[Qemu-devel] Re: [PATCH 02/21] introduce QEMUAccel and fill it with interrupt specific driver

[Anthony Liguori <aliguori@us.ibm.com>]

Glauber Costa wrote:
> From: Glauber Costa <gcosta@redhat.com>
>
> This patch introduces QEMUAccel, a placeholder for function pointers
> that aims at helping qemu to abstract accelerators such as kqemu and
> kvm (actually, the 'accelerator' name was proposed by avi kivity, since
> he loves referring to kvm that way).
>
> To begin with, the accelerator is given the opportunity to register a
> cpu_interrupt function, to be called after the raw cpu_interrupt.
> This has the side effect of, for the kqemu accelerator, calling kqemu_cpu_interrupt
> everytime, which didn't use to happen. But looking at the code, this seems safe to me.
>
> This patch applies on raw qemu.
>
> Signed-off-by: Glauber Costa <glommer@redhat.com>
> Signed-off-by: Dmitry Baryshkov <dmitry.baryshkov@siemens.com>
> ---
>  Makefile.target |    2 +-
>  accel.c         |   17 +++++++++++++++++
>  accel.h         |   20 ++++++++++++++++++++
>  exec.c          |    3 +++
>  kqemu.c         |    9 +++++++++
>  vl.c            |   17 +++++------------
>  6 files changed, 55 insertions(+), 13 deletions(-)
>  create mode 100644 accel.c
>  create mode 100644 accel.h
>
> diff --git a/Makefile.target b/Makefile.target
> index e2edf9d..623ecd8 100644
> --- a/Makefile.target
> +++ b/Makefile.target
> @@ -188,7 +188,7 @@ all: $(PROGS)
>  #########################################################
>  # cpu emulator library
>  LIBOBJS=exec.o kqemu.o translate-all.o cpu-exec.o\
> -        translate.o host-utils.o
> +        translate.o host-utils.o accel.o
>  ifdef CONFIG_DYNGEN_OP
>  exec.o: dyngen-opc.h
>  LIBOBJS+=op.o
> diff --git a/accel.c b/accel.c
> new file mode 100644
> index 0000000..d30460d
> --- /dev/null
> +++ b/accel.c
> @@ -0,0 +1,17 @@
> +#include "hw/hw.h"
> +#include "accel.h"
> +
> +QEMUAccel *current_accel;
> +
> +int _accel_nop(void)
> +{
> +    return 0;
> +}
> +
> +#define accel_nop ((void *)_accel_nop)
> +
> +/* Accelerator wrapper for the no-accel (raw qemu) case */
> +QEMUAccel noaccel = {
> +    .cpu_interrupt = accel_nop,
> +};
> +
> diff --git a/accel.h b/accel.h
> new file mode 100644
> index 0000000..8e5ddc6
> --- /dev/null
> +++ b/accel.h
> @@ -0,0 +1,20 @@
> +#ifndef _ACCEL_H_
> +#define _ACCEL_H_
> +
> +typedef struct QEMUAccel {
> +    void (*cpu_interrupt)(CPUState *env);
> +} QEMUAccel;
> +
> +extern QEMUAccel *current_accel;
> +extern QEMUAccel noaccel;
> +
> +static inline void register_qemu_accel(QEMUAccel *accel)
> +{
> +    current_accel = accel;
> +}
> +
> +static inline void accel_cpu_interrupt(CPUState *env)
> +{
> +    current_accel->cpu_interrupt(env);
> +}
> +#endif
> diff --git a/exec.c b/exec.c
> index 1cad0be..21253cc 100644
> --- a/exec.c
> +++ b/exec.c
> @@ -43,6 +43,8 @@
>  #include <qemu.h>
>  #endif
>
> +#include "accel.h"
> +
>  //#define DEBUG_TB_INVALIDATE
>  //#define DEBUG_FLUSH
>  //#define DEBUG_TLB
> @@ -1430,6 +1432,7 @@ void cpu_single_step(CPUState *env, int enabled)
>          tb_flush(env);
>      }
>  #endif
> +    accel_cpu_interrupt(env);
>  }
>
>  /* enable or disable low levels log */
> diff --git a/kqemu.c b/kqemu.c
> index 9b52237..87c06cd 100644
> --- a/kqemu.c
> +++ b/kqemu.c
> @@ -50,6 +50,7 @@
>  #include <unistd.h>
>  #include <fcntl.h>
>  #include "kqemu.h"
> +#include "accel.h"
>
>  #ifdef _WIN32
>  #define KQEMU_DEVICE "\\\\.\\kqemu"
> @@ -150,6 +151,8 @@ static void kqemu_update_cpuid(CPUState *env)
>         accelerated code */
>  }
>
> +QEMUAccel kqemu_accel;
> +
>  int kqemu_start(void)
>  {
>      struct kqemu_init kinit;
> @@ -232,6 +235,7 @@ int kqemu_start(void)
>      }
>      nb_pages_to_flush = 0;
>      nb_ram_pages_to_update = 0;
> +    register_qemu_accel(&kqemu_accel);
>
>      qpi_init();
>      return 0;
> @@ -243,6 +247,11 @@ void kqemu_init_env(CPUState *env)
>      env->kqemu_enabled = kqemu_allowed;
>  }
>
> +QEMUAccel kqemu_accel = {
> +    .cpu_interrupt = kqemu_cpu_interrupt,
> +};
>   

The hook kqemu uses is for CPU_INTERRUPT_EXIT which is basically 
intended to allow IO to run.  The fact that it's done via cpu_interrupt 
is somewhat hackish.

I think a better hook would be something like run_io_handlers.  Since 
this is called from signal handlers, we have to be clear that this code 
needs to be signal safe.  This is compared to say a generic 
cpu_interrupt hook which doesn't necessarily need to be signal safe.

Regards,

Anthony Liguori