Skip to contentSkip to conversation

Fixes for running a big-endian kernel on Cubieboard2

PATCH Superseded Michael Weiser 4 participants 16 messages Jul 21, 2016 – Jul 29, 2016

Subsystems (8)

v2 of 5 Superseded Latest is v5 →

lore
Share & tools
lore
#0 lore raw

v2 cover Fixes for running a big-endian kernel on Cubieboard2

Hello,

the following patches fix up endianness problems with DMA descriptors in the sunxi-mmc and stmmac drivers preventing a big-endian kernel from working correctly on the Cubieboard2. The final patch enables the big-endian kernel configuration option for sunxi devices in order to be able to compile a big-endian kernel for the Cubieboard2. All of these have been tested successfull on a Cubieboard2 Dual-Card.

Changes since v1:
- fixed checkpatch niggles
- added respective Cc:s

Regards,
Michael

#1 lore raw

PATCH v2 1/3 mmc: sunxi-mmc: change idma descriptor to __le32

The sunxi-mmc driver does not take into account the processor may be big endian when writing the DMA descriptors. This causes cards not to be detected when running a big-endian kernel. Change the descriptors for IDMA to use __le32 and ensure they are suitably swapped before writing. Tested successfully on the Cubieboard2.

Cc Ulf Hansson <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Chen-Yu Tsai <[email protected]>
Cc linux-mmc at vger.kernel.org
---
 drivers/mmc/host/sunxi-mmc.c | 27 +++++++++++++++------------
 1 file changed, 15 insertions(+), 12 deletions(-)
drivers/mmc/host/sunxi-mmc.c +15 -12
index 2ee4c21..1e6df8a 100644
@@ -223,10 +223,10 @@ struct sunxi_mmc_clk_delay {
 };
 
 struct sunxi_idma_des {
-	u32	config;
-	u32	buf_size;
-	u32	buf_addr_ptr1;
-	u32	buf_addr_ptr2;
+	__le32 config;
+	__le32 buf_size;
+	__le32 buf_addr_ptr1;
+	__le32 buf_addr_ptr2;
 };
 
 struct sunxi_mmc_host {
@@ -325,22 +325,25 @@ static void sunxi_mmc_init_idma_des(struct sunxi_mmc_host *host,
 	int i, max_len = (1 << host->idma_des_size_bits);
 
 	for (i = 0; i < data->sg_len; i++) {
-		pdes[i].config = SDXC_IDMAC_DES0_CH | SDXC_IDMAC_DES0_OWN |
-				 SDXC_IDMAC_DES0_DIC;
+		pdes[i].config = cpu_to_le32(SDXC_IDMAC_DES0_CH |
+					     SDXC_IDMAC_DES0_OWN |
+					     SDXC_IDMAC_DES0_DIC);
 
 		if (data->sg[i].length == max_len)
 			pdes[i].buf_size = 0; /* 0 == max_len */
 		else
-			pdes[i].buf_size = data->sg[i].length;
+			pdes[i].buf_size = cpu_to_le32(data->sg[i].length);
 
 		next_desc += sizeof(struct sunxi_idma_des);
-		pdes[i].buf_addr_ptr1 = sg_dma_address(&data->sg[i]);
-		pdes[i].buf_addr_ptr2 = (u32)next_desc;
+		pdes[i].buf_addr_ptr1 =
+			cpu_to_le32(sg_dma_address(&data->sg[i]));
+		pdes[i].buf_addr_ptr2 = cpu_to_le32((u32)next_desc);
 	}
 
-	pdes[0].config |= SDXC_IDMAC_DES0_FD;
-	pdes[i - 1].config |= SDXC_IDMAC_DES0_LD | SDXC_IDMAC_DES0_ER;
-	pdes[i - 1].config &= ~SDXC_IDMAC_DES0_DIC;
+	pdes[0].config |= cpu_to_le32(SDXC_IDMAC_DES0_FD);
+	pdes[i - 1].config |= cpu_to_le32(SDXC_IDMAC_DES0_LD |
+					  SDXC_IDMAC_DES0_ER);
+	pdes[i - 1].config &= cpu_to_le32(~SDXC_IDMAC_DES0_DIC);
 	pdes[i - 1].buf_addr_ptr2 = 0;
 
