Methods, apparatuses, and systems for zero silent data corruption (ZDC) compiler technique
Methods, apparatuses, systems, and implementations of a zero silent data corruption (ZDC) compiler technique are disclosed. The ZDC technique may use an effective instruction duplication approach to protect programs from soft errors. The ZDC may also provide an effective control flow checking mechanism to detect most control flow errors. The ZDC technique may provide a failure percentage close to zero while incurring a lower performance overhead than prior art systems. The ZDC may also be effectively applied in a multi-thread environment.
1. A computing system configured to detect and reduce silent data corruptions, the computing system comprising:
at least one memory device;
at least one processor; and
at least one physical computer readable medium coupled to the at least one memory device, the at least one physical computer readable medium comprising computer executable instructions that when executed by the at least one processor are configured to:
perform a store instruction checking operation, wherein the store instruction checking operation comprises instructions to reload a value of a store instruction into a value register as a checking load instruction, and compare the checking load instruction with the store instruction;
perform a load instruction duplication operation, wherein the load instruction duplication operation comprises instructions to duplicate a load instruction from a load store queue;
perform a compare instruction and branch instruction duplication operation, wherein the compare instruction and branch instruction duplication operation comprise instructions to duplicate a compare instruction, save the compare instruction as a value in a compare destination register and a value in a compare check register, duplicate a branch instruction, and conditionally invert the value of the compare destination register based on a direction of the branch instruction and the duplicate of the branch instruction;
perform a branch direction checking operation, wherein the branch direction checking operation comprises instructions to check a basic destination block by comparing one or more static signatures;
perform a register file checking operation, wherein the register file checking operation comprises instructions to compare a value in a shadow register with a master register after a store instruction; and
in response to determining an inconsistency during execution of the instructions, transfer execution to a separate diagnostic block to correct any error or inconsistency.
2. The computing system of claim 1 , where the executable instructions are further configured to:
compare a load destination register value with a shadow load destination register value;
determine whether the load destination register value matches the shadow load destination register value; and
execute a load diagnostic process when the load destination register value does not match the shadow load destination register value.
3. The computing system of claim 2 , where the load diagnostic process comprises instructions to:
check a base address register and a shadow base address register for errors; and
avoid a propagation of a detected error to a storage device.
4. The computing system of claim 3 , where the load diagnostic process comprises instructions to:
re-execute redundant load instructions when an error is not detected.
5. The computing system of claim 1 , where the executable instructions are further configured to:
execute a swap instruction to remove data present in a memory storage location into a register and write a register value into the memory storage location;
compare a store destination register value with a shadow store destination register value;
determine whether the store destination register value matches the shadow store destination register value; and
execute a store diagnostic process when the store destination register value does not match the shadow store destination register value.
6. The computing system of claim 5 , where the store diagnostic process comprises instructions to:
check a base address register and a shadow base address register for errors; and
avoid a propagation of a detected error to a storage device.
7. The computing system of claim 6 , where the store diagnostic process comprises instructions to:
re-execute the swap instruction when an error is not detected.
8. A method of detecting and reducing silent data corruptions, the method comprising:
performing a store instruction checking operation, wherein the store instruction checking operation comprises instructions to reload a value of a store instruction into a value register as a checking load instruction, and compare the checking load instruction with the store instruction;
performing a load instruction duplication operation, wherein the load instruction duplication operation comprises instructions to duplicate a load instruction from a load store queue;
performing a compare instruction and branch instruction duplication operation, wherein the compare instruction and branch instruction duplication operation comprises instructions to duplicate a compare instruction, save the compare instruction as a value in a compare destination register and a value in a compare check register, duplicate a branch instruction, and conditionally invert the value of the compare destination register based on a direction of the branch instruction and the duplicate of the branch instruction;
performing a branch direction checking operation, wherein the branch direction checking operation comprises instructions to check a basic destination block by comparing one or more static signatures;
performing a register file checking operation, wherein the register file checking operation comprises instructions to compare a value in a shadow register with a master register after a store instruction; and
in response to determining an inconsistency during execution of the instructions, transfer execution to a separate diagnostic block to correct any error or inconsistency.
9. The method of claim 8 , further comprising:
comparing a load destination register value with a shadow load destination register value;
determining whether the load destination register value matches the shadow load destination register value; and
executing a load diagnostic process when the load destination register value does not match the shadow load destination register value.
10. The method of claim 9 , where the load diagnostic process comprises:
checking a base address register and a shadow base address register for errors; and
avoiding a propagation of a detected error to a storage device.
11. The method of claim 10 , where the load diagnostic process comprises:
re-executing redundant load instructions when an error is not detected.
12. The method of claim 8 , further comprising:
executing a swap instruction to remove data present in a memory storage location into a register and write a register value into the memory storage location;
comparing a store destination register value with a shadow store destination register value;
determining whether the store destination register value matches the shadow store destination register value; and
executing a store diagnostic process when the store destination register value does not match the shadow store destination register value.
13. The method of claim 12 , where the store diagnostic process comprises:
checking a base address register and a shadow base address register for errors; and
avoiding a propagation of a detected error to a storage device.
14. The method of claim 13 , where the store diagnostic process comprises:
re-executing the swap instruction when an error is not detected.
15. An apparatus for detecting and reducing silent data corruptions, the apparatus comprising: at least one processor configured to:
perform a store instruction checking operation, wherein the store instruction checking operation comprises instructions to reload a value of a store instruction into a value register as a checking load instruction, and compare the checking load instruction with the store instruction;
perform a load instruction duplication operation, wherein the load instruction duplication operation comprises instructions to duplicate a load instruction from a load store queue;
perform a compare instruction and branch instruction duplication operation, wherein the compare instruction and branch instruction duplication operation comprises instructions to duplicate a compare instruction, save the compare instruction as a value in a compare destination register and a value in a compare check register, duplicate a branch instruction, and conditionally invert the value of the compare destination register based on a direction of the branch instruction and the duplicate of the branch instruction;
perform a branch direction checking operation, wherein the branch direction checking operation comprises instructions to check a basic destination block by comparing one or more static signatures;
perform a register file checking operation, wherein the register file checking operation comprises instructions to compare a value in a shadow register with a master register after a store instruction; and
in response to determining an inconsistency during execution of the instructions, transfer execution to a separate diagnostic block to correct any error or inconsistency.