Skip to contentSkip to conversation

net: ethernet: mtk_eth_soc: improve device tree handling

PATCH Superseded Daniel Golle 1 participant 5 messages Jun 30, 2025 – Jul 1, 2025

Subsystems (4)

v3 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v3 cover net: ethernet: mtk_eth_soc: improve device tree handling

This series further improves the mtk_eth_soc driver in preparation to complete upstream support for the MediaTek MT7988 SoC family.

Frank Wunderlich's previous attempt to have the ethernet node included in mt7988a.dtsi and cover support for MT7988 in the device tree bindings was criticized for the way mtk_eth_soc references SRAM in device tree[1].

Having a 2nd 'reg' property, like introduced by commit ebb1e4f9cf38 ("net: ethernet: mtk_eth_soc: add support for in-SoC SRAM") isn't acceptable and a dedicated "mmio-sram" node should be used instead.

Support for the hard-coded offset and including the SRAM region as part of the Ethernet's "reg" MMIO space will still be required in order to support existing legacy device trees of the MT7981 and MT7986 SoC families.

While at it also replace confusing error messages when using legacy device trees without "interrupt-names" with a warning informing users that they are using a legacy device tree.

[1]: https://patchwork.ozlabs.org/comment/3533543/

Daniel Golle (3):
net: ethernet: mtk_eth_soc: improve support for named interrupts
net: ethernet: mtk_eth_soc: fix kernel-doc comment
net: ethernet: mtk_eth_soc: use genpool allocator for SRAM

 drivers/net/ethernet/mediatek/mtk_eth_soc.c | 132 +++++++++++++-------
 drivers/net/ethernet/mediatek/mtk_eth_soc.h |   5 +-
 2 files changed, 94 insertions(+), 43 deletions(-)
Signature · 1 line
2.50.0
#1 lore raw

PATCH v3 1/3 net: ethernet: mtk_eth_soc: improve support for named interrupts

Use platform_get_irq_byname_optional() to avoid outputting error messages when using legacy device trees which rely identifying interrupts only by index. Instead, output a warning notifying the user to update their device tree.

---
v3: unchanged
v2: unchanged

 drivers/net/ethernet/mediatek/mtk_eth_soc.c | 15 ++++++++++-----
 1 file changed, 10 insertions(+), 5 deletions(-)
drivers/net/ethernet/mediatek/mtk_eth_soc.c +10 -5
index f8a907747db4..8f55069441f4 100644
@@ -3341,17 +3341,22 @@ static int mtk_get_irqs(struct platform_device *pdev, struct mtk_eth *eth)
 	int i;
 
 	/* future SoCs beginning with MT7988 should use named IRQs in dts */
-	eth->irq[MTK_FE_IRQ_TX] = platform_get_irq_byname(pdev, "fe1");
-	eth->irq[MTK_FE_IRQ_RX] = platform_get_irq_byname(pdev, "fe2");
+	eth->irq[MTK_FE_IRQ_TX] = platform_get_irq_byname_optional(pdev, "fe1");
+	eth->irq[MTK_FE_IRQ_RX] = platform_get_irq_byname_optional(pdev, "fe2");
 	if (eth->irq[MTK_FE_IRQ_TX] >= 0 && eth->irq[MTK_FE_IRQ_RX] >= 0)
 		return 0;
 
-	/* only use legacy mode if platform_get_irq_byname returned -ENXIO */
+	/* only use legacy mode if platform_get_irq_byname_optional returned -ENXIO */
 	if (eth->irq[MTK_FE_IRQ_TX] != -ENXIO)
-		return eth->irq[MTK_FE_IRQ_TX];
+		return dev_err_probe(&pdev->dev, eth->irq[MTK_FE_IRQ_TX],
+				     "Error requesting FE TX IRQ\n");
 
 	if (eth->irq[MTK_FE_IRQ_RX] != -ENXIO)
-		return eth->irq[MTK_FE_IRQ_RX];
+		return dev_err_probe(&pdev->dev, eth->irq[MTK_FE_IRQ_RX],
+				     "Error requesting FE RX IRQ\n");
+
+	if (!MTK_HAS_CAPS(eth->soc->caps, MTK_SHARED_INT))
+		dev_warn(&pdev->dev, "legacy DT: missing interrupt-names.");
 
 	/* legacy way:
 	 * On MTK_SHARED_INT SoCs (MT7621 + MT7628) the first IRQ is taken
Signature · 1 line
2.50.0
#2 lore raw

PATCH v3 2/3 net: ethernet: mtk_eth_soc: fix kernel-doc comment

In reply to Daniel Golle

Fix and add some missing field descriptions to kernel-doc comment of struct mtk_eth.

