[PATCH v1 0/5] NVIDIA Tegra AHB DMA controller driver

[PATCH v1 0/5] NVIDIA Tegra AHB DMA controller driver

[Dmitry Osipenko]

NVIDIA Tegra20/30 SoC's have AHB DMA controller. It has 4 DMA channels,
supports AHB <-> Memory and Memory <-> Memory transfers, slave / master
modes. This driver is primarily supposed to be used by gpu/host1x in a
master mode, performing 3D HW context stores.

Dmitry Osipenko (5):
  clk: tegra: Add AHB DMA clock entry
  clk: tegra: Bump SCLK clock rate to 216MHz on Tegra20
  dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller
  dmaengine: Add driver for NVIDIA Tegra AHB DMA controller
  ARM: dts: tegra: Add AHB DMA controller nodes

 .../bindings/dma/nvidia,tegra20-ahbdma.txt         |  23 +
 arch/arm/boot/dts/tegra20.dtsi                     |   9 +
 arch/arm/boot/dts/tegra30.dtsi                     |   9 +
 drivers/clk/tegra/clk-id.h                         |   1 +
 drivers/clk/tegra/clk-tegra-periph.c               |   1 +
 drivers/clk/tegra/clk-tegra20.c                    |   8 +-
 drivers/clk/tegra/clk-tegra30.c                    |   2 +
 drivers/dma/Kconfig                                |   9 +
 drivers/dma/Makefile                               |   1 +
 drivers/dma/tegra20-ahb-dma.c                      | 679 +++++++++++++++++++++
 10 files changed, 741 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
 create mode 100644 drivers/dma/tegra20-ahb-dma.c

-- 
2.14.1

[PATCH v1 0/5] NVIDIA Tegra AHB DMA controller driver

[Dmitry Osipenko]

NVIDIA Tegra20/30 SoC's have AHB DMA controller. It has 4 DMA channels,
supports AHB <-> Memory and Memory <-> Memory transfers, slave / master
modes. This driver is primarily supposed to be used by gpu/host1x in a
master mode, performing 3D HW context stores.

Dmitry Osipenko (5):
  clk: tegra: Add AHB DMA clock entry
  clk: tegra: Bump SCLK clock rate to 216MHz on Tegra20
  dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller
  dmaengine: Add driver for NVIDIA Tegra AHB DMA controller
  ARM: dts: tegra: Add AHB DMA controller nodes

 .../bindings/dma/nvidia,tegra20-ahbdma.txt         |  23 +
 arch/arm/boot/dts/tegra20.dtsi                     |   9 +
 arch/arm/boot/dts/tegra30.dtsi                     |   9 +
 drivers/clk/tegra/clk-id.h                         |   1 +
 drivers/clk/tegra/clk-tegra-periph.c               |   1 +
 drivers/clk/tegra/clk-tegra20.c                    |   8 +-
 drivers/clk/tegra/clk-tegra30.c                    |   2 +
 drivers/dma/Kconfig                                |   9 +
 drivers/dma/Makefile                               |   1 +
 drivers/dma/tegra20-ahb-dma.c                      | 679 +++++++++++++++++++++
 10 files changed, 741 insertions(+), 1 deletion(-)
 create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
 create mode 100644 drivers/dma/tegra20-ahb-dma.c

-- 
2.14.1

[PATCH v1 1/5] clk: tegra: Add AHB DMA clock entry

[Dmitry Osipenko]

AHB DMA presents on Tegra20/30. Add missing entries, so that driver
for AHB DMA could be implemented.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/clk/tegra/clk-id.h           | 1 +
 drivers/clk/tegra/clk-tegra-periph.c | 1 +
 drivers/clk/tegra/clk-tegra20.c      | 6 ++++++
 drivers/clk/tegra/clk-tegra30.c      | 2 ++
 4 files changed, 10 insertions(+)

diff --git a/drivers/clk/tegra/clk-id.h b/drivers/clk/tegra/clk-id.h
index 689f344377a7..c1661b47bbda 100644
--- a/drivers/clk/tegra/clk-id.h
+++ b/drivers/clk/tegra/clk-id.h
@@ -12,6 +12,7 @@ enum clk_id {
 	tegra_clk_amx,
 	tegra_clk_amx1,
 	tegra_clk_apb2ape,
+	tegra_clk_ahbdma,
 	tegra_clk_apbdma,
 	tegra_clk_apbif,
 	tegra_clk_ape,
diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c
index 848255cc0209..95a3d8c95f06 100644
--- a/drivers/clk/tegra/clk-tegra-periph.c
+++ b/drivers/clk/tegra/clk-tegra-periph.c
@@ -823,6 +823,7 @@ static struct tegra_periph_init_data gate_clks[] = {
 	GATE("timer", "clk_m", 5, 0, tegra_clk_timer, CLK_IS_CRITICAL),
 	GATE("isp", "clk_m", 23, 0, tegra_clk_isp, 0),
 	GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0),
+	GATE("ahbdma", "clk_m", 33, 0, tegra_clk_ahbdma, 0),
 	GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0),
 	GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0),
 	GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0),
diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
index 837e5cbd60e9..e76c0d292ca7 100644
--- a/drivers/clk/tegra/clk-tegra20.c
+++ b/drivers/clk/tegra/clk-tegra20.c
@@ -449,6 +449,7 @@ static struct tegra_devclk devclks[] __initdata = {
 	{ .con_id = "audio", .dt_id = TEGRA20_CLK_AUDIO },
 	{ .con_id = "audio_2x", .dt_id = TEGRA20_CLK_AUDIO_2X },
 	{ .dev_id = "tegra20-ac97", .dt_id = TEGRA20_CLK_AC97 },
+	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA20_CLK_AHBDMA },
 	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA20_CLK_APBDMA },
 	{ .dev_id = "rtc-tegra", .dt_id = TEGRA20_CLK_RTC },
 	{ .dev_id = "timer", .dt_id = TEGRA20_CLK_TIMER },
@@ -806,6 +807,11 @@ static void __init tegra20_periph_clk_init(void)
 				    clk_base, 0, 3, periph_clk_enb_refcnt);
 	clks[TEGRA20_CLK_AC97] = clk;
 
+	/* ahbdma */
+	clk = tegra_clk_register_periph_gate("ahbdma", "hclk", 0, clk_base,
+				    0, 33, periph_clk_enb_refcnt);
+	clks[TEGRA20_CLK_AHBDMA] = clk;
+
 	/* apbdma */
 	clk = tegra_clk_register_periph_gate("apbdma", "pclk", 0, clk_base,
 				    0, 34, periph_clk_enb_refcnt);
diff --git a/drivers/clk/tegra/clk-tegra30.c b/drivers/clk/tegra/clk-tegra30.c
index a2d163f759b4..e99701557f29 100644
--- a/drivers/clk/tegra/clk-tegra30.c
+++ b/drivers/clk/tegra/clk-tegra30.c
@@ -612,6 +612,7 @@ static struct tegra_devclk devclks[] __initdata = {
 	{ .con_id = "fuse_burn", .dev_id = "fuse-tegra", .dt_id = TEGRA30_CLK_FUSE_BURN },
 	{ .con_id = "apbif", .dev_id = "tegra30-ahub", .dt_id = TEGRA30_CLK_APBIF },
 	{ .con_id = "hda2hdmi", .dev_id = "tegra30-hda", .dt_id = TEGRA30_CLK_HDA2HDMI },
+	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA30_CLK_AHBDMA },
 	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA30_CLK_APBDMA },
 	{ .dev_id = "rtc-tegra", .dt_id = TEGRA30_CLK_RTC },
 	{ .dev_id = "timer", .dt_id = TEGRA30_CLK_TIMER },
@@ -788,6 +789,7 @@ static struct tegra_clk tegra30_clks[tegra_clk_max] __initdata = {
 	[tegra_clk_extern3] = { .dt_id = TEGRA30_CLK_EXTERN3, .present = true },
 	[tegra_clk_disp1] = { .dt_id = TEGRA30_CLK_DISP1, .present = true },
 	[tegra_clk_disp2] = { .dt_id = TEGRA30_CLK_DISP2, .present = true },
+	[tegra_clk_ahbdma] = { .dt_id = TEGRA30_CLK_AHBDMA, .present = true },
 	[tegra_clk_apbdma] = { .dt_id = TEGRA30_CLK_APBDMA, .present = true },
 	[tegra_clk_rtc] = { .dt_id = TEGRA30_CLK_RTC, .present = true },
 	[tegra_clk_timer] = { .dt_id = TEGRA30_CLK_TIMER, .present = true },
-- 
2.14.1

[PATCH v1 2/5] clk: tegra: Bump SCLK clock rate to 216MHz on Tegra20

[Dmitry Osipenko]

AHB DMA is a running on 1/2 of SCLK rate, APB on 1/4. Increasing SCLK rate
results in an increased DMA transfer rate.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/clk/tegra/clk-tegra20.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
index e76c0d292ca7..c511716093e2 100644
--- a/drivers/clk/tegra/clk-tegra20.c
+++ b/drivers/clk/tegra/clk-tegra20.c
@@ -1031,7 +1031,7 @@ static struct tegra_clk_init_table init_table[] __initdata = {
 	{ TEGRA20_CLK_PLL_P_OUT3, TEGRA20_CLK_CLK_MAX, 72000000, 1 },
 	{ TEGRA20_CLK_PLL_P_OUT4, TEGRA20_CLK_CLK_MAX, 24000000, 1 },
 	{ TEGRA20_CLK_PLL_C, TEGRA20_CLK_CLK_MAX, 600000000, 1 },
-	{ TEGRA20_CLK_PLL_C_OUT1, TEGRA20_CLK_CLK_MAX, 120000000, 1 },
+	{ TEGRA20_CLK_PLL_C_OUT1, TEGRA20_CLK_CLK_MAX, 216000000, 1 },
 	{ TEGRA20_CLK_SCLK, TEGRA20_CLK_PLL_C_OUT1, 0, 1 },
 	{ TEGRA20_CLK_HCLK, TEGRA20_CLK_CLK_MAX, 0, 1 },
 	{ TEGRA20_CLK_PCLK, TEGRA20_CLK_CLK_MAX, 60000000, 1 },
-- 
2.14.1

[PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
on Tegra20/30 SoC's.

Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
---
 .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
 1 file changed, 23 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt

diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
new file mode 100644
index 000000000000..2af9aa76ae11
--- /dev/null
+++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
@@ -0,0 +1,23 @@
+* NVIDIA Tegra AHB DMA controller
+
+Required properties:
+- compatible:	Must be "nvidia,tegra20-ahbdma"
+- reg:		Should contain registers base address and length.
+- interrupts:	Should contain one entry, DMA controller interrupt.
+- clocks:	Should contain one entry, DMA controller clock.
+- resets :	Should contain one entry, DMA controller reset.
+- #dma-cells:	Should be <1>. The cell represents DMA request select value
+		for the peripheral. For more details consult the Tegra TRM's
+		documentation, in particular AHB DMA channel control register
+		REQ_SEL field.
+
+Example:
+
+ahbdma: ahbdma@60008000  {
+	compatible = "nvidia,tegra20-ahbdma";
+	reg = <0x60008000 0x2000>;
+	interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
+	clocks = <&tegra_car TEGRA20_CLK_AHBDMA>;
+	resets = <&tegra_car 33>;
+	#dma-cells = <1>;
+};
-- 
2.14.1

[PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
on Tegra20/30 SoC's.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
 1 file changed, 23 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt

diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
new file mode 100644
index 000000000000..2af9aa76ae11
--- /dev/null
+++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
@@ -0,0 +1,23 @@
+* NVIDIA Tegra AHB DMA controller
+
+Required properties:
+- compatible:	Must be "nvidia,tegra20-ahbdma"
+- reg:		Should contain registers base address and length.
+- interrupts:	Should contain one entry, DMA controller interrupt.
+- clocks:	Should contain one entry, DMA controller clock.
+- resets :	Should contain one entry, DMA controller reset.
+- #dma-cells:	Should be <1>. The cell represents DMA request select value
+		for the peripheral. For more details consult the Tegra TRM's
+		documentation, in particular AHB DMA channel control register
+		REQ_SEL field.
+
+Example:
+
+ahbdma: ahbdma@60008000  {
+	compatible = "nvidia,tegra20-ahbdma";
+	reg = <0x60008000 0x2000>;
+	interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
+	clocks = <&tegra_car TEGRA20_CLK_AHBDMA>;
+	resets = <&tegra_car 33>;
+	#dma-cells = <1>;
+};
-- 
2.14.1

[PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
doesn't yet implement transfers larger than 64K and scatter-gather
transfers that have NENT > 1, HW doesn't have native support for these
cases.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/Kconfig           |   9 +
 drivers/dma/Makefile          |   1 +
 drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 689 insertions(+)
 create mode 100644 drivers/dma/tegra20-ahb-dma.c

diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index fadc4d8783bd..937c110a715b 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -503,6 +503,15 @@ config TXX9_DMAC
 	  Support the TXx9 SoC internal DMA controller.  This can be
 	  integrated in chips such as the Toshiba TX4927/38/39.
 
+config TEGRA20_AHB_DMA
+	tristate "NVIDIA Tegra20 AHB DMA support"
+	depends on ARCH_TEGRA
+	select DMA_ENGINE
+	help
+	  Enable support for the NVIDIA Tegra20 AHB DMA controller driver.
+	  This DMA controller transfers data from memory to AHB peripherals
+	  or vice versa, it supports memory to memory data transfer as well.
+
 config TEGRA20_APB_DMA
 	bool "NVIDIA Tegra20 APB DMA support"
 	depends on ARCH_TEGRA
diff --git a/drivers/dma/Makefile b/drivers/dma/Makefile
index f08f8de1b567..82d5d36b0379 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -61,6 +61,7 @@ obj-$(CONFIG_STE_DMA40) += ste_dma40.o ste_dma40_ll.o
 obj-$(CONFIG_STM32_DMA) += stm32-dma.o
 obj-$(CONFIG_S3C24XX_DMAC) += s3c24xx-dma.o
 obj-$(CONFIG_TXX9_DMAC) += txx9dmac.o
+obj-$(CONFIG_TEGRA20_AHB_DMA) += tegra20-ahb-dma.o
 obj-$(CONFIG_TEGRA20_APB_DMA) += tegra20-apb-dma.o
 obj-$(CONFIG_TEGRA210_ADMA) += tegra210-adma.o
 obj-$(CONFIG_TIMB_DMA) += timb_dma.o
diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
new file mode 100644
index 000000000000..8316d64e35e1
--- /dev/null
+++ b/drivers/dma/tegra20-ahb-dma.c
@@ -0,0 +1,679 @@
+/*
+ * Copyright 2017 Dmitry Osipenko <digetx@gmail.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope it will be useful, but WITHOUT
+ * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
+ * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <linux/clk.h>
+#include <linux/delay.h>
+#include <linux/interrupt.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/of_device.h>
+#include <linux/of_dma.h>
+#include <linux/platform_device.h>
+#include <linux/reset.h>
+#include <linux/slab.h>
+#include <linux/spinlock.h>
+
+#include "dmaengine.h"
+
+#define TEGRA_AHBDMA_CMD			0x0
+#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
+
+#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
+#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
+
+#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
+
+#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
+#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
+#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
+#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
+#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
+#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
+#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
+#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
+#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
+
+#define TEGRA_AHBDMA_CHANNEL_STA		0x4
+#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
+
+#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
+
+#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
+#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
+#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
+#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
+#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
+#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
+
+#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
+
+#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
+
+#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
+						BIT(DMA_MEM_TO_DEV)
+
+struct tegra_ahbdma_tx_desc {
+	struct dma_async_tx_descriptor desc;
+	struct tasklet_struct tasklet;
+	struct list_head node;
+	enum dma_transfer_direction dir;
+	dma_addr_t mem_paddr;
+	unsigned long flags;
+	size_t size;
+	bool in_fly;
+	bool cyclic;
+};
+
+struct tegra_ahbdma_chan {
+	struct dma_chan dma_chan;
+	struct list_head active_list;
+	struct list_head pending_list;
+	struct completion idling;
+	void __iomem *regs;
+	spinlock_t lock;
+	unsigned int id;
+};
+
+struct tegra_ahbdma {
+	struct tegra_ahbdma_chan channels[4];
+	struct dma_device dma_dev;
+	struct reset_control *rst;
+	struct clk *clk;
+	void __iomem *regs;
+};
+
+static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
+{
+	return container_of(dev, struct tegra_ahbdma, dma_dev);
+}
+
+static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
+{
+	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
+}
+
+static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
+				struct dma_async_tx_descriptor *tx)
+{
+	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
+}
+
+static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
+				   struct tegra_ahbdma_tx_desc *tx)
+{
+	u32 csr;
+
+	writel_relaxed(tx->mem_paddr,
+		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
+
+	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
+
+	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
+	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
+	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
+	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
+	csr |= tx->size - sizeof(u32);
+
+	if (tx->dir == DMA_DEV_TO_MEM)
+		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
+
+	if (!tx->cyclic)
+		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
+
+	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
+
+	tx->in_fly = true;
+}
+
+static void tegra_ahbdma_tasklet(unsigned long data)
+{
+	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
+	struct dma_async_tx_descriptor *desc = &tx->desc;
+
+	dmaengine_desc_get_callback_invoke(desc, NULL);
+
+	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
+		kfree(tx);
+}
+
+static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
+				      struct tegra_ahbdma_tx_desc *tx)
+{
+	struct dma_async_tx_descriptor *desc = &tx->desc;
+	bool reuse = dmaengine_desc_test_reuse(desc);
+	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
+	bool completed = !tx->cyclic;
+
+	if (completed)
+		dma_cookie_complete(desc);
+
+	if (interrupt)
+		tasklet_schedule(&tx->tasklet);
+
+	if (completed) {
+		list_del(&tx->node);
+
+		if (reuse)
+			tx->in_fly = false;
+
+		if (!interrupt && !reuse)
+			kfree(tx);
+	}
+
+	return completed;
+}
+
+static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
+{
+	struct tegra_ahbdma_tx_desc *tx;
+
+	tx = list_first_entry_or_null(&chan->active_list,
+				      struct tegra_ahbdma_tx_desc,
+				      node);
+	if (tx)
+		tegra_ahbdma_submit_tx(chan, tx);
+
+	return !!tx;
+}
+
+static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
+{
+	struct tegra_ahbdma_tx_desc *tx;
+	unsigned long flags;
+	u32 status;
+
+	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
+	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
+		return;
+
+	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
+		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
+
+	spin_lock_irqsave(&chan->lock, flags);
+
+	if (!completion_done(&chan->idling)) {
+		tx = list_first_entry(&chan->active_list,
+				      struct tegra_ahbdma_tx_desc,
+				      node);
+
+		if (tegra_ahbdma_tx_completed(chan, tx) &&
+		    !tegra_ahbdma_next_tx_issued(chan))
+			complete_all(&chan->idling);
+	}
+
+	spin_unlock_irqrestore(&chan->lock, flags);
+}
+
+static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
+{
+	struct tegra_ahbdma *tdma = dev_id;
+	unsigned int i;
+
+	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
+		tegra_ahbdma_handle_channel(&tdma->channels[i]);
+
+	return IRQ_HANDLED;
+}
+
+static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
+{
+	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
+	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
+	dma_cookie_t cookie;
+
+	cookie = dma_cookie_assign(desc);
+
+	spin_lock_irq(&chan->lock);
+	list_add_tail(&tx->node, &chan->pending_list);
+	spin_unlock_irq(&chan->lock);
+
+	return cookie;
+}
+
+static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
+{
+	kfree(to_ahbdma_tx_desc(desc));
+
+	return 0;
+}
+
+static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
+					struct dma_chan *chan,
+					struct scatterlist *sgl,
+					unsigned int sg_len,
+					enum dma_transfer_direction dir,
+					unsigned long flags,
+					void *context)
+{
+	struct tegra_ahbdma_tx_desc *tx;
+
+	/* unimplemented */
+	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
+		return NULL;
+
+	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
+	if (!tx)
+		return NULL;
+
+	dma_async_tx_descriptor_init(&tx->desc, chan);
+
+	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
+	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
+	tx->mem_paddr		= sg_dma_address(sgl);
+	tx->size		= sg_dma_len(sgl);
+	tx->flags		= flags;
+	tx->dir			= dir;
+
+	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
+
+	return &tx->desc;
+}
+
+static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
+					struct dma_chan *chan,
+					dma_addr_t buf_addr,
+					size_t buf_len,
+					size_t period_len,
+					enum dma_transfer_direction dir,
+					unsigned long flags)
+{
+	struct tegra_ahbdma_tx_desc *tx;
+
+	/* unimplemented */
+	if (buf_len != period_len || buf_len > SZ_64K)
+		return NULL;
+
+	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
+	if (!tx)
+		return NULL;
+
+	dma_async_tx_descriptor_init(&tx->desc, chan);
+
+	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
+	tx->mem_paddr		= buf_addr;
+	tx->size		= buf_len;
+	tx->flags		= flags;
+	tx->cyclic		= true;
+	tx->dir			= dir;
+
+	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
+
+	return &tx->desc;
+}
+
+static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+	struct tegra_ahbdma_tx_desc *tx;
+	struct list_head *entry, *tmp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ahbdma_chan->lock, flags);
+
+	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
+		list_move_tail(entry, &ahbdma_chan->active_list);
+
+	if (completion_done(&ahbdma_chan->idling)) {
+		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
+					      struct tegra_ahbdma_tx_desc,
+					      node);
+		if (tx) {
+			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
+			reinit_completion(&ahbdma_chan->idling);
+		}
+	}
+
+	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
+}
+
+static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
+					      dma_cookie_t cookie,
+					      struct dma_tx_state *state)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+	struct tegra_ahbdma_tx_desc *tx;
+	enum dma_status cookie_status;
+	unsigned long flags;
+	size_t residual;
+	u32 status;
+
+	spin_lock_irqsave(&ahbdma_chan->lock, flags);
+
+	cookie_status = dma_cookie_status(chan, cookie, state);
+	if (cookie_status != DMA_COMPLETE) {
+		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
+			if (tx->desc.cookie == cookie)
+				goto found;
+		}
+	}
+
+	goto unlock;
+
+found:
+	if (tx->in_fly) {
+		status = readl_relaxed(
+			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
+		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
+
+		residual = status;
+	} else
+		residual = tx->size;
+
+	dma_set_residue(state, residual);
+
+unlock:
+	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
+
+	return cookie_status;
+}
+
+static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+	struct tegra_ahbdma_tx_desc *tx;
+	struct list_head *entry, *tmp;
+	u32 csr;
+
+	spin_lock_irq(&ahbdma_chan->lock);
+
+	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
+	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
+
+	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
+
+	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
+		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
+		list_del(entry);
+		kfree(tx);
+	}
+
+	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
+		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
+		list_del(entry);
+		kfree(tx);
+	}
+
+	complete_all(&ahbdma_chan->idling);
+
+	spin_unlock_irq(&ahbdma_chan->lock);
+
+	return 0;
+}
+
+static int tegra_ahbdma_config(struct dma_chan *chan,
+			       struct dma_slave_config *sconfig)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+	enum dma_transfer_direction dir = sconfig->direction;
+	u32 burst, ahb_seq, ahb_addr;
+
+	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
+	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
+		return -EINVAL;
+
+	if (dir == DMA_DEV_TO_MEM) {
+		burst    = sconfig->src_maxburst;
+		ahb_addr = sconfig->src_addr;
+	} else {
+		burst    = sconfig->dst_maxburst;
+		ahb_addr = sconfig->dst_addr;
+	}
+
+	switch (burst) {
+	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
+	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
+	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
+	default:
+		return -EINVAL;
+	}
+
+	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
+	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
+	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
+
+	writel_relaxed(ahb_seq,
+		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
+
+	writel_relaxed(ahb_addr,
+		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
+
+	return 0;
+}
+
+static void tegra_ahbdma_synchronize(struct dma_chan *chan)
+{
+	wait_for_completion(&to_ahbdma_chan(chan)->idling);
+}
+
+static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+	struct tegra_ahbdma_tx_desc *tx;
+	struct list_head *entry, *tmp;
+
+	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
+		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
+		list_del(entry);
+		kfree(tx);
+	}
+}
+
+static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
+				      unsigned int chan_id)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
+	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
+	struct dma_device *dma_dev = &tdma->dma_dev;
+
+	INIT_LIST_HEAD(&ahbdma_chan->active_list);
+	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
+	init_completion(&ahbdma_chan->idling);
+	spin_lock_init(&ahbdma_chan->lock);
+	complete(&ahbdma_chan->idling);
+
+	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
+	ahbdma_chan->id = chan_id;
+
+	dma_cookie_init(dma_chan);
+	dma_chan->device = dma_dev;
+
+	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
+}
+
+static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
+					      struct of_dma *ofdma)
+{
+	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
+	struct dma_chan *chan;
+	u32 csr;
+
+	chan = dma_get_any_slave_channel(&tdma->dma_dev);
+	if (!chan)
+		return NULL;
+
+	/* enable channels flow control */
+	if (dma_spec->args_count == 1) {
+		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
+		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;
+
+		writel_relaxed(csr,
+			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
+	}
+
+	return chan;
+}
+
+static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
+{
+	int err;
+
+	err = reset_control_assert(tdma->rst);
+	if (err) {
+		dev_err(dev, "Failed to assert reset: %d\n", err);
+		return err;
+	}
+
+	err = clk_prepare_enable(tdma->clk);
+	if (err) {
+		dev_err(dev, "Failed to enable clock: %d\n", err);
+		return err;
+	}
+
+	usleep_range(1000, 2000);
+
+	err = reset_control_deassert(tdma->rst);
+	if (err) {
+		dev_err(dev, "Failed to deassert reset: %d\n", err);
+		return err;
+	}
+
+	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
+
+	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
+		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
+		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
+		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
+		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
+
+	return 0;
+}
+
+static int tegra_ahbdma_probe(struct platform_device *pdev)
+{
+	struct dma_device *dma_dev;
+	struct tegra_ahbdma *tdma;
+	struct resource *res_regs;
+	unsigned int i;
+	int irq;
+	int err;
+
+	tdma = devm_kzalloc(&pdev->dev, sizeof(*tdma), GFP_KERNEL);
+	if (!tdma)
+		return -ENOMEM;
+
+	irq = platform_get_irq(pdev, 0);
+	if (irq < 0) {
+		dev_err(&pdev->dev, "Failed to get IRQ\n");
+		return irq;
+	}
+
+	err = devm_request_irq(&pdev->dev, irq, tegra_ahbdma_isr, 0,
+			       dev_name(&pdev->dev), tdma);
+	if (err) {
+		dev_err(&pdev->dev, "Failed to request IRQ\n");
+		return -ENODEV;
+	}
+
+	res_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	if (!res_regs)
+		return -ENODEV;
+
+	tdma->regs = devm_ioremap_resource(&pdev->dev, res_regs);
+	if (IS_ERR(tdma->regs))
+		return PTR_ERR(tdma->regs);
+
+	tdma->clk = devm_clk_get(&pdev->dev, NULL);
+	if (IS_ERR(tdma->clk)) {
+		dev_err(&pdev->dev, "Failed to get AHB-DMA clock\n");
+		return PTR_ERR(tdma->clk);
+	}
+
+	tdma->rst = devm_reset_control_get(&pdev->dev, NULL);
+	if (IS_ERR(tdma->rst)) {
+		dev_err(&pdev->dev, "Failed to get AHB-DMA reset\n");
+		return PTR_ERR(tdma->rst);
+	}
+
+	err = tegra_ahbdma_init_hw(tdma, &pdev->dev);
+	if (err)
+		return err;
+
+	dma_dev = &tdma->dma_dev;
+
+	INIT_LIST_HEAD(&dma_dev->channels);
+
+	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
+		tegra_ahbdma_init_channel(tdma, i);
+
+	dma_cap_set(DMA_PRIVATE, dma_dev->cap_mask);
+	dma_cap_set(DMA_CYCLIC, dma_dev->cap_mask);
+	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);
+
+	dma_dev->max_burst		= 8;
+	dma_dev->directions		= TEGRA_AHBDMA_DIRECTIONS;
+	dma_dev->src_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
+	dma_dev->dst_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
+	dma_dev->descriptor_reuse	= true;
+	dma_dev->residue_granularity	= DMA_RESIDUE_GRANULARITY_BURST;
+	dma_dev->device_free_chan_resources = tegra_ahbdma_free_chan_resources;
+	dma_dev->device_prep_slave_sg	= tegra_ahbdma_prep_slave_sg;
+	dma_dev->device_prep_dma_cyclic	= tegra_ahbdma_prep_dma_cyclic;
+	dma_dev->device_terminate_all	= tegra_ahbdma_terminate_all;
+	dma_dev->device_issue_pending	= tegra_ahbdma_issue_pending;
+	dma_dev->device_tx_status	= tegra_ahbdma_tx_status;
+	dma_dev->device_config		= tegra_ahbdma_config;
+	dma_dev->device_synchronize	= tegra_ahbdma_synchronize;
+	dma_dev->dev			= &pdev->dev;
+
+	err = dma_async_device_register(dma_dev);
+	if (err) {
+		dev_err(&pdev->dev, "Device registration failed %d\n", err);
+		return err;
+	}
+
+	err = of_dma_controller_register(pdev->dev.of_node,
+					 tegra_ahbdma_of_xlate, tdma);
+	if (err) {
+		dev_err(&pdev->dev, "OF registration failed %d\n", err);
+		dma_async_device_unregister(dma_dev);
+		return err;
+	}
+
+	platform_set_drvdata(pdev, tdma);
+
+	return 0;
+}
+
+static int tegra_ahbdma_remove(struct platform_device *pdev)
+{
+	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
+
+	of_dma_controller_free(pdev->dev.of_node);
+	dma_async_device_unregister(&tdma->dma_dev);
+	clk_disable_unprepare(tdma->clk);
+
+	return 0;
+}
+
+static const struct of_device_id tegra_ahbdma_of_match[] = {
+	{ .compatible = "nvidia,tegra20-ahbdma" },
+	{ },
+};
+MODULE_DEVICE_TABLE(of, tegra_ahbdma_of_match);
+
+static struct platform_driver tegra_ahbdma_driver = {
+	.driver = {
+		.name	= "tegra-ahbdma",
+		.of_match_table = tegra_ahbdma_of_match,
+	},
+	.probe	= tegra_ahbdma_probe,
+	.remove	= tegra_ahbdma_remove,
+};
+module_platform_driver(tegra_ahbdma_driver);
+
+MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
+MODULE_AUTHOR("Dmitry Osipenko <digetx@gmail.com>");
+MODULE_LICENSE("GPL");
-- 
2.14.1

[PATCH v1 5/5] ARM: dts: tegra: Add AHB DMA controller nodes

[Dmitry Osipenko]

Add AHB DMA controller nodes to Tegra20/30 DT's.

Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
---
 arch/arm/boot/dts/tegra20.dtsi | 9 +++++++++
 arch/arm/boot/dts/tegra30.dtsi | 9 +++++++++
 2 files changed, 18 insertions(+)

diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
index f1579c9a7ef4..d0c0c26427f7 100644
--- a/arch/arm/boot/dts/tegra20.dtsi
+++ b/arch/arm/boot/dts/tegra20.dtsi
@@ -200,6 +200,15 @@
 		reg = <0x60007000 0x1000>;
 	};
 
+	ahbdma: ahbdma@60008000 {
+		compatible = "nvidia,tegra20-ahbdma";
+		reg = <0x60008000 0x2000>;
+		interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
+		clocks = <&tegra_car TEGRA20_CLK_AHBDMA>;
+		resets = <&tegra_car 33>;
+		#dma-cells = <1>;
+	};
+
 	apbdma: dma@6000a000 {
 		compatible = "nvidia,tegra20-apbdma";
 		reg = <0x6000a000 0x1200>;
diff --git a/arch/arm/boot/dts/tegra30.dtsi b/arch/arm/boot/dts/tegra30.dtsi
index 13960fda7471..0800d9a8c546 100644
--- a/arch/arm/boot/dts/tegra30.dtsi
+++ b/arch/arm/boot/dts/tegra30.dtsi
@@ -291,6 +291,15 @@
 		reg = <0x60007000 0x1000>;
 	};
 
+	ahbdma: ahbdma@60008000 {
+		compatible = "nvidia,tegra20-ahbdma";
+		reg = <0x60008000 0x2000>;
+		interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
+		clocks = <&tegra_car TEGRA30_CLK_AHBDMA>;
+		resets = <&tegra_car 33>;
+		#dma-cells = <1>;
+	};
+
 	apbdma: dma@6000a000 {
 		compatible = "nvidia,tegra30-apbdma", "nvidia,tegra20-apbdma";
 		reg = <0x6000a000 0x1400>;
-- 
2.14.1

[PATCH v1 5/5] ARM: dts: tegra: Add AHB DMA controller nodes

[Dmitry Osipenko]

Add AHB DMA controller nodes to Tegra20/30 DT's.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 arch/arm/boot/dts/tegra20.dtsi | 9 +++++++++
 arch/arm/boot/dts/tegra30.dtsi | 9 +++++++++
 2 files changed, 18 insertions(+)

diff --git a/arch/arm/boot/dts/tegra20.dtsi b/arch/arm/boot/dts/tegra20.dtsi
index f1579c9a7ef4..d0c0c26427f7 100644
--- a/arch/arm/boot/dts/tegra20.dtsi
+++ b/arch/arm/boot/dts/tegra20.dtsi
@@ -200,6 +200,15 @@
 		reg = <0x60007000 0x1000>;
 	};
 
+	ahbdma: ahbdma@60008000 {
+		compatible = "nvidia,tegra20-ahbdma";
+		reg = <0x60008000 0x2000>;
+		interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
+		clocks = <&tegra_car TEGRA20_CLK_AHBDMA>;
+		resets = <&tegra_car 33>;
+		#dma-cells = <1>;
+	};
+
 	apbdma: dma@6000a000 {
 		compatible = "nvidia,tegra20-apbdma";
 		reg = <0x6000a000 0x1200>;
diff --git a/arch/arm/boot/dts/tegra30.dtsi b/arch/arm/boot/dts/tegra30.dtsi
index 13960fda7471..0800d9a8c546 100644
--- a/arch/arm/boot/dts/tegra30.dtsi
+++ b/arch/arm/boot/dts/tegra30.dtsi
@@ -291,6 +291,15 @@
 		reg = <0x60007000 0x1000>;
 	};
 
+	ahbdma: ahbdma@60008000 {
+		compatible = "nvidia,tegra20-ahbdma";
+		reg = <0x60008000 0x2000>;
+		interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
+		clocks = <&tegra_car TEGRA30_CLK_AHBDMA>;
+		resets = <&tegra_car 33>;
+		#dma-cells = <1>;
+	};
+
 	apbdma: dma@6000a000 {
 		compatible = "nvidia,tegra30-apbdma", "nvidia,tegra20-apbdma";
 		reg = <0x6000a000 0x1400>;
-- 
2.14.1

Re: [PATCH v1 1/5] clk: tegra: Add AHB DMA clock entry

[Peter De Schrijver]

On Tue, Sep 26, 2017 at 02:22:02AM +0300, Dmitry Osipenko wrote:
> AHB DMA presents on Tegra20/30. Add missing entries, so that driver
> for AHB DMA could be implemented.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/clk/tegra/clk-id.h           | 1 +
>  drivers/clk/tegra/clk-tegra-periph.c | 1 +
>  drivers/clk/tegra/clk-tegra20.c      | 6 ++++++
>  drivers/clk/tegra/clk-tegra30.c      | 2 ++
>  4 files changed, 10 insertions(+)
> 
> diff --git a/drivers/clk/tegra/clk-id.h b/drivers/clk/tegra/clk-id.h
> index 689f344377a7..c1661b47bbda 100644
> --- a/drivers/clk/tegra/clk-id.h
> +++ b/drivers/clk/tegra/clk-id.h
> @@ -12,6 +12,7 @@ enum clk_id {
>  	tegra_clk_amx,
>  	tegra_clk_amx1,
>  	tegra_clk_apb2ape,
> +	tegra_clk_ahbdma,
>  	tegra_clk_apbdma,
>  	tegra_clk_apbif,
>  	tegra_clk_ape,
> diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c
> index 848255cc0209..95a3d8c95f06 100644
> --- a/drivers/clk/tegra/clk-tegra-periph.c
> +++ b/drivers/clk/tegra/clk-tegra-periph.c
> @@ -823,6 +823,7 @@ static struct tegra_periph_init_data gate_clks[] = {
>  	GATE("timer", "clk_m", 5, 0, tegra_clk_timer, CLK_IS_CRITICAL),
>  	GATE("isp", "clk_m", 23, 0, tegra_clk_isp, 0),
>  	GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0),
> +	GATE("ahbdma", "clk_m", 33, 0, tegra_clk_ahbdma, 0),

Parent for this should be hclk on Tegra30 and later chips as well..

>  	GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0),
>  	GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0),
>  	GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0),
> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
> index 837e5cbd60e9..e76c0d292ca7 100644
> --- a/drivers/clk/tegra/clk-tegra20.c
> +++ b/drivers/clk/tegra/clk-tegra20.c
> @@ -449,6 +449,7 @@ static struct tegra_devclk devclks[] __initdata = {
>  	{ .con_id = "audio", .dt_id = TEGRA20_CLK_AUDIO },
>  	{ .con_id = "audio_2x", .dt_id = TEGRA20_CLK_AUDIO_2X },
>  	{ .dev_id = "tegra20-ac97", .dt_id = TEGRA20_CLK_AC97 },
> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA20_CLK_AHBDMA },

This isn't needed if you use DT bindings to get the clock handle.

