[PATCH] serial: imx: Only enable UART DMA on i.MX6Q and i.MX53

On Sun, Feb 05, 2017 at 02:36:10PM -0200, Fabio Estevam wrote:

On Sun, Feb 5, 2017 at 12:56 PM, Fabio Estevam [off-list ref] wrote:

quoted

[Sorry, for the top-posting]


imx6sx uart should be of a IMX6Q_UART instead.


Does this patch fix the issue?

It seems this dts change will not fix the problem.

quoted

From: Kai Ruhnau <redacted>

In 1c06bde6 with the introduction of DMA support for the i.MX53 platform,
DMA was enabled for all platforms instead. On an i.MX6SX (which is a
IMX21_UART), this breaks gpsd talking to an UART module with

 imx-uart 21f4000.serial: DMA transaction error.

This patch makes sure, UART DMA is only activated for i.MX53 and i.MX6Q.

Signed-off-by: Kai Ruhnau <redacted>
Reviewed-by: Marek Vasut <marex@denx.de>
Fixes: 1c06bde643d0 ("Allowing UART DMA to be configured on i.MX53")
Cc: <redacted> # 4.9+
Cc: Fabio Estevam <redacted>
Cc: Martyn Welch <martyn.welch@collabora.co.uk>
Cc: Shawn Guo <shawnguo@kernel.org>
---
 drivers/tty/serial/imx.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/tty/serial/imx.c b/drivers/tty/serial/imx.c
index a70356d..8e2b5f2 100644
--- a/drivers/tty/serial/imx.c
+++ b/drivers/tty/serial/imx.c

@@ -1269,7 +1269,8 @@ static int imx_startup(struct uart_port *port)
         writel(temp & ~UCR4_DREN, sport->port.membase + UCR4);

         /* Can we enable the DMA support? */
-       if (!uart_console(port) && !sport->dma_is_inited)
+       if ((is_imx6q_uart(sport)  || is_imx53_uart(sport)) &&
+           !uart_console(port) && !sport->dma_is_inited)
                 imx_uart_dma_init(sport);

Looks like imx serial DMA is broken in 4.9.

There are probably two separate problems:

A)
I also observed the DMA transaction error message when I used the ROM
SDMA scripts. Seems to affect single- and multi-core systems and is
fixed by: github.com/Freescale/linux-fslc/commit/a2ec8df37f5d (?)
This is also not RS-485 specific, but affects RS-232 too.

B)
The RS-485 DMA problem I reported, which only affects multi-core
systems.
RS-232 DMA is not affected.

There is also this report from Clemens: https://lkml.org/lkml/2017/1/4/579

Could anyone run a git bisect?

I reverted tty/serial/imx.c to the first implementation of RS-485,
namely to commit 17b8f2a3fdca ("serial: imx: add support for half duplex
rs485"). Also had to add "fsl," in front of uart-has-rts-cts in the DT.
But it does not work either (same behavior). Looks like RS-485 DMA never
worked on SMP systems..?

It's still a mistery to me what is causing the xmit->tail index to walk
over the xmit->head index, leading to the rest of the whole
UART_XMIT_SIZE buffer to be sent out. But it must be connected to the
transmit-complete interrupt, because if I disable it, this does not
happen.

Thanks,
Clemens