Re: [PATCH v2 05/06] input synaptics-rmi4: Add firmware update support

[Christopher Heiny <cheiny@synaptics.com>]

[I sent this last Thursday, but it never showed up on the input list. 
I'm assuming nobody else saw it.]


On 07/31/2014 02:58 PM, Dmitry Torokhov wrote:
> On Thu, Jul 31, 2014 at 02:43:47PM -0700, Christopher Heiny wrote:
>> On 07/31/2014 02:19 PM, Dmitry Torokhov wrote:
>>> On Thu, Jul 31, 2014 at 02:00:14PM -0700, Christopher Heiny wrote:
>>>>> On 07/31/2014 10:53 AM, Dmitry Torokhov wrote:
>>>>>>> Hi Christopher,
>>>>>>>
>>>>>>> On Wed, Mar 12, 2014 at 06:53:56PM -0700, Christopher Heiny wrote:
>>>>>>>>> Add support for updating firmware on RMI4 devices with V5 bootloader.
>>>>>>>
>>>>>>> I am wondering why F34 is not following the staindard RMI function
>>>>>>> implementation. By that I mean that it does not declare itself as struct
>>>>>>> rmi_function_handler and does not rely on RMI core to bind itself to the device
>>>>>>> if device supports it.
>>>>>
>>>>> Hi Dmitry,
>>>>>
>>>>> We originally had an F34 implementation that followed the RMI4
>>>>> function standard and exposed most of the basic F34 operations via
>>>>> sysfs. However, we got feedback (both on LKML and offline) (a)
>>>>> recommending to use request_firmware, and (b) improve reflash times
>>>>> while (c) reducing impact on boot time, and (d) "get rid of all that
>>>>> sysfs crap" (paraphrased, but close to it).
>>>>>
>>>>> So after looking at how some other drivers use request_firmware, we
>>>>> came up with the current approach.  Switching to request_firmware
>>>>> definitely sped up the reflash times!  Including a check to see if
>>>>> firmware update is required before setting up the RMI4
>>>>> sensor/function structures also significantly reduced boot times.
>>>
>>> I am not suggesting you stop using request-firmware or introduce
>>> bazillion of new sysfs attributes. I just wondered why you have manual
>>> "binding" of F34 functionality instead of standrad RMI4 function
>>> binding, liek you do for F01, F11 and so forth.
>>
>> Sorry!  My answer wasn't very clear on that part, was it?
>>
>> The manual binding gets the reflash (if required) done very early in
>> the boot/probe process.  This eliminates the need to set up the
>> whole sensor + functions structure, tear it down in order to
>> reflash, and then build it all back up again.  It is felt that the
>> time savings is significant, especially on highly featured products.
>
> I am sorry but I have hard time accepting this argument. How often do
> you reflash devices during normal operation and how long does it take to
> initialize the device compared to getting entire userspace up and
> running to be able to actually supply or serve flash data (even without
> using usermode helper to flash you need filesystem with the firmware to
> be mounted)?

That was my argument exactly, but that was the direction we were pushed. 
  I'd much rather implement it as we discussed offline earlier this 
week.  If you were to say: "I'm sorry, but this simply can't be merged 
as it stands." you wouldn't get any argument from me on technical 
grounds. There might be people who will argue about the additional 
calendar time it would take to restructure it, though.