On Mon, Mar 25, 2019 at 05:49:07PM -0700, Brian Norris wrote: > > Not to say that this shouldn't be fixed if at all possible, but realize > > that this is not the "normal" case of "we do not break userspace" here, > > given the tool involved, and the apis being used. > > I think I sort of understand what you're going for here, but can you > elaborate so I don't have to assume? To be specific, tools at the "very low level" that are used to configure the kernel for userspace, or to interact with the kernel such that other programs work on top of things properly (like udev), we have broken apis such that we can fix issues with old mistakes and move on to more secure or "correct" apis. So we do change things at this layer at times, and normally no one notices as they have moved on to newer userspace programs in the past 3 years. Sometimes we do keep kernel support for really old userspace systems, until they die out, and then we can finally drop the kernel code. Again, udev has done this over the years as we worked to figure out a sane way to handle stuff. We have rolled back kernel changes until people finally killed their old Fedora 3 ppc32 systems for example :) To be honest, with the USB3 support added to usbip, no one noticed that things broke, and the fact that it took 4 years to notice implies that maybe it wasn't that big of a deal as no one used this. But, as you show, that assumption was not correct, so if we can fix things, we should. Did that help? thanks, greg k-h
