Post a Comment
News and New Products
Breaking the time barrier
By Robert Cravotta -- EDN, 10/2/2003
Sometimes, host debugging and single-stepping through your software code is insufficient to uncover the reason that your embedded system is failing. Some failures require you to observe the system under real-time operating conditions to track down the cause. In these situations, it is usually necessary to understand what the system was doing just before the failure becomes apparent. A capability you need for debugging real-time anomalous behavior is to capture and store your processor’s execution data at and before the moment the anomalous event occurs within your system.
A real-time execution trace buffer allows you to capture what your processor is doing on a cycle-by-cycle basis. Trace memory can exist as an on-chip memory block or as an off-chip buffer as part of a debugging probe or an in-circuit emulator. The size of an on-chip trace buffer is generally limited to a few thousand bytes because of the expense of including the trace memory in the end device. Many external trace-buffer options offer as much as a few megabytes of buffer space. To capture the execution data you most need to analyze within the trace buffer, you may need to tune your trigger criteria for capturing the processor activity over several execution iterations. With smaller trace buffers, you may have to spend more iterations and effort to properly tune your trigger criteria. The Green Hills SuperTrace Probe greatly expands the amount of trace data that you can capture and analyze by providing a 1-Gbyte buffer that can accept and store hundreds of millions of processor execution cycle data in real time at clock speeds exceeding 300 MHz.
Another capability you need for debugging real-time anomalous behavior is to meaningfully analyze the execution trace data. Debugging software can usually associate the trace-buffer data with assembler and source-code statements to provide a low-level analysis of the trace data. The Green Hills TimeMachine software adds a new twist to analyzing your system using the captured trace-buffer data - you can step through and run your program code backward. It also allows you to examine registers, variables, or memory locations, use complex execution and data breakpoints, and run and step forward through your program as though you were operating on a live target with RTOS awareness and MMU protection.
Together, TimeMachine and the SuperTrace probe can provide you higher level analysis of your system without intrusive instrumentation by reconstructing the system operation from the trace data. It can graphically display function execution over time as well as operating-system events (with the Integrity RTOS only), such as kernel-service calls, interrupts, exceptions, and context switches. The trace-display tool enhances the source-code correlation by supporting high-level state-machine-transition triggers and by providing virtual execution to maintain an exact correlation between the low-level machine instructions and high-level debugging. The trace display tool can also derive from the trace data a nonintrusive analysis for execution time, call count, call graph, and code coverage of the system.
With the large trace buffer and virtual-program execution from the trace-buffer data, a team of developers could simultaneously analyze the same execution scenario from different perspectives. Green Hills will begin shipping the SuperTrace Probe and TimeMachine software, including the Multi debugger, in October for $9900 and $17,900, respectively. Both products support ARM 7, 9, and 10 processors with ETM (embedded-trace-macrocell) ports and PowerPC405 and 440 processors.
Green Hills Software, 1-805-965-6044, www.ghs.com.
