From a8e76b8635c6daf65fb029ea4476ef777768c092 Mon Sep 17 00:00:00 2001
From: Sujuan Chen <sujuan.chen@mediatek.com>
Date: Thu, 5 Jan 2023 16:43:57 +0800
Subject: [PATCH 3015/3015] mt76: connac: wed: add wed rx copy skb and revert
 page_pool

Signed-off-by: Sujuan Chen <sujuan.chen@mediatek.com>
---
 dma.c         | 138 ++++++++++++++++++++++++++++++++------------------
 mac80211.c    |  57 ---------------------
 mt76.h        |  22 +-------
 mt7915/main.c |  26 ++++------
 mt7915/mmio.c |  74 +++++++++++++++++++--------
 mt7921/main.c |  31 ++----------
 usb.c         |  42 +++++++--------
 7 files changed, 180 insertions(+), 210 deletions(-)

diff --git a/dma.c b/dma.c
index 76af6506..79da538c 100644
--- a/dma.c
+++ b/dma.c
@@ -173,7 +173,7 @@ mt76_free_pending_rxwi(struct mt76_dev *dev)
 	local_bh_disable();
 	while ((r = __mt76_get_rxwi(dev)) != NULL) {
 		if (r->ptr)
-			mt76_put_page_pool_buf(r->ptr, false);
+			skb_free_frag(r->ptr);
 		kfree(r);
 	}
 	local_bh_enable();
@@ -208,11 +208,11 @@ mt76_dma_queue_reset(struct mt76_dev *dev, struct mt76_queue *q)
 
 static int
 mt76_dma_add_rx_buf(struct mt76_dev *dev, struct mt76_queue *q,
-		    struct mt76_queue_buf *buf, void *data)
+		    struct mt76_queue_buf *buf, void *data,
+		    struct mt76_rxwi_cache *rxwi)
 {
 	struct mt76_desc *desc = &q->desc[q->head];
 	struct mt76_queue_entry *entry = &q->entry[q->head];
-	struct mt76_rxwi_cache *rxwi = NULL;
 	u32 buf1 = 0, ctrl;
 	int idx = q->head;
 	int rx_token;
@@ -220,9 +220,11 @@ mt76_dma_add_rx_buf(struct mt76_dev *dev, struct mt76_queue *q,
 	ctrl = FIELD_PREP(MT_DMA_CTL_SD_LEN0, buf[0].len);
 
 	if (mt76_queue_is_wed_rx(q)) {
-		rxwi = mt76_get_rxwi(dev);
-		if (!rxwi)
-			return -ENOMEM;
+		if (!rxwi) {
+			rxwi = mt76_get_rxwi(dev);
+			if (!rxwi)
+				return -ENOMEM;
+		}
 
 		rx_token = mt76_rx_token_consume(dev, data, rxwi, buf->addr);
 		if (rx_token < 0) {
@@ -387,7 +389,7 @@ mt76_dma_tx_cleanup(struct mt76_dev *dev, struct mt76_queue *q, bool flush)
 
 static void *
 mt76_dma_get_buf(struct mt76_dev *dev, struct mt76_queue *q, int idx,
-		 int *len, u32 *info, bool *more, bool *drop)
+		 int *len, u32 *info, bool *more, bool *drop, bool flush)
 {
 	struct mt76_queue_entry *e = &q->entry[idx];
 	struct mt76_desc *desc = &q->desc[idx];
@@ -435,15 +437,47 @@ mt76_dma_get_buf(struct mt76_dev *dev, struct mt76_queue *q, int idx,
 		if (!r)
 			return NULL;
 
-		dma_sync_single_for_cpu(dev->dma_dev, r->dma_addr,
-				SKB_WITH_OVERHEAD(q->buf_size),
-				page_pool_get_dma_dir(q->page_pool));
+		dma_unmap_single(dev->dma_dev, r->dma_addr,
+				 SKB_WITH_OVERHEAD(q->buf_size),
+				 DMA_FROM_DEVICE);
 
-		buf = r->ptr;
-		r->dma_addr = 0;
-		r->ptr = NULL;
+		if (flush) {
+			buf = r->ptr;
+			r->dma_addr = 0;
+			r->ptr = NULL;
 
-		mt76_put_rxwi(dev, r);
+			mt76_put_rxwi(dev, r);
+		} else {
+			struct mt76_queue_buf qbuf;
+
+			buf = page_frag_alloc(&q->rx_page, q->buf_size, GFP_ATOMIC);
+			if (!buf)
+				return NULL;
+
+			memcpy(buf, r->ptr, SKB_WITH_OVERHEAD(q->buf_size));
+
+			r->dma_addr = dma_map_single(dev->dma_dev, r->ptr,
+						     SKB_WITH_OVERHEAD(q->buf_size),
+						     DMA_FROM_DEVICE);
+			if (unlikely(dma_mapping_error(dev->dma_dev, r->dma_addr))) {
+				skb_free_frag(r->ptr);
+				mt76_put_rxwi(dev, r);
+				return NULL;
+			}
+
+			qbuf.addr = r->dma_addr;
+			qbuf.len = SKB_WITH_OVERHEAD(q->buf_size);
+			qbuf.skip_unmap = false;
+
+			if (mt76_dma_add_rx_buf(dev, q, &qbuf, r->ptr, r) < 0) {
+				dma_unmap_single(dev->dma_dev, r->dma_addr,
+						 SKB_WITH_OVERHEAD(q->buf_size),
+						 DMA_FROM_DEVICE);
+				skb_free_frag(r->ptr);
+				mt76_put_rxwi(dev, r);
+				return NULL;
+			}
+		}
 
 		if (drop) {
 			u32 ctrl = le32_to_cpu(READ_ONCE(desc->ctrl));
@@ -463,9 +497,9 @@ mt76_dma_get_buf(struct mt76_dev *dev, struct mt76_queue *q, int idx,
 	} else {
 		buf = e->buf;
 		e->buf = NULL;
-		dma_sync_single_for_cpu(dev->dma_dev, e->dma_addr[0],
-				SKB_WITH_OVERHEAD(q->buf_size),
-				page_pool_get_dma_dir(q->page_pool));
+		dma_unmap_single(dev->dma_dev, e->dma_addr[0],
+				 SKB_WITH_OVERHEAD(q->buf_size),
+				 DMA_FROM_DEVICE);
 	}
 
 	return buf;
@@ -489,7 +523,7 @@ mt76_dma_dequeue(struct mt76_dev *dev, struct mt76_queue *q, bool flush,
 	q->tail = (q->tail + 1) % q->ndesc;
 	q->queued--;
 
-	return mt76_dma_get_buf(dev, q, idx, len, info, more, drop);
+	return mt76_dma_get_buf(dev, q, idx, len, info, more, drop, flush);
 }
 
 static int
@@ -616,11 +650,12 @@ free_skb:
 }
 
 static int
-mt76_dma_rx_fill(struct mt76_dev *dev, struct mt76_queue *q,
-		 bool allow_direct)
+mt76_dma_rx_fill(struct mt76_dev *dev, struct mt76_queue *q)
 {
 	int len = SKB_WITH_OVERHEAD(q->buf_size);
-	int frames = 0;
+	int frames = 0, offset = q->buf_offset;
+	dma_addr_t addr;
+	bool flags = false;
 
 	if (!q->ndesc)
 		return 0;
@@ -628,31 +663,35 @@ mt76_dma_rx_fill(struct mt76_dev *dev, struct mt76_queue *q,
 	spin_lock_bh(&q->lock);
 
 	while (q->queued < q->ndesc - 1) {
-		enum dma_data_direction dir;
 		struct mt76_queue_buf qbuf;
-		dma_addr_t addr;
-		int offset;
-		void *buf;
+		void *buf = NULL;
 
-		buf = mt76_get_page_pool_buf(q, &offset, q->buf_size);
+		buf = page_frag_alloc(&q->rx_page, q->buf_size, GFP_ATOMIC);
 		if (!buf)
 			break;
 
-		addr = page_pool_get_dma_addr(virt_to_head_page(buf)) + offset;
-		dir = page_pool_get_dma_dir(q->page_pool);
-		dma_sync_single_for_device(dev->dma_dev, addr, len, dir);
+		addr = dma_map_single(dev->dma_dev, buf, len, DMA_FROM_DEVICE);
+		if (unlikely(dma_mapping_error(dev->dma_dev, addr))) {
+			skb_free_frag(buf);
+			break;
+		}
 
-		qbuf.addr = addr + q->buf_offset;
-		qbuf.len = len - q->buf_offset;
+		qbuf.addr = addr + offset;
+		qbuf.len = len - offset;
 		qbuf.skip_unmap = false;
-		if (mt76_dma_add_rx_buf(dev, q, &qbuf, buf) < 0) {
-			mt76_put_page_pool_buf(buf, allow_direct);
+		if (mt76_dma_add_rx_buf(dev, q, &qbuf, buf, NULL) < 0) {
+			dma_unmap_single(dev->dma_dev, addr, len,
+					 DMA_FROM_DEVICE);
+			skb_free_frag(buf);
 			break;
 		}
 		frames++;
 	}
 
-	if (frames)
+	flags = (q->flags & MT_QFLAG_WED) &&
+		FIELD_GET(MT_QFLAG_WED_TYPE, q->flags) == MT76_WED_Q_RX;
+
+	if (frames || flags)
 		mt76_dma_kick_queue(dev, q);
 
 	spin_unlock_bh(&q->lock);
@@ -690,7 +729,7 @@ int mt76_dma_wed_setup(struct mt76_dev *dev, struct mt76_queue *q, bool reset)
 		/* WED txfree queue needs ring to be initialized before setup */
 		q->flags = 0;
 		mt76_dma_queue_reset(dev, q);
-		mt76_dma_rx_fill(dev, q, false);
+		mt76_dma_rx_fill(dev, q);
 		q->flags = flags;
 
 		ret = mtk_wed_device_txfree_ring_setup(wed, q->regs);
@@ -738,10 +777,6 @@ mt76_dma_alloc_queue(struct mt76_dev *dev, struct mt76_queue *q,
 	if (!q->entry)
 		return -ENOMEM;
 
-	ret = mt76_create_page_pool(dev, q);
-	if (ret)
-		return ret;
-
 	ret = mt76_dma_wed_setup(dev, q, false);
 	if (ret)
 		return ret;
@@ -755,6 +790,7 @@ mt76_dma_alloc_queue(struct mt76_dev *dev, struct mt76_queue *q,
 static void
 mt76_dma_rx_cleanup(struct mt76_dev *dev, struct mt76_queue *q)
 {
+	struct page *page;
 	void *buf;
 	bool more;
 
@@ -768,7 +804,7 @@ mt76_dma_rx_cleanup(struct mt76_dev *dev, struct mt76_queue *q)
 		if (!buf)
 			break;
 
-		mt76_put_page_pool_buf(buf, false);
+		skb_free_frag(buf);
 	} while (1);
 
 	if (q->rx_head) {
@@ -777,6 +813,15 @@ mt76_dma_rx_cleanup(struct mt76_dev *dev, struct mt76_queue *q)
 	}
 
 	spin_unlock_bh(&q->lock);
+
+	if (FIELD_GET(MT_QFLAG_WED_TYPE, q->flags) != MT76_WED_Q_RX) {
+		if (!q->rx_page.va)
+			return;
+
+		page = virt_to_page(q->rx_page.va);
+		__page_frag_cache_drain(page, q->rx_page.pagecnt_bias);
+		memset(&q->rx_page, 0, sizeof(q->rx_page));
+	}
 }
 
 static void
@@ -797,7 +842,7 @@ mt76_dma_rx_reset(struct mt76_dev *dev, enum mt76_rxq_id qid)
 	mt76_dma_wed_setup(dev, q, true);
 	if (q->flags != MT_WED_Q_TXFREE) {
 		mt76_dma_sync_idx(dev, q);
-		mt76_dma_rx_fill(dev, q, false);
+		mt76_dma_rx_fill(dev, q);
 	}
 }
 
@@ -815,7 +860,7 @@ mt76_add_fragment(struct mt76_dev *dev, struct mt76_queue *q, void *data,
 
 		skb_add_rx_frag(skb, nr_frags, page, offset, len, q->buf_size);
 	} else {
-		mt76_put_page_pool_buf(data, true);
+		skb_free_frag(data);
 	}
 
 	if (more)
@@ -888,7 +933,6 @@ mt76_dma_rx_process(struct mt76_dev *dev, struct mt76_queue *q, int budget)
 			goto free_frag;
 
 		skb_reserve(skb, q->buf_offset);
-		skb_mark_for_recycle(skb);
 
 		*(u32 *)skb->cb = info;
 
@@ -904,10 +948,10 @@ mt76_dma_rx_process(struct mt76_dev *dev, struct mt76_queue *q, int budget)
 		continue;
 
 free_frag:
-		mt76_put_page_pool_buf(data, true);
+		skb_free_frag(data);
 	}
 
-	mt76_dma_rx_fill(dev, q, true);
+	mt76_dma_rx_fill(dev, q);
 	return done;
 }
 
@@ -952,7 +996,7 @@ mt76_dma_init(struct mt76_dev *dev,
 
 	mt76_for_each_q_rx(dev, i) {
 		netif_napi_add(&dev->napi_dev, &dev->napi[i], poll);
-		mt76_dma_rx_fill(dev, &dev->q_rx[i], false);
+		mt76_dma_rx_fill(dev, &dev->q_rx[i]);
 		napi_enable(&dev->napi[i]);
 	}
 
@@ -1003,8 +1047,6 @@ void mt76_dma_cleanup(struct mt76_dev *dev)
 
 		netif_napi_del(&dev->napi[i]);
 		mt76_dma_rx_cleanup(dev, q);
-
-		page_pool_destroy(q->page_pool);
 	}
 
 	mt76_free_pending_txwi(dev);
diff --git a/mac80211.c b/mac80211.c
index 95861d47..90447472 100644
--- a/mac80211.c
+++ b/mac80211.c
@@ -4,7 +4,6 @@
  */
 #include <linux/sched.h>
 #include <linux/of.h>
-#include <net/page_pool.h>
 #include "mt76.h"
 
 #define CHAN2G(_idx, _freq) {			\
@@ -569,47 +568,6 @@ void mt76_unregister_phy(struct mt76_phy *phy)
 }
 EXPORT_SYMBOL_GPL(mt76_unregister_phy);
 
-int mt76_create_page_pool(struct mt76_dev *dev, struct mt76_queue *q)
-{
-	struct page_pool_params pp_params = {
-		.order = 0,
-		.flags = PP_FLAG_PAGE_FRAG,
-		.nid = NUMA_NO_NODE,
-		.dev = dev->dma_dev,
-	};
-	int idx = q - dev->q_rx;
-
-	switch (idx) {
-	case MT_RXQ_MAIN:
-	case MT_RXQ_BAND1:
-	case MT_RXQ_BAND2:
-		pp_params.pool_size = 256;
-		break;
-	default:
-		pp_params.pool_size = 16;
-		break;
-	}
-
-	if (mt76_is_mmio(dev)) {
-		/* rely on page_pool for DMA mapping */
-		pp_params.flags |= PP_FLAG_DMA_MAP | PP_FLAG_DMA_SYNC_DEV;
-		pp_params.dma_dir = DMA_FROM_DEVICE;
-		pp_params.max_len = PAGE_SIZE;
-		pp_params.offset = 0;
-	}
-
-	q->page_pool = page_pool_create(&pp_params);
-	if (IS_ERR(q->page_pool)) {
-		int err = PTR_ERR(q->page_pool);
-
-		q->page_pool = NULL;
-		return err;
-	}
-
-	return 0;
-}
-EXPORT_SYMBOL_GPL(mt76_create_page_pool);
-
 struct mt76_dev *
 mt76_alloc_device(struct device *pdev, unsigned int size,
 		  const struct ieee80211_ops *ops,
@@ -1745,21 +1703,6 @@ void mt76_ethtool_worker(struct mt76_ethtool_worker_info *wi,
 }
 EXPORT_SYMBOL_GPL(mt76_ethtool_worker);
 
-void mt76_ethtool_page_pool_stats(struct mt76_dev *dev, u64 *data, int *index)
-{
-#ifdef CONFIG_PAGE_POOL_STATS
-	struct page_pool_stats stats = {};
-	int i;
-
-	mt76_for_each_q_rx(dev, i)
-		page_pool_get_stats(dev->q_rx[i].page_pool, &stats);
-
-	page_pool_ethtool_stats_get(data, &stats);
-	*index += page_pool_ethtool_stats_get_count();
-#endif
-}
-EXPORT_SYMBOL_GPL(mt76_ethtool_page_pool_stats);
-
 enum mt76_dfs_state mt76_phy_dfs_state(struct mt76_phy *phy)
 {
 	struct ieee80211_hw *hw = phy->hw;
diff --git a/mt76.h b/mt76.h
index 22c6531a..19100d62 100644
--- a/mt76.h
+++ b/mt76.h
@@ -210,7 +210,7 @@ struct mt76_queue {
 
 	dma_addr_t desc_dma;
 	struct sk_buff *rx_head;
-	struct page_pool *page_pool;
+	struct page_frag_cache rx_page;
 };
 
 struct mt76_mcu_ops {
@@ -1458,7 +1458,6 @@ mt76u_bulk_msg(struct mt76_dev *dev, void *data, int len, int *actual_len,
 	return usb_bulk_msg(udev, pipe, data, len, actual_len, timeout);
 }
 
