Low priority: Theoretical deadlock when dumping exit info

[kren1]

Describe the bug
erts_exit can acquire lock in different order than other code, which might be a potential for deadlock

To Reproduce
I haven't observed this dynamically

Here's a theoretical example I see in the code:

Thread 1: lock(&smq_mtx) -> lock(&export_staging_lock)

This goes via the error condition here, where the dumping of information eventually calls export_table_sz())

Thread 2: lock(&export_staging_lock) -> lock(&smq_mtx)

This goes via index_put_entry, which can fail and lead to smq_mtx:

Expected behavior

Locks are acquired in the same order

Affected versions
Ran on OTP 26, 12509834ed

Additional context

This was found by Infer and I'm not familiar enough with the code to tell if there are some implicit invariants that prevent this sort of deadlock. There are ~242 examples of this all related to this exit dumping of info, maybe I haven't described the best one above. This seems quite unlikely to observe dynamically, so I thought you might be interested, but let me know if this is too niche/small sort of issue to be interesting.

report.json.zip
59004b2add47e12806aaaaa43834b8) is the full report of all the issue (you can download infer for a bit nicer exploring of issues). The issue I described is at index 4

[sverker]

This scenario cannot happen as there are two threads doing crash dumping which there are protection against.

However, export_table_sz() should avoid locking during dump like export_info() does. Made a fix for that in #8521.