[PATCHv5 0/3] Introduce the /proc/socinfo and use it to export OMAP data

[Saravana Kannan]

On 03/01/2011 07:35 PM, Ryan Mallon wrote:
> On 03/02/2011 04:21 PM, Saravana Kannan wrote:
>> On 03/01/2011 07:11 PM, Ryan Mallon wrote:
>>> On 03/02/2011 03:55 PM, Saravana Kannan wrote:
>>>> On 03/01/2011 06:41 PM, Ryan Mallon wrote:
>>>>> On 03/02/2011 03:23 PM, Saravana Kannan wrote:
>>>>>> I don't have any attachment to the "arch" file suggestion. If there
>>>>>> is a
>>>>>> better solution to identify the different implementations of socinfo
>>>>>> without having to maintain some "unique id" list in the kernel,
>>>>>> then I'm
>>>>>> all for it. But cpuinfo is not it.
>>>>>
>>>>> Sorry I am confusing the 'arch' and 'mach' bits here. I definitely have
>>>>> an objection to having an 'arch' file (i.e. ARM). A 'mach' (i.e. omap)
>>>>> file makes a bit more sense, but should probably be called 'mach'
>>>>> rather
>>>>> than 'arch' to avoid this confusion :-).
>>>>
>>>> Sorry for the confusion. Sure, I don't care much for the filename as
>>>> long as we can all agree on it. I care more about the content of the
>>>> file (using names very close to xxxx in mach-xxxx). I like "soc-family"
>>>> better since it's generic enough to not force, say omap3 and omap4, to
>>>> report different values.
>>>>
>>>> Linus Walleij, Eduardo, Maxime, Andrei,
>>>>
>>>> Would like to hear your opinion on the file name (soc-family vs. mach vs
>>>> <somethingelse>) and the path /sys/devices/system/soc/.
>>>
>>> 'family' sounds good. I don't think we need the 'soc-' prefix on
>>> filenames if they are already in /sys/devices/system/soc/.
>>
>> Makes sense. We can drop the soc- prefix. So the contenders left: family
>> vs<somethingelse>. Would still be nice if the other folks chime in.
>>
>>>> If we settle on this, may be it would be easier to get this through.
>>>>
>>>>> I still think it is a solution in search of a problem though. What
>>>>> userspace programs need to know what specific SoC they are on? My
>>>>> feeling is that if userspace needs to know this information, then it is
>>>>> probably dicking around with things that should be managed by the
>>>>> kernel. Differences in available peripherals, etc can be determined by
>>>>> looking at existing sysfs files.
>>>>
>>>> I certainly have seen several use cases. Couple of easy examples:
>>>>
>>>> * A lot of test scripts would find this very useful. For example, some
>>>> clock (present is all/most MSMs) shouldn't be tested on some SOCs as it
>>>> would lock up the system if you try to turn it off while the CPU is
>>>> running.
>>>
>>> I don't follow here. Do you mean a struct clk clock or something else?
>>> Why is userspace allowed to disable a clock which will effectively hang
>>> the system? :-).
>>
>> Ah, sorry. Didn't give enough details. To give some context, I manage
>> the clock stuff for MSM. The MSM clock driver exports clock control thru
>> debugfs. We have test scripts that bang the clocks to test them. Each
>> SoC has a different set of "touch me and you die" clocks that the test
>> script shouldn't mess with. This socinfo would be useful for those test
>> cases.
>
> Ah, okay. This is still within a single SoC family though since we don't
> yet (AFAIK) support mutliple SoCs in a single kernel.

Yes, my example was within a single SoC family. But since this is user 
space example, it doesn't matter if a single kernel can support multiple 
SoCs/SoC families. We could still have one userspace code that might 
want to support multiple SoC families.

Anyway, I think we are in agreement here. So, will stop discussing this 
point.

>>>> * Some of the user space tools might want to report different "product
>>>> id/type" (nothing to do with USB, etc) depending on what SOC it is
>>>> running on.
>>>
>>> This makes more sense. It would actually be useful for custom USB
>>> devices (gadget) which can be done from user space.
>>
>> Hmm... didn't know USB devices/gadgets could be handled from userspace.
>
> The gadgetfs driver allows for writing custom usb device
> implementations. The SoC info could be used to set the USB
> vendor/product id. Again, I see this more useful within the SoC family
> (ie at91sam9260 vs at91sam9263) rather than between families. From an
> embedded perspective at least, I think it is unlikely for an application
> to need to work on multiple SoC families.
>
> The only real objection I have to adding the SoC family information is
> basically to discourage it being abused by userspace. I can see it being
> useful in debug situations, but I can also see stupid userspace
> applications explicitly testing for some particular SoC, rather than
> more correctly (IMHO) checking for presence of certain drivers etc.

True, but so many other things could be misused by stupid userspace 
programs. When there are legitimate usecases, I think we shouldn't 
prevent them just because we think a stupid userspace program could 
misuse it.

Again, although you might not be gung-ho about this, I think I have at 
least made you indifferent/mildly supportive to adding socinfo. If you 
don't mind, I would like to wait for others to chime in before 
continuing this discussion.

-Saravana
-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.