[RFC 0/6] rework sched_domain topology description

[Vincent Guittot]

On 19 March 2014 20:15, Dietmar Eggemann <dietmar.eggemann at arm.com> wrote:
> On 17/03/14 11:52, Peter Zijlstra wrote:
>> On Wed, Mar 12, 2014 at 01:28:07PM +0000, Dietmar Eggemann wrote:
> [...]
>>> By making it robust, I guess you mean that the core scheduler has to
>>> check that the provided set-ups are sane, something like the following
>>> code snippet in sd_init()
>>>
>>> if (WARN_ONCE(tl->sd_flags & ~TOPOLOGY_SD_FLAGS,
>>>              "wrong sd_flags in topology description\n"))
>>>      tl->sd_flags &= ~TOPOLOGY_SD_FLAGS;
>>>
>>> but for per cpu set-up's.
>>
>> So a domain is principally a group of CPUs with the same properties.
>> However per-cpu domain attributes allows you to specify different domain
>> properties within the one domain mask.
>>
>> That's completely broken.
>>
>> So the way to validate something like that would be:
>>
>>       cpu = cpumask_first(tl->mask());
>>       flags = tl->flags(cpu);
>>
>>       for (;cpu = cpumask_next(cpu, tl->mask()), cpu < nr_cpu_ids;)
>>               BUG_ON(tl->flags(cpu) != flags);
>>
>> Or something along those lines.
>
> I tried this idea inside sd_init() on top of Vincent's V3 and it's doing
> its job.
>
>>
>> But for me its far easier to think in the simple one domain one flags
>> scenario. The whole degenerate folding is a very simple optimization
>> simply removing redundant levels.
>>
>
> For me, the approach with the 'int cpu' parameter in the flag function is
> easier to understand. One of the things I had to grasp though was the fact that
> we can only specify SD_SHARE_FOO flags and not SD_NOT_SHARE_FOO per domain.

Looking at you test below, the solution without cpu argument is more
readable for me because you don't have to handle 2 cpu args that can
vary when setting your level. I'm afraid that the flags function will
become quite complex and unreadable with a cpu arg. And this
additional cpu arg doesn't give any benefit.

Vincent

>
> -- >8 --
>
> Subject: [PATCH] sched: check that the sd_flags are consistent in one domain
>
> ---
>  arch/arm/kernel/topology.c |   13 +++++++++----
>  include/linux/sched.h      |    6 +++---
>  kernel/sched/core.c        |   11 +++++++++--
>  3 files changed, 21 insertions(+), 9 deletions(-)
>
> diff --git a/arch/arm/kernel/topology.c b/arch/arm/kernel/topology.c
> index 71e1fec6d31a..425f133c690d 100644
> --- a/arch/arm/kernel/topology.c
> +++ b/arch/arm/kernel/topology.c
> @@ -275,15 +275,20 @@ void store_cpu_topology(unsigned int cpuid)
>                 cpu_topology[cpuid].socket_id, mpidr);
>  }
>
> -static inline const int cpu_corepower_flags(void)
> +//static inline const int cpu_corepower_flags(void)
> +//{
> +//     return SD_SHARE_PKG_RESOURCES  | SD_SHARE_POWERDOMAIN;
> +//}
> +
> +static inline const int arm_cpu_core_flags(int cpu)
>  {
> -       return SD_SHARE_PKG_RESOURCES  | SD_SHARE_POWERDOMAIN;
> +       return (cpu < 2) ? SD_SHARE_PKG_RESOURCES : SD_SHARE_PKG_RESOURCES  | SD_SHARE_POWERDOMAIN;
>  }
>
>  static struct sched_domain_topology_level arm_topology[] = {
>  #ifdef CONFIG_SCHED_MC
> -       { cpu_corepower_mask, cpu_corepower_flags, SD_INIT_NAME(GMC) },
> -       { cpu_coregroup_mask, cpu_core_flags, SD_INIT_NAME(MC) },
> +//     { cpu_corepower_mask, cpu_corepower_flags, SD_INIT_NAME(GMC) },
> +       { cpu_coregroup_mask, arm_cpu_core_flags, SD_INIT_NAME(MC) },
>  #endif
>         { cpu_cpu_mask, SD_INIT_NAME(DIE) },
>         { NULL, },
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index 05ce264e5144..45e5aa3d3e80 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -870,14 +870,14 @@ enum cpu_idle_type {
>  #define SD_NUMA                        0x4000  /* cross-node balancing */
>
>  #ifdef CONFIG_SCHED_SMT
> -static inline const int cpu_smt_flags(void)
> +static inline const int cpu_smt_flags(int cpu)
>  {
>         return SD_SHARE_CPUPOWER | SD_SHARE_PKG_RESOURCES;
>  }
>  #endif
>
>  #ifdef CONFIG_SCHED_MC
> -static inline const int cpu_core_flags(void)
> +static inline const int cpu_core_flags(int cpu)
>  {
>         return SD_SHARE_PKG_RESOURCES;
>  }
> @@ -990,7 +990,7 @@ void free_sched_domains(cpumask_var_t doms[], unsigned int ndoms);
>  bool cpus_share_cache(int this_cpu, int that_cpu);
>
>  typedef const struct cpumask *(*sched_domain_mask_f)(int cpu);
> -typedef const int (*sched_domain_flags_f)(void);
> +typedef const int (*sched_domain_flags_f)(int cpu);
>
>  #define SDTL_OVERLAP   0x01
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index f2ee6c72b13a..6b8ba837977c 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -5968,7 +5968,7 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
>         sd_weight = cpumask_weight(tl->mask(cpu));
>
>         if (tl->sd_flags)
> -               sd_flags = (*tl->sd_flags)();
> +               sd_flags = (*tl->sd_flags)(cpu);
>         if (WARN_ONCE(sd_flags & ~TOPOLOGY_SD_FLAGS,
>                         "wrong sd_flags in topology description\n"))
>                 sd_flags &= ~TOPOLOGY_SD_FLAGS;
> @@ -6044,9 +6044,16 @@ sd_init(struct sched_domain_topology_level *tl, int cpu)
>                 sd->idle_idx = 1;
>         }
>
> +       if (tl->sd_flags) {
> +              int flags = (*tl->sd_flags)(cpumask_first(tl->mask(cpu)));
> +
> +              for (;cpu = cpumask_next(cpu, tl->mask(cpu)), cpu < nr_cpu_ids;)
> +                      BUG_ON((*tl->sd_flags)(cpu) != flags);
> +       }
> +
>         sd->private = &tl->data;
>
> -       return sd;
> +       return sd;
>  }
>
>  /*
> --
> 1.7.9.5
>
>
>
>
>