Platforms

• Added support for the ARM Cortex A55
• Added support for the imx8mp-evk platform
• Added support for additional RPI4 variants
• Added support for the Odroid C4
• Added support for the Avnet MaaXBoard
• Added support for arm_hyp on qemu-arm-virt platfrom with cortex-a15 CPU
• Added support for qemu-riscv-virt
• Added support for the Pine64 Star64
• Added support for the TQMa8XQP 1GiB module
• Remove imx31/kzm platform support. This platform is being removed as it is sufficiently old and unused.
• Remove ARM1136JF_S and ARMv6 support. This architecture version is being removed as it is sufficiently old and unused. See RFC-8.
• Remove ARMv6 specific configs: KernelGlobalsFrame and KernelDangerousCodeInjectionOnUndefInstr. This removes the constant seL4_GlobalsFrame from libsel4 as well as the IPC buffer in GlobalsFrame caveat from CAVEATS.md
• rpi3+rpi4: Mark first memory page as reserved

Arm

• Enabled access to seL4_VCPUReg_VMPIDR and seL4_VCPUReg_VMPIDR_EL2 for all hypervisor configurations. Previously this register was only accessible for SMP kernel configurations. Non-SMP configurations can still require access when wanting to control the value of MPIDR that the guest reads. Note that the initial value for new seL4_ARM_VCPUs for this register is 0 which isn't a legal value for MPIDR_EL1 on AArch64. It may be necessary for the register to be explicitly initialized by user level before launching a thread associated with the new seL4_ARM_VCPU.

• Allow changing the VCPU of active thread: call vcpu_switch in associateVCPUTCB. This guarantees that the correct VCPU will be activated when the kernel finishes execution. Previously, changing the VCPU of the current thread would result in no VCPU being active.

• benchmarking: use write-through kernel log buffer

• arm_hyp: Access SPSR via non-banked instructions

• generic_timer: force timer to de-assert IRQ.

• No special handling for edge-triggered IRQs.

  • Clearing the pending state only has an effect if the IRQ state is active-and-pending, which happens for edge-triggered interrupts if another edge happens on the IRQ line for the currently active interrupt. This window is small enough to ignore, at worst user space will get another notification, which is harmless.

  If unnecessary notifications are unwanted, the pending state should be cleared during seL4_IRQHandler_Ack(), as that covers a much bigger window. However, edge-triggered interrupts are not expected to happen often. Making all interrupt handling slightly faster and the code simpler is the better trade-off.

• Make write-only mapping message consistent. There is a warning when creating a write-only mapping on AArch32/AArch64. This message is now the same in all variants.

AArch32

• Implement KernelArmExportPTMRUser and KernelArmExportVTMRUser options for Arm generic timer use access on AArch32.
• AArch32 VM fault messages now deliver original (untranslated) faulting IP in a hypervisor context, matching AArch64 behaviour.
• Fix single stepping on ARMv7
• TLB: only perform TLB lockdown for Cortex A8.
  • The code previously used the same instructions for Cortex A8 and A9, but the Cortex A8 instructions are undocumented for A9, and A9 provides a slightly different TLB interface. As far as we can tell, the instructions were simply ignored by the supported A8 platforms, so there was no current correctness issue. Since the instructions had no effect, we removed A9 TLB lockdown support. This potential issue was discovered and reported by the UK's National Cyber Security Centre (NCSC).
• TLB: guard TLB lockdown count.
  • lockTLBEntry uses the global tlbLockCount as input without checking bounds. This is fine, because the function is called at most 2 times per core, but this is only apparent when checking the entire possible calling context. Make this bound obvious locally by doing nothing if the function is called with values of tlbLockCount of 2 or greater. This is safe, because TLB lockdown is a performance change only. Also add an assert for debug mode, because we want to know if calling context ever changes. This potential issue was reported by The UK's National Cyber Security Centre (NCSC).

AArch64

• Add AARCH64_verified.cmake config for functional correctness on AArch64
• Rename libsel4 config option AARCH64_VSPACE_S2_START_L1 to CONFIG_AARCH64_VSPACE_S2_START_L1 to be namespace compliant.
• Added SMC Capability (smc_cap) and SMC forwarding for AArch64 platforms. See RFC-9.
• Remove VSpace object types in AArch64: seL4_ARM_PageDirectory and seL4_ARM_PageUpperDirectory. See also the corresponding RFC. The functionality previously provided by these types will be provided by the existing seL4_ARM_PageTable object type. This allows for a simpler API and enables a smaller kernel implementation that will be easier to verify. libsel4 provides a source compatibility translation that maps the old libsel4 names and constants the new ones in <sel4/sel4_arch/deprecated.h>. There are some exceptional cases where kernel behavior has changed:
  • A Page directory and page table are now the same kind of object and can be mapped at any page table level. If the lookup for the provided address stops at a slot that can map a page directory, it will map the object as a page directory. If there already is a page directory mapped, the lookup will proceed to the next level and it will map as a page table instead of returning an error.
• Removed user address space reserved slots restriction on 40bit PA platforms when KernelArmHypervisorSupport is set. This change is reflected in the definition of the seL4_UserTop constant that holds the largest user virtual address.
• Added support for GICv3 virtualization, tested on iMX8QXP
• Implemented a signal fastpath on AArch64. Must be enabled explicitly with the KernelSignalFastpath config option.
• Implemented a virtual memory fault fastpath on AArch64. Must be enabled explicitly with the KernelExceptionFastpath config option.
• Add option KernelAArch64UserCacheEnable for user cache maintenance.

  • Enables user level access to DC CVAU, DC CIVAC, DC CVAC, and IC IVAU which are cache maintenance operations for the data caches and instruction caches underlying Normal memory and also access to the read-only cache-type register CTR_EL0 that provides cache type information. The ArmV8-A architecture allows access from EL0 as fast cache maintenance operations improves DMA performance in user-level device drivers.

    These instructions are a subset of the available cache maintenance instructions as they can only address lines by virtual address (VA). They also require that the VA provided refers to a valid mapping with at least read permissions. This corresponds to lines that the EL0 could already affect via regular operation and so it's not expected to break any cache-partitioning scheme.

    The config option allows this policy to be selected for a particular kernel configuration, but it is default enabled as this has been the existing behavior for current aarch64,hyp configurations and have not been explicitly disabled in non-hyp configurations.

• make error reporting consistent: Report the VSpace cap (1) as invalid, instead of the frame cap (0) in ARMFrameInvocation to stay consistent with the other architectures.
• vcpu: only trap WFx instructions from VCPUs. When KernelArmDisableWFIWFETraps is disabled (trapping of WFI/WFE is enabled), the kernel traps WFx instructions from both native and vCPU threads. This change brings the code in line with the config description.

RISC-V

• Remove the ability for user-space on RISC-V platforms to access the core-local interrupt controller (CLINT). The CLINT contains memory-mapped registers that the kernel depends on for timer interrupts and hence should not be accessible by user-space.
• Add configuration option KernelRiscvUseClintMtime for faster access of hardware timestamp on RISC-V platforms. The configuration option is not enabled by default as it requires access to the CLINT which depends on the platform and whether the M-mode firmware allows S-mode to access the CLINT. For example, newer versions of OpenSBI (1.0 and above) do not allow direct access of the CLINT.
• Rename object interface files include/interfaces/sel4.xml, arch_include/*/interfaces/sel4arch.xml, and sel4_arch_include/*/interfaces/sel4arch.xml to include/interfaces/object-api.xml, arch_include/*/interfaces/object-api-arch.xml, and sel4_arch_include/*/interfaces/object-api-sel4-arch.xml, respectively.
• Improve PLIC driver API and documentation
• Fix getMaxUsToTicks function tor return time instead of ticks.
• Improve RISC-V PTE compliance: keep D, A, and U bits cleared.

Intel

• libsel4: add enum for EPT attributes.
• VTX: fix EPT cache attribute setting. Previously the only effectively possible value was EPTWriteBack.
• Add kernel support for 64-bit VMs
• Provide CONFIG_X86_64_VTX_64BIT_GUESTS for 64-bit VM support
• Only access real IOAPIC registers. IOAPICS can have varying numbers of lines attached. The actual number can be accessed in the top 16 bits of the version register. Rather than assuming fixed 24 lines per IRQ, read the actual number and use that.

