Skip to contentSkip to conversation

at_hdmac: move to generic DMA binding

RFC PATCH Superseded ludovic.desroches at atmel.com 2 participants 4 messages Apr 8, 2013

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

RFC PATCH v2 1/1 at_hdmac: move to generic DMA binding

From: Ludovic Desroches <[email protected]>

---

Hi,

Here is a second try to move at_hdmac to generic DMA binding. I have updated bindings according to Arnd comments ie I have removed chunk transfer size.

I have added the implementation but I am not very happy with the translation function. I have tried to not break old stuff: slave ask for a channel giving an atslave structure (passed through pdata) as a parameter for the filter function which saves this structure into chan->private. Then chan->private contains the configuration for the channel CFG register which is done when calling device_alloc_chan_resources.

If I allocate the atslave structure in the xlate function, where should be the right place to deallocate it? For the moment, I choose to add the atslave structure to the at_dma_chan structure but I am not happy with writing the channel configuration register into the xlate function. I would like to keep the same path as before.

Regards

Ludovic

 .../devicetree/bindings/dma/atmel-dma.txt          | 27 +++++++-
 arch/arm/boot/dts/sama5d3.dtsi                     |  2 +
 drivers/dma/at_hdmac.c                             | 71 ++++++++++++++++++++--
 drivers/dma/at_hdmac_regs.h                        |  7 +++
 4 files changed, 99 insertions(+), 8 deletions(-)
Documentation/devicetree/bindings/dma/atmel-dma.txt +24 -2
index 3c046ee..2d6f0f3 100644
@@ -4,11 +4,34 @@ Required properties:
 - compatible: Should be "atmel,<chip>-dma"
 - reg: Should contain DMA registers location and length
 - interrupts: Should contain DMA interrupt
+- #dma-cells: Must be <2>
 
-Examples:
+Example:
 
-dma at ffffec00 {
+dma0: dma at ffffec00 {
 	compatible = "atmel,at91sam9g45-dma";
 	reg = <0xffffec00 0x200>;
 	interrupts = <21>;
+	#dma-cells = <2>;
+};
+
+DMA clients connected to the Atmel DMA controller must use the format
+described in the dma.txt file, using a three-cell specifier for each channel.
+The three cells in order are:
+
+1. A phandle pointing to the DMA controller
+2. The memory interface (16 most significant bits), the peripheral interface
+(16 less significant bits)
+3. The peripheral identifier (can be different for tx and rx)
+
+Example:
+
+i2c0 at i2c@f8010000 {
+	compatible = "atmel,at91sam9x5-i2c";
+	reg = <0xf8010000 0x100>;
+	interrupts = <9 4 6>;
+	dmas = <&dma0 1 7>,
+	       <&dma0 1 8>;
+	dma-names = "tx", "rx"
 };
arch/arm/boot/dts/sama5d3.dtsi +2 -0
index 39b0458..9232143 100644
@@ -349,6 +349,7 @@
 				reg = <0xffffe600 0x200>;
 				interrupts = <30 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;
 			};
 
 			dma1: dma-controller at ffffe800 {
@@ -356,6 +357,7 @@
 				reg = <0xffffe800 0x200>;
 				interrupts = <31 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;
 			};
 
 			ramc0: ramc at ffffea00 {
drivers/dma/at_hdmac.c +65 -6

113 lines not on this page · raw message

drivers/dma/at_hdmac_regs.h +7 -0
index 0eb3c13..e3d2f12 100644
@@ -220,6 +220,8 @@ enum atc_status {
  * @device: parent device
  * @ch_regs: memory mapped register base
  * @mask: channel index in a mask
+ * @per_if: peripheral interface
+ * @mem_if: memory interface
  * @status: transmit status information from irq/prep* functions
  *                to tasklet (use atomic operations)
  * @tasklet: bottom half to finish transaction work
@@ -227,6 +229,8 @@ enum atc_status {
  * @save_dscr: for cyclic operations, preserve next descriptor address in
  *             the cyclic list on suspend/resume cycle
  * @dma_sconfig: configuration for slave transfers, passed via DMA_SLAVE_CONFIG
+ * @atslave: hardware configuration for slave transfers which is not passed via
+ * DMA_SLAVE_CONFIG
  * @lock: serializes enqueue/dequeue operations to descriptors lists
  * @active_list: list of descriptors dmaengine is being running on
  * @queue: list of descriptors ready to be submitted to engine
@@ -238,11 +242,14 @@ struct at_dma_chan {
 	struct at_dma		*device;
 	void __iomem		*ch_regs;
 	u8			mask;
+	u8			per_if;
+	u8			mem_if;
 	unsigned long		status;
 	struct tasklet_struct	tasklet;
 	u32			save_cfg;
 	u32			save_dscr;
 	struct dma_slave_config dma_sconfig;
+	struct at_dma_slave	atslave;
 
 	spinlock_t		lock;
 
Signature · 1 line
1.7.11.3
#1 lore raw

RFC PATCH v2 1/1 at_hdmac: move to generic DMA binding

Quoted 17 lines · ludovic.desroches at atmel.com

From: Ludovic Desroches <[email protected]>

---

Hi,

Here is a second try to move at_hdmac to generic DMA binding. I have updated bindings according to Arnd comments ie I have removed chunk transfer size.

