Re: [PATCH v2 02/19] spi: dw: Add Tx/Rx finish wait methods to the MID DMA

[Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>]

On Fri, May 15, 2020 at 01:47:41PM +0300, Serge Semin wrote:
> Since DMA transfers are performed asynchronously with actual SPI
> transaction, then even if DMA transfers are finished it doesn't mean
> all data is actually pushed to the SPI bus. Some data might still be
> in the controller FIFO. This is specifically true for Tx-only
> transfers. In this case if the next SPI transfer is recharged while
> a tail of the previous one is still in FIFO, we'll loose that tail
> data. In order to fix this lets add the wait procedure of the Tx/Rx
> SPI transfers completion after the corresponding DMA transactions
> are finished.

General question, doesn't spi core provides us some helpers like
spi_delay_exec()?

> Co-developed-by: Georgy Vlasov <Georgy.Vlasov@xxxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Georgy Vlasov <Georgy.Vlasov@xxxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Serge Semin <Sergey.Semin@xxxxxxxxxxxxxxxxxxxx>
> Cc: Ramil Zaripov <Ramil.Zaripov@xxxxxxxxxxxxxxxxxxxx>
> Cc: Alexey Malahov <Alexey.Malahov@xxxxxxxxxxxxxxxxxxxx>
> Cc: Thomas Bogendoerfer <tsbogend@xxxxxxxxxxxxxxxx>
> Cc: Paul Burton <paulburton@xxxxxxxxxx>
> Cc: Ralf Baechle <ralf@xxxxxxxxxxxxxx>
> Cc: Arnd Bergmann <arnd@xxxxxxxx>
> Cc: Allison Randal <allison@xxxxxxxxxxx>
> Cc: Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>
> Cc: Gareth Williams <gareth.williams.jx@xxxxxxxxxxx>
> Cc: Rob Herring <robh+dt@xxxxxxxxxx>
> Cc: linux-mips@xxxxxxxxxxxxxxx
> Cc: devicetree@xxxxxxxxxxxxxxx
> 
> ---
> 
> Changelog v2:
> - Use conditional statement instead of the ternary operator in the ref
>   clock getter.

> - Move the patch to the head of the series so one could be picked up to
>   the stable kernels as a fix.

You forgot a Fixes tag.

> ---
>  drivers/spi/spi-dw-mid.c | 50 ++++++++++++++++++++++++++++++++++++++++
>  drivers/spi/spi-dw.h     | 10 ++++++++
>  2 files changed, 60 insertions(+)
> 
> diff --git a/drivers/spi/spi-dw-mid.c b/drivers/spi/spi-dw-mid.c
> index 177e1f5ec62b..7a5ae1506365 100644
> --- a/drivers/spi/spi-dw-mid.c
> +++ b/drivers/spi/spi-dw-mid.c
> @@ -16,7 +16,9 @@
>  #include <linux/irqreturn.h>
>  #include <linux/pci.h>
>  #include <linux/platform_data/dma-dw.h>

> +#include <linux/delay.h>

Keep it in order.

>  
> +#define WAIT_RETRIES	5
>  #define RX_BUSY		0
>  #define TX_BUSY		1
>  
> @@ -141,6 +143,28 @@ static enum dma_slave_buswidth convert_dma_width(u32 dma_width) {
>  	return DMA_SLAVE_BUSWIDTH_UNDEFINED;
>  }
>  
> +static inline bool dw_spi_dma_tx_busy(struct dw_spi *dws)
> +{
> +	return !(dw_readl(dws, DW_SPI_SR) & SR_TF_EMPT);
> +}
> +
> +static void dw_spi_dma_wait_tx_done(struct dw_spi *dws)
> +{
> +	int retry = WAIT_RETRIES;
> +	unsigned long ns;
> +
> +	ns = (NSEC_PER_SEC / spi_get_clk(dws)) * dws->n_bytes * BITS_PER_BYTE;
> +	ns *= dw_readl(dws, DW_SPI_TXFLR);
> +

> +	while (dw_spi_dma_tx_busy(dws) && retry--)
> +		ndelay(ns);

This misses power management for CPU and do you really need this to be atomic?
At the end why not to use readx_poll_timeout() ?

> +	if (retry < 0) {

Usually we do

	unsigned int retries = NNNN;

	do {
		...
	} while (--retries);
	if (!retries)
		...

But in any case, see above.

> +		dev_err(&dws->master->dev, "Tx hanged up\n");
> +		dws->master->cur_msg->status = -EIO;
> +	}
> +}

Same comments to Rx part.

> +
>  /*
>   * dws->dma_chan_busy is set before the dma transfer starts, callback for tx
>   * channel will clear a corresponding bit.
> @@ -149,6 +173,8 @@ static void dw_spi_dma_tx_done(void *arg)
>  {
>  	struct dw_spi *dws = arg;
>  
> +	dw_spi_dma_wait_tx_done(dws);
> +
>  	clear_bit(TX_BUSY, &dws->dma_chan_busy);
>  	if (test_bit(RX_BUSY, &dws->dma_chan_busy))
>  		return;
> @@ -188,6 +214,28 @@ static struct dma_async_tx_descriptor *dw_spi_dma_prepare_tx(struct dw_spi *dws,
>  	return txdesc;
>  }
>  
> +static inline bool dw_spi_dma_rx_busy(struct dw_spi *dws)
> +{
> +	return !!(dw_readl(dws, DW_SPI_SR) & SR_RF_NOT_EMPT);
> +}
> +
> +static void dw_spi_dma_wait_rx_done(struct dw_spi *dws)
> +{
> +	int retry = WAIT_RETRIES;
> +	unsigned long ns;
> +
> +	ns = (NSEC_PER_SEC / spi_get_clk(dws)) * dws->n_bytes * BITS_PER_BYTE;
> +	ns *= dw_readl(dws, DW_SPI_RXFLR);
> +
> +	while (dw_spi_dma_rx_busy(dws) && retry--)
> +		ndelay(ns);
> +
> +	if (retry < 0) {
> +		dev_err(&dws->master->dev, "Rx hanged up\n");
> +		dws->master->cur_msg->status = -EIO;
> +	}
> +}
> +
>  /*
>   * dws->dma_chan_busy is set before the dma transfer starts, callback for rx
>   * channel will clear a corresponding bit.
> @@ -196,6 +244,8 @@ static void dw_spi_dma_rx_done(void *arg)
>  {
>  	struct dw_spi *dws = arg;
>  
> +	dw_spi_dma_wait_rx_done(dws);
> +
>  	clear_bit(RX_BUSY, &dws->dma_chan_busy);
>  	if (test_bit(TX_BUSY, &dws->dma_chan_busy))
>  		return;
> diff --git a/drivers/spi/spi-dw.h b/drivers/spi/spi-dw.h
> index e92d43b9a9e6..81364f501b7e 100644
> --- a/drivers/spi/spi-dw.h
> +++ b/drivers/spi/spi-dw.h
> @@ -210,6 +210,16 @@ static inline void spi_set_clk(struct dw_spi *dws, u16 div)
>  	dw_writel(dws, DW_SPI_BAUDR, div);
>  }
>  
> +static inline u32 spi_get_clk(struct dw_spi *dws)
> +{
> +	u32 div = dw_readl(dws, DW_SPI_BAUDR);
> +
> +	if (!div)
> +		return 0;
> +
> +	return dws->max_freq / div;
> +}
> +
>  /* Disable IRQ bits */
>  static inline void spi_mask_intr(struct dw_spi *dws, u32 mask)
>  {
> -- 
> 2.25.1
> 

-- 
With Best Regards,
Andy Shevchenko