Re: [PATCH v2 2/2] mtd: spi-nor: sst: Add support for Global Unlock on sst26vf

[Tudor.Ambarus]

On 1/20/21 5:49 PM, Michael Walle wrote:
> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> 
> Am 2021-01-20 16:39, schrieb Tudor.Ambarus@xxxxxxxxxxxxx:
>> On 1/20/21 5:02 PM, Michael Walle wrote:
>>> EXTERNAL EMAIL: Do not click links or open attachments unless you know
>>> the content is safe
>>>
>>> Am 2021-01-20 15:52, schrieb Tudor.Ambarus@xxxxxxxxxxxxx:
>>>> On 1/20/21 4:05 PM, Michael Walle wrote:
>>>>>> diff --git a/drivers/mtd/spi-nor/sst.c b/drivers/mtd/spi-nor/sst.c
>>>>>> index 00e48da0744a..d6e1396abb96 100644
>>>>>> --- a/drivers/mtd/spi-nor/sst.c
>>>>>> +++ b/drivers/mtd/spi-nor/sst.c
>>>>>> @@ -8,6 +8,39 @@
>>>>>>
>>>>>>  #include "core.h"
>>>>>>
>>>>>> +static int sst26vf_lock(struct spi_nor *nor, loff_t ofs, uint64_t
>>>>>> len)
>>>>>> +{
>>>>>> +     return -EOPNOTSUPP;
>>>>>> +}
>>>>>> +
>>>>>> +static int sst26vf_unlock(struct spi_nor *nor, loff_t ofs,
>>>>>> uint64_t
>>>>>> len)
>>>>>> +{
>>>>>> +     if (ofs == 0 && len == nor->params->size)
>>>>>> +             return spi_nor_global_block_unlock(nor);
>>>>>
>>>>>
>>>>> Some blocks might not be unlocked because they are permanently
>>>>> locked. Does it make sense to read BPNV of the control register
>>>>> and add a debug message here?
>>>>
>>>> It would, yes. If any block is permanently locked in the unlock_all
>>>> case,
>>>> I'll just print a dbg message and return -EINVAL. Sounds good?
>>>
>>> spi_nor_sr_unlock(), atmel_at25fs_unlock() and
>>> atmel_global_unprotect()
>>> will return -EIO in case the SR wasn't writable.
>>
>> You mean in the spi_nor_write_sr_and_check() calls. -EIO is fine
>> there if what we wrote is different than what we read back, it would
>> indicate an IO error.
>>
>> GBULK command clears all the write-protection bits in the Block
>> Protection register, except for those bits that have been permanently
>> locked down. So even if we have few blocks permanently locked, i.e.
>> CR.BPNV == 1, the GBULK can clear the protection for the remaining
>> blocks. So not really an IO error, but rather an -EINVAL, because
>> the user asks to unlock more than we can.
> 
> Doesn't EINVAL indicate wrong parameters, but does nothing? In this
> case, unlock would be partially successful.
> 
yes, that's what I said I'll do: "If any block is permanently locked
in the unlock_all case, I'll just print a dbg message and return -EINVAL",
without sending a GBULK cmd. Caller wrongly asks to unlock all, when we
can just unlock partial memory. 

It's similar to what is at:
https://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux.git/tree/drivers/mtd/spi-nor/core.c?h=spi-nor/next#n1946

> In any case, my point was that depending on the underlying locking
> ops, either -EIO or -EINVAL is returned if spi_nor_unlock() fails
> for the same reason, that is unlock() wasn't possible because of
> some sort of hardware write protection. And IMHO it should return
> the same errno (whatever the correct errno is in this case).
> 

But the reasons are different: 1/caller wrongly asks to unlock
more than we can, thus -EINVAL 2/ -EIO when we don't read what
we expect to read.

Cheers,
ta