[PATCH v1 0/2] spi: tegra210-quad: Improve timeout handling under high system load

[PATCH v1 0/2] spi: tegra210-quad: Improve timeout handling under high system load

[Vishwaroop A]

Hi,

This patch series addresses timeout handling issues in the Tegra QSPI driver
that occur under high system load conditions. We've observed that when CPUs
are saturated (due to error injection, RAS firmware activity, or general CPU
contention), QSPI interrupt handlers can be delayed, causing spurious transfer
failures even though the hardware completed the operation successfully.

Patch 1 fixes a stale pointer issue by ensuring curr_xfer is cleared on timeout
and checked when the IRQ thread finally runs. It also ensures interrupts are
properly cleared on failure paths.

Patch 2 adds hardware status checking on timeout. Before failing a transfer,
the driver now reads QSPI_TRANS_STATUS to verify if the hardware actually
completed the operation. If so, it manually invokes the completion handler
instead of failing the transfer. This distinguishes genuine hardware timeouts
from delayed/lost interrupts.

These changes have been tested in production environments under various high
load scenarios including RAS testing and CPU saturation workloads.

Testing:
- Verified normal operation under light load
- Tested under heavy CPU load with concurrent workloads
- Validated with RAS firmware activity and error injection
- Confirmed no regressions in existing timeout behavior

Thierry Reding (1):
  spi: tegra210-quad: Fix timeout handling

Vishwaroop A (1):
  spi: tegra210-quad: Check hardware status on timeout

 drivers/spi/spi-tegra210-quad.c | 195 ++++++++++++++++++++++++++------
 1 file changed, 138 insertions(+), 57 deletions(-)

-- 
2.34.1

[PATCH v1 1/2] spi: tegra210-quad: Fix timeout handling

[Vishwaroop A]

When the CPU that the QSPI interrupt handler runs on (typically CPU 0)
is excessively busy, it can lead to rare cases of the IRQ thread not
running before the transfer timeout is reached.

While handling the timeouts, any pending transfers are cleaned up and
the message that they correspond to is marked as failed, which leaves
the curr_xfer field pointing at stale memory.

To avoid this, clear curr_xfer to NULL upon timeout and check for this
condition when the IRQ thread is finally run.

While at it, also make sure to clear interrupts on failure so that new
interrupts can be run.

A better, more involved, fix would move the interrupt clearing into a
hard IRQ handler. Ideally we would also want to signal that the IRQ
thread no longer needs to be run after the timeout is hit to avoid the
extra check for a valid transfer.

Fixes: 921fc1838fb0 ("spi: tegra210-quad: Add support for Tegra210 QSPI controller")
Signed-off-by: Thierry Reding <treding@nvidia.com>
Signed-off-by: Vishwaroop A <va@nvidia.com>
---
 drivers/spi/spi-tegra210-quad.c | 31 ++++++++++++++++++++++++-------
 1 file changed, 24 insertions(+), 7 deletions(-)

diff --git a/drivers/spi/spi-tegra210-quad.c b/drivers/spi/spi-tegra210-quad.c
index 3be7499db21e..c2f880d08109 100644
--- a/drivers/spi/spi-tegra210-quad.c
+++ b/drivers/spi/spi-tegra210-quad.c
@@ -1024,8 +1024,10 @@ static void tegra_qspi_handle_error(struct tegra_qspi *tqspi)
 	dev_err(tqspi->dev, "error in transfer, fifo status 0x%08x\n", tqspi->status_reg);
 	tegra_qspi_dump_regs(tqspi);
 	tegra_qspi_flush_fifos(tqspi, true);
-	if (device_reset(tqspi->dev) < 0)
+	if (device_reset(tqspi->dev) < 0) {
 		dev_warn_once(tqspi->dev, "device reset failed\n");
+		tegra_qspi_mask_clear_irq(tqspi);
+	}
 }
 
 static void tegra_qspi_transfer_end(struct spi_device *spi)
@@ -1175,12 +1177,14 @@ static int tegra_qspi_combined_seq_xfer(struct tegra_qspi *tqspi,
 							  QSPI_DMA_CTL);
 				}
 