 	/*
Signature · 1 line
2.9.0
#2 lore raw

PATCH 1/3 mmc: sunxi-mmc: change idma descriptor to __le32

Quoted 11 lines · Michael Weiser

The sunxi-mmc driver does not take into account the processor may be big endian when writing the DMA descriptors. This causes cards not to be detected when running a big-endian kernel. Change the descriptors for IDMA to use __le32 and ensure they are suitably swapped before writing. Tested successfully on the Cubieboard2.

Cc Ulf Hansson <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Chen-Yu Tsai <[email protected]>
Cc linux-mmc at vger.kernel.org

Thanks,
Maxime

Signature · 10 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20160722/f4b399ca/attachment.sig>
#3 lore raw

PATCH v2 2/3 stmmac: change dma descriptors to __le32

In reply to Michael Weiser

The stmmac driver does not take into account the processor may be big endian when writing the DMA descriptors. This causes the ethernet interface not to be initialised correctly when running a big-endian kernel. Change the descriptors for DMA to use __le32 and ensure they are suitably swapped before writing. Tested successfully on the Cubieboard2.

Cc Giuseppe Cavallaro <[email protected]>
Cc Alexandre Torgue <[email protected]>
Cc netdev at vger.kernel.org
---
 drivers/net/ethernet/stmicro/stmmac/chain_mode.c   | 55 ++++++++++----------
 drivers/net/ethernet/stmicro/stmmac/descs.h        | 20 ++++----
 drivers/net/ethernet/stmicro/stmmac/descs_com.h    | 48 +++++++++--------
 drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c | 60 +++++++++++-----------
 drivers/net/ethernet/stmicro/stmmac/enh_desc.c     | 55 ++++++++++----------
 drivers/net/ethernet/stmicro/stmmac/norm_desc.c    | 48 ++++++++---------
 drivers/net/ethernet/stmicro/stmmac/ring_mode.c    | 39 +++++++-------
 drivers/net/ethernet/stmicro/stmmac/stmmac_main.c  | 53 ++++++++++---------
 8 files changed, 196 insertions(+), 182 deletions(-)
drivers/net/ethernet/stmicro/stmmac/chain_mode.c +29 -26
index b3e669a..026e8e9 100644
@@ -34,7 +34,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
 	unsigned int entry = priv->cur_tx;
 	struct dma_desc *desc = priv->dma_tx + entry;
 	unsigned int nopaged_len = skb_headlen(skb);
-	unsigned int bmax;
+	unsigned int bmax, des2;
 	unsigned int i = 1, len;
 
 	if (priv->plat->enh_desc)
@@ -44,11 +44,12 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
 
 	len = nopaged_len - bmax;
 
-	desc->des2 = dma_map_single(priv->device, skb->data,
-				    bmax, DMA_TO_DEVICE);
-	if (dma_mapping_error(priv->device, desc->des2))
+	des2 = dma_map_single(priv->device, skb->data,
+			      bmax, DMA_TO_DEVICE);
+	desc->des2 = cpu_to_le32(des2);
+	if (dma_mapping_error(priv->device, des2))
 		return -1;
-	priv->tx_skbuff_dma[entry].buf = desc->des2;
+	priv->tx_skbuff_dma[entry].buf = des2;
 	priv->tx_skbuff_dma[entry].len = bmax;
 	/* do not close the descriptor and do not set own bit */
 	priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum, STMMAC_CHAIN_MODE,
@@ -60,12 +61,13 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
 		desc = priv->dma_tx + entry;
 
