Processor, computer system, and method for flushing hierarchical cache structure based on a designated key identification code and a designated address

This application claims priority of China Patent Application No. 202111374225.X, filed on Nov. 19, 2021, the entirety of which is incorporated by reference herein.

The present application relates to management technology for a hierarchical cache structure of a processor.

In a computer system, memory devices may be classified into tiers. The higher-level memory has higher speed, lower latency, but lower capacity. The memory hierarchy of most computer systems has the following four levels (ordered from top to bottom): registers; caches; a system memory (a main memory, such as a DRAM); and disks (SSD or HD).

In particular, caches may also be arranged hierarchically. From the high-access speed to the low-access speed, the caches include: the level 1 cache (L1); the level 2 cache (L2); and the level 3 cache (L3, also known as the last level cache, or LLC for short). The management of the hierarchical cache structure will significantly affect system performance.

In order to protect confidential and sensitive data, a total memory encryption technology has been developed to use different keys to encrypt different parts of a system memory. Thus, the management of the system memory depends on the keys (in granularity of keys). The management with granularity of keys, however, is not applied to the hierarchical cache structure, so an operating system (OS) is incapable of managing the hierarchical cache structure in granularity of keys.

This case proposes a management technology that manages a hierarchical cache structure in granularity of keys.

A processor in accordance with an exemplary embodiment of the present application includes a first core, and a last-level cache. The first core includes a decoder, a memory ordering buffer (MOB for short), and a first in-core cache module. In response to an Instruction Set Architecture (ISA) instruction that requests to flush a hierarchical cache structure according to a designated key identification code and a designated address, the decoder outputs at least one microinstruction. According to the at least one microinstruction, a flushing request with the designated key identification code and the designated address is provided to the first in-core cache module through the memory ordering buffer, and then the first in-core cache module further provides the flushing request to the last-level cache. In response to the flushing request, the last-level cache searches itself for a matching cache line that matches the designated key identification code and the designated address, and flushes the matching cache line.

A detailed description is given in the following embodiments with reference to the accompanying drawings.

The present application can be more fully understood by reading the subsequent detailed description and examples with references made to the accompanying drawings, wherein:

FIG. 1 shows an instruction format for the ISA instruction CFLUSHKEYID in accordance with an exemplary embodiment of the present application;

FIG. 2 illustrates a cache line format 200 of a hierarchical cache structure in accordance with an exemplary embodiment of the present application;

FIG. 3 is a block diagram illustrating a processor 300 and a core core_ 1 thereon in accordance with an exemplary embodiment of the present application; and

FIG. 4 illustrates a computer system 400 in accordance with an exemplary embodiment of the present application.

The following description is made for the purpose of illustrating the general principles of the invention and should not be taken in a limiting sense. The scope of the invention is best determined by reference to the appended claims.

At present, a computer system usually has a total memory encryption design, which uses different keys to encrypt the different parts of a system memory to increase the security of the computer system. The keys for the encryption of the different storage areas of the system memory each may be represented by a particular key identification code (key ID). The computer system may use a key table to store the keys of the different key IDs. During data encryption, the key table is checked to obtain the key corresponding to the entered key ID.

Considering the multi-key encryption of the system memory, in the present application, the key ID is one of the parameters used in the management of a hierarchical cache structure. The hierarchical cache structure may include level 1, level 2 and level 3 caches (L1, L2 and L3). Based on the total memory encryption, the hierarchical cache structure in the present application may be flushed according to a designated key ID and a designated address.

In an exemplary embodiment, the present application proposes a processor, which uses an instruction set architecture (ISA) instruction CFLUSHKEYID to manage its hierarchical cache structure to flush a cache line matching both a designated key ID Key_ID_S and a designated address Addr_S.

A modern operating system generally uses a virtual memory management mechanism, and a memory management unit (MMU) of a central processing unit (CPU) supports the transform from a virtual address (VA) to a physical address (PA). The designated address (Addr_S) designated in the instruction CFLUSHKEYID may be a designated physical address (PA) or a designated virtual address (VA). If the designated address is a virtual address, the memory management unit (such as a memory ordering buffer) operates to transform the virtual address into a physical address, and then performs a cache line flushing operation. If the updated content in the matching cache line has not been written back to the system memory, the matching cache line has to be written back to the system memory prior to being flushed. The instruction set architecture supported by the processor is not limited, it may be x86 architecture, Advanced RISC Machine (abbreviated as ARM) architecture, MIPS (Microprocessor without Interlocked Pipeline Stages) instruction set architecture, RISC-V (RISC-Five) Instruction Set Architecture, SPARC Instruction Set Architecture, IBM Power Instruction Set Architecture, or others.

FIG. 1 shows an instruction format for the ISA instruction CFLUSHKEYID in accordance with an exemplary embodiment of the present application. In addition to the opcode 102 for recognizing the instruction CFLUSHKEYID, two operands 104 and 106 are required. The operand 104 indicates the designated key ID (Key_ID_S). The operand 106 indicates the designated address (Addr_S). The processor compares the designated key ID (Key_ID_S) with the key ID (Key_ID) obtained from each cache line, matching the designated address (Addr_S), in the hierarchical cache structure. If the obtained key ID Key_ID equals to the designated key ID Key_ID_S, the corresponding cache line is flushed. The operands 104 and 106 may be implemented in various ways. Referring to the instruction format 108 , corresponding to the ISA instruction CFLUSHKEYID, the operand 104 indicates a register number or a system memory address (r/m), and the operand 106 indicates a register number (reg). According to the operand 104 , the designated key ID (Key_ID_S) is obtained from a register or the system memory (r/m). According to the operand 106 , the designated address (Addr_S) is obtained from a register (reg). Referring to the instruction format 110 , corresponding to the ISA instruction CFLUSHKEYID, the operand 104 indicates a register number (reg), and the operand 106 indicates a register number or a system memory address (r/m). According to the operand 104 , the designated key ID (Key_ID_S) is obtained from a register (reg). According to the operand 106 , the designated address (Addr_S) is obtained from a register or the system memory (r/m). Referring to the instruction format 112 , corresponding to the ISA instruction CFLUSHKEYID, the operand 104 indicates a register number or a system memory address (r/m), and the operand 106 is an immediate data (imm16). According to the operand 104 , the designated key ID (Key_ID_S) is obtained from a register or the system memory (r/m). According to the operand 106 , an immediate data is obtained and interpreted as the designated address (Addr_S). Another instruction format of the ISA instruction CFLUSHKEYID identified by the opcode 102 may use just a single operand to indicate both the designated key ID (Key_ID_S) and the designated address (Addr_S). According to the single operand, the designated key ID (Key_ID_S) and the designated address (Addr_S) are obtained from a register (reg), the system memory (m), or interpreted from an immediate data. In some exemplary embodiments, some instructions for setting the registers, the system memory address, or the immediate data to get ready the designated key ID (Key_ID_S) and the designated address (Addr_S) are coded prior to the ISA instruction CFLUSHKEYID. The ISA instruction CFLUSHKEYID obtains the designated key ID (Key_ID_S) and the designated address (Addr_S) through its operands 104 and 106 .

FIG. 2 illustrates a cache line format 200 of a hierarchical cache structure in accordance with an exemplary embodiment of the present application. Referring to the cache line format 200 of the hierarchical cache structure, the field 202 shows a valid bit (VALID), using ‘0’ to indicate that the cache line is invalid, and using ‘1’ to indicate that the cache line is valid. The field 204 shows a key ID Key_ID, and the field 206 shows a tag. The hierarchical cache structure may be searched for matching cache lines matching the designated key ID (Key_ID_S) and matching the designated physical address (PA).

The cache line searching may involve the following steps. In step 1 , the hierarchical cache structure generates a tag and an index according to the designated physical address (PA). Specifically, a physical address (PA) may be divided into sections, wherein one section shows a tag and one section shows an index. Thus, the hierarchical cache structure may extract a tag and an index from the designated physical address (PA). In step 2 , the hierarchical cache structure searches itself for matching cache lines matching the designated key ID Key_ID_S, the tag, and the index. Specifically, the hierarchical cache structure is first searched according to the index, and there may be at least one cache line matching the index. Referring to each cache line matching the index, a key ID Key_ID in the field 204 and a tag in the field 206 are compared with the designated key ID Key_ID_S and the tag corresponding to the designated physical address. If they are all the same, the checked cache line is the matching cache line. Otherwise, the checked cache line is not the matching cache line. As for how to search the hierarchical cache structure according to the index, it is the general knowledge of those skilled in the art, and details are not described here.

In the present application, to manage a hierarchical cache structure in granularity of keys, a key ID (Key_ID) field may be added to each cache line, and the hierarchical cache structure may be modified accordingly.

In an exemplary embodiment, the present application designs the microcode (UCODE) of the processor for execution of the instruction CFLUSHKEYID, and may further modify the processor hardware with the UCODE design.

FIG. 3 is a block diagram illustrating a processor 300 and a core core_ 1 thereon in accordance with an exemplary embodiment of the present application. The illustrated hierarchical cache structure includes level 1, 2 and 3 caches (L1, L2 and L3). The L1 and L2 form an in-core cache module of the core core_ 1 . The level 3 cache L3 is the last level cache (LLC for short) that may be shared with the other cores.

After being loaded from a system memory 302 into an instruction cache 304 , at least one instruction is decoded by a decoder 306 , wherein an instruction CFLUSHKEYID is included in the at least one instruction. The decoder 306 includes an instruction buffer (XIB for short) 308 and an instruction translator (XLATE for short) 310 . The instruction buffer (XIB) 308 identifies the instruction CFLUSHKEYID proposed in the present application, and the instruction translator (XLATE) 310 translates the instruction CFLUSHKEYID into at least one microinstruction that may be recognized by the pipelined hardware to drive the pipelined hardware to flush the matching cache lines in L1, L2 and L3. The matching cache lines matches the designated key ID Key_ID_S as well as the designated address Addr_S indicated by the instruction CFLUSHKEYID. In an exemplary embodiment, the XLATE 310 recognizes the opcode 102 of the instruction CFLUSHKEYID, and translates the instruction CFLUSHKEYID into at least one microinstruction, recognizable by the pipelined hardware, based on the microcode UCODE stored in a microcode memory. According to a register alias table (RAT) 312 , the at least one microinstruction is stored in the reservation station (RS) 314 for further utilization. The at least one microinstruction includes a flushing microinstruction. According to the flushing microinstruction stored in the RS 314 , a memory ordering buffer (MOB) 316 is triggered to operate the hierarchical cache structure to perform a flushing operation. In an exemplary embodiment, the decoded at least one microinstruction further includes microinstruction(s) for exception checking (e.g., privilege level checking), memory address jumping (e.g., jumping to the instruction following the ISA invalidation instruction), and so on.

The memory ordering buffer (MOB) 316 is generally used as a communication interface between the core core_ 1 and the memories (e.g., registers Reg, the L1, L2 and L3, and system memory 302 ). FIG. 3 specifically illustrates a microinstruction design in the present application, showing how to flush matching cache lines, matching the designated key ID (Key_ID_S) and the designated address (Addr_S), in the L1, L2, and L3 through the memory ordering buffer (MOB) 316 .

As shown, the reservation station (RS) 314 outputs the flushing microinstruction (including the opcode 318 , and operands 320 and 322 ) to the memory ordering buffer (MOB) 316 . After identifying the opcode 318 , as indicated by the operands 320 and 322 , the designated key ID (Key_ID_S) and the designated address (Addr_S) are obtained from a register Reg or the system memory 302 through the communication interface implemented by the memory ordering buffer (MOB) 316 . In another exemplary embodiments ( 112 of FIG. 1 ), the designated address (Addr_S) is interpreted from the immediate data (imm16). As mentioned above, in an exemplary embodiment, the instruction CFLUSHKEYID uses only one operand, and the operand may be divided into two sections of information, one section indicates the designated key ID (Key_ID_S), and the other section indicates the designated address (Addr_S). The operand may record a register number, a system memory address, or an immediate data. Through the operand, the designated key ID (Key_ID_S) and the designated address (Addr_S) are read from a register (reg) or the system memory (m), or interpreted from an immediate data (imm16). In an exemplary embodiment, the designated address Addr_S is a virtual address VA, and is translated to a physical address PA through the memory ordering buffer 316 . Through the memory ordering buffer 316 , a flushing request 324 is provided to the level 1 cache (L1), and then to the level 2 cache (L2), and finally to the level 3 cache (L3).

