Optimize PointerIsMine() for the secondary allocator

[hygoni]

As the commit 30a6aaf66c5bd3c35d63fe37a37131cbd3e7c05b in PR #72, the secondary allocator can be optimized in the same way. However, unlike the primary allocator, there is no way to limit address space for the secondary allocator.

The optimization would be to make the secondary allocator to take address space range just like the primary one.

[hygoni]

Just realzing that this approach is not as easy as because mmap() system calls does not support mmap()ing within specific address range :(

Another approach might be to implement own address space allocator and use MAP_FIXED,

...Or just use a map (hashmap, treemap, bitmap, etc.) that translates an address to boolean type 'allocated from the secondary or not'

[hygoni]

Another approach might be to implement own address space allocator and use MAP_FIXED,

• [ ] Implement red-black tree or AVL tree in sanitizer_common
• [ ] Implement address space allocator using the RBtree
• [ ] mmap() 2T of VA space first, and mmap(MAP_FIXED) only within that address space

.....and probably (maybe too far?) introduce RCU based balanced tree for better performance.

This is not a small optimization...

[hygoni]

Resolved by METAlloc metadata management scheme