Re: [Qemu-devel] [PATCH] pseries: Implement RTAS system-reboot call

["Andreas Färber" <afaerber@suse.de>]

Am 28.03.2012 03:05, schrieb David Gibson:
> This patch adds the PAPR defined RTAS system-reboot call to the pseries
> machine emulation, providing the guest with a way to trigger a reboot.
> This exposes a bug in the pseries VIO code which means CRQs are not
> properly reset on a system reset.  This patch also fixes that bug by
> adding a suitable reset handler.
> 
> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>

Patch overall looks good to me, except for some minor things.

> 
> Conflicts:
> 
> 	hw/spapr_vio.c

Please drop conflicts from commit messages (but no need to resend).

> ---
>  hw/spapr_rtas.c |   14 ++++++++++++++
>  hw/spapr_vio.c  |   33 +++++++++++++++++++++++++--------
>  2 files changed, 39 insertions(+), 8 deletions(-)
> 
> diff --git a/hw/spapr_rtas.c b/hw/spapr_rtas.c
> index 0946585..480a4ae 100644
> --- a/hw/spapr_rtas.c
> +++ b/hw/spapr_rtas.c
> @@ -112,6 +112,19 @@ static void rtas_power_off(sPAPREnvironment *spapr,
>      rtas_st(rets, 0, 0);
>  }
>  
> +static void rtas_system_reboot(sPAPREnvironment *spapr,
> +                               uint32_t token, uint32_t nargs,
> +                               target_ulong args,
> +                               uint32_t nret, target_ulong rets)
> +{
> +    if (nargs != 0 || nret != 1) {
> +        rtas_st(rets, 0, -3);
> +        return;
> +    }
> +    qemu_system_reset_request();
> +    rtas_st(rets, 0, 0);
> +}
> +
>  static void rtas_query_cpu_stopped_state(sPAPREnvironment *spapr,
>                                           uint32_t token, uint32_t nargs,
>                                           target_ulong args,
> @@ -294,6 +307,7 @@ static void core_rtas_register_types(void)
>      spapr_rtas_register("get-time-of-day", rtas_get_time_of_day);
>      spapr_rtas_register("set-time-of-day", rtas_set_time_of_day);
>      spapr_rtas_register("power-off", rtas_power_off);
> +    spapr_rtas_register("system-reboot", rtas_system_reboot);
>      spapr_rtas_register("query-cpu-stopped-state",
>                          rtas_query_cpu_stopped_state);
>      spapr_rtas_register("start-cpu", rtas_start_cpu);

This part looks good to go.

> diff --git a/hw/spapr_vio.c b/hw/spapr_vio.c
> index dbf5a90..200f27c 100644
> --- a/hw/spapr_vio.c
> +++ b/hw/spapr_vio.c
> @@ -435,12 +435,13 @@ static target_ulong h_reg_crq(CPUPPCState *env, sPAPREnvironment *spapr,
>  
>      /* Check if device supports CRQs */
>      if (!dev->crq.SendFunc) {
> +        hcall_dprintf("h_reg_crq, device does not support CRQ\n");
>          return H_NOT_FOUND;
>      }
>  
> -

Was this whitespace change intentional for consistency?

>      /* Already a queue ? */
>      if (dev->crq.qsize) {
> +        hcall_dprintf("h_reg_crq, CRQ already registered\n");

Adding dprintfs for CRQs seems only indirectly related. I'd suggest a
separate patch and using __func__ so that function name and comment
don't start to differ.

>          return H_RESOURCE;
>      }
>      dev->crq.qladdr = queue_addr;
> @@ -453,6 +454,17 @@ static target_ulong h_reg_crq(CPUPPCState *env, sPAPREnvironment *spapr,
>      return H_SUCCESS;
>  }
>  
> +static target_ulong free_crq(VIOsPAPRDevice *dev)
> +{
> +    dev->crq.qladdr = 0;
> +    dev->crq.qsize = 0;
> +    dev->crq.qnext = 0;
> +
> +    dprintf("CRQ for dev 0x%" PRIx32 " freed\n", dev->reg);
> +
> +    return H_SUCCESS;
> +}
> +
>  static target_ulong h_free_crq(CPUPPCState *env, sPAPREnvironment *spapr,
>                                 target_ulong opcode, target_ulong *args)
>  {
> @@ -465,13 +477,7 @@ static target_ulong h_free_crq(CPUPPCState *env, sPAPREnvironment *spapr,
>          return H_PARAMETER;
>      }
>  
> -    dev->crq.qladdr = 0;
> -    dev->crq.qsize = 0;
> -    dev->crq.qnext = 0;
> -
> -    dprintf("CRQ for dev 0x" TARGET_FMT_lx " freed\n", reg);
> -
> -    return H_SUCCESS;
> +    return free_crq(dev);
>  }
>  
>  static target_ulong h_send_crq(CPUPPCState *env, sPAPREnvironment *spapr,

Here freeing crq is moved to its own static function...

> @@ -649,6 +655,15 @@ static int spapr_vio_check_reg(VIOsPAPRDevice *sdev)
>      return 0;
>  }
>  
> +static void spapr_vio_busdev_reset(void *opaque)
> +{
> +    VIOsPAPRDevice *dev = (VIOsPAPRDevice *)opaque;
> +
> +    if (dev->crq.qsize) {
> +        free_crq(dev);
> +    }
> +}
> +

...so that it can be used...

>  static int spapr_vio_busdev_init(DeviceState *qdev)
>  {
>      VIOsPAPRDevice *dev = (VIOsPAPRDevice *)qdev;
> @@ -677,6 +692,8 @@ static int spapr_vio_busdev_init(DeviceState *qdev)
>  
>      rtce_init(dev);
>  
> +    qemu_register_reset(spapr_vio_busdev_reset, dev);
> +
>      return pc->init(dev);
>  }
>  

... for a new reset handler.

Would you mind splitting this off into a preceding patch? Cleaning up /
freeing resources on reset seems a good idea independent of the new
hcall (thinking of reset from monitor interface).

I'll look into rebasing ppc-next meanwhile.

Andreas

-- 
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg