Re: [PULL 06/28] spapr: Set VSMT to smp_threads by default

[Laurent Vivier <lvivier@redhat.com>]

On 08/11/2019 15:26, David Gibson wrote:
> On Fri, Nov 08, 2019 at 02:11:03PM +0100, Laurent Vivier wrote:
>> On 24/10/2019 10:17, David Gibson wrote:
>>> From: Greg Kurz <groug@kaod.org>
>>>
>>> Support for setting VSMT is available in KVM since linux-4.13. Most distros
>>> that support KVM on POWER already have it. It thus seem reasonable enough
>>> to have the default machine to set VSMT to smp_threads.
>>>
>>> This brings contiguous VCPU ids and thus brings their upper bound down to
>>> the machine's max_cpus. This is especially useful for XIVE KVM devices,
>>> which may thus allocate only one VP descriptor per VCPU.
>>>
>>> Signed-off-by: Greg Kurz <groug@kaod.org>
>>> Message-Id: <157010411885.246126.12610015369068227139.stgit@bahia.lan>
>>> Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
>>> ---
>>>  hw/ppc/spapr.c         | 7 ++++++-
>>>  include/hw/ppc/spapr.h | 1 +
>>>  2 files changed, 7 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
>>> index 4eb97d3a9b..428b834f30 100644
>>> --- a/hw/ppc/spapr.c
>>> +++ b/hw/ppc/spapr.c
>>> @@ -2496,6 +2496,7 @@ static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
>>>  static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
>>>  {
>>>      MachineState *ms = MACHINE(spapr);
>>> +    SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
>>>      Error *local_err = NULL;
>>>      bool vsmt_user = !!spapr->vsmt;
>>>      int kvm_smt = kvmppc_smt_threads();
>>> @@ -2522,7 +2523,7 @@ static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
>>>              goto out;
>>>          }
>>>          /* In this case, spapr->vsmt has been set by the command line */
>>> -    } else {
>>> +    } else if (!smc->smp_threads_vsmt) {
>>>          /*
>>>           * Default VSMT value is tricky, because we need it to be as
>>>           * consistent as possible (for migration), but this requires
>>> @@ -2531,6 +2532,8 @@ static void spapr_set_vsmt_mode(SpaprMachineState *spapr, Error **errp)
>>>           * overwhelmingly common case in production systems.
>>>           */
>>>          spapr->vsmt = MAX(8, smp_threads);
>>> +    } else {
>>> +        spapr->vsmt = smp_threads;
>>>      }
>>>  
>>>      /* KVM: If necessary, set the SMT mode: */
>>> @@ -4438,6 +4441,7 @@ static void spapr_machine_class_init(ObjectClass *oc, void *data)
>>>      smc->irq = &spapr_irq_dual;
>>>      smc->dr_phb_enabled = true;
>>>      smc->linux_pci_probe = true;
>>> +    smc->smp_threads_vsmt = true;
>>>  }
>>>  
>>>  static const TypeInfo spapr_machine_info = {
>>> @@ -4505,6 +4509,7 @@ static void spapr_machine_4_1_class_options(MachineClass *mc)
>>>  
>>>      spapr_machine_4_2_class_options(mc);
>>>      smc->linux_pci_probe = false;
>>> +    smc->smp_threads_vsmt = false;
>>>      compat_props_add(mc->compat_props, hw_compat_4_1, hw_compat_4_1_len);
>>>      compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
>>>  }
>>> diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
>>> index cbd1a4c9f3..2009eb64f9 100644
>>> --- a/include/hw/ppc/spapr.h
>>> +++ b/include/hw/ppc/spapr.h
>>> @@ -122,6 +122,7 @@ struct SpaprMachineClass {
>>>      bool broken_host_serial_model; /* present real host info to the guest */
>>>      bool pre_4_1_migration; /* don't migrate hpt-max-page-size */
>>>      bool linux_pci_probe;
>>> +    bool smp_threads_vsmt; /* set VSMT to smp_threads by default */
>>>  
>>>      void (*phb_placement)(SpaprMachineState *spapr, uint32_t index,
>>>                            uint64_t *buid, hwaddr *pio, 
>>>
>>
>> This patch breaks tests/migration-test on P8 host with kernel older than
>> 4.3 because it tries by default to set the VSMT to 1.
>>
>> qemu-system-ppc64: Failed to set KVM's VSMT mode to 1 (errno -22)
>> On PPC, a VM with 1 threads/core on a host with 8 threads/core requires
>> the use of VSMT mode 1.
>> This KVM seems to be too old to support VSMT.
>>
>> As this is clearly intentional, is there a way to fix migration-test?
> 
> Hrm.  I believe the argument for this was that the broken kernels were
> old enough we didn't care.  What platform are you testing on where
> you're hitting this?
> 

I'm going to propose a patch to fix this problem.
(it was on RHEL7)

Thanks,
Laurent