When preparing model answers to technical MCTP and PLDM questions, it helps to connect core topics—MCTP EIDs and packet assembly, PLDM Instance IDs, sensor thresholds and hysteresis, PLDM firmware update, and OpenBMC pldmd—directly to their specification numbers. Based on the DMTF PMCI standards, this post provides concise, specification-backed model answers to 19 common technical questions.

Summary

Most MCTP/PLDM questions fall into five groups: MCTP packets and addressing (EIDs and tags), PLDM common rules (TID and Instance ID), monitoring (sensors, thresholds, and events), firmware update (Type 5 sequence), and the layer-by-layer order for debugging a missing response. Citing the underlying specification numbers (DSP0236, DSP0240, DSP0248, DSP0267) makes each answer much stronger.

What questions come up in MCTP/PLDM interviews?

One-line answer: They fall into five groups: MCTP transport, PLDM Base common rules, monitoring (sensors and events), firmware update, and layer-by-layer debugging (including the link to OpenBMC and Redfish).

The 19 questions are organized by topic group as follows:

GroupQuestions and Core Topics
MCTP transportQ1–Q8 (medium independence, roles, SOM/EOM/Pkt Seq, Msg Tag/TO, MCTP EIDs, bindings, Control messages)
PLDM Base common rulesQ9–Q11 (discovery commands, PLDM Instance ID and retries, timing PN1 and PT1–PT4)
PLDM monitoring (sensors and events)Q12–Q15 (numeric vs. state sensors, sensor threshold hysteresis, event reporting, PDR repository)
PLDM firmware updateQ16–Q17 (Type 5 sequence, failure and abort handling)
Layer-by-layer debuggingQ18 (bottom-up check from bus to MCTP, PLDM, and application service)
OpenBMC and Redfish linkQ19 (AF_MCTP, mctpd, libpldm, OpenBMC pldmd, bmcweb, Type 6 RDE)

How do you answer MCTP transport questions?

One-line answer: Explain medium independence, packet assembly with SOM/EOM/Pkt Seq, Msg Tag and TO bits, MCTP EID assignment, and transport binding differences based on DSP0236.

Q1. Why does the BMC manage devices with MCTP/PLDM requests and responses instead of reading device memory directly?
  • MCTP is medium-independent: it uses the same message format over SMBus/I2C, I3C, and PCIe VDM, so management software does not have to be rewritten per bus.
  • If register layouts differ per device, the BMC would need a per-device map; with PLDM, the device describes itself with PDRs and the BMC asks with standard commands (GetPDR, GetSensorReading), so adding a device changes BMC code less.
  • Request/response messages carry an explicit Completion Code, so success, failure, or an unsupported command come back clearly.

Basis: DSP0236, DSP0240, DSP0248

Q2. What are the respective roles of MCTP and PLDM?

Basis: DSP0236, DSP0239

Q3. How does MCTP split large messages into packets and reassemble them?
  • The header includes SOM (start) and EOM (end) bits; SOM=1 and EOM=1 indicates a single-packet message that needs no reassembly.
  • The baseline transmission unit is 64 bytes.
  • The receiver starts assembly only on a SOM packet; middle or end packets without a preceding SOM are silently dropped, and a new SOM during assembly discards the in-progress assembly and starts again.

Basis: DSP0236 8.7–8.9

Q4. Why is Pkt Seq 2 bits wide and what does it detect?
  • It increments modulo 4 per packet in a multi-packet message, allowing the receiver to detect up to three successive missing packets.
  • An unexpected sequence number terminates assembly; MCTP does not use out-of-order packets for message assembly.

Basis: DSP0236 header field table, 8.9

Q5. Why are the Message Tag and Tag Owner (TO) bit needed?
  • Msg Tag (3 bits) distinguishes several messages in flight with the same peer; TO and the tag stay the same from SOM to EOM.
  • TO=1 means “I originated this tag (requester side)”; the responder returns the same tag value with TO=0.
  • A TO=0 packet with a tag the receiver did not issue or no longer expects is dropped (“Bad, unexpected, or expired message tag”).

Basis: DSP0236 header field table, 8.7

Q6. Who assigns EIDs and how?
  • The bus owner manages MCTP EID assignment: Set Endpoint ID (0x01) assigns an EID, and Get Endpoint ID (0x02) checks it.
  • EID 0 is the null EID, and 0xFF is the broadcast EID (reserved).

