* and the University of New Mexico. You can find out more at
* http://www.v3vee.org
*
- * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
- * Copyright (c) 2008, Lei Xia <lxia@northwestern.edu>
- * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
+ * Copyright (c) 2010, Lei Xia <lxia@northwestern.edu>
+ * Copyright (c) 2010, Cui Zheng <cuizheng@cs.unm.edu>
+ * Copyright (c) 2010, The V3VEE Project <http://www.v3vee.org>
* All rights reserved.
*
- * Author: Jack Lange <jarusl@cs.northwestern.edu>
- * Lei Xia <lxia@northwestern.edu>
+ * Author: Lei Xia <lxia@northwestern.edu>
+ * Cui Zheng <cuizheng@cs.unm.edu>
*
*
* This is free software. You are permitted to use,
#include <palacios/vmm.h>
#include <palacios/vmm_dev_mgr.h>
#include <devices/lnx_virtio_pci.h>
-#include <devices/lnx_virtio_nic.h>
#include <palacios/vm_guest_mem.h>
-
+#include <palacios/vmm_sprintf.h>
+#include <palacios/vmm_vnet.h>
+#include <palacios/vmm_lock.h>
+#include <palacios/vmm_util.h>
#include <devices/pci.h>
-#ifndef CONFIG_DEBUG_VIRTIO_BLK
+#ifndef CONFIG_DEBUG_VIRTIO_NET
#undef PrintDebug
#define PrintDebug(fmt, args...)
#endif
-#define NIC_STATUS_OK 0
-#define NIC_STATUS_ERR 1
-#define NIC_STATUS_NOT_SUPPORTED 2
-
-
-/* The feature bitmap for virtio net */
-#define VIRTIO_NET_F_CSUM 0 /* Host handles pkts w/ partial csum */
-#define VIRTIO_NET_F_GUEST_CSUM 1 /* Guest handles pkts w/ partial csum */
-#define VIRTIO_NET_F_MAC 5 /* Host has given MAC address. */
-#define VIRTIO_NET_F_GSO 6 /* Host handles pkts w/ any GSO type */
-#define VIRTIO_NET_F_GUEST_TSO4 7 /* Guest can handle TSOv4 in. */
-#define VIRTIO_NET_F_GUEST_TSO6 8 /* Guest can handle TSOv6 in. */
-#define VIRTIO_NET_F_GUEST_ECN 9 /* Guest can handle TSO[6] w/ ECN in. */
-#define VIRTIO_NET_F_GUEST_UFO 10 /* Guest can handle UFO in. */
-#define VIRTIO_NET_F_HOST_TSO4 11 /* Host can handle TSOv4 in. */
-#define VIRTIO_NET_F_HOST_TSO6 12 /* Host can handle TSOv6 in. */
-#define VIRTIO_NET_F_HOST_ECN 13 /* Host can handle TSO[6] w/ ECN in. */
-#define VIRTIO_NET_F_HOST_UFO 14 /* Host can handle UFO in. */
-#define VIRTIO_NET_F_MRG_RXBUF 15 /* Host can merge receive buffers. */
-#define VIRTIO_NET_F_STATUS 16 /* virtio_net_config.status available */
-#define VIRTIO_NET_F_CTRL_VQ 17 /* Control channel available */
-#define VIRTIO_NET_F_CTRL_RX 18 /* Control channel RX mode support */
-#define VIRTIO_NET_F_CTRL_VLAN 19 /* Control channel VLAN filtering */
-#define VIRTIO_NET_F_CTRL_RX_EXTRA 20 /* Extra RX mode control support */
#define VIRTIO_NET_S_LINK_UP 1 /* Link is up */
-
-/* Maximum packet size we can receive from tap device: header + 64k */
#define VIRTIO_NET_MAX_BUFSIZE (sizeof(struct virtio_net_hdr) + (64 << 10))
-
struct virtio_net_hdr {
-#define VIRTIO_NET_HDR_F_NEEDS_CSUM 1 /* Use csum_start, csum_offset */
uint8_t flags;
#define VIRTIO_NET_HDR_GSO_NONE 0 /* Not a GSO frame */
-#define VIRTIO_NET_HDR_GSO_TCPV4 1 /* GSO frame, IPv4 TCP (TSO) */
-#define VIRTIO_NET_HDR_GSO_UDP 3 /* GSO frame, IPv4 UDP (UFO) */
-#define VIRTIO_NET_HDR_GSO_TCPV6 4 /* GSO frame, IPv6 TCP */
-#define VIRTIO_NET_HDR_GSO_ECN 0x80 /* TCP has ECN set */
uint8_t gso_type;
-
uint16_t hdr_len; /* Ethernet + IP + tcp/udp hdrs */
uint16_t gso_size; /* Bytes to append to hdr_len per frame */
- uint16_t csum_start; /* Position to start checksumming from */
- uint16_t csum_offset; /* Offset after that to place checksum */
+ uint16_t csum_start; /* Position to start checksumming from */
+ uint16_t csum_offset; /* Offset after that to place checksum */
}__attribute__((packed));
-
-
-
-#define QUEUE_SIZE 256
-#define CTRL_QUEUE_SIZE 64
-
-struct v3_net_ops {
- int (*send)(uint8_t * buf, uint32_t count, void * private_data);
- int (*receive)(uint8_t * buf, uint32_t count, void * private_data);
+/* 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 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. */
+
struct virtio_net_config
{
- uint8_t mac[ETH_ALEN];
- // See VIRTIO_NET_F_STATUS and VIRTIO_NET_S_* above
+ uint8_t mac[ETH_ALEN]; /* VIRTIO_NET_F_MAC */
uint16_t status;
} __attribute__((packed));
+struct virtio_dev_state {
+ struct vm_device * pci_bus;
+ struct list_head dev_list;
+ struct v3_vm_info *vm;
+};
+
struct virtio_net_state {
struct virtio_net_config net_cfg;
struct virtio_config virtio_cfg;
- struct vm_device * pci_bus;
- struct pci_device * pci_dev;
+ struct vm_device * dev;
+ struct pci_device * pci_dev;
+ int io_range_size;
- struct virtio_queue rx_vq; //index 0, rvq in Linux virtio driver, handle packet to guest
- struct virtio_queue tx_vq; //index 1, svq in Linux virtio driver, handle packet from guest
- struct virtio_queue ctrl_vq; //index 2, ctrol info from guest
+ 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 v3_net_ops * net_ops;
+ int buffed_rx;
+ int tx_disabled; /* stop TX pkts from guest */
+ uint16_t cur_notify_tx_idx; /*for used in update_tx_queue */
- int io_range_size;
+ uint64_t pkt_sent, pkt_recv, pkt_drop;
+ uint64_t tx_stop_times, rx_stop_times, tx_poll_times, rx_ipi_num;
+
+ struct v3_dev_net_ops * net_ops;
+ v3_lock_t rx_lock, tx_lock;
- void *private_data;
+ 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 -1;
+ return 0;
}
-static int virtio_reset(struct vm_device * dev)
+static int virtio_init_state(struct virtio_net_state * virtio)
{
- struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data;
+ virtio->rx_vq.queue_size = RX_QUEUE_SIZE;
+ virtio->tx_vq.queue_size = TX_QUEUE_SIZE;
+ virtio->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
virtio->rx_vq.ring_desc_addr = 0;
virtio->rx_vq.ring_avail_addr = 0;
virtio->ctrl_vq.pfn = 0;
virtio->ctrl_vq.cur_avail_idx = 0;
- virtio->virtio_cfg.status = VIRTIO_NET_S_LINK_UP;
virtio->virtio_cfg.pci_isr = 0;
- virtio->private_data = NULL;
-
- return 0;
-}
-
-static int read_op(struct vm_device * dev, uint8_t * buf, uint32_t len)
-{
- struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data;
- int ret = -1;
-
- PrintDebug("Receving pkt from guest\n");
-
- ret = virtio->net_ops->receive(buf, len, virtio->private_data);
- return ret;
-}
-
-
-static int write_op(struct vm_device *dev, uint8_t *buf, uint32_t len)
-{
- struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data;
- int ret = -1;
+
+ virtio->virtio_cfg.host_features = 0; // (1 << VIRTIO_NET_F_MAC);
- PrintDebug("Receving pkt from guest\n");
+ 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");
+ }
- ret = virtio->net_ops->send(buf, len, virtio->private_data);
+ virtio->pkt_sent = virtio->pkt_recv = virtio->pkt_drop = 0;
+ virtio->buffed_rx = 0;
- return ret;
+ return 0;
}
-
-//sending guest's packet to network sink
-static int handle_pkt_write(struct vm_device *dev, struct virtio_net_hdr *hdr,
- struct vring_desc *buf_desc, uint8_t *status)
+static int pkt_tx(struct guest_info *core, struct virtio_net_state * virtio, struct vring_desc * buf_desc)
{
- //struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data;
uint8_t * buf = NULL;
+ uint32_t len = buf_desc->length;
- PrintDebug("Handling Virtio Net write\n");
-
