Re: [PATCH v3 3/3] mtd: spi-nor: otp: implement erase for Winbond and similar flashes

[Michael Walle]

Am 2021-05-20 19:51, schrieb Pratyush Yadav:
> On 20/05/21 05:58PM, Michael Walle wrote:
> > Winbond flashes with OTP support provide a command to erase the OTP
> > data. This might come in handy during development.
> >
> > This was tested with a Winbond W25Q32JW on a LS1028A SoC with the
> > NXP FSPI controller.
> >
> > Signed-off-by: Michael Walle <michael@xxxxxxxx>
> > ---
> [...]
> > diff --git a/drivers/mtd/spi-nor/otp.c b/drivers/mtd/spi-nor/otp.c
> > index ec0c1b33f7cc..2dc315b6bffc 100644
> > --- a/drivers/mtd/spi-nor/otp.c
> > +++ b/drivers/mtd/spi-nor/otp.c
> > @@ -111,6 +111,34 @@ int spi_nor_otp_write_secr(struct spi_nor *nor, 
> > loff_t addr, size_t len,
> >  	return ret ?: written;
> >  }
> >
> > +/**
> > + * spi_nor_otp_erase_secr() - erase a security register
> > + * @nor:        pointer to 'struct spi_nor'
> > + * @addr:       offset of the security register to be erased
> > + *
> > + * Erase a security register by using the SPINOR_OP_ESECR command. 
> > This method
> > + * is used on GigaDevice and Winbond flashes to erase OTP data.
> > + *
> > + * Return: 0 on success, -errno otherwise
> > + */
> > +int spi_nor_otp_erase_secr(struct spi_nor *nor, loff_t addr)
> > +{
> > +	u8 erase_opcode = nor->erase_opcode;
> > +	int ret;
> > +
> > +	ret = spi_nor_write_enable(nor);
> > +	if (ret)
> > +		return ret;
> > +
> > +	nor->erase_opcode = SPINOR_OP_ESECR;
> > +	ret = spi_nor_erase_sector(nor, addr);
> > +	nor->erase_opcode = erase_opcode;
> > +	if (ret)
> > +		return ret;
> > +
> > +	return spi_nor_wait_till_ready(nor);
>
> The datasheet for W25Q32JW says in Section 8.2.29:
>
>   The Security Register Lock Bits (LB3-1) in the Status Register-2 can
>   be used to OTP protect the security registers. Once a lock bit is set
>   to 1, the corresponding security register will be permanently locked,
>   Erase Security Register instruction to that register will be ignored
>
> So if the region is locked, the flash will happily accept the erase and
> simply do nothing. So will the program. So when the OTP region is 
> locked
> and someone does an erase-program cycle, they will think their data 
> went
> through even though it was simply thrown away by the flash.

Btw, this is also how it is handled for the "normal" data write 
protection.
If block protection bits are set, we happily accept erase and write
commands although the flash will discard the commands.

> I think you should check that bit before doing these operations to make
> sure it is actually allowed. If it isn't, return an error code (-EPERM
> maybe).

It's not that easy. You'd have to check the entire range, not just a 
part
of it. Otherwise you'll end up with half of the data committed to the 
flash
and the other half rejected with EPERM. So you'd have to do in
spi_nor_mtd_otp_erase() and spi_nor_mtd_otp_write().

I'm not opposed to it, but I tried to behave in the same way as the
other MTD chips which supports the OTP ops and I thought they would
ignore it, too. But I just had another look:
 - chips/cfi_cmdset_0001.c: seems to return -EROFS (partial writes are
   possible?). I couldn't find any datasheets where command 0xc0 is
   used to write OTP data. Thus I can only guess that status bit 1 is
   program failed or something like that [1]. All the datasheets I found
   will tell you its bit 0.
 - chips/cfi_cmdset_0002.c: seems to ignore errors
 - nand/onenand/onenand_base.c: I really don't know..

If no one opposes I'll add that check with EROFS.

-michael

[1] 
https://elixir.bootlin.com/linux/v5.13-rc2/source/drivers/mtd/chips/cfi_cmdset_0001.c#L1606