[RFC] Add Arm cpu topology definition

Vincent Guittot vincent.guittot at linaro.org
Fri Jun 17 02:54:14 EDT 2011


On 16 June 2011 21:40, Stephen Boyd <sboyd at codeaurora.org> wrote:
> On 06/16/2011 01:49 AM, Vincent Guittot wrote:
>> +config SCHED_MC
>> +     bool "Multi-core scheduler support"
>> +     depends on SMP && ARM_CPU_TOPOLOGY
>> +     default n
>> +     help
>> +       Multi-core scheduler support improves the CPU scheduler's decision
>> +       making when dealing with multi-core CPU chips at a cost of slightly
>> +       increased overhead in some places. If unsure say N here.
>> +
>> +config SCHED_SMT
>> +     bool "SMT scheduler support"
>> +     depends on SMP && ARM_CPU_TOPOLOGY
>> +     default n
>> +     help
>> +       Improves the CPU scheduler's decision making when dealing with
>> +       MultiThreading at a cost of slightly increased overhead in some
>> +       places. If unsure say N here.
>> +
>
> The default is already n so you can drop those two lines.
>

ok

>> +              * This is a multiprocessor system
>> +              * multiprocessor format & multiprocessor mode field are set
>> +              */
>> +
>> +             if (mpidr & (0x1 << 24)) {
>> +                     /* core performance interdependency */
>> +                     cpuid_topo->thread_id = (mpidr & 0x3);
>> +                     cpuid_topo->core_id =  ((mpidr >> 8) & 0xF);
>> +                     cpuid_topo->socket_id = ((mpidr >> 16) & 0xFF);
>> +             } else {
>> +                     /* normal core interdependency */
>> +                     cpuid_topo->thread_id = -1;
>> +                     cpuid_topo->core_id = (mpidr & 0x3);
>> +                     cpuid_topo->socket_id = ((mpidr >> 8) & 0xF);
>> +             }
>> +
>
> The ARM ARM says these fields are IMPLEMENTATION DEFINED meaning that
> different vendors may attribute different meaning to these fields if
> they wish. Does that mean this should be a platform_*() function?
>

The ARM ARM also provides a recommended use of the fields of this
register and the TRM of each Cortex adds some details. On the cortex
A9, each platform can only set the value of the Cluster ID with the
CLUSTERID pins. I have tried to consolidate the value of MPIDR  across
several platforms and they all match with the description.

Have you got an example of a MPIDR register which doesn't match with
the implementation ?

> --
> Sent by an employee of the Qualcomm Innovation Center, Inc.
> The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.
>
>



More information about the linux-arm-kernel mailing list