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 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_3)) {
271 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
272 				       1 << VMXNET3_REV_3);
273 		hw->version = VMXNET3_REV_3 + 1;
274 	} else if (ver & (1 << VMXNET3_REV_2)) {
275 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
276 				       1 << VMXNET3_REV_2);
277 		hw->version = VMXNET3_REV_2 + 1;
278 	} else if (ver & (1 << VMXNET3_REV_1)) {
279 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_VRRS,
280 				       1 << VMXNET3_REV_1);
281 		hw->version = VMXNET3_REV_1 + 1;
282 	} else {
283 		PMD_INIT_LOG(ERR, "Incompatible hardware version: %d", ver);
284 		return -EIO;
285 	}
286 
287 	PMD_INIT_LOG(DEBUG, "Using device version %d\n", hw->version);
288 
289 	/* Check UPT version compatibility with driver. */
290 	ver = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_UVRS);
291 	PMD_INIT_LOG(DEBUG, "UPT hardware version : %d", ver);
292 	if (ver & 0x1)
293 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_UVRS, 1);
294 	else {
295 		PMD_INIT_LOG(ERR, "Incompatible UPT version.");
296 		return -EIO;
297 	}
298 
299 	/* Getting MAC Address */
300 	mac_lo = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_MACL);
301 	mac_hi = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_MACH);
302 	memcpy(hw->perm_addr, &mac_lo, 4);
303 	memcpy(hw->perm_addr + 4, &mac_hi, 2);
304 
305 	/* Allocate memory for storing MAC addresses */
306 	eth_dev->data->mac_addrs = rte_zmalloc("vmxnet3", ETHER_ADDR_LEN *
307 					       VMXNET3_MAX_MAC_ADDRS, 0);
308 	if (eth_dev->data->mac_addrs == NULL) {
309 		PMD_INIT_LOG(ERR,
310 			     "Failed to allocate %d bytes needed to store MAC addresses",
311 			     ETHER_ADDR_LEN * VMXNET3_MAX_MAC_ADDRS);
312 		return -ENOMEM;
313 	}
314 	/* Copy the permanent MAC address */
315 	ether_addr_copy((struct ether_addr *) hw->perm_addr,
316 			&eth_dev->data->mac_addrs[0]);
317 
318 	PMD_INIT_LOG(DEBUG, "MAC Address : %02x:%02x:%02x:%02x:%02x:%02x",
319 		     hw->perm_addr[0], hw->perm_addr[1], hw->perm_addr[2],
320 		     hw->perm_addr[3], hw->perm_addr[4], hw->perm_addr[5]);
321 
322 	/* Flag to call rte_eth_dev_release_port() in rte_eth_dev_close(). */
323 	eth_dev->data->dev_flags |= RTE_ETH_DEV_CLOSE_REMOVE;
324 
325 	/* Put device in Quiesce Mode */
326 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
327 
328 	/* allow untagged pkts */
329 	VMXNET3_SET_VFTABLE_ENTRY(hw->shadow_vfta, 0);
330 
331 	hw->txdata_desc_size = VMXNET3_VERSION_GE_3(hw) ?
332 		eth_vmxnet3_txdata_get(hw) : sizeof(struct Vmxnet3_TxDataDesc);
333 
334 	hw->rxdata_desc_size = VMXNET3_VERSION_GE_3(hw) ?
335 		VMXNET3_DEF_RXDATA_DESC_SIZE : 0;
336 	RTE_ASSERT((hw->rxdata_desc_size & ~VMXNET3_RXDATA_DESC_SIZE_MASK) ==
337 		   hw->rxdata_desc_size);
338 
339 	/* clear shadow stats */
340 	memset(hw->saved_tx_stats, 0, sizeof(hw->saved_tx_stats));
341 	memset(hw->saved_rx_stats, 0, sizeof(hw->saved_rx_stats));
342 
343 	/* clear snapshot stats */
344 	memset(hw->snapshot_tx_stats, 0, sizeof(hw->snapshot_tx_stats));
345 	memset(hw->snapshot_rx_stats, 0, sizeof(hw->snapshot_rx_stats));
346 
347 	/* set the initial link status */
348 	memset(&link, 0, sizeof(link));
349 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
350 	link.link_speed = ETH_SPEED_NUM_10G;
351 	link.link_autoneg = ETH_LINK_FIXED;
352 	rte_eth_linkstatus_set(eth_dev, &link);
353 
354 	return 0;
355 }
356 
357 static int
358 eth_vmxnet3_dev_uninit(struct rte_eth_dev *eth_dev)
359 {
360 	struct vmxnet3_hw *hw = eth_dev->data->dev_private;
361 
362 	PMD_INIT_FUNC_TRACE();
363 
364 	if (rte_eal_process_type() != RTE_PROC_PRIMARY)
365 		return 0;
366 
367 	if (hw->adapter_stopped == 0) {
368 		PMD_INIT_LOG(DEBUG, "Device has not been closed.");
369 		return -EBUSY;
370 	}
371 
372 	eth_dev->dev_ops = NULL;
373 	eth_dev->rx_pkt_burst = NULL;
374 	eth_dev->tx_pkt_burst = NULL;
375 	eth_dev->tx_pkt_prepare = NULL;
376 
377 	return 0;
378 }
379 
380 static int eth_vmxnet3_pci_probe(struct rte_pci_driver *pci_drv __rte_unused,
381 	struct rte_pci_device *pci_dev)
382 {
383 	return rte_eth_dev_pci_generic_probe(pci_dev,
384 		sizeof(struct vmxnet3_hw), eth_vmxnet3_dev_init);
385 }
386 
387 static int eth_vmxnet3_pci_remove(struct rte_pci_device *pci_dev)
388 {
389 	return rte_eth_dev_pci_generic_remove(pci_dev, eth_vmxnet3_dev_uninit);
390 }
391 
392 static struct rte_pci_driver rte_vmxnet3_pmd = {
393 	.id_table = pci_id_vmxnet3_map,
394 	.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_INTR_LSC,
395 	.probe = eth_vmxnet3_pci_probe,
396 	.remove = eth_vmxnet3_pci_remove,
397 };
398 
399 static int
400 vmxnet3_dev_configure(struct rte_eth_dev *dev)
401 {
402 	const struct rte_memzone *mz;
403 	struct vmxnet3_hw *hw = dev->data->dev_private;
404 	size_t size;
405 
406 	PMD_INIT_FUNC_TRACE();
407 
408 	if (dev->data->nb_tx_queues > VMXNET3_MAX_TX_QUEUES ||
409 	    dev->data->nb_rx_queues > VMXNET3_MAX_RX_QUEUES) {
410 		PMD_INIT_LOG(ERR, "ERROR: Number of queues not supported");
411 		return -EINVAL;
412 	}
413 
414 	if (!rte_is_power_of_2(dev->data->nb_rx_queues)) {
415 		PMD_INIT_LOG(ERR, "ERROR: Number of rx queues not power of 2");
416 		return -EINVAL;
417 	}
418 
419 	size = dev->data->nb_rx_queues * sizeof(struct Vmxnet3_TxQueueDesc) +
420 		dev->data->nb_tx_queues * sizeof(struct Vmxnet3_RxQueueDesc);
421 
422 	if (size > UINT16_MAX)
423 		return -EINVAL;
424 
425 	hw->num_rx_queues = (uint8_t)dev->data->nb_rx_queues;
426 	hw->num_tx_queues = (uint8_t)dev->data->nb_tx_queues;
427 
428 	/*
429 	 * Allocate a memzone for Vmxnet3_DriverShared - Vmxnet3_DSDevRead
430 	 * on current socket
431 	 */
432 	mz = gpa_zone_reserve(dev, sizeof(struct Vmxnet3_DriverShared),
433 			      "shared", rte_socket_id(), 8, 1);
434 
435 	if (mz == NULL) {
436 		PMD_INIT_LOG(ERR, "ERROR: Creating shared zone");
437 		return -ENOMEM;
438 	}
439 	memset(mz->addr, 0, mz->len);
440 
441 	hw->shared = mz->addr;
442 	hw->sharedPA = mz->iova;
443 
444 	/*
445 	 * Allocate a memzone for Vmxnet3_RxQueueDesc - Vmxnet3_TxQueueDesc
446 	 * on current socket.
