Implement a simple interval tree.

[wllenyj]

A special interval tree implementation for VMM resource management.
It's not designed as a generic interval tree, but specialized for VMM
resource management.

Signed-off-by: wllenyj wllenyj@linux.alibaba.com

[sboeuf]

I'm curious why we need such a complex and rich Interval tree? I'm asking because it's not obvious which benefits this will bring. Could you elaborate on the need for this and include it part of the documentation (in the code) and also as part of the git commit.

[jiangliu]

It will be used to manage PCI bus resources, such as mmio/pio address ranges. We are refactoring PCI crate from cloud hypervisor.

[sameo]

It will be used to manage PCI bus resources, such as mmio/pio address ranges. We are refactoring PCI crate from cloud hypervisor.

I think @sboeuf question is similar to the one I asked upstream.
The code from this PR looks nice, but we are missing the requirements behind the complexity it carries. It would be good to e.g. document them from this PR. Personally I'd like to understand:

• What are the VMM specific requirements for managing an interval tree? What are we trying to solve with that PR compared to the current cloud hypervisor vm-allocator?
• How would that work with an actual device creation? Showing the workflow for (device resources creation -> device resources translation to Constraint -> device resources allocation from this allocator) would be nice.
• It seems that each node on this tree holds a (range, NodeState<T>) couple. Basically you're trying to map a range to potentially some object. It would be very useful to get some VMM related examples of what T would be. Something like a PCI bar? Or a chunk of RAM?
• Why isn't a simpler BTreeMap<Range, NodeState<T>>, not good enough?

[jiangliu]

Here is an example to use the interval tree to implement an epoll manager with support of device hotplug.
https://github.com/jiangliu/vmm-epoll/blob/v2/src/interval_tree_mt.rs

Actually the device manager should use an interval tree instead of an btree to map address to registered DeviceIo objects.

[sameo]

I'm merging this PR into the dragonball branch in order to allow @jiangliu to submit the vm-pci code from dragonball.