Re: [PATCH] kgdboc: fix return value of __setup handler

[Randy Dunlap]

Hi Doug,

On 3/8/22 08:04, Doug Anderson wrote:
> Hi,
> 
> On Mon, Mar 7, 2022 at 7:32 PM Randy Dunlap <rdunlap@xxxxxxxxxxxxx> wrote:
>>
>> __setup() handlers should return 1 to indicate that the boot option
>> has been handled. A return of 0 causes the boot option/value to be
>> listed as an Unknown kernel parameter and added to init's (limited)
>> environment strings. So return 1 from kgdboc_option_setup().
> 
> This took me about 20 minutes to trace through the code to confirm,
> but it appears you're correct. It's pretty twisted that early_param()
> and __setup(), both of which add things to the same list, work exactly
> opposite here. :( Any chance I could convince you to:
> 
> 1. Add a comment before the definition of __setup_param() explaining
> that 0 means error and 1 means no error. There's a comment next to
> early_param() that _implies_ that setup is the opposite(), but it'd be
> nice to see documentation of __setup(). I know __setup() is supposed
> to be "only for core code", but still seems like we could document it.

I have already done this. The patch is in Andrew's mmotm tree (patch queue).

> 2. Add something to your commit message helping someone find the place
> where the return value is checked. Basically just mention
> obsolete_checksetup() to give people a hint.
> 

Sure, no problem. Good idea.

> 
>> Unknown kernel command line parameters "BOOT_IMAGE=/boot/bzImage-517rc7
>>   kgdboc=kbd kgdbts=", will be passed to user space.
>>
>>  Run /sbin/init as init process
>>    with arguments:
>>      /sbin/init
>>    with environment:
>>      HOME=/
>>      TERM=linux
>>      BOOT_IMAGE=/boot/bzImage-517rc7
>>      kgdboc=kbd
>>      kgdbts=
>>
>> Fixes: 1cd25cbb2fed ("kgdboc: Fix warning with module build")
> 
> Are you certain about this "Fixes" line? That commit was just code
> motion to move the code inside the #ifdef. It sure looks like it was
> broken even before this.
> 

Yes, but I am not enough of a git user to be able to backtrack
to see where this code was added. :(
(help?)

> 
>> Signed-off-by: Randy Dunlap <rdunlap@xxxxxxxxxxxxx>
>> Reported-by: Igor Zhbanov <i.zhbanov@xxxxxxxxxxxx>
>> Link: lore.kernel.org/r/64644a2f-4a20-bab3-1e15-3b2cdd0defe3@xxxxxxxxxxxx
>> Cc: Laura Abbott <labbott@xxxxxxxxxx>
>> Cc: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
>> Cc: Jiri Slaby <jirislaby@xxxxxxxxxx>
>> Cc: kgdb-bugreport@xxxxxxxxxxxxxxxxxxxxx
>> Cc: Jason Wessel <jason.wessel@xxxxxxxxxxxxx>
>> Cc: Daniel Thompson <daniel.thompson@xxxxxxxxxx>
>> Cc: Douglas Anderson <dianders@xxxxxxxxxxxx>
>> Cc: linux-serial@xxxxxxxxxxxxxxx
>> ---
>>  drivers/tty/serial/kgdboc.c |    6 +++---
>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> --- lnx-517-rc7.orig/drivers/tty/serial/kgdboc.c
>> +++ lnx-517-rc7/drivers/tty/serial/kgdboc.c
>> @@ -403,16 +403,16 @@ static int kgdboc_option_setup(char *opt
>>  {
>>         if (!opt) {
>>                 pr_err("config string not provided\n");
>> -               return -EINVAL;
>> +               return 1;
> 
> Shouldn't it return 0 in the error cases? If __setup() functions are
> supposed to return "1" no matter what then what was the purpose of
> having a return value in the first place?

It should return 0 if the string(s) should be added to init's arg or env
strings, which is probably very rare. I don't know why it has a return
value in the first place. Someone else has already suggested that __setup()
functions should be void. Maybe they should one day, but that's a much
larger patch.

I'll send a v2.

thanks.
-- 
~Randy