l?r, 17.01.2004 kl. 09.48 skrev Jean Delvare: > > sorry to bother you again, but this didn't tell you anything about > > what's going on? > > Sorry for the delay. Too much work, not enough time. You know the story > I guess. I do. Sorry for buggin' you again, I just figured you had forgot me when you answered other mails on LKML. > > > I forgot to mention it in my last mail, but I sometime has to > > > reload the modules before "sensors" finds any sensor. > > As we load sensor chip drivers, we make sure that the chip we want to > handle is what we think it is. Technically, this means reading from a > few registers and compare the values with what we would expect for this > chip. So the same read errors that make your hardware monitoring values > jump make the chip identification fail sometimes. I see. > > > Attached three runs. Seems to be some read errors :( On these three > > > runs I got first three bus-collisions, then one, and last two. > > Not all read errors are detected as bus collisions. Anyway, you got > loads of 'XX' as I expected, "moving" from run to run, which means that > your i2c bus is unreliable. > > My conclusion would be: bad hardware design generates noise on the i2c > bus, resulting in read errors. I don't doubt that. > > > > Did you have to enable any particular option in MBM? > > > > > > Nah, it just worked :) > > I asked Alex van Kaam (MBM's brilliant author) about his strategy with > bus collisions. To make it short, he explained he resets the bus on > problems. If you confirm that the smbus MBM detected was Via, Alex will > send me his code so that I can compare with ours, just in case. But I > doubt it'll change anything (our driver is working, it's just a matter > of how errors are - or aren't - handled). Yes, it's a Via Apollo Pro 133A. > Basically we don't handle read errors at all (because it is so rare). > Handling them correctly would make all chip drivers (and possibly i2c > core and bus drivers as well) more complex. I'm not sure it's worth it. > > That said, a similar problem was reported with W83L785TS-S chips (A7N8X > motherboards). I think that the cause is different (BIOS trying to > access the bus at the same time we do) but the consequences are the > same. I plan to add basic error handling to this specific chip driver > (don't know when I'll find the time to do so though). If it works and > could be extended to other drivers as well, we'll consider it. > > > > I guess this is unsolvable, but I just wanted to hear what you say. > > > Kinda weird it works so well with MBM, but that's ok. It's just for > > > fun I want it to work. > > I think it is work-around-able, but doubt it's worth it. Anyway, thanks > for reporting, as it increased our knowledge of the topic. Thanks for your help :) Best regards, Stian
