Tags: firecracker-microvm/firecracker
Tags
Added * [#4987](#4987): Reset physical counter register (`CNTPCT_EL0`) on VM startup. This avoids VM reading the host physical counter value. This is only possible on 6.4 and newer kernels. For older kernels physical counter will still be passed to the guest unmodified. See more info [here](https://github.com/firecracker-microvm/firecracker/blob/main/docs/prod-host-setup.md#arm-only-vm-physical-counter-behaviour) * [#5088](#5088): Added AMD Genoa as a supported and tested platform for Firecracker. Changed * [#4913](#4913): Removed unnecessary fields (`max_connections` and `max_pending_resets`) from the snapshot format, bumping the snapshot version to 5.0.0. Users need to regenerate snapshots. * [#4926](#4926): Replace underlying implementation for seccompiler from in house one in favor of `libseccomp` which produces smaller and more optimized BPF code. Deprecated Removed Fixed * [#4921](#4921): Fixed swagger `CpuConfig` definition to include missing aarch64-specific fields. * [#4916](#4916): Fixed `IovDeque` implementation to work with any host page size. This fixes virtio-net device on non 4K host kernels. * [#4991](#4991): Fixed `mem_size_mib` and `track_dirty_pages` being mandatory for all `PATCH /machine-config` requests. Now, they can be omitted which leaves these parts of the machine configuration unchanged. * [#5007](#5007): Fixed watchdog softlockup warning on x86_64 guests when a vCPU is paused during GDB debugging. * [#5021](#5021) If a balloon device is inflated post UFFD-backed snapshot restore, Firecracker now causes `remove` UFFD messages to be sent to the UFFD handler. Previously, no such message would be sent. * [#5034](#5034): Fix an integer underflow in the jailer when computing the value it passes to Firecracker's `--parent-cpu-time-us` values, which caused development builds of Firecracker to crash (but production builds were unaffected as underflows do not panic in release mode). * [#5045](#5045): Fixed an issue where firecracker intermittently receives SIGHUP when using jailer with `--new-pid-ns` but without `--daemonize`. * [#4995](#4995): Firecracker no longer overwrites CPUID leaf 0x80000000 when running AMD hardware, meaning the guest can now discover a greater range of CPUID leaves in the extended function range (this range is host kernel dependent). * [#5046](#5046): Retry KVM_CREATE_VM on EINTR that occasionally happen on heavily loaded hosts to improve reliability of microVM creation. * [#5052](#5052): Build the empty seccomp policy as default for debug builds to avoid crashes on syscalls introduced by debug assertions from Rust 1.80.0.
Added * [#4834](#4834): Add `VIRTIO_NET_F_RX_MRGBUF` support to the `virtio-net` device. When this feature is negotiated, guest `virtio-net` driver can perform more efficient memory management which in turn improves RX and TX performance. * [#4460](#4460): Add a call to [`KVM_KVMCLOCK_CTRL`](https://docs.kernel.org/virt/kvm/api.html#kvm-kvmclock-ctrl) after pausing vCPUs on x86_64 architectures. This ioctl sets a flag in the KVM state of the vCPU indicating that it has been paused by the host userspace. In guests that use kvmclock, the soft lockup watchdog checks this flag. If it is set, it won't trigger the lockup condition. Calling the ioctl for guests that don't use kvmclock will fail. These failures are not fatal. We log the failure and increase the `vcpu.kvmclock_ctrl_fails` metric. * [#4869](#4869): Added support for Aarch64 systems which feature CPU caches with a number of sets higher than `u16::MAX`. * [#4797](#4797), [#4854](#4854): Added GDB debugging support for a microVM guest kernel. Please see our [GDB debugging documentation](docs/gdb-debugging.md) for more information. Changed * [#4844](#4844): Upgrade `virtio-net` device to use `readv` syscall to avoid unnecessary memory copies on RX path, increasing the RX performance. Deprecated Removed * [#4804](#4804): Drop Support for guest kernel 4.14. Linux 4.14 reached end-of-life in [January 2024](https://lore.kernel.org/lkml/2024011046-ecology-tiptoeing-ce50@gregkh/) The minimum supported guest kernel now is 5.10. Fixed * [#4796](#4796): Fixed Vsock not notifying guest about `TRANSPORT_RESET_EVENT` event after snapshot restore. This resulted in guest waiting indefinitely on a connection which was reset during snapshot creation. * [#4790](#4790): v1.9.0 was missing most of the debugging information in the debuginfo file, due to a change in the Cargo defaults. This has been corrected. * [#4826](#4826): Add missing configuration of tap offload features when restoring from a snapshot. Setting the features was previously [moved](49ed5ea) from net device creation to device activation time, but it was not reflected in the restore path. This was leading to inability to connect to the restored VM if the offload features were used.
Fixed * [#4824](#4824): Add missing configuration of tap offload features when restoring from a snapshot. Setting the features was previously [moved](49ed5ea) from net device creation to device activation time, but it was not reflected in the restore path. This was leading to inability to connect to the restored VM if the offload features were used. * [#4829](#4829): v1.9.0 was missing most of the debugging information in the debuginfo file, due to a change in the Cargo defaults. This has been corrected.
