Method for dirty-page tracking and full memory mirroring redundancy in a fault-tolerant server

1. A method of transferring memory from an active to a standby memory in an FT (Fault Tolerant) Server system comprising the steps of:

Reserving a portion of memory using BIOS of the FT Server system;

Loading and initializing an FT Kernel Mode Driver into memory;

Loading and initializing an FT virtual machine Manager (FTVMM) including the Second Level Address Table (SLAT) into the reserved portion of memory and synchronizing all processors in the FTVMM;

Tracking the OS (Operating System), driver, software and Hypervisor memory accesses using the FTVMM's SLAT in Reserved Memory;

Tracking Guest VM (Virtual Machine) memory accesses by tracking all pages of the SLAT associated with the guest and intercepting the Hypervisor writes to memory pages that constitute the SLAT;

Entering Brownout—level 0 , by executing a full memory copy while keeping track of the dirty bits (D-Bits) in the SLAT to track memory writes by all software in the FT Server;

Clearing all of the D-Bits in the FTVMM SLAT and each Guest's current SLAT;

Entering Brownout of phases 1 - 4 tracking all D-Bits by:

Collecting the D-Bits;

Invalidating all processors cached translations for the FTVMM SLAT and each current Guest's SLAT;

Copying the data of the modified memory pages from the active memory to the second Subsystem memory;

Entering Blackout by pausing all the processors in the FTVMM, except Processor # 0 , and disabling interrupts in the FT Driver for processor # 0 ;

Copying the collected data from active to the mirror memory;

Collecting the final set of recently dirtied pages including stack and volatile pages;

Transferring the set of final pages to the mirror memory using an FPGA (Field Programmable Gate Array);

Using the FT Server FPGA and Firmware SMM (System Management Module) to enter the state of Mirrored Execution;

Completing the Blackout portion of the operation by terminating and unloading the FTVMM;

Returning control to the FT Kernel Mode Driver; and

Resuming normal FT System operation.

2. The method of claim 1 wherein FTVMM code remains in Reserved Memory following the resumption of normal FT System operation.

3. The method of claim 1 wherein the reserved memory is a memory region from ¼ GByte to 2.25 GByte of contiguous physical memory, located up to and including the maximum existing physical memory address in the FT Server.

4. The method of claim 1 wherein moving the data from the active memory uses an FPGA.

5. The method of claim 1 wherein the return of control to the FT Kernel Mode Driver is by the FT Kernel Mode Driver causing a VMEXIT event to occur.

6. The method of claim 1 wherein the FT Kernel Mode Driver returns to the condition of normal Guest VM operation on each processor by performing a VMPTRLD (Virtual Machine Pointer Load) instruction with the current guest VMCS (VM Control Structure) as an operand to make the current guest VMCS current and active.

7. The method of claim 1 where the collecting of D-bits is performed either periodically from the SLAT or continuously using Page Modification Logging (PML).

8. A method of transferring memory from an active to a standby memory in an FT (Fault Tolerant) Server system comprising the steps of:

Reserving a portion of memory using BIOS of the FT Server system at boot time;

Loading and initializing an FT Kernel Mode Driver;

Loading and initializing an FT virtual machine Manager (FTVMM) including the SLAT into the reserved portion of memory and synchronizing all processors in the FTVMM;

Tracking memory accesses using the SLAT in Reserved Memory;

Entering Brownout—level 0 , by executing a full memory copy while keeping track of the dirty bits (D-Bits) in the SLAT;

Clearing all of the D-Bits in the SLAT;

Entering Brownout of phases 1 - 4 tracking all D-Bits by:

Collecting the D-Bits;

Invalidating all processors cached translations for SLAT;

Copying the data of the modified memory pages from the active memory to the second Subsystem memory;

Entering Blackout by pausing all the processors in the FTVMM, except Processor # 0 , and disabling interrupts in the FT Driver for processor # 0 ;

Copying the collected data from active to the mirror memory;

Collecting the final set of recently dirtied pages;

Transferring the set of final pages to the mirror memory using an FPGA (Field Programmable Gate Array);

Using the FT Server FPGA and Firmware SMM to enter the state of Mirrored Execution;

Unloading the FTVMM and exiting into FT Kernel Mode Driver to cease operation of the FTVMM.

9. The method of claim 8 wherein synchronizing of all processors in the FTVMM causes FTVMM code to execute as hypervisor on all processors.

10. The method of claim 8 wherein the FTVMM tracks memory accesses using the SLAT in Reserved Memory.

11. The method of claim 8 wherein the full memory copy of system memory to the standby memory is performed by the FT Driver.

12. The method of claim 11 , wherein the copying the data of the modified memory pages from the active memory to the second Subsystem memory begins with the FTVMM obtaining a copy of all pages dirtied during the full memory copy by the FT Driver.

13. The method of claim 12 , wherein the FTVMM provides dirty page bit map data to the FT Driver.

14. The method of claim 13 , wherein the providing of the dirty page bit map to the FT Driver occurs for any pages dirtied during the previous copying of dirty page data by the FT Driver.

15. The method of claim 8 , wherein the SMM enters the state of Mirrored Execution by restoring Processor State from memory.

16. The method of claim 15 , wherein the SMM returns to the FT Driver after Processor State is restored, completing entrance into Mirrored Execution.

17. The method of claim 8 where the collecting of D-bits is performed either periodically from the SLAT or continuously using PML.