The flushing request 324 carries the designated key ID (Key_ID_S) and the physical address (PA) corresponding to the designated address (Addr_S). According to the designated key ID (Key_ID_S) and the physical address (PA) carried in the flushing request 324 , the level 3 cache (L3) uses the aforementioned cache line searching method to find a matching cache line and flush it. Note that if the matching cache line has not been stored back to the system memory 302 , storing the matching cache line back to the system memory 302 is required before flushing the matching cache line. In an exemplary embodiment, a valid bit (VALID) of the matching cache line is de-asserted to flush the matching cache line. In an exemplary embodiment, when no matching cache line is found from the level 3 cache (L3) according to the aforementioned cache line searching method, the level 3 cache (L3) does no further actions and the instruction CFLUSHKEYID is completed.

A symbol (hereinafter referred to as a matching symbol) of the matching cache line found from the level 3 cache (L3) will be used in searching the level 2 cache (L2) and the level 1 cache (L1). Generally, in a hierarchical cache structure, each cache line is marked with a symbol. At the different cache levels, the cache lines matching the same key ID and the same address are marked with the same symbol. In an exemplary embodiment, a symbol includes information of a key ID (Key_ID), a tag, an index, and the like. In another exemplary embodiment, a symbol includes information of physical address (PA).

The level 3 cache (L3) sends a snoop request 328 to a snoop filter 326 , wherein the snoop request 328 carries a matching symbol. Such a snoop request carrying a matching symbol is provided to the level 2 cache (L2) through the snoop filter 326 . The level 2 cache (L2) flushes a cache line with the same matching symbol, and further provides the snoop request carrying the matching symbol to the level 1 cache (L1). The level 1 cache (L1) also flushes a cache line with the same matching symbol. In this manner, in the whole hierarchical cache structure including L1, L2, and L3, all cache lines matching the designated key identifier (Key_ID_S) and the designated address (Addr_S) are indeed flushed.

FIG. 4 illustrates a computer system 400 in accordance with an exemplary embodiment of the present application. The die Die_ 1 includes a processor Processor_ 1 , and the die Die_ 2 includes a processor Proessor_ 2 . Each processor Processor_ 1 /Proessor_ 2 includes multiple cores core_ 1 ˜core_N. Each core includes an in-core cache module (e.g., each formed by a level 1 cache L1 and a level 2 cache L2). In the processor Processor_ 1 , the multiple cores core_ 1 ˜core_N share the same last-level cache (e.g., the level 3 cache L3) LLC_ 1 , and a snoop filter Snoop_ 1 is paired with the last-level cache LLC_ 1 . In the processor Processor_ 2 , the multiple cores core_ 1 ˜core_N share the same last-level cache LLC_ 2 , and a snoop filter Snoop_ 2 is paired with the last-level cache LLC_ 2 . The two processors Processor_ 1 and Processor_ 2 on the two dies Die_ 1 and Die_ 2 share a system memory Sys_Mem. The illustration shows that a core core_ 1 of the processor Processor_ 1 on the die Die_ 1 executes the instruction CFLUSHKEYID proposed in the present application, which indicates a designated key ID (Key_ID_S) and a designated address (Addr_S).

Through the communication interface provided by the memory ordering buffer MOB, the core core_i transmits a flushing request 402 that indicates a designated key ID (Key_ID_S) and a physical address corresponding to a designated address (Addr_S) to an in-core cache module 404 , and the in-core cache module 404 further passes the flushing request 402 to the last level cache LLC_ 1 . According to the designated key ID (Key_ID_S) and the designated address (Addr_S) indicated by the flushing request 402 , the last-level cache LLC_ 1 searches itself to find the matching cache line and flushes it. The last-level cache LLC_ 1 loads a symbol of the matching cache line to a snoop request 406 and provides the snoop request 406 to the snoop filter snoop_ 1 . The snoop filter snoop_ 1 receives the snoop request 406 and passes it to all in-core cache modules of the different cores core_ 1 ˜core_N of the processor Processor_ 1 . Accordingly, the matching cache lines in the in-core cache modules of all cores core_ 1 ˜core_N of the processor Processor_ 1 are flushed. In this way, in the processor Processor_ 1 , all matching cache lines (matching the designated key ID (Key_ID_S) and the designated address (Addr_S)) in each of the in-core cache modules of the different cores core_ 1 ˜core_N and in the last-level cache LLC_ 1 shared by all cores core_ 1 ˜core_N are flushed consistently.

