-off-by Christian Borntraeger<[email protected]>

---
  hw/s390-virtio.c   |   21 +++++++++++++++++++++
  target-s390x/cpu.h |   12 ++++++++++++
  target-s390x/kvm.c |   21 +++++++++++++++++----
  3 files changed, 50 insertions(+), 4 deletions(-)

Index: b/hw/s390-virtio.c
===================================================================

hw/s390-virtio.c +9 -0

@@ -130,6 +130,26 @@ int s390_virtio_hypercall(CPUState *env,
      return r;
  }
 
+/*
+ * The number of running CPUs. On s390 a shutdown is the state of all CPUs
+ * being either stopped or disabled (for interrupts) waiting. We have to
+ * track this number to call the shutdown sequence accordingly. This
+ * number is modified either on startup or while holding the big qemu lock.
+ */
+static unsigned s390_running_cpus;
+
+void kvm_s390_add_running_cpu(CPUState *env)

hw/s390-virtio.c +2 -0

+
  /* PC hardware initialisation */
  static void s390_init(ram_addr_t my_ram_size,
                        const char *boot_device,

@@ -189,6 +209,7 @@ static void s390_init(ram_addr_t my_ram_
 
      env->halted = 0;
      env->exception_index = 0;
+    kvm_s390_add_running_cpu(env);
 
      if (kernel_filename) {
          kernel_size = load_image(kernel_filename, qemu_get_ram_ptr(0));

Index: b/target-s390x/cpu.h
===================================================================

target-s390x/cpu.h +2 -0

@@ -292,6 +292,8 @@ void kvm_s390_interrupt(CPUState *env, i
  void kvm_s390_virtio_irq(CPUState *env, int config_change, uint64_t token);
  void kvm_s390_interrupt_internal(CPUState *env, int type, uint32_t parm,
                                   uint64_t parm64, int vm);
+void kvm_s390_add_running_cpu(CPUState *env);
+unsigned kvm_s390_del_running_cpu(CPUState *env);
  #else
  static inline void kvm_s390_interrupt(CPUState *env, int type, uint32_t code)
  {

@@ -307,6 +309,16 @@ static inline void kvm_s390_interrupt_in
                                                 int vm)
  {

target-s390x/cpu.h +1 -0

+
  #endif
  CPUState *s390_cpu_addr2state(uint16_t cpu_addr);
 

Index: b/target-s390x/kvm.c
===================================================================

target-s390x/kvm.c +6 -0

@@ -185,6 +185,12 @@ void kvm_s390_interrupt_internal(CPUStat
          return;
      }
 
+    /*
+     * We can only deliver interrupts to (interrupt) enabled CPUs.
+     * We dont call kvm_s390_add_running_cpu here, since CPUs in enabled wait
+     * will wait inside the kernel (no exit). Therefore, the targeted
+     * CPUs was neither disabled waiting or stopped for qemu.
+     */
      env->halted = 0;
      env->exception_index = -1;
      qemu_cpu_kick(env);

On s390 a shutdown is the state of all CPUs being either stopped or disabled (for interrupts) waiting. We have to track the overall number of running CPUs to call the shutdown sequence accordingly. This patch implements the counting and shutdown handling for the kvm path in qemu.
Lets also wrap changes to env->halted and env->exception_index.

Signed-off-by Christian Borntraeger<[email protected]>

---
  hw/s390-virtio.c   |   32 ++++++++++++++++++++++++++++++--
  target-s390x/cpu.h |    2 ++
  target-s390x/kvm.c |   19 +++++++------------
  3 files changed, 39 insertions(+), 14 deletions(-)

Index: b/hw/s390-virtio.c
===================================================================

hw/s390-virtio.c +30 -2

@@ -130,6 +130,34 @@ int s390_virtio_hypercall(CPUState *env,
      return r;
  }
 
