Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Fix NativeAOT ThunksPool thunk data block size handling. (#110732)
* Fix NativeAOT ThunksPool thunk data block size handling. #88710 made a change in TunkPool.cs moving away from using a page size define to calling ThunkBlockSize to get to end of thunk data block where a common stub address get stored. This change is not equivalent on platforms where the thunk blocks are laid out in pair where a stub thunk blocks are followed by a data thunk block and all gets mapped from file. This is the schema used on Windows platforms. In that layout schema the ThunkBlockSize is 2 * page size meaning that the calculation getting to the end of the thunk data block will move to the end of next thunk stub that is RX memory and storing the common stub address at that location will trigger an AV. This works on iOS since it reports its ThunkBlockSize as one page but that is not totally correct since it uses 2 pages, just that they are allocated in the same way as FEATURE_RX_THUNKS, all thunk stubs blocks followed by all thunk data blocks. The reason why this works is because it only maps the thunk stubs from file, reporting a ThunkBlockSize that is inline with what gets map:ed from file, but then there is a special handling in PalAllocateThunksFromTemplate on iOS that virutal alloc template size * 2, mapping the first template size bytes from the file and the rest are kept as its thunk data blocks. This commit adds a new function returning the size of the thunk data block and use that when calculating the end of the data block instead of using the ThunkBlockSize since its reflects the size of each block getting mapped from file, on Windows platforms that is stub+data, 2 * page size. * Switch to ThunkDataBlockSizeMask in one place. * Include pointer size slot in data block size calculation. Co-authored-by: Jan Kotas <jkotas@microsoft.com>
- Loading branch information