xref: /dpdk/drivers/net/vmxnet3/vmxnet3_ethdev.c (revision b467d382)
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_SCATTER |	\
56 	 DEV_RX_OFFLOAD_IPV4_CKSUM |	\
57 	 DEV_RX_OFFLOAD_UDP_CKSUM |	\
58 	 DEV_RX_OFFLOAD_TCP_CKSUM |	\
59 	 DEV_RX_OFFLOAD_TCP_LRO |	\
60 	 DEV_RX_OFFLOAD_JUMBO_FRAME)
61 
62 static int eth_vmxnet3_dev_init(struct rte_eth_dev *eth_dev);
63 static int eth_vmxnet3_dev_uninit(struct rte_eth_dev *eth_dev);
64 static int vmxnet3_dev_configure(struct rte_eth_dev *dev);
65 static int vmxnet3_dev_start(struct rte_eth_dev *dev);
66 static void vmxnet3_dev_stop(struct rte_eth_dev *dev);
67 static void vmxnet3_dev_close(struct rte_eth_dev *dev);
68 static void vmxnet3_dev_set_rxmode(struct vmxnet3_hw *hw, uint32_t feature, int set);
69 static void vmxnet3_dev_promiscuous_enable(struct rte_eth_dev *dev);
70 static void vmxnet3_dev_promiscuous_disable(struct rte_eth_dev *dev);
71 static void vmxnet3_dev_allmulticast_enable(struct rte_eth_dev *dev);
72 static void vmxnet3_dev_allmulticast_disable(struct rte_eth_dev *dev);
73 static int __vmxnet3_dev_link_update(struct rte_eth_dev *dev,
74 				     int wait_to_complete);
75 static int vmxnet3_dev_link_update(struct rte_eth_dev *dev,
76 				   int wait_to_complete);
77 static void vmxnet3_hw_stats_save(struct vmxnet3_hw *hw);
78 static int vmxnet3_dev_stats_get(struct rte_eth_dev *dev,
79 				  struct rte_eth_stats *stats);
80 static void vmxnet3_dev_stats_reset(struct rte_eth_dev *dev);
81 static int vmxnet3_dev_xstats_get_names(struct rte_eth_dev *dev,
82 					struct rte_eth_xstat_name *xstats,
83 					unsigned int n);
84 static int vmxnet3_dev_xstats_get(struct rte_eth_dev *dev,
85 				  struct rte_eth_xstat *xstats, unsigned int n);
86 static void vmxnet3_dev_info_get(struct rte_eth_dev *dev,
87 				 struct rte_eth_dev_info *dev_info);
88 static const uint32_t *
89 vmxnet3_dev_supported_ptypes_get(struct rte_eth_dev *dev);
90 static int vmxnet3_dev_vlan_filter_set(struct rte_eth_dev *dev,
91 				       uint16_t vid, int on);
92 static int vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask);
93 static int vmxnet3_mac_addr_set(struct rte_eth_dev *dev,
94 				 struct ether_addr *mac_addr);
95 static void vmxnet3_interrupt_handler(void *param);
96 
97 int vmxnet3_logtype_init;
98 int vmxnet3_logtype_driver;
99 
100 /*
101  * The set of PCI devices this driver supports
102  */
103 #define VMWARE_PCI_VENDOR_ID 0x15AD
104 #define VMWARE_DEV_ID_VMXNET3 0x07B0
105 static const struct rte_pci_id pci_id_vmxnet3_map[] = {
106 	{ RTE_PCI_DEVICE(VMWARE_PCI_VENDOR_ID, VMWARE_DEV_ID_VMXNET3) },
107 	{ .vendor_id = 0, /* sentinel */ },
108 };
109 
110 static const struct eth_dev_ops vmxnet3_eth_dev_ops = {
111 	.dev_configure        = vmxnet3_dev_configure,
112 	.dev_start            = vmxnet3_dev_start,
113 	.dev_stop             = vmxnet3_dev_stop,
114 	.dev_close            = vmxnet3_dev_close,
115 	.promiscuous_enable   = vmxnet3_dev_promiscuous_enable,
116 	.promiscuous_disable  = vmxnet3_dev_promiscuous_disable,
117 	.allmulticast_enable  = vmxnet3_dev_allmulticast_enable,
118 	.allmulticast_disable = vmxnet3_dev_allmulticast_disable,
119 	.link_update          = vmxnet3_dev_link_update,
120 	.stats_get            = vmxnet3_dev_stats_get,
121 	.xstats_get_names     = vmxnet3_dev_xstats_get_names,
122 	.xstats_get           = vmxnet3_dev_xstats_get,
123 	.stats_reset          = vmxnet3_dev_stats_reset,
124 	.mac_addr_set         = vmxnet3_mac_addr_set,
125 	.dev_infos_get        = vmxnet3_dev_info_get,
126 	.dev_supported_ptypes_get = vmxnet3_dev_supported_ptypes_get,
127 	.vlan_filter_set      = vmxnet3_dev_vlan_filter_set,
128 	.vlan_offload_set     = vmxnet3_dev_vlan_offload_set,
129 	.rx_queue_setup       = vmxnet3_dev_rx_queue_setup,
130 	.rx_queue_release     = vmxnet3_dev_rx_queue_release,
131 	.tx_queue_setup       = vmxnet3_dev_tx_queue_setup,
132 	.tx_queue_release     = vmxnet3_dev_tx_queue_release,
133 };
134 
135 struct vmxnet3_xstats_name_off {
136 	char name[RTE_ETH_XSTATS_NAME_SIZE];
137 	unsigned int offset;
138 };
139 
140 /* tx_qX_ is prepended to the name string here */
141 static const struct vmxnet3_xstats_name_off vmxnet3_txq_stat_strings[] = {
142 	{"drop_total",         offsetof(struct vmxnet3_txq_stats, drop_total)},
143 	{"drop_too_many_segs", offsetof(struct vmxnet3_txq_stats, drop_too_many_segs)},
144 	{"drop_tso",           offsetof(struct vmxnet3_txq_stats, drop_tso)},
145 	{"tx_ring_full",       offsetof(struct vmxnet3_txq_stats, tx_ring_full)},
146 };
147 
148 /* rx_qX_ is prepended to the name string here */
149 static const struct vmxnet3_xstats_name_off vmxnet3_rxq_stat_strings[] = {
150 	{"drop_total",           offsetof(struct vmxnet3_rxq_stats, drop_total)},
151 	{"drop_err",             offsetof(struct vmxnet3_rxq_stats, drop_err)},
152 	{"drop_fcs",             offsetof(struct vmxnet3_rxq_stats, drop_fcs)},
153 	{"rx_buf_alloc_failure", offsetof(struct vmxnet3_rxq_stats, rx_buf_alloc_failure)},
154 };
155 
156 static const struct rte_memzone *
157 gpa_zone_reserve(struct rte_eth_dev *dev, uint32_t size,
158 		 const char *post_string, int socket_id,
159 		 uint16_t align, bool reuse)
160 {
161 	char z_name[RTE_MEMZONE_NAMESIZE];
162 	const struct rte_memzone *mz;
163 
164 	snprintf(z_name, sizeof(z_name), "eth_p%d_%s",
165 			dev->data->port_id, post_string);
166 
167 	mz = rte_memzone_lookup(z_name);
168 	if (!reuse) {
169 		if (mz)
170 			rte_memzone_free(mz);
171 		return rte_memzone_reserve_aligned(z_name, size, socket_id,
172 				RTE_MEMZONE_IOVA_CONTIG, align);
173 	}
174 
175 	if (mz)
176 		return mz;
177 
178 	return rte_memzone_reserve_aligned(z_name, size, socket_id,
179 			RTE_MEMZONE_IOVA_CONTIG, align);
180 }
181 
182 /*
183  * This function is based on vmxnet3_disable_intr()
184  */
185 static void
186 vmxnet3_disable_intr(struct vmxnet3_hw *hw)
187 {
188 	int i;
189 
190 	PMD_INIT_FUNC_TRACE();
191 
192 	hw->shared->devRead.intrConf.intrCtrl |= VMXNET3_IC_DISABLE_ALL;
193 	for (i = 0; i < hw->num_intrs; i++)
194 		VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + i * 8, 1);
195 }
196 
197 static void
198 vmxnet3_enable_intr(struct vmxnet3_hw *hw)
199 {
200 	int i;
201 
202 	PMD_INIT_FUNC_TRACE();
203 
204 	hw->shared->devRead.intrConf.intrCtrl &= ~VMXNET3_IC_DISABLE_ALL;
205 	for (i = 0; i < hw->num_intrs; i++)
206 		VMXNET3_WRITE_BAR0_REG(hw, VMXNET3_REG_IMR + i * 8, 0);
207 }
208 
209 /*
210  * Gets tx data ring descriptor size.