I have added the implementation but I am not very happy with the translation function. I have tried to not break old stuff: slave ask for a channel giving an atslave structure (passed through pdata) as a parameter for the filter function which saves this structure into chan->private. Then chan->private contains the configuration for the channel CFG register which is done when calling device_alloc_chan_resources.

Quoted 5 lines

If I allocate the atslave structure in the xlate function, where should be the right place to deallocate it? For the moment, I choose to add the atslave structure to the at_dma_chan structure but I am not happy with writing the channel configuration register into the xlate function. I would like to keep the same path as before.

so allow with devm_

Quoted 57 lines

Regards

Ludovic

 .../devicetree/bindings/dma/atmel-dma.txt          | 27 +++++++-
 arch/arm/boot/dts/sama5d3.dtsi                     |  2 +
 drivers/dma/at_hdmac.c                             | 71 ++++++++++++++++++++--
 drivers/dma/at_hdmac_regs.h                        |  7 +++
 4 files changed, 99 insertions(+), 8 deletions(-)
Documentation/devicetree/bindings/dma/atmel-dma.txt +24 -2
index 3c046ee..2d6f0f3 100644
@@ -4,11 +4,34 @@ Required properties:
 - compatible: Should be "atmel,<chip>-dma"
 - reg: Should contain DMA registers location and length
 - interrupts: Should contain DMA interrupt
+- #dma-cells: Must be <2>
 
-Examples:
+Example:
 
-dma at ffffec00 {
+dma0: dma at ffffec00 {
 	compatible = "atmel,at91sam9g45-dma";
 	reg = <0xffffec00 0x200>;
 	interrupts = <21>;
+	#dma-cells = <2>;
+};
+
+DMA clients connected to the Atmel DMA controller must use the format
+described in the dma.txt file, using a three-cell specifier for each channel.
+The three cells in order are:
+
+1. A phandle pointing to the DMA controller
+2. The memory interface (16 most significant bits), the peripheral interface
+(16 less significant bits)
+3. The peripheral identifier (can be different for tx and rx)
+
+Example:
+
+i2c0 at i2c@f8010000 {
+	compatible = "atmel,at91sam9x5-i2c";
+	reg = <0xf8010000 0x100>;
+	interrupts = <9 4 6>;
+	dmas = <&dma0 1 7>,
+	       <&dma0 1 8>;
+	dma-names = "tx", "rx"
 };
arch/arm/boot/dts/sama5d3.dtsi +0 -0
index 39b0458..9232143 100644

arch/arm/boot/dts/sama5d3.dtsi line 352

Quoted 4 lines
arch/arm/boot/dts/sama5d3.dtsi +1 -0
@@ -349,6 +349,7 @@
 				reg = <0xffffe600 0x200>;
 				interrupts = <30 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;

??

2 dma-cells?

arch/arm/boot/dts/sama5d3.dtsi line 360

Quoted 8 lines
arch/arm/boot/dts/sama5d3.dtsi +1 -0
 			};
 
 			dma1: dma-controller at ffffe800 {
@@ -356,6 +357,7 @@
 				reg = <0xffffe800 0x200>;
 				interrupts = <31 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;

ditto

Quoted 112 lines
arch/arm/boot/dts/sama5d3.dtsi +0 -0
 			};
 
 			ramc0: ramc at ffffea00 {
drivers/dma/at_hdmac.c +56 -6
index 8415467..ec7f561 100644
@@ -24,6 +24,7 @@
 #include <linux/slab.h>
 #include <linux/of.h>
 #include <linux/of_device.h>
+#include <linux/of_dma.h>
 
 #include "at_hdmac_regs.h"
 #include "dmaengine.h"
@@ -676,7 +677,7 @@ atc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		ctrlb |=  ATC_DST_ADDR_MODE_FIXED
 			| ATC_SRC_ADDR_MODE_INCR
 			| ATC_FC_MEM2PER
-			| ATC_SIF(AT_DMA_MEM_IF) | ATC_DIF(AT_DMA_PER_IF);
+			| ATC_SIF(atchan->mem_if) | ATC_DIF(atchan->per_if);
 		reg = sconfig->dst_addr;
 		for_each_sg(sgl, sg, sg_len, i) {
 			struct at_desc	*desc;
@@ -715,7 +716,7 @@ atc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		ctrlb |=  ATC_DST_ADDR_MODE_INCR
 			| ATC_SRC_ADDR_MODE_FIXED
 			| ATC_FC_PER2MEM
-			| ATC_SIF(AT_DMA_PER_IF) | ATC_DIF(AT_DMA_MEM_IF);
+			| ATC_SIF(atchan->per_if) | ATC_DIF(atchan->mem_if);
 