 		if (len > bmax) {
-			desc->des2 = dma_map_single(priv->device,
-						    (skb->data + bmax * i),
-						    bmax, DMA_TO_DEVICE);
-			if (dma_mapping_error(priv->device, desc->des2))
+			des2 = dma_map_single(priv->device,
+					      (skb->data + bmax * i),
+					      bmax, DMA_TO_DEVICE);
+			desc->des2 = cpu_to_le32(des2);
+			if (dma_mapping_error(priv->device, des2))
 				return -1;
-			priv->tx_skbuff_dma[entry].buf = desc->des2;
+			priv->tx_skbuff_dma[entry].buf = des2;
 			priv->tx_skbuff_dma[entry].len = bmax;
 			priv->hw->desc->prepare_tx_desc(desc, 0, bmax, csum,
 							STMMAC_CHAIN_MODE, 1,
@@ -73,12 +75,13 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
 			len -= bmax;
 			i++;
 		} else {
-			desc->des2 = dma_map_single(priv->device,
-						    (skb->data + bmax * i), len,
-						    DMA_TO_DEVICE);
-			if (dma_mapping_error(priv->device, desc->des2))
+			des2 = dma_map_single(priv->device,
+					      (skb->data + bmax * i), len,
+					      DMA_TO_DEVICE);
+			desc->des2 = cpu_to_le32(des2);
+			if (dma_mapping_error(priv->device, des2))
 				return -1;
-			priv->tx_skbuff_dma[entry].buf = desc->des2;
+			priv->tx_skbuff_dma[entry].buf = des2;
 			priv->tx_skbuff_dma[entry].len = len;
 			/* last descriptor can be set now */
 			priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
@@ -119,19 +122,19 @@ static void stmmac_init_dma_chain(void *des, dma_addr_t phy_addr,
 		struct dma_extended_desc *p = (struct dma_extended_desc *)des;
 		for (i = 0; i < (size - 1); i++) {
 			dma_phy += sizeof(struct dma_extended_desc);
-			p->basic.des3 = (unsigned int)dma_phy;
+			p->basic.des3 = cpu_to_le32((unsigned int)dma_phy);
 			p++;
 		}
-		p->basic.des3 = (unsigned int)phy_addr;
+		p->basic.des3 = cpu_to_le32((unsigned int)phy_addr);
 
 	} else {
 		struct dma_desc *p = (struct dma_desc *)des;
 		for (i = 0; i < (size - 1); i++) {
 			dma_phy += sizeof(struct dma_desc);
-			p->des3 = (unsigned int)dma_phy;
+			p->des3 = cpu_to_le32((unsigned int)dma_phy);
 			p++;
 		}
-		p->des3 = (unsigned int)phy_addr;
+		p->des3 = cpu_to_le32((unsigned int)phy_addr);
 	}
 }
 
@@ -144,10 +147,10 @@ static void stmmac_refill_desc3(void *priv_ptr, struct dma_desc *p)
 		 * 1588-2002 time stamping is enabled, hence reinitialize it
 		 * to keep explicit chaining in the descriptor.
 		 */
-		p->des3 = (unsigned int)(priv->dma_rx_phy +
-					 (((priv->dirty_rx) + 1) %
-					  DMA_RX_SIZE) *
-					 sizeof(struct dma_desc));
+		p->des3 = cpu_to_le32((unsigned int)(priv->dma_rx_phy +
+				      (((priv->dirty_rx) + 1) %
+				       DMA_RX_SIZE) *
+				      sizeof(struct dma_desc)));
 }
 
 static void stmmac_clean_desc3(void *priv_ptr, struct dma_desc *p)
@@ -161,9 +164,9 @@ static void stmmac_clean_desc3(void *priv_ptr, struct dma_desc *p)
 		 * 1588-2002 time stamping is enabled, hence reinitialize it
 		 * to keep explicit chaining in the descriptor.
 		 */
-		p->des3 = (unsigned int)((priv->dma_tx_phy +
-					  ((priv->dirty_tx + 1) % DMA_TX_SIZE))
-					  * sizeof(struct dma_desc));
+		p->des3 = cpu_to_le32((unsigned int)((priv->dma_tx_phy +
+				      ((priv->dirty_tx + 1) % DMA_TX_SIZE))
+				      * sizeof(struct dma_desc)));
 }
 
