[v2,2/3] dmaengine: Add driver for NVIDIA Tegra AHB DMA controller

Commit Message

Dmitry Osipenko Oct. 3, 2017, 11:58 p.m. UTC

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, mem-to-mem isn't implemented as well.

Signed-off-by: Dmitry Osipenko <[e-mail]>
---
 drivers/dma/Kconfig           |  10 +
 drivers/dma/Makefile          |   1 +
 drivers/dma/tegra20-ahb-dma.c | 630 ++++++++++++++++++++++++++++++++++++++++++
 3 files changed, 641 insertions(+)
 create mode 100644 drivers/dma/tegra20-ahb-dma.c

Comments

Jon Hunter Oct. 6, 2017, 1:11 p.m. UTC | #1

On 04/10/17 00:58, 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, mem-to-mem isn't implemented as well.

The APB DMA does not have h/w support for sg-transfers either, but
transfer request are placed on a list. Can we not do the same for AHB?

> Signed-off-by: Dmitry Osipenko <[e-mail]>
> ---
>  drivers/dma/Kconfig           |  10 +
>  drivers/dma/Makefile          |   1 +
>  drivers/dma/tegra20-ahb-dma.c | 630 ++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 641 insertions(+)
>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
> 
> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
> index 04e381b522b4..7d132aa85174 100644
> --- a/drivers/dma/Kconfig
> +++ b/drivers/dma/Kconfig
> @@ -512,6 +512,16 @@ 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 || COMPILE_TEST
> +	select DMA_ENGINE
> +	select DMA_VIRTUAL_CHANNELS
> +	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 a145ad1426bc..f3d284bf6d65 100644
> --- a/drivers/dma/Makefile
> +++ b/drivers/dma/Makefile
> @@ -62,6 +62,7 @@ obj-$(CONFIG_STM32_DMA) += stm32-dma.o
>  obj-$(CONFIG_STM32_DMAMUX) += stm32-dmamux.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..2d176a5536aa
> --- /dev/null
> +++ b/drivers/dma/tegra20-ahb-dma.c
> @@ -0,0 +1,630 @@
> +/*
> + * Copyright 2017 Dmitry Osipenko <[e-mail]>
> + *
> + * 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/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 <dt-bindings/dma/tegra-ahb-dma.h>
> +
> +#include "virt-dma.h"
> +
> +#define AHBDMA_CMD			0x0
> +#define AHBDMA_CMD_ENABLE		BIT(31)
> +
> +#define AHBDMA_IRQ_ENB_MASK		0x20
> +#define AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
> +
> +#define AHBDMA_CH_BASE(ch)		(0x1000 + (ch) * 0x20)
> +
> +#define AHBDMA_CH_CSR			0x0
> +#define AHBDMA_CH_ADDR_WRAP		BIT(18)

This should be under the AHB_SEQ register.

> +#define AHBDMA_CH_FLOW			BIT(24)
> +#define AHBDMA_CH_ONCE			BIT(26)
> +#define AHBDMA_CH_DIR_TO_XMB		BIT(27)
> +#define AHBDMA_CH_IE_EOC		BIT(30)
> +#define AHBDMA_CH_ENABLE		BIT(31)
> +#define AHBDMA_CH_REQ_SEL_SHIFT		16
> +#define AHBDMA_CH_WCOUNT_MASK		GENMASK(15, 2)

I know that it makes the definitions longer and maybe this is why you
avoided it, but typically we have the bit field names have the register
name as the prefix. So the above would all start AHBDMA_CH_CSR_xxxx.

> +
> +#define AHBDMA_CH_STA			0x4
> +#define AHBDMA_CH_IS_EOC		BIT(30)
> +
> +#define AHBDMA_CH_AHB_PTR		0x10
> +
> +#define AHBDMA_CH_AHB_SEQ		0x14
> +#define AHBDMA_CH_INTR_ENB		BIT(31)
> +#define AHBDMA_CH_AHB_BURST_SHIFT	24
> +#define AHBDMA_CH_AHB_BURST_1		2
> +#define AHBDMA_CH_AHB_BURST_4		3
> +#define AHBDMA_CH_AHB_BURST_8		4
> +
> +#define AHBDMA_CH_XMB_PTR		0x18
> +
> +#define AHBDMA_BUS_WIDTH		BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
> +
> +#define AHBDMA_DIRECTIONS		BIT(DMA_DEV_TO_MEM) | \
> +					BIT(DMA_MEM_TO_DEV)
> +
> +#define AHBDMA_BURST_COMPLETE_TIME	20
> +
> +struct tegra_ahbdma_tx_desc {
> +	struct virt_dma_desc vdesc;
> +	dma_addr_t mem_addr;
> +	phys_addr_t ahb_addr;
> +	u32 ahb_seq;
> +	u32 csr;
> +};
> +
> +struct tegra_ahbdma_chan {
> +	struct tegra_ahbdma_tx_desc *active_tx;
> +	struct virt_dma_chan vchan;
> +	struct completion idling;
> +	void __iomem *regs;
> +	phys_addr_t ahb_addr;
> +	u32 ahb_seq;
> +	u32 csr;
> +	unsigned int of_req_sel;
> +	bool of_slave;
> +};
> +
> +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_chan *to_ahbdma_chan(struct dma_chan *chan)
> +{
> +	return container_of(chan, struct tegra_ahbdma_chan, vchan.chan);
> +}
> +
> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
> +						struct virt_dma_desc *vdesc)
> +{
> +	return container_of(vdesc, struct tegra_ahbdma_tx_desc, vdesc);
> +}
> +
> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
> +						struct tegra_ahbdma_chan *chan)
> +{
> +	struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
> +
> +	if (vdesc)
> +		list_del(&vdesc->node);
> +
> +	return vdesc ? to_ahbdma_tx_desc(vdesc) : NULL;
> +}
> +
> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
> +
> +	if (tx) {
> +		writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
> +		writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
> +		writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
> +		writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
> +
> +		reinit_completion(&chan->idling);

Should this be done before actually starting the DMA?

> +	} else
> +		complete_all(&chan->idling);

Matching curly brackets needed here (per checkpatch).

> +
> +	chan->active_tx = tx;

Should this be done right before the if-statement?

> +}
> +
> +static bool tegra_ahbdma_clear_interrupt(struct tegra_ahbdma_chan *chan)
> +{
> +	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
> +
> +	if (status & AHBDMA_CH_IS_EOC) {
> +		writel_relaxed(AHBDMA_CH_IS_EOC, chan->regs + AHBDMA_CH_STA);
> +
> +		return true;
> +	}
> +
> +	return false;
> +}
> +
> +static bool tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +	unsigned long flags;
> +	bool intr = false;
> +	bool cyclic;
> +
> +	spin_lock_irqsave(&chan->vchan.lock, flags);
> +
> +	tx = chan->active_tx;
> +	if (tx)
> +		intr = tegra_ahbdma_clear_interrupt(chan);
> +
> +	if (intr) {
> +		cyclic = !(tx->csr & AHBDMA_CH_ONCE);
> +
> +		if (!cyclic)
> +			tegra_ahbdma_issue_next_tx(chan);
> +
> +		if (cyclic)
> +			vchan_cyclic_callback(&tx->vdesc);
> +		else
> +			vchan_cookie_complete(&tx->vdesc);

Looks like the above could be combined into a single if-statement.

> +	}
> +
> +	spin_unlock_irqrestore(&chan->vchan.lock, flags);
> +
> +	return intr;
> +}
> +
> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
> +{
> +	struct tegra_ahbdma *tdma = dev_id;
> +	bool handled;
> +
> +	handled  = tegra_ahbdma_handle_channel(&tdma->channels[0]);
> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[1]);
> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[2]);
> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[3]);
> +
> +	return handled ? IRQ_HANDLED : IRQ_NONE;
> +}
> +
> +static void tegra_ahbdma_tx_desc_free(struct virt_dma_desc *vdesc)
> +{
> +	kfree(to_ahbdma_tx_desc(vdesc));
> +}
> +
> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep(
> +					struct dma_chan *chan,
> +					enum dma_transfer_direction dir,
> +					unsigned long flags,
> +					dma_addr_t paddr,
> +					size_t size,
> +					bool cyclic)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	u32 csr = ahbdma_chan->csr;
> +
> +	/* size and alignments should fulfill HW requirements */
> +	if (size < 4 || size & 3 || paddr & 3)
> +		return NULL;
> +
> +	tx = kzalloc(sizeof(*tx), GFP_NOWAIT);
> +	if (!tx)
> +		return NULL;
> +
> +	if (dir == DMA_DEV_TO_MEM)
> +		csr |= AHBDMA_CH_DIR_TO_XMB;
> +
> +	if (!cyclic)
> +		csr |= AHBDMA_CH_ONCE;
> +
> +	tx->csr = csr | (size - sizeof(u32));
> +	tx->ahb_seq = ahbdma_chan->ahb_seq;
> +	tx->ahb_addr = ahbdma_chan->ahb_addr;
> +	tx->mem_addr = paddr;
> +
> +	return vchan_tx_prep(&ahbdma_chan->vchan, &tx->vdesc, flags);
> +}
> +
> +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)
> +{
> +	/* unimplemented */
> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
> +		return NULL;

I think that this warrants a better comment. It is clear from the
changelog but not found just browsing the source.

> +
> +	return tegra_ahbdma_prep(chan, dir, flags, sg_dma_address(sgl),
> +				 sg_dma_len(sgl), false);
> +}
> +
> +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)
> +{
> +	/* unimplemented */
> +	if (buf_len != period_len || buf_len > SZ_64K)
> +		return NULL;

Same here w.r.t the comment.

> +
> +	return tegra_ahbdma_prep(chan, dir, flags, buf_addr, buf_len, true);
> +}
> +
> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct virt_dma_chan *vchan = &ahbdma_chan->vchan;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&vchan->lock, flags);
> +
> +	if (vchan_issue_pending(vchan) && !ahbdma_chan->active_tx)
> +		tegra_ahbdma_issue_next_tx(ahbdma_chan);
> +
> +	spin_unlock_irqrestore(&vchan->lock, flags);
> +}
> +
> +static size_t tegra_ahbdma_residual(struct tegra_ahbdma_chan *chan)
> +{
> +	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
> +
> +	return (status & AHBDMA_CH_WCOUNT_MASK);
> +}
> +
> +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;
> +	struct virt_dma_desc *vdesc;
> +	enum dma_status cookie_status;
> +	unsigned long flags;
> +	size_t residual;
> +
> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
> +
> +	cookie_status = dma_cookie_status(chan, cookie, state);
> +	if (cookie_status == DMA_COMPLETE)
> +		goto unlock;
> +
> +	vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
> +	if (!vdesc)
> +		residual = 0;

Matching curly brackets needed.

> +	else {
> +		tx = to_ahbdma_tx_desc(vdesc);
> +
> +		if (tx == ahbdma_chan->active_tx)
> +			residual = tegra_ahbdma_residual(ahbdma_chan);
> +		else
> +			residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
> +
> +		residual += sizeof(u32);
> +	}
> +
> +	dma_set_residue(state, residual);

I believe residue needs to be bytes.

> +
> +unlock:
> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.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);
> +	unsigned long flags;
> +	LIST_HEAD(head);
> +	u32 csr;
> +
> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
> +
> +	csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
> +	writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
> +		       ahbdma_chan->regs + AHBDMA_CH_CSR);
> +
> +	if (ahbdma_chan->active_tx) {
> +		udelay(AHBDMA_BURST_COMPLETE_TIME);

Why not poll the status register and wait for the channel to stop?

> +
> +		writel_relaxed(AHBDMA_CH_IS_EOC,
> +			       ahbdma_chan->regs + AHBDMA_CH_STA);
> +
> +		ahbdma_chan->active_tx = NULL;
> +	}
> +
> +	vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
> +	complete_all(&ahbdma_chan->idling);
> +
> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
> +
> +	vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
> +
> +	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, csr;
> +	unsigned int slave_id;
> +	phys_addr_t ahb_addr;
> +
> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
> +		return -EINVAL;
> +
> +	switch (dir) {
> +	case DMA_DEV_TO_MEM:
> +		burst    = sconfig->src_maxburst;
> +		ahb_addr = sconfig->src_addr;
> +		break;
> +	case DMA_MEM_TO_DEV:
> +		burst    = sconfig->dst_maxburst;
> +		ahb_addr = sconfig->dst_addr;
> +		break;
> +	default:
> +		return -EINVAL;
> +	}
> +
> +	switch (burst) {
> +	case 1:
> +		burst = AHBDMA_CH_AHB_BURST_1;
> +		break;
> +	case 4:
> +		burst = AHBDMA_CH_AHB_BURST_4;
> +		break;
> +	case 8:
> +		burst = AHBDMA_CH_AHB_BURST_8;
> +		break;
> +	default:
> +		return -EINVAL;
> +	}
> +
> +	if (ahb_addr & 3)
> +		return -EINVAL;
> +
> +	ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
> +	ahb_seq |= AHBDMA_CH_INTR_ENB;
> +
> +	csr  = AHBDMA_CH_ENABLE;
> +	csr |= AHBDMA_CH_IE_EOC;
> +
> +	if (ahbdma_chan->of_slave || sconfig->device_fc) {
> +		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
> +			slave_id = ahbdma_chan->of_req_sel;
> +		else
> +			slave_id = sconfig->slave_id;
> +
> +		if (slave_id > 15)
> +			return -EINVAL;

Why not ...

	if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
		slave_id = ahbdma_chan->of_req_sel;
	else if (slave_id = sconfig->slave_id < TEGRA_AHBDMA_REQ_N_A)
		slave_id = sconfig->slave_id;
	else
		return -EINVAL;

> +
> +		ahb_seq |= AHBDMA_CH_ADDR_WRAP;
> +
> +		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
> +		csr |= AHBDMA_CH_FLOW;
> +	}
> +
> +	ahbdma_chan->csr = csr;
> +	ahbdma_chan->ahb_seq = ahb_seq;
> +	ahbdma_chan->ahb_addr = ahb_addr;
> +
> +	return 0;
> +}
> +
> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +
> +	wait_for_completion(&ahbdma_chan->idling);
> +	vchan_synchronize(&ahbdma_chan->vchan);
> +}
> +
> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
> +{
> +	vchan_free_chan_resources(to_virt_chan(chan));
> +}
> +
> +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_device *dma_dev = &tdma->dma_dev;
> +
> +	vchan_init(&ahbdma_chan->vchan, dma_dev);
> +	init_completion(&ahbdma_chan->idling);
> +	complete(&ahbdma_chan->idling);
> +
> +	ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
> +	ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
> +	ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
> +}
> +
> +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;
> +
> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
> +	if (!chan)
> +		return NULL;
> +
> +	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];

Test for args[0] < TEGRA_AHBDMA_REQ_N_A?

> +	to_ahbdma_chan(chan)->of_slave = true;

Is this really needed? Doesn't a value of 0..TEGRA_AHBDMA_REQ_N_A-1 tell
us it is valid?

> +
> +	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);

Clock disable?

> +		return err;
> +	}
> +
> +	writel_relaxed(AHBDMA_CMD_ENABLE, tdma->regs + AHBDMA_CMD);
> +
> +	writel_relaxed(AHBDMA_IRQ_ENB_CH(0) |
> +		       AHBDMA_IRQ_ENB_CH(1) |
> +		       AHBDMA_IRQ_ENB_CH(2) |
> +		       AHBDMA_IRQ_ENB_CH(3),
> +		       tdma->regs + 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;

Nit-pick ... put irq and err on one line.

> +
> +	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		= AHBDMA_DIRECTIONS;
> +	dma_dev->src_addr_widths	= AHBDMA_BUS_WIDTH;
> +	dma_dev->dst_addr_widths	= 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);

Clock disable?

> +		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);

Clock disable?

> +		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 <[e-mail]>");
> +MODULE_LICENSE("GPL");
> 

Cheers
Jon

Dmitry Osipenko Oct. 6, 2017, 3:26 p.m. UTC | #2

On 06.10.2017 16:11, Jon Hunter wrote:
> 
> On 04/10/17 00:58, 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, mem-to-mem isn't implemented as well.
> 
> The APB DMA does not have h/w support for sg-transfers either, but
> transfer request are placed on a list. Can we not do the same for AHB?
> 

We can, but I'm not going to implement it without a use-case. It could be done
later if needed.

>> Signed-off-by: Dmitry Osipenko <[e-mail]>
>> ---
>>  drivers/dma/Kconfig           |  10 +
>>  drivers/dma/Makefile          |   1 +
>>  drivers/dma/tegra20-ahb-dma.c | 630 ++++++++++++++++++++++++++++++++++++++++++
>>  3 files changed, 641 insertions(+)
>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>
>> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
>> index 04e381b522b4..7d132aa85174 100644
>> --- a/drivers/dma/Kconfig
>> +++ b/drivers/dma/Kconfig
>> @@ -512,6 +512,16 @@ 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 || COMPILE_TEST
>> +	select DMA_ENGINE
>> +	select DMA_VIRTUAL_CHANNELS
>> +	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 a145ad1426bc..f3d284bf6d65 100644
>> --- a/drivers/dma/Makefile
>> +++ b/drivers/dma/Makefile
>> @@ -62,6 +62,7 @@ obj-$(CONFIG_STM32_DMA) += stm32-dma.o
>>  obj-$(CONFIG_STM32_DMAMUX) += stm32-dmamux.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..2d176a5536aa
>> --- /dev/null
>> +++ b/drivers/dma/tegra20-ahb-dma.c
>> @@ -0,0 +1,630 @@
>> +/*
>> + * Copyright 2017 Dmitry Osipenko <[e-mail]>
>> + *
>> + * 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/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 <dt-bindings/dma/tegra-ahb-dma.h>
>> +
>> +#include "virt-dma.h"
>> +
>> +#define AHBDMA_CMD			0x0
>> +#define AHBDMA_CMD_ENABLE		BIT(31)
>> +
>> +#define AHBDMA_IRQ_ENB_MASK		0x20
>> +#define AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
>> +
>> +#define AHBDMA_CH_BASE(ch)		(0x1000 + (ch) * 0x20)
>> +
>> +#define AHBDMA_CH_CSR			0x0
>> +#define AHBDMA_CH_ADDR_WRAP		BIT(18)
> 
> This should be under the AHB_SEQ register.
> 

Indeed

>> +#define AHBDMA_CH_FLOW			BIT(24)
>> +#define AHBDMA_CH_ONCE			BIT(26)
>> +#define AHBDMA_CH_DIR_TO_XMB		BIT(27)
>> +#define AHBDMA_CH_IE_EOC		BIT(30)
>> +#define AHBDMA_CH_ENABLE		BIT(31)
>> +#define AHBDMA_CH_REQ_SEL_SHIFT		16
>> +#define AHBDMA_CH_WCOUNT_MASK		GENMASK(15, 2)
> 
> I know that it makes the definitions longer and maybe this is why you
> avoided it, but typically we have the bit field names have the register
> name as the prefix. So the above would all start AHBDMA_CH_CSR_xxxx.
> 

Okay

>> +
>> +#define AHBDMA_CH_STA			0x4
>> +#define AHBDMA_CH_IS_EOC		BIT(30)
>> +
>> +#define AHBDMA_CH_AHB_PTR		0x10
>> +
>> +#define AHBDMA_CH_AHB_SEQ		0x14
>> +#define AHBDMA_CH_INTR_ENB		BIT(31)
>> +#define AHBDMA_CH_AHB_BURST_SHIFT	24
>> +#define AHBDMA_CH_AHB_BURST_1		2
>> +#define AHBDMA_CH_AHB_BURST_4		3
>> +#define AHBDMA_CH_AHB_BURST_8		4
>> +
>> +#define AHBDMA_CH_XMB_PTR		0x18
>> +
>> +#define AHBDMA_BUS_WIDTH		BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +
>> +#define AHBDMA_DIRECTIONS		BIT(DMA_DEV_TO_MEM) | \
>> +					BIT(DMA_MEM_TO_DEV)
>> +
>> +#define AHBDMA_BURST_COMPLETE_TIME	20
>> +
>> +struct tegra_ahbdma_tx_desc {
>> +	struct virt_dma_desc vdesc;
>> +	dma_addr_t mem_addr;
>> +	phys_addr_t ahb_addr;
>> +	u32 ahb_seq;
>> +	u32 csr;
>> +};
>> +
>> +struct tegra_ahbdma_chan {
>> +	struct tegra_ahbdma_tx_desc *active_tx;
>> +	struct virt_dma_chan vchan;
>> +	struct completion idling;
>> +	void __iomem *regs;
>> +	phys_addr_t ahb_addr;
>> +	u32 ahb_seq;
>> +	u32 csr;
>> +	unsigned int of_req_sel;
>> +	bool of_slave;
>> +};
>> +
>> +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_chan *to_ahbdma_chan(struct dma_chan *chan)
>> +{
>> +	return container_of(chan, struct tegra_ahbdma_chan, vchan.chan);
>> +}
>> +
>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>> +						struct virt_dma_desc *vdesc)
>> +{
>> +	return container_of(vdesc, struct tegra_ahbdma_tx_desc, vdesc);
>> +}
>> +
>> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
>> +						struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
>> +
>> +	if (vdesc)
>> +		list_del(&vdesc->node);
>> +
>> +	return vdesc ? to_ahbdma_tx_desc(vdesc) : NULL;
>> +}
>> +
>> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
>> +
>> +	if (tx) {
>> +		writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
>> +		writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
>> +		writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
>> +		writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
>> +
>> +		reinit_completion(&chan->idling);
> 
> Should this be done before actually starting the DMA?
> 

Doesn't matter, this code is protected with a spinlock. I prefer mine variant,
if you don't mind.

