Re: [PATCH RFC] i2c: Add a void pointer to i2c_device_id

[Uwe Kleine-König]

Hello Andy,

On Mon, Apr 29, 2024 at 01:28:32PM +0300, Andy Shevchenko wrote:
> On Mon, Apr 29, 2024 at 12:21:05PM +0200, Uwe Kleine-König wrote:
> > On Mon, Apr 29, 2024 at 11:54:29AM +0300, Andy Shevchenko wrote:
> > > On Fri, Apr 26, 2024 at 11:38:33PM +0200, Uwe Kleine-König wrote:
> 
> ...
> 
> > > >  static const struct i2c_device_id wlf_gf_module_id[] = {
> > > > -	{ "wlf-gf-module", 0 },
> > > > +	{ "wlf-gf-module", },
> > > 
> > > In such cases the inner comma is redundant as well.
> > 
> > I would tend to keep the comma, but no strong opinion on my side.
> 
> It's just a confusing leftover in my opinion.
> 
> > If another member init is added later, the line has to be touched
> > anyhow, but in the layout:
> > 
> > 	... = {
> > 		{
> > 			"wlf-gf-module",
> > 		},
> > 		{ }
> > 	}
> > 
> > I'd keep it for sure.
> 
> That's not what I object. Here I am 100% with you.

OK, agreed. I'm not sure yet if I prefer

	static const struct i2c_device_id wlf_gf_module_id[] = {
		{ "wlf-gf-module" },
		{ }
	};

or

	static const struct i2c_device_id wlf_gf_module_id[] = {
		{ .name = "wlf-gf-module" },
		{ }
	};

> > > In general idea might be okay, but I always have the same Q (do we have it
> > > being clarified in the documentation, btw): is an ID table the ABI or not?
> > > In another word, how should we treat the changes there, because ID tables
> > > are being used by the user space tools.
> > 
> > Note that the layout doesn't change and the traditional interpretation
> > of the data still works fine. Or do you see something that I miss?
> 
> Do we have any configurations / architectures / etc when
> sizeof(kernel_ulong_t) != sizeof(void *) ? If not, we are fine.

According to https://wiki.debian.org/ArchitectureSpecificsMemo (my goto
address for such questions) we have sizeof(void *) == sizeof(long) on
all archs. Also storing a pointer in today's struct
i2c_device_id::driver_data is so common that it should be safe to assume
that sizeof(void *) <= sizeof(kernel_ulong_t). And that <= is enough
that the union doesn't get bigger.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | https://www.pengutronix.de/ |

Attachment: signature.asc
Description: PGP signature