 const struct stmmac_mode_ops chain_mode_ops = {
drivers/net/ethernet/stmicro/stmmac/descs.h +10 -10
index 2e4c171..4000af4 100644
@@ -87,7 +87,7 @@
 #define	TDES0_ERROR_SUMMARY		BIT(15)
 #define	TDES0_IP_HEADER_ERROR		BIT(16)
 #define	TDES0_TIME_STAMP_STATUS		BIT(17)
-#define	TDES0_OWN			BIT(31)
+#define	TDES0_OWN			((u32)BIT(31))	/* silence sparse */
 /* TDES1 */
 #define	TDES1_BUFFER1_SIZE_MASK		GENMASK(10, 0)
 #define	TDES1_BUFFER2_SIZE_MASK		GENMASK(21, 11)
@@ -130,7 +130,7 @@
 #define	ETDES0_FIRST_SEGMENT		BIT(28)
 #define	ETDES0_LAST_SEGMENT		BIT(29)
 #define	ETDES0_INTERRUPT		BIT(30)
-#define	ETDES0_OWN			BIT(31)
+#define	ETDES0_OWN			((u32)BIT(31))	/* silence sparse */
 /* TDES1 */
 #define	ETDES1_BUFFER1_SIZE_MASK	GENMASK(12, 0)
 #define	ETDES1_BUFFER2_SIZE_MASK	GENMASK(28, 16)
@@ -166,19 +166,19 @@
 
 /* Basic descriptor structure for normal and alternate descriptors */
 struct dma_desc {
-	unsigned int des0;
-	unsigned int des1;
-	unsigned int des2;
-	unsigned int des3;
+	__le32 des0;
+	__le32 des1;
+	__le32 des2;
+	__le32 des3;
 };
 
 /* Extended descriptor structure (e.g. >= databook 3.50a) */
 struct dma_extended_desc {
 	struct dma_desc basic;	/* Basic descriptors */
-	unsigned int des4;	/* Extended Status */
-	unsigned int des5;	/* Reserved */
-	unsigned int des6;	/* Tx/Rx Timestamp Low */
-	unsigned int des7;	/* Tx/Rx Timestamp High */
+	__le32 des4;	/* Extended Status */
+	__le32 des5;	/* Reserved */
+	__le32 des6;	/* Tx/Rx Timestamp Low */
+	__le32 des7;	/* Tx/Rx Timestamp High */
 };
 
 /* Transmit checksum insertion control */
drivers/net/ethernet/stmicro/stmmac/descs_com.h +26 -22

115 lines not on this page · raw message

drivers/net/ethernet/stmicro/stmmac/dwmac4_descs.c +31 -29

174 lines not on this page · raw message

drivers/net/ethernet/stmicro/stmmac/enh_desc.c +28 -27

164 lines not on this page · raw message

drivers/net/ethernet/stmicro/stmmac/norm_desc.c +25 -23

155 lines not on this page · raw message

drivers/net/ethernet/stmicro/stmmac/ring_mode.c +21 -18
index 7723b5d..9983ce9 100644
@@ -34,7 +34,7 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
 	unsigned int entry = priv->cur_tx;
 	struct dma_desc *desc;
 	unsigned int nopaged_len = skb_headlen(skb);
-	unsigned int bmax, len;
+	unsigned int bmax, len, des2;
 
 	if (priv->extend_desc)
 		desc = (struct dma_desc *)(priv->dma_etx + entry);
@@ -50,16 +50,17 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
 
 	if (nopaged_len > BUF_SIZE_8KiB) {
 
-		desc->des2 = dma_map_single(priv->device, skb->data,
-					    bmax, DMA_TO_DEVICE);
-		if (dma_mapping_error(priv->device, desc->des2))
+		des2 = dma_map_single(priv->device, skb->data, bmax,
+				      DMA_TO_DEVICE);
+		desc->des2 = cpu_to_le32(des2);
+		if (dma_mapping_error(priv->device, des2))
 			return -1;
 
-		priv->tx_skbuff_dma[entry].buf = desc->des2;
+		priv->tx_skbuff_dma[entry].buf = des2;
 		priv->tx_skbuff_dma[entry].len = bmax;
 		priv->tx_skbuff_dma[entry].is_jumbo = true;
 