In particular, through a channel 408 between the die Die_ 1 and the die Die_ 2 (in one exemplary embodiment, the channel 408 is an internal bus), the last level cache LLC_ 1 provides the flushing request 402 (with the designated key ID Key_ID_S and the physical address (PA) corresponding to the designated address (Addr_S)) received from the in-core cache module 404 to the last level cache LLC_ 2 . According to the designated key ID Key_ID_S and the designated address (Addr_S) indicated by the flushing request 402 , the last-level cache LLC_ 2 searches itself to find a matching cache line and flushes it. A symbol of the matching cache line found from the last-level cache LLC_ 2 no doubt equals to the matching symbol determined by the last level cache LLC_ 1 , and is carried by a snoop request 410 . The snoop filter snoop_ 2 receives the snoop request 410 from the last-level cache LLC_ 2 and passes it to all in-core cache modules of the different cores core_ 1 ˜core_N of the processor Processor_ 2 . Accordingly, all matching cache lines in the in-core cache modules of the different cores core_ 1 ˜core_N of the processor Processor_ 2 are flushed. In this way, in the processor Processor_ 2 , all matching cache lines (matching the designated key ID (Key_ID_S) and the designated address (Addr_S)) in the in-core cache modules of the different cores core_ 1 ˜core_N and in the last-level cache LLC_ 2 shared by all cores core_ 1 ˜core_N are flushed consistently.

The technique of the present application flushes all matching cache lines (matching the designated key ID (Key_ID_S) and the designated address (Addr_S)) in the whole hierarchical cache structure of the entire computer system 400 without omission. Any use of an ISA instruction to complete the flushing of the hierarchical cache structure based on of the designated key ID (Key_ID_S) and the designated address (Addr_S) is within the field of the present application.

Based on the instruction format 108 of FIG. 1 and the techniques described in FIGS. 1 to 4 , how an instruction CFLUSHKEYID in the present application is executed is described in the following. In an exemplary embodiment, the sequence of instructions to be executed is:

MOV EAX,KeyID  (1)

MOV EBX,ProcessMemory_VA  (2)

CFLUSHKEYID EAX,EBX  (3)

After being compiled, the instructions (1), (2), and (3) are loaded into the system memory 302 in FIG. 3 . The processor 300 sequentially reads the instructions (1) and (2) from the system memory 302 and executes them. A key ID KeyID is loaded into a register EAX. A virtual address ProcessMemory_VA is loaded into a register EBX. Then, the processor 300 executes the instruction (3).

Referring to FIG. 3 , the processor 300 loads the instruction (3) from the system memory 302 into the instruction cache 304 . The decoder 306 translates the instruction (3) into at least one microinstruction according to the recognized opcode 102 shown in FIG. 1 , and stores the at least one microinstruction into a corresponding reserved station (RS) 314 as indicated by a register alias table (RAT) 312 . When a flushing microinstruction in the at least one microinstruction satisfies an execution condition, the reservation station (RS) 314 provides the flushing microinstruction to the memory ordering buffer (MOB) 316 for execution. After recognizing the opcode 318 , the memory ordering buffer (MOB) 316 obtains the key ID KeyID and the virtual address ProcessMemory_VA from the registers EAX and EBX according to the operands 320 and 322 . The memory ordering buffer (MOB) 316 translates the virtual address ProcessMemory_VA to a physical address ProcessMemory_PA. Then, the memory order buffer (MOB) 316 generates a flushing request 324 carrying the key ID KeyID and the physical address ProcessMemory_PA, and passes the flushing request 324 to the level 1 cache (L1), the level 2 cache (L2), and finally to the level 3 cache (L3).

