Date:	Tue, 12 Jan 2016 13:18:25 -0800
From:	Peter Hurley <peter@...leysoftware.com>
To:	Sebastian Frias <sf84@...oste.net>
Cc:	Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
	linux-serial@...r.kernel.org, LKML <linux-kernel@...r.kernel.org>,
	mason <slash.tmp@...e.fr>,
	Måns Rullgård <mans@...sr.com>
Subject: Re: [RFC PATCH] always probe UART HW when options are not specified

Hi Sebastian,

On 01/12/2016 06:14 AM, Sebastian Frias wrote:
> On 01/11/2016 08:06 PM, Peter Hurley wrote:
>> On 01/11/2016 09:56 AM, Sebastian Frias wrote:
>>> On 01/11/2016 05:11 PM, Peter Hurley wrote:
>>>> On 01/11/2016 07:07 AM, Sebastian Frias wrote:
>>>>> On 12/22/2015 06:56 PM, Sebastian Frias wrote:
>>>>>>
>>>>>>        OF_EARLYCON_DECLARE(rt2880, "ralink,rt2880-uart",
>>>>>> early_serial8250_setup);
>>>>
>>>> There is no support for this uart in 8250 earlycon; the registers
>>>> need remapped.
>>>
>>> Ok, two questions then:
>>> 1) If the UART is not supported on 8250 earlycon, what is the
>>> suggested/advised solution? Using just "earlyprintk"?
>>
>> I don't have enough information to suggest what you "should" use
>> here.
>>
>> Is this going to be a shipping product?
>> Is it single-core?
>> etc.
>>
>> And what is your purpose for outputting early boot information
>> before loading the serial driver which does provide console output?
>>
> 
> No, it is not for production, just for debug, but we would like to understand if there is a standard way of doing so, so that whenever somebody ask us for "early print", we can provide with the good way.
> I know we can always provide with hacks, I'm just wondering if there's a "standard way".

Ok.

The "standard" solution is either command-line or DT earlycon (which
we've established isn't working for your setup).

Let's fix that so that both of those options work.


>>> 2) What would it take to make the "rt2880" work with the 8250
>>> earlycon? I mean, it is already pretty much supported in there, what
>>> would be missing? (I don't see why it blocks on earlycon_map) And
>>> would it be worth doing?
>>
>> The rt2880 does not have the same register locations as a 8250.
>> The 8250 port driver remaps all register accesses with a LUT.
>>
>> Adding support would be trivial.
> 
> Ok, I will see if I can find some commit that does something similar to get some inspiration.

I sent you a patch yesterday that adds earlycon support for
this uart (so command line should work), plus the series I sent today
should get DT earlycon up and running. Please test both.


>>>>>> at the end of the file, trying to mimic commit
>>>>>> d05f15707bb7659d2b863fafa1a918f286d74a63
>>>>>>
>>>>>> I'm still trying to figure out the right bootargs, so that's why both
>>>>>> "earlycon" and "console" are there. Suggestions welcome.
>>>>
>>>> Just 'earlycon' triggers the attempted registration of earlycon matching the
>>>> compatible string of the stdout-path node.
>>>>
>>>> The empty 'console' in bootargs is doing nothing.
>>>
>>> Ok, thanks.
>>>
>>> So, just to recap.
>>> We would like to understand what is the right way of doing this:
>>>
>>> - we are using 8250 (rt288x variant) UART: CONFIG_SERIAL_8250_RT288X=y
>>> - the UART hardware is setup prior to Linux boot
>>> - we don't want Linux to change the UART settings, just to pick up
>>> whatever settings the UART has and take over the UART.
>>> There were two replies to that, one by Greg Kroah-Hartman
>>> (http://www.spinics.net/lists/linux-serial/msg20278.html) and one by
>>> you, where you suggested we use "console=uart", but as I reported
>>> (http://www.spinics.net/lists/linux-serial/msg20307.html) it does not
>>> work, you replied that iotype and mmio are not optional but mandatory
>>> (http://www.spinics.net/lists/linux-serial/msg20310.html), and I
>>> wondered if it was really necessary to duplicate data that is already
>>> on the DT among other questions
>>
>> At the time, I didn't know you were describing your h/w with DT.
> 
> Oh, sorry for the inconvenience then.

No big deal, I was just explaining the earlier advice.


>> If you use the console command line (console= or earlycon=) to start
>> an earlycon, then the uart address and iotype are mandatory.
>> For this usage, earlycon matching is attempted with every EARLYCON_DECLARE().
>>
>> If you use plain "earlycon" on the command line, that will attempt to
>> register the uart described by stdout-path in DT. For this usage,
>> earlycon matching is attempted with the compatible string of every
>> OF_EARLYCON_DECLARE().
> 
> I see, and since rt288x variant is not fully supported (no OF_EARLYCON_DECLARE) "earlycon" fails.

Yes.

