Re: [PATCH] x86/EPT: flush cache when (potentially) limiting cachability

From: "Jan Beulich" <JBeulich@suse.com>
To: SongtaoX Liu <songtaox.liu@intel.com>
Cc: Kevin Tian <kevin.tian@intel.com>, "keir@xen.org" <keir@xen.org>,
	"tim@xen.org" <tim@xen.org>, Eddie Dong <eddie.dong@intel.com>,
	Jun Nakajima <jun.nakajima@intel.com>,
	"xen-devel@lists.xenproject.org" <xen-devel@lists.xenproject.org>,
	Yang Z Zhang <yang.z.zhang@intel.com>,
	Robert Hu <robert.hu@intel.com>,
	LongtaoX Pang <longtaox.pang@intel.com>
Subject: Re: [PATCH] x86/EPT: flush cache when (potentially) limiting cachability
Date: Tue, 17 Jun 2014 08:11:35 +0100	[thread overview]
Message-ID: <53A00647020000780001ADD2@mail.emea.novell.com> (raw)
In-Reply-To: <582FB90AB890394081254B69739046FC0EAD8C9D@SHSMSX101.ccr.corp.intel.com>

>>> On 17.06.14 at 04:19, <songtaox.liu@intel.com> wrote:
>> >>> On 12.06.14 at 03:28, <songtaox.liu@intel.com> wrote:
>> >> From: Jan Beulich [mailto:jbeulich@suse.com]
>> >> >>> "Liu, SongtaoX" <songtaox.liu@intel.com> 05/30/14 9:35 AM >>>
>> >> >Sorry to update.
>> >> >The bug refers to
>> >> http://bugzilla.xensource.com/bugzilla/show_bug.cgi?id=1887.
>> >> >L2 guest vnc console hangs at black screen with cursor flashing at
>> >> >the top
>> > left
>> >> corner.
>> >> >After several minutes, the Seabios screen shows up, but the screen
>> >> >still
>> > flush
>> >> slowly, the guest could not boot up.
>> >>
>> >> That would then indeed be an indication of excessive flushes. I.e.
>> >> you may
>> > want
>> >> to simply track down which one (and why) gets triggered too often.
>> >>
>> > How to get that information? It is hard for me to do it since I am not
>> > a Xen expert. But if you can provide a patch help to debug, I would happy 
> to
>> do it.
>> 
>> The original patch makes quite clear where to put the instrumentation; I'm
>> attaching a patch anyway for you to try. This (obviously) is against the 
> current
>> staging tree - while it ought to apply fine on master, it certainly won't 
> apply on
>> 4.4 or older without adjustments.
>> 
> Env: Xen.git, master, commit: 7f8d8abcf6dfb85fae591a547b24f9b27d92272c, with 
> your patch,
> L2 guest with "hap=1", after creating the L2, "xl dmesg" in L0 shows: 
> (XEN) chg[000020] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[000040] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[000080] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[000100] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[000200] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[000400] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[000800] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[001000] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[002000] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[004000] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[008000] d1 [hvm_set_guest_pat+0xad/0xb9]
> (XEN) chg[008000] d1 [hvm_set_guest_pat+0xad/0xb9]

This looks like it's not really a proper part of the log (the last two
lines being duplicated). Which underlines that you should supply
(attach!) full logs rather than decide on your own which part(s)
might be relevant.

_Assuming_ the tail above really is the tail of the log, I can't see
why the guest wouldn't come up: 32k cache flushes aren't the
end of the world. There ought to be something else...

Jan