From mboxrd@z Thu Jan 1 00:00:00 1970 From: Konrad Rzeszutek Wilk Subject: Re: [PATCH 0/4] xen: events: improve event channel IRQ allocation strategy. Date: Tue, 11 Jan 2011 13:34:59 -0500 Message-ID: <20110111183459.GA29378@dumpdata.com> References: <1294766389.3831.5902.camel@zakaz.uk.xensource.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Return-path: Content-Disposition: inline In-Reply-To: <1294766389.3831.5902.camel@zakaz.uk.xensource.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xensource.com Errors-To: xen-devel-bounces@lists.xensource.com To: Ian Campbell Cc: Jeremy Fitzhardinge , xen-devel , Stefano Stabellini List-Id: xen-devel@lists.xenproject.org On Tue, Jan 11, 2011 at 05:19:49PM +0000, Ian Campbell wrote: > The following series changes the Xen event channel infrastructure to > make better use of the core IRQ allocation interfaces and simplifies > our attempts to avoid clashes between GSIs and event channels. > > The two most interesting changes are: > > * PV domU IRQ numbers are now completely dynamically allocated, > including those associated with PCI passthrough devices. Since > the guest sees no equivalent to GSI space there is no need to do > anything clever in this case like maintaining a 1-1 mapping > between GSI and IRQ. (this is impossible anyway since a PV guest > has no idea what the largest possible GSI is). > > * PV dom0 and HVM guests now completely segregate GSIs from > dynamically allocated IRQs. In this case IRQs for GSIs are > allocated 1-1 beneath nr_irq_gsi (similar to native) and dynamic > IRQs (event channels, MSIs etc) are allocated above. This > simplifies the IRQ allocator code enormously. > > The series has been tested as: > * PV guest with a PCI passthrough device. Which type of domain0? A 2.6.37 + Xen PCI backend + your patches? Or the older 2.6.32? > * PV domain 0. > * HVM guest with and without XENFEAT_hvm_pirqs. > > It appears to do the correct thing in each case. Nice.. > > I also tried HVM with PCI passthrough, which wasn't too successful due > to my hardware not having an IOMMU, but it did appear to setup the IRQ > mapping correctly at least. > > I don't intend this for 2.6.38 at this stage since it looks like: > 67b0ea2bdcd7 xen/irq: Don't fall over when nr_irqs_gsi > nr_irqs. > d1b758ebc2a8 xen/irq: Cleanup the find_unbound_irq > are sufficient (BTW this series follows those). > > Ian. > > > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.xensource.com > http://lists.xensource.com/xen-devel