Re: [PATCH v9 10/10] powerpc/perf: Thread imc cpuhotplug support
From: Thomas Gleixner
Date: Tue Jun 06 2017 - 06:27:44 EST
On Mon, 5 Jun 2017, Anju T Sudhakar wrote:
> static void thread_imc_mem_alloc(int cpu_id)
> {
> - u64 ldbar_addr, ldbar_value;
> int phys_id = topology_physical_package_id(cpu_id);
>
> per_cpu_add[cpu_id] = (u64)alloc_pages_exact_nid(phys_id,
> (size_t)IMC_THREAD_COUNTER_MEM, GFP_KERNEL | __GFP_ZERO);
It took me a while to understand that per_cpu_add is an array and not a new
fangled per cpu function which adds something to a per cpu variable.
So why is that storing the address as u64?
And why is this a NR_CPUS sized array instead of a regular per cpu variable?
> +}
> +
> +static void thread_imc_update_ldbar(unsigned int cpu_id)
> +{
> + u64 ldbar_addr, ldbar_value;
> +
> + if (per_cpu_add[cpu_id] == 0)
> + thread_imc_mem_alloc(cpu_id);
So if that allocation fails then you happily proceed. Optmistic
programming, right?
> +
> ldbar_addr = (u64)virt_to_phys((void *)per_cpu_add[cpu_id]);
Ah, it's stored as u64 because you have to convert it back to a void
pointer at the place where it is actually used. Interesting approach.
> ldbar_value = (ldbar_addr & (u64)THREAD_IMC_LDBAR_MASK) |
> - (u64)THREAD_IMC_ENABLE;
> + (u64)THREAD_IMC_ENABLE;
> mtspr(SPRN_LDBAR, ldbar_value);
> }
>
> @@ -442,6 +450,26 @@ static void core_imc_change_cpu_context(int old_cpu, int new_cpu)
> perf_pmu_migrate_context(&core_imc_pmu->pmu, old_cpu, new_cpu);
> }
>
> +static int ppc_thread_imc_cpu_online(unsigned int cpu)
> +{
> + thread_imc_update_ldbar(cpu);
> + return 0;
> +}
> +
> +static int ppc_thread_imc_cpu_offline(unsigned int cpu)
> +{
> + mtspr(SPRN_LDBAR, 0);
> + return 0;
> +}
> +
> +void thread_imc_cpu_init(void)
> +{
> + cpuhp_setup_state(CPUHP_AP_PERF_POWERPC_THREAD_IMC_ONLINE,
> + "perf/powerpc/imc_thread:online",
> + ppc_thread_imc_cpu_online,
> + ppc_thread_imc_cpu_offline);
> +}
> +
> static int ppc_core_imc_cpu_online(unsigned int cpu)
> {
> const struct cpumask *l_cpumask;
> @@ -953,6 +981,9 @@ int __init init_imc_pmu(struct imc_events *events, int idx,
> if (ret)
> return ret;
> break;
> + case IMC_DOMAIN_THREAD:
> + thread_imc_cpu_init();
> + break;
> default:
> return -1; /* Unknown domain */
> }
Just a general observation. If anything in init_imc_pmu() fails, then all
the hotplug callbacks are staying registered, at least I haven't seen
anything undoing it.
Thanks,
tglx