211  */
212 static uint16_t
213 eth_vmxnet3_txdata_get(struct vmxnet3_hw *hw)
214 {
215 	uint16 txdata_desc_size;
216 
217 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
218 			       VMXNET3_CMD_GET_TXDATA_DESC_SIZE);
219 	txdata_desc_size = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
220 
221 	return (txdata_desc_size < VMXNET3_TXDATA_DESC_MIN_SIZE ||
222 		txdata_desc_size > VMXNET3_TXDATA_DESC_MAX_SIZE ||
223 		txdata_desc_size & VMXNET3_TXDATA_DESC_SIZE_MASK) ?
224 		sizeof(struct Vmxnet3_TxDataDesc) : txdata_desc_size;
225 }
226 
227 /*
228  * It returns 0 on success.
229  */
230 static int
231 eth_vmxnet3_dev_init(struct rte_eth_dev *eth_dev)
232 {
233 	struct rte_pci_device *pci_dev;
234 	struct vmxnet3_hw *hw = eth_dev->data->dev_private;
235 	uint32_t mac_hi, mac_lo, ver;
236 	struct rte_eth_link link;
237 
238 	PMD_INIT_FUNC_TRACE();
239 
240 	eth_dev->dev_ops = &vmxnet3_eth_dev_ops;
241 	eth_dev->rx_pkt_burst = &vmxnet3_recv_pkts;
242 	eth_dev->tx_pkt_burst = &vmxnet3_xmit_pkts;
243 	eth_dev->tx_pkt_prepare = vmxnet3_prep_pkts;
244 	pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
245 
246 	/*
247 	 * for secondary processes, we don't initialize any further as primary
248 	 * has already done this work.
249 	 */
250 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
251 		return 0;
252 
253 	rte_eth_copy_pci_info(eth_dev, pci_dev);
254 
255 	/* Vendor and Device ID need to be set before init of shared code */
256 	hw->device_id = pci_dev->id.device_id;
257 	hw->vendor_id = pci_dev->id.vendor_id;
258 	hw->hw_addr0 = (void *)pci_dev->mem_resource[0].addr;
259 	hw->hw_addr1 = (void *)pci_dev->mem_resource[1].addr;
260 
261 	hw->num_rx_queues = 1;
262 	hw->num_tx_queues = 1;
263 	hw->bufs_per_pkt = 1;
264 
265 	/* Check h/w version compatibility with driver. */
266 	ver = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_VRRS);
267 	PMD_INIT_LOG(DEBUG, "Hardware version : %d", ver);
268 
269 	if (ver & (1 << VMXNET3_REV_3)) {
270 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
271 				       1 << VMXNET3_REV_3);
272 		hw->version = VMXNET3_REV_3 + 1;
273 	} else if (ver & (1 << VMXNET3_REV_2)) {
274 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
275 				       1 << VMXNET3_REV_2);
276 		hw->version = VMXNET3_REV_2 + 1;
277 	} else if (ver & (1 << VMXNET3_REV_1)) {
278 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
279 				       1 << VMXNET3_REV_1);
280 		hw->version = VMXNET3_REV_1 + 1;
281 	} else {
282 		PMD_INIT_LOG(ERR, "Incompatible hardware version: %d", ver);
283 		return -EIO;
284 	}
285 
286 	PMD_INIT_LOG(DEBUG, "Using device version %d\n", hw->version);
287 
288 	/* Check UPT version compatibility with driver. */
289 	ver = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_UVRS);
290 	PMD_INIT_LOG(DEBUG, "UPT hardware version : %d", ver);
291 	if (ver & 0x1)
292 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_UVRS, 1);
293 	else {
294 		PMD_INIT_LOG(ERR, "Incompatible UPT version.");
295 		return -EIO;
296 	}
297 
298 	/* Getting MAC Address */
299 	mac_lo = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_MACL);
300 	mac_hi = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_MACH);
301 	memcpy(hw->perm_addr, &mac_lo, 4);
302 	memcpy(hw->perm_addr + 4, &mac_hi, 2);
303 
304 	/* Allocate memory for storing MAC addresses */
305 	eth_dev->data->mac_addrs = rte_zmalloc("vmxnet3", ETHER_ADDR_LEN *
306 					       VMXNET3_MAX_MAC_ADDRS, 0);
307 	if (eth_dev->data->mac_addrs == NULL) {
308 		PMD_INIT_LOG(ERR,
309 			     "Failed to allocate %d bytes needed to store MAC addresses",
310 			     ETHER_ADDR_LEN * VMXNET3_MAX_MAC_ADDRS);
311 		return -ENOMEM;
312 	}
313 	/* Copy the permanent MAC address */
314 	ether_addr_copy((struct ether_addr *) hw->perm_addr,
315 			&eth_dev->data->mac_addrs[0]);
316 
317 	PMD_INIT_LOG(DEBUG, "MAC Address : %02x:%02x:%02x:%02x:%02x:%02x",
318 		     hw->perm_addr[0], hw->perm_addr[1], hw->perm_addr[2],
319 		     hw->perm_addr[3], hw->perm_addr[4], hw->perm_addr[5]);
320 
321 	/* Put device in Quiesce Mode */
322 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
323 
324 	/* allow untagged pkts */
325 	VMXNET3_SET_VFTABLE_ENTRY(hw->shadow_vfta, 0);
326 
327 	hw->txdata_desc_size = VMXNET3_VERSION_GE_3(hw) ?
328 		eth_vmxnet3_txdata_get(hw) : sizeof(struct Vmxnet3_TxDataDesc);
329 
330 	hw->rxdata_desc_size = VMXNET3_VERSION_GE_3(hw) ?
