[PATCH 1/2] serial: use dmaengine_get_dma_device() instead of chan->device->dev

[PATCH 1/2] serial: use dmaengine_get_dma_device() instead of chan->device->dev

[Frank.Li]

From: Frank Li <Frank.Li@nxp.com>

Replace direct dma_chan::device::dev access with the proper
dmaengine_get_dma_device() for consumer API

chan->device->dev is not always the device used for DMA mapping.
Some DMA engines support per-channel IOMMU mappings, so different
channels may use different DMA devices.  dmaengine_get_dma_device()
returns the correct device for each channel.

This also prepares for making the DMA engine provider data structures
private. DMA consumers should not access DMA engine internals directly.

Assisted-by: LLM
Signed-off-by: Frank Li <Frank.Li@nxp.com>
---
Check all unders driver/tty/ by rename device to _device in dma_chan struct
with all yes config.

Still few left, one for copy_align access, which need new API. some filter
DMA channel, need new API to get provider devices.

In https://lore.kernel.org/imx/67966b47-22cb-4b07-80c7-2044f557dbcb@app.fastmail.com/
There are better idea to move map singe/sg into dma prep functions.

But it takes more times, not straightforward works. Let replace this firstly,
---
 drivers/tty/serial/8250/8250_dma.c  | 21 +++++++++++++--------
 drivers/tty/serial/8250/8250_omap.c |  4 ++--
 drivers/tty/serial/amba-pl011.c     | 10 +++++-----
 drivers/tty/serial/pch_uart.c       |  2 +-
 drivers/tty/serial/samsung_tty.c    | 27 +++++++++++++++------------
 drivers/tty/serial/sh-sci.c         | 14 ++++++++------
 6 files changed, 44 insertions(+), 34 deletions(-)

diff --git a/drivers/tty/serial/8250/8250_dma.c b/drivers/tty/serial/8250/8250_dma.c
index 5a83e5269b415..dc7fb3db1584d 100644
--- a/drivers/tty/serial/8250/8250_dma.c
+++ b/drivers/tty/serial/8250/8250_dma.c
@@ -19,7 +19,7 @@ static void __dma_tx_complete(void *param)
 	unsigned long	flags;
 	int		ret;
 
-	dma_sync_single_for_cpu(dma->txchan->device->dev, dma->tx_addr,
+	dma_sync_single_for_cpu(dmaengine_get_dma_device(dma->txchan), dma->tx_addr,
 				UART_XMIT_SIZE, DMA_TO_DEVICE);
 
 	uart_port_lock_irqsave(&p->port, &flags);
@@ -136,7 +136,7 @@ int serial8250_tx_dma(struct uart_8250_port *p)
 
 	dma->tx_cookie = dmaengine_submit(desc);
 
-	dma_sync_single_for_device(dma->txchan->device->dev, dma->tx_addr,
+	dma_sync_single_for_device(dmaengine_get_dma_device(dma->txchan), dma->tx_addr,
 				   UART_XMIT_SIZE, DMA_TO_DEVICE);
 
 	dma_async_issue_pending(dma->txchan);
@@ -223,6 +223,8 @@ EXPORT_SYMBOL_GPL(serial8250_rx_dma_flush);
 int serial8250_request_dma(struct uart_8250_port *p)
 {
 	struct uart_8250_dma	*dma = p->dma;
+	struct device *rx_dev;
+	struct device *tx_dev;
 	phys_addr_t rx_dma_addr = dma->rx_dma_addr ?
 				  dma->rx_dma_addr : p->port.mapbase;
 	phys_addr_t tx_dma_addr = dma->tx_dma_addr ?
@@ -282,11 +284,14 @@ int serial8250_request_dma(struct uart_8250_port *p)
 
 	dmaengine_slave_config(dma->txchan, &dma->txconf);
 
+	rx_dev = dmaengine_get_dma_device(dma->rxchan);
+	tx_dev = dmaengine_get_dma_device(dma->txchan);
+
 	/* RX buffer */
 	if (!dma->rx_size)
 		dma->rx_size = PAGE_SIZE;
 
