Re: [PATCH v2 1/2] Make the batch size of the percpu_counterconfigurable

From: Andrew Morton
Date: Wed May 22 2013 - 03:20:46 EST


On Tue, 21 May 2013 17:43:10 -0700 Tim Chen <tim.c.chen@xxxxxxxxxxxxxxx> wrote:

>
> I'll spin off another version of the patch later to add the
> memory-hotplug notifier. In the mean time, does the following looks
> good to you?
>
> ...
>
> --- a/include/linux/mman.h
> +++ b/include/linux/mman.h
> @@ -10,12 +10,30 @@
> extern int sysctl_overcommit_memory;
> extern int sysctl_overcommit_ratio;
> extern struct percpu_counter vm_committed_as;
> +#ifdef CONFIG_SMP
> +extern int vm_committed_as_batch;
> +
> +static inline void mm_compute_batch(void)
> +{
> + int nr = num_present_cpus();
> + int batch = max(32, nr*2);
> +
> + /* batch size set to 0.4% of (total memory/#cpus) */
> + vm_committed_as_batch = max((int) (totalram_pages/nr) / 256, batch);

Use max_t() here.

That expression will overflow when the machine has two exabytes of RAM ;)

> +}
> +#else
> +#define vm_committed_as_batch 0
> +
> +static inline void mm_compute_batch(void)
> +{
> +}
> +#endif

I think it would be better if all the above was not inlined. There's
no particular reason to inline it, and putting it here requires that
mman.h include a bunch more header files (which the patch forgot to
do).

> unsigned long vm_memory_committed(void);
>
> static inline void vm_acct_memory(long pages)
> {
> - percpu_counter_add(&vm_committed_as, pages);
> + __percpu_counter_add(&vm_committed_as, pages, vm_committed_as_batch);
> }
>
> static inline void vm_unacct_memory(long pages)
> diff --git a/mm/mmap.c b/mm/mmap.c
> index f681e18..55c8773 100644
> --- a/mm/mmap.c
> +++ b/mm/mmap.c
> @@ -3145,11 +3145,15 @@ void mm_drop_all_locks(struct mm_struct *mm)
> /*
> * initialise the VMA slab
> */
> +
> +int vm_committed_as_batch;
> +
> void __init mmap_init(void)
> {
> int ret;
>
> ret = percpu_counter_init(&vm_committed_as, 0);
> + mm_compute_batch();
> VM_BUG_ON(ret);
> }
>
> diff --git a/mm/nommu.c b/mm/nommu.c
> index 298884d..9ad16ba 100644
> --- a/mm/nommu.c
> +++ b/mm/nommu.c
> @@ -527,11 +527,15 @@ SYSCALL_DEFINE1(brk, unsigned long, brk)
> /*
> * initialise the VMA and region record slabs
> */
> +
> +int vm_committed_as_batch;

This definition duplicates the one in mmap.c?

> void __init mmap_init(void)
> {
> int ret;
>
> ret = percpu_counter_init(&vm_committed_as, 0);
> + mm_compute_batch();
> VM_BUG_ON(ret);
> vm_region_jar = KMEM_CACHE(vm_region, SLAB_PANIC);
> }

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/