On 04/19/2016 01:37 AM, Matwey V. Kornilov wrote: > 2016-04-19 6:01 GMT+03:00 Peter Hurley <peter@xxxxxxxxxxxxxxxxxx>: >> On 04/18/2016 07:18 AM, Matwey V. Kornilov wrote: >>> 2016-04-18 9:55 GMT+03:00 Jiri Slaby <jslaby@xxxxxxx>: >>>> On 04/17/2016, 02:33 PM, Matwey V. Kornilov wrote: >>>>> Hello, >>>>> >>>>> When running 4.6-rc3 (please find used config attached) on BeagleBone Black with 8250_omap serial driver I see the following hung. >> > > Hello, > >> What uart # (0-5) are you using? > > I am using uart5 (ttyS5). > >> Are you using not-yet-mainline patches to 8250 driver (like for gpio modem control)? > > No, I use vanilla 4.6-rc3 (and 4.5 behaviour is the same). > >> >> >>>>> When test application a.out run twice (test.c source code attached here) the following happens: >>>>> >>>>> # ./a.out # ./a.out >>>>> (hung here, may be interrupted with Ctrl-C) >>>>> >>>>> Trace from echo t > /proc/sysrq-trigger is the following: >>>>> >>>>> апр 17 12:29:23 nohostname kernel: a.out S c0c3544c 0 922 882 0x00000000 >>>>> апр 17 12:29:23 nohostname kernel: [<c0c3544c>] (__schedule) from [<c0c3597c>] (schedule+0x5c/0xd0) >>>>> апр 17 12:29:23 nohostname kernel: [<c0c3597c>] (schedule) from [<c0794820>] (tty_port_block_til_ready+0x11c/0x2a0) >>>>> апр 17 12:29:23 nohostname kernel: [<c0794820>] (tty_port_block_til_ready) from [<c07abfb0>] (uart_open+0x11c/0x158) >>>> ... >>>>> >>>>> I don't think that open() user-space call should be blocked under such circumstances, so I think that this is rather an issue within the kernel. >>>>> I've found that the effect dissapears when termios calls are dropped from test.c >> >> Assuming you are _not_ using gpio modem control, this shouldn't happen >> because the driver should report TIOCM_CAR always on if DCD is not >> pinned out for exactly this reason. > > As Jiri said, CLOCAL prevents open() from blocking. However, my uart > doesn't have any modem control pins in device tree only rts/cts. Yes CLOCAL prevents waiting for carrier, but regardless, if the h/w does not have DCD then the driver should always return TIOCM_CAR. I tested this last night on a beaglebone, and the MSR read always returns not asserted for DCD, so the 8250_omap driver should be fixing this to TIOCM_CAR. Unfortunately, the situation is not that simple because uart1 can have modem control signals (including DCD) so the lack of pins will need to be described by DT. Regards, Peter Hurley >> >> IOW, could be driver bug. >> >> Regards, >> Peter Hurley >> >> >>>> I suppose the memset in the program cleared CLOCAL and the subsequent >>>> open waits for carrier... >>>> >>> >>> The same program doesn't hang when fdti_sio based /dev/ttyUSB is being used. >>> >>>> thanks, >>>> -- >>>> js >>>> suse labs >>>> >>> >>> >>> >> > > > -- To unsubscribe from this list: send the line "unsubscribe linux-serial" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
