Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 1 | Change Log & Release Notes |
| 2 | ========================== |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 3 | |
| 4 | Please note that the Trusted Firmware-A Tests version follows the Trusted |
| 5 | Firmware-A version for simplicity. At any point in time, TF-A Tests version |
| 6 | `x.y` aims at testing TF-A version `x.y`. Different versions of TF-A and TF-A |
| 7 | Tests are not guaranteed to be compatible. This also means that a version |
| 8 | upgrade on the TF-A-Tests side might not necessarily introduce any new feature. |
| 9 | |
Bipin Ravi | 024685e | 2021-05-06 14:32:52 -0500 | [diff] [blame^] | 10 | Version 2.5 |
| 11 | ----------- |
| 12 | |
| 13 | New features |
| 14 | ^^^^^^^^^^^^ |
| 15 | - More tests are made available in this release to help validate the |
| 16 | functionalities in the following areas: |
| 17 | - True Random Number Generator (TRNG) test scenarios. |
| 18 | - Multicore / Power State Controller Interface (PSCI) tests. |
| 19 | - v8.6 Activity Monitors Unit (AMU) enhancements test scenarios. |
| 20 | - Secure Partition Manager (SPM) / Firmware Framework (FF-A) v1.0 testing. |
| 21 | - Interrupt Handling between Non-secure and Secure world. |
| 22 | - Direct messages and memory sharing between Secure Partitions(SP). |
| 23 | - Many tests to exercise FF-A v1.0 ABIs. |
| 24 | - SPM saving/restoring the NS SIMD context enabling a normal world FF-A |
| 25 | endpoint (TFTF) and a secure partition to use SIMD vectors and |
| 26 | instructions independently. |
| 27 | |
| 28 | TFTF |
| 29 | ~~~~ |
| 30 | |
| 31 | - SPM / FF-A v1.0 testing. |
| 32 | - Refactor FF-A memory sharing tests |
| 33 | - Created helper functions to initialize ffa_memory_region and to send |
| 34 | the respective memory region to the SP, making it possible to reuse |
| 35 | the logic in SP-to-SP memory share tests. |
| 36 | - Added comments to document relevant aspects about memory sharing. |
| 37 | |
| 38 | - Trigger direct messaging between SPs. |
| 39 | - Use cactus command 'CACTUS_REQ_ECHO_SEND_CMD' to make cactus SPs |
| 40 | communicate with each other using direct message interfaces. |
| 41 | |
| 42 | - Added helpers for SPM tests. |
| 43 | - Checking SPMC has expected FFA_VERSION. |
| 44 | - Checking that expected FF-A endpoints are deployed in the system. |
| 45 | - Getting global TFTF mailbox. |
| 46 | |
| 47 | - Replace '.inst' AArch64 machine directives with CPU Memory Tagging Extension |
| 48 | instructions in 'test_mte_instructions' function. |
| 49 | |
| 50 | - Add build option for Arm Feature Modifiers. |
| 51 | - This patch adds a new ARM_ARCH_FEATURE build option to add support |
| 52 | for compiler's feature modifiers. |
| 53 | |
| 54 | - Enable 8 cores support for Theodul DSU(DynamIQ Shared Unit) for the |
| 55 | Total Compute (TC0) platform. |
| 56 | |
| 57 | - New tests: |
| 58 | |
| 59 | - Remove redundant code and add better tests for TRNG SMCs. |
| 60 | - Tests that the Version, Features, and RND calls conform to the spec. |
| 61 | |
| 62 | - New tests for v8.6 AMU enhancements (FEAT_AMUv1p1) |
| 63 | - Make sure AMU offsets are being saved and restored properly. |
| 64 | |
| 65 | - Tests to request SP-to-SP memory share. |
| 66 | |
| 67 | - SP-to-SP direct messaging deadlock test. |
| 68 | - TFTF sends CACTUS_REQ_DEADLOCK_CMD to cactus SP. |
| 69 | |
| 70 | Cactus(Secure-EL1 test partition) |
| 71 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 72 | |
| 73 | - Enable managed exit for primary cactus secure partition. |
| 74 | |
| 75 | - Helper commands needed for interrupt testing. |
| 76 | |
| 77 | - Add handler from managed exit FIQ interrupt. |
| 78 | |
| 79 | - Make ffa_id global. |
| 80 | |
| 81 | - Implement HF_INTERRUPT_ENABLE Hafnium hypervisor call wrapper. With this |
| 82 | service, a secure partition calls into the SPMC to enable/disable a |
| 83 | particular virtual interrupt. |
| 84 | |
| 85 | - Invalidate the data cache for the cactus image. |
| 86 | |
| 87 | - Helper commands needed for interrupt testing. |
| 88 | - CACTUS_SLEEP_CMD & CACTUS_INTERRUPT_CMD added. |
| 89 | |
| 90 | - Decouple exception handling from tftf framework. |
| 91 | - With new interrupt related tests coming up in Cactus, added separate |
| 92 | exception handler code for irq/fiq in Cactus. |
| 93 | |
| 94 | - Hypervisor calls moved to a separate module. |
| 95 | |
| 96 | - Add secondary entry point register function. |
| 97 | |
| 98 | - Declare third SP instance as UP SP. |
| 99 | |
| 100 | - Provision a cold boot path for secondary cores (or secondary pinned |
| 101 | execution contexts). |
| 102 | |
| 103 | - Tidy message loop, commands definitions, direct messaging API definitions. |
| 104 | |
| 105 | - Helpers for error logging after FF-A calls. |
| 106 | |
| 107 | - Properly placing Cactus test files. |
| 108 | |
| 109 | - Tidying FF-A Memory Sharing tests. |
| 110 | |
| 111 | - Use CACTUS_ECHO_CMD in direct message tests. |
| 112 | |
| 113 | - Refactor handling of commands. |
| 114 | - Added helper macros to define a command handler, build a command table |
| 115 | in which each element is a pair of the handler and respective command |
| 116 | ID. Available tests have been moved to their own command handler. |
| 117 | |
| 118 | - Extend arguments in commands responses. |
| 119 | - In the test commands framework, added template to extend number of |
| 120 | values to include in a command response. |
| 121 | |
| 122 | - Check FF-A return is a valid direct response. |
| 123 | - Added a helper function to check if return of FFA_MSG_SEND_DIRECT_REQ |
| 124 | is FFA_MSG_SEND_DIRECT_RESP. |
| 125 | |
| 126 | - FFA_MSG_DIRECT_RESP call extended to use 5 registers. |
| 127 | |
| 128 | - Added accessors for arguments from FF-A calls. |
| 129 | - Some accessors for arguments from FF-A calls, namely for func id, error |
| 130 | code, and direct message destination/source. |
| 131 | |
| 132 | - Use virtual counter for sp_sleep. |
| 133 | - Changes sp_sleep() to use virtual counter instead of physical counter. |
| 134 | |
| 135 | - Checks if SIMD vectors are preserved in the normal world while transitioning |
| 136 | from normal world to secure world and back to normal world. |
| 137 | |
| 138 | - Tidying common code to tftf and cactus. |
| 139 | |
| 140 | - Refactor cactus_test_cmds.h to incorporate static inline functions instead |
