Re: [PATCH V3] PR gdb/33747: gdb/ser-unix: modernize Linux custom baud rate support

[Sunil Kumar Dora <sunilkumar.dora@windriver.com>]

Thanks Maciej.

On 3/20/2026 6:54 AM, Maciej W. Rozycki wrote:
> CAUTION: This email comes from a non Wind River email account!
> Do not click links or open attachments unless you recognize the sender and know the content is safe.
>
> On Thu, 12 Mar 2026, sunilkumar.dora@windriver.com wrote:
>
>> diff --git a/gdb/ser-unix.c b/gdb/ser-unix.c
>> index c295a9c5ba1..a15f4cc2162 100644
>> --- a/gdb/ser-unix.c
>> +++ b/gdb/ser-unix.c
>> @@ -510,36 +510,59 @@ set_baudcode_baudrate (struct serial *scb, int baud_code)
>>
>>   #if HAVE_CUSTOM_BAUDRATE_SUPPORT && defined(BOTHER)
>>
>> -/* Set a custom baud rate using the termios BOTHER.  */
>> +/* Set a custom baud rate.
>> +
>> +   Prefer the POSIX cfsetispeed/cfsetospeed interface when it accepts
>> +   arbitrary baud rates.  Otherwise fall back to Linux-specific termios2
>> +   (BOTHER) or legacy termios interfaces.  */
>>
>>   static void
>>   set_custom_baudrate_linux (int fd, int rate)
>>   {
>> -#ifdef TCGETS2
>> -  struct termios2 tio;
>> -  const unsigned long req_get = TCGETS2;
>> -  const unsigned long req_set = TCSETS2;
>> -#else
>> +#if defined(HAVE_CFSETSPEED_ARBITRARY)
>>     struct termios tio;
>> -  const unsigned long req_get = TCGETS;
>> -  const unsigned long req_set = TCSETS;
>> -#endif
>> +  if (tcgetattr (fd, &tio) < 0)
>> +    perror_with_name (_("Cannot get current baud rate"));
>> +
>> +  cfsetispeed (&tio, rate);
>> +  cfsetospeed (&tio, rate);
>> +
>> +  if (tcsetattr (fd, TCSANOW, &tio) < 0)
>> +    perror_with_name (_("Cannot set custom baud rate"));
>   It doesn't appear to me this code should be wrapped in:
>
> #if ... && defined(BOTHER)
>
> or for that matter be a part of `set_custom_baudrate_linux' (which AFAICT
> should be left alone, or please point me at what actually you have changed
> there that is not pure code massaging) as this is now platform-agnostic
> (and I believe GNU Hurd has always supported arbitrary baud rates using
> this interface).  I'm leaving it up to you to come up with a name for this
> new variant of the function.
>
>   Also ISTM that HAVE_CUSTOM_BAUDRATE_SUPPORT now also needs to be set if
> HAVE_CFSETSPEED_ARBITRARY rather than relying on BOTHER to be defined as
> well, which may not necessarily be the case (cf. GNU Hurd again).
To clarify the change in set_custom_baudrate_linux: on musl-based
systems BOTHER is visible via <asm/termbits.h>, included through the
HAVE_ASM_TERMIOS_H path in ser-unix.c, so the function does get
compiled. The #else branch then fails to compile because musl's
struct termios exposes the speed fields only as private
__c_ispeed/__c_ospeed, not as c_ispeed/c_ospeed.

Following your suggestion, the fix belongs at the macro level.
Requiring HAVE_STRUCT_TERMIOS_C_OSPEED alongside BOTHER in the
HAVE_CUSTOM_BAUDRATE_SUPPORT guard ensures the Linux path is simply
not compiled on musl systems where those fields are unavailable,
leaving set_custom_baudrate_linux completely unchanged.

In V4:

   - set_custom_baudrate_linux is left completely unchanged.

   - HAVE_CUSTOM_BAUDRATE_SUPPORT is updated to require
     HAVE_STRUCT_TERMIOS_C_OSPEED alongside BOTHER, so the Linux
     path is not compiled on musl where struct termios lacks those
     fields.

   - New independent function set_custom_baudrate_posix(), guarded
     by #if HAVE_CFSETSPEED_ARBITRARY, outside the defined(BOTHER)
     block, covering systems like GNU Hurd.

   - HAVE_CUSTOM_BAUDRATE_SUPPORT extended to also fire when
     HAVE_CFSETSPEED_ARBITRARY is defined.

   - The #else error() fallback is removed. Unsupported platforms
     are already handled in rate_to_code() and
     hardwire_setbaudrate().

V4 will be posted shortly.
>
>> +#else
>> +  error (_("Custom baud rate not supported on this platform"));
>> +#endif
>   This shouldn't be needed as this code is not supposed to be ever compiled
> if !HAVE_CUSTOM_BAUDRATE_SUPPORT (IOW don't set the macro instead if the
> conditions required are not met).
Fixed in V4.

Thanks,
Sunil Dora
>
>    Maciej