[dpdk-dev] [PATCH] eal: register rte_panic user callback
Burakov, Anatoly
anatoly.burakov at intel.com
Wed Mar 7 12:29:47 CET 2018
On 07-Mar-18 9:59 AM, Thomas Monjalon wrote:
> 07/03/2018 10:05, Burakov, Anatoly:
>> On 07-Mar-18 8:32 AM, Thomas Monjalon wrote:
>>> Hi,
>>>
>>> 06/03/2018 19:28, Arnon Warshavsky:
>>>> The use case addressed here is dpdk environment init
>>>> aborting the process due to panic,
>>>> preventing the calling process from running its own tear-down actions.
>>>
>>> Thank you for working on this long standing issue.
>>>
>>>> A preferred, though ABI breaking solution would be
>>>> to have the environment init always return a value
>>>> rather than abort upon distress.
>>>
>>> Yes, it is the preferred solution.
>>> We should not use exit (panic & co) inside a library.
>>> It is important enough to break the API.
>>
>> +1, panic exists mostly for historical reasons AFAIK. it's a pity i
>> didn't think of it at the time of submitting the memory hotplug RFC,
>> because i now hit the same issue with the v1 - we might panic while
>> holding a lock, and didn't realize that it was an API break to change
>> this behavior.
>>
>> Can this really go into current release without deprecation notices?
>
> If such an exception is done, it must be approved by the technical board.
> We need to check few criterias:
> - which functions need to be changed
> - how the application is impacted
> - what is the urgency
>
> If a panic is removed and the application is not already checking some
> error code, the execution will continue without considering the error.
>
> Some rte_panic could be probably removed without any impact on applications.
> Some rte_panic could wait for 18.08 with a notice in 18.05.
> If some rte_panic cannot wait, it must be discussed specifically.
>
Can we add a compile warning for adding new rte_panic's into code? It's
a nice tool while debugging, but it probably shouldn't be in any new
production code.
--
Thanks,
Anatoly
More information about the dev
mailing list