Ok just to clarify it.
Are you ok with the last patch or should I change anything else?

ok useing _linux_ is possible (new patch).
But putting -std=c99 is a bad idea for other reasons also (sea below).



make  all-recursive
make[1]: Entering directory `/mnt/ssd/projects/avrdude/avrdude'
Making all in .
make[2]: Entering directory `/mnt/ssd/projects/avrdude/avrdude'
gcc -DHAVE_CONFIG_H -I.  -DCONFIG_DIR=\"/usr/local/etc\"  -Wall -Wno-pointer-sign -std=c99 -g -O2 -MT avrdude-main.o -MD -MP -MF .deps/avrdude-main.Tpo -c -o avrdude-main.o `test -f 'main.c' || echo './'`main.c
In file included from avr.h:26:0,
                 from main.c:47:
avrpart.h:166:29: error: ‘PATH_MAX’ undeclared here (not in a function)
   char          config_file[PATH_MAX]; /* config file where defined */
                             ^
main.c: In function ‘main’:
main.c:456:3: warning: implicit declaration of function ‘getopt’ [-Wimplicit-function-declaration]
   while ((ch = getopt(argc,argv,"?b:B:c:C:DeE:Fi:l:np:OP:qstU:uvVx:yY:")) != -1) {
   ^
main.c:460:27: error: ‘optarg’ undeclared (first use in this function)
         baudrate = strtol(optarg, &e, 0);
                           ^
main.c:460:27: note: each undeclared identifier is reported only once for each function it appears in
main.c:660:36: error: ‘S_IFREG’ undeclared (first use in this function)
     if ((rc < 0) || ((sb.st_mode & S_IFREG) == 0)) {
                                    ^
main.c:891:7: warning: implicit declaration of function ‘strdup’ [-Wimplicit-function-declaration]
       if ((upd->memtype = strdup(mtype)) == NULL) {
       ^
main.c:891:25: warning: assignment makes pointer from integer without a cast [enabled by default]
       if ((upd->memtype = strdup(mtype)) == NULL) {
                         ^
main.c:1032:5: warning: implicit declaration of function ‘usleep’ [-Wimplicit-function-declaration]
     usleep(waittime);
     ^
main.c:1166:9: warning: implicit declaration of function ‘strcasecmp’ [-Wimplicit-function-declaration]
         if ((strcasecmp(m->desc, memname) == 0) && (upd->op == DEVICE_WRITE)) {
         ^
main.c:324:11: warning: unused variable ‘usr_config’ [-Wunused-variable]
   char    usr_config[PATH_MAX]; /* per-user config file */
           ^
main.c:323:11: warning: unused variable ‘sys_config’ [-Wunused-variable]
   char    sys_config[PATH_MAX]; /* system wide config file */
           ^
make[2]: * [avrdude-main.o] Error 1
make[2]: Leaving directory `/mnt/ssd/projects/avrdude/avrdude'
make[1]: * [all-recursive] Error 1
make[1]: Leaving directory `/mnt/ssd/projects/avrdude/avrdude'
make: * [all] Error 2


(file #30876)

And one more patch file.

- removed OSX stuff
- linux still works (tested)
- hopefully removed all unrelated changes




(file #30875)

The last days I put a bit more work into this and think I did some progress.
The custom_baud.patch adds not only the three rate mentioned in the topic. It enables avrdude to set "every" custom baud rate.
It is "every" because of kernel/hardware limitations.

The patch supports mac os x and linuxs.
For mac os x 921600 is the maximum baud rate, it seem to be a kernel/hardware limitation.
For linux I tested it up to 2M baud on different debian and arch systems and a fedora virtualbox.

For all not os x or linux systems the patch does not change anything, here the standard posix way is used.
This is mostly because of my missing knowledge of BSDs.

The Patch also fixes the issue of checking the return values of cfsetospeed() and  cfsetispeed() you mentioned before.

Custom baud rates are useful for avrdude because it makes programming this serial bootloaders faster even if the maximum baud rate of the avr is kind of odd.
 
So this patch makes baudrate.patch obsolete.

(file #30874)

I have cross-checked myself.

OSX uses the BSD model where Bxxxx == xxxx, so no special
treatment is necessary.  (They don't have Bxxxx constants for
the baudrates you mention but for other rates like 230400 etc.)
Their struct termios has baudrate entries separate from the
historic c_*flags fields.

Solaris uses the same model as Linux where the baudrate is
encoded in the lower bits of c_cflags, but it only has standard
baudrates up to B38400 anyway.

So it's only Linux that needs the patch at all.

After some more review, it appears that (at least some versions
of) Linux don't follow the idea of using Bxxxx == xxxx, so I'm
probably going to accept your patch.

Curious, what's the point of adding more baudrates?

The code is supposed to allow arbitrary baudrates already as
it is.  The historical Bxxxx constants are mainly there to
support historic code; on all modern systems, Bxxxx is just
the numerical value xxxx anyway.  (The Single Unix Specification
only allows for the use of Bxxxx constants as the portable
method; but then, baud rates beyond 38400 Bd are inherently
non-portable anyway.)