>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA20_CLK_APBDMA },
>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA20_CLK_RTC },
>  	{ .dev_id = "timer", .dt_id = TEGRA20_CLK_TIMER },
> @@ -806,6 +807,11 @@ static void __init tegra20_periph_clk_init(void)
>  				    clk_base, 0, 3, periph_clk_enb_refcnt);
>  	clks[TEGRA20_CLK_AC97] = clk;
>  
> +	/* ahbdma */
> +	clk = tegra_clk_register_periph_gate("ahbdma", "hclk", 0, clk_base,
> +				    0, 33, periph_clk_enb_refcnt);
> +	clks[TEGRA20_CLK_AHBDMA] = clk;
> +

You can use the generic definition here if you correct the entry above.

>  	/* apbdma */
>  	clk = tegra_clk_register_periph_gate("apbdma", "pclk", 0, clk_base,
>  				    0, 34, periph_clk_enb_refcnt);
> diff --git a/drivers/clk/tegra/clk-tegra30.c b/drivers/clk/tegra/clk-tegra30.c
> index a2d163f759b4..e99701557f29 100644
> --- a/drivers/clk/tegra/clk-tegra30.c
> +++ b/drivers/clk/tegra/clk-tegra30.c
> @@ -612,6 +612,7 @@ static struct tegra_devclk devclks[] __initdata = {
>  	{ .con_id = "fuse_burn", .dev_id = "fuse-tegra", .dt_id = TEGRA30_CLK_FUSE_BURN },
>  	{ .con_id = "apbif", .dev_id = "tegra30-ahub", .dt_id = TEGRA30_CLK_APBIF },
>  	{ .con_id = "hda2hdmi", .dev_id = "tegra30-hda", .dt_id = TEGRA30_CLK_HDA2HDMI },
> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA30_CLK_AHBDMA },

Same as for Tegra20.

>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA30_CLK_APBDMA },
>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA30_CLK_RTC },
>  	{ .dev_id = "timer", .dt_id = TEGRA30_CLK_TIMER },
> @@ -788,6 +789,7 @@ static struct tegra_clk tegra30_clks[tegra_clk_max] __initdata = {
>  	[tegra_clk_extern3] = { .dt_id = TEGRA30_CLK_EXTERN3, .present = true },
>  	[tegra_clk_disp1] = { .dt_id = TEGRA30_CLK_DISP1, .present = true },
>  	[tegra_clk_disp2] = { .dt_id = TEGRA30_CLK_DISP2, .present = true },
> +	[tegra_clk_ahbdma] = { .dt_id = TEGRA30_CLK_AHBDMA, .present = true },
>  	[tegra_clk_apbdma] = { .dt_id = TEGRA30_CLK_APBDMA, .present = true },
>  	[tegra_clk_rtc] = { .dt_id = TEGRA30_CLK_RTC, .present = true },
>  	[tegra_clk_timer] = { .dt_id = TEGRA30_CLK_TIMER, .present = true },

Cheers,

Peter.

Re: [PATCH v1 1/5] clk: tegra: Add AHB DMA clock entry

[Peter De Schrijver]

On Tue, Sep 26, 2017 at 02:22:02AM +0300, Dmitry Osipenko wrote:
> AHB DMA presents on Tegra20/30. Add missing entries, so that driver
> for AHB DMA could be implemented.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/clk/tegra/clk-id.h           | 1 +
>  drivers/clk/tegra/clk-tegra-periph.c | 1 +
>  drivers/clk/tegra/clk-tegra20.c      | 6 ++++++
>  drivers/clk/tegra/clk-tegra30.c      | 2 ++
>  4 files changed, 10 insertions(+)
> 
> diff --git a/drivers/clk/tegra/clk-id.h b/drivers/clk/tegra/clk-id.h
> index 689f344377a7..c1661b47bbda 100644
> --- a/drivers/clk/tegra/clk-id.h
> +++ b/drivers/clk/tegra/clk-id.h
> @@ -12,6 +12,7 @@ enum clk_id {
>  	tegra_clk_amx,
>  	tegra_clk_amx1,
>  	tegra_clk_apb2ape,
> +	tegra_clk_ahbdma,
>  	tegra_clk_apbdma,
>  	tegra_clk_apbif,
>  	tegra_clk_ape,
> diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c
> index 848255cc0209..95a3d8c95f06 100644
> --- a/drivers/clk/tegra/clk-tegra-periph.c
> +++ b/drivers/clk/tegra/clk-tegra-periph.c
> @@ -823,6 +823,7 @@ static struct tegra_periph_init_data gate_clks[] = {
>  	GATE("timer", "clk_m", 5, 0, tegra_clk_timer, CLK_IS_CRITICAL),
>  	GATE("isp", "clk_m", 23, 0, tegra_clk_isp, 0),
>  	GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0),
> +	GATE("ahbdma", "clk_m", 33, 0, tegra_clk_ahbdma, 0),

Parent for this should be hclk on Tegra30 and later chips as well..

>  	GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0),
>  	GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0),
>  	GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0),
> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
> index 837e5cbd60e9..e76c0d292ca7 100644
> --- a/drivers/clk/tegra/clk-tegra20.c
> +++ b/drivers/clk/tegra/clk-tegra20.c
> @@ -449,6 +449,7 @@ static struct tegra_devclk devclks[] __initdata = {
>  	{ .con_id = "audio", .dt_id = TEGRA20_CLK_AUDIO },
>  	{ .con_id = "audio_2x", .dt_id = TEGRA20_CLK_AUDIO_2X },
>  	{ .dev_id = "tegra20-ac97", .dt_id = TEGRA20_CLK_AC97 },
> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA20_CLK_AHBDMA },

This isn't needed if you use DT bindings to get the clock handle.

>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA20_CLK_APBDMA },
>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA20_CLK_RTC },
>  	{ .dev_id = "timer", .dt_id = TEGRA20_CLK_TIMER },
> @@ -806,6 +807,11 @@ static void __init tegra20_periph_clk_init(void)
>  				    clk_base, 0, 3, periph_clk_enb_refcnt);
>  	clks[TEGRA20_CLK_AC97] = clk;
>  
> +	/* ahbdma */
> +	clk = tegra_clk_register_periph_gate("ahbdma", "hclk", 0, clk_base,
> +				    0, 33, periph_clk_enb_refcnt);
> +	clks[TEGRA20_CLK_AHBDMA] = clk;
> +

You can use the generic definition here if you correct the entry above.

>  	/* apbdma */
>  	clk = tegra_clk_register_periph_gate("apbdma", "pclk", 0, clk_base,
>  				    0, 34, periph_clk_enb_refcnt);
> diff --git a/drivers/clk/tegra/clk-tegra30.c b/drivers/clk/tegra/clk-tegra30.c
> index a2d163f759b4..e99701557f29 100644
> --- a/drivers/clk/tegra/clk-tegra30.c
> +++ b/drivers/clk/tegra/clk-tegra30.c
> @@ -612,6 +612,7 @@ static struct tegra_devclk devclks[] __initdata = {
>  	{ .con_id = "fuse_burn", .dev_id = "fuse-tegra", .dt_id = TEGRA30_CLK_FUSE_BURN },
>  	{ .con_id = "apbif", .dev_id = "tegra30-ahub", .dt_id = TEGRA30_CLK_APBIF },
>  	{ .con_id = "hda2hdmi", .dev_id = "tegra30-hda", .dt_id = TEGRA30_CLK_HDA2HDMI },
> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA30_CLK_AHBDMA },

Same as for Tegra20.

>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA30_CLK_APBDMA },
>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA30_CLK_RTC },
>  	{ .dev_id = "timer", .dt_id = TEGRA30_CLK_TIMER },
> @@ -788,6 +789,7 @@ static struct tegra_clk tegra30_clks[tegra_clk_max] __initdata = {
>  	[tegra_clk_extern3] = { .dt_id = TEGRA30_CLK_EXTERN3, .present = true },
>  	[tegra_clk_disp1] = { .dt_id = TEGRA30_CLK_DISP1, .present = true },
>  	[tegra_clk_disp2] = { .dt_id = TEGRA30_CLK_DISP2, .present = true },
> +	[tegra_clk_ahbdma] = { .dt_id = TEGRA30_CLK_AHBDMA, .present = true },
>  	[tegra_clk_apbdma] = { .dt_id = TEGRA30_CLK_APBDMA, .present = true },
>  	[tegra_clk_rtc] = { .dt_id = TEGRA30_CLK_RTC, .present = true },
>  	[tegra_clk_timer] = { .dt_id = TEGRA30_CLK_TIMER, .present = true },

Cheers,

Peter.

Re: [PATCH v1 2/5] clk: tegra: Bump SCLK clock rate to 216MHz on Tegra20

[Peter De Schrijver]

On Tue, Sep 26, 2017 at 02:22:03AM +0300, Dmitry Osipenko wrote:
> AHB DMA is a running on 1/2 of SCLK rate, APB on 1/4. Increasing SCLK rate
> results in an increased DMA transfer rate.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>

Acked-By: Peter De Schrijver <pdeschrijver@nvidia.com>

> ---
>  drivers/clk/tegra/clk-tegra20.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
> index e76c0d292ca7..c511716093e2 100644
> --- a/drivers/clk/tegra/clk-tegra20.c
> +++ b/drivers/clk/tegra/clk-tegra20.c
> @@ -1031,7 +1031,7 @@ static struct tegra_clk_init_table init_table[] __initdata = {
>  	{ TEGRA20_CLK_PLL_P_OUT3, TEGRA20_CLK_CLK_MAX, 72000000, 1 },
>  	{ TEGRA20_CLK_PLL_P_OUT4, TEGRA20_CLK_CLK_MAX, 24000000, 1 },
>  	{ TEGRA20_CLK_PLL_C, TEGRA20_CLK_CLK_MAX, 600000000, 1 },
> -	{ TEGRA20_CLK_PLL_C_OUT1, TEGRA20_CLK_CLK_MAX, 120000000, 1 },
> +	{ TEGRA20_CLK_PLL_C_OUT1, TEGRA20_CLK_CLK_MAX, 216000000, 1 },
>  	{ TEGRA20_CLK_SCLK, TEGRA20_CLK_PLL_C_OUT1, 0, 1 },
>  	{ TEGRA20_CLK_HCLK, TEGRA20_CLK_CLK_MAX, 0, 1 },
>  	{ TEGRA20_CLK_PCLK, TEGRA20_CLK_CLK_MAX, 60000000, 1 },
> -- 
> 2.14.1
> 

Re: [PATCH v1 2/5] clk: tegra: Bump SCLK clock rate to 216MHz on Tegra20

[Peter De Schrijver]

On Tue, Sep 26, 2017 at 02:22:03AM +0300, Dmitry Osipenko wrote:
> AHB DMA is a running on 1/2 of SCLK rate, APB on 1/4. Increasing SCLK rate
> results in an increased DMA transfer rate.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>

Acked-By: Peter De Schrijver <pdeschrijver@nvidia.com>

> ---
>  drivers/clk/tegra/clk-tegra20.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
> index e76c0d292ca7..c511716093e2 100644
> --- a/drivers/clk/tegra/clk-tegra20.c
> +++ b/drivers/clk/tegra/clk-tegra20.c
> @@ -1031,7 +1031,7 @@ static struct tegra_clk_init_table init_table[] __initdata = {
>  	{ TEGRA20_CLK_PLL_P_OUT3, TEGRA20_CLK_CLK_MAX, 72000000, 1 },
>  	{ TEGRA20_CLK_PLL_P_OUT4, TEGRA20_CLK_CLK_MAX, 24000000, 1 },
>  	{ TEGRA20_CLK_PLL_C, TEGRA20_CLK_CLK_MAX, 600000000, 1 },
> -	{ TEGRA20_CLK_PLL_C_OUT1, TEGRA20_CLK_CLK_MAX, 120000000, 1 },
> +	{ TEGRA20_CLK_PLL_C_OUT1, TEGRA20_CLK_CLK_MAX, 216000000, 1 },
>  	{ TEGRA20_CLK_SCLK, TEGRA20_CLK_PLL_C_OUT1, 0, 1 },
>  	{ TEGRA20_CLK_HCLK, TEGRA20_CLK_CLK_MAX, 0, 1 },
>  	{ TEGRA20_CLK_PCLK, TEGRA20_CLK_CLK_MAX, 60000000, 1 },
> -- 
> 2.14.1
> 

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

Hi Dmitry,

On 26/09/17 00:22, Dmitry Osipenko wrote:
> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
> doesn't yet implement transfers larger than 64K and scatter-gather
> transfers that have NENT > 1, HW doesn't have native support for these
> cases.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/Kconfig           |   9 +
>  drivers/dma/Makefile          |   1 +
>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 689 insertions(+)
>  create mode 100644 drivers/dma/tegra20-ahb-dma.c

...

> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
> new file mode 100644
> index 000000000000..8316d64e35e1
> --- /dev/null
> +++ b/drivers/dma/tegra20-ahb-dma.c
> @@ -0,0 +1,679 @@
> +/*
> + * Copyright 2017 Dmitry Osipenko <digetx@gmail.com>
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms and conditions of the GNU General Public License,
> + * version 2, as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope it will be useful, but WITHOUT
> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
> + * more details.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
> + */
> +
> +#include <linux/clk.h>
> +#include <linux/delay.h>
> +#include <linux/interrupt.h>
> +#include <linux/io.h>
> +#include <linux/module.h>
> +#include <linux/of_device.h>
> +#include <linux/of_dma.h>
> +#include <linux/platform_device.h>
> +#include <linux/reset.h>
> +#include <linux/slab.h>
> +#include <linux/spinlock.h>
> +
> +#include "dmaengine.h"
> +
> +#define TEGRA_AHBDMA_CMD			0x0
> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
> +
> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
> +
> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
> +
> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
> +
> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
> +
> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
> +
> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
> +
> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
> +
> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
> +
> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
> +						BIT(DMA_MEM_TO_DEV)
> +
> +struct tegra_ahbdma_tx_desc {
> +	struct dma_async_tx_descriptor desc;
> +	struct tasklet_struct tasklet;
> +	struct list_head node;

Any reason why we cannot use the virt-dma framework for this driver? I
would hope it would simplify the driver a bit.

> +	enum dma_transfer_direction dir;
> +	dma_addr_t mem_paddr;
> +	unsigned long flags;
> +	size_t size;
> +	bool in_fly;
> +	bool cyclic;
> +};
> +
> +struct tegra_ahbdma_chan {
> +	struct dma_chan dma_chan;
> +	struct list_head active_list;
> +	struct list_head pending_list;
> +	struct completion idling;
> +	void __iomem *regs;
> +	spinlock_t lock;
> +	unsigned int id;
> +};
> +
> +struct tegra_ahbdma {
> +	struct tegra_ahbdma_chan channels[4];
> +	struct dma_device dma_dev;
> +	struct reset_control *rst;
> +	struct clk *clk;
> +	void __iomem *regs;
> +};
> +
> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
> +{
> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
> +}
> +
> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
> +{
> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
> +}
> +
> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
> +				struct dma_async_tx_descriptor *tx)
> +{
> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
> +}
> +
> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
> +				   struct tegra_ahbdma_tx_desc *tx)
> +{
> +	u32 csr;
> +
> +	writel_relaxed(tx->mem_paddr,
> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
> +
> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +
> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
> +	csr |= tx->size - sizeof(u32);
> +
> +	if (tx->dir == DMA_DEV_TO_MEM)
> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
> +
> +	if (!tx->cyclic)
> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
> +
> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +
> +	tx->in_fly = true;
> +}
> +
> +static void tegra_ahbdma_tasklet(unsigned long data)
> +{
> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
> +	struct dma_async_tx_descriptor *desc = &tx->desc;
> +
> +	dmaengine_desc_get_callback_invoke(desc, NULL);
> +
> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
> +		kfree(tx);
> +}
> +
> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
> +				      struct tegra_ahbdma_tx_desc *tx)
> +{
> +	struct dma_async_tx_descriptor *desc = &tx->desc;
> +	bool reuse = dmaengine_desc_test_reuse(desc);
> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
> +	bool completed = !tx->cyclic;
> +
> +	if (completed)
> +		dma_cookie_complete(desc);
> +
> +	if (interrupt)
> +		tasklet_schedule(&tx->tasklet);
> +
> +	if (completed) {
> +		list_del(&tx->node);
> +
> +		if (reuse)
> +			tx->in_fly = false;
> +
> +		if (!interrupt && !reuse)
> +			kfree(tx);
> +	}
> +
> +	return completed;
> +}
> +
> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +
> +	tx = list_first_entry_or_null(&chan->active_list,
> +				      struct tegra_ahbdma_tx_desc,
> +				      node);
> +	if (tx)
> +		tegra_ahbdma_submit_tx(chan, tx);
> +
> +	return !!tx;
> +}
> +
> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +	unsigned long flags;
> +	u32 status;
> +
> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
> +		return;
> +
> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
> +
> +	spin_lock_irqsave(&chan->lock, flags);
> +
> +	if (!completion_done(&chan->idling)) {
> +		tx = list_first_entry(&chan->active_list,
> +				      struct tegra_ahbdma_tx_desc,
> +				      node);
> +
> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
> +		    !tegra_ahbdma_next_tx_issued(chan))
> +			complete_all(&chan->idling);
> +	}
> +
> +	spin_unlock_irqrestore(&chan->lock, flags);
> +}
> +
> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
> +{
> +	struct tegra_ahbdma *tdma = dev_id;
> +	unsigned int i;
> +
> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
> +{
> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
> +	dma_cookie_t cookie;
> +
> +	cookie = dma_cookie_assign(desc);
> +
> +	spin_lock_irq(&chan->lock);
> +	list_add_tail(&tx->node, &chan->pending_list);
> +	spin_unlock_irq(&chan->lock);
> +
> +	return cookie;
> +}
> +
> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
> +{
> +	kfree(to_ahbdma_tx_desc(desc));
> +
> +	return 0;
> +}
> +
> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
> +					struct dma_chan *chan,
> +					struct scatterlist *sgl,
> +					unsigned int sg_len,
> +					enum dma_transfer_direction dir,
> +					unsigned long flags,
> +					void *context)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +
> +	/* unimplemented */
> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
> +		return NULL;
> +
> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
> +	if (!tx)
> +		return NULL;
> +
> +	dma_async_tx_descriptor_init(&tx->desc, chan);
> +
> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
> +	tx->mem_paddr		= sg_dma_address(sgl);
> +	tx->size		= sg_dma_len(sgl);
> +	tx->flags		= flags;
> +	tx->dir			= dir;
> +
> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
> +
> +	return &tx->desc;
> +}
> +
> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
> +					struct dma_chan *chan,
> +					dma_addr_t buf_addr,
> +					size_t buf_len,
> +					size_t period_len,
> +					enum dma_transfer_direction dir,
> +					unsigned long flags)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +
> +	/* unimplemented */
> +	if (buf_len != period_len || buf_len > SZ_64K)
> +		return NULL;
> +
> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
> +	if (!tx)
> +		return NULL;
> +
> +	dma_async_tx_descriptor_init(&tx->desc, chan);
> +
> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
> +	tx->mem_paddr		= buf_addr;
> +	tx->size		= buf_len;
> +	tx->flags		= flags;
> +	tx->cyclic		= true;
> +	tx->dir			= dir;
> +
> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
> +
> +	return &tx->desc;
> +}
> +
> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	struct list_head *entry, *tmp;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
> +		list_move_tail(entry, &ahbdma_chan->active_list);
> +
> +	if (completion_done(&ahbdma_chan->idling)) {
> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
> +					      struct tegra_ahbdma_tx_desc,
> +					      node);
> +		if (tx) {
> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
> +			reinit_completion(&ahbdma_chan->idling);
> +		}
> +	}
> +
> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
> +}
> +
> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
> +					      dma_cookie_t cookie,
> +					      struct dma_tx_state *state)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	enum dma_status cookie_status;
> +	unsigned long flags;
> +	size_t residual;
> +	u32 status;
> +
> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
> +
> +	cookie_status = dma_cookie_status(chan, cookie, state);
> +	if (cookie_status != DMA_COMPLETE) {
> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
> +			if (tx->desc.cookie == cookie)
> +				goto found;
> +		}
> +	}
> +
> +	goto unlock;
> +
> +found:
> +	if (tx->in_fly) {
> +		status = readl_relaxed(
> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
> +
> +		residual = status;
> +	} else
> +		residual = tx->size;
> +
> +	dma_set_residue(state, residual);
> +
> +unlock:
> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
> +
> +	return cookie_status;
> +}
> +
> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	struct list_head *entry, *tmp;
> +	u32 csr;
> +
> +	spin_lock_irq(&ahbdma_chan->lock);
> +
> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
> +
> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
> +		list_del(entry);
> +		kfree(tx);
> +	}
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
> +		list_del(entry);
> +		kfree(tx);
> +	}
> +
> +	complete_all(&ahbdma_chan->idling);
> +
> +	spin_unlock_irq(&ahbdma_chan->lock);
> +
> +	return 0;
> +}
> +
> +static int tegra_ahbdma_config(struct dma_chan *chan,
> +			       struct dma_slave_config *sconfig)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	enum dma_transfer_direction dir = sconfig->direction;
> +	u32 burst, ahb_seq, ahb_addr;
> +
> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
> +		return -EINVAL;
> +
> +	if (dir == DMA_DEV_TO_MEM) {
> +		burst    = sconfig->src_maxburst;
> +		ahb_addr = sconfig->src_addr;
> +	} else {
> +		burst    = sconfig->dst_maxburst;
> +		ahb_addr = sconfig->dst_addr;
> +	}
> +
> +	switch (burst) {
> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
> +	default:
> +		return -EINVAL;
> +	}
> +
> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
> +
> +	writel_relaxed(ahb_seq,
> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
> +
> +	writel_relaxed(ahb_addr,
> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> +
> +	return 0;
> +}
> +
> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
> +{
> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
> +}
> +
> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	struct list_head *entry, *tmp;
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
> +		list_del(entry);
> +		kfree(tx);
> +	}
> +}
> +
> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
> +				      unsigned int chan_id)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
> +	struct dma_device *dma_dev = &tdma->dma_dev;
> +
> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
> +	init_completion(&ahbdma_chan->idling);
> +	spin_lock_init(&ahbdma_chan->lock);
> +	complete(&ahbdma_chan->idling);
> +
> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
> +	ahbdma_chan->id = chan_id;
> +
> +	dma_cookie_init(dma_chan);
> +	dma_chan->device = dma_dev;
> +
> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
> +}
> +
> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
> +					      struct of_dma *ofdma)
> +{
> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
> +	struct dma_chan *chan;
> +	u32 csr;
> +
> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
> +	if (!chan)
> +		return NULL;
> +
> +	/* enable channels flow control */
> +	if (dma_spec->args_count == 1) {

The DT doc says #dma-cells should be '1' and so if not equal 1, is this
not an error?

> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;

What about the TRIG_REQ field?

> +
> +		writel_relaxed(csr,
> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +	}
> +	
> +	return chan;
> +}
> +
> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
> +{
> +	int err;
> +
> +	err = reset_control_assert(tdma->rst);
> +	if (err) {
> +		dev_err(dev, "Failed to assert reset: %d\n", err);
> +		return err;
> +	}
> +
> +	err = clk_prepare_enable(tdma->clk);
> +	if (err) {
> +		dev_err(dev, "Failed to enable clock: %d\n", err);
> +		return err;
> +	}
> +
> +	usleep_range(1000, 2000);
> +
> +	err = reset_control_deassert(tdma->rst);
> +	if (err) {
> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
> +		return err;
> +	}
> +
> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
> +
> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
> +
> +	return 0;
> +}

Personally I would use the pm_runtime callbacks for this sort of thing
and ...

> +static int tegra_ahbdma_probe(struct platform_device *pdev)
> +{
> +	struct dma_device *dma_dev;
> +	struct tegra_ahbdma *tdma;
> +	struct resource *res_regs;
> +	unsigned int i;
> +	int irq;
> +	int err;
> +
> +	tdma = devm_kzalloc(&pdev->dev, sizeof(*tdma), GFP_KERNEL);
> +	if (!tdma)
> +		return -ENOMEM;
> +
> +	irq = platform_get_irq(pdev, 0);
> +	if (irq < 0) {
> +		dev_err(&pdev->dev, "Failed to get IRQ\n");
> +		return irq;
> +	}
> +
> +	err = devm_request_irq(&pdev->dev, irq, tegra_ahbdma_isr, 0,
> +			       dev_name(&pdev->dev), tdma);
> +	if (err) {
> +		dev_err(&pdev->dev, "Failed to request IRQ\n");
> +		return -ENODEV;
> +	}
> +
> +	res_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	if (!res_regs)
> +		return -ENODEV;
> +
> +	tdma->regs = devm_ioremap_resource(&pdev->dev, res_regs);
> +	if (IS_ERR(tdma->regs))
> +		return PTR_ERR(tdma->regs);
> +
> +	tdma->clk = devm_clk_get(&pdev->dev, NULL);
> +	if (IS_ERR(tdma->clk)) {
> +		dev_err(&pdev->dev, "Failed to get AHB-DMA clock\n");
> +		return PTR_ERR(tdma->clk);
> +	}
> +
> +	tdma->rst = devm_reset_control_get(&pdev->dev, NULL);
> +	if (IS_ERR(tdma->rst)) {
> +		dev_err(&pdev->dev, "Failed to get AHB-DMA reset\n");
> +		return PTR_ERR(tdma->rst);
> +	}
> +
> +	err = tegra_ahbdma_init_hw(tdma, &pdev->dev);
> +	if (err)
> +		return err;

... here is looks like we turn the clocks on and leave them on. I would
rather that we turn them on when the DMA channel is requested and turn
them off again when freed. Again would be good to use pm_runtime APIs
for this.

> +	dma_dev = &tdma->dma_dev;
> +
> +	INIT_LIST_HEAD(&dma_dev->channels);
> +
> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
> +		tegra_ahbdma_init_channel(tdma, i);
> +
> +	dma_cap_set(DMA_PRIVATE, dma_dev->cap_mask);
> +	dma_cap_set(DMA_CYCLIC, dma_dev->cap_mask);
> +	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);
> +
> +	dma_dev->max_burst		= 8;
> +	dma_dev->directions		= TEGRA_AHBDMA_DIRECTIONS;
> +	dma_dev->src_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
> +	dma_dev->dst_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
> +	dma_dev->descriptor_reuse	= true;
> +	dma_dev->residue_granularity	= DMA_RESIDUE_GRANULARITY_BURST;
> +	dma_dev->device_free_chan_resources = tegra_ahbdma_free_chan_resources;
> +	dma_dev->device_prep_slave_sg	= tegra_ahbdma_prep_slave_sg;
> +	dma_dev->device_prep_dma_cyclic	= tegra_ahbdma_prep_dma_cyclic;
> +	dma_dev->device_terminate_all	= tegra_ahbdma_terminate_all;
> +	dma_dev->device_issue_pending	= tegra_ahbdma_issue_pending;
> +	dma_dev->device_tx_status	= tegra_ahbdma_tx_status;
> +	dma_dev->device_config		= tegra_ahbdma_config;
> +	dma_dev->device_synchronize	= tegra_ahbdma_synchronize;
> +	dma_dev->dev			= &pdev->dev;
> +
> +	err = dma_async_device_register(dma_dev);
> +	if (err) {
> +		dev_err(&pdev->dev, "Device registration failed %d\n", err);
> +		return err;
> +	}
> +
> +	err = of_dma_controller_register(pdev->dev.of_node,
> +					 tegra_ahbdma_of_xlate, tdma);
> +	if (err) {
> +		dev_err(&pdev->dev, "OF registration failed %d\n", err);
> +		dma_async_device_unregister(dma_dev);
> +		return err;
> +	}
> +
> +	platform_set_drvdata(pdev, tdma);
> +
> +	return 0;
> +}
> +
> +static int tegra_ahbdma_remove(struct platform_device *pdev)
> +{
> +	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
> +
> +	of_dma_controller_free(pdev->dev.of_node);
> +	dma_async_device_unregister(&tdma->dma_dev);
> +	clk_disable_unprepare(tdma->clk);
> +
> +	return 0;
> +}
> +
> +static const struct of_device_id tegra_ahbdma_of_match[] = {
> +	{ .compatible = "nvidia,tegra20-ahbdma" },
> +	{ },
> +};
> +MODULE_DEVICE_TABLE(of, tegra_ahbdma_of_match);
> +
> +static struct platform_driver tegra_ahbdma_driver = {
> +	.driver = {
> +		.name	= "tegra-ahbdma",
> +		.of_match_table = tegra_ahbdma_of_match,

It would be nice to have suspend/resume handler too. We could do a
similar thing to the APB dma driver.

> +	},
> +	.probe	= tegra_ahbdma_probe,
> +	.remove	= tegra_ahbdma_remove,
> +};
> +module_platform_driver(tegra_ahbdma_driver);
> +
> +MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
> +MODULE_AUTHOR("Dmitry Osipenko <digetx@gmail.com>");
> +MODULE_LICENSE("GPL");

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

Hi Dmitry,

On 26/09/17 00:22, Dmitry Osipenko wrote:
> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
> doesn't yet implement transfers larger than 64K and scatter-gather
> transfers that have NENT > 1, HW doesn't have native support for these
> cases.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/Kconfig           |   9 +
>  drivers/dma/Makefile          |   1 +
>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 689 insertions(+)
>  create mode 100644 drivers/dma/tegra20-ahb-dma.c

...

> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
> new file mode 100644
> index 000000000000..8316d64e35e1
> --- /dev/null
> +++ b/drivers/dma/tegra20-ahb-dma.c
> @@ -0,0 +1,679 @@
> +/*
> + * Copyright 2017 Dmitry Osipenko <digetx@gmail.com>
> + *
> + * This program is free software; you can redistribute it and/or modify it
> + * under the terms and conditions of the GNU General Public License,
> + * version 2, as published by the Free Software Foundation.
> + *
> + * This program is distributed in the hope it will be useful, but WITHOUT
> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
> + * more details.
> + *
> + * You should have received a copy of the GNU General Public License
> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
> + */
> +
> +#include <linux/clk.h>
> +#include <linux/delay.h>
> +#include <linux/interrupt.h>
> +#include <linux/io.h>
> +#include <linux/module.h>
> +#include <linux/of_device.h>
> +#include <linux/of_dma.h>
> +#include <linux/platform_device.h>
> +#include <linux/reset.h>
> +#include <linux/slab.h>
> +#include <linux/spinlock.h>
> +
> +#include "dmaengine.h"
> +
> +#define TEGRA_AHBDMA_CMD			0x0
> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
> +
> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
> +
> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
> +
> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
> +
> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
> +
> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
> +
> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
> +
> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
> +
> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
> +
> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
> +						BIT(DMA_MEM_TO_DEV)
> +
> +struct tegra_ahbdma_tx_desc {
> +	struct dma_async_tx_descriptor desc;
> +	struct tasklet_struct tasklet;
> +	struct list_head node;

Any reason why we cannot use the virt-dma framework for this driver? I
would hope it would simplify the driver a bit.

> +	enum dma_transfer_direction dir;
> +	dma_addr_t mem_paddr;
> +	unsigned long flags;
> +	size_t size;
> +	bool in_fly;
> +	bool cyclic;
> +};
> +
> +struct tegra_ahbdma_chan {
> +	struct dma_chan dma_chan;
> +	struct list_head active_list;
> +	struct list_head pending_list;
> +	struct completion idling;
> +	void __iomem *regs;
> +	spinlock_t lock;
> +	unsigned int id;
> +};
> +
> +struct tegra_ahbdma {
> +	struct tegra_ahbdma_chan channels[4];
> +	struct dma_device dma_dev;
> +	struct reset_control *rst;
> +	struct clk *clk;
> +	void __iomem *regs;
> +};
> +
> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
> +{
> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
> +}
> +
> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
> +{
> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
> +}
> +
> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
> +				struct dma_async_tx_descriptor *tx)
> +{
> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
> +}
> +
> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
> +				   struct tegra_ahbdma_tx_desc *tx)
> +{
> +	u32 csr;
> +
> +	writel_relaxed(tx->mem_paddr,
> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
> +
> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +
> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
> +	csr |= tx->size - sizeof(u32);
> +
> +	if (tx->dir == DMA_DEV_TO_MEM)
> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
> +
> +	if (!tx->cyclic)
> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
> +
> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +
> +	tx->in_fly = true;
> +}
> +
> +static void tegra_ahbdma_tasklet(unsigned long data)
> +{
> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
> +	struct dma_async_tx_descriptor *desc = &tx->desc;
> +
> +	dmaengine_desc_get_callback_invoke(desc, NULL);
> +
> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
> +		kfree(tx);
> +}
> +
> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
> +				      struct tegra_ahbdma_tx_desc *tx)
> +{
> +	struct dma_async_tx_descriptor *desc = &tx->desc;
> +	bool reuse = dmaengine_desc_test_reuse(desc);
> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
> +	bool completed = !tx->cyclic;
> +
> +	if (completed)
> +		dma_cookie_complete(desc);
> +
> +	if (interrupt)
> +		tasklet_schedule(&tx->tasklet);
> +
> +	if (completed) {
> +		list_del(&tx->node);
> +
> +		if (reuse)
> +			tx->in_fly = false;
> +
> +		if (!interrupt && !reuse)
> +			kfree(tx);
> +	}
> +
> +	return completed;
> +}
> +
> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +
> +	tx = list_first_entry_or_null(&chan->active_list,
> +				      struct tegra_ahbdma_tx_desc,
> +				      node);
> +	if (tx)
> +		tegra_ahbdma_submit_tx(chan, tx);
> +
> +	return !!tx;
> +}
> +
> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +	unsigned long flags;
> +	u32 status;
> +
> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
> +		return;
> +
> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
> +
> +	spin_lock_irqsave(&chan->lock, flags);
> +
> +	if (!completion_done(&chan->idling)) {
> +		tx = list_first_entry(&chan->active_list,
> +				      struct tegra_ahbdma_tx_desc,
> +				      node);
> +
> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
> +		    !tegra_ahbdma_next_tx_issued(chan))
> +			complete_all(&chan->idling);
> +	}
> +
> +	spin_unlock_irqrestore(&chan->lock, flags);
> +}
> +
> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
> +{
> +	struct tegra_ahbdma *tdma = dev_id;
> +	unsigned int i;
> +
> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
> +
> +	return IRQ_HANDLED;
> +}
> +
> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
> +{
> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
> +	dma_cookie_t cookie;
> +
> +	cookie = dma_cookie_assign(desc);
> +
> +	spin_lock_irq(&chan->lock);
> +	list_add_tail(&tx->node, &chan->pending_list);
> +	spin_unlock_irq(&chan->lock);
> +
> +	return cookie;
> +}
> +
> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
> +{
> +	kfree(to_ahbdma_tx_desc(desc));
> +
> +	return 0;
> +}
> +
> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
> +					struct dma_chan *chan,
> +					struct scatterlist *sgl,
> +					unsigned int sg_len,
> +					enum dma_transfer_direction dir,
> +					unsigned long flags,
> +					void *context)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +
> +	/* unimplemented */
> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
> +		return NULL;
> +
> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
> +	if (!tx)
> +		return NULL;
> +
> +	dma_async_tx_descriptor_init(&tx->desc, chan);
> +
> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
> +	tx->mem_paddr		= sg_dma_address(sgl);
> +	tx->size		= sg_dma_len(sgl);
> +	tx->flags		= flags;
> +	tx->dir			= dir;
> +
> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
> +
> +	return &tx->desc;
> +}
> +
> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
> +					struct dma_chan *chan,
> +					dma_addr_t buf_addr,
> +					size_t buf_len,
> +					size_t period_len,
> +					enum dma_transfer_direction dir,
> +					unsigned long flags)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +
> +	/* unimplemented */
> +	if (buf_len != period_len || buf_len > SZ_64K)
> +		return NULL;
> +
> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
> +	if (!tx)
> +		return NULL;
> +
> +	dma_async_tx_descriptor_init(&tx->desc, chan);
> +
> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
> +	tx->mem_paddr		= buf_addr;
> +	tx->size		= buf_len;
> +	tx->flags		= flags;
> +	tx->cyclic		= true;
> +	tx->dir			= dir;
> +
> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
> +
> +	return &tx->desc;
> +}
> +
> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	struct list_head *entry, *tmp;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
> +		list_move_tail(entry, &ahbdma_chan->active_list);
> +
> +	if (completion_done(&ahbdma_chan->idling)) {
> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
> +					      struct tegra_ahbdma_tx_desc,
> +					      node);
> +		if (tx) {
> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
> +			reinit_completion(&ahbdma_chan->idling);
> +		}
> +	}
> +
> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
> +}
> +
> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
> +					      dma_cookie_t cookie,
> +					      struct dma_tx_state *state)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	enum dma_status cookie_status;
> +	unsigned long flags;
> +	size_t residual;
> +	u32 status;
> +
> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
> +
> +	cookie_status = dma_cookie_status(chan, cookie, state);
> +	if (cookie_status != DMA_COMPLETE) {
> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
> +			if (tx->desc.cookie == cookie)
> +				goto found;
> +		}
> +	}
> +
> +	goto unlock;
> +
> +found:
> +	if (tx->in_fly) {
> +		status = readl_relaxed(
> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
> +
> +		residual = status;
> +	} else
> +		residual = tx->size;
> +
> +	dma_set_residue(state, residual);
> +
> +unlock:
> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
> +
> +	return cookie_status;
> +}
> +
> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	struct list_head *entry, *tmp;
> +	u32 csr;
> +
> +	spin_lock_irq(&ahbdma_chan->lock);
> +
> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
> +
> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
> +		list_del(entry);
> +		kfree(tx);
> +	}
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
> +		list_del(entry);
> +		kfree(tx);
> +	}
> +
> +	complete_all(&ahbdma_chan->idling);
> +
> +	spin_unlock_irq(&ahbdma_chan->lock);
> +
> +	return 0;
> +}
> +
> +static int tegra_ahbdma_config(struct dma_chan *chan,
> +			       struct dma_slave_config *sconfig)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	enum dma_transfer_direction dir = sconfig->direction;
> +	u32 burst, ahb_seq, ahb_addr;
> +
> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
> +		return -EINVAL;
> +
> +	if (dir == DMA_DEV_TO_MEM) {
> +		burst    = sconfig->src_maxburst;
> +		ahb_addr = sconfig->src_addr;
> +	} else {
> +		burst    = sconfig->dst_maxburst;
> +		ahb_addr = sconfig->dst_addr;
> +	}
> +
> +	switch (burst) {
> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
> +	default:
> +		return -EINVAL;
> +	}
> +
> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
> +
> +	writel_relaxed(ahb_seq,
> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
> +
> +	writel_relaxed(ahb_addr,
> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> +
> +	return 0;
> +}
> +
> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
> +{
> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
> +}
> +
> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	struct list_head *entry, *tmp;
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
> +		list_del(entry);
> +		kfree(tx);
> +	}
> +}
> +
> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
> +				      unsigned int chan_id)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
> +	struct dma_device *dma_dev = &tdma->dma_dev;
> +
> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
> +	init_completion(&ahbdma_chan->idling);
> +	spin_lock_init(&ahbdma_chan->lock);
> +	complete(&ahbdma_chan->idling);
> +
> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
> +	ahbdma_chan->id = chan_id;
> +
> +	dma_cookie_init(dma_chan);
> +	dma_chan->device = dma_dev;
> +
> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
> +}
> +
> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
> +					      struct of_dma *ofdma)
> +{
> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
> +	struct dma_chan *chan;
> +	u32 csr;
> +
> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
> +	if (!chan)
> +		return NULL;
> +
> +	/* enable channels flow control */
> +	if (dma_spec->args_count == 1) {

The DT doc says #dma-cells should be '1' and so if not equal 1, is this
not an error?

> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;

What about the TRIG_REQ field?

> +
> +		writel_relaxed(csr,
> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
> +	}
> +	
> +	return chan;
> +}
> +
> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
> +{
> +	int err;
> +
> +	err = reset_control_assert(tdma->rst);
> +	if (err) {
> +		dev_err(dev, "Failed to assert reset: %d\n", err);
> +		return err;
> +	}
> +
> +	err = clk_prepare_enable(tdma->clk);
> +	if (err) {
> +		dev_err(dev, "Failed to enable clock: %d\n", err);
> +		return err;
> +	}
> +
> +	usleep_range(1000, 2000);
> +
> +	err = reset_control_deassert(tdma->rst);
> +	if (err) {
> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
> +		return err;
> +	}
> +
> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
> +
> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
> +
> +	return 0;
> +}

Personally I would use the pm_runtime callbacks for this sort of thing
and ...

> +static int tegra_ahbdma_probe(struct platform_device *pdev)
> +{
> +	struct dma_device *dma_dev;
> +	struct tegra_ahbdma *tdma;
> +	struct resource *res_regs;
> +	unsigned int i;
> +	int irq;
> +	int err;
> +
> +	tdma = devm_kzalloc(&pdev->dev, sizeof(*tdma), GFP_KERNEL);
> +	if (!tdma)
> +		return -ENOMEM;
> +
> +	irq = platform_get_irq(pdev, 0);
> +	if (irq < 0) {
> +		dev_err(&pdev->dev, "Failed to get IRQ\n");
> +		return irq;
> +	}
> +
> +	err = devm_request_irq(&pdev->dev, irq, tegra_ahbdma_isr, 0,
> +			       dev_name(&pdev->dev), tdma);
> +	if (err) {
> +		dev_err(&pdev->dev, "Failed to request IRQ\n");
> +		return -ENODEV;
> +	}
> +
> +	res_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
> +	if (!res_regs)
> +		return -ENODEV;
> +
> +	tdma->regs = devm_ioremap_resource(&pdev->dev, res_regs);
> +	if (IS_ERR(tdma->regs))
> +		return PTR_ERR(tdma->regs);
> +
> +	tdma->clk = devm_clk_get(&pdev->dev, NULL);
> +	if (IS_ERR(tdma->clk)) {
> +		dev_err(&pdev->dev, "Failed to get AHB-DMA clock\n");
> +		return PTR_ERR(tdma->clk);
> +	}
> +
> +	tdma->rst = devm_reset_control_get(&pdev->dev, NULL);
> +	if (IS_ERR(tdma->rst)) {
> +		dev_err(&pdev->dev, "Failed to get AHB-DMA reset\n");
> +		return PTR_ERR(tdma->rst);
> +	}
> +
> +	err = tegra_ahbdma_init_hw(tdma, &pdev->dev);
> +	if (err)
> +		return err;

... here is looks like we turn the clocks on and leave them on. I would
rather that we turn them on when the DMA channel is requested and turn
them off again when freed. Again would be good to use pm_runtime APIs
for this.

> +	dma_dev = &tdma->dma_dev;
> +
> +	INIT_LIST_HEAD(&dma_dev->channels);
> +
> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
> +		tegra_ahbdma_init_channel(tdma, i);
> +
> +	dma_cap_set(DMA_PRIVATE, dma_dev->cap_mask);
> +	dma_cap_set(DMA_CYCLIC, dma_dev->cap_mask);
> +	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);
> +
> +	dma_dev->max_burst		= 8;
> +	dma_dev->directions		= TEGRA_AHBDMA_DIRECTIONS;
> +	dma_dev->src_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
> +	dma_dev->dst_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
> +	dma_dev->descriptor_reuse	= true;
> +	dma_dev->residue_granularity	= DMA_RESIDUE_GRANULARITY_BURST;
> +	dma_dev->device_free_chan_resources = tegra_ahbdma_free_chan_resources;
> +	dma_dev->device_prep_slave_sg	= tegra_ahbdma_prep_slave_sg;
> +	dma_dev->device_prep_dma_cyclic	= tegra_ahbdma_prep_dma_cyclic;
> +	dma_dev->device_terminate_all	= tegra_ahbdma_terminate_all;
> +	dma_dev->device_issue_pending	= tegra_ahbdma_issue_pending;
> +	dma_dev->device_tx_status	= tegra_ahbdma_tx_status;
> +	dma_dev->device_config		= tegra_ahbdma_config;
> +	dma_dev->device_synchronize	= tegra_ahbdma_synchronize;
> +	dma_dev->dev			= &pdev->dev;
> +
> +	err = dma_async_device_register(dma_dev);
> +	if (err) {
> +		dev_err(&pdev->dev, "Device registration failed %d\n", err);
> +		return err;
> +	}
> +
> +	err = of_dma_controller_register(pdev->dev.of_node,
> +					 tegra_ahbdma_of_xlate, tdma);
> +	if (err) {
> +		dev_err(&pdev->dev, "OF registration failed %d\n", err);
> +		dma_async_device_unregister(dma_dev);
> +		return err;
> +	}
> +
> +	platform_set_drvdata(pdev, tdma);
> +
> +	return 0;
> +}
> +
> +static int tegra_ahbdma_remove(struct platform_device *pdev)
> +{
> +	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
> +
> +	of_dma_controller_free(pdev->dev.of_node);
> +	dma_async_device_unregister(&tdma->dma_dev);
> +	clk_disable_unprepare(tdma->clk);
> +
> +	return 0;
> +}
> +
> +static const struct of_device_id tegra_ahbdma_of_match[] = {
> +	{ .compatible = "nvidia,tegra20-ahbdma" },
> +	{ },
> +};
> +MODULE_DEVICE_TABLE(of, tegra_ahbdma_of_match);
> +
> +static struct platform_driver tegra_ahbdma_driver = {
> +	.driver = {
> +		.name	= "tegra-ahbdma",
> +		.of_match_table = tegra_ahbdma_of_match,

It would be nice to have suspend/resume handler too. We could do a
similar thing to the APB dma driver.

> +	},
> +	.probe	= tegra_ahbdma_probe,
> +	.remove	= tegra_ahbdma_remove,
> +};
> +module_platform_driver(tegra_ahbdma_driver);
> +
> +MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
> +MODULE_AUTHOR("Dmitry Osipenko <digetx@gmail.com>");
> +MODULE_LICENSE("GPL");

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 1/5] clk: tegra: Add AHB DMA clock entry

[Dmitry Osipenko]

On 26.09.2017 12:56, Peter De Schrijver wrote:
> On Tue, Sep 26, 2017 at 02:22:02AM +0300, Dmitry Osipenko wrote:
>> AHB DMA presents on Tegra20/30. Add missing entries, so that driver
>> for AHB DMA could be implemented.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>> ---
>>  drivers/clk/tegra/clk-id.h           | 1 +
>>  drivers/clk/tegra/clk-tegra-periph.c | 1 +
>>  drivers/clk/tegra/clk-tegra20.c      | 6 ++++++
>>  drivers/clk/tegra/clk-tegra30.c      | 2 ++
>>  4 files changed, 10 insertions(+)
>>
>> diff --git a/drivers/clk/tegra/clk-id.h b/drivers/clk/tegra/clk-id.h
>> index 689f344377a7..c1661b47bbda 100644
>> --- a/drivers/clk/tegra/clk-id.h
>> +++ b/drivers/clk/tegra/clk-id.h
>> @@ -12,6 +12,7 @@ enum clk_id {
>>  	tegra_clk_amx,
>>  	tegra_clk_amx1,
>>  	tegra_clk_apb2ape,
>> +	tegra_clk_ahbdma,
>>  	tegra_clk_apbdma,
>>  	tegra_clk_apbif,
>>  	tegra_clk_ape,
>> diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c
>> index 848255cc0209..95a3d8c95f06 100644
>> --- a/drivers/clk/tegra/clk-tegra-periph.c
>> +++ b/drivers/clk/tegra/clk-tegra-periph.c
>> @@ -823,6 +823,7 @@ static struct tegra_periph_init_data gate_clks[] = {
>>  	GATE("timer", "clk_m", 5, 0, tegra_clk_timer, CLK_IS_CRITICAL),
>>  	GATE("isp", "clk_m", 23, 0, tegra_clk_isp, 0),
>>  	GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0),
>> +	GATE("ahbdma", "clk_m", 33, 0, tegra_clk_ahbdma, 0),
> 
> Parent for this should be hclk on Tegra30 and later chips as well..
> 

It looks like other clocks have a wrong parent too here, aren't they? Like for
example "apbdma" should have "pclk" as a parent, isn't it?

>>  	GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0),
>>  	GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0),
>>  	GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0),
>> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
>> index 837e5cbd60e9..e76c0d292ca7 100644
>> --- a/drivers/clk/tegra/clk-tegra20.c
>> +++ b/drivers/clk/tegra/clk-tegra20.c
>> @@ -449,6 +449,7 @@ static struct tegra_devclk devclks[] __initdata = {
>>  	{ .con_id = "audio", .dt_id = TEGRA20_CLK_AUDIO },
>>  	{ .con_id = "audio_2x", .dt_id = TEGRA20_CLK_AUDIO_2X },
>>  	{ .dev_id = "tegra20-ac97", .dt_id = TEGRA20_CLK_AC97 },
>> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA20_CLK_AHBDMA },
> 
> This isn't needed if you use DT bindings to get the clock handle.
> 

Yes, I added it for consistency. Shouldn't we get rid of that all legacy stuff
already?

>>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA20_CLK_APBDMA },
>>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA20_CLK_RTC },
>>  	{ .dev_id = "timer", .dt_id = TEGRA20_CLK_TIMER },
>> @@ -806,6 +807,11 @@ static void __init tegra20_periph_clk_init(void)
>>  				    clk_base, 0, 3, periph_clk_enb_refcnt);
>>  	clks[TEGRA20_CLK_AC97] = clk;
>>  
>> +	/* ahbdma */
>> +	clk = tegra_clk_register_periph_gate("ahbdma", "hclk", 0, clk_base,
>> +				    0, 33, periph_clk_enb_refcnt);
>> +	clks[TEGRA20_CLK_AHBDMA] = clk;
>> +
> 
> You can use the generic definition here if you correct the entry above.
> 

Good point, same applies to "apbdma". Thank you for the suggestion.

>>  	/* apbdma */
>>  	clk = tegra_clk_register_periph_gate("apbdma", "pclk", 0, clk_base,
>>  				    0, 34, periph_clk_enb_refcnt);
>> diff --git a/drivers/clk/tegra/clk-tegra30.c b/drivers/clk/tegra/clk-tegra30.c
>> index a2d163f759b4..e99701557f29 100644
>> --- a/drivers/clk/tegra/clk-tegra30.c
>> +++ b/drivers/clk/tegra/clk-tegra30.c
>> @@ -612,6 +612,7 @@ static struct tegra_devclk devclks[] __initdata = {
>>  	{ .con_id = "fuse_burn", .dev_id = "fuse-tegra", .dt_id = TEGRA30_CLK_FUSE_BURN },
>>  	{ .con_id = "apbif", .dev_id = "tegra30-ahub", .dt_id = TEGRA30_CLK_APBIF },
>>  	{ .con_id = "hda2hdmi", .dev_id = "tegra30-hda", .dt_id = TEGRA30_CLK_HDA2HDMI },
>> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA30_CLK_AHBDMA },
> 
> Same as for Tegra20.
> 
>>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA30_CLK_APBDMA },
>>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA30_CLK_RTC },
>>  	{ .dev_id = "timer", .dt_id = TEGRA30_CLK_TIMER },
>> @@ -788,6 +789,7 @@ static struct tegra_clk tegra30_clks[tegra_clk_max] __initdata = {
>>  	[tegra_clk_extern3] = { .dt_id = TEGRA30_CLK_EXTERN3, .present = true },
>>  	[tegra_clk_disp1] = { .dt_id = TEGRA30_CLK_DISP1, .present = true },
>>  	[tegra_clk_disp2] = { .dt_id = TEGRA30_CLK_DISP2, .present = true },
>> +	[tegra_clk_ahbdma] = { .dt_id = TEGRA30_CLK_AHBDMA, .present = true },
>>  	[tegra_clk_apbdma] = { .dt_id = TEGRA30_CLK_APBDMA, .present = true },
>>  	[tegra_clk_rtc] = { .dt_id = TEGRA30_CLK_RTC, .present = true },
>>  	[tegra_clk_timer] = { .dt_id = TEGRA30_CLK_TIMER, .present = true },
> 
> Cheers,
> 
> Peter.
> 


-- 
Dmitry

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 26/09/17 00:22, Dmitry Osipenko wrote:
> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
> on Tegra20/30 SoC's.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>  1 file changed, 23 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> 
> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> new file mode 100644
> index 000000000000..2af9aa76ae11
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> @@ -0,0 +1,23 @@
> +* NVIDIA Tegra AHB DMA controller
> +
> +Required properties:
> +- compatible:	Must be "nvidia,tegra20-ahbdma"
> +- reg:		Should contain registers base address and length.
> +- interrupts:	Should contain one entry, DMA controller interrupt.
> +- clocks:	Should contain one entry, DMA controller clock.
> +- resets :	Should contain one entry, DMA controller reset.
> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
> +		for the peripheral. For more details consult the Tegra TRM's
> +		documentation, in particular AHB DMA channel control register
> +		REQ_SEL field.

What about the TRIG_SEL field? Do we need to handle this here as well?

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 26/09/17 00:22, Dmitry Osipenko wrote:
> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
> on Tegra20/30 SoC's.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>  1 file changed, 23 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> 
> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> new file mode 100644
> index 000000000000..2af9aa76ae11
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> @@ -0,0 +1,23 @@
> +* NVIDIA Tegra AHB DMA controller
> +
> +Required properties:
> +- compatible:	Must be "nvidia,tegra20-ahbdma"
> +- reg:		Should contain registers base address and length.
> +- interrupts:	Should contain one entry, DMA controller interrupt.
> +- clocks:	Should contain one entry, DMA controller clock.
> +- resets :	Should contain one entry, DMA controller reset.
> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
> +		for the peripheral. For more details consult the Tegra TRM's
> +		documentation, in particular AHB DMA channel control register
> +		REQ_SEL field.

What about the TRIG_SEL field? Do we need to handle this here as well?

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 26.09.2017 17:50, Jon Hunter wrote:
> 
> On 26/09/17 00:22, Dmitry Osipenko wrote:
>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>> on Tegra20/30 SoC's.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>> ---
>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>  1 file changed, 23 insertions(+)
>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>
>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> new file mode 100644
>> index 000000000000..2af9aa76ae11
>> --- /dev/null
>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> @@ -0,0 +1,23 @@
>> +* NVIDIA Tegra AHB DMA controller
>> +
>> +Required properties:
>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>> +- reg:		Should contain registers base address and length.
>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>> +- clocks:	Should contain one entry, DMA controller clock.
>> +- resets :	Should contain one entry, DMA controller reset.
>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>> +		for the peripheral. For more details consult the Tegra TRM's
>> +		documentation, in particular AHB DMA channel control register
>> +		REQ_SEL field.
> 
> What about the TRIG_SEL field? Do we need to handle this here as well?
> 

I've followed APB DMA here, that HW also has TRIG_SEL but ignores it for some
reason. I think technically it should be present in the binding, yeah.

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

Hi Jon,

On 26.09.2017 17:45, Jon Hunter wrote:
> Hi Dmitry,
> 
> On 26/09/17 00:22, Dmitry Osipenko wrote:
>> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
>> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
>> doesn't yet implement transfers larger than 64K and scatter-gather
>> transfers that have NENT > 1, HW doesn't have native support for these
>> cases.
>>
>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>> ---
>>  drivers/dma/Kconfig           |   9 +
>>  drivers/dma/Makefile          |   1 +
>>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>>  3 files changed, 689 insertions(+)
>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
> 
> ...
> 
>> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
>> new file mode 100644
>> index 000000000000..8316d64e35e1
>> --- /dev/null
>> +++ b/drivers/dma/tegra20-ahb-dma.c
>> @@ -0,0 +1,679 @@
>> +/*
>> + * Copyright 2017 Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>> + *
>> + * This program is free software; you can redistribute it and/or modify it
>> + * under the terms and conditions of the GNU General Public License,
>> + * version 2, as published by the Free Software Foundation.
>> + *
>> + * This program is distributed in the hope it will be useful, but WITHOUT
>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>> + * more details.
>> + *
>> + * You should have received a copy of the GNU General Public License
>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>> + */
>> +
>> +#include <linux/clk.h>
>> +#include <linux/delay.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/io.h>
>> +#include <linux/module.h>
>> +#include <linux/of_device.h>
>> +#include <linux/of_dma.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/reset.h>
>> +#include <linux/slab.h>
>> +#include <linux/spinlock.h>
>> +
>> +#include "dmaengine.h"
>> +
>> +#define TEGRA_AHBDMA_CMD			0x0
>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>> +
>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
>> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
>> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
>> +
>> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +
>> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
>> +						BIT(DMA_MEM_TO_DEV)
>> +
>> +struct tegra_ahbdma_tx_desc {
>> +	struct dma_async_tx_descriptor desc;
>> +	struct tasklet_struct tasklet;
>> +	struct list_head node;
> 
> Any reason why we cannot use the virt-dma framework for this driver? I
> would hope it would simplify the driver a bit.
> 

IIUC virt-dma is supposed to provide virtually unlimited number of channels.
I've looked at it and decided that it would just add unnecessary functionality
and, as a result, complexity. As I wrote in the cover-letter, it is supposed
that this driver would have only one consumer - the host1x. It shouldn't be
difficult to implement virt-dma later, if desired.  But again it is very
unlikely that it would be needed.

>> +	enum dma_transfer_direction dir;
>> +	dma_addr_t mem_paddr;
>> +	unsigned long flags;
>> +	size_t size;
>> +	bool in_fly;
>> +	bool cyclic;
>> +};
>> +
>> +struct tegra_ahbdma_chan {
>> +	struct dma_chan dma_chan;
>> +	struct list_head active_list;
>> +	struct list_head pending_list;
>> +	struct completion idling;
>> +	void __iomem *regs;
>> +	spinlock_t lock;
>> +	unsigned int id;
>> +};
>> +
>> +struct tegra_ahbdma {
>> +	struct tegra_ahbdma_chan channels[4];
>> +	struct dma_device dma_dev;
>> +	struct reset_control *rst;
>> +	struct clk *clk;
>> +	void __iomem *regs;
>> +};
>> +
>> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
>> +{
>> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
>> +}
>> +
>> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
>> +{
>> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
>> +}
>> +
>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>> +				struct dma_async_tx_descriptor *tx)
>> +{
>> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
>> +}
>> +
>> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
>> +				   struct tegra_ahbdma_tx_desc *tx)
>> +{
>> +	u32 csr;
>> +
>> +	writel_relaxed(tx->mem_paddr,
>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
>> +
>> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +
>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
>> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
>> +	csr |= tx->size - sizeof(u32);
>> +
>> +	if (tx->dir == DMA_DEV_TO_MEM)
>> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>> +
>> +	if (!tx->cyclic)
>> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
>> +
>> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +
>> +	tx->in_fly = true;
>> +}
>> +
>> +static void tegra_ahbdma_tasklet(unsigned long data)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>> +
>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>> +
>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>> +		kfree(tx);
>> +}
>> +
>> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
>> +				      struct tegra_ahbdma_tx_desc *tx)
>> +{
>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>> +	bool reuse = dmaengine_desc_test_reuse(desc);
>> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
>> +	bool completed = !tx->cyclic;
>> +
>> +	if (completed)
>> +		dma_cookie_complete(desc);
>> +
>> +	if (interrupt)
>> +		tasklet_schedule(&tx->tasklet);
>> +
>> +	if (completed) {
>> +		list_del(&tx->node);
>> +
>> +		if (reuse)
>> +			tx->in_fly = false;
>> +
>> +		if (!interrupt && !reuse)
>> +			kfree(tx);
>> +	}
>> +
>> +	return completed;
>> +}
>> +
>> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	tx = list_first_entry_or_null(&chan->active_list,
>> +				      struct tegra_ahbdma_tx_desc,
>> +				      node);
>> +	if (tx)
>> +		tegra_ahbdma_submit_tx(chan, tx);
>> +
>> +	return !!tx;
>> +}
>> +
>> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	unsigned long flags;
>> +	u32 status;
>> +
>> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
>> +		return;
>> +
>> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>> +
>> +	spin_lock_irqsave(&chan->lock, flags);
>> +
>> +	if (!completion_done(&chan->idling)) {
>> +		tx = list_first_entry(&chan->active_list,
>> +				      struct tegra_ahbdma_tx_desc,
>> +				      node);
>> +
>> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
>> +		    !tegra_ahbdma_next_tx_issued(chan))
>> +			complete_all(&chan->idling);
>> +	}
>> +
>> +	spin_unlock_irqrestore(&chan->lock, flags);
>> +}
>> +
>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>> +{
>> +	struct tegra_ahbdma *tdma = dev_id;
>> +	unsigned int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
>> +
>> +	return IRQ_HANDLED;
>> +}
>> +
>> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
>> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
>> +	dma_cookie_t cookie;
>> +
>> +	cookie = dma_cookie_assign(desc);
>> +
>> +	spin_lock_irq(&chan->lock);
>> +	list_add_tail(&tx->node, &chan->pending_list);
>> +	spin_unlock_irq(&chan->lock);
>> +
>> +	return cookie;
>> +}
>> +
>> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
>> +{
>> +	kfree(to_ahbdma_tx_desc(desc));
>> +
>> +	return 0;
>> +}
>> +
>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
>> +					struct dma_chan *chan,
>> +					struct scatterlist *sgl,
>> +					unsigned int sg_len,
>> +					enum dma_transfer_direction dir,
>> +					unsigned long flags,
>> +					void *context)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	/* unimplemented */
>> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>> +		return NULL;
>> +
>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>> +	if (!tx)
>> +		return NULL;
>> +
>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>> +
>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
>> +	tx->mem_paddr		= sg_dma_address(sgl);
>> +	tx->size		= sg_dma_len(sgl);
>> +	tx->flags		= flags;
>> +	tx->dir			= dir;
>> +
>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>> +
>> +	return &tx->desc;
>> +}
>> +
>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>> +					struct dma_chan *chan,
>> +					dma_addr_t buf_addr,
>> +					size_t buf_len,
>> +					size_t period_len,
>> +					enum dma_transfer_direction dir,
>> +					unsigned long flags)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	/* unimplemented */
>> +	if (buf_len != period_len || buf_len > SZ_64K)
>> +		return NULL;
>> +
>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>> +	if (!tx)
>> +		return NULL;
>> +
>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>> +
>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>> +	tx->mem_paddr		= buf_addr;
>> +	tx->size		= buf_len;
>> +	tx->flags		= flags;
>> +	tx->cyclic		= true;
>> +	tx->dir			= dir;
>> +
>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>> +
>> +	return &tx->desc;
>> +}
>> +
>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +	unsigned long flags;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>> +
>> +	if (completion_done(&ahbdma_chan->idling)) {
>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>> +					      struct tegra_ahbdma_tx_desc,
>> +					      node);
>> +		if (tx) {
>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
>> +			reinit_completion(&ahbdma_chan->idling);
>> +		}
>> +	}
>> +
>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>> +}
>> +
>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>> +					      dma_cookie_t cookie,
>> +					      struct dma_tx_state *state)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	enum dma_status cookie_status;
>> +	unsigned long flags;
>> +	size_t residual;
>> +	u32 status;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>> +	if (cookie_status != DMA_COMPLETE) {
>> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
>> +			if (tx->desc.cookie == cookie)
>> +				goto found;
>> +		}
>> +	}
>> +
>> +	goto unlock;
>> +
>> +found:
>> +	if (tx->in_fly) {
>> +		status = readl_relaxed(
>> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>> +
>> +		residual = status;
>> +	} else
>> +		residual = tx->size;
>> +
>> +	dma_set_residue(state, residual);
>> +
>> +unlock:
>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>> +
>> +	return cookie_status;
>> +}
>> +
>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +	u32 csr;
>> +
>> +	spin_lock_irq(&ahbdma_chan->lock);
>> +
>> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
>> +
>> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>> +		list_del(entry);
>> +		kfree(tx);
>> +	}
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>> +		list_del(entry);
>> +		kfree(tx);
>> +	}
>> +
>> +	complete_all(&ahbdma_chan->idling);
>> +
>> +	spin_unlock_irq(&ahbdma_chan->lock);
>> +
>> +	return 0;
>> +}
>> +
>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>> +			       struct dma_slave_config *sconfig)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	enum dma_transfer_direction dir = sconfig->direction;
>> +	u32 burst, ahb_seq, ahb_addr;
>> +
>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +		return -EINVAL;
>> +
>> +	if (dir == DMA_DEV_TO_MEM) {
>> +		burst    = sconfig->src_maxburst;
>> +		ahb_addr = sconfig->src_addr;
>> +	} else {
>> +		burst    = sconfig->dst_maxburst;
>> +		ahb_addr = sconfig->dst_addr;
>> +	}
>> +
>> +	switch (burst) {
>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>> +
>> +	writel_relaxed(ahb_seq,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>> +
>> +	writel_relaxed(ahb_addr,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>> +
>> +	return 0;
>> +}
>> +
>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>> +{
>> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
>> +}
>> +
>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>> +		list_del(entry);
>> +		kfree(tx);
>> +	}
>> +}
>> +
>> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
>> +				      unsigned int chan_id)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
>> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
>> +	struct dma_device *dma_dev = &tdma->dma_dev;
>> +
>> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
>> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
>> +	init_completion(&ahbdma_chan->idling);
>> +	spin_lock_init(&ahbdma_chan->lock);
>> +	complete(&ahbdma_chan->idling);
>> +
>> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
>> +	ahbdma_chan->id = chan_id;
>> +
>> +	dma_cookie_init(dma_chan);
>> +	dma_chan->device = dma_dev;
>> +
>> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
>> +}
>> +
>> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
>> +					      struct of_dma *ofdma)
>> +{
>> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
>> +	struct dma_chan *chan;
>> +	u32 csr;
>> +
>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>> +	if (!chan)
>> +		return NULL;
>> +
>> +	/* enable channels flow control */
>> +	if (dma_spec->args_count == 1) {
> 
> The DT doc says #dma-cells should be '1' and so if not equal 1, is this
> not an error?
> 

I wanted to differentiate slave/master modes here. But if we'd want to add
TRIG_SEL as another cell, then it probably would worth to implement a custom DMA
configure options, like documentation suggests - to wrap generic
dma_slave_config into the custom one. On the other hand that probably would add
an unused functionality to the driver.

>> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
>> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;
> 
> What about the TRIG_REQ field?
> 

Not implemented, there is no test case for it yet.

>> +
>> +		writel_relaxed(csr,
>> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +	}
>> +	
>> +	return chan;
>> +}
>> +
>> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
>> +{
>> +	int err;
>> +
>> +	err = reset_control_assert(tdma->rst);
>> +	if (err) {
>> +		dev_err(dev, "Failed to assert reset: %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	err = clk_prepare_enable(tdma->clk);
>> +	if (err) {
>> +		dev_err(dev, "Failed to enable clock: %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	usleep_range(1000, 2000);
>> +
>> +	err = reset_control_deassert(tdma->rst);
>> +	if (err) {
>> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
>> +
>> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
>> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
>> +
>> +	return 0;
>> +}
> 
> Personally I would use the pm_runtime callbacks for this sort of thing
> and ...
> 

I decided that it probaby would be better to implement PM later if needed. I'm
not sure whether DMA controller consumes any substantial amounts of power while
idling. If it's not, why bother? Unnecessary power managment would just cause
CPU to waste its cycles (and power) doing PM.

>> +static int tegra_ahbdma_probe(struct platform_device *pdev)
>> +{
>> +	struct dma_device *dma_dev;
>> +	struct tegra_ahbdma *tdma;
>> +	struct resource *res_regs;
>> +	unsigned int i;
>> +	int irq;
>> +	int err;
>> +
>> +	tdma = devm_kzalloc(&pdev->dev, sizeof(*tdma), GFP_KERNEL);
>> +	if (!tdma)
>> +		return -ENOMEM;
>> +
>> +	irq = platform_get_irq(pdev, 0);
>> +	if (irq < 0) {
>> +		dev_err(&pdev->dev, "Failed to get IRQ\n");
>> +		return irq;
>> +	}
>> +
>> +	err = devm_request_irq(&pdev->dev, irq, tegra_ahbdma_isr, 0,
>> +			       dev_name(&pdev->dev), tdma);
>> +	if (err) {
>> +		dev_err(&pdev->dev, "Failed to request IRQ\n");
>> +		return -ENODEV;
>> +	}
>> +
>> +	res_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>> +	if (!res_regs)
>> +		return -ENODEV;
>> +
>> +	tdma->regs = devm_ioremap_resource(&pdev->dev, res_regs);
>> +	if (IS_ERR(tdma->regs))
>> +		return PTR_ERR(tdma->regs);
>> +
>> +	tdma->clk = devm_clk_get(&pdev->dev, NULL);
>> +	if (IS_ERR(tdma->clk)) {
>> +		dev_err(&pdev->dev, "Failed to get AHB-DMA clock\n");
>> +		return PTR_ERR(tdma->clk);
>> +	}
>> +
>> +	tdma->rst = devm_reset_control_get(&pdev->dev, NULL);
>> +	if (IS_ERR(tdma->rst)) {
>> +		dev_err(&pdev->dev, "Failed to get AHB-DMA reset\n");
>> +		return PTR_ERR(tdma->rst);
>> +	}
>> +
>> +	err = tegra_ahbdma_init_hw(tdma, &pdev->dev);
>> +	if (err)
>> +		return err;
> 
> ... here is looks like we turn the clocks on and leave them on. I would
> rather that we turn them on when the DMA channel is requested and turn
> them off again when freed. Again would be good to use pm_runtime APIs
> for this.
> 

Again not sure about it :)

>> +	dma_dev = &tdma->dma_dev;
>> +
>> +	INIT_LIST_HEAD(&dma_dev->channels);
>> +
>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>> +		tegra_ahbdma_init_channel(tdma, i);
>> +
>> +	dma_cap_set(DMA_PRIVATE, dma_dev->cap_mask);
>> +	dma_cap_set(DMA_CYCLIC, dma_dev->cap_mask);
>> +	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);
>> +
>> +	dma_dev->max_burst		= 8;
>> +	dma_dev->directions		= TEGRA_AHBDMA_DIRECTIONS;
>> +	dma_dev->src_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
>> +	dma_dev->dst_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
>> +	dma_dev->descriptor_reuse	= true;
>> +	dma_dev->residue_granularity	= DMA_RESIDUE_GRANULARITY_BURST;
>> +	dma_dev->device_free_chan_resources = tegra_ahbdma_free_chan_resources;
>> +	dma_dev->device_prep_slave_sg	= tegra_ahbdma_prep_slave_sg;
>> +	dma_dev->device_prep_dma_cyclic	= tegra_ahbdma_prep_dma_cyclic;
>> +	dma_dev->device_terminate_all	= tegra_ahbdma_terminate_all;
>> +	dma_dev->device_issue_pending	= tegra_ahbdma_issue_pending;
>> +	dma_dev->device_tx_status	= tegra_ahbdma_tx_status;
>> +	dma_dev->device_config		= tegra_ahbdma_config;
>> +	dma_dev->device_synchronize	= tegra_ahbdma_synchronize;
>> +	dma_dev->dev			= &pdev->dev;
>> +
>> +	err = dma_async_device_register(dma_dev);
>> +	if (err) {
>> +		dev_err(&pdev->dev, "Device registration failed %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	err = of_dma_controller_register(pdev->dev.of_node,
>> +					 tegra_ahbdma_of_xlate, tdma);
>> +	if (err) {
>> +		dev_err(&pdev->dev, "OF registration failed %d\n", err);
>> +		dma_async_device_unregister(dma_dev);
>> +		return err;
>> +	}
>> +
>> +	platform_set_drvdata(pdev, tdma);
>> +
>> +	return 0;
>> +}
>> +
>> +static int tegra_ahbdma_remove(struct platform_device *pdev)
>> +{
>> +	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
>> +
>> +	of_dma_controller_free(pdev->dev.of_node);
>> +	dma_async_device_unregister(&tdma->dma_dev);
>> +	clk_disable_unprepare(tdma->clk);
>> +
>> +	return 0;
>> +}
>> +
>> +static const struct of_device_id tegra_ahbdma_of_match[] = {
>> +	{ .compatible = "nvidia,tegra20-ahbdma" },
>> +	{ },
>> +};
>> +MODULE_DEVICE_TABLE(of, tegra_ahbdma_of_match);
>> +
>> +static struct platform_driver tegra_ahbdma_driver = {
>> +	.driver = {
>> +		.name	= "tegra-ahbdma",
>> +		.of_match_table = tegra_ahbdma_of_match,
> 
> It would be nice to have suspend/resume handler too. We could do a
> similar thing to the APB dma driver.
> 

It is not stricly necessary because LP0 isn't implemented by the core arch. I've
tested LP1 and it works fine that way. I'd prefer to implement suspend/resume
later, we can't really test it properly without LP0.

>> +	},
>> +	.probe	= tegra_ahbdma_probe,
>> +	.remove	= tegra_ahbdma_remove,
>> +};
>> +module_platform_driver(tegra_ahbdma_driver);
>> +
>> +MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
>> +MODULE_AUTHOR("Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>");
>> +MODULE_LICENSE("GPL");

-- 
Dmitry
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

Hi Jon,

On 26.09.2017 17:45, Jon Hunter wrote:
> Hi Dmitry,
> 
> On 26/09/17 00:22, Dmitry Osipenko wrote:
>> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
>> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
>> doesn't yet implement transfers larger than 64K and scatter-gather
>> transfers that have NENT > 1, HW doesn't have native support for these
>> cases.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>> ---
>>  drivers/dma/Kconfig           |   9 +
>>  drivers/dma/Makefile          |   1 +
>>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>>  3 files changed, 689 insertions(+)
>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
> 
> ...
> 
>> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
>> new file mode 100644
>> index 000000000000..8316d64e35e1
>> --- /dev/null
>> +++ b/drivers/dma/tegra20-ahb-dma.c
>> @@ -0,0 +1,679 @@
>> +/*
>> + * Copyright 2017 Dmitry Osipenko <digetx@gmail.com>
>> + *
>> + * This program is free software; you can redistribute it and/or modify it
>> + * under the terms and conditions of the GNU General Public License,
>> + * version 2, as published by the Free Software Foundation.
>> + *
>> + * This program is distributed in the hope it will be useful, but WITHOUT
>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>> + * more details.
>> + *
>> + * You should have received a copy of the GNU General Public License
>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>> + */
>> +
>> +#include <linux/clk.h>
>> +#include <linux/delay.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/io.h>
>> +#include <linux/module.h>
>> +#include <linux/of_device.h>
>> +#include <linux/of_dma.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/reset.h>
>> +#include <linux/slab.h>
>> +#include <linux/spinlock.h>
>> +
>> +#include "dmaengine.h"
>> +
>> +#define TEGRA_AHBDMA_CMD			0x0
>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>> +
>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
>> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
>> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
>> +
>> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +
>> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
>> +						BIT(DMA_MEM_TO_DEV)
>> +
>> +struct tegra_ahbdma_tx_desc {
>> +	struct dma_async_tx_descriptor desc;
>> +	struct tasklet_struct tasklet;
>> +	struct list_head node;
> 
> Any reason why we cannot use the virt-dma framework for this driver? I
> would hope it would simplify the driver a bit.
> 

IIUC virt-dma is supposed to provide virtually unlimited number of channels.
I've looked at it and decided that it would just add unnecessary functionality
and, as a result, complexity. As I wrote in the cover-letter, it is supposed
that this driver would have only one consumer - the host1x. It shouldn't be
difficult to implement virt-dma later, if desired.  But again it is very
unlikely that it would be needed.

