X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fvnet%2Fvnet_core.c;h=7d9942d38f7a0cf1d4d597bd385b746e6c0c06ea;hb=cdaef2158977795b123c9dba6816eb6e9d54810d;hp=9bf044a29fec8b95f754dec621beb4765bf1da98;hpb=d12cdf5b213b2b51d525914a852ceacd2c198b5a;p=palacios.git diff --git a/palacios/src/vnet/vnet_core.c b/palacios/src/vnet/vnet_core.c index 9bf044a..7d9942d 100644 --- a/palacios/src/vnet/vnet_core.c +++ b/palacios/src/vnet/vnet_core.c @@ -10,7 +10,7 @@ * Copyright (c) 2010, Lei Xia * Copyright (c) 2009, Yuan Tang * Copyright (c) 2009, The V3VEE Project - * All rights reserved. + * All rights reserved * * Author: Lei Xia * Yuan Tang @@ -20,18 +20,25 @@ */ #include -#include -#include +#include +#include +#include + #include -#include -#include #ifndef V3_CONFIG_DEBUG_VNET #undef PrintDebug #define PrintDebug(fmt, args...) #endif -int v3_net_debug = 0; +#define VNET_NUM_TX_KICK_THREADS 1 + +#define VNET_ADAPTIVE_TX_KICK 0 // 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 { uint8_t dst_mac[ETH_ALEN]; @@ -45,6 +52,12 @@ struct vnet_dev { 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; @@ -68,6 +81,8 @@ struct vnet_route_info { struct vnet_dev * dst_dev; struct vnet_dev * src_dev; + uint32_t idx; + struct list_head node; struct list_head match_node; // used for route matching }; @@ -88,36 +103,35 @@ struct queue_entry{ uint32_t size_alloc; }; -#define VNET_QUEUE_SIZE 1024 -struct vnet_queue { - struct queue_entry buf[VNET_QUEUE_SIZE]; - int head, tail; - int count; - v3_lock_t lock; -}; static struct { struct list_head routes; struct list_head devs; - - int num_routes; - int num_devs; + + uint8_t status; + + uint32_t num_routes; + uint32_t route_idx; + uint32_t num_devs; + uint32_t dev_idx; struct vnet_brg_dev * bridge; - v3_lock_t lock; + vnet_lock_t lock; struct vnet_stat stats; - void * 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[VNET_NUM_TX_KICK_THREADS]; struct hashtable * route_cache; + } vnet_state; #ifdef V3_CONFIG_DEBUG_VNET -static inline void mac_to_string(uint8_t * mac, char * buf) { +static inline void mac2str(uint8_t * mac, char * buf) { snprintf(buf, 100, "%2x:%2x:%2x:%2x:%2x:%2x", mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]); @@ -126,16 +140,16 @@ static inline void mac_to_string(uint8_t * mac, char * buf) { static void print_route(struct v3_vnet_route * route){ char str[50]; - mac_to_string(route->src_mac, str); - PrintDebug("Src Mac (%s), src_qual (%d)\n", + mac2str(route->src_mac, str); + PrintDebug(VM_NONE, VCORE_NONE, "Src Mac (%s), src_qual (%d)\n", str, route->src_mac_qual); - mac_to_string(route->dst_mac, str); - PrintDebug("Dst Mac (%s), dst_qual (%d)\n", + mac2str(route->dst_mac, str); + PrintDebug(VM_NONE, VCORE_NONE, "Dst Mac (%s), dst_qual (%d)\n", str, route->dst_mac_qual); - PrintDebug("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); - PrintDebug("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); } @@ -143,14 +157,13 @@ static void print_route(struct v3_vnet_route * route){ static void dump_routes(){ struct vnet_route_info *route; - int i = 0; - PrintDebug("\n========Dump routes starts ============\n"); + PrintDebug(VM_NONE, VCORE_NONE, "\n========Dump routes starts ============\n"); list_for_each_entry(route, &(vnet_state.routes), node) { - PrintDebug("\nroute %d:\n", i++); + PrintDebug(VM_NONE, VCORE_NONE, "\nroute %d:\n", route->idx); print_route(&(route->route_def)); if (route->route_def.dst_type == LINK_INTERFACE) { - PrintDebug("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), @@ -158,7 +171,7 @@ static void dump_routes(){ } } - PrintDebug("\n========Dump routes end ============\n"); + PrintDebug(VM_NONE, VCORE_NONE, "\n========Dump routes end ============\n"); } #endif @@ -171,7 +184,7 @@ static void dump_routes(){ static inline uint_t hash_fn(addr_t hdr_ptr) { uint8_t * hdr_buf = (uint8_t *)hdr_ptr; - return v3_hash_buffer(hdr_buf, VNET_HASH_SIZE); + return vnet_hash_buffer(hdr_buf, VNET_HASH_SIZE); } static inline int hash_eq(addr_t key1, addr_t key2) { @@ -181,8 +194,8 @@ static inline int hash_eq(addr_t key1, addr_t key2) { static int add_route_to_cache(const struct v3_vnet_pkt * pkt, struct route_list * routes) { memcpy(routes->hash_buf, pkt->hash_buf, VNET_HASH_SIZE); - if (v3_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"); + if (vnet_htable_insert(vnet_state.route_cache, (addr_t)routes->hash_buf, (addr_t)routes) == 0) { + PrintError(VM_NONE, VCORE_NONE, "VNET/P Core: Failed to insert new route entry to the cache\n"); return -1; } @@ -190,15 +203,15 @@ static int add_route_to_cache(const struct v3_vnet_pkt * pkt, struct route_list } static int clear_hash_cache() { - v3_free_htable(vnet_state.route_cache, 1, 1); - vnet_state.route_cache = v3_create_htable(0, &hash_fn, &hash_eq); + vnet_free_htable(vnet_state.route_cache, 1, 1); + vnet_state.route_cache = vnet_create_htable(0, &hash_fn, &hash_eq); return 0; } static int look_into_cache(const struct v3_vnet_pkt * pkt, struct route_list ** routes) { - *routes = (struct route_list *)v3_htable_search(vnet_state.route_cache, (addr_t)(pkt->hash_buf)); + *routes = (struct route_list *)vnet_htable_search(vnet_state.route_cache, (addr_t)(pkt->hash_buf)); return 0; } @@ -208,10 +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; @@ -230,6 +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; @@ -245,13 +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; + } - new_route = (struct vnet_route_info *)V3_Malloc(sizeof(struct vnet_route_info)); memset(new_route, 0, sizeof(struct vnet_route_info)); #ifdef V3_CONFIG_DEBUG_VNET - PrintDebug("VNET/P Core: add_route_entry:\n"); + PrintDebug(VM_NONE, VCORE_NONE, "VNET/P Core: add_route_entry:\n"); print_route(&route); #endif @@ -273,30 +295,56 @@ int v3_vnet_add_route(struct v3_vnet_route route) { } - flags = v3_lock_irqsave(vnet_state.lock); + flags = vnet_lock_irqsave(vnet_state.lock); list_add(&(new_route->node), &(vnet_state.routes)); - clear_hash_cache(); + new_route->idx = ++ vnet_state.route_idx; + vnet_state.num_routes ++; + + vnet_unlock_irqrestore(vnet_state.lock, flags); - v3_unlock_irqrestore(vnet_state.lock, flags); - + clear_hash_cache(); #ifdef V3_CONFIG_DEBUG_VNET dump_routes(); #endif - return 0; + return new_route->idx; } -/* delete all route entries with specfied src or dst device id */ -static void inline del_routes_by_dev(int dev_id){ +void v3_vnet_del_route(uint32_t route_idx){ struct vnet_route_info * route = NULL; - unsigned long flags; + vnet_intr_flags_t flags; - flags = v3_lock_irqsave(vnet_state.lock); + 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)); + 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, *tmp_route; + vnet_intr_flags_t flags; + + flags = vnet_lock_irqsave(vnet_state.lock); + + 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 && @@ -304,178 +352,394 @@ static void inline del_routes_by_dev(int dev_id){ list_del(&(route->node)); list_del(&(route->match_node)); - V3_Free(route); + Vnet_Free(route); } } + + vnet_unlock_irqrestore(vnet_state.lock, flags); +} + + +// 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; + } - v3_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) { +#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]; - - mac_to_string(hdr->src_mac, src_str); - mac_to_string(hdr->dst_mac, dst_str); - PrintDebug("VNET/P Core: match_route. pkt: SRC(%s), DEST(%s)\n", src_str, dst_str); + char dst_str[32], src_str[32]; + mac2str(hdr->src_mac, src_str); + mac2str(hdr->dst_mac, 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); - } - } - - 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); - } - } + list_add(&(route->match_node), &match_list); + num_matches++; + } - // Default route - if ( (memcmp(route_def->src_mac, hdr->src_mac, 6) == 0) && - (route_def->dst_mac_qual == MAC_NONE)) { - UPDATE_MATCHES(4); - } } - PrintDebug("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)); - matches = (struct route_list *)V3_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); -int vnet_tx_one_pkt(struct v3_vnet_pkt * pkt, void * private_data) { + 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(); - V3_Net_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(v3_net_debug >= 4){ - v3_hexdump(pkt->data, pkt->size, NULL, 0); + + 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); } - flags = v3_lock_irqsave(vnet_state.lock); + flags = vnet_lock_irqsave(vnet_state.lock); vnet_state.stats.rx_bytes += pkt->size; vnet_state.stats.rx_pkts++; look_into_cache(pkt, &matched_routes); + if (matched_routes == NULL) { - PrintDebug("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 { - PrintDebug("VNET/P Core: Could not find route for packet... discards packet\n"); - v3_unlock_irqrestore(vnet_state.lock, flags); + 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?*/ } } - v3_unlock_irqrestore(vnet_state.lock, flags); + vnet_unlock_irqrestore(vnet_state.lock, flags); - PrintDebug("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]; @@ -486,30 +750,30 @@ int vnet_tx_one_pkt(struct v3_vnet_pkt * pkt, void * private_data) { pkt->dst_id = route->route_def.dst_id; if (bridge == NULL) { - V3_Net_Print(2, "VNET/P Core: No active bridge to sent data to\n"); - continue; + Vnet_Print(2, "VNET/P Core: No active bridge to sent data to\n"); + continue; } if(bridge->brg_ops.input(bridge->vm, pkt, bridge->private_data) < 0){ - V3_Net_Print(2, "VNET/P Core: Packet not sent properly to bridge\n"); + Vnet_Print(2, "VNET/P Core: Packet not sent properly to bridge\n"); continue; } vnet_state.stats.tx_bytes += pkt->size; vnet_state.stats.tx_pkts ++; } else if (route->route_def.dst_type == LINK_INTERFACE) { if (route->dst_dev == NULL){ - V3_Net_Print(2, "VNET/P Core: No active device to sent data to\n"); + Vnet_Print(2, "VNET/P Core: No active device to sent data to\n"); continue; } if(route->dst_dev->dev_ops.input(route->dst_dev->vm, pkt, route->dst_dev->private_data) < 0) { - V3_Net_Print(2, "VNET/P Core: Packet not sent properly\n"); + Vnet_Print(2, "VNET/P Core: Packet not sent properly\n"); continue; } vnet_state.stats.tx_bytes += pkt->size; vnet_state.stats.tx_pkts ++; } else { - PrintError("VNET/P Core: Wrong dst type\n"); + Vnet_Print(0, "VNET/P Core: Wrong dst type\n"); } } @@ -517,99 +781,55 @@ int vnet_tx_one_pkt(struct v3_vnet_pkt * pkt, void * private_data) { } -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 = v3_lock_irqsave(q->lock); - - if (q->count >= VNET_QUEUE_SIZE){ - V3_Net_Print(1, "VNET Queue overflow!\n"); - v3_unlock_irqrestore(q->lock, flags); - return -1; - } - - q->count ++; - entry = &(q->buf[q->tail++]); - q->tail %= VNET_QUEUE_SIZE; - - v3_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){ - V3_FreePages(V3_PAddr(entry->data), (entry->size_alloc / PAGE_SIZE)); - entry->data = NULL; - } - - num_pages = 1 + (pkt->size / PAGE_SIZE); - entry->data = V3_VAddr(V3_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); - V3_Net_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 *)V3_Malloc(sizeof(struct vnet_dev)); + new_dev = (struct vnet_dev *)Vnet_Malloc(sizeof(struct vnet_dev)); if (new_dev == NULL) { - PrintError("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 = quotepoll = poll_state; - flags = v3_lock_irqsave(vnet_state.lock); + 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.num_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"); } - v3_unlock_irqrestore(vnet_state.lock, flags); + 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) { - PrintError("VNET/P Core: Device Already exists\n"); + Vnet_Print(0, "VNET/P Core: Device Already exists\n"); return -1; } - PrintDebug("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; } @@ -617,28 +837,32 @@ 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; - flags = v3_lock_irqsave(vnet_state.lock); + stop_vnet_kick_threads(); + + flags = vnet_lock_irqsave(vnet_state.lock); dev = dev_by_id(dev_id); if (dev != NULL){ list_del(&(dev->node)); - del_routes_by_dev(dev_id); + //del_routes_by_dev(dev_id); + vnet_state.num_devs --; } - v3_unlock_irqrestore(vnet_state.lock, flags); + vnet_unlock_irqrestore(vnet_state.lock, flags); + + start_vnet_kick_threads(); - V3_Free(dev); + Vnet_Free(dev); - PrintDebug("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; } int v3_vnet_stat(struct vnet_stat * stats){ - stats->rx_bytes = vnet_state.stats.rx_bytes; stats->rx_pkts = vnet_state.stats.rx_pkts; stats->tx_bytes = vnet_state.stats.tx_bytes; @@ -647,22 +871,22 @@ int v3_vnet_stat(struct vnet_stat * stats){ return 0; } -static void free_devices(){ - struct vnet_dev * dev = NULL; +static void deinit_devices_list(){ + 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)); - V3_Free(dev); + Vnet_Free(dev); } } -static void free_routes(){ - struct vnet_route_info * route = NULL; +static void deinit_routes_list(){ + 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)); - V3_Free(route); + Vnet_Free(route); } } @@ -670,26 +894,26 @@ 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; - flags = v3_lock_irqsave(vnet_state.lock); + flags = vnet_lock_irqsave(vnet_state.lock); if (vnet_state.bridge == NULL) { bridge_free = 1; vnet_state.bridge = (void *)1; } - v3_unlock_irqrestore(vnet_state.lock, flags); + 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 *)V3_Malloc(sizeof(struct vnet_brg_dev)); + 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; } @@ -701,51 +925,135 @@ int v3_vnet_add_bridge(struct v3_vm_info * vm, tmp_bridge->type = type; /* make this atomic to avoid possible race conditions */ - flags = v3_lock_irqsave(vnet_state.lock); + flags = vnet_lock_irqsave(vnet_state.lock); vnet_state.bridge = tmp_bridge; - v3_unlock_irqrestore(vnet_state.lock, flags); + vnet_unlock_irqrestore(vnet_state.lock, flags); return 0; } -#if 0 -static int vnet_tx_flush(void *args){ - unsigned long flags; - struct queue_entry * entry; - struct vnet_queue * q = &(vnet_state.pkt_q); - V3_Print("VNET/P Handing Pkt Thread Starting ....\n"); +void v3_vnet_del_bridge(uint8_t type) { + 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_unlock_irqrestore(vnet_state.lock, flags); + + if (tmp_bridge) { + Vnet_Free(tmp_bridge); + } +} + - //V3_THREAD_SLEEP(); - /* we need thread sleep/wakeup in Palacios */ - while(1){ - flags = v3_lock_irqsave(q->lock); +/* 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; + uint64_t noprogress_count; - if (q->count <= 0){ - v3_unlock_irqrestore(q->lock, flags); - v3_yield(NULL); - //V3_THREAD_SLEEP(); - }else { - q->count --; - entry = &(q->buf[q->head++]); - q->head %= VNET_QUEUE_SIZE; + Vnet_Print(0, "VNET/P Polling Thread Starting ....\n"); - v3_unlock_irqrestore(q->lock, flags); + // 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 - /* this is ugly, but should happen very unlikely */ - while(!entry->use); - vnet_tx_one_pkt(&(entry->pkt), NULL); + struct v3_queue * tq = v3_create_queue(); - /* asynchronizely release allocated memory for buffer entry here */ - entry->use = 0; + if (!tq) { + PrintError(VM_NONE, VCORE_NONE, "VNET/P polling thread cannot allocate queue\n"); + return -1; + } - V3_Net_Print(2, "vnet_tx_flush: pkt (size %d)\n", entry->pkt.size); + 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); + } + + + if (more) { + noprogress_count=0; + } else { + if ( ! ((noprogress_count+1) < noprogress_count)) { + noprogress_count++; + } + } + + // 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; + + for (i=0; i