KASSERT_WARN for asserting malloc(M_WAITOK) not in a non-sleepable thread

[Davide Italiano]

On Thu, Sep 25, 2014 at 9:14 AM, Adrian Chadd <adrian at freebsd.org> wrote:
> Hi,
>
> Please bring in KASSERT_WARN().
>
> I'm grown up enough to use KASSERT_WARN() along with handling the
> invariant check myself in code. Having KASSERT_WARN() means I can add
> in this rather than printf()s or device_printf()'s with various knobs
> to remove it.
>
> (This is absolutely _not_ the "should KASSERT() optionally just log"
> argument. I'm not going to get into that a second time.)
>
>

If you put a KASSERT() inside your code -- probably you should be
careful enough to put that iff you're sure that it should be always
verified. No exceptions.
People tend to be very lazy (including me). I don't expect everybody
diligently upgrading KASSERT_WARN to KASSERT. So KASSERT_WARN start
becoming more and more widespread, and people realize all of these
need to be upgraded to KASSERT or removed. This generally happens
after years. Yet. Another. Crusade.
There's a lot of work in the kernel to remove old/wrong/naive  KPI
from the kernel. jhb@ is looking at timeout()-> callout() conversion.
I'm personally looking at dev_clone() removal. There are a lot of
other examples.
Adding KASSERT_WARN is a step backward, not a step forward, IMHO.
That said, if you want to pollute the kernel, fine. I expressed my
opinion, and I'm personally not happy about this, but I never stated
I'm gonna stop you from doing that.

Thanks,

--
Davide