Stack check for AP_STACK

[simonmar]

This is a bug I uncovered in the interpreter, that I think is also present in the compiler.

When compiling code for a function or thunk, we aggregate the stack usage from case continuations up to the top of the function/thunk and perform a single stack check. This normally works fine: we know the maximum amount of stack that will be required in the evaluation of this function/thunk, and we check for it up front.

However, this goes wrong if the evaluation is suspended by an asynchronous exception: the stack is broken into chunks and stored in AP_STACK objects, which may later be resumed. On resumption, we might not have enough stack any more: the code might now be running on another stack entirely, even.

Our proposed solution is as follows:

• Continuations that require more than a certain amount of stack (say 4k) do their own stack checks.

• an AP_STACK object records the amount of stack available at the time it was suspended, if the amount is <4k. On resumption of an AP_STACK, we ensure that at least this amount of stack is available. (there's a spare half-word field in AP_STACK that we can use for this).

The 4k limit is important: it puts a bound on the amount of stack growth due to evaluating an AP_STACK. Essentially the 4k limit is large enough that it almost never happens.

[simonmar]

Partly done:

Tue Sep 11 14:02:28 BST 2007  Simon Marlow <simonmar@microsoft.com>
  * FIX #1466 (partly), which was causing concprog001(ghci) to fail
  An AP_STACK now ensures that there is at least AP_STACK_SPLIM words of
  stack headroom available after unpacking the payload.  Continuations
  that require more than AP_STACK_SPLIM words of stack must do their own
  stack checks instead of aggregating their stack usage into the parent
  frame.  I have made this change for the interpreter, but not for
  compiled code yet - we should do this in the glorious rewrite of the
  code generator.

[simonmar]

Do this as part of the new back end

[simonmar]

Bumping to 6.14.1 as this is work to do in the new codegen.

[morabbin]

Bump; does this work in the new codegen as indicated above?

[simonmar]

Replying to morabbin:

Bump; does this work in the new codegen as indicated above?

No, not done yet.