[PATCH v6 00/16] NVIDIA Tegra APB DMA driver fixes and improvements

[PATCH v6 00/16] NVIDIA Tegra APB DMA driver fixes and improvements

[Dmitry Osipenko]

Hello,

This series fixes some problems that I spotted recently, secondly the
driver's code gets a cleanup. Please review and apply, thanks in advance!

Changelog:

v6: - Added stable tag and changed order of the patch "Prevent race
      conditions of tasklet vs free list", making it patch #2, as was
      requested by Jon Hunter in the review comment to v5.

    - Factored out the tdc->config_init cleanup into separate patch, as was
      requested by Jon Hunter in the review comment to v5:

        dmaengine: tegra-apb: Remove unneeded initialization of tdc->config_init

    - Added new very minor patch to enable compile-testing for the driver:

        dmaengine: tegra-apb: Support COMPILE_TEST

v5: - Fixed touching hardware registers after RPM-suspending in the patch
      "Keep clock enabled only during of DMA transfer", now RPM is kept
      resumed in the tegra_dma_terminate_all() while needed. Thanks to
      Jon Hunter for pointing at this problem in a review comment to v4.

    - The "Clean up runtime PM teardown" patch is replaced with the "Remove
      assumptions about unavailable runtime PM" patch because I recalled that
      now RPM is always available on all Tegra SoCs.

    - The "Clean up suspend-resume" patch got a minor improvement, now
      tasklet_kill() is invoked before checking of the busy state in
      tegra_dma_dev_suspend(), this should allow us to catch problems if DMA
      callback issues a new DMA transfer.

    - Added Jon's acks to the reviewed patches.

v4: - Addressed Jon's request to *not* remove the runtime PM usage, instead
      there is now new patch that makes RPM more practical:

        dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

    - Added new minor patch to clean up RPM's teardown:

        dmaengine: tegra-apb: Clean up runtime PM teardown

v3: - In the review comment to v1 Michał Mirosław suggested that "Prevent
      race conditions on channel's freeing" does changes that deserve to
      be separated into two patches. I factored out and improved tasklet
      releasing into this new patch:

        dmaengine: tegra-apb: Clean up tasklet releasing

    - The "Fix use-after-free" patch got an improved commit message.

v2: - I took another look at the driver and spotted few more things that
      could be improved, which resulted in these new patches:

        dmaengine: tegra-apb: Remove runtime PM usage
        dmaengine: tegra-apb: Clean up suspend-resume
        dmaengine: tegra-apb: Add missing of_dma_controller_free
        dmaengine: tegra-apb: Allow to compile as a loadable kernel module
        dmaengine: tegra-apb: Remove MODULE_ALIAS


Dmitry Osipenko (16):
  dmaengine: tegra-apb: Fix use-after-free
  dmaengine: tegra-apb: Prevent race conditions of tasklet vs free list
  dmaengine: tegra-apb: Implement synchronization hook
  dmaengine: tegra-apb: Prevent race conditions on channel's freeing
  dmaengine: tegra-apb: Clean up tasklet releasing
  dmaengine: tegra-apb: Use devm_platform_ioremap_resource
  dmaengine: tegra-apb: Use devm_request_irq
  dmaengine: tegra-apb: Fix coding style problems
  dmaengine: tegra-apb: Remove unneeded initialization of
    tdc->config_init
  dmaengine: tegra-apb: Remove assumptions about unavailable runtime PM
  dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer
  dmaengine: tegra-apb: Clean up suspend-resume
  dmaengine: tegra-apb: Add missing of_dma_controller_free
  dmaengine: tegra-apb: Allow to compile as a loadable kernel module
  dmaengine: tegra-apb: Remove MODULE_ALIAS
  dmaengine: tegra-apb: Support COMPILE_TEST

 drivers/dma/Kconfig           |   4 +-
 drivers/dma/tegra20-apb-dma.c | 507 +++++++++++++++++-----------------
 2 files changed, 261 insertions(+), 250 deletions(-)

-- 
2.24.0

[PATCH v6 01/16] dmaengine: tegra-apb: Fix use-after-free

[Dmitry Osipenko]

I was doing some experiments with I2C and noticed that Tegra APB DMA
driver crashes sometime after I2C DMA transfer termination. The crash
happens because tegra_dma_terminate_all() bails out immediately if pending
list is empty, and thus, it doesn't release the half-completed descriptors
which are getting re-used before ISR tasklet kicks-in.

 tegra-i2c 7000c400.i2c: DMA transfer timeout
 elants_i2c 0-0010: elants_i2c_irq: failed to read data: -110
 ------------[ cut here ]------------
 WARNING: CPU: 0 PID: 142 at lib/list_debug.c:45 __list_del_entry_valid+0x45/0xac
 list_del corruption, ddbaac44->next is LIST_POISON1 (00000100)
 Modules linked in:
 CPU: 0 PID: 142 Comm: kworker/0:2 Not tainted 5.5.0-rc2-next-20191220-00175-gc3605715758d-dirty #538
 Hardware name: NVIDIA Tegra SoC (Flattened Device Tree)
 Workqueue: events_freezable_power_ thermal_zone_device_check
 [<c010e5c5>] (unwind_backtrace) from [<c010a1c5>] (show_stack+0x11/0x14)
 [<c010a1c5>] (show_stack) from [<c0973925>] (dump_stack+0x85/0x94)
 [<c0973925>] (dump_stack) from [<c011f529>] (__warn+0xc1/0xc4)
 [<c011f529>] (__warn) from [<c011f7e9>] (warn_slowpath_fmt+0x61/0x78)
 [<c011f7e9>] (warn_slowpath_fmt) from [<c042497d>] (__list_del_entry_valid+0x45/0xac)
 [<c042497d>] (__list_del_entry_valid) from [<c047a87f>] (tegra_dma_tasklet+0x5b/0x154)
 [<c047a87f>] (tegra_dma_tasklet) from [<c0124799>] (tasklet_action_common.constprop.0+0x41/0x7c)
 [<c0124799>] (tasklet_action_common.constprop.0) from [<c01022ab>] (__do_softirq+0xd3/0x2a8)
 [<c01022ab>] (__do_softirq) from [<c0124683>] (irq_exit+0x7b/0x98)
 [<c0124683>] (irq_exit) from [<c0168c19>] (__handle_domain_irq+0x45/0x80)
 [<c0168c19>] (__handle_domain_irq) from [<c043e429>] (gic_handle_irq+0x45/0x7c)
 [<c043e429>] (gic_handle_irq) from [<c0101aa5>] (__irq_svc+0x65/0x94)
 Exception stack(0xde2ebb90 to 0xde2ebbd8)

Cc: <stable@vger.kernel.org>
Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 4 ----
 1 file changed, 4 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 3a45079d11ec..319f31d27014 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -756,10 +756,6 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
 	bool was_busy;
 
 	spin_lock_irqsave(&tdc->lock, flags);
-	if (list_empty(&tdc->pending_sg_req)) {
-		spin_unlock_irqrestore(&tdc->lock, flags);
-		return 0;
-	}
 
 	if (!tdc->busy)
 		goto skip_dma_stop;
-- 
2.24.0

[PATCH v6 02/16] dmaengine: tegra-apb: Prevent race conditions of tasklet vs free list

[Dmitry Osipenko]

The interrupt handler puts a half-completed DMA descriptor on a free list
and then schedules tasklet to process bottom half of the descriptor that
executes client's callback, this creates possibility to pick up the busy
descriptor from the free list. Thus, let's disallow descriptor's re-use
until it is fully processed.

Cc: <stable@vger.kernel.org>
Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 319f31d27014..4a750e29bfb5 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -281,7 +281,7 @@ static struct tegra_dma_desc *tegra_dma_desc_get(
 
 	/* Do not allocate if desc are waiting for ack */
 	list_for_each_entry(dma_desc, &tdc->free_dma_desc, node) {
-		if (async_tx_test_ack(&dma_desc->txd)) {
+		if (async_tx_test_ack(&dma_desc->txd) && !dma_desc->cb_count) {
 			list_del(&dma_desc->node);
 			spin_unlock_irqrestore(&tdc->lock, flags);
 			dma_desc->txd.flags = 0;
-- 
2.24.0

[PATCH v6 03/16] dmaengine: tegra-apb: Implement synchronization hook

[Dmitry Osipenko]

The ISR tasklet could be kept scheduled after DMA transfer termination,
let's add synchronization hook which blocks until tasklet is finished.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 4a750e29bfb5..f56881500a23 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -798,6 +798,13 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
 	return 0;
 }
 
+static void tegra_dma_synchronize(struct dma_chan *dc)
+{
+	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
+
+	tasklet_kill(&tdc->tasklet);
+}
+
 static unsigned int tegra_dma_sg_bytes_xferred(struct tegra_dma_channel *tdc,
 					       struct tegra_dma_sg_req *sg_req)
 {
@@ -1506,6 +1513,7 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	tdma->dma_dev.residue_granularity = DMA_RESIDUE_GRANULARITY_BURST;
 	tdma->dma_dev.device_config = tegra_dma_slave_config;
 	tdma->dma_dev.device_terminate_all = tegra_dma_terminate_all;
+	tdma->dma_dev.device_synchronize = tegra_dma_synchronize;
 	tdma->dma_dev.device_tx_status = tegra_dma_tx_status;
 	tdma->dma_dev.device_issue_pending = tegra_dma_issue_pending;
 
-- 
2.24.0

[PATCH v6 04/16] dmaengine: tegra-apb: Prevent race conditions on channel's freeing

[Dmitry Osipenko]

It's incorrect to check the channel's "busy" state without taking a lock.
That shouldn't cause any real troubles, nevertheless it's always better
not to have any race conditions in the code.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index f56881500a23..766c2c9eac8e 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1294,8 +1294,7 @@ static void tegra_dma_free_chan_resources(struct dma_chan *dc)
 
 	dev_dbg(tdc2dev(tdc), "Freeing channel %d\n", tdc->id);
 
-	if (tdc->busy)
-		tegra_dma_terminate_all(dc);
+	tegra_dma_terminate_all(dc);
 
 	spin_lock_irqsave(&tdc->lock, flags);
 	list_splice_init(&tdc->pending_sg_req, &sg_req_list);
-- 
2.24.0

[PATCH v6 05/16] dmaengine: tegra-apb: Clean up tasklet releasing

[Dmitry Osipenko]

There is no need to kill tasklet when driver's probe fails because tasklet
can't be scheduled at this time. It is also cleaner to kill tasklet on
channel's freeing rather than to kill it on driver's removal, otherwise
tasklet could perform a dummy execution after channel's releasing, which
isn't very nice.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 6 +-----
 1 file changed, 1 insertion(+), 5 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 766c2c9eac8e..aafad50d075e 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1287,7 +1287,6 @@ static void tegra_dma_free_chan_resources(struct dma_chan *dc)
 	struct tegra_dma_sg_req *sg_req;
 	struct list_head dma_desc_list;
 	struct list_head sg_req_list;
-	unsigned long flags;
 
 	INIT_LIST_HEAD(&dma_desc_list);
 	INIT_LIST_HEAD(&sg_req_list);
@@ -1295,15 +1294,14 @@ static void tegra_dma_free_chan_resources(struct dma_chan *dc)
 	dev_dbg(tdc2dev(tdc), "Freeing channel %d\n", tdc->id);
 
 	tegra_dma_terminate_all(dc);
+	tasklet_kill(&tdc->tasklet);
 
-	spin_lock_irqsave(&tdc->lock, flags);
 	list_splice_init(&tdc->pending_sg_req, &sg_req_list);
 	list_splice_init(&tdc->free_sg_req, &sg_req_list);
 	list_splice_init(&tdc->free_dma_desc, &dma_desc_list);
 	INIT_LIST_HEAD(&tdc->cb_desc);
 	tdc->config_init = false;
 	tdc->isr_handler = NULL;
-	spin_unlock_irqrestore(&tdc->lock, flags);
 
 	while (!list_empty(&dma_desc_list)) {
 		dma_desc = list_first_entry(&dma_desc_list,
@@ -1542,7 +1540,6 @@ static int tegra_dma_probe(struct platform_device *pdev)
 		struct tegra_dma_channel *tdc = &tdma->channels[i];
 
 		free_irq(tdc->irq, tdc);
-		tasklet_kill(&tdc->tasklet);
 	}
 
 	pm_runtime_disable(&pdev->dev);
@@ -1562,7 +1559,6 @@ static int tegra_dma_remove(struct platform_device *pdev)
 	for (i = 0; i < tdma->chip_data->nr_channels; ++i) {
 		tdc = &tdma->channels[i];
 		free_irq(tdc->irq, tdc);
-		tasklet_kill(&tdc->tasklet);
 	}
 
 	pm_runtime_disable(&pdev->dev);
-- 
2.24.0

[PATCH v6 06/16] dmaengine: tegra-apb: Use devm_platform_ioremap_resource

[Dmitry Osipenko]

Use devm_platform_ioremap_resource to keep code cleaner a tad.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index aafad50d075e..f44291207928 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1402,8 +1402,7 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	tdma->chip_data = cdata;
 	platform_set_drvdata(pdev, tdma);
 
-	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
-	tdma->base_addr = devm_ioremap_resource(&pdev->dev, res);
+	tdma->base_addr = devm_platform_ioremap_resource(pdev, 0);
 	if (IS_ERR(tdma->base_addr))
 		return PTR_ERR(tdma->base_addr);
 
-- 
2.24.0

[PATCH v6 07/16] dmaengine: tegra-apb: Use devm_request_irq

[Dmitry Osipenko]

Use resource-managed variant of request_irq for brevity.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 35 +++++++++++------------------------
 1 file changed, 11 insertions(+), 24 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index f44291207928..dff21e80ffa4 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -182,7 +182,6 @@ struct tegra_dma_channel {
 	char			name[12];
 	bool			config_init;
 	int			id;
-	int			irq;
 	void __iomem		*chan_addr;
 	spinlock_t		lock;
 	bool			busy;
@@ -1380,7 +1379,6 @@ static const struct tegra_dma_chip_data tegra148_dma_chip_data = {
 
 static int tegra_dma_probe(struct platform_device *pdev)
 {
-	struct resource *res;
 	struct tegra_dma *tdma;
 	int ret;
 	int i;
@@ -1446,25 +1444,27 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	INIT_LIST_HEAD(&tdma->dma_dev.channels);
 	for (i = 0; i < cdata->nr_channels; i++) {
 		struct tegra_dma_channel *tdc = &tdma->channels[i];
+		int irq;
 
 		tdc->chan_addr = tdma->base_addr +
 				 TEGRA_APBDMA_CHANNEL_BASE_ADD_OFFSET +
 				 (i * cdata->channel_reg_size);
 
