riscv cbo: fix cache underflushing

[br]

Summary

Fix cache underflushing

Reported by: jhb

[jrtc27]

sys/riscv/riscv/cbo.c

63

Invalidating outside the given range is dangerous; any adjacent data risks getting corrupted. Though arm64 seems to do that in its assembly too...

[br]

Indeed. I do find however that this actually fixes my issue with if_xae when tag alignment requirement for TX/RX buffers is downgraded to lower values.

[mhorne]

I think I need some more context to understand this diff.

In D54531#1246777, @br wrote:

Indeed. I do find however that this actually fixes my issue with if_xae when tag alignment requirement for TX/RX buffers is downgraded to lower values.

Is this not a driver bug?

sys/riscv/riscv/cbo.c

83

What is the difference between __align_down and masking to the line size?

[jrtc27]

sys/riscv/riscv/cbo.c

83

None, except that jhb@ is on a question to replace a bunch of vm_offset_t with void * or similar, and __align_down works with pointer types, but masking would require casting to vm_offset_t (or vm_pointer_t aka uintptr_t for CheriBSD).