This is an automated email from the ASF dual-hosted git repository.
xiaoxiang pushed a commit to branch master
in repository https://gitbox.apache.org/repos/asf/nuttx.git
commit 05e620d12bad8a7b6915f5b3ba100b28df5b733c
Author: David Sidrane <david.sidr...@nscdg.com>
AuthorDate: Mon Nov 20 08:16:03 2023 -0800
imxrt:Serial Use smart invalidate
---
arch/arm/src/imxrt/imxrt_serial.c | 55 ++++++++++++++++++++++++++++++++++-----
1 file changed, 48 insertions(+), 7 deletions(-)
diff --git a/arch/arm/src/imxrt/imxrt_serial.c
b/arch/arm/src/imxrt/imxrt_serial.c
index 6926cd8da5..ec8ee84abd 100644
--- a/arch/arm/src/imxrt/imxrt_serial.c
+++ b/arch/arm/src/imxrt/imxrt_serial.c
@@ -770,6 +770,10 @@ struct imxrt_uart_s
DMACH_HANDLE rxdma; /* currently-open receive DMA stream */
bool rxenable; /* DMA-based reception en/disable */
uint32_t rxdmanext; /* Next byte in the DMA buffer to be read
*/
+#ifdef CONFIG_ARMV7M_DCACHE
+ uint32_t rxdmaavail; /* Number of bytes available without need
to
+ * to invalidate the data cache */
+#endif
char *const rxfifo; /* Receive DMA buffer */
#endif
};
@@ -2086,6 +2090,9 @@ static int imxrt_dma_setup(struct uart_dev_s *dev)
*/
priv->rxdmanext = 0;
+#ifdef CONFIG_ARMV7M_DCACHE
+ priv->rxdmaavail = 0;
+#endif
/* Enable receive Rx DMA for the UART */
@@ -2907,7 +2914,6 @@ static bool imxrt_rxflowcontrol(struct uart_dev_s *dev,
static int imxrt_dma_receive(struct uart_dev_s *dev, unsigned int *status)
{
struct imxrt_uart_s *priv = (struct imxrt_uart_s *)dev;
- static uint32_t last_nextrx = -1;
uint32_t nextrx = imxrt_dma_nextrx(priv);
int c = 0;
@@ -2915,17 +2921,49 @@ static int imxrt_dma_receive(struct uart_dev_s *dev,
unsigned int *status)
if (nextrx != priv->rxdmanext)
{
- /* Now we must ensure the cache is updated if the DMA has
- * updated again.
+#ifdef CONFIG_ARMV7M_DCACHE
+ /* If the data cache is enabled, then we will also need to manage
+ * cache coherency. Are any bytes available in the currently coherent
+ * region of the data cache?
*/
- if (last_nextrx != nextrx)
+ if (priv->rxdmaavail == 0)
{
- up_invalidate_dcache((uintptr_t)priv->rxfifo,
- (uintptr_t)priv->rxfifo + RXDMA_BUFFER_SIZE);
- last_nextrx = nextrx;
+ uint32_t rxdmaavail;
+ uintptr_t addr;
+
+ /* No.. then we will have to invalidate additional space in the Rx
+ * DMA buffer.
+ */
+
+ if (nextrx > priv->rxdmanext)
+ {
+ /* Number of available bytes */
+
+ rxdmaavail = nextrx - priv->rxdmanext;
+ }
+ else
+ {
+ /* Number of available bytes up to the end of RXDMA buffer */
+
+ rxdmaavail = RXDMA_BUFFER_SIZE - priv->rxdmanext;
+ }
+
+ /* Invalidate the DMA buffer range */
+
+ addr = (uintptr_t)&priv->rxfifo[priv->rxdmanext];
+ up_invalidate_dcache(addr, addr + rxdmaavail);
+
+ /* We don't need to invalidate the data cache for the next
+ * rxdmaavail number of next bytes.
+ */
+
+ priv->rxdmaavail = rxdmaavail;
}
+ priv->rxdmaavail--;
+#endif
+
/* Now read from the DMA buffer */
c = priv->rxfifo[priv->rxdmanext];
@@ -2990,6 +3028,9 @@ static void imxrt_dma_reenable(struct imxrt_uart_s *priv)
*/
priv->rxdmanext = 0;
+#ifdef CONFIG_ARMV7M_DCACHE
+ priv->rxdmaavail = 0;
+#endif
/* Start the DMA channel, and arrange for callbacks at the half and
* full points in the FIFO. This ensures that we have half a FIFO
