Re: [PATCH 3/9] hw/block/nvme: support per-namespace smart log

[Klaus Jensen]

On Oct  1 09:20, Keith Busch wrote:
> On Thu, Oct 01, 2020 at 06:10:57AM +0200, Klaus Jensen wrote:
> > On Sep 30 15:04, Keith Busch wrote:
> > > Let the user specify a specific namespace if they want to get access
> > > stats for a specific namespace.
> > > 
> > 
> > I don't think this makes sense for v1.3+.
> > 
> > NVM Express v1.3d, Section 5.14.1.2: "There is no namespace specific
> > information defined in the SMART / Health log page in this revision of
> > the specification.  therefore the controller log page and namespace
> > specific log page contain identical information".
> > 
> > I have no idea why the TWG decided this, but that's the way it is ;)
> 
> I don't think they did that. The behavior you're referring to is specific to
> controllers that operate a particular way: "If the log page is not supported 
> on
> a per namespace basis ...". They were trying to provide a spec compliant way
> for controllers to return a success status if you supplied a valid NSID when
> the controller doesn't support per-namespace smart data..
> 
> The previous paragraph is more clear on this: "The controller may also support
> requesting the log page on a per namespace basis, as indicated by bit 0 of the
> LPA field in the Identify Controller data structure".

OK, so I agree that it makes sense for it to be supported on a per
namespace basis, but I think the spec is just keeping the door open for
future namespace specific stuff in the log page - currently there is
none.

Figure 94 (the actual SMART log page) says that the Data Units
Read/Written are controller wide, so there really is no namespace
specific information. Maybe this could be in the context of shared
namespaces? How would a controller know how much data has been
read/written from it without asking the other controllers? What if a
controller is detached from the namespace - you'd lose those numbers.

signature.asc
Description: PGP signature