[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [v6][PATCH 2/2] xen:vtd: missing RMRR mapping while share EPT
On 2014/9/18 17:09, Jan Beulich wrote: On 30.07.14 at 03:36, <tiejun.chen@xxxxxxxxx> wrote:--- a/xen/drivers/passthrough/vtd/iommu.c +++ b/xen/drivers/passthrough/vtd/iommu.c @@ -1867,8 +1867,14 @@ static int rmrr_identity_mapping(struct domain *d, while ( base_pfn < end_pfn ) { - if ( intel_iommu_map_page(d, base_pfn, base_pfn, - IOMMUF_readable|IOMMUF_writable) ) + if ( iommu_use_hap_pt(d) ) + { + ASSERT(!iommu_passthrough || !is_hardware_domain(d)); + if ( set_identity_p2m_entry(d, base_pfn) ) + return -1; + } + else if ( intel_iommu_map_page(d, base_pfn, base_pfn, + IOMMUF_readable|IOMMUF_writable) ) return -1; base_pfn++; }So I gave this a try on the one box I have which exposes RMRRs (since those are for USB devices I also used your patch to drop the USB special casing as done in your later patch series, and I further had to fiddle with vtd_ept_page_compatible() in order to get page table sharing to actually work on that box [I'll send the resulting patch later]) - with the result that passing through an affected USB controller (as expected) doesn't work anymore. Which With or without these two patches, USB always can be passed through in my platform. Note I'm using ubuntu as a Guest OS. raises the question why the two patches alone would work at all. Could you please share information on the address ranges specified by the RMRR(s) in your case? I simply wonder whether things just (XEN) [VT-D]dmar.c:807: found ACPI_DMAR_RMRR: (XEN) [VT-D]dmar.c:383: endpoint: 0000:00:14.0(XEN) [VT-D]dmar.c:676: RMRR region: base_addr ab805000 end_address ab818fff (XEN) [VT-D]dmar.c:807: found ACPI_DMAR_RMRR: (XEN) [VT-D]dmar.c:383: endpoint: 0000:00:02.0(XEN) [VT-D]dmar.c:676: RMRR region: base_addr ad000000 end_address af7fffff happen to work for you on the particular system(s) you're testing on, as I'd generally expect an address space collision to be possible for any RMRR. I think you understand the consequences: If the series here has no way of reliably working without the other one, "iommu=no-sharept" This already is our known way but we need to support the PT in both shared and non-shared cases. is going to be the solution for you, at once being one more argument towards dropping page table sharing altogether. The one argument in favor of the two patches here would be that they at least detect the collision now, thus forcing people to suppress page table sharing. Sorry this sort of estimate is out of the scope I can answer properly. Maybe Yang or Kevin can do follow this if possible. But what's worse, I can't see how the non-sharing case is being handled correctly either (independent of the series here): rmrr_identity_mapping() blindly overwrites what may already be in the page tables, breaking consistency with the CPU-side P2M (iiuc this is a problem even for PV, including Dom0). Plus there's nothing being done to prevent subsequent overwriting of these 1:1 entries by "normal" P2M manipulations. All in all another I'm not sure this as well. Yang and Kevin, What are your comments about this point? argument not to allow (at least by default) passing through of devices associated with one or more RMRRs. Here I have to wait Kevin and Yang's comments since they're familiar with these internals than me :), then I can try to figure out appropriate ways to fix these arguments if they really exist. Thanks Tiejun Jan _______________________________________________ Xen-devel mailing list Xen-devel@xxxxxxxxxxxxx http://lists.xen.org/xen-devel
|
Lists.xenproject.org is hosted with RackSpace, monitoring our |