xref: /dpdk/drivers/net/avp/avp_ethdev.c (revision 200bc52e5aa0d72e70464c9cd22b55cf536ed13c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2013-2017 Wind River Systems, Inc.
3  */
4 
5 #include <stdint.h>
6 #include <string.h>
7 #include <stdio.h>
8 #include <errno.h>
9 #include <unistd.h>
10 
11 #include <rte_ethdev_driver.h>
12 #include <rte_ethdev_pci.h>
13 #include <rte_memcpy.h>
14 #include <rte_string_fns.h>
15 #include <rte_malloc.h>
16 #include <rte_atomic.h>
17 #include <rte_branch_prediction.h>
18 #include <rte_pci.h>
19 #include <rte_bus_pci.h>
20 #include <rte_ether.h>
21 #include <rte_common.h>
22 #include <rte_cycles.h>
23 #include <rte_spinlock.h>
24 #include <rte_byteorder.h>
25 #include <rte_dev.h>
26 #include <rte_memory.h>
27 #include <rte_eal.h>
28 #include <rte_io.h>
29 
30 #include "rte_avp_common.h"
31 #include "rte_avp_fifo.h"
32 
33 #include "avp_logs.h"
34 
35 int avp_logtype_driver;
36 
37 static int avp_dev_create(struct rte_pci_device *pci_dev,
38 			  struct rte_eth_dev *eth_dev);
39 
40 static int avp_dev_configure(struct rte_eth_dev *dev);
41 static int avp_dev_start(struct rte_eth_dev *dev);
42 static void avp_dev_stop(struct rte_eth_dev *dev);
43 static void avp_dev_close(struct rte_eth_dev *dev);
44 static void avp_dev_info_get(struct rte_eth_dev *dev,
45 			     struct rte_eth_dev_info *dev_info);
46 static int avp_vlan_offload_set(struct rte_eth_dev *dev, int mask);
47 static int avp_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete);
48 static void avp_dev_promiscuous_enable(struct rte_eth_dev *dev);
49 static void avp_dev_promiscuous_disable(struct rte_eth_dev *dev);
50 
51 static int avp_dev_rx_queue_setup(struct rte_eth_dev *dev,
52 				  uint16_t rx_queue_id,
53 				  uint16_t nb_rx_desc,
54 				  unsigned int socket_id,
55 				  const struct rte_eth_rxconf *rx_conf,
56 				  struct rte_mempool *pool);
57 
58 static int avp_dev_tx_queue_setup(struct rte_eth_dev *dev,
59 				  uint16_t tx_queue_id,
60 				  uint16_t nb_tx_desc,
61 				  unsigned int socket_id,
62 				  const struct rte_eth_txconf *tx_conf);
63 
64 static uint16_t avp_recv_scattered_pkts(void *rx_queue,
65 					struct rte_mbuf **rx_pkts,
66 					uint16_t nb_pkts);
67 
68 static uint16_t avp_recv_pkts(void *rx_queue,
69 			      struct rte_mbuf **rx_pkts,
70 			      uint16_t nb_pkts);
71 
72 static uint16_t avp_xmit_scattered_pkts(void *tx_queue,
73 					struct rte_mbuf **tx_pkts,
74 					uint16_t nb_pkts);
75 
76 static uint16_t avp_xmit_pkts(void *tx_queue,
77 			      struct rte_mbuf **tx_pkts,
78 			      uint16_t nb_pkts);
79 
80 static void avp_dev_rx_queue_release(void *rxq);
81 static void avp_dev_tx_queue_release(void *txq);
82 
83 static int avp_dev_stats_get(struct rte_eth_dev *dev,
84 			      struct rte_eth_stats *stats);
85 static void avp_dev_stats_reset(struct rte_eth_dev *dev);
86 
87 
88 #define AVP_MAX_RX_BURST 64
89 #define AVP_MAX_TX_BURST 64
90 #define AVP_MAX_MAC_ADDRS 1
91 #define AVP_MIN_RX_BUFSIZE RTE_ETHER_MIN_LEN
92 
93 
94 /*
95  * Defines the number of microseconds to wait before checking the response
96  * queue for completion.
97  */
98 #define AVP_REQUEST_DELAY_USECS (5000)
99 
100 /*
101  * Defines the number times to check the response queue for completion before
102  * declaring a timeout.
103  */
104 #define AVP_MAX_REQUEST_RETRY (100)
105 
106 /* Defines the current PCI driver version number */
107 #define AVP_DPDK_DRIVER_VERSION RTE_AVP_CURRENT_GUEST_VERSION
108 
109 /*
110  * The set of PCI devices this driver supports
111  */
112 static const struct rte_pci_id pci_id_avp_map[] = {
113 	{ .vendor_id = RTE_AVP_PCI_VENDOR_ID,
114 	  .device_id = RTE_AVP_PCI_DEVICE_ID,
115 	  .subsystem_vendor_id = RTE_AVP_PCI_SUB_VENDOR_ID,
116 	  .subsystem_device_id = RTE_AVP_PCI_SUB_DEVICE_ID,
117 	  .class_id = RTE_CLASS_ANY_ID,
118 	},
119 
120 	{ .vendor_id = 0, /* sentinel */
121 	},
122 };
123 
124 /*
125  * dev_ops for avp, bare necessities for basic operation
126  */
127 static const struct eth_dev_ops avp_eth_dev_ops = {
128 	.dev_configure       = avp_dev_configure,
129 	.dev_start           = avp_dev_start,
130 	.dev_stop            = avp_dev_stop,
131 	.dev_close           = avp_dev_close,
132 	.dev_infos_get       = avp_dev_info_get,
133 	.vlan_offload_set    = avp_vlan_offload_set,
134 	.stats_get           = avp_dev_stats_get,
135 	.stats_reset         = avp_dev_stats_reset,
136 	.link_update         = avp_dev_link_update,
137 	.promiscuous_enable  = avp_dev_promiscuous_enable,
138 	.promiscuous_disable = avp_dev_promiscuous_disable,
139 	.rx_queue_setup      = avp_dev_rx_queue_setup,
140 	.rx_queue_release    = avp_dev_rx_queue_release,
141 	.tx_queue_setup      = avp_dev_tx_queue_setup,
142 	.tx_queue_release    = avp_dev_tx_queue_release,
143 };
144 
145 /**@{ AVP device flags */
146 #define AVP_F_PROMISC (1 << 1)
147 #define AVP_F_CONFIGURED (1 << 2)
148 #define AVP_F_LINKUP (1 << 3)
149 #define AVP_F_DETACHED (1 << 4)
150 /**@} */
151 
152 /* Ethernet device validation marker */
153 #define AVP_ETHDEV_MAGIC 0x92972862
154 
155 /*
156  * Defines the AVP device attributes which are attached to an RTE ethernet
157  * device
158  */
159 struct avp_dev {
160 	uint32_t magic; /**< Memory validation marker */
161 	uint64_t device_id; /**< Unique system identifier */
162 	struct rte_ether_addr ethaddr; /**< Host specified MAC address */
163 	struct rte_eth_dev_data *dev_data;
164 	/**< Back pointer to ethernet device data */
165 	volatile uint32_t flags; /**< Device operational flags */
166 	uint16_t port_id; /**< Ethernet port identifier */
167 	struct rte_mempool *pool; /**< pkt mbuf mempool */
168 	unsigned int guest_mbuf_size; /**< local pool mbuf size */
169 	unsigned int host_mbuf_size; /**< host mbuf size */
170 	unsigned int max_rx_pkt_len; /**< maximum receive unit */
171 	uint32_t host_features; /**< Supported feature bitmap */
172 	uint32_t features; /**< Enabled feature bitmap */
173 	unsigned int num_tx_queues; /**< Negotiated number of transmit queues */
174 	unsigned int max_tx_queues; /**< Maximum number of transmit queues */
175 	unsigned int num_rx_queues; /**< Negotiated number of receive queues */
176 	unsigned int max_rx_queues; /**< Maximum number of receive queues */
177 
178 	struct rte_avp_fifo *tx_q[RTE_AVP_MAX_QUEUES]; /**< TX queue */
179 	struct rte_avp_fifo *rx_q[RTE_AVP_MAX_QUEUES]; /**< RX queue */
180 	struct rte_avp_fifo *alloc_q[RTE_AVP_MAX_QUEUES];
181 	/**< Allocated mbufs queue */
182 	struct rte_avp_fifo *free_q[RTE_AVP_MAX_QUEUES];
183 	/**< To be freed mbufs queue */
184 
185 	/* mutual exclusion over the 'flag' and 'resp_q/req_q' fields */
186 	rte_spinlock_t lock;
187 
188 	/* For request & response */
189 	struct rte_avp_fifo *req_q; /**< Request queue */
190 	struct rte_avp_fifo *resp_q; /**< Response queue */
191 	void *host_sync_addr; /**< (host) Req/Resp Mem address */
192 	void *sync_addr; /**< Req/Resp Mem address */
193 	void *host_mbuf_addr; /**< (host) MBUF pool start address */
194 	void *mbuf_addr; /**< MBUF pool start address */
195 } __rte_cache_aligned;
196 
197 /* RTE ethernet private data */
198 struct avp_adapter {
199 	struct avp_dev avp;
200 } __rte_cache_aligned;
201 
202 
203 /* 32-bit MMIO register write */
204 #define AVP_WRITE32(_value, _addr) rte_write32_relaxed((_value), (_addr))
205 
206 /* 32-bit MMIO register read */
207 #define AVP_READ32(_addr) rte_read32_relaxed((_addr))
208 
209 /* Macro to cast the ethernet device private data to a AVP object */
210 #define AVP_DEV_PRIVATE_TO_HW(adapter) \
211 	(&((struct avp_adapter *)adapter)->avp)
212 
213 /*
214  * Defines the structure of a AVP device queue for the purpose of handling the
215  * receive and transmit burst callback functions
216  */
217 struct avp_queue {
218 	struct rte_eth_dev_data *dev_data;
219 	/**< Backpointer to ethernet device data */
220 	struct avp_dev *avp; /**< Backpointer to AVP device */
221 	uint16_t queue_id;
222 	/**< Queue identifier used for indexing current queue */
223 	uint16_t queue_base;
224 	/**< Base queue identifier for queue servicing */
225 	uint16_t queue_limit;
226 	/**< Maximum queue identifier for queue servicing */
227 
228 	uint64_t packets;
229 	uint64_t bytes;
230 	uint64_t errors;
231 };
232 
233 /* send a request and wait for a response
234  *
235  * @warning must be called while holding the avp->lock spinlock.
236  */
237 static int
238 avp_dev_process_request(struct avp_dev *avp, struct rte_avp_request *request)
239 {
240 	unsigned int retry = AVP_MAX_REQUEST_RETRY;
241 	void *resp_addr = NULL;
242 	unsigned int count;
243 	int ret;
244 
245 	PMD_DRV_LOG(DEBUG, "Sending request %u to host\n", request->req_id);
246 
247 	request->result = -ENOTSUP;
248 
249 	/* Discard any stale responses before starting a new request */
250 	while (avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1))
251 		PMD_DRV_LOG(DEBUG, "Discarding stale response\n");
252 
253 	rte_memcpy(avp->sync_addr, request, sizeof(*request));
254 	count = avp_fifo_put(avp->req_q, &avp->host_sync_addr, 1);
255 	if (count < 1) {
256 		PMD_DRV_LOG(ERR, "Cannot send request %u to host\n",
257 			    request->req_id);
258 		ret = -EBUSY;
259 		goto done;
260 	}
261 
262 	while (retry--) {
263 		/* wait for a response */
264 		usleep(AVP_REQUEST_DELAY_USECS);
265 
266 		count = avp_fifo_count(avp->resp_q);
267 		if (count >= 1) {
268 			/* response received */
269 			break;
270 		}
271 
272 		if ((count < 1) && (retry == 0)) {
273 			PMD_DRV_LOG(ERR, "Timeout while waiting for a response for %u\n",
274 				    request->req_id);
275 			ret = -ETIME;
276 			goto done;
277 		}
278 	}
279 
280 	/* retrieve the response */
281 	count = avp_fifo_get(avp->resp_q, (void **)&resp_addr, 1);
282 	if ((count != 1) || (resp_addr != avp->host_sync_addr)) {
283 		PMD_DRV_LOG(ERR, "Invalid response from host, count=%u resp=%p host_sync_addr=%p\n",
284 			    count, resp_addr, avp->host_sync_addr);
285 		ret = -ENODATA;
286 		goto done;
287 	}
288 
289 	/* copy to user buffer */
290 	rte_memcpy(request, avp->sync_addr, sizeof(*request));
291 	ret = 0;
292 
293 	PMD_DRV_LOG(DEBUG, "Result %d received for request %u\n",
294 		    request->result, request->req_id);
295 
296 done:
297 	return ret;
298 }
299 
300 static int
301 avp_dev_ctrl_set_link_state(struct rte_eth_dev *eth_dev, unsigned int state)
302 {
303 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
304 	struct rte_avp_request request;
305 	int ret;
306 
307 	/* setup a link state change request */
308 	memset(&request, 0, sizeof(request));
309 	request.req_id = RTE_AVP_REQ_CFG_NETWORK_IF;
310 	request.if_up = state;
311 
312 	ret = avp_dev_process_request(avp, &request);
313 
314 	return ret == 0 ? request.result : ret;
315 }
316 
317 static int
318 avp_dev_ctrl_set_config(struct rte_eth_dev *eth_dev,
319 			struct rte_avp_device_config *config)
320 {
321 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
322 	struct rte_avp_request request;
323 	int ret;
324 
325 	/* setup a configure request */
326 	memset(&request, 0, sizeof(request));
327 	request.req_id = RTE_AVP_REQ_CFG_DEVICE;
328 	memcpy(&request.config, config, sizeof(request.config));
329 
330 	ret = avp_dev_process_request(avp, &request);
331 
332 	return ret == 0 ? request.result : ret;
333 }
334 
335 static int
336 avp_dev_ctrl_shutdown(struct rte_eth_dev *eth_dev)
337 {
338 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
339 	struct rte_avp_request request;
340 	int ret;
341 
342 	/* setup a shutdown request */
343 	memset(&request, 0, sizeof(request));
344 	request.req_id = RTE_AVP_REQ_SHUTDOWN_DEVICE;
345 
346 	ret = avp_dev_process_request(avp, &request);
347 
348 	return ret == 0 ? request.result : ret;
349 }
350 
351 /* translate from host mbuf virtual address to guest virtual address */
352 static inline void *
353 avp_dev_translate_buffer(struct avp_dev *avp, void *host_mbuf_address)
354 {
355 	return RTE_PTR_ADD(RTE_PTR_SUB(host_mbuf_address,
356 				       (uintptr_t)avp->host_mbuf_addr),
357 			   (uintptr_t)avp->mbuf_addr);
358 }
359 
360 /* translate from host physical address to guest virtual address */
361 static void *
362 avp_dev_translate_address(struct rte_eth_dev *eth_dev,
363 			  rte_iova_t host_phys_addr)
364 {
365 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
366 	struct rte_mem_resource *resource;
367 	struct rte_avp_memmap_info *info;
368 	struct rte_avp_memmap *map;
369 	off_t offset;
370 	void *addr;
371 	unsigned int i;
372 
373 	addr = pci_dev->mem_resource[RTE_AVP_PCI_MEMORY_BAR].addr;
374 	resource = &pci_dev->mem_resource[RTE_AVP_PCI_MEMMAP_BAR];
375 	info = (struct rte_avp_memmap_info *)resource->addr;
376 
377 	offset = 0;
378 	for (i = 0; i < info->nb_maps; i++) {
379 		/* search all segments looking for a matching address */
380 		map = &info->maps[i];
381 
382 		if ((host_phys_addr >= map->phys_addr) &&
383 			(host_phys_addr < (map->phys_addr + map->length))) {
384 			/* address is within this segment */
385 			offset += (host_phys_addr - map->phys_addr);
386 			addr = RTE_PTR_ADD(addr, (uintptr_t)offset);
387 
388 			PMD_DRV_LOG(DEBUG, "Translating host physical 0x%" PRIx64 " to guest virtual 0x%p\n",
389 				    host_phys_addr, addr);
390 
391 			return addr;
392 		}
393 		offset += map->length;
394 	}
395 
396 	return NULL;
397 }
398 
399 /* verify that the incoming device version is compatible with our version */
400 static int
401 avp_dev_version_check(uint32_t version)
402 {
403 	uint32_t driver = RTE_AVP_STRIP_MINOR_VERSION(AVP_DPDK_DRIVER_VERSION);
404 	uint32_t device = RTE_AVP_STRIP_MINOR_VERSION(version);
405 
406 	if (device <= driver) {
407 		/* the host driver version is less than or equal to ours */
408 		return 0;
409 	}
410 
411 	return 1;
412 }
413 
414 /* verify that memory regions have expected version and validation markers */
415 static int
416 avp_dev_check_regions(struct rte_eth_dev *eth_dev)
417 {
418 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
419 	struct rte_avp_memmap_info *memmap;
420 	struct rte_avp_device_info *info;
421 	struct rte_mem_resource *resource;
422 	unsigned int i;
423 
424 	/* Dump resource info for debug */
425 	for (i = 0; i < PCI_MAX_RESOURCE; i++) {
426 		resource = &pci_dev->mem_resource[i];
427 		if ((resource->phys_addr == 0) || (resource->len == 0))
428 			continue;
429 
430 		PMD_DRV_LOG(DEBUG, "resource[%u]: phys=0x%" PRIx64 " len=%" PRIu64 " addr=%p\n",
431 			    i, resource->phys_addr,
432 			    resource->len, resource->addr);
433 
434 		switch (i) {
435 		case RTE_AVP_PCI_MEMMAP_BAR:
436 			memmap = (struct rte_avp_memmap_info *)resource->addr;
437 			if ((memmap->magic != RTE_AVP_MEMMAP_MAGIC) ||
438 			    (memmap->version != RTE_AVP_MEMMAP_VERSION)) {
439 				PMD_DRV_LOG(ERR, "Invalid memmap magic 0x%08x and version %u\n",
440 					    memmap->magic, memmap->version);
441 				return -EINVAL;
442 			}
443 			break;
444 
445 		case RTE_AVP_PCI_DEVICE_BAR:
446 			info = (struct rte_avp_device_info *)resource->addr;
447 			if ((info->magic != RTE_AVP_DEVICE_MAGIC) ||
448 			    avp_dev_version_check(info->version)) {
449 				PMD_DRV_LOG(ERR, "Invalid device info magic 0x%08x or version 0x%08x > 0x%08x\n",
450 					    info->magic, info->version,
451 					    AVP_DPDK_DRIVER_VERSION);
452 				return -EINVAL;
453 			}
454 			break;
455 
456 		case RTE_AVP_PCI_MEMORY_BAR:
457 		case RTE_AVP_PCI_MMIO_BAR:
458 			if (resource->addr == NULL) {
459 				PMD_DRV_LOG(ERR, "Missing address space for BAR%u\n",
460 					    i);
461 				return -EINVAL;
462 			}
463 			break;
464 
465 		case RTE_AVP_PCI_MSIX_BAR:
466 		default:
467 			/* no validation required */
468 			break;
469 		}
470 	}
471 
472 	return 0;
473 }
474 
475 static int
476 avp_dev_detach(struct rte_eth_dev *eth_dev)
477 {
478 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
479 	int ret;
480 
481 	PMD_DRV_LOG(NOTICE, "Detaching port %u from AVP device 0x%" PRIx64 "\n",
482 		    eth_dev->data->port_id, avp->device_id);
483 
484 	rte_spinlock_lock(&avp->lock);
485 
486 	if (avp->flags & AVP_F_DETACHED) {
487 		PMD_DRV_LOG(NOTICE, "port %u already detached\n",
488 			    eth_dev->data->port_id);
489 		ret = 0;
490 		goto unlock;
491 	}
492 
493 	/* shutdown the device first so the host stops sending us packets. */
494 	ret = avp_dev_ctrl_shutdown(eth_dev);
495 	if (ret < 0) {
496 		PMD_DRV_LOG(ERR, "Failed to send/recv shutdown to host, ret=%d\n",
497 			    ret);
498 		avp->flags &= ~AVP_F_DETACHED;
499 		goto unlock;
500 	}
501 
502 	avp->flags |= AVP_F_DETACHED;
503 	rte_wmb();
504 
505 	/* wait for queues to acknowledge the presence of the detach flag */
506 	rte_delay_ms(1);
507 
508 	ret = 0;
509 
510 unlock:
511 	rte_spinlock_unlock(&avp->lock);
512 	return ret;
513 }
514 
515 static void
516 _avp_set_rx_queue_mappings(struct rte_eth_dev *eth_dev, uint16_t rx_queue_id)
517 {
518 	struct avp_dev *avp =
519 		AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
520 	struct avp_queue *rxq;
521 	uint16_t queue_count;
522 	uint16_t remainder;
523 
524 	rxq = (struct avp_queue *)eth_dev->data->rx_queues[rx_queue_id];
525 
526 	/*
527 	 * Must map all AVP fifos as evenly as possible between the configured
528 	 * device queues.  Each device queue will service a subset of the AVP
529 	 * fifos. If there is an odd number of device queues the first set of
530 	 * device queues will get the extra AVP fifos.
531 	 */
532 	queue_count = avp->num_rx_queues / eth_dev->data->nb_rx_queues;
533 	remainder = avp->num_rx_queues % eth_dev->data->nb_rx_queues;
534 	if (rx_queue_id < remainder) {
535 		/* these queues must service one extra FIFO */
536 		rxq->queue_base = rx_queue_id * (queue_count + 1);
537 		rxq->queue_limit = rxq->queue_base + (queue_count + 1) - 1;
538 	} else {
539 		/* these queues service the regular number of FIFO */
540 		rxq->queue_base = ((remainder * (queue_count + 1)) +
541 				   ((rx_queue_id - remainder) * queue_count));
542 		rxq->queue_limit = rxq->queue_base + queue_count - 1;
543 	}
544 
545 	PMD_DRV_LOG(DEBUG, "rxq %u at %p base %u limit %u\n",
546 		    rx_queue_id, rxq, rxq->queue_base, rxq->queue_limit);
547 
548 	rxq->queue_id = rxq->queue_base;
549 }
550 
551 static void
552 _avp_set_queue_counts(struct rte_eth_dev *eth_dev)
553 {
554 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
555 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
556 	struct rte_avp_device_info *host_info;
557 	void *addr;
558 
559 	addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
560 	host_info = (struct rte_avp_device_info *)addr;
561 
562 	/*
563 	 * the transmit direction is not negotiated beyond respecting the max
564 	 * number of queues because the host can handle arbitrary guest tx
565 	 * queues (host rx queues).
