Message-ID: <168244293875.1741095.10502498932946558516.stgit@firesoul>
Date:   Tue, 25 Apr 2023 19:15:38 +0200
From:   Jesper Dangaard Brouer <brouer@...hat.com>
To:     Ilias Apalodimas <ilias.apalodimas@...aro.org>,
        netdev@...r.kernel.org, Eric Dumazet <eric.dumazet@...il.com>,
        linux-mm@...ck.org, Mel Gorman <mgorman@...hsingularity.net>
Cc:     Jesper Dangaard Brouer <brouer@...hat.com>, lorenzo@...nel.org,
        Toke Høiland-Jørgensen <toke@...hat.com>,
        linyunsheng@...wei.com, bpf@...r.kernel.org,
        "David S. Miller" <davem@...emloft.net>,
        Jakub Kicinski <kuba@...nel.org>,
        Paolo Abeni <pabeni@...hat.com>,
        Andrew Morton <akpm@...ux-foundation.org>, willy@...radead.org
Subject: [PATCH RFC net-next/mm V1 1/3] page_pool: Remove workqueue in new
 shutdown scheme

This removes the workqueue scheme that periodically tests when
inflight reach zero such that page_pool memory can be freed.

This change adds code to fast-path free checking for a shutdown flags
bit after returning PP pages.

Performance is very important for PP, as the fast path is used for
XDP_DROP use-cases where NIC drivers recycle PP pages directly into PP
alloc cache.

The goal were that this code change should have zero impact on this
fast-path. The slight code reorg of likely() are deliberate. Micro
benchmarking done via kernel module[1] on x86_64, shows this code
change only cost a single instruction extra (approx 0.3 nanosec on CPU
E5-1650 @3.60GHz).

It is possible to make this code zero impact via static_key, but that
change is split out into the next patch, as we are unsure if it is
worth the complexity.

[1] https://github.com/netoptimizer/prototype-kernel/blob/master/kernel/lib/bench_page_pool_simple.c

Signed-off-by: Jesper Dangaard Brouer <brouer@...hat.com>
---
 include/net/page_pool.h |    9 +++----
 net/core/page_pool.c    |   59 +++++++++++++++++++----------------------------
 2 files changed, 28 insertions(+), 40 deletions(-)

diff --git a/include/net/page_pool.h b/include/net/page_pool.h
index c8ec2f34722b..a71c0f2695b0 100644
--- a/include/net/page_pool.h
+++ b/include/net/page_pool.h
@@ -50,6 +50,9 @@
 				 PP_FLAG_DMA_SYNC_DEV |\
 				 PP_FLAG_PAGE_FRAG)
 
+/* Internal flag: PP in shutdown phase, waiting for inflight pages */
+#define PP_FLAG_SHUTDOWN	BIT(8)
+
 /*
  * Fast allocation side cache array/stack
  *
@@ -151,11 +154,6 @@ static inline u64 *page_pool_ethtool_stats_get(u64 *data, void *stats)
 struct page_pool {
 	struct page_pool_params p;
 
-	struct delayed_work release_dw;
-	void (*disconnect)(void *);
-	unsigned long defer_start;
-	unsigned long defer_warn;
-
 	u32 pages_state_hold_cnt;
 	unsigned int frag_offset;
 	struct page *frag_page;
@@ -165,6 +163,7 @@ struct page_pool {
 	/* these stats are incremented while in softirq context */
 	struct page_pool_alloc_stats alloc_stats;
 #endif
+	void (*disconnect)(void *);
 	u32 xdp_mem_id;
 
 	/*
diff --git a/net/core/page_pool.c b/net/core/page_pool.c
index e212e9d7edcb..ce7e8dda6403 100644
--- a/net/core/page_pool.c
+++ b/net/core/page_pool.c
@@ -23,9 +23,6 @@
 
 #include <trace/events/page_pool.h>
 
-#define DEFER_TIME (msecs_to_jiffies(1000))
-#define DEFER_WARN_INTERVAL (60 * HZ)
-
 #define BIAS_MAX	LONG_MAX
 
 #ifdef CONFIG_PAGE_POOL_STATS
@@ -380,6 +377,10 @@ static struct page *__page_pool_alloc_pages_slow(struct page_pool *pool,
 	struct page *page;
 	int i, nr_pages;
 
