X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fvnet%2Fvnet_core.c;h=70ba3826993c17953e51ab1db967f59f1f3795fc;hb=ea68fe042705986beaec92f33db0395b816c2e38;hp=46690294165b5f50eac65f501ba7ef2f8c24bba1;hpb=020108512c01b113a5e8cca88a8facdc5ba5488a;p=palacios.git diff --git a/palacios/src/vnet/vnet_core.c b/palacios/src/vnet/vnet_core.c index 4669029..70ba382 100644 --- a/palacios/src/vnet/vnet_core.c +++ b/palacios/src/vnet/vnet_core.c @@ -27,10 +27,17 @@ #include #ifndef V3_CONFIG_DEBUG_VNET -#undef Vnet_Debug -#define Vnet_Debug(fmt, args...) +#undef PrintDebug +#define PrintDebug(fmt, args...) #endif +#define VNET_NUM_TX_KICK_THREADS 1 + +#define VNET_ADAPTIVE_TX_KICK 1 // set to 1 to try to sleep when there is nothing to do +#define VNET_NOPROGRESS_LIMIT 1000 // ... after this many tries +#define VNET_YIELD_USEC 1000 // ... and go to sleep for this long + + int net_debug = 0; struct eth_hdr { @@ -53,7 +60,7 @@ struct vnet_dev { void * private_data; - struct list_head node; + struct list_head node; } __attribute__((packed)); @@ -116,9 +123,10 @@ static struct { /* device queue that are waiting to be polled */ struct v3_queue * poll_devs; - struct vnet_thread * pkt_flush_thread; + struct vnet_thread * pkt_flush_thread[VNET_NUM_TX_KICK_THREADS]; struct hashtable * route_cache; + } vnet_state; @@ -133,15 +141,15 @@ static void print_route(struct v3_vnet_route * route){ char str[50]; mac2str(route->src_mac, str); - Vnet_Debug("Src Mac (%s), src_qual (%d)\n", + PrintDebug(VM_NONE, VCORE_NONE, "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(VM_NONE, VCORE_NONE, "Dst Mac (%s), dst_qual (%d)\n", str, route->dst_mac_qual); - Vnet_Debug("Src dev id (%d), src type (%d)", + PrintDebug(VM_NONE, VCORE_NONE, "Src dev id (%d), src type (%d)", route->src_id, route->src_type); - Vnet_Debug("Dst dev id (%d), dst type (%d)\n", + PrintDebug(VM_NONE, VCORE_NONE, "Dst dev id (%d), dst type (%d)\n", route->dst_id, route->dst_type); } @@ -149,13 +157,13 @@ static void print_route(struct v3_vnet_route * route){ static void dump_routes(){ struct vnet_route_info *route; - Vnet_Debug("\n========Dump routes starts ============\n"); + PrintDebug(VM_NONE, VCORE_NONE, "\n========Dump routes starts ============\n"); list_for_each_entry(route, &(vnet_state.routes), node) { - Vnet_Debug("\nroute %d:\n", route->idx); + PrintDebug(VM_NONE, VCORE_NONE, "\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(VM_NONE, VCORE_NONE, "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), @@ -163,7 +171,7 @@ static void dump_routes(){ } } - Vnet_Debug("\n========Dump routes end ============\n"); + PrintDebug(VM_NONE, VCORE_NONE, "\n========Dump routes end ============\n"); } #endif @@ -187,7 +195,7 @@ static int add_route_to_cache(const struct v3_vnet_pkt * pkt, struct route_list memcpy(routes->hash_buf, pkt->hash_buf, VNET_HASH_SIZE); if (vnet_htable_insert(vnet_state.route_cache, (addr_t)routes->hash_buf, (addr_t)routes) == 0) { - PrintError("VNET/P Core: Failed to insert new route entry to the cache\n"); + PrintError(VM_NONE, VCORE_NONE, "VNET/P Core: Failed to insert new route entry to the cache\n"); return -1; } @@ -213,10 +221,9 @@ static struct vnet_dev * dev_by_id(int idx) { 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; @@ -235,6 +242,10 @@ static struct vnet_dev * dev_by_mac(uint8_t * mac) { } +static int start_vnet_kick_threads(void); +static int stop_vnet_kick_threads(void); + + int v3_vnet_find_dev(uint8_t * mac) { struct vnet_dev * dev = NULL; @@ -250,13 +261,19 @@ int v3_vnet_find_dev(uint8_t * mac) { 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(VM_NONE, VCORE_NONE, "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(VM_NONE, VCORE_NONE, "VNET/P Core: add_route_entry:\n"); print_route(&route); #endif @@ -298,12 +315,12 @@ int v3_vnet_add_route(struct v3_vnet_route route) { 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) { - V3_Print("v3_vnet_del_route, route idx: %d\n", route->idx); + Vnet_Print(0, "v3_vnet_del_route, route idx: %d\n", route->idx); if(route->idx == route_idx){ list_del(&(route->node)); Vnet_Free(route); @@ -322,12 +339,12 @@ void v3_vnet_del_route(uint32_t route_idx){ /* 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 && @@ -338,148 +355,363 @@ static void inline del_routes_by_dev(int dev_id){ Vnet_Free(route); } } - + 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 - */ -static struct route_list * match_route(const struct v3_vnet_pkt * pkt) { + +// Match classes, must be in order +#define NUM_MATCH_CLASSES 4 +#define NUM_MATCH_CLASSES_BOUND 3 +#define NONE 0 +#define NOT 1 +#define ANY 2 +#define DIRECT 3 + + +static inline uint8_t match_mac(uint8_t test_mac[ETH_ALEN], + uint8_t route_mac[ETH_ALEN], + uint8_t route_qual) +{ + switch (route_qual) { + case MAC_NOSET: + return NONE; + break; + case MAC_NONE: + return NONE; + break; + case MAC_ANY: + return ANY; + break; + case MAC_NOT: + if (memcmp(test_mac,route_mac,ETH_ALEN)) { + return NOT; + } else { + return NONE; + } + break; + case MAC_ADDR: + if (memcmp(test_mac,route_mac,ETH_ALEN)) { + return NONE; + } else { + return DIRECT; + } + break; + default: + PrintError(VM_NONE, VCORE_NONE, "Unknown qualifier %u\n",route_qual); + return NONE; + break; + } + +} + +#define QUAL_TO_STR(q) ( \ +(q)==MAC_NOSET ? "MAC_NOSET" : \ +(q)==MAC_NONE ? "MAC_NONE" : \ +(q)==MAC_ANY ? "MAC_ANY" : \ +(q)==MAC_NOT ? "MAC_NOT" : \ +(q)==MAC_ADDR ? "MAC_ADDR" : \ +"***UNDEFINED****" \ + ) \ + +#define MATCH_CLASS_TO_STR(c) ( \ +(c)==NONE ? "NONE" : \ +(c)==NOT ? "NOT" : \ +(c)==ANY ? "ANY" : \ +(c)==DIRECT ? "DIRECT" : \ +"***UNDEFINED****" \ + ) \ + + + +/* + +Original priority behavior... + +priority src srcqual dst dstqual +3 ANY ANY +4 X NONE +5 ANY X NOT +5 X NOT ANY +6 X ~NOT ANY +6 ANY X ~NOT +7 X ~NOT X NOT +7 X NOT X ~NOT +8 X ~NOT X ~NOT +8 X ~NOT X ~NOT + +*/ + +/* + Current priority order is given in the following table +*/ + +// [src][dst] => priority +static int priority_map[NUM_MATCH_CLASSES][NUM_MATCH_CLASSES] = +{ + [NONE] = { [ 0 ... NUM_MATCH_CLASSES_BOUND ] = -1}, // ignore if it's not a source match + [NOT][NONE] = -1, // ignore it if there is no destination match + [NOT][NOT] = 3, + [NOT][ANY] = 5, + [NOT][DIRECT] = 7, + [ANY][NONE] = -1, // ignore if there is no destination match + [ANY][NOT] = 5, + [ANY][ANY] = 6, + [ANY][DIRECT] = 6, + [DIRECT][NONE] = -1, // ignore if there is no destination match + [DIRECT][NOT] = 7, + [DIRECT][ANY] = 8, + [DIRECT][DIRECT] = 8, +}; + + + + +static inline int match_priority(uint8_t src_mac[ETH_ALEN], + uint8_t dst_mac[ETH_ALEN], + uint8_t route_src_mac[ETH_ALEN], + uint8_t route_src_qual, + uint8_t route_dst_mac[ETH_ALEN], + uint8_t route_dst_qual) + +{ + + return priority_map[match_mac(src_mac,route_src_mac,route_src_qual)][match_mac(dst_mac,route_dst_mac,route_dst_qual)]; +} + + +/* + Route matching will return the list of the highest priority routes that + match. It's a list because it's possible to have multiple high priority routes + */ +static struct route_list * match_route(const struct v3_vnet_pkt * pkt) +{ + int i; struct vnet_route_info * route = NULL; struct route_list * matches = NULL; int num_matches = 0; - int max_rank = 0; + int max_priority = -1; struct list_head match_list; struct