1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2015-2019 Netronome Systems, Inc. */
3 
4 /*
5  * nfp_net_common.c
6  * Netronome network device driver: Common functions between PF and VF
7  * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
8  *          Jason McMullan <jason.mcmullan@netronome.com>
9  *          Rolf Neugebauer <rolf.neugebauer@netronome.com>
10  *          Brad Petrus <brad.petrus@netronome.com>
11  *          Chris Telfer <chris.telfer@netronome.com>
12  */
13 
14 #include <linux/bitfield.h>
15 #include <linux/bpf.h>
16 #include <linux/module.h>
17 #include <linux/kernel.h>
18 #include <linux/init.h>
19 #include <linux/fs.h>
20 #include <linux/netdevice.h>
21 #include <linux/etherdevice.h>
22 #include <linux/interrupt.h>
23 #include <linux/ip.h>
24 #include <linux/ipv6.h>
25 #include <linux/mm.h>
26 #include <linux/overflow.h>
27 #include <linux/page_ref.h>
28 #include <linux/pci.h>
29 #include <linux/pci_regs.h>
30 #include <linux/ethtool.h>
31 #include <linux/log2.h>
32 #include <linux/if_vlan.h>
33 #include <linux/if_bridge.h>
34 #include <linux/random.h>
35 #include <linux/vmalloc.h>
36 #include <linux/ktime.h>
37 
38 #include <net/tls.h>
39 #include <net/vxlan.h>
40 #include <net/xdp_sock_drv.h>
41 #include <net/xfrm.h>
42 
43 #include "nfpcore/nfp_dev.h"
44 #include "nfpcore/nfp_nsp.h"
45 #include "ccm.h"
46 #include "nfp_app.h"
47 #include "nfp_net_ctrl.h"
48 #include "nfp_net.h"
49 #include "nfp_net_dp.h"
50 #include "nfp_net_sriov.h"
51 #include "nfp_net_xsk.h"
52 #include "nfp_port.h"
53 #include "crypto/crypto.h"
54 #include "crypto/fw.h"
55 
56 /**
57  * nfp_net_get_fw_version() - Read and parse the FW version
58  * @fw_ver:	Output fw_version structure to read to
59  * @ctrl_bar:	Mapped address of the control BAR
60  */
61 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
62 			    void __iomem *ctrl_bar)
63 {
64 	u32 reg;
65 
66 	reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
67 	put_unaligned_le32(reg, fw_ver);
68 }
69 
70 u32 nfp_qcp_queue_offset(const struct nfp_dev_info *dev_info, u16 queue)
71 {
72 	queue &= dev_info->qc_idx_mask;
73 	return dev_info->qc_addr_offset + NFP_QCP_QUEUE_ADDR_SZ * queue;
74 }
75 
76 /* Firmware reconfig
77  *
78  * Firmware reconfig may take a while so we have two versions of it -
79  * synchronous and asynchronous (posted).  All synchronous callers are holding
80  * RTNL so we don't have to worry about serializing them.
81  */
82 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
83 {
84 	nn_writel(nn, NFP_NET_CFG_UPDATE, update);
85 	/* ensure update is written before pinging HW */
86 	nn_pci_flush(nn);
87 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
88 	nn->reconfig_in_progress_update = update;
89 }
90 
91 /* Pass 0 as update to run posted reconfigs. */
92 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
93 {
94 	update |= nn->reconfig_posted;
95 	nn->reconfig_posted = 0;
96 
97 	nfp_net_reconfig_start(nn, update);
98 
99 	nn->reconfig_timer_active = true;
100 	mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
101 }
102 
103 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
104 {
105 	u32 reg;
106 
107 	reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
108 	if (reg == 0)
109 		return true;
110 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
111 		nn_err(nn, "Reconfig error (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
112 		       reg, nn->reconfig_in_progress_update,
113 		       nn_readl(nn, NFP_NET_CFG_CTRL));
114 		return true;
115 	} else if (last_check) {
116 		nn_err(nn, "Reconfig timeout (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
117 		       reg, nn->reconfig_in_progress_update,
118 		       nn_readl(nn, NFP_NET_CFG_CTRL));
119 		return true;
120 	}
121 
122 	return false;
123 }
124 
125 static bool __nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
126 {
127 	bool timed_out = false;
128 	int i;
129 
130 	/* Poll update field, waiting for NFP to ack the config.
131 	 * Do an opportunistic wait-busy loop, afterward sleep.
132 	 */
133 	for (i = 0; i < 50; i++) {
134 		if (nfp_net_reconfig_check_done(nn, false))
135 			return false;
136 		udelay(4);
137 	}
138 
139 	while (!nfp_net_reconfig_check_done(nn, timed_out)) {
140 		usleep_range(250, 500);
141 		timed_out = time_is_before_eq_jiffies(deadline);
142 	}
143 
144 	return timed_out;
145 }
146 
147 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
148 {
149 	if (__nfp_net_reconfig_wait(nn, deadline))
150 		return -EIO;
151 
152 	if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
153 		return -EIO;
154 
155 	return 0;
156 }
157 
158 static void nfp_net_reconfig_timer(struct timer_list *t)
159 {
160 	struct nfp_net *nn = from_timer(nn, t, reconfig_timer);
161 
162 	spin_lock_bh(&nn->reconfig_lock);
163 
164 	nn->reconfig_timer_active = false;
165 
166 	/* If sync caller is present it will take over from us */
167 	if (nn->reconfig_sync_present)
168 		goto done;
169 
170 	/* Read reconfig status and report errors */
171 	nfp_net_reconfig_check_done(nn, true);
172 
173 	if (nn->reconfig_posted)
174 		nfp_net_reconfig_start_async(nn, 0);
175 done:
176 	spin_unlock_bh(&nn->reconfig_lock);
177 }
178 
179 /**
180  * nfp_net_reconfig_post() - Post async reconfig request
181  * @nn:      NFP Net device to reconfigure
182  * @update:  The value for the update field in the BAR config
183  *
184  * Record FW reconfiguration request.  Reconfiguration will be kicked off
185  * whenever reconfiguration machinery is idle.  Multiple requests can be
186  * merged together!
187  */
188 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
189 {
190 	spin_lock_bh(&nn->reconfig_lock);
191 
192 	/* Sync caller will kick off async reconf when it's done, just post */
193 	if (nn->reconfig_sync_present) {
194 		nn->reconfig_posted |= update;
195 		goto done;
196 	}
197 
198 	/* Opportunistically check if the previous command is done */
199 	if (!nn->reconfig_timer_active ||
200 	    nfp_net_reconfig_check_done(nn, false))
201 		nfp_net_reconfig_start_async(nn, update);
202 	else
203 		nn->reconfig_posted |= update;
204 done:
205 	spin_unlock_bh(&nn->reconfig_lock);
206 }
207 
208 static void nfp_net_reconfig_sync_enter(struct nfp_net *nn)
209 {
210 	bool cancelled_timer = false;
211 	u32 pre_posted_requests;
212 
213 	spin_lock_bh(&nn->reconfig_lock);
214 
215 	WARN_ON(nn->reconfig_sync_present);
216 	nn->reconfig_sync_present = true;
217 
218 	if (nn->reconfig_timer_active) {
219 		nn->reconfig_timer_active = false;
220 		cancelled_timer = true;
221 	}
222 	pre_posted_requests = nn->reconfig_posted;
223 	nn->reconfig_posted = 0;
224 
225 	spin_unlock_bh(&nn->reconfig_lock);
226 
227 	if (cancelled_timer) {
228 		del_timer_sync(&nn->reconfig_timer);
229 		nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
230 	}
231 
232 	/* Run the posted reconfigs which were issued before we started */
233 	if (pre_posted_requests) {
234 		nfp_net_reconfig_start(nn, pre_posted_requests);
235 		nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
236 	}
237 }
238 
239 static void nfp_net_reconfig_wait_posted(struct nfp_net *nn)
240 {
241 	nfp_net_reconfig_sync_enter(nn);
242 
243 	spin_lock_bh(&nn->reconfig_lock);
244 	nn->reconfig_sync_present = false;
245 	spin_unlock_bh(&nn->reconfig_lock);
246 }
247 
248 /**
249  * __nfp_net_reconfig() - Reconfigure the firmware
250  * @nn:      NFP Net device to reconfigure
251  * @update:  The value for the update field in the BAR config
252  *
253  * Write the update word to the BAR and ping the reconfig queue.  The
254  * poll until the firmware has acknowledged the update by zeroing the
255  * update word.
256  *
257  * Return: Negative errno on error, 0 on success
258  */
259 int __nfp_net_reconfig(struct nfp_net *nn, u32 update)
260 {
261 	int ret;
262 
263 	nfp_net_reconfig_sync_enter(nn);
264 
265 	nfp_net_reconfig_start(nn, update);
266 	ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
267 
268 	spin_lock_bh(&nn->reconfig_lock);
269 
270 	if (nn->reconfig_posted)
271 		nfp_net_reconfig_start_async(nn, 0);
272 
273 	nn->reconfig_sync_present = false;
274 
275 	spin_unlock_bh(&nn->reconfig_lock);
276 
277 	return ret;
278 }
279 
280 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
281 {
282 	int ret;
283 
284 	nn_ctrl_bar_lock(nn);
285 	ret = __nfp_net_reconfig(nn, update);
286 	nn_ctrl_bar_unlock(nn);
287 
288 	return ret;
289 }
290 
291 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size)
292 {
293 	if (nn->tlv_caps.mbox_len < NFP_NET_CFG_MBOX_SIMPLE_VAL + data_size) {
294 		nn_err(nn, "mailbox too small for %u of data (%u)\n",
295 		       data_size, nn->tlv_caps.mbox_len);
296 		return -EIO;
297 	}
298 
299 	nn_ctrl_bar_lock(nn);
300 	return 0;
301 }
302 
303 /**
304  * nfp_net_mbox_reconfig() - Reconfigure the firmware via the mailbox
305  * @nn:        NFP Net device to reconfigure
306  * @mbox_cmd:  The value for the mailbox command
307  *
308  * Helper function for mailbox updates
309  *
310  * Return: Negative errno on error, 0 on success
311  */
312 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd)
313 {
314 	u32 mbox = nn->tlv_caps.mbox_off;
315 	int ret;
316 
317 	nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
318 
319 	ret = __nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
320 	if (ret) {
321 		nn_err(nn, "Mailbox update error\n");
322 		return ret;
323 	}
324 
325 	return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
326 }
327 
328 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 mbox_cmd)
329 {
330 	u32 mbox = nn->tlv_caps.mbox_off;
331 
332 	nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
333 
334 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_MBOX);
335 }
336 
337 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn)
338 {
339 	u32 mbox = nn->tlv_caps.mbox_off;
340 
341 	nfp_net_reconfig_wait_posted(nn);
342 
343 	return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
344 }
345 
346 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd)
347 {
348 	int ret;
349 
350 	ret = nfp_net_mbox_reconfig(nn, mbox_cmd);
351 	nn_ctrl_bar_unlock(nn);
352 	return ret;
353 }
354 
355 /* Interrupt configuration and handling
356  */
357 
358 /**
359  * nfp_net_irqs_alloc() - allocates MSI-X irqs
360  * @pdev:        PCI device structure
361  * @irq_entries: Array to be initialized and used to hold the irq entries
362  * @min_irqs:    Minimal acceptable number of interrupts
363  * @wanted_irqs: Target number of interrupts to allocate
364  *
365  * Return: Number of irqs obtained or 0 on error.
366  */
367 unsigned int
368 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
369 		   unsigned int min_irqs, unsigned int wanted_irqs)
370 {
371 	unsigned int i;
372 	int got_irqs;
373 
374 	for (i = 0; i < wanted_irqs; i++)
375 		irq_entries[i].entry = i;
376 
377 	got_irqs = pci_enable_msix_range(pdev, irq_entries,
378 					 min_irqs, wanted_irqs);
379 	if (got_irqs < 0) {
380 		dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
381 			min_irqs, wanted_irqs, got_irqs);
382 		return 0;
383 	}
384 
385 	if (got_irqs < wanted_irqs)
386 		dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
387 			 wanted_irqs, got_irqs);
388 
389 	return got_irqs;
390 }
391 
392 /**
393  * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
394  * @nn:		 NFP Network structure
395  * @irq_entries: Table of allocated interrupts
396  * @n:		 Size of @irq_entries (number of entries to grab)
397  *
398  * After interrupts are allocated with nfp_net_irqs_alloc() this function
399  * should be called to assign them to a specific netdev (port).
400  */
401 void
402 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
403 		    unsigned int n)
404 {
405 	struct nfp_net_dp *dp = &nn->dp;
406 
407 	nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
408 	dp->num_r_vecs = nn->max_r_vecs;
409 
410 	memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
411 
412 	if (dp->num_rx_rings > dp->num_r_vecs ||
413 	    dp->num_tx_rings > dp->num_r_vecs)
414 		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
415 			 dp->num_rx_rings, dp->num_tx_rings,
416 			 dp->num_r_vecs);
417 
418 	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
419 	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
420 	dp->num_stack_tx_rings = dp->num_tx_rings;
421 }
422 
423 /**
424  * nfp_net_irqs_disable() - Disable interrupts
425  * @pdev:        PCI device structure
426  *
427  * Undoes what @nfp_net_irqs_alloc() does.
428  */
429 void nfp_net_irqs_disable(struct pci_dev *pdev)
430 {
431 	pci_disable_msix(pdev);
432 }
433 
434 /**
435  * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
436  * @irq:      Interrupt
437  * @data:     Opaque data structure
438  *
439  * Return: Indicate if the interrupt has been handled.
440  */
441 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
442 {
443 	struct nfp_net_r_vector *r_vec = data;
444 
445 	/* Currently we cannot tell if it's a rx or tx interrupt,
446 	 * since dim does not need accurate event_ctr to calculate,
447 	 * we just use this counter for both rx and tx dim.
448 	 */
449 	r_vec->event_ctr++;
450 
451 	napi_schedule_irqoff(&r_vec->napi);
452 
453 	/* The FW auto-masks any interrupt, either via the MASK bit in
454 	 * the MSI-X table or via the per entry ICR field.  So there
455 	 * is no need to disable interrupts here.
456 	 */
457 	return IRQ_HANDLED;
458 }
459 
460 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
461 {
462 	struct nfp_net_r_vector *r_vec = data;
463 
464 	tasklet_schedule(&r_vec->tasklet);
465 
466 	return IRQ_HANDLED;
467 }
468 
469 /**
470  * nfp_net_read_link_status() - Reread link status from control BAR
471  * @nn:       NFP Network structure
472  */
473 static void nfp_net_read_link_status(struct nfp_net *nn)
474 {
475 	unsigned long flags;
476 	bool link_up;
477 	u16 sts;
478 
479 	spin_lock_irqsave(&nn->link_status_lock, flags);
480 
481 	sts = nn_readw(nn, NFP_NET_CFG_STS);
482 	link_up = !!(sts & NFP_NET_CFG_STS_LINK);
483 
484 	if (nn->link_up == link_up)
485 		goto out;
486 
487 	nn->link_up = link_up;
488 	if (nn->port) {
489 		set_bit(NFP_PORT_CHANGED, &nn->port->flags);
490 		if (nn->port->link_cb)
491 			nn->port->link_cb(nn->port);
492 	}
493 
494 	if (nn->link_up) {
495 		netif_carrier_on(nn->dp.netdev);
496 		netdev_info(nn->dp.netdev, "NIC Link is Up\n");
497 	} else {
498 		netif_carrier_off(nn->dp.netdev);
499 		netdev_info(nn->dp.netdev, "NIC Link is Down\n");
500 	}
501 out:
502 	spin_unlock_irqrestore(&nn->link_status_lock, flags);
503 }
504 
505 /**
506  * nfp_net_irq_lsc() - Interrupt service routine for link state changes
507  * @irq:      Interrupt
508  * @data:     Opaque data structure
509  *
510  * Return: Indicate if the interrupt has been handled.
511  */
512 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
513 {
514 	struct nfp_net *nn = data;
515 	struct msix_entry *entry;
516 
517 	entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
518 
519 	nfp_net_read_link_status(nn);
520 
521 	nfp_net_irq_unmask(nn, entry->entry);
522 
523 	return IRQ_HANDLED;
524 }
525 
526 /**
527  * nfp_net_irq_exn() - Interrupt service routine for exceptions
528  * @irq:      Interrupt
529  * @data:     Opaque data structure
530  *
531  * Return: Indicate if the interrupt has been handled.
532  */
533 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
534 {
535 	struct nfp_net *nn = data;
536 
537 	nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
538 	/* XXX TO BE IMPLEMENTED */
539 	return IRQ_HANDLED;
540 }
541 
542 /**
543  * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
544  * @nn:		NFP Network structure
545  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
546  * @format:	printf-style format to construct the interrupt name
547  * @name:	Pointer to allocated space for interrupt name
548  * @name_sz:	Size of space for interrupt name
549  * @vector_idx:	Index of MSI-X vector used for this interrupt
550  * @handler:	IRQ handler to register for this interrupt
551  */
552 static int
553 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
554 			const char *format, char *name, size_t name_sz,
555 			unsigned int vector_idx, irq_handler_t handler)
556 {
557 	struct msix_entry *entry;
558 	int err;
559 
560 	entry = &nn->irq_entries[vector_idx];
561 
562 	snprintf(name, name_sz, format, nfp_net_name(nn));
563 	err = request_irq(entry->vector, handler, 0, name, nn);
564 	if (err) {
565 		nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
566 		       entry->vector, err);
567 		return err;
568 	}
569 	nn_writeb(nn, ctrl_offset, entry->entry);
570 	nfp_net_irq_unmask(nn, entry->entry);
571 
572 	return 0;
573 }
574 
575 /**
576  * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
577  * @nn:		NFP Network structure
578  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
579  * @vector_idx:	Index of MSI-X vector used for this interrupt
580  */
581 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
582 				 unsigned int vector_idx)
583 {
584 	nn_writeb(nn, ctrl_offset, 0xff);
585 	nn_pci_flush(nn);
586 	free_irq(nn->irq_entries[vector_idx].vector, nn);
587 }
588 
589 struct sk_buff *
590 nfp_net_tls_tx(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
591 	       struct sk_buff *skb, u64 *tls_handle, int *nr_frags)
592 {
593 #ifdef CONFIG_TLS_DEVICE
594 	struct nfp_net_tls_offload_ctx *ntls;
595 	struct sk_buff *nskb;
596 	bool resync_pending;
597 	u32 datalen, seq;
598 
599 	if (likely(!dp->ktls_tx))
600 		return skb;
601 	if (!tls_is_skb_tx_device_offloaded(skb))
602 		return skb;
603 
604 	datalen = skb->len - skb_tcp_all_headers(skb);
605 	seq = ntohl(tcp_hdr(skb)->seq);
606 	ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX);
607 	resync_pending = tls_offload_tx_resync_pending(skb->sk);
608 	if (unlikely(resync_pending || ntls->next_seq != seq)) {
609 		/* Pure ACK out of order already */
610 		if (!datalen)
611 			return skb;
612 
613 		u64_stats_update_begin(&r_vec->tx_sync);
614 		r_vec->tls_tx_fallback++;
615 		u64_stats_update_end(&r_vec->tx_sync);
616 
617 		nskb = tls_encrypt_skb(skb);
618 		if (!nskb) {
619 			u64_stats_update_begin(&r_vec->tx_sync);
620 			r_vec->tls_tx_no_fallback++;
621 			u64_stats_update_end(&r_vec->tx_sync);
622 			return NULL;
623 		}
624 		/* encryption wasn't necessary */
625 		if (nskb == skb)
626 			return skb;
627 		/* we don't re-check ring space */
628 		if (unlikely(skb_is_nonlinear(nskb))) {
629 			nn_dp_warn(dp, "tls_encrypt_skb() produced fragmented frame\n");
630 			u64_stats_update_begin(&r_vec->tx_sync);
631 			r_vec->tx_errors++;
632 			u64_stats_update_end(&r_vec->tx_sync);
633 			dev_kfree_skb_any(nskb);
634 			return NULL;
635 		}
636 
637 		/* jump forward, a TX may have gotten lost, need to sync TX */
638 		if (!resync_pending && seq - ntls->next_seq < U32_MAX / 4)
639 			tls_offload_tx_resync_request(nskb->sk, seq,
640 						      ntls->next_seq);
641 
642 		*nr_frags = 0;
643 		return nskb;
644 	}
645 
646 	if (datalen) {
647 		u64_stats_update_begin(&r_vec->tx_sync);
648 		if (!skb_is_gso(skb))
649 			r_vec->hw_tls_tx++;
650 		else
651 			r_vec->hw_tls_tx += skb_shinfo(skb)->gso_segs;
652 		u64_stats_update_end(&r_vec->tx_sync);
653 	}
654 
655 	memcpy(tls_handle, ntls->fw_handle, sizeof(ntls->fw_handle));
656 	ntls->next_seq += datalen;
657 #endif
658 	return skb;
659 }
660 
661 void nfp_net_tls_tx_undo(struct sk_buff *skb, u64 tls_handle)
662 {
663 #ifdef CONFIG_TLS_DEVICE
664 	struct nfp_net_tls_offload_ctx *ntls;
665 	u32 datalen, seq;
666 
667 	if (!tls_handle)
668 		return;
669 	if (WARN_ON_ONCE(!tls_is_skb_tx_device_offloaded(skb)))
670 		return;
671 
672 	datalen = skb->len - skb_tcp_all_headers(skb);
673 	seq = ntohl(tcp_hdr(skb)->seq);
674 
675 	ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX);
676 	if (ntls->next_seq == seq + datalen)
677 		ntls->next_seq = seq;
678 	else
679 		WARN_ON_ONCE(1);
680 #endif
681 }
682 
683 static void nfp_net_tx_timeout(struct net_device *netdev, unsigned int txqueue)
684 {
685 	struct nfp_net *nn = netdev_priv(netdev);
686 
687 	nn_warn(nn, "TX watchdog timeout on ring: %u\n", txqueue);
688 }
689 
690 /* Receive processing */
691 static unsigned int
692 nfp_net_calc_fl_bufsz_data(struct nfp_net_dp *dp)
693 {
694 	unsigned int fl_bufsz = 0;
695 
696 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
697 		fl_bufsz += NFP_NET_MAX_PREPEND;
698 	else
699 		fl_bufsz += dp->rx_offset;
700 	fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
701 
702 	return fl_bufsz;
703 }
704 
705 static unsigned int nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
706 {
707 	unsigned int fl_bufsz;
708 
709 	fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
710 	fl_bufsz += dp->rx_dma_off;
711 	fl_bufsz += nfp_net_calc_fl_bufsz_data(dp);
712 
713 	fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
714 	fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
715 
716 	return fl_bufsz;
717 }
718 
719 static unsigned int nfp_net_calc_fl_bufsz_xsk(struct nfp_net_dp *dp)
720 {
721 	unsigned int fl_bufsz;
722 
723 	fl_bufsz = XDP_PACKET_HEADROOM;
724 	fl_bufsz += nfp_net_calc_fl_bufsz_data(dp);
725 
726 	return fl_bufsz;
727 }
728 
729 /* Setup and Configuration
730  */
731 
732 /**
733  * nfp_net_vecs_init() - Assign IRQs and setup rvecs.
734  * @nn:		NFP Network structure
735  */
736 static void nfp_net_vecs_init(struct nfp_net *nn)
737 {
738 	int numa_node = dev_to_node(&nn->pdev->dev);
739 	struct nfp_net_r_vector *r_vec;
740 	unsigned int r;
741 
742 	nn->lsc_handler = nfp_net_irq_lsc;
743 	nn->exn_handler = nfp_net_irq_exn;
744 
745 	for (r = 0; r < nn->max_r_vecs; r++) {
746 		struct msix_entry *entry;
747 
748 		entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r];
749 
750 		r_vec = &nn->r_vecs[r];
751 		r_vec->nfp_net = nn;
752 		r_vec->irq_entry = entry->entry;
753 		r_vec->irq_vector = entry->vector;
754 
755 		if (nn->dp.netdev) {
756 			r_vec->handler = nfp_net_irq_rxtx;
757 		} else {
758 			r_vec->handler = nfp_ctrl_irq_rxtx;
759 
760 			__skb_queue_head_init(&r_vec->queue);
761 			spin_lock_init(&r_vec->lock);
762 			tasklet_setup(&r_vec->tasklet, nn->dp.ops->ctrl_poll);
763 			tasklet_disable(&r_vec->tasklet);
764 		}
765 
766 		cpumask_set_cpu(cpumask_local_spread(r, numa_node), &r_vec->affinity_mask);
767 	}
768 }
769 
770 static void
771 nfp_net_napi_add(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec, int idx)
772 {
773 	if (dp->netdev)
774 		netif_napi_add(dp->netdev, &r_vec->napi,
775 			       nfp_net_has_xsk_pool_slow(dp, idx) ? dp->ops->xsk_poll : dp->ops->poll);
776 	else
777 		tasklet_enable(&r_vec->tasklet);
778 }
779 
780 static void
781 nfp_net_napi_del(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec)
782 {
783 	if (dp->netdev)
784 		netif_napi_del(&r_vec->napi);
785 	else
786 		tasklet_disable(&r_vec->tasklet);
787 }
788 
789 static void
790 nfp_net_vector_assign_rings(struct nfp_net_dp *dp,
791 			    struct nfp_net_r_vector *r_vec, int idx)
792 {
793 	r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL;
794 	r_vec->tx_ring =
795 		idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL;
796 
797 	r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ?
798 		&dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL;
799 
800 	if (nfp_net_has_xsk_pool_slow(dp, idx) || r_vec->xsk_pool) {
801 		r_vec->xsk_pool = dp->xdp_prog ? dp->xsk_pools[idx] : NULL;
802 
803 		if (r_vec->xsk_pool)
804 			xsk_pool_set_rxq_info(r_vec->xsk_pool,
805 					      &r_vec->rx_ring->xdp_rxq);
806 
807 		nfp_net_napi_del(dp, r_vec);
808 		nfp_net_napi_add(dp, r_vec, idx);
809 	}
810 }
811 
812 static int
813 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
814 		       int idx)
815 {
816 	int err;
817 
818 	nfp_net_napi_add(&nn->dp, r_vec, idx);
819 
820 	snprintf(r_vec->name, sizeof(r_vec->name),
821 		 "%s-rxtx-%d", nfp_net_name(nn), idx);
822 	err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name,
823 			  r_vec);
824 	if (err) {
825 		nfp_net_napi_del(&nn->dp, r_vec);
826 		nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector);
827 		return err;
828 	}
829 	disable_irq(r_vec->irq_vector);
830 
831 	irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask);
832 
833 	nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector,
834 	       r_vec->irq_entry);
835 
836 	return 0;
837 }
838 
839 static void
840 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
841 {
842 	irq_set_affinity_hint(r_vec->irq_vector, NULL);
843 	nfp_net_napi_del(&nn->dp, r_vec);
844 	free_irq(r_vec->irq_vector, r_vec);
845 }
846 
847 /**
848  * nfp_net_rss_write_itbl() - Write RSS indirection table to device
849  * @nn:      NFP Net device to reconfigure
850  */
851 void nfp_net_rss_write_itbl(struct nfp_net *nn)
852 {
853 	int i;
854 
855 	for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
856 		nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
857 			  get_unaligned_le32(nn->rss_itbl + i));
858 }
859 
860 /**
861  * nfp_net_rss_write_key() - Write RSS hash key to device
862  * @nn:      NFP Net device to reconfigure
863  */
864 void nfp_net_rss_write_key(struct nfp_net *nn)
865 {
866 	int i;
867 
868 	for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4)
869 		nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
870 			  get_unaligned_le32(nn->rss_key + i));
871 }
872 
873 /**
874  * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
875  * @nn:      NFP Net device to reconfigure
876  */
877 void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
878 {
879 	u8 i;
880 	u32 factor;
881 	u32 value;
882 
883 	/* Compute factor used to convert coalesce '_usecs' parameters to
884 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
885 	 * count.
886 	 */
887 	factor = nn->tlv_caps.me_freq_mhz / 16;
888 
889 	/* copy RX interrupt coalesce parameters */
890 	value = (nn->rx_coalesce_max_frames << 16) |
891 		(factor * nn->rx_coalesce_usecs);
892 	for (i = 0; i < nn->dp.num_rx_rings; i++)
893 		nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
894 
895 	/* copy TX interrupt coalesce parameters */
896 	value = (nn->tx_coalesce_max_frames << 16) |
897 		(factor * nn->tx_coalesce_usecs);
898 	for (i = 0; i < nn->dp.num_tx_rings; i++)
899 		nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
900 }
901 
902 /**
903  * nfp_net_write_mac_addr() - Write mac address to the device control BAR
904  * @nn:      NFP Net device to reconfigure
905  * @addr:    MAC address to write
906  *
907  * Writes the MAC address from the netdev to the device control BAR.  Does not
908  * perform the required reconfig.  We do a bit of byte swapping dance because
909  * firmware is LE.
910  */
911 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr)
912 {
913 	nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr));
914 	nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4));
915 }
916 
917 /**
918  * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
919  * @nn:      NFP Net device to reconfigure
920  *
921  * Warning: must be fully idempotent.
922  */
923 static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
924 {
925 	u32 new_ctrl, update;
926 	unsigned int r;
927 	int err;
928 
929 	new_ctrl = nn->dp.ctrl;
930 	new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
931 	update = NFP_NET_CFG_UPDATE_GEN;
932 	update |= NFP_NET_CFG_UPDATE_MSIX;
933 	update |= NFP_NET_CFG_UPDATE_RING;
934 
935 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
936 		new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
937 
938 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
939 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
940 
941 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
942 	err = nfp_net_reconfig(nn, update);
943 	if (err)
944 		nn_err(nn, "Could not disable device: %d\n", err);
945 
946 	for (r = 0; r < nn->dp.num_rx_rings; r++) {
947 		nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]);
948 		if (nfp_net_has_xsk_pool_slow(&nn->dp, nn->dp.rx_rings[r].idx))
949 			nfp_net_xsk_rx_bufs_free(&nn->dp.rx_rings[r]);
950 	}
951 	for (r = 0; r < nn->dp.num_tx_rings; r++)
952 		nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]);
953 	for (r = 0; r < nn->dp.num_r_vecs; r++)
954 		nfp_net_vec_clear_ring_data(nn, r);
955 
956 	nn->dp.ctrl = new_ctrl;
957 }
958 
959 /**
960  * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
961  * @nn:      NFP Net device to reconfigure
962  */
963 static int nfp_net_set_config_and_enable(struct nfp_net *nn)
964 {
965 	u32 bufsz, new_ctrl, update = 0;
966 	unsigned int r;
967 	int err;
968 
969 	new_ctrl = nn->dp.ctrl;
970 
971 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) {
972 		nfp_net_rss_write_key(nn);
973 		nfp_net_rss_write_itbl(nn);
974 		nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
975 		update |= NFP_NET_CFG_UPDATE_RSS;
976 	}
977 
978 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) {
979 		nfp_net_coalesce_write_cfg(nn);
980 		update |= NFP_NET_CFG_UPDATE_IRQMOD;
981 	}
982 
983 	for (r = 0; r < nn->dp.num_tx_rings; r++)
984 		nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r);
985 	for (r = 0; r < nn->dp.num_rx_rings; r++)
986 		nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r);
987 
988 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE,
989 		  U64_MAX >> (64 - nn->dp.num_tx_rings));
990 
991 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE,
992 		  U64_MAX >> (64 - nn->dp.num_rx_rings));
993 
994 	if (nn->dp.netdev)
995 		nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
996 
997 	nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu);
998 
999 	bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA;
1000 	nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz);
1001 
1002 	/* Enable device */
1003 	new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
1004 	update |= NFP_NET_CFG_UPDATE_GEN;
1005 	update |= NFP_NET_CFG_UPDATE_MSIX;
1006 	update |= NFP_NET_CFG_UPDATE_RING;
1007 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
1008 		new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
1009 
1010 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
1011 	nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, nn->dp.ctrl_w1);
1012 	err = nfp_net_reconfig(nn, update);
1013 	if (err) {
1014 		nfp_net_clear_config_and_disable(nn);
1015 		return err;
1016 	}
1017 
1018 	nn->dp.ctrl = new_ctrl;
1019 
1020 	for (r = 0; r < nn->dp.num_rx_rings; r++)
1021 		nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]);
1022 
1023 	return 0;
1024 }
1025 
1026 /**
1027  * nfp_net_close_stack() - Quiesce the stack (part of close)
1028  * @nn:	     NFP Net device to reconfigure
1029  */
1030 static void nfp_net_close_stack(struct nfp_net *nn)
1031 {
1032 	struct nfp_net_r_vector *r_vec;
1033 	unsigned int r;
1034 
1035 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
1036 	netif_carrier_off(nn->dp.netdev);
1037 	nn->link_up = false;
1038 
1039 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
1040 		r_vec = &nn->r_vecs[r];
1041 
1042 		disable_irq(r_vec->irq_vector);
1043 		napi_disable(&r_vec->napi);
1044 
1045 		if (r_vec->rx_ring)
1046 			cancel_work_sync(&r_vec->rx_dim.work);
1047 
1048 		if (r_vec->tx_ring)
1049 			cancel_work_sync(&r_vec->tx_dim.work);
1050 	}
1051 
1052 	netif_tx_disable(nn->dp.netdev);
1053 }
1054 
1055 /**
1056  * nfp_net_close_free_all() - Free all runtime resources
1057  * @nn:      NFP Net device to reconfigure
1058  */
1059 static void nfp_net_close_free_all(struct nfp_net *nn)
1060 {
1061 	unsigned int r;
1062 
1063 	nfp_net_tx_rings_free(&nn->dp);
1064 	nfp_net_rx_rings_free(&nn->dp);
1065 
1066 	for (r = 0; r < nn->dp.num_r_vecs; r++)
1067 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
1068 
1069 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
1070 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
1071 }
1072 
1073 /**
1074  * nfp_net_netdev_close() - Called when the device is downed
1075  * @netdev:      netdev structure
1076  */
1077 static int nfp_net_netdev_close(struct net_device *netdev)
1078 {
1079 	struct nfp_net *nn = netdev_priv(netdev);
1080 
1081 	/* Step 1: Disable RX and TX rings from the Linux kernel perspective
1082 	 */
1083 	nfp_net_close_stack(nn);
1084 
1085 	/* Step 2: Tell NFP
1086 	 */
1087 	nfp_net_clear_config_and_disable(nn);
1088 	nfp_port_configure(netdev, false);
1089 
1090 	/* Step 3: Free resources
1091 	 */
1092 	nfp_net_close_free_all(nn);
1093 
1094 	nn_dbg(nn, "%s down", netdev->name);
1095 	return 0;
1096 }
1097 
1098 void nfp_ctrl_close(struct nfp_net *nn)
1099 {
1100 	int r;
1101 
1102 	rtnl_lock();
1103 
1104 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
1105 		disable_irq(nn->r_vecs[r].irq_vector);
1106 		tasklet_disable(&nn->r_vecs[r].tasklet);
1107 	}
1108 
1109 	nfp_net_clear_config_and_disable(nn);
1110 
1111 	nfp_net_close_free_all(nn);
1112 
1113 	rtnl_unlock();
1114 }
1115 
1116 static void nfp_net_rx_dim_work(struct work_struct *work)
1117 {
1118 	struct nfp_net_r_vector *r_vec;
1119 	unsigned int factor, value;
1120 	struct dim_cq_moder moder;
1121 	struct nfp_net *nn;
1122 	struct dim *dim;
1123 
1124 	dim = container_of(work, struct dim, work);
1125 	moder = net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
1126 	r_vec = container_of(dim, struct nfp_net_r_vector, rx_dim);
1127 	nn = r_vec->nfp_net;
1128 
1129 	/* Compute factor used to convert coalesce '_usecs' parameters to
1130 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
1131 	 * count.
1132 	 */
1133 	factor = nn->tlv_caps.me_freq_mhz / 16;
1134 	if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts))
1135 		return;
1136 
1137 	/* copy RX interrupt coalesce parameters */
1138 	value = (moder.pkts << 16) | (factor * moder.usec);
1139 	nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(r_vec->rx_ring->idx), value);
1140 	(void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD);
1141 
1142 	dim->state = DIM_START_MEASURE;
1143 }
1144 
1145 static void nfp_net_tx_dim_work(struct work_struct *work)
1146 {
1147 	struct nfp_net_r_vector *r_vec;
1148 	unsigned int factor, value;
1149 	struct dim_cq_moder moder;
1150 	struct nfp_net *nn;
1151 	struct dim *dim;
1152 
1153 	dim = container_of(work, struct dim, work);
1154 	moder = net_dim_get_tx_moderation(dim->mode, dim->profile_ix);
1155 	r_vec = container_of(dim, struct nfp_net_r_vector, tx_dim);
1156 	nn = r_vec->nfp_net;
1157 
1158 	/* Compute factor used to convert coalesce '_usecs' parameters to
1159 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
1160 	 * count.
1161 	 */
1162 	factor = nn->tlv_caps.me_freq_mhz / 16;
1163 	if (nfp_net_coalesce_para_check(factor * moder.usec, moder.pkts))
1164 		return;
1165 
1166 	/* copy TX interrupt coalesce parameters */
1167 	value = (moder.pkts << 16) | (factor * moder.usec);
1168 	nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(r_vec->tx_ring->idx), value);
1169 	(void)nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_IRQMOD);
1170 
1171 	dim->state = DIM_START_MEASURE;
1172 }
1173 
1174 /**
1175  * nfp_net_open_stack() - Start the device from stack's perspective
1176  * @nn:      NFP Net device to reconfigure
1177  */
1178 static void nfp_net_open_stack(struct nfp_net *nn)
1179 {
1180 	struct nfp_net_r_vector *r_vec;
1181 	unsigned int r;
1182 
1183 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
1184 		r_vec = &nn->r_vecs[r];
1185 
1186 		if (r_vec->rx_ring) {
1187 			INIT_WORK(&r_vec->rx_dim.work, nfp_net_rx_dim_work);
1188 			r_vec->rx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
1189 		}
1190 
1191 		if (r_vec->tx_ring) {
1192 			INIT_WORK(&r_vec->tx_dim.work, nfp_net_tx_dim_work);
1193 			r_vec->tx_dim.mode = DIM_CQ_PERIOD_MODE_START_FROM_EQE;
1194 		}
1195 
1196 		napi_enable(&r_vec->napi);
1197 		enable_irq(r_vec->irq_vector);
1198 	}
1199 
1200 	netif_tx_wake_all_queues(nn->dp.netdev);
1201 
1202 	enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
1203 	nfp_net_read_link_status(nn);
1204 }
1205 
1206 static int nfp_net_open_alloc_all(struct nfp_net *nn)
1207 {
1208 	int err, r;
1209 
1210 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
1211 				      nn->exn_name, sizeof(nn->exn_name),
1212 				      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
1213 	if (err)
1214 		return err;
1215 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
1216 				      nn->lsc_name, sizeof(nn->lsc_name),
1217 				      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
1218 	if (err)
1219 		goto err_free_exn;
1220 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
1221 
1222 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
1223 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
1224 		if (err)
1225 			goto err_cleanup_vec_p;
1226 	}
1227 
1228 	err = nfp_net_rx_rings_prepare(nn, &nn->dp);
1229 	if (err)
1230 		goto err_cleanup_vec;
1231 
1232 	err = nfp_net_tx_rings_prepare(nn, &nn->dp);
1233 	if (err)
1234 		goto err_free_rx_rings;
1235 
1236 	for (r = 0; r < nn->max_r_vecs; r++)
1237 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
1238 
1239 	return 0;
1240 
1241 err_free_rx_rings:
1242 	nfp_net_rx_rings_free(&nn->dp);
1243 err_cleanup_vec:
1244 	r = nn->dp.num_r_vecs;
1245 err_cleanup_vec_p:
1246 	while (r--)
1247 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
1248 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
1249 err_free_exn:
1250 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
1251 	return err;
1252 }
1253 
1254 static int nfp_net_netdev_open(struct net_device *netdev)
1255 {
1256 	struct nfp_net *nn = netdev_priv(netdev);
1257 	int err;
1258 
1259 	/* Step 1: Allocate resources for rings and the like
1260 	 * - Request interrupts
1261 	 * - Allocate RX and TX ring resources
1262 	 * - Setup initial RSS table
1263 	 */
1264 	err = nfp_net_open_alloc_all(nn);
1265 	if (err)
1266 		return err;
1267 
1268 	err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
1269 	if (err)
1270 		goto err_free_all;
1271 
1272 	err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
1273 	if (err)
1274 		goto err_free_all;
1275 
1276 	/* Step 2: Configure the NFP
1277 	 * - Ifup the physical interface if it exists
1278 	 * - Enable rings from 0 to tx_rings/rx_rings - 1.
1279 	 * - Write MAC address (in case it changed)
1280 	 * - Set the MTU
1281 	 * - Set the Freelist buffer size
1282 	 * - Enable the FW
1283 	 */
1284 	err = nfp_port_configure(netdev, true);
1285 	if (err)
1286 		goto err_free_all;
1287 
1288 	err = nfp_net_set_config_and_enable(nn);
1289 	if (err)
1290 		goto err_port_disable;
1291 
1292 	/* Step 3: Enable for kernel
1293 	 * - put some freelist descriptors on each RX ring
1294 	 * - enable NAPI on each ring
1295 	 * - enable all TX queues
1296 	 * - set link state
1297 	 */
1298 	nfp_net_open_stack(nn);
1299 
1300 	return 0;
1301 
1302 err_port_disable:
1303 	nfp_port_configure(netdev, false);
1304 err_free_all:
1305 	nfp_net_close_free_all(nn);
1306 	return err;
1307 }
1308 
1309 int nfp_ctrl_open(struct nfp_net *nn)
1310 {
1311 	int err, r;
1312 
1313 	/* ring dumping depends on vNICs being opened/closed under rtnl */
1314 	rtnl_lock();
1315 
1316 	err = nfp_net_open_alloc_all(nn);
1317 	if (err)
1318 		goto err_unlock;
1319 
1320 	err = nfp_net_set_config_and_enable(nn);
1321 	if (err)
1322 		goto err_free_all;
1323 
1324 	for (r = 0; r < nn->dp.num_r_vecs; r++)
1325 		enable_irq(nn->r_vecs[r].irq_vector);
1326 
1327 	rtnl_unlock();
1328 
1329 	return 0;
1330 
1331 err_free_all:
1332 	nfp_net_close_free_all(nn);
1333 err_unlock:
1334 	rtnl_unlock();
1335 	return err;
1336 }
1337 
1338 int nfp_net_sched_mbox_amsg_work(struct nfp_net *nn, u32 cmd, const void *data, size_t len,
1339 				 int (*cb)(struct nfp_net *, struct nfp_mbox_amsg_entry *))
1340 {
1341 	struct nfp_mbox_amsg_entry *entry;
1342 
1343 	entry = kmalloc(sizeof(*entry) + len, GFP_ATOMIC);
1344 	if (!entry)
1345 		return -ENOMEM;
1346 
1347 	memcpy(entry->msg, data, len);
1348 	entry->cmd = cmd;
1349 	entry->cfg = cb;
1350 
1351 	spin_lock_bh(&nn->mbox_amsg.lock);
1352 	list_add_tail(&entry->list, &nn->mbox_amsg.list);
1353 	spin_unlock_bh(&nn->mbox_amsg.lock);
1354 
1355 	schedule_work(&nn->mbox_amsg.work);
1356 
1357 	return 0;
1358 }
1359 
1360 static void nfp_net_mbox_amsg_work(struct work_struct *work)
1361 {
1362 	struct nfp_net *nn = container_of(work, struct nfp_net, mbox_amsg.work);
1363 	struct nfp_mbox_amsg_entry *entry, *tmp;
1364 	struct list_head tmp_list;
1365 
1366 	INIT_LIST_HEAD(&tmp_list);
1367 
1368 	spin_lock_bh(&nn->mbox_amsg.lock);
1369 	list_splice_init(&nn->mbox_amsg.list, &tmp_list);
1370 	spin_unlock_bh(&nn->mbox_amsg.lock);
1371 
1372 	list_for_each_entry_safe(entry, tmp, &tmp_list, list) {
1373 		int err = entry->cfg(nn, entry);
1374 
1375 		if (err)
1376 			nn_err(nn, "Config cmd %d to HW failed %d.\n", entry->cmd, err);
1377 
1378 		list_del(&entry->list);
1379 		kfree(entry);
1380 	}
1381 }
1382 
1383 static int nfp_net_mc_cfg(struct nfp_net *nn, struct nfp_mbox_amsg_entry *entry)
1384 {
1385 	unsigned char *addr = entry->msg;
1386 	int ret;
1387 
1388 	ret = nfp_net_mbox_lock(nn, NFP_NET_CFG_MULTICAST_SZ);
1389 	if (ret)
1390 		return ret;
1391 
1392 	nn_writel(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_HI,
1393 		  get_unaligned_be32(addr));
1394 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_MULTICAST_MAC_LO,
1395 		  get_unaligned_be16(addr + 4));
1396 
1397 	return nfp_net_mbox_reconfig_and_unlock(nn, entry->cmd);
1398 }
1399 
1400 static int nfp_net_mc_sync(struct net_device *netdev, const unsigned char *addr)
1401 {
1402 	struct nfp_net *nn = netdev_priv(netdev);
1403 
1404 	if (netdev_mc_count(netdev) > NFP_NET_CFG_MAC_MC_MAX) {
1405 		nn_err(nn, "Requested number of MC addresses (%d) exceeds maximum (%d).\n",
1406 		       netdev_mc_count(netdev), NFP_NET_CFG_MAC_MC_MAX);
1407 		return -EINVAL;
1408 	}
1409 
1410 	return nfp_net_sched_mbox_amsg_work(nn, NFP_NET_CFG_MBOX_CMD_MULTICAST_ADD, addr,
1411 					    NFP_NET_CFG_MULTICAST_SZ, nfp_net_mc_cfg);
1412 }
1413 
1414 static int nfp_net_mc_unsync(struct net_device *netdev, const unsigned char *addr)
1415 {
1416 	struct nfp_net *nn = netdev_priv(netdev);
1417 
1418 	return nfp_net_sched_mbox_amsg_work(nn, NFP_NET_CFG_MBOX_CMD_MULTICAST_DEL, addr,
1419 					    NFP_NET_CFG_MULTICAST_SZ, nfp_net_mc_cfg);
1420 }
1421 
1422 static void nfp_net_set_rx_mode(struct net_device *netdev)
1423 {
1424 	struct nfp_net *nn = netdev_priv(netdev);
1425 	u32 new_ctrl, new_ctrl_w1;
1426 
1427 	new_ctrl = nn->dp.ctrl;
1428 	new_ctrl_w1 = nn->dp.ctrl_w1;
1429 
1430 	if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI)
1431 		new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC;
1432 	else
1433 		new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC;
1434 
1435 	if (netdev->flags & IFF_ALLMULTI)
1436 		new_ctrl_w1 &= ~NFP_NET_CFG_CTRL_MCAST_FILTER;
1437 	else
1438 		new_ctrl_w1 |= nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER;
1439 
1440 	if (netdev->flags & IFF_PROMISC) {
1441 		if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
1442 			new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
1443 		else
1444 			nn_warn(nn, "FW does not support promiscuous mode\n");
1445 	} else {
1446 		new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
1447 	}
1448 
1449 	if ((nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER) &&
