On Sun, Jan 25, 2026 at 03:07:41PM +0100, Danilo Krummrich wrote: > On Sun Jan 25, 2026 at 2:22 PM CET, Laurent Pinchart wrote: > > It's the wrong solution for most cases, if not all. It will spread in > > drivers > > and then become another big piece of technical debt we'll wish we had never > > merged. > > It is a matter of how the revocable pattern is adopted. I.e. I don't think > drivers should create instances of revocable (device) resources by themselves. > Instead, I think it should be up to the corresponding subsystems to adopt the > pattern in the way necessary and make it accessible to drivers instead. > > > We know what the right solution to the cdev race is > > So, what is it? Assuming that this is what you are referring to, how do you > prevent accesses to (potentially freed) device resources after the bus device > has been unbound from the driver for subsystems that may still call back into > the driver after device unbind?
I've answered this question in another e-mail in this thread, see https://lore.kernel.org/all/20260129010822.GA3310904@killaraus/ -- Regards, Laurent Pinchart
