- 1. Introduction
- 2. Foreword
- 3. Vocabulary
- 3.1. The architectural test
- 3.2. The RISC-V architectural test pool
- 3.3. The RISC-V architectural test suite
- 3.4. The test case
- 3.5. The test case signature
- 3.6. The test signature
- 3.7. The test suite signature
- 3.8. The target shell
- 3.9. The test target
- 3.10. The RISC-V processor (device) configuration
- 3.11. The architectural test framework
- 4. Architectural test pool
- Appendix A: Example ISA test add-01.S
This document contains the RISC-V architectural test pool structure and architectural test format specification which shall be used as a reference document for those who write or are going to write tests for the RISC-V architectural test pool and for those who are going to use the architectural test pool in their own architectural test framework.
-
It includes, as example, source code listing and detailed description of one architectural test
Framework specification which includes description of how the architectural test suites are built and used for the appropriate RISC-V configurations is given in the complementary Framework Specification document. This document is made freely available under a [app_cc_by_4.0].
This document is intended for design and verification engineers who wish to develop new architectural tests and also for those who wish to write or adapt their own test framework.
This is a draft document; it partially documents what exists, and partially documents the longer-term goal. As such, this document still under review and its content will change. Its primary aim is to get a long-term stable version of the spec and to give test authors sufficient lead time to prepare test authoring tools and strategies.
This document has been created by the following people (in alphabetical order of surname).
-
Allen Baum
-
Jeremy Bennett
-
Radek Hajek
-
Premysl Vaclavik
| Revision | Date | Author | Modification |
|---|---|---|---|
1.2.6 Draft |
24 September 2020 |
Neel Gala |
Replaced Compliance with Architecture/Architectural. Minor beautification in macro definitions. |
1.2.5 Draft |
22 Jan 2020 |
Allen Baum |
* removed references to test pool reference doc, mentioned that the framework will generate it * clarified that macros defined in a test may be used in a test * minor clarifications, consistency changes, added page breaks |
1.2.4 Draft |
08 Jan 2020 |
Allen Baum |
* typos fixed * added RVTEST_BASEUPD macro * added explanations for each macro * clarified restrictions on #ifdefs * added comment that test cases with identical conditions should be combined into a single case * documented that test case first parameter should match the #ifdef parameter that precedes it |
1.2.3 Draft |
02 Dec 2019 |
Allen Baum |
* modified macro names to conformn to riscof naming convention of model specific vs. pre-defined * add more complete list of macros, their uses, parameters, and whether they are required or optional * minor structural changes (moving sentences, renumbering) and typo fixes * clarified impact of debug macros * clarified how SIGUPD and BASEUPD must be used, fixed parameter description |
1.2.2 Draft |
21 Nov 2019 |
Allen Baum |
* remove section about test taxonomy, binary tests, emulated ops * clarify/fix boundary between test target and framework responsibilities (split test target into test target and test shell) * remove To Be discussed items that have been discussed * remove default case condition; if conditions are unchaged, part of same case * minor grammatical changes related to the above |
1.2.1 Draft |
19 Nov 2019 |
Allen Baum |
* spec/TestFormatSpec.adoc: changed the format of the signature to fixed 32b data size only extracted from COMPLIANCE_DATA_BEGIN/END range. * made test suite subdirectories upper case, with sub-extensions camel case * updated example to match most recent riscof implement macros * fix format so Appendix is now in TOC * moved note about multiple test cases in a test closer to definition * fixed cut/paste error in example of test pool * more gramatical fixes, clarifications added * added To Be Discussed items regarding emulated instruction and binary tests * added graphic of test suite/test_pool/test/test_case hierarchy |
1.2.1 Draft |
12 Oct 2019 |
Allen Baum |
minor grammar, wording, syntax corrections, added detail and clarification from suggestions by Paul Donahue |
1.2 Draft |
12 Sep 2019 |
Allen Baum |
minor grammar, wording, syntax corrections, added detail and clarification Added detail regarding the 2 approaches for test selection: central database, or embedded conditions embedded in macros Added detail of proposed standard macros RVTEST_SIGBASE, RVTEST_SIGUPD, RVTEST_CASE More explanation of spec status in initial future work paragraph (i.e. goal, not yet accomplished) Removed many "to Be Discussed items and made them official Removed options, made POR for test selection and standard macros RVTEST_SIGBASE, RVTEST_SIGUPD, RVTEST_CASE Removed prohibition on absolute addresses Clarified which test suites a test should be in where they are dependent on multiple extensions Clarified use of includes and macros (and documented exsiting deviations) Clarified use of YAML files Added detail to description and uses of common compliance test pool reference document |
1.1 Draft |
15 Feb 2019 |
Radek Hajek |
Appendix A: example assertions update |
1.0 Draft |
10 Dec 2018 |
Radek Hajek, Premysl Vaclavik |
First version of the document under this file name. Document may contain some segments of the README.adoc from the compatibility reasons. |
The architectural test pool shall become a complete set of architectural tests which will allow developers to build an architectural test suite for any legal RISC-V configuration. The architectural tests will be very likely written by various authors and therefore it is very important to define the architectural test pool structure and architectural test form, which will be obligatory for all tests. Unification of tests will guarantee optimal architectural test pool management and also better quality and readability of the tests. Last but not least, it will simplify the process of adding new tests into the existing architectural test pool and the formal revision process.
