Skip to contentSkip to conversation

net: ethernet: mtk_eth_soc: handle dma buffer size soc specific

PATCH Applied Frank Wunderlich 4 participants 4 messages Jun 3, 2024 – Jun 6, 2024

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 net: ethernet: mtk_eth_soc: handle dma buffer size soc specific

From: Frank Wunderlich <[email protected]>

The mainline MTK ethernet driver suffers long time from rarly but annoying tx queue timeouts. We think that this is caused by fixed dma sizes hardcoded for all SoCs.

We suspect this problem arises from a low level of free TX DMADs, the TX Ring alomost full.

The transmit timeout is caused by the Tx queue not waking up. The Tx queue stops when the free counter is less than ring->thres, and it will wake up once the free counter is greater than ring->thres. If the CPU is too late to wake up the Tx queues, it may cause a transmit timeout.
Therefore, we increased the TX and RX DMADs to improve this error situation.

Use the dma-size implementation from SDK in a per SoC manner. In difference to SDK we have no RSS feature yet, so all RX/TX sizes should be raised from 512 to 2048 byte except fqdma on mt7988 to avoid the tx timeout issue.

Fixes 656e705243fd ("net-next: mediatek: add support for MT7623 ethernet")

---
based on SDK:

https://git01.mediatek.com/plugins/gitiles/openwrt/feeds/mtk-openwrt-feeds/+/fac194d6253d339e15c651c052b532a449a04d6e

v3:
- rephrase commit message with some information from mtk
- change the TX/RX DMA Size from 512 to 2048 for all platforms. When rss gets upstream TX Size should be increases to 4096 and the RX Size decreased to 1024 for MT798[1/6/8].
- drop fq_dma_size from rt5350/mt7628 as this does not have QDMA v2:
- fix unused variable 'addr' in 32bit build

---
 drivers/net/ethernet/mediatek/mtk_eth_soc.c | 104 +++++++++++++-------
 drivers/net/ethernet/mediatek/mtk_eth_soc.h |   9 +-
 2 files changed, 77 insertions(+), 36 deletions(-)
drivers/net/ethernet/mediatek/mtk_eth_soc.c +70 -34
index cae46290a7ae..c84ce54a84a0 100644
@@ -1131,9 +1131,9 @@ static int mtk_init_fq_dma(struct mtk_eth *eth)
 {
 	const struct mtk_soc_data *soc = eth->soc;
 	dma_addr_t phy_ring_tail;
-	int cnt = MTK_QDMA_RING_SIZE;
+	int cnt = soc->tx.fq_dma_size;
 	dma_addr_t dma_addr;
-	int i;
+	int i, j, len;
 
 	if (MTK_HAS_CAPS(eth->soc->caps, MTK_SRAM))
 		eth->scratch_ring = eth->sram_base;
@@ -1142,40 +1142,46 @@ static int mtk_init_fq_dma(struct mtk_eth *eth)
 						       cnt * soc->tx.desc_size,
 						       &eth->phy_scratch_ring,
 						       GFP_KERNEL);
+
 	if (unlikely(!eth->scratch_ring))
 		return -ENOMEM;
 
-	eth->scratch_head = kcalloc(cnt, MTK_QDMA_PAGE_SIZE, GFP_KERNEL);
-	if (unlikely(!eth->scratch_head))
-		return -ENOMEM;
+	phy_ring_tail = eth->phy_scratch_ring + soc->tx.desc_size * (cnt - 1);
 
-	dma_addr = dma_map_single(eth->dma_dev,
-				  eth->scratch_head, cnt * MTK_QDMA_PAGE_SIZE,
-				  DMA_FROM_DEVICE);
-	if (unlikely(dma_mapping_error(eth->dma_dev, dma_addr)))
-		return -ENOMEM;
+	for (j = 0; j < DIV_ROUND_UP(soc->tx.fq_dma_size, MTK_FQ_DMA_LENGTH); j++) {
+		len = min_t(int, cnt - j * MTK_FQ_DMA_LENGTH, MTK_FQ_DMA_LENGTH);
+		eth->scratch_head[j] = kcalloc(len, MTK_QDMA_PAGE_SIZE, GFP_KERNEL);
 