>>> (http://www.spinics.net/lists/linux-serial/msg20383.html), like how
>>> are DT described drivers supposed to interact with the
>>> "console="/"earlycon=" commandlines
>>
>> They don't; those are orthogonal.
> 
> Ok.
> 
>>
>>> , or, the contradiction between
>>> "console=ttyS0" means '9600n81' and "if unspecified [the uart
>>> options], the h/w is not re-initialized">
>>
>> I thought I was clear on that: "console=ttyS0" initializes the h/w
>> to 9600n81 *because there are already existing users that must not break*.
>>
>> "console=uart,..." probes the h/w
>> *because there are already existing users that must not break *
> 
> Thanks, I had misunderstood.
> 
>>
>>> So, for us, it is still not clear what is the recommended way of
>>> achieving our goal above, and it seems it is not clear what does
>>> "console=ttyS0" is supposed to do, hardcode ('9600n81') or probe
>>> ('the h/w is not re-initialized')
>>
>> The DT way will be simplest at this point because you won't
>> have to write console handover matching for "console=rt288x,..."
>>
>> With DT (ie, stdout-path) earlycon, when a serial driver loads,
>> an attempt is made to cross-reference any existing console with
>> the node that is loading and will do a console takeover from
>> a running earlycon for a matching uart node.
>>
>> There is a bug with DT earlycon though.
>> If you have a dummy console that loads, the DT earlycon is
>> disabled at that point because boot consoles are disabled when
>> "real" consoles load.
> 
> I'm sorry for my ignorance, but what is a "dummy console"? Under what circumstances would it load and this bug be seen?

The dummy console is a hack used to preserve the priority of the
display console in the event no console is specified.

Here's an example:
[    0.000000] Booting Linux on physical CPU 0x0
[    0.000000] Initializing cgroup subsys cpuset
[    0.000000] Initializing cgroup subsys cpu
[    0.000000] Initializing cgroup subsys cpuacct
[    0.000000] Linux version 4.4.0-rc6+ (peter@...r) (gcc version 4.8.2 (Ubuntu/Linaro 4.8.2-16ubuntu4) ) #86 PREEMPT Tue Jan 12 12:30:39 PST 2016
[    0.000000] CPU: ARMv7 Processor [413fc082] revision 2 (ARMv7), cr=10c5387d
[    0.000000] CPU: PIPT / VIPT nonaliasing data cache, VIPT aliasing instruction cache
[    0.000000] Machine model: TI AM335x BeagleBone Black
[    0.000000] earlycon: omap8250 at MMIO 0x44e09000 (options '')
[    0.000000] bootconsole [omap8250] enabled
[    0.000000] cma: Reserved 16 MiB at 0x9f000000
[    0.000000] Memory policy: Data cache writeback
[    0.000000] CPU: All CPU(s) started in SVC mode.
[    0.000000] AM335X ES2.1 (sgx neon )
[    0.000000] Built 1 zonelists in Zone order, mobility grouping on.  Total pages: 129920
[    0.000000] Kernel command line: earlycon root=/dev/mmcblk0p2 ro fixrtc rootfstype=ext4 rootwait
[    0.000000] PID hash table entries: 2048 (order: 1, 8192 bytes)
[    0.000000] Dentry cache hash table entries: 65536 (order: 6, 262144 bytes)
[    0.000000] Inode-cache hash table entries: 32768 (order: 5, 131072 bytes)
[    0.000000] Memory: 486332K/524288K available (10582K kernel code, 862K rwdata, 3396K rodata, 596K init, 871K bss, 21572K reserved, 16384K cma-reserved, 0K highmem)
[    0.000000] Virtual kernel memory layout:
[    0.000000]     vector  : 0xffff0000 - 0xffff1000   (   4 kB)
[    0.000000]     fixmap  : 0xffc00000 - 0xfff00000   (3072 kB)
[    0.000000]     vmalloc : 0xe0800000 - 0xff800000   ( 496 MB)
[    0.000000]     lowmem  : 0xc0000000 - 0xe0000000   ( 512 MB)
[    0.000000]     pkmap   : 0xbfe00000 - 0xc0000000   (   2 MB)
[    0.000000]     modules : 0xbf000000 - 0xbfe00000   (  14 MB)
[    0.000000]       .text : 0xc0008000 - 0xc0daebf0   (13979 kB)
[    0.000000]       .init : 0xc0daf000 - 0xc0e44000   ( 596 kB)
[    0.000000]       .data : 0xc0e44000 - 0xc0f1bb3c   ( 863 kB)
[    0.000000]        .bss : 0xc0f1e000 - 0xc0ff7d84   ( 872 kB)
[    0.000000] Preemptible hierarchical RCU implementation.
[    0.000000]  Build-time adjustment of leaf fanout to 32.
[    0.000000] NR_IRQS:16 nr_irqs:16 16
[    0.000000] IRQ: Found an INTC at 0xfa200000 (revision 5.0) with 128 interrupts
[    0.000000] OMAP clockevent source: timer2 at 24000000 Hz
[    0.000014] sched_clock: 32 bits at 24MHz, resolution 41ns, wraps every 89478484971ns
[    0.008083] clocksource: timer1: mask: 0xffffffff max_cycles: 0xffffffff, max_idle_ns: 79635851949 ns
[    0.017570] OMAP clocksource: timer1 at 24000000 Hz
[    0.022969] Console: colour dummy device 80x30
[    0.027570] console [tty0] enabled
[    0.031100] bootconsole [omap8250] disabled
Ubuntu 14.04.3 LTS black ttyS0

black login:


As you can see, there's boot console but no regular console until
getty spawns.

I've fixed the problem once, but that broke some setups and had to be
reverted so I need to rethink an alternate approach. A unified, coherent
solution is elusive because of the many different possible console setups.

Regards,
Peter Hurley

Powered by blists - more mailing lists