xref: /dpdk/drivers/net/ixgbe/ixgbe_pf.c (revision 4e2cf9af)
1 /* SPDX-License-Identifier: BSD-3-Clause
2  * Copyright(c) 2010-2016 Intel Corporation
3  */
4 
5 #include <stdio.h>
6 #include <errno.h>
7 #include <stdint.h>
8 #include <stdlib.h>
9 #include <unistd.h>
10 #include <stdarg.h>
11 #include <inttypes.h>
12 
13 #include <rte_interrupts.h>
14 #include <rte_log.h>
15 #include <rte_debug.h>
16 #include <rte_eal.h>
17 #include <rte_ether.h>
18 #include <rte_ethdev_driver.h>
19 #include <rte_memcpy.h>
20 #include <rte_malloc.h>
21 #include <rte_random.h>
22 
23 #include "base/ixgbe_common.h"
24 #include "ixgbe_ethdev.h"
25 #include "rte_pmd_ixgbe.h"
26 
27 #define IXGBE_MAX_VFTA     (128)
28 #define IXGBE_VF_MSG_SIZE_DEFAULT 1
29 #define IXGBE_VF_GET_QUEUE_MSG_SIZE 5
30 #define IXGBE_ETHERTYPE_FLOW_CTRL 0x8808
31 
32 static inline uint16_t
33 dev_num_vf(struct rte_eth_dev *eth_dev)
34 {
35 	struct rte_pci_device *pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
36 
37 	return pci_dev->max_vfs;
38 }
39 
40 static inline
41 int ixgbe_vf_perm_addr_gen(struct rte_eth_dev *dev, uint16_t vf_num)
42 {
43 	unsigned char vf_mac_addr[RTE_ETHER_ADDR_LEN];
44 	struct ixgbe_vf_info *vfinfo =
45 		*IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
46 	uint16_t vfn;
47 
48 	for (vfn = 0; vfn < vf_num; vfn++) {
49 		rte_eth_random_addr(vf_mac_addr);
50 		/* keep the random address as default */
51 		memcpy(vfinfo[vfn].vf_mac_addresses, vf_mac_addr,
52 			   RTE_ETHER_ADDR_LEN);
53 	}
54 
55 	return 0;
56 }
57 
58 static inline int
59 ixgbe_mb_intr_setup(struct rte_eth_dev *dev)
60 {
61 	struct ixgbe_interrupt *intr =
62 		IXGBE_DEV_PRIVATE_TO_INTR(dev->data->dev_private);
63 
64 	intr->mask |= IXGBE_EICR_MAILBOX;
65 
66 	return 0;
67 }
68 
69 void ixgbe_pf_host_init(struct rte_eth_dev *eth_dev)
70 {
71 	struct ixgbe_vf_info **vfinfo =
72 		IXGBE_DEV_PRIVATE_TO_P_VFDATA(eth_dev->data->dev_private);
73 	struct ixgbe_mirror_info *mirror_info =
74 	IXGBE_DEV_PRIVATE_TO_PFDATA(eth_dev->data->dev_private);
75 	struct ixgbe_uta_info *uta_info =
76 	IXGBE_DEV_PRIVATE_TO_UTA(eth_dev->data->dev_private);
77 	struct ixgbe_hw *hw =
78 		IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
79 	uint16_t vf_num;
80 	uint8_t nb_queue;
81 
82 	PMD_INIT_FUNC_TRACE();
83 
84 	RTE_ETH_DEV_SRIOV(eth_dev).active = 0;
85 	vf_num = dev_num_vf(eth_dev);
86 	if (vf_num == 0)
87 		return;
88 
89 	*vfinfo = rte_zmalloc("vf_info", sizeof(struct ixgbe_vf_info) * vf_num, 0);
90 	if (*vfinfo == NULL)
91 		rte_panic("Cannot allocate memory for private VF data\n");
92 
93 	rte_eth_switch_domain_alloc(&(*vfinfo)->switch_domain_id);
94 
95 	memset(mirror_info, 0, sizeof(struct ixgbe_mirror_info));
96 	memset(uta_info, 0, sizeof(struct ixgbe_uta_info));
97 	hw->mac.mc_filter_type = 0;
98 
99 	if (vf_num >= ETH_32_POOLS) {
100 		nb_queue = 2;
101 		RTE_ETH_DEV_SRIOV(eth_dev).active = ETH_64_POOLS;
102 	} else if (vf_num >= ETH_16_POOLS) {
103 		nb_queue = 4;
104 		RTE_ETH_DEV_SRIOV(eth_dev).active = ETH_32_POOLS;
105 	} else {
106 		nb_queue = 8;
107 		RTE_ETH_DEV_SRIOV(eth_dev).active = ETH_16_POOLS;
108 	}
109 
110 	RTE_ETH_DEV_SRIOV(eth_dev).nb_q_per_pool = nb_queue;
111 	RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx = vf_num;
112 	RTE_ETH_DEV_SRIOV(eth_dev).def_pool_q_idx = (uint16_t)(vf_num * nb_queue);
113 
114 	ixgbe_vf_perm_addr_gen(eth_dev, vf_num);
115 
116 	/* init_mailbox_params */
117 	hw->mbx.ops.init_params(hw);
118 
119 	/* set mb interrupt mask */
120 	ixgbe_mb_intr_setup(eth_dev);
121 }
122 
123 void ixgbe_pf_host_uninit(struct rte_eth_dev *eth_dev)
124 {
125 	struct ixgbe_vf_info **vfinfo;
126 	uint16_t vf_num;
127 	int ret;
128 
129 	PMD_INIT_FUNC_TRACE();
130 
131 	RTE_ETH_DEV_SRIOV(eth_dev).active = 0;