-	phy_ring_tail = eth->phy_scratch_ring + soc->tx.desc_size * (cnt - 1);
+		if (unlikely(!eth->scratch_head[j]))
+			return -ENOMEM;
 
-	for (i = 0; i < cnt; i++) {
-		dma_addr_t addr = dma_addr + i * MTK_QDMA_PAGE_SIZE;
-		struct mtk_tx_dma_v2 *txd;
+		dma_addr = dma_map_single(eth->dma_dev,
+					  eth->scratch_head[j], len * MTK_QDMA_PAGE_SIZE,
+					  DMA_FROM_DEVICE);
 
-		txd = eth->scratch_ring + i * soc->tx.desc_size;
-		txd->txd1 = addr;
-		if (i < cnt - 1)
-			txd->txd2 = eth->phy_scratch_ring +
-				    (i + 1) * soc->tx.desc_size;
+		if (unlikely(dma_mapping_error(eth->dma_dev, dma_addr)))
+			return -ENOMEM;
 
-		txd->txd3 = TX_DMA_PLEN0(MTK_QDMA_PAGE_SIZE);
-		if (MTK_HAS_CAPS(soc->caps, MTK_36BIT_DMA))
-			txd->txd3 |= TX_DMA_PREP_ADDR64(addr);
-		txd->txd4 = 0;
-		if (mtk_is_netsys_v2_or_greater(eth)) {
-			txd->txd5 = 0;
-			txd->txd6 = 0;
-			txd->txd7 = 0;
-			txd->txd8 = 0;
+		for (i = 0; i < cnt; i++) {
+			struct mtk_tx_dma_v2 *txd;
+
+			txd = eth->scratch_ring + (j * MTK_FQ_DMA_LENGTH + i) * soc->tx.desc_size;
+			txd->txd1 = dma_addr + i * MTK_QDMA_PAGE_SIZE;
+			if (j * MTK_FQ_DMA_LENGTH + i < cnt)
+				txd->txd2 = eth->phy_scratch_ring +
+					    (j * MTK_FQ_DMA_LENGTH + i + 1) * soc->tx.desc_size;
+
+			txd->txd3 = TX_DMA_PLEN0(MTK_QDMA_PAGE_SIZE);
+			if (MTK_HAS_CAPS(soc->caps, MTK_36BIT_DMA))
+				txd->txd3 |= TX_DMA_PREP_ADDR64(dma_addr + i * MTK_QDMA_PAGE_SIZE);
+
+			txd->txd4 = 0;
+			if (mtk_is_netsys_v2_or_greater(eth)) {
+				txd->txd5 = 0;
+				txd->txd6 = 0;
+				txd->txd7 = 0;
+				txd->txd8 = 0;
+			}
 		}
 	}
 
@@ -2457,7 +2463,7 @@ static int mtk_tx_alloc(struct mtk_eth *eth)
 	if (MTK_HAS_CAPS(soc->caps, MTK_QDMA))
 		ring_size = MTK_QDMA_RING_SIZE;
 	else
-		ring_size = MTK_DMA_SIZE;
+		ring_size = soc->tx.dma_size;
 
 	ring->buf = kcalloc(ring_size, sizeof(*ring->buf),
 			       GFP_KERNEL);
@@ -2465,8 +2471,8 @@ static int mtk_tx_alloc(struct mtk_eth *eth)
 		goto no_tx_mem;
 
 	if (MTK_HAS_CAPS(soc->caps, MTK_SRAM)) {
-		ring->dma = eth->sram_base + ring_size * sz;
-		ring->phys = eth->phy_scratch_ring + ring_size * (dma_addr_t)sz;
+		ring->dma = eth->sram_base + soc->tx.fq_dma_size * sz;
+		ring->phys = eth->phy_scratch_ring + soc->tx.fq_dma_size * (dma_addr_t)sz;
 	} else {
 		ring->dma = dma_alloc_coherent(eth->dma_dev, ring_size * sz,
 					       &ring->phys, GFP_KERNEL);
@@ -2588,6 +2594,7 @@ static void mtk_tx_clean(struct mtk_eth *eth)
 static int mtk_rx_alloc(struct mtk_eth *eth, int ring_no, int rx_flag)
 {
 	const struct mtk_reg_map *reg_map = eth->soc->reg_map;
+	const struct mtk_soc_data *soc = eth->soc;
 	struct mtk_rx_ring *ring;
 	int rx_data_len, rx_dma_size, tx_ring_size;
 	int i;
@@ -2595,7 +2602,7 @@ static int mtk_rx_alloc(struct mtk_eth *eth, int ring_no, int rx_flag)
 	if (MTK_HAS_CAPS(eth->soc->caps, MTK_QDMA))
 		tx_ring_size = MTK_QDMA_RING_SIZE;
 	else
-		tx_ring_size = MTK_DMA_SIZE;
+		tx_ring_size = soc->tx.dma_size;
 