-	dma->rx_buf = dma_alloc_coherent(dma->rxchan->device->dev, dma->rx_size,
+	dma->rx_buf = dma_alloc_coherent(rx_dev, dma->rx_size,
 					&dma->rx_addr, GFP_KERNEL);
 	if (!dma->rx_buf) {
 		ret = -ENOMEM;
@@ -294,12 +299,12 @@ int serial8250_request_dma(struct uart_8250_port *p)
 	}
 
 	/* TX buffer */
-	dma->tx_addr = dma_map_single(dma->txchan->device->dev,
+	dma->tx_addr = dma_map_single(tx_dev,
 					p->port.state->port.xmit_buf,
 					UART_XMIT_SIZE,
 					DMA_TO_DEVICE);
-	if (dma_mapping_error(dma->txchan->device->dev, dma->tx_addr)) {
-		dma_free_coherent(dma->rxchan->device->dev, dma->rx_size,
+	if (dma_mapping_error(tx_dev, dma->tx_addr)) {
+		dma_free_coherent(rx_dev, dma->rx_size,
 				  dma->rx_buf, dma->rx_addr);
 		ret = -ENOMEM;
 		goto err;
@@ -326,14 +331,14 @@ void serial8250_release_dma(struct uart_8250_port *p)
 	/* Release RX resources */
 	dmaengine_terminate_sync(dma->rxchan);
 	dma->rx_running = 0;
-	dma_free_coherent(dma->rxchan->device->dev, dma->rx_size, dma->rx_buf,
+	dma_free_coherent(dmaengine_get_dma_device(dma->rxchan), dma->rx_size, dma->rx_buf,
 			  dma->rx_addr);
 	dma_release_channel(dma->rxchan);
 	dma->rxchan = NULL;
 
 	/* Release TX resources */
 	dmaengine_terminate_sync(dma->txchan);
-	dma_unmap_single(dma->txchan->device->dev, dma->tx_addr,
+	dma_unmap_single(dmaengine_get_dma_device(dma->txchan), dma->tx_addr,
 			 UART_XMIT_SIZE, DMA_TO_DEVICE);
 	dma_release_channel(dma->txchan);
 	dma->txchan = NULL;
diff --git a/drivers/tty/serial/8250/8250_omap.c b/drivers/tty/serial/8250/8250_omap.c
index ceecb39fb82da..3a1473bf54d0f 100644
--- a/drivers/tty/serial/8250/8250_omap.c
+++ b/drivers/tty/serial/8250/8250_omap.c
@@ -1076,7 +1076,7 @@ static void omap_8250_dma_tx_complete(void *param)
 	bool			en_thri = false;
 	struct omap8250_priv	*priv = p->port.private_data;
 
-	dma_sync_single_for_cpu(dma->txchan->device->dev, dma->tx_addr,
+	dma_sync_single_for_cpu(dmaengine_get_dma_device(dma->txchan), dma->tx_addr,
 				UART_XMIT_SIZE, DMA_TO_DEVICE);
 
 	guard(uart_port_lock_irqsave)(&p->port);
@@ -1193,7 +1193,7 @@ static int omap_8250_tx_dma(struct uart_8250_port *p)
 
 	dma->tx_cookie = dmaengine_submit(desc);
 
-	dma_sync_single_for_device(dma->txchan->device->dev, dma->tx_addr,
+	dma_sync_single_for_device(dmaengine_get_dma_device(dma->txchan), dma->tx_addr,
 				   UART_XMIT_SIZE, DMA_TO_DEVICE);
 
 	dma_async_issue_pending(dma->txchan);
diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
index c4824c201e1c3..86729ddbdadc8 100644
--- a/drivers/tty/serial/amba-pl011.c
+++ b/drivers/tty/serial/amba-pl011.c
@@ -435,7 +435,7 @@ static int pl011_fifo_to_tty(struct uart_amba_port *uap)
 static int pl011_dmabuf_init(struct dma_chan *chan, struct pl011_dmabuf *db,
 			     enum dma_data_direction dir)
 {
-	db->buf = dma_alloc_coherent(chan->device->dev, PL011_DMA_BUFFER_SIZE,
+	db->buf = dma_alloc_coherent(dmaengine_get_dma_device(chan), PL011_DMA_BUFFER_SIZE,
 				     &db->dma, GFP_KERNEL);
 	if (!db->buf)
 		return -ENOMEM;
@@ -448,7 +448,7 @@ static void pl011_dmabuf_free(struct dma_chan *chan, struct pl011_dmabuf *db,
 			      enum dma_data_direction dir)
 {
 	if (db->buf) {
-		dma_free_coherent(chan->device->dev,
+		dma_free_coherent(dmaengine_get_dma_device(chan),
 				  PL011_DMA_BUFFER_SIZE, db->buf, db->dma);
 	}
 }
@@ -609,7 +609,7 @@ static void pl011_dma_tx_callback(void *data)
 
 	uart_port_lock_irqsave(&uap->port, &flags);
 	if (uap->dmatx.queued)
-		dma_unmap_single(dmatx->chan->device->dev, dmatx->dma,
+		dma_unmap_single(dmaengine_get_dma_device(dmatx->chan), dmatx->dma,
 				 dmatx->len, DMA_TO_DEVICE);
 
 	dmacr = uap->dmacr;
@@ -867,7 +867,7 @@ __acquires(&uap->port.lock)
 	dmaengine_terminate_async(uap->dmatx.chan);
 
 	if (uap->dmatx.queued) {
-		dma_unmap_single(uap->dmatx.chan->device->dev, uap->dmatx.dma,
+		dma_unmap_single(dmaengine_get_dma_device(uap->dmatx.chan), uap->dmatx.dma,
 				 uap->dmatx.len, DMA_TO_DEVICE);
 		uap->dmatx.queued = false;
 		uap->dmacr &= ~UART011_TXDMAE;
@@ -1249,7 +1249,7 @@ static void pl011_dma_shutdown(struct uart_amba_port *uap)
 		/* In theory, this should already be done by pl011_dma_flush_buffer */
 		dmaengine_terminate_sync(uap->dmatx.chan);
 		if (uap->dmatx.queued) {
-			dma_unmap_single(uap->dmatx.chan->device->dev,
+			dma_unmap_single(dmaengine_get_dma_device(uap->dmatx.chan),
 					 uap->dmatx.dma, uap->dmatx.len,
 					 DMA_TO_DEVICE);
 			uap->dmatx.queued = false;
diff --git a/drivers/tty/serial/pch_uart.c b/drivers/tty/serial/pch_uart.c
index db5c62b0322f6..5e8540e835351 100644
--- a/drivers/tty/serial/pch_uart.c
+++ b/drivers/tty/serial/pch_uart.c
@@ -656,7 +656,7 @@ static bool filter(struct dma_chan *chan, void *slave)
 	struct pch_dma_slave *param = slave;
 
 	if ((chan->chan_id == param->chan_id) && (param->dma_dev ==
-						  chan->device->dev)) {
+						  dmaengine_get_dma_device(chan))) {
 		chan->private = param;
 		return true;
 	} else {
diff --git a/drivers/tty/serial/samsung_tty.c b/drivers/tty/serial/samsung_tty.c
index 63d0232dffc20..640f85e2d636b 100644
--- a/drivers/tty/serial/samsung_tty.c
+++ b/drivers/tty/serial/samsung_tty.c
@@ -300,7 +300,7 @@ static void s3c24xx_serial_stop_tx(struct uart_port *port)
 		dmaengine_pause(dma->tx_chan);
 		dmaengine_tx_status(dma->tx_chan, dma->tx_cookie, &state);
 		dmaengine_terminate_all(dma->tx_chan);
-		dma_sync_single_for_cpu(dma->tx_chan->device->dev,
+		dma_sync_single_for_cpu(dmaengine_get_dma_device(dma->tx_chan),
 					dma->tx_transfer_addr, dma->tx_size,
 					DMA_TO_DEVICE);
 		async_tx_ack(dma->tx_desc);
@@ -333,7 +333,7 @@ static void s3c24xx_serial_tx_dma_complete(void *args)
 	count = dma->tx_bytes_requested - state.residue;
 	async_tx_ack(dma->tx_desc);
 
