On Tue, Jul 02, 2019 at 10:36:59AM +0800, JC Kuo wrote: > On 7/1/19 4:52 PM, Greg KH wrote: > > On Mon, Jul 01, 2019 at 04:48:48PM +0800, JC Kuo wrote: > >> When usb-storage driver detects a UAS capable device, it ignores the > >> device if CONFIG_USB_UAS is enabled. usb-storage driver assumes uas > >> driver certainly will be loaded. However, it's possible that uas > >> driver will not be loaded, for example, uas kernel module is not > >> installed properly or it is in modprobe blacklist. > >> > >> In case of uas driver not being loaded, the UAS capable device will > >> not fallback to work at Bulk-only-transfer mode. The device just > >> disappears without any notification to user/userspace. > >> > >> This commit changes usb-storage driver to skip UAS capable device > >> only when uas driver is already loaded to make sure the device will > >> at least work with Bulk protocol. > > > > But what happens if the driver is loaded afterward, because 'modprobe' > > was called by the driver core (or it should have been, because this is a > > device that supports that protocol)? > If uas driver is loaded after usb-storage driver probed the device, > the device will still work with Bulk-only protocol, though it can't > make uses of streams. Which is not a good thing, and is what the original code was there to prevent happening. > > I think you just broke working systems :( > > > > Why wouldn't the UAS driver get loaded automatically if it is configured > > in the system as it is today? > An user might want to completely disable uas for some reason so he/she > adds "blacklist uas" to modprobe conf file. I think in case of this, > usb-storage driver has to enable this device with the legacy Bulk-only > protocol instead of ignoring the device. Why would they want to do that? Where are people doing this in ways that breaks their systems? > As an alternative to this patch, I thought I could get uas driver > loaded before usb-storage driver so I tried moving the functions in > drivers/usb/storage/uas-detect.h into uas.c and letting usb-storage > links uas_use_uas_driver() of uas.ko. However, that didn't work > because uas driver actually depends on usb-storage driver for > usb_stor_adjust_quirks(). There will be a recursive dependency. > > Please let me know if there is better approach to avoid the issue. If users blacklist the uas driver, that's their choice and they should rebuild their kernel :) Or better yet, talk to us to get the issue fixed for why they would want to blacklist such a driver. As it is, this patch is not acceptable. thanks, greg k-h
