Re: [patch] Real-Time Preemption, -VP-2.6.9-rc4-mm1-U3

From: Mark_H_Johnson
Date: Fri Oct 15 2004 - 12:03:55 EST


>i have released the -U3 PREEMPT_REALTIME patch:
>
>
http://redhat.com/~mingo/voluntary-preempt/voluntary-preempt-2.6.9-rc4-mm1-U3

>
Built without any problems, booted as single user and had no errors until I
started running init scripts.

First attempt - When I got to S10network start, had a couple
messages displayed about sleeping in invalid contexts (scrolled by too
fast).
It ended with a kernel bug, looked similar to what I reported before. The
new data displayed included:

preempt count: 4
entry 1: schedule+0x46/0x810 / (ksoftirqd+0xd4/0xf0)
entry 2: _spin_lock_irqsave+0x1f/0x80 / (schedule+0x108/0x810)
entry 3: _spin_lock+0x67/0x70 / (load_balance_newidle+0x66/0xc0)
entry 4: _spin_lock+0x67/0x70 / (die_nmi+0x1b/0xa0)

System locked up, hardware reset to reboot. After reboot, noticed
I did capture some data, will send that separately.

Second attempt - basically the same symptom. The kernel BUG output was
incomplete with a "preempt count: 4" but only one entry displayed
(same as entry 1 above).

System locked up, hardware reset to reboot. Will boot with max_cpus=1
to see if I get farther.

Third attempt - had two outputs of tracing data while starting the
network interface. Tried a couple ping commands with numeric destinations
and the machine had no response (displayed) on the first one, Ctrl-C to
abort. On the second try, the system responded the same until I hit Ctrl-C,
no response. Alt-SysRq-T did display task data, ping was in D mode, and
call trace was pretty deep. Thought it was somewhat odd that it (and
several
tasks above it) had a preempt count:1 with
entry 1: _spin_lock+0x1f/0x70 / (__handle_sysrq+0x1f/0xf0)
not sure if that is the expected behavior or not.

Alt-SysRq-S had the same traceback as I reported before. Rebooted with
Alt-SysRq-B.

As mentioned before, will send the system log data in a separate
message.

--Mark H Johnson
<mailto:Mark_H_Johnson@xxxxxxxxxxxx>

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/