132 	RTE_ETH_DEV_SRIOV(eth_dev).nb_q_per_pool = 0;
133 	RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx = 0;
134 	RTE_ETH_DEV_SRIOV(eth_dev).def_pool_q_idx = 0;
135 
136 	vf_num = dev_num_vf(eth_dev);
137 	if (vf_num == 0)
138 		return;
139 
140 	vfinfo = IXGBE_DEV_PRIVATE_TO_P_VFDATA(eth_dev->data->dev_private);
141 	if (*vfinfo == NULL)
142 		return;
143 
144 	ret = rte_eth_switch_domain_free((*vfinfo)->switch_domain_id);
145 	if (ret)
146 		PMD_INIT_LOG(WARNING, "failed to free switch domain: %d", ret);
147 
148 	rte_free(*vfinfo);
149 	*vfinfo = NULL;
150 }
151 
152 static void
153 ixgbe_add_tx_flow_control_drop_filter(struct rte_eth_dev *eth_dev)
154 {
155 	struct ixgbe_hw *hw =
156 		IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
157 	struct ixgbe_filter_info *filter_info =
158 		IXGBE_DEV_PRIVATE_TO_FILTER_INFO(eth_dev->data->dev_private);
159 	uint16_t vf_num;
160 	int i;
161 	struct ixgbe_ethertype_filter ethertype_filter;
162 
163 	if (!hw->mac.ops.set_ethertype_anti_spoofing) {
164 		PMD_DRV_LOG(INFO, "ether type anti-spoofing is not supported.\n");
165 		return;
166 	}
167 
168 	i = ixgbe_ethertype_filter_lookup(filter_info,
169 					  IXGBE_ETHERTYPE_FLOW_CTRL);
170 	if (i >= 0) {
171 		PMD_DRV_LOG(ERR, "A ether type filter entity for flow control already exists!\n");
172 		return;
173 	}
174 
175 	ethertype_filter.ethertype = IXGBE_ETHERTYPE_FLOW_CTRL;
176 	ethertype_filter.etqf = IXGBE_ETQF_FILTER_EN |
177 				IXGBE_ETQF_TX_ANTISPOOF |
178 				IXGBE_ETHERTYPE_FLOW_CTRL;
179 	ethertype_filter.etqs = 0;
180 	ethertype_filter.conf = TRUE;
181 	i = ixgbe_ethertype_filter_insert(filter_info,
182 					  &ethertype_filter);
183 	if (i < 0) {
184 		PMD_DRV_LOG(ERR, "Cannot find an unused ether type filter entity for flow control.\n");
185 		return;
186 	}
187 
188 	IXGBE_WRITE_REG(hw, IXGBE_ETQF(i),
189 			(IXGBE_ETQF_FILTER_EN |
190 			IXGBE_ETQF_TX_ANTISPOOF |
191 			IXGBE_ETHERTYPE_FLOW_CTRL));
192 
193 	vf_num = dev_num_vf(eth_dev);
194 	for (i = 0; i < vf_num; i++)
195 		hw->mac.ops.set_ethertype_anti_spoofing(hw, true, i);
196 }
197 
198 int ixgbe_pf_host_configure(struct rte_eth_dev *eth_dev)
199 {
200 	uint32_t vtctl, fcrth;
201 	uint32_t vfre_slot, vfre_offset;
202 	uint16_t vf_num;
203 	const uint8_t VFRE_SHIFT = 5;  /* VFRE 32 bits per slot */
204 	const uint8_t VFRE_MASK = (uint8_t)((1U << VFRE_SHIFT) - 1);
205 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
206 	uint32_t gpie, gcr_ext;
207 	uint32_t vlanctrl;
208 	int i;
209 
210 	vf_num = dev_num_vf(eth_dev);
211 	if (vf_num == 0)
212 		return -1;
213 
214 	/* enable VMDq and set the default pool for PF */
215 	vtctl = IXGBE_READ_REG(hw, IXGBE_VT_CTL);
216 	vtctl |= IXGBE_VMD_CTL_VMDQ_EN;
217 	vtctl &= ~IXGBE_VT_CTL_POOL_MASK;
218 	vtctl |= RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx
219 		<< IXGBE_VT_CTL_POOL_SHIFT;
220 	vtctl |= IXGBE_VT_CTL_REPLEN;
221 	IXGBE_WRITE_REG(hw, IXGBE_VT_CTL, vtctl);
222 
223 	vfre_offset = vf_num & VFRE_MASK;
224 	vfre_slot = (vf_num >> VFRE_SHIFT) > 0 ? 1 : 0;
225 
226 	/* Enable pools reserved to PF only */
227 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(vfre_slot), (~0U) << vfre_offset);
228 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(vfre_slot ^ 1), vfre_slot - 1);
229 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(vfre_slot), (~0U) << vfre_offset);
230 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(vfre_slot ^ 1), vfre_slot - 1);
231 
232 	/* PFDMA Tx General Switch Control Enables VMDQ loopback */
233 	IXGBE_WRITE_REG(hw, IXGBE_PFDTXGSWC, IXGBE_PFDTXGSWC_VT_LBEN);
234 
235 	/* clear VMDq map to perment rar 0 */
236 	hw->mac.ops.clear_vmdq(hw, 0, IXGBE_CLEAR_VMDQ_ALL);
237 
238 	/* clear VMDq map to scan rar 127 */
