xref: /dpdk/drivers/net/vmxnet3/vmxnet3_ethdev.c (revision 200bc52e5aa0d72e70464c9cd22b55cf536ed13c)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2015 Intel Corporation
3  */
4 
5 #include <sys/queue.h>
6 #include <stdio.h>
7 #include <errno.h>
8 #include <stdint.h>
9 #include <string.h>
10 #include <unistd.h>
11 #include <stdarg.h>
12 #include <fcntl.h>
13 #include <inttypes.h>
14 #include <rte_byteorder.h>
15 #include <rte_common.h>
16 #include <rte_cycles.h>
17 
18 #include <rte_interrupts.h>
19 #include <rte_log.h>
20 #include <rte_debug.h>
21 #include <rte_pci.h>
22 #include <rte_bus_pci.h>
23 #include <rte_branch_prediction.h>
24 #include <rte_memory.h>
25 #include <rte_memzone.h>
26 #include <rte_eal.h>
27 #include <rte_alarm.h>
28 #include <rte_ether.h>
29 #include <rte_ethdev_driver.h>
30 #include <rte_ethdev_pci.h>
31 #include <rte_string_fns.h>
32 #include <rte_malloc.h>
33 #include <rte_dev.h>
34 
35 #include "base/vmxnet3_defs.h"
36 
37 #include "vmxnet3_ring.h"
38 #include "vmxnet3_logs.h"
39 #include "vmxnet3_ethdev.h"
40 
41 #define PROCESS_SYS_EVENTS 0
42 
43 #define	VMXNET3_TX_MAX_SEG	UINT8_MAX
44 
45 #define VMXNET3_TX_OFFLOAD_CAP		\
46 	(DEV_TX_OFFLOAD_VLAN_INSERT |	\
47 	 DEV_TX_OFFLOAD_IPV4_CKSUM |	\
48 	 DEV_TX_OFFLOAD_TCP_CKSUM |	\
49 	 DEV_TX_OFFLOAD_UDP_CKSUM |	\
50 	 DEV_TX_OFFLOAD_TCP_TSO |	\
51 	 DEV_TX_OFFLOAD_MULTI_SEGS)
52 
53 #define VMXNET3_RX_OFFLOAD_CAP		\
54 	(DEV_RX_OFFLOAD_VLAN_STRIP |	\
55 	 DEV_RX_OFFLOAD_VLAN_FILTER |   \
56 	 DEV_RX_OFFLOAD_SCATTER |	\
57 	 DEV_RX_OFFLOAD_IPV4_CKSUM |	\
58 	 DEV_RX_OFFLOAD_UDP_CKSUM |	\
59 	 DEV_RX_OFFLOAD_TCP_CKSUM |	\
60 	 DEV_RX_OFFLOAD_TCP_LRO |	\
61 	 DEV_RX_OFFLOAD_JUMBO_FRAME)
62 
63 static int eth_vmxnet3_dev_init(struct rte_eth_dev *eth_dev);
64 static int eth_vmxnet3_dev_uninit(struct rte_eth_dev *eth_dev);
65 static int vmxnet3_dev_configure(struct rte_eth_dev *dev);
66 static int vmxnet3_dev_start(struct rte_eth_dev *dev);
67 static void vmxnet3_dev_stop(struct rte_eth_dev *dev);
68 static void vmxnet3_dev_close(struct rte_eth_dev *dev);
69 static void vmxnet3_dev_set_rxmode(struct vmxnet3_hw *hw, uint32_t feature, int set);
70 static void vmxnet3_dev_promiscuous_enable(struct rte_eth_dev *dev);
71 static void vmxnet3_dev_promiscuous_disable(struct rte_eth_dev *dev);
72 static void vmxnet3_dev_allmulticast_enable(struct rte_eth_dev *dev);
73 static void vmxnet3_dev_allmulticast_disable(struct rte_eth_dev *dev);
74 static int __vmxnet3_dev_link_update(struct rte_eth_dev *dev,
75 				     int wait_to_complete);
76 static int vmxnet3_dev_link_update(struct rte_eth_dev *dev,
77 				   int wait_to_complete);
78 static void vmxnet3_hw_stats_save(struct vmxnet3_hw *hw);
79 static int vmxnet3_dev_stats_get(struct rte_eth_dev *dev,
80 				  struct rte_eth_stats *stats);
81 static void vmxnet3_dev_stats_reset(struct rte_eth_dev *dev);
82 static int vmxnet3_dev_xstats_get_names(struct rte_eth_dev *dev,
83 					struct rte_eth_xstat_name *xstats,
84 					unsigned int n);
85 static int vmxnet3_dev_xstats_get(struct rte_eth_dev *dev,
86 				  struct rte_eth_xstat *xstats, unsigned int n);
87 static void vmxnet3_dev_info_get(struct rte_eth_dev *dev,
88 				 struct rte_eth_dev_info *dev_info);
89 static const uint32_t *
90 vmxnet3_dev_supported_ptypes_get(struct rte_eth_dev *dev);
91 static int vmxnet3_dev_vlan_filter_set(struct rte_eth_dev *dev,
92 				       uint16_t vid, int on);
93 static int vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
94 static int vmxnet3_mac_addr_set(struct rte_eth_dev *dev,
95 				 struct rte_ether_addr *mac_addr);
96 static void vmxnet3_interrupt_handler(void *param);
97 
98 int vmxnet3_logtype_init;
99 int vmxnet3_logtype_driver;
100 
101 /*
102  * The set of PCI devices this driver supports
103  */
104 #define VMWARE_PCI_VENDOR_ID 0x15AD
105 #define VMWARE_DEV_ID_VMXNET3 0x07B0
106 static const struct rte_pci_id pci_id_vmxnet3_map[] = {
107 	{ RTE_PCI_DEVICE(VMWARE_PCI_VENDOR_ID, VMWARE_DEV_ID_VMXNET3) },
108 	{ .vendor_id = 0, /* sentinel */ },
109 };
110 
111 static const struct eth_dev_ops vmxnet3_eth_dev_ops = {
112 	.dev_configure        = vmxnet3_dev_configure,
113 	.dev_start            = vmxnet3_dev_start,
114 	.dev_stop             = vmxnet3_dev_stop,
115 	.dev_close            = vmxnet3_dev_close,
116 	.promiscuous_enable   = vmxnet3_dev_promiscuous_enable,
117 	.promiscuous_disable  = vmxnet3_dev_promiscuous_disable,
118 	.allmulticast_enable  = vmxnet3_dev_allmulticast_enable,
119 	.allmulticast_disable = vmxnet3_dev_allmulticast_disable,
120 	.link_update          = vmxnet3_dev_link_update,
121 	.stats_get            = vmxnet3_dev_stats_get,
122 	.xstats_get_names     = vmxnet3_dev_xstats_get_names,
123 	.xstats_get           = vmxnet3_dev_xstats_get,
124 	.stats_reset          = vmxnet3_dev_stats_reset,
125 	.mac_addr_set         = vmxnet3_mac_addr_set,
126 	.dev_infos_get        = vmxnet3_dev_info_get,
127 	.dev_supported_ptypes_get = vmxnet3_dev_supported_ptypes_get,
128 	.vlan_filter_set      = vmxnet3_dev_vlan_filter_set,
129 	.vlan_offload_set     = vmxnet3_dev_vlan_offload_set,
130 	.rx_queue_setup       = vmxnet3_dev_rx_queue_setup,
131 	.rx_queue_release     = vmxnet3_dev_rx_queue_release,
132 	.tx_queue_setup       = vmxnet3_dev_tx_queue_setup,
133 	.tx_queue_release     = vmxnet3_dev_tx_queue_release,
134 };
135 
136 struct vmxnet3_xstats_name_off {
137 	char name[RTE_ETH_XSTATS_NAME_SIZE];
138 	unsigned int offset;
139 };
140 
141 /* tx_qX_ is prepended to the name string here */
142 static const struct vmxnet3_xstats_name_off vmxnet3_txq_stat_strings[] = {
143 	{"drop_total",         offsetof(struct vmxnet3_txq_stats, drop_total)},
144 	{"drop_too_many_segs", offsetof(struct vmxnet3_txq_stats, drop_too_many_segs)},
145 	{"drop_tso",           offsetof(struct vmxnet3_txq_stats, drop_tso)},
146 	{"tx_ring_full",       offsetof(struct vmxnet3_txq_stats, tx_ring_full)},
147 };
148 
149 /* rx_qX_ is prepended to the name string here */
150 static const struct vmxnet3_xstats_name_off vmxnet3_rxq_stat_strings[] = {
151 	{"drop_total",           offsetof(struct vmxnet3_rxq_stats, drop_total)},
152 	{"drop_err",             offsetof(struct vmxnet3_rxq_stats, drop_err)},
153 	{"drop_fcs",             offsetof(struct vmxnet3_rxq_stats, drop_fcs)},
154 	{"rx_buf_alloc_failure", offsetof(struct vmxnet3_rxq_stats, rx_buf_alloc_failure)},
155 };
156 
157 static const struct rte_memzone *
158 gpa_zone_reserve(struct rte_eth_dev *dev, uint32_t size,
159 		 const char *post_string, int socket_id,
160 		 uint16_t align, bool reuse)
161 {
162 	char z_name[RTE_MEMZONE_NAMESIZE];
163 	const struct rte_memzone *mz;
164 
165 	snprintf(z_name, sizeof(z_name), "eth_p%d_%s",
166 			dev->data->port_id, post_string);
167 
168 	mz = rte_memzone_lookup(z_name);
169 	if (!reuse) {
170 		if (mz)
171 			rte_memzone_free(mz);
172 		return rte_memzone_reserve_aligned(z_name, size, socket_id,
173 				RTE_MEMZONE_IOVA_CONTIG, align);
174 	}
175 
176 	if (mz)
177 		return mz;
178 
179 	return rte_memzone_reserve_aligned(z_name, size, socket_id,
180 			RTE_MEMZONE_IOVA_CONTIG, align);
181 }
182 
183 /*
184  * This function is based on vmxnet3_disable_intr()
185  */
186 static void
187 vmxnet3_disable_intr(struct vmxnet3_hw *hw)
188 {
189 	int i;
190 
191 	PMD_INIT_FUNC_TRACE();
192 
193 	hw->shared->devRead.intrConf.intrCtrl |= VMXNET3_IC_DISABLE_ALL;
194 	for (i = 0; i < hw->num_intrs; i++)
195 		VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + i * 8, 1);
196 }
197 
198 static void
199 vmxnet3_enable_intr(struct vmxnet3_hw *hw)
200 {
201 	int i;
202 
203 	PMD_INIT_FUNC_TRACE();
204 
205 	hw->shared->devRead.intrConf.intrCtrl &= ~VMXNET3_IC_DISABLE_ALL;
206 	for (i = 0; i < hw->num_intrs; i++)
207 		VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + i * 8, 0);
208 }
209 
210 /*
211  * Gets tx data ring descriptor size.
212  */
213 static uint16_t
214 eth_vmxnet3_txdata_get(struct vmxnet3_hw *hw)
215 {
216 	uint16 txdata_desc_size;
217 
218 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
219 			       VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
220 	txdata_desc_size = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
221 
222 	return (txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE ||
223 		txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE ||
224 		txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK) ?
225 		sizeof(struct Vmxnet3_TxDataDesc) : txdata_desc_size;
226 }
227 
228 /*
229  * It returns 0 on success.
230  */
231 static int
232 eth_vmxnet3_dev_init(struct rte_eth_dev *eth_dev)
233 {
234 	struct rte_pci_device *pci_dev;
235 	struct vmxnet3_hw *hw = eth_dev->data->dev_private;
236 	uint32_t mac_hi, mac_lo, ver;
237 	struct rte_eth_link link;
238 
239 	PMD_INIT_FUNC_TRACE();
240 
241 	eth_dev->dev_ops = &vmxnet3_eth_dev_ops;
242 	eth_dev->rx_pkt_burst = &vmxnet3_recv_pkts;
243 	eth_dev->tx_pkt_burst = &vmxnet3_xmit_pkts;
244 	eth_dev->tx_pkt_prepare = vmxnet3_prep_pkts;
245 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
246 
247 	/*
248 	 * for secondary processes, we don't initialize any further as primary
249 	 * has already done this work.
250 	 */
251 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
252 		return 0;
253 
254 	rte_eth_copy_pci_info(eth_dev, pci_dev);
255 
256 	/* Vendor and Device ID need to be set before init of shared code */
257 	hw->device_id = pci_dev->id.device_id;
258 	hw->vendor_id = pci_dev->id.vendor_id;
259 	hw->hw_addr0 = (void *)pci_dev->mem_resource[0].addr;
260 	hw->hw_addr1 = (void *)pci_dev->mem_resource[1].addr;
261 
262 	hw->num_rx_queues = 1;
263 	hw->num_tx_queues = 1;
264 	hw->bufs_per_pkt = 1;
265 
266 	/* Check h/w version compatibility with driver. */
267 	ver = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_VRRS);
268 	PMD_INIT_LOG(DEBUG, "Hardware version : %d", ver);
269 
270 	if (ver & (1 << VMXNET3_REV_4)) {
271 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
272 				       1 << VMXNET3_REV_4);
273 		hw->version = VMXNET3_REV_4 + 1;
274 	} else if (ver & (1 << VMXNET3_REV_3)) {
275 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
276 				       1 << VMXNET3_REV_3);
277 		hw->version = VMXNET3_REV_3 + 1;
278 	} else if (ver & (1 << VMXNET3_REV_2)) {
279 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
280 				       1 << VMXNET3_REV_2);
281 		hw->version = VMXNET3_REV_2 + 1;
282 	} else if (ver & (1 << VMXNET3_REV_1)) {
283 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
284 				       1 << VMXNET3_REV_1);
285 		hw->version = VMXNET3_REV_1 + 1;
286 	} else {
287 		PMD_INIT_LOG(ERR, "Incompatible hardware version: %d", ver);
288 		return -EIO;
289 	}
290 
291 	PMD_INIT_LOG(DEBUG, "Using device version %d\n", hw->version);
292 
293 	/* Check UPT version compatibility with driver. */
294 	ver = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_UVRS);
295 	PMD_INIT_LOG(DEBUG, "UPT hardware version : %d", ver);
296 	if (ver & 0x1)
297 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_UVRS, 1);
298 	else {
299 		PMD_INIT_LOG(ERR, "Incompatible UPT version.");
300 		return -EIO;
301 	}
302 
303 	/* Getting MAC Address */
304 	mac_lo = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_MACL);
305 	mac_hi = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_MACH);
306 	memcpy(hw->perm_addr, &mac_lo, 4);
307 	memcpy(hw->perm_addr + 4, &mac_hi, 2);
308 
309 	/* Allocate memory for storing MAC addresses */
310 	eth_dev->data->mac_addrs = rte_zmalloc("vmxnet3", RTE_ETHER_ADDR_LEN *
311 					       VMXNET3_MAX_MAC_ADDRS, 0);
312 	if (eth_dev->data->mac_addrs == NULL) {
313 		PMD_INIT_LOG(ERR,
314 			     "Failed to allocate %d bytes needed to store MAC addresses",
315 			     RTE_ETHER_ADDR_LEN * VMXNET3_MAX_MAC_ADDRS);
316 		return -ENOMEM;
317 	}
318 	/* Copy the permanent MAC address */
319 	rte_ether_addr_copy((struct rte_ether_addr *)hw->perm_addr,
320 			&eth_dev->data->mac_addrs[0]);
321 
322 	PMD_INIT_LOG(DEBUG, "MAC Address : %02x:%02x:%02x:%02x:%02x:%02x",
323 		     hw->perm_addr[0], hw->perm_addr[1], hw->perm_addr[2],
324 		     hw->perm_addr[3], hw->perm_addr[4], hw->perm_addr[5]);
325 
326 	/* Flag to call rte_eth_dev_release_port() in rte_eth_dev_close(). */
327 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
328 
329 	/* Put device in Quiesce Mode */
330 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
331 
332 	/* allow untagged pkts */
333 	VMXNET3_SET_VFTABLE_ENTRY(hw->shadow_vfta, 0);
334 
335 	hw->txdata_desc_size = VMXNET3_VERSION_GE_3(hw) ?
336 		eth_vmxnet3_txdata_get(hw) : sizeof(struct Vmxnet3_TxDataDesc);
337 
338 	hw->rxdata_desc_size = VMXNET3_VERSION_GE_3(hw) ?
339 		VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
340 	RTE_ASSERT((hw->rxdata_desc_size & ~VMXNET3_RXDATA_DESC_SIZE_MASK) ==
341 		   hw->rxdata_desc_size);
342 
343 	/* clear shadow stats */
344 	memset(hw->saved_tx_stats, 0, sizeof(hw->saved_tx_stats));
345 	memset(hw->saved_rx_stats, 0, sizeof(hw->saved_rx_stats));
346 
347 	/* clear snapshot stats */
348 	memset(hw->snapshot_tx_stats, 0, sizeof(hw->snapshot_tx_stats));
349 	memset(hw->snapshot_rx_stats, 0, sizeof(hw->snapshot_rx_stats));
350 
351 	/* set the initial link status */
352 	memset(&link, 0, sizeof(link));
353 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
354 	link.link_speed = ETH_SPEED_NUM_10G;
355 	link.link_autoneg = ETH_LINK_FIXED;
356 	rte_eth_linkstatus_set(eth_dev, &link);
357 
358 	return 0;
359 }
360 
361 static int
362 eth_vmxnet3_dev_uninit(struct rte_eth_dev *eth_dev)
363 {
364 	struct vmxnet3_hw *hw = eth_dev->data->dev_private;
365 
366 	PMD_INIT_FUNC_TRACE();
367 
368 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
369 		return 0;
370 
371 	if (hw->adapter_stopped == 0) {
372 		PMD_INIT_LOG(DEBUG, "Device has not been closed.");
373 		return -EBUSY;
374 	}
375 
376 	eth_dev->dev_ops = NULL;
377 	eth_dev->rx_pkt_burst = NULL;
378 	eth_dev->tx_pkt_burst = NULL;
379 	eth_dev->tx_pkt_prepare = NULL;
380 
381 	return 0;
382 }
383 
384 static int eth_vmxnet3_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
385 	struct rte_pci_device *pci_dev)
386 {
387 	return rte_eth_dev_pci_generic_probe(pci_dev,
388 		sizeof(struct vmxnet3_hw), eth_vmxnet3_dev_init);
389 }
390 
391 static int eth_vmxnet3_pci_remove(struct rte_pci_device *pci_dev)
392 {
393 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_vmxnet3_dev_uninit);
394 }
395 
396 static struct rte_pci_driver rte_vmxnet3_pmd = {
397 	.id_table = pci_id_vmxnet3_map,
398 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
399 	.probe = eth_vmxnet3_pci_probe,
400 	.remove = eth_vmxnet3_pci_remove,
401 };
402 
403 static int
404 vmxnet3_dev_configure(struct rte_eth_dev *dev)
405 {
406 	const struct rte_memzone *mz;
407 	struct vmxnet3_hw *hw = dev->data->dev_private;
408 	size_t size;
409 
410 	PMD_INIT_FUNC_TRACE();
411 
412 	if (dev->data->nb_tx_queues > VMXNET3_MAX_TX_QUEUES ||
413 	    dev->data->nb_rx_queues > VMXNET3_MAX_RX_QUEUES) {
414 		PMD_INIT_LOG(ERR, "ERROR: Number of queues not supported");
415 		return -EINVAL;
416 	}
417 
418 	if (!rte_is_power_of_2(dev->data->nb_rx_queues)) {
419 		PMD_INIT_LOG(ERR, "ERROR: Number of rx queues not power of 2");
420 		return -EINVAL;
421 	}
422 
423 	size = dev->data->nb_rx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
424 		dev->data->nb_tx_queues * sizeof(struct Vmxnet3_RxQueueDesc);
425 
426 	if (size > UINT16_MAX)
427 		return -EINVAL;
428 
429 	hw->num_rx_queues = (uint8_t)dev->data->nb_rx_queues;
430 	hw->num_tx_queues = (uint8_t)dev->data->nb_tx_queues;
431 
432 	/*
433 	 * Allocate a memzone for Vmxnet3_DriverShared - Vmxnet3_DSDevRead
434 	 * on current socket
435 	 */
436 	mz = gpa_zone_reserve(dev, sizeof(struct Vmxnet3_DriverShared),
437 			      "shared", rte_socket_id(), 8, 1);
438 
439 	if (mz == NULL) {
440 		PMD_INIT_LOG(ERR, "ERROR: Creating shared zone");
441 		return -ENOMEM;
442 	}
443 	memset(mz->addr, 0, mz->len);
444 
445 	hw->shared = mz->addr;
446 	hw->sharedPA = mz->iova;
447 
448 	/*
449 	 * Allocate a memzone for Vmxnet3_RxQueueDesc - Vmxnet3_TxQueueDesc
450 	 * on current socket.
451 	 *
452 	 * We cannot reuse this memzone from previous allocation as its size
453 	 * depends on the number of tx and rx queues, which could be different
454 	 * from one config to another.
455 	 */
456 	mz = gpa_zone_reserve(dev, size, "queuedesc", rte_socket_id(),
457 			      VMXNET3_QUEUE_DESC_ALIGN, 0);
458 	if (mz == NULL) {
459 		PMD_INIT_LOG(ERR, "ERROR: Creating queue descriptors zone");
460 		return -ENOMEM;
461 	}
462 	memset(mz->addr, 0, mz->len);
463 
464 	hw->tqd_start = (Vmxnet3_TxQueueDesc *)mz->addr;
465 	hw->rqd_start = (Vmxnet3_RxQueueDesc *)(hw->tqd_start + hw->num_tx_queues);
466 
467 	hw->queueDescPA = mz->iova;
468 	hw->queue_desc_len = (uint16_t)size;
469 
470 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
471 		/* Allocate memory structure for UPT1_RSSConf and configure */
472 		mz = gpa_zone_reserve(dev, sizeof(struct VMXNET3_RSSConf),
473 				      "rss_conf", rte_socket_id(),
474 				      RTE_CACHE_LINE_SIZE, 1);
475 		if (mz == NULL) {
476 			PMD_INIT_LOG(ERR,
477 				     "ERROR: Creating rss_conf structure zone");
478 			return -ENOMEM;
479 		}
480 		memset(mz->addr, 0, mz->len);
481 
482 		hw->rss_conf = mz->addr;
483 		hw->rss_confPA = mz->iova;
484 	}
485 
486 	return 0;
487 }
488 
489 static void
490 vmxnet3_write_mac(struct vmxnet3_hw *hw, const uint8_t *addr)
491 {
492 	uint32_t val;
493 
494 	PMD_INIT_LOG(DEBUG,
495 		     "Writing MAC Address : %02x:%02x:%02x:%02x:%02x:%02x",
496 		     addr[0], addr[1], addr[2],
497 		     addr[3], addr[4], addr[5]);
498 
499 	memcpy(&val, addr, 4);
500 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_MACL, val);
501 
502 	memcpy(&val, addr + 4, 2);
503 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_MACH, val);
504 }
505 
506 static int
507 vmxnet3_dev_setup_memreg(struct rte_eth_dev *dev)
508 {
509 	struct vmxnet3_hw *hw = dev->data->dev_private;
510 	Vmxnet3_DriverShared *shared = hw->shared;
511 	Vmxnet3_CmdInfo *cmdInfo;
512 	struct rte_mempool *mp[VMXNET3_MAX_RX_QUEUES];
513 	uint8_t index[VMXNET3_MAX_RX_QUEUES + VMXNET3_MAX_TX_QUEUES];
514 	uint32_t num, i, j, size;
515 
516 	if (hw->memRegsPA == 0) {
517 		const struct rte_memzone *mz;
518 
519 		size = sizeof(Vmxnet3_MemRegs) +
520 			(VMXNET3_MAX_RX_QUEUES + VMXNET3_MAX_TX_QUEUES) *
521 			sizeof(Vmxnet3_MemoryRegion);
522 
523 		mz = gpa_zone_reserve(dev, size, "memRegs", rte_socket_id(), 8,
524 				      1);
525 		if (mz == NULL) {
526 			PMD_INIT_LOG(ERR, "ERROR: Creating memRegs zone");
527 			return -ENOMEM;
528 		}
529 		memset(mz->addr, 0, mz->len);
530 		hw->memRegs = mz->addr;
531 		hw->memRegsPA = mz->iova;
532 	}
533 
534 	num = hw->num_rx_queues;
535 
536 	for (i = 0; i < num; i++) {
537 		vmxnet3_rx_queue_t *rxq = dev->data->rx_queues[i];
538 
539 		mp[i] = rxq->mp;
540 		index[i] = 1 << i;
541 	}
542 
543 	/*
544 	 * The same mempool could be used by multiple queues. In such a case,
545 	 * remove duplicate mempool entries. Only one entry is kept with
546 	 * bitmask indicating queues that are using this mempool.
547 	 */
548 	for (i = 1; i < num; i++) {
549 		for (j = 0; j < i; j++) {
550 			if (mp[i] == mp[j]) {
551 				mp[i] = NULL;
552 				index[j] |= 1 << i;
553 				break;
554 			}
555 		}
556 	}
557 
558 	j = 0;
559 	for (i = 0; i < num; i++) {
560 		if (mp[i] == NULL)
561 			continue;
562 
563 		Vmxnet3_MemoryRegion *mr = &hw->memRegs->memRegs[j];
564 
565 		mr->startPA =
566 			(uintptr_t)STAILQ_FIRST(&mp[i]->mem_list)->iova;
567 		mr->length = STAILQ_FIRST(&mp[i]->mem_list)->len <= INT32_MAX ?
568 			STAILQ_FIRST(&mp[i]->mem_list)->len : INT32_MAX;
569 		mr->txQueueBits = index[i];
570 		mr->rxQueueBits = index[i];
571 
572 		PMD_INIT_LOG(INFO,
573 			     "index: %u startPA: %" PRIu64 " length: %u, "
574 			     "rxBits: %x",
575 			     j, mr->startPA, mr->length, mr->rxQueueBits);
576 		j++;
577 	}
578 	hw->memRegs->numRegs = j;
579 	PMD_INIT_LOG(INFO, "numRegs: %u", j);
580 
581 	size = sizeof(Vmxnet3_MemRegs) +
582 		(j - 1) * sizeof(Vmxnet3_MemoryRegion);
583 
584 	cmdInfo = &shared->cu.cmdInfo;
585 	cmdInfo->varConf.confVer = 1;
586 	cmdInfo->varConf.confLen = size;
587 	cmdInfo->varConf.confPA = hw->memRegsPA;
588 
589 	return 0;
590 }
591 
592 static int
593 vmxnet3_setup_driver_shared(struct rte_eth_dev *dev)
594 {
595 	struct rte_eth_conf port_conf = dev->data->dev_conf;
596 	struct vmxnet3_hw *hw = dev->data->dev_private;
597 	uint32_t mtu = dev->data->mtu;
598 	Vmxnet3_DriverShared *shared = hw->shared;
599 	Vmxnet3_DSDevRead *devRead = &shared->devRead;
600 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
601 	uint32_t i;
602 	int ret;
603 
604 	hw->mtu = mtu;
605 
606 	shared->magic = VMXNET3_REV1_MAGIC;
607 	devRead->misc.driverInfo.version = VMXNET3_DRIVER_VERSION_NUM;
608 
609 	/* Setting up Guest OS information */
610 	devRead->misc.driverInfo.gos.gosBits   = sizeof(void *) == 4 ?
611 		VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64;
612 	devRead->misc.driverInfo.gos.gosType   = VMXNET3_GOS_TYPE_LINUX;
613 	devRead->misc.driverInfo.vmxnet3RevSpt = 1;
614 	devRead->misc.driverInfo.uptVerSpt     = 1;
615 
616 	devRead->misc.mtu = rte_le_to_cpu_32(mtu);
617 	devRead->misc.queueDescPA  = hw->queueDescPA;
618 	devRead->misc.queueDescLen = hw->queue_desc_len;
619 	devRead->misc.numTxQueues  = hw->num_tx_queues;
620 	devRead->misc.numRxQueues  = hw->num_rx_queues;
621 
622 	/*
623 	 * Set number of interrupts to 1
624 	 * PMD by default disables all the interrupts but this is MUST
625 	 * to activate device. It needs at least one interrupt for
626 	 * link events to handle
627 	 */
628 	hw->num_intrs = devRead->intrConf.numIntrs = 1;
629 	devRead->intrConf.intrCtrl |= VMXNET3_IC_DISABLE_ALL;
630 
631 	for (i = 0; i < hw->num_tx_queues; i++) {
632 		Vmxnet3_TxQueueDesc *tqd = &hw->tqd_start[i];
633 		vmxnet3_tx_queue_t *txq  = dev->data->tx_queues[i];
634 
635 		txq->shared = &hw->tqd_start[i];
636 
637 		tqd->ctrl.txNumDeferred  = 0;
638 		tqd->ctrl.txThreshold    = 1;
639 		tqd->conf.txRingBasePA   = txq->cmd_ring.basePA;
640 		tqd->conf.compRingBasePA = txq->comp_ring.basePA;
641 		tqd->conf.dataRingBasePA = txq->data_ring.basePA;
642 
643 		tqd->conf.txRingSize   = txq->cmd_ring.size;
644 		tqd->conf.compRingSize = txq->comp_ring.size;
645 		tqd->conf.dataRingSize = txq->data_ring.size;
646 		tqd->conf.txDataRingDescSize = txq->txdata_desc_size;
647 		tqd->conf.intrIdx      = txq->comp_ring.intr_idx;
648 		tqd->status.stopped    = TRUE;
649 		tqd->status.error      = 0;
650 		memset(&tqd->stats, 0, sizeof(tqd->stats));
651 	}
652 
653 	for (i = 0; i < hw->num_rx_queues; i++) {
654 		Vmxnet3_RxQueueDesc *rqd  = &hw->rqd_start[i];
655 		vmxnet3_rx_queue_t *rxq   = dev->data->rx_queues[i];
656 
657 		rxq->shared = &hw->rqd_start[i];
658 
659 		rqd->conf.rxRingBasePA[0] = rxq->cmd_ring[0].basePA;
660 		rqd->conf.rxRingBasePA[1] = rxq->cmd_ring[1].basePA;
661 		rqd->conf.compRingBasePA  = rxq->comp_ring.basePA;
662 
663 		rqd->conf.rxRingSize[0]   = rxq->cmd_ring[0].size;
664 		rqd->conf.rxRingSize[1]   = rxq->cmd_ring[1].size;
665 		rqd->conf.compRingSize    = rxq->comp_ring.size;
666 		rqd->conf.intrIdx         = rxq->comp_ring.intr_idx;
667 		if (VMXNET3_VERSION_GE_3(hw)) {
668 			rqd->conf.rxDataRingBasePA = rxq->data_ring.basePA;
669 			rqd->conf.rxDataRingDescSize = rxq->data_desc_size;
670 		}
671 		rqd->status.stopped       = TRUE;
672 		rqd->status.error         = 0;
673 		memset(&rqd->stats, 0, sizeof(rqd->stats));
674 	}
675 
676 	/* RxMode set to 0 of VMXNET3_RXM_xxx */
677 	devRead->rxFilterConf.rxMode = 0;
678 
679 	/* Setting up feature flags */
680 	if (rx_offloads & DEV_RX_OFFLOAD_CHECKSUM)
681 		devRead->misc.uptFeatures |= VMXNET3_F_RXCSUM;
682 
683 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
684 		devRead->misc.uptFeatures |= VMXNET3_F_LRO;
685 		devRead->misc.maxNumRxSG = 0;
686 	}
687 
688 	if (port_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
689 		ret = vmxnet3_rss_configure(dev);
690 		if (ret != VMXNET3_SUCCESS)
691 			return ret;
692 
693 		devRead->misc.uptFeatures |= VMXNET3_F_RSS;
694 		devRead->rssConfDesc.confVer = 1;
695 		devRead->rssConfDesc.confLen = sizeof(struct VMXNET3_RSSConf);
696 		devRead->rssConfDesc.confPA  = hw->rss_confPA;
697 	}
698 
699 	ret = vmxnet3_dev_vlan_offload_set(dev,
700 			ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK);
701 	if (ret)
702 		return ret;
703 
704 	vmxnet3_write_mac(hw, dev->data->mac_addrs->addr_bytes);
705 
706 	return VMXNET3_SUCCESS;
707 }
708 
709 /*
710  * Configure device link speed and setup link.
711  * Must be called after eth_vmxnet3_dev_init. Other wise it might fail
712  * It returns 0 on success.
713  */
714 static int
715 vmxnet3_dev_start(struct rte_eth_dev *dev)
716 {
717 	int ret;
718 	struct vmxnet3_hw *hw = dev->data->dev_private;
719 
720 	PMD_INIT_FUNC_TRACE();
721 
722 	/* Save stats before it is reset by CMD_ACTIVATE */
723 	vmxnet3_hw_stats_save(hw);
724 
725 	ret = vmxnet3_setup_driver_shared(dev);
726 	if (ret != VMXNET3_SUCCESS)
727 		return ret;
728 
729 	/* check if lsc interrupt feature is enabled */
730 	if (dev->data->dev_conf.intr_conf.lsc) {
731 		struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
732 
733 		/* Setup interrupt callback  */
734 		rte_intr_callback_register(&pci_dev->intr_handle,
735 					   vmxnet3_interrupt_handler, dev);
736 
737 		if (rte_intr_enable(&pci_dev->intr_handle) < 0) {
738 			PMD_INIT_LOG(ERR, "interrupt enable failed");
739 			return -EIO;
740 		}
741 	}
742 
743 	/* Exchange shared data with device */
744 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL,
745 			       VMXNET3_GET_ADDR_LO(hw->sharedPA));
746 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAH,
747 			       VMXNET3_GET_ADDR_HI(hw->sharedPA));
748 
749 	/* Activate device by register write */
750 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_ACTIVATE_DEV);
751 	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
752 
753 	if (ret != 0) {
754 		PMD_INIT_LOG(ERR, "Device activation: UNSUCCESSFUL");
755 		return -EINVAL;
756 	}
757 
758 	/* Setup memory region for rx buffers */
759 	ret = vmxnet3_dev_setup_memreg(dev);
760 	if (ret == 0) {
761 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
762 				       VMXNET3_CMD_REGISTER_MEMREGS);
763 		ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
764 		if (ret != 0)
765 			PMD_INIT_LOG(DEBUG,
766 				     "Failed in setup memory region cmd\n");
767 		ret = 0;
768 	} else {
769 		PMD_INIT_LOG(DEBUG, "Failed to setup memory region\n");
770 	}
771 
772 	if (VMXNET3_VERSION_GE_4(hw)) {
773 		/* Check for additional RSS  */
774 		ret = vmxnet3_v4_rss_configure(dev);
775 		if (ret != VMXNET3_SUCCESS) {
776 			PMD_INIT_LOG(ERR, "Failed to configure v4 RSS");
777 			return ret;
778 		}
779 	}
780 
781 	/* Disable interrupts */
782 	vmxnet3_disable_intr(hw);
783 
784 	/*
785 	 * Load RX queues with blank mbufs and update next2fill index for device
786 	 * Update RxMode of the device
787 	 */
788 	ret = vmxnet3_dev_rxtx_init(dev);
789 	if (ret != VMXNET3_SUCCESS) {
790 		PMD_INIT_LOG(ERR, "Device queue init: UNSUCCESSFUL");
791 		return ret;
792 	}
793 
794 	hw->adapter_stopped = FALSE;
795 
796 	/* Setting proper Rx Mode and issue Rx Mode Update command */
797 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_UCAST | VMXNET3_RXM_BCAST, 1);
798 
799 	if (dev->data->dev_conf.intr_conf.lsc) {
800 		vmxnet3_enable_intr(hw);
801 
802 		/*
803 		 * Update link state from device since this won't be
804 		 * done upon starting with lsc in use. This is done
805 		 * only after enabling interrupts to avoid any race
806 		 * where the link state could change without an
807 		 * interrupt being fired.
808 		 */
809 		__vmxnet3_dev_link_update(dev, 0);
810 	}
811 
812 	return VMXNET3_SUCCESS;
813 }
814 
815 /*
816  * Stop device: disable rx and tx functions to allow for reconfiguring.
817  */
818 static void
819 vmxnet3_dev_stop(struct rte_eth_dev *dev)
820 {
821 	struct rte_eth_link link;
822 	struct vmxnet3_hw *hw = dev->data->dev_private;
823 
824 	PMD_INIT_FUNC_TRACE();
825 
826 	if (hw->adapter_stopped == 1) {
827 		PMD_INIT_LOG(DEBUG, "Device already stopped.");
828 		return;
829 	}
830 
831 	/* disable interrupts */
832 	vmxnet3_disable_intr(hw);
833 
834 	if (dev->data->dev_conf.intr_conf.lsc) {
835 		struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
836 
837 		rte_intr_disable(&pci_dev->intr_handle);
838 
839 		rte_intr_callback_unregister(&pci_dev->intr_handle,
840 					     vmxnet3_interrupt_handler, dev);
841 	}
842 
843 	/* quiesce the device first */
844 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
845 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL, 0);
846 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAH, 0);
847 
848 	/* reset the device */
849 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
850 	PMD_INIT_LOG(DEBUG, "Device reset.");
851 
852 	vmxnet3_dev_clear_queues(dev);
853 
854 	/* Clear recorded link status */
855 	memset(&link, 0, sizeof(link));
856 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
857 	link.link_speed = ETH_SPEED_NUM_10G;
858 	link.link_autoneg = ETH_LINK_FIXED;
859 	rte_eth_linkstatus_set(dev, &link);
860 
861 	hw->adapter_stopped = 1;
862 }
863 
864 static void
865 vmxnet3_free_queues(struct rte_eth_dev *dev)
866 {
867 	int i;
868 
869 	PMD_INIT_FUNC_TRACE();
870 
871 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
872 		void *rxq = dev->data->rx_queues[i];
873 
874 		vmxnet3_dev_rx_queue_release(rxq);
875 	}
876 	dev->data->nb_rx_queues = 0;
877 
878 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
879 		void *txq = dev->data->tx_queues[i];
880 
881 		vmxnet3_dev_tx_queue_release(txq);
882 	}
883 	dev->data->nb_tx_queues = 0;
884 }
885 
886 /*
887  * Reset and stop device.