| 141 | of macros to enforce type checking. |
| 142 | |
| 143 | - Removed reference to Hafnium in name from helper function and macro to |
| 144 | make them generic. |
| 145 | |
| 146 | - For consistency added the cmd id 'CACTUS_MEM_SEND_CMD'. |
| 147 | |
| 148 | - Add command to request memory sharing between SPs. |
| 149 | |
| 150 | - Add & handle commands 'CACTUS_REQ_ECHO_CMD' and 'CACTUS_ECHO_CMD'. |
| 151 | |
| 152 | - Update README with list of sample partitions. |
| 153 | |
| 154 | - Remove reference to PSA from xml test file. |
| 155 | |
| 156 | - Reduce tests verbosity in release mode. |
| 157 | - Update few NOTICE messages to VERBOSE/INFO. |
| 158 | |
| 159 | - Fix conversion issues on cactus responses. |
| 160 | |
| 161 | - Create RXTX map/configure helper macros and use them. |
| 162 | |
| 163 | - Update OP-TEE version used for testing to 3.10. |
| 164 | - SPMC as S-EL1 tests using OP-TEE depend on a static binary stored as |
| 165 | a CI file. This binary corresponds to a build of OP-TEE v3.10. |
| 166 | |
| 167 | - Add uart2 to device-regions node. |
| 168 | - First SP no longer has an open access to the full system peripheral |
| 169 | range and devices must be explicitly declared in the SP manifest. |
| 170 | |
| 171 | - New tests: |
| 172 | |
| 173 | - Test for exercising SMMUv3 driver to perform stage2 translation. |
| 174 | |
| 175 | - Test handling of non-secure interrupt while running SP. |
| 176 | |
| 177 | - Add secondary cores direct messaging test for SPM. |
| 178 | |
| 179 | - Testing deadlock by FF-A direct message. |
| 180 | - Added command CACTUS_DEADLOCK_CMD to file cactus_test_cmds.h to create |
| 181 | a deadlock scenario using FF-A direct message interfaces. |
| 182 | |
| 183 | - Test SP-to-SP memory share operations |
| 184 | - Handle 'CACTUS_REQ_MEM_SEND_CMD' by sending memory to the receiver SP. |
| 185 | |
| 186 | - Implemented test to validate FFA_RXTX_MAP ABI. |
| 187 | |
Manish V Badarkhe | 983f195 | 2020-10-04 18:33:44 +0100 | [diff] [blame] | 188 | Version 2.4 |
| 189 | ----------- |
| 190 | |
| 191 | New features |
| 192 | ^^^^^^^^^^^^ |
| 193 | - More tests are made available in this release to help validate the |
| 194 | functionalities in the following areas: |
| 195 | - SMCCC. |
| 196 | - New architecture specific features. |
| 197 | - FF-A features. |
| 198 | - New platform ports. |
| 199 | |
| 200 | - Various improvements to test framework and test suite such as documentation, |
| 201 | removing un-necessary dependencies, etc. |
| 202 | |
| 203 | TFTF |
| 204 | ~~~~ |
| 205 | |
| 206 | - Remove dependencies from FVP to generic code by converting some FVP platform |
| 207 | specific macros to the common macros. |
| 208 | |
| 209 | - Remove make as a package dependency to compile TF-A test code. |
| 210 | |
| 211 | - Move defaults values and macro defs in a separate folder from Makefile. |
| 212 | |
| 213 | - Allow alternate stdout to be used apart from pl011 UART. |
| 214 | |
| 215 | - Get FVP platform's topology from build options to make FVP platform |
| 216 | configuration more flexible and eliminate test errors when the platform |
| 217 | is configured with number of CPUs less than default values in the makefile. |
| 218 | |
| 219 | - Update the FIP corrupt address which is used to corrupt BL2 image that helps |
| 220 | to trigger firmware update process. |
| 221 | |
| 222 | - Add explicit barrier before sev() in tftf_send_event_common API to avoid |
| 223 | core hang. |
| 224 | |
| 225 | - Align output properly on issuing make help_tests by removing dashes |
| 226 | and sort tests. |
| 227 | |
| 228 | - Moved a few FVP and Juno specific defined from common header files to platform |
| 229 | specific header files. |
| 230 | |
| 231 | - Replace SPCI with PSA FF-A in code as SPCI is now called as FF-A. |
| 232 | |
| 233 | - Add owner field to sp_layout generation to differentiate owner of SP which |
| 234 | could either be Silicon Provider or Platform provider. |
| 235 | |
| 236 | - Add v8.5 Branch Target Identifier(BTI) support in TFTF. |
| 237 | |
| 238 | - Remove dependency on SYS_CNT_BASE1 to read the memory mapped timers. |
| 239 | |
| 240 | - Enables SError aborts for all CPUs, during their power on sequence. |
| 241 | |
| 242 | - Documentation: |
| 243 | |
| 244 | - Use conditional assignment on sphinx variables so that they can be |
| 245 | overwritten by environment and/or command line. |
| 246 | |
| 247 | - Add support for documentation build as a target in Makefile. |
| 248 | |
| 249 | - Update list of maintainers. |
| 250 | |
| 251 | - Update documentation to explain how to locally build the documentation. |
| 252 | |
| 253 | - Add .editorconfig from TF-A to define the coding style. |
| 254 | |
| 255 | - Fix documentation to include 'path/to' prefix when specifying tftf.bin on |
| 256 | make fip cmd. |
| 257 | |
| 258 | - Use docker to build documentation. |
| 259 | |
| 260 | - Replace SPCI with PSA FF-A in documentation as SPCI is now called |
| 261 | as FF-A. |
| 262 | |
| 263 | - NVIDIA Tegra194: |
| 264 | |
| 265 | - Skip CPU suspend tests requiring SGI as wake source as Tegra194 platforms |
| 266 | do not support CPU suspend power down and cannot be woken up with an SGI. |
| 267 | |
| 268 | - Disable some system suspend test cases. |
| 269 | |
| 270 | - Create dummy SMMU context for system resume to allow the System Resume |
| 271 | Firmware to complete without any errors or warnings. |
| 272 | |
| 273 | - Increase RTC step value to 5ms as RTC consumes 250us for each register |
| 274 | read/write. Increase the step value to 5ms to cover all the register |
| 275 | read/write in program_timer(). |
| 276 | |
| 277 | - Skip some timer framework validation tests as CPUs on Tegra194 platforms |
| 278 | cannot be woken up with the RTC timer interrupt after power off. |
| 279 | |
| 280 | - Introduce per-CPU Hypervisor Timer Interrupt ID. |
| 281 | |
| 282 | - Skip PSCI STAT tests requiring PSTATE_TYPE_POWERDOWN as Tegra194 platforms |
| 283 | do not support CPU suspend with state type as PSTATE_TYPE_POWERDOWN. |
| 284 | |
| 285 | - Disable boot requirement tests as Tegra194 platforms do not support memory |
| 286 | mapped timers. |
| 287 | |
| 288 | - Skips the test "Create all power states and validate EL3 power state parsing" |
| 289 | from the "EL3 power state parser validation" test suite as it is not in |
| 290 | sync with this expectation. |
| 291 | |
