Hi Alexander, David, On Fri, 16 Nov 2007 00:15:21 +0100, Alexander Kiel wrote: > Hi David, > > > I understand the problem now. In the w83627ehf driver, there is this code: > > static ssize_t > > store_pwm_enable(struct device *dev, struct device_attribute *attr, > > const char *buf, size_t count) > > > > ... > > > > if (!val || (val > 2)) /* only modes 1 and 2 are supported */ > > return -EINVAL; > > > > That is the "write error: invalid argument" error. The reason this > > code rejects a write of 0 to pwm_enable is because the w83627ehf does > > not have a true "disable" mode. Some other chips will set the pwm > > output to 100% when they are disabled. The w83627ehf doesn't have a > > disable mode. Of course, setting it to manual and storing 255 in the > > pwm would do the same thing. > > > > That's what pwmconfig (lm_sensors version 2.10.4) does: > > # Try pwmN_enable=0 > > echo 0 > $ENABLE 2> /dev/null > > if [ "`cat $ENABLE`" -eq 0 ] > > then > > # Success > > return 0 > > fi > > > > # It didn't work, try pwmN_enable=1 pwmN=255 > > echo 1 > $ENABLE 2> /dev/null > > echo $MAX > $1 > > if [ "`cat $ENABLE`" -eq 1 -a "`cat $1`" -ge 190 ] > > then > > # Success > > return 0 > > fi Note that the failure can be triggered by two conditions: pwmN_enable not set to 1, or pwmN less than 190. In Alexander's case, pwmN_enable has the expected value: > hwmon0/device/pwm1_enable stuck to 1 Which means that the problem is that writing 255 to pwm1 did not work. No idea why... I agree that the error message could be made clearer, I'll improve it now. > > > > So I *think* the problem is in pwmconfig. But yes, manually writing 0 > > to pwm_enable will give you an -EINVAL. > > > > To be completely honest, I haven't run pwmconfig in a while. I know it > > generally works, but someone should probably post a patch to get it to > > work smoothly with w83627ehf and w83627dhg chips. (I could do that... > > given lots of time :-) It already works just fine for me. I am unable to reproduce Alexander's problem. > > Ok the problem with pwmconfig is the following: > > echo 0 > $ENABLE 2> /dev/null outputs actually 0. This is the 0 from the > echo. echo 3 > $ENABLE 2> /dev/null would output 3. I have no clue why > this is the case. How did you try that, on the command line? $ENABLE is a variable in the pwmconfig script, on the command line it's undefined, so that can't work. Your original report doesn't show any "0" printed on the screen. > > This 0 causes the line echo $MAX > $1 to set pwm1 to 0 instead of 255. I very much doubt it. Both statements are separate, I fail to see how whatever the first one does could have any influence on the second. > If I add a > /dev/null to the line echo 0 > $ENABLE 2> /dev/null it > works as it should. > > So the diff is: > ---------------------------------------------------------------------- > 127c127 > < echo 0 > $ENABLE > /dev/null 2> /dev/null > --- > > echo 0 > $ENABLE 2> /dev/null > ---------------------------------------------------------------------- > > Can you submit this patch? I don't have a reasonable access to the > lm-sensors community. This patch is not correct, it breaks pwmconfig more than it fixes it. As explained above, the problem you have is not with setting pwm1_enable to 1 (that works) but presumably with setting pwm1 to 255. So you're trying to fix the wrong line of the script. -- Jean Delvare
