[PATCH] mtd: spi-nor: Add support for BoHong bh25q128as

[Michael Walle]

Hi David,

Am 2021-05-10 13:04, schrieb David Bauer:
> On 5/10/21 12:56 PM, Michael Walle wrote:
>> Am 2021-05-10 12:27, schrieb David Bauer:
>>> On 5/10/21 11:35 AM, Michael Walle wrote:
>>>> Am 2021-05-10 11:28, schrieb David Bauer:
>>>>> On 5/10/21 10:00 AM, Michael Walle wrote
>>>>> 
>>>>> [...]
>>>>> 
>>>>>>> +static const struct flash_info bohong_parts[] = {
>>>>>>> +    /* BoHong Microelectronics */
>>>>>>> +    { "bh25q128as", INFO(0x684018, 0, 64 * 1024, 256,
>>>>>> 
>>>>>> I couldn't find "BoHong" in JEP106BC. 0x68 (without continuation 
>>>>>> codes)
>>>>>> is "Convex Computer". So this is wrong. OTOH I'm not sure, how 
>>>>>> many
>>>>>> SPI flashes "convex computer" have, if any ;) This company was 
>>>>>> brought
>>>>>> by HP in the end.
>>>>>> 
>>>>>> In any case, this patch depends on how we handle continuation 
>>>>>> codes or
>>>>>> if we can handle them at all. Or if this flash just lie about its
>>>>>> manufacturer id and don't and CC.
>>>>> 
>>>>> First of all, BoHong and Boya microelectronics seems to be the same
>>>>> company, as their datasheets seem to copy each other. There's not 
>>>>> much
>>>>> information about either of both, so I'd say that's a fair 
>>>>> assumption.
>>>>> 
>>>>> Regarding the continuation codes, Boya is listed in bank nine, 
>>>>> however
>>>>> in this case I should currently read an all 0x7f ID shouldn't I?
>>>> 
>>>> I'd guess so, yes.
>>>> 
>>>>> The datasheet also only specifies 3 bytes as a return value for
>>>>> register 0x9fh :(
>>>> 
>>>> Yeah. So, this flash falls into the same category "simply hijacks
>>>> a manuf id" as all the other flashes.
>>> 
>>> From a quick check, this is also be the case for GigaDevices and XMC.
>>> 
>>> My spontaneous idea would be to extend support for JEDEC IDs to read
>>> the up to 9 banks of the vendor ID and fix up the existing offenders.
>> 
>> you mean gigadevices and xmc? I'd presume they are also lacking the
>> continuation bytes.
> 
> Correct, same story with them.
> 
>> 
>>> To not break existing boards, we could either skip the continuation
>>> bytes of the kernel ID definitions for all flash chips or flag the
>>> already existing ones and only perform this on such flagged chips.
>>> 
>>> Personally, I'd say that only performing this on existing chips would
>>> be better, as new vendors with this violation scheme might probably
>>> appear and cause conflicts.
>>> 
>>> As we still lack auto detection for new chips with that, configuring
>>> the flash chip used with the chip name via DT would allow to set the
>>> exact chip used and also validate if the manufacturer / product after
>>> the continuation bits matches the one read from the chip.
>>> 
>>> What do you think?
>> 
>> If you'd ask me, unless there is a real world conflict, I'd just go
>> ahead and add them as is. If there is a conflict we'd need to find
>> a per device resolution for it.
> 
> Okay, I'll resend a v2 with the removed copyright then.

Could you also apply my SFDP patch [1] and send the dump (if there
is any)? Unfortunately, I can't think of a good way to do that along
with the patch and if this in some way regarded as copyrighted material.
So feel free to send it to me privately. I'm starting to build a
database.

>> There is another problem: shared device ids per vendor. Some (most?)
>> flash vendor share device ids on "similar" flashes, which we still
>> need tell apart in the kernel. So technically, this is the same 
>> problem
>> as with non-existing continuation bytes. Two different flashes sharing
>> the same id. For now, we look for differences in the SFDP.
>> 
>> Right now, the flash is probed first by its id and the the SFDP is
>> read and parsed. There are ideas to first read the SFDP. Having this
>> might come in handy here, too. Eg. we could fingerprint the flash
>> by its SFDP.
> 
> I was also thinking about that and we're also already bitten by 
> identical
> JEDEC IDs for different models. It's really a pity that there is no 
> real
> unique model identifier for us to use, which is not hacked to support
> legacy implementations like with this chip :(

Unfortunately, they aren't legacy and new chips still have this
behavior..

-michael

[1] 
https://lore.kernel.org/linux-mtd/20210503155651.30889-1-michael@walle.cc/