Re: [PATCHv13 00/11] block write streams with nvme fdp

[Christoph Hellwig <hch@lst.de>]

On Thu, Dec 12, 2024 at 11:21:07AM +0530, Kanchan Joshi wrote:
> On 12/11/2024 12:43 PM, Christoph Hellwig wrote:
> > The changes looks good to me.  I'll ty to send out an API for querying
> > the paramters in the next so that we don't merge the granularity as dead
> > code
> 
> What do you have in mind for this API. New fcntl or ioctl?
> With ability limited to querying only or....

Yes, new fcntl or ioctl, query the number of streams and (nominal)
stream granularity as a start.  There is a few other things that
might be useful:

 - check if the size is just nominal or not, aka if the device can
   shorten it.  FDP currently allows for that, but given that
   notifications for that are out of bounds it makes a complete mess
   of software actually trying to use it for more than simple hot/cold
   separation like cachelib, so we find a way to query that in the
   spec.
 - reporting of the remaining capacity per stream, although I'm not
   sure if that should be in the same ioctl/fcntl, or better done
   using a separate interface that has the stream number as input
   and the capacity as out, which would seem a lot simpler