Skip to contentSkip to conversation

Tegra GCPDMA: Add dma-channel-mask support

PATCH RESEND Superseded Akhil R 5 participants 11 messages Oct 20, 2022 – Nov 1, 2022

Subsystems (4)

v2 of 4 Superseded Latest is v4 →

lore
Share & tools
lore

Akhil R

author
#0 lore raw

v2 cover Tegra GCPDMA: Add dma-channel-mask support

Read dma-channel-mask from device tree and register only the specified channels. This is useful to reserve some channels for the firmware.

Also update the channel number and interrupts to include all 32 channels. The current driver was using only 31 channels as one channel was reserved for firmware. Now with this change, the driver can align more to the actual hardware.

v1->v2:
* Reversed the operands and used BIT macro in 'if' condition.
* Fixed warning reported-by: kernel test robot <[email protected]>

Akhil R (3):
dt-bindings: dmaengine: Add dma-channel-mask to Tegra GPCDMA
arm64: tegra: Add dma-channel-mask in GPCDMA node
dmaengine: tegra: Add support for dma-channel-mask

 .../bindings/dma/nvidia,tegra186-gpc-dma.yaml |  7 +++-
 arch/arm64/boot/dts/nvidia/tegra186.dtsi      |  4 +-
 arch/arm64/boot/dts/nvidia/tegra194.dtsi      |  4 +-
 arch/arm64/boot/dts/nvidia/tegra234.dtsi      |  4 +-
 drivers/dma/tegra186-gpc-dma.c                | 37 +++++++++++++++----
 5 files changed, 45 insertions(+), 11 deletions(-)
Signature · 1 line
2.17.1
author
#1 lore raw

PATCH RESEND v2 1/3 dt-bindings: dmaengine: Add dma-channel-mask to Tegra GPCDMA

Add dma-channel-mask property in Tegra GPCDMA document.

The property would help to specify the channels to be used in kernel and reserve few for the firmware. This was previously achieved by limiting the channel number to 31 in the driver. Now since we can list all 32 channels, update the interrupts property as well to list all 32 interrupts.

---
 .../devicetree/bindings/dma/nvidia,tegra186-gpc-dma.yaml   | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)
Documentation/devicetree/bindings/dma/nvidia,tegra186-gpc-dma.yaml +6 -1
index c8894476b6ab..851bd50ee67f 100644
@@ -39,7 +39,7 @@ properties:
       Should contain all of the per-channel DMA interrupts in
       ascending order with respect to the DMA channel index.
     minItems: 1
-    maxItems: 31
+    maxItems: 32
 
   resets:
     maxItems: 1
@@ -52,6 +52,9 @@ properties:
 
   dma-coherent: true
 
+  dma-channel-mask:
+    maxItems: 1
+
 required:
   - compatible
   - reg
@@ -60,6 +63,7 @@ required:
   - reset-names
   - "#dma-cells"
   - iommus
+  - dma-channel-mask
 
 additionalProperties: false
 
@@ -108,5 +112,6 @@ examples:
         #dma-cells = <1>;
         iommus = <&smmu TEGRA186_SID_GPCDMA_0>;
         dma-coherent;
+        dma-channel-mask = <0xfffffffe>;
     };
 ...
Signature · 1 line
2.17.1
Acked-by on v2 1/3
#2 lore raw

Re: [PATCH RESEND v2 1/3] dt-bindings: dmaengine: Add dma-channel-mask to Tegra GPCDMA

Quoted 11 lines · Akhil R

Add dma-channel-mask property in Tegra GPCDMA document.

The property would help to specify the channels to be used in kernel and reserve few for the firmware. This was previously achieved by limiting the channel number to 31 in the driver. Now since we can list all 32 channels, update the interrupts property as well to list all 32 interrupts.

---

Best regards,
Krzysztof

author
#3 lore raw

PATCH RESEND v2 2/3 arm64: tegra: Add dma-channel-mask in GPCDMA node

In reply to Akhil R

Add dma-channel-mask property in Tegra GPCDMA device tree node.

The property would help to specify the channels to be used in kernel and reserve few for the firmware. This was previously achieved by limiting the channel number to 31 in the driver. Now since we can list all 32 channels, update the interrupts property as well to list all 32 interrupts.

