Merge branch 'page_pool-allow-direct-bulk-recycling'

Alexander Lobakin says:

====================
page_pool: allow direct bulk recycling

Previously, there was no reliable way to check whether it's safe to use
direct PP cache. The drivers were passing @allow_direct to the PP
recycling functions and that was it. Bulk recycling is used by
xdp_return_frame_bulk() on .ndo_xdp_xmit() frames completion where
the page origin is unknown, thus the direct recycling has never been
tried.
Now that we have at least 2 ways of checking if we're allowed to perform
direct recycling -- pool->p.napi (Jakub) and pool->cpuid (Lorenzo), we
can use them when doing bulk recycling as well. Just move that logic
from the skb core to the PP core and call it before
__page_pool_put_page() every time @allow_direct is false.
Under high .ndo_xdp_xmit() traffic load, the win is 2-3% Pps assuming
the sending driver uses xdp_return_frame_bulk() on Tx completion.
====================

Link: https://lore.kernel.org/r/20240329165507.3240110-1-aleksander.lobakin@intel.comSigned-off-by: Jakub Kicinski <kuba@kernel.org>

include/linux/skbuff.h

@@ -3510,25 +3510,25 @@ int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
 		    unsigned int headroom);
 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
 			 struct bpf_prog *prog);
-bool napi_pp_put_page(struct page *page, bool napi_safe);
+bool napi_pp_put_page(struct page *page);
 
 static inline void
-skb_page_unref(const struct sk_buff *skb, struct page *page, bool napi_safe)
+skb_page_unref(const struct sk_buff *skb, struct page *page)
 {
 #ifdef CONFIG_PAGE_POOL
-	if (skb->pp_recycle && napi_pp_put_page(page, napi_safe))
+	if (skb->pp_recycle && napi_pp_put_page(page))
 		return;
 #endif
 	put_page(page);
 }
 
 static inline void
-napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe)
+napi_frag_unref(skb_frag_t *frag, bool recycle)
 {
 	struct page *page = skb_frag_page(frag);
 
 #ifdef CONFIG_PAGE_POOL
-	if (recycle && napi_pp_put_page(page, napi_safe))
+	if (recycle && napi_pp_put_page(page))
 		return;
 #endif
 	put_page(page);
@@ -3544,7 +3544,7 @@ napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe)
  */
 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle)
 {
-	napi_frag_unref(frag, recycle, false);
+	napi_frag_unref(frag, recycle);
 }
 
 /**

net/core/page_pool.c

@@ -690,8 +690,7 @@ __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() &&
-		    page_pool_recycle_in_cache(page, pool))
+		if (allow_direct && page_pool_recycle_in_cache(page, pool))
 			return NULL;
 
 		/* Page found as candidate for recycling */
@@ -716,9 +715,35 @@ __page_pool_put_page(struct page_pool *pool, struct page *page,
 	return NULL;
 }
 
+static bool page_pool_napi_local(const struct page_pool *pool)
+{
+	const struct napi_struct *napi;
+	u32 cpuid;
+
+	if (unlikely(!in_softirq()))
+		return false;
+
+	/* Allow direct recycle if we have reasons to believe that we are
+	 * in the same context as the consumer would run, so there's
+	 * no possible race.
+	 * __page_pool_put_page() makes sure we're not in hardirq context
+	 * and interrupts are enabled prior to accessing the cache.
+	 */
+	cpuid = smp_processor_id();
+	if (READ_ONCE(pool->cpuid) == cpuid)
+		return true;
+
+	napi = READ_ONCE(pool->p.napi);
+
+	return napi && READ_ONCE(napi->list_owner) == cpuid;
+}
+
 void page_pool_put_unrefed_page(struct page_pool *pool, struct page *page,
 				unsigned int dma_sync_size, bool allow_direct)
 {
+	if (!allow_direct)
+		allow_direct = page_pool_napi_local(pool);
+
 	page = __page_pool_put_page(pool, page, dma_sync_size, allow_direct);
 	if (page && !page_pool_recycle_in_ring(pool, page)) {
 		/* Cache full, fallback to free pages */
@@ -747,8 +772,11 @@ void page_pool_put_page_bulk(struct page_pool *pool, void **data,
 			     int count)
 {
 	int i, bulk_len = 0;
+	bool allow_direct;
 	bool in_softirq;
 
+	allow_direct = page_pool_napi_local(pool);
+
 	for (i = 0; i < count; i++) {
 		struct page *page = virt_to_head_page(data[i]);
 
@@ -756,13 +784,13 @@ void page_pool_put_page_bulk(struct page_pool *pool, void **data,
 		if (!page_pool_is_last_ref(page))
 			continue;
 