The architectural test is a nonfunctional testing technique which is done to validate whether the system developed meets the prescribed standard or not. In this particular case the golden reference is the RISC-V ISA standard.
For purpose of this document we understand that the architectural test is a single test which represents the minimum test code that can be compiled and run. It is written in assembler code and its product is a test signature. A architectural test may consist of several test cases.
The RISC-V architectural test pool consists of all approved architectural tests that can be assembled by the test framework, forming the architectural test suite. The RISC-V architectural test pool must be test target independent (so, should correctly run on any compliant target). Note that this nonfunctional testing is not a substitute for verification or device test.
The RISC-V architectural test suite is a group of tests selected from the architectural test pool to test adherence for the specific RISC-V configuration. Test results are obtained in the form of a test suite signature. Selection of tests is performed based on the target’s asserted configuration, and the spec, Execution Environment or platform requirements. Compliant processor or processor models shall exhibit the same test suite signature as the golden reference test suite signature for the specific configuration being tested.
A test case is part of the architectural test that tests just one feature of the specification.
Note: a single test can contain multiple test cases, each of which can have its own test inclusion condition (as defined by the cond_str parameter of the RVTEST_CASE macro.
The test case signature is represented by single or multiple values. Values are written to memory at the address starting at the address specified by the RVMODEL_DATA_BEGIN and ending at RVMODEL_DATA_END. Signatures can be generated most easily using the RVTEST_SIGUPD macro.
The test signature is a characteristic value which is generated by the architectural test run. The test signature may consist of several test case signatures, prefixed with a separate line containing the name of the test and a unique value indicating its version (e.g. git checkin hash). The test target is responsible for extracting values from memory and properly formatting them, using metadata provided to it by the framework using the RVMODEL_DATA_BEGIN and RVMODEL_DATA_END macros. Test case signature values are written one per line, starting with the most-significant byte on the left-hand side with the format <hex_value> where the length of value will be 32 bits (so 8 characters), regardless of the actual value length computed by the test. Furthermore, the signature should always begin at a 16-byte (128-bit) boundary and the size of the signature should be a multiple of 16-bytes (i.e. it should also end at a 16-byte boundary).
The test suite signature is defined as a set of test signatures valid for given architectural test suite. It represents the test signature of the particular RISC-V configuration selected for the architectural test.
The target shell is the software and hardware environment around the test target that enables it to communicate with the framework, including assembling and linking tests, loading tests into memory, executing tests, and extracting the signature. The input to the target shell is a .S architectural test file, and the output is a test signature.
The test target can be either a RISC-V Instruction Set Simulator (ISS), a RISC-V emulator, a RISC-V RTL model running on an HDL simulator, a RISC-V FPGA implementation or a physical chip. Each of the target types offers specific features and represents specific interface challenges. It is a role of the target shell to handle different targets while using the same architectural test pool as a test source.
The RISC-V ISA specification allows many optional instructions, registers, and other features. Production directed targets typically have a fixed subset of available options. A simulator, on the other hand, may implement all known options which may be constrained to mimic the behavior of the RISC-V processor with the particular configuration. It is a role of the Architectural Test Framework to build and use the architectural test suite suitable for the selected RISC-V configuration.
