Adding OTP-only device to MTD or CHAR subsystem?

[Srinivas Kandagatla]

Hi Scott,

Sorry for long delay in replying.. :-)
On 29/12/15 00:32, Scott Branden wrote:
> +Srinivas/Maxime
>
> On 15-12-28 03:38 PM, Arnd Bergmann wrote:
>> On Monday 28 December 2015 15:21:08 Scott Branden wrote:
>>> Greg/Brian/Arnd,
>>>
>>> We have OTP device drivers for accessing OTP memory in our SoCs.
>>>
>>> I looking for the right place and model to place such OTP device
>>> drivers.
>>>
>>> 1) Should we follow the bfin-otp model in drivers/char?  This doesn't
>>> seem like the right place to put it although following the bfin example
>>> is quite simple to implement.  We actually had a custom set of Ioctl's
>>> that I changed to use the standard file access model used by the bfin
>>> driver.  But a custom util is still needed to issue an OTPLOCK command.
>>>    I'm guess mtd-utils has such abilities (or should).
>>>
>>> 2) Instead, should we start adding OTP-only drivers into the MTD
>>> subsystem?  Onenand and CFI based MTD devices already have OTP
>>> programmable regions.  If we created a new OTP device type in the MTD
>>> subsystem this looks like a good thing to do.  mtd-utils could/should be
>>> used to access the OTP device then along with standard fileio
>>> operations.
>>>
>>> 3) Or some other suggestion of where to place OTP device drivers?
>>
>> I think drivers/nvmem is now the right place for this.
> Thanks for the pointer Arnd.  Too bad an extension wasn't added in MTD
> but at least there is some sort of a model in place now.  The mtd-abi.h
> would have been a good thing to use for OTPLOCK as well as the remainder
> of the functionality in MTD OTP.  The mapping of nvmem to properties to
> user settings seems like a useful feature though.
>
> Hi Srinivas/Maxime,
> The nvmem drivers don't seem to support row unlocking of OTP.  Having
> raw write access to the OTP is not a desirable feature.  Are there plans
> in place to add any additional functionality to the nvm driver model
> right now?
NVMEM was started to be simple interface based on regmap, As of today we 
did not have any users which wanted lock feature, but Am open for more 
discussion on this if we want to get this feature via nvmem.

Any ideas and patches are welcome.

Few ideas from my side though,
 From Kernel side: introduce nvmem_lock()/nvmem_unlock() apis which 
would set the ranges as required, and the providers can use regmap 
callbacks writeable_reg()/readable_reg() in regmap configs to enforce 
the lock range

 From userspace side, As of today nvmem only provides an binary file 
interface, which we could probably extend to device based and provide 
userspace abi.

Lets continue discussion !!

--srini



>>
>>     Arnd
>>
> Regards,
>   Scott