The realtime kernel is unstable on VIA MII (modules crash, i/o problems, does not shut down)

[Johan Ehnberg]

I tried the kernel on my Via MII 1,2 Ghz Mini-ITX, but encountered quite a few problems.

The kernel reports a lot of crashes, which are most probably the cause of much of this. See the attachment.
I can't get good latencies with the programs I want, such as jackd. In fact, latencies are better with the generic kernel.
I can't get good I/O performance like with the generic kernel, such as streaming video to mythfrontends.
I can't shut down the computer, it hangs at one point. A Ctrl+C gets me a bit further but it still hangs.

I'd be happy to help out with testing -rt on this hardware. I will also soon test the Hardy kernel to see if there's any difference.

[Alessio Igor Bogani (abogani)]

Thank you for taking the time to report this bug and helping to make Ubuntu better. It would help us greatly if you could test with Jaunty Jackalope! Thanks again and we appreciate your help.

[Johan Ehnberg (johan-ehnberg)]

Jaunty beta is a big improvement here. I tested it for a day and it runs stable, it even handled the oom nicely. But it still is not ready. There was one serious oops and the latency is pretty much the same as with the generic kernel. In fact, I got more xruns with rt in jackd than I get with intrepid generic.

[Johan Ehnberg (johan-ehnberg)]

The oops seems to be a bit similar as the one in bug #341468. Also, I believe this should be tested on another MII board - could someone confirm?

[Alessio Igor Bogani (abogani)]

@Johan,

Could you pinpoint the video card and the video module in use, please?

Thanks!

[Johan Ehnberg (johan-ehnberg)]

(note: these are from the stable 8.10-generic, ask for the -rt if necessary as it requires some time on my part)

lspci -v:
01:00.0 VGA compatible controller: VIA Technologies, Inc. VT8623 [Apollo CLE266] integrated CastleRock graphics (rev 03)
 Subsystem: VIA Technologies, Inc. VT8623 [Apollo CLE266] integrated CastleRock graphics
 Flags: bus master, 66MHz, medium devsel, latency 32, IRQ 12
 Memory at d8000000 (32-bit, prefetchable) [size=64M]
 Memory at dc000000 (32-bit, non-prefetchable) [size=16M]
 [virtual] Expansion ROM at dd000000 [disabled] [size=64K]
 Capabilities: <access denied>
 Kernel driver in use: vt8623fb
 Kernel modules: vt8623fb

And the X driver is openchrome, reporting:
(--) CHROME(0): Chipset: CLE266
(--) CHROME(0): Chipset revision: 17
(--) CHROME(0): Probed amount of VideoRAM = 65536 kB
...
(WW) CHROME(0): Manufacturer plainly copied main PCI IDs to subsystem/card IDs.
(--) CHROME(0): Detected VIA VT3122 (CLE266)-EPIA M/MII/....

I also noted a new thing in the jaunty kernel output: The CPU should be supported:
[ 0.000000] KERNEL supported cpus:
[ 0.000000] Intel GenuineIntel
[ 0.000000] AMD AuthenticAMD
[ 0.000000] NSC Geode by NSC
[ 0.000000] Cyrix CyrixInstead
[ 0.000000] Centaur CentaurHauls
[ 0.000000] Transmeta GenuineTMx86
[ 0.000000] Transmeta TransmetaCPU
[ 0.000000] UMC UMC UMC UMC

/proc/cpuinfo snippet:
vendor_id : CentaurHauls
cpu family : 6
model : 9
model name : VIA Nehemiah
stepping : 8
cpu MHz : 1199.778
cache size : 64 KB

[Alessio Igor Bogani (abogani)]

@Johan,

You should do this test:

1) Locate via.ko (of the 2.6.28-3-rt kernel) module and move it out (in you home for example)
2) Boot in that kernel:
2a) Execute sudo depmod -a
2b) Configure Xorg for use vesa video driver

Reboot again and try (check with lsmod that via.ko isn't loaded).

Thanks!

[Johan Ehnberg (johan-ehnberg)]

Thanks for the effort - I tried that! However, the oops still happens. I also noticed there's no need to use the vesa driver, chrome still works but without some acceleration. (The oops is very early on, way before the X driver could play any role in this.)

I'm now running Jaunty Beta and have the -rt kernel alongside so testing should be faster.

If it makes any difference, the box is running off of usb flash, curretly in a 2x8GB mostly raid0 configuration.

[Alessio Igor Bogani (abogani)]

Johan,

Do you already tried Karmic?

Thanks!

[Johan Ehnberg (johan-ehnberg)]

So far so good - the kernel does not produce any oops during bootup. I'll make it the default kernel and see if something turns up during regular use. When I get the time, I'll make sure the performance issue is gone as well.

[Johan Ehnberg (johan-ehnberg)]

I've used the 2.6.31-7-rt and 9-rt kernels for mythTV and as a live synth module, with no oopses. I can get fairly low latencies too, and the remaining challenges (such as fluidsynth zombifying) don't seem to be a kernel problem. I consider this bug fixed as of Karmic beta. Great work, wherever it happened!

[Alessio Igor Bogani (abogani)]

Johan,

Thank you for your great work as tester!