RFR: 8286876: NMT.test_unaliged_block_address_vm_assert fails if using clang toolchain [v2]
Johan Sjölen
jsjolen at openjdk.org
Thu Dec 8 14:10:20 UTC 2022
On Mon, 5 Dec 2022 13:16:37 GMT, Johan Sjölen <jsjolen at openjdk.org> wrote:
>> This fixes a bug by removing some UB which Clang used in order to make a test fail. Specifically, Clang optimized the `is_aligned` call in ` MallocHeader::assert_block_integrity` to `true`, because an unaligned `this` is UB. This is fixed by making a static method, essentially. The new code liberally uses `reinterpret_cast<uintptr_t>()`, this is a specifically mentioned as legal use on https://en.cppreference.com/w/cpp/language/reinterpret_cast .
>>
>> **The idea is essentially:** Make sure the pointer looks OK before casting it to a `MallocHeader*`.
>>
>> I also changed the definition of `malloc_header`:
>>
>> ```c++
>> // old
>> return (MallocHeader*)((char*)memblock - sizeof(MallocHeader));
>> // new
>> return &(((MallocHeader*)memblock)[-1]);
>>
>>
>> The previous definition was UB, because you shouldn't cast from a pointer with smaller alignment to one with larger alignment requirements.
>>
>> I also made `MallocHeader::print_block_on_error` static, as the `this` argument was always equal to the `bad_address`.
>
> Johan Sjölen has updated the pull request incrementally with one additional commit since the last revision:
>
> Generate assert_block_integrity thru templated function
Thanks for the comments, I implemented the suggested changes. I think we're on the same page, in that the API needed some slight changes but that it was mostly naming that was an issue?
-------------
PR: https://git.openjdk.org/jdk/pull/11465
More information about the hotspot-runtime-dev
mailing list