Processor structure and method for maintaining and restoring precise state at any instruction boundary

In a processor having means for issuing instructions, means for executing instructions, and a memory store for storing data within said processor, a method of restoring machine state in a processor at an instruction boundary;

• said method comprising the steps of:
• storing predetermined machine state parameters in registers in said data store of said processor prior to execution of a predetermined plurality of speculatively issued predicted instructions;
• monitoring instruction execution to detect an execution exception or fault condition;
• saving an instruction identifier of each instruction that results in said execution exception or fault condition;
• if more than one executed instruction results in an execution exception or fault condition, determining which of said more than one excepting or faulting instructions occurred sequentially earlier;
• using said earliest excepting or faulting instruction identifier as an endpoint to backup the program counter in said processor;
• if said earliest faulting instruction is a checkpointed instruction, then restoring said machine state information stored in said checkpoint prior to execution of said faulting instruction and decrementing the processor program counter to the instruction serial number of said faulting instruction to restore said prior machine state;
• and
• if one of said predetermined executable instructions for which a machine state information has been stored is sequentially interposed between said faulting instruction and the last issued instruction, then (i) first, backing-up said processor to the closest checkpoint to said faulting instruction after said faulting instruction, and (ii) second, backstepping said processor to restore processor state to the state that existed just prior to execution of said faulting instruction by updating register resources and (ii) decrementing the processor program counter to the instruction serial number of said faulting instruction.

A method of restoring machine state at an instruction boundary;

• said method comprising the steps of:
• for any instruction which may modify architectural control registers, including instructions which may have the side-effect of modifying control register values, choosing to either (i) synchronize said machine, or (ii) generate and store an instruction checkpoint in data storage structure within said processor to preserve machine state prior to execution of said instruction so that control register updates are not needed;
• for any instruction that creates a program counter discontinuity including any speculative instruction sequence, generating and storing an instruction checkpoint in data storage within said processor to preserve architectural and machine state prior to execution of said instruction so that program counter values correspond to instructions within an in-order sequence of instructions and so that the correct program counter can be reconstructed without re-issue and execution of the instruction that caused the fault;
• monitoring instruction execution to detect an execution exception or fault condition;
• saving an instruction identifier of each instruction that results in said execution exception or fault condition;
• if more than one executed instruction results in an execution exception or fault condition, determining which of said more than one excepting or faulting instructions occurred sequentially earlier;
• using said earliest excepting or faulting instruction identifier as an endpoint to backup the program counter in said processor;
• if said earliest faulting instruction is a checkpointed instruction, then restoring said machine state information stored in the checkpoint prior to execution of said faulting instruction and decrementing the processor program counter to the instruction serial number of said faulting instruction to restore said prior machine state;
• and
• if one of said predetermined executable instructions for which a machine state information has been stored is sequentially interposed between said faulting instruction and the last issued instruction, then (i) first, backing-up said processor to the closest checkpoint to said faulting instruction after said faulting instruction, (ii) second, backstepping said processor to restore processor state to the state that existed just prior to execution of said faulting instruction by updating register resources, and oil) decrementing the processor program counter to the instruction serial number of said faulting instruction.

A method of choosing which instructions in a machine are to be designated as requiring syncing of said machine before execution, said method comprising the steps of:

• determining the degree of speculation for said instruction.;
• determining the limits of logic complexity that it is desirable to implement including determining the maximum circuit area that it is desirable to allocated to implement the required logic that will be needed to handle all of the instructions speculatively;
• establishing predetermined trade-off parameters including a weighting factor for each said trade-off parameter;
• computing a performance index for instruction to determine if the instruction is one that should require syncing on the basis of said determinations and said criteria.

In a processor having means for issuing instructions, means for executing instructions, and a memory store for storing data within said processor, a apparatus for restoring a prior machine state in a processor at any instruction boundary comprising:

• means for allocating an instruction serial number for each issued instruction;
• means for storing machine state information in a checkpoint within said data store in said processor prior to execution only for a first predetermined set of executable instructions that alter machine state, said predetermined set of executable instructions that alter machine state being fewer than the set of all of the executable instructions that alter machine state implemented within said processor;
• means for identifying the instruction causing said fault, and if multiple exceptions or faults occurred simultaneously then identifying the in order sequentially earliest instruction causing a fault or exception;
• means for restoring said machine state information stored prior to execution of said faulting instruction and decrementing the processor program counter to the instruction serial number of said faulting instruction to restore said prior machine state if said earliest faulting instruction is a checkpointed instruction;
• means for backtracking said processor to an earlier state if one of said predetermined executable instructions for which a machine state information has been stored is sequentially interposed between said faulting instruction and the last issued instruction;
• said means for backtracking including means for backing-up said processor to the closest checkpoint to said faulting instruction sequentially after said faulting instruction;
• means for backstepping said processor to restore processor state to the state that existed just prior to execution of said faulting instruction by updating register resources;
• end means for decrementing the processor program counter to the instruction serial number of said faulting instruction.