-	dma_sync_single_for_cpu(dma->tx_chan->device->dev,
+	dma_sync_single_for_cpu(dmaengine_get_dma_device(dma->tx_chan),
 				dma->tx_transfer_addr, dma->tx_size,
 				DMA_TO_DEVICE);
 
@@ -435,7 +435,7 @@ static int s3c24xx_serial_start_tx_dma(struct s3c24xx_uart_port *ourport,
 	dma->tx_size = count & ~(dma_get_cache_alignment() - 1);
 	dma->tx_transfer_addr = dma->tx_addr + tail;
 
-	dma_sync_single_for_device(dma->tx_chan->device->dev,
+	dma_sync_single_for_device(dmaengine_get_dma_device(dma->tx_chan),
 				   dma->tx_transfer_addr, dma->tx_size,
 				   DMA_TO_DEVICE);
 
@@ -509,7 +509,7 @@ static void s3c24xx_uart_copy_rx_to_tty(struct s3c24xx_uart_port *ourport,
 	if (!count)
 		return;
 
-	dma_sync_single_for_cpu(dma->rx_chan->device->dev, dma->rx_addr,
+	dma_sync_single_for_cpu(dmaengine_get_dma_device(dma->rx_chan), dma->rx_addr,
 				dma->rx_size, DMA_FROM_DEVICE);
 
 	ourport->port.icount.rx += count;
@@ -631,7 +631,7 @@ static void s3c64xx_start_rx_dma(struct s3c24xx_uart_port *ourport)
 {
 	struct s3c24xx_uart_dma *dma = ourport->dma;
 
-	dma_sync_single_for_device(dma->rx_chan->device->dev, dma->rx_addr,
+	dma_sync_single_for_device(dmaengine_get_dma_device(dma->rx_chan), dma->rx_addr,
 				   dma->rx_size, DMA_FROM_DEVICE);
 
 	dma->rx_desc = dmaengine_prep_slave_single(dma->rx_chan,
@@ -1101,20 +1101,23 @@ static int s3c24xx_serial_request_dma(struct s3c24xx_uart_port *p)
 		goto err_release_tx;
 	}
 
-	dma->rx_addr = dma_map_single(dma->rx_chan->device->dev, dma->rx_buf,
+	struct device *rx_dev = dmaengine_get_dma_device(dma->rx_chan);
+	struct device *tx_dev = dmaengine_get_dma_device(dma->tx_chan);
+
+	dma->rx_addr = dma_map_single(rx_dev, dma->rx_buf,
 				      dma->rx_size, DMA_FROM_DEVICE);
-	if (dma_mapping_error(dma->rx_chan->device->dev, dma->rx_addr)) {
+	if (dma_mapping_error(rx_dev, dma->rx_addr)) {
 		reason = "DMA mapping error for RX buffer";
 		ret = -EIO;
 		goto err_free_rx;
 	}
 
