[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
RE: [Xen-ia64-devel] Xen/IA64 Healthiness Report -Cset#10811:254c090854de
- To: "Isaku Yamahata" <yamahata@xxxxxxxxxxxxx>
- From: "You, Yongkang" <yongkang.you@xxxxxxxxx>
- Date: Fri, 28 Jul 2006 08:55:36 +0800
- Cc: xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
- Delivery-date: Thu, 27 Jul 2006 17:56:18 -0700
- List-id: Discussion of the ia64 port of Xen <xen-ia64-devel.lists.xensource.com>
- Thread-index: Acaxbjnm+8kOTLeyQPOj47iZNBeU6QAcDdRw
- Thread-topic: [Xen-ia64-devel] Xen/IA64 Healthiness Report -Cset#10811:254c090854de
Thank you for investigating the VTI booting issues. In my side, this issue is a
little hard to reproduce. Actually we only see a few times in the past 2 weeks.
Do you often meet this situation?
For the credit schedule, we are preparing to switch schedule from bvt to credit
in our nightly testing. But we also meet some unstable issues in VTI. I want to
have more trying and report some bugs, for example, we found "Oops: time before
its due... " in SMP VTI when using credit. Today we will try more testing with
credit and report the issues.
Could somebody give some comments to this patch? Maybe it fix some potential
issue for VTI domain.
Yongkang (Kangkang) 永康
>From: Isaku Yamahata [mailto:yamahata@xxxxxxxxxxxxx]
>Sent: 2006年7月27日 19:15
>To: You, Yongkang
>Subject: Re: [Xen-ia64-devel] Xen/IA64 Healthiness Report
>On Thu, Jul 27, 2006 at 05:43:36PM +0800, You, Yongkang wrote:
>> Other Issue:
>> 1. SMP_VTI_Boot may stop to boot after "Calibrating delay loop..."
>I'm seeing a similar issue.
>My case is UP VTI domain. VTI domain doesn't share pcpu with dom0.
> By inserting debug print code to vtm_set_itv(), vtm_set_itm() and
> vtm_timer_fn(), I found that vtm_set_itv() wasn't called.
> vITV.m bit isn't cleared so that VTI domain doesn't receives
> timer interrupt. Thus the calibrating delay loop isn't terminated.
> On the other hand Linux/ia64 surely clears itv.m in ia64_cpu_local_tick()
> which is called before the calibration.
>- struct arch_vmx_struct::vtm
> VTI domain code seems to assume that vtime_t::vtm_timer is running on
> same physical cpu of its vcpu.
> Probably the attached patch is needed for credit scheduler.
> It is only compile tested.
>> 2. SMP Xen0 couldn't see all vcpus, no matter dom0_max_vcpus=4 or 2.
>> "xm vcpu-list" can show the other vcpus status are "P"
>xenLinux default config has CONFIG_NR_CPUS=4.
>Does increasing it up to 8 or 16 make difference?
Xen-ia64-devel mailing list