>> +	} else
>> +		complete_all(&chan->idling);
> 
> Matching curly brackets needed here (per checkpatch).
> 

Interestingly my checkpatch doesn't complain about it. But I see that it is
documented in the codingstyle, if I remember correctly it was the right style
before.. Will correct it.

>> +
>> +	chan->active_tx = tx;
> 
> Should this be done right before the if-statement?
> 

Same as for reinit_completion.

>> +}
>> +
>> +static bool tegra_ahbdma_clear_interrupt(struct tegra_ahbdma_chan *chan)
>> +{
>> +	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>> +
>> +	if (status & AHBDMA_CH_IS_EOC) {
>> +		writel_relaxed(AHBDMA_CH_IS_EOC, chan->regs + AHBDMA_CH_STA);
>> +
>> +		return true;
>> +	}
>> +
>> +	return false;
>> +}
>> +
>> +static bool tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	unsigned long flags;
>> +	bool intr = false;
>> +	bool cyclic;
>> +
>> +	spin_lock_irqsave(&chan->vchan.lock, flags);
>> +
>> +	tx = chan->active_tx;
>> +	if (tx)
>> +		intr = tegra_ahbdma_clear_interrupt(chan);
>> +
>> +	if (intr) {
>> +		cyclic = !(tx->csr & AHBDMA_CH_ONCE);
>> +
>> +		if (!cyclic)
>> +			tegra_ahbdma_issue_next_tx(chan);
>> +
>> +		if (cyclic)
>> +			vchan_cyclic_callback(&tx->vdesc);
>> +		else
>> +			vchan_cookie_complete(&tx->vdesc);
> 
> Looks like the above could be combined into a single if-statement.
> 

Okay

>> +	}
>> +
>> +	spin_unlock_irqrestore(&chan->vchan.lock, flags);
>> +
>> +	return intr;
>> +}
>> +
>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>> +{
>> +	struct tegra_ahbdma *tdma = dev_id;
>> +	bool handled;
>> +
>> +	handled  = tegra_ahbdma_handle_channel(&tdma->channels[0]);
>> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[1]);
>> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[2]);
>> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[3]);
>> +
>> +	return handled ? IRQ_HANDLED : IRQ_NONE;
>> +}
>> +
>> +static void tegra_ahbdma_tx_desc_free(struct virt_dma_desc *vdesc)
>> +{
>> +	kfree(to_ahbdma_tx_desc(vdesc));
>> +}
>> +
>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep(
>> +					struct dma_chan *chan,
>> +					enum dma_transfer_direction dir,
>> +					unsigned long flags,
>> +					dma_addr_t paddr,
>> +					size_t size,
>> +					bool cyclic)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct tegra_ahbdma_tx_desc *tx;
>> +	u32 csr = ahbdma_chan->csr;
>> +
>> +	/* size and alignments should fulfill HW requirements */
>> +	if (size < 4 || size & 3 || paddr & 3)
>> +		return NULL;
>> +
>> +	tx = kzalloc(sizeof(*tx), GFP_NOWAIT);
>> +	if (!tx)
>> +		return NULL;
>> +
>> +	if (dir == DMA_DEV_TO_MEM)
>> +		csr |= AHBDMA_CH_DIR_TO_XMB;
>> +
>> +	if (!cyclic)
>> +		csr |= AHBDMA_CH_ONCE;
>> +
>> +	tx->csr = csr | (size - sizeof(u32));
>> +	tx->ahb_seq = ahbdma_chan->ahb_seq;
>> +	tx->ahb_addr = ahbdma_chan->ahb_addr;
>> +	tx->mem_addr = paddr;
>> +
>> +	return vchan_tx_prep(&ahbdma_chan->vchan, &tx->vdesc, flags);
>> +}
>> +
>> +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)
>> +{
>> +	/* unimplemented */
>> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>> +		return NULL;
> 
> I think that this warrants a better comment. It is clear from the
> changelog but not found just browsing the source.
> 

Okay

>> +
>> +	return tegra_ahbdma_prep(chan, dir, flags, sg_dma_address(sgl),
>> +				 sg_dma_len(sgl), false);
>> +}
>> +
>> +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)
>> +{
>> +	/* unimplemented */
>> +	if (buf_len != period_len || buf_len > SZ_64K)
>> +		return NULL;
> 
> Same here w.r.t the comment.
> 

Okay

>> +
>> +	return tegra_ahbdma_prep(chan, dir, flags, buf_addr, buf_len, true);
>> +}
>> +
>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +	struct virt_dma_chan *vchan = &ahbdma_chan->vchan;
>> +	unsigned long flags;
>> +
>> +	spin_lock_irqsave(&vchan->lock, flags);
>> +
>> +	if (vchan_issue_pending(vchan) && !ahbdma_chan->active_tx)
>> +		tegra_ahbdma_issue_next_tx(ahbdma_chan);
>> +
>> +	spin_unlock_irqrestore(&vchan->lock, flags);
>> +}
>> +
>> +static size_t tegra_ahbdma_residual(struct tegra_ahbdma_chan *chan)
>> +{
>> +	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>> +
>> +	return (status & AHBDMA_CH_WCOUNT_MASK);
>> +}
>> +
>> +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;
>> +	struct virt_dma_desc *vdesc;
>> +	enum dma_status cookie_status;
>> +	unsigned long flags;
>> +	size_t residual;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>> +
>> +	cookie_status = dma_cookie_status(chan, cookie, state);
>> +	if (cookie_status == DMA_COMPLETE)
>> +		goto unlock;
>> +
>> +	vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
>> +	if (!vdesc)
>> +		residual = 0;
> 
> Matching curly brackets needed.
> 

Okay

>> +	else {
>> +		tx = to_ahbdma_tx_desc(vdesc);
>> +
>> +		if (tx == ahbdma_chan->active_tx)
>> +			residual = tegra_ahbdma_residual(ahbdma_chan);
>> +		else
>> +			residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>> +
>> +		residual += sizeof(u32);
>> +	}
>> +
>> +	dma_set_residue(state, residual);
> 
> I believe residue needs to be bytes.
> 

It is in bytes. WCOUNT_MASK is 0xFFFC, words count is aligned to a word size.

>> +
>> +unlock:
>> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.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);
>> +	unsigned long flags;
>> +	LIST_HEAD(head);
>> +	u32 csr;
>> +
>> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>> +
>> +	csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
>> +	writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
>> +		       ahbdma_chan->regs + AHBDMA_CH_CSR);
>> +
>> +	if (ahbdma_chan->active_tx) {
>> +		udelay(AHBDMA_BURST_COMPLETE_TIME);
> 
> Why not poll the status register and wait for the channel to stop?
> 

That probably would also work. But I'm not sure whether status depends on the
channels "enable" state..

>> +
>> +		writel_relaxed(AHBDMA_CH_IS_EOC,
>> +			       ahbdma_chan->regs + AHBDMA_CH_STA);
>> +
>> +		ahbdma_chan->active_tx = NULL;
>> +	}
>> +
>> +	vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
>> +	complete_all(&ahbdma_chan->idling);
>> +
>> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
>> +
>> +	vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
>> +
>> +	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, csr;
>> +	unsigned int slave_id;
>> +	phys_addr_t ahb_addr;
>> +
>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>> +		return -EINVAL;
>> +
>> +	switch (dir) {
>> +	case DMA_DEV_TO_MEM:
>> +		burst    = sconfig->src_maxburst;
>> +		ahb_addr = sconfig->src_addr;
>> +		break;
>> +	case DMA_MEM_TO_DEV:
>> +		burst    = sconfig->dst_maxburst;
>> +		ahb_addr = sconfig->dst_addr;
>> +		break;
>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	switch (burst) {
>> +	case 1:
>> +		burst = AHBDMA_CH_AHB_BURST_1;
>> +		break;
>> +	case 4:
>> +		burst = AHBDMA_CH_AHB_BURST_4;
>> +		break;
>> +	case 8:
>> +		burst = AHBDMA_CH_AHB_BURST_8;
>> +		break;
>> +	default:
>> +		return -EINVAL;
>> +	}
>> +
>> +	if (ahb_addr & 3)
>> +		return -EINVAL;
>> +
>> +	ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
>> +	ahb_seq |= AHBDMA_CH_INTR_ENB;
>> +
>> +	csr  = AHBDMA_CH_ENABLE;
>> +	csr |= AHBDMA_CH_IE_EOC;
>> +
>> +	if (ahbdma_chan->of_slave || sconfig->device_fc) {
>> +		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>> +			slave_id = ahbdma_chan->of_req_sel;
>> +		else
>> +			slave_id = sconfig->slave_id;
>> +
>> +		if (slave_id > 15)
>> +			return -EINVAL;
> 
> Why not ...
> 
> 	if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
> 		slave_id = ahbdma_chan->of_req_sel;
> 	else if (slave_id = sconfig->slave_id < TEGRA_AHBDMA_REQ_N_A)
> 		slave_id = sconfig->slave_id;
> 	else
> 		return -EINVAL;
> 

Because I'm finding variant like yours more difficult to read. I'll stick with
my variant if you don't mind.

>> +
>> +		ahb_seq |= AHBDMA_CH_ADDR_WRAP;
>> +
>> +		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
>> +		csr |= AHBDMA_CH_FLOW;
>> +	}
>> +
>> +	ahbdma_chan->csr = csr;
>> +	ahbdma_chan->ahb_seq = ahb_seq;
>> +	ahbdma_chan->ahb_addr = ahb_addr;
>> +
>> +	return 0;
>> +}
>> +
>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>> +{
>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>> +
>> +	wait_for_completion(&ahbdma_chan->idling);
>> +	vchan_synchronize(&ahbdma_chan->vchan);
>> +}
>> +
>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>> +{
>> +	vchan_free_chan_resources(to_virt_chan(chan));
>> +}
>> +
>> +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_device *dma_dev = &tdma->dma_dev;
>> +
>> +	vchan_init(&ahbdma_chan->vchan, dma_dev);
>> +	init_completion(&ahbdma_chan->idling);
>> +	complete(&ahbdma_chan->idling);
>> +
>> +	ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
>> +	ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
>> +	ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
>> +}
>> +
>> +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;
>> +
>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>> +	if (!chan)
>> +		return NULL;
>> +
>> +	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
> 
> Test for args[0] < TEGRA_AHBDMA_REQ_N_A?
> 

It would duplicate slave_id checking done in tegra_ahbdma_config(), so not
needed here.

>> +	to_ahbdma_chan(chan)->of_slave = true;
> 
> Is this really needed? Doesn't a value of 0..TEGRA_AHBDMA_REQ_N_A-1 tell
> us it is valid?
> 

I think we should enforce channels flow control in a case of OF xlate'd channel,
no? To avoid abusing channels usage by client. Seems tegra_ahbdma_config isn't
correct, should be:

	if (ahbdma_chan->of_slave || sconfig->device_fc) {
-		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
+		if (ahbdma_chan->of_slave)
			slave_id = ahbdma_chan->of_req_sel;
		else
			slave_id = sconfig->slave_id;

		if (slave_id >= TEGRA_AHBDMA_REQ_N_A)
			return -EINVAL;

I'm finding OF-requsted channel + ability of DMA API to override requested
channels parameters quite vague. So I'm not exactly sure how to handle it
correctly. It looks like each driver is free to do its own thing, which is kinda
a mess. Suggestions?

>> +
>> +	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);
> 
> Clock disable?
> 

Yep

>> +		return err;
>> +	}
>> +
>> +	writel_relaxed(AHBDMA_CMD_ENABLE, tdma->regs + AHBDMA_CMD);
>> +
>> +	writel_relaxed(AHBDMA_IRQ_ENB_CH(0) |
>> +		       AHBDMA_IRQ_ENB_CH(1) |
>> +		       AHBDMA_IRQ_ENB_CH(2) |
>> +		       AHBDMA_IRQ_ENB_CH(3),
>> +		       tdma->regs + 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;
> 
> Nit-pick ... put irq and err on one line.
> 

Okay

>> +
>> +	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		= AHBDMA_DIRECTIONS;
>> +	dma_dev->src_addr_widths	= AHBDMA_BUS_WIDTH;
>> +	dma_dev->dst_addr_widths	= 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);
> 
> Clock disable?
> 

Yeah, I also noticed it a day ago. Thank you very much for the review!

>> +		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);
> 
> Clock disable?
> 
>> +		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 <[e-mail]>");
>> +MODULE_LICENSE("GPL");
>>

--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Jon Hunter Oct. 6, 2017, 3:50 p.m. UTC | #3

On 06/10/17 16:26, Dmitry Osipenko wrote:
> On 06.10.2017 16:11, Jon Hunter wrote:
>>
>> On 04/10/17 00:58, 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, mem-to-mem isn't implemented as well.
>>
>> The APB DMA does not have h/w support for sg-transfers either, but
>> transfer request are placed on a list. Can we not do the same for AHB?
>>
> 
> We can, but I'm not going to implement it without a use-case. It could be done
> later if needed.

OK, that's fine, maybe state that above.

[...]

>>> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
>>> +
>>> +	if (tx) {
>>> +		writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
>>> +		writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
>>> +		writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
>>> +		writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
>>> +
>>> +		reinit_completion(&chan->idling);
>>
>> Should this be done before actually starting the DMA?

OK, then that's fine.

[...]

>>> +	else {
>>> +		tx = to_ahbdma_tx_desc(vdesc);
>>> +
>>> +		if (tx == ahbdma_chan->active_tx)
>>> +			residual = tegra_ahbdma_residual(ahbdma_chan);
>>> +		else
>>> +			residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>>> +
>>> +		residual += sizeof(u32);
>>> +	}
>>> +
>>> +	dma_set_residue(state, residual);
>>
>> I believe residue needs to be bytes.

Oops yes indeed!

>>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +	unsigned long flags;
>>> +	LIST_HEAD(head);
>>> +	u32 csr;
>>> +
>>> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>>> +
>>> +	csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
>>> +	writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
>>> +		       ahbdma_chan->regs + AHBDMA_CH_CSR);
>>> +
>>> +	if (ahbdma_chan->active_tx) {
>>> +		udelay(AHBDMA_BURST_COMPLETE_TIME);
>>
>> Why not poll the status register and wait for the channel to stop?
>>
> 
> That probably would also work. But I'm not sure whether status depends on the
> channels "enable" state..

Well if it is not enabled, then we probably don't care about the state.
However, a quick test should tell us.

>>> +
>>> +		writel_relaxed(AHBDMA_CH_IS_EOC,
>>> +			       ahbdma_chan->regs + AHBDMA_CH_STA);
>>> +
>>> +		ahbdma_chan->active_tx = NULL;
>>> +	}
>>> +
>>> +	vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
>>> +	complete_all(&ahbdma_chan->idling);
>>> +
>>> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
>>> +
>>> +	vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
>>> +
>>> +	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, csr;
>>> +	unsigned int slave_id;
>>> +	phys_addr_t ahb_addr;
>>> +
>>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>> +		return -EINVAL;
>>> +
>>> +	switch (dir) {
>>> +	case DMA_DEV_TO_MEM:
>>> +		burst    = sconfig->src_maxburst;
>>> +		ahb_addr = sconfig->src_addr;
>>> +		break;
>>> +	case DMA_MEM_TO_DEV:
>>> +		burst    = sconfig->dst_maxburst;
>>> +		ahb_addr = sconfig->dst_addr;
>>> +		break;
>>> +	default:
>>> +		return -EINVAL;
>>> +	}
>>> +
>>> +	switch (burst) {
>>> +	case 1:
>>> +		burst = AHBDMA_CH_AHB_BURST_1;
>>> +		break;
>>> +	case 4:
>>> +		burst = AHBDMA_CH_AHB_BURST_4;
>>> +		break;
>>> +	case 8:
>>> +		burst = AHBDMA_CH_AHB_BURST_8;
>>> +		break;
>>> +	default:
>>> +		return -EINVAL;
>>> +	}
>>> +
>>> +	if (ahb_addr & 3)
>>> +		return -EINVAL;
>>> +
>>> +	ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
>>> +	ahb_seq |= AHBDMA_CH_INTR_ENB;
>>> +
>>> +	csr  = AHBDMA_CH_ENABLE;
>>> +	csr |= AHBDMA_CH_IE_EOC;
>>> +
>>> +	if (ahbdma_chan->of_slave || sconfig->device_fc) {
>>> +		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>>> +			slave_id = ahbdma_chan->of_req_sel;
>>> +		else
>>> +			slave_id = sconfig->slave_id;
>>> +
>>> +		if (slave_id > 15)
>>> +			return -EINVAL;
>>
>> Why not ...
>>
>> 	if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>> 		slave_id = ahbdma_chan->of_req_sel;
>> 	else if (slave_id = sconfig->slave_id < TEGRA_AHBDMA_REQ_N_A)
>> 		slave_id = sconfig->slave_id;
>> 	else
>> 		return -EINVAL;
>>
> 
> Because I'm finding variant like yours more difficult to read. I'll stick with
> my variant if you don't mind.

