Problem with 4.6-rc2

On 18/04/2016 21:23, Jean Delvare wrote:

Hi Peter, Mason,

On Mon, 18 Apr 2016 20:29:47 +0200, Mason wrote:

quoted

On 18/04/2016 20:06, Peter Hurley wrote:

quoted

On 04/18/2016 09:24 AM, Mason wrote:

quoted

I'm running into this panic. I will take a closer look tomorrow.
Any ideas?

commit 8d2acdb9fc3a544ab0442634531834d6007b5467
Author: Jean Delvare [off-list ref]
Date:   Mon Feb 22 09:00:39 2016 +0100

    serial: 8250: Add hardware dependency to RT288X option
    
    Kconfig option SERIAL_8250_RT288X seems to be only relevant on MIPS
    platforms, so do not present it on other architectures, unless
    build-testing.
    
    Signed-off-by: Jean Delvare [off-list ref]
    Cc: Mans Rullgard [off-list ref]
    Cc: Jiri Slaby [off-list ref]
    Acked-by: John Crispin [off-list ref]
    Signed-off-by: Greg Kroah-Hartman [off-list ref]


Not sure why Greg picked this up over my objections though

Greg applied v1 of my patch. In v2 I added ARCH_TANGO as a possible
dependency, but Greg did pick the update. Mason, is your target machine
an ARCH_TANGO machine, or something else?

Correct, ARCH_TANGO.
http://lxr.free-electrons.com/source/arch/arm/mach-tango/Kconfig

quoted

Peter,

Thanks for pointing out the problem.

Mason, did you check if reverting this commit and re-enabling
SERIAL_8250_RT288X actually solves your problem?

I will test tomorrow. But I'm quite confident that enabling
SERIAL_8250_RT288X will make the problem go away (see below).

quoted

My ARM-based SoC uses this hardware (Palmchip BK-3103) and it
now panics on boot.

Can we revert 8d2acdb9fc3a in time for v4.6?

If not selecting SERIAL_8250_RT288X results in a crash at boot on some
systems, then reverting my commit is not the proper fix. Even after
reverting, you can still omit selecting the option, and get the same
crash. In other words, my commit is not introducing the crash, it must
have been there lurking before.

I wanted to have
select SERIAL_8250_RT288X if SERIAL_8250
in my platform Kconfig, but Arnd shot that down :-(
(What good is a SoC without a console?)

http://thread.gmane.org/gmane.linux.ports.arm.kernel/444131/focus=444197

"Picking SERIAL_8250 but not SERIAL_8250_RT288X makes the kernel
panic. Doesn't that qualify for selecting it?"

The problem has existed for a while.

Whatever code crashes in this case should be made more robust to
properly deal with the situation. Or if it is too much work or too ugly
(not being familiar with the code, I have no idea), then we could
finally go with Peter's earlier proposal of dropping the
SERIAL_8250_RT288X option altogether and unconditionally including the
code in question. In fact I think Peter was supposed to send a patch
doing exactly that.

Regards.