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_vf_rep_open(struct net_device *dev)
68 {
69 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
70 	struct bnxt *bp = vf_rep->bp;
71 
72 	/* Enable link and TX only if the parent PF is open. */
73 	if (netif_running(bp->dev)) {
74 		netif_carrier_on(dev);
75 		netif_tx_start_all_queues(dev);
76 	}
77 	return 0;
78 }
79 
80 static int bnxt_vf_rep_close(struct net_device *dev)
81 {
82 	netif_carrier_off(dev);
83 	netif_tx_disable(dev);
84 
85 	return 0;
86 }
87 
88 static netdev_tx_t bnxt_vf_rep_xmit(struct sk_buff *skb,
89 				    struct net_device *dev)
90 {
91 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
92 	int rc, len = skb->len;
93 
94 	skb_dst_drop(skb);
95 	dst_hold((struct dst_entry *)vf_rep->dst);
96 	skb_dst_set(skb, (struct dst_entry *)vf_rep->dst);
97 	skb->dev = vf_rep->dst->u.port_info.lower_dev;
98 
99 	rc = dev_queue_xmit(skb);
100 	if (!rc) {
101 		vf_rep->tx_stats.packets++;
102 		vf_rep->tx_stats.bytes += len;
103 	}
104 	return rc;
105 }
106 
107 static void
108 bnxt_vf_rep_get_stats64(struct net_device *dev,
109 			struct rtnl_link_stats64 *stats)
110 {
111 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
112 
113 	stats->rx_packets = vf_rep->rx_stats.packets;
114 	stats->rx_bytes = vf_rep->rx_stats.bytes;
115 	stats->tx_packets = vf_rep->tx_stats.packets;
116 	stats->tx_bytes = vf_rep->tx_stats.bytes;
117 }
118 
119 static int bnxt_vf_rep_setup_tc_block_cb(enum tc_setup_type type,
120 					 void *type_data,
121 					 void *cb_priv)
122 {
123 	struct bnxt_vf_rep *vf_rep = cb_priv;
124 	struct bnxt *bp = vf_rep->bp;
125 	int vf_fid = bp->pf.vf[vf_rep->vf_idx].fw_fid;
126 
127 	if (!bnxt_tc_flower_enabled(vf_rep->bp) ||
128 	    !tc_cls_can_offload_and_chain0(bp->dev, type_data))
129 		return -EOPNOTSUPP;
130 
131 	switch (type) {
132 	case TC_SETUP_CLSFLOWER:
133 		return bnxt_tc_setup_flower(bp, vf_fid, type_data);
134 	default:
135 		return -EOPNOTSUPP;
136 	}
137 }
138 
139 static int bnxt_vf_rep_setup_tc_block(struct net_device *dev,
140 				      struct tc_block_offload *f)
141 {
142 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
143 
144 	if (f->binder_type != TCF_BLOCK_BINDER_TYPE_CLSACT_INGRESS)
145 		return -EOPNOTSUPP;
146 
147 	switch (f->command) {
148 	case TC_BLOCK_BIND:
149 		return tcf_block_cb_register(f->block,
150 					     bnxt_vf_rep_setup_tc_block_cb,
151 					     vf_rep, vf_rep);
152 	case TC_BLOCK_UNBIND:
153 		tcf_block_cb_unregister(f->block,
154 					bnxt_vf_rep_setup_tc_block_cb, vf_rep);
155 		return 0;
156 	default:
157 		return -EOPNOTSUPP;
158 	}
159 }
160 
161 static int bnxt_vf_rep_setup_tc(struct net_device *dev, enum tc_setup_type type,
162 				void *type_data)
163 {
164 	switch (type) {
165 	case TC_SETUP_BLOCK:
166 		return bnxt_vf_rep_setup_tc_block(dev, type_data);
167 	default:
168 		return -EOPNOTSUPP;
169 	}
170 }
171 
172 struct net_device *bnxt_get_vf_rep(struct bnxt *bp, u16 cfa_code)
173 {
174 	u16 vf_idx;
175 
176 	if (cfa_code && bp->cfa_code_map && BNXT_PF(bp)) {
177 		vf_idx = bp->cfa_code_map[cfa_code];
178 		if (vf_idx != VF_IDX_INVALID)
179 			return bp->vf_reps[vf_idx]->dev;
180 	}
181 	return NULL;
182 }
183 
184 void bnxt_vf_rep_rx(struct bnxt *bp, struct sk_buff *skb)
185 {
186 	struct bnxt_vf_rep *vf_rep = netdev_priv(skb->dev);
187 	struct bnxt_vf_rep_stats *rx_stats;
188 
189 	rx_stats = &vf_rep->rx_stats;
190 	vf_rep->rx_stats.bytes += skb->len;
191 	vf_rep->rx_stats.packets++;
192 
193 	netif_receive_skb(skb);
194 }
195 
196 static int bnxt_vf_rep_get_phys_port_name(struct net_device *dev, char *buf,
197 					  size_t len)
198 {
199 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
200 	struct pci_dev *pf_pdev = vf_rep->bp->pdev;
201 	int rc;
202 
203 	rc = snprintf(buf, len, "pf%dvf%d", PCI_FUNC(pf_pdev->devfn),
204 		      vf_rep->vf_idx);
205 	if (rc >= len)
206 		return -EOPNOTSUPP;
207 	return 0;
208 }
209 
210 static void bnxt_vf_rep_get_drvinfo(struct net_device *dev,
211 				    struct ethtool_drvinfo *info)
212 {
213 	strlcpy(info->driver, DRV_MODULE_NAME, sizeof(info->driver));
214 	strlcpy(info->version, DRV_MODULE_VERSION, sizeof(info->version));
215 }
216 
217 static int bnxt_vf_rep_port_attr_get(struct net_device *dev,
218 				     struct switchdev_attr *attr)
219 {
220 	struct bnxt_vf_rep *vf_rep = netdev_priv(dev);
221 
222 	/* as only PORT_PARENT_ID is supported currently use common code
223 	 * between PF and VF-rep for now.
