X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fdevices%2Flnx_virtio_nic.c;h=400cb0c0aad3ab77ecc5e3e0d8762caad9267d74;hb=9a32111c4074aafd55cd9590a24bd5c751a6fe61;hp=56bf74e8d208b17c06d4bf62292b92084168f7ba;hpb=e49df33562b190c348b3a117044136c704442c6b;p=palacios.git diff --git a/palacios/src/devices/lnx_virtio_nic.c b/palacios/src/devices/lnx_virtio_nic.c index 56bf74e..400cb0c 100644 --- a/palacios/src/devices/lnx_virtio_nic.c +++ b/palacios/src/devices/lnx_virtio_nic.c @@ -1,3 +1,4 @@ + /* * This file is part of the Palacios Virtual Machine Monitor developed * by the V3VEE Project with funding from the United States National @@ -25,59 +26,80 @@ #include #include #include -#include +#include #include #include #include #include +#include -#ifndef CONFIG_DEBUG_VIRTIO_NET +#ifndef V3_CONFIG_DEBUG_VIRTIO_NET #undef PrintDebug #define PrintDebug(fmt, args...) #endif -#define VIRTIO_NET_S_LINK_UP 1 /* Link is up */ +#ifndef V3_CONFIG_VNET +static int net_debug = 0; +#endif + +#define TX_QUEUE_SIZE 4096 +#define RX_QUEUE_SIZE 4096 +#define CTRL_QUEUE_SIZE 64 + +/* The feature bitmap for virtio nic + * from Linux */ +#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 */ + +/* Port to get virtio config */ +#define VIRTIO_NET_CONFIG 20 + #define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10)) +/* for gso_type in virtio_net_hdr */ +#define VIRTIO_NET_HDR_GSO_NONE 0 +#define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */ +#define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */ +#define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */ +#define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */ -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 */ +/* for flags in virtio_net_hdr */ +#define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */ + + +/* First element of the scatter-gather list, used with GSO or CSUM features */ +struct virtio_net_hdr +{ + uint8_t flags; + 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 */ }__attribute__((packed)); -/* This is the version of the header to use when the MRG_RXBUF +/* The header to use when the MRG_RXBUF * feature has been negotiated. */ struct virtio_net_hdr_mrg_rxbuf { - struct virtio_net_hdr hdr; - uint16_t num_buffers; /* Number of merged rx buffers */ + struct virtio_net_hdr hdr; + uint16_t num_buffers; /* Number of merged rx buffers */ }; - -#define TX_QUEUE_SIZE 64 -#define RX_QUEUE_SIZE 1024 -#define CTRL_QUEUE_SIZE 64 - -#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. */ - -/* this is not how virtio supposed to be, - * we may need a separately implemented virtio_pci - * In order to make guest to get virtio MAC from host - * I added it here -- Lei - */ - #define VIRTIO_NET_CONFIG 20 - - struct virtio_net_config { uint8_t mac[ETH_ALEN]; /* VIRTIO_NET_F_MAC */ @@ -89,44 +111,41 @@ struct virtio_dev_state { struct list_head dev_list; struct v3_vm_info *vm; - char mac[ETH_ALEN]; + uint8_t mac[ETH_ALEN]; }; struct virtio_net_state { struct virtio_net_config net_cfg; struct virtio_config virtio_cfg; + struct v3_vm_info * vm; struct vm_device * dev; struct pci_device * pci_dev; int io_range_size; + + uint16_t status; - struct virtio_queue rx_vq; /* idx 0, pkts to guest */ - struct virtio_queue tx_vq; /* idx 1, pkts from guest */ - struct virtio_queue ctrl_vq; /* idx 2 */ + struct virtio_queue rx_vq; /* idx 0*/ + struct virtio_queue tx_vq; /* idx 1*/ + struct virtio_queue ctrl_vq; /* idx 2*/ - int buffed_rx; - int tx_disabled; /* stop TX pkts from guest */ + uint8_t mergeable_rx_bufs; - uint64_t pkt_sent, pkt_recv, pkt_drop; + struct v3_timer * timer; + struct nic_statistics stats; struct v3_dev_net_ops * net_ops; v3_lock_t rx_lock, tx_lock; + uint8_t tx_notify, rx_notify; + uint32_t tx_pkts, rx_pkts; + uint64_t