447 	 *
448 	 * We cannot reuse this memzone from previous allocation as its size
449 	 * depends on the number of tx and rx queues, which could be different
450 	 * from one config to another.
451 	 */
452 	mz = gpa_zone_reserve(dev, size, "queuedesc", rte_socket_id(),
453 			      VMXNET3_QUEUE_DESC_ALIGN, 0);
454 	if (mz == NULL) {
455 		PMD_INIT_LOG(ERR, "ERROR: Creating queue descriptors zone");
456 		return -ENOMEM;
457 	}
458 	memset(mz->addr, 0, mz->len);
459 
460 	hw->tqd_start = (Vmxnet3_TxQueueDesc *)mz->addr;
461 	hw->rqd_start = (Vmxnet3_RxQueueDesc *)(hw->tqd_start + hw->num_tx_queues);
462 
463 	hw->queueDescPA = mz->iova;
464 	hw->queue_desc_len = (uint16_t)size;
465 
466 	if (dev->data->dev_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
467 		/* Allocate memory structure for UPT1_RSSConf and configure */
468 		mz = gpa_zone_reserve(dev, sizeof(struct VMXNET3_RSSConf),
469 				      "rss_conf", rte_socket_id(),
470 				      RTE_CACHE_LINE_SIZE, 1);
471 		if (mz == NULL) {
472 			PMD_INIT_LOG(ERR,
473 				     "ERROR: Creating rss_conf structure zone");
474 			return -ENOMEM;
475 		}
476 		memset(mz->addr, 0, mz->len);
477 
478 		hw->rss_conf = mz->addr;
479 		hw->rss_confPA = mz->iova;
480 	}
481 
482 	return 0;
483 }
484 
485 static void
486 vmxnet3_write_mac(struct vmxnet3_hw *hw, const uint8_t *addr)
487 {
488 	uint32_t val;
489 
490 	PMD_INIT_LOG(DEBUG,
491 		     "Writing MAC Address : %02x:%02x:%02x:%02x:%02x:%02x",
492 		     addr[0], addr[1], addr[2],
493 		     addr[3], addr[4], addr[5]);
494 
495 	memcpy(&val, addr, 4);
496 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_MACL, val);
497 
498 	memcpy(&val, addr + 4, 2);
499 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_MACH, val);
500 }
501 
502 static int
503 vmxnet3_dev_setup_memreg(struct rte_eth_dev *dev)
504 {
505 	struct vmxnet3_hw *hw = dev->data->dev_private;
506 	Vmxnet3_DriverShared *shared = hw->shared;
507 	Vmxnet3_CmdInfo *cmdInfo;
508 	struct rte_mempool *mp[VMXNET3_MAX_RX_QUEUES];
509 	uint8_t index[VMXNET3_MAX_RX_QUEUES + VMXNET3_MAX_TX_QUEUES];
510 	uint32_t num, i, j, size;
511 
512 	if (hw->memRegsPA == 0) {
513 		const struct rte_memzone *mz;
514 
515 		size = sizeof(Vmxnet3_MemRegs) +
516 			(VMXNET3_MAX_RX_QUEUES + VMXNET3_MAX_TX_QUEUES) *
517 			sizeof(Vmxnet3_MemoryRegion);
518 
519 		mz = gpa_zone_reserve(dev, size, "memRegs", rte_socket_id(), 8,
520 				      1);
521 		if (mz == NULL) {
522 			PMD_INIT_LOG(ERR, "ERROR: Creating memRegs zone");
523 			return -ENOMEM;
524 		}
525 		memset(mz->addr, 0, mz->len);
526 		hw->memRegs = mz->addr;
527 		hw->memRegsPA = mz->iova;
528 	}
529 
530 	num = hw->num_rx_queues;
531 
532 	for (i = 0; i < num; i++) {
533 		vmxnet3_rx_queue_t *rxq = dev->data->rx_queues[i];
534 
535 		mp[i] = rxq->mp;
536 		index[i] = 1 << i;
537 	}
538 
539 	/*
540 	 * The same mempool could be used by multiple queues. In such a case,
541 	 * remove duplicate mempool entries. Only one entry is kept with
542 	 * bitmask indicating queues that are using this mempool.
543 	 */
544 	for (i = 1; i < num; i++) {
545 		for (j = 0; j < i; j++) {
546 			if (mp[i] == mp[j]) {
547 				mp[i] = NULL;
548 				index[j] |= 1 << i;
549 				break;
550 			}
551 		}
552 	}
553 
554 	j = 0;
555 	for (i = 0; i < num; i++) {
556 		if (mp[i] == NULL)
557 			continue;
558 
559 		Vmxnet3_MemoryRegion *mr = &hw->memRegs->memRegs[j];
560 
561 		mr->startPA =
562 			(uintptr_t)STAILQ_FIRST(&mp[i]->mem_list)->iova;
563 		mr->length = STAILQ_FIRST(&mp[i]->mem_list)->len <= INT32_MAX ?
564 			STAILQ_FIRST(&mp[i]->mem_list)->len : INT32_MAX;
565 		mr->txQueueBits = index[i];
566 		mr->rxQueueBits = index[i];
567 
568 		PMD_INIT_LOG(INFO,
569 			     "index: %u startPA: %" PRIu64 " length: %u, "
570 			     "rxBits: %x",
571 			     j, mr->startPA, mr->length, mr->rxQueueBits);
572 		j++;
573 	}
574 	hw->memRegs->numRegs = j;
575 	PMD_INIT_LOG(INFO, "numRegs: %u", j);
576 
577 	size = sizeof(Vmxnet3_MemRegs) +
578 		(j - 1) * sizeof(Vmxnet3_MemoryRegion);
579 
580 	cmdInfo = &shared->cu.cmdInfo;
581 	cmdInfo->varConf.confVer = 1;
582 	cmdInfo->varConf.confLen = size;
583 	cmdInfo->varConf.confPA = hw->memRegsPA;
584 
585 	return 0;
586 }
587 
588 static int
589 vmxnet3_setup_driver_shared(struct rte_eth_dev *dev)
590 {
591 	struct rte_eth_conf port_conf = dev->data->dev_conf;
592 	struct vmxnet3_hw *hw = dev->data->dev_private;
593 	uint32_t mtu = dev->data->mtu;
594 	Vmxnet3_DriverShared *shared = hw->shared;
595 	Vmxnet3_DSDevRead *devRead = &shared->devRead;
596 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
597 	uint32_t i;
598 	int ret;
599 
600 	hw->mtu = mtu;
601 
602 	shared->magic = VMXNET3_REV1_MAGIC;
603 	devRead->misc.driverInfo.version = VMXNET3_DRIVER_VERSION_NUM;
604 
605 	/* Setting up Guest OS information */
606 	devRead->misc.driverInfo.gos.gosBits   = sizeof(void *) == 4 ?
607 		VMXNET3_GOS_BITS_32 : VMXNET3_GOS_BITS_64;
608 	devRead->misc.driverInfo.gos.gosType   = VMXNET3_GOS_TYPE_LINUX;
609 	devRead->misc.driverInfo.vmxnet3RevSpt = 1;
610 	devRead->misc.driverInfo.uptVerSpt     = 1;
611 
612 	devRead->misc.mtu = rte_le_to_cpu_32(mtu);
613 	devRead->misc.queueDescPA  = hw->queueDescPA;
614 	devRead->misc.queueDescLen = hw->queue_desc_len;
615 	devRead->misc.numTxQueues  = hw->num_tx_queues;
616 	devRead->misc.numRxQueues  = hw->num_rx_queues;
617 
618 	/*
619 	 * Set number of interrupts to 1
620 	 * PMD by default disables all the interrupts but this is MUST
621 	 * to activate device. It needs at least one interrupt for
622 	 * link events to handle
623 	 */
624 	hw->num_intrs = devRead->intrConf.numIntrs = 1;
625 	devRead->intrConf.intrCtrl |= VMXNET3_IC_DISABLE_ALL;
626 
627 	for (i = 0; i < hw->num_tx_queues; i++) {
628 		Vmxnet3_TxQueueDesc *tqd = &hw->tqd_start[i];
629 		vmxnet3_tx_queue_t *txq  = dev->data->tx_queues[i];
630 
631 		txq->shared = &hw->tqd_start[i];
632 
633 		tqd->ctrl.txNumDeferred  = 0;
634 		tqd->ctrl.txThreshold    = 1;
635 		tqd->conf.txRingBasePA   = txq->cmd_ring.basePA;
636 		tqd->conf.compRingBasePA = txq->comp_ring.basePA;
637 		tqd->conf.dataRingBasePA = txq->data_ring.basePA;
638 
639 		tqd->conf.txRingSize   = txq->cmd_ring.size;
640 		tqd->conf.compRingSize = txq->comp_ring.size;
641 		tqd->conf.dataRingSize = txq->data_ring.size;
642 		tqd->conf.txDataRingDescSize = txq->txdata_desc_size;
643 		tqd->conf.intrIdx      = txq->comp_ring.intr_idx;
644 		tqd->status.stopped    = TRUE;
645 		tqd->status.error      = 0;
646 		memset(&tqd->stats, 0, sizeof(tqd->stats));
647 	}
648 
649 	for (i = 0; i < hw->num_rx_queues; i++) {
650 		Vmxnet3_RxQueueDesc *rqd  = &hw->rqd_start[i];
651 		vmxnet3_rx_queue_t *rxq   = dev->data->rx_queues[i];
652 
653 		rxq->shared = &hw->rqd_start[i];
654 
655 		rqd->conf.rxRingBasePA[0] = rxq->cmd_ring[0].basePA;
656 		rqd->conf.rxRingBasePA[1] = rxq->cmd_ring[1].basePA;
657 		rqd->conf.compRingBasePA  = rxq->comp_ring.basePA;
658 
659 		rqd->conf.rxRingSize[0]   = rxq->cmd_ring[0].size;
660 		rqd->conf.rxRingSize[1]   = rxq->cmd_ring[1].size;
661 		rqd->conf.compRingSize    = rxq->comp_ring.size;
662 		rqd->conf.intrIdx         = rxq->comp_ring.intr_idx;
663 		if (VMXNET3_VERSION_GE_3(hw)) {
664 			rqd->conf.rxDataRingBasePA = rxq->data_ring.basePA;
665 			rqd->conf.rxDataRingDescSize = rxq->data_desc_size;
666 		}
667 		rqd->status.stopped       = TRUE;
668 		rqd->status.error         = 0;
669 		memset(&rqd->stats, 0, sizeof(rqd->stats));
670 	}
671 
672 	/* RxMode set to 0 of VMXNET3_RXM_xxx */
673 	devRead->rxFilterConf.rxMode = 0;
674 
675 	/* Setting up feature flags */
676 	if (rx_offloads & DEV_RX_OFFLOAD_CHECKSUM)
677 		devRead->misc.uptFeatures |= VMXNET3_F_RXCSUM;
678 
679 	if (rx_offloads & DEV_RX_OFFLOAD_TCP_LRO) {
680 		devRead->misc.uptFeatures |= VMXNET3_F_LRO;
681 		devRead->misc.maxNumRxSG = 0;
682 	}
683 
684 	if (port_conf.rxmode.mq_mode == ETH_MQ_RX_RSS) {
685 		ret = vmxnet3_rss_configure(dev);
686 		if (ret != VMXNET3_SUCCESS)
687 			return ret;
688 
689 		devRead->misc.uptFeatures |= VMXNET3_F_RSS;
690 		devRead->rssConfDesc.confVer = 1;
691 		devRead->rssConfDesc.confLen = sizeof(struct VMXNET3_RSSConf);
692 		devRead->rssConfDesc.confPA  = hw->rss_confPA;
693 	}
694 
695 	ret = vmxnet3_dev_vlan_offload_set(dev,
696 			ETH_VLAN_STRIP_MASK | ETH_VLAN_FILTER_MASK);
697 	if (ret)
698 		return ret;
699 
700 	vmxnet3_write_mac(hw, dev->data->mac_addrs->addr_bytes);
701 
702 	return VMXNET3_SUCCESS;
703 }
704 
705 /*
706  * Configure device link speed and setup link.
707  * Must be called after eth_vmxnet3_dev_init. Other wise it might fail
708  * It returns 0 on success.
709  */
710 static int
711 vmxnet3_dev_start(struct rte_eth_dev *dev)
712 {
713 	int ret;
714 	struct vmxnet3_hw *hw = dev->data->dev_private;
715 
716 	PMD_INIT_FUNC_TRACE();
717 
718 	/* Save stats before it is reset by CMD_ACTIVATE */
719 	vmxnet3_hw_stats_save(hw);
720 
721 	ret = vmxnet3_setup_driver_shared(dev);
722 	if (ret != VMXNET3_SUCCESS)
723 		return ret;
724 
725 	/* check if lsc interrupt feature is enabled */
726 	if (dev->data->dev_conf.intr_conf.lsc) {
727 		struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
728 
729 		/* Setup interrupt callback  */
730 		rte_intr_callback_register(&pci_dev->intr_handle,
731 					   vmxnet3_interrupt_handler, dev);
732 
733 		if (rte_intr_enable(&pci_dev->intr_handle) < 0) {
734 			PMD_INIT_LOG(ERR, "interrupt enable failed");
735 			return -EIO;
736 		}
737 	}
738 
739 	/* Exchange shared data with device */
740 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL,
741 			       VMXNET3_GET_ADDR_LO(hw->sharedPA));
742 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAH,
743 			       VMXNET3_GET_ADDR_HI(hw->sharedPA));
744 
745 	/* Activate device by register write */
746 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_ACTIVATE_DEV);
747 	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
748 
749 	if (ret != 0) {
750 		PMD_INIT_LOG(ERR, "Device activation: UNSUCCESSFUL");
751 		return -EINVAL;
752 	}
753 
754 	/* Setup memory region for rx buffers */
755 	ret = vmxnet3_dev_setup_memreg(dev);
756 	if (ret == 0) {
757 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
758 				       VMXNET3_CMD_REGISTER_MEMREGS);
759 		ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
760 		if (ret != 0)
761 			PMD_INIT_LOG(DEBUG,
762 				     "Failed in setup memory region cmd\n");
763 		ret = 0;
764 	} else {
765 		PMD_INIT_LOG(DEBUG, "Failed to setup memory region\n");
766 	}
767 
768 	/* Disable interrupts */
769 	vmxnet3_disable_intr(hw);
770 
771 	/*
772 	 * Load RX queues with blank mbufs and update next2fill index for device
773 	 * Update RxMode of the device
774 	 */
775 	ret = vmxnet3_dev_rxtx_init(dev);
776 	if (ret != VMXNET3_SUCCESS) {
777 		PMD_INIT_LOG(ERR, "Device queue init: UNSUCCESSFUL");
778 		return ret;
779 	}
780 
781 	hw->adapter_stopped = FALSE;
782 
783 	/* Setting proper Rx Mode and issue Rx Mode Update command */
784 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_UCAST | VMXNET3_RXM_BCAST, 1);
785 
786 	if (dev->data->dev_conf.intr_conf.lsc) {
787 		vmxnet3_enable_intr(hw);
788 
789 		/*
790 		 * Update link state from device since this won't be
791 		 * done upon starting with lsc in use. This is done
792 		 * only after enabling interrupts to avoid any race
793 		 * where the link state could change without an
794 		 * interrupt being fired.
795 		 */
796 		__vmxnet3_dev_link_update(dev, 0);
797 	}
798 
799 	return VMXNET3_SUCCESS;
800 }
801 
802 /*
803  * Stop device: disable rx and tx functions to allow for reconfiguring.
804  */
805 static void
806 vmxnet3_dev_stop(struct rte_eth_dev *dev)
807 {
808 	struct rte_eth_link link;
809 	struct vmxnet3_hw *hw = dev->data->dev_private;
810 
811 	PMD_INIT_FUNC_TRACE();
812 
813 	if (hw->adapter_stopped == 1) {
814 		PMD_INIT_LOG(DEBUG, "Device already stopped.");
815 		return;
816 	}
817 
818 	/* disable interrupts */
819 	vmxnet3_disable_intr(hw);
820 
821 	if (dev->data->dev_conf.intr_conf.lsc) {
822 		struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
823 
824 		rte_intr_disable(&pci_dev->intr_handle);
825 
826 		rte_intr_callback_unregister(&pci_dev->intr_handle,
827 					     vmxnet3_interrupt_handler, dev);
828 	}
829 
830 	/* quiesce the device first */
831 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_QUIESCE_DEV);
832 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAL, 0);
833 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_DSAH, 0);
834 
835 	/* reset the device */
836 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_RESET_DEV);
837 	PMD_INIT_LOG(DEBUG, "Device reset.");
838 
839 	vmxnet3_dev_clear_queues(dev);
840 
841 	/* Clear recorded link status */
842 	memset(&link, 0, sizeof(link));
843 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
844 	link.link_speed = ETH_SPEED_NUM_10G;
845 	link.link_autoneg = ETH_LINK_FIXED;
846 	rte_eth_linkstatus_set(dev, &link);
847 
848 	hw->adapter_stopped = 1;
849 }
850 
851 static void
852 vmxnet3_free_queues(struct rte_eth_dev *dev)
853 {
854 	int i;
855 
856 	PMD_INIT_FUNC_TRACE();
857 
858 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
859 		void *rxq = dev->data->rx_queues[i];
860 
861 		vmxnet3_dev_rx_queue_release(rxq);
862 	}
863 	dev->data->nb_rx_queues = 0;
864 
865 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
866 		void *txq = dev->data->tx_queues[i];
867 
868 		vmxnet3_dev_tx_queue_release(txq);
869 	}
870 	dev->data->nb_tx_queues = 0;
871 }
872 
873 /*
874  * Reset and stop device.
