Most recent kernel where this bug did *NOT* occur: 2.6.20
Distribution: openSUSE 10.2
Hardware Environment: x86_64 (AMD64 Athlon) single-proc, 1 GB RAM
Problem Description: Any kernel after 2.6.20 hangs during boot.
Steps to reproduce: Build a kernel and try to boot it.
I built a kernel with CONFIG_ACPI_DEBUG=y and booted
with acpi_dbg_level=0x10 acpi_dbg_layer=01x1ff.
Boot log output is here:
does it work with "acpi=off"?
> Linux version 2.6.20-git1
Can you try anything newer, such as 2.6.21-rc3, or the latest -mm tree?
2.6.21-rc3: acpi=off: still hangs
pci=nomsi: still hangs
acpi=off pci=nomsi: still hangs
build with CONFIG_PCI_MSI=n: still hangs
2.6.21-rc3-mm2: acpi=off pci=nomsi: still hangs
build with CONFIG_PCIEPORTBUS=n (PCI Express support): hangs
add acpi=off: BOOTS
Duplicated that success on 2.6.21-rc3.
Summary: Combination of PCIEPORTBUS and ACPI causes a hang during boot,
but building a kernel without PCI Express support and booting with
"apci=off" allows me to boot successfully.
There are multiple boot logs for this (all concatenated in one file)
at: http://www.xenotime.net/linux/logs/2621-rc3mm2.bootlog .
Created attachment 10670 [details]
working 2.6.20 boot log
Created attachment 10671 [details]
2.6.20 config file
Created attachment 10672 [details]
failing 2.6.21-rc3 config file
Has PCIEPORTBUS enabled. If I disable this config option and use acpi=off,
that kernel will boot.
2.6.20 does boot with "acpi=off" but 2.6.21 does not.
Git bisect found nothing, so I went back to comparing config files and
kernel boot parameters, which led to User Error. All of my later kernels
were using "netconsole=<parameters>". When I remove this, they all boot fine,
so apparently there is some problem with netconsole and not ACPI. :)