- if (guest_pa_to_host_va(dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
+ if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
PrintError("Could not translate buffer address\n");
- return -1;
+ return -ERR_VIRTIO_OTHER;
}
- PrintDebug("Length=%d\n", buf_desc->length);
-
- if (write_op(dev, buf, buf_desc->length) == -1) {
- *status = NIC_STATUS_ERR;
- return -1;
- } else {
- *status = NIC_STATUS_OK;
- }
-
- PrintDebug("Returning Status: %d\n", *status);
-
- return 0;
+ return virtio->net_ops->send(buf, len, virtio->backend_data, NULL);
}
-
-//get packet from network, and send to guest
-static int handle_pkt_read(struct vm_device *dev, struct virtio_net_hdr *hdr,
- struct vring_desc *buf_desc, uint8_t *status)
+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)
{
- //struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data;
- uint8_t * buf = NULL;
-
- PrintDebug("Handling Virtio Net read\n");
+ uint32_t len;
+ uint8_t * desc_buf = NULL;
- if (guest_pa_to_host_va(dev->vm, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) {
+ if (v3_gpa_to_hva(core, desc->addr_gpa, (addr_t *)&(desc_buf)) == -1) {
PrintError("Could not translate buffer address\n");
return -1;
}
+ len = (desc->length < buf_len)?(desc->length - offset):buf_len;
+ memcpy(desc_buf+offset, buf, len);
- PrintDebug("Length=%d\n", buf_desc->length);
-
- if (read_op(dev, buf, buf_desc->length) == -1) {
- *status = NIC_STATUS_ERR;
- return -1;
- } else {
- *status = NIC_STATUS_OK;
- }
-
- PrintDebug("Returning Status: %d\n", *status);
-
- return 0;
+ return len;
}
-static int get_desc_count(struct virtio_queue * q, int index)
-{
+
+static int get_desc_count(struct virtio_queue * q, int index) {
struct vring_desc * tmp_desc = &(q->desc[index]);
int cnt = 1;
return cnt;
}
+static inline void enable_cb(struct virtio_queue *queue){
+ queue->used->flags &= ~ VRING_NO_NOTIFY_FLAG;
+}
-static int handle_ctrl(struct vm_device * dev) {
-
+static inline void disable_cb(struct virtio_queue *queue) {
+ queue->used->flags |= VRING_NO_NOTIFY_FLAG;
+}
- return 0;
+/* interrupt the guest, so the guest core get EXIT to Palacios
+ * this happens when there are either incoming pkts for the guest
+ * or the guest can start TX pkts again */
+static inline void notify_guest(struct virtio_net_state * virtio){
+ v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0);
}
-// TODO: handle receiving, not done yet
-//send packet to guest
-static int handle_pkt_rx(struct vm_device * dev)
+
+/* guest free some pkts from rx queue */
+static int handle_rx_kick(struct guest_info *core, struct virtio_net_state * virtio)
{
+ unsigned long flags;
- if (handle_pkt_read(dev, NULL, 0, NULL) == -1) {
- PrintError("Error handling nic operation\n");
- return -1;
- }
+ 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;
}
-//get packet from guest
-static int handle_pkt_tx(struct vm_device * dev)
+#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) {
+
+ return 0;
+}
+
+static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virtio_state)
{
- struct virtio_net_state *virtio = (struct virtio_net_state *)dev->private_data;
- struct virtio_queue *q = &(virtio->rx_vq);
+ struct virtio_queue * q = &(virtio_state->tx_vq);
+ struct virtio_net_hdr * hdr = NULL;
+ int recved = 0;
+ unsigned long flags;
+
+ if (!q->ring_avail_addr)
+ return -ERR_VIRTIO_TXQ_NOSET;
- PrintDebug("VIRTIO NIC KICK: cur_index=%d (mod=%d), avail_index=%d\n",
- q->cur_avail_idx, q->cur_avail_idx % QUEUE_SIZE, q->avail->index);
+ if(virtio_state->tx_disabled)
