[U-Boot] [PATCH v4 09/15] dm: Cast away the const-ness of the global_data pointer

[Simon Glass]

Hi Marek,

On 17 June 2014 23:36, Simon Glass <sjg at chromium.org> wrote:
> Hi Marek,
>
> On 17 June 2014 16:42, Marek Vasut <marex at denx.de> wrote:
>> On Thursday, June 12, 2014 at 05:26:50 AM, Simon Glass wrote:
>>> Hi Marek,
>>>
>>> On 8 June 2014 01:00, Marek Vasut <marex at denx.de> wrote:
>>> > On Friday, June 06, 2014 at 09:13:26 PM, Simon Glass wrote:
>>> >> In a very few cases we need to adjust the driver model root device, such
>>> >> as when setting it up at initialisation. Add a macro to make this
>>> >> easier.
>>> >>
>>> >> Signed-off-by: Simon Glass <sjg at chromium.org>
>>> >> ---
>>> >
>>> > [...]
>>> >
>>> >> -     ret = device_bind_by_name(NULL, &root_info, &gd->dm_root);
>>> >> +     ret = device_bind_by_name(NULL, &root_info, &DM_ROOT());
>>> >
>>> > [...]
>>> >
>>> >> +/* Cast away any volatile pointer */
>>> >> +#define DM_ROOT()            (((gd_t *)gd)->dm_root)
>>> >> +#define DM_UCLASS_ROOT()             (((gd_t *)gd)->uclass_root)
>>> >
>>> > Can you implement this "DM_ROOT()" macro as a function instead ? I
>>> > believe that'd be much nicer , would have typechecking etc., usual
>>> > stuff.
>>>
>>> I had a look at this. I don't see how I can do it, but you might have
>>> ideas.
>>>
>>> In this function call I need to pass a pointer without its const bit.
>>> I made it const since nothing should change the dm root except driver
>>> model itself.
>>
>> All right, I will just be doing a little guesswork here. Shall you not have some
>> functions to change the DM root then instead of exposing the DM root ? Such
>> functions shall not be exported publicly , but available via separate header
>> file too ?
>
> Firstly I should retitle the commit - it is of course the volatile
> that I'm trying to get rid of as it says in the commit message. I
> might have just been trying to confuse everyone.
>
> I have code like this:
>
> INIT_LIST_HEAD(&gd->uclass_root);
>
> I don't see how I can turn this into a function. I suppose I could
> init the list head in a temporary variable and then assign it to
> gd->uclass_root, perhaps with memcpy(), but that doesn't really seem
> any better to me.
>
>>
>> Sorry, I might be completely wrong.
>
> Well I'm not saying it's pretty, so if you have time and come up with
> a better option please let me know.
>
> Also, this is not exported 'publicly'. It is in device-internal.h

Please let me know on this one. If you have a better option please let me know.

Otherwise I'd like to do a dm pull request soon to get these fixes
into the release.

Regards,
Simon