 	if (rx_flag == MTK_RX_FLAGS_QDMA) {
 		if (ring_no)
@@ -2610,7 +2617,7 @@ static int mtk_rx_alloc(struct mtk_eth *eth, int ring_no, int rx_flag)
 		rx_dma_size = MTK_HW_LRO_DMA_SIZE;
 	} else {
 		rx_data_len = ETH_DATA_LEN;
-		rx_dma_size = MTK_DMA_SIZE;
+		rx_dma_size = soc->rx.dma_size;
 	}
 
 	ring->frag_size = mtk_max_frag_size(rx_data_len);
@@ -3139,7 +3146,10 @@ static void mtk_dma_free(struct mtk_eth *eth)
 			mtk_rx_clean(eth, &eth->rx_ring[i], false);
 	}
 
-	kfree(eth->scratch_head);
+	for (i = 0; i < DIV_ROUND_UP(soc->tx.fq_dma_size, MTK_FQ_DMA_LENGTH); i++) {
+		kfree(eth->scratch_head[i]);
+		eth->scratch_head[i] = NULL;
+	}
 }
 
 static bool mtk_hw_reset_check(struct mtk_eth *eth)
@@ -5052,11 +5062,14 @@ static const struct mtk_soc_data mt2701_data = {
 		.desc_size = sizeof(struct mtk_tx_dma),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
 		.irq_done_mask = MTK_RX_DONE_INT,
 		.dma_l4_valid = RX_DMA_L4_VALID,
+		.dma_size = MTK_DMA_SIZE(2K),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
 	},
@@ -5076,11 +5089,14 @@ static const struct mtk_soc_data mt7621_data = {
 		.desc_size = sizeof(struct mtk_tx_dma),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
 		.irq_done_mask = MTK_RX_DONE_INT,
 		.dma_l4_valid = RX_DMA_L4_VALID,
+		.dma_size = MTK_DMA_SIZE(2K),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
 	},
@@ -5102,11 +5118,14 @@ static const struct mtk_soc_data mt7622_data = {
 		.desc_size = sizeof(struct mtk_tx_dma),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
 		.irq_done_mask = MTK_RX_DONE_INT,
 		.dma_l4_valid = RX_DMA_L4_VALID,
+		.dma_size = MTK_DMA_SIZE(2K),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
 	},
@@ -5127,11 +5146,14 @@ static const struct mtk_soc_data mt7623_data = {
 		.desc_size = sizeof(struct mtk_tx_dma),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
 		.irq_done_mask = MTK_RX_DONE_INT,
 		.dma_l4_valid = RX_DMA_L4_VALID,
+		.dma_size = MTK_DMA_SIZE(2K),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
 	},
@@ -5150,11 +5172,14 @@ static const struct mtk_soc_data mt7629_data = {
 		.desc_size = sizeof(struct mtk_tx_dma),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
 		.irq_done_mask = MTK_RX_DONE_INT,
 		.dma_l4_valid = RX_DMA_L4_VALID,
+		.dma_size = MTK_DMA_SIZE(2K),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
 	},
@@ -5176,6 +5201,8 @@ static const struct mtk_soc_data mt7981_data = {
 		.desc_size = sizeof(struct mtk_tx_dma_v2),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
 		.dma_len_offset = 8,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
@@ -5183,6 +5210,7 @@ static const struct mtk_soc_data mt7981_data = {
 		.dma_l4_valid = RX_DMA_L4_VALID_V2,
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
 	},
 };
 
@@ -5202,6 +5230,8 @@ static const struct mtk_soc_data mt7986_data = {
 		.desc_size = sizeof(struct mtk_tx_dma_v2),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
 		.dma_len_offset = 8,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
@@ -5209,6 +5239,7 @@ static const struct mtk_soc_data mt7986_data = {
 		.dma_l4_valid = RX_DMA_L4_VALID_V2,
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
 	},
 };
 
@@ -5228,6 +5259,8 @@ static const struct mtk_soc_data mt7988_data = {
 		.desc_size = sizeof(struct mtk_tx_dma_v2),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
 		.dma_len_offset = 8,
+		.dma_size = MTK_DMA_SIZE(2K),
+		.fq_dma_size = MTK_DMA_SIZE(4K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma_v2),
@@ -5235,6 +5268,7 @@ static const struct mtk_soc_data mt7988_data = {
 		.dma_l4_valid = RX_DMA_L4_VALID_V2,
 		.dma_max_len = MTK_TX_DMA_BUF_LEN_V2,
 		.dma_len_offset = 8,
+		.dma_size = MTK_DMA_SIZE(2K),
 	},
 };
 
@@ -5249,6 +5283,7 @@ static const struct mtk_soc_data rt5350_data = {
 		.desc_size = sizeof(struct mtk_tx_dma),
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
 	},
 	.rx = {
 		.desc_size = sizeof(struct mtk_rx_dma),
@@ -5256,6 +5291,7 @@ static const struct mtk_soc_data rt5350_data = {
 		.dma_l4_valid = RX_DMA_L4_VALID_PDMA,
 		.dma_max_len = MTK_TX_DMA_BUF_LEN,
 		.dma_len_offset = 16,
+		.dma_size = MTK_DMA_SIZE(2K),
 	},
 };
 
drivers/net/ethernet/mediatek/mtk_eth_soc.h +7 -2
index 4eab30b44070..f5174f6cb1bb 100644
@@ -32,7 +32,9 @@
 #define MTK_TX_DMA_BUF_LEN	0x3fff
 #define MTK_TX_DMA_BUF_LEN_V2	0xffff
 #define MTK_QDMA_RING_SIZE	2048
-#define MTK_DMA_SIZE		512
+#define MTK_DMA_SIZE(x)		(SZ_##x)
+#define MTK_FQ_DMA_HEAD		32
+#define MTK_FQ_DMA_LENGTH	2048
 #define MTK_RX_ETH_HLEN		(ETH_HLEN + ETH_FCS_LEN)
 #define MTK_RX_HLEN		(NET_SKB_PAD + MTK_RX_ETH_HLEN + NET_IP_ALIGN)
 #define MTK_DMA_DUMMY_DESC	0xffffffff
@@ -1176,6 +1178,8 @@ struct mtk_soc_data {
 		u32	desc_size;
 		u32	dma_max_len;
 		u32	dma_len_offset;
+		u32	dma_size;
+		u32	fq_dma_size;
 	} tx;
 	struct {
 		u32	desc_size;
@@ -1183,6 +1187,7 @@ struct mtk_soc_data {
 		u32	dma_l4_valid;
 		u32	dma_max_len;
 		u32	dma_len_offset;
+		u32	dma_size;
 	} rx;
 };
 
@@ -1264,7 +1269,7 @@ struct mtk_eth {
 	struct napi_struct		rx_napi;
 	void				*scratch_ring;
 	dma_addr_t			phy_scratch_ring;
-	void				*scratch_head;
+	void				*scratch_head[MTK_FQ_DMA_HEAD];
 	struct clk			*clks[MTK_CLK_MAX];
 
 	struct mii_bus			*mii_bus;
Signature · 1 line
2.34.1
Reviewed-by on v3 1/1
#1 lore raw
Quoted 60 lines · Frank Wunderlich
drivers/net/ethernet/mediatek/mtk_eth_soc.c +22 -26
@@ -1142,40 +1142,46 @@ static int mtk_init_fq_dma(struct mtk_eth *eth)
 						       cnt * soc->tx.desc_size,
 						       &eth->phy_scratch_ring,
 						       GFP_KERNEL);
+
 	if (unlikely(!eth->scratch_ring))
 		return -ENOMEM;
 