past_us; + void * backend_data; struct virtio_dev_state * virtio_dev; struct list_head dev_link; }; -/* virtio nic error type */ -#define ERR_VIRTIO_OTHER 1 -#define ERR_VIRTIO_RXQ_FULL 2 -#define ERR_VIRTIO_RXQ_NOSET 3 -#define ERR_VIRTIO_TXQ_NOSET 4 -#define ERR_VIRTIO_TXQ_FULL 5 -#define ERR_VIRTIO_TXQ_DISABLED 6 - - - static int virtio_init_state(struct virtio_net_state * virtio) { @@ -153,60 +172,73 @@ static int virtio_init_state(struct virtio_net_state * virtio) virtio->ctrl_vq.cur_avail_idx = 0; virtio->virtio_cfg.pci_isr = 0; + + virtio->mergeable_rx_bufs = 1; virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC); + if(virtio->mergeable_rx_bufs) { + virtio->virtio_cfg.host_features |= (1 << VIRTIO_NET_F_MRG_RXBUF); + } 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"); } - virtio->pkt_sent = virtio->pkt_recv = virtio->pkt_drop = 0; - virtio->buffed_rx = 0; - return 0; } -static int -pkt_tx(struct guest_info * core, - 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) { uint8_t * buf = NULL; uint32_t len = buf_desc->length; if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) { - PrintError("Could not translate buffer address\n"); - return -ERR_VIRTIO_OTHER; + PrintDebug("Could not translate buffer address\n"); + return -1; + } + + V3_Net_Print(2, "Virtio-NIC: virtio_tx: size: %d\n", len); + if(net_debug >= 4){ + v3_hexdump(buf, len, NULL, 0); } - return virtio->net_ops->send(buf, len, virtio->backend_data); + if(virtio->net_ops->send(buf, len, virtio->backend_data) < 0){ + virtio->stats.tx_dropped ++; + return -1; + } + + virtio->stats.tx_pkts ++; + virtio->stats.tx_bytes += len; + + return 0; } -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) -{ +/*copy data into ring buffer */ +static inline 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 dst_offset){ uint32_t len; uint8_t * desc_buf = NULL; - + if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) { - PrintError("Could not translate buffer address\n"); + PrintDebug("Could not translate buffer address\n"); return -1; } - len = (desc->length < buf_len)?(desc->length - offset):buf_len; - memcpy(desc_buf+offset, buf, len); + len = (desc->length < (buf_len+dst_offset))?(desc->length - dst_offset):buf_len; + memcpy(desc_buf + dst_offset, buf, len); return len; } -static int get_desc_count(struct virtio_queue * q, int index) { +static inline int get_desc_count(struct virtio_queue * q, int index) { struct vring_desc * tmp_desc = &(q->desc[index]); int cnt = 1; @@ -219,69 +251,41 @@ static int get_desc_count(struct virtio_queue * q, int index) { } static inline void enable_cb(struct virtio_queue *queue){ - queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG; + if(queue->used){ + queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG; + } } 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 - * this happens when there are either incoming pkts for the guest - * or the guest can start TX pkts again */ -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); -} - - -/* guest free some pkts from rx queue */ -static int handle_rx_kick(struct guest_info *core, - struct virtio_net_state * virtio) -{ - unsigned long flags; - - flags = v3_lock_irqsave(virtio->rx_lock); - - if(virtio->net_ops->start_rx != NULL){ - virtio->net_ops->start_rx(virtio->backend_data); + if(queue->used){ + queue->used->flags |= VRING_NO_NOTIFY_FLAG; } - //disable_cb(&virtio->rx_vq); - - v3_unlock_irqrestore(virtio->rx_lock, flags); - - return 0; -} - - -static int handle_ctrl(struct guest_info *core, - struct virtio_net_state * virtio) { - - return 0; } -static int handle_pkt_tx(struct guest_info *core, - struct virtio_net_state * virtio_state) +static int handle_pkt_tx(struct guest_info * core, + struct virtio_net_state * virtio_state, + int quote) { - struct virtio_queue * q = &(virtio_state->tx_vq); - struct virtio_net_hdr * hdr = NULL; - int recved = 0; + struct virtio_queue *q = &(virtio_state->tx_vq); + int txed = 0, left = 0; unsigned long flags; - if (!q->ring_avail_addr) - return -ERR_VIRTIO_TXQ_NOSET; - - if(virtio_state->tx_disabled) - return -ERR_VIRTIO_TXQ_DISABLED; + if (!q->ring_avail_addr) { + return -1; + } flags = v3_lock_irqsave(virtio_state->tx_lock); while (q->cur_avail_idx != q->avail->index) { + struct virtio_net_hdr_mrg_rxbuf * hdr = NULL; 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]; int desc_cnt = get_desc_count(q, desc_idx); - uint32_t req_len = 0; - int i = 0; + + if(desc_cnt != 2){ + PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt); + goto exit_error; + } hdr_desc = &(q->desc[desc_idx]); if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) { @@ -289,51 +293,54 @@ static int handle_pkt_tx(struct guest_info *core, goto exit_error; } - hdr = (struct virtio_net_hdr*)hdr_addr; + hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr; desc_idx = hdr_desc->next; - if(desc_cnt > 2){ - PrintError("VNIC: merged rx buffer not supported\n"); - goto exit_error; - } - - /* 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 (pkt_tx(core, virtio_state, buf_desc) == -1) { - PrintError("Error handling nic operation\n"); - goto exit_error; - } + V3_Net_Print(2, "Virtio NIC: TX hdr count : %d\n", hdr->num_buffers); - req_len += buf_desc->length; - desc_idx = buf_desc->next; + /* here we assumed that one ethernet pkt is not splitted into multiple buffer */ + struct vring_desc * buf_desc = &(q->desc[desc_idx]); + if (tx_one_pkt(core, virtio_state, buf_desc) == -1) { + PrintError("Virtio NIC: Fails to send packet\n"); } - virtio_state->pkt_sent ++; - recved ++; - - 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->index ++; + if(buf_desc->next & VIRTIO_NEXT_FLAG){ + PrintError("Virtio NIC: TX more buffer need to read\n"); + } + + 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 = + buf_desc->length; /* What do we set this to???? */ + q->used->index ++; q->cur_avail_idx ++; + + if(++txed >= quote && quote > 0){ + left = (q->cur_avail_idx != q->avail->index); + break; + } } - + v3_unlock_irqrestore(virtio_state->tx_lock, flags); - - if(!recved) - return 0; - - if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev); + + 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->stats.rx_interrupts ++; } - return 0; + if(txed > 0) { + V3_Net_Print(2, "Virtio Handle TX: txed pkts: %d, left %d\n", txed, left); + } -exit_error: - + return left; + + exit_error: + v3_unlock_irqrestore(virtio_state->tx_lock, flags); - return -ERR_VIRTIO_OTHER; + return -1; } @@ -384,8 +391,9 @@ static int virtio_io_write(struct guest_info *core, struct virtio_net_state * virtio = (struct virtio_net_state *)private_data; int port_idx = port % virtio->io_range_size; - PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data, - port, port_idx, length, *(uint32_t *)src); + 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: @@ -406,15 +414,23 @@ static int virtio_io_write(struct guest_info *core, uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector; switch (queue_idx) { case 0: - virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr); - //disable_cb(&virtio->rx_vq); + virtio_setup_queue(core, virtio, + &virtio->rx_vq, + pfn, page_addr); break; case 1: - virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr); - //disable_cb(&virtio->tx_vq); + virtio_setup_queue(core, virtio, + &virtio->tx_vq, + pfn, page_addr); + if(virtio->tx_notify == 0){ + disable_cb(&virtio->tx_vq); + } + virtio->status = 1; break; case 2: - virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr); + virtio_setup_queue(core, virtio, + &virtio->ctrl_vq, + pfn, page_addr); break; default: break; @@ -434,19 +450,18 @@ static int virtio_io_write(struct guest_info *core, { uint16_t queue_idx = *(uint16_t *)src; if (queue_idx == 0){ - handle_rx_kick(core, virtio); + /* receive queue refill */ + virtio->stats.tx_interrupts ++; } else if (queue_idx == 1){ - if (handle_pkt_tx(core, virtio) == -1) { - PrintError("Could not handle NIC Notification\n"); + if (handle_pkt_tx(core, virtio, 0) < 0) { + PrintError("Virtio NIC: Error to handle packet TX\n"); return -1; } + virtio->stats.tx_interrupts ++; } else if (queue_idx == 2){ - if (handle_ctrl(core, virtio) == -1) { - PrintError("Could not handle NIC Notification\n"); - return -1; - } + /* ctrl */ } else { - PrintError("Wrong queue index %d\n", queue_idx); + PrintError("Virtio NIC: Wrong queue index %d\n", queue_idx); } break; } @@ -454,7 +469,6 @@ static int virtio_io_write(struct guest_info *core, case VIRTIO_STATUS_PORT: virtio->virtio_cfg.status = *(uint8_t *)src; if (virtio->virtio_cfg.status == 0) { - PrintDebug("Resetting device\n"); virtio_init_state(virtio); } break; @@ -479,21 +493,21 @@ static int virtio_io_read(struct guest_info *core, int port_idx = port % virtio->io_range_size; uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector; - PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", private_data, - port, port_idx, length); + PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", + private_data, port, port_idx, length); switch (port_idx) { case HOST_FEATURES_PORT: if (length != 4) { - PrintError("Illegal read length for host features\n"); - return -1; + PrintError("Virtio NIC: Illegal read length for host features\n"); + //return -1; } *(uint32_t *)dst = virtio->virtio_cfg.host_features; break; case VRING_PG_NUM_PORT: if (length != 4) { - PrintError("Illegal read length for page frame number\n"); + PrintError("Virtio NIC: Illegal read length for page frame number\n"); return -1; } switch (queue_idx) { @@ -513,7 +527,7 @@ static int virtio_io_read(struct guest_info *core, case VRING_SIZE_PORT: if (length != 2) { - PrintError("Illegal read length for vring size\n"); + PrintError("Virtio NIC: Illegal read length for vring size\n"); return -1; } switch (queue_idx) { @@ -533,7 +547,7 @@ static int virtio_io_read(struct guest_info *core, case VIRTIO_STATUS_PORT: if (length != 1) { - PrintError("Illegal read length for status\n"); + PrintError("Virtio NIC: Illegal read length for status\n"); return -1; } *(uint8_t *)dst = virtio->virtio_cfg.status; @@ -542,7 +556,8 @@ static int virtio_io_read(struct guest_info *core, 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); + v3_pci_lower_irq(virtio->virtio_dev->pci_bus, + 0, virtio->pci_dev); break; case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN: @@ -550,7 +565,8 @@ static int virtio_io_read(struct guest_info *core, break; default: - PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx); + PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", + port_idx); return -1; } @@ -558,110 +574,163 @@ static int virtio_io_read(struct guest_info *core, } +/* 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 = size; - uint32_t offset = 0; unsigned long flags; - int ret_val = -ERR_VIRTIO_OTHER; - int raw = 1; - -#ifndef CONFIG_DEBUG_VIRTIO_NET - { - PrintDebug("Virtio-NIC: virtio_rx: size: %d\n", size); - //v3_hexdump(buf, size, NULL, 0); - } -#endif + uint8_t kick_guest = 0; - flags = v3_lock_irqsave(virtio->rx_lock); + V3_Net_Print(2, "Virtio NIC: virtio_rx: size: %d\n", size); - virtio->pkt_recv ++; - if (!raw) - data_len -= hdr_len; + if (!q->ring_avail_addr) { + V3_Net_Print(2, "Virtio NIC: RX Queue not set\n"); + virtio->stats.rx_dropped ++; + + return -1; + } - if (!raw) - memcpy(&hdr, buf, sizeof(struct virtio_net_hdr_mrg_rxbuf)); - else - memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf)); + memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf)); - if (q->ring_avail_addr == 0) { - PrintError("Queue is not set\n"); - ret_val = -ERR_VIRTIO_RXQ_NOSET; - goto exit; - } + flags = v3_lock_irqsave(virtio->rx_lock); 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; + uint16_t buf_idx; + struct vring_desc * buf_desc; + uint32_t hdr_len, len; + uint32_t offset = 0; + + hdr_len = (virtio->mergeable_rx_bufs)? + sizeof(struct virtio_net_hdr_mrg_rxbuf): + sizeof(struct virtio_net_hdr); + + if(virtio->mergeable_rx_bufs){/* merged buffer */ + struct vring_desc * hdr_desc; + uint16_t old_idx = q->cur_avail_idx; + + buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; + hdr_desc = &(q->desc[buf_idx]); + + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), + virtio, hdr_desc, buf, size, hdr_len); + if(len < 0){ + goto err_exit; + } + offset += len; - hdr_desc = &(q->desc[hdr_idx]); - if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) { - PrintError("Could not translate receive buffer address\n"); - goto 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; - } + 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 = hdr_len + offset; + q->cur_avail_idx ++; + hdr.num_buffers ++; + + while(offset < size) { + buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; + buf_desc = &(q->desc[buf_idx]); + + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), + virtio, buf_desc, buf+offset, size-offset, 0); + if (len < 0){ + V3_Net_Print(2, "Virtio NIC: merged buffer, %d buffer size %d\n", + hdr.num_buffers, len); + q->cur_avail_idx = old_idx; + goto err_exit; + } + offset += len; + buf_desc->flags &= ~VIRTIO_NEXT_FLAG; + + q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; + q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len; + q->cur_avail_idx ++; + + hdr.num_buffers ++; + } - 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; + copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), + virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0); + q->used->index += hdr.num_buffers; + }else{ + buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; 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); + /* copy header */ + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), + virtio, buf_desc, (uchar_t *)&(hdr.hdr), hdr_len, 0); + if(len < hdr_len){ + V3_Net_Print(2, "Virtio NIC: rx copy header error %d, hdr_len %d\n", + len, hdr_len); + goto err_exit; + } + + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), + virtio, buf_desc, buf, size, hdr_len); + if(len < 0){ + V3_Net_Print(2, "Virtio NIC: rx copy data error %d\n", len); + goto err_exit; + } offset += len; - if (offset < data_len) { - buf_desc->flags = VIRTIO_NEXT_FLAG; + + /* copy rest of data */ + while(offset < size && + (buf_desc->flags & VIRTIO_NEXT_FLAG)){ + buf_desc = &(q->desc[buf_desc->next]); + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), + virtio, buf_desc, buf+offset, size-offset, 0); + if (len < 0) { + break; + } + offset += len; } - 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++; - - /* if there are certain num of pkts in the RX queue, notify guest - * so guest will exit to palacios - * when it returns, guest gets the virtio rx interrupt */ - if((++virtio->buffed_rx > q->queue_size/5) && - (q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - if(virtio->virtio_dev->vm->cores[0].cpu_id != V3_Get_CPU()){ - notify_guest(virtio); + buf_desc->flags &= ~VIRTIO_NEXT_FLAG; + + if(offset < size){ + V3_Net_Print(2, "Virtio NIC: rx not enough ring buffer, buffer size %d\n", + len); + goto err_exit; } - virtio->buffed_rx = 0; - } + + 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 = size + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */ + q->used->index ++; + q->cur_avail_idx ++; + } + + virtio->stats.rx_pkts ++; + virtio->stats.rx_bytes += size; } else { - virtio->pkt_drop++; - /* RX queue is full, tell backend to stop RX on this device */ - virtio->net_ops->stop_rx(virtio->backend_data); - enable_cb(&virtio->rx_vq); - - ret_val = -ERR_VIRTIO_RXQ_FULL; - goto exit; + V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n"); + virtio->stats.rx_dropped ++; + + /* kick guest to refill RX queue */ + kick_guest = 1; } - if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line); + v3_unlock_irqrestore(virtio->rx_lock, flags); + + if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG) || kick_guest) { + V3_Net_Print(2, "Virtio NIC: RX Raising IRQ %d\n", + virtio->pci_dev->config_header.intr_line); + + virtio->virtio_cfg.pci_isr = 0x1; v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); - virtio->virtio_cfg.pci_isr = 0x1; + virtio->stats.rx_interrupts ++; } - ret_val = offset; + /* notify guest if it is in guest mode */ + if((kick_guest || virtio->rx_notify == 1) && + V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].pcpu_id){ + v3_interrupt_cpu(virtio->virtio_dev->vm, + virtio->virtio_dev->vm->cores[0].pcpu_id, + 0); + } -exit: + return 0; +err_exit: + virtio->stats.rx_dropped ++; v3_unlock_irqrestore(virtio->rx_lock, flags); - return ret_val; + return -1; } static int virtio_free(struct virtio_dev_state * virtio) { @@ -687,45 +756,17 @@ static struct v3_device_ops dev_ops = { }; -/* TODO: Issue here: which vm info it needs? calling VM or the device's own VM? */ -static void virtio_nic_poll(struct v3_vm_info * vm, void * data){ - struct virtio_net_state * virtio = (struct virtio_net_state *)data; - - handle_pkt_tx(&(vm->cores[0]), virtio); -} - -static void virtio_start_tx(void * data){ - struct virtio_net_state * virtio = (struct virtio_net_state *)data; - unsigned long flags; +static int virtio_poll(int quote, void * data){ + struct virtio_net_state * virtio = (struct virtio_net_state *)data; - flags = v3_lock_irqsave(virtio->tx_lock); - virtio->tx_disabled = 0; + if (virtio->status) { - /* notify the device's guest to start sending pkt */ - if(virtio->virtio_dev->vm->cores[0].cpu_id != V3_Get_CPU()){ - notify_guest(virtio); - } - v3_unlock_irqrestore(virtio->tx_lock, flags); -} - -static void virtio_stop_tx(void * data){ - struct virtio_net_state * virtio = (struct virtio_net_state *)data; - unsigned long flags; - - flags = v3_lock_irqsave(virtio->tx_lock); - virtio->tx_disabled = 1; - - /* stop the guest to exit to palacios for sending pkt? */ - if(virtio->virtio_dev->vm->cores[0].cpu_id != V3_Get_CPU()){ - disable_cb(&virtio->tx_vq); - } + return handle_pkt_tx(&(virtio->vm->cores[0]), virtio, quote); + } - v3_unlock_irqrestore(virtio->tx_lock, flags); + return 0; } - - - static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_state * net_state) { @@ -735,15 +776,16 @@ static int register_dev(struct virtio_dev_state * virtio, int tmp_ports = num_ports; int i; - // 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 + /* This gets the number of ports, rounded up to a power of 2 */ + net_state->io_range_size = 1; 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; } @@ -752,7 +794,8 @@ static int register_dev(struct virtio_dev_state * virtio, bars[i].type = PCI_BAR_NONE; } - PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size); + PrintDebug("Virtio NIC: io_range_size = %d\n", + net_state->io_range_size); bars[0].type = PCI_BAR_IO; bars[0].default_base_port = -1; @@ -762,7 +805,7 @@ static int register_dev(struct virtio_dev_state * virtio, bars[0].private_data = net_state; pci_dev = v3_pci_register_device(virtio->pci_bus, PCI_STD_DEVICE, - 