1450 	    __dev_mc_sync(netdev, nfp_net_mc_sync, nfp_net_mc_unsync))
1451 		netdev_err(netdev, "Sync mc address failed\n");
1452 
1453 	if (new_ctrl == nn->dp.ctrl && new_ctrl_w1 == nn->dp.ctrl_w1)
1454 		return;
1455 
1456 	if (new_ctrl != nn->dp.ctrl)
1457 		nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
1458 	if (new_ctrl_w1 != nn->dp.ctrl_w1)
1459 		nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, new_ctrl_w1);
1460 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
1461 
1462 	nn->dp.ctrl = new_ctrl;
1463 	nn->dp.ctrl_w1 = new_ctrl_w1;
1464 }
1465 
1466 static void nfp_net_rss_init_itbl(struct nfp_net *nn)
1467 {
1468 	int i;
1469 
1470 	for (i = 0; i < sizeof(nn->rss_itbl); i++)
1471 		nn->rss_itbl[i] =
1472 			ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
1473 }
1474 
1475 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
1476 {
1477 	struct nfp_net_dp new_dp = *dp;
1478 
1479 	*dp = nn->dp;
1480 	nn->dp = new_dp;
1481 
1482 	nn->dp.netdev->mtu = new_dp.mtu;
1483 
1484 	if (!netif_is_rxfh_configured(nn->dp.netdev))
1485 		nfp_net_rss_init_itbl(nn);
1486 }
1487 
1488 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
1489 {
1490 	unsigned int r;
1491 	int err;
1492 
1493 	nfp_net_dp_swap(nn, dp);
1494 
1495 	for (r = 0; r <	nn->max_r_vecs; r++)
1496 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
1497 
1498 	err = netif_set_real_num_queues(nn->dp.netdev,
1499 					nn->dp.num_stack_tx_rings,
1500 					nn->dp.num_rx_rings);
1501 	if (err)
1502 		return err;
1503 
1504 	return nfp_net_set_config_and_enable(nn);
1505 }
1506 
1507 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
1508 {
1509 	struct nfp_net_dp *new;
1510 
1511 	new = kmalloc(sizeof(*new), GFP_KERNEL);
1512 	if (!new)
1513 		return NULL;
1514 
1515 	*new = nn->dp;
1516 
1517 	new->xsk_pools = kmemdup(new->xsk_pools,
1518 				 array_size(nn->max_r_vecs,
1519 					    sizeof(new->xsk_pools)),
1520 				 GFP_KERNEL);
1521 	if (!new->xsk_pools) {
1522 		kfree(new);
1523 		return NULL;
1524 	}
1525 
1526 	/* Clear things which need to be recomputed */
1527 	new->fl_bufsz = 0;
1528 	new->tx_rings = NULL;
1529 	new->rx_rings = NULL;
1530 	new->num_r_vecs = 0;
1531 	new->num_stack_tx_rings = 0;
1532 	new->txrwb = NULL;
1533 	new->txrwb_dma = 0;
1534 
1535 	return new;
1536 }
1537 
1538 static void nfp_net_free_dp(struct nfp_net_dp *dp)
1539 {
1540 	kfree(dp->xsk_pools);
1541 	kfree(dp);
1542 }
1543 
1544 static int
1545 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
1546 		     struct netlink_ext_ack *extack)
1547 {
1548 	unsigned int r, xsk_min_fl_bufsz;
1549 
1550 	/* XDP-enabled tests */
1551 	if (!dp->xdp_prog)
1552 		return 0;
1553 	if (dp->fl_bufsz > PAGE_SIZE) {
1554 		NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
1555 		return -EINVAL;
1556 	}
1557 	if (dp->num_tx_rings > nn->max_tx_rings) {
1558 		NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
1559 		return -EINVAL;
1560 	}
1561 
1562 	xsk_min_fl_bufsz = nfp_net_calc_fl_bufsz_xsk(dp);
1563 	for (r = 0; r < nn->max_r_vecs; r++) {
1564 		if (!dp->xsk_pools[r])
1565 			continue;
1566 
1567 		if (xsk_pool_get_rx_frame_size(dp->xsk_pools[r]) < xsk_min_fl_bufsz) {
1568 			NL_SET_ERR_MSG_MOD(extack,
1569 					   "XSK buffer pool chunk size too small");
1570 			return -EINVAL;
1571 		}
1572 	}
1573 
1574 	return 0;
1575 }
1576 
1577 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
1578 			  struct netlink_ext_ack *extack)
1579 {
1580 	int r, err;
1581 
1582 	dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
1583 
1584 	dp->num_stack_tx_rings = dp->num_tx_rings;
1585 	if (dp->xdp_prog)
1586 		dp->num_stack_tx_rings -= dp->num_rx_rings;
1587 
1588 	dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
1589 
1590 	err = nfp_net_check_config(nn, dp, extack);
1591 	if (err)
1592 		goto exit_free_dp;
1593 
1594 	if (!netif_running(dp->netdev)) {
1595 		nfp_net_dp_swap(nn, dp);
1596 		err = 0;
1597 		goto exit_free_dp;
1598 	}
1599 
1600 	/* Prepare new rings */
1601 	for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
1602 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
1603 		if (err) {
1604 			dp->num_r_vecs = r;
1605 			goto err_cleanup_vecs;
1606 		}
1607 	}
1608 
1609 	err = nfp_net_rx_rings_prepare(nn, dp);
1610 	if (err)
1611 		goto err_cleanup_vecs;
1612 
1613 	err = nfp_net_tx_rings_prepare(nn, dp);
1614 	if (err)
1615 		goto err_free_rx;
1616 
1617 	/* Stop device, swap in new rings, try to start the firmware */
1618 	nfp_net_close_stack(nn);
1619 	nfp_net_clear_config_and_disable(nn);
1620 
1621 	err = nfp_net_dp_swap_enable(nn, dp);
1622 	if (err) {
1623 		int err2;
1624 
1625 		nfp_net_clear_config_and_disable(nn);
1626 
1627 		/* Try with old configuration and old rings */
1628 		err2 = nfp_net_dp_swap_enable(nn, dp);
1629 		if (err2)
1630 			nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
1631 			       err, err2);
1632 	}
1633 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
1634 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
1635 
1636 	nfp_net_rx_rings_free(dp);
1637 	nfp_net_tx_rings_free(dp);
1638 
1639 	nfp_net_open_stack(nn);
1640 exit_free_dp:
1641 	nfp_net_free_dp(dp);
1642 
1643 	return err;
1644 
1645 err_free_rx:
1646 	nfp_net_rx_rings_free(dp);
1647 err_cleanup_vecs:
1648 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
1649 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
1650 	nfp_net_free_dp(dp);
1651 	return err;
1652 }
1653 
1654 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
1655 {
1656 	struct nfp_net *nn = netdev_priv(netdev);
1657 	struct nfp_net_dp *dp;
1658 	int err;
1659 
1660 	err = nfp_app_check_mtu(nn->app, netdev, new_mtu);
1661 	if (err)
1662 		return err;
1663 
1664 	dp = nfp_net_clone_dp(nn);
1665 	if (!dp)
1666 		return -ENOMEM;
1667 
1668 	dp->mtu = new_mtu;
1669 
1670 	return nfp_net_ring_reconfig(nn, dp, NULL);
1671 }
1672 
1673 static int
1674 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
1675 {
1676 	const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD;
1677 	struct nfp_net *nn = netdev_priv(netdev);
1678 	int err;
1679 
1680 	/* Priority tagged packets with vlan id 0 are processed by the
1681 	 * NFP as untagged packets
1682 	 */
1683 	if (!vid)
1684 		return 0;
1685 
1686 	err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ);
1687 	if (err)
1688 		return err;
1689 
1690 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
1691 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
1692 		  ETH_P_8021Q);
1693 
1694 	return nfp_net_mbox_reconfig_and_unlock(nn, cmd);
1695 }
1696 
1697 static int
1698 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
1699 {
1700 	const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL;
1701 	struct nfp_net *nn = netdev_priv(netdev);
1702 	int err;
1703 
1704 	/* Priority tagged packets with vlan id 0 are processed by the
1705 	 * NFP as untagged packets
1706 	 */
1707 	if (!vid)
1708 		return 0;
1709 
1710 	err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ);
1711 	if (err)
1712 		return err;
1713 
1714 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
1715 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
1716 		  ETH_P_8021Q);
1717 
1718 	return nfp_net_mbox_reconfig_and_unlock(nn, cmd);
1719 }
1720 
1721 static void nfp_net_stat64(struct net_device *netdev,
1722 			   struct rtnl_link_stats64 *stats)
1723 {
1724 	struct nfp_net *nn = netdev_priv(netdev);
1725 	int r;
1726 
1727 	/* Collect software stats */
1728 	for (r = 0; r < nn->max_r_vecs; r++) {
1729 		struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
1730 		u64 data[3];
1731 		unsigned int start;
1732 
1733 		do {
1734 			start = u64_stats_fetch_begin(&r_vec->rx_sync);
1735 			data[0] = r_vec->rx_pkts;
1736 			data[1] = r_vec->rx_bytes;
1737 			data[2] = r_vec->rx_drops;
1738 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
1739 		stats->rx_packets += data[0];
1740 		stats->rx_bytes += data[1];
1741 		stats->rx_dropped += data[2];
1742 
1743 		do {
1744 			start = u64_stats_fetch_begin(&r_vec->tx_sync);
1745 			data[0] = r_vec->tx_pkts;
1746 			data[1] = r_vec->tx_bytes;
1747 			data[2] = r_vec->tx_errors;
1748 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
1749 		stats->tx_packets += data[0];
1750 		stats->tx_bytes += data[1];
1751 		stats->tx_errors += data[2];
1752 	}
1753 
1754 	/* Add in device stats */
1755 	stats->multicast += nn_readq(nn, NFP_NET_CFG_STATS_RX_MC_FRAMES);
1756 	stats->rx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_RX_DISCARDS);
1757 	stats->rx_errors += nn_readq(nn, NFP_NET_CFG_STATS_RX_ERRORS);
1758 
1759 	stats->tx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_TX_DISCARDS);
1760 	stats->tx_errors += nn_readq(nn, NFP_NET_CFG_STATS_TX_ERRORS);
1761 }
1762 
1763 static int nfp_net_set_features(struct net_device *netdev,
1764 				netdev_features_t features)
1765 {
1766 	netdev_features_t changed = netdev->features ^ features;
1767 	struct nfp_net *nn = netdev_priv(netdev);
1768 	u32 new_ctrl;
1769 	int err;
1770 
1771 	/* Assume this is not called with features we have not advertised */
1772 
1773 	new_ctrl = nn->dp.ctrl;
1774 
1775 	if (changed & NETIF_F_RXCSUM) {
1776 		if (features & NETIF_F_RXCSUM)
1777 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
1778 		else
1779 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
1780 	}
1781 
1782 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
1783 		if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
1784 			new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
1785 		else
1786 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
1787 	}
1788 
1789 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
1790 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
1791 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
1792 					      NFP_NET_CFG_CTRL_LSO;
1793 		else
1794 			new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
1795 	}
1796 
1797 	if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
1798 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
1799 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?:
1800 				    NFP_NET_CFG_CTRL_RXVLAN;
1801 		else
1802 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN_ANY;
1803 	}
1804 
1805 	if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
1806 		if (features & NETIF_F_HW_VLAN_CTAG_TX)
1807 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?:
1808 				    NFP_NET_CFG_CTRL_TXVLAN;
1809 		else
1810 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN_ANY;
1811 	}
1812 
1813 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
1814 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
1815 			new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
1816 		else
1817 			new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
1818 	}
1819 
1820 	if (changed & NETIF_F_HW_VLAN_STAG_RX) {
1821 		if (features & NETIF_F_HW_VLAN_STAG_RX)
1822 			new_ctrl |= NFP_NET_CFG_CTRL_RXQINQ;
1823 		else
1824 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ;
1825 	}
1826 
1827 	if (changed & NETIF_F_SG) {
1828 		if (features & NETIF_F_SG)
1829 			new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
1830 		else
1831 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
1832 	}
1833 
1834 	err = nfp_port_set_features(netdev, features);
1835 	if (err)
1836 		return err;
1837 
1838 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
1839 	       netdev->features, features, changed);
1840 
1841 	if (new_ctrl == nn->dp.ctrl)
1842 		return 0;
1843 
1844 	nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
1845 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
1846 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
1847 	if (err)
1848 		return err;
1849 
1850 	nn->dp.ctrl = new_ctrl;
1851 
1852 	return 0;
1853 }
1854 
1855 static netdev_features_t
1856 nfp_net_fix_features(struct net_device *netdev,
1857 		     netdev_features_t features)
1858 {
1859 	if ((features & NETIF_F_HW_VLAN_CTAG_RX) &&
1860 	    (features & NETIF_F_HW_VLAN_STAG_RX)) {
1861 		if (netdev->features & NETIF_F_HW_VLAN_CTAG_RX) {
1862 			features &= ~NETIF_F_HW_VLAN_CTAG_RX;
1863 			netdev->wanted_features &= ~NETIF_F_HW_VLAN_CTAG_RX;
1864 			netdev_warn(netdev,
1865 				    "S-tag and C-tag stripping can't be enabled at the same time. Enabling S-tag stripping and disabling C-tag stripping\n");
1866 		} else if (netdev->features & NETIF_F_HW_VLAN_STAG_RX) {
1867 			features &= ~NETIF_F_HW_VLAN_STAG_RX;
1868 			netdev->wanted_features &= ~NETIF_F_HW_VLAN_STAG_RX;
1869 			netdev_warn(netdev,
1870 				    "S-tag and C-tag stripping can't be enabled at the same time. Enabling C-tag stripping and disabling S-tag stripping\n");
1871 		}
1872 	}
1873 	return features;
1874 }
1875 
1876 static netdev_features_t
1877 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
1878 		       netdev_features_t features)
1879 {
1880 	u8 l4_hdr;
1881 
1882 	/* We can't do TSO over double tagged packets (802.1AD) */
1883 	features &= vlan_features_check(skb, features);
1884 
1885 	if (!skb->encapsulation)
1886 		return features;
1887 
1888 	/* Ensure that inner L4 header offset fits into TX descriptor field */
1889 	if (skb_is_gso(skb)) {
1890 		u32 hdrlen;
1891 
1892 		hdrlen = skb_inner_tcp_all_headers(skb);
1893 
1894 		/* Assume worst case scenario of having longest possible
1895 		 * metadata prepend - 8B
1896 		 */
1897 		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8))
1898 			features &= ~NETIF_F_GSO_MASK;
1899 	}
1900 
1901 	if (xfrm_offload(skb))
1902 		return features;
1903 
1904 	/* VXLAN/GRE check */
1905 	switch (vlan_get_protocol(skb)) {
1906 	case htons(ETH_P_IP):
1907 		l4_hdr = ip_hdr(skb)->protocol;
1908 		break;
1909 	case htons(ETH_P_IPV6):
1910 		l4_hdr = ipv6_hdr(skb)->nexthdr;
1911 		break;
1912 	default:
1913 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
1914 	}
1915 
1916 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
1917 	    skb->inner_protocol != htons(ETH_P_TEB) ||
1918 	    (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
1919 	    (l4_hdr == IPPROTO_UDP &&
1920 	     (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
1921 	      sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
1922 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
1923 
1924 	return features;
1925 }
1926 
1927 static int
1928 nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len)
1929 {
1930 	struct nfp_net *nn = netdev_priv(netdev);
1931 	int n;
1932 
1933 	/* If port is defined, devlink_port is registered and devlink core
1934 	 * is taking care of name formatting.
1935 	 */
1936 	if (nn->port)
1937 		return -EOPNOTSUPP;
1938 
1939 	if (nn->dp.is_vf || nn->vnic_no_name)
1940 		return -EOPNOTSUPP;
1941 
1942 	n = snprintf(name, len, "n%d", nn->id);
1943 	if (n >= len)
1944 		return -EINVAL;
1945 
1946 	return 0;
1947 }
1948 
1949 static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf)
1950 {
1951 	struct bpf_prog *prog = bpf->prog;
1952 	struct nfp_net_dp *dp;
1953 	int err;
1954 
1955 	if (!prog == !nn->dp.xdp_prog) {
1956 		WRITE_ONCE(nn->dp.xdp_prog, prog);
1957 		xdp_attachment_setup(&nn->xdp, bpf);
1958 		return 0;
1959 	}
1960 
1961 	dp = nfp_net_clone_dp(nn);
1962 	if (!dp)
1963 		return -ENOMEM;
1964 
1965 	dp->xdp_prog = prog;
1966 	dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
1967 	dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
1968 	dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
1969 
1970 	/* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
1971 	err = nfp_net_ring_reconfig(nn, dp, bpf->extack);
1972 	if (err)
1973 		return err;
1974 
1975 	xdp_attachment_setup(&nn->xdp, bpf);
1976 	return 0;
1977 }
1978 
1979 static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf)
1980 {
1981 	int err;
1982 
1983 	err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack);
1984 	if (err)
1985 		return err;
1986 
1987 	xdp_attachment_setup(&nn->xdp_hw, bpf);
1988 	return 0;
1989 }
1990 
1991 static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
1992 {
1993 	struct nfp_net *nn = netdev_priv(netdev);
1994 
1995 	switch (xdp->command) {
1996 	case XDP_SETUP_PROG:
1997 		return nfp_net_xdp_setup_drv(nn, xdp);
1998 	case XDP_SETUP_PROG_HW:
1999 		return nfp_net_xdp_setup_hw(nn, xdp);
2000 	case XDP_SETUP_XSK_POOL:
2001 		return nfp_net_xsk_setup_pool(netdev, xdp->xsk.pool,
2002 					      xdp->xsk.queue_id);
2003 	default:
2004 		return nfp_app_bpf(nn->app, nn, xdp);
2005 	}
2006 }
2007 
2008 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
2009 {
2010 	struct nfp_net *nn = netdev_priv(netdev);
2011 	struct sockaddr *saddr = addr;
2012 	int err;
2013 
2014 	err = eth_prepare_mac_addr_change(netdev, addr);
2015 	if (err)
2016 		return err;
2017 
2018 	nfp_net_write_mac_addr(nn, saddr->sa_data);
2019 
2020 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
2021 	if (err)
2022 		return err;
2023 
2024 	eth_commit_mac_addr_change(netdev, addr);
2025 
2026 	return 0;
2027 }
2028 
2029 static int nfp_net_bridge_getlink(struct sk_buff *skb, u32 pid, u32 seq,
2030 				  struct net_device *dev, u32 filter_mask,
2031 				  int nlflags)
2032 {
2033 	struct nfp_net *nn = netdev_priv(dev);
2034 	u16 mode;
2035 
2036 	if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA))
2037 		return -EOPNOTSUPP;
2038 
2039 	mode = (nn->dp.ctrl & NFP_NET_CFG_CTRL_VEPA) ?
2040 	       BRIDGE_MODE_VEPA : BRIDGE_MODE_VEB;
2041 
2042 	return ndo_dflt_bridge_getlink(skb, pid, seq, dev, mode, 0, 0,
2043 				       nlflags, filter_mask, NULL);
2044 }
2045 
2046 static int nfp_net_bridge_setlink(struct net_device *dev, struct nlmsghdr *nlh,
2047 				  u16 flags, struct netlink_ext_ack *extack)
2048 {
2049 	struct nfp_net *nn = netdev_priv(dev);
2050 	struct nlattr *attr, *br_spec;
2051 	int rem, err;
2052 	u32 new_ctrl;
2053 	u16 mode;
2054 
2055 	if (!(nn->cap & NFP_NET_CFG_CTRL_VEPA))
2056 		return -EOPNOTSUPP;
2057 
2058 	br_spec = nlmsg_find_attr(nlh, sizeof(struct ifinfomsg), IFLA_AF_SPEC);
2059 	if (!br_spec)
2060 		return -EINVAL;
2061 
2062 	nla_for_each_nested(attr, br_spec, rem) {
2063 		if (nla_type(attr) != IFLA_BRIDGE_MODE)
2064 			continue;
2065 
2066 		if (nla_len(attr) < sizeof(mode))
2067 			return -EINVAL;
2068 
2069 		new_ctrl = nn->dp.ctrl;
2070 		mode = nla_get_u16(attr);
2071 		if (mode == BRIDGE_MODE_VEPA)
2072 			new_ctrl |= NFP_NET_CFG_CTRL_VEPA;
2073 		else if (mode == BRIDGE_MODE_VEB)
2074 			new_ctrl &= ~NFP_NET_CFG_CTRL_VEPA;
2075 		else
2076 			return -EOPNOTSUPP;
2077 
2078 		if (new_ctrl == nn->dp.ctrl)
2079 			return 0;
2080 
2081 		nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2082 		err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
2083 		if (!err)
2084 			nn->dp.ctrl = new_ctrl;
2085 
2086 		return err;
2087 	}
2088 
2089 	return -EINVAL;
2090 }
2091 
2092 const struct net_device_ops nfp_nfd3_netdev_ops = {
2093 	.ndo_init		= nfp_app_ndo_init,
2094 	.ndo_uninit		= nfp_app_ndo_uninit,
2095 	.ndo_open		= nfp_net_netdev_open,
2096 	.ndo_stop		= nfp_net_netdev_close,
2097 	.ndo_start_xmit		= nfp_net_tx,
2098 	.ndo_get_stats64	= nfp_net_stat64,
2099 	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
2100 	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
2101 	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
2102 	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
2103 	.ndo_set_vf_rate	= nfp_app_set_vf_rate,
2104 	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
2105 	.ndo_set_vf_trust	= nfp_app_set_vf_trust,
2106 	.ndo_get_vf_config	= nfp_app_get_vf_config,
2107 	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
2108 	.ndo_setup_tc		= nfp_port_setup_tc,
2109 	.ndo_tx_timeout		= nfp_net_tx_timeout,
2110 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
2111 	.ndo_change_mtu		= nfp_net_change_mtu,
2112 	.ndo_set_mac_address	= nfp_net_set_mac_address,
2113 	.ndo_set_features	= nfp_net_set_features,
2114 	.ndo_fix_features	= nfp_net_fix_features,
2115 	.ndo_features_check	= nfp_net_features_check,
2116 	.ndo_get_phys_port_name	= nfp_net_get_phys_port_name,
2117 	.ndo_bpf		= nfp_net_xdp,
2118 	.ndo_xsk_wakeup		= nfp_net_xsk_wakeup,
2119 	.ndo_bridge_getlink     = nfp_net_bridge_getlink,
2120 	.ndo_bridge_setlink     = nfp_net_bridge_setlink,
2121 };
2122 
2123 const struct net_device_ops nfp_nfdk_netdev_ops = {
2124 	.ndo_init		= nfp_app_ndo_init,
2125 	.ndo_uninit		= nfp_app_ndo_uninit,
2126 	.ndo_open		= nfp_net_netdev_open,
2127 	.ndo_stop		= nfp_net_netdev_close,
2128 	.ndo_start_xmit		= nfp_net_tx,
2129 	.ndo_get_stats64	= nfp_net_stat64,
2130 	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
2131 	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
2132 	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
2133 	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
2134 	.ndo_set_vf_rate	= nfp_app_set_vf_rate,
2135 	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
2136 	.ndo_set_vf_trust	= nfp_app_set_vf_trust,
2137 	.ndo_get_vf_config	= nfp_app_get_vf_config,
2138 	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
2139 	.ndo_setup_tc		= nfp_port_setup_tc,
2140 	.ndo_tx_timeout		= nfp_net_tx_timeout,
2141 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
2142 	.ndo_change_mtu		= nfp_net_change_mtu,
2143 	.ndo_set_mac_address	= nfp_net_set_mac_address,
2144 	.ndo_set_features	= nfp_net_set_features,
2145 	.ndo_fix_features	= nfp_net_fix_features,
2146 	.ndo_features_check	= nfp_net_features_check,
2147 	.ndo_get_phys_port_name	= nfp_net_get_phys_port_name,
2148 	.ndo_bpf		= nfp_net_xdp,
2149 	.ndo_bridge_getlink     = nfp_net_bridge_getlink,
2150 	.ndo_bridge_setlink     = nfp_net_bridge_setlink,
2151 };
2152 
2153 static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table)
2154 {
2155 	struct nfp_net *nn = netdev_priv(netdev);
2156 	int i;
2157 
2158 	BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
2159 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2) {
2160 		struct udp_tunnel_info ti0, ti1;
2161 
2162 		udp_tunnel_nic_get_port(netdev, table, i, &ti0);
2163 		udp_tunnel_nic_get_port(netdev, table, i + 1, &ti1);
2164 
2165 		nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(ti0.port),
2166 			  be16_to_cpu(ti1.port) << 16 | be16_to_cpu(ti0.port));
2167 	}
2168 
2169 	return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_VXLAN);
2170 }
2171 
2172 static const struct udp_tunnel_nic_info nfp_udp_tunnels = {
2173 	.sync_table     = nfp_udp_tunnel_sync,
2174 	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
2175 			  UDP_TUNNEL_NIC_INFO_OPEN_ONLY,
2176 	.tables         = {
2177 		{
2178 			.n_entries      = NFP_NET_N_VXLAN_PORTS,
2179 			.tunnel_types   = UDP_TUNNEL_TYPE_VXLAN,
2180 		},
2181 	},
2182 };
2183 
2184 /**
2185  * nfp_net_info() - Print general info about the NIC
2186  * @nn:      NFP Net device to reconfigure
2187  */
2188 void nfp_net_info(struct nfp_net *nn)
2189 {
2190 	nn_info(nn, "NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
2191 		nn->dp.is_vf ? "VF " : "",
2192 		nn->dp.num_tx_rings, nn->max_tx_rings,
2193 		nn->dp.num_rx_rings, nn->max_rx_rings);
2194 	nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
2195 		nn->fw_ver.extend, nn->fw_ver.class,
2196 		nn->fw_ver.major, nn->fw_ver.minor,
2197 		nn->max_mtu);
2198 	nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
2199 		nn->cap,
2200 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
2201 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
2202 		nn->cap & NFP_NET_CFG_CTRL_L2MC     ? "L2MCFILT " : "",
2203 		nn->cap & NFP_NET_CFG_CTRL_RXCSUM   ? "RXCSUM "   : "",
2204 		nn->cap & NFP_NET_CFG_CTRL_TXCSUM   ? "TXCSUM "   : "",
2205 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN   ? "RXVLAN "   : "",
2206 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN   ? "TXVLAN "   : "",
2207 		nn->cap & NFP_NET_CFG_CTRL_RXQINQ   ? "RXQINQ "   : "",
2208 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ? "RXVLANv2 "   : "",
2209 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2   ? "TXVLANv2 "   : "",
2210 		nn->cap & NFP_NET_CFG_CTRL_SCATTER  ? "SCATTER "  : "",
2211 		nn->cap & NFP_NET_CFG_CTRL_GATHER   ? "GATHER "   : "",
2212 		nn->cap & NFP_NET_CFG_CTRL_LSO      ? "TSO1 "     : "",
2213 		nn->cap & NFP_NET_CFG_CTRL_LSO2     ? "TSO2 "     : "",
2214 		nn->cap & NFP_NET_CFG_CTRL_RSS      ? "RSS1 "     : "",
2215 		nn->cap & NFP_NET_CFG_CTRL_RSS2     ? "RSS2 "     : "",
2216 		nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "",
2217 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
2218 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
2219 		nn->cap & NFP_NET_CFG_CTRL_TXRWB    ? "TXRWB "    : "",
2220 		nn->cap & NFP_NET_CFG_CTRL_VEPA     ? "VEPA "     : "",
2221 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
2222 		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
2223 		nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ?
2224 						      "RXCSUM_COMPLETE " : "",
2225 		nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "",
2226 		nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER ? "MULTICAST_FILTER " : "",
2227 		nfp_app_extra_cap(nn->app, nn));
2228 }
2229 
2230 /**
2231  * nfp_net_alloc() - Allocate netdev and related structure
2232  * @pdev:         PCI device
2233  * @dev_info:     NFP ASIC params
2234  * @ctrl_bar:     PCI IOMEM with vNIC config memory
2235  * @needs_netdev: Whether to allocate a netdev for this vNIC
2236  * @max_tx_rings: Maximum number of TX rings supported by device
2237  * @max_rx_rings: Maximum number of RX rings supported by device
2238  *
2239  * This function allocates a netdev device and fills in the initial
2240  * part of the @struct nfp_net structure.  In case of control device
2241  * nfp_net structure is allocated without the netdev.
2242  *
2243  * Return: NFP Net device structure, or ERR_PTR on error.
2244  */
2245 struct nfp_net *
2246 nfp_net_alloc(struct pci_dev *pdev, const struct nfp_dev_info *dev_info,
2247 	      void __iomem *ctrl_bar, bool needs_netdev,
2248 	      unsigned int max_tx_rings, unsigned int max_rx_rings)
2249 {
2250 	u64 dma_mask = dma_get_mask(&pdev->dev);
2251 	struct nfp_net *nn;
2252 	int err;
2253 
2254 	if (needs_netdev) {
2255 		struct net_device *netdev;
2256 
2257 		netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
2258 					    max_tx_rings, max_rx_rings);
2259 		if (!netdev)
2260 			return ERR_PTR(-ENOMEM);
2261 
2262 		SET_NETDEV_DEV(netdev, &pdev->dev);
2263 		nn = netdev_priv(netdev);
2264 		nn->dp.netdev = netdev;
2265 	} else {
2266 		nn = vzalloc(sizeof(*nn));
2267 		if (!nn)
2268 			return ERR_PTR(-ENOMEM);
2269 	}
2270 
2271 	nn->dp.dev = &pdev->dev;
2272 	nn->dp.ctrl_bar = ctrl_bar;
2273 	nn->dev_info = dev_info;
2274 	nn->pdev = pdev;
2275 	nfp_net_get_fw_version(&nn->fw_ver, ctrl_bar);
2276 
2277 	switch (FIELD_GET(NFP_NET_CFG_VERSION_DP_MASK, nn->fw_ver.extend)) {
2278 	case NFP_NET_CFG_VERSION_DP_NFD3:
2279 		nn->dp.ops = &nfp_nfd3_ops;
2280 		break;
2281 	case NFP_NET_CFG_VERSION_DP_NFDK:
2282 		if (nn->fw_ver.major < 5) {
2283 			dev_err(&pdev->dev,
2284 				"NFDK must use ABI 5 or newer, found: %d\n",
2285 				nn->fw_ver.major);
2286 			err = -EINVAL;
2287 			goto err_free_nn;
2288 		}
2289 		nn->dp.ops = &nfp_nfdk_ops;
2290 		break;
2291 	default:
2292 		err = -EINVAL;
2293 		goto err_free_nn;
2294 	}
2295 
2296 	if ((dma_mask & nn->dp.ops->dma_mask) != dma_mask) {
2297 		dev_err(&pdev->dev,
2298 			"DMA mask of loaded firmware: %llx, required DMA mask: %llx\n",
2299 			nn->dp.ops->dma_mask, dma_mask);
2300 		err = -EINVAL;
2301 		goto err_free_nn;
2302 	}
2303 
2304 	nn->max_tx_rings = max_tx_rings;
2305 	nn->max_rx_rings = max_rx_rings;
2306 
2307 	nn->dp.num_tx_rings = min_t(unsigned int,
2308 				    max_tx_rings, num_online_cpus());
2309 	nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings,
2310 				 netif_get_num_default_rss_queues());
2311 
2312 	nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings);
2313 	nn->dp.num_r_vecs = min_t(unsigned int,
2314 				  nn->dp.num_r_vecs, num_online_cpus());
2315 	nn->max_r_vecs = nn->dp.num_r_vecs;
2316 
2317 	nn->dp.xsk_pools = kcalloc(nn->max_r_vecs, sizeof(nn->dp.xsk_pools),
2318 				   GFP_KERNEL);
2319 	if (!nn->dp.xsk_pools) {
2320 		err = -ENOMEM;
2321 		goto err_free_nn;
2322 	}
2323 
2324 	nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
2325 	nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
2326 
2327 	sema_init(&nn->bar_lock, 1);
2328 
2329 	spin_lock_init(&nn->reconfig_lock);
2330 	spin_lock_init(&nn->link_status_lock);
2331 
2332 	timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0);
2333 
2334 	err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar,
2335 				     &nn->tlv_caps);
2336 	if (err)
2337 		goto err_free_nn;
2338 
2339 	err = nfp_ccm_mbox_alloc(nn);
2340 	if (err)
2341 		goto err_free_nn;
2342 
2343 	return nn;
2344 
2345 err_free_nn:
2346 	if (nn->dp.netdev)
2347 		free_netdev(nn->dp.netdev);
2348 	else
2349 		vfree(nn);
2350 	return ERR_PTR(err);
2351 }
2352 
2353 /**
2354  * nfp_net_free() - Undo what @nfp_net_alloc() did
2355  * @nn:      NFP Net device to reconfigure
2356  */
2357 void nfp_net_free(struct nfp_net *nn)
2358 {
2359 	WARN_ON(timer_pending(&nn->reconfig_timer) || nn->reconfig_posted);
2360 	nfp_ccm_mbox_free(nn);
2361 
2362 	kfree(nn->dp.xsk_pools);
2363 	if (nn->dp.netdev)
2364 		free_netdev(nn->dp.netdev);
2365 	else
2366 		vfree(nn);
2367 }
2368 
2369 /**
2370  * nfp_net_rss_key_sz() - Get current size of the RSS key
2371  * @nn:		NFP Net device instance
2372  *
2373  * Return: size of the RSS key for currently selected hash function.
