+
/*
* This file is part of the Palacios Virtual Machine Monitor developed
* by the V3VEE Project with funding from the United States National
#include <devices/lnx_virtio_pci.h>
#include <palacios/vm_guest_mem.h>
#include <palacios/vmm_sprintf.h>
-#include <palacios/vmm_vnet.h>
+#include <vnet/vnet.h>
#include <palacios/vmm_lock.h>
#include <palacios/vmm_util.h>
#include <devices/pci.h>
#include <palacios/vmm_time.h>
-#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
} __attribute__((packed));
struct virtio_dev_state {
+
struct vm_device * pci_bus;
struct list_head dev_list;
struct v3_vm_info *vm;
+ enum {GUEST_DRIVEN=0, VMM_DRIVEN, ADAPTIVE} model;
+ uint64_t lower_thresh_pps, upper_thresh_pps, period_us;
+
uint8_t mac[ETH_ALEN];
};
struct virtio_net_state {
+
struct virtio_net_config net_cfg;
struct virtio_config virtio_cfg;
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*/
- struct v3_timer * timer;
- void * poll_thread;
+ uint8_t mergeable_rx_bufs;
+ struct v3_timer * timer;
struct nic_statistics stats;
struct v3_dev_net_ops * net_ops;
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)){
- PrintError("Virtio NIC: Failure to init locks for net_state\n");
+ PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Failure to init locks for net_state\n");
+ }
+
+ return 0;
+}
+
+static int virtio_deinit_state(struct guest_info *core, struct virtio_net_state *ns)
+{
+ if (ns->timer) {
+ v3_remove_timer(core,ns->timer);
}
+ v3_lock_deinit(&(ns->rx_lock));
+ v3_lock_deinit(&(ns->tx_lock));
+
+
return 0;
}
{
uint8_t * buf = NULL;
uint32_t len = buf_desc->length;
- int synchronize = 1; // (virtio->tx_notify == 1)?1:0;
if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
- PrintDebug("Could not translate buffer address\n");
+ PrintDebug(core->vm_info, core, "Could not translate buffer address\n");
return -1;
}
-
+
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);
}
- if(virtio->net_ops->send(buf, len, synchronize, virtio->backend_data) < 0){
+ 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;
}
+/*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;
-
+
if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
- PrintDebug("Could not translate buffer address\n");
+ PrintDebug(core->vm_info, core, "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 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,
- struct virtio_net_state * virtio_state)
+ struct virtio_net_state * virtio_state,
+ int quote)
{
- struct virtio_queue *q = &(virtio_state->tx_vq);
- int txed = 0;
+ struct virtio_queue * q;
+ int txed = 0, left = 0;
unsigned long flags;
+ q = &(virtio_state->tx_vq);
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 *hdr = NULL;
+ while (1) {
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);
+ uint16_t desc_idx, tmp_idx;
+ int desc_cnt;
+
+ flags = v3_lock_irqsave(virtio_state->tx_lock);
- if(desc_cnt > 2){
- PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
- goto exit_error;
+ if(q->cur_avail_idx == q->avail->index ||
+ (quote > 0 && txed >= quote)) {
+ left = (q->cur_avail_idx != q->avail->index);
+ v3_unlock_irqrestore(virtio_state->tx_lock, flags);
+ break;
}
+
+ desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
+ tmp_idx = q->cur_avail_idx ++;
+
+ v3_unlock_irqrestore(virtio_state->tx_lock, flags);
- 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;
+ desc_cnt = get_desc_count(q, desc_idx);
+ if(desc_cnt != 2){
+ PrintError(core->vm_info, core, "VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
}
- hdr = (struct virtio_net_hdr *)hdr_addr;
- desc_idx = hdr_desc->next;
+ hdr_desc = &(q->desc[desc_idx]);
+ if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) != -1) {
+ struct virtio_net_hdr_mrg_rxbuf * hdr;
+ struct vring_desc * buf_desc;
+
+ hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr;
+ desc_idx = hdr_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: Error handling nic operation\n");
- goto exit_error;
+ /* here we assumed that one ethernet pkt is not splitted into multiple buffer */
+ buf_desc = &(q->desc[desc_idx]);
+ if (tx_one_pkt(core, virtio_state, buf_desc) == -1) {
+ PrintError(core->vm_info, core, "Virtio NIC: Fails to send packet\n");
+ }
+ } else {
+ PrintError(core->vm_info, core, "Could not translate block header address\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???? */
+
+ flags = v3_lock_irqsave(virtio_state->tx_lock);
+
+ q->used->ring[q->used->index % q->queue_size].id =
+ q->avail->ring[tmp_idx % q->queue_size];
+
q->used->index ++;
- q->cur_avail_idx ++;
+ v3_unlock_irqrestore(virtio_state->tx_lock, flags);
txed ++;
}
-
- v3_unlock_irqrestore(virtio_state->tx_lock, flags);
-
- //virtio_state->virtio_cfg.pci_isr == 0 &&
+
if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
- v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
+ v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus,
+ virtio_state->pci_dev, 0);
virtio_state->virtio_cfg.pci_isr = 0x1;
-
virtio_state->stats.rx_interrupts ++;
}
- if(txed > 0) {
- V3_Net_Print(2, "Virtio Handle TX: txed pkts: %d\n", txed);
- }
-
- return 0;
-
-exit_error:
-
- v3_unlock_irqrestore(virtio_state->tx_lock, flags);
- return -1;
+ return left;
}
// round up to next page boundary.
queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
- PrintError("Could not translate ring descriptor address\n");
+ PrintError(core->vm_info, core, "Could not translate ring descriptor address\n");
return -1;
}
if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
- PrintError("Could not translate ring available address\n");
+ PrintError(core->vm_info, core, "Could not translate ring available address\n");
return -1;
}
if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
- PrintError("Could not translate ring used address\n");
+ PrintError(core->vm_info, core, "Could not translate ring used address\n");
return -1;
}
- PrintDebug("RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
+ PrintDebug(core->vm_info, core, "RingDesc_addr=%p, Avail_addr=%p, Used_addr=%p\n",
(void *)(queue->ring_desc_addr),
(void *)(queue->ring_avail_addr),
(void *)(queue->ring_used_addr));
- PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
+ PrintDebug(core->vm_info, core, "RingDesc=%p, Avail=%p, Used=%p\n",
queue->desc, queue->avail, queue->used);
return 0;
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(core->vm_info, core, "VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n",
+ private_data, port, port_idx,
+ length, *(uint32_t *)src);
switch (port_idx) {
case GUEST_FEATURES_PORT:
if (length != 4) {
- PrintError("Illegal write length for guest features\n");
+ PrintError(core->vm_info, core, "Illegal write length for guest features\n");
return -1;
}
virtio->virtio_cfg.guest_features = *(uint32_t *)src;
case VRING_PG_NUM_PORT:
if (length != 4) {
- PrintError("Illegal write length for page frame number\n");
+ PrintError(core->vm_info, core, "Illegal write length for page frame number\n");
return -1;
}
addr_t pfn = *(uint32_t *)src;
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);
}
+ 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;
case VRING_Q_SEL_PORT:
virtio->virtio_cfg.vring_queue_selector = *(uint16_t *)src;
if (virtio->virtio_cfg.vring_queue_selector > 2) {
- PrintError("Virtio NIC: wrong queue idx: %d\n",
+ PrintError(core->vm_info, core, "Virtio NIC: wrong queue idx: %d\n",
virtio->virtio_cfg.vring_queue_selector);
return -1;
}
/* receive queue refill */
virtio->stats.tx_interrupts ++;
} else if (queue_idx == 1){
- if (handle_pkt_tx(core, virtio) == -1) {
- PrintError("Could not handle Virtio NIC tx kick\n");
+ if (handle_pkt_tx(core, virtio, 0) < 0) {
+ PrintError(core->vm_info, core, "Virtio NIC: Error to handle packet TX\n");
return -1;
}
virtio->stats.tx_interrupts ++;
} else if (queue_idx == 2){
/* ctrl */
} else {
- PrintError("Wrong queue index %d\n", queue_idx);
+ PrintError(core->vm_info, core, "Virtio NIC: Wrong queue index %d\n", queue_idx);
}
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 0x%x (index =%d), length=%d\n", private_data,
- port, port_idx, length);
+ PrintDebug(core->vm_info, core, "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");
+ PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for host features\n");
//return -1;
}
*(uint32_t *)dst = virtio->virtio_cfg.host_features;
case VRING_PG_NUM_PORT:
if (length != 4) {
- PrintError("Illegal read length for page frame number\n");
+ PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for page frame number\n");
return -1;
}
switch (queue_idx) {
case VRING_SIZE_PORT:
if (length != 2) {
- PrintError("Illegal read length for vring size\n");
+ PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for vring size\n");
return -1;
}
switch (queue_idx) {
case VIRTIO_STATUS_PORT:
if (length != 1) {
- PrintError("Illegal read length for status\n");
+ PrintError(core->vm_info, core, "Virtio NIC: Illegal read length for status\n");
return -1;