566 	 */
567 	avp->num_tx_queues = eth_dev->data->nb_tx_queues;
568 
569 	/*
570 	 * the receive direction is more restrictive.  The host requires a
571 	 * minimum number of guest rx queues (host tx queues) therefore
572 	 * negotiate a value that is at least as large as the host minimum
573 	 * requirement.  If the host and guest values are not identical then a
574 	 * mapping will be established in the receive_queue_setup function.
575 	 */
576 	avp->num_rx_queues = RTE_MAX(host_info->min_rx_queues,
577 				     eth_dev->data->nb_rx_queues);
578 
579 	PMD_DRV_LOG(DEBUG, "Requesting %u Tx and %u Rx queues from host\n",
580 		    avp->num_tx_queues, avp->num_rx_queues);
581 }
582 
583 static int
584 avp_dev_attach(struct rte_eth_dev *eth_dev)
585 {
586 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
587 	struct rte_avp_device_config config;
588 	unsigned int i;
589 	int ret;
590 
591 	PMD_DRV_LOG(NOTICE, "Attaching port %u to AVP device 0x%" PRIx64 "\n",
592 		    eth_dev->data->port_id, avp->device_id);
593 
594 	rte_spinlock_lock(&avp->lock);
595 
596 	if (!(avp->flags & AVP_F_DETACHED)) {
597 		PMD_DRV_LOG(NOTICE, "port %u already attached\n",
598 			    eth_dev->data->port_id);
599 		ret = 0;
600 		goto unlock;
601 	}
602 
603 	/*
604 	 * make sure that the detached flag is set prior to reconfiguring the
605 	 * queues.
606 	 */
607 	avp->flags |= AVP_F_DETACHED;
608 	rte_wmb();
609 
610 	/*
611 	 * re-run the device create utility which will parse the new host info
612 	 * and setup the AVP device queue pointers.
613 	 */
614 	ret = avp_dev_create(RTE_ETH_DEV_TO_PCI(eth_dev), eth_dev);
615 	if (ret < 0) {
616 		PMD_DRV_LOG(ERR, "Failed to re-create AVP device, ret=%d\n",
617 			    ret);
618 		goto unlock;
619 	}
620 
621 	if (avp->flags & AVP_F_CONFIGURED) {
622 		/*
623 		 * Update the receive queue mapping to handle cases where the
624 		 * source and destination hosts have different queue
625 		 * requirements.  As long as the DETACHED flag is asserted the
626 		 * queue table should not be referenced so it should be safe to
627 		 * update it.
628 		 */
629 		_avp_set_queue_counts(eth_dev);
630 		for (i = 0; i < eth_dev->data->nb_rx_queues; i++)
631 			_avp_set_rx_queue_mappings(eth_dev, i);
632 
633 		/*
634 		 * Update the host with our config details so that it knows the
635 		 * device is active.
636 		 */
637 		memset(&config, 0, sizeof(config));
638 		config.device_id = avp->device_id;
639 		config.driver_type = RTE_AVP_DRIVER_TYPE_DPDK;
640 		config.driver_version = AVP_DPDK_DRIVER_VERSION;
641 		config.features = avp->features;
642 		config.num_tx_queues = avp->num_tx_queues;
643 		config.num_rx_queues = avp->num_rx_queues;
644 		config.if_up = !!(avp->flags & AVP_F_LINKUP);
645 
646 		ret = avp_dev_ctrl_set_config(eth_dev, &config);
647 		if (ret < 0) {
648 			PMD_DRV_LOG(ERR, "Config request failed by host, ret=%d\n",
649 				    ret);
650 			goto unlock;
651 		}
652 	}
653 
654 	rte_wmb();
655 	avp->flags &= ~AVP_F_DETACHED;
656 
657 	ret = 0;
658 
659 unlock:
660 	rte_spinlock_unlock(&avp->lock);
661 	return ret;
662 }
663 
664 static void
665 avp_dev_interrupt_handler(void *data)
666 {
667 	struct rte_eth_dev *eth_dev = data;
668 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
669 	void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
670 	uint32_t status, value;
671 	int ret;
672 
673 	if (registers == NULL)
674 		rte_panic("no mapped MMIO register space\n");
675 
676 	/* read the interrupt status register
677 	 * note: this register clears on read so all raised interrupts must be
678 	 *    handled or remembered for later processing
679 	 */
680 	status = AVP_READ32(
681 		RTE_PTR_ADD(registers,
682 			    RTE_AVP_INTERRUPT_STATUS_OFFSET));
683 
684 	if (status & RTE_AVP_MIGRATION_INTERRUPT_MASK) {
685 		/* handle interrupt based on current status */
686 		value = AVP_READ32(
687 			RTE_PTR_ADD(registers,
688 				    RTE_AVP_MIGRATION_STATUS_OFFSET));
689 		switch (value) {
690 		case RTE_AVP_MIGRATION_DETACHED:
691 			ret = avp_dev_detach(eth_dev);
692 			break;
693 		case RTE_AVP_MIGRATION_ATTACHED:
694 			ret = avp_dev_attach(eth_dev);
695 			break;
696 		default:
697 			PMD_DRV_LOG(ERR, "unexpected migration status, status=%u\n",
698 				    value);
699 			ret = -EINVAL;
700 		}
701 
702 		/* acknowledge the request by writing out our current status */
703 		value = (ret == 0 ? value : RTE_AVP_MIGRATION_ERROR);
704 		AVP_WRITE32(value,
705 			    RTE_PTR_ADD(registers,
706 					RTE_AVP_MIGRATION_ACK_OFFSET));
707 
708 		PMD_DRV_LOG(NOTICE, "AVP migration interrupt handled\n");
709 	}
710 
711 	if (status & ~RTE_AVP_MIGRATION_INTERRUPT_MASK)
712 		PMD_DRV_LOG(WARNING, "AVP unexpected interrupt, status=0x%08x\n",
713 			    status);
714 
715 	/* re-enable UIO interrupt handling */
716 	ret = rte_intr_enable(&pci_dev->intr_handle);
717 	if (ret < 0) {
718 		PMD_DRV_LOG(ERR, "Failed to re-enable UIO interrupts, ret=%d\n",
719 			    ret);
720 		/* continue */
721 	}
722 }
723 
724 static int
725 avp_dev_enable_interrupts(struct rte_eth_dev *eth_dev)
726 {
727 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
728 	void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
729 	int ret;
730 
731 	if (registers == NULL)
732 		return -EINVAL;
733 
734 	/* enable UIO interrupt handling */
735 	ret = rte_intr_enable(&pci_dev->intr_handle);
736 	if (ret < 0) {
737 		PMD_DRV_LOG(ERR, "Failed to enable UIO interrupts, ret=%d\n",
738 			    ret);
739 		return ret;
740 	}
741 
742 	/* inform the device that all interrupts are enabled */
743 	AVP_WRITE32(RTE_AVP_APP_INTERRUPTS_MASK,
744 		    RTE_PTR_ADD(registers, RTE_AVP_INTERRUPT_MASK_OFFSET));
745 
746 	return 0;
747 }
748 
749 static int
750 avp_dev_disable_interrupts(struct rte_eth_dev *eth_dev)
751 {
752 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
753 	void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
754 	int ret;
755 
756 	if (registers == NULL)
757 		return 0;
758 
759 	/* inform the device that all interrupts are disabled */
760 	AVP_WRITE32(RTE_AVP_NO_INTERRUPTS_MASK,
761 		    RTE_PTR_ADD(registers, RTE_AVP_INTERRUPT_MASK_OFFSET));
762 
763 	/* enable UIO interrupt handling */
764 	ret = rte_intr_disable(&pci_dev->intr_handle);
765 	if (ret < 0) {
766 		PMD_DRV_LOG(ERR, "Failed to disable UIO interrupts, ret=%d\n",
767 			    ret);
768 		return ret;
769 	}
770 
771 	return 0;
772 }
773 
774 static int
775 avp_dev_setup_interrupts(struct rte_eth_dev *eth_dev)
776 {
777 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
778 	int ret;
779 
780 	/* register a callback handler with UIO for interrupt notifications */
781 	ret = rte_intr_callback_register(&pci_dev->intr_handle,
782 					 avp_dev_interrupt_handler,
783 					 (void *)eth_dev);
784 	if (ret < 0) {
785 		PMD_DRV_LOG(ERR, "Failed to register UIO interrupt callback, ret=%d\n",
786 			    ret);
787 		return ret;
788 	}
789 
790 	/* enable interrupt processing */
791 	return avp_dev_enable_interrupts(eth_dev);
792 }
793 
794 static int
795 avp_dev_migration_pending(struct rte_eth_dev *eth_dev)
796 {
797 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
798 	void *registers = pci_dev->mem_resource[RTE_AVP_PCI_MMIO_BAR].addr;
799 	uint32_t value;
800 
801 	if (registers == NULL)
802 		return 0;
803 
804 	value = AVP_READ32(RTE_PTR_ADD(registers,
805 				       RTE_AVP_MIGRATION_STATUS_OFFSET));
806 	if (value == RTE_AVP_MIGRATION_DETACHED) {
807 		/* migration is in progress; ack it if we have not already */
808 		AVP_WRITE32(value,
809 			    RTE_PTR_ADD(registers,
810 					RTE_AVP_MIGRATION_ACK_OFFSET));
811 		return 1;
812 	}
813 	return 0;
814 }
815 
816 /*
817  * create a AVP device using the supplied device info by first translating it
818  * to guest address space(s).