224 	 */
225 	return bnxt_port_attr_get(vf_rep->bp, attr);
226 }
227 
228 static const struct switchdev_ops bnxt_vf_rep_switchdev_ops = {
229 	.switchdev_port_attr_get	= bnxt_vf_rep_port_attr_get
230 };
231 
232 static const struct ethtool_ops bnxt_vf_rep_ethtool_ops = {
233 	.get_drvinfo		= bnxt_vf_rep_get_drvinfo
234 };
235 
236 static const struct net_device_ops bnxt_vf_rep_netdev_ops = {
237 	.ndo_open		= bnxt_vf_rep_open,
238 	.ndo_stop		= bnxt_vf_rep_close,
239 	.ndo_start_xmit		= bnxt_vf_rep_xmit,
240 	.ndo_get_stats64	= bnxt_vf_rep_get_stats64,
241 	.ndo_setup_tc		= bnxt_vf_rep_setup_tc,
242 	.ndo_get_phys_port_name = bnxt_vf_rep_get_phys_port_name
243 };
244 
245 bool bnxt_dev_is_vf_rep(struct net_device *dev)
246 {
247 	return dev->netdev_ops == &bnxt_vf_rep_netdev_ops;
248 }
249 
250 /* Called when the parent PF interface is closed:
251  * As the mode transition from SWITCHDEV to LEGACY
252  * happens under the rtnl_lock() this routine is safe
253  * under the rtnl_lock()
254  */
255 void bnxt_vf_reps_close(struct bnxt *bp)
256 {
257 	struct bnxt_vf_rep *vf_rep;
258 	u16 num_vfs, i;
259 
260 	if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV)
261 		return;
262 
263 	num_vfs = pci_num_vf(bp->pdev);
264 	for (i = 0; i < num_vfs; i++) {
265 		vf_rep = bp->vf_reps[i];
266 		if (netif_running(vf_rep->dev))
267 			bnxt_vf_rep_close(vf_rep->dev);
268 	}
269 }
270 
271 /* Called when the parent PF interface is opened (re-opened):
272  * As the mode transition from SWITCHDEV to LEGACY
273  * happen under the rtnl_lock() this routine is safe
274  * under the rtnl_lock()
275  */
276 void bnxt_vf_reps_open(struct bnxt *bp)
277 {
278 	int i;
279 
280 	if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV)
281 		return;
282 
283 	for (i = 0; i < pci_num_vf(bp->pdev); i++)
284 		bnxt_vf_rep_open(bp->vf_reps[i]->dev);
285 }
286 
287 static void __bnxt_vf_reps_destroy(struct bnxt *bp)
288 {
289 	u16 num_vfs = pci_num_vf(bp->pdev);
290 	struct bnxt_vf_rep *vf_rep;
291 	int i;
292 
293 	for (i = 0; i < num_vfs; i++) {
294 		vf_rep = bp->vf_reps[i];
295 		if (vf_rep) {
296 			dst_release((struct dst_entry *)vf_rep->dst);
297 
298 			if (vf_rep->tx_cfa_action != CFA_HANDLE_INVALID)
299 				hwrm_cfa_vfr_free(bp, vf_rep->vf_idx);
300 
301 			if (vf_rep->dev) {
302 				/* if register_netdev failed, then netdev_ops
303 				 * would have been set to NULL
304 				 */
305 				if (vf_rep->dev->netdev_ops)
306 					unregister_netdev(vf_rep->dev);
307 				free_netdev(vf_rep->dev);
308 			}
309 		}
310 	}
311 
312 	kfree(bp->vf_reps);
313 	bp->vf_reps = NULL;
314 }
315 
316 void bnxt_vf_reps_destroy(struct bnxt *bp)
317 {
318 	bool closed = false;
319 
320 	if (bp->eswitch_mode != DEVLINK_ESWITCH_MODE_SWITCHDEV)
321 		return;
322 
323 	if (!bp->vf_reps)
324 		return;
325 
326 	/* Ensure that parent PF's and VF-reps' RX/TX has been quiesced
327 	 * before proceeding with VF-rep cleanup.
328 	 */
329 	rtnl_lock();
330 	if (netif_running(bp->dev)) {
331 		bnxt_close_nic(bp, false, false);
332 		closed = true;
333 	}
334 	/* un-publish cfa_code_map so that RX path can't see it anymore */
335 	kfree(bp->cfa_code_map);
336 	bp->cfa_code_map = NULL;
337 	bp->eswitch_mode = DEVLINK_ESWITCH_MODE_LEGACY;
338 
339 	if (closed)
340 		bnxt_open_nic(bp, false, false);
341 	rtnl_unlock();
342 
343 	/* Need to call vf_reps_destroy() outside of rntl_lock
344 	 * as unregister_netdev takes rtnl_lock
345 	 */
346 	__bnxt_vf_reps_destroy(bp);
347 }
348 
349 /* Use the OUI of the PF's perm addr and report the same mac addr
350  * for the same VF-rep each time
351  */
352 static void bnxt_vf_rep_eth_addr_gen(u8 *src_mac, u16 vf_idx, u8 *mac)
353 {
354 	u32 addr;
355 
356 	ether_addr_copy(mac, src_mac);
357 
358 	addr = jhash(src_mac, ETH_ALEN, 0) + vf_idx;
359 	mac[3] = (u8)(addr & 0xFF);
360 	mac[4] = (u8)((addr >> 8) & 0xFF);
361 	mac[5] = (u8)((addr >> 16) & 0xFF);
362 }
363 
364 static void bnxt_vf_rep_netdev_init(struct bnxt *bp, struct bnxt_vf_rep *vf_rep,
365 				    struct net_device *dev)
366 {
367 	struct net_device *pf_dev = bp->dev;
368 
369 	dev->netdev_ops = &bnxt_vf_rep_netdev_ops;
370 	dev->ethtool_ops = &bnxt_vf_rep_ethtool_ops;
371 	SWITCHDEV_SET_OPS(dev, &bnxt_vf_rep_switchdev_ops);
372 	/* Just inherit all the featues of the parent PF as the VF-R
373 	 * uses the RX/TX rings of the parent PF
374 	 */
375 	dev->hw_features = pf_dev->hw_features;
376 	dev->gso_partial_features = pf_dev->gso_partial_features;
377 	dev->vlan_features = pf_dev->vlan_features;
378 	dev->hw_enc_features = pf_dev->hw_enc_features;
379 	dev->features |= pf_dev->features;
380 	bnxt_vf_rep_eth_addr_gen(bp->pf.mac_addr, vf_rep->vf_idx,
381 				 dev->perm_addr);
382 	ether_addr_copy(dev->dev_addr, dev->perm_addr);
383 }
384 
385 static int bnxt_pcie_dsn_get(struct bnxt *bp, u8 dsn[])
386 {
387 	struct pci_dev *pdev = bp->pdev;
388 	int pos = pci_find_ext_capability(pdev, PCI_EXT_CAP_ID_DSN);
389 	u32 dw;
390 
391 	if (!pos) {
392 		netdev_info(bp->dev, "Unable do read adapter's DSN");
393 		return -EOPNOTSUPP;
394 	}
395 
396 	/* DSN (two dw) is at an offset of 4 from the cap pos */
397 	pos += 4;
398 	pci_read_config_dword(pdev, pos, &dw);
399 	put_unaligned_le32(dw, &dsn[0]);
400 	pci_read_config_dword(pdev, pos + 4, &dw);
401 	put_unaligned_le32(dw, &dsn[4]);
402 	return 0;
403 }
404 
405 static int bnxt_vf_reps_create(struct bnxt *bp)
406 {
407 	u16 *cfa_code_map = NULL, num_vfs = pci_num_vf(bp->pdev);
408 	struct bnxt_vf_rep *vf_rep;
409 	struct net_device *dev;
410 	int rc, i;
411 
412 	bp->vf_reps = kcalloc(num_vfs, sizeof(vf_rep), GFP_KERNEL);
413 	if (!bp->vf_reps)
414 		return -ENOMEM;
415 
416 	/* storage for cfa_code to vf-idx mapping */
417 	cfa_code_map = kmalloc(sizeof(*bp->cfa_code_map) * MAX_CFA_CODE,
418 			       GFP_KERNEL);
419 	if (!cfa_code_map) {
420 		rc = -ENOMEM;
421 		goto err;
422 	}
