On 01/19/2018 08:44 PM, Andrey Konovalov wrote: > Instead of __asan_report_load_n_noabort and __asan_report_store_n_noabort > callbacks Clang emits differently named __asan_report_loadN_noabort and > __asan_report_storeN_noabort (similar to __asan_loadN/storeN_noabort, whose > names both GCC and Clang agree on). > > Add callback implementation for __asan_report_loadN/storeN_noabort. > This made me wonder why this wasn't observed before. So I noticed that inline instrumentation with -fsanitize=kernel-addresss is broken in clang, and clang never calls __asan_report*() functions. I see that you guys fixed this just yesterday https://reviews.llvm.org/D42384 . But it seems that you didn't fix the rest of "if (CompileKernel)" crap. Clang generates "__asan_report_[load,store]N*" instead of "__asan_report_[load,store]_n*" only because of this idiocy: const std::string SuffixStr = CompileKernel ? "N" : "_n"; See https://github.com/llvm-mirror/llvm/blob/ca19eaabd75f55865efd321b7a6f1d4ba3db8bc8/lib/Transforms/Instrumentation/AddressSanitizer.cpp#L2250 Note that SuffixStr is used *only* for __asan_report_* callbacks, which makes no sense because we never ever had __asan_report* callbacks with "N" suffix. So I think that you should just fix the llvm here. And there is probably one more "if (CompileKernel)" crap in runOnModule() which breaks globals instrumentation. -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
