Re: [PATCH for-6.2 v2 07/11] machine: Prefer cores over sockets in smp parsing since 6.2

["wangyanan (Y)" <wangyanan55@huawei.com>]

On 2021/7/20 1:13, Andrew Jones wrote:
> On Mon, Jul 19, 2021 at 11:20:39AM +0800, Yanan Wang wrote:
>> In the real SMP hardware topology world, it's much more likely that
>> we have high cores-per-socket counts and few sockets totally. While
>> the current preference of sockets over cores in smp parsing results
>> in a virtual cpu topology with low cores-per-sockets counts and a
>> large number of sockets, which is just contrary to the real world.
>>
>> Given that it is better to make the virtual cpu topology be more
>> reflective of the real world and also for the sake of compatibility,
>> we start to prefer cores over sockets over threads in smp parsing
>> since machine type 6.2 for different arches.
>>
>> In this patch, a boolean "smp_prefer_sockets" is added, and we only
>> enable the old preference on older machines and enable the new one
>> since type 6.2 for all arches by using the machine compat mechanism.
>>
>> Suggested-by: Daniel P. Berrange <berrange@redhat.com>
>> Signed-off-by: Yanan Wang <wangyanan55@huawei.com>
>> ---
>>   hw/arm/virt.c              |  1 +
>>   hw/core/machine.c          | 59 +++++++++++++++++++++++++++++---------
>>   hw/i386/pc_piix.c          |  1 +
>>   hw/i386/pc_q35.c           |  1 +
>>   hw/ppc/spapr.c             |  1 +
>>   hw/s390x/s390-virtio-ccw.c |  1 +
>>   include/hw/boards.h        |  1 +
>>   qemu-options.hx            |  4 ++-
>>   8 files changed, 55 insertions(+), 14 deletions(-)
>>
>> diff --git a/hw/arm/virt.c b/hw/arm/virt.c
>> index 01165f7f53..7babea40dc 100644
>> --- a/hw/arm/virt.c
>> +++ b/hw/arm/virt.c
>> @@ -2797,6 +2797,7 @@ static void virt_machine_6_1_options(MachineClass *mc)
>>   {
>>       virt_machine_6_2_options(mc);
>>       compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
>> +    mc->smp_prefer_sockets = true;
>>   }
>>   DEFINE_VIRT_MACHINE(6, 1)
>>   
>> diff --git a/hw/core/machine.c b/hw/core/machine.c
>> index 63439c4a6d..c074425015 100644
>> --- a/hw/core/machine.c
>> +++ b/hw/core/machine.c
>> @@ -744,6 +744,22 @@ void machine_set_cpu_numa_node(MachineState *machine,
>>       }
>>   }
>>   
>> +/*
>> + * smp_parse - Generic function used to parse the given SMP configuration
>> + *
>> + * The topology parameters must be specified equal to or great than one
>> + * or just omitted, explicit configuration like "cpus=0" is not allowed.
>> + * The omitted parameters will be calculated based on the provided ones.
>> + *
>> + * maxcpus will default to the value of cpus if omitted and will be used
>> + * to compute the missing sockets/cores/threads. cpus will be calculated
>> + * from the computed parametrs if omitted.
> parameters
>
> Or how about something like this:
>
> When both maxcpus and cpus are omitted maxcpus will be calculated from the
> given parameters and cpus will be set equal to maxcpus. When only one of
> maxcpus and cpus is given then the omitted one will be set to its given
> counterpart's value. Both maxcpus and cpus may be specified, but cpus must
> be less than or equal to maxcpus.
Yes, this is much clearer.
Now I understand why you suggested in patch #4 to change the format from
   cpus = cpus > 0 ? cpus : sockets * dies * cores * threads;
   maxcpus = maxcpus > 0 ? maxcpus : cpus;
to
   maxcpus = maxcpus > 0 ? maxcpus : sockets * dies * cores * threads;
   cpus = cpus > 0 ? cpus : maxcpus;