Ha! I prefer mine :-)

>>> +
>>> +		ahb_seq |= AHBDMA_CH_ADDR_WRAP;
>>> +
>>> +		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
>>> +		csr |= AHBDMA_CH_FLOW;
>>> +	}
>>> +
>>> +	ahbdma_chan->csr = csr;
>>> +	ahbdma_chan->ahb_seq = ahb_seq;
>>> +	ahbdma_chan->ahb_addr = ahb_addr;
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>> +{
>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +
>>> +	wait_for_completion(&ahbdma_chan->idling);
>>> +	vchan_synchronize(&ahbdma_chan->vchan);
>>> +}
>>> +
>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>> +{
>>> +	vchan_free_chan_resources(to_virt_chan(chan));
>>> +}
>>> +
>>> +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_device *dma_dev = &tdma->dma_dev;
>>> +
>>> +	vchan_init(&ahbdma_chan->vchan, dma_dev);
>>> +	init_completion(&ahbdma_chan->idling);
>>> +	complete(&ahbdma_chan->idling);
>>> +
>>> +	ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
>>> +	ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
>>> +	ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
>>> +}
>>> +
>>> +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;
>>> +
>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>> +	if (!chan)
>>> +		return NULL;
>>> +
>>> +	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
>>
>> Test for args[0] < TEGRA_AHBDMA_REQ_N_A?
>>
> 
> It would duplicate slave_id checking done in tegra_ahbdma_config(), so not
> needed here.

But surely we should not let them request a channel in the first place?

>>> +	to_ahbdma_chan(chan)->of_slave = true;
>>
>> Is this really needed? Doesn't a value of 0..TEGRA_AHBDMA_REQ_N_A-1 tell
>> us it is valid?
>>
> 
> I think we should enforce channels flow control in a case of OF xlate'd channel,
> no? To avoid abusing channels usage by client. Seems tegra_ahbdma_config isn't
> correct, should be:

Absolutely. However, I don't see the need for the additional 'of_slave'
variable. If we validate the slave id here, we can get rid of the extra
variable. It does not simplify the code really by adding this IMO.

> 	if (ahbdma_chan->of_slave || sconfig->device_fc) {
> -		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
> +		if (ahbdma_chan->of_slave)
> 			slave_id = ahbdma_chan->of_req_sel;
> 		else
> 			slave_id = sconfig->slave_id;
> 
> 		if (slave_id >= TEGRA_AHBDMA_REQ_N_A)
> 			return -EINVAL;
> 
> I'm finding OF-requsted channel + ability of DMA API to override requested
> channels parameters quite vague. So I'm not exactly sure how to handle it
> correctly. It looks like each driver is free to do its own thing, which is kinda
> a mess. Suggestions?

I think it is fine how you have it and limit OF-requested channels to
actual REQ_SEL values.

Cheers
Jon

Dmitry Osipenko Oct. 6, 2017, 5:23 p.m. UTC | #4

On 06.10.2017 18:50, Jon Hunter wrote:
> 
> On 06/10/17 16:26, Dmitry Osipenko wrote:
>> On 06.10.2017 16:11, Jon Hunter wrote:
>>>
>>> On 04/10/17 00:58, 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, mem-to-mem isn't implemented as well.
>>>
>>> The APB DMA does not have h/w support for sg-transfers either, but
>>> transfer request are placed on a list. Can we not do the same for AHB?
>>>
>>
>> We can, but I'm not going to implement it without a use-case. It could be done
>> later if needed.
> 
> OK, that's fine, maybe state that above.
> 

Well, I think it is explicitly mentioned in the commit message.

"Driver doesn't yet implement transfers larger than 64K and scatter-gather
transfers that have NENT > 1".

Isn't it enough?

> [...]
> 
>>>> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
>>>> +
>>>> +	if (tx) {
>>>> +		writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
>>>> +		writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
>>>> +		writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
>>>> +		writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
>>>> +
>>>> +		reinit_completion(&chan->idling);
>>>
>>> Should this be done before actually starting the DMA?
> 
> OK, then that's fine.
> 
> [...]
> 
>>>> +	else {
>>>> +		tx = to_ahbdma_tx_desc(vdesc);
>>>> +
>>>> +		if (tx == ahbdma_chan->active_tx)
>>>> +			residual = tegra_ahbdma_residual(ahbdma_chan);
>>>> +		else
>>>> +			residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>>>> +
>>>> +		residual += sizeof(u32);
>>>> +	}
>>>> +
>>>> +	dma_set_residue(state, residual);
>>>
>>> I believe residue needs to be bytes.
> 
> Oops yes indeed!
> 
>>>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +	unsigned long flags;
>>>> +	LIST_HEAD(head);
>>>> +	u32 csr;
>>>> +
>>>> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>>>> +
>>>> +	csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
>>>> +	writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
>>>> +		       ahbdma_chan->regs + AHBDMA_CH_CSR);
>>>> +
>>>> +	if (ahbdma_chan->active_tx) {
>>>> +		udelay(AHBDMA_BURST_COMPLETE_TIME);
>>>
>>> Why not poll the status register and wait for the channel to stop?
>>>
>>
>> That probably would also work. But I'm not sure whether status depends on the
>> channels "enable" state..
> 
> Well if it is not enabled, then we probably don't care about the state.
> However, a quick test should tell us.
> 

Okay, I'll try to check it.

>>>> +
>>>> +		writel_relaxed(AHBDMA_CH_IS_EOC,
>>>> +			       ahbdma_chan->regs + AHBDMA_CH_STA);
>>>> +
>>>> +		ahbdma_chan->active_tx = NULL;
>>>> +	}
>>>> +
>>>> +	vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
>>>> +	complete_all(&ahbdma_chan->idling);
>>>> +
>>>> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
>>>> +
>>>> +	vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
>>>> +
>>>> +	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, csr;
>>>> +	unsigned int slave_id;
>>>> +	phys_addr_t ahb_addr;
>>>> +
>>>> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>>> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>>> +		return -EINVAL;
>>>> +
>>>> +	switch (dir) {
>>>> +	case DMA_DEV_TO_MEM:
>>>> +		burst    = sconfig->src_maxburst;
>>>> +		ahb_addr = sconfig->src_addr;
>>>> +		break;
>>>> +	case DMA_MEM_TO_DEV:
>>>> +		burst    = sconfig->dst_maxburst;
>>>> +		ahb_addr = sconfig->dst_addr;
>>>> +		break;
>>>> +	default:
>>>> +		return -EINVAL;
>>>> +	}
>>>> +
>>>> +	switch (burst) {
>>>> +	case 1:
>>>> +		burst = AHBDMA_CH_AHB_BURST_1;
>>>> +		break;
>>>> +	case 4:
>>>> +		burst = AHBDMA_CH_AHB_BURST_4;
>>>> +		break;
>>>> +	case 8:
>>>> +		burst = AHBDMA_CH_AHB_BURST_8;
>>>> +		break;
>>>> +	default:
>>>> +		return -EINVAL;
>>>> +	}
>>>> +
>>>> +	if (ahb_addr & 3)
>>>> +		return -EINVAL;
>>>> +
>>>> +	ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
>>>> +	ahb_seq |= AHBDMA_CH_INTR_ENB;
>>>> +
>>>> +	csr  = AHBDMA_CH_ENABLE;
>>>> +	csr |= AHBDMA_CH_IE_EOC;
>>>> +
>>>> +	if (ahbdma_chan->of_slave || sconfig->device_fc) {
>>>> +		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>>>> +			slave_id = ahbdma_chan->of_req_sel;
>>>> +		else
>>>> +			slave_id = sconfig->slave_id;
>>>> +
>>>> +		if (slave_id > 15)
>>>> +			return -EINVAL;
>>>
>>> Why not ...
>>>
>>> 	if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>>> 		slave_id = ahbdma_chan->of_req_sel;
>>> 	else if (slave_id = sconfig->slave_id < TEGRA_AHBDMA_REQ_N_A)
>>> 		slave_id = sconfig->slave_id;
>>> 	else
>>> 		return -EINVAL;
>>>
>>
>> Because I'm finding variant like yours more difficult to read. I'll stick with
>> my variant if you don't mind.
> 
> Ha! I prefer mine :-)
> 

Okay, I'll switch to yours ;)

>>>> +
>>>> +		ahb_seq |= AHBDMA_CH_ADDR_WRAP;
>>>> +
>>>> +		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
>>>> +		csr |= AHBDMA_CH_FLOW;
>>>> +	}
>>>> +
>>>> +	ahbdma_chan->csr = csr;
>>>> +	ahbdma_chan->ahb_seq = ahb_seq;
>>>> +	ahbdma_chan->ahb_addr = ahb_addr;
>>>> +
>>>> +	return 0;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>>> +{
>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +
>>>> +	wait_for_completion(&ahbdma_chan->idling);
>>>> +	vchan_synchronize(&ahbdma_chan->vchan);
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>>> +{
>>>> +	vchan_free_chan_resources(to_virt_chan(chan));
>>>> +}
>>>> +
>>>> +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_device *dma_dev = &tdma->dma_dev;
>>>> +
>>>> +	vchan_init(&ahbdma_chan->vchan, dma_dev);
>>>> +	init_completion(&ahbdma_chan->idling);
>>>> +	complete(&ahbdma_chan->idling);
>>>> +
>>>> +	ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
>>>> +	ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
>>>> +	ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
>>>> +}
>>>> +
>>>> +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;
>>>> +
>>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>>> +	if (!chan)
>>>> +		return NULL;
>>>> +
>>>> +	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
>>>
>>> Test for args[0] < TEGRA_AHBDMA_REQ_N_A?
>>>
>>
>> It would duplicate slave_id checking done in tegra_ahbdma_config(), so not
>> needed here.
> 
> But surely we should not let them request a channel in the first place?
> 

If allowing client to disable flow control is okay, as you mentioned below, then
I agree that it is fine. I'll make this change.

>>>> +	to_ahbdma_chan(chan)->of_slave = true;
>>>
>>> Is this really needed? Doesn't a value of 0..TEGRA_AHBDMA_REQ_N_A-1 tell
>>> us it is valid?
>>>
>>
>> I think we should enforce channels flow control in a case of OF xlate'd channel,
>> no? To avoid abusing channels usage by client. Seems tegra_ahbdma_config isn't
>> correct, should be:
> 
> Absolutely. However, I don't see the need for the additional 'of_slave'
> variable. If we validate the slave id here, we can get rid of the extra
> variable. It does not simplify the code really by adding this IMO.
> 

'of_slave' enforces flow control enable. If I understand you correctly, you are
suggesting that it is okay to leave ability for clients to override flow
control. Well, that's probably is fine indeed, just keep an eye on client drivers.

>> 	if (ahbdma_chan->of_slave || sconfig->device_fc) {
>> -		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>> +		if (ahbdma_chan->of_slave)
>> 			slave_id = ahbdma_chan->of_req_sel;
>> 		else
>> 			slave_id = sconfig->slave_id;
>>
>> 		if (slave_id >= TEGRA_AHBDMA_REQ_N_A)
>> 			return -EINVAL;
>>
>> I'm finding OF-requsted channel + ability of DMA API to override requested
>> channels parameters quite vague. So I'm not exactly sure how to handle it
>> correctly. It looks like each driver is free to do its own thing, which is kinda
>> a mess. Suggestions?
> 
> I think it is fine how you have it and limit OF-requested channels to
> actual REQ_SEL values.
> 

Okay.
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Dmitry Osipenko Oct. 6, 2017, 6:02 p.m. UTC | #5

On 06.10.2017 18:50, Jon Hunter wrote:
> 
> On 06/10/17 16:26, Dmitry Osipenko wrote:
>> On 06.10.2017 16:11, Jon Hunter wrote:
>>>
>>> On 04/10/17 00:58, 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, mem-to-mem isn't implemented as well.
>>>
>>> The APB DMA does not have h/w support for sg-transfers either, but
>>> transfer request are placed on a list. Can we not do the same for AHB?
>>>
>>
>> We can, but I'm not going to implement it without a use-case. It could be done
>> later if needed.
> 
> OK, that's fine, maybe state that above.
> 

It just occurred to me that you are meaning to state that there is no use-case,
I'll mention it in the commit message.
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Dmitry Osipenko Oct. 6, 2017, 7:11 p.m. UTC | #6

On 04.10.2017 02:58, 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, mem-to-mem isn't implemented as well.
> 
> Signed-off-by: Dmitry Osipenko <[e-mail]>
> ---
>  drivers/dma/Kconfig           |  10 +
>  drivers/dma/Makefile          |   1 +
>  drivers/dma/tegra20-ahb-dma.c | 630 ++++++++++++++++++++++++++++++++++++++++++
>  3 files changed, 641 insertions(+)
>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
> 
> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
> index 04e381b522b4..7d132aa85174 100644
> --- a/drivers/dma/Kconfig
> +++ b/drivers/dma/Kconfig
> @@ -512,6 +512,16 @@ 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 || COMPILE_TEST
> +	select DMA_ENGINE
> +	select DMA_VIRTUAL_CHANNELS
> +	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 a145ad1426bc..f3d284bf6d65 100644
> --- a/drivers/dma/Makefile
> +++ b/drivers/dma/Makefile
> @@ -62,6 +62,7 @@ obj-$(CONFIG_STM32_DMA) += stm32-dma.o
>  obj-$(CONFIG_STM32_DMAMUX) += stm32-dmamux.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..2d176a5536aa
> --- /dev/null
> +++ b/drivers/dma/tegra20-ahb-dma.c
> @@ -0,0 +1,630 @@
> +/*
> + * Copyright 2017 Dmitry Osipenko <[e-mail]>
> + *
> + * 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/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 <dt-bindings/dma/tegra-ahb-dma.h>
> +
> +#include "virt-dma.h"
> +
> +#define AHBDMA_CMD			0x0
> +#define AHBDMA_CMD_ENABLE		BIT(31)
> +
> +#define AHBDMA_IRQ_ENB_MASK		0x20
> +#define AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
> +
> +#define AHBDMA_CH_BASE(ch)		(0x1000 + (ch) * 0x20)
> +
> +#define AHBDMA_CH_CSR			0x0
> +#define AHBDMA_CH_ADDR_WRAP		BIT(18)
> +#define AHBDMA_CH_FLOW			BIT(24)
> +#define AHBDMA_CH_ONCE			BIT(26)
> +#define AHBDMA_CH_DIR_TO_XMB		BIT(27)
> +#define AHBDMA_CH_IE_EOC		BIT(30)
> +#define AHBDMA_CH_ENABLE		BIT(31)
> +#define AHBDMA_CH_REQ_SEL_SHIFT		16
> +#define AHBDMA_CH_WCOUNT_MASK		GENMASK(15, 2)
> +
> +#define AHBDMA_CH_STA			0x4
> +#define AHBDMA_CH_IS_EOC		BIT(30)
> +
> +#define AHBDMA_CH_AHB_PTR		0x10
> +
> +#define AHBDMA_CH_AHB_SEQ		0x14
> +#define AHBDMA_CH_INTR_ENB		BIT(31)
> +#define AHBDMA_CH_AHB_BURST_SHIFT	24
> +#define AHBDMA_CH_AHB_BURST_1		2
> +#define AHBDMA_CH_AHB_BURST_4		3
> +#define AHBDMA_CH_AHB_BURST_8		4
> +
> +#define AHBDMA_CH_XMB_PTR		0x18
> +
> +#define AHBDMA_BUS_WIDTH		BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
> +
> +#define AHBDMA_DIRECTIONS		BIT(DMA_DEV_TO_MEM) | \
> +					BIT(DMA_MEM_TO_DEV)
> +
> +#define AHBDMA_BURST_COMPLETE_TIME	20
> +
> +struct tegra_ahbdma_tx_desc {
> +	struct virt_dma_desc vdesc;
> +	dma_addr_t mem_addr;
> +	phys_addr_t ahb_addr;
> +	u32 ahb_seq;
> +	u32 csr;
> +};
> +
> +struct tegra_ahbdma_chan {
> +	struct tegra_ahbdma_tx_desc *active_tx;
> +	struct virt_dma_chan vchan;
> +	struct completion idling;
> +	void __iomem *regs;
> +	phys_addr_t ahb_addr;
> +	u32 ahb_seq;
> +	u32 csr;
> +	unsigned int of_req_sel;
> +	bool of_slave;
> +};
> +
> +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_chan *to_ahbdma_chan(struct dma_chan *chan)
> +{
> +	return container_of(chan, struct tegra_ahbdma_chan, vchan.chan);
> +}
> +
> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
> +						struct virt_dma_desc *vdesc)
> +{
> +	return container_of(vdesc, struct tegra_ahbdma_tx_desc, vdesc);
> +}
> +
> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
> +						struct tegra_ahbdma_chan *chan)
> +{
> +	struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
> +
> +	if (vdesc)
> +		list_del(&vdesc->node);

I just noticed that this is incorrect. Node must be deleted after TX completion,
otherwise vchan_find_desc won't find TX and residual won't be reported by
dmaengine_tx_status.

Jon, I think you ADMA driver has the same issue, as well as several other DMA
drivers that use virt-dma.

