Re: [PATCH 06/19] mtd: spi-nor: swp: Explain the MEMLOCK ioctl implementation behaviour

[Miquel Raynal <miquel.raynal@bootlin.com>]

On 18/11/2025 at 10:53:42 +01, "Michael Walle" <mwalle@kernel.org> wrote:

> On Fri Nov 14, 2025 at 6:53 PM CET, Miquel Raynal wrote:
>> Add comments about how these requests are actually handled in the SPI
>> NOR core. Their behaviour was not entirely clear to me at first, and
>> explaining them in plain English sounds the way to go.
>>
>> Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>
>> ---
>>  drivers/mtd/spi-nor/swp.c | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>>
>> diff --git a/drivers/mtd/spi-nor/swp.c b/drivers/mtd/spi-nor/swp.c
>> index 9bc5a356444665ad8824e9e12d679fd551b3e67d..ede03f26de3c65ff53c1cb888c2c43aea268b85a 100644
>> --- a/drivers/mtd/spi-nor/swp.c
>> +++ b/drivers/mtd/spi-nor/swp.c
>> @@ -341,6 +341,14 @@ static int spi_nor_sr_is_locked(struct spi_nor *nor, loff_t ofs, u64 len)
>>  	return spi_nor_is_locked_sr(nor, ofs, len, nor->bouncebuf[0]);
>>  }
>>  
>> +/*
>> + * These ioctls behave according to the following rules:
>> + * ->lock(): Never locks more than what is requested, ie. may lock less
>
> That behavior sounds so wrong... The user requests a region to be
> locked, and it isn't actually locked.

Agreed. I really got puzzled by that.

>> + * ->unlock(): Never unlocks more than what is requested, ie. may unlock less
>
> That seems somewhat sane.
>
> Maybe we should return -EINVAL if ofs or ofs+len aren't at sector
> boundaries. Yeah it's a change in the UAPI, but I'm not sure the
> current behavior is not harmful and misleading.

I would even go further and propose to return -EINVAL whenever the
request is not exactly doable. Being at a block boundary is not enough,
as there are many boundaries we cannot describe with just 4 protection
bits.

But this is somewhat a uAPI change indeed. So in the first place, I will
keep this comment. But if we feel like we should make the uAPI stricter,
it can come on top. Doing this would require a broad acknowledgement.

>> + * -is_locked(): Checks if the region is *fully* locked, returns false otherwise.
>> + *               This feeback may be misleading because users may get an "unlocked"
>> + *               status even though a subpart of the region is effectively locked.
>> + */
>>  static const struct spi_nor_locking_ops spi_nor_sr_locking_ops = {
>>  	.lock = spi_nor_sr_lock,
>>  	.unlock = spi_nor_sr_unlock,
>
> Anyway, as it is how it's currently behaving:
>
> Reviewed-by: Michael Walle <mwalle@kernel.org>

Thanks!
Miquèl