X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fdevices%2Flnx_virtio_nic.c;h=952cc52132ac24fb07a1338ffed42273eca8b75b;hb=2334b64dae68f806dcaf81292e87e90562bfde75;hp=9c55d418ddd24ce17bf31c97f64b094b69a005a5;hpb=afca783e03bd737fe7bb02d498644df30bf5d4d9;p=palacios.git diff --git a/palacios/src/devices/lnx_virtio_nic.c b/palacios/src/devices/lnx_virtio_nic.c index 9c55d41..952cc52 100644 --- a/palacios/src/devices/lnx_virtio_nic.c +++ b/palacios/src/devices/lnx_virtio_nic.c @@ -25,7 +25,7 @@ #include #include #include -#include +#include #include #include #include @@ -33,11 +33,15 @@ #include -#ifndef CONFIG_DEBUG_VIRTIO_NET +#ifndef V3_CONFIG_DEBUG_VIRTIO_NET #undef PrintDebug #define PrintDebug(fmt, args...) #endif +#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 @@ -117,13 +121,17 @@ struct virtio_net_state { struct vm_device * dev; struct pci_device * pci_dev; int io_range_size; + + uint16_t status; struct virtio_queue rx_vq; /* idx 0*/ struct virtio_queue tx_vq; /* idx 1*/ struct virtio_queue ctrl_vq; /* idx 2*/ + uint8_t mergeable_rx_bufs; + struct v3_timer * timer; - void * poll_thread; + struct vnet_thread * poll_thread; struct nic_statistics stats; @@ -165,11 +173,13 @@ 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); - // (1 << VIRTIO_NET_F_GSO) | - // (1 << VIRTIO_NET_F_HOST_UFO) | - // (1 << VIRTIO_NET_F_HOST_TSO4); + 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)){ @@ -185,7 +195,7 @@ static int tx_one_pkt(struct guest_info * core, { uint8_t * buf = NULL; uint32_t len = buf_desc->length; - int synchronize = 1; // (virtio->tx_notify == 1)?1:0; + int synchronize = virtio->tx_notify; if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) { PrintDebug("Could not translate buffer address\n"); @@ -193,7 +203,7 @@ static int tx_one_pkt(struct guest_info * core, } V3_Net_Print(2, "Virtio-NIC: virtio_tx: size: %d\n", len); - if(v3_net_debug >= 4){ + if(net_debug >= 4){ v3_hexdump(buf, len, NULL, 0); } @@ -209,13 +219,13 @@ static int tx_one_pkt(struct guest_info * core, } +/*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 offset) -{ + 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; @@ -223,8 +233,8 @@ static inline int copy_data_to_desc(struct guest_info * core, 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; } @@ -243,11 +253,15 @@ static inline 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; + if(queue->used){ + queue->used->flags |= VRING_NO_NOTIFY_FLAG; + } } static int handle_pkt_tx(struct guest_info * core, @@ -255,7 +269,7 @@ static int handle_pkt_tx(struct guest_info * core, { struct virtio_queue *q = &(virtio_state->tx_vq); int txed = 0; - unsigned long flags; + unsigned long flags; if (!q->ring_avail_addr) { return -1; @@ -263,13 +277,13 @@ static int handle_pkt_tx(struct guest_info * core, flags = v3_lock_irqsave(virtio_state->tx_lock); while (q->cur_avail_idx != q->avail->index) { - struct virtio_net_hdr *hdr = NULL; + 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); - if(desc_cnt > 2){ + if(desc_cnt != 2){ PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt); goto exit_error; } @@ -280,15 +294,20 @@ 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; + V3_Net_Print(2, "Virtio NIC: TX hdr count : %d\n", hdr->num_buffers); + /* 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: Error handling nic operation\n"); goto exit_error; } + 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???? */ @@ -300,12 +319,11 @@ static int handle_pkt_tx(struct guest_info * core, } v3_unlock_irqrestore(virtio_state->tx_lock, flags); - - if (virtio_state->virtio_cfg.pci_isr == 0 && - 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; + 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 ++; } @@ -369,8 +387,9 @@ 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: @@ -391,17 +410,23 @@ 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); + 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); + virtio_setup_queue(core, virtio, + &virtio->tx_vq, + pfn, page_addr); if(virtio->tx_notify == 0){ disable_cb(&virtio->tx_vq); - V3_THREAD_WAKEUP(virtio->poll_thread); + vnet_thread_wakeup(virtio->poll_thread); } 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; @@ -464,8 +489,8 @@ 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: @@ -527,7 +552,8 @@ 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: @@ -535,7 +561,8 @@ 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; } @@ -548,118 +575,155 @@ 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; + uint8_t kick_guest = 0; V3_Net_Print(2, "Virtio-NIC: virtio_rx: size: %d\n", size); - if(v3_net_debug >= 4){ - v3_hexdump(buf, size, NULL, 0); - } - 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) { + if (!q->ring_avail_addr) { V3_Net_Print(2, "Virtio NIC: RX Queue not set\n"); virtio->stats.rx_dropped ++; - goto err_exit; + + return -1; } + 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 buf_idx = 0; - uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; - struct vring_desc * hdr_desc = NULL; - struct vring_desc * buf_desc = NULL; - uint32_t 