Recovery unit for mirrored processors

Claim 3

A method as recited in claim 1, wherein said recovery sequence includes the steps of reading data subsets from the last checkpointing step; regenerating ECCs for the data subsets; correcting any errors detected by said ECC regeneration step; updating working registers of each processor with the corrected data subsets; rewriting the corrected data subsets back into checkpoint arrays rereading the rewritten data; rechecking, by ECC circuitry, said rewritten data; and effecting a check-stop operation if a previously corrected error persists.

Claim 4

A method as recited in claim 1, wherein each of said current state registers having a plurality of rows and latches, further comprising the step of selectively disabling of clocks of the current state latches to save power by using a decoded address input to the current state registers.

Claim 5

A method as recited in claim 3, further including the step of selectively disabling the decoders as determined by valid bits associated with the address information.

Claim 6

A method as recited in claim 1, wherein the checkpointing step further includes the step of selectively disabling of clocks of the checkpoint state latches to save power by using a decoded address input to the checkpoint state registers.

Claim 7

A method as recited in claim 5, further including the step of selectively disabling the decoders as determined by valid bits associated with the address information.

Claim 8

A method as recited in claim 1, wherein each said checkpoint state register portion in each partition is augmented by a single bit register associated with said checkpoint state register portion, said register having its inverted output connected to its input, said register initially containing a value of zero and thereafter providing a modulo-2 count of the number of times that said checkpoint state register portion has been stored to.

Claim 9

A method as recited in claim 7, further including the step of comparing the counter outputs from both partitions for each checkpoint state register in the checkpoint state array to determine whether both partitions have performed identical checkpointing operations.

Claim 10

A method as recited in claim 8, further including the step of checkstopping the processor if the checkpointing operations in both partitions are determined to be different.

Claim 11

A recovery unit for a processor system including first and second identical processors running identical code in lock-step mode, comprising: one or more of checkpoint registers, each of said checkpoint registers having first and second partitions; means for separately storing identical operand data received from each of two processors; means for initiating a fail-safe, no-fault error detection and recovery procedure, said creating means further comprising means for partitioning the operand data received from each of said first and second processors into respective first and second pluralities of data subsets, means for generating error correction codes (ECCs) for each data subset of said first and second pluralities of data subsets, and means for bit interlacing the data subsets and corresponding ECCs to form an interlaced entity for each of said first and second processors, and means for dividing each interlaced entity into first and second partitions and loading the divided interlaced entities into corresponding partitions of first and second sets of current state registers; and means for concluding said fail-safe, no-fault error detection and recovery procedure, said concluding means further comprising means for comparing corresponding portions of the divided interlaced entities of each of said first and second processors, means for checkpointing, if no errors are detected by said means for comparing, said first partition of an interlaced entity from said first processor stored in a first partition of said first set of current state registers into a corresponding first partition of a checkpoint register sniffed by address data from said first processor, and checkpoint said second partition of the interlaced entity from the second processor stored in a second partition of said second set of current state registers into said second partition of the checkpoint register specified by said address data from said second processor, wherein said address data from said first and second processors is known to have been equal at the time of the comparison is made by said means for comparing, and means for recovery initiated by said means for comparing if errors are detected by said means for comparing.

Claim 12

A recovery unit for a processor system including first and second identical processors running identical code in lock-step mode, comprising: one or more of checkpoint registers, each of said checkpoint registers having first and second partitions; means for separately storing identical operand data received from each of two processors; means for initiating a fail-safe, no-fault error detection and recovery procedure, said creating means further comprising means for partitioning the operand data received from each of said first and second processors into respective first and second pluralities of data subsets, means for generating error correction codes (ECCs) for each data subset of said first and second pluralities of data subsets, and means for bit interlacing the data subsets and corresponding ECCs to form an interlaced entity for each of said first and second processors, and means for dividing each interlaced entity into first and second partitions and loading the divided interlaced entities into corresponding partitions of first and second sets of current state registers; and means for concluding said fail-safe, no-fault error detection and recovery procedure, said concluding means further comprising means for comparing corresponding portions of the divided interlaced entities of each of said first and second processors, means for checkpointing, if no errors are detected by said means for comparing, said first partition of an interlaced entity from said first processor stored in a first partition of said first set of current state registers into a corresponding first partition of a checkpoint register sniffed by address data from said first processor, and checkpoint said second partition of the interlaced entity from the second processor stored in a second partition of said second set of current state registers into said second partition of the checkpoint register specified by said address data from said second processor, wherein said address data from said first and second processors is known to have been equal at the time of the comparison is made by said means for comparing, and means for recovery initiated by said means for comparing if errors are detected by said means for comparing.

Claim 13

A recovery unit as recited in claim 10, wherein said recovery means includes an instruction unit state machine and an execution unit state machine cooperating with the recovery unit.

Claim 14

A recovery unit as recited in claim 10, wherein said means for checkpointing includes means for latching each of said current state registers with a corresponding plurality of checkpoint registers, and wherein said current state registers and checkpoint registers are doubleword registers.

Claim 15

A recovery unit as recited in claim 12, further including means for comparing information in said current state registers and checkpoint registers of each processor.

Claim 16

A recovery unit as recited in claim 10, further including means for gating clocks of each row of current state register latches of each processor.

Claim 17

A recovery unit as recited in claim 14, further including counting means coupled to each said register wherein said counting means is incremented whenever the register to which it is coupled is stored into.

Claim 18

A recovery unit as recited in claim 15, further including means for comparing counter outputs generated by the gated clocks for each row of registers.

Claim 19

A recovery unit as recited in claim 13, further including means for gating clocks of checkpoint state latches in the checkpoint state registers.