X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fdevices%2Flnx_virtio_nic.c;h=bb13a6959e54f553105c625f693c012c9d8a70ee;hb=16047acf61f82cb4ae33bcb84b221c2098215172;hp=d3feff759c8e339d67ae909d2739a4bb5737209d;hpb=bbb2c127976c9cbb64af9e753a70f1bdc6310704;p=palacios.git diff --git a/palacios/src/devices/lnx_virtio_nic.c b/palacios/src/devices/lnx_virtio_nic.c index d3feff7..bb13a69 100644 --- a/palacios/src/devices/lnx_virtio_nic.c +++ b/palacios/src/devices/lnx_virtio_nic.c @@ -7,13 +7,13 @@ * and the University of New Mexico. You can find out more at * http://www.v3vee.org * - * Copyright (c) 2008, Jack Lange - * Copyright (c) 2008, Lei Xia - * Copyright (c) 2008, The V3VEE Project + * Copyright (c) 2010, Lei Xia + * Copyright (c) 2010, Cui Zheng + * Copyright (c) 2010, The V3VEE Project * All rights reserved. * - * Author: Jack Lange - * Lei Xia + * Author: Lei Xia + * Cui Zheng * * * This is free software. You are permitted to use, @@ -25,8 +25,12 @@ #include #include #include - +#include +#include +#include #include +#include +#include #ifndef CONFIG_DEBUG_VIRTIO_NET @@ -34,121 +38,87 @@ #define PrintDebug(fmt, args...) #endif -/* The feature bitmap for virtio net */ -#define VIRTIO_NET_F_CSUM 0 /* Host handles pkts w/ partial csum */ -#define VIRTIO_NET_F_GUEST_CSUM 1 /* Guest handles pkts w/ partial csum */ -#define VIRTIO_NET_F_MAC 5 /* Host has given MAC address. */ -#define VIRTIO_NET_F_GSO 6 /* Host handles pkts w/ any GSO type */ -#define VIRTIO_NET_F_GUEST_TSO4 7 /* Guest can handle TSOv4 in. */ -#define VIRTIO_NET_F_GUEST_TSO6 8 /* Guest can handle TSOv6 in. */ -#define VIRTIO_NET_F_GUEST_ECN 9 /* Guest can handle TSO[6] w/ ECN in. */ -#define VIRTIO_NET_F_GUEST_UFO 10 /* Guest can handle UFO in. */ -#define VIRTIO_NET_F_HOST_TSO4 11 /* Host can handle TSOv4 in. */ -#define VIRTIO_NET_F_HOST_TSO6 12 /* Host can handle TSOv6 in. */ -#define VIRTIO_NET_F_HOST_ECN 13 /* Host can handle TSO[6] w/ ECN in. */ -#define VIRTIO_NET_F_HOST_UFO 14 /* Host can handle UFO in. */ -#define VIRTIO_NET_F_MRG_RXBUF 15 /* Host can merge receive buffers. */ -#define VIRTIO_NET_F_STATUS 16 /* virtio_net_config.status available */ -#define VIRTIO_NET_F_CTRL_VQ 17 /* Control channel available */ -#define VIRTIO_NET_F_CTRL_RX 18 /* Control channel RX mode support */ -#define VIRTIO_NET_F_CTRL_VLAN 19 /* Control channel VLAN filtering */ -#define VIRTIO_NET_F_CTRL_RX_EXTRA 20 /* Extra RX mode control support */ -#define VIRTIO_NET_S_LINK_UP 1 /* Link is up */ - -/* Maximum packet size we can receive from tap device: header + 64k */ #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10)) struct virtio_net_hdr { uint8_t flags; - -#define VIRTIO_NET_HDR_GSO_NONE 0 /* Not a GSO frame */ + uint8_t gso_type; - uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */ uint16_t gso_size; /* Bytes to append to hdr_len per frame */ - uint16_t csum_start; /* Position to start checksumming from */ - uint16_t csum_offset; /* Offset after that to place checksum */ + uint16_t csum_start; /* Position to start checksumming from */ + uint16_t csum_offset; /* Offset after that to place checksum */ }__attribute__((packed)); - +struct virtio_net_hdr_mrg_rxbuf { + struct virtio_net_hdr hdr; + uint16_t num_buffers; /* Number of merged rx buffers */ +}; -#define QUEUE_SIZE 256 + +#define TX_QUEUE_SIZE 256 +#define RX_QUEUE_SIZE 4096 #define CTRL_QUEUE_SIZE 64 -#define ETH_ALEN 6 +#define VIRTIO_NET_F_MRG_RXBUF 15 /* Host can merge receive buffers. */ +#define VIRTIO_NET_F_MAC 5 /* Host has given MAC address. */ +#define VIRTIO_NET_F_GSO 6 /* Host handles pkts w/ any GSO type */ +#define VIRTIO_NET_F_HOST_TSO4 11 /* Host can handle TSOv4 in. */ +#define VIRTIO_NET_F_HOST_UFO 14 /* Host can handle UFO in. */ + +/* Port to get virtio config */ +#define VIRTIO_NET_CONFIG 20 struct virtio_net_config { - uint8_t mac[ETH_ALEN]; - // See VIRTIO_NET_F_STATUS and VIRTIO_NET_S_* above + uint8_t mac[ETH_ALEN]; /* VIRTIO_NET_F_MAC */ uint16_t status; } __attribute__((packed)); struct virtio_dev_state { struct vm_device * pci_bus; struct list_head dev_list; - struct guest_info * vm; + struct v3_vm_info *vm; + + uint8_t mac[ETH_ALEN]; }; struct virtio_net_state { struct virtio_net_config net_cfg; struct virtio_config virtio_cfg; - //struct vm_device * pci_bus; - struct pci_device * pci_dev; + struct vm_device * dev; + struct pci_device * pci_dev; + int io_range_size; - struct virtio_queue rx_vq; //index 0, rvq in Linux virtio driver, handle packet to guest - struct virtio_queue tx_vq; //index 1, svq in Linux virtio driver, handle packet from guest - struct virtio_queue ctrl_vq; //index 2, ctrol info from guest + struct virtio_queue rx_vq; /* idx 0*/ + struct virtio_queue tx_vq; /* idx 1*/ + struct virtio_queue ctrl_vq; /* idx 2*/ + + struct v3_timer * timer; + + struct nic_statistics statistics; struct v3_dev_net_ops * net_ops; + v3_lock_t rx_lock, tx_lock; - void * backend_data; + uint8_t tx_notify, rx_notify; + uint32_t tx_pkts, rx_pkts; + uint64_t past_ms; + void * backend_data; struct virtio_dev_state * virtio_dev; - struct list_head dev_link; - - int io_range_size; - - void * private_data; }; -#if 1 -//Temporarly for debug -static void print_packet(uchar_t *pkt, int size) { - PrintDebug("Vnet: print_data_packet: size: %d\n", size); - v3_hexdump(pkt, size, NULL, 0); -} - -static int send(uint8_t * buf, uint32_t count, void * private_data, struct vm_device *dest_dev) -{ - PrintDebug("Virito NIC: In sending stub\n"); - print_packet(buf, count); - - return count; -} - -static int receive(uint8_t * buf, uint32_t count, void * private_data, struct vm_device *src_dev) +static int virtio_init_state(struct virtio_net_state * virtio) { - PrintDebug("Virito NIC: In receiveing stub\n"); + virtio->rx_vq.queue_size = RX_QUEUE_SIZE; + virtio->tx_vq.queue_size = TX_QUEUE_SIZE; + virtio->ctrl_vq.queue_size = CTRL_QUEUE_SIZE; - return 0; -} - - -#endif - -static int virtio_free(struct vm_device * dev) -{ - - return -1; -} - -static int virtio_reset(struct virtio_net_state * virtio) -{ virtio->rx_vq.ring_desc_addr = 0; virtio->rx_vq.ring_avail_addr = 0; virtio->rx_vq.ring_used_addr = 0; @@ -167,160 +137,67 @@ static int virtio_reset(struct virtio_net_state * virtio) virtio->ctrl_vq.pfn = 0; virtio->ctrl_vq.cur_avail_idx = 0; - virtio->virtio_cfg.host_features = 0; - //virtio->virtio_cfg.status = VIRTIO_NET_S_LINK_UP; virtio->virtio_cfg.pci_isr = 0; - virtio->private_data = NULL; + + virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC) | + (1 << VIRTIO_NET_F_HOST_UFO) | + (1 << VIRTIO_NET_F_HOST_TSO4); + + if ((v3_lock_init(&(virtio->rx_lock)) == -1) || + (v3_lock_init(&(virtio->tx_lock)) == -1)){ + PrintError("Virtio NIC: Failure to init locks for net_state\n"); + } return 0; } - -//sending guest's packet to network sink -static int pkt_write(struct virtio_net_state * virtio, struct vring_desc * buf_desc) +static int tx_one_pkt(struct guest_info * core, + struct virtio_net_state * virtio, + struct vring_desc * buf_desc) { - //struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data; uint8_t * buf = NULL; uint32_t len = buf_desc->length; - PrintDebug("Handling Virtio Net write\n"); - - if (guest_pa_to_host_va(virtio->virtio_dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) { - PrintError("Could not translate buffer address\n"); + if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) { + PrintDebug("Could not translate buffer address\n"); return -1; } - PrintDebug("Length=%d\n", buf_desc->length); - PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d, buf address: %p, send address: %p\n", buf_desc, - (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next, buf, virtio->net_ops->send); + if(virtio->net_ops->send(buf, len, virtio->backend_data) >= 0){ + virtio->statistics.tx_pkts ++; + virtio->statistics.tx_bytes += len; - if (virtio->net_ops->send(buf, len, (void *)virtio, NULL) == -1) { - return -1; + return 0; } - return 0; -} - - -static int build_receive_header(struct virtio_net_hdr * hdr, const void * buf, int raw) { - hdr->flags = 0; - - if (!raw) { - memcpy(hdr, buf, sizeof(struct virtio_net_hdr)); - } else { - memset(hdr, 0, sizeof(struct virtio_net_hdr)); - } + virtio->statistics.tx_dropped ++; - return 0; + return -1; } -//sending guest's packet to network sink -static int copy_data_to_desc(struct virtio_net_state * virtio_state, struct vring_desc * desc, uchar_t * buf, uint_t buf_len) +static int +copy_data_to_desc(struct guest_info * core, + struct virtio_net_state * virtio_state, + struct vring_desc * desc, + uchar_t * buf, + uint_t buf_len, + uint_t offset) { uint32_t len; uint8_t * desc_buf = NULL; - if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) { + if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) { PrintError("Could not translate buffer address\n"); return -1; } - - len = (desc->length < buf_len)?desc->length:buf_len; - - memcpy(desc_buf, buf, len); - - PrintDebug("Length=%d\n", len); + len = (desc->length < buf_len)?