239 	IXGBE_WRITE_REG(hw, IXGBE_MPSAR_LO(hw->mac.num_rar_entries), 0);
240 	IXGBE_WRITE_REG(hw, IXGBE_MPSAR_HI(hw->mac.num_rar_entries), 0);
241 
242 	/* set VMDq map to default PF pool */
243 	hw->mac.ops.set_vmdq(hw, 0, RTE_ETH_DEV_SRIOV(eth_dev).def_vmdq_idx);
244 
245 	/*
246 	 * SW msut set GCR_EXT.VT_Mode the same as GPIE.VT_Mode
247 	 */
248 	gcr_ext = IXGBE_READ_REG(hw, IXGBE_GCR_EXT);
249 	gcr_ext &= ~IXGBE_GCR_EXT_VT_MODE_MASK;
250 
251 	gpie = IXGBE_READ_REG(hw, IXGBE_GPIE);
252 	gpie &= ~IXGBE_GPIE_VTMODE_MASK;
253 	gpie |= IXGBE_GPIE_MSIX_MODE | IXGBE_GPIE_PBA_SUPPORT;
254 
255 	switch (RTE_ETH_DEV_SRIOV(eth_dev).active) {
256 	case ETH_64_POOLS:
257 		gcr_ext |= IXGBE_GCR_EXT_VT_MODE_64;
258 		gpie |= IXGBE_GPIE_VTMODE_64;
259 		break;
260 	case ETH_32_POOLS:
261 		gcr_ext |= IXGBE_GCR_EXT_VT_MODE_32;
262 		gpie |= IXGBE_GPIE_VTMODE_32;
263 		break;
264 	case ETH_16_POOLS:
265 		gcr_ext |= IXGBE_GCR_EXT_VT_MODE_16;
266 		gpie |= IXGBE_GPIE_VTMODE_16;
267 		break;
268 	}
269 
270 	IXGBE_WRITE_REG(hw, IXGBE_GCR_EXT, gcr_ext);
271 	IXGBE_WRITE_REG(hw, IXGBE_GPIE, gpie);
272 
273 	/*
274 	 * enable vlan filtering and allow all vlan tags through
275 	 */
276 	vlanctrl = IXGBE_READ_REG(hw, IXGBE_VLNCTRL);
277 	vlanctrl |= IXGBE_VLNCTRL_VFE; /* enable vlan filters */
278 	IXGBE_WRITE_REG(hw, IXGBE_VLNCTRL, vlanctrl);
279 
280 	/* VFTA - enable all vlan filters */
281 	for (i = 0; i < IXGBE_MAX_VFTA; i++)
282 		IXGBE_WRITE_REG(hw, IXGBE_VFTA(i), 0xFFFFFFFF);
283 
284 	/* Enable MAC Anti-Spoofing */
285 	hw->mac.ops.set_mac_anti_spoofing(hw, FALSE, vf_num);
286 
287 	/* set flow control threshold to max to avoid tx switch hang */
288 	for (i = 0; i < IXGBE_DCB_MAX_TRAFFIC_CLASS; i++) {
289 		IXGBE_WRITE_REG(hw, IXGBE_FCRTL_82599(i), 0);
290 		fcrth = IXGBE_READ_REG(hw, IXGBE_RXPBSIZE(i)) - 32;
291 		IXGBE_WRITE_REG(hw, IXGBE_FCRTH_82599(i), fcrth);
292 	}
293 
294 	ixgbe_add_tx_flow_control_drop_filter(eth_dev);
295 
296 	return 0;
297 }
298 
299 static void
300 set_rx_mode(struct rte_eth_dev *dev)
301 {
302 	struct rte_eth_dev_data *dev_data = dev->data;
303 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
304 	u32 fctrl, vmolr = IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE;
305 	uint16_t vfn = dev_num_vf(dev);
306 
307 	/* Check for Promiscuous and All Multicast modes */
308 	fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
309 
310 	/* set all bits that we expect to always be set */
311 	fctrl &= ~IXGBE_FCTRL_SBP; /* disable store-bad-packets */
312 	fctrl |= IXGBE_FCTRL_BAM;
313 
314 	/* clear the bits we are changing the status of */
315 	fctrl &= ~(IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
316 
317 	if (dev_data->promiscuous) {
318 		fctrl |= (IXGBE_FCTRL_UPE | IXGBE_FCTRL_MPE);
319 		vmolr |= (IXGBE_VMOLR_ROPE | IXGBE_VMOLR_MPE);
320 	} else {
321 		if (dev_data->all_multicast) {
322 			fctrl |= IXGBE_FCTRL_MPE;
323 			vmolr |= IXGBE_VMOLR_MPE;
324 		} else {
325 			vmolr |= IXGBE_VMOLR_ROMPE;
326 		}
327 	}
328 
329 	if (hw->mac.type != ixgbe_mac_82598EB) {
330 		vmolr |= IXGBE_READ_REG(hw, IXGBE_VMOLR(vfn)) &
331 			 ~(IXGBE_VMOLR_MPE | IXGBE_VMOLR_ROMPE |
332 			   IXGBE_VMOLR_ROPE);
333 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vfn), vmolr);
334 	}
335 
336 	IXGBE_WRITE_REG(hw, IXGBE_FCTRL, fctrl);
337 
338 	ixgbe_vlan_hw_strip_config(dev);
339 }
340 
341 static inline void
342 ixgbe_vf_reset_event(struct rte_eth_dev *dev, uint16_t vf)
343 {
344 	struct ixgbe_hw *hw =
345 		IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
346 	struct ixgbe_vf_info *vfinfo =
347 		*(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
348 	int rar_entry = hw->mac.num_rar_entries - (vf + 1);
349 	uint32_t vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
350 
351 	vmolr |= (IXGBE_VMOLR_ROPE |
