This page provides detailed information about the Imperas Instruction Set Simulator for the RISC-V RV64IMC processor core.
Processor IP owner is RISC-V Foundation. More information is available from them here.
The Imperas Instruction Set Simulator (ISS) is a product of Imperas Software Ltd. It is a binary licensed by Imperas as a commercial product. It is also available in OVP packages for evaluation and demonstration.
The Imperas ISS uses the OVP Fast Processor Models and dynamically loads them as selected.The OVP Fast Processor Models are written using the OVP VMI API that provides a Virtual Machine Interface that defines the behavior of the processor.The VMI API makes a clear line between model and simulator allowing very good optimization and world class high speed performance.
The processor models are provided as a binary shared object and are also available as source (different models have different licensing conditions). This allows the download and use of the model binary or the use of the source to explore and modify the model.
The models have been run through an extensive QA and regression testing process.
If you develop your own processor models using the OVP VMI APIs then they can be used with the Imperas ISS.
Traditionally, processor models and simulators make use of one thread on the host PC. Imperas have developed a technology, called QuantumLeap, that makes use of the many host cores found in modern PC/workstations to achieve industry leading simulation performance. To find out about the Imperas parallel simulation lookup Imperas QuantumLeap. There are videos of QuantumLeap on ARM here, and MIPS here. For press information related to QuantumLeap for ARM click here or for MIPS click here. Many of the OVP Fast Processor Models have been qualified to work with QuantumLeap - this is indicated for this model below.
This ISS executes instructions of the target architecture and provides an interface for debug access. An interface to GDB is provided and this allows the connection of many industry standard debuggers that use the GDB/RSP interface. For more information watch the OVP video here.
The ISS also works with the Imperas Multicore Debugger and advanced Verification, Analysis and Profiling tools.
The ISS is downloadable (needs registration and to be logged in) in package Demo_Processors for Windows32 and for Linux32. Note that the ISS is also available for 64 bit hosts as part of the commercial products from Imperas.
This ISS uses the CPU with Model Variant name: RV64IMC
RISC-V RV64IMC 64-bit processor model
This Model is released under the Open Source Apache 2.0
Extensions Enabled by Default:
The model has the following architectural extensions enabled, and the following bits in the misa CSR Extensions field will be set upon reset:
misa bit 2: extension C (compressed instructions)
misa bit 8: RV32I/RV64I/RV128I base integer instruction set
misa bit 12: extension M (integer multiply/divide instructions)
misa bit 18: extension S (Supervisor mode)
misa bit 20: extension U (User mode)
To specify features that can be dynamically enabled or disabled by writes to the misa register in addition to those listed above, use parameter "add_Extensions_mask". This is a string parameter containing the feature letters to add; for example, value "DV" indicates that double-precision floating point and the Vector Extension can be enabled or disabled by writes to the misa register, if supported on this variant.
Legacy parameter "misa_Extensions_mask" can also be used. This Uns32-valued parameter specifies all writable bits in the misa Extensions field, replacing any permitted bits defined in the base variant.
Note that any features that are indicated as present in the misa mask but absent in the misa will be ignored. See the next section.
Available Extensions Not Enabled by Default:
The following extensions are supported by the model, but not enabled by default in this variant:
misa bit 0: extension A (atomic instructions)
misa bit 1: extension B (bit manipulation extension)
misa bit 3: extension D (double-precision floating point)
misa bit 4: RV32E base integer instruction set (embedded)
misa bit 5: extension F (single-precision floating point)
misa bit 7: extension H (hypervisor)
misa bit 10: extension K (cryptographic)
misa bit 13: extension N (user-level interrupts)
misa bit 21: extension V (vector extension)
misa bit 23: extension X (non-standard extensions present)
To add features from this list to the base variant, use parameter "add_Extensions". This is a string parameter containing the feature letters to add; for example, value "DV" indicates that double-precision floating point and the Vector Extension should be enabled, if they are currently absent and are available on this variant.
Legacy parameter "misa_Extensions" can also be used. This Uns32-valued parameter specifies the reset value for the misa CSR Extensions field, replacing any permitted bits defined in the base variant.
On this variant, the Machine trap-vector base-address register (mtvec) is writable. It can instead be configured as read-only using parameter "mtvec_is_ro".
