Re: [RFC PATCH 1/2] i2c: Add i2c_verify_device_id() to verify device id

[Peter Rosin]

On 2018-03-19 21:50, Guenter Roeck wrote:
> On Mon, Mar 19, 2018 at 08:55:55PM +0100, Peter Rosin wrote:
>> On 2018-03-19 19:48, Guenter Roeck wrote:
>>> On Mon, Mar 19, 2018 at 05:47:05PM +0100, Peter Rosin wrote:
>>>> I also wonder if NXP will ever release a chip with part-id 0 and
>>>> die-revision 0? If not, an all zero struct i2c_device_identity
>>>> could be used instead of manufacturer_id 0xffff and that would
>>>> simplify the pca954x driver code a bit more. But I guess we can
>>>> never know the answer to that question. And even if we did, the
>>>> answer might change later. But it would be nice...
>>>>
>>>
>>> That would be nice. You could ask at i2c.support@xxxxxxx, but I guess
>>> it would always be somewhat risky since the standard doesn't restrict
>>> its use, and some product manager at NXP might decide in the future
>>> that a device ID of 0x00 would be "cool".
>>
>> No need to bother NXP, PCA9848 has already claimed 0-0-0. Sigh.
>>
>> But while I googled that I found old datasheets for the chips PCA9672
>> through PCA9675 which use a different layout for the three bytes in
>> the device id. They have 8 manufacturer bits, 7 category bits, 6 bits
>> of feature indication and then 3 bits of revision. The top category
>> bits are zero so it is compatible for NXP chips. But since noone else
>> has implemented this, it is probably safe, but still a little bit
>> disturbing.
>>
> 
> The PCA9570 datasheet is especially interesting.
> 
> "9 bits with the part identification, assigned by manufacturer, the 7 MSBs with
> the category ID and the 6 LSBs with the feature ID (for example PCA9570 4-bit I/O
> expander)"
> 
> Maybe there is a magic compression scheme to squash 7 MSBs and 6 LSBs into
> a 9-bit field, or the category and feature IDs have quite some overlap,
> or the document would benefit from some proof-reading.

Hmm, I think I'll put my money on the compression thing.

>> I also found that NXP apparently uses the same part id (0x100) and die
>> revision (0) for PCA9570 and PCA9670. That seems odd.
>>
> ... especially since one has 4 channels and the other has 8 channels.
> It would be interesting to see if reality and datasheets match; this might
> as well be a curt-and-paste error. Of course, it might as well be that both
> chips use the same die and that some pins are just not exposed on the 4 channel
> version.
> 
>> Example old datasheet (2006):
>> https://www.digchip.com/datasheets/download_datasheet.php?id=1098812&part-number=PCA9672
>>
> On the other side this has been corrected in more recent datasheet versions,
> so I would not be too concerned about that.

PCA9671 is still using the old bit field layout in its most recent
datasheet (2011). I couldn't find anything newer anyway, and the
link seems canonical enough...

https://www.nxp.com/docs/en/data-sheet/PCA9671.pdf

Enough of this, but I'm going to send a note to NXP about the issues
we did find.

Cheers,
Peter