Looks sensible. Just one question as highlighted below.

I assume it works :-)

tests/configs/base_config.py (Diff revision 1)

Why has this disappeared?

Show all issues

1. Nilay Vaish April 17, 2013, 9:14 a.m. (April 17, 2013, 9:14 a.m.)

   The call to init_cpu(), as of now, only creates the interrupt controller
   for the cpu. When we switch from one cpu with another, we hand over the
   interrupt controller as well. So, as such that call was not required.
   Additionally, the interrupt controller in x86 architecture is a Pio Device.
   The init() function of the Pio Device class checks whether or not it is
   connected to something. For a cpu that is not initially operational,
   the interrupt controller is not connected to anything. This resulted
   in a panic(). Hence, I removed that call.

2. Andreas Hansson April 17, 2013, 8:47 p.m. (April 17, 2013, 8:47 p.m.)

   Ok. Thanks for the clarification.
   
   I'd be keen to know if Andreas S has any input on this one. Is it needed for the KVM cpu or similar?

3. Andreas Sandberg April 18, 2013, 2 a.m. (April 18, 2013, 2 a.m.)

   I just had a quick look. I'd prefer if you fix this checking if the CPU is switched out in init_cpu instead. The intention of init_cpu is to have a common cpu initialization hook that can be overloaded by test cases, removing it from the system initialization code clearly defeats that purpose. In the future, I'd like to move thread creation from run.py to this method.

tests/configs/base_config.py (Diff revision 1)

I assume the switchroo tests are unchanged for ARM with this change?

LGTM. I think I would have preferred to have the connectAllPorts call in init_system, but the current version is OK.

Minor thing, but officially there should only be one keyword (x86 in this case). I'd suggest baking the regression bit into the title.