Re: [Kgdb-bugreport] [PATCH 4/5] KGDB-8250: refactor configuration

[Jan Kiszka <jan.kiszka@web.de>]

Sergei Shtylyov wrote:
> Hello.
> 
> Jan Kiszka wrote:
> 
>> Sorry, previous version was missing some __init[data] attributes which
>> were dropped in an intermediate stage. Here comes an updated patch:
> 
>> <---snip--->
> 
>> This major refactoring of the quite complex kgdb8250 configuration does
>> the following:
> 
>>  - ensures that static configurations according to SERIAL_PORT_DFNS are
>>    always loaded first
>>  - tries to pull more accurate configuration via serial8250_get_port_def
>>    if simple-config is used
>>  - detects empty/invalid simple-configs
>>  - enforces KGDB_PORT_NUM <= SERIAL_8250_NR_UARTS at kconfig level
>>  - removes kgdb8250_add_port and its hook in serial_core (calling
>>    serial8250_get_port_def in demand should provide us the same
>>    information)
> 
>    You left powerpc-lite.patch broken with this change as it has
> multiple calls to kgdb8250_add_port()...

I see. But I wonder if there ever was a real need for these hooks (in
2.4 times?): If I look at bamboo_early_serial_map() e.g., I find it
calling into early_serial_setup() which fills serial8250_ports[] - and
that content is now retrieved via serial8250_get_port_def() when we
parse the runtime or build-time provided parameters (port number &
baudrate).

> 
>> Signed-off-by: Jan Kiszka <jan.kiszka@web.de>
> 
>> Index: b/drivers/serial/serial_core.c
>> ===================================================================
>> --- a/drivers/serial/serial_core.c
>> +++ b/drivers/serial/serial_core.c
> [...]
>> @@ -2370,12 +2369,6 @@ int uart_add_one_port(struct uart_driver
>>       */
>>      port->flags &= ~UPF_DEAD;
>>  
>> -#if defined(CONFIG_KGDB_8250)
>> -    /* Add any 8250-like ports we find later. */
>> -    if (port->type <= PORT_MAX_8250)
>> -        kgdb8250_add_port(port->line, port);
>> -#endif
>> -
> 
>    I'm afraid this wasn't correct from the very start since this can add
> ports with .iotype that 8250_kgdb.c does not support. So, nothing to
> regret here...

I think a lot of cruft piled up in the kgdb patches over their long life. :)

Thanks for your feedback!

Jan