331 		VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
332 	RTE_ASSERT((hw->rxdata_desc_size & ~VMXNET3_RXDATA_DESC_SIZE_MASK) ==
333 		   hw->rxdata_desc_size);
334 
335 	/* clear shadow stats */
336 	memset(hw->saved_tx_stats, 0, sizeof(hw->saved_tx_stats));
337 	memset(hw->saved_rx_stats, 0, sizeof(hw->saved_rx_stats));
338 
339 	/* clear snapshot stats */
340 	memset(hw->snapshot_tx_stats, 0, sizeof(hw->snapshot_tx_stats));
341 	memset(hw->snapshot_rx_stats, 0, sizeof(hw->snapshot_rx_stats));
342 
343 	/* set the initial link status */
344 	memset(&link, 0, sizeof(link));
345 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
346 	link.link_speed = ETH_SPEED_NUM_10G;
347 	link.link_autoneg = ETH_LINK_FIXED;
348 	rte_eth_linkstatus_set(eth_dev, &link);
349 
350 	return 0;
351 }
352 
353 static int
354 eth_vmxnet3_dev_uninit(struct rte_eth_dev *eth_dev)
355 {
356 	struct vmxnet3_hw *hw = eth_dev->data->dev_private;
357 
358 	PMD_INIT_FUNC_TRACE();
359 
360 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
361 		return 0;
362 
363 	if (hw->adapter_stopped == 0) {
364 		PMD_INIT_LOG(DEBUG, "Device has not been closed.");
365 		return -EBUSY;
366 	}
367 
368 	eth_dev->dev_ops = NULL;
369 	eth_dev->rx_pkt_burst = NULL;
370 	eth_dev->tx_pkt_burst = NULL;
371 	eth_dev->tx_pkt_prepare = NULL;
372 
373 	return 0;
374 }
375 
376 static int eth_vmxnet3_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
377 	struct rte_pci_device *pci_dev)
378 {
379 	return rte_eth_dev_pci_generic_probe(pci_dev,
380 		sizeof(struct vmxnet3_hw), eth_vmxnet3_dev_init);
381 }
382 
383 static int eth_vmxnet3_pci_remove(struct rte_pci_device *pci_dev)
384 {
385 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_vmxnet3_dev_uninit);
386 }
387 
388 static struct rte_pci_driver rte_vmxnet3_pmd = {
389 	.id_table = pci_id_vmxnet3_map,
390 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
391 	.probe = eth_vmxnet3_pci_probe,
392 	.remove = eth_vmxnet3_pci_remove,
393 };
394 
395 static int
396 vmxnet3_dev_configure(struct rte_eth_dev *dev)
397 {
398 	const struct rte_memzone *mz;
399 	struct vmxnet3_hw *hw = dev->data->dev_private;
400 	size_t size;
401 
402 	PMD_INIT_FUNC_TRACE();
403 
404 	if (dev->data->nb_tx_queues > VMXNET3_MAX_TX_QUEUES ||
405 	    dev->data->nb_rx_queues > VMXNET3_MAX_RX_QUEUES) {
406 		PMD_INIT_LOG(ERR, "ERROR: Number of queues not supported");
407 		return -EINVAL;
408 	}
409 
410 	if (!rte_is_power_of_2(dev->data->nb_rx_queues)) {
411 		PMD_INIT_LOG(ERR, "ERROR: Number of rx queues not power of 2");
412 		return -EINVAL;
413 	}
414 
415 	size = dev->data->nb_rx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
416 		dev->data->nb_tx_queues * sizeof(struct Vmxnet3_RxQueueDesc);
417 
418 	if (size > UINT16_MAX)
419 		return -EINVAL;
420 
421 	hw->num_rx_queues = (uint8_t)dev->data->nb_rx_queues;
422 	hw->num_tx_queues = (uint8_t)dev->data->nb_tx_queues;
423 
424 	/*
425 	 * Allocate a memzone for Vmxnet3_DriverShared - Vmxnet3_DSDevRead
426 	 * on current socket
427 	 */
428 	mz = gpa_zone_reserve(dev, sizeof(struct Vmxnet3_DriverShared),
429 			      "shared", rte_socket_id(), 8, 1);
430 
431 	if (mz == NULL) {
432 		PMD_INIT_LOG(ERR, "ERROR: Creating shared zone");
433 		return -ENOMEM;
434 	}
435 	memset(mz->addr, 0, mz->len);
436 
437 	hw->shared = mz->addr;
438 	hw->sharedPA = mz->iova;
439 
440 	/*
441 	 * Allocate a memzone for Vmxnet3_RxQueueDesc - Vmxnet3_TxQueueDesc
442 	 * on current socket.
443 	 *
444 	 * We cannot reuse this memzone from previous allocation as its size
445 	 * depends on the number of tx and rx queues, which could be different
446 	 * from one config to another.
447 	 */
448 	mz = gpa_zone_reserve(dev, size, "queuedesc", rte_socket_id(),
449 			      VMXNET3_QUEUE_DESC_ALIGN, 0);
450 	if (mz == NULL) {
451 		PMD_INIT_LOG(ERR, "ERROR: Creating queue descriptors zone");
452 		return -ENOMEM;
453 	}
454 	memset(mz->addr, 0, mz->len);
455 
456 	hw->tqd_start = (Vmxnet3_TxQueueDesc *)mz->addr;
457 	hw->rqd_start = (Vmxnet3_RxQueueDesc *)(hw->tqd_start + hw->num_tx_queues);
458 
459 	hw->queueDescPA = mz->iova;
460 	hw->queue_desc_len = (uint16_t)size;
461 
462 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
463 		/* Allocate memory structure for UPT1_RSSConf and configure */
464 		mz = gpa_zone_reserve(dev, sizeof(struct VMXNET3_RSSConf),
465 				      "rss_conf", rte_socket_id(),
466 				      RTE_CACHE_LINE_SIZE, 1);
467 		if (mz == NULL) {
468 			PMD_INIT_LOG(ERR,
469 				     "ERROR: Creating rss_conf structure zone");
470 			return -ENOMEM;
471 		}
472 		memset(mz->addr, 0, mz->len);
473 
474 		hw->rss_conf = mz->addr;
475 		hw->rss_confPA = mz->iova;
476 	}
477 
478 	return 0;
479 }
480 
481 static void
482 vmxnet3_write_mac(struct vmxnet3_hw *hw, const uint8_t *addr)
483 {
484 	uint32_t val;
485 
486 	PMD_INIT_LOG(DEBUG,
487 		     "Writing MAC Address : %02x:%02x:%02x:%02x:%02x:%02x",
488 		     addr[0], addr[1], addr[2],
489 		     addr[3], addr[4], addr[5]);
490 
491 	memcpy(&val, addr, 4);
492 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_MACL, val);
493 
494 	memcpy(&val, addr + 4, 2);
495 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_MACH, val);
496 }
497 
498 static int
499 vmxnet3_dev_setup_memreg(struct rte_eth_dev *dev)
500 {
501 	struct vmxnet3_hw *hw = dev->data->dev_private;
502 	Vmxnet3_DriverShared *shared = hw->shared;
503 	Vmxnet3_CmdInfo *cmdInfo;
504 	struct rte_mempool *mp[VMXNET3_MAX_RX_QUEUES];
505 	uint8_t index[VMXNET3_MAX_RX_QUEUES + VMXNET3_MAX_TX_QUEUES];
506 	uint32_t num, i, j, size;
507 
508 	if (hw->memRegsPA == 0) {
509 		const struct rte_memzone *mz;
510 
511 		size = sizeof(Vmxnet3_MemRegs) +
512 			(VMXNET3_MAX_RX_QUEUES + VMXNET3_MAX_TX_QUEUES) *
513 			sizeof(Vmxnet3_MemoryRegion);
514 
515 		mz = gpa_zone_reserve(dev, size, "memRegs", rte_socket_id(), 8,
516 				      1);
517 		if (mz == NULL) {
518 			PMD_INIT_LOG(ERR, "ERROR: Creating memRegs zone");
519 			return -ENOMEM;
520 		}
521 		memset(mz->addr, 0, mz->len);
522 		hw->memRegs = mz->addr;
523 		hw->memRegsPA = mz->iova;
524 	}
525 
526 	num = hw->num_rx_queues;
527 
528 	for (i = 0; i < num; i++) {
529 		vmxnet3_rx_queue_t *rxq = dev->data->rx_queues[i];
530 
531 		mp[i] = rxq->mp;
532 		index[i] = 1 << i;
533 	}
534 
535 	/*
536 	 * The same mempool could be used by multiple queues. In such a case,
537 	 * remove duplicate mempool entries. Only one entry is kept with
538 	 * bitmask indicating queues that are using this mempool.
