Re: [Qemu-ppc] [RFC PATCH v0 4/6] spapr: CPU hotplug support

[Bharata B Rao]

On Fri, Feb 26, 2016 at 02:51:41PM +1100, David Gibson wrote:
> On Thu, Feb 25, 2016 at 09:52:40PM +0530, Bharata B Rao wrote:
> > Set up device tree entries for the hotplugged CPU core and use the
> > exising EPOW event infrastructure to send CPU hotplug notification to
> > the guest.
> > 
> > Signed-off-by: Bharata B Rao <address@hidden>
> > ---
> >  hw/ppc/spapr.c         | 136 
> > ++++++++++++++++++++++++++++++++++++++++++++++++-
> >  hw/ppc/spapr_events.c  |   3 ++
> >  hw/ppc/spapr_rtas.c    |  24 +++++++++
> >  include/hw/ppc/spapr.h |   1 +
> >  4 files changed, 163 insertions(+), 1 deletion(-)
> > 
> > diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> > index 1f0d232..780cd00 100644
> > --- a/hw/ppc/spapr.c
> > +++ b/hw/ppc/spapr.c
> > @@ -603,6 +603,18 @@ static void spapr_populate_cpu_dt(CPUState *cs, void 
> > *fdt, int offset,
> >      size_t page_sizes_prop_size;
> >      uint32_t vcpus_per_socket = smp_threads * smp_cores;
> >      uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
> > +    sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
> > +    sPAPRDRConnector *drc;
> > +    sPAPRDRConnectorClass *drck;
> > +    int drc_index;
> > +
> > +    if (smc->dr_cpu_enabled) {
> > +        drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
> > +        g_assert(drc);
> > +        drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
> > +        drc_index = drck->get_index(drc);
> > +        _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", 
> > drc_index)));
> > +    }
> >  
> >      /* Note: we keep CI large pages off for now because a 64K capable guest
> >       * provisioned with large pages might otherwise try to map a qemu
> > @@ -987,6 +999,16 @@ static void spapr_finalize_fdt(sPAPRMachineState 
> > *spapr,
> >          _FDT(spapr_drc_populate_dt(fdt, 0, NULL, 
> > SPAPR_DR_CONNECTOR_TYPE_LMB));
> >      }
> >  
> > +    if (smc->dr_cpu_enabled) {
> > +        int offset = fdt_path_offset(fdt, "/cpus");
> > +        ret = spapr_drc_populate_dt(fdt, offset, NULL,
> > +                                    SPAPR_DR_CONNECTOR_TYPE_CPU);
> > +        if (ret < 0) {
> > +            fprintf(stderr, "Couldn't set up CPU DR device tree 
> > properties\n");
> > +            exit(1);
> > +        }
> > +    }
> > +
> >      _FDT((fdt_pack(fdt)));
> >  
> >      if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
> > @@ -1759,6 +1781,7 @@ static void ppc_spapr_init(MachineState *machine)
> >      char *filename;
> >      int spapr_cores = smp_cpus / smp_threads;
> >      int spapr_max_cores = max_cpus / smp_threads;
> > +    int smt = kvmppc_smt_threads();
> >  
> >      msi_supported = true;
> >  
> > @@ -1813,6 +1836,15 @@ static void ppc_spapr_init(MachineState *machine)
> >          spapr_validate_node_memory(machine, &error_fatal);
> >      }
> >  
> > +    if (smc->dr_cpu_enabled) {
> > +        for (i = 0; i < spapr_max_cores; i++) {
> > +            sPAPRDRConnector *drc =
> > +                spapr_dr_connector_new(OBJECT(spapr),
> > +                                       SPAPR_DR_CONNECTOR_TYPE_CPU, i * 
> > smt);
> > +            qemu_register_reset(spapr_drc_reset, drc);
> > +        }
> > +    }
> > +
> >      /* init CPUs */
> >      if (machine->cpu_model == NULL) {
> >          machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
> > @@ -2247,6 +2279,88 @@ out:
> >      error_propagate(errp, local_err);
> >  }
> >  
> > +static void *spapr_populate_hotplug_cpu_dt(DeviceState *dev, CPUState *cs,
> > +                                           int *fdt_offset,
> > +                                           sPAPRMachineState *spapr)
> > +{
> > +    PowerPCCPU *cpu = POWERPC_CPU(cs);
