1 /* Broadcom NetXtreme-C/E network driver.
2  *
3  * Copyright (c) 2016-2017 Broadcom Limited
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 #include <linux/pci.h>
10 #include <linux/netdevice.h>
11 #include <linux/etherdevice.h>
12 #include <linux/rtnetlink.h>
13 #include <linux/jhash.h>
14 #include <net/pkt_cls.h>
15 
16 #include "bnxt_hsi.h"
17 #include "bnxt.h"
18 #include "bnxt_vfr.h"
19 #include "bnxt_devlink.h"
20 #include "bnxt_tc.h"
21 
22 #ifdef CONFIG_BNXT_SRIOV
23 
24 #define CFA_HANDLE_INVALID		0xffff
25 #define VF_IDX_INVALID			0xffff
26 
27 static int hwrm_cfa_vfr_alloc(struct bnxt *bp, u16 vf_idx,
28 			      u16 *tx_cfa_action, u16 *rx_cfa_code)
29 {
30 	struct hwrm_cfa_vfr_alloc_output *resp = bp->hwrm_cmd_resp_addr;
31 	struct hwrm_cfa_vfr_alloc_input req = { 0 };
32 	int rc;
33 
34 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_VFR_ALLOC, -1, -1);
35 	req.vf_id = cpu_to_le16(vf_idx);
36 	sprintf(req.vfr_name, "vfr%d", vf_idx);
37 
38 	mutex_lock(&bp->hwrm_cmd_lock);
39 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
40 	if (!rc) {
41 		*tx_cfa_action = le16_to_cpu(resp->tx_cfa_action);
42 		*rx_cfa_code = le16_to_cpu(resp->rx_cfa_code);
43 		netdev_dbg(bp->dev, "tx_cfa_action=0x%x, rx_cfa_code=0x%x",
44 			   *tx_cfa_action, *rx_cfa_code);
45 	} else {
46 		netdev_info(bp->dev, "%s error rc=%d", __func__, rc);
47 	}
48 
49 	mutex_unlock(&bp->hwrm_cmd_lock);
50 	return rc;
51 }
52 
53 static int hwrm_cfa_vfr_free(struct bnxt *bp, u16 vf_idx)
54 {
55 	struct hwrm_cfa_vfr_free_input req = { 0 };
56 	int rc;
57 
58 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_CFA_VFR_FREE, -1, -1);
59 	sprintf(req.vfr_name, "vfr%d", vf_idx);
60 
61 	rc = hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
62 	if (rc)
63 		netdev_info(bp->dev, "%s error rc=%d", __func__, rc);
64 	return rc;
65 }
66 
67 static int bnxt_hwrm_vfr_qcfg(struct bnxt *bp, struct bnxt_vf_rep *vf_rep,
68 			      u16 *max_mtu)
69 {
70 	struct hwrm_func_qcfg_output *resp = bp->hwrm_cmd_resp_addr;
71 	struct hwrm_func_qcfg_input req = {0};
72 	u16 mtu;
73 	int rc;
74 
75 	bnxt_hwrm_cmd_hdr_init(bp, &req, HWRM_FUNC_QCFG, -1, -1);
76 	req.fid = cpu_to_le16(bp->pf.vf[vf_rep->vf_idx].fw_fid);
77 
78 	mutex_lock(&bp->hwrm_cmd_lock);
79 
80 	rc = _hwrm_send_message(bp, &req, sizeof(req), HWRM_CMD_TIMEOUT);
81 	if (!rc) {
82 		mtu = le16_to_cpu(resp->max_mtu_configured);
83 		if (!mtu)
84 			*max_mtu = BNXT_MAX_MTU;
85 		else
86 			*max_mtu = mtu;
87 	}
88 	mutex_unlock(&bp->hwrm_cmd_lock);
89 	return rc;
90 }
91 
92 static int bnxt_vf_rep_open(struct net_device *dev)
93 {
94 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
95 	struct bnxt *bp = vf_rep->bp;
96 
97 	/* Enable link and TX only if the parent PF is open. */
98 	if (netif_running(bp->dev)) {
99 		netif_carrier_on(dev);
100 		netif_tx_start_all_queues(dev);
101 	}
102 	return 0;
103 }
104 
105 static int bnxt_vf_rep_close(struct net_device *dev)
106 {
107 	netif_carrier_off(dev);
108 	netif_tx_disable(dev);
109 
110 	return 0;
111 }
112 
113 static netdev_tx_t bnxt_vf_rep_xmit(struct sk_buff *skb,
114 				    struct net_device *dev)
115 {
116 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
