xref: /linux/drivers/net/ethernet/broadcom/bnxt/bnxt_sriov.c (revision 0883c2c06fb5bcf5b9e008270827e63c09a88c1e)
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_vf_ndo_prep(struct bnxt *bp, int vf_id)
23 {
24 	if (!test_bit(BNXT_STATE_OPEN, &bp->state)) {
25 		netdev_err(bp->dev, "vf ndo called though PF is down\n");
26 		return -EINVAL;
27 	}
28 	if (!bp->pf.active_vfs) {
29 		netdev_err(bp->dev, "vf ndo called though sriov is disabled\n");
30 		return -EINVAL;
31 	}
32 	if (vf_id >= bp->pf.max_vfs) {
33 		netdev_err(bp->dev, "Invalid VF id %d\n", vf_id);
34 		return -EINVAL;
35 	}
36 	return 0;
37 }
38 
39 int bnxt_set_vf_spoofchk(struct net_device *dev, int vf_id, bool setting)
40 {
41 	struct hwrm_func_cfg_input req = {0};
42 	struct bnxt *bp = netdev_priv(dev);
43 	struct bnxt_vf_info *vf;
44 	bool old_setting = false;
45 	u32 func_flags;
46 	int rc;
47 
48 	rc = bnxt_vf_ndo_prep(bp, vf_id);
49 	if (rc)
50 		return rc;
51 
52 	vf = &bp->pf.vf[vf_id];
53 	if (vf->flags & BNXT_VF_SPOOFCHK)
54 		old_setting = true;
55 	if (old_setting == setting)
56 		return 0;
57 
58 	func_flags = vf->func_flags;
59 	if (setting)
60 		func_flags |= FUNC_CFG_REQ_FLAGS_SRC_MAC_ADDR_CHECK;
61 	else
62 		func_flags &= ~FUNC_CFG_REQ_FLAGS_SRC_MAC_ADDR_CHECK;
63 	/*TODO: if the driver supports VLAN filter on guest VLAN,
64 	 * the spoof check should also include vlan anti-spoofing
65 	 */
66 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
67 	req.fid = cpu_to_le16(vf->fw_fid);
68 	req.flags = cpu_to_le32(func_flags);
69 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
70 	if (!rc) {
71 		vf->func_flags = func_flags;
72 		if (setting)
73 			vf->flags |= BNXT_VF_SPOOFCHK;
74 		else
75 			vf->flags &= ~BNXT_VF_SPOOFCHK;
76 	}
77 	return rc;
78 }
79 
80 int bnxt_get_vf_config(struct net_device *dev, int vf_id,
81 		       struct ifla_vf_info *ivi)
82 {
83 	struct bnxt *bp = netdev_priv(dev);
84 	struct bnxt_vf_info *vf;
85 	int rc;
86 
87 	rc = bnxt_vf_ndo_prep(bp, vf_id);
88 	if (rc)
89 		return rc;
90 
91 	ivi->vf = vf_id;
92 	vf = &bp->pf.vf[vf_id];
93 
94 	memcpy(&ivi->mac, vf->mac_addr, ETH_ALEN);
95 	ivi->max_tx_rate = vf->max_tx_rate;
96 	ivi->min_tx_rate = vf->min_tx_rate;
97 	ivi->vlan = vf->vlan;
98 	ivi->qos = vf->flags & BNXT_VF_QOS;
99 	ivi->spoofchk = vf->flags & BNXT_VF_SPOOFCHK;
100 	if (!(vf->flags & BNXT_VF_LINK_FORCED))
101 		ivi->linkstate = IFLA_VF_LINK_STATE_AUTO;
102 	else if (vf->flags & BNXT_VF_LINK_UP)
103 		ivi->linkstate = IFLA_VF_LINK_STATE_ENABLE;
104 	else
105 		ivi->linkstate = IFLA_VF_LINK_STATE_DISABLE;
106 
107 	return 0;
108 }
109 
110 int bnxt_set_vf_mac(struct net_device *dev, int vf_id, u8 *mac)
111 {
112 	struct hwrm_func_cfg_input req = {0};
113 	struct bnxt *bp = netdev_priv(dev);
114 	struct bnxt_vf_info *vf;
115 	int rc;
116 
117 	rc = bnxt_vf_ndo_prep(bp, vf_id);
118 	if (rc)
119 		return rc;
120 	/* reject bc or mc mac addr, zero mac addr means allow
121 	 * VF to use its own mac addr
122 	 */
123 	if (is_multicast_ether_addr(mac)) {
124 		netdev_err(dev, "Invalid VF ethernet address\n");
125 		return -EINVAL;
126 	}
127 	vf = &bp->pf.vf[vf_id];
128 
129 	memcpy(vf->mac_addr, mac, ETH_ALEN);
130 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
131 	req.fid = cpu_to_le16(vf->fw_fid);
132 	req.flags = cpu_to_le32(vf->func_flags);
133 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_DFLT_MAC_ADDR);
134 	memcpy(req.dflt_mac_addr, mac, ETH_ALEN);
135 	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
136 }
137 
138 int bnxt_set_vf_vlan(struct net_device *dev, int vf_id, u16 vlan_id, u8 qos)
139 {
140 	struct hwrm_func_cfg_input req = {0};
141 	struct bnxt *bp = netdev_priv(dev);