0, 4/*PCI_AUTO_DEV_NUM*/, 0, + 0, PCI_AUTO_DEV_NUM, 0, "LNX_VIRTIO_NIC", bars, NULL, NULL, NULL, net_state); @@ -791,12 +834,83 @@ static int register_dev(struct virtio_dev_state * virtio, virtio_init_state(net_state); + V3_Print("Virtio NIC: Registered Intr Line %d\n", pci_dev->config_header.intr_line); + /* 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 10000 /*us*/ + +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_us; + static int profile_ms = 0; + + if(!net_state->status){ /* VNIC is not in working status */ + return; + } + + period_us = (1000*cpu_cycles)/cpu_freq; + net_state->past_us += period_us; + + if(net_state->past_us > PROFILE_PERIOD){ + uint32_t tx_rate, rx_rate; + + tx_rate = (net_state->stats.tx_pkts - net_state->tx_pkts)/(net_state->past_us/1000); /* pkts/per ms */ + rx_rate = (net_state->stats.rx_pkts - net_state->rx_pkts)/(net_state->past_us/1000); + + net_state->tx_pkts = net_state->stats.tx_pkts; + net_state->rx_pkts = net_state->stats.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){ + V3_Print("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){ + V3_Print("Virtio NIC: Switch RX to VMM notify mode\n"); + net_state->rx_notify = 1; + } + + net_state->past_us = 0; + } + + profile_ms += period_us/1000; + if(profile_ms > 20000){ + V3_Net_Print(1, "Virtio NIC: TX: Pkt: %lld, Bytes: %lld\n\t\tRX Pkt: %lld. Bytes: %lld\n\t\tDropped: tx %lld, rx %lld\nInterrupts: tx %d, rx %d\nTotal Exit: %lld\n", + net_state->stats.tx_pkts, net_state->stats.tx_bytes, + net_state->stats.rx_pkts, net_state->stats.rx_bytes, + net_state->stats.tx_dropped, net_state->stats.rx_dropped, + net_state->stats.tx_interrupts, net_state->stats.rx_interrupts, + net_state->vm->cores[0].num_exits); + profile_ms = 0; + } +} + +static struct v3_timer_ops timer_ops = { + .update_timer = virtio_nic_timer, +}; + + static int connect_fn(struct v3_vm_info * info, void * frontend_data, struct v3_dev_net_ops * ops, @@ -808,58 +922,38 @@ static int connect_fn(struct v3_vm_info * info, memset(net_state, 0, sizeof(struct virtio_net_state)); register_dev(virtio, net_state); + net_state->vm = info; 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->start_tx = virtio_start_tx; - ops->stop_tx = virtio_stop_tx; - ops->frontend_data = net_state; - memcpy(ops->fnt_mac, virtio->mac, ETH_ALEN); + ops->poll = virtio_poll; + ops->config.frontend_data = net_state; + ops->config.poll = 1; + ops->config.quote = 64; + ops->config.fnt_mac = V3_Malloc(ETH_ALEN); + memcpy(ops->config.fnt_mac, virtio->mac, ETH_ALEN); return 0; } - -static int str2mac(char * macstr, char * mac){ - char hex[2], *s = macstr; - int i = 0; - - while(s){ - memcpy(hex, s, 2); - mac[i++] = (char)atox(hex); - if (i == ETH_ALEN) return 0; - s=strchr(s, ':'); - if(s) s++; - } - - return -1; -} - -static inline void random_ethaddr(uchar_t * addr) -{ - uint64_t val; - - /* using current rdtsc as random number */ - rdtscll(val); - *(uint64_t *)addr = val; - - addr [0] &= 0xfe; /* clear multicast bit */ - addr [0] |= 0x02; /* set local assignment bit (IEEE802) */ -} - - 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 * dev_id = v3_cfg_val(cfg, "ID"); - char * macstr = v3_cfg_val(cfg, "mac"); + char macstr[128]; + char * str = v3_cfg_val(cfg, "mac"); + memcpy(macstr, str, strlen(str)); if (pci_bus == NULL) { - PrintError("Virtio NIC: VirtIO devices require a PCI Bus"); + PrintError("Virtio NIC: Virtio device require a PCI Bus"); return -1; } @@ -873,7 +967,6 @@ static int virtio_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) { if (macstr != NULL && !str2mac(macstr, virtio_state->mac)) { PrintDebug("Virtio NIC: Mac specified %s\n", macstr); }else { - PrintDebug("Virtio NIC: MAC not specified\n"); random_ethaddr(virtio_state->mac); }