 	/* TX buffer */
-	dma->tx_addr = dma_map_single(dma->tx_chan->device->dev,
+	dma->tx_addr = dma_map_single(tx_dev,
 				      p->port.state->port.xmit_buf,
 				      UART_XMIT_SIZE,
 				      DMA_TO_DEVICE);
-	if (dma_mapping_error(dma->tx_chan->device->dev, dma->tx_addr)) {
+	if (dma_mapping_error(tx_dev, dma->tx_addr)) {
 		reason = "DMA mapping error for TX buffer";
 		ret = -EIO;
 		goto err_unmap_rx;
@@ -1123,7 +1126,7 @@ static int s3c24xx_serial_request_dma(struct s3c24xx_uart_port *p)
 	return 0;
 
 err_unmap_rx:
-	dma_unmap_single(dma->rx_chan->device->dev, dma->rx_addr,
+	dma_unmap_single(dmaengine_get_dma_device(dma->rx_chan), dma->rx_addr,
 			 dma->rx_size, DMA_FROM_DEVICE);
 err_free_rx:
 	kfree(dma->rx_buf);
@@ -1143,7 +1146,7 @@ static void s3c24xx_serial_release_dma(struct s3c24xx_uart_port *p)
 
 	if (dma->rx_chan) {
 		dmaengine_terminate_all(dma->rx_chan);
-		dma_unmap_single(dma->rx_chan->device->dev, dma->rx_addr,
+		dma_unmap_single(dmaengine_get_dma_device(dma->rx_chan), dma->rx_addr,
 				 dma->rx_size, DMA_FROM_DEVICE);
 		kfree(dma->rx_buf);
 		dma_release_channel(dma->rx_chan);
@@ -1152,7 +1155,7 @@ static void s3c24xx_serial_release_dma(struct s3c24xx_uart_port *p)
 
 	if (dma->tx_chan) {
 		dmaengine_terminate_all(dma->tx_chan);
-		dma_unmap_single(dma->tx_chan->device->dev, dma->tx_addr,
+		dma_unmap_single(dmaengine_get_dma_device(dma->tx_chan), dma->tx_addr,
 				 UART_XMIT_SIZE, DMA_TO_DEVICE);
 		dma_release_channel(dma->tx_chan);
 		dma->tx_chan = NULL;
diff --git a/drivers/tty/serial/sh-sci.c b/drivers/tty/serial/sh-sci.c
index 50ae9aae6b614..bd58bbfa65aa7 100644
--- a/drivers/tty/serial/sh-sci.c
+++ b/drivers/tty/serial/sh-sci.c
@@ -1495,7 +1495,7 @@ static void sci_dma_rx_release(struct sci_port *s)
 	uart_port_unlock_irqrestore(port, flags);
 
 	dmaengine_terminate_sync(chan);
-	dma_free_coherent(chan->device->dev, s->buf_len_rx * 2, s->rx_buf[0],
+	dma_free_coherent(dmaengine_get_dma_device(chan), s->buf_len_rx * 2, s->rx_buf[0],
 			  sg_dma_address(&s->sg_rx[0]));
 	dma_release_channel(chan);
 }
@@ -1591,7 +1591,7 @@ static void sci_dma_tx_release(struct sci_port *s)
 	s->chan_tx_saved = s->chan_tx = NULL;
 	s->cookie_tx = -EINVAL;
 	dmaengine_terminate_sync(chan);
-	dma_unmap_single(chan->device->dev, s->tx_dma_addr, UART_XMIT_SIZE,
+	dma_unmap_single(dmaengine_get_dma_device(chan), s->tx_dma_addr, UART_XMIT_SIZE,
 			 DMA_TO_DEVICE);
 	dma_release_channel(chan);
 }
@@ -1676,7 +1676,7 @@ static void sci_dma_tx_work_fn(struct work_struct *work)
 		goto switch_to_pio;
 	}
 
-	dma_sync_single_for_device(chan->device->dev, buf, s->tx_dma_len,
+	dma_sync_single_for_device(dmaengine_get_dma_device(chan), buf, s->tx_dma_len,
 				   DMA_TO_DEVICE);
 
