Remove page locking from pmap_mincore().

[markj]

Summary

It is useless in mincore(2)'s default mode of operation. If we are in
the non-default mode, re-look up the page in the pmap after acquiring
the page's object lock.

Sprinkle some style fixes in vm_map_mincore(). If the approach is
acceptable, I will convert the other implementations of pmap_mincore()
and remove vm_page_pa_tryrelock().

[markj]

Restore the original comment for the mincore_mapped case, I didn't
mean to rewrite it.

[kib]

I think the approach of relying on the object lock is fine.

sys/vm/vm_mmap.c

882 ↗

(On Diff #62651)

Where does this loop terminate if m->object == object and not NULL ?

[markj]

kern_proc_vmmap_resident() requires an update as well.

[markj]

• Convert other pmaps.
• Remove vm_page_pa_tryrelock().
• Initialize pa = 0. Some platforms do not implement pmap_mincore().

[alc]

sys/vm/vm_mmap.c

798 ↗

(On Diff #62731)

Shouldn't this be:

end = round_page(addr0 + len);

[alc]

sys/amd64/amd64/pmap.c

8407 ↗

(On Diff #62731)

This is redundant.

[markj]

sys/vm/vm_mmap.c

798 ↗

(On Diff #62731)

I think so.

882 ↗

(On Diff #62651)

I forgot to submit this comment apparently:

Sorry, I probably don't understand the question. The loop condition is m->object != object. object != NULL implies that object is write-locked, so if m->object == object the page identity cannot change.

[markj]

• Remove an unnecessary assignment in pmap_mincore().
• Fix the calculation of end.