-		desc->des3 = desc->des2 + BUF_SIZE_4KiB;
+		desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
 		priv->hw->desc->prepare_tx_desc(desc, 1, bmax, csum,
 						STMMAC_RING_MODE, 0, false);
 		priv->tx_skbuff[entry] = NULL;
@@ -70,26 +71,28 @@ static int stmmac_jumbo_frm(void *p, struct sk_buff *skb, int csum)
 		else
 			desc = priv->dma_tx + entry;
 
-		desc->des2 = dma_map_single(priv->device, skb->data + bmax,
-					    len, DMA_TO_DEVICE);
-		if (dma_mapping_error(priv->device, desc->des2))
+		des2 = dma_map_single(priv->device, skb->data + bmax, len,
+				      DMA_TO_DEVICE);
+		desc->des2 = cpu_to_le32(des2);
+		if (dma_mapping_error(priv->device, des2))
 			return -1;
-		priv->tx_skbuff_dma[entry].buf = desc->des2;
+		priv->tx_skbuff_dma[entry].buf = des2;
 		priv->tx_skbuff_dma[entry].len = len;
 		priv->tx_skbuff_dma[entry].is_jumbo = true;
 
-		desc->des3 = desc->des2 + BUF_SIZE_4KiB;
+		desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
 		priv->hw->desc->prepare_tx_desc(desc, 0, len, csum,
 						STMMAC_RING_MODE, 1, true);
 	} else {
-		desc->des2 = dma_map_single(priv->device, skb->data,
-					    nopaged_len, DMA_TO_DEVICE);
-		if (dma_mapping_error(priv->device, desc->des2))
+		des2 = dma_map_single(priv->device, skb->data,
+				      nopaged_len, DMA_TO_DEVICE);
+		desc->des2 = cpu_to_le32(des2);
+		if (dma_mapping_error(priv->device, des2))
 			return -1;
-		priv->tx_skbuff_dma[entry].buf = desc->des2;
+		priv->tx_skbuff_dma[entry].buf = des2;
 		priv->tx_skbuff_dma[entry].len = nopaged_len;
 		priv->tx_skbuff_dma[entry].is_jumbo = true;
-		desc->des3 = desc->des2 + BUF_SIZE_4KiB;
+		desc->des3 = cpu_to_le32(des2 + BUF_SIZE_4KiB);
 		priv->hw->desc->prepare_tx_desc(desc, 1, nopaged_len, csum,
 						STMMAC_RING_MODE, 0, true);
 	}
@@ -115,13 +118,13 @@ static void stmmac_refill_desc3(void *priv_ptr, struct dma_desc *p)
 
 	/* Fill DES3 in case of RING mode */
 	if (priv->dma_buf_sz >= BUF_SIZE_8KiB)
-		p->des3 = p->des2 + BUF_SIZE_8KiB;
+		p->des3 = cpu_to_le32(le32_to_cpu(p->des2) + BUF_SIZE_8KiB);
 }
 
 /* In ring mode we need to fill the desc3 because it is used as buffer */
 static void stmmac_init_desc3(struct dma_desc *p)
 {
-	p->des3 = p->des2 + BUF_SIZE_8KiB;
+	p->des3 = cpu_to_le32(le32_to_cpu(p->des2) + BUF_SIZE_8KiB);
 }
 
 static void stmmac_clean_desc3(void *priv_ptr, struct dma_desc *p)
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c +26 -27

112 lines not on this page · raw message

Signature · 1 line
2.9.0
#4 lore raw

Re: [PATCH 2/3] stmmac: change dma descriptors to __le32

Quoted 6 lines · Michael Weiser

The stmmac driver does not take into account the processor may be big endian when writing the DMA descriptors. This causes the ethernet interface not to be initialised correctly when running a big-endian kernel. Change the descriptors for DMA to use __le32 and ensure they are suitably swapped before writing. Tested successfully on the Cubieboard2.

Thx for the effort on big endian platform.

