On 4/20/22 16:36, Uwe Kleine-König wrote:
> Hello Ian,
>
> On Wed, Apr 20, 2022 at 09:57:27AM -0400, Ian Cowan wrote:
>> On Wed, Apr 20, 2022 at 08:47:11AM +0200, Uwe Kleine-König wrote:
>>> On Tue, Apr 19, 2022 at 03:21:28PM -0400, Ian Cowan wrote:
>>>> Removed an unnecessary semicolon at the end of a macro call
>>>>
>>>> Signed-off-by: Ian Cowan <ian@xxxxxxxxxxxxxxxx>
>>>> ---
>>>> drivers/staging/fbtft/fbtft.h | 2 +-
>>>> 1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/staging/fbtft/fbtft.h b/drivers/staging/fbtft/fbtft.h
>>>> index 2c2b5f1c1df3..aa66760e1a9c 100644
>>>> --- a/drivers/staging/fbtft/fbtft.h
>>>> +++ b/drivers/staging/fbtft/fbtft.h
>>>> @@ -277,7 +277,7 @@ static const struct of_device_id dt_ids[] = { \
>>>> { .compatible = _compatible }, \
>>>> {}, \
>>>> }; \
>>>> -MODULE_DEVICE_TABLE(of, dt_ids);
>>>> +MODULE_DEVICE_TABLE(of, dt_ids)
>>>
>>> In fact the ; after MODULE_DEVICE_TABLE is necessary. There is only a
>>> single instance in the kernel without a semicolon[1]. That's in
>>> drivers/pci/controller/pcie-microchip-host.c and this only works because
>>> this driver cannot be compiled as a module and so MODULE_DEVICE_TABLE
>>> evaluates to nothing. Will send a patch for that one.
>
Indeed. I was curious about this so I went to look at the driver code.
For this particular driver it may be not necessary, but that's just due
how these fbtft drivers define their MODULE_DEVICE_TABLE(), using a lot
of macro layers.
As an example, drivers/staging/fbtft/fb_agm1264k-fl.c does the following:
FBTFT_REGISTER_DRIVER(DRVNAME, "displaytronic,fb_agm1264k-fl", &display);
which is defined as:
#define FBTFT_REGISTER_DRIVER(_name, _compatible, _display) \
... \
FBTFT_DT_TABLE(_compatible) \
...
which in turn is defined as:
#define FBTFT_DT_TABLE(_compatible) \
static const struct of_device_id dt_ids[] = { \
{ .compatible = _compatible }, \
{}, \
}; \
MODULE_DEVICE_TABLE(of, dt_ids);
so it seems that it builds, just because the semicolon for the expression
is the one that's after the FBTFT_REGISTER_DRIVER(); in the driver.
> FTR: Patch was sent: https://lore.kernel.org/linux-pci/20220420065832.14173-1-u.kleine-koenig@xxxxxxxxxxxxxx
>
>> When I built this, it appeared to succeed. I used the command "make
>> M=/drivers/staging/fbtft modules". Is this incorrect? For reference this
>> is my first patch so it's highly likely I did this incorrectly.
>
You are just changing a header file though, did you also enable one of the
fbtft drivers as a module to see if those build? But as said, by looking at
the code it seems that should build correctly.
I agree with Uwe though that is less confusing to have a semicolon after
the MODULE_DEVICE_TABLE(), but I'm not the driver maintainer to decide.
> I don't know for sure, but I'd have said that the M= stuff is for
> out-of-tree modules only.
>
It does work, I use M= to build drivers in mainline that are configured
to build as a module all the time.
--
Best regards,
Javier Martinez Canillas
Linux Engineering
Red Hat