 	desc->callback = sci_dma_tx_complete;
@@ -1832,12 +1832,13 @@ static void sci_request_dma(struct uart_port *port)
 	chan = sci_request_dma_chan(port, DMA_MEM_TO_DEV);
 	dev_dbg(port->dev, "%s: TX: got channel %p\n", __func__, chan);
 	if (chan) {
+		struct device *dma_dev = dmaengine_get_dma_device(chan);
 		/* UART circular tx buffer is an aligned page. */
-		s->tx_dma_addr = dma_map_single(chan->device->dev,
+		s->tx_dma_addr = dma_map_single(dma_dev,
 						tport->xmit_buf,
 						UART_XMIT_SIZE,
 						DMA_TO_DEVICE);
-		if (dma_mapping_error(chan->device->dev, s->tx_dma_addr)) {
+		if (dma_mapping_error(dma_dev, s->tx_dma_addr)) {
 			dev_warn(port->dev, "Failed mapping Tx DMA descriptor\n");
 			dma_release_channel(chan);
 		} else {
@@ -1853,12 +1854,13 @@ static void sci_request_dma(struct uart_port *port)
 	chan = sci_request_dma_chan(port, DMA_DEV_TO_MEM);
 	dev_dbg(port->dev, "%s: RX: got channel %p\n", __func__, chan);
 	if (chan) {
+		struct device *dma_dev = dmaengine_get_dma_device(chan);
 		unsigned int i;
 		dma_addr_t dma;
 		void *buf;
 
 		s->buf_len_rx = 2 * max_t(size_t, 16, port->fifosize);
-		buf = dma_alloc_coherent(chan->device->dev, s->buf_len_rx * 2,
+		buf = dma_alloc_coherent(dma_dev, s->buf_len_rx * 2,
 					 &dma, GFP_KERNEL);
 		if (!buf) {
 			dev_warn(port->dev,
-- 
2.43.0

[PATCH 2/2] serial: amba-pl011: use dmaengine public API instead of raw ops

[Frank.Li]

From: Frank Li <Frank.Li@nxp.com>

Replace direct calls to the internal DMA device ops
(device->device_tx_status() and device->device_issue_pending())
with the proper public DMA engine wrappers:
  - dmaengine_tx_status()
  - dma_async_issue_pending()

Drivers must not call device ops directly; the wrappers provide
locking, debug checks, and a stable interface that hides internal
implementation details of the DMA controller driver.

Signed-off-by: Frank Li <Frank.Li@nxp.com>
---
 drivers/tty/serial/amba-pl011.c | 9 ++++-----
 1 file changed, 4 insertions(+), 5 deletions(-)

diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
index 86729ddbdadc8..cd9128b213c94 100644
--- a/drivers/tty/serial/amba-pl011.c
+++ b/drivers/tty/serial/amba-pl011.c
@@ -721,7 +721,7 @@ static int pl011_dma_tx_refill(struct uart_amba_port *uap)
 	dmaengine_submit(desc);
 
 	/* Fire the DMA transaction */
-	dma_dev->device_issue_pending(chan);
+	dma_async_issue_pending(chan);
 
 	uap->dmacr |= UART011_TXDMAE;
 	pl011_write(uap->dmacr, uap, REG_DMACR);
@@ -1011,8 +1011,7 @@ static void pl011_dma_rx_irq(struct uart_amba_port *uap)
 	 */
 	if (dmaengine_pause(rxchan))
 		dev_err(uap->port.dev, "unable to pause DMA transfer\n");
-	dmastat = rxchan->device->device_tx_status(rxchan,
-						   dmarx->cookie, &state);
+	dmastat = dmaengine_tx_status(rxchan, dmarx->cookie, &state);
 	if (dmastat != DMA_PAUSED)
 		dev_err(uap->port.dev, "unable to pause DMA transfer\n");
 
