Hi Marek,
Le 24/08/2015 12:48, Marek Vasut a écrit :
> On Monday, August 24, 2015 at 12:13:58 PM, Cyrille Pitchen wrote:
>> The number of dummy cycles used during Fast Read commands can be reduced
>> to improve transfer performances. Each manufacturer has a dedicated set of
>> registers to provide the memory with the exact number of dummy cycles it
>> should expect. Both the memory and the (Q)SPI controller must agree on
>> this number of dummy cycles.
>>
>> The number of dummy cycles can be found into the memory datasheet and
>> mostly depends on the SPI clock frequency, the Fast Read op code and the
>> Single/Dual Data Rate mode.
>>
>> Probing JEDEC Serial Flash Discoverable Parameters (SFDP) tables would
>> only provide the driver with a high enough number of dummy cycles for each
>> Fast Read command to be used for all clock frequencies: this solution
>> would not be optimized.
>>
>> Signed-off-by: Cyrille Pitchen <cyrille.pitchen@xxxxxxxxx>
>
> Hi!
>
>> drivers/mtd/spi-nor/spi-nor.c | 97
>> ++++++++++++++++++++++++++++++++++--------- include/linux/mtd/spi-nor.h
>> | 2 +
>> 2 files changed, 80 insertions(+), 19 deletions(-)
>>
>> diff --git a/drivers/mtd/spi-nor/spi-nor.c b/drivers/mtd/spi-nor/spi-nor.c
>> index e2a6029dc056..869e098a6841 100644
>> --- a/drivers/mtd/spi-nor/spi-nor.c
>> +++ b/drivers/mtd/spi-nor/spi-nor.c
>> @@ -119,24 +119,6 @@ static int read_cr(struct spi_nor *nor)
>> }
>>
>> /*
>> - * Dummy Cycle calculation for different type of read.
>> - * It can be used to support more commands with
>> - * different dummy cycle requirements.
>> - */
>> -static inline int spi_nor_read_dummy_cycles(struct spi_nor *nor)
>> -{
>> - switch (nor->flash_read) {
>> - case SPI_NOR_FAST:
>> - case SPI_NOR_DUAL:
>> - case SPI_NOR_QUAD:
>> - return 8;
>> - case SPI_NOR_NORMAL:
>> - return 0;
>> - }
>> - return 0;
>> -}
>
> You can probably just soup up this function so that it sets the
> nor->read_dummy, no ?
>
Actually, this is what the patch does: spi_nor_read_dummy_cycles() was reused
and enhanced few lines below where you've pointed out the
"switch (nor->flash_read)" block should be move after the else block.
I think when I wrote the code I've chosen to move the definition of this
function instead of adding forward declarations of functions such as read_cr()
or write_sr_cr(), which are now called by micron_set_dummy_cycles().
>> -/*
>> * Write status register 1 byte
>> * Returns negative if error occurred.
>> */
>> @@ -1012,6 +994,81 @@ static int set_quad_mode(struct spi_nor *nor, struct
>> flash_info *info) }
>> }
>>
>> +static int micron_set_dummy_cycles(struct spi_nor *nor)
>> +{
>> + int ret;
>> + u8 val, mask;
>> +
>> + /* read the Volatile Configuration Register (VCR) */
>
> NIT: If this is a sentence, start it with capital letter and end it with
> fullstop :)
>
done for the next version
>> + ret = nor->read_reg(nor, SPINOR_OP_RD_VCR, &val, 1);
>> + if (ret < 0) {
>> + dev_err(nor->dev, "error %d reading VCR\n", ret);
>> + return ret;
>> + }
>> +
>> + write_enable(nor);
>> +
>> + /* update the number of dummy into the VCR */
>
> DTTO
>
done for the next version
>> + mask = GENMASK(7, 4);
>> + val &= ~mask;
>> + val |= (nor->read_dummy << 4) & mask;
>> + ret = nor->write_reg(nor, SPINOR_OP_WR_VCR, &val, 1, 0);
>> + if (ret < 0) {
>> + dev_err(nor->dev, "error while writing VCR register\n");
>> + return ret;
>> + }
>> +
>> + ret = spi_nor_wait_till_ready(nor);
>> + if (ret)
>> + return ret;
>> +
>> + return 0;
>> +}
>> +
>> +/*
>> + * Dummy Cycle calculation for different type of read.
>> + * It can be used to support more commands with
>> + * different dummy cycle requirements.
>> + */
>> +static int spi_nor_read_dummy_cycles(struct spi_nor *nor,
>> + const struct flash_info *info)
>> +{
>> + struct device_node *np = nor->dev->of_node;
>> + u32 num_dummy_cycles;
>> +
>> + if (np && !of_property_read_u32(np, "m25p,num-dummy-cycles",
>> + &num_dummy_cycles)) {
>> + nor->read_dummy = num_dummy_cycles;
>> +
>> + /*
>> + * This switch block might be moved after the if...then...else
>> + * statement but it was not tested with all Spansion or Micron
>> + * memories.
>> + * Now the "m25p,num-dummy-cycles" property needs to be
>> + * explicitly set in the device tree so the switch statement is
>> + * executed. This should avoid unwanted side effects and keep
>> + * backward compatibility.
>> + */
>> + switch (JEDEC_MFR(info)) {
>> + case CFI_MFR_ST:
>> + return micron_set_dummy_cycles(nor);
>> + default:
>
> If you do have m25p,num-dummy-cycles set for non-micron flash, you have a
> problem here I believe.
>
>> + break;
>> + }
>> + } else {
>
> The solution would be to drop this else {} bit here, so that if you fail in
> the DT-based configuration, you fall back to this old behavior. What do you
> think please ? :)
>
Good idea!
I also add a trace for the default case of "switch (JEDEC_MFR(info))":
dev_warn(dev, "can't set the number of dummy cycles\n");
So the user is notified that the driver could not use the value of
"m25p,num-dummy-cycles" from the DT before falling back to the legacy
code.
>> + switch (nor->flash_read) {
>> + case SPI_NOR_FAST:
>> + case SPI_NOR_DUAL:
>> + case SPI_NOR_QUAD:
>> + nor->read_dummy = 8;
>> + case SPI_NOR_NORMAL:
>> + nor->read_dummy = 0;
>> + }
>> + }
>> +
>> + return 0;
>> +}
>
> [...]
>
thanks for the review!
Best regards,
Cyrille
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
