master

Fflyking001virtio: add gpu snapshot vmstate

fb5bd8ba创建于 4 天前历史提交

文件	最后提交记录	最后更新时间
aarch64	machine: gate guest memory writes by migration state 1. Determine boot, SMBIOS, and ACPI guest memory writes from the VM migration state in shared machine helpers. 2. Skip direct-boot and DTB guest memory writes during incoming restore while preserving fwcfg-based setup. 3. Keep reset-time direct-boot reload behavior so guest memory is rewritten after migration completes. (cherry picked from commit d955b95d449169eae4b42f09ff65e5f3c00f60fc)	4 天前
micro_common	machine: expose guest RAM snapshot query data Implement the snapshot memory QMP queries for standard and micro machines. The mapping query returns RAM host addresses, sizes, page sizes, and flat image offsets. The page-state query returns resident and zero-page bitmaps using the same flat RAM order. Residency is derived from /proc/self/pagemap rather than mincore, so pages that are present in RAM or swapped out are both reported as resident. Swapped pages are read back in before zero-page classification so the empty bitmap remains exact. All snapshot memory query paths require the VM to be paused so external tools observe a stable memory layout and page-state view. The dirty bitmap command uses the UFFD query-and-reset helper introduced earlier. (cherry picked from commit 8eb89af3594843237af9edcc64d6cc34b0ab4f4f)	4 天前
standard_common	machine: expose guest RAM snapshot query data Implement the snapshot memory QMP queries for standard and micro machines. The mapping query returns RAM host addresses, sizes, page sizes, and flat image offsets. The page-state query returns resident and zero-page bitmaps using the same flat RAM order. Residency is derived from /proc/self/pagemap rather than mincore, so pages that are present in RAM or swapped out are both reported as resident. Swapped pages are read back in before zero-page classification so the empty bitmap remains exact. All snapshot memory query paths require the VM to be paused so external tools observe a stable memory layout and page-state view. The dirty bitmap command uses the UFFD query-and-reset helper introduced earlier. (cherry picked from commit 8eb89af3594843237af9edcc64d6cc34b0ab4f4f)	4 天前
x86_64	fix(snapshot): allow syscalls and ioctls required by snapshot restore in seccomp Snapshot restore from file was killed by the seccomp filter because several syscalls and ioctls used on the restore path were not allowlisted: - Define the KVM_SET_PIT2, KVM_SET_CLOCK and KVM_SET_IRQCHIP ioctl numbers in the hypervisor crate and allow them in the micro and standard VM x86_64 ioctl allowlists (the SET counterparts of the GET ioctls already allowed for snapshot save) - Micro VM: allow KVM setup ioctls re-issued during device restore (KVM_SET_USER_MEMORY_REGION, KVM_IOEVENTFD, KVM_IRQFD, KVM_SET_GSI_ROUTING, KVM_SIGNAL_MSI), the VHOST_GET_FEATURES and VHOST_RESET_OWNER ioctls, and SYS_clock_nanosleep used by glibc sleep/backoff - Micro and standard VM: allow additional futex ops observed on glibc hosts during restore (FUTEX_WAKE_BITSET, FUTEX_REQUEUE and, on gnu, FUTEX_LOCK_PI2) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com> Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> (cherry picked from commit 6eb05950dda68aea3d2dd4497762ed95a64ad139)	4 天前
error.rs	Hypervisor: Remove hypervisor directory Remove the hypervisor directory and cleans up all references to it. All features and functionality previously located in hypervisor have been fully migrated to the hypervisor_refactor directory. So it is safe to remove the directory. Signed-off-by: Jinhao Gao <gaojinhao@huawei.com>	2 年前
lib.rs	virtio: add gpu snapshot vmstate 1. Add virtio-gpu vmstate metadata and framebuffer blob serialization. 2. Share GPU runtime state between the device and IO handler for snapshot save and restore. 3. Register virtio-gpu on the normal device vmstate path and allow its large state payload. Co-Authored-By: GPT-5.5 (cherry picked from commit a6e015e708effa1ee2f4e37324c1777eb5962dcd)	4 天前