352 			IXGBE_VMOLR_BAM | IXGBE_VMOLR_AUPE);
353 	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
354 
355 	IXGBE_WRITE_REG(hw, IXGBE_VMVIR(vf), 0);
356 
357 	/* reset multicast table array for vf */
358 	vfinfo[vf].num_vf_mc_hashes = 0;
359 
360 	/* reset rx mode */
361 	set_rx_mode(dev);
362 
363 	hw->mac.ops.clear_rar(hw, rar_entry);
364 }
365 
366 static inline void
367 ixgbe_vf_reset_msg(struct rte_eth_dev *dev, uint16_t vf)
368 {
369 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
370 	uint32_t reg;
371 	uint32_t reg_offset, vf_shift;
372 	const uint8_t VFRE_SHIFT = 5;  /* VFRE 32 bits per slot */
373 	const uint8_t VFRE_MASK = (uint8_t)((1U << VFRE_SHIFT) - 1);
374 	uint8_t  nb_q_per_pool;
375 	int i;
376 
377 	vf_shift = vf & VFRE_MASK;
378 	reg_offset = (vf >> VFRE_SHIFT) > 0 ? 1 : 0;
379 
380 	/* enable transmit for vf */
381 	reg = IXGBE_READ_REG(hw, IXGBE_VFTE(reg_offset));
382 	reg |= (reg | (1 << vf_shift));
383 	IXGBE_WRITE_REG(hw, IXGBE_VFTE(reg_offset), reg);
384 
385 	/* enable all queue drop for IOV */
386 	nb_q_per_pool = RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
387 	for (i = vf * nb_q_per_pool; i < (vf + 1) * nb_q_per_pool; i++) {
388 		IXGBE_WRITE_FLUSH(hw);
389 		reg = IXGBE_QDE_ENABLE | IXGBE_QDE_WRITE;
390 		reg |= i << IXGBE_QDE_IDX_SHIFT;
391 		IXGBE_WRITE_REG(hw, IXGBE_QDE, reg);
392 	}
393 
394 	/* enable receive for vf */
395 	reg = IXGBE_READ_REG(hw, IXGBE_VFRE(reg_offset));
396 	reg |= (reg | (1 << vf_shift));
397 	IXGBE_WRITE_REG(hw, IXGBE_VFRE(reg_offset), reg);
398 
399 	/* Enable counting of spoofed packets in the SSVPC register */
400 	reg = IXGBE_READ_REG(hw, IXGBE_VMECM(reg_offset));
401 	reg |= (1 << vf_shift);
402 	IXGBE_WRITE_REG(hw, IXGBE_VMECM(reg_offset), reg);
403 
404 	ixgbe_vf_reset_event(dev, vf);
405 }
406 
407 static int
408 ixgbe_disable_vf_mc_promisc(struct rte_eth_dev *dev, uint32_t vf)
409 {
410 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
411 	uint32_t vmolr;
412 
413 	vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
414 
415 	PMD_DRV_LOG(INFO, "VF %u: disabling multicast promiscuous\n", vf);
416 
417 	vmolr &= ~IXGBE_VMOLR_MPE;
418 
419 	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
420 
421 	return 0;
422 }
423 
424 static int
425 ixgbe_vf_reset(struct rte_eth_dev *dev, uint16_t vf, uint32_t *msgbuf)
426 {
427 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
428 	struct ixgbe_vf_info *vfinfo =
429 		*(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
430 	unsigned char *vf_mac = vfinfo[vf].vf_mac_addresses;
431 	int rar_entry = hw->mac.num_rar_entries - (vf + 1);
432 	uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
433 
434 	ixgbe_vf_reset_msg(dev, vf);
435 
436 	hw->mac.ops.set_rar(hw, rar_entry, vf_mac, vf, IXGBE_RAH_AV);
437 
438 	/* Disable multicast promiscuous at reset */
439 	ixgbe_disable_vf_mc_promisc(dev, vf);
440 
441 	/* reply to reset with ack and vf mac address */
442 	msgbuf[0] = IXGBE_VF_RESET | IXGBE_VT_MSGTYPE_ACK;
443 	rte_memcpy(new_mac, vf_mac, RTE_ETHER_ADDR_LEN);
444 	/*
445 	 * Piggyback the multicast filter type so VF can compute the
446 	 * correct vectors
447 	 */
448 	msgbuf[3] = hw->mac.mc_filter_type;
449 	ixgbe_write_mbx(hw, msgbuf, IXGBE_VF_PERMADDR_MSG_LEN, vf);
450 
451 	return 0;
452 }
453 
454 static int
455 ixgbe_vf_set_mac_addr(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
456 {
457 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
458 	struct ixgbe_vf_info *vfinfo =
459 		*(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
460 	int rar_entry = hw->mac.num_rar_entries - (vf + 1);
461 	uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
462 
463 	if (rte_is_valid_assigned_ether_addr(
464 			(struct rte_ether_addr *)new_mac)) {
465 		rte_memcpy(vfinfo[vf].vf_mac_addresses, new_mac, 6);
466 		return hw->mac.ops.set_rar(hw, rar_entry, new_mac, vf, IXGBE_RAH_AV);
467 	}
468 	return -1;
469 }
470 
471 static int
472 ixgbe_vf_set_multicast(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
473 {
474 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
475 	struct ixgbe_vf_info *vfinfo =
476 		*(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
477 	int nb_entries = (msgbuf[0] & IXGBE_VT_MSGINFO_MASK) >>
478 		IXGBE_VT_MSGINFO_SHIFT;
479 	uint16_t *hash_list = (uint16_t *)&msgbuf[1];
480 	uint32_t mta_idx;
481 	uint32_t mta_shift;
482 	const uint32_t IXGBE_MTA_INDEX_MASK = 0x7F;
483 	const uint32_t IXGBE_MTA_BIT_SHIFT = 5;
484 	const uint32_t IXGBE_MTA_BIT_MASK = (0x1 << IXGBE_MTA_BIT_SHIFT) - 1;
485 	uint32_t reg_val;
486 	int i;
487 	u32 vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
488 
489 	/* Disable multicast promiscuous first */
490 	ixgbe_disable_vf_mc_promisc(dev, vf);
491 
492 	/* only so many hash values supported */
493 	nb_entries = RTE_MIN(nb_entries, IXGBE_MAX_VF_MC_ENTRIES);
494 
495 	/* store the mc entries  */
496 	vfinfo->num_vf_mc_hashes = (uint16_t)nb_entries;
497 	for (i = 0; i < nb_entries; i++) {
498 		vfinfo->vf_mc_hashes[i] = hash_list[i];
499 	}
500 
501 	if (nb_entries == 0) {
502 		vmolr &= ~IXGBE_VMOLR_ROMPE;
503 		IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
504 		return 0;
505 	}
506 
507 	for (i = 0; i < vfinfo->num_vf_mc_hashes; i++) {
508 		mta_idx = (vfinfo->vf_mc_hashes[i] >> IXGBE_MTA_BIT_SHIFT)
509 				& IXGBE_MTA_INDEX_MASK;
510 		mta_shift = vfinfo->vf_mc_hashes[i] & IXGBE_MTA_BIT_MASK;
511 		reg_val = IXGBE_READ_REG(hw, IXGBE_MTA(mta_idx));
512 		reg_val |= (1 << mta_shift);
513 		IXGBE_WRITE_REG(hw, IXGBE_MTA(mta_idx), reg_val);
514 	}
515 
516 	vmolr |= IXGBE_VMOLR_ROMPE;
517 	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
518 
519 	return 0;
520 }
521 
522 static int
523 ixgbe_vf_set_vlan(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
524 {
525 	int add, vid;
526 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
527 	struct ixgbe_vf_info *vfinfo =
528 		*(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
529 
530 	add = (msgbuf[0] & IXGBE_VT_MSGINFO_MASK)
531 		>> IXGBE_VT_MSGINFO_SHIFT;
532 	vid = (msgbuf[1] & IXGBE_VLVF_VLANID_MASK);
533 
534 	if (add)
535 		vfinfo[vf].vlan_count++;
536 	else if (vfinfo[vf].vlan_count)
537 		vfinfo[vf].vlan_count--;
538 	return hw->mac.ops.set_vfta(hw, vid, vf, (bool)add, false);
539 }
540 
541 static int
542 ixgbe_set_vf_lpe(struct rte_eth_dev *dev, __rte_unused uint32_t vf, uint32_t *msgbuf)
543 {
544 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
545 	uint32_t new_mtu = msgbuf[1];
546 	uint32_t max_frs;
547 	uint32_t hlreg0;
548 	int max_frame = new_mtu + RTE_ETHER_HDR_LEN + RTE_ETHER_CRC_LEN;
549 
550 	/* X540 and X550 support jumbo frames in IOV mode */
551 	if (hw->mac.type != ixgbe_mac_X540 &&
552 		hw->mac.type != ixgbe_mac_X550 &&
553 		hw->mac.type != ixgbe_mac_X550EM_x &&
554 		hw->mac.type != ixgbe_mac_X550EM_a)
555 		return -1;
556 
557 	if (max_frame < RTE_ETHER_MIN_LEN ||
558 			max_frame > RTE_ETHER_MAX_JUMBO_FRAME_LEN)
559 		return -1;
560 
561 	max_frs = (IXGBE_READ_REG(hw, IXGBE_MAXFRS) &
562 		   IXGBE_MHADD_MFS_MASK) >> IXGBE_MHADD_MFS_SHIFT;
563 	if (max_frs < new_mtu) {
564 		hlreg0 = IXGBE_READ_REG(hw, IXGBE_HLREG0);
565 		if (new_mtu > RTE_ETHER_MAX_LEN) {
566 			dev->data->dev_conf.rxmode.offloads |=
567 				DEV_RX_OFFLOAD_JUMBO_FRAME;
568 			hlreg0 |= IXGBE_HLREG0_JUMBOEN;
569 		} else {
570 			dev->data->dev_conf.rxmode.offloads &=
571 				~DEV_RX_OFFLOAD_JUMBO_FRAME;
572 			hlreg0 &= ~IXGBE_HLREG0_JUMBOEN;
573 		}
574 		IXGBE_WRITE_REG(hw, IXGBE_HLREG0, hlreg0);
575 
576 		max_frs = new_mtu << IXGBE_MHADD_MFS_SHIFT;
577 		IXGBE_WRITE_REG(hw, IXGBE_MAXFRS, max_frs);
578 	}
579 
580 	return 0;
581 }
582 
583 static int
584 ixgbe_negotiate_vf_api(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
585 {
586 	uint32_t api_version = msgbuf[1];
587 	struct ixgbe_vf_info *vfinfo =
588 		*IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
589 
590 	switch (api_version) {
591 	case ixgbe_mbox_api_10:
592 	case ixgbe_mbox_api_11:
593 	case ixgbe_mbox_api_12:
594 	case ixgbe_mbox_api_13:
595 		vfinfo[vf].api_version = (uint8_t)api_version;
596 		return 0;
597 	default:
598 		break;
599 	}
600 
601 	PMD_DRV_LOG(ERR, "Negotiate invalid api version %u from VF %d\n",
602 		api_version, vf);
603 
604 	return -1;
605 }
606 
607 static int
608 ixgbe_get_vf_queues(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
609 {
610 	struct ixgbe_vf_info *vfinfo =
611 		*IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
612 	uint32_t default_q = vf * RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
613 	struct rte_eth_conf *eth_conf;
614 	struct rte_eth_vmdq_dcb_tx_conf *vmdq_dcb_tx_conf;
615 	u8 num_tcs;
616 	struct ixgbe_hw *hw;
617 	u32 vmvir;
618 #define IXGBE_VMVIR_VLANA_MASK		0xC0000000
619 #define IXGBE_VMVIR_VLAN_VID_MASK	0x00000FFF
620 #define IXGBE_VMVIR_VLAN_UP_MASK	0x0000E000
621 #define VLAN_PRIO_SHIFT			13
622 	u32 vlana;
623 	u32 vid;
624 	u32 user_priority;
625 
626 	/* Verify if the PF supports the mbox APIs version or not */
627 	switch (vfinfo[vf].api_version) {
628 	case ixgbe_mbox_api_20:
629 	case ixgbe_mbox_api_11:
630 	case ixgbe_mbox_api_12:
631 	case ixgbe_mbox_api_13:
632 		break;
633 	default:
634 		return -1;
635 	}
636 
637 	/* Notify VF of Rx and Tx queue number */
638 	msgbuf[IXGBE_VF_RX_QUEUES] = RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
639 	msgbuf[IXGBE_VF_TX_QUEUES] = RTE_ETH_DEV_SRIOV(dev).nb_q_per_pool;
640 
641 	/* Notify VF of default queue */
642 	msgbuf[IXGBE_VF_DEF_QUEUE] = default_q;
643 
644 	/* Notify VF of number of DCB traffic classes */
645 	eth_conf = &dev->data->dev_conf;
646 	switch (eth_conf->txmode.mq_mode) {
647 	case ETH_MQ_TX_NONE:
648 	case ETH_MQ_TX_DCB:
649 		PMD_DRV_LOG(ERR, "PF must work with virtualization for VF %u"
650 			", but its tx mode = %d\n", vf,
651 			eth_conf->txmode.mq_mode);
652 		return -1;
653 
654 	case ETH_MQ_TX_VMDQ_DCB:
655 		vmdq_dcb_tx_conf = &eth_conf->tx_adv_conf.vmdq_dcb_tx_conf;
656 		switch (vmdq_dcb_tx_conf->nb_queue_pools) {
657 		case ETH_16_POOLS:
658 			num_tcs = ETH_8_TCS;
659 			break;
660 		case ETH_32_POOLS:
661 			num_tcs = ETH_4_TCS;
662 			break;
663 		default:
664 			return -1;
665 		}
666 		break;
667 
668 	/* ETH_MQ_TX_VMDQ_ONLY,  DCB not enabled */
669 	case ETH_MQ_TX_VMDQ_ONLY:
670 		hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
671 		vmvir = IXGBE_READ_REG(hw, IXGBE_VMVIR(vf));
672 		vlana = vmvir & IXGBE_VMVIR_VLANA_MASK;
673 		vid = vmvir & IXGBE_VMVIR_VLAN_VID_MASK;
674 		user_priority =
675 			(vmvir & IXGBE_VMVIR_VLAN_UP_MASK) >> VLAN_PRIO_SHIFT;
676 		if ((vlana == IXGBE_VMVIR_VLANA_DEFAULT) &&
677 			((vid !=  0) || (user_priority != 0)))
678 			num_tcs = 1;
679 		else
680 			num_tcs = 0;
681 		break;
682 
683 	default:
684 		PMD_DRV_LOG(ERR, "PF work with invalid mode = %d\n",
685 			eth_conf->txmode.mq_mode);
686 		return -1;
687 	}
688 	msgbuf[IXGBE_VF_TRANS_VLAN] = num_tcs;
689 
690 	return 0;
691 }
692 
693 static int
694 ixgbe_set_vf_mc_promisc(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
695 {
696 	struct ixgbe_vf_info *vfinfo =
697 		*(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
698 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
699 	int xcast_mode = msgbuf[1];	/* msgbuf contains the flag to enable */
700 	u32 vmolr, fctrl, disable, enable;