| 292 | - Moved reset, timers. wake, watchdog drivers from Tegra194 specific folder to |
| 293 | common driver folder so that these drivers can be used for other NVIDIA platforms. |
| 294 | |
| 295 | - New tests: |
| 296 | |
| 297 | - Add test for SDEI RM_ANY routing mode. |
| 298 | |
| 299 | - Add initial platform support for TC0. |
| 300 | |
| 301 | - Add SMC fuzzing module test. |
| 302 | |
| 303 | - Add test case for SMCCC_ARCH_SOC_ID feature. |
| 304 | |
| 305 | - Add test that supports ARMv8.6-FGT in TF-A. |
| 306 | |
| 307 | - Add test that supports ARMv8.6-ECV in TF-A. |
| 308 | |
| 309 | - Add test for FFA_VERSION interface. |
| 310 | |
| 311 | - Add test for FFA_FEATURES interface. |
| 312 | |
| 313 | - Add console driver for the TI UART 16550. |
| 314 | |
| 315 | - Add tests for FF-A memory sharing interfaces between tftf |
| 316 | and cactus secure partitions. |
| 317 | |
| 318 | - NVIDIA Tegra194: |
| 319 | |
| 320 | - Introduce platform port for Tegra194 to to initialize the tftf |
| 321 | framework and execute tests on the CPUs. |
| 322 | |
| 323 | - Introduce power management support. |
| 324 | |
| 325 | - Introduce support for RTC as wake source. |
| 326 | |
| 327 | - Introduce system reset functionality test. |
| 328 | |
| 329 | - Introduce watchdog timer test. |
| 330 | |
| 331 | - Introduce support for NVIDIA Denver CPUs. |
| 332 | |
| 333 | - Introduce RAS uncorrectable error injection test. |
| 334 | |
| 335 | - Introduce tests to verify the Video Memory resize interface. |
| 336 | |
| 337 | - Introduce test to inject RAS corrected errors for all supported |
| 338 | nodes from all CPUs. |
| 339 | |
| 340 | - Introduce a test to get return value from SMC SiP function |
| 341 | TEGRA_SIP_GET_SMMU_PER. |
| 342 | |
| 343 | - NVIDIA Tegra196: |
| 344 | |
| 345 | - Introduce initial support for Tegra186 platforms. |
| 346 | |
| 347 | - NVIDIA Tegra210: |
| 348 | |
| 349 | - Introduce initial support for Tegra210 platforms. |
| 350 | |
| 351 | Secure partition - Cactus |
| 352 | ~~~~~~~~~~~~~~~~~~~~~~~~~ |
| 353 | |
| 354 | - TFTF doesn't need to boot Secondary Cactus as Hafnium now boots all |
| 355 | partitions according to "boot-order" field value in the partition |
| 356 | manifests. |
| 357 | |
| 358 | - Remove test files related to deprecated SPCI Alpha specification and |
| 359 | SPRT interface. |
| 360 | |
| 361 | - Select different stdout device at runtime as primary VM can access |
| 362 | to UART while secondary VM's use hypervisor call to SPM for debug |
| 363 | logging. |
| 364 | |
| 365 | - An SP maps its RX/TX buffers in its EL1&0 Stage-1 translation regime. |
| 366 | The same RX/TX buffers are mapped by the SPMC in the SP's EL1&0 |
| 367 | Stage-2 translation regime during boot time. |
| 368 | |
| 369 | - Update memory/device region nodes in manifest. Memory region has 3 |
| 370 | entries such as RX buffer, TX buffer and dummy. These memory region |
| 371 | entries are mapped with attributes as "RX buffer: read-only", |
| 372 | "TX buffer: read-write" and "dummy: read-write-execute". |
| 373 | Device region mapped with read-write attribute. |
| 374 | |
| 375 | - Create tertiary partition without RX_TX region specified to test the |
| 376 | RXTX_MAP API. |
| 377 | |
| 378 | - Add third partition to ffa_partition_info_get test to test that a |
| 379 | partition can successfully get information about the third cactus |
| 380 | partition. |
| 381 | |
| 382 | - Map RXTX region to third partition to point the mailbox to this RXTX |
| 383 | region. |
| 384 | |
| 385 | - Adjust the number of EC context to max number of PEs as per the FF-A |
| 386 | specification mandating that a SP must either "Implement as many ECs |
| 387 | as the number of PEs (in case of a "multi-processor" SP with pinned |
| 388 | contexts)" or "Implement a single EC (in case of a migratable |
| 389 | "uni-processor" SP). |
| 390 | |
| 391 | - Updated cactus test payload and TFTF ids as it is decided to have |
| 392 | secure partition FF-A ids in the range from 0x8001 to 0xfffe, 0x8000 |
| 393 | and 0xffff FF-A ids are reserved for the SPMC and the SPMD respectively |
| 394 | and in the non-secure worlds, FF-A id 0 is reserved for the hypervisor |
| 395 | and 1 to 0x7fff FF-A ids are reserved for VMs. |
| 396 | |
| 397 | - Break the message loop on bad message request instead of replying |
| 398 | with the FF-A error ABI to the SPMC. |
| 399 | |
| 400 | - Remove deprecated hypervisor calls spm_vm_get_count and spm_vcpu_get_count. |
| 401 | Instead use FFA_PARTITION_INFO_GET discovery ABI. |
| 402 | |
| 403 | - Implement hvc call 'SPM_INTERRUPT_GET' to get interrupt id. |
| 404 | |
| 405 | - Re-structure platform dependent files by moving platform dependent files |
| 406 | and macros to platform specific folder. |
| 407 | |
| 408 | - Adjust partition info get properties to support receipt of direct |
| 409 | message request. |
| 410 | |
| 411 | - New tests: |
| 412 | |
| 413 | - Add FFA Version Test. |
| 414 | |
| 415 | - Add FFA_FEATURES test. |
| 416 | |
| 417 | - Add FFA_MEM_SHARE test |
| 418 | |
| 419 | - Add FFA_MEM_LEND test. |
| 420 | |
| 421 | - Add FFA_MEM_DONATE test. |
| 422 | |
| 423 | - Add FFA_PARTITION_INFO_GET test. |
| 424 | |
| 425 | - Add exception/interrupt framework. |
| 426 | |
| 427 | - Add cactus support for TC0 platform. |
| 428 | |
| 429 | Issues resolved since last release |
| 430 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 431 | |
| 432 | - Update link to SMCCC specification. |
| 433 | |
| 434 | - Trim down the top-level readme file to give brief overview of the project |
| 435 | and also fix/update a number of broken/out-dated links in it. |
| 436 | |
| 437 | - Bug fix in Multicore IRQ spurious test. |
| 438 | |
| 439 | - Fix memory regions mapping with no NS bit set. |
| 440 | |
| 441 | - Reenable PSCI NODE_HW_STATE test which was disabled earlier due to |
| 442 | outdated SCP firmware. |
| 443 | |
| 444 | - Fix Aarch32 zeromem() function by avoiding infinite loop in 'zeromem' |
| 445 | function and optimizing 'memcpy4' function. |
| 446 | |
| 447 | - Add missing help_tests info on help target in the top-level Makefile. |
| 448 | |
| 449 | - Trim down the readme file as it does not need to provide detailed |
| 450 | information, instead it can simply be a landing page providing a brief |
| 451 | overview of the project and redirecting the reader to RTD for further |
| 452 | information. |
| 453 | |
