So, I've been poking around at our debugging support lately, and I
had remembered Nathan complaining about ptrace(2), specifically that
it pretty much sucks when it comes to multiple LWPs.

It seems to me that we could simply add a ptrace(2) request to "set
which LWP this request is for".

To clarify:

	* Those ptrace(2) requests that would cause a single-threaded
	  program to stop, all LWPs should be stopped in the multi-LWP
	  case.

	* Those ptrace(2) requests that would cause a single-threaded
	  program to continue running should cause all LWPs to continue
	  running in the multi-LWP case.

	* {READ,WRITE}{I,D} operate on the address space, and are thus
	  global to the process anyway.

	* PT_{GET,SET}*REGS are per-LWP, so we need to identify which
	  LWP we want.

So, we would add:

	PT_GET_LWP	<- report back which LWP is being operated on
	PT_SET_LWP	<- set which LWP to operate on

GDB could easily handle issuing the PT_{GET,SET}_LWP commands before
fetching/storing registers in the traced process.

-- 
        -- Jason R. Thorpe <thorpej@wasabisystems.com>