-	eth->scratch_head = kcalloc(cnt, MTK_QDMA_PAGE_SIZE, GFP_KERNEL);
-	if (unlikely(!eth->scratch_head))
-		return -ENOMEM;
+	phy_ring_tail = eth->phy_scratch_ring + soc->tx.desc_size * (cnt - 1);
 
-	dma_addr = dma_map_single(eth->dma_dev,
-				  eth->scratch_head, cnt * MTK_QDMA_PAGE_SIZE,
-				  DMA_FROM_DEVICE);
-	if (unlikely(dma_mapping_error(eth->dma_dev, dma_addr)))
-		return -ENOMEM;
+	for (j = 0; j < DIV_ROUND_UP(soc->tx.fq_dma_size, MTK_FQ_DMA_LENGTH); j++) {
+		len = min_t(int, cnt - j * MTK_FQ_DMA_LENGTH, MTK_FQ_DMA_LENGTH);
+		eth->scratch_head[j] = kcalloc(len, MTK_QDMA_PAGE_SIZE, GFP_KERNEL);
 
-	phy_ring_tail = eth->phy_scratch_ring + soc->tx.desc_size * (cnt - 1);
+		if (unlikely(!eth->scratch_head[j]))
+			return -ENOMEM;
 
-	for (i = 0; i < cnt; i++) {
-		dma_addr_t addr = dma_addr + i * MTK_QDMA_PAGE_SIZE;
-		struct mtk_tx_dma_v2 *txd;
+		dma_addr = dma_map_single(eth->dma_dev,
+					  eth->scratch_head[j], len * MTK_QDMA_PAGE_SIZE,
+					  DMA_FROM_DEVICE);
 
-		txd = eth->scratch_ring + i * soc->tx.desc_size;
-		txd->txd1 = addr;
-		if (i < cnt - 1)
-			txd->txd2 = eth->phy_scratch_ring +
-				    (i + 1) * soc->tx.desc_size;
+		if (unlikely(dma_mapping_error(eth->dma_dev, dma_addr)))
+			return -ENOMEM;
 
-		txd->txd3 = TX_DMA_PLEN0(MTK_QDMA_PAGE_SIZE);
-		if (MTK_HAS_CAPS(soc->caps, MTK_36BIT_DMA))
-			txd->txd3 |= TX_DMA_PREP_ADDR64(addr);
-		txd->txd4 = 0;
-		if (mtk_is_netsys_v2_or_greater(eth)) {
-			txd->txd5 = 0;
-			txd->txd6 = 0;
-			txd->txd7 = 0;
-			txd->txd8 = 0;
+		for (i = 0; i < cnt; i++) {
+			struct mtk_tx_dma_v2 *txd;
+
+			txd = eth->scratch_ring + (j * MTK_FQ_DMA_LENGTH + i) * soc->tx.desc_size;
+			txd->txd1 = dma_addr + i * MTK_QDMA_PAGE_SIZE;
+			if (j * MTK_FQ_DMA_LENGTH + i < cnt)
+				txd->txd2 = eth->phy_scratch_ring +
+					    (j * MTK_FQ_DMA_LENGTH + i + 1) * soc->tx.desc_size;
+
+			txd->txd3 = TX_DMA_PLEN0(MTK_QDMA_PAGE_SIZE);

drivers/net/ethernet/mediatek/mtk_eth_soc.c static int mtk_init_fq_dma(struct mtk_eth *eth) · line 1184

Quoted 10 lines
drivers/net/ethernet/mediatek/mtk_eth_soc.c +10 -0
+			if (MTK_HAS_CAPS(soc->caps, MTK_36BIT_DMA))
+				txd->txd3 |= TX_DMA_PREP_ADDR64(dma_addr + i * MTK_QDMA_PAGE_SIZE);
+
+			txd->txd4 = 0;
+			if (mtk_is_netsys_v2_or_greater(eth)) {
+				txd->txd5 = 0;
+				txd->txd6 = 0;
+				txd->txd7 = 0;
+				txd->txd8 = 0;
+			}

This block of change was a bit hard to understand what was going on, but I think I get the result is that you end up allocating different set of scratch_head per size vs the original only having one scratch_head per device?

