Re: [v3 1/2] mtd: nand: Add Cadence NAND controller driver

[Dmitry Osipenko]

14.06.2019 18:09, Piotr Sroka ÐÐÑÐÑ:

Commit description is mandatory.

> Signed-off-by: Piotr Sroka <piotrs@xxxxxxxxxxx>
> ---

[snip]

> +
> +/* Cadnence NAND flash controller capabilities get from driver data. */
> +struct cadence_nand_dt_devdata {
> +	/* Skew value of the output signals of the NAND Flash interface. */
> +	u32 if_skew;
> +	/* It informs if aging feature in the DLL PHY supported. */
> +	u8 phy_dll_aging;
> +	/*
> +	 * It informs if per bit deskew for read and write path in
> +	 * the PHY is supported.
> +	 */
> +	u8 phy_per_bit_deskew;
> +	/* It informs if slave DMA interface is connected to DMA engine. */
> +	u8 has_dma;

There is no needed to dedicate 8 bits to a variable if you only care about a single
bit. You may write this as:

bool has_dma : 1;

[snip]

> +static struct
> +cdns_nand_chip *to_cdns_nand_chip(struct nand_chip *chip)
> +{
> +	return container_of(chip, struct cdns_nand_chip, chip);
> +}
> +
> +static struct
> +cdns_nand_ctrl *to_cdns_nand_ctrl(struct nand_controller *controller)
> +{
> +	return container_of(controller, struct cdns_nand_ctrl, controller);
> +}

It's better to inline explicitly such cases because they won't get inlined with some
kernel configurations, like enabled ftracing for example.

> +static bool
> +cadence_nand_dma_buf_ok(struct cdns_nand_ctrl *cdns_ctrl, const void *buf,
> +			u32 buf_len)
> +{
> +	u8 data_dma_width = cdns_ctrl->caps2.data_dma_width;
> +
> +	return buf && virt_addr_valid(buf) &&
> +		likely(IS_ALIGNED((uintptr_t)buf, data_dma_width)) &&
> +		likely(IS_ALIGNED(buf_len, data_dma_width));
> +}
> +
> +static int cadence_nand_wait_for_value(struct cdns_nand_ctrl *cdns_ctrl,
> +				       u32 reg_offset, u32 timeout_us,
> +				       u32 mask, bool is_clear)
> +{
> +	u32 val;
> +	int ret = 0;
> +
> +	ret = readl_poll_timeout(cdns_ctrl->reg + reg_offset,
> +				 val, !(val & mask) == is_clear,
> +				 10, timeout_us);

Apparently you don't care about having memory barrier here, hence
readl_relaxed_poll_timeout().

> +	if (ret < 0) {
> +		dev_err(cdns_ctrl->dev,
> +			"Timeout while waiting for reg %x with mask %x is clear %d\n",
> +			reg_offset, mask, is_clear);
> +	}
> +
> +	return ret;
> +}
> +
> +static int cadence_nand_set_ecc_enable(struct cdns_nand_ctrl *cdns_ctrl,
> +				       bool enable)
> +{
> +	u32 reg;
> +
> +	if (cadence_nand_wait_for_value(cdns_ctrl, CTRL_STATUS,
> +					1000000,
> +					CTRL_STATUS_CTRL_BUSY, true))
> +		return -ETIMEDOUT;
> +
> +	reg = readl(cdns_ctrl->reg + ECC_CONFIG_0);
> +
> +	if (enable)
> +		reg |= ECC_CONFIG_0_ECC_EN;
> +	else
> +		reg &= ~ECC_CONFIG_0_ECC_EN;
> +
> +	writel(reg, cdns_ctrl->reg + ECC_CONFIG_0);

And here.. looks like there is no need for the memory barries, hence use the relaxed
versions of readl/writel. Same for the rest of the patch.

