Generic heap calculation

[michael-white-inspire]

Perhaps not an issue, but maybe just a request for documentation or comments in the code. The 'generic' machine uses the following formula for heap calculation:

• 0x8000 + (0x1000 * hartcount) + (SBI_TLB_INFO_SIZE * hartcount * hartcount)

What is the reason for the square of the hart count? The reason I'm asking is that for large number of harts, this square factor along with SBI_TLB_INFO_SIZE can become quite large.

Thanks.

[SiFiveHolland]

To avoid deadlock with the current design, there needs to be space in the TLB FIFOs for all possible combinations of source and destination hart. TLB operations are synchronous -- consider what happens if every hart sends a broadcast TLB flush at the same time. If it makes more sense, one of those hart_counts could be replaced with generic_tlb_num_entries().