} __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;
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;
}
uint32_t len = buf_desc->length;
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;
}
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+dst_offset))?(desc->length - dst_offset):buf_len;
desc_cnt = get_desc_count(q, desc_idx);
if(desc_cnt != 2){
- PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
+ PrintError(core->vm_info, core, "VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
}
hdr_desc = &(q->desc[desc_idx]);
/* 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("Virtio NIC: Fails to send packet\n");
+ PrintError(core->vm_info, core, "Virtio NIC: Fails to send packet\n");
}
} else {
- PrintError("Could not translate block header address\n");
+ PrintError(core->vm_info, core, "Could not translate block header address\n");
}
flags = v3_lock_irqsave(virtio_state->tx_lock);
// 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",
+ 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;
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;
}
virtio->stats.tx_interrupts ++;
} else if (queue_idx == 1){
if (handle_pkt_tx(core, virtio, 0) < 0) {
- PrintError("Virtio NIC: Error to handle packet TX\n");
+ 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("Virtio NIC: 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",
+ 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("Virtio NIC: 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("Virtio NIC: 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("Virtio NIC: 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("Virtio NIC: 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;
break;
default:
- PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n",
+ PrintError(core->vm_info, core, "Virtio NIC: Read of Unhandled Virtio Read:%d\n",
port_idx);
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;
}
bars[i].type = PCI_BAR_NONE;
}
- PrintDebug("Virtio NIC: io_range_size = %d\n",
+ PrintDebug(VM_NONE, VCORE_NONE, "Virtio NIC: io_range_size = %d\n",
net_state->io_range_size);
bars[0].type = PCI_BAR_IO;
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("Virtio NIC: Registered Intr Line %d\n", pci_dev->config_header.intr_line);
+ 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;
}
-#if 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(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;
}
- 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;
}
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,
static struct v3_timer_ops timer_ops = {
.update_timer = virtio_nic_timer,
};
-#endif
static int connect_fn(struct v3_vm_info * info,
void * frontend_data,
struct virtio_net_state * net_state = (struct virtio_net_state *)V3_Malloc(sizeof(struct virtio_net_state));
if (!net_state) {
- PrintError("Cannot allocate in connect\n");
+ PrintError(info, VCORE_NONE, "Cannot allocate in connect\n");
return -1;
}
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);
+ 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->poll = virtio_poll;
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);
+ }
+
+ 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 device 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("Cannot allocate in init\n");
+ PrintError(vm, VCORE_NONE, "Cannot allocate in init\n");
return -1;
}
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;
}