Hi Boris,
2016-01-14 10:19 GMT+01:00 Boris Brezillon <boris.brezillon@xxxxxxxxxxxxxxxxxx>:
> On Wed, 13 Jan 2016 17:34:15 +0100
> Romain Izard <romain.izard.pro@xxxxxxxxx> wrote:
>
>> Starting with the SAMA5D2, there is a new revision of the Atmel PMECC
>> controller that can correct 32 bits in each sector. This controller is
>> not 100% compatible with the previous revision that corrected a maximum
>> of 24 bits by sector, as some register addresses overlap.
>>
>> Using information from the device tree, we can configure the driver to
>> work with both versions.
>>
>> Signed-off-by: Romain Izard <romain.izard.pro@xxxxxxxxx>
>> ---
>> .../devicetree/bindings/mtd/atmel-nand.txt | 7 +++++--
>> drivers/mtd/nand/atmel_nand.c | 23 +++++++++++++++++++++-
>> drivers/mtd/nand/atmel_nand_ecc.h | 8 ++++++--
>> 3 files changed, 33 insertions(+), 5 deletions(-)
>>
>> diff --git a/Documentation/devicetree/bindings/mtd/atmel-nand.txt b/Documentation/devicetree/bindings/mtd/atmel-nand.txt
>> index 89b0db9801b0..90887b430f03 100644
>> --- a/Documentation/devicetree/bindings/mtd/atmel-nand.txt
>> +++ b/Documentation/devicetree/bindings/mtd/atmel-nand.txt
>> @@ -1,7 +1,10 @@
>> Atmel NAND flash
>>
>> Required properties:
>> -- compatible : should be "atmel,at91rm9200-nand" or "atmel,sama5d4-nand".
>> +- compatible: The possible values are:
>> + "atmel,at91rm9200-nand"
>> + "atmel,sama5d2-nand"
>> + "atmel,sama5d4-nand"
>> - reg : should specify localbus address and size used for the chip,
>> and hardware ECC controller if available.
>> If the hardware ECC is PMECC, it should contain address and size for
>> @@ -22,7 +25,7 @@ Optional properties:
>> Supported values are: "none", "soft", "hw", "hw_syndrome", "hw_oob_first",
>> "soft_bch".
>> - atmel,has-pmecc : boolean to enable Programmable Multibit ECC hardware.
>> - Only supported by at91sam9x5 or later sam9 product.
>> + Supported by AT91SAM9x5 or later SAM9 chips, and SAMA5 chips.
>> - atmel,pmecc-cap : error correct capability for Programmable Multibit ECC
>> Controller. Supported values are: 2, 4, 8, 12, 24.
>> - atmel,pmecc-sector-size : sector size for ECC computation. Supported values
>> diff --git a/drivers/mtd/nand/atmel_nand.c b/drivers/mtd/nand/atmel_nand.c
>> index e5d7e7e63f49..6fe50e2d291f 100644
>> --- a/drivers/mtd/nand/atmel_nand.c
>> +++ b/drivers/mtd/nand/atmel_nand.c
>> @@ -65,6 +65,7 @@ module_param(on_flash_bbt, int, 0);
>>
>> struct atmel_nand_caps {
>> bool pmecc_correct_erase_page;
>> + uint8_t pmecc_max_correction;
>> };
>>
>> struct atmel_nand_nfc_priv {
>> @@ -146,6 +147,7 @@ struct atmel_nand_host {
>> int pmecc_cw_len; /* Length of codeword */
>>
>> void __iomem *pmerrloc_base;
>> + void __iomem *pmerrloc_el_base;
>> void __iomem *pmecc_rom_base;
>>
>> /* lookup table for alpha_to and index_of */
>> @@ -817,7 +819,7 @@ static void pmecc_correct_data(struct mtd_info *mtd, uint8_t *buf, uint8_t *ecc,
>> sector_size = host->pmecc_sector_size;
>>
>> while (err_nbr) {
>> - tmp = pmerrloc_readl_el_relaxed(host->pmerrloc_base, i) - 1;
>> + tmp = pmerrloc_readl_el_relaxed(host->pmerrloc_el_base, i) - 1;
>> byte_pos = tmp / 8;
>> bit_pos = tmp % 8;
>>
>> @@ -1208,6 +1210,8 @@ static int atmel_pmecc_nand_init_params(struct platform_device *pdev,
>> err_no = PTR_ERR(host->pmerrloc_base);
>> goto err;
>> }
>
> How about putting the comment you added below here:
>
> /*
> * The ATMEL_PMERRLOC_ELx register location depends on the number of
> * bits corrected by the PMECC controller. Pre-calculate the
> * pmerrloc_el_base according to the PMECC capabilities.
