Do I have to shutdown IO-APIC before Local APIC

[wjian8]

Hi guys,

I found a bug of kernel that seems be related to the shutdown sequence between io-apic and local-apic .

I had reported it at https://lkml.org/lkml/2016/6/27/64 LKML: "Wei, Jiangang": Re: RFC: Fix kdump failed with 'notsc' .

I don't find any descriptions about the sequence of disable IO-APIC and Local APIC in "Intel 64 and IA-32 Architectures software developer's manual volume 3A".

Only erratum AVR31 for "Intel Atom Processor C2000 Product Family Specification Update".

AVR31. Interrupts That Target an APIC That is Being Disabled May Result in a System Hang

Problem: Interrupts that target a Logical Processor whose Local APIC is either in the process of being hardware disabled by clearing by bit 11 in the IA32_APIC_BASE_MSR or software

disabled by clearing bit 8 In the Spurious-Interrupt Vector Register at offset 0F0H from the APIC base are neither delivered nor discarded.

Implication: When this erratum occurs, the processor may hang.

Workaround: None identified. Software must follow the recommendation that all interrupt sources that target an APIC must be masked or changed to no longer target the APIC before

the APIC is disabled.

Status: For the steppings affected, see Table 1, "Errata Summary Table" on page 9.

My machine is Lenovo's QiTianM4340,

and the CPU is Intel(R) Core(TM) i5-3470 CPU @ 3.20GHz， 4Cores.

Do I have to shutdown IO-APIC before Local APIC for Core(TM) series ?

[idata]

Hello, pandawei:

Are you developing something? How did you find out about the possible bug?

Regards,

Amy.

[wjian8]

I just want to fix the failure of kdump with notsc.

I found the bad commit by bisect in kernel upstream.

And I wan to know Other intel processor must disable I/O APIC firstly?

Can you give me a determinate answer to the problem?

BTW, I have sent a patch to fix , but so far, no response. https://lkml.org/lkml/2016/6/29/18 https://lkml.org/lkml/2016/6/29/18

[wjian8]

Ping ..., any feedback.

Thanks

[idata]

What would be the purpose of disabling I/O APIC?

Are you developing something based on this controller? Software? Hardware?

Regards,

Amy.

[wjian8]

When kernel encounter panic (crash), it will disable APIC in the first kernel, and then start the capture kernel that not be interrupted.

Nope, I just want to fix a bug in kernel, which seems be related to the shutdown sequence between IO-APIC and Local APIC.

I cann't find any SPECs for it, but only erratum AVR31 for "Intel Atom Processor C2000 Product Family Specification Update".My question is , Must we disable I/O APIC firstly for any other intel processor

The bug I have show you in the above message.

[idata]

Thank you for the clarification.

Let me review your case, I will update this thread as soon as possible.

Regards,

Amy.

[Ronny_G_Intel]

Hi pandawei,

Are you reporting this issue as a bug on the CPU or as a bug for the Kernel shutdown sequence between io-apic and local-apic?

If you already submitted a patch for the Kernel, I would assume this pointing out to the Kernel and not to the CPU, is this a correct assumption?

Or do you want to know if you have to shutdown IO-APIC before Local APIC?

Thanks,

Ronny G

By the way, I am moving this community over to the Processors forum.

[wjian8]

Thanks for your reply.

I want to know if I have to shutdown IO-APIC before Local APIC.

But It seems that there's no consensus on the shutdown sequence .

Indeed, I put forward a new solution,

which can fix the dump-capture kernel hang and avoid touching the shutdown sequence between io-apic and local-apic .

url: https://lkml.org/lkml/2016/7/11/220 LKML: "Wei, Jiangang": Re: [PATCH v2] kexec: Fix kdump failure with notsc

[Ronny_G_Intel]

Hi pandawei,

Thanks for the information. We appreciate that you share your findings on this community.

And yes, It seems that there is no consensus on the shutdown sequence.

Regards,

Ronny G