Page Allocator bugs

[Knight-Ops]

Chapter 3 onward there is a bug in the Page Allocator that rears it's ugly head when you try to allocate more pages than are possible. Since for i in 0..num_pages - pages is not using checked_sub You can underflow num_pages and end up iterating way past the end of allocable memory, this will likely result in heap corruption for any allocation greater than the number of pages remaining in the heap. This is particularly troublesome because it is in an unsafe block.

Additionally, for i in 0..num_pages - pages should be for i in 0..=num_pages - pages so we can use the last page. While a minor grievance, I actually implemented this on a microcontroller that only had a single page of heap memory, which you couldn't allocate because the range doesn't include the last page in the heap... needless to say, this was detrimental to the usefulness of the allocator.

https://github.com/sgmarz/osblog/blob/master/risc_v/chapters/ch3/src/page.rs#L137

[sgmarz]

I will have to look into this. Thank you.