> +
> +	return vdesc ? to_ahbdma_tx_desc(vdesc) : NULL;
> +}
> +
> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
> +
> +	if (tx) {
> +		writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
> +		writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
> +		writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
> +		writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
> +
> +		reinit_completion(&chan->idling);
> +	} else
> +		complete_all(&chan->idling);
> +
> +	chan->active_tx = tx;
> +}
> +
> +static bool tegra_ahbdma_clear_interrupt(struct tegra_ahbdma_chan *chan)
> +{
> +	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
> +
> +	if (status & AHBDMA_CH_IS_EOC) {
> +		writel_relaxed(AHBDMA_CH_IS_EOC, chan->regs + AHBDMA_CH_STA);
> +
> +		return true;
> +	}
> +
> +	return false;
> +}
> +
> +static bool tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
> +{
> +	struct tegra_ahbdma_tx_desc *tx;
> +	unsigned long flags;
> +	bool intr = false;
> +	bool cyclic;
> +
> +	spin_lock_irqsave(&chan->vchan.lock, flags);
> +
> +	tx = chan->active_tx;
> +	if (tx)
> +		intr = tegra_ahbdma_clear_interrupt(chan);
> +
> +	if (intr) {
> +		cyclic = !(tx->csr & AHBDMA_CH_ONCE);
> +
> +		if (!cyclic)
> +			tegra_ahbdma_issue_next_tx(chan);
> +
> +		if (cyclic)
> +			vchan_cyclic_callback(&tx->vdesc);
> +		else
> +			vchan_cookie_complete(&tx->vdesc);
> +	}
> +
> +	spin_unlock_irqrestore(&chan->vchan.lock, flags);
> +
> +	return intr;
> +}
> +
> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
> +{
> +	struct tegra_ahbdma *tdma = dev_id;
> +	bool handled;
> +
> +	handled  = tegra_ahbdma_handle_channel(&tdma->channels[0]);
> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[1]);
> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[2]);
> +	handled |= tegra_ahbdma_handle_channel(&tdma->channels[3]);
> +
> +	return handled ? IRQ_HANDLED : IRQ_NONE;
> +}
> +
> +static void tegra_ahbdma_tx_desc_free(struct virt_dma_desc *vdesc)
> +{
> +	kfree(to_ahbdma_tx_desc(vdesc));
> +}
> +
> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep(
> +					struct dma_chan *chan,
> +					enum dma_transfer_direction dir,
> +					unsigned long flags,
> +					dma_addr_t paddr,
> +					size_t size,
> +					bool cyclic)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct tegra_ahbdma_tx_desc *tx;
> +	u32 csr = ahbdma_chan->csr;
> +
> +	/* size and alignments should fulfill HW requirements */
> +	if (size < 4 || size & 3 || paddr & 3)
> +		return NULL;
> +
> +	tx = kzalloc(sizeof(*tx), GFP_NOWAIT);
> +	if (!tx)
> +		return NULL;
> +
> +	if (dir == DMA_DEV_TO_MEM)
> +		csr |= AHBDMA_CH_DIR_TO_XMB;
> +
> +	if (!cyclic)
> +		csr |= AHBDMA_CH_ONCE;
> +
> +	tx->csr = csr | (size - sizeof(u32));
> +	tx->ahb_seq = ahbdma_chan->ahb_seq;
> +	tx->ahb_addr = ahbdma_chan->ahb_addr;
> +	tx->mem_addr = paddr;
> +
> +	return vchan_tx_prep(&ahbdma_chan->vchan, &tx->vdesc, flags);
> +}
> +
> +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)
> +{
> +	/* unimplemented */
> +	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
> +		return NULL;
> +
> +	return tegra_ahbdma_prep(chan, dir, flags, sg_dma_address(sgl),
> +				 sg_dma_len(sgl), false);
> +}
> +
> +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)
> +{
> +	/* unimplemented */
> +	if (buf_len != period_len || buf_len > SZ_64K)
> +		return NULL;
> +
> +	return tegra_ahbdma_prep(chan, dir, flags, buf_addr, buf_len, true);
> +}
> +
> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +	struct virt_dma_chan *vchan = &ahbdma_chan->vchan;
> +	unsigned long flags;
> +
> +	spin_lock_irqsave(&vchan->lock, flags);
> +
> +	if (vchan_issue_pending(vchan) && !ahbdma_chan->active_tx)
> +		tegra_ahbdma_issue_next_tx(ahbdma_chan);
> +
> +	spin_unlock_irqrestore(&vchan->lock, flags);
> +}
> +
> +static size_t tegra_ahbdma_residual(struct tegra_ahbdma_chan *chan)
> +{
> +	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
> +
> +	return (status & AHBDMA_CH_WCOUNT_MASK);
> +}
> +
> +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;
> +	struct virt_dma_desc *vdesc;
> +	enum dma_status cookie_status;
> +	unsigned long flags;
> +	size_t residual;
> +
> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
> +
> +	cookie_status = dma_cookie_status(chan, cookie, state);
> +	if (cookie_status == DMA_COMPLETE)
> +		goto unlock;
> +
> +	vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
> +	if (!vdesc)
> +		residual = 0;
> +	else {
> +		tx = to_ahbdma_tx_desc(vdesc);
> +
> +		if (tx == ahbdma_chan->active_tx)
> +			residual = tegra_ahbdma_residual(ahbdma_chan);
> +		else
> +			residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
> +
> +		residual += sizeof(u32);
> +	}
> +
> +	dma_set_residue(state, residual);
> +
> +unlock:
> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.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);
> +	unsigned long flags;
> +	LIST_HEAD(head);
> +	u32 csr;
> +
> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
> +
> +	csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
> +	writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
> +		       ahbdma_chan->regs + AHBDMA_CH_CSR);
> +
> +	if (ahbdma_chan->active_tx) {
> +		udelay(AHBDMA_BURST_COMPLETE_TIME);
> +
> +		writel_relaxed(AHBDMA_CH_IS_EOC,
> +			       ahbdma_chan->regs + AHBDMA_CH_STA);
> +
> +		ahbdma_chan->active_tx = NULL;
> +	}
> +
> +	vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
> +	complete_all(&ahbdma_chan->idling);
> +
> +	spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
> +
> +	vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
> +
> +	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, csr;
> +	unsigned int slave_id;
> +	phys_addr_t ahb_addr;
> +
> +	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
> +	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
> +		return -EINVAL;
> +
> +	switch (dir) {
> +	case DMA_DEV_TO_MEM:
> +		burst    = sconfig->src_maxburst;
> +		ahb_addr = sconfig->src_addr;
> +		break;
> +	case DMA_MEM_TO_DEV:
> +		burst    = sconfig->dst_maxburst;
> +		ahb_addr = sconfig->dst_addr;
> +		break;
> +	default:
> +		return -EINVAL;
> +	}
> +
> +	switch (burst) {
> +	case 1:
> +		burst = AHBDMA_CH_AHB_BURST_1;
> +		break;
> +	case 4:
> +		burst = AHBDMA_CH_AHB_BURST_4;
> +		break;
> +	case 8:
> +		burst = AHBDMA_CH_AHB_BURST_8;
> +		break;
> +	default:
> +		return -EINVAL;
> +	}
> +
> +	if (ahb_addr & 3)
> +		return -EINVAL;
> +
> +	ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
> +	ahb_seq |= AHBDMA_CH_INTR_ENB;
> +
> +	csr  = AHBDMA_CH_ENABLE;
> +	csr |= AHBDMA_CH_IE_EOC;
> +
> +	if (ahbdma_chan->of_slave || sconfig->device_fc) {
> +		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
> +			slave_id = ahbdma_chan->of_req_sel;
> +		else
> +			slave_id = sconfig->slave_id;
> +
> +		if (slave_id > 15)
> +			return -EINVAL;
> +
> +		ahb_seq |= AHBDMA_CH_ADDR_WRAP;
> +
> +		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
> +		csr |= AHBDMA_CH_FLOW;
> +	}
> +
> +	ahbdma_chan->csr = csr;
> +	ahbdma_chan->ahb_seq = ahb_seq;
> +	ahbdma_chan->ahb_addr = ahb_addr;
> +
> +	return 0;
> +}
> +
> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
> +{
> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
> +
> +	wait_for_completion(&ahbdma_chan->idling);
> +	vchan_synchronize(&ahbdma_chan->vchan);
> +}
> +
> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
> +{
> +	vchan_free_chan_resources(to_virt_chan(chan));
> +}
> +
> +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_device *dma_dev = &tdma->dma_dev;
> +
> +	vchan_init(&ahbdma_chan->vchan, dma_dev);
> +	init_completion(&ahbdma_chan->idling);
> +	complete(&ahbdma_chan->idling);
> +
> +	ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
> +	ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
> +	ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
> +}
> +
> +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;
> +
> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
> +	if (!chan)
> +		return NULL;
> +
> +	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
> +	to_ahbdma_chan(chan)->of_slave = true;
> +
> +	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(AHBDMA_CMD_ENABLE, tdma->regs + AHBDMA_CMD);
> +
> +	writel_relaxed(AHBDMA_IRQ_ENB_CH(0) |
> +		       AHBDMA_IRQ_ENB_CH(1) |
> +		       AHBDMA_IRQ_ENB_CH(2) |
> +		       AHBDMA_IRQ_ENB_CH(3),
> +		       tdma->regs + 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		= AHBDMA_DIRECTIONS;
> +	dma_dev->src_addr_widths	= AHBDMA_BUS_WIDTH;
> +	dma_dev->dst_addr_widths	= 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 <[e-mail]>");
> +MODULE_LICENSE("GPL");
> 

--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Dmitry Osipenko Oct. 6, 2017, 7:25 p.m. UTC | #7

On 06.10.2017 20:23, Dmitry Osipenko wrote:
> On 06.10.2017 18:50, Jon Hunter wrote:
>>
>> On 06/10/17 16:26, Dmitry Osipenko wrote:
>>> On 06.10.2017 16:11, Jon Hunter wrote:
>>>>
>>>> On 04/10/17 00:58, 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, mem-to-mem isn't implemented as well.
>>>>
>>>> The APB DMA does not have h/w support for sg-transfers either, but
>>>> transfer request are placed on a list. Can we not do the same for AHB?
>>>>
>>>
>>> We can, but I'm not going to implement it without a use-case. It could be done
>>> later if needed.
>>
>> OK, that's fine, maybe state that above.
>>
> 
> Well, I think it is explicitly mentioned in the commit message.
> 
> "Driver doesn't yet implement transfers larger than 64K and scatter-gather
> transfers that have NENT > 1".
> 
> Isn't it enough?
> 
>> [...]
>>
>>>>> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
>>>>> +{
>>>>> +	struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
>>>>> +
>>>>> +	if (tx) {
>>>>> +		writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
>>>>> +		writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
>>>>> +		writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
>>>>> +		writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
>>>>> +
>>>>> +		reinit_completion(&chan->idling);
>>>>
>>>> Should this be done before actually starting the DMA?
>>
>> OK, then that's fine.
>>
>> [...]
>>
>>>>> +	else {
>>>>> +		tx = to_ahbdma_tx_desc(vdesc);
>>>>> +
>>>>> +		if (tx == ahbdma_chan->active_tx)
>>>>> +			residual = tegra_ahbdma_residual(ahbdma_chan);
>>>>> +		else
>>>>> +			residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>>>>> +
>>>>> +		residual += sizeof(u32);
>>>>> +	}
>>>>> +
>>>>> +	dma_set_residue(state, residual);
>>>>
>>>> I believe residue needs to be bytes.
>>
>> Oops yes indeed!
>>
>>>>> +static int tegra_ahbdma_terminate_all(struct dma_chan *chan)
>>>>> +{
>>>>> +	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>>> +	unsigned long flags;
>>>>> +	LIST_HEAD(head);
>>>>> +	u32 csr;
>>>>> +
>>>>> +	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>>>>> +
>>>>> +	csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
>>>>> +	writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
>>>>> +		       ahbdma_chan->regs + AHBDMA_CH_CSR);
>>>>> +
>>>>> +	if (ahbdma_chan->active_tx) {
>>>>> +		udelay(AHBDMA_BURST_COMPLETE_TIME);
>>>>
>>>> Why not poll the status register and wait for the channel to stop?
>>>>
>>>
>>> That probably would also work. But I'm not sure whether status depends on the
>>> channels "enable" state..
>>
>> Well if it is not enabled, then we probably don't care about the state.
>> However, a quick test should tell us.
>>
> 
> Okay, I'll try to check it.
>
I have checked the 'busy' bit and it indeed reflects the real channels busy
state. I'll change code to poll the busy bit instead of the delaying. Thank you
for the suggestion!
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Dmitry Osipenko Oct. 7, 2017, 12:43 p.m. UTC | #8

Hello, Eric

On 07.10.2017 09:21, Eric Pilmore wrote:
> 
> 
> On Fri, Oct 6, 2017 at 12:11 PM, Dmitry Osipenko <[e-mail]
> <mailto:[e-mail]>> wrote:
> 
>     On 04.10.2017 02:58, 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, mem-to-mem isn't implemented as well.
>     >
>     > Signed-off-by: Dmitry Osipenko <[e-mail] <mailto:[e-mail]>>
>     > ---
>     >  drivers/dma/Kconfig           |  10 +
>     >  drivers/dma/Makefile          |   1 +
>     >  drivers/dma/tegra20-ahb-dma.c | 630
>     ++++++++++++++++++++++++++++++++++++++++++
>     >  3 files changed, 641 insertions(+)
>     >  create mode 100644 drivers/dma/tegra20-ahb-dma.c
>     >
>     > diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
>     > index 04e381b522b4..7d132aa85174 100644
>     > --- a/drivers/dma/Kconfig
>     > +++ b/drivers/dma/Kconfig
>     > @@ -512,6 +512,16 @@ 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 || COMPILE_TEST
>     > +     select DMA_ENGINE
>     > +     select DMA_VIRTUAL_CHANNELS
>     > +     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 a145ad1426bc..f3d284bf6d65 100644
>     > --- a/drivers/dma/Makefile
>     > +++ b/drivers/dma/Makefile
>     > @@ -62,6 +62,7 @@ obj-$(CONFIG_STM32_DMA) += stm32-dma.o
>     >  obj-$(CONFIG_STM32_DMAMUX) += stm32-dmamux.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..2d176a5536aa
>     > --- /dev/null
>     > +++ b/drivers/dma/tegra20-ahb-dma.c
>     > @@ -0,0 +1,630 @@
>     > +/*
>     > + * Copyright 2017 Dmitry Osipenko <[e-mail]
>     <mailto:[e-mail]>>
>     > + *
>     > + * 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/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 <dt-bindings/dma/tegra-ahb-dma.h>
>     > +
>     > +#include "virt-dma.h"
>     > +
>     > +#define AHBDMA_CMD                   0x0
>     > +#define AHBDMA_CMD_ENABLE            BIT(31)
>     > +
>     > +#define AHBDMA_IRQ_ENB_MASK          0x20
>     > +#define AHBDMA_IRQ_ENB_CH(ch)                BIT(ch)
>     > +
>     > +#define AHBDMA_CH_BASE(ch)           (0x1000 + (ch) * 0x20)
>     > +
>     > +#define AHBDMA_CH_CSR                        0x0
>     > +#define AHBDMA_CH_ADDR_WRAP          BIT(18)
>     > +#define AHBDMA_CH_FLOW                       BIT(24)
>     > +#define AHBDMA_CH_ONCE                       BIT(26)
>     > +#define AHBDMA_CH_DIR_TO_XMB         BIT(27)
>     > +#define AHBDMA_CH_IE_EOC             BIT(30)
>     > +#define AHBDMA_CH_ENABLE             BIT(31)
>     > +#define AHBDMA_CH_REQ_SEL_SHIFT              16
>     > +#define AHBDMA_CH_WCOUNT_MASK                GENMASK(15, 2)
>     > +
>     > +#define AHBDMA_CH_STA                        0x4
>     > +#define AHBDMA_CH_IS_EOC             BIT(30)
>     > +
>     > +#define AHBDMA_CH_AHB_PTR            0x10
>     > +
>     > +#define AHBDMA_CH_AHB_SEQ            0x14
>     > +#define AHBDMA_CH_INTR_ENB           BIT(31)
>     > +#define AHBDMA_CH_AHB_BURST_SHIFT    24
>     > +#define AHBDMA_CH_AHB_BURST_1                2
>     > +#define AHBDMA_CH_AHB_BURST_4                3
>     > +#define AHBDMA_CH_AHB_BURST_8                4
>     > +
>     > +#define AHBDMA_CH_XMB_PTR            0x18
>     > +
>     > +#define AHBDMA_BUS_WIDTH             BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>     > +
>     > +#define AHBDMA_DIRECTIONS            BIT(DMA_DEV_TO_MEM) | \
>     > +                                     BIT(DMA_MEM_TO_DEV)
>     > +
>     > +#define AHBDMA_BURST_COMPLETE_TIME   20
>     > +
>     > +struct tegra_ahbdma_tx_desc {
>     > +     struct virt_dma_desc vdesc;
>     > +     dma_addr_t mem_addr;
>     > +     phys_addr_t ahb_addr;
>     > +     u32 ahb_seq;
>     > +     u32 csr;
>     > +};
>     > +
>     > +struct tegra_ahbdma_chan {
>     > +     struct tegra_ahbdma_tx_desc *active_tx;
>     > +     struct virt_dma_chan vchan;
>     > +     struct completion idling;
>     > +     void __iomem *regs;
>     > +     phys_addr_t ahb_addr;
>     > +     u32 ahb_seq;
>     > +     u32 csr;
>     > +     unsigned int of_req_sel;
>     > +     bool of_slave;
>     > +};
>     > +
>     > +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_chan *to_ahbdma_chan(struct dma_chan *chan)
>     > +{
>     > +     return container_of(chan, struct tegra_ahbdma_chan, vchan.chan);
>     > +}
>     > +
>     > +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>     > +                                             struct virt_dma_desc *vdesc)
>     > +{
>     > +     return container_of(vdesc, struct tegra_ahbdma_tx_desc, vdesc);
>     > +}
>     > +
>     > +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
>     > +                                             struct tegra_ahbdma_chan *chan)
>     > +{
>     > +     struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
>     > +
>     > +     if (vdesc)
>     > +             list_del(&vdesc->node);
> 
>     I just noticed that this is incorrect. Node must be deleted after TX completion,
>     otherwise vchan_find_desc won't find TX and residual won't be reported by
>     dmaengine_tx_status.
> 
>     Jon, I think you ADMA driver has the same issue, as well as several other DMA
>     drivers that use virt-dma.
> 
>     > +
>     > +     return vdesc ? to_ahbdma_tx_desc(vdesc) : NULL;
>     > +}
>     > +
>     > +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
>     > +{
>     > +     struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
>     > +
>     > +     if (tx) {
>     > +             writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
>     > +             writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
>     > +             writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
>     > +             writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
>     > +
>     > +             reinit_completion(&chan->idling);
>     > +     } else
>     > +             complete_all(&chan->idling);
>     > +
>     > +     chan->active_tx = tx;
>     > +}
>     > +
>     > +static bool tegra_ahbdma_clear_interrupt(struct tegra_ahbdma_chan *chan)
>     > +{
>     > +     u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>     > +
>     > +     if (status & AHBDMA_CH_IS_EOC) {
>     > +             writel_relaxed(AHBDMA_CH_IS_EOC, chan->regs + AHBDMA_CH_STA);
>     > +
>     > +             return true;
>     > +     }
>     > +
>     > +     return false;
>     > +}
>     > +
>     > +static bool tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>     > +{
>     > +     struct tegra_ahbdma_tx_desc *tx;
>     > +     unsigned long flags;
>     > +     bool intr = false;
>     > +     bool cyclic;
>     > +
>     > +     spin_lock_irqsave(&chan->vchan.lock, flags);
>     > +
>     > +     tx = chan->active_tx;
>     > +     if (tx)
>     > +             intr = tegra_ahbdma_clear_interrupt(chan);
>     > +
>     > +     if (intr) {
>     > +             cyclic = !(tx->csr & AHBDMA_CH_ONCE);
>     > +
>     > +             if (!cyclic)
>     > +                     tegra_ahbdma_issue_next_tx(chan);
>     > +
>     > +             if (cyclic)
>     > +                     vchan_cyclic_callback(&tx->vdesc);
>     > +             else
>     > +                     vchan_cookie_complete(&tx->vdesc);
>     > +     }
>     > +
>     > +     spin_unlock_irqrestore(&chan->vchan.lock, flags);
>     > +
>     > +     return intr;
>     > +}
>     > +
>     > +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>     > +{
>     > +     struct tegra_ahbdma *tdma = dev_id;
>     > +     bool handled;
>     > +
>     > +     handled  = tegra_ahbdma_handle_channel(&tdma->channels[0]);
>     > +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[1]);
>     > +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[2]);
>     > +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[3]);
>     > +
>     > +     return handled ? IRQ_HANDLED : IRQ_NONE;
>     > +}
>     > +
>     > +static void tegra_ahbdma_tx_desc_free(struct virt_dma_desc *vdesc)
>     > +{
>     > +     kfree(to_ahbdma_tx_desc(vdesc));
> 
> 
> 
> Can do devm_kfree() here instead.  See devm_kzalloc() comment below and create a
> chan2dev function. Then
> Add the following field to your desc structure.  This will get set when the
> descriptor is created.
> 
> struct tegra_ahmdma_tx_desc {
>      ....
>      struct tegra_ahbdma_chan *tchan;   /* see tegra_ahbdma_prep() */
>      ....
> };
> 
> struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(vdesc);
> struct device *dev = chan2dev(&tx->tchan->vchan.chan);
> devm_kfree(dev, tx);
> 

Unfortunately I'm thinking that your proposal isn't correct:

