Re: [PATCH REVIEW] e4000: convert DVB tuner to I2C driver model

[Antti Palosaari <crope@iki.fi>]

On 16.10.2013 20:33, Michael Krufky wrote:
> On Wed, Oct 16, 2013 at 1:22 PM, Antti Palosaari <crope@iki.fi> wrote:
>> On 16.10.2013 20:19, Michael Krufky wrote:
>>>
>>> On Wed, Oct 16, 2013 at 1:09 PM, Jean Delvare <khali@linux-fr.org> wrote:
>>>>
>>>> Hi Michael,
>>>>
>>>> On Wed, 16 Oct 2013 13:04:42 -0400, Michael Krufky wrote:
>>>>>
>>>>> YIKES!!  i2c_new_probed_device() does indeed probe the hardware --
>>>>> this is unacceptable, as such an action can damage the ic.
>>>>>
>>>>> Is there some additional information that I'm missing that lets this
>>>>> perform an attach without probe?
>>>>
>>>>
>>>> Oh, i2c_new_probed_device() probes the device, what a surprise! :D
>>>>
>>>> Try, I don't know, i2c_new_device() maybe if you don't want the
>>>> probe? ;)
>>>>
>>>> --
>>>> Jean Delvare
>>>
>>>
>>> OK, so to confirm that I follow correctly, one can use
>>> i2c_new_device() to attach the sub-driver without probing, and the
>>> line that ensures that the correct sub-driver gets attached is
>>> "strlcpy(info.type, "e4000", I2C_NAME_SIZE);"  ??
>>>
>>> We're matching based on a string?  I think that's kinda yucky, but if
>>> that's what we're doing in i2c nowadays then I'm OK with it.
>>>
>>> If not, what prevents the wrong sub-driver from attaching to a device?
>>>    ...or conversely, how does the right sub-driver know which device to
>>> attach to?
>>
>>
>> Yes, it is that string. Driver has that string as a ID table entry. Then you
>> issue i2c_new_device() call with string and it attachs driver when strings
>> match.
>>
>>
>>> Again, if I'm asking "stupid questions" just point me to the
>>> documentation.
>>>
>>> -Mike
>>>
>>
>> regards
>> Antti
>
>
> OK, I get it and it does seem OK.  I'm just curious what kind of
> impact this refactoring would have over something like the
> b2c2-flexcop-fe driver, who does not know which ic's to attach based
> on device ids, but it does probe a few frontend combinations one after
> another, in an order that the driver authors knew was safe.  I'd
> imaging that we'd write some helper abstraction function to switch out
> the info.type string as each driver gets probed?  I think that can get
> quite ugly, but I know that the general population thinks dvb_attach()
> is even uglier, so maybe this could be the right path...
>
> Wanna take a crack at b2c2-flexcop-fe?
>
> -Mike
>

heh, look the rtl28xxu driver in question, it does same. It probes maybe 
total 10 tuners - checking their device ID too. Probing plain I2C device 
address and making devision from that is simply dead idea. "Standard" 
I2C address range for these silicon tuners are 0x60-0x64 => need to read 
chip ID in order to detect => i2c_new_probed_device() is quite useless.

regards
Antti

-- 
http://palosaari.fi/