Re: [pchecks v1 2/4] Use raw cpu ops for calls that would trigger with checks

[Ingo Molnar <mingo@kernel.org>]

* Christoph Lameter <cl@linux.com> wrote:

> These location triggered during testing with KVM.
> 
> These are fetches without preemption off where we judged that
> to be more performance efficient or where other means of
> providing synchronization (BH handling) are available.
> 
> Signed-off-by: Christoph Lameter <cl@linux.com>
> 
> Index: linux/include/linux/topology.h
> ===================================================================
> --- linux.orig/include/linux/topology.h	2013-09-12 13:26:29.216103951 -0500
> +++ linux/include/linux/topology.h	2013-09-12 13:41:30.762358687 -0500
> @@ -182,7 +182,7 @@ DECLARE_PER_CPU(int, numa_node);
>  /* Returns the number of the current Node. */
>  static inline int numa_node_id(void)
>  {
> -	return __this_cpu_read(numa_node);
> +	return raw_cpu_read(numa_node);
>  }
>  #endif
>  
> @@ -239,7 +239,7 @@ static inline void set_numa_mem(int node
>  /* Returns the number of the nearest Node with memory */
>  static inline int numa_mem_id(void)
>  {
> -	return __this_cpu_read(_numa_mem_);
> +	return raw_cpu_read(_numa_mem_);
>  }

These are generic primitives used in quite a few places and it can easily 
be a bug to use numa_node_id() in a preemptible section - and this patch 
would hide that fact.

So the correct way to do it is to have checking in these and to introduce 
raw_numa_node_id()/raw_numa_mem_id() and change eventual KVM (and any 
other) preemptible-section use of numa_node_id() to raw_numa_node_id() and 
explain why it's safe to do it.

Thanks,

	Ingo