Olivier Deprez | 55b74f8 | 2021-11-10 11:32:31 +0100 | [diff] [blame] | 1 | # Change log |
| 2 | |
Olivier Deprez | 79e9522 | 2022-05-09 17:21:08 +0200 | [diff] [blame] | 3 | ## v2.7 |
| 4 | #### Highlights |
| 5 | |
| 6 | * Boot protocol (FF-A v1.1 EAC0) |
| 7 | * The SPMC primarily supports passing the SP manifest address at boot time. |
| 8 | * In a secure partition package, partition manifest and image offsets are |
| 9 | configurable. |
| 10 | * Allows for larger partition manifest sizes. |
| 11 | * Setup and discovery (FF-A v1.1 EAC0) |
| 12 | * FFA_VERSION is forwarded from SPMD to SPMC. SPMC records the version of |
| 13 | a normal world endpoint. |
| 14 | * Added UUID to partition info descriptors. |
| 15 | * Introduced count flag to FFA_PARTITION_INFO_GET. |
| 16 | * Interrupt handling (FF-A v1.1 Beta0) |
| 17 | * Physical GIC registers trapped when accessed from secure partitions. |
| 18 | * Priority mask register saved/restored on world switches. |
| 19 | * Interrupts masked before resuming a pre-empted vCPU. |
| 20 | * Implemented implicit secure interrupt completion signal. |
| 21 | * Allow unused GICR frame for non-existent PEs. |
| 22 | * Notifications (FF-A v1.1 EAC0) |
| 23 | * Implemented notification pending interrupt and additional test coverage. |
| 24 | * MTE stack tagging |
| 25 | * Implemented FEAT_MTE2 stack tagging support at S-EL2. |
| 26 | * Core stacks marked as normal tagged memory. A synchronous abort triggers |
| 27 | on a load/store tag check failure. |
| 28 | * This permits detection of wrong operations affecting buffers allocated |
| 29 | from the stack. |
| 30 | * FF-A v1.0 compliance |
| 31 | * Check composite memory region offset is defined in FF-A memory sharing. |
| 32 | * Check sender and receiver memory attributes in a FF-A memory sharing |
| 33 | operation match the attributes expected in the Hafnium implementation. |
| 34 | * Fix clear memory bit use in FF-A memory sharing from NWd to SWd. |
| 35 | * Prevent FF-A memory sharing from a SP to a NS endpoint. |
| 36 | * Reject a FF-A memory retrieve operation with the 'Address Range Alignment |
| 37 | Hint' bit set (not supported by the implementation). |
| 38 | * Refine usage of FF-A memory sharing 'clear memory flag'. |
| 39 | * Misc |
| 40 | * Improved extended memory address ranges support: |
| 41 | * 52 bits PA (FEAT_LPA/FEAT_LPA2) architecture extension detected |
| 42 | results in limiting the EL2 Stage-1 physical address range to 48 bits. |
| 43 | * In the FF-A memory sharing operations, harden address width checks on |
| 44 | buffer mapping. |
| 45 | * Improved MP SP and S-EL0 partitions support |
| 46 | * The physical core index is passed to a SP vCPU0 on booting. |
| 47 | * Added MP SP and S-EL0 partitions boot test coverage. |
| 48 | * Emulate SMCCC VERSION to the primary VM. |
| 49 | * Memory config registers (non-secure and secure virtualization control and |
| 50 | translation table base) moved to the vCPU context. |
| 51 | * EL2 stage 1 mapping extended to 1TB to support systems with physical |
| 52 | address space larger than 512GB. |
| 53 | * FFA_RUN ABI hardened to check the vCPU index matches the PE index onto |
| 54 | which a vCPU is requested to run. |
| 55 | * Fixed missing ISB after CPTR_EL2 update upon PE initialization. |
| 56 | * Fixed stage 2 default shareability to inner shareable (from non-shareable) |
| 57 | to better support vCPU migration. |
| 58 | * Fixed manifest structure allocation from BSS rather than stack |
| 59 | at initialization. |
| 60 | * Fixed an issue with FF-A memory reclaim executed after memory donate |
| 61 | resulting in a returned error code. |
| 62 | * Build and test environment |
| 63 | * Add the ability to use an out-of-tree toolchain. |
| 64 | * Primary intent is to permit building Hafnium on Aarch64 hosts. |
| 65 | * CI runs using the toolchain versioned in prebuilts submodule. |
| 66 | A developer can still use this version as well. |
| 67 | * Introduce an assert macro enabled by a build option on the command line. |
