Would it work to call this from within init_mparams instead? That way it'd get called for all entrypoints, not just malloc.

In that case, it shouldn't do ensure_initialization itself, but just do its work after init_mparams has done its work.

Hmmm, I've though about it, but there are some clues:

1. ensure_initialization is used for mspace also (mspace is like arena in ptmalloc for different continuous space of allocations). At now, mspace is disabled, but if someone want to enable it, there will some unobvious problems. And it is logically incorrect.
2. Currently, ensure_initialization is disabled in dlmalloc, because it is inside #if USE_LOCKS that is also disabled. And it is called into sys_alloc. But for our purpose it is important to have it initialized before main logic of chunk choosing in dlmalloc being called. => in case of ensure_initialization we will need two augmentation (into ensure_initialization and dlmalloc) instead of one in current case.
3. On the other side this initialization into dlmalloc seems correct, because other entry points (realloc, calloc) in all important CFG paths will call it before any chunk manipulation. The only exception here is free, but it is UB to call free without any previous malloc.

So, yes, it possible to move top chunk initialization into ensure_initialization, but it seems that it is better leave it as now.

What do you think?

Thanks for looking into that! I think what you have right now looks like a reasonable approach then.