-		res = platform_get_resource(pdev, IORESOURCE_IRQ, i);
-		if (!res) {
-			ret = -EINVAL;
+		irq = platform_get_irq(pdev, i);
+		if (irq < 0) {
+			ret = irq;
 			dev_err(&pdev->dev, "No irq resource for chan %d\n", i);
-			goto err_irq;
+			goto err_pm_disable;
 		}
-		tdc->irq = res->start;
+
 		snprintf(tdc->name, sizeof(tdc->name), "apbdma.%d", i);
-		ret = request_irq(tdc->irq, tegra_dma_isr, 0, tdc->name, tdc);
+		ret = devm_request_irq(&pdev->dev, irq, tegra_dma_isr, 0,
+				       tdc->name, tdc);
 		if (ret) {
 			dev_err(&pdev->dev,
 				"request_irq failed with err %d channel %d\n",
 				ret, i);
-			goto err_irq;
+			goto err_pm_disable;
 		}
 
 		tdc->dma_chan.device = &tdma->dma_dev;
@@ -1517,7 +1517,7 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	if (ret < 0) {
 		dev_err(&pdev->dev,
 			"Tegra20 APB DMA driver registration failed %d\n", ret);
-		goto err_irq;
+		goto err_pm_disable;
 	}
 
 	ret = of_dma_controller_register(pdev->dev.of_node,
@@ -1534,13 +1534,7 @@ static int tegra_dma_probe(struct platform_device *pdev)
 
 err_unregister_dma_dev:
 	dma_async_device_unregister(&tdma->dma_dev);
-err_irq:
-	while (--i >= 0) {
-		struct tegra_dma_channel *tdc = &tdma->channels[i];
-
-		free_irq(tdc->irq, tdc);
-	}
-
+err_pm_disable:
 	pm_runtime_disable(&pdev->dev);
 	if (!pm_runtime_status_suspended(&pdev->dev))
 		tegra_dma_runtime_suspend(&pdev->dev);
@@ -1550,16 +1544,9 @@ static int tegra_dma_probe(struct platform_device *pdev)
 static int tegra_dma_remove(struct platform_device *pdev)
 {
 	struct tegra_dma *tdma = platform_get_drvdata(pdev);
-	int i;
-	struct tegra_dma_channel *tdc;
 
 	dma_async_device_unregister(&tdma->dma_dev);
 
-	for (i = 0; i < tdma->chip_data->nr_channels; ++i) {
-		tdc = &tdma->channels[i];
-		free_irq(tdc->irq, tdc);
-	}
-
 	pm_runtime_disable(&pdev->dev);
 	if (!pm_runtime_status_suspended(&pdev->dev))
 		tegra_dma_runtime_suspend(&pdev->dev);
-- 
2.24.0

[PATCH v6 08/16] dmaengine: tegra-apb: Fix coding style problems

[Dmitry Osipenko]

This patch fixes few dozens of coding style problems reported by
checkpatch and prettifies code where makes sense.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 275 ++++++++++++++++++----------------
 1 file changed, 144 insertions(+), 131 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index dff21e80ffa4..b4535b3a07ce 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -59,7 +59,7 @@
 #define TEGRA_APBDMA_STATUS_COUNT_MASK		0xFFFC
 
 #define TEGRA_APBDMA_CHAN_CSRE			0x00C
-#define TEGRA_APBDMA_CHAN_CSRE_PAUSE		(1 << 31)
+#define TEGRA_APBDMA_CHAN_CSRE_PAUSE		BIT(31)
 
 /* AHB memory address */
 #define TEGRA_APBDMA_CHAN_AHBPTR		0x010
@@ -120,21 +120,21 @@ struct tegra_dma;
  * @support_separate_wcount_reg: Support separate word count register.
  */
 struct tegra_dma_chip_data {
-	int nr_channels;
-	int channel_reg_size;
-	int max_dma_count;
+	unsigned int nr_channels;
+	unsigned int channel_reg_size;
+	unsigned int max_dma_count;
 	bool support_channel_pause;
 	bool support_separate_wcount_reg;
 };
 
 /* DMA channel registers */
 struct tegra_dma_channel_regs {
-	unsigned long	csr;
-	unsigned long	ahb_ptr;
-	unsigned long	apb_ptr;
-	unsigned long	ahb_seq;
-	unsigned long	apb_seq;
-	unsigned long	wcount;
+	u32 csr;
+	u32 ahb_ptr;
+	u32 apb_ptr;
+	u32 ahb_seq;
+	u32 apb_seq;
+	u32 wcount;
 };
 
 /*
@@ -168,7 +168,7 @@ struct tegra_dma_desc {
 	struct list_head		node;
 	struct list_head		tx_list;
 	struct list_head		cb_node;
-	int				cb_count;
+	unsigned int			cb_count;
 };
 
 struct tegra_dma_channel;
@@ -181,7 +181,7 @@ struct tegra_dma_channel {
 	struct dma_chan		dma_chan;
 	char			name[12];
 	bool			config_init;
-	int			id;
+	unsigned int		id;
 	void __iomem		*chan_addr;
 	spinlock_t		lock;
 	bool			busy;
@@ -201,7 +201,7 @@ struct tegra_dma_channel {
 	/* Channel-slave specific configuration */
 	unsigned int slave_id;
 	struct dma_slave_config dma_sconfig;
-	struct tegra_dma_channel_regs	channel_reg;
+	struct tegra_dma_channel_regs channel_reg;
 };
 
 /* tegra_dma: Tegra DMA specific information */
@@ -239,7 +239,7 @@ static inline u32 tdma_read(struct tegra_dma *tdma, u32 reg)
 }
 
 static inline void tdc_write(struct tegra_dma_channel *tdc,
-		u32 reg, u32 val)
+			     u32 reg, u32 val)
 {
 	writel(val, tdc->chan_addr + reg);
 }
@@ -254,8 +254,8 @@ static inline struct tegra_dma_channel *to_tegra_dma_chan(struct dma_chan *dc)
 	return container_of(dc, struct tegra_dma_channel, dma_chan);
 }
 
-static inline struct tegra_dma_desc *txd_to_tegra_dma_desc(
-		struct dma_async_tx_descriptor *td)
+static inline struct tegra_dma_desc *
+txd_to_tegra_dma_desc(struct dma_async_tx_descriptor *td)
 {
 	return container_of(td, struct tegra_dma_desc, txd);
 }
@@ -270,8 +270,7 @@ static int tegra_dma_runtime_suspend(struct device *dev);
 static int tegra_dma_runtime_resume(struct device *dev);
 
 /* Get DMA desc from free list, if not there then allocate it.  */
-static struct tegra_dma_desc *tegra_dma_desc_get(
-		struct tegra_dma_channel *tdc)
+static struct tegra_dma_desc *tegra_dma_desc_get(struct tegra_dma_channel *tdc)
 {
 	struct tegra_dma_desc *dma_desc;
 	unsigned long flags;
@@ -298,11 +297,12 @@ static struct tegra_dma_desc *tegra_dma_desc_get(
 	dma_async_tx_descriptor_init(&dma_desc->txd, &tdc->dma_chan);
 	dma_desc->txd.tx_submit = tegra_dma_tx_submit;
 	dma_desc->txd.flags = 0;
+
 	return dma_desc;
 }
 
 static void tegra_dma_desc_put(struct tegra_dma_channel *tdc,
-		struct tegra_dma_desc *dma_desc)
+			       struct tegra_dma_desc *dma_desc)
 {
 	unsigned long flags;
 
@@ -313,29 +313,29 @@ static void tegra_dma_desc_put(struct tegra_dma_channel *tdc,
 	spin_unlock_irqrestore(&tdc->lock, flags);
 }
 
-static struct tegra_dma_sg_req *tegra_dma_sg_req_get(
-		struct tegra_dma_channel *tdc)
+static struct tegra_dma_sg_req *
+tegra_dma_sg_req_get(struct tegra_dma_channel *tdc)
 {
-	struct tegra_dma_sg_req *sg_req = NULL;
+	struct tegra_dma_sg_req *sg_req;
 	unsigned long flags;
 
 	spin_lock_irqsave(&tdc->lock, flags);
 	if (!list_empty(&tdc->free_sg_req)) {
-		sg_req = list_first_entry(&tdc->free_sg_req,
-					typeof(*sg_req), node);
+		sg_req = list_first_entry(&tdc->free_sg_req, typeof(*sg_req),
+					  node);
 		list_del(&sg_req->node);
 		spin_unlock_irqrestore(&tdc->lock, flags);
 		return sg_req;
 	}
 	spin_unlock_irqrestore(&tdc->lock, flags);
 
-	sg_req = kzalloc(sizeof(struct tegra_dma_sg_req), GFP_NOWAIT);
+	sg_req = kzalloc(sizeof(*sg_req), GFP_NOWAIT);
 
 	return sg_req;
 }
 
 static int tegra_dma_slave_config(struct dma_chan *dc,
-		struct dma_slave_config *sconfig)
+				  struct dma_slave_config *sconfig)
 {
 	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
 
@@ -352,11 +352,12 @@ static int tegra_dma_slave_config(struct dma_chan *dc,
 		tdc->slave_id = sconfig->slave_id;
 	}
 	tdc->config_init = true;
+
 	return 0;
 }
 
 static void tegra_dma_global_pause(struct tegra_dma_channel *tdc,
-	bool wait_for_burst_complete)
+				   bool wait_for_burst_complete)
 {
 	struct tegra_dma *tdma = tdc->tdma;
 
@@ -391,13 +392,13 @@ static void tegra_dma_global_resume(struct tegra_dma_channel *tdc)
 }
 
 static void tegra_dma_pause(struct tegra_dma_channel *tdc,
-	bool wait_for_burst_complete)
+			    bool wait_for_burst_complete)
 {
 	struct tegra_dma *tdma = tdc->tdma;
 
 	if (tdma->chip_data->support_channel_pause) {
 		tdc_write(tdc, TEGRA_APBDMA_CHAN_CSRE,
-				TEGRA_APBDMA_CHAN_CSRE_PAUSE);
+			  TEGRA_APBDMA_CHAN_CSRE_PAUSE);
 		if (wait_for_burst_complete)
 			udelay(TEGRA_APBDMA_BURST_COMPLETE_TIME);
 	} else {
@@ -409,17 +410,15 @@ static void tegra_dma_resume(struct tegra_dma_channel *tdc)
 {
 	struct tegra_dma *tdma = tdc->tdma;
 
-	if (tdma->chip_data->support_channel_pause) {
+	if (tdma->chip_data->support_channel_pause)
 		tdc_write(tdc, TEGRA_APBDMA_CHAN_CSRE, 0);
-	} else {
+	else
 		tegra_dma_global_resume(tdc);
-	}
 }
 
 static void tegra_dma_stop(struct tegra_dma_channel *tdc)
 {
-	u32 csr;
-	u32 status;
+	u32 csr, status;
 
 	/* Disable interrupts */
 	csr = tdc_read(tdc, TEGRA_APBDMA_CHAN_CSR);
@@ -440,7 +439,7 @@ static void tegra_dma_stop(struct tegra_dma_channel *tdc)
 }
 
 static void tegra_dma_start(struct tegra_dma_channel *tdc,
-		struct tegra_dma_sg_req *sg_req)
+			    struct tegra_dma_sg_req *sg_req)
 {
 	struct tegra_dma_channel_regs *ch_regs = &sg_req->ch_regs;
 
@@ -454,11 +453,11 @@ static void tegra_dma_start(struct tegra_dma_channel *tdc,
 
 	/* Start DMA */
 	tdc_write(tdc, TEGRA_APBDMA_CHAN_CSR,
-				ch_regs->csr | TEGRA_APBDMA_CSR_ENB);
+		  ch_regs->csr | TEGRA_APBDMA_CSR_ENB);
 }
 
 static void tegra_dma_configure_for_next(struct tegra_dma_channel *tdc,
-		struct tegra_dma_sg_req *nsg_req)
+					 struct tegra_dma_sg_req *nsg_req)
 {
 	unsigned long status;
 
@@ -492,9 +491,9 @@ static void tegra_dma_configure_for_next(struct tegra_dma_channel *tdc,
 	tdc_write(tdc, TEGRA_APBDMA_CHAN_AHBPTR, nsg_req->ch_regs.ahb_ptr);
 	if (tdc->tdma->chip_data->support_separate_wcount_reg)
 		tdc_write(tdc, TEGRA_APBDMA_CHAN_WCOUNT,
-						nsg_req->ch_regs.wcount);
+			  nsg_req->ch_regs.wcount);
 	tdc_write(tdc, TEGRA_APBDMA_CHAN_CSR,
-				nsg_req->ch_regs.csr | TEGRA_APBDMA_CSR_ENB);
+		  nsg_req->ch_regs.csr | TEGRA_APBDMA_CSR_ENB);
 	nsg_req->configured = true;
 	nsg_req->words_xferred = 0;
 
@@ -508,8 +507,7 @@ static void tdc_start_head_req(struct tegra_dma_channel *tdc)
 	if (list_empty(&tdc->pending_sg_req))
 		return;
 
-	sg_req = list_first_entry(&tdc->pending_sg_req,
-					typeof(*sg_req), node);
+	sg_req = list_first_entry(&tdc->pending_sg_req, typeof(*sg_req), node);
 	tegra_dma_start(tdc, sg_req);
 	sg_req->configured = true;
 	sg_req->words_xferred = 0;
@@ -518,34 +516,35 @@ static void tdc_start_head_req(struct tegra_dma_channel *tdc)
 
 static void tdc_configure_next_head_desc(struct tegra_dma_channel *tdc)
 {
-	struct tegra_dma_sg_req *hsgreq;
-	struct tegra_dma_sg_req *hnsgreq;
+	struct tegra_dma_sg_req *hsgreq, *hnsgreq;
 
 	if (list_empty(&tdc->pending_sg_req))
 		return;
 
 	hsgreq = list_first_entry(&tdc->pending_sg_req, typeof(*hsgreq), node);
 	if (!list_is_last(&hsgreq->node, &tdc->pending_sg_req)) {
-		hnsgreq = list_first_entry(&hsgreq->node,
-					typeof(*hnsgreq), node);
+		hnsgreq = list_first_entry(&hsgreq->node, typeof(*hnsgreq),
+					   node);
 		tegra_dma_configure_for_next(tdc, hnsgreq);
 	}
 }
 