+	/* API usage BUG: PP in shutdown phase, cannot alloc new pages */
+	if (WARN_ON(pool->p.flags & PP_FLAG_SHUTDOWN))
+		return NULL;
+
 	/* Don't support bulk alloc for high-order pages */
 	if (unlikely(pp_order))
 		return __page_pool_alloc_page_order(pool, gfp);
@@ -489,10 +490,6 @@ void page_pool_release_page(struct page_pool *pool, struct page *page)
 	page_pool_set_dma_addr(page, 0);
 skip_dma_unmap:
 	page_pool_clear_pp_info(page);
-
-	/* This may be the last page returned, releasing the pool, so
-	 * it is not safe to reference pool afterwards.
-	 */
 	count = atomic_inc_return_relaxed(&pool->pages_state_release_cnt);
 	trace_page_pool_state_release(pool, page, count);
 }
@@ -535,7 +532,7 @@ static bool page_pool_recycle_in_ring(struct page_pool *pool, struct page *page)
 static bool page_pool_recycle_in_cache(struct page *page,
 				       struct page_pool *pool)
 {
-	if (unlikely(pool->alloc.count == PP_ALLOC_CACHE_SIZE)) {
+	if (pool->alloc.count == PP_ALLOC_CACHE_SIZE) {
 		recycle_stat_inc(pool, cache_full);
 		return false;
 	}
@@ -546,6 +543,8 @@ static bool page_pool_recycle_in_cache(struct page *page,
 	return true;
 }
 
+static void page_pool_shutdown_attempt(struct page_pool *pool);
+
 /* If the page refcnt == 1, this will try to recycle the page.
  * if PP_FLAG_DMA_SYNC_DEV is set, we'll try to sync the DMA area for
  * the configured size min(dma_sync_size, pool->max_len).
@@ -572,7 +571,8 @@ __page_pool_put_page(struct page_pool *pool, struct page *page,
 			page_pool_dma_sync_for_device(pool, page,
 						      dma_sync_size);
 
-		if (allow_direct && in_softirq() &&
+		/* During PP shutdown, no direct recycle must occur */
+		if (likely(allow_direct && in_softirq()) &&
 		    page_pool_recycle_in_cache(page, pool))
 			return NULL;
 
@@ -609,6 +609,8 @@ void page_pool_put_defragged_page(struct page_pool *pool, struct page *page,
 		recycle_stat_inc(pool, ring_full);
 		page_pool_return_page(pool, page);
 	}
+	if (pool->p.flags & PP_FLAG_SHUTDOWN)
+		page_pool_shutdown_attempt(pool);
 }
 EXPORT_SYMBOL(page_pool_put_defragged_page);
 
@@ -648,13 +650,17 @@ void page_pool_put_page_bulk(struct page_pool *pool, void **data,
 
 	/* Hopefully all pages was return into ptr_ring */
 	if (likely(i == bulk_len))
-		return;
+		goto out;
 
 	/* ptr_ring cache full, free remaining pages outside producer lock
 	 * since put_page() with refcnt == 1 can be an expensive operation
 	 */
 	for (; i < bulk_len; i++)
 		page_pool_return_page(pool, data[i]);
+
+out:
+	if (pool->p.flags & PP_FLAG_SHUTDOWN)
+		page_pool_shutdown_attempt(pool);
 }
 EXPORT_SYMBOL(page_pool_put_page_bulk);
 
@@ -808,27 +814,10 @@ static int page_pool_release(struct page_pool *pool)
 	return inflight;
 }
 
-static void page_pool_release_retry(struct work_struct *wq)
+noinline
+static void page_pool_shutdown_attempt(struct page_pool *pool)
 {
-	struct delayed_work *dwq = to_delayed_work(wq);
-	struct page_pool *pool = container_of(dwq, typeof(*pool), release_dw);
-	int inflight;
-
-	inflight = page_pool_release(pool);
-	if (!inflight)
-		return;
-
-	/* Periodic warning */
-	if (time_after_eq(jiffies, pool->defer_warn)) {
-		int sec = (s32)((u32)jiffies - (u32)pool->defer_start) / HZ;
-
-		pr_warn("%s() stalled pool shutdown %d inflight %d sec\n",
-			__func__, inflight, sec);
-		pool->defer_warn = jiffies + DEFER_WARN_INTERVAL;
-	}
-
-	/* Still not ready to be disconnected, retry later */
-	schedule_delayed_work(&pool->release_dw, DEFER_TIME);
+	page_pool_release(pool);
 }
 
 void page_pool_use_xdp_mem(struct page_pool *pool, void (*disconnect)(void *),
@@ -868,11 +857,11 @@ void page_pool_destroy(struct page_pool *pool)
 	if (!page_pool_release(pool))
 		return;
 
-	pool->defer_start = jiffies;
-	pool->defer_warn  = jiffies + DEFER_WARN_INTERVAL;
-
-	INIT_DELAYED_WORK(&pool->release_dw, page_pool_release_retry);
-	schedule_delayed_work(&pool->release_dw, DEFER_TIME);
+	/* PP have pages inflight, thus cannot immediately release memory.
+	 * Enter into shutdown phase, and retry release to handle races.
+	 */
+	pool->p.flags |= PP_FLAG_SHUTDOWN;
+	page_pool_shutdown_attempt(pool);
 }
 EXPORT_SYMBOL(page_pool_destroy);
 

Powered by blists - more mailing lists