1. Select the target and information

Results describe the receiving interface and controller. Interfaces can differ; with CC.CSS=110b, CSI and the enabled command-set profile also affect support. Another interface’s directory is not interchangeable.

CSI

Command Set Identifier

Use LID 00h. LPA.MLPS=1 guarantees Supported Log Pages, while MLPS=0 may still provide it and does not negate all logs. Figure 338 LPA.LPEDS bit 2 describes extended offsets/length; per-LID LSUPP and IOS describe log support and index-offset support.

There are 256 four-byte entries. LID n is at byte 4n; bit 0 is LSUPP, bit 1 IOS, and LIDSP is log-specific. The host should ignore other fields when LSUPP=0.

2. Query sequence and interpretation

Select the controller and command set, read 00h, and inspect the target entry. Then apply that log’s scope, length and parameters. Rediscover capability after relevant firmware or configuration changes.

Success returns a capability directory, not the logs themselves. LSUPP=1 for 81h establishes Sanitize Status availability, not that sanitization has run or completed.

3. Handle missing or inconsistent evidence

An unsupported LID normally returns Invalid Log Page (1/09h). Using unsupported OT=1 returns Invalid Field (0/02h), even when the LID is supported. These are different failures.

Compare LSUPP, IOS and LPA against a valid request. Supported logs need not support index offsets; an invalid request does not by itself contradict LSUPP.

First establish the same interface, CSI and configuration period, then verify four-byte entry addressing.

CSI

Command Set Identifier

1. What differs

LID 01h is controller-scoped. LID 02h may accept namespace requests, but Base 2.4 defines no namespace-specific SMART information, so controller and namespace reports contain identical information rather than independent usage totals.

Each 01h entry is 64 bytes, with ECNT, SQID, CID, status, parameter location and NSID. The 512-byte 02h page contains warnings, temperature, spare, endurance estimate and cumulative usage/error counters.

SQID

Submission Queue Identifier

CID

Command Identifier, unique among outstanding commands in one SQ

2. How to choose and verify

ELPE+1 gives the maximum Error Information entry count. The LPA SMART bit determines whether per-namespace requests are accepted; establish support before interpreting the data.

Preserve the CQE, correlate 01h to the command, then sample 02h for health context. Record times to avoid pairing an old error with an unrelated current warning.

CQE

Completion Queue Entry

A successful read does not mean an error-free device. ECNT=0 marks an invalid entry; Percentage Used=100 describes estimated endurance consumption, not necessarily device failure.

3. Conditions that shape the comparison

A specific-NSID SMART request without support returns Invalid Field (0/02h). Historical status in the log is distinct from the status of the current Get Log Page command.

The SMART Error Log Entries total is not the number of entries currently visible in 01h. Finite capacity and reset clearing can make the lifetime total larger.

1. What differs

03h is domain or subsystem scoped; 04h describes attached-namespace changes for this controller; 05h describes its commands for the selected command set.

Read CAFS, NAFS and FRS1–7 in 03h; NSIDs and the leading FFFFFFFFh overflow marker in 04h; and CSUPP, effects, CSE/CSER and CSP in 05h.

2. How to choose and verify

Check 00h and relevant Identify capabilities. Slot count and read-only restrictions come from FRMW, not merely the number of nonzero strings in 03h.

Choose by purpose: compare 03h with Identify.FR after activation; read 04h and rediscover namespaces after notification; use 05h before a management operation and refresh affected capabilities afterward.

NAFS=2 identifies a slot awaiting an applicable CLR, not the currently running slot. NSID 7 in 04h signals change without specifying whether it was formatted or deleted.

3. Conditions that shape the comparison

Separate retrieval errors from reported state. Unsupported LIDs return 1/09h; invalid defined parameters follow 0/02h rules. Empty slots or empty change lists are not themselves retrieval failures.

Correlate the active slot with FR, changed IDs with fresh Identify, and CSUPP with valid command behavior. The three logs need not contain matching values.

1. What differs

Self-test scope depends on capabilities such as DSTO.SDSO. PEL is subsystem-scoped; 81h uses NSID to distinguish subsystem and namespace targets. The querying controller is not necessarily the full affected scope.

For 06h read DSTOS before progress and validity-qualified results. For 0Dh establish or obtain a reporting context and traverse lengths. For 81h inspect SSTAT before SPROG and outcome fields.

2. How to choose and verify

Check self-test OACS, PEL LPA and subsystem/namespace sanitize capabilities, then the relevant LIDs. A readable status log does not establish every operation method.

Poll 81h after sanitize acceptance for background completion; inspect the newest valid 06h result after self-test; use PEL for supported historical reset or management events.

Successful Get establishes retrieval, not a passed test or successful sanitize. Missing PEL events require checking event support, logging conditions and retained history.

3. Conditions that shape the comparison

Unsupported LIDs return 1/09h; invalid PEL context sequencing can return 0/0Ch. Operation failures in returned data do not turn a successful retrieval CQE into failure.

CQE

Completion Queue Entry

Correlate target, time and latest results. An old test result or previous sanitize success is not proof of success for the current operation.

1. What differs

Establish the management object: endurance group, NVM set, lockdown selection/scope, controller-accessed boot partitions, or domain/subsystem media units.

Inspect group warnings/spare/endurance in 09h, set latency state/events in 0Ah/0Bh, prohibited opcodes/FIDs in 14h, partition identity/protection in 15h, and media-unit identity/association/state in 10h.

2. How to choose and verify

Check 00h and relevant Identify capabilities. The main LIDs are 09h, 0Ah/0Bh, 14h, 15h and 10h. LSI can mean different things for different logs.

Choose the LID by question, map resource IDs through Identify, then apply the log’s selectors and structure. Group-3 health requires ENDGID=3, not an arbitrary NSID substituted into LSI.

Results describe the selected object. One group warning does not establish identical media faults in every namespace; lockdown entries are prohibitions, not execution history.

3. Conditions that shape the comparison

Distinguish unsupported LIDs (1/09h) from invalid resource selectors governed by each log’s defined fields and specific status. Do not classify all selector failures as Invalid Namespace.

Correlate resource association, reported state and behavior. Prohibited FIDs should match lockdown policy; media-unit state must be related to namespace dependencies.

1. Select the target and information

Figure 209 lists scopes, while individual fields may define more specific ones. Read both the directory and the selected LID’s exceptions.

Check LSUPP, LPA, multi-domain support and whether CSI/UUID selection applies. Together they determine valid query combinations.

Controller/subsystem logs normally use NSID=0 or FFFFFFFFh. LSI, LSP, CSI and UIDX are separate selectors, not extensions of NSID.

2. Query sequence and interpretation

Name the target object first, then map it to the specified fields. Fill unused fields according to reserved-field rules.

A valid request returns the selected scope. Aggregate SMART and group endurance are different views; copying an aggregate does not create independent namespace statistics.

3. Handle missing or inconsistent evidence

Under the common rule, controller/subsystem logs with another NSID return Invalid Field (0/02h). Other scopes use their LID-specific and NSID rules.

Retain command selectors, Identify associations and returned data so comparisons refer to the same object.

Check the LID’s scope first; a nonzero NSID alone does not establish namespace-specific information.

1. Prepare the operation

Offsets locate data within the log, not host memory. OT=0 uses bytes; OT=1 uses log-defined entry indices. Do not mix their units.

Check LPA.LPEDS and per-LID IOS, then size transfers using MDTS and log-specific length rules.

NUMD=(NUMDU<<16)|NUMDL normally encodes bytes/4−1. LPOU/LPOL form the 64-bit offset; byte offsets require Dword alignment, with further log-specific constraints.

2. Sequence and completion conditions

Success covers the requested valid range. Unless otherwise specified, bytes beyond the log end are undefined and are not additional log content.

3. Handle unmet conditions

An offset beyond the log or OT=1 without IOS returns 0/02h. For nonzero low two byte-offset bits, the controller may reject or operate as if those bits were zero; neither outcome is universally mandatory.

Verify total length, offsets/NUMD and generation or reporting context. Complete coverage and one coherent snapshot are separate requirements.

1. Establish the source and scope

LSI is LID-specific; UIDX selects a UUID-list definition; RAE controls the associated event. Not every log uses all three.

Check whether LSI is defined and UUID selection supported. RAE behavior depends on the event’s acknowledgment rules.

2. Fields, units and worked interpretation

LSI is CDW11[31:16], UIDX CDW14[6:0] with zero meaning no UUID selection, and RAE CDW10[15]. UIDX is an index, not the UUID itself.

Fix the target and UUID definition, then decide whether the event should remain pending. Record RAE for each chunk and follow the selected log’s acknowledgment procedure.

Successful RAE=1 reads retain the event; successful RAE=0 reads clear it as defined. Failed reads must retain it; an attempted read is not acknowledgment.

3. Conditions that change the interpretation

Invalid UUID selection follows 0/02h rules in §8.1.31; invalid LSI follows the LID definition. Changing RAE does not fix the selected resource.

Compare the target, UUID mapping and pending-event state. Identical returned bytes with different RAE can produce different subsequent notification behavior.

1. Preserve the evidence first

Consistency is log-specific: Telemetry has generations, PEL reporting contexts. Other dynamic logs do not gain cross-command atomic snapshots by assumption.

Check the supported log structure/version and its generation, total length, availability or context action. Common command fields alone are insufficient.

Retain Telemetry generation and area boundaries, or PEL context/TLL/event lengths and versions, together with each chunk’s offset, length and acquisition time.

2. Work through the possible causes

Check generation/context continuity before blaming the parser.

Read the header, use the same capture/context and recheck the header where applicable. A changed generation requires reacquisition or an explicit consistency limitation, not just replacement of the header.

3. Connect observations to specification conditions

Initial TCDGN=10 and final=11 leave intervening chunks unproven even when all Gets succeed. Retain consistency evidence with the resulting report.

