illumos gate

Rich Lowe wrote:

At present, we assert that q_len is not 0 on SPARC only, because SPARC seems to have asm implementations of kstat_q_* for some reason.

If the assert is a valid thing to do (and I say "assert", it appears to actually be not conditional on DEBUG), it should be done on both platforms. If invalid, on neither. It definitely should not be platform specific while in a common interface

to be exact, on sparcv9 the asm code behaves as having VERIFY, while corresponding C code has ASSERT. I think it is not reasonable to dump asm code as its there for performance reasons and assuming its valid reason, I think, the correct course of action would be use VERIFY in C code as well.

After careful reading of how the SPARC code is implementing those kstat_* functlion, I have found an way to implement ASSERT like behavior, so we can continue to have ASSERT in C code and have both platforms to behave the same way.