Jason Wessel wrote:
quoted text > Ingo Molnar wrote:
>> * Jan Kiszka <jan.kiszka@web.de> wrote:
>>
>> =20
>>> As most changes are tightly coupled, this refactoring patch for
>>> KGDB_8250 as well as the core and the new KGDBOC driver comes as a
>>> single chunk. The changes are:
>>> - Reorganized configuration: I/O drivers can be independently
>>> configured as module or built-in
>>> - Dynamic reconfiguration for KGDB_8250 (just like for KGDBOC)
>>> - Reworked KGDB_8250 configuration string format
>>> - attachwait removed, arming the debugger via assigning an I/O drive=
r
quoted text >>> implies "attachwait"
>>> - Cleaned up I/O driver managment of the core
>>> - Matured the various boot-up, configure, unconfigure code paths for=
quoted text >>> both I/O drivers
>>> - IRQ vs. KGDB_CONSOLE-output SMP race fixed for KGDB_8250
>>> - Reduced and cleaned up hooks into serial_core/8250
>>> - Kconfig cleanups
>>> =20
>> great stuff! I've picked these cleanups up into x86.git. (will pick up=
=20
quoted text >> Jason's commits too)
>>
>> Could you try something else too perhaps, which would be way useful fo=
r=20
quoted text >> me: to add a sysctl flag (or something like that) to change kgdboc to =
quoted text >> accept a Ctrl-C and break into kgdb mode? [this means a simple Ctrl-C =
on=20
quoted text >> a kgdboc line would break into KGDB as well - but that would be an=20
>> acceptable price.] Right now kgdboc just hangs when gdb attaches - i=20
>> have to generate a SysRq sequence via a terminal emulator to break it =
quoted text >> into KGDB mode.
>> =20
>=20
> FYI, even if you were to hack in a control-c vs sysrq, gdb will still
And quite a bit of "hacking" would be required here, because the sysrq
hook is built upon the assumption that a hardware break starts the
sequence, not that just a single special character comes in. Changing
this means changing the drivers, not just some central hook function.
quoted text > hang on connect because it does not issue a break of any kind when it
> connects. It assumes the connection is in a usable state.
>=20
> The proxy spliter automatically sends the break (or in the current case=
quoted text > the sysrq g)
>> This would make kgdboc way more practical without having to resort to =
a=20
quoted text >> protocol splitting proxy, etc.
>>
>> Ingo
>> =20
>=20
>=20
> Jason.
Jan
