Hi Hans, Darrick: * Hans de Goede <j.w.r.degoede at hhs.nl> [2007-07-26 18:58:18 +0200]: > Darrick J. Wong wrote: > > On Thu, Jul 26, 2007 at 08:08:09AM -0700, Juerg Haefliger wrote: > >> Hi Darrick, > >> > >> I guess I'm not seeing the big picture :-) The problem you're trying to > >> solve is that the reading of the registers takes a long time and needs to > >> be able to sleep? > > > > Yes, and also that sleeping in the sysfs read function is not desirable. > > Personally, I was fine with that, but Mr. de Goede was concerned about it, > > so I decided to try to fix it. I admit, the dual-delayed-workqueue method > > is a bit strange, but it solves the reading problem at a slight cost of > > waking up the CPU even if nobody's watching the sensors. > > > > I just read in another thread, that other hwmon drivers suffer from similar > problems, so I guess userspace will just have to be able to deal with this, > and the best fix is not to fix it. Sorry to waist everybody's time on this. Yes, and in this case the cure is worse than the disease. As I mentioned in another thread recently, most I2C bus drivers are very slow and polled, using %100 cpu while doing transfers. To put that kind of load on a system periodically just to keep the app from blocking would be IMHO a terrible tradeoff. The root cause of the speed problem is in the I2C bus drivers, not the hwmon drivers. > You could still try to shave some time off by not reading all the limits etc > each update. Notwithstanding what I wrote above, this is still a good idea. But don't go too far with it... we do still want to read/update the limits occasionally. Sometimes that's the only way a user can find out that the BIOS or ACPI or whatever is modifying chip values behind our backs. Regards, -- Mark M. Hoffman mhoffman at lightlink.com
