## @file # # Technical notes for the virtio-net driver. # # Copyright (C) 2013, Red Hat, Inc. # # This program and the accompanying materials are licensed and made available # under the terms and conditions of the BSD License which accompanies this # distribution. The full text of the license may be found at # http://opensource.org/licenses/bsd-license.php # # THE PROGRAM IS DISTRIBUTED UNDER THE BSD LICENSE ON AN "AS IS" BASIS, WITHOUT # WARRANTIES OR REPRESENTATIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED. # ## Disclaimer ---------- All statements concerning standards and specifications are informative and not normative. They are made in good faith. Corrections are most welcome on the edk2-devel mailing list. The following documents have been perused while writing the driver and this document: - Unified Extensible Firmware Interface Specification, Version 2.3.1, Errata C; June 27, 2012 - Driver Writer's Guide for UEFI 2.3.1, 03/08/2012, Version 1.01; - Virtio PCI Card Specification, v0.9.5 DRAFT, 2012 May 7. Summary ------- The VirtioNetDxe UEFI_DRIVER implements the Simple Network Protocol for virtio-net devices. Higher level protocols are automatically installed on top of it by the DXE Core / the ConnectController() boot service, enabling for virtio-net devices eg. DHCP configuration, TCP transfers with edk2 StdLib applications, and PXE booting in OVMF. UEFI driver structure --------------------- A driver instance, belonging to a given virtio-net device, can be in one of four states at any time. The states stack up as follows below. The state transitions are labeled with the primary function (and its important callees faithfully indented) that implement the transition. | ^ | | [DriverBinding.c] | | [DriverBinding.c] VirtioNetDriverBindingStart | | VirtioNetDriverBindingStop VirtioNetSnpPopulate | | VirtioNetSnpEvacuate VirtioNetGetFeatures | | v | +-------------------------+ | EfiSimpleNetworkStopped | +-------------------------+ | ^ [SnpStart.c] | | [SnpStop.c] VirtioNetStart | | VirtioNetStop | | v | +-------------------------+ | EfiSimpleNetworkStarted | +-------------------------+ | ^ [SnpInitialize.c] | | [SnpShutdown.c] VirtioNetInitialize | | VirtioNetShutdown VirtioNetInitRing {Rx, Tx} | | VirtioNetShutdownRx [SnpSharedHelpers.c] VirtioRingInit | | VirtIo->UnmapSharedBuffer VirtioRingMap | | VirtIo->FreeSharedPages VirtioNetInitTx | | VirtioNetShutdownTx [SnpSharedHelpers.c] VirtIo->AllocateShare... | | VirtIo->UnmapSharedBuffer VirtioMapAllBytesInSh... | | VirtIo->FreeSharedPages VirtioNetInitRx | | VirtioNetUninitRing [SnpSharedHelpers.c] VirtIo->AllocateShare... | | {Tx, Rx} VirtioMapAllBytesInSh... | | VirtIo->UnmapSharedBuffer | | VirtioRingUninit v | +-----------------------------+ | EfiSimpleNetworkInitialized | +-----------------------------+ The state at the top means "nonexistent" and is hence unnamed on the diagram -- a driver instance actually doesn't exist at that point. The transition functions out of and into that state implement the Driver Binding Protocol. The lower three states characterize an existent driver instance and are all states defined by the Simple Network Protocol. The transition functions between them are member functions of the Simple Network Protocol. Each transition function validates its expected source state and its parameters. For example, VirtioNetDriverBindingStop will refuse to disconnect from the controller unless it's in EfiSimpleNetworkStopped. Driver instance states (Simple Network Protocol) ------------------------------------------------ In the EfiSimpleNetworkStopped state, the virtio-net device is (has been) re-set. No resources are allocated for networking / traffic purposes. The MAC address and other device attributes have been retrieved from the device (this is necessary for completing the VirtioNetDriverBindingStart transition). The EfiSimpleNetworkStarted is completely identical to the EfiSimpleNetworkStopped state for virtio-net, in the functional and resource-usage sense. This state is mandated / provided by the Simple Network Protocol for flexibility that the virtio-net driver doesn't exploit. In particular, the EfiSimpleNetworkStarted state is the target of the Shutdown SNP member function, and must therefore correspond to a hardware configuration where "[it] is safe for another driver to initialize". (Clearly another UEFI driver could not do that due to the exclusivity of the driver binding that VirtioNetDriverBindingStart() installs, but a later OS driver might qualify.) The EfiSimpleNetworkInitialized state is the live state of the virtio NIC / the driver instance. Virtio and other resources required for network traffic have been allocated, and the following SNP member functions are available (in addition to VirtioNetShutdown which leaves the state): - VirtioNetReceive [SnpReceive.c]: poll the virtio NIC for an Rx packet that may have arrived asynchronously; - VirtioNetTransmit [SnpTransmit.c]: queue a Tx packet for asynchronous transmission (meant to be used together with VirtioNetGetStatus); - VirtioNetGetStatus [SnpGetStatus.c]: query link status and status of pending Tx packets; - VirtioNetMcastIpToMac [SnpMcastIpToMac.c]: transform a multicast IPv4/IPv6 address into a multicast MAC address; - VirtioNetReceiveFilters [SnpReceiveFilters.c]: emulate unicast / multicast / broadcast filter configuration (not their actual effect -- a more liberal filter setting than requested is allowed by the UEFI specification). The following SNP member functions are not supported [SnpUnsupported.c]: - VirtioNetReset: reinitialize the virtio NIC without shutting it down (a loop from/to EfiSimpleNetworkInitialized); - VirtioNetStationAddress: assign a new MAC address to the virtio NIC, - VirtioNetStatistics: collect statistics, - VirtioNetNvData: access non-volatile data on the virtio NIC. Missing support for these functions is allowed by the UEFI specification and doesn't seem to trip up higher level protocols. Events and task priority levels ------------------------------- The UEFI specification defines a sophisticated mechanism for asynchronous events / callbacks (see "6.1 Event, Timer, and Task Priority Services" for details). Such callbacks work like software interrupts, and some notion of locking / masking is important to implement critical sections (atomic or exclusive access to data or a device). This notion is defined as Task Priority Levels. The virtio-net driver for OVMF must concern itself with events for two reasons: - The Simple Network Protocol provides its clients with a (non-optional) WAIT type event called WaitForPacket: it allows them to check or wait for Rx packets by polling or blocking on this event. (This functionality overlaps with the Receive member function.) The event is available to clients starting with EfiSimpleNetworkStopped (inclusive). The virtio-net driver is informed about such client polling or blockage by receiving an asynchronous callback (a software interrupt). In the callback function the driver must interrogate the driver instance state, and if it is EfiSimpleNetworkInitialized, access the Rx queue and see if any packets are available for consumption. If so, it must signal the WaitForPacket WAIT type event, waking the client. For simplicity and safety, all parts of the virtio-net driver that access any bit of the driver instance (data or device) run at the TPL_CALLBACK level. This is the highest level allowed for an SNP implementation, and all code protected in this manner satisfies even stricter non-blocking requirements than what's documented for TPL_CALLBACK. The task priority level for the WaitForPacket callback too is set by the driver, the choice is TPL_CALLBACK again. This in effect serializes the WaitForPacket callback (VirtioNetIsPacketAvailable [Events.c]) with "normal" parts of the driver. - According to the Driver Writer's Guide, a network driver should install a callback function for the global EXIT_BOOT_SERVICES event (a special NOTIFY type event). When the ExitBootServices() boot service has cleaned up internal firmware state and is about to pass control to the OS, any network driver has to stop any in-flight DMA transfers, lest it corrupts OS memory. For this reason EXIT_BOOT_SERVICES is emitted and the network driver must abort in-flight DMA transfers. This callback (VirtioNetExitBoot) is synchronized with the rest of the driver code just the same as explained for WaitForPacket. In EfiSimpleNetworkInitialized state it resets the virtio NIC, halting all data transfer. After the callback returns, no further driver code is expected to be scheduled. Virtio internals -- Rx ---------------------- Requests (Rx and Tx alike) are always submitted by the guest and processed by the host. For Tx, processing means transmission. For Rx, processing means filling in the request with an incoming packet. Submitted requests exist on the "Available Ring", and answered (processed) requests show up on the "Used Ring". Packet data includes the media (Ethernet) header: destination MAC, source MAC, and Ethertype (14 bytes total). The following structures implement packet reception. Most of them are defined in the Virtio specification, the only driver-specific trait here is the static pre-configuration of the two-part descriptor chains, in VirtioNetInitRx. The diagram is simplified. Available Index Available Index last processed incremented by the host by the guest v -------> v Available +-------+-------+-------+-------+-------+ Ring |DescIdx|DescIdx|DescIdx|DescIdx|DescIdx| +-------+-------+-------+-------+-------+ =D6 =D2 D2 D3 D4 D5 D6 D7 Descr. +----------+----------++----------+----------++----------+----------+ Table |Adr:Len:Nx|Adr:Len:Nx||Adr:Len:Nx|Adr:Len:Nx||Adr:Len:Nx|Adr:Len:Nx| +----------+----------++----------+----------++----------+----------+ =A2 =D3 =A3 =A4 =D5 =A5 =A6 =D7 =A7 A2 A3 A4 A5 A6 A7 Receive +---------------+---------------+---------------+ Destination |vnet hdr:packet|vnet hdr:packet|vnet hdr:packet| Area +---------------+---------------+---------------+ Used Index Used Index incremented last processed by the guest by the host v -------> v Used +-----------+-----------+-----------+-----------+-----------+ Ring |DescIdx:Len|DescIdx:Len|DescIdx:Len|DescIdx:Len|DescIdx:Len| +-----------+-----------+-----------+-----------+-----------+ =D4 In VirtioNetInitRx, the guest allocates the fixed size Receive Destination Area, which accommodates all packets delivered asynchronously by the host. To each packet, a slice of this area is dedicated; each slice is further subdivided into virtio-net request header and network packet data. The (device-physical) addresses of these sub-slices