888  */
889 static void
890 vmxnet3_dev_close(struct rte_eth_dev *dev)
891 {
892 	PMD_INIT_FUNC_TRACE();
893 
894 	vmxnet3_dev_stop(dev);
895 	vmxnet3_free_queues(dev);
896 }
897 
898 static void
899 vmxnet3_hw_tx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
900 			struct UPT1_TxStats *res)
901 {
902 #define VMXNET3_UPDATE_TX_STAT(h, i, f, r)		\
903 		((r)->f = (h)->tqd_start[(i)].stats.f +	\
904 			(h)->saved_tx_stats[(i)].f)
905 
906 	VMXNET3_UPDATE_TX_STAT(hw, q, ucastPktsTxOK, res);
907 	VMXNET3_UPDATE_TX_STAT(hw, q, mcastPktsTxOK, res);
908 	VMXNET3_UPDATE_TX_STAT(hw, q, bcastPktsTxOK, res);
909 	VMXNET3_UPDATE_TX_STAT(hw, q, ucastBytesTxOK, res);
910 	VMXNET3_UPDATE_TX_STAT(hw, q, mcastBytesTxOK, res);
911 	VMXNET3_UPDATE_TX_STAT(hw, q, bcastBytesTxOK, res);
912 	VMXNET3_UPDATE_TX_STAT(hw, q, pktsTxError, res);
913 	VMXNET3_UPDATE_TX_STAT(hw, q, pktsTxDiscard, res);
914 
915 #undef VMXNET3_UPDATE_TX_STAT
916 }
917 
918 static void
919 vmxnet3_hw_rx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
920 			struct UPT1_RxStats *res)
921 {
922 #define VMXNET3_UPDATE_RX_STAT(h, i, f, r)		\
923 		((r)->f = (h)->rqd_start[(i)].stats.f +	\
924 			(h)->saved_rx_stats[(i)].f)
925 
926 	VMXNET3_UPDATE_RX_STAT(hw, q, ucastPktsRxOK, res);
927 	VMXNET3_UPDATE_RX_STAT(hw, q, mcastPktsRxOK, res);
928 	VMXNET3_UPDATE_RX_STAT(hw, q, bcastPktsRxOK, res);
929 	VMXNET3_UPDATE_RX_STAT(hw, q, ucastBytesRxOK, res);
930 	VMXNET3_UPDATE_RX_STAT(hw, q, mcastBytesRxOK, res);
931 	VMXNET3_UPDATE_RX_STAT(hw, q, bcastBytesRxOK, res);
932 	VMXNET3_UPDATE_RX_STAT(hw, q, pktsRxError, res);
933 	VMXNET3_UPDATE_RX_STAT(hw, q, pktsRxOutOfBuf, res);
934 
935 #undef VMXNET3_UPDATE_RX_STAT
936 }
937 
938 static void
939 vmxnet3_tx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
940 					struct UPT1_TxStats *res)
941 {
942 		vmxnet3_hw_tx_stats_get(hw, q, res);
943 
944 #define VMXNET3_REDUCE_SNAPSHOT_TX_STAT(h, i, f, r)	\
945 		((r)->f -= (h)->snapshot_tx_stats[(i)].f)
946 
947 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, ucastPktsTxOK, res);
948 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, mcastPktsTxOK, res);
949 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, bcastPktsTxOK, res);
950 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, ucastBytesTxOK, res);
951 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, mcastBytesTxOK, res);
952 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, bcastBytesTxOK, res);
953 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, pktsTxError, res);
954 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, pktsTxDiscard, res);
955 
956 #undef VMXNET3_REDUCE_SNAPSHOT_TX_STAT
957 }
958 
959 static void
960 vmxnet3_rx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
961 					struct UPT1_RxStats *res)
962 {
963 		vmxnet3_hw_rx_stats_get(hw, q, res);
964 
965 #define VMXNET3_REDUCE_SNAPSHOT_RX_STAT(h, i, f, r)	\
966 		((r)->f -= (h)->snapshot_rx_stats[(i)].f)
967 
968 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, ucastPktsRxOK, res);
969 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, mcastPktsRxOK, res);
970 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, bcastPktsRxOK, res);
971 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, ucastBytesRxOK, res);
972 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, mcastBytesRxOK, res);
973 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, bcastBytesRxOK, res);
974 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, pktsRxError, res);
975 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, pktsRxOutOfBuf, res);
976 
977 #undef VMXNET3_REDUCE_SNAPSHOT_RX_STAT
978 }
979 
980 static void
981 vmxnet3_hw_stats_save(struct vmxnet3_hw *hw)
982 {
983 	unsigned int i;
984 
985 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
986 
987 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
988 
989 	for (i = 0; i < hw->num_tx_queues; i++)
990 		vmxnet3_hw_tx_stats_get(hw, i, &hw->saved_tx_stats[i]);
991 	for (i = 0; i < hw->num_rx_queues; i++)
992 		vmxnet3_hw_rx_stats_get(hw, i, &hw->saved_rx_stats[i]);
993 }
994 
995 static int
996 vmxnet3_dev_xstats_get_names(struct rte_eth_dev *dev,
997 			     struct rte_eth_xstat_name *xstats_names,
998 			     unsigned int n)
999 {
1000 	unsigned int i, t, count = 0;
1001 	unsigned int nstats =
1002 		dev->data->nb_tx_queues * RTE_DIM(vmxnet3_txq_stat_strings) +
1003 		dev->data->nb_rx_queues * RTE_DIM(vmxnet3_rxq_stat_strings);
1004 
1005 	if (!xstats_names || n < nstats)
1006 		return nstats;
1007 
1008 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1009 		if (!dev->data->rx_queues[i])
1010 			continue;
1011 
1012 		for (t = 0; t < RTE_DIM(vmxnet3_rxq_stat_strings); t++) {
1013 			snprintf(xstats_names[count].name,
1014 				 sizeof(xstats_names[count].name),
1015 				 "rx_q%u_%s", i,
1016 				 vmxnet3_rxq_stat_strings[t].name);
1017 			count++;
1018 		}
1019 	}
1020 
1021 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1022 		if (!dev->data->tx_queues[i])
1023 			continue;
1024 
1025 		for (t = 0; t < RTE_DIM(vmxnet3_txq_stat_strings); t++) {
1026 			snprintf(xstats_names[count].name,
1027 				 sizeof(xstats_names[count].name),
1028 				 "tx_q%u_%s", i,
1029 				 vmxnet3_txq_stat_strings[t].name);
1030 			count++;
1031 		}
1032 	}
1033 
1034 	return count;
1035 }
1036 
1037 static int
1038 vmxnet3_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1039 		       unsigned int n)
1040 {
1041 	unsigned int i, t, count = 0;
1042 	unsigned int nstats =
1043 		dev->data->nb_tx_queues * RTE_DIM(vmxnet3_txq_stat_strings) +
1044 		dev->data->nb_rx_queues * RTE_DIM(vmxnet3_rxq_stat_strings);
1045 
1046 	if (n < nstats)
1047 		return nstats;
1048 
1049 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1050 		struct vmxnet3_rx_queue *rxq = dev->data->rx_queues[i];
1051 
1052 		if (rxq == NULL)
1053 			continue;
1054 
1055 		for (t = 0; t < RTE_DIM(vmxnet3_rxq_stat_strings); t++) {
1056 			xstats[count].value = *(uint64_t *)(((char *)&rxq->stats) +
1057 				vmxnet3_rxq_stat_strings[t].offset);
1058 			xstats[count].id = count;
1059 			count++;
1060 		}
1061 	}
1062 
1063 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1064 		struct vmxnet3_tx_queue *txq = dev->data->tx_queues[i];
1065 
1066 		if (txq == NULL)
1067 			continue;
1068 
1069 		for (t = 0; t < RTE_DIM(vmxnet3_txq_stat_strings); t++) {
1070 			xstats[count].value = *(uint64_t *)(((char *)&txq->stats) +
1071 				vmxnet3_txq_stat_strings[t].offset);
1072 			xstats[count].id = count;
1073 			count++;
1074 		}
1075 	}
1076 
1077 	return count;
1078 }
1079 
1080 static int
1081 vmxnet3_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1082 {
1083 	unsigned int i;
1084 	struct vmxnet3_hw *hw = dev->data->dev_private;
1085 	struct UPT1_TxStats txStats;
1086 	struct UPT1_RxStats rxStats;
1087 
1088 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
1089 
1090 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
1091 	for (i = 0; i < hw->num_tx_queues; i++) {
1092 		vmxnet3_tx_stats_get(hw, i, &txStats);
1093 
1094 		stats->q_opackets[i] = txStats.ucastPktsTxOK +
1095 			txStats.mcastPktsTxOK +
1096 			txStats.bcastPktsTxOK;
1097 
1098 		stats->q_obytes[i] = txStats.ucastBytesTxOK +
1099 			txStats.mcastBytesTxOK +
1100 			txStats.bcastBytesTxOK;
1101 
1102 		stats->opackets += stats->q_opackets[i];
1103 		stats->obytes += stats->q_obytes[i];
1104 		stats->oerrors += txStats.pktsTxError + txStats.pktsTxDiscard;
1105 	}
1106 
1107 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_RX_QUEUES);
1108 	for (i = 0; i < hw->num_rx_queues; i++) {
1109 		vmxnet3_rx_stats_get(hw, i, &rxStats);
1110 
1111 		stats->q_ipackets[i] = rxStats.ucastPktsRxOK +
1112 			rxStats.mcastPktsRxOK +
1113 			rxStats.bcastPktsRxOK;
1114 
1115 		stats->q_ibytes[i] = rxStats.ucastBytesRxOK +
1116 			rxStats.mcastBytesRxOK +
1117 			rxStats.bcastBytesRxOK;
1118 
1119 		stats->ipackets += stats->q_ipackets[i];
1120 		stats->ibytes += stats->q_ibytes[i];
1121 
1122 		stats->q_errors[i] = rxStats.pktsRxError;
1123 		stats->ierrors += rxStats.pktsRxError;
1124 		stats->imissed += rxStats.pktsRxOutOfBuf;
1125 	}
1126 
1127 	return 0;
1128 }
1129 
1130 static void
1131 vmxnet3_dev_stats_reset(struct rte_eth_dev *dev)
1132 {
1133 	unsigned int i;
1134 	struct vmxnet3_hw *hw = dev->data->dev_private;
1135 	struct UPT1_TxStats txStats = {0};
1136 	struct UPT1_RxStats rxStats = {0};
1137 
1138 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
1139 
1140 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
1141 
1142 	for (i = 0; i < hw->num_tx_queues; i++) {
1143 		vmxnet3_hw_tx_stats_get(hw, i, &txStats);
1144 		memcpy(&hw->snapshot_tx_stats[i], &txStats,
1145 			sizeof(hw->snapshot_tx_stats[0]));
1146 	}
1147 	for (i = 0; i < hw->num_rx_queues; i++) {
1148 		vmxnet3_hw_rx_stats_get(hw, i, &rxStats);
1149 		memcpy(&hw->snapshot_rx_stats[i], &rxStats,
1150 			sizeof(hw->snapshot_rx_stats[0]));
1151 	}
1152 }
1153 
1154 static void
1155 vmxnet3_dev_info_get(struct rte_eth_dev *dev __rte_unused,
1156 		     struct rte_eth_dev_info *dev_info)
1157 {
1158 	struct vmxnet3_hw *hw = dev->data->dev_private;
1159 
1160 	dev_info->max_rx_queues = VMXNET3_MAX_RX_QUEUES;
1161 	dev_info->max_tx_queues = VMXNET3_MAX_TX_QUEUES;
1162 	dev_info->min_rx_bufsize = 1518 + RTE_PKTMBUF_HEADROOM;
1163 	dev_info->max_rx_pktlen = 16384; /* includes CRC, cf MAXFRS register */
1164 	dev_info->speed_capa = ETH_LINK_SPEED_10G;
1165 	dev_info->max_mac_addrs = VMXNET3_MAX_MAC_ADDRS;
1166 
1167 	dev_info->flow_type_rss_offloads = VMXNET3_RSS_OFFLOAD_ALL;
1168 
1169 	if (VMXNET3_VERSION_GE_4(hw)) {
1170 		dev_info->flow_type_rss_offloads |= VMXNET3_V4_RSS_MASK;
1171 	}
1172 
1173 	dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
1174 		.nb_max = VMXNET3_RX_RING_MAX_SIZE,
1175 		.nb_min = VMXNET3_DEF_RX_RING_SIZE,
1176 		.nb_align = 1,
1177 	};
1178 
1179 	dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
1180 		.nb_max = VMXNET3_TX_RING_MAX_SIZE,
1181 		.nb_min = VMXNET3_DEF_TX_RING_SIZE,
1182 		.nb_align = 1,
1183 		.nb_seg_max = VMXNET3_TX_MAX_SEG,
1184 		.nb_mtu_seg_max = VMXNET3_MAX_TXD_PER_PKT,
1185 	};
1186 
1187 	dev_info->rx_offload_capa = VMXNET3_RX_OFFLOAD_CAP;
1188 	dev_info->rx_queue_offload_capa = 0;
1189 	dev_info->tx_offload_capa = VMXNET3_TX_OFFLOAD_CAP;
1190 	dev_info->tx_queue_offload_capa = 0;
1191 }
1192 
1193 static const uint32_t *
1194 vmxnet3_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1195 {
1196 	static const uint32_t ptypes[] = {
1197 		RTE_PTYPE_L3_IPV4_EXT,
1198 		RTE_PTYPE_L3_IPV4,
1199 		RTE_PTYPE_UNKNOWN
1200 	};
1201 
1202 	if (dev->rx_pkt_burst == vmxnet3_recv_pkts)
1203 		return ptypes;
1204 	return NULL;
1205 }
1206 
1207 static int
1208 vmxnet3_mac_addr_set(struct rte_eth_dev *dev, struct rte_ether_addr *mac_addr)
1209 {
1210 	struct vmxnet3_hw *hw = dev->data->dev_private;
1211 
1212 	rte_ether_addr_copy(mac_addr, (struct rte_ether_addr *)(hw->perm_addr));
1213 	vmxnet3_write_mac(hw, mac_addr->addr_bytes);
1214 	return 0;
1215 }
1216 
1217 /* return 0 means link status changed, -1 means not changed */
1218 static int
1219 __vmxnet3_dev_link_update(struct rte_eth_dev *dev,
1220 			  __rte_unused int wait_to_complete)
1221 {
1222 	struct vmxnet3_hw *hw = dev->data->dev_private;
1223 	struct rte_eth_link link;
1224 	uint32_t ret;
1225 
1226 	memset(&link, 0, sizeof(link));
1227 
1228 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
1229 	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
1230 
1231 	if (ret & 0x1)
1232 		link.link_status = ETH_LINK_UP;
1233 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1234 	link.link_speed = ETH_SPEED_NUM_10G;
1235 	link.link_autoneg = ETH_LINK_FIXED;
1236 
1237 	return rte_eth_linkstatus_set(dev, &link);
1238 }
1239 
1240 static int
1241 vmxnet3_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
1242 {
1243 	/* Link status doesn't change for stopped dev */
1244 	if (dev->data->dev_started == 0)
1245 		return -1;
1246 
1247 	return __vmxnet3_dev_link_update(dev, wait_to_complete);
1248 }
1249 
1250 /* Updating rxmode through Vmxnet3_DriverShared structure in adapter */
1251 static void
1252 vmxnet3_dev_set_rxmode(struct vmxnet3_hw *hw, uint32_t feature, int set)
1253 {
1254 	struct Vmxnet3_RxFilterConf *rxConf = &hw->shared->devRead.rxFilterConf;
1255 
1256 	if (set)
1257 		rxConf->rxMode = rxConf->rxMode | feature;
1258 	else
1259 		rxConf->rxMode = rxConf->rxMode & (~feature);
1260 
1261 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_UPDATE_RX_MODE);
1262 }
1263 
1264 /* Promiscuous supported only if Vmxnet3_DriverShared is initialized in adapter */
1265 static void
1266 vmxnet3_dev_promiscuous_enable(struct rte_eth_dev *dev)
