[dpdk-dev] rte_eal_init() alternative?

[Panu Matilainen]

On 10/08/2015 05:58 PM, Montorsi, Francesco wrote:
> Hi,
>
>> -----Original Message-----
>> From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
>> Sent: mercoledì 2 settembre 2015 15:10
>> To: Montorsi, Francesco <fmontorsi at empirix.com>
>> Cc: dev at dpdk.org; Bruce Richardson <bruce.richardson at intel.com>
>> Subject: Re: [dpdk-dev] rte_eal_init() alternative?
>>
>> 2015-09-02 13:56, Bruce Richardson:
>>> On Wed, Sep 02, 2015 at 12:49:40PM +0000, Montorsi, Francesco wrote:
>>>> Hi all,
>>>>
>>>> Currently it seems that the only way to initialize EAL is using rte_eal_init()
>> function, correct?
>>>>
>>>> I have the problem that rte_eal_init() will call rte_panic() whenever
>> something fails to initialize or in other cases it will call exit().
>>>> In my application, I would rather like to attempt DPDK initialization. If it
>> fails I don't want to exit.
>>>> Unfortunately I cannot even copy&paste the rte_eal_init() code into my
>> application (removing rte_panic and exit calls) since it uses a lot of DPDK
>> internal private functions.
>>>>
>>>> I think that my requirements (avoid abort/exit calls when init fails) is a
>> basic requirement... would you accept a patch that adds an alternative
>> rte_eal_init() function that just returns an error code upon failure, instead of
>> immediately exiting?
>>>>
>>>> Thanks for your hard work!
>>>>
>>>> Francesco Montorsi
>>>>
>>> I, for one, would welcome such a patch. I think the code is overly
>>> quick in many places to panic or exit the app, when an error code would be
>> more appropriate.
>>> Feel free to also look at other libraries in DPDK too, if you like :-)
>>
>> Yes but please, do not create an alternative init function.
>> We just need to replace panic/exit with error codes and be sure that apps
>> and examples handle them correctly.
>
> To maintain compatibility with existing applications I think that
> perhaps the best would be to have a core initialization function
> rte_eal_init_raw() that never calls rte_panic() and returns an error
> code. Then we can maintain compatibility having an rte_eal_init()
> function that does call rte_panic() if rte_eal_init_raw() fails.

Note that callers are already required to check rte_eal_init() return 
code for errors, and any app failing to do so would be buggy to begin 
with. So just turning the panics into error returns is not an 
incompatible change.

I agree with Thomas here, lets just fix rte_eal_init() to do the right 
thing instead of adding alternatives just for the error return. 
Especially when _raw() in the name suggests that is not the thing you'd 
commonly want to use.

> Something like the attached patch.

It seems the patch missed the boat :)

> Note that the attached patch exposes also a way to skip the
> argv/argc configuration process by directly providing a populated
> configuration structure...
> Let me know what you think about it (the patch is just a draft and
> needs more work).

Can't comment on what I've not seen, but based on comments seen on this 
list, having an alternative way to initialize with structures would be 
welcomed by many. The downside is that those structures will need to be 
exposed in the API forever which means any changes there are subject to 
the ABI process.

	- Panu -


> Thanks,
> Francesco
>
>
>
> 	
>