eth_hdr * hdr = (struct eth_hdr *)(pkt->data); - // uint8_t src_type = pkt->src_type; - // uint32_t src_link = pkt->src_id; + // + // + // NOTE: USING THE MATCH_NODE in the route list to record a match list + // IS A DISASTER WAITING TO HAPPEN + // + #ifdef V3_CONFIG_DEBUG_VNET { - char dst_str[100]; - char src_str[100]; - + char dst_str[32], src_str[32]; 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(VM_NONE, VCORE_NONE, "VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str); } #endif - - INIT_LIST_HEAD(&match_list); -#define UPDATE_MATCHES(rank) do { \ - if (max_rank < (rank)) { \ - max_rank = (rank); \ - INIT_LIST_HEAD(&match_list); \ - \ - list_add(&(route->match_node), &match_list); \ - num_matches = 1; \ - } else if (max_rank == (rank)) { \ - list_add(&(route->match_node), &match_list); \ - num_matches++; \ - } \ - } while (0) + INIT_LIST_HEAD(&match_list); + - list_for_each_entry(route, &(vnet_state.routes), node) { + struct v3_vnet_route * route_def = &(route->route_def); + + int priority; + + priority = match_priority(hdr->src_mac, + hdr->dst_mac, + route_def->src_mac, + route_def->src_mac_qual, + route_def->dst_mac, + route_def->dst_mac_qual); -/* - // CHECK SOURCE TYPE HERE - if ( (route_def->src_type != LINK_ANY) && - ( (route_def->src_type != src_type) || - ( (route_def->src_id != src_link) && - (route_def->src_id != -1)))) { - continue; + + +#ifdef V3_CONFIG_DEBUG_VNET + { + char dst_str[32]; + char src_str[32]; + + mac2str(route_def->src_mac, src_str); + mac2str(route_def->dst_mac, dst_str); + + PrintDebug(VM_NONE, VCORE_NONE, "Tested match against SRC(%s) SRC_QUAL(%s), DEST(%s) DST_QUAL(%s): " + "SRC_MATCH=%s DEST_MATCH=%s PRIORITY=%d\n", + src_str, QUAL_TO_STR(route_def->src_mac_qual), + dst_str, QUAL_TO_STR(route_def->dst_mac_qual), + MATCH_CLASS_TO_STR(match_mac(hdr->src_mac,route_def->src_mac,route_def->src_mac_qual)), + MATCH_CLASS_TO_STR(match_mac(hdr->dst_mac,route_def->dst_mac,route_def->dst_mac_qual)), + priority); } -*/ +#endif - if ((route_def->dst_mac_qual == MAC_ANY) && - (route_def->src_mac_qual == MAC_ANY)) { - UPDATE_MATCHES(3); + if (priority<0) { + PrintDebug(VM_NONE, VCORE_NONE, "No match to this rule\n"); + continue; } - - if (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) { - if (route_def->src_mac_qual != MAC_NOT) { - if (route_def->dst_mac_qual == MAC_ANY) { - UPDATE_MATCHES(6); - } else if (route_def->dst_mac_qual != MAC_NOT && - memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) { - UPDATE_MATCHES(8); - } + + if (priority > max_priority) { + PrintDebug(VM_NONE, VCORE_NONE, "New highest priority match, reseting list\n"); + max_priority = priority; + + struct vnet_route_info *my_route, *tmp_route; + + list_for_each_entry_safe(my_route, tmp_route, &match_list,match_node) { + list_del(&(my_route->match_node)); } - } + + list_add(&(route->match_node), &match_list); + num_matches = 1; - if (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0) { - if (route_def->dst_mac_qual != MAC_NOT) { - if (route_def->src_mac_qual == MAC_ANY) { - UPDATE_MATCHES(6); - } else if ((route_def->src_mac_qual != MAC_NOT) && - (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) { - UPDATE_MATCHES(8); - } - } - } + } else if (priority == max_priority) { + PrintDebug(VM_NONE, VCORE_NONE, "Equal priority match, adding to list\n"); - if ((route_def->dst_mac_qual == MAC_NOT) && - (memcmp(route_def->dst_mac, hdr->dst_mac, 6) != 0)) { - if (route_def->src_mac_qual == MAC_ANY) { - UPDATE_MATCHES(5); - } else if ((route_def->src_mac_qual != MAC_NOT) && - (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0)) { - UPDATE_MATCHES(7); - } - } + list_add(&(route->match_node), &match_list); + num_matches++; + } - if ((route_def->src_mac_qual == MAC_NOT) && - (memcmp(route_def->src_mac, hdr->src_mac, 6) != 0)) { - if (route_def->dst_mac_qual == MAC_ANY) { - UPDATE_MATCHES(5); - } else if ((route_def->dst_mac_qual != MAC_NOT) && - (memcmp(route_def->dst_mac, hdr->dst_mac, 6) == 0)) { - UPDATE_MATCHES(7); - } - } - - // Default route - if ( (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) && - (route_def->dst_mac_qual == MAC_NONE)) { - UPDATE_MATCHES(4); - } } - Vnet_Debug("VNET/P Core: match_route: Matches=%d\n", num_matches); + PrintDebug(VM_NONE, VCORE_NONE, "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(VM_NONE, VCORE_NONE, "VNET/P Core: Unable to allocate matches\n"); + return NULL; + } + matches->num_routes = num_matches; - { - int i = 0; - list_for_each_entry(route, &match_list, match_node) { + i=0; + list_for_each_entry(route, &match_list, match_node) { + if (i==num_matches) { + // the list should never have more than num_matches on it... + PrintError(VM_NONE, VCORE_NONE, "Weird list behavior\n"); + break; + } else { matches->routes[i++] = route; } + } return matches; } +int v3_vnet_query_header(uint8_t src_mac[ETH_ALEN], + uint8_t dest_mac[ETH_ALEN], + 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; + void *flags; + + p.size=14; + p.data=p.header; + memcpy(p.header,dest_mac,ETH_ALEN); + memcpy(p.header+ETH_ALEN,src_mac,ETH_ALEN); + memset(p.header+12,0,2); + + p.src_type = LINK_EDGE; + p.src_id = 0; + + memcpy(header->src_mac,src_mac,ETH_ALEN); + memcpy(header->dst_mac,dest_mac,ETH_ALEN); + + + flags = vnet_lock_irqsave(vnet_state.lock); + + look_into_cache(&p,&routes); + + if (!routes) { + routes = match_route(&p); + if (!routes) { + vnet_unlock_irqrestore(vnet_state.lock,flags); + PrintError(VM_NONE, VCORE_NONE, "Cannot match route\n"); + header->header_type=VNET_HEADER_NOMATCH; + header->header_len=0; + return -1; + } else { + add_route_to_cache(&p,routes); + } + } + + vnet_unlock_irqrestore(vnet_state.lock,flags); + + if (routes->num_routes<1) { + PrintError(VM_NONE, VCORE_NONE, "Less than one route\n"); + header->header_type=VNET_HEADER_NOMATCH; + header->header_len=0; + return -1; + } + + if (routes->num_routes>1) { + PrintError(VM_NONE, VCORE_NONE, "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(VM_NONE, VCORE_NONE, "Unknown destination type\n"); + return -1; + break; + + } + +} + + + 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); + if(net_debug >= 4){ v3_hexdump(pkt->data, pkt->size, NULL, 0); } @@ -490,15 +722,16 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { 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(VM_NONE, VCORE_NONE, "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(VM_NONE, VCORE_NONE, "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?*/ } @@ -506,7 +739,7 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { vnet_unlock_irqrestore(vnet_state.lock, flags); - Vnet_Debug("VNET/P Core: send pkt route matches %d\n", matched_routes->num_routes); + PrintDebug(VM_NONE, VCORE_NONE, "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]; @@ -549,43 +782,54 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac, - struct v3_vnet_dev_ops *ops, int quote, int poll_state, + 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); + memcpy(new_dev->mac_addr, mac, ETH_ALEN); 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 = quotequote = quotepoll = poll_state; + stop_vnet_kick_threads(); + flags = vnet_lock_irqsave(vnet_state.lock); if (dev_by_mac(mac) == NULL) { 