Referring to FIGS. 2 and 3 , the level 3 cache (L3) obtains a tag ProcessMemory_TAG and an index ProcessMemory_INDEX corresponding to the physical address ProcessMemory_PA. The level 3 cache (L3) first searches itself according to the index ProcessMemory_INDEX for at least one valid cache line, and then compares the key ID (Key_ID 204 ) and the tag ( 206 ) of the valid cache lines with the key ID KeyID and the tag ProcessMemory_TAG to determine the matching cache line. Then the matching cache line is flushed. The level 3 cache (L3) sends a snoop request 328 to the snoop filter 326 , wherein the snoop request 328 carries a matching symbol that contains information of the key ID KeyID, the tag ProcessMemory_TAG, the index ProcessMemory_INDEX, etc. Such a snoop request with a matching symbol is provided to the level 2 cache (L2) through the snoop filter 326 . According to the matching symbol, the level 2 cache (L2) searches itself for a matching cache line and flushes it. Then, the level 2 cache (L2) passes the flushing request with the matching symbol to the level 1 cache (L1). According to the matching symbol, the level 1 cache (L1) searches itself for a matching cache line and flushes it. In this exemplary embodiment, all matching cache lines (matching the key ID KeyID and the virtual address ProcessMemory_VA) in the whole hierarchical cache structure including L1, L2 and L3 are flushed consistently.

Referring to FIG. 4 , it is assumed that the core core_i of the processor Processor_ 1 of the computer system 400 executes the aforementioned sequence of instructions (1), (2), and (3). Through the communication interface built through the memory ordering buffer MOB, the core core_i transmits a flushing request 402 that indicates the key ID KeyID and the physical address ProcessMemory_PA corresponding to the virtual address ProcessMemory_VA to the in-core cache module 404 , and then to the last level cache LLC_ 1 . According to the key ID KeyID and the physical address ProcessMemory_PA indicated by the flushing request 402 , the last-level cache LLC_ 1 searches itself for a matching cache line and flushes it. A matching symbol (containing information of the key ID KeyID, the tag ProcessMemory_TAG, the index ProcessMemory_INDEX, and other information) of the matching cache line found from the last-level cache LLC_ 1 may be carried by a snoop request 406 . The last-level cache LLC_ 1 provides the snoop request 406 to the snoop filter snoop_ 1 , and then the snoop filter snoop_ 1 passes the snoop request 406 to the in-core cache modules of the cores core_ 1 ˜core_N in the processor Processor_ 1 . Thus, all matching cache lines (matching the key ID KeyID and the virtual address ProcessMemory_VA) in the in-core cache modules of the different cores core_ 1 ˜core_N of the processor Processor_ 1 and in the last-level cache LLC_ 1 are flushed consistently

In addition, through the channel 408 between the die Die_ 1 and the die Die_ 2 (in an exemplary embodiment, the channel 408 is an internal bus), the last level cache LLC_ 1 outputs the flushing request 402 (received from the in-core cache module 404 and carrying the designated key ID KEYID and the physical address ProcessMemory_PA of the designated virtual address ProcessMemory_VA) to the last level cache LLC_ 2 . According to the designated key ID KEYID and the physical address ProcessMemory_PA carried by the flushing request 402 , the last-level cache LLC_ 2 searches itself for the matching cache line and flushes it. The last-level cache LLC_ 2 finds a matching symbol (containing information of a key ID KeyID, a tag ProcessMemory_TAG, an index ProcessMemory_INDEX, and so on) which is the same as that found by the last-level cache LLC_ 1 . The matching symbol is carried by a flushing request 410 to be passed from the last-level cache LLC_ 2 to the snoop filter snoop_ 2 . The snoop filter snoop_ 2 passes the snoop request 410 to all in-core modules of the different cores core_ 1 to core_N of the processor Processor_ 2 . In this manner, the matching cache lines (matching the designated key ID KeyID and the designated virtual address ProcessMemory_VA) in the in-core cache modules of all cores core_ 1 ˜core_N of the processor Processor_ 2 as well as the matching cache lines in the last-level cache LLC_ 2 shared by the different cores core_ 1 ˜core_N are flushed consistently.

According to the technology of the present application, the hierarchical cache structure is managed in granularity of the keys. When a total memory encryption function is enabled, the operating system may manage the hierarchical cache structure in granularity of the keys.

While the invention has been described by way of example and in terms of the preferred embodiments, it should be understood that the invention is not limited to the disclosed embodiments. On the contrary, it is intended to cover various modifications and similar arrangements (as would be apparent to those skilled in the art). Therefore, the scope of the appended claims should be accorded the broadest interpretation so as to encompass all such modifications and similar arrangements.