Oliver Neukum <oneukum@xxxxxxx> writes: > On Tuesday 05 March 2013 11:07:05 Bjørn Mork wrote: >> "Fangxiaozhi (Franko)" <fangxiaozhi@xxxxxxxxxx> writes: >> >> > ------ commit 200e0d99 and commit cd060956, only put the switch >> > command into kernel, instead of userspace usb_modeswitch utility. >> >> Yes. And that is the problem. It was agreed years ago that this >> functionality belongs in userspace. Ref e.g >> https://lkml.org/lkml/2009/12/15/332 >> https://lkml.org/lkml/2010/4/19/216 >> https://lkml.org/lkml/2012/2/28/569 >> >> Please note the re-occurrence of this discussion, despite the fact that >> Matthew's message from 2009 should be quite clear. > > Since 2009 the facts have been changing. Basically we are encountering > two trends > > 1. Power save has become more important > 2. We are seeing devices that are switchable, yet include interfaces other > than communication and virtual storage, foremost real storage in the > form of a microSD-reader This changed with Windows8. You will see _less_ switchable devices in the future. Mode switching is now considered legacy functionality intended for OSes released before 2012. > Whenever we cut power to a switchable device it reverts to the power-on > state. That involves not only the communication functionality (which we > could live with, as it cannot handle loss of power anyway) but also other > interfaces. In addition we need to deal with resets which may or may not > make the device revert to power-on > > This is a basic problem of the design. If you want reset_resume() for power > loss to work you need to do it in kernel space for the same reason we restore > the configuration in kernel space before we rebind the drivers and before > we thaw user space. > > However, this argues not for doing the switch simply in the storage driver > but to switch in the core. OK, I think I understand. You really want us to completely ignore the unswitched mode and just seemlessly restore the switched mode, the same way we just silently fetch the device descriptor and restore the configuration. Well, that does sound interesting. But I don't think we want the tables hard coded in the USB core? Maybe a new userspace ABI or a new driver API? I am not sure this is worth it though. The "other OS" does not support this either. Microsoft's solution to the problem is getting rid of the whole mode switching hell, requiring the devices to support MBIM in their default mode. I believe we are better off working on improving the MBIM userspace support than trying to work around this self imposed firmware issue. Bjørn -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html