[PATCH 1/2] serial_ns16550: move switch from ns16550_{read, write}() to ns16550_probe()

[Antony Pavlov]

On 4 August 2011 11:19, Sascha Hauer <s.hauer at pengutronix.de> wrote:

>>
>> +#define NS16550_READ_WRITE_UART_FUNC(pfx, cmdr, cmdw, mask) \
>> +static unsigned int ns16550_generic_##pfx##_read(unsigned long base, \
>> +             unsigned char reg_idx)          \
>> +{                                            \
>> +     return cmdr((char *)base + reg_idx);    \
>> +}                                            \
>> +                                             \
>> +static void ns16550_generic_##pfx##_write(unsigned int val, unsigned long base, \
>> +             unsigned char reg_idx)                  \
>> +{                                                    \
>> +     cmdw((mask val), (char *)base + reg_idx);       \
>> +}
>> +
>> +NS16550_READ_WRITE_UART_FUNC(8bit, readb, writeb, 0xff &)
>> +NS16550_READ_WRITE_UART_FUNC(16bit, readw, writew, 0xffff &)
>> +NS16550_READ_WRITE_UART_FUNC(32bit, readl, writel, )
>
> Please don't do such preprocessor tricks if not necessary. we can afford
> the 8 additional lines for making this readable.

approx. > 16 additional lines

>> +     if (plat->reg_read == NULL) {
>> +             int width = dev->resource[0].flags & IORESOURCE_MEM_TYPE_MASK;
>> +
>> +             switch (width) {
...
>> +     }
>> +
>> +     if (plat->reg_write == NULL) {
>> +             int width = dev->resource[0].flags & IORESOURCE_MEM_TYPE_MASK;
>> +
>> +             switch (width) {
...
>> +             }
>> +     }
>
> Passing one of the register access function without the other shouldn't
> be a valid usecase. Jean is right, there should be only one if().

Agree.

> Generally platform_data shouldn't be used after device probe and for
> sure it shouldn't be modified. This may be the point where this driver
> needs a private data struct.

I like this point of view.
We must add private data struct and copy plat->width, plat->reg_read etc there.

-- 
Best regards,
  Antony Pavlov