-void mt76_ethtool_page_pool_stats(struct mt76_dev *dev, u64 *data, int *index);
 void mt76_ethtool_worker(struct mt76_ethtool_worker_info *wi,
 			 struct mt76_sta_stats *stats, bool eht);
 int mt76_skb_adjust_pad(struct sk_buff *skb, int pad);
@@ -1570,25 +1569,6 @@ void __mt76_set_tx_blocked(struct mt76_dev *dev, bool blocked);
 struct mt76_rxwi_cache *mt76_rx_token_release(struct mt76_dev *dev, int token);
 int mt76_rx_token_consume(struct mt76_dev *dev, void *ptr,
 			  struct mt76_rxwi_cache *r, dma_addr_t phys);
-int mt76_create_page_pool(struct mt76_dev *dev, struct mt76_queue *q);
-static inline void mt76_put_page_pool_buf(void *buf, bool allow_direct)
-{
-	struct page *page = virt_to_head_page(buf);
-
-	page_pool_put_full_page(page->pp, page, allow_direct);
-}
-
-static inline void *
-mt76_get_page_pool_buf(struct mt76_queue *q, u32 *offset, u32 size)
-{
-	struct page *page;
-
-	page = page_pool_dev_alloc_frag(q->page_pool, offset, size);
-	if (!page)
-		return NULL;
-
-	return page_address(page) + *offset;
-}
 
 static inline void mt76_set_tx_blocked(struct mt76_dev *dev, bool blocked)
 {
diff --git a/mt7915/main.c b/mt7915/main.c
index a4d1b816..21caddbe 100644
--- a/mt7915/main.c
+++ b/mt7915/main.c
@@ -1359,22 +1359,19 @@ void mt7915_get_et_strings(struct ieee80211_hw *hw,
 			   struct ieee80211_vif *vif,
 			   u32 sset, u8 *data)
 {
-	if (sset != ETH_SS_STATS)
-		return;
-
-	memcpy(data, *mt7915_gstrings_stats, sizeof(mt7915_gstrings_stats));
-	data += sizeof(mt7915_gstrings_stats);
-	page_pool_ethtool_stats_get_strings(data);
+	if (sset == ETH_SS_STATS)
+		memcpy(data, *mt7915_gstrings_stats,
+		       sizeof(mt7915_gstrings_stats));
 }
 
 static
 int mt7915_get_et_sset_count(struct ieee80211_hw *hw,
 			     struct ieee80211_vif *vif, int sset)
 {
-	if (sset != ETH_SS_STATS)
-		return 0;
+	if (sset == ETH_SS_STATS)
+		return MT7915_SSTATS_LEN;
 
-	return MT7915_SSTATS_LEN + page_pool_ethtool_stats_get_count();
+	return 0;
 }
 
 static void mt7915_ethtool_worker(void *wi_data, struct ieee80211_sta *sta)
@@ -1402,7 +1399,7 @@ void mt7915_get_et_stats(struct ieee80211_hw *hw,
 	};
 	struct mib_stats *mib = &phy->mib;
 	/* See mt7915_ampdu_stat_read_phy, etc */
-	int i, ei = 0, stats_size;
+	int i, ei = 0;
 
 	mutex_lock(&dev->mt76.mutex);
 
@@ -1483,12 +1480,9 @@ void mt7915_get_et_stats(struct ieee80211_hw *hw,
 		return;
 
 	ei += wi.worker_stat_count;
-
-	mt76_ethtool_page_pool_stats(&dev->mt76, &data[ei], &ei);
-
-	stats_size = MT7915_SSTATS_LEN + page_pool_ethtool_stats_get_count();
-	if (ei != stats_size)
-		dev_err(dev->mt76.dev, "ei: %d size: %d", ei, stats_size);
+	if (ei != MT7915_SSTATS_LEN)
+		dev_err(dev->mt76.dev, "ei: %d  MT7915_SSTATS_LEN: %d",
+			ei, (int)MT7915_SSTATS_LEN);
 }
 
 static void
diff --git a/mt7915/mmio.c b/mt7915/mmio.c
index 87cff5d4..b6adfd0b 100644
--- a/mt7915/mmio.c
+++ b/mt7915/mmio.c
@@ -581,9 +581,14 @@ static void mt7915_mmio_wed_offload_disable(struct mtk_wed_device *wed)
 static void mt7915_mmio_wed_release_rx_buf(struct mtk_wed_device *wed)
 {
 	struct mt7915_dev *dev;
+	struct page *page;
+	u32 length;
 	int i;
 
 	dev = container_of(wed, struct mt7915_dev, mt76.mmio.wed);
+	length = SKB_DATA_ALIGN(NET_SKB_PAD + wed->wlan.rx_size +
+				sizeof(struct skb_shared_info));
+
 	for (i = 0; i < dev->mt76.rx_token_size; i++) {
 		struct mt76_rxwi_cache *r;
 