-		page = __page_pool_put_page(pool, page, -1, false);
+		page = __page_pool_put_page(pool, page, -1, allow_direct);
 		/* Approved for bulk recycling in ptr_ring cache */
 		if (page)
 			data[bulk_len++] = page;
 	}
 
-	if (unlikely(!bulk_len))
+	if (!bulk_len)
 		return;
 
 	/* Bulk producer into ptr_ring page_pool cache */
@@ -969,7 +997,7 @@ void page_pool_use_xdp_mem(struct page_pool *pool, void (*disconnect)(void *),
 static void page_pool_disable_direct_recycling(struct page_pool *pool)
 {
 	/* Disable direct recycling based on pool->cpuid.
-	 * Paired with READ_ONCE() in napi_pp_put_page().
+	 * Paired with READ_ONCE() in page_pool_napi_local().
 	 */
 	WRITE_ONCE(pool->cpuid, -1);
 

net/core/skbuff.c

@@ -1004,11 +1004,8 @@ int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
 EXPORT_SYMBOL(skb_cow_data_for_xdp);
 
 #if IS_ENABLED(CONFIG_PAGE_POOL)
-bool napi_pp_put_page(struct page *page, bool napi_safe)
+bool napi_pp_put_page(struct page *page)
 {
-	bool allow_direct = false;
-	struct page_pool *pp;
-
 	page = compound_head(page);
 
 	/* page->pp_magic is OR'ed with PP_SIGNATURE after the allocation
@@ -1021,39 +1018,18 @@ bool napi_pp_put_page(struct page *page, bool napi_safe)
 	if (unlikely(!is_pp_page(page)))
 		return false;
 
-	pp = page->pp;
-
-	/* Allow direct recycle if we have reasons to believe that we are
-	 * in the same context as the consumer would run, so there's
-	 * no possible race.
-	 * __page_pool_put_page() makes sure we're not in hardirq context
-	 * and interrupts are enabled prior to accessing the cache.
-	 */
-	if (napi_safe || in_softirq()) {
-		const struct napi_struct *napi = READ_ONCE(pp->p.napi);
-		unsigned int cpuid = smp_processor_id();
-
-		allow_direct = napi && READ_ONCE(napi->list_owner) == cpuid;
-		allow_direct |= READ_ONCE(pp->cpuid) == cpuid;
-	}
-
-	/* Driver set this to memory recycling info. Reset it on recycle.
-	 * This will *not* work for NIC using a split-page memory model.
-	 * The page will be returned to the pool here regardless of the
-	 * 'flipped' fragment being in use or not.
-	 */
-	page_pool_put_full_page(pp, page, allow_direct);
+	page_pool_put_full_page(page->pp, page, false);
 
 	return true;
 }
 EXPORT_SYMBOL(napi_pp_put_page);
 #endif
 
-static bool skb_pp_recycle(struct sk_buff *skb, void *data, bool napi_safe)
+static bool skb_pp_recycle(struct sk_buff *skb, void *data)
 {
 	if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
 		return false;
-	return napi_pp_put_page(virt_to_page(data), napi_safe);
+	return napi_pp_put_page(virt_to_page(data));
 }
 
 /**
@@ -1095,12 +1071,12 @@ static void skb_kfree_head(void *head, unsigned int end_offset)
 		kfree(head);
 }
 
-static void skb_free_head(struct sk_buff *skb, bool napi_safe)
+static void skb_free_head(struct sk_buff *skb)
 {
 	unsigned char *head = skb->head;
 
 	if (skb->head_frag) {
-		if (skb_pp_recycle(skb, head, napi_safe))
+		if (skb_pp_recycle(skb, head))
 			return;
 		skb_free_frag(head);
 	} else {
@@ -1108,8 +1084,7 @@ static void skb_free_head(struct sk_buff *skb, bool napi_safe)
 	}
 }
 
-static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason,
-			     bool napi_safe)
+static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason)
 {
 	struct skb_shared_info *shinfo = skb_shinfo(skb);
 	int i;
@@ -1126,13 +1101,13 @@ static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason,
 	}
 
 	for (i = 0; i < shinfo->nr_frags; i++)
-		napi_frag_unref(&shinfo->frags[i], skb->pp_recycle, napi_safe);
+		napi_frag_unref(&shinfo->frags[i], skb->pp_recycle);
 
 free_head:
 	if (shinfo->frag_list)
 		kfree_skb_list_reason(shinfo->frag_list, reason);
 