are denoted with A2, A3, A4 and so on. Importantly, an even-subscript "A" always belongs to a virtio-net request header, while an odd-subscript "A" always belongs to a packet sub-slice. Furthermore, the guest lays out a static pattern in the Descriptor Table. For each packet that can be in-flight or already arrived from the host, VirtioNetInitRx sets up a separate, two-part descriptor chain. For packet N, the Nth descriptor chain is set up as follows: - the first (=head) descriptor, with even index, points to the fixed-size sub-slice receiving the virtio-net request header, - the second descriptor (with odd index) points to the fixed (1514 byte) size sub-slice receiving the packet data, - a link from the first (head) descriptor in the chain is established to the second (tail) descriptor in the chain. Finally, the guest populates the Available Ring with the indices of the head descriptors. All descriptor indices on both the Available Ring and the Used Ring are even. Packet reception occurs as follows: - The host consumes a descriptor index off the Available Ring. This index is even (=2*N), and fingers the head descriptor of the chain belonging to packet N. - The host reads the descriptors D(2*N) and -- following the Next link there --- D(2*N+1), and stores the virtio-net request header at A(2*N), and the packet data at A(2*N+1). - The host places the index of the head descriptor, 2*N, onto the Used Ring, and sets the Len field in the same Used Ring Element to the total number of bytes transferred for the entire descriptor chain. This enables the guest to identify the length of Rx packets. - VirtioNetReceive polls the Used Ring. If a new Used Ring Element shows up, it copies the data out to the caller, and recycles the index of the head descriptor (ie. 2*N) to the Available Ring. - Because the host can process (answer) Rx requests in any order theoretically, the order of head descriptor indices on each of the Available Ring and the Used Ring is virtually random. (Except right after the initial population in VirtioNetInitRx, when the Available Ring is full and increasing, and the Used Ring is empty.) - If the Available Ring is empty, the host is forced to drop packets. If the Used Ring is empty, VirtioNetReceive returns EFI_NOT_READY (no packet available). Virtio internals -- Tx ---------------------- The transmission structure erected by VirtioNetInitTx is similar, it differs in the following: - There is no Receive Destination Area. - Each head descriptor, D(2*N), points to a read-only virtio-net request header that is shared by all of the head descriptors. This virtio-net request header is never modified by the host. - Each tail descriptor is re-pointed to the device-mapped address of the caller-supplied packet buffer whenever VirtioNetTransmit places the corresponding head descriptor on the Available Ring. A reverse mapping, from the device-mapped address to the caller-supplied packet address, is saved in an associative data structure that belongs to the driver instance. - Per spec, the caller is responsible to hang on to the unmodified packet buffer until it is reported transmitted by VirtioNetGetStatus. Steps of packet transmission: - Client code calls VirtioNetTransmit. VirtioNetTransmit tracks free descriptor chains by keeping the indices of their head descriptors in a stack that is private to the driver instance. All elements of the stack are even. - If the stack is empty (that is, each descriptor chain, in isolation, is either pending transmission, or has been processed by the host but not yet recycled by a VirtioNetGetStatus call), then VirtioNetTransmit returns EFI_NOT_READY. - Otherwise the index of a free chain's head descriptor is popped from the stack. The linked tail descriptor is re-pointed as discussed above. The head descriptor's index is pushed on the Available Ring. - The host moves the head descriptor index from the Available Ring to the Used Ring when it transmits the packet. - Client code calls VirtioNetGetStatus. In case the Used Ring is empty, the function reports no Tx completion. Otherwise, a head descriptor's index is consumed from the Used Ring and recycled to the private stack. The client code's original packet buffer address is calculated by fetching the device-mapped address from the tail descriptor (where it has been stored at VirtioNetTransmit time), and by looking up the device-mapped address in the associative data structure. The reverse-mapped packet buffer address is returned to the caller. - The Len field of the Used Ring Element is not checked. The host is assumed to have transmitted the entire packet -- VirtioNetTransmit had forced it below 1514 bytes (inclusive). The Virtio specification suggests this packet size is always accepted (and a lower MTU could be encountered on any later hop as well). Additionally, there's no good way to report a short transmit via VirtioNetGetStatus; EFI_DEVICE_ERROR seems too serious from the specification and higher level protocols could interpret it as a fatal condition. - The host can theoretically reorder head descriptor indices when moving them from the Available Ring to the Used Ring (out of order transmission). Because of this (and the choice of a stack over a list for free descriptor chain tracking) the order of head descriptor indices on either Ring is unpredictable.