[RFC] Add Arm cpu topology definition
Vincent Guittot
vincent.guittot at linaro.org
Thu Jun 16 08:05:45 EDT 2011
On 16 June 2011 12:49, Daniel Lezcano <daniel.lezcano at linaro.org> wrote:
> On 06/16/2011 10:49 AM, Vincent Guittot wrote:
>>
>> The affinity between Arm processors is defined in the MPIDR register.
>> We can identify which processors are in the same cluster,
>> and which ones have performance interdependency. The cpu topology
>> of an Arm platform can be set thanks to this register and this topology
>> is then used by sched_mc and sched_smt.
>>
>> Signed-off-by: Vincent Guittot<vincent.guittot at linaro.org>
>> ---
>> arch/arm/Kconfig | 26 ++++++++
>> arch/arm/include/asm/topology.h | 33 ++++++++++
>> arch/arm/kernel/Makefile | 1 +
>> arch/arm/kernel/smp.c | 6 ++
>> arch/arm/kernel/topology.c | 133
>> +++++++++++++++++++++++++++++++++++++++
>> 5 files changed, 199 insertions(+), 0 deletions(-)
>> create mode 100644 arch/arm/kernel/topology.c
>>
>> diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
>> index 9adc278..bacf9af 100644
>> --- a/arch/arm/Kconfig
>> +++ b/arch/arm/Kconfig
>> @@ -219,6 +219,24 @@ source "kernel/Kconfig.freezer"
>>
>> menu "System Type"
>>
>> +config SCHED_MC
>> + bool "Multi-core scheduler support"
>> + depends on SMP&& ARM_CPU_TOPOLOGY
>
> ARM_CPU_TOPOLOGY depends on SMP, so the check can be reduced to
>
> depends on ARM_CPU_TOPOLOGY
you're right
>>
>> + 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
>
> depends on SMT && ARM_CPU_TOPOLOGY ?
SMP is the right one but it can be reduced to : depends on ARM_CPU_TOPOLOGY
like SCHED_MC,
>
>> + 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.
>> +
>> config MMU
>> bool "MMU-based Paged Memory Management Support"
>> default y
>> @@ -1062,6 +1080,14 @@ if !MMU
>> source "arch/arm/Kconfig-nommu"
>> endif
>>
>> +config ARM_CPU_TOPOLOGY
>> + bool "Support cpu topology definition"
>> + depends on SMP&& CPU_V7
>> + help
>> + Support Arm cpu topology definition. The MPIDR register defines
>> + affinity between processors which is used to set the cpu
>> + topology of an Arm System.
>> +
>> config ARM_ERRATA_411920
>> bool "ARM errata: Invalidation of the Instruction Cache operation
>> can fail"
>> depends on CPU_V6 || CPU_V6K
>> diff --git a/arch/arm/include/asm/topology.h
>> b/arch/arm/include/asm/topology.h
>> index accbd7c..cb90d0a 100644
>> --- a/arch/arm/include/asm/topology.h
>> +++ b/arch/arm/include/asm/topology.h
>> @@ -1,6 +1,39 @@
>> #ifndef _ASM_ARM_TOPOLOGY_H
>> #define _ASM_ARM_TOPOLOGY_H
>>
>> +#ifdef CONFIG_ARM_CPU_TOPOLOGY
>> +
>> +#include<linux/cpumask.h>
>> +
>> +struct cputopo_arm {
>> + int thread_id;
>> + int core_id;
>> + int socket_id;
>
> I am not sure how that deals with the rest of the functions prototype but
> wouldn't u16 be more adequate ?
>
I have used int to be aligned on register size and minimize register
manipulation
>> + cpumask_t thread_sibling;
>> + cpumask_t core_sibling;
>> +};
>> +
>> +extern struct cputopo_arm cpu_topology[NR_CPUS];
>> +
>> +#define topology_physical_package_id(cpu)
>> (cpu_topology[cpu].socket_id)
>> +#define topology_core_id(cpu) (cpu_topology[cpu].core_id)
>> +#define topology_core_cpumask(cpu)
>> (&(cpu_topology[cpu].core_sibling))
>> +#define topology_thread_cpumask(cpu)
>> (&(cpu_topology[cpu].thread_sibling))
>> +
>> +#define mc_capable() (cpu_topology[0].socket_id != -1)
>> +#define smt_capable() (cpu_topology[0].thread_id != -1)
>> +
>> +void init_cpu_topology(void);
>> +void store_cpu_topology(unsigned int cpuid);
>> +const struct cpumask *cpu_coregroup_mask(unsigned int cpu);
>> +
>> +#else
>> +
>> +#define init_cpu_topology() {};
>> +#define store_cpu_topology(cpuid) {};
>
> AFAIK the convention is to declare static inline noop functions.
