Remove some lingering page lock usages.

[markj]

Summary

I neglected to remove vm_page_change_lock(), which still had a couple of
consumers in the contig reclaim code.

Fix object locking. The page lock no longer protects the page's
<object, pindex> tuple as of r352110. Instead we must unconditionally
acquire the object lock and validate the page's object pointer.

Remove an unnecessary second check for page wirings in
vm_page_scan_contig(). Both checks are racy so it doesn't really make
sense to repeat them. Races are handled properly in
vm_page_reclaim_run().

Test Plan

Peter tested the patch.

[kib]

sys/vm/vm_page.c

2544 ↗

(On Diff #67045)

Why don't you relock the right object, instead of giving up ?

[markj]

Retry in vm_page_scan_contig() if the object pointer changed.

[jeff]

sys/vm/vm_page.c

2563 ↗

(On Diff #67216)

I didn't notice this before. What happened to the wired check?

2680 ↗

(On Diff #67216)

If you trybusy first you can be sure the object is either identical or NULL. I am likely to make this change eventually.

[markj]

sys/vm/vm_page.c

2563 ↗

(On Diff #67216)

There were two wired checks in this function, I removed one. There is no purpose to having two checks, since there's nothing preventing a wiring from appearing after this function returns and before the reclamation begins. vm_page_reclaim_contig() handles such races.