diff options
Diffstat (limited to 'src/vnet/devices/ssvm')
| -rw-r--r-- | src/vnet/devices/ssvm/node.c | 343 | ||||
| -rw-r--r-- | src/vnet/devices/ssvm/ssvm_eth.c | 491 | ||||
| -rw-r--r-- | src/vnet/devices/ssvm/ssvm_eth.h | 141 |
3 files changed, 975 insertions, 0 deletions
diff --git a/src/vnet/devices/ssvm/node.c b/src/vnet/devices/ssvm/node.c new file mode 100644 index 00000000000..3a695b1d8c0 --- /dev/null +++ b/src/vnet/devices/ssvm/node.c @@ -0,0 +1,343 @@ +/* + * Copyright (c) 2015 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include "ssvm_eth.h" + +vlib_node_registration_t ssvm_eth_input_node; + +typedef struct +{ + u32 next_index; + u32 sw_if_index; +} ssvm_eth_input_trace_t; + +/* packet trace format function */ +static u8 * +format_ssvm_eth_input_trace (u8 * s, va_list * args) +{ + CLIB_UNUSED (vlib_main_t * vm) = va_arg (*args, vlib_main_t *); + CLIB_UNUSED (vlib_node_t * node) = va_arg (*args, vlib_node_t *); + ssvm_eth_input_trace_t *t = va_arg (*args, ssvm_eth_input_trace_t *); + + s = format (s, "SSVM_ETH_INPUT: sw_if_index %d, next index %d", + t->sw_if_index, t->next_index); + return s; +} + +vlib_node_registration_t ssvm_eth_input_node; + +#define foreach_ssvm_eth_input_error \ +_(NO_BUFFERS, "Rx packet drops (no buffers)") + +typedef enum +{ +#define _(sym,str) SSVM_ETH_INPUT_ERROR_##sym, + foreach_ssvm_eth_input_error +#undef _ + SSVM_ETH_INPUT_N_ERROR, +} ssvm_eth_input_error_t; + +static char *ssvm_eth_input_error_strings[] = { +#define _(sym,string) string, + foreach_ssvm_eth_input_error +#undef _ +}; + +typedef enum +{ + SSVM_ETH_INPUT_NEXT_DROP, + SSVM_ETH_INPUT_NEXT_ETHERNET_INPUT, + SSVM_ETH_INPUT_NEXT_IP4_INPUT, + SSVM_ETH_INPUT_NEXT_IP6_INPUT, + SSVM_ETH_INPUT_NEXT_MPLS_INPUT, + SSVM_ETH_INPUT_N_NEXT, +} ssvm_eth_input_next_t; + +static inline uword +ssvm_eth_device_input (ssvm_eth_main_t * em, + ssvm_private_t * intfc, vlib_node_runtime_t * node) +{ + ssvm_shared_header_t *sh = intfc->sh; + vlib_main_t *vm = em->vlib_main; + unix_shared_memory_queue_t *q; + ssvm_eth_queue_elt_t *elt, *elts; + u32 elt_index; + u32 my_pid = intfc->my_pid; + int rx_queue_index; + u32 n_to_alloc = VLIB_FRAME_SIZE * 2; + u32 n_allocated, n_present_in_cache; + u32 next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT; + vlib_buffer_free_list_t *fl; + u32 n_left_to_next, *to_next; + u32 next0; + u32 n_buffers; + u32 n_available; + u32 bi0, saved_bi0; + vlib_buffer_t *b0, *prev; + u32 saved_cache_size = 0; + ethernet_header_t *eh0; + u16 type0; + u32 n_rx_bytes = 0, l3_offset0; + u32 cpu_index = os_get_cpu_number (); + u32 trace_cnt __attribute__ ((unused)) = vlib_get_trace_count (vm, node); + volatile u32 *lock; + u32 *elt_indices; + uword n_trace = vlib_get_trace_count (vm, node); + + /* Either side down? buh-bye... */ + if (pointer_to_uword (sh->opaque[MASTER_ADMIN_STATE_INDEX]) == 0 || + pointer_to_uword (sh->opaque[SLAVE_ADMIN_STATE_INDEX]) == 0) + return 0; + + if (intfc->i_am_master) + q = (unix_shared_memory_queue_t *) (sh->opaque[TO_MASTER_Q_INDEX]); + else + q = (unix_shared_memory_queue_t *) (sh->opaque[TO_SLAVE_Q_INDEX]); + + /* Nothing to do? */ + if (q->cursize == 0) + return 0; + + fl = vlib_buffer_get_free_list (vm, VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX); + + vec_reset_length (intfc->rx_queue); + + lock = (u32 *) q; + while (__sync_lock_test_and_set (lock, 1)) + ; + while (q->cursize > 0) + { + unix_shared_memory_queue_sub_raw (q, (u8 *) & elt_index); + ASSERT (elt_index < 2048); + vec_add1 (intfc->rx_queue, elt_index); + } + CLIB_MEMORY_BARRIER (); + *lock = 0; + + n_present_in_cache = vec_len (em->buffer_cache); + + if (vec_len (em->buffer_cache) < vec_len (intfc->rx_queue) * 2) + { + vec_validate (em->buffer_cache, + n_to_alloc + vec_len (em->buffer_cache) - 1); + n_allocated = + vlib_buffer_alloc (vm, &em->buffer_cache[n_present_in_cache], + n_to_alloc); + + n_present_in_cache += n_allocated; + _vec_len (em->buffer_cache) = n_present_in_cache; + } + + elts = (ssvm_eth_queue_elt_t *) (sh->opaque[CHUNK_POOL_INDEX]); + + n_buffers = vec_len (intfc->rx_queue); + rx_queue_index = 0; + + while (n_buffers > 0) + { + vlib_get_next_frame (vm, node, next_index, to_next, n_left_to_next); + + while (n_buffers > 0 && n_left_to_next > 0) + { + elt = elts + intfc->rx_queue[rx_queue_index]; + + saved_cache_size = n_present_in_cache; + if (PREDICT_FALSE (saved_cache_size == 0)) + { + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + goto out; + } + saved_bi0 = bi0 = em->buffer_cache[--n_present_in_cache]; + b0 = vlib_get_buffer (vm, bi0); + prev = 0; + + while (1) + { + vlib_buffer_init_for_free_list (b0, fl); + + b0->current_data = elt->current_data_hint; + b0->current_length = elt->length_this_buffer; + b0->total_length_not_including_first_buffer = + elt->total_length_not_including_first_buffer; + + clib_memcpy (b0->data + b0->current_data, elt->data, + b0->current_length); + + if (PREDICT_FALSE (prev != 0)) + prev->next_buffer = bi0; + + if (PREDICT_FALSE (elt->flags & SSVM_BUFFER_NEXT_PRESENT)) + { + prev = b0; + if (PREDICT_FALSE (n_present_in_cache == 0)) + { + vlib_put_next_frame (vm, node, next_index, + n_left_to_next); + goto out; + } + bi0 = em->buffer_cache[--n_present_in_cache]; + b0 = vlib_get_buffer (vm, bi0); + } + else + break; + } + + saved_cache_size = n_present_in_cache; + + to_next[0] = saved_bi0; + to_next++; + n_left_to_next--; + + b0 = vlib_get_buffer (vm, saved_bi0); + eh0 = vlib_buffer_get_current (b0); + + type0 = clib_net_to_host_u16 (eh0->type); + + next0 = SSVM_ETH_INPUT_NEXT_ETHERNET_INPUT; + + if (type0 == ETHERNET_TYPE_IP4) + next0 = SSVM_ETH_INPUT_NEXT_IP4_INPUT; + else if (type0 == ETHERNET_TYPE_IP6) + next0 = SSVM_ETH_INPUT_NEXT_IP6_INPUT; + else if (type0 == ETHERNET_TYPE_MPLS_UNICAST) + next0 = SSVM_ETH_INPUT_NEXT_MPLS_INPUT; + + l3_offset0 = ((next0 == SSVM_ETH_INPUT_NEXT_IP4_INPUT || + next0 == SSVM_ETH_INPUT_NEXT_IP6_INPUT || + next0 == SSVM_ETH_INPUT_NEXT_MPLS_INPUT) ? + sizeof (ethernet_header_t) : 0); + + n_rx_bytes += b0->current_length + + b0->total_length_not_including_first_buffer; + + b0->current_data += l3_offset0; + b0->current_length -= l3_offset0; + b0->flags = VLIB_BUFFER_TOTAL_LENGTH_VALID; + + vnet_buffer (b0)->sw_if_index[VLIB_RX] = intfc->vlib_hw_if_index; + vnet_buffer (b0)->sw_if_index[VLIB_TX] = (u32) ~ 0; + + /* + * Turn this on if you run into + * "bad monkey" contexts, and you want to know exactly + * which nodes they've visited... See main.c... + */ + VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b0); + + if (PREDICT_FALSE (n_trace > 0)) + { + ssvm_eth_input_trace_t *tr; + + vlib_trace_buffer (vm, node, next0, b0, /* follow_chain */ 1); + vlib_set_trace_count (vm, node, --n_trace); + + tr = vlib_add_trace (vm, node, b0, sizeof (*tr)); + + tr->next_index = next0; + tr->sw_if_index = intfc->vlib_hw_if_index; + } + + vlib_validate_buffer_enqueue_x1 (vm, node, next_index, + to_next, n_left_to_next, + bi0, next0); + n_buffers--; + rx_queue_index++; + } + + vlib_put_next_frame (vm, node, next_index, n_left_to_next); + } + +out: + if (em->buffer_cache) + _vec_len (em->buffer_cache) = saved_cache_size; + else + ASSERT (saved_cache_size == 0); + + ssvm_lock (sh, my_pid, 2); + + ASSERT (vec_len (intfc->rx_queue) > 0); + + n_available = (u32) pointer_to_uword (sh->opaque[CHUNK_POOL_NFREE]); + elt_indices = (u32 *) (sh->opaque[CHUNK_POOL_FREELIST_INDEX]); + + clib_memcpy (&elt_indices[n_available], intfc->rx_queue, + vec_len (intfc->rx_queue) * sizeof (u32)); + + n_available += vec_len (intfc->rx_queue); + sh->opaque[CHUNK_POOL_NFREE] = uword_to_pointer (n_available, void *); + + ssvm_unlock (sh); + + vlib_error_count (vm, node->node_index, SSVM_ETH_INPUT_ERROR_NO_BUFFERS, + n_buffers); + + vlib_increment_combined_counter + (vnet_get_main ()->interface_main.combined_sw_if_counters + + VNET_INTERFACE_COUNTER_RX, cpu_index, + intfc->vlib_hw_if_index, rx_queue_index, n_rx_bytes); + + return rx_queue_index; +} + +static uword +ssvm_eth_input_node_fn (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * frame) +{ + ssvm_eth_main_t *em = &ssvm_eth_main; + ssvm_private_t *intfc; + uword n_rx_packets = 0; + + vec_foreach (intfc, em->intfcs) + { + n_rx_packets += ssvm_eth_device_input (em, intfc, node); + } + + return n_rx_packets; +} + +/* *INDENT-OFF* */ +VLIB_REGISTER_NODE (ssvm_eth_input_node) = { + .function = ssvm_eth_input_node_fn, + .name = "ssvm_eth_input", + .vector_size = sizeof (u32), + .format_trace = format_ssvm_eth_input_trace, + .type = VLIB_NODE_TYPE_INPUT, + .state = VLIB_NODE_STATE_DISABLED, + + .n_errors = ARRAY_LEN(ssvm_eth_input_error_strings), + .error_strings = ssvm_eth_input_error_strings, + + .n_next_nodes = SSVM_ETH_INPUT_N_NEXT, + + /* edit / add dispositions here */ + .next_nodes = { + [SSVM_ETH_INPUT_NEXT_DROP] = "error-drop", + [SSVM_ETH_INPUT_NEXT_ETHERNET_INPUT] = "ethernet-input", + [SSVM_ETH_INPUT_NEXT_IP4_INPUT] = "ip4-input", + [SSVM_ETH_INPUT_NEXT_IP6_INPUT] = "ip6-input", + [SSVM_ETH_INPUT_NEXT_MPLS_INPUT] = "mpls-input", + }, +}; + +VLIB_NODE_FUNCTION_MULTIARCH (ssvm_eth_input_node, ssvm_eth_input_node_fn) +/* *INDENT-ON* */ + + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/devices/ssvm/ssvm_eth.c b/src/vnet/devices/ssvm/ssvm_eth.c new file mode 100644 index 00000000000..db4fafa9a14 --- /dev/null +++ b/src/vnet/devices/ssvm/ssvm_eth.c @@ -0,0 +1,491 @@ +/* + * Copyright (c) 2015 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#include "ssvm_eth.h" + +ssvm_eth_main_t ssvm_eth_main; + +#define foreach_ssvm_eth_tx_func_error \ +_(RING_FULL, "Tx packet drops (ring full)") \ +_(NO_BUFFERS, "Tx packet drops (no buffers)") \ +_(ADMIN_DOWN, "Tx packet drops (admin down)") + +typedef enum +{ +#define _(f,s) SSVM_ETH_TX_ERROR_##f, + foreach_ssvm_eth_tx_func_error +#undef _ + SSVM_ETH_TX_N_ERROR, +} ssvm_eth_tx_func_error_t; + +static u32 ssvm_eth_flag_change (vnet_main_t * vnm, + vnet_hw_interface_t * hi, u32 flags); + +int +ssvm_eth_create (ssvm_eth_main_t * em, u8 * name, int is_master) +{ + ssvm_private_t *intfc; + void *oldheap; + clib_error_t *e; + unix_shared_memory_queue_t *q; + ssvm_shared_header_t *sh; + ssvm_eth_queue_elt_t *elts; + u32 *elt_indices; + u8 enet_addr[6]; + int i, rv; + + vec_add2 (em->intfcs, intfc, 1); + + intfc->ssvm_size = em->segment_size; + intfc->i_am_master = 1; + intfc->name = name; + intfc->my_pid = getpid (); + if (is_master == 0) + { + rv = ssvm_slave_init (intfc, 20 /* timeout in seconds */ ); + if (rv < 0) + return rv; + goto create_vnet_interface; + } + + intfc->requested_va = em->next_base_va; + em->next_base_va += em->segment_size; + rv = ssvm_master_init (intfc, intfc - em->intfcs /* master index */ ); + + if (rv < 0) + return rv; + + /* OK, segment created, set up queues and so forth. */ + + sh = intfc->sh; + oldheap = ssvm_push_heap (sh); + + q = unix_shared_memory_queue_init (em->queue_elts, sizeof (u32), + 0 /* consumer pid not interesting */ , + 0 /* signal not sent */ ); + sh->opaque[TO_MASTER_Q_INDEX] = (void *) q; + q = unix_shared_memory_queue_init (em->queue_elts, sizeof (u32), + 0 /* consumer pid not interesting */ , + 0 /* signal not sent */ ); + sh->opaque[TO_SLAVE_Q_INDEX] = (void *) q; + + /* + * Preallocate the requested number of buffer chunks + * There must be a better way to do this, etc. + * Add some slop to avoid pool reallocation, which will not go well + */ + elts = 0; + elt_indices = 0; + + vec_validate_aligned (elts, em->nbuffers - 1, CLIB_CACHE_LINE_BYTES); + vec_validate_aligned (elt_indices, em->nbuffers - 1, CLIB_CACHE_LINE_BYTES); + + for (i = 0; i < em->nbuffers; i++) + elt_indices[i] = i; + + sh->opaque[CHUNK_POOL_INDEX] = (void *) elts; + sh->opaque[CHUNK_POOL_FREELIST_INDEX] = (void *) elt_indices; + sh->opaque[CHUNK_POOL_NFREE] = (void *) (uword) em->nbuffers; + + ssvm_pop_heap (oldheap); + +create_vnet_interface: + + sh = intfc->sh; + + memset (enet_addr, 0, sizeof (enet_addr)); + enet_addr[0] = 2; + enet_addr[1] = 0xFE; + enet_addr[2] = is_master; + enet_addr[5] = sh->master_index; + + e = ethernet_register_interface + (em->vnet_main, ssvm_eth_device_class.index, intfc - em->intfcs, + /* ethernet address */ enet_addr, + &intfc->vlib_hw_if_index, ssvm_eth_flag_change); + + if (e) + { + clib_error_report (e); + /* $$$$ unmap offending region? */ + return VNET_API_ERROR_INVALID_INTERFACE; + } + + /* Declare link up */ + vnet_hw_interface_set_flags (em->vnet_main, intfc->vlib_hw_if_index, + VNET_HW_INTERFACE_FLAG_LINK_UP); + + /* Let the games begin... */ + if (is_master) + sh->ready = 1; + return 0; +} + +static clib_error_t * +ssvm_config (vlib_main_t * vm, unformat_input_t * input) +{ + u8 *name; + int is_master = 1; + int i, rv; + ssvm_eth_main_t *em = &ssvm_eth_main; + + while (unformat_check_input (input) != UNFORMAT_END_OF_INPUT) + { + if (unformat (input, "base-va %llx", &em->next_base_va)) + ; + else if (unformat (input, "segment-size %lld", &em->segment_size)) + em->segment_size = 1ULL << (max_log2 (em->segment_size)); + else if (unformat (input, "nbuffers %lld", &em->nbuffers)) + ; + else if (unformat (input, "queue-elts %lld", &em->queue_elts)) + ; + else if (unformat (input, "slave")) + is_master = 0; + else if (unformat (input, "%s", &name)) + vec_add1 (em->names, name); + else + break; + } + + /* No configured instances, we're done... */ + if (vec_len (em->names) == 0) + return 0; + + for (i = 0; i < vec_len (em->names); i++) + { + rv = ssvm_eth_create (em, em->names[i], is_master); + if (rv < 0) + return clib_error_return (0, "ssvm_eth_create '%s' failed, error %d", + em->names[i], rv); + } + + vlib_node_set_state (vm, ssvm_eth_input_node.index, + VLIB_NODE_STATE_POLLING); + + return 0; +} + +VLIB_CONFIG_FUNCTION (ssvm_config, "ssvm_eth"); + + +static clib_error_t * +ssvm_eth_init (vlib_main_t * vm) +{ + ssvm_eth_main_t *em = &ssvm_eth_main; + + if (((sizeof (ssvm_eth_queue_elt_t) / CLIB_CACHE_LINE_BYTES) + * CLIB_CACHE_LINE_BYTES) != sizeof (ssvm_eth_queue_elt_t)) + clib_warning ("ssvm_eth_queue_elt_t size %d not a multiple of %d", + sizeof (ssvm_eth_queue_elt_t), CLIB_CACHE_LINE_BYTES); + + em->vlib_main = vm; + em->vnet_main = vnet_get_main (); + em->elog_main = &vm->elog_main; + + /* default config param values... */ + + em->next_base_va = 0x600000000ULL; + /* + * Allocate 2 full superframes in each dir (256 x 2 x 2 x 2048 bytes), + * 2mb; double that so we have plenty of space... 4mb + */ + em->segment_size = 8 << 20; + em->nbuffers = 1024; + em->queue_elts = 512; + return 0; +} + +VLIB_INIT_FUNCTION (ssvm_eth_init); + +static char *ssvm_eth_tx_func_error_strings[] = { +#define _(n,s) s, + foreach_ssvm_eth_tx_func_error +#undef _ +}; + +static u8 * +format_ssvm_eth_device_name (u8 * s, va_list * args) +{ + u32 i = va_arg (*args, u32); + + s = format (s, "ssvmEthernet%d", i); + return s; +} + +static u8 * +format_ssvm_eth_device (u8 * s, va_list * args) +{ + s = format (s, "SSVM Ethernet"); + return s; +} + +static u8 * +format_ssvm_eth_tx_trace (u8 * s, va_list * args) +{ + s = format (s, "Unimplemented..."); + return s; +} + + +static uword +ssvm_eth_interface_tx (vlib_main_t * vm, + vlib_node_runtime_t * node, vlib_frame_t * f) +{ + ssvm_eth_main_t *em = &ssvm_eth_main; + vnet_interface_output_runtime_t *rd = (void *) node->runtime_data; + ssvm_private_t *intfc = vec_elt_at_index (em->intfcs, rd->dev_instance); + ssvm_shared_header_t *sh = intfc->sh; + unix_shared_memory_queue_t *q; + u32 *from; + u32 n_left; + ssvm_eth_queue_elt_t *elts, *elt, *prev_elt; + u32 my_pid = intfc->my_pid; + vlib_buffer_t *b0; + u32 bi0; + u32 size_this_buffer; + u32 chunks_this_buffer; + u8 i_am_master = intfc->i_am_master; + u32 elt_index; + int is_ring_full, interface_down; + int i; + volatile u32 *queue_lock; + u32 n_to_alloc = VLIB_FRAME_SIZE; + u32 n_allocated, n_present_in_cache, n_available; + u32 *elt_indices; + + if (i_am_master) + q = (unix_shared_memory_queue_t *) sh->opaque[TO_SLAVE_Q_INDEX]; + else + q = (unix_shared_memory_queue_t *) sh->opaque[TO_MASTER_Q_INDEX]; + + queue_lock = (u32 *) q; + + from = vlib_frame_vector_args (f); + n_left = f->n_vectors; + is_ring_full = 0; + interface_down = 0; + + n_present_in_cache = vec_len (em->chunk_cache); + + /* admin / link up/down check */ + if (sh->opaque[MASTER_ADMIN_STATE_INDEX] == 0 || + sh->opaque[SLAVE_ADMIN_STATE_INDEX] == 0) + { + interface_down = 1; + goto out; + } + + ssvm_lock (sh, my_pid, 1); + + elts = (ssvm_eth_queue_elt_t *) (sh->opaque[CHUNK_POOL_INDEX]); + elt_indices = (u32 *) (sh->opaque[CHUNK_POOL_FREELIST_INDEX]); + n_available = (u32) pointer_to_uword (sh->opaque[CHUNK_POOL_NFREE]); + + if (n_present_in_cache < n_left * 2) + { + vec_validate (em->chunk_cache, n_to_alloc + n_present_in_cache - 1); + + n_allocated = n_to_alloc < n_available ? n_to_alloc : n_available; + + if (PREDICT_TRUE (n_allocated > 0)) + { + clib_memcpy (&em->chunk_cache[n_present_in_cache], + &elt_indices[n_available - n_allocated], + sizeof (u32) * n_allocated); + } + + n_present_in_cache += n_allocated; + n_available -= n_allocated; + sh->opaque[CHUNK_POOL_NFREE] = uword_to_pointer (n_available, void *); + _vec_len (em->chunk_cache) = n_present_in_cache; + } + + ssvm_unlock (sh); + + while (n_left) + { + bi0 = from[0]; + b0 = vlib_get_buffer (vm, bi0); + + size_this_buffer = vlib_buffer_length_in_chain (vm, b0); + chunks_this_buffer = (size_this_buffer + (SSVM_BUFFER_SIZE - 1)) + / SSVM_BUFFER_SIZE; + + /* If we're not going to be able to enqueue the buffer, tail drop. */ + if (q->cursize >= q->maxsize) + { + is_ring_full = 1; + break; + } + + prev_elt = 0; + elt_index = ~0; + for (i = 0; i < chunks_this_buffer; i++) + { + if (PREDICT_FALSE (n_present_in_cache == 0)) + goto out; + + elt_index = em->chunk_cache[--n_present_in_cache]; + elt = elts + elt_index; + + elt->type = SSVM_PACKET_TYPE; + elt->flags = 0; + elt->total_length_not_including_first_buffer = + b0->total_length_not_including_first_buffer; + elt->length_this_buffer = b0->current_length; + elt->current_data_hint = b0->current_data; + elt->owner = !i_am_master; + elt->tag = 1; + + clib_memcpy (elt->data, b0->data + b0->current_data, + b0->current_length); + + if (PREDICT_FALSE (prev_elt != 0)) + prev_elt->next_index = elt - elts; + + if (PREDICT_FALSE (i < (chunks_this_buffer - 1))) + { + elt->flags = SSVM_BUFFER_NEXT_PRESENT; + ASSERT (b0->flags & VLIB_BUFFER_NEXT_PRESENT); + b0 = vlib_get_buffer (vm, b0->next_buffer); + } + prev_elt = elt; + } + + while (__sync_lock_test_and_set (queue_lock, 1)) + ; + + unix_shared_memory_queue_add_raw (q, (u8 *) & elt_index); + CLIB_MEMORY_BARRIER (); + *queue_lock = 0; + + from++; + n_left--; + } + +out: + if (PREDICT_FALSE (n_left)) + { + if (is_ring_full) + vlib_error_count (vm, node->node_index, SSVM_ETH_TX_ERROR_RING_FULL, + n_left); + else if (interface_down) + vlib_error_count (vm, node->node_index, SSVM_ETH_TX_ERROR_ADMIN_DOWN, + n_left); + else + vlib_error_count (vm, node->node_index, SSVM_ETH_TX_ERROR_NO_BUFFERS, + n_left); + + vlib_buffer_free (vm, from, n_left); + } + else + vlib_buffer_free (vm, vlib_frame_vector_args (f), f->n_vectors); + + if (PREDICT_TRUE (vec_len (em->chunk_cache))) + _vec_len (em->chunk_cache) = n_present_in_cache; + + return f->n_vectors; +} + +static void +ssvm_eth_clear_hw_interface_counters (u32 instance) +{ + /* Nothing for now */ +} + +static clib_error_t * +ssvm_eth_interface_admin_up_down (vnet_main_t * vnm, u32 hw_if_index, + u32 flags) +{ + vnet_hw_interface_t *hif = vnet_get_hw_interface (vnm, hw_if_index); + uword is_up = (flags & VNET_SW_INTERFACE_FLAG_ADMIN_UP) != 0; + ssvm_eth_main_t *em = &ssvm_eth_main; + ssvm_private_t *intfc = vec_elt_at_index (em->intfcs, hif->dev_instance); + ssvm_shared_header_t *sh; + + /* publish link-state in shared-memory, to discourage buffer-wasting */ + sh = intfc->sh; + if (intfc->i_am_master) + sh->opaque[MASTER_ADMIN_STATE_INDEX] = (void *) is_up; + else + sh->opaque[SLAVE_ADMIN_STATE_INDEX] = (void *) is_up; + + return 0; +} + +static clib_error_t * +ssvm_eth_subif_add_del_function (vnet_main_t * vnm, + u32 hw_if_index, + struct vnet_sw_interface_t *st, int is_add) +{ + /* Nothing for now */ + return 0; +} + +/* + * Dynamically redirect all pkts from a specific interface + * to the specified node + */ +static void +ssvm_eth_set_interface_next_node (vnet_main_t * vnm, u32 hw_if_index, + u32 node_index) +{ + ssvm_eth_main_t *em = &ssvm_eth_main; + vnet_hw_interface_t *hw = vnet_get_hw_interface (vnm, hw_if_index); + ssvm_private_t *intfc = pool_elt_at_index (em->intfcs, hw->dev_instance); + + /* Shut off redirection */ + if (node_index == ~0) + { + intfc->per_interface_next_index = node_index; + return; + } + + intfc->per_interface_next_index = + vlib_node_add_next (em->vlib_main, ssvm_eth_input_node.index, node_index); +} + +static u32 +ssvm_eth_flag_change (vnet_main_t * vnm, vnet_hw_interface_t * hi, u32 flags) +{ + /* nothing for now */ + return 0; +} + +/* *INDENT-OFF* */ +VNET_DEVICE_CLASS (ssvm_eth_device_class) = { + .name = "ssvm-eth", + .tx_function = ssvm_eth_interface_tx, + .tx_function_n_errors = SSVM_ETH_TX_N_ERROR, + .tx_function_error_strings = ssvm_eth_tx_func_error_strings, + .format_device_name = format_ssvm_eth_device_name, + .format_device = format_ssvm_eth_device, + .format_tx_trace = format_ssvm_eth_tx_trace, + .clear_counters = ssvm_eth_clear_hw_interface_counters, + .admin_up_down_function = ssvm_eth_interface_admin_up_down, + .subif_add_del_function = ssvm_eth_subif_add_del_function, + .rx_redirect_to_node = ssvm_eth_set_interface_next_node, +}; + +VLIB_DEVICE_TX_FUNCTION_MULTIARCH (ssvm_eth_device_class, + ssvm_eth_interface_tx) +/* *INDENT-ON* */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ diff --git a/src/vnet/devices/ssvm/ssvm_eth.h