Values written to "mtvec" are masked using the value 0xfffffffffffffffd. A different mask of writable bits may be specified using parameter "mtvec_mask" if required. In addition, when Vectored interrupt mode is enabled, parameter "tvec_align" may be used to specify additional hardware-enforced base address alignment. In this variant, "tvec_align" defaults to 0, implying no alignment constraint.
The initial value of "mtvec" is 0x0. A different value may be specified using parameter "mtvec" if required.
Values written to "stvec" are masked using the value 0xfffffffffffffffd. A different mask of writable bits may be specified using parameter "stvec_mask" if required. parameter "tvec_align" may be used to specify additional hardware-enforced base address alignment in the same manner as for the "mtvec" register, described above.
On reset, the model will restart at address 0x0. A different reset address may be specified using parameter "reset_address" or applied using optional input port "reset_addr" if required.
On an NMI, the model will restart at address 0x0. A different NMI address may be specified using parameter "nmi_address" or applied using optional input port "nmi_addr" if required.
WFI will halt the processor until an interrupt occurs. It can instead be configured as a NOP using parameter "wfi_is_nop". WFI timeout wait is implemented with a time limit of 0 (i.e. WFI causes an Illegal Instruction trap in Supervisor mode when mstatus.TW=1).
The "cycle" CSR is implemented in this variant. Set parameter "cycle_undefined" to True to instead specify that "cycle" is unimplemented and reads of it should trap to Machine mode.
The "time" CSR is implemented in this variant. Set parameter "time_undefined" to True to instead specify that "time" is unimplemented and reads of it should trap to Machine mode. Usually, the value of the "time" CSR should be provided by the platform - see notes below about the artifact "CSR" bus for information about how this is done.
The "instret" CSR is implemented in this variant. Set parameter "instret_undefined" to True to instead specify that "instret" is unimplemented and reads of it should trap to Machine mode.
A 16-bit ASID is implemented. Use parameter "ASID_bits" to specify a different implemented ASID size if required.
This variant supports address translation modes 0, 8 and 9. Use parameter "Sv_modes" to specify a bit mask of different modes if required.
TLB behavior is controlled by parameter "ASIDCacheSize". If this parameter is 0, then an unlimited number of TLB entries will be maintained concurrently. If this parameter is non-zero, then only TLB entries for up to "ASIDCacheSize" different ASIDs will be maintained concurrently initially; as new ASIDs are used, TLB entries for less-recently used ASIDs are deleted, which improves model performance in some cases. If the model detects that the TLB entry cache is too small (entry ejections are very frequent), it will increase the cache size automatically. In this variant, "ASIDCacheSize" is 8
Unaligned memory accesses are not supported by this variant. Set parameter "unaligned" to "T" to enable such accesses.
16 PMP entries are implemented by this variant. Use parameter "PMP_registers" to specify a different number of PMP entries; set the parameter to 0 to disable the PMP unit. The PMP grain size (G) is 0, meaning that PMP regions as small as 4 bytes are implemented. Use parameter "PMP_grain" to specify a different grain size if required. Unaligned PMP accesses are not decomposed into separate aligned accesses; use parameter "PMP_decompose" to modify this behavior if required.
The model can be configured to implement a Core Local Interrupt Controller (CLIC) using parameter "CLICLEVELS"; when non-zero, the CLIC is present with the specified number of interrupt levels (2-256), as described in the RISC-V Core-Local Interrupt Controller specification, and further parameters are made available to configure other aspects of the CLIC. "CLICLEVELS" is zero in this variant, indicating that a CLIC is not implemented.
The "reset" port is an active-high reset input. The processor is halted when "reset" goes high and resumes execution from the reset address specified using the "reset_address" parameter or "reset_addr" port when the signal goes low. The "mcause" register is cleared to zero.
The "nmi" port is an active-high NMI input. The processor resumes execution from the address specified using the "nmi_address" parameter or "nmi_addr" port when the NMI signal goes high. The "mcause" register is cleared to zero.
