In current implementation if some data still exists in Tx FIFO it can be silently flushed, i.e. dropped on disabling of the controller, which happens when writing 0 to DW_SPI_SSIENR (it happens in the beginning of new transfer)
So add wait for current transmit operation to complete to be sure that current transmit operation is finished before new one.
Signed-off-by: Eugeniy Paltsev Eugeniy.Paltsev@synopsys.com --- Changes v1->v2: * Use readl_poll_timeout macros instead of custom code.
drivers/spi/designware_spi.c | 15 +++++++++++++++ 1 file changed, 15 insertions(+)
diff --git a/drivers/spi/designware_spi.c b/drivers/spi/designware_spi.c index c501aeea16..5e196b21c9 100644 --- a/drivers/spi/designware_spi.c +++ b/drivers/spi/designware_spi.c @@ -18,6 +18,7 @@ #include <spi.h> #include <fdtdec.h> #include <linux/compat.h> +#include <linux/iopoll.h> #include <asm/io.h>
DECLARE_GLOBAL_DATA_PTR; @@ -342,6 +343,7 @@ static int dw_spi_xfer(struct udevice *dev, unsigned int bitlen, u8 *rx = din; int ret = 0; u32 cr0 = 0; + u32 val; u32 cs;
/* spi core configured to do 8 bit transfers */ @@ -394,6 +396,19 @@ static int dw_spi_xfer(struct udevice *dev, unsigned int bitlen, /* Start transfer in a polling loop */ ret = poll_transfer(priv);
+ /* + * Wait for current transmit operation to complete. + * Otherwise if some data still exists in Tx FIFO it can be + * silently flushed, i.e. dropped on disabling of the controller, + * which happens when writing 0 to DW_SPI_SSIENR which happens + * in the beginning of new transfer. + */ + if (readl_poll_timeout(priv->regs + DW_SPI_SR, val, + !(val & SR_TF_EMPT) || (val & SR_BUSY), + RX_TIMEOUT * 1000)) { + ret = -ETIMEDOUT; + } + return ret; }