Re: [Xen-devel] [PATCH v2] Re-enable LTR/OBFF when device is owned by pciback

["Zhang, Xiantao" <xiantao.zhang@xxxxxxxxx>]

Hi, Jan
It is over-engineering to add PCI-e support to Qemu  for enabling this feature. 
 Basically, LTR/OBFF is a host PM feature,  and this patch only make this 
feature still work after the device is assigned to one guest.   Our goal is not 
to enable guest's LTR/OBFF support, and just let host not lose this advanced PM 
feature. Maybe we can move the logic to hypervisor like today's ATS feature 
does,  and once hypervisor finds the assigned device has this feature,  it can 
determine whether needs to enable it or not.   Thanks!
Xiantao

> -----Original Message-----
> From: Jan Beulich [mailto:JBeulich@xxxxxxxx]
> Sent: Wednesday, May 23, 2012 11:08 PM
> To: Zhang, Xiantao; Hao, Xudong
> Cc: xen-devel; Konrad Rzeszutek Wilk
> Subject: RE: [Xen-devel] [PATCH v2] Re-enable LTR/OBFF when device is
> owned by pciback
> 
> >>> On 23.05.12 at 16:45, "Zhang, Xiantao" <xiantao.zhang@xxxxxxxxx>
> wrote:
> > Okay,  I understand your concern about this implementation, and you think
> > ltr/obff should be enabled by guest instead of host.   AS we know,
> LTR/OBFF
> > is a special PCI-e feature for optimizing the whole system's  PM, and
> > to enable it, the whole system(including root port and all upstream
> > bridges or
> > ports) should have these two capabilities. If anyone in this chain
> > doesn't have or enable  these two capabilities, the whole system would fail
> to
> > benefit from this optimized PM.   In addition, since Qemu doesn't have PCI-
> e
> > support, and also hasn't such  LTR/OBFF capabilities supported,  guest
> > shouldn't be able to enable this feature for its any device.
> > Certainly, we can change linux's pci interface and para-virtualize
> > this two features, in this way, guest can ask host to enable this
> > feature, but how to handle Windows guest's case ?  We can't change or
> > para-virtualize Windows's PCI interface, I think. Do you have good
> suggestion or proposals?  Thanks!
> 
> Fix qemu to support PCI-e (and then make use of the to be added PV
> interface).
> 
> Jan


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