Re: [Uclinux-dist-devel] [PATCH v3] add the driver for Analog Devices Blackfin on-chip CAN controllers

[Wolfgang Grandegger <wg-5Yr1BZd7O62+XT7JhA+gdA@public.gmane.org>]

Mike Frysinger wrote:
> On Thu, Dec 10, 2009 at 05:20, Wolfgang Grandegger wrote:
>> Mike Frysinger wrote:
>>> On Thu, Dec 10, 2009 at 04:11, Wolfgang Grandegger wrote:
>>>> Barry Song wrote:
>>>>> +#include <linux/module.h>
>>>>> +#include <linux/init.h>
>>>>> +#include <linux/kernel.h>
>>>>> +#include <linux/types.h>
>>>> I think you don't need "types.h" as the code no longer uses "uint*_t".
>>> linux/types.h declares all types, like u* which this driver still uses
>> I just remember that "linux/types.h" needs to be added for the uint*_t
>> types. At a first glance I do not see __u8/u8 being defined in that
>> header file but I might have missed something.
> 
> you need to follow the include paths

I thought I did. Could you point me to the relevant location?

>>>> Well, I'm still not a friend of the following inline functions,
>>>> especially the *one-liners* which are called just *once*. With the usage
>>>> of structs they seem even more useless.
>>> seems like it would make more sense to not even use the read/write
>>> functions either.  just declare the regs as volatile and assign/read
>>> the struct directly.
>> Two times no. Don't use volatile and proper accessor functions. See:
>>
>> http://lxr.linux.no/#linux+v2.6.32/Documentation/volatile-considered-harmful.txt
> 
> too bad the document is largely irrelevant (all but one paragraph)
> because this is how volatiles were designed in the first place --
> hardware I/O registers.  the CAN implementation here is Blackfin
> specific and not going to be use elsewhere, so other architectures are
> irrelevant.  the resulting C code would certainly look a hell of a lot
> more natural without the useless I/O accessor functions, and be much
> tighter.

Well, so far *no* volatiles have been used in the BFIN CAN driver. But
if you tell me that they are really required for blackfin... I can't
really judge.

> at any rate, the common Blackfin I/O accessor functions force a lot of
> useless overhead when used here as they're designed for async memory,
> not MMRs.  the driver needs to be switched to the bfin_read/bfin_write
> MMR functions.

I just brought up this issue in another mail.

Wolfgang.

> 
>