142 	struct bnxt_vf_info *vf;
143 	u16 vlan_tag;
144 	int rc;
145 
146 	rc = bnxt_vf_ndo_prep(bp, vf_id);
147 	if (rc)
148 		return rc;
149 
150 	/* TODO: needed to implement proper handling of user priority,
151 	 * currently fail the command if there is valid priority
152 	 */
153 	if (vlan_id > 4095 || qos)
154 		return -EINVAL;
155 
156 	vf = &bp->pf.vf[vf_id];
157 	vlan_tag = vlan_id;
158 	if (vlan_tag == vf->vlan)
159 		return 0;
160 
161 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
162 	req.fid = cpu_to_le16(vf->fw_fid);
163 	req.flags = cpu_to_le32(vf->func_flags);
164 	req.dflt_vlan = cpu_to_le16(vlan_tag);
165 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_DFLT_VLAN);
166 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
167 	if (!rc)
168 		vf->vlan = vlan_tag;
169 	return rc;
170 }
171 
172 int bnxt_set_vf_bw(struct net_device *dev, int vf_id, int min_tx_rate,
173 		   int max_tx_rate)
174 {
175 	struct hwrm_func_cfg_input req = {0};
176 	struct bnxt *bp = netdev_priv(dev);
177 	struct bnxt_vf_info *vf;
178 	u32 pf_link_speed;
179 	int rc;
180 
181 	rc = bnxt_vf_ndo_prep(bp, vf_id);
182 	if (rc)
183 		return rc;
184 
185 	vf = &bp->pf.vf[vf_id];
186 	pf_link_speed = bnxt_fw_to_ethtool_speed(bp->link_info.link_speed);
187 	if (max_tx_rate > pf_link_speed) {
188 		netdev_info(bp->dev, "max tx rate %d exceed PF link speed for VF %d\n",
189 			    max_tx_rate, vf_id);
190 		return -EINVAL;
191 	}
192 
193 	if (min_tx_rate > pf_link_speed || min_tx_rate > max_tx_rate) {
194 		netdev_info(bp->dev, "min tx rate %d is invalid for VF %d\n",
195 			    min_tx_rate, vf_id);
196 		return -EINVAL;
197 	}
198 	if (min_tx_rate == vf->min_tx_rate && max_tx_rate == vf->max_tx_rate)
199 		return 0;
200 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
201 	req.fid = cpu_to_le16(vf->fw_fid);
202 	req.flags = cpu_to_le32(vf->func_flags);
203 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_MAX_BW);
204 	req.max_bw = cpu_to_le32(max_tx_rate);
205 	req.enables |= cpu_to_le32(FUNC_CFG_REQ_ENABLES_MIN_BW);
206 	req.min_bw = cpu_to_le32(min_tx_rate);
207 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
208 	if (!rc) {
209 		vf->min_tx_rate = min_tx_rate;
210 		vf->max_tx_rate = max_tx_rate;
211 	}
212 	return rc;
213 }
214 
215 int bnxt_set_vf_link_state(struct net_device *dev, int vf_id, int link)
216 {
217 	struct bnxt *bp = netdev_priv(dev);
218 	struct bnxt_vf_info *vf;
219 	int rc;
220 
221 	rc = bnxt_vf_ndo_prep(bp, vf_id);
222 	if (rc)
223 		return rc;
224 
225 	vf = &bp->pf.vf[vf_id];
226 
227 	vf->flags &= ~(BNXT_VF_LINK_UP | BNXT_VF_LINK_FORCED);
228 	switch (link) {
229 	case IFLA_VF_LINK_STATE_AUTO:
230 		vf->flags |= BNXT_VF_LINK_UP;
231 		break;
232 	case IFLA_VF_LINK_STATE_DISABLE:
233 		vf->flags |= BNXT_VF_LINK_FORCED;
234 		break;
235 	case IFLA_VF_LINK_STATE_ENABLE:
236 		vf->flags |= BNXT_VF_LINK_UP | BNXT_VF_LINK_FORCED;
237 		break;
238 	default:
239 		netdev_err(bp->dev, "Invalid link option\n");
240 		rc = -EINVAL;
241 		break;
242 	}
243 	/* CHIMP TODO: send msg to VF to update new link state */
244 
245 	return rc;
246 }
247 
248 static int bnxt_set_vf_attr(struct bnxt *bp, int num_vfs)
249 {
250 	int i;
251 	struct bnxt_vf_info *vf;
252 
253 	for (i = 0; i < num_vfs; i++) {
254 		vf = &bp->pf.vf[i];
255 		memset(vf, 0, sizeof(*vf));
256 		vf->flags = BNXT_VF_QOS | BNXT_VF_LINK_UP;
257 	}
258 	return 0;
259 }
260 
261 static int bnxt_hwrm_func_vf_resource_free(struct bnxt *bp, int num_vfs)
262 {
263 	int i, rc = 0;
264 	struct bnxt_pf_info *pf = &bp->pf;
265 	struct hwrm_func_vf_resc_free_input req = {0};
266 
267 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_VF_RESC_FREE, -1, -1);