701 
702 	switch (vfinfo[vf].api_version) {
703 	case ixgbe_mbox_api_12:
704 		/* promisc introduced in 1.3 version */
705 		if (xcast_mode == IXGBEVF_XCAST_MODE_PROMISC)
706 			return -EOPNOTSUPP;
707 		break;
708 		/* Fall threw */
709 	case ixgbe_mbox_api_13:
710 		break;
711 	default:
712 		return -1;
713 	}
714 
715 	if (vfinfo[vf].xcast_mode == xcast_mode)
716 		goto out;
717 
718 	switch (xcast_mode) {
719 	case IXGBEVF_XCAST_MODE_NONE:
720 		disable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE |
721 			  IXGBE_VMOLR_MPE | IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
722 		enable = 0;
723 		break;
724 	case IXGBEVF_XCAST_MODE_MULTI:
725 		disable = IXGBE_VMOLR_MPE | IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
726 		enable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE;
727 		break;
728 	case IXGBEVF_XCAST_MODE_ALLMULTI:
729 		disable = IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
730 		enable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE | IXGBE_VMOLR_MPE;
731 		break;
732 	case IXGBEVF_XCAST_MODE_PROMISC:
733 		if (hw->mac.type <= ixgbe_mac_82599EB)
734 			return -1;
735 
736 		fctrl = IXGBE_READ_REG(hw, IXGBE_FCTRL);
737 		if (!(fctrl & IXGBE_FCTRL_UPE)) {
738 			/* VF promisc requires PF in promisc */
739 			PMD_DRV_LOG(ERR,
740 			       "Enabling VF promisc requires PF in promisc\n");
741 			return -1;
742 		}
743 
744 		disable = 0;
745 		enable = IXGBE_VMOLR_BAM | IXGBE_VMOLR_ROMPE |
746 			 IXGBE_VMOLR_MPE | IXGBE_VMOLR_UPE | IXGBE_VMOLR_VPE;
747 		break;
748 	default:
749 		return -1;
750 	}
751 
752 	vmolr = IXGBE_READ_REG(hw, IXGBE_VMOLR(vf));
753 	vmolr &= ~disable;
754 	vmolr |= enable;
755 	IXGBE_WRITE_REG(hw, IXGBE_VMOLR(vf), vmolr);
756 	vfinfo[vf].xcast_mode = xcast_mode;
757 
758 out:
759 	msgbuf[1] = xcast_mode;
760 
761 	return 0;
762 }
763 
764 static int
765 ixgbe_set_vf_macvlan_msg(struct rte_eth_dev *dev, uint32_t vf, uint32_t *msgbuf)
766 {
767 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
768 	struct ixgbe_vf_info *vf_info =
769 		*(IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private));
770 	uint8_t *new_mac = (uint8_t *)(&msgbuf[1]);
771 	int index = (msgbuf[0] & IXGBE_VT_MSGINFO_MASK) >>
772 		    IXGBE_VT_MSGINFO_SHIFT;
773 
774 	if (index) {
775 		if (!rte_is_valid_assigned_ether_addr(
776 			(struct rte_ether_addr *)new_mac)) {
777 			PMD_DRV_LOG(ERR, "set invalid mac vf:%d\n", vf);
778 			return -1;
779 		}
780 
781 		vf_info[vf].mac_count++;
782 
783 		hw->mac.ops.set_rar(hw, vf_info[vf].mac_count,
784 				new_mac, vf, IXGBE_RAH_AV);
785 	} else {
786 		if (vf_info[vf].mac_count) {
787 			hw->mac.ops.clear_rar(hw, vf_info[vf].mac_count);
788 			vf_info[vf].mac_count = 0;
789 		}
790 	}
791 	return 0;
792 }
793 
794 static int
795 ixgbe_rcv_msg_from_vf(struct rte_eth_dev *dev, uint16_t vf)
796 {
797 	uint16_t mbx_size = IXGBE_VFMAILBOX_SIZE;
798 	uint16_t msg_size = IXGBE_VF_MSG_SIZE_DEFAULT;
799 	uint32_t msgbuf[IXGBE_VFMAILBOX_SIZE];
800 	int32_t retval;
801 	struct ixgbe_hw *hw = IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
802 	struct ixgbe_vf_info *vfinfo =
803 		*IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
804 	struct rte_pmd_ixgbe_mb_event_param ret_param;
805 
806 	retval = ixgbe_read_mbx(hw, msgbuf, mbx_size, vf);
807 	if (retval) {
808 		PMD_DRV_LOG(ERR, "Error mbx recv msg from VF %d", vf);
809 		return retval;
810 	}
811 
812 	/* do nothing with the message already been processed */
813 	if (msgbuf[0] & (IXGBE_VT_MSGTYPE_ACK | IXGBE_VT_MSGTYPE_NACK))
814 		return retval;
815 
816 	/* flush the ack before we write any messages back */
817 	IXGBE_WRITE_FLUSH(hw);
818 
819 	/**
820 	 * initialise structure to send to user application
