Re: [Qemu-ppc] [PATCH v4 2/9] spapr: move the IRQ server number mapping under the machine

[David Gibson]

On Thu, Mar 30, 2017 at 03:04:47PM +0200, Cédric Le Goater wrote:
> On 03/30/2017 08:46 AM, David Gibson wrote:
> > On Wed, Mar 29, 2017 at 03:53:24PM +0200, Cédric Le Goater wrote:
> >> This is the second step to abstract the IRQ 'server' number of the
> >> XICS layer. Now that the prereq cleanups have been done in the
> >> previous patch, we can move down the 'cpu_dt_id' to 'cpu_index'
> >> mapping in the sPAPR machine handler.
> >>
> >> Signed-off-by: Cédric Le Goater <address@hidden>
> >> Reviewed-by: David Gibson <address@hidden>
> >> ---
> >>  hw/intc/xics_spapr.c    | 5 ++---
> >>  hw/ppc/spapr.c          | 3 ++-
> >>  hw/ppc/spapr_cpu_core.c | 2 +-
> >>  3 files changed, 5 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/hw/intc/xics_spapr.c b/hw/intc/xics_spapr.c
> >> index 58f100d379cb..f05308b897f2 100644
> >> --- a/hw/intc/xics_spapr.c
> >> +++ b/hw/intc/xics_spapr.c
> >> @@ -52,9 +52,8 @@ static target_ulong h_cppr(PowerPCCPU *cpu, 
> >> sPAPRMachineState *spapr,
> >>  static target_ulong h_ipi(PowerPCCPU *cpu, sPAPRMachineState *spapr,
> >>                            target_ulong opcode, target_ulong *args)
> >>  {
> >> -    target_ulong server = xics_get_cpu_index_by_dt_id(args[0]);
> >>      target_ulong mfrr = args[1];
> >> -    ICPState *icp = xics_icp_get(XICS_FABRIC(spapr), server);
> >> +    ICPState *icp = xics_icp_get(XICS_FABRIC(spapr), args[0]);
> >>  
> >>      if (!icp) {
> >>          return H_PARAMETER;
> >> @@ -122,7 +121,7 @@ static void rtas_set_xive(PowerPCCPU *cpu, 
> >> sPAPRMachineState *spapr,
> >>      }
> >>  
> >>      nr = rtas_ld(args, 0);
> >> -    server = xics_get_cpu_index_by_dt_id(rtas_ld(args, 1));
> >> +    server = rtas_ld(args, 1);
> >>      priority = rtas_ld(args, 2);
> >>  
> >>      if (!ics_valid_irq(ics, nr) || !xics_icp_get(XICS_FABRIC(spapr), 
> >> server)
> >> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> >> index 8aecea3dd10c..b9f7f8607869 100644
> >> --- a/hw/ppc/spapr.c
> >> +++ b/hw/ppc/spapr.c
> >> @@ -3024,9 +3024,10 @@ static void spapr_ics_resend(XICSFabric *dev)
> >>      ics_resend(spapr->ics);
> >>  }
> >>  
> >> -static ICPState *spapr_icp_get(XICSFabric *xi, int server)
> >> +static ICPState *spapr_icp_get(XICSFabric *xi, int cpu_dt_id)
> >>  {
> >>      sPAPRMachineState *spapr = SPAPR_MACHINE(xi);
> >> +    int server = xics_get_cpu_index_by_dt_id(cpu_dt_id);
> > 
> > The idea is good, but this is a bad name (as it was in the original
> > version, too).  The whole point here is that the XICS server number
> > (as it appears in the ICS registers) is the input to this function,
> > and we no longer assume that is equal to cpu_index.
> > 
> > Seems we could just get the cpu object by dt_id here, then go to 
> > ICP(cpu->intc).
> 
> yes that would be nice and this is exactly what pnv does now, but 
> this is only possible because the PnvICPState objects are allocated 
> from under PnvCore. This is not the case for sPAPR yet.
> 
> Currently, when the sPAPR core are realized, we call spapr_cpu_init() 
> which first gets its ICP with :
> 
>       xics_icp_get(xi, cpu->cpu_dt_id);
> 
> so we cannot use ICP(cpu->intc) in this XICSFabric handler, it is not
> assigned yet. It only will be at the end of spapr_cpu_init() when 
> 
>       xics_cpu_setup(xi, cpu, icp);
> 
> is called.

Uh.. but spapr_cpu_init() has the spapr pointer and can obviously get
the cpu_index.  So it could look up the right ICP in the array easily
enough instead of using xics_icp_get().

-- 
David Gibson                    | I'll have my music baroque, and my code
david AT gibson.dropbear.id.au  | minimalist, thank you.  NOT _the_ _other_
                                | _way_ _around_!
http://www.ozlabs.org/~dgibson

signature.asc
Description: PGP signature