On Thu, Dec 21, 2023 at 11:01:56AM +0000, Robin Murphy wrote: > On 2023-12-21 10:42 am, Lukas Wunner wrote: > > On Wed, Dec 13, 2023 at 11:54:05AM +0000, Robin Murphy wrote: > > > I think if we want to ensure ATCs are invalidated on hot-unplug we need an > > > additional pre-removal notifier to take care of that, and that step would > > > then want to distinguish between an orderly removal where cleaning up is > > > somewhat meaningful, and a surprise removal where it definitely isn't. > > > > Even if a user starts the process for orderly removal, the device may be > > surprise-removed *during* that process. So we cannot assume that the > > device is actually accessible if orderly removal has been initiated. > > If the form factor supports surprise removal, the device may be gone > > at any time. > > Sure, whatever we do there's always going to be some unavoidable > time-of-check-to-time-of-use race window so we can never guarantee that > sending a request to the device will succeed. I was just making the point > that if we *have* already detected a surprise removal, then cleaning up its > leftover driver model state should still generate a BUS_NOTIFY_REMOVE_DEVICE > call, but in that case we can know there's no point trying to send any > requests to the device that's already gone. Right, using pci_dev_is_disconnected() as a *speedup* when we definitely know the device is gone, that's perfectly fine. So in that sense the proposed patch is acceptable *after* this series has been extended to make sure hard lockups can *never* occur on unplug. Thanks, Lukas