Quoted 6 lines
Cc Giuseppe Cavallaro <[email protected]>
Cc Alexandre Torgue <[email protected]>
Cc netdev at vger.kernel.org

---

...

drivers/net/ethernet/stmicro/stmmac/stmmac_main.c static void sysfs_display_ring(void *head, int size, int extend_desc, · line 2883

Quoted 13 lines
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c +5 -3
@@ -2880,14 +2876,17 @@ static void sysfs_display_ring(void *head, int size, int extend_desc,
 			x = *(u64 *) ep;
 			seq_printf(seq, "%d [0x%x]: 0x%x 0x%x 0x%x 0x%x\n",
 				   i, (unsigned int)virt_to_phys(ep),
-				   ep->basic.des0, ep->basic.des1,
-				   ep->basic.des2, ep->basic.des3);
-			ep++;
+				   le32_to_cpu(ep->basic.des0),
+				   le32_to_cpu(ep->basic.des1),
+				   le32_to_cpu(ep->basic.des2),
+				   le32_to_cpu(ep->basic.des3));
+			ep++);

there is a build problem here.

Pls fix it.

Peppe

#5 lore raw

Re: [PATCH 2/3] stmmac: change dma descriptors to __le32

Hi Giuseppe,

Quoted 4 lines · Giuseppe CAVALLARO

drivers/net/ethernet/stmicro/stmmac/stmmac_main.c static void sysfs_display_ring(void *head, int size, int extend_desc, · line 2883

Quoted 2 lines
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c +1 -1
-			ep++;
+			ep++);

there is a build problem here.
Pls fix it.

Thanks for your review. I've got it fixed locally. An updated patchset will be forthcoming as soon as I've sorted a problem with sun4i-emac which I've been asked to look into as well.

Signature · 2 lines
Thanks,
Michael
#6 lore raw

Re: [PATCH 2/3] stmmac: change dma descriptors to __le32

Quoted 4 lines · Michael Weiser

Hi Giuseppe,

On Tue, Jul 26, 2016 at 02:13:45PM +0200, Giuseppe CAVALLARO wrote:

Quoted 9 lines
Quoted 4 lines

drivers/net/ethernet/stmicro/stmmac/stmmac_main.c static void sysfs_display_ring(void *head, int size, int extend_desc, · line 2883

Quoted 2 lines
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c +1 -1
-			ep++;
+			ep++);

there is a build problem here.
Pls fix it.

Thanks for your review. I've got it fixed locally. An updated patchset will be forthcoming as soon as I've sorted a problem with sun4i-emac which I've been asked to look into as well.

ok, as you send the new version I will do a run on ARM box based too

peppe

#7 lore raw

PATCH v2 3/3 ARM: sunxi: enable big-endian

In reply to Michael Weiser

At least the Cubieboard2 incarnation of sunxi works correctly in big-endian mode. Allow the selection of big-endian.

Cc Maxime Ripard <[email protected]>
Cc Chen-Yu Tsai <[email protected]>
Cc linux-arm-kernel at lists.infradead.org
---
 arch/arm/mach-sunxi/Kconfig | 1 +
 1 file changed, 1 insertion(+)
arch/arm/mach-sunxi/Kconfig +1 -0
index c124d65..4dcbbfe 100644
@@ -1,6 +1,7 @@
 menuconfig ARCH_SUNXI
 	bool "Allwinner SoCs"
 	depends on ARCH_MULTI_V7
+	select ARCH_SUPPORTS_BIG_ENDIAN
 	select ARCH_REQUIRE_GPIOLIB
 	select ARCH_HAS_RESET_CONTROLLER
 	select CLKSRC_MMIO
Signature · 1 line
2.9.0
#8 lore raw

PATCH 3/3 ARM: sunxi: enable big-endian

Hi,

Quoted 7 lines · Michael Weiser

At least the Cubieboard2 incarnation of sunxi works correctly in big-endian mode. Allow the selection of big-endian.

Cc Maxime Ripard <[email protected]>
Cc Chen-Yu Tsai <[email protected]>
Cc linux-arm-kernel at lists.infradead.org

With what devices has it been tested? The one doing DMA will probably be the one who fail, have you tested the DMA engine? EMAC? GMAC?

Thanks,
Maxime

Signature · 10 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20160722/0e68325a/attachment.sig>
#9 lore raw

PATCH v2 3/3 ARM: sunxi: enable big-endian

Hi Maxime,

Quoted 4 lines · Maxime Ripard
Quoted 2 lines

At least the Cubieboard2 incarnation of sunxi works correctly in big-endian mode. Allow the selection of big-endian.

With what devices has it been tested? The one doing DMA will probably be the one who fail, have you tested the DMA engine? EMAC? GMAC?

I've tested with a Cubieboard2 Two-Card (on-board EMMC replaced with a second slot). I ran into and fixed DMA problems with stmmac (that's GMAC) and sunxi-mmc. Patches are submitted as part of this patchset. Endianness fixes to stmmac went much farther than just GMAC but remain untested since the Cubieboard2 only uses GMAC.

I have also successfully used SATA, SPI, UART, security system (with cryptodev), the hwmon part of the touchscreen controller, pinctrl with a custom interrupt pin via fdt for ethernet via SPI (ENC28J60), watchdog as reset controller, onboard RTC and external battery-backed DS1307 RTC via I2C. I have not deliberately/synthetically tested the DMA engine. It's enabled in the my kernel config and seems happy but was not selected by any other driver so seems to be unused currently.

I was reluctant to enable big endian support this generically but found no better way. Since board specifics are in device tree now, I can't really do a && CONFIG_MACH_CUBIEBOARD2 or somesuch. Or can I?

Signature · 2 lines
Thanks,
Michael
#10 lore raw

PATCH 3/3 ARM: sunxi: enable big-endian

Quoted 15 lines · Michael Weiser

Hi Maxime,

Quoted 4 lines · Maxime Ripard
Quoted 2 lines

At least the Cubieboard2 incarnation of sunxi works correctly in big-endian mode. Allow the selection of big-endian.

With what devices has it been tested? The one doing DMA will probably be the one who fail, have you tested the DMA engine? EMAC? GMAC?

I've tested with a Cubieboard2 Two-Card (on-board EMMC replaced with a second slot). I ran into and fixed DMA problems with stmmac (that's GMAC) and sunxi-mmc. Patches are submitted as part of this patchset. Endianness fixes to stmmac went much farther than just GMAC but remain untested since the Cubieboard2 only uses GMAC.

Quoted 7 lines

I have also successfully used SATA, SPI, UART, security system (with cryptodev), the hwmon part of the touchscreen controller, pinctrl with a custom interrupt pin via fdt for ethernet via SPI (ENC28J60), watchdog as reset controller, onboard RTC and external battery-backed DS1307 RTC via I2C. I have not deliberately/synthetically tested the DMA engine. It's enabled in the my kernel config and seems happy but was not selected by any other driver so seems to be unused currently.

You can test the DMA using the audio components, either the codec, i2s or spdif. EMAC would be worth testing too.

Quoted 3 lines

I was reluctant to enable big endian support this generically but found no better way. Since board specifics are in device tree now, I can't really do a && CONFIG_MACH_CUBIEBOARD2 or somesuch. Or can I?

Not really, but it doesn't really matter, since all those drivers are at the SoC level and should affect all boards equally. And you have MACH_SUN7I :)

Maxime

Signature · 10 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20160725/5783092e/attachment.sig>
#11 lore raw

PATCH v2 3/3 ARM: sunxi: enable big-endian

Hi Maxime,

Quoted 5 lines · Maxime Ripard
Quoted 3 lines

via I2C. I have not deliberately/synthetically tested the DMA engine. It's enabled in the my kernel config and seems happy but was not selected by any other driver so seems to be unused currently.

You can test the DMA using the audio components, either the codec, i2s or spdif.

I've enabled the codec and it's currently playing some King Khan to me.