819  */
820 static int
821 avp_dev_create(struct rte_pci_device *pci_dev,
822 	       struct rte_eth_dev *eth_dev)
823 {
824 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
825 	struct rte_avp_device_info *host_info;
826 	struct rte_mem_resource *resource;
827 	unsigned int i;
828 
829 	resource = &pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR];
830 	if (resource->addr == NULL) {
831 		PMD_DRV_LOG(ERR, "BAR%u is not mapped\n",
832 			    RTE_AVP_PCI_DEVICE_BAR);
833 		return -EFAULT;
834 	}
835 	host_info = (struct rte_avp_device_info *)resource->addr;
836 
837 	if ((host_info->magic != RTE_AVP_DEVICE_MAGIC) ||
838 		avp_dev_version_check(host_info->version)) {
839 		PMD_DRV_LOG(ERR, "Invalid AVP PCI device, magic 0x%08x version 0x%08x > 0x%08x\n",
840 			    host_info->magic, host_info->version,
841 			    AVP_DPDK_DRIVER_VERSION);
842 		return -EINVAL;
843 	}
844 
845 	PMD_DRV_LOG(DEBUG, "AVP host device is v%u.%u.%u\n",
846 		    RTE_AVP_GET_RELEASE_VERSION(host_info->version),
847 		    RTE_AVP_GET_MAJOR_VERSION(host_info->version),
848 		    RTE_AVP_GET_MINOR_VERSION(host_info->version));
849 
850 	PMD_DRV_LOG(DEBUG, "AVP host supports %u to %u TX queue(s)\n",
851 		    host_info->min_tx_queues, host_info->max_tx_queues);
852 	PMD_DRV_LOG(DEBUG, "AVP host supports %u to %u RX queue(s)\n",
853 		    host_info->min_rx_queues, host_info->max_rx_queues);
854 	PMD_DRV_LOG(DEBUG, "AVP host supports features 0x%08x\n",
855 		    host_info->features);
856 
857 	if (avp->magic != AVP_ETHDEV_MAGIC) {
858 		/*
859 		 * First time initialization (i.e., not during a VM
860 		 * migration)
861 		 */
862 		memset(avp, 0, sizeof(*avp));
863 		avp->magic = AVP_ETHDEV_MAGIC;
864 		avp->dev_data = eth_dev->data;
865 		avp->port_id = eth_dev->data->port_id;
866 		avp->host_mbuf_size = host_info->mbuf_size;
867 		avp->host_features = host_info->features;
868 		rte_spinlock_init(&avp->lock);
869 		memcpy(&avp->ethaddr.addr_bytes[0],
870 		       host_info->ethaddr, RTE_ETHER_ADDR_LEN);
871 		/* adjust max values to not exceed our max */
872 		avp->max_tx_queues =
873 			RTE_MIN(host_info->max_tx_queues, RTE_AVP_MAX_QUEUES);
874 		avp->max_rx_queues =
875 			RTE_MIN(host_info->max_rx_queues, RTE_AVP_MAX_QUEUES);
876 	} else {
877 		/* Re-attaching during migration */
878 
879 		/* TODO... requires validation of host values */
880 		if ((host_info->features & avp->features) != avp->features) {
881 			PMD_DRV_LOG(ERR, "AVP host features mismatched; 0x%08x, host=0x%08x\n",
882 				    avp->features, host_info->features);
883 			/* this should not be possible; continue for now */
884 		}
885 	}
886 
887 	/* the device id is allowed to change over migrations */
888 	avp->device_id = host_info->device_id;
889 
890 	/* translate incoming host addresses to guest address space */
891 	PMD_DRV_LOG(DEBUG, "AVP first host tx queue at 0x%" PRIx64 "\n",
892 		    host_info->tx_phys);
893 	PMD_DRV_LOG(DEBUG, "AVP first host alloc queue at 0x%" PRIx64 "\n",
894 		    host_info->alloc_phys);
895 	for (i = 0; i < avp->max_tx_queues; i++) {
896 		avp->tx_q[i] = avp_dev_translate_address(eth_dev,
897 			host_info->tx_phys + (i * host_info->tx_size));
898 
899 		avp->alloc_q[i] = avp_dev_translate_address(eth_dev,
900 			host_info->alloc_phys + (i * host_info->alloc_size));
901 	}
902 
903 	PMD_DRV_LOG(DEBUG, "AVP first host rx queue at 0x%" PRIx64 "\n",
904 		    host_info->rx_phys);
905 	PMD_DRV_LOG(DEBUG, "AVP first host free queue at 0x%" PRIx64 "\n",
906 		    host_info->free_phys);
907 	for (i = 0; i < avp->max_rx_queues; i++) {
908 		avp->rx_q[i] = avp_dev_translate_address(eth_dev,
909 			host_info->rx_phys + (i * host_info->rx_size));
910 		avp->free_q[i] = avp_dev_translate_address(eth_dev,
911 			host_info->free_phys + (i * host_info->free_size));
912 	}
913 
914 	PMD_DRV_LOG(DEBUG, "AVP host request queue at 0x%" PRIx64 "\n",
915 		    host_info->req_phys);
916 	PMD_DRV_LOG(DEBUG, "AVP host response queue at 0x%" PRIx64 "\n",
917 		    host_info->resp_phys);
918 	PMD_DRV_LOG(DEBUG, "AVP host sync address at 0x%" PRIx64 "\n",
919 		    host_info->sync_phys);
920 	PMD_DRV_LOG(DEBUG, "AVP host mbuf address at 0x%" PRIx64 "\n",
921 		    host_info->mbuf_phys);
922 	avp->req_q = avp_dev_translate_address(eth_dev, host_info->req_phys);
923 	avp->resp_q = avp_dev_translate_address(eth_dev, host_info->resp_phys);
924 	avp->sync_addr =
925 		avp_dev_translate_address(eth_dev, host_info->sync_phys);
926 	avp->mbuf_addr =
927 		avp_dev_translate_address(eth_dev, host_info->mbuf_phys);
928 
929 	/*
930 	 * store the host mbuf virtual address so that we can calculate
931 	 * relative offsets for each mbuf as they are processed
932 	 */
933 	avp->host_mbuf_addr = host_info->mbuf_va;
934 	avp->host_sync_addr = host_info->sync_va;
935 
936 	/*
937 	 * store the maximum packet length that is supported by the host.
938 	 */
939 	avp->max_rx_pkt_len = host_info->max_rx_pkt_len;
940 	PMD_DRV_LOG(DEBUG, "AVP host max receive packet length is %u\n",
941 				host_info->max_rx_pkt_len);
942 
943 	return 0;
944 }
945 
946 /*
947  * This function is based on probe() function in avp_pci.c
948  * It returns 0 on success.
949  */
950 static int
951 eth_avp_dev_init(struct rte_eth_dev *eth_dev)
952 {
953 	struct avp_dev *avp =
954 		AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
955 	struct rte_pci_device *pci_dev;
956 	int ret;
957 
958 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
959 	eth_dev->dev_ops = &avp_eth_dev_ops;
960 	eth_dev->rx_pkt_burst = &avp_recv_pkts;
961 	eth_dev->tx_pkt_burst = &avp_xmit_pkts;
962 
963 	if (rte_eal_process_type() != RTE_PROC_PRIMARY) {
964 		/*
965 		 * no setup required on secondary processes.  All data is saved
966 		 * in dev_private by the primary process. All resource should
967 		 * be mapped to the same virtual address so all pointers should
968 		 * be valid.
969 		 */
970 		if (eth_dev->data->scattered_rx) {
971 			PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
972 			eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
973 			eth_dev->tx_pkt_burst = avp_xmit_scattered_pkts;
974 		}
975 		return 0;
976 	}
977 
978 	rte_eth_copy_pci_info(eth_dev, pci_dev);
979 
980 	/* Check current migration status */
981 	if (avp_dev_migration_pending(eth_dev)) {
982 		PMD_DRV_LOG(ERR, "VM live migration operation in progress\n");
983 		return -EBUSY;
984 	}
985 
986 	/* Check BAR resources */
987 	ret = avp_dev_check_regions(eth_dev);
988 	if (ret < 0) {
989 		PMD_DRV_LOG(ERR, "Failed to validate BAR resources, ret=%d\n",
990 			    ret);
991 		return ret;
992 	}
993 
994 	/* Enable interrupts */
995 	ret = avp_dev_setup_interrupts(eth_dev);
996 	if (ret < 0) {
997 		PMD_DRV_LOG(ERR, "Failed to enable interrupts, ret=%d\n", ret);
998 		return ret;
999 	}
1000 
1001 	/* Handle each subtype */
1002 	ret = avp_dev_create(pci_dev, eth_dev);
1003 	if (ret < 0) {
1004 		PMD_DRV_LOG(ERR, "Failed to create device, ret=%d\n", ret);
1005 		return ret;
1006 	}
1007 
1008 	/* Allocate memory for storing MAC addresses */
1009 	eth_dev->data->mac_addrs = rte_zmalloc("avp_ethdev",
1010 					RTE_ETHER_ADDR_LEN, 0);
1011 	if (eth_dev->data->mac_addrs == NULL) {
1012 		PMD_DRV_LOG(ERR, "Failed to allocate %d bytes needed to store MAC addresses\n",
1013 			    RTE_ETHER_ADDR_LEN);
1014 		return -ENOMEM;
1015 	}
1016 
1017 	/* Get a mac from device config */
1018 	rte_ether_addr_copy(&avp->ethaddr, &eth_dev->data->mac_addrs[0]);
1019 
1020 	return 0;
1021 }
1022 
1023 static int
1024 eth_avp_dev_uninit(struct rte_eth_dev *eth_dev)
1025 {
1026 	int ret;
1027 
1028 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
1029 		return -EPERM;
1030 
1031 	if (eth_dev->data == NULL)
1032 		return 0;
1033 
1034 	ret = avp_dev_disable_interrupts(eth_dev);
1035 	if (ret != 0) {
1036 		PMD_DRV_LOG(ERR, "Failed to disable interrupts, ret=%d\n", ret);
1037 		return ret;
1038 	}
1039 
1040 	return 0;
1041 }
1042 
1043 static int
1044 eth_avp_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
1045 		  struct rte_pci_device *pci_dev)
1046 {
1047 	return rte_eth_dev_pci_generic_probe(pci_dev, sizeof(struct avp_adapter),
1048 			eth_avp_dev_init);
1049 }
1050 
1051 static int
1052 eth_avp_pci_remove(struct rte_pci_device *pci_dev)
1053 {
1054 	return rte_eth_dev_pci_generic_remove(pci_dev,
1055 					      eth_avp_dev_uninit);
1056 }
1057 
1058 static struct rte_pci_driver rte_avp_pmd = {
1059 	.id_table = pci_id_avp_map,
1060 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING,
1061 	.probe = eth_avp_pci_probe,
1062 	.remove = eth_avp_pci_remove,
1063 };
1064 
1065 static int
1066 avp_dev_enable_scattered(struct rte_eth_dev *eth_dev,
1067 			 struct avp_dev *avp)
1068 {
1069 	unsigned int max_rx_pkt_len;
1070 
1071 	max_rx_pkt_len = eth_dev->data->dev_conf.rxmode.max_rx_pkt_len;
1072 
1073 	if ((max_rx_pkt_len > avp->guest_mbuf_size) ||
1074 	    (max_rx_pkt_len > avp->host_mbuf_size)) {
1075 		/*
1076 		 * If the guest MTU is greater than either the host or guest
1077 		 * buffers then chained mbufs have to be enabled in the TX
1078 		 * direction.  It is assumed that the application will not need
1079 		 * to send packets larger than their max_rx_pkt_len (MRU).
1080 		 */
1081 		return 1;
1082 	}
1083 
1084 	if ((avp->max_rx_pkt_len > avp->guest_mbuf_size) ||
1085 	    (avp->max_rx_pkt_len > avp->host_mbuf_size)) {
1086 		/*
1087 		 * If the host MRU is greater than its own mbuf size or the
1088 		 * guest mbuf size then chained mbufs have to be enabled in the
1089 		 * RX direction.
1090 		 */
1091 		return 1;
1092 	}
1093 
1094 	return 0;
1095 }
1096 
1097 static int
1098 avp_dev_rx_queue_setup(struct rte_eth_dev *eth_dev,
1099 		       uint16_t rx_queue_id,
1100 		       uint16_t nb_rx_desc,
1101 		       unsigned int socket_id,
1102 		       const struct rte_eth_rxconf *rx_conf,
1103 		       struct rte_mempool *pool)
1104 {
1105 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1106 	struct rte_pktmbuf_pool_private *mbp_priv;
1107 	struct avp_queue *rxq;
1108 
1109 	if (rx_queue_id >= eth_dev->data->nb_rx_queues) {
1110 		PMD_DRV_LOG(ERR, "RX queue id is out of range: rx_queue_id=%u, nb_rx_queues=%u\n",
1111 			    rx_queue_id, eth_dev->data->nb_rx_queues);
1112 		return -EINVAL;
1113 	}
1114 
1115 	/* Save mbuf pool pointer */
1116 	avp->pool = pool;
1117 
1118 	/* Save the local mbuf size */
1119 	mbp_priv = rte_mempool_get_priv(pool);
1120 	avp->guest_mbuf_size = (uint16_t)(mbp_priv->mbuf_data_room_size);
1121 	avp->guest_mbuf_size -= RTE_PKTMBUF_HEADROOM;
1122 
1123 	if (avp_dev_enable_scattered(eth_dev, avp)) {
1124 		if (!eth_dev->data->scattered_rx) {
1125 			PMD_DRV_LOG(NOTICE, "AVP device configured for chained mbufs\n");
1126 			eth_dev->data->scattered_rx = 1;
1127 			eth_dev->rx_pkt_burst = avp_recv_scattered_pkts;
1128 			eth_dev->tx_pkt_burst = avp_xmit_scattered_pkts;
1129 		}
1130 	}
1131 
1132 	PMD_DRV_LOG(DEBUG, "AVP max_rx_pkt_len=(%u,%u) mbuf_size=(%u,%u)\n",
1133 		    avp->max_rx_pkt_len,
1134 		    eth_dev->data->dev_conf.rxmode.max_rx_pkt_len,
1135 		    avp->host_mbuf_size,
1136 		    avp->guest_mbuf_size);
1137 
1138 	/* allocate a queue object */
1139 	rxq = rte_zmalloc_socket("ethdev RX queue", sizeof(struct avp_queue),
1140 				 RTE_CACHE_LINE_SIZE, socket_id);
1141 	if (rxq == NULL) {
1142 		PMD_DRV_LOG(ERR, "Failed to allocate new Rx queue object\n");
1143 		return -ENOMEM;
1144 	}
1145 
1146 	/* save back pointers to AVP and Ethernet devices */
1147 	rxq->avp = avp;
1148 	rxq->dev_data = eth_dev->data;
1149 	eth_dev->data->rx_queues[rx_queue_id] = (void *)rxq;
1150 
1151 	/* setup the queue receive mapping for the current queue. */
1152 	_avp_set_rx_queue_mappings(eth_dev, rx_queue_id);
1153 
1154 	PMD_DRV_LOG(DEBUG, "Rx queue %u setup at %p\n", rx_queue_id, rxq);
1155 
1156 	(void)nb_rx_desc;
1157 	(void)rx_conf;
1158 	return 0;
1159 }
1160 
1161 static int
1162 avp_dev_tx_queue_setup(struct rte_eth_dev *eth_dev,
1163 		       uint16_t tx_queue_id,
1164 		       uint16_t nb_tx_desc,
1165 		       unsigned int socket_id,
1166 		       const struct rte_eth_txconf *tx_conf)
1167 {
1168 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1169 	struct avp_queue *txq;
1170 
1171 	if (tx_queue_id >= eth_dev->data->nb_tx_queues) {
1172 		PMD_DRV_LOG(ERR, "TX queue id is out of range: tx_queue_id=%u, nb_tx_queues=%u\n",
1173 			    tx_queue_id, eth_dev->data->nb_tx_queues);
1174 		return -EINVAL;
1175 	}
1176 
1177 	/* allocate a queue object */
1178 	txq = rte_zmalloc_socket("ethdev TX queue", sizeof(struct avp_queue),
1179 				 RTE_CACHE_LINE_SIZE, socket_id);
1180 	if (txq == NULL) {
1181 		PMD_DRV_LOG(ERR, "Failed to allocate new Tx queue object\n");
1182 		return -ENOMEM;
1183 	}
1184 
1185 	/* only the configured set of transmit queues are used */
1186 	txq->queue_id = tx_queue_id;
1187 	txq->queue_base = tx_queue_id;
1188 	txq->queue_limit = tx_queue_id;
1189 
1190 	/* save back pointers to AVP and Ethernet devices */
1191 	txq->avp = avp;
1192 	txq->dev_data = eth_dev->data;
1193 	eth_dev->data->tx_queues[tx_queue_id] = (void *)txq;
1194 
1195 	PMD_DRV_LOG(DEBUG, "Tx queue %u setup at %p\n", tx_queue_id, txq);
1196 
1197 	(void)nb_tx_desc;
1198 	(void)tx_conf;
1199 	return 0;
1200 }
1201 
1202 static inline int
1203 _avp_cmp_ether_addr(struct rte_ether_addr *a, struct rte_ether_addr *b)
1204 {
1205 	uint16_t *_a = (uint16_t *)&a->addr_bytes[0];
1206 	uint16_t *_b = (uint16_t *)&b->addr_bytes[0];
1207 	return (_a[0] ^ _b[0]) | (_a[1] ^ _b[1]) | (_a[2] ^ _b[2]);
1208 }
1209 
1210 static inline int
1211 _avp_mac_filter(struct avp_dev *avp, struct rte_mbuf *m)
1212 {
1213 	struct rte_ether_hdr *eth = rte_pktmbuf_mtod(m, struct rte_ether_hdr *);
1214 
1215 	if (likely(_avp_cmp_ether_addr(&avp->ethaddr, &eth->d_addr) == 0)) {
1216 		/* allow all packets destined to our address */
1217 		return 0;
1218 	}
1219 
1220 	if (likely(rte_is_broadcast_ether_addr(&eth->d_addr))) {
1221 		/* allow all broadcast packets */
1222 		return 0;
1223 	}
1224 
1225 	if (likely(rte_is_multicast_ether_addr(&eth->d_addr))) {
1226 		/* allow all multicast packets */
1227 		return 0;
1228 	}
1229 
1230 	if (avp->flags & AVP_F_PROMISC) {
1231 		/* allow all packets when in promiscuous mode */
1232 		return 0;
1233 	}
1234 
1235 	return -1;
1236 }
1237 
1238 #ifdef RTE_LIBRTE_AVP_DEBUG_BUFFERS
1239 static inline void
1240 __avp_dev_buffer_sanity_check(struct avp_dev *avp, struct rte_avp_desc *buf)
1241 {
1242 	struct rte_avp_desc *first_buf;
1243 	struct rte_avp_desc *pkt_buf;
1244 	unsigned int pkt_len;
1245 	unsigned int nb_segs;
1246 	void *pkt_data;
1247 	unsigned int i;
1248 
1249 	first_buf = avp_dev_translate_buffer(avp, buf);
1250 
1251 	i = 0;
1252 	pkt_len = 0;
1253 	nb_segs = first_buf->nb_segs;
1254 	do {
1255 		/* Adjust pointers for guest addressing */
1256 		pkt_buf = avp_dev_translate_buffer(avp, buf);
1257 		if (pkt_buf == NULL)
1258 			rte_panic("bad buffer: segment %u has an invalid address %p\n",
1259 				  i, buf);
1260 		pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1261 		if (pkt_data == NULL)
1262 			rte_panic("bad buffer: segment %u has a NULL data pointer\n",
1263 				  i);
1264 		if (pkt_buf->data_len == 0)
1265 			rte_panic("bad buffer: segment %u has 0 data length\n",
1266 				  i);
1267 		pkt_len += pkt_buf->data_len;
1268 		nb_segs--;
1269 		i++;
1270 
1271 	} while (nb_segs && (buf = pkt_buf->next) != NULL);
1272 
1273 	if (nb_segs != 0)
1274 		rte_panic("bad buffer: expected %u segments found %u\n",
1275 			  first_buf->nb_segs, (first_buf->nb_segs - nb_segs));
1276 	if (pkt_len != first_buf->pkt_len)
1277 		rte_panic("bad buffer: expected length %u found %u\n",
1278 			  first_buf->pkt_len, pkt_len);
1279 }
1280 
1281 #define avp_dev_buffer_sanity_check(a, b) \
1282 	__avp_dev_buffer_sanity_check((a), (b))
1283 
1284 #else /* RTE_LIBRTE_AVP_DEBUG_BUFFERS */
1285 
1286 #define avp_dev_buffer_sanity_check(a, b) do {} while (0)
1287 
1288 #endif
1289 
1290 /*
1291  * Copy a host buffer chain to a set of mbufs.	This function assumes that
1292  * there exactly the required number of mbufs to copy all source bytes.
1293  */
1294 static inline struct rte_mbuf *
1295 avp_dev_copy_from_buffers(struct avp_dev *avp,
1296 			  struct rte_avp_desc *buf,
1297 			  struct rte_mbuf **mbufs,
1298 			  unsigned int count)
1299 {
1300 	struct rte_mbuf *m_previous = NULL;
1301 	struct rte_avp_desc *pkt_buf;
1302 	unsigned int total_length = 0;
1303 	unsigned int copy_length;
1304 	unsigned int src_offset;
1305 	struct rte_mbuf *m;
1306 	uint16_t ol_flags;
1307 	uint16_t vlan_tci;
1308 	void *pkt_data;
1309 	unsigned int i;
1310 
1311 	avp_dev_buffer_sanity_check(avp, buf);
1312 
1313 	/* setup the first source buffer */
1314 	pkt_buf = avp_dev_translate_buffer(avp, buf);
1315 	pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1316 	total_length = pkt_buf->pkt_len;
1317 	src_offset = 0;
1318 
1319 	if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
1320 		ol_flags = PKT_RX_VLAN;
1321 		vlan_tci = pkt_buf->vlan_tci;
1322 	} else {
1323 		ol_flags = 0;
1324 		vlan_tci = 0;
1325 	}
1326 
1327 	for (i = 0; (i < count) && (buf != NULL); i++) {
1328 		/* fill each destination buffer */
1329 		m = mbufs[i];
1330 
1331 		if (m_previous != NULL)
1332 			m_previous->next = m;
1333 
1334 		m_previous = m;
1335 
1336 		do {
1337 			/*
1338 			 * Copy as many source buffers as will fit in the
1339 			 * destination buffer.
1340 			 */
1341 			copy_length = RTE_MIN((avp->guest_mbuf_size -
1342 					       rte_pktmbuf_data_len(m)),
1343 					      (pkt_buf->data_len -
1344 					       src_offset));
1345 			rte_memcpy(RTE_PTR_ADD(rte_pktmbuf_mtod(m, void *),
1346 					       rte_pktmbuf_data_len(m)),
1347 				   RTE_PTR_ADD(pkt_data, src_offset),
1348 				   copy_length);
1349 			rte_pktmbuf_data_len(m) += copy_length;
1350 			src_offset += copy_length;
1351 
1352 			if (likely(src_offset == pkt_buf->data_len)) {
1353 				/* need a new source buffer */
1354 				buf = pkt_buf->next;
1355 				if (buf != NULL) {
1356 					pkt_buf = avp_dev_translate_buffer(
1357 						avp, buf);
1358 					pkt_data = avp_dev_translate_buffer(
1359 						avp, pkt_buf->data);
1360 					src_offset = 0;
1361 				}
1362 			}
1363 
1364 			if (unlikely(rte_pktmbuf_data_len(m) ==
1365 				     avp->guest_mbuf_size)) {
1366 				/* need a new destination mbuf */
1367 				break;
1368 			}
1369 
1370 		} while (buf != NULL);
1371 	}
1372 
1373 	m = mbufs[0];
1374 	m->ol_flags = ol_flags;
1375 	m->nb_segs = count;
1376 	rte_pktmbuf_pkt_len(m) = total_length;
1377 	m->vlan_tci = vlan_tci;
1378 
1379 	__rte_mbuf_sanity_check(m, 1);
1380 
1381 	return m;
1382 }
1383 
1384 static uint16_t
1385 avp_recv_scattered_pkts(void *rx_queue,
1386 			struct rte_mbuf **rx_pkts,
1387 			uint16_t nb_pkts)
1388 {
1389 	struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1390 	struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
1391 	struct rte_mbuf *mbufs[RTE_AVP_MAX_MBUF_SEGMENTS];
1392 	struct avp_dev *avp = rxq->avp;
1393 	struct rte_avp_desc *pkt_buf;
1394 	struct rte_avp_fifo *free_q;
1395 	struct rte_avp_fifo *rx_q;
1396 	struct rte_avp_desc *buf;
1397 	unsigned int count, avail, n;
1398 	unsigned int guest_mbuf_size;
1399 	struct rte_mbuf *m;
1400 	unsigned int required;
1401 	unsigned int buf_len;
1402 	unsigned int port_id;
1403 	unsigned int i;
1404 
1405 	if (unlikely(avp->flags & AVP_F_DETACHED)) {
1406 		/* VM live migration in progress */
1407 		return 0;
1408 	}
1409 
1410 	guest_mbuf_size = avp->guest_mbuf_size;
1411 	port_id = avp->port_id;
1412 	rx_q = avp->rx_q[rxq->queue_id];
1413 	free_q = avp->free_q[rxq->queue_id];
1414 
1415 	/* setup next queue to service */
1416 	rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
1417 		(rxq->queue_id + 1) : rxq->queue_base;
1418 
1419 	/* determine how many slots are available in the free queue */
1420 	count = avp_fifo_free_count(free_q);
1421 
1422 	/* determine how many packets are available in the rx queue */
1423 	avail = avp_fifo_count(rx_q);
1424 
1425 	/* determine how many packets can be received */
1426 	count = RTE_MIN(count, avail);
1427 	count = RTE_MIN(count, nb_pkts);
1428 	count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
1429 
1430 	if (unlikely(count == 0)) {
1431 		/* no free buffers, or no buffers on the rx queue */
1432 		return 0;
1433 	}
1434 
1435 	/* retrieve pending packets */
1436 	n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
1437 	PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
1438 		   count, rx_q);
1439 
1440 	count = 0;
1441 	for (i = 0; i < n; i++) {
1442 		/* prefetch next entry while processing current one */
1443 		if (i + 1 < n) {
1444 			pkt_buf = avp_dev_translate_buffer(avp,
1445 							   avp_bufs[i + 1]);
1446 			rte_prefetch0(pkt_buf);
1447 		}
1448 		buf = avp_bufs[i];
1449 
1450 		/* Peek into the first buffer to determine the total length */
1451 		pkt_buf = avp_dev_translate_buffer(avp, buf);
1452 		buf_len = pkt_buf->pkt_len;
1453 
1454 		/* Allocate enough mbufs to receive the entire packet */
1455 		required = (buf_len + guest_mbuf_size - 1) / guest_mbuf_size;
1456 		if (rte_pktmbuf_alloc_bulk(avp->pool, mbufs, required)) {
1457 			rxq->dev_data->rx_mbuf_alloc_failed++;
1458 			continue;
1459 		}
1460 
1461 		/* Copy the data from the buffers to our mbufs */
1462 		m = avp_dev_copy_from_buffers(avp, buf, mbufs, required);
1463 
1464 		/* finalize mbuf */
1465 		m->port = port_id;
1466 
1467 		if (_avp_mac_filter(avp, m) != 0) {
1468 			/* silently discard packets not destined to our MAC */
1469 			rte_pktmbuf_free(m);
1470 			continue;
1471 		}
1472 
1473 		/* return new mbuf to caller */
1474 		rx_pkts[count++] = m;
1475 		rxq->bytes += buf_len;
1476 	}
1477 
1478 	rxq->packets += count;
1479 
1480 	/* return the buffers to the free queue */
1481 	avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
1482 
1483 	return count;
1484 }
1485 
1486 
1487 static uint16_t
1488 avp_recv_pkts(void *rx_queue,
1489 	      struct rte_mbuf **rx_pkts,
1490 	      uint16_t nb_pkts)
1491 {
1492 	struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1493 	struct rte_avp_desc *avp_bufs[AVP_MAX_RX_BURST];
1494 	struct avp_dev *avp = rxq->avp;
1495 	struct rte_avp_desc *pkt_buf;
1496 	struct rte_avp_fifo *free_q;
1497 	struct rte_avp_fifo *rx_q;
1498 	unsigned int count, avail, n;
1499 	unsigned int pkt_len;
1500 	struct rte_mbuf *m;
1501 	char *pkt_data;
1502 	unsigned int i;
1503 
1504 	if (unlikely(avp->flags & AVP_F_DETACHED)) {
1505 		/* VM live migration in progress */
1506 		return 0;
1507 	}
1508 
1509 	rx_q = avp->rx_q[rxq->queue_id];
1510 	free_q = avp->free_q[rxq->queue_id];
1511 
1512 	/* setup next queue to service */
1513 	rxq->queue_id = (rxq->queue_id < rxq->queue_limit) ?
1514 		(rxq->queue_id + 1) : rxq->queue_base;
1515 
1516 	/* determine how many slots are available in the free queue */
1517 	count = avp_fifo_free_count(free_q);
1518 
1519 	/* determine how many packets are available in the rx queue */
1520 	avail = avp_fifo_count(rx_q);
1521 
1522 	/* determine how many packets can be received */
1523 	count = RTE_MIN(count, avail);
1524 	count = RTE_MIN(count, nb_pkts);
1525 	count = RTE_MIN(count, (unsigned int)AVP_MAX_RX_BURST);
1526 
1527 	if (unlikely(count == 0)) {
1528 		/* no free buffers, or no buffers on the rx queue */
1529 		return 0;
1530 	}
1531 
1532 	/* retrieve pending packets */
1533 	n = avp_fifo_get(rx_q, (void **)&avp_bufs, count);
1534 	PMD_RX_LOG(DEBUG, "Receiving %u packets from Rx queue at %p\n",
1535 		   count, rx_q);
1536 
1537 	count = 0;
1538 	for (i = 0; i < n; i++) {
1539 		/* prefetch next entry while processing current one */
1540 		if (i < n - 1) {
1541 			pkt_buf = avp_dev_translate_buffer(avp,
1542 							   avp_bufs[i + 1]);
1543 			rte_prefetch0(pkt_buf);
1544 		}
1545 
1546 		/* Adjust host pointers for guest addressing */
1547 		pkt_buf = avp_dev_translate_buffer(avp, avp_bufs[i]);
1548 		pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1549 		pkt_len = pkt_buf->pkt_len;
1550 
1551 		if (unlikely((pkt_len > avp->guest_mbuf_size) ||
1552 			     (pkt_buf->nb_segs > 1))) {
1553 			/*
1554 			 * application should be using the scattered receive
1555 			 * function
1556 			 */
1557 			rxq->errors++;
1558 			continue;
1559 		}
1560 
1561 		/* process each packet to be transmitted */
1562 		m = rte_pktmbuf_alloc(avp->pool);
1563 		if (unlikely(m == NULL)) {
1564 			rxq->dev_data->rx_mbuf_alloc_failed++;
1565 			continue;
1566 		}
1567 
1568 		/* copy data out of the host buffer to our buffer */
1569 		m->data_off = RTE_PKTMBUF_HEADROOM;
1570 		rte_memcpy(rte_pktmbuf_mtod(m, void *), pkt_data, pkt_len);
1571 
1572 		/* initialize the local mbuf */
1573 		rte_pktmbuf_data_len(m) = pkt_len;
1574 		rte_pktmbuf_pkt_len(m) = pkt_len;
1575 		m->port = avp->port_id;
1576 
1577 		if (pkt_buf->ol_flags & RTE_AVP_RX_VLAN_PKT) {
1578 			m->ol_flags = PKT_RX_VLAN;
1579 			m->vlan_tci = pkt_buf->vlan_tci;
1580 		}
1581 
1582 		if (_avp_mac_filter(avp, m) != 0) {
1583 			/* silently discard packets not destined to our MAC */
1584 			rte_pktmbuf_free(m);
1585 			continue;
1586 		}
1587 
1588 		/* return new mbuf to caller */
1589 		rx_pkts[count++] = m;
1590 		rxq->bytes += pkt_len;
1591 	}
1592 
1593 	rxq->packets += count;
1594 
1595 	/* return the buffers to the free queue */
1596 	avp_fifo_put(free_q, (void **)&avp_bufs[0], n);
1597 
1598 	return count;
1599 }
1600 
1601 /*
1602  * Copy a chained mbuf to a set of host buffers.  This function assumes that
1603  * there are sufficient destination buffers to contain the entire source
1604  * packet.
1605  */
1606 static inline uint16_t
1607 avp_dev_copy_to_buffers(struct avp_dev *avp,
1608 			struct rte_mbuf *mbuf,
1609 			struct rte_avp_desc **buffers,
1610 			unsigned int count)
1611 {
1612 	struct rte_avp_desc *previous_buf = NULL;
1613 	struct rte_avp_desc *first_buf = NULL;
1614 	struct rte_avp_desc *pkt_buf;
1615 	struct rte_avp_desc *buf;
1616 	size_t total_length;
1617 	struct rte_mbuf *m;
1618 	size_t copy_length;
1619 	size_t src_offset;
1620 	char *pkt_data;
1621 	unsigned int i;
1622 
1623 	__rte_mbuf_sanity_check(mbuf, 1);
1624 
1625 	m = mbuf;
1626 	src_offset = 0;
1627 	total_length = rte_pktmbuf_pkt_len(m);
1628 	for (i = 0; (i < count) && (m != NULL); i++) {
1629 		/* fill each destination buffer */
1630 		buf = buffers[i];
1631 
1632 		if (i < count - 1) {
1633 			/* prefetch next entry while processing this one */
1634 			pkt_buf = avp_dev_translate_buffer(avp, buffers[i + 1]);
1635 			rte_prefetch0(pkt_buf);
1636 		}
1637 
1638 		/* Adjust pointers for guest addressing */
1639 		pkt_buf = avp_dev_translate_buffer(avp, buf);
1640 		pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1641 
1642 		/* setup the buffer chain */
1643 		if (previous_buf != NULL)
1644 			previous_buf->next = buf;
1645 		else
1646 			first_buf = pkt_buf;
1647 
1648 		previous_buf = pkt_buf;
1649 
1650 		do {
1651 			/*
1652 			 * copy as many source mbuf segments as will fit in the
1653 			 * destination buffer.
