Neil:
> The fastest would be to create an immortal block as part of the BSS (uninitialized data).

That's an interesting idea, definitely worth exploring and we can probably get some perf win out of it. And yes, using the frozen modules is definitely a step forward and we can leverage that to move these instances into the rodata section of the binary.
 
> I had started doing an experiment with the arena approach before I noticed Eddie's comment about it.  I would like to see his version.

What I had written up is slightly different from what you mentioned. I was mostly concerned about having a small object that we did not reach through the GC roots. If this small object would get a reference count bump, the whole arena would Copy on Write.

I added a commit to the PR with this Arena Immortalization so that you could take a look at the implementation: https://github.com/python/cpython/pull/19474/commits/b29c8ffd3faf99fc5c9885d2a4c6c3c6d5768c8c

The idea is to walk all the arena's pools to mark them as immortal by using a new word added to pool_header. This word would help us identify if the pool is immortal on every pyalloc and pydealloc.

I still get some tests breaking with this, I haven't tried to debug it though