875  */
876 static void
877 vmxnet3_dev_close(struct rte_eth_dev *dev)
878 {
879 	PMD_INIT_FUNC_TRACE();
880 
881 	vmxnet3_dev_stop(dev);
882 	vmxnet3_free_queues(dev);
883 }
884 
885 static void
886 vmxnet3_hw_tx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
887 			struct UPT1_TxStats *res)
888 {
889 #define VMXNET3_UPDATE_TX_STAT(h, i, f, r)		\
890 		((r)->f = (h)->tqd_start[(i)].stats.f +	\
891 			(h)->saved_tx_stats[(i)].f)
892 
893 	VMXNET3_UPDATE_TX_STAT(hw, q, ucastPktsTxOK, res);
894 	VMXNET3_UPDATE_TX_STAT(hw, q, mcastPktsTxOK, res);
895 	VMXNET3_UPDATE_TX_STAT(hw, q, bcastPktsTxOK, res);
896 	VMXNET3_UPDATE_TX_STAT(hw, q, ucastBytesTxOK, res);
897 	VMXNET3_UPDATE_TX_STAT(hw, q, mcastBytesTxOK, res);
898 	VMXNET3_UPDATE_TX_STAT(hw, q, bcastBytesTxOK, res);
899 	VMXNET3_UPDATE_TX_STAT(hw, q, pktsTxError, res);
900 	VMXNET3_UPDATE_TX_STAT(hw, q, pktsTxDiscard, res);
901 
902 #undef VMXNET3_UPDATE_TX_STAT
903 }
904 
905 static void
906 vmxnet3_hw_rx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
907 			struct UPT1_RxStats *res)
908 {
909 #define VMXNET3_UPDATE_RX_STAT(h, i, f, r)		\
910 		((r)->f = (h)->rqd_start[(i)].stats.f +	\
911 			(h)->saved_rx_stats[(i)].f)
912 
913 	VMXNET3_UPDATE_RX_STAT(hw, q, ucastPktsRxOK, res);
914 	VMXNET3_UPDATE_RX_STAT(hw, q, mcastPktsRxOK, res);
915 	VMXNET3_UPDATE_RX_STAT(hw, q, bcastPktsRxOK, res);
916 	VMXNET3_UPDATE_RX_STAT(hw, q, ucastBytesRxOK, res);
917 	VMXNET3_UPDATE_RX_STAT(hw, q, mcastBytesRxOK, res);
918 	VMXNET3_UPDATE_RX_STAT(hw, q, bcastBytesRxOK, res);
919 	VMXNET3_UPDATE_RX_STAT(hw, q, pktsRxError, res);
920 	VMXNET3_UPDATE_RX_STAT(hw, q, pktsRxOutOfBuf, res);
921 
922 #undef VMXNET3_UPDATE_RX_STAT
923 }
924 
925 static void
926 vmxnet3_tx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
927 					struct UPT1_TxStats *res)
928 {
929 		vmxnet3_hw_tx_stats_get(hw, q, res);
930 
931 #define VMXNET3_REDUCE_SNAPSHOT_TX_STAT(h, i, f, r)	\
932 		((r)->f -= (h)->snapshot_tx_stats[(i)].f)
933 
934 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, ucastPktsTxOK, res);
935 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, mcastPktsTxOK, res);
936 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, bcastPktsTxOK, res);
937 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, ucastBytesTxOK, res);
938 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, mcastBytesTxOK, res);
939 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, bcastBytesTxOK, res);
940 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, pktsTxError, res);
941 	VMXNET3_REDUCE_SNAPSHOT_TX_STAT(hw, q, pktsTxDiscard, res);
942 
943 #undef VMXNET3_REDUCE_SNAPSHOT_TX_STAT
944 }
945 
946 static void
947 vmxnet3_rx_stats_get(struct vmxnet3_hw *hw, unsigned int q,
948 					struct UPT1_RxStats *res)
949 {
950 		vmxnet3_hw_rx_stats_get(hw, q, res);
951 
952 #define VMXNET3_REDUCE_SNAPSHOT_RX_STAT(h, i, f, r)	\
953 		((r)->f -= (h)->snapshot_rx_stats[(i)].f)
954 
955 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, ucastPktsRxOK, res);
956 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, mcastPktsRxOK, res);
957 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, bcastPktsRxOK, res);
958 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, ucastBytesRxOK, res);
959 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, mcastBytesRxOK, res);
960 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, bcastBytesRxOK, res);
961 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, pktsRxError, res);
962 	VMXNET3_REDUCE_SNAPSHOT_RX_STAT(hw, q, pktsRxOutOfBuf, res);
963 
964 #undef VMXNET3_REDUCE_SNAPSHOT_RX_STAT
965 }
966 
967 static void
968 vmxnet3_hw_stats_save(struct vmxnet3_hw *hw)
969 {
970 	unsigned int i;
971 
972 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
973 
974 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
975 
976 	for (i = 0; i < hw->num_tx_queues; i++)
977 		vmxnet3_hw_tx_stats_get(hw, i, &hw->saved_tx_stats[i]);
978 	for (i = 0; i < hw->num_rx_queues; i++)
979 		vmxnet3_hw_rx_stats_get(hw, i, &hw->saved_rx_stats[i]);
980 }
981 
982 static int
983 vmxnet3_dev_xstats_get_names(struct rte_eth_dev *dev,
984 			     struct rte_eth_xstat_name *xstats_names,
985 			     unsigned int n)
986 {
987 	unsigned int i, t, count = 0;
988 	unsigned int nstats =
989 		dev->data->nb_tx_queues * RTE_DIM(vmxnet3_txq_stat_strings) +
990 		dev->data->nb_rx_queues * RTE_DIM(vmxnet3_rxq_stat_strings);
991 
992 	if (!xstats_names || n < nstats)
993 		return nstats;
994 
995 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
996 		if (!dev->data->rx_queues[i])
997 			continue;
998 
999 		for (t = 0; t < RTE_DIM(vmxnet3_rxq_stat_strings); t++) {
1000 			snprintf(xstats_names[count].name,
1001 				 sizeof(xstats_names[count].name),
1002 				 "rx_q%u_%s", i,
1003 				 vmxnet3_rxq_stat_strings[t].name);
1004 			count++;
1005 		}
1006 	}
1007 
1008 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1009 		if (!dev->data->tx_queues[i])
1010 			continue;
1011 
1012 		for (t = 0; t < RTE_DIM(vmxnet3_txq_stat_strings); t++) {
1013 			snprintf(xstats_names[count].name,
1014 				 sizeof(xstats_names[count].name),
1015 				 "tx_q%u_%s", i,
1016 				 vmxnet3_txq_stat_strings[t].name);
1017 			count++;
1018 		}
1019 	}
1020 
1021 	return count;
1022 }
1023 
1024 static int
1025 vmxnet3_dev_xstats_get(struct rte_eth_dev *dev, struct rte_eth_xstat *xstats,
1026 		       unsigned int n)
1027 {
1028 	unsigned int i, t, count = 0;
1029 	unsigned int nstats =
1030 		dev->data->nb_tx_queues * RTE_DIM(vmxnet3_txq_stat_strings) +
1031 		dev->data->nb_rx_queues * RTE_DIM(vmxnet3_rxq_stat_strings);
1032 
1033 	if (n < nstats)
1034 		return nstats;
1035 
1036 	for (i = 0; i < dev->data->nb_rx_queues; i++) {
1037 		struct vmxnet3_rx_queue *rxq = dev->data->rx_queues[i];
1038 
1039 		if (rxq == NULL)
1040 			continue;
1041 
1042 		for (t = 0; t < RTE_DIM(vmxnet3_rxq_stat_strings); t++) {
1043 			xstats[count].value = *(uint64_t *)(((char *)&rxq->stats) +
1044 				vmxnet3_rxq_stat_strings[t].offset);
1045 			xstats[count].id = count;
1046 			count++;
1047 		}
1048 	}
1049 
1050 	for (i = 0; i < dev->data->nb_tx_queues; i++) {
1051 		struct vmxnet3_tx_queue *txq = dev->data->tx_queues[i];
1052 
1053 		if (txq == NULL)
1054 			continue;
1055 
1056 		for (t = 0; t < RTE_DIM(vmxnet3_txq_stat_strings); t++) {
1057 			xstats[count].value = *(uint64_t *)(((char *)&txq->stats) +
1058 				vmxnet3_txq_stat_strings[t].offset);
1059 			xstats[count].id = count;
1060 			count++;
1061 		}
1062 	}
1063 
1064 	return count;
1065 }
1066 
1067 static int
1068 vmxnet3_dev_stats_get(struct rte_eth_dev *dev, struct rte_eth_stats *stats)
1069 {
1070 	unsigned int i;
1071 	struct vmxnet3_hw *hw = dev->data->dev_private;
1072 	struct UPT1_TxStats txStats;
1073 	struct UPT1_RxStats rxStats;
1074 
1075 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
1076 
1077 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
1078 	for (i = 0; i < hw->num_tx_queues; i++) {
1079 		vmxnet3_tx_stats_get(hw, i, &txStats);
1080 
1081 		stats->q_opackets[i] = txStats.ucastPktsTxOK +
1082 			txStats.mcastPktsTxOK +
1083 			txStats.bcastPktsTxOK;
1084 
1085 		stats->q_obytes[i] = txStats.ucastBytesTxOK +
1086 			txStats.mcastBytesTxOK +
1087 			txStats.bcastBytesTxOK;
1088 
1089 		stats->opackets += stats->q_opackets[i];
1090 		stats->obytes += stats->q_obytes[i];
1091 		stats->oerrors += txStats.pktsTxError + txStats.pktsTxDiscard;
1092 	}
1093 
1094 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_RX_QUEUES);
1095 	for (i = 0; i < hw->num_rx_queues; i++) {
1096 		vmxnet3_rx_stats_get(hw, i, &rxStats);
1097 
1098 		stats->q_ipackets[i] = rxStats.ucastPktsRxOK +
1099 			rxStats.mcastPktsRxOK +
1100 			rxStats.bcastPktsRxOK;
1101 
1102 		stats->q_ibytes[i] = rxStats.ucastBytesRxOK +
1103 			rxStats.mcastBytesRxOK +
1104 			rxStats.bcastBytesRxOK;
1105 
1106 		stats->ipackets += stats->q_ipackets[i];
1107 		stats->ibytes += stats->q_ibytes[i];
1108 
1109 		stats->q_errors[i] = rxStats.pktsRxError;
1110 		stats->ierrors += rxStats.pktsRxError;
1111 		stats->imissed += rxStats.pktsRxOutOfBuf;
1112 	}
1113 
1114 	return 0;
1115 }
1116 
1117 static void
1118 vmxnet3_dev_stats_reset(struct rte_eth_dev *dev)
1119 {
1120 	unsigned int i;
1121 	struct vmxnet3_hw *hw = dev->data->dev_private;
1122 	struct UPT1_TxStats txStats;
1123 	struct UPT1_RxStats rxStats;
1124 
1125 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_STATS);
1126 
1127 	RTE_BUILD_BUG_ON(RTE_ETHDEV_QUEUE_STAT_CNTRS < VMXNET3_MAX_TX_QUEUES);
1128 
1129 	for (i = 0; i < hw->num_tx_queues; i++) {
1130 		vmxnet3_hw_tx_stats_get(hw, i, &txStats);
1131 		memcpy(&hw->snapshot_tx_stats[i], &txStats,
1132 			sizeof(hw->snapshot_tx_stats[0]));
1133 	}
1134 	for (i = 0; i < hw->num_rx_queues; i++) {
1135 		vmxnet3_hw_rx_stats_get(hw, i, &rxStats);
1136 		memcpy(&hw->snapshot_rx_stats[i], &rxStats,
1137 			sizeof(hw->snapshot_rx_stats[0]));
1138 	}
1139 }
1140 
1141 static void
1142 vmxnet3_dev_info_get(struct rte_eth_dev *dev __rte_unused,
1143 		     struct rte_eth_dev_info *dev_info)
1144 {
1145 	dev_info->max_rx_queues = VMXNET3_MAX_RX_QUEUES;
1146 	dev_info->max_tx_queues = VMXNET3_MAX_TX_QUEUES;
1147 	dev_info->min_rx_bufsize = 1518 + RTE_PKTMBUF_HEADROOM;
1148 	dev_info->max_rx_pktlen = 16384; /* includes CRC, cf MAXFRS register */
1149 	dev_info->speed_capa = ETH_LINK_SPEED_10G;
1150 	dev_info->max_mac_addrs = VMXNET3_MAX_MAC_ADDRS;