1) virt-dma manages descriptor allocations for us here, all desc's are free'd by
vchan_free_chan_resources on channels release

2) we want to release all channels descriptors when *channel* is released, using
devm_* just doesn't make sense

> 
>     > +}
>     > +
>     > +static struct dma_async_tx_descriptor *tegra_ahbdma_prep(
>     > +                                     struct dma_chan *chan,
>     > +                                     enum dma_transfer_direction dir,
>     > +                                     unsigned long flags,
>     > +                                     dma_addr_t paddr,
>     > +                                     size_t size,
>     > +                                     bool cyclic)
>     > +{
>     > +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>     > +     struct tegra_ahbdma_tx_desc *tx;
>     > +     u32 csr = ahbdma_chan->csr;
>     > +
>     > +     /* size and alignments should fulfill HW requirements */
>     > +     if (size < 4 || size & 3 || paddr & 3)
>     > +             return NULL;
>     > +
>     > +     tx = kzalloc(sizeof(*tx), GFP_NOWAIT);
> 
> 
> 
> How about using devm_kzalloc() here?  You can get access to your "dev" with a
> function like the following:
> 
> static inline struct device *chan2dev(struct dma_chan *chan)
> {
>     return &chan->dev->device;
> }
> 
> 
>     > +     if (!tx)
>     > +             return NULL;
>     > +
>     > +     if (dir == DMA_DEV_TO_MEM)
>     > +             csr |= AHBDMA_CH_DIR_TO_XMB;
>     > +
>     > +     if (!cyclic)
>     > +             csr |= AHBDMA_CH_ONCE;
>     > +
>     > +     tx->csr = csr | (size - sizeof(u32));
>     > +     tx->ahb_seq = ahbdma_chan->ahb_seq;
>     > +     tx->ahb_addr = ahbdma_chan->ahb_addr;
>     > +     tx->mem_addr = paddr;
> 
> 
> 
> Add setting of suggested new field:
>     tx->tchan = ahbdma_chan;
> 
> 
>     > +
>     > +     return vchan_tx_prep(&ahbdma_chan->vchan, &tx->vdesc, flags);
>     > +}
>     > +
>     > +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)
>     > +{
>     > +     /* unimplemented */
>     > +     if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>     > +             return NULL;
>     > +
>     > +     return tegra_ahbdma_prep(chan, dir, flags, sg_dma_address(sgl),
>     > +                              sg_dma_len(sgl), false);
>     > +}
>     > +
>     > +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)
>     > +{
>     > +     /* unimplemented */
>     > +     if (buf_len != period_len || buf_len > SZ_64K)
>     > +             return NULL;
>     > +
>     > +     return tegra_ahbdma_prep(chan, dir, flags, buf_addr, buf_len, true);
>     > +}
>     > +
>     > +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>     > +{
>     > +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>     > +     struct virt_dma_chan *vchan = &ahbdma_chan->vchan;
>     > +     unsigned long flags;
>     > +
>     > +     spin_lock_irqsave(&vchan->lock, flags);
>     > +
>     > +     if (vchan_issue_pending(vchan) && !ahbdma_chan->active_tx)
>     > +             tegra_ahbdma_issue_next_tx(ahbdma_chan);
>     > +
>     > +     spin_unlock_irqrestore(&vchan->lock, flags);
>     > +}
>     > +
>     > +static size_t tegra_ahbdma_residual(struct tegra_ahbdma_chan *chan)
>     > +{
>     > +     u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>     > +
>     > +     return (status & AHBDMA_CH_WCOUNT_MASK);
>     > +}
>     > +
>     > +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;
>     > +     struct virt_dma_desc *vdesc;
>     > +     enum dma_status cookie_status;
>     > +     unsigned long flags;
>     > +     size_t residual;
>     > +
>     > +     spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>     > +
>     > +     cookie_status = dma_cookie_status(chan, cookie, state);
>     > +     if (cookie_status == DMA_COMPLETE)
>     > +             goto unlock;
>     > +
>     > +     vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
>     > +     if (!vdesc)
>     > +             residual = 0;
>     > +     else {
>     > +             tx = to_ahbdma_tx_desc(vdesc);
>     > +
>     > +             if (tx == ahbdma_chan->active_tx)
>     > +                     residual = tegra_ahbdma_residual(ahbdma_chan);
>     > +             else
>     > +                     residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>     > +
>     > +             residual += sizeof(u32);
>     > +     }
>     > +
>     > +     dma_set_residue(state, residual);
>     > +
>     > +unlock:
>     > +     spin_unlock_irqrestore(&ahbdma_chan->vchan.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);
>     > +     unsigned long flags;
>     > +     LIST_HEAD(head);
>     > +     u32 csr;
>     > +
>     > +     spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>     > +
>     > +     csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
>     > +     writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
>     > +                    ahbdma_chan->regs + AHBDMA_CH_CSR);
>     > +
>     > +     if (ahbdma_chan->active_tx) {
>     > +             udelay(AHBDMA_BURST_COMPLETE_TIME);
>     > +
>     > +             writel_relaxed(AHBDMA_CH_IS_EOC,
>     > +                            ahbdma_chan->regs + AHBDMA_CH_STA);
>     > +
>     > +             ahbdma_chan->active_tx = NULL;
>     > +     }
>     > +
>     > +     vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
>     > +     complete_all(&ahbdma_chan->idling);
>     > +
>     > +     spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
>     > +
>     > +     vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
>     > +
>     > +     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, csr;
>     > +     unsigned int slave_id;
>     > +     phys_addr_t ahb_addr;
>     > +
>     > +     if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>     > +         sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>     > +             return -EINVAL;
>     > +
>     > +     switch (dir) {
>     > +     case DMA_DEV_TO_MEM:
>     > +             burst    = sconfig->src_maxburst;
>     > +             ahb_addr = sconfig->src_addr;
>     > +             break;
>     > +     case DMA_MEM_TO_DEV:
>     > +             burst    = sconfig->dst_maxburst;
>     > +             ahb_addr = sconfig->dst_addr;
>     > +             break;
>     > +     default:
>     > +             return -EINVAL;
>     > +     }
>     > +
>     > +     switch (burst) {
>     > +     case 1:
>     > +             burst = AHBDMA_CH_AHB_BURST_1;
>     > +             break;
>     > +     case 4:
>     > +             burst = AHBDMA_CH_AHB_BURST_4;
>     > +             break;
>     > +     case 8:
>     > +             burst = AHBDMA_CH_AHB_BURST_8;
>     > +             break;
>     > +     default:
>     > +             return -EINVAL;
>     > +     }
>     > +
>     > +     if (ahb_addr & 3)
>     > +             return -EINVAL;
>     > +
>     > +     ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
>     > +     ahb_seq |= AHBDMA_CH_INTR_ENB;
>     > +
>     > +     csr  = AHBDMA_CH_ENABLE;
>     > +     csr |= AHBDMA_CH_IE_EOC;
>     > +
>     > +     if (ahbdma_chan->of_slave || sconfig->device_fc) {
>     > +             if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>     > +                     slave_id = ahbdma_chan->of_req_sel;
>     > +             else
>     > +                     slave_id = sconfig->slave_id;
>     > +
>     > +             if (slave_id > 15)
>     > +                     return -EINVAL;
>     > +
>     > +             ahb_seq |= AHBDMA_CH_ADDR_WRAP;
>     > +
>     > +             csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
>     > +             csr |= AHBDMA_CH_FLOW;
>     > +     }
>     > +
>     > +     ahbdma_chan->csr = csr;
>     > +     ahbdma_chan->ahb_seq = ahb_seq;
>     > +     ahbdma_chan->ahb_addr = ahb_addr;
>     > +
>     > +     return 0;
>     > +}
>     > +
>     > +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>     > +{
>     > +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>     > +
>     > +     wait_for_completion(&ahbdma_chan->idling);
>     > +     vchan_synchronize(&ahbdma_chan->vchan);
>     > +}
>     > +
>     > +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>     > +{
>     > +     vchan_free_chan_resources(to_virt_chan(chan));
>     > +}
>     > +
>     > +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_device *dma_dev = &tdma->dma_dev;
>     > +
>     > +     vchan_init(&ahbdma_chan->vchan, dma_dev);
>     > +     init_completion(&ahbdma_chan->idling);
>     > +     complete(&ahbdma_chan->idling);
>     > +
>     > +     ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
>     > +     ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
>     > +     ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
>     > +}
>     > +
>     > +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;
>     > +
>     > +     chan = dma_get_any_slave_channel(&tdma->dma_dev);
>     > +     if (!chan)
>     > +             return NULL;
>     > +
>     > +     to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
>     > +     to_ahbdma_chan(chan)->of_slave = true;
>     > +
>     > +     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(AHBDMA_CMD_ENABLE, tdma->regs + AHBDMA_CMD);
>     > +
>     > +     writel_relaxed(AHBDMA_IRQ_ENB_CH(0) |
>     > +                    AHBDMA_IRQ_ENB_CH(1) |
>     > +                    AHBDMA_IRQ_ENB_CH(2) |
>     > +                    AHBDMA_IRQ_ENB_CH(3),
>     > +                    tdma->regs + 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             = AHBDMA_DIRECTIONS;
>     > +     dma_dev->src_addr_widths        = AHBDMA_BUS_WIDTH;
>     > +     dma_dev->dst_addr_widths        = 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 <[e-mail]
>     <mailto:[e-mail]>>");
>     > +MODULE_LICENSE("GPL");
>     >
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Eric Pilmore Oct. 7, 2017, 2:42 p.m. UTC | #9

Hi Dmitry,

My impression of virt-dma was that although it manages the allocations of the descriptors it does this through the wrappers which ultimately call into the underlying driver’s (e.g. Tegra) wrapper functions to do the actual kzalloc/kfree. I’m only suggesting to replace the kzalloc/kfree calls that the Tegra driver is already doing for the descriptors with devm_kzalloc/devm_free instead.

Not a critical change. Kzalloc/kfree are fine as-is. Usage of the devm_ functions instead is just more as a precaution to ensure when the driver is unloaded any forgotten allocations are removed/freed and we don’t end up with “lost” memory. At least this is my understanding of the devm_ memory allocation functions. Although maybe I’m missing something.

Regards,
Eric



Sent from my iPhone

> On Oct 7, 2017, at 5:43 AM, Dmitry Osipenko <[e-mail]> wrote:
> 
> Hello, Eric
> 
>> On 07.10.2017 09:21, Eric Pilmore wrote:
>> 
>> 
>> On Fri, Oct 6, 2017 at 12:11 PM, Dmitry Osipenko <[e-mail]
>> <mailto:[e-mail]>> wrote:
>> 
>>>    On 04.10.2017 02:58, 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, mem-to-mem isn't implemented as well.
>>> 
>>> Signed-off-by: Dmitry Osipenko <[e-mail] <mailto:[e-mail]>>
>>> ---
>>>   drivers/dma/Kconfig           |  10 +
>>>   drivers/dma/Makefile          |   1 +
>>>   drivers/dma/tegra20-ahb-dma.c | 630
>>    ++++++++++++++++++++++++++++++++++++++++++
>>>   3 files changed, 641 insertions(+)
>>>   create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>> 
>>> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
>>> index 04e381b522b4..7d132aa85174 100644
>>> --- a/drivers/dma/Kconfig
>>> +++ b/drivers/dma/Kconfig
>>> @@ -512,6 +512,16 @@ 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 || COMPILE_TEST
>>> +     select DMA_ENGINE
>>> +     select DMA_VIRTUAL_CHANNELS
>>> +     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 a145ad1426bc..f3d284bf6d65 100644
>>> --- a/drivers/dma/Makefile
>>> +++ b/drivers/dma/Makefile
>>> @@ -62,6 +62,7 @@ obj-$(CONFIG_STM32_DMA) += stm32-dma.o
>>>   obj-$(CONFIG_STM32_DMAMUX) += stm32-dmamux.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..2d176a5536aa
>>> --- /dev/null
>>> +++ b/drivers/dma/tegra20-ahb-dma.c
>>> @@ -0,0 +1,630 @@
>>> +/*
>>> + * Copyright 2017 Dmitry Osipenko <[e-mail]
>>    <mailto:[e-mail]>>
>>> + *
>>> + * 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/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 <dt-bindings/dma/tegra-ahb-dma.h>
>>> +
>>> +#include "virt-dma.h"
>>> +
>>> +#define AHBDMA_CMD                   0x0
>>> +#define AHBDMA_CMD_ENABLE            BIT(31)
>>> +
>>> +#define AHBDMA_IRQ_ENB_MASK          0x20
>>> +#define AHBDMA_IRQ_ENB_CH(ch)                BIT(ch)
>>> +
>>> +#define AHBDMA_CH_BASE(ch)           (0x1000 + (ch) * 0x20)
>>> +
>>> +#define AHBDMA_CH_CSR                        0x0
>>> +#define AHBDMA_CH_ADDR_WRAP          BIT(18)
>>> +#define AHBDMA_CH_FLOW                       BIT(24)
>>> +#define AHBDMA_CH_ONCE                       BIT(26)
>>> +#define AHBDMA_CH_DIR_TO_XMB         BIT(27)
>>> +#define AHBDMA_CH_IE_EOC             BIT(30)
>>> +#define AHBDMA_CH_ENABLE             BIT(31)
>>> +#define AHBDMA_CH_REQ_SEL_SHIFT              16
>>> +#define AHBDMA_CH_WCOUNT_MASK                GENMASK(15, 2)
>>> +
>>> +#define AHBDMA_CH_STA                        0x4
>>> +#define AHBDMA_CH_IS_EOC             BIT(30)
>>> +
>>> +#define AHBDMA_CH_AHB_PTR            0x10
>>> +
>>> +#define AHBDMA_CH_AHB_SEQ            0x14
>>> +#define AHBDMA_CH_INTR_ENB           BIT(31)
>>> +#define AHBDMA_CH_AHB_BURST_SHIFT    24
>>> +#define AHBDMA_CH_AHB_BURST_1                2
>>> +#define AHBDMA_CH_AHB_BURST_4                3
>>> +#define AHBDMA_CH_AHB_BURST_8                4
>>> +
>>> +#define AHBDMA_CH_XMB_PTR            0x18
>>> +
>>> +#define AHBDMA_BUS_WIDTH             BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>>> +
>>> +#define AHBDMA_DIRECTIONS            BIT(DMA_DEV_TO_MEM) | \
>>> +                                     BIT(DMA_MEM_TO_DEV)
>>> +
>>> +#define AHBDMA_BURST_COMPLETE_TIME   20
>>> +
>>> +struct tegra_ahbdma_tx_desc {
>>> +     struct virt_dma_desc vdesc;
>>> +     dma_addr_t mem_addr;
>>> +     phys_addr_t ahb_addr;
>>> +     u32 ahb_seq;
>>> +     u32 csr;
>>> +};
>>> +
>>> +struct tegra_ahbdma_chan {
>>> +     struct tegra_ahbdma_tx_desc *active_tx;
>>> +     struct virt_dma_chan vchan;
>>> +     struct completion idling;
>>> +     void __iomem *regs;
>>> +     phys_addr_t ahb_addr;
>>> +     u32 ahb_seq;
>>> +     u32 csr;
>>> +     unsigned int of_req_sel;
>>> +     bool of_slave;
>>> +};
>>> +
>>> +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_chan *to_ahbdma_chan(struct dma_chan *chan)
>>> +{
>>> +     return container_of(chan, struct tegra_ahbdma_chan, vchan.chan);
>>> +}
>>> +
>>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>>> +                                             struct virt_dma_desc *vdesc)
>>> +{
>>> +     return container_of(vdesc, struct tegra_ahbdma_tx_desc, vdesc);
>>> +}
>>> +
>>> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
>>> +                                             struct tegra_ahbdma_chan *chan)
>>> +{
>>> +     struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
>>> +
>>> +     if (vdesc)
>>> +             list_del(&vdesc->node);
>> 
>>    I just noticed that this is incorrect. Node must be deleted after TX completion,
>>    otherwise vchan_find_desc won't find TX and residual won't be reported by
>>    dmaengine_tx_status.
>> 
>>    Jon, I think you ADMA driver has the same issue, as well as several other DMA
>>    drivers that use virt-dma.
>> 
>>> +
>>> +     return vdesc ? to_ahbdma_tx_desc(vdesc) : NULL;
>>> +}
>>> +
>>> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +     struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
>>> +
>>> +     if (tx) {
>>> +             writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
>>> +             writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
>>> +             writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
>>> +             writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
>>> +
>>> +             reinit_completion(&chan->idling);
>>> +     } else
>>> +             complete_all(&chan->idling);
>>> +
>>> +     chan->active_tx = tx;
>>> +}
>>> +
>>> +static bool tegra_ahbdma_clear_interrupt(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +     u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>>> +
>>> +     if (status & AHBDMA_CH_IS_EOC) {
>>> +             writel_relaxed(AHBDMA_CH_IS_EOC, chan->regs + AHBDMA_CH_STA);
>>> +
>>> +             return true;
>>> +     }
>>> +
>>> +     return false;
>>> +}
>>> +
>>> +static bool tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +     struct tegra_ahbdma_tx_desc *tx;
>>> +     unsigned long flags;
>>> +     bool intr = false;
>>> +     bool cyclic;
>>> +
>>> +     spin_lock_irqsave(&chan->vchan.lock, flags);
>>> +
>>> +     tx = chan->active_tx;
>>> +     if (tx)
>>> +             intr = tegra_ahbdma_clear_interrupt(chan);
>>> +
>>> +     if (intr) {
>>> +             cyclic = !(tx->csr & AHBDMA_CH_ONCE);
>>> +
>>> +             if (!cyclic)
>>> +                     tegra_ahbdma_issue_next_tx(chan);
>>> +
>>> +             if (cyclic)
>>> +                     vchan_cyclic_callback(&tx->vdesc);
>>> +             else
>>> +                     vchan_cookie_complete(&tx->vdesc);
>>> +     }
>>> +
>>> +     spin_unlock_irqrestore(&chan->vchan.lock, flags);
>>> +
>>> +     return intr;
>>> +}
>>> +
>>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>>> +{
>>> +     struct tegra_ahbdma *tdma = dev_id;
>>> +     bool handled;
>>> +
>>> +     handled  = tegra_ahbdma_handle_channel(&tdma->channels[0]);
>>> +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[1]);
>>> +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[2]);
>>> +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[3]);
>>> +
>>> +     return handled ? IRQ_HANDLED : IRQ_NONE;
>>> +}
>>> +
>>> +static void tegra_ahbdma_tx_desc_free(struct virt_dma_desc *vdesc)
>>> +{
>>> +     kfree(to_ahbdma_tx_desc(vdesc));
>> 
>> 
>> 
>> Can do devm_kfree() here instead.  See devm_kzalloc() comment below and create a
>> chan2dev function. Then
>> Add the following field to your desc structure.  This will get set when the
>> descriptor is created.
>> 
>> struct tegra_ahmdma_tx_desc {
>>      ....
>>      struct tegra_ahbdma_chan *tchan;   /* see tegra_ahbdma_prep() */
>>      ....
>> };
>> 
>> struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(vdesc);
>> struct device *dev = chan2dev(&tx->tchan->vchan.chan);
>> devm_kfree(dev, tx);
>> 
> 
> Unfortunately I'm thinking that your proposal isn't correct:
> 
> 1) virt-dma manages descriptor allocations for us here, all desc's are free'd by
> vchan_free_chan_resources on channels release
> 
> 2) we want to release all channels descriptors when *channel* is released, using
> devm_* just doesn't make sense
> 
>> 
>>> +}
>>> +
>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep(
>>> +                                     struct dma_chan *chan,
>>> +                                     enum dma_transfer_direction dir,
>>> +                                     unsigned long flags,
>>> +                                     dma_addr_t paddr,
>>> +                                     size_t size,
>>> +                                     bool cyclic)
>>> +{
>>> +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +     struct tegra_ahbdma_tx_desc *tx;
>>> +     u32 csr = ahbdma_chan->csr;
>>> +
>>> +     /* size and alignments should fulfill HW requirements */
>>> +     if (size < 4 || size & 3 || paddr & 3)
>>> +             return NULL;
>>> +
>>> +     tx = kzalloc(sizeof(*tx), GFP_NOWAIT);
>> 
>> 
>> 
>> How about using devm_kzalloc() here?  You can get access to your "dev" with a
>> function like the following:
>> 
>> static inline struct device *chan2dev(struct dma_chan *chan)
>> {
>>     return &chan->dev->device;
>> }
>> 
>> 
>>> +     if (!tx)
>>> +             return NULL;
>>> +
>>> +     if (dir == DMA_DEV_TO_MEM)
>>> +             csr |= AHBDMA_CH_DIR_TO_XMB;
>>> +
>>> +     if (!cyclic)
>>> +             csr |= AHBDMA_CH_ONCE;
>>> +
>>> +     tx->csr = csr | (size - sizeof(u32));
>>> +     tx->ahb_seq = ahbdma_chan->ahb_seq;
>>> +     tx->ahb_addr = ahbdma_chan->ahb_addr;
>>> +     tx->mem_addr = paddr;
>> 
>> 
>> 
>> Add setting of suggested new field:
>>     tx->tchan = ahbdma_chan;
>> 
>> 
>>> +
>>> +     return vchan_tx_prep(&ahbdma_chan->vchan, &tx->vdesc, flags);
>>> +}
>>> +
>>> +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)
>>> +{
>>> +     /* unimplemented */
>>> +     if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>>> +             return NULL;
>>> +
>>> +     return tegra_ahbdma_prep(chan, dir, flags, sg_dma_address(sgl),
>>> +                              sg_dma_len(sgl), false);
>>> +}
>>> +
>>> +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)
>>> +{
>>> +     /* unimplemented */
>>> +     if (buf_len != period_len || buf_len > SZ_64K)
>>> +             return NULL;
>>> +
>>> +     return tegra_ahbdma_prep(chan, dir, flags, buf_addr, buf_len, true);
>>> +}
>>> +
>>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>>> +{
>>> +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +     struct virt_dma_chan *vchan = &ahbdma_chan->vchan;
>>> +     unsigned long flags;
>>> +
>>> +     spin_lock_irqsave(&vchan->lock, flags);
>>> +
>>> +     if (vchan_issue_pending(vchan) && !ahbdma_chan->active_tx)
>>> +             tegra_ahbdma_issue_next_tx(ahbdma_chan);
>>> +
>>> +     spin_unlock_irqrestore(&vchan->lock, flags);
>>> +}
>>> +
>>> +static size_t tegra_ahbdma_residual(struct tegra_ahbdma_chan *chan)
>>> +{
>>> +     u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>>> +
>>> +     return (status & AHBDMA_CH_WCOUNT_MASK);
>>> +}
>>> +
>>> +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;
>>> +     struct virt_dma_desc *vdesc;
>>> +     enum dma_status cookie_status;
>>> +     unsigned long flags;
>>> +     size_t residual;
>>> +
>>> +     spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>>> +
>>> +     cookie_status = dma_cookie_status(chan, cookie, state);
>>> +     if (cookie_status == DMA_COMPLETE)
>>> +             goto unlock;
>>> +
>>> +     vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
>>> +     if (!vdesc)
>>> +             residual = 0;
>>> +     else {
>>> +             tx = to_ahbdma_tx_desc(vdesc);
>>> +
>>> +             if (tx == ahbdma_chan->active_tx)
>>> +                     residual = tegra_ahbdma_residual(ahbdma_chan);
>>> +             else
>>> +                     residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>>> +
>>> +             residual += sizeof(u32);
>>> +     }
>>> +
>>> +     dma_set_residue(state, residual);
>>> +
>>> +unlock:
>>> +     spin_unlock_irqrestore(&ahbdma_chan->vchan.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);
>>> +     unsigned long flags;
>>> +     LIST_HEAD(head);
>>> +     u32 csr;
>>> +
>>> +     spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>>> +
>>> +     csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
>>> +     writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
>>> +                    ahbdma_chan->regs + AHBDMA_CH_CSR);
>>> +
>>> +     if (ahbdma_chan->active_tx) {
>>> +             udelay(AHBDMA_BURST_COMPLETE_TIME);
>>> +
>>> +             writel_relaxed(AHBDMA_CH_IS_EOC,
>>> +                            ahbdma_chan->regs + AHBDMA_CH_STA);
>>> +
>>> +             ahbdma_chan->active_tx = NULL;
>>> +     }
>>> +
>>> +     vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
>>> +     complete_all(&ahbdma_chan->idling);
>>> +
>>> +     spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
>>> +
>>> +     vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
>>> +
>>> +     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, csr;
>>> +     unsigned int slave_id;
>>> +     phys_addr_t ahb_addr;
>>> +
>>> +     if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>> +         sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>> +             return -EINVAL;
>>> +
>>> +     switch (dir) {
>>> +     case DMA_DEV_TO_MEM:
>>> +             burst    = sconfig->src_maxburst;
>>> +             ahb_addr = sconfig->src_addr;
>>> +             break;
>>> +     case DMA_MEM_TO_DEV:
>>> +             burst    = sconfig->dst_maxburst;
>>> +             ahb_addr = sconfig->dst_addr;
>>> +             break;
>>> +     default:
>>> +             return -EINVAL;
>>> +     }
>>> +
>>> +     switch (burst) {
>>> +     case 1:
>>> +             burst = AHBDMA_CH_AHB_BURST_1;
>>> +             break;
>>> +     case 4:
>>> +             burst = AHBDMA_CH_AHB_BURST_4;
>>> +             break;
>>> +     case 8:
>>> +             burst = AHBDMA_CH_AHB_BURST_8;
>>> +             break;
>>> +     default:
>>> +             return -EINVAL;
>>> +     }
>>> +
>>> +     if (ahb_addr & 3)
>>> +             return -EINVAL;
>>> +
>>> +     ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
>>> +     ahb_seq |= AHBDMA_CH_INTR_ENB;
>>> +
>>> +     csr  = AHBDMA_CH_ENABLE;
>>> +     csr |= AHBDMA_CH_IE_EOC;
>>> +
>>> +     if (ahbdma_chan->of_slave || sconfig->device_fc) {
>>> +             if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>>> +                     slave_id = ahbdma_chan->of_req_sel;
>>> +             else
>>> +                     slave_id = sconfig->slave_id;
>>> +
>>> +             if (slave_id > 15)
>>> +                     return -EINVAL;
>>> +
>>> +             ahb_seq |= AHBDMA_CH_ADDR_WRAP;
>>> +
>>> +             csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
>>> +             csr |= AHBDMA_CH_FLOW;
>>> +     }
>>> +
>>> +     ahbdma_chan->csr = csr;
>>> +     ahbdma_chan->ahb_seq = ahb_seq;
>>> +     ahbdma_chan->ahb_addr = ahb_addr;
>>> +
>>> +     return 0;
>>> +}
>>> +
>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>> +{
>>> +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>> +
>>> +     wait_for_completion(&ahbdma_chan->idling);
>>> +     vchan_synchronize(&ahbdma_chan->vchan);
>>> +}
>>> +
>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>> +{
>>> +     vchan_free_chan_resources(to_virt_chan(chan));
>>> +}
>>> +
>>> +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_device *dma_dev = &tdma->dma_dev;
>>> +
>>> +     vchan_init(&ahbdma_chan->vchan, dma_dev);
>>> +     init_completion(&ahbdma_chan->idling);
>>> +     complete(&ahbdma_chan->idling);
>>> +
>>> +     ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
>>> +     ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
>>> +     ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
>>> +}
>>> +
>>> +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;
>>> +
>>> +     chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>> +     if (!chan)
>>> +             return NULL;
>>> +
>>> +     to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
>>> +     to_ahbdma_chan(chan)->of_slave = true;
>>> +
>>> +     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(AHBDMA_CMD_ENABLE, tdma->regs + AHBDMA_CMD);
>>> +
>>> +     writel_relaxed(AHBDMA_IRQ_ENB_CH(0) |
>>> +                    AHBDMA_IRQ_ENB_CH(1) |
>>> +                    AHBDMA_IRQ_ENB_CH(2) |
>>> +                    AHBDMA_IRQ_ENB_CH(3),
>>> +                    tdma->regs + 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             = AHBDMA_DIRECTIONS;
>>> +     dma_dev->src_addr_widths        = AHBDMA_BUS_WIDTH;
>>> +     dma_dev->dst_addr_widths        = 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 <[e-mail]
>>    <mailto:[e-mail]>>");
>>> +MODULE_LICENSE("GPL");
>>> 
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Dmitry Osipenko Oct. 7, 2017, 3:50 p.m. UTC | #10