}
*(uint8_t *)dst = virtio->virtio_cfg.status;
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,
+ virtio->pci_dev, 0);
break;
- case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN:
+ case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN - 1:
*(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);
+ PrintError(core->vm_info, core, "Virtio NIC: Read of Unhandled Virtio Read:%d\n",
+ port_idx);
return -1;
}
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));
+ V3_Net_Print(2, "Virtio NIC: virtio_rx: size: %d\n", size);
- 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;
+
+ 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 ++;
+ }
-#if 0 /* merged buffer */
- 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);
- 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;
}
- V3_Net_Print(2, "pci_isr %d, virtio flags %d\n", virtio->virtio_cfg.pci_isr, q->avail->flags);
- //virtio->virtio_cfg.pci_isr == 0 &&
+ v3_unlock_irqrestore(virtio->rx_lock, flags);
- if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
- V3_Net_Print(2, "Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
+ 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);
-
+ v3_pci_raise_irq(virtio->virtio_dev->pci_bus, virtio->pci_dev, 0);
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;
list_for_each_entry_safe(backend, tmp, &(virtio->dev_list), dev_link) {
-
- // unregister from PCI
-
+ virtio_deinit_state(&(virtio->vm->cores[0]),backend);
list_del(&(backend->dev_link));
V3_Free(backend);
}
V3_Free(virtio);
+
return 0;
}
};
-static int virtio_tx_flush(void * args){
- struct virtio_net_state *virtio = (struct virtio_net_state *)args;
+static int virtio_poll(int quote, void * data){
+ struct virtio_net_state * virtio = (struct virtio_net_state *)data;
- V3_Print("Virtio TX Poll Thread Starting for %s\n", virtio->vm->name);
+ if (virtio->status) {
- while(1){
- if(virtio->tx_notify == 0){
- handle_pkt_tx(&(virtio->vm->cores[0]), virtio);
- v3_yield(NULL);
- }else {
- V3_THREAD_SLEEP();
- }
- }
+ return handle_pkt_tx(&(virtio->vm->cores[0]), virtio, quote);
+ }
return 0;
}
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
- */
+ * 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;
}
bars[i].type = PCI_BAR_NONE;
}
- PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
+ PrintDebug(VM_NONE, VCORE_NONE, "Virtio NIC: io_range_size = %d\n",
+ net_state->io_range_size);
bars[0].type = PCI_BAR_IO;
bars[0].default_base_port = -1;
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);
+ NULL, NULL, NULL, NULL, net_state);
if (!pci_dev) {
- PrintError("Virtio NIC: Could not register PCI Device\n");
+ PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Could not register PCI Device\n");
return -1;
}
- PrintDebug("Virtio NIC: registered to PCI bus\n");
+ PrintDebug(VM_NONE, VCORE_NONE, "Virtio NIC: registered to PCI bus\n");
pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
virtio_init_state(net_state);
+ V3_Print(VM_NONE, VCORE_NONE, "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*/
+#define RATE_UPPER_THRESHOLD_DEFAULT 10000 /* 10000 pkts per second, around 100Mbits */
+#define RATE_LOWER_THRESHOLD_DEFAULT 1000 /* 1000 pkts per second, around 10Mbits */
+#define PROFILE_PERIOD_DEFAULT 10000 /* us */
static void virtio_nic_timer(struct guest_info * core,
uint64_t cpu_cycles, uint64_t cpu_freq,
struct virtio_net_state * net_state = (struct virtio_net_state *)priv_data;
uint64_t period_us;
static int profile_ms = 0;
+ uint64_t target_period_us = net_state->virtio_dev->period_us;
+ uint64_t upper_thresh_pps = net_state->virtio_dev->upper_thresh_pps;
+ uint64_t lower_thresh_pps = net_state->virtio_dev->lower_thresh_pps;
+
+
