[PATCH] nvme: fix command effects log lifetime for multipath heads

[Keith Busch]

On Wed, Aug 26, 2026 at 03:55:43PM +0800, Yao Sang wrote:
> Hi Keith,
> 
> Thanks for the review.
> 
> Do you mean the whole IOCS entry should be shared, including CSUPP?
> 
> The Commands Supported and Effects log is controller scoped (NVMe Base
> Specification 2.3, Figure 206), and CSUPP reports command support for that
> controller (Figure 214). Zone Append, for example, is optional per I/O
> controller (ZNS 1.4, Figure 9). That seems to allow different paths to the
> same shared namespace to report different CSUPP values.

I don't think we (Linux) can work with a multipath setup where the IO
commands supported are path specific. The path selection criteria is not
based on the opcode.