>> +	enum dma_transfer_direction dir;
>> +	dma_addr_t mem_paddr;
>> +	unsigned long flags;
>> +	size_t size;
>> +	bool in_fly;
>> +	bool cyclic;
>> +};
>> +
>> +struct tegra_ahbdma_chan {
>> +	struct dma_chan dma_chan;
>> +	struct list_head active_list;
>> +	struct list_head pending_list;
>> +	struct completion idling;
>> +	void __iomem *regs;
>> +	spinlock_t lock;
>> +	unsigned int id;
>> +};
>> +
>> +struct tegra_ahbdma {
>> +	struct tegra_ahbdma_chan channels[4];
>> +	struct dma_device dma_dev;
>> +	struct reset_control *rst;
>> +	struct clk *clk;
>> +	void __iomem *regs;
>> +};
>> +
>> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
>> +{
>> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
>> +}
>> +
>> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
>> +{
>> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
>> +}
>> +
>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>> +				struct dma_async_tx_descriptor *tx)
>> +{
>> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
>> +}
>> +
>> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
>> +				   struct tegra_ahbdma_tx_desc *tx)
>> +{
>> +	u32 csr;
>> +
>> +	writel_relaxed(tx->mem_paddr,
>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
>> +
>> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +
>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
>> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
>> +	csr |= tx->size - sizeof(u32);
>> +
>> +	if (tx->dir == DMA_DEV_TO_MEM)
>> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>> +
>> +	if (!tx->cyclic)
>> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
>> +
>> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +
>> +	tx->in_fly = true;
>> +}
>> +
>> +static void tegra_ahbdma_tasklet(unsigned long data)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>> +
>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>> +
>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>> +		kfree(tx);
>> +}
>> +
>> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
>> +				      struct tegra_ahbdma_tx_desc *tx)
>> +{
>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>> +	bool reuse = dmaengine_desc_test_reuse(desc);
>> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
>> +	bool completed = !tx->cyclic;
>> +
>> +	if (completed)
>> +		dma_cookie_complete(desc);
>> +
>> +	if (interrupt)
>> +		tasklet_schedule(&tx->tasklet);
>> +
>> +	if (completed) {
>> +		list_del(&tx->node);
>> +
>> +		if (reuse)
>> +			tx->in_fly = false;
>> +
>> +		if (!interrupt && !reuse)
>> +			kfree(tx);
>> +	}
>> +
>> +	return completed;
>> +}
>> +
>> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	tx = list_first_entry_or_null(&chan->active_list,
>> +				      struct tegra_ahbdma_tx_desc,
>> +				      node);
>> +	if (tx)
>> +		tegra_ahbdma_submit_tx(chan, tx);
>> +
>> +	return !!tx;
>> +}
>> +
>> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	unsigned long flags;
>> +	u32 status;
>> +
>> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
>> +		return;
>> +
>> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>> +
>> +	spin_lock_irqsave(&chan->lock, flags);
>> +
>> +	if (!completion_done(&chan->idling)) {
>> +		tx = list_first_entry(&chan->active_list,
>> +				      struct tegra_ahbdma_tx_desc,
>> +				      node);
>> +
>> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
>> +		    !tegra_ahbdma_next_tx_issued(chan))
>> +			complete_all(&chan->idling);
>> +	}
>> +
>> +	spin_unlock_irqrestore(&chan->lock, flags);
>> +}
>> +
>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>> +{
>> +	struct tegra_ahbdma *tdma = dev_id;
>> +	unsigned int i;
>> +
>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
>> +
>> +	return IRQ_HANDLED;
>> +}
>> +
>> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
>> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
>> +	dma_cookie_t cookie;
>> +
>> +	cookie = dma_cookie_assign(desc);
>> +
>> +	spin_lock_irq(&chan->lock);
>> +	list_add_tail(&tx->node, &chan->pending_list);
>> +	spin_unlock_irq(&chan->lock);
>> +
>> +	return cookie;
>> +}
>> +
>> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
>> +{
>> +	kfree(to_ahbdma_tx_desc(desc));
>> +
>> +	return 0;
>> +}
>> +
>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
>> +					struct dma_chan *chan,
>> +					struct scatterlist *sgl,
>> +					unsigned int sg_len,
>> +					enum dma_transfer_direction dir,
>> +					unsigned long flags,
>> +					void *context)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	/* unimplemented */
>> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>> +		return NULL;
>> +
>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>> +	if (!tx)
>> +		return NULL;
>> +
>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>> +
>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
>> +	tx->mem_paddr		= sg_dma_address(sgl);
>> +	tx->size		= sg_dma_len(sgl);
>> +	tx->flags		= flags;
>> +	tx->dir			= dir;
>> +
>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>> +
>> +	return &tx->desc;
>> +}
>> +
>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>> +					struct dma_chan *chan,
>> +					dma_addr_t buf_addr,
>> +					size_t buf_len,
>> +					size_t period_len,
>> +					enum dma_transfer_direction dir,
>> +					unsigned long flags)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	/* unimplemented */
>> +	if (buf_len != period_len || buf_len > SZ_64K)
>> +		return NULL;
>> +
>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>> +	if (!tx)
>> +		return NULL;
>> +
>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>> +
>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>> +	tx->mem_paddr		= buf_addr;
>> +	tx->size		= buf_len;
>> +	tx->flags		= flags;
>> +	tx->cyclic		= true;
>> +	tx->dir			= dir;
>> +
>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>> +
>> +	return &tx->desc;
>> +}
>> +
>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +	unsigned long flags;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>> +
>> +	if (completion_done(&ahbdma_chan->idling)) {
>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>> +					      struct tegra_ahbdma_tx_desc,
>> +					      node);
>> +		if (tx) {
>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
>> +			reinit_completion(&ahbdma_chan->idling);
>> +		}
>> +	}
>> +
>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>> +}
>> +
>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>> +					      dma_cookie_t cookie,
>> +					      struct dma_tx_state *state)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	enum dma_status cookie_status;
>> +	unsigned long flags;
>> +	size_t residual;
>> +	u32 status;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>> +	if (cookie_status != DMA_COMPLETE) {
>> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
>> +			if (tx->desc.cookie == cookie)
>> +				goto found;
>> +		}
>> +	}
>> +
>> +	goto unlock;
>> +
>> +found:
>> +	if (tx->in_fly) {
>> +		status = readl_relaxed(
>> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>> +
>> +		residual = status;
>> +	} else
>> +		residual = tx->size;
>> +
>> +	dma_set_residue(state, residual);
>> +
>> +unlock:
>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>> +
>> +	return cookie_status;
>> +}
>> +
>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +	u32 csr;
>> +
>> +	spin_lock_irq(&ahbdma_chan->lock);
>> +
>> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
>> +
>> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>> +		list_del(entry);
>> +		kfree(tx);
>> +	}
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>> +		list_del(entry);
>> +		kfree(tx);
>> +	}
>> +
>> +	complete_all(&ahbdma_chan->idling);
>> +
>> +	spin_unlock_irq(&ahbdma_chan->lock);
>> +
>> +	return 0;
>> +}
>> +
>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>> +			       struct dma_slave_config *sconfig)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	enum dma_transfer_direction dir = sconfig->direction;
>> +	u32 burst, ahb_seq, ahb_addr;
>> +
>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +		return -EINVAL;
>> +
>> +	if (dir == DMA_DEV_TO_MEM) {
>> +		burst    = sconfig->src_maxburst;
>> +		ahb_addr = sconfig->src_addr;
>> +	} else {
>> +		burst    = sconfig->dst_maxburst;
>> +		ahb_addr = sconfig->dst_addr;
>> +	}
>> +
>> +	switch (burst) {
>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>> +
>> +	writel_relaxed(ahb_seq,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>> +
>> +	writel_relaxed(ahb_addr,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>> +
>> +	return 0;
>> +}
>> +
>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>> +{
>> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
>> +}
>> +
>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>> +		list_del(entry);
>> +		kfree(tx);
>> +	}
>> +}
>> +
>> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
>> +				      unsigned int chan_id)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
>> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
>> +	struct dma_device *dma_dev = &tdma->dma_dev;
>> +
>> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
>> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
>> +	init_completion(&ahbdma_chan->idling);
>> +	spin_lock_init(&ahbdma_chan->lock);
>> +	complete(&ahbdma_chan->idling);
>> +
>> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
>> +	ahbdma_chan->id = chan_id;
>> +
>> +	dma_cookie_init(dma_chan);
>> +	dma_chan->device = dma_dev;
>> +
>> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
>> +}
>> +
>> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
>> +					      struct of_dma *ofdma)
>> +{
>> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
>> +	struct dma_chan *chan;
>> +	u32 csr;
>> +
>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>> +	if (!chan)
>> +		return NULL;
>> +
>> +	/* enable channels flow control */
>> +	if (dma_spec->args_count == 1) {
> 
> The DT doc says #dma-cells should be '1' and so if not equal 1, is this
> not an error?
> 

I wanted to differentiate slave/master modes here. But if we'd want to add
TRIG_SEL as another cell, then it probably would worth to implement a custom DMA
configure options, like documentation suggests - to wrap generic
dma_slave_config into the custom one. On the other hand that probably would add
an unused functionality to the driver.

>> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
>> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;
> 
> What about the TRIG_REQ field?
> 

Not implemented, there is no test case for it yet.

>> +
>> +		writel_relaxed(csr,
>> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>> +	}
>> +	
>> +	return chan;
>> +}
>> +
>> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
>> +{
>> +	int err;
>> +
>> +	err = reset_control_assert(tdma->rst);
>> +	if (err) {
>> +		dev_err(dev, "Failed to assert reset: %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	err = clk_prepare_enable(tdma->clk);
>> +	if (err) {
>> +		dev_err(dev, "Failed to enable clock: %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	usleep_range(1000, 2000);
>> +
>> +	err = reset_control_deassert(tdma->rst);
>> +	if (err) {
>> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
>> +
>> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
>> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
>> +
>> +	return 0;
>> +}
> 
> Personally I would use the pm_runtime callbacks for this sort of thing
> and ...
> 

I decided that it probaby would be better to implement PM later if needed. I'm
not sure whether DMA controller consumes any substantial amounts of power while
idling. If it's not, why bother? Unnecessary power managment would just cause
CPU to waste its cycles (and power) doing PM.

>> +static int tegra_ahbdma_probe(struct platform_device *pdev)
>> +{
>> +	struct dma_device *dma_dev;
>> +	struct tegra_ahbdma *tdma;
>> +	struct resource *res_regs;
>> +	unsigned int i;
>> +	int irq;
>> +	int err;
>> +
>> +	tdma = devm_kzalloc(&pdev->dev, sizeof(*tdma), GFP_KERNEL);
>> +	if (!tdma)
>> +		return -ENOMEM;
>> +
>> +	irq = platform_get_irq(pdev, 0);
>> +	if (irq < 0) {
>> +		dev_err(&pdev->dev, "Failed to get IRQ\n");
>> +		return irq;
>> +	}
>> +
>> +	err = devm_request_irq(&pdev->dev, irq, tegra_ahbdma_isr, 0,
>> +			       dev_name(&pdev->dev), tdma);
>> +	if (err) {
>> +		dev_err(&pdev->dev, "Failed to request IRQ\n");
>> +		return -ENODEV;
>> +	}
>> +
>> +	res_regs = platform_get_resource(pdev, IORESOURCE_MEM, 0);
>> +	if (!res_regs)
>> +		return -ENODEV;
>> +
>> +	tdma->regs = devm_ioremap_resource(&pdev->dev, res_regs);
>> +	if (IS_ERR(tdma->regs))
>> +		return PTR_ERR(tdma->regs);
>> +
>> +	tdma->clk = devm_clk_get(&pdev->dev, NULL);
>> +	if (IS_ERR(tdma->clk)) {
>> +		dev_err(&pdev->dev, "Failed to get AHB-DMA clock\n");
>> +		return PTR_ERR(tdma->clk);
>> +	}
>> +
>> +	tdma->rst = devm_reset_control_get(&pdev->dev, NULL);
>> +	if (IS_ERR(tdma->rst)) {
>> +		dev_err(&pdev->dev, "Failed to get AHB-DMA reset\n");
>> +		return PTR_ERR(tdma->rst);
>> +	}
>> +
>> +	err = tegra_ahbdma_init_hw(tdma, &pdev->dev);
>> +	if (err)
>> +		return err;
> 
> ... here is looks like we turn the clocks on and leave them on. I would
> rather that we turn them on when the DMA channel is requested and turn
> them off again when freed. Again would be good to use pm_runtime APIs
> for this.
> 

Again not sure about it :)

>> +	dma_dev = &tdma->dma_dev;
>> +
>> +	INIT_LIST_HEAD(&dma_dev->channels);
>> +
>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>> +		tegra_ahbdma_init_channel(tdma, i);
>> +
>> +	dma_cap_set(DMA_PRIVATE, dma_dev->cap_mask);
>> +	dma_cap_set(DMA_CYCLIC, dma_dev->cap_mask);
>> +	dma_cap_set(DMA_SLAVE, dma_dev->cap_mask);
>> +
>> +	dma_dev->max_burst		= 8;
>> +	dma_dev->directions		= TEGRA_AHBDMA_DIRECTIONS;
>> +	dma_dev->src_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
>> +	dma_dev->dst_addr_widths	= TEGRA_AHBDMA_BUS_WIDTH;
>> +	dma_dev->descriptor_reuse	= true;
>> +	dma_dev->residue_granularity	= DMA_RESIDUE_GRANULARITY_BURST;
>> +	dma_dev->device_free_chan_resources = tegra_ahbdma_free_chan_resources;
>> +	dma_dev->device_prep_slave_sg	= tegra_ahbdma_prep_slave_sg;
>> +	dma_dev->device_prep_dma_cyclic	= tegra_ahbdma_prep_dma_cyclic;
>> +	dma_dev->device_terminate_all	= tegra_ahbdma_terminate_all;
>> +	dma_dev->device_issue_pending	= tegra_ahbdma_issue_pending;
>> +	dma_dev->device_tx_status	= tegra_ahbdma_tx_status;
>> +	dma_dev->device_config		= tegra_ahbdma_config;
>> +	dma_dev->device_synchronize	= tegra_ahbdma_synchronize;
>> +	dma_dev->dev			= &pdev->dev;
>> +
>> +	err = dma_async_device_register(dma_dev);
>> +	if (err) {
>> +		dev_err(&pdev->dev, "Device registration failed %d\n", err);
>> +		return err;
>> +	}
>> +
>> +	err = of_dma_controller_register(pdev->dev.of_node,
>> +					 tegra_ahbdma_of_xlate, tdma);
>> +	if (err) {
>> +		dev_err(&pdev->dev, "OF registration failed %d\n", err);
>> +		dma_async_device_unregister(dma_dev);
>> +		return err;
>> +	}
>> +
>> +	platform_set_drvdata(pdev, tdma);
>> +
>> +	return 0;
>> +}
>> +
>> +static int tegra_ahbdma_remove(struct platform_device *pdev)
>> +{
>> +	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
>> +
>> +	of_dma_controller_free(pdev->dev.of_node);
>> +	dma_async_device_unregister(&tdma->dma_dev);
>> +	clk_disable_unprepare(tdma->clk);
>> +
>> +	return 0;
>> +}
>> +
>> +static const struct of_device_id tegra_ahbdma_of_match[] = {
>> +	{ .compatible = "nvidia,tegra20-ahbdma" },
>> +	{ },
>> +};
>> +MODULE_DEVICE_TABLE(of, tegra_ahbdma_of_match);
>> +
>> +static struct platform_driver tegra_ahbdma_driver = {
>> +	.driver = {
>> +		.name	= "tegra-ahbdma",
>> +		.of_match_table = tegra_ahbdma_of_match,
> 
> It would be nice to have suspend/resume handler too. We could do a
> similar thing to the APB dma driver.
> 

It is not stricly necessary because LP0 isn't implemented by the core arch. I've
tested LP1 and it works fine that way. I'd prefer to implement suspend/resume
later, we can't really test it properly without LP0.

>> +	},
>> +	.probe	= tegra_ahbdma_probe,
>> +	.remove	= tegra_ahbdma_remove,
>> +};
>> +module_platform_driver(tegra_ahbdma_driver);
>> +
>> +MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
>> +MODULE_AUTHOR("Dmitry Osipenko <digetx@gmail.com>");
>> +MODULE_LICENSE("GPL");

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

Hi Dmitry,

On 26/09/17 17:06, Dmitry Osipenko wrote:
> Hi Jon,
> 
> On 26.09.2017 17:45, Jon Hunter wrote:
>> Hi Dmitry,
>>
>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
>>> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
>>> doesn't yet implement transfers larger than 64K and scatter-gather
>>> transfers that have NENT > 1, HW doesn't have native support for these
>>> cases.
>>>
>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>> ---
>>>  drivers/dma/Kconfig           |   9 +
>>>  drivers/dma/Makefile          |   1 +
>>>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>>>  3 files changed, 689 insertions(+)
>>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>
>> ...
>>
>>> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
>>> new file mode 100644
>>> index 000000000000..8316d64e35e1
>>> --- /dev/null
>>> +++ b/drivers/dma/tegra20-ahb-dma.c
>>> @@ -0,0 +1,679 @@
>>> +/*
>>> + * Copyright 2017 Dmitry Osipenko <digetx@gmail.com>
>>> + *
>>> + * This program is free software; you can redistribute it and/or modify it
>>> + * under the terms and conditions of the GNU General Public License,
>>> + * version 2, as published by the Free Software Foundation.
>>> + *
>>> + * This program is distributed in the hope it will be useful, but WITHOUT
>>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>>> + * more details.
>>> + *
>>> + * You should have received a copy of the GNU General Public License
>>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>>> + */
>>> +
>>> +#include <linux/clk.h>
>>> +#include <linux/delay.h>
>>> +#include <linux/interrupt.h>
>>> +#include <linux/io.h>
>>> +#include <linux/module.h>
>>> +#include <linux/of_device.h>
>>> +#include <linux/of_dma.h>
>>> +#include <linux/platform_device.h>
>>> +#include <linux/reset.h>
>>> +#include <linux/slab.h>
>>> +#include <linux/spinlock.h>
>>> +
>>> +#include "dmaengine.h"
>>> +
>>> +#define TEGRA_AHBDMA_CMD			0x0
>>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>>> +
>>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
>>> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
>>> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
>>> +
>>> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>>> +
>>> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
>>> +						BIT(DMA_MEM_TO_DEV)
>>> +
>>> +struct tegra_ahbdma_tx_desc {
>>> +	struct dma_async_tx_descriptor desc;
>>> +	struct tasklet_struct tasklet;
>>> +	struct list_head node;
>>
>> Any reason why we cannot use the virt-dma framework for this driver? I
>> would hope it would simplify the driver a bit.
>>
> 
> IIUC virt-dma is supposed to provide virtually unlimited number of channels.
> I've looked at it and decided that it would just add unnecessary functionality
> and, as a result, complexity. As I wrote in the cover-letter, it is supposed
> that this driver would have only one consumer - the host1x. It shouldn't be
> difficult to implement virt-dma later, if desired.  But again it is very
> unlikely that it would be needed.

I think that the biggest benefit is that is simplifies the linked list
management. See the tegra210-adma driver.

>>> +	enum dma_transfer_direction dir;
>>> +	dma_addr_t mem_paddr;
>>> +	unsigned long flags;
>>> +	size_t size;
>>> +	bool in_fly;
>>> +	bool cyclic;
>>> +};
>>> +
>>> +struct tegra_ahbdma_chan {
>>> +	struct dma_chan dma_chan;
>>> +	struct list_head active_list;
>>> +	struct list_head pending_list;
>>> +	struct completion idling;
>>> +	void __iomem *regs;
>>> +	spinlock_t lock;
>>> +	unsigned int id;
>>> +};
>>> +
>>> +struct tegra_ahbdma {
>>> +	struct tegra_ahbdma_chan channels[4];
>>> +	struct dma_device dma_dev;
>>> +	struct reset_control *rst;
>>> +	struct clk *clk;
>>> +	void __iomem *regs;
>>> +};
>>> +
>>> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
>>> +{
>>> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
>>> +}
>>> +
>>> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
>>> +{
>>> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
>>> +}
>>> +
>>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>>> +				struct dma_async_tx_descriptor *tx)
>>> +{
>>> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
>>> +}
>>> +
>>> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
>>> +				   struct tegra_ahbdma_tx_desc *tx)
>>> +{
>>> +	u32 csr;
>>> +
>>> +	writel_relaxed(tx->mem_paddr,
>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
>>> +
>>> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +
>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
>>> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
>>> +	csr |= tx->size - sizeof(u32);
>>> +
>>> +	if (tx->dir == DMA_DEV_TO_MEM)
>>> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>> +
>>> +	if (!tx->cyclic)
>>> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
>>> +
>>> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +
>>> +	tx->in_fly = true;
>>> +}
>>> +
>>> +static void tegra_ahbdma_tasklet(unsigned long data)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>> +
>>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>>> +
>>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>>> +		kfree(tx);
>>> +}
>>> +
>>> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
>>> +				      struct tegra_ahbdma_tx_desc *tx)
>>> +{
>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>> +	bool reuse = dmaengine_desc_test_reuse(desc);
>>> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
>>> +	bool completed = !tx->cyclic;
>>> +
>>> +	if (completed)
>>> +		dma_cookie_complete(desc);
>>> +
>>> +	if (interrupt)
>>> +		tasklet_schedule(&tx->tasklet);
>>> +
>>> +	if (completed) {
>>> +		list_del(&tx->node);
>>> +
>>> +		if (reuse)
>>> +			tx->in_fly = false;
>>> +
>>> +		if (!interrupt && !reuse)
>>> +			kfree(tx);
>>> +	}
>>> +
>>> +	return completed;
>>> +}
>>> +
>>> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +
>>> +	tx = list_first_entry_or_null(&chan->active_list,
>>> +				      struct tegra_ahbdma_tx_desc,
>>> +				      node);
>>> +	if (tx)
>>> +		tegra_ahbdma_submit_tx(chan, tx);
>>> +
>>> +	return !!tx;
>>> +}
>>> +
>>> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	unsigned long flags;
>>> +	u32 status;
>>> +
>>> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
>>> +		return;
>>> +
>>> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>> +
>>> +	spin_lock_irqsave(&chan->lock, flags);
>>> +
>>> +	if (!completion_done(&chan->idling)) {
>>> +		tx = list_first_entry(&chan->active_list,
>>> +				      struct tegra_ahbdma_tx_desc,
>>> +				      node);
>>> +
>>> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
>>> +		    !tegra_ahbdma_next_tx_issued(chan))
>>> +			complete_all(&chan->idling);
>>> +	}
>>> +
>>> +	spin_unlock_irqrestore(&chan->lock, flags);
>>> +}
>>> +
>>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>>> +{
>>> +	struct tegra_ahbdma *tdma = dev_id;
>>> +	unsigned int i;
>>> +
>>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>>> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
>>> +
>>> +	return IRQ_HANDLED;
>>> +}
>>> +
>>> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
>>> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
>>> +	dma_cookie_t cookie;
>>> +
>>> +	cookie = dma_cookie_assign(desc);
>>> +
>>> +	spin_lock_irq(&chan->lock);
>>> +	list_add_tail(&tx->node, &chan->pending_list);
>>> +	spin_unlock_irq(&chan->lock);
>>> +
>>> +	return cookie;
>>> +}
>>> +
>>> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
>>> +{
>>> +	kfree(to_ahbdma_tx_desc(desc));
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
>>> +					struct dma_chan *chan,
>>> +					struct scatterlist *sgl,
>>> +					unsigned int sg_len,
>>> +					enum dma_transfer_direction dir,
>>> +					unsigned long flags,
>>> +					void *context)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +
>>> +	/* unimplemented */
>>> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>>> +		return NULL;
>>> +
>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>> +	if (!tx)
>>> +		return NULL;
>>> +
>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>> +
>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
>>> +	tx->mem_paddr		= sg_dma_address(sgl);
>>> +	tx->size		= sg_dma_len(sgl);
>>> +	tx->flags		= flags;
>>> +	tx->dir			= dir;
>>> +
>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>> +
>>> +	return &tx->desc;
>>> +}
>>> +
>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>>> +					struct dma_chan *chan,
>>> +					dma_addr_t buf_addr,
>>> +					size_t buf_len,
>>> +					size_t period_len,
>>> +					enum dma_transfer_direction dir,
>>> +					unsigned long flags)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +
>>> +	/* unimplemented */
>>> +	if (buf_len != period_len || buf_len > SZ_64K)
>>> +		return NULL;
>>> +
>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>> +	if (!tx)
>>> +		return NULL;
>>> +
>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>> +
>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>> +	tx->mem_paddr		= buf_addr;
>>> +	tx->size		= buf_len;
>>> +	tx->flags		= flags;
>>> +	tx->cyclic		= true;
>>> +	tx->dir			= dir;
>>> +
>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>> +
>>> +	return &tx->desc;
>>> +}
>>> +
>>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	struct list_head *entry, *tmp;
>>> +	unsigned long flags;
>>> +
>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>>> +
>>> +	if (completion_done(&ahbdma_chan->idling)) {
>>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>>> +					      struct tegra_ahbdma_tx_desc,
>>> +					      node);
>>> +		if (tx) {
>>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
>>> +			reinit_completion(&ahbdma_chan->idling);
>>> +		}
>>> +	}
>>> +
>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>> +}
>>> +
>>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>>> +					      dma_cookie_t cookie,
>>> +					      struct dma_tx_state *state)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	enum dma_status cookie_status;
>>> +	unsigned long flags;
>>> +	size_t residual;
>>> +	u32 status;
>>> +
>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>> +
>>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>>> +	if (cookie_status != DMA_COMPLETE) {
>>> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
>>> +			if (tx->desc.cookie == cookie)
>>> +				goto found;
>>> +		}
>>> +	}
>>> +
>>> +	goto unlock;
>>> +
>>> +found:
>>> +	if (tx->in_fly) {
>>> +		status = readl_relaxed(
>>> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>> +
>>> +		residual = status;
>>> +	} else
>>> +		residual = tx->size;
>>> +
>>> +	dma_set_residue(state, residual);
>>> +
>>> +unlock:
>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>> +
>>> +	return cookie_status;
>>> +}
>>> +
>>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	struct list_head *entry, *tmp;
>>> +	u32 csr;
>>> +
>>> +	spin_lock_irq(&ahbdma_chan->lock);
>>> +
>>> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
>>> +
>>> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>> +		list_del(entry);
>>> +		kfree(tx);
>>> +	}
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>> +		list_del(entry);
>>> +		kfree(tx);
>>> +	}
>>> +
>>> +	complete_all(&ahbdma_chan->idling);
>>> +
>>> +	spin_unlock_irq(&ahbdma_chan->lock);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>>> +			       struct dma_slave_config *sconfig)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	enum dma_transfer_direction dir = sconfig->direction;
>>> +	u32 burst, ahb_seq, ahb_addr;
>>> +
>>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>> +		return -EINVAL;
>>> +
>>> +	if (dir == DMA_DEV_TO_MEM) {
>>> +		burst    = sconfig->src_maxburst;
>>> +		ahb_addr = sconfig->src_addr;
>>> +	} else {
>>> +		burst    = sconfig->dst_maxburst;
>>> +		ahb_addr = sconfig->dst_addr;
>>> +	}
>>> +
>>> +	switch (burst) {
>>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
>>> +	default:
>>> +		return -EINVAL;
>>> +	}
>>> +
>>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>>> +
>>> +	writel_relaxed(ahb_seq,
>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>>> +
>>> +	writel_relaxed(ahb_addr,
>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>> +{
>>> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
>>> +}
>>> +
>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	struct list_head *entry, *tmp;
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>> +		list_del(entry);
>>> +		kfree(tx);
>>> +	}
>>> +}
>>> +
>>> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
>>> +				      unsigned int chan_id)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
>>> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
>>> +	struct dma_device *dma_dev = &tdma->dma_dev;
>>> +
>>> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
>>> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
>>> +	init_completion(&ahbdma_chan->idling);
>>> +	spin_lock_init(&ahbdma_chan->lock);
>>> +	complete(&ahbdma_chan->idling);
>>> +
>>> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
>>> +	ahbdma_chan->id = chan_id;
>>> +
>>> +	dma_cookie_init(dma_chan);
>>> +	dma_chan->device = dma_dev;
>>> +
>>> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
>>> +}
>>> +
>>> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
>>> +					      struct of_dma *ofdma)
>>> +{
>>> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
>>> +	struct dma_chan *chan;
>>> +	u32 csr;
>>> +
>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>> +	if (!chan)
>>> +		return NULL;
>>> +
>>> +	/* enable channels flow control */
>>> +	if (dma_spec->args_count == 1) {
>>
>> The DT doc says #dma-cells should be '1' and so if not equal 1, is this
>> not an error?
>>
> 
> I wanted to differentiate slave/master modes here. But if we'd want to add
> TRIG_SEL as another cell, then it probably would worth to implement a custom DMA
> configure options, like documentation suggests - to wrap generic
> dma_slave_config into the custom one. On the other hand that probably would add
> an unused functionality to the driver.
> 
>>> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
>>> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;
>>
>> What about the TRIG_REQ field?
>>
> 
> Not implemented, there is no test case for it yet.
> 
>>> +
>>> +		writel_relaxed(csr,
>>> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +	}
>>> +	
>>> +	return chan;
>>> +}
>>> +
>>> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
>>> +{
>>> +	int err;
>>> +
>>> +	err = reset_control_assert(tdma->rst);
>>> +	if (err) {
>>> +		dev_err(dev, "Failed to assert reset: %d\n", err);
>>> +		return err;
>>> +	}
>>> +
>>> +	err = clk_prepare_enable(tdma->clk);
>>> +	if (err) {
>>> +		dev_err(dev, "Failed to enable clock: %d\n", err);
>>> +		return err;
>>> +	}
>>> +
>>> +	usleep_range(1000, 2000);
>>> +
>>> +	err = reset_control_deassert(tdma->rst);
>>> +	if (err) {
>>> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
>>> +		return err;
>>> +	}
>>> +
>>> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
>>> +
>>> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
>>> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
>>> +
>>> +	return 0;
>>> +}
>>
>> Personally I would use the pm_runtime callbacks for this sort of thing
>> and ...
>>
> 
> I decided that it probaby would be better to implement PM later if needed. I'm
> not sure whether DMA controller consumes any substantial amounts of power while
> idling. If it's not, why bother? Unnecessary power managment would just cause
> CPU to waste its cycles (and power) doing PM.

Yes it probably does not but it is easy to do and so even though there
are probably a ton of other clocks left running, I still think it is
good practice.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

Hi Dmitry,

On 26/09/17 17:06, Dmitry Osipenko wrote:
> Hi Jon,
> 
> On 26.09.2017 17:45, Jon Hunter wrote:
>> Hi Dmitry,
>>
>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
>>> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
>>> doesn't yet implement transfers larger than 64K and scatter-gather
>>> transfers that have NENT > 1, HW doesn't have native support for these
>>> cases.
>>>
>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>> ---
>>>  drivers/dma/Kconfig           |   9 +
>>>  drivers/dma/Makefile          |   1 +
>>>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>>>  3 files changed, 689 insertions(+)
>>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>
>> ...
>>
>>> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
>>> new file mode 100644
>>> index 000000000000..8316d64e35e1
>>> --- /dev/null
>>> +++ b/drivers/dma/tegra20-ahb-dma.c
>>> @@ -0,0 +1,679 @@
>>> +/*
>>> + * Copyright 2017 Dmitry Osipenko <digetx@gmail.com>
>>> + *
>>> + * This program is free software; you can redistribute it and/or modify it
>>> + * under the terms and conditions of the GNU General Public License,
>>> + * version 2, as published by the Free Software Foundation.
>>> + *
>>> + * This program is distributed in the hope it will be useful, but WITHOUT
>>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>>> + * more details.
>>> + *
>>> + * You should have received a copy of the GNU General Public License
>>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>>> + */
>>> +
>>> +#include <linux/clk.h>
>>> +#include <linux/delay.h>
>>> +#include <linux/interrupt.h>
>>> +#include <linux/io.h>
>>> +#include <linux/module.h>
>>> +#include <linux/of_device.h>
>>> +#include <linux/of_dma.h>
>>> +#include <linux/platform_device.h>
>>> +#include <linux/reset.h>
>>> +#include <linux/slab.h>
>>> +#include <linux/spinlock.h>
>>> +
>>> +#include "dmaengine.h"
>>> +
>>> +#define TEGRA_AHBDMA_CMD			0x0
>>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>>> +
>>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
>>> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
>>> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
>>> +
>>> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
>>> +
>>> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>>> +
>>> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
>>> +						BIT(DMA_MEM_TO_DEV)
>>> +
>>> +struct tegra_ahbdma_tx_desc {
>>> +	struct dma_async_tx_descriptor desc;
>>> +	struct tasklet_struct tasklet;
>>> +	struct list_head node;
>>
>> Any reason why we cannot use the virt-dma framework for this driver? I
>> would hope it would simplify the driver a bit.
>>
> 
> IIUC virt-dma is supposed to provide virtually unlimited number of channels.
> I've looked at it and decided that it would just add unnecessary functionality
> and, as a result, complexity. As I wrote in the cover-letter, it is supposed
> that this driver would have only one consumer - the host1x. It shouldn't be
> difficult to implement virt-dma later, if desired.  But again it is very
> unlikely that it would be needed.

I think that the biggest benefit is that is simplifies the linked list
management. See the tegra210-adma driver.

