1.. SPDX-License-Identifier: GPL-2.0 2 3.. _virtio: 4 5=============== 6Virtio on Linux 7=============== 8 9Introduction 10============ 11 12Virtio is an open standard that defines a protocol for communication 13between drivers and devices of different types, see Chapter 5 ("Device 14Types") of the virtio spec (`[1]`_). Originally developed as a standard 15for paravirtualized devices implemented by a hypervisor, it can be used 16to interface any compliant device (real or emulated) with a driver. 17 18For illustrative purposes, this document will focus on the common case 19of a Linux kernel running in a virtual machine and using paravirtualized 20devices provided by the hypervisor, which exposes them as virtio devices 21via standard mechanisms such as PCI. 22 23 24Device - Driver communication: virtqueues 25========================================= 26 27Although the virtio devices are really an abstraction layer in the 28hypervisor, they're exposed to the guest as if they are physical devices 29using a specific transport method -- PCI, MMIO or CCW -- that is 30orthogonal to the device itself. The virtio spec defines these transport 31methods in detail, including device discovery, capabilities and 32interrupt handling. 33 34The communication between the driver in the guest OS and the device in 35the hypervisor is done through shared memory (that's what makes virtio 36devices so efficient) using specialized data structures called 37virtqueues, which are actually ring buffers [#f1]_ of buffer descriptors 38similar to the ones used in a network device: 39 40.. kernel-doc:: include/uapi/linux/virtio_ring.h 41 :identifiers: struct vring_desc 42 43All the buffers the descriptors point to are allocated by the guest and 44used by the host either for reading or for writing but not for both. 45 46Refer to Chapter 2.5 ("Virtqueues") of the virtio spec (`[1]`_) for the 47reference definitions of virtqueues and "Virtqueues and virtio ring: How 48the data travels" blog post (`[2]`_) for an illustrated overview of how 49the host device and the guest driver communicate. 50 51The :c:type:`vring_virtqueue` struct models a virtqueue, including the 52ring buffers and management data. Embedded in this struct is the 53:c:type:`virtqueue` struct, which is the data structure that's 54ultimately used by virtio drivers: 55 56.. kernel-doc:: include/linux/virtio.h 57 :identifiers: struct virtqueue 58 59The callback function pointed by this struct is triggered when the 60device has consumed the buffers provided by the driver. More 61specifically, the trigger will be an interrupt issued by the hypervisor 62(see vring_interrupt()). Interrupt request handlers are registered for 63a virtqueue during the virtqueue setup process (transport-specific). 64 65.. kernel-doc:: drivers/virtio/virtio_ring.c 66 :identifiers: vring_interrupt 67 68 69Device discovery and probing 70============================ 71 72In the kernel, the virtio core contains the virtio bus driver and 73transport-specific drivers like `virtio-pci` and `virtio-mmio`. Then 74there are individual virtio drivers for specific device types that are 75registered to the virtio bus driver. 76 77How a virtio device is found and configured by the kernel depends on how 78the hypervisor defines it. Taking the `QEMU virtio-console 79<https://gitlab.com/qemu-project/qemu/-/blob/master/hw/char/virtio-console.c>`__ 80device as an example. When using PCI as a transport method, the device 81will present itself on the PCI bus with vendor 0x1af4 (Red Hat, Inc.) 82and device id 0x1003 (virtio console), as defined in the spec, so the 83kernel will detect it as it would do with any other PCI device. 84 85During the PCI enumeration process, if a device is found to match the 86virtio-pci driver (according to the virtio-pci device table, any PCI 87device with vendor id = 0x1af4):: 88 89 /* Qumranet donated their vendor ID for devices 0x1000 thru 0x10FF. */ 90 static const struct pci_device_id virtio_pci_id_table[] = { 91 { PCI_DEVICE(PCI_VENDOR_ID_REDHAT_QUMRANET, PCI_ANY_ID) }, 92 { 0 } 93 }; 94 95then the virtio-pci driver is probed and, if the probing goes well, the 96device is registered to the virtio bus:: 97 98 static int virtio_pci_probe(struct pci_dev *pci_dev, 99 const struct pci_device_id *id) 100 { 101 ... 102 103 if (force_legacy) { 104 rc = virtio_pci_legacy_probe(vp_dev); 105 /* Also try modern mode if we can't map BAR0 (no IO space). */ 106 if (rc == -ENODEV || rc == -ENOMEM) 107 rc = virtio_pci_modern_probe(vp_dev); 108 if (rc) 109 goto err_probe; 110 } else { 111 rc = virtio_pci_modern_probe(vp_dev); 112 if (rc == -ENODEV) 113 rc = virtio_pci_legacy_probe(vp_dev); 114 if (rc) 115 goto err_probe; 116 } 117 118 ... 119 120 rc = register_virtio_device(&vp_dev->vdev); 121 122When the device is registered to the virtio bus the kernel will look 123for a driver in the bus that can handle the device and call that 124driver's ``probe`` method. 125 126It's at this stage that the virtqueues will be allocated and configured 127by calling the appropriate ``virtio_find`` helper function, such as 128virtio_find_single_vq() or virtio_find_vqs(), which will end up 129calling a transport-specific ``find_vqs`` method. 130 131 132References 133========== 134 135_`[1]` Virtio Spec v1.2: 136https://docs.oasis-open.org/virtio/virtio/v1.2/virtio-v1.2.html 137 138Check for later versions of the spec as well. 139 140_`[2]` Virtqueues and virtio ring: How the data travels 141https://www.redhat.com/en/blog/virtqueues-and-virtio-ring-how-data-travels 142 143.. rubric:: Footnotes 144 145.. [#f1] that's why they may be also referred to as virtrings. 146