268 
269 	mutex_lock(&bp->hwrm_cmd_lock);
270 	for (i = pf->first_vf_id; i < pf->first_vf_id + num_vfs; i++) {
271 		req.vf_id = cpu_to_le16(i);
272 		rc = _hwrm_send_message(bp, &req, sizeof(req),
273 					HWRM_CMD_TIMEOUT);
274 		if (rc)
275 			break;
276 	}
277 	mutex_unlock(&bp->hwrm_cmd_lock);
278 	return rc;
279 }
280 
281 static void bnxt_free_vf_resources(struct bnxt *bp)
282 {
283 	struct pci_dev *pdev = bp->pdev;
284 	int i;
285 
286 	kfree(bp->pf.vf_event_bmap);
287 	bp->pf.vf_event_bmap = NULL;
288 
289 	for (i = 0; i < 4; i++) {
290 		if (bp->pf.hwrm_cmd_req_addr[i]) {
291 			dma_free_coherent(&pdev->dev, BNXT_PAGE_SIZE,
292 					  bp->pf.hwrm_cmd_req_addr[i],
293 					  bp->pf.hwrm_cmd_req_dma_addr[i]);
294 			bp->pf.hwrm_cmd_req_addr[i] = NULL;
295 		}
296 	}
297 
298 	kfree(bp->pf.vf);
299 	bp->pf.vf = NULL;
300 }
301 
302 static int bnxt_alloc_vf_resources(struct bnxt *bp, int num_vfs)
303 {
304 	struct pci_dev *pdev = bp->pdev;
305 	u32 nr_pages, size, i, j, k = 0;
306 
307 	bp->pf.vf = kcalloc(num_vfs, sizeof(struct bnxt_vf_info), GFP_KERNEL);
308 	if (!bp->pf.vf)
309 		return -ENOMEM;
310 
311 	bnxt_set_vf_attr(bp, num_vfs);
312 
313 	size = num_vfs * BNXT_HWRM_REQ_MAX_SIZE;
314 	nr_pages = size / BNXT_PAGE_SIZE;
315 	if (size & (BNXT_PAGE_SIZE - 1))
316 		nr_pages++;
317 
318 	for (i = 0; i < nr_pages; i++) {
319 		bp->pf.hwrm_cmd_req_addr[i] =
320 			dma_alloc_coherent(&pdev->dev, BNXT_PAGE_SIZE,
321 					   &bp->pf.hwrm_cmd_req_dma_addr[i],
322 					   GFP_KERNEL);
323 
324 		if (!bp->pf.hwrm_cmd_req_addr[i])
325 			return -ENOMEM;
326 
327 		for (j = 0; j < BNXT_HWRM_REQS_PER_PAGE && k < num_vfs; j++) {
328 			struct bnxt_vf_info *vf = &bp->pf.vf[k];
329 
330 			vf->hwrm_cmd_req_addr = bp->pf.hwrm_cmd_req_addr[i] +
331 						j * BNXT_HWRM_REQ_MAX_SIZE;
332 			vf->hwrm_cmd_req_dma_addr =
333 				bp->pf.hwrm_cmd_req_dma_addr[i] + j *
334 				BNXT_HWRM_REQ_MAX_SIZE;
335 			k++;
336 		}
337 	}
338 
339 	/* Max 128 VF's */
340 	bp->pf.vf_event_bmap = kzalloc(16, GFP_KERNEL);
341 	if (!bp->pf.vf_event_bmap)
342 		return -ENOMEM;
343 
344 	bp->pf.hwrm_cmd_req_pages = nr_pages;
345 	return 0;
346 }
347 
348 static int bnxt_hwrm_func_buf_rgtr(struct bnxt *bp)
349 {
350 	struct hwrm_func_buf_rgtr_input req = {0};
351 
352 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_BUF_RGTR, -1, -1);
353 
354 	req.req_buf_num_pages = cpu_to_le16(bp->pf.hwrm_cmd_req_pages);
355 	req.req_buf_page_size = cpu_to_le16(BNXT_PAGE_SHIFT);
356 	req.req_buf_len = cpu_to_le16(BNXT_HWRM_REQ_MAX_SIZE);
357 	req.req_buf_page_addr0 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[0]);
358 	req.req_buf_page_addr1 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[1]);
359 	req.req_buf_page_addr2 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[2]);
360 	req.req_buf_page_addr3 = cpu_to_le64(bp->pf.hwrm_cmd_req_dma_addr[3]);
361 
362 	return hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
363 }
364 
365 /* only call by PF to reserve resources for VF */
366 static int bnxt_hwrm_func_cfg(struct bnxt *bp, int num_vfs)
367 {
368 	u32 rc = 0, mtu, i;
369 	u16 vf_tx_rings, vf_rx_rings, vf_cp_rings, vf_stat_ctx, vf_vnics;
370 	u16 vf_ring_grps;
371 	struct hwrm_func_cfg_input req = {0};
372 	struct bnxt_pf_info *pf = &bp->pf;
373 
374 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_CFG, -1, -1);
375 
376 	/* Remaining rings are distributed equally amongs VF's for now */
377 	/* TODO: the following workaroud is needed to restrict total number
378 	 * of vf_cp_rings not exceed number of HW ring groups. This WA should
379 	 * be removed once new HWRM provides HW ring groups capability in
380 	 * hwrm_func_qcap.
