Re: [patch] voluntary-preempt-2.6.9-rc1-bk12-R5

[Rui Nuno Capela]

Hi,

I'm back with some news :)


>* Ingo Molnar wrote:
>
>> * Rui Nuno Capela wrote:
>>
>> > OK, could just someone with a P4 HT/SMP box hand me their working
>> > kernel .config file for me to try? That could be a good starting
>> > point, if not a plain baseline.
>>
>> I'll try the latest VP kernel (-R9) on a P4/HT SMP box in a minute and
>> will send you a .config if it works. [...]
>
> P4/HT SMP works fine here - config attached.
>
> since your lockups occur under X, could you try to disable DRI/DRM in
> your XConfig? Also, would it be possible to connect that box to another
> box via a serial line and enable the kernel's serial console via the
> 'console=ttyS0,38400 console=tty' boot option and run 'minicom' on that
> other box, set the serial line to 38400 baud there too and capture all
> kernel messages that occur when the lockups happens? Also enable the NMI
> watchdog via nmi_watchdog=1.
>


OK. Spent the whole late night building several R9's, enabling modules one
by one, and testing most options in a incremental fashion. Didn't tried
the serial console yet but I found something that's otherwise specific to
my hardware setup, but only exposed with the VP SMP configuration, as I've
been recently reporting.

So the base experiments were conducted by applying on:

    linux-2.6.9-rc1.tar.bz2

the follwing patches in sequence:

    patch-2.6.9-rc1-bk12.bz2
    voluntary-preempt-2.6.9-rc1-bk12-R9

Here are my summary report:

1) My first experiment was using Ingo's .config exactly as he gave me,
with the sole exception that I switched reiserfs support liked in, as
needed by primary/boot disk partitions. It boots OK, fine but missing
alomost every needed module/device support.

2) Switched module (un|auto)loading support. OK.

3) Several config iterations later, which were taken by switching in some
modules I found obvious to be included one by one, I've finally reached
the showstopper: USB support.

4) Then I wandered: the problem must be in one of plugged USB devices. And
right I was: my WACOM GRAPHIRE2 USB tablet was the culprit. Strange
enough, the hid and wacom modules weren't compiled in yet. Some more
iterations later, and it didn't matter if those modules are in or not: if
the tablet is plugged in at boot time the VP+SMP combination freezes.

5) Incidentally I found that I must unplug the tablet at boot time of
freshly built VP+SMP kernel. Then I found that installing the linuxwacom
project [http://linuxwacom.sourceforge.net] drivers, which adds some
changes to mousedev (built-in), evdev and wacom kernel modules, I end up
with a kernel that I can boot and run later already with the tablet
plugged in.

6) Now that had found the major showstopper, I decided to go for audio:
among some other thingies, switched ALSA sound modules on, included the
realtime-lsm patch and built what comes to be my latest VP+SMP working
kernel. And it boots OK. Great.

7) Now let's start jackd... start some client applications, hear some
sound, and... horror! The system hangs completetly. The time it takes to
hang is by no means deterministic. Soon or later it hangs. Hard-reboot is
always the only way around, no magic-sysrq :( Gasp, I've seen this before.

8) Indeed, only by disabling both softirq and hardirq preeemption I get an
usable VP+SMP kernel. But that's no surprise either, it has been always
like that until Q3, which was the latest VP+SMP combination that didn't
suffer with the Wacom tablet presence at boot/init time. I only hoped the
(soft|hard)irq trouble would be solved by R9 time.

Nevertheless, I'm now a lil'bit happier, now that I've catched up on this
VP promise :)

Thanks for the patience.
-- 
rncbc aka Rui Nuno Capela
rncbc@xxxxxxxxx

P.S. Included is my latest .config.

Attachment: config-2.6.9-rc1-bk12-R9.6smp.gz
Description: application/gzip-compressed