(desc->length - offset):buf_len; + memcpy(desc_buf+offset, buf, len); return len; } -//send data to guest -static int send_pkt_to_guest(struct vm_device * dev, uchar_t * buf, uint_t size, int raw, void * private_data) -{ - // TODO: This should not be like this - struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data; - struct virtio_queue * q = &(virtio->rx_vq); - struct virtio_net_hdr hdr; - uint32_t hdr_len = sizeof(struct virtio_net_hdr); - uint32_t data_len = size; - uint32_t offset = 0; - - PrintDebug("VIRTIO Handle RX: cur_index=%d (mod=%d), avail_index=%d\n", - q->cur_avail_idx, q->cur_avail_idx % q->queue_size, q->avail->index); - - - - if (!raw) { - data_len -= hdr_len; - } - - build_receive_header(&hdr, buf, 1); - - //queue is not set yet - if (q->ring_avail_addr == 0) { - PrintError("Queue is not set\n"); - return -1; - } - - - if (q->cur_avail_idx < q->avail->index) { - addr_t hdr_addr = 0; - uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; - uint16_t buf_idx = 0; - struct vring_desc * hdr_desc = NULL; - - PrintDebug("Descriptor index=%d\n", q->cur_avail_idx % q->queue_size); - - hdr_desc = &(q->desc[hdr_idx]); - - PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc, - (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next); - - if (guest_pa_to_host_va(virtio->virtio_dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) { - PrintError("Could not translate receive buffer address\n"); - return -1; - } - - //copy header to the header descriptor - memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr)); - - //copy data to the next descriptors - for (buf_idx = 0; offset < data_len; buf_idx = q->desc[hdr_idx].next) { - struct vring_desc * buf_desc = &(q->desc[buf_idx]); - uint32_t len = 0; - - buf_desc->flags = VIRTIO_NEXT_FLAG; - - len = copy_data_to_desc(virtio, buf_desc, buf + offset, data_len - offset); - - offset += len; - buf_desc->length = len; // TODO: do we need this? - } - - q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; - q->used->ring[q->used->index % q->queue_size].length = data_len; // What do we set this to???? - - q->used->index++; - q->cur_avail_idx++; - } - - if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line); - v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); - virtio->virtio_cfg.pci_isr = 0x1; - } - - return offset; -} - - -int virtio_send(struct vm_device * dev, uchar_t * buf, uint_t size) { - return send_pkt_to_guest(dev, buf, size, 1, NULL); -} - - static int get_desc_count(struct virtio_queue * q, int index) { struct vring_desc * tmp_desc = &(q->desc[index]); int cnt = 1; @@ -333,23 +210,49 @@ static int get_desc_count(struct virtio_queue * q, int index) { return cnt; } +static inline void enable_cb(struct virtio_queue *queue){ + queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG; +} -static int handle_ctrl(struct virtio_net_state * dev) { - return 0; +static inline void disable_cb(struct virtio_queue *queue) { + queue->used->flags |= VRING_NO_NOTIFY_FLAG; +} + + +/* interrupt the guest, so the guest core get EXIT to Palacios */ +static inline void notify_guest(struct virtio_net_state * virtio){ + v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0); } -//get packet from guest -static int handle_pkt_tx(struct virtio_net_state * virtio_state) + +/* guest free some pkts for rx queue */ +static int handle_rx_queue_kick(struct guest_info * core, + struct virtio_net_state * virtio) { - //struct virtio_net_state *virtio = (struct virtio_net_state *)dev->private_data; - struct virtio_queue * q = &(virtio_state->tx_vq); + return 0; +} + + +static int handle_ctrl(struct guest_info * core, + struct virtio_net_state * virtio) { + + return 0; +} - PrintDebug("VIRTIO NIC pkt_tx: cur_index=%d (mod=%d), avail_index=%d\n", - q->cur_avail_idx, q->cur_avail_idx % q->queue_size, q->avail->index); +static int handle_pkt_tx(struct