Re: HW communication debugging interface - ideas?

[John Fastabend <john.fastabend@gmail.com>]

On 15-10-06 01:14 AM, Jiri Pirko wrote:
> Mon, Oct 05, 2015 at 05:47:09PM CEST, john.fastabend@gmail.com wrote:
>> On 15-10-05 08:35 AM, Jiri Pirko wrote:
>>> Mon, Oct 05, 2015 at 05:29:09PM CEST, john.fastabend@gmail.com wrote:
>>>> On 15-10-05 08:18 AM, Jiri Pirko wrote:
>>>>> Mon, Oct 05, 2015 at 04:58:42PM CEST, andrew@lunn.ch wrote:
>>>>>> On Mon, Oct 05, 2015 at 04:55:42PM +0200, Jiri Pirko wrote:
>>>>>>> Mon, Oct 05, 2015 at 04:49:41PM CEST, andrew@lunn.ch wrote:
>>>>>>>>>> Are you referring here to messages of the EMAD protocol ?
>>>>>>>>
>>>>>>>> I know nothing about this protocol.....
>>>>>>>>
>>>>>>>> Does it at least use standard Ethernet framing? Source and Destination
>>>>>>>> header and an EtherType which mean EMAD?
>>>>>>>
>>>>>>> Yep, but that does not really matter. I believe we should find debugging
>>>>>>> interface which is protocol agnostic. Just arbitrary messages
>>>>>>> monitoring.
>>>>>>
>>>>>> Hi Jiri
>>>>>>
>>>>>> O.K, it is just that you mentioned wireshark. Passing the frames to
>>>>>> network interface taps would make this trivial.
>>>>>
>>>>> That is true. But using netlink+nlmon would do the same.
>>>>
>>>> Also I guess if you go this direction you want to make it generic
>>>> enough for any drivers to use it to snoop software/firmware msgs. This
>>>> is common across many devices.
>>>
>>> Yes, definitelly, this should be something generic to be usable for
>>> every device type.
>>>
>>>
>>>>
>>>> In the past though I've just used ethtool dump commands and some
>>>> "scripts" on top of this to debug devices. And when it got really
>>>> bad wrote some throw away code to debug my issue. I guess it might
>>>> be nice to have something in the kernel to improve this but have
>>>> you considered using the tracing features that already exist?
>>>
>>> Which ones do you have in mind?
>>>
>>
>> I was thinking something like kprobes+bpf to dump a trace and
>> then a lua script in wireshark to parse the input and pretty
>> print it for users. This might get you good-enough support without
>> having to carry it around in the kernel just so we can debug
>> the devices. We could build some libs/pkgs around it in userspace
>> and get it published somewhere so we can all work on it together.
> 
> Well, I was thinking rather about some standard interface, not dependent
> on actual kernel internals.
> 


Sure just throwing out an idea. I suspect whatever interface you have
will include the vendor-id or some other identifier and a set of
parsers in user space to pretty print the msg.

.John