Re: Aw: Re: serial console broken in v3.17-rc6 ?

From: Peter Hurley
Date: Wed Oct 01 2014 - 11:35:50 EST


On 10/01/2014 10:46 AM, Helge Deller wrote:
> Hi Peter,
>
>>> Serial: 8250/16550 driver, 4 ports, IRQ sharing enabled
>>> serial8250: ttyS0 at I/O 0x3f8 (irq = 3, base_baud = 115200) is a 16550A
>>>
>>> The source code for this driver is in drivers/parisc/superio.c,
>>> see e.g. function superio_serial_init().
>>> Maybe something is missing in here which was done to the other serial drivers?
>>
>> Ok. So still the 8250 driver.
>>
>> What are the remote line settings? (speed/parity/bits/stop)
>
> 9600,8,N,1
>
>> See if reverting commit ae84db9661cafc63d179e1d985a2c5b841ff0ac4,
>> 'serial: core: Preserve termios c_cflag for console resume', has some effect.
>> I'm not seeing how this commit would affect your setup, but if it does, I can
>> provide a debug patch to find out what userspace is doing.
>
> No, this did not fixed it.
>
>> If not, then I think you'll need to bisect drivers/tty/serial. I tried to
>> analyze if the new generic earlycon was somehow triggering but I got lost
>> multiple times in the static analysis.
>>
>> BTW, since superio_serial_init() assigns the port type as PORT_16550A,
>> superio_serial_init() should be doing:
>>
>> diff --git a/drivers/parisc/superio.c b/drivers/parisc/superio.c
>> index a042d06..36b3adb 100644
>> --- a/drivers/parisc/superio.c
>> +++ b/drivers/parisc/superio.c
>> @@ -395,7 +395,7 @@ static void __init superio_serial_init(void)
>> serial_port.iotype = UPIO_PORT;
>> serial_port.type = PORT_16550A;
>> serial_port.uartclk = 115200*16;
>> - serial_port.fifosize = 16;
>> + serial_port.flags = UPF_FIXED_TYPE;
>>
>> /* serial port #1 */
>> serial_port.iobase = sio_dev.sp1_base;
>>
>> to properly initialize the 8250 port definition. Right now, the fifosize
>> has no effect because the fifo capabilities is not on.
>>
>> Also, consider adding
>>
>> - serial_port.flags = UPF_FIXED_TYPE;
>> + serial_port.flags = UPF_FIXED_PORT | UPF_FIXED_TYPE;
>>
>> to disable reprogramming the irq and ioport from setserial.
>
> That was partly solving the problem!
> By looking at the other drivers, I found that they set the UPF_BOOT_AUTOCONF flag too.
> So, changing it to
> - serial_port.fifosize = 16;
> + serial_port.flags = UPF_FIXED_PORT | UPF_FIXED_TYPE | UPF_BOOT_AUTOCONF;
> solved the problem :-)
>
> Might this be the right fix? If yes, I can push it upstream.

Yes, that's one possible correct fix.

I didn't notice that the superio probe code did not claim the io region for
the serial ports; that's what UPF_BOOT_AUTOCONF does for you.

The regression must be from a change on the parisc arch regarding how/whether io
ports are mapped into the vm.


> All those flags like UPF_BOOT_AUTOCONF seem to be poorly documented btw.

Yeah, you're right about that. It's part of the legacy of one of the oldest
subsystems that's changed and grown by fits and starts; not all of the parts
are completely understood by one person.


>>> Interestingly the *same* kernel don't show this problem on another (older) parisc
>>> machine, but this older machine uses another serial driver:
>>> This (working) machine is using the drivers/tty/serial/8250/8250_gsc.c driver.
>>> Serial: 8250/16550 driver, 4 ports, IRQ sharing enabled
>>> 2:0:4: ttyS0 at MMIO 0xf0105800 (irq = 18, base_baud = 454545) is a 16550A
>>> 5:0:2: ttyS1 at MMIO 0xf0202800 (irq = 25, base_baud = 454545) is a 16550A
>>
>> PS - Unrelated note. This base_baud is junk ---------------^
>
> Is it?
> The base_baud value is calculated as "port->uartclk / 16" in drivers/tty/serial/serial_core.c.
>
> Looking at the machine specific driver code we have:
> +++ b/drivers/tty/serial/8250/8250_gsc.c
> @@ -56,12 +56,12 @@ static int __init serial_init_chip(struct parisc_device *dev)
> /* 7.272727MHz on Lasi. Assumed the same for Dino, Wax and Timi. */
> uart.port.uartclk = (dev->id.sversion != 0xad) ? /* XX */
> 7272727 : 1843200;
>
> which comes from the description of this chip which can be found in section 7.4 in this PDF file:
> https://parisc.wiki.kernel.org/images-parisc/7/79/Lasi_ers.pdf (page "64 of 111" or 76):
> 7.4 Differences from NS16550A
> The Lasi serial port is intended to function just like the real National NS16550A. This includes
> behavior that often seems rather stupid. The National NS16550A was chosen over the WD16C552
> because the National part came first, and the WD part is supposed to be compatible. There are a few
> minor differences, however, between the NS16550A and this implementation.
> Baud clock generation is an area of slight differences from the NS16550A. The baudrate reference
> frequency is taken from the 40 MHz IO system clock. The frequency 7.2727 MHz is generated by
> dividing 40 MHz by 5.5.
>
> So, probably the calculation of "port->uartclk / 16" is wrong for this chip?

Wow. No, I was wrong; it's the correct base_baud. I hadn't expected that uartclk value :)

Regards,
Peter Hurley
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/