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(VM_NONE, VCORE_NONE,"VNET/P: Device with the same MAC has already been added\n"); } vnet_unlock_irqrestore(vnet_state.lock, flags); + start_vnet_kick_threads(); + /* if the device was found previosly the id should still be 0 */ if (new_dev->dev_id == 0) { Vnet_Print(0, "VNET/P Core: Device Already exists\n"); return -1; } - Vnet_Debug("VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id); + PrintDebug(VM_NONE, VCORE_NONE, "VNET/P Core: Add Device: dev_id %d\n", new_dev->dev_id); return new_dev->dev_id; } @@ -593,7 +837,9 @@ int v3_vnet_add_dev(struct v3_vm_info * vm, uint8_t * mac, int v3_vnet_del_dev(int dev_id){ struct vnet_dev * dev = NULL; - unsigned long flags; + vnet_intr_flags_t flags; + + stop_vnet_kick_threads(); flags = vnet_lock_irqsave(vnet_state.lock); @@ -606,9 +852,11 @@ int v3_vnet_del_dev(int dev_id){ vnet_unlock_irqrestore(vnet_state.lock, flags); + start_vnet_kick_threads(); + Vnet_Free(dev); - Vnet_Debug("VNET/P Core: Remove Device: dev_id %d\n", dev_id); + PrintDebug(VM_NONE, VCORE_NONE, "VNET/P Core: Removed Device: dev_id %d\n", dev_id); return 0; } @@ -624,18 +872,18 @@ int v3_vnet_stat(struct vnet_stat * stats){ } 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); @@ -646,7 +894,7 @@ int v3_vnet_add_bridge(struct v3_vm_info * vm, 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; @@ -658,14 +906,14 @@ int v3_vnet_add_bridge(struct v3_vm_info * vm, vnet_unlock_irqrestore(vnet_state.lock, flags); if (bridge_free == 0) { - PrintError("VNET/P Core: Bridge already set\n"); + PrintError(VM_NONE, VCORE_NONE, "VNET/P Core: Bridge already set\n"); return -1; } tmp_bridge = (struct vnet_brg_dev *)Vnet_Malloc(sizeof(struct vnet_brg_dev)); if (tmp_bridge == NULL) { - PrintError("Malloc Fails\n"); + PrintError(VM_NONE, VCORE_NONE, "VNET/P Core: Unable to allocate new bridge\n"); vnet_state.bridge = NULL; return -1; } @@ -686,14 +934,14 @@ int v3_vnet_add_bridge(struct v3_vm_info * vm, 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); @@ -708,36 +956,104 @@ void v3_vnet_del_bridge(uint8_t type) { * that runs on multiple cores * or it could be running on a dedicated side core */ -static int vnet_tx_flush(void *args){ +static int vnet_tx_flush(void * args){ struct vnet_dev * dev = NULL; - int ret; + int more; + int rc; + uint64_t noprogress_count; Vnet_Print(0, "VNET/P Polling Thread Starting ....\n"); - /* we need thread sleep/wakeup in Palacios */ - while(!vnet_thread_should_stop()){ - dev = (struct vnet_dev *)v3_dequeue(vnet_state.poll_devs); - if(dev != NULL){ - if(dev->poll && dev->dev_ops.poll != NULL){ - ret = dev->dev_ops.poll(dev->vm, dev->quote, dev->private_data); + // 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 (ret < 0){ - PrintDebug("VNET/P: poll from device %p error!\n", dev); + struct v3_queue * tq = v3_create_queue(); + + if (!tq) { + PrintError(VM_NONE, VCORE_NONE, "VNET/P polling thread cannot allocate queue\n"); + return -1; + } + + noprogress_count = 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); + } - v3_enqueue(vnet_state.poll_devs, (addr_t)dev); + + if (more) { + noprogress_count=0; + } else { + if ( ! ((noprogress_count+1) < noprogress_count)) { + noprogress_count++; } - }else { /* no device needs to be polled */ - /* sleep here? */ - Vnet_Yield(); } + + // adaptively yield + if ((!VNET_ADAPTIVE_TX_KICK) || (noprogress_count < VNET_NOPROGRESS_LIMIT)) { + V3_Yield(); + } else { + V3_Sleep(VNET_YIELD_USEC); + } + } + Vnet_Free(tq); + + Vnet_Print(0, "VNET/P Polling Thread Done.\n"); + return 0; } +static int start_vnet_kick_threads() +{ + int i; -int v3_init_vnet() { + for (i=0; i