1151 
1152 	dev_info->flow_type_rss_offloads = VMXNET3_RSS_OFFLOAD_ALL;
1153 
1154 	dev_info->rx_desc_lim = (struct rte_eth_desc_lim) {
1155 		.nb_max = VMXNET3_RX_RING_MAX_SIZE,
1156 		.nb_min = VMXNET3_DEF_RX_RING_SIZE,
1157 		.nb_align = 1,
1158 	};
1159 
1160 	dev_info->tx_desc_lim = (struct rte_eth_desc_lim) {
1161 		.nb_max = VMXNET3_TX_RING_MAX_SIZE,
1162 		.nb_min = VMXNET3_DEF_TX_RING_SIZE,
1163 		.nb_align = 1,
1164 		.nb_seg_max = VMXNET3_TX_MAX_SEG,
1165 		.nb_mtu_seg_max = VMXNET3_MAX_TXD_PER_PKT,
1166 	};
1167 
1168 	dev_info->rx_offload_capa = VMXNET3_RX_OFFLOAD_CAP;
1169 	dev_info->rx_queue_offload_capa = 0;
1170 	dev_info->tx_offload_capa = VMXNET3_TX_OFFLOAD_CAP;
1171 	dev_info->tx_queue_offload_capa = 0;
1172 }
1173 
1174 static const uint32_t *
1175 vmxnet3_dev_supported_ptypes_get(struct rte_eth_dev *dev)
1176 {
1177 	static const uint32_t ptypes[] = {
1178 		RTE_PTYPE_L3_IPV4_EXT,
1179 		RTE_PTYPE_L3_IPV4,
1180 		RTE_PTYPE_UNKNOWN
1181 	};
1182 
1183 	if (dev->rx_pkt_burst == vmxnet3_recv_pkts)
1184 		return ptypes;
1185 	return NULL;
1186 }
1187 
1188 static int
1189 vmxnet3_mac_addr_set(struct rte_eth_dev *dev, struct ether_addr *mac_addr)
1190 {
1191 	struct vmxnet3_hw *hw = dev->data->dev_private;
1192 
1193 	ether_addr_copy(mac_addr, (struct ether_addr *)(hw->perm_addr));
1194 	vmxnet3_write_mac(hw, mac_addr->addr_bytes);
1195 	return 0;
1196 }
1197 
1198 /* return 0 means link status changed, -1 means not changed */
1199 static int
1200 __vmxnet3_dev_link_update(struct rte_eth_dev *dev,
1201 			  __rte_unused int wait_to_complete)
1202 {
1203 	struct vmxnet3_hw *hw = dev->data->dev_private;
1204 	struct rte_eth_link link;
1205 	uint32_t ret;
1206 
1207 	memset(&link, 0, sizeof(link));
1208 
1209 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_GET_LINK);
1210 	ret = VMXNET3_READ_BAR1_REG(hw, VMXNET3_REG_CMD);
1211 
1212 	if (ret & 0x1)
1213 		link.link_status = ETH_LINK_UP;
1214 	link.link_duplex = ETH_LINK_FULL_DUPLEX;
1215 	link.link_speed = ETH_SPEED_NUM_10G;
1216 	link.link_autoneg = ETH_LINK_FIXED;
1217 
1218 	return rte_eth_linkstatus_set(dev, &link);
1219 }
1220 
1221 static int
1222 vmxnet3_dev_link_update(struct rte_eth_dev *dev, int wait_to_complete)
1223 {
1224 	/* Link status doesn't change for stopped dev */
1225 	if (dev->data->dev_started == 0)
1226 		return -1;
1227 
1228 	return __vmxnet3_dev_link_update(dev, wait_to_complete);
1229 }
1230 
1231 /* Updating rxmode through Vmxnet3_DriverShared structure in adapter */
1232 static void
1233 vmxnet3_dev_set_rxmode(struct vmxnet3_hw *hw, uint32_t feature, int set)
1234 {
1235 	struct Vmxnet3_RxFilterConf *rxConf = &hw->shared->devRead.rxFilterConf;
1236 
1237 	if (set)
1238 		rxConf->rxMode = rxConf->rxMode | feature;
1239 	else
1240 		rxConf->rxMode = rxConf->rxMode & (~feature);
1241 
1242 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD, VMXNET3_CMD_UPDATE_RX_MODE);
1243 }
1244 
1245 /* Promiscuous supported only if Vmxnet3_DriverShared is initialized in adapter */
1246 static void
1247 vmxnet3_dev_promiscuous_enable(struct rte_eth_dev *dev)
1248 {
1249 	struct vmxnet3_hw *hw = dev->data->dev_private;
1250 	uint32_t *vf_table = hw->shared->devRead.rxFilterConf.vfTable;
1251 
1252 	memset(vf_table, 0, VMXNET3_VFT_TABLE_SIZE);
1253 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 1);
1254 
1255 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1256 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1257 }
1258 
1259 /* Promiscuous supported only if Vmxnet3_DriverShared is initialized in adapter */
1260 static void
1261 vmxnet3_dev_promiscuous_disable(struct rte_eth_dev *dev)
1262 {
1263 	struct vmxnet3_hw *hw = dev->data->dev_private;
1264 	uint32_t *vf_table = hw->shared->devRead.rxFilterConf.vfTable;
1265 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1266 
1267 	if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1268 		memcpy(vf_table, hw->shadow_vfta, VMXNET3_VFT_TABLE_SIZE);
1269 	else
1270 		memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);
1271 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_PROMISC, 0);
1272 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1273 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1274 }
1275 
1276 /* Allmulticast supported only if Vmxnet3_DriverShared is initialized in adapter */
1277 static void
1278 vmxnet3_dev_allmulticast_enable(struct rte_eth_dev *dev)
1279 {
1280 	struct vmxnet3_hw *hw = dev->data->dev_private;
1281 
1282 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_ALL_MULTI, 1);
1283 }
1284 
1285 /* Allmulticast supported only if Vmxnet3_DriverShared is initialized in adapter */
1286 static void
1287 vmxnet3_dev_allmulticast_disable(struct rte_eth_dev *dev)
1288 {
1289 	struct vmxnet3_hw *hw = dev->data->dev_private;
1290 
1291 	vmxnet3_dev_set_rxmode(hw, VMXNET3_RXM_ALL_MULTI, 0);
1292 }
1293 
1294 /* Enable/disable filter on vlan */
1295 static int
1296 vmxnet3_dev_vlan_filter_set(struct rte_eth_dev *dev, uint16_t vid, int on)
1297 {
1298 	struct vmxnet3_hw *hw = dev->data->dev_private;
1299 	struct Vmxnet3_RxFilterConf *rxConf = &hw->shared->devRead.rxFilterConf;
1300 	uint32_t *vf_table = rxConf->vfTable;
1301 
1302 	/* save state for restore */
1303 	if (on)
1304 		VMXNET3_SET_VFTABLE_ENTRY(hw->shadow_vfta, vid);
1305 	else
1306 		VMXNET3_CLEAR_VFTABLE_ENTRY(hw->shadow_vfta, vid);
1307 
1308 	/* don't change active filter if in promiscuous mode */
1309 	if (rxConf->rxMode & VMXNET3_RXM_PROMISC)
1310 		return 0;
1311 
1312 	/* set in hardware */
1313 	if (on)
1314 		VMXNET3_SET_VFTABLE_ENTRY(vf_table, vid);
1315 	else
1316 		VMXNET3_CLEAR_VFTABLE_ENTRY(vf_table, vid);
1317 
1318 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1319 			       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1320 	return 0;
1321 }
1322 
1323 static int
1324 vmxnet3_dev_vlan_offload_set(struct rte_eth_dev *dev, int mask)
1325 {
1326 	struct vmxnet3_hw *hw = dev->data->dev_private;
1327 	Vmxnet3_DSDevRead *devRead = &hw->shared->devRead;
1328 	uint32_t *vf_table = devRead->rxFilterConf.vfTable;
1329 	uint64_t rx_offloads = dev->data->dev_conf.rxmode.offloads;
1330 
1331 	if (mask & ETH_VLAN_STRIP_MASK) {
1332 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_STRIP)
1333 			devRead->misc.uptFeatures |= UPT1_F_RXVLAN;
1334 		else
1335 			devRead->misc.uptFeatures &= ~UPT1_F_RXVLAN;
1336 
1337 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1338 				       VMXNET3_CMD_UPDATE_FEATURE);
1339 	}
1340 
1341 	if (mask & ETH_VLAN_FILTER_MASK) {
1342 		if (rx_offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
1343 			memcpy(vf_table, hw->shadow_vfta, VMXNET3_VFT_TABLE_SIZE);
1344 		else
1345 			memset(vf_table, 0xff, VMXNET3_VFT_TABLE_SIZE);
1346 
1347 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1348 				       VMXNET3_CMD_UPDATE_VLAN_FILTERS);
1349 	}
1350 
1351 	return 0;
1352 }
1353 
1354 static void
1355 vmxnet3_process_events(struct rte_eth_dev *dev)
1356 {
1357 	struct vmxnet3_hw *hw = dev->data->dev_private;
1358 	uint32_t events = hw->shared->ecr;
1359 
1360 	if (!events)
1361 		return;
1362 
1363 	/*
1364 	 * ECR bits when written with 1b are cleared. Hence write
1365 	 * events back to ECR so that the bits which were set will be reset.
1366 	 */
1367 	VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_ECR, events);
1368 
1369 	/* Check if link state has changed */
1370 	if (events & VMXNET3_ECR_LINK) {
1371 		PMD_DRV_LOG(DEBUG, "Process events: VMXNET3_ECR_LINK event");
1372 		if (vmxnet3_dev_link_update(dev, 0) == 0)
1373 			_rte_eth_dev_callback_process(dev,
1374 						      RTE_ETH_EVENT_INTR_LSC,
1375 						      NULL);
1376 	}
1377 
1378 	/* Check if there is an error on xmit/recv queues */
1379 	if (events & (VMXNET3_ECR_TQERR | VMXNET3_ECR_RQERR)) {
1380 		VMXNET3_WRITE_BAR1_REG(hw, VMXNET3_REG_CMD,
1381 				       VMXNET3_CMD_GET_QUEUE_STATUS);
1382 
1383 		if (hw->tqd_start->status.stopped)
1384 			PMD_DRV_LOG(ERR, "tq error 0x%x",
1385 				    hw->tqd_start->status.error);
1386 
1387 		if (hw->rqd_start->status.stopped)
1388 			PMD_DRV_LOG(ERR, "rq error 0x%x",
1389 				     hw->rqd_start->status.error);
1390 
1391 		/* Reset the device */
1392 		/* Have to reset the device */
1393 	}
1394 
1395 	if (events & VMXNET3_ECR_DIC)
1396 		PMD_DRV_LOG(DEBUG, "Device implementation change event.");
1397 
1398 	if (events & VMXNET3_ECR_DEBUG)
1399 		PMD_DRV_LOG(DEBUG, "Debug event generated by device.");
1400 }
1401 
1402 static void
1403 vmxnet3_interrupt_handler(void *param)
1404 {
1405 	struct rte_eth_dev *dev = param;
1406 	struct rte_pci_device *pci_dev = RTE_DEV_TO_PCI(dev->device);
1407 
1408 	vmxnet3_process_events(dev);
1409 
1410 	if (rte_intr_enable(&pci_dev->intr_handle) < 0)
1411 		PMD_DRV_LOG(ERR, "interrupt enable failed");
1412 }
1413 
1414 RTE_PMD_REGISTER_PCI(net_vmxnet3, rte_vmxnet3_pmd);
1415 RTE_PMD_REGISTER_PCI_TABLE(net_vmxnet3, pci_id_vmxnet3_map);
1416 RTE_PMD_REGISTER_KMOD_DEP(net_vmxnet3, "* igb_uio | uio_pci_generic | vfio-pci");
1417 
1418 RTE_INIT(vmxnet3_init_log)
1419 {
1420 	vmxnet3_logtype_init = rte_log_register("pmd.net.vmxnet3.init");
1421 	if (vmxnet3_logtype_init >= 0)
1422 		rte_log_set_level(vmxnet3_logtype_init, RTE_LOG_NOTICE);
1423 	vmxnet3_logtype_driver = rte_log_register("pmd.net.vmxnet3.driver");
1424 	if (vmxnet3_logtype_driver >= 0)
1425 		rte_log_set_level(vmxnet3_logtype_driver, RTE_LOG_NOTICE);
1426 }
1427