block_size_max can be larger

[dignod]

tlsf/tlsf.c

Line 341 in deff9ab

static const size_t block_size_max = tlsf_cast(size_t, 1) << FL_INDEX_MAX;

The calculation of block_size_max doesnt take into account the fact that we still have the full range of second-level indices for FL_INDEX_MAX first-level index. Lets take a look on indices of current block_size_max block:
fl==FL_INDEX_MAX
sl==0

We can observe that we don`t use a range 2^fl - (2^(fl + 1) - 1) effectively, as we set block_size_max to the smallest possible size in this range, although we still can use all possible second-level ranges for fl=FL_INDEX_MAX, but the last one, easily without being afraid to get out of bounds. To summarize, indices we really wanna have for the block of size block_size_max are:
fl==FL_INDEX_MAX
sl==SL_INDEX_MAX

The correct calculation for block_size_max now has to be:
block_size_max = (1 << FL_INDEX_MAX) + ((1 << (FL_INDEX_MAX - SL_INDEX_COUNT_LOG2)) *(SL_INDEX_COUNT - 1))

This calculation provides us largest possible block size as smallest possible block for fl==FL_INDEX_MAX, sl==SL_INDEX_MAX, which will not get out of bounds for both mapping_search and mapping_insert. We are able to verify that this size is really the largest possible block size by checking that the next possible block size will "overflow" fl index.