Basis: DSP0236

Q7. How do the SMBus/I2C, I3C and PCIe VDM bindings differ?
  • SMBus/I2C binding (DSP0237): MCTP packets use SMBus command code 0x0F; for serial bus basics, see I2C vs SPI vs UART: How to Compare Them in an Embedded Interview.
  • I3C binding (DSP0233): target-to-controller transfer has IBI (In-Band Interrupt) mode and polling mode, and the target must provide both.
  • PCIe VDM binding (DSP0238): uses DMTF Vendor ID 1AB4h and routing types Route to Root Complex, Route by ID, and Broadcast from Root Complex.

Basis: DSP0237, DSP0233, DSP0238

Q8. When are MCTP Control messages used?
  • Common discovery and configuration commands include Set Endpoint ID (0x01), Get Endpoint ID (0x02), Get Endpoint UUID (0x03), Get MCTP Version Support (0x04), and Get Message Type Support (0x05).
  • A natural flow is to first check with Get Message Type Support whether the device supports PLDM (0x01), then move on to PLDM commands.

Basis: DSP0236, DSP0239

How do you answer PLDM Base common-rule questions?

One-line answer: Explain initial discovery commands (GetTID, GetPLDMTypes, GetPLDMVersion, GetPLDMCommands), the 5-bit PLDM Instance ID retry rules, and timing parameters (PN1, PT1–PT4) based on DSP0240.

Q9. What do you query first when meeting a PLDM device?

Basis: DSP0240, libpldm base.h

Q10. What is the PLDM Instance ID and how does it relate to retries?
  • It is a 5-bit value that pairs a request with its response; the response carries the same PLDM Instance ID.
  • Retries reuse the same Instance ID, and a new Instance ID is used only after retries for that request are finished.
  • A responder that already processed a request successfully and receives it again with the same Instance ID does not execute it again and returns SUCCESS; PLDM responses themselves are not retried (DSP0240 also lists “Bad Instance ID”).

Basis: DSP0240

Q11. How do you explain PLDM timing (PN1 and PT1–PT4)?
  • PN1 request retries: minimum 2 (a minimum of three tries in total: the original try plus two retries), all within PT3 max of the initial request.
  • PT1 request-to-response time: max 100 ms; PT4 transmission delay: max 100 ms; PT3 Instance ID expiration interval: min 5 s, max 6 s.
  • PT2 (the time a requester waits before retrying) is not a single number but a range derived from PT1, PT3, and PT4.

Basis: DSP0240 1.2.1 Table 6

How do you answer PLDM monitoring (sensor and event) questions?

One-line answer: Cover numeric vs. state sensors, upper/lower thresholds and sensor threshold hysteresis on recovery transitions, event messages vs. polling, and self-describing PDRs based on DSP0248.

Q12. How do numeric sensors and state sensors differ?
  • Numeric sensors read a value such as temperature or voltage with GetSensorReading (0x11) and are described by a Numeric Sensor PDR.
  • State sensors report one of a defined set of states with GetStateSensorReadings and are described by a State Sensor PDR.

Basis: DSP0248, libpldm platform.h

Q13. How do thresholds and hysteresis work?
  • Upper and lower thresholds each have warning, critical, and fatal levels, ordered lower fatal < … < upper fatal, and are read or written with GetSensorThresholds (0x12) and SetSensorThresholds (0x13).
  • Sensor threshold hysteresis is subtracted from the upper thresholds and added to the lower thresholds; it applies only on the transition from a more severe to a less severe state, keeping the state from flickering near a boundary (commands GetSensorHysteresis 0x15 and SetSensorHysteresis).
  • EventState includes the effect of hysteresis, whereas PresentState does not.

Basis: DSP0248

Q14. How are device events received: polling or PlatformEventMessage?
  • The terminus can send events to the event receiver with PlatformEventMessage (0x0A).
  • The receiver can also fetch queued events with PollForPlatformEventMessage.
  • Receiving events only on changes rather than reading sensors every time can reduce bus traffic.

Basis: DSP0248

Q15. Why is the PDR repository needed?

Basis: DSP0248

How do you answer PLDM firmware update questions?

One-line answer: Describe the UA and FD roles, the PLDM firmware update sequence where the FD pulls image data via RequestFirmwareData, and abort handling via CancelUpdate/CancelUpdateComponent and the FD internal timer based on DSP0267.

Q16. In what order does a PLDM Type 5 firmware update proceed?
  • Roles: UA (Update Agent, typically the BMC) and FD (Firmware Device). Before starting, QueryDeviceIdentifiers (0x01) and GetFirmwareParameters (0x02) identify the device and current versions.
  • UA → FD: RequestUpdate (0x10) → (optional) package data → PassComponentTable (0x13) → per-component UpdateComponent (0x14).
  • FD → UA: RequestFirmwareData (0x15) repeated as the FD pulls the image → TransferComplete (0x16) → VerifyComplete (0x17) → ApplyComplete (0x18).
  • UA → FD: ActivateFirmware (0x1A), then GetStatus (0x1B) to check status periodically; see What are PLDM types? Base, monitoring (PDRs and sensors), firmware update and RDE for an overview.

