Re: [PACTH v2] spi: imx: add 16/32 bits per word support for target mode

[Frank Li <Frank.li@nxp.com>]

On Tue, Oct 28, 2025 at 09:29:02AM +0000, Carlos Song wrote:
>
>
> > -----Original Message-----
> > From: Frank Li <frank.li@nxp.com>
> > Sent: Tuesday, October 28, 2025 2:50 AM
> > To: Carlos Song <carlos.song@nxp.com>
> > Cc: mkl@pengutronix.de; broonie@kernel.org; shawnguo@kernel.org;
> > s.hauer@pengutronix.de; kernel@pengutronix.de; festevam@gmail.com;
> > linux-spi@vger.kernel.org; imx@lists.linux.dev;
> > linux-arm-kernel@lists.infradead.org; linux-kernel@vger.kernel.org
> > Subject: Re: [PACTH v2] spi: imx: add 16/32 bits per word support for target
> > mode
> >
> > On Mon, Oct 27, 2025 at 03:39:36AM +0000, Carlos Song wrote:
> > >
> > >
> > > > -----Original Message-----
> > > > From: Frank Li <frank.li@nxp.com>
> > > > Sent: Friday, October 24, 2025 9:25 PM
> > > > To: Carlos Song <carlos.song@nxp.com>
> > > > Cc: mkl@pengutronix.de; broonie@kernel.org; shawnguo@kernel.org;
> > > > s.hauer@pengutronix.de; kernel@pengutronix.de; festevam@gmail.com;
> > > > linux-spi@vger.kernel.org; imx@lists.linux.dev;
> > > > linux-arm-kernel@lists.infradead.org; linux-kernel@vger.kernel.org
> > > > Subject: Re: [PACTH v2] spi: imx: add 16/32 bits per word support
> > > > for target mode
> > > >
> > > > On Fri, Oct 24, 2025 at 07:31:07PM +0800, carlos.song@nxp.com wrote:
> > > > > From: Carlos Song <carlos.song@nxp.com>
> > > > >
> > > > > Now for ECSPI only support 8 bits per word in target mode.
> > > > > Enable 16/32 bits per word support for spi-imx target mode.
> > > > >
> > > > > Signed-off-by: Carlos Song <carlos.song@nxp.com>
> > > > > Signed-off-by: Clark Wang <xiaoning.wang@nxp.com>
> > > > > ---
> > > > >  drivers/spi/spi-imx.c | 12 ++++++++++--
> > > > >  1 file changed, 10 insertions(+), 2 deletions(-)
> > > > >
> > > > > diff --git a/drivers/spi/spi-imx.c b/drivers/spi/spi-imx.c index
> > > > > 155ddeb8fcd4..017f83f5dfdf 100644
> > > > > --- a/drivers/spi/spi-imx.c
> > > > > +++ b/drivers/spi/spi-imx.c
> > > > > @@ -424,8 +424,12 @@ static void spi_imx_buf_tx_swap(struct
> > > > > spi_imx_data *spi_imx)
> > > > >
> > > > >  static void mx53_ecspi_rx_target(struct spi_imx_data *spi_imx)  {
> > > > > -	u32 val = ioread32be(spi_imx->base + MXC_CSPIRXDATA);
> > > > > +	u32 val = readl(spi_imx->base + MXC_CSPIRXDATA);
> > > > >
> > > > > +	if (spi_imx->bits_per_word <= 8)
> > > > > +		swab32s(&val);
> > > > > +	else if (spi_imx->bits_per_word <= 16)
> > > > > +		swahw32s(&val);
> > > >
> > > > Needn't swap when bits_per_word > 24, like 32? or our hardware max
> > > > support to 16?
> > > >
> > > > Frank
> > >
> > > I think we don't need. We are reading/writing FIFO by 32-bits word. In
> > > this case, we just need keep the default byte order.
> > >
> > > Is it more reasonable only support bits per word=8/16/32 for target mode?
> > > -b 8/16/32 can cover most use case.
> >
> > yes, I only can't understand why needn't swap at 32bits workds, but other
> > needs.
> >
> > FIFO 31..24  23..16  15..8  7..0
> >       B0      B1     B2      B3
> >
> > next you
> >
> > in memory val is
> >
> > 0x0: B3
> > 0x1: B2
> > 0x2: B1
> > 0x3: B0
> >
> > swab32s() to  bits_per_work 8
> >
> > 0x0: B0
> > 0x1: B1
> > 0x2: B2
> > 0x3: B3
> >
> > if bits per_word 16
> >
> > val
> >
> > 0x0: X
> > 0x1: X
> > 0x2: B1
> > 0x3: B0
> >
> > after swahw32s change to
> >
> > 0x0: B1
> > 0x1: B0
> > 0x2: x
> > 0x3: x
> >
> > B0 and B1 still be swapped.
> >
> > Does SPI defined term word fixed big-endian?
> >
> > Frank
> >
>
> No, SPI doesn't define this. This swap is caused by the SPI-IMX FIFO design.
> SPI-IMX FIFO is fixed little-endian.
>
> One example to easy understand:
>
> If buf is:
>
> 0XB0B1B2B3
>
> So data in TXFIFO is:
> FIFO 31..24  23..16  15..8  7..0
>   B0        B1		B2   B3
>
> So we can always see data on SPI bus:
> 0xB3B2B1B0
>
> For this patch:
> bits per word= 8: it means one byte is one word, so it should not effect by big endian or little endian, so every bytes should keep the same order between memory and SPI bus.
> We write FIFO by 32 bits word, So we should swap every bytes in this word from 0XB0B1B2B3 to 0XB3B2B1B0
>
> Data in TXFIFO is
> FIFO 31..24  23..16  15..8  7..0
>   B3        B2		B1  B0
> Then we can see data on SPI bus is
> 0XB0B1B2B3
>
> For bits per word = 16, it means two bytes in one word, so every 2 bytes is one word, the word should be a little endian word,
>
> so we should swap half word from
> 0XB0B1B2B3 to 0XB2B3B0B1
>
> Then data in TXFIFO is:
> FIFO 31..24  23..16  15..8  7..0
>   B2        B3		B0  B1
> Then we can see data on SPI bus is
> 0XB1B0B3B2
>
> For bis per word =32, it means for bytes in one word, so every 4 bytes is one word. The whole word is a little endian.
> So we keep the default order.
> 0XB0B1B2B3
>
> FIFO 31..24  23..16  15..8  7..0
>   B0        B1		B2  B3
>
> So we can see the data on bus is:
> 0xB3B2B1B0