> */
>
Here, this comment is paraphrasing the code. That's not very useful.
>> + host->pmerrloc_el_base = host->pmerrloc_base + ATMEL_PMERRLOC_SIGMAx +
>> + (host->caps->pmecc_max_correction + 1) * 4;
>>
>> if (!host->has_no_lookup_table) {
>> regs_rom = platform_get_resource(pdev, IORESOURCE_MEM, 3);
>> @@ -2307,17 +2311,34 @@ static int atmel_nand_remove(struct platform_device *pdev)
>> return 0;
>> }
>>
>> +/*
>> + * AT91RM9200 does not have PMECC or PMECC Errloc peripherals for
>> + * BCH ECC. Combined with the "atmel,has-pmecc", it is used to describe
>> + * devices from the SAM9 family that have those.
>> + */
>> static struct atmel_nand_caps at91rm9200_caps = {
>> .pmecc_correct_erase_page = false,
>> + .pmecc_max_correction = 24,
>> };
>>
>> static struct atmel_nand_caps sama5d4_caps = {
>> .pmecc_correct_erase_page = true,
>> + .pmecc_max_correction = 24,
>> +};
>> +
>> +/*
>> + * The PMECC Errloc controller starting in SAMA5D2 is not compatible,
>> + * as the increased correction strength requires more registers.
>> + */
>> +static struct atmel_nand_caps sama5d2_caps = {
>> + .pmecc_correct_erase_page = true,
>> + .pmecc_max_correction = 32,
>> };
>>
>> static const struct of_device_id atmel_nand_dt_ids[] = {
>> { .compatible = "atmel,at91rm9200-nand", .data = &at91rm9200_caps },
>> { .compatible = "atmel,sama5d4-nand", .data = &sama5d4_caps },
>> + { .compatible = "atmel,sama5d2-nand", .data = &sama5d2_caps },
>> { /* sentinel */ }
>> };
>>
>> diff --git a/drivers/mtd/nand/atmel_nand_ecc.h b/drivers/mtd/nand/atmel_nand_ecc.h
>> index 668e7358f19b..ec964c43c932 100644
>> --- a/drivers/mtd/nand/atmel_nand_ecc.h
>> +++ b/drivers/mtd/nand/atmel_nand_ecc.h
>> @@ -108,7 +108,11 @@
>> #define PMERRLOC_ERR_NUM_MASK (0x1f << 8)
>> #define PMERRLOC_CALC_DONE (1 << 0)
>> #define ATMEL_PMERRLOC_SIGMAx 0x028 /* Error location SIGMA x */
>> -#define ATMEL_PMERRLOC_ELx 0x08c /* Error location x */
>> +
>> +/*
>> + * The ATMEL_PMERRLOC_ELx register location depends from the number of
>> + * bits corrected by the PMECC controller. Do not use it.
>> + */
>>
>> /* Register access macros for PMECC */
>> #define pmecc_readl_relaxed(addr, reg) \
>> @@ -136,7 +140,7 @@
>> readl_relaxed((addr) + ATMEL_PMERRLOC_SIGMAx + ((n) * 4))
>>
>> #define pmerrloc_readl_el_relaxed(addr, n) \
>> - readl_relaxed((addr) + ATMEL_PMERRLOC_ELx + ((n) * 4))
>> + readl_relaxed((addr) + ((n) * 4))
>
> Another option would have been to pass the maximum ECC strength here,
> and modify the ATMEL_PMERRLOC_ELx macro to take a max_strength argument
> and return the adjusted ELx offset.
> Anyway, both solutions work for me.
>
The current solution makes it easier to keep within the 80 character limit.
The memory/computing tradeoff is so small it won't change anything.
>>
>> /* Galois field dimension */
>> #define PMECC_GF_DIMENSION_13 13
Best regards,
--
Romain Izard
--
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