@@ -591,50 +596,79 @@ static void mt7915_mmio_wed_release_rx_buf(struct mtk_wed_device *wed)
 		if (!r || !r->ptr)
 			continue;
 
-		mt76_put_page_pool_buf(r->ptr, false);
+		dma_unmap_single(dev->mt76.dma_dev, r->dma_addr,
+				 wed->wlan.rx_size, DMA_FROM_DEVICE);
+		skb_free_frag(r->ptr);
 		r->ptr = NULL;
 
 		mt76_put_rxwi(&dev->mt76, r);
 	}
 
 	mt76_free_pending_rxwi(&dev->mt76);
+
+	mt76_for_each_q_rx(&dev->mt76, i) {
+		struct mt76_queue *q = &dev->mt76.q_rx[i];
+
+		if (mt76_queue_is_wed_rx(q)) {
+			if (!q->rx_page.va)
+				continue;
+
+			page = virt_to_page(q->rx_page.va);
+			__page_frag_cache_drain(page, q->rx_page.pagecnt_bias);
+			memset(&q->rx_page, 0, sizeof(q->rx_page));
+		}
+	}
+
+	if (!wed->rx_buf_ring.rx_page.va)
+		return;
+
+	page = virt_to_page(wed->rx_buf_ring.rx_page.va);
+	__page_frag_cache_drain(page, wed->rx_buf_ring.rx_page.pagecnt_bias);
+	memset(&wed->rx_buf_ring.rx_page, 0, sizeof(wed->rx_buf_ring.rx_page));
 }
 
 static u32 mt7915_mmio_wed_init_rx_buf(struct mtk_wed_device *wed, int size)
 {
 	struct mtk_rxbm_desc *desc = wed->rx_buf_ring.desc;
-	struct mt76_rxwi_cache *r = NULL;
 	struct mt7915_dev *dev;
-	struct mt76_queue *q;
-	int i, len;
+	u32 length;
+	int i;
 
 	dev = container_of(wed, struct mt7915_dev, mt76.mmio.wed);
-	q = &dev->mt76.q_rx[MT_RXQ_MAIN];
-	len = SKB_WITH_OVERHEAD(q->buf_size);
+	length = SKB_DATA_ALIGN(NET_SKB_PAD + wed->wlan.rx_size +
+				sizeof(struct skb_shared_info));
 
 	for (i = 0; i < size; i++) {
-		enum dma_data_direction dir;
-		dma_addr_t addr;
-		u32 offset;
+		struct mt76_rxwi_cache *r = mt76_get_rxwi(&dev->mt76);
+		dma_addr_t phy_addr;
 		int token;
-		void *buf;
+		void *ptr;
 
-		r = mt76_get_rxwi(&dev->mt76);
 		if (!r)
 			goto unmap;
 
-		buf = mt76_get_page_pool_buf(q, &offset, q->buf_size);
-		if (!buf)
+		ptr = page_frag_alloc(&wed->rx_buf_ring.rx_page, length, GFP_ATOMIC);
+		if (!ptr) {
+			mt76_put_rxwi(&dev->mt76, r);
 			goto unmap;
+		}
 
-		addr = page_pool_get_dma_addr(virt_to_head_page(buf)) + offset;
-		dir = page_pool_get_dma_dir(q->page_pool);
-		dma_sync_single_for_device(dev->mt76.dma_dev, addr, len, dir);
+		phy_addr = dma_map_single(dev->mt76.dma_dev, ptr,
+					wed->wlan.rx_size,
+					DMA_TO_DEVICE);
 
-		desc->buf0 = cpu_to_le32(addr);
-		token = mt76_rx_token_consume(&dev->mt76, buf, r, addr);
+		if (unlikely(dma_mapping_error(dev->mt76.dev, phy_addr))) {
+			skb_free_frag(ptr);
+			mt76_put_rxwi(&dev->mt76, r);
+			goto unmap;
+		}
+		desc->buf0 = cpu_to_le32(phy_addr);
+		token = mt76_rx_token_consume(&dev->mt76, ptr, r, phy_addr);
 		if (token < 0) {
-			mt76_put_page_pool_buf(buf, false);
+			dma_unmap_single(dev->mt76.dma_dev, phy_addr,
+					 wed->wlan.rx_size, DMA_TO_DEVICE);
+			skb_free_frag(ptr);
+			mt76_put_rxwi(&dev->mt76, r);
 			goto unmap;
 		}
 