539 	 */
540 	for (i = 1; i < num; i++) {
541 		for (j = 0; j < i; j++) {
542 			if (mp[i] == mp[j]) {
543 				mp[i] = NULL;
544 				index[j] |= 1 << i;
545 				break;
546 			}
547 		}
548 	}
549 
550 	j = 0;
551 	for (i = 0; i < num; i++) {
552 		if (mp[i] == NULL)
553 			continue;
554 
555 		Vmxnet3_MemoryRegion *mr = &hw->memRegs->memRegs[j];
556 
557 		mr->startPA =
558 			(uintptr_t)STAILQ_FIRST(&mp[i]->mem_list)->iova;
559 		mr->length = STAILQ_FIRST(&mp[i]->mem_list)->len <= INT32_MAX ?
560 			STAILQ_FIRST(&mp[i]->mem_list)->len : INT32_MAX;
561 		mr->txQueueBits = index[i];
562 		mr->rxQueueBits = index[i];
563 
564 		PMD_INIT_LOG(INFO,
565 			     "index: %u startPA: %" PRIu64 " length: %u, "
566 			     "rxBits: %x",
567 			     j, mr->startPA, mr->length, mr->rxQueueBits);
568 		j++;
569 	}
570 	hw->memRegs->numRegs = j;
571 	PMD_INIT_LOG(INFO, "numRegs: %u", j);
572 
573 	size = sizeof(Vmxnet3_MemRegs) +
574 		(j - 1) * sizeof(Vmxnet3_MemoryRegion);
575 
576 	cmdInfo = &shared->cu.cmdInfo;
577 	cmdInfo->varConf.confVer = 1;
578 	cmdInfo->varConf.confLen = size;
579 	cmdInfo->varConf.confPA = hw->memRegsPA;
580 
581 	return 0;
582 }
583 
584 static int
585 vmxnet3_setup_driver_shared(struct rte_eth_dev *dev)
586 {
587 	struct rte_eth_conf port_conf = dev->data->dev_conf;
588 	struct vmxnet3_hw *hw = dev->data->dev_private;
589 	uint32_t mtu = dev->data->mtu;
590 	Vmxnet3_DriverShared *shared = hw->shared;
591 	Vmxnet3_DSDevRead *devRead = &shared->devRead;
592 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
593 	uint32_t i;
594 	int ret;
595 
596 	hw->mtu = mtu;
597 
598 	shared->magic = VMXNET3_REV1_MAGIC;
599 	devRead->misc.driverInfo.version = VMXNET3_DRIVER_VERSION_NUM;
600 
601 	/* Setting up Guest OS information */
602 	devRead->misc.driverInfo.gos.gosBits   = sizeof(void *) == 4 ?
603 		VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64;
604 	devRead->misc.driverInfo.gos.gosType   = VMXNET3_GOS_TYPE_LINUX;
605 	devRead->misc.driverInfo.vmxnet3RevSpt = 1;
606 	devRead->misc.driverInfo.uptVerSpt     = 1;
607 
608 	devRead->misc.mtu = rte_le_to_cpu_32(mtu);
609 	devRead->misc.queueDescPA  = hw->queueDescPA;
610 	devRead->misc.queueDescLen = hw->queue_desc_len;
611 	devRead->misc.numTxQueues  = hw->num_tx_queues;
612 	devRead->misc.numRxQueues  = hw->num_rx_queues;
613 
614 	/*
615 	 * Set number of interrupts to 1
616 	 * PMD by default disables all the interrupts but this is MUST
617 	 * to activate device. It needs at least one interrupt for
618 	 * link events to handle
619 	 */
620 	hw->num_intrs = devRead->intrConf.numIntrs = 1;
621 	devRead->intrConf.intrCtrl |= VMXNET3_IC_DISABLE_ALL;
622 
623 	for (i = 0; i < hw->num_tx_queues; i++) {
624 		Vmxnet3_TxQueueDesc *tqd = &hw->tqd_start[i];
625 		vmxnet3_tx_queue_t *txq  = dev->data->tx_queues[i];
626 
627 		txq->shared = &hw->tqd_start[i];
628 
629 		tqd->ctrl.txNumDeferred  = 0;
630 		tqd->ctrl.txThreshold    = 1;
631 		tqd->conf.txRingBasePA   = txq->cmd_ring.basePA;
632 		tqd->conf.compRingBasePA = txq->comp_ring.basePA;
633 		tqd->conf.dataRingBasePA = txq->data_ring.basePA;
634 
635 		tqd->conf.txRingSize   = txq->cmd_ring.size;
636 		tqd->conf.compRingSize = txq->comp_ring.size;
637 		tqd->conf.dataRingSize = txq->data_ring.size;
638 		tqd->conf.txDataRingDescSize = txq->txdata_desc_size;
639 		tqd->conf.intrIdx      = txq->comp_ring.intr_idx;
640 		tqd->status.stopped    = TRUE;
641 		tqd->status.error      = 0;
642 		memset(&tqd->stats, 0, sizeof(tqd->stats));
643 	}
644 
645 	for (i = 0; i < hw->num_rx_queues; i++) {
646 		Vmxnet3_RxQueueDesc *rqd  = &hw->rqd_start[i];
647 		vmxnet3_rx_queue_t *rxq   = dev->data->rx_queues[i];
648 
649 		rxq->shared = &hw->rqd_start[i];
650 
651 		rqd->conf.rxRingBasePA[0] = rxq->cmd_ring[0].basePA;
652 		rqd->conf.rxRingBasePA[1] = rxq->cmd_ring[1].basePA;
653 		rqd->conf.compRingBasePA  = rxq->comp_ring.basePA;
654 
655 		rqd->conf.rxRingSize[0]   = rxq->cmd_ring[0].size;
656 		rqd->conf.rxRingSize[1]   = rxq->cmd_ring[1].size;
657 		rqd->conf.compRingSize    = rxq->comp_ring.size;
658 		rqd->conf.intrIdx         = rxq->comp_ring.intr_idx;
659 		if (VMXNET3_VERSION_GE_3(hw)) {
660 			rqd->conf.rxDataRingBasePA = rxq->data_ring.basePA;
661 			rqd->conf.rxDataRingDescSize = rxq->data_desc_size;
662 		}
663 		rqd->status.stopped       = TRUE;
664 		rqd->status.error         = 0;
665 		memset(&rqd->stats, 0, sizeof(rqd->stats));
666 	}
667 
668 	/* RxMode set to 0 of VMXNET3_RXM_xxx */
669 	devRead->rxFilterConf.rxMode = 0;
670 
671 	/* Setting up feature flags */
672 	if (rx_offloads & DEV_RX_OFFLOAD_CHECKSUM)
673 		devRead->misc.uptFeatures |= VMXNET3_F_RXCSUM;
674 
675 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
676 		devRead->misc.uptFeatures |= VMXNET3_F_LRO;
677 		devRead->misc.maxNumRxSG = 0;
678 	}
679 
680 	if (port_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
681 		ret = vmxnet3_rss_configure(dev);
682 		if (ret != VMXNET3_SUCCESS)
683 			return ret;
684 
685 		devRead->misc.uptFeatures |= VMXNET3_F_RSS;
686 		devRead->rssConfDesc.confVer = 1;
687 		devRead->rssConfDesc.confLen = sizeof(struct VMXNET3_RSSConf);
688 		devRead->rssConfDesc.confPA  = hw->rss_confPA;
689 	}
690 
691 	ret = vmxnet3_dev_vlan_offload_set(dev,
692 			ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK);
693 	if (ret)
694 		return ret;
695 
696 	vmxnet3_write_mac(hw, dev->data->mac_addrs->addr_bytes);
697 
698 	return VMXNET3_SUCCESS;
699 }
700 
701 /*
702  * Configure device link speed and setup link.
