On Wed, Jun 17, 2020 at 12:53:03PM -0700, Rajat Jain wrote: > Hi Greg, Christoph, > > On Wed, Jun 17, 2020 at 12:31 AM Christoph Hellwig <hch@xxxxxxxxxxxxx> wrote: > > > > On Tue, Jun 16, 2020 at 12:27:35PM -0700, Rajat Jain wrote: > > > Need clarification. The flag "untrusted" is currently a part of > > > pci_dev struct, and is populated within the PCI subsystem. > > > > Yes, and that is the problem. > > > > > > > > 1) Is your suggestion to move this flag as well as the attribute to > > > device core (in "struct device")? This would allow other buses to > > > populate/use this flag if they want. By default it'll be set to 0 for > > > all devices (PCI subsystem will populate it based on platform info, > > > like it does today). > > > > > > OR > > > > > > 2) Are you suggesting to keep the "untrusted" flag within PCI, but > > > attach the sysfs attribute to the base device? (&pci_dev->dev)? > > > > (1). As for IOMMUs and userspace policy it really should not matter > > what bus a device is on if it is external and not trustworthy. > > Sure. I can move the flag to the "struct device" (and likely call > it "external" instead of "untrusted" so as to make it suitable for > more use cases later). The buses can fill this up if they know which > devices are external and which ones are not (otherwise it will be 0 by > default). The PCI can fill this up like it does today, from platform > info (ACPI / Device tree). Greg, how does this sound? That's fine, convert USB over to use it at the same time if you get the chance :) thanks, greg k-h