 		reg = sconfig->src_addr;
 		for_each_sg(sgl, sg, sg_len, i) {
@@ -821,8 +822,8 @@ atc_dma_cyclic_fill_desc(struct dma_chan *chan, struct at_desc *desc,
 		desc->lli.ctrlb = ATC_DST_ADDR_MODE_FIXED
 				| ATC_SRC_ADDR_MODE_INCR
 				| ATC_FC_MEM2PER
-				| ATC_SIF(AT_DMA_MEM_IF)
-				| ATC_DIF(AT_DMA_PER_IF);
+				| ATC_SIF(atchan->mem_if)
+				| ATC_DIF(atchan->per_if);
 		break;
 
 	case DMA_DEV_TO_MEM:
@@ -832,8 +833,8 @@ atc_dma_cyclic_fill_desc(struct dma_chan *chan, struct at_desc *desc,
 		desc->lli.ctrlb = ATC_DST_ADDR_MODE_INCR
 				| ATC_SRC_ADDR_MODE_FIXED
 				| ATC_FC_PER2MEM
-				| ATC_SIF(AT_DMA_PER_IF)
-				| ATC_DIF(AT_DMA_MEM_IF);
+				| ATC_SIF(atchan->per_if)
+				| ATC_DIF(atchan->mem_if);
 		break;
 
 	default:
@@ -1189,6 +1190,55 @@ static void atc_free_chan_resources(struct dma_chan *chan)
 	dev_vdbg(chan2dev(chan), "free_chan_resources: done\n");
 }
 
+static bool at_dma_filter(struct dma_chan *chan, void *dma_dev)
+{
+	if (dma_dev == chan->device->dev)
+		return true;
+	else
+		return false;
+}
+
+static struct dma_chan *at_dma_xlate(struct of_phandle_args *dma_spec,
+				     struct of_dma *of_dma)
+{
+	struct dma_chan *chan;
+	struct at_dma_chan *atchan;
+	struct at_dma_slave *atslave;
+	dma_cap_mask_t mask;
+	unsigned int per_id;
+	struct platform_device *dmac_pdev;
+
+	if (dma_spec->args_count != 2)
+		return NULL;
+
+	dmac_pdev = of_find_device_by_node(dma_spec->np);
+
+	dma_cap_zero(mask);
+	dma_cap_set(DMA_SLAVE, mask);
+
+	chan = dma_request_channel(mask, at_dma_filter, &dmac_pdev->dev);
+	if (!chan)
+		return NULL;
+
+	atchan = to_at_dma_chan(chan);
+	atchan->per_if = dma_spec->args[0] & 0xff;
+	atchan->mem_if = (dma_spec->args[0] >> 16) & 0xff;
+
+	atslave = &atchan->atslave;
+	/*
+	 * We can fill both SRC_PER and DST_PER, one of these fields will be
+	 * ignored depending on DMA transfer direction.
+	 */
+	per_id = dma_spec->args[1];
+	atslave->cfg = ATC_FIFOCFG_HALFFIFO | ATC_DST_H2SEL_HW
+		      | ATC_SRC_H2SEL_HW | ATC_DST_PER(per_id)
+		      | ATC_SRC_PER(per_id);
+	atslave->dma_dev = &dmac_pdev->dev;
+	chan->private = atslave;
+	channel_writel(atchan, CFG, atslave->cfg);
+
+	return chan;
+}
 
 /*--  Module Management  -----------------------------------------------*/
 

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1442

Quoted 4 lines
drivers/dma/at_hdmac.c +1 -0
@@ -1389,6 +1439,15 @@ static int __init at_dma_probe(struct platform_device *pdev)
 
 	dma_async_device_register(&atdma->dma_common);
 
+	if (pdev->dev.of_node) {

more simple return if NULL

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1443

Quoted 1 line
drivers/dma/at_hdmac.c +1 -0
+		printk("=== of_dma_controller_register ===\n");

drop this

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1448

Quoted 5 lines
drivers/dma/at_hdmac.c +5 -0
+		err = of_dma_controller_register(pdev->dev.of_node,
+						 at_dma_xlate, atdma);
+		if (err && err != -ENODEV)
+			dev_err(&pdev->dev,
+				"could not register of_dma_controller\n");

if err you need to return err

Quoted 45 lines
drivers/dma/at_hdmac.c +2 -0
+	}
+
 	return 0;
 
 err_pool_create:
drivers/dma/at_hdmac_regs.h +7 -0
index 0eb3c13..e3d2f12 100644
@@ -220,6 +220,8 @@ enum atc_status {
  * @device: parent device
  * @ch_regs: memory mapped register base
  * @mask: channel index in a mask
+ * @per_if: peripheral interface
+ * @mem_if: memory interface
  * @status: transmit status information from irq/prep* functions
  *                to tasklet (use atomic operations)
  * @tasklet: bottom half to finish transaction work
@@ -227,6 +229,8 @@ enum atc_status {
  * @save_dscr: for cyclic operations, preserve next descriptor address in
  *             the cyclic list on suspend/resume cycle
  * @dma_sconfig: configuration for slave transfers, passed via DMA_SLAVE_CONFIG
+ * @atslave: hardware configuration for slave transfers which is not passed via
+ * DMA_SLAVE_CONFIG
  * @lock: serializes enqueue/dequeue operations to descriptors lists
  * @active_list: list of descriptors dmaengine is being running on
  * @queue: list of descriptors ready to be submitted to engine
@@ -238,11 +242,14 @@ struct at_dma_chan {
 	struct at_dma		*device;
 	void __iomem		*ch_regs;
 	u8			mask;
+	u8			per_if;
+	u8			mem_if;
 	unsigned long		status;
 	struct tasklet_struct	tasklet;
 	u32			save_cfg;
 	u32			save_dscr;
 	struct dma_slave_config dma_sconfig;
+	struct at_dma_slave	atslave;
 
 	spinlock_t		lock;
 
Signature · 1 line
1.7.11.3
#2 lore raw

RFC PATCH v2 1/1 at_hdmac: move to generic DMA binding

Quoted 18 lines · Jean-Christophe PLAGNIOL-VILLARD
Quoted 17 lines · ludovic.desroches at atmel.com

From: Ludovic Desroches <[email protected]>

---

Hi,

Here is a second try to move at_hdmac to generic DMA binding. I have updated bindings according to Arnd comments ie I have removed chunk transfer size.

I have added the implementation but I am not very happy with the translation function. I have tried to not break old stuff: slave ask for a channel giving an atslave structure (passed through pdata) as a parameter for the filter function which saves this structure into chan->private. Then chan->private contains the configuration for the channel CFG register which is done when calling device_alloc_chan_resources.

Quoted 6 lines
Quoted 5 lines

If I allocate the atslave structure in the xlate function, where should be the right place to deallocate it? For the moment, I choose to add the atslave structure to the at_dma_chan structure but I am not happy with writing the channel configuration register into the xlate function. I would like to keep the same path as before.

so allow with devm_

To my mind it's channel related and not controller related. If the channel is released, will the deallocation be done automatically?

Quoted 63 lines
Quoted 57 lines

Regards

Ludovic

 .../devicetree/bindings/dma/atmel-dma.txt          | 27 +++++++-
 arch/arm/boot/dts/sama5d3.dtsi                     |  2 +
 drivers/dma/at_hdmac.c                             | 71 ++++++++++++++++++++--
 drivers/dma/at_hdmac_regs.h                        |  7 +++
 4 files changed, 99 insertions(+), 8 deletions(-)
Documentation/devicetree/bindings/dma/atmel-dma.txt +24 -2
index 3c046ee..2d6f0f3 100644
@@ -4,11 +4,34 @@ Required properties:
 - compatible: Should be "atmel,<chip>-dma"
 - reg: Should contain DMA registers location and length
 - interrupts: Should contain DMA interrupt
+- #dma-cells: Must be <2>
 
-Examples:
+Example:
 
-dma at ffffec00 {
+dma0: dma at ffffec00 {
 	compatible = "atmel,at91sam9g45-dma";
 	reg = <0xffffec00 0x200>;
 	interrupts = <21>;
+	#dma-cells = <2>;
+};
+
+DMA clients connected to the Atmel DMA controller must use the format
+described in the dma.txt file, using a three-cell specifier for each channel.
+The three cells in order are:
+
+1. A phandle pointing to the DMA controller
+2. The memory interface (16 most significant bits), the peripheral interface
+(16 less significant bits)
+3. The peripheral identifier (can be different for tx and rx)
+
+Example:
+
+i2c0 at i2c@f8010000 {
+	compatible = "atmel,at91sam9x5-i2c";
+	reg = <0xf8010000 0x100>;
+	interrupts = <9 4 6>;
+	dmas = <&dma0 1 7>,
+	       <&dma0 1 8>;
+	dma-names = "tx", "rx"
 };
arch/arm/boot/dts/sama5d3.dtsi +0 -0
index 39b0458..9232143 100644

arch/arm/boot/dts/sama5d3.dtsi line 352

Quoted 4 lines
arch/arm/boot/dts/sama5d3.dtsi +1 -0
@@ -349,6 +349,7 @@
 				reg = <0xffffe600 0x200>;
 				interrupts = <30 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;

??

Quoted 1 line

2 dma-cells?

Yes 2 dma-cells, mistake in the patch, the work is still in progress.

Quoted 120 lines

arch/arm/boot/dts/sama5d3.dtsi line 360

Quoted 8 lines
arch/arm/boot/dts/sama5d3.dtsi +1 -0
 			};
 
 			dma1: dma-controller at ffffe800 {
@@ -356,6 +357,7 @@
 				reg = <0xffffe800 0x200>;
 				interrupts = <31 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;

ditto

Quoted 111 lines
arch/arm/boot/dts/sama5d3.dtsi +0 -0
 			};
 
 			ramc0: ramc at ffffea00 {
drivers/dma/at_hdmac.c +56 -6
index 8415467..ec7f561 100644
@@ -24,6 +24,7 @@
 #include <linux/slab.h>
 #include <linux/of.h>
 #include <linux/of_device.h>
+#include <linux/of_dma.h>
 
 #include "at_hdmac_regs.h"
 #include "dmaengine.h"
@@ -676,7 +677,7 @@ atc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		ctrlb |=  ATC_DST_ADDR_MODE_FIXED
 			| ATC_SRC_ADDR_MODE_INCR
 			| ATC_FC_MEM2PER
-			| ATC_SIF(AT_DMA_MEM_IF) | ATC_DIF(AT_DMA_PER_IF);
+			| ATC_SIF(atchan->mem_if) | ATC_DIF(atchan->per_if);
 		reg = sconfig->dst_addr;
 		for_each_sg(sgl, sg, sg_len, i) {
 			struct at_desc	*desc;
@@ -715,7 +716,7 @@ atc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		ctrlb |=  ATC_DST_ADDR_MODE_INCR
 			| ATC_SRC_ADDR_MODE_FIXED
 			| ATC_FC_PER2MEM
-			| ATC_SIF(AT_DMA_PER_IF) | ATC_DIF(AT_DMA_MEM_IF);
+			| ATC_SIF(atchan->per_if) | ATC_DIF(atchan->mem_if);
 
 		reg = sconfig->src_addr;
 		for_each_sg(sgl, sg, sg_len, i) {
@@ -821,8 +822,8 @@ atc_dma_cyclic_fill_desc(struct dma_chan *chan, struct at_desc *desc,
 		desc->lli.ctrlb = ATC_DST_ADDR_MODE_FIXED
 				| ATC_SRC_ADDR_MODE_INCR
 				| ATC_FC_MEM2PER
-				| ATC_SIF(AT_DMA_MEM_IF)
-				| ATC_DIF(AT_DMA_PER_IF);
+				| ATC_SIF(atchan->mem_if)
+				| ATC_DIF(atchan->per_if);
 		break;
 
 	case DMA_DEV_TO_MEM:
@@ -832,8 +833,8 @@ atc_dma_cyclic_fill_desc(struct dma_chan *chan, struct at_desc *desc,
 		desc->lli.ctrlb = ATC_DST_ADDR_MODE_INCR
 				| ATC_SRC_ADDR_MODE_FIXED
 				| ATC_FC_PER2MEM
-				| ATC_SIF(AT_DMA_PER_IF)
-				| ATC_DIF(AT_DMA_MEM_IF);
+				| ATC_SIF(atchan->per_if)
+				| ATC_DIF(atchan->mem_if);
 		break;
 
 	default:
@@ -1189,6 +1190,55 @@ static void atc_free_chan_resources(struct dma_chan *chan)
 	dev_vdbg(chan2dev(chan), "free_chan_resources: done\n");
 }
 
+static bool at_dma_filter(struct dma_chan *chan, void *dma_dev)
+{
+	if (dma_dev == chan->device->dev)
+		return true;
+	else
+		return false;
+}
+
+static struct dma_chan *at_dma_xlate(struct of_phandle_args *dma_spec,
+				     struct of_dma *of_dma)
+{
+	struct dma_chan *chan;
+	struct at_dma_chan *atchan;
+	struct at_dma_slave *atslave;
+	dma_cap_mask_t mask;
+	unsigned int per_id;
+	struct platform_device *dmac_pdev;
+
+	if (dma_spec->args_count != 2)
+		return NULL;
+
+	dmac_pdev = of_find_device_by_node(dma_spec->np);
+
+	dma_cap_zero(mask);
+	dma_cap_set(DMA_SLAVE, mask);
+
+	chan = dma_request_channel(mask, at_dma_filter, &dmac_pdev->dev);
+	if (!chan)
+		return NULL;
+
+	atchan = to_at_dma_chan(chan);
+	atchan->per_if = dma_spec->args[0] & 0xff;
+	atchan->mem_if = (dma_spec->args[0] >> 16) & 0xff;
+
+	atslave = &atchan->atslave;
+	/*
+	 * We can fill both SRC_PER and DST_PER, one of these fields will be
+	 * ignored depending on DMA transfer direction.
+	 */
+	per_id = dma_spec->args[1];
+	atslave->cfg = ATC_FIFOCFG_HALFFIFO | ATC_DST_H2SEL_HW
+		      | ATC_SRC_H2SEL_HW | ATC_DST_PER(per_id)
+		      | ATC_SRC_PER(per_id);
+	atslave->dma_dev = &dmac_pdev->dev;
+	chan->private = atslave;
+	channel_writel(atchan, CFG, atslave->cfg);
+
+	return chan;
+}
 
 /*--  Module Management  -----------------------------------------------*/
 
Quoted 8 lines

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1442

Quoted 5 lines
drivers/dma/at_hdmac.c +1 -0
@@ -1389,6 +1439,15 @@ static int __init at_dma_probe(struct platform_device *pdev)
 
 	dma_async_device_register(&atdma->dma_common);
 
+	if (pdev->dev.of_node) {

more simple return if NULL

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1443

Quoted 1 line
drivers/dma/at_hdmac.c +1 -0
+		printk("=== of_dma_controller_register ===\n");

drop this

of course, I'll drop it for the submission

Quoted 6 lines

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1448

Quoted 5 lines
drivers/dma/at_hdmac.c +5 -0
+		err = of_dma_controller_register(pdev->dev.of_node,
+						 at_dma_xlate, atdma);
+		if (err && err != -ENODEV)
+			dev_err(&pdev->dev,
+				"could not register of_dma_controller\n");

if err you need to return err

ok

Quoted 45 lines
Quoted 45 lines
drivers/dma/at_hdmac.c +2 -0
+	}
+
 	return 0;
 
 err_pool_create:
drivers/dma/at_hdmac_regs.h +7 -0
index 0eb3c13..e3d2f12 100644
@@ -220,6 +220,8 @@ enum atc_status {
  * @device: parent device
  * @ch_regs: memory mapped register base
  * @mask: channel index in a mask
+ * @per_if: peripheral interface
+ * @mem_if: memory interface
  * @status: transmit status information from irq/prep* functions
  *                to tasklet (use atomic operations)
  * @tasklet: bottom half to finish transaction work
@@ -227,6 +229,8 @@ enum atc_status {
  * @save_dscr: for cyclic operations, preserve next descriptor address in
  *             the cyclic list on suspend/resume cycle
  * @dma_sconfig: configuration for slave transfers, passed via DMA_SLAVE_CONFIG
+ * @atslave: hardware configuration for slave transfers which is not passed via
+ * DMA_SLAVE_CONFIG
  * @lock: serializes enqueue/dequeue operations to descriptors lists
  * @active_list: list of descriptors dmaengine is being running on
  * @queue: list of descriptors ready to be submitted to engine
@@ -238,11 +242,14 @@ struct at_dma_chan {
 	struct at_dma		*device;
 	void __iomem		*ch_regs;
 	u8			mask;
+	u8			per_if;
+	u8			mem_if;
 	unsigned long		status;
 	struct tasklet_struct	tasklet;
 	u32			save_cfg;
 	u32			save_dscr;
 	struct dma_slave_config dma_sconfig;
+	struct at_dma_slave	atslave;
 
 	spinlock_t		lock;
 
Signature · 1 line
1.7.11.3
#3 lore raw

RFC PATCH v2 1/1 at_hdmac: move to generic DMA binding

Quoted 24 lines · Ludovic Desroches
Quoted 23 lines · Jean-Christophe PLAGNIOL-VILLARD
Quoted 22 lines · ludovic.desroches at atmel.com

From: Ludovic Desroches <[email protected]>

---

Hi,

Here is a second try to move at_hdmac to generic DMA binding. I have updated bindings according to Arnd comments ie I have removed chunk transfer size.

I have added the implementation but I am not very happy with the translation function. I have tried to not break old stuff: slave ask for a channel giving an atslave structure (passed through pdata) as a parameter for the filter function which saves this structure into chan->private. Then chan->private contains the configuration for the channel CFG register which is done when calling device_alloc_chan_resources.

If I allocate the atslave structure in the xlate function, where should be the right place to deallocate it? For the moment, I choose to add the atslave structure to the at_dma_chan structure but I am not happy with writing the channel configuration register into the xlate function. I would like to keep the same path as before.

Quoted 4 lines
Quoted 1 line

so allow with devm_

To my mind it's channel related and not controller related. If the channel is released, will the deallocation be done automatically?

when the device is free yes

so free on slave device it's ok

Best Regards,
J.

Quoted 253 lines
Quoted 64 lines
Quoted 58 lines

Regards

Ludovic

 .../devicetree/bindings/dma/atmel-dma.txt          | 27 +++++++-
 arch/arm/boot/dts/sama5d3.dtsi                     |  2 +
 drivers/dma/at_hdmac.c                             | 71 ++++++++++++++++++++--
 drivers/dma/at_hdmac_regs.h                        |  7 +++
 4 files changed, 99 insertions(+), 8 deletions(-)
Documentation/devicetree/bindings/dma/atmel-dma.txt +24 -2
index 3c046ee..2d6f0f3 100644
@@ -4,11 +4,34 @@ Required properties:
 - compatible: Should be "atmel,<chip>-dma"
 - reg: Should contain DMA registers location and length
 - interrupts: Should contain DMA interrupt
+- #dma-cells: Must be <2>
 
-Examples:
+Example:
 
-dma at ffffec00 {
+dma0: dma at ffffec00 {
 	compatible = "atmel,at91sam9g45-dma";
 	reg = <0xffffec00 0x200>;
 	interrupts = <21>;
+	#dma-cells = <2>;
+};
+
+DMA clients connected to the Atmel DMA controller must use the format
+described in the dma.txt file, using a three-cell specifier for each channel.
+The three cells in order are:
+
+1. A phandle pointing to the DMA controller
+2. The memory interface (16 most significant bits), the peripheral interface
+(16 less significant bits)
+3. The peripheral identifier (can be different for tx and rx)
+
+Example:
+
+i2c0 at i2c@f8010000 {
+	compatible = "atmel,at91sam9x5-i2c";
+	reg = <0xf8010000 0x100>;
+	interrupts = <9 4 6>;
+	dmas = <&dma0 1 7>,
+	       <&dma0 1 8>;
+	dma-names = "tx", "rx"
 };
arch/arm/boot/dts/sama5d3.dtsi +0 -0
index 39b0458..9232143 100644

arch/arm/boot/dts/sama5d3.dtsi line 352

Quoted 4 lines
arch/arm/boot/dts/sama5d3.dtsi +1 -0
@@ -349,6 +349,7 @@
 				reg = <0xffffe600 0x200>;
 				interrupts = <30 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;

??

Quoted 1 line

2 dma-cells?

Yes 2 dma-cells, mistake in the patch, the work is still in progress.

Quoted 120 lines

arch/arm/boot/dts/sama5d3.dtsi line 360

Quoted 8 lines
arch/arm/boot/dts/sama5d3.dtsi +1 -0
 			};
 
 			dma1: dma-controller at ffffe800 {
@@ -356,6 +357,7 @@
 				reg = <0xffffe800 0x200>;
 				interrupts = <31 4 0>;
 				#dma-cells = <1>;
+				#dma-cells = <2>;

ditto

Quoted 111 lines
arch/arm/boot/dts/sama5d3.dtsi +0 -0
 			};
 
 			ramc0: ramc at ffffea00 {
drivers/dma/at_hdmac.c +56 -6
index 8415467..ec7f561 100644
@@ -24,6 +24,7 @@
 #include <linux/slab.h>
 #include <linux/of.h>
 #include <linux/of_device.h>
+#include <linux/of_dma.h>
 
 #include "at_hdmac_regs.h"
 #include "dmaengine.h"
@@ -676,7 +677,7 @@ atc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		ctrlb |=  ATC_DST_ADDR_MODE_FIXED
 			| ATC_SRC_ADDR_MODE_INCR
 			| ATC_FC_MEM2PER
-			| ATC_SIF(AT_DMA_MEM_IF) | ATC_DIF(AT_DMA_PER_IF);
+			| ATC_SIF(atchan->mem_if) | ATC_DIF(atchan->per_if);
 		reg = sconfig->dst_addr;
 		for_each_sg(sgl, sg, sg_len, i) {
 			struct at_desc	*desc;
@@ -715,7 +716,7 @@ atc_prep_slave_sg(struct dma_chan *chan, struct scatterlist *sgl,
 		ctrlb |=  ATC_DST_ADDR_MODE_INCR
 			| ATC_SRC_ADDR_MODE_FIXED
 			| ATC_FC_PER2MEM
-			| ATC_SIF(AT_DMA_PER_IF) | ATC_DIF(AT_DMA_MEM_IF);
+			| ATC_SIF(atchan->per_if) | ATC_DIF(atchan->mem_if);
 
 		reg = sconfig->src_addr;
 		for_each_sg(sgl, sg, sg_len, i) {
@@ -821,8 +822,8 @@ atc_dma_cyclic_fill_desc(struct dma_chan *chan, struct at_desc *desc,
 		desc->lli.ctrlb = ATC_DST_ADDR_MODE_FIXED
 				| ATC_SRC_ADDR_MODE_INCR
 				| ATC_FC_MEM2PER
-				| ATC_SIF(AT_DMA_MEM_IF)
-				| ATC_DIF(AT_DMA_PER_IF);
+				| ATC_SIF(atchan->mem_if)
+				| ATC_DIF(atchan->per_if);
 		break;
 
 	case DMA_DEV_TO_MEM:
@@ -832,8 +833,8 @@ atc_dma_cyclic_fill_desc(struct dma_chan *chan, struct at_desc *desc,
 		desc->lli.ctrlb = ATC_DST_ADDR_MODE_INCR
 				| ATC_SRC_ADDR_MODE_FIXED
 				| ATC_FC_PER2MEM
-				| ATC_SIF(AT_DMA_PER_IF)
-				| ATC_DIF(AT_DMA_MEM_IF);
+				| ATC_SIF(atchan->per_if)
+				| ATC_DIF(atchan->mem_if);
 		break;
 
 	default:
@@ -1189,6 +1190,55 @@ static void atc_free_chan_resources(struct dma_chan *chan)
 	dev_vdbg(chan2dev(chan), "free_chan_resources: done\n");
 }
 
