From mboxrd@z Thu Jan 1 00:00:00 1970 From: Boris Ostrovsky Subject: Re: [PATCH v3 5/7] libxl/libxc: Move libxl_get_cpu_topology()'s hypercall buffer management to libxc Date: Mon, 23 Feb 2015 12:59:49 -0500 Message-ID: <54EB6A95.2050108__44067.369742743$1424714511$gmane$org@oracle.com> References: <1423512275-6531-1-git-send-email-boris.ostrovsky@oracle.com> <1423512275-6531-6-git-send-email-boris.ostrovsky@oracle.com> <1424709863.27930.214.camel@citrix.com> <54EB5AC9.7070409@oracle.com> <1424711698.27930.236.camel@citrix.com> Mime-Version: 1.0 Content-Type: text/plain; charset="us-ascii"; Format="flowed" Content-Transfer-Encoding: 7bit Return-path: In-Reply-To: <1424711698.27930.236.camel@citrix.com> List-Unsubscribe: , List-Post: List-Help: List-Subscribe: , Sender: xen-devel-bounces@lists.xen.org Errors-To: xen-devel-bounces@lists.xen.org To: Ian Campbell Cc: wei.liu2@citrix.com, ufimtseva@gmail.com, port-xen@netbsd.org, stefano.stabellini@eu.citrix.com, andrew.cooper3@citrix.com, dario.faggioli@citrix.com, ian.jackson@eu.citrix.com, xen-devel@lists.xen.org, jbeulich@suse.com, keir@xen.org List-Id: xen-devel@lists.xenproject.org On 02/23/2015 12:14 PM, Ian Campbell wrote: > On Mon, 2015-02-23 at 11:52 -0500, Boris Ostrovsky wrote: >> On 02/23/2015 11:44 AM, Ian Campbell wrote: >>> On Mon, 2015-02-09 at 15:04 -0500, Boris Ostrovsky wrote: >>> >>> What is the rationale for this change? >> libxl is not the right place to handle hypervisor-specific details like >> buffer management (most, if not all, of other services that libxl >> provides push these sort of things to libxc). > It is acceptable for callers to do the buffer management themselves in > principal. The primary reason to do so would be high frequency calls > where the bouncing would be unacceptable overhead on every iteration > (i.e. it allows callers to preallocate a single buffer). > > There aren't many such interfaces though and as you say most of them are > in libxc (as it happens). > > Nonetheless the argument for this changw should be made in terms of the > interface not being called frequently and therefore being tolerant of > the bouncing overhead which simplifies things for the caller at the > expense of the performance. > > That's assuming these interfaces are indeed tolerant of the performance > hit... These interfaces should be called once per domain creation or as a query from commandline so they don't appear to be not performance-critical. -boris