1654 			 */
1655 			copy_length = RTE_MIN((avp->host_mbuf_size -
1656 					       pkt_buf->data_len),
1657 					      (rte_pktmbuf_data_len(m) -
1658 					       src_offset));
1659 			rte_memcpy(RTE_PTR_ADD(pkt_data, pkt_buf->data_len),
1660 				   RTE_PTR_ADD(rte_pktmbuf_mtod(m, void *),
1661 					       src_offset),
1662 				   copy_length);
1663 			pkt_buf->data_len += copy_length;
1664 			src_offset += copy_length;
1665 
1666 			if (likely(src_offset == rte_pktmbuf_data_len(m))) {
1667 				/* need a new source buffer */
1668 				m = m->next;
1669 				src_offset = 0;
1670 			}
1671 
1672 			if (unlikely(pkt_buf->data_len ==
1673 				     avp->host_mbuf_size)) {
1674 				/* need a new destination buffer */
1675 				break;
1676 			}
1677 
1678 		} while (m != NULL);
1679 	}
1680 
1681 	first_buf->nb_segs = count;
1682 	first_buf->pkt_len = total_length;
1683 
1684 	if (mbuf->ol_flags & PKT_TX_VLAN_PKT) {
1685 		first_buf->ol_flags |= RTE_AVP_TX_VLAN_PKT;
1686 		first_buf->vlan_tci = mbuf->vlan_tci;
1687 	}
1688 
1689 	avp_dev_buffer_sanity_check(avp, buffers[0]);
1690 
1691 	return total_length;
1692 }
1693 
1694 
1695 static uint16_t
1696 avp_xmit_scattered_pkts(void *tx_queue,
1697 			struct rte_mbuf **tx_pkts,
1698 			uint16_t nb_pkts)
1699 {
1700 	struct rte_avp_desc *avp_bufs[(AVP_MAX_TX_BURST *
1701 				       RTE_AVP_MAX_MBUF_SEGMENTS)];
1702 	struct avp_queue *txq = (struct avp_queue *)tx_queue;
1703 	struct rte_avp_desc *tx_bufs[AVP_MAX_TX_BURST];
1704 	struct avp_dev *avp = txq->avp;
1705 	struct rte_avp_fifo *alloc_q;
1706 	struct rte_avp_fifo *tx_q;
1707 	unsigned int count, avail, n;
1708 	unsigned int orig_nb_pkts;
1709 	struct rte_mbuf *m;
1710 	unsigned int required;
1711 	unsigned int segments;
1712 	unsigned int tx_bytes;
1713 	unsigned int i;
1714 
1715 	orig_nb_pkts = nb_pkts;
1716 	if (unlikely(avp->flags & AVP_F_DETACHED)) {
1717 		/* VM live migration in progress */
1718 		/* TODO ... buffer for X packets then drop? */
1719 		txq->errors += nb_pkts;
1720 		return 0;
1721 	}
1722 
1723 	tx_q = avp->tx_q[txq->queue_id];
1724 	alloc_q = avp->alloc_q[txq->queue_id];
1725 
1726 	/* limit the number of transmitted packets to the max burst size */
1727 	if (unlikely(nb_pkts > AVP_MAX_TX_BURST))
1728 		nb_pkts = AVP_MAX_TX_BURST;
1729 
1730 	/* determine how many buffers are available to copy into */
1731 	avail = avp_fifo_count(alloc_q);
1732 	if (unlikely(avail > (AVP_MAX_TX_BURST *
1733 			      RTE_AVP_MAX_MBUF_SEGMENTS)))
1734 		avail = AVP_MAX_TX_BURST * RTE_AVP_MAX_MBUF_SEGMENTS;
1735 
1736 	/* determine how many slots are available in the transmit queue */
1737 	count = avp_fifo_free_count(tx_q);
1738 
1739 	/* determine how many packets can be sent */
1740 	nb_pkts = RTE_MIN(count, nb_pkts);
1741 
1742 	/* determine how many packets will fit in the available buffers */
1743 	count = 0;
1744 	segments = 0;
1745 	for (i = 0; i < nb_pkts; i++) {
1746 		m = tx_pkts[i];
1747 		if (likely(i < (unsigned int)nb_pkts - 1)) {
1748 			/* prefetch next entry while processing this one */
1749 			rte_prefetch0(tx_pkts[i + 1]);
1750 		}
1751 		required = (rte_pktmbuf_pkt_len(m) + avp->host_mbuf_size - 1) /
1752 			avp->host_mbuf_size;
1753 
1754 		if (unlikely((required == 0) ||
1755 			     (required > RTE_AVP_MAX_MBUF_SEGMENTS)))
1756 			break;
1757 		else if (unlikely(required + segments > avail))
1758 			break;
1759 		segments += required;
1760 		count++;
1761 	}
1762 	nb_pkts = count;
1763 
1764 	if (unlikely(nb_pkts == 0)) {
1765 		/* no available buffers, or no space on the tx queue */
1766 		txq->errors += orig_nb_pkts;
1767 		return 0;
1768 	}
1769 
1770 	PMD_TX_LOG(DEBUG, "Sending %u packets on Tx queue at %p\n",
1771 		   nb_pkts, tx_q);
1772 
1773 	/* retrieve sufficient send buffers */
1774 	n = avp_fifo_get(alloc_q, (void **)&avp_bufs, segments);
1775 	if (unlikely(n != segments)) {
1776 		PMD_TX_LOG(DEBUG, "Failed to allocate buffers "
1777 			   "n=%u, segments=%u, orig=%u\n",
1778 			   n, segments, orig_nb_pkts);
1779 		txq->errors += orig_nb_pkts;
1780 		return 0;
1781 	}
1782 
1783 	tx_bytes = 0;
1784 	count = 0;
1785 	for (i = 0; i < nb_pkts; i++) {
1786 		/* process each packet to be transmitted */
1787 		m = tx_pkts[i];
1788 
1789 		/* determine how many buffers are required for this packet */
1790 		required = (rte_pktmbuf_pkt_len(m) + avp->host_mbuf_size - 1) /
1791 			avp->host_mbuf_size;
1792 
1793 		tx_bytes += avp_dev_copy_to_buffers(avp, m,
1794 						    &avp_bufs[count], required);
1795 		tx_bufs[i] = avp_bufs[count];
1796 		count += required;
1797 
1798 		/* free the original mbuf */
1799 		rte_pktmbuf_free(m);
1800 	}
1801 
1802 	txq->packets += nb_pkts;
1803 	txq->bytes += tx_bytes;
1804 
1805 #ifdef RTE_LIBRTE_AVP_DEBUG_BUFFERS
1806 	for (i = 0; i < nb_pkts; i++)
1807 		avp_dev_buffer_sanity_check(avp, tx_bufs[i]);
1808 #endif
1809 
1810 	/* send the packets */
1811 	n = avp_fifo_put(tx_q, (void **)&tx_bufs[0], nb_pkts);
1812 	if (unlikely(n != orig_nb_pkts))
1813 		txq->errors += (orig_nb_pkts - n);
1814 
1815 	return n;
1816 }
1817 
1818 
1819 static uint16_t
1820 avp_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
1821 {
1822 	struct avp_queue *txq = (struct avp_queue *)tx_queue;
1823 	struct rte_avp_desc *avp_bufs[AVP_MAX_TX_BURST];
1824 	struct avp_dev *avp = txq->avp;
1825 	struct rte_avp_desc *pkt_buf;
1826 	struct rte_avp_fifo *alloc_q;
1827 	struct rte_avp_fifo *tx_q;
1828 	unsigned int count, avail, n;
1829 	struct rte_mbuf *m;
1830 	unsigned int pkt_len;
1831 	unsigned int tx_bytes;
1832 	char *pkt_data;
1833 	unsigned int i;
1834 
1835 	if (unlikely(avp->flags & AVP_F_DETACHED)) {
1836 		/* VM live migration in progress */
1837 		/* TODO ... buffer for X packets then drop?! */
1838 		txq->errors++;
1839 		return 0;
1840 	}
1841 
1842 	tx_q = avp->tx_q[txq->queue_id];
1843 	alloc_q = avp->alloc_q[txq->queue_id];
1844 
1845 	/* limit the number of transmitted packets to the max burst size */
1846 	if (unlikely(nb_pkts > AVP_MAX_TX_BURST))
1847 		nb_pkts = AVP_MAX_TX_BURST;
1848 
1849 	/* determine how many buffers are available to copy into */
1850 	avail = avp_fifo_count(alloc_q);
1851 
1852 	/* determine how many slots are available in the transmit queue */
1853 	count = avp_fifo_free_count(tx_q);
1854 
1855 	/* determine how many packets can be sent */
1856 	count = RTE_MIN(count, avail);
1857 	count = RTE_MIN(count, nb_pkts);
1858 
1859 	if (unlikely(count == 0)) {
1860 		/* no available buffers, or no space on the tx queue */
1861 		txq->errors += nb_pkts;
1862 		return 0;
1863 	}
1864 
1865 	PMD_TX_LOG(DEBUG, "Sending %u packets on Tx queue at %p\n",
1866 		   count, tx_q);
1867 
1868 	/* retrieve sufficient send buffers */
1869 	n = avp_fifo_get(alloc_q, (void **)&avp_bufs, count);
1870 	if (unlikely(n != count)) {
1871 		txq->errors++;
1872 		return 0;
1873 	}
1874 
1875 	tx_bytes = 0;
1876 	for (i = 0; i < count; i++) {
1877 		/* prefetch next entry while processing the current one */
1878 		if (i < count - 1) {
1879 			pkt_buf = avp_dev_translate_buffer(avp,
1880 							   avp_bufs[i + 1]);
1881 			rte_prefetch0(pkt_buf);
1882 		}
1883 
1884 		/* process each packet to be transmitted */
1885 		m = tx_pkts[i];
1886 
1887 		/* Adjust pointers for guest addressing */
1888 		pkt_buf = avp_dev_translate_buffer(avp, avp_bufs[i]);
1889 		pkt_data = avp_dev_translate_buffer(avp, pkt_buf->data);
1890 		pkt_len = rte_pktmbuf_pkt_len(m);
1891 
1892 		if (unlikely((pkt_len > avp->guest_mbuf_size) ||
1893 					 (pkt_len > avp->host_mbuf_size))) {
1894 			/*
1895 			 * application should be using the scattered transmit
1896 			 * function; send it truncated to avoid the performance
1897 			 * hit of having to manage returning the already
1898 			 * allocated buffer to the free list.  This should not
1899 			 * happen since the application should have set the
1900 			 * max_rx_pkt_len based on its MTU and it should be
1901 			 * policing its own packet sizes.