703  * Must be called after eth_vmxnet3_dev_init. Other wise it might fail
704  * It returns 0 on success.
705  */
706 static int
707 vmxnet3_dev_start(struct rte_eth_dev *dev)
708 {
709 	int ret;
710 	struct vmxnet3_hw *hw = dev->data->dev_private;
711 
712 	PMD_INIT_FUNC_TRACE();
713 
714 	/* Save stats before it is reset by CMD_ACTIVATE */
715 	vmxnet3_hw_stats_save(hw);
716 
717 	ret = vmxnet3_setup_driver_shared(dev);
718 	if (ret != VMXNET3_SUCCESS)
719 		return ret;
720 
721 	/* check if lsc interrupt feature is enabled */
722 	if (dev->data->dev_conf.intr_conf.lsc) {
723 		struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
724 
725 		/* Setup interrupt callback  */
726 		rte_intr_callback_register(&pci_dev->intr_handle,
727 					   vmxnet3_interrupt_handler, dev);
728 
729 		if (rte_intr_enable(&pci_dev->intr_handle) < 0) {
730 			PMD_INIT_LOG(ERR, "interrupt enable failed");
731 			return -EIO;
732 		}
733 	}
734 
735 	/* Exchange shared data with device */
736 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL,
737 			       VMXNET3_GET_ADDR_LO(hw->sharedPA));
738 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAH,
739 			       VMXNET3_GET_ADDR_HI(hw->sharedPA));
740 
741 	/* Activate device by register write */
742 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_ACTIVATE_DEV);
743 	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
744 
745 	if (ret != 0) {
746 		PMD_INIT_LOG(ERR, "Device activation: UNSUCCESSFUL");
747 		return -EINVAL;
748 	}
749 
750 	/* Setup memory region for rx buffers */
751 	ret = vmxnet3_dev_setup_memreg(dev);
752 	if (ret == 0) {
753 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
754 				       VMXNET3_CMD_REGISTER_MEMREGS);
755 		ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
756 		if (ret != 0)
757 			PMD_INIT_LOG(DEBUG,
758 				     "Failed in setup memory region cmd\n");
759 		ret = 0;
760 	} else {
761 		PMD_INIT_LOG(DEBUG, "Failed to setup memory region\n");
762 	}
763 
764 	/* Disable interrupts */
765 	vmxnet3_disable_intr(hw);
766 
767 	/*
768 	 * Load RX queues with blank mbufs and update next2fill index for device
769 	 * Update RxMode of the device
770 	 */
771 	ret = vmxnet3_dev_rxtx_init(dev);
772 	if (ret != VMXNET3_SUCCESS) {
773 		PMD_INIT_LOG(ERR, "Device queue init: UNSUCCESSFUL");
774 		return ret;
775 	}
776 
777 	hw->adapter_stopped = FALSE;
778 
779 	/* Setting proper Rx Mode and issue Rx Mode Update command */
780 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_UCAST | VMXNET3_RXM_BCAST, 1);
781 
782 	if (dev->data->dev_conf.intr_conf.lsc) {
783 		vmxnet3_enable_intr(hw);
784 
785 		/*
786 		 * Update link state from device since this won't be
787 		 * done upon starting with lsc in use. This is done
788 		 * only after enabling interrupts to avoid any race
789 		 * where the link state could change without an
790 		 * interrupt being fired.
791 		 */
792 		__vmxnet3_dev_link_update(dev, 0);
793 	}
794 
795 	return VMXNET3_SUCCESS;
796 }
797 
798 /*
799  * Stop device: disable rx and tx functions to allow for reconfiguring.
800  */
801 static void
802 vmxnet3_dev_stop(struct rte_eth_dev *dev)
803 {
804 	struct rte_eth_link link;
805 	struct vmxnet3_hw *hw = dev->data->dev_private;
806 
807 	PMD_INIT_FUNC_TRACE();
808 
809 	if (hw->adapter_stopped == 1) {
810 		PMD_INIT_LOG(DEBUG, "Device already stopped.");
811 		return;
812 	}
813 
814 	/* disable interrupts */
815 	vmxnet3_disable_intr(hw);
816 
817 	if (dev->data->dev_conf.intr_conf.lsc) {
818 		struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
819 
820 		rte_intr_disable(&pci_dev->intr_handle);
821 
822 		rte_intr_callback_unregister(&pci_dev->intr_handle,
823 					     vmxnet3_interrupt_handler, dev);
824 	}
825 
826 	/* quiesce the device first */
827 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
828 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL, 0);
829 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAH, 0);
830 
831 	/* reset the device */
832 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
833 	PMD_INIT_LOG(DEBUG, "Device reset.");
834 
835 	vmxnet3_dev_clear_queues(dev);
836 
837 	/* Clear recorded link status */
838 	memset(&link, 0, sizeof(link));
839 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
840 	link.link_speed = ETH_SPEED_NUM_10G;
841 	link.link_autoneg = ETH_LINK_FIXED;
842 	rte_eth_linkstatus_set(dev, &link);
843 
844 	hw->adapter_stopped = 1;
845 }
846 
847 static void
848 vmxnet3_free_queues(struct rte_eth_dev *dev)
849 {
850 	int i;
851 
852 	PMD_INIT_FUNC_TRACE();
853 
854 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
855 		void *rxq = dev->data->rx_queues[i];
856 
857 		vmxnet3_dev_rx_queue_release(rxq);
858 	}
859 	dev->data->nb_rx_queues = 0;
860 
861 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
862 		void *txq = dev->data->tx_queues[i];
863 
864 		vmxnet3_dev_tx_queue_release(txq);
865 	}
866 	dev->data->nb_tx_queues = 0;
867 }
868 
869 /*
870  * Reset and stop device.
871  */
872 static void
873 vmxnet3_dev_close(struct rte_eth_dev *dev)
874 {
875 	PMD_INIT_FUNC_TRACE();
876 
877 	vmxnet3_dev_stop(dev);
878 	vmxnet3_free_queues(dev);
879 
880 	/*
881 	 * flag to rte_eth_dev_close() that it should release the port resources
882 	 * (calling rte_eth_dev_release_port()) in addition to closing it.