#12 lore raw

PATCH 3/3 ARM: sunxi: enable big-endian

Quoted 15 lines · Michael Weiser

Hi Maxime,

Quoted 5 lines · Maxime Ripard
Quoted 3 lines

via I2C. I have not deliberately/synthetically tested the DMA engine. It's enabled in the my kernel config and seems happy but was not selected by any other driver so seems to be unused currently.

You can test the DMA using the audio components, either the codec, i2s or spdif.

I've enabled the codec and it's currently playing some King Khan to me. From looking at sun4i-dma.c it seems that all communication with the silicon is via registers, including transfer of the DMA descriptors, and all is done using {read,write}l{,_relaxed} already.

Quoted 4 lines
Quoted 1 line

EMAC would be worth testing too.

As far as I gather, the Cubieboard2 does not route the EMAC pins anywhere. So I can't connect a PHY to it for testing.

The EMAC pins are muxed with the GMAC pins. All you need to do is disable the GMAC, enable the EMAC, and use emac_pins_a for the pinctrl.

ChenYu

#13 lore raw

PATCH v2 3/3 ARM: sunxi: enable big-endian

Hi ChenYu,

Quoted 4 lines · Chen-Yu Tsai
Quoted 2 lines

As far as I gather, the Cubieboard2 does not route the EMAC pins anywhere. So I can't connect a PHY to it for testing.

The EMAC pins are muxed with the GMAC pins. All you need to do is disable the GMAC, enable the EMAC, and use emac_pins_a for the pinctrl.

Okay, I've done that by copying the relevant dts sections from sun4i-a10-cubieboard.dts. It initialises and transmits packets fine. But it doesn't seem to accept incoming packets. I suspect that programming of its own MAC address (which the driver chooses randomly) suffers from an endianness issue. I'm somewhat stumped as to what is causing it since all accesses use {read,write}l. Interrupt handling is another possible culprit. I'm still digging.

Signature · 2 lines
Thanks,
Michael
#14 lore raw

PATCH 3/3 ARM: sunxi: enable big-endian

Hi,

Quoted 8 lines · Michael Weiser

Hi ChenYu,

Quoted 4 lines · Chen-Yu Tsai
Quoted 2 lines

As far as I gather, the Cubieboard2 does not route the EMAC pins anywhere. So I can't connect a PHY to it for testing.

The EMAC pins are muxed with the GMAC pins. All you need to do is disable the GMAC, enable the EMAC, and use emac_pins_a for the pinctrl.

Quoted 7 lines

Okay, I've done that by copying the relevant dts sections from sun4i-a10-cubieboard.dts. It initialises and transmits packets fine. But it doesn't seem to accept incoming packets. I suspect that programming of its own MAC address (which the driver chooses randomly) suffers from an endianness issue. I'm somewhat stumped as to what is causing it since all accesses use {read,write}l. Interrupt handling is another possible culprit. I'm still digging.

Maybe a good way to confirm that would be to generate outgoing traffic, dump that and see what the mac address is? If it's not programmed properly, you should see it in the packets.

Maxime

Signature · 10 lines
Maxime Ripard, Free Electrons
Embedded Linux and Kernel engineering
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20160729/ba256feb/attachment-0001.sig>
#15 lore raw

PATCH v2 3/3 ARM: sunxi: enable big-endian

Hi Maxime,

Quoted 6 lines · Maxime Ripard
Quoted 3 lines

it doesn't seem to accept incoming packets. I suspect that programming of its own MAC address (which the driver chooses randomly) suffers from an endianness issue. I'm somewhat stumped as to what is causing it since

Maybe a good way to confirm that would be to generate outgoing traffic, dump that and see what the mac address is? If it's not programmed properly, you should see it in the packets.

That's what I've done: When trying to ping another machine, with tcpdump I see ARP requests coming from the cubieboard with the right MAC and I see the target machine sending ARP responses to the right MAC. But I do not see them coming in with tcpdump on the cubieboard.

The same process works fine with a little-endian kernel (same .config and dtb) and userland, so I'm sure it's an endianess issue somewhere.

I'll do some more comparison of little- and big-endian kernel behaviour and dig a bit deeper into the code.

Signature · 2 lines
Thanks,
Michael