AW: (EXT) Re: [PATCH 2/2] mtd: spi-nor: micron-st: Add support for output-driver-strength

[Alexander Stein]

Hello Pratyush,

> On 08/10/21 01:18PM, Pratyush Yadav wrote:
> From: Pratyush Yadav <p.yadav at ti.com>
> > 
> > Micron flashes support this by the Bits [2:0] in the Enhanced Volatile
> > Configuration Register.
> > Checked datasheets:
> > - n25q_128mb_3v_65nm.pdf
> > - mt25t-qljs-L512-xBA-xxT.pdf
> 
> What does changing the impedance do? Does it improve latency? If we use 
> suboptimal impedance, will the flash still keep working correctly?
> 
> In other words, you need to justify why this patch is needed.

Hardware guys told me this will affect the signal qualitiy when the flash is
sending data. This depends among others on line length. If set incorrectly
voltage overshoots upon 0->1 or 1-> switch might occur. To answer to your
question if the flash will work with incorrect settings: It depends. It might
working the short term, but might fail in the long term.
Also this is completly unrelated to latency.

> > +static int micron_read_evcr(struct spi_nor *nor)
> > +{
> > +	int ret;
> > +
> > +	if (nor->spimem) {
> > +		struct spi_mem_op op =
> > +			SPI_MEM_OP(SPI_MEM_OP_CMD(SPINOR_OP_RD_EVCR, 1),
> > +				   SPI_MEM_OP_NO_ADDR,
> > +				   SPI_MEM_OP_NO_DUMMY,
> > +				   SPI_MEM_OP_DATA_IN(1, nor->bouncebuf, 1));
> 
> Are you always guaranteed to be in 1S-1S-1S mode during register write?
> 
> I would suggest calling spi_nor_spimem_setup_op() before this so that it 
> sets up all the buswidths correctly based on nor->reg_proto.

Thanks, I wasn't aware of that. I'll do that.

> > +
> > +		ret = spi_mem_exec_op(nor->spimem, &op);
> > +	} else {
> > +		ret = nor->controller_ops->read_reg(nor, SPINOR_OP_RD_EVCR, nor->bouncebuf,
> 1);
> 
> Split into 2 lines?

Will do.

> > +	}
> > +
> > +	if (ret < 0) {
> > +		dev_err(nor->dev, "error %d reading EVCR\n", ret);
> > +		return ret;
> > +	}
> > +
> > +	return nor->bouncebuf[0];
> > +}
> > +
> > +/*
> > + * Write Micron enhanced volatile configuration register
> > + * Return negative if error occurred or configuration register value
> > + */
> > +static int micron_write_evcr(struct spi_nor *nor, u8 evcr)
> > +{
> > +	nor->bouncebuf[0] = evcr;
> > +
> > +	spi_nor_write_enable(nor);
> 
> Check return code.

Will do.

> > +
> > +	if (nor->spimem) {
> > +		struct spi_mem_op op =
> > +			SPI_MEM_OP(SPI_MEM_OP_CMD(SPINOR_OP_WD_EVCR, 1),
> > +				   SPI_MEM_OP_NO_ADDR,
> > +				   SPI_MEM_OP_NO_DUMMY,
> > +				   SPI_MEM_OP_DATA_OUT(1, nor->bouncebuf, 1));
> 
> Same as above.

Will do.

> > +
> > +		return spi_mem_exec_op(nor->spimem, &op);
> > +	}
> > +
> > +	return nor->controller_ops->write_reg(nor, SPINOR_OP_WD_EVCR, nor->bouncebuf,
> 1);
> 
> Same as above. Split into 2 lines?

Will do.

> > +}
> > +
> > +/*
> > + * Supported values from Enahanced Volatile COnfiguration Register (Bits
> 2:0)
> > + */
> > +static const struct micron_drive_strength drive_strength_data[] = {
> > +	{ .ohms = 90, .val = 1 },
> > +	{ .ohms = 45, .val = 3 },
> > +	{ .ohms = 20, .val = 5 },
> > +	{ .ohms = 30, .val = 7 },
> > +};
> > +
> > +static struct micron_drive_strength const *micron_st_find_drive_strength_entry(u32
> ohms)
> > +{
> > +	int i;
> > +
> > +	for (i = 0; i < ARRAY_SIZE(drive_strength_data); i++) {
> > +		if (ohms == drive_strength_data[i].ohms)
> > +			return &drive_strength_data[i];
> > +	}
> > +	return NULL;
> > +}
> > +
> > +static void micron_st_post_sfdp(struct spi_nor *nor)
> > +{
> > +	struct device_node *np = spi_nor_get_flash_node(nor);
> > +	u32 ohms;
> > +
> > +	if (!np)
> > +		return;
> > +
> > +	if (!of_property_read_u32(np, "output-driver-strength", &ohms)) {
> > +		struct micron_drive_strength const *entry =
> > +			micron_st_find_drive_strength_entry(ohms);
> > +
> > +		if (entry) {
> > +			int evcrr = micron_read_evcr(nor);
> > +
> > +			if (evcrr >= 0) {
> 
> This is a bit confusing. Can you instead do:
> 
> if (evcrr < 0)
> 	return evcrr;
> 
> ...

Will do.

> > +				u8 evcr = (u8)(evcrr & 0xf8) | entry->val;
> 
> Don't use magic numbers. Define a bitmask, preferably using GENMASK().

Will do.

> > +
> > +				micron_write_evcr(nor, evcr);
> 
> Check return code. Do we need to abort flash probe if this fails, or can 
> we live with it, despite the suboptimal impedance?

Well, it's hard to say. It might work, see above. At least this should raise a warning
if setting the driver strength for some reason.

> > +				dev_dbg(nor->dev, "%s: EVCR 0x%x\n", __func__,
> > +					(u32)micron_read_evcr(nor));
> > +			}
> > +		} else {
> > +			dev_warn(nor->dev,
> > +				"Invalid output-driver-strength property specified: %u",
> > +				ohms);
> > +		}
> > +	}
> > +}
> > +
> >  static const struct spi_nor_fixups micron_st_fixups = {
> >  	.default_init = micron_st_default_init,
> > +	.post_sfdp = micron_st_post_sfdp,
> 
> NACK. Not every Micron flash has the EVCR register. For example, the 
> Micron MT35 flash family does not have an EVCR and the output drive 
> strength is programmed in a separate register. Set this only for the 
> flashes that need this.

Thanks for that information. I do not have the MT35 datasheet, so I wasn't aware of it.
I'll stick to MT25Q for now, which is the part number my datasheet is for.

Thanks for your feedback.

Best regards,
Alexander