| 68 | Assertions are checked by default. Production builds can optionally |
| 69 | disable assertions. |
| 70 | * Added manifest options to permit loading VMs using an FF-A manifest. |
| 71 | * CI |
| 72 | * Added job running the Hypervisor + SPMC configuration on patch |
| 73 | submissions. |
| 74 | * FVP |
| 75 | * Enable secure memory option. |
| 76 | * Remove restriction on speculative execution options. |
| 77 | * Updated to use model version 11.17 build 21. |
| 78 | * Updated linux submodule to v5.10. |
| 79 | * VHE EL0 partitions tests automated through jenkins. |
| 80 | |
| 81 | #### Known limitations: |
| 82 | * FF-A v1.1 EAC0 implementation is partial mainly on interrupt handling and |
| 83 | memory sharing. |
| 84 | * Hafnium limits physical interrupt IDs to 64. The legacy virtual interrupt |
| 85 | controller driver limits to 64. The recent addition of physical interrupt |
| 86 | handling in the SPMC through the GIC assumes a 1:1 mapping of a physical |
| 87 | interrupt ID to a virtual interrupt ID. |
| 88 | * Secure timer virtualization is not supported. |
| 89 | * The security state of memory or device region cannot be specified in a SP |
| 90 | manifest. |
| 91 | |
Olivier Deprez | 55b74f8 | 2021-11-10 11:32:31 +0100 | [diff] [blame] | 92 | ## v2.6 |
| 93 | #### Highlights |
| 94 | * FF-A Setup and discovery |
| 95 | * FF-A build time version updated to v1.1. |
| 96 | * Managed exit and notifications feature support enabled in SP manifests. |
| 97 | * Updated FFA_FEATURES to permit discovery of managed exit, schedule receiver, |
| 98 | and notification pending interrupt IDs. |
| 99 | * FFA_PARTITION_INFO_GET updated to permit managed exit and notification |
| 100 | support discovery. |
| 101 | * FFA_SPM_ID_GET added to permit discovering the SPMC endpoint ID (or the |
| 102 | SPMD ID at the secure physical FF-A instance). |
| 103 | * FFA_RXTX_UNMAP implementation added. |
| 104 | * FF-A v1.1 notifications |
| 105 | * Added ABIs permitting VM (or OS kernel) to SP, and SP to SP asynchronous |
| 106 | signaling. |
| 107 | * Added generation of scheduler receiver (NS physical) and notification |
| 108 | pending (secure virtual) interrupts. |
| 109 | * The schedule receiver interrupt is donated from the secure world SGI |
| 110 | interrupt ID range. |
| 111 | * FF-A v1.1 interrupt handling |
| 112 | * Added a GIC driver at S-EL2 permitting to trap and handle non-secure and |
| 113 | secure interrupts while the secure world runs. |
| 114 | * Added forwarding and handling of a secure interrupt while the normal world |
| 115 | runs. |
| 116 | * Added secure interrupt forwarding to the secure partition that had the |
| 117 | interrupt registered in its partition manifest. |
| 118 | * The interrupt deactivation happens through the Hafnium para-virtualized |
| 119 | interrupt controller interface. |
| 120 | * vCPU states, run time models and SP scheduling model are revisited as per |
| 121 | FF-A v1.1 Beta0 specification (see 'Known limitations' section below). |
| 122 | * S-EL0 partitions support |
| 123 | * Added support for VHE architecture extension in the secure world (through |
| 124 | a build option). |
| 125 | * A partition bootstraps as an S-EL0 partition based on the exception-level |
| 126 | field in the FF-A manifest. |
| 127 | * It permits the implementation of applications on top of Hafnium without |
| 128 | relying on an operating system at S-EL1. |
| 129 | * It leverages the EL2&0 Stage-1 translation regime. Apps use FF-A |
| 130 | ABIs through the SVC conduit. |
| 131 | * Added FF-A v1.1 FFA_MEM_PERM_GET/SET ABIs permitting run-time update of |
| 132 | memory region permissions. |
| 133 | * It supersedes the existing S-EL1 shim architecture (without removing its |
| 134 | support). |
| 135 | * S-EL1 SP, S-EL0 SP or former S-EL0 SP+shim can all co-exist in the same |
| 136 | system. |
| 137 | * SVE |
| 138 | * Support for saving/restoring the SVE live state such that S-EL2/Hafnium |
| 139 | preserves the normal world state on world switches. |
