Re: [Xen-ia64-devel] [PATCH] Fixed VTI domain destruction - Masaki Kanno - com.xensource.lists.xen-ia64-devel

9 messages in com.xensource.lists.xen-ia64-develRe: [Xen-ia64-devel] [PATCH] Fixed VT...

From	Sent On	Attachments
Masaki Kanno	22 Mar 2006 04:23	.patch
Alex Williamson	22 Mar 2006 08:06
You, Yongkang	22 Mar 2006 22:16
Masaki Kanno	22 Mar 2006 23:28
You, Yongkang	23 Mar 2006 00:03
Zhang, Xiantao	24 Mar 2006 03:30
Masaki Kanno	24 Mar 2006 03:55
Tian, Kevin	24 Mar 2006 06:31
Masaki Kanno	24 Mar 2006 06:57

Subject:	Re: [Xen-ia64-devel] [PATCH] Fixed VTI domain destruction
From:	Masaki Kanno (kann...@jp.fujitsu.com)
Date:	03/24/2006 06:57:51 AM
List:	com.xensource.lists.xen-ia64-devel

Hi Kevin, Hi Xiantao,

Thank you for detailed information. I may sleep happily today. :-)

Best regards, Kan

Tian, Kevin wrote:

From: Masaki Kanno

Sent: 2006年3月24日 19:56

Hi Xiantao,

That's great!! What was the cause of a issue?

Hi, Kan, Fortunately we've cleaned up the patch and sent it out. The major problem is that there're two different irq handlers one for VTI and another for non-VTI, due to some historical reason disappearing now. The VTI version wasn't enhanced to handle IA64_IPI_VECTOR when host SMP was added. So when destroying domU, tlb_flush_all will send out an IA64_IPI_VECTOR to the LP where domVTI is running. Since VTI path doesn't handle that IPI vector, it's instead injected back to dom0 and handle_IPI is not called. When this scenario happens, dom0 will fall into dead loop to wait call function started. That's it. :-)

Now we can destroy domU (domU+domVTI), and domVTI (two domVTIs) safely. You may give it a try to see whether working in your side.

Thanks, Kevin

Best regards, Kan

Zhang, Xiantao wrote:

Hi Masaki, Two domain destroy issue has been fixed. The patch will be sent out

next week. Your destroy patch is very helpful to community developers :)

Thanks & Best regards -Xiantao

-----Original Message----- From: xen-...@lists.xensource.com [mailto:xen-...@lists.xensource.com] On Behalf Of You, Yongkang Sent: 2006定3埖23晩 16:04 To: Masaki Kanno Cc: xen-...@lists.xensource.com Subject: RE: [Xen-ia64-devel] [PATCH] Fixed VTI domain destruction

Masaki,

Your patch is really important and great helpful! Needn't sorry. ;) I am playing some auto testing to try to give more accurate result in time.

I also found another similar bug before that patch. If I create 1 xenU and 1 VTI, destroying xenU will cause the whole system hang.

Best Regards, Yongkang (Kangkang) 喟慎

-----Original Message----- From: Masaki Kanno [mailto:kann...@jp.fujitsu.com] Sent: 2006定3埖23晩 15:29 To: You, Yongkang Cc: xen-...@lists.xensource.com Subject: Re: [Xen-ia64-devel] [PATCH] Fixed VTI domain destruction

Hi Yongkang,

I'm sorry. I was hasty. I carried out a test to repeat that created/destroyed only one VTI domain. I don't know the cause why a system was hung up.

Best regards, Kan

You, Yongkang wrote:

Hi Masaki,

It is a great patch! I have tried the latest Cset about the destroying

VTI

Domain. But a bug induced in. :)

If 1 VTI domain, it is okay. But if 2 VTI domain is created, destroy 1

VTI

will

hang whole system.

Best Regards, Yongkang (Kangkang) 牋風

-----Original Message----- From: xen-...@lists.xensource.com [mailto:xen-...@lists.xensource.com] On Behalf

Masaki

Kanno Sent: 2006協3・Q2賦 20:24 To: xen-...@lists.xensource.com Subject: [Xen-ia64-devel] [PATCH] Fixed VTI domain destruction

Hi,

This patch fixed the VTI domain destruction processing. I tested creation/destruction domU. The "domU" which I tested is VTI domain and non VTI domain. There was not the memory leak issue, and VTI domain is not left in a zombie state.

Signed-off-by: Masaki Kanno <kann...@jp.fujitsu.com>

Best regards, Kan