> > +    DeviceClass *dc = DEVICE_GET_CLASS(cs);
> > +    int id = ppc_get_vcpu_dt_id(cpu);
> > +    void *fdt;
> > +    int offset, fdt_size;
> > +    char *nodename;
> > +
> > +    fdt = create_device_tree(&fdt_size);
> > +    nodename = g_strdup_printf("address@hidden", dc->fw_name, id);
> > +    offset = fdt_add_subnode(fdt, 0, nodename);
> > +
> > +    spapr_populate_cpu_dt(cs, fdt, offset, spapr);
> > +    g_free(nodename);
> > +
> > +    *fdt_offset = offset;
> > +    return fdt;
> > +}
> > +
> > +static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
> > +                            Error **errp)
> > +{
> > +    sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
> > +    sPAPRMachineState *ms = SPAPR_MACHINE(qdev_get_machine());
> > +    sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
> > +    PowerPCCPU *cpu = &core->threads[0];
> > +    CPUState *cs = CPU(cpu);
> > +    int id = ppc_get_vcpu_dt_id(cpu);
> > +    sPAPRDRConnector *drc =
> > +        spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, id);
> > +    sPAPRDRConnectorClass *drck;
> > +    Error *local_err = NULL;
> > +    void *fdt = NULL;
> > +    int fdt_offset = 0;
> > +
> > +    if (!smc->dr_cpu_enabled) {
> > +        /*
> > +         * This is a cold plugged CPU core but the machine doesn't support
> > +         * DR. So skip the hotplug path ensuring that the core is brought
> > +         * up online with out an associated DR connector.
> > +         */
> > +        return;
> > +    }
> > +
> > +    g_assert(drc);
> > +
> > +    /*
> > +     * Setup CPU DT entries only for hotplugged CPUs. For boot time or
> > +     * coldplugged CPUs DT entries are setup in spapr_finalize_fdt().
> > +     */
> > +    if (dev->hotplugged) {
> > +        fdt = spapr_populate_hotplug_cpu_dt(dev, cs, &fdt_offset, ms);
> > +        dev->hotplugged = true;
> > +    }
> > +
> > +    drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
> > +    drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
> > +    if (local_err) {
> > +        g_free(fdt);
> > +        error_propagate(errp, local_err);
> > +        return;
> > +    }
> > +
> > +    if (dev->hotplugged) {
> > +        /*
> > +         * Send hotplug notification interrupt to the guest only in case
> > +         * of hotplugged CPUs.
> > +         */
> > +        spapr_hotplug_req_add_by_index(drc);
> > +    } else {
> > +        /*
> > +         * Set the right DRC states for cold plugged CPU.
> > +         */
> > +        drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
> > +        drck->set_isolation_state(drc, 
> > SPAPR_DR_ISOLATION_STATE_UNISOLATED);
> > +    }
> > +    return;
> > +}
> > +
> >  static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
> >                                        DeviceState *dev, Error **errp)
> >  {
> > @@ -2291,8 +2405,25 @@ static void spapr_machine_device_plug(HotplugHandler 
> > *hotplug_dev,
> >      } else if (object_dynamic_cast(OBJECT(dev), TYPE_CPU)) {
> >          CPUState *cs = CPU(dev);
> >          PowerPCCPU *cpu = POWERPC_CPU(cs);
> > +        int i;
> > +
> > +        if (!smc->dr_cpu_enabled && dev->hotplugged) {
> > +            error_setg(errp, "CPU hotplug not supported for this machine");
> > +            return;
> > +        }
> > +
> > +        /* Set NUMA node for the added CPUs  */
> > +        for (i = 0; i < nb_numa_nodes; i++) {
> > +            if (test_bit(cs->cpu_index, numa_info[i].node_cpu)) {
> > +                cs->numa_node = i;
> > +                break;
> > +            }
> > +        }
> >  
> >          spapr_cpu_init(ms, cpu, errp);
> > +        spapr_cpu_reset(cpu);
> 
> It looks to me like all of this setup should be either in the last
> patch, or in this one, not split between them.