>>> +	enum dma_transfer_direction dir;
>>> +	dma_addr_t mem_paddr;
>>> +	unsigned long flags;
>>> +	size_t size;
>>> +	bool in_fly;
>>> +	bool cyclic;
>>> +};
>>> +
>>> +struct tegra_ahbdma_chan {
>>> +	struct dma_chan dma_chan;
>>> +	struct list_head active_list;
>>> +	struct list_head pending_list;
>>> +	struct completion idling;
>>> +	void __iomem *regs;
>>> +	spinlock_t lock;
>>> +	unsigned int id;
>>> +};
>>> +
>>> +struct tegra_ahbdma {
>>> +	struct tegra_ahbdma_chan channels[4];
>>> +	struct dma_device dma_dev;
>>> +	struct reset_control *rst;
>>> +	struct clk *clk;
>>> +	void __iomem *regs;
>>> +};
>>> +
>>> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
>>> +{
>>> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
>>> +}
>>> +
>>> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
>>> +{
>>> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
>>> +}
>>> +
>>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>>> +				struct dma_async_tx_descriptor *tx)
>>> +{
>>> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
>>> +}
>>> +
>>> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
>>> +				   struct tegra_ahbdma_tx_desc *tx)
>>> +{
>>> +	u32 csr;
>>> +
>>> +	writel_relaxed(tx->mem_paddr,
>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
>>> +
>>> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +
>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
>>> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
>>> +	csr |= tx->size - sizeof(u32);
>>> +
>>> +	if (tx->dir == DMA_DEV_TO_MEM)
>>> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>> +
>>> +	if (!tx->cyclic)
>>> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
>>> +
>>> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +
>>> +	tx->in_fly = true;
>>> +}
>>> +
>>> +static void tegra_ahbdma_tasklet(unsigned long data)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>> +
>>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>>> +
>>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>>> +		kfree(tx);
>>> +}
>>> +
>>> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
>>> +				      struct tegra_ahbdma_tx_desc *tx)
>>> +{
>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>> +	bool reuse = dmaengine_desc_test_reuse(desc);
>>> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
>>> +	bool completed = !tx->cyclic;
>>> +
>>> +	if (completed)
>>> +		dma_cookie_complete(desc);
>>> +
>>> +	if (interrupt)
>>> +		tasklet_schedule(&tx->tasklet);
>>> +
>>> +	if (completed) {
>>> +		list_del(&tx->node);
>>> +
>>> +		if (reuse)
>>> +			tx->in_fly = false;
>>> +
>>> +		if (!interrupt && !reuse)
>>> +			kfree(tx);
>>> +	}
>>> +
>>> +	return completed;
>>> +}
>>> +
>>> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +
>>> +	tx = list_first_entry_or_null(&chan->active_list,
>>> +				      struct tegra_ahbdma_tx_desc,
>>> +				      node);
>>> +	if (tx)
>>> +		tegra_ahbdma_submit_tx(chan, tx);
>>> +
>>> +	return !!tx;
>>> +}
>>> +
>>> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	unsigned long flags;
>>> +	u32 status;
>>> +
>>> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
>>> +		return;
>>> +
>>> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>> +
>>> +	spin_lock_irqsave(&chan->lock, flags);
>>> +
>>> +	if (!completion_done(&chan->idling)) {
>>> +		tx = list_first_entry(&chan->active_list,
>>> +				      struct tegra_ahbdma_tx_desc,
>>> +				      node);
>>> +
>>> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
>>> +		    !tegra_ahbdma_next_tx_issued(chan))
>>> +			complete_all(&chan->idling);
>>> +	}
>>> +
>>> +	spin_unlock_irqrestore(&chan->lock, flags);
>>> +}
>>> +
>>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>>> +{
>>> +	struct tegra_ahbdma *tdma = dev_id;
>>> +	unsigned int i;
>>> +
>>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>>> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
>>> +
>>> +	return IRQ_HANDLED;
>>> +}
>>> +
>>> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
>>> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
>>> +	dma_cookie_t cookie;
>>> +
>>> +	cookie = dma_cookie_assign(desc);
>>> +
>>> +	spin_lock_irq(&chan->lock);
>>> +	list_add_tail(&tx->node, &chan->pending_list);
>>> +	spin_unlock_irq(&chan->lock);
>>> +
>>> +	return cookie;
>>> +}
>>> +
>>> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
>>> +{
>>> +	kfree(to_ahbdma_tx_desc(desc));
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
>>> +					struct dma_chan *chan,
>>> +					struct scatterlist *sgl,
>>> +					unsigned int sg_len,
>>> +					enum dma_transfer_direction dir,
>>> +					unsigned long flags,
>>> +					void *context)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +
>>> +	/* unimplemented */
>>> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>>> +		return NULL;
>>> +
>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>> +	if (!tx)
>>> +		return NULL;
>>> +
>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>> +
>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
>>> +	tx->mem_paddr		= sg_dma_address(sgl);
>>> +	tx->size		= sg_dma_len(sgl);
>>> +	tx->flags		= flags;
>>> +	tx->dir			= dir;
>>> +
>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>> +
>>> +	return &tx->desc;
>>> +}
>>> +
>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>>> +					struct dma_chan *chan,
>>> +					dma_addr_t buf_addr,
>>> +					size_t buf_len,
>>> +					size_t period_len,
>>> +					enum dma_transfer_direction dir,
>>> +					unsigned long flags)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +
>>> +	/* unimplemented */
>>> +	if (buf_len != period_len || buf_len > SZ_64K)
>>> +		return NULL;
>>> +
>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>> +	if (!tx)
>>> +		return NULL;
>>> +
>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>> +
>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>> +	tx->mem_paddr		= buf_addr;
>>> +	tx->size		= buf_len;
>>> +	tx->flags		= flags;
>>> +	tx->cyclic		= true;
>>> +	tx->dir			= dir;
>>> +
>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>> +
>>> +	return &tx->desc;
>>> +}
>>> +
>>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	struct list_head *entry, *tmp;
>>> +	unsigned long flags;
>>> +
>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>>> +
>>> +	if (completion_done(&ahbdma_chan->idling)) {
>>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>>> +					      struct tegra_ahbdma_tx_desc,
>>> +					      node);
>>> +		if (tx) {
>>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
>>> +			reinit_completion(&ahbdma_chan->idling);
>>> +		}
>>> +	}
>>> +
>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>> +}
>>> +
>>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>>> +					      dma_cookie_t cookie,
>>> +					      struct dma_tx_state *state)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	enum dma_status cookie_status;
>>> +	unsigned long flags;
>>> +	size_t residual;
>>> +	u32 status;
>>> +
>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>> +
>>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>>> +	if (cookie_status != DMA_COMPLETE) {
>>> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
>>> +			if (tx->desc.cookie == cookie)
>>> +				goto found;
>>> +		}
>>> +	}
>>> +
>>> +	goto unlock;
>>> +
>>> +found:
>>> +	if (tx->in_fly) {
>>> +		status = readl_relaxed(
>>> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>> +
>>> +		residual = status;
>>> +	} else
>>> +		residual = tx->size;
>>> +
>>> +	dma_set_residue(state, residual);
>>> +
>>> +unlock:
>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>> +
>>> +	return cookie_status;
>>> +}
>>> +
>>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	struct list_head *entry, *tmp;
>>> +	u32 csr;
>>> +
>>> +	spin_lock_irq(&ahbdma_chan->lock);
>>> +
>>> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
>>> +
>>> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>> +		list_del(entry);
>>> +		kfree(tx);
>>> +	}
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>> +		list_del(entry);
>>> +		kfree(tx);
>>> +	}
>>> +
>>> +	complete_all(&ahbdma_chan->idling);
>>> +
>>> +	spin_unlock_irq(&ahbdma_chan->lock);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>>> +			       struct dma_slave_config *sconfig)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	enum dma_transfer_direction dir = sconfig->direction;
>>> +	u32 burst, ahb_seq, ahb_addr;
>>> +
>>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>> +		return -EINVAL;
>>> +
>>> +	if (dir == DMA_DEV_TO_MEM) {
>>> +		burst    = sconfig->src_maxburst;
>>> +		ahb_addr = sconfig->src_addr;
>>> +	} else {
>>> +		burst    = sconfig->dst_maxburst;
>>> +		ahb_addr = sconfig->dst_addr;
>>> +	}
>>> +
>>> +	switch (burst) {
>>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
>>> +	default:
>>> +		return -EINVAL;
>>> +	}
>>> +
>>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>>> +
>>> +	writel_relaxed(ahb_seq,
>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>>> +
>>> +	writel_relaxed(ahb_addr,
>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>> +{
>>> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
>>> +}
>>> +
>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	struct tegra_ahbdma_tx_desc *tx;
>>> +	struct list_head *entry, *tmp;
>>> +
>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>> +		list_del(entry);
>>> +		kfree(tx);
>>> +	}
>>> +}
>>> +
>>> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
>>> +				      unsigned int chan_id)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
>>> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
>>> +	struct dma_device *dma_dev = &tdma->dma_dev;
>>> +
>>> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
>>> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
>>> +	init_completion(&ahbdma_chan->idling);
>>> +	spin_lock_init(&ahbdma_chan->lock);
>>> +	complete(&ahbdma_chan->idling);
>>> +
>>> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
>>> +	ahbdma_chan->id = chan_id;
>>> +
>>> +	dma_cookie_init(dma_chan);
>>> +	dma_chan->device = dma_dev;
>>> +
>>> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
>>> +}
>>> +
>>> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
>>> +					      struct of_dma *ofdma)
>>> +{
>>> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
>>> +	struct dma_chan *chan;
>>> +	u32 csr;
>>> +
>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>> +	if (!chan)
>>> +		return NULL;
>>> +
>>> +	/* enable channels flow control */
>>> +	if (dma_spec->args_count == 1) {
>>
>> The DT doc says #dma-cells should be '1' and so if not equal 1, is this
>> not an error?
>>
> 
> I wanted to differentiate slave/master modes here. But if we'd want to add
> TRIG_SEL as another cell, then it probably would worth to implement a custom DMA
> configure options, like documentation suggests - to wrap generic
> dma_slave_config into the custom one. On the other hand that probably would add
> an unused functionality to the driver.
> 
>>> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
>>> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;
>>
>> What about the TRIG_REQ field?
>>
> 
> Not implemented, there is no test case for it yet.
> 
>>> +
>>> +		writel_relaxed(csr,
>>> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>> +	}
>>> +	
>>> +	return chan;
>>> +}
>>> +
>>> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
>>> +{
>>> +	int err;
>>> +
>>> +	err = reset_control_assert(tdma->rst);
>>> +	if (err) {
>>> +		dev_err(dev, "Failed to assert reset: %d\n", err);
>>> +		return err;
>>> +	}
>>> +
>>> +	err = clk_prepare_enable(tdma->clk);
>>> +	if (err) {
>>> +		dev_err(dev, "Failed to enable clock: %d\n", err);
>>> +		return err;
>>> +	}
>>> +
>>> +	usleep_range(1000, 2000);
>>> +
>>> +	err = reset_control_deassert(tdma->rst);
>>> +	if (err) {
>>> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
>>> +		return err;
>>> +	}
>>> +
>>> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
>>> +
>>> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
>>> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
>>> +
>>> +	return 0;
>>> +}
>>
>> Personally I would use the pm_runtime callbacks for this sort of thing
>> and ...
>>
> 
> I decided that it probaby would be better to implement PM later if needed. I'm
> not sure whether DMA controller consumes any substantial amounts of power while
> idling. If it's not, why bother? Unnecessary power managment would just cause
> CPU to waste its cycles (and power) doing PM.

Yes it probably does not but it is easy to do and so even though there
are probably a ton of other clocks left running, I still think it is
good practice.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 27.09.2017 00:37, Jon Hunter wrote:
> Hi Dmitry,
> 
> On 26/09/17 17:06, Dmitry Osipenko wrote:
>> Hi Jon,
>>
>> On 26.09.2017 17:45, Jon Hunter wrote:
>>> Hi Dmitry,
>>>
>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
>>>> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
>>>> doesn't yet implement transfers larger than 64K and scatter-gather
>>>> transfers that have NENT > 1, HW doesn't have native support for these
>>>> cases.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>> ---
>>>>  drivers/dma/Kconfig           |   9 +
>>>>  drivers/dma/Makefile          |   1 +
>>>>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>>>>  3 files changed, 689 insertions(+)
>>>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>>
>>> ...
>>>
>>>> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
>>>> new file mode 100644
>>>> index 000000000000..8316d64e35e1
>>>> --- /dev/null
>>>> +++ b/drivers/dma/tegra20-ahb-dma.c
>>>> @@ -0,0 +1,679 @@
>>>> +/*
>>>> + * Copyright 2017 Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>> + *
>>>> + * This program is free software; you can redistribute it and/or modify it
>>>> + * under the terms and conditions of the GNU General Public License,
>>>> + * version 2, as published by the Free Software Foundation.
>>>> + *
>>>> + * This program is distributed in the hope it will be useful, but WITHOUT
>>>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>>>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>>>> + * more details.
>>>> + *
>>>> + * You should have received a copy of the GNU General Public License
>>>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>>>> + */
>>>> +
>>>> +#include <linux/clk.h>
>>>> +#include <linux/delay.h>
>>>> +#include <linux/interrupt.h>
>>>> +#include <linux/io.h>
>>>> +#include <linux/module.h>
>>>> +#include <linux/of_device.h>
>>>> +#include <linux/of_dma.h>
>>>> +#include <linux/platform_device.h>
>>>> +#include <linux/reset.h>
>>>> +#include <linux/slab.h>
>>>> +#include <linux/spinlock.h>
>>>> +
>>>> +#include "dmaengine.h"
>>>> +
>>>> +#define TEGRA_AHBDMA_CMD			0x0
>>>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>>>> +
>>>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>>>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>>>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>>>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>>>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>>>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>>>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>>>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>>>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>>>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
>>>> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
>>>> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
>>>> +
>>>> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>>>> +
>>>> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
>>>> +						BIT(DMA_MEM_TO_DEV)
>>>> +
>>>> +struct tegra_ahbdma_tx_desc {
>>>> +	struct dma_async_tx_descriptor desc;
>>>> +	struct tasklet_struct tasklet;
>>>> +	struct list_head node;
>>>
>>> Any reason why we cannot use the virt-dma framework for this driver? I
>>> would hope it would simplify the driver a bit.
>>>
>>
>> IIUC virt-dma is supposed to provide virtually unlimited number of channels.
>> I've looked at it and decided that it would just add unnecessary functionality
>> and, as a result, complexity. As I wrote in the cover-letter, it is supposed
>> that this driver would have only one consumer - the host1x. It shouldn't be
>> difficult to implement virt-dma later, if desired.  But again it is very
>> unlikely that it would be needed.
> 
> I think that the biggest benefit is that is simplifies the linked list
> management. See the tegra210-adma driver.
> 

I'll take a more thorough look at it. Thank you for suggestion.

>>>> +	enum dma_transfer_direction dir;
>>>> +	dma_addr_t mem_paddr;
>>>> +	unsigned long flags;
>>>> +	size_t size;
>>>> +	bool in_fly;
>>>> +	bool cyclic;
>>>> +};
>>>> +
>>>> +struct tegra_ahbdma_chan {
>>>> +	struct dma_chan dma_chan;
>>>> +	struct list_head active_list;
>>>> +	struct list_head pending_list;
>>>> +	struct completion idling;
>>>> +	void __iomem *regs;
>>>> +	spinlock_t lock;
>>>> +	unsigned int id;
>>>> +};
>>>> +
>>>> +struct tegra_ahbdma {
>>>> +	struct tegra_ahbdma_chan channels[4];
>>>> +	struct dma_device dma_dev;
>>>> +	struct reset_control *rst;
>>>> +	struct clk *clk;
>>>> +	void __iomem *regs;
>>>> +};
>>>> +
>>>> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
>>>> +{
>>>> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
>>>> +}
>>>> +
>>>> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
>>>> +{
>>>> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
>>>> +}
>>>> +
>>>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>>>> +				struct dma_async_tx_descriptor *tx)
>>>> +{
>>>> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
>>>> +				   struct tegra_ahbdma_tx_desc *tx)
>>>> +{
>>>> +	u32 csr;
>>>> +
>>>> +	writel_relaxed(tx->mem_paddr,
>>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
>>>> +
>>>> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +
>>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>>> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
>>>> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
>>>> +	csr |= tx->size - sizeof(u32);
>>>> +
>>>> +	if (tx->dir == DMA_DEV_TO_MEM)
>>>> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>>> +
>>>> +	if (!tx->cyclic)
>>>> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
>>>> +
>>>> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +
>>>> +	tx->in_fly = true;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_tasklet(unsigned long data)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>>> +
>>>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>>>> +
>>>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>>>> +		kfree(tx);
>>>> +}
>>>> +
>>>> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
>>>> +				      struct tegra_ahbdma_tx_desc *tx)
>>>> +{
>>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>>> +	bool reuse = dmaengine_desc_test_reuse(desc);
>>>> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
>>>> +	bool completed = !tx->cyclic;
>>>> +
>>>> +	if (completed)
>>>> +		dma_cookie_complete(desc);
>>>> +
>>>> +	if (interrupt)
>>>> +		tasklet_schedule(&tx->tasklet);
>>>> +
>>>> +	if (completed) {
>>>> +		list_del(&tx->node);
>>>> +
>>>> +		if (reuse)
>>>> +			tx->in_fly = false;
>>>> +
>>>> +		if (!interrupt && !reuse)
>>>> +			kfree(tx);
>>>> +	}
>>>> +
>>>> +	return completed;
>>>> +}
>>>> +
>>>> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +
>>>> +	tx = list_first_entry_or_null(&chan->active_list,
>>>> +				      struct tegra_ahbdma_tx_desc,
>>>> +				      node);
>>>> +	if (tx)
>>>> +		tegra_ahbdma_submit_tx(chan, tx);
>>>> +
>>>> +	return !!tx;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	unsigned long flags;
>>>> +	u32 status;
>>>> +
>>>> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>>> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
>>>> +		return;
>>>> +
>>>> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
>>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>>> +
>>>> +	spin_lock_irqsave(&chan->lock, flags);
>>>> +
>>>> +	if (!completion_done(&chan->idling)) {
>>>> +		tx = list_first_entry(&chan->active_list,
>>>> +				      struct tegra_ahbdma_tx_desc,
>>>> +				      node);
>>>> +
>>>> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
>>>> +		    !tegra_ahbdma_next_tx_issued(chan))
>>>> +			complete_all(&chan->idling);
>>>> +	}
>>>> +
>>>> +	spin_unlock_irqrestore(&chan->lock, flags);
>>>> +}
>>>> +
>>>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>>>> +{
>>>> +	struct tegra_ahbdma *tdma = dev_id;
>>>> +	unsigned int i;
>>>> +
>>>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>>>> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
>>>> +
>>>> +	return IRQ_HANDLED;
>>>> +}
>>>> +
>>>> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
>>>> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
>>>> +	dma_cookie_t cookie;
>>>> +
>>>> +	cookie = dma_cookie_assign(desc);
>>>> +
>>>> +	spin_lock_irq(&chan->lock);
>>>> +	list_add_tail(&tx->node, &chan->pending_list);
>>>> +	spin_unlock_irq(&chan->lock);
>>>> +
>>>> +	return cookie;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
>>>> +{
>>>> +	kfree(to_ahbdma_tx_desc(desc));
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
>>>> +					struct dma_chan *chan,
>>>> +					struct scatterlist *sgl,
>>>> +					unsigned int sg_len,
>>>> +					enum dma_transfer_direction dir,
>>>> +					unsigned long flags,
>>>> +					void *context)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +
>>>> +	/* unimplemented */
>>>> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>>>> +		return NULL;
>>>> +
>>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>>> +	if (!tx)
>>>> +		return NULL;
>>>> +
>>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>>> +
>>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>>> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
>>>> +	tx->mem_paddr		= sg_dma_address(sgl);
>>>> +	tx->size		= sg_dma_len(sgl);
>>>> +	tx->flags		= flags;
>>>> +	tx->dir			= dir;
>>>> +
>>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>>> +
>>>> +	return &tx->desc;
>>>> +}
>>>> +
>>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>>>> +					struct dma_chan *chan,
>>>> +					dma_addr_t buf_addr,
>>>> +					size_t buf_len,
>>>> +					size_t period_len,
>>>> +					enum dma_transfer_direction dir,
>>>> +					unsigned long flags)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +
>>>> +	/* unimplemented */
>>>> +	if (buf_len != period_len || buf_len > SZ_64K)
>>>> +		return NULL;
>>>> +
>>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>>> +	if (!tx)
>>>> +		return NULL;
>>>> +
>>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>>> +
>>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>>> +	tx->mem_paddr		= buf_addr;
>>>> +	tx->size		= buf_len;
>>>> +	tx->flags		= flags;
>>>> +	tx->cyclic		= true;
>>>> +	tx->dir			= dir;
>>>> +
>>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>>> +
>>>> +	return &tx->desc;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	struct list_head *entry, *tmp;
>>>> +	unsigned long flags;
>>>> +
>>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>>>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>>>> +
>>>> +	if (completion_done(&ahbdma_chan->idling)) {
>>>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>>>> +					      struct tegra_ahbdma_tx_desc,
>>>> +					      node);
>>>> +		if (tx) {
>>>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
>>>> +			reinit_completion(&ahbdma_chan->idling);
>>>> +		}
>>>> +	}
>>>> +
>>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>>> +}
>>>> +
>>>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>>>> +					      dma_cookie_t cookie,
>>>> +					      struct dma_tx_state *state)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	enum dma_status cookie_status;
>>>> +	unsigned long flags;
>>>> +	size_t residual;
>>>> +	u32 status;
>>>> +
>>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>>> +
>>>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>>>> +	if (cookie_status != DMA_COMPLETE) {
>>>> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
>>>> +			if (tx->desc.cookie == cookie)
>>>> +				goto found;
>>>> +		}
>>>> +	}
>>>> +
>>>> +	goto unlock;
>>>> +
>>>> +found:
>>>> +	if (tx->in_fly) {
>>>> +		status = readl_relaxed(
>>>> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>>> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>>> +
>>>> +		residual = status;
>>>> +	} else
>>>> +		residual = tx->size;
>>>> +
>>>> +	dma_set_residue(state, residual);
>>>> +
>>>> +unlock:
>>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>>> +
>>>> +	return cookie_status;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	struct list_head *entry, *tmp;
>>>> +	u32 csr;
>>>> +
>>>> +	spin_lock_irq(&ahbdma_chan->lock);
>>>> +
>>>> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
>>>> +
>>>> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
>>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>>> +		list_del(entry);
>>>> +		kfree(tx);
>>>> +	}
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>>> +		list_del(entry);
>>>> +		kfree(tx);
>>>> +	}
>>>> +
>>>> +	complete_all(&ahbdma_chan->idling);
>>>> +
>>>> +	spin_unlock_irq(&ahbdma_chan->lock);
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>>>> +			       struct dma_slave_config *sconfig)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	enum dma_transfer_direction dir = sconfig->direction;
>>>> +	u32 burst, ahb_seq, ahb_addr;
>>>> +
>>>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>>> +		return -EINVAL;
>>>> +
>>>> +	if (dir == DMA_DEV_TO_MEM) {
>>>> +		burst    = sconfig->src_maxburst;
>>>> +		ahb_addr = sconfig->src_addr;
>>>> +	} else {
>>>> +		burst    = sconfig->dst_maxburst;
>>>> +		ahb_addr = sconfig->dst_addr;
>>>> +	}
>>>> +
>>>> +	switch (burst) {
>>>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>>>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>>>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
>>>> +	default:
>>>> +		return -EINVAL;
>>>> +	}
>>>> +
>>>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>>>> +
>>>> +	writel_relaxed(ahb_seq,
>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>>>> +
>>>> +	writel_relaxed(ahb_addr,
>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>>> +{
>>>> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	struct list_head *entry, *tmp;
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>>> +		list_del(entry);
>>>> +		kfree(tx);
>>>> +	}
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
>>>> +				      unsigned int chan_id)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
>>>> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
>>>> +	struct dma_device *dma_dev = &tdma->dma_dev;
>>>> +
>>>> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
>>>> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
>>>> +	init_completion(&ahbdma_chan->idling);
>>>> +	spin_lock_init(&ahbdma_chan->lock);
>>>> +	complete(&ahbdma_chan->idling);
>>>> +
>>>> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
>>>> +	ahbdma_chan->id = chan_id;
>>>> +
>>>> +	dma_cookie_init(dma_chan);
>>>> +	dma_chan->device = dma_dev;
>>>> +
>>>> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
>>>> +}
>>>> +
>>>> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
>>>> +					      struct of_dma *ofdma)
>>>> +{
>>>> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
>>>> +	struct dma_chan *chan;
>>>> +	u32 csr;
>>>> +
>>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>>> +	if (!chan)
>>>> +		return NULL;
>>>> +
>>>> +	/* enable channels flow control */
>>>> +	if (dma_spec->args_count == 1) {
>>>
>>> The DT doc says #dma-cells should be '1' and so if not equal 1, is this
>>> not an error?
>>>
>>
>> I wanted to differentiate slave/master modes here. But if we'd want to add
>> TRIG_SEL as another cell, then it probably would worth to implement a custom DMA
>> configure options, like documentation suggests - to wrap generic
>> dma_slave_config into the custom one. On the other hand that probably would add
>> an unused functionality to the driver.
>>
>>>> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
>>>> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;
>>>
>>> What about the TRIG_REQ field?
>>>
>>
>> Not implemented, there is no test case for it yet.
>>
>>>> +
>>>> +		writel_relaxed(csr,
>>>> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +	}
>>>> +	
>>>> +	return chan;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
>>>> +{
>>>> +	int err;
>>>> +
>>>> +	err = reset_control_assert(tdma->rst);
>>>> +	if (err) {
>>>> +		dev_err(dev, "Failed to assert reset: %d\n", err);
>>>> +		return err;
>>>> +	}
>>>> +
>>>> +	err = clk_prepare_enable(tdma->clk);
>>>> +	if (err) {
>>>> +		dev_err(dev, "Failed to enable clock: %d\n", err);
>>>> +		return err;
>>>> +	}
>>>> +
>>>> +	usleep_range(1000, 2000);
>>>> +
>>>> +	err = reset_control_deassert(tdma->rst);
>>>> +	if (err) {
>>>> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
>>>> +		return err;
>>>> +	}
>>>> +
>>>> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
>>>> +
>>>> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
>>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
>>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
>>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
>>>> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
>>>> +
>>>> +	return 0;
>>>> +}
>>>
>>> Personally I would use the pm_runtime callbacks for this sort of thing
>>> and ...
>>>
>>
>> I decided that it probaby would be better to implement PM later if needed. I'm
>> not sure whether DMA controller consumes any substantial amounts of power while
>> idling. If it's not, why bother? Unnecessary power managment would just cause
>> CPU to waste its cycles (and power) doing PM.
> 
> Yes it probably does not but it is easy to do and so even though there
> are probably a ton of other clocks left running, I still think it is
> good practice.
> 

Okay, I'll take a look into implementing PM. Disabling AHBDMA clock won't stop
the actual clock, but only gate it to the controller.

Thank you for the comments!

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 27.09.2017 00:37, Jon Hunter wrote:
> Hi Dmitry,
> 
> On 26/09/17 17:06, Dmitry Osipenko wrote:
>> Hi Jon,
>>
>> On 26.09.2017 17:45, Jon Hunter wrote:
>>> Hi Dmitry,
>>>
>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>> AHB DMA controller presents on Tegra20/30 SoC's, it supports transfers
>>>> memory <-> AHB bus peripherals as well as mem-to-mem transfers. Driver
>>>> doesn't yet implement transfers larger than 64K and scatter-gather
>>>> transfers that have NENT > 1, HW doesn't have native support for these
>>>> cases.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>> ---
>>>>  drivers/dma/Kconfig           |   9 +
>>>>  drivers/dma/Makefile          |   1 +
>>>>  drivers/dma/tegra20-ahb-dma.c | 679 ++++++++++++++++++++++++++++++++++++++++++
>>>>  3 files changed, 689 insertions(+)
>>>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>>
>>> ...
>>>
>>>> diff --git a/drivers/dma/tegra20-ahb-dma.c b/drivers/dma/tegra20-ahb-dma.c
>>>> new file mode 100644
>>>> index 000000000000..8316d64e35e1
>>>> --- /dev/null
>>>> +++ b/drivers/dma/tegra20-ahb-dma.c
>>>> @@ -0,0 +1,679 @@
>>>> +/*
>>>> + * Copyright 2017 Dmitry Osipenko <digetx@gmail.com>
>>>> + *
>>>> + * This program is free software; you can redistribute it and/or modify it
>>>> + * under the terms and conditions of the GNU General Public License,
>>>> + * version 2, as published by the Free Software Foundation.
>>>> + *
>>>> + * This program is distributed in the hope it will be useful, but WITHOUT
>>>> + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
>>>> + * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
>>>> + * more details.
>>>> + *
>>>> + * You should have received a copy of the GNU General Public License
>>>> + * along with this program.  If not, see <http://www.gnu.org/licenses/>.
>>>> + */
>>>> +
>>>> +#include <linux/clk.h>
>>>> +#include <linux/delay.h>
>>>> +#include <linux/interrupt.h>
>>>> +#include <linux/io.h>
>>>> +#include <linux/module.h>
>>>> +#include <linux/of_device.h>
>>>> +#include <linux/of_dma.h>
>>>> +#include <linux/platform_device.h>
>>>> +#include <linux/reset.h>
>>>> +#include <linux/slab.h>
>>>> +#include <linux/spinlock.h>
>>>> +
>>>> +#include "dmaengine.h"
>>>> +
>>>> +#define TEGRA_AHBDMA_CMD			0x0
>>>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>>>> +
>>>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>>>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>>>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>>>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>>>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>>>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>>>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>>>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>>>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>>>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_STA		0x4
>>>> +#define TEGRA_AHBDMA_CHANNEL_IS_EOC		BIT(30)
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_PTR		0x10
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_SEQ		0x14
>>>> +#define TEGRA_AHBDMA_CHANNEL_INTR_ENB		BIT(31)
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT	24
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_1	2
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_4	3
>>>> +#define TEGRA_AHBDMA_CHANNEL_AHB_BURST_8	4
>>>> +
>>>> +#define TEGRA_AHBDMA_CHANNEL_XMB_PTR		0x18
>>>> +
>>>> +#define TEGRA_AHBDMA_BUS_WIDTH			BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>>>> +
>>>> +#define TEGRA_AHBDMA_DIRECTIONS			BIT(DMA_DEV_TO_MEM) | \
>>>> +						BIT(DMA_MEM_TO_DEV)
>>>> +
>>>> +struct tegra_ahbdma_tx_desc {
>>>> +	struct dma_async_tx_descriptor desc;
>>>> +	struct tasklet_struct tasklet;
>>>> +	struct list_head node;
>>>
>>> Any reason why we cannot use the virt-dma framework for this driver? I
>>> would hope it would simplify the driver a bit.
>>>
>>
>> IIUC virt-dma is supposed to provide virtually unlimited number of channels.
>> I've looked at it and decided that it would just add unnecessary functionality
>> and, as a result, complexity. As I wrote in the cover-letter, it is supposed
>> that this driver would have only one consumer - the host1x. It shouldn't be
>> difficult to implement virt-dma later, if desired.  But again it is very
>> unlikely that it would be needed.
> 
> I think that the biggest benefit is that is simplifies the linked list
> management. See the tegra210-adma driver.
> 

I'll take a more thorough look at it. Thank you for suggestion.

>>>> +	enum dma_transfer_direction dir;
>>>> +	dma_addr_t mem_paddr;
>>>> +	unsigned long flags;
>>>> +	size_t size;
>>>> +	bool in_fly;
>>>> +	bool cyclic;
>>>> +};
>>>> +
>>>> +struct tegra_ahbdma_chan {
>>>> +	struct dma_chan dma_chan;
>>>> +	struct list_head active_list;
>>>> +	struct list_head pending_list;
>>>> +	struct completion idling;
>>>> +	void __iomem *regs;
>>>> +	spinlock_t lock;
>>>> +	unsigned int id;
>>>> +};
>>>> +
>>>> +struct tegra_ahbdma {
>>>> +	struct tegra_ahbdma_chan channels[4];
>>>> +	struct dma_device dma_dev;
>>>> +	struct reset_control *rst;
>>>> +	struct clk *clk;
>>>> +	void __iomem *regs;
>>>> +};
>>>> +
>>>> +static inline struct tegra_ahbdma *to_ahbdma(struct dma_device *dev)
>>>> +{
>>>> +	return container_of(dev, struct tegra_ahbdma, dma_dev);
>>>> +}
>>>> +
>>>> +static inline struct tegra_ahbdma_chan *to_ahbdma_chan(struct dma_chan *chan)
>>>> +{
>>>> +	return container_of(chan, struct tegra_ahbdma_chan, dma_chan);
>>>> +}
>>>> +
>>>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>>>> +				struct dma_async_tx_descriptor *tx)
>>>> +{
>>>> +	return container_of(tx, struct tegra_ahbdma_tx_desc, desc);
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_submit_tx(struct tegra_ahbdma_chan *chan,
>>>> +				   struct tegra_ahbdma_tx_desc *tx)
>>>> +{
>>>> +	u32 csr;
>>>> +
>>>> +	writel_relaxed(tx->mem_paddr,
>>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_XMB_PTR);
>>>> +
>>>> +	csr = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +
>>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>>> +	csr |= TEGRA_AHBDMA_CHANNEL_ENABLE;
>>>> +	csr |= TEGRA_AHBDMA_CHANNEL_IE_EOC;
>>>> +	csr |= tx->size - sizeof(u32);
>>>> +
>>>> +	if (tx->dir == DMA_DEV_TO_MEM)
>>>> +		csr |= TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB;
>>>> +
>>>> +	if (!tx->cyclic)
>>>> +		csr |= TEGRA_AHBDMA_CHANNEL_ONCE;
>>>> +
>>>> +	writel_relaxed(csr, chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +
>>>> +	tx->in_fly = true;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_tasklet(unsigned long data)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>>> +
>>>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>>>> +
>>>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>>>> +		kfree(tx);
>>>> +}
>>>> +
>>>> +static bool tegra_ahbdma_tx_completed(struct tegra_ahbdma_chan *chan,
>>>> +				      struct tegra_ahbdma_tx_desc *tx)
>>>> +{
>>>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>>>> +	bool reuse = dmaengine_desc_test_reuse(desc);
>>>> +	bool interrupt = tx->flags & DMA_PREP_INTERRUPT;
>>>> +	bool completed = !tx->cyclic;
>>>> +
>>>> +	if (completed)
>>>> +		dma_cookie_complete(desc);
>>>> +
>>>> +	if (interrupt)
>>>> +		tasklet_schedule(&tx->tasklet);
>>>> +
>>>> +	if (completed) {
>>>> +		list_del(&tx->node);
>>>> +
>>>> +		if (reuse)
>>>> +			tx->in_fly = false;
>>>> +
>>>> +		if (!interrupt && !reuse)
>>>> +			kfree(tx);
>>>> +	}
>>>> +
>>>> +	return completed;
>>>> +}
>>>> +
>>>> +static bool tegra_ahbdma_next_tx_issued(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +
>>>> +	tx = list_first_entry_or_null(&chan->active_list,
>>>> +				      struct tegra_ahbdma_tx_desc,
>>>> +				      node);
>>>> +	if (tx)
>>>> +		tegra_ahbdma_submit_tx(chan, tx);
>>>> +
>>>> +	return !!tx;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	unsigned long flags;
>>>> +	u32 status;
>>>> +
>>>> +	status = readl_relaxed(chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>>> +	if (!(status & TEGRA_AHBDMA_CHANNEL_IS_EOC))
>>>> +		return;
>>>> +
>>>> +	writel_relaxed(TEGRA_AHBDMA_CHANNEL_IS_EOC,
>>>> +		       chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>>> +
>>>> +	spin_lock_irqsave(&chan->lock, flags);
>>>> +
>>>> +	if (!completion_done(&chan->idling)) {
>>>> +		tx = list_first_entry(&chan->active_list,
>>>> +				      struct tegra_ahbdma_tx_desc,
>>>> +				      node);
>>>> +
>>>> +		if (tegra_ahbdma_tx_completed(chan, tx) &&
>>>> +		    !tegra_ahbdma_next_tx_issued(chan))
>>>> +			complete_all(&chan->idling);
>>>> +	}
>>>> +
>>>> +	spin_unlock_irqrestore(&chan->lock, flags);
>>>> +}
>>>> +
>>>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>>>> +{
>>>> +	struct tegra_ahbdma *tdma = dev_id;
>>>> +	unsigned int i;
>>>> +
>>>> +	for (i = 0; i < ARRAY_SIZE(tdma->channels); i++)
>>>> +		tegra_ahbdma_handle_channel(&tdma->channels[i]);
>>>> +
>>>> +	return IRQ_HANDLED;
>>>> +}
>>>> +
>>>> +static dma_cookie_t tegra_ahbdma_tx_submit(struct dma_async_tx_descriptor *desc)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(desc);
>>>> +	struct tegra_ahbdma_chan *chan = to_ahbdma_chan(desc->chan);
>>>> +	dma_cookie_t cookie;
>>>> +
>>>> +	cookie = dma_cookie_assign(desc);
>>>> +
>>>> +	spin_lock_irq(&chan->lock);
>>>> +	list_add_tail(&tx->node, &chan->pending_list);
>>>> +	spin_unlock_irq(&chan->lock);
>>>> +
>>>> +	return cookie;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_tx_desc_free(struct dma_async_tx_descriptor *desc)
>>>> +{
>>>> +	kfree(to_ahbdma_tx_desc(desc));
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_slave_sg(
>>>> +					struct dma_chan *chan,
>>>> +					struct scatterlist *sgl,
>>>> +					unsigned int sg_len,
>>>> +					enum dma_transfer_direction dir,
>>>> +					unsigned long flags,
>>>> +					void *context)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +
>>>> +	/* unimplemented */
>>>> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>>>> +		return NULL;
>>>> +
>>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>>> +	if (!tx)
>>>> +		return NULL;
>>>> +
>>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>>> +
>>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>>> +	tx->desc.desc_free	= tegra_ahbdma_tx_desc_free;
>>>> +	tx->mem_paddr		= sg_dma_address(sgl);
>>>> +	tx->size		= sg_dma_len(sgl);
>>>> +	tx->flags		= flags;
>>>> +	tx->dir			= dir;
>>>> +
>>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>>> +
>>>> +	return &tx->desc;
>>>> +}
>>>> +
>>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>>>> +					struct dma_chan *chan,
>>>> +					dma_addr_t buf_addr,
>>>> +					size_t buf_len,
>>>> +					size_t period_len,
>>>> +					enum dma_transfer_direction dir,
>>>> +					unsigned long flags)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +
>>>> +	/* unimplemented */
>>>> +	if (buf_len != period_len || buf_len > SZ_64K)
>>>> +		return NULL;
>>>> +
>>>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>>>> +	if (!tx)
>>>> +		return NULL;
>>>> +
>>>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>>>> +
>>>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>>>> +	tx->mem_paddr		= buf_addr;
>>>> +	tx->size		= buf_len;
>>>> +	tx->flags		= flags;
>>>> +	tx->cyclic		= true;
>>>> +	tx->dir			= dir;
>>>> +
>>>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
>>>> +
>>>> +	return &tx->desc;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	struct list_head *entry, *tmp;
>>>> +	unsigned long flags;
>>>> +
>>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>>>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>>>> +
>>>> +	if (completion_done(&ahbdma_chan->idling)) {
>>>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>>>> +					      struct tegra_ahbdma_tx_desc,
>>>> +					      node);
>>>> +		if (tx) {
>>>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
>>>> +			reinit_completion(&ahbdma_chan->idling);
>>>> +		}
>>>> +	}
>>>> +
>>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>>> +}
>>>> +
>>>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>>>> +					      dma_cookie_t cookie,
>>>> +					      struct dma_tx_state *state)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	enum dma_status cookie_status;
>>>> +	unsigned long flags;
>>>> +	size_t residual;
>>>> +	u32 status;
>>>> +
>>>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>>>> +
>>>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>>>> +	if (cookie_status != DMA_COMPLETE) {
>>>> +		list_for_each_entry(tx, &ahbdma_chan->active_list, node) {
>>>> +			if (tx->desc.cookie == cookie)
>>>> +				goto found;
>>>> +		}
>>>> +	}
>>>> +
>>>> +	goto unlock;
>>>> +
>>>> +found:
>>>> +	if (tx->in_fly) {
>>>> +		status = readl_relaxed(
>>>> +			ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_STA);
>>>> +		status  &= TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK;
>>>> +
>>>> +		residual = status;
>>>> +	} else
>>>> +		residual = tx->size;
>>>> +
>>>> +	dma_set_residue(state, residual);
>>>> +
>>>> +unlock:
>>>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>>>> +
>>>> +	return cookie_status;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	struct list_head *entry, *tmp;
>>>> +	u32 csr;
>>>> +
>>>> +	spin_lock_irq(&ahbdma_chan->lock);
>>>> +
>>>> +	csr = readl_relaxed(ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +	csr &= ~TEGRA_AHBDMA_CHANNEL_ENABLE;
>>>> +
>>>> +	writel_relaxed(csr, ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->active_list) {
>>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>>> +		list_del(entry);
>>>> +		kfree(tx);
>>>> +	}
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>>> +		list_del(entry);
>>>> +		kfree(tx);
>>>> +	}
>>>> +
>>>> +	complete_all(&ahbdma_chan->idling);
>>>> +
>>>> +	spin_unlock_irq(&ahbdma_chan->lock);
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>>>> +			       struct dma_slave_config *sconfig)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	enum dma_transfer_direction dir = sconfig->direction;
>>>> +	u32 burst, ahb_seq, ahb_addr;
>>>> +
>>>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>>> +		return -EINVAL;
>>>> +
>>>> +	if (dir == DMA_DEV_TO_MEM) {
>>>> +		burst    = sconfig->src_maxburst;
>>>> +		ahb_addr = sconfig->src_addr;
>>>> +	} else {
>>>> +		burst    = sconfig->dst_maxburst;
>>>> +		ahb_addr = sconfig->dst_addr;
>>>> +	}
>>>> +
>>>> +	switch (burst) {
>>>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>>>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>>>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
>>>> +	default:
>>>> +		return -EINVAL;
>>>> +	}
>>>> +
>>>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>>>> +
>>>> +	writel_relaxed(ahb_seq,
>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>>>> +
>>>> +	writel_relaxed(ahb_addr,
>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>>> +{
>>>> +	wait_for_completion(&to_ahbdma_chan(chan)->idling);
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	struct tegra_ahbdma_tx_desc *tx;
>>>> +	struct list_head *entry, *tmp;
>>>> +
>>>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list) {
>>>> +		tx = list_entry(entry, struct tegra_ahbdma_tx_desc, node);
>>>> +		list_del(entry);
>>>> +		kfree(tx);
>>>> +	}
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_init_channel(struct tegra_ahbdma *tdma,
>>>> +				      unsigned int chan_id)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = &tdma->channels[chan_id];
>>>> +	struct dma_chan *dma_chan = &ahbdma_chan->dma_chan;
>>>> +	struct dma_device *dma_dev = &tdma->dma_dev;
>>>> +
>>>> +	INIT_LIST_HEAD(&ahbdma_chan->active_list);
>>>> +	INIT_LIST_HEAD(&ahbdma_chan->pending_list);
>>>> +	init_completion(&ahbdma_chan->idling);
>>>> +	spin_lock_init(&ahbdma_chan->lock);
>>>> +	complete(&ahbdma_chan->idling);
>>>> +
>>>> +	ahbdma_chan->regs = tdma->regs + TEGRA_AHBDMA_CHANNEL_BASE(chan_id);
>>>> +	ahbdma_chan->id = chan_id;
>>>> +
>>>> +	dma_cookie_init(dma_chan);
>>>> +	dma_chan->device = dma_dev;
>>>> +
>>>> +	list_add_tail(&dma_chan->device_node, &dma_dev->channels);
>>>> +}
>>>> +
>>>> +static struct dma_chan *tegra_ahbdma_of_xlate(struct of_phandle_args *dma_spec,
>>>> +					      struct of_dma *ofdma)
>>>> +{
>>>> +	struct tegra_ahbdma *tdma = ofdma->of_dma_data;
>>>> +	struct dma_chan *chan;
>>>> +	u32 csr;
>>>> +
>>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>>> +	if (!chan)
>>>> +		return NULL;
>>>> +
>>>> +	/* enable channels flow control */
>>>> +	if (dma_spec->args_count == 1) {
>>>
>>> The DT doc says #dma-cells should be '1' and so if not equal 1, is this
>>> not an error?
>>>
>>
>> I wanted to differentiate slave/master modes here. But if we'd want to add
>> TRIG_SEL as another cell, then it probably would worth to implement a custom DMA
>> configure options, like documentation suggests - to wrap generic
>> dma_slave_config into the custom one. On the other hand that probably would add
>> an unused functionality to the driver.
>>
>>>> +		csr  = TEGRA_AHBDMA_CHANNEL_FLOW;
>>>> +		csr |= dma_spec->args[0] << TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT;
>>>
>>> What about the TRIG_REQ field?
>>>
>>
>> Not implemented, there is no test case for it yet.
>>
>>>> +
>>>> +		writel_relaxed(csr,
>>>> +			to_ahbdma_chan(chan)->regs + TEGRA_AHBDMA_CHANNEL_CSR);
>>>> +	}
>>>> +	
>>>> +	return chan;
>>>> +}
>>>> +
>>>> +static int tegra_ahbdma_init_hw(struct tegra_ahbdma *tdma, struct device *dev)
>>>> +{
>>>> +	int err;
>>>> +
>>>> +	err = reset_control_assert(tdma->rst);
>>>> +	if (err) {
>>>> +		dev_err(dev, "Failed to assert reset: %d\n", err);
>>>> +		return err;
>>>> +	}
>>>> +
>>>> +	err = clk_prepare_enable(tdma->clk);
>>>> +	if (err) {
>>>> +		dev_err(dev, "Failed to enable clock: %d\n", err);
>>>> +		return err;
>>>> +	}
>>>> +
>>>> +	usleep_range(1000, 2000);
>>>> +
>>>> +	err = reset_control_deassert(tdma->rst);
>>>> +	if (err) {
>>>> +		dev_err(dev, "Failed to deassert reset: %d\n", err);
>>>> +		return err;
>>>> +	}
>>>> +
>>>> +	writel_relaxed(TEGRA_AHBDMA_CMD_ENABLE, tdma->regs + TEGRA_AHBDMA_CMD);
>>>> +
>>>> +	writel_relaxed(TEGRA_AHBDMA_IRQ_ENB_CH(0) |
>>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(1) |
>>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(2) |
>>>> +		       TEGRA_AHBDMA_IRQ_ENB_CH(3),
>>>> +		       tdma->regs + TEGRA_AHBDMA_IRQ_ENB_MASK);
>>>> +
>>>> +	return 0;
>>>> +}
>>>
>>> Personally I would use the pm_runtime callbacks for this sort of thing
>>> and ...
>>>
>>
>> I decided that it probaby would be better to implement PM later if needed. I'm
>> not sure whether DMA controller consumes any substantial amounts of power while
>> idling. If it's not, why bother? Unnecessary power managment would just cause
>> CPU to waste its cycles (and power) doing PM.
> 
> Yes it probably does not but it is easy to do and so even though there
> are probably a ton of other clocks left running, I still think it is
> good practice.
> 

