On Sat, 27 Feb 1999, Jason Thorpe wrote:

> [ Can we follow this up to tech-kern next? ]

  Certainly.

> The alpha witches to proc0's stack (its entire context, actually) in
> switch_exit().  proc0 _CAN'T_ be running alreay when you call switch_exit(),
> so I really don't see how there is a "2 proesses using same stack" problem.

  I looked at the alpha at the time, but didn't completely understand what
it was doing.  I went back and looked at it some more and have a better
idea of what it is doing now.

> Really, proc0 is FOR this.. it's the primary processor's idle context.
> 
> Oh...wait... I see.  "Eek."  This trick mostly works other places because
> the stack isn't actually USED (i.e. no software TLB miss handler).

  Aha!  I now see where the mips went wrong.  When the mips was switched
to use the proc0 stack, it was *resetting* the stack pointer, which then
started scribbling over the current stack frames in use by uvm_scheduler.
It should have been using the *current* stack pointer for proc0.  Doing
this now allows a profiled kernel to work.

Michael