Huge chunk of time spent on _mi_page_retire normal?

Hi, I am using mimalloc in a programming language runtime, which relies on closures heavily. While the performance is not bad at all, when profiling, I notice that a significant chunk of time is spent on _mi_page_retire. The language runtime does allocate and deallocate closure objects at a very frequent rate hence this behavior makes sense to me on a high level. However, my intuition is that: wouldn't it be great if mimalloc can just let emptied-out pages lying around without retiring them, which I imagine would cut down the _mi_page_retire cost?

I am using the library version v1.7.5. The allocation is mostly performed via aligned_alloc calls (64-byte alignment).

Interestingly, profiling results across different running sessions vary as well, which confuses me even more. Some guidance on this is much appreciated. Thanks!

sample running session 1

sample running session 2

Start by reading the implementation and profiling behavior around _mi_page_retire, including how aligned_alloc with 64-byte alignment reaches that path. Compare the two reported sessions and determine whether page retirement is expected and what explains the differing costs; document the guidance or a reproducible issue if behavior is abnormal.