+ return -ERR_VIRTIO_TXQ_DISABLED;
- while (q->cur_avail_idx < q->avail->index) {
+ flags = v3_lock_irqsave(virtio_state->tx_lock);
+ while (q->cur_avail_idx != q->avail->index) {
struct vring_desc * hdr_desc = NULL;
- struct vring_desc * buf_desc = NULL;
- struct vring_desc * status_desc = NULL;
- struct virtio_net_hdr hdr;
addr_t hdr_addr = 0;
- uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
+ uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
int desc_cnt = get_desc_count(q, desc_idx);
- int i = 0;
- uint8_t * status_ptr = NULL;
- uint8_t status = NIC_STATUS_OK;
uint32_t req_len = 0;
-
- PrintDebug("Descriptor Count=%d, index=%d\n", desc_cnt, q->cur_avail_idx % QUEUE_SIZE);
+ int i = 0;
hdr_desc = &(q->desc[desc_idx]);
-
- PrintDebug("Header Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", hdr_desc,
- (void *)(hdr_desc->addr_gpa), hdr_desc->length, hdr_desc->flags, hdr_desc->next);
-
- if (guest_pa_to_host_va(dev->vm, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
+ if (v3_gpa_to_hva(core, hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
PrintError("Could not translate block header address\n");
- return -1;
+ goto exit_error;
}
- // We copy the block op header out because we are going to modify its contents
- memcpy(&hdr, (void *)hdr_addr, sizeof(struct virtio_net_hdr));
-
- PrintDebug("NIC Op Hdr (ptr=%p) type=%d, sector=%p\n", (void *)hdr_addr, hdr.hdr_len, (void *)hdr.csum_start);
-
+ hdr = (struct virtio_net_hdr*)hdr_addr;
desc_idx = hdr_desc->next;
- for (i = 0; i < desc_cnt - 2; i++) {
- uint8_t tmp_status = NIC_STATUS_OK;
-
- buf_desc = &(q->desc[desc_idx]);
-
- PrintDebug("Buffer Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", buf_desc,
- (void *)(buf_desc->addr_gpa), buf_desc->length, buf_desc->flags, buf_desc->next);
+ if(desc_cnt > 2){
+ PrintError("VNIC: merged rx buffer not supported\n");
+ goto exit_error;
+ }
- if (handle_pkt_write(dev, &hdr, buf_desc, &tmp_status) == -1) {
+ /* here we assumed that one ethernet pkt is not splitted into multiple virtio buffer */
+ for (i = 0; i < desc_cnt - 1; i++) {
+ struct vring_desc * buf_desc = &(q->desc[desc_idx]);
+ if (pkt_tx(core, virtio_state, buf_desc) == -1) {
PrintError("Error handling nic operation\n");
- return -1;
- }
-
- if (tmp_status != NIC_STATUS_OK) {
- status = tmp_status;
+ goto exit_error;
}
req_len += buf_desc->length;
desc_idx = buf_desc->next;
}
+ virtio_state->pkt_sent ++;
+ recved ++;
- status_desc = &(q->desc[desc_idx]);
+ q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
+ q->used->ring[q->used->index % q->queue_size].length = req_len; // What do we set this to????
+ q->used->index ++;
+
+ q->cur_avail_idx ++;
+ }
- PrintDebug("Status Descriptor (ptr=%p) gpa=%p, len=%d, flags=%x, next=%d\n", status_desc,
- (void *)(status_desc->addr_gpa), status_desc->length, status_desc->flags, status_desc->next);
+ v3_unlock_irqrestore(virtio_state->tx_lock, flags);
- if (guest_pa_to_host_va(dev->vm, status_desc->addr_gpa, (addr_t *)&(status_ptr)) == -1) {
- PrintError("Could not translate status address\n");
- return -1;
- }
+ if(!recved)
+ return 0;
+
+ if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
+ v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
+ virtio_state->virtio_cfg.pci_isr = 0x1;
+ }
- req_len += status_desc->length;
- *status_ptr = status;
- q->used->ring[q->used->index % QUEUE_SIZE].id = q->avail->ring[q->cur_avail_idx % QUEUE_SIZE];
- q->used->ring[q->used->index % QUEUE_SIZE].length = req_len; // What do we set this to????