| 454 | - Fix maximum number of CPUs in DSU cluster by setting maximum number of CPUs |
| 455 | in DSU cluster to 8. |
| 456 | |
Madhukar Pappireddy | 6ec644e | 2020-04-08 15:44:38 -0500 | [diff] [blame] | 457 | Version 2.3 |
| 458 | ----------- |
| 459 | |
| 460 | New features |
| 461 | ^^^^^^^^^^^^ |
| 462 | |
| 463 | - More tests are made available in this release to help validate |
| 464 | the functionality of TF-A. |
| 465 | |
| 466 | - CI upgraded to use GCC 9.2-2019.12 toolchain for tf-a-tests. |
| 467 | |
| 468 | - Various improvements to test framework and test suite. |
| 469 | |
| 470 | TFTF |
| 471 | ~~~~ |
| 472 | |
| 473 | - Support for extended register usage as per SMCCC v1.2 specification. |
| 474 | |
| 475 | - Support for FVP platforms with SMT capabilities. |
| 476 | |
| 477 | - Improved support for documentation through addition of basic Sphinx |
| 478 | configuration and Makefile similar to TF-A repository. |
| 479 | |
| 480 | - Enhancement to libc library synchronous to TF-A code base. |
| 481 | |
| 482 | - ARMv8.3-PAuth enabled for all FWU tests in TFTF. |
| 483 | |
| 484 | - TFTF made RFC 4122 compliant by converting UUIDs to network order format. |
| 485 | |
| 486 | - Build improvement by deprecating custom AARCH64/AARCH32 macros in favor of |
| 487 | __arch64__ macro provided by compiler. |
| 488 | |
| 489 | - Support for HVC as a SMCCC conduit in TFTF. |
| 490 | |
| 491 | - New tests: |
| 492 | |
| 493 | - AArch32 tests for checking if PMU counters leak in secure world. |
| 494 | |
| 495 | - Add new debug filesystem (debugfs) test. |
| 496 | |
| 497 | - Add a SPCI direct messaging test targeting bare-metal cactus SP. |
| 498 | |
| 499 | |
| 500 | Secure partitions |
| 501 | ~~~~~~~~~~~~~~~~~ |
| 502 | |
| 503 | Cactus |
| 504 | ~~~~~~ |
| 505 | |
| 506 | - Several build improvements and symbol relocation fixup to make it position |
| 507 | independent executable. |
| 508 | |
| 509 | - Update of sample manifest to SPCI Beta1 format. |
| 510 | |
| 511 | - Support for generating JSON file as required by TF-A. |
| 512 | |
| 513 | Issues resolved since last release |
| 514 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 515 | |
| 516 | - Makefile bug fix for performing parallel builds. |
| 517 | |
| 518 | - Add missing D-cache invalidation of RW memory in tftf_entrypoint to safeguard |
| 519 | against possible corruption. |
| 520 | |
| 521 | - Fixes in GIC drivers to support base addresses beyond 4G range. |
| 522 | |
| 523 | - Fix build with XML::LibXML 2.0202 Perl module |
| 524 | |
| 525 | Known issues and limitations |
| 526 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 527 | |
| 528 | The sections below list the known issues and limitations of each test image |
| 529 | provided in this repository. Unless and otherwise stated, issues and limitations |
| 530 | stated in previous release continue to exist in this release. |
| 531 | |
| 532 | TFTF |
| 533 | ~~~~ |
| 534 | - NODE_HW_STATE test has been temporarily disabled for sgi575 platform due to a |
| 535 | dependency on SCP binaries version 2.5 |
| 536 | |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 537 | Version 2.2 |
| 538 | ----------- |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 539 | |
| 540 | New features |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 541 | ^^^^^^^^^^^^ |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 542 | |
| 543 | - A wide range of tests are made available in this release to help validate |
| 544 | the functionality of TF-A. |
| 545 | |
| 546 | - Various improvements to test framework and test suite. |
| 547 | |
| 548 | TFTF |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 549 | ~~~~ |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 550 | |
| 551 | - Enhancement to xlat table library synchronous to TF-A code base. |
| 552 | |
| 553 | - Enabled strict alignment checks (SCTLR.A & SCTLR.SA) in all images. |
| 554 | |
| 555 | - Support for a simple console driver. Currently it serves as a placeholder |
| 556 | with empty functions. |
| 557 | |
| 558 | - A topology helper API is added in the framework to get parent node info. |
| 559 | |
| 560 | - Support for FVP with clusters having upto 8 CPUs. |
| 561 | |
| 562 | - Enhanced linker script to separate code and RO data sections. |
| 563 | |
| 564 | - Relax SMC calls tests. The SMCCC specification recommends Trusted OSes to |
| 565 | mitigate the risk of leaking information by either preserving the register |
| 566 | state over the call, or returning a constant value, such as zero, in each |
| 567 | register. Tests only allowed the former behaviour and have been extended to |
| 568 | allow the latter as well. |
| 569 | |
| 570 | - Pointer Authentication enabled on warm boot path with individual APIAKey |
| 571 | generation for each CPU. |
| 572 | |
| 573 | - New tests: |
| 574 | |
| 575 | - Basic unit tests for xlat table library v2. |
| 576 | |
| 577 | - Tests for validating SVE support in TF-A. |
| 578 | |
| 579 | - Stress tests for dynamic xlat table library. |
| 580 | |
| 581 | - PSCI test to measure latencies when turning ON a cluster. |
| 582 | |
| 583 | - Series of AArch64 tests that stress the secure world to leak sensitive |
| 584 | counter values. |
| 585 | |
| 586 | - Test to validate PSCI SYSTEM_RESET call. |
| 587 | |
| 588 | - Basic tests to validate Memory Tagging Extensions are being enabled and |
| 589 | ensuring no undesired leak of sensitive data occurs. |
| 590 | |
| 591 | - Enhanced tests: |
| 592 | |
| 593 | - Improved tests for Pointer Authentication support. Checks are performed |
| 594 | to see if pointer authentication keys are accessible as well as validate |
| 595 | if secure keys are being leaked after a PSCI version call or TSP call. |
| 596 | |
| 597 | - Improved AMU test to remove unexecuted code iterating over Group1 counters |
| 598 | and fix the conditional check of AMU Group0 counter value. |
| 599 | |
| 600 | Secure partitions |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 601 | ~~~~~~~~~~~~~~~~~ |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 602 | |
| 603 | A new Secure Partition Quark is introduced in this release. |
| 604 | |
| 605 | Quark |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 606 | ~~~~~ |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 607 | |
| 608 | The Quark test secure partition provided is a simple service which returns a |