883 	 */
884 	dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
885 }
886 
887 static void
888 vmxnet3_hw_tx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
889 			struct UPT1_TxStats *res)
890 {
891 #define VMXNET3_UPDATE_TX_STAT(h, i, f, r)		\
892 		((r)->f = (h)->tqd_start[(i)].stats.f +	\
893 			(h)->saved_tx_stats[(i)].f)
894 
895 	VMXNET3_UPDATE_TX_STAT(hw, q, ucastPktsTxOK, res);
896 	VMXNET3_UPDATE_TX_STAT(hw, q, mcastPktsTxOK, res);
897 	VMXNET3_UPDATE_TX_STAT(hw, q, bcastPktsTxOK, res);
898 	VMXNET3_UPDATE_TX_STAT(hw, q, ucastBytesTxOK, res);
899 	VMXNET3_UPDATE_TX_STAT(hw, q, mcastBytesTxOK, res);
900 	VMXNET3_UPDATE_TX_STAT(hw, q, bcastBytesTxOK, res);
901 	VMXNET3_UPDATE_TX_STAT(hw, q, pktsTxError, res);
902 	VMXNET3_UPDATE_TX_STAT(hw, q, pktsTxDiscard, res);
903 
904 #undef VMXNET3_UPDATE_TX_STAT
905 }
906 
907 static void
908 vmxnet3_hw_rx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
909 			struct UPT1_RxStats *res)
910 {
911 #define VMXNET3_UPDATE_RX_STAT(h, i, f, r)		\
912 		((r)->f = (h)->rqd_start[(i)].stats.f +	\
913 			(h)->saved_rx_stats[(i)].f)
914 
915 	VMXNET3_UPDATE_RX_STAT(hw, q, ucastPktsRxOK, res);
916 	VMXNET3_UPDATE_RX_STAT(hw, q, mcastPktsRxOK, res);
917 	VMXNET3_UPDATE_RX_STAT(hw, q, bcastPktsRxOK, res);
918 	VMXNET3_UPDATE_RX_STAT(hw, q, ucastBytesRxOK, res);
919 	VMXNET3_UPDATE_RX_STAT(hw, q, mcastBytesRxOK, res);
920 	VMXNET3_UPDATE_RX_STAT(hw, q, bcastBytesRxOK, res);
921 	VMXNET3_UPDATE_RX_STAT(hw, q, pktsRxError, res);
922 	VMXNET3_UPDATE_RX_STAT(hw, q, pktsRxOutOfBuf, res);
923 
924 #undef VMXNET3_UPDATE_RX_STAT
925 }
926 
927 static void
928 vmxnet3_tx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
929 					struct UPT1_TxStats *res)
930 {
931 		vmxnet3_hw_tx_stats_get(hw, q, res);
932 
933 #define VMXNET3_REDUCE_SNAPSHOT_TX_STAT(h, i, f, r)	\
934 		((r)->f -= (h)->snapshot_tx_stats[(i)].f)
935 
936 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, ucastPktsTxOK, res);
937 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, mcastPktsTxOK, res);
938 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, bcastPktsTxOK, res);
939 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, ucastBytesTxOK, res);
940 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, mcastBytesTxOK, res);
941 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, bcastBytesTxOK, res);
942 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, pktsTxError, res);
943 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, pktsTxDiscard, res);
944 
945 #undef VMXNET3_REDUCE_SNAPSHOT_TX_STAT
946 }
947 
948 static void
949 vmxnet3_rx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
950 					struct UPT1_RxStats *res)
951 {
952 		vmxnet3_hw_rx_stats_get(hw, q, res);
953 
954 #define VMXNET3_REDUCE_SNAPSHOT_RX_STAT(h, i, f, r)	\
955 		((r)->f -= (h)->snapshot_rx_stats[(i)].f)
956 
957 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, ucastPktsRxOK, res);
958 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, mcastPktsRxOK, res);
959 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, bcastPktsRxOK, res);
960 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, ucastBytesRxOK, res);
961 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, mcastBytesRxOK, res);
962 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, bcastBytesRxOK, res);
963 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, pktsRxError, res);
964 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, pktsRxOutOfBuf, res);
965 
966 #undef VMXNET3_REDUCE_SNAPSHOT_RX_STAT
967 }
968 
969 static void
970 vmxnet3_hw_stats_save(struct vmxnet3_hw *hw)
971 {
972 	unsigned int i;
973 
974 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
975 
976 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
977 
978 	for (i = 0; i < hw->num_tx_queues; i++)
979 		vmxnet3_hw_tx_stats_get(hw, i, &hw->saved_tx_stats[i]);
980 	for (i = 0; i < hw->num_rx_queues; i++)
981 		vmxnet3_hw_rx_stats_get(hw, i, &hw->saved_rx_stats[i]);
982 }
983 
984 static int
985 vmxnet3_dev_xstats_get_names(struct rte_eth_dev *dev,
986 			     struct rte_eth_xstat_name *xstats_names,
987 			     unsigned int n)
988 {
989 	unsigned int i, t, count = 0;
990 	unsigned int nstats =
991 		dev->data->nb_tx_queues * RTE_DIM(vmxnet3_txq_stat_strings) +
992 		dev->data->nb_rx_queues * RTE_DIM(vmxnet3_rxq_stat_strings);
993 
994 	if (!xstats_names || n < nstats)
995 		return nstats;
996 
997 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
998 		if (!dev->data->rx_queues[i])
999 			continue;
1000 
1001 		for (t = 0; t < RTE_DIM(vmxnet3_rxq_stat_strings); t++) {
1002 			snprintf(xstats_names[count].name,
1003 				 sizeof(xstats_names[count].name),
1004 				 "rx_q%u_%s", i,
1005 				 vmxnet3_rxq_stat_strings[t].name);
1006 			count++;
1007 		}
1008 	}
1009 
1010 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1011 		if (!dev->data->tx_queues[i])
1012 			continue;
1013 
1014 		for (t = 0; t < RTE_DIM(vmxnet3_txq_stat_strings); t++) {
1015 			snprintf(xstats_names[count].name,
1016 				 sizeof(xstats_names[count].name),
1017 				 "tx_q%u_%s", i,
1018 				 vmxnet3_txq_stat_strings[t].name);
1019 			count++;
1020 		}
1021 	}
1022 
1023 	return count;
1024 }
1025 
1026 static int
1027 vmxnet3_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1028 		       unsigned int n)
1029 {
1030 	unsigned int i, t, count = 0;
1031 	unsigned int nstats =
1032 		dev->data->nb_tx_queues * RTE_DIM(vmxnet3_txq_stat_strings) +
1033 		dev->data->nb_rx_queues * RTE_DIM(vmxnet3_rxq_stat_strings);
1034 
1035 	if (n < nstats)
1036 		return nstats;
1037 
1038 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1039 		struct vmxnet3_rx_queue *rxq = dev->data->rx_queues[i];
1040 
1041 		if (rxq == NULL)
1042 			continue;
1043 
1044 		for (t = 0; t < RTE_DIM(vmxnet3_rxq_stat_strings); t++) {
1045 			xstats[count].value = *(uint64_t *)(((char *)&rxq->stats) +
1046 				vmxnet3_rxq_stat_strings[t].offset);
1047 			xstats[count].id = count;
1048 			count++;
1049 		}
1050 	}
1051 
1052 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1053 		struct vmxnet3_tx_queue *txq = dev->data->tx_queues[i];
1054 
1055 		if (txq == NULL)
1056 			continue;
1057 
1058 		for (t = 0; t < RTE_DIM(vmxnet3_txq_stat_strings); t++) {
