Failure analysis: Board that says "Failed 9/6/06"

Mitch Bradley wmb at firmworks.com
Mon Sep 18 05:48:23 EDT 2006


This reports my debug results for the failed board that was sent to me 
last week.

Identification:  The small bar code tag on the bottom says 0017C4000958

Failure description: The board had a sticky note that says "Failed 
9/6/06".  There was no indication of the system configuration or the 
software being used when the "failure" was detected.  So it's difficult 
to know what "failure" means.

Configuration: The board had two wireless antennae installed.  The SPI 
FLASH contains Insyde BIOS.

Initial tests:
a) Attach CRT and serial port and power on - Inysde BIOS screen comes on.

b) Try to boot my diags off USB key - VSA USB enumeration does not 
report the existence of the USB key, either directly connected or via a 
powered USB 2.0 hub

c) Connect a USB keyboard (directly, no hub) and get control of BIOS via 
F1.  Change boot order so USB HD is first (instead of floppy first).

d) Disconnect keyboard, insert USB key.  VSA USB enumeration still does 
not see it.

e) Plug in ROM emulator in PLCC socket, loaded with my LB + diagnostic 
payload.  Gets about halfway through LB startup then dies after "Call 
real_mode_switch_call_vsm"

f) Another try dies shortly thereafter, saying "PCI: Sanity check failed"

g) Suspecting memory flakiness, I apply the RAM timing patch described 
in trac #108.  (The board has Hynix RAMs).

h) Now my Open Firmware diag boots just fine, turns on the video screen, 
etc.

i) I use it to boot Linux (build from yesterday) from USB key.  vmlinuz 
and initrd load just fine, Linux starts running.  But then I get a bunch 
of messages from the USB stack :  hub 1-0:1.0: connect-debounce failed, 
port 4 disabled
The messages never stop, they just keep coming out every second or so.

So something is funny in USB land.

j) I go back to my Open Firmware diagnostic and look at the USB device 
tree.  Aha!  The Marvell wireless device is not showing up in the tree.

Next step (after getting some sleep): Figure out what's wrong with the 
Marvell wireless.





More information about the Devel mailing list