ext/systemc/src/sysc/qt/README.PORT

Date: Tue, 11 Jan 94 13:23:11 -0800
From: "pardo@cs.washington.edu" <pardo@meitner.cs.washington.edu>

>[What's needed to get `qt' on an i860-based machine?]

Almost certainly "some assembly required" (pun accepted).

To write a cswap port, you need to understand the context switching
model.  Turn to figure 2 in the QT TR.  Here's about what the assembly
code looks like to implement that:

	qt_cswap:
		adjust stack pointer
		save callee-save registers on to old's stack
		argument register <- old sp
		sp <- new sp
		(*helper)(args...)
		restore callee-save registers from new's stack
		unadjust stack pointer
		return

Once more in slow motion:

	- `old' thread calls context switch routine (new, a0, a1, h)
	- cswap routine saves registers that have useful values
	- cswap routine switches to new stack
	- cswap routine calls helper function (*h)(old, a0, a1)
	- when helper returns, cswap routine restores registers
	  that were saved the last time `new' was suspended
	- cswap routine returns to whatever `new' routine called the
	  context switch routine

There's a few tricks here.  First, how do you start a thread running
for the very first time?  Answer is: fake some stuff on the stack
so it *looks* like it was called from the middle of some routine.
When the new thread is restarted, it is treated like any other
thread.  It just so happens that it's never really run before, but
you can't tell that because the saved state makes it look like like
it's been run.  The return pc is set to point at a little stub of
assembly code that loads up registers with the right values and
then calls `only'.

Second, I advise you to forget about varargs routines (at least
until you get single-arg routines up and running).

Third, on most machines `qt_abort' is the same as `qt_cswap' except
that it need not save any callee-save registers.

Fourth, `qt_cswap' needs to save and restore any floating-point
registers that are callee-save (see your processor handbook).  On
some machines, *no* floating-point registers are callee-save, so
`qt_cswap' is exactly the same as the integer-only cswap routine.

I suggest staring at the MIPS code for a few minutes.  It's "mostly"
generic RISC code, so it gets a lot of the flavor across without
getting too bogged down in little nitty details.



Now for a bit more detail:  The stack is laid out to hold callee-save
registers.  On many machines, I implemented fp cswap as save fp
regs, call integer cswap, and when integer cswap returns (when the
thread wakes up again), restore fp regs.

For thread startup, I figure out some callee-save registers that
I use to hold parameters to the startup routine (`only').  When
the thread is being started it doesn't have any saved registers
that need to be restored, but I go ahead and let the integer context
switch routine restore some registers then "return" to the stub
code.  The stub code then copies the "callee save" registers to
argument registers and calls the startup routine.  That keeps the
stub code pretty darn simple.

For each machine I need to know the machine's procedure calling
convention before I write a port.  I figure out how many callee-save
registers are there and allocate enough stack space for those
registers.  I also figure out how parameters are passed, since I
will need to call the helper function.  On most RISC machines, I
just need to put the old sp in the 0'th arg register and then call
indirect through the 3rd arg register; the 1st and 2nd arg registers
are already set up correctly.  Likewise, I don't touch the return
value register between the helper's return and the context switch
routine's return.

I have a bunch of macros set up to do the stack initialization.
The easiest way to debug this stuff is to go ahead and write a C
routine to do stack initialization.  Once you're happy with it you
can turn it in to a macro.

In general there's a lot of ugly macros, but most of them do simple
things like return constants, etc.  Any time you're looking at it
and it looks confusing you just need to remember "this is actually
simple code, the only tricky thing is calling the helper between
the stack switch and the new thread's register restore."


You will almost certainly need to write the assembly code fragment
that starts a thread.  You might be able to do a lot of the context
switch code with `setjmp' and `longjmp', if they *happen* to have
the "right" implementation.  But getting all the details right (the
helper can return a value to the new thread's cswap routine caller)
is probaby trickier than writing code that does the minimum and
thus doesn't have any extra instructions (or generality) to cause
problems.

I don't know of any ports besides those included with the source
code distribution.   If you send me a port I will hapily add it to
the distribution.

Let me know as you have questions and/or comments.

	;-D on  ( Now *that*'s a switch... )  Pardo