423 	for (i = 0; i < MAX_CFA_CODE; i++)
424 		cfa_code_map[i] = VF_IDX_INVALID;
425 
426 	for (i = 0; i < num_vfs; i++) {
427 		dev = alloc_etherdev(sizeof(*vf_rep));
428 		if (!dev) {
429 			rc = -ENOMEM;
430 			goto err;
431 		}
432 
433 		vf_rep = netdev_priv(dev);
434 		bp->vf_reps[i] = vf_rep;
435 		vf_rep->dev = dev;
436 		vf_rep->bp = bp;
437 		vf_rep->vf_idx = i;
438 		vf_rep->tx_cfa_action = CFA_HANDLE_INVALID;
439 
440 		/* get cfa handles from FW */
441 		rc = hwrm_cfa_vfr_alloc(bp, vf_rep->vf_idx,
442 					&vf_rep->tx_cfa_action,
443 					&vf_rep->rx_cfa_code);
444 		if (rc) {
445 			rc = -ENOLINK;
446 			goto err;
447 		}
448 		cfa_code_map[vf_rep->rx_cfa_code] = vf_rep->vf_idx;
449 
450 		vf_rep->dst = metadata_dst_alloc(0, METADATA_HW_PORT_MUX,
451 						 GFP_KERNEL);
452 		if (!vf_rep->dst) {
453 			rc = -ENOMEM;
454 			goto err;
455 		}
456 		/* only cfa_action is needed to mux a packet while TXing */
457 		vf_rep->dst->u.port_info.port_id = vf_rep->tx_cfa_action;
458 		vf_rep->dst->u.port_info.lower_dev = bp->dev;
459 
460 		bnxt_vf_rep_netdev_init(bp, vf_rep, dev);
461 		rc = register_netdev(dev);
462 		if (rc) {
463 			/* no need for unregister_netdev in cleanup */
464 			dev->netdev_ops = NULL;
465 			goto err;
466 		}
467 	}
468 
469 	/* Read the adapter's DSN to use as the eswitch switch_id */
470 	rc = bnxt_pcie_dsn_get(bp, bp->switch_id);
471 	if (rc)
472 		goto err;
473 
474 	/* publish cfa_code_map only after all VF-reps have been initialized */
475 	bp->cfa_code_map = cfa_code_map;
476 	bp->eswitch_mode = DEVLINK_ESWITCH_MODE_SWITCHDEV;
477 	netif_keep_dst(bp->dev);
478 	return 0;
479 
480 err:
481 	netdev_info(bp->dev, "%s error=%d", __func__, rc);
482 	kfree(cfa_code_map);
483 	__bnxt_vf_reps_destroy(bp);
484 	return rc;
485 }
486 
487 /* Devlink related routines */
488 int bnxt_dl_eswitch_mode_get(struct devlink *devlink, u16 *mode)
489 {
490 	struct bnxt *bp = bnxt_get_bp_from_dl(devlink);
491 
492 	*mode = bp->eswitch_mode;
493 	return 0;
494 }
495 
496 int bnxt_dl_eswitch_mode_set(struct devlink *devlink, u16 mode)
497 {
498 	struct bnxt *bp = bnxt_get_bp_from_dl(devlink);
499 	int rc = 0;
500 
501 	mutex_lock(&bp->sriov_lock);
502 	if (bp->eswitch_mode == mode) {
503 		netdev_info(bp->dev, "already in %s eswitch mode",
504 			    mode == DEVLINK_ESWITCH_MODE_LEGACY ?
505 			    "legacy" : "switchdev");
506 		rc = -EINVAL;
507 		goto done;
508 	}
509 
510 	switch (mode) {
511 	case DEVLINK_ESWITCH_MODE_LEGACY:
512 		bnxt_vf_reps_destroy(bp);
513 		break;
514 
515 	case DEVLINK_ESWITCH_MODE_SWITCHDEV:
516 		if (pci_num_vf(bp->pdev) == 0) {
517 			netdev_info(bp->dev,
518 				    "Enable VFs before setting switchdev mode");
519 			rc = -EPERM;
520 			goto done;
521 		}
522 		rc = bnxt_vf_reps_create(bp);
523 		break;
524 
525 	default:
526 		rc = -EINVAL;
527 		goto done;
528 	}
529 done:
530 	mutex_unlock(&bp->sriov_lock);
531 	return rc;
532 }
533 
534 #endif
535