e5500 CORE OVERVIEW
- Highlighting data path and instruction path
- Changes from e500mc to e5500
e5500 HYPERVISOR STATE
- Processor privilege levels state machine, user, guest OS, hypervisor
- Bare-metal operation
- Collaboration between guest OS and hypervisor to reload TLBs
- Directed interrupts
- Messaging within a coherency domain
- Filtering incoming messages
PIPELINE
- Instruction pipeline operation, dual issue, out of order execution
- Issue queue resource requirements
- Dispatch conditions, completion conditions
- Execution and context serializations, purpose of the isync instruction
- Branch management: dynamic prediction, BTB
- Link stack
- Segment target index cache (STIC) and segment target address cache (STAC)
- Guarded memory
DATA AND INSTRUCTION PATHS
- Implementation of a spin lock routine
- Decorated storage facility
- Memory barriers
- List insertion in a multicore system
COMPUTATION MODES
- Selecting 32-bit thread mode or 64-bit thread mode
- Computing effective addresses
- 64-bit arithmetic instructions
FLOATING POINT UNIT
- FPU operation: FPSCR register, IEEE vs non-IEEE mode
- Float load / store instructions
- Float arithmetic instructions
- Convert instructions
- Fully pipelined FPU
THE EXCEPTION MECHANISM
- Exception management: building the handler table through IVPR,IVOR registers
- Recoverable vs non recoverable exceptions
- Requirements to support exception nesting
- Exception priorities
- Interrupt proxy
- Multicore exceptions, doorbells and messages
- Integrated timers
- Reset sequence, initialization requirements
THE MEMORY MANAGEMENT UNIT
- MMU objectives definition
- Address translation, understanding the interim 48-bit virtual address
- Process protection through TID
- Two-level MMU architecture, level-1 TLBs and level-2 TLBs
- TLB organization, TLB software management, MAS registers
- Software TLB reload, clarifying the hardware assistance to select the victim in L2TLB0
- Managing a page descriptor table in a SMP system, tlbivax instruction
- Virtualization fault, managing the MMU at hypervisor level
- External PID load and store instructions
- TLB parity protection, multiple-hit detection
L1 AND L2 CACHES, SNOOPING
- e5500 L1 cache
- L2 cache organization
- Hit under miss and miss under miss
- Store miss merging
- Dynamic Harvard implementation
- Write shadow mode
- MESI snooping sequences involving two e5500 and a PCI Express master
- Data & instruction prefetch instructions
- Cache entry locking
- Stashing capability
- L1 and L2 error checking and correction, error injection
DEBUG
- Performance monitor
- Nexus debug unit
- Instruction and data breakpoints, programming address ranges
- Debug data acquisition message
- Debug Notify Halt instruction
- Nexus trace
POWER MANAGEMENT
- Connection to platform PM unit
- Power states
- Wake-up interrupt