[PATCH 2/2] dmaengine: dw-axi-dmac: add reset control support

Artem Shimko posted 2 patches 3 months, 4 weeks ago
There is a newer version of this series
[PATCH 2/2] dmaengine: dw-axi-dmac: add reset control support
Posted by Artem Shimko 3 months, 4 weeks ago
Add proper reset control handling to the AXI DMA driver to ensure
reliable initialization and power management. The driver now manages
resets during probe, remove, and system suspend/resume operations.

The implementation stores reset control in the chip structure and adds
reset assert/deassert calls at the appropriate points: resets are
deasserted during probe after clock acquisition, asserted during remove
and error cleanup, and properly managed during suspend/resume cycles.
Additionally, proper error handling is implemented for reset control
operations to ensure robust behavior.

This ensures the controller is properly reset during power transitions
and prevents potential issues with incomplete initialization.

Signed-off-by: Artem Shimko <a.shimko.dev@gmail.com>
---
 .../dma/dw-axi-dmac/dw-axi-dmac-platform.c    | 41 ++++++++++++-------
 drivers/dma/dw-axi-dmac/dw-axi-dmac.h         |  2 +
 2 files changed, 28 insertions(+), 15 deletions(-)

diff --git a/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c b/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
index 8b7cf3baf5d3..3f4dd2178498 100644
--- a/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
+++ b/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
@@ -1321,6 +1321,9 @@ static int axi_dma_suspend(struct device *dev)
 	axi_dma_irq_disable(chip);
 	axi_dma_disable(chip);
 
+	if (chip->has_resets)
+		reset_control_assert(chip->resets);
+
 	clk_disable_unprepare(chip->core_clk);
 	clk_disable_unprepare(chip->cfgr_clk);
 
@@ -1340,6 +1343,9 @@ static int axi_dma_resume(struct device *dev)
 	if (ret < 0)
 		return ret;
 
+	if (chip->has_resets)
+		reset_control_deassert(chip->resets);
+
 	axi_dma_enable(chip);
 	axi_dma_irq_enable(chip);
 
@@ -1455,7 +1461,6 @@ static int dw_probe(struct platform_device *pdev)
 	struct axi_dma_chip *chip;
 	struct dw_axi_dma *dw;
 	struct dw_axi_dma_hcfg *hdata;
-	struct reset_control *resets;
 	unsigned int flags;
 	u32 i;
 	int ret;
@@ -1487,16 +1492,6 @@ static int dw_probe(struct platform_device *pdev)
 			return PTR_ERR(chip->apb_regs);
 	}
 
-	if (flags & AXI_DMA_FLAG_HAS_RESETS) {
-		resets = devm_reset_control_array_get_exclusive(&pdev->dev);
-		if (IS_ERR(resets))
-			return PTR_ERR(resets);
-
-		ret = reset_control_deassert(resets);
-		if (ret)
-			return ret;
-	}
-
 	chip->dw->hdata->use_cfg2 = !!(flags & AXI_DMA_FLAG_USE_CFG2);
 
 	chip->core_clk = devm_clk_get(chip->dev, "core-clk");
@@ -1507,18 +1502,31 @@ static int dw_probe(struct platform_device *pdev)
 	if (IS_ERR(chip->cfgr_clk))
 		return PTR_ERR(chip->cfgr_clk);
 
+	chip->has_resets = !!(flags & AXI_DMA_FLAG_HAS_RESETS);
+	if (chip->has_resets) {
+		chip->resets = devm_reset_control_array_get_exclusive(&pdev->dev);
+		if (IS_ERR(chip->resets))
+			return PTR_ERR(chip->resets);
+
+		ret = reset_control_deassert(chip->resets);
+		if (ret)
+			return dev_err_probe(&pdev->dev, ret, "Failed to deassert resets\n");
+	}
+
 	ret = parse_device_properties(chip);
 	if (ret)
-		return ret;
+		goto err_exit;
 
 	dw->chan = devm_kcalloc(chip->dev, hdata->nr_channels,
 				sizeof(*dw->chan), GFP_KERNEL);
-	if (!dw->chan)
-		return -ENOMEM;
+	if (!dw->chan) {
+		ret = -ENOMEM;
+		goto err_exit;
+	}
 
 	ret = axi_req_irqs(pdev, chip);
 	if (ret)
-		return ret;
+		goto err_exit;
 
