The problem is that, other than perhaps the stack walk, I don't think any of these costs can be avoided. Afterall, we need to save the thread state to the TSO in order to safely GC (since TSOs are GC roots). I suspect it would also be necessary to lock the capability.
Is there a way to do it lazily? By say lazily i meaning, can we just
#13944: Introduce synchronized FFI -------------------------------------+------------------------------------- Reporter: winter | Owner: (none) Type: feature request | Status: new Priority: normal | Milestone: 8.4.1 Component: Compiler | Version: 8.0.1 Resolution: | Keywords: Operating System: Unknown/Multiple | Architecture: | Unknown/Multiple Type of failure: None/Unknown | Test Case: Blocked By: | Blocking: Related Tickets: | Differential Rev(s): Wiki Page: | -------------------------------------+------------------------------------- Comment (by bgamari): perform the FFI call and mark the capacity with some kind of SYNC flag. If a GC thread is asking for sync during FFI, then let it perform these stack saving for us. Otherwise we just happily continue, and saving all the overheads. I'll assume by "it" you mean the thread-state save. I don't believe it is possible; afterall, there may be thread state (which may contain reference to heap objects) sitting in machine registers. -- Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/13944#comment:8 GHC http://www.haskell.org/ghc/ The Glasgow Haskell Compiler