Mdk 10.1 Community freezes on boot

for us simple beings do we?

no lapic and noapic for kernel appends
if bios has some power options will it make a diff if we disable them oh gurus?

Originally posted by opjose I've seen many a kernel post to the effect that the noapic option should indeed shut off the local apic as well.

I haven't read enough code to be sure exactly how the two options interact, but I don't see anything on the path into detect_init_APIC that depends on the noapic option. Also, I can tell you how the hardware works: the ioapic requires a local apic on each CPU to talk to, so if you shut off the local apic, you necessarily have to shut them both off. However, a local apic can do some limited things without an ioapic. In fact, in a non-multiprocessor-capable system, there may be no ioapic, but the local apic can still be used. I think only the timer part of it is significantly useful without an ioapic.

Also you've pretty much confirmed that the kernel is disregarding the nolapic parameter at boot, as all that patch does is to FORCE apic off totally. Note that the patch does not seem to disable LOCAL apic, but rather it specifies no_apic ... + if (enable_local_apic != 1) + goto no_apic; If Local APIC is not enabled on the command line explicitly, disable APIC altogether...

No, you are reading code way too superficially here. The whole function you are looking at is all about the LOCAL apic, even though it doesn't have the word "local" in every label and variable name. Check the message that is printed when you reach the label no_apic. The ioapic stuff is in io-apic.c.

Sheez this fix is so simple! I wish Mandrake would include it in their release. You may want to post it on their bugtracker as well as this seems to be a rather large problem which needs to be addressed. [/B]

Yes, I'll do that soon. It's a trivial backport from 2.6.9, so they shouldn't have any problem including it in an update kernel.

Originally posted by aus9 for us simple beings do we? no lapic and noapic for kernel appends if bios has some power options will it make a diff if we disable them oh gurus?

Sorry, I don't understand the question. This thread is about one specific issue, where the local apic must not be enabled on certain machines, but some kernel versions try to do it anyway, and the "nolapic" option that is supposed to prevent this behavior doesn't work on some of those kernel versions.

I have no idea whether you have this issue or not. Your profile says you're running Slackware 10.0, and I have no clue what kernel that has. If you see "Local APIC disabled by BIOS -- reenabling" in your boot messages and you have lots of problems after that, maybe you have the issue.

If you do have the issue, the only solutions I know of at the moment are to (a) patch your kernel and recompile, or (b) reconfigure your kernel with some options turned off (sorry, I haven't dug up the details on which ones) and recompile, or (c) upgrade or downgrade to a kernel version that works -- a 2.6.9-based kernel should work, for example.

Originally posted by TimMann I haven't read enough code to be sure exactly how the two options interact, but I don't see anything on the path into detect_init_APIC that depends on the noapic option. Also, I can tell you how the hardware works: the ioapic requires a local apic on each CPU to talk to, so if you shut off the local apic, you necessarily have to shut them both off. However, a local apic can do some limited things without an ioapic. In fact, in a non-multiprocessor-capable system, there may be no ioapic, but the local apic can still be used. I think only the timer part of it is significantly useful without an ioapic.

Yeah, I don't doubt you... merely that a lot of contradictory posts appeared, probably from people who had not looked at the code.

Certainly I haven't had any reason to peruse the APIC code to date...

No, you are reading code way too superficially here. The whole function you are looking at is all about the LOCAL apic, even though it doesn't have the word "local" in every label and variable name. Check the message that is printed when you reach the label no_apic. The ioapic stuff is in io-apic.c.

Ah!

I had assumed that it was calling another function... my bad.

Yes, I'll do that soon. It's a trivial backport from 2.6.9, so they shouldn't have any problem including it in an update kernel.

PLEASE, PLEASE, PLEASE do so!

The shear amount of posts here about "Mandrake Freezing on boot" is enormous and seems to revolve around this issue.

This would be a very welcome fix indeed!!!

TimMann

well spotted, I am a ho....I am getting a new computer soon as wanted to try out mdk.
Actually I will have a burner so may try out heaps but thats another story.

FYI, I've filed this bug in Mandrake's bugzilla at http://qa.mandrakesoft.com/show_bug.cgi?id=12561

Those Aholes:

Did you see the response?

You can drop the theatrics (this problem has been known on cooker for quite a while, but kernel maintainers changed during community), and test this boot ISO image (for people for whom the issue affects installation): http://plf.zarb.org/~bgmilne/10.1/bo...local-apic.iso Since this didn't help on the machine I was testing on (but http://plf.zarb.org/~bgmilne/10.1/bo...c-disabled.iso did), I didn't make it available at the time (19 Oct or so). Of course, we would still need updated kernel packages ... but I wonder if anyone has tried booting an smp-kernel on the affected machines.

And MY response to this stupidity!

Re: http://qa.mandrakesoft.com/show_bug.cgi?id=12561 Theatrics? Are you kidding? This problem is plaguing a lot of people's and does not only affect laptops. There have been no errata's addressing it and it is quite severe. Yeah it breaks SMP, but the problem is usually not with SMP systems. Someone really dropped the ball on this one, citing that it was a known problem on cooker is a rather trite response.