From: Lei Xia Date: Mon, 2 May 2011 15:50:30 +0000 (-0500) Subject: Add large MTU support into Virtio NIC X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=commitdiff_plain;h=70842a596167a7cfb7d8c8121cf0d5d2350079ea;p=palacios.git Add large MTU support into Virtio NIC asychronized pkt sending in VNET/P --- diff --git a/palacios/src/devices/lnx_virtio_nic.c b/palacios/src/devices/lnx_virtio_nic.c index 9c55d41..baf3369 100644 --- a/palacios/src/devices/lnx_virtio_nic.c +++ b/palacios/src/devices/lnx_virtio_nic.c @@ -117,11 +117,15 @@ struct virtio_net_state { struct vm_device * dev; struct pci_device * pci_dev; int io_range_size; + + uint16_t status; struct virtio_queue rx_vq; /* idx 0*/ struct virtio_queue tx_vq; /* idx 1*/ struct virtio_queue ctrl_vq; /* idx 2*/ + uint8_t mergeable_rx_bufs; + struct v3_timer * timer; void * poll_thread; @@ -165,11 +169,13 @@ static int virtio_init_state(struct virtio_net_state * virtio) virtio->ctrl_vq.cur_avail_idx = 0; virtio->virtio_cfg.pci_isr = 0; + + virtio->mergeable_rx_bufs = 0; virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC); - // (1 << VIRTIO_NET_F_GSO) | - // (1 << VIRTIO_NET_F_HOST_UFO) | - // (1 << VIRTIO_NET_F_HOST_TSO4); + if(virtio->mergeable_rx_bufs) { + virtio->virtio_cfg.host_features |= (1 << VIRTIO_NET_F_MRG_RXBUF); + } if ((v3_lock_init(&(virtio->rx_lock)) == -1) || (v3_lock_init(&(virtio->tx_lock)) == -1)){ @@ -185,7 +191,7 @@ static int tx_one_pkt(struct guest_info * core, { uint8_t * buf = NULL; uint32_t len = buf_desc->length; - int synchronize = 1; // (virtio->tx_notify == 1)?1:0; + int synchronize = virtio->tx_notify; if (v3_gpa_to_hva(core, buf_desc->addr_gpa, (addr_t *)&(buf)) == -1) { PrintDebug("Could not translate buffer address\n"); @@ -214,8 +220,7 @@ static inline int copy_data_to_desc(struct guest_info * core, struct vring_desc * desc, uchar_t * buf, uint_t buf_len, - uint_t offset) -{ + uint_t dst_offset){ uint32_t len; uint8_t * desc_buf = NULL; @@ -223,8 +228,8 @@ static inline int copy_data_to_desc(struct guest_info * core, PrintDebug("Could not translate buffer address\n"); return -1; } - len = (desc->length < buf_len)?(desc->length - offset):buf_len; - memcpy(desc_buf+offset, buf, len); + len = (desc->length < buf_len)?(desc->length - dst_offset):buf_len; + memcpy(desc_buf+dst_offset, buf, len); return len; } @@ -255,7 +260,7 @@ static int handle_pkt_tx(struct guest_info * core, { struct virtio_queue *q = &(virtio_state->tx_vq); int txed = 0; - unsigned long flags; + unsigned long flags; if (!q->ring_avail_addr) { return -1; @@ -263,7 +268,7 @@ static int handle_pkt_tx(struct guest_info * core, flags = v3_lock_irqsave(virtio_state->tx_lock); while (q->cur_avail_idx != q->avail->index) { - struct virtio_net_hdr *hdr = NULL; + struct virtio_net_hdr_mrg_rxbuf * hdr = NULL; struct vring_desc * hdr_desc = NULL; addr_t hdr_addr = 0; uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; @@ -280,15 +285,20 @@ static int handle_pkt_tx(struct guest_info * core, goto exit_error; } - hdr = (struct virtio_net_hdr *)hdr_addr; + hdr = (struct virtio_net_hdr_mrg_rxbuf *)hdr_addr; desc_idx = hdr_desc->next; + V3_Net_Print(2, "Virtio NIC: TX hdr count : %d\n", hdr->num_buffers); + /* here we assumed that one ethernet pkt is not splitted into multiple buffer */ struct vring_desc * buf_desc = &(q->desc[desc_idx]); if (tx_one_pkt(core, virtio_state, buf_desc) == -1) { PrintError("Virtio NIC: Error handling nic operation\n"); goto exit_error; } + if(buf_desc->next & VIRTIO_NEXT_FLAG){ + V3_Net_Print(2, "Virtio NIC: TX more buffer need to read\n"); + } 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 = buf_desc->length; /* What do we set this to???? */ @@ -300,9 +310,8 @@ static int handle_pkt_tx(struct guest_info * core, } v3_unlock_irqrestore(virtio_state->tx_lock, flags); - - if (virtio_state->virtio_cfg.pci_isr == 0 && - txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { + + if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev); virtio_state->virtio_cfg.pci_isr = 0x1; @@ -548,9 +557,7 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { struct virtio_net_state * virtio = (struct virtio_net_state *)private_data; struct virtio_queue * q = &(virtio->rx_vq); struct virtio_net_hdr_mrg_rxbuf hdr; - uint32_t hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); uint32_t data_len; - //uint32_t offset = 0; unsigned long flags; V3_Net_Print(2, "Virtio-NIC: virtio_rx: size: %d\n", size); @@ -575,6 +582,7 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; struct vring_desc * hdr_desc = NULL; struct vring_desc * buf_desc = NULL; + uint32_t hdr_len = 0; uint32_t len; hdr_desc = &(q->desc[hdr_idx]); @@ -584,44 +592,64 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { goto err_exit; } -#if 0 /* merged buffer */ - for(buf_idx = hdr_desc->next; offset < data_len; buf_idx = q->desc[hdr_idx].next) { - uint32_t len = 0; - buf_desc = &(q->desc[buf_idx]); + hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf); - len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset, 0); - offset += len; - if (offset < data_len) { - buf_desc->flags = VIRTIO_NEXT_FLAG; - } - buf_desc->length = len; - hdr.num_buffers ++; - } - buf_desc->flags &= ~VIRTIO_NEXT_FLAG; - memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr_mrg_rxbuf)); -#endif + if(virtio->mergeable_rx_bufs){/* merged buffer */ + uint32_t offset = 0; + len = 0; + hdr.num_buffers = 0; - hdr.num_buffers = 1; - memcpy((void *)hdr_addr, &hdr, sizeof(struct virtio_net_hdr_mrg_rxbuf)); - if (data_len == 0) { + hdr_desc = &(q->desc[buf_idx]); hdr_desc->flags &= ~VIRTIO_NEXT_FLAG; - } - buf_idx = hdr_desc->next; - buf_desc = &(q->desc[buf_idx]); - len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf, data_len, 0); - if (len < data_len) { - V3_Net_Print(2, "Virtio NIC: ring buffer len less than pkt size, merged buffer not supported\n"); - virtio->stats.rx_dropped ++; - - goto err_exit; - } - buf_desc->flags &= ~VIRTIO_NEXT_FLAG; + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, buf, data_len, hdr_len); + offset += len; - 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 = data_len + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */ - q->used->index++; - q->cur_avail_idx++; + hdr.num_buffers ++; + 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 = hdr_len + len; + q->cur_avail_idx ++; + + while(offset < data_len) { + buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size]; + buf_desc = &(q->desc[buf_idx]); + + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset, 0); + if (len <= 0){ + V3_Net_Print(2, "Virtio NIC:merged buffer, %d buffer size %d\n", hdr.num_buffers, data_len); + virtio->stats.rx_dropped ++; + goto err_exit; + } + offset += len; + buf_desc->flags &= ~VIRTIO_NEXT_FLAG; + + hdr.num_buffers ++; + q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size]; + q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len; + q->cur_avail_idx ++; + } + q->used->index += hdr.num_buffers; + copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0); + }else{ + hdr_desc = &(q->desc[buf_idx]); + copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0); + + buf_idx = hdr_desc->next; + buf_desc = &(q->desc[buf_idx]); + len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf, data_len, 0); + if (len < data_len) { + V3_Net_Print(2, "Virtio NIC: ring buffer len less than pkt size, merged buffer not supported, buffer size %d\n", len); + virtio->stats.rx_dropped ++; + + goto err_exit; + } + buf_desc->flags &= ~VIRTIO_NEXT_FLAG; + + 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 = data_len + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */ + q->used->index++; + q->cur_avail_idx++; + } virtio->stats.rx_pkts ++; virtio->stats.rx_bytes += size; @@ -638,9 +666,8 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { goto err_exit; } - if (virtio->virtio_cfg.pci_isr == 0 && - !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { - PrintDebug("Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line); + if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) { + V3_Net_Print(2, "Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line); virtio->virtio_cfg.pci_isr = 0x1; v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev); @@ -651,8 +678,8 @@ static int virtio_rx(uint8_t * buf, uint32_t size, void * private_data) { v3_unlock_irqrestore(virtio->rx_lock, flags); /* notify guest if it is in guest mode */ - /* ISSUE: What is gonna happen if guest thread is running on the same core as this thread? */ - if(virtio->rx_notify == 1){ + if(virtio->rx_notify == 1 && + V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].cpu_id){ v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].cpu_id, 0); } @@ -791,7 +818,6 @@ static void virtio_nic_timer(struct guest_info * core, period_us = (1000*cpu_cycles)/cpu_freq; net_state->past_us += period_us; -#if 0 if(net_state->past_us > PROFILE_PERIOD){ uint32_t tx_rate, rx_rate; @@ -826,7 +852,6 @@ static void virtio_nic_timer(struct guest_info * core, net_state->past_us = 0; } -#endif profile_ms += period_us/1000; if(profile_ms > 20000){ diff --git a/palacios/src/palacios/vmm_vnet_core.c b/palacios/src/palacios/vmm_vnet_core.c index 4b54d71..92f8467 100644 --- a/palacios/src/palacios/vmm_vnet_core.c +++ b/palacios/src/palacios/vmm_vnet_core.c @@ -84,10 +84,11 @@ struct route_list { struct queue_entry{ uint8_t use; struct v3_vnet_pkt pkt; - uint8_t data[ETHERNET_PACKET_LEN]; + uint8_t * data; + uint32_t size_alloc; }; -#define VNET_QUEUE_SIZE 10240 +#define VNET_QUEUE_SIZE 1024 struct vnet_queue { struct queue_entry buf[VNET_QUEUE_SIZE]; int head, tail; @@ -520,6 +521,7 @@ 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); @@ -538,6 +540,20 @@ static int vnet_pkt_enqueue(struct v3_vnet_pkt * pkt){ /* 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); @@ -719,6 +735,8 @@ static int vnet_tx_flush(void *args){ /* 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; V3_Net_Print(2, "vnet_tx_flush: pkt (size %d)\n", entry->pkt.size);