The architectural test framework selects and configures the architectural test suite from the architectural test pool for the selected test target based on both the specific architectural choices made by an implementation and those required by the Execution Environment It causes the target shell to build, execute, and report a signature. The architectural test framework then compares reported signatures, inserts test part names and version numbers and summarizes differences (or lack of them) into a RISC-V test report. The primary role of the well-defined architectural test pool structure is to provide the tests in a form suitable for the Architectural Test Framework selection engine.
The structure of architectural tests in the architectural test pool shall be based on defined RISC-V extensions and privileged mode selection. This will provide a good overview of which parts of the ISA specification are already covered in the architectural test suite, and which tests are suitable for certain configurations. The architectural test pool has this structure:
architectural-tests-suite (root) |-- <architecture>_<mode>/<feature(s)>, where <architecture> is [ RV32I | RV64I | RV32E ] <mode> is [ M | MU | MS | MSU ], where M Machine mode tests - tests execute in M-mode only MU Machine/User mode tests - tests execute in both M- & U-modes (S-mode may exist) MS Machine/Supv mode tests - tests execute in both M- & S-modes (not U-mode) MSU All mode tests - tests execute in all of M-, S-, & U-Modes <feature(s)> are the lettered extension [A | B | C | M ...] or subextension [Zifencei | Zam | ...] when the tests involve extensions, or more general names when tests cut across extension definitionss (e.g. Priv, Interrupt, Vm). The feature string consists of an initial capital letter, followed by any further letters in lower case.
Note that this structure is for organizational purposes, not functional purposes, although full test names will take advantage of it.
Tests that will be executed in different modes, even if the results are identical, should be replicated in each mode directory, e.g. RV32I_M/, RV32I_MS/, and RV32I_MU/. These tests are typically those involving trapping behavior, e.g load, store, and privilged ops.
The naming convention of a single test:
<test objective>-<test number>.S
-
test objective – an aspect that the test is focused on. A test objective may be an instruction for ISA tests (ADD, SUB, …), or a characteristic covering multiple instructions, e.g. exception event (misaligned fetch, misalign load/store) and others.
-
test number – number of the test. It is expected that multiple tests may be specified for one test objective. We recommend to break down complex tests into a set of small tests. A simple rule of thumb is one simple test objective = one simple test. The code becomes more readable and the test of the objective can be improved just by adding test cases. The typical example are instruction tests for the F extension.
-
A test name shall not include an ISA category as part of its name (i.e. the directory, subdirectory names).
Experience has shown that including ISA category in the test name leads to very long test names. Instead, we have introduced the test pool structure where the full name is composed of the test path in the test pool structure and the simple test name.
Since full names can be reconstructed easily it is not necessary to include the path in test names.
There are both pre-defined and model-specific macros which shall be used in every test to guarantee their portability. In addition, there are both pre-defined and model specific macros that are not required, but may be used in tests for either convenience or debugging purposes.
These macros are be defined in the file compilance_test.h by the author of the test. A significant amount of the framework shall depend on the existence of these macros.
RVTEST_ISA(isa_str)-
-
defines the Test Virtual Machine (TVM, the ISA being tested)
-
empty macro to specify the isa required for compilation of the test.
-
this is mandated to be present at the start of the test.
-
RVTEST_CODE_BEGIN-
-
start of code (test) section
-
macro to indicate test code start add and where test startup routine is inserted.
-
no part of the test-code section should precede this macro
-
this macro includes an initialization routine which pre-loads all the GPRs with unique values (not
0xdeadbeef). Register t0 and t1 are initialized to point to the labels :rvtest_code_beginandrvtest_code_endrespectively. -
the macros contains a label
rvtest_code_beginafter the above initilization routine to mark the begining of the actual test.
-
RVTEST_CODE_END-
-
end of code (test) section
-
macro to indicate test code end.
-
no part of the test-code section should follow after this macro.
-
the macro enforces a 16-byte boundary alignment
-
the macro also inlcudes the label
rvtest_code_endwhich marks the end of the actual test. -
if trap handling is enabled, this macro contains the entire trap handler code required by the test.
