Re: Using algo-bit in another i2c algo

[Jean Delvare <khali-PUYAD+kWke1g9hUCZPvPmw@public.gmane.org>]

On Wed, 10 Mar 2010 14:16:18 -0500, Alex Deucher wrote:
> I tested the second patch with a trivial patch to the radeon drm:
> 
> --- a/drivers/gpu/drm/radeon/radeon_i2c.c
> +++ b/drivers/gpu/drm/radeon/radeon_i2c.c
> @@ -892,9 +892,9 @@ struct radeon_i2c_chan *radeon_i2c_create(struct
> drm_device *dev,
>          * make this, 2 jiffies is a lot more reliable */
>         i2c->algo.radeon.bit_data.timeout = 2;
>         i2c->algo.radeon.bit_data.data = i2c;
> -       ret = i2c_bit_add_bus(&i2c->algo.radeon.bit_adapter);
> +       ret = i2c_bit_prepare_bus(&i2c->algo.radeon.bit_adapter);
>         if (ret) {
> -               DRM_ERROR("Failed to register internal bit i2c %s\n", name);
> +               DRM_ERROR("Failed to prepare internal bit i2c %s\n", name);
>                 goto out_free;
>         }
>         /* set the radeon i2c adapter */
> 
> However, just calling i2c_bit_prepare_bus does not appear to be
> enough. as I get the following oops:
> (...)

Hmm, sorry, I have probably not been clear about the intent of
i2c_bit_prepare_bus(). It does _not_ give you a usable i2c_adapter. The
actual initialization happens in i2c_register_adapter(), in i2c-core.
What it gives you is a usable i2c _algorithm_. That is, you can call
bit_adapter.algo->master_xfer(), presumably from within your actual
(registered) adapter's master_xfer() function.

Now, I have to admit that it is pretty confusing that you call a
"prepare" function on an i2c_adapter and you can't use it. And looking
a bit more into it... Even calling bit_adapter.algo->master_xfer()
assumes an initialized i2c_adapter, at least for error messages. So my
proposal was probably not such a good idea, at least not with the
current i2c-algo-bit implementation, and probably not without a big
rework of the i2c algorithm structure itself: at the moment, i2c
algorithms really assume that they have an i2c_adapter associated _and_
that this i2c_adapter has a device associated.

Maybe we could move the core of i2c-algo-bit to a library so that it is
easier to reuse. But that would be a lot more work, so we will only do
this there is a clear benefit. At this point, it seems cheaper to
pursue the first option (pre- and post-xfer hooks) if that works for
you.

-- 
Jean Delvare