On Fri, Sep 02, 2022 at 03:00:26AM +0000, Limonciello, Mario wrote: > [Public] > > Hi, > > A sysfs file /sys/bus/thunderbolt/devices/domainX/iommu_dma_protection is exported from the kernel and used by userspace to make judgments on whether to automatically authorize PCIe tunnels for USB4 devices. > Before kernel 5.19 this file was only populated on Intel USB4 and TBT3 controllers, but starting with 5.19 it also populates for non-Intel as well. So that's a new kernel feature? > This is accomplished by an assertion from the IOMMU subsystem that it's safe to do so by a combination of firmware and hardware. > > Here is the patch series on top of 5.15.64: > > 3f6634d997dba8140b3a7cba01776b9638d70dff > ed36d04e8f8d7b00db451b0fa56a54e8e02ec43e > d0be55fbeb6ac694d15af5d1aad19cdec8cd64e5 > f316ba0a8814f4c91e80a435da3421baf0ddd24c > f1ca70717bcb4525e29da422f3d280acbddb36fe > bfb3ba32061da1a9217ef6d02fbcffb528e4c8df > 418e0a3551bbef5b221705b0e5b8412cdc0afd39 > acdb89b6c87a2d7b5c48a82756e6f5c6f599f60a > ea4692c75e1c63926e4fb0728f5775ef0d733888 > 86eaf4a5b4312bea8676fb79399d9e08b53d8e71 > > Can you please consider backporting them to 5.15.y+? I don't understand why all of the string helpers are needed just for the last commit, are you sure this is all necessary? And again, this feels like new features are being added that are much more than just a "new device id added". Why not just use 5.19 for this hardware? thanks, greg k-h
