* Copyright (c) 2010, Lei Xia <lxia@northwestern.edu>
* Copyright (c) 2009, Yuan Tang <ytang@northwestern.edu>
* Copyright (c) 2009, The V3VEE Project <http://www.v3vee.org>
- * All rights reserved.
+ * All rights reserved
*
* Author: Lei Xia <lxia@northwestern.edu>
* Yuan Tang <ytang@northwestern.edu>
#include <vnet/vnet_host.h>
#include <vnet/vnet_vmm.h>
+#include <palacios/vmm_queue.h>
+
#ifndef V3_CONFIG_DEBUG_VNET
-#undef Vnet_Debug
-#define Vnet_Debug(fmt, args...)
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
#endif
+#define VNET_YIELD_USEC 1000
+
int net_debug = 0;
struct eth_hdr {
uint8_t mac_addr[ETH_ALEN];
struct v3_vm_info * vm;
struct v3_vnet_dev_ops dev_ops;
+
+ int poll;
+
+#define VNET_MAX_QUOTE 64
+ int quote;
+
void * private_data;
struct list_head node;
uint32_t size_alloc;
};
-#define VNET_QUEUE_SIZE 1024
-struct vnet_queue {
- struct queue_entry buf[VNET_QUEUE_SIZE];
- int head, tail;
- int count;
- vnet_lock_t lock;
-};
static struct {
struct list_head routes;
vnet_lock_t lock;
struct vnet_stat stats;
- struct vnet_thread * pkt_flush_thread;
+ /* device queue that are waiting to be polled */
+ struct v3_queue * poll_devs;
- struct vnet_queue pkt_q;
+ struct vnet_thread * pkt_flush_thread;
struct hashtable * route_cache;
} vnet_state;
char str[50];
mac2str(route->src_mac, str);
- Vnet_Debug("Src Mac (%s), src_qual (%d)\n",
+ PrintDebug("Src Mac (%s), src_qual (%d)\n",
str, route->src_mac_qual);
mac2str(route->dst_mac, str);
- Vnet_Debug("Dst Mac (%s), dst_qual (%d)\n",
+ PrintDebug("Dst Mac (%s), dst_qual (%d)\n",
str, route->dst_mac_qual);
- Vnet_Debug("Src dev id (%d), src type (%d)",
+ PrintDebug("Src dev id (%d), src type (%d)",
route->src_id,
route->src_type);
- Vnet_Debug("Dst dev id (%d), dst type (%d)\n",
+ PrintDebug("Dst dev id (%d), dst type (%d)\n",
route->dst_id,
route->dst_type);
}
static void dump_routes(){
struct vnet_route_info *route;
- int i = 0;
- Vnet_Debug("\n========Dump routes starts ============\n");
+ PrintDebug("\n========Dump routes starts ============\n");
list_for_each_entry(route, &(vnet_state.routes), node) {
- Vnet_Debug("\nroute %d:\n", i++);
+ PrintDebug("\nroute %d:\n", route->idx);
print_route(&(route->route_def));
if (route->route_def.dst_type == LINK_INTERFACE) {
- Vnet_Debug("dst_dev (%p), dst_dev_id (%d), dst_dev_ops(%p), dst_dev_data (%p)\n",
+ PrintDebug("dst_dev (%p), dst_dev_id (%d), dst_dev_ops(%p), dst_dev_data (%p)\n",
route->dst_dev,
route->dst_dev->dev_id,
(void *)&(route->dst_dev->dev_ops),
}
}
- Vnet_Debug("\n========Dump routes end ============\n");
+ PrintDebug("\n========Dump routes end ============\n");
}
#endif
struct vnet_dev * dev = NULL;
list_for_each_entry(dev, &(vnet_state.devs), node) {
- int dev_id = dev->dev_id;
-
- if (dev_id == idx)
+ if (dev->dev_id == idx) {
return dev;
+ }
}
return NULL;
int v3_vnet_add_route(struct v3_vnet_route route) {
struct vnet_route_info * new_route = NULL;
- unsigned long flags;
+ vnet_intr_flags_t flags;
new_route = (struct vnet_route_info *)Vnet_Malloc(sizeof(struct vnet_route_info));
+
+ if (!new_route) {
+ PrintError("Cannot allocate new route\n");
+ return -1;
+ }
+
memset(new_route, 0, sizeof(struct vnet_route_info));
#ifdef V3_CONFIG_DEBUG_VNET
- Vnet_Debug("VNET/P Core: add_route_entry:\n");
+ PrintDebug("VNET/P Core: add_route_entry:\n");
print_route(&route);
#endif
void v3_vnet_del_route(uint32_t route_idx){
struct vnet_route_info * route = NULL;
- unsigned long flags;
