When a transfer has cs_change set and is not the last one of a message, tegra_spi_transfer_one_message() ends it with tegra_spi_transfer_end(). That function deasserts CS and stores the deasserted state in tspi->command1_reg (SPI_CS_SW_VAL). The next transfer is not the first one of the message, so tegra_spi_setup_transfer_one() reuses command1_reg unchanged: the controller clocks that transfer out with CS inactive. The slave never sees the data and no error is reported. Every driver that splits a message into "write payload, toggle CS, write trigger" hits this. mcp251xfd does it for every CAN frame, so an MCP2518FD never transmits: the frame reaches the chip's RAM, but the write that sets UINC|TXREQ in the TX FIFO control register is lost. Reception uses single-transfer messages and still works. Before commit f3e182c33e53 ("spi: tegra114: de-assert CS before SPI mode change") the deassertion did not touch command1_reg and the next transfer asserted CS on its own. Mark the following transfer as the first one of a message again, so that it takes the setup path that asserts CS. Tested on a reComputer Orin J40 Mini with an MCP2518FD on spi1.0. No frame was transmitted before this change; with it CAN traffic runs in both directions with no errors. Fixes: f3e182c33e53 ("spi: tegra114: de-assert CS before SPI mode change") Cc: stable@vger.kernel.org Signed-off-by: Roman Putintsev --- drivers/spi/spi-tegra114.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/spi/spi-tegra114.c b/drivers/spi/spi-tegra114.c index f43f2df..d2fda21 100644 --- a/drivers/spi/spi-tegra114.c +++ b/drivers/spi/spi-tegra114.c @@ -1116,6 +1116,8 @@ complete_xfer: } else if (xfer->cs_change) { tegra_spi_transfer_end(spi); spi_transfer_delay_exec(xfer); + /* CS was deasserted, assert it for the next transfer */ + is_first_msg = true; } } base-commit: 20eef29b7c8465ac98d7dd6d6f2ac07a8d8056ca