Re: [PATCH] spi: atmel-quadspi: Add support for sama7g5 QSPI

[<Tudor.Ambarus@xxxxxxxxxxxxx>]

On 12/14/21 3:34 PM, Tudor Ambarus wrote:
> The sama7g5 QSPI controller uses dedicated clocks for the
> QSPI Controller Interface and the QSPI Controller Core, and
> requires synchronization before accessing registers or bit
> fields.
> 
> QSPI_SR.SYNCBSY must be zero before accessing any of the bits:
> QSPI_CR.QSPIEN, QSPI_CR.QSPIDIS, QSPI_CR.SRFRSH, QSPI_CR.SWRST,
> QSPI_CR.UPDCFG, QSPI_CR.STTFR, QSPI_CR.RTOUT, QSPI_CR.LASTXFER.
> 
> Also, the QSPI controller core configuration can be updated by
> writing the QSPI_CR.UPDCFG bit to ‘1’. This is needed by the
> following registers: QSPI_MR, QSPI_SCR, QSPI_IAR, QSPI_WICR,
> QSPI_IFR, QSPI_RICR, QSPI_SMR, QSPI_SKR,QSPI_REFRESH, QSPI_WRACNT
> QSPI_PCALCFG.
> 
> The Octal SPI supports frequencies up to 200 MHZ DDR. The need
> for output impedance calibration arises. To avoid the degradation
> of the signal quality, a PAD calibration cell is used to adjust
> the output impedance to the driven I/Os.
> 
> The transmission flow requires different sequences for setting
> the configuration and for the actual transfer, than what is in
> the sama5d2 and sam9x60 versions of the IP. Different interrupts
> are handled. aq->ops->set_cfg() and aq->ops->transfer() are
> introduced to help differentiating the flows.
> 
> Tested single and octal SPI mode with mx66lm1g45g.
> 
> Signed-off-by: Tudor Ambarus <tudor.ambarus@xxxxxxxxxxxxx>
> ---
>  drivers/spi/atmel-quadspi.c | 921 ++++++++++++++++++++++++++++++++++--
>  1 file changed, 869 insertions(+), 52 deletions(-)
> 
> diff --git a/drivers/spi/atmel-quadspi.c b/drivers/spi/atmel-quadspi.c
> index 92d9610df1fd..8980a729dd53 100644
> --- a/drivers/spi/atmel-quadspi.c
> +++ b/drivers/spi/atmel-quadspi.c
> @@ -11,11 +11,15 @@

cut

> +static int atmel_qspi_reg_sync(struct atmel_qspi *aq)
> +{
> +	u32 val;
> +	int ret;
> +
> +	ret = readl_poll_timeout(aq->regs + QSPI_SR2, val,
> +				 !(val & QSPI_SR2_SYNCBSY), 40,
> +				 ATMEL_QSPI_SYNC_TIMEOUT);
> +	return ret;

return readl_poll_timeout();

> +}

cut

> +static int atmel_qspi_dma_xfer(struct atmel_qspi *aq, struct dma_chan *chan,
> +			       dma_addr_t dma_dst, dma_addr_t dma_src,
> +			       unsigned int len)
> +{
> +	struct dma_async_tx_descriptor *tx;
> +	dma_cookie_t cookie;
> +	int ret;
> +
> +	tx = dmaengine_prep_dma_memcpy(chan, dma_dst, dma_src, len,
> +				       DMA_PREP_INTERRUPT | DMA_CTRL_ACK);
> +	if (!tx) {
> +		dev_err(&aq->pdev->dev, "device_prep_dma_memcpy error\n");
> +		return -EIO;
> +	}
> +
> +	reinit_completion(&aq->dma_completion);
> +	tx->callback = atmel_qspi_dma_callback;
> +	tx->callback_param = aq;
> +	cookie = tx->tx_submit(tx);
> +	ret = dma_submit_error(cookie);
> +	if (ret) {
> +		dev_err(&aq->pdev->dev, "dma_submit_error %d\n", cookie);
> +		return ret;
> +	}
> +
> +	dma_async_issue_pending(chan);
> +	ret = wait_for_completion_timeout(&aq->dma_completion,
> +					  msecs_to_jiffies(20 * ATMEL_QSPI_TIMEOUT));

20 * ATMEL_QSPI_TIMEOUT, this is a leftover from a debug session.

I'll send a new version to fix these small issues.

Cheers,
ta