1. Establish the source and scope

Opcode interpretation depends on the Admin/I/O command set; equal numeric values need not mean the same operation across sets.

Check capabilities such as OACS, ONCS and SGLS and the namespace CSI. Not every command uses every common field.

CSI

Command Set Identifier

2. Fields, units and worked interpretation

CDW0 contains OPC, FUSE, PSDT and CID. NSID selects the target, MPTR/DPTR describe data, and CDW10–15 are command-specific. The common command is 64 bytes.

Select command set, opcode/support, legal NSID/parameters, data/pointers and a unique outstanding CID, then submit. PCIe Admin commands use PRPs, not arbitrary SGLs.

Match completion by SQID+CID. Results may reside in the host buffer or CQE DW0/1, not solely Status.

3. Conditions that change the interpretation

Unsupported opcode uses Invalid Command Opcode (0/01h). Invalid defined fields generally use Invalid Field (0/02h), subject to specific error definitions.

Cross-check command-set context, support and payload size. A valid DPTR address does not prove sufficient described data length.

1. Distinguish the failure conditions

Host construction requirements and controller validation requirements are separate.

The host shall zero reserved fields; the recipient need not check reserved bits/bytes/fields. A reserved coded value in a defined command field shall be reported as an error.

Unsupported/reserved opcode: 0/01h. Illegal defined field: generally 0/02h unless a specific status applies. Multiple faults usually permit choosing one; nonzero reserved bits do not universally mandate 0/02h.

2. Establish the cause from evidence

Use Base §1.4.1 reserved semantics, opcode support and the individual command's field restrictions.

Start from a zeroed SQE and populate used fields. Change one item per error exercise and record whether the requirement says shall or should.

3. Outcome and follow-up checks

A legal combination follows normal completion. Lack of reserved-bit checking does not make a nonzero host value legal.

Record both host violation and controller detection obligations to avoid labeling permitted nonchecking as a firmware bug.

First ask whether the modified item is a reserved bit or a reserved encoding within a defined field.

1. Distinguish the failure conditions

Active state is controller-relative. Accessibility through another path does not establish activity on this controller.

Distinguish zero, valid allocated, inactive and FFFFFFFFh identifiers. Each command defines the scope of FFFFFFFFh.

Figure 93 defaults: inactive on an NSID-using command returns 0/02h, invalid returns 0/0Bh. Unsupported broadcast or nonzero NSID on an unused field returns 0/02h. Command-specific exceptions take precedence.

2. Establish the cause from evidence

Use active/allocated lists and the command's NSID rules, then check command-specific exceptions.

Determine whether NSID is used, then permitted namespace state and broadcast support. Some Identify CNS values intentionally query allocated/inactive namespaces.

CNS

Controller or Namespace Structure, selecting Identify data

3. Outcome and follow-up checks

Legal targets follow the command's completion rules. Some queries define special replies for absent/inactive objects that override the generic rule.

Correlate the original NSID, the list at that time and CNS/FID/opcode rules. A current list alone may not explain a past request.

CNS

Controller or Namespace Structure, selecting Identify data

FID

Feature Identifier

First check whether the command uses the generic rule or defines an explicit exception.

1. Establish the source and scope

SQHD describes the identified SQ, P the CQ slot, and Status the command identified by SQID+CID.

Interpret using current queue configuration and the common CQE layout, not a feature-selected meaning of SQHD.

2. Fields, units and worked interpretation

DW2 contains SQHD/SQID; DW3 contains CID/P/Status; DW0/1 carry command-specific results. SQHD reports consumption when that CQE was constructed.

Validate P, match SQID+CID, decode status/results, update SQ slot availability and finally release the CQ slot.

Simulated data: the previously known Head of one SQ is 4. A valid CQE identifies that SQ and reports CID=9 and SQHD=8. With no wrap in between, SQHD establishes that the controller consumed entries in slots 4–7, allowing those SQ slots to be reused; CID and Status identify and report the result of command 9. Other consumed commands need not have completed, and their data buffers cannot be reclaimed merely because their SQ slots are reusable. Figure 98 notes that SQHD is captured when this CQE is created: by the time the host reads it, the controller’s actual Head might have advanced to 10, which this CQE alone cannot establish. Read Base 2.4 §3.3.1.2 for submission and consumption, Figure 98 for SQHD and SQID, then Figure 99 for CID, P and Status.

