Re: [Qemu-devel] [Qemu-ppc] [PATCH 1/1] ppc: spapr: Move VCPU ID calculation into sPAPR

[Greg Kurz <groug@kaod.org>]

[-- Attachment #1: Type: text/plain, Size: 5215 bytes --]

On Tue, 5 Sep 2017 09:31:47 +1000
Sam Bobroff <sam.bobroff@au1.ibm.com> wrote:

> On Mon, Sep 04, 2017 at 12:18:57PM +0200, Greg Kurz wrote:
> > On Thu, 31 Aug 2017 16:38:46 +1000
> > Sam Bobroff <sam.bobroff@au1.ibm.com> wrote:
> >   
> > > Move the calculation of a CPU's VCPU ID out of the generic PPC code
> > > (ppc_cpu_realizefn()) and into sPAPR specific code
> > > (spapr_cpu_core_realize()) where it belongs.
> > > 
> > > Unfortunately, due to the way things are ordered, we still need to
> > > default the VCPU ID in ppc_cpu_realizfn() but at least doing that
> > > doesn't require any interaction with sPAPR.
> > > 
> > > Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
> > > ---
> > > This is follow up work arising from my work to clean up the way CPU VCPU IDs are
> > > handled on PowerPC. It had looked like it would be difficult to move the actual
> > > VCPU ID calculation out of generic code but it turned out to be OK.
> > > 
> > > It's based on dgibson/ppc-for-2.11.
> > > 
> > > Cheers,
> > > Sam.
> > > 
> > >  hw/ppc/spapr_cpu_core.c     | 11 +++++++++++
> > >  target/ppc/translate_init.c | 18 +++---------------
> > >  2 files changed, 14 insertions(+), 15 deletions(-)
> > > 
> > > diff --git a/hw/ppc/spapr_cpu_core.c b/hw/ppc/spapr_cpu_core.c
> > > index 5e319d9bbb..84dcc6e264 100644
> > > --- a/hw/ppc/spapr_cpu_core.c
> > > +++ b/hw/ppc/spapr_cpu_core.c
> > > @@ -211,6 +211,7 @@ error:
> > >  
> > >  static void spapr_cpu_core_realize(DeviceState *dev, Error **errp)
> > >  {
> > > +    sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
> > >      sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
> > >      sPAPRCPUCoreClass *scc = SPAPR_CPU_CORE_GET_CLASS(OBJECT(dev));
> > >      CPUCore *cc = CPU_CORE(OBJECT(dev));
> > > @@ -237,6 +238,16 @@ static void spapr_cpu_core_realize(DeviceState *dev, Error **errp)
> > >          cs = CPU(obj);
> > >          cpu = POWERPC_CPU(cs);
> > >          cs->cpu_index = cc->core_id + i;
> > > +        cpu->vcpu_id = (cc->core_id * spapr->vsmt / smp_threads) + i;  
> > 
> > According to what we currently have in the generic code, this should be:
> > 
> > ((cc->core_id + i) / smp_threads) * spapr->vsmt + ((cc->core_id + i) % smp_threads)
> > 
> > but since cc->core_id is a multiple of smp_threads and i < cc->nr_threads <= smp_threads,
> > then we can indeed simplify the computation as you did. :)
> > 
> > \o/
> > 
> > Reviewed-by: Greg Kurz <groug@kaod.org>  
> 
> Thanks for the review, and the nice simplification was David's idea :-)
> 

Yeah, David said so earlier... I probably missed the mail with the
suggestion while I was on vacation. :)

> > > +        if (kvm_enabled() && !kvm_vcpu_id_is_valid(cpu->vcpu_id)) {
> > > +            error_setg(&local_err, "Can't create CPU with id %d in KVM",
> > > +                       cpu->vcpu_id);
> > > +            error_append_hint(&local_err, "Adjust the number of cpus to %d "
> > > +                              "or try to raise the number of threads per core\n",
> > > +                              cpu->vcpu_id * smp_threads / spapr->vsmt);
> > > +            goto err;
> > > +        }
> > > +
> > >  
> > >          /* Set NUMA node for the threads belonged to core  */
> > >          cpu->node_id = sc->node_id;
> > > diff --git a/target/ppc/translate_init.c b/target/ppc/translate_init.c
> > > index 7f6a349e43..1f7286c893 100644
> > > --- a/target/ppc/translate_init.c
> > > +++ b/target/ppc/translate_init.c
> > > @@ -9903,28 +9903,15 @@ static void ppc_cpu_realizefn(DeviceState *dev, Error **errp)
> > >      PowerPCCPU *cpu = POWERPC_CPU(dev);
> > >      PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
> > >      Error *local_err = NULL;
> > > -#if !defined(CONFIG_USER_ONLY)
> > > -    int max_smt = kvmppc_smt_threads();
> > > -#endif
> > >  
> > >      cpu_exec_realizefn(cs, &local_err);
> > >      if (local_err != NULL) {
> > >          error_propagate(errp, local_err);
> > >          return;
> > >      }
> > > -
> > > -#if !defined(CONFIG_USER_ONLY)
> > > -    cpu->vcpu_id = (cs->cpu_index / smp_threads) * max_smt
> > > -        + (cs->cpu_index % smp_threads);
> > > -
> > > -    if (kvm_enabled() && !kvm_vcpu_id_is_valid(cpu->vcpu_id)) {
> > > -        error_setg(errp, "Can't create CPU with id %d in KVM", cpu->vcpu_id);
> > > -        error_append_hint(errp, "Adjust the number of cpus to %d "
> > > -                          "or try to raise the number of threads per core\n",
> > > -                          cpu->vcpu_id * smp_threads / max_smt);
> > > -        goto unrealize;
> > > +    if (cpu->vcpu_id == UNASSIGNED_CPU_INDEX) {
> > > +        cpu->vcpu_id = cs->cpu_index;
> > >      }
> > > -#endif
> > >  
> > >      if (tcg_enabled()) {
> > >          if (ppc_fixup_cpu(cpu) != 0) {
> > > @@ -10625,6 +10612,7 @@ static void ppc_cpu_initfn(Object *obj)
> > >      CPUPPCState *env = &cpu->env;
> > >  
> > >      cs->env_ptr = env;
> > > +    cpu->vcpu_id = UNASSIGNED_CPU_INDEX;
> > >  
> > >      env->msr_mask = pcc->msr_mask;
> > >      env->mmu_model = pcc->mmu_model;  
> >   
> 
> 


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 195 bytes --]