Content changes need not produce an error CQE. Invalid PEL context operations can produce 0/0Ch, which is distinct from a legitimate generation change.

Treat headers, chunks and query order as one evidence set. The last capability or generation value cannot retroactively validate every earlier chunk.

1. Mechanism and scope

Error Information is controller-scoped; PEL contains supported subsystem events accessed through reporting contexts. Capacity and read lifecycles differ.

Error entries are newest first with ECNT identity. PEL traversal uses EHL, EL, ET and ETR; events are not fixed 64-byte error entries.

2. Understand it through actions and results

ELPE+1 sets Error Information capacity. For PEL inspect support, report length/event count and maximum capacity; SMART totals do not size the current buffer.

Record valid entries and boundary identities, then correlate overlapping samples. Mark lost history explicitly rather than inventing missing records or their order.

When Error Information is full, the specification recommends inserting the new entry and discarding the oldest. A complete finite PEL report is not the device’s entire lifetime history.

3. Avoid a misleading conclusion

Full storage does not automatically make a valid Get fail. Inspect retention behavior; invalid parameters or contexts are separate command errors.

ECNT/SMART totals may diverge from visible entry counts. Explain the difference using capacity and retention, not an expectation of unlimited historical preservation.

1. Identify the trigger and affected objects

Retention may apply to a whole log or individual fields. Establish reset coverage in multi-controller and multi-domain configurations.

Use log-specific retention text and reset rules. Figure 209’s Restore to Default Content concerns manufacturing defaults, not clearing on every reset.

2. State changes and recovery

Compare error entries, their ECNT, SMART totals, current temperature, PEL events and PEL context separately. Proximity does not imply identical retention.

Capture the same target before reset, perform the specified event, recover and read again. Record concurrent activation, format or manufacturing-default restoration as separate causes.

Error entries should clear on CLR/power cycle while ECNT persists. SMART lifetime data persists across power cycles unless a field says otherwise. Persistent PEL events do not make reporting contexts persistent.

3. Checks across reset or power loss

4. Verify retention and recovery

Comparing values creates no status. A failed post-reset Get first requires checking readiness, selectors and context; it is not evidence of cleared data.

Compare field semantics and validity rather than requiring byte-identical buffers. Persistent counters can coexist with newly sampled current state.

First determine whether the discrepancy concerns stored content, an invalidated context or an interrupted query.

1. Preserve the evidence first

Fix interface, controller, CSI, UUID selection and configuration time. With CC.CSS=110b, a command set not enabled by the profile is treated as unsupported.

Preserve LSUPP/IOS, CSUPP, Identify capability bits and FID 19h. Log support and opcode support are different declarations.

FID

Feature Identifier

Retain all relevant selectors, length and data pointers, then decode CQE SCT/SC rather than only a tool’s error string.

SCT

Status Code Type

2. Work through the possible causes

First verify the same interface and command-set configuration on both sides of the comparison.

Construct a request meeting capability, scope and state requirements, execute, then exclude invalid fields, lockdown, namespace readiness and intervening changes before declaring a contradiction.

3. Connect observations to specification conditions

In a stable context, unsupported status for a valid advertised operation is a contradiction to investigate. A legitimate restrictive status is not equivalent to lack of support.

Unsupported LID uses 1/09h; unsupported opcode uses 0/01h; 0/02h calls for parameter review. Preserve specific exceptions and allowed choices for multiple errors.

Compare declarations, behavior and timing. After activation, rediscover support instead of imposing an old table on new firmware.

1. Select the target and information

Admin and I/O opcodes occupy separate regions; the I/O region also requires the correct command set. CSP can name several possible scopes because parameters affect actual impact.

First check Figure 338 LPA.CSES bit 1 for LID 05h availability and fix command-set/UUID selection. CSES advertises the log, while per-opcode CSUPP advertises each command. CSP=0 means unreported scope, not no impact.

Figure 216 locates the four-byte entry: Admin uses 4×opcode and I/O uses 1024+4×opcode, in bytes. Figure 217 decodes its support/effect fields. Admin opcode 02h is Get Log Page, whereas NVM I/O opcode 02h is Read; identical numbers do not identify the same entry.

2. Query sequence and interpretation

Select the entry and coordinate work under CSE/CSER recommendations. Use a supported nonzero CSER relaxation; otherwise use CSE. Rediscover potentially changed capabilities or inventory afterward.

NIC=1 permits inventory or multi-namespace capability changes; it does not guarantee one new namespace per call. CSUPP=0 requires all other entry fields to be zero.

3. Handle missing or inconsistent evidence

Retrieval errors follow Get Log Page; operation errors follow the command. Submission recommendations do not authorize inventing a mandatory status for every violation.

Correlate effects with before/after Identify, inventory and data. Support declarations should agree; actual effects depend on this invocation’s parameters.

First establish Admin/I/O region, opcode, byte offset and query success. Log-relative position, segment-buffer position and entry bit position are distinct coordinates; confusing them can select another command or field.