NVMe Cancel support
I am scheduled to present on the topic of NVMe Cancel support next week in the LSF/MM/BPF IO Track.
Here's an abstract for my talk.
The NVMe Cancel command was added to the NVMe spec with TP4097a back in the NVMe 2.0 time frame[1].
At that time I was working for a storage array company and was personally involved with the effort to add support for this protocol into the storage array's controller.
After moving to Red Hat in the fall of 2021 I started working with my engineering team at Red Hat to add support for TP4097a to the Linux host.
My talk next week will present the latest version of Red Hat's patches - which are based upon our previous versions[2] submitted for comments - and discuss some of the obstacles and challenges we've encountered during implementation.
Almost all of these obstacles have to do with testing. We've had these patches working in our lab for more than a year now, and we are happily able to send Cancel commands to the storage array with no problems. However, these tests are not sufficient for upstream acceptance and we've spent the last few months developing the nvmet patches needed to sufficiently test this feature with a nvmet soft target..
I want to share those nvmet patches with the community and discuss what the possible next steps could be to get this feature accepted upstream.
Many thanks to Chris Leech <[email protected]> and Maurizio Lombardi <[email protected]> who have done the lions share of this work.
[1] https://people.redhat.com/jmeneghi/LSFMM_2025/TP4097a_Abort_Enhancements_2021.07.26-Ratified.pdf
[2] https://lore.kernel.org/linux-nvme/[email protected]/
Signature · 4 lines
John A. Meneghini Senior Principal Platform Storage Engineer RHEL SST - Platform Storage Group [email protected]