Re: [PATCH] mtd: spi-nor: Return error when nor->addr_width not match the device size

From: Tudor.Ambarus
Date: Thu Nov 15 2018 - 05:54:47 EST


Hi, Liu, Boris, Cyrille,

On 11/14/2018 03:51 PM, Boris Brezillon wrote:
> On Wed, 14 Nov 2018 20:56:05 +0800
> Liu Xiang <liu.xiang6@xxxxxxxxxx> wrote:
>
>> In is25lp256, the DWORD1 of JEDEC Basic Flash Parameter Header
>> is 0xfff920e5. So the DWORD1[18:17] Address Bytes bits are 0b00,

Liu, can you point us to a datasheet that has the JEDEC BFPT tables described? I
couldn't find one ...

>> means that 3-Byte only addressing.
>
> According to your other patch this NOR supports 4B opcode, which means
> the SFDP table is wrong.
>
>> But the device size is larger
>> than 16MB, nor->addr_width must be 4 to access the whole address.
>> An error should be returned when nor->addr_width not match
>
> ^does not
>
>> the device size in spi_nor_parse_sfdp().
>>
>> Suggested-by: Boris Brezillon <boris.brezillon@xxxxxxxxxxx>
>> Signed-off-by: Liu Xiang <liu.xiang6@xxxxxxxxxx>
>> ---
>> drivers/mtd/spi-nor/spi-nor.c | 4 ++++
>> 1 file changed, 4 insertions(+)
>>
>> diff --git a/drivers/mtd/spi-nor/spi-nor.c b/drivers/mtd/spi-nor/spi-nor.c
>> index 3eba13a..77eaf22 100644
>> --- a/drivers/mtd/spi-nor/spi-nor.c
>> +++ b/drivers/mtd/spi-nor/spi-nor.c
>> @@ -2669,6 +2669,10 @@ static int spi_nor_parse_bfpt(struct spi_nor *nor,
>> }
>> params->size >>= 3; /* Convert to bytes. */
>>
>> + /*if the device exceeds 16MiB, addr_width must be 4*/
>
> Please add a white space after '/*' and before '*/':
>
> /* If the device exceeds 16MiB, ->addr_width must be 4. */
>
>> + if ((params->size > 0x1000000) && (nor->addr_width == 3))
>
> Parens are not needed around sub-conditions:
>
> if (params->size > 0x1000000 && nor->addr_width == 3)
>
>> + return -EINVAL;
>> +
>
> I'm not sure this is correct. Looks like some NORs only support 3B
> opcodes but have a "4-byte addressing" mode (see set_4byte() [1]).
> Don't know what's reported by the BFPT section in this case though
> (BFPT_DWORD1_ADDRESS_BYTES_3_ONLY or BFPT_DWORD1_ADDRESS_BYTES_3_OR_4).

Boris, this is in close relation with your second patch: [PATCH v3 2/2] mtd:
spi-nor: Use 4B opcodes when the NOR advertises both 3B and 4B.

When looking again at this, I would say that for the flashes that have a "4-byte
addressing" mode, but just 3B opcodes, I would expect the DWORD1[18:17] to be of
value BFPT_DWORD1_ADDRESS_BYTES_3_OR_4 (enters 4-Byte mode on command - uses 3B
opcodes).

If BFPT_DWORD1_ADDRESS_BYTES_3_OR_4 and 4B opcodes, then we can query BFPT
DWORD16[31:24]: it should have value xx1x_xxxxb to indicate that 4B opcodes are
supported. But which 4B opcodes are supported? Do all 3B opcodes have a 4B
opcode correspondent if SFDP 4-byte table is not available? This might be a good
assumption, but I can't see it anywhere in jesd216c.

Cyrille, when looking at ba3ae6a1d4c ("mtd: spi-nor: add a stateless method to
support memory size above 128Mib") I see that all flashes that declare
SPI_NOR_4B_OPCODES will have the 3B opcodes converted to 4B opcodes, assuming
that all 3B opcodes have a 4B correspondent. Is this assumption a general rule,
or it's just for the Spansion flashes?

Cheers,
ta

>
> Anyway, I think this check should be moved here [2] to cover the
> non-SFDP case.
>
>> /* Fast Read settings. */
>> for (i = 0; i < ARRAY_SIZE(sfdp_bfpt_reads); i++) {
>> const struct sfdp_bfpt_read *rd = &sfdp_bfpt_reads[i];
>
> [1]https://elixir.bootlin.com/linux/v4.20-rc2/source/drivers/mtd/spi-nor/spi-nor.c#L278
> [2]https://elixir.bootlin.com/linux/v4.20-rc2/source/drivers/mtd/spi-nor/spi-nor.c#L3758
>