Two test binaries have been compiled on lip3 and executed on lip2. The test binaries come from -O2 compilation of this source:

Add a footnote.

I can't find a way to exploit this observation about free_down in a way that makes the code smaller, so abandoning.

Use switch in RB_REMOVE_COLOR instead of several if-elses. Only modify each color-bit-pair once instead of modifying left and then right. Reduces size to 8766 from 8830 on lip3.

Do not worry about passing to iommu_gas_match_one a range wholly beyond lowaddr. Since that function checks the upper end of the range against lowaddr, it'll get rejected there.

Find a better (smaller) way to set 'child' in RB_INSERT_COLOR. Reduce lip3 size from 8830 to 8734.

Apply the same trick to INSERT_COLOR. This assumes that the two children of elm cannot both be red. I assert that than cannot be.

Undo last revision. Can't be assigning to RB_BITS. Still, a 16 byte shrinkage.

Shrink the size a little more with REMOVE_COLOR adjustments.

Purify the patch.

Merge-in overlapping changes. Add manpage symbolic link.

Add a manpage symlink.

Add a symlink.

Correct variable names.

Move the left/right path walking code into the insert function.

Add augmentation to the RB tree example.

Tweak text, as directed.

Rename variables in REMOVE_COLOR (elm->child, sib->elm) to make it look more like the INSERT_COLOR code.

Update the augmentation function in a way that causes iommu_gas.o to shrink somewhat.

Make the third parameter to RB_ROTATE_{LEFT,RIGHT} an input parameter instead of an output parameter, to avoid a few pointer dereferences where we already have the value in a register. Dropping a new such assignments makes things a tiny bit faster (and the code a tiny bit bigger, because, who knows).

Add to the tree.3 man page. Drop counters.

Drop some redundant assignments from the color functions.

Make it easier for a compiler to erase the final loops in RB_INSERT_COLOR and RB_REMOVE_COLOR when RB_AUGMENT_CHECK is false.

Move the updates of augmentation data to RB_INSERT_COLOR and RB_REMOVE_COLOR, from RB_INSERT and RB_REMOVE, to avoid two separate climbs up toward the root.

Restore performance of augment function. Update augmentation before color-rebalancing in rb_insert/rb_delete.

Add braces.

Enhance comments.

Fix indent.

Replace uppermatch and lowermatch with a single, first-fit, nonrecursive match function.

In D35424#803805, @andrew wrote:

I think the ITS driver just needs a mask in the iommu_map_msi call. It looks like it correctly sets the physical address on the vm_page it creates, but then assumes iommu_map_msi will mask off the upper bits.

Add an entry count output to CTR4()

Update for merge lowermatch patch.

Update to merge in changes to lowermatch.

Merge with an overlapping commit.

Remove a couple of changes that leaked in from another patch.

Typo. Compiles and boots.

Toss all my parts out.

Merge 'insert' into 'one'. Don't let a->size change on failed lookup. Don't miss the chance to accept a shortened lookup when the second 'start' value exceeds end.

Update comments only.

Add CTR3(KTR_SPARE5)

Add a sum-of-tree-sizes counter value.

Use counters.

Add recommended comment.

Remove the material part of this patch, because a stealth reviewer finds a bug in it, and leave only the style bits which do simplify some obscure code.

Recover missing "- offset".

Remove 'offset' in computing initial size, but not when when examining iommu_map product.