-static inline int get_current_xferred_count(struct tegra_dma_channel *tdc,
-	struct tegra_dma_sg_req *sg_req, unsigned long status)
+static inline unsigned int
+get_current_xferred_count(struct tegra_dma_channel *tdc,
+			  struct tegra_dma_sg_req *sg_req,
+			  unsigned long status)
 {
 	return sg_req->req_len - (status & TEGRA_APBDMA_STATUS_COUNT_MASK) - 4;
 }
 
 static void tegra_dma_abort_all(struct tegra_dma_channel *tdc)
 {
-	struct tegra_dma_sg_req *sgreq;
 	struct tegra_dma_desc *dma_desc;
+	struct tegra_dma_sg_req *sgreq;
 
 	while (!list_empty(&tdc->pending_sg_req)) {
-		sgreq = list_first_entry(&tdc->pending_sg_req,
-						typeof(*sgreq), node);
+		sgreq = list_first_entry(&tdc->pending_sg_req, typeof(*sgreq),
+					 node);
 		list_move_tail(&sgreq->node, &tdc->free_sg_req);
 		if (sgreq->last_sg) {
 			dma_desc = sgreq->dma_desc;
@@ -555,7 +554,7 @@ static void tegra_dma_abort_all(struct tegra_dma_channel *tdc)
 			/* Add in cb list if it is not there. */
 			if (!dma_desc->cb_count)
 				list_add_tail(&dma_desc->cb_node,
-							&tdc->cb_desc);
+					      &tdc->cb_desc);
 			dma_desc->cb_count++;
 		}
 	}
@@ -563,9 +562,10 @@ static void tegra_dma_abort_all(struct tegra_dma_channel *tdc)
 }
 
 static bool handle_continuous_head_request(struct tegra_dma_channel *tdc,
-		struct tegra_dma_sg_req *last_sg_req, bool to_terminate)
+					   struct tegra_dma_sg_req *last_sg_req,
+					   bool to_terminate)
 {
-	struct tegra_dma_sg_req *hsgreq = NULL;
+	struct tegra_dma_sg_req *hsgreq;
 
 	if (list_empty(&tdc->pending_sg_req)) {
 		dev_err(tdc2dev(tdc), "DMA is running without req\n");
@@ -589,14 +589,15 @@ static bool handle_continuous_head_request(struct tegra_dma_channel *tdc,
 	/* Configure next request */
 	if (!to_terminate)
 		tdc_configure_next_head_desc(tdc);
+
 	return true;
 }
 
 static void handle_once_dma_done(struct tegra_dma_channel *tdc,
-	bool to_terminate)
+				 bool to_terminate)
 {
-	struct tegra_dma_sg_req *sgreq;
 	struct tegra_dma_desc *dma_desc;
+	struct tegra_dma_sg_req *sgreq;
 
 	tdc->busy = false;
 	sgreq = list_first_entry(&tdc->pending_sg_req, typeof(*sgreq), node);
@@ -622,10 +623,10 @@ static void handle_once_dma_done(struct tegra_dma_channel *tdc,
 }
 
 static void handle_cont_sngl_cycle_dma_done(struct tegra_dma_channel *tdc,
-		bool to_terminate)
+					    bool to_terminate)
 {
-	struct tegra_dma_sg_req *sgreq;
 	struct tegra_dma_desc *dma_desc;
+	struct tegra_dma_sg_req *sgreq;
 	bool st;
 
 	sgreq = list_first_entry(&tdc->pending_sg_req, typeof(*sgreq), node);
@@ -657,13 +658,13 @@ static void tegra_dma_tasklet(unsigned long data)
 	struct tegra_dma_channel *tdc = (struct tegra_dma_channel *)data;
 	struct dmaengine_desc_callback cb;
 	struct tegra_dma_desc *dma_desc;
+	unsigned int cb_count;
 	unsigned long flags;
-	int cb_count;
 
 	spin_lock_irqsave(&tdc->lock, flags);
 	while (!list_empty(&tdc->cb_desc)) {
-		dma_desc  = list_first_entry(&tdc->cb_desc,
-					typeof(*dma_desc), cb_node);
+		dma_desc = list_first_entry(&tdc->cb_desc, typeof(*dma_desc),
+					    cb_node);
 		list_del(&dma_desc->cb_node);
 		dmaengine_desc_get_callback(&dma_desc->txd, &cb);
 		cb_count = dma_desc->cb_count;
@@ -681,8 +682,8 @@ static void tegra_dma_tasklet(unsigned long data)
 static irqreturn_t tegra_dma_isr(int irq, void *dev_id)
 {
 	struct tegra_dma_channel *tdc = dev_id;
-	unsigned long status;
 	unsigned long flags;
+	u32 status;
 
 	spin_lock_irqsave(&tdc->lock, flags);
 
@@ -697,8 +698,9 @@ static irqreturn_t tegra_dma_isr(int irq, void *dev_id)
 	}
 
 	spin_unlock_irqrestore(&tdc->lock, flags);
-	dev_info(tdc2dev(tdc),
-		"Interrupt already served status 0x%08lx\n", status);
+	dev_info(tdc2dev(tdc), "Interrupt already served status 0x%08x\n",
+		 status);
+
 	return IRQ_NONE;
 }
 
@@ -714,6 +716,7 @@ static dma_cookie_t tegra_dma_tx_submit(struct dma_async_tx_descriptor *txd)
 	cookie = dma_cookie_assign(&dma_desc->txd);
 	list_splice_tail_init(&dma_desc->tx_list, &tdc->pending_sg_req);
 	spin_unlock_irqrestore(&tdc->lock, flags);
+
 	return cookie;
 }
 
@@ -747,11 +750,10 @@ static void tegra_dma_issue_pending(struct dma_chan *dc)
 static int tegra_dma_terminate_all(struct dma_chan *dc)
 {
 	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
-	struct tegra_dma_sg_req *sgreq;
 	struct tegra_dma_desc *dma_desc;
+	struct tegra_dma_sg_req *sgreq;
 	unsigned long flags;
-	unsigned long status;
-	unsigned long wcount;
+	u32 status, wcount;
 	bool was_busy;
 
 	spin_lock_irqsave(&tdc->lock, flags);
@@ -777,8 +779,8 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
 	tegra_dma_stop(tdc);
 
 	if (!list_empty(&tdc->pending_sg_req) && was_busy) {
-		sgreq = list_first_entry(&tdc->pending_sg_req,
-					typeof(*sgreq), node);
+		sgreq = list_first_entry(&tdc->pending_sg_req, typeof(*sgreq),
+					 node);
 		sgreq->dma_desc->bytes_transferred +=
 				get_current_xferred_count(tdc, sgreq, wcount);
 	}
@@ -788,12 +790,13 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
 	tegra_dma_abort_all(tdc);
 
 	while (!list_empty(&tdc->cb_desc)) {
-		dma_desc  = list_first_entry(&tdc->cb_desc,
-					typeof(*dma_desc), cb_node);
+		dma_desc = list_first_entry(&tdc->cb_desc, typeof(*dma_desc),
+					    cb_node);
 		list_del(&dma_desc->cb_node);
 		dma_desc->cb_count = 0;
 	}
 	spin_unlock_irqrestore(&tdc->lock, flags);
+
 	return 0;
 }
 
@@ -807,7 +810,7 @@ static void tegra_dma_synchronize(struct dma_chan *dc)
 static unsigned int tegra_dma_sg_bytes_xferred(struct tegra_dma_channel *tdc,
 					       struct tegra_dma_sg_req *sg_req)
 {
-	unsigned long status, wcount = 0;
+	u32 status, wcount = 0;
 
 	if (!list_is_first(&sg_req->node, &tdc->pending_sg_req))
 		return 0;
@@ -864,7 +867,8 @@ static unsigned int tegra_dma_sg_bytes_xferred(struct tegra_dma_channel *tdc,
 }
 
 static enum dma_status tegra_dma_tx_status(struct dma_chan *dc,
-	dma_cookie_t cookie, struct dma_tx_state *txstate)
+					   dma_cookie_t cookie,
+					   struct dma_tx_state *txstate)
 {
 	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
 	struct tegra_dma_desc *dma_desc;
@@ -911,11 +915,12 @@ static enum dma_status tegra_dma_tx_status(struct dma_chan *dc,
 
 	trace_tegra_dma_tx_status(&tdc->dma_chan, cookie, txstate);
 	spin_unlock_irqrestore(&tdc->lock, flags);
+
 	return ret;
 }
 
-static inline int get_bus_width(struct tegra_dma_channel *tdc,
-		enum dma_slave_buswidth slave_bw)
+static inline unsigned int get_bus_width(struct tegra_dma_channel *tdc,
+					 enum dma_slave_buswidth slave_bw)
 {
 	switch (slave_bw) {
 	case DMA_SLAVE_BUSWIDTH_1_BYTE:
@@ -928,16 +933,17 @@ static inline int get_bus_width(struct tegra_dma_channel *tdc,
 		return TEGRA_APBDMA_APBSEQ_BUS_WIDTH_64;
 	default:
 		dev_warn(tdc2dev(tdc),
-			"slave bw is not supported, using 32bits\n");
+			 "slave bw is not supported, using 32bits\n");
 		return TEGRA_APBDMA_APBSEQ_BUS_WIDTH_32;
 	}
 }
 
-static inline int get_burst_size(struct tegra_dma_channel *tdc,
-	u32 burst_size, enum dma_slave_buswidth slave_bw, int len)
+static inline unsigned int get_burst_size(struct tegra_dma_channel *tdc,
+					  u32 burst_size,
+					  enum dma_slave_buswidth slave_bw,
+					  u32 len)
 {
-	int burst_byte;
-	int burst_ahb_width;
+	unsigned int burst_byte, burst_ahb_width;
 
 	/*
 	 * burst_size from client is in terms of the bus_width.
@@ -964,9 +970,12 @@ static inline int get_burst_size(struct tegra_dma_channel *tdc,
 }
 
 static int get_transfer_param(struct tegra_dma_channel *tdc,
-	enum dma_transfer_direction direction, unsigned long *apb_addr,
-	unsigned long *apb_seq,	unsigned long *csr, unsigned int *burst_size,
-	enum dma_slave_buswidth *slave_bw)
+			      enum dma_transfer_direction direction,
+			      u32 *apb_addr,
+			      u32 *apb_seq,
+			      u32 *csr,
+			      unsigned int *burst_size,
+			      enum dma_slave_buswidth *slave_bw)
 {
 	switch (direction) {
 	case DMA_MEM_TO_DEV:
@@ -987,13 +996,15 @@ static int get_transfer_param(struct tegra_dma_channel *tdc,
 
 	default:
 		dev_err(tdc2dev(tdc), "DMA direction is not supported\n");
-		return -EINVAL;
+		break;
 	}
+
 	return -EINVAL;
 }
 
 static void tegra_dma_prep_wcount(struct tegra_dma_channel *tdc,
-	struct tegra_dma_channel_regs *ch_regs, u32 len)
+				  struct tegra_dma_channel_regs *ch_regs,
+				  u32 len)
 {
 	u32 len_field = (len - 4) & 0xFFFC;
 
@@ -1003,20 +1014,23 @@ static void tegra_dma_prep_wcount(struct tegra_dma_channel *tdc,
 		ch_regs->csr |= len_field;
 }
 
-static struct dma_async_tx_descriptor *tegra_dma_prep_slave_sg(
-	struct dma_chan *dc, struct scatterlist *sgl, unsigned int sg_len,
-	enum dma_transfer_direction direction, unsigned long flags,
-	void *context)
+static struct dma_async_tx_descriptor *
+tegra_dma_prep_slave_sg(struct dma_chan *dc,
+			struct scatterlist *sgl,
+			unsigned int sg_len,
+			enum dma_transfer_direction direction,
+			unsigned long flags,
+			void *context)
 {
 	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
+	struct tegra_dma_sg_req *sg_req = NULL;
+	u32 csr, ahb_seq, apb_ptr, apb_seq;
+	enum dma_slave_buswidth slave_bw;
 	struct tegra_dma_desc *dma_desc;
-	unsigned int i;
-	struct scatterlist *sg;
-	unsigned long csr, ahb_seq, apb_ptr, apb_seq;
 	struct list_head req_list;
-	struct tegra_dma_sg_req  *sg_req = NULL;
-	u32 burst_size;
-	enum dma_slave_buswidth slave_bw;
+	struct scatterlist *sg;
+	unsigned int burst_size;
+	unsigned int i;
 
 	if (!tdc->config_init) {
 		dev_err(tdc2dev(tdc), "DMA channel is not configured\n");
@@ -1028,7 +1042,7 @@ static struct dma_async_tx_descriptor *tegra_dma_prep_slave_sg(
 	}
 
 	if (get_transfer_param(tdc, direction, &apb_ptr, &apb_seq, &csr,
-				&burst_size, &slave_bw) < 0)
+			       &burst_size, &slave_bw) < 0)
 		return NULL;
 
 	INIT_LIST_HEAD(&req_list);
@@ -1074,7 +1088,7 @@ static struct dma_async_tx_descriptor *tegra_dma_prep_slave_sg(
 		len = sg_dma_len(sg);
 
 		if ((len & 3) || (mem & 3) ||
-				(len > tdc->tdma->chip_data->max_dma_count)) {
+		    len > tdc->tdma->chip_data->max_dma_count) {
 			dev_err(tdc2dev(tdc),
 				"DMA length/memory address is not supported\n");
 			tegra_dma_desc_put(tdc, dma_desc);
@@ -1126,20 +1140,21 @@ static struct dma_async_tx_descriptor *tegra_dma_prep_slave_sg(
 	return &dma_desc->txd;
 }
 
-static struct dma_async_tx_descriptor *tegra_dma_prep_dma_cyclic(
-	struct dma_chan *dc, dma_addr_t buf_addr, size_t buf_len,
-	size_t period_len, enum dma_transfer_direction direction,
-	unsigned long flags)
+static struct dma_async_tx_descriptor *
+tegra_dma_prep_dma_cyclic(struct dma_chan *dc, dma_addr_t buf_addr,
+			  size_t buf_len,
+			  size_t period_len,
+			  enum dma_transfer_direction direction,
+			  unsigned long flags)
 {
 	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
-	struct tegra_dma_desc *dma_desc = NULL;
 	struct tegra_dma_sg_req *sg_req = NULL;
-	unsigned long csr, ahb_seq, apb_ptr, apb_seq;
-	int len;
-	size_t remain_len;
-	dma_addr_t mem = buf_addr;
-	u32 burst_size;
+	u32 csr, ahb_seq, apb_ptr, apb_seq;
 	enum dma_slave_buswidth slave_bw;
+	struct tegra_dma_desc *dma_desc;
+	dma_addr_t mem = buf_addr;
+	unsigned int burst_size;
+	size_t len, remain_len;
 
 	if (!buf_len || !period_len) {
 		dev_err(tdc2dev(tdc), "Invalid buffer/period len\n");
@@ -1173,13 +1188,13 @@ static struct dma_async_tx_descriptor *tegra_dma_prep_dma_cyclic(
 
 	len = period_len;
 	if ((len & 3) || (buf_addr & 3) ||
-			(len > tdc->tdma->chip_data->max_dma_count)) {
+	    len > tdc->tdma->chip_data->max_dma_count) {
 		dev_err(tdc2dev(tdc), "Req len/mem address is not correct\n");
 		return NULL;
 	}
 
 	if (get_transfer_param(tdc, direction, &apb_ptr, &apb_seq, &csr,
-				&burst_size, &slave_bw) < 0)
+			       &burst_size, &slave_bw) < 0)
 		return NULL;
 
 	ahb_seq = TEGRA_APBDMA_AHBSEQ_INTR_ENB;
@@ -1303,8 +1318,8 @@ static void tegra_dma_free_chan_resources(struct dma_chan *dc)
 	tdc->isr_handler = NULL;
 
 	while (!list_empty(&dma_desc_list)) {
-		dma_desc = list_first_entry(&dma_desc_list,
-					typeof(*dma_desc), node);
+		dma_desc = list_first_entry(&dma_desc_list, typeof(*dma_desc),
+					    node);
 		list_del(&dma_desc->node);
 		kfree(dma_desc);
 	}
@@ -1323,8 +1338,8 @@ static struct dma_chan *tegra_dma_of_xlate(struct of_phandle_args *dma_spec,
 					   struct of_dma *ofdma)
 {
 	struct tegra_dma *tdma = ofdma->of_dma_data;
-	struct dma_chan *chan;
 	struct tegra_dma_channel *tdc;
+	struct dma_chan *chan;
 
 	if (dma_spec->args[0] > TEGRA_APBDMA_CSR_REQ_SEL_MASK) {
 		dev_err(tdma->dev, "Invalid slave id: %d\n", dma_spec->args[0]);
@@ -1379,20 +1394,16 @@ static const struct tegra_dma_chip_data tegra148_dma_chip_data = {
 
 static int tegra_dma_probe(struct platform_device *pdev)
 {
+	const struct tegra_dma_chip_data *cdata;
 	struct tegra_dma *tdma;
+	unsigned int i;
+	size_t size;
 	int ret;
-	int i;
-	const struct tegra_dma_chip_data *cdata;
 
 	cdata = of_device_get_match_data(&pdev->dev);
-	if (!cdata) {
-		dev_err(&pdev->dev, "Error: No device match data found\n");
-		return -ENODEV;
-	}
+	size = struct_size(tdma, channels, cdata->nr_channels);
 