-				/* Reset controller if timeout happens */
-				if (device_reset(tqspi->dev) < 0)
-					dev_warn_once(tqspi->dev,
-						      "device reset failed\n");
-				ret = -EIO;
-				goto exit;
+			/* Reset controller if timeout happens */
+			if (device_reset(tqspi->dev) < 0) {
+				dev_warn_once(tqspi->dev,
+					      "device reset failed\n");
+				tegra_qspi_mask_clear_irq(tqspi);
+			}
+			ret = -EIO;
+			goto exit;
 			}
 
 			if (tqspi->tx_status ||  tqspi->rx_status) {
@@ -1196,6 +1200,7 @@ static int tegra_qspi_combined_seq_xfer(struct tegra_qspi *tqspi,
 			goto exit;
 		}
 		msg->actual_length += xfer->len;
+		tqspi->curr_xfer = NULL;
 		if (!xfer->cs_change && transfer_phase == DATA_TRANSFER) {
 			tegra_qspi_transfer_end(spi);
 			spi_transfer_delay_exec(xfer);
@@ -1205,6 +1210,7 @@ static int tegra_qspi_combined_seq_xfer(struct tegra_qspi *tqspi,
 	ret = 0;
 
 exit:
+	tqspi->curr_xfer = NULL;
 	msg->status = ret;
 
 	return ret;
@@ -1290,6 +1296,8 @@ static int tegra_qspi_non_combined_seq_xfer(struct tegra_qspi *tqspi,
 		msg->actual_length += xfer->len + dummy_bytes;
 
 complete_xfer:
+		tqspi->curr_xfer = NULL;
+
 		if (ret < 0) {
 			tegra_qspi_transfer_end(spi);
 			spi_transfer_delay_exec(xfer);
@@ -1480,6 +1488,15 @@ static irqreturn_t tegra_qspi_isr_thread(int irq, void *context_data)
 {
 	struct tegra_qspi *tqspi = context_data;
 
+	/*
+	 * Occasionally the IRQ thread takes a long time to wake up (usually
+	 * when the CPU that it's running on is excessively busy) and we have
+	 * already reached the timeout before and cleaned up the timed out
+	 * transfer. Avoid any processing in that case and bail out early.
+	 */
+	if (tqspi->curr_xfer == NULL)
+		return IRQ_NONE;
+
 	tqspi->status_reg = tegra_qspi_readl(tqspi, QSPI_FIFO_STATUS);
 
 	if (tqspi->cur_direction & DATA_DIR_TX)
-- 
2.17.1

[PATCH v1 2/2] spi: tegra210-quad: Check hardware status on timeout

[Vishwaroop A]

Under high system load, QSPI interrupts can be delayed or blocked on the
target CPU, causing wait_for_completion_timeout() to report failure even
though the hardware successfully completed the transfer. This has been
observed in production during error injection, RAS firmware activity, and
CPU saturation scenarios.

When a timeout occurs, check the QSPI_RDY bit in QSPI_TRANS_STATUS to
determine if the hardware actually completed the transfer. If so, manually
invoke the completion handler to process the transfer successfully instead
of failing it.

This distinguishes lost/delayed interrupts from real hardware timeouts,
preventing unnecessary failures of transfers that completed successfully.

Signed-off-by: Vishwaroop A <va@nvidia.com>
---
 drivers/spi/spi-tegra210-quad.c | 164 ++++++++++++++++++++++----------
 1 file changed, 114 insertions(+), 50 deletions(-)

diff --git a/drivers/spi/spi-tegra210-quad.c b/drivers/spi/spi-tegra210-quad.c
index c2f880d08109..757e9fe23e0e 100644
--- a/drivers/spi/spi-tegra210-quad.c
+++ b/drivers/spi/spi-tegra210-quad.c
@@ -1019,17 +1019,22 @@ static void tegra_qspi_dump_regs(struct tegra_qspi *tqspi)
 		tegra_qspi_readl(tqspi, QSPI_FIFO_STATUS));
 }
 
-static void tegra_qspi_handle_error(struct tegra_qspi *tqspi)
+static void tegra_qspi_reset(struct tegra_qspi *tqspi)
 {
-	dev_err(tqspi->dev, "error in transfer, fifo status 0x%08x\n", tqspi->status_reg);
-	tegra_qspi_dump_regs(tqspi);
-	tegra_qspi_flush_fifos(tqspi, true);
 	if (device_reset(tqspi->dev) < 0) {
 		dev_warn_once(tqspi->dev, "device reset failed\n");
 		tegra_qspi_mask_clear_irq(tqspi);
 	}
 }
 
+static void tegra_qspi_handle_error(struct tegra_qspi *tqspi)
+{
+	dev_err(tqspi->dev, "error in transfer, fifo status 0x%08x\n", tqspi->status_reg);
+	tegra_qspi_dump_regs(tqspi);
+	tegra_qspi_flush_fifos(tqspi, true);
+	tegra_qspi_reset(tqspi);
+}
+
 static void tegra_qspi_transfer_end(struct spi_device *spi)
 {
 	struct tegra_qspi *tqspi = spi_controller_get_devdata(spi->controller);
@@ -1043,6 +1048,49 @@ static void tegra_qspi_transfer_end(struct spi_device *spi)
 	tegra_qspi_writel(tqspi, tqspi->def_command1_reg, QSPI_COMMAND1);
 }
 
+static irqreturn_t handle_cpu_based_xfer(struct tegra_qspi *tqspi);
+static irqreturn_t handle_dma_based_xfer(struct tegra_qspi *tqspi);
+
+/**
+ * tegra_qspi_handle_timeout - Handle transfer timeout with hardware check
+ * @tqspi: QSPI controller instance
+ *
+ * When a timeout occurs but hardware has completed the transfer (interrupt
+ * was lost or delayed), manually trigger transfer completion processing.
+ * This avoids failing transfers that actually succeeded.
+ *
+ * Returns: 0 if transfer was completed, -ETIMEDOUT if real timeout
+ */
+static int tegra_qspi_handle_timeout(struct tegra_qspi *tqspi)
+{
+	irqreturn_t ret;
+	u32 status;
+
+	/* Check if hardware actually completed the transfer */
+	status = tegra_qspi_readl(tqspi, QSPI_TRANS_STATUS);
+	if (!(status & QSPI_RDY))
+		return -ETIMEDOUT;
+
+	/*
+	 * Hardware completed but interrupt was lost/delayed. Manually
+	 * process the completion by calling the appropriate handler.
+	 */
+	dev_warn_ratelimited(tqspi->dev,
+			     "QSPI interrupt timeout, but transfer complete\n");
+
+	/* Clear the transfer status */
+	status = tegra_qspi_readl(tqspi, QSPI_TRANS_STATUS);
+	tegra_qspi_writel(tqspi, status, QSPI_TRANS_STATUS);
+
+	/* Manually trigger completion handler */
+	if (!tqspi->is_curr_dma_xfer)
+		ret = handle_cpu_based_xfer(tqspi);
+	else
+		ret = handle_dma_based_xfer(tqspi);
+
+	return (ret == IRQ_HANDLED) ? 0 : -EIO;
+}
+
 static u32 tegra_qspi_cmd_config(bool is_ddr, u8 bus_width, u8 len)
 {
 	u32 cmd_config = 0;
@@ -1074,6 +1122,30 @@ static u32 tegra_qspi_addr_config(bool is_ddr, u8 bus_width, u8 len)
 	return addr_config;
 }
 
+static void tegra_qspi_dma_stop(struct tegra_qspi *tqspi)
+{
+	u32 value;
+
+	if ((tqspi->cur_direction & DATA_DIR_TX) && tqspi->tx_dma_chan)
+		dmaengine_terminate_all(tqspi->tx_dma_chan);
+
+	if ((tqspi->cur_direction & DATA_DIR_RX) && tqspi->rx_dma_chan)
+		dmaengine_terminate_all(tqspi->rx_dma_chan);
+
+	value = tegra_qspi_readl(tqspi, QSPI_DMA_CTL);
+	value &= ~QSPI_DMA_EN;
+	tegra_qspi_writel(tqspi, value, QSPI_DMA_CTL);
+}
+
+static void tegra_qspi_pio_stop(struct tegra_qspi *tqspi)
+{
+	u32 value;
+
+	value = tegra_qspi_readl(tqspi, QSPI_COMMAND1);
+	value &= ~QSPI_PIO;
+	tegra_qspi_writel(tqspi, value, QSPI_COMMAND1);
+}
+
 static int tegra_qspi_combined_seq_xfer(struct tegra_qspi *tqspi,
 					struct spi_message *msg)
 {
@@ -1081,7 +1153,7 @@ static int tegra_qspi_combined_seq_xfer(struct tegra_qspi *tqspi,
 	struct spi_transfer *xfer;
 	struct spi_device *spi = msg->spi;
 	u8 transfer_phase = 0;
-	u32 cmd1 = 0, dma_ctl = 0;
+	u32 cmd1 = 0;
 	int ret = 0;
 	u32 address_value = 0;
 	u32 cmd_config = 0, addr_config = 0;
@@ -1148,43 +1220,28 @@ static int tegra_qspi_combined_seq_xfer(struct tegra_qspi *tqspi,
 					QSPI_DMA_TIMEOUT);
 
