From 78b8b86fce85c8ff6fad935bd9bd8b2df0e151ab Mon Sep 17 00:00:00 2001
From: Bo Jiao <Bo.Jiao@mediatek.com>
Date: Mon, 6 Feb 2023 13:37:23 +0800
Subject: [PATCH 65/98] wifi: mt76: rework wed rx flow

Signed-off-by: Bo Jiao <Bo.Jiao@mediatek.com>
Change-Id: Icd787345c811cb5ad30d9c7c1c5f9e5298bd3be6
---
 dma.c           | 125 +++++++++++++++++++++++++++++++-----------------
 mac80211.c      |   2 +-
 mmio.c          |  58 ++++++++++++++--------
 mt76.h          |  23 +++++----
 mt7915/mmio.c   |   3 +-
 mt7915/mt7915.h |   1 +
 tx.c            |  16 +++----
 7 files changed, 143 insertions(+), 85 deletions(-)

diff --git a/dma.c b/dma.c
index 06b76ea..f48ec57 100644
--- a/dma.c
+++ b/dma.c
@@ -64,17 +64,17 @@ mt76_alloc_txwi(struct mt76_dev *dev)
 	return t;
 }
 
-static struct mt76_txwi_cache *
+static struct mt76_rxwi_cache *
 mt76_alloc_rxwi(struct mt76_dev *dev)
 {
-	struct mt76_txwi_cache *t;
+	struct mt76_rxwi_cache *r;
 
-	t = kzalloc(L1_CACHE_ALIGN(sizeof(*t)), GFP_ATOMIC);
-	if (!t)
+	r = kzalloc(L1_CACHE_ALIGN(sizeof(*r)), GFP_ATOMIC);
+	if (!r)
 		return NULL;
 
-	t->ptr = NULL;
-	return t;
+	r->ptr = NULL;
+	return r;
 }
 
 static struct mt76_txwi_cache *
@@ -93,20 +93,20 @@ __mt76_get_txwi(struct mt76_dev *dev)
 	return t;
 }
 
-static struct mt76_txwi_cache *
+static struct mt76_rxwi_cache *
 __mt76_get_rxwi(struct mt76_dev *dev)
 {
-	struct mt76_txwi_cache *t = NULL;
+	struct mt76_rxwi_cache *r = NULL;
 
-	spin_lock(&dev->wed_lock);
+	spin_lock(&dev->lock);
 	if (!list_empty(&dev->rxwi_cache)) {
-		t = list_first_entry(&dev->rxwi_cache, struct mt76_txwi_cache,
+		r = list_first_entry(&dev->rxwi_cache, struct mt76_rxwi_cache,
 				     list);
-		list_del(&t->list);
+		list_del(&r->list);
 	}
-	spin_unlock(&dev->wed_lock);
+	spin_unlock(&dev->lock);
 
-	return t;
+	return r;
 }
 
 static struct mt76_txwi_cache *
@@ -120,13 +120,13 @@ mt76_get_txwi(struct mt76_dev *dev)
 	return mt76_alloc_txwi(dev);
 }
 
-struct mt76_txwi_cache *
+struct mt76_rxwi_cache *
 mt76_get_rxwi(struct mt76_dev *dev)
 {
-	struct mt76_txwi_cache *t = __mt76_get_rxwi(dev);
+	struct mt76_rxwi_cache *r = __mt76_get_rxwi(dev);
 
-	if (t)
-		return t;
+	if (r)
+		return r;
 
 	return mt76_alloc_rxwi(dev);
 }
@@ -145,14 +145,14 @@ mt76_put_txwi(struct mt76_dev *dev, struct mt76_txwi_cache *t)
 EXPORT_SYMBOL_GPL(mt76_put_txwi);
 
 void
-mt76_put_rxwi(struct mt76_dev *dev, struct mt76_txwi_cache *t)
+mt76_put_rxwi(struct mt76_dev *dev, struct mt76_rxwi_cache *r)
 {
-	if (!t)
+	if (!r)
 		return;
 
-	spin_lock(&dev->wed_lock);
-	list_add(&t->list, &dev->rxwi_cache);
-	spin_unlock(&dev->wed_lock);
+	spin_lock(&dev->lock);
+	list_add(&r->list, &dev->rxwi_cache);
+	spin_unlock(&dev->lock);
 }
 EXPORT_SYMBOL_GPL(mt76_put_rxwi);
 
@@ -173,13 +173,13 @@ mt76_free_pending_txwi(struct mt76_dev *dev)
 void
 mt76_free_pending_rxwi(struct mt76_dev *dev)
 {
-	struct mt76_txwi_cache *t;
+	struct mt76_rxwi_cache *r;
 
 	local_bh_disable();
-	while ((t = __mt76_get_rxwi(dev)) != NULL) {
-		if (t->ptr)
-			skb_free_frag(t->ptr);
-		kfree(t);
+	while ((r = __mt76_get_rxwi(dev)) != NULL) {
+		if (r->ptr)
+			skb_free_frag(r->ptr);
+		kfree(r);
 	}
 	local_bh_enable();
 }
@@ -227,10 +227,10 @@ 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_queue_entry *entry = &q->entry[q->head];
-	struct mt76_txwi_cache *txwi = NULL;
 	struct mt76_desc *desc;
 	u32 buf1 = 0, ctrl;
 	int idx = q->head;
@@ -248,13 +248,15 @@ 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)) {
-		txwi = mt76_get_rxwi(dev);
-		if (!txwi)
-			return -ENOMEM;
+		if (!rxwi) {
+			rxwi = mt76_get_rxwi(dev);
+			if (!rxwi)
+				return -ENOMEM;
+		}
 
-		rx_token = mt76_rx_token_consume(dev, data, txwi, buf->addr);
+		rx_token = mt76_rx_token_consume(dev, data, rxwi, buf->addr);
 		if (rx_token < 0) {
-			mt76_put_rxwi(dev, txwi);
+			mt76_put_rxwi(dev, rxwi);
 			return -ENOMEM;
 		}
 
@@ -270,7 +272,7 @@ mt76_dma_add_rx_buf(struct mt76_dev *dev, struct mt76_queue *q,
 done:
 	entry->dma_addr[0] = buf->addr;
 	entry->dma_len[0] = buf->len;
-	entry->txwi = txwi;
+	entry->rxwi = rxwi;
 	entry->buf = data;
 	entry->wcid = 0xffff;
 	entry->skip_buf1 = true;
@@ -412,7 +414,7 @@ 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];
@@ -440,20 +442,53 @@ mt76_dma_get_buf(struct mt76_dev *dev, struct mt76_queue *q, int idx,
 	if (mt76_queue_is_wed_rx(q)) {
 		u32 buf1 = le32_to_cpu(desc->buf1);
 		u32 token = FIELD_GET(MT_DMA_CTL_TOKEN, buf1);
-		struct mt76_txwi_cache *t = mt76_rx_token_release(dev, token);
+		struct mt76_rxwi_cache *r = mt76_rx_token_release(dev, token);
 
-		if (!t)
+		if (!r)
 			return NULL;
 
-		dma_unmap_single(dev->dma_dev, t->dma_addr,
+		dma_unmap_single(dev->dma_dev, r->dma_addr,
 				 SKB_WITH_OVERHEAD(q->buf_size),
 				 DMA_FROM_DEVICE);
 
-		buf = t->ptr;
-		t->dma_addr = 0;
-		t->ptr = NULL;
+		if (flush) {
+			buf = r->ptr;
+			r->dma_addr = 0;
+			r->ptr = NULL;
+
+			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;
+			}
+		}
 
-		mt76_put_rxwi(dev, t);
 		if (drop)
 			*drop |= !!(buf1 & MT_DMA_CTL_WO_DROP);
 	} else {
@@ -490,7 +525,7 @@ 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
@@ -658,7 +693,7 @@ mt76_dma_rx_fill(struct mt76_dev *dev, struct mt76_queue *q)
 done:
 		qbuf.len = len - offset;
 		qbuf.skip_unmap = false;
-		if (mt76_dma_add_rx_buf(dev, q, &qbuf, buf) < 0) {
+		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);
diff --git a/mac80211.c b/mac80211.c
index 3715c73..4552bc2 100644
--- a/mac80211.c
+++ b/mac80211.c
@@ -575,7 +575,6 @@ mt76_alloc_device(struct device *pdev, unsigned int size,
 	spin_lock_init(&dev->lock);
 	spin_lock_init(&dev->cc_lock);
 	spin_lock_init(&dev->status_lock);
-	spin_lock_init(&dev->wed_lock);
 	mutex_init(&dev->mutex);
 	init_waitqueue_head(&dev->tx_wait);
 
@@ -608,6 +607,7 @@ mt76_alloc_device(struct device *pdev, unsigned int size,
 	INIT_LIST_HEAD(&dev->txwi_cache);
 	INIT_LIST_HEAD(&dev->rxwi_cache);
 	dev->token_size = dev->drv->token_size;
+	dev->rx_token_size = dev->drv->rx_token_size;
 
 	for (i = 0; i < ARRAY_SIZE(dev->q_rx); i++)
 		skb_queue_head_init(&dev->rx_skb[i]);
diff --git a/mmio.c b/mmio.c
index 5fb8392..f7495f6 100644
--- a/mmio.c
+++ b/mmio.c
@@ -89,28 +89,45 @@ EXPORT_SYMBOL_GPL(mt76_set_irq_mask);
 void mt76_mmio_wed_release_rx_buf(struct mtk_wed_device *wed)
 {
 	struct mt76_dev *dev = container_of(wed, struct mt76_dev, mmio.wed);
-	u32 length;
+	struct page *page;
 	int i;
 
-	length = SKB_DATA_ALIGN(NET_SKB_PAD + wed->wlan.rx_size +
-				sizeof(struct skb_shared_info));
-
 	for (i = 0; i < dev->rx_token_size; i++) {
-		struct mt76_txwi_cache *t;
+		struct mt76_rxwi_cache *r;
 
-		t = mt76_rx_token_release(dev, i);
-		if (!t || !t->ptr)
+		r = mt76_rx_token_release(dev, i);
+		if (!r || !r->ptr)
 			continue;
 
-		dma_unmap_single(dev->dma_dev, t->dma_addr,
+		dma_unmap_single(dev->dma_dev, r->dma_addr,
 				 wed->wlan.rx_size, DMA_FROM_DEVICE);
-		__free_pages(virt_to_page(t->ptr), get_order(length));
-		t->ptr = NULL;
+		skb_free_frag(r->ptr);
+		r->ptr = NULL;
 
-		mt76_put_rxwi(dev, t);
+		mt76_put_rxwi(dev, r);
 	}
 
 	mt76_free_pending_rxwi(dev);
+
+	mt76_for_each_q_rx(dev, i) {
+		struct mt76_queue *q = &dev->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));
 }
 EXPORT_SYMBOL_GPL(mt76_mmio_wed_release_rx_buf);
 
@@ -125,18 +142,17 @@ u32 mt76_mmio_wed_init_rx_buf(struct mtk_wed_device *wed, int size)
 				sizeof(struct skb_shared_info));
 
 	for (i = 0; i < size; i++) {
-		struct mt76_txwi_cache *t = mt76_get_rxwi(dev);
+		struct mt76_rxwi_cache *r = mt76_get_rxwi(dev);
 		dma_addr_t phy_addr;
-		struct page *page;
 		int token;
 		void *ptr;
 
-		if (!t)
+		if (!r)
 			goto unmap;
 
-		page = __dev_alloc_pages(GFP_KERNEL, get_order(length));
-		if (!page) {
-			mt76_put_rxwi(dev, t);
+		ptr = page_frag_alloc(&wed->rx_buf_ring.rx_page, length, GFP_ATOMIC);
+		if (!ptr) {
+			mt76_put_rxwi(dev, r);
  			goto unmap;
 		}
 