821 	 * will return response from user in retval field
822 	 */
823 	ret_param.retval = RTE_PMD_IXGBE_MB_EVENT_PROCEED;
824 	ret_param.vfid = vf;
825 	ret_param.msg_type = msgbuf[0] & 0xFFFF;
826 	ret_param.msg = (void *)msgbuf;
827 
828 	/* perform VF reset */
829 	if (msgbuf[0] == IXGBE_VF_RESET) {
830 		int ret = ixgbe_vf_reset(dev, vf, msgbuf);
831 
832 		vfinfo[vf].clear_to_send = true;
833 
834 		/* notify application about VF reset */
835 		rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX,
836 					      &ret_param);
837 		return ret;
838 	}
839 
840 	/**
841 	 * ask user application if we allowed to perform those functions
842 	 * if we get ret_param.retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED
843 	 * then business as usual,
844 	 * if 0, do nothing and send ACK to VF
845 	 * if ret_param.retval > 1, do nothing and send NAK to VF
846 	 */
847 	rte_eth_dev_callback_process(dev, RTE_ETH_EVENT_VF_MBOX, &ret_param);
848 
849 	retval = ret_param.retval;
850 
851 	/* check & process VF to PF mailbox message */
852 	switch ((msgbuf[0] & 0xFFFF)) {
853 	case IXGBE_VF_SET_MAC_ADDR:
854 		if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
855 			retval = ixgbe_vf_set_mac_addr(dev, vf, msgbuf);
856 		break;
857 	case IXGBE_VF_SET_MULTICAST:
858 		if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
859 			retval = ixgbe_vf_set_multicast(dev, vf, msgbuf);
860 		break;
861 	case IXGBE_VF_SET_LPE:
862 		if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
863 			retval = ixgbe_set_vf_lpe(dev, vf, msgbuf);
864 		break;
865 	case IXGBE_VF_SET_VLAN:
866 		if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
867 			retval = ixgbe_vf_set_vlan(dev, vf, msgbuf);
868 		break;
869 	case IXGBE_VF_API_NEGOTIATE:
870 		retval = ixgbe_negotiate_vf_api(dev, vf, msgbuf);
871 		break;
872 	case IXGBE_VF_GET_QUEUES:
873 		retval = ixgbe_get_vf_queues(dev, vf, msgbuf);
874 		msg_size = IXGBE_VF_GET_QUEUE_MSG_SIZE;
875 		break;
876 	case IXGBE_VF_UPDATE_XCAST_MODE:
877 		if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
878 			retval = ixgbe_set_vf_mc_promisc(dev, vf, msgbuf);
879 		break;
880 	case IXGBE_VF_SET_MACVLAN:
881 		if (retval == RTE_PMD_IXGBE_MB_EVENT_PROCEED)
882 			retval = ixgbe_set_vf_macvlan_msg(dev, vf, msgbuf);
883 		break;
884 	default:
885 		PMD_DRV_LOG(DEBUG, "Unhandled Msg %8.8x", (unsigned)msgbuf[0]);
886 		retval = IXGBE_ERR_MBX;
887 		break;
888 	}
889 
890 	/* response the VF according to the message process result */
891 	if (retval)
892 		msgbuf[0] |= IXGBE_VT_MSGTYPE_NACK;
893 	else
894 		msgbuf[0] |= IXGBE_VT_MSGTYPE_ACK;
895 
896 	msgbuf[0] |= IXGBE_VT_MSGTYPE_CTS;
897 
898 	ixgbe_write_mbx(hw, msgbuf, msg_size, vf);
899 
900 	return retval;
901 }
902 
903 static inline void
904 ixgbe_rcv_ack_from_vf(struct rte_eth_dev *dev, uint16_t vf)
905 {
906 	uint32_t msg = IXGBE_VT_MSGTYPE_NACK;
907 	struct ixgbe_hw *hw =
908 		IXGBE_DEV_PRIVATE_TO_HW(dev->data->dev_private);
909 	struct ixgbe_vf_info *vfinfo =
910 		*IXGBE_DEV_PRIVATE_TO_P_VFDATA(dev->data->dev_private);
911 
912 	if (!vfinfo[vf].clear_to_send)
913 		ixgbe_write_mbx(hw, &msg, 1, vf);
914 }
915 
916 void ixgbe_pf_mbx_process(struct rte_eth_dev *eth_dev)
917 {
918 	uint16_t vf;
919 	struct ixgbe_hw *hw =
920 		IXGBE_DEV_PRIVATE_TO_HW(eth_dev->data->dev_private);
921 
922 	for (vf = 0; vf < dev_num_vf(eth_dev); vf++) {
923 		/* check & process vf function level reset */
924 		if (!ixgbe_check_for_rst(hw, vf))
925 			ixgbe_vf_reset_event(eth_dev, vf);
926 
927 		/* check & process vf mailbox messages */
928 		if (!ixgbe_check_for_msg(hw, vf))
929 			ixgbe_rcv_msg_from_vf(eth_dev, vf);
930 
931 		/* check & process acks from vf */
932 		if (!ixgbe_check_for_ack(hw, vf))
933 			ixgbe_rcv_ack_from_vf(eth_dev, vf);
934 	}
935 }
936