[PATCH v15 6/8] mtd: spi-nor: Retain nor->addr_width at 4BAIT parse

[Takahiro Kuwano]

On 7/22/2022 1:20 PM, Tudor.Ambarus at microchip.com wrote:
> On 7/22/22 07:00, Takahiro Kuwano wrote:
> 
> Good morning, Takahiro!
> 
Good morning, Tudor!

>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
>>
>> On 7/22/2022 1:06 AM, Tudor.Ambarus at microchip.com wrote:
>>> On 5/23/22 10:49, Michael Walle wrote:
>>>> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
>>>>
>>>> Am 2022-05-14 05:51, schrieb Takahiro Kuwano:
>>>>> On 5/13/2022 6:40 PM, Michael Walle wrote:
>>>>>> [btw the subject still has the old name of the addr_width]
>>>>>>
>>>>> Yes, it must be fixed in next rev.
>>>>>
>>>>>> Am 2022-05-13 03:26, schrieb Takahiro Kuwano:
>>>>>>> On 5/13/2022 7:14 AM, Michael Walle wrote:
>>>>>>>> Am 2022-05-10 00:10, schrieb tkuw584924 at gmail.com:
>>>>>>>>> From: Takahiro Kuwano <Takahiro.Kuwano at infineon.com>
>>>>>>>>>
>>>>>>>>> In 4BAIT parse, keep nor->params->addr_width because it may be used
>>>>>>>>> as
>>>>>>>>> current address mode in SMPT parse later on.
>>>>>>>>
>>>>>>>> Mh I'm not sure this is needed at all.
>>>>>>>>
>>>>>>>> SFDP spec says
>>>>>>>>   Variable address length (the current setting of the address
>>>>>>>>   length mode defines the address length)
>>>>>>>>
>>>>>>>> and
>>>>>>>>   When the length is defined as variable, the software or hardware
>>>>>>>>   controlling the memory is aware of the address length mode last
>>>>>>>>   set in the memory device and this same length of address.
>>>>>>>>
>>>>>>>> We don't set any address mode until all the SFDP parsing is
>>>>>>>> over. Therefore we should always be in 3 byte mode, no?
>>>>>>>>
>>>>>>> Actually there are some devices that have variable address length but
>>>>>>> 4 byte mode by default (I will work on those devices after this
>>>>>>> series
>>>>>>> is settled). To support such case, I prefer to use
>>>>>>> params->addr_nbytes
>>>>>>> as current address mode so that I can fix it in post_bfpt_fixup()
>>>>>>> hook.
>>>>>>
>>>>>> Are there public datasheets available? So these devices have a 3 byte
>>>>> I will send datasheets to you in another email. At this point, only
>>>>> summary datasheet is available in website.
>>>>>
>>>>>> and a 4 byte mode, but after reset, they are in the 4 byte mode? Looks
>>>>> Yes.
>>>>>
>>>>>> like it should be fixed in a different way. I'm not sure the "current
>>>>>> mode" handling is correct.
>>>>>>
>>>>> Yes, we may want to introduce a new flag like SPI_NOR_4BAM_DEFAULT and
>>>>> check
>>>>> the flag in BFPT parse. Once I send another series, please review.
>>>>>
>>>>>> We need to differentiate between the mode the flash currently is using
>>>>>> (nor->addr_nbytes) and the mode parsed by SFDP (params->addr_nbytes).
>>>>>>
>>>>> The flash's address mode affects the address length of Non-4B opcodes,
>>>>> including read/write any register ops used in SMPT parse and Infineon
>>>>> (spansion) specific hooks.
>>>>>
>>>>> The 4B opcodes always take address length of 4 regardless of flash's
>>>>> address mode. In these Infineon chips, 4B opcodes for read/program/
>>>>> erase are available and 4BAIT advertises them. We don't have to enter
>>>>> 4 byte address mode for read/program/erase.
>>>>
>>>> btw. this is a pity. you are using the stateless 4b opcodes but
>>>> then you don't provide stateless opcodes for the read any register
>>>> op :/
>>>>
>>>>> So, I think we need to differentiate between address length for
>>>>> read/program/erase and flash's default address mode.
>>>>
>>>> Or we keep them in sync. E.g. switch to 4bytes mode if we are
>>>> using the 4 byte. Granted, that sounds like a hack :)
>>>>
>>>>> Obviously we are using nor->addr_nbytes as address length for read/
>>>>> program/erase and should keep this usage.
>>>>
>>>> Yes, I wasn't aware that we actually two different runtime
>>>> parameters:
>>>>  - the read/program/erase address width, also used with the
>>>>    4b opcodes
>>>>  - internal mode 3b/4b. Up until now, this wasn't an issue
>>>>    because either the mode was switched or the 4b opcodes
>>>>    were used. So this was mutually exclusive. Now we have
>>>>    flashes which uses 4b opcodes _and_ we need the state
>>>>    of the internal mode.
>>>>
>>>> I can't think of a good solution for now. Need to think
>>>> more about this, but I'm pretty busy at the moment.
>>>> What I think is clear is that we need two different modes
>>>> here in the spi_nor struct. nor->addr_nbytes for the
>>>> read/program/erase opcodes and nor->address_mode or similar
>>>> which tracks the SPI flash's internal address mode.
>>>
>>> Hi, Takahiro,
>>>
>>> Can we determine the flash's internal address mode by querying
>>> the flash at run-time? Is this possible on Semper flashes?
>>>
>> CFR2V[7] has current address mode, but to read that, we need to
>> issue Read Any Register which address length relies on current
>> address mode. Chicken-and-egg...
>>
> 
> I see. What happens if we issue the Read Any Register command with
> the wrong address internal mode? Will I read just 0xff?
> For example try reading CFR2V[7] using addr_nbytes of value 3, and
> if it fails, to read it again but this time using addr_nbytes of
> value 4.
> 
It's undetermined. In case we issue Read Any Register with 3B address,
the host controller outputs 4 bytes (opcode + address) then inputs
1 byte. If the device is in 4B address mode at this time, the host
controller inputs the 1 byte while device is still waits for LSB of
address so IO is not driven by device.

Thanks,
Takahiro