-
RVTEST_DATA_BEGIN-
-
marks the begining of the test data section
-
used to provided initialized data regions to be used by the test
-
this region starts at a 16-byte boundary
-
the start of this is macro can be addressed using the label:
rvtest_data_begin -
when trap handling is enabled, this macro also includes the following labels :
-
trapreg_sv: This region is used to save the temporary registers used in the trap-handler code
-
tramptbl_sv: This region is used to save the contents of the test-target’s initial code-section which is overwritten with the necessary trampoline table.
-
mtvec_save: a double-word region to save the test-target specific mtvec register
-
mscratch_save: a double-word region to save the test-target specific mscratch register
-
-
RVTEST_DATA_END-
-
this macros marks the end of the test input data section.
-
the start of this macro can be addressed using the label:
rvtest_data_end
-
RVTEST_CASE(CaseName, CondStr)-
-
execute this case only if condition in cond_str are met
-
caseName is arbitrary string
-
condStr is evaluated to determine if the test-case is enabled and sets name variable
-
condStr can also define compile time macros required for the test-case to be enabled.
-
the test-case must be delimited with an #ifdef CaseName/#endif pair
-
the format of CondStr can be found in https://riscof.readthedocs.io/en/latest/cond_spec.html#cond-spec
-
These macros are be defined by the owner of the test target in the file model_test.h. These macros are required to define the signature regions and also the logic required to halt/exit the test.
RVMODEL_DATA_BEGIN-
-
This macro marks the start of signature regions. The test-target should use this macro to create a label to indicate the begining of the signature region. For example :
.globl begin_signature; begin_signature. This macro must also begin at a 16-byte boundary and must not include anything else.
-
RVMODEL_DATA_END-
-
This macros marks the end of the signature-region. The test-target must declare any labels required to indicate the end of the signature region. For example :
.globl end_signature; end_signature. This label must be at a 16-byte boundary. The entire signature region must be included within the RVMODEL_DATA_BEGIN macro and the start of the RVMODEL_DATA_END macro. The RVMODEL_DATA_END macro can also contain other target specific data regions and initializations but only after the end of the signature.
-
RVMODEL_HALT-
-
This macros must define the test-target halt mechanism. This macro is called when the test is to be terminated either due to completion or dur to unsupported behavior. This macro could also include routines to dump the signature region to a file on the host system which can be used for comparison.
-
RVTEST_SIGBASE(BaseReg,Val)-
-
defines the base register used to update signature values
-
Register BaseReg is loaded with value Val
-
hidden_offset is initialized to zero
-
RVTEST_SIGUPD(BaseReg, SigReg [, Offset])-
-
if Offset is present in the arguments, hidden_offset if set to Offset
-
Sigreg is stored at hidden_offset[BaseReg]
-
hidden_offset is post incremented so repeated uses store signature values sequentially
-
RVTEST_BASEUPD(BaseReg[oldBase[,newOff]])-
-
[moves &] updates BaseReg past stored signature
-
Register BaseReg is loaded with the oldReg+newOff+hidden_offset
-
BaseReg is used if oldBase isn’t specified; 0 is used if newOff isn’t specified
-
hidden_offset is re-initialized to 0 afterwards
-
RVMODEL_BOOT-
-
contains boot code for the test-target; may include emulation code or trap stub. If the test-target enforces alignment or value restrictions on the mtvec csr, it is required that this macro sets the value of mtvec to a region which is readable and writable by the machine mode. May include code to copy the data sections from boot device to ram. Or any other code that needs to be run prior to running the tests.
-
RVMODEL_IO_INIT-
-
initializes IO for debug output
-
this must be invoked if any of the other RV_MODEL_IO_* macros are used
-
RVMODEL_IO_ASSERT_GPR_EQ(ScrReg, Reg, Value)-
-
debug assertion that GPR should have value
-
outputs a debug message if Reg!=Value
-
ScrReg is a scratch register used by the output routine; its final value cannot be guaranteed
-
Can be used to help debug what tests have passed/failed
-
RVMODEL_IO_WRITE_STR(ScrReg, String)-
-
output debug string, using a scratch register
-
outputs the message String
-
ScrReg is a scratch register used by the output routine; its final value cannot be guaranteed
-
RVMODEL_SET_MSW_INT-
-
This macro needs to include a routine to set the machine software interrupt.
-
Currently the test forces an empty macro if a target does not declare this. Future tests may change this.
-
RVMODEL_CLEAR_MSW_INT-
-
This macro needs to include a routine to clear the machine software interrupt.
