[U-Boot] [PATCH v3 01/10] dm: i2c: Add a uclass for I2C

[Masahiro Yamada <yamada.m@jp.panasonic.com>]

Hi Simon,



On Wed, 3 Dec 2014 19:36:15 -0700
Simon Glass <sjg@chromium.org> wrote:

> Hi Masahiro,
> 
> On 3 December 2014 at 19:01, Masahiro Yamada <yamada.m@jp.panasonic.com> wrote:
> > Hi Simon,
> >
> >
> > More comments again.
> >
> >
> > On Mon, 24 Nov 2014 11:57:15 -0700
> >> +
> >> +static int i2c_probe_chip(struct udevice *bus, uint chip_addr)
> >> +{
> >> +     struct dm_i2c_ops *ops = i2c_get_ops(bus);
> >> +     struct i2c_msg msg[1];
> >> +     uint8_t ch = 0;
> >> +
> >> +     if (!ops->xfer)
> >> +             return -ENOSYS;
> >> +
> >> +     msg->addr = chip_addr;
> >> +     msg->flags = 0;
> >> +     msg->len = 1;
> >> +     msg->buf = &ch;
> >> +
> >> +     return ops->xfer(bus, msg, 1);
> >> +}
> >
> > i2c_probe_chip() issues a write transaction with one length,
> > but a read transaction should be used.
> >
> > For most of chips, the first write data is the first byte of
> > the offset address, so no real data will be written into the chip.
> >
> > But it is possible to have offset_address_length == 0.
> >
> > The read transaction is always safer than the write transaction.
> >
> 
> I originally had a probe method to allow the driver to decide what to
> do. Many drivers can in fact just write the address and don't need a
> data / offset byte. But not all. With a read byte the tegra driver
> takes ages to respond and the probe takes a lot longer than it should.

I can't believe this.
Reading one byte should be done pretty fast.
If not, the driver implemetation has crewed up.


> Yes I agree that read is safer, but even safer is nothing. Perhaps I
> should go back to having a probe_chip() method? Or at least it could
> be optional.

Safety takes precedence over efficiency.

I don't think we should go back to a probe_chip().

Probing happens only one time before the use of I2C.
Even if it takes longer, no problem.



Best Regards
Masahiro Yamada