 			if (WARN_ON_ONCE(ret == 0)) {
-				dev_err_ratelimited(tqspi->dev,
-						    "QSPI Transfer failed with timeout\n");
-				if (tqspi->is_curr_dma_xfer) {
-					if ((tqspi->cur_direction & DATA_DIR_TX) &&
-					    tqspi->tx_dma_chan)
-						dmaengine_terminate_all(tqspi->tx_dma_chan);
-					if ((tqspi->cur_direction & DATA_DIR_RX) &&
-					    tqspi->rx_dma_chan)
-						dmaengine_terminate_all(tqspi->rx_dma_chan);
-				}
-
-				/* Abort transfer by resetting pio/dma bit */
-				if (!tqspi->is_curr_dma_xfer) {
-					cmd1 = tegra_qspi_readl
-							(tqspi,
-							 QSPI_COMMAND1);
-					cmd1 &= ~QSPI_PIO;
-					tegra_qspi_writel
-							(tqspi, cmd1,
-							 QSPI_COMMAND1);
-				} else {
-					dma_ctl = tegra_qspi_readl
-							(tqspi,
-							 QSPI_DMA_CTL);
-					dma_ctl &= ~QSPI_DMA_EN;
-					tegra_qspi_writel(tqspi, dma_ctl,
-							  QSPI_DMA_CTL);
+				/*
+				 * Check if hardware completed the transfer
+				 * even though interrupt was lost or delayed.
+				 * If so, process the completion and continue.
+				 */
+				ret = tegra_qspi_handle_timeout(tqspi);
+				if (ret < 0) {
+					/* Real timeout - clean up and fail */
+					dev_err(tqspi->dev, "transfer timeout\n");
+
+					/* Abort transfer by resetting pio/dma bit */
+					if (tqspi->is_curr_dma_xfer)
+						tegra_qspi_dma_stop(tqspi);
+					else
+						tegra_qspi_pio_stop(tqspi);
+
+					/* Reset controller if timeout happens */
+					tegra_qspi_reset(tqspi);
+
+					ret = -EIO;
+					goto exit;
 				}
-
-			/* Reset controller if timeout happens */
-			if (device_reset(tqspi->dev) < 0) {
-				dev_warn_once(tqspi->dev,
-					      "device reset failed\n");
-				tegra_qspi_mask_clear_irq(tqspi);
-			}
-			ret = -EIO;
-			goto exit;
 			}
 
 			if (tqspi->tx_status ||  tqspi->rx_status) {
@@ -1275,16 +1332,23 @@ static int tegra_qspi_non_combined_seq_xfer(struct tegra_qspi *tqspi,
 		ret = wait_for_completion_timeout(&tqspi->xfer_completion,
 						  QSPI_DMA_TIMEOUT);
 		if (WARN_ON(ret == 0)) {
-			dev_err(tqspi->dev, "transfer timeout\n");
-			if (tqspi->is_curr_dma_xfer) {
-				if ((tqspi->cur_direction & DATA_DIR_TX) && tqspi->tx_dma_chan)
-					dmaengine_terminate_all(tqspi->tx_dma_chan);
-				if ((tqspi->cur_direction & DATA_DIR_RX) && tqspi->rx_dma_chan)
-					dmaengine_terminate_all(tqspi->rx_dma_chan);
+			/*
+			 * Check if hardware completed the transfer even though
+			 * interrupt was lost or delayed. If so, process the
+			 * completion and continue.
+			 */
+			ret = tegra_qspi_handle_timeout(tqspi);
+			if (ret < 0) {
+				/* Real timeout - clean up and fail */
+				dev_err(tqspi->dev, "transfer timeout\n");
+
+				if (tqspi->is_curr_dma_xfer)
+					tegra_qspi_dma_stop(tqspi);
+
+				tegra_qspi_handle_error(tqspi);
+				ret = -EIO;
+				goto complete_xfer;
 			}
-			tegra_qspi_handle_error(tqspi);
-			ret = -EIO;
-			goto complete_xfer;
 		}
 