381 	 */
382 	vf_cp_rings = min_t(u16, pf->max_cp_rings, pf->max_stat_ctxs);
383 	vf_cp_rings = (vf_cp_rings - bp->cp_nr_rings) / num_vfs;
384 	/* TODO: restore this logic below once the WA above is removed */
385 	/* vf_cp_rings = (pf->max_cp_rings - bp->cp_nr_rings) / num_vfs; */
386 	vf_stat_ctx = (pf->max_stat_ctxs - bp->num_stat_ctxs) / num_vfs;
387 	if (bp->flags & BNXT_FLAG_AGG_RINGS)
388 		vf_rx_rings = (pf->max_rx_rings - bp->rx_nr_rings * 2) /
389 			      num_vfs;
390 	else
391 		vf_rx_rings = (pf->max_rx_rings - bp->rx_nr_rings) / num_vfs;
392 	vf_ring_grps = (bp->pf.max_hw_ring_grps - bp->rx_nr_rings) / num_vfs;
393 	vf_tx_rings = (pf->max_tx_rings - bp->tx_nr_rings) / num_vfs;
394 
395 	req.enables = cpu_to_le32(FUNC_CFG_REQ_ENABLES_MTU |
396 				  FUNC_CFG_REQ_ENABLES_MRU |
397 				  FUNC_CFG_REQ_ENABLES_NUM_RSSCOS_CTXS |
398 				  FUNC_CFG_REQ_ENABLES_NUM_STAT_CTXS |
399 				  FUNC_CFG_REQ_ENABLES_NUM_CMPL_RINGS |
400 				  FUNC_CFG_REQ_ENABLES_NUM_TX_RINGS |
401 				  FUNC_CFG_REQ_ENABLES_NUM_RX_RINGS |
402 				  FUNC_CFG_REQ_ENABLES_NUM_L2_CTXS |
403 				  FUNC_CFG_REQ_ENABLES_NUM_VNICS |
404 				  FUNC_CFG_REQ_ENABLES_NUM_HW_RING_GRPS);
405 
406 	mtu = bp->dev->mtu + ETH_HLEN + ETH_FCS_LEN + VLAN_HLEN;
407 	req.mru = cpu_to_le16(mtu);
408 	req.mtu = cpu_to_le16(mtu);
409 
410 	req.num_rsscos_ctxs = cpu_to_le16(1);
411 	req.num_cmpl_rings = cpu_to_le16(vf_cp_rings);
412 	req.num_tx_rings = cpu_to_le16(vf_tx_rings);
413 	req.num_rx_rings = cpu_to_le16(vf_rx_rings);
414 	req.num_hw_ring_grps = cpu_to_le16(vf_ring_grps);
415 	req.num_l2_ctxs = cpu_to_le16(4);
416 	vf_vnics = 1;
417 
418 	req.num_vnics = cpu_to_le16(vf_vnics);
419 	/* FIXME spec currently uses 1 bit for stats ctx */
420 	req.num_stat_ctxs = cpu_to_le16(vf_stat_ctx);
421 
422 	mutex_lock(&bp->hwrm_cmd_lock);
423 	for (i = 0; i < num_vfs; i++) {
424 		req.fid = cpu_to_le16(pf->first_vf_id + i);
425 		rc = _hwrm_send_message(bp, &req, sizeof(req),
426 					HWRM_CMD_TIMEOUT);
427 		if (rc)
428 			break;
429 		pf->active_vfs = i + 1;
430 		pf->vf[i].fw_fid = le16_to_cpu(req.fid);
431 	}
432 	mutex_unlock(&bp->hwrm_cmd_lock);
433 	if (!rc) {
434 		pf->max_tx_rings -= vf_tx_rings * num_vfs;
435 		pf->max_rx_rings -= vf_rx_rings * num_vfs;
436 		pf->max_hw_ring_grps -= vf_ring_grps * num_vfs;
437 		pf->max_cp_rings -= vf_cp_rings * num_vfs;
438 		pf->max_rsscos_ctxs -= num_vfs;
439 		pf->max_stat_ctxs -= vf_stat_ctx * num_vfs;
440 		pf->max_vnics -= vf_vnics * num_vfs;
441 	}
442 	return rc;
443 }
444 
445 static int bnxt_sriov_enable(struct bnxt *bp, int *num_vfs)
446 {
447 	int rc = 0, vfs_supported;
448 	int min_rx_rings, min_tx_rings, min_rss_ctxs;
449 	int tx_ok = 0, rx_ok = 0, rss_ok = 0;
450 
451 	/* Check if we can enable requested num of vf's. At a mininum
452 	 * we require 1 RX 1 TX rings for each VF. In this minimum conf
453 	 * features like TPA will not be available.
454 	 */
455 	vfs_supported = *num_vfs;
456 
457 	while (vfs_supported) {
458 		min_rx_rings = vfs_supported;
459 		min_tx_rings = vfs_supported;
460 		min_rss_ctxs = vfs_supported;
461 
462 		if (bp->flags & BNXT_FLAG_AGG_RINGS) {
463 			if (bp->pf.max_rx_rings - bp->rx_nr_rings * 2 >=
464 			    min_rx_rings)
465 				rx_ok = 1;
466 		} else {
467 			if (bp->pf.max_rx_rings - bp->rx_nr_rings >=
468 			    min_rx_rings)
469 				rx_ok = 1;
470 		}
471 
472 		if (bp->pf.max_tx_rings - bp->tx_nr_rings >= min_tx_rings)
473 			tx_ok = 1;
474 
475 		if (bp->pf.max_rsscos_ctxs - bp->rsscos_nr_ctxs >= min_rss_ctxs)
476 			rss_ok = 1;
477 
478 		if (tx_ok && rx_ok && rss_ok)
479 			break;
480 
481 		vfs_supported--;
482 	}
483 
484 	if (!vfs_supported) {
485 		netdev_err(bp->dev, "Cannot enable VF's as all resources are used by PF\n");
486 		return -EINVAL;
487 	}
488 
489 	if (vfs_supported != *num_vfs) {
490 		netdev_info(bp->dev, "Requested VFs %d, can enable %d\n",
491 			    *num_vfs, vfs_supported);
492 		*num_vfs = vfs_supported;
493 	}
494 
495 	rc = bnxt_alloc_vf_resources(bp, *num_vfs);
496 	if (rc)
497 		goto err_out1;
498 
499 	/* Reserve resources for VFs */
500 	rc = bnxt_hwrm_func_cfg(bp, *num_vfs);
501 	if (rc)
502 		goto err_out2;
503 
504 	/* Register buffers for VFs */
505 	rc = bnxt_hwrm_func_buf_rgtr(bp);
506 	if (rc)
507 		goto err_out2;
508 
509 	rc = pci_enable_sriov(bp->pdev, *num_vfs);
510 	if (rc)
511 		goto err_out2;
512 
513 	return 0;
514 
515 err_out2:
516 	/* Free the resources reserved for various VF's */
517 	bnxt_hwrm_func_vf_resource_free(bp, *num_vfs);
518 
519 err_out1:
520 	bnxt_free_vf_resources(bp);
521 
522 	return rc;
523 }
524 
525 static int bnxt_hwrm_fwd_async_event_cmpl(struct bnxt *bp,
526 					  struct bnxt_vf_info *vf,
527 					  u16 event_id)
528 {
529 	int rc = 0;
530 	struct hwrm_fwd_async_event_cmpl_input req = {0};
531 	struct hwrm_fwd_async_event_cmpl_output *resp = bp->hwrm_cmd_resp_addr;
532 	struct hwrm_async_event_cmpl *async_cmpl;
533 
534 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FWD_ASYNC_EVENT_CMPL, -1, -1);
535 	if (vf)
536 		req.encap_async_event_target_id = cpu_to_le16(vf->fw_fid);
537 	else
538 		/* broadcast this async event to all VFs */
539 		req.encap_async_event_target_id = cpu_to_le16(0xffff);
540 	async_cmpl = (struct hwrm_async_event_cmpl *)req.encap_async_event_cmpl;
541 	async_cmpl->type =
542 		cpu_to_le16(HWRM_ASYNC_EVENT_CMPL_TYPE_HWRM_ASYNC_EVENT);
543 	async_cmpl->event_id = cpu_to_le16(event_id);
544 
545 	mutex_lock(&bp->hwrm_cmd_lock);
546 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
547 
548 	if (rc) {
549 		netdev_err(bp->dev, "hwrm_fwd_async_event_cmpl failed. rc:%d\n",
550 			   rc);
551 		goto fwd_async_event_cmpl_exit;
552 	}
553 
554 	if (resp->error_code) {
555 		netdev_err(bp->dev, "hwrm_fwd_async_event_cmpl error %d\n",
556 			   resp->error_code);
557 		rc = -1;
558 	}
559 
560 fwd_async_event_cmpl_exit:
561 	mutex_unlock(&bp->hwrm_cmd_lock);
562 	return rc;
563 }
564 
565 void bnxt_sriov_disable(struct bnxt *bp)
566 {
567 	u16 num_vfs = pci_num_vf(bp->pdev);
568 
569 	if (!num_vfs)
570 		return;
571 
572 	if (pci_vfs_assigned(bp->pdev)) {
573 		bnxt_hwrm_fwd_async_event_cmpl(
574 			bp, NULL,
575 			HWRM_ASYNC_EVENT_CMPL_EVENT_ID_PF_DRVR_UNLOAD);
576 		netdev_warn(bp->dev, "Unable to free %d VFs because some are assigned to VMs.\n",
577 			    num_vfs);
578 	} else {
