* [PATCH v2 0/3] Add DMA support for i.MX94 and improve AHB read performance
@ 2026-07-21 21:01 han.xu
2026-07-21 21:01 ` [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties han.xu
` (2 more replies)
0 siblings, 3 replies; 6+ messages in thread
From: han.xu @ 2026-07-21 21:01 UTC (permalink / raw)
To: Rob Herring, Krzysztof Kozlowski, Conor Dooley, Frank Li,
Sascha Hauer, Pengutronix Kernel Team, Fabio Estevam, Haibo Chen,
Mark Brown
Cc: devicetree, imx, linux-arm-kernel, linux-kernel, linux-spi,
Han Xu
This series adds DMA support to the NXP XSPI controller on i.MX94
and enables the use of DMA for AHB read operations to improve
performance.
On i.MX943 EVK, the measured read throughput improves 4x higher, from
approximately 70 MB/s to 290 MB/s when DMA is used.
Changes in v2:
- Reorder the patch series.
- Adjust DT binding patch subject lines to match SPI subsystem conventions.
- Update commit message to include ~4x AHB read performance improvement.
- Rename DMA timeout macro to indicate time unit and document timeout selection.
- Simplify DMA channel initialization flow.
- Add CPU fallback path when DMA read fails.
- Clarify/fix DMA alignment handling based on DMA engine requirements.
- Drop unrelated formatting change in probe().
Signed-off-by: Han Xu <han.xu@nxp.com>
---
Han Xu (3):
spi: dt-bindings: nxp,imx94-xspi: add DMA properties
spi: nxp-xspi: Improve AHB read performance with DMA
arm64: dts: imx94-xspi: add the DMA channels
.../devicetree/bindings/spi/nxp,imx94-xspi.yaml | 10 ++
arch/arm64/boot/dts/freescale/imx94.dtsi | 4 +
drivers/spi/spi-nxp-xspi.c | 130 ++++++++++++++++++++-
3 files changed, 141 insertions(+), 3 deletions(-)
---
base-commit: 2c2900017f2486e7b46795e80d7c8f3f8fbe0236
change-id: 20260621-xspi_dma_upstream-324c07d4c116
Best regards,
--
Han Xu <han.xu@nxp.com>
^ permalink raw reply [flat|nested] 6+ messages in thread
* [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties
2026-07-21 21:01 [PATCH v2 0/3] Add DMA support for i.MX94 and improve AHB read performance han.xu
@ 2026-07-21 21:01 ` han.xu
2026-07-21 21:35 ` Frank Li
2026-07-22 16:46 ` Conor Dooley
2026-07-21 21:01 ` [PATCH v2 2/3] spi: nxp-xspi: Improve AHB read performance with DMA han.xu
2026-07-21 21:01 ` [PATCH v2 3/3] arm64: dts: imx94-xspi: add the DMA channels han.xu
2 siblings, 2 replies; 6+ messages in thread
From: han.xu @ 2026-07-21 21:01 UTC (permalink / raw)
To: Rob Herring, Krzysztof Kozlowski, Conor Dooley, Frank Li,
Sascha Hauer, Pengutronix Kernel Team, Fabio Estevam, Haibo Chen,
Mark Brown
Cc: devicetree, imx, linux-arm-kernel, linux-kernel, linux-spi,
Han Xu
From: Han Xu <han.xu@nxp.com>
Add dmas and dma-names to describe TX and RX DMA channels for the i.MX94
XSPI controller.