Perhaps you can explain, but we're now allocating a bunch of different scratch_head pointers.. However, in the patch, the only places that we modify scratch_head appear to be the allocation path and the free path.. but I can't seem to understand how that would impact the users of scratch head? I guess it changes the dma_addr which then changes the txd values we program?

Ok.

I sort of understand whats going on here, but it was a fair bit to fully grok this flow.

Overall, I'm no expert on the part or DMA here, but:

#2 lore raw
Quoted 92 lines · Jacob Keller

External email : Please do not click links or open attachments until you have verified the sender or the content.

Review static int mtk_init_fq_dma(struct mtk_eth

Quoted 1 line · Frank Wunderlich
(quoted diff) +0 -0
@@ -1142,40 +1142,46 @@ static int mtk_init_fq_dma(struct mtk_eth

*eth)

drivers/net/ethernet/mediatek/mtk_eth_soc.c static int mtk_init_fq_dma(struct mtk_eth *eth)

Quoted 11 lines
drivers/net/ethernet/mediatek/mtk_eth_soc.c +2 -3
        cnt * soc->tx.desc_size,
        &eth->phy_scratch_ring,
        GFP_KERNEL);
+
 if (unlikely(!eth->scratch_ring))
 return -ENOMEM;
 
-eth->scratch_head = kcalloc(cnt, MTK_QDMA_PAGE_SIZE, GFP_KERNEL);
-if (unlikely(!eth->scratch_head))
-return -ENOMEM;
+phy_ring_tail = eth->phy_scratch_ring + soc->tx.desc_size * (cnt -

1);

drivers/net/ethernet/mediatek/mtk_eth_soc.c static int mtk_init_fq_dma(struct mtk_eth *eth)

Quoted 7 lines
drivers/net/ethernet/mediatek/mtk_eth_soc.c +1 -5
-dma_addr = dma_map_single(eth->dma_dev,
-  eth->scratch_head, cnt * MTK_QDMA_PAGE_SIZE,
-  DMA_FROM_DEVICE);
-if (unlikely(dma_mapping_error(eth->dma_dev, dma_addr)))
-return -ENOMEM;
+for (j = 0; j < DIV_ROUND_UP(soc->tx.fq_dma_size,

MTK_FQ_DMA_LENGTH); j++) {

Review static int mtk_init_fq_dma(struct mtk_eth · line 1152

Quoted 2 lines
(quoted diff) +2 -0
+len = min_t(int, cnt - j * MTK_FQ_DMA_LENGTH, MTK_FQ_DMA_LENGTH);
+eth->scratch_head[j] = kcalloc(len, MTK_QDMA_PAGE_SIZE,

GFP_KERNEL);

Review static int mtk_init_fq_dma(struct mtk_eth · line 1156

Quoted 2 lines
(quoted diff) +0 -1
-phy_ring_tail = eth->phy_scratch_ring + soc->tx.desc_size * (cnt -

1);

Quoted 30 lines
drivers/net/ethernet/mediatek/mtk_eth_soc.c +10 -17
+if (unlikely(!eth->scratch_head[j]))
+return -ENOMEM;
 
-for (i = 0; i < cnt; i++) {
-dma_addr_t addr = dma_addr + i * MTK_QDMA_PAGE_SIZE;
-struct mtk_tx_dma_v2 *txd;
+dma_addr = dma_map_single(eth->dma_dev,
+  eth->scratch_head[j], len * MTK_QDMA_PAGE_SIZE,
+  DMA_FROM_DEVICE);
 
-txd = eth->scratch_ring + i * soc->tx.desc_size;
-txd->txd1 = addr;
-if (i < cnt - 1)
-txd->txd2 = eth->phy_scratch_ring +
-    (i + 1) * soc->tx.desc_size;
+if (unlikely(dma_mapping_error(eth->dma_dev, dma_addr)))
+return -ENOMEM;
 
