pmap_enter_{l2,pde}: correct the handling of an error case

[alc]

Summary

When pmap_enter_object()'s call to pmap_enter_{l2,pde}() fails to create a managed mapping within the kernel address space due to the inability to allocate a PV entry, it needs to remove the kernel page table page from the pmap's trie of idle page table pages. Previously, it did not.

[andrew]

sys/arm64/arm64/pmap.c

5835

Can you use ADDR_IS_KERNEL here?

[alc]

Use ADDR_IS_KERNEL.

[andrew]

Is my reading of the code correct where uwptpg == NULL in the new else case because it's only set for userspace maps? If so it means we will call pmap_remove_pt_page.

[markj]

In D51353#1173164, @andrew wrote:

Is my reading of the code correct where uwptpg == NULL in the new else case because it's only set for userspace maps?

That's my understanding.

[alc]

I have two related observations:

1. !ADDR_IS_KERNEL() appears more frequently than ADDR_IS_KERNEL(). @andrew What would you think about adding ADDR_IS_USER()?
2. @kib @markj What do you think about adding these macros to amd64 to eliminate trivial differences between the pmap implementations?

[kib]

In D51353#1173408, @alc wrote:

1. @kib @markj What do you think about adding these macros to amd64 to eliminate trivial differences between the pmap implementations?

I do not see why not. It is trivial macro, although I prefer a different way to write it for amd64, like

#define ADDR_IS_KERNEL(va) ((vm_offset_t)(va) >= kva_layout.kva_min)