Let arm64 pmap_qenter() and pmap_kenter() unconditionally set NX.

[markj]

Test Plan

I booted a ThunderX with this diff applied.

[val_packett.cool]

Diff broken by rS354193 'Set the userspace execute never bit on kernel mappings' (looks like that sets UXN in the else case instead).

[markj]

Rebase.

[alc]

sys/arm64/arm64/pmap.c

1260 ↗

(On Diff #63930)

While you're here, I'd like to suggest that it would good from a stylistic standpoint to add ATTR_AP(ATTR_AP_RW).

[alc]

sys/arm64/arm64/pmap.c

1378 ↗

(On Diff #63930)

Just an FYI, and not a comment on this patch. I think that we should be implementing break-before-make here. Occasionally, we have used pmap_qenter() to replace a valid mapping. That is why I did not change this to the simpler pmap_store() when I introduced pmap_store().

[markj]

Explicitly specify ATTR_AP_RW for mappings created by pmap_kenter().

[markj]

sys/arm64/arm64/pmap.c

1378 ↗

(On Diff #63930)

I think this can happen with some implementations of sf_buf_map() (though not on arm64). Are there other examples?

[alc]

sys/arm64/arm64/pmap.c

1378 ↗

(On Diff #63930)

I won't swear to it, but I thought that I recently saw pmap_qenter() being used to remap a virtual page to a different physical page on a different NUMA node.

[andrew]

Can you commit this or shall I commit D22241 first? They conflict as they both change pmap_qenter() and pmap_kenter().

[markj]

In D22141#487424, @andrew wrote:

Can you commit this or shall I commit D22241 first? They conflict as they both change pmap_qenter() and pmap_kenter().

Please go ahead and commit your patch, I will come back to this next week.