| 609 | magic number. Further, a simple test is added to test if Quark is functional. |
| 610 | |
| 611 | Issues resolved since last release |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 612 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 613 | |
| 614 | - Bug fix in libc memchr implementation. |
| 615 | |
| 616 | - Bug fix in calculation of number of CPUs. |
| 617 | |
| 618 | - Streamlined SMC WORKAROUND_2 test and fixed a false fail on Cortex-A76 CPU. |
| 619 | |
| 620 | - Pointer Authentication support is now available for secondary CPUs and the |
| 621 | corresponding tests are stable in this release. |
| 622 | |
| 623 | Known issues and limitations |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 624 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 625 | |
| 626 | The sections below list the known issues and limitations of each test image |
| 627 | provided in this repository. Unless and otherwise stated, issues and limitations |
| 628 | stated in previous release continue to exist in this release. |
| 629 | |
| 630 | TFTF |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 631 | ~~~~ |
Madhukar Pappireddy | 0468ac0 | 2019-10-10 02:30:40 -0500 | [diff] [blame] | 632 | - Multicore spurious interrupt test is observed to have unstable behavior. As a |
| 633 | temporary solution, this test is skipped for AArch64 Juno configurations. |
| 634 | |
| 635 | - Generating SVE instructions requires `O3` compilation optimization. Since the |
| 636 | current build structure does not allow compilation flag modification for |
| 637 | specific files, the function which tests support for SVE has been pre-compiled |
| 638 | and added as an assembly file. |
| 639 | |
| 640 | |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 641 | |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 642 | Version 2.1 |
| 643 | ----------- |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 644 | |
| 645 | New features |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 646 | ^^^^^^^^^^^^ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 647 | |
| 648 | - Add initial support for testing Secure Partition Client Interface (SPCI) |
| 649 | and Secure Partition Run-Time (SPRT) standards. |
| 650 | |
| 651 | Exercise the full communication flow throughout the software stack, involving: |
| 652 | |
| 653 | - A Secure-EL0 test partition as the Trusted World agent. |
| 654 | |
| 655 | - TFTF as the Normal World agent. |
| 656 | |
| 657 | - The Secure Partition Manager (SPM) in TF-A. |
| 658 | |
| 659 | - Various stability improvements, code refactoring and clean ups. |
| 660 | |
| 661 | TFTF |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 662 | ~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 663 | |
| 664 | - Reorganize tests build infrastructure to allow the selection of a subset of |
| 665 | tests. |
| 666 | |
| 667 | - Reorganize the platform layer for improved clarity and simplicity. |
| 668 | |
| 669 | - Sanitise inclusion of drivers header files. |
| 670 | |
| 671 | - Enhance the test report format for improved clarity and conciseness. |
| 672 | |
| 673 | - Dump CPU registers when hitting an unexpected exception. Previously, this |
| 674 | would silently loop forever. |
| 675 | |
| 676 | - Import libc from TF-A to better align the two code bases. |
| 677 | |
| 678 | - New tests: |
| 679 | |
| 680 | - SPM tests for exercising communication through either the MM or SPCI/SPRT |
| 681 | interfaces. |
| 682 | |
| 683 | - SMC calling convention tests. |
| 684 | |
| 685 | - Initial tests for Armv8.3 Pointer Authentication support (experimental). |
| 686 | |
| 687 | - New platform ports: |
| 688 | |
| 689 | - `Arm SGI-575`_ FVP. |
| 690 | |
| 691 | - Hikey960 board (experimental). |
| 692 | |
| 693 | - `Arm Neoverse Reference Design N1 Edge (RD-N1-Edge)`_ FVP (experimental). |
| 694 | |
| 695 | Secure partitions |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 696 | ~~~~~~~~~~~~~~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 697 | |
| 698 | We now have 3 Secure Partitions to test the SPM implementation in TF-A. |
| 699 | |
| 700 | Cactus-MM |
| 701 | ''''''''' |
| 702 | |
| 703 | The Cactus test secure partition provided in version 2.0 has been renamed into |
| 704 | "*Cactus-MM*". It is still responsible for testing the SPM implementation based |
| 705 | on the Arm Management Mode Interface. |
| 706 | |
| 707 | Cactus |
| 708 | '''''' |
| 709 | |
| 710 | This is a new test secure partition (as the former "*Cactus*" has been renamed |
| 711 | into "*Cactus-MM*", see above). |
| 712 | |
| 713 | Unlike *Cactus-MM*, this image tests the SPM implementation based on the SPCI |
| 714 | and SPRT draft specifications. |
| 715 | |
| 716 | It runs in Secure-EL0 and performs the following tasks: |
| 717 | |
| 718 | - Test that TF-A has correctly setup the secure partition environment (access |
| 719 | to cache maintenance operations, to floating point registers, etc.) |
| 720 | |
| 721 | - Test that TF-A accepts to change data access permissions and instruction |
| 722 | permissions on behalf of Cactus for memory regions the latter owns. |
| 723 | |
| 724 | - Test communication with SPM through SPCI/SPRT interfaces. |
| 725 | |
| 726 | Ivy |
| 727 | ''' |
| 728 | |
| 729 | This is also a new test secure partition. It is provided in order to test |
| 730 | multiple partitions support in TF-A. It is derived from Cactus and essentially |
| 731 | provides the same services but with different identifiers at the moment. |
| 732 | |
| 733 | EL3 payload |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 734 | ~~~~~~~~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 735 | |
| 736 | - New platform ports: |
| 737 | |
| 738 | - `Arm SGI-575`_ FVP. |
| 739 | |
| 740 | - `Arm Neoverse Reference Design N1 Edge (RD-N1-Edge)`_ FVP (experimental). |
| 741 | |
| 742 | Issues resolved since last release |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 743 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 744 | |
| 745 | - The GICv2 spurious IRQ test is no longer Juno-specific. It is now only |
| 746 | GICv2-specific. |
| 747 | |
| 748 | - The manual tests in AArch32 state now work properly. After investigation, |
| 749 | we identified that this issue was not AArch32 specific but concerned any |
| 750 | test relying on state information persisting across reboots. It was due to |
| 751 | an incorrect build configuration. |
| 752 | |
