by Alex Rozenman, Vladimir Pilko, and Nilay Mitash, Mentor Graphics
Introduction
With the SoCs now supporting Multi-Core processors, complex ASICs and combinations that include systems on a board, SoC implementations now become an ever growing challenge for software development. Software development has to be supported not only by the inclusion of an RTOS, but, many SoC providers now have to leverage upon the Bare-Metal concept to achieve the necessary demands of today's SoCs. However, there still exists a huge chasm in the software development arena that addresses both the need to be able to verify not only the sw/hw interactions, but, also the software itself in a hardware context. This has become almost a necessity in today's "security" based systems marketplace.
As large and complex SoCs trend deeper into embedded environments, they need to be seamlessly implemented and contained, for verification of both software and hardware in the same context. The following sections cover the underlying technologies used in context to each of the areas and explain the mechanisms for achieving verification of the software itself.
What Is Aspect Oriented Software Verification?
In almost every intelligent applicable software function, there are always locations that have meaningful implications. A reflective functional equivalent of these locations would create the verification logic. For example, an assertion in the code can be reflectively mapped to its verification logic, making sure that the conditions that cause the assertion to fire are captured in the verification logic. Each such verification logic is strongly anchored to its aspect in the original software.
From AOP (Aspect Oriented Programming) point of view, the verification locations can be seen as "join points", verification logic as a set of "advices" and the verification script as the AOP program itself.
The following code locations/anchors that are found in almost all software code could be used:
- Absolute execution addresses (PC)
- Source locations (file:line)
- Function entry/exit
- State change points (watch points)
- Peripheral device accesses and programming
- Interrupts
- CPU exceptions
In typical software, verification logic may include assertions, invariant conditions checkers, print outs for flow tests, fault injection or an external stimulus. We will use the term "verification script" to denote a program that attaches the verification logic to its corresponding locations.
The meta-information generated from a standard compilation of the target software is used by the verification scripts. The verification scripts commonly have access to the meta-information. The meta-information access allows it to explore the software structure and attach advices (verification logic) to the selected locations.
Differentiating Aspect-oriented Software Verification From Aop In Hardware Verification Environments
Using an aspect-oriented approach to verifying software allows the injection of additional verification code without altering the basic execution of the software itself, from the perspective of the system. As such, this approach has the advantage of being orthogonal to the functionality being verified. This is different from the typical view of hardware verification, where we build a testbench to model the environment in which the hardware will operate.
One value in using an approach like UVM for developing a verification environment for hardware is that you can use object-oriented techniques to modify the functionality and structure of your testbench without having to rewrite your code. Using AOP in this context is not recommended, nor required, because the injected code directly affects the functionality of testbench without providing sufficient visibility to the user.
Enabling Non-intrusive Jit Based Aspect Oriented Verification Environment
In a QEMU based processor modeling technology (ISS) based on JIT (Just-in-Time) compilation infrastructure, the JIT compiler is a binary translation tool, which is able to convert target binary code (e.g. ARM) into a host binary code (x86). The conversion is performed block by block and on demand. When the target execution reaches an instruction for the first time, only then, a short sequence of instructions is translated to the host binary code. The size of the compilation depends on many different factors, for example, blocks cannot include jump instruction, except, only at the end.
Running target software under JIT based processor model allows injecting verification logic non-intrusively and exclusively into the simulator context. The non-intrusive injection implies that verification logic does not affect any aspect of target behavior, target timing and power. The execution order of the target code is also left unmodified even when modeled with multiple threads or multi-core execution. Target debug information (DWARF: a widely used, standardized debugging data format generated by software compilers) is used in order to explore the software structure and provide reflection information to the verification scripts.
The aspect oriented JIT compiler provides a non-intrusive way of creating various test enablers. Some of the direct advantages are the capability to use this technology for tracing function call, enabling specific profiling, targeting coverage, and, inserting tracepoints (unlike breakpoints, tracepoints are software 'triggers' that do not stop the simulation). Tracepoints can be used to inject verification logic into arbitrary places in the target software. By accessing the DWARF debug data that contains information on the software structure, types, objects and source locations, the source code lines and function names can be converted into the needed addresses. A small "C" compiler is used to capture the tracepoint actions specified by "C" programs.
Applying The Aop Verification As An Advanced Software Unit Test Environment
As the software content in systems grows, more often than not, it is the software components itself that need to be verified while in in-system execution, i.e., without disturbing the overall functionality. Software based aspect oriented verification capabilities best suit this need to create a unit test environment.
Typical software unit test environment requires the ability to execute specific functions in specific predictable environments, be able to randomize arguments and state, collect execution results, compare them with a predefined gold values, and, all the while collect code coverage during the entire process. There is a very big intersection between this requirement and the abilities provided by a JIT based aspect oriented verification environment. The JIT based aspect oriented verification methodology is then able to initiate the needed unit test flow by injecting the needed advices and the verification logic that entirely replace the normal execution flow. Software coverage is almost a by-product of this methodology.
