Re: [PATCH 1/2] add cpu_thread_id to struct cpuinfo_x86

[Andrew Cooper <andrew.cooper3@citrix.com>]

On 03/09/18 15:52, Juergen Gross wrote:
> On 03/09/18 16:47, Jan Beulich wrote:
>>>>> On 03.09.18 at 15:53, <jgross@suse.com> wrote:
>>> On 03/09/18 15:46, Jan Beulich wrote:
>>>>>>> On 31.08.18 at 18:22, <jgross@suse.com> wrote:
>>>>> Add the thread-id to the cpu config data and an accessor macro
>>>>> cpu_to_thread().
>>>>>
>>>>> Signed-off-by: Juergen Gross <jgross@suse.com>
>>>>> ---
>>>>>  xen/arch/x86/cpu/common.c       |  1 +
>>>>>  xen/arch/x86/smpboot.c          | 10 ++++++++++
>>>>>  xen/include/asm-arm/processor.h |  1 +
>>>>>  xen/include/asm-x86/processor.h |  2 ++
>>>>>  xen/include/public/sysctl.h     |  1 +
>>>>>  5 files changed, 15 insertions(+)
>>>>>
>>>>> diff --git a/xen/arch/x86/cpu/common.c b/xen/arch/x86/cpu/common.c
>>>>> index 057859ab14..f626a6a510 100644
>>>>> --- a/xen/arch/x86/cpu/common.c
>>>>> +++ b/xen/arch/x86/cpu/common.c
>>>>> @@ -434,6 +434,7 @@ void identify_cpu(struct cpuinfo_x86 *c)
>>>>>  	c->phys_proc_id = XEN_INVALID_SOCKET_ID;
>>>>>  	c->cpu_core_id = XEN_INVALID_CORE_ID;
>>>>>  	c->compute_unit_id = INVALID_CUID;
>>>>> +	c->cpu_thread_id = XEN_INVALID_THREAD_ID;
>>>>>  	memset(&c->x86_capability, 0, sizeof c->x86_capability);
>>>>>  
>>>>>  	generic_identify(c);
>>>>> diff --git a/xen/arch/x86/smpboot.c b/xen/arch/x86/smpboot.c
>>>>> index 7e76cc3d68..2c6a40c543 100644
>>>>> --- a/xen/arch/x86/smpboot.c
>>>>> +++ b/xen/arch/x86/smpboot.c
>>>>> @@ -247,6 +247,8 @@ static void set_cpu_sibling_map(unsigned int cpu)
>>>>>      cpumask_set_cpu(cpu, per_cpu(cpu_core_mask, cpu));
>>>>>      cpumask_set_cpu(cpu, per_cpu(cpu_sibling_mask, cpu));
>>>>>  
>>>>> +    c[cpu].cpu_thread_id = 0;
>>>>> +
>>>>>      if ( c[cpu].x86_num_siblings > 1 )
>>>>>      {
>>>>>          for_each_cpu ( i, &cpu_sibling_setup_map )
>>>>> @@ -270,6 +272,14 @@ static void set_cpu_sibling_map(unsigned int cpu)
>>>>>                         "CPU%u: unclear relationship with CPU%u\n",
>>>>>                         cpu, i);
>>>>>          }
>>>>> +
>>>>> +        for_each_cpu ( i, per_cpu(cpu_sibling_mask, cpu) )
>>>>> +        {
>>>>> +            if ( cpu == i )
>>>>> +                break;
>>>>> +            c[cpu].cpu_thread_id++;
>>>> No. This wants doing alongside setting of the other similar fields,
>>>> i.e. in (at least) detect_extended_topology() and detect_ht(). It
>>>> is the hardware which tells us what the thread ID is.
>>> I wasn't able to find the related information for AMD cpus. In case it
>>> is in the specs it is hidden very well.
>> What was used for compute units in Fam16 and earlier is used for
>> hyperthreads in Fam17 (and going forward, from what iirc Brian
>> has said).
> And what about hyperthreads in Fam16 and earlier? Or are "compute units"
> just another name for the threads and I can rename that field?

Fam 15/16 "Compute Units" are complicated.

The have a shared L1 I$, fetch and branch prediction, separate integer
pipelines, but a shared FPU pipeline.

Half of the pipeline looks very much like threads, and half looks like
separate cores.

~Andrew

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel