[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <bff45ab9-2818-4b37-837e-f18ffcab8f47@intel.com>
Date: Thu, 15 Feb 2024 14:41:52 +0100
From: Alexander Lobakin <aleksander.lobakin@...el.com>
To: Lorenzo Bianconi <lorenzo@...nel.org>
CC: <netdev@...r.kernel.org>, <lorenzo.bianconi@...hat.com>,
<kuba@...nel.org>, <davem@...emloft.net>, <edumazet@...gle.com>,
<pabeni@...hat.com>, <hawk@...nel.org>, <ilias.apalodimas@...aro.org>,
<linyunsheng@...wei.com>, <toke@...hat.com>
Subject: Re: [RFC net-next] net: page_pool: fix recycle stats for percpu
page_pool allocator
From: Lorenzo Bianconi <lorenzo@...nel.org>
Date: Wed, 14 Feb 2024 19:08:40 +0100
> Use global page_pool_recycle_stats percpu counter for percpu page_pool
> allocator.
>
> Signed-off-by: Lorenzo Bianconi <lorenzo@...nel.org>
> ---
> net/core/page_pool.c | 18 +++++++++++++-----
> 1 file changed, 13 insertions(+), 5 deletions(-)
>
> diff --git a/net/core/page_pool.c b/net/core/page_pool.c
> index 6e0753e6a95b..1bb83b6e7a61 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_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 +222,19 @@ 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 (cpuid < 0) {
TBH I don't like the idea of assuming that only system page_pools might
be created with cpuid >= 0.
For example, if I have an Rx queue always pinned to one CPU, I might
want to create a PP for this queue with the cpuid set already to save
some cycles when recycling. We might also reuse cpuid later for some
more optimizations or features.
Maybe add a new PP_FLAG indicating that system percpu PP stats should be
used?
> + pool->recycle_stats = alloc_percpu(struct page_pool_recycle_stats);
> + if (!pool->recycle_stats)
> + return -ENOMEM;
> + } else {
> + pool->recycle_stats = &pp_recycle_stats;
> + }
> #endif
>
> if (ptr_ring_init(&pool->ring, ring_qsize, GFP_KERNEL) < 0) {
> #ifdef CONFIG_PAGE_POOL_STATS
> - free_percpu(pool->recycle_stats);
> + if (cpuid < 0)
> + free_percpu(pool->recycle_stats);
> #endif
> return -ENOMEM;
> }
> @@ -251,7 +258,8 @@ 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->cpuid < 0)
> + free_percpu(pool->recycle_stats);
> #endif
> }
>
Thanks,
Olek
Powered by blists - more mailing lists