I've got my usb mouse working with the new moused. Most of the code in the module is ripped from usbhidctl(1). Now that I *actually* know how things work, I found that the read() wasn't actually one with side effects. Good.

You'll need the new moused framework for this module.

Basic instructions:

1. Download my moused framework (newpsm project on this site)
2. Recompile your kernel without "device ums"
3. Build moused, build the usbmouse module.
4. Run moused: ./moused -m usbmouse -d /dev/uhid0

If it doesn't work, let me know.

moused-usbmouse.tar.gz

Future plans:

• Rewrite ugly bits
• Port ums(4) hacks for broken mice
• Test with more mice
• Write uhid joypad support

I started working on usb (usbhid) mouse support for my newpsm/newmoused project. So far so good, I can probe /dev/uhidX devices and figure out if it's a mouse or not. Thankfully, usb hid is very easy. Unfortunately, FreeBSD (NetBSD's) implementation is somewhat crazy with the voodoo.

To see what I mean, look at: src/usr.sbin/usbhidct/usbhid.c.

Look near line 230 (the only read() call in the file). Notice the read() call, but 'dbuf' is NEVER used meaningfully. More specifically, the read is mostly ignored and a loop is done over the 'hids' list. Looks like read()s on usbhid devices pushes data to special places in memory which can be found by using hid_start_parse(). A read() shouldn't be doing this. This is the job for an ioctl() or something. read(), to me, says "give me data so I can use it" - not "update some magic places in memory, thanks"

Eek.