Embedded Systems Design - Europe - August/September 2007 - (Page 33)

software contend for access to a capture buffer. To meet these performance requirements and minimize unwanted serialization, the TDR library should take advantage of “atomic” functions provided by the hardware and operating system. For example, in Linux, the atomic_set() and atomic_inc_return() functions can index into the buffer and freeze the buffer prior to initiating the dump. The function TDR_ThreadRemoveIdentity() is used when a thread exits normally. It removes the identifying information from the trace buffer so that its entries won’t be confused with a later instance of the same software. The TDR_BufferDestroy() function is used when the system shuts down normally. It removes the buffer from RAM and unregisters it with the TDR system. If a process/subsystem terminates without calling TDR_BufferDestroy(), the buffer becomes orphaned. The TDR subsystem should detect orphaned buffers but wait several minutes before destroying them. This waiting period gives the system a chance to recognize a catastrophic error that may be related to the orphaned buffer. Two mechanisms for recognizing an orphaned buffer include monitoring the operating system process that created the buffer and inspecting the buffer itself to see if it’s still being referenced. ORGANIZING THE BUFFERS The capture buffers reside in RAM until they’re dumped to nonvolatile storage. Each buffer must be accessible both from the process that creates it and the dump agent that generates the dump. For most RTOS environments, the buffers can be allocated dynamically from the heap when the system is started. For heavyweight tasking environments, the buffers may need to be placed into shared memory or mapped into the memory space of a device driver that performs the dump. Figure 2 illustrates two options for organizing the buffers and how the data is grouped within. Option 1 groups all of the data from the same process into one buffer. This means that if a thread calls a function in library #1 and then a function in library #2, each function places its data into the common capture buffer. For option 2, the data is grouped by module. If a thread calls a function in library #1 and then a function in library #2, each function places its data into a separate capture buffer dedicated to the library. For debugging, option 1 is frequently the best choice. Developers are most interested in tracing the logic as it executes on the system. Placing all entries from a process into the same buffer makes it easy to follow each thread’s execution and observe the ordering of inputs and relationships of actions and outputs. Option 2 can also provide this information, but requires developers to time-correlate entries in different buffers. This increases the amount of processing that must occur during debugging. The most elegant way to accomplish option 1 is to associate the buffer’s Nothing empowers performance like PowerMP! The PowerMP is designed to provide off-the-shelf, off-the-chart performance for your critical computing needs in all kinds of demanding environments. Each PowerMP system is a high-performance, low-cost COTS-based multiprocessor computing solution leveraging industry standards with Intel and/ or PowerPC architecture. It includes tools geared for such tasks as real-time performance analysis, remote control operations and monitoring system management. The PowerMP6 - an Intel “plenty-core”, ready-to-use solution When you place a premium on software productivity and performance turn to the turnkey computer system with dazzling performance—PowerMP6. The PowerMP6 is made of multiple Dual-Core Intel Xeon Processor-based boards in a 19-inch rack. Powered by Red Hat Enterprise Linux, it supports software productivity and portability through an extensive set of open source and commercial tools and libraries. PowerMP6 is available in various customized configurations of up to eleven processor boards in a single chassis… The PowerMP4 - RapidIO™ entry range system The PowerMP4 fills the embedded industry's need for reliability, increased bandwidth and faster bus speeds. It combines PowerPC (for legacy Altivec based signal processing) and Intel Architecture processors (for software productivity and connectivity). PowerMP4's RapidIO™ high performance and packet-switched interconnect technology will meet your demanding embedded system needs. The PowerMP9 - RapidIO™ PowerPC 64-bit system This member of the PowerMP family is geared towards high performance PowerPC computing with demanding memory bandwidth requirements. Featuring computing boards with a dual PowerPC 970 processor architecture, PowerMP9 flies high above competitor's solutions, thanks to the highest PowerPC memory bandwidth offered by this processor. This solution in its rugged conduction cooled version makes it the ideal embedded computers for airborne signal processing applications. It features Linux on PPC64 and can also host Gedae, a framework for data flow processing mapping on parallel computer hardware. The specialists in embedded performance Thales Computers is unequalled in the development of commercial and ruggedized VMEbus and CompactPCI systems solutions based on PowerPC and Intel microprocessors. Thales' products are optimized for a wide variety of applications in the military, aerospace, transportation, communications, and industrial markets and are used by blue chip customers worldwide, especially those who put a premium on long term support for computers. For more information please contact: Vincent Chuffart Tel: 33 (0)4 98 16 34 31 e-mail: vincent.chuffart@thalescomputers.fr www.thalescomputers.com www.embedded.com/europe | embedded systems design europe | AUGUST – SEPTEMBER 2007 33 030-031-032-033-034-035_ESDE.ind33 33 5/09/07 12:31:48 http://www.thalescomputers.com http://www.thalescomputers.com http://www.embedded.com/europe

Table of Contents Feed for the Digital Edition of Embedded Systems Design - Europe - August/September 2007

Contents News Cover Feature: Annual Study Uncovers the Embedded Market DSP Serves the Convergence Needs of Small Business Embedded Test Offers Unique Value for Serial I/O The Software Detective: First-Fault Data Capture Boards May Shrink But Performance Doesn't New Products Advertising Contacts

Embedded Systems Design - Europe - August/September 2007

For optimal viewing of this digital publication, please enable JavaScript and then refresh the page. If you would like to try to load the digital publication without using Flash Player detection, please click here.