+ vnet_intr_flags_t flags;
flags = vnet_lock_irqsave(vnet_state.lock);
list_for_each_entry(route, &(vnet_state.routes), node) {
+ Vnet_Print(0, "v3_vnet_del_route, route idx: %d\n", route->idx);
if(route->idx == route_idx){
list_del(&(route->node));
- list_del(&(route->match_node));
- Vnet_Free(route);
+ Vnet_Free(route);
+ break;
}
}
vnet_unlock_irqrestore(vnet_state.lock, flags);
+ clear_hash_cache();
+
+#ifdef V3_CONFIG_DEBUG_VNET
+ dump_routes();
+#endif
}
/* delete all route entries with specfied src or dst device id */
static void inline del_routes_by_dev(int dev_id){
- struct vnet_route_info * route = NULL;
- unsigned long flags;
+ struct vnet_route_info * route, *tmp_route;
+ vnet_intr_flags_t flags;
flags = vnet_lock_irqsave(vnet_state.lock);
- list_for_each_entry(route, &(vnet_state.routes), node) {
+ list_for_each_entry_safe(route, tmp_route, &(vnet_state.routes), node) {
if((route->route_def.dst_type == LINK_INTERFACE &&
route->route_def.dst_id == dev_id) ||
(route->route_def.src_type == LINK_INTERFACE &&
vnet_unlock_irqrestore(vnet_state.lock, flags);
}
+
+
+
/* At the end allocate a route_list
* This list will be inserted into the cache so we don't need to free it
*/
mac2str(hdr->src_mac, src_str);
mac2str(hdr->dst_mac, dst_str);
- Vnet_Debug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
+ PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str);
}
#endif
}
}
- Vnet_Debug("VNET/P Core: match_route: Matches=%d\n", num_matches);
+ PrintDebug("VNET/P Core: match_route: Matches=%d\n", num_matches);
- if (num_matches == 0) {
+ if (num_matches <= 0) {
return NULL;
}
matches = (struct route_list *)Vnet_Malloc(sizeof(struct route_list) +
(sizeof(struct vnet_route_info *) * num_matches));
+
+ if (!matches) {
+ PrintError("VNET/P Core: Unable to allocate matches\n");
+ return NULL;
+ }
+
matches->num_routes = num_matches;
{
return matches;
}
+int v3_vnet_query_header(uint8_t src_mac[6],
+ uint8_t dest_mac[6],
+ int recv, // 0 = send, 1=recv
+ struct v3_vnet_header *header)
+{
+ struct route_list *routes;
+ struct vnet_route_info *r;
+ struct v3_vnet_pkt p;
+
+ p.size=14;
+ p.data=p.header;
+ memcpy(p.header,dest_mac,6);
+ memcpy(p.header+6,src_mac,6);
+ memset(p.header+12,0,2);
+
+ p.src_type = LINK_EDGE;
+ p.src_id = 0;
+
+ memcpy(header->src_mac,src_mac,6);
+ memcpy(header->dst_mac,dest_mac,6);
+
+
+ look_into_cache(&p,&routes);
+
+ if (!routes) {
+ routes = match_route(&p);
+ if (!routes) {
+ PrintError("Cannot match route\n");
+ header->header_type=VNET_HEADER_NOMATCH;
+ header->header_len=0;
+ return -1;
+ } else {
+ add_route_to_cache(&p,routes);
+ }
+ }
+
+ if (routes->num_routes<1) {
+ PrintError("Less than one route\n");
+ header->header_type=VNET_HEADER_NOMATCH;
+ header->header_len=0;
+ return -1;
+ }
+
+ if (routes->num_routes>1) {
+ PrintError("More than one route, building header for the first one only\n");
+ }
+
+ r=routes->routes[0];
+
+ switch (r->route_def.dst_type) {
+ case LINK_EDGE: {
+ // switch based on the link type
+ // for mac-in-udp, we would want to generate a mac, ip, and udp header
+ // direct transmission
+
+ // for now we will say we have no encapsulation
+ //
+ header->header_type=VNET_HEADER_NONE;
+ header->header_len=0;
+ header->src_mac_qual=r->route_def.src_mac_qual;
+ header->dst_mac_qual=r->route_def.dst_mac_qual;
+
+ }
+
+ return 0;
+ break;
+
+
+ case LINK_INTERFACE:
+ // direct transmission
+ // let's guess that it goes to the same interface...
+ header->header_type=VNET_HEADER_NONE;
+ header->header_len=0;
+ header->src_mac_qual=r->route_def.src_mac_qual;
+ header->dst_mac_qual=r->route_def.dst_mac_qual;
+
+ return 0;
+ break;
+
+ default:
+ PrintError("Unknown destination type\n");
+ return -1;
+ break;
+
+ }
+
+}
+
+
-int vnet_tx_one_pkt(struct v3_vnet_pkt * pkt, void * private_data) {
+
+int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) {
struct route_list * matched_routes = NULL;
- unsigned long flags;
+ vnet_intr_flags_t flags;
int i;
int cpu = V3_Get_CPU();
+
Vnet_Print(2, "VNET/P Core: cpu %d: pkt (size %d, src_id:%d, src_type: %d, dst_id: %d, dst_type: %d)\n",
- cpu, pkt->size, pkt->src_id,
- pkt->src_type, pkt->dst_id, pkt->dst_type);
+ cpu, pkt->size, pkt->src_id,
+ pkt->src_type, pkt->dst_id, pkt->dst_type);
+
if(net_debug >= 4){
- v3_hexdump(pkt->data, pkt->size, NULL, 0);
+ v3_hexdump(pkt->data, pkt->size, NULL, 0);
}
flags = vnet_lock_irqsave(vnet_state.lock);
vnet_state.stats.rx_pkts++;
look_into_cache(pkt, &matched_routes);
+
if (matched_routes == NULL) {
- Vnet_Debug("VNET/P Core: send pkt Looking into routing table\n");
+ PrintDebug("VNET/P Core: sending pkt - matching route\n");
matched_routes = match_route(pkt);
if (matched_routes) {
add_route_to_cache(pkt, matched_routes);
} else {
- Vnet_Debug("VNET/P Core: Could not find route for packet... discards packet\n");
+ PrintDebug("VNET/P Core: Could not find route for packet... discarding packet\n");
vnet_unlock_irqrestore(vnet_state.lock, flags);
return 0; /* do we return -1 here?*/
}
vnet_unlock_irqrestore(vnet_state.lock, flags);
- Vnet_Debug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes);
+ PrintDebug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes);
for (i = 0; i < matched_routes->num_routes; i++) {
struct vnet_route_info * route = matched_routes->routes[i];
}
-static int vnet_pkt_enqueue(struct v3_vnet_pkt * pkt){
- unsigned long flags;
- struct queue_entry * entry;
- struct vnet_queue * q = &(vnet_state.pkt_q);
- uint16_t num_pages;
-
- flags = vnet_lock_irqsave(q->lock);
-
- if (q->count >= VNET_QUEUE_SIZE){
- Vnet_Print(1, "VNET Queue overflow!\n");
- vnet_unlock_irqrestore(q->lock, flags);
- return -1;
- }
-
- q->count ++;
- entry = &(q->buf[q->tail++]);
- q->tail %= VNET_QUEUE_SIZE;
-
- vnet_unlock_irqrestore(q->lock, flags);
-
- /* this is ugly, but should happen very unlikely */
- while(entry->use);
-
- if(entry->size_alloc < pkt->size){
- if(entry->data != NULL){
- Vnet_FreePages(Vnet_PAddr(entry->data), (entry->size_alloc / PAGE_SIZE));
- entry->data = NULL;
- }
-
- num_pages = 1 + (pkt->size / PAGE_SIZE);
- entry->data = Vnet_VAddr(Vnet_AllocPages(num_pages));
- if(entry->data == NULL){
- return -1;
- }
- entry->size_alloc = PAGE_SIZE * num_pages;
- }
-
- entry->pkt.data = entry->data;
- memcpy(&(entry->pkt), pkt, sizeof(struct v3_vnet_pkt));
- memcpy(entry->data, pkt->data, pkt->size);
-
- entry->use = 1;
-
- return 0;
-}
-
-
-int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data, int synchronize) {
- if(synchronize){
- vnet_tx_one_pkt(pkt, NULL);
- }else {
- vnet_pkt_enqueue(pkt);
- Vnet_Print(2, "VNET/P Core: Put pkt into Queue: pkt size %d\n", pkt->size);
- }
-
- return 0;
-}
-
int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac,
- struct v3_vnet_dev_ops *ops,
+ struct v3_vnet_dev_ops * ops, int quote, int poll_state,
void * priv_data){
struct vnet_dev * new_dev = NULL;
- unsigned long flags;
+ vnet_intr_flags_t flags;
new_dev = (struct vnet_dev *)Vnet_Malloc(sizeof(struct vnet_dev));