-
Currently the test forces an empty macro if a target does not declare this. Future tests may change this.
-
RVMODEL_CLEAR_MTIMER_INT-
-
This macro needs to include a routine to clear the machine timer interrupt.
-
Currently the test forces an empty macro if a target does not declare this. Future tests may change this.
-
RVMODEL_CLEAR_MEXT_INT-
-
This macro needs to include a routine to clear the machine external interrupt.
-
Currently the test forces an empty macro if a target does not declare this. Future tests may change this.
-
The test must define a rvtest_entry_point label to indicate the location to be used by the linker
as the entry point in the test. Generally, this would be before the RVMODEL_BOOT macro and should
belong to the text.init section.
All tests shall use a signature approach. Each test shall be written in the same style, with defined mandatory items. The test structure of an architectural test shall have the following sections in the order as follows:
-
Header + license (including a specification link, a brief test description and RVTEST_ISA macro)).
-
Includes of header files (see Common Header Files section).
-
Test Virtual Machine (TVM) specification,
-
Test code between “RVTEST_CODE_BEGIN” and “RVTEST_CODE_END”.
-
Input data section, marked with "RVTEST_DATA_BEGIN" and "RVTEST_DATA_END".
-
Output data section between “RVMODEL_DATA_BEGIN” and “RVMODEL_DATA_END”.
- Note
-
Note that there is no requirement that the code or scratch data sections must be contiguous in memory, or that they be located before or after data or code sections (configured by embedded directives recognized by the linker)
There are the following common rules that shall be applied to each architectural test:
-
Always use “//” as commentary. “#” should be used only for includes and defines.
-
As part of the initialization code, all GPRs are preloaded with unique predefined values (which is not
0xdeadbeef). However, t0 is initialized withrvtest_code_beginand t1 is initialized withrvtest_data_begin. -
The signature section of every test is pre-loaded with the word
0xdeadbeef -
The signature region should always begin at a 16-byte boundary
-
A test shall be divided into logical blocks (test cases) according to the test goals. Test cases are enclosed in an
#ifdef <CaseName>, #endifpair and begin with the RVTEST_CASE(CaseName,CondStr) macro that specifies the test case name, and a string that defines the conditions under which that Test case can be selected for assembly and execution. Those conditions will be collected and used to generate the database which in turn is used to select tests for inclusion in the test suite for this target. -
Tests should use the RVTEST_SIGBASE(BaseReg,Val) macro to define the GPR used as a pointer to the output signature area, and its initial value. It can be used multiple times within a test to reassign the output area or change the base register. This value will be used by the invocations of the RVTEST_SIGUPD macro.
-
Tests should use the RVTEST_SIGUPD(BaseReg, SigReg, ScratchReg, Value) macro to store signature values using (only) the base register defined in the most recently encountered RVTEST_SIGBASE(BaseReg,Val) macro. Repeated uses will automatically have an increasing offset that is managed by the macro.
-
Uses of RVTEST_SIGUPD shall always be preceded sometime in the test case by RVTEST_SIGBASE.
-
Tests that use SIGUPD inside a loop or in any section of code that will be repeated (e.g. traps) must use the BASEUPD macro between each loop iteration or repeated code to ensure static values of the base and offset don’t overwrite older values.
-
-
When macros are needed for debug purposes, only macros from model_test.h shall be used. Note that using this feature shall not affect the signature results of the test run.
-
Test shall not include other tests (e.g. #include “../add.S”) to prevent non-complete tests, compilation issues, and problems with code maintenance.
-
Tests and test cases shall be skipped if not required for a specific model test configuration based on test conditions defined in the RVTEST_CASE macro. Tests that are selected may be further configured using variables (e.g. XLEN) which are passed into the tests and used to compile them. In either case, those conditions and variables are derived from the YAML specification of the device and execution environment that are passed into the framework. The flow is to run an architectural test suite built by the Architectural Test Framework from the architectural test pool to determine which tests and test cases to run.
-
Tests shall not depend on tool specific features. For example, tests shall avoid usage of internal GCC macros (e..g. __risc_xlen), specific syntax (char 'a' instead of 'a) or simulator features (e.g. tohost) etc.
-
A test will end by either jumping to or implicitly reaching the
RVTEST_CODE_ENDmacro (i.e. rvtest_code_end label). TheRVTEST_CODE_ENDmacro is always followed by theRVMODEL_HALTmacro. -
Macros defined outside of a test shall only be defined in specific predefined header files (see Common Header Files below), and once they are in use, they may be modified only if the function of all affected tests remains unchanged. It is acceptable that macros use may lead to operand repetition (register X is used every time).
-
The aim of this restriction is to have test code more readable and to avoid side effects which may occur when different contributors will include new architectural tests or updates of existing ones in the architectural test pool. This measure results from the negative experience, where the architectural test suite could be used just for one target while the architectural test code changes were necessary to have it also running for other targets.
-
-
All contents of the signature region must always be initialized to
0xdeadbeef. -
The result of no operation should be stored in the signature even though not register has been altered.
-
Pseudo ops other than
liandlawhich can map to multiple standard instruction sequences should not be used. -
The actual test-section of the assembly must always start with the
RVTEST_CODE_BEGINwhich contains a routine to initialize the registers to specific values.
Each test shall include only the following header files:
-
model_test.h – defines target-specific macros, both required and optional: (e.g. RVMODEL_xxx)
-
arch_test.h – defines pre-defined test macros both required and optional: (e.g. RVTEST_xxx)
The inclusion of the arch_test.h should always occur after the model_test.h file.
Important points to be noted regarding header files :
-
Adding new header files is forbidden in the test. It may lead to macro redefinition and compilation issues.
-
Macros maybe defined and used inside a test, as they will not be defined and used outside that specific test.
The framework will import files that describe
-
the implemented, target-specific configuration parameters in YAML format
-
the required, platform-specific configuration parameters in YAML format
The framework will generate intermediate files, including a Test Database YAML file that selects tests from the test pool to generate a test suite for the target.
The framework will also invoke the target shell as appropriate to cause tests to be built, loaded, executed, and results reported.
The YAML files define both the values of those conditions and values that can be used by the framework to configure tests (e.g. format of WARL CSR fields). Tests should not have #if, #ifdef, etc. for conditional assembly except those that surround RVMODEL_CASE macros Instead, each of those should be a separate test case whose conditions are defined in the common reference document entry for that test and test case number.
#This assembly file tests the add instruction of the RISC-V I extension for the add covergroup.
#include "model_test.h" #include "arch_test.h"
RVTEST_ISA("RV32I")
RVMODEL_BOOT
RVTEST_CODE_BEGIN
#ifdef TEST_CASE_1 // this test is meant for devices implementing rv32I extension and requires enabling the compile // macro TEST_CASE_1. This test will contribute to the "add" coverage label. RVTEST_CASE(0,"//check ISA:=regex(.*32.*);check ISA:=regex(.*I.*);def TEST_CASE_1=True;",add)
RVTEST_SIGBASE( x16,signature_x16_1) // x16 will point to signature_x16_1 label in the signature region
TEST_RR_OP(add, x9, x4, x6, 0x80000005, 0x80000000, 0x00000005, x16, 0, x24) TEST_RR_OP(add, x5, x5, x14, 0xfffffeff, 0x00000000, 0xfffffeff, x16, 4, x24) ... ...
// this will change the signature base register to x3. x3 will not point to signature_x3_0 in // the signature region RVTEST_SIGBASE( x3,signature_x3_0) // continue with new test cases .. TEST_RR_OP(add, x4, x24, x27, 0x55555955, 0x00000400, 0x55555555, x3, 0, x5) ... ...
RVTEST_CODE_END RVMODEL_HALT
RVTEST_DATA_BEGIN rvtest_data: .word 0xbabecafe RVTEST_DATA_END
RVMODEL_DATA_BEGIN
.align 4
signature_x16_0:
.fill 0*(XLEN/32),4,0xdeadbeef
signature_x16_1:
.fill 16*(XLEN/32),4,0xdeadbeef
signature_x3_0:
.fill 86*(XLEN/32),4,0xdeadbeef
#ifdef rvtest_mtrap_routine
mtrap_sigptr:
.fill 64*(XLEN/32),4,0xdeadbeef
#endif
#ifdef rvtest_gpr_save
gpr_save:
.fill 32*(XLEN/32),4,0xdeadbeef
#endif
RVMODEL_DATA_END