Comment 34 for bug 842560

> IIRC upstart nukes some processes after initrd has run and just before the
> rootfs is mounted.

No, it does not. the /usr/share/initramfs/scripts/init-bottom/udev script signals udev to quit with 'udevadm control --exit', which causes udev to signal its workers and wait up to 60 seconds for them to finish up.

The logs for this (apparently widely reproducible) bug show that a udev thread spends a full 60 seconds waiting for the bnx2 firmware to be loaded, and at the end it times out and kills the process, leading to these messages:

[ 2.218105] bnx2 0000:09:00.0: eth0: Broadcom NetXtreme II BCM5708 1000Base-T (B2) PCI-X 64-bit 133MHz found at mem f4000000, IRQ 16, node addr 00:1c:23:bd:ed:e3
[ 2.218178] bnx2 0000:05:00.0: PCI INT A -> GSI 16 (level, low) -> IRQ 16
<snip>
[ 62.944047] bnx2: Can't load firmware file "bnx2/bnx2-mips-06-6.2.1.fw"
[ 62.944087] bnx2 0000:05:00.0: PCI INT A disabled
[ 62.944107] bnx2: probe of 0000:05:00.0 failed with error -2

I don't know what this worker thread is doing while it's supposed to be loading this firmware. I also don't have any hardware to reproduce this on. Were you able to reproduce it with your AMD server?