On Sat, Jul 29, 2006 at 02:32:02PM +0300, Shem Multinymous wrote: > On 7/29/06, Vojtech Pavlik <vojtech@xxxxxxx> wrote: > > >I think we're hitting a fundamental problem with sysfs/hotplug/udev > >here. It was created to get fixed, non-changing names of devices in > >/dev, so that they'd be easy to enter into configuration files. > > > >Yet applications today want automatic discovery of devices and don't > >want to rely on udev getting the names right. > > > >We should make our minds up, and decide whether we want the 'devices are > >in /dev and applications just need to open the filename' or the 'an > >application will find the device itself' approach. > > I think what people want from device choice is a reasonable default > plus a convenient way to override things. The former is handled nicely > by distributions' udev rules, while the latter is best done by > providing fixed paths. As an end-user, if I know my favorite joystick > is on a specific USB port (hence a specific syfs directory), then I > want to tell neverball "use that one" without setting up nasty udev > rules or playing major:minor matchup. Yes, that's bypassing the Proper > Udevian Way of Doing Things, but it's so much easier and Unix-like > that we really should make it possible (though not by default!). IMO the right way here would be to have a nice GUI for configuring udev included with the distro, that'd let you browse the sysfs tree and point'n'click to create the rule you need. > Security issues aside (for a moment): > Is there any reason not to provide real device inodes on sysfs, > instead of just a textual /sys/foo/dev? And then, maybe udev should > symlink to those device files under /sys instead of creating its own? > This would tie the two systems together rather elegantly. The reason behind this was to force people NOT use sysfs directly when interfacing to the OS. ;) Because sysfs wasn't intended to be an API you can rely on, one that's fixed in stone and cannot be changed for compatibility reasons. I believe it failed in that respect. > >This reminds me very much of the Joerg Schilling discussion (flamewar) > >of enumerating CD-burners. Most people on the kernel mailing list just > >wanted to enter the name of the device node on the cdrecord command > >line. Yet Joerg insisted that the application should do the discovery. > > I think there's a lot more to *that* flamewar - such as unwavering > belief in generic scsi... Sure. It was just one of the points raised there. > >HDAPS, as explained above, doesn't have huge latency impact. The reason > >to have high update rates for input devices (mice nowadays run at 100 Hz > >refresh usually, gaming mice up to 1 kHz), is to not introduce > >additional delay to the user->computer->user closed control loop. > > > >The less delay, the better stability of the control loop and the better > >results in the game. The limiting factor is usually 3D rendering, but a > >10 Hz joystick will still kill the experience by inducing a much larger > >delay. > > Yes, I understand. I just pointed out that in the specific case of > system accelerometer readouts, either the readouts change very slowly > or your laptop is being rattled into an early death. You want the frequent readouts even for slow changes of the direction of gravity there, that's what I wanted to say. > >Sort of a 'reverse select'. > > Exactly. -- Vojtech Pavlik Director SuSE Labs - To unsubscribe from this list: send the line "unsubscribe linux-acpi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html