Re: call for testers: altq in current

From: Eygene Ryabinkin <rea-fbsd_at_codelabs.ru>
Date: Mon, 30 Apr 2007 15:27:26 +0400
Nate, good day.

Sat, Apr 14, 2007 at 01:27:42AM +0400, Eygene Ryabinkin wrote:
> I am just using the defaults for the -CURRENT. Can not verify
> them now -- my -CURRENT is crashing with the modem link, so
> I am either writing mails or doing the tests, sorry.

OK, I had cured the modem crash and coincidently it was related to
your changes: the problem was in the altq_subr.c. The behaviour of
the calls to machclk_init() was "check if we have machclk_freq ==
0 and invoke the machclk_init()". And the first action of machclk_init()
was to initialise the tbr_callout and then the machclk_freq was set
(or not, but the tbr_callout was initialized in any way).

But your change introduced another way for the 'machclk_freq' to
be set. And the bad thing was that the tbr_callout was not initialized
and will never be: your hook set the machclk_freq to some value and
machclk_init() was never called. So it gave me the uninitialized
callout with the wrong (NULL) c_mtx and bad flags. And when the
NULL mutex was freed in the softclock() the kernel panic'ed. There
were message in -current from me about this (subject started with
'mtx_unlock(NULL)' posted at 21st Apr 2007). I just did the very
rough and incorrect patch and John Baldwin kindly pointed me that
it was incorrect.

The patch that fixes the root of the problem is attached: it just
decouples the callout initialization from machclk_freq initialization.

I am CC'ing this to John Baldwin and Robert Watson, because they
were involved into the discuission about my previous wrong fix.  I
still have a question: maybe the initialization of the tbr_callout
in my patch should be protected with some mutex? I don't feel that
it is the case, because for the current code is seems to be unrelevant:
the worst thing will be the double initialization of the callout,
but maybe the mutex will be good for the future. Any ideas?

> > On the new code but without loading cpufreq and leaving the freq at 2200
> > Mhz, do you get the right numbers?  Are they constant?
> 
> Monday will reveal the things. Will post an update.

Was not able to test the things on Monday. But will try to do it
on this week. Sorry for it: many other tasks waited my attention
:(( Maybe the weird speeds were related to the uninitialized
tbr_callout(), though I am not sure.
-- 
Eygene

Received on Wed May 02 2007 - 09:42:07 UTC

This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:39:09 UTC