import Data.Binary hiding (get)
import Data.Int
import qualified Data.Map as M
+import qualified Data.Set as S
import qualified Data.ByteString.Lazy as B
import Data.Maybe
import Control.Monad.State
-- (offset in bytecode, offset to jump target, ins)
type OffIns = (Int, Maybe BBEnd, Instruction)
-type Targets = [BlockID]
-type BBState = Targets
+type Target = BlockID
+type BBState = S.Set Target
type AnalyseState = State BBState [OffIns]
+emptyBasicBlock :: BasicBlock
+emptyBasicBlock = BasicBlock
+ { code = []
+ , bblength = 0
+ , successor = Return }
+
printMapBB :: MapBB -> IO ()
printMapBB hmap = do
printfBb "BlockIDs: "
printMapBB' [] _ = return ()
printMapBB' (i:is) hmap' = case M.lookup i hmap' of
Just bb -> do
- printfBb $ "Block " ++ show i ++ "\n"
+ printfBb $ "Block " ++ show i ++ ". len: " ++ (show $ bblength bb) ++ "\n"
mapM_ (printfBb . flip (++) "\n" . (++) "\t" . show) $ code bb
printfBb $ case successor bb of
Return -> ""
let nametype = methodNameType methoddirect
let argscount = methodGetArgsCount nametype + (if isStatic then 0 else 1)
+ let exceptionMap :: ExceptionMap
+ exceptionMap = foldl f M.empty $ codeExceptions decoded
+ where
+ f emap ce =
+ if M.member key emap
+ then M.adjust (value:) key emap
+ else M.insert key [value] emap
+ where
+ key = (&&&) eStartPC eEndPC ce
+ value = (&&&) g eHandlerPC ce
+ where
+ g ce' = case eCatchType ce' of
+ 0 -> B.empty
+ x -> buildClassID cls x
+
let msig = methodSignature method
printfBb $ printf "BB: analysing \"%s\"\n" $ toString (methodname `B.append` ": " `B.append` encode msig)
printMapBB mapbb
- -- small example how to get information about
- -- exceptions of a method
- -- TODO: remove ;-)
- let (Just m) = lookupMethodSig methodname sig cls
- case attrByName m "Code" of
- Nothing ->
- printfBb $ printf "exception: no handler for this method\n"
- Just exceptionstream ->
- printfBb $ printf "exception: \"%s\"\n" (show $ codeExceptions $ decodeMethod exceptionstream)
- return $ RawMethod mapbb locals stacks argscount codelen
+ return $ RawMethod mapbb exceptionMap locals stacks argscount codelen
testCFG :: Code -> MapBB
-testCFG = buildCFG . codeInstructions
-
-buildCFG :: [Instruction] -> MapBB
-buildCFG xs = buildCFG' M.empty xs' xs'
+testCFG c = buildCFG (codeInstructions c) (codeExceptions c)
where
- xs' :: [OffIns]
- xs' = evalState (calculateInstructionOffset xs >>= markBackwardTargets) []
-
--- get already calculated jmp-targets and mark the predecessor of the
--- target-instruction as "FallThrough". we just care about backwards
--- jumps here (forward jumps are handled in buildCFG')
-markBackwardTargets :: [OffIns] -> AnalyseState
-markBackwardTargets [] = return []
-markBackwardTargets (x:[]) = return [x]
-markBackwardTargets (x@(x_off,x_bbend,x_ins):y@(y_off,_,_):xs) = do
- rest <- markBackwardTargets (y:xs)
- targets <- get
- let isTarget = y_off `elem` targets
- x_new = case x_bbend of
- Just _ -> x -- already marked, don't change
- Nothing -> if isTarget then checkX y_off else x
- checkX w16 = case x_bbend of
- Nothing -> (x_off, Just $ FallThrough w16, x_ins) -- mark previous insn
- _ -> error "basicblock: something is wrong"
- return $ x_new:rest
-
-
-
-buildCFG' :: MapBB -> [OffIns] -> [OffIns] -> MapBB
-buildCFG' hmap [] _ = hmap
-buildCFG' hmap ((off, entry, _):xs) insns = buildCFG' (insertlist entryi hmap) xs insns
- where
- insertlist :: [BlockID] -> MapBB -> MapBB
- insertlist [] hmap' = hmap'