+#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
- q->used->index++;
- q->cur_avail_idx++;
- }
-
- 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->pci_bus, 0, virtio->pci_dev);
- virtio->virtio_cfg.pci_isr = 0x1;
- }
return 0;
+
+exit_error:
+
+ v3_unlock_irqrestore(virtio_state->tx_lock, flags);
+ return -ERR_VIRTIO_OTHER;
}
-static int virtio_setup_queue(struct vm_device * dev, struct virtio_queue *queue, addr_t pfn, addr_t page_addr)
-{
+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) {
queue->pfn = pfn;
- queue->ring_desc_addr = page_addr ;
- queue->ring_avail_addr = page_addr + (QUEUE_SIZE * sizeof(struct vring_desc));
- queue->ring_used_addr = (queue->ring_avail_addr + \
- sizeof(struct vring_avail) + \
- (QUEUE_SIZE * sizeof(uint16_t)));
-
+ queue->ring_desc_addr = page_addr;
+ queue->ring_avail_addr = page_addr + (queue->queue_size * sizeof(struct vring_desc));
+ queue->ring_used_addr = ((queue->ring_avail_addr) +
+ (sizeof(struct vring_avail)) +
+ (queue->queue_size * sizeof(uint16_t)));
+
// round up to next page boundary.
queue->ring_used_addr = (queue->ring_used_addr + 0xfff) & ~0xfff;
-
- if (guest_pa_to_host_va(dev->vm, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
+ if (v3_gpa_to_hva(core, queue->ring_desc_addr, (addr_t *)&(queue->desc)) == -1) {
PrintError("Could not translate ring descriptor address\n");
return -1;
}
-
- if (guest_pa_to_host_va(dev->vm, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
+ if (v3_gpa_to_hva(core, queue->ring_avail_addr, (addr_t *)&(queue->avail)) == -1) {
PrintError("Could not translate ring available address\n");
return -1;
}
-
- if (guest_pa_to_host_va(dev->vm, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
+ if (v3_gpa_to_hva(core, queue->ring_used_addr, (addr_t *)&(queue->used)) == -1) {
PrintError("Could not translate ring used address\n");
return -1;
}
PrintDebug("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));
-
+ (void *)(queue->ring_desc_addr),
+ (void *)(queue->ring_avail_addr),
+ (void *)(queue->ring_used_addr));
+
PrintDebug("RingDesc=%p, Avail=%p, Used=%p\n",
- queue->desc, queue->avail, queue->used);
-
+ queue->desc, queue->avail, queue->used);
+
return 0;
}
-
-
-static int virtio_io_write(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 vm_device * dev = (struct vm_device *)private_data;
- struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data;
+ struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
int port_idx = port % virtio->io_range_size;
- PrintDebug("VIRTIO NIC Write for port %d (index=%d) len=%d, value=%x\n",
+ 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:
if (length != 4) {
PrintError("Illegal write length for guest features\n");
return -1;
- }
-
+ }
virtio->virtio_cfg.guest_features = *(uint32_t *)src;
- PrintDebug("Setting Guest Features to %x\n", virtio->virtio_cfg.guest_features);
-
break;
+
case VRING_PG_NUM_PORT:
- if (length == 4) {
- addr_t pfn = *(uint32_t *)src;
- addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
-
- uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
- switch (queue_idx) {
- case 0:
- virtio_setup_queue(dev, &virtio->rx_vq, pfn, page_addr);
- break;
- case 1:
- virtio_setup_queue(dev, &virtio->tx_vq, pfn, page_addr);
- break;
- case 2:
- virtio_setup_queue(dev, &virtio->ctrl_vq, pfn, page_addr);
- break;
-
- default:
- break;
- }
- } else {
+ if (length != 4) {
PrintError("Illegal write length for page frame number\n");
return -1;
}
+ addr_t pfn = *(uint32_t *)src;
+ addr_t page_addr = (pfn << VIRTIO_PAGE_SHIFT);
+ uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
+ switch (queue_idx) {
+ case 0:
+ virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
+ disable_cb(&virtio->rx_vq);
+ 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);
+ break;
+ default:
+ break;
+ }
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 device only uses 3 queue, selected %d\n",
+ PrintError("Virtio NIC: wrong queue idx: %d\n",
virtio->virtio_cfg.vring_queue_selector);
