The game from the bug is doing (rather far away) out of bound read access. However, it is lucky to hit accessible memory on Windows and used to on Wine before commit 354a8bb1f4a65bdec052606f2799db9e2907b5b1. The app creates some heaps with initial commit size of 0x1000 which after the blamed commit results in total heap sizes (and growing heap sizes) not aligned to 64k which introduces memory fragmentation with uncommitted holes, which is making the crash in the game (and possibly others doing out of bound access) more likely. More detailed testing shows that total heap size should always be 64 aligned. The added tests (as well as one existing todo) also cover committed sizes which are not matching Windows now. I prepared the patches which fix that part also (https://gitlab.winehq.org/gofman/wine/-/commits/ntdll_heap_sizes) but I think those are not safe during the code freeze. "ntdll: Round commit size to page size instead of allocation size." looks rather obvious, but it reduces the committed size and thus begs for introducing new regressions under similar pattern. The commit after that looks weird, but it matches the tests which show that Windows actually commits more than needed (so it is better to have both of those patches or none; now we usually commit more because of excessive commit size aligment to 64k). I'd send this remaining part after the code freeze. Both this MR and that fuller one fix the crash in game here. -- https://gitlab.winehq.org/wine/wine/-/merge_requests/9752#note_125405