MCS

• Rename seL4_TimeoutMsg to seL4_Timeout_Msg to make it consistent with the naming of other messages.
• Correct the minimum size of a scheduling context. This changes the value of seL4_MinSchedContextBits.
• Correct check for message length in SchedControl_ConfigureFlags
• Allow lazy SchedContext rebind. Before, binding a scheduling context to a TCB was not allowed if the SC was bound to a notification object. Also, binding an SC to a notification was not allowed if that scheduling context was already bound to a TCB. Without these restriction it is much easier to move scheduling contexts around: In effect having a SC bound on both the TCB and a notification acts as if the thread is running on a donated SC which will be returned when the tasks calls Recv/Wait, which is done by maybeReturnSchedContext().
• Only charge budgets for non-idle thread SCs
• ARM+MCS: Introduce TIMER_OVERHEAD_TICKS. For ARM currently TIMER_PRECISION exists, but that is in microseconds and not fine-grained enough. TIMER_OVERHEAD_TICKS is needed to make periodic tasks synchronous with the system clock. If this value is non-zero every period will be extended with the overhead of taking an interrupt and reading the system clock. To avoid this drift, the configured value should be set to at least the average overhead.
• SMP: Do not use cross-node ksCurTime assuming they are in sync (which they are not), instead use NODE_STATE(ksCurTime).
• SMP: Add clock synchronisation test on boot
• SMP: Fix scheduling context use-after-free

Other Changes

• boot: Introduce seL4_BootInfoFrameSize and seL4_BootInfoFrameBits for user land so there is no longer a need to hard-code a 4 KiByte assumption. Remove BI_FRAME_SIZE_BITS.
• Fix: Don't clobber msgInfo register in PageGetAddress, ASIDControlInvocation, ARMCBInvocation, A32PageDirectoryGetStatusBits, X86PortIn, WriteVMCS, ReadVMCS, ConfigureSingleStepping, and GetBreakpoint.
• libsel4: Make bootinfo consistent. Some slot positions in the rootnode would depend on configuration. However that makes it difficult to add new root caps, especially if multiple caps only exist based on configuration. Make all caps always there, but null if not configured.
• libsel4: Eliminate unnamed enums
• Improved consistency and completeness of user manual
• Added manual for bitfield generator
• bitfield_gen: allow non-contiguous tag fields. A tagged union can now optionally use multiple fields to indicate the tag. These are called "sliced" tags in the code. The tag fields have to be at the same position and width in each block of the tagged unions, and all tag fields have to be within the same word. See the manual for details.
• Make CONFIG_PRINTING and CONFIG_DEBUG_BUILD usable independently from each other
• Overall debug printing improvements
• Fix invisible chars due to ANSI escape codes. On terminals with black background some debug output was invisible due to black foreground colour. Use bold instead.
• Rename libsel4 config option ENABLE_SMP_SUPPORT to CONFIG_ENABLE_SMP_SUPPORT to be namespace compliant.
• Removed obsolete define HAVE_AUTOCONF
• Remove userError from seL4_ReplyRecv path, because it too often incorrectly warns about legitimate operations
• Update GDB macros
• Add support for cmake --install <dir> for final build and config artefacts
• cmake: support supplying custom device trees overrides
• cmake: provide gen_config.json with kernel config settings
• Provide platform_gen.json in addition to platform_gen.yaml
• Allow build with GNU binutils >= 2.38
• Allow compilation with clang 12
• cmake: detect x86 cross-compiler for Arm host (e.g. on Apple M1)
• Consistently use /usr/bin/env for bash/sh invocations
• Require Python 3 consistently everywhere
• Improved support for compiling on MacOS
• General minor build system improvements and clean-up
• Set up automated tests for CI and GitHub pull requests on seL4
• Add vulnerability disclosure policy

Upgrade Notes

• The change in seL4_MinSchedContextBits can lead to failure where code previously created scheduling contexts with size seL4_MinSchedContextBits and expected more than the 2 minimum refills to be available for that size. Either use a larger size (the previous value 8 of seL4_MinSchedContextBits) in retyping, or request fewer refills.

Full changelog

Refer to the git log in https://github.com/seL4/seL4 using git log 12.1.0..13.0.0

More details

See the 13.0.0 manual included in the release or ask on the mailing list!