| 140 | * Secure partitions are permitted to use FP/SIMD while normal world uses |
| 141 | SVE/SIMD/FP on the same core. |
| 142 | * The SVE NS live state comprises FPCR/FPSR/FFR/p[16]/Z[32] registers. |
| 143 | * LLVM/Clang 12 |
| 144 | * The toolchain stored in prebuilts submodule is updated to LLVM 12.0.5. |
| 145 | * Build/static analyzer fixes done in the top and third party projects. |
| 146 | * Linux sources (used by the test infrastructure) are updated to 5.4.148. |
| 147 | The linux test kernel module build is updated to only depend on LLVM |
| 148 | toolchain. |
| 149 | * Hafnium CI improvements |
| 150 | * Added two configurations permitting Hafnium testing in the secure world. |
| 151 | * First configuration launches both the Hypervisor in the normal world |
| 152 | and the SPMC in the secure world. This permits thorough FF-A ABI testing |
| 153 | among normal and secure world endpoints. |
| 154 | * The second configuration launches the SPMC alone for component testing |
| 155 | or SP to SP ABI testing. |
| 156 | * Hafnium CI Qemu version updated to v6.0.0 (implements VHE and FEAT_SEL2 |
| 157 | extensions). |
| 158 | * FF-A compliance fixes |
| 159 | * Added checks for valid memory permissions values in manifest memory and |
| 160 | device regions declarations. |
| 161 | * FFA_FEATURES fixed to state indirect messages are not supported by |
| 162 | the SPMC. |
| 163 | * Limit an SP to emit a direct request to another SP only. |
| 164 | * Memory sharing: fixed input validation and return values. |
| 165 | * FFA_RXTX_MAP fixed returned error codes. |
| 166 | * FFA_MSG_WAIT input parameters check hardened. |
| 167 | |
| 168 | #### Known limitations: |
| 169 | * S-EL0 partitions/VHE: the feature is in an experimental stage and not all use |
| 170 | cases have been implemented or tested. Normal world to SP and SP to SP memory |
| 171 | sharing is not tested. Interrupt handling is not tested. |
| 172 | * The current implementation does not support handling a secure interrupt that |
| 173 | is triggered while currently handling a secure interrupt. This restricts to |
| 174 | scenarios described in Table 8.13 and Table 8.14 of the FF-A v1.1 Beta0 |
| 175 | specification. Priority Mask Register is not saved/restored during context |
| 176 | switching while handling secure interrupt. |
| 177 | * Hafnium CI: scenarios involving the Hypervisor are left as test harness |
| 178 | purposes only, not meant for production use cases. |
Daniel Boulby | ccb7068 | 2021-04-28 19:40:39 +0100 | [diff] [blame] | 179 | |
| 180 | ## v2.5 |
| 181 | #### Highlights |
| 182 | * BTI/Pointer authentication support |
| 183 | * Add branch protection build option for FEAT_PAuth and FEAT_BTI to the |
| 184 | clang command line. This only affects the S-EL2 image. |
| 185 | * Enable pointer authentication by supplying a platform defined pseudo |
| 186 | random key. |
| 187 | * Enable BTI by setting the guarded page bit in MMU descriptors for |
| 188 | executable pages. |
| 189 | * SMMUv3.2 S-EL2 support |
| 190 | * Add support for SMMUv3 driver to perform stage 2 translation, protection |
| 191 | and isolation of upstream peripheral device's DMA transactions. |
| 192 | * FF-A v1.0 Non-secure interrupt handling |
| 193 | * Trap physical interrupts to S-EL2 when running a SP. |
| 194 | * Handle non secure interrupts that occur while an SP is executing, |
| 195 | performing managed exit if supported. |
| 196 | * Add basic support for the GICv3 interrupt controller for the AArch64 |
| 197 | platform. |
| 198 | * FF-A power management support at boot time |
| 199 | * Provide platform-independent power management implementations for the |
| 200 | Hypervisor and SPMC. |
| 201 | * Implement the FFA_SECONDARY_EP_REGISTER interface for an MP SP or SPMC |
| 202 | to register the secondary core cold boot entry point for each of their |
| 203 | execution contexts. |
| 204 | * Introduce a generic "SPMD handler" to process the power management events |
| 205 | that may be conveyed from SPMD to SPMC, such as core off. |
