X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fdevices%2Flnx_virtio_nic.c;h=bb13a6959e54f553105c625f693c012c9d8a70ee;hb=259a6afe752758f5ef79b19d12b599a6419cecc9;hp=4f86bbe57ae299dbb653fff060ed35b1e32a2b1c;hpb=4c4382c3fa7e7e87c260b6bb473220d0081ca1d5;p=palacios.git diff --git a/palacios/src/devices/lnx_virtio_nic.c b/palacios/src/devices/lnx_virtio_nic.c index 4f86bbe..bb13a69 100644 --- a/palacios/src/devices/lnx_virtio_nic.c +++ b/palacios/src/devices/lnx_virtio_nic.c @@ -29,6 +29,8 @@ #include #include #include +#include +#include #ifndef CONFIG_DEBUG_VIRTIO_NET @@ -36,13 +38,12 @@ #define PrintDebug(fmt, args...) #endif -#define VIRTIO_NET_S_LINK_UP 1 /* Link is up */ #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 */ @@ -51,23 +52,24 @@ struct virtio_net_hdr { }__attribute__((packed)); -/* This is the version of 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 */ }; -#define TX_QUEUE_SIZE 64 -#define RX_QUEUE_SIZE 1024 +#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 { @@ -79,6 +81,8 @@ struct virtio_dev_state { struct vm_device * pci_bus; struct list_head dev_list; struct v3_vm_info *vm; + + uint8_t mac[ETH_ALEN]; }; struct virtio_net_state { @@ -89,40 +93,26 @@ struct virtio_net_state { struct pci_device * pci_dev; int io_range_size; - 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 */ - uint16_t cur_notify_tx_idx; /*for used in update_tx_queue */ + struct v3_timer * timer; - uint64_t pkt_sent, pkt_recv, pkt_drop; - uint64_t tx_stop_times, rx_stop_times, tx_poll_times, rx_ipi_num; + struct nic_statistics statistics; 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_ms; + 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_free(struct vm_device * dev) -{ - - return 0; -} - static int virtio_init_state(struct virtio_net_state * virtio) { virtio->rx_vq.queue_size = RX_QUEUE_SIZE; @@ -149,39 +139,50 @@ static int virtio_init_state(struct virtio_net_state * virtio) virtio->virtio_cfg.pci_isr = 0; - virtio->virtio_cfg.host_features = 0; // (1 << VIRTIO_NET_F_MAC); + 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)){ + (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; } - return virtio->net_ops->send(buf, len, virtio->backend_data, NULL); + if(virtio->net_ops->send(buf, len, virtio->backend_data) >= 0){ + virtio->statistics.tx_pkts ++; + virtio->statistics.tx_bytes += len; + + return 0; + } + + virtio->statistics.tx_dropped ++; + + return -1; } -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) +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; @@ -217,61 +218,38 @@ 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 */ + +/* 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); } -/* guest free some pkts from rx queue */ -static int handle_rx_kick(struct guest_info *core, struct virtio_net_state * virtio) +/* guest free some pkts for rx queue */ +static int handle_rx_queue_kick(struct guest_info * core, + struct virtio_net_state * virtio) { - unsigned long flags; - - flags = v3_lock_irqsave(virtio->rx_lock); - - virtio->net_ops->start_rx(virtio->backend_data); - disable_cb(&virtio->rx_vq); - - v3_unlock_irqrestore(virtio->rx_lock, flags); - return 0; } -#ifdef CONFIG_VNET_PROFILE -static void print_profile_info(struct virtio_net_state *virtio){ - PrintError("Virtio NIC: %p, sent: %lld, rxed: %lld, dropped: %lld, \ - tx_stop: %lld, rx_stop: %lld, poll_time: %lld, rx_ipi: %lld\n", - virtio, - virtio->pkt_sent, - virtio->pkt_recv, - virtio->pkt_drop, - virtio->tx_stop_times, - virtio->rx_stop_times, - virtio->tx_poll_times, - virtio->rx_ipi_num); -} -#endif -static int handle_ctrl(struct guest_info *core, struct virtio_net_state * virtio) { +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) { - 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); + struct virtio_net_hdr *hdr = NULL; + int txed = 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) { @@ -288,18 +266,18 @@ static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virt goto exit_error; } - hdr = (struct virtio_net_hdr*)hdr_addr; + hdr = (struct virtio_net_hdr *)hdr_addr; desc_idx = hdr_desc->next; if(desc_cnt > 2){ - PrintError("VNIC: merged rx buffer not supported\n"); + PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt); 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) { + if (tx_one_pkt(core, virtio_state, buf_desc) == -1) { PrintError("Error handling nic operation\n"); goto exit_error; } @@ -307,173 +285,38 @@ static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virt req_len += buf_desc->length; desc_idx = buf_desc->next; } - 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->ring[q->used->index % q->queue_size].length = req_len; /* What do we set this to???? */ q->used->index ++; q->cur_avail_idx ++; + + txed ++; } v3_unlock_irqrestore(virtio_state->tx_lock, flags); - - if(!recved) - return 0; - if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { + 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; - } - - -#ifdef CONFIG_VNET_PROFILE - static long min = 1024, max = 0, total=0; - static int i=0; - total += recved; - i ++; - if(recved > max) max = recved; - if(recved < min) min = recved; - if(total > 100000) { - PrintError("VNIC: TX polling: %ld, min %ld, max %ld, avg: %ld pkts\n", total, min, max, total/i); - min = 1024; - max = 0; - i = 1; - total = 0; - } -#endif - - - return 0; - -exit_error: - - v3_unlock_irqrestore(virtio_state->tx_lock, flags); - return -ERR_VIRTIO_OTHER; -} - - -#if 0 //for multicore VNET -/* used for poll pkt from virtio nic by VNET - * only when vnet is running on sidecore */ -static int handle_pkt_tx_sidecore(struct guest_info *core, struct virtio_net_state * virtio) -{ - struct virtio_queue * q = &(virtio->tx_vq); - struct virtio_net_hdr * hdr = NULL; - int recved = 0; - unsigned long flags; - - if (!q->ring_avail_addr) { - return -ERR_VIRTIO_TXQ_NOSET; - } - - flags = v3_lock_irqsave(virtio->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]; - int desc_cnt = get_desc_count(q, desc_idx); - uint32_t req_len = 0; - int i = 0; - - hdr_desc = &(q->desc[desc_idx]); - if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) { - PrintError("Could not translate block header address\n"); - goto exit_error; - } - hdr = (struct virtio_net_hdr*)hdr_addr; - desc_idx = hdr_desc->next; - - if(desc_cnt > 2){ - PrintError("VNIC: merged rx buffer not supported\n"); - goto exit_error; - } - - for (i = 0; i < desc_cnt - 1; i++) { - struct vring_desc * buf_desc = &(q->desc[desc_idx]); - if (pkt_tx_sidecore(core, virtio, buf_desc) < 0) { - PrintError("Error handling nic operation\n"); - goto exit_error; - } - - req_len += buf_desc->length; - desc_idx = buf_desc->next; - } - recved ++; - - q->cur_avail_idx ++; + virtio_state->statistics.interrupts ++; } - if(recved) PrintDebug("VNIC: Tx polling %d pkts\n", recved); - - v3_unlock_irqrestore(virtio->tx_lock, flags); - return 0; exit_error: - v3_unlock_irqrestore(virtio->tx_lock, flags); - return -ERR_VIRTIO_OTHER; + v3_unlock_irqrestore(virtio_state->tx_lock, flags); + return -1; } -/* called by VNET, to notify Virtio to update the tx_queue used index - * used only when vnet running on sidecore */ -static int update_tx_queue(struct virtio_net_state * virtio, int handled_pkt){ - struct virtio_queue * q = &(virtio->tx_vq); - unsigned long flags; - - if (!q->ring_avail_addr) { - return -ERR_VIRTIO_TXQ_NOSET; - } - - flags = v3_lock_irqsave(virtio->tx_lock); - while(handled_pkt > 0){ - q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[virtio->cur_notify_tx_idx % q->queue_size]; - q->used->ring[q->used->index % q->queue_size].length = req_len; // TODO: - q->used->index ++; - virtio->cur_notify_tx_idx ++; - handled_pkt --; - } - v3_unlock_irqrestore(virtio->tx_lock, flags); - - if (handled_pkt && (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG))) { - v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); - virtio->virtio_cfg.pci_isr = 0x1; - - /* do we need to notify here? */ - notify_guest(virtio); - } - - virtio->pkt_sent += handled_pkt; - -#ifdef CONFIG_VNET_PROFILE - if (virtio->pkt_sent % 50000 == 0){ - long cur_time, time; - rdtscll(cur_time); - time = cur_time - virtio_state->last_sent_time; - PrintError("Virtio NIC: last sent 50000 cycles: %ld\n",time); - PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n", - virtio->pkt_sent, - virtio->pkt_recv, - virtio->pkt_drop); - rdtscll(virtio->last_sent_time); - } -#endif - - - return 0; -} -#endif - 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) { + 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; @@ -510,12 +353,13 @@ static int virtio_setup_queue(struct guest_info *core, return 0; } -static int virtio_io_write(struct guest_info *core, 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 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); @@ -539,11 +383,9 @@ static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, u switch (queue_idx) { case 0: virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr); - disable_cb(&virtio->rx_vq); break; case 1: virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr); - disable_cb(&virtio->tx_vq); break; case 2: virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr); @@ -566,17 +408,18 @@ static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, u { uint16_t queue_idx = *(uint16_t *)src; if (queue_idx == 0){ - handle_rx_kick(core, virtio); - PrintError("rx kick\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){ - PrintError("tx