I'm using kdb-v1.7-2.4.2-pre1 with 2.4.1 (which applied cleanly!) on a
UP system and it makes it into kdb on panic. I'm also using getty,
not minigetty but getty, but I've also disabled it in the passed and
it is not strictly needed to use kdb. That is, I think kdb performs
I/O to the serial device regardless of it's open state in Linux.
Note: for me, hardware flow control wires (DCD/DTR) were needed to get
getty to wake up.
My config params are:
# CONFIG_KDB_OFF is not set
# CONFIG_FRAME_POINTER is not set
> On Wed, 21 Feb 2001 20:37:37 -0600,
> Ken Preslan <kpreslan@xxxxxxxxxxx> wrote:
> >I'm using kdb-v1.7-2.4.0.gz off the FTP site on a 2.4.1 kernel. I have
> >a serial console for my machine setup (with mingetty on the tty device to
> >keep it open). When the machine is running, I can use ctrl-A to break into
> >KDB. For some reason, ctrl-A doesn't work after machine panics (right when
> >I need it :-).
> If the machine is spinning with interrupts disabled then the IRQ from
> the serial line is not being processed so ctrl-A does not get in. This
> is a standard problem with the serial line and keyboard, you need an
> initial interrupt to get the kernel's attention. Once kdb is running
> it uses polled I/O but the first input needs a working IRQ.
> The solution is to use the NMI watchdog which always interrupts, as
> long as the hardware is not totally frozen. SMP boxes automatically
> have the NMI watchdog. The kdb patch includes NMI for UP boxes, under
> the processor type menu.
> Mind you, a panic should drop into kdb after printing the registers.
> Did you select "kdb off by default" in the config or boot with