@@ -1066,7 +1065,7 @@ static void pl011_dma_rx_callback(void *data)
 	 * Rx data can be taken by the UART interrupts during
 	 * the DMA irq handler. So we check the residue here.
 	 */
-	rxchan->device->device_tx_status(rxchan, dmarx->cookie, &state);
+	dmaengine_tx_status(rxchan, dmarx->cookie, &state);
 	pending = dbuf->len - state.residue;
 	BUG_ON(pending > PL011_DMA_BUFFER_SIZE);
 	/* Then we terminate the transfer - we now know our residue */
@@ -1124,7 +1123,7 @@ static void pl011_dma_rx_poll(struct timer_list *t)
 	struct dma_tx_state state;
 
 	dbuf = dmarx->use_buf_b ? &uap->dmarx.dbuf_b : &uap->dmarx.dbuf_a;
-	rxchan->device->device_tx_status(rxchan, dmarx->cookie, &state);
+	dmaengine_tx_status(rxchan, dmarx->cookie, &state);
 	if (likely(state.residue < dmarx->last_residue)) {
 		dmataken = dbuf->len - dmarx->last_residue;
 		size = dmarx->last_residue - state.residue;
-- 
2.43.0

Re: [PATCH 2/2] serial: amba-pl011: use dmaengine public API instead of raw ops

[sashiko-bot]

> From: Frank Li <Frank.Li@nxp.com>
> 
> Replace direct calls to the internal DMA device ops
> (device->device_tx_status() and device->device_issue_pending())
> with the proper public DMA engine wrappers:
>   - dmaengine_tx_status()
>   - dma_async_issue_pending()
> 
> Drivers must not call device ops directly; the wrappers provide
> locking, debug checks, and a stable interface that hides internal
> implementation details of the DMA controller driver.
> 
> Signed-off-by: Frank Li <Frank.Li@nxp.com>

Sashiko has reviewed this patch and found no issues. It looks great!

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260925200850.1678179-1-Frank.Li@oss.nxp.com?part=2

Re: [PATCH 1/2] serial: use dmaengine_get_dma_device() instead of chan->device->dev

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] dma-mapping: asymmetric dma_map_single/unmap pair in pl011
--

commit fb77536d75ed603638aebc851f73960ddf8efe10
Author: Frank Li <Frank.Li@nxp.com>

serial: use dmaengine_get_dma_device() instead of chan->device->dev

This commit replaces direct access to chan->device->dev with the
dmaengine_get_dma_device() API across multiple serial drivers. This correctly
handles per-channel IOMMU mappings and prepares for making DMA engine
provider structures private.

Link: https://lore.kernel.org/imx/67966b47-22cb-4b07-80c7-2044f557dbcb@app.fastmail.com/

> diff --git a/drivers/tty/serial/amba-pl011.c b/drivers/tty/serial/amba-pl011.c
> index c4824c201e1c3..86729ddbdadc8 100644
> --- a/drivers/tty/serial/amba-pl011.c
> +++ b/drivers/tty/serial/amba-pl011.c

[ ... ]

> @@ -609,7 +609,7 @@ static void pl011_dma_tx_callback(void *data)
>  
>  	uart_port_lock_irqsave(&uap->port, &flags);
>  	if (uap->dmatx.queued)
> -		dma_unmap_single(dmatx->chan->device->dev, dmatx->dma,
> +		dma_unmap_single(dmaengine_get_dma_device(dmatx->chan), dmatx->dma,

[Severity: High]
Does this change create an asymmetric map/unmap pair in the PL011 driver?

By updating pl011_dma_tx_callback() to use dmaengine_get_dma_device() for the
unmap operation, it appears to no longer match the map operation in
pl011_dma_tx_refill() which was not updated in this patch:

drivers/tty/serial/amba-pl011.c:pl011_dma_tx_refill() {
    ...
	dmatx->dma = dma_map_single(dma_dev->dev, dmatx->buf, count,
				    DMA_TO_DEVICE);
    ...
}

The DMA API requires the exact same device pointer to be used for both
mapping and unmapping. On platforms where the DMA engine provides per-channel
devices, dmaengine_get_dma_device() will return a different device pointer
than the raw device pointer used during mapping.

Could this break DMA tracking or cause IOMMU faults during standard TX
operations? Should the map operation in pl011_dma_tx_refill() be updated to
use the new API as well?

>  				 dmatx->len, DMA_TO_DEVICE);
>  
>  	dmacr = uap->dmacr;

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260925200850.1678179-1-Frank.Li@oss.nxp.com?part=1

Re: [PATCH 1/2] serial: use dmaengine_get_dma_device() instead of chan->device->dev

[Andy Shevchenko]

On Fri, Sep 25, 2026 at 04:08:30PM -0400, Frank.Li@oss.nxp.com wrote:
> From: Frank Li <Frank.Li@nxp.com>
> 
> Replace direct dma_chan::device::dev access with the proper
> dmaengine_get_dma_device() for consumer API
> 
> chan->device->dev is not always the device used for DMA mapping.
> Some DMA engines support per-channel IOMMU mappings, so different
> channels may use different DMA devices.  dmaengine_get_dma_device()
> returns the correct device for each channel.
> 
> This also prepares for making the DMA engine provider data structures
> private. DMA consumers should not access DMA engine internals directly.

...

>  	/* RX buffer */
>  	if (!dma->rx_size)
>  		dma->rx_size = PAGE_SIZE;
>  
> -	dma->rx_buf = dma_alloc_coherent(dma->rxchan->device->dev, dma->rx_size,
> +	dma->rx_buf = dma_alloc_coherent(rx_dev, dma->rx_size,
>  					&dma->rx_addr, GFP_KERNEL);

Now one parameter can be moved up and positive outcome the split becomes logical
(on a logic boundaries).

>  	if (!dma->rx_buf) {
>  		ret = -ENOMEM;

...

>  	/* TX buffer */
> -	dma->tx_addr = dma_map_single(dma->txchan->device->dev,
> +	dma->tx_addr = dma_map_single(tx_dev,
>  					p->port.state->port.xmit_buf,
>  					UART_XMIT_SIZE,
>  					DMA_TO_DEVICE);

You can fix indentation while at it.

> -	if (dma_mapping_error(dma->txchan->device->dev, dma->tx_addr)) {
> -		dma_free_coherent(dma->rxchan->device->dev, dma->rx_size,
> +	if (dma_mapping_error(tx_dev, dma->tx_addr)) {
> +		dma_free_coherent(rx_dev, dma->rx_size,
>  				  dma->rx_buf, dma->rx_addr);
>  		ret = -ENOMEM;

...

>  	/* Release RX resources */
>  	dmaengine_terminate_sync(dma->rxchan);
>  	dma->rx_running = 0;
> -	dma_free_coherent(dma->rxchan->device->dev, dma->rx_size, dma->rx_buf,
> +	dma_free_coherent(dmaengine_get_dma_device(dma->rxchan), dma->rx_size, dma->rx_buf,
>  			  dma->rx_addr);

And here the last parameter of the previous line can be moved to the next line.

...

> +	db->buf = dma_alloc_coherent(dmaengine_get_dma_device(chan), PL011_DMA_BUFFER_SIZE,
>  				     &db->dma, GFP_KERNEL);

^^^ (1)

>  	if (!db->buf)
>  		return -ENOMEM;

...

>  {
>  	if (db->buf) {
> -		dma_free_coherent(chan->device->dev,
> +		dma_free_coherent(dmaengine_get_dma_device(chan),
>  				  PL011_DMA_BUFFER_SIZE, db->buf, db->dma);

Perhaps you want both (1) and this be consistent, either (1) be rewrapped,
or this one

		dma_free_coherent(dmaengine_get_dma_device(chan), PL011_DMA_BUFFER_SIZE,
				  db->buf, db->dma);

>  	}

...

>  	struct pch_dma_slave *param = slave;
>  
>  	if ((chan->chan_id == param->chan_id) && (param->dma_dev ==
> -						  chan->device->dev)) {
> +						  dmaengine_get_dma_device(chan))) {

Even original code has broken indentation. What about rewrapping it?

	if ((chan->chan_id == param->chan_id) &&
	    (param->dma_dev == dmaengine_get_dma_device(chan))) {

>  		chan->private = param;
>  		return true;

-- 
With Best Regards,
Andy Shevchenko