+/*
+ * The number of running CPUs. On s390 a shutdown is the state of all CPUs
+ * being either stopped or disabled (for interrupts) waiting. We have to
+ * track this number to call the shutdown sequence accordingly. This
+ * number is modified either on startup or while holding the big qemu lock.
+ */
+static unsigned s390_running_cpus;
+
+void s390_add_running_cpu(CPUState *env)
+{
+    if (env->halted) {
+        s390_running_cpus++;
+        env->halted = 0;
+        env->exception_index = -1;
+    }
+}
+
+unsigned s390_del_running_cpu(CPUState *env)
+{
+    if (env->halted == 0) {
+        assert(s390_running_cpus>= 1);
+        s390_running_cpus--;
+        env->halted = 1;
+        env->exception_index = EXCP_HLT;
+    }
+    return s390_running_cpus;
+}
+
  /* PC hardware initialisation */
  static void s390_init(ram_addr_t my_ram_size,
                        const char *boot_device,

@@ -187,8 +215,8 @@ static void s390_init(ram_addr_t my_ram_
          tmp_env->storage_keys = storage_keys;
      }
 
-    env->halted = 0;
-    env->exception_index = 0;
+    /* One CPU has to run */
+    s390_add_running_cpu(env);
 
      if (kernel_filename) {
          kernel_size = load_image(kernel_filename, qemu_get_ram_ptr(0));

Index: b/target-s390x/cpu.h
===================================================================

target-s390x/cpu.h +2 -0

@@ -309,6 +309,8 @@ static inline void kvm_s390_interrupt_in
  }
  #endif
  CPUState *s390_cpu_addr2state(uint16_t cpu_addr);
+void s390_add_running_cpu(CPUState *env);
+unsigned s390_del_running_cpu(CPUState *env);
 
  /* from s390-virtio-bus */
  extern const target_phys_addr_t virtio_size;

Index: b/target-s390x/kvm.c
===================================================================

target-s390x/kvm.c +7 -10

@@ -185,8 +185,7 @@ void kvm_s390_interrupt_internal(CPUStat
          return;
      }
 
-    env->halted = 0;
-    env->exception_index = -1;
+    s390_add_running_cpu(env);
      qemu_cpu_kick(env);
 
      kvmint.type = type;

@@ -299,8 +298,7 @@ static int handle_diag(CPUState *env, st
  static int s390_cpu_restart(CPUState *env)
  {
      kvm_s390_interrupt(env, KVM_S390_RESTART, 0);
-    env->halted = 0;
-    env->exception_index = -1;
+    s390_add_running_cpu(env);
      qemu_cpu_kick(env);
      dprintf("DONE: SIGP cpu restart: %p\n", env);
      return 0;

@@ -425,17 +423,16 @@ static int handle_intercept(CPUState *en
              r = handle_instruction(env, run);
              break;
          case ICPT_WAITPSW:
-            /* XXX What to do on system shutdown? */
-            env->halted = 1;
-            env->exception_index = EXCP_HLT;
+        case ICPT_CPU_STOP:
+            if (s390_del_running_cpu(env) == 0) {
+                qemu_system_shutdown_request();
+            }
+            r = EXCP_HALTED;
              break;
          case ICPT_SOFT_INTERCEPT:
              fprintf(stderr, "KVM unimplemented icpt SOFT\n");
              exit(1);
              break;
-        case ICPT_CPU_STOP:
-            qemu_system_shutdown_request();
-            break;
          case ICPT_IO:
              fprintf(stderr, "KVM unimplemented icpt IO\n");
              exit(1);

Hrm. Are you sure that this doesn't break anything? The rest looks good.

On s390 a shutdown is the state of all CPUs being either stopped or disabled (for interrupts) waiting. We have to track the overall number of running CPUs to call the shutdown sequence accordingly. This patch implements the counting and shutdown handling for the kvm path in qemu.
Lets also wrap changes to env->halted and env->exception_index.

Signed-off-by Christian Borntraeger <[email protected]>