1059 			xstats[count].value = *(uint64_t *)(((char *)&txq->stats) +
1060 				vmxnet3_txq_stat_strings[t].offset);
1061 			xstats[count].id = count;
1062 			count++;
1063 		}
1064 	}
1065 
1066 	return count;
1067 }
1068 
1069 static int
1070 vmxnet3_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1071 {
1072 	unsigned int i;
1073 	struct vmxnet3_hw *hw = dev->data->dev_private;
1074 	struct UPT1_TxStats txStats;
1075 	struct UPT1_RxStats rxStats;
1076 
1077 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
1078 
1079 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
1080 	for (i = 0; i < hw->num_tx_queues; i++) {
1081 		vmxnet3_tx_stats_get(hw, i, &txStats);
1082 
1083 		stats->q_opackets[i] = txStats.ucastPktsTxOK +
1084 			txStats.mcastPktsTxOK +
1085 			txStats.bcastPktsTxOK;
1086 
1087 		stats->q_obytes[i] = txStats.ucastBytesTxOK +
1088 			txStats.mcastBytesTxOK +
1089 			txStats.bcastBytesTxOK;
1090 
1091 		stats->opackets += stats->q_opackets[i];
1092 		stats->obytes += stats->q_obytes[i];
1093 		stats->oerrors += txStats.pktsTxError + txStats.pktsTxDiscard;
1094 	}
1095 
1096 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_RX_QUEUES);
1097 	for (i = 0; i < hw->num_rx_queues; i++) {
1098 		vmxnet3_rx_stats_get(hw, i, &rxStats);
1099 
1100 		stats->q_ipackets[i] = rxStats.ucastPktsRxOK +
1101 			rxStats.mcastPktsRxOK +
1102 			rxStats.bcastPktsRxOK;
1103 
1104 		stats->q_ibytes[i] = rxStats.ucastBytesRxOK +
1105 			rxStats.mcastBytesRxOK +
1106 			rxStats.bcastBytesRxOK;
1107 
1108 		stats->ipackets += stats->q_ipackets[i];
1109 		stats->ibytes += stats->q_ibytes[i];
1110 
1111 		stats->q_errors[i] = rxStats.pktsRxError;
1112 		stats->ierrors += rxStats.pktsRxError;
1113 		stats->imissed += rxStats.pktsRxOutOfBuf;
1114 	}
1115 
1116 	return 0;
1117 }
1118 
1119 static void
1120 vmxnet3_dev_stats_reset(struct rte_eth_dev *dev)
1121 {
1122 	unsigned int i;
1123 	struct vmxnet3_hw *hw = dev->data->dev_private;
1124 	struct UPT1_TxStats txStats;
1125 	struct UPT1_RxStats rxStats;
1126 
1127 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
1128 
1129 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
1130 
1131 	for (i = 0; i < hw->num_tx_queues; i++) {
1132 		vmxnet3_hw_tx_stats_get(hw, i, &txStats);
1133 		memcpy(&hw->snapshot_tx_stats[i], &txStats,
1134 			sizeof(hw->snapshot_tx_stats[0]));
1135 	}
1136 	for (i = 0; i < hw->num_rx_queues; i++) {
1137 		vmxnet3_hw_rx_stats_get(hw, i, &rxStats);
1138 		memcpy(&hw->snapshot_rx_stats[i], &rxStats,
1139 			sizeof(hw->snapshot_rx_stats[0]));
1140 	}
1141 }
1142 
1143 static void
1144 vmxnet3_dev_info_get(struct rte_eth_dev *dev __rte_unused,
1145 		     struct rte_eth_dev_info *dev_info)
1146 {
1147 	dev_info->max_rx_queues = VMXNET3_MAX_RX_QUEUES;
1148 	dev_info->max_tx_queues = VMXNET3_MAX_TX_QUEUES;
1149 	dev_info->min_rx_bufsize = 1518 + RTE_PKTMBUF_HEADROOM;
1150 	dev_info->max_rx_pktlen = 16384; /* includes CRC, cf MAXFRS register */
1151 	dev_info->speed_capa = ETH_LINK_SPEED_10G;
1152 	dev_info->max_mac_addrs = VMXNET3_MAX_MAC_ADDRS;
1153 
1154 	dev_info->flow_type_rss_offloads = VMXNET3_RSS_OFFLOAD_ALL;
1155 
1156 	dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
1157 		.nb_max = VMXNET3_RX_RING_MAX_SIZE,
1158 		.nb_min = VMXNET3_DEF_RX_RING_SIZE,
1159 		.nb_align = 1,
1160 	};
1161 
1162 	dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
1163 		.nb_max = VMXNET3_TX_RING_MAX_SIZE,
1164 		.nb_min = VMXNET3_DEF_TX_RING_SIZE,
1165 		.nb_align = 1,
1166 		.nb_seg_max = VMXNET3_TX_MAX_SEG,
1167 		.nb_mtu_seg_max = VMXNET3_MAX_TXD_PER_PKT,
1168 	};
1169 
1170 	dev_info->rx_offload_capa = VMXNET3_RX_OFFLOAD_CAP;
1171 	dev_info->rx_queue_offload_capa = 0;
1172 	dev_info->tx_offload_capa = VMXNET3_TX_OFFLOAD_CAP;
1173 	dev_info->tx_queue_offload_capa = 0;
1174 }
1175 
1176 static const uint32_t *
1177 vmxnet3_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1178 {
1179 	static const uint32_t ptypes[] = {
1180 		RTE_PTYPE_L3_IPV4_EXT,
1181 		RTE_PTYPE_L3_IPV4,
1182 		RTE_PTYPE_UNKNOWN
1183 	};
1184 
1185 	if (dev->rx_pkt_burst == vmxnet3_recv_pkts)
1186 		return ptypes;
1187 	return NULL;
1188 }
1189 
1190 static int
1191 vmxnet3_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1192 {
1193 	struct vmxnet3_hw *hw = dev->data->dev_private;
1194 
1195 	ether_addr_copy(mac_addr, (struct ether_addr *)(hw->perm_addr));
1196 	vmxnet3_write_mac(hw, mac_addr->addr_bytes);
1197 	return 0;
1198 }
1199 
1200 /* return 0 means link status changed, -1 means not changed */
1201 static int
1202 __vmxnet3_dev_link_update(struct rte_eth_dev *dev,
1203 			  __rte_unused int wait_to_complete)
1204 {
1205 	struct vmxnet3_hw *hw = dev->data->dev_private;
1206 	struct rte_eth_link link;
1207 	uint32_t ret;
1208 
1209 	memset(&link, 0, sizeof(link));
1210 
1211 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
1212 	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
1213 
1214 	if (ret & 0x1)
1215 		link.link_status = ETH_LINK_UP;
1216 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1217 	link.link_speed = ETH_SPEED_NUM_10G;
1218 	link.link_autoneg = ETH_LINK_FIXED;
1219 
1220 	return rte_eth_linkstatus_set(dev, &link);
1221 }
1222 
1223 static int
1224 vmxnet3_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
1225 {
1226 	/* Link status doesn't change for stopped dev */
1227 	if (dev->data->dev_started == 0)
1228 		return -1;
1229 
1230 	return __vmxnet3_dev_link_update(dev, wait_to_complete);
1231 }
1232 
1233 /* Updating rxmode through Vmxnet3_DriverShared structure in adapter */
1234 static void
1235 vmxnet3_dev_set_rxmode(struct vmxnet3_hw *hw, uint32_t feature, int set)
1236 {
1237 	struct Vmxnet3_RxFilterConf *rxConf = &hw->shared->devRead.rxFilterConf;
1238 
1239 	if (set)
1240 		rxConf->rxMode = rxConf->rxMode | feature;
1241 	else
1242 		rxConf->rxMode = rxConf->rxMode & (~feature);
1243 
1244 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_UPDATE_RX_MODE);
1245 }
1246 
1247 /* Promiscuous supported only if Vmxnet3_DriverShared is initialized in adapter */
1248 static void
1249 vmxnet3_dev_promiscuous_enable(struct rte_eth_dev *dev)
1250 {
1251 	struct vmxnet3_hw *hw = dev->data->dev_private;