+ 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;
-
- 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);
+ if (net_state->past_us > target_period_us) {
+
+ uint64_t tx_count, rx_count;
+ uint64_t lb_tx_count, lb_rx_count;
+ uint64_t ub_tx_count, ub_rx_count;
+
+ lb_tx_count = lb_rx_count = (lower_thresh_pps * 1000000) / net_state->past_us; // packets expected in this interval
+ ub_tx_count = ub_rx_count = (upper_thresh_pps * 1000000) / net_state->past_us;
+
+ tx_count = net_state->stats.tx_pkts - net_state->tx_pkts;
+ rx_count = net_state->stats.rx_pkts - net_state->rx_pkts;
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");
+ if(tx_count > ub_tx_count && net_state->tx_notify == 1) {
+ PrintDebug(core->vm_info, core, "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);
}
- if(tx_rate < RATE_LOWER_THRESHOLD && net_state->tx_notify == 0){
- V3_Print("Virtio NIC: Switch TX to Guest driven mode\n");
+ if(tx_count < lb_tx_count && net_state->tx_notify == 0) {
+ PrintDebug(core->vm_info, core, "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");
+ if(rx_count > ub_rx_count && net_state->rx_notify == 1) {
+ PrintDebug(core->vm_info, core, "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");
+ if(rx_count < lb_rx_count && net_state->rx_notify == 0) {
+ V3_Print(core->vm_info, core, "Virtio NIC: Switch RX to VMM notify mode\n");
net_state->rx_notify = 1;
}
net_state->past_us = 0;
}
-#endif
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",
+ PrintDebug(core->vm_info, core, "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,
.update_timer = virtio_nic_timer,
};
-
static int connect_fn(struct v3_vm_info * info,
void * frontend_data,
struct v3_dev_net_ops * ops,
struct virtio_dev_state * virtio = (struct virtio_dev_state *)frontend_data;
struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
+ if (!net_state) {
+ PrintError(info, VCORE_NONE, "Cannot allocate in connect\n");
+ return -1;
+ }
+
memset(net_state, 0, sizeof(struct virtio_net_state));
register_dev(virtio, net_state);
net_state->net_ops = ops;
net_state->backend_data = private_data;
net_state->virtio_dev = virtio;
- net_state->tx_notify = 0;
- net_state->rx_notify = 0;
-
- net_state->timer = v3_add_timer(&(info->cores[0]),&timer_ops,net_state);
+
+ switch (virtio->model) {
+ case GUEST_DRIVEN:
+ V3_Print(info, VCORE_NONE, "Virtio NIC: Guest-driven operation\n");
+ net_state->tx_notify = 1;
+ net_state->rx_notify = 1;
+ break;
+ case VMM_DRIVEN:
+ V3_Print(info, VCORE_NONE, "Virtio NIC: VMM-driven operation\n");
+ net_state->tx_notify = 0;
+ net_state->rx_notify = 0;
+ break;
+ case ADAPTIVE: {
+ V3_Print(info, VCORE_NONE, "Virtio NIC: Adaptive operation (begins in guest-driven mode)\n");
+ net_state->tx_notify = 1;
+ net_state->rx_notify = 1;
+
+ net_state->timer = v3_add_timer(&(info->cores[0]), &timer_ops,net_state);
+
+ }
+ break;
+
+ default:
+ V3_Print(info, VCORE_NONE, "Virtio NIC: Unknown model, using GUEST_DRIVEN\n");
+ net_state->tx_notify = 1;
+ net_state->rx_notify = 1;
+ break;
+ }
+
ops->recv = virtio_rx;
- 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);
+
+ if (!ops->config.fnt_mac) {
+ PrintError(info, VCORE_NONE, "Cannot allocate in connect\n");
+ // should unregister here
+ return -1;
+ }
+
+ memcpy(ops->config.fnt_mac, virtio->mac, ETH_ALEN);
+
+ return 0;
+}
+
+static int setup_perf_model(struct virtio_dev_state *virtio_state, v3_cfg_tree_t *t)
+{
+ char *mode = v3_cfg_val(t,"mode");
+
+ // defaults
+ virtio_state->model = GUEST_DRIVEN;
+ virtio_state->lower_thresh_pps = RATE_LOWER_THRESHOLD_DEFAULT;
+ virtio_state->upper_thresh_pps = RATE_UPPER_THRESHOLD_DEFAULT;
+ virtio_state->period_us = PROFILE_PERIOD_DEFAULT;
+
+
