On Wed, Jun 21, 2017 at 10:23 AM, Baolin Wang <baolin.wang@xxxxxxxxxxxxxx> wrote: > This patch adds the I2C controller driver for Spreadtrum platform. Needs more work. See my comments below. > +#include <linux/init.h> > +#include <linux/module.h> Since your answer to the comment about arch_initcall you perhaps need to get rid of tristate (use bool) and drop module.h here and all leftovers like MODULE_*() calls. > +#include <linux/slab.h> Should we include this still explicitly (after kernel.h)? > + > +#define I2C_CTL 0x0 > +#define I2C_ADDR_CFG 0x4 > +#define I2C_COUNT 0x8 > +#define I2C_RX 0xC > +#define I2C_TX 0x10 > +#define I2C_STATUS 0x14 > +#define I2C_HSMODE_CFG 0x18 > +#define I2C_VERSION 0x1C > +#define ADDR_DVD0 0x20 > +#define ADDR_DVD1 0x24 > +#define ADDR_STA0_DVD 0x28 > +#define ADDR_RST 0x2C 1. Please, use fixed sized values 0x00 and so on. 2. Please, low case. > +#define SPRD_I2C_TIMEOUT (msecs_to_jiffies(1000)) Redundant parens. > +static void sprd_i2c_dump_reg(struct sprd_i2c *i2c_dev) If you switch your driver to regmap API, you may drop this function completely. > + writel(tmp & (~STP_EN), i2c_dev->base + I2C_CTL); Redundant parens. Also pay attention to other similar places. > +static void sprd_i2c_write_bytes(struct sprd_i2c *i2c_dev, u8 *buf, u32 len) So, isn't better to provide a writesb(), readsb() to ia64 instead of doing below? > +{ > + u32 i = 0; > + > + for (i = 0; i < len; i++) { > + writel(buf[i], i2c_dev->base + I2C_TX); > + dev_dbg(&i2c_dev->adap.dev, "write bytes[%d] = %x\n", i, buf[i]); Moreover, don't waste lines in the kernel log buffer by doing this. Choose either %*ph (up to 64 bytes) or print_hex_dump() (Same for _read_bytes() below) > + } > +} > +static void sprd_i2c_set_full_thld(struct sprd_i2c *i2c_dev, u32 full_thld) > +{ > + unsigned int tmp = readl(i2c_dev->base + I2C_CTL); > + > + tmp &= ~(0xf << FIFO_AF_LVL); Magic. Define it using GENMASK() > + tmp |= (full_thld << FIFO_AF_LVL); Redundant parens. > + tmp &= ~(0xf << FIFO_AE_LVL); > + tmp |= (empty_thld << FIFO_AE_LVL); Same. > +static void sprd_i2c_opt_mode(struct sprd_i2c *i2c_dev, int rw) > +{ > + unsigned int cmd = readl(i2c_dev->base + I2C_CTL) & (~I2C_MODE); Redundant parens. > + writel((cmd | rw << 3), i2c_dev->base + I2C_CTL); Ditto. It's a C, and not a LISP :-) > +} > +static void sprd_i2c_data_transfer(struct sprd_i2c *i2c_dev) > +{ > + if ((msg->flags & I2C_M_RD)) { Redundant parens. > + /* Reset rx/tx fifos */ Noise. > + /* Set device address */ Ditto. > + /* Set I2C read or write bytes count */ Ditto. > + if ((msg->flags & I2C_M_RD)) { > + /* Set read operation */ > + sprd_i2c_opt_mode(i2c_dev, 1); > + /* Last byte transmission should excute stop operation */ > + sprd_i2c_send_stop(i2c_dev, 1); Same comments as above. > + } else { > + /* Set write operation */ Noise. > + /* Last byte transmission should excute stop operation */ > + if (is_last_msg) > + sprd_i2c_send_stop(i2c_dev, 1); > + else > + sprd_i2c_send_stop(i2c_dev, 0); sprd_i2c_send_stop(i2c_dev, !!is_last_msg); Though, consider to make is_last_msg boolean. > +static int sprd_i2c_master_xfer(struct