return -1;
}
-
break;
- case VRING_Q_NOTIFY_PORT:
- PrintDebug("Handling Kick\n");
- uint16_t queue_idx = *(uint16_t *)src;
- if (queue_idx == 0){
- if (handle_pkt_rx(dev) == -1) {
- PrintError("Could not handle NIC Notification\n");
- return -1;
- }
- }else if (queue_idx == 1){
- if (handle_pkt_tx(dev) == -1) {
+
+ case VRING_Q_NOTIFY_PORT:
+ {
+ uint16_t queue_idx = *(uint16_t *)src;
+ if (queue_idx == 0){
+ handle_rx_kick(core, virtio);
+ PrintError("rx kick\n");
+ } else if (queue_idx == 1){
+ PrintError("tx kick\n");
+ if (handle_pkt_tx(core, virtio) == -1) {
PrintError("Could not handle NIC Notification\n");
return -1;
}
- }else if (queue_idx == 2){
- if (handle_ctrl(dev) == -1) {
+ } else if (queue_idx == 2){
+ if (handle_ctrl(core, virtio) == -1) {
PrintError("Could not handle NIC Notification\n");
return -1;
}
- }else {
- PrintError("Virtio NIC device only uses 3 queue, selected %d\n",
- queue_idx);
+ } else {
+ PrintError("Wrong queue index %d\n", queue_idx);
+ }
+ break;
}
-
- break;
+
case VIRTIO_STATUS_PORT:
virtio->virtio_cfg.status = *(uint8_t *)src;
-
if (virtio->virtio_cfg.status == 0) {
PrintDebug("Resetting device\n");
- virtio_reset(dev);
+ virtio_init_state(virtio);
}
-
break;
case VIRTIO_ISR_PORT:
virtio->virtio_cfg.pci_isr = *(uint8_t *)src;
break;
+
default:
return -1;
break;
return length;
}
-
-static int virtio_io_read(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 vm_device * dev = (struct vm_device *)private_data;
- struct virtio_net_state * virtio = (struct virtio_net_state *)dev->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 Read for port %d (index =%d), length=%d\n",
+ PrintDebug("Virtio NIC %p: Read for port %d (index =%d), length=%d\n", private_data,
port, port_idx, length);
-
+
switch (port_idx) {
case HOST_FEATURES_PORT:
if (length != 4) {
PrintError("Illegal read length for host features\n");
return -1;
}
-
*(uint32_t *)dst = virtio->virtio_cfg.host_features;
-
break;
- case VRING_PG_NUM_PORT:
+ case VRING_PG_NUM_PORT:
if (length != 4) {
PrintError("Illegal read length for page frame number\n");
return -1;
}
-
switch (queue_idx) {
case 0:
- *(uint32_t *)dst = virtio->rx_vq.pfn;
- break;
- case 1:
- *(uint32_t *)dst = virtio->tx_vq.pfn;
- break;
- case 2:
- *(uint32_t *)dst = virtio->ctrl_vq.pfn;
- break;
- default:
- break;
- }
-
+ *(uint32_t *)dst = virtio->rx_vq.pfn;
+ break;
+ case 1:
+ *(uint32_t *)dst = virtio->tx_vq.pfn;
+ break;
+ case 2:
+ *(uint32_t *)dst = virtio->ctrl_vq.pfn;
+ break;
+ default:
+ break;
+ }
break;
+
case VRING_SIZE_PORT:
if (length != 2) {
PrintError("Illegal read length for vring size\n");
return -1;
}
-
switch (queue_idx) {
case 0:
- *(uint16_t *)dst = virtio->rx_vq.queue_size;
- break;
- case 1:
- *(uint32_t *)dst = virtio->tx_vq.queue_size;
- break;
- case 2:
- *(uint32_t *)dst = virtio->ctrl_vq.queue_size;
- break;
- default:
- break;
- }
-
+ *(uint16_t *)dst = virtio->rx_vq.queue_size;
+ break;
+ case 1:
+ *(uint16_t *)dst = virtio->tx_vq.queue_size;
+ break;
+ case 2:
+ *(uint16_t *)dst = virtio->ctrl_vq.queue_size;
+ break;
+ default:
+ break;
+ }
break;
+
case VIRTIO_STATUS_PORT:
if (length != 1) {
PrintError("Illegal read length for status\n");
return -1;
}
-
*(uint8_t *)dst = virtio->virtio_cfg.status;
break;
-
+
case VIRTIO_ISR_PORT:
*(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
virtio->virtio_cfg.pci_isr = 0;
- v3_pci_lower_irq(virtio->pci_bus, 0, virtio->pci_dev);
+ v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
break;
default:
- PrintError("Read of Unhandled Virtio Read\n");
- return -1;
+ PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx);
+ return -1;
}
return length;
}
+static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) {
+ struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
+ struct virtio_queue * q = &(virtio->rx_vq);
+ struct virtio_net_hdr_mrg_rxbuf hdr;
+ uint32_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+ uint32_t data_len = size;
+ uint32_t offset = 0;
+ unsigned long flags;
+ int ret_val = -ERR_VIRTIO_OTHER;
+ int raw = 1;
+
+ 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));
+
+ if (q->ring_avail_addr == 0) {
+ PrintError("Queue is not set\n");
+ ret_val = -ERR_VIRTIO_RXQ_NOSET;
+ goto exit;
+ }
+
+ 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;
+
+ hdr_desc = &(q->desc[hdr_idx]);
+ if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
+ PrintError("Could not translate receive buffer address\n");
+ goto exit;
+ }
+ hdr.num_buffers = 1;
+ memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr_mrg_rxbuf));
+ if (offset >= data_len) {
+ hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
+ }
+
+ struct vring_desc * buf_desc = NULL;
+ for (buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) {
+ uint32_t len = 0;
+ 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;
+ }
+ buf_desc->length = len;
+ }
+ buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
+
+ q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
+ q->used->ring[q->used->index % q->queue_size].length = data_len + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */
+ q->used->index++;
+ q->cur_avail_idx++;
+
+ /* if there are certain num of pkts in the RX queue, notify guest
+ * so guest will exit to palacios
+ * when it returns, guest gets the virtio rx interrupt */
+ if((++virtio->buffed_rx > q->queue_size/5) &&
+ (q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
+ if(virtio->virtio_dev->vm->cores[0].cpu_id != V3_Get_CPU()){
+ notify_guest(virtio);
+ virtio->rx_ipi_num ++;
+ }
+ virtio->buffed_rx = 0;
+ }
+ } 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 ++;
+
+ ret_val = -ERR_VIRTIO_RXQ_FULL;
+ goto 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;
+ }
+
+ ret_val = offset;
+
+exit:
+
+ v3_unlock_irqrestore(virtio->rx_lock, flags);
+
+ return ret_val;
+}
+
+
static struct v3_device_ops dev_ops = {
.free = virtio_free,
- .reset = virtio_reset,
+ .reset = NULL,
.start = NULL,
.stop = NULL,
};
-int v3_virtio_register_nic(struct vm_device *dev, struct v3_net_ops *ops, void *private_data) {
- struct virtio_net_state * virtio = (struct virtio_net_state *)dev->private_data;
-
- virtio->net_ops = ops;
+/* TODO: Issue here: which vm info it needs? calling VM or the device's own VM? */
+static void virtio_nic_poll(struct v3_vm_info *vm, void *data){
+ struct virtio_net_state *virtio = (struct virtio_net_state *)data;
+
+ handle_pkt_tx(&(vm->cores[0]), virtio);
- return 0;
-}
+ virtio->tx_poll_times ++;
+#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);
+ }
+#endif
+}
-static int virtio_init(struct guest_info * vm, void *cfg_data) {
- struct vm_device * pci_bus = v3_find_dev(vm, (char *)cfg_data);
- struct virtio_net_state * virtio_state = NULL;
- struct pci_device * pci_dev = NULL;
+static void virtio_start_tx(void *data){
+ struct virtio_net_state * virtio = (struct virtio_net_state *)data;
- PrintDebug("Initializing VIRTIO Network device\n");
+ /* do we need a lock here? */
+ virtio->tx_disabled = 0;
- if (pci_bus == NULL) {
- PrintError("VirtIO network devices require a PCI Bus");
- return -1;
+ /* 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);
}
-
- virtio_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
- memset(virtio_state, 0, sizeof(struct virtio_net_state));
+}
- struct vm_device * dev = v3_allocate_device("LNX_VIRTIO_NIC", &dev_ops, virtio_state);
- if (v3_attach_device(vm, dev) == -1) {
- PrintError("Could not attach device %s\n", "LNX_VIRTIO_NIC");
- return -1;
+static void virtio_stop_tx(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);
}
+ virtio->tx_stop_times ++;
+}
- // PCI initialization
- {
- struct v3_pci_bar bars[6];
- int num_ports = sizeof(struct virtio_config);
- int tmp_ports = num_ports;
- int i;
+
- // This gets the number of ports, rounded up to a power of 2
- virtio_state->io_range_size = 1; // must be a power of 2
- while (tmp_ports > 0) {
- tmp_ports >>= 1;
- virtio_state->io_range_size <<= 1;
- }
+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];