Okay, I'll take a look into implementing PM. Disabling AHBDMA clock won't stop
the actual clock, but only gate it to the controller.

Thank you for the comments!

-- 
Dmitry

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 26.09.2017 17:50, Jon Hunter wrote:
> 
> On 26/09/17 00:22, Dmitry Osipenko wrote:
>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>> on Tegra20/30 SoC's.
>>
>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>> ---
>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>  1 file changed, 23 insertions(+)
>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>
>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> new file mode 100644
>> index 000000000000..2af9aa76ae11
>> --- /dev/null
>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> @@ -0,0 +1,23 @@
>> +* NVIDIA Tegra AHB DMA controller
>> +
>> +Required properties:
>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>> +- reg:		Should contain registers base address and length.
>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>> +- clocks:	Should contain one entry, DMA controller clock.
>> +- resets :	Should contain one entry, DMA controller reset.
>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>> +		for the peripheral. For more details consult the Tegra TRM's
>> +		documentation, in particular AHB DMA channel control register
>> +		REQ_SEL field.
> 
> What about the TRIG_SEL field? Do we need to handle this here as well?
> 

Actually, DMA transfer trigger isn't related a hardware description. It's up to
software to decide what trigger to select. So it shouldn't be in the binding.

And I think the same applies to requester... any objections?

-- 
Dmitry
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 26.09.2017 17:50, Jon Hunter wrote:
> 
> On 26/09/17 00:22, Dmitry Osipenko wrote:
>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>> on Tegra20/30 SoC's.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>> ---
>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>  1 file changed, 23 insertions(+)
>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>
>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> new file mode 100644
>> index 000000000000..2af9aa76ae11
>> --- /dev/null
>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> @@ -0,0 +1,23 @@
>> +* NVIDIA Tegra AHB DMA controller
>> +
>> +Required properties:
>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>> +- reg:		Should contain registers base address and length.
>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>> +- clocks:	Should contain one entry, DMA controller clock.
>> +- resets :	Should contain one entry, DMA controller reset.
>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>> +		for the peripheral. For more details consult the Tegra TRM's
>> +		documentation, in particular AHB DMA channel control register
>> +		REQ_SEL field.
> 
> What about the TRIG_SEL field? Do we need to handle this here as well?
> 

Actually, DMA transfer trigger isn't related a hardware description. It's up to
software to decide what trigger to select. So it shouldn't be in the binding.

And I think the same applies to requester... any objections?

-- 
Dmitry

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 27/09/17 02:57, Dmitry Osipenko wrote:
> On 26.09.2017 17:50, Jon Hunter wrote:
>>
>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>> on Tegra20/30 SoC's.
>>>
>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>> ---
>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>  1 file changed, 23 insertions(+)
>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>
>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>> new file mode 100644
>>> index 000000000000..2af9aa76ae11
>>> --- /dev/null
>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>> @@ -0,0 +1,23 @@
>>> +* NVIDIA Tegra AHB DMA controller
>>> +
>>> +Required properties:
>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>> +- reg:		Should contain registers base address and length.
>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>> +- clocks:	Should contain one entry, DMA controller clock.
>>> +- resets :	Should contain one entry, DMA controller reset.
>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>> +		for the peripheral. For more details consult the Tegra TRM's
>>> +		documentation, in particular AHB DMA channel control register
>>> +		REQ_SEL field.
>>
>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>
> 
> Actually, DMA transfer trigger isn't related a hardware description. It's up to
> software to decide what trigger to select. So it shouldn't be in the binding.

I think it could be, if say a board wanted a GPIO to trigger a transfer.

> And I think the same applies to requester... any objections?

Well, the REQ_SEL should definitely be in the binding.

Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
like we never bothered with it for the APB DMA and so maybe no ones uses
this.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 27/09/17 02:57, Dmitry Osipenko wrote:
> On 26.09.2017 17:50, Jon Hunter wrote:
>>
>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>> on Tegra20/30 SoC's.
>>>
>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>> ---
>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>  1 file changed, 23 insertions(+)
>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>
>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>> new file mode 100644
>>> index 000000000000..2af9aa76ae11
>>> --- /dev/null
>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>> @@ -0,0 +1,23 @@
>>> +* NVIDIA Tegra AHB DMA controller
>>> +
>>> +Required properties:
>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>> +- reg:		Should contain registers base address and length.
>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>> +- clocks:	Should contain one entry, DMA controller clock.
>>> +- resets :	Should contain one entry, DMA controller reset.
>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>> +		for the peripheral. For more details consult the Tegra TRM's
>>> +		documentation, in particular AHB DMA channel control register
>>> +		REQ_SEL field.
>>
>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>
> 
> Actually, DMA transfer trigger isn't related a hardware description. It's up to
> software to decide what trigger to select. So it shouldn't be in the binding.

I think it could be, if say a board wanted a GPIO to trigger a transfer.

> And I think the same applies to requester... any objections?

Well, the REQ_SEL should definitely be in the binding.

Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
like we never bothered with it for the APB DMA and so maybe no ones uses
this.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 1/5] clk: tegra: Add AHB DMA clock entry

[Peter De Schrijver]

On Tue, Sep 26, 2017 at 05:46:01PM +0300, Dmitry Osipenko wrote:
> On 26.09.2017 12:56, Peter De Schrijver wrote:
> > On Tue, Sep 26, 2017 at 02:22:02AM +0300, Dmitry Osipenko wrote:
> >> AHB DMA presents on Tegra20/30. Add missing entries, so that driver
> >> for AHB DMA could be implemented.
> >>
> >> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> >> ---
> >>  drivers/clk/tegra/clk-id.h           | 1 +
> >>  drivers/clk/tegra/clk-tegra-periph.c | 1 +
> >>  drivers/clk/tegra/clk-tegra20.c      | 6 ++++++
> >>  drivers/clk/tegra/clk-tegra30.c      | 2 ++
> >>  4 files changed, 10 insertions(+)
> >>
> >> diff --git a/drivers/clk/tegra/clk-id.h b/drivers/clk/tegra/clk-id.h
> >> index 689f344377a7..c1661b47bbda 100644
> >> --- a/drivers/clk/tegra/clk-id.h
> >> +++ b/drivers/clk/tegra/clk-id.h
> >> @@ -12,6 +12,7 @@ enum clk_id {
> >>  	tegra_clk_amx,
> >>  	tegra_clk_amx1,
> >>  	tegra_clk_apb2ape,
> >> +	tegra_clk_ahbdma,
> >>  	tegra_clk_apbdma,
> >>  	tegra_clk_apbif,
> >>  	tegra_clk_ape,
> >> diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c
> >> index 848255cc0209..95a3d8c95f06 100644
> >> --- a/drivers/clk/tegra/clk-tegra-periph.c
> >> +++ b/drivers/clk/tegra/clk-tegra-periph.c
> >> @@ -823,6 +823,7 @@ static struct tegra_periph_init_data gate_clks[] = {
> >>  	GATE("timer", "clk_m", 5, 0, tegra_clk_timer, CLK_IS_CRITICAL),
> >>  	GATE("isp", "clk_m", 23, 0, tegra_clk_isp, 0),
> >>  	GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0),
> >> +	GATE("ahbdma", "clk_m", 33, 0, tegra_clk_ahbdma, 0),
> > 
> > Parent for this should be hclk on Tegra30 and later chips as well..
> > 
> 
> It looks like other clocks have a wrong parent too here, aren't they? Like for
> example "apbdma" should have "pclk" as a parent, isn't it?
> 

Yes. That is correct.

> >>  	GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0),
> >>  	GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0),
> >>  	GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0),
> >> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
> >> index 837e5cbd60e9..e76c0d292ca7 100644
> >> --- a/drivers/clk/tegra/clk-tegra20.c
> >> +++ b/drivers/clk/tegra/clk-tegra20.c
> >> @@ -449,6 +449,7 @@ static struct tegra_devclk devclks[] __initdata = {
> >>  	{ .con_id = "audio", .dt_id = TEGRA20_CLK_AUDIO },
> >>  	{ .con_id = "audio_2x", .dt_id = TEGRA20_CLK_AUDIO_2X },
> >>  	{ .dev_id = "tegra20-ac97", .dt_id = TEGRA20_CLK_AC97 },
> >> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA20_CLK_AHBDMA },
> > 
> > This isn't needed if you use DT bindings to get the clock handle.
> > 
> 
> Yes, I added it for consistency. Shouldn't we get rid of that all legacy stuff
> already?
> 

We probably should, but we can start by not adding more :)

> >>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA20_CLK_APBDMA },
> >>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA20_CLK_RTC },
> >>  	{ .dev_id = "timer", .dt_id = TEGRA20_CLK_TIMER },
> >> @@ -806,6 +807,11 @@ static void __init tegra20_periph_clk_init(void)
> >>  				    clk_base, 0, 3, periph_clk_enb_refcnt);
> >>  	clks[TEGRA20_CLK_AC97] = clk;
> >>  
> >> +	/* ahbdma */
> >> +	clk = tegra_clk_register_periph_gate("ahbdma", "hclk", 0, clk_base,
> >> +				    0, 33, periph_clk_enb_refcnt);
> >> +	clks[TEGRA20_CLK_AHBDMA] = clk;
> >> +
> > 
> > You can use the generic definition here if you correct the entry above.
> > 
> 
> Good point, same applies to "apbdma". Thank you for the suggestion.
> 

Indeed.

Peter.

Re: [PATCH v1 1/5] clk: tegra: Add AHB DMA clock entry

[Peter De Schrijver]

On Tue, Sep 26, 2017 at 05:46:01PM +0300, Dmitry Osipenko wrote:
> On 26.09.2017 12:56, Peter De Schrijver wrote:
> > On Tue, Sep 26, 2017 at 02:22:02AM +0300, Dmitry Osipenko wrote:
> >> AHB DMA presents on Tegra20/30. Add missing entries, so that driver
> >> for AHB DMA could be implemented.
> >>
> >> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> >> ---
> >>  drivers/clk/tegra/clk-id.h           | 1 +
> >>  drivers/clk/tegra/clk-tegra-periph.c | 1 +
> >>  drivers/clk/tegra/clk-tegra20.c      | 6 ++++++
> >>  drivers/clk/tegra/clk-tegra30.c      | 2 ++
> >>  4 files changed, 10 insertions(+)
> >>
> >> diff --git a/drivers/clk/tegra/clk-id.h b/drivers/clk/tegra/clk-id.h
> >> index 689f344377a7..c1661b47bbda 100644
> >> --- a/drivers/clk/tegra/clk-id.h
> >> +++ b/drivers/clk/tegra/clk-id.h
> >> @@ -12,6 +12,7 @@ enum clk_id {
> >>  	tegra_clk_amx,
> >>  	tegra_clk_amx1,
> >>  	tegra_clk_apb2ape,
> >> +	tegra_clk_ahbdma,
> >>  	tegra_clk_apbdma,
> >>  	tegra_clk_apbif,
> >>  	tegra_clk_ape,
> >> diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c
> >> index 848255cc0209..95a3d8c95f06 100644
> >> --- a/drivers/clk/tegra/clk-tegra-periph.c
> >> +++ b/drivers/clk/tegra/clk-tegra-periph.c
> >> @@ -823,6 +823,7 @@ static struct tegra_periph_init_data gate_clks[] = {
> >>  	GATE("timer", "clk_m", 5, 0, tegra_clk_timer, CLK_IS_CRITICAL),
> >>  	GATE("isp", "clk_m", 23, 0, tegra_clk_isp, 0),
> >>  	GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0),
> >> +	GATE("ahbdma", "clk_m", 33, 0, tegra_clk_ahbdma, 0),
> > 
> > Parent for this should be hclk on Tegra30 and later chips as well..
> > 
> 
> It looks like other clocks have a wrong parent too here, aren't they? Like for
> example "apbdma" should have "pclk" as a parent, isn't it?
> 

Yes. That is correct.

> >>  	GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0),
> >>  	GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0),
> >>  	GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0),
> >> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
> >> index 837e5cbd60e9..e76c0d292ca7 100644
> >> --- a/drivers/clk/tegra/clk-tegra20.c
> >> +++ b/drivers/clk/tegra/clk-tegra20.c
> >> @@ -449,6 +449,7 @@ static struct tegra_devclk devclks[] __initdata = {
> >>  	{ .con_id = "audio", .dt_id = TEGRA20_CLK_AUDIO },
> >>  	{ .con_id = "audio_2x", .dt_id = TEGRA20_CLK_AUDIO_2X },
> >>  	{ .dev_id = "tegra20-ac97", .dt_id = TEGRA20_CLK_AC97 },
> >> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA20_CLK_AHBDMA },
> > 
> > This isn't needed if you use DT bindings to get the clock handle.
> > 
> 
> Yes, I added it for consistency. Shouldn't we get rid of that all legacy stuff
> already?
> 

We probably should, but we can start by not adding more :)

> >>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA20_CLK_APBDMA },
> >>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA20_CLK_RTC },
> >>  	{ .dev_id = "timer", .dt_id = TEGRA20_CLK_TIMER },
> >> @@ -806,6 +807,11 @@ static void __init tegra20_periph_clk_init(void)
> >>  				    clk_base, 0, 3, periph_clk_enb_refcnt);
> >>  	clks[TEGRA20_CLK_AC97] = clk;
> >>  
> >> +	/* ahbdma */
> >> +	clk = tegra_clk_register_periph_gate("ahbdma", "hclk", 0, clk_base,
> >> +				    0, 33, periph_clk_enb_refcnt);
> >> +	clks[TEGRA20_CLK_AHBDMA] = clk;
> >> +
> > 
> > You can use the generic definition here if you correct the entry above.
> > 
> 
> Good point, same applies to "apbdma". Thank you for the suggestion.
> 

Indeed.

Peter.

Re: [PATCH v1 1/5] clk: tegra: Add AHB DMA clock entry

[Dmitry Osipenko]

On 27.09.2017 11:36, Peter De Schrijver wrote:
> On Tue, Sep 26, 2017 at 05:46:01PM +0300, Dmitry Osipenko wrote:
>> On 26.09.2017 12:56, Peter De Schrijver wrote:
>>> On Tue, Sep 26, 2017 at 02:22:02AM +0300, Dmitry Osipenko wrote:
>>>> AHB DMA presents on Tegra20/30. Add missing entries, so that driver
>>>> for AHB DMA could be implemented.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>> ---
>>>>  drivers/clk/tegra/clk-id.h           | 1 +
>>>>  drivers/clk/tegra/clk-tegra-periph.c | 1 +
>>>>  drivers/clk/tegra/clk-tegra20.c      | 6 ++++++
>>>>  drivers/clk/tegra/clk-tegra30.c      | 2 ++
>>>>  4 files changed, 10 insertions(+)
>>>>
>>>> diff --git a/drivers/clk/tegra/clk-id.h b/drivers/clk/tegra/clk-id.h
>>>> index 689f344377a7..c1661b47bbda 100644
>>>> --- a/drivers/clk/tegra/clk-id.h
>>>> +++ b/drivers/clk/tegra/clk-id.h
>>>> @@ -12,6 +12,7 @@ enum clk_id {
>>>>  	tegra_clk_amx,
>>>>  	tegra_clk_amx1,
>>>>  	tegra_clk_apb2ape,
>>>> +	tegra_clk_ahbdma,
>>>>  	tegra_clk_apbdma,
>>>>  	tegra_clk_apbif,
>>>>  	tegra_clk_ape,
>>>> diff --git a/drivers/clk/tegra/clk-tegra-periph.c b/drivers/clk/tegra/clk-tegra-periph.c
>>>> index 848255cc0209..95a3d8c95f06 100644
>>>> --- a/drivers/clk/tegra/clk-tegra-periph.c
>>>> +++ b/drivers/clk/tegra/clk-tegra-periph.c
>>>> @@ -823,6 +823,7 @@ static struct tegra_periph_init_data gate_clks[] = {
>>>>  	GATE("timer", "clk_m", 5, 0, tegra_clk_timer, CLK_IS_CRITICAL),
>>>>  	GATE("isp", "clk_m", 23, 0, tegra_clk_isp, 0),
>>>>  	GATE("vcp", "clk_m", 29, 0, tegra_clk_vcp, 0),
>>>> +	GATE("ahbdma", "clk_m", 33, 0, tegra_clk_ahbdma, 0),
>>>
>>> Parent for this should be hclk on Tegra30 and later chips as well..
>>>
>>
>> It looks like other clocks have a wrong parent too here, aren't they? Like for
>> example "apbdma" should have "pclk" as a parent, isn't it?
>>
> 
> Yes. That is correct.
> 

Okay, I'll fix it in V2.

>>>>  	GATE("apbdma", "clk_m", 34, 0, tegra_clk_apbdma, 0),
>>>>  	GATE("kbc", "clk_32k", 36, TEGRA_PERIPH_ON_APB | TEGRA_PERIPH_NO_RESET, tegra_clk_kbc, 0),
>>>>  	GATE("fuse", "clk_m", 39, TEGRA_PERIPH_ON_APB, tegra_clk_fuse, 0),
>>>> diff --git a/drivers/clk/tegra/clk-tegra20.c b/drivers/clk/tegra/clk-tegra20.c
>>>> index 837e5cbd60e9..e76c0d292ca7 100644
>>>> --- a/drivers/clk/tegra/clk-tegra20.c
>>>> +++ b/drivers/clk/tegra/clk-tegra20.c
>>>> @@ -449,6 +449,7 @@ static struct tegra_devclk devclks[] __initdata = {
>>>>  	{ .con_id = "audio", .dt_id = TEGRA20_CLK_AUDIO },
>>>>  	{ .con_id = "audio_2x", .dt_id = TEGRA20_CLK_AUDIO_2X },
>>>>  	{ .dev_id = "tegra20-ac97", .dt_id = TEGRA20_CLK_AC97 },
>>>> +	{ .dev_id = "tegra-ahbdma", .dt_id = TEGRA20_CLK_AHBDMA },
>>>
>>> This isn't needed if you use DT bindings to get the clock handle.
>>>
>>
>> Yes, I added it for consistency. Shouldn't we get rid of that all legacy stuff
>> already?
>>
> 
> We probably should, but we can start by not adding more :)
> 

Sure ;)

>>>>  	{ .dev_id = "tegra-apbdma", .dt_id = TEGRA20_CLK_APBDMA },
>>>>  	{ .dev_id = "rtc-tegra", .dt_id = TEGRA20_CLK_RTC },
>>>>  	{ .dev_id = "timer", .dt_id = TEGRA20_CLK_TIMER },
>>>> @@ -806,6 +807,11 @@ static void __init tegra20_periph_clk_init(void)
>>>>  				    clk_base, 0, 3, periph_clk_enb_refcnt);
>>>>  	clks[TEGRA20_CLK_AC97] = clk;
>>>>  
>>>> +	/* ahbdma */
>>>> +	clk = tegra_clk_register_periph_gate("ahbdma", "hclk", 0, clk_base,
>>>> +				    0, 33, periph_clk_enb_refcnt);
>>>> +	clks[TEGRA20_CLK_AHBDMA] = clk;
>>>> +
>>>
>>> You can use the generic definition here if you correct the entry above.
>>>
>>
>> Good point, same applies to "apbdma". Thank you for the suggestion.
>>
> 
> Indeed.

-- 
Dmitry

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 27.09.2017 11:34, Jon Hunter wrote:
> 
> On 27/09/17 02:57, Dmitry Osipenko wrote:
>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>
>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>> on Tegra20/30 SoC's.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>> ---
>>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>  1 file changed, 23 insertions(+)
>>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>> new file mode 100644
>>>> index 000000000000..2af9aa76ae11
>>>> --- /dev/null
>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>> @@ -0,0 +1,23 @@
>>>> +* NVIDIA Tegra AHB DMA controller
>>>> +
>>>> +Required properties:
>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>> +- reg:		Should contain registers base address and length.
>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>> +		documentation, in particular AHB DMA channel control register
>>>> +		REQ_SEL field.
>>>
>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>
>>
>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>> software to decide what trigger to select. So it shouldn't be in the binding.
> 
> I think it could be, if say a board wanted a GPIO to trigger a transfer.
> 

GPIO isn't a very good example, there is no "GPIO" trigger. To me all triggers
are software-defined, so that software could create transfer chains.

>> And I think the same applies to requester... any objections?
> 
> Well, the REQ_SEL should definitely be in the binding.
> 

Okay.

> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
> like we never bothered with it for the APB DMA and so maybe no ones uses
> this.
> 

-- 
Dmitry

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 27/09/17 13:12, Dmitry Osipenko wrote:
> On 27.09.2017 11:34, Jon Hunter wrote:
>>
>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>
>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>> on Tegra20/30 SoC's.
>>>>>
>>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>>> ---
>>>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>>  1 file changed, 23 insertions(+)
>>>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>
>>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>> new file mode 100644
>>>>> index 000000000000..2af9aa76ae11
>>>>> --- /dev/null
>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>> @@ -0,0 +1,23 @@
>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>> +
>>>>> +Required properties:
>>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>>> +- reg:		Should contain registers base address and length.
>>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>>> +		documentation, in particular AHB DMA channel control register
>>>>> +		REQ_SEL field.
>>>>
>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>
>>>
>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>
>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>
> 
> GPIO isn't a very good example, there is no "GPIO" trigger. To me all triggers
> are software-defined, so that software could create transfer chains.

TRM shows the following in the APBDMA_TRIG_REG_0 ...

"XRQ_A: XRQ.A (GPIOA) (Hardware initiated DMA request)"

Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 27/09/17 13:12, Dmitry Osipenko wrote:
> On 27.09.2017 11:34, Jon Hunter wrote:
>>
>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>
>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>> on Tegra20/30 SoC's.
>>>>>
>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>> ---
>>>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>>  1 file changed, 23 insertions(+)
>>>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>
>>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>> new file mode 100644
>>>>> index 000000000000..2af9aa76ae11
>>>>> --- /dev/null
>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>> @@ -0,0 +1,23 @@
>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>> +
>>>>> +Required properties:
>>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>>> +- reg:		Should contain registers base address and length.
>>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>>> +		documentation, in particular AHB DMA channel control register
>>>>> +		REQ_SEL field.
>>>>
>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>
>>>
>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>
>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>
> 
> GPIO isn't a very good example, there is no "GPIO" trigger. To me all triggers
> are software-defined, so that software could create transfer chains.

TRM shows the following in the APBDMA_TRIG_REG_0 ...

"XRQ_A: XRQ.A (GPIOA) (Hardware initiated DMA request)"

Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 27/09/17 14:44, Jon Hunter wrote:
> 
> On 27/09/17 13:12, Dmitry Osipenko wrote:
>> On 27.09.2017 11:34, Jon Hunter wrote:
>>>
>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>
>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>> on Tegra20/30 SoC's.
>>>>>>
>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>> ---
>>>>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>>>  1 file changed, 23 insertions(+)
>>>>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>
>>>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>> new file mode 100644
>>>>>> index 000000000000..2af9aa76ae11
>>>>>> --- /dev/null
>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>> @@ -0,0 +1,23 @@
>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>> +
>>>>>> +Required properties:
>>>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>>>> +- reg:		Should contain registers base address and length.
>>>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>>>> +		documentation, in particular AHB DMA channel control register
>>>>>> +		REQ_SEL field.
>>>>>
>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>
>>>>
>>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>
>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>
>>
>> GPIO isn't a very good example, there is no "GPIO" trigger. To me all triggers
>> are software-defined, so that software could create transfer chains.
> 
> TRM shows the following in the APBDMA_TRIG_REG_0 ...
> 
> "XRQ_A: XRQ.A (GPIOA) (Hardware initiated DMA request)"

Furthermore there are timer and hw-semaphore triggers as well.

Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 27/09/17 14:44, Jon Hunter wrote:
> 
> On 27/09/17 13:12, Dmitry Osipenko wrote:
>> On 27.09.2017 11:34, Jon Hunter wrote:
>>>
>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>
>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>> on Tegra20/30 SoC's.
>>>>>>
>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>> ---
>>>>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>>>  1 file changed, 23 insertions(+)
>>>>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>
>>>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>> new file mode 100644
>>>>>> index 000000000000..2af9aa76ae11
>>>>>> --- /dev/null
>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>> @@ -0,0 +1,23 @@
>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>> +
>>>>>> +Required properties:
>>>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>>>> +- reg:		Should contain registers base address and length.
>>>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>>>> +		documentation, in particular AHB DMA channel control register
>>>>>> +		REQ_SEL field.
>>>>>
>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>
>>>>
>>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>
>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>
>>
>> GPIO isn't a very good example, there is no "GPIO" trigger. To me all triggers
>> are software-defined, so that software could create transfer chains.
> 
> TRM shows the following in the APBDMA_TRIG_REG_0 ...
> 
> "XRQ_A: XRQ.A (GPIOA) (Hardware initiated DMA request)"

Furthermore there are timer and hw-semaphore triggers as well.

Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 27.09.2017 16:46, Jon Hunter wrote:
> 
> On 27/09/17 14:44, Jon Hunter wrote:
>>
>> On 27/09/17 13:12, Dmitry Osipenko wrote:
>>> On 27.09.2017 11:34, Jon Hunter wrote:
>>>>
>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>
>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>> on Tegra20/30 SoC's.
>>>>>>>
>>>>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>>>>> ---
>>>>>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>>>>  1 file changed, 23 insertions(+)
>>>>>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>
>>>>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>> new file mode 100644
>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>> --- /dev/null
>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>> @@ -0,0 +1,23 @@
>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>> +
>>>>>>> +Required properties:
>>>>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>>>>> +- reg:		Should contain registers base address and length.
>>>>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>>>>> +		documentation, in particular AHB DMA channel control register
>>>>>>> +		REQ_SEL field.
>>>>>>
>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>
>>>>>
>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>
>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>
>>>
>>> GPIO isn't a very good example, there is no "GPIO" trigger. To me all triggers
>>> are software-defined, so that software could create transfer chains.
>>
>> TRM shows the following in the APBDMA_TRIG_REG_0 ...
>>
>> "XRQ_A: XRQ.A (GPIOA) (Hardware initiated DMA request)"
> 
> Furthermore there are timer and hw-semaphore triggers as well.
> 

Aha, I wasn't sure about what XRQ is. AHB DMA doesn't have XRQ.A as a trigger,
but XRQ.C/D which I suppose corresponds to GPIO C/D.

Timer and hw-semaphore are more questionable, aren't semaphores software-only
triggerable?

-- 
Dmitry
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 27.09.2017 16:46, Jon Hunter wrote:
> 
> On 27/09/17 14:44, Jon Hunter wrote:
>>
>> On 27/09/17 13:12, Dmitry Osipenko wrote:
>>> On 27.09.2017 11:34, Jon Hunter wrote:
>>>>
>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>
>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>> on Tegra20/30 SoC's.
>>>>>>>
>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>> ---
>>>>>>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>>>>  1 file changed, 23 insertions(+)
>>>>>>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>
>>>>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>> new file mode 100644
>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>> --- /dev/null
>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>> @@ -0,0 +1,23 @@
>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>> +
>>>>>>> +Required properties:
>>>>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>>>>> +- reg:		Should contain registers base address and length.
>>>>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>>>>> +		documentation, in particular AHB DMA channel control register
>>>>>>> +		REQ_SEL field.
>>>>>>
>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>
>>>>>
>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>
>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>
>>>
>>> GPIO isn't a very good example, there is no "GPIO" trigger. To me all triggers
>>> are software-defined, so that software could create transfer chains.
>>
>> TRM shows the following in the APBDMA_TRIG_REG_0 ...
>>
>> "XRQ_A: XRQ.A (GPIOA) (Hardware initiated DMA request)"
> 
> Furthermore there are timer and hw-semaphore triggers as well.
> 

Aha, I wasn't sure about what XRQ is. AHB DMA doesn't have XRQ.A as a trigger,
but XRQ.C/D which I suppose corresponds to GPIO C/D.

Timer and hw-semaphore are more questionable, aren't semaphores software-only
triggerable?

-- 
Dmitry

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Stephen Boyd]

On 09/27, Jon Hunter wrote:
> 
> 
> Well, the REQ_SEL should definitely be in the binding.
> 
> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks

Wrong Stephen?

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 28/09/17 00:32, Stephen Boyd wrote:
> On 09/27, Jon Hunter wrote:
>>
>>
>> Well, the REQ_SEL should definitely be in the binding.
>>
>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
> 
> Wrong Stephen?

Indeed! With all the folks in copy, I assumed we had the right one :-)

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 28/09/17 00:32, Stephen Boyd wrote:
> On 09/27, Jon Hunter wrote:
>>
>>
>> Well, the REQ_SEL should definitely be in the binding.
>>
>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
> 
> Wrong Stephen?

Indeed! With all the folks in copy, I assumed we had the right one :-)

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Vinod Koul]

On Tue, Sep 26, 2017 at 02:22:05AM +0300, Dmitry Osipenko wrote:

> +config TEGRA20_AHB_DMA
> +	tristate "NVIDIA Tegra20 AHB DMA support"
> +	depends on ARCH_TEGRA

Can we add COMPILE_TEST, helps me compile drivers

> +#include <linux/clk.h>
> +#include <linux/delay.h>
> +#include <linux/interrupt.h>
> +#include <linux/io.h>
> +#include <linux/module.h>
> +#include <linux/of_device.h>
> +#include <linux/of_dma.h>
> +#include <linux/platform_device.h>
> +#include <linux/reset.h>
> +#include <linux/slab.h>
> +#include <linux/spinlock.h>

no vchan.h, so i presume we are not using that here, any reason why?

> +
> +#include "dmaengine.h"
> +
> +#define TEGRA_AHBDMA_CMD			0x0
> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
> +
> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
> +
> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
> +
> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC

GENMASK() ?

> +static void tegra_ahbdma_tasklet(unsigned long data)
> +{
> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
> +	struct dma_async_tx_descriptor *desc = &tx->desc;
> +
> +	dmaengine_desc_get_callback_invoke(desc, NULL);
> +
> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
> +		kfree(tx);

lot of code here can be reduced if we use vchan

> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
> +					struct dma_chan *chan,
> +					dma_addr_t buf_addr,
> +					size_t buf_len,
> +					size_t period_len,
> +					enum dma_transfer_direction dir,
> +					unsigned long flags)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +
> +	/* unimplemented */
> +	if (buf_len != period_len || buf_len > SZ_64K)
> +		return NULL;
> +
> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
> +	if (!tx)
> +		return NULL;
> +
> +	dma_async_tx_descriptor_init(&tx->desc, chan);
> +
> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
> +	tx->mem_paddr		= buf_addr;
> +	tx->size		= buf_len;
> +	tx->flags		= flags;
> +	tx->cyclic		= true;
> +	tx->dir			= dir;
> +
> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);

why not precalulcate the register settings here. While submitting you are in
hot path keeping dmaengine idle so faster you can submit, better the perf

> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	struct list_head *entry, *tmp;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
> +
> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
> +		list_move_tail(entry, &ahbdma_chan->active_list);
> +
> +	if (completion_done(&ahbdma_chan->idling)) {
> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
> +					      struct tegra_ahbdma_tx_desc,
> +					      node);
> +		if (tx) {
> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);

what is chan is already running?

> +			reinit_completion(&ahbdma_chan->idling);
> +		}
> +	}
> +
> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
> +}
> +
> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
> +					      dma_cookie_t cookie,
> +					      struct dma_tx_state *state)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	enum dma_status cookie_status;
> +	unsigned long flags;
> +	size_t residual;
> +	u32 status;
> +
> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
> +
> +	cookie_status = dma_cookie_status(chan, cookie, state);
> +	if (cookie_status != DMA_COMPLETE) {

residue can be NULL so check it before proceeding ahead

> +static int tegra_ahbdma_config(struct dma_chan *chan,
> +			       struct dma_slave_config *sconfig)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	enum dma_transfer_direction dir = sconfig->direction;
> +	u32 burst, ahb_seq, ahb_addr;
> +
> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
> +		return -EINVAL;
> +
> +	if (dir == DMA_DEV_TO_MEM) {
> +		burst    = sconfig->src_maxburst;
> +		ahb_addr = sconfig->src_addr;
> +	} else {
> +		burst    = sconfig->dst_maxburst;
> +		ahb_addr = sconfig->dst_addr;
> +	}
> +
> +	switch (burst) {
> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;

pls make this statement and break on subsequent lines, readablity matters

> +	default:
> +		return -EINVAL;
> +	}
> +
> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
> +
> +	writel_relaxed(ahb_seq,
> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
> +
> +	writel_relaxed(ahb_addr,
> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);

oh no, you don't write to HW here. This can be called anytime when you have
txn running! You should save these and use them in prep_ calls.

> +static int tegra_ahbdma_remove(struct platform_device *pdev)
> +{
> +	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
> +
> +	of_dma_controller_free(pdev->dev.of_node);
> +	dma_async_device_unregister(&tdma->dma_dev);
> +	clk_disable_unprepare(tdma->clk);

not ensuring tasklets are killed and irq is freed so no more tasklets can
run? I think that needs to be done...

> +MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
> +MODULE_AUTHOR("Dmitry Osipenko <digetx@gmail.com>");
> +MODULE_LICENSE("GPL");

MODULE_ALIAS?

-- 
~Vinod

Re: [PATCH v1 0/5] NVIDIA Tegra AHB DMA controller driver

[Vinod Koul]

On Tue, Sep 26, 2017 at 02:22:01AM +0300, Dmitry Osipenko wrote:
> NVIDIA Tegra20/30 SoC's have AHB DMA controller. It has 4 DMA channels,
> supports AHB <-> Memory and Memory <-> Memory transfers, slave / master
> modes. This driver is primarily supposed to be used by gpu/host1x in a
> master mode, performing 3D HW context stores.
> 
> Dmitry Osipenko (5):
>   clk: tegra: Add AHB DMA clock entry
>   clk: tegra: Bump SCLK clock rate to 216MHz on Tegra20
>   dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller
>   dmaengine: Add driver for NVIDIA Tegra AHB DMA controller
>   ARM: dts: tegra: Add AHB DMA controller nodes

I don't think they are dependent, so consider sending them separately

-- 
~Vinod

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 28.09.2017 12:29, Vinod Koul wrote:
> On Tue, Sep 26, 2017 at 02:22:05AM +0300, Dmitry Osipenko wrote:
> 
>> +config TEGRA20_AHB_DMA
>> +	tristate "NVIDIA Tegra20 AHB DMA support"
>> +	depends on ARCH_TEGRA
> 
> Can we add COMPILE_TEST, helps me compile drivers
> 

Good point.

>> +#include <linux/clk.h>
>> +#include <linux/delay.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/io.h>
>> +#include <linux/module.h>
>> +#include <linux/of_device.h>
>> +#include <linux/of_dma.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/reset.h>
>> +#include <linux/slab.h>
>> +#include <linux/spinlock.h>
> 
> no vchan.h, so i presume we are not using that here, any reason why?
> 

Jon Hunter asked the same question, I already reworked driver to use the
virt-dma. Turned out it is a really neat helper, -100 lines of driver code.

>> +
>> +#include "dmaengine.h"
>> +
>> +#define TEGRA_AHBDMA_CMD			0x0
>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>> +
>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
> 
> GENMASK() ?
> 

Okay.

>> +static void tegra_ahbdma_tasklet(unsigned long data)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>> +
>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>> +
>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>> +		kfree(tx);
> 
> lot of code here can be reduced if we use vchan
> 

+1

>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>> +					struct dma_chan *chan,
>> +					dma_addr_t buf_addr,
>> +					size_t buf_len,
>> +					size_t period_len,
>> +					enum dma_transfer_direction dir,
>> +					unsigned long flags)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	/* unimplemented */
>> +	if (buf_len != period_len || buf_len > SZ_64K)
>> +		return NULL;
>> +
>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>> +	if (!tx)
>> +		return NULL;
>> +
>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>> +
>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>> +	tx->mem_paddr		= buf_addr;
>> +	tx->size		= buf_len;
>> +	tx->flags		= flags;
>> +	tx->cyclic		= true;
>> +	tx->dir			= dir;
>> +
>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
> 
> why not precalulcate the register settings here. While submitting you are in
> hot path keeping dmaengine idle so faster you can submit, better the perf
> 

I may argue that the perf impact isn't measurable, but I agree that
precalculated register value would be a bit cleaner. Thanks for the suggestion.

>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +	unsigned long flags;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>> +
>> +	if (completion_done(&ahbdma_chan->idling)) {
>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>> +					      struct tegra_ahbdma_tx_desc,
>> +					      node);
>> +		if (tx) {
>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
> 
> what is chan is already running?
> 

It can't run here, we just checked whether it is idling. That would be a HW bug.

>> +			reinit_completion(&ahbdma_chan->idling);
>> +		}
>> +	}
>> +
>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>> +}
>> +
>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>> +					      dma_cookie_t cookie,
>> +					      struct dma_tx_state *state)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	enum dma_status cookie_status;
>> +	unsigned long flags;
>> +	size_t residual;
>> +	u32 status;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>> +	if (cookie_status != DMA_COMPLETE) {
> 
> residue can be NULL so check it before proceeding ahead
> 

Yeah, I noticed it too and fixed it in the upcoming V2 yesterday.

>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>> +			       struct dma_slave_config *sconfig)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	enum dma_transfer_direction dir = sconfig->direction;
>> +	u32 burst, ahb_seq, ahb_addr;
>> +
>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +		return -EINVAL;
>> +
>> +	if (dir == DMA_DEV_TO_MEM) {
>> +		burst    = sconfig->src_maxburst;
>> +		ahb_addr = sconfig->src_addr;
>> +	} else {
>> +		burst    = sconfig->dst_maxburst;
>> +		ahb_addr = sconfig->dst_addr;
>> +	}
>> +
>> +	switch (burst) {
>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
> 
> pls make this statement and break on subsequent lines, readablity matters
> 

Okay.

>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>> +
>> +	writel_relaxed(ahb_seq,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>> +
>> +	writel_relaxed(ahb_addr,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> 
> oh no, you don't write to HW here. This can be called anytime when you have
> txn running! You should save these and use them in prep_ calls.
> 

Okay.

>> +static int tegra_ahbdma_remove(struct platform_device *pdev)
>> +{
>> +	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
>> +
>> +	of_dma_controller_free(pdev->dev.of_node);
>> +	dma_async_device_unregister(&tdma->dma_dev);
>> +	clk_disable_unprepare(tdma->clk);
> 
> not ensuring tasklets are killed and irq is freed so no more tasklets can
> run? I think that needs to be done...
> 

Already fixed in V2 by using vchan_synchronize() that kills tasklet in
tegra_ahbdma_synchronize(). DMA core invokes synchronization upon channels
resource freeing.

>> +MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
>> +MODULE_AUTHOR("Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>");
>> +MODULE_LICENSE("GPL");
> 
> MODULE_ALIAS?
> 

Not needed, driver is "OF-only". It's default alias is "tegra20-ahb-dma".

-- 
Dmitry
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 28.09.2017 12:29, Vinod Koul wrote:
> On Tue, Sep 26, 2017 at 02:22:05AM +0300, Dmitry Osipenko wrote:
> 
>> +config TEGRA20_AHB_DMA
>> +	tristate "NVIDIA Tegra20 AHB DMA support"
>> +	depends on ARCH_TEGRA
> 
> Can we add COMPILE_TEST, helps me compile drivers
> 

Good point.

>> +#include <linux/clk.h>
>> +#include <linux/delay.h>
>> +#include <linux/interrupt.h>
>> +#include <linux/io.h>
>> +#include <linux/module.h>
>> +#include <linux/of_device.h>
>> +#include <linux/of_dma.h>
>> +#include <linux/platform_device.h>
>> +#include <linux/reset.h>
>> +#include <linux/slab.h>
>> +#include <linux/spinlock.h>
> 
> no vchan.h, so i presume we are not using that here, any reason why?
> 

Jon Hunter asked the same question, I already reworked driver to use the
virt-dma. Turned out it is a really neat helper, -100 lines of driver code.

>> +
>> +#include "dmaengine.h"
>> +
>> +#define TEGRA_AHBDMA_CMD			0x0
>> +#define TEGRA_AHBDMA_CMD_ENABLE			BIT(31)
>> +
>> +#define TEGRA_AHBDMA_IRQ_ENB_MASK		0x20
>> +#define TEGRA_AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_BASE(ch)		(0x1000 + (ch) * 0x20)
>> +
>> +#define TEGRA_AHBDMA_CHANNEL_CSR		0x0
>> +#define TEGRA_AHBDMA_CHANNEL_ADDR_WRAP		BIT(18)
>> +#define TEGRA_AHBDMA_CHANNEL_FLOW		BIT(24)
>> +#define TEGRA_AHBDMA_CHANNEL_ONCE		BIT(26)
>> +#define TEGRA_AHBDMA_CHANNEL_DIR_TO_XMB		BIT(27)
>> +#define TEGRA_AHBDMA_CHANNEL_IE_EOC		BIT(30)
>> +#define TEGRA_AHBDMA_CHANNEL_ENABLE		BIT(31)
>> +#define TEGRA_AHBDMA_CHANNEL_REQ_SEL_SHIFT	16
>> +#define TEGRA_AHBDMA_CHANNEL_WCOUNT_MASK	0xFFFC
> 
> GENMASK() ?
> 

Okay.

>> +static void tegra_ahbdma_tasklet(unsigned long data)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx = (struct tegra_ahbdma_tx_desc *)data;
>> +	struct dma_async_tx_descriptor *desc = &tx->desc;
>> +
>> +	dmaengine_desc_get_callback_invoke(desc, NULL);
>> +
>> +	if (!tx->cyclic && !dmaengine_desc_test_reuse(desc))
>> +		kfree(tx);
> 
> lot of code here can be reduced if we use vchan
> 

+1

>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep_dma_cyclic(
>> +					struct dma_chan *chan,
>> +					dma_addr_t buf_addr,
>> +					size_t buf_len,
>> +					size_t period_len,
>> +					enum dma_transfer_direction dir,
>> +					unsigned long flags)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +
>> +	/* unimplemented */
>> +	if (buf_len != period_len || buf_len > SZ_64K)
>> +		return NULL;
>> +
>> +	tx = kzalloc(sizeof(*tx), GFP_KERNEL);
>> +	if (!tx)
>> +		return NULL;
>> +
>> +	dma_async_tx_descriptor_init(&tx->desc, chan);
>> +
>> +	tx->desc.tx_submit	= tegra_ahbdma_tx_submit;
>> +	tx->mem_paddr		= buf_addr;
>> +	tx->size		= buf_len;
>> +	tx->flags		= flags;
>> +	tx->cyclic		= true;
>> +	tx->dir			= dir;
>> +
>> +	tasklet_init(&tx->tasklet, tegra_ahbdma_tasklet, (unsigned long)tx);
> 
> why not precalulcate the register settings here. While submitting you are in
> hot path keeping dmaengine idle so faster you can submit, better the perf
> 

I may argue that the perf impact isn't measurable, but I agree that
precalculated register value would be a bit cleaner. Thanks for the suggestion.

>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	struct list_head *entry, *tmp;
>> +	unsigned long flags;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	list_for_each_safe(entry, tmp, &ahbdma_chan->pending_list)
>> +		list_move_tail(entry, &ahbdma_chan->active_list);
>> +
>> +	if (completion_done(&ahbdma_chan->idling)) {
>> +		tx = list_first_entry_or_null(&ahbdma_chan->active_list,
>> +					      struct tegra_ahbdma_tx_desc,
>> +					      node);
>> +		if (tx) {
>> +			tegra_ahbdma_submit_tx(ahbdma_chan, tx);
> 
> what is chan is already running?
> 

It can't run here, we just checked whether it is idling. That would be a HW bug.

>> +			reinit_completion(&ahbdma_chan->idling);
>> +		}
>> +	}
>> +
>> +	spin_unlock_irqrestore(&ahbdma_chan->lock, flags);
>> +}
>> +
>> +static enum dma_status tegra_ahbdma_tx_status(struct dma_chan *chan,
>> +					      dma_cookie_t cookie,
>> +					      struct dma_tx_state *state)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	enum dma_status cookie_status;
>> +	unsigned long flags;
>> +	size_t residual;
>> +	u32 status;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->lock, flags);
>> +
>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>> +	if (cookie_status != DMA_COMPLETE) {
> 
> residue can be NULL so check it before proceeding ahead
> 

Yeah, I noticed it too and fixed it in the upcoming V2 yesterday.

>> +static int tegra_ahbdma_config(struct dma_chan *chan,
>> +			       struct dma_slave_config *sconfig)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	enum dma_transfer_direction dir = sconfig->direction;
>> +	u32 burst, ahb_seq, ahb_addr;
>> +
>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +		return -EINVAL;
>> +
>> +	if (dir == DMA_DEV_TO_MEM) {
>> +		burst    = sconfig->src_maxburst;
>> +		ahb_addr = sconfig->src_addr;
>> +	} else {
>> +		burst    = sconfig->dst_maxburst;
>> +		ahb_addr = sconfig->dst_addr;
>> +	}
>> +
>> +	switch (burst) {
>> +	case 1: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_1; break;
>> +	case 4: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_4; break;
>> +	case 8: burst = TEGRA_AHBDMA_CHANNEL_AHB_BURST_8; break;
> 
> pls make this statement and break on subsequent lines, readablity matters
> 

Okay.

>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>> +
>> +	writel_relaxed(ahb_seq,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>> +
>> +	writel_relaxed(ahb_addr,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> 
> oh no, you don't write to HW here. This can be called anytime when you have
> txn running! You should save these and use them in prep_ calls.
> 

Okay.

>> +static int tegra_ahbdma_remove(struct platform_device *pdev)
>> +{
>> +	struct tegra_ahbdma *tdma = platform_get_drvdata(pdev);
>> +
>> +	of_dma_controller_free(pdev->dev.of_node);
>> +	dma_async_device_unregister(&tdma->dma_dev);
>> +	clk_disable_unprepare(tdma->clk);
> 
> not ensuring tasklets are killed and irq is freed so no more tasklets can
> run? I think that needs to be done...
> 

Already fixed in V2 by using vchan_synchronize() that kills tasklet in
tegra_ahbdma_synchronize(). DMA core invokes synchronization upon channels
resource freeing.

>> +MODULE_DESCRIPTION("NVIDIA Tegra AHB DMA Controller driver");
>> +MODULE_AUTHOR("Dmitry Osipenko <digetx@gmail.com>");
>> +MODULE_LICENSE("GPL");
> 
> MODULE_ALIAS?
> 

Not needed, driver is "OF-only". It's default alias is "tegra20-ahb-dma".

-- 
Dmitry

Re: [PATCH v1 0/5] NVIDIA Tegra AHB DMA controller driver

[Dmitry Osipenko]

On 28.09.2017 12:31, Vinod Koul wrote:
> On Tue, Sep 26, 2017 at 02:22:01AM +0300, Dmitry Osipenko wrote:
>> NVIDIA Tegra20/30 SoC's have AHB DMA controller. It has 4 DMA channels,
>> supports AHB <-> Memory and Memory <-> Memory transfers, slave / master
>> modes. This driver is primarily supposed to be used by gpu/host1x in a
>> master mode, performing 3D HW context stores.
>>
>> Dmitry Osipenko (5):
>>   clk: tegra: Add AHB DMA clock entry
>>   clk: tegra: Bump SCLK clock rate to 216MHz on Tegra20
>>   dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller
>>   dmaengine: Add driver for NVIDIA Tegra AHB DMA controller
>>   ARM: dts: tegra: Add AHB DMA controller nodes
> 
> I don't think they are dependent, so consider sending them separately
> 

Well, they are dependent in a sense of making driver usable. Only the "SCLK rate
bump" patch isn't strictly needed.

Splitting this series won't cause building failures, but all pieces should be in
place for the working driver. So I suppose it is okay if clk patches would get
in earlier than the others, I'll split the series.

Thank you for the review.

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 28.09.2017 12:29, Vinod Koul wrote:
>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>> +
>> +	writel_relaxed(ahb_seq,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>> +
>> +	writel_relaxed(ahb_addr,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> 
> oh no, you don't write to HW here. This can be called anytime when you have
> txn running! You should save these and use them in prep_ calls.
> 

BTW, some of the DMA drivers have exactly the same problem. I now see that it is
actually documented explicitly in provider.txt, but that's inconsistent across
the actual drivers.

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 28.09.2017 12:29, Vinod Koul wrote:
>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>> +
>> +	writel_relaxed(ahb_seq,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>> +
>> +	writel_relaxed(ahb_addr,
>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> 
> oh no, you don't write to HW here. This can be called anytime when you have
> txn running! You should save these and use them in prep_ calls.
> 

BTW, some of the DMA drivers have exactly the same problem. I now see that it is
actually documented explicitly in provider.txt, but that's inconsistent across
the actual drivers.

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 28.09.2017 17:06, Dmitry Osipenko wrote:
> On 28.09.2017 12:29, Vinod Koul wrote:
>>> +	default:
>>> +		return -EINVAL;
>>> +	}
>>> +
>>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>>> +
>>> +	writel_relaxed(ahb_seq,
>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>>> +
>>> +	writel_relaxed(ahb_addr,
>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>>
>> oh no, you don't write to HW here. This can be called anytime when you have
>> txn running! You should save these and use them in prep_ calls.
>>
> 
> BTW, some of the DMA drivers have exactly the same problem. I now see that it is
> actually documented explicitly in provider.txt, but that's inconsistent across
> the actual drivers.
> 

Also, shouldn't prep_ and dma_slave_config be protected with locking? I don't
see DMA core doing any locking and seems none of the drivers too.

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Vinod Koul]

On Thu, Sep 28, 2017 at 05:06:03PM +0300, Dmitry Osipenko wrote:
> On 28.09.2017 12:29, Vinod Koul wrote:
> >> +	default:
> >> +		return -EINVAL;
> >> +	}
> >> +
> >> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
> >> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
> >> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
> >> +
> >> +	writel_relaxed(ahb_seq,
> >> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
> >> +
> >> +	writel_relaxed(ahb_addr,
> >> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> > 
> > oh no, you don't write to HW here. This can be called anytime when you have
> > txn running! You should save these and use them in prep_ calls.
> > 
> 
> BTW, some of the DMA drivers have exactly the same problem. I now see that it is
> actually documented explicitly in provider.txt, but that's inconsistent across
> the actual drivers.

yeah they need to be fixed!

-- 
~Vinod

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Vinod Koul]

On Thu, Sep 28, 2017 at 05:35:59PM +0300, Dmitry Osipenko wrote:
> On 28.09.2017 17:06, Dmitry Osipenko wrote:
> > On 28.09.2017 12:29, Vinod Koul wrote:
> >>> +	default:
> >>> +		return -EINVAL;
> >>> +	}
> >>> +
> >>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
> >>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
> >>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
> >>> +
> >>> +	writel_relaxed(ahb_seq,
> >>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
> >>> +
> >>> +	writel_relaxed(ahb_addr,
> >>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> >>
> >> oh no, you don't write to HW here. This can be called anytime when you have
> >> txn running! You should save these and use them in prep_ calls.
> >>
> > 
> > BTW, some of the DMA drivers have exactly the same problem. I now see that it is
> > actually documented explicitly in provider.txt, but that's inconsistent across
> > the actual drivers.
> > 
> 
> Also, shouldn't prep_ and dma_slave_config be protected with locking? I don't
> see DMA core doing any locking and seems none of the drivers too.

In prep when you modify the list yes (with vchan I suspect that maybe taken
care), but in general yes driver needs to do that

-- 
~Vinod
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Vinod Koul]

On Thu, Sep 28, 2017 at 05:35:59PM +0300, Dmitry Osipenko wrote:
> On 28.09.2017 17:06, Dmitry Osipenko wrote:
> > On 28.09.2017 12:29, Vinod Koul wrote:
> >>> +	default:
> >>> +		return -EINVAL;
> >>> +	}
> >>> +
> >>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
> >>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
> >>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
> >>> +
> >>> +	writel_relaxed(ahb_seq,
> >>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
> >>> +
> >>> +	writel_relaxed(ahb_addr,
> >>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
> >>
> >> oh no, you don't write to HW here. This can be called anytime when you have
> >> txn running! You should save these and use them in prep_ calls.
> >>
> > 
> > BTW, some of the DMA drivers have exactly the same problem. I now see that it is
> > actually documented explicitly in provider.txt, but that's inconsistent across
> > the actual drivers.
> > 
> 
> Also, shouldn't prep_ and dma_slave_config be protected with locking? I don't
> see DMA core doing any locking and seems none of the drivers too.

In prep when you modify the list yes (with vchan I suspect that maybe taken
care), but in general yes driver needs to do that

-- 
~Vinod

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 28.09.2017 19:22, Vinod Koul wrote:
> On Thu, Sep 28, 2017 at 05:35:59PM +0300, Dmitry Osipenko wrote:
>> On 28.09.2017 17:06, Dmitry Osipenko wrote:
>>> On 28.09.2017 12:29, Vinod Koul wrote:
>>>>> +	default:
>>>>> +		return -EINVAL;
>>>>> +	}
>>>>> +
>>>>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>>>>> +
>>>>> +	writel_relaxed(ahb_seq,
>>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>>>>> +
>>>>> +	writel_relaxed(ahb_addr,
>>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>>>>
>>>> oh no, you don't write to HW here. This can be called anytime when you have
>>>> txn running! You should save these and use them in prep_ calls.
>>>>
>>>
>>> BTW, some of the DMA drivers have exactly the same problem. I now see that it is
>>> actually documented explicitly in provider.txt, but that's inconsistent across
>>> the actual drivers.
>>>
>>
>> Also, shouldn't prep_ and dma_slave_config be protected with locking? I don't
>> see DMA core doing any locking and seems none of the drivers too.
> 
> In prep when you modify the list yes (with vchan I suspect that maybe taken
> care), but in general yes driver needs to do that
> 

I meant that one CPU could modify channels config, while other CPU is preparing
the new TX using config that is in process of the modification. On the other
hand, this looks like something that DMA client should take care of.

-- 
Dmitry

Re: [PATCH v1 4/5] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 28.09.2017 19:22, Vinod Koul wrote:
> On Thu, Sep 28, 2017 at 05:35:59PM +0300, Dmitry Osipenko wrote:
>> On 28.09.2017 17:06, Dmitry Osipenko wrote:
>>> On 28.09.2017 12:29, Vinod Koul wrote:
>>>>> +	default:
>>>>> +		return -EINVAL;
>>>>> +	}
>>>>> +
>>>>> +	ahb_seq  = burst << TEGRA_AHBDMA_CHANNEL_AHB_BURST_SHIFT;
>>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_ADDR_WRAP;
>>>>> +	ahb_seq |= TEGRA_AHBDMA_CHANNEL_INTR_ENB;
>>>>> +
>>>>> +	writel_relaxed(ahb_seq,
>>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_SEQ);
>>>>> +
>>>>> +	writel_relaxed(ahb_addr,
>>>>> +		       ahbdma_chan->regs + TEGRA_AHBDMA_CHANNEL_AHB_PTR);
>>>>
>>>> oh no, you don't write to HW here. This can be called anytime when you have
>>>> txn running! You should save these and use them in prep_ calls.
>>>>
>>>
>>> BTW, some of the DMA drivers have exactly the same problem. I now see that it is
>>> actually documented explicitly in provider.txt, but that's inconsistent across
>>> the actual drivers.
>>>
>>
>> Also, shouldn't prep_ and dma_slave_config be protected with locking? I don't
>> see DMA core doing any locking and seems none of the drivers too.
> 
> In prep when you modify the list yes (with vchan I suspect that maybe taken
> care), but in general yes driver needs to do that
> 

I meant that one CPU could modify channels config, while other CPU is preparing
the new TX using config that is in process of the modification. On the other
hand, this looks like something that DMA client should take care of.

-- 
Dmitry

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Stephen Warren]

On 09/27/2017 02:34 AM, Jon Hunter wrote:
> 
> On 27/09/17 02:57, Dmitry Osipenko wrote:
>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>
>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>> on Tegra20/30 SoC's.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>> ---
>>>>   .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>   1 file changed, 23 insertions(+)
>>>>   create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>> new file mode 100644
>>>> index 000000000000..2af9aa76ae11
>>>> --- /dev/null
>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>> @@ -0,0 +1,23 @@
>>>> +* NVIDIA Tegra AHB DMA controller
>>>> +
>>>> +Required properties:
>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>> +- reg:		Should contain registers base address and length.
>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>> +		documentation, in particular AHB DMA channel control register
>>>> +		REQ_SEL field.
>>>
>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>
>>
>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>> software to decide what trigger to select. So it shouldn't be in the binding.
> 
> I think it could be, if say a board wanted a GPIO to trigger a transfer.
> 
>> And I think the same applies to requester... any objections?
> 
> Well, the REQ_SEL should definitely be in the binding.
> 
> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
> like we never bothered with it for the APB DMA and so maybe no ones uses
> this.

I don't think TRIG_SEL should be in the binding, at least at present. 
While TRIG_SEL certainly is something used to configure the transfer, I 
believe the semantics of the current DMA binding only cover DMA 
transfers that are initiated when SW desires, rather than being a 
combination of after SW programs the transfer plus some other HW event. 
So, we always use a default/hard-coded TRIG_SEL value. As such, there's 
no need for a TRIG_SEL value in DT. There's certainly no known use-case 
that requires a non-default TRIG_SEL value at present. We could add an 
extra #dma-cells value later if we find a use for it, and the semantics 
of that use-case make sense to add it to the DMA specifier, rather than 
some other separate higher-level property/driver/...

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Stephen Warren]

On 09/27/2017 02:34 AM, Jon Hunter wrote:
> 
> On 27/09/17 02:57, Dmitry Osipenko wrote:
>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>
>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>> on Tegra20/30 SoC's.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>> ---
>>>>   .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>>>   1 file changed, 23 insertions(+)
>>>>   create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>
>>>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>> new file mode 100644
>>>> index 000000000000..2af9aa76ae11
>>>> --- /dev/null
>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>> @@ -0,0 +1,23 @@
>>>> +* NVIDIA Tegra AHB DMA controller
>>>> +
>>>> +Required properties:
>>>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>>>> +- reg:		Should contain registers base address and length.
>>>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>>>> +- clocks:	Should contain one entry, DMA controller clock.
>>>> +- resets :	Should contain one entry, DMA controller reset.
>>>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>>>> +		for the peripheral. For more details consult the Tegra TRM's
>>>> +		documentation, in particular AHB DMA channel control register
>>>> +		REQ_SEL field.
>>>
>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>
>>
>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>> software to decide what trigger to select. So it shouldn't be in the binding.
> 
> I think it could be, if say a board wanted a GPIO to trigger a transfer.
> 
>> And I think the same applies to requester... any objections?
> 
> Well, the REQ_SEL should definitely be in the binding.
> 
> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
> like we never bothered with it for the APB DMA and so maybe no ones uses
> this.

I don't think TRIG_SEL should be in the binding, at least at present. 
While TRIG_SEL certainly is something used to configure the transfer, I 
believe the semantics of the current DMA binding only cover DMA 
transfers that are initiated when SW desires, rather than being a 
combination of after SW programs the transfer plus some other HW event. 
So, we always use a default/hard-coded TRIG_SEL value. As such, there's 
no need for a TRIG_SEL value in DT. There's certainly no known use-case 
that requires a non-default TRIG_SEL value at present. We could add an 
extra #dma-cells value later if we find a use for it, and the semantics 
of that use-case make sense to add it to the DMA specifier, rather than 
some other separate higher-level property/driver/...

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 29.09.2017 22:30, Stephen Warren wrote:
> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>
>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>
>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>> on Tegra20/30 SoC's.
>>>>>
>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>> ---
>>>>>   .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>> ++++++++++++++++++++++
>>>>>   1 file changed, 23 insertions(+)
>>>>>   create mode 100644
>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>
>>>>> diff --git
>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>> new file mode 100644
>>>>> index 000000000000..2af9aa76ae11
>>>>> --- /dev/null
>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>> @@ -0,0 +1,23 @@
>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>> +
>>>>> +Required properties:
>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>> +- reg:        Should contain registers base address and length.
>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select value
>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>> +        documentation, in particular AHB DMA channel control register
>>>>> +        REQ_SEL field.
>>>>
>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>
>>>
>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>
>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>
>>> And I think the same applies to requester... any objections?
>>
>> Well, the REQ_SEL should definitely be in the binding.
>>
>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>> like we never bothered with it for the APB DMA and so maybe no ones uses
>> this.
> 
> I don't think TRIG_SEL should be in the binding, at least at present. While
> TRIG_SEL certainly is something used to configure the transfer, I believe the
> semantics of the current DMA binding only cover DMA transfers that are initiated
> when SW desires, rather than being a combination of after SW programs the
> transfer plus some other HW event. So, we always use a default/hard-coded
> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
> certainly no known use-case that requires a non-default TRIG_SEL value at
> present. We could add an extra #dma-cells value later if we find a use for it,
> and the semantics of that use-case make sense to add it to the DMA specifier,
> rather than some other separate higher-level property/driver/...

Thank you for the comment. If we'd want to extend the binding further with the
trigger, how to differentiate trigger from the requester in a case of a single
#data-cell?

Of course realistically a chance that the further extension would be needed is
very-very low, so we may defer the efforts to solve that question and for now
make driver aware of the potential #dma-cells extension.

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Stephen Warren]

On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
> On 29.09.2017 22:30, Stephen Warren wrote:
>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>
>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>
>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>> on Tegra20/30 SoC's.
>>>>>>
>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>> ---
>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>> ++++++++++++++++++++++
>>>>>>    1 file changed, 23 insertions(+)
>>>>>>    create mode 100644
>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>
>>>>>> diff --git
>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>> new file mode 100644
>>>>>> index 000000000000..2af9aa76ae11
>>>>>> --- /dev/null
>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>> @@ -0,0 +1,23 @@
>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>> +
>>>>>> +Required properties:
>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>> +- reg:        Should contain registers base address and length.
>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select value
>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>> +        REQ_SEL field.
>>>>>
>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>
>>>>
>>>> Actually, DMA transfer trigger isn't related a hardware description. It's up to
>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>
>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>
>>>> And I think the same applies to requester... any objections?
>>>
>>> Well, the REQ_SEL should definitely be in the binding.
>>>
>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>> this.
>>
>> I don't think TRIG_SEL should be in the binding, at least at present. While
>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>> semantics of the current DMA binding only cover DMA transfers that are initiated
>> when SW desires, rather than being a combination of after SW programs the
>> transfer plus some other HW event. So, we always use a default/hard-coded
>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>> certainly no known use-case that requires a non-default TRIG_SEL value at
>> present. We could add an extra #dma-cells value later if we find a use for it,
>> and the semantics of that use-case make sense to add it to the DMA specifier,
>> rather than some other separate higher-level property/driver/...
> 
> Thank you for the comment. If we'd want to extend the binding further with the
> trigger, how to differentiate trigger from the requester in a case of a single
> #data-cell?
> 
> Of course realistically a chance that the further extension would be needed is
> very-very low, so we may defer the efforts to solve that question and for now
> make driver aware of the potential #dma-cells extension.

The request selector cell isn't optional, so is always present. If we 
later add an optional trig_sel cell, we'll either have:

#dma-cells=<1>: req_sel

or:

#dma-cells=<2>: req_sel, trig_sel

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 02.10.2017 20:05, Stephen Warren wrote:
> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>> On 29.09.2017 22:30, Stephen Warren wrote:
>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>
>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>
>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>> on Tegra20/30 SoC's.
>>>>>>>
>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>> ---
>>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>> ++++++++++++++++++++++
>>>>>>>    1 file changed, 23 insertions(+)
>>>>>>>    create mode 100644
>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>
>>>>>>> diff --git
>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>> new file mode 100644
>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>> --- /dev/null
>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>> @@ -0,0 +1,23 @@
>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>> +
>>>>>>> +Required properties:
>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>> value
>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>> +        REQ_SEL field.
>>>>>>
>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>
>>>>>
>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>> up to
>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>
>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>
>>>>> And I think the same applies to requester... any objections?
>>>>
>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>
>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>> this.
>>>
>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>> when SW desires, rather than being a combination of after SW programs the
>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>> rather than some other separate higher-level property/driver/...
>>
>> Thank you for the comment. If we'd want to extend the binding further with the
>> trigger, how to differentiate trigger from the requester in a case of a single
>> #data-cell?
>>
>> Of course realistically a chance that the further extension would be needed is
>> very-very low, so we may defer the efforts to solve that question and for now
>> make driver aware of the potential #dma-cells extension.
> 
> The request selector cell isn't optional, so is always present. If we later add
> an optional trig_sel cell, we'll either have:
> 
> #dma-cells=<1>: req_sel
> 
> or:
> 
> #dma-cells=<2>: req_sel, trig_sel

Why request sel. couldn't be optional? Could you please elaborate a bit more?

I think possible options are:

#dma-cells=<1>: req_sel
#dma-cells=<1>: trig_sel
#dma-cells=<2>: req_sel, trig_sel

The only difference between request and trigger is that trigger issues the whole
transfer, while request only a single burst. Isn't it possible to have a case in
HW for the "trigger-only" option? If not or it's a rareness, then I agree that
REQ_SEL must be mandatory.

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 03/10/17 00:02, Dmitry Osipenko wrote:
> On 02.10.2017 20:05, Stephen Warren wrote:
>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>
>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>
>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>> ---
>>>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>> ++++++++++++++++++++++
>>>>>>>>    1 file changed, 23 insertions(+)
>>>>>>>>    create mode 100644
>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>
>>>>>>>> diff --git
>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> new file mode 100644
>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>> --- /dev/null
>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>> +
>>>>>>>> +Required properties:
>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>> value
>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>> +        REQ_SEL field.
>>>>>>>
>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>
>>>>>>
>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>> up to
>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>
>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>
>>>>>> And I think the same applies to requester... any objections?
>>>>>
>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>
>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>> this.
>>>>
>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>> when SW desires, rather than being a combination of after SW programs the
>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>> rather than some other separate higher-level property/driver/...
>>>
>>> Thank you for the comment. If we'd want to extend the binding further with the
>>> trigger, how to differentiate trigger from the requester in a case of a single
>>> #data-cell?
>>>
>>> Of course realistically a chance that the further extension would be needed is
>>> very-very low, so we may defer the efforts to solve that question and for now
>>> make driver aware of the potential #dma-cells extension.
>>
>> The request selector cell isn't optional, so is always present. If we later add
>> an optional trig_sel cell, we'll either have:
>>
>> #dma-cells=<1>: req_sel
>>
>> or:
>>
>> #dma-cells=<2>: req_sel, trig_sel
> 
> Why request sel. couldn't be optional? Could you please elaborate a bit more?
> 
> I think possible options are:
> 
> #dma-cells=<1>: req_sel
> #dma-cells=<1>: trig_sel

With the above, how would you know that it is the req_sel or trig_sel
that is specified?

> #dma-cells=<2>: req_sel, trig_sel
> 
> The only difference between request and trigger is that trigger issues the whole
> transfer, while request only a single burst. Isn't it possible to have a case in
> HW for the "trigger-only" option? If not or it's a rareness, then I agree that
> REQ_SEL must be mandatory.