1267 {
1268 	struct vmxnet3_hw *hw = dev->data->dev_private;
1269 	uint32_t *vf_table = hw->shared->devRead.rxFilterConf.vfTable;
1270 
1271 	memset(vf_table, 0, VMXNET3_VFT_TABLE_SIZE);
1272 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 1);
1273 
1274 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1275 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1276 }
1277 
1278 /* Promiscuous supported only if Vmxnet3_DriverShared is initialized in adapter */
1279 static void
1280 vmxnet3_dev_promiscuous_disable(struct rte_eth_dev *dev)
1281 {
1282 	struct vmxnet3_hw *hw = dev->data->dev_private;
1283 	uint32_t *vf_table = hw->shared->devRead.rxFilterConf.vfTable;
1284 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1285 
1286 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1287 		memcpy(vf_table, hw->shadow_vfta, VMXNET3_VFT_TABLE_SIZE);
1288 	else
1289 		memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);
1290 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 0);
1291 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1292 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1293 }
1294 
1295 /* Allmulticast supported only if Vmxnet3_DriverShared is initialized in adapter */
1296 static void
1297 vmxnet3_dev_allmulticast_enable(struct rte_eth_dev *dev)
1298 {
1299 	struct vmxnet3_hw *hw = dev->data->dev_private;
1300 
1301 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_ALL_MULTI, 1);
1302 }
1303 
1304 /* Allmulticast supported only if Vmxnet3_DriverShared is initialized in adapter */
1305 static void
1306 vmxnet3_dev_allmulticast_disable(struct rte_eth_dev *dev)
1307 {
1308 	struct vmxnet3_hw *hw = dev->data->dev_private;
1309 
1310 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_ALL_MULTI, 0);
1311 }
1312 
1313 /* Enable/disable filter on vlan */
1314 static int
1315 vmxnet3_dev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vid, int on)
1316 {
1317 	struct vmxnet3_hw *hw = dev->data->dev_private;
1318 	struct Vmxnet3_RxFilterConf *rxConf = &hw->shared->devRead.rxFilterConf;
1319 	uint32_t *vf_table = rxConf->vfTable;
1320 
1321 	/* save state for restore */
1322 	if (on)
1323 		VMXNET3_SET_VFTABLE_ENTRY(hw->shadow_vfta, vid);
1324 	else
1325 		VMXNET3_CLEAR_VFTABLE_ENTRY(hw->shadow_vfta, vid);
1326 
1327 	/* don't change active filter if in promiscuous mode */
1328 	if (rxConf->rxMode & VMXNET3_RXM_PROMISC)
1329 		return 0;
1330 
1331 	/* set in hardware */
1332 	if (on)
1333 		VMXNET3_SET_VFTABLE_ENTRY(vf_table, vid);
1334 	else
1335 		VMXNET3_CLEAR_VFTABLE_ENTRY(vf_table, vid);
1336 
1337 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1338 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1339 	return 0;
1340 }
1341 
1342 static int
1343 vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1344 {
1345 	struct vmxnet3_hw *hw = dev->data->dev_private;
1346 	Vmxnet3_DSDevRead *devRead = &hw->shared->devRead;
1347 	uint32_t *vf_table = devRead->rxFilterConf.vfTable;
1348 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1349 
1350 	if (mask & ETH_VLAN_STRIP_MASK) {
1351 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1352 			devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1353 		else
1354 			devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;
1355 
1356 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1357 				       VMXNET3_CMD_UPDATE_FEATURE);
1358 	}
1359 
1360 	if (mask & ETH_VLAN_FILTER_MASK) {
1361 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1362 			memcpy(vf_table, hw->shadow_vfta, VMXNET3_VFT_TABLE_SIZE);
1363 		else
1364 			memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);
1365 
1366 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1367 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1368 	}
1369 
1370 	return 0;
1371 }
1372 
1373 static void
1374 vmxnet3_process_events(struct rte_eth_dev *dev)
1375 {
1376 	struct vmxnet3_hw *hw = dev->data->dev_private;
1377 	uint32_t events = hw->shared->ecr;
1378 
1379 	if (!events)
1380 		return;
1381 
1382 	/*
1383 	 * ECR bits when written with 1b are cleared. Hence write
1384 	 * events back to ECR so that the bits which were set will be reset.
1385 	 */
1386 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_ECR, events);
1387 
1388 	/* Check if link state has changed */
1389 	if (events & VMXNET3_ECR_LINK) {
1390 		PMD_DRV_LOG(DEBUG, "Process events: VMXNET3_ECR_LINK event");
1391 		if (vmxnet3_dev_link_update(dev, 0) == 0)
1392 			_rte_eth_dev_callback_process(dev,
1393 						      RTE_ETH_EVENT_INTR_LSC,
1394 						      NULL);
1395 	}
1396 
1397 	/* Check if there is an error on xmit/recv queues */
1398 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
1399 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1400 				       VMXNET3_CMD_GET_QUEUE_STATUS);
1401 
1402 		if (hw->tqd_start->status.stopped)
1403 			PMD_DRV_LOG(ERR, "tq error 0x%x",
1404 				    hw->tqd_start->status.error);
1405 
1406 		if (hw->rqd_start->status.stopped)
1407 			PMD_DRV_LOG(ERR, "rq error 0x%x",
1408 				     hw->rqd_start->status.error);
1409 
1410 		/* Reset the device */
1411 		/* Have to reset the device */
1412 	}
1413 
1414 	if (events & VMXNET3_ECR_DIC)
1415 		PMD_DRV_LOG(DEBUG, "Device implementation change event.");
1416 
1417 	if (events & VMXNET3_ECR_DEBUG)
1418 		PMD_DRV_LOG(DEBUG, "Debug event generated by device.");
1419 }
1420 
1421 static void
1422 vmxnet3_interrupt_handler(void *param)
1423 {
1424 	struct rte_eth_dev *dev = param;
1425 	struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
1426 
1427 	vmxnet3_process_events(dev);
1428 
1429 	if (rte_intr_enable(&pci_dev->intr_handle) < 0)
1430 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1431 }
1432 
1433 RTE_PMD_REGISTER_PCI(net_vmxnet3, rte_vmxnet3_pmd);
1434 RTE_PMD_REGISTER_PCI_TABLE(net_vmxnet3, pci_id_vmxnet3_map);
1435 RTE_PMD_REGISTER_KMOD_DEP(net_vmxnet3, "* igb_uio | uio_pci_generic | vfio-pci");
1436 
1437 RTE_INIT(vmxnet3_init_log)
1438 {
1439 	vmxnet3_logtype_init = rte_log_register("pmd.net.vmxnet3.init");
1440 	if (vmxnet3_logtype_init >= 0)
1441 		rte_log_set_level(vmxnet3_logtype_init, RTE_LOG_NOTICE);
1442 	vmxnet3_logtype_driver = rte_log_register("pmd.net.vmxnet3.driver");
1443 	if (vmxnet3_logtype_driver >= 0)
1444 		rte_log_set_level(vmxnet3_logtype_driver, RTE_LOG_NOTICE);
1445 }
1446