Hi Greg, On Nov 20, 2007 5:06 PM, Greg KH <greg at kroah.com> wrote: > On Tue, Nov 20, 2007 at 04:24:19PM -0400, Felipe Balbi wrote: > > > > > > My question is _why_ does it need to be a kernel driver, why can't this > > > all be done in userspace instead? > > > > On musb driver I was using a kernel driver with a sysfs interface and > > a hacked usb cable with 1k resistor between D+/D- and a switchable > > load between Vbus and GROUND (500mA and 100mA). > > Of course this code is musb specific and shouldn't even get integrated > > to linux-omap nor to mainline. > > Why not, have you posted it anywhere? That sounds a bit useful to have, > although why it too would need to be a kernel driver is a bit strange. The driver was mainly written by Tony Lindgren, I re-worked it a bit to include id-pin test as well and you can get it from n800 linux kernel source posted by nokia at http://repository.maemo.org/pool/maemo3.2/free/source/kernel-source-rx-34_2.6.18.orig.tar.gz One should wait until next release come up (OS2008) then you'll get kernel 2.6.21. > > > Having a proper device for running loopback tests (it this is the same > > loopback test I'm wondering :-p) > > would allow us to provide "portable" code that would work with any > > HCD, not only this or that one. > > Ah, I see, your hacked cable is not a "real" device that the HCD could > discover at all. yup. > > And have you looked at the usbtest module? It already provides such a > driver, that works for all HCDs today. I think I was using it with a userspace app which I can't remember the name. > > > Using a sysfs interface (aka linux device driver), looks better to me, > > and that's what I plan to do for USB HSET and USB Command Verifier > > tools for linux. If you pay attention on windows, those tools provides > > a different Host Controller Driver from where they can issue the > > commands they want. > > Are you thinking to use sysfs to send USB commands to a device? You do > know that sysfs is not made for this, right? :) Yeah... and I'm not wondering to use this way. But a sysfs_*_show(), i.e. cat sysfs entry, start a test scenario which would internally handle all usb commands. Or maybe echo testN > $sysfs_entry > > > Well, that's only my opinion. Userspace should only handle the results > > reported by sysfs interfaces or whatever the driver wanna do. > > Again, at first glance, this driver looks like it can be written in > userspace, using usbfs, without any kernel code at all. That's my main > question. If it can be done completely in userspace with usbfs, I can't see why not. But I can't also see why not putting in the kernel and allowing users to build it as a module. > > thanks, > > greg k-h > -- Best Regards, Felipe Balbi felipebalbi at users.sourceforge.net
