[PATCH] nvme-tcp: add additional info for nvme_tcp_timeout log

[Sagi Grimberg]

> Hi Sagi.
> 
> The previous msg:
> nvme nvme0: queue 0: timeout request 0x0 type 4
> 
> We found it more useful to have also:
> cmd opc that was timed out

That is fine

> command id that was timed out

That is the lower 12 bits of the tag.
You can introduce a generic helper to extract genctr+cid from a tag.

> The timeout value for this specific io - in case of a timeout - it make sense to know the timeout value itself
> more info about the blk rq - such as:
> 	a. time of a request that was allocated for the IO that was timed out
> 	b. time of the timed out IO that was submitted to the device

I don't understand what added value that is providing you.
An IO timeout is usually (if not universally) >> the delta from tag
allocation to submission.

This is debug information, and in my mind the proper place for this type
of debug is blktrace.

> The log is already exists, why not provide additional info that can help with io timeouts debug?

We can say that about a lot of things.

There is negative value from spam information in the log which is debug
in nature and is separate from the normal error/warn log information.

If this information existed in the form of trace points would that solve
your problem? Keith had a patch for nvme-tcp trace points, which can
provide this type of information.