On 07.10.2017 17:42, Eric Pilmore (GigaIO) wrote:
> Hi Dmitry,
> 
> My impression of virt-dma was that although it manages the allocations of the descriptors it does this through the wrappers which ultimately call into the underlying driver’s (e.g. Tegra) wrapper functions to do the actual kzalloc/kfree. I’m only suggesting to replace the kzalloc/kfree calls that the Tegra driver is already doing for the descriptors with devm_kzalloc/devm_free instead. 

The calls are happening in both directions. Yeah, pretty sure I understand what
you are suggesting.

> Not a critical change. Kzalloc/kfree are fine as-is. Usage of the devm_ functions instead is just more as a precaution to ensure when the driver is unloaded any forgotten allocations are removed/freed and we don’t end up with “lost” memory. At least this is my understanding of the devm_ memory allocation functions. Although maybe I’m missing something.

Well, we shouldn't be caution of a leaking descriptors possibility because it
would be a kinda very severe bug that would affect other drivers as well, that
bug must be fixed properly instead of trying to mask it. So a use of
client-managed allocations isn't necessary in this case, it would only add some
extra burden to the allocations without any profit.

I think client-managed allocations should be only used for the static
allocations, like those that are done on drivers probe and are alive till
drivers removal. Simply because of an extra resources usage per-allocation and
increased allocation latency.

Anyway, thank you for looking at the patch! Please let me know if you'll spot
anything else that potentially could be improved ;)

