witness_warn: Redirect output through kwarn(9)

[cem]

Summary

Follow-up to D6134. I'm not sure I love the change since the kwarn
functionality is really all about per-message. Alternatively, the individual
output could be accrued in an sbuf before sending it over to kwarn.

Previous revision: https://reviews.freebsd.org/D6134

[jhb]

Hmm, the rate limiting stuff might chop one of these messages in half though? (That is, truncate the larger message by having the throttling kick in one of the "middle" lines) Also, if you have 'drop to ddb for a warning' enabled you'd drop to DDB for each line rather than at the end. I appreciate the thought, but I think witness_warn is just not a great match for kwarn unfortunately (even though I had originally thought it would be).

Let's do this. First, send a mail to arch@ saying that we want to fix KASSERT, but that if people want optional warnings you have a KWARN ready to go. However, I'd like to get the first patch into fix KASSERT no matter what.

[cem]

Cast printer function for silly caller that doesn't use int result.

[cem]

In D6306#133863, @jhb wrote:

Hmm, the rate limiting stuff might chop one of these messages in half though? (That is, truncate the larger message by having the throttling kick in one of the "middle" lines)

Yes, exactly.

Also, if you have 'drop to ddb for a warning' enabled you'd drop to DDB for each line rather than at the end. I appreciate the thought, but I think witness_warn is just not a great match for kwarn unfortunately (even though I had originally thought it would be).

Right. We could compile output from individual witness_warn() calls into a single sbuf, then kwarn on that.

Let's do this. First, send a mail to arch@ saying that we want to fix KASSERT, but that if people want optional warnings you have a KWARN ready to go. However, I'd like to get the first patch into fix KASSERT no matter what.

Ok.