From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Jan Beulich" Subject: Re: [PATCH v2] Re-enable LTR/OBFF when device is owned by pciback Date: Tue, 08 May 2012 10:41:44 +0100 Message-ID: <4FA906780200007800082364@nat28.tlf.novell.com> References: <403610A45A2B5242BD291EDAE8B37D300FDAD88C@SHSMSX102.ccr.corp.intel.com> <20120504132046.GD26418@phenom.dumpdata.com> <403610A45A2B5242BD291EDAE8B37D300FDADF10@SHSMSX102.ccr.corp.intel.com> <20120507135410.GD361@phenom.dumpdata.com> <403610A45A2B5242BD291EDAE8B37D300FDBDA79@SHSMSX102.ccr.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <403610A45A2B5242BD291EDAE8B37D300FDBDA79@SHSMSX102.ccr.corp.intel.com> Content-Disposition: inline List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Xudong Hao , Konrad Rzeszutek Wilk Cc: xen-devel List-Id: xen-devel@lists.xenproject.org >>> On 08.05.12 at 11:05, "Hao, Xudong" wrote: >> -----Original Message----- >> From: Konrad Rzeszutek Wilk [mailto:konrad.wilk@oracle.com] >> On Sun, May 06, 2012 at 07:35:47AM +0000, Hao, Xudong wrote: >> > > From: Konrad Rzeszutek Wilk [mailto:konrad.wilk@oracle.com] >> > > Don't you need to disable this when the device is un-assigned from the >> guest? >> > > >> > >> > I don't think this need to be disabled when device is un-assigned from > guest. >> If host want to use device again, the host device driver need re-load, so >> whether disabling ltr/obff is up to host device driver. >> >> What if the driver isn't doing that? > > Make it clear, here host side do not be considered, host has its own driver. > The only thing is to make sure ltr/obff is enabled before assigning guest, so > that benefit on power. > > Since device is owned by pciback again when it un-assigned from guest, we > need not disable explicitly. As you didn't answer my respective earlier question - _if_ this is a feature needing enabling (and parameter tweaking), I'd imply there are possible incompatibilities (i.e. reasons for not enabling this always), and hence this shouldn't be done universally (and with fixed values for the parameters) _and_ should be properly reset. Jan