[lm-sensors] latest findings - my older posting from ~ 11 days

[r.marek@sh.cvut.cz (Rudolf Marek)]

Hi all,

> Hi folks,
> dear Rudolf,
> I think I've solved the issue. As it is a very "heuristic" approach I can only 
> tell what I did and you can give me a commentary on it saying it makes no 
> sense this way or whatsoever ...
> 
> Nevertheless, I keep crash-looped now doing 
> 
> while true; do
> 	i2cdump -y 0 0x2d b > /dev/null
> done
> 
> for more than 105000 times (the best value I had had reached before was around 
> 20000, and only with debugging turned on) so I think it is fair to say things 
> have improved significantly at least.
> 
> I readily mentioned that I had had the impression that things were less 
> problematic with debugging turned on. Now, what does debugging do? It does 
> many printk's or printf's thereby making the process slower.
> 
> Rudolf pointed me to the file in charge: i2c-amd756.c. I simply added a single 
> msleep(1);

Good.

> 
> /usr/src/linux/drivers/i2c/busses/i2c-amd756.c (referring to linux-2.6.11.4):
> 
> static int amd756_transaction(struct i2c_adapter *adap)
> {
>         int temp;
>         int result = 0;
>         int timeout = 0;
> 
>         dev_dbg(&adap->dev, "Transaction (pre): GS=%04x, GE=%04x, ADD=%04x, "
>                 "DAT=%04x\n", inw_p(SMB_GLOBAL_STATUS),
>                 inw_p(SMB_GLOBAL_ENABLE), inw_p(SMB_HOST_ADDRESS),
>                 inb_p(SMB_HOST_DATA));  
> 
>         /* Make sure the SMBus host is ready to start transmitting */
>         /* but always wait one millisecond just in case ... */
>>>      msleep(1); <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<  Here!
>         if ((temp = inw_p(SMB_GLOBAL_STATUS)) & (GS_HST_STS | GS_SMB_STS)) {
> 
> and, until now, this seems to do the trick (knock on wood!!!). Tell me why 
> this cannot happen - as a matter of fact it seems to work. The explanation is 
> up to you, folks ;-).

I'm still waiting for Raphael to contact someone at AMD, that was the reason for 
the quiet in this case...

This is "sleep" is good idea, there must be something wrong with the controller, 
  as we already know ...

Explanation hmm :) hard to tell...

I checked the sequence of command and nothing really strange...

Have you some crash even with this delay?

Thanks
Regards
Rudolf