On Sun, 2011-05-08 at 11:48 -0700, Yinghai Lu wrote: > When do pci remove/rescan on system that have more iommus, got > > [ 894.089745] Set context mapping for c4:00.0 > [ 894.110890] mpt2sas3: Allocated physical memory: size(4293 kB) > [ 894.112556] mpt2sas3: Current Controller Queue Depth(1883), Max Controller Queue Depth(2144) > [ 894.127278] mpt2sas3: Scatter Gather Elements per IO(128) > [ 894.361295] DRHD: handling fault status reg 2 > [ 894.364053] DMAR:[DMA Read] Request device [c4:00.0] fault addr fffbe000 > [ 894.364056] DMAR:[fault reason 02] Present bit in context entry is cl > > it turns out when remove/rescan, pci dev will be freed and will get another new dev. > but drhd units still keep old one... so dmar_find_matched_drhd_unit will > return wrong drhd and iommu for the device that is not on first iommu. > > So need to update devices in drhd_units during pci remove/rescan. > > Could save domain/bus/device/function aside in the list and according that info > restore new dev to drhd_units later. > Then dmar_find_matched_drdh_unit and device_to_iommu could return right drhd and iommu. I think I'd vote for saving some kind of representation of the bus hierarchy, we probably don't need to list every possible individual device. Leaving a broken pointer around to be matched up and restored later just seems like a continuation of an idea that was bad to begin with. Thanks, Alex -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