-	tdma = devm_kzalloc(&pdev->dev,
-			    struct_size(tdma, channels, cdata->nr_channels),
-			    GFP_KERNEL);
+	tdma = devm_kzalloc(&pdev->dev, size, GFP_KERNEL);
 	if (!tdma)
 		return -ENOMEM;
 
@@ -1424,10 +1435,8 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	else
 		ret = pm_runtime_get_sync(&pdev->dev);
 
-	if (ret < 0) {
-		pm_runtime_disable(&pdev->dev);
-		return ret;
-	}
+	if (ret < 0)
+		goto err_pm_disable;
 
 	/* Reset DMA controller */
 	reset_control_assert(tdma->rst);
@@ -1470,13 +1479,13 @@ static int tegra_dma_probe(struct platform_device *pdev)
 		tdc->dma_chan.device = &tdma->dma_dev;
 		dma_cookie_init(&tdc->dma_chan);
 		list_add_tail(&tdc->dma_chan.device_node,
-				&tdma->dma_dev.channels);
+			      &tdma->dma_dev.channels);
 		tdc->tdma = tdma;
 		tdc->id = i;
 		tdc->slave_id = TEGRA_APBDMA_SLAVE_ID_INVALID;
 
 		tasklet_init(&tdc->tasklet, tegra_dma_tasklet,
-				(unsigned long)tdc);
+			     (unsigned long)tdc);
 		spin_lock_init(&tdc->lock);
 
 		INIT_LIST_HEAD(&tdc->pending_sg_req);
@@ -1528,16 +1537,19 @@ static int tegra_dma_probe(struct platform_device *pdev)
 		goto err_unregister_dma_dev;
 	}
 
-	dev_info(&pdev->dev, "Tegra20 APB DMA driver register %d channels\n",
-			cdata->nr_channels);
+	dev_info(&pdev->dev, "Tegra20 APB DMA driver registered %u channels\n",
+		 cdata->nr_channels);
+
 	return 0;
 
 err_unregister_dma_dev:
 	dma_async_device_unregister(&tdma->dma_dev);
+
 err_pm_disable:
 	pm_runtime_disable(&pdev->dev);
 	if (!pm_runtime_status_suspended(&pdev->dev))
 		tegra_dma_runtime_suspend(&pdev->dev);
+
 	return ret;
 }
 
@@ -1557,7 +1569,7 @@ static int tegra_dma_remove(struct platform_device *pdev)
 static int tegra_dma_runtime_suspend(struct device *dev)
 {
 	struct tegra_dma *tdma = dev_get_drvdata(dev);
-	int i;
+	unsigned int i;
 
 	tdma->reg_gen = tdma_read(tdma, TEGRA_APBDMA_GENERAL);
 	for (i = 0; i < tdma->chip_data->nr_channels; i++) {
@@ -1586,7 +1598,8 @@ static int tegra_dma_runtime_suspend(struct device *dev)
 static int tegra_dma_runtime_resume(struct device *dev)
 {
 	struct tegra_dma *tdma = dev_get_drvdata(dev);
-	int i, ret;
+	unsigned int i;
+	int ret;
 
 	ret = clk_prepare_enable(tdma->dma_clk);
 	if (ret < 0) {
@@ -1614,7 +1627,7 @@ static int tegra_dma_runtime_resume(struct device *dev)
 		tdc_write(tdc, TEGRA_APBDMA_CHAN_AHBSEQ, ch_reg->ahb_seq);
 		tdc_write(tdc, TEGRA_APBDMA_CHAN_AHBPTR, ch_reg->ahb_ptr);
 		tdc_write(tdc, TEGRA_APBDMA_CHAN_CSR,
-			(ch_reg->csr & ~TEGRA_APBDMA_CSR_ENB));
+			  ch_reg->csr & ~TEGRA_APBDMA_CSR_ENB);
 	}
 
 	return 0;
-- 
2.24.0

[PATCH v6 09/16] dmaengine: tegra-apb: Remove unneeded initialization of tdc->config_init

[Dmitry Osipenko]

There is no need to re-initialize the already initialized variables.
The tdc->config_init=false after driver's probe and after channel's
freeing, so there is no need to re-initialize it on the channel's
allocation.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index b4535b3a07ce..7158bd3145c4 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1284,7 +1284,6 @@ static int tegra_dma_alloc_chan_resources(struct dma_chan *dc)
 	int ret;
 
 	dma_cookie_init(&tdc->dma_chan);
-	tdc->config_init = false;
 
 	ret = pm_runtime_get_sync(tdma->dev);
 	if (ret < 0)
-- 
2.24.0

[PATCH v6 10/16] dmaengine: tegra-apb: Remove assumptions about unavailable runtime PM

[Dmitry Osipenko]

The runtime PM is always available on all Tegra SoCs since the commit
40b2bb1b132a ("ARM: tegra: enforce PM requirement"), so there is no
need to handle the case of unavailable RPM in the code anymore.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 10 +---------
 1 file changed, 1 insertion(+), 9 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 7158bd3145c4..22b88ccff05d 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1429,11 +1429,8 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	spin_lock_init(&tdma->global_lock);
 
 	pm_runtime_enable(&pdev->dev);
-	if (!pm_runtime_enabled(&pdev->dev))
-		ret = tegra_dma_runtime_resume(&pdev->dev);
-	else
-		ret = pm_runtime_get_sync(&pdev->dev);
 
+	ret = pm_runtime_get_sync(&pdev->dev);
 	if (ret < 0)
 		goto err_pm_disable;
 
@@ -1546,8 +1543,6 @@ static int tegra_dma_probe(struct platform_device *pdev)
 
 err_pm_disable:
 	pm_runtime_disable(&pdev->dev);
-	if (!pm_runtime_status_suspended(&pdev->dev))
-		tegra_dma_runtime_suspend(&pdev->dev);
 
 	return ret;
 }
@@ -1557,10 +1552,7 @@ static int tegra_dma_remove(struct platform_device *pdev)
 	struct tegra_dma *tdma = platform_get_drvdata(pdev);
 
 	dma_async_device_unregister(&tdma->dma_dev);
-
 	pm_runtime_disable(&pdev->dev);
-	if (!pm_runtime_status_suspended(&pdev->dev))
-		tegra_dma_runtime_suspend(&pdev->dev);
 
 	return 0;
 }
-- 
2.24.0

[PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Dmitry Osipenko]

It's a bit impractical to enable hardware's clock at the time of DMA
channel's allocation because most of DMA client drivers allocate DMA
channel at the time of the driver's probing, and thus, DMA clock is kept
always-enabled in practice, defeating the whole purpose of runtime PM.

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 47 ++++++++++++++++++++++++-----------
 1 file changed, 32 insertions(+), 15 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 22b88ccff05d..0ee28d8e3c96 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -436,6 +436,8 @@ static void tegra_dma_stop(struct tegra_dma_channel *tdc)
 		tdc_write(tdc, TEGRA_APBDMA_CHAN_STATUS, status);
 	}
 	tdc->busy = false;
+
+	pm_runtime_put(tdc->tdma->dev);
 }
 
 static void tegra_dma_start(struct tegra_dma_channel *tdc,
@@ -500,18 +502,25 @@ static void tegra_dma_configure_for_next(struct tegra_dma_channel *tdc,
 	tegra_dma_resume(tdc);
 }
 
-static void tdc_start_head_req(struct tegra_dma_channel *tdc)
+static bool tdc_start_head_req(struct tegra_dma_channel *tdc)
 {
 	struct tegra_dma_sg_req *sg_req;
+	int err;
 
 	if (list_empty(&tdc->pending_sg_req))
-		return;
+		return false;
+
+	err = pm_runtime_get_sync(tdc->tdma->dev);
+	if (WARN_ON_ONCE(err < 0))
+		return false;
 
 	sg_req = list_first_entry(&tdc->pending_sg_req, typeof(*sg_req), node);
 	tegra_dma_start(tdc, sg_req);
 	sg_req->configured = true;
 	sg_req->words_xferred = 0;
 	tdc->busy = true;
+
+	return true;
 }
 
 static void tdc_configure_next_head_desc(struct tegra_dma_channel *tdc)
@@ -615,6 +624,8 @@ static void handle_once_dma_done(struct tegra_dma_channel *tdc,
 	}
 	list_add_tail(&sgreq->node, &tdc->free_sg_req);
 
+	pm_runtime_put(tdc->tdma->dev);
+
 	/* Do not start DMA if it is going to be terminate */
 	if (to_terminate || list_empty(&tdc->pending_sg_req))
 		return;
@@ -730,9 +741,7 @@ static void tegra_dma_issue_pending(struct dma_chan *dc)
 		dev_err(tdc2dev(tdc), "No DMA request\n");
 		goto end;
 	}
-	if (!tdc->busy) {
-		tdc_start_head_req(tdc);
-
+	if (!tdc->busy && tdc_start_head_req(tdc)) {
 		/* Continuous single mode: Configure next req */
 		if (tdc->cyclic) {
 			/*
@@ -775,6 +784,13 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
 	else
 		wcount = status;
 
+	/*
+	 * tegra_dma_stop() will drop the RPM's usage refcount, but
+	 * tegra_dma_resume() touches hardware and thus we should keep
+	 * the DMA clock active while it's needed.
+	 */
+	pm_runtime_get(tdc->tdma->dev);
+
 	was_busy = tdc->busy;
 	tegra_dma_stop(tdc);
 
@@ -786,6 +802,8 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
 	}
 	tegra_dma_resume(tdc);
 
+	pm_runtime_put(tdc->tdma->dev);
+
 skip_dma_stop:
 	tegra_dma_abort_all(tdc);
 
@@ -1280,22 +1298,15 @@ tegra_dma_prep_dma_cyclic(struct dma_chan *dc, dma_addr_t buf_addr,
 static int tegra_dma_alloc_chan_resources(struct dma_chan *dc)
 {
 	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
-	struct tegra_dma *tdma = tdc->tdma;
-	int ret;
 
 	dma_cookie_init(&tdc->dma_chan);
 
-	ret = pm_runtime_get_sync(tdma->dev);
-	if (ret < 0)
-		return ret;
-
 	return 0;
 }
 
 static void tegra_dma_free_chan_resources(struct dma_chan *dc)
 {
 	struct tegra_dma_channel *tdc = to_tegra_dma_chan(dc);
-	struct tegra_dma *tdma = tdc->tdma;
 	struct tegra_dma_desc *dma_desc;
 	struct tegra_dma_sg_req *sg_req;
 	struct list_head dma_desc_list;
@@ -1328,7 +1339,6 @@ static void tegra_dma_free_chan_resources(struct dma_chan *dc)
 		list_del(&sg_req->node);
 		kfree(sg_req);
 	}
-	pm_runtime_put(tdma->dev);
 
 	tdc->slave_id = TEGRA_APBDMA_SLAVE_ID_INVALID;
 }
@@ -1428,6 +1438,11 @@ static int tegra_dma_probe(struct platform_device *pdev)
 
 	spin_lock_init(&tdma->global_lock);
 
+	ret = clk_prepare(tdma->dma_clk);
+	if (ret)
+		return ret;
+
+	pm_runtime_irq_safe(&pdev->dev);
 	pm_runtime_enable(&pdev->dev);
 
 	ret = pm_runtime_get_sync(&pdev->dev);
@@ -1543,6 +1558,7 @@ static int tegra_dma_probe(struct platform_device *pdev)
 
 err_pm_disable:
 	pm_runtime_disable(&pdev->dev);
+	clk_unprepare(tdma->dma_clk);
 
 	return ret;
 }
@@ -1553,6 +1569,7 @@ static int tegra_dma_remove(struct platform_device *pdev)
 
 	dma_async_device_unregister(&tdma->dma_dev);
 	pm_runtime_disable(&pdev->dev);
+	clk_unprepare(tdma->dma_clk);
 
 	return 0;
 }
@@ -1581,7 +1598,7 @@ static int tegra_dma_runtime_suspend(struct device *dev)
 						  TEGRA_APBDMA_CHAN_WCOUNT);
 	}
 
-	clk_disable_unprepare(tdma->dma_clk);
+	clk_disable(tdma->dma_clk);
 
 	return 0;
 }
@@ -1592,7 +1609,7 @@ static int tegra_dma_runtime_resume(struct device *dev)
 	unsigned int i;
 	int ret;
 
-	ret = clk_prepare_enable(tdma->dma_clk);
+	ret = clk_enable(tdma->dma_clk);
 	if (ret < 0) {
 		dev_err(dev, "clk_enable failed: %d\n", ret);
 		return ret;
-- 
2.24.0

[PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Dmitry Osipenko]

It is enough to check whether hardware is busy on suspend and to reset
it across of suspend-resume because channel's configuration is fully
re-programmed on each DMA transaction anyways and because save-restore
of an active channel won't end up well without pausing transfer prior to
saving of the state (note that all channels shall be idling at the time of
suspend, so save-restore is not needed at all).

Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 133 ++++++++++++++++++----------------
 1 file changed, 69 insertions(+), 64 deletions(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 0ee28d8e3c96..4a371891b7c4 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1401,6 +1401,36 @@ static const struct tegra_dma_chip_data tegra148_dma_chip_data = {
 	.support_separate_wcount_reg = true,
 };
 
+static int tegra_dma_init_hw(struct tegra_dma *tdma)
+{
+	int err;
+
+	err = reset_control_assert(tdma->rst);
+	if (err) {
+		dev_err(tdma->dev, "failed to assert reset: %d\n", err);
+		return err;
+	}
+
+	err = clk_enable(tdma->dma_clk);
+	if (err) {
+		dev_err(tdma->dev, "failed to enable clk: %d\n", err);
+		return err;
+	}
+
+	/* reset DMA controller */
+	udelay(2);
+	reset_control_deassert(tdma->rst);
+
+	/* enable global DMA registers */
+	tdma_write(tdma, TEGRA_APBDMA_GENERAL, TEGRA_APBDMA_GENERAL_ENABLE);
+	tdma_write(tdma, TEGRA_APBDMA_CONTROL, 0);
+	tdma_write(tdma, TEGRA_APBDMA_IRQ_MASK_SET, 0xFFFFFFFF);
+
+	clk_disable(tdma->dma_clk);
+
+	return 0;
+}
+
 static int tegra_dma_probe(struct platform_device *pdev)
 {
 	const struct tegra_dma_chip_data *cdata;
@@ -1442,25 +1472,13 @@ static int tegra_dma_probe(struct platform_device *pdev)
 	if (ret)
 		return ret;
 
+	ret = tegra_dma_init_hw(tdma);
+	if (ret)
+		goto err_clk_unprepare;
+
 	pm_runtime_irq_safe(&pdev->dev);
 	pm_runtime_enable(&pdev->dev);
 
