1. Establish the source and scope

These reads observe one controller without changing queues or media. A version number does not establish support for every optional feature.

VS.MJR/MNR/TER encode the version. CAP.MQES+1 limits I/O queue entries, CQR specifies physical-contiguity requirements, and CSS describes command-set support.

2. Fields, units and worked interpretation

CAP.TO uses 500 ms units. Doorbell stride is 2^(2+DSTRD) bytes; supported pages range from 2^(12+MPSMIN) to 2^(12+MPSMAX) bytes. Enable timeout also requires CRTO.

Map the NVMe registers, read VS/CAP, select legal MPS/CSS/AMS and ready mode, then configure queues. If IOCSS=1, discover combinations with Identify CNS 1Ch after enabling.

CNS

Controller or Namespace Structure, selecting Identify data

The result is configuration information, not a CQE. For example, MQES=03FFh means up to 1024 slots; DSTRD=2 means adjacent doorbells are 16 bytes apart.

CQE

Completion Queue Entry

3. Conditions that change the interpretation

Reading read-only registers has no Invalid Field CQE. Later programming unsupported CC.MPS/AMS values can be undefined behavior, not a condition with a guaranteed error completion.

CQE

Completion Queue Entry

Cross-check CAP.CSS against CC.CSS and Identify combinations, and MQES against Create Queue QSIZE. Admin queues have their separate 4096-slot AQA limit.

1. Prepare the operation

This controls the controller's management interface. A wrong queue address may prevent every subsequent Identify or Get Features from completing.

Check CAP's MPS range, CSS, AMS and CRMS, together with AQA's Admin queue limits.

AQA.ASQS/ACQS encode entries−1. ASQ/ACQ contain physical bases aligned to CC.MPS. Configure CSS, MPS, AMS and CRIME in CC; NVM I/O uses IOSQES=6 and IOCQES=4.

2. Sequence and completion conditions

Wait for the previous reset to reach RDY=0. Allocate contiguous Admin queues and clear Admin CQ phases. Program AQA/ASQ/ACQ and legal CC fields with EN=0, then enable and wait for RDY=1. Set I/O entry sizes before Create I/O Queue.

CQ

Completion Queue

The Admin queues become usable for Identify. I/O queues are not automatically created, and independent mode does not imply that every medium is ready.

3. Handle unmet conditions

Enabling with an AQA size of zero, changing MPS while enabled or selecting unsupported MPS/AMS has undefined results. Do not invent Invalid Field CQEs for these register violations.

Validate register encodings and then a correctly identified Identify completion. CC.EN=1 alone does not establish successful initialization.

1. Prepare the operation

This establishes readiness of the target controller; media availability also depends on the selected ready mode.

Read CAP.CRMS, CC.CRIME, CRTO.CRWMT/CRIMT and CAP.TO. CAP.TO alone does not cover every enable timeout.

2. Sequence and completion conditions

3. Handle unmet conditions

Register enabling has no CQE. While media is unready, eligible commands may return Namespace Not Ready (SCT=0, SC=82h) or Admin Command Media Not Ready (0/24h).

SCT

Status Code Type

Correlate timing, CRIME, command type and status. Figure 84's eligible Admin commands and restrictions do not permit rejecting every Admin command temporarily.

4. Correlate error completions and records

Enabling has no DNR/More. Eligible media-not-ready responses may have DNR=0 during the allowed Independent-mode interval; that transient explanation cannot continue with DNR=0 beyond CRWMT. More reflects additional Error Information.

DNR

Do Not Retry, a completion retry indication

For an initialization-deadline failure meeting §3.5.4.1, a controller supporting PEL shall record an NVM Subsystem Hardware Error Event with Controller Ready Timeout Exceeded. This is not a record for every successful enable.

1. Prepare the operation

Clearing EN resets this controller and all its I/O queues; it does not merely pause one SQ.

SQ

Submission Queue

CAP.TO defines the worst-case wait for RDY 1→0 after EN is cleared. Preserve CC/CSTS snapshots too.

In the normal enabled state, initiate EN 1→0 with RDY=1; RDY=0 indicates readiness to enable again.

2. Sequence and completion conditions

Stop new submissions, clear EN and poll RDY. Once zero, reinitialize Admin CQ and host tracking, configure CC and enable again.

RDY=0 establishes disable completion; receiving a CQE for every old command is not required. Old I/O queues are invalid.

3. Handle unmet conditions

Setting EN 0→1 while RDY remains 1 has undefined results. This is not an Admin request requiring Command Sequence Error.

Check EN/RDY transitions and retirement of old queues. Residual data in an old queue does not prove it is reusable.

1. When commands may be submitted

This applies to Admin and I/O submission on this PCIe controller. Configuring Admin queue registers during initialization is different from command submission.

Check CSTS.RDY, CC.EN and queue existence. Even RDY=1 requires a created queue and consideration of shutdown state.

Wait for RDY=1 and a valid queue lifetime, make the SQE visible to the device, then update SQ Tail.

SQE

Submission Queue Entry

2. Doorbells do not replace readiness

SQ Tail announces submissions; CQ Head announces consumed completions. Neither is a command for making the controller ready.

