From 412a0e175069102761e85d4518c50dfb9b03519b Mon Sep 17 00:00:00 2001 From: Sascha Hauer Date: Mon, 31 Aug 2015 17:09:23 +0200 Subject: spi: i.MX: optimize transfers for ECSPI v2.3 Instead of writing one word to the txfifo and then wait until one is received in the rxfifo we can write while the txfifos are not full and read as long the rxfifos contain data. This makes transfers for the m25p80 driver around 7 times faster here. Unlike the last version this time we only optimize the common case with 8 bits per word and SPI_LSB_FIRST cleared. The other cases would require more bit shuffling of the data. Signed-off-by: Sascha Hauer --- drivers/spi/imx_spi.c | 73 ++++++++++++++++++++++++++++++++++++++++++++++++++- 1 file changed, 72 insertions(+), 1 deletion(-) (limited to 'drivers/spi') diff --git a/drivers/spi/imx_spi.c b/drivers/spi/imx_spi.c index 9a3f05cdec..2e489674d8 100644 --- a/drivers/spi/imx_spi.c +++ b/drivers/spi/imx_spi.c @@ -380,6 +380,77 @@ static void imx_spi_do_transfer(struct spi_device *spi) } } +static int cspi_2_3_xchg_burst(struct spi_device *spi) +{ + struct imx_spi *imx = container_of(spi->master, struct imx_spi, master); + int now, txlen, rxlen; + u32 ctrl; + void __iomem *base = imx->regs; + + now = min(imx->xfer_len, 512); + now >>= 2; + + if (!now) + return 0; + + txlen = rxlen = now; + + ctrl = readl(base + CSPI_2_3_CTRL); + ctrl &= ~(0xfff << CSPI_2_3_CTRL_BL_OFFSET); + ctrl |= ((txlen * 32) - 1) << CSPI_2_3_CTRL_BL_OFFSET; + ctrl |= 1 << 3; + writel(ctrl, base + CSPI_2_3_CTRL); + + while (txlen || rxlen) { + u32 status = readl(base + CSPI_2_3_STAT); + + if (txlen && !(status & CSPI_2_3_STAT_TF)) { + if (imx->tx_buf) { + u32 data = swab32(*(u32 *)imx->tx_buf); + writel(data, base + CSPI_2_3_TXDATA); + imx->tx_buf += sizeof(u32); + } else { + writel(0, base + CSPI_2_3_TXDATA); + } + txlen--; + } + + if (rxlen && (status & CSPI_2_3_STAT_RR)) { + u32 data = readl(base + CSPI_2_3_RXDATA); + + if (imx->rx_buf) { + *(u32 *)imx->rx_buf = swab32(data); + imx->rx_buf += sizeof(u32); + } + + rxlen--; + } + } + + imx->xfer_len -= now * 4; + + return now; +} + +static void cspi_2_3_do_transfer(struct spi_device *spi) +{ + struct imx_spi *imx = container_of(spi->master, struct imx_spi, master); + u32 ctrl; + + if (imx->bits_per_word == 8 && !(spi->mode & SPI_LSB_FIRST)) + while (cspi_2_3_xchg_burst(spi) > 0); + + if (!imx->xfer_len) + return; + + ctrl = readl(imx->regs + CSPI_2_3_CTRL); + ctrl &= ~(0xfff << CSPI_2_3_CTRL_BL_OFFSET); + ctrl |= (spi->bits_per_word - 1) << CSPI_2_3_CTRL_BL_OFFSET; + writel(ctrl, imx->regs + CSPI_2_3_CTRL); + + imx_spi_do_transfer(spi); +} + static int imx_spi_transfer(struct spi_device *spi, struct spi_message *mesg) { struct imx_spi *imx = container_of(spi->master, struct imx_spi, master); @@ -437,7 +508,7 @@ static __maybe_unused struct spi_imx_devtype_data spi_imx_devtype_data_0_7 = { static __maybe_unused struct spi_imx_devtype_data spi_imx_devtype_data_2_3 = { .chipselect = cspi_2_3_chipselect, - .do_transfer = imx_spi_do_transfer, + .do_transfer = cspi_2_3_do_transfer, .xchg_single = cspi_2_3_xchg_single, }; -- cgit v1.2.3