PLDM Type 5 firmware update UA and FD command sequence

Basis: DSP0267 1.3.0, libpldm firmware_update.h

Q17. How are failures or aborts handled during a firmware update?
  • The UA can cancel the whole update mode with CancelUpdate (0x1D) or only the current component with CancelUpdateComponent (0x1C).
  • The FD has an internal timer that ends update mode, so it is not stuck if the UA disappears.
  • Check the FD’s current state with GetStatus, then decide whether to retry or cancel.

Basis: DSP0267

In what order do you say you would debug a missing response?

One-line answer: Because MCTP does not report dropped packets, verify from the bottom layer up: physical bus → MCTP (EID, routes, tags, sequence) → PLDM (Completion Code, Instance ID) → upper management service.

Q18. In what order do you debug when no response arrives?
  • ① Bus stage: physical-layer errors (data integrity, framing, packet size) drop packets right there, so check wiring, addressing, and bus state first.
  • ② MCTP stage: an unroutable EID, a non-matching destination EID, an unsupported header version or transmission unit, a bad tag, or an out-of-order Pkt Seq are all silently dropped. On a Linux environment where MCTP is available, check interfaces, EIDs, and routes with mctp link, mctp address, mctp route, and mctp neigh (see How are MCTP and PLDM handled on Linux? AF_MCTP sockets, mctp, mctpd, libpldm and pldmd).
  • ③ PLDM stage: if a response arrives but failed, inspect the Completion Code (SUCCESS 0x00, ERROR 0x01, ERROR_INVALID_DATA 0x02, ERROR_INVALID_LENGTH 0x03, ERROR_NOT_READY 0x04, ERROR_UNSUPPORTED_PLDM_CMD 0x05, ERROR_INVALID_PLDM_TYPE 0x20). If no response arrives at all, check the Instance ID retry and expiration rules (Q10, Q11).
  • ④ Application stage: only then look at the management service on top (for example, pldmd). Because the base MCTP protocol does not handle flow or congestion control and does not notify about dropped packets, always check from the lower layers upward.

Basis: DSP0236 8.7–8.10, DSP0240, libpldm base.h, CodeConstruct/mctp

Layer-by-layer debugging order from bus through MCTP, PLDM, and application

One-line answer: Over kernel AF_MCTP and mctpd, libpldm and OpenBMC pldmd handle PLDM messages, bmcweb exposes internal services as Redfish APIs, and PLDM Type 6 RDE lets devices exchange Redfish data.

Q19. How does it connect to OpenBMC and Redfish?
  • 30-second answer: OpenBMC has the PLDM encode/decode library libpldm and the OpenBMC PLDM daemon pldmd (openbmc/pldm), while the MCTP side uses the kernel AF_MCTP and mctp/mctpd (CodeConstruct/mctp); see How are MCTP and PLDM handled on Linux? AF_MCTP sockets, mctp, mctpd, libpldm and pldmd.
  • The bmcweb README describes it as a “DBus to Redfish translator”: data gathered by BMC internal services is exposed externally through the Redfish API.
  • PLDM Type 6 RDE (DSP0218, “PLDM for Redfish Device Enablement”) lets a device exchange Redfish data using BEJ encoding and a dictionary.

Basis: DSP0218, openbmc/pldm, openbmc/bmcweb, CodeConstruct/mctp

What should you avoid when answering?

One-line answer: Do not make up numbers, timers, or codes that are not in the specification; if unsure, state which specification document you would check.

  • Do not guess or invent unstated command codes or timer values.
  • Keep silent packet drops at the MCTP transport layer (DSP0236) distinct from explicit Completion Code errors at the PLDM layer (DSP0240).
  • When unsure of an exact parameter or field definition, state which specification document you would check: DSP0236 (MCTP Base), DSP0240 (PLDM Base), DSP0248 (Monitoring and Control), or DSP0267 (Firmware Update).

One-line answer: Read the companion posts covering MCTP/PLDM fundamentals, PLDM types, Linux sockets and daemons, and embedded serial bus comparisons.

What are the sources?

One-line answer: This post is based on the DMTF PMCI specifications and official OpenBMC, CodeConstruct, and Linux kernel repositories and documentation.