 		if (tqspi->tx_status ||  tqspi->rx_status) {
-- 
2.17.1

Re: [PATCH v1 1/2] spi: tegra210-quad: Fix timeout handling

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 2863 bytes --]

On Fri, Oct 10, 2025 at 03:20:00PM +0000, Vishwaroop A wrote:
> When the CPU that the QSPI interrupt handler runs on (typically CPU 0)
> is excessively busy, it can lead to rare cases of the IRQ thread not
> running before the transfer timeout is reached.
> 
> While handling the timeouts, any pending transfers are cleaned up and
> the message that they correspond to is marked as failed, which leaves
> the curr_xfer field pointing at stale memory.
> 
> To avoid this, clear curr_xfer to NULL upon timeout and check for this
> condition when the IRQ thread is finally run.
> 
> While at it, also make sure to clear interrupts on failure so that new
> interrupts can be run.
> 
> A better, more involved, fix would move the interrupt clearing into a
> hard IRQ handler. Ideally we would also want to signal that the IRQ
> thread no longer needs to be run after the timeout is hit to avoid the
> extra check for a valid transfer.
> 
> Fixes: 921fc1838fb0 ("spi: tegra210-quad: Add support for Tegra210 QSPI controller")
> Signed-off-by: Thierry Reding <treding@nvidia.com>
> Signed-off-by: Vishwaroop A <va@nvidia.com>
> ---
>  drivers/spi/spi-tegra210-quad.c | 31 ++++++++++++++++++++++++-------
>  1 file changed, 24 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/spi/spi-tegra210-quad.c b/drivers/spi/spi-tegra210-quad.c
> index 3be7499db21e..c2f880d08109 100644
> --- a/drivers/spi/spi-tegra210-quad.c
> +++ b/drivers/spi/spi-tegra210-quad.c
> @@ -1024,8 +1024,10 @@ static void tegra_qspi_handle_error(struct tegra_qspi *tqspi)
>  	dev_err(tqspi->dev, "error in transfer, fifo status 0x%08x\n", tqspi->status_reg);
>  	tegra_qspi_dump_regs(tqspi);
>  	tegra_qspi_flush_fifos(tqspi, true);
> -	if (device_reset(tqspi->dev) < 0)
> +	if (device_reset(tqspi->dev) < 0) {
>  		dev_warn_once(tqspi->dev, "device reset failed\n");
> +		tegra_qspi_mask_clear_irq(tqspi);
> +	}
>  }
>  
>  static void tegra_qspi_transfer_end(struct spi_device *spi)
> @@ -1175,12 +1177,14 @@ static int tegra_qspi_combined_seq_xfer(struct tegra_qspi *tqspi,
>  							  QSPI_DMA_CTL);
>  				}
>  
> -				/* Reset controller if timeout happens */
> -				if (device_reset(tqspi->dev) < 0)
> -					dev_warn_once(tqspi->dev,
> -						      "device reset failed\n");
> -				ret = -EIO;
> -				goto exit;
> +			/* Reset controller if timeout happens */
> +			if (device_reset(tqspi->dev) < 0) {
> +				dev_warn_once(tqspi->dev,
> +					      "device reset failed\n");
> +				tegra_qspi_mask_clear_irq(tqspi);
> +			}
> +			ret = -EIO;
> +			goto exit;

Vishwaroop,

It looks like somehow this hunk got messed up when you applied the patch
to your tree. The entire block above needs to remain indented like it
was. The only addition was the curly brackets and the call to
tegra_qspi_mask_clear_irq().

Thierry

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]