[PATCH] nvme: print disposition on err req completion

[Chaitanya Kulkarni]

On 6/7/2023 10:06 AM, Engel, Amit wrote:
> That’s correct, my original proposal was to have a dedicated tracepoint for
> nvme_end_req/nvme_retry_req/nvme_failover_req.
> 
> But, ck proposed a bit different approach and
> We agreed to go with optionally printing disposition value only
> when disposition != COMPLETE in the trace_nvme_complete_rq()
> 
> Thanks,
> Amit
> -----Original Message-----
> From: Sagi Grimberg <sagi at grimberg.me>
> Sent: Wednesday, 7 June 2023 18:08
> To: Christoph Hellwig <hch at lst.de>; Chaitanya Kulkarni <kch at nvidia.com>
> Cc: linux-nvme at lists.infradead.org; Engel, Amit <Amit.Engel at Dell.com>; kbusch at kernel.org
> Subject: Re: [PATCH] nvme: print disposition on err req completion
> 
> 
> [EXTERNAL EMAIL]
> 
> 
>>> Currently we don't print disposition in the request completion path
>>> in nvme_complete_rq() tracepoint, so when request fails it is not
>>> straight forward to understand the diposition making it hard to debug
>>> the problems in the field.
>>>
>>> Update nvme_complete_rq() tracepoint to print the disposition when
>>> request fails, since disposition value will always be 0 when request
>>> status = 0.
>>
>> Hmm, not sure this is the right way to approach it.  I'd rather redo
>> how the trace points work by having a class and the ondividual
>> tracepoints in nvme_end_req/nvme_retry_req/nvme_failover_req.
> 
> :)
> 
> This was the original proposal (posted as a question to the list), and I proposed this approach. You don't like it?
> 
> I kinda like the explicit disposition output.

I thought it could save a code to add class.

I'll send V2 with above approach.

-ck