Re: [PATCH] of: Custom printk format specifier for device node

[Pantelis Antoniou <pantelis.antoniou@konsulko.com>]

Hi Grant,

> On Mar 31, 2015, at 20:02 , Grant Likely <grant.likely@secretlab.ca> wrote:
> 
> Hi Pantelis,
> 
> Thanks for the quick reply. Comments below...
> 
> On Tue, 31 Mar 2015 13:03:05 +0300
> , Pantelis Antoniou <pantelis.antoniou@konsulko.com>
> wrote:
>> Hi Grant,
>> 
>>> On Mar 30, 2015, at 22:04 , Grant Likely <grant.likely@secretlab.ca> wrote:
>>> 
>>> On Thu, 22 Jan 2015 22:31:46 +0200
>>> , Pantelis Antoniou <panto@antoniou-consulting.com>
>>> wrote:
>>>> Hi Joe,
>>>> 
>>>>> On Jan 21, 2015, at 19:37 , Joe Perches <joe@perches.com> wrote:
>>>>> 
>>>>> On Wed, 2015-01-21 at 19:06 +0200, Pantelis Antoniou wrote:
>>>>>> 90% of the usage of device node's full_name is printing it out
>>>>>> in a kernel message. Preparing for the eventual delayed allocation
>>>>>> introduce a custom printk format specifier that is both more
>>>>>> compact and more pleasant to the eye.
>>>>>> 
>>>>>> For instance typical use is:
>>>>>> 	pr_info("Frobbing node %s\n", node->full_name);
>>>>>> 
>>>>>> Which can be written now as:
>>>>>> 	pr_info("Frobbing node %pO\n", node);
>>>> 
>>>>> Still disliking use of %p0.
>>>>> 
>>>> 
>>>> pO - Open Firmware
>>>> 
>>>> pT for tree is bad, cause we plan to use a tree type in the future in OF.
>>> 
>>> So, here's a radical thought. How about we reserve '%pO' for objects, as
>>> in kobjects.  We'll use extra flags to narrow down specifically to
>>> device tree nodes, but we could teach vsprintf() to treat a plain '%pO'
>>> as plain kobject pointer, and if it is able to recognize the kobj_type,
>>> then run a specific decoder to format it.
>>> 
>>> This also gives us a namespace for various kinds of firmware data
>>> output. %Od could be a struct device, %On for device tree node, %Oa for
>>> an ACPI node, etc.
>>> 
>> 
>> I’m fine with this. I also have a patch to turn an overlay to a kobj
>> so this fits naturally.
>> 
>> OTOH if we do this, I would expect to rework the custom printk infrastructure
>> to be more generic.
>> 
>> IMHO having the format specifier and the format print methods in lib/vsprintf.c
>> is not very nice.
>> 
>> How about having a way to register object printk handlers with something like that?
>> We could put that in a special linker section and have the printk method pass control
>> there.
>> 
>> PRINTK_OBJFMT(’n’, printk_objfmt_device_node);
>> 
>> We might have to make a few printk methods public however.
> 
> Honestly, I think trying to add registration is an overengineered
> solution at this point. We're not hitting a wall on the complexity of
> vsprintf.c, and having them all in one place helps to ensure we don't
> have conflicts.
> 
>> 
>>> I've dropped the refcount decoder. I know it is useful for debugging the
>>> core DT code, but it isn't something that will be used generally. Plus
>>> the returned value cannot be relied upon to be stable because there may
>>> be other code currently iterating over the tree.
>>> 
>> 
>> Yeah, I know it’s not something to rely on. If we do %pOk to be kobj
>> debug I can add it back in.
> 
> Yes, that would be a good place to have refcount output.
> 
>>> +Device tree nodes:
>>> +
>>> +	%pOn[fnpPcCFr]
>>> +
>>> +	For printing device tree nodes. The optional arguments are:
>>> +	    f device node full_name
>>> +	    n device node name
>>> +	    p device node phandle
>>> +	    P device node path spec (name + @unit)
>>> +	    F device node flags
>>> +	    c major compatible string
>>> +	    C full compatible string
>>> +	Without any arguments prints full_name (same as %pOnf)
>>> +	The separator when using multiple arguments is ‘:’
>> ^ separator is ‘.'
> 
> ? I'm confused? The separator that I'm using is a colon. ':'  Where do
> you see ','? I don't think ',' would be a good separator because it
> appears in node names and compatible strings. Originally, I think you
> were using pipe '|', but my personal opinion is that ':' is better
> because there is already precidence as a separator.
> 

Ugh, -EJETLAG.

You’re correct, sorry for the confusion.

> g.

Regards

— Pantelis