When guest os crash, there is no way to notify qemu on the host quickly (within 1s), and it is
not convenient for libvirt management. We suggest to register a IO port (with specific addr) when
qemu process be launched, on the other hand, a linux module which is comprised of kprobe hooks will
be inserted in VM. When guest linux os crash, kprobe hooks write IO port mentioned above to notify
qemu. With this approach, VM management component (e.g.libvirt) can monitor VM running state in time.

Signed-off-by: Huang Yong<huang.yong@zte.com.cn>
---
 vl.c | 35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/vl.c b/vl.c
index 0b4ed52..2760c52 100644
--- a/vl.c
+++ b/vl.c
@@ -129,9 +129,12 @@ int main(int argc, char **argv)
 #include "sysemu/replay.h"
 #include "qapi/qmp/qerror.h"
 #include "sysemu/iothread.h"
+#include "exec/memory.h"
+#include "qapi/error.h"
 
 #define MAX_VIRTIO_CONSOLES 1
 #define MAX_SCLP_CONSOLES 1
+#define CRASH_IO_PORT (0xcff0)
 
 static const char *data_dir[16];
 static int data_dir_idx;
@@ -2939,6 +2942,37 @@ static int qemu_read_default_config_file(void)
     return 0;
 }
 
+static MemoryRegion crash_io;
+
+static uint64_t crash_io_read(void *opaque, hwaddr addr,
+                              unsigned int size)
+{
+    return 0;
+}
+
+static void crash_io_write(void *opaque, hwaddr addr,
+                           uint64_t value, unsigned int size)
+{
+    qapi_event_send_guest_panicked(GUEST_PANIC_ACTION_PAUSE,
+                                   false, NULL, &error_abort);
+}
+
+static const MemoryRegionOps crash_io_ops = {
+    .read = crash_io_read,
+    .write = crash_io_write,
+    .endianness = DEVICE_LITTLE_ENDIAN,
+};
+
+static void init_crash_port(int port)
+{
+    int addr = port;
+    MemoryRegion *system_io = get_system_io();
+
+    memory_region_init_io(&crash_io, NULL, &crash_io_ops,
+                          NULL, "crash_event_port", 2);
+    memory_region_add_subregion(system_io, addr, &crash_io);
+}
+
 int main(int argc, char **argv, char **envp)
 {
     int i;
@@ -4706,6 +4740,7 @@ int main(int argc, char **argv, char **envp)
 
     os_setup_post();
 
+    init_crash_port(CRASH_IO_PORT);
     main_loop();
     replay_disable_events();
     iothread_stop_all();
-- 
1.8.3.1

On 12/04/2017 10:59, Huang Yong wrote:
> When guest os crash, there is no way to notify qemu on the host quickly (within 1s), and it is
> not convenient for libvirt management. We suggest to register a IO port (with specific addr) when
> qemu process be launched, on the other hand, a linux module which is comprised of kprobe hooks will
> be inserted in VM. When guest linux os crash, kprobe hooks write IO port mentioned above to notify
> qemu. With this approach, VM management component (e.g.libvirt) can monitor VM running state in time.

QEMU already supports this using the "pvpanic" device.  There is also a
driver in drivers/platform/x86/pvpanic.c.

Paolo

> Signed-off-by: Huang Yong<huang.yong@zte.com.cn>
> ---
>  vl.c | 35 +++++++++++++++++++++++++++++++++++
>  1 file changed, 35 insertions(+)
> 
> diff --git a/vl.c b/vl.c
> index 0b4ed52..2760c52 100644
> --- a/vl.c
> +++ b/vl.c
> @@ -129,9 +129,12 @@ int main(int argc, char **argv)
>  #include "sysemu/replay.h"
>  #include "qapi/qmp/qerror.h"
>  #include "sysemu/iothread.h"
> +#include "exec/memory.h"
> +#include "qapi/error.h"
>  
>  #define MAX_VIRTIO_CONSOLES 1
>  #define MAX_SCLP_CONSOLES 1
> +#define CRASH_IO_PORT (0xcff0)
>  
>  static const char *data_dir[16];
>  static int data_dir_idx;
> @@ -2939,6 +2942,37 @@ static int qemu_read_default_config_file(void)
>      return 0;
>  }
>  
> +static MemoryRegion crash_io;
> +
> +static uint64_t crash_io_read(void *opaque, hwaddr addr,
> +                              unsigned int size)
> +{
> +    return 0;
> +}
> +
> +static void crash_io_write(void *opaque, hwaddr addr,
> +                           uint64_t value, unsigned int size)
> +{
> +    qapi_event_send_guest_panicked(GUEST_PANIC_ACTION_PAUSE,
> +                                   false, NULL, &error_abort);
> +}
> +
> +static const MemoryRegionOps crash_io_ops = {
> +    .read = crash_io_read,
> +    .write = crash_io_write,
> +    .endianness = DEVICE_LITTLE_ENDIAN,
> +};
> +
> +static void init_crash_port(int port)
> +{
> +    int addr = port;
> +    MemoryRegion *system_io = get_system_io();
> +
> +    memory_region_init_io(&crash_io, NULL, &crash_io_ops,
> +                          NULL, "crash_event_port", 2);
> +    memory_region_add_subregion(system_io, addr, &crash_io);
> +}
> +
>  int main(int argc, char **argv, char **envp)
>  {
>      int i;
> @@ -4706,6 +4740,7 @@ int main(int argc, char **argv, char **envp)
>  
>      os_setup_post();
>  
> +    init_crash_port(CRASH_IO_PORT);
>      main_loop();
>      replay_disable_events();
>      iothread_stop_all();
>