Re: [PATCH v3] Input: add bu21029 touch driver

From: Dmitry Torokhov
Date: Wed May 16 2018 - 13:11:29 EST


On Wed, May 16, 2018 at 01:24:24PM +0000, Jonas Mark (BT-FIR/ENG1) wrote:
> Hello Andy,
>
> > > Add Rohm BU21029 resistive touch panel controller support with I2C
> > > interface.
> >
> > > +#include <linux/of.h>
> >
> > This becomes redundant (see below).
>
> Removed.
>
> > > +#define STOP_DELAY_US 50L
> > > +#define START_DELAY_MS 2L
> > > +#define BUF_LEN 8L
> >
> > No need to use L for such small numbers. Integer promotion is a part
> > of C standard.
>
> OK.
>
> > > +#define SCALE_12BIT (1 << 12)
> > > +#define MAX_12BIT ((1 << 12) - 1)
> >
> > BIT(12)
> > GENMASK(11, 0)
>
> We are not convinced that we should use BIT() and GENMASK() here.
>
> The reason is that SCALE_12BIT is actually not used as a bit but as an
> input value for DIV_ROUND_CLOSEST. We think that the BIT() macro will
> hide the meaning of the value.
>
> MAX_12BIT is also a value and not a bit mask. Thus, we also think that
> using the GENMASK() macro will hide its purpose. Also, the
> documentation of GENMASK() says that it is a mask and not a value.

I agree.

>
> > > +static int bu21029_touch_report(struct bu21029_ts_data *bu21029)
> > > +{
> > > + struct i2c_client *i2c = bu21029->client;
> > > + u8 buf[BUF_LEN];
> > > + int error = bu21029_touch_report(bu21029);
> >
> > > +
> >
> > Redundant empty line.
>
> Removed.
>
> > > + if (error) {
> >
> > > + dev_err(&i2c->dev, "failed to report (error: %d)\n", error);
> >
> > Potential spamming case.
> >
> > > + return IRQ_NONE;
> > > + }
>
> You are right, we will remove the error message.
>
> > > +static void bu21029_stop_chip(struct input_dev *dev)
> > > +{
> > > + struct bu21029_ts_data *bu21029 = input_get_drvdata(dev);
> > > +
> > > + disable_irq(bu21029->client->irq);
> > > + del_timer_sync(&bu21029->timer);
> > > +
> > > + /* put chip into reset */
> > > + gpiod_set_value_cansleep(bu21029->reset_gpios, 1);
> >
> > > + udelay(STOP_DELAY_US);
> >
> > udelay() ?!
> >
> > > +}
>
> According to the datasheet disabling the chip will take 30 microseconds.
> In the defines we added a buffer of 20 microseconds and thus
> STOP_DELAY_US is 50. The function guarantees that the chip is stopped
> before it returns.
>
> We think that it is ok to use udelay() here because in normal operation
> the chip is not stopped. It is only stopped when loading or unloading
> the driver, or when the system suspends.
>
> We would like to keep it like it is.

The issue is not with having delay here, but the kind of delay you are
using: udelay makes CPU spin for given amount of time; you really want
msleep() or usleep_range() here.

>
> > > +static int bu21029_start_chip(struct input_dev *dev)
> > > +{
> >
> > > + u16 hwid;
> > > +
> > > + /* take chip out of reset */
> > > + gpiod_set_value_cansleep(bu21029->reset_gpios, 0);
> >
> > > + mdelay(START_DELAY_MS);
> >
> > mdelay()?!

Same here - replace with msleep().

> >
> > > +
> > > + error = i2c_smbus_read_i2c_block_data(i2c,
> > > + BU21029_HWID_REG,
> > > + 2,
> > > + (u8 *)&hwid);
> > > + if (error < 0) {
> > > + dev_err(&i2c->dev, "failed to read HW ID\n");
> > > + goto out;
> > > + }
>
> After de-asserting the reset chip takes 1 millisecond until it is
> operational. We added a 1 millisecond buffer to it. Thus,
> START_DELAY_MS is 2.
>
> The following I2C read will not succeed without waiting for the chip
> being ready.
>
> > > + if (cpu_to_be16(hwid) != SUPPORTED_HWID) {
> >
> > Hmm... Why cpu_to_be16() is required?
> >
> > > + dev_err(&i2c->dev, "unsupported HW ID 0x%x\n", hwid);
> > > + error = -ENODEV;
> > > + goto out;
> > > + }
> > > +}
>
> You are right, it works but what we meant to do here is to convert the
> chip's value (big endian) into the CPU endianness. We will change it to
> be16_to_cpu().
>
> > > +static int bu21029_parse_dt(struct bu21029_ts_data *bu21029)
> >
> > You can get rid of DT requirement by...
> >
> > > +{
> > > + struct device *dev = &bu21029->client->dev;
> > > + struct device_node *np = dev->of_node;
> > > + u32 val32;
> > > + int error;
> >
> > > + if (!np) {
> > > + dev_err(dev, "no device tree data\n");
> > > + return -EINVAL;
> > > + }
> >
> > (this becomes redundant)
> >
> > > +
> > > + bu21029->reset_gpios = devm_gpiod_get(dev, "reset",
> > GPIOD_OUT_HIGH);
> > > + if (IS_ERR(bu21029->reset_gpios)) {
> > > + error = PTR_ERR(bu21029->reset_gpios);
> > > + if (error != -EPROBE_DEFER)
> > > + dev_err(dev, "invalid 'reset-gpios':%d\n", error);
> > > + return error;
> > > + }
> > > +
> >
> > > + if (of_property_read_u32(np, "rohm,x-plate-ohms", &val32)) {
> >
> > ...simple calling device_property_read_u32() instead.
> >
> > > + dev_err(dev, "invalid 'x-plate-ohms' supplied\n");
> > > + return -EINVAL;
> > > + }
> > > + bu21029->x_plate_ohms = val32;
> > > +
> > > + touchscreen_parse_properties(bu21029->in_dev, false, &bu21029->prop);
> > > +
> > > + return 0;
> > > +}
>
> Thank you, changed.
>
> > > +#ifdef CONFIG_PM_SLEEP
> >
> > Instead...
> >
> > > +static int bu21029_suspend(struct device *dev)
> >
> > ...use __maby_unused annotation.
> >
> > > +static int bu21029_resume(struct device *dev)
> >
> > Ditto.
>
> OK, added.

You also need to drop #ifdef CONFIG_SLEEP. That's the point: we want to
reduce amount of conditionally compiled code and use __maybe_unused to
shut off compiler warning about some functions not used in certain
configurations. We rely on linker to eliminate dead functions from
executable.

Thanks.

--
Dmitry