| 753 | - Cactus-MM now successfully links with GNU toolchain 7.3.1. |
| 754 | |
| 755 | Known issues and limitations |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 756 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 757 | |
| 758 | The sections below lists the known issues and limitations of each test image |
| 759 | provided in this repository. |
| 760 | |
| 761 | TFTF |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 762 | ~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 763 | |
| 764 | The TFTF test image might be conceptually sub-divided further in 2 parts: the |
| 765 | tests themselves, and the test framework they are based upon. |
| 766 | |
| 767 | Test framework |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 768 | ~~~~~~~~~~~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 769 | |
| 770 | - Some stability issues. |
| 771 | |
| 772 | - No mechanism to abort tests when they time out (e.g. this could be |
| 773 | implemented using a watchdog). |
| 774 | |
| 775 | - No convenient way to include or exclude tests on a per-platform basis. |
| 776 | |
| 777 | - Power domains and affinity levels are considered equivalent but they may |
| 778 | not necessarily be. |
| 779 | |
| 780 | - Need to provide better support to alleviate duplication of test code. There |
| 781 | are some recurrent test patterns for which helper functions should be |
| 782 | provided. For example, bringing up all CPUs on the platform and executing the |
| 783 | same function on all of them, or programming an interrupt and waiting for it |
| 784 | to trigger. |
| 785 | |
| 786 | - Every CPU that participates in a test must return from the test function. If |
| 787 | it does not - e.g. because it powered itself off for testing purposes - then |
| 788 | the test framework will wait forever for this CPU. This limitation is too |
| 789 | restrictive for some tests. |
| 790 | |
| 791 | - No protection against interrupted flash operations. If the target is reset |
| 792 | while some data is written to flash, the test framework might behave |
| 793 | incorrectly on reset. |
| 794 | |
| 795 | - When compiling the code, if the generation of the ``tests_list.c`` and/or |
| 796 | ``tests_list.h`` files fails, the build process is not aborted immediately |
| 797 | and will only fail later on. |
| 798 | |
| 799 | - The directory layout requires further improvements. Most of the test |
| 800 | framework code has been moved under the ``tftf/`` directory to better isolate |
| 801 | it but this effort is not complete. As a result, there are still some TFTF |
| 802 | files scattered around. |
| 803 | |
| 804 | - Pointer Authentication testing is experimental and incomplete at this stage. |
| 805 | It is only enabled on the primary CPU on the cold boot. |
| 806 | |
| 807 | Tests |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 808 | ~~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 809 | |
| 810 | - Some tests are implemented for AArch64 only and are skipped on AArch32. |
| 811 | |
| 812 | - Some tests are not robust enough: |
| 813 | |
| 814 | - Some tests might hang in some circumstances. For example, they might wait |
| 815 | forever for a condition to become true. |
| 816 | |
| 817 | - Some tests rely on arbitrary time delays instead of proper synchronization |
| 818 | when executing order-sensitive steps. |
| 819 | |
| 820 | - Some tests have been implemented in a practical manner: they seem to work |
| 821 | on actual hardware but they make assumptions that are not guaranteed by |
| 822 | the Arm architecture. Therefore, they might fail on some other platforms. |
| 823 | |
| 824 | - PSCI stress tests are very unreliable and will often hang. The root cause is |
| 825 | not known for sure but this might be due to bad synchronization between CPUs. |
| 826 | |
| 827 | - The GICv2 spurious IRQ test sometimes fails with the following error message: |
| 828 | |
| 829 | ``SMC @ lead CPU returned 0xFFFFFFFF 0x8 0xC`` |
| 830 | |
| 831 | The root cause is unknown. |
| 832 | |
| 833 | - The FWU tests take a long time to complete. This is because they wait for the |
| 834 | watchdog to reset the system. On FVP, TF-A configures the watchdog period to |
| 835 | about 4 min. This limit is excessive for an automated testing context and |
| 836 | leaves the user without feedback and unable to determine if the tests are |
| 837 | proceeding properly. |
| 838 | |
| 839 | - The test "Target timer to a power down cpu" sometimes fails with the |
| 840 | following error message: |
| 841 | |
| 842 | ``Expected timer switch: 4 Actual: 3`` |
| 843 | |
| 844 | The root cause is unknown. |
| 845 | |
| 846 | FWU images |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 847 | ~~~~~~~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 848 | |
| 849 | - The FWU tests do not work on the revC of the Base AEM FVP. They only work on |
| 850 | the revB. |
| 851 | |
| 852 | - NS-BL1U and NS-BL2U images reuse TFTF-specific code for legacy reasons. This |
| 853 | is not a clean design and may cause confusion. |
| 854 | |
| 855 | Test secure partitions (Cactus, Cactus-MM, Ivy) |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 856 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 857 | |
| 858 | - This is experimental code. It's likely to change a lot as the secure |
| 859 | partition software architecture evolves. |
| 860 | |
| 861 | - Supported on AArch64 FVP platform only. |
| 862 | |
| 863 | All test images |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 864 | ~~~~~~~~~~~~~~~ |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 865 | |
| 866 | - TF-A Tests are derived from a fork of TF-A so: |
| 867 | |
| 868 | - they've got some code in common but lag behind on some features. |
| 869 | |
| 870 | - there might still be some irrelevant references to TF-A. |
| 871 | |
| 872 | - Some design issues. |
| 873 | E.g. TF-A Tests inherited from the I/O layer of TF-A, which still needs a |
| 874 | major rework. |
| 875 | |
| 876 | - Cannot build TF-A Tests with Clang. Only GCC is supported. |
| 877 | |
| 878 | - The build system does not cope well with parallel building. The user should |
| 879 | not attempt to run multiple jobs in parallel with the ``-j`` option of `GNU |
| 880 | make`. |
| 881 | |
| 882 | - The build system does not properly track build options. A clean build must be |
| 883 | performed every time a build option changes. |