1902 			 */
1903 			txq->errors++;
1904 			pkt_len = RTE_MIN(avp->guest_mbuf_size,
1905 					  avp->host_mbuf_size);
1906 		}
1907 
1908 		/* copy data out of our mbuf and into the AVP buffer */
1909 		rte_memcpy(pkt_data, rte_pktmbuf_mtod(m, void *), pkt_len);
1910 		pkt_buf->pkt_len = pkt_len;
1911 		pkt_buf->data_len = pkt_len;
1912 		pkt_buf->nb_segs = 1;
1913 		pkt_buf->next = NULL;
1914 
1915 		if (m->ol_flags & PKT_TX_VLAN_PKT) {
1916 			pkt_buf->ol_flags |= RTE_AVP_TX_VLAN_PKT;
1917 			pkt_buf->vlan_tci = m->vlan_tci;
1918 		}
1919 
1920 		tx_bytes += pkt_len;
1921 
1922 		/* free the original mbuf */
1923 		rte_pktmbuf_free(m);
1924 	}
1925 
1926 	txq->packets += count;
1927 	txq->bytes += tx_bytes;
1928 
1929 	/* send the packets */
1930 	n = avp_fifo_put(tx_q, (void **)&avp_bufs[0], count);
1931 
1932 	return n;
1933 }
1934 
1935 static void
1936 avp_dev_rx_queue_release(void *rx_queue)
1937 {
1938 	struct avp_queue *rxq = (struct avp_queue *)rx_queue;
1939 	struct avp_dev *avp = rxq->avp;
1940 	struct rte_eth_dev_data *data = avp->dev_data;
1941 	unsigned int i;
1942 
1943 	for (i = 0; i < avp->num_rx_queues; i++) {
1944 		if (data->rx_queues[i] == rxq)
1945 			data->rx_queues[i] = NULL;
1946 	}
1947 }
1948 
1949 static void
1950 avp_dev_tx_queue_release(void *tx_queue)
1951 {
1952 	struct avp_queue *txq = (struct avp_queue *)tx_queue;
1953 	struct avp_dev *avp = txq->avp;
1954 	struct rte_eth_dev_data *data = avp->dev_data;
1955 	unsigned int i;
1956 
1957 	for (i = 0; i < avp->num_tx_queues; i++) {
1958 		if (data->tx_queues[i] == txq)
1959 			data->tx_queues[i] = NULL;
1960 	}
1961 }
1962 
1963 static int
1964 avp_dev_configure(struct rte_eth_dev *eth_dev)
1965 {
1966 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
1967 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
1968 	struct rte_avp_device_info *host_info;
1969 	struct rte_avp_device_config config;
1970 	int mask = 0;
1971 	void *addr;
1972 	int ret;
1973 
1974 	rte_spinlock_lock(&avp->lock);
1975 	if (avp->flags & AVP_F_DETACHED) {
1976 		PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
1977 		ret = -ENOTSUP;
1978 		goto unlock;
1979 	}
1980 
1981 	addr = pci_dev->mem_resource[RTE_AVP_PCI_DEVICE_BAR].addr;
1982 	host_info = (struct rte_avp_device_info *)addr;
1983 
1984 	/* Setup required number of queues */
1985 	_avp_set_queue_counts(eth_dev);
1986 
1987 	mask = (ETH_VLAN_STRIP_MASK |
1988 		ETH_VLAN_FILTER_MASK |
1989 		ETH_VLAN_EXTEND_MASK);
1990 	ret = avp_vlan_offload_set(eth_dev, mask);
1991 	if (ret < 0) {
1992 		PMD_DRV_LOG(ERR, "VLAN offload set failed by host, ret=%d\n",
1993 			    ret);
1994 		goto unlock;
1995 	}
1996 
1997 	/* update device config */
1998 	memset(&config, 0, sizeof(config));
1999 	config.device_id = host_info->device_id;
2000 	config.driver_type = RTE_AVP_DRIVER_TYPE_DPDK;
2001 	config.driver_version = AVP_DPDK_DRIVER_VERSION;
2002 	config.features = avp->features;
2003 	config.num_tx_queues = avp->num_tx_queues;
2004 	config.num_rx_queues = avp->num_rx_queues;
2005 
2006 	ret = avp_dev_ctrl_set_config(eth_dev, &config);
2007 	if (ret < 0) {
2008 		PMD_DRV_LOG(ERR, "Config request failed by host, ret=%d\n",
2009 			    ret);
2010 		goto unlock;
2011 	}
2012 
2013 	avp->flags |= AVP_F_CONFIGURED;
2014 	ret = 0;
2015 
2016 unlock:
2017 	rte_spinlock_unlock(&avp->lock);
2018 	return ret;
2019 }
2020 
2021 static int
2022 avp_dev_start(struct rte_eth_dev *eth_dev)
2023 {
2024 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2025 	int ret;
2026 
2027 	rte_spinlock_lock(&avp->lock);
2028 	if (avp->flags & AVP_F_DETACHED) {
2029 		PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2030 		ret = -ENOTSUP;
2031 		goto unlock;
2032 	}
2033 
2034 	/* update link state */
2035 	ret = avp_dev_ctrl_set_link_state(eth_dev, 1);
2036 	if (ret < 0) {
2037 		PMD_DRV_LOG(ERR, "Link state change failed by host, ret=%d\n",
2038 			    ret);
2039 		goto unlock;
2040 	}
2041 
2042 	/* remember current link state */
2043 	avp->flags |= AVP_F_LINKUP;
2044 
2045 	ret = 0;
2046 
2047 unlock:
2048 	rte_spinlock_unlock(&avp->lock);
2049 	return ret;
2050 }
2051 
2052 static void
2053 avp_dev_stop(struct rte_eth_dev *eth_dev)
2054 {
2055 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2056 	int ret;
2057 
2058 	rte_spinlock_lock(&avp->lock);
2059 	if (avp->flags & AVP_F_DETACHED) {
2060 		PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2061 		goto unlock;
2062 	}
2063 
2064 	/* remember current link state */
2065 	avp->flags &= ~AVP_F_LINKUP;
2066 
2067 	/* update link state */
2068 	ret = avp_dev_ctrl_set_link_state(eth_dev, 0);
2069 	if (ret < 0) {
2070 		PMD_DRV_LOG(ERR, "Link state change failed by host, ret=%d\n",
2071 			    ret);
2072 	}
2073 
2074 unlock:
2075 	rte_spinlock_unlock(&avp->lock);
2076 }
2077 
2078 static void
2079 avp_dev_close(struct rte_eth_dev *eth_dev)
2080 {
2081 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2082 	int ret;
2083 
2084 	rte_spinlock_lock(&avp->lock);
2085 	if (avp->flags & AVP_F_DETACHED) {
2086 		PMD_DRV_LOG(ERR, "Operation not supported during VM live migration\n");
2087 		goto unlock;
2088 	}
2089 
2090 	/* remember current link state */
2091 	avp->flags &= ~AVP_F_LINKUP;
2092 	avp->flags &= ~AVP_F_CONFIGURED;
2093 
2094 	ret = avp_dev_disable_interrupts(eth_dev);
2095 	if (ret < 0) {
2096 		PMD_DRV_LOG(ERR, "Failed to disable interrupts\n");
2097 		/* continue */
2098 	}
2099 
2100 	/* update device state */
2101 	ret = avp_dev_ctrl_shutdown(eth_dev);
2102 	if (ret < 0) {
2103 		PMD_DRV_LOG(ERR, "Device shutdown failed by host, ret=%d\n",
2104 			    ret);
2105 		/* continue */
2106 	}
2107 
2108 unlock:
2109 	rte_spinlock_unlock(&avp->lock);
2110 }
2111 
2112 static int
2113 avp_dev_link_update(struct rte_eth_dev *eth_dev,
2114 					__rte_unused int wait_to_complete)
2115 {
2116 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2117 	struct rte_eth_link *link = &eth_dev->data->dev_link;
2118 
2119 	link->link_speed = ETH_SPEED_NUM_10G;
2120 	link->link_duplex = ETH_LINK_FULL_DUPLEX;
2121 	link->link_status = !!(avp->flags & AVP_F_LINKUP);
2122 
2123 	return -1;
2124 }
2125 
2126 static void
2127 avp_dev_promiscuous_enable(struct rte_eth_dev *eth_dev)
2128 {
2129 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2130 
2131 	rte_spinlock_lock(&avp->lock);
2132 	if ((avp->flags & AVP_F_PROMISC) == 0) {
2133 		avp->flags |= AVP_F_PROMISC;
2134 		PMD_DRV_LOG(DEBUG, "Promiscuous mode enabled on %u\n",
2135 			    eth_dev->data->port_id);
2136 	}
2137 	rte_spinlock_unlock(&avp->lock);
2138 }
2139 
2140 static void
2141 avp_dev_promiscuous_disable(struct rte_eth_dev *eth_dev)
2142 {
2143 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2144 
2145 	rte_spinlock_lock(&avp->lock);
2146 	if ((avp->flags & AVP_F_PROMISC) != 0) {
2147 		avp->flags &= ~AVP_F_PROMISC;
2148 		PMD_DRV_LOG(DEBUG, "Promiscuous mode disabled on %u\n",
2149 			    eth_dev->data->port_id);
2150 	}
2151 	rte_spinlock_unlock(&avp->lock);
2152 }
2153 
2154 static void
2155 avp_dev_info_get(struct rte_eth_dev *eth_dev,
2156 		 struct rte_eth_dev_info *dev_info)
2157 {
2158 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2159 
2160 	dev_info->max_rx_queues = avp->max_rx_queues;
2161 	dev_info->max_tx_queues = avp->max_tx_queues;
2162 	dev_info->min_rx_bufsize = AVP_MIN_RX_BUFSIZE;
2163 	dev_info->max_rx_pktlen = avp->max_rx_pkt_len;
2164 	dev_info->max_mac_addrs = AVP_MAX_MAC_ADDRS;
2165 	if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
2166 		dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
2167 		dev_info->tx_offload_capa = DEV_TX_OFFLOAD_VLAN_INSERT;
2168 	}
2169 }
2170 
2171 static int
2172 avp_vlan_offload_set(struct rte_eth_dev *eth_dev, int mask)
2173 {
2174 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2175 	struct rte_eth_conf *dev_conf = &eth_dev->data->dev_conf;
2176 	uint64_t offloads = dev_conf->rxmode.offloads;
2177 
2178 	if (mask & ETH_VLAN_STRIP_MASK) {
2179 		if (avp->host_features & RTE_AVP_FEATURE_VLAN_OFFLOAD) {
2180 			if (offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
2181 				avp->features |= RTE_AVP_FEATURE_VLAN_OFFLOAD;
2182 			else
2183 				avp->features &= ~RTE_AVP_FEATURE_VLAN_OFFLOAD;
2184 		} else {
2185 			PMD_DRV_LOG(ERR, "VLAN strip offload not supported\n");
2186 		}
2187 	}
2188 
2189 	if (mask & ETH_VLAN_FILTER_MASK) {
2190 		if (offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
2191 			PMD_DRV_LOG(ERR, "VLAN filter offload not supported\n");
2192 	}
2193 
2194 	if (mask & ETH_VLAN_EXTEND_MASK) {
2195 		if (offloads & DEV_RX_OFFLOAD_VLAN_EXTEND)
2196 			PMD_DRV_LOG(ERR, "VLAN extend offload not supported\n");
2197 	}
2198 
2199 	return 0;
2200 }
2201 
2202 static int
2203 avp_dev_stats_get(struct rte_eth_dev *eth_dev, struct rte_eth_stats *stats)
2204 {
2205 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2206 	unsigned int i;
2207 
2208 	for (i = 0; i < avp->num_rx_queues; i++) {
2209 		struct avp_queue *rxq = avp->dev_data->rx_queues[i];
2210 
2211 		if (rxq) {
2212 			stats->ipackets += rxq->packets;
2213 			stats->ibytes += rxq->bytes;
2214 			stats->ierrors += rxq->errors;
2215 
2216 			stats->q_ipackets[i] += rxq->packets;
2217 			stats->q_ibytes[i] += rxq->bytes;
2218 			stats->q_errors[i] += rxq->errors;
2219 		}
2220 	}
2221 
2222 	for (i = 0; i < avp->num_tx_queues; i++) {
2223 		struct avp_queue *txq = avp->dev_data->tx_queues[i];
2224 
2225 		if (txq) {
2226 			stats->opackets += txq->packets;
2227 			stats->obytes += txq->bytes;
2228 			stats->oerrors += txq->errors;
2229 
2230 			stats->q_opackets[i] += txq->packets;
2231 			stats->q_obytes[i] += txq->bytes;
2232 			stats->q_errors[i] += txq->errors;
2233 		}
2234 	}
2235 
2236 	return 0;
2237 }
2238 
2239 static void
2240 avp_dev_stats_reset(struct rte_eth_dev *eth_dev)
2241 {
2242 	struct avp_dev *avp = AVP_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
2243 	unsigned int i;
2244 
2245 	for (i = 0; i < avp->num_rx_queues; i++) {
2246 		struct avp_queue *rxq = avp->dev_data->rx_queues[i];
2247 
2248 		if (rxq) {
2249 			rxq->bytes = 0;
2250 			rxq->packets = 0;
2251 			rxq->errors = 0;
2252 		}
2253 	}
2254 
2255 	for (i = 0; i < avp->num_tx_queues; i++) {
2256 		struct avp_queue *txq = avp->dev_data->tx_queues[i];
2257 
2258 		if (txq) {
2259 			txq->bytes = 0;
2260 			txq->packets = 0;
2261 			txq->errors = 0;
2262 		}
2263 	}
2264 }
2265 
2266 RTE_PMD_REGISTER_PCI(net_avp, rte_avp_pmd);
2267 RTE_PMD_REGISTER_PCI_TABLE(net_avp, pci_id_avp_map);
2268 
2269 RTE_INIT(avp_init_log)
2270 {
2271 	avp_logtype_driver = rte_log_register("pmd.net.avp.driver");
2272 	if (avp_logtype_driver >= 0)
2273 		rte_log_set_level(avp_logtype_driver, RTE_LOG_NOTICE);
2274 }
2275