>
> static inline void init_cpu_topology(void) { };
> static inline void store_cpu_topology(unsigned int cpuid) { };
>
ok
>> +
>> +#endif
>> +
>> #include<asm-generic/topology.h>
>>
>> #endif /* _ASM_ARM_TOPOLOGY_H */
>> diff --git a/arch/arm/kernel/Makefile b/arch/arm/kernel/Makefile
>> index a5b31af..816a481 100644
>> --- a/arch/arm/kernel/Makefile
>> +++ b/arch/arm/kernel/Makefile
>> @@ -61,6 +61,7 @@ obj-$(CONFIG_IWMMXT) += iwmmxt.o
>> obj-$(CONFIG_CPU_HAS_PMU) += pmu.o
>> obj-$(CONFIG_HW_PERF_EVENTS) += perf_event.o
>> AFLAGS_iwmmxt.o := -Wa,-mcpu=iwmmxt
>> +obj-$(CONFIG_ARM_CPU_TOPOLOGY) += topology.o
>>
>> ifneq ($(CONFIG_ARCH_EBSA110),y)
>> obj-y += io.o
>> diff --git a/arch/arm/kernel/smp.c b/arch/arm/kernel/smp.c
>> index 344e52b..3e8dc3b 100644
>> --- a/arch/arm/kernel/smp.c
>> +++ b/arch/arm/kernel/smp.c
>> @@ -31,6 +31,7 @@
>> #include<asm/cacheflush.h>
>> #include<asm/cpu.h>
>> #include<asm/cputype.h>
>> +#include<asm/topology.h>
>> #include<asm/mmu_context.h>
>> #include<asm/pgtable.h>
>> #include<asm/pgalloc.h>
>> @@ -268,6 +269,9 @@ static void __cpuinit smp_store_cpu_info(unsigned int
>> cpuid)
>> struct cpuinfo_arm *cpu_info =&per_cpu(cpu_data, cpuid);
>>
>> cpu_info->loops_per_jiffy = loops_per_jiffy;
>> +
>> + store_cpu_topology(cpuid);
>> +
>> }
>
> If the store_cpu_topology function is called once, can it be changed to a
> __cpuinit function, declared as a subsys_initcall and removed from here ?
>
it must be called once on each cpu before the call of sched_init_smp
>> /*
>> @@ -354,6 +358,8 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
>> {
>> unsigned int ncores = num_possible_cpus();
>>
>> + init_cpu_topology();
>
> Why do you need to call the init function here ?
>
this function must be called before the 1st call to smp_store_cpu_info
> On the other architecture I see:
>
> static int __init topology_init(void)
> {
> ...
> }
>
> subsys_initcall(topology_init);
>
> Isn't possible to use the same way ? (with the benefit to save two
> declarations in the header).
>
>
> [ ... ]
>
>> +
>> +struct cputopo_arm cpu_topology[NR_CPUS];
>
> IMO, you can define it static here no ?
>
This array is used by "#define topology_xxx" and "#define xx_capable"
which are used by the scheduler and the topology driver
>> +
>> +const struct cpumask *cpu_coregroup_mask(unsigned int cpu)
>> +{
>> + return&(cpu_topology[cpu].core_sibling);
>> +}
>> +
>> +/*
>> + * store_cpu_topology is called at boot when only one cpu is running
>> + * and with the mutex cpu_hotplug.lock locked, when several cpus have
>> booted,
>> + * which prevents simultaneous write access to cpu_topology array
>> + */
>> +void store_cpu_topology(unsigned int cpuid)
>> +{
>> + struct cputopo_arm *cpuid_topo =&(cpu_topology[cpuid]);
>> + unsigned int mpidr;
>> + unsigned int cpu;
>> +
>> + /* If the cpu topology has been already set, just return */
>> + if (cpuid_topo->core_id != -1)
>> + return;
>
> If the code calls store_cpu_topology but with no effect because it was
> already called before, that means it shouldn't be called at all, no ?
> IMHO, this test should be removed or at least add a WARN_ONCE.
>
We will call smp_store_cpu_info each time the cpu will be plugged. But
once set, we don't need to update topology information
>> +
>> + mpidr = hard_smp_mpidr();
>> +
>> + /* create cpu topology mapping */
>> + if (mpidr& (0x3<< 30)) {
>> + /*
>> + * 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);
>> + }
>> + } else {
>> + /*
>> + * This is an uniprocessor system
>> + * we are in multiprocessor format but uniprocessor system
>> + * or in the old uniprocessor format
>> + */
>> +
>> + cpuid_topo->thread_id = -1;
>> + cpuid_topo->core_id = 0;
>> + cpuid_topo->socket_id = -1;
>> + }
>> +
>> + /* update core and thread sibling masks */
>> + for_each_possible_cpu(cpu) {
>> + struct cputopo_arm *cpu_topo =&(cpu_topology[cpu]);
>> +
>> + if (cpuid_topo->socket_id == cpu_topo->socket_id) {
>> + cpumask_set_cpu(cpuid,&cpu_topo->core_sibling);
>> + if (cpu != cpuid)
>> + cpumask_set_cpu(cpu,
>> + &cpuid_topo->core_sibling);
>> +
>> + if (cpuid_topo->core_id == cpu_topo->core_id) {
>> + cpumask_set_cpu(cpuid,
>> + &cpu_topo->thread_sibling);
>> + if (cpu != cpuid)
>> + cpumask_set_cpu(cpu,
>> +
>> &cpuid_topo->thread_sibling);
>> + }
>> + }
>> + }
>> + smp_wmb();
>> +
>> + printk(KERN_INFO "cpu %u : thread %d cpu %d, socket %d, mpidr
>> %x\n",
>> + cpuid, cpu_topology[cpuid].thread_id,
>> + cpu_topology[cpuid].core_id,
>> + cpu_topology[cpuid].socket_id, mpidr);
>> +
>> +}
>> +
>> +/*
>> + * init_cpu_topology is called at boot when only one cpu is running
>> + * which prevent simultaneous write access to cpu_topology array
>> + */
>> +void init_cpu_topology(void)
>> +{
>> + unsigned int cpu;
>> +
>> + /* init core mask */
>> + for_each_possible_cpu(cpu) {
>> + struct cputopo_arm *cpu_topo =&(cpu_topology[cpu]);
>> +
>> + cpu_topo->thread_id = -1;
>> + cpu_topo->core_id = -1;
>
> nit : extra space
ok
>>
>> + cpu_topo->socket_id = -1;
>> + cpumask_clear(&cpu_topo->core_sibling);
>> + cpumask_clear(&cpu_topo->thread_sibling);
>> + }
>> + smp_wmb();
>> +}
>
>
More information about the linux-arm-kernel
mailing list