Figure 1: This shows a snapshot of all the function calls and their time duration in the Mentor software profiler
 |
Using Jit And Aop For Software Trace And Analysis
Graphical viewers and visualization is a powerful tool for better understanding of software execution flow in a greater level of detail. JIT based aspect oriented verification environments allow to trace many kinds of software and hardware events non-intrusively without any need in code instrumentation. The following capabilities can be very efficiently traced:
- Function entry and exit to show the call graph
- CPU state (showing IDLE/BUSY/exceptions)
- Caches/Snoops state and activity
- Peripheral state and activity
Graphical analysis of both hardware and software together provides broad opportunities to find bottle necks, performance and power issues in a complex SoC system. Cache and snoop analysis can identify cache ineffective code fragments, false sharing and even hardto- debug software bug related to cache control. CPU state visualization together with function calls can help to identify interrupt latency bottle necks, unneeded preemption and dead locks in the code. Peripheral tracing may help to debug driver code, analyze bus contention and power issues.
As the simulation proceeds, the captured data can be displayed in a graphical format that shows the timelines of execution and their results. In Figure 1, the top portion (Function Calls) shows the calls to various functions as different compiled instructions for the CPU are executed. The state of the CPU is also captured, to show that the CPU is actually BUSY or IDLE during that time, shown by header CORE_STATE. The ISR plot shows when the Interrupt Service Routine triggered along with the SysTick provided by the ARM SystemC Fast Model. These plots enable intelligent debug of the software execution on the CPU.
Using Jit And Aop For Software Function Profiling
Another very important technique used in software verification is profiling. Profiling may be performed with respect to time, cache-hit, cache-miss, software-hardware execution and power. Software function profiling constitutes of tracing/profiling of instructions across single functions, functions stacks, and stack fragments.
A typical profiling analysis report contains details of functions self-time (i.e. amount of time consumed by instructions which belongs to the function itself) and functions total-time (i.e. actual duration of the function execution from entry till exit). This type of profiling is slightly enhanced, since it not only covers the execution of the function in the CPU context, but, also with respect to the effects of the instruction to and its return from an I/O.
Profiling enables the software developer to visualize the round-trip and nested function calls, and, its effects of a particular instruction executed on the target. Other traditional areas of software profiling, for example, Multi- Core processor execution, time and overlap of shared functions, are still available, and remain software centric.
JIT based aspect oriented verification environment allows to perform the software profiling completely non-intrusively requiring very small attention from the user. All the needed information can be automatically extracted from the meta-data available for the verification script. Ability to perform function profiling without explicit recompilation that also keeps the software semantics unmodified becomes a unique feature available for customers.
This plot clearly shows when a main_loop call made recursive calls, for example, to cs3_isr_ systick, which in turn calls udivdi3 functions, and when they return. After the final call to write happened, the recursive call ended and returned to the main_loop function.
Figure 2: A detailed analysis profiled for nested function calls.
 |
Enabling Software Coverage
Coverage in Software has been around for many years. Software code coverage is used to determine the quality of test-benches, flow tests, unit tests and the tested code itself. In addition it can be used to highlight "dead code", a major requirement in security based verification. The following types of coverage are well known:
- Instruction: Checks if instructions were executed
- Branch: Checks if conditional branches were both taken and not taken
- Block: Checks if lexical blocks were executed
- Path: Checks if all possible block combinations are executed
- Logical expression: Check if all possible combinations of logical conditions were executed
Simple evaluation of structural coverage is not sufficient. To this effect, coverage also needs to include MC/DC (modified condition/decision coverage). MC/DC will provide an additional means of verifying software coverage in security sensitive applications.
All coverage information can be collected and assimilated with the hardware coverage information, thus, providing a comprehensive coverage of the entire sw/hw system.
Conclusion
In today's verification environment, software verification still lags behind the hardware verification. Hardware verification has a proven and beaten path pervasively adopted in the form of UVM.
Software verification tends to be a collection of in-house and vendor tools, with all its intellectual property focused on the creativity of Software Validation Engineers. However, software verification is quickly becoming one of the biggest challenges for complex ASIC platforms, SoCs and systems. In many cases, the SoCs and systems platforms are rated as life-threatening, such as medical devices, lethal and destructive, such as smart armaments, affect our daily lives, such as automotive and transportation, or data handlers, such as networking applications. In all of these, software is quietly becoming a dominant design-in and reliability factor. To set right a misalignment of directional vector, after deployment, of say a satellite in space, the only means available may be a pure software intervention.
It is important to note, that, the article has attempted to address the need for software verification by taking the route of aspect oriented methodology, but, clearly differentiates the needs for hardware verification from this approach. The objective of this article is to enable the flexibility of software aspect oriented verification for complex platforms that get affected by any intrusive functions or logic, and can lead to fatal exceptions in normal software execution.
Mentor Vista tools have taken a bold step in understanding these requirements and providing tools and methodologies that will enable a software developer to release products with confidence and assurance.
References:
- Aspect oriented Programming: https://en.wikipedia.org/wiki/Aspect-oriented_programming
- Mentor Vista tools
- Mentor Embedded CodeSourcery tools
Back to Top