Hi Joerg, On 07/27/17 at 05:55pm, Joerg Roedel wrote: > On Fri, Jul 21, 2017 at 04:59:08PM +0800, Baoquan He wrote: > > AMD pointed out it's unsafe to update the device-table while iommu > > is enabled. It turns out that device-table pointer update is split > > up into two 32bit writes in the IOMMU hardware. So updating it while > > the IOMMU is enabled could have some nasty side effects. > > > > The only way to work around this is to allocate the device-table below > > 4GB if translation is pre-enabled in kdump kernel. If allocation failed, > > still use the old one. > > Not only for the kdump kernel. The old device table must also be below > 4GB so that its pointer can be updated with a 32bit write. > > If the old table is above 4GB you still need the second write to zero > the upper parts of the pointer in hardware.
Do you mean the allocation of amd_iommu_dev_table in early_amd_iommu_init() also need be addressed for 1st kernel? Seems we don't make sure that for 1st kernel, like adding GFP_DMA32 flag when allocate amd_iommu_dev_table in amd_iommu_dev_table early_amd_iommu_init().
