Re: [KGDB] Make kgdb get in sync with it's I/O drivers for the breakpoint

[George Anzinger <george@mvista.com>]

Tom Rini wrote:
> Hello.  The following interdiff, vs current kgdb-2 CVS makes kgdb core
> and I/O drivers get in sync in order to cause a breakpoint.  This kills
> off the init/main.c change, and makes way for doing things much earlier,
> if other support exists.  What would be left, tangentally, is some sort
> of queue to register with, so we can handle the case of KGDBOE on a
> pcmcia card.  George? Amit? Comments ?

Well a simple but dumb way is to poll using the timer list, i.e. set up a timer 
at the first entry were things "might" work and if the driver is not yet, do a 
timer to come back in 1 tick, and keep doing it for each tick until it is 
available.  This puts it all on the kgdb side.

The other way is with a call back list which would be managed by common OE code. 
  This would put most of the code in that area.  I tend to like call back lists 
that one registers for by passing in a structure which contains a "list_head" 
member.  That way there is no memory allocation on either end.  The manager, on 
a register call, just puts the new structure in its call back list.  The struct 
would have the list_head member and a function member, and the function would be 
called with the struct address as its only parameter.  This allows for an 
expanded struct if more complex info is needed.
~
-- 
George Anzinger   george@mvista.com
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml