On Mon, 2014-10-20 at 21:03 +0200, Knut Omang wrote: > On Mon, 2014-10-20 at 20:18 +0200, Jan Kiszka wrote: > > On 2014-10-20 17:15, Knut Omang wrote: > > > On Mon, 2014-10-20 at 17:23 +0300, Michael S. Tsirkin wrote: > > >> WRT intel_iommu, it does not yet seem to be as fully functional as I > > >> hoped. People also discussed the best way to handle virtio versus iommu > > >> (it bypasses it ATM). > > >> I'd like to suggest we hide the iommu from the command line > > >> help for 2.2, this way people don't activate it mistakenly. > > >> > > >> Let's make it more complete and then enable for 2.3. > > >> > > >> Thoughts? > > > > > > Note that you have to explicitly enable iommu support in the guest > > > (intel_iommu=on on the boot command line in the Linux case) for it to > > > have any effect apart from being visible in the DMAR table and logging > > > so it should not really do any harm. > > > > > > From my perspective the feature works well and I have been running a few > > > virtual machines with div.network workload stable using the additional 4 > > > patches referred to in this post (Jan's two for interrupt remapping and > > > two bug fixes and enhancements from me for running behind bridges) : > > > > > > https://lists.nongnu.org/archive/html/qemu-devel/2014-08/msg02986.html > > > > > > Latest version here: > > > > > > https://github.com/knuto/qemu/tree/sriov_patches_v2 > > > > > > Give me a hint and I can rebase and post the two iommu patches, I > > > believe Jan wanted to do some more work on the interrupt remapping > > > first. > > > > You should avoid to depend on my series regarding upstreaming of fixes > > or features that can be done independently. Did your bridging fixes > > depend on IR? Can you break them up? > > Yes, that was my intention with the comment (to rebase to make my two > commits independent of your interrupt remapping commits) but I realize > the language in my comment was not clear - sorry for the confusion,..
Jan, https://github.com/knuto/qemu/tree/vtd_patches_v2 now contains my iommu fixes right on qemu master (which I just posted to the list) with your interrupt remapping patches rebased on top of this. This was a a bit of work after all so maybe you can save some time from that. Knut