+ int num_ports = sizeof(struct virtio_config);
+ 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
+ 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 ((num_ports & ((virtio_state->io_range_size >> 1) - 1)) == 0) {
- virtio_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 ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
+ net_state->io_range_size >>= 1;
+ }
+
+ for (i = 0; i < 6; i++) {
+ bars[i].type = PCI_BAR_NONE;
+ }
+
+ 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;
+ bars[0].num_ports = net_state->io_range_size;
+ bars[0].io_read = virtio_io_read;
+ bars[0].io_write = virtio_io_write;
+ 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,
+ "LNX_VIRTIO_NIC", bars,
+ NULL, NULL, NULL, net_state);
+
+ if (!pci_dev) {
+ PrintError("Virtio NIC: Could not register PCI Device\n");
+ return -1;
+ }
- for (i = 0; i < 6; i++) {
- bars[i].type = PCI_BAR_NONE;
- }
+ PrintDebug("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;
+
- PrintDebug("Virtio-NIC io_range_size = %d\n", virtio_state->io_range_size);
+ pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
+ pci_dev->config_header.class = PCI_CLASS_NETWORK;
+ pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
+ pci_dev->config_header.subsystem_id = VIRTIO_NET_SUBDEVICE_ID;
+ pci_dev->config_header.intr_pin = 1;
+ pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
- bars[0].type = PCI_BAR_IO;
- bars[0].default_base_port = -1;
- bars[0].num_ports = virtio_state->io_range_size;
+ net_state->pci_dev = pci_dev;
+ net_state->virtio_dev = virtio;
- bars[0].io_read = virtio_io_read;
- bars[0].io_write = virtio_io_write;
- bars[0].private_data = dev;
+ 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);
+
+ virtio_init_state(net_state);
- pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE,
- 0, PCI_AUTO_DEV_NUM, 0,
- "LNX_VIRTIO_NIC", bars,
- NULL, NULL, NULL, dev, NULL);
+#if 0 //temporary hacking LX
+ vnic_states[num_vnic ++] = net_state;
+ PrintError("VNIC: num of vnic %d\n", num_vnic);
+#endif
- if (!pci_dev) {
- PrintError("Could not register PCI Device\n");
- return -1;
- }
-
- pci_dev->config_header.vendor_id = VIRTIO_VENDOR_ID;
- pci_dev->config_header.subsystem_vendor_id = VIRTIO_SUBVENDOR_ID;
-
+ return 0;
+}
- pci_dev->config_header.device_id = VIRTIO_NET_DEV_ID;
- pci_dev->config_header.class = PCI_CLASS_NETWORK;
- pci_dev->config_header.subclass = PCI_NET_SUBCLASS_OTHER;
+static int connect_fn(struct v3_vm_info * info,
+ void * frontend_data,
+ struct v3_dev_net_ops * ops,
+ v3_cfg_tree_t * cfg,
+ void * private_data) {
+ 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));
- // TODO:how to define new one for virtio net device
- pci_dev->config_header.subsystem_id = VIRTIO_BLOCK_SUBDEVICE_ID;
+ 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;
- pci_dev->config_header.intr_pin = 1;
+ 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;
- pci_dev->config_header.max_latency = 1; // ?? (qemu does it...)
+ return 0;
+}
+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");
- virtio_state->pci_dev = pci_dev;
- virtio_state->pci_bus = pci_bus;
- }
+ PrintDebug("Virtio NIC: Initializing VIRTIO Network device: %s\n", name);
- virtio_state->virtio_cfg.host_features = 0; //no features support now
+ if (pci_bus == NULL) {
+ PrintError("Virtio NIC: VirtIO devices require a PCI Bus");
+ return -1;
+ }
- virtio_state->rx_vq.queue_size = QUEUE_SIZE;
- virtio_state->tx_vq.queue_size = QUEUE_SIZE;
- virtio_state->ctrl_vq.queue_size = CTRL_QUEUE_SIZE;
-
+ virtio_state = (struct virtio_dev_state *)V3_Malloc(sizeof(struct virtio_dev_state));
+ memset(virtio_state, 0, sizeof(struct virtio_dev_state));
- virtio_reset(dev);
+ INIT_LIST_HEAD(&(virtio_state->dev_list));
+ virtio_state->pci_bus = pci_bus;
+ virtio_state->vm = vm;
-// TODO: net ops
- virtio_state->net_ops = NULL;
+ 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);
+ 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);
+ return -1;
+ }
+
return 0;
}
-
device_register("LNX_VIRTIO_NIC", virtio_init)
-
+