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

[Antony Pavlov]

On 4 August 2011 04:42, Jean-Christophe PLAGNIOL-VILLARD
<plagnioj at jcrosoft.com> wrote:
> On 22:37 Wed 03 Aug     , Antony Pavlov wrote:
>> Signed-off-by: Antony Pavlov <antonynpavlov at gmail.com>
>> ---
>>  drivers/serial/serial_ns16550.c |   87 ++++++++++++++++++++++++---------------
>>  1 files changed, 54 insertions(+), 33 deletions(-)
>>

>> +     plat->reg_write(val, (unsigned long)dev->priv, off);
> no it must stay void __iomem*

plat->reg_write() get the second argument 'unsigned long base'.

Moreover, using 'void *' in arithmetic is a poor practice.

Fortunally for us, in GCC incrementing a void pointer adds one to the
value (sizeof(void) == 1).

Add '-pedantic' flag to gcc and compile current 'next' barebox.

You will see something like this:
In function ‘ns16550_read’:
drivers/serial/serial_ns16550.c:72: warning: pointer of type ‘void *’
used in arithmetic

>> +     if (plat->reg_read == NULL) {
>> +             int width = dev->resource[0].flags & IORESOURCE_MEM_TYPE_MASK;
>> +
>> +             switch (width) {
>> +             default:
...
>> +             }
>> +     }
>> +
>> +     if (plat->reg_write == NULL) {
>> +             int width = dev->resource[0].flags & IORESOURCE_MEM_TYPE_MASK;
> why do it twice?

You are right.
width calculation or even separate reg_write & reg_read handling  can
be combined.

-- 
Best regards,
  Antony Pavlov