What you suggest seems more reasonable according to above doc,
I will adjust the related part as you suggested.
>> + *
>> + * In calculation of omitted arch-netural sockets/cores/threads, we prefer
>> + * sockets over cores over threads before 6.2, while prefer cores over
> while preferring
>
>> + * sockets over threads since 6.2 on. The arch-specific dies will directly
> s/on//
Right.
>> + * default to 1 if omitted.
>> + */
>>   static void smp_parse(MachineState *ms, SMPConfiguration *config, Error **errp)
>>   {
>>       MachineClass *mc = MACHINE_GET_CLASS(ms);
>> @@ -772,19 +788,36 @@ static void smp_parse(MachineState *ms, SMPConfiguration *config, Error **errp)
>>   
>>       maxcpus = maxcpus > 0 ? maxcpus : cpus;
>>   
>> -    /* compute missing values, prefer sockets over cores over threads */
>> -    if (sockets == 0) {
>> -        cores = cores > 0 ? cores : 1;
>> -        threads = threads > 0 ? threads : 1;
>> -        sockets = maxcpus / (dies * cores * threads);
>> -        sockets = sockets > 0 ? sockets : 1;
>> -    } else if (cores == 0) {
>> -        threads = threads > 0 ? threads : 1;
>> -        cores = maxcpus / (sockets * dies * threads);
>> -        cores = cores > 0 ? cores : 1;
>> -    } else if (threads == 0) {
>> -        threads = maxcpus / (sockets * dies * cores);
>> -        threads = threads > 0 ? threads : 1;
>> +    /* prefer sockets over cores over threads before 6.2 */
>> +    if (mc->smp_prefer_sockets) {
> please move the comment into the if, so...
>
>> +        if (sockets == 0) {
>> +            cores = cores > 0 ? cores : 1;
>> +            threads = threads > 0 ? threads : 1;
>> +            sockets = maxcpus / (dies * cores * threads);
>> +            sockets = sockets > 0 ? sockets : 1;
>> +        } else if (cores == 0) {
>> +            threads = threads > 0 ? threads : 1;
>> +            cores = maxcpus / (sockets * dies * threads);
>> +            cores = cores > 0 ? cores : 1;
>> +        } else if (threads == 0) {
>> +            threads = maxcpus / (sockets * dies * cores);
>> +            threads = threads > 0 ? threads : 1;
>> +        }
>> +    /* prefer cores over sockets over threads since 6.2 */
> ...here we can put the comment in the else
Ok
>> +    } else {
>> +        if (cores == 0) {
>> +            sockets = sockets > 0 ? sockets : 1;
>> +            threads = threads > 0 ? threads : 1;
>> +            cores = maxcpus / (sockets * dies * threads);
>> +            cores = cores > 0 ? cores : 1;
>> +        } else if (sockets == 0) {
>> +            threads = threads > 0 ? threads : 1;
>> +            sockets = maxcpus / (dies * cores * threads);
>> +            sockets = sockets > 0 ? sockets : 1;
>> +        } else if (threads == 0) {
>> +            threads = maxcpus / (sockets * dies * cores);
>> +            threads = threads > 0 ? threads : 1;
>> +        }
>>       }
>>   
>>       /* use the computed parameters to calculate the omitted cpus */
>> diff --git a/hw/i386/pc_piix.c b/hw/i386/pc_piix.c
>> index fd5c2277f2..9b811fc6ca 100644
>> --- a/hw/i386/pc_piix.c
>> +++ b/hw/i386/pc_piix.c
>> @@ -432,6 +432,7 @@ static void pc_i440fx_6_1_machine_options(MachineClass *m)
>>       m->is_default = false;
>>       compat_props_add(m->compat_props, hw_compat_6_1, hw_compat_6_1_len);
>>       compat_props_add(m->compat_props, pc_compat_6_1, pc_compat_6_1_len);
>> +    m->smp_prefer_sockets = true;
>>   }
>>   
>>   DEFINE_I440FX_MACHINE(v6_1, "pc-i440fx-6.1", NULL,
>> diff --git a/hw/i386/pc_q35.c b/hw/i386/pc_q35.c
>> index b45903b15e..88efb7fde4 100644
>> --- a/hw/i386/pc_q35.c
>> +++ b/hw/i386/pc_q35.c
>> @@ -372,6 +372,7 @@ static void pc_q35_6_1_machine_options(MachineClass *m)
>>       m->alias = NULL;
>>       compat_props_add(m->compat_props, hw_compat_6_1, hw_compat_6_1_len);
>>       compat_props_add(m->compat_props, pc_compat_6_1, pc_compat_6_1_len);
>> +    m->smp_prefer_sockets = true;
>>   }
>>   
>>   DEFINE_Q35_MACHINE(v6_1, "pc-q35-6.1", NULL,
>> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
>> index d39fd4e644..a481fade51 100644
>> --- a/hw/ppc/spapr.c
>> +++ b/hw/ppc/spapr.c
>> @@ -4702,6 +4702,7 @@ static void spapr_machine_6_1_class_options(MachineClass *mc)
>>   {
>>       spapr_machine_6_2_class_options(mc);
>>       compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
>> +    mc->smp_prefer_sockets = true;
>>   }
>>   
>>   DEFINE_SPAPR_MACHINE(6_1, "6.1", false);
>> diff --git a/hw/s390x/s390-virtio-ccw.c b/hw/s390x/s390-virtio-ccw.c
>> index 4d25278cf2..b40e647883 100644
>> --- a/hw/s390x/s390-virtio-ccw.c
>> +++ b/hw/s390x/s390-virtio-ccw.c
>> @@ -809,6 +809,7 @@ static void ccw_machine_6_1_class_options(MachineClass *mc)
>>   {
>>       ccw_machine_6_2_class_options(mc);
>>       compat_props_add(mc->compat_props, hw_compat_6_1, hw_compat_6_1_len);
>> +    mc->smp_prefer_sockets = true;
>>   }
>>   DEFINE_CCW_MACHINE(6_1, "6.1", false);
>>   
>> diff --git a/include/hw/boards.h b/include/hw/boards.h
>> index 2832f0f8aa..8df885c9d2 100644
>> --- a/include/hw/boards.h
>> +++ b/include/hw/boards.h
>> @@ -247,6 +247,7 @@ struct MachineClass {
>>       bool nvdimm_supported;
>>       bool numa_mem_supported;
>>       bool smp_dies_supported;
>> +    bool smp_prefer_sockets;
>>       bool auto_enable_numa;
>>       const char *default_ram_id;
>>   
>> diff --git a/qemu-options.hx b/qemu-options.hx
>> index 0c9ddc0274..6ef57e838c 100644
>> --- a/qemu-options.hx
>> +++ b/qemu-options.hx
>> @@ -227,7 +227,9 @@ SRST
>>       from those which are given. Historically preference was given to the
>>       coarsest topology parameters when computing missing values (ie sockets
>>       preferred over cores, which were preferred over threads), however, this
>> -    behaviour is considered liable to change.
>> +    behaviour is considered liable to change. The historical preference of
>> +    sockets over cores over threads works before 6.2, and a new preference
>> +    of cores over sockets over threads starts to work since 6.2 on.
> Prior to 6.2 the preference was sockets over cores over threads. Since 6.2
> the preference is cores over sockets over threads.
It's more concise and clearer.

Thanks,
Yanan
.
>>   ERST
>>   
>>   DEF("numa", HAS_ARG, QEMU_OPTION_numa,
>> -- 
>> 2.19.1
>>
> Thanks,
> drew
>
> .