From: Lei Xia Date: Thu, 25 Mar 2010 22:15:53 +0000 (-0500) Subject: Update on VNET X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=commitdiff_plain;h=a04b0778a7f74005560b99255de2c6d2df4d5292;p=palacios.git Update on VNET Add asychronous IPI for VNET --- diff --git a/palacios/include/palacios/vmm.h b/palacios/include/palacios/vmm.h index bbc4eba..67b60c8 100644 --- a/palacios/include/palacios/vmm.h +++ b/palacios/include/palacios/vmm.h @@ -202,6 +202,14 @@ struct guest_info; +#define V3_lapic_send_ipi(cpu, vector) \ + do { \ + extern struct v3_os_hooks * os_hooks; \ + if ((os_hooks) && (os_hooks)->lapic_send_ipi) { \ + (os_hooks)->lapic_send_ipi(cpu, vector); \ + } \ + } while (0) + typedef enum v3_vm_class {V3_INVALID_VM, V3_PC_VM, V3_CRAY_VM} v3_vm_class_t; @@ -267,12 +275,9 @@ struct v3_os_hooks { void (*interrupt_cpu)(struct v3_vm_info * vm, int logical_cpu); void (*call_on_cpu)(int logical_cpu, void (*fn)(void * arg), void * arg); void (*start_thread_on_cpu)(int logical_cpu, int (*fn)(void * arg), void * arg, char * thread_name); -}; - - - - + void (*lapic_send_ipi)(unsigned int cpu, unsigned int vector); +}; // diff --git a/palacios/include/palacios/vmm_vnet.h b/palacios/include/palacios/vmm_vnet.h index 7298918..0e9bc17 100644 --- a/palacios/include/palacios/vmm_vnet.h +++ b/palacios/include/palacios/vmm_vnet.h @@ -26,6 +26,9 @@ #include + +#define V3_VNET_POLLING_VECTOR 50 + typedef enum {MAC_ANY=0, MAC_NOT, MAC_NONE, MAC_ADDR} mac_type_t; //for 'src_mac_qual' and 'dst_mac_qual' typedef enum {LINK_INTERFACE=0, LINK_EDGE, LINK_ANY} link_type_t; //for 'type' and 'src_type' in struct routing @@ -89,25 +92,41 @@ struct v3_vnet_profile{ }; #endif +struct v3_vnet_bridge_input_args{ + struct v3_vm_info * vm; + struct v3_vnet_pkt *vnet_pkts; + uint16_t pkt_num; + void * private_data; +}; int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void *private_data); -int v3_vnet_add_route(struct v3_vnet_route route); - -//int v3_vnet_del_route(); - +void v3_vnet_send_pkt_xcall(void * data); +int v3_vnet_add_route(struct v3_vnet_route route); int V3_init_vnet(); int v3_vnet_add_bridge(struct v3_vm_info * vm, - int (*input)(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, void * private_data), - void * priv_data); + int (*input)(struct v3_vm_info * vm, struct v3_vnet_pkt pkt[], uint16_t pkt_num, void * private_data), + void (*xcall_input)(void *data), + uint16_t max_delayed_pkts, + long max_latency, + void * priv_data); int v3_vnet_add_dev(struct v3_vm_info *info, uint8_t mac[6], int (*dev_input)(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, void * private_data), void * priv_data); +void v3_vnet_heartbeat(struct guest_info *core); + + +int v3_vnet_disable_bridge(); +int v3_vnet_enable_bridge(); + +void v3_vnet_bridge_polling(); + +int v3_vnet_bridge_rx(uchar_t *buf, uint16_t size, uint16_t src_link); #endif diff --git a/palacios/src/devices/lnx_virtio_nic.c b/palacios/src/devices/lnx_virtio_nic.c index 7bcd9bb..557d3e8 100644 --- a/palacios/src/devices/lnx_virtio_nic.c +++ b/palacios/src/devices/lnx_virtio_nic.c @@ -204,6 +204,7 @@ static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virt struct virtio_queue * q = &(virtio_state->tx_vq); struct virtio_net_hdr * hdr = NULL; + PrintDebug("Virtio NIC: TX: running on cpu: %d\n", V3_Get_CPU()); while (q->cur_avail_idx != q->avail->index) { struct vring_desc * hdr_desc = NULL; @@ -221,7 +222,7 @@ static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virt hdr = (struct virtio_net_hdr*)hdr_addr; desc_idx = hdr_desc->next; - + for (i = 0; i < desc_cnt - 1; i++) { struct vring_desc * buf_desc = &(q->desc[desc_idx]); if (pkt_tx(core, virtio_state, buf_desc) == -1) { @@ -246,6 +247,15 @@ static int handle_pkt_tx(struct guest_info *core, struct virtio_net_state * virt virtio_state->virtio_cfg.pci_isr = 0x1; } +#ifdef CONFIG_VNET_PROFILE + if (virtio_state->pkt_sent % 10000 == 0) + PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n", + virtio_state->pkt_sent, + virtio_state->pkt_recv, + virtio_state->pkt_drop); + +#endif + return 0; } @@ -482,8 +492,8 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { int raw = 1; flags = v3_lock_irqsave(virtio->lock); - - PrintDebug("VIRTIO NIC: receiving packet to virtio nic %p, size:%d\n", virtio, size); + + PrintDebug("VIRTIO NIC: RX on cpu %d to virtio nic %p, size:%d\n", V3_Get_CPU(), virtio, size); virtio->pkt_recv ++; @@ -535,6 +545,7 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { q->cur_avail_idx++; } else { virtio->pkt_drop++; + goto exit; } if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { @@ -546,6 +557,15 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { ret_val = offset; exit: + +#ifdef CONFIG_VNET_PROFILE + if (virtio->pkt_recv % 10000 == 0) + PrintError("Virtio NIC: sent: %ld, rxed: %ld, dropped: %ld\n", + virtio->pkt_sent, + virtio->pkt_recv, + virtio->pkt_drop); + +#endif v3_unlock_irqrestore(virtio->lock, flags); diff --git a/palacios/src/devices/lnx_virtio_symmod.c b/palacios/src/devices/lnx_virtio_symmod.c index 46b9ecb..62ed8d3 100644 --- a/palacios/src/devices/lnx_virtio_symmod.c +++ b/palacios/src/devices/lnx_virtio_symmod.c @@ -103,7 +103,7 @@ static int handle_xfer_kick(struct guest_info * core, struct virtio_sym_state * PrintDebug("SYMMOD: VIRTIO SYMMOD Kick on loader queue\n"); - while (q->cur_avail_idx ! q->avail->index) { + while (q->cur_avail_idx != q->avail->index) { struct vring_desc * hdr_desc = NULL; struct vring_desc * buf_desc = NULL; struct vring_desc * status_desc = NULL; diff --git a/palacios/src/devices/lnx_virtio_vnet.c b/palacios/src/devices/lnx_virtio_vnet.c index bfbc56c..f88bf28 100644 --- a/palacios/src/devices/lnx_virtio_vnet.c +++ b/palacios/src/devices/lnx_virtio_vnet.c @@ -8,10 +8,12 @@ * http://www.v3vee.org * * Copyright (c) 2008, Jack Lange + * Copyright (c) 2008, Lei Xia * Copyright (c) 2008, The V3VEE Project * All rights reserved. * * Author: Jack Lange + * Lei Xia * * This is free software. You are permitted to use, * redistribute, and modify it as specified in the file "V3VEE_LICENSE". @@ -32,7 +34,8 @@ #endif -#define QUEUE_SIZE 128 +#define QUEUE_SIZE 8192 +#define CMD_QUEUE_SIZE 128 #define NUM_QUEUES 3 struct vnet_config { @@ -59,6 +62,9 @@ struct virtio_vnet_state { int io_range_size; v3_lock_t lock; + + ulong_t pkt_sent, pkt_recv, pkt_drop; + int ready; }; #define VNET_GET_ROUTES 10 @@ -75,7 +81,8 @@ struct vnet_ctrl_hdr { uint32_t num_cmds; } __attribute__((packed)); -struct vnet_virtio_pkt { + +struct vnet_bridge_pkt { uint32_t link_id; uint32_t pkt_size; uint8_t pkt[ETHERNET_PACKET_LEN]; @@ -84,14 +91,14 @@ struct vnet_virtio_pkt { static int virtio_reset(struct virtio_vnet_state * vnet_state) { - memset(vnet_state->queue, 0, sizeof(struct virtio_queue) * 2); + memset(vnet_state->queue, 0, sizeof(struct virtio_queue) * NUM_QUEUES); vnet_state->cur_queue = &(vnet_state->queue[0]); vnet_state->virtio_cfg.status = 0; vnet_state->virtio_cfg.pci_isr = 0; - vnet_state->queue[0].queue_size = QUEUE_SIZE; + vnet_state->queue[0].queue_size = CMD_QUEUE_SIZE; vnet_state->queue[1].queue_size = QUEUE_SIZE; vnet_state->queue[2].queue_size = QUEUE_SIZE; @@ -215,11 +222,16 @@ static int handle_cmd_kick(struct guest_info * core, struct virtio_vnet_state * } -static int vnet_pkt_input_cb(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, void * private_data){ +static int vnet_pkt_input_cb(struct v3_vm_info * vm, struct v3_vnet_pkt vnet_pkts[], uint16_t pkt_num, void * private_data){ struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data; struct