All other interrupt ports are active high. For each implemented privileged execution level, there are by default input ports for software interrupt, timer interrupt and external interrupt; for example, for Machine mode, these are called "MSWInterrupt", "MTimerInterrupt" and "MExternalInterrupt", respectively. When the N extension is implemented, ports are also present for User mode. Parameter "unimp_int_mask" allows the default behavior to be changed to exclude certain interrupt ports. The parameter value is a mask in the same format as the "mip" CSR; any interrupt corresponding to a non-zero bit in this mask will be removed from the processor and read as zero in "mip", "mie" and "mideleg" CSRs (and Supervisor and User mode equivalents if implemented).
Parameter "external_int_id" can be used to enable extra interrupt ID input ports on each hart. If the parameter is True then when an external interrupt is applied the value on the ID port is sampled and used to fill the Exception Code field in the "mcause" CSR (or the equivalent CSR for other execution levels). For Machine mode, the extra interrupt ID port is called "MExternalInterruptID".
The "deferint" port is an active-high artifact input that, when written to 1, prevents any pending-and-enabled interrupt being taken (normally, such an interrupt would be taken on the next instruction after it becomes pending-and-enabled). The purpose of this signal is to enable alignment with hardware models in step-and-compare usage.
The model can be configured to implement Debug mode using parameter "debug_mode". This implements features described in Chapter 4 of the RISC-V External Debug Support specification with version specified by parameter "debug_version" (see References). Some aspects of this mode are not defined in the specification because they are implementation-specific; the model provides infrastructure to allow implementation of a Debug Module using a custom harness. Features added are described below.
Parameter "debug_mode" can be used to specify three different behaviors, as follows:
1. If set to value "vector", then operations that would cause entry to Debug mode result in the processor jumping to the address specified by the "debug_address" parameter. It will execute at this address, in Debug mode, until a "dret" instruction causes return to non-Debug mode. Any exception generated during this execution will cause a jump to the address specified by the "dexc_address" parameter.
2. If set to value "interrupt", then operations that would cause entry to Debug mode result in the processor simulation call (e.g. opProcessorSimulate) returning, with a stop reason of OP_SR_INTERRUPT. In this usage scenario, the Debug Module is implemented in the simulation harness.
3. If set to value "halt", then operations that would cause entry to Debug mode result in the processor halting. Depending on the simulation environment, this might cause a return from the simulation call with a stop reason of OP_SR_HALT, or debug mode might be implemented by another platform component which then restarts the debugged processor again.
Debug State Entry:
The specification does not define how Debug mode is implemented. In this model, Debug mode is enabled by a Boolean pseudo-register, "DM". When "DM" is True, the processor is in Debug mode. When "DM" is False, mode is defined by "mstatus" in the usual way.
Entry to Debug mode can be performed in any of these ways:
1. By writing True to register "DM" (e.g. using opProcessorRegWrite) followed by simulation of at least one cycle (e.g. using opProcessorSimulate), dcsr cause will be reported as trigger;
2. By writing a 1 then 0 to net "haltreq" (using opNetWrite) followed by simulation of at least one cycle (e.g. using opProcessorSimulate);
3. By writing a 1 to net "resethaltreq" (using opNetWrite) while the "reset" signal undergoes a negedge transition, followed by simulation of at least one cycle (e.g. using opProcessorSimulate);
4. By executing an "ebreak" instruction when Debug mode entry for the current processor mode is enabled by dcsr.ebreakm, dcsr.ebreaks or dcsr.ebreaku.
In all cases, the processor will save required state in "dpc" and "dcsr" and then perform actions described above, depending in the value of the "debug_mode" parameter.
Debug State Exit:
Exit from Debug mode can be performed in any of these ways:
1. By writing False to register "DM" (e.g. using opProcessorRegWrite) followed by simulation of at least one cycle (e.g. using opProcessorSimulate);
2. By executing an "dret" instruction when Debug mode.
In both cases, the processor will perform the steps described in section 4.6 (Resume) of the Debug specification.
When Debug mode is enabled, registers "dcsr", "dpc", "dscratch0" and "dscratch1" are implemented as described in the specification. These may be manipulated externally by a Debug Module using opProcessorRegRead or opProcessorRegWrite; for example, the Debug Module could write "dcsr" to enable "ebreak" instruction behavior as described above, or read and write "dpc" to emulate stepping over an "ebreak" instruction prior to resumption from Debug mode.
Debug Mode Execution:
The specification allows execution of code fragments in Debug mode. A Debug Module implementation can cause execution in Debug mode by the following steps:
1. Write the address of a Program Buffer to the program counter using opProcessorPCSet;
2. If "debug_mode" is set to "halt", write 0 to pseudo-register "DMStall" (to leave halted state);
3. If entry to Debug mode was handled by exiting the simulation callback, call opProcessorSimulate or opRootModuleSimulate to resume simulation.
Debug mode will be re-entered in these cases:
1. By execution of an "ebreak" instruction; or:
2. By execution of an instruction that causes an exception.
In both cases, the processor will either jump to the debug exception address, or return control immediately to the harness, with stopReason of OP_SR_INTERRUPT, or perform a halt, depending on the value of the "debug_mode" parameter.
Debug Single Step:
When in Debug mode, the processor or harness can cause a single instruction to be executed on return from that mode by setting dcsr.step. After one non-Debug-mode instruction has been executed, control will be returned to the harness. The processor will remain in single-step mode until dcsr.step is cleared.
Port "DM" is an output signal that indicates whether the processor is in Debug mode
Port "haltreq" is a rising-edge-triggered signal that triggers entry to Debug mode (see above).
Port "resethaltreq" is a level-sensitive signal that triggers entry to Debug mode after reset (see above).
This model is configured with a trigger module, implementing a subset of the behavior described in Chapter 5 of the RISC-V External Debug Support specification with version specified by parameter "debug_version" (see References).
Trigger Module Restrictions:
The model currently supports tdata1 of type 0, type 2 (mcontrol), type 3 (icount), type 4 (itrigger), type 5 (etrigger) and type 6 (mcontrol6). icount triggers are implemented for a single instruction only, with count hard-wired to 1 and automatic zeroing of mode bits when the trigger fires.
Trigger Module Parameters:
Parameter "trigger_num" is used to specify the number of implemented triggers. In this variant, "trigger_num" is 4.
Parameter "tinfo" is used to specify the value of the read-only "tinfo" register, which indicates the trigger types supported. In this variant, "tinfo" is 0x7d.
Parameter "tinfo_undefined" is used to specify whether the "tinfo" register is undefined, in which case reads of it trap to Machine mode. In this variant, "tinfo_undefined" is 0.
Parameter "tcontrol_undefined" is used to specify whether the "tcontrol" register is undefined, in which case accesses to it trap to Machine mode. In this variant, "tcontrol_undefined" is 0.
Parameter "mcontext_undefined" is used to specify whether the "mcontext" register is undefined, in which case accesses to it trap to Machine mode. In this variant, "mcontext_undefined" is 0.
Parameter "scontext_undefined" is used to specify whether the "scontext" register is undefined, in which case accesses to it trap to Machine mode. In this variant, "scontext_undefined" is 0.
Parameter "mscontext_undefined" is used to specify whether the "mscontext" register is undefined, in which case accesses to it trap to Machine mode. In this variant, "mscontext_undefined" is 0.
Parameter "amo_trigger" is used to specify whether load/store triggers are activated for AMO instructions. In this variant, "amo_trigger" is 0.
Parameter "no_hit" is used to specify whether the "hit" bit in tdata1 is unimplemented. In this variant, "no_hit" is 0.
Parameter "no_sselect_2" is used to specify whether the "sselect" field in "textra32"/"textra64" registers is unable to hold value 2 (indicating match by ASID is not allowed). In this variant, "no_sselect_2" is 0.
Parameter "mcontext_bits" is used to specify the number of writable bits in the "mcontext" register. In this variant, "mcontext_bits" is 13.
Parameter "scontext_bits" is used to specify the number of writable bits in the "scontext" register. In this variant, "scontext_bits" is 34.
Parameter "mvalue_bits" is used to specify the number of writable bits in the "mvalue" field in "textra32"/"textra64" registers; if zero, the "mselect" field is tied to zero. In this variant, "mvalue_bits" is 13.
Parameter "svalue_bits" is used to specify the number of writable bits in the "svalue" field in "textra32"/"textra64" registers; if zero, the "sselect" is tied to zero. In this variant, "svalue_bits" is 34.
Parameter "mcontrol_maskmax" is used to specify the value of field "maskmax" in the "mcontrol" register. In this variant, "mcontrol_maskmax" is 63.