---
 arch/arm64/boot/dts/nvidia/tegra186.dtsi | 4 +++-
 arch/arm64/boot/dts/nvidia/tegra194.dtsi | 4 +++-
 arch/arm64/boot/dts/nvidia/tegra234.dtsi | 4 +++-
 3 files changed, 9 insertions(+), 3 deletions(-)
arch/arm64/boot/dts/nvidia/tegra186.dtsi +3 -1
index 6602fe421ee8..db479064ff72 100644
@@ -78,7 +78,8 @@
 		reg = <0x0 0x2600000 0x0 0x210000>;
 		resets = <&bpmp TEGRA186_RESET_GPCDMA>;
 		reset-names = "gpcdma";
-		interrupts = <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,
+		interrupts = <GIC_SPI 75 IRQ_TYPE_LEVEL_HIGH>,
+			     <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,
 			     <GIC_SPI 77 IRQ_TYPE_LEVEL_HIGH>,
 			     <GIC_SPI 78 IRQ_TYPE_LEVEL_HIGH>,
 			     <GIC_SPI 79 IRQ_TYPE_LEVEL_HIGH>,
@@ -112,6 +113,7 @@
 		#dma-cells = <1>;
 		iommus = <&smmu TEGRA186_SID_GPCDMA_0>;
 		dma-coherent;
+		dma-channel-mask = <0xfffffffe>;
 		status = "okay";
 	};
 
arch/arm64/boot/dts/nvidia/tegra194.dtsi +3 -1
index 41f3a7e188d0..b009f8145016 100644
@@ -142,7 +142,8 @@
 			reg = <0x2600000 0x210000>;
 			resets = <&bpmp TEGRA194_RESET_GPCDMA>;
 			reset-names = "gpcdma";
-			interrupts = <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,
+			interrupts = <GIC_SPI 75 IRQ_TYPE_LEVEL_HIGH>,
+				     <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,
 				     <GIC_SPI 77 IRQ_TYPE_LEVEL_HIGH>,
 				     <GIC_SPI 78 IRQ_TYPE_LEVEL_HIGH>,
 				     <GIC_SPI 79 IRQ_TYPE_LEVEL_HIGH>,
@@ -176,6 +177,7 @@
 			#dma-cells = <1>;
 			iommus = <&smmu TEGRA194_SID_GPCDMA_0>;
 			dma-coherent;
+			dma-channel-mask = <0xfffffffe>;
 			status = "okay";
 		};
 
arch/arm64/boot/dts/nvidia/tegra234.dtsi +3 -1
index 0170bfa8a467..ccc1a4bd094d 100644
@@ -27,7 +27,8 @@
 			reg = <0x2600000 0x210000>;
 			resets = <&bpmp TEGRA234_RESET_GPCDMA>;
 			reset-names = "gpcdma";
-			interrupts = <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,
+			interrupts = <GIC_SPI 75 IRQ_TYPE_LEVEL_HIGH>,
+				     <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,
 				     <GIC_SPI 77 IRQ_TYPE_LEVEL_HIGH>,
 				     <GIC_SPI 78 IRQ_TYPE_LEVEL_HIGH>,
 				     <GIC_SPI 79 IRQ_TYPE_LEVEL_HIGH>,
@@ -60,6 +61,7 @@
 				     <GIC_SPI 106 IRQ_TYPE_LEVEL_HIGH>;
 			#dma-cells = <1>;
 			iommus = <&smmu_niso0 TEGRA234_SID_GPCDMA>;
+			dma-channel-mask = <0xfffffffe>;
 			dma-coherent;
 		};
 
Signature · 1 line
2.17.1
maintainer
#4 lore raw

Re: [PATCH RESEND v2 2/3] arm64: tegra: Add dma-channel-mask in GPCDMA node

Quoted 20 lines · Akhil R

Add dma-channel-mask property in Tegra GPCDMA device tree node.

The property would help to specify the channels to be used in kernel and reserve few for the firmware. This was previously achieved by limiting the channel number to 31 in the driver. Now since we can list all 32 channels, update the interrupts property as well to list all 32 interrupts.

---
 arch/arm64/boot/dts/nvidia/tegra186.dtsi | 4 +++-
 arch/arm64/boot/dts/nvidia/tegra194.dtsi | 4 +++-
 arch/arm64/boot/dts/nvidia/tegra234.dtsi | 4 +++-
 3 files changed, 9 insertions(+), 3 deletions(-)
arch/arm64/boot/dts/nvidia/tegra186.dtsi +0 -0
index 6602fe421ee8..db479064ff72 100644

arch/arm64/boot/dts/nvidia/tegra186.dtsi line 82

Quoted 6 lines
arch/arm64/boot/dts/nvidia/tegra186.dtsi +2 -1
@@ -78,7 +78,8 @@
 		reg = <0x0 0x2600000 0x0 0x210000>;
 		resets = <&bpmp TEGRA186_RESET_GPCDMA>;
 		reset-names = "gpcdma";
-		interrupts = <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,
+		interrupts = <GIC_SPI 75 IRQ_TYPE_LEVEL_HIGH>,
+			     <GIC_SPI 76 IRQ_TYPE_LEVEL_HIGH>,

Adding an interrupt onto the beginning? Also an ABI issue.

If you want to break the ABI, you have to explain that you are and why it is okay on that platform.

Rob

author
#5 lore raw

PATCH RESEND v2 3/3 dmaengine: tegra: Add support for dma-channel-mask

In reply to Akhil R

Add support for dma-channel-mask so that only the specified channels are used. This helps to reserve some channels for the firmware.

This was initially achieved by limiting the channel number to 31 in the driver and adjusting the register address to skip channel0 which was reserved for a firmware. Now, with this change, the driver can align more to the actual hardware which has 32 channels.

---
 drivers/dma/tegra186-gpc-dma.c | 37 +++++++++++++++++++++++++++-------
 1 file changed, 30 insertions(+), 7 deletions(-)
drivers/dma/tegra186-gpc-dma.c +30 -7
index fa9bda4a2bc6..1d1180db6d4e 100644
@@ -161,7 +161,10 @@
 #define TEGRA_GPCDMA_BURST_COMPLETION_TIMEOUT	5000 /* 5 msec */
 
 /* Channel base address offset from GPCDMA base address */
-#define TEGRA_GPCDMA_CHANNEL_BASE_ADD_OFFSET	0x20000
+#define TEGRA_GPCDMA_CHANNEL_BASE_ADDR_OFFSET	0x10000
+
+/* Default channel mask reserving channel0 */
+#define TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK	0xfffffffe
 
 struct tegra_dma;
 struct tegra_dma_channel;
@@ -246,6 +249,7 @@ struct tegra_dma {
 	const struct tegra_dma_chip_data *chip_data;
 	unsigned long sid_m2d_reserved;
 	unsigned long sid_d2m_reserved;
+	u32 chan_mask;
 	void __iomem *base_addr;
 	struct device *dev;
 	struct dma_device dma_dev;
@@ -1288,7 +1292,7 @@ static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec,
 }
 
 static const struct tegra_dma_chip_data tegra186_dma_chip_data = {
-	.nr_channels = 31,
+	.nr_channels = 32,
 	.channel_reg_size = SZ_64K,
 	.max_dma_count = SZ_1G,
 	.hw_support_pause = false,
@@ -1296,7 +1300,7 @@ static const struct tegra_dma_chip_data tegra186_dma_chip_data = {
 };
 
 static const struct tegra_dma_chip_data tegra194_dma_chip_data = {
-	.nr_channels = 31,
+	.nr_channels = 32,
 	.channel_reg_size = SZ_64K,
 	.max_dma_count = SZ_1G,
 	.hw_support_pause = true,
@@ -1304,7 +1308,7 @@ static const struct tegra_dma_chip_data tegra194_dma_chip_data = {
 };
 
 static const struct tegra_dma_chip_data tegra234_dma_chip_data = {
-	.nr_channels = 31,
+	.nr_channels = 32,
 	.channel_reg_size = SZ_64K,
 	.max_dma_count = SZ_1G,
 	.hw_support_pause = true,
@@ -1380,15 +1384,28 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	}
 	stream_id = iommu_spec->ids[0] & 0xffff;
 
+	ret = device_property_read_u32(&pdev->dev, "dma-channel-mask",
+				       &tdma->chan_mask);
+	if (ret) {
+		dev_warn(&pdev->dev,
+			 "Missing dma-channel-mask property, using default channel mask %#x\n",
+			 TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK);
+		tdma->chan_mask = TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK;
+	}
+
 	INIT_LIST_HEAD(&tdma->dma_dev.channels);
 	for (i = 0; i < cdata->nr_channels; i++) {
 		struct tegra_dma_channel *tdc = &tdma->channels[i];
 
+		/* Check for channel mask */
+		if (!(tdma->chan_mask & BIT(i)))
+			continue;
+
 		tdc->irq = platform_get_irq(pdev, i);
 		if (tdc->irq < 0)
 			return tdc->irq;
 
-		tdc->chan_base_offset = TEGRA_GPCDMA_CHANNEL_BASE_ADD_OFFSET +
+		tdc->chan_base_offset = TEGRA_GPCDMA_CHANNEL_BASE_ADDR_OFFSET +
 					i * cdata->channel_reg_size;
 		snprintf(tdc->name, sizeof(tdc->name), "gpcdma.%d", i);
 		tdc->tdma = tdma;
@@ -1449,8 +1466,8 @@ static int tegra_dma_probe(struct platform_device *pdev)
 		return ret;
 	}
 
-	dev_info(&pdev->dev, "GPC DMA driver register %d channels\n",
-		 cdata->nr_channels);
+	dev_info(&pdev->dev, "GPC DMA driver register %lu channels\n",
+		 hweight_long(tdma->chan_mask));
 
 	return 0;
 }
@@ -1473,6 +1490,9 @@ static int __maybe_unused tegra_dma_pm_suspend(struct device *dev)
 	for (i = 0; i < tdma->chip_data->nr_channels; i++) {
 		struct tegra_dma_channel *tdc = &tdma->channels[i];
 
+		if (!(tdma->chan_mask & BIT(i)))
+			continue;
+
 		if (tdc->dma_desc) {
 			dev_err(tdma->dev, "channel %u busy\n", i);
 			return -EBUSY;
@@ -1492,6 +1512,9 @@ static int __maybe_unused tegra_dma_pm_resume(struct device *dev)
 	for (i = 0; i < tdma->chip_data->nr_channels; i++) {
 		struct tegra_dma_channel *tdc = &tdma->channels[i];
 
+		if (!(tdma->chan_mask & BIT(i)))
+			continue;
+
 		tegra_dma_program_sid(tdc, tdc->stream_id);
 	}
 
Signature · 1 line
2.17.1
maintainer
#6 lore raw

Re: [PATCH RESEND v2 3/3] dmaengine: tegra: Add support for dma-channel-mask

Quoted 39 lines · Akhil R

Add support for dma-channel-mask so that only the specified channels are used. This helps to reserve some channels for the firmware.

This was initially achieved by limiting the channel number to 31 in the driver and adjusting the register address to skip channel0 which was reserved for a firmware. Now, with this change, the driver can align more to the actual hardware which has 32 channels.

---
 drivers/dma/tegra186-gpc-dma.c | 37 +++++++++++++++++++++++++++-------
 1 file changed, 30 insertions(+), 7 deletions(-)
drivers/dma/tegra186-gpc-dma.c +5 -1
index fa9bda4a2bc6..1d1180db6d4e 100644
@@ -161,7 +161,10 @@
 #define TEGRA_GPCDMA_BURST_COMPLETION_TIMEOUT	5000 /* 5 msec */
 
 /* Channel base address offset from GPCDMA base address */
-#define TEGRA_GPCDMA_CHANNEL_BASE_ADD_OFFSET	0x20000
+#define TEGRA_GPCDMA_CHANNEL_BASE_ADDR_OFFSET	0x10000
+
+/* Default channel mask reserving channel0 */
+#define TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK	0xfffffffe
 
 struct tegra_dma;
 struct tegra_dma_channel;
@@ -246,6 +249,7 @@ struct tegra_dma {
 	const struct tegra_dma_chip_data *chip_data;
 	unsigned long sid_m2d_reserved;
 	unsigned long sid_d2m_reserved;
+	u32 chan_mask;
 	void __iomem *base_addr;
 	struct device *dev;
 	struct dma_device dma_dev;

drivers/dma/tegra186-gpc-dma.c static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec, · line 1295

Quoted 5 lines
drivers/dma/tegra186-gpc-dma.c +1 -1
@@ -1288,7 +1292,7 @@ static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec,
 }
 
 static const struct tegra_dma_chip_data tegra186_dma_chip_data = {
-	.nr_channels = 31,
+	.nr_channels = 32,

This is an ABI break. A new kernel with an old DTB will use 32 channels instead of 31. You should leave this and use the dma-channel-mask to enable all 32 channels.

Rob

author
#7 lore raw

RE: [PATCH RESEND v2 3/3] dmaengine: tegra: Add support for dma-channel-mask

Quoted 45 lines · Akhil R

drivers/dma/tegra186-gpc-

Quoted 15 lines

Add support for dma-channel-mask so that only the specified channels are used. This helps to reserve some channels for the firmware.

This was initially achieved by limiting the channel number to 31 in the driver and adjusting the register address to skip channel0 which was reserved for a firmware. Now, with this change, the driver can align more to the actual hardware which has 32 channels.

---
 drivers/dma/tegra186-gpc-dma.c | 37 +++++++++++++++++++++++++++-------
 1 file changed, 30 insertions(+), 7 deletions(-)
drivers/dma/tegra186-gpc- renamed from drivers/dma/tegra186-gpc-dma.c +0 -0

dma.c

drivers/dma/tegra186-gpc-dma.c line 161

Quoted 5 lines

index fa9bda4a2bc6..1d1180db6d4e 100644

drivers/dma/tegra186-gpc-dma.c +0 -0
@@ -161,7 +161,10 @@
 #define TEGRA_GPCDMA_BURST_COMPLETION_TIMEOUT        5000 /* 5

msec */

drivers/dma/tegra186-gpc-dma.c static struct dma_chan *tegra_dma_of_xlate(struct

Quoted 19 lines
drivers/dma/tegra186-gpc-dma.c +5 -1
 /* Channel base address offset from GPCDMA base address */
-#define TEGRA_GPCDMA_CHANNEL_BASE_ADD_OFFSET 0x20000
+#define TEGRA_GPCDMA_CHANNEL_BASE_ADDR_OFFSET        0x10000
+
+/* Default channel mask reserving channel0 */
+#define TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK    0xfffffffe
 
 struct tegra_dma;
 struct tegra_dma_channel;
@@ -246,6 +249,7 @@ struct tegra_dma {
      const struct tegra_dma_chip_data *chip_data;
      unsigned long sid_m2d_reserved;
      unsigned long sid_d2m_reserved;
+     u32 chan_mask;
      void __iomem *base_addr;
      struct device *dev;
      struct dma_device dma_dev;
@@ -1288,7 +1292,7 @@ static struct dma_chan *tegra_dma_of_xlate(struct

of_phandle_args *dma_spec,

Quoted 2 lines
drivers/dma/tegra186-gpc-dma.c +0 -0
 }
 
Quoted 8 lines

drivers/dma/tegra186-gpc-dma.c static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec, · line 1295

Quoted 3 lines
drivers/dma/tegra186-gpc-dma.c +1 -1
 static const struct tegra_dma_chip_data tegra186_dma_chip_data = {
-     .nr_channels = 31,
+     .nr_channels = 32,

This is an ABI break. A new kernel with an old DTB will use 32 channels instead of 31. You should leave this and use the dma-channel-mask to enable all 32 channels.

Hi Rob,

If using an old DTB, tdma->chan_mask will be default to 0xfffffffe since it would not have the dma-channel-mask property. The driver would still use 31 channels even if it uses an old DTB. Shouldn't it prevent the ABI break?

Regards,
Akhil

maintainer
#8 lore raw

Re: [PATCH RESEND v2 3/3] dmaengine: tegra: Add support for dma-channel-mask

Quoted 53 lines · Akhil R
Quoted 45 lines · Akhil R

drivers/dma/tegra186-gpc-

Quoted 15 lines

Add support for dma-channel-mask so that only the specified channels are used. This helps to reserve some channels for the firmware.

This was initially achieved by limiting the channel number to 31 in the driver and adjusting the register address to skip channel0 which was reserved for a firmware. Now, with this change, the driver can align more to the actual hardware which has 32 channels.

---
  drivers/dma/tegra186-gpc-dma.c | 37 +++++++++++++++++++++++++++-------
  1 file changed, 30 insertions(+), 7 deletions(-)
drivers/dma/tegra186-gpc- renamed from drivers/dma/tegra186-gpc-dma.c +0 -0

dma.c

drivers/dma/tegra186-gpc-dma.c line 161

Quoted 5 lines

index fa9bda4a2bc6..1d1180db6d4e 100644

drivers/dma/tegra186-gpc-dma.c +0 -0
@@ -161,7 +161,10 @@
  #define TEGRA_GPCDMA_BURST_COMPLETION_TIMEOUT        5000 /* 5

msec */

drivers/dma/tegra186-gpc-dma.c static struct dma_chan *tegra_dma_of_xlate(struct

Quoted 19 lines
drivers/dma/tegra186-gpc-dma.c +5 -1
  /* Channel base address offset from GPCDMA base address */
-#define TEGRA_GPCDMA_CHANNEL_BASE_ADD_OFFSET 0x20000
+#define TEGRA_GPCDMA_CHANNEL_BASE_ADDR_OFFSET        0x10000
+
+/* Default channel mask reserving channel0 */
+#define TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK    0xfffffffe
 
  struct tegra_dma;
  struct tegra_dma_channel;
@@ -246,6 +249,7 @@ struct tegra_dma {
       const struct tegra_dma_chip_data *chip_data;
       unsigned long sid_m2d_reserved;
       unsigned long sid_d2m_reserved;
+     u32 chan_mask;
       void __iomem *base_addr;
       struct device *dev;
       struct dma_device dma_dev;
@@ -1288,7 +1292,7 @@ static struct dma_chan *tegra_dma_of_xlate(struct

of_phandle_args *dma_spec,

Quoted 2 lines
drivers/dma/tegra186-gpc-dma.c +0 -0
  }
 
Quoted 8 lines

drivers/dma/tegra186-gpc-dma.c static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec, · line 1295

Quoted 3 lines
drivers/dma/tegra186-gpc-dma.c +1 -1
  static const struct tegra_dma_chip_data tegra186_dma_chip_data = {
-     .nr_channels = 31,
+     .nr_channels = 32,

This is an ABI break. A new kernel with an old DTB will use 32 channels instead of 31. You should leave this and use the dma-channel-mask to enable all 32 channels.

Quoted 6 lines

Hi Rob,

If using an old DTB, tdma->chan_mask will be default to 0xfffffffe since it would not have the dma-channel-mask property. The driver would still use 31 channels even if it uses an old DTB. Shouldn't it prevent the ABI break?

Unfortunately no. Yes for an old DTB without the dma-channel-mask property, we set the channel mask to 0xfffffffe, but this is not correct because it only has 31 interrupts/channels and not 32. So I think we will need to use of_irq_count() here.

Jon

Signature · 1 line
nvpublic
author
#9 lore raw

RE: [PATCH RESEND v2 3/3] dmaengine: tegra: Add support for dma-channel-mask

Quoted 63 lines · Akhil R
Quoted 9 lines
Quoted 9 lines · Akhil R
Quoted 8 lines

Add support for dma-channel-mask so that only the specified channels are used. This helps to reserve some channels for the firmware.

This was initially achieved by limiting the channel number to 31 in the driver and adjusting the register address to skip channel0 which was reserved for a firmware. Now, with this change, the driver can align more to the actual hardware which has 32 channels.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
---
  drivers/dma/tegra186-gpc-dma.c | 37 +++++++++++++++++++++++++++----

---

Quoted 18 lines
Quoted 18 lines

drivers/dma/tegra186-gpc-

Quoted 3 lines
  1 file changed, 30 insertions(+), 7 deletions(-)
drivers/dma/tegra186-gpc- renamed from drivers/dma/tegra186-gpc-dma.c +0 -0

dma.c

drivers/dma/tegra186-gpc-dma.c line 161

Quoted 5 lines

index fa9bda4a2bc6..1d1180db6d4e 100644

drivers/dma/tegra186-gpc-dma.c +0 -0
@@ -161,7 +161,10 @@
  #define TEGRA_GPCDMA_BURST_COMPLETION_TIMEOUT        5000 /* 5

msec */

Quoted 8 lines
drivers/dma/tegra186-gpc-dma.c +4 -1
  /* Channel base address offset from GPCDMA base address */
-#define TEGRA_GPCDMA_CHANNEL_BASE_ADD_OFFSET 0x20000
+#define TEGRA_GPCDMA_CHANNEL_BASE_ADDR_OFFSET        0x10000
+
+/* Default channel mask reserving channel0 */
+#define TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK    0xfffffffe
 
Quoted 11 lines
Quoted 11 lines

drivers/dma/tegra186-gpc-dma.c static struct dma_chan

Quoted 11 lines
drivers/dma/tegra186-gpc-dma.c +1 -0
  struct tegra_dma;
  struct tegra_dma_channel;
@@ -246,6 +249,7 @@ struct tegra_dma {
       const struct tegra_dma_chip_data *chip_data;
       unsigned long sid_m2d_reserved;
       unsigned long sid_d2m_reserved;
+     u32 chan_mask;
       void __iomem *base_addr;
       struct device *dev;
       struct dma_device dma_dev;
@@ -1288,7 +1292,7 @@ static struct dma_chan

*tegra_dma_of_xlate(struct

Quoted 11 lines
Quoted 11 lines

of_phandle_args *dma_spec,

drivers/dma/tegra186-gpc-dma.c static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec, · line 1295

Quoted 5 lines
drivers/dma/tegra186-gpc-dma.c +1 -1
  }
 
  static const struct tegra_dma_chip_data tegra186_dma_chip_data = {
-     .nr_channels = 31,
+     .nr_channels = 32,

This is an ABI break. A new kernel with an old DTB will use 32 channels instead of 31. You should leave this and use the dma-channel-mask to enable all 32 channels.

Quoted 6 lines

Hi Rob,

If using an old DTB, tdma->chan_mask will be default to 0xfffffffe since it would not have the dma-channel-mask property. The driver would still use 31 channels even if it uses an old DTB. Shouldn't it prevent the ABI break?

Quoted 5 lines

Unfortunately no. Yes for an old DTB without the dma-channel-mask property, we set the channel mask to 0xfffffffe, but this is not correct because it only has 31 interrupts/channels and not 32. So I think we will need to use of_irq_count() here.

Shall I put it in a way that only the used interrupts are mentioned in the DT? With this I can revert the interrupt change in the DT and would not break the ABI as well.

The code would look something like this.

int chan_count = 0;

if (of_irq_count(pdev->dev.of_node) != hweight_long(tdma->chan_mask)) {
        dev_err(&pdev->dev, "Interrupt count doesn't match with channels\n");
        return -EINVAL;
}
for (i = 0; i < cdata->nr_channels; i++) {
        struct tegra_dma_channel *tdc = &tdma->channels[i];
    
        /* Check for channel mask */
        if (!(tdma->chan_mask & BIT(i)))
            continue;

        tdc->irq = platform_get_irq(pdev, chan_count);
        chan_count++;
        if (tdc->irq < 0)
            return tdc->irq;
    ...
    ...
}

Regards,
Akhil

#10 lore raw

Re: [PATCH RESEND v2 3/3] dmaengine: tegra: Add support for dma-channel-mask

Quoted 82 lines · Akhil R
Quoted 63 lines · Akhil R
Quoted 9 lines
Quoted 9 lines · Akhil R
Quoted 8 lines

Add support for dma-channel-mask so that only the specified channels are used. This helps to reserve some channels for the firmware.

This was initially achieved by limiting the channel number to 31 in the driver and adjusting the register address to skip channel0 which was reserved for a firmware. Now, with this change, the driver can align more to the actual hardware which has 32 channels.

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
---
  drivers/dma/tegra186-gpc-dma.c | 37 +++++++++++++++++++++++++++----

---

Quoted 18 lines
Quoted 18 lines

drivers/dma/tegra186-gpc-

Quoted 3 lines
  1 file changed, 30 insertions(+), 7 deletions(-)
drivers/dma/tegra186-gpc- renamed from drivers/dma/tegra186-gpc-dma.c +0 -0

dma.c

drivers/dma/tegra186-gpc-dma.c line 161

Quoted 5 lines

index fa9bda4a2bc6..1d1180db6d4e 100644

drivers/dma/tegra186-gpc-dma.c +0 -0
@@ -161,7 +161,10 @@
  #define TEGRA_GPCDMA_BURST_COMPLETION_TIMEOUT        5000 /* 5

msec */

Quoted 8 lines
drivers/dma/tegra186-gpc-dma.c +4 -1
  /* Channel base address offset from GPCDMA base address */
-#define TEGRA_GPCDMA_CHANNEL_BASE_ADD_OFFSET 0x20000
+#define TEGRA_GPCDMA_CHANNEL_BASE_ADDR_OFFSET        0x10000
+
+/* Default channel mask reserving channel0 */
+#define TEGRA_GPCDMA_DEFAULT_CHANNEL_MASK    0xfffffffe
 
