[PATCH 0/2] crypto: add new driver for Marvell CESA

[gregory.clement@free-electrons.com (Gregory CLEMENT)]

On 17/04/2015 17:49, Maxime Ripard wrote:
> On Fri, Apr 17, 2015 at 05:01:55PM +0200, Gregory CLEMENT wrote:
>> On 17/04/2015 16:50, Maxime Ripard wrote:
>>> On Fri, Apr 17, 2015 at 04:40:43PM +0200, Gregory CLEMENT wrote:
>>>> Hi Maxime,
>>>>
>>>> On 17/04/2015 16:32, Maxime Ripard wrote:
>>>>> On Fri, Apr 17, 2015 at 04:19:22PM +0200, Boris Brezillon wrote:
>>>>>> Hi Gregory,
>>>>>>
>>>>>> On Fri, 17 Apr 2015 15:01:01 +0200
>>>>>> Gregory CLEMENT <gregory.clement@free-electrons.com> wrote:
>>>>>>
>>>>>>> Hi Boris,
>>>>>>>
>>>>>>> On 17/04/2015 10:39, Boris Brezillon wrote:
>>>>>>>> On Fri, 17 Apr 2015 10:33:56 +0200
>>>>>>>> Boris Brezillon <boris.brezillon@free-electrons.com> wrote:
>>>>>>>>
>>>>>>>>> Hi Jason,
>>>>>>>>>
>>>>>>>>> On Mon, 13 Apr 2015 20:11:46 +0000
>>>>>>>>> Jason Cooper <jason@lakedaemon.net> wrote:
>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>> I'd appreciate if we'd look into it.  I understand from on-list and
>>>>>>>>>>>> off-list discussion that the rewrite was unavoidable.  So I'm willing to
>>>>>>>>>>>> concede that.  Giving people time to migrate from old to new while still
>>>>>>>>>>>> being able to update for other security fixes seems reasonable.
>>>>>>>>>>>
>>>>>>>>>>> Jason, what do you think of the approach above? 
>>>>>>>>>>
>>>>>>>>>> I say keep it simple.  We shouldn't use the DT changes to trigger one
>>>>>>>>>> vice the other.  We need to be able to build both, but only load one at
>>>>>>>>>> a time.  If that's anything other than simple to do, then we make it a
>>>>>>>>>> Kconfig binary choice and move on.
>>>>>>>>>
>>>>>>>>> Actually I was planning to handle it with a Kconfig dependency rule
>>>>>>>>> (NEW_DRIVER depends on !OLD_DRIVER and OLD_DRIVER depends
>>>>>>>>> on !NEW_DRIVER).
>>>>>>>>> I don't know how to make it a runtime check without adding new
>>>>>>>>> compatible strings for the kirkwood, dove and orion platforms, and I'm
>>>>>>>>> sure sure this is a good idea.
>>>>>>>>   ^ not
>>>>>>>>
>>>>>>>>> Do you have any ideas ?
>>>>>>>
>>>>>>> You use devm_ioremap_resource() in the new driver, so if the old one
>>>>>>> is already loaded the memory region will be already hold and the new
>>>>>>> driver will simply fail during the probe. So for this part it is OK.
>>>>>>
>>>>>> I like the idea :-).
>>>>>
>>>>> Not really, how do you know which device is going to be probed? For
>>>>> that matter, it's pretty much random, and you have no control over it.
>>>>>
>>>>> Why not just have a choice option, and select which one you want to
>>>>> enable?
>>>>
>>>> Because you can't prevent an user to build a module, then modifying the
>>>> configuration and building the other module.
>>>
>>> Well, actually, you don't even know if it's going to be a module. You
>>> might very well have both drivers compiled statically in the kernel
>>> image, and this is where the trouble begins.
>>
>> No it won't be possible, Boris already speak about this issue (see below):
>> "Actually I was planning to handle it with a Kconfig dependency rule
>> (NEW_DRIVER depends on !OLD_DRIVER and OLD_DRIVER depends
>> on !NEW_DRIVER)."
> 
> Which is a circular dependency and won't work.

Indeed.

Boris what about using choice/endchoice ?

Thanks,

Gregory



> 
>>>> So even if there is a choice at build time, and I think that it is
>>>> something expected for the v2, we still need preventing having the
>>>> both drivers trying accessing the same hardware in the same time.
>>>
>>> Of course, but this is already there, and doesn't really address the
>>> same issue.
>>
>> This was the only issue remaining, (see below again):
>> "I don't know how to make it a runtime check ". And my last emails
>> was bout it.
> 
> Ok, my bad then :)
> 
> Maxime
> 


-- 
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com