[PATCH] D23354: [asan] Support dynamic shadow address instrumentation

[Etienne Bergeron via llvm-commits]

etienneb added a comment.

Thanks Anna. I modified this patch to include your changes.

If you have some cycles, can you run the unittests with both patches on iOS.
I can't test on that platform.

- https://reviews.llvm.org/D23354
- https://reviews.llvm.org/D25104

Tests are fine on windows 32/64 and linux 64.

In https://reviews.llvm.org/D23354#557270, @zaks.anna wrote:

> When testing with this patch on the Mac, I ran into a problem using -asan-force-dynamic-shadow. Specifically, Mapping.OrShadowOffset was not inferred correctly because it relies on checking kDynamicShadowSentinel and not the ClForceDynamicShadow. How about using ClForceDynamicShadow when setting the Mapping.Offset and gating the maybeInsertDynamicShadowAtFunctionEntry solely on the value of the Mapping.Offset?
>
>   @@ -434,13 +434,19 @@ static ShadowMapping getShadowMapping(Triple &TargetTriple, int LongSize,
>   ...
>        else
>          Mapping.Offset = kDefaultShadowOffset64;
>      }
>   
>   +  if (ClForceDynamicShadow) {
>   +    Mapping.Offset = kDynamicShadowSentinel;
>   +  }
>   +
>      Mapping.Scale = kDefaultShadowScale;
>      if (ClMappingScale.getNumOccurrences() > 0) {
>        Mapping.Scale = ClMappingScale;
>   @@ -1802,7 +1808,7 @@ bool AddressSanitizer::maybeInsertAsanInitAtFunctionEntry(Function &F) {
>   
>    void AddressSanitizer::maybeInsertDynamicShadowAtFunctionEntry(Function &F) {
>      // Generate code only when dynamic addressing is needed.
>   -  if (!ClForceDynamicShadow && Mapping.Offset != kDynamicShadowSentinel)
>   +  if (Mapping.Offset != kDynamicShadowSentinel)
>        return;
>   
>      IRBuilder<> IRB(&F.front().front());
>  
>
>
> If you'd like, I can commit this as a separate patch.





https://reviews.llvm.org/D23354