-	skb_free_head(skb, napi_safe);
+	skb_free_head(skb);
 exit:
 	/* When we clone an SKB we copy the reycling bit. The pp_recycle
 	 * bit is only set on the head though, so in order to avoid races
@@ -1193,12 +1168,11 @@ void skb_release_head_state(struct sk_buff *skb)
 }
 
 /* Free everything but the sk_buff shell. */
-static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason,
-			    bool napi_safe)
+static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason)
 {
 	skb_release_head_state(skb);
 	if (likely(skb->head))
-		skb_release_data(skb, reason, napi_safe);
+		skb_release_data(skb, reason);
 }
 
 /**
@@ -1212,7 +1186,7 @@ static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason,
 
 void __kfree_skb(struct sk_buff *skb)
 {
-	skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED, false);
+	skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
 	kfree_skbmem(skb);
 }
 EXPORT_SYMBOL(__kfree_skb);
@@ -1269,7 +1243,7 @@ static void kfree_skb_add_bulk(struct sk_buff *skb,
 		return;
 	}
 
-	skb_release_all(skb, reason, false);
+	skb_release_all(skb, reason);
 	sa->skb_array[sa->skb_count++] = skb;
 
 	if (unlikely(sa->skb_count == KFREE_SKB_BULK_SIZE)) {
@@ -1443,7 +1417,7 @@ EXPORT_SYMBOL(consume_skb);
 void __consume_stateless_skb(struct sk_buff *skb)
 {
 	trace_consume_skb(skb, __builtin_return_address(0));
-	skb_release_data(skb, SKB_CONSUMED, false);
+	skb_release_data(skb, SKB_CONSUMED);
 	kfree_skbmem(skb);
 }
 
@@ -1470,7 +1444,7 @@ static void napi_skb_cache_put(struct sk_buff *skb)
 
 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason)
 {
-	skb_release_all(skb, reason, true);
+	skb_release_all(skb, reason);
 	napi_skb_cache_put(skb);
 }
 
@@ -1508,7 +1482,7 @@ void napi_consume_skb(struct sk_buff *skb, int budget)
 		return;
 	}
 
-	skb_release_all(skb, SKB_CONSUMED, !!budget);
+	skb_release_all(skb, SKB_CONSUMED);
 	napi_skb_cache_put(skb);
 }
 EXPORT_SYMBOL(napi_consume_skb);
@@ -1639,7 +1613,7 @@ EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
  */
 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
 {
-	skb_release_all(dst, SKB_CONSUMED, false);
+	skb_release_all(dst, SKB_CONSUMED);
 	return __skb_clone(dst, src);
 }
 EXPORT_SYMBOL_GPL(skb_morph);
@@ -2271,9 +2245,9 @@ int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
 		if (skb_has_frag_list(skb))
 			skb_clone_fraglist(skb);
 
-		skb_release_data(skb, SKB_CONSUMED, false);
+		skb_release_data(skb, SKB_CONSUMED);
 	} else {
-		skb_free_head(skb, false);
+		skb_free_head(skb);
 	}
 	off = (data + nhead) - skb->head;
 
@@ -6574,12 +6548,12 @@ static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
 			skb_frag_ref(skb, i);
 		if (skb_has_frag_list(skb))
 			skb_clone_fraglist(skb);
-		skb_release_data(skb, SKB_CONSUMED, false);
+		skb_release_data(skb, SKB_CONSUMED);
 	} else {
 		/* we can reuse existing recount- all we did was
 		 * relocate values
 		 */
-		skb_free_head(skb, false);
+		skb_free_head(skb);
 	}
 
 	skb->head = data;
@@ -6714,7 +6688,7 @@ static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
 		skb_kfree_head(data, size);
 		return -ENOMEM;
 	}
-	skb_release_data(skb, SKB_CONSUMED, false);
+	skb_release_data(skb, SKB_CONSUMED);
 
 	skb->head = data;
 	skb->head_frag = 0;

net/ipv4/esp4.c

@@ -114,7 +114,7 @@ static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb)
 	 */
 	if (req->src != req->dst)
 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
-			skb_page_unref(skb, sg_page(sg), false);
+			skb_page_unref(skb, sg_page(sg));
 }
 
 #ifdef CONFIG_INET_ESPINTCP

net/ipv6/esp6.c

@@ -131,7 +131,7 @@ static void esp_ssg_unref(struct xfrm_state *x, void *tmp, struct sk_buff *skb)
 	 */
 	if (req->src != req->dst)
 		for (sg = sg_next(req->src); sg; sg = sg_next(sg))
-			skb_page_unref(skb, sg_page(sg), false);
+			skb_page_unref(skb, sg_page(sg));
 }
 
 #ifdef CONFIG_INET6_ESPINTCP