On Mon, Feb 12, 2024 at 02:04:31PM -0500, Alan Stern wrote: > On Mon, Feb 12, 2024 at 01:22:42PM +0100, Oliver Neukum wrote: > > On 04.02.24 11:52, Laurent Pinchart wrote: > > > Hi Ricardo, > > > > > > Thank you for the patch. > > > > Hi, > > > > sorry for commenting on this late, but this patch has > > a fundamental issue. In fact this issue is the reason the > > handling for quirks is in usbcore at all. > > > > If you leave the setting/clearing of this flag to a driver you > > are introducing a race condition. The driver may or may not be > > present at the time a device is enumerated. And you have > > no idea how long the autosuspend delay is on a system > > and what its default policy is regarding suspending > > devices. > > That means that a device can have been suspended and > > resumed before it is probed. On a device that needs > > RESET_RESUME, we are in trouble. > > Not necessarily. If the driver knows that one of these devices may > already have been suspend and resumed, it can issue its own preemptive > reset at probe time. > > > The inverse issue will arise if a device does not react > > well to RESET_RESUME. You cannot rule out that a device > > that must not be reset will be reset. > > That's a separate issue, with its own list of potential problems. > > > I am sorry, but it seems to me that the exceptions need > > to go into usbcore. > > If we do then we may want to come up with a better scheme for seeing > which devices need to have a quirk flag set. A static listing probably > won't be good enough; the decision may have to be made dynamically. I don't mind either way personally. Oliver, could you try to find a good solution with Ricardo ? I'll merge the outcome. -- Regards, Laurent Pinchart