-	ret = pm_runtime_get_sync(&pdev->dev);
-	if (ret < 0)
-		goto err_pm_disable;
-
-	/* Reset DMA controller */
-	reset_control_assert(tdma->rst);
-	udelay(2);
-	reset_control_deassert(tdma->rst);
-
-	/* Enable global DMA registers */
-	tdma_write(tdma, TEGRA_APBDMA_GENERAL, TEGRA_APBDMA_GENERAL_ENABLE);
-	tdma_write(tdma, TEGRA_APBDMA_CONTROL, 0);
-	tdma_write(tdma, TEGRA_APBDMA_IRQ_MASK_SET, 0xFFFFFFFFul);
-
-	pm_runtime_put(&pdev->dev);
-
 	INIT_LIST_HEAD(&tdma->dma_dev.channels);
 	for (i = 0; i < cdata->nr_channels; i++) {
 		struct tegra_dma_channel *tdc = &tdma->channels[i];
@@ -1558,6 +1576,8 @@ static int tegra_dma_probe(struct platform_device *pdev)
 
 err_pm_disable:
 	pm_runtime_disable(&pdev->dev);
+
+err_clk_unprepare:
 	clk_unprepare(tdma->dma_clk);
 
 	return ret;
@@ -1577,26 +1597,6 @@ static int tegra_dma_remove(struct platform_device *pdev)
 static int tegra_dma_runtime_suspend(struct device *dev)
 {
 	struct tegra_dma *tdma = dev_get_drvdata(dev);
-	unsigned int i;
-
-	tdma->reg_gen = tdma_read(tdma, TEGRA_APBDMA_GENERAL);
-	for (i = 0; i < tdma->chip_data->nr_channels; i++) {
-		struct tegra_dma_channel *tdc = &tdma->channels[i];
-		struct tegra_dma_channel_regs *ch_reg = &tdc->channel_reg;
-
-		/* Only save the state of DMA channels that are in use */
-		if (!tdc->config_init)
-			continue;
-
-		ch_reg->csr = tdc_read(tdc, TEGRA_APBDMA_CHAN_CSR);
-		ch_reg->ahb_ptr = tdc_read(tdc, TEGRA_APBDMA_CHAN_AHBPTR);
-		ch_reg->apb_ptr = tdc_read(tdc, TEGRA_APBDMA_CHAN_APBPTR);
-		ch_reg->ahb_seq = tdc_read(tdc, TEGRA_APBDMA_CHAN_AHBSEQ);
-		ch_reg->apb_seq = tdc_read(tdc, TEGRA_APBDMA_CHAN_APBSEQ);
-		if (tdma->chip_data->support_separate_wcount_reg)
-			ch_reg->wcount = tdc_read(tdc,
-						  TEGRA_APBDMA_CHAN_WCOUNT);
-	}
 
 	clk_disable(tdma->dma_clk);
 
@@ -1606,46 +1606,51 @@ static int tegra_dma_runtime_suspend(struct device *dev)
 static int tegra_dma_runtime_resume(struct device *dev)
 {
 	struct tegra_dma *tdma = dev_get_drvdata(dev);
-	unsigned int i;
-	int ret;
 
-	ret = clk_enable(tdma->dma_clk);
-	if (ret < 0) {
-		dev_err(dev, "clk_enable failed: %d\n", ret);
-		return ret;
-	}
+	return clk_enable(tdma->dma_clk);
+}
 
-	tdma_write(tdma, TEGRA_APBDMA_GENERAL, tdma->reg_gen);
-	tdma_write(tdma, TEGRA_APBDMA_CONTROL, 0);
-	tdma_write(tdma, TEGRA_APBDMA_IRQ_MASK_SET, 0xFFFFFFFFul);
+static int __maybe_unused tegra_dma_dev_suspend(struct device *dev)
+{
+	struct tegra_dma *tdma = dev_get_drvdata(dev);
+	unsigned long flags;
+	unsigned int i;
+	bool busy;
 
 	for (i = 0; i < tdma->chip_data->nr_channels; i++) {
 		struct tegra_dma_channel *tdc = &tdma->channels[i];
-		struct tegra_dma_channel_regs *ch_reg = &tdc->channel_reg;
-
-		/* Only restore the state of DMA channels that are in use */
-		if (!tdc->config_init)
-			continue;
-
-		if (tdma->chip_data->support_separate_wcount_reg)
-			tdc_write(tdc, TEGRA_APBDMA_CHAN_WCOUNT,
-				  ch_reg->wcount);
-		tdc_write(tdc, TEGRA_APBDMA_CHAN_APBSEQ, ch_reg->apb_seq);
-		tdc_write(tdc, TEGRA_APBDMA_CHAN_APBPTR, ch_reg->apb_ptr);
-		tdc_write(tdc, TEGRA_APBDMA_CHAN_AHBSEQ, ch_reg->ahb_seq);
-		tdc_write(tdc, TEGRA_APBDMA_CHAN_AHBPTR, ch_reg->ahb_ptr);
-		tdc_write(tdc, TEGRA_APBDMA_CHAN_CSR,
-			  ch_reg->csr & ~TEGRA_APBDMA_CSR_ENB);
+
+		tasklet_kill(&tdc->tasklet);
+
+		spin_lock_irqsave(&tdc->lock, flags);
+		busy = tdc->busy;
+		spin_unlock_irqrestore(&tdc->lock, flags);
+
+		if (busy) {
+			dev_err(tdma->dev, "channel %u busy\n", i);
+			return -EBUSY;
+		}
 	}
 
-	return 0;
+	return pm_runtime_force_suspend(dev);
+}
+
+static int __maybe_unused tegra_dma_dev_resume(struct device *dev)
+{
+	struct tegra_dma *tdma = dev_get_drvdata(dev);
+	int err;
+
+	err = tegra_dma_init_hw(tdma);
+	if (err)
+		return err;
+
+	return pm_runtime_force_resume(dev);
 }
 
 static const struct dev_pm_ops tegra_dma_dev_pm_ops = {
 	SET_RUNTIME_PM_OPS(tegra_dma_runtime_suspend, tegra_dma_runtime_resume,
 			   NULL)
-	SET_SYSTEM_SLEEP_PM_OPS(pm_runtime_force_suspend,
-				pm_runtime_force_resume)
+	SET_SYSTEM_SLEEP_PM_OPS(tegra_dma_dev_suspend, tegra_dma_dev_resume)
 };
 
 static const struct of_device_id tegra_dma_of_match[] = {
-- 
2.24.0

[PATCH v6 13/16] dmaengine: tegra-apb: Add missing of_dma_controller_free

[Dmitry Osipenko]

The DMA controller shall be released on driver's removal.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 4a371891b7c4..8f4281ab6296 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1587,6 +1587,7 @@ static int tegra_dma_remove(struct platform_device *pdev)
 {
 	struct tegra_dma *tdma = platform_get_drvdata(pdev);
 
+	of_dma_controller_free(pdev->dev.of_node);
 	dma_async_device_unregister(&tdma->dma_dev);
 	pm_runtime_disable(&pdev->dev);
 	clk_unprepare(tdma->dma_clk);
-- 
2.24.0

[PATCH v6 14/16] dmaengine: tegra-apb: Allow to compile as a loadable kernel module

[Dmitry Osipenko]

The driver's removal was fixed by a recent commit and module load/unload
is working well now, tested on Tegra30.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/Kconfig | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index 5142da401db3..98daf3aafdcc 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -616,7 +616,7 @@ config TXX9_DMAC
 	  integrated in chips such as the Toshiba TX4927/38/39.
 
 config TEGRA20_APB_DMA
-	bool "NVIDIA Tegra20 APB DMA support"
+	tristate "NVIDIA Tegra20 APB DMA support"
 	depends on ARCH_TEGRA
 	select DMA_ENGINE
 	help
-- 
2.24.0

[PATCH v6 15/16] dmaengine: tegra-apb: Remove MODULE_ALIAS

[Dmitry Osipenko]

Tegra APB DMA driver is an Open Firmware driver, so it uses OF alias
naming scheme which overrides MODULE_ALIAS, meaning that MODULE_ALIAS
does nothing and could be removed safely.

Acked-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
---
 drivers/dma/tegra20-apb-dma.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 8f4281ab6296..756a5dbef118 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -1684,7 +1684,6 @@ static struct platform_driver tegra_dmac_driver = {
 
 module_platform_driver(tegra_dmac_driver);
 
-MODULE_ALIAS("platform:tegra20-apbdma");
 MODULE_DESCRIPTION("NVIDIA Tegra APB DMA Controller driver");
 MODULE_AUTHOR("Laxman Dewangan <ldewangan@nvidia.com>");
 MODULE_LICENSE("GPL v2");
-- 
2.24.0

[PATCH v6 16/16] dmaengine: tegra-apb: Support COMPILE_TEST

[Dmitry Osipenko]

There is nothing arch-specific in the driver's code, so let's enable
compile-testing for the driver.

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

diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
index 98daf3aafdcc..7fc725d928b2 100644
--- a/drivers/dma/Kconfig
+++ b/drivers/dma/Kconfig
@@ -617,7 +617,7 @@ config TXX9_DMAC
 
 config TEGRA20_APB_DMA
 	tristate "NVIDIA Tegra20 APB DMA support"
-	depends on ARCH_TEGRA
+	depends on ARCH_TEGRA || COMPILE_TEST
 	select DMA_ENGINE
 	help
 	  Support for the NVIDIA Tegra20 APB DMA controller driver. The
-- 
2.24.0

Re: [PATCH v6 08/16] dmaengine: tegra-apb: Fix coding style problems

[Jon Hunter]

On 30/01/2020 04:37, Dmitry Osipenko wrote:
> This patch fixes few dozens of coding style problems reported by
> checkpatch and prettifies code where makes sense.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/tegra20-apb-dma.c | 275 ++++++++++++++++++----------------
>  1 file changed, 144 insertions(+), 131 deletions(-)

Thanks!

Acked-by: Jon Hunter <jonathanh@nvidia.com>

Cheers
Jon

-- 
nvpublic

Re: [PATCH v6 09/16] dmaengine: tegra-apb: Remove unneeded initialization of tdc->config_init

[Jon Hunter]

On 30/01/2020 04:37, Dmitry Osipenko wrote:
> There is no need to re-initialize the already initialized variables.
> The tdc->config_init=false after driver's probe and after channel's
> freeing, so there is no need to re-initialize it on the channel's
> allocation.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/tegra20-apb-dma.c | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
> index b4535b3a07ce..7158bd3145c4 100644
> --- a/drivers/dma/tegra20-apb-dma.c
> +++ b/drivers/dma/tegra20-apb-dma.c
> @@ -1284,7 +1284,6 @@ static int tegra_dma_alloc_chan_resources(struct dma_chan *dc)
>  	int ret;
>  
>  	dma_cookie_init(&tdc->dma_chan);
> -	tdc->config_init = false;
>  
>  	ret = pm_runtime_get_sync(tdma->dev);
>  	if (ret < 0)
> 

Acked-by: Jon Hunter <jonathanh@nvidia.com>

Cheers
Jon

-- 
nvpublic

Re: [PATCH v6 10/16] dmaengine: tegra-apb: Remove assumptions about unavailable runtime PM

[Jon Hunter]

On 30/01/2020 04:37, Dmitry Osipenko wrote:
> The runtime PM is always available on all Tegra SoCs since the commit
> 40b2bb1b132a ("ARM: tegra: enforce PM requirement"), so there is no
> need to handle the case of unavailable RPM in the code anymore.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/tegra20-apb-dma.c | 10 +---------
>  1 file changed, 1 insertion(+), 9 deletions(-)
> 
> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
> index 7158bd3145c4..22b88ccff05d 100644
> --- a/drivers/dma/tegra20-apb-dma.c
> +++ b/drivers/dma/tegra20-apb-dma.c
> @@ -1429,11 +1429,8 @@ static int tegra_dma_probe(struct platform_device *pdev)
>  	spin_lock_init(&tdma->global_lock);
>  
>  	pm_runtime_enable(&pdev->dev);
> -	if (!pm_runtime_enabled(&pdev->dev))
> -		ret = tegra_dma_runtime_resume(&pdev->dev);
> -	else
> -		ret = pm_runtime_get_sync(&pdev->dev);
>  
> +	ret = pm_runtime_get_sync(&pdev->dev);
>  	if (ret < 0)
>  		goto err_pm_disable;
>  
> @@ -1546,8 +1543,6 @@ static int tegra_dma_probe(struct platform_device *pdev)
>  
>  err_pm_disable:
>  	pm_runtime_disable(&pdev->dev);
> -	if (!pm_runtime_status_suspended(&pdev->dev))
> -		tegra_dma_runtime_suspend(&pdev->dev);
>  
>  	return ret;
>  }
> @@ -1557,10 +1552,7 @@ static int tegra_dma_remove(struct platform_device *pdev)
>  	struct tegra_dma *tdma = platform_get_drvdata(pdev);
>  
>  	dma_async_device_unregister(&tdma->dma_dev);
> -
>  	pm_runtime_disable(&pdev->dev);
> -	if (!pm_runtime_status_suspended(&pdev->dev))
> -		tegra_dma_runtime_suspend(&pdev->dev);
>  
>  	return 0;
>  }

I wonder if we need to make the pm_runtime_put a pm_runtime_put_sync or
call pm_runtime_barrier() here, to ensure it has called the callback
when freeing the channel? Otherwise this is fine.

Jon

-- 
nvpublic

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Jon Hunter]

On 30/01/2020 04:37, Dmitry Osipenko wrote:
> It's a bit impractical to enable hardware's clock at the time of DMA
> channel's allocation because most of DMA client drivers allocate DMA
> channel at the time of the driver's probing, and thus, DMA clock is kept
> always-enabled in practice, defeating the whole purpose of runtime PM.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/tegra20-apb-dma.c | 47 ++++++++++++++++++++++++-----------
>  1 file changed, 32 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
> index 22b88ccff05d..0ee28d8e3c96 100644
> --- a/drivers/dma/tegra20-apb-dma.c
> +++ b/drivers/dma/tegra20-apb-dma.c
> @@ -436,6 +436,8 @@ static void tegra_dma_stop(struct tegra_dma_channel *tdc)
>  		tdc_write(tdc, TEGRA_APBDMA_CHAN_STATUS, status);
>  	}
>  	tdc->busy = false;
> +
> +	pm_runtime_put(tdc->tdma->dev);
>  }
>  
>  static void tegra_dma_start(struct tegra_dma_channel *tdc,
> @@ -500,18 +502,25 @@ static void tegra_dma_configure_for_next(struct tegra_dma_channel *tdc,
>  	tegra_dma_resume(tdc);
>  }
>  
> -static void tdc_start_head_req(struct tegra_dma_channel *tdc)
> +static bool tdc_start_head_req(struct tegra_dma_channel *tdc)
>  {
>  	struct tegra_dma_sg_req *sg_req;
> +	int err;
>  
>  	if (list_empty(&tdc->pending_sg_req))
> -		return;
> +		return false;
> +
> +	err = pm_runtime_get_sync(tdc->tdma->dev);
> +	if (WARN_ON_ONCE(err < 0))
> +		return false;
>  
>  	sg_req = list_first_entry(&tdc->pending_sg_req, typeof(*sg_req), node);
>  	tegra_dma_start(tdc, sg_req);
>  	sg_req->configured = true;
>  	sg_req->words_xferred = 0;
>  	tdc->busy = true;
> +
> +	return true;
>  }
>  
>  static void tdc_configure_next_head_desc(struct tegra_dma_channel *tdc)
> @@ -615,6 +624,8 @@ static void handle_once_dma_done(struct tegra_dma_channel *tdc,
>  	}
>  	list_add_tail(&sgreq->node, &tdc->free_sg_req);
>  
> +	pm_runtime_put(tdc->tdma->dev);
> +
>  	/* Do not start DMA if it is going to be terminate */
>  	if (to_terminate || list_empty(&tdc->pending_sg_req))
>  		return;
> @@ -730,9 +741,7 @@ static void tegra_dma_issue_pending(struct dma_chan *dc)
>  		dev_err(tdc2dev(tdc), "No DMA request\n");
>  		goto end;
>  	}
> -	if (!tdc->busy) {
> -		tdc_start_head_req(tdc);
> -
> +	if (!tdc->busy && tdc_start_head_req(tdc)) {
>  		/* Continuous single mode: Configure next req */
>  		if (tdc->cyclic) {
>  			/*
> @@ -775,6 +784,13 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
>  	else
>  		wcount = status;
>  
> +	/*
> +	 * tegra_dma_stop() will drop the RPM's usage refcount, but
> +	 * tegra_dma_resume() touches hardware and thus we should keep
> +	 * the DMA clock active while it's needed.
> +	 */
> +	pm_runtime_get(tdc->tdma->dev);
> +