if (new_dev == NULL) {
- Vnet_Print(0, "Malloc fails\n");
+ Vnet_Print(0, "VNET/P Core: Unable to allocate a new device\n");
return -1;
}
memcpy(new_dev->mac_addr, mac, 6);
new_dev->dev_ops.input = ops->input;
+ new_dev->dev_ops.poll = ops->poll;
new_dev->private_data = priv_data;
new_dev->vm = vm;
new_dev->dev_id = 0;
+ new_dev->quote = quote<VNET_MAX_QUOTE ? quote : VNET_MAX_QUOTE;
+ new_dev->poll = poll_state;
flags = vnet_lock_irqsave(vnet_state.lock);
list_add(&(new_dev->node), &(vnet_state.devs));
new_dev->dev_id = ++ vnet_state.dev_idx;
vnet_state.num_devs ++;
+
+ if(new_dev->poll) {
+ v3_enqueue(vnet_state.poll_devs, (addr_t)new_dev);
+ }
+ } else {
+ PrintError("VNET/P: Device with the same MAC has already been added\n");
}
vnet_unlock_irqrestore(vnet_state.lock, flags);
return -1;
}
- Vnet_Debug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id);
+ PrintDebug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id);
return new_dev->dev_id;
}
int v3_vnet_del_dev(int dev_id){
struct vnet_dev * dev = NULL;
- unsigned long flags;
+ vnet_intr_flags_t flags;
flags = vnet_lock_irqsave(vnet_state.lock);
Vnet_Free(dev);
- Vnet_Debug("VNET/P Core: Remove Device: dev_id %d\n", dev_id);
+ PrintDebug("VNET/P Core: Removed Device: dev_id %d\n", dev_id);
return 0;
}
}
static void deinit_devices_list(){
- struct vnet_dev * dev = NULL;
+ struct vnet_dev * dev, * tmp;
- list_for_each_entry(dev, &(vnet_state.devs), node) {
+ list_for_each_entry_safe(dev, tmp, &(vnet_state.devs), node) {
list_del(&(dev->node));
Vnet_Free(dev);
}
}
static void deinit_routes_list(){
- struct vnet_route_info * route = NULL;
+ struct vnet_route_info * route, * tmp;
- list_for_each_entry(route, &(vnet_state.routes), node) {
+ list_for_each_entry_safe(route, tmp, &(vnet_state.routes), node) {
list_del(&(route->node));
list_del(&(route->match_node));
Vnet_Free(route);
struct v3_vnet_bridge_ops * ops,
uint8_t type,
void * priv_data) {
- unsigned long flags;
+ vnet_intr_flags_t flags;
int bridge_free = 0;
struct vnet_brg_dev * tmp_bridge = NULL;
tmp_bridge = (struct vnet_brg_dev *)Vnet_Malloc(sizeof(struct vnet_brg_dev));
if (tmp_bridge == NULL) {
- PrintError("Malloc Fails\n");
+ PrintError("VNET/P Core: Unable to allocate new bridge\n");
vnet_state.bridge = NULL;
return -1;
}
void v3_vnet_del_bridge(uint8_t type) {
- unsigned long flags;
+ vnet_intr_flags_t flags;
struct vnet_brg_dev * tmp_bridge = NULL;
flags = vnet_lock_irqsave(vnet_state.lock);
if (vnet_state.bridge != NULL && vnet_state.bridge->type == type) {
tmp_bridge = vnet_state.bridge;
- vnet_state.bridge = NULL;
+ vnet_state.bridge = NULL;
}
vnet_unlock_irqrestore(vnet_state.lock, flags);
}
-static int vnet_tx_flush(void *args){
- unsigned long flags;
- struct queue_entry * entry;
- struct vnet_queue * q = &(vnet_state.pkt_q);
+/* can be instanieoued to multiple threads
+ * that runs on multiple cores
+ * or it could be running on a dedicated side core
+ */
+static int vnet_tx_flush(void * args){
+ struct vnet_dev * dev = NULL;
+ int more;
+ int rc;
- Vnet_Print(0, "VNET/P Handing Pkt Thread Starting ....\n");
+ Vnet_Print(0, "VNET/P Polling Thread Starting ....\n");
- /* we need thread sleep/wakeup in Palacios */
- while(!vnet_thread_should_stop()){
- flags = vnet_lock_irqsave(q->lock);
+ // since there are multiple instances of this thread, and only
+ // one queue of pollable devices, our model here will be to synchronize
+ // on that queue, removing devices as we go, and keeping them
+ // then putting them back on the queue when we are done
+ // in this way, multiple instances of this function will never
+ // be polling the same device at the same time
- if (q->count <= 0){
- vnet_unlock_irqrestore(q->lock, flags);
- Vnet_Yield();
- }else {
- q->count --;
- entry = &(q->buf[q->head++]);
- q->head %= VNET_QUEUE_SIZE;
+ struct v3_queue * tq = v3_create_queue();
- vnet_unlock_irqrestore(q->lock, flags);
+ if (!tq) {
+ PrintError("VNET/P polling thread cannot allocate queue\n");
+ return -1;
+ }
- /* this is ugly, but should happen very unlikely */
- while(!entry->use);
- vnet_tx_one_pkt(&(entry->pkt), NULL);
- /* asynchronizely release allocated memory for buffer entry here */
- entry->use = 0;
+ while (!vnet_thread_should_stop()) {
+
+ more=0; // will indicate if any device has more work for us to do
+
+ while ((dev = (struct vnet_dev *)v3_dequeue(vnet_state.poll_devs))) {
+ // we are handling this device
+ v3_enqueue(tq,(addr_t)dev);
+
+ if (dev->poll && dev->dev_ops.poll) {
+ // The device's poll function MUST NOT BLOCK
+ rc = dev->dev_ops.poll(dev->vm, dev->quote, dev->private_data);
+
+ if (rc<0) {
+ Vnet_Print(0, "VNET/P: poll from device %p error (ignoring) !\n", dev);
+ } else {
+ more |= rc;
+ }
+ }
+ }
+
+ while ((dev = (struct vnet_dev *)v3_dequeue(tq))) {
+ // now someone else can handle it
+ v3_enqueue(vnet_state.poll_devs, (addr_t)dev);
+ }
- Vnet_Print(2, "vnet_tx_flush: pkt (size %d)\n", entry->pkt.size);
+ // Yield regardless of whether we handled any devices - need
+ // to allow other threads to run
+ if (more) {
+ // we have more to do, so we want to get back asap
+ V3_Yield();
+ } else {
+ // put ourselves briefly to sleep if we we don't have more
+ V3_Yield_Timed(VNET_YIELD_USEC);
}
+
}
+ Vnet_Free(tq);
+
+ Vnet_Print(0, "VNET/P Polling Thread Done.\n");
+
return 0;
}
vnet_state.num_routes = 0;
if (vnet_lock_init(&(vnet_state.lock)) == -1){
- PrintError("VNET/P Core: Fails to initiate lock\n");
+ PrintError("VNET/P: Fails to initiate lock\n");
}
vnet_state.route_cache = vnet_create_htable(0, &hash_fn, &hash_eq);
if (vnet_state.route_cache == NULL) {
- PrintError("VNET/P Core: Fails to initiate route cache\n");
+ PrintError("VNET/P: Fails to initiate route cache\n");
return -1;
}
- vnet_lock_init(&(vnet_state.pkt_q.lock));
+ vnet_state.poll_devs = v3_create_queue();
- vnet_state.pkt_flush_thread = vnet_start_thread(vnet_tx_flush, NULL, "VNET_Pkts");
+ vnet_state.pkt_flush_thread = vnet_start_thread(vnet_tx_flush, NULL, "vnetd-1");
- Vnet_Debug("VNET/P Core is initiated\n");
+ PrintDebug("VNET/P is initiated\n");
return 0;
}
void v3_deinit_vnet(){
- vnet_lock_deinit(&(vnet_state.lock));
+ PrintDebug("Stopping flush thread\n");
+ // This will pause until the flush thread is gone
+ vnet_thread_stop(vnet_state.pkt_flush_thread);
+ // At this point there should be no lock-holder
+
+ Vnet_Free(vnet_state.poll_devs);
+
- deinit_devices_list();
+ PrintDebug("Deiniting Device List\n");
+ // close any devices we have open
+ deinit_devices_list();
+
+ PrintDebug("Deiniting Route List\n");
+ // remove any routes we have
deinit_routes_list();
+ PrintDebug("Freeing hash table\n");
+ // remove the hash table
vnet_free_htable(vnet_state.route_cache, 1, 1);
- Vnet_Free(vnet_state.bridge);
+
+
+ PrintDebug("Removing Bridge\n");
+ // remove bridge if it was added
+ if (vnet_state.bridge) {
+ Vnet_Free(vnet_state.bridge);
+ }
+
+ PrintDebug("Deleting lock\n");
+ // eliminate the lock
+ vnet_lock_deinit(&(vnet_state.lock));
+
}
