Re: [PATCH v5 4/4] dmaengine: zynqmp_dma: Add per-channel reset support

From: Golla, Nagendra

Date: Thu Nov 19 2026 - 06:54:48 EST


Hi Frank,

On 11/18/2026 10:45 PM, Frank Li wrote:
On Wed, Nov 18, 2026 at 11:54:30AM +0530, Golla, Nagendra wrote:
Hi Frank,

On 11/17/2026 2:06 AM, Frank Li wrote:
On Fri, Nov 06, 2026 at 03:40:26PM +0530, Golla Nagendra wrote:
Versal Gen2 and Versal Net expose a dedicated reset line per ZDMA
channel. Trigger a reset pulse during probe using device_reset().

For Versal Net and Versal Gen2, reset is a required DT property, so
probe returns an error if the reset property is missing. For legacy
SoCs, the reset path is skipped.

Signed-off-by: Golla Nagendra <nagendra.golla@xxxxxxx>
---
drivers/dma/xilinx/zynqmp_dma.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)

diff --git a/drivers/dma/xilinx/zynqmp_dma.c b/drivers/dma/xilinx/zynqmp_dma.c
index cb0eb13f9b05..12696fa44c66 100644
--- a/drivers/dma/xilinx/zynqmp_dma.c
+++ b/drivers/dma/xilinx/zynqmp_dma.c
@@ -19,6 +19,7 @@
#include <linux/clk.h>
#include <linux/io-64-nonatomic-lo-hi.h>
#include <linux/pm_runtime.h>
+#include <linux/reset.h>

#include "../dmaengine.h"

@@ -937,6 +938,14 @@ static int zynqmp_dma_chan_probe(struct zynqmp_dma_device *zdev,
if (IS_ERR(chan->regs))
return PTR_ERR(chan->regs);

+ match_data = of_device_get_match_data(&pdev->dev);
+ if (match_data && match_data->has_reset) {
+ err = device_reset(&pdev->dev);
+ if (err)
+ return dev_err_probe(&pdev->dev, err,
+ "failed to reset channel\n");
+ }
+

Use device_reset_optional() will simple code. Your binding already verify
dts, needn't check again.

Thanks for the review

The binding schema restricts the resets property to Versal Gen2 and Versal
Net, where it is required.

An unconditional device_reset_optional() does not enforce this required
property for Versal Gen2 and Versal Net devices in alignment with the DT
binding schema. If the required reset is missing, device_reset_optional()
would allow the driver to continue probing.

Yes, but I suppose function doesn't work without reset at dts, which is
same as that driver is not probed. And new dts require pass CHECK_DTBS.

Thanks for the response

However, if a DTB without the reset property is used, device_reset_optional() would allow probe to succeed. Without the reset, we cannot ensure the channel is in a known state, and a subsequent DMA transfer could fail.

Since the reset is required for Versal Gen 2 and Versal Net, we think it would be clearer to detect the missing reset and fail in the probe, rather than allow a possible failure during a later transfer.




Whereas, in the current approach, the match-data check limits reset handling
to Versal Gen2 and Versal Net. Calling device_reset() for those compatibles
enforces the required reset property and ensures that probe fails if the
required reset is missing or cannot be applied, in alignment with the DT
binding schema.

For legacy compatibles, the driver skips reset handling even if an old or
unvalidated DTB contains an unexpected resets property.

Fixes these dts.


Agreed, we are taking care of this as well. The invalid DTB case was an auxiliary point.


Hence, we prefer the current approach. Could you please advise if you see
any concerns?

Similar case is about clk_get_optional(). Only check critical problem, like
out of ranges.

Using optional does check critical problem, when the reset is present. But doesn't enforce the reset as required.

Thanks,
Nagendra

Frank


Thanks,
Nagendra

Frank

chan->bus_width = ZYNQMP_DMA_BUS_WIDTH_64;
chan->dst_burst_len = ZYNQMP_DMA_MAX_DST_BURST_LEN;
chan->src_burst_len = ZYNQMP_DMA_MAX_SRC_BURST_LEN;
@@ -952,7 +961,6 @@ static int zynqmp_dma_chan_probe(struct zynqmp_dma_device *zdev,
return -EINVAL;
}

- match_data = of_device_get_match_data(&pdev->dev);
if (match_data)
chan->irq_offset = match_data->offset;


--
2.43.7