579 		pci_disable_sriov(bp->pdev);
580 		/* Free the HW resources reserved for various VF's */
581 		bnxt_hwrm_func_vf_resource_free(bp, num_vfs);
582 	}
583 
584 	bnxt_free_vf_resources(bp);
585 
586 	bp->pf.active_vfs = 0;
587 	/* Reclaim all resources for the PF. */
588 	bnxt_hwrm_func_qcaps(bp);
589 }
590 
591 int bnxt_sriov_configure(struct pci_dev *pdev, int num_vfs)
592 {
593 	struct net_device *dev = pci_get_drvdata(pdev);
594 	struct bnxt *bp = netdev_priv(dev);
595 
596 	if (!(bp->flags & BNXT_FLAG_USING_MSIX)) {
597 		netdev_warn(dev, "Not allow SRIOV if the irq mode is not MSIX\n");
598 		return 0;
599 	}
600 
601 	rtnl_lock();
602 	if (!netif_running(dev)) {
603 		netdev_warn(dev, "Reject SRIOV config request since if is down!\n");
604 		rtnl_unlock();
605 		return 0;
606 	}
607 	bp->sriov_cfg = true;
608 	rtnl_unlock();
609 
610 	if (pci_vfs_assigned(bp->pdev)) {
611 		netdev_warn(dev, "Unable to configure SRIOV since some VFs are assigned to VMs.\n");
612 		num_vfs = 0;
613 		goto sriov_cfg_exit;
614 	}
615 
616 	/* Check if enabled VFs is same as requested */
617 	if (num_vfs && num_vfs == bp->pf.active_vfs)
618 		goto sriov_cfg_exit;
619 
620 	/* if there are previous existing VFs, clean them up */
621 	bnxt_sriov_disable(bp);
622 	if (!num_vfs)
623 		goto sriov_cfg_exit;
624 
625 	bnxt_sriov_enable(bp, &num_vfs);
626 
627 sriov_cfg_exit:
628 	bp->sriov_cfg = false;
629 	wake_up(&bp->sriov_cfg_wait);
630 
631 	return num_vfs;
632 }
633 
634 static int bnxt_hwrm_fwd_resp(struct bnxt *bp, struct bnxt_vf_info *vf,
635 			      void *encap_resp, __le64 encap_resp_addr,
636 			      __le16 encap_resp_cpr, u32 msg_size)
637 {
638 	int rc = 0;
639 	struct hwrm_fwd_resp_input req = {0};
640 	struct hwrm_fwd_resp_output *resp = bp->hwrm_cmd_resp_addr;
641 
642 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FWD_RESP, -1, -1);
643 
644 	/* Set the new target id */
645 	req.target_id = cpu_to_le16(vf->fw_fid);
646 	req.encap_resp_target_id = cpu_to_le16(vf->fw_fid);
647 	req.encap_resp_len = cpu_to_le16(msg_size);
648 	req.encap_resp_addr = encap_resp_addr;
649 	req.encap_resp_cmpl_ring = encap_resp_cpr;
650 	memcpy(req.encap_resp, encap_resp, msg_size);
651 
652 	mutex_lock(&bp->hwrm_cmd_lock);
653 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
654 
655 	if (rc) {
656 		netdev_err(bp->dev, "hwrm_fwd_resp failed. rc:%d\n", rc);
657 		goto fwd_resp_exit;
658 	}
659 
660 	if (resp->error_code) {
661 		netdev_err(bp->dev, "hwrm_fwd_resp error %d\n",
662 			   resp->error_code);
663 		rc = -1;
664 	}
665 
666 fwd_resp_exit:
667 	mutex_unlock(&bp->hwrm_cmd_lock);
668 	return rc;
669 }
670 
671 static int bnxt_hwrm_fwd_err_resp(struct bnxt *bp, struct bnxt_vf_info *vf,
672 				  u32 msg_size)
673 {
674 	int rc = 0;
675 	struct hwrm_reject_fwd_resp_input req = {0};
676 	struct hwrm_reject_fwd_resp_output *resp = bp->hwrm_cmd_resp_addr;
677 
678 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_REJECT_FWD_RESP, -1, -1);
679 	/* Set the new target id */
680 	req.target_id = cpu_to_le16(vf->fw_fid);
681 	req.encap_resp_target_id = cpu_to_le16(vf->fw_fid);
682 	memcpy(req.encap_request, vf->hwrm_cmd_req_addr, msg_size);
683 
684 	mutex_lock(&bp->hwrm_cmd_lock);
685 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
686 
687 	if (rc) {
688 		netdev_err(bp->dev, "hwrm_fwd_err_resp failed. rc:%d\n", rc);
689 		goto fwd_err_resp_exit;
690 	}
691 
692 	if (resp->error_code) {
