#include <palacios/vmm_lock.h>
#include <palacios/vmm_util.h>
#include <devices/pci.h>
+#include <palacios/vmm_ethernet.h>
+#include <palacios/vmm_time.h>
#ifndef CONFIG_DEBUG_VIRTIO_NET
#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 */
}__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
{
struct vm_device * pci_bus;
struct list_head dev_list;
struct v3_vm_info *vm;
+
+ uint8_t mac[ETH_ALEN];
};
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 */
+ struct v3_timer * timer;
- uint64_t pkt_sent, pkt_recv, pkt_drop;
+ 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;
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)){
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,
+static int tx_one_pkt(struct guest_info * core,
struct virtio_net_state * virtio,
struct vring_desc * buf_desc)
{
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;
+ }
+
+ if(virtio->net_ops->send(buf, len, virtio->backend_data) >= 0){
+ virtio->statistics.tx_pkts ++;
+ virtio->statistics.tx_bytes += len;
+
+ return 0;
}
- return virtio->net_ops->send(buf, len, virtio->backend_data);
+ virtio->statistics.tx_dropped ++;
+
+ return -1;
}
}
-/* 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,
+/* 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;
}
-static int handle_ctrl(struct guest_info *core,
+static int handle_ctrl(struct guest_info * core,
struct virtio_net_state * virtio) {
return 0;
}
-static int handle_pkt_tx(struct guest_info *core,
+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) {
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;
}
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;
+
+ virtio_state->statistics.interrupts ++;
}
return 0;
exit_error:
v3_unlock_irqrestore(virtio_state->tx_lock, flags);
- return -ERR_VIRTIO_OTHER;
+ return -1;
}
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);
{
uint16_t queue_idx = *(uint16_t *)src;
if (queue_idx == 0){
- handle_rx_kick(core, virtio);
+ 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(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 {
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;
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) {
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;
}
+/* 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;
-#ifndef CONFIG_DEBUG_VIRTIO_NET
- {
- PrintDebug("Virtio-NIC: virtio_rx: size: %d\n", size);
- //v3_hexdump(buf, size, NULL, 0);
- }
+#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);
- virtio->pkt_recv ++;
- if (!raw)
- data_len -= hdr_len;
-
- 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){
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));
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->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->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 ++;
+ }
+
+ 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);
}
- ret_val = offset;
+ return 0;
-exit:
+err_exit:
v3_unlock_irqrestore(virtio->rx_lock, flags);
- return ret_val;
+ return -1;
}
+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,
-};
+ 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);
+ // unregister from PCI
+
+ list_del(&(backend->dev_link));
+ V3_Free(backend);
+ }
+
+ V3_Free(virtio);
+ return 0;
}
-static void virtio_start_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 = 0;
+static struct v3_device_ops dev_ops = {
+ .free = (int (*)(void *))virtio_free,
+};
- /* 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){
+static void virtio_nic_poll(struct v3_vm_info * vm, int budget, 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);
+ if(virtio->tx_notify == 0){
+ handle_pkt_tx(&(vm->cores[0]), virtio);
}
-
- v3_unlock_irqrestore(virtio->tx_lock, flags);
}
-
-
-
static int register_dev(struct virtio_dev_state * virtio,
struct virtio_net_state * net_state)
{
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;
}
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);
+ /* 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,
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;
}
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");
-
- PrintDebug("Virtio NIC: Initializing VIRTIO Network device: %s\n", dev_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");
virtio_state->pci_bus = pci_bus;
virtio_state->vm = vm;
+ 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) {