>> On Oct 7, 2017, at 5:43 AM, Dmitry Osipenko <[e-mail]> wrote:
>>
>> Hello, Eric
>>
>>> On 07.10.2017 09:21, Eric Pilmore wrote:
>>>
>>>
>>> On Fri, Oct 6, 2017 at 12:11 PM, Dmitry Osipenko <[e-mail]
>>> <mailto:[e-mail]>> wrote:
>>>
>>>>    On 04.10.2017 02:58, 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, mem-to-mem isn't implemented as well.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <[e-mail] <mailto:[e-mail]>>
>>>> ---
>>>>   drivers/dma/Kconfig           |  10 +
>>>>   drivers/dma/Makefile          |   1 +
>>>>   drivers/dma/tegra20-ahb-dma.c | 630
>>>    ++++++++++++++++++++++++++++++++++++++++++
>>>>   3 files changed, 641 insertions(+)
>>>>   create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>>>
>>>> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
>>>> index 04e381b522b4..7d132aa85174 100644
>>>> --- a/drivers/dma/Kconfig
>>>> +++ b/drivers/dma/Kconfig
>>>> @@ -512,6 +512,16 @@ 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 || COMPILE_TEST
>>>> +     select DMA_ENGINE
>>>> +     select DMA_VIRTUAL_CHANNELS
>>>> +     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 a145ad1426bc..f3d284bf6d65 100644
>>>> --- a/drivers/dma/Makefile
>>>> +++ b/drivers/dma/Makefile
>>>> @@ -62,6 +62,7 @@ obj-$(CONFIG_STM32_DMA) += stm32-dma.o
>>>>   obj-$(CONFIG_STM32_DMAMUX) += stm32-dmamux.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..2d176a5536aa
>>>> --- /dev/null
>>>> +++ b/drivers/dma/tegra20-ahb-dma.c
>>>> @@ -0,0 +1,630 @@
>>>> +/*
>>>> + * Copyright 2017 Dmitry Osipenko <[e-mail]
>>>    <mailto:[e-mail]>>
>>>> + *
>>>> + * 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/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 <dt-bindings/dma/tegra-ahb-dma.h>
>>>> +
>>>> +#include "virt-dma.h"
>>>> +
>>>> +#define AHBDMA_CMD                   0x0
>>>> +#define AHBDMA_CMD_ENABLE            BIT(31)
>>>> +
>>>> +#define AHBDMA_IRQ_ENB_MASK          0x20
>>>> +#define AHBDMA_IRQ_ENB_CH(ch)                BIT(ch)
>>>> +
>>>> +#define AHBDMA_CH_BASE(ch)           (0x1000 + (ch) * 0x20)
>>>> +
>>>> +#define AHBDMA_CH_CSR                        0x0
>>>> +#define AHBDMA_CH_ADDR_WRAP          BIT(18)
>>>> +#define AHBDMA_CH_FLOW                       BIT(24)
>>>> +#define AHBDMA_CH_ONCE                       BIT(26)
>>>> +#define AHBDMA_CH_DIR_TO_XMB         BIT(27)
>>>> +#define AHBDMA_CH_IE_EOC             BIT(30)
>>>> +#define AHBDMA_CH_ENABLE             BIT(31)
>>>> +#define AHBDMA_CH_REQ_SEL_SHIFT              16
>>>> +#define AHBDMA_CH_WCOUNT_MASK                GENMASK(15, 2)
>>>> +
>>>> +#define AHBDMA_CH_STA                        0x4
>>>> +#define AHBDMA_CH_IS_EOC             BIT(30)
>>>> +
>>>> +#define AHBDMA_CH_AHB_PTR            0x10
>>>> +
>>>> +#define AHBDMA_CH_AHB_SEQ            0x14
>>>> +#define AHBDMA_CH_INTR_ENB           BIT(31)
>>>> +#define AHBDMA_CH_AHB_BURST_SHIFT    24
>>>> +#define AHBDMA_CH_AHB_BURST_1                2
>>>> +#define AHBDMA_CH_AHB_BURST_4                3
>>>> +#define AHBDMA_CH_AHB_BURST_8                4
>>>> +
>>>> +#define AHBDMA_CH_XMB_PTR            0x18
>>>> +
>>>> +#define AHBDMA_BUS_WIDTH             BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
>>>> +
>>>> +#define AHBDMA_DIRECTIONS            BIT(DMA_DEV_TO_MEM) | \
>>>> +                                     BIT(DMA_MEM_TO_DEV)
>>>> +
>>>> +#define AHBDMA_BURST_COMPLETE_TIME   20
>>>> +
>>>> +struct tegra_ahbdma_tx_desc {
>>>> +     struct virt_dma_desc vdesc;
>>>> +     dma_addr_t mem_addr;
>>>> +     phys_addr_t ahb_addr;
>>>> +     u32 ahb_seq;
>>>> +     u32 csr;
>>>> +};
>>>> +
>>>> +struct tegra_ahbdma_chan {
>>>> +     struct tegra_ahbdma_tx_desc *active_tx;
>>>> +     struct virt_dma_chan vchan;
>>>> +     struct completion idling;
>>>> +     void __iomem *regs;
>>>> +     phys_addr_t ahb_addr;
>>>> +     u32 ahb_seq;
>>>> +     u32 csr;
>>>> +     unsigned int of_req_sel;
>>>> +     bool of_slave;
>>>> +};
>>>> +
>>>> +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_chan *to_ahbdma_chan(struct dma_chan *chan)
>>>> +{
>>>> +     return container_of(chan, struct tegra_ahbdma_chan, vchan.chan);
>>>> +}
>>>> +
>>>> +static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
>>>> +                                             struct virt_dma_desc *vdesc)
>>>> +{
>>>> +     return container_of(vdesc, struct tegra_ahbdma_tx_desc, vdesc);
>>>> +}
>>>> +
>>>> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
>>>> +                                             struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +     struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
>>>> +
>>>> +     if (vdesc)
>>>> +             list_del(&vdesc->node);
>>>
>>>    I just noticed that this is incorrect. Node must be deleted after TX completion,
>>>    otherwise vchan_find_desc won't find TX and residual won't be reported by
>>>    dmaengine_tx_status.
>>>
>>>    Jon, I think you ADMA driver has the same issue, as well as several other DMA
>>>    drivers that use virt-dma.
>>>
>>>> +
>>>> +     return vdesc ? to_ahbdma_tx_desc(vdesc) : NULL;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +     struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
>>>> +
>>>> +     if (tx) {
>>>> +             writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
>>>> +             writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
>>>> +             writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
>>>> +             writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
>>>> +
>>>> +             reinit_completion(&chan->idling);
>>>> +     } else
>>>> +             complete_all(&chan->idling);
>>>> +
>>>> +     chan->active_tx = tx;
>>>> +}
>>>> +
>>>> +static bool tegra_ahbdma_clear_interrupt(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +     u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>>>> +
>>>> +     if (status & AHBDMA_CH_IS_EOC) {
>>>> +             writel_relaxed(AHBDMA_CH_IS_EOC, chan->regs + AHBDMA_CH_STA);
>>>> +
>>>> +             return true;
>>>> +     }
>>>> +
>>>> +     return false;
>>>> +}
>>>> +
>>>> +static bool tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +     struct tegra_ahbdma_tx_desc *tx;
>>>> +     unsigned long flags;
>>>> +     bool intr = false;
>>>> +     bool cyclic;
>>>> +
>>>> +     spin_lock_irqsave(&chan->vchan.lock, flags);
>>>> +
>>>> +     tx = chan->active_tx;
>>>> +     if (tx)
>>>> +             intr = tegra_ahbdma_clear_interrupt(chan);
>>>> +
>>>> +     if (intr) {
>>>> +             cyclic = !(tx->csr & AHBDMA_CH_ONCE);
>>>> +
>>>> +             if (!cyclic)
>>>> +                     tegra_ahbdma_issue_next_tx(chan);
>>>> +
>>>> +             if (cyclic)
>>>> +                     vchan_cyclic_callback(&tx->vdesc);
>>>> +             else
>>>> +                     vchan_cookie_complete(&tx->vdesc);
>>>> +     }
>>>> +
>>>> +     spin_unlock_irqrestore(&chan->vchan.lock, flags);
>>>> +
>>>> +     return intr;
>>>> +}
>>>> +
>>>> +static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
>>>> +{
>>>> +     struct tegra_ahbdma *tdma = dev_id;
>>>> +     bool handled;
>>>> +
>>>> +     handled  = tegra_ahbdma_handle_channel(&tdma->channels[0]);
>>>> +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[1]);
>>>> +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[2]);
>>>> +     handled |= tegra_ahbdma_handle_channel(&tdma->channels[3]);
>>>> +
>>>> +     return handled ? IRQ_HANDLED : IRQ_NONE;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_tx_desc_free(struct virt_dma_desc *vdesc)
>>>> +{
>>>> +     kfree(to_ahbdma_tx_desc(vdesc));
>>>
>>>
>>>
>>> Can do devm_kfree() here instead.  See devm_kzalloc() comment below and create a
>>> chan2dev function. Then
>>> Add the following field to your desc structure.  This will get set when the
>>> descriptor is created.
>>>
>>> struct tegra_ahmdma_tx_desc {
>>>      ....
>>>      struct tegra_ahbdma_chan *tchan;   /* see tegra_ahbdma_prep() */
>>>      ....
>>> };
>>>
>>> struct tegra_ahbdma_tx_desc *tx = to_ahbdma_tx_desc(vdesc);
>>> struct device *dev = chan2dev(&tx->tchan->vchan.chan);
>>> devm_kfree(dev, tx);
>>>
>>
>> Unfortunately I'm thinking that your proposal isn't correct:
>>
>> 1) virt-dma manages descriptor allocations for us here, all desc's are free'd by
>> vchan_free_chan_resources on channels release
>>
>> 2) we want to release all channels descriptors when *channel* is released, using
>> devm_* just doesn't make sense
>>
>>>
>>>> +}
>>>> +
>>>> +static struct dma_async_tx_descriptor *tegra_ahbdma_prep(
>>>> +                                     struct dma_chan *chan,
>>>> +                                     enum dma_transfer_direction dir,
>>>> +                                     unsigned long flags,
>>>> +                                     dma_addr_t paddr,
>>>> +                                     size_t size,
>>>> +                                     bool cyclic)
>>>> +{
>>>> +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +     struct tegra_ahbdma_tx_desc *tx;
>>>> +     u32 csr = ahbdma_chan->csr;
>>>> +
>>>> +     /* size and alignments should fulfill HW requirements */
>>>> +     if (size < 4 || size & 3 || paddr & 3)
>>>> +             return NULL;
>>>> +
>>>> +     tx = kzalloc(sizeof(*tx), GFP_NOWAIT);
>>>
>>>
>>>
>>> How about using devm_kzalloc() here?  You can get access to your "dev" with a
>>> function like the following:
>>>
>>> static inline struct device *chan2dev(struct dma_chan *chan)
>>> {
>>>     return &chan->dev->device;
>>> }
>>>
>>>
>>>> +     if (!tx)
>>>> +             return NULL;
>>>> +
>>>> +     if (dir == DMA_DEV_TO_MEM)
>>>> +             csr |= AHBDMA_CH_DIR_TO_XMB;
>>>> +
>>>> +     if (!cyclic)
>>>> +             csr |= AHBDMA_CH_ONCE;
>>>> +
>>>> +     tx->csr = csr | (size - sizeof(u32));
>>>> +     tx->ahb_seq = ahbdma_chan->ahb_seq;
>>>> +     tx->ahb_addr = ahbdma_chan->ahb_addr;
>>>> +     tx->mem_addr = paddr;
>>>
>>>
>>>
>>> Add setting of suggested new field:
>>>     tx->tchan = ahbdma_chan;
>>>
>>>
>>>> +
>>>> +     return vchan_tx_prep(&ahbdma_chan->vchan, &tx->vdesc, flags);
>>>> +}
>>>> +
>>>> +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)
>>>> +{
>>>> +     /* unimplemented */
>>>> +     if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
>>>> +             return NULL;
>>>> +
>>>> +     return tegra_ahbdma_prep(chan, dir, flags, sg_dma_address(sgl),
>>>> +                              sg_dma_len(sgl), false);
>>>> +}
>>>> +
>>>> +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)
>>>> +{
>>>> +     /* unimplemented */
>>>> +     if (buf_len != period_len || buf_len > SZ_64K)
>>>> +             return NULL;
>>>> +
>>>> +     return tegra_ahbdma_prep(chan, dir, flags, buf_addr, buf_len, true);
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
>>>> +{
>>>> +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +     struct virt_dma_chan *vchan = &ahbdma_chan->vchan;
>>>> +     unsigned long flags;
>>>> +
>>>> +     spin_lock_irqsave(&vchan->lock, flags);
>>>> +
>>>> +     if (vchan_issue_pending(vchan) && !ahbdma_chan->active_tx)
>>>> +             tegra_ahbdma_issue_next_tx(ahbdma_chan);
>>>> +
>>>> +     spin_unlock_irqrestore(&vchan->lock, flags);
>>>> +}
>>>> +
>>>> +static size_t tegra_ahbdma_residual(struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +     u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
>>>> +
>>>> +     return (status & AHBDMA_CH_WCOUNT_MASK);
>>>> +}
>>>> +
>>>> +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;
>>>> +     struct virt_dma_desc *vdesc;
>>>> +     enum dma_status cookie_status;
>>>> +     unsigned long flags;
>>>> +     size_t residual;
>>>> +
>>>> +     spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>>>> +
>>>> +     cookie_status = dma_cookie_status(chan, cookie, state);
>>>> +     if (cookie_status == DMA_COMPLETE)
>>>> +             goto unlock;
>>>> +
>>>> +     vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
>>>> +     if (!vdesc)
>>>> +             residual = 0;
>>>> +     else {
>>>> +             tx = to_ahbdma_tx_desc(vdesc);
>>>> +
>>>> +             if (tx == ahbdma_chan->active_tx)
>>>> +                     residual = tegra_ahbdma_residual(ahbdma_chan);
>>>> +             else
>>>> +                     residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>>>> +
>>>> +             residual += sizeof(u32);
>>>> +     }
>>>> +
>>>> +     dma_set_residue(state, residual);
>>>> +
>>>> +unlock:
>>>> +     spin_unlock_irqrestore(&ahbdma_chan->vchan.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);
>>>> +     unsigned long flags;
>>>> +     LIST_HEAD(head);
>>>> +     u32 csr;
>>>> +
>>>> +     spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
>>>> +
>>>> +     csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
>>>> +     writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
>>>> +                    ahbdma_chan->regs + AHBDMA_CH_CSR);
>>>> +
>>>> +     if (ahbdma_chan->active_tx) {
>>>> +             udelay(AHBDMA_BURST_COMPLETE_TIME);
>>>> +
>>>> +             writel_relaxed(AHBDMA_CH_IS_EOC,
>>>> +                            ahbdma_chan->regs + AHBDMA_CH_STA);
>>>> +
>>>> +             ahbdma_chan->active_tx = NULL;
>>>> +     }
>>>> +
>>>> +     vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
>>>> +     complete_all(&ahbdma_chan->idling);
>>>> +
>>>> +     spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
>>>> +
>>>> +     vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
>>>> +
>>>> +     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, csr;
>>>> +     unsigned int slave_id;
>>>> +     phys_addr_t ahb_addr;
>>>> +
>>>> +     if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
>>>> +         sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
>>>> +             return -EINVAL;
>>>> +
>>>> +     switch (dir) {
>>>> +     case DMA_DEV_TO_MEM:
>>>> +             burst    = sconfig->src_maxburst;
>>>> +             ahb_addr = sconfig->src_addr;
>>>> +             break;
>>>> +     case DMA_MEM_TO_DEV:
>>>> +             burst    = sconfig->dst_maxburst;
>>>> +             ahb_addr = sconfig->dst_addr;
>>>> +             break;
>>>> +     default:
>>>> +             return -EINVAL;
>>>> +     }
>>>> +
>>>> +     switch (burst) {
>>>> +     case 1:
>>>> +             burst = AHBDMA_CH_AHB_BURST_1;
>>>> +             break;
>>>> +     case 4:
>>>> +             burst = AHBDMA_CH_AHB_BURST_4;
>>>> +             break;
>>>> +     case 8:
>>>> +             burst = AHBDMA_CH_AHB_BURST_8;
>>>> +             break;
>>>> +     default:
>>>> +             return -EINVAL;
>>>> +     }
>>>> +
>>>> +     if (ahb_addr & 3)
>>>> +             return -EINVAL;
>>>> +
>>>> +     ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
>>>> +     ahb_seq |= AHBDMA_CH_INTR_ENB;
>>>> +
>>>> +     csr  = AHBDMA_CH_ENABLE;
>>>> +     csr |= AHBDMA_CH_IE_EOC;
>>>> +
>>>> +     if (ahbdma_chan->of_slave || sconfig->device_fc) {
>>>> +             if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
>>>> +                     slave_id = ahbdma_chan->of_req_sel;
>>>> +             else
>>>> +                     slave_id = sconfig->slave_id;
>>>> +
>>>> +             if (slave_id > 15)
>>>> +                     return -EINVAL;
>>>> +
>>>> +             ahb_seq |= AHBDMA_CH_ADDR_WRAP;
>>>> +
>>>> +             csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
>>>> +             csr |= AHBDMA_CH_FLOW;
>>>> +     }
>>>> +
>>>> +     ahbdma_chan->csr = csr;
>>>> +     ahbdma_chan->ahb_seq = ahb_seq;
>>>> +     ahbdma_chan->ahb_addr = ahb_addr;
>>>> +
>>>> +     return 0;
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_synchronize(struct dma_chan *chan)
>>>> +{
>>>> +     struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
>>>> +
>>>> +     wait_for_completion(&ahbdma_chan->idling);
>>>> +     vchan_synchronize(&ahbdma_chan->vchan);
>>>> +}
>>>> +
>>>> +static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
>>>> +{
>>>> +     vchan_free_chan_resources(to_virt_chan(chan));
>>>> +}
>>>> +
>>>> +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_device *dma_dev = &tdma->dma_dev;
>>>> +
>>>> +     vchan_init(&ahbdma_chan->vchan, dma_dev);
>>>> +     init_completion(&ahbdma_chan->idling);
>>>> +     complete(&ahbdma_chan->idling);
>>>> +
>>>> +     ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
>>>> +     ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
>>>> +     ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
>>>> +}
>>>> +
>>>> +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;
>>>> +
>>>> +     chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>>> +     if (!chan)
>>>> +             return NULL;
>>>> +
>>>> +     to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
>>>> +     to_ahbdma_chan(chan)->of_slave = true;
>>>> +
>>>> +     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(AHBDMA_CMD_ENABLE, tdma->regs + AHBDMA_CMD);
>>>> +
>>>> +     writel_relaxed(AHBDMA_IRQ_ENB_CH(0) |
>>>> +                    AHBDMA_IRQ_ENB_CH(1) |
>>>> +                    AHBDMA_IRQ_ENB_CH(2) |
>>>> +                    AHBDMA_IRQ_ENB_CH(3),
>>>> +                    tdma->regs + 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             = AHBDMA_DIRECTIONS;
>>>> +     dma_dev->src_addr_widths        = AHBDMA_BUS_WIDTH;
>>>> +     dma_dev->dst_addr_widths        = 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 <[e-mail]
>>>    <mailto:[e-mail]>>");
>>>> +MODULE_LICENSE("GPL");
>>>>

--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Jon Hunter Oct. 9, 2017, 9:43 a.m. UTC | #11

On 06/10/17 20:11, Dmitry Osipenko wrote:
> On 04.10.2017 02:58, 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, mem-to-mem isn't implemented as well.
>>
>> Signed-off-by: Dmitry Osipenko <[e-mail]>
>> ---
>>  drivers/dma/Kconfig           |  10 +
>>  drivers/dma/Makefile          |   1 +
>>  drivers/dma/tegra20-ahb-dma.c | 630 ++++++++++++++++++++++++++++++++++++++++++
>>  3 files changed, 641 insertions(+)
>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c


...

>> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
>> +						struct tegra_ahbdma_chan *chan)
>> +{
>> +	struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
>> +
>> +	if (vdesc)
>> +		list_del(&vdesc->node);
> 
> I just noticed that this is incorrect. Node must be deleted after TX completion,
> otherwise vchan_find_desc won't find TX and residual won't be reported by
> dmaengine_tx_status.
> 
> Jon, I think you ADMA driver has the same issue, as well as several other DMA
> drivers that use virt-dma.

Actually, I think that the above is correct. If vchan_find_desc() finds
the descriptor then this indicates that the transfer has not started yet
and so the residual is equal to the transfer size.

If you look at the adma driver, you will see the if vchan_find_desc()
does not find the descriptor, then we check to see if the current
transfer is the one we are querying and if so return the bytes
remaining. Looking at this driver again, what you have in
tegra_ahbdma_tx_status() does not look correct. You should have
something like ...

	vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
	if (vdesc) {
		tx = to_ahbdma_tx_desc(vdesc);
		residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
	} else if (ahbdma_chan->tx_active && 	
		   ahbdma_chan->tx_active->vd.tx.cookie == cookie) {
		residual = tegra_ahbdma_residual(ahbdma_chan);
	} else {
		residual = 0;
	}

Cheers
Jon

Jon Hunter Oct. 9, 2017, 9:51 a.m. UTC | #12

On 06/10/17 18:23, Dmitry Osipenko wrote:
> On 06.10.2017 18:50, Jon Hunter wrote:
>> On 06/10/17 16:26, Dmitry Osipenko wrote:
>>> On 06.10.2017 16:11, Jon Hunter wrote:
>>>> On 04/10/17 00:58, Dmitry Osipenko wrote:

...

>>>>> +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;
>>>>> +
>>>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>>>> +	if (!chan)
>>>>> +		return NULL;
>>>>> +
>>>>> +	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
>>>>
>>>> Test for args[0] < TEGRA_AHBDMA_REQ_N_A?
>>>>
>>>
>>> It would duplicate slave_id checking done in tegra_ahbdma_config(), so not
>>> needed here.
>>
>> But surely we should not let them request a channel in the first place?
>>
> 
> If allowing client to disable flow control is okay, as you mentioned below, then
> I agree that it is fine. I'll make this change.
> 
>>>>> +	to_ahbdma_chan(chan)->of_slave = true;
>>>>
>>>> Is this really needed? Doesn't a value of 0..TEGRA_AHBDMA_REQ_N_A-1 tell
>>>> us it is valid?
>>>>
>>>
>>> I think we should enforce channels flow control in a case of OF xlate'd channel,
>>> no? To avoid abusing channels usage by client. Seems tegra_ahbdma_config isn't
>>> correct, should be:
>>
>> Absolutely. However, I don't see the need for the additional 'of_slave'
>> variable. If we validate the slave id here, we can get rid of the extra
>> variable. It does not simplify the code really by adding this IMO.
>>
> 
> 'of_slave' enforces flow control enable. If I understand you correctly, you are
> suggesting that it is okay to leave ability for clients to override flow
> control. Well, that's probably is fine indeed, just keep an eye on client drivers.
Nope :-)

I am simply saying that we do not need this 'of_slave' variable in
addition to the 'of_req_sel'. If we verify that 'args[0] <
TEGRA_AHBDMA_REQ_N_A' in this xlate function and set 'of_req_sel =
args[0]', then in tegra_ahbdma_config() we just have ...

	if (ahbdma_chan->of_req_sel || sconfig->device_fc) {
		if (ahbdma_chan->of_req_sel)
			slave_id = ahbdma_chan->of_req_sel;
		else if (sconfig->slave_id < TEGRA_AHBDMA_REQ_N_A)
			slave_id = sconfig->slave_id;
		else
			return -EINVAL;

		ahb_seq |= AHBDMA_CH_ADDR_WRAP;

		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
		csr |= AHBDMA_CH_FLOW;
	}

Cheers
Jon

Vinod Koul Oct. 9, 2017, 10:39 a.m. UTC | #13

On Mon, Oct 09, 2017 at 10:43:54AM +0100, Jon Hunter wrote:
> 
> 
> On 06/10/17 20:11, Dmitry Osipenko wrote:
> > On 04.10.2017 02:58, 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, mem-to-mem isn't implemented as well.
> >>
> >> Signed-off-by: Dmitry Osipenko <[e-mail]>
> >> ---
> >>  drivers/dma/Kconfig           |  10 +
> >>  drivers/dma/Makefile          |   1 +
> >>  drivers/dma/tegra20-ahb-dma.c | 630 ++++++++++++++++++++++++++++++++++++++++++
> >>  3 files changed, 641 insertions(+)
> >>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
> 
> 
> ...
> 
> >> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
> >> +						struct tegra_ahbdma_chan *chan)
> >> +{
> >> +	struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
> >> +
> >> +	if (vdesc)
> >> +		list_del(&vdesc->node);
> > 
> > I just noticed that this is incorrect. Node must be deleted after TX completion,
> > otherwise vchan_find_desc won't find TX and residual won't be reported by
> > dmaengine_tx_status.
> > 
> > Jon, I think you ADMA driver has the same issue, as well as several other DMA
> > drivers that use virt-dma.
> 
> Actually, I think that the above is correct. If vchan_find_desc() finds
> the descriptor then this indicates that the transfer has not started yet
> and so the residual is equal to the transfer size.

That is correct, so you can do a quick calculation and find the residue only
for current one

> If you look at the adma driver, you will see the if vchan_find_desc()
> does not find the descriptor, then we check to see if the current
> transfer is the one we are querying and if so return the bytes
> remaining. Looking at this driver again, what you have in
> tegra_ahbdma_tx_status() does not look correct. You should have
> something like ...
> 
> 	vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
> 	if (vdesc) {
> 		tx = to_ahbdma_tx_desc(vdesc);
> 		residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
> 	} else if (ahbdma_chan->tx_active && 	
> 		   ahbdma_chan->tx_active->vd.tx.cookie == cookie) {
> 		residual = tegra_ahbdma_residual(ahbdma_chan);
> 	} else {
> 		residual = 0;
> 	}
> 
> Cheers
> Jon
> 
> -- 
> nvpublic

Dmitry Osipenko Oct. 9, 2017, 11:43 a.m. UTC | #14

On 09.10.2017 13:39, Vinod Koul wrote:
> On Mon, Oct 09, 2017 at 10:43:54AM +0100, Jon Hunter wrote:
>>
>>
>> On 06/10/17 20:11, Dmitry Osipenko wrote:
>>> On 04.10.2017 02:58, 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, mem-to-mem isn't implemented as well.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <[e-mail]>
>>>> ---
>>>>  drivers/dma/Kconfig           |  10 +
>>>>  drivers/dma/Makefile          |   1 +
>>>>  drivers/dma/tegra20-ahb-dma.c | 630 ++++++++++++++++++++++++++++++++++++++++++
>>>>  3 files changed, 641 insertions(+)
>>>>  create mode 100644 drivers/dma/tegra20-ahb-dma.c
>>
>>
>> ...
>>
>>>> +static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
>>>> +						struct tegra_ahbdma_chan *chan)
>>>> +{
>>>> +	struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
>>>> +
>>>> +	if (vdesc)
>>>> +		list_del(&vdesc->node);
>>>
>>> I just noticed that this is incorrect. Node must be deleted after TX completion,
>>> otherwise vchan_find_desc won't find TX and residual won't be reported by
>>> dmaengine_tx_status.
>>>
>>> Jon, I think you ADMA driver has the same issue, as well as several other DMA
>>> drivers that use virt-dma.
>>
>> Actually, I think that the above is correct. If vchan_find_desc() finds
>> the descriptor then this indicates that the transfer has not started yet
>> and so the residual is equal to the transfer size.
> 
> That is correct, so you can do a quick calculation and find the residue only
> for current one
> 

Yeah, the problem was that the current *active* one wasn't found.

