From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S934049AbcAYRwy (ORCPT ); Mon, 25 Jan 2016 12:52:54 -0500 Received: from mail.kmu-office.ch ([178.209.48.109]:47353 "EHLO mail.kmu-office.ch" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S933852AbcAYRwv (ORCPT ); Mon, 25 Jan 2016 12:52:51 -0500 MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Date: Mon, 25 Jan 2016 09:50:33 -0800 From: Stefan Agner To: Marcel Ziswiler Cc: linux-arm-kernel@lists.infradead.org, Greg Kroah-Hartman , linux-kernel@vger.kernel.org, linux-serial@vger.kernel.org, Jiri Slaby , =?UTF-8?Q?Petr_=C5=A0tetiar?= , Lucas Stach Subject: Re: [PATCH] serial: imx: fix dcd interrupt firing during probe In-Reply-To: <1451531565-1233-1-git-send-email-marcel.ziswiler@toradex.com> References: <1451531565-1233-1-git-send-email-marcel.ziswiler@toradex.com> Message-ID: User-Agent: Roundcube Webmail/1.1.3 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Marcel, [also added Lucas] On 2015-12-30 19:12, Marcel Ziswiler wrote: > Continuing on the Apalis iMX6 mainlining work started by Petr Štetiar 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] [] (dump_backtrace) from [] > (show_stack+0x18/0x1c) > [ 1.150722] r7:00000000 r6:c0d20848 r5:00000000 r4:00000000 > [ 1.150739] [] (show_stack) from [] > (dump_stack+0x8c/0xa4) > [ 1.150760] [] (dump_stack) from [] > (__lock_acquire+0x1d88/0x1eb4) > [ 1.150774] r7:00000000 r6:c0d063f8 r5:c155f834 r4:e503bc20 > [ 1.150785] [] (__lock_acquire) from [] > (lock_acquire+0x74/0x94) > [ 1.150803] r10:c0d6eb5e r9:e597b300 r8:00000019 r7:00000001 > r6:00000001 r5:60000193 > [ 1.150808] r4:00000000 > [ 1.150821] [] (lock_acquire) from [] > (_raw_spin_lock_irqsave+0x40/0x54) > [ 1.150834] r7:000052c8 r6:c0481470 r5:40000193 r4:e503bc10 > [ 1.150852] [] (_raw_spin_lock_irqsave) from [] > (imx_rxint+0x20/0x2a4) > [ 1.150862] r6:00000000 r5:000052d0 r4:e503bc10 > [ 1.150874] [] (imx_rxint) from [] > (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() 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... > > Signed-off-by: Marcel Ziswiler > > --- > > 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 > clk_disable_unprepare(sport->clk_ipg); > > /*