guest_info * core, + struct virtio_net_state * virtio_state) +{ + struct virtio_queue *q = &(virtio_state->tx_vq); + struct virtio_net_hdr *hdr = NULL; + int txed = 0; + unsigned long flags; - struct virtio_net_hdr * hdr = NULL; + if (!q->ring_avail_addr) { + return -1; + } - while (q->cur_avail_idx < q->avail->index) { + flags = v3_lock_irqsave(virtio_state->tx_lock); + while (q->cur_avail_idx != q->avail->index) { struct vring_desc * hdr_desc = NULL; addr_t hdr_addr = 0; uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; @@ -357,34 +260,26 @@ static int handle_pkt_tx(struct virtio_net_state * virtio_state) uint32_t req_len = 0; int i = 0; - PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % q->queue_size); - hdr_desc = &(q->desc[desc_idx]); - - PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc, - (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next); - - if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) { + if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) { PrintError("Could not translate block header address\n"); - return -1; + goto exit_error; } - //memcpy(&hdr, (void *)hdr_addr, sizeof(struct virtio_net_hdr)); - hdr = (struct virtio_net_hdr*)hdr_addr; - - PrintDebug("NIC Op Hdr (ptr=%p) header len =%x\n", (void *)hdr_addr, (int)hdr->hdr_len); - + hdr = (struct virtio_net_hdr *)hdr_addr; desc_idx = hdr_desc->next; - - for (i = 0; i < desc_cnt - 1; i++) { - struct vring_desc * buf_desc = &(q->desc[desc_idx]); - PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc, - (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next); + if(desc_cnt > 2){ + PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt); + goto exit_error; + } - if (pkt_write(virtio_state, buf_desc) == -1) { + /* here we assumed that one ethernet pkt is not splitted into multiple virtio buffer */ + for (i = 0; i < desc_cnt - 1; i++) { + struct vring_desc * buf_desc = &(q->desc[desc_idx]); + if (tx_one_pkt(core, virtio_state, buf_desc) == -1) { PrintError("Error handling nic operation\n"); - return -1; + goto exit_error; } req_len += buf_desc->length; @@ -392,23 +287,36 @@ static int handle_pkt_tx(struct virtio_net_state * virtio_state) } q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; - q->used->ring[q->used->index % q->queue_size].length = req_len; // What do we set this to???? + q->used->ring[q->used->index % q->queue_size].length = req_len; /* What do we set this to???? */ + q->used->index ++; + + q->cur_avail_idx ++; - q->used->index++; - q->cur_avail_idx++; + txed ++; } - if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - PrintDebug("Raising IRQ %d\n", virtio_state->pci_dev->config_header.intr_line); + v3_unlock_irqrestore(virtio_state->tx_lock, flags); + + if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev); virtio_state->virtio_cfg.pci_isr = 0x1; + + virtio_state->statistics.interrupts ++; } return 0; + +exit_error: + + v3_unlock_irqrestore(virtio_state->tx_lock, flags); + return -1; } -static int virtio_setup_queue(struct virtio_net_state * virtio_state, struct virtio_queue * queue, addr_t pfn, addr_t page_addr) { +static int virtio_setup_queue(struct guest_info *core, + struct virtio_net_state * virtio_state, + struct virtio_queue * queue, + addr_t pfn, addr_t page_addr) { queue->pfn = pfn; queue->ring_desc_addr = page_addr; @@ -416,23 +324,20 @@ static int virtio_setup_queue(struct virtio_net_state * virtio_state, struct vir queue->ring_used_addr = ((queue->ring_avail_addr) + (sizeof(struct vring_avail)) + (queue->queue_size * sizeof(uint16_t))); - + // round up to next page boundary. queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff; - - if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) { + if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) { PrintError("Could not translate ring descriptor address\n"); return -1; } - - if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) { + if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) { PrintError("Could not translate ring available address\n"); return -1; } - - if (guest_pa_to_host_va(virtio_state->virtio_dev->vm, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) { + if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) { PrintError("Could not translate ring used address\n"); return -1; } @@ -448,105 +353,92 @@ static int virtio_setup_queue(struct virtio_net_state * virtio_state, struct vir return 0; } - - -static int virtio_io_write(uint16_t port, void * src, uint_t length, void * private_data) +static int virtio_io_write(struct guest_info *core, + uint16_t port, void * src, + uint_t length, void * private_data) { - //struct vm_device * dev = (struct vm_device *)private_data; struct virtio_net_state * virtio = (struct virtio_net_state *)private_data; int port_idx = port % virtio->io_range_size; - - PrintDebug("VIRTIO NIC Write for port %d (index=%d) len=%d, value=%x\n", + PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data, port, port_idx, length, *(uint32_t *)src); - switch (port_idx) { case GUEST_FEATURES_PORT: if (length != 4) { PrintError("Illegal write length for guest features\n"); return -1; - } - + } virtio->virtio_cfg.guest_features = *(uint32_t *)src; - PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features); - break; + case VRING_PG_NUM_PORT: - if (length != 4) { PrintError("Illegal write length for page frame number\n"); return -1; } - addr_t pfn = *(uint32_t *)src; addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT); uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector; - switch (queue_idx) { case 0: - virtio_setup_queue(virtio, &virtio->rx_vq, pfn, page_addr); + virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr); break; case 1: - virtio_setup_queue(virtio, &virtio->tx_vq, pfn, page_addr); + virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr); break; case 2: - virtio_setup_queue(virtio, &virtio->ctrl_vq, pfn, page_addr); - break; - + virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr); + break; default: break; } - break; + case VRING_Q_SEL_PORT: virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src; - if (virtio->virtio_cfg.vring_queue_selector > 2) { - PrintError("Virtio NIC device only uses 3 queue, selected %d\n", + PrintError("Virtio NIC: wrong queue idx: %d\n", virtio->virtio_cfg.vring_queue_selector); return -1; } - break; + case VRING_Q_NOTIFY_PORT: { - uint16_t queue_idx = *(uint16_t *)src; - - PrintDebug("Handling Kick\n"); - + uint16_t queue_idx = *(uint16_t *)src; if (queue_idx == 0){ - PrintError("receive queue notification\n"); + if(handle_rx_queue_kick(core, virtio) == -1){ + PrintError("Could not handle Virtio NIC rx kick\n"); + return -1; + } } else if (queue_idx == 1){ - if (handle_pkt_tx(virtio) == -1) { - PrintError("Could not handle NIC Notification\n"); + if (handle_pkt_tx(core, virtio) == -1) { + PrintError("Could not handle Virtio NIC tx kick\n"); return -1; } } else if (queue_idx == 2){ - if (handle_ctrl(virtio) == -1) { - PrintError("Could not handle NIC Notification\n"); + if (handle_ctrl(core, virtio) == -1) { + PrintError("Could not handle Virtio NIC ctrl kick\n"); return -1; } } else { - PrintError("Virtio NIC device only uses 3 queue, selected %d\n", - queue_idx); - } - + PrintError("Wrong queue index %d\n", queue_idx); + } break; } + case VIRTIO_STATUS_PORT: virtio->virtio_cfg.status = *(uint8_t *)src; - if (virtio->virtio_cfg.status == 0) { - PrintDebug("Resetting device\n"); - virtio_reset(virtio); + virtio_init_state(virtio); } - break; case VIRTIO_ISR_PORT: virtio->virtio_cfg.pci_isr = *(uint8_t *)src; break; + default: return -1; break; @@ -555,15 +447,15 @@ static int virtio_io_write(uint16_t port, void * src, uint_t length, void * priv return length; } - -static int virtio_io_read(uint16_t port, void * dst, uint_t length, void * private_data) +static int virtio_io_read(struct guest_info *core, + uint16_t port, void * dst, + uint_t length, void * private_data) { - //struct vm_device * dev = (struct vm_device *)private_data; struct virtio_net_state * virtio = (struct virtio_net_state *)private_data; int port_idx = port % virtio->io_range_size; uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector; - PrintDebug("VIRTIO NIC Read for port %d (index =%d), length=%d", + PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", private_data, port, port_idx, length); switch (port_idx) { @@ -572,19 +464,14 @@ static int virtio_io_read(uint16_t port, void * dst, uint_t length, void * priva PrintError("Illegal read length for host features\n"); return -1; } - *(uint32_t *)dst = virtio->virtio_cfg.host_features; - - PrintDebug("value=0x%x\n", *(uint32_t *)dst); - break; - case VRING_PG_NUM_PORT: + case VRING_PG_NUM_PORT: if (length != 4) { PrintError("Illegal read length for page frame number\n"); return -1; } - switch (queue_idx) { case 0: *(uint32_t *)dst = virtio->rx_vq.pfn; @@ -598,17 +485,13 @@ static int virtio_io_read(uint16_t port, void * dst, uint_t length, void * priva default: break; } - - PrintDebug(", value=0x%x\n", *(uint32_t *)dst); - break; - case VRING_SIZE_PORT: + case VRING_SIZE_PORT: if (length != 2) { PrintError("Illegal read length for vring size\n"); return -1; } - switch (queue_idx) { case 0: *(uint16_t *)dst = virtio->rx_vq.queue_size; @@ -622,33 +505,28 @@ static int virtio_io_read(uint16_t port, void * dst, uint_t length, void * priva default: break; } - - PrintDebug("queue index: %d, value=0x%x\n", (int)queue_idx, *(uint16_t *)dst); - break; - case VIRTIO_STATUS_PORT: + case VIRTIO_STATUS_PORT: if (length != 1) { PrintError("Illegal read length for status\n"); return -1; } - *(uint8_t *)dst = virtio->virtio_cfg.status; - - PrintDebug(", value=0x%x\n", *(uint8_t *)dst); break; - + case VIRTIO_ISR_PORT: *(uint8_t *)dst = virtio->virtio_cfg.pci_isr; virtio->virtio_cfg.pci_isr = 0; v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); + break; - PrintDebug(", value=0x%x\n", *(uint8_t *)dst); - + case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN: + *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG]; break; default: - PrintError("Read of Unhandled Virtio Read\n"); + PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx); return -1; } @@ -656,15 +534,132 @@ static int virtio_io_read(uint16_t port, void * dst, uint_t length, void * priva } +/* receiving raw ethernet pkt from backend */ +static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { + struct virtio_net_state * virtio = (struct virtio_net_state *)private_data; + struct virtio_queue * q = &(virtio->rx_vq); + struct virtio_net_hdr_mrg_rxbuf hdr; + uint32_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); + uint32_t data_len; + uint32_t offset = 0; + unsigned long flags; + +#ifdef CONFIG_DEBUG_VIRTIO_NET + PrintDebug("Virtio-NIC: virtio_rx: size: %d\n", size); + v3_hexdump(buf, size, NULL, 0); +#endif + + flags = v3_lock_irqsave(virtio->rx_lock); + + data_len = size; + memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf)); + + if (q->ring_avail_addr == 0) { + PrintDebug("Queue is not set\n"); + goto err_exit; + } + + if (q->cur_avail_idx != q->avail->index){ + addr_t hdr_addr = 0; + uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; + uint16_t buf_idx = 0; + struct vring_desc * hdr_desc = NULL; + + hdr_desc = &(q->desc[hdr_idx]); + if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) { + PrintDebug("Could not translate receive buffer address\n"); + goto err_exit; + } + hdr.num_buffers = 1; + memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr_mrg_rxbuf)); + if (offset >= data_len) { + hdr_desc->flags &= ~VIRTIO_NEXT_FLAG; + } + + struct vring_desc * buf_desc = NULL; + for (buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) { + uint32_t len = 0; + buf_desc = &(q->desc[buf_idx]); + + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset, 0); + offset += len; + if (offset < data_len) { + buf_desc->flags = VIRTIO_NEXT_FLAG; + } + buf_desc->length = len; + } + buf_desc->flags &= ~VIRTIO_NEXT_FLAG; + + q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; + q->used->ring[q->used->index % q->queue_size].length = data_len + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */ + q->used->index++; + q->cur_avail_idx++; + + virtio->statistics.rx_pkts ++; + virtio->statistics.rx_bytes += size; + } else { + virtio->statistics.rx_dropped ++; + + goto err_exit; + } + + if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { + PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line); + + v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); + virtio->virtio_cfg.pci_isr = 0x1; + virtio->statistics.interrupts ++; + } + + v3_unlock_irqrestore(virtio->rx_lock, flags); + + /* notify guest if guest is running */ + if(virtio->rx_notify == 1){ + v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0); + } + + return 0; + +err_exit: + + v3_unlock_irqrestore(virtio->rx_lock, flags); + + return -1; +} + +static int virtio_free(struct virtio_dev_state * virtio) { + struct virtio_net_state * backend = NULL; + struct virtio_net_state * tmp = NULL; + + + list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) { + + // unregister from PCI + + list_del(&(backend->dev_link)); + V3_Free(backend); + } + + V3_Free(virtio); + return 0; +} + + static struct v3_device_ops dev_ops = { - .free = virtio_free, - .reset = NULL, - .start = NULL, - .stop = NULL, + .free = (int (*)(void *))virtio_free, }; -static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_state * net_state) +static void virtio_nic_poll(struct v3_vm_info * vm, int budget, void * data){ + struct virtio_net_state * virtio = (struct virtio_net_state *)data; + + if(virtio->tx_notify == 0){ + handle_pkt_tx(&(vm->cores[0]), virtio); + } +} + +static int register_dev(struct virtio_dev_state * virtio, + struct virtio_net_state * net_state) { struct pci_device * pci_dev = NULL; struct v3_pci_bar bars[6]; @@ -674,19 +669,18 @@ static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_stat // This gets the number of ports, rounded up to a power of 2 net_state->io_range_size = 1; // must be a power of 2 - while (tmp_ports > 0) { tmp_ports >>= 1; net_state->io_range_size <<= 1; } - // this is to account for any low order bits being set in num_ports - // if there are none, then num_ports was already a power of 2 so we shift right to reset it + /* this is to account for any low order bits being set in num_ports + * if there are none, then num_ports was already a power of 2 so we shift right to reset it + */ if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) { net_state->io_range_size >>= 1; } - for (i = 0; i < 6; i++) { bars[i].type = PCI_BAR_NONE; } @@ -696,22 +690,21 @@ static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_stat bars[0].type = PCI_BAR_IO; bars[0].default_base_port = -1; bars[0].num_ports = net_state->io_range_size; - bars[0].io_read = virtio_io_read; bars[0].io_write = virtio_io_write; bars[0].private_data = net_state; pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE, - 0, PCI_AUTO_DEV_NUM, 0, + 0, 4/*PCI_AUTO_DEV_NUM*/, 0, "LNX_VIRTIO_NIC", bars, NULL, NULL, NULL, net_state); if (!pci_dev) { - PrintError("Could not register PCI Device\n"); + PrintError("Virtio NIC: Could not register PCI Device\n"); return -1; } - PrintDebug("Virtio-NIC registered to PCI bus\n"); + PrintDebug("Virtio NIC: registered to PCI bus\n"); pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID; pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID; @@ -719,34 +712,80 @@ static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_stat pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID; pci_dev->config_header.class = PCI_CLASS_NETWORK; - pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER; - - pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;; - + pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER; + pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID; pci_dev->config_header.intr_pin = 1; - pci_dev->config_header.max_latency = 1; // ?? (qemu does it...) - net_state->pci_dev = pci_dev; - //net_state->pci_bus = pci_bus; - - net_state->virtio_cfg.host_features = 0; //no features support now - - net_state->rx_vq.queue_size = QUEUE_SIZE; - net_state->tx_vq.queue_size = QUEUE_SIZE; - net_state->ctrl_vq.queue_size = CTRL_QUEUE_SIZE; - net_state->virtio_dev = virtio; - virtio_reset(net_state); + memcpy(net_state->net_cfg.mac, virtio->mac, 6); + + virtio_init_state(net_state); + + /* Add backend to list of devices */ + list_add(&(net_state->dev_link), &(virtio->dev_list)); return 0; } +#define RATE_UPPER_THRESHOLD 10 /* 10000 pkts per second, around 100Mbits */ +#define RATE_LOWER_THRESHOLD 1 +#define PROFILE_PERIOD 50 /*50ms*/ + +/* Timer Functions */ +static void virtio_nic_timer(struct guest_info * core, + uint64_t cpu_cycles, uint64_t cpu_freq, + void * priv_data) { + struct virtio_net_state * net_state = (struct virtio_net_state *)priv_data; + uint64_t period_ms; + + period_ms = cpu_cycles/cpu_freq; + net_state->past_ms += period_ms; + + if(net_state->past_ms > PROFILE_PERIOD){ + uint32_t tx_rate, rx_rate; + + tx_rate = (net_state->statistics.tx_pkts - net_state->tx_pkts)/net_state->past_ms; /* pkts/per ms */ + rx_rate = (net_state->statistics.rx_pkts - net_state->rx_pkts)/net_state->past_ms; + + net_state->tx_pkts = net_state->statistics.tx_pkts; + net_state->rx_pkts = net_state->statistics.rx_pkts; + + if(tx_rate > RATE_UPPER_THRESHOLD && net_state->tx_notify == 1){ + V3_Print("Virtio NIC: Switch TX to VMM driven mode\n"); + disable_cb(&(net_state->tx_vq)); + net_state->tx_notify = 0; + } + + if(tx_rate < RATE_LOWER_THRESHOLD && net_state->tx_notify == 0){ + V3_Print("Virtio NIC: Switch TX to Guest driven mode\n"); + enable_cb(&(net_state->tx_vq)); + net_state->tx_notify = 1; + } + + if(rx_rate > RATE_UPPER_THRESHOLD && net_state->rx_notify == 1){ + PrintDebug("Virtio NIC: Switch RX to VMM None notify mode\n"); + net_state->rx_notify = 0; + } + + if(rx_rate < RATE_LOWER_THRESHOLD && net_state->rx_notify == 0){ + PrintDebug("Virtio NIC: Switch RX to VMM notify mode\n"); + net_state->rx_notify = 1; + } + + net_state->past_ms = 0; + } +} + + +static struct v3_timer_ops timer_ops = { + .update_timer = virtio_nic_timer, +}; -static int connect_fn(struct guest_info * info, +static int connect_fn(struct v3_vm_info * info, void * frontend_data, struct v3_dev_net_ops * ops, v3_cfg_tree_t * cfg, @@ -755,25 +794,34 @@ static int connect_fn(struct guest_info * info, struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state)); memset(net_state, 0, sizeof(struct virtio_net_state)); - register_dev(virtio, net_state); net_state->net_ops = ops; net_state->backend_data = private_data; + net_state->virtio_dev = virtio; + net_state->tx_notify = 1; + net_state->rx_notify = 1; + + net_state->timer = v3_add_timer(&(info->cores[0]),&timer_ops,net_state); + + ops->recv = virtio_rx; + ops->poll = virtio_nic_poll; + ops->frontend_data = net_state; + memcpy(ops->fnt_mac, virtio->mac, ETH_ALEN); return 0; } - -static int virtio_init(struct guest_info * vm, v3_cfg_tree_t * cfg) { +static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) { struct vm_device * pci_bus = v3_find_dev(vm, v3_cfg_val(cfg, "bus")); struct virtio_dev_state * virtio_state = NULL; - char * name = v3_cfg_val(cfg, "name"); - - PrintDebug("Initializing VIRTIO Network device\n"); + char * dev_id = v3_cfg_val(cfg, "ID"); + char macstr[128]; + char * str = v3_cfg_val(cfg, "mac"); + memcpy(macstr, str, strlen(str)); if (pci_bus == NULL) { - PrintError("VirtIO devices require a PCI Bus"); + PrintError("Virtio NIC: VirtIO devices require a PCI Bus"); return -1; } @@ -784,37 +832,33 @@ static int virtio_init(struct guest_info * vm, v3_cfg_tree_t * cfg) { virtio_state->pci_bus = pci_bus; virtio_state->vm = vm; - struct vm_device * dev = v3_allocate_device(name, &dev_ops, virtio_state); - if (v3_attach_device(vm, dev) == -1) { - PrintError("Could not attach device %s\n", name); - return -1; + if (macstr != NULL && !str2mac(macstr, virtio_state->mac)) { + PrintDebug("Virtio NIC: Mac specified %s\n", macstr); + PrintDebug("MAC: %x:%x:%x:%x:%x:%x\n", virtio_state->mac[0], + virtio_state->mac[1], + virtio_state->mac[2], + virtio_state->mac[3], + virtio_state->mac[4], + virtio_state->mac[5]); + }else { + PrintDebug("Virtio NIC: MAC not specified\n"); + random_ethaddr(virtio_state->mac); } - if (v3_dev_add_net_frontend(vm, name, connect_fn, (void *)virtio_state) == -1) { - PrintError("Could not register %s as net frontend\n", name); + struct vm_device * dev = v3_add_device(vm, dev_id, &dev_ops, virtio_state); + + if (dev == NULL) { + PrintError("Virtio NIC: Could not attach device %s\n", dev_id); + V3_Free(virtio_state); return -1; } - - //for temporary testing, add a backend - #if 1 - - struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state)); - memset(net_state, 0, sizeof(struct virtio_net_state)); - - net_state->net_ops = (struct v3_dev_net_ops *)V3_Malloc(sizeof(struct v3_dev_net_ops)); - - net_state->net_ops->send = &send; - net_state->net_ops->receive = &receive; - - register_dev(virtio_state, net_state); + if (v3_dev_add_net_frontend(vm, dev_id, connect_fn, (void *)virtio_state) == -1) { + PrintError("Virtio NIC: Could not register %s as net frontend\n", dev_id); + v3_remove_device(dev); + return -1; + } - PrintDebug("Virtio NIC After register Device: queue size: %d, %d\n", - net_state->rx_vq.queue_size, net_state->tx_vq.queue_size); - - - #endif - return 0; }