Would it work and make it simpler to just enable in the issue_pending
and disable in the handle_once_dma_done or terminate_all?

diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
index 3a45079d11ec..86bbb45da93d 100644
--- a/drivers/dma/tegra20-apb-dma.c
+++ b/drivers/dma/tegra20-apb-dma.c
@@ -616,9 +616,14 @@ static void handle_once_dma_done(struct
tegra_dma_channel *tdc,
        list_add_tail(&sgreq->node, &tdc->free_sg_req);

        /* Do not start DMA if it is going to be terminate */
-       if (to_terminate || list_empty(&tdc->pending_sg_req))
+       if (to_terminate)
                return;

+       if (list_empty(&tdc->pending_sg_req)) {
+               pm_runtime_put(tdc->tdma->dev);
+               return;
+       }
+
        tdc_start_head_req(tdc);
 }

@@ -729,6 +734,11 @@ static void tegra_dma_issue_pending(struct dma_chan
*dc)
                goto end;
        }
        if (!tdc->busy) {
+               if (pm_runtime_get_sync(tdc->tdma->dev) < 0) {
+                       dev_err(tdc2dev(tdc), "Failed to enable DMA!\n");
+                       goto end;
+               }
+
                tdc_start_head_req(tdc);

                /* Continuous single mode: Configure next req */
@@ -788,6 +798,7 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
                                get_current_xferred_count(tdc, sgreq,
wcount);
        }
        tegra_dma_resume(tdc);
+       pm_runtime_put(tdc->tdma->dev);

 skip_dma_stop:
        tegra_dma_abort_all(tdc);

-- 
nvpublic

Re: [PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Jon Hunter]

On 30/01/2020 04:38, Dmitry Osipenko wrote:
> It is enough to check whether hardware is busy on suspend and to reset
> it across of suspend-resume because channel's configuration is fully
> re-programmed on each DMA transaction anyways and because save-restore
> of an active channel won't end up well without pausing transfer prior to
> saving of the state (note that all channels shall be idling at the time of
> suspend, so save-restore is not needed at all).

I guess if we ever wanted to support SNDRV_PCM_INFO_PAUSE for audio and
support the pause callback, then saving and restoring the channels could
be needed. Right now I believe that it will just terminate_all transfers
for audio on entering suspend. Any value in keeping this?

Jon

-- 
nvpublic

Re: [PATCH v6 16/16] dmaengine: tegra-apb: Support COMPILE_TEST

[Jon Hunter]

On 30/01/2020 04:38, Dmitry Osipenko wrote:
> There is nothing arch-specific in the driver's code, so let's enable
> compile-testing for the driver.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/Kconfig | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/dma/Kconfig b/drivers/dma/Kconfig
> index 98daf3aafdcc..7fc725d928b2 100644
> --- a/drivers/dma/Kconfig
> +++ b/drivers/dma/Kconfig
> @@ -617,7 +617,7 @@ config TXX9_DMAC
>  
>  config TEGRA20_APB_DMA
>  	tristate "NVIDIA Tegra20 APB DMA support"
> -	depends on ARCH_TEGRA
> +	depends on ARCH_TEGRA || COMPILE_TEST
>  	select DMA_ENGINE
>  	help
>  	  Support for the NVIDIA Tegra20 APB DMA controller driver. The

Acked-by: Jon Hunter <jonathanh@nvidia.com>

Cheers
Jon

-- 
nvpublic

Re: [PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Dmitry Osipenko]

30.01.2020 17:09, Jon Hunter пишет:
> 
> On 30/01/2020 04:38, Dmitry Osipenko wrote:
>> It is enough to check whether hardware is busy on suspend and to reset
>> it across of suspend-resume because channel's configuration is fully
>> re-programmed on each DMA transaction anyways and because save-restore
>> of an active channel won't end up well without pausing transfer prior to
>> saving of the state (note that all channels shall be idling at the time of
>> suspend, so save-restore is not needed at all).
> 
> I guess if we ever wanted to support SNDRV_PCM_INFO_PAUSE for audio and
> support the pause callback, then saving and restoring the channels could
> be needed. Right now I believe that it will just terminate_all transfers
> for audio on entering suspend. Any value in keeping this?

Indeed, looks like [1] pauses DMA during suspend if SNDRV_PCM_INFO_PAUSE
is supported.

[1]
https://elixir.bootlin.com/linux/v5.5/source/sound/core/pcm_dmaengine.c#L199

So we'll need to save-restore context only if DMA is in a paused state
during suspend, I'll adjust this patch to do that and will see if
enabling SNDRV_PCM_INFO_PAUSE works.

Re: [PATCH v6 10/16] dmaengine: tegra-apb: Remove assumptions about unavailable runtime PM

[Dmitry Osipenko]

30.01.2020 17:09, Jon Hunter пишет:
> 
> On 30/01/2020 04:37, Dmitry Osipenko wrote:
>> The runtime PM is always available on all Tegra SoCs since the commit
>> 40b2bb1b132a ("ARM: tegra: enforce PM requirement"), so there is no
>> need to handle the case of unavailable RPM in the code anymore.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>> ---
>>  drivers/dma/tegra20-apb-dma.c | 10 +---------
>>  1 file changed, 1 insertion(+), 9 deletions(-)
>>
>> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
>> index 7158bd3145c4..22b88ccff05d 100644
>> --- a/drivers/dma/tegra20-apb-dma.c
>> +++ b/drivers/dma/tegra20-apb-dma.c
>> @@ -1429,11 +1429,8 @@ static int tegra_dma_probe(struct platform_device *pdev)
>>  	spin_lock_init(&tdma->global_lock);
>>  
>>  	pm_runtime_enable(&pdev->dev);
>> -	if (!pm_runtime_enabled(&pdev->dev))
>> -		ret = tegra_dma_runtime_resume(&pdev->dev);
>> -	else
>> -		ret = pm_runtime_get_sync(&pdev->dev);
>>  
>> +	ret = pm_runtime_get_sync(&pdev->dev);
>>  	if (ret < 0)
>>  		goto err_pm_disable;
>>  
>> @@ -1546,8 +1543,6 @@ static int tegra_dma_probe(struct platform_device *pdev)
>>  
>>  err_pm_disable:
>>  	pm_runtime_disable(&pdev->dev);
>> -	if (!pm_runtime_status_suspended(&pdev->dev))
>> -		tegra_dma_runtime_suspend(&pdev->dev);
>>  
>>  	return ret;
>>  }
>> @@ -1557,10 +1552,7 @@ static int tegra_dma_remove(struct platform_device *pdev)
>>  	struct tegra_dma *tdma = platform_get_drvdata(pdev);
>>  
>>  	dma_async_device_unregister(&tdma->dma_dev);
>> -
>>  	pm_runtime_disable(&pdev->dev);
>> -	if (!pm_runtime_status_suspended(&pdev->dev))
>> -		tegra_dma_runtime_suspend(&pdev->dev);
>>  
>>  	return 0;
>>  }
> 
> I wonder if we need to make the pm_runtime_put a pm_runtime_put_sync or
> call pm_runtime_barrier() here, to ensure it has called the callback
> when freeing the channel? Otherwise this is fine.

The explicit RPM syncing shouldn't be needed because
pm_runtime_disable() takes care of doing that for us, please take a look at:

https://elixir.bootlin.com/linux/v5.5/source/drivers/base/power/runtime.c#L1360

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Dmitry Osipenko]

30.01.2020 17:09, Jon Hunter пишет:
> 
> On 30/01/2020 04:37, Dmitry Osipenko wrote:
>> It's a bit impractical to enable hardware's clock at the time of DMA
>> channel's allocation because most of DMA client drivers allocate DMA
>> channel at the time of the driver's probing, and thus, DMA clock is kept
>> always-enabled in practice, defeating the whole purpose of runtime PM.
>>
>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>> ---
>>  drivers/dma/tegra20-apb-dma.c | 47 ++++++++++++++++++++++++-----------
>>  1 file changed, 32 insertions(+), 15 deletions(-)
>>
>> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
>> index 22b88ccff05d..0ee28d8e3c96 100644
>> --- a/drivers/dma/tegra20-apb-dma.c
>> +++ b/drivers/dma/tegra20-apb-dma.c
>> @@ -436,6 +436,8 @@ static void tegra_dma_stop(struct tegra_dma_channel *tdc)
>>  		tdc_write(tdc, TEGRA_APBDMA_CHAN_STATUS, status);
>>  	}
>>  	tdc->busy = false;
>> +
>> +	pm_runtime_put(tdc->tdma->dev);
>>  }
>>  
>>  static void tegra_dma_start(struct tegra_dma_channel *tdc,
>> @@ -500,18 +502,25 @@ static void tegra_dma_configure_for_next(struct tegra_dma_channel *tdc,
>>  	tegra_dma_resume(tdc);
>>  }
>>  
>> -static void tdc_start_head_req(struct tegra_dma_channel *tdc)
>> +static bool tdc_start_head_req(struct tegra_dma_channel *tdc)
>>  {
>>  	struct tegra_dma_sg_req *sg_req;
>> +	int err;
>>  
>>  	if (list_empty(&tdc->pending_sg_req))
>> -		return;
>> +		return false;
>> +
>> +	err = pm_runtime_get_sync(tdc->tdma->dev);
>> +	if (WARN_ON_ONCE(err < 0))
>> +		return false;
>>  
>>  	sg_req = list_first_entry(&tdc->pending_sg_req, typeof(*sg_req), node);
>>  	tegra_dma_start(tdc, sg_req);
>>  	sg_req->configured = true;
>>  	sg_req->words_xferred = 0;
>>  	tdc->busy = true;
>> +
>> +	return true;
>>  }
>>  
>>  static void tdc_configure_next_head_desc(struct tegra_dma_channel *tdc)
>> @@ -615,6 +624,8 @@ static void handle_once_dma_done(struct tegra_dma_channel *tdc,
>>  	}
>>  	list_add_tail(&sgreq->node, &tdc->free_sg_req);
>>  
>> +	pm_runtime_put(tdc->tdma->dev);
>> +
>>  	/* Do not start DMA if it is going to be terminate */
>>  	if (to_terminate || list_empty(&tdc->pending_sg_req))
>>  		return;
>> @@ -730,9 +741,7 @@ static void tegra_dma_issue_pending(struct dma_chan *dc)
>>  		dev_err(tdc2dev(tdc), "No DMA request\n");
>>  		goto end;
>>  	}
>> -	if (!tdc->busy) {
>> -		tdc_start_head_req(tdc);
>> -
>> +	if (!tdc->busy && tdc_start_head_req(tdc)) {
>>  		/* Continuous single mode: Configure next req */
>>  		if (tdc->cyclic) {
>>  			/*
>> @@ -775,6 +784,13 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
>>  	else
>>  		wcount = status;
>>  
>> +	/*
>> +	 * tegra_dma_stop() will drop the RPM's usage refcount, but
>> +	 * tegra_dma_resume() touches hardware and thus we should keep
>> +	 * the DMA clock active while it's needed.
>> +	 */
>> +	pm_runtime_get(tdc->tdma->dev);
>> +
> 
> Would it work and make it simpler to just enable in the issue_pending
> and disable in the handle_once_dma_done or terminate_all?
> 
> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
> index 3a45079d11ec..86bbb45da93d 100644
> --- a/drivers/dma/tegra20-apb-dma.c
> +++ b/drivers/dma/tegra20-apb-dma.c
> @@ -616,9 +616,14 @@ static void handle_once_dma_done(struct
> tegra_dma_channel *tdc,
>         list_add_tail(&sgreq->node, &tdc->free_sg_req);
> 
>         /* Do not start DMA if it is going to be terminate */
> -       if (to_terminate || list_empty(&tdc->pending_sg_req))
> +       if (to_terminate)
>                 return;
> 
> +       if (list_empty(&tdc->pending_sg_req)) {
> +               pm_runtime_put(tdc->tdma->dev);
> +               return;
> +       }
> +
>         tdc_start_head_req(tdc);
>  }
> 
> @@ -729,6 +734,11 @@ static void tegra_dma_issue_pending(struct dma_chan
> *dc)
>                 goto end;
>         }
>         if (!tdc->busy) {
> +               if (pm_runtime_get_sync(tdc->tdma->dev) < 0) {
> +                       dev_err(tdc2dev(tdc), "Failed to enable DMA!\n");
> +                       goto end;
> +               }
> +
>                 tdc_start_head_req(tdc);
> 
>                 /* Continuous single mode: Configure next req */
> @@ -788,6 +798,7 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
>                                 get_current_xferred_count(tdc, sgreq,
> wcount);
>         }
>         tegra_dma_resume(tdc);
> +       pm_runtime_put(tdc->tdma->dev);
> 
>  skip_dma_stop:
>         tegra_dma_abort_all(tdc);
> 

The tegra_dma_stop() should put RPM anyways, which is missed in yours
sample. Please see handle_continuous_head_request().

I'm also finding the explicit get/put a bit easier to follow in the
code, don't you think so?

Re: [PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Dmitry Osipenko]

30.01.2020 19:08, Dmitry Osipenko пишет:
> 30.01.2020 17:09, Jon Hunter пишет:
>>
>> On 30/01/2020 04:38, Dmitry Osipenko wrote:
>>> It is enough to check whether hardware is busy on suspend and to reset
>>> it across of suspend-resume because channel's configuration is fully
>>> re-programmed on each DMA transaction anyways and because save-restore
>>> of an active channel won't end up well without pausing transfer prior to
>>> saving of the state (note that all channels shall be idling at the time of
>>> suspend, so save-restore is not needed at all).
>>
>> I guess if we ever wanted to support SNDRV_PCM_INFO_PAUSE for audio and
>> support the pause callback, then saving and restoring the channels could
>> be needed. Right now I believe that it will just terminate_all transfers
>> for audio on entering suspend. Any value in keeping this?
> 
> Indeed, looks like [1] pauses DMA during suspend if SNDRV_PCM_INFO_PAUSE
> is supported.
> 
> [1]
> https://elixir.bootlin.com/linux/v5.5/source/sound/core/pcm_dmaengine.c#L199
> 
> So we'll need to save-restore context only if DMA is in a paused state
> during suspend, I'll adjust this patch to do that and will see if
> enabling SNDRV_PCM_INFO_PAUSE works.

I started to look at it and found that the .device_pause() hook isn't
implemented by the driver. So, it's fine to remove the context's
save-restore for now.

Jon, what about to keep this patch as-is? Later on I'll take a look at
implementing the proper pausing functionality and try to cleanup code a
bit further (remove the free list usage, etc).