@@ -146,17 +162,17 @@ u32 mt76_mmio_wed_init_rx_buf(struct mtk_wed_device *wed, int size)
 
 		if (unlikely(dma_mapping_error(dev->dev, phy_addr))) {
 			skb_free_frag(ptr);
-			mt76_put_rxwi(dev, t);
+			mt76_put_rxwi(dev, r);
 			goto unmap;
 		}
 
 		desc->buf0 = cpu_to_le32(phy_addr);
-		token = mt76_rx_token_consume(dev, ptr, t, phy_addr);
+		token = mt76_rx_token_consume(dev, ptr, r, phy_addr);
 		if (token < 0) {
 			dma_unmap_single(dev->dma_dev, phy_addr,
 					 wed->wlan.rx_size, DMA_TO_DEVICE);
-			__free_pages(page, get_order(length));
-			mt76_put_rxwi(dev, t);
+			skb_free_frag(ptr);
+			mt76_put_rxwi(dev, r);
 			goto unmap;
 		}
 
diff --git a/mt76.h b/mt76.h
index 3af97e5..b960f3d 100644
--- a/mt76.h
+++ b/mt76.h
@@ -200,6 +200,7 @@ struct mt76_queue_entry {
 	};
 	union {
 		struct mt76_txwi_cache *txwi;
+		struct mt76_rxwi_cache *rxwi;
 		struct urb *urb;
 		int buf_sz;
 	};
@@ -410,10 +411,14 @@ struct mt76_txwi_cache {
 	struct list_head list;
 	dma_addr_t dma_addr;
 
-	union {
-		struct sk_buff *skb;
-		void *ptr;
-	};
+	struct sk_buff *skb;
+};
+
+struct mt76_rxwi_cache {
+	struct list_head list;
+	dma_addr_t dma_addr;
+
+	void *ptr;
 };
 
 struct mt76_rx_tid {
@@ -499,6 +504,7 @@ struct mt76_driver_ops {
 	u16 txwi_size;
 	u16 token_size;
 	u8 mcs_rates;
+	u16 rx_token_size;
 
 	void (*update_survey)(struct mt76_phy *phy);
 
@@ -858,7 +864,6 @@ struct mt76_dev {
 
 	struct ieee80211_hw *hw;
 
-	spinlock_t wed_lock;
 	spinlock_t lock;
 	spinlock_t cc_lock;
 
@@ -1521,8 +1526,8 @@ mt76_tx_status_get_hw(struct mt76_dev *dev, struct sk_buff *skb)
 }
 
 void mt76_put_txwi(struct mt76_dev *dev, struct mt76_txwi_cache *t);
-void mt76_put_rxwi(struct mt76_dev *dev, struct mt76_txwi_cache *t);
-struct mt76_txwi_cache *mt76_get_rxwi(struct mt76_dev *dev);
+void mt76_put_rxwi(struct mt76_dev *dev, struct mt76_rxwi_cache *r);
+struct mt76_rxwi_cache *mt76_get_rxwi(struct mt76_dev *dev);
 void mt76_free_pending_rxwi(struct mt76_dev *dev);
 void mt76_rx_complete(struct mt76_dev *dev, struct sk_buff_head *frames,
 		      struct napi_struct *napi);
@@ -1703,9 +1708,9 @@ struct mt76_txwi_cache *
 mt76_token_release(struct mt76_dev *dev, int token, bool *wake);
 int mt76_token_consume(struct mt76_dev *dev, struct mt76_txwi_cache **ptxwi);
 void __mt76_set_tx_blocked(struct mt76_dev *dev, bool blocked);
-struct mt76_txwi_cache *mt76_rx_token_release(struct mt76_dev *dev, int token);
+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_txwi_cache *r, dma_addr_t phys);
+			  struct mt76_rxwi_cache *r, dma_addr_t phys);
 
 static inline void mt76_set_tx_blocked(struct mt76_dev *dev, bool blocked)
 {
diff --git a/mt7915/mmio.c b/mt7915/mmio.c
index 85cb3fe..690cac5 100644
--- a/mt7915/mmio.c
+++ b/mt7915/mmio.c
@@ -687,7 +687,7 @@ int mt7915_mmio_wed_init(struct mt7915_dev *dev, void *pdev_ptr,
 	wed->wlan.reset = mt7915_mmio_wed_reset;
 	wed->wlan.reset_complete = mt7915_mmio_wed_reset_complete;
 