Signed-off-by: Han Xu <han.xu@nxp.com>
---
Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml | 10 ++++++++++
1 file changed, 10 insertions(+)
diff --git a/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml b/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
index 16a0598c6d033..ccf841f194c06 100644
--- a/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
+++ b/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
@@ -30,6 +30,16 @@ properties:
- const: base
- const: mmap
+ dmas:
+ items:
+ - description: Transmit DMA
+ - description: Receive DMA
+
+ dma-names:
+ items:
+ - const: tx
+ - const: rx
+
interrupts:
items:
- description: interrupt for EENV0
--
2.34.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH v2 2/3] spi: nxp-xspi: Improve AHB read performance with DMA
2026-07-21 21:01 [PATCH v2 0/3] Add DMA support for i.MX94 and improve AHB read performance han.xu
2026-07-21 21:01 ` [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties han.xu
@ 2026-07-21 21:01 ` han.xu
2026-07-21 21:01 ` [PATCH v2 3/3] arm64: dts: imx94-xspi: add the DMA channels han.xu
2 siblings, 0 replies; 6+ messages in thread
From: han.xu @ 2026-07-21 21:01 UTC (permalink / raw)
To: Rob Herring, Krzysztof Kozlowski, Conor Dooley, Frank Li,
Sascha Hauer, Pengutronix Kernel Team, Fabio Estevam, Haibo Chen,
Mark Brown
Cc: devicetree, imx, linux-arm-kernel, linux-kernel, linux-spi,
Han Xu
From: Han Xu <han.xu@nxp.com>
Use eDMA for XSPI AHB read to improve the performance.
Read with CPU
root@imx943evk:~# dd if=/dev/mtd0 of=/dev/null bs=32M count=1
1+0 records in
1+0 records out
33554432 bytes (34 MB, 32 MiB) copied, 0.479719 s, 69.9 MB/s
Read with DMA
root@imx943evk:~# dd if=/dev/mtd0 of=/dev/null bs=32M count=1
1+0 records in
1+0 records out
33554432 bytes (34 MB, 32 MiB) copied, 0.115788 s, 290 MB/s
Signed-off-by: Han Xu <han.xu@nxp.com>
---
drivers/spi/spi-nxp-xspi.c | 130 +++++++++++++++++++++++++++++++++++++++++++--
1 file changed, 127 insertions(+), 3 deletions(-)
diff --git a/drivers/spi/spi-nxp-xspi.c b/drivers/spi/spi-nxp-xspi.c
index 037eac24e6fd2..0658c0557ff17 100644
--- a/drivers/spi/spi-nxp-xspi.c
+++ b/drivers/spi/spi-nxp-xspi.c
@@ -35,6 +35,8 @@
#include <linux/clk.h>
#include <linux/completion.h>
#include <linux/delay.h>
+#include <linux/dmaengine.h>
+#include <linux/dma-mapping.h>
#include <linux/err.h>
#include <linux/errno.h>
#include <linux/interrupt.h>
@@ -303,6 +305,12 @@
(opr)) << (((idx) % 2) * OPRND_SHIFT))
#define NXP_XSPI_MIN_IOMAP SZ_4M
+/*
+ * It takes 0.5s to access the entire area with DMA,
+ * leaving a 10x safety margin
+ */
+#define NXP_XSPI_DMA_TOUT_MS 5000
+#define NXP_XSPI_DMA_TOUT msecs_to_jiffies(NXP_XSPI_DMA_TOUT_MS)
#define NXP_XSPI_MAX_CHIPSELECT 2
#define POLL_TOUT_US 5000
@@ -336,6 +344,9 @@ struct nxp_xspi {
/* mutex lock for each operation */
struct mutex lock;
int selected;
+ int dma_copy_align;
+ struct dma_chan *dma_rx;
+ struct completion dma_rx_c;
#define XSPI_DTR_PROTO BIT(0)
int flags;
/* Save the previous operation clock rate */
@@ -796,6 +807,82 @@ static int nxp_xspi_ahb_read(struct nxp_xspi *xspi, const struct spi_mem_op *op)
return 0;
}
+static int nxp_xspi_dma_init(struct device *dev, struct nxp_xspi *xspi)
+{
+ struct dma_chan *rx;
+
+ rx = dma_request_chan(dev, "rx");
+ if (IS_ERR(rx)) {
+ int ret = PTR_ERR(rx);
+
+ if (ret == -EPROBE_DEFER)
+ return ret;
+ dev_dbg(dev, "NO DMA RX channel, falling back to CPU read\n");
+ xspi->dma_copy_align = 1;
+ return 0;
+ }
+
+ xspi->dma_rx = rx;
+ init_completion(&xspi->dma_rx_c);
+ xspi->dma_copy_align = 1 << rx->device->copy_align;
+
+ return 0;
+}
+
+static void nxp_xspi_dma_rx_callback(void *data)
+{
+ struct nxp_xspi *xspi = data;
+
+ complete(&xspi->dma_rx_c);
+}
+
+static int nxp_xspi_ahb_dma_read(struct nxp_xspi *xspi,
+ const struct spi_mem_op *op)
+{
+ struct dma_async_tx_descriptor *desc;
+ struct dma_chan *chan = xspi->dma_rx;
+ unsigned long timeout;
+ dma_addr_t dma_addr;
+ int ret = 0;
+
+ dma_addr = dma_map_single(chan->device->dev,
+ op->data.buf.in, op->data.nbytes,
+ DMA_FROM_DEVICE);
+
+ if (dma_mapping_error(chan->device->dev, dma_addr)) {
+ dev_err(xspi->dev, "failed to map DMA buffer for AHB read\n");
+ return -ENOMEM;
+ }
+
+ desc = dmaengine_prep_dma_memcpy(chan, dma_addr,
+ xspi->memmap_phy + op->addr.val,
+ op->data.nbytes,
+ DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
+ if (!desc) {
+ dev_err(xspi->dev, "failed to prepare AHB RX DMA descriptor\n");
+ ret = -EIO;
+ goto err_unmap;
+ }
+ desc->callback = nxp_xspi_dma_rx_callback;
+ desc->callback_param = xspi;
+ reinit_completion(&xspi->dma_rx_c);
+ dmaengine_submit(desc);
+ dma_async_issue_pending(chan);
+
+ timeout = wait_for_completion_timeout(&xspi->dma_rx_c, NXP_XSPI_DMA_TOUT);
+ if (!timeout) {
+ dev_err(xspi->dev, "AHB RX DMA timeout\n");
+ dmaengine_terminate_sync(chan);
+ ret = -ETIMEDOUT;
+ }
+
+err_unmap:
+ dma_unmap_single(chan->device->dev, dma_addr,
+ op->data.nbytes, DMA_FROM_DEVICE);
+
+ return ret;
+}
+
static int nxp_xspi_fill_txfifo(struct nxp_xspi *xspi,
const struct spi_mem_op *op)
{
@@ -1016,10 +1103,21 @@ static int nxp_xspi_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
* all use IP write.
*/
if ((op->data.dir == SPI_MEM_DATA_IN) && !needs_ip_only(xspi)
- && ((op->addr.val + op->data.nbytes) <= xspi->memmap_phy_size))
- err = nxp_xspi_ahb_read(xspi, op);
- else
+ && ((op->addr.val + op->data.nbytes) <= xspi->memmap_phy_size)) {
+ /* use DMA for transfers no less than ahb_buf_size, when DMA is available */
+ if (xspi->dma_rx && op->data.nbytes >= xspi->devtype_data->ahb_buf_size) {
+ err = nxp_xspi_ahb_dma_read(xspi, op);
+ if (err) {
+ dev_warn_ratelimited(xspi->dev,
+ "DMA read failed (%d), falling back to CPU read\n", err);
+ err = nxp_xspi_ahb_read(xspi, op);
+ }
+ } else {
+ err = nxp_xspi_ahb_read(xspi, op);
+ }
+ } else {
err = nxp_xspi_do_op(xspi, op);
+ }
nxp_xspi_sw_reset(xspi);
@@ -1029,6 +1127,7 @@ static int nxp_xspi_exec_op(struct spi_mem *mem, const struct spi_mem_op *op)
static int nxp_xspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
{
struct nxp_xspi *xspi = spi_controller_get_devdata(mem->spi->controller);
+ int alignment = xspi->dma_copy_align;
if (op->data.dir == SPI_MEM_DATA_OUT) {
if (op->data.nbytes > xspi->devtype_data->txfifo)
@@ -1042,6 +1141,22 @@ static int nxp_xspi_adjust_op_size(struct spi_mem *mem, struct spi_mem_op *op)
if (!needs_ip_only(xspi) && (op->addr.val < xspi->memmap_phy_size)
&& ((op->addr.val + op->data.nbytes) > xspi->memmap_phy_size))
op->data.nbytes = xspi->memmap_phy_size - op->addr.val;
+
+ /*
+ * For AHB DMA read, align the transfer to the DMA alignment
+ * boundaries. If the start address is unaligned, shorten this
+ * transfer so the next one starts on an aligned boundary.
