On 08/26/2006 01:34 AM, Hans Verkuil wrote:
>On Saturday 26 August 2006 01:55, Chris Schanzle wrote:
>
>
>>Oh, I hate to be the first bearer of bad news...ran with the patch on
>>pristine 0.4.6 sources, encoder died on first or second recording
>>(need to reboot).
>>
>>
>
>Well, it's only half bad since the DMA errors that used to produce MPEG
>corruption are now handled flawlessly, you bomb out on the same DMA
>problem that I haven't been able to fix yet. I think you never got that
>far in the past because the MPEG corruption happened much more
>frequently.
>
Yes, I understand this is CLEARLY progress and is very much appreciated.
>
>What I find very strange is how you can hit that error that I can only
>get by doing really heavy stress testing.
>
>Can you give me a precise description of the hardware you are running
>on? Esp. the chipset and any other cards you have.
>
Asus A8V Deluxe, rev 2.0 motherboard, VIA K8T800Pro/VT8237 chipset, 1 GB
RAM, AMD64 3400+, Hauppauge PVR-250, PCHDTV HD-3000, NVIDIA GeForce FX
5700LE AGP 8X, hda and hdb are Hitachi HDS725050KLAT80 500GB ATA disks,
XFS filesystem on main video partition made of LVM stripe, NEC ND-3500AG
DVD drive (hdc).
Recall trac ticket 48 where I tried an NForce4 Asus A8N Premium + AMD64
X2 4400+ and had the same problems (not to mention worse HD video
playback due to no xv support on pci-express nvidia card).
No other PCI cards (I took out SMC 2802W wireless card a couple months ago).
OS is Fedora Core 4, running 2.6.15-1.1833_FC4smp (I use the SMP kernel
so I could swap in a dual-core AMD 4400+ for testing). Mostly
ATrpms-based Mythtv installation (hand-compiled ivtv drivers, lirc).
> Also the output of cat /proc/interrupts.
>
CPU0
0: 22604092 IO-APIC-edge timer
1: 11250 IO-APIC-edge i8042
4: 4 IO-APIC-edge lirc_serial
8: 1 IO-APIC-edge rtc
9: 0 IO-APIC-level acpi
12: 52202 IO-APIC-edge i8042
14: 1291287 IO-APIC-edge ide0
15: 1963629 IO-APIC-edge ide1
169: 0 IO-APIC-level libata
177: 1266010 IO-APIC-level SysKonnect SK-98xx, ivtv0
185: 676859 IO-APIC-level VIA8237
193: 274267 IO-APIC-level cx88[0], cx88[0]
201: 0 IO-APIC-level ehci_hcd:usb1, uhci_hcd:usb2,
uhci_hcd:usb3, uhci_hcd:usb4, uhci_hcd:usb5
209: 3400621 IO-APIC-level ohci1394, nvidia
NMI: 0
LOC: 22603471
ERR: 0
MIS: 0
>And are you running a cpuspeed daemon? If so, which daemon are you using?
>
>
Yes. Fedora's default "cpuspeed" by Carl Thompson (cpuspeed -h) version
1.2.1 with these arguments: -d -n -p 60 90
Sometimes I'd limit the max CPU speed by adding "-M 2000000" to take off
about 10c temperature (very low-speed fan on cpu heatsink, but temps
never over 60c even when run full speed).
cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies
2200000 2000000 1800000 1000000
>Was this test done with MythTV?
>
Yes. I probably left commercial dection enabled during recording, and
watched some other prerecorded HDTV, perhaps watched some live HDTV
material (which myth records to and plays back from disk, including DMA
from the HD3000). I basically threw a variety of things into the mix to
try to make it break. Sorry if that wasn't what I was suppose to do. :-)
> If you just capture using 'cat', do you get the same behavior?
>
(cat to a file.) It appears to work properly. With mythbackend
stopped, cpuspeed running (and I doubt ever increasing the frequency
from 1000000 since the load average was typically 0.00, max at 0.10), no
messages were logged in a 1 hour timeframe.
I'd be happy to try to make it break without cpuspeed running, but that
doesn't sound like the root cause.
I'll agree I can really pound my system with I/O and DMA traffic, so if
we can get it working here, I think we'll have a very bullet-proof
software package. But since it's typical to get bursts of load by
transcoding, mythfilldatabase (really pounds the disks with the db), or
updatedb, bug reports will continue until we get this all solved.
Thanks again!
Chris
_______________________________________________
ivtv-devel mailing list
[email protected]
http://ivtvdriver.org/mailman/listinfo/ivtv-devel