Quoted 11 lines
Quoted 11 lines

drivers/dma/tegra186-gpc-dma.c static struct dma_chan

Quoted 11 lines
drivers/dma/tegra186-gpc-dma.c +1 -0
  struct tegra_dma;
  struct tegra_dma_channel;
@@ -246,6 +249,7 @@ struct tegra_dma {
       const struct tegra_dma_chip_data *chip_data;
       unsigned long sid_m2d_reserved;
       unsigned long sid_d2m_reserved;
+     u32 chan_mask;
       void __iomem *base_addr;
       struct device *dev;
       struct dma_device dma_dev;
@@ -1288,7 +1292,7 @@ static struct dma_chan

*tegra_dma_of_xlate(struct

Quoted 11 lines
Quoted 11 lines

of_phandle_args *dma_spec,

drivers/dma/tegra186-gpc-dma.c static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec, · line 1295

Quoted 5 lines
drivers/dma/tegra186-gpc-dma.c +1 -1
  }
 
  static const struct tegra_dma_chip_data tegra186_dma_chip_data = {
-     .nr_channels = 31,
+     .nr_channels = 32,

This is an ABI break. A new kernel with an old DTB will use 32 channels instead of 31. You should leave this and use the dma-channel-mask to enable all 32 channels.

Quoted 6 lines

Hi Rob,

If using an old DTB, tdma->chan_mask will be default to 0xfffffffe since it would not have the dma-channel-mask property. The driver would still use 31 channels even if it uses an old DTB. Shouldn't it prevent the ABI break?

Quoted 5 lines

Unfortunately no. Yes for an old DTB without the dma-channel-mask property, we set the channel mask to 0xfffffffe, but this is not correct because it only has 31 interrupts/channels and not 32. So I think we will need to use of_irq_count() here.

Shall I put it in a way that only the used interrupts are mentioned in the DT? With this I can revert the interrupt change in the DT and would not break the ABI as well.

The code would look something like this.

int chan_count = 0;

if (of_irq_count(pdev->dev.of_node) != hweight_long(tdma->chan_mask)) {
        dev_err(&pdev->dev, "Interrupt count doesn't match with channels\n");
        return -EINVAL;
}
Quoted 14 lines
for (i = 0; i < cdata->nr_channels; i++) {
        struct tegra_dma_channel *tdc = &tdma->channels[i];
    
        /* Check for channel mask */
        if (!(tdma->chan_mask & BIT(i)))
            continue;

        tdc->irq = platform_get_irq(pdev, chan_count);
        chan_count++;
        if (tdc->irq < 0)
            return tdc->irq;
    ...
    ...
}

This is all getting quite complicated for what is essentially a bug fix. The root of this problem is that the original bindings were simply wrong and didn't accurately represent the hardware.

The GPC DMA controller has 32 channels and each channel has one DMA interrupt. So the right way to describe this in DT is by listing all of the interrupts. If the firmware then ends up using one of those channels the right way is not to make it seem like the controller supports only 31 channels, but rather to mark the used channels as reserved.

So I think the bottom line here is that the original binding has a bug that we're trying to address. Any workaround for this is problematic, and I think breaking backwards-compatibility is the cleanest solution here.

Now, GPC DMA was introduced on Tegra186 and the DT bindings were added in 5.19. Any products released with this IP were released with kernels prior to 5.19 and bindings that were never in-tree. Any of those products that are supported upstream we know have replaceable DTB images (i.e. by default they are flashed at the same time as the kernel image), so breaking DT ABI should be okay.

In fact, there are slight incompatibilities between product and upstream DT bindings (we're in the process of removing those as part of the upstreaming process), so being able to reflash the DTB is a prerequisite for supporting upstream Linux on these devices.

I think going forward we should be extra careful about these kinds of problems, so that we don't have to keep dealing with fallout like this.

Rob, given the above, do you think it's okay to proceed with the ABI break?

Thierry