Re: PCU panic on V240

On Dec 5, 2013, at 2:11 PM, matthew green <[email protected]> wrote:

> 
>> panic: kernel diagnostic assertion "!cpu_intr_p()" failed: file "/work/build/src/sys/kern/subr_xcall.c", line 351
> [ .. ]
>> This is the high priority xcall from subr_pcu.c and
>> curcpu()->ci_idepth is zero.
> 
> this doesn't make sense.  ci_idepth should be zero here, and that
> should force cpu_intr_p() to return false, and thus the ASSERT()
> shouldn't fire.  has something changed ci_idepth between the check
> and ddb???  yhuck.

It comes from Rev. 1.17+1.18 of kern/subr_xcall.c:

-       xc_sih = softint_establish(SOFTINT_CLOCK | SOFTINT_MPSAFE,
+       xc_sih = softint_establish(SOFTINT_SERIAL | SOFTINT_MPSAFE,
            xc__highpri_intr, NULL);

Reverting it to Rev. 1.16 all is fine.  Is there some difference
regarding interruption between the softint levels CLOCK and SERIAL?

>> What is going wrong here?
> 
> assuming this has nothing to do with PCU, something very very
> strange and unknown.

Not PCU, passive serialization (subr_pserialize.c) -- my fault.

--
J. Hannken-Illjes - [email protected] - TU Braunschweig (Germany)