virtio_queue * q = &(vnet_state->queue[RECV_QUEUE]); int ret_val = -1; unsigned long flags; + uint16_t sent; + struct v3_vnet_pkt *pkt; + + if(pkt_num <= 0) + return 0; flags = v3_lock_irqsave(vnet_state->lock); @@ -228,35 +240,49 @@ static int vnet_pkt_input_cb(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, goto exit; } + PrintDebug("VNET Bridge: RX: running on cpu: %d, num of pkts: %d\n", V3_Get_CPU(), pkt_num); - if (q->cur_avail_idx != q->avail->index) { - uint16_t pkt_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; - struct vring_desc * pkt_desc = NULL; - struct vnet_virtio_pkt * virtio_pkt = NULL; + for(sent = 0; sent < pkt_num; sent ++) { + pkt = &vnet_pkts[sent]; + vnet_state->pkt_recv ++; + if (q->cur_avail_idx != q->avail->index) { + uint16_t pkt_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; + struct vring_desc * pkt_desc = NULL; + struct vnet_bridge_pkt * virtio_pkt = NULL; - pkt_desc = &(q->desc[pkt_idx]); - PrintDebug("VNET Bridge RX: buffer desc len: %d\n", pkt_desc->length); + //if(q->cur_avail_idx % 100 == 0) + // PrintError("cur_avai_idx %d, idx: %d\n", q->cur_avail_idx, q->avail->index); - if (guest_pa_to_host_va(&(vm->cores[0]), pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) { - PrintError("Could not translate buffer address\n"); - return -1; - } + pkt_desc = &(q->desc[pkt_idx]); + PrintDebug("VNET Bridge RX: buffer desc len: %d\n", pkt_desc->length); - PrintDebug("VNET Bridge: RX: pkt sent to guest pkt size: %d, dst link: %d\n", pkt->size, pkt->dst_id); + if (guest_pa_to_host_va(&(vm->cores[0]), pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) { + PrintError("Could not translate buffer address\n"); + goto exit; + } + + PrintDebug("VNET Bridge: RX: pkt sent to guest pkt size: %d, dst link: %d\n", pkt->size, pkt->dst_id); - // Fill in dst packet buffer - virtio_pkt->link_id = pkt->dst_id; - virtio_pkt->pkt_size = pkt->size; - memcpy(virtio_pkt->pkt, pkt->data, pkt->size); + // Fill in dst packet buffer + virtio_pkt->link_id = pkt->dst_id; + virtio_pkt->pkt_size = pkt->size; + memcpy(virtio_pkt->pkt, pkt->data, pkt->size); - q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; - q->used->ring[q->used->index % q->queue_size].length = sizeof(struct vnet_virtio_pkt); // This should be the total length of data sent to guest (header+pkt_data) + q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; + q->used->ring[q->used->index % q->queue_size].length = sizeof(struct vnet_bridge_pkt); // This should be the total length of data sent to guest (header+pkt_data) + + q->used->index++; + q->cur_avail_idx++; + } else { + //PrintError("VNET Bridge: guest RX buffer full: cur_avai_idx %d, idx: %d\nDisable Bridge\n", q->cur_avail_idx, q->avail->index); + vnet_state->pkt_drop ++; + v3_vnet_disable_bridge(); + } + } - q->used->index++; - q->cur_avail_idx++; - } else { - PrintError("Packet buffer overflow in the guest: cur_avai_idx %d, idx: %d\n", q->cur_avail_idx, q->avail->index); + if(sent == 0){ + goto exit; } if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { @@ -268,67 +294,92 @@ static int vnet_pkt_input_cb(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, ret_val = 0; exit: + +#ifdef CONFIG_VNET_PROFILE + if (vnet_state->pkt_recv % 10000 == 0) + PrintError("Vnet Bridge: sent: %ld, rxed: %ld, dropped: %ld\n", + vnet_state->pkt_sent, + vnet_state->pkt_recv, + vnet_state->pkt_drop); + +#endif + v3_unlock_irqrestore(vnet_state->lock, flags); return ret_val; } +static void vnet_pkt_input_xcall(void *data){ + struct v3_vnet_bridge_input_args *args = (struct v3_vnet_bridge_input_args *)data; + + vnet_pkt_input_cb(args->vm, args->vnet_pkts, args->pkt_num, args->private_data); +} static int handle_pkt_kick(struct guest_info *core, struct virtio_vnet_state * vnet_state) { struct virtio_queue * q = &(vnet_state->queue[XMIT_QUEUE]); - struct v3_vnet_pkt pkt; while (q->cur_avail_idx != q->avail->index) { uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; struct vring_desc * pkt_desc = NULL; - struct vnet_virtio_pkt * virtio_pkt = NULL; + struct vnet_bridge_pkt * virtio_pkt = NULL; pkt_desc = &(q->desc[desc_idx]); PrintDebug("VNET Bridge: Handle TX desc buf_len: %d\n", pkt_desc->length); - - //PrintDebug("q: %p, %p, %p, %d\n", q, vnet_state->queue, &vnet_state->queue[XMIT_QUEUE], XMIT_QUEUE); - //PrintDebug("q->used: %p\n", q->used); - //PrintDebug("q->used->ring %p\n", q->used->ring); - + if (guest_pa_to_host_va(core, pkt_desc->addr_gpa, (addr_t *)&(virtio_pkt)) == -1) { PrintError("Could not translate buffer address\n"); return -1; } - PrintDebug("VNET Bridge: TX: pkt size: %d, dst link: %d\n", virtio_pkt->pkt_size, virtio_pkt->link_id); - - pkt.size = virtio_pkt->pkt_size; - pkt.src_id = virtio_pkt->link_id; - pkt.src_type = LINK_EDGE; - memcpy(pkt.header, virtio_pkt->pkt, ETHERNET_HEADER_LEN); - pkt.data = virtio_pkt->pkt; - - v3_vnet_send_pkt(&pkt, NULL); - - q = (struct virtio_queue *)((addr_t)vnet_state->queue + XMIT_QUEUE*sizeof(struct virtio_queue)); - - PrintDebug("After q: %p, , %p, %p, %d\n", q, vnet_state->queue, &(vnet_state->queue[XMIT_QUEUE]), XMIT_QUEUE); - //PrintDebug("After q->used: %p\n", q->used); - //PrintDebug("After q->used->ring %p\n", q->used->ring); + PrintDebug("VNET Bridge: TX: on cpu %d pkt size: %d, dst link: %d\n", cpu, virtio_pkt->pkt_size, virtio_pkt->link_id); + v3_vnet_bridge_rx(virtio_pkt->pkt, virtio_pkt->pkt_size, virtio_pkt->link_id); + q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; q->used->ring[q->used->index % q->queue_size].length = pkt_desc->length; // What do we set this to???? q->used->index++; + vnet_state->pkt_sent ++; + q->cur_avail_idx++; } - if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev); - vnet_state->virtio_cfg.pci_isr = 0x1; + //interrupt the vnet to poll pkts + int cpu = V3_Get_CPU(); + cpu = (cpu == 0)?1:0; + V3_lapic_send_ipi(cpu, V3_VNET_POLLING_VECTOR); + + if((vnet_state->pkt_sent % (QUEUE_SIZE/20)) == 0) { //optimized for guest's, batch the interrupts + if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { + v3_pci_raise_irq(vnet_state->pci_bus, 0, vnet_state->pci_dev); + vnet_state->virtio_cfg.pci_isr = 0x1; + } } +#ifdef CONFIG_VNET_PROFILE + if (vnet_state->pkt_sent % 10000 == 0) + PrintError("Vnet Bridge: sent: %ld, rxed: %ld, dropped: %ld\n", + vnet_state->pkt_sent, + vnet_state->pkt_recv, + vnet_state->pkt_drop); + +#endif + return 0; } -static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) { + +static int handle_rx_kick(struct guest_info *core, struct virtio_vnet_state * vnet_state) +{ + v3_vnet_enable_bridge(); + //PrintError("Enable Bridge\n"); + + return 0; +} + +static int vnet_virtio_io_write(struct guest_info * core, uint16_t port, void * src, uint_t length, void * private_data) { struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data; int port_idx = port % vnet_state->io_range_size; @@ -385,6 +436,9 @@ static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, PrintDebug("VNET Bridge: RingDesc=%p, Avail=%p, Used=%p\n", vnet_state->cur_queue->desc, vnet_state->cur_queue->avail, vnet_state->cur_queue->used); + if(vnet_state->queue[RECV_QUEUE].avail != NULL){ + vnet_state->ready = 1; + } } else { PrintError("Illegal write length for page frame number\n"); return -1; @@ -416,9 +470,12 @@ static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, if (handle_pkt_kick(core, vnet_state) == -1){ PrintError("Could not handle Virtio VNET TX\n"); return -1; - } + } } else if (queue_idx == 2) { - PrintDebug("VNET Bridge: receive kick on RX Queue\n"); + if (handle_rx_kick(core, vnet_state) == -1){ + PrintError("Could not handle Virtio RX buffer refills Kick\n"); + return -1; + } } else { PrintError("VNET Bridge: Kick on invalid queue (%d)\n", queue_idx); return -1; @@ -448,7 +505,7 @@ static int virtio_io_write(struct guest_info * core, uint16_t port, void * src, } -static int virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) { +static int vnet_virtio_io_read(struct guest_info * core, uint16_t port, void * dst, uint_t length, void * private_data) { struct virtio_vnet_state * vnet_state = (struct virtio_vnet_state *)private_data; int port_idx = port % vnet_state->io_range_size; @@ -529,7 +586,6 @@ static struct v3_device_ops dev_ops = { .stop = NULL, }; - static int dev_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_vnet_state * vnet_state = NULL; @@ -584,8 +640,8 @@ static int dev_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) { bars[0].type = PCI_BAR_IO; bars[0].default_base_port = -1; bars[0].num_ports = vnet_state->io_range_size; - bars[0].io_read = virtio_io_read; - bars[0].io_write = virtio_io_write; + bars[0].io_read = vnet_virtio_io_read; + bars[0].io_write = vnet_virtio_io_write; bars[0].private_data = vnet_state; pci_dev = v3_pci_register_device(pci_bus, PCI_STD_DEVICE, @@ -615,7 +671,8 @@ static int dev_init(struct v3_vm_info * vm, v3_cfg_tree_t * cfg) { virtio_reset(vnet_state); V3_Print("Registering Virtio device as vnet bridge\n"); - v3_vnet_add_bridge(vm, vnet_pkt_input_cb, (void *)vnet_state); + v3_vnet_add_bridge(vm, vnet_pkt_input_cb, vnet_pkt_input_xcall, 5, 1000000, (void *)vnet_state); + return 0; } diff --git a/palacios/src/palacios/vmm_vnet.c b/palacios/src/palacios/vmm_vnet.c index 38a9087..e5c17bc 100644 --- a/palacios/src/palacios/vmm_vnet.c +++ b/palacios/src/palacios/vmm_vnet.c @@ -55,10 +55,29 @@ struct vnet_dev { } __attribute__((packed)); +#define BRIDGE_BUF_SIZE 1024 +struct bridge_pkts_buf { + int start, end; + int num; + v3_lock_t lock; + struct v3_vnet_pkt pkts[BRIDGE_BUF_SIZE]; + uint8_t datas[ETHERNET_PACKET_LEN*BRIDGE_BUF_SIZE]; +}; + struct vnet_brg_dev { struct v3_vm_info * vm; - int (*input)(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, void * private_data); + int (*input)(struct v3_vm_info * vm, struct v3_vnet_pkt pkt[], uint16_t pkt_num, void * private_data); + void (*xcall_input)(void *data); + + struct bridge_pkts_buf recv_buf; //packets from Vnet to vnet_bridge device + + struct bridge_pkts_buf send_buf; //packets from vnet_bridge device to Vnet + + int disabled; + + uint16_t max_delayed_pkts; + long max_latency; //in cycles void * private_data; } __attribute__((packed)); @@ -77,8 +96,6 @@ struct vnet_route_info { }; - - struct route_list { uint8_t hash_buf[VNET_HASH_SIZE]; @@ -109,7 +126,7 @@ static struct { #ifdef CONFIG_DEBUG_VNET static inline void mac_to_string(char mac[6], char * buf) { - snprintf(buf, 50, "%x:%x:%x:%x:%x:%x", + snprintf(buf, 100, "%d:%d:%d:%d:%d:%d", mac[0], mac[1], mac[2], mac[3], mac[4], mac[5]); } @@ -168,7 +185,6 @@ static inline int hash_eq(addr_t key1, addr_t key2) { return (memcmp((uint8_t *)key1, (uint8_t *)key2, VNET_HASH_SIZE) == 0); } - 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); @@ -291,8 +307,8 @@ static struct route_list * match_route(const struct v3_vnet_pkt * pkt) { #ifdef CONFIG_DEBUG_VNET { - char dst_str[50]; - char src_str[50]; + char dst_str[100]; + char src_str[100]; mac_to_string(hdr->src_mac, src_str); mac_to_string(hdr->dst_mac, dst_str); @@ -388,8 +404,8 @@ static struct route_list * match_route(const struct v3_vnet_pkt * pkt) { return NULL; } - matches = V3_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)); matches->num_routes = num_matches; @@ -404,6 +420,146 @@ static struct route_list * match_route(const struct v3_vnet_pkt * pkt) { } +static int flush_bridge_pkts(struct vnet_brg_dev *bridge){ + unsigned long flags; + int num, start, send; + struct v3_vnet_bridge_input_args args; + int cpu_id = bridge->vm->cores[0].cpu_id; + int current_core = V3_Get_CPU(); + + if (bridge == NULL) { + PrintDebug("VNET: No bridge to sent data to links\n"); + return -1; + } + + flags = v3_lock_irqsave(bridge->recv_buf.lock); + + num = bridge->recv_buf.num; + start = bridge->recv_buf.start; + + bridge->recv_buf.num -= num; + bridge->recv_buf.start += num; + bridge->recv_buf.start %= BRIDGE_BUF_SIZE; + + v3_unlock_irqrestore(bridge->recv_buf.lock, flags); + + + if(bridge->disabled){ + PrintDebug("VNET: In flush bridge pkts: Bridge is disabled\n"); + return -1; + } + + if(num <= 2 && num > 0){ + PrintDebug("VNET: In flush bridge pkts: %d\n", num); + } + + if(num > 0) { + PrintDebug("VNET: In flush bridge pkts to bridge, cur_cpu %d, brige_core: %d\n", current_core, cpu_id); + if (current_core == cpu_id){ + if ((start + num) < BRIDGE_BUF_SIZE){ + bridge->input(bridge->vm, &(bridge->recv_buf.pkts[start]), num, bridge->private_data); + }else { + bridge->input(bridge->vm, &(bridge->recv_buf.pkts[start]), (BRIDGE_BUF_SIZE - start), bridge->private_data); + send = num - (BRIDGE_BUF_SIZE - start); + bridge->input(bridge->vm, &(bridge->recv_buf.pkts[0]), send, bridge->private_data); + } + }else { + args.vm = bridge->vm; + args.private_data = bridge->private_data; + + if ((start + num) < BRIDGE_BUF_SIZE){ + args.pkt_num = num; + args.vnet_pkts = &(bridge->recv_buf.pkts[start]); + V3_Call_On_CPU(cpu_id, bridge->xcall_input, (void *)&args); + }else { + args.pkt_num = BRIDGE_BUF_SIZE - start; + args.vnet_pkts = &(bridge->recv_buf.pkts[start]); + V3_Call_On_CPU(cpu_id, bridge->xcall_input, (void *)&args); + + send = num - (BRIDGE_BUF_SIZE - start); + args.pkt_num = send; + args.vnet_pkts = &(bridge->recv_buf.pkts[0]); + V3_Call_On_CPU(cpu_id, bridge->xcall_input, (void *)&args); + } + } + + PrintDebug("VNET: flush bridge pkts %d\n", num); + } + + return 0; +} + + +static int send_to_bridge(struct v3_vnet_pkt * pkt){ + struct vnet_brg_dev *bridge = vnet_state.bridge; + int cpu_id = bridge->vm->cores[0].cpu_id; + struct v3_vnet_bridge_input_args args; + + if (bridge == NULL) { + PrintDebug("VNET: No bridge to sent data to links\n"); + return -1; + } + + if(bridge->max_delayed_pkts <= 1){ + if(bridge->disabled){ + PrintDebug("VNET: Bridge diabled\n"); + return -1; + } + + args.pkt_num = 1; + args.vm = bridge->vm; + args.vnet_pkts = pkt; + args.private_data = bridge->private_data; + + V3_Call_On_CPU(cpu_id, bridge->xcall_input, (void *)&args); + PrintDebug("VNET: sent one packet to the bridge\n"); + return 0; + } + + unsigned long flags; + int end, num=0; + struct v3_vnet_pkt *buf; + + PrintDebug("VNET: send_to_bridge\n"); + + flags = v3_lock_irqsave(bridge->recv_buf.lock); + + if(bridge->disabled && bridge->recv_buf.num >= BRIDGE_BUF_SIZE){ + PrintDebug("Bridge diabled and bridge receive buffer full\n"); + v3_unlock_irqrestore(bridge->recv_buf.lock, flags);//maybe should move this after copy + num = bridge->recv_buf.num; + goto exit; + } + + end = bridge->recv_buf.end; + buf = &(bridge->recv_buf.pkts[end]); + + bridge->recv_buf.num ++; + bridge->recv_buf.end ++; + bridge->recv_buf.end %= BRIDGE_BUF_SIZE; + + num = bridge->recv_buf.num; + + v3_unlock_irqrestore(bridge->recv_buf.lock, flags);//maybe should move this after copy + + + buf->size = pkt->size; + buf->dst_id = pkt->dst_id; + buf->src_id = pkt->src_id; + buf->src_type = pkt->src_type; + buf->dst_type = pkt->dst_type; + memcpy(buf->header, pkt->header, ETHERNET_HEADER_LEN); + memcpy(buf->data, pkt->data, pkt->size); + +exit: + + if (num >= bridge->max_delayed_pkts){ + flush_bridge_pkts(bridge); + } + + return 0; +} + int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { struct route_list * matched_routes = NULL; unsigned long flags; @@ -412,12 +568,13 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { #ifdef CONFIG_DEBUG_VNET { struct eth_hdr * hdr = (struct eth_hdr *)(pkt->header); - char dest_str[30]; - char src_str[30]; + char dest_str[100]; + char src_str[100]; mac_to_string(hdr->src_mac, src_str); mac_to_string(hdr->dst_mac, dest_str); - PrintDebug("Vnet: HandleDataOverLink. SRC(%s), DEST(%s)\n", src_str, dest_str); + int cpu = V3_Get_CPU(); + PrintDebug("Vnet: on cpu %d, HandleDataOverLink. SRC(%s), DEST(%s), pkt size: %d\n", cpu, src_str, dest_str, pkt->size); } #endif @@ -429,7 +586,7 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { PrintDebug("Vnet: send pkt Looking into routing table\n"); matched_routes = match_route(pkt); - + if (matched_routes) { add_route_to_cache(pkt, matched_routes); } else { @@ -441,7 +598,6 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { v3_unlock_irqrestore(vnet_state.lock, flags); - PrintDebug("Vnet: send pkt route matches %d\n", matched_routes->num_routes); for (i = 0; i < matched_routes->num_routes; i++) { @@ -451,16 +607,11 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { pkt->dst_type = LINK_EDGE; pkt->dst_id = route->route_def.dst_id; - if (vnet_state.bridge == NULL) { - PrintDebug("VNET: No bridge to sent data to links\n"); + if (send_to_bridge(pkt) == -1) { + PrintDebug("VNET: Packet not sent properly to bridge\n"); continue; - } - - if (vnet_state.bridge->input(vnet_state.bridge->vm, pkt, vnet_state.bridge->private_data) == -1) { - PrintDebug("VNET: Packet not sent properly\n"); - continue; - } - + } + } else if (route->route_def.dst_type == LINK_INTERFACE) { if (route->dst_dev->input(route->dst_dev->vm, pkt, route->dst_dev->private_data) == -1) { PrintDebug("VNET: Packet not sent properly\n"); @@ -471,12 +622,95 @@ int v3_vnet_send_pkt(struct v3_vnet_pkt * pkt, void * private_data) { continue; } - PrintDebug("Vnet: HandleOnePacket: Forward packet according to Route\n"); + PrintDebug("Vnet: v3_vnet_send_pkt: Forward packet according to Route %d\n", i); } return 0; } +void v3_vnet_send_pkt_xcall(void * data){ + struct v3_vnet_pkt * pkt = (struct v3_vnet_pkt *)data; + v3_vnet_send_pkt(pkt, NULL); +} + + +void v3_vnet_bridge_polling() +{ + unsigned long flags; + int num, start; + struct v3_vnet_pkt *buf; + struct vnet_brg_dev *bridge = vnet_state.bridge; + + PrintDebug("In vnet bridge pollling: cpu %d\n", V3_Get_CPU()); + + if(bridge == NULL){ + PrintDebug("VNET: Bridge is not set\n"); + return; + } + + flags = v3_lock_irqsave(bridge->send_buf.lock); + + num = bridge->send_buf.num; + start = bridge->send_buf.start; + + PrintDebug("VNET: bridge polling pkts %d\n", num); + + while(num > 0) { + buf = &(bridge->send_buf.pkts[bridge->send_buf.start]); + + v3_vnet_send_pkt(buf, NULL); + + bridge->send_buf.num --; + bridge->send_buf.start ++; + bridge->send_buf.start %= BRIDGE_BUF_SIZE; + num --; + } + + v3_unlock_irqrestore(bridge->send_buf.lock, flags); + + return; +} + + +int v3_vnet_bridge_rx(uchar_t *buf, uint16_t size, uint16_t src_link){ + struct vnet_brg_dev *bridge = vnet_state.bridge; + unsigned long flags; + int end; + struct v3_vnet_pkt *pkt; + + if (bridge == NULL) { + PrintDebug("VNET: No bridge is set\n"); + return -1; + } + + flags = v3_lock_irqsave(bridge->send_buf.lock); + + end = bridge->send_buf.end; + pkt = &(bridge->send_buf.pkts[end]); + + if(bridge->send_buf.num > BRIDGE_BUF_SIZE){ + PrintDebug("VNET: bridge rx: buffer full\n"); + goto exit; + } + + bridge->send_buf.num ++; + bridge->send_buf.end ++; + bridge->send_buf.end %= BRIDGE_BUF_SIZE; + + pkt->size = size; + pkt->src_id = src_link; + pkt->src_type = LINK_EDGE; + memcpy(pkt->header, buf, ETHERNET_HEADER_LEN); + memcpy(pkt->data, buf, size); + +exit: + + v3_unlock_irqrestore(bridge->send_buf.lock, flags); + + return 0; +} + + int v3_vnet_add_dev(struct v3_vm_info *vm, uint8_t mac[6], int (*netif_input)(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, void * private_data), void * priv_data){ @@ -518,12 +752,35 @@ int v3_vnet_add_dev(struct v3_vm_info *vm, uint8_t mac[6], } +void v3_vnet_heartbeat(struct guest_info *core){ + static long last_time, cur_time; + + if(vnet_state.bridge == NULL) + return; + + if(vnet_state.bridge->max_delayed_pkts <= 1) + return; + + if(V3_Get_CPU() != vnet_state.bridge->vm->cores[0].cpu_id){ + rdtscll(cur_time); + } + + if ((cur_time - last_time) >= vnet_state.bridge->max_latency) { + last_time = cur_time; + flush_bridge_pkts(vnet_state.bridge); + } +} + int v3_vnet_add_bridge(struct v3_vm_info * vm, - int (*input)(struct v3_vm_info * vm, struct v3_vnet_pkt * pkt, void * private_data), + int (*input)(struct v3_vm_info * vm, struct v3_vnet_pkt pkt[], uint16_t pkt_num, void * private_data), + void (*xcall_input)(void *data), + uint16_t max_delayed_pkts, + long max_latency, void * priv_data) { unsigned long flags; int bridge_free = 0; struct vnet_brg_dev * tmp_bridge = NULL; + int i; flags = v3_lock_irqsave(vnet_state.lock); @@ -549,8 +806,34 @@ int v3_vnet_add_bridge(struct v3_vm_info * vm, tmp_bridge->vm = vm; tmp_bridge->input = input; + tmp_bridge->xcall_input = xcall_input; tmp_bridge->private_data = priv_data; + tmp_bridge->disabled = 0; + + //initial receving buffer + tmp_bridge->recv_buf.start = 0; + tmp_bridge->recv_buf.end = 0; + tmp_bridge->recv_buf.num = 0; + if(v3_lock_init(&(tmp_bridge->recv_buf.lock)) == -1){ + PrintError("VNET: add bridge, error to initiate recv buf lock\n"); + } + tmp_bridge->max_delayed_pkts = (max_delayed_pktsmax_latency = max_latency; + for(i = 0; irecv_buf.pkts[i].data = &(tmp_bridge->recv_buf.datas[i*ETHERNET_PACKET_LEN]); + } + //initial sending buffer + tmp_bridge->send_buf.start = 0; + tmp_bridge->send_buf.end = 0; + tmp_bridge->send_buf.num = 0; + if(v3_lock_init(&(tmp_bridge->send_buf.lock)) == -1){ + PrintError("VNET: add bridge, error to initiate send buf lock\n"); + } + for(i = 0; isend_buf.pkts[i].data = &(tmp_bridge->send_buf.datas[i*ETHERNET_PACKET_LEN]); + } + // make this atomic to avoid possible race conditions flags = v3_lock_irqsave(vnet_state.lock); vnet_state.bridge = tmp_bridge; @@ -559,6 +842,38 @@ int v3_vnet_add_bridge(struct v3_vm_info * vm, return 0; } + +int v3_vnet_disable_bridge() { + unsigned long flags; + + flags = v3_lock_irqsave(vnet_state.lock); + + if (vnet_state.bridge != NULL) { + vnet_state.bridge->disabled = 1; + } + + v3_unlock_irqrestore(vnet_state.lock, flags); + + return 0; +} + + +int v3_vnet_enable_bridge() { + unsigned long flags; + + flags = v3_lock_irqsave(vnet_state.lock); + + if (vnet_state.bridge != NULL) { + vnet_state.bridge->disabled = 0; + } + + v3_unlock_irqrestore(vnet_state.lock, flags); + + return 0; +} + + + int V3_init_vnet() { INIT_LIST_HEAD(&(vnet_state.routes)); diff --git a/palacios/src/palacios/vmx_handler.c b/palacios/src/palacios/vmx_handler.c index a5f6bc8..2f64f83 100644 --- a/palacios/src/palacios/vmx_handler.c +++ b/palacios/src/palacios/vmx_handler.c @@ -37,6 +37,9 @@ #endif +#ifdef CONFIG_VNET +#include +#endif /* At this point the GPRs are already copied into the guest_info state */ int v3_handle_vmx_exit(struct guest_info * info, struct vmx_exit_info * exit_info) { @@ -222,6 +225,12 @@ int v3_handle_vmx_exit(struct guest_info * info, struct vmx_exit_info * exit_inf return -1; } + +#ifdef CONFIG_VNET + v3_vnet_heartbeat(info); +#endif + + #ifdef CONFIG_TELEMETRY if (info->vm_info->enable_telemetry) { v3_telemetry_end_exit(info, exit_info->exit_reason);