+ // overrides
+ if (mode) {
+ if (!strcasecmp(mode,"guest-driven")) {
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting static GUEST_DRIVEN mode of operation (latency optimized)\n");
+ virtio_state->model=GUEST_DRIVEN;
+ } else if (!strcasecmp(mode, "vmm-driven")) {
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting static VMM_DRIVEN mode of operation (throughput optimized)\n");
+ virtio_state->model=VMM_DRIVEN;
+ } else if (!strcasecmp(mode, "adaptive")) {
+ char *s;
+
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: Setting dynamic ADAPTIVE mode of operation\n");
+ virtio_state->model=ADAPTIVE;
+
+ if (!(s=v3_cfg_val(t,"upper"))) {
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No upper bound given, using default\n");
+ } else {
+ virtio_state->upper_thresh_pps = atoi(s);
+ }
+ if (!(s=v3_cfg_val(t,"lower"))) {
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No lower bound given, using default\n");
+ } else {
+ virtio_state->lower_thresh_pps = atoi(s);
+ }
+ if (!(s=v3_cfg_val(t,"period"))) {
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No period given, using default\n");
+ } else {
+ virtio_state->period_us = atoi(s);
+ }
- net_state->poll_thread = V3_CREATE_THREAD(virtio_tx_flush, (void *)net_state, "Virtio_Poll");
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: lower_thresh_pps=%llu, upper_thresh_pps=%llu, period_us=%llu\n",
+ virtio_state->lower_thresh_pps,
+ virtio_state->upper_thresh_pps,
+ virtio_state->period_us);
+ } else {
+ PrintError(VM_NONE, VCORE_NONE, "Virtio NIC: Unknown mode of operation '%s', using default (guest-driven)\n",mode);
+ virtio_state->model=GUEST_DRIVEN;
+ }
+ } else {
+ V3_Print(VM_NONE, VCORE_NONE, "Virtio NIC: No model given, using default (guest-driven)\n");
+ }
return 0;
+
}
+/*
+ <device class="LNX_VIRTIO_NIC" id="nic">
+ <bus>pci-bus-to-attach-to</bus> // required
+ <mac>mac address</mac> // if ommited with pic one
+ <model mode="guest-driven|vmm-driven|adaptive" upper="pkts_per_sec" lower="pkts" period="us" />
+ </device>
+*/
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[128];
- char * str = v3_cfg_val(cfg, "mac");
- memcpy(macstr, str, strlen(str));
-
+ char * mac = v3_cfg_val(cfg, "mac");
+ v3_cfg_tree_t *model = v3_cfg_subtree(cfg,"model");
+
if (pci_bus == NULL) {
- PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
+ PrintError(vm, VCORE_NONE, "Virtio NIC: Virtio device require a PCI Bus");
return -1;
}
virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
+
+ if (!virtio_state) {
+ PrintError(vm, VCORE_NONE, "Cannot allocate in init\n");
+ return -1;
+ }
+
memset(virtio_state, 0, sizeof(struct virtio_dev_state));
INIT_LIST_HEAD(&(virtio_state->dev_list));
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);
- }else {
+ if (mac) {
+ if (!str2mac(mac, virtio_state->mac)) {
+ PrintDebug(vm, VCORE_NONE, "Virtio NIC: Mac specified %s\n", mac);
+ } else {
+ PrintError(vm, VCORE_NONE, "Virtio NIC: Mac specified is incorrect, picking a randoom mac\n");
+ random_ethaddr(virtio_state->mac);
+ }
+ } else {
+ PrintDebug(vm, VCORE_NONE, "Virtio NIC: no mac specified, so picking a random mac\n");
random_ethaddr(virtio_state->mac);
}
+ if (setup_perf_model(virtio_state,model)<0) {
+ PrintError(vm, VCORE_NONE, "Cannnot setup performance model\n");
+ V3_Free(virtio_state);
+ return -1;
+ }
+
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);
+ PrintError(vm, VCORE_NONE, "Virtio NIC: Could not attach device %s\n", dev_id);
V3_Free(virtio_state);
return -1;
}
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);
+ PrintError(vm, VCORE_NONE, "Virtio NIC: Could not register %s as net frontend\n", dev_id);
v3_remove_device(dev);
return -1;
}