-txd->txd3 = TX_DMA_PLEN0(MTK_QDMA_PAGE_SIZE);
-if (MTK_HAS_CAPS(soc->caps, MTK_36BIT_DMA))
-txd->txd3 |= TX_DMA_PREP_ADDR64(addr);
-txd->txd4 = 0;
-if (mtk_is_netsys_v2_or_greater(eth)) {
-txd->txd5 = 0;
-txd->txd6 = 0;
-txd->txd7 = 0;
-txd->txd8 = 0;
+for (i = 0; i < cnt; i++) {
+struct mtk_tx_dma_v2 *txd;
+

drivers/net/ethernet/mediatek/mtk_eth_soc.c static int mtk_init_fq_dma(struct mtk_eth *eth)

Quoted 10 lines
drivers/net/ethernet/mediatek/mtk_eth_soc.c +9 -0
+txd = eth->scratch_ring + (j * MTK_FQ_DMA_LENGTH + i) * soc-
 tx.desc_size;
+txd->txd1 = dma_addr + i * MTK_QDMA_PAGE_SIZE;
+if (j * MTK_FQ_DMA_LENGTH + i < cnt)
+txd->txd2 = eth->phy_scratch_ring +
+    (j * MTK_FQ_DMA_LENGTH + i + 1) * soc->tx.desc_size;
+
+txd->txd3 = TX_DMA_PLEN0(MTK_QDMA_PAGE_SIZE);
+if (MTK_HAS_CAPS(soc->caps, MTK_36BIT_DMA))
+txd->txd3 |= TX_DMA_PREP_ADDR64(dma_addr + i *

MTK_QDMA_PAGE_SIZE);

drivers/net/ethernet/mediatek/mtk_eth_soc.c static int mtk_init_fq_dma(struct mtk_eth *eth) · line 1184

Quoted 8 lines
drivers/net/ethernet/mediatek/mtk_eth_soc.c +8 -0
+
+txd->txd4 = 0;
+if (mtk_is_netsys_v2_or_greater(eth)) {
+txd->txd5 = 0;
+txd->txd6 = 0;
+txd->txd7 = 0;
+txd->txd8 = 0;
+}

This block of change was a bit hard to understand what was going on, but
I think I get the result is that you end up allocating different set of
scratch_head per size vs the original only having one scratch_head per
device?

Quoted 10 lines

Perhaps you can explain, but we're now allocating a bunch of different
scratch_head pointers.. However, in the patch, the only places that we
modify scratch_head appear to be the allocation path and the free path..
but I can't seem to understand how that would impact the users of scratch head? I guess it changes the dma_addr which then changes the txd
values we program?

In our hardware design, we need to allocate a large number of fq_dma buffers for buffering in the hardware-accelerated path. Each fq_dma buffer requires 2048 bytes of memory from the kernel. However, the driver can only request up to 4 MB of contiguous memory at a time if we do not want to request a large contiguous memory from the CMA allocator. Therefore, in the previous driver code, we could only allocate 2048 fq_dma buffers (2048 * 2048 bytes = 4 MB).

With the MT7988, the Ethernet bandwidth has increased to 2*10 Gbps, which means we need to allocate more fq_dma buffers (increased to 4096) to handle the buffering. Consequently, we need to modify the driver code to allocate multiple contiguous memory and assign them into the fq_dma ring.

Quoted 9 lines

Ok.

I sort of understand whats going on here, but it was a fair bit to fully
grok this flow.

Overall, I'm no expert on the part or DMA here, but:

#3 lore raw
In reply to Frank Wunderlich
This patch was applied to netdev/net.git (main) by David S. Miller <[email protected]>:

Hello:

This patch was applied to netdev/net.git (main)
by David S. Miller <[email protected]>:

Quoted 10 lines · On Mon, 3 Jun 2024 21:25:05 +0200 you wrote:

From: Frank Wunderlich <[email protected]>

The mainline MTK ethernet driver suffers long time from rarly but annoying tx queue timeouts. We think that this is caused by fixed dma sizes hardcoded for all SoCs.

We suspect this problem arises from a low level of free TX DMADs, the TX Ring alomost full.

[...]

Here is the summary with links:
- [net,v3] net: ethernet: mtk_eth_soc: handle dma buffer size soc specific

    https://git.kernel.org/netdev/net/c/c57e55819443

You are awesome, thank you!

Signature · 2 lines
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html