ST67W611M1 T01 SDK 2.0.106 recurring Load Access Fault during TCP socket operation
We are observing an intermittent internal crash and automatic restart of the ST67W611M1 while using the T01 mission firmware with X-CUBE-ST67W61.
Configuration
Module: ST67W611M1
X-CUBE-ST67W61 middleware: v1.3.0
Mission profile: T01
AT version: 1.0.0
SDK version: 2.0.106
Wi-Fi MAC version: 1.6.44
BT Controller version: 1.6.128
BT Stack version: 1.10.83
Module build date: Mar 20 2026
Host MCU: STM32U5 series
Interface between STM32 and ST67W611M1: SPI
Architecture:
- Wi-Fi and TCP socket provided by ST67W611M1 T01 firmware
- TLS runs on STM32 using MbedTLS
- MQTT 3.1.1 runs on STM32
- X.509 authentication is handled on the STM32 side
Problem
During continuous TCP traffic, the ST67W611M1 intermittently generates a RISC-V exception, produces a QCC74x core dump, and restarts.
The recurring exception is:
exception_entry mcause=30000005 mepc:62fd0de2 mtval:842a0748 Load access fault
The corresponding ST67 firmware build ID is:
551e2178cb2e0e5f41ed380892ed67064c504657
The same fault signature has been reproduced multiple times.
QoS 1 and QoS 0 testing
Initially, the problem was reproduced while MQTT telemetry was using QoS 1.
In several occurrences the module crashed while the STM32 was waiting for network data associated with the MQTT acknowledgment path.
To determine whether the failure was related specifically to MQTT PUBACK processing, we changed telemetry publishing to QoS 0.
With QoS 0:
- No PUBACK is expected.
- The host does not wait for a PUBACK.
- MQTT keep-alive uses PINGREQ/PINGRESP.
- The same ST67 exception still occurs.
The QoS 0 test reproduced exactly the same recurring fault:
mcause=30000005 mepc:62fd0de2 mtval:842a0748
Therefore, our testing indicates that MQTT QoS 1/PUBACK is not required to reproduce the ST67 crash.
Because TLS and MQTT execute on the STM32 host, the ST67 only handles the underlying TCP/socket traffic for this connection.
Additional observation
We have also intermittently observed SPI interface messages similar to:
Invalid magic 0xa0d0
Invalid magic 0x0
These do not occur for every transaction and we have not established whether they are related to the internal ST67 exception.
We would appreciate confirmation of whether these SPI framing errors could indicate an SPI RDY/CS/timing synchronization problem that could contribute to the NCP failure.
NCP restart recovery
The STM32 detects the subsequent ST67 "ready" indication and recognizes that the module has restarted.
The host then invalidates the pre-restart TCP socket, waits for Wi-Fi recovery, and establishes a completely new TCP/TLS/MQTT session.
This recovery mechanism is working.
Our primary concern is identifying why the ST67 firmware itself generates the Load Access Fault.
Request for ST investigation
Could you please:
- Symbolicate the faulting instruction address:
mepc = 0x62FD0DE2
using the T01 SDK 2.0.106 ELF corresponding to build ID:
551e2178cb2e0e5f41ed380892ed67064c504657
- Identify the internal function/source location associated with this address.
- Advise what type of invalid memory access is represented by:
mcause = 0x30000005 mtval = 0x842A0748
for this particular firmware build.
- Confirm whether this is a known issue in T01 mission firmware SDK 2.0.106.
- Confirm whether a newer T01 mission firmware containing a relevant fix is available.
- Advise whether the observed SPI "Invalid magic" events could be related to this fault.
- Provide the recommended SPI clock, CS timing, SPI_RDY handling requirements and any other host-interface requirements for reliable T01 operation.
- Advise what additional crash-specific diagnostic information is required. We can provide a sanitized QCC74x exception/core-dump section without application payloads, cloud details, credentials, MQTT topics, or other project-specific information.
Fault Logs:
ST67W611M1
X-CUBE-ST67W61: 1.3.0
AT Version: 1.0.0
SDK Version: 2.0.106
Mission profile: T01
Build Date: Mar 20 2026
Fault:
exception_entry
mcause=30000005
mepc:62fd0de2
mtval:842a0748
Load access fault
build_id:
551e2178cb2e0e5f41ed380892ed67064c504657
QCC74x COREDUMP:
[include the crash dump here only after checking it for
credentials, payloads and project-specific strings]
