Make changes to address feedback.

Rename (function) ilog2 so that it differs from (macro) ilog2. Rename the other functions too, and make the parameter name the same across all ilog2 functions/macros.

Drop the ilog2 definition from log2.h.

Remove unneeded KASSERT.

@whu: This change would drop the definition of ilog2 that you included in sys/dev/mana/gdma_util.h. However, that function, called ilog2(x), actually returns x + log2(x), because it initializes 'log' to x, and not to 0.

powerpc64 is a trickier beast. You need to 'make TARGET=powerpc TARGET_ARCH=powerpc64 ...' (explicitly specify TARGET_ARCH)

In D45170#1030344, @jhibbits wrote:

PPC LGTM. If you haven't, test a buildkernel with MPC85XX (powerpc) and QORIQ64 (powerpc64) just to be safe.

Change ilog2 to ilog2l where appropriate. Drop bslr, since it's just another name for ilog2l.

Others have used the ilog2 name for the same function. Drop those other definitions.

Update to leave out vm_radix.c.

Calculate the loop bound a little more efficiently. Reorder some computations within the loop to use fewer registers.

Move macro definitions, as suggested.

Restore pmap_load calls dropped from pmap_unwire_ptp.

Combine two short lines into one.

Refresh, seeking review.

Fix typos identified by @markj.

Edit a couple of comments.

Add assertion. Add/update some comments.

Fix a comment, suggested by @alc.

For @alc, change a semicolon to empty braces.

As @alc suggests, remove a redundant assertion (aren't they all redundant?), and a pointless NULL check, and a comment which in no way improved performance.

Apply @alc suggestions. Remove some inessential parts.

Fix an assertion identified by @markj.

Address feedback from @alc.

Re-enter an iterator for checking empty blocks.

Add a missing assignment.

Apply suggestions from @markj.

Don't modify m_ret in the loop, in case iteration happens and that breaks FOREACH. Save the potential return value in m_val instead.

The reporter of this bug:

In D41846#955841, @pkubaj wrote:

Well, you need to ask re@ and since it was your fix, you should probably be the one to do it.

In D41846#955679, @pkubaj wrote:

Can you also MFS to releng/14.0?

As @alc requested, drop the changes to the vm_page struct definition and the increment of the version number that accompanied it.

Are there any more issues to address before I commit this?

Remove needless #include, __unused, blank line.

Bump __FreeBSD_version.
Reshuffle vm_page members as suggested.
Inline all the vm_radix functions that are wrappers for pctrie functions, leaving only a bit of storage management code in vm_radix.c.

Rearrange ordering in vm_page one more time - to move pindex to the start and have order right after it. Without introducing gaps.

Move order and several others up with pindex to avoid creating a gap, and because future changes are likely to access 'order' in vm_radix calls, and having 'order' next to 'pindex' will be cache friendly. Didn't move enough up with pindex last time.

Rearrange the members of vm_page. Move pindex to the beginning to avoid adding and subtracting in the vm_radix functions. Move order and 3 others up with pindex to avoid creating a gap, and because future changes are likely to access 'order' in vm_radix calls, and having 'order' next to 'pindex' will be cache friendly. Make order signed because, first, it is only ever assigned to, compare with, or printed as a signed value and, second, because negative values will be part of a planned future change.