+ * Otherwise, if the length is unaligned, round it down.
+ */
+ if (xspi->dma_rx && !needs_ip_only(xspi) &&
+ (op->data.nbytes >= xspi->devtype_data->ahb_buf_size)) {
+ if (op->addr.val % alignment)
+ op->data.nbytes =
+ ALIGN(op->addr.val, alignment) - op->addr.val;
+ else if (op->data.nbytes % alignment)
+ op->data.nbytes =
+ ALIGN_DOWN(op->data.nbytes, alignment);
+ }
}
return 0;
@@ -1203,6 +1318,10 @@ static void nxp_xspi_cleanup(void *data)
if (xspi->ahb_addr)
iounmap(xspi->ahb_addr);
+ if (xspi->dma_rx) {
+ dmaengine_terminate_sync(xspi->dma_rx);
+ dma_release_channel(xspi->dma_rx);
+ }
}
static int nxp_xspi_probe(struct platform_device *pdev)
@@ -1252,6 +1371,11 @@ static int nxp_xspi_probe(struct platform_device *pdev)
if (irq < 0)
return dev_err_probe(dev, irq, "Failed to get irq source");
+ /* DMA is optional, failure(other than -EPROBE_DEFER) falls back to CPU */
+ ret = nxp_xspi_dma_init(dev, xspi);
+ if (ret == -EPROBE_DEFER)
+ return ret;
+
pm_runtime_set_autosuspend_delay(dev, XSPI_RPM_TIMEOUT_MS);
pm_runtime_use_autosuspend(dev);
ret = devm_pm_runtime_enable(dev);
--
2.34.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* [PATCH v2 3/3] arm64: dts: imx94-xspi: add the DMA channels
2026-07-21 21:01 [PATCH v2 0/3] Add DMA support for i.MX94 and improve AHB read performance han.xu
2026-07-21 21:01 ` [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties han.xu
2026-07-21 21:01 ` [PATCH v2 2/3] spi: nxp-xspi: Improve AHB read performance with DMA han.xu
@ 2026-07-21 21:01 ` han.xu
2 siblings, 0 replies; 6+ messages in thread
From: han.xu @ 2026-07-21 21:01 UTC (permalink / raw)
To: Rob Herring, Krzysztof Kozlowski, Conor Dooley, Frank Li,
Sascha Hauer, Pengutronix Kernel Team, Fabio Estevam, Haibo Chen,
Mark Brown
Cc: devicetree, imx, linux-arm-kernel, linux-kernel, linux-spi,
Han Xu
From: Han Xu <han.xu@nxp.com>
Add the DMA channels for iMX94 XSPI controller.
Signed-off-by: Han Xu <han.xu@nxp.com>
---
arch/arm64/boot/dts/freescale/imx94.dtsi | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/arch/arm64/boot/dts/freescale/imx94.dtsi b/arch/arm64/boot/dts/freescale/imx94.dtsi
index a6cb5a6e848b3..70c9dbf10fa5e 100644
--- a/arch/arm64/boot/dts/freescale/imx94.dtsi
+++ b/arch/arm64/boot/dts/freescale/imx94.dtsi
@@ -827,6 +827,8 @@ xspi1: spi@42b90000 {
#size-cells = <0>;
clocks = <&scmi_clk IMX94_CLK_XSPI1>;
clock-names = "per";
+ dmas = <&edma2 27 0 0>, <&edma2 28 0 FSL_EDMA_RX>;
+ dma-names = "tx", "rx";
status = "disabled";
};
@@ -843,6 +845,8 @@ xspi2: spi@42be0000 {
#size-cells = <0>;
clocks = <&scmi_clk IMX94_CLK_XSPI2>;
clock-names = "per";
+ dmas = <&edma4 42 0 0>, <&edma4 43 0 FSL_EDMA_RX>;
+ dma-names = "tx", "rx";
status = "disabled";
};
};
--
2.34.1
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties
2026-07-21 21:01 ` [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties han.xu
@ 2026-07-21 21:35 ` Frank Li
2026-07-22 16:46 ` Conor Dooley
1 sibling, 0 replies; 6+ messages in thread
From: Frank Li @ 2026-07-21 21:35 UTC (permalink / raw)
To: han.xu
Cc: Rob Herring, Krzysztof Kozlowski, Conor Dooley, Frank Li,
Sascha Hauer, Pengutronix Kernel Team, Fabio Estevam, Haibo Chen,
Mark Brown, devicetree, imx, linux-arm-kernel, linux-kernel,
linux-spi, Han Xu
On Tue, Jul 21, 2026 at 04:01:42PM -0500, han.xu@oss.nxp.com wrote:
> From: Han Xu <han.xu@nxp.com>
>
> Add dmas and dma-names to describe TX and RX DMA channels for the i.MX94
> XSPI controller.
