On 31.07.2019 10:58, Andrew Cooper wrote: > On 31/07/2019 09:34, Jan Beulich wrote: >> On 30.07.2019 19:56, Roman Shaposhnik wrote: >>> On Fri, Jul 26, 2019 at 1:06 AM Jan Beulich <jbeul...@suse.com> wrote: >>>> On 23.07.2019 20:25, Roman Shaposhnik wrote: >>>>> Interestingly enough, adding iommu_inclusive_mapping=1 AND iommu=debug >>>>> booted the system just fine. >>>> Btw (I've noticed this only now) - are you saying without "iommu=debug" >>>> the box does _not_ boot fine, despite the other option? >>> Yes. But it made sense to me since iommu=debug (as per your >>> explanation) overwhelms the CPU and I guess adding >>> iommu_inclusive_mapping=1 avoids the code path that does it? >> I'm afraid I don't follow: My question was whether >> "iommu_inclusive_mapping=1" alone would not allow the box to boot. >> Without "iommu=debug" there's no excessive logging afaict, no >> matter what other IOMMU options you use. > > Without inclusive mappings, the system boots but the screen is junk and > there are DMA faults all over the place. With inclusive mappings, it > all "works" fine. > > With debug enabled, the DMA faults are spat out to the console for a > short while, until an APIC error occurs and the system wedges completely.
Right - the verbosity _with_ "iommu=debug" may be a problem. Hence me wondering whether the system indeed wouldn't boot _without_ that option. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel