+CC lkml,Arnd, Rob
On Friday 06 November 2015 12:20 AM, Alexey Brodkin wrote:
> Hi Vineet,
>
> During OpenWRT upsreaming process one interesting topic was raised.
> See in the middle of https://lists.openwrt.org/pipermail/openwrt-devel/2015-November/036959.html
>
> In Device Tree descriptions for our boards we don't use "model" property
> even though it is a required one as specified by ePAPR, see
> http://free-electrons.com/~thomas/pub/Power_ePAPR_APPROVED_v1.0.pdf,
> page 39 "Table 3-1 Root node properties".
>
> Instead we put 2 items in "compatible" property.
>
> For example:
> ------------------->8----------------
> compatible = "snps,axs101", "snps,arc-sdp";
> ------------------->8----------------
>
> And from ePAPR standpoint it makes sense to split contents of that "compatible"
> property in 2:
> ------------------->8----------------
> compatible = "snps,arc-sdp";
> model = "snps,axs101";
> ------------------->8----------------
It seems model is just a descriptive label and we can surely add them to existing DT.
compatible on the other hand is more fundamental used for exact comparisons etc
and follows the vendor,device convention.
It is pretty common for compatible to have multiple strings for exactly the same
reason as I have them here. Both axs101 and axs103 are based on sdp thus we want
the ability to have both pieces of information and use as needed.
While doing some other DT research recently, I found some of the best basic DT
documentation is a somewhat misnamed in-kernel document
Documentation/devicetree/booting-without-of.txt
> But I do see problems with implementation of that thing.
> Today we have a luxury of selection of AXS init functionality
> based on that compatible value and if "snps,axs101" goes in
> "model" then we'll need to add some more code in
> arch/arc/plat-axs10x/axs10x.c that reads "model" value with
> of_get_property() and then compare to "axs10{1|3}".
>
> Any thoughts?
>
> -Alexey
On Friday 06 November 2015 04:45:24 Vineet Gupta wrote:
> >
> > During OpenWRT upsreaming process one interesting topic was raised.
> > See in the middle of https://lists.openwrt.org/pipermail/openwrt-devel/2015-November/036959.html
> >
> > In Device Tree descriptions for our boards we don't use "model" property
> > even though it is a required one as specified by ePAPR, see
> > http://free-electrons.com/~thomas/pub/Power_ePAPR_APPROVED_v1.0.pdf,
> > page 39 "Table 3-1 Root node properties".
> >
> > Instead we put 2 items in "compatible" property.
> >
> > For example:
> > ------------------->8----------------
> > compatible = "snps,axs101", "snps,arc-sdp";
> > ------------------->8----------------
> >
> > And from ePAPR standpoint it makes sense to split contents of that "compatible"
> > property in 2:
> > ------------------->8----------------
> > compatible = "snps,arc-sdp";
> > model = "snps,axs101";
> > ------------------->8----------------
>
> It seems model is just a descriptive label and we can surely add them to existing DT.
> compatible on the other hand is more fundamental used for exact comparisons etc
> and follows the vendor,device convention.
> It is pretty common for compatible to have multiple strings for exactly the same
> reason as I have them here. Both axs101 and axs103 are based on sdp thus we want
> the ability to have both pieces of information and use as needed.
Correct.
The model should also be a human readable name of the machine, just one
string like "Synapsys AXS101 Development Board" (or whatever that is called).
> > But I do see problems with implementation of that thing.
> > Today we have a luxury of selection of AXS init functionality
> > based on that compatible value and if "snps,axs101" goes in
> > "model" then we'll need to add some more code in
> > arch/arc/plat-axs10x/axs10x.c that reads "model" value with
> > of_get_property() and then compare to "axs10{1|3}".
> >
> > Any thoughts?
That should use the of_machine_is_compatible() helper. The model string
is for printing in dmesg or /proc/cpuinfo.
Arnd
On 06.11.2015 09:59, Arnd Bergmann wrote:
> On Friday 06 November 2015 04:45:24 Vineet Gupta wrote:
>>>
>>> During OpenWRT upsreaming process one interesting topic was raised.
>>> See in the middle of https://lists.openwrt.org/pipermail/openwrt-devel/2015-November/036959.html
>>>
>>> In Device Tree descriptions for our boards we don't use "model" property
>>> even though it is a required one as specified by ePAPR, see
>>> http://free-electrons.com/~thomas/pub/Power_ePAPR_APPROVED_v1.0.pdf,
>>> page 39 "Table 3-1 Root node properties".