| 884 | |
| 885 | - UUIDs are not compliant to RFC 4122. |
| 886 | |
| 887 | - No floating point support. The code is compiled with GCC flag |
| 888 | ``-mgeneral-regs-only``, which prevents the compiler from generating code |
| 889 | that accesses floating point registers. This might limit some test scenarios. |
| 890 | |
| 891 | - The documentation is too lightweight. |
| 892 | |
| 893 | - Missing instruction barriers in some places before reading the system counter |
| 894 | value. As a result, the CPU could speculatively read it and any delay loop |
| 895 | calculations might be off (because based on stale values). We need to examine |
| 896 | all such direct reads of the ``CNTPCT_EL0`` register and replace them with a |
| 897 | call to ``syscounter_read()`` where appropriate. |
| 898 | |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 899 | Version 2.0 |
| 900 | ----------- |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 901 | |
| 902 | New features |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 903 | ^^^^^^^^^^^^ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 904 | |
| 905 | This is the first public release of the Trusted Firmware-A Tests source code. |
| 906 | |
| 907 | TFTF |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 908 | ~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 909 | |
| 910 | - Provides a baremetal test framework to exercise TF-A features through its |
| 911 | ``SMC`` interface. |
| 912 | |
| 913 | - Integrates easily with TF-A: the TFTF binary is packaged in the FIP image |
| 914 | as a ``BL33`` component. |
| 915 | |
| 916 | - Standalone binary that runs on the target without human intervention (except |
| 917 | for some specific tests that require a manual target reset). |
| 918 | |
| 919 | - Designed for multi-core testing. The various sub-frameworks allow maximum |
| 920 | parallelism in order to stress the firmware. |
| 921 | |
| 922 | - Displays test results on the UART output. This may then be parsed by an |
| 923 | external tool and integrated in a continuous integration system. |
| 924 | |
| 925 | - Supports running in AArch64 (NS-EL2 or NS-EL1) and AArch32 states. |
| 926 | |
| 927 | - Supports parsing a tests manifest (XML file) listing the tests to include in |
| 928 | the binary. |
| 929 | |
| 930 | - Detects most platform features at run time (e.g. topology, GIC version, ...). |
| 931 | |
| 932 | - Provides a topology enumeration framework. Allows tests to easily go through |
| 933 | affinity levels and power domain nodes. |
| 934 | |
| 935 | - Provides an event framework to synchronize CPU operations in a multi-core |
| 936 | context. |
| 937 | |
| 938 | - Provides a timer framework. Relies on a single global timer to generate |
| 939 | interrupts for all CPUs in the system. This allows tests to easily program |
| 940 | interrupts on demand to use as a wake-up event source to come out of CPU |
| 941 | suspend state for example. |
| 942 | |
| 943 | - Provides a power-state enumeration framework. Abstracts the valid power |
| 944 | states supported on the platform. |
| 945 | |
| 946 | - Provides helper functions for power management operations (CPU hotplug, |
| 947 | CPU suspend, system suspend, ...) with proper saving of the hardware state. |
| 948 | |
| 949 | - Supports rebooting the platform at the end of each test for greater |
| 950 | independence between tests. |
| 951 | |
| 952 | - Supports interrupting and resuming a test session. This relies on storing |
| 953 | test results in non-volatile memory (e.g. flash). |
| 954 | |
| 955 | FWU images |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 956 | ~~~~~~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 957 | |
| 958 | - Provides example code to exercise the Firmware Update feature of TF-A. |
| 959 | |
| 960 | - Tests the robustness of the FWU state machine implemented in the TF-A by |
| 961 | sending valid and invalid authentication, copy and image execution requests |
| 962 | to the TF-A BL1 image. |
| 963 | |
| 964 | EL3 test payload |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 965 | ~~~~~~~~~~~~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 966 | |
| 967 | - Tests the ability of TF-A to load an EL3 payload. |
| 968 | |
| 969 | Cactus test secure partition |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 970 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 971 | |
| 972 | - Tests that TF-A has correctly setup the secure partition environment: it |
| 973 | should be allowed to perform cache maintenance operations, access floating |
| 974 | point registers, etc. |
| 975 | |
| 976 | - Tests the ability of a secure partition to request changing data access |
| 977 | permissions and instruction permissions of memory regions it owns. |
| 978 | |
| 979 | - Tests the ability of a secure partition to handle StandaloneMM requests. |
| 980 | |
| 981 | Known issues and limitations |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 982 | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 983 | |
| 984 | The sections below lists the known issues and limitations of each test image |
| 985 | provided in this repository. |
| 986 | |
| 987 | TFTF |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 988 | ~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 989 | |
| 990 | The TFTF test image might be conceptually sub-divided further in 2 parts: the |
| 991 | tests themselves, and the test framework they are based upon. |
| 992 | |
| 993 | Test framework |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 994 | ~~~~~~~~~~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 995 | |
| 996 | - Some stability issues. |
| 997 | |
| 998 | - No mechanism to abort tests when they time out (e.g. this could be |
| 999 | implemented using a watchdog). |
| 1000 | |
| 1001 | - No convenient way to include or exclude tests on a per-platform basis. |
| 1002 | |
| 1003 | - Power domains and affinity levels are considered equivalent but they may |
| 1004 | not necessarily be. |
| 1005 | |
| 1006 | - Need to provide better support to alleviate duplication of test code. There |
| 1007 | are some recurrent test patterns for which helper functions should be |
| 1008 | provided. For example, bringing up all CPUs on the platform and executing the |
| 1009 | same function on all of them, or programming an interrupt and waiting for it |
| 1010 | to trigger. |
| 1011 | |
