Re: [Xen-users] [Xen VGA Passthrough] DomU performance inversely proportional to VM memory size when VGA passthrough is enabled

[Andrej Komelj <akomelj@xxxxxxxxx>]

Hi Ian,

I knew I needed another set of eyes to go through everything. ;-)

YES, disabling HAP solved it!!!

I remember trying it once in the past few days of banging my head
against the table, but I guess I had set shadow memory size to some
low value at that time and the effect wasn't noticeable. With plenty
of shadow memory and no HAP, my domU is usable again. :-D

Thank you very much for you effort!

Best regards,
Andrej

On Sun, Sep 14, 2014 at 9:58 PM, Ian Campbell <ian.campbell@xxxxxxxxxx> wrote:
> On Sun, 2014-09-14 at 21:40 +0200, Andrej Komelj wrote:
>> Does anybody know what else I could try/check?
>
> This sounds pretty odd to me. As a wild stab in the dark I wonder if
> enabling passthrough means that some operation now has to be trapped and
> handled in the hypervisor in a way which requires walking over the
> entire (populated) address space, that would be an operation which was
> proportional to the RAM size and if it was happening often could explain
> what you are seeing.
>
> I'm not really up to date on x86 HVM stuff, but for example I think
> passthrough vs. none makes some difference to the default cacheability
> attributes used for RAM in the second stage page tables, and there are
> some cache enable/disable operations which then get more/less expensive
> due to that (I'm really fuzzy on the details as you might be able to
> tell!).
>
> From your logs it looks like you are running with HAP enabled. It might
> be worth trying without (i.e. using shadow mode) just to see if it is
> hap specific or not. (I think hap=0 in your guest cfg does this)
>
> After that you may want to post to the devel list and see if anyone has
> any ideas.
>
> Ian.
>

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