Patchwork s390: Fix cpu shutdown for KVM

login
register
mail settings
Submitter Christian Borntraeger
Date Oct. 4, 2011, 1:47 p.m.
Message ID <4E8B0E8C.9030609@de.ibm.com>
Download mbox | patch
Permalink /patch/117622/
State New
Headers show

Comments

Christian Borntraeger - Oct. 4, 2011, 1:47 p.m.
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 <borntraeger@de.ibm.com>
---
 hw/s390-virtio.c   |   21 +++++++++++++++++++++
 target-s390x/cpu.h |   12 ++++++++++++
 target-s390x/kvm.c |   21 +++++++++++++++++----
 3 files changed, 50 insertions(+), 4 deletions(-)
Peter Maydell - Oct. 4, 2011, 1:56 p.m.
On 4 October 2011 14:47, Christian Borntraeger <borntraeger@de.ibm.com> wrote:
> +     * 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"

> @@ -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");

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

-- PMM
Christian Borntraeger - Oct. 4, 2011, 2:16 p.m.
On 04/10/11 15:56, Peter Maydell wrote:
> On 4 October 2011 14:47, Christian Borntraeger <borntraeger@de.ibm.com> wrote:
>> +     * 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"

Ok. Alex can you fix that up or do you want a new patch?

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

Yes, at the moment they are identical.
I am still thinking about some additional changes that will make them separate again
due to their usage in Linux:
- disabled wait usually indicates a kernel panic
- stop is called during cpu hot unplug and during shutdown for the last cpu

So on disabled wait we might want to perform extra logging etc, but I dont know yet.
Should I merge them anyway?

Christian
Alexander Graf - Oct. 4, 2011, 2:37 p.m.
On 10/04/2011 03:47 PM, Christian Borntraeger wrote:
> 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<borntraeger@de.ibm.com>
> ---
>   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
> ===================================================================
> --- a/hw/s390-virtio.c
> +++ b/hw/s390-virtio.c
> @@ -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)
> +{
> +    assert(env->halted == 0);
> +    s390_running_cpus++;
> +}
> +
> +unsigned kvm_s390_del_running_cpu(CPUState *env)
> +{
> +    assert(s390_running_cpus>= 1);
> +    return --s390_running_cpus;
> +}

Almost. It has nothing to do with KVM though. Please keep the 
abstraction intact!

> +
>   /* 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
> ===================================================================
> --- a/target-s390x/cpu.h
> +++ b/target-s390x/cpu.h
> @@ -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)
>   {
>   }
> +
> +static inline void kvm_s390_add_running_cpu(CPUState *env)
> +{
> +}
> +
> +static inline unsigned kvm_s390_del_running_cpu(CPUState *env)
> +{
> +    return 0;
> +}

The functions are defined in s390-virtio.c and thus are available 
anywhere now. No need to special-case KVM anywhere.

> +
>   #endif
>   CPUState *s390_cpu_addr2state(uint16_t cpu_addr);
>
> Index: b/target-s390x/kvm.c
> ===================================================================
> --- a/target-s390x/kvm.c
> +++ b/target-s390x/kvm.c
> @@ -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);
> @@ -301,6 +307,7 @@ static int s390_cpu_restart(CPUState *en
>       kvm_s390_interrupt(env, KVM_S390_RESTART, 0);
>       env->halted = 0;
>       env->exception_index = -1;
> +    kvm_s390_add_running_cpu(env);

Hrm. Is add_running correct here? Maybe we should add a wrapper around 
the halted = 0 parts in s390-virtio.c and do the increase of the counter 
there? That way we can make sure that transitioning from 0 -> 0 doesn't 
increase the counter.


Alex

Patch

Index: b/hw/s390-virtio.c
===================================================================
--- a/hw/s390-virtio.c
+++ b/hw/s390-virtio.c
@@ -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)
+{
+    assert(env->halted == 0);
+    s390_running_cpus++;
+}
+
+unsigned kvm_s390_del_running_cpu(CPUState *env)
+{
+    assert(s390_running_cpus >= 1);
+    return --s390_running_cpus;
+}
+
 /* 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
===================================================================
--- a/target-s390x/cpu.h
+++ b/target-s390x/cpu.h
@@ -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)
 {
 }
+
+static inline void kvm_s390_add_running_cpu(CPUState *env)
+{
+}
+
+static inline unsigned kvm_s390_del_running_cpu(CPUState *env)
+{
+    return 0;
+}
+
 #endif
 CPUState *s390_cpu_addr2state(uint16_t cpu_addr);
 
Index: b/target-s390x/kvm.c
===================================================================
--- a/target-s390x/kvm.c
+++ b/target-s390x/kvm.c
@@ -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);
@@ -301,6 +307,7 @@  static int s390_cpu_restart(CPUState *en
     kvm_s390_interrupt(env, KVM_S390_RESTART, 0);
     env->halted = 0;
     env->exception_index = -1;
+    kvm_s390_add_running_cpu(env);
     qemu_cpu_kick(env);
     dprintf("DONE: SIGP cpu restart: %p\n", env);
     return 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");
@@ -468,8 +483,6 @@  int kvm_arch_handle_exit(CPUState *env, 
 
     if (ret == 0) {
         ret = EXCP_INTERRUPT;
-    } else if (ret > 0) {
-        ret = 0;
     }
     return ret;
 }