693 		netdev_err(bp->dev, "hwrm_fwd_err_resp error %d\n",
694 			   resp->error_code);
695 		rc = -1;
696 	}
697 
698 fwd_err_resp_exit:
699 	mutex_unlock(&bp->hwrm_cmd_lock);
700 	return rc;
701 }
702 
703 static int bnxt_hwrm_exec_fwd_resp(struct bnxt *bp, struct bnxt_vf_info *vf,
704 				   u32 msg_size)
705 {
706 	int rc = 0;
707 	struct hwrm_exec_fwd_resp_input req = {0};
708 	struct hwrm_exec_fwd_resp_output *resp = bp->hwrm_cmd_resp_addr;
709 
710 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_EXEC_FWD_RESP, -1, -1);
711 	/* Set the new target id */
712 	req.target_id = cpu_to_le16(vf->fw_fid);
713 	req.encap_resp_target_id = cpu_to_le16(vf->fw_fid);
714 	memcpy(req.encap_request, vf->hwrm_cmd_req_addr, msg_size);
715 
716 	mutex_lock(&bp->hwrm_cmd_lock);
717 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
718 
719 	if (rc) {
720 		netdev_err(bp->dev, "hwrm_exec_fw_resp failed. rc:%d\n", rc);
721 		goto exec_fwd_resp_exit;
722 	}
723 
724 	if (resp->error_code) {
725 		netdev_err(bp->dev, "hwrm_exec_fw_resp error %d\n",
726 			   resp->error_code);
727 		rc = -1;
728 	}
729 
730 exec_fwd_resp_exit:
731 	mutex_unlock(&bp->hwrm_cmd_lock);
732 	return rc;
733 }
734 
735 static int bnxt_vf_validate_set_mac(struct bnxt *bp, struct bnxt_vf_info *vf)
736 {
737 	u32 msg_size = sizeof(struct hwrm_cfa_l2_filter_alloc_input);
738 	struct hwrm_cfa_l2_filter_alloc_input *req =
739 		(struct hwrm_cfa_l2_filter_alloc_input *)vf->hwrm_cmd_req_addr;
740 
741 	if (!is_valid_ether_addr(vf->mac_addr) ||
742 	    ether_addr_equal((const u8 *)req->l2_addr, vf->mac_addr))
743 		return bnxt_hwrm_exec_fwd_resp(bp, vf, msg_size);
744 	else
745 		return bnxt_hwrm_fwd_err_resp(bp, vf, msg_size);
746 }
747 
748 static int bnxt_vf_set_link(struct bnxt *bp, struct bnxt_vf_info *vf)
749 {
750 	int rc = 0;
751 
752 	if (!(vf->flags & BNXT_VF_LINK_FORCED)) {
753 		/* real link */
754 		rc = bnxt_hwrm_exec_fwd_resp(
755 			bp, vf, sizeof(struct hwrm_port_phy_qcfg_input));
756 	} else {
757 		struct hwrm_port_phy_qcfg_output phy_qcfg_resp;
758 		struct hwrm_port_phy_qcfg_input *phy_qcfg_req;
759 
760 		phy_qcfg_req =
761 		(struct hwrm_port_phy_qcfg_input *)vf->hwrm_cmd_req_addr;
762 		mutex_lock(&bp->hwrm_cmd_lock);
763 		memcpy(&phy_qcfg_resp, &bp->link_info.phy_qcfg_resp,
764 		       sizeof(phy_qcfg_resp));
765 		mutex_unlock(&bp->hwrm_cmd_lock);
766 		phy_qcfg_resp.seq_id = phy_qcfg_req->seq_id;
767 
768 		if (vf->flags & BNXT_VF_LINK_UP) {
769 			/* if physical link is down, force link up on VF */
770 			if (phy_qcfg_resp.link ==
771 			    PORT_PHY_QCFG_RESP_LINK_NO_LINK) {
772 				phy_qcfg_resp.link =
773 					PORT_PHY_QCFG_RESP_LINK_LINK;
774 				phy_qcfg_resp.link_speed = cpu_to_le16(
775 					PORT_PHY_QCFG_RESP_LINK_SPEED_10GB);
776 				phy_qcfg_resp.duplex =
777 					PORT_PHY_QCFG_RESP_DUPLEX_FULL;
778 				phy_qcfg_resp.pause =
779 					(PORT_PHY_QCFG_RESP_PAUSE_TX |
780 					 PORT_PHY_QCFG_RESP_PAUSE_RX);
781 			}
782 		} else {
783 			/* force link down */
784 			phy_qcfg_resp.link = PORT_PHY_QCFG_RESP_LINK_NO_LINK;
785 			phy_qcfg_resp.link_speed = 0;
786 			phy_qcfg_resp.duplex = PORT_PHY_QCFG_RESP_DUPLEX_HALF;
787 			phy_qcfg_resp.pause = 0;
788 		}
789 		rc = bnxt_hwrm_fwd_resp(bp, vf, &phy_qcfg_resp,
790 					phy_qcfg_req->resp_addr,
791 					phy_qcfg_req->cmpl_ring,
792 					sizeof(phy_qcfg_resp));
793 	}
794 	return rc;
795 }
796 
797 static int bnxt_vf_req_validate_snd(struct bnxt *bp, struct bnxt_vf_info *vf)
798 {
799 	int rc = 0;
800 	struct input *encap_req = vf->hwrm_cmd_req_addr;
801 	u32 req_type = le16_to_cpu(encap_req->req_type);
802 
803 	switch (req_type) {
804 	case HWRM_CFA_L2_FILTER_ALLOC:
805 		rc = bnxt_vf_validate_set_mac(bp, vf);
806 		break;
807 	case HWRM_FUNC_CFG:
808 		/* TODO Validate if VF is allowed to change mac address,
809 		 * mtu, num of rings etc
810 		 */
811 		rc = bnxt_hwrm_exec_fwd_resp(
812 			bp, vf, sizeof(struct hwrm_func_cfg_input));
813 		break;
814 	case HWRM_PORT_PHY_QCFG:
815 		rc = bnxt_vf_set_link(bp, vf);
816 		break;
817 	default:
818 		break;
819 	}
820 	return rc;
821 }
822 
823 void bnxt_hwrm_exec_fwd_req(struct bnxt *bp)
824 {
825 	u32 i = 0, active_vfs = bp->pf.active_vfs, vf_id;
826 
827 	/* Scan through VF's and process commands */
828 	while (1) {
829 		vf_id = find_next_bit(bp->pf.vf_event_bmap, active_vfs, i);
830 		if (vf_id >= active_vfs)
831 			break;
832 
833 		clear_bit(vf_id, bp->pf.vf_event_bmap);
834 		bnxt_vf_req_validate_snd(bp, &bp->pf.vf[vf_id]);
835 		i = vf_id + 1;
836 	}
837 }
838 
839 void bnxt_update_vf_mac(struct bnxt *bp)
840 {
841 	struct hwrm_func_qcaps_input req = {0};
842 	struct hwrm_func_qcaps_output *resp = bp->hwrm_cmd_resp_addr;
843 
844 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCAPS, -1, -1);
845 	req.fid = cpu_to_le16(0xffff);
846 
847 	mutex_lock(&bp->hwrm_cmd_lock);
848 	if (_hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT))
849 		goto update_vf_mac_exit;
850 
851 	/* Store MAC address from the firmware.  There are 2 cases:
852 	 * 1. MAC address is valid.  It is assigned from the PF and we
853 	 *    need to override the current VF MAC address with it.
854 	 * 2. MAC address is zero.  The VF will use a random MAC address by
855 	 *    default but the stored zero MAC will allow the VF user to change
856 	 *    the random MAC address using ndo_set_mac_address() if he wants.
857 	 */
858 	if (!ether_addr_equal(resp->mac_address, bp->vf.mac_addr))
859 		memcpy(bp->vf.mac_addr, resp->mac_address, ETH_ALEN);
860 
861 	/* overwrite netdev dev_addr with admin VF MAC */
862 	if (is_valid_ether_addr(bp->vf.mac_addr))
863 		memcpy(bp->dev->dev_addr, bp->vf.mac_addr, ETH_ALEN);
864 update_vf_mac_exit:
865 	mutex_unlock(&bp->hwrm_cmd_lock);
866 }
867 
868 int bnxt_approve_mac(struct bnxt *bp, u8 *mac)
869 {
870 	struct hwrm_func_vf_cfg_input req = {0};
871 	int rc = 0;
872 
873 	if (!BNXT_VF(bp))
874 		return 0;
875 
876 	if (bp->hwrm_spec_code < 0x10202) {
877 		if (is_valid_ether_addr(bp->vf.mac_addr))
878 			rc = -EADDRNOTAVAIL;
879 		goto mac_done;
880 	}
881 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_VF_CFG, -1, -1);
882 	req.enables = cpu_to_le32(FUNC_VF_CFG_REQ_ENABLES_DFLT_MAC_ADDR);
883 	memcpy(req.dflt_mac_addr, mac, ETH_ALEN);
884 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
885 mac_done:
886 	if (rc) {
887 		rc = -EADDRNOTAVAIL;
888 		netdev_warn(bp->dev, "VF MAC address %pM not approved by the PF\n",
889 			    mac);
890 	}
891 	return rc;
892 }
893 #else
894 
895 void bnxt_sriov_disable(struct bnxt *bp)
896 {
897 }
898 
899 void bnxt_hwrm_exec_fwd_req(struct bnxt *bp)
900 {
901 	netdev_err(bp->dev, "Invalid VF message received when SRIOV is not enable\n");
902 }
903 
904 void bnxt_update_vf_mac(struct bnxt *bp)
905 {
906 }
907 
908 int bnxt_approve_mac(struct bnxt *bp, u8 *mac)
909 {
910 	return 0;
911 }
912 #endif
913