117 	int rc, len = skb->len;
118 
119 	skb_dst_drop(skb);
120 	dst_hold((struct dst_entry *)vf_rep->dst);
121 	skb_dst_set(skb, (struct dst_entry *)vf_rep->dst);
122 	skb->dev = vf_rep->dst->u.port_info.lower_dev;
123 
124 	rc = dev_queue_xmit(skb);
125 	if (!rc) {
126 		vf_rep->tx_stats.packets++;
127 		vf_rep->tx_stats.bytes += len;
128 	}
129 	return rc;
130 }
131 
132 static void
133 bnxt_vf_rep_get_stats64(struct net_device *dev,
134 			struct rtnl_link_stats64 *stats)
135 {
136 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
137 
138 	stats->rx_packets = vf_rep->rx_stats.packets;
139 	stats->rx_bytes = vf_rep->rx_stats.bytes;
140 	stats->tx_packets = vf_rep->tx_stats.packets;
141 	stats->tx_bytes = vf_rep->tx_stats.bytes;
142 }
143 
144 static int bnxt_vf_rep_setup_tc_block_cb(enum tc_setup_type type,
145 					 void *type_data,
146 					 void *cb_priv)
147 {
148 	struct bnxt_vf_rep *vf_rep = cb_priv;
149 	struct bnxt *bp = vf_rep->bp;
150 	int vf_fid = bp->pf.vf[vf_rep->vf_idx].fw_fid;
151 
152 	if (!bnxt_tc_flower_enabled(vf_rep->bp) ||
153 	    !tc_cls_can_offload_and_chain0(bp->dev, type_data))
154 		return -EOPNOTSUPP;
155 
156 	switch (type) {
157 	case TC_SETUP_CLSFLOWER:
158 		return bnxt_tc_setup_flower(bp, vf_fid, type_data);
159 	default:
160 		return -EOPNOTSUPP;
161 	}
162 }
163 
164 static int bnxt_vf_rep_setup_tc(struct net_device *dev, enum tc_setup_type type,
165 				void *type_data)
166 {
167 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
168 
169 	switch (type) {
170 	case TC_SETUP_BLOCK:
171 		return flow_block_cb_setup_simple(type_data, NULL,
172 						  bnxt_vf_rep_setup_tc_block_cb,
173 						  vf_rep, vf_rep, true);
174 	default:
175 		return -EOPNOTSUPP;
176 	}
177 }
178 
179 struct net_device *bnxt_get_vf_rep(struct bnxt *bp, u16 cfa_code)
180 {
181 	u16 vf_idx;
182 
183 	if (cfa_code && bp->cfa_code_map && BNXT_PF(bp)) {
184 		vf_idx = bp->cfa_code_map[cfa_code];
185 		if (vf_idx != VF_IDX_INVALID)
186 			return bp->vf_reps[vf_idx]->dev;
187 	}
188 	return NULL;
189 }
190 
191 void bnxt_vf_rep_rx(struct bnxt *bp, struct sk_buff *skb)
192 {
193 	struct bnxt_vf_rep *vf_rep = netdev_priv(skb->dev);
194 
195 	vf_rep->rx_stats.bytes += skb->len;
196 	vf_rep->rx_stats.packets++;
197 
198 	netif_receive_skb(skb);
199 }
200 
201 static int bnxt_vf_rep_get_phys_port_name(struct net_device *dev, char *buf,
202 					  size_t len)
203 {
204 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
205 	struct pci_dev *pf_pdev = vf_rep->bp->pdev;
206 	int rc;
207 
208 	rc = snprintf(buf, len, "pf%dvf%d", PCI_FUNC(pf_pdev->devfn),
209 		      vf_rep->vf_idx);
210 	if (rc >= len)
211 		return -EOPNOTSUPP;
212 	return 0;
213 }
214 
215 static void bnxt_vf_rep_get_drvinfo(struct net_device *dev,
216 				    struct ethtool_drvinfo *info)
217 {
218 	strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
219 	strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
220 }
221 
222 static int bnxt_vf_rep_get_port_parent_id(struct net_device *dev,
223 					  struct netdev_phys_item_id *ppid)
224 {
225 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
226 
227 	/* as only PORT_PARENT_ID is supported currently use common code
228 	 * between PF and VF-rep for now.