I think that what Stephen is proposing is that for now we go with
'#dma-cells=<1>' and if we ever need to support the trigger cell we
could add support for '#dma-cells=<2>'. So with this proposal the
'req_sel' would always be required for both '#dma-cells=<1>' and
'#dma-cells=<2>'. Even if the req_sel is not actually used but the
'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
value (eg. -1) where we know to ignore it.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 03/10/17 00:02, Dmitry Osipenko wrote:
> On 02.10.2017 20:05, Stephen Warren wrote:
>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>
>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>
>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>> ---
>>>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>> ++++++++++++++++++++++
>>>>>>>>    1 file changed, 23 insertions(+)
>>>>>>>>    create mode 100644
>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>
>>>>>>>> diff --git
>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> new file mode 100644
>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>> --- /dev/null
>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>> +
>>>>>>>> +Required properties:
>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>> value
>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>> +        REQ_SEL field.
>>>>>>>
>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>
>>>>>>
>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>> up to
>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>
>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>
>>>>>> And I think the same applies to requester... any objections?
>>>>>
>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>
>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>> this.
>>>>
>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>> when SW desires, rather than being a combination of after SW programs the
>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>> rather than some other separate higher-level property/driver/...
>>>
>>> Thank you for the comment. If we'd want to extend the binding further with the
>>> trigger, how to differentiate trigger from the requester in a case of a single
>>> #data-cell?
>>>
>>> Of course realistically a chance that the further extension would be needed is
>>> very-very low, so we may defer the efforts to solve that question and for now
>>> make driver aware of the potential #dma-cells extension.
>>
>> The request selector cell isn't optional, so is always present. If we later add
>> an optional trig_sel cell, we'll either have:
>>
>> #dma-cells=<1>: req_sel
>>
>> or:
>>
>> #dma-cells=<2>: req_sel, trig_sel
> 
> Why request sel. couldn't be optional? Could you please elaborate a bit more?
> 
> I think possible options are:
> 
> #dma-cells=<1>: req_sel
> #dma-cells=<1>: trig_sel

With the above, how would you know that it is the req_sel or trig_sel
that is specified?

> #dma-cells=<2>: req_sel, trig_sel
> 
> The only difference between request and trigger is that trigger issues the whole
> transfer, while request only a single burst. Isn't it possible to have a case in
> HW for the "trigger-only" option? If not or it's a rareness, then I agree that
> REQ_SEL must be mandatory.

I think that what Stephen is proposing is that for now we go with
'#dma-cells=<1>' and if we ever need to support the trigger cell we
could add support for '#dma-cells=<2>'. So with this proposal the
'req_sel' would always be required for both '#dma-cells=<1>' and
'#dma-cells=<2>'. Even if the req_sel is not actually used but the
'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
value (eg. -1) where we know to ignore it.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 03/10/17 00:02, Dmitry Osipenko wrote:
> On 02.10.2017 20:05, Stephen Warren wrote:
>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>
>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that pres=
ents
>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>
>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>> ---
>>>>>>>> =C2=A0=C2=A0 .../bindings/dma/nvidia,tegra20-ahbdma.txt=C2=A0=C2=
=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 | 23
>>>>>>>> ++++++++++++++++++++++
>>>>>>>> =C2=A0=C2=A0 1 file changed, 23 insertions(+)
>>>>>>>> =C2=A0=C2=A0 create mode 100644
>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>
>>>>>>>> diff --git
>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>> new file mode 100644
>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>> --- /dev/null
>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.=
txt
>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>> +
>>>>>>>> +Required properties:
>>>>>>>> +- compatible:=C2=A0=C2=A0=C2=A0 Must be "nvidia,tegra20-ahbdma"
>>>>>>>> +- reg:=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 Should contain r=
egisters base address and length.
>>>>>>>> +- interrupts:=C2=A0=C2=A0=C2=A0 Should contain one entry, DMA con=
troller interrupt.
>>>>>>>> +- clocks:=C2=A0=C2=A0=C2=A0 Should contain one entry, DMA control=
ler clock.
>>>>>>>> +- resets :=C2=A0=C2=A0=C2=A0 Should contain one entry, DMA contro=
ller reset.
>>>>>>>> +- #dma-cells:=C2=A0=C2=A0=C2=A0 Should be <1>. The cell represent=
s DMA request select
>>>>>>>> value
>>>>>>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 for the peripheral. Fo=
r more details consult the Tegra TRM's
>>>>>>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 documentation, in part=
icular AHB DMA channel control register
>>>>>>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 REQ_SEL field.
>>>>>>>
>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as we=
ll?
>>>>>>>
>>>>>>
>>>>>> Actually, DMA transfer trigger isn't related a hardware description.=
 It's
>>>>>> up to
>>>>>> software to decide what trigger to select. So it shouldn't be in the=
 binding.
>>>>>
>>>>> I think it could be, if say a board wanted a GPIO to trigger a transf=
er.
>>>>>
>>>>>> And I think the same applies to requester... any objections?
>>>>>
>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>
>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>> like we never bothered with it for the APB DMA and so maybe no ones u=
ses
>>>>> this.
>>>>
>>>> I don't think TRIG_SEL should be in the binding, at least at present. =
While
>>>> TRIG_SEL certainly is something used to configure the transfer, I beli=
eve the
>>>> semantics of the current DMA binding only cover DMA transfers that are=
 initiated
>>>> when SW desires, rather than being a combination of after SW programs =
the
>>>> transfer plus some other HW event. So, we always use a default/hard-co=
ded
>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. T=
here's
>>>> certainly no known use-case that requires a non-default TRIG_SEL value=
 at
>>>> present. We could add an extra #dma-cells value later if we find a use=
 for it,
>>>> and the semantics of that use-case make sense to add it to the DMA spe=
cifier,
>>>> rather than some other separate higher-level property/driver/...
>>>
>>> Thank you for the comment. If we'd want to extend the binding further w=
ith the
>>> trigger, how to differentiate trigger from the requester in a case of a=
 single
>>> #data-cell?
>>>
>>> Of course realistically a chance that the further extension would be ne=
eded is
>>> very-very low, so we may defer the efforts to solve that question and f=
or now
>>> make driver aware of the potential #dma-cells extension.
>>
>> The request selector cell isn't optional, so is always present. If we la=
ter add
>> an optional trig_sel cell, we'll either have:
>>
>> #dma-cells=3D<1>: req_sel
>>
>> or:
>>
>> #dma-cells=3D<2>: req_sel, trig_sel
>=20
> Why request sel. couldn't be optional? Could you please elaborate a bit m=
ore?
>=20
> I think possible options are:
>=20
> #dma-cells=3D<1>: req_sel
> #dma-cells=3D<1>: trig_sel

With the above, how would you know that it is the req_sel or trig_sel
that is specified?

> #dma-cells=3D<2>: req_sel, trig_sel
>=20
> The only difference between request and trigger is that trigger issues th=
e whole
> transfer, while request only a single burst. Isn't it possible to have a =
case in
> HW for the "trigger-only" option? If not or it's a rareness, then I agree=
 that
> REQ_SEL must be mandatory.

I think that what Stephen is proposing is that for now we go with
'#dma-cells=3D<1>' and if we ever need to support the trigger cell we
could add support for '#dma-cells=3D<2>'. So with this proposal the
'req_sel' would always be required for both '#dma-cells=3D<1>' and
'#dma-cells=3D<2>'. Even if the req_sel is not actually used but the
'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
value (eg. -1) where we know to ignore it.

Cheers
Jon

--=20
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 03.10.2017 13:32, Jon Hunter wrote:
> 
> 
> On 03/10/17 00:02, Dmitry Osipenko wrote:
>> On 02.10.2017 20:05, Stephen Warren wrote:
>>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>>
>>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>>
>>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
>>>>>>>>> ---
>>>>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>>> ++++++++++++++++++++++
>>>>>>>>>    1 file changed, 23 insertions(+)
>>>>>>>>>    create mode 100644
>>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>
>>>>>>>>> diff --git
>>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> new file mode 100644
>>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>>> --- /dev/null
>>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>>> +
>>>>>>>>> +Required properties:
>>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>>> value
>>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>>> +        REQ_SEL field.
>>>>>>>>
>>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>>
>>>>>>>
>>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>>> up to
>>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>>
>>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>>
>>>>>>> And I think the same applies to requester... any objections?
>>>>>>
>>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>>
>>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>>> this.
>>>>>
>>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>>> when SW desires, rather than being a combination of after SW programs the
>>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>>> rather than some other separate higher-level property/driver/...
>>>>
>>>> Thank you for the comment. If we'd want to extend the binding further with the
>>>> trigger, how to differentiate trigger from the requester in a case of a single
>>>> #data-cell?
>>>>
>>>> Of course realistically a chance that the further extension would be needed is
>>>> very-very low, so we may defer the efforts to solve that question and for now
>>>> make driver aware of the potential #dma-cells extension.
>>>
>>> The request selector cell isn't optional, so is always present. If we later add
>>> an optional trig_sel cell, we'll either have:
>>>
>>> #dma-cells=<1>: req_sel
>>>
>>> or:
>>>
>>> #dma-cells=<2>: req_sel, trig_sel
>>
>> Why request sel. couldn't be optional? Could you please elaborate a bit more?
>>
>> I think possible options are:
>>
>> #dma-cells=<1>: req_sel
>> #dma-cells=<1>: trig_sel
> 
> With the above, how would you know that it is the req_sel or trig_sel
> that is specified?
> 
>> #dma-cells=<2>: req_sel, trig_sel
>>
>> The only difference between request and trigger is that trigger issues the whole
>> transfer, while request only a single burst. Isn't it possible to have a case in
>> HW for the "trigger-only" option? If not or it's a rareness, then I agree that
>> REQ_SEL must be mandatory.
> 
> I think that what Stephen is proposing is that for now we go with
> '#dma-cells=<1>' and if we ever need to support the trigger cell we
> could add support for '#dma-cells=<2>'. So with this proposal the
> 'req_sel' would always be required for both '#dma-cells=<1>' and
> '#dma-cells=<2>'. Even if the req_sel is not actually used but the
> 'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
> value (eg. -1) where we know to ignore it.
> 

Okay, I see now. Thank you for the clarification, but then we should have that
pre-defined value declared in the binding?

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 03.10.2017 13:32, Jon Hunter wrote:
> 
> 
> On 03/10/17 00:02, Dmitry Osipenko wrote:
>> On 02.10.2017 20:05, Stephen Warren wrote:
>>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>>
>>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>>
>>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>>> ---
>>>>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>>> ++++++++++++++++++++++
>>>>>>>>>    1 file changed, 23 insertions(+)
>>>>>>>>>    create mode 100644
>>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>
>>>>>>>>> diff --git
>>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> new file mode 100644
>>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>>> --- /dev/null
>>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>>> +
>>>>>>>>> +Required properties:
>>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>>> value
>>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>>> +        REQ_SEL field.
>>>>>>>>
>>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>>
>>>>>>>
>>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>>> up to
>>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>>
>>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>>
>>>>>>> And I think the same applies to requester... any objections?
>>>>>>
>>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>>
>>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>>> this.
>>>>>
>>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>>> when SW desires, rather than being a combination of after SW programs the
>>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>>> rather than some other separate higher-level property/driver/...
>>>>
>>>> Thank you for the comment. If we'd want to extend the binding further with the
>>>> trigger, how to differentiate trigger from the requester in a case of a single
>>>> #data-cell?
>>>>
>>>> Of course realistically a chance that the further extension would be needed is
>>>> very-very low, so we may defer the efforts to solve that question and for now
>>>> make driver aware of the potential #dma-cells extension.
>>>
>>> The request selector cell isn't optional, so is always present. If we later add
>>> an optional trig_sel cell, we'll either have:
>>>
>>> #dma-cells=<1>: req_sel
>>>
>>> or:
>>>
>>> #dma-cells=<2>: req_sel, trig_sel
>>
>> Why request sel. couldn't be optional? Could you please elaborate a bit more?
>>
>> I think possible options are:
>>
>> #dma-cells=<1>: req_sel
>> #dma-cells=<1>: trig_sel
> 
> With the above, how would you know that it is the req_sel or trig_sel
> that is specified?
> 
>> #dma-cells=<2>: req_sel, trig_sel
>>
>> The only difference between request and trigger is that trigger issues the whole
>> transfer, while request only a single burst. Isn't it possible to have a case in
>> HW for the "trigger-only" option? If not or it's a rareness, then I agree that
>> REQ_SEL must be mandatory.
> 
> I think that what Stephen is proposing is that for now we go with
> '#dma-cells=<1>' and if we ever need to support the trigger cell we
> could add support for '#dma-cells=<2>'. So with this proposal the
> 'req_sel' would always be required for both '#dma-cells=<1>' and
> '#dma-cells=<2>'. Even if the req_sel is not actually used but the
> 'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
> value (eg. -1) where we know to ignore it.
> 

Okay, I see now. Thank you for the clarification, but then we should have that
pre-defined value declared in the binding?

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 03/10/17 13:07, Dmitry Osipenko wrote:
> On 03.10.2017 13:32, Jon Hunter wrote:
>>
>>
>> On 03/10/17 00:02, Dmitry Osipenko wrote:
>>> On 02.10.2017 20:05, Stephen Warren wrote:
>>>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>>>
>>>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>>>> ---
>>>>>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>>>> ++++++++++++++++++++++
>>>>>>>>>>    1 file changed, 23 insertions(+)
>>>>>>>>>>    create mode 100644
>>>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>>
>>>>>>>>>> diff --git
>>>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> new file mode 100644
>>>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>>>> --- /dev/null
>>>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>>>> +
>>>>>>>>>> +Required properties:
>>>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>>>> value
>>>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>>>> +        REQ_SEL field.
>>>>>>>>>
>>>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>>>
>>>>>>>>
>>>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>>>> up to
>>>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>>>
>>>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>>>
>>>>>>>> And I think the same applies to requester... any objections?
>>>>>>>
>>>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>>>
>>>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>>>> this.
>>>>>>
>>>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>>>> when SW desires, rather than being a combination of after SW programs the
>>>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>>>> rather than some other separate higher-level property/driver/...
>>>>>
>>>>> Thank you for the comment. If we'd want to extend the binding further with the
>>>>> trigger, how to differentiate trigger from the requester in a case of a single
>>>>> #data-cell?
>>>>>
>>>>> Of course realistically a chance that the further extension would be needed is
>>>>> very-very low, so we may defer the efforts to solve that question and for now
>>>>> make driver aware of the potential #dma-cells extension.
>>>>
>>>> The request selector cell isn't optional, so is always present. If we later add
>>>> an optional trig_sel cell, we'll either have:
>>>>
>>>> #dma-cells=<1>: req_sel
>>>>
>>>> or:
>>>>
>>>> #dma-cells=<2>: req_sel, trig_sel
>>>
>>> Why request sel. couldn't be optional? Could you please elaborate a bit more?
>>>
>>> I think possible options are:
>>>
>>> #dma-cells=<1>: req_sel
>>> #dma-cells=<1>: trig_sel
>>
>> With the above, how would you know that it is the req_sel or trig_sel
>> that is specified?
>>
>>> #dma-cells=<2>: req_sel, trig_sel
>>>
>>> The only difference between request and trigger is that trigger issues the whole
>>> transfer, while request only a single burst. Isn't it possible to have a case in
>>> HW for the "trigger-only" option? If not or it's a rareness, then I agree that
>>> REQ_SEL must be mandatory.
>>
>> I think that what Stephen is proposing is that for now we go with
>> '#dma-cells=<1>' and if we ever need to support the trigger cell we
>> could add support for '#dma-cells=<2>'. So with this proposal the
>> 'req_sel' would always be required for both '#dma-cells=<1>' and
>> '#dma-cells=<2>'. Even if the req_sel is not actually used but the
>> 'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
>> value (eg. -1) where we know to ignore it.
>>
> 
> Okay, I see now. Thank you for the clarification, but then we should have that
> pre-defined value declared in the binding?

I would have thought it should be in the dt-binding header.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 03/10/17 13:07, Dmitry Osipenko wrote:
> On 03.10.2017 13:32, Jon Hunter wrote:
>>
>>
>> On 03/10/17 00:02, Dmitry Osipenko wrote:
>>> On 02.10.2017 20:05, Stephen Warren wrote:
>>>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>>>
>>>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>>>> ---
>>>>>>>>>>    .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>>>> ++++++++++++++++++++++
>>>>>>>>>>    1 file changed, 23 insertions(+)
>>>>>>>>>>    create mode 100644
>>>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>>
>>>>>>>>>> diff --git
>>>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> new file mode 100644
>>>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>>>> --- /dev/null
>>>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>>>> +
>>>>>>>>>> +Required properties:
>>>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>>>> value
>>>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>>>> +        REQ_SEL field.
>>>>>>>>>
>>>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>>>
>>>>>>>>
>>>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>>>> up to
>>>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>>>
>>>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>>>
>>>>>>>> And I think the same applies to requester... any objections?
>>>>>>>
>>>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>>>
>>>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>>>> this.
>>>>>>
>>>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>>>> when SW desires, rather than being a combination of after SW programs the
>>>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>>>> rather than some other separate higher-level property/driver/...
>>>>>
>>>>> Thank you for the comment. If we'd want to extend the binding further with the
>>>>> trigger, how to differentiate trigger from the requester in a case of a single
>>>>> #data-cell?
>>>>>
>>>>> Of course realistically a chance that the further extension would be needed is
>>>>> very-very low, so we may defer the efforts to solve that question and for now
>>>>> make driver aware of the potential #dma-cells extension.
>>>>
>>>> The request selector cell isn't optional, so is always present. If we later add
>>>> an optional trig_sel cell, we'll either have:
>>>>
>>>> #dma-cells=<1>: req_sel
>>>>
>>>> or:
>>>>
>>>> #dma-cells=<2>: req_sel, trig_sel
>>>
>>> Why request sel. couldn't be optional? Could you please elaborate a bit more?
>>>
>>> I think possible options are:
>>>
>>> #dma-cells=<1>: req_sel
>>> #dma-cells=<1>: trig_sel
>>
>> With the above, how would you know that it is the req_sel or trig_sel
>> that is specified?
>>
>>> #dma-cells=<2>: req_sel, trig_sel
>>>
>>> The only difference between request and trigger is that trigger issues the whole
>>> transfer, while request only a single burst. Isn't it possible to have a case in
>>> HW for the "trigger-only" option? If not or it's a rareness, then I agree that
>>> REQ_SEL must be mandatory.
>>
>> I think that what Stephen is proposing is that for now we go with
>> '#dma-cells=<1>' and if we ever need to support the trigger cell we
>> could add support for '#dma-cells=<2>'. So with this proposal the
>> 'req_sel' would always be required for both '#dma-cells=<1>' and
>> '#dma-cells=<2>'. Even if the req_sel is not actually used but the
>> 'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
>> value (eg. -1) where we know to ignore it.
>>
> 
> Okay, I see now. Thank you for the clarification, but then we should have that
> pre-defined value declared in the binding?

I would have thought it should be in the dt-binding header.

Cheers
Jon

-- 
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Jon Hunter]

On 03/10/17 13:07, Dmitry Osipenko wrote:
> On 03.10.2017 13:32, Jon Hunter wrote:
>>
>>
>> On 03/10/17 00:02, Dmitry Osipenko wrote:
>>> On 02.10.2017 20:05, Stephen Warren wrote:
>>>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>>>
>>>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that pr=
esents
>>>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>>>> ---
>>>>>>>>>> =C2=A0=C2=A0 .../bindings/dma/nvidia,tegra20-ahbdma.txt=C2=A0=C2=
=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 | 23
>>>>>>>>>> ++++++++++++++++++++++
>>>>>>>>>> =C2=A0=C2=A0 1 file changed, 23 insertions(+)
>>>>>>>>>> =C2=A0=C2=A0 create mode 100644
>>>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>>
>>>>>>>>>> diff --git
>>>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.tx=
t
>>>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.tx=
t
>>>>>>>>>> new file mode 100644
>>>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>>>> --- /dev/null
>>>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdm=
a.txt
>>>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>>>> +
>>>>>>>>>> +Required properties:
>>>>>>>>>> +- compatible:=C2=A0=C2=A0=C2=A0 Must be "nvidia,tegra20-ahbdma"
>>>>>>>>>> +- reg:=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 Should contain=
 registers base address and length.
>>>>>>>>>> +- interrupts:=C2=A0=C2=A0=C2=A0 Should contain one entry, DMA c=
ontroller interrupt.
>>>>>>>>>> +- clocks:=C2=A0=C2=A0=C2=A0 Should contain one entry, DMA contr=
oller clock.
>>>>>>>>>> +- resets :=C2=A0=C2=A0=C2=A0 Should contain one entry, DMA cont=
roller reset.
>>>>>>>>>> +- #dma-cells:=C2=A0=C2=A0=C2=A0 Should be <1>. The cell represe=
nts DMA request select
>>>>>>>>>> value
>>>>>>>>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 for the peripheral. =
For more details consult the Tegra TRM's
>>>>>>>>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 documentation, in pa=
rticular AHB DMA channel control register
>>>>>>>>>> +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 REQ_SEL field.
>>>>>>>>>
>>>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as =
well?
>>>>>>>>>
>>>>>>>>
>>>>>>>> Actually, DMA transfer trigger isn't related a hardware descriptio=
n. It's
>>>>>>>> up to
>>>>>>>> software to decide what trigger to select. So it shouldn't be in t=
he binding.
>>>>>>>
>>>>>>> I think it could be, if say a board wanted a GPIO to trigger a tran=
sfer.
>>>>>>>
>>>>>>>> And I think the same applies to requester... any objections?
>>>>>>>
>>>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>>>
>>>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Look=
s
>>>>>>> like we never bothered with it for the APB DMA and so maybe no ones=
 uses
>>>>>>> this.
>>>>>>
>>>>>> I don't think TRIG_SEL should be in the binding, at least at present=
. While
>>>>>> TRIG_SEL certainly is something used to configure the transfer, I be=
lieve the
>>>>>> semantics of the current DMA binding only cover DMA transfers that a=
re initiated
>>>>>> when SW desires, rather than being a combination of after SW program=
s the
>>>>>> transfer plus some other HW event. So, we always use a default/hard-=
coded
>>>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT.=
 There's
>>>>>> certainly no known use-case that requires a non-default TRIG_SEL val=
ue at
>>>>>> present. We could add an extra #dma-cells value later if we find a u=
se for it,
>>>>>> and the semantics of that use-case make sense to add it to the DMA s=
pecifier,
>>>>>> rather than some other separate higher-level property/driver/...
>>>>>
>>>>> Thank you for the comment. If we'd want to extend the binding further=
 with the
>>>>> trigger, how to differentiate trigger from the requester in a case of=
 a single
>>>>> #data-cell?
>>>>>
>>>>> Of course realistically a chance that the further extension would be =
needed is
>>>>> very-very low, so we may defer the efforts to solve that question and=
 for now
>>>>> make driver aware of the potential #dma-cells extension.
>>>>
>>>> The request selector cell isn't optional, so is always present. If we =
later add
>>>> an optional trig_sel cell, we'll either have:
>>>>
>>>> #dma-cells=3D<1>: req_sel
>>>>
>>>> or:
>>>>
>>>> #dma-cells=3D<2>: req_sel, trig_sel
>>>
>>> Why request sel. couldn't be optional? Could you please elaborate a bit=
 more?
>>>
>>> I think possible options are:
>>>
>>> #dma-cells=3D<1>: req_sel
>>> #dma-cells=3D<1>: trig_sel
>>
>> With the above, how would you know that it is the req_sel or trig_sel
>> that is specified?
>>
>>> #dma-cells=3D<2>: req_sel, trig_sel
>>>
>>> The only difference between request and trigger is that trigger issues =
the whole
>>> transfer, while request only a single burst. Isn't it possible to have =
a case in
>>> HW for the "trigger-only" option? If not or it's a rareness, then I agr=
ee that
>>> REQ_SEL must be mandatory.
>>
>> I think that what Stephen is proposing is that for now we go with
>> '#dma-cells=3D<1>' and if we ever need to support the trigger cell we
>> could add support for '#dma-cells=3D<2>'. So with this proposal the
>> 'req_sel' would always be required for both '#dma-cells=3D<1>' and
>> '#dma-cells=3D<2>'. Even if the req_sel is not actually used but the
>> 'trig_sel' is, the user would have to set 'req_sel' to some pre-defined
>> value (eg. -1) where we know to ignore it.
>>
>=20
> Okay, I see now. Thank you for the clarification, but then we should have=
 that
> pre-defined value declared in the binding?

I would have thought it should be in the dt-binding header.

Cheers
Jon

--=20
nvpublic

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Stephen Warren]

On 10/03/2017 04:32 AM, Jon Hunter wrote:
> 
> 
> On 03/10/17 00:02, Dmitry Osipenko wrote:
>> On 02.10.2017 20:05, Stephen Warren wrote:
>>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>>
>>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>>
>>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>>
>>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>>> ---
>>>>>>>>>     .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>>> ++++++++++++++++++++++
>>>>>>>>>     1 file changed, 23 insertions(+)
>>>>>>>>>     create mode 100644
>>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>
>>>>>>>>> diff --git
>>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> new file mode 100644
>>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>>> --- /dev/null
>>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>>> +
>>>>>>>>> +Required properties:
>>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>>> value
>>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>>> +        REQ_SEL field.
>>>>>>>>
>>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>>
>>>>>>>
>>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>>> up to
>>>>>>> software to decide what trigger to select. So it shouldn't be in the binding.
>>>>>>
>>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>>
>>>>>>> And I think the same applies to requester... any objections?
>>>>>>
>>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>>
>>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>>> this.
>>>>>
>>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>>> semantics of the current DMA binding only cover DMA transfers that are initiated
>>>>> when SW desires, rather than being a combination of after SW programs the
>>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>>> present. We could add an extra #dma-cells value later if we find a use for it,
>>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>>> rather than some other separate higher-level property/driver/...
>>>>
>>>> Thank you for the comment. If we'd want to extend the binding further with the
>>>> trigger, how to differentiate trigger from the requester in a case of a single
>>>> #data-cell?
>>>>
>>>> Of course realistically a chance that the further extension would be needed is
>>>> very-very low, so we may defer the efforts to solve that question and for now
>>>> make driver aware of the potential #dma-cells extension.
>>>
>>> The request selector cell isn't optional, so is always present. If we later add
>>> an optional trig_sel cell, we'll either have:
>>>
>>> #dma-cells=<1>: req_sel
>>>
>>> or:
>>>
>>> #dma-cells=<2>: req_sel, trig_sel
>>
>> Why request sel. couldn't be optional? Could you please elaborate a bit more?

The documentation currently says it's mandatory, and DT bindings must be 
evolved in a backwards-compatible fashion.

>> I think possible options are:
>>
>> #dma-cells=<1>: req_sel
>> #dma-cells=<1>: trig_sel
> 
> With the above, how would you know that it is the req_sel or trig_sel
> that is specified?

Also, if req_sel were optional, then it'd be impossible to distinguish 
between those cases, so we can't design a binding like that. In general, 
when adding extra optional cells to an #xxx-cells style binding, then 
whenever cell N is present, all cells before cell N must be present even 
if optional.

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 03.10.2017 18:38, Stephen Warren wrote:
> On 10/03/2017 04:32 AM, Jon Hunter wrote:
>>
>>
>> On 03/10/17 00:02, Dmitry Osipenko wrote:
>>> On 02.10.2017 20:05, Stephen Warren wrote:
>>>> On 09/29/2017 09:11 PM, Dmitry Osipenko wrote:
>>>>> On 29.09.2017 22:30, Stephen Warren wrote:
>>>>>> On 09/27/2017 02:34 AM, Jon Hunter wrote:
>>>>>>>
>>>>>>> On 27/09/17 02:57, Dmitry Osipenko wrote:
>>>>>>>> On 26.09.2017 17:50, Jon Hunter wrote:
>>>>>>>>>
>>>>>>>>> On 26/09/17 00:22, Dmitry Osipenko wrote:
>>>>>>>>>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>>>>>>>>>> on Tegra20/30 SoC's.
>>>>>>>>>>
>>>>>>>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>>>>>>>> ---
>>>>>>>>>>     .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23
>>>>>>>>>> ++++++++++++++++++++++
>>>>>>>>>>     1 file changed, 23 insertions(+)
>>>>>>>>>>     create mode 100644
>>>>>>>>>> Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>>
>>>>>>>>>> diff --git
>>>>>>>>>> a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> new file mode 100644
>>>>>>>>>> index 000000000000..2af9aa76ae11
>>>>>>>>>> --- /dev/null
>>>>>>>>>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>>>>>>>>> @@ -0,0 +1,23 @@
>>>>>>>>>> +* NVIDIA Tegra AHB DMA controller
>>>>>>>>>> +
>>>>>>>>>> +Required properties:
>>>>>>>>>> +- compatible:    Must be "nvidia,tegra20-ahbdma"
>>>>>>>>>> +- reg:        Should contain registers base address and length.
>>>>>>>>>> +- interrupts:    Should contain one entry, DMA controller interrupt.
>>>>>>>>>> +- clocks:    Should contain one entry, DMA controller clock.
>>>>>>>>>> +- resets :    Should contain one entry, DMA controller reset.
>>>>>>>>>> +- #dma-cells:    Should be <1>. The cell represents DMA request select
>>>>>>>>>> value
>>>>>>>>>> +        for the peripheral. For more details consult the Tegra TRM's
>>>>>>>>>> +        documentation, in particular AHB DMA channel control register
>>>>>>>>>> +        REQ_SEL field.
>>>>>>>>>
>>>>>>>>> What about the TRIG_SEL field? Do we need to handle this here as well?
>>>>>>>>>
>>>>>>>>
>>>>>>>> Actually, DMA transfer trigger isn't related a hardware description. It's
>>>>>>>> up to
>>>>>>>> software to decide what trigger to select. So it shouldn't be in the
>>>>>>>> binding.
>>>>>>>
>>>>>>> I think it could be, if say a board wanted a GPIO to trigger a transfer.
>>>>>>>
>>>>>>>> And I think the same applies to requester... any objections?
>>>>>>>
>>>>>>> Well, the REQ_SEL should definitely be in the binding.
>>>>>>>
>>>>>>> Laxman, Stephen, what are your thoughts on the TRIG_SEL field? Looks
>>>>>>> like we never bothered with it for the APB DMA and so maybe no ones uses
>>>>>>> this.
>>>>>>
>>>>>> I don't think TRIG_SEL should be in the binding, at least at present. While
>>>>>> TRIG_SEL certainly is something used to configure the transfer, I believe the
>>>>>> semantics of the current DMA binding only cover DMA transfers that are
>>>>>> initiated
>>>>>> when SW desires, rather than being a combination of after SW programs the
>>>>>> transfer plus some other HW event. So, we always use a default/hard-coded
>>>>>> TRIG_SEL value. As such, there's no need for a TRIG_SEL value in DT. There's
>>>>>> certainly no known use-case that requires a non-default TRIG_SEL value at
>>>>>> present. We could add an extra #dma-cells value later if we find a use for
>>>>>> it,
>>>>>> and the semantics of that use-case make sense to add it to the DMA specifier,
>>>>>> rather than some other separate higher-level property/driver/...
>>>>>
>>>>> Thank you for the comment. If we'd want to extend the binding further with the
>>>>> trigger, how to differentiate trigger from the requester in a case of a single
>>>>> #data-cell?
>>>>>
>>>>> Of course realistically a chance that the further extension would be needed is
>>>>> very-very low, so we may defer the efforts to solve that question and for now
>>>>> make driver aware of the potential #dma-cells extension.
>>>>
>>>> The request selector cell isn't optional, so is always present. If we later add
>>>> an optional trig_sel cell, we'll either have:
>>>>
>>>> #dma-cells=<1>: req_sel
>>>>
>>>> or:
>>>>
>>>> #dma-cells=<2>: req_sel, trig_sel
>>>
>>> Why request sel. couldn't be optional? Could you please elaborate a bit more?
> 
> The documentation currently says it's mandatory, and DT bindings must be evolved
> in a backwards-compatible fashion.
> 
>>> I think possible options are:
>>>
>>> #dma-cells=<1>: req_sel
>>> #dma-cells=<1>: trig_sel
>>
>> With the above, how would you know that it is the req_sel or trig_sel
>> that is specified?
> 
> Also, if req_sel were optional, then it'd be impossible to distinguish between
> those cases, so we can't design a binding like that. In general, when adding
> extra optional cells to an #xxx-cells style binding, then whenever cell N is
> present, all cells before cell N must be present even if optional.

I've checked how extending of #dma-cells looks like and it is indeed a good
variant since it preserves backward compatibility. Thank you and Jon for the
comments and suggestions, I'll send out v2 soon.

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Rob Herring]

On Tue, Sep 26, 2017 at 02:22:04AM +0300, Dmitry Osipenko wrote:
> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
> on Tegra20/30 SoC's.
> 
> Signed-off-by: Dmitry Osipenko <digetx-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> ---
>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>  1 file changed, 23 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> 
> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> new file mode 100644
> index 000000000000..2af9aa76ae11
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> @@ -0,0 +1,23 @@
> +* NVIDIA Tegra AHB DMA controller
> +
> +Required properties:
> +- compatible:	Must be "nvidia,tegra20-ahbdma"
> +- reg:		Should contain registers base address and length.
> +- interrupts:	Should contain one entry, DMA controller interrupt.
> +- clocks:	Should contain one entry, DMA controller clock.
> +- resets :	Should contain one entry, DMA controller reset.
> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
> +		for the peripheral. For more details consult the Tegra TRM's
> +		documentation, in particular AHB DMA channel control register
> +		REQ_SEL field.
> +
> +Example:
> +
> +ahbdma: ahbdma@60008000  {

Use standard node names. dma-controller in this case.

> +	compatible = "nvidia,tegra20-ahbdma";
> +	reg = <0x60008000 0x2000>;
> +	interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
> +	clocks = <&tegra_car TEGRA20_CLK_AHBDMA>;
> +	resets = <&tegra_car 33>;
> +	#dma-cells = <1>;
> +};
> -- 
> 2.14.1
> 
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Rob Herring]

On Tue, Sep 26, 2017 at 02:22:04AM +0300, Dmitry Osipenko wrote:
> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
> on Tegra20/30 SoC's.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>  1 file changed, 23 insertions(+)
>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> 
> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> new file mode 100644
> index 000000000000..2af9aa76ae11
> --- /dev/null
> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
> @@ -0,0 +1,23 @@
> +* NVIDIA Tegra AHB DMA controller
> +
> +Required properties:
> +- compatible:	Must be "nvidia,tegra20-ahbdma"
> +- reg:		Should contain registers base address and length.
> +- interrupts:	Should contain one entry, DMA controller interrupt.
> +- clocks:	Should contain one entry, DMA controller clock.
> +- resets :	Should contain one entry, DMA controller reset.
> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
> +		for the peripheral. For more details consult the Tegra TRM's
> +		documentation, in particular AHB DMA channel control register
> +		REQ_SEL field.
> +
> +Example:
> +
> +ahbdma: ahbdma@60008000  {

Use standard node names. dma-controller in this case.

> +	compatible = "nvidia,tegra20-ahbdma";
> +	reg = <0x60008000 0x2000>;
> +	interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
> +	clocks = <&tegra_car TEGRA20_CLK_AHBDMA>;
> +	resets = <&tegra_car 33>;
> +	#dma-cells = <1>;
> +};
> -- 
> 2.14.1
> 

Re: [PATCH v1 3/5] dt-bindings: Add DT bindings for NVIDIA Tegra AHB DMA controller

[Dmitry Osipenko]

On 05.10.2017 23:33, Rob Herring wrote:
> On Tue, Sep 26, 2017 at 02:22:04AM +0300, Dmitry Osipenko wrote:
>> Document DT bindings for NVIDIA Tegra AHB DMA controller that presents
>> on Tegra20/30 SoC's.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>> ---
>>  .../bindings/dma/nvidia,tegra20-ahbdma.txt         | 23 ++++++++++++++++++++++
>>  1 file changed, 23 insertions(+)
>>  create mode 100644 Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>>
>> diff --git a/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> new file mode 100644
>> index 000000000000..2af9aa76ae11
>> --- /dev/null
>> +++ b/Documentation/devicetree/bindings/dma/nvidia,tegra20-ahbdma.txt
>> @@ -0,0 +1,23 @@
>> +* NVIDIA Tegra AHB DMA controller
>> +
>> +Required properties:
>> +- compatible:	Must be "nvidia,tegra20-ahbdma"
>> +- reg:		Should contain registers base address and length.
>> +- interrupts:	Should contain one entry, DMA controller interrupt.
>> +- clocks:	Should contain one entry, DMA controller clock.
>> +- resets :	Should contain one entry, DMA controller reset.
>> +- #dma-cells:	Should be <1>. The cell represents DMA request select value
>> +		for the peripheral. For more details consult the Tegra TRM's
>> +		documentation, in particular AHB DMA channel control register
>> +		REQ_SEL field.
>> +
>> +Example:
>> +
>> +ahbdma: ahbdma@60008000  {
> 
> Use standard node names. dma-controller in this case.
> 

Okay, I'll change it in v3. Thank you for the comment.

>> +	compatible = "nvidia,tegra20-ahbdma";
>> +	reg = <0x60008000 0x2000>;
>> +	interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH>;
>> +	clocks = <&tegra_car TEGRA20_CLK_AHBDMA>;
>> +	resets = <&tegra_car 33>;
>> +	#dma-cells = <1>;
>> +};
>> -- 
>> 2.14.1
>>