kick\n"); if (handle_pkt_tx(core, virtio) == -1) { - PrintError("Could not handle NIC Notification\n"); + PrintError("Could not handle Virtio NIC tx kick\n"); return -1; } } else if (queue_idx == 2){ if (handle_ctrl(core, virtio) == -1) { - PrintError("Could not handle NIC Notification\n"); + PrintError("Could not handle Virtio NIC ctrl kick\n"); return -1; } } else { @@ -588,7 +431,6 @@ static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, u 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; @@ -605,13 +447,15 @@ static int virtio_io_write(struct guest_info *core, uint16_t port, void * src, u return length; } -static int virtio_io_read(struct guest_info *core, 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 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 %p: Read for port %d (index =%d), length=%d\n", private_data, + PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", private_data, port, port_idx, length); switch (port_idx) { @@ -677,6 +521,10 @@ static int virtio_io_read(struct guest_info *core, uint16_t port, void * dst, ui v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); break; + case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN: + *(uint8_t *)dst = virtio->net_cfg.mac[port_idx-VIRTIO_NET_CONFIG]; + break; + default: PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx); return -1; @@ -686,32 +534,29 @@ static int virtio_io_read(struct guest_info *core, uint16_t port, void * dst, ui } +/* 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 data_len; uint32_t offset = 0; unsigned long flags; - int ret_val = -ERR_VIRTIO_OTHER; - int raw = 1; - flags = v3_lock_irqsave(virtio->rx_lock); +#ifdef CONFIG_DEBUG_VIRTIO_NET + PrintDebug("Virtio-NIC: virtio_rx: size: %d\n", size); + v3_hexdump(buf, size, NULL, 0); +#endif - virtio->pkt_recv ++; - if (!raw) - data_len -= hdr_len; + flags = v3_lock_irqsave(virtio->rx_lock); - if (!raw) - memcpy(&hdr, buf, sizeof(struct virtio_net_hdr_mrg_rxbuf)); - else - memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf)); + data_len = size; + 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; + PrintDebug("Queue is not set\n"); + goto err_exit; } if (q->cur_avail_idx != q->avail->index){ @@ -722,8 +567,8 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { 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; + 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)); @@ -750,243 +595,71 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_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); - virtio->rx_ipi_num ++; - } - virtio->buffed_rx = 0; - } + virtio->statistics.rx_pkts ++; + virtio->statistics.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); - - virtio->rx_stop_times ++; + virtio->statistics.rx_dropped ++; - ret_val = -ERR_VIRTIO_RXQ_FULL; - goto exit; + 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 ++; } - ret_val = offset; - -exit: - v3_unlock_irqrestore(virtio->rx_lock, flags); - - return ret_val; -} - - -#if 0 /* for encapuslation */ -/* virtio RX with encapulation version */ -static int virtio_rx_encap(uint8_t * data, - uint32_t size, - void * encap_header, - uint16_t encap_len, - 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 + encap_len; - unsigned long flags; - int ret_val = -ERR_VIRTIO_OTHER; - - if (q->ring_avail_addr == 0) { - PrintError("Queue is not set\n"); - ret_val = -ERR_VIRTIO_RXQ_NOSET; - goto exit; - } - - memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf)); - - 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; - uint32_t offset = 0; - - hdr_desc = &(q->desc[hdr_idx]); - if (guest_pa_to_host_va(&(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; - } - - struct vring_desc * buf_desc = NULL; - uint32_t len = 0; - buf_idx = hdr_desc->next; - buf_desc = &(q->desc[buf_idx]); - - if(hdr_len > 0 && !encap_header) { - len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, encap_header, encap_len, 0); - offset += len; - } - - len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, data + offset, data_len - offset, offset); - offset += len; - if(data_len < offset)/* if there is large pkt, need merge more buffer */ - PrintDebug("Virtio NIC: data pkt larger than RX queue buffer\n"); - - buf_desc->length = offset; - 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++; - - /* notify guest - * when it returns from EXIT, guest gets the virtio rx interrupt */ - if((++virtio->buffed_rx > q->queue_size/2) && - (q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - if(virtio->virtio_dev->vm->cores[0].cpu_id != V3_Get_CPU()){ - notify_guest(virtio); - } - virtio->buffed_rx = 0; - } - ret_val = offset; - } else { - virtio->pkt_drop++; - ret_val = -ERR_VIRTIO_RXQ_FULL; - goto exit; + /* 