229 	 */
230 	return bnxt_get_port_parent_id(vf_rep->bp->dev, ppid);
231 }
232 
233 static const struct ethtool_ops bnxt_vf_rep_ethtool_ops = {
234 	.get_drvinfo		= bnxt_vf_rep_get_drvinfo
235 };
236 
237 static const struct net_device_ops bnxt_vf_rep_netdev_ops = {
238 	.ndo_open		= bnxt_vf_rep_open,
239 	.ndo_stop		= bnxt_vf_rep_close,
240 	.ndo_start_xmit		= bnxt_vf_rep_xmit,
241 	.ndo_get_stats64	= bnxt_vf_rep_get_stats64,
242 	.ndo_setup_tc		= bnxt_vf_rep_setup_tc,
243 	.ndo_get_port_parent_id	= bnxt_vf_rep_get_port_parent_id,
244 	.ndo_get_phys_port_name = bnxt_vf_rep_get_phys_port_name
245 };
246 
247 bool bnxt_dev_is_vf_rep(struct net_device *dev)
248 {
249 	return dev->netdev_ops == &bnxt_vf_rep_netdev_ops;
250 }
251 
252 /* Called when the parent PF interface is closed:
253  * As the mode transition from SWITCHDEV to LEGACY
254  * happens under the rtnl_lock() this routine is safe
255  * under the rtnl_lock()
256  */
257 void bnxt_vf_reps_close(struct bnxt *bp)
258 {
259 	struct bnxt_vf_rep *vf_rep;
260 	u16 num_vfs, i;
261 
262 	if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV)
263 		return;
264 
265 	num_vfs = pci_num_vf(bp->pdev);
266 	for (i = 0; i < num_vfs; i++) {
267 		vf_rep = bp->vf_reps[i];
268 		if (netif_running(vf_rep->dev))
269 			bnxt_vf_rep_close(vf_rep->dev);
270 	}
271 }
272 
273 /* Called when the parent PF interface is opened (re-opened):
274  * As the mode transition from SWITCHDEV to LEGACY
275  * happen under the rtnl_lock() this routine is safe
276  * under the rtnl_lock()
277  */
278 void bnxt_vf_reps_open(struct bnxt *bp)
279 {
280 	int i;
281 
282 	if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV)
283 		return;
284 
285 	for (i = 0; i < pci_num_vf(bp->pdev); i++)
286 		bnxt_vf_rep_open(bp->vf_reps[i]->dev);
287 }
288 
289 static void __bnxt_vf_reps_destroy(struct bnxt *bp)
290 {
291 	u16 num_vfs = pci_num_vf(bp->pdev);
292 	struct bnxt_vf_rep *vf_rep;
293 	int i;
294 
295 	for (i = 0; i < num_vfs; i++) {
296 		vf_rep = bp->vf_reps[i];
297 		if (vf_rep) {
298 			dst_release((struct dst_entry *)vf_rep->dst);
299 
300 			if (vf_rep->tx_cfa_action != CFA_HANDLE_INVALID)
301 				hwrm_cfa_vfr_free(bp, vf_rep->vf_idx);
302 
303 			if (vf_rep->dev) {
304 				/* if register_netdev failed, then netdev_ops
305 				 * would have been set to NULL
306 				 */
307 				if (vf_rep->dev->netdev_ops)
308 					unregister_netdev(vf_rep->dev);
309 				free_netdev(vf_rep->dev);
310 			}
311 		}
312 	}
313 
314 	kfree(bp->vf_reps);
315 	bp->vf_reps = NULL;
316 }
317 
318 void bnxt_vf_reps_destroy(struct bnxt *bp)
319 {
320 	bool closed = false;
321 
322 	if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV)
323 		return;
324 
325 	if (!bp->vf_reps)
326 		return;
327 
328 	/* Ensure that parent PF's and VF-reps' RX/TX has been quiesced
329 	 * before proceeding with VF-rep cleanup.
