Re: [RFC PATCH 2/3] mtd: spi-nor: sfdp: fix spi_nor_read_sfdp()

[Michael Walle <michael@walle.cc>]

Am 2021-03-16 12:04, schrieb Pratyush Yadav:
> On 12/03/21 08:05PM, Michael Walle wrote:
>> If spi_nor_read_sfdp() is used after probe, we have to set read_proto
>> and the read dirmap.
>> 
>> Signed-off-by: Michael Walle <michael@walle.cc>
>> ---
>>  drivers/mtd/spi-nor/sfdp.c | 8 ++++++++
>>  1 file changed, 8 insertions(+)
>> 
>> diff --git a/drivers/mtd/spi-nor/sfdp.c b/drivers/mtd/spi-nor/sfdp.c
>> index b1814afefc33..47634ec9b899 100644
>> --- a/drivers/mtd/spi-nor/sfdp.c
>> +++ b/drivers/mtd/spi-nor/sfdp.c
>> @@ -179,19 +179,27 @@ static int spi_nor_read_sfdp(struct spi_nor 
>> *nor, u32 addr,
>>  			     size_t len, void *buf)
>>  {
>>  	u8 addr_width, read_opcode, read_dummy;
>> +	struct spi_mem_dirmap_desc *rdesc;
>> +	enum spi_nor_protocol read_proto;
>>  	int ret;
>> 
>>  	read_opcode = nor->read_opcode;
>> +	read_proto = nor->read_proto;
>> +	rdesc = nor->dirmap.rdesc;
>>  	addr_width = nor->addr_width;
>>  	read_dummy = nor->read_dummy;
>> 
>>  	nor->read_opcode = SPINOR_OP_RDSFDP;
>> +	nor->read_proto = SNOR_PROTO_1_1_1;
>> +	nor->dirmap.rdesc = NULL;
>>  	nor->addr_width = 3;
>>  	nor->read_dummy = 8;
> 
> NACK. You can't assume the device is still in 1S-1S-1S mode after 
> probe.
> For example, the s28hs512t flash is switched to 8D-8D-8D mode by the
> time the probe finishes so this would be an invalid command. Same for
> any flash that goes into a stateful mode.

I see.

> And you can't even keep using nor->read_proto to read SFDP because the
> Read SFDP command might not be supported in all modes. xSPI spec
> (JESD251) says that the Read SFDP command is optional in 8D-8D-8D mode.
> 
> I think the best approach for this would be to cache the entire SFDP
> table at parse time. This obviously comes with a memory overhead but I
> don't think it would be too big. For example, the sfdp table on
> s28hs512t is 491 bytes and it has 6 tables. Anyway, if the memory usage
> is too much of a problem we can put the feature behind a config.

I don't like to have it a config option, because then, if you really
need it, i.e. some user has an unknown flash, it might not be there.

The next question would be, should I leave the current parsing code
as is or should I also change that to use the sftp data cache. I'd
prefer to leave it as is.

-michael

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/