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

["Jan Beulich" <jbeulich@xxxxxxxx>]

>>> "Liu, SongtaoX" <songtaox.liu@xxxxxxxxx> 06/18/14 5:11 AM >>>
>> >>> On 17.06.14 at 10:33, <songtaox.liu@xxxxxxxxx> wrote:
>> > Full L0 serial log as attached, it was from L0 boot to L2 guest creation.
>> > After creating L2, it takes nearly 15 minutes to print the last line of "
>> > (XEN) chg[040000] d1 [hvm_set_guest_pat+0xad/0xb9]".
>> > No other logs printed out in 20 minutes.
>> 
>> So then I suppose the problem are the two hvm_set_guest_pat() invocations
>> from vvmx.c. Thinking about it, the call from that function to
>> memory_type_changed() seems unnecessary - PAT_CR contents don't
>> influence the decision taken by epte_get_entry_emt(). Tim, Kevin, you had
>> looked at the original patch - would you agree?
>> Songtao - mind giving removal of that call a try to at least verify the 
>> theory?
> 
>I guess your meaning is to remove flush_all(FLUSH_CACHE) in function 
>memory_type_changed.

No, I clearly told you to try removing the call of memory_type_changed() from
hvm_set_guest_pat(). What you did is definitely wrong.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel