From f5ba5ffe80c52e448bf162686c21c82f166ab6c0 Mon Sep 17 00:00:00 2001 From: Philipp Bartsch Date: Thu, 13 Jun 2019 14:42:10 +0000 Subject: [PATCH 1/2] fix zero terminated string issue In C "strings" are zero terminated. Functions like strcmp/strncmp/memcmp/... work on them. We have to be careful to not ignore the last byte. --- llvm_mode/compare-transform-pass.so.cc | 14 +++++++------- 1 file changed, 7 insertions(+), 7 deletions(-) diff --git a/llvm_mode/compare-transform-pass.so.cc b/llvm_mode/compare-transform-pass.so.cc index c89655eab..d9a1f9452 100644 --- a/llvm_mode/compare-transform-pass.so.cc +++ b/llvm_mode/compare-transform-pass.so.cc @@ -184,6 +184,7 @@ bool CompareTransform::transformCmps(Module &M, const bool processStrcmp, const Value *Str1P = callInst->getArgOperand(0), *Str2P = callInst->getArgOperand(1); StringRef Str1, Str2, ConstStr; + std::string TmpConstStr; Value *VarStr; bool HasStr1 = getConstantStringInfo(Str1P, Str1); getConstantStringInfo(Str2P, Str2); @@ -202,21 +203,20 @@ bool CompareTransform::transformCmps(Module &M, const bool processStrcmp, const } if (HasStr1) { - ConstStr = Str1; + TmpConstStr = Str1.str(); VarStr = Str2P; constLen = isMemcmp ? sizedLen : GetStringLength(Str1P); } else { - ConstStr = Str2; + TmpConstStr = Str2.str(); VarStr = Str1P; constLen = isMemcmp ? sizedLen : GetStringLength(Str2P); } - /* bugfix thanks to pbst */ - /* ignore terminating '\0' in string for strcmp */ - if (!isSizedcmp && constLen > 0) { - constLen--; - } + /* properly handle zero terminated C strings by adding the terminating 0 to + * the StringRef (in comparison to std::string a StringRef has built-in + * runtime bounds checking, which makes debugging easier) */ + TmpConstStr.append("\0", 1); ConstStr = StringRef(TmpConstStr); if (isSizedcmp && constLen > sizedLen) { constLen = sizedLen; From 7b5905bda6f632425b7c3b4b8ee698cba65dcd48 Mon Sep 17 00:00:00 2001 From: Philipp Bartsch Date: Mon, 17 Jun 2019 02:00:14 +0000 Subject: [PATCH 2/2] llvm_mode/split-switches-pass: add checks Add extra check to allow early exist in trivial cases that would sometimes lead to crashes. --- llvm_mode/split-switches-pass.so.cc | 11 +++++++++-- 1 file changed, 9 insertions(+), 2 deletions(-) diff --git a/llvm_mode/split-switches-pass.so.cc b/llvm_mode/split-switches-pass.so.cc index 45847f885..4c28f34ce 100644 --- a/llvm_mode/split-switches-pass.so.cc +++ b/llvm_mode/split-switches-pass.so.cc @@ -87,6 +87,7 @@ BasicBlock* SplitSwitchesTransform::switchConvert(CaseVector Cases, std::vector< std::vector setSizes; std::vector> byteSets(BytesInValue, std::set()); + assert(ValTypeBitWidth >= 8 && ValTypeBitWidth <= 64); /* for each of the possible cases we iterate over all bytes of the values * build a set of possible values at each byte position in byteSets */ @@ -98,6 +99,8 @@ BasicBlock* SplitSwitchesTransform::switchConvert(CaseVector Cases, std::vector< } } + /* find the index of the first byte position that was not yet checked. then + * save the number of possible values at that byte position */ unsigned smallestIndex = 0; unsigned smallestSize = 257; for(unsigned i = 0; i < byteSets.size(); i++) { @@ -235,9 +238,13 @@ bool SplitSwitchesTransform::splitSwitches(Module &M) { /* this is the value we are switching on */ Value *Val = SI->getCondition(); BasicBlock* Default = SI->getDefaultDest(); + unsigned bitw = Val->getType()->getIntegerBitWidth(); - /* If there is only the default destination, don't bother with the code below. */ - if (!SI->getNumCases()) { + errs() << "switch: " << SI->getNumCases() << " cases " << bitw << " bit\n"; + + /* If there is only the default destination or the condition checks 8 bit or less, don't bother with the code below. */ + if (!SI->getNumCases() || bitw <= 8) { + errs() << "skip trivial switch..\n"; continue; }