blob: 4fdfbd5d6e0d7b313c2b33e38174122d4ac06855 [file] [log] [blame]
Mohsin Kazmie7cde312018-06-26 17:20:11 +02001/*
2 *------------------------------------------------------------------
3 * vhost-user-input
4 *
5 * Copyright (c) 2014-2018 Cisco and/or its affiliates.
6 * Licensed under the Apache License, Version 2.0 (the "License");
7 * you may not use this file except in compliance with the License.
8 * You may obtain a copy of the License at:
9 *
10 * http://www.apache.org/licenses/LICENSE-2.0
11 *
12 * Unless required by applicable law or agreed to in writing, software
13 * distributed under the License is distributed on an "AS IS" BASIS,
14 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
15 * See the License for the specific language governing permissions and
16 * limitations under the License.
17 *------------------------------------------------------------------
18 */
19
20#include <fcntl.h> /* for open */
21#include <sys/ioctl.h>
22#include <sys/socket.h>
23#include <sys/un.h>
24#include <sys/stat.h>
25#include <sys/types.h>
26#include <sys/uio.h> /* for iovec */
27#include <netinet/in.h>
28#include <sys/vfs.h>
29
30#include <linux/if_arp.h>
31#include <linux/if_tun.h>
32
33#include <vlib/vlib.h>
34#include <vlib/unix/unix.h>
35
36#include <vnet/ip/ip.h>
37
38#include <vnet/ethernet/ethernet.h>
39#include <vnet/devices/devices.h>
40#include <vnet/feature/feature.h>
41
42#include <vnet/devices/virtio/vhost_user.h>
43#include <vnet/devices/virtio/vhost_user_inline.h>
44
45/*
46 * When an RX queue is down but active, received packets
47 * must be discarded. This value controls up to how many
48 * packets will be discarded during each round.
49 */
50#define VHOST_USER_DOWN_DISCARD_COUNT 256
51
52/*
53 * When the number of available buffers gets under this threshold,
54 * RX node will start discarding packets.
55 */
56#define VHOST_USER_RX_BUFFER_STARVATION 32
57
58/*
59 * On the receive side, the host should free descriptors as soon
60 * as possible in order to avoid TX drop in the VM.
61 * This value controls the number of copy operations that are stacked
62 * before copy is done for all and descriptors are given back to
63 * the guest.
64 * The value 64 was obtained by testing (48 and 128 were not as good).
65 */
66#define VHOST_USER_RX_COPY_THRESHOLD 64
67
68vlib_node_registration_t vhost_user_input_node;
69
70#define foreach_vhost_user_input_func_error \
71 _(NO_ERROR, "no error") \
72 _(NO_BUFFER, "no available buffer") \
73 _(MMAP_FAIL, "mmap failure") \
74 _(INDIRECT_OVERFLOW, "indirect descriptor overflows table") \
75 _(UNDERSIZED_FRAME, "undersized ethernet frame received (< 14 bytes)") \
76 _(FULL_RX_QUEUE, "full rx queue (possible driver tx drop)")
77
78typedef enum
79{
80#define _(f,s) VHOST_USER_INPUT_FUNC_ERROR_##f,
81 foreach_vhost_user_input_func_error
82#undef _
83 VHOST_USER_INPUT_FUNC_N_ERROR,
84} vhost_user_input_func_error_t;
85
86static __clib_unused char *vhost_user_input_func_error_strings[] = {
87#define _(n,s) s,
88 foreach_vhost_user_input_func_error
89#undef _
90};
91
92static_always_inline void
93vhost_user_rx_trace (vhost_trace_t * t,
94 vhost_user_intf_t * vui, u16 qid,
95 vlib_buffer_t * b, vhost_user_vring_t * txvq)
96{
97 vhost_user_main_t *vum = &vhost_user_main;
98 u32 last_avail_idx = txvq->last_avail_idx;
99 u32 desc_current = txvq->avail->ring[last_avail_idx & txvq->qsz_mask];
100 vring_desc_t *hdr_desc = 0;
101 virtio_net_hdr_mrg_rxbuf_t *hdr;
102 u32 hint = 0;
103
Dave Barachb7b92992018-10-17 10:38:51 -0400104 clib_memset (t, 0, sizeof (*t));
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200105 t->device_index = vui - vum->vhost_user_interfaces;
106 t->qid = qid;
107
108 hdr_desc = &txvq->desc[desc_current];
109 if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT)
110 {
111 t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_INDIRECT;
112 /* Header is the first here */
113 hdr_desc = map_guest_mem (vui, txvq->desc[desc_current].addr, &hint);
114 }
115 if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT)
116 {
117 t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SIMPLE_CHAINED;
118 }
119 if (!(txvq->desc[desc_current].flags & VIRTQ_DESC_F_NEXT) &&
120 !(txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT))
121 {
122 t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_SINGLE_DESC;
123 }
124
125 t->first_desc_len = hdr_desc ? hdr_desc->len : 0;
126
127 if (!hdr_desc || !(hdr = map_guest_mem (vui, hdr_desc->addr, &hint)))
128 {
129 t->virtio_ring_flags |= 1 << VIRTIO_TRACE_F_MAP_ERROR;
130 }
131 else
132 {
133 u32 len = vui->virtio_net_hdr_sz;
134 memcpy (&t->hdr, hdr, len > hdr_desc->len ? hdr_desc->len : len);
135 }
136}
137
138static_always_inline u32
139vhost_user_input_copy (vhost_user_intf_t * vui, vhost_copy_t * cpy,
140 u16 copy_len, u32 * map_hint)
141{
142 void *src0, *src1, *src2, *src3;
143 if (PREDICT_TRUE (copy_len >= 4))
144 {
145 if (PREDICT_FALSE (!(src2 = map_guest_mem (vui, cpy[0].src, map_hint))))
146 return 1;
147 if (PREDICT_FALSE (!(src3 = map_guest_mem (vui, cpy[1].src, map_hint))))
148 return 1;
149
150 while (PREDICT_TRUE (copy_len >= 4))
151 {
152 src0 = src2;
153 src1 = src3;
154
155 if (PREDICT_FALSE
156 (!(src2 = map_guest_mem (vui, cpy[2].src, map_hint))))
157 return 1;
158 if (PREDICT_FALSE
159 (!(src3 = map_guest_mem (vui, cpy[3].src, map_hint))))
160 return 1;
161
162 CLIB_PREFETCH (src2, 64, LOAD);
163 CLIB_PREFETCH (src3, 64, LOAD);
164
Dave Barach178cf492018-11-13 16:34:13 -0500165 clib_memcpy_fast ((void *) cpy[0].dst, src0, cpy[0].len);
166 clib_memcpy_fast ((void *) cpy[1].dst, src1, cpy[1].len);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200167 copy_len -= 2;
168 cpy += 2;
169 }
170 }
171 while (copy_len)
172 {
173 if (PREDICT_FALSE (!(src0 = map_guest_mem (vui, cpy->src, map_hint))))
174 return 1;
Dave Barach178cf492018-11-13 16:34:13 -0500175 clib_memcpy_fast ((void *) cpy->dst, src0, cpy->len);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200176 copy_len -= 1;
177 cpy += 1;
178 }
179 return 0;
180}
181
182/**
183 * Try to discard packets from the tx ring (VPP RX path).
184 * Returns the number of discarded packets.
185 */
186static_always_inline u32
187vhost_user_rx_discard_packet (vlib_main_t * vm,
188 vhost_user_intf_t * vui,
189 vhost_user_vring_t * txvq, u32 discard_max)
190{
191 /*
192 * On the RX side, each packet corresponds to one descriptor
193 * (it is the same whether it is a shallow descriptor, chained, or indirect).
194 * Therefore, discarding a packet is like discarding a descriptor.
195 */
196 u32 discarded_packets = 0;
197 u32 avail_idx = txvq->avail->idx;
198 while (discarded_packets != discard_max)
199 {
200 if (avail_idx == txvq->last_avail_idx)
201 goto out;
202
203 u16 desc_chain_head =
204 txvq->avail->ring[txvq->last_avail_idx & txvq->qsz_mask];
205 txvq->last_avail_idx++;
206 txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].id =
207 desc_chain_head;
208 txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].len = 0;
209 vhost_user_log_dirty_ring (vui, txvq,
210 ring[txvq->last_used_idx & txvq->qsz_mask]);
211 txvq->last_used_idx++;
212 discarded_packets++;
213 }
214
215out:
216 CLIB_MEMORY_BARRIER ();
217 txvq->used->idx = txvq->last_used_idx;
218 vhost_user_log_dirty_ring (vui, txvq, idx);
219 return discarded_packets;
220}
221
222/*
223 * In case of overflow, we need to rewind the array of allocated buffers.
224 */
225static __clib_unused void
226vhost_user_input_rewind_buffers (vlib_main_t * vm,
227 vhost_cpu_t * cpu, vlib_buffer_t * b_head)
228{
229 u32 bi_current = cpu->rx_buffers[cpu->rx_buffers_len];
230 vlib_buffer_t *b_current = vlib_get_buffer (vm, bi_current);
231 b_current->current_length = 0;
232 b_current->flags = 0;
233 while (b_current != b_head)
234 {
235 cpu->rx_buffers_len++;
236 bi_current = cpu->rx_buffers[cpu->rx_buffers_len];
237 b_current = vlib_get_buffer (vm, bi_current);
238 b_current->current_length = 0;
239 b_current->flags = 0;
240 }
241 cpu->rx_buffers_len++;
242}
243
244static __clib_unused u32
245vhost_user_if_input (vlib_main_t * vm,
246 vhost_user_main_t * vum,
247 vhost_user_intf_t * vui,
248 u16 qid, vlib_node_runtime_t * node,
249 vnet_hw_interface_rx_mode mode)
250{
251 vhost_user_vring_t *txvq = &vui->vrings[VHOST_VRING_IDX_TX (qid)];
Damjan Marion9af45042018-11-21 09:51:42 +0100252 vnet_feature_main_t *fm = &feature_main;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200253 u16 n_rx_packets = 0;
254 u32 n_rx_bytes = 0;
255 u16 n_left;
256 u32 n_left_to_next, *to_next;
257 u32 next_index = VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT;
258 u32 n_trace = vlib_get_trace_count (vm, node);
259 u32 map_hint = 0;
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100260 vhost_cpu_t *cpu = &vum->cpus[vm->thread_index];
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200261 u16 copy_len = 0;
Damjan Marion9af45042018-11-21 09:51:42 +0100262 u8 feature_arc_idx = fm->device_input_feature_arc_index;
263 u32 current_config_index = ~(u32) 0;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200264
Yichen Wang28812a02018-08-28 23:05:27 -0700265 /* The descriptor table is not ready yet */
266 if (PREDICT_FALSE (txvq->avail == 0))
267 return 0;
268
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200269 {
270 /* do we have pending interrupts ? */
271 vhost_user_vring_t *rxvq = &vui->vrings[VHOST_VRING_IDX_RX (qid)];
272 f64 now = vlib_time_now (vm);
273
274 if ((txvq->n_since_last_int) && (txvq->int_deadline < now))
275 vhost_user_send_call (vm, txvq);
276
277 if ((rxvq->n_since_last_int) && (rxvq->int_deadline < now))
278 vhost_user_send_call (vm, rxvq);
279 }
280
281 /*
282 * For adaptive mode, it is optimized to reduce interrupts.
283 * If the scheduler switches the input node to polling due
284 * to burst of traffic, we tell the driver no interrupt.
285 * When the traffic subsides, the scheduler switches the node back to
286 * interrupt mode. We must tell the driver we want interrupt.
287 */
288 if (PREDICT_FALSE (mode == VNET_HW_INTERFACE_RX_MODE_ADAPTIVE))
289 {
290 if ((node->flags &
291 VLIB_NODE_FLAG_SWITCH_FROM_POLLING_TO_INTERRUPT_MODE) ||
292 !(node->flags &
293 VLIB_NODE_FLAG_SWITCH_FROM_INTERRUPT_TO_POLLING_MODE))
294 /* Tell driver we want notification */
295 txvq->used->flags = 0;
296 else
297 /* Tell driver we don't want notification */
298 txvq->used->flags = VRING_USED_F_NO_NOTIFY;
299 }
300
301 if (PREDICT_FALSE (txvq->avail->flags & 0xFFFE))
302 return 0;
303
304 n_left = (u16) (txvq->avail->idx - txvq->last_avail_idx);
305
306 /* nothing to do */
307 if (PREDICT_FALSE (n_left == 0))
308 return 0;
309
310 if (PREDICT_FALSE (!vui->admin_up || !(txvq->enabled)))
311 {
312 /*
313 * Discard input packet if interface is admin down or vring is not
314 * enabled.
315 * "For example, for a networking device, in the disabled state
316 * client must not supply any new RX packets, but must process
317 * and discard any TX packets."
318 */
319 vhost_user_rx_discard_packet (vm, vui, txvq,
320 VHOST_USER_DOWN_DISCARD_COUNT);
321 return 0;
322 }
323
324 if (PREDICT_FALSE (n_left == (txvq->qsz_mask + 1)))
325 {
326 /*
327 * Informational error logging when VPP is not
328 * receiving packets fast enough.
329 */
330 vlib_error_count (vm, node->node_index,
331 VHOST_USER_INPUT_FUNC_ERROR_FULL_RX_QUEUE, 1);
332 }
333
334 if (n_left > VLIB_FRAME_SIZE)
335 n_left = VLIB_FRAME_SIZE;
336
337 /*
338 * For small packets (<2kB), we will not need more than one vlib buffer
339 * per packet. In case packets are bigger, we will just yeld at some point
340 * in the loop and come back later. This is not an issue as for big packet,
341 * processing cost really comes from the memory copy.
342 * The assumption is that big packets will fit in 40 buffers.
343 */
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100344 if (PREDICT_FALSE (cpu->rx_buffers_len < n_left + 1 ||
345 cpu->rx_buffers_len < 40))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200346 {
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100347 u32 curr_len = cpu->rx_buffers_len;
348 cpu->rx_buffers_len +=
349 vlib_buffer_alloc_from_free_list (vm, cpu->rx_buffers + curr_len,
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200350 VHOST_USER_RX_BUFFERS_N - curr_len,
351 VLIB_BUFFER_DEFAULT_FREE_LIST_INDEX);
352
353 if (PREDICT_FALSE
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100354 (cpu->rx_buffers_len < VHOST_USER_RX_BUFFER_STARVATION))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200355 {
356 /* In case of buffer starvation, discard some packets from the queue
357 * and log the event.
358 * We keep doing best effort for the remaining packets. */
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100359 u32 flush = (n_left + 1 > cpu->rx_buffers_len) ?
360 n_left + 1 - cpu->rx_buffers_len : 1;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200361 flush = vhost_user_rx_discard_packet (vm, vui, txvq, flush);
362
363 n_left -= flush;
364 vlib_increment_simple_counter (vnet_main.
365 interface_main.sw_if_counters +
366 VNET_INTERFACE_COUNTER_DROP,
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100367 vm->thread_index, vui->sw_if_index,
368 flush);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200369
370 vlib_error_count (vm, vhost_user_input_node.index,
371 VHOST_USER_INPUT_FUNC_ERROR_NO_BUFFER, flush);
372 }
373 }
374
Damjan Marion9af45042018-11-21 09:51:42 +0100375 if (PREDICT_FALSE (vnet_have_features (feature_arc_idx, vui->sw_if_index)))
376 {
377 vnet_feature_config_main_t *cm;
378 cm = &fm->feature_config_mains[feature_arc_idx];
379 current_config_index = vec_elt (cm->config_index_by_sw_if_index,
380 vui->sw_if_index);
381 vnet_get_config_data (&cm->config_main, &current_config_index,
382 &next_index, 0);
383 }
384
Damjan Marion92825382018-11-21 10:03:44 +0100385 vlib_get_new_next_frame (vm, node, next_index, to_next, n_left_to_next);
386
387 if (next_index == VNET_DEVICE_INPUT_NEXT_ETHERNET_INPUT)
388 {
389 /* give some hints to ethernet-input */
390 vlib_next_frame_t *nf;
391 vlib_frame_t *f;
392 ethernet_input_frame_t *ef;
393 nf = vlib_node_runtime_get_next_frame (vm, node, next_index);
394 f = vlib_get_frame (vm, nf->frame_index);
395 f->flags = ETH_INPUT_FRAME_F_SINGLE_SW_IF_IDX;
396
397 ef = vlib_frame_scalar_args (f);
398 ef->sw_if_index = vui->sw_if_index;
399 ef->hw_if_index = vui->hw_if_index;
400 }
401
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200402 while (n_left > 0)
403 {
Damjan Marion92825382018-11-21 10:03:44 +0100404 vlib_buffer_t *b_head, *b_current;
405 u32 bi_current;
406 u16 desc_current;
407 u32 desc_data_offset;
408 vring_desc_t *desc_table = txvq->desc;
Damjan Marion6a8bfd42018-11-21 09:54:41 +0100409
Damjan Marion92825382018-11-21 10:03:44 +0100410 if (PREDICT_FALSE (cpu->rx_buffers_len <= 1))
Damjan Marion6a8bfd42018-11-21 09:54:41 +0100411 {
Damjan Marion92825382018-11-21 10:03:44 +0100412 /* Not enough rx_buffers
413 * Note: We yeld on 1 so we don't need to do an additional
414 * check for the next buffer prefetch.
415 */
416 n_left = 0;
417 break;
Damjan Marion6a8bfd42018-11-21 09:54:41 +0100418 }
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200419
Damjan Marion92825382018-11-21 10:03:44 +0100420 desc_current = txvq->avail->ring[txvq->last_avail_idx & txvq->qsz_mask];
421 cpu->rx_buffers_len--;
422 bi_current = cpu->rx_buffers[cpu->rx_buffers_len];
423 b_head = b_current = vlib_get_buffer (vm, bi_current);
424 to_next[0] = bi_current; //We do that now so we can forget about bi_current
425 to_next++;
426 n_left_to_next--;
427
428 vlib_prefetch_buffer_with_index
429 (vm, cpu->rx_buffers[cpu->rx_buffers_len - 1], LOAD);
430
431 /* Just preset the used descriptor id and length for later */
432 txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].id =
433 desc_current;
434 txvq->used->ring[txvq->last_used_idx & txvq->qsz_mask].len = 0;
435 vhost_user_log_dirty_ring (vui, txvq,
436 ring[txvq->last_used_idx & txvq->qsz_mask]);
437
438 /* The buffer should already be initialized */
439 b_head->total_length_not_including_first_buffer = 0;
440 b_head->flags |= VLIB_BUFFER_TOTAL_LENGTH_VALID;
441
442 if (PREDICT_FALSE (n_trace))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200443 {
Damjan Marion92825382018-11-21 10:03:44 +0100444 //TODO: next_index is not exactly known at that point
445 vlib_trace_buffer (vm, node, next_index, b_head,
446 /* follow_chain */ 0);
447 vhost_trace_t *t0 =
448 vlib_add_trace (vm, node, b_head, sizeof (t0[0]));
449 vhost_user_rx_trace (t0, vui, qid, b_head, txvq);
450 n_trace--;
451 vlib_set_trace_count (vm, node, n_trace);
452 }
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200453
Damjan Marion92825382018-11-21 10:03:44 +0100454 /* This depends on the setup but is very consistent
455 * So I think the CPU branch predictor will make a pretty good job
456 * at optimizing the decision. */
457 if (txvq->desc[desc_current].flags & VIRTQ_DESC_F_INDIRECT)
458 {
459 desc_table = map_guest_mem (vui, txvq->desc[desc_current].addr,
460 &map_hint);
461 desc_current = 0;
462 if (PREDICT_FALSE (desc_table == 0))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200463 {
Damjan Marion92825382018-11-21 10:03:44 +0100464 vlib_error_count (vm, node->node_index,
465 VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL, 1);
466 goto out;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200467 }
Damjan Marion92825382018-11-21 10:03:44 +0100468 }
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200469
Damjan Marion92825382018-11-21 10:03:44 +0100470 if (PREDICT_TRUE (vui->is_any_layout) ||
471 (!(desc_table[desc_current].flags & VIRTQ_DESC_F_NEXT)))
472 {
473 /* ANYLAYOUT or single buffer */
474 desc_data_offset = vui->virtio_net_hdr_sz;
475 }
476 else
477 {
478 /* CSR case without ANYLAYOUT, skip 1st buffer */
479 desc_data_offset = desc_table[desc_current].len;
480 }
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200481
Damjan Marion92825382018-11-21 10:03:44 +0100482 while (1)
483 {
484 /* Get more input if necessary. Or end of packet. */
485 if (desc_data_offset == desc_table[desc_current].len)
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200486 {
Damjan Marion92825382018-11-21 10:03:44 +0100487 if (PREDICT_FALSE (desc_table[desc_current].flags &
488 VIRTQ_DESC_F_NEXT))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200489 {
Damjan Marion92825382018-11-21 10:03:44 +0100490 desc_current = desc_table[desc_current].next;
491 desc_data_offset = 0;
492 }
493 else
494 {
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200495 goto out;
496 }
497 }
498
Damjan Marion92825382018-11-21 10:03:44 +0100499 /* Get more output if necessary. Or end of packet. */
500 if (PREDICT_FALSE
501 (b_current->current_length == VLIB_BUFFER_DATA_SIZE))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200502 {
Damjan Marion92825382018-11-21 10:03:44 +0100503 if (PREDICT_FALSE (cpu->rx_buffers_len == 0))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200504 {
Damjan Marion92825382018-11-21 10:03:44 +0100505 /* Cancel speculation */
506 to_next--;
507 n_left_to_next++;
508
509 /*
510 * Checking if there are some left buffers.
511 * If not, just rewind the used buffers and stop.
512 * Note: Scheduled copies are not cancelled. This is
513 * not an issue as they would still be valid. Useless,
514 * but valid.
515 */
516 vhost_user_input_rewind_buffers (vm, cpu, b_head);
517 n_left = 0;
518 goto stop;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200519 }
520
Damjan Marion92825382018-11-21 10:03:44 +0100521 /* Get next output */
522 cpu->rx_buffers_len--;
523 u32 bi_next = cpu->rx_buffers[cpu->rx_buffers_len];
524 b_current->next_buffer = bi_next;
525 b_current->flags |= VLIB_BUFFER_NEXT_PRESENT;
526 bi_current = bi_next;
527 b_current = vlib_get_buffer (vm, bi_current);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200528 }
529
Damjan Marion92825382018-11-21 10:03:44 +0100530 /* Prepare a copy order executed later for the data */
531 vhost_copy_t *cpy = &cpu->copy[copy_len];
532 copy_len++;
533 u32 desc_data_l = desc_table[desc_current].len - desc_data_offset;
534 cpy->len = VLIB_BUFFER_DATA_SIZE - b_current->current_length;
535 cpy->len = (cpy->len > desc_data_l) ? desc_data_l : cpy->len;
536 cpy->dst = (uword) (vlib_buffer_get_current (b_current) +
537 b_current->current_length);
538 cpy->src = desc_table[desc_current].addr + desc_data_offset;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200539
Damjan Marion92825382018-11-21 10:03:44 +0100540 desc_data_offset += cpy->len;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200541
Damjan Marion92825382018-11-21 10:03:44 +0100542 b_current->current_length += cpy->len;
543 b_head->total_length_not_including_first_buffer += cpy->len;
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200544 }
Damjan Marion92825382018-11-21 10:03:44 +0100545
546 out:
547
548 n_rx_bytes += b_head->total_length_not_including_first_buffer;
549 n_rx_packets++;
550
551 b_head->total_length_not_including_first_buffer -=
552 b_head->current_length;
553
554 /* consume the descriptor and return it as used */
555 txvq->last_avail_idx++;
556 txvq->last_used_idx++;
557
558 VLIB_BUFFER_TRACE_TRAJECTORY_INIT (b_head);
559
560 vnet_buffer (b_head)->sw_if_index[VLIB_RX] = vui->sw_if_index;
561 vnet_buffer (b_head)->sw_if_index[VLIB_TX] = (u32) ~ 0;
562 b_head->error = 0;
563
564 if (current_config_index != ~(u32) 0)
565 {
566 b_head->current_config_index = current_config_index;
567 vnet_buffer (b_head)->feature_arc_index = feature_arc_idx;
568 }
569
570 n_left--;
571
572 /*
573 * Although separating memory copies from virtio ring parsing
574 * is beneficial, we can offer to perform the copies from time
575 * to time in order to free some space in the ring.
