Hi, > Hi A'rpi, nice to see you there :) :) > > I've spent few hours experiencing with the watchdog feature of the > > Fujitsu Siemens server board (don't ask the model name, what i know is > > that it uses serverworks CSB5 chipset) using lm_sensors 2.7.0's fscscy > > driver.(I needed it because i have random hangup (once a week, so hard > > to debug...) on a server). > > > > Ok, so there is /proc/sys/dev/sensors/fscscy-i2c-0-73/wdog, containing > > 3 0..255 values for the 3 watchdog registers. > > > > The first one is the time counter, it counts backwards (seems > > write-only, at least you can't read the current counter back). It's in > > 2 seconds base, so writting 30 there means 60 seconds delay. It seems > > whole 0..255 range is supported, so up to 510 seconds. writting 0 > > means immediate hardware reset. > > According to the docs, this is read and write. I'll check the code and > update if required. Yes, in viewpoint of code (and you :)) it's R/W. But the value you can read back is the same as you wrote there, not the current value of the counter (as I expected). > > The second number is unknown to me, it > > doesn't matter what value i put there. > > According to the docs again, it is supposed to be a "state" register, so > it's probably meant to be read from, not written to (also the same docs > say it's read and write). It's always 0. Even if i write there something, i got 0 back. Maybe it's non-zero at the monment of reset :))) > > The third is the control > > register, with flags 16, 32 and 128. If i write only 16 or 144 > > (128+16), it means system reset when the counter reaches 0. If i OR > > 32, it has no effect. > > Do you mean that the 6th bit has no effect, or that setting it to 1 > disables the watchdog? 5th, not 6th Seems it disables the watchdog, at least when i set it, bit 4 has no effect. So only 'working' values for me were 16 and 144 (=128+16), for other values it does nothing when the counter reach 0. (at least i see nothing happening) > > So, the world's simples watchdog using this mainboard: > > > > while true ; do > > echo 30 0 16 > /proc/sys/dev/sensors/fscscy-i2c-0-73/wdog > > sleep 10 > > done ^^^ this is the main point, to get it working :) actually i'm not interested (and i have no time) to do more experimenting with wdog, with the above 4-liner it works as expected form a watchdog. > > it does hardware reset after 1 minutes, if this script is killed or > > system hangup occurs. > > > > Also note, that BIOS has a strange setting, named OS Boot Retry Count, > > set to 0 by default, it changes watchdog behaviour to power off > > instead of reset.(0=poweroff 1..7=reset). It took me a while to find > > this... > > Thanks for reporting your experience (and success). Never used a > watchdog myself, but I know how it works and your explanations make > sense. Same here, I've never used such thing, and I always hoped i will never have to use such thing, but this bastard mainboard keeps crashing and doing unpredictable hangup (tried various stresstests etc) so i had to do sth :) > > Please add the above to the documentation (doc/chips/fscscy), so i can > > save a few hours of resetting for other people :) > > It will be done :) thanks :) i hope it helps somebody... before i started to experiment on a live production server i searched through the net for this info but found nothing :( A'rpi / Astral & ESP-team -- Developer of MPlayer G2, the Movie Framework for all - http://www.MPlayerHQ.hu