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) { - V3_Net_Print(2, "Virtio NIC: Could not translate receive buffer address\n"); - virtio->stats.rx_dropped ++; - goto err_exit; - } + 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; -#if 0 /* merged buffer */ - for(buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) { - uint32_t len = 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 = 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 ++; + } + + 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); - offset += len; - if (offset < data_len) { - buf_desc->flags = VIRTIO_NEXT_FLAG; + /* 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; } - buf_desc->length = len; - hdr.num_buffers ++; - } - buf_desc->flags &= ~VIRTIO_NEXT_FLAG; - memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr_mrg_rxbuf)); -#endif - hdr.num_buffers = 1; - memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr_mrg_rxbuf)); - if (data_len == 0) { - hdr_desc->flags &= ~VIRTIO_NEXT_FLAG; - } + 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; - buf_idx = hdr_desc->next; - buf_desc = &(q->desc[buf_idx]); - len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf, data_len, 0); - if (len < data_len) { - V3_Net_Print(2, "Virtio NIC: ring buffer len less than pkt size, merged buffer not supported\n"); - virtio->stats.rx_dropped ++; - - goto err_exit; - } - 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->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(offset < size){ + V3_Net_Print(2, "Virtio NIC: rx not enough ring buffer, buffer size %d\n", + len); + goto err_exit; + } + + 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 { V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n"); - virtio->stats.rx_dropped ++; + virtio->stats.rx_dropped ++; - /* kick guest to refill the queue */ - virtio->virtio_cfg.pci_isr = 0x1; - v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); - v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0); - virtio->stats.rx_interrupts ++; - - goto err_exit; + /* kick guest to refill RX queue */ + kick_guest = 1; } - if (virtio->virtio_cfg.pci_isr == 0 && - !(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->stats.rx_interrupts ++; } - v3_unlock_irqrestore(virtio->rx_lock, flags); - /* notify guest if it is in guest mode */ - /* ISSUE: What is gonna happen if guest thread is running on the same core as this thread? */ - if(virtio->rx_notify == 1){ - v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0); + 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); } return 0; err_exit: - + virtio->stats.rx_dropped ++; v3_unlock_irqrestore(virtio->rx_lock, flags); return -1; @@ -691,14 +755,15 @@ static struct v3_device_ops dev_ops = { static int virtio_tx_flush(void * args){ struct virtio_net_state *virtio = (struct virtio_net_state *)args; - V3_Print("Virtio TX Poll Thread Starting for %s\n", virtio->vm->name); + V3_Print("Virtio TX Poll Thread Starting for %s\n", + virtio->vm->name); while(1){ if(virtio->tx_notify == 0){ handle_pkt_tx(&(virtio->vm->cores[0]), virtio); v3_yield(NULL); }else { - V3_THREAD_SLEEP(); + vnet_thread_sleep(-1); } } @@ -714,7 +779,7 @@ 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 + /* 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; @@ -722,8 +787,8 @@ static int register_dev(struct virtio_dev_state * virtio, } /* 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 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; } @@ -732,7 +797,8 @@ 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; @@ -777,6 +843,7 @@ static int register_dev(struct virtio_dev_state * virtio, return 0; } + #define RATE_UPPER_THRESHOLD 10 /* 10000 pkts per second, around 100Mbits */ #define RATE_LOWER_THRESHOLD 1 #define PROFILE_PERIOD 10000 /*us*/ @@ -788,10 +855,13 @@ static void virtio_nic_timer(struct guest_info * core, 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 0 if(net_state->past_us > PROFILE_PERIOD){ uint32_t tx_rate, rx_rate; @@ -805,7 +875,7 @@ static void virtio_nic_timer(struct guest_info * core, V3_Print("Virtio NIC: Switch TX to VMM driven mode\n"); disable_cb(&(net_state->tx_vq)); net_state->tx_notify = 0; - V3_THREAD_WAKEUP(net_state->poll_thread); + vnet_thread_wakeup(net_state->poll_thread); } if(tx_rate < RATE_LOWER_THRESHOLD && net_state->tx_notify == 0){ @@ -826,7 +896,6 @@ static void virtio_nic_timer(struct guest_info * core, net_state->past_us = 0; } -#endif profile_ms += period_us/1000; if(profile_ms > 20000){ @@ -863,13 +932,17 @@ static int connect_fn(struct v3_vm_info * info, net_state->tx_notify = 0; net_state->rx_notify = 0; - net_state->timer = v3_add_timer(&(info->cores[0]),&timer_ops,net_state); + net_state->timer = v3_add_timer(&(info->cores[0]), + &timer_ops,net_state); ops->recv = virtio_rx; ops->frontend_data = net_state; memcpy(ops->fnt_mac, virtio->mac, ETH_ALEN); - net_state->poll_thread = V3_CREATE_THREAD(virtio_tx_flush, (void *)net_state, "Virtio_Poll"); + net_state->poll_thread = vnet_start_thread(virtio_tx_flush, + (void *)net_state, "Virtio_Poll"); + + net_state->status = 1; return 0; }