>
> Signed-off-by: Han Xu <han.xu@nxp.com>
> ---
Reviewed-by: Frank Li <Frank.Li@nxp.com>
> Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml | 10 ++++++++++
> 1 file changed, 10 insertions(+)
>
> diff --git a/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml b/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
> index 16a0598c6d033..ccf841f194c06 100644
> --- a/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
> +++ b/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
> @@ -30,6 +30,16 @@ properties:
> - const: base
> - const: mmap
>
> + dmas:
> + items:
> + - description: Transmit DMA
> + - description: Receive DMA
> +
> + dma-names:
> + items:
> + - const: tx
> + - const: rx
> +
> interrupts:
> items:
> - description: interrupt for EENV0
>
> --
> 2.34.1
>
>
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties
2026-07-21 21:01 ` [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties han.xu
2026-07-21 21:35 ` Frank Li
@ 2026-07-22 16:46 ` Conor Dooley
1 sibling, 0 replies; 6+ messages in thread
From: Conor Dooley @ 2026-07-22 16:46 UTC (permalink / raw)
To: han.xu
Cc: Rob Herring, Krzysztof Kozlowski, Conor Dooley, Frank Li,
Sascha Hauer, Pengutronix Kernel Team, Fabio Estevam, Haibo Chen,
Mark Brown, devicetree, imx, linux-arm-kernel, linux-kernel,
linux-spi, Han Xu
[-- Attachment #1: Type: text/plain, Size: 1146 bytes --]
On Tue, Jul 21, 2026 at 04:01:42PM -0500, han.xu@oss.nxp.com wrote:
> From: Han Xu <han.xu@nxp.com>
>
> Add dmas and dma-names to describe TX and RX DMA channels for the i.MX94
> XSPI controller.
>
> Signed-off-by: Han Xu <han.xu@nxp.com>
Acked-by: Conor Dooley <conor.dooley@microchip.com>
pw-bot: not-applicable
> ---
> Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml | 10 ++++++++++
> 1 file changed, 10 insertions(+)
>
> diff --git a/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml b/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
> index 16a0598c6d033..ccf841f194c06 100644
> --- a/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
> +++ b/Documentation/devicetree/bindings/spi/nxp,imx94-xspi.yaml
> @@ -30,6 +30,16 @@ properties:
> - const: base
> - const: mmap
>
> + dmas:
> + items:
> + - description: Transmit DMA
> + - description: Receive DMA
> +
> + dma-names:
> + items:
> + - const: tx
> + - const: rx
> +
> interrupts:
> items:
> - description: interrupt for EENV0
>
> --
> 2.34.1
>
[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 228 bytes --]
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2026-07-22 16:46 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-07-21 21:01 [PATCH v2 0/3] Add DMA support for i.MX94 and improve AHB read performance han.xu
2026-07-21 21:01 ` [PATCH v2 1/3] spi: dt-bindings: nxp,imx94-xspi: add DMA properties han.xu
2026-07-21 21:35 ` Frank Li
2026-07-22 16:46 ` Conor Dooley
2026-07-21 21:01 ` [PATCH v2 2/3] spi: nxp-xspi: Improve AHB read performance with DMA han.xu
2026-07-21 21:01 ` [PATCH v2 3/3] arm64: dts: imx94-xspi: add the DMA channels han.xu
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox