xref: /openbmc/linux/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c (revision e4781421e883340b796da5a724bda7226817990b)
1 /* Broadcom NetXtreme-C/E network driver.
2  *
3  * Copyright (c) 2014-2016 Broadcom Corporation
4  *
5  * This program is free software; you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation.
8  */
9 
10 #include <linux/module.h>
11 #include <linux/pci.h>
12 #include <linux/netdevice.h>
13 #include <linux/if_vlan.h>
14 #include <linux/interrupt.h>
15 #include <linux/etherdevice.h>
16 #include "bnxt_hsi.h"
17 #include "bnxt.h"
18 #include "bnxt_sriov.h"
19 #include "bnxt_ethtool.h"
20 
21 #ifdef CONFIG_BNXT_SRIOV
22 static int bnxt_hwrm_fwd_async_event_cmpl(struct bnxt *bp,
23 					  struct bnxt_vf_info *vf, u16 event_id)
24 {
25 	struct hwrm_fwd_async_event_cmpl_output *resp = bp->hwrm_cmd_resp_addr;
26 	struct hwrm_fwd_async_event_cmpl_input req = {0};
27 	struct hwrm_async_event_cmpl *async_cmpl;
28 	int rc = 0;
29 
30 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FWD_ASYNC_EVENT_CMPL, -1, -1);
31 	if (vf)
32 		req.encap_async_event_target_id = cpu_to_le16(vf->fw_fid);
33 	else
34 		/* broadcast this async event to all VFs */
35 		req.encap_async_event_target_id = cpu_to_le16(0xffff);
36 	async_cmpl = (struct hwrm_async_event_cmpl *)req.encap_async_event_cmpl;
37 	async_cmpl->type = cpu_to_le16(ASYNC_EVENT_CMPL_TYPE_HWRM_ASYNC_EVENT);
38 	async_cmpl->event_id = cpu_to_le16(event_id);
39 
40 	mutex_lock(&bp->hwrm_cmd_lock);
41 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
42 
43 	if (rc) {
44 		netdev_err(bp->dev, "hwrm_fwd_async_event_cmpl failed. rc:%d\n",
45 			   rc);
46 		goto fwd_async_event_cmpl_exit;
47 	}
48 
49 	if (resp->error_code) {
50 		netdev_err(bp->dev, "hwrm_fwd_async_event_cmpl error %d\n",
51 			   resp->error_code);
52 		rc = -1;
53 	}
54 
55 fwd_async_event_cmpl_exit:
56 	mutex_unlock(&bp->hwrm_cmd_lock);
57 	return rc;
58 }
59 
60 static int bnxt_vf_ndo_prep(struct bnxt *bp, int vf_id)
61 {
62 	if (!test_bit(BNXT_STATE_OPEN, &bp->state)) {
63 		netdev_err(bp->dev, "vf ndo called though PF is down\n");
64 		return -EINVAL;
65 	}
66 	if (!bp->pf.active_vfs) {
67 		netdev_err(bp->dev, "vf ndo called though sriov is disabled\n");
68 		return -EINVAL;
69 	}
70 	if (vf_id >= bp->pf.max_vfs) {
71 		netdev_err(bp->dev, "Invalid VF id %d\n", vf_id);
72 		return -EINVAL;
73 	}
74 	return 0;
75 }
76 
77 int bnxt_set_vf_spoofchk(struct net_device *dev, int vf_id, bool setting)
78 {
79 	struct hwrm_func_cfg_input req = {0};
80 	struct bnxt *bp = netdev_priv(dev);
81 	struct bnxt_vf_info *vf;
82 	bool old_setting = false;
83 	u32 func_flags;
84 	int rc;
85 
86 	rc = bnxt_vf_ndo_prep(bp, vf_id);
87 	if (rc)
88 		return rc;
89 
90 	vf = &bp->pf.vf[vf_id];
91 	if (vf->flags & BNXT_VF_SPOOFCHK)
92 		old_setting = true;
93 	if (old_setting == setting)
94 		return 0;
95 
96 	func_flags = vf->func_flags;
97 	if (setting)
98 		func_flags |= FUNC_CFG_REQ_FLAGS_SRC_MAC_ADDR_CHECK;
99 	else
100 		func_flags &= ~FUNC_CFG_REQ_FLAGS_SRC_MAC_ADDR_CHECK;
101 	/*TODO: if the driver supports VLAN filter on guest VLAN,
102 	 * the spoof check should also include vlan anti-spoofing
103 	 */
104 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
105 	req.fid = cpu_to_le16(vf->fw_fid);
106 	req.flags = cpu_to_le32(func_flags);
107 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
108 	if (!rc) {
109 		vf->func_flags = func_flags;
110 		if (setting)
111 			vf->flags |= BNXT_VF_SPOOFCHK;
112 		else
113 			vf->flags &= ~BNXT_VF_SPOOFCHK;
114 	}
115 	return rc;
116 }
117 
118 int bnxt_get_vf_config(struct net_device *dev, int vf_id,
119 		       struct ifla_vf_info *ivi)
120 {
121 	struct bnxt *bp = netdev_priv(dev);
122 	struct bnxt_vf_info *vf;
123 	int rc;
124 
125 	rc = bnxt_vf_ndo_prep(bp, vf_id);
126 	if (rc)
127 		return rc;
128 
129 	ivi->vf = vf_id;
130 	vf = &bp->pf.vf[vf_id];
131 
132 	memcpy(&ivi->mac, vf->mac_addr, ETH_ALEN);
133 	ivi->max_tx_rate = vf->max_tx_rate;
134 	ivi->min_tx_rate = vf->min_tx_rate;
135 	ivi->vlan = vf->vlan;
136 	ivi->qos = vf->flags & BNXT_VF_QOS;
137 	ivi->spoofchk = vf->flags & BNXT_VF_SPOOFCHK;
138 	if (!(vf->flags & BNXT_VF_LINK_FORCED))
139 		ivi->linkstate = IFLA_VF_LINK_STATE_AUTO;
140 	else if (vf->flags & BNXT_VF_LINK_UP)
141 		ivi->linkstate = IFLA_VF_LINK_STATE_ENABLE;
142 	else
143 		ivi->linkstate = IFLA_VF_LINK_STATE_DISABLE;
144 
145 	return 0;
146 }
147 
148 int bnxt_set_vf_mac(struct net_device *dev, int vf_id, u8 *mac)
149 {
150 	struct hwrm_func_cfg_input req = {0};
151 	struct bnxt *bp = netdev_priv(dev);
152 	struct bnxt_vf_info *vf;
153 	int rc;
154 
155 	rc = bnxt_vf_ndo_prep(bp, vf_id);
156 	if (rc)
157 		return rc;
158 	/* reject bc or mc mac addr, zero mac addr means allow
159 	 * VF to use its own mac addr
160 	 */
161 	if (is_multicast_ether_addr(mac)) {
162 		netdev_err(dev, "Invalid VF ethernet address\n");
163 		return -EINVAL;
164 	}
165 	vf = &bp->pf.vf[vf_id];
166 
167 	memcpy(vf->mac_addr, mac, ETH_ALEN);
168 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
169 	req.fid = cpu_to_le16(vf->fw_fid);
170 	req.flags = cpu_to_le32(vf->func_flags);
171 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_DFLT_MAC_ADDR);
172 	memcpy(req.dflt_mac_addr, mac, ETH_ALEN);
173 	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
174 }
175 
176 int bnxt_set_vf_vlan(struct net_device *dev, int vf_id, u16 vlan_id, u8 qos,
177 		     __be16 vlan_proto)
178 {
179 	struct hwrm_func_cfg_input req = {0};
180 	struct bnxt *bp = netdev_priv(dev);
181 	struct bnxt_vf_info *vf;
182 	u16 vlan_tag;
183 	int rc;
184 
185 	if (bp->hwrm_spec_code < 0x10201)
186 		return -ENOTSUPP;
187 
188 	if (vlan_proto != htons(ETH_P_8021Q))
189 		return -EPROTONOSUPPORT;
190 
191 	rc = bnxt_vf_ndo_prep(bp, vf_id);
192 	if (rc)
193 		return rc;
194 
195 	/* TODO: needed to implement proper handling of user priority,
196 	 * currently fail the command if there is valid priority
197 	 */
198 	if (vlan_id > 4095 || qos)
199 		return -EINVAL;
200 
201 	vf = &bp->pf.vf[vf_id];
202 	vlan_tag = vlan_id;
203 	if (vlan_tag == vf->vlan)
204 		return 0;
205 
206 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
207 	req.fid = cpu_to_le16(vf->fw_fid);
208 	req.flags = cpu_to_le32(vf->func_flags);
209 	req.dflt_vlan = cpu_to_le16(vlan_tag);
210 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_DFLT_VLAN);
211 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
212 	if (!rc)
213 		vf->vlan = vlan_tag;
214 	return rc;
215 }
216 
217 int bnxt_set_vf_bw(struct net_device *dev, int vf_id, int min_tx_rate,
218 		   int max_tx_rate)
219 {
220 	struct hwrm_func_cfg_input req = {0};
221 	struct bnxt *bp = netdev_priv(dev);
222 	struct bnxt_vf_info *vf;
223 	u32 pf_link_speed;
224 	int rc;
225 
226 	rc = bnxt_vf_ndo_prep(bp, vf_id);
227 	if (rc)
228 		return rc;
229 
230 	vf = &bp->pf.vf[vf_id];
231 	pf_link_speed = bnxt_fw_to_ethtool_speed(bp->link_info.link_speed);
232 	if (max_tx_rate > pf_link_speed) {
233 		netdev_info(bp->dev, "max tx rate %d exceed PF link speed for VF %d\n",
234 			    max_tx_rate, vf_id);
235 		return -EINVAL;
236 	}
237 
238 	if (min_tx_rate > pf_link_speed || min_tx_rate > max_tx_rate) {
239 		netdev_info(bp->dev, "min tx rate %d is invalid for VF %d\n",
240 			    min_tx_rate, vf_id);
241 		return -EINVAL;
242 	}
243 	if (min_tx_rate == vf->min_tx_rate && max_tx_rate == vf->max_tx_rate)
244 		return 0;
245 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
246 	req.fid = cpu_to_le16(vf->fw_fid);
247 	req.flags = cpu_to_le32(vf->func_flags);
248 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_MAX_BW);
249 	req.max_bw = cpu_to_le32(max_tx_rate);
250 	req.enables |= cpu_to_le32(FUNC_CFG_REQ_ENABLES_MIN_BW);
251 	req.min_bw = cpu_to_le32(min_tx_rate);
252 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
253 	if (!rc) {
254 		vf->min_tx_rate = min_tx_rate;
255 		vf->max_tx_rate = max_tx_rate;
256 	}
257 	return rc;
258 }
259 
260 int bnxt_set_vf_link_state(struct net_device *dev, int vf_id, int link)
261 {
262 	struct bnxt *bp = netdev_priv(dev);
263 	struct bnxt_vf_info *vf;
264 	int rc;
265 
266 	rc = bnxt_vf_ndo_prep(bp, vf_id);
267 	if (rc)
268 		return rc;
269 
270 	vf = &bp->pf.vf[vf_id];
271 
272 	vf->flags &= ~(BNXT_VF_LINK_UP | BNXT_VF_LINK_FORCED);
273 	switch (link) {
274 	case IFLA_VF_LINK_STATE_AUTO:
275 		vf->flags |= BNXT_VF_LINK_UP;
276 		break;
277 	case IFLA_VF_LINK_STATE_DISABLE:
278 		vf->flags |= BNXT_VF_LINK_FORCED;
279 		break;
280 	case IFLA_VF_LINK_STATE_ENABLE:
281 		vf->flags |= BNXT_VF_LINK_UP | BNXT_VF_LINK_FORCED;
282 		break;
283 	default:
284 		netdev_err(bp->dev, "Invalid link option\n");
285 		rc = -EINVAL;
286 		break;
287 	}
288 	if (vf->flags & (BNXT_VF_LINK_UP | BNXT_VF_LINK_FORCED))
289 		rc = bnxt_hwrm_fwd_async_event_cmpl(bp, vf,
290 			ASYNC_EVENT_CMPL_EVENT_ID_LINK_STATUS_CHANGE);
291 	return rc;
292 }
293 
294 static int bnxt_set_vf_attr(struct bnxt *bp, int num_vfs)
295 {
296 	int i;
297 	struct bnxt_vf_info *vf;
298 
299 	for (i = 0; i < num_vfs; i++) {
300 		vf = &bp->pf.vf[i];
301 		memset(vf, 0, sizeof(*vf));
302 		vf->flags = BNXT_VF_QOS | BNXT_VF_LINK_UP;
303 	}
304 	return 0;
305 }
306 
307 static int bnxt_hwrm_func_vf_resource_free(struct bnxt *bp, int num_vfs)
308 {
309 	int i, rc = 0;
310 	struct bnxt_pf_info *pf = &bp->pf;
311 	struct hwrm_func_vf_resc_free_input req = {0};
312 
313 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_VF_RESC_FREE, -1, -1);
314 
315 	mutex_lock(&bp->hwrm_cmd_lock);
316 	for (i = pf->first_vf_id; i < pf->first_vf_id + num_vfs; i++) {
317 		req.vf_id = cpu_to_le16(i);
318 		rc = _hwrm_send_message(bp, &req, sizeof(req),
319 					HWRM_CMD_TIMEOUT);
320 		if (rc)
321 			break;
322 	}
323 	mutex_unlock(&bp->hwrm_cmd_lock);
324 	return rc;
325 }
326 
327 static void bnxt_free_vf_resources(struct bnxt *bp)
328 {
329 	struct pci_dev *pdev = bp->pdev;
330 	int i;
331 
332 	kfree(bp->pf.vf_event_bmap);
333 	bp->pf.vf_event_bmap = NULL;
334 
335 	for (i = 0; i < 4; i++) {
336 		if (bp->pf.hwrm_cmd_req_addr[i]) {
337 			dma_free_coherent(&pdev->dev, BNXT_PAGE_SIZE,
338 					  bp->pf.hwrm_cmd_req_addr[i],
339 					  bp->pf.hwrm_cmd_req_dma_addr[i]);
340 			bp->pf.hwrm_cmd_req_addr[i] = NULL;
341 		}
342 	}
343 
344 	kfree(bp->pf.vf);
345 	bp->pf.vf = NULL;
346 }
347 
348 static int bnxt_alloc_vf_resources(struct bnxt *bp, int num_vfs)
349 {
350 	struct pci_dev *pdev = bp->pdev;
351 	u32 nr_pages, size, i, j, k = 0;
352 
353 	bp->pf.vf = kcalloc(num_vfs, sizeof(struct bnxt_vf_info), GFP_KERNEL);
354 	if (!bp->pf.vf)
355 		return -ENOMEM;
356 
357 	bnxt_set_vf_attr(bp, num_vfs);
358 
359 	size = num_vfs * BNXT_HWRM_REQ_MAX_SIZE;
360 	nr_pages = size / BNXT_PAGE_SIZE;
361 	if (size & (BNXT_PAGE_SIZE - 1))
362 		nr_pages++;
363 
364 	for (i = 0; i < nr_pages; i++) {
365 		bp->pf.hwrm_cmd_req_addr[i] =
366 			dma_alloc_coherent(&pdev->dev, BNXT_PAGE_SIZE,
367 					   &bp->pf.hwrm_cmd_req_dma_addr[i],
368 					   GFP_KERNEL);
369 
370 		if (!bp->pf.hwrm_cmd_req_addr[i])
371 			return -ENOMEM;
372 
373 		for (j = 0; j < BNXT_HWRM_REQS_PER_PAGE && k < num_vfs; j++) {
374 			struct bnxt_vf_info *vf = &bp->pf.vf[k];
375 
376 			vf->hwrm_cmd_req_addr = bp->pf.hwrm_cmd_req_addr[i] +
377 						j * BNXT_HWRM_REQ_MAX_SIZE;
378 			vf->hwrm_cmd_req_dma_addr =
379 				bp->pf.hwrm_cmd_req_dma_addr[i] + j *
380 				BNXT_HWRM_REQ_MAX_SIZE;
381 			k++;
382 		}
383 	}
384 
385 	/* Max 128 VF's */
386 	bp->pf.vf_event_bmap = kzalloc(16, GFP_KERNEL);
387 	if (!bp->pf.vf_event_bmap)
388 		return -ENOMEM;
389 
390 	bp->pf.hwrm_cmd_req_pages = nr_pages;
391 	return 0;
392 }
393 
394 static int bnxt_hwrm_func_buf_rgtr(struct bnxt *bp)
395 {
396 	struct hwrm_func_buf_rgtr_input req = {0};
397 
398 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_BUF_RGTR, -1, -1);
399 
400 	req.req_buf_num_pages = cpu_to_le16(bp->pf.hwrm_cmd_req_pages);
401 	req.req_buf_page_size = cpu_to_le16(BNXT_PAGE_SHIFT);
402 	req.req_buf_len = cpu_to_le16(BNXT_HWRM_REQ_MAX_SIZE);
403 	req.req_buf_page_addr0 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[0]);
404 	req.req_buf_page_addr1 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[1]);
405 	req.req_buf_page_addr2 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[2]);
406 	req.req_buf_page_addr3 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[3]);
407 
408 	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
409 }
410 
411 /* only call by PF to reserve resources for VF */
412 static int bnxt_hwrm_func_cfg(struct bnxt *bp, int num_vfs)
413 {
414 	u32 rc = 0, mtu, i;
415 	u16 vf_tx_rings, vf_rx_rings, vf_cp_rings, vf_stat_ctx, vf_vnics;
416 	u16 vf_ring_grps;
417 	struct hwrm_func_cfg_input req = {0};
418 	struct bnxt_pf_info *pf = &bp->pf;
419 	int total_vf_tx_rings = 0;
420 
421 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
422 
423 	/* Remaining rings are distributed equally amongs VF's for now */
424 	vf_cp_rings = (pf->max_cp_rings - bp->cp_nr_rings) / num_vfs;
425 	vf_stat_ctx = (pf->max_stat_ctxs - bp->num_stat_ctxs) / num_vfs;
426 	if (bp->flags & BNXT_FLAG_AGG_RINGS)
427 		vf_rx_rings = (pf->max_rx_rings - bp->rx_nr_rings * 2) /
428 			      num_vfs;
429 	else
430 		vf_rx_rings = (pf->max_rx_rings - bp->rx_nr_rings) / num_vfs;
431 	vf_ring_grps = (bp->pf.max_hw_ring_grps - bp->rx_nr_rings) / num_vfs;
432 	vf_tx_rings = (pf->max_tx_rings - bp->tx_nr_rings) / num_vfs;
433 	vf_vnics = (pf->max_vnics - bp->nr_vnics) / num_vfs;
434 	vf_vnics = min_t(u16, vf_vnics, vf_rx_rings);
435 
436 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_MTU |
437 				  FUNC_CFG_REQ_ENABLES_MRU |
438 				  FUNC_CFG_REQ_ENABLES_NUM_RSSCOS_CTXS |
439 				  FUNC_CFG_REQ_ENABLES_NUM_STAT_CTXS |
440 				  FUNC_CFG_REQ_ENABLES_NUM_CMPL_RINGS |
441 				  FUNC_CFG_REQ_ENABLES_NUM_TX_RINGS |
442 				  FUNC_CFG_REQ_ENABLES_NUM_RX_RINGS |
443 				  FUNC_CFG_REQ_ENABLES_NUM_L2_CTXS |
444 				  FUNC_CFG_REQ_ENABLES_NUM_VNICS |
445 				  FUNC_CFG_REQ_ENABLES_NUM_HW_RING_GRPS);
446 
447 	mtu = bp->dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
448 	req.mru = cpu_to_le16(mtu);
449 	req.mtu = cpu_to_le16(mtu);
450 
451 	req.num_rsscos_ctxs = cpu_to_le16(1);
452 	req.num_cmpl_rings = cpu_to_le16(vf_cp_rings);
453 	req.num_tx_rings = cpu_to_le16(vf_tx_rings);
454 	req.num_rx_rings = cpu_to_le16(vf_rx_rings);
455 	req.num_hw_ring_grps = cpu_to_le16(vf_ring_grps);
456 	req.num_l2_ctxs = cpu_to_le16(4);
457 
458 	req.num_vnics = cpu_to_le16(vf_vnics);
459 	/* FIXME spec currently uses 1 bit for stats ctx */
460 	req.num_stat_ctxs = cpu_to_le16(vf_stat_ctx);
461 
462 	mutex_lock(&bp->hwrm_cmd_lock);
463 	for (i = 0; i < num_vfs; i++) {
464 		int vf_tx_rsvd = vf_tx_rings;
465 
466 		req.fid = cpu_to_le16(pf->first_vf_id + i);
467 		rc = _hwrm_send_message(bp, &req, sizeof(req),
468 					HWRM_CMD_TIMEOUT);
469 		if (rc)
470 			break;
471 		pf->active_vfs = i + 1;
472 		pf->vf[i].fw_fid = le16_to_cpu(req.fid);
473 		rc = __bnxt_hwrm_get_tx_rings(bp, pf->vf[i].fw_fid,
474 					      &vf_tx_rsvd);
475 		if (rc)
476 			break;
477 		total_vf_tx_rings += vf_tx_rsvd;
478 	}
479 	mutex_unlock(&bp->hwrm_cmd_lock);
480 	if (!rc) {
481 		pf->max_tx_rings -= total_vf_tx_rings;
482 		pf->max_rx_rings -= vf_rx_rings * num_vfs;
483 		pf->max_hw_ring_grps -= vf_ring_grps * num_vfs;
484 		pf->max_cp_rings -= vf_cp_rings * num_vfs;
485 		pf->max_rsscos_ctxs -= num_vfs;
486 		pf->max_stat_ctxs -= vf_stat_ctx * num_vfs;
487 		pf->max_vnics -= vf_vnics * num_vfs;
488 	}
489 	return rc;
490 }
491 
492 static int bnxt_sriov_enable(struct bnxt *bp, int *num_vfs)
493 {
494 	int rc = 0, vfs_supported;
495 	int min_rx_rings, min_tx_rings, min_rss_ctxs;
496 	int tx_ok = 0, rx_ok = 0, rss_ok = 0;
497 
498 	/* Check if we can enable requested num of vf's. At a mininum
499 	 * we require 1 RX 1 TX rings for each VF. In this minimum conf
500 	 * features like TPA will not be available.
501 	 */
502 	vfs_supported = *num_vfs;
503 
504 	while (vfs_supported) {
505 		min_rx_rings = vfs_supported;
506 		min_tx_rings = vfs_supported;
507 		min_rss_ctxs = vfs_supported;
508 
509 		if (bp->flags & BNXT_FLAG_AGG_RINGS) {
510 			if (bp->pf.max_rx_rings - bp->rx_nr_rings * 2 >=
511 			    min_rx_rings)
512 				rx_ok = 1;
513 		} else {
514 			if (bp->pf.max_rx_rings - bp->rx_nr_rings >=
515 			    min_rx_rings)
516 				rx_ok = 1;
517 		}
518 		if (bp->pf.max_vnics - bp->nr_vnics < min_rx_rings)
519 			rx_ok = 0;
520 
521 		if (bp->pf.max_tx_rings - bp->tx_nr_rings >= min_tx_rings)
522 			tx_ok = 1;
523 
524 		if (bp->pf.max_rsscos_ctxs - bp->rsscos_nr_ctxs >= min_rss_ctxs)
525 			rss_ok = 1;
526 
527 		if (tx_ok && rx_ok && rss_ok)
528 			break;
529 
530 		vfs_supported--;
531 	}
532 
533 	if (!vfs_supported) {
534 		netdev_err(bp->dev, "Cannot enable VF's as all resources are used by PF\n");
535 		return -EINVAL;
536 	}
537 
538 	if (vfs_supported != *num_vfs) {
539 		netdev_info(bp->dev, "Requested VFs %d, can enable %d\n",
540 			    *num_vfs, vfs_supported);
541 		*num_vfs = vfs_supported;
542 	}
543 
544 	rc = bnxt_alloc_vf_resources(bp, *num_vfs);
545 	if (rc)
546 		goto err_out1;
547 
548 	/* Reserve resources for VFs */
549 	rc = bnxt_hwrm_func_cfg(bp, *num_vfs);
550 	if (rc)
551 		goto err_out2;
552 
553 	/* Register buffers for VFs */
554 	rc = bnxt_hwrm_func_buf_rgtr(bp);
555 	if (rc)
556 		goto err_out2;
557 
558 	rc = pci_enable_sriov(bp->pdev, *num_vfs);
559 	if (rc)
560 		goto err_out2;
561 
562 	return 0;
563 
564 err_out2:
565 	/* Free the resources reserved for various VF's */
566 	bnxt_hwrm_func_vf_resource_free(bp, *num_vfs);
567 
568 err_out1:
569 	bnxt_free_vf_resources(bp);
570 
571 	return rc;
572 }
573 
574 void bnxt_sriov_disable(struct bnxt *bp)
575 {
576 	u16 num_vfs = pci_num_vf(bp->pdev);
577 
578 	if (!num_vfs)
579 		return;
580 
581 	if (pci_vfs_assigned(bp->pdev)) {
582 		bnxt_hwrm_fwd_async_event_cmpl(
583 			bp, NULL, ASYNC_EVENT_CMPL_EVENT_ID_PF_DRVR_UNLOAD);
584 		netdev_warn(bp->dev, "Unable to free %d VFs because some are assigned to VMs.\n",
585 			    num_vfs);
586 	} else {
587 		pci_disable_sriov(bp->pdev);
588 		/* Free the HW resources reserved for various VF's */
589 		bnxt_hwrm_func_vf_resource_free(bp, num_vfs);
590 	}
591 
592 	bnxt_free_vf_resources(bp);
593 
594 	bp->pf.active_vfs = 0;
595 	/* Reclaim all resources for the PF. */
596 	rtnl_lock();
597 	bnxt_restore_pf_fw_resources(bp);
598 	rtnl_unlock();
599 }
600 
601 int bnxt_sriov_configure(struct pci_dev *pdev, int num_vfs)
602 {
603 	struct net_device *dev = pci_get_drvdata(pdev);
604 	struct bnxt *bp = netdev_priv(dev);
605 
606 	if (!(bp->flags & BNXT_FLAG_USING_MSIX)) {
607 		netdev_warn(dev, "Not allow SRIOV if the irq mode is not MSIX\n");
608 		return 0;
609 	}
610 
611 	rtnl_lock();
612 	if (!netif_running(dev)) {
613 		netdev_warn(dev, "Reject SRIOV config request since if is down!\n");
614 		rtnl_unlock();
615 		return 0;
616 	}
617 	bp->sriov_cfg = true;
618 	rtnl_unlock();
619 
620 	if (pci_vfs_assigned(bp->pdev)) {
621 		netdev_warn(dev, "Unable to configure SRIOV since some VFs are assigned to VMs.\n");
622 		num_vfs = 0;
623 		goto sriov_cfg_exit;
624 	}
625 
626 	/* Check if enabled VFs is same as requested */
627 	if (num_vfs && num_vfs == bp->pf.active_vfs)
628 		goto sriov_cfg_exit;
629 
630 	/* if there are previous existing VFs, clean them up */
631 	bnxt_sriov_disable(bp);
632 	if (!num_vfs)
633 		goto sriov_cfg_exit;
634 
635 	bnxt_sriov_enable(bp, &num_vfs);
636 
637 sriov_cfg_exit:
638 	bp->sriov_cfg = false;
639 	wake_up(&bp->sriov_cfg_wait);
640 
641 	return num_vfs;
642 }
643 
644 static int bnxt_hwrm_fwd_resp(struct bnxt *bp, struct bnxt_vf_info *vf,
645 			      void *encap_resp, __le64 encap_resp_addr,
646 			      __le16 encap_resp_cpr, u32 msg_size)
647 {
648 	int rc = 0;
649 	struct hwrm_fwd_resp_input req = {0};
650 	struct hwrm_fwd_resp_output *resp = bp->hwrm_cmd_resp_addr;
651 
652 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FWD_RESP, -1, -1);
653 
654 	/* Set the new target id */
655 	req.target_id = cpu_to_le16(vf->fw_fid);
656 	req.encap_resp_target_id = cpu_to_le16(vf->fw_fid);
657 	req.encap_resp_len = cpu_to_le16(msg_size);
658 	req.encap_resp_addr = encap_resp_addr;
659 	req.encap_resp_cmpl_ring = encap_resp_cpr;
660 	memcpy(req.encap_resp, encap_resp, msg_size);
661 
662 	mutex_lock(&bp->hwrm_cmd_lock);
663 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
664 
665 	if (rc) {
666 		netdev_err(bp->dev, "hwrm_fwd_resp failed. rc:%d\n", rc);
667 		goto fwd_resp_exit;
668 	}
669 
670 	if (resp->error_code) {
671 		netdev_err(bp->dev, "hwrm_fwd_resp error %d\n",
672 			   resp->error_code);
673 		rc = -1;
674 	}
675 
676 fwd_resp_exit:
677 	mutex_unlock(&bp->hwrm_cmd_lock);
678 	return rc;
679 }
680 
681 static int bnxt_hwrm_fwd_err_resp(struct bnxt *bp, struct bnxt_vf_info *vf,
682 				  u32 msg_size)
683 {
684 	int rc = 0;
685 	struct hwrm_reject_fwd_resp_input req = {0};
686 	struct hwrm_reject_fwd_resp_output *resp = bp->hwrm_cmd_resp_addr;
687 
688 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_REJECT_FWD_RESP, -1, -1);
689 	/* Set the new target id */
690 	req.target_id = cpu_to_le16(vf->fw_fid);
691 	req.encap_resp_target_id = cpu_to_le16(vf->fw_fid);
692 	memcpy(req.encap_request, vf->hwrm_cmd_req_addr, msg_size);
693 
694 	mutex_lock(&bp->hwrm_cmd_lock);
695 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
696 
697 	if (rc) {
698 		netdev_err(bp->dev, "hwrm_fwd_err_resp failed. rc:%d\n", rc);
699 		goto fwd_err_resp_exit;
700 	}
701 
702 	if (resp->error_code) {
703 		netdev_err(bp->dev, "hwrm_fwd_err_resp error %d\n",
704 			   resp->error_code);
705 		rc = -1;
706 	}
707 
708 fwd_err_resp_exit:
709 	mutex_unlock(&bp->hwrm_cmd_lock);
710 	return rc;
711 }
712 
713 static int bnxt_hwrm_exec_fwd_resp(struct bnxt *bp, struct bnxt_vf_info *vf,
714 				   u32 msg_size)
715 {
716 	int rc = 0;
717 	struct hwrm_exec_fwd_resp_input req = {0};
718 	struct hwrm_exec_fwd_resp_output *resp = bp->hwrm_cmd_resp_addr;
719 
720 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_EXEC_FWD_RESP, -1, -1);
721 	/* Set the new target id */
722 	req.target_id = cpu_to_le16(vf->fw_fid);
723 	req.encap_resp_target_id = cpu_to_le16(vf->fw_fid);
724 	memcpy(req.encap_request, vf->hwrm_cmd_req_addr, msg_size);
725 
726 	mutex_lock(&bp->hwrm_cmd_lock);
727 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
728 
729 	if (rc) {
730 		netdev_err(bp->dev, "hwrm_exec_fw_resp failed. rc:%d\n", rc);
731 		goto exec_fwd_resp_exit;
732 	}
733 
734 	if (resp->error_code) {
735 		netdev_err(bp->dev, "hwrm_exec_fw_resp error %d\n",
736 			   resp->error_code);
737 		rc = -1;
738 	}
739 
740 exec_fwd_resp_exit:
741 	mutex_unlock(&bp->hwrm_cmd_lock);
742 	return rc;
743 }
744 
745 static int bnxt_vf_validate_set_mac(struct bnxt *bp, struct bnxt_vf_info *vf)
746 {
747 	u32 msg_size = sizeof(struct hwrm_cfa_l2_filter_alloc_input);
748 	struct hwrm_cfa_l2_filter_alloc_input *req =
749 		(struct hwrm_cfa_l2_filter_alloc_input *)vf->hwrm_cmd_req_addr;
750 
751 	if (!is_valid_ether_addr(vf->mac_addr) ||
752 	    ether_addr_equal((const u8 *)req->l2_addr, vf->mac_addr))
753 		return bnxt_hwrm_exec_fwd_resp(bp, vf, msg_size);
754 	else
755 		return bnxt_hwrm_fwd_err_resp(bp, vf, msg_size);
756 }
757 
758 static int bnxt_vf_set_link(struct bnxt *bp, struct bnxt_vf_info *vf)
759 {
760 	int rc = 0;
761 
762 	if (!(vf->flags & BNXT_VF_LINK_FORCED)) {
763 		/* real link */
764 		rc = bnxt_hwrm_exec_fwd_resp(
765 			bp, vf, sizeof(struct hwrm_port_phy_qcfg_input));
766 	} else {
767 		struct hwrm_port_phy_qcfg_output phy_qcfg_resp;
768 		struct hwrm_port_phy_qcfg_input *phy_qcfg_req;
769 
770 		phy_qcfg_req =
771 		(struct hwrm_port_phy_qcfg_input *)vf->hwrm_cmd_req_addr;
772 		mutex_lock(&bp->hwrm_cmd_lock);
773 		memcpy(&phy_qcfg_resp, &bp->link_info.phy_qcfg_resp,
774 		       sizeof(phy_qcfg_resp));
775 		mutex_unlock(&bp->hwrm_cmd_lock);
776 		phy_qcfg_resp.seq_id = phy_qcfg_req->seq_id;
777 
778 		if (vf->flags & BNXT_VF_LINK_UP) {
779 			/* if physical link is down, force link up on VF */
780 			if (phy_qcfg_resp.link !=
781 			    PORT_PHY_QCFG_RESP_LINK_LINK) {
782 				phy_qcfg_resp.link =
783 					PORT_PHY_QCFG_RESP_LINK_LINK;
784 				phy_qcfg_resp.link_speed = cpu_to_le16(
785 					PORT_PHY_QCFG_RESP_LINK_SPEED_10GB);
786 				phy_qcfg_resp.duplex =
787 					PORT_PHY_QCFG_RESP_DUPLEX_FULL;
788 				phy_qcfg_resp.pause =
789 					(PORT_PHY_QCFG_RESP_PAUSE_TX |
790 					 PORT_PHY_QCFG_RESP_PAUSE_RX);
791 			}
792 		} else {
793 			/* force link down */
794 			phy_qcfg_resp.link = PORT_PHY_QCFG_RESP_LINK_NO_LINK;
795 			phy_qcfg_resp.link_speed = 0;
796 			phy_qcfg_resp.duplex = PORT_PHY_QCFG_RESP_DUPLEX_HALF;
797 			phy_qcfg_resp.pause = 0;
798 		}
799 		rc = bnxt_hwrm_fwd_resp(bp, vf, &phy_qcfg_resp,
800 					phy_qcfg_req->resp_addr,
801 					phy_qcfg_req->cmpl_ring,