>> If you look at the adma driver, you will see the if vchan_find_desc()
>> does not find the descriptor, then we check to see if the current
>> transfer is the one we are querying and if so return the bytes
>> remaining. Looking at this driver again, what you have in
>> tegra_ahbdma_tx_status() does not look correct. You should have
>> something like ...
>>
>> 	vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
>> 	if (vdesc) {
>> 		tx = to_ahbdma_tx_desc(vdesc);
>> 		residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
>> 	} else if (ahbdma_chan->tx_active && 	
>> 		   ahbdma_chan->tx_active->vd.tx.cookie == cookie) {
>> 		residual = tegra_ahbdma_residual(ahbdma_chan);
>> 	} else {
>> 		residual = 0;
>> 	}

I've moved list_del() to the TX completion handler, but I agree that it is not
entirely correct even though that solution works fine. Your variant also works
and does the right thing, thanks!

ADMA driver is correct. Sorry for the 'false alarm' :)
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Dmitry Osipenko Oct. 9, 2017, 1:28 p.m. UTC | #15

On 09.10.2017 12:51, Jon Hunter wrote:
> 
> 
> On 06/10/17 18:23, Dmitry Osipenko wrote:
>> On 06.10.2017 18:50, Jon Hunter wrote:
>>> On 06/10/17 16:26, Dmitry Osipenko wrote:
>>>> On 06.10.2017 16:11, Jon Hunter wrote:
>>>>> On 04/10/17 00:58, Dmitry Osipenko wrote:
> 
> ...
> 
>>>>>> +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;
>>>>>> +
>>>>>> +	chan = dma_get_any_slave_channel(&tdma->dma_dev);
>>>>>> +	if (!chan)
>>>>>> +		return NULL;
>>>>>> +
>>>>>> +	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
>>>>>
>>>>> Test for args[0] < TEGRA_AHBDMA_REQ_N_A?
>>>>>
>>>>
>>>> It would duplicate slave_id checking done in tegra_ahbdma_config(), so not
>>>> needed here.
>>>
>>> But surely we should not let them request a channel in the first place?
>>>
>>
>> If allowing client to disable flow control is okay, as you mentioned below, then
>> I agree that it is fine. I'll make this change.
>>
>>>>>> +	to_ahbdma_chan(chan)->of_slave = true;
>>>>>
>>>>> Is this really needed? Doesn't a value of 0..TEGRA_AHBDMA_REQ_N_A-1 tell
>>>>> us it is valid?
>>>>>
>>>>
>>>> I think we should enforce channels flow control in a case of OF xlate'd channel,
>>>> no? To avoid abusing channels usage by client. Seems tegra_ahbdma_config isn't
>>>> correct, should be:
>>>
>>> Absolutely. However, I don't see the need for the additional 'of_slave'
>>> variable. If we validate the slave id here, we can get rid of the extra
>>> variable. It does not simplify the code really by adding this IMO.
>>>
>>
>> 'of_slave' enforces flow control enable. If I understand you correctly, you are
>> suggesting that it is okay to leave ability for clients to override flow
>> control. Well, that's probably is fine indeed, just keep an eye on client drivers.
> Nope :-)
> 
> I am simply saying that we do not need this 'of_slave' variable in
> addition to the 'of_req_sel'. If we verify that 'args[0] <
> TEGRA_AHBDMA_REQ_N_A' in this xlate function and set 'of_req_sel =
> args[0]', then in tegra_ahbdma_config() we just have ...
> 
> 	if (ahbdma_chan->of_req_sel || sconfig->device_fc) {
> 		if (ahbdma_chan->of_req_sel)
> 			slave_id = ahbdma_chan->of_req_sel;
> 		else if (sconfig->slave_id < TEGRA_AHBDMA_REQ_N_A)
> 			slave_id = sconfig->slave_id;
> 		else
> 			return -EINVAL;
> 
> 		ahb_seq |= AHBDMA_CH_ADDR_WRAP;
> 
> 		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
> 		csr |= AHBDMA_CH_FLOW;
> 	}
> 

Okay
--
To unsubscribe from this list: send the line "unsubscribe dmaengine" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index 04e381b522b4..7d132aa85174 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -512,6 +512,16 @@  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 || COMPILE_TEST
+	select DMA_ENGINE
+	select DMA_VIRTUAL_CHANNELS
+	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 a145ad1426bc..f3d284bf6d65 100644
--- a/drivers/dma/Makefile
+++ b/drivers/dma/Makefile
@@ -62,6 +62,7 @@  obj-$(CONFIG_STM32_DMA) += stm32-dma.o
 obj-$(CONFIG_STM32_DMAMUX) += stm32-dmamux.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..2d176a5536aa
--- /dev/null
+++ b/drivers/dma/tegra20-ahb-dma.c
@@ -0,0 +1,630 @@ 
+/*
+ * Copyright 2017 Dmitry Osipenko <[e-mail]>
+ *
+ * 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/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 <dt-bindings/dma/tegra-ahb-dma.h>
+
+#include "virt-dma.h"
+
+#define AHBDMA_CMD			0x0
+#define AHBDMA_CMD_ENABLE		BIT(31)
+
+#define AHBDMA_IRQ_ENB_MASK		0x20
+#define AHBDMA_IRQ_ENB_CH(ch)		BIT(ch)
+
+#define AHBDMA_CH_BASE(ch)		(0x1000 + (ch) * 0x20)
+
+#define AHBDMA_CH_CSR			0x0
+#define AHBDMA_CH_ADDR_WRAP		BIT(18)
+#define AHBDMA_CH_FLOW			BIT(24)
+#define AHBDMA_CH_ONCE			BIT(26)
+#define AHBDMA_CH_DIR_TO_XMB		BIT(27)
+#define AHBDMA_CH_IE_EOC		BIT(30)
+#define AHBDMA_CH_ENABLE		BIT(31)
+#define AHBDMA_CH_REQ_SEL_SHIFT		16
+#define AHBDMA_CH_WCOUNT_MASK		GENMASK(15, 2)
+
+#define AHBDMA_CH_STA			0x4
+#define AHBDMA_CH_IS_EOC		BIT(30)
+
+#define AHBDMA_CH_AHB_PTR		0x10
+
+#define AHBDMA_CH_AHB_SEQ		0x14
+#define AHBDMA_CH_INTR_ENB		BIT(31)
+#define AHBDMA_CH_AHB_BURST_SHIFT	24
+#define AHBDMA_CH_AHB_BURST_1		2
+#define AHBDMA_CH_AHB_BURST_4		3
+#define AHBDMA_CH_AHB_BURST_8		4
+
+#define AHBDMA_CH_XMB_PTR		0x18
+
+#define AHBDMA_BUS_WIDTH		BIT(DMA_SLAVE_BUSWIDTH_4_BYTES)
+
+#define AHBDMA_DIRECTIONS		BIT(DMA_DEV_TO_MEM) | \
+					BIT(DMA_MEM_TO_DEV)
+
+#define AHBDMA_BURST_COMPLETE_TIME	20
+
+struct tegra_ahbdma_tx_desc {
+	struct virt_dma_desc vdesc;
+	dma_addr_t mem_addr;
+	phys_addr_t ahb_addr;
+	u32 ahb_seq;
+	u32 csr;
+};
+
+struct tegra_ahbdma_chan {
+	struct tegra_ahbdma_tx_desc *active_tx;
+	struct virt_dma_chan vchan;
+	struct completion idling;
+	void __iomem *regs;
+	phys_addr_t ahb_addr;
+	u32 ahb_seq;
+	u32 csr;
+	unsigned int of_req_sel;
+	bool of_slave;
+};
+
+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_chan *to_ahbdma_chan(struct dma_chan *chan)
+{
+	return container_of(chan, struct tegra_ahbdma_chan, vchan.chan);
+}
+
+static inline struct tegra_ahbdma_tx_desc *to_ahbdma_tx_desc(
+						struct virt_dma_desc *vdesc)
+{
+	return container_of(vdesc, struct tegra_ahbdma_tx_desc, vdesc);
+}
+
+static struct tegra_ahbdma_tx_desc *tegra_ahbdma_get_next_tx(
+						struct tegra_ahbdma_chan *chan)
+{
+	struct virt_dma_desc *vdesc = vchan_next_desc(&chan->vchan);
+
+	if (vdesc)
+		list_del(&vdesc->node);
+
+	return vdesc ? to_ahbdma_tx_desc(vdesc) : NULL;
+}
+
+static void tegra_ahbdma_issue_next_tx(struct tegra_ahbdma_chan *chan)
+{
+	struct tegra_ahbdma_tx_desc *tx = tegra_ahbdma_get_next_tx(chan);
+
+	if (tx) {
+		writel_relaxed(tx->ahb_seq,  chan->regs + AHBDMA_CH_AHB_SEQ);
+		writel_relaxed(tx->ahb_addr, chan->regs + AHBDMA_CH_AHB_PTR);
+		writel_relaxed(tx->mem_addr, chan->regs + AHBDMA_CH_XMB_PTR);
+		writel_relaxed(tx->csr,      chan->regs + AHBDMA_CH_CSR);
+
+		reinit_completion(&chan->idling);
+	} else
+		complete_all(&chan->idling);
+
+	chan->active_tx = tx;
+}
+
+static bool tegra_ahbdma_clear_interrupt(struct tegra_ahbdma_chan *chan)
+{
+	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
+
+	if (status & AHBDMA_CH_IS_EOC) {
+		writel_relaxed(AHBDMA_CH_IS_EOC, chan->regs + AHBDMA_CH_STA);
+
+		return true;
+	}
+
+	return false;
+}
+
+static bool tegra_ahbdma_handle_channel(struct tegra_ahbdma_chan *chan)
+{
+	struct tegra_ahbdma_tx_desc *tx;
+	unsigned long flags;
+	bool intr = false;
+	bool cyclic;
+
+	spin_lock_irqsave(&chan->vchan.lock, flags);
+
+	tx = chan->active_tx;
+	if (tx)
+		intr = tegra_ahbdma_clear_interrupt(chan);
+
+	if (intr) {
+		cyclic = !(tx->csr & AHBDMA_CH_ONCE);
+
+		if (!cyclic)
+			tegra_ahbdma_issue_next_tx(chan);
+
+		if (cyclic)
+			vchan_cyclic_callback(&tx->vdesc);
+		else
+			vchan_cookie_complete(&tx->vdesc);
+	}
+
+	spin_unlock_irqrestore(&chan->vchan.lock, flags);
+
+	return intr;
+}
+
+static irqreturn_t tegra_ahbdma_isr(int irq, void *dev_id)
+{
+	struct tegra_ahbdma *tdma = dev_id;
+	bool handled;
+
+	handled  = tegra_ahbdma_handle_channel(&tdma->channels[0]);
+	handled |= tegra_ahbdma_handle_channel(&tdma->channels[1]);
+	handled |= tegra_ahbdma_handle_channel(&tdma->channels[2]);
+	handled |= tegra_ahbdma_handle_channel(&tdma->channels[3]);
+
+	return handled ? IRQ_HANDLED : IRQ_NONE;
+}
+
+static void tegra_ahbdma_tx_desc_free(struct virt_dma_desc *vdesc)
+{
+	kfree(to_ahbdma_tx_desc(vdesc));
+}
+
+static struct dma_async_tx_descriptor *tegra_ahbdma_prep(
+					struct dma_chan *chan,
+					enum dma_transfer_direction dir,
+					unsigned long flags,
+					dma_addr_t paddr,
+					size_t size,
+					bool cyclic)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+	struct tegra_ahbdma_tx_desc *tx;
+	u32 csr = ahbdma_chan->csr;
+
+	/* size and alignments should fulfill HW requirements */
+	if (size < 4 || size & 3 || paddr & 3)
+		return NULL;
+
+	tx = kzalloc(sizeof(*tx), GFP_NOWAIT);
+	if (!tx)
+		return NULL;
+
+	if (dir == DMA_DEV_TO_MEM)
+		csr |= AHBDMA_CH_DIR_TO_XMB;
+
+	if (!cyclic)
+		csr |= AHBDMA_CH_ONCE;
+
+	tx->csr = csr | (size - sizeof(u32));
+	tx->ahb_seq = ahbdma_chan->ahb_seq;
+	tx->ahb_addr = ahbdma_chan->ahb_addr;
+	tx->mem_addr = paddr;
+
+	return vchan_tx_prep(&ahbdma_chan->vchan, &tx->vdesc, flags);
+}
+
+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)
+{
+	/* unimplemented */
+	if (sg_len != 1 || sg_dma_len(sgl) > SZ_64K)
+		return NULL;
+
+	return tegra_ahbdma_prep(chan, dir, flags, sg_dma_address(sgl),
+				 sg_dma_len(sgl), false);
+}
+
+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)
+{
+	/* unimplemented */
+	if (buf_len != period_len || buf_len > SZ_64K)
+		return NULL;
+
+	return tegra_ahbdma_prep(chan, dir, flags, buf_addr, buf_len, true);
+}
+
+static void tegra_ahbdma_issue_pending(struct dma_chan *chan)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+	struct virt_dma_chan *vchan = &ahbdma_chan->vchan;
+	unsigned long flags;
+
+	spin_lock_irqsave(&vchan->lock, flags);
+
+	if (vchan_issue_pending(vchan) && !ahbdma_chan->active_tx)
+		tegra_ahbdma_issue_next_tx(ahbdma_chan);
+
+	spin_unlock_irqrestore(&vchan->lock, flags);
+}
+
+static size_t tegra_ahbdma_residual(struct tegra_ahbdma_chan *chan)
+{
+	u32 status = readl_relaxed(chan->regs + AHBDMA_CH_STA);
+
+	return (status & AHBDMA_CH_WCOUNT_MASK);
+}
+
+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;
+	struct virt_dma_desc *vdesc;
+	enum dma_status cookie_status;
+	unsigned long flags;
+	size_t residual;
+
+	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
+
+	cookie_status = dma_cookie_status(chan, cookie, state);
+	if (cookie_status == DMA_COMPLETE)
+		goto unlock;
+
+	vdesc = vchan_find_desc(&ahbdma_chan->vchan, cookie);
+	if (!vdesc)
+		residual = 0;
+	else {
+		tx = to_ahbdma_tx_desc(vdesc);
+
+		if (tx == ahbdma_chan->active_tx)
+			residual = tegra_ahbdma_residual(ahbdma_chan);
+		else
+			residual = tx->csr & AHBDMA_CH_WCOUNT_MASK;
+
+		residual += sizeof(u32);
+	}
+
+	dma_set_residue(state, residual);
+
+unlock:
+	spin_unlock_irqrestore(&ahbdma_chan->vchan.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);
+	unsigned long flags;
+	LIST_HEAD(head);
+	u32 csr;
+
+	spin_lock_irqsave(&ahbdma_chan->vchan.lock, flags);
+
+	csr = readl_relaxed(ahbdma_chan->regs + AHBDMA_CH_CSR);
+	writel_relaxed(csr & ~AHBDMA_CH_ENABLE,
+		       ahbdma_chan->regs + AHBDMA_CH_CSR);
+
+	if (ahbdma_chan->active_tx) {
+		udelay(AHBDMA_BURST_COMPLETE_TIME);
+
+		writel_relaxed(AHBDMA_CH_IS_EOC,
+			       ahbdma_chan->regs + AHBDMA_CH_STA);
+
+		ahbdma_chan->active_tx = NULL;
+	}
+
+	vchan_get_all_descriptors(&ahbdma_chan->vchan, &head);
+	complete_all(&ahbdma_chan->idling);
+
+	spin_unlock_irqrestore(&ahbdma_chan->vchan.lock, flags);
+
+	vchan_dma_desc_free_list(&ahbdma_chan->vchan, &head);
+
+	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, csr;
+	unsigned int slave_id;
+	phys_addr_t ahb_addr;
+
+	if (sconfig->src_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES ||
+	    sconfig->dst_addr_width != DMA_SLAVE_BUSWIDTH_4_BYTES)
+		return -EINVAL;
+
+	switch (dir) {
+	case DMA_DEV_TO_MEM:
+		burst    = sconfig->src_maxburst;
+		ahb_addr = sconfig->src_addr;
+		break;
+	case DMA_MEM_TO_DEV:
+		burst    = sconfig->dst_maxburst;
+		ahb_addr = sconfig->dst_addr;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	switch (burst) {
+	case 1:
+		burst = AHBDMA_CH_AHB_BURST_1;
+		break;
+	case 4:
+		burst = AHBDMA_CH_AHB_BURST_4;
+		break;
+	case 8:
+		burst = AHBDMA_CH_AHB_BURST_8;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	if (ahb_addr & 3)
+		return -EINVAL;
+
+	ahb_seq  = burst << AHBDMA_CH_AHB_BURST_SHIFT;
+	ahb_seq |= AHBDMA_CH_INTR_ENB;
+
+	csr  = AHBDMA_CH_ENABLE;
+	csr |= AHBDMA_CH_IE_EOC;
+
+	if (ahbdma_chan->of_slave || sconfig->device_fc) {
+		if (ahbdma_chan->of_req_sel < TEGRA_AHBDMA_REQ_N_A)
+			slave_id = ahbdma_chan->of_req_sel;
+		else
+			slave_id = sconfig->slave_id;
+
+		if (slave_id > 15)
+			return -EINVAL;
+
+		ahb_seq |= AHBDMA_CH_ADDR_WRAP;
+
+		csr |= slave_id << AHBDMA_CH_REQ_SEL_SHIFT;
+		csr |= AHBDMA_CH_FLOW;
+	}
+
+	ahbdma_chan->csr = csr;
+	ahbdma_chan->ahb_seq = ahb_seq;
+	ahbdma_chan->ahb_addr = ahb_addr;
+
+	return 0;
+}
+
+static void tegra_ahbdma_synchronize(struct dma_chan *chan)
+{
+	struct tegra_ahbdma_chan *ahbdma_chan = to_ahbdma_chan(chan);
+
+	wait_for_completion(&ahbdma_chan->idling);
+	vchan_synchronize(&ahbdma_chan->vchan);
+}
+
+static void tegra_ahbdma_free_chan_resources(struct dma_chan *chan)
+{
+	vchan_free_chan_resources(to_virt_chan(chan));
+}
+
+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_device *dma_dev = &tdma->dma_dev;
+
+	vchan_init(&ahbdma_chan->vchan, dma_dev);
+	init_completion(&ahbdma_chan->idling);
+	complete(&ahbdma_chan->idling);
+
+	ahbdma_chan->regs = tdma->regs + AHBDMA_CH_BASE(chan_id);
+	ahbdma_chan->vchan.desc_free = tegra_ahbdma_tx_desc_free;
+	ahbdma_chan->of_req_sel = TEGRA_AHBDMA_REQ_N_A;
+}
+
+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;
+
+	chan = dma_get_any_slave_channel(&tdma->dma_dev);
+	if (!chan)
+		return NULL;
+
+	to_ahbdma_chan(chan)->of_req_sel = dma_spec->args[0];
+	to_ahbdma_chan(chan)->of_slave = true;
+
+	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(AHBDMA_CMD_ENABLE, tdma->regs + AHBDMA_CMD);
+
+	writel_relaxed(AHBDMA_IRQ_ENB_CH(0) |
+		       AHBDMA_IRQ_ENB_CH(1) |
+		       AHBDMA_IRQ_ENB_CH(2) |
+		       AHBDMA_IRQ_ENB_CH(3),
+		       tdma->regs + 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		= AHBDMA_DIRECTIONS;
+	dma_dev->src_addr_widths	= AHBDMA_BUS_WIDTH;
+	dma_dev->dst_addr_widths	= 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 <[e-mail]>");
+MODULE_LICENSE("GPL");