Re: [PATCH net-next] net: page_pool: fix recycle stats for system page_pool allocator

[Yunsheng Lin <linyunsheng@huawei.com>]

On 2024/2/16 17:25, Lorenzo Bianconi wrote:
> Use global percpu page_pool_recycle_stats counter for system page_pool
> allocator instead of allocating a separate percpu variable for each
> (also percpu) page pool instance.

I may missed some obvious discussion in previous version due to spring
holiday.

> 
> Reviewed-by: Toke Hoiland-Jorgensen <toke@redhat.com>
> Signed-off-by: Lorenzo Bianconi <lorenzo@kernel.org>
> ---
>  include/net/page_pool/types.h |  5 +++--
>  net/core/dev.c                |  1 +
>  net/core/page_pool.c          | 22 +++++++++++++++++-----
>  3 files changed, 21 insertions(+), 7 deletions(-)
> 
> diff --git a/include/net/page_pool/types.h b/include/net/page_pool/types.h
> index 3828396ae60c..2515cca6518b 100644
> --- a/include/net/page_pool/types.h
> +++ b/include/net/page_pool/types.h
> @@ -18,8 +18,9 @@
>  					* Please note DMA-sync-for-CPU is still
>  					* device driver responsibility
>  					*/
> -#define PP_FLAG_ALL		(PP_FLAG_DMA_MAP |\
> -				 PP_FLAG_DMA_SYNC_DEV)
> +#define PP_FLAG_SYSTEM_POOL	BIT(2) /* Global system page_pool */
> +#define PP_FLAG_ALL		(PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV | \
> +				 PP_FLAG_SYSTEM_POOL)
>  
>  /*
>   * Fast allocation side cache array/stack
> diff --git a/net/core/dev.c b/net/core/dev.c
> index cc9c2eda65ac..c588808be77f 100644
> --- a/net/core/dev.c
> +++ b/net/core/dev.c
> @@ -11738,6 +11738,7 @@ static int net_page_pool_create(int cpuid)
>  #if IS_ENABLED(CONFIG_PAGE_POOL)
>  	struct page_pool_params page_pool_params = {
>  		.pool_size = SYSTEM_PERCPU_PAGE_POOL_SIZE,
> +		.flags = PP_FLAG_SYSTEM_POOL,
>  		.nid = NUMA_NO_NODE,
>  	};
>  	struct page_pool *pp_ptr;
> diff --git a/net/core/page_pool.c b/net/core/page_pool.c
> index 89c835fcf094..8f0c4e76181b 100644
> --- a/net/core/page_pool.c
> +++ b/net/core/page_pool.c
> @@ -31,6 +31,8 @@
>  #define BIAS_MAX	(LONG_MAX >> 1)
>  
>  #ifdef CONFIG_PAGE_POOL_STATS
> +static DEFINE_PER_CPU(struct page_pool_recycle_stats, pp_system_recycle_stats);
> +
>  /* alloc_stat_inc is intended to be used in softirq context */
>  #define alloc_stat_inc(pool, __stat)	(pool->alloc_stats.__stat++)
>  /* recycle_stat_inc is safe to use when preemption is possible. */
> @@ -220,14 +222,23 @@ static int page_pool_init(struct page_pool *pool,
>  	pool->has_init_callback = !!pool->slow.init_callback;
>  
>  #ifdef CONFIG_PAGE_POOL_STATS
> -	pool->recycle_stats = alloc_percpu(struct page_pool_recycle_stats);
> -	if (!pool->recycle_stats)
> -		return -ENOMEM;
> +	if (!(pool->p.flags & PP_FLAG_SYSTEM_POOL)) {
> +		pool->recycle_stats = alloc_percpu(struct page_pool_recycle_stats);
> +		if (!pool->recycle_stats)
> +			return -ENOMEM;
> +	} else {
> +		/* For system page pool instance we use a singular stats object
> +		 * instead of allocating a separate percpu variable for each
> +		 * (also percpu) page pool instance.
> +		 */
> +		pool->recycle_stats = &pp_system_recycle_stats;

Do we need to return -EINVAL here if page_pool_init() is called with
pool->p.flags & PP_FLAG_SYSTEM_POOL being true and cpuid being a valid
cpu?
If yes, it seems we may be able to use the cpuid to decide if we need
to allocate a new pool->recycle_stats without adding a new flag.

If no, the API for page_pool_create_percpu() seems a litte weird as it
relies on the user calling it correctly.

Also, do we need to enforce that page_pool_create_percpu() is only called
once for the same cpu? if no, we may have two page_pool instance sharing
the same stats.

> +	}
>  #endif
>  
>  	if (ptr_ring_init(&pool->ring, ring_qsize, GFP_KERNEL) < 0) {
>  #ifdef CONFIG_PAGE_POOL_STATS
> -		free_percpu(pool->recycle_stats);
> +		if (!(pool->p.flags & PP_FLAG_SYSTEM_POOL))
> +			free_percpu(pool->recycle_stats);
>  #endif
>  		return -ENOMEM;
>  	}
> @@ -251,7 +262,8 @@ static void page_pool_uninit(struct page_pool *pool)
>  		put_device(pool->p.dev);
>  
>  #ifdef CONFIG_PAGE_POOL_STATS
> -	free_percpu(pool->recycle_stats);
> +	if (!(pool->p.flags & PP_FLAG_SYSTEM_POOL))
> +		free_percpu(pool->recycle_stats);
>  #endif
>  }
>  
>