+static bool at_dma_filter(struct dma_chan *chan, void *dma_dev)
+{
+	if (dma_dev == chan->device->dev)
+		return true;
+	else
+		return false;
+}
+
+static struct dma_chan *at_dma_xlate(struct of_phandle_args *dma_spec,
+				     struct of_dma *of_dma)
+{
+	struct dma_chan *chan;
+	struct at_dma_chan *atchan;
+	struct at_dma_slave *atslave;
+	dma_cap_mask_t mask;
+	unsigned int per_id;
+	struct platform_device *dmac_pdev;
+
+	if (dma_spec->args_count != 2)
+		return NULL;
+
+	dmac_pdev = of_find_device_by_node(dma_spec->np);
+
+	dma_cap_zero(mask);
+	dma_cap_set(DMA_SLAVE, mask);
+
+	chan = dma_request_channel(mask, at_dma_filter, &dmac_pdev->dev);
+	if (!chan)
+		return NULL;
+
+	atchan = to_at_dma_chan(chan);
+	atchan->per_if = dma_spec->args[0] & 0xff;
+	atchan->mem_if = (dma_spec->args[0] >> 16) & 0xff;
+
+	atslave = &atchan->atslave;
+	/*
+	 * We can fill both SRC_PER and DST_PER, one of these fields will be
+	 * ignored depending on DMA transfer direction.
+	 */
+	per_id = dma_spec->args[1];
+	atslave->cfg = ATC_FIFOCFG_HALFFIFO | ATC_DST_H2SEL_HW
+		      | ATC_SRC_H2SEL_HW | ATC_DST_PER(per_id)
+		      | ATC_SRC_PER(per_id);
+	atslave->dma_dev = &dmac_pdev->dev;
+	chan->private = atslave;
+	channel_writel(atchan, CFG, atslave->cfg);
+
+	return chan;
+}
 
 /*--  Module Management  -----------------------------------------------*/
 
Quoted 8 lines

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1442

Quoted 5 lines
drivers/dma/at_hdmac.c +1 -0
@@ -1389,6 +1439,15 @@ static int __init at_dma_probe(struct platform_device *pdev)
 
 	dma_async_device_register(&atdma->dma_common);
 
+	if (pdev->dev.of_node) {

more simple return if NULL

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1443

Quoted 1 line
drivers/dma/at_hdmac.c +1 -0
+		printk("=== of_dma_controller_register ===\n");

drop this

of course, I'll drop it for the submission

Quoted 6 lines

drivers/dma/at_hdmac.c static int __init at_dma_probe(struct platform_device *pdev) · line 1448

Quoted 5 lines
drivers/dma/at_hdmac.c +5 -0
+		err = of_dma_controller_register(pdev->dev.of_node,
+						 at_dma_xlate, atdma);
+		if (err && err != -ENODEV)
+			dev_err(&pdev->dev,
+				"could not register of_dma_controller\n");

if err you need to return err

ok

Quoted 45 lines
Quoted 45 lines
drivers/dma/at_hdmac.c +2 -0
+	}
+
 	return 0;
 
 err_pool_create:
drivers/dma/at_hdmac_regs.h +7 -0
index 0eb3c13..e3d2f12 100644
@@ -220,6 +220,8 @@ enum atc_status {
  * @device: parent device
  * @ch_regs: memory mapped register base
  * @mask: channel index in a mask
+ * @per_if: peripheral interface
+ * @mem_if: memory interface
  * @status: transmit status information from irq/prep* functions
  *                to tasklet (use atomic operations)
  * @tasklet: bottom half to finish transaction work
@@ -227,6 +229,8 @@ enum atc_status {
  * @save_dscr: for cyclic operations, preserve next descriptor address in
  *             the cyclic list on suspend/resume cycle
  * @dma_sconfig: configuration for slave transfers, passed via DMA_SLAVE_CONFIG
+ * @atslave: hardware configuration for slave transfers which is not passed via
+ * DMA_SLAVE_CONFIG
  * @lock: serializes enqueue/dequeue operations to descriptors lists
  * @active_list: list of descriptors dmaengine is being running on
  * @queue: list of descriptors ready to be submitted to engine
@@ -238,11 +242,14 @@ struct at_dma_chan {
 	struct at_dma		*device;
 	void __iomem		*ch_regs;
 	u8			mask;
+	u8			per_if;
+	u8			mem_if;
 	unsigned long		status;
 	struct tasklet_struct	tasklet;
 	u32			save_cfg;
 	u32			save_dscr;
 	struct dma_slave_config dma_sconfig;
+	struct at_dma_slave	atslave;
 
 	spinlock_t		lock;
 
Signature · 1 line
1.7.11.3