arm64 pmap: Handle kernel mappings in pmap_remove_l2().

[markj]

Test Plan

Only compile-tested so far.

[markj]

• Ensure that the existing mapping was destroyed.

[markj]

• Fix KASSERT() format string.

[kib]

sys/arm64/arm64/pmap.c

2412 ↗

(On Diff #50383)

This is the 'break before make', right ? Might be mention the term in the comment above ?

So what would happen with the parallel accesses to the l2 range meantime ?

2417 ↗

(On Diff #50383)

... or kernel.

[markj]

sys/arm64/arm64/pmap.c

2412 ↗

(On Diff #50383)

Right, the "break" happens in pmap_remove_l2(). I'll update the comment here.

I believe it will cause a data abort. As we discussed in email, I believe this can already happen with pmap_demote_l2(). My intent here is to fix the problem with superpages in exec_map, which is pageable.

[markj]

• Address kib's feedback.

[alc]

sys/arm64/arm64/pmap.c

2401 ↗

(On Diff #50419)

I think that this should just be

newl2 = ml3pa | L2_TABLE;

[alc]

Aside from my one comment, this looks fine. Once that comment is addressed, consider this change "Accepted" by me.

[andrew]

sys/arm64/arm64/pmap.c

2401 ↗

(On Diff #50419)

Yes. ATTR_DEFAULT and ATTR_IDX are only needed on last level entries. The bits they set are ignored on table entries.

[markj]

• Remove flags that don't make sense for a table descriptor.