Does SPI require send MSB/LSB first of a CPU's word?

Frank

>
> The RX handle is the same with TX. From bus to FIFO is a little word, so swap in memory.
>
> Carlos
>
> > > Carlos
> > > > >  	if (spi_imx->rx_buf) {
> > > > >  		int n_bytes = spi_imx->target_burst % sizeof(val);
> > > > >
> > > > > @@ -453,12 +457,16 @@ static void mx53_ecspi_tx_target(struct
> > > > spi_imx_data *spi_imx)
> > > > >  	if (spi_imx->tx_buf) {
> > > > >  		memcpy(((u8 *)&val) + sizeof(val) - n_bytes,
> > > > >  		       spi_imx->tx_buf, n_bytes);
> > > > > +		if (spi_imx->bits_per_word <= 8)
> > > > > +			swab32s(&val);
> > > > > +		else if (spi_imx->bits_per_word <= 16)
> > > > > +			swahw32s(&val);
> > > > >  		spi_imx->tx_buf += n_bytes;
> > > > >  	}
> > > > >
> > > > >  	spi_imx->count -= n_bytes;
> > > > >
> > > > > -	iowrite32be(val, spi_imx->base + MXC_CSPITXDATA);
> > > > > +	writel(val, spi_imx->base + MXC_CSPITXDATA);
> > > > >  }
> > > > >
> > > > >  /* MX51 eCSPI */
> > > > > --
> > > > > 2.34.1
> > > > >