b/src/vnet/devices/ssvm/ssvm_eth.h new file mode 100644 index 00000000000..f877df3cd33 --- /dev/null +++ b/src/vnet/devices/ssvm/ssvm_eth.h @@ -0,0 +1,141 @@ +/* + * Copyright (c) 2015 Cisco and/or its affiliates. + * Licensed under the Apache License, Version 2.0 (the "License"); + * you may not use this file except in compliance with the License. + * You may obtain a copy of the License at: + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +#ifndef __included_ssvm_eth_h__ +#define __included_ssvm_eth_h__ + +#include <vnet/vnet.h> + +#include <vppinfra/elog.h> +#include <vppinfra/error.h> +#include <vppinfra/format.h> +#include <vppinfra/hash.h> +#include <vppinfra/vec.h> +#include <vppinfra/elog.h> +#include <vlib/vlib.h> +#include <vnet/ethernet/ethernet.h> +#include <vnet/devices/devices.h> +#include <vnet/ip/ip.h> +#include <vnet/pg/pg.h> +#include <vlibmemory/unix_shared_memory_queue.h> + +#include <svm/ssvm.h> + +extern vnet_device_class_t ssvm_eth_device_class; +extern vlib_node_registration_t ssvm_eth_input_node; + +#define SSVM_BUFFER_SIZE \ + (VLIB_BUFFER_DATA_SIZE + VLIB_BUFFER_PRE_DATA_SIZE) +#define SSVM_PACKET_TYPE 1 + +typedef struct +{ + /* Type of queue element */ + u8 type; + u8 flags; +#define SSVM_BUFFER_NEXT_PRESENT (1<<0) + u8 owner; + u8 tag; + i16 current_data_hint; + u16 length_this_buffer; + u16 total_length_not_including_first_buffer; + u16 pad; + u32 next_index; + /* offset 16 */ + u8 data[SSVM_BUFFER_SIZE]; + /* pad to an even multiple of 64 octets */ + u8 pad2[CLIB_CACHE_LINE_BYTES - 16]; +} ssvm_eth_queue_elt_t; + +typedef struct +{ + /* vector of point-to-point connections */ + ssvm_private_t *intfcs; + + u32 *buffer_cache; + u32 *chunk_cache; + + /* Configurable parameters */ + /* base address for next placement */ + u64 next_base_va; + u64 segment_size; + u64 nbuffers; + u64 queue_elts; + + /* Segment names */ + u8 **names; + + /* convenience */ + vlib_main_t *vlib_main; + vnet_main_t *vnet_main; + elog_main_t *elog_main; +} ssvm_eth_main_t; + +ssvm_eth_main_t ssvm_eth_main; + +typedef enum +{ + CHUNK_POOL_FREELIST_INDEX = 0, + CHUNK_POOL_INDEX, + CHUNK_POOL_NFREE, + TO_MASTER_Q_INDEX, + TO_SLAVE_Q_INDEX, + MASTER_ADMIN_STATE_INDEX, + SLAVE_ADMIN_STATE_INDEX, +} ssvm_eth_opaque_index_t; + +/* + * debug scaffolding. + */ +static inline void +ssvm_eth_validate_freelists (int need_lock) +{ +#if CLIB_DEBUG > 0 + ssvm_eth_main_t *em = &ssvm_eth_main; + ssvm_private_t *intfc; + ssvm_shared_header_t *sh; + u32 *elt_indices; + u32 n_available; + int i; + + for (i = 0; i < vec_len (em->intfcs); i++) + { + intfc = em->intfcs + i; + sh = intfc->sh; + u32 my_pid = intfc->my_pid; + + if (need_lock) + ssvm_lock (sh, my_pid, 15); + + elt_indices = (u32 *) (sh->opaque[CHUNK_POOL_FREELIST_INDEX]); + n_available = (u32) (uword) (sh->opaque[CHUNK_POOL_NFREE]); + + for (i = 0; i < n_available; i++) + ASSERT (elt_indices[i] < 2048); + + if (need_lock) + ssvm_unlock (sh); + } +#endif +} + +#endif /* __included_ssvm_eth_h__ */ + +/* + * fd.io coding-style-patch-verification: ON + * + * Local Variables: + * eval: (c-set-style "gnu") + * End: + */ |