It is possible to enable model debug messages in various categories. This can be done statically using the "override_debugMask" parameter, or dynamically using the "debugflags" command. Enabled messages are specified using a bitmask value, as follows:
Value 0x002: enable debugging of PMP and virtual memory state;
Value 0x004: enable debugging of interrupt state.
All other bits in the debug bitmask are reserved and must not be set to non-zero values.
This model implements a number of non-architectural pseudo-registers and other features to facilitate integration.
CSR Register External Implementation:
If parameter "enable_CSR_bus" is True, an artifact 16-bit bus "CSR" is enabled. Slave callbacks installed on this bus can be used to implement modified CSR behavior (use opBusSlaveNew or icmMapExternalMemory, depending on the client API). A CSR with index 0xABC is mapped on the bus at address 0xABC0; as a concrete example, implementing CSR "time" (number 0xC01) externally requires installation of callbacks at address 0xC010 on the CSR bus.
Instruction pipelines are not modeled in any way. All instructions are assumed to complete immediately. This means that instruction barrier instructions (e.g. fence.i) are treated as NOPs, with the exception of any Illegal Instruction behavior, which is modeled.
Caches and write buffers are not modeled in any way. All loads, fetches and stores complete immediately and in order, and are fully synchronous. Data barrier instructions (e.g. fence) are treated as NOPs, with the exception of any Illegal Instruction behavior, which is modeled.
Real-world timing effects are not modeled: all instructions are assumed to complete in a single cycle.
Hardware Performance Monitor registers are not implemented and hardwired to zero.
The TLB is architecturally-accurate but not device accurate. This means that all TLB maintenance and address translation operations are fully implemented but the cache is larger than in the real device.
All instructions have been extensively tested by Imperas, using tests generated specifically for this model and also reference tests from https://github.com/riscv/riscv-tests.
Also reference tests have been used from various sources including:
The Imperas OVPsim RISC-V models are used in the RISC-V Foundation Compliance Framework as a functional Golden Reference:
where the simulated model is used to provide the reference signatures for compliance testing. The Imperas OVPsim RISC-V models are used as reference in both open source and commercial instruction stream test generators for hardware design verification, for example:
http://valtrix.in/sting from Valtrix
https://github.com/google/riscv-dv from Google
The Imperas OVPsim RISC-V models are also used by commercial and open source RISC-V Core RTL developers as a reference to ensure correct functionality of their IP.
The Model details are based upon the following specifications:
RISC-V Instruction Set Manual, Volume I: User-Level ISA (User Architecture Version 20190305-Base-Ratification)
RISC-V Instruction Set Manual, Volume II: Privileged Architecture (Privileged Architecture Version 20190405-Priv-MSU-Ratification)
The CPU model being used is downloadable (needs registration and to be logged in) in package riscv.model for Windows32 and for Linux32. Note that the CPU model is also available for 64 bit hosts as part of the commercial products from Imperas.
OVP simulator downloadable (needs registration and to be logged in) in package OVPsim for Windows32 and for Linux32. Note that the OVP simulator is also available for 64 bit hosts as part of the commercial products from Imperas.
OVP Download page here.
OVP documentation that provides overview information on processor models is available OVP_Guide_To_Using_Processor_Models.pdf.
Full model specific documentation on the variant (RV64IMC) being used in this ISS is available OVP_Model_Specific_Information_riscv_RV64IMC.pdf.
For more information on the Imperas ISS see the Imperas site and on the OVP Fast Processor model see the OVPworld site.
Location: The Fast Processor Model source and object file is found in the installation VLNV tree: riscv.ovpworld.org/processor/riscv/1.0
Processor Endian-ness: This model can be set to either endian-ness (normally by a pin, or the ELF code).
Processor ELF Code: The ELF code for this model is: 0xf3
QuantumLeap Support: The processor model is qualified to run in a QuantumLeap enabled simulator.
Information on the RV64IMC OVP Fast Processor Model can also be found on other web sites::
www.ovpworld.org has the library pages http://www.ovpworld.org/library/wikka.php?wakka=CategoryProcessor
www.imperas.com has more information on the model library
http://www.ovpworld.org: VMI Programmers Views (VMI VIEW) API Reference Guide.
http://www.ovpworld.org: Creating Instruction Accurate Processor models using the VMI API
Currently available Instruction Set Simulator (ISS) Families.