3. Conditions that change the interpretation

Decode SCT and SC together; P is not an error code. Malformed CQEs or unknown CIDs require checking memory/lifetimes, not inventing a new NVMe status.

Distinguish SQ consumption, command completion and data-buffer reclamation. Reusable SQ slots do not permit early release of command data.

1. Establish the source and scope

Status belongs to this completion and must be interpreted with opcode, command set and failure conditions.

Base defines common SCT/SC values; use NVM definitions for its specific codes, not another command set's table.

2. Fields, units and worked interpretation

In CQE DW3, SCT is bits 27:25, SC bits 24:17 and P bit 16. In the final 16-bit word, SCT occupies bits 11:9 and SC bits 8:1.

Confirm data width/endianness, extract SCT/SC, select the correct table, then inspect DNR, More, CRD and command-specific results.

DNR

Do Not Retry, a completion retry indication

SCT=0, SC=00h means Successful Completion, but matching and phase still matter; zero low bits alone do not establish a new valid success.

3. Conditions that change the interpretation

For example, 0/01h is Invalid Command Opcode and 1/01h Invalid Queue Identifier. SCT 2 is Media/Data Integrity, SCT 3 Path Related and SCT 7 Vendor Specific.

Preserve raw CQE bytes and decoded values. Tools that remove P or repack status cannot have their integers compared blindly.

1. Mechanism and scope

Decode the same CQE; DNR's identical-command retry semantics cover any controller in the same subsystem.

DNR does not establish that data was unchanged. More=1 points to additional status in LID 01h, not another completion.

LID

Log Page Identifier

2. Understand it through actions and results

The common bits always exist; CRD use additionally depends on Host Behavior Support.ACRE and Identify.CRDT1–3.

Decode the cause; with DNR=0, address transient conditions, respect valid CRD guidance and assess retry safety. For non-idempotent work, determine whether the earlier operation took effect.

Only the new CQE establishes retry success. DNR=0 means may succeed, not immediate or duplicate-free success.

3. Avoid a misleading conclusion

Repeating unchanged invalid parameters is not useful. Apply explicit DNR rules: Command Interrupted (0/21h) requires DNR=0 and may be returned only with ACRE enabled.

More=1 must correspond to associated additional Error Information. Interpret CRD only when DNR/ACRE make it applicable.

1. Preserve the evidence first

This involves CQ memory, host Head/expected phase and interrupt routing, not necessarily execution failure.

Check Create CQ.IEN/IV, the selected MSI/MSI-X mode, masks and interrupt features.

MSI

Message Signaled Interrupt

MSI-X

Message Signaled Interrupts Extension

For MSI-X inspect Function Mask, vector mask and PBA; ordinary MSI uses different INTMS/INTMC rules. CQE validity still depends on location and phase.

MSI-X

Message Signaled Interrupts Extension

MSI

Message Signaled Interrupt

2. Work through the possible causes

First check CQ/slot selection and expected phase before assuming interrupt loss.

Inspect the correct CQ Head and phase first. Process valid CQEs; investigate missing interrupts through IEN, IV, masks, coalescing and the host handler.

3. Connect observations to specification conditions

Polling can consume a valid completion without a one-to-one interrupt. Interrupt count need not equal completion count.

Missing an interrupt does not itself carry an NVMe error status; the CQE may be successful. Do not invent Internal Error for it.

Compare posting, masking, interrupt delivery and Head-update times. An IRQ is notification, not the completion payload itself.

1. Which stages are ordered

Independent commands need not complete FIFO even within one SQ; nearby submission times across SQs do not establish dependency.

Check for a fused operation or command-specific ordering rule; FUSES advertises fused support.

Distinguish submitted, consumed, processing and completed stages. SQHD only reports consumption.

2. How the host handles dependencies

If B depends on A, wait for A's successful completion before submitting B. Apply Flush/FUA separately for persistence.

Independent Read B may legally complete before Read A; match by SQID/CID rather than submission-array order.

Out-of-order completion alone has no error status. Violations of defined multicommand sequences follow their own errors.

Test required causal relations rather than treating apparent reordering alone as noncompliance.

1. Mechanism and scope

Examples include CQ-before-SQ creation, SQ-before-CQ deletion, successful Set Features before affected new commands, and supported fused pairs.

FUSE=01b/10b marks the first/second command. In PCIe they must be adjacent in one SQ and submitted by one Tail update.