-	dev->mt76.rx_token_size = wed->wlan.rx_npkt;
+	dev->mt76.rx_token_size += wed->wlan.rx_npkt;
 
 	if (mtk_wed_device_attach(wed))
 		return 0;
@@ -893,6 +893,7 @@ struct mt7915_dev *mt7915_mmio_probe(struct device *pdev,
 				SURVEY_INFO_TIME_RX |
 				SURVEY_INFO_TIME_BSS_RX,
 		.token_size = MT7915_TOKEN_SIZE,
+		.rx_token_size = MT7915_RX_TOKEN_SIZE;
 		.tx_prepare_skb = mt7915_tx_prepare_skb,
 		.tx_complete_skb = mt76_connac_tx_complete_skb,
 		.rx_skb = mt7915_queue_rx_skb,
diff --git a/mt7915/mt7915.h b/mt7915/mt7915.h
index d317c52..91eb5ad 100644
--- a/mt7915/mt7915.h
+++ b/mt7915/mt7915.h
@@ -62,6 +62,7 @@
 #define MT7915_EEPROM_BLOCK_SIZE	16
 #define MT7915_HW_TOKEN_SIZE		4096
 #define MT7915_TOKEN_SIZE		8192
+#define MT7915_RX_TOKEN_SIZE		4096
 
 #define MT7915_CFEND_RATE_DEFAULT	0x49	/* OFDM 24M */
 #define MT7915_CFEND_RATE_11B		0x03	/* 11B LP, 11M */
diff --git a/tx.c b/tx.c
index 1809b03..74bf0de 100644
--- a/tx.c
+++ b/tx.c
@@ -843,16 +843,16 @@ int mt76_token_consume(struct mt76_dev *dev, struct mt76_txwi_cache **ptxwi)
 EXPORT_SYMBOL_GPL(mt76_token_consume);
 
 int mt76_rx_token_consume(struct mt76_dev *dev, void *ptr,
-			  struct mt76_txwi_cache *t, dma_addr_t phys)
+			  struct mt76_rxwi_cache *r, dma_addr_t phys)
 {
 	int token;
 
 	spin_lock_bh(&dev->rx_token_lock);
-	token = idr_alloc(&dev->rx_token, t, 0, dev->rx_token_size,
+	token = idr_alloc(&dev->rx_token, r, 0, dev->rx_token_size,
 			  GFP_ATOMIC);
 	if (token >= 0) {
-		t->ptr = ptr;
-		t->dma_addr = phys;
+		r->ptr = ptr;
+		r->dma_addr = phys;
 	}
 	spin_unlock_bh(&dev->rx_token_lock);
 
@@ -889,15 +889,15 @@ mt76_token_release(struct mt76_dev *dev, int token, bool *wake)
 }
 EXPORT_SYMBOL_GPL(mt76_token_release);
 
-struct mt76_txwi_cache *
+struct mt76_rxwi_cache *
 mt76_rx_token_release(struct mt76_dev *dev, int token)
 {
-	struct mt76_txwi_cache *t;
+	struct mt76_rxwi_cache *r;
 
 	spin_lock_bh(&dev->rx_token_lock);
-	t = idr_remove(&dev->rx_token, token);
+	r = idr_remove(&dev->rx_token, token);
 	spin_unlock_bh(&dev->rx_token_lock);
 
-	return t;
+	return r;
 }
 EXPORT_SYMBOL_GPL(mt76_rx_token_release);
-- 
2.18.0