 	INIT_LIST_HEAD(&dw->dma.channels);
 	for (i = 0; i < hdata->nr_channels; i++) {
@@ -1605,6 +1613,9 @@ static int dw_probe(struct platform_device *pdev)
 
 err_pm_disable:
 	pm_runtime_disable(chip->dev);
+err_exit:
+	if (chip->has_resets)
+		reset_control_assert(chip->resets);
 
 	return ret;
 }
diff --git a/drivers/dma/dw-axi-dmac/dw-axi-dmac.h b/drivers/dma/dw-axi-dmac/dw-axi-dmac.h
index b842e6a8d90d..56dc3d75fe92 100644
--- a/drivers/dma/dw-axi-dmac/dw-axi-dmac.h
+++ b/drivers/dma/dw-axi-dmac/dw-axi-dmac.h
@@ -71,6 +71,8 @@ struct axi_dma_chip {
 	struct clk		*core_clk;
 	struct clk		*cfgr_clk;
 	struct dw_axi_dma	*dw;
+	struct reset_control	*resets;
+	bool			has_resets;
 };
 
 /* LLI == Linked List Item */
-- 
2.43.0
Re: [PATCH 2/2] dmaengine: dw-axi-dmac: add reset control support
Posted by Philipp Zabel 3 months, 3 weeks ago
On So, 2025-10-12 at 13:00 +0300, Artem Shimko wrote:
> Add proper reset control handling to the AXI DMA driver to ensure
> reliable initialization and power management. The driver now manages
> resets during probe, remove, and system suspend/resume operations.
> 
> The implementation stores reset control in the chip structure and adds
> reset assert/deassert calls at the appropriate points: resets are
> deasserted during probe after clock acquisition, asserted during remove
> and error cleanup, and properly managed during suspend/resume cycles.
> Additionally, proper error handling is implemented for reset control
> operations to ensure robust behavior.
> 
> This ensures the controller is properly reset during power transitions
> and prevents potential issues with incomplete initialization.
> 
> Signed-off-by: Artem Shimko <a.shimko.dev@gmail.com>
> ---
>  .../dma/dw-axi-dmac/dw-axi-dmac-platform.c    | 41 ++++++++++++-------
>  drivers/dma/dw-axi-dmac/dw-axi-dmac.h         |  2 +
>  2 files changed, 28 insertions(+), 15 deletions(-)
> 
> diff --git a/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c b/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
> index 8b7cf3baf5d3..3f4dd2178498 100644
> --- a/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
> +++ b/drivers/dma/dw-axi-dmac/dw-axi-dmac-platform.c
> @@ -1321,6 +1321,9 @@ static int axi_dma_suspend(struct device *dev)
>  	axi_dma_irq_disable(chip);
>  	axi_dma_disable(chip);
>  
> +	if (chip->has_resets)
> +		reset_control_assert(chip->resets);

reset_control_assert/deassert() handle NULL pointers, so you could drop
the chip->has_resets flag and just

	reset_control_assert(chip->resets);

unconditionally.

> +
>  	clk_disable_unprepare(chip->core_clk);
>  	clk_disable_unprepare(chip->cfgr_clk);
>  
> @@ -1340,6 +1343,9 @@ static int axi_dma_resume(struct device *dev)
>  	if (ret < 0)
>  		return ret;
>  
> +	if (chip->has_resets)
> +		reset_control_deassert(chip->resets);
> +

Same as above.

>  	axi_dma_enable(chip);
>  	axi_dma_irq_enable(chip);
>  
> @@ -1455,7 +1461,6 @@ static int dw_probe(struct platform_device *pdev)
>  	struct axi_dma_chip *chip;
>  	struct dw_axi_dma *dw;
>  	struct dw_axi_dma_hcfg *hdata;
> -	struct reset_control *resets;
>  	unsigned int flags;
>  	u32 i;
>  	int ret;
> @@ -1487,16 +1492,6 @@ static int dw_probe(struct platform_device *pdev)
>  			return PTR_ERR(chip->apb_regs);
>  	}
>  
> -	if (flags & AXI_DMA_FLAG_HAS_RESETS) {
> -		resets = devm_reset_control_array_get_exclusive(&pdev->dev);
> -		if (IS_ERR(resets))
> -			return PTR_ERR(resets);
> -
> -		ret = reset_control_deassert(resets);
> -		if (ret)
> -			return ret;
> -	}
> -
>  	chip->dw->hdata->use_cfg2 = !!(flags & AXI_DMA_FLAG_USE_CFG2);
>  
>  	chip->core_clk = devm_clk_get(chip->dev, "core-clk");
> @@ -1507,18 +1502,31 @@ static int dw_probe(struct platform_device *pdev)
>  	if (IS_ERR(chip->cfgr_clk))
>  		return PTR_ERR(chip->cfgr_clk);
>  
> +	chip->has_resets = !!(flags & AXI_DMA_FLAG_HAS_RESETS);
> +	if (chip->has_resets) {
> +		chip->resets = devm_reset_control_array_get_exclusive(&pdev->dev);
> +		if (IS_ERR(chip->resets))
> +			return PTR_ERR(chip->resets);
> +
> +		ret = reset_control_deassert(chip->resets);
> +		if (ret)
> +			return dev_err_probe(&pdev->dev, ret, "Failed to deassert resets\n");
> +	}
> +