Re: [PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Dmitry Osipenko]

30.01.2020 21:06, Dmitry Osipenko пишет:
> 30.01.2020 19:08, Dmitry Osipenko пишет:
>> 30.01.2020 17:09, Jon Hunter пишет:
>>>
>>> On 30/01/2020 04:38, Dmitry Osipenko wrote:
>>>> It is enough to check whether hardware is busy on suspend and to reset
>>>> it across of suspend-resume because channel's configuration is fully
>>>> re-programmed on each DMA transaction anyways and because save-restore
>>>> of an active channel won't end up well without pausing transfer prior to
>>>> saving of the state (note that all channels shall be idling at the time of
>>>> suspend, so save-restore is not needed at all).
>>>
>>> I guess if we ever wanted to support SNDRV_PCM_INFO_PAUSE for audio and
>>> support the pause callback, then saving and restoring the channels could
>>> be needed. Right now I believe that it will just terminate_all transfers
>>> for audio on entering suspend. Any value in keeping this?
>>
>> Indeed, looks like [1] pauses DMA during suspend if SNDRV_PCM_INFO_PAUSE
>> is supported.
>>
>> [1]
>> https://elixir.bootlin.com/linux/v5.5/source/sound/core/pcm_dmaengine.c#L199
>>
>> So we'll need to save-restore context only if DMA is in a paused state
>> during suspend, I'll adjust this patch to do that and will see if
>> enabling SNDRV_PCM_INFO_PAUSE works.
> 
> I started to look at it and found that the .device_pause() hook isn't
> implemented by the driver. So, it's fine to remove the context's
> save-restore for now.
> 
> Jon, what about to keep this patch as-is? Later on I'll take a look at
> implementing the proper pausing functionality and try to cleanup code a
> bit further (remove the free list usage, etc).

Ah, only T114+ supports the per-channel pausing. So, I won't care about
implementing the device_pause(), let's leave it to somebody else :)

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Jon Hunter]

On 30/01/2020 16:11, Dmitry Osipenko wrote:
> 30.01.2020 17:09, Jon Hunter пишет:
>>
>> On 30/01/2020 04:37, Dmitry Osipenko wrote:
>>> It's a bit impractical to enable hardware's clock at the time of DMA
>>> channel's allocation because most of DMA client drivers allocate DMA
>>> channel at the time of the driver's probing, and thus, DMA clock is kept
>>> always-enabled in practice, defeating the whole purpose of runtime PM.
>>>
>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>> ---
>>>  drivers/dma/tegra20-apb-dma.c | 47 ++++++++++++++++++++++++-----------
>>>  1 file changed, 32 insertions(+), 15 deletions(-)
>>>
>>> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
>>> index 22b88ccff05d..0ee28d8e3c96 100644
>>> --- a/drivers/dma/tegra20-apb-dma.c
>>> +++ b/drivers/dma/tegra20-apb-dma.c
>>> @@ -436,6 +436,8 @@ static void tegra_dma_stop(struct tegra_dma_channel *tdc)
>>>  		tdc_write(tdc, TEGRA_APBDMA_CHAN_STATUS, status);
>>>  	}
>>>  	tdc->busy = false;
>>> +
>>> +	pm_runtime_put(tdc->tdma->dev);
>>>  }
>>>  
>>>  static void tegra_dma_start(struct tegra_dma_channel *tdc,
>>> @@ -500,18 +502,25 @@ static void tegra_dma_configure_for_next(struct tegra_dma_channel *tdc,
>>>  	tegra_dma_resume(tdc);
>>>  }
>>>  
>>> -static void tdc_start_head_req(struct tegra_dma_channel *tdc)
>>> +static bool tdc_start_head_req(struct tegra_dma_channel *tdc)
>>>  {
>>>  	struct tegra_dma_sg_req *sg_req;
>>> +	int err;
>>>  
>>>  	if (list_empty(&tdc->pending_sg_req))
>>> -		return;
>>> +		return false;
>>> +
>>> +	err = pm_runtime_get_sync(tdc->tdma->dev);
>>> +	if (WARN_ON_ONCE(err < 0))
>>> +		return false;
>>>  
>>>  	sg_req = list_first_entry(&tdc->pending_sg_req, typeof(*sg_req), node);
>>>  	tegra_dma_start(tdc, sg_req);
>>>  	sg_req->configured = true;
>>>  	sg_req->words_xferred = 0;
>>>  	tdc->busy = true;
>>> +
>>> +	return true;
>>>  }
>>>  
>>>  static void tdc_configure_next_head_desc(struct tegra_dma_channel *tdc)
>>> @@ -615,6 +624,8 @@ static void handle_once_dma_done(struct tegra_dma_channel *tdc,
>>>  	}
>>>  	list_add_tail(&sgreq->node, &tdc->free_sg_req);
>>>  
>>> +	pm_runtime_put(tdc->tdma->dev);
>>> +
>>>  	/* Do not start DMA if it is going to be terminate */
>>>  	if (to_terminate || list_empty(&tdc->pending_sg_req))
>>>  		return;
>>> @@ -730,9 +741,7 @@ static void tegra_dma_issue_pending(struct dma_chan *dc)
>>>  		dev_err(tdc2dev(tdc), "No DMA request\n");
>>>  		goto end;
>>>  	}
>>> -	if (!tdc->busy) {
>>> -		tdc_start_head_req(tdc);
>>> -
>>> +	if (!tdc->busy && tdc_start_head_req(tdc)) {
>>>  		/* Continuous single mode: Configure next req */
>>>  		if (tdc->cyclic) {
>>>  			/*
>>> @@ -775,6 +784,13 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
>>>  	else
>>>  		wcount = status;
>>>  
>>> +	/*
>>> +	 * tegra_dma_stop() will drop the RPM's usage refcount, but
>>> +	 * tegra_dma_resume() touches hardware and thus we should keep
>>> +	 * the DMA clock active while it's needed.
>>> +	 */
>>> +	pm_runtime_get(tdc->tdma->dev);
>>> +
>>
>> Would it work and make it simpler to just enable in the issue_pending
>> and disable in the handle_once_dma_done or terminate_all?
>>
>> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
>> index 3a45079d11ec..86bbb45da93d 100644
>> --- a/drivers/dma/tegra20-apb-dma.c
>> +++ b/drivers/dma/tegra20-apb-dma.c
>> @@ -616,9 +616,14 @@ static void handle_once_dma_done(struct
>> tegra_dma_channel *tdc,
>>         list_add_tail(&sgreq->node, &tdc->free_sg_req);
>>
>>         /* Do not start DMA if it is going to be terminate */
>> -       if (to_terminate || list_empty(&tdc->pending_sg_req))
>> +       if (to_terminate)
>>                 return;
>>
>> +       if (list_empty(&tdc->pending_sg_req)) {
>> +               pm_runtime_put(tdc->tdma->dev);
>> +               return;
>> +       }
>> +
>>         tdc_start_head_req(tdc);
>>  }
>>
>> @@ -729,6 +734,11 @@ static void tegra_dma_issue_pending(struct dma_chan
>> *dc)
>>                 goto end;
>>         }
>>         if (!tdc->busy) {
>> +               if (pm_runtime_get_sync(tdc->tdma->dev) < 0) {
>> +                       dev_err(tdc2dev(tdc), "Failed to enable DMA!\n");
>> +                       goto end;
>> +               }
>> +
>>                 tdc_start_head_req(tdc);
>>
>>                 /* Continuous single mode: Configure next req */
>> @@ -788,6 +798,7 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
>>                                 get_current_xferred_count(tdc, sgreq,
>> wcount);
>>         }
>>         tegra_dma_resume(tdc);
>> +       pm_runtime_put(tdc->tdma->dev);
>>
>>  skip_dma_stop:
>>         tegra_dma_abort_all(tdc);
>>
> 
> The tegra_dma_stop() should put RPM anyways, which is missed in yours
> sample. Please see handle_continuous_head_request().

Yes and that is deliberate. The cyclic transfers the transfers *should*
not stop until terminate_all is called. The tegra_dma_stop in
handle_continuous_head_request() is an error condition and so I am not
sure it is actually necessary to call pm_runtime_put() here.

> I'm also finding the explicit get/put a bit easier to follow in the
> code, don't you think so?

I can see that, but I was thinking that in the case of cyclic transfers,
it should only really be necessary to call the get/put at the beginning
and end. So in my mind there should only be two exit points which are
the ISR handler for SG and terminate_all for SG and cyclic.

Jon

-- 
nvpublic

Re: [PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Jon Hunter]

On 30/01/2020 18:26, Dmitry Osipenko wrote:
> 30.01.2020 21:06, Dmitry Osipenko пишет:
>> 30.01.2020 19:08, Dmitry Osipenko пишет:
>>> 30.01.2020 17:09, Jon Hunter пишет:
>>>>
>>>> On 30/01/2020 04:38, Dmitry Osipenko wrote:
>>>>> It is enough to check whether hardware is busy on suspend and to reset
>>>>> it across of suspend-resume because channel's configuration is fully
>>>>> re-programmed on each DMA transaction anyways and because save-restore
>>>>> of an active channel won't end up well without pausing transfer prior to
>>>>> saving of the state (note that all channels shall be idling at the time of
>>>>> suspend, so save-restore is not needed at all).
>>>>
>>>> I guess if we ever wanted to support SNDRV_PCM_INFO_PAUSE for audio and
>>>> support the pause callback, then saving and restoring the channels could
>>>> be needed. Right now I believe that it will just terminate_all transfers
>>>> for audio on entering suspend. Any value in keeping this?
>>>
>>> Indeed, looks like [1] pauses DMA during suspend if SNDRV_PCM_INFO_PAUSE
>>> is supported.
>>>
>>> [1]
>>> https://elixir.bootlin.com/linux/v5.5/source/sound/core/pcm_dmaengine.c#L199
>>>
>>> So we'll need to save-restore context only if DMA is in a paused state
>>> during suspend, I'll adjust this patch to do that and will see if
>>> enabling SNDRV_PCM_INFO_PAUSE works.
>>
>> I started to look at it and found that the .device_pause() hook isn't
>> implemented by the driver. So, it's fine to remove the context's
>> save-restore for now.
>>
>> Jon, what about to keep this patch as-is? Later on I'll take a look at
>> implementing the proper pausing functionality and try to cleanup code a
>> bit further (remove the free list usage, etc).
> 
> Ah, only T114+ supports the per-channel pausing. So, I won't care about
> implementing the device_pause(), let's leave it to somebody else :)

That's fine. We have lived with out it for this long.

Jon

-- 
nvpublic

Re: [PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Jon Hunter]

On 30/01/2020 04:38, Dmitry Osipenko wrote:
> It is enough to check whether hardware is busy on suspend and to reset
> it across of suspend-resume because channel's configuration is fully
> re-programmed on each DMA transaction anyways and because save-restore
> of an active channel won't end up well without pausing transfer prior to
> saving of the state (note that all channels shall be idling at the time of
> suspend, so save-restore is not needed at all).

Maybe just update the commit message to state that the pause callback is
not implemented and channel pause is not supported prior to T114. And
then ...

Acked-by: Jon Hunter <jonathanh@nvidia.com>

Cheers
Jon

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Dmitry Osipenko]

30.01.2020 21:45, Jon Hunter пишет:
> 
> On 30/01/2020 16:11, Dmitry Osipenko wrote:
>> 30.01.2020 17:09, Jon Hunter пишет:
>>>
>>> On 30/01/2020 04:37, Dmitry Osipenko wrote:
>>>> It's a bit impractical to enable hardware's clock at the time of DMA
>>>> channel's allocation because most of DMA client drivers allocate DMA
>>>> channel at the time of the driver's probing, and thus, DMA clock is kept
>>>> always-enabled in practice, defeating the whole purpose of runtime PM.
>>>>
>>>> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
>>>> ---
>>>>  drivers/dma/tegra20-apb-dma.c | 47 ++++++++++++++++++++++++-----------
>>>>  1 file changed, 32 insertions(+), 15 deletions(-)
>>>>
>>>> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
>>>> index 22b88ccff05d..0ee28d8e3c96 100644
>>>> --- a/drivers/dma/tegra20-apb-dma.c
>>>> +++ b/drivers/dma/tegra20-apb-dma.c
>>>> @@ -436,6 +436,8 @@ static void tegra_dma_stop(struct tegra_dma_channel *tdc)
>>>>  		tdc_write(tdc, TEGRA_APBDMA_CHAN_STATUS, status);
>>>>  	}
>>>>  	tdc->busy = false;
>>>> +
>>>> +	pm_runtime_put(tdc->tdma->dev);
>>>>  }
>>>>  
>>>>  static void tegra_dma_start(struct tegra_dma_channel *tdc,
>>>> @@ -500,18 +502,25 @@ static void tegra_dma_configure_for_next(struct tegra_dma_channel *tdc,
>>>>  	tegra_dma_resume(tdc);
>>>>  }
>>>>  
>>>> -static void tdc_start_head_req(struct tegra_dma_channel *tdc)
>>>> +static bool tdc_start_head_req(struct tegra_dma_channel *tdc)
>>>>  {
>>>>  	struct tegra_dma_sg_req *sg_req;
>>>> +	int err;
>>>>  
>>>>  	if (list_empty(&tdc->pending_sg_req))
>>>> -		return;
>>>> +		return false;
>>>> +
>>>> +	err = pm_runtime_get_sync(tdc->tdma->dev);
>>>> +	if (WARN_ON_ONCE(err < 0))
>>>> +		return false;
>>>>  
>>>>  	sg_req = list_first_entry(&tdc->pending_sg_req, typeof(*sg_req), node);
>>>>  	tegra_dma_start(tdc, sg_req);
>>>>  	sg_req->configured = true;
>>>>  	sg_req->words_xferred = 0;
>>>>  	tdc->busy = true;
>>>> +
>>>> +	return true;
>>>>  }
>>>>  
>>>>  static void tdc_configure_next_head_desc(struct tegra_dma_channel *tdc)
>>>> @@ -615,6 +624,8 @@ static void handle_once_dma_done(struct tegra_dma_channel *tdc,
>>>>  	}
>>>>  	list_add_tail(&sgreq->node, &tdc->free_sg_req);
>>>>  
>>>> +	pm_runtime_put(tdc->tdma->dev);
>>>> +
>>>>  	/* Do not start DMA if it is going to be terminate */
>>>>  	if (to_terminate || list_empty(&tdc->pending_sg_req))
>>>>  		return;
>>>> @@ -730,9 +741,7 @@ static void tegra_dma_issue_pending(struct dma_chan *dc)
>>>>  		dev_err(tdc2dev(tdc), "No DMA request\n");
>>>>  		goto end;
>>>>  	}
>>>> -	if (!tdc->busy) {
>>>> -		tdc_start_head_req(tdc);
>>>> -
>>>> +	if (!tdc->busy && tdc_start_head_req(tdc)) {
>>>>  		/* Continuous single mode: Configure next req */
>>>>  		if (tdc->cyclic) {
>>>>  			/*
>>>> @@ -775,6 +784,13 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
>>>>  	else
>>>>  		wcount = status;
>>>>  
>>>> +	/*
>>>> +	 * tegra_dma_stop() will drop the RPM's usage refcount, but
>>>> +	 * tegra_dma_resume() touches hardware and thus we should keep
>>>> +	 * the DMA clock active while it's needed.
>>>> +	 */
>>>> +	pm_runtime_get(tdc->tdma->dev);
>>>> +
>>>
>>> Would it work and make it simpler to just enable in the issue_pending
>>> and disable in the handle_once_dma_done or terminate_all?
>>>
>>> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
>>> index 3a45079d11ec..86bbb45da93d 100644
>>> --- a/drivers/dma/tegra20-apb-dma.c
>>> +++ b/drivers/dma/tegra20-apb-dma.c
>>> @@ -616,9 +616,14 @@ static void handle_once_dma_done(struct
>>> tegra_dma_channel *tdc,
>>>         list_add_tail(&sgreq->node, &tdc->free_sg_req);
>>>
>>>         /* Do not start DMA if it is going to be terminate */
>>> -       if (to_terminate || list_empty(&tdc->pending_sg_req))
>>> +       if (to_terminate)
>>>                 return;
>>>
>>> +       if (list_empty(&tdc->pending_sg_req)) {
>>> +               pm_runtime_put(tdc->tdma->dev);
>>> +               return;
>>> +       }
>>> +
>>>         tdc_start_head_req(tdc);
>>>  }
>>>
>>> @@ -729,6 +734,11 @@ static void tegra_dma_issue_pending(struct dma_chan
>>> *dc)
>>>                 goto end;
>>>         }
>>>         if (!tdc->busy) {
>>> +               if (pm_runtime_get_sync(tdc->tdma->dev) < 0) {
>>> +                       dev_err(tdc2dev(tdc), "Failed to enable DMA!\n");
>>> +                       goto end;
>>> +               }
>>> +
>>>                 tdc_start_head_req(tdc);
>>>
>>>                 /* Continuous single mode: Configure next req */
>>> @@ -788,6 +798,7 @@ static int tegra_dma_terminate_all(struct dma_chan *dc)
>>>                                 get_current_xferred_count(tdc, sgreq,
>>> wcount);
>>>         }
>>>         tegra_dma_resume(tdc);
>>> +       pm_runtime_put(tdc->tdma->dev);
>>>
>>>  skip_dma_stop:
>>>         tegra_dma_abort_all(tdc);
>>>
>>
>> The tegra_dma_stop() should put RPM anyways, which is missed in yours
>> sample. Please see handle_continuous_head_request().
> 
> Yes and that is deliberate. The cyclic transfers the transfers *should*
> not stop until terminate_all is called. The tegra_dma_stop in
> handle_continuous_head_request() is an error condition and so I am not
> sure it is actually necessary to call pm_runtime_put() here.