2. Understand it through actions and results

Check FUSES and individual feature/command rules; no global ordered-mode bit resolves every case.

Establish ordinary dependencies by waiting for completion; submit fused pairs under their special rules with separate CQEs. Do not assume the controller orders arbitrary concurrent requests for the host.

Commands submitted after successful Set Features use the new value; previously submitted ones may use old or new settings. Drain first when testing a clean transition.

3. Avoid a misleading conclusion

A missing adjacent fused partner uses 0/0Ah. Queue dependency errors use Create/Delete-specific statuses, not universally 0/0Ch.

Trace completion-to-submission boundaries. FUA controls persistence, not these dependencies.

1. Mechanism and scope

Limits may affect an SQ, shared CQ or internal resources; not every limit is controller-wide failure.

Fetching submitted SQEs uses a vendor-specific algorithm; arbitration selects where candidate processing starts. Fetch and execution start are distinct.

2. Understand it through actions and results

The host uses queue size/SQHD and unique CIDs. Do not import another transport's MAXCMD flow-control assumptions into PCIe.

The host must not overwrite unconsumed SQEs; the controller must not post into a full CQ and may pause related further processing while independent SQs continue.

Progress resumes as capacity becomes available. A freed SQ slot can belong to a still-outstanding command because consumption is not completion.

3. Avoid a misleading conclusion

Normal resource limits do not automatically require error CQEs; pointer misuse, CID reuse and command-specific limits have separate rules.

Use evidence for each stage; submitted count does not establish simultaneous execution of every command.

1. What differs

Arbitration governs candidate processing starts, not completion order or a fixed IOPS ratio.

FID 01h provides AB and HPW/MPW/LPW; QPRIO applies in WRR. AB=7 means unlimited burst, otherwise 2^AB; weights encode value+1.

FID

Feature Identifier

2. How to choose and verify

All controllers support Round Robin. Discover optional WRR via CAP.AMS and select it with CC.AMS.

Select CC.AMS while disabled, set FID 01h after enabling, create SQ priorities and observe under continuously eligible workloads.

RR gives all SQs, including Admin, equal round-robin priority. WRR prioritizes Admin, then Urgent, then weighted High/Medium/Low service with within-class rules.

3. Conditions that shape the comparison

Unsupported CC.AMS programming is undefined, not a Set Features Invalid Field completion. Invalid defined FID 01h parameters follow command rules.

Get FID 01h verifies configuration only. Service time, media contention and CQ pressure affect throughput, so IOPS ratios alone cannot validate WRR.

1. Mechanism and scope

QPRIO is an SQ-level setting, not a per-Read/Write priority field.

Create SQ CDW11.QPRIO encodes Urgent 00b, High 01b, Medium 10b and Low 11b; FID 01h weights the latter three.

2. Understand it through actions and results

QPRIO is used only when WRR is supported and selected through CC.AMS; otherwise it shall be ignored.

Create SQs with suitable QPRIO. Change an existing queue through reconstruction, not by editing an old Create SQE in host memory.

Urgent outranks weighted service but remains below Admin. Persistent Urgent work may starve lower classes; High has no universal fixed minimum bandwidth.

3. Avoid a misleading conclusion

Ignoring QPRIO under RR is correct, not a failed priority feature. Invalid queue parameters still follow Create statuses.

Preserve each SQ's QPRIO, CC.AMS and FID 01h together; none alone explains priority behavior.

1. Preserve the evidence first

Observe competing SQs on one controller after excluding CQ Full, dependencies, unequal service costs and insufficient offered work.

Record CAP.AMS, CC.AMS, FID 01h AB/weights and each SQ's QPRIO.

In WRR, processing starts per round are limited by remaining credits and AB. Ready candidates matter, not merely a nonempty SQ.

2. Work through the possible causes

First verify continuously eligible high-priority candidates and available CQ space.

Use comparable commands, sustained supply and nonblocking CQs. Compare processing-start/scheduler evidence where available; CQEs alone cannot reconstruct all internal choices.

3. Connect observations to specification conditions

High/Low weights of 4:1 describe service credits, not an exact completion ratio in every short interval.

Measurement itself has no error CQE. Verify configuration success and distinguish shall/may requirements; unsuitable workload conditions do not establish noncompliance.

A noncompliance claim needs a contradiction between a defined selection rule and evidence, not merely slower high-priority work.