>>>
>>> Instead we put 2 items in "compatible" property.
>>>
>>> For example:
>>> ------------------->8----------------
>>> compatible = "snps,axs101", "snps,arc-sdp";
>>> ------------------->8----------------
>>>
>>> And from ePAPR standpoint it makes sense to split contents of that "compatible"
>>> property in 2:
>>> ------------------->8----------------
>>> compatible = "snps,arc-sdp";
>>> model = "snps,axs101";
>>> ------------------->8----------------
>>
>> It seems model is just a descriptive label and we can surely add them to existing DT.
>> compatible on the other hand is more fundamental used for exact comparisons etc
>> and follows the vendor,device convention.
>> It is pretty common for compatible to have multiple strings for exactly the same
>> reason as I have them here. Both axs101 and axs103 are based on sdp thus we want
>> the ability to have both pieces of information and use as needed.
>
> Correct.
>
> The model should also be a human readable name of the machine, just one
> string like "Synapsys AXS101 Development Board" (or whatever that is called).
This contradicts ePAPR, which says the model's recommended* format is the same as
the compatible one's (<vendor>,<model>). Most PowerPC and some MIPS dts files
follow that, while ARM(64) uses the free text form.
To me it looks like the intended usage was
model = <actual_model>; compatible = <platform>;
but the actual usage in arm is
model = <human readable string>; compatible = <actual_model>, <platform>;
Of course for changing this in the existing dts files it might be a bit late, but it
would be good to decide which of these two is the actually expected format.
It also is a required property, and we have a few boards not having a model property,
including the example in Documentation/devicetree/usage-model.txt.
Jonas
* compatible strings are also only "recommended" to be in that format.
On Friday 06 November 2015 04:27 PM, Jonas Gorski wrote:
> On 06.11.2015 09:59, Arnd Bergmann wrote:
>> On Friday 06 November 2015 04:45:24 Vineet Gupta wrote:
>>>>
>>>> During OpenWRT upsreaming process one interesting topic was raised.
>>>> See in the middle of https://lists.openwrt.org/pipermail/openwrt-devel/2015-November/036959.html
>>>>
>>>> In Device Tree descriptions for our boards we don't use "model" property
>>>> even though it is a required one as specified by ePAPR, see
>>>> http://free-electrons.com/~thomas/pub/Power_ePAPR_APPROVED_v1.0.pdf,
>>>> page 39 "Table 3-1 Root node properties".
>>>>
>>>> Instead we put 2 items in "compatible" property.
>>>>
>>>> For example:
>>>> ------------------->8----------------
>>>> compatible = "snps,axs101", "snps,arc-sdp";
>>>> ------------------->8----------------
>>>>
>>>> And from ePAPR standpoint it makes sense to split contents of that "compatible"
>>>> property in 2:
>>>> ------------------->8----------------
>>>> compatible = "snps,arc-sdp";
>>>> model = "snps,axs101";
>>>> ------------------->8----------------
>>>
>>> It seems model is just a descriptive label and we can surely add them to existing DT.
>>> compatible on the other hand is more fundamental used for exact comparisons etc
>>> and follows the vendor,device convention.
>>> It is pretty common for compatible to have multiple strings for exactly the same
>>> reason as I have them here. Both axs101 and axs103 are based on sdp thus we want
>>> the ability to have both pieces of information and use as needed.
>>
>> Correct.
>>
>> The model should also be a human readable name of the machine, just one
>> string like "Synapsys AXS101 Development Board" (or whatever that is called).
>
> This contradicts ePAPR, which says the model's recommended* format is the same as
> the compatible one's (<vendor>,<model>). Most PowerPC and some MIPS dts files
> follow that, while ARM(64) uses the free text form.
>
> To me it looks like the intended usage was
> model = <actual_model>; compatible = <platform>;
> but the actual usage in arm is
> model = <human readable string>; compatible = <actual_model>, <platform>;
>
> Of course for changing this in the existing dts files it might be a bit late, but it
> would be good to decide which of these two is the actually expected format.
>
> It also is a required property, and we have a few boards not having a model property,
> including the example in Documentation/devicetree/usage-model.txt.
>
>
> Jonas
>
> * compatible strings are also only "recommended" to be in that format.
>
Alexey, can u please rework the DT files per Jonas' suggestion above.
Thx,
-Vineet