| 文件 | 最后提交记录 | 最后更新时间 |
|---|---|---|
virtio-snd: remove unnecessary mut The patch introduces helper function to get unmutable stream. Then update the code by removing unnecessary mut. Signed-off-by: chaisong <chai_song@126.com> (cherry picked from commit dabe80d661c881bde0e973727ca795a5752fbdd9) | 4 天前 | |
virtio-balloon: support memory snapshot of virtio-balloon Signed-off-by: HZ-Xiaohu <huzihao5@huawei.com> Signed-off-by: Fei Xu <xufei30@huawei.com> | 4 个月前 | |
virtio-blk: fix an deadlock problem when deactivate block When deactivate block device in vcpu thread, it will do the follow steps: 1) lock qcow2 driver 2) wait inflight io count decrease to zero For disk iothread, it trigger the flush metadata timer: 1) lock qcow2 driver stuck, wait vcpu thread unlock it When decrease inflight io count, it needs to run on disk iothead which has been stuck. So, it cause an deadlock. Do not lock qcow2 driver before inflight io count decrease to zero when deactivate block. Signed-off-by: Yan Wang <wangyan122@huawei.com> Signed-off-by: Fei Xu <xufei30@huawei.com> | 2 个月前 | |
virtio-gpu: optimize resource image snapshot 1. Skip saving resource image blobs when framebuffer resources directly reference guest memory. 2. Rebuild omitted resource images from guest backing memory during restore. Co-Authored-By: Codex <codex@openai.com> (cherry picked from commit b41fd4aa0fb61fc4cb405c80b53e02e40bbe5de8) | 4 天前 | |
virtio_input: remove needless Arc Signed-off-by: h957891668 <957891668@qq.com> (cherry picked from commit a2bf1d0d7023b95e9d0fd859d908497bd4e7c0bc) | 4 天前 | |
pluggable_addr: Fix virtio pluggable address base Use max size is not correct, as the machine ram does not alway start from address 0. Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com> Signed-off-by: Fei Xu <xufei30@huawei.com> | 2 个月前 | |
virtio: gate optional device modules 1. Add feature gates for virtio balloon, memory, serial, input, multitouch, and vhost-user-fs devices across the root, machine, and virtio crates. 2. Gate virtio module exports and vhost-user-fs exports so disabled features are not compiled into the default binary. 3. Keep machine device dispatch unified while returning explicit not-enabled errors from disabled virtio device paths. 4. Provide QMP fallback helpers for disabled balloon and virtio-mem support and verify the new feature gates. (cherry picked from commit 3d1b70499b4aaef84685a6e1a921aa2e14f07799) | 4 天前 | |
touchpad: Add BTN_TOOL_DOUBLETAP and BTN_TOOL_TRIPLETAP support Added events handling for BTN_TOOL_DOUBLETAP and BTN_TOOL_TRIPLETAP, fixing the unresponsive two-finger and three-finger gestures on linux vm. Signed-off-by: chaisong <chai_song@126.com> (cherry picked from commit 03005267361d4c1177e8841f5fcb178d4bb1ba91) | 4 天前 | |
feat(mmds): intercept MMDS-destined packets in the virtio-net TX handler Add mmds_tcp_stack: Arc<Mutex<Option<MmdsTcpStack>>> and iface_id to NetIoQueue; add mmds_store: Option<Arc<Mutex<MmdsStore>>> to Net. In the TX path, check the Ethernet frame's destination IP against the MMDS endpoint address before forwarding to TAP. On match: copy the full packet, call MmdsTcpStack::process(), and inject each response frame into the guest RX virtqueue via inject_rx_packet(). Mark the TX descriptor consumed without forwarding to TAP. inject_rx_packet() uses try_lock on the RX queue to avoid deadlocking with the RX event handler; it silently drops the response if no RX buffer is available. The Arc<Mutex<Option<...>>> pattern lets mmds-config swap or clear the stack atomically at runtime without restarting the device. Also extract complete_elem() to deduplicate virtqueue completion logic shared between the RX and TX paths. (cherry picked from commit 4a3afecd463e58d04be7cf9dad326ab56e67fb12) | 4 天前 | |
virtio: support read-only pmem file backend Support using a memory-backend-file as a read-only backing file for virtio-pmem private mappings. When a virtio-pmem file backend is configured with share=off, open the existing backing file read-only, skip create, set_len and unlink handling, and validate that the file is large enough for the configured pmem size. The pmem mapping remains writable for the guest. With share=off it is mapped privately, so guest writes are handled by the private mapping and are not written back to the backing file. With share=on, keep the persistent file backend behavior and allow the backing file to be opened writable and extended when needed. This lets virtio-pmem use immutable image files without triggering host-side writeback or overlayfs copy-up. Signed-off-by: Yekelu <yekelu1@huawei.com> (cherry picked from commit 1716986788bdb8448e21cdc64b3198b6cf3486e2) | 4 天前 | |
migration: add restored version when restoring vm state Add restored version when restoring vm state. Signed-off-by: Yan Wang <wangyan122@huawei.com> | 4 个月前 | |
Revise some comments. Signed-off-by: HZ-Xiaohu <huzihao5@huawei.com>Signed-off-by: Fei Xu <xufei30@huawei.com> | 2 个月前 | |
virtio-serial: reduce the queue size Current code uses 256 queue size for all queues. It's too large for virtio serial and the driver consumes longer time. According to QEMU code, 32 for control queues and 128 for port queues is enough. This can reduce about 1.5s boot-up time. In addition use hashmap to index the specific port via port number. Signed-off-by: Zhao Yi Min <zhaoyimin1@huawei.com> | 4 个月前 |