330 	 */
331 	rtnl_lock();
332 	if (netif_running(bp->dev)) {
333 		bnxt_close_nic(bp, false, false);
334 		closed = true;
335 	}
336 	/* un-publish cfa_code_map so that RX path can't see it anymore */
337 	kfree(bp->cfa_code_map);
338 	bp->cfa_code_map = NULL;
339 	bp->eswitch_mode = DEVLINK_ESWITCH_MODE_LEGACY;
340 
341 	if (closed)
342 		bnxt_open_nic(bp, false, false);
343 	rtnl_unlock();
344 
345 	/* Need to call vf_reps_destroy() outside of rntl_lock
346 	 * as unregister_netdev takes rtnl_lock
347 	 */
348 	__bnxt_vf_reps_destroy(bp);
349 }
350 
351 /* Use the OUI of the PF's perm addr and report the same mac addr
352  * for the same VF-rep each time
353  */
354 static void bnxt_vf_rep_eth_addr_gen(u8 *src_mac, u16 vf_idx, u8 *mac)
355 {
356 	u32 addr;
357 
358 	ether_addr_copy(mac, src_mac);
359 
360 	addr = jhash(src_mac, ETH_ALEN, 0) + vf_idx;
361 	mac[3] = (u8)(addr & 0xFF);
362 	mac[4] = (u8)((addr >> 8) & 0xFF);
363 	mac[5] = (u8)((addr >> 16) & 0xFF);
364 }
365 
366 static void bnxt_vf_rep_netdev_init(struct bnxt *bp, struct bnxt_vf_rep *vf_rep,
367 				    struct net_device *dev)
368 {
369 	struct net_device *pf_dev = bp->dev;
370 	u16 max_mtu;
371 
372 	dev->netdev_ops = &bnxt_vf_rep_netdev_ops;
373 	dev->ethtool_ops = &bnxt_vf_rep_ethtool_ops;
374 	/* Just inherit all the featues of the parent PF as the VF-R
375 	 * uses the RX/TX rings of the parent PF
376 	 */
377 	dev->hw_features = pf_dev->hw_features;
378 	dev->gso_partial_features = pf_dev->gso_partial_features;
379 	dev->vlan_features = pf_dev->vlan_features;
380 	dev->hw_enc_features = pf_dev->hw_enc_features;
381 	dev->features |= pf_dev->features;
382 	bnxt_vf_rep_eth_addr_gen(bp->pf.mac_addr, vf_rep->vf_idx,
383 				 dev->perm_addr);
384 	ether_addr_copy(dev->dev_addr, dev->perm_addr);
385 	/* Set VF-Rep's max-mtu to the corresponding VF's max-mtu */
386 	if (!bnxt_hwrm_vfr_qcfg(bp, vf_rep, &max_mtu))
387 		dev->max_mtu = max_mtu;
388 	dev->min_mtu = ETH_ZLEN;
389 }
390 
391 static int bnxt_vf_reps_create(struct bnxt *bp)
392 {
393 	u16 *cfa_code_map = NULL, num_vfs = pci_num_vf(bp->pdev);
394 	struct bnxt_vf_rep *vf_rep;
395 	struct net_device *dev;
396 	int rc, i;
397 
398 	bp->vf_reps = kcalloc(num_vfs, sizeof(vf_rep), GFP_KERNEL);
399 	if (!bp->vf_reps)
400 		return -ENOMEM;
401 
402 	/* storage for cfa_code to vf-idx mapping */
403 	cfa_code_map = kmalloc_array(MAX_CFA_CODE, sizeof(*bp->cfa_code_map),
404 				     GFP_KERNEL);
405 	if (!cfa_code_map) {
406 		rc = -ENOMEM;
407 		goto err;
408 	}
409 	for (i = 0; i < MAX_CFA_CODE; i++)
410 		cfa_code_map[i] = VF_IDX_INVALID;
411 
412 	for (i = 0; i < num_vfs; i++) {
413 		dev = alloc_etherdev(sizeof(*vf_rep));
414 		if (!dev) {
415 			rc = -ENOMEM;
416 			goto err;
417 		}
418 
