Bug 10780 ("nohz=off"required) - kernel will boot with "nohz=off" only
Summary: kernel will boot with "nohz=off" only
Status: CLOSED CODE_FIX
Alias: "nohz=off"required
Product: Platform Specific/Hardware
Classification: Unclassified
Component: x86-64 (show other bugs)
Hardware: All Linux
: P1 high
Assignee: platform_x86_64@kernel-bugs.osdl.org
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2008-05-23 01:57 UTC by Michael Fischer
Modified: 2010-01-19 17:35 UTC (History)
7 users (show)

See Also:
Kernel Version: 2.6.25.4-2-vanilla
Subsystem:
Regression: No
Bisected commit-id:


Attachments
output of dmesg -s 1000000, gzipped (9.06 KB, application/x-gzip)
2008-05-23 14:51 UTC, Michael Fischer
Details

Description Michael Fischer 2008-05-23 01:57:22 UTC
Latest working kernel version: none
Earliest failing kernel version: 2.6.25.4-2
Distribution: openSUSE
Hardware Environment: AMD ATI SB600 chipset on main board
Software Environment: openSUSE 11.0 x64_86 or i386
Problem Description: system goes into black screen - no movement mode somewhen during the boot process. When booting with "nohz=off" everything works just fine. This holds true for x86_64 as well as for i386 kernels on x86_64 hardware and also for i386 kernels on i386 hardware. It is however not distribution dependent, one will run into the same problem using 2.6.24 or later kernels on e.g. ubuntu.

Steps to reproduce: reboot the system without "nohz=off", it will get into a hanging state every time
Comment 1 Andrew Morton 2008-05-23 09:48:10 UTC
I reassigned this to x86_64 - it's unclear why it was directed at serial ata.

So nohz has never worked on that machine?
Comment 2 Michael Fischer 2008-05-23 10:31:57 UTC
No, never worked, but if you google around you'll see there are plenty of issues around the AMD ATI SB600 chipset support in the current linux kernel. Just most users do not even know what their chipset is. From looking at their product name and description one can then from the vendor site deduct the common denominator: SB600. BTW I am getting vibes and waves that SB700 is not much better supported but I have no system myself for testing it on. The reason, why I directed it at serial ata is because I didn't find a better category (like my first wild guess). What happens before the system comes to a standstill are extremely! slow accesses both on IDE (CD-ROM) side and the serial ata side (hard disk) und this from the very beginning of the boot process.
Comment 3 Michael Fischer 2008-05-23 10:34:52 UTC
But you could assign the bug to i386 as well, it's the same situation there.
Comment 4 Adrian Bunk 2008-05-23 14:32:42 UTC
(In reply to comment #2)
> No, never worked, but if you google around you'll see there are plenty of
> issues around the AMD ATI SB600 chipset support in the current linux kernel.
>...

Kernels <= 2.6.24 contained several bad bugs on SB600 and Google will likely find many people running into them, but 2.6.25 is much better.

Please attach the output of "dmesg -s 1000000" when booting with nohz=off.
Comment 5 Michael Fischer 2008-05-23 14:51:44 UTC
Created attachment 16260 [details]
output of dmesg -s 1000000, gzipped

as requested
Comment 6 Greg Kroah-Hartman 2008-05-23 15:46:43 UTC
Note, this was originally posted in Novell's bugzilla at https://bugzilla.novell.com/show_bug.cgi?id=388583

I'll track this here for now until there is a resolution.
Comment 7 Adrian Bunk 2008-05-25 12:06:26 UTC
I don't see any obvious problem in the dmesg of the working kernel.

Thomas, can you look at this bug?
Comment 8 Thomas Gleixner 2008-05-26 00:58:53 UTC
Michael,
can you please disable CONFIG_CPUIDLE in the kernel config ?

Thanks,
       tglx
Comment 9 Michael Fischer 2008-05-26 01:16:03 UTC
Don't take me for rude but my time schedule does not allow for making kernels. I did not even make the vanilla kernel we are talking about here. What I can offer (as I got virtually unlimited WAN bandwith in my office) is to download any kernel you would like me to test from whatever URL you provide me with. What also can do (within reason) try other boot parameters and report the effect. Sorry!
Comment 10 Michael Fischer 2008-05-26 02:27:07 UTC
Just for completeness: download a kernel means either an RPM for opensuse 11.0 or a deb for ubuntu 8.04. Maybe Greg can let the Novell people build a CONFIG_CPUIDLE=n vanilla kernel, which I am going to test?
Comment 11 Greg Kroah-Hartman 2008-07-24 09:51:50 UTC
No, I can't provide such a kernel rpm package at the moment, due to travel and conferences right now.  Can you just rebuild your kernel by hand?
Comment 12 Vedran Miletić 2009-09-11 20:33:54 UTC
I'm pretty sure this is no longer relevant, because I have seen at least 5 different mainboards with SB600 running lastest distributions without problems, and there was no need to disable anything.

Note You need to log in before you can comment on or make changes to this bug.