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); } - virtio->pkt_recv ++; - 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; - } - -#ifdef CONFIG_VNET_PROFILE - if (virtio->pkt_recv % 50000 == 0){ - long cur_time, time; - rdtscll(cur_time); - time = cur_time - virtio->last_recv_time; - PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n", - virtio->pkt_sent, - virtio->pkt_recv, - virtio->pkt_drop); - rdtscll(virtio->last_recv_time); - } -#endif + return 0; -exit: +err_exit: v3_unlock_irqrestore(virtio->rx_lock, flags); - return ret_val; + return -1; } -#endif +static int virtio_free(struct virtio_dev_state * virtio) { + struct virtio_net_state * backend = NULL; + struct virtio_net_state * tmp = NULL; -static struct v3_device_ops dev_ops = { - .free = virtio_free, - .reset = NULL, - .start = NULL, - .stop = NULL, -}; - -#if 0 //temporary hacking LX -static struct virtio_net_state *vnic_states[2] = {NULL, NULL}; -static int num_vnic = 0; - -void vnic_polling(void *data){ - struct v3_vm_info *info = (struct v3_vm_info *)data; - if(vnic_states[0] != NULL && info == vnic_states[0]->virtio_dev->vm){ - handle_pkt_tx(&(info->cores[0]), vnic_states[0]); - } - - if(vnic_states[1] != NULL && info == vnic_states[1]->virtio_dev->vm){ - handle_pkt_tx(&(info->cores[0]), vnic_states[1]); - } -} -#endif + list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) { -/* 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); - - virtio->tx_poll_times ++; + // unregister from PCI -#ifdef CONFIG_VNET_PROFILE - static uint64_t last_time = 0; - uint64_t time; - rdtscll(time); - if((time - last_time) > 5000000000){ - last_time = time; - print_profile_info(virtio); + list_del(&(backend->dev_link)); + V3_Free(backend); } -#endif -} - -#if 0 /* NAPI */ -/* tx one pkt from guest */ -static int virtio_tx_pkt(struct guest_info *core, struct virtio_net_state * virtio_state){ - -} - -static void virtio_nic_poll(struct v3_vm_info *vm, void *data, int budget){ - + V3_Free(virtio); + return 0; } -#endif - -static void virtio_start_tx(void *data){ - struct virtio_net_state * virtio = (struct virtio_net_state *)data; +static struct v3_device_ops dev_ops = { + .free = (int (*)(void *))virtio_free, +}; - /* do we need a lock here? */ - virtio->tx_disabled = 0; - /* notify the device's guest it can start sending pkt */ - if(virtio->virtio_dev->vm->cores[0].cpu_id != V3_Get_CPU()){ - notify_guest(virtio); - } -} - -static void virtio_stop_tx(void *data){ +static void virtio_nic_poll(struct v3_vm_info * vm, int budget, void * data){ struct virtio_net_state * virtio = (struct virtio_net_state *)data; - /* do we need a lock here? */ - virtio->tx_disabled = 1; - - /* how do we 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); + if(virtio->tx_notify == 0){ + handle_pkt_tx(&(vm->cores[0]), virtio); } - - virtio->tx_stop_times ++; } - - - -static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_state * net_state) +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]; @@ -1001,8 +674,9 @@ static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_stat 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; } @@ -1046,21 +720,71 @@ static int register_dev(struct virtio_dev_state * virtio, struct virtio_net_stat net_state->pci_dev = pci_dev; net_state->virtio_dev = virtio; - uchar_t mac[6] = {0x11,0x11,0x11,0x11,0x11,0x11}; - memcpy(net_state->net_cfg.mac, mac, 6); - - memcpy(pci_dev->config_data, net_state->net_cfg.mac, ETH_ALEN); - + memcpy(net_state->net_cfg.mac, virtio->mac, 6); + virtio_init_state(net_state); -#if 0 //temporary hacking LX - vnic_states[num_vnic ++] = net_state; - PrintError("VNIC: num of vnic %d\n", num_vnic); -#endif + /* 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 v3_vm_info * info, void * frontend_data, struct v3_dev_net_ops * ops, @@ -1074,12 +798,16 @@ static int connect_fn(struct v3_vm_info * 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); return 0; } @@ -1087,9 +815,10 @@ static int connect_fn(struct v3_vm_info * info, 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("Virtio NIC: Initializing VIRTIO Network device: %s\n", name); + 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 NIC: VirtIO devices require a PCI Bus"); @@ -1103,14 +832,30 @@ static int virtio_init(struct v3_vm_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("Virtio NIC: Could not attach device %s\n", name); + 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); + } + + 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; } - if (v3_dev_add_net_frontend(vm, name, connect_fn, (void *)virtio_state) == -1) { - PrintError("Virtio NIC: Could not register %s as net frontend\n", name); + 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; }