| 206 | * FF-A Direct message interfaces |
| 207 | * Introduce SP to SP direct messaging. |
| 208 | * Fix bug in the MP SP to UP SP direct response handling. |
| 209 | * FF-A Memory sharing interfaces |
| 210 | * Introduce SP to SP memory sharing. |
| 211 | * When a sender of a memory management operation reclaims memory, set the |
| 212 | memory regions permissions back to it's original configuration. |
| 213 | * Require default permissions to be supplied to the function |
| 214 | 'ffa_memory_permissions_to_mode', so in the case where no permissions are |
| 215 | specified for a memory operation, the data and instruction permissions can |
| 216 | be set to the default. |
| 217 | * Encode Bit[63] of the memory region handle according to if the handle is |
| 218 | allocated by the Hypervisor or SPMC. |
| 219 | * FF-A v1.0 spec compliance |
| 220 | * Return INVALID_PARAMETER error code instead of NOT_SUPPORTED for direct |
| 221 | messaging interfaces when an invalid sender or receiver id is given. |
| 222 | * Check that reserved parameter registers are 0 when invoking direct |
| 223 | messaging ABI interfaces. |
| 224 | * For SMC32 compliant direct message interfaces, only copy 32-bits |
| 225 | parameter values. |
| 226 | * Change the FF-A error codes to 32-bit to match the FF-A specification. |
| 227 | * Fix consistency with maintaining the calling convention bit of the |
| 228 | func id between the ffa_handler and the FFA_FEATURES function. |
| 229 | * Remove primary VM dependencies in the SPMC |
| 230 | * Treat normal world as primary VM when running in the secure world. |
| 231 | * Create an SPMC boot flow. |
| 232 | * Hafnium CI |
| 233 | * Enable Hafnium CI to include tests for Hafnium SPMC. |
| 234 | * Add basic exception handler to service VM's. |
| 235 | * SIMD support |
| 236 | * Add saving/restoring of other world FP/NEON/SIMD state when entering and |
| 237 | exiting the SPMC. |
| 238 | * SPMC early boot cache fix |
| 239 | * Import data cache clean and invalidation helpers from TF-A project and |
| 240 | provide an arch module for cache operations. |
| 241 | * Invalidate the SPMC image in the data cache at boot time to prevent |
| 242 | potential access to stale cache entries left by earlier boots stages. |
| 243 | * Misc and bug fixes |
| 244 | * Complete vCPU state save prior to normal world exit. |
| 245 | * Update S-EL2 Stage-1 page table shareability from outer to inner. |
| 246 | * Add PL011 UART initialization code to set the IDRD and FBRD registers |
| 247 | according to the UART clock and baud rate specified at build time. |
| 248 | * License script checker fixes. |
| 249 | |
| 250 | #### Known limitations: |
| 251 | * Secure interrupts not supported. |
| 252 | * FF-A indirect message interface not supported in the secure world. |
| 253 | * Only supporting models of MultiProcessor SP (vCPUs pinned to physical |
| 254 | CPUs) or UniProcessor SP (single vCPU). |
| 255 | * The first secure partition booted must be a MP SP. |
| 256 | * FFA_RXTX_UNMAP not implemented. |
| 257 | * Use of an alternate caller provided buffer from RX/TX buffers for memory |
| 258 | sharing operations is not implemented. |
| 259 | * A memory retrieve request to SPMC does not support the caller endpoint to |
| 260 | provide the range of IPA addresses to map the region to. |
Olivier Deprez | 62405ad | 2020-10-26 18:51:49 +0100 | [diff] [blame] | 261 | |
| 262 | ## v2.4 |
| 263 | |
| 264 | This is the first drop to implement the TrustZone secure side S-EL2 firmware |
Olivier Deprez | 410a3ac | 2020-11-04 13:54:15 +0100 | [diff] [blame] | 265 | (SPM Core component) complying with FF-A v1.0. |
Olivier Deprez | 62405ad | 2020-10-26 18:51:49 +0100 | [diff] [blame] | 266 | It is a companion to the broader TF-A v2.4 release. |
| 267 | The normal world Hypervisor is maintained functional along with the |
| 268 | Hafnium CI test suite. |
| 269 | |
Daniel Boulby | ccb7068 | 2021-04-28 19:40:39 +0100 | [diff] [blame] | 270 | #### Highlights |
Olivier Deprez | 62405ad | 2020-10-26 18:51:49 +0100 | [diff] [blame] | 271 | * FF-A v1.0 Setup and discovery interface |
| 272 | * Hypervisor implementation re-used and extended to the SPMC and SPs. |
| 273 | * Added partition info get ABI and appropriate properties response depending |
| 274 | on partition capabilities (PVM, Secondary VM or Secure Partitions). |
| 275 | * FF-A device-tree manifest parsing. |
| 276 | * FF-A partitions can declare memory/device regions, and RX/TX buffers that |
| 277 | the SPMC sets up in the SP EL1&0 Stage-2 translation regime at boot time. |
| 278 | * FF-A IDs normal and secure world split ranges. |
| 279 | * The SPMC maps the Hypervisor (or OS kernel) RX/TX buffers as non-secure |
| 280 | buffers in its EL2 Stage-1 translation regime on FFA_RXTX_MAP ABI |
| 281 | invocation from the non-secure physical FF-A instance. |
| 282 | * FF-A v1.0 Direct message interface |
| 283 | * Added implementation for the normal world Hypervisor and test cases. |
| 284 | * Implementation extended to the SPMC and SPs. |
| 285 | * Direct message requests emitted from the PVM to a Secondary VM or a |
| 286 | Secure Partition (or OS Kernel to a Secure Partition). Direct message |
| 287 | responses emitted from Secondary VMs and Secure Partitions to the PVM. |
| 288 | * The secure world represents the "other world" (normal world Hypervisor |
| 289 | or OS kernel) vCPUs in an abstract "Hypervisor VM". |
| 290 | * FF-A v1.0 memory sharing |
| 291 | * Hypervisor implementation re-used and extended to the SPMC and SPs. |
| 292 | * A NS buffer can be shared/lent/donated by a VM to a SP (or OS Kernel |
| 293 | to a SP). |
| 294 | * The secure world configures Stage-1 NS IPA output to access the NS PA |
| 295 | space. |
| 296 | * The secure world represents the "other world" (normal world Hypervisor |
| 297 | or OS kernel) memory pages in an abstract "Hypervisor VM" and tracks |
| 298 | memory sharing permissions from incoming normal world requests. |
| 299 | * Secure world enablement |
Olivier Deprez | 410a3ac | 2020-11-04 13:54:15 +0100 | [diff] [blame] | 300 | * Secure Partitions booted in sequence on their primary execution context, |
| 301 | according to the boot order field in their partition manifest. |
| 302 | This happens during the secure boot process before the normal world |
| 303 | actually runs. |
Olivier Deprez | 62405ad | 2020-10-26 18:51:49 +0100 | [diff] [blame] | 304 | * The SPMC implements the logic to receive FF-A messages through the EL3 |
| 305 | SPMD, process them, and either return to the SPMD (and normal world) or |
| 306 | resume a Secure Partition. |
| 307 | * Extract NS bit from HPFAR_EL2 on Stage-2 page fault. |
| 308 | * Prevent setup of LOR regions in SWd. |
| 309 | * Avoid direct PSCI calls down to EL3. |
| 310 | * Platforms |
| 311 | * Added Arm FVP secure Hafnium build support. |
| 312 | * Added Arm TC0 "Total Compute" secure Hafnium build support. |
| 313 | * Other improvements |
| 314 | * Re-hosting to trustedfirmware.org |
| 315 | * busy_secondary timer increased to improve CI stability. |
| 316 | * Removed legacy Hypervisor calls. |
| 317 | * Fix CPTR_EL2 TTA bit position. |
| 318 | * Report FAR_EL2 on injecting EL1 exception. |
Daniel Boulby | ccb7068 | 2021-04-28 19:40:39 +0100 | [diff] [blame] | 319 | #### Known limitations: |
| 320 | * Not all fields of the FF-A manifest are actually processed by the Hafnium |
| 321 | device-tree parser. |
| 322 | * SP to SP communication not supported. |
| 323 | * SP to SP memory sharing not supported. |
| 324 | * S-EL1 and SIMD contexts shall be saved/restored by EL3. |
| 325 | * Multi-endpoint memory sharing not supported. |
| 326 | * Interrupt management limited to trapping physical interrupts to |
| 327 | the first S-EL1 SP. Physical interrupt trapping at S-EL2 planned as |
| 328 | next release improvement. |
| 329 | * Validation mostly performed using first SP Execution Context (vCPU0). More |
| 330 | comprehensive multicore enablement planned as next release improvement. |