i2c_adapter *i2c_adap, > + struct i2c_msg *msgs, int num) > +{ > + struct sprd_i2c *i2c_dev = i2c_adap->algo_data; > + int im, ret; > + > + ret = pm_runtime_get_sync(i2c_dev->dev); > + if (ret < 0) > + return ret; > + > + for (im = 0; im != num; im++) im < num - 1, and... ret = sprd_i2c_handle_msg(i2c_adap, &msgs[im], 0); ... ret = sprd_i2c_handle_msg(i2c_adap, &msgs[im++], 1); ...and we clearly see that you have messed up with returned code on each itteration here > + pm_runtime_mark_last_busy(i2c_dev->dev); > + pm_runtime_put_autosuspend(i2c_dev->dev); > + return (ret >= 0) ? im : ret; Usual pattern is ret < 0 ? ret : im; > +static void sprd_i2c_set_clk(struct sprd_i2c *i2c_dev, unsigned int freq) > +{ > + u32 apb_clk = i2c_dev->src_clk; > + u32 i2c_dvd = apb_clk / (4 * freq) - 1; > + u32 high = ((i2c_dvd << 1) * 2) / 5; > + u32 low = ((i2c_dvd << 1) * 3) / 5; > + u32 div0 = (high & 0xffff) << 16 | (low & 0xffff); > + u32 div1 = (high & 0xffff0000) | ((low & 0xffff0000) >> 16); Magic masks all over. Also it needs a comment what formula is used and how. > + > + writel(div0, i2c_dev->base + ADDR_DVD0); > + writel(div1, i2c_dev->base + ADDR_DVD1); > + > + if (freq == 400000) > + writel((6 * apb_clk) / 10000000, i2c_dev->base + ADDR_STA0_DVD); > + else if (freq == 100000) > + writel((4 * apb_clk) / 1000000, i2c_dev->base + ADDR_STA0_DVD); What about 3400000? What about the rest of the speeds, shouldn't you return an error from here? > +} > + > +static void sprd_i2c_enable(struct sprd_i2c *i2c_dev) > +{ > + unsigned int tmp = readl(i2c_dev->base + I2C_CTL); > + > + tmp &= ~(I2C_EN | I2C_TRIM_OPT | (0xF << FIFO_AF_LVL) | > + (0xF << FIFO_AE_LVL)); Magic masks (I saw them already above). > + /* Set rx fifo full data threshold */ Drop noise comments. They don't bring any value since you have nice function names. > + sprd_i2c_set_full_thld(i2c_dev, I2C_FIFO_FULL_THLD); > + /* Set tx fifo empty data threshold */ > + sprd_i2c_set_empty_thld(i2c_dev, I2C_FIFO_EMPTY_THLD); > + > + sprd_i2c_set_clk(i2c_dev, i2c_dev->bus_freq); > + /* Reset rx/tx fifo */ > + sprd_i2c_reset_fifo(i2c_dev); > + sprd_i2c_clear_irq(i2c_dev); > +static int sprd_i2c_clk_init(struct sprd_i2c *i2c_dev) > +{ > + struct clk *clk_i2c, *clk_parent; > + struct device_node *np = i2c_dev->adap.dev.of_node; > + > + clk_i2c = of_clk_get_by_name(np, "i2c"); What the issue to use resource agnostic API here, i.e. devm_clk_get() ? > + clk_parent = of_clk_get_by_name(np, "source"); Ditto. > + i2c_dev->clk = of_clk_get_by_name(np, "enable"); Ditto. > + if (!of_property_read_u32(dev->of_node, "clock-frequency", &prop)) > + i2c_dev->bus_freq = prop; > + > + sprd_i2c_clk_init(i2c_dev); > + platform_set_drvdata(pdev, i2c_dev); > + > + ret = clk_prepare_enable(i2c_dev->clk); > + if (ret) > + return ret; > + > + sprd_i2c_enable(i2c_dev); > + > +error: I would put it as err_rpm_put: > + pm_runtime_put_noidle(i2c_dev->dev); > + pm_runtime_disable(i2c_dev->dev); > + clk_disable_unprepare(i2c_dev->clk); > + return ret; > +} -- With Best Regards, Andy Shevchenko -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html