@@ -646,8 +680,6 @@ static u32 mt7915_mmio_wed_init_rx_buf(struct mtk_wed_device *wed, int size)
 	return 0;
 
 unmap:
-	if (r)
-		mt76_put_rxwi(&dev->mt76, r);
 	mt7915_mmio_wed_release_rx_buf(wed);
 	return -ENOMEM;
 }
diff --git a/mt7921/main.c b/mt7921/main.c
index 75eaf86c..1434ca79 100644
--- a/mt7921/main.c
+++ b/mt7921/main.c
@@ -1090,34 +1090,17 @@ static void
 mt7921_get_et_strings(struct ieee80211_hw *hw, struct ieee80211_vif *vif,
 		      u32 sset, u8 *data)
 {
-	struct mt7921_dev *dev = mt7921_hw_dev(hw);
-
 	if (sset != ETH_SS_STATS)
 		return;
 
 	memcpy(data, *mt7921_gstrings_stats, sizeof(mt7921_gstrings_stats));
-
-	if (mt76_is_sdio(&dev->mt76))
-		return;
-
-	data += sizeof(mt7921_gstrings_stats);
-	page_pool_ethtool_stats_get_strings(data);
 }
 
 static int
 mt7921_get_et_sset_count(struct ieee80211_hw *hw, struct ieee80211_vif *vif,
 			 int sset)
 {
-	struct mt7921_dev *dev = mt7921_hw_dev(hw);
-
-	if (sset != ETH_SS_STATS)
-		return 0;
-
-	if (mt76_is_sdio(&dev->mt76))
-		return ARRAY_SIZE(mt7921_gstrings_stats);
-
-	return ARRAY_SIZE(mt7921_gstrings_stats) +
-	       page_pool_ethtool_stats_get_count();
+	return sset == ETH_SS_STATS ? ARRAY_SIZE(mt7921_gstrings_stats) : 0;
 }
 
 static void
@@ -1137,7 +1120,6 @@ void mt7921_get_et_stats(struct ieee80211_hw *hw, struct ieee80211_vif *vif,
 			 struct ethtool_stats *stats, u64 *data)
 {
 	struct mt7921_vif *mvif = (struct mt7921_vif *)vif->drv_priv;
-	int stats_size = ARRAY_SIZE(mt7921_gstrings_stats);
 	struct mt7921_phy *phy = mt7921_hw_phy(hw);
 	struct mt7921_dev *dev = phy->dev;
 	struct mib_stats *mib = &phy->mib;
@@ -1193,14 +1175,9 @@ void mt7921_get_et_stats(struct ieee80211_hw *hw, struct ieee80211_vif *vif,
 		return;
 
 	ei += wi.worker_stat_count;
-
-	if (!mt76_is_sdio(&dev->mt76)) {
-		mt76_ethtool_page_pool_stats(&dev->mt76, &data[ei], &ei);
-		stats_size += page_pool_ethtool_stats_get_count();
-	}
-
-	if (ei != stats_size)
-		dev_err(dev->mt76.dev, "ei: %d  SSTATS_LEN: %d", ei, stats_size);
+	if (ei != ARRAY_SIZE(mt7921_gstrings_stats))
+		dev_err(dev->mt76.dev, "ei: %d  SSTATS_LEN: %zu",
+			ei, ARRAY_SIZE(mt7921_gstrings_stats));
 }
 
 static u64
diff --git a/usb.c b/usb.c
index b88959ef..3e281715 100644
--- a/usb.c
+++ b/usb.c
@@ -319,27 +319,29 @@ mt76u_set_endpoints(struct usb_interface *intf,
 
 static int
 mt76u_fill_rx_sg(struct mt76_dev *dev, struct mt76_queue *q, struct urb *urb,
-		 int nsgs)
+		 int nsgs, gfp_t gfp)
 {
 	int i;
 
 	for (i = 0; i < nsgs; i++) {
+		struct page *page;
 		void *data;
 		int offset;
 
-		data = mt76_get_page_pool_buf(q, &offset, q->buf_size);
+		data = page_frag_alloc(&q->rx_page, q->buf_size, gfp);
 		if (!data)
 			break;
 
-		sg_set_page(&urb->sg[i], virt_to_head_page(data), q->buf_size,
-			    offset);
+		page = virt_to_head_page(data);
+		offset = data - page_address(page);
+		sg_set_page(&urb->sg[i], page, q->buf_size, offset);
 	}
 
 	if (i < nsgs) {
 		int j;
 
 		for (j = nsgs; j < urb->num_sgs; j++)
-			mt76_put_page_pool_buf(sg_virt(&urb->sg[j]), false);
+			skb_free_frag(sg_virt(&urb->sg[j]));
 		urb->num_sgs = i;
 	}
 