Violating readiness or queue-lifetime preconditions has no guaranteed CQE. A disabled Admin Queue cannot be relied upon to report the misuse.

EN=1, RDY=0 and a timeout first indicate an initialization-state issue; an illegally early Identify is not evidence that Identify is unsupported.

1. Preserve the evidence first

The failure may affect controller-wide command service. CFS and an individual namespace media error are not interchangeable.

Observe CSTS.CFS with RDY, CC.EN and the applicable timeout. No Identify optional-capability bit permits ignoring CFS.

Preserve registers and available CQEs/logs first. Collect diagnostics if the Admin path still works, without waiting indefinitely for an inaccessible Get Log Page.

2. Work through the possible causes

First confirm valid register access and an actual CFS value of 1. An all-ones failed read is not evidence that every status bit is asserted.

Perform Controller Level Reset as appropriate to the failure and accessible interface, then reinitialize. An unresponsive interface requires platform recovery; a CC.EN reset is not guaranteed to repair every fault.

3. Connect observations to specification conditions

Verify RDY, Identify and command completion on new queues after recovery. Clearing CFS alone does not validate data integrity.

Internal Error (SCT=0, SC=06h) may apply when affected commands can still complete. Serious failure preventing CQE communication may set CFS; do not require a CQE for every command.

Correlate CFS, initialization deadlines and hardware records at the same time. One Invalid Field completion is not itself controller-fatal evidence.

1. Mechanism and scope

The modes are Controller Ready With Media and Controller Ready Independent of Media, not arbitrary degraded-ready encodings.

CRIME=0 selects With Media; CRIME=1 selects Independent. RDY, CRTO.CRIMT/CRWMT and Figure 84's command classes determine permitted work.

2. Understand it through actions and results

CAP.CRMS.CRWMS/CRIMS advertise support. With CAP.CRMS=11b, CC.CRIME selects the mode.

Select before enable. After Independent-mode RDY=1, perform media-independent discovery; defer eligible media operations based on transient status and track CRWMT from enable.

Normal With Media readiness includes required media readiness. Independent RDY=1 initially establishes interface and eligible-command readiness, not readiness of every namespace.

3. Avoid a misleading conclusion

Eligible cases may return 0/82h or 0/24h. The latter is restricted to appropriate Admin commands in Independent mode. Continuing the same transient DNR=0 explanation beyond the media deadline violates that timing rule.

Cross-check CAP support, CC selection, timeouts and behavior. CSTS.PP is a separate processing-pause indication, not a third ready mode.

1. What differs

Enable and CC.EN clearing target a controller. NVM Subsystem Reset covers its defined subsystem/domain scope. Shutdown also distinguishes controller and subsystem types.

Enable is EN 0→1; Controller Reset is EN 1→0. Writing 4E564D65h to supported NSSR.NSSRC requests subsystem reset. Normal controller shutdown uses CC.SHN=01b and waits for SHST=10b.

2. How to choose and verify

CAP.NSSRS advertises NSSR support; shutdown capability and CSTS.ST distinguish shutdown scope.

Coordinate I/O, notify shutdown and await completion before normal power removal. Select a suitable reset for recovery, then rebuild the operating environment.

Enable uses RDY=1; disable/reset uses its corresponding state; shutdown uses SHST=10b. Completed shutdown does not mean every register has its reset value.

3. Conditions that shape the comparison

Register operations have no common CQE. During shutdown, commands may return Commands Aborted due to Power Loss Notification (0/05h); that code is not universal for commands interrupted by reset.

Record the trigger and scope before comparing retained state. A log saying only “reset” cannot explain differences in AQA, PCIe state or features.

1. Identify the trigger and affected objects

Here Controller Reset specifically means clearing CC.EN. Do not mix in the register-retention rules of FLR or NVM Subsystem Reset.

FLR

Function Level Reset

Use §3.7.2's exceptions and register Reset columns; for features examine scope, saveability and Figures 126/127/466.

2. State changes and recovery

This reset preserves AQA/ASQ/ACQ, specified PMR registers and CMBMSC, but still resets Admin Queue pointers. All I/O queues are deleted and CC selections require reconfiguration. Saveable features restore Saved, or Default without Saved, under reset-scope rules; unsaveable features use their persistence rules. Reconfigure HMB.

PMR

Persistent Memory Region

HMB

Host Memory Buffer

Wait for RDY=0, verify retained Admin memory remains valid and initialize Admin CQ phases to zero. Configure and enable, rediscover needed capabilities/features, renegotiate Number of Queues, then create CQs before SQs.

New commands complete in the new queue lifetime. Reusing a QID or address does not resume the old queue.

3. Verify retention and recovery

There is no universal Reset Status owed for every old command. Invalid reconstruction parameters follow Create-command rules; writing doorbells for stale memory is not a recovery method.

Compare register values, Get Features and actual queue capability. Retention of AQA is distinct from validity of Admin CQ contents.

First check retirement of old outstanding tracking and phase reinitialization. A reused CID must not resolve to a command from the previous lifetime.

CID

Command Identifier, unique among outstanding commands in one SQ