802 					sizeof(phy_qcfg_resp));
803 	}
804 	return rc;
805 }
806 
807 static int bnxt_vf_req_validate_snd(struct bnxt *bp, struct bnxt_vf_info *vf)
808 {
809 	int rc = 0;
810 	struct input *encap_req = vf->hwrm_cmd_req_addr;
811 	u32 req_type = le16_to_cpu(encap_req->req_type);
812 
813 	switch (req_type) {
814 	case HWRM_CFA_L2_FILTER_ALLOC:
815 		rc = bnxt_vf_validate_set_mac(bp, vf);
816 		break;
817 	case HWRM_FUNC_CFG:
818 		/* TODO Validate if VF is allowed to change mac address,
819 		 * mtu, num of rings etc
820 		 */
821 		rc = bnxt_hwrm_exec_fwd_resp(
822 			bp, vf, sizeof(struct hwrm_func_cfg_input));
823 		break;
824 	case HWRM_PORT_PHY_QCFG:
825 		rc = bnxt_vf_set_link(bp, vf);
826 		break;
827 	default:
828 		break;
829 	}
830 	return rc;
831 }
832 
833 void bnxt_hwrm_exec_fwd_req(struct bnxt *bp)
834 {
835 	u32 i = 0, active_vfs = bp->pf.active_vfs, vf_id;
836 
837 	/* Scan through VF's and process commands */
838 	while (1) {
839 		vf_id = find_next_bit(bp->pf.vf_event_bmap, active_vfs, i);
840 		if (vf_id >= active_vfs)
841 			break;
842 
843 		clear_bit(vf_id, bp->pf.vf_event_bmap);
844 		bnxt_vf_req_validate_snd(bp, &bp->pf.vf[vf_id]);
845 		i = vf_id + 1;
846 	}
847 }
848 
849 void bnxt_update_vf_mac(struct bnxt *bp)
850 {
851 	struct hwrm_func_qcaps_input req = {0};
852 	struct hwrm_func_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
853 
854 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCAPS, -1, -1);
855 	req.fid = cpu_to_le16(0xffff);
856 
857 	mutex_lock(&bp->hwrm_cmd_lock);
858 	if (_hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT))
859 		goto update_vf_mac_exit;
860 
861 	/* Store MAC address from the firmware.  There are 2 cases:
862 	 * 1. MAC address is valid.  It is assigned from the PF and we
863 	 *    need to override the current VF MAC address with it.
864 	 * 2. MAC address is zero.  The VF will use a random MAC address by
865 	 *    default but the stored zero MAC will allow the VF user to change
866 	 *    the random MAC address using ndo_set_mac_address() if he wants.
867 	 */
868 	if (!ether_addr_equal(resp->mac_address, bp->vf.mac_addr))
869 		memcpy(bp->vf.mac_addr, resp->mac_address, ETH_ALEN);
870 
871 	/* overwrite netdev dev_addr with admin VF MAC */
872 	if (is_valid_ether_addr(bp->vf.mac_addr))
873 		memcpy(bp->dev->dev_addr, bp->vf.mac_addr, ETH_ALEN);
874 update_vf_mac_exit:
875 	mutex_unlock(&bp->hwrm_cmd_lock);
876 }
877 
878 int bnxt_approve_mac(struct bnxt *bp, u8 *mac)
879 {
880 	struct hwrm_func_vf_cfg_input req = {0};
881 	int rc = 0;
882 
883 	if (!BNXT_VF(bp))
884 		return 0;
885 
886 	if (bp->hwrm_spec_code < 0x10202) {
887 		if (is_valid_ether_addr(bp->vf.mac_addr))
888 			rc = -EADDRNOTAVAIL;
889 		goto mac_done;
890 	}
891 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_VF_CFG, -1, -1);
892 	req.enables = cpu_to_le32(FUNC_VF_CFG_REQ_ENABLES_DFLT_MAC_ADDR);
893 	memcpy(req.dflt_mac_addr, mac, ETH_ALEN);
894 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
895 mac_done:
896 	if (rc) {
897 		rc = -EADDRNOTAVAIL;
898 		netdev_warn(bp->dev, "VF MAC address %pM not approved by the PF\n",
899 			    mac);
900 	}
901 	return rc;
902 }
903 #else
904 
905 void bnxt_sriov_disable(struct bnxt *bp)
906 {
907 }
908 
909 void bnxt_hwrm_exec_fwd_req(struct bnxt *bp)
910 {
911 	netdev_err(bp->dev, "Invalid VF message received when SRIOV is not enable\n");
912 }
913 
914 void bnxt_update_vf_mac(struct bnxt *bp)
915 {
916 }
917 
918 int bnxt_approve_mac(struct bnxt *bp, u8 *mac)
919 {
920 	return 0;
921 }
922 #endif
923