Re: [PATCH] nvme-pci: add NVMe controller statistics

[Tokunori Ikegami <ikegami.t@gmail.com>]

On 2025/05/29 5:35, Keith Busch wrote:
> Specifically on sysfs attributes, the intended use was one file, one
> attribute, one value. That makes programatically parsing these much
> easier. So what you'd do instead is create 4 new files called
> "timeouts", "aborts", "resets", and "disables", and each just print out
> a sinlge number instead of this more complex output. If you're concerned
> with polluting the nvme device's sysfs directory, you can also make a
> sub-directory called "stats" to collect such things.
Just done as mentioned by the version patch.
> That principle hasn't always been strictly followed, but let's not
> introduce new deviants.
I see.
> As to what you're tracking and reporting here, I'm not convinced it's
> useful. It's just a snapshot of what's happened for the lifetime of that
> instance, which may not be very long.

Just changed the attributes as read-write device attribute then user can 
reset the counter value if needed and accumulate the counter value on 
user application.
By the way I think any unstable issue behavior caused on a drive the 
statistics counters can be used to check the warning behavior cause but 
it is okay as not be very long for the lifetime.
Thank you.