- insertlist (y:ys) hmap' = insertlist ys newhmap
- where
- newhmap = if M.member y hmap' then hmap' else M.insert y value hmap'
- value = parseBasicBlock y insns
- entryi :: [BlockID]
- entryi = if off == 0 then 0:ys else ys -- also consider the entrypoint
+ buildCFG :: [Instruction] -> [CodeException] -> MapBB
+ buildCFG xs excps = execState (mapM buildCFG' $ alltargets ++ handlerEntries) M.empty
where
- ys = case entry of
- Just (TwoTarget t1 t2) -> [t1, t2]
- Just (OneTarget t) -> [t]
- Just (FallThrough t) -> [t]
- Just Return -> []
- Nothing -> []
+ (offins, targets) = runState (calculateInstructionOffset tryBlocks xs) S.empty
+ alltargets = S.toList $ S.insert 0 targets
+ tryBlocks = map (fromIntegral . eStartPC) excps
+ handlerEntries = map (fromIntegral . eHandlerPC) excps
+ buildCFG' :: Int -> State MapBB ()
+ buildCFG' off = do
+ let value = parseBasicBlock off offins
+ modify (M.insert off value)
parseBasicBlock :: Int -> [OffIns] -> BasicBlock
-parseBasicBlock i insns = BasicBlock insonly endblock
+parseBasicBlock i insns = emptyBasicBlock
+ { code = zip offsets insonly
+ , bblength = lastoff - i + (insnLength lastins)
+ , successor = endblock }
where
(lastblock, is) = takeWhilePlusOne validins omitins insns
- (_, _, insonly) = unzip3 is
- (_, Just endblock, _) = fromJust lastblock
+ (offsets, _, insonly) = unzip3 is
+ (lastoff, Just endblock, lastins) = fromJust lastblock
-- also take last (non-matched) element and return it
takeWhilePlusOne :: (a -> Bool) -> (a -> Bool) -> [a] -> (Maybe a, [a])
omitins (off, _, _) = off < i
-calculateInstructionOffset :: [Instruction] -> AnalyseState
-calculateInstructionOffset = cio' (0, Nothing, NOP)
+calculateInstructionOffset :: [BlockID] -> [Instruction] -> AnalyseState
+calculateInstructionOffset exstarts = cio' 0
where
- addW16Signed :: Int -> Word16 -> Int
addW16Signed i w16 = i + fromIntegral s16
where s16 = fromIntegral w16 :: Int16
- cio' :: OffIns -> [Instruction] -> AnalyseState
+ cio' :: Int -> [Instruction] -> AnalyseState
cio' _ [] = return $ []
- cio' (off,_,_) (x:xs) = case x of
+ cio' off (x:xs) = case x of
IF _ w16 -> twotargets w16
IF_ICMP _ w16 -> twotargets w16
IF_ACMP _ w16 -> twotargets w16
IFNONNULL w16 -> twotargets w16
IFNULL w16 -> twotargets w16
GOTO w16 -> onetarget w16
+ ATHROW -> notarget
IRETURN -> notarget
ARETURN -> notarget
RETURN -> notarget
- _ -> ((off, Nothing, x):) <$> next
+ _ -> if newoffset `elem` exstarts
+ then do
+ modify (S.insert newoffset)
+ ((off, Just $ OneTarget newoffset, x):) <$> next
+ else normalins
where
+ normalins = do
+ tailinsns <- next -- eval remaining instructions
+ isNextInsATarget <- (S.member newoffset) <$> get
+ let bbtyp = if isNextInsATarget
+ then Just $ FallThrough newoffset
+ else Nothing
+ return $ (off, bbtyp, x):tailinsns
notarget = ((off, Just Return, x):) <$> next
onetarget w16 = do
let jump = off `addW16Signed` w16
- modify (jump:)
+ modify (S.insert jump)
((off, Just $ OneTarget jump, x):) <$> next
twotargets w16 = do
let nojump = off + 3
- modify (nojump:)
+ modify (S.insert nojump)
let jump = off `addW16Signed` w16
- modify (jump:)
+ modify (S.insert jump)
((off, Just $ TwoTarget nojump jump, x):) <$> next
next = cio' newoffset xs
- newoffset = (off + insnLength x, Nothing, NOP)
+ newoffset = off + insLen
+ insLen = insnLength x
-- TODO(bernhard): does GHC memomize results? i.e. does it calculate the size
-- of `NOP' only once?