On 01/25/2016 09:50 AM, Stefan Agner wrote:

Hi Marcel,

[also added Lucas]

On 2015-12-30 19:12, Marcel Ziswiler wrote:

quoted

Continuing on the Apalis iMX6 mainlining work started by Petr ?tetiar I
noticed that it only boots when rebooting from NXP's downstream 3.14.28
kernel or mainline otherwise (e.g. when cold booted) I did not get any
serial debug output at all. Enabling earlyprintk I got the following:

[    1.136806] INFO: trying to register non-static key.
[    1.136871] 2020000.serial: ttymxc0 at MMIO 0x2020000 (irq = 25,
base_baud = 5000000) is a IMX
[    1.150651] the code is fine but needs lockdep annotation.
[    1.150653] turning off the locking correctness validator.
[    1.150664] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 4.4.0-rc6-
next-20151223-00002-g6008f30-dirty #27
[    1.150668] Hardware name: Freescale i.MX6 Quad/DualLite (Device
Tree)
[    1.150675] Backtrace:
[    1.150701] [<c010c45c>] (dump_backtrace) from [<c010c654>]
(show_stack+0x18/0x1c)
[    1.150722]  r7:00000000 r6:c0d20848 r5:00000000 r4:00000000
[    1.150739] [<c010c63c>] (show_stack) from [<c03dadcc>]
(dump_stack+0x8c/0xa4)
[    1.150760] [<c03dad40>] (dump_stack) from [<c016b60c>]
(__lock_acquire+0x1d88/0x1eb4)
[    1.150774]  r7:00000000 r6:c0d063f8 r5:c155f834 r4:e503bc20
[    1.150785] [<c0169884>] (__lock_acquire) from [<c016bf54>]
(lock_acquire+0x74/0x94)
[    1.150803]  r10:c0d6eb5e r9:e597b300 r8:00000019 r7:00000001
r6:00000001 r5:60000193
[    1.150808]  r4:00000000
[    1.150821] [<c016bee0>] (lock_acquire) from [<c08d49d4>]
(_raw_spin_lock_irqsave+0x40/0x54)
[    1.150834]  r7:000052c8 r6:c0481470 r5:40000193 r4:e503bc10
[    1.150852] [<c08d4994>] (_raw_spin_lock_irqsave) from [<c0481470>]
(imx_rxint+0x20/0x2a4)
[    1.150862]  r6:00000000 r5:000052d0 r4:e503bc10
[    1.150874] [<c0481450>] (imx_rxint) from [<c0482540>]
(imx_int+0x170/0x1f4)

Debugging a little further I noticed this actually happening after
calling devm_request_irq() during serial_imx_probe() even before
uart_add_one_port()

The imx driver is in the minority by claiming the irq @ probe time.
The vast majority of uart drivers claim the irq in their startup() methods.

Regards,
Peter Hurley

quoted

did initialise the spin_lock acquired in
imx_rxint(). Turns out the data carrier detect interrupt fired
immediately upon requesting interrupts. This patch fixes this by
explicitly disabling data carrier detect before requesting interrupts.

I actually hit a very similar issue on the Colibri iMX6 and hence looked
into this too. In our downstream tree, I fixed it in imx_startup, which
is probably the wrong place actually...

quoted

Signed-off-by: Marcel Ziswiler <redacted>

---

 drivers/tty/serial/imx.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/tty/serial/imx.c b/drivers/tty/serial/imx.c
index 9362f54c..b1588f9 100644
--- a/drivers/tty/serial/imx.c
+++ b/drivers/tty/serial/imx.c

@@ -2032,6 +2032,11 @@ static int serial_imx_probe(struct platform_device *pdev)
 		 UCR1_TXMPTYEN | UCR1_RTSDEN);
 	writel_relaxed(reg, sport->port.membase + UCR1);
 
+	/* Disable data carrier detect before requesting interrupts */
+	reg = readl_relaxed(sport->port.membase + UCR3);
+	reg &= ~(UCR3_DCD);
+	writel_relaxed(reg, sport->port.membase + UCR3);
+

Note that UCR3_DCD has a different meaning in DCE mode. As a DCE, the
bit should be kept 1 (we assume that the carrier is detected by
default...)

The dte_mode bit should be initialized at that point, hence just making
that code conditional should account for that.

What I also noticed is that the RI signal suffers the very same issue.
We do not use the signal on our Colibri board, but we should disable
that interrupt too.

--
Stefan

quoted

 	clk_disable_unprepare(sport->clk_ipg);
 
 	/*