1252 	uint32_t *vf_table = hw->shared->devRead.rxFilterConf.vfTable;
1253 
1254 	memset(vf_table, 0, VMXNET3_VFT_TABLE_SIZE);
1255 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 1);
1256 
1257 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1258 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1259 }
1260 
1261 /* Promiscuous supported only if Vmxnet3_DriverShared is initialized in adapter */
1262 static void
1263 vmxnet3_dev_promiscuous_disable(struct rte_eth_dev *dev)
1264 {
1265 	struct vmxnet3_hw *hw = dev->data->dev_private;
1266 	uint32_t *vf_table = hw->shared->devRead.rxFilterConf.vfTable;
1267 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1268 
1269 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1270 		memcpy(vf_table, hw->shadow_vfta, VMXNET3_VFT_TABLE_SIZE);
1271 	else
1272 		memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);
1273 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 0);
1274 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1275 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1276 }
1277 
1278 /* Allmulticast supported only if Vmxnet3_DriverShared is initialized in adapter */
1279 static void
1280 vmxnet3_dev_allmulticast_enable(struct rte_eth_dev *dev)
1281 {
1282 	struct vmxnet3_hw *hw = dev->data->dev_private;
1283 
1284 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_ALL_MULTI, 1);
1285 }
1286 
1287 /* Allmulticast supported only if Vmxnet3_DriverShared is initialized in adapter */
1288 static void
1289 vmxnet3_dev_allmulticast_disable(struct rte_eth_dev *dev)
1290 {
1291 	struct vmxnet3_hw *hw = dev->data->dev_private;
1292 
1293 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_ALL_MULTI, 0);
1294 }
1295 
1296 /* Enable/disable filter on vlan */
1297 static int
1298 vmxnet3_dev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vid, int on)
1299 {
1300 	struct vmxnet3_hw *hw = dev->data->dev_private;
1301 	struct Vmxnet3_RxFilterConf *rxConf = &hw->shared->devRead.rxFilterConf;
1302 	uint32_t *vf_table = rxConf->vfTable;
1303 
1304 	/* save state for restore */
1305 	if (on)
1306 		VMXNET3_SET_VFTABLE_ENTRY(hw->shadow_vfta, vid);
1307 	else
1308 		VMXNET3_CLEAR_VFTABLE_ENTRY(hw->shadow_vfta, vid);
1309 
1310 	/* don't change active filter if in promiscuous mode */
1311 	if (rxConf->rxMode & VMXNET3_RXM_PROMISC)
1312 		return 0;
1313 
1314 	/* set in hardware */
1315 	if (on)
1316 		VMXNET3_SET_VFTABLE_ENTRY(vf_table, vid);
1317 	else
1318 		VMXNET3_CLEAR_VFTABLE_ENTRY(vf_table, vid);
1319 
1320 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1321 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1322 	return 0;
1323 }
1324 
1325 static int
1326 vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1327 {
1328 	struct vmxnet3_hw *hw = dev->data->dev_private;
1329 	Vmxnet3_DSDevRead *devRead = &hw->shared->devRead;
1330 	uint32_t *vf_table = devRead->rxFilterConf.vfTable;
1331 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1332 
1333 	if (mask & ETH_VLAN_STRIP_MASK) {
1334 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1335 			devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1336 		else
1337 			devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;
1338 
1339 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1340 				       VMXNET3_CMD_UPDATE_FEATURE);
1341 	}
1342 
1343 	if (mask & ETH_VLAN_FILTER_MASK) {
1344 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1345 			memcpy(vf_table, hw->shadow_vfta, VMXNET3_VFT_TABLE_SIZE);
1346 		else
1347 			memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);
1348 
1349 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1350 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1351 	}
1352 
1353 	return 0;
1354 }
1355 
1356 static void
1357 vmxnet3_process_events(struct rte_eth_dev *dev)
1358 {
1359 	struct vmxnet3_hw *hw = dev->data->dev_private;
1360 	uint32_t events = hw->shared->ecr;
1361 
1362 	if (!events)
1363 		return;
1364 
1365 	/*
1366 	 * ECR bits when written with 1b are cleared. Hence write
1367 	 * events back to ECR so that the bits which were set will be reset.
1368 	 */
1369 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_ECR, events);
1370 
1371 	/* Check if link state has changed */
1372 	if (events & VMXNET3_ECR_LINK) {
1373 		PMD_DRV_LOG(DEBUG, "Process events: VMXNET3_ECR_LINK event");
1374 		if (vmxnet3_dev_link_update(dev, 0) == 0)
1375 			_rte_eth_dev_callback_process(dev,
1376 						      RTE_ETH_EVENT_INTR_LSC,
1377 						      NULL);
1378 	}
1379 
1380 	/* Check if there is an error on xmit/recv queues */
1381 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
1382 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1383 				       VMXNET3_CMD_GET_QUEUE_STATUS);
1384 
1385 		if (hw->tqd_start->status.stopped)
1386 			PMD_DRV_LOG(ERR, "tq error 0x%x",
1387 				    hw->tqd_start->status.error);
1388 
1389 		if (hw->rqd_start->status.stopped)
1390 			PMD_DRV_LOG(ERR, "rq error 0x%x",
1391 				     hw->rqd_start->status.error);
1392 
1393 		/* Reset the device */
1394 		/* Have to reset the device */
1395 	}
1396 
1397 	if (events & VMXNET3_ECR_DIC)
1398 		PMD_DRV_LOG(DEBUG, "Device implementation change event.");
1399 
1400 	if (events & VMXNET3_ECR_DEBUG)
1401 		PMD_DRV_LOG(DEBUG, "Debug event generated by device.");
1402 }
1403 
1404 static void
1405 vmxnet3_interrupt_handler(void *param)
1406 {
1407 	struct rte_eth_dev *dev = param;
1408 	struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
1409 
1410 	vmxnet3_process_events(dev);
1411 
1412 	if (rte_intr_enable(&pci_dev->intr_handle) < 0)
1413 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1414 }
1415 
1416 RTE_PMD_REGISTER_PCI(net_vmxnet3, rte_vmxnet3_pmd);
1417 RTE_PMD_REGISTER_PCI_TABLE(net_vmxnet3, pci_id_vmxnet3_map);
1418 RTE_PMD_REGISTER_KMOD_DEP(net_vmxnet3, "* igb_uio | uio_pci_generic | vfio-pci");
1419 
1420 RTE_INIT(vmxnet3_init_log)
1421 {
1422 	vmxnet3_logtype_init = rte_log_register("pmd.net.vmxnet3.init");
1423 	if (vmxnet3_logtype_init >= 0)
1424 		rte_log_set_level(vmxnet3_logtype_init, RTE_LOG_NOTICE);
1425 	vmxnet3_logtype_driver = rte_log_register("pmd.net.vmxnet3.driver");
1426 	if (vmxnet3_logtype_driver >= 0)
1427 		rte_log_set_level(vmxnet3_logtype_driver, RTE_LOG_NOTICE);
1428 }
1429