Further reorganize pmap_invalidate TLB code.

[kib]

Summary

Split calculation of mask for shootdown IPI and local invalidation. Reorder IPI before local.

Test Plan

Peter, could you, please, test this, same as D17184.

[alc]

sys/amd64/amd64/pmap.c

1810

Is there a reason that you're now doing the remote invalidations before the local one? Two of the functions are now doing remote invalidations first, and one is doing local invalidations first.

[kib]

Move all IPIs before local, as advertised.

[pho]

I ran tests with vm.pmap.pcid_enabled set to 0 and 1. The tests included all threaded tests and all of the mmap() tests.
No problems seen.

[mjg]

I started getting semi-reliable tinderbox failures which I bisected to this change.

The box in question is a 4-socket broadwell with:
vm.pmap.invpcid_works: 1
vm.pmap.pcid_enabled: 1
vm.pmap.pg_ps_enabled: 1
vm.pmap.pat_works: 1

Most of the time the bug can be reproduced while building i386 LINT kernels and manifests itself with objcopy complaining about passed files (objcopy: file format not recognized, files change at random).

head r338993 with this change reverted works fine.

[kib]

Try this please P225

[mjg]

the patch works

[markj]

This revision can be closed. (I'd do it, but phabricator doesn't give me the option.)