The way I have done the functional split is like this:

- Introduce the spapr-cpu-core device as the 2nd patch in the series
- The previous patch converts the boot CPU initialization into
  spapr-cpu-core creation. So until this point, we can just boot a guest
  with this CPUs created as core devices.
- Then this patch adds the bits needed to actually perform  a device_add
  operation of core device.

But I am open to re-arrange things if you feel it is necessary.

> 
> > +    } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
> > +        spapr_core_plug(hotplug_dev, dev, errp);
> >      }
> >  }
> >  
> > @@ -2308,7 +2439,8 @@ static HotplugHandler 
> > *spapr_get_hotpug_handler(MachineState *machine,
> >                                               DeviceState *dev)
> >  {
> >      if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
> > -        object_dynamic_cast(OBJECT(dev), TYPE_CPU)) {
> > +        object_dynamic_cast(OBJECT(dev), TYPE_CPU) ||
> > +        object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
> >          return HOTPLUG_HANDLER(machine);
> >      }
> >      return NULL;
> > @@ -2352,6 +2484,7 @@ static void spapr_machine_class_init(ObjectClass *oc, 
> > void *data)
> >      mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
> >  
> >      smc->dr_lmb_enabled = true;
> > +    smc->dr_cpu_enabled = true;
> >      fwc->get_dev_path = spapr_get_fw_dev_path;
> >      nc->nmi_monitor_handler = spapr_nmi;
> >  }
> > @@ -2431,6 +2564,7 @@ static void 
> > spapr_machine_2_5_class_options(MachineClass *mc)
> >  
> >      spapr_machine_2_6_class_options(mc);
> >      smc->use_ohci_by_default = true;
> > +    smc->dr_cpu_enabled = false;
> >      SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
> >  }
> >  
> > diff --git a/hw/ppc/spapr_events.c b/hw/ppc/spapr_events.c
> > index f5eac4b..e50bb16 100644
> > --- a/hw/ppc/spapr_events.c
> > +++ b/hw/ppc/spapr_events.c
> > @@ -437,6 +437,9 @@ static void spapr_hotplug_req_event(uint8_t hp_id, 
> > uint8_t hp_action,
> >      case SPAPR_DR_CONNECTOR_TYPE_LMB:
> >          hp->hotplug_type = RTAS_LOG_V6_HP_TYPE_MEMORY;
> >          break;
> > +    case SPAPR_DR_CONNECTOR_TYPE_CPU:
> > +        hp->hotplug_type = RTAS_LOG_V6_HP_TYPE_CPU;
> > +        break;
> >      default:
> >          /* we shouldn't be signaling hotplug events for resources
> >           * that don't support them
> > diff --git a/hw/ppc/spapr_rtas.c b/hw/ppc/spapr_rtas.c
> > index b7c5ebd..cc0369e 100644
> > --- a/hw/ppc/spapr_rtas.c
> > +++ b/hw/ppc/spapr_rtas.c
> > @@ -34,6 +34,7 @@
> >  
> >  #include "hw/ppc/spapr.h"
> >  #include "hw/ppc/spapr_vio.h"
> > +#include "hw/ppc/ppc.h"
> >  #include "qapi-event.h"
> >  #include "hw/boards.h"
> >  
> > @@ -161,6 +162,27 @@ static void rtas_query_cpu_stopped_state(PowerPCCPU 
> > *cpu_,
> >      rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR);
> >  }
> >  
> > +/*
> > + * Set the timebase offset of the CPU to that of first CPU.
> > + * This helps hotplugged CPU to have the correct timebase offset.
> > + */
> > +static void spapr_cpu_update_tb_offset(PowerPCCPU *cpu)
> > +{
> > +    PowerPCCPU *fcpu = POWERPC_CPU(first_cpu);
> > +
> > +    cpu->env.tb_env->tb_offset = fcpu->env.tb_env->tb_offset;
> > +}
> > +
> > +static void spapr_cpu_set_endianness(PowerPCCPU *cpu)
> > +{
> > +    PowerPCCPU *fcpu = POWERPC_CPU(first_cpu);
> > +    PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(fcpu);
> > +
> > +    if (!pcc->interrupts_big_endian(fcpu)) {
> > +        cpu->env.spr[SPR_LPCR] |= LPCR_ILE;
> > +    }
> > +}
> > +
> 
> Any particular reason for doing these things at rtas_start_cpu() time,
> but other initialization at plug time?  Could you consolidate it to
> one place or the other?

Those board specific things that are needed to be done have been consolidated
into spapr_cpu_init() which will be called from the plug handler. We have
discussed this earlier at:

https://lists.nongnu.org/archive/html/qemu-devel/2015-02/msg04399.html

It has been a while but there was a good reason why setting endianness
here rather than in plug handler is necessary. W/o this LE hotplug on guests
wouldn't work, I will dig up and come back on what exactly necessiated
this change.

Regards,
Bharata.