> +	return 0;
> +}
> +
> +static void cadence_nand_set_ecc_strength(struct cdns_nand_ctrl *cdns_ctrl,
> +					  u8 corr_str_idx)
> +{
> +	u32 reg;
> +
> +	if (cdns_ctrl->curr_corr_str_idx == corr_str_idx)
> +		return;
> +
> +	reg = readl(cdns_ctrl->reg + ECC_CONFIG_0);
> +	reg &= ~ECC_CONFIG_0_CORR_STR;
> +	reg |= FIELD_PREP(ECC_CONFIG_0_CORR_STR, corr_str_idx);
> +	writel(reg, cdns_ctrl->reg + ECC_CONFIG_0);
> +
> +	cdns_ctrl->curr_corr_str_idx = corr_str_idx;
> +}
> +
> +static u8 cadence_nand_get_ecc_strength_idx(struct cdns_nand_ctrl *cdns_ctrl,
> +					    u8 strength)
> +{
> +	u8 i, corr_str_idx = 0;
> +
> +	for (i = 0; i < BCH_MAX_NUM_CORR_CAPS; i++) {
> +		if (cdns_ctrl->ecc_strengths[i] == strength) {
> +			corr_str_idx = i;
> +			break;
> +		}
> +	}

Is it a error case when i == BCH_MAX_NUM_CORR_CAPS?

> +	return corr_str_idx;
> +}
> +
> +static int cadence_nand_set_skip_marker_val(struct cdns_nand_ctrl *cdns_ctrl,
> +					    u16 marker_value)
> +{
> +	u32 reg = 0;
> +
> +	if (cadence_nand_wait_for_value(cdns_ctrl, CTRL_STATUS,
> +					1000000,
> +					CTRL_STATUS_CTRL_BUSY, true))
> +		return -ETIMEDOUT;
> +
> +	reg = readl(cdns_ctrl->reg + SKIP_BYTES_CONF);
> +	reg &= ~SKIP_BYTES_MARKER_VALUE;
> +	reg |= FIELD_PREP(SKIP_BYTES_MARKER_VALUE,
> +		    marker_value);
> +
> +	writel(reg, cdns_ctrl->reg + SKIP_BYTES_CONF);
> +
> +	return 0;
> +}
> +
> +static int cadence_nand_set_skip_bytes_conf(struct cdns_nand_ctrl *cdns_ctrl,
> +					    u8 num_of_bytes,
> +					    u32 offset_value,
> +					    int enable)
> +{
> +	u32 reg = 0;
> +	u32 skip_bytes_offset = 0;

Please don't initialize variables if not necessary. You could also write this in a
single line.

	u32 skip_bytes_offset, reg;

Same for the rest of the patch.

> +	if (cadence_nand_wait_for_value(cdns_ctrl, CTRL_STATUS,
> +					1000000,
> +					CTRL_STATUS_CTRL_BUSY, true))
> +		return -ETIMEDOUT;
> +
> +	if (!enable) {
> +		num_of_bytes = 0;
> +		offset_value = 0;
> +	}
> +
> +	reg = readl(cdns_ctrl->reg + SKIP_BYTES_CONF);
> +	reg &= ~SKIP_BYTES_NUM_OF_BYTES;
> +	reg |= FIELD_PREP(SKIP_BYTES_NUM_OF_BYTES,
> +		    num_of_bytes);
> +	skip_bytes_offset = FIELD_PREP(SKIP_BYTES_OFFSET_VALUE,
> +				       offset_value);
> +
> +	writel(reg, cdns_ctrl->reg + SKIP_BYTES_CONF);
> +	writel(skip_bytes_offset, cdns_ctrl->reg + SKIP_BYTES_OFFSET);
> +
> +	return 0;
> +}
> +
> +/* Fucntions enables/disables hardware detection of erased data */

s/Fucntions/Function/, please use spellchecker. I'd also recommend to start all
single-line comments with a lower case (and without a dot in the end) because it is a
more common style in the kernel and is a bit easier for the eyes.

[snip]