But then tegra_dma_stop() shouldn't unset the "busy" mark.

>> I'm also finding the explicit get/put a bit easier to follow in the
>> code, don't you think so?
> 
> I can see that, but I was thinking that in the case of cyclic transfers,
> it should only really be necessary to call the get/put at the beginning
> and end. So in my mind there should only be two exit points which are
> the ISR handler for SG and terminate_all for SG and cyclic.

Alright, I'll update this patch.

Re: [PATCH v6 12/16] dmaengine: tegra-apb: Clean up suspend-resume

[Dmitry Osipenko]

30.01.2020 22:00, Jon Hunter пишет:
> 
> On 30/01/2020 04:38, Dmitry Osipenko wrote:
>> It is enough to check whether hardware is busy on suspend and to reset
>> it across of suspend-resume because channel's configuration is fully
>> re-programmed on each DMA transaction anyways and because save-restore
>> of an active channel won't end up well without pausing transfer prior to
>> saving of the state (note that all channels shall be idling at the time of
>> suspend, so save-restore is not needed at all).
> 
> Maybe just update the commit message to state that the pause callback is
> not implemented and channel pause is not supported prior to T114. And
> then ...
> 
> Acked-by: Jon Hunter <jonathanh@nvidia.com>

Thanks, I'll update the message.

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Jon Hunter]

On 30/01/2020 20:04, Dmitry Osipenko wrote:

...

>>> The tegra_dma_stop() should put RPM anyways, which is missed in yours
>>> sample. Please see handle_continuous_head_request().
>>
>> Yes and that is deliberate. The cyclic transfers the transfers *should*
>> not stop until terminate_all is called. The tegra_dma_stop in
>> handle_continuous_head_request() is an error condition and so I am not
>> sure it is actually necessary to call pm_runtime_put() here.
> 
> But then tegra_dma_stop() shouldn't unset the "busy" mark.

True.

>>> I'm also finding the explicit get/put a bit easier to follow in the
>>> code, don't you think so?
>>
>> I can see that, but I was thinking that in the case of cyclic transfers,
>> it should only really be necessary to call the get/put at the beginning
>> and end. So in my mind there should only be two exit points which are
>> the ISR handler for SG and terminate_all for SG and cyclic.
> 
> Alright, I'll update this patch.

Hmmm ... I am wondering if we should not mess with that and leave how
you have it.

-- 
nvpublic

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Jon Hunter]

On 30/01/2020 04:37, Dmitry Osipenko wrote:
> It's a bit impractical to enable hardware's clock at the time of DMA
> channel's allocation because most of DMA client drivers allocate DMA
> channel at the time of the driver's probing, and thus, DMA clock is kept
> always-enabled in practice, defeating the whole purpose of runtime PM.
> 
> Signed-off-by: Dmitry Osipenko <digetx@gmail.com>
> ---
>  drivers/dma/tegra20-apb-dma.c | 47 ++++++++++++++++++++++++-----------
>  1 file changed, 32 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/dma/tegra20-apb-dma.c b/drivers/dma/tegra20-apb-dma.c
> index 22b88ccff05d..0ee28d8e3c96 100644
> --- a/drivers/dma/tegra20-apb-dma.c
> +++ b/drivers/dma/tegra20-apb-dma.c
> @@ -436,6 +436,8 @@ static void tegra_dma_stop(struct tegra_dma_channel *tdc)
>  		tdc_write(tdc, TEGRA_APBDMA_CHAN_STATUS, status);
>  	}
>  	tdc->busy = false;
> +
> +	pm_runtime_put(tdc->tdma->dev);

There are only 3 places where tegra_dma_stop is called, does it simplify
the code if we move the pm_runtime_put() outside of tegra_dma_stop? In
other words, everywhere there is a tegra_dma_stop, afterwards we then
call pm_runtime_put?

This would allow us to get rid of the extra pm_runtime_get in
terminate_all.

Jon

-- 
nvpublic

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Dmitry Osipenko]

31.01.2020 12:02, Jon Hunter пишет:
> 
> On 30/01/2020 20:04, Dmitry Osipenko wrote:
> 
> ...
> 
>>>> The tegra_dma_stop() should put RPM anyways, which is missed in yours
>>>> sample. Please see handle_continuous_head_request().
>>>
>>> Yes and that is deliberate. The cyclic transfers the transfers *should*
>>> not stop until terminate_all is called. The tegra_dma_stop in
>>> handle_continuous_head_request() is an error condition and so I am not
>>> sure it is actually necessary to call pm_runtime_put() here.
>>
>> But then tegra_dma_stop() shouldn't unset the "busy" mark.
> 
> True.
> 
>>>> I'm also finding the explicit get/put a bit easier to follow in the
>>>> code, don't you think so?
>>>
>>> I can see that, but I was thinking that in the case of cyclic transfers,
>>> it should only really be necessary to call the get/put at the beginning
>>> and end. So in my mind there should only be two exit points which are
>>> the ISR handler for SG and terminate_all for SG and cyclic.
>>
>> Alright, I'll update this patch.
> 
> Hmmm ... I am wondering if we should not mess with that and leave how
> you have it.

I took another look and seems my current v6 should be more correct because:

1. If "busy" is unset in tegra_dma_stop(), then the RPM should be put
there since tegra_dma_terminate_all() won't put RPM in this case:

	if (!tdc->busy)
		goto skip_dma_stop;

2. We can't move the "busy" unsetting into the terminate because then
tegra_dma_stop() will be invoked twice. Although, one option could be to
remove the tegra_dma_stop() from the error paths of
handle_continuous_head_request(), but I'm not sure that this is correct
to do.

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Dmitry Osipenko]

31.01.2020 17:22, Dmitry Osipenko пишет:
> 31.01.2020 12:02, Jon Hunter пишет:
>>
>> On 30/01/2020 20:04, Dmitry Osipenko wrote:
>>
>> ...
>>
>>>>> The tegra_dma_stop() should put RPM anyways, which is missed in yours
>>>>> sample. Please see handle_continuous_head_request().
>>>>
>>>> Yes and that is deliberate. The cyclic transfers the transfers *should*
>>>> not stop until terminate_all is called. The tegra_dma_stop in
>>>> handle_continuous_head_request() is an error condition and so I am not
>>>> sure it is actually necessary to call pm_runtime_put() here.
>>>
>>> But then tegra_dma_stop() shouldn't unset the "busy" mark.
>>
>> True.
>>
>>>>> I'm also finding the explicit get/put a bit easier to follow in the
>>>>> code, don't you think so?
>>>>
>>>> I can see that, but I was thinking that in the case of cyclic transfers,
>>>> it should only really be necessary to call the get/put at the beginning
>>>> and end. So in my mind there should only be two exit points which are
>>>> the ISR handler for SG and terminate_all for SG and cyclic.
>>>
>>> Alright, I'll update this patch.
>>
>> Hmmm ... I am wondering if we should not mess with that and leave how
>> you have it.
> 
> I took another look and seems my current v6 should be more correct because:
> 
> 1. If "busy" is unset in tegra_dma_stop(), then the RPM should be put
> there since tegra_dma_terminate_all() won't put RPM in this case:
> 
> 	if (!tdc->busy)
> 		goto skip_dma_stop;
> 
> 2. We can't move the "busy" unsetting into the terminate because then
> tegra_dma_stop() will be invoked twice. Although, one option could be to
> remove the tegra_dma_stop() from the error paths of
> handle_continuous_head_request(), but I'm not sure that this is correct
> to do.

Jon, I realized that my v6 variant is wrong too because
tegra_dma_terminate_all() -> tdc->isr_handler() will put RPM, and thus,
the RPM enable-count will be wrecked in this case.

I'm now leaning to adopt yours variant and simply remove the
tegra_dma_stop() from handle_continuous_head_request() because there
shouldn't be any harm in keeping DMA active in the case of error
condition. Besides, these error conditions are very extreme cases that
should never happen in practice.

The "list_empty(&tdc->pending_sg_req)" error seems couldn't ever happen
at all, I'll remove it in v7.

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Jon Hunter]

On 01/02/2020 15:13, Dmitry Osipenko wrote:
> 31.01.2020 17:22, Dmitry Osipenko пишет:
>> 31.01.2020 12:02, Jon Hunter пишет:
>>>
>>> On 30/01/2020 20:04, Dmitry Osipenko wrote:
>>>
>>> ...
>>>
>>>>>> The tegra_dma_stop() should put RPM anyways, which is missed in yours
>>>>>> sample. Please see handle_continuous_head_request().
>>>>>
>>>>> Yes and that is deliberate. The cyclic transfers the transfers *should*
>>>>> not stop until terminate_all is called. The tegra_dma_stop in
>>>>> handle_continuous_head_request() is an error condition and so I am not
>>>>> sure it is actually necessary to call pm_runtime_put() here.
>>>>
>>>> But then tegra_dma_stop() shouldn't unset the "busy" mark.
>>>
>>> True.
>>>
>>>>>> I'm also finding the explicit get/put a bit easier to follow in the
>>>>>> code, don't you think so?
>>>>>
>>>>> I can see that, but I was thinking that in the case of cyclic transfers,
>>>>> it should only really be necessary to call the get/put at the beginning
>>>>> and end. So in my mind there should only be two exit points which are
>>>>> the ISR handler for SG and terminate_all for SG and cyclic.
>>>>
>>>> Alright, I'll update this patch.
>>>
>>> Hmmm ... I am wondering if we should not mess with that and leave how
>>> you have it.
>>
>> I took another look and seems my current v6 should be more correct because:
>>
>> 1. If "busy" is unset in tegra_dma_stop(), then the RPM should be put
>> there since tegra_dma_terminate_all() won't put RPM in this case:
>>
>> 	if (!tdc->busy)
>> 		goto skip_dma_stop;
>>
>> 2. We can't move the "busy" unsetting into the terminate because then
>> tegra_dma_stop() will be invoked twice. Although, one option could be to
>> remove the tegra_dma_stop() from the error paths of
>> handle_continuous_head_request(), but I'm not sure that this is correct
>> to do.
> 
> Jon, I realized that my v6 variant is wrong too because
> tegra_dma_terminate_all() -> tdc->isr_handler() will put RPM, and thus,
> the RPM enable-count will be wrecked in this case.

Did you see my other suggestion to move the pm_runtime_put() outside of
tegra_dma_stop? There are only a few call sites for tegra_dma_stop and
so if we call pm_runtime_put() after calling tegra_dma_stop this should
simplify matters.

Jon

-- 
nvpublic

Re: [PATCH v6 11/16] dmaengine: tegra-apb: Keep clock enabled only during of DMA transfer

[Dmitry Osipenko]

03.02.2020 14:37, Jon Hunter пишет:
> 
> On 01/02/2020 15:13, Dmitry Osipenko wrote:
>> 31.01.2020 17:22, Dmitry Osipenko пишет:
>>> 31.01.2020 12:02, Jon Hunter пишет:
>>>>
>>>> On 30/01/2020 20:04, Dmitry Osipenko wrote:
>>>>
>>>> ...
>>>>
>>>>>>> The tegra_dma_stop() should put RPM anyways, which is missed in yours
>>>>>>> sample. Please see handle_continuous_head_request().
>>>>>>
>>>>>> Yes and that is deliberate. The cyclic transfers the transfers *should*
>>>>>> not stop until terminate_all is called. The tegra_dma_stop in
>>>>>> handle_continuous_head_request() is an error condition and so I am not
>>>>>> sure it is actually necessary to call pm_runtime_put() here.
>>>>>
>>>>> But then tegra_dma_stop() shouldn't unset the "busy" mark.
>>>>
>>>> True.
>>>>
>>>>>>> I'm also finding the explicit get/put a bit easier to follow in the
>>>>>>> code, don't you think so?
>>>>>>
>>>>>> I can see that, but I was thinking that in the case of cyclic transfers,
>>>>>> it should only really be necessary to call the get/put at the beginning
>>>>>> and end. So in my mind there should only be two exit points which are
>>>>>> the ISR handler for SG and terminate_all for SG and cyclic.
>>>>>
>>>>> Alright, I'll update this patch.
>>>>
>>>> Hmmm ... I am wondering if we should not mess with that and leave how
>>>> you have it.
>>>
>>> I took another look and seems my current v6 should be more correct because:
>>>
>>> 1. If "busy" is unset in tegra_dma_stop(), then the RPM should be put
>>> there since tegra_dma_terminate_all() won't put RPM in this case:
>>>
>>> 	if (!tdc->busy)
>>> 		goto skip_dma_stop;
>>>
>>> 2. We can't move the "busy" unsetting into the terminate because then
>>> tegra_dma_stop() will be invoked twice. Although, one option could be to
>>> remove the tegra_dma_stop() from the error paths of
>>> handle_continuous_head_request(), but I'm not sure that this is correct
>>> to do.
>>
>> Jon, I realized that my v6 variant is wrong too because
>> tegra_dma_terminate_all() -> tdc->isr_handler() will put RPM, and thus,
>> the RPM enable-count will be wrecked in this case.
> 
> Did you see my other suggestion to move the pm_runtime_put() outside of
> tegra_dma_stop?

Yes, but seems I skimmed too quickly through the lines and failed to
recognize the point you made.

> There are only a few call sites for tegra_dma_stop and
> so if we call pm_runtime_put() after calling tegra_dma_stop this should
> simplify matters.

This is somewhat similar to what I made in the v7. Instead of adding
pm_runtime_put() after each tegra_dma_stop(), I removed the
tegra_dma_stop().

Looking at it once again, perhaps indeed it will be better to leave the
relevant tegra_dma_stop() in place (the irrelevant could be removed).

Please take a look at the v7, I'll drop the "[PATCH v7 13/19] dmaengine:
tegra-apb: Don't stop cyclic DMA in a case of error condition" and make
v8 after yours review of the v7. Thanks in advance!