Added * [#4687](#4687): Added VMGenID support for microVMs running on ARM hosts with 6.1 guest kernels. Support for VMGenID via DeviceTree bindings exists only on mainline 6.10 Linux onwards. Users of Firecracker will need to backport the relevant patches on top of their 6.1 kernels to make use of the feature. * [#4732](#4732), [#4733](#4733), [#4741](#4741), [#4746](#4746): Added official support for 6.1 microVM guest kernels. Changed Deprecated * Support for guest kernel 4.14 is now deprecated. We will completely remove 4.14 support with Firecracker version v1.10 Removed * [#4689](#4689): Drop support for host kernel 4.14. Linux 4.14 reached end-of-life in [January 2024](https://lore.kernel.org/lkml/2024011046-ecology-tiptoeing-ce50@gregkh/). The minimum supported kernel now is 5.10. Guest kernel 4.14 is still supported. Fixed * [4680](#4680): Fixed an issue ([#4659](#4659)) where the virtio-net device implementation would always assume the guest accepts all VirtIO features the device offers. This is always true with the Linux guest kernels we are testing but other kernels, like FreeBSD make different assumptions. This PR fixes the emulation code to set the TAP features based on the features accepted by the guest.
Added * [#4428](#4428): Added ACPI support to Firecracker for x86_64 microVMs. Currently, we pass ACPI tables with information about the available vCPUs, interrupt controllers, VirtIO and legacy x86 devices to the guest. This allows booting kernels without MPTable support. Please see our [kernel policy documentation](docs/kernel-policy.md) for more information regarding relevant kernel configurations. * [#4487](#4487): Added support for the Virtual Machine Generation Identifier (VMGenID) device on x86_64 platforms. VMGenID is a virtual device that allows VMMs to notify guests when they are resumed from a snapshot. Linux includes VMGenID support since version 5.18. It uses notifications from the device to reseed its internal CSPRNG. Please refer to [snapshot support](docs/snapshotting/snapshot-support.md) and [random for clones](docs/snapshotting/random-for-clones.md) documention for more info on VMGenID. VMGenID state is part of the snapshot format of Firecracker. As a result, Firecracker snapshot version is now 2.0.0. Changed * [#4492](#4492): Changed `--config` parameter of `cpu-template-helper` optional. Users no longer need to prepare kernel, rootfs and Firecracker configuration files to use `cpu-template-helper`. * [#4537](#4537) Changed T2CL template to pass through bit 27 and 28 of `MSR_IA32_ARCH_CAPABILITIES` (`RFDS_NO` and `RFDS_CLEAR`) since KVM consider they are able to be passed through and T2CL isn't designed for secure snapshot migration between different processors. * [#4537](#4537) Changed T2S template to set bit 27 of `MSR_IA32_ARCH_CAPABILITIES` (`RFDS_NO`) to 1 since it assumes that the fleet only consists of processors that are not affected by RFDS. * [#4388](#4388): Avoid setting `kvm_immediate_exit` to 1 if are already handling an exit, or if the vCPU is stopped. This avoids a spurious KVM exit upon restoring snapshots. * [#4567](#4567): Do not initialize vCPUs in powered-off state upon snapshot restore. No functional change, as vCPU initialization is only relevant for the booted case (where the guest expects CPUs to be powered off). Deprecated * Firecracker's `--start-time-cpu-us` and `--start-time-us` parameters are deprecated and will be removed in v2.0 or later. They are used by the jailer to pass the value that should be subtracted from the (CPU) time, when emitting the `start_time_us` and `start_time_cpu_us` metrics. These parameters were never meant to be used by end customers, and we recommend doing any such time adjustments outside Firecracker. * Booting with microVM kernels that rely on MPTable on x86_64 is deprecated and support will be removed in v2.0 or later. We suggest to users of Firecracker to use guest kernels with ACPI support. For x86_64 microVMs, ACPI will be the only way Firecracker passes hardware information to the guest once MPTable support is removed. Fixed * [#4526](#4526): Added a check in the network TX path that the size of the network frames the guest passes to us is not bigger than the maximum frame the device expects to handle. On the TX path, we copy frames destined to MMDS from guest memory to Firecracker memory. Without the check, a mis-behaving virtio-net driver could cause an increase in the memory footprint of the Firecracker process. Now, if we receive such a frame, we ignore it and increase `Net::tx_malformed_frames` metric. * [#4536](#4536): Make the first differential snapshot taken after a full snapshot contain only the set of memory pages changed since the full snapshot. Previously, these differential snapshots would contain all memory pages. This will result in potentially much smaller differential snapshots after a full snapshot. * [#4578](#4578): Fix UFFD support not being forward-compatible with new ioctl options introduced in Linux 6.6. See also bytecodealliance/userfaultfd-rs#61. * [#4630](#4630): On x86_64, when taking a snapshot, if a vCPU has MSR_IA32_TSC_DEADLINE set to 0, Firecracker will replace it with the MSR_IA32_TSC value from the same vCPU. This is to guarantee that the vCPU will continue receiving TSC interrupts after restoring from the snapshot even if an interrupt is lost when taking a snapshot. * [#4666](#4666): Fixed Firecracker sometimes restoring `MSR_IA32_TSC_DEADLINE` before `MSR_IA32_TSC`. Now it always restores `MSR_IA32_TSC_DEADLINE` MSR after `MSR_IA32_TSC`, as KVM relies on the guest TSC for correct restoration of `MSR_IA32_TSC_DEADLINE`. This fixed guests using the `TSC_DEADLINE` hardware feature receiving incorrect timer interrupts after snapshot restoration, which could lead to them seemingly getting stuck in sleep-related syscalls (see also #4099).
PreviousNext