419 		vf_rep = netdev_priv(dev);
420 		bp->vf_reps[i] = vf_rep;
421 		vf_rep->dev = dev;
422 		vf_rep->bp = bp;
423 		vf_rep->vf_idx = i;
424 		vf_rep->tx_cfa_action = CFA_HANDLE_INVALID;
425 
426 		/* get cfa handles from FW */
427 		rc = hwrm_cfa_vfr_alloc(bp, vf_rep->vf_idx,
428 					&vf_rep->tx_cfa_action,
429 					&vf_rep->rx_cfa_code);
430 		if (rc) {
431 			rc = -ENOLINK;
432 			goto err;
433 		}
434 		cfa_code_map[vf_rep->rx_cfa_code] = vf_rep->vf_idx;
435 
436 		vf_rep->dst = metadata_dst_alloc(0, METADATA_HW_PORT_MUX,
437 						 GFP_KERNEL);
438 		if (!vf_rep->dst) {
439 			rc = -ENOMEM;
440 			goto err;
441 		}
442 		/* only cfa_action is needed to mux a packet while TXing */
443 		vf_rep->dst->u.port_info.port_id = vf_rep->tx_cfa_action;
444 		vf_rep->dst->u.port_info.lower_dev = bp->dev;
445 
446 		bnxt_vf_rep_netdev_init(bp, vf_rep, dev);
447 		rc = register_netdev(dev);
448 		if (rc) {
449 			/* no need for unregister_netdev in cleanup */
450 			dev->netdev_ops = NULL;
451 			goto err;
452 		}
453 	}
454 
455 	/* publish cfa_code_map only after all VF-reps have been initialized */
456 	bp->cfa_code_map = cfa_code_map;
457 	bp->eswitch_mode = DEVLINK_ESWITCH_MODE_SWITCHDEV;
458 	netif_keep_dst(bp->dev);
459 	return 0;
460 
461 err:
462 	netdev_info(bp->dev, "%s error=%d", __func__, rc);
463 	kfree(cfa_code_map);
464 	__bnxt_vf_reps_destroy(bp);
465 	return rc;
466 }
467 
468 /* Devlink related routines */
469 int bnxt_dl_eswitch_mode_get(struct devlink *devlink, u16 *mode)
470 {
471 	struct bnxt *bp = bnxt_get_bp_from_dl(devlink);
472 
473 	*mode = bp->eswitch_mode;
474 	return 0;
475 }
476 
477 int bnxt_dl_eswitch_mode_set(struct devlink *devlink, u16 mode,
478 			     struct netlink_ext_ack *extack)
479 {
480 	struct bnxt *bp = bnxt_get_bp_from_dl(devlink);
481 	int rc = 0;
482 
483 	mutex_lock(&bp->sriov_lock);
484 	if (bp->eswitch_mode == mode) {
485 		netdev_info(bp->dev, "already in %s eswitch mode",
486 			    mode == DEVLINK_ESWITCH_MODE_LEGACY ?
487 			    "legacy" : "switchdev");
488 		rc = -EINVAL;
489 		goto done;
490 	}
491 
492 	switch (mode) {
493 	case DEVLINK_ESWITCH_MODE_LEGACY:
494 		bnxt_vf_reps_destroy(bp);
495 		break;
496 
497 	case DEVLINK_ESWITCH_MODE_SWITCHDEV:
498 		if (bp->hwrm_spec_code < 0x10803) {
499 			netdev_warn(bp->dev, "FW does not support SRIOV E-Switch SWITCHDEV mode\n");
500 			rc = -ENOTSUPP;
501 			goto done;
502 		}
503 
504 		if (pci_num_vf(bp->pdev) == 0) {
505 			netdev_info(bp->dev, "Enable VFs before setting switchdev mode");
506 			rc = -EPERM;
507 			goto done;
508 		}
509 		rc = bnxt_vf_reps_create(bp);
510 		break;
511 
512 	default:
513 		rc = -EINVAL;
514 		goto done;
515 	}
516 done:
517 	mutex_unlock(&bp->sriov_lock);
518 	return rc;
519 }
520 
521 #endif
522