---
v3: unchanged
v2: unchanged

 drivers/net/ethernet/mediatek/mtk_eth_soc.h | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
drivers/net/ethernet/mediatek/mtk_eth_soc.h +2 -1
index 9261c0e13b59..1ad9075a9b69 100644
@@ -1243,8 +1243,9 @@ struct mtk_soc_data {
 /* struct mtk_eth -	This is the main datasructure for holding the state
  *			of the driver
  * @dev:		The device pointer
- * @dev:		The device pointer used for dma mapping/alloc
+ * @dma_dev:		The device pointer used for dma mapping/alloc
  * @base:		The mapped register i/o base
+ * @sram_base:		The mapped SRAM base
  * @page_lock:		Make sure that register operations are atomic
  * @tx_irq__lock:	Make sure that IRQ register operations are atomic
  * @rx_irq__lock:	Make sure that IRQ register operations are atomic
Signature · 1 line
2.50.0
#3 lore raw

PATCH v3 3/3 net: ethernet: mtk_eth_soc: use genpool allocator for SRAM

In reply to Daniel Golle

Use a dedicated "mmio-sram" and the genpool allocator instead of open-coding SRAM allocation for DMA rings.
Keep support for legacy device trees but notify the user via a warning to update.

---
v3: fix resource leak on error in mtk_probe()
v2: fix return type of mtk_dma_ring_alloc() in case of error

 drivers/net/ethernet/mediatek/mtk_eth_soc.c | 117 +++++++++++++-------
 drivers/net/ethernet/mediatek/mtk_eth_soc.h |   4 +-
 2 files changed, 83 insertions(+), 38 deletions(-)
drivers/net/ethernet/mediatek/mtk_eth_soc.c +80 -37
index 8f55069441f4..29b95d96f3d4 100644
@@ -27,6 +27,7 @@
 #include <net/dsa.h>
 #include <net/dst_metadata.h>
 #include <net/page_pool/helpers.h>
+#include <linux/genalloc.h>
 
 #include "mtk_eth_soc.h"
 #include "mtk_wed.h"
@@ -1267,6 +1268,45 @@ static void *mtk_max_lro_buf_alloc(gfp_t gfp_mask)
 	return (void *)data;
 }
 
+static bool mtk_use_legacy_sram(struct mtk_eth *eth)
+{
+	return !eth->sram_pool && MTK_HAS_CAPS(eth->soc->caps, MTK_SRAM);
+}
+
+static void *mtk_dma_ring_alloc(struct mtk_eth *eth, size_t size,
+				dma_addr_t *dma_handle)
+{
+	void *dma_ring;
+
+	if (WARN_ON(mtk_use_legacy_sram(eth)))
+		return NULL;
+
+	if (eth->sram_pool) {
+		dma_ring = (void *)gen_pool_alloc(eth->sram_pool, size);
+		if (!dma_ring)
+			return dma_ring;
+		*dma_handle = gen_pool_virt_to_phys(eth->sram_pool,
+						    (unsigned long)dma_ring);
+	} else {
+		dma_ring = dma_alloc_coherent(eth->dma_dev, size, dma_handle,
+					      GFP_KERNEL);
+	}
+
+	return dma_ring;
+}
+
+static void mtk_dma_ring_free(struct mtk_eth *eth, size_t size, void *dma_ring,
+			      dma_addr_t dma_handle)
+{
+	if (WARN_ON(mtk_use_legacy_sram(eth)))
+		return;
+
+	if (eth->sram_pool)
+		gen_pool_free(eth->sram_pool, (unsigned long)dma_ring, size);
+	else
+		dma_free_coherent(eth->dma_dev, size, dma_ring, dma_handle);
+}
+
 /* the qdma core needs scratch memory to be setup */
 static int mtk_init_fq_dma(struct mtk_eth *eth)
 {
@@ -1276,13 +1316,12 @@ static int mtk_init_fq_dma(struct mtk_eth *eth)
 	dma_addr_t dma_addr;
 	int i, j, len;
 
-	if (MTK_HAS_CAPS(eth->soc->caps, MTK_SRAM))
+	if (!mtk_use_legacy_sram(eth)) {
+		eth->scratch_ring = mtk_dma_ring_alloc(eth, cnt * soc->tx.desc_size,
+						       &eth->phy_scratch_ring);
+	} else {
 		eth->scratch_ring = eth->sram_base;
-	else
-		eth->scratch_ring = dma_alloc_coherent(eth->dma_dev,
-						       cnt * soc->tx.desc_size,
-						       &eth->phy_scratch_ring,
-						       GFP_KERNEL);
+	}
 
 	if (unlikely(!eth->scratch_ring))
 		return -ENOMEM;
@@ -2620,12 +2659,11 @@ static int mtk_tx_alloc(struct mtk_eth *eth)
 	if (!ring->buf)
 		goto no_tx_mem;
 
-	if (MTK_HAS_CAPS(soc->caps, MTK_SRAM)) {
+	if (!mtk_use_legacy_sram(eth)) {
+		ring->dma = mtk_dma_ring_alloc(eth, ring_size * sz, &ring->phys);
+	} else {
 		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);
 	}
 
 	if (!ring->dma)
@@ -2726,9 +2764,9 @@ static void mtk_tx_clean(struct mtk_eth *eth)
 		kfree(ring->buf);
 		ring->buf = NULL;
 	}
-	if (!MTK_HAS_CAPS(soc->caps, MTK_SRAM) && ring->dma) {
-		dma_free_coherent(eth->dma_dev,
-				  ring->dma_size * soc->tx.desc_size,
+
+	if (!mtk_use_legacy_sram(eth) && ring->dma) {
+		mtk_dma_ring_free(eth, ring->dma_size * soc->tx.desc_size,
 				  ring->dma, ring->phys);
 		ring->dma = NULL;
 	}
@@ -2793,6 +2831,9 @@ static int mtk_rx_alloc(struct mtk_eth *eth, int ring_no, int rx_flag)
 		ring->dma = dma_alloc_coherent(eth->dma_dev,
 				rx_dma_size * eth->soc->rx.desc_size,
 				&ring->phys, GFP_KERNEL);
+	} else if (eth->sram_pool) {
+		ring->dma = mtk_dma_ring_alloc(eth, rx_dma_size * eth->soc->rx.desc_size,
+					       &ring->phys);
 	} else {
 		struct mtk_tx_ring *tx_ring = &eth->tx_ring;
 
@@ -2921,6 +2962,11 @@ static void mtk_rx_clean(struct mtk_eth *eth, struct mtk_rx_ring *ring, bool in_
 				  ring->dma_size * eth->soc->rx.desc_size,
 				  ring->dma, ring->phys);
 		ring->dma = NULL;
+	} else if (!mtk_use_legacy_sram(eth) && ring->dma) {
+		mtk_dma_ring_free(eth,
+				  ring->dma_size * eth->soc->rx.desc_size,
+				  ring->dma, ring->phys);
+		ring->dma = NULL;
 	}
 
 	if (ring->page_pool) {
@@ -3287,9 +3333,8 @@ static void mtk_dma_free(struct mtk_eth *eth)
 			netdev_tx_reset_subqueue(eth->netdev[i], j);
 	}
 
-	if (!MTK_HAS_CAPS(soc->caps, MTK_SRAM) && eth->scratch_ring) {
-		dma_free_coherent(eth->dma_dev,
-				  MTK_QDMA_RING_SIZE * soc->tx.desc_size,
+	if (!mtk_use_legacy_sram(eth) && eth->scratch_ring) {
+		mtk_dma_ring_free(eth, soc->tx.fq_dma_size * soc->tx.desc_size,
 				  eth->scratch_ring, eth->phy_scratch_ring);
 		eth->scratch_ring = NULL;
 		eth->phy_scratch_ring = 0;
@@ -5009,7 +5054,7 @@ static int mtk_sgmii_init(struct mtk_eth *eth)
 
 static int mtk_probe(struct platform_device *pdev)
 {
-	struct resource *res = NULL, *res_sram;
+	struct resource *res = NULL;
 	struct device_node *mac_np;
 	struct mtk_eth *eth;
 	int err, i;
@@ -5029,20 +5074,6 @@ static int mtk_probe(struct platform_device *pdev)
 	if (MTK_HAS_CAPS(eth->soc->caps, MTK_SOC_MT7628))
 		eth->ip_align = NET_IP_ALIGN;
 
-	if (MTK_HAS_CAPS(eth->soc->caps, MTK_SRAM)) {
-		/* SRAM is actual memory and supports transparent access just like DRAM.
-		 * Hence we don't require __iomem being set and don't need to use accessor
-		 * functions to read from or write to SRAM.
-		 */
-		if (mtk_is_netsys_v3_or_greater(eth)) {
-			eth->sram_base = (void __force *)devm_platform_ioremap_resource(pdev, 1);
-			if (IS_ERR(eth->sram_base))
-				return PTR_ERR(eth->sram_base);
-		} else {
-			eth->sram_base = (void __force *)eth->base + MTK_ETH_SRAM_OFFSET;
-		}
-	}
-
 	if (MTK_HAS_CAPS(eth->soc->caps, MTK_36BIT_DMA)) {
 		err = dma_set_mask(&pdev->dev, DMA_BIT_MASK(36));
 		if (!err)
@@ -5117,16 +5148,28 @@ static int mtk_probe(struct platform_device *pdev)
 			err = -EINVAL;
 			goto err_destroy_sgmii;
 		}
+
 		if (MTK_HAS_CAPS(eth->soc->caps, MTK_SRAM)) {
-			if (mtk_is_netsys_v3_or_greater(eth)) {
-				res_sram = platform_get_resource(pdev, IORESOURCE_MEM, 1);
-				if (!res_sram) {
+			eth->sram_pool = of_gen_pool_get(pdev->dev.of_node, "sram", 0);
+			if (!eth->sram_pool) {
+				if (!mtk_is_netsys_v3_or_greater(eth)) {
+					/*
+					 * Legacy support for missing 'sram' node in DT.
+					 * SRAM is actual memory and supports transparent access
+					 * just like DRAM. Hence we don't require __iomem being
+					 * set and don't need to use accessor functions to read from
+					 * or write to SRAM.
+					 */
+					eth->sram_base = (void __force *)eth->base +
+							 MTK_ETH_SRAM_OFFSET;
+					eth->phy_scratch_ring = res->start + MTK_ETH_SRAM_OFFSET;
+					dev_warn(&pdev->dev,
+						 "legacy DT: using hard-coded SRAM offset.\n");
+				} else {
+					dev_err(&pdev->dev, "Could not get SRAM pool\n");
 					err = -EINVAL;
 					goto err_destroy_sgmii;
 				}
-				eth->phy_scratch_ring = res_sram->start;
-			} else {
-				eth->phy_scratch_ring = res->start + MTK_ETH_SRAM_OFFSET;
 			}
 		}
 	}
drivers/net/ethernet/mediatek/mtk_eth_soc.h +3 -1
index 1ad9075a9b69..0104659e37f0 100644
@@ -1245,7 +1245,8 @@ struct mtk_soc_data {
  * @dev:		The device pointer
  * @dma_dev:		The device pointer used for dma mapping/alloc
  * @base:		The mapped register i/o base
- * @sram_base:		The mapped SRAM base
+ * @sram_base:		The mapped SRAM base (deprecated)
+ * @sram_pool:		Pointer to SRAM pool used for DMA descriptor rings
  * @page_lock:		Make sure that register operations are atomic
  * @tx_irq__lock:	Make sure that IRQ register operations are atomic
  * @rx_irq__lock:	Make sure that IRQ register operations are atomic
@@ -1292,6 +1293,7 @@ struct mtk_eth {
 	struct device			*dma_dev;
 	void __iomem			*base;
 	void				*sram_base;
+	struct gen_pool			*sram_pool;
 	spinlock_t			page_lock;
 	spinlock_t			tx_irq_lock;
 	spinlock_t			rx_irq_lock;
Signature · 1 line
2.50.0
#4 lore raw

Re: [PATCH net-next v3 3/3] net: ethernet: mtk_eth_soc: use genpool allocator for SRAM

Quoted 8 lines · Daniel Golle

Use a dedicated "mmio-sram" and the genpool allocator instead of open-coding SRAM allocation for DMA rings.
Keep support for legacy device trees but notify the user via a warning to update.

Quoted 8 lines

---
v3: fix resource leak on error in mtk_probe()
v2: fix return type of mtk_dma_ring_alloc() in case of error

 drivers/net/ethernet/mediatek/mtk_eth_soc.c | 117 +++++++++++++-------
 drivers/net/ethernet/mediatek/mtk_eth_soc.h |   4 +-
 2 files changed, 83 insertions(+), 38 deletions(-)

I just noticed I forgot to 'git add' the Kconfig change adding 'select GENERIC_ALLOCATOR' for NET_MEDIATEK_SOC... I will include that in v4 and also add a patch dropping the open coded static partitioning of the SRAM in favor of always using the gen_pool allocator even with legacy device trees.

Meanwhile I would still be thankful to see this patch reviewed as apart from the Kconfig addition it won't change.