From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jan Kiszka Subject: Re: [PATCH 0/4 v9] MSI-X MMIO support for KVM Date: Tue, 06 Sep 2011 10:36:23 +0200 Message-ID: <4E65DB87.5090506@siemens.com> References: <1298019193-12742-1-git-send-email-sheng@linux.intel.com> <20110222180858.GA5784@amt.cnet> <4E5F6C20.7020701@siemens.com> <4E65D149.1070807@redhat.com> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Cc: Marcelo Tosatti , "kvm@vger.kernel.org" , "Michael S. Tsirkin" , Alex Williamson To: Avi Kivity Return-path: Received: from goliath.siemens.de ([192.35.17.28]:33940 "EHLO goliath.siemens.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1753296Ab1IFIga (ORCPT ); Tue, 6 Sep 2011 04:36:30 -0400 In-Reply-To: <4E65D149.1070807@redhat.com> Sender: kvm-owner@vger.kernel.org List-ID: On 2011-09-06 09:52, Avi Kivity wrote: > On 09/01/2011 02:27 PM, Jan Kiszka wrote: >> On 2011-02-22 19:08, Marcelo Tosatti wrote: >>> On Fri, Feb 18, 2011 at 04:53:09PM +0800, Sheng Yang wrote: >>>> Sorry for the long delay, just come back from vacation... >>>> >>>> Change from v8: >>>> 1. Update struct kvm_run to contain MSI-X routing update exit specific >>>> information. >>>> 2. Fix a mmio_nr counting bug. >>>> >>>> Notice this patchset still based on 2.6.37 due to a block bug on assigned >>>> device in the upstream now. >>>> >>>> Sheng Yang (4): >>>> KVM: Move struct kvm_io_device to kvm_host.h >>>> KVM: Add kvm_io_ext_data to IO handler >>>> KVM: Emulate MSI-X table in kernel >>>> KVM: Add documents for MSI-X MMIO API >>> >>> Looks good to me. >> >> What happened to this series and [1]? Forgotten because Sheng switched >> his job? > > I actually have guilty feelings about it now and then. It's just hard > to merge such a complicated change with a non-trivial userspace interface. I agree it's tricky. It will surely require rebasing anyway, thus also a proper re-review. I was also wondering (without locking into the dirty details yet) if that interface couldn't be made more generic to allow fast masking for virtio/vhost as well. > >> I just realized that MSI-X per-vector masking for assigned devices is >> not only slow (takes user space exits) but just simply broken (masking >> bit is not evaluated and applied to the device). > > Isn't that just a bug in qemu? Or something else? The fact the the per-vectors masks aren't evaluated indicates that it's simply not implemented yet. Jan -- Siemens AG, Corporate Technology, CT T DE IT 1 Corporate Competence Center Embedded Linux