576 */
577 if (PREDICT_FALSE (copy_len >= VHOST_USER_RX_COPY_THRESHOLD))
578 {
579 if (PREDICT_FALSE (vhost_user_input_copy (vui, cpu->copy,
580 copy_len, &map_hint)))
581 {
582 vlib_error_count (vm, node->node_index,
583 VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL, 1);
584 }
585 copy_len = 0;
586
587 /* give buffers back to driver */
588 CLIB_MEMORY_BARRIER ();
589 txvq->used->idx = txvq->last_used_idx;
590 vhost_user_log_dirty_ring (vui, txvq, idx);
591 }
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200592 }
Damjan Marion92825382018-11-21 10:03:44 +0100593stop:
594 vlib_put_next_frame (vm, node, next_index, n_left_to_next);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200595
596 /* Do the memory copies */
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100597 if (PREDICT_FALSE (vhost_user_input_copy (vui, cpu->copy, copy_len,
598 &map_hint)))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200599 {
600 vlib_error_count (vm, node->node_index,
601 VHOST_USER_INPUT_FUNC_ERROR_MMAP_FAIL, 1);
602 }
603
604 /* give buffers back to driver */
605 CLIB_MEMORY_BARRIER ();
606 txvq->used->idx = txvq->last_used_idx;
607 vhost_user_log_dirty_ring (vui, txvq, idx);
608
609 /* interrupt (call) handling */
610 if ((txvq->callfd_idx != ~0) &&
611 !(txvq->avail->flags & VRING_AVAIL_F_NO_INTERRUPT))
612 {
613 txvq->n_since_last_int += n_rx_packets;
614
615 if (txvq->n_since_last_int > vum->coalesce_frames)
616 vhost_user_send_call (vm, txvq);
617 }
618
619 /* increase rx counters */
620 vlib_increment_combined_counter
621 (vnet_main.interface_main.combined_sw_if_counters
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100622 + VNET_INTERFACE_COUNTER_RX, vm->thread_index, vui->sw_if_index,
623 n_rx_packets, n_rx_bytes);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200624
Damjan Marion7e0b17d2018-11-20 21:07:03 +0100625 vnet_device_increment_rx_packets (vm->thread_index, n_rx_packets);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200626
627 return n_rx_packets;
628}
629
630VLIB_NODE_FN (vhost_user_input_node) (vlib_main_t * vm,
631 vlib_node_runtime_t * node,
632 vlib_frame_t * frame)
633{
634 vhost_user_main_t *vum = &vhost_user_main;
635 uword n_rx_packets = 0;
636 vhost_user_intf_t *vui;
637 vnet_device_input_runtime_t *rt =
638 (vnet_device_input_runtime_t *) node->runtime_data;
639 vnet_device_and_queue_t *dq;
640
641 vec_foreach (dq, rt->devices_and_queues)
642 {
Sirshak Das5b718d52018-10-12 09:38:27 -0500643 if ((node->state == VLIB_NODE_STATE_POLLING) ||
644 clib_atomic_swap_acq_n (&dq->interrupt_pending, 0))
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200645 {
646 vui =
647 pool_elt_at_index (vum->vhost_user_interfaces, dq->dev_instance);
Damjan Marionbefe6912018-11-22 15:45:54 +0100648 n_rx_packets += vhost_user_if_input (vm, vum, vui, dq->queue_id, node,
649 dq->mode);
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200650 }
651 }
652
653 return n_rx_packets;
654}
655
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200656/* *INDENT-OFF* */
657VLIB_REGISTER_NODE (vhost_user_input_node) = {
658 .type = VLIB_NODE_TYPE_INPUT,
659 .name = "vhost-user-input",
660 .sibling_of = "device-input",
661
662 /* Will be enabled if/when hardware is detected. */
663 .state = VLIB_NODE_STATE_DISABLED,
664
665 .format_buffer = format_ethernet_header_with_length,
666 .format_trace = format_vhost_trace,
667
668 .n_errors = VHOST_USER_INPUT_FUNC_N_ERROR,
669 .error_strings = vhost_user_input_func_error_strings,
670};
671/* *INDENT-ON* */
Mohsin Kazmie7cde312018-06-26 17:20:11 +0200672
673/*
674 * fd.io coding-style-patch-verification: ON
675 *
676 * Local Variables:
677 * eval: (c-set-style "gnu")
678 * End:
679 */