RE: [Xen-ia64-devel] [PATCH] NEW_TLBFLUSH_CLOCK_PERIOD_SOFTIRQ is notregistered.

["Xu, Anthony" <anthony.xu@xxxxxxxxx>]

Isaku Yamahata write on 2007年1月29日 15:35:
> Hi Anthony.
> 
> On Mon, Jan 29, 2007 at 02:31:14PM +0800, Xu, Anthony wrote:
> 
>> Could you elaberate the purpose of new_tlbflush_clock_period?
> 
> Tlb flush clock is the optimization to try to remove unnecessary
> flush of mCPU tlb cache or vhpt by NEED_FLUSH().
> It uses a single counter, tlbflush_time, which is incremented when
> mCPU tlb cache or vhpt is flushed.
> When the counter overflows, something must be done.
> new_tlbflush_clock_period() does it.

I don't catch why it can optimize NEED_FLUSH.
Can you  explain more detail?

Anthony.
Thanks,



> 
> 
>> BTW, Intel experienced domain panic or domain disppear after the
>> very long run(after 24 hours). We are investigating this.
> 
> only VT-i domain or only domU or both?
We found that VTI domain panic and disappear after long run.
As kouya pointed out, it seems not related with new_tlbflush_clock_period().




> 
> You can disable tlb flush clock by setting xen_ia64_tlbflush_clock=n
> in xen's configuration file so that you can determine whether the root
> cause is tlbflush clock or not.
> I didn't recently compiled with xen_ia64_tlbflush_clock=n though.
> 
> Or you can shorten the time to call new_tlbflush_clock_period() by
> defining WRAP_MASK as 0x000003FFU (or smaller value) in
> xen/arch/ia64/xen/flushtlb.c.
> Thus you can distinguish the root cause from tlbflush clock.
> 
> 
>> Kouya SHIMURA write on 2007年1月26日 19:06:
>>> Hi
>>> 
>>> NEW_TLBFLUSH_CLOCK_PERIOD_SOFTIRQ is used but not registered.
>>> I've never experienced but system will panic in the very long run.
>>> I wonder why Isaku missed it.
> 
> Ouch!
> It doesn't cause panic, but it just becomes slow.
> So I missed it.

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel