diff --git a/Control/Concurrent/Raw.hs b/Control/Concurrent/Raw.hs
new file mode 100644
--- /dev/null
+++ b/Control/Concurrent/Raw.hs
@@ -0,0 +1,21 @@
+{-# LANGUAGE UnicodeSyntax, NoImplicitPrelude, MagicHash, UnboxedTuples #-}
+
+module Control.Concurrent.Raw ( rawForkIO, rawForkOn ) where
+
+import Data.Function ( ($) )
+import GHC.IO        ( IO(IO) )
+import GHC.Exts      ( Int(I#), fork#, forkOn# )
+import GHC.Conc      ( ThreadId(ThreadId) )
+
+-- A version of forkIO that does not include the outer exception
+-- handler: saves a bit of time when we will be installing our own
+-- exception handler.
+{-# INLINE rawForkIO #-}
+rawForkIO ∷ IO () → IO ThreadId
+rawForkIO action = IO $ \s →
+   case (fork# action s) of (# s1, tid #) → (# s1, ThreadId tid #)
+
+{-# INLINE rawForkOn #-}
+rawForkOn ∷ Int → IO () → IO ThreadId
+rawForkOn (I# cpu) action = IO $ \s →
+   case (forkOn# cpu action s) of (# s1, tid #) → (# s1, ThreadId tid #)
diff --git a/Control/Concurrent/Thread.hs b/Control/Concurrent/Thread.hs
--- a/Control/Concurrent/Thread.hs
+++ b/Control/Concurrent/Thread.hs
@@ -1,5 +1,9 @@
-{-# LANGUAGE NoImplicitPrelude, UnicodeSyntax, RankNTypes #-}
+{-# LANGUAGE CPP, NoImplicitPrelude, UnicodeSyntax, RankNTypes #-}
 
+#if __GLASGOW_HASKELL__ >= 701
+{-# LANGUAGE Trustworthy #-}
+#endif
+
 --------------------------------------------------------------------------------
 -- |
 -- Module     : Control.Concurrent.Thread
@@ -52,9 +56,7 @@
 --------------------------------------------------------------------------------
 
 -- from base:
-import qualified Control.Concurrent ( forkIO
-                                    , forkOS
-                                    , forkOn
+import qualified Control.Concurrent ( forkOS
                                     , forkIOWithUnmask
                                     , forkOnWithUnmask
                                     )
@@ -70,7 +72,10 @@
 -- from base-unicode-symbols:
 import Data.Function.Unicode        ( (∘) )
 
+-- from threads:
+import Control.Concurrent.Raw       ( rawForkIO, rawForkOn )
 
+
 --------------------------------------------------------------------------------
 -- * Forking threads
 --------------------------------------------------------------------------------
@@ -79,7 +84,7 @@
 -- a computation that when executed blocks until the thread terminates
 -- then returns the final value of the thread.
 forkIO ∷ IO α → IO (ThreadId, IO (Result α))
-forkIO = fork Control.Concurrent.forkIO
+forkIO = fork rawForkIO
 
 -- | Like @Control.Concurrent.'Control.Concurrent.forkOS'@ but returns
 -- a computation that when executed blocks until the thread terminates
@@ -91,7 +96,7 @@
 -- a computation that when executed blocks until the thread terminates
 -- then returns the final value of the thread.
 forkOn ∷ Int → IO α → IO (ThreadId, IO (Result α))
-forkOn = fork ∘ Control.Concurrent.forkOn
+forkOn = fork ∘ rawForkOn
 
 -- | Like @Control.Concurrent.'Control.Concurrent.forkIOWithUnmask'@ but returns
 -- a computation that when executed blocks until the thread terminates
diff --git a/Control/Concurrent/Thread/Group.hs b/Control/Concurrent/Thread/Group.hs
--- a/Control/Concurrent/Thread/Group.hs
+++ b/Control/Concurrent/Thread/Group.hs
@@ -5,6 +5,10 @@
            , RankNTypes
   #-}
 
+#if __GLASGOW_HASKELL__ >= 701
+{-# LANGUAGE Trustworthy #-}
+#endif
+
 --------------------------------------------------------------------------------
 -- |
 -- Module     : Control.Concurrent.Thread.Group
@@ -47,9 +51,7 @@
 --------------------------------------------------------------------------------
 
 -- from base:
-import qualified Control.Concurrent     ( forkIO
-                                        , forkOS
-                                        , forkOn
+import qualified Control.Concurrent     ( forkOS
                                         , forkIOWithUnmask
                                         , forkOnWithUnmask
                                         )
@@ -75,7 +77,7 @@
 
 -- from threads:
 import Control.Concurrent.Thread        ( Result )
-
+import Control.Concurrent.Raw           ( rawForkIO, rawForkOn )
 #ifdef __HADDOCK__
 import qualified Control.Concurrent.Thread as Thread ( forkIO
                                                      , forkOS
@@ -134,7 +136,7 @@
 -- | Same as @Control.Concurrent.Thread.'Thread.forkIO'@ but additionaly adds
 -- the thread to the group.
 forkIO ∷ ThreadGroup → IO α → IO (ThreadId, IO (Result α))
-forkIO = fork Control.Concurrent.forkIO
+forkIO = fork rawForkIO
 
 -- | Same as @Control.Concurrent.Thread.'Thread.forkOS'@ but additionaly adds
 -- the thread to the group.
@@ -144,7 +146,7 @@
 -- | Same as @Control.Concurrent.Thread.'Thread.forkOn'@ but
 -- additionaly adds the thread to the group.
 forkOn ∷ Int → ThreadGroup → IO α → IO (ThreadId, IO (Result α))
-forkOn = fork ∘ Control.Concurrent.forkOn
+forkOn = fork ∘ rawForkOn
 
 -- | Same as @Control.Concurrent.Thread.'Thread.forkIOWithUnmask'@ but
 -- additionaly adds the thread to the group.
diff --git a/threads.cabal b/threads.cabal
--- a/threads.cabal
+++ b/threads.cabal
@@ -1,5 +1,5 @@
 name:          threads
-version:       0.5
+version:       0.5.0.1
 cabal-version: >= 1.9.2
 build-type:    Custom
 stability:     experimental
@@ -45,11 +45,12 @@
 -------------------------------------------------------------------------------
 
 library
-  build-depends: base                 >= 4.4   && < 4.6
+  build-depends: base                 >= 4.4   && < 4.7
                , base-unicode-symbols >= 0.1.1 && < 0.3
-               , stm                  >= 2.1   && < 2.4
+               , stm                  >= 2.1   && < 2.5
   exposed-modules: Control.Concurrent.Thread
                  , Control.Concurrent.Thread.Group
+  other-modules:   Control.Concurrent.Raw
   ghc-options: -Wall
 
 -------------------------------------------------------------------------------
@@ -61,10 +62,10 @@
   ghc-options:    -Wall -threaded
 
   build-depends: threads
-               , base                 >= 4.4   && < 4.6
+               , base                 >= 4.4   && < 4.7
                , base-unicode-symbols >= 0.1.1 && < 0.3
-               , stm                  >= 2.1   && < 2.4
+               , stm                  >= 2.1   && < 2.5
                , concurrent-extra     >= 0.5.1 && < 0.8
                , HUnit                >= 1.2.2 && < 1.3
-               , test-framework       >= 0.2.4 && < 0.6
+               , test-framework       >= 0.2.4 && < 0.7
                , test-framework-hunit >= 0.2.4 && < 0.3