
On Wed, Feb 13, 2019 at 9:21 AM André Przywara andre.przywara@arm.com wrote:
On 09/02/2019 13:14, Jagan Teki wrote:
Add A31 spi controller support for existing sun4i_spi driver via driver data, this would simply add A31 register along with proper register bits via enum sets.
Signed-off-by: Jagan Teki jagan@amarulasolutions.com
drivers/spi/Kconfig | 4 +- drivers/spi/sun4i_spi.c | 84 ++++++++++++++++++++++++++++++++++++++++- 2 files changed, 85 insertions(+), 3 deletions(-)
diff --git a/drivers/spi/Kconfig b/drivers/spi/Kconfig index ac7fbab841..15207d23c1 100644 --- a/drivers/spi/Kconfig +++ b/drivers/spi/Kconfig @@ -223,9 +223,9 @@ config STM32_QSPI this ST IP core.
config SUN4I_SPI
bool "Allwinner A10 SoCs SPI controller"
bool "Allwinner A10/A31 SoCs SPI controller" help
SPI driver for Allwinner sun4i, sun5i and sun7i SoCs
This enables using the SPI controller on the Allwinner A10/A31 SoCs.
config TEGRA114_SPI bool "nVidia Tegra114 SPI driver" diff --git a/drivers/spi/sun4i_spi.c b/drivers/spi/sun4i_spi.c index aeed68764c..36f2215f7d 100644 --- a/drivers/spi/sun4i_spi.c +++ b/drivers/spi/sun4i_spi.c @@ -24,6 +24,7 @@ #include <spi.h> #include <errno.h> #include <fdt_support.h> +#include <reset.h> #include <wait_bit.h>
#include <asm/bitops.h> @@ -84,6 +85,18 @@ #define SUN4I_FIFO_STA_TF_CNT_MASK 0x7f #define SUN4I_FIFO_STA_TF_CNT_BITS 16
+/* sun6i spi registers */ +#define SUN6I_GBL_CTL_REG 0x04 +#define SUN6I_TFR_CTL_REG 0x08 +#define SUN6I_FIFO_CTL_REG 0x18 +#define SUN6I_FIFO_STA_REG 0x1c +#define SUN6I_CLK_CTL_REG 0x24 +#define SUN6I_BURST_CNT_REG 0x30 +#define SUN6I_XMIT_CNT_REG 0x34 +#define SUN6I_BURST_CTL_REG 0x38 +#define SUN6I_TXDATA_REG 0x200 +#define SUN6I_RXDATA_REG 0x300
#define SUN4I_SPI_MAX_RATE 24000000 #define SUN4I_SPI_MIN_RATE 3000 #define SUN4I_SPI_DEFAULT_RATE 1000000 @@ -106,6 +119,7 @@ enum sun4i_spi_regs { /* sun spi register bits */ enum sun4i_spi_bits { SPI_GCR_TP,
SPI_GCR_SRST, SPI_TCR_CPHA, SPI_TCR_CPOL, SPI_TCR_CS_ACTIVE_LOW,
@@ -133,6 +147,7 @@ struct sun4i_spi_platdata { struct sun4i_spi_priv { struct sun4i_spi_variant *variant; struct clk clk_ahb, clk_mod;
struct reset_ctl reset; u32 base_addr; u32 freq; u32 mode;
@@ -255,7 +270,10 @@ static int sun4i_spi_parse_pins(struct udevice *dev) if (pin < 0) break;
sunxi_gpio_set_cfgpin(pin, SUNXI_GPC_SPI0);
if (IS_ENABLED(CONFIG_MACH_SUN50I))
sunxi_gpio_set_cfgpin(pin, SUN50I_GPC_SPI0);
else
sunxi_gpio_set_cfgpin(pin, SUNXI_GPC_SPI0); sunxi_gpio_set_drv(pin, drive); sunxi_gpio_set_pull(pin, pull); }
@@ -271,6 +289,8 @@ static inline int sun4i_spi_set_clock(struct udevice *dev, bool enable) if (!enable) { clk_disable(&priv->clk_ahb); clk_disable(&priv->clk_mod);
if (reset_valid(&priv->reset))
reset_assert(&priv->reset); return 0; }
@@ -286,8 +306,18 @@ static inline int sun4i_spi_set_clock(struct udevice *dev, bool enable) goto err_ahb; }
if (reset_valid(&priv->reset)) {
ret = reset_deassert(&priv->reset);
if (ret) {
dev_err(dev, "failed to deassert reset\n");
goto err_mod;
}
}
return 0;
+err_mod:
clk_disable(&priv->clk_mod);
err_ahb: clk_disable(&priv->clk_ahb); return ret; @@ -328,6 +358,12 @@ static int sun4i_spi_probe(struct udevice *bus) return ret; }
ret = reset_get_by_index(bus, 0, &priv->reset);
if (ret && ret != -ENOENT) {
dev_err(dev, "failed to get reset\n");
return ret;
}
sun4i_spi_parse_pins(bus); priv->variant = plat->variant;
@@ -351,6 +387,10 @@ static int sun4i_spi_claim_bus(struct udevice *dev) SUN4I_CTL_ENABLE | SUN4I_CTL_MASTER | variant->bits[SPI_GCR_TP]);
if (IS_ENABLED(CONFIG_SUNXI_GEN_SUN6I))
setbits_le32(priv->base_addr + variant->regs[SPI_GCR],
variant->bits[SPI_GCR_SRST]);
Just a note: This makes the driver only support one class of device, set at compile time. Which means we could have used an #ifdef approach from the beginning .... Can you make this dependent on some bits in variant? That should solve it.
I believe the argument for IS_ENABLED from working in the kernel is that it still gives you compile-time checks, whereas #ifdef doesn't.
Though I believe your ultimate goal is to have one image for all? In that case checking against the variants at runtime is better.
ChenYu