Why is this moved down here?

>  	ret = parse_device_properties(chip);
>  	if (ret)
> -		return ret;
> +		goto err_exit;
>  
>  	dw->chan = devm_kcalloc(chip->dev, hdata->nr_channels,
>  				sizeof(*dw->chan), GFP_KERNEL);
> -	if (!dw->chan)
> -		return -ENOMEM;
> +	if (!dw->chan) {
> +		ret = -ENOMEM;
> +		goto err_exit;
> +	}
>  
>  	ret = axi_req_irqs(pdev, chip);
>  	if (ret)
> -		return ret;
> +		goto err_exit;
>  
>  	INIT_LIST_HEAD(&dw->dma.channels);
>  	for (i = 0; i < hdata->nr_channels; i++) {
> @@ -1605,6 +1613,9 @@ static int dw_probe(struct platform_device *pdev)
>  
>  err_pm_disable:
>  	pm_runtime_disable(chip->dev);
> +err_exit:
> +	if (chip->has_resets)
> +		reset_control_assert(chip->resets);

If it is ok to keep the module in reset, shouldn't the reset control be
asserted on device remove() as well?

regards
Philipp
Re: [PATCH 2/2] dmaengine: dw-axi-dmac: add reset control support
Posted by Artem Shimko 3 months, 3 weeks ago
Hi Philipp,

On Mon, Oct 13, 2025 at 11:41 AM Philipp Zabel <p.zabel@pengutronix.de> wrote:
> reset_control_assert/deassert() handle NULL pointers, so you could drop
> the chip->has_resets flag and just
>
>         reset_control_assert(chip->resets);
>
> unconditionally.

Thanks, I'll fix that

> Why is this moved down here?

Reset operations typically require clock signals to be available. By
moving reset after clock
acquisition (devm_clk_get), we ensure that the clock is ready to
operate when reset is performed.

> If it is ok to keep the module in reset, shouldn't the reset control be
> asserted on device remove() as well?

dw_remove() has axi_dma_suspend() function, which is where the reset assertion
occurs via reset_control_assert().

Best regards,
Artem Shimko
Re: [PATCH 2/2] dmaengine: dw-axi-dmac: add reset control support
Posted by Philipp Zabel 3 months, 3 weeks ago
On Mo, 2025-10-13 at 17:56 +0300, Artem Shimko wrote:

[...]
> dw_remove() has axi_dma_suspend() function, which is where the reset assertion
> occurs via reset_control_assert().

It looks to me like dw_remove() is now missing a deassert before
accessing registers, in case the device is removed while runtime
suspended.

regards
Philipp
Re: [PATCH 2/2] dmaengine: dw-axi-dmac: add reset control support
Posted by Artem Shimko 3 months, 3 weeks ago
Hi Philipp,

On Mon, Oct 13, 2025 at 6:41 PM Philipp Zabel <p.zabel@pengutronix.de> wrote:
>
> On Mo, 2025-10-13 at 17:56 +0300, Artem Shimko wrote:
>
> [...]
> > dw_remove() has axi_dma_suspend() function, which is where the reset assertion
> > occurs via reset_control_assert().
>
> It looks to me like dw_remove() is now missing a deassert before
> accessing registers, in case the device is removed while runtime
> suspended.
>
> regards
> Philipp

oh, I see, it looks like we have to insert reset_control_deassert(chip->resets);
just after clk_prepare_enable(chip->cfgr_clk);...

You're absolutely right!

Just realized I may have been too quick with v2. Will wait for more
feedback and send v3.

Best regards,
Artem Shimko