@@ -352,16 +354,15 @@ mt76u_fill_rx_sg(struct mt76_dev *dev, struct mt76_queue *q, struct urb *urb,
 
 static int
 mt76u_refill_rx(struct mt76_dev *dev, struct mt76_queue *q,
-		struct urb *urb, int nsgs)
+		struct urb *urb, int nsgs, gfp_t gfp)
 {
 	enum mt76_rxq_id qid = q - &dev->q_rx[MT_RXQ_MAIN];
-	int offset;
 
 	if (qid == MT_RXQ_MAIN && dev->usb.sg_en)
-		return mt76u_fill_rx_sg(dev, q, urb, nsgs);
+		return mt76u_fill_rx_sg(dev, q, urb, nsgs, gfp);
 
 	urb->transfer_buffer_length = q->buf_size;
-	urb->transfer_buffer = mt76_get_page_pool_buf(q, &offset, q->buf_size);
+	urb->transfer_buffer = page_frag_alloc(&q->rx_page, q->buf_size, gfp);
 
 	return urb->transfer_buffer ? 0 : -ENOMEM;
 }
@@ -399,7 +400,7 @@ mt76u_rx_urb_alloc(struct mt76_dev *dev, struct mt76_queue *q,
 	if (err)
 		return err;
 
-	return mt76u_refill_rx(dev, q, e->urb, sg_size);
+	return mt76u_refill_rx(dev, q, e->urb, sg_size, GFP_KERNEL);
 }
 
 static void mt76u_urb_free(struct urb *urb)
@@ -407,10 +408,10 @@ static void mt76u_urb_free(struct urb *urb)
 	int i;
 
 	for (i = 0; i < urb->num_sgs; i++)
-		mt76_put_page_pool_buf(sg_virt(&urb->sg[i]), false);
+		skb_free_frag(sg_virt(&urb->sg[i]));
 
 	if (urb->transfer_buffer)
-		mt76_put_page_pool_buf(urb->transfer_buffer, false);
+		skb_free_frag(urb->transfer_buffer);
 
 	usb_free_urb(urb);
 }
@@ -546,8 +547,6 @@ mt76u_process_rx_entry(struct mt76_dev *dev, struct urb *urb,
 		len -= data_len;
 		nsgs++;
 	}
-
-	skb_mark_for_recycle(skb);
 	dev->drv->rx_skb(dev, MT_RXQ_MAIN, skb, NULL);
 
 	return nsgs;
@@ -613,7 +612,7 @@ mt76u_process_rx_queue(struct mt76_dev *dev, struct mt76_queue *q)
 
 		count = mt76u_process_rx_entry(dev, urb, q->buf_size);
 		if (count > 0) {
-			err = mt76u_refill_rx(dev, q, urb, count);
+			err = mt76u_refill_rx(dev, q, urb, count, GFP_ATOMIC);
 			if (err < 0)
 				break;
 		}
@@ -664,10 +663,6 @@ mt76u_alloc_rx_queue(struct mt76_dev *dev, enum mt76_rxq_id qid)
 	struct mt76_queue *q = &dev->q_rx[qid];
 	int i, err;
 
-	err = mt76_create_page_pool(dev, q);
-	if (err)
-		return err;
-
 	spin_lock_init(&q->lock);
 	q->entry = devm_kcalloc(dev->dev,
 				MT_NUM_RX_ENTRIES, sizeof(*q->entry),
@@ -696,6 +691,7 @@ EXPORT_SYMBOL_GPL(mt76u_alloc_mcu_queue);
 static void
 mt76u_free_rx_queue(struct mt76_dev *dev, struct mt76_queue *q)
 {
+	struct page *page;
 	int i;
 
 	for (i = 0; i < q->ndesc; i++) {
@@ -705,7 +701,13 @@ mt76u_free_rx_queue(struct mt76_dev *dev, struct mt76_queue *q)
 		mt76u_urb_free(q->entry[i].urb);
 		q->entry[i].urb = NULL;
 	}
-	page_pool_destroy(q->page_pool);
+
+	if (!q->rx_page.va)
+		return;
+
+	page = virt_to_page(q->rx_page.va);
+	__page_frag_cache_drain(page, q->rx_page.pagecnt_bias);
+	memset(&q->rx_page, 0, sizeof(q->rx_page));
 }
 
 static void mt76u_free_rx(struct mt76_dev *dev)
-- 
2.18.0