Re: [PATCH] hw/nvme: remove param zoned.auto_transition

[Klaus Jensen]

On Aug 12 13:01, Niklas Cassel wrote:
> The intention of the Zoned Namespace Command Set Specification was
> never to make an automatic zone transition optional.
> 
> Excerpt from the nvmexpress.org zns mailing list:
> """
> A question came up internally on the differences between ZNS and ZAC/ZBC
> that asked about when a controller should transitions a specific zone in
> the Implicitly Opened state to Closed state.
> 
> For example, consider a ZNS SSD that supports a max of 20 active zones,
> and a max of 10 open zones, which has the following actions occur:
> 
> First, the host writes to ten empty zones, thereby transitioning 10 zones
> to the Implicitly Opened state.
> 
> Second, the host issues a write to an 11th empty zone.
> 
> Given that state, my understanding of the second part is that the ZNS SSD
> chooses one of the previously 10 zones, and transition the chosen zone to
> the Closed state, and then proceeds to write to the new zone which also
> implicitly transition it from the Empty state to the Impl. Open state.
> After this, there would be 11 active zones in total, 10 in impl. Open
> state, and one in closed state.
> 
> The above assumes that a ZNS SSD will always transition an implicitly
> opened zone to closed state when required to free up resources when
> another zone is opened. However, it isn’t strictly said in the ZNS spec.
> 
> The paragraph that should cover it is defined in section
> 2.1.1.4.1 – Managing Resources:
> The controller may transition zones in the ZSIO:Implicitly Opened state
> to the ZSC:Closed state for resource management purposes.
> 
> However, it doesn’t say “when” it should occur. Thus, as the text stand,
> it could be misinterpreted that the controller shouldn’t do close a zone
> to make room for a new zone. The issue with this, is that it makes the
> point of having implicitly managed zones moot.
> 
> The ZAC/ZBC specs is more specific and clarifies when a zone should be
> closed. I think it would be natural to the same here.
> """
> 
> While the Zoned Namespace Command Set Specification hasn't received an
> errata yet, it is quite clear that the intention was that an automatic
> zone transition was never supposed to be optional, as then the whole
> point of having implictly open zones would be pointless. Therefore,
> remove the param zoned.auto_transition, as this was never supposed to
> be controller implementation specific.
> 
> Signed-off-by: Niklas Cassel <niklas.cassel@wdc.com>
> ---
>  hw/nvme/ctrl.c | 35 ++++++++++++-----------------------
>  hw/nvme/nvme.h |  1 -
>  2 files changed, 12 insertions(+), 24 deletions(-)
> 

No quarrel from me, but we need to deprecate the parameter first.

However, I agree that it is a misinterpretation of the spec, so I'm ok
with removing the code using the parameter so it doesnt do anything.

Please do that, but retain the parameter and add it in
docs/about/deprecated.rst and remove the documentation related to the
option.

signature.asc
Description: PGP signature