In my opinion all three numbers (udelay, mdelay{not used?], timeout)
should be
absolute times, not related to jiffies.
I guess it's supposed to be 10 ms units for timeout.
udelay is obviously the half-clock time (i2c clock freq = 1,000,000 / (2
* udelay))
Whatever the answer is, we should put it as a comment in i2c-algo-bit.h.
We do periodically get asked what the units for the 3 vaules are.
Most drivers have timeout = 100. That's 1 second. That's forever,
as you have found out if you ever probed a stuck bus.
As Dori Eldar points out (see my mail from a couple of weeks ago,
or TODO in i2c), i2c has no minimum frequency but smbus has
a min. frequency of 10 kHz. There's no way to enforce that in
i2c-algo-bit today. It would be nice to have some way to
tell i2c-algo-bit to enforce SMBus timing rules.
What mechanism could be used to set a timeout
of 100 microsec?
Albert Cranford wrote:
>
> Thanks for review. I'll pass your questions on to the
> group to clarify/answer.
>
> The current I2C CVS tree looks like:
> if (start+adap->timeout <= jiffies) {
> return -ETIMEDOUT;
> }
> but I think I was told not to submit the change since
> Linus's tree was correct.
> if (time_after_eq(jiffies, start+adap->timeout)) {
> return -ETIMEDOUT;
> }
> Albert
>
> Petr Vandrovec wrote:
> >
> > On 11 Jul 02 at 1:26, Albert Cranford wrote:
> > > Hello Linus,
> > > Could you please apply these 3 patches toward 2.5.26.
> > > They include Config.in updates, additions in i2c-id.h
> > > for "Video for Linux" and a compatibility fix for
> > > i2c-algo-bit.c
> >
> > Hi,
> > is timeout field in i2c_algo_bit_data supposed to be in jiffies
> > (like it is currently used) or in 10ms units? If it is supposed
> > to be in jiffies (other algos do not care about timeout field),
> > there is dozen of places (all callers of i2c_bit_add_bus) which
> > get it wrong.
> >
> > Next suspicious thing is that 'timeout' field from i2c_adapter
> > structure is not used at all - only i2c_control is willing to
> > set it, but nobody reads this field. Should not i2c_bit_add_bus
> > copy its timeout into this field, and then use adap->timeout instead
> > of bitadap->timeout in its sclhi() procedure?
> > Thanks,
> > Petr Vandrovec
> > vandrove at vc.cvut.cz
> >
>
> --
> Albert Cranford Deerfield Beach FL USA
> ac9410 at bellsouth.net