| 1012 | - Every CPU that participates in a test must return from the test function. If |
| 1013 | it does not - e.g. because it powered itself off for testing purposes - then |
| 1014 | the test framework will wait forever for this CPU. This limitation is too |
| 1015 | restrictive for some tests. |
| 1016 | |
| 1017 | - No protection against interrupted flash operations. If the target is reset |
| 1018 | while some data is written to flash, the test framework might behave |
| 1019 | incorrectly on reset. |
| 1020 | |
| 1021 | - When compiling the code, if the generation of the tests_list.c and/or |
| 1022 | tests_list.h files fails, the build process is not aborted immediately and |
| 1023 | will only fail later on. |
| 1024 | |
| 1025 | - The directory layout is confusing. Most of the test framework code has been |
| 1026 | moved under the ``tftf/`` directory to better isolate it but this effort is |
| 1027 | not complete. As a result, there are still some TFTF files scattered around. |
| 1028 | |
| 1029 | Tests |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 1030 | ~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 1031 | |
| 1032 | - Some tests are implemented for AArch64 only and are skipped on AArch32. |
| 1033 | |
| 1034 | - Some tests are not robust enough: |
| 1035 | |
| 1036 | - Some tests might hang in some circumstances. For example, they might wait |
| 1037 | forever for a condition to become true. |
| 1038 | |
| 1039 | - Some tests rely on arbitrary time delays instead of proper synchronization |
| 1040 | when executing order-sensitive steps. |
| 1041 | |
| 1042 | - Some tests have been implemented in a practical manner: they seem to work |
| 1043 | on actual hardware but they make assumptions that are not guaranteed by |
| 1044 | the Arm architecture. Therefore, they might fail on some other platforms. |
| 1045 | |
| 1046 | - PSCI stress tests are very unreliable and will often hang. The root cause is |
| 1047 | not known for sure but this might be due to bad synchronization between CPUs. |
| 1048 | |
| 1049 | - The GICv2 spurious IRQ test is Juno-specific. In reality, it should only be |
| 1050 | GICv2-specific. It should be reworked to remove any platform-specific |
| 1051 | assumption. |
| 1052 | |
| 1053 | - The GICv2 spurious IRQ test sometimes fails with the following error message: |
| 1054 | |
| 1055 | ``SMC @ lead CPU returned 0xFFFFFFFF 0x8 0xC`` |
| 1056 | |
| 1057 | The root cause is unknown. |
| 1058 | |
| 1059 | - The manual tests in AArch32 mode do not work properly. They save some state |
| 1060 | information into non-volatile memory in order to detect the reset reason but |
| 1061 | this state does not appear to be retained. As a result, these tests keep |
| 1062 | resetting infinitely. |
| 1063 | |
| 1064 | - The FWU tests take a long time to complete. This is because they wait for the |
| 1065 | watchdog to reset the system. On FVP, TF-A configures the watchdog period to |
| 1066 | about 4 min. This is way too long in an automated testing context. Besides, |
| 1067 | the user gets not feedback, which may let them think that the tests are not |
| 1068 | working properly. |
| 1069 | |
| 1070 | - The test "Target timer to a power down cpu" sometimes fails with the |
| 1071 | following error message: |
| 1072 | |
| 1073 | ``Expected timer switch: 4 Actual: 3`` |
| 1074 | |
| 1075 | The root cause is unknown. |
| 1076 | |
| 1077 | FWU images |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 1078 | ~~~~~~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 1079 | |
Sandrine Bailleux | 7af6c6d | 2018-10-22 17:09:07 +0200 | [diff] [blame] | 1080 | - The FWU tests do not work on the revC of the Base AEM FVP. They only work on |
| 1081 | the revB. |
| 1082 | |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 1083 | - NS-BL1U and NS-BL2U images reuse TFTF-specific code for legacy reasons. This |
| 1084 | is not a clean design and may cause confusion. |
| 1085 | |
| 1086 | Cactus test secure partition |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 1087 | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 1088 | |
| 1089 | - Cactus is experimental code. It's likely to change a lot as the secure |
| 1090 | partition software architecture evolves. |
| 1091 | |
| 1092 | - Fails to link with GNU toolchain 7.3.1. |
| 1093 | |
| 1094 | - Cactus is supported on AArch64 FVP platform only. |
| 1095 | |
| 1096 | All test images |
Paul Beesley | ec7988c | 2019-10-24 11:57:00 +0000 | [diff] [blame] | 1097 | ~~~~~~~~~~~~~~~ |
Sandrine Bailleux | 3cd87d7 | 2018-10-09 11:12:55 +0200 | [diff] [blame] | 1098 | |
| 1099 | - TF-A Tests are derived from a fork of TF-A so: |
| 1100 | |
| 1101 | - they've got some code in common but lag behind on some features. |
| 1102 | |
| 1103 | - there might still be some irrelevant references to TF-A. |
| 1104 | |
| 1105 | - Some design issues. |
| 1106 | E.g. TF-A Tests inherited from the I/O layer of TF-A, which still needs a |
| 1107 | major rework. |
| 1108 | |
| 1109 | - Cannot build TF-A Tests with Clang. Only GCC is supported. |
| 1110 | |
| 1111 | - The build system does not cope well with parallel building. The user should |
| 1112 | not attempt to run multiple jobs in parallel with the ``-j`` option of `GNU |
| 1113 | make`. |
| 1114 | |
| 1115 | - The build system does not properly track build options. A clean build must be |
| 1116 | performed every time a build option changes. |
| 1117 | |
| 1118 | - SMCCC v2 is not properly supported. |
| 1119 | |
| 1120 | - UUIDs are not compliant to RFC 4122. |
| 1121 | |
| 1122 | - No floating point support. The code is compiled with GCC flag |
| 1123 | ``-mgeneral-regs-only``, which prevents the compiler from generating code |
| 1124 | that accesses floating point registers. This might limit some test scenarios. |
| 1125 | |
| 1126 | - The documentation is too lightweight. |
| 1127 | |
| 1128 | -------------- |
| 1129 | |
Madhukar Pappireddy | 6ec644e | 2020-04-08 15:44:38 -0500 | [diff] [blame] | 1130 | *Copyright (c) 2018-2020, Arm Limited. All rights reserved.* |
Sandrine Bailleux | 34da066 | 2019-03-28 09:34:55 +0100 | [diff] [blame] | 1131 | |
| 1132 | .. _Arm Neoverse Reference Design N1 Edge (RD-N1-Edge): https://developer.arm.com/products/system-design/reference-design/neoverse-reference-design |
| 1133 | .. _Arm SGI-575: https://developer.arm.com/products/system-design/fixed-virtual-platforms |