Non-intrusive trace
ETM generates trace packets in hardware, in parallel with core execution. No code instrumentation, no probe effect, no timing distortion.
Non-intrusive hardware trace acquisition for Arm®-based SoCs
ChipTrace uses the Arm® CoreSight™ on-chip debug infrastructure (ETM, ETF, CSTF, ETR, TPIU) to capture instruction-level execution traces at full CPU clock speed. The trace path is purely hardware: zero instrumentation, zero CPU overhead, cycle-accurate timestamps.
Ref: Arm IHI 0029E — CoreSight Architecture Specification v3.0
Arm, Cortex and CoreSight are trademarks or registered trademarks of Arm Limited (or its subsidiaries or affiliates) in the US and/or elsewhere. ChipTrace is an independent product and is not affiliated with or endorsed by Arm Limited.
ETM generates trace packets in hardware, in parallel with core execution. No code instrumentation, no probe effect, no timing distortion.
Trace data is output on the ATB (AMBA Trace Bus) at core clock speed. No sampling, no dropped packets when ETF buffering is correctly sized.
ChipTrace configures the on-chip CoreSight components directly via APB register access: ETM,STM, TMC (ETF,ETB,ETR), CSTF funnels, Replicator, Timestamp genererator. No proprietary hardware required.
ChipTrace integrates with Lauterbach TRACE32 via PRACTICE scripts for CoreSight path setup. Configuration follows the Lauterbach app note "Setup of the Debugger for a CoreSight System" (app_arm_coresight.pdf).
ASIL-D projects on S32G gateway SoCs require proof of deterministic execution paths. Hardware trace provides objective evidence of code coverage and worst-case execution time without altering the test conditions.
Certification authorities accept hardware trace as structural coverage evidence (MC/DC, statement, branch). The non-intrusive nature of ETM trace means test results are not invalidated by probe effects.
Measure actual context switch latency, ISR entry-to-exit time, and lock hold duration from the trace stream. Results are cycle-accurate and obtained on the production binary, not an instrumented build.
On heterogeneous SoCs (e.g. S32G2: A53 + M7), ChipTrace correlates traces from all cores using the shared TSG timestamp. Inter-core communication sequences (shared memory, mailbox interrupts) appear on a single unified timeline.
| Parameter | ChipTrace (ETM/ETR hardware trace) | Software instrumentation |
|---|---|---|
| Capture overhead | No CPU cycle cost, ETM snoops the pipeline in parallel. Bound by AXI bus bandwidth to the ETR and buffer drain rate, not by a physical trace port | 2 to 15% depending on instrumentation density (probe effect) |
| Timing resolution | Cycle-accurate via the Time Stamp Generator, provided the trace clock source is calibrated | Microsecond range (SysTick resolution) |
| Granularity | Instruction-level, every branch captured | Function-level, depends on where instrumentation points are placed |
| Capture depth / continuity | Configurable buffer topology: circular (snapshot on trigger), FIFO (continuous capture gated by drain path), or scatter-gather (extends depth across discontiguous memory when contiguous DRAM isn't available) | Depends on log buffer sizing and storage medium chosen by the instrumentation |
| Source code access | Capture works on stripped binaries; meaningful analysis still needs symbols | Required at build time to insert instrumentation points |
| RTOS/OS awareness | Requires tracing CONTEXTIDR correlated with OS symbols in the host-side analysis tool | Native if instrumentation is placed in the kernel/scheduler |
Network processor
4x A53 (1.6 GHz)
Full CoreSight topology
LS1043A RM, debug ch.
Validated
Applications processor
2-4x A35 (1.2 GHz) + M4F
CoreSight ETM, funnel, TPIU
i.MX 8QXP RM, debug ch.
Validated
High-performance MCU
M7 (480 MHz) + optional M4
ETM, ITM/DWT, SWO
RM0433, debug ch.
Validated
SoC FPGA (ZCU102/106)
4x A53 (1.5 GHz) + 2x R5F (600 MHz)
ETM, ETF, funnel, TPIU, FTM
Zynq US+ TRM UG1085
Validated