+ +
- -
Systèmes d'Exploitation
Calendrier  Détails
Calendrier  Détails
Processeurs ARM
Calendrier  Détails
Processeurs PowerPC
Calendrier  Détails
Calendrier  Détails
+ +
> >
- -

ac6 >> ac6-training >> PowerPC System Design Télécharger le catalogue Télécharger la page Ecrivez nous Version imprimable

PG2 PowerPC System Design

This course has been designed for developers involved in a PowerPC development who want to understand generic mechanisms

  • This course explains the objectives of mechanisms used to boost the performance and the way they are implemented in various PowerPCs : cache / cache coherency, pipeline, MMU, exceptions.
  • This gives to the attendees a wider overview of the state of the art in these domains.
  • The course details the instructions required to write program in supervisor mode to adapt the behaviour of the core to specific needs.
  • Task switch requirements are highlighted.
  • Debug facilities implemented in PowerPCs (hardware breakpoints, real-time trace, watchpoints) are studied through the use of Lauterbach TRACE32 debugger.
A lot of programming examples have been developed by ACSYS to explain the PowerPC assembly language.
•  They have been developed with GNU compiler and are executed under Lauterbach debugger.
A more detailed course description is available on request at training@ac6-training.com
  • Basic knowledge of processor or DSP.

  • PowerPC programming environment : 32-bit PowerPC architecture, Book E, 64-bit architecture
  • Register set, GPR vs SPR, HID registers
  • Data type instantiation for PowerPC
  • Pointers management (Addressing modes)
  • User and supervisor functions call and return (EABI, C-to-assembly interface)
  • Sections, benefits of small data sections
  • Locating code and data in memory , linker command file
  • Reset, what is to be done before calling the main() : Cstart program
  • Superscalar operation
  • Mechanisms used to boost performance : branch prediction, branch target address cache, link stack
  • Guidelines to optimize execution time
  • Serializations, isync instruction, determining when this instruction is really required
  • Highlighting the frequency domains present in PowerPC : core and bus interface
  • Decoupling the core from cache and bus through load and store buffers
  • Default ordering of load and store transactions
  • Enforcing the ordering through eieio (called mbar in Book E) and sync (called msync in Book E) instructions
  • Purpose of the Guarded attribute
  • Consequence for high level development of IO drivers
  • Requirements for kernels enabling dynamic memory mapping
  • Single process multi-thread versus multiprocess multi-thread kernels
  • Objectives of the MMU : page protection, definition of page attribute, address translation
  • Segment and page translation
  • Table search mechanisms : benefits of a software table search
  • Operation of TLB caches
  • TLB programming, static initialization
  • Introduction to cache memory
  • Cache organization
  • Write policies
  • Replacement algorithms
  • Data flow between external main memory
  • Distinguishing private memory that is accessed only by the core and shared memory that can be accessed by the core and other masters (DMA or CPU)
  • Software enforced coherency
  • Hardware enforced coherency
  • Software exceptions vs interrupts
  • Save / restore registers
  • Organization of an exception handler : prolog, body and epilog
  • How to find the cause of the exception, syndrome registers
  • Design of a generic exception handler based on a vector table
  • Interrupt management, addition of a critical interrupt in Book E
  • Integrated interrupt controller
  • Requirements for interrupt nesting
  • Management of boolean semaphores, lwarx / stwcx. instruction pair
  • Stack switch, use of SPRG registers
  • Definition of the set of registers that determine the stack state
  • Management of task lists in single and multi processor systems
  • On-chip debug logic
  • Restrictions of JTAG debug
  • Hardware breakpoints
  • Real-time trace
  • Debugging software when caches are active
  • The performance monitor