2374  */
2375 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn)
2376 {
2377 	switch (nn->rss_hfunc) {
2378 	case ETH_RSS_HASH_TOP:
2379 		return NFP_NET_CFG_RSS_KEY_SZ;
2380 	case ETH_RSS_HASH_XOR:
2381 		return 0;
2382 	case ETH_RSS_HASH_CRC32:
2383 		return 4;
2384 	}
2385 
2386 	nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc);
2387 	return 0;
2388 }
2389 
2390 /**
2391  * nfp_net_rss_init() - Set the initial RSS parameters
2392  * @nn:	     NFP Net device to reconfigure
2393  */
2394 static void nfp_net_rss_init(struct nfp_net *nn)
2395 {
2396 	unsigned long func_bit, rss_cap_hfunc;
2397 	u32 reg;
2398 
2399 	/* Read the RSS function capability and select first supported func */
2400 	reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP);
2401 	rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg);
2402 	if (!rss_cap_hfunc)
2403 		rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC,
2404 					  NFP_NET_CFG_RSS_TOEPLITZ);
2405 
2406 	func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS);
2407 	if (func_bit == NFP_NET_CFG_RSS_HFUNCS) {
2408 		dev_warn(nn->dp.dev,
2409 			 "Bad RSS config, defaulting to Toeplitz hash\n");
2410 		func_bit = ETH_RSS_HASH_TOP_BIT;
2411 	}
2412 	nn->rss_hfunc = 1 << func_bit;
2413 
2414 	netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn));
2415 
2416 	nfp_net_rss_init_itbl(nn);
2417 
2418 	/* Enable IPv4/IPv6 TCP by default */
2419 	nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
2420 		      NFP_NET_CFG_RSS_IPV6_TCP |
2421 		      NFP_NET_CFG_RSS_IPV4_UDP |
2422 		      NFP_NET_CFG_RSS_IPV6_UDP |
2423 		      FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) |
2424 		      NFP_NET_CFG_RSS_MASK;
2425 }
2426 
2427 /**
2428  * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
2429  * @nn:	     NFP Net device to reconfigure
2430  */
2431 static void nfp_net_irqmod_init(struct nfp_net *nn)
2432 {
2433 	nn->rx_coalesce_usecs      = 50;
2434 	nn->rx_coalesce_max_frames = 64;
2435 	nn->tx_coalesce_usecs      = 50;
2436 	nn->tx_coalesce_max_frames = 64;
2437 
2438 	nn->rx_coalesce_adapt_on   = true;
2439 	nn->tx_coalesce_adapt_on   = true;
2440 }
2441 
2442 static void nfp_net_netdev_init(struct nfp_net *nn)
2443 {
2444 	struct net_device *netdev = nn->dp.netdev;
2445 
2446 	nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
2447 
2448 	netdev->mtu = nn->dp.mtu;
2449 
2450 	/* Advertise/enable offloads based on capabilities
2451 	 *
2452 	 * Note: netdev->features show the currently enabled features
2453 	 * and netdev->hw_features advertises which features are
2454 	 * supported.  By default we enable most features.
2455 	 */
2456 	if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
2457 		netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
2458 
2459 	netdev->hw_features = NETIF_F_HIGHDMA;
2460 	if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) {
2461 		netdev->hw_features |= NETIF_F_RXCSUM;
2462 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
2463 	}
2464 	if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
2465 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2466 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
2467 	}
2468 	if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
2469 		netdev->hw_features |= NETIF_F_SG;
2470 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER;
2471 	}
2472 	if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
2473 	    nn->cap & NFP_NET_CFG_CTRL_LSO2) {
2474 		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2475 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
2476 					 NFP_NET_CFG_CTRL_LSO;
2477 	}
2478 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY)
2479 		netdev->hw_features |= NETIF_F_RXHASH;
2480 
2481 #ifdef CONFIG_NFP_NET_IPSEC
2482 	if (nn->cap_w1 & NFP_NET_CFG_CTRL_IPSEC)
2483 		netdev->hw_features |= NETIF_F_HW_ESP | NETIF_F_HW_ESP_TX_CSUM;
2484 #endif
2485 
2486 	if (nn->cap & NFP_NET_CFG_CTRL_VXLAN) {
2487 		if (nn->cap & NFP_NET_CFG_CTRL_LSO) {
2488 			netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL |
2489 					       NETIF_F_GSO_UDP_TUNNEL_CSUM |
2490 					       NETIF_F_GSO_PARTIAL;
2491 			netdev->gso_partial_features = NETIF_F_GSO_UDP_TUNNEL_CSUM;
2492 		}
2493 		netdev->udp_tunnel_nic_info = &nfp_udp_tunnels;
2494 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN;
2495 	}
2496 	if (nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
2497 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
2498 			netdev->hw_features |= NETIF_F_GSO_GRE;
2499 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_NVGRE;
2500 	}
2501 	if (nn->cap & (NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE))
2502 		netdev->hw_enc_features = netdev->hw_features;
2503 
2504 	netdev->vlan_features = netdev->hw_features;
2505 
2506 	if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN_ANY) {
2507 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
2508 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXVLAN_V2 ?:
2509 			       NFP_NET_CFG_CTRL_RXVLAN;
2510 	}
2511 	if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN_ANY) {
2512 		if (nn->cap & NFP_NET_CFG_CTRL_LSO2) {
2513 			nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
2514 		} else {
2515 			netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
2516 			nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_TXVLAN_V2 ?:
2517 				       NFP_NET_CFG_CTRL_TXVLAN;
2518 		}
2519 	}
2520 	if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) {
2521 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
2522 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
2523 	}
2524 	if (nn->cap & NFP_NET_CFG_CTRL_RXQINQ) {
2525 		netdev->hw_features |= NETIF_F_HW_VLAN_STAG_RX;
2526 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXQINQ;
2527 	}
2528 
2529 	netdev->features = netdev->hw_features;
2530 
2531 	if (nfp_app_has_tc(nn->app) && nn->port)
2532 		netdev->hw_features |= NETIF_F_HW_TC;
2533 
2534 	/* C-Tag strip and S-Tag strip can't be supported simultaneously,
2535 	 * so enable C-Tag strip and disable S-Tag strip by default.
2536 	 */
2537 	netdev->features &= ~NETIF_F_HW_VLAN_STAG_RX;
2538 	nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_RXQINQ;
2539 
2540 	netdev->xdp_features = NETDEV_XDP_ACT_BASIC;
2541 	if (nn->app && nn->app->type->id == NFP_APP_BPF_NIC)
2542 		netdev->xdp_features |= NETDEV_XDP_ACT_HW_OFFLOAD;
2543 
2544 	/* Finalise the netdev setup */
2545 	switch (nn->dp.ops->version) {
2546 	case NFP_NFD_VER_NFD3:
2547 		netdev->netdev_ops = &nfp_nfd3_netdev_ops;
2548 		netdev->xdp_features |= NETDEV_XDP_ACT_XSK_ZEROCOPY;
2549 		break;
2550 	case NFP_NFD_VER_NFDK:
2551 		netdev->netdev_ops = &nfp_nfdk_netdev_ops;
2552 		break;
2553 	}
2554 
2555 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
2556 
2557 	/* MTU range: 68 - hw-specific max */
2558 	netdev->min_mtu = ETH_MIN_MTU;
2559 	netdev->max_mtu = nn->max_mtu;
2560 
2561 	netif_set_tso_max_segs(netdev, NFP_NET_LSO_MAX_SEGS);
2562 
2563 	netif_carrier_off(netdev);
2564 
2565 	nfp_net_set_ethtool_ops(netdev);
2566 }
2567 
2568 static int nfp_net_read_caps(struct nfp_net *nn)
2569 {
2570 	/* Get some of the read-only fields from the BAR */
2571 	nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
2572 	nn->cap_w1 = nn_readl(nn, NFP_NET_CFG_CAP_WORD1);
2573 	nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
2574 
2575 	/* ABI 4.x and ctrl vNIC always use chained metadata, in other cases
2576 	 * we allow use of non-chained metadata if RSS(v1) is the only
2577 	 * advertised capability requiring metadata.
2578 	 */
2579 	nn->dp.chained_metadata_format = nn->fw_ver.major == 4 ||
2580 					 !nn->dp.netdev ||
2581 					 !(nn->cap & NFP_NET_CFG_CTRL_RSS) ||
2582 					 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META;
2583 	/* RSS(v1) uses non-chained metadata format, except in ABI 4.x where
2584 	 * it has the same meaning as RSSv2.
2585 	 */
2586 	if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4)
2587 		nn->cap &= ~NFP_NET_CFG_CTRL_RSS;
2588 
2589 	/* Determine RX packet/metadata boundary offset */
2590 	if (nn->fw_ver.major >= 2) {
2591 		u32 reg;
2592 
2593 		reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
2594 		if (reg > NFP_NET_MAX_PREPEND) {
2595 			nn_err(nn, "Invalid rx offset: %d\n", reg);
2596 			return -EINVAL;
2597 		}
2598 		nn->dp.rx_offset = reg;
2599 	} else {
2600 		nn->dp.rx_offset = NFP_NET_RX_OFFSET;
2601 	}
2602 
2603 	/* Mask out NFD-version-specific features */
2604 	nn->cap &= nn->dp.ops->cap_mask;
2605 
2606 	/* For control vNICs mask out the capabilities app doesn't want. */
2607 	if (!nn->dp.netdev)
2608 		nn->cap &= nn->app->type->ctrl_cap_mask;
2609 
2610 	return 0;
2611 }
2612 
2613 /**
2614  * nfp_net_init() - Initialise/finalise the nfp_net structure
2615  * @nn:		NFP Net device structure
2616  *
2617  * Return: 0 on success or negative errno on error.
2618  */
2619 int nfp_net_init(struct nfp_net *nn)
2620 {
2621 	int err;
2622 
2623 	nn->dp.rx_dma_dir = DMA_FROM_DEVICE;
2624 
2625 	err = nfp_net_read_caps(nn);
2626 	if (err)
2627 		return err;
2628 
2629 	/* Set default MTU and Freelist buffer size */
2630 	if (!nfp_net_is_data_vnic(nn) && nn->app->ctrl_mtu) {
2631 		nn->dp.mtu = min(nn->app->ctrl_mtu, nn->max_mtu);
2632 	} else if (nn->max_mtu < NFP_NET_DEFAULT_MTU) {
2633 		nn->dp.mtu = nn->max_mtu;
2634 	} else {
2635 		nn->dp.mtu = NFP_NET_DEFAULT_MTU;
2636 	}
2637 	nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp);
2638 
2639 	if (nfp_app_ctrl_uses_data_vnics(nn->app))
2640 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_CMSG_DATA;
2641 
2642 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) {
2643 		nfp_net_rss_init(nn);
2644 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?:
2645 					 NFP_NET_CFG_CTRL_RSS;
2646 	}
2647 
2648 	/* Allow L2 Broadcast and Multicast through by default, if supported */
2649 	if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
2650 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC;
2651 
2652 	/* Allow IRQ moderation, if supported */
2653 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
2654 		nfp_net_irqmod_init(nn);
2655 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
2656 	}
2657 
2658 	/* Enable TX pointer writeback, if supported */
2659 	if (nn->cap & NFP_NET_CFG_CTRL_TXRWB)
2660 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXRWB;
2661 
2662 	if (nn->cap_w1 & NFP_NET_CFG_CTRL_MCAST_FILTER)
2663 		nn->dp.ctrl_w1 |= NFP_NET_CFG_CTRL_MCAST_FILTER;
2664 
2665 	/* Stash the re-configuration queue away.  First odd queue in TX Bar */
2666 	nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
2667 
2668 	/* Make sure the FW knows the netdev is supposed to be disabled here */
2669 	nn_writel(nn, NFP_NET_CFG_CTRL, 0);
2670 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2671 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2672 	nn_writel(nn, NFP_NET_CFG_CTRL_WORD1, 0);
2673 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
2674 				   NFP_NET_CFG_UPDATE_GEN);
2675 	if (err)
2676 		return err;
2677 
2678 	if (nn->dp.netdev) {
2679 		nfp_net_netdev_init(nn);
2680 
2681 		err = nfp_ccm_mbox_init(nn);
2682 		if (err)
2683 			return err;
2684 
2685 		err = nfp_net_tls_init(nn);
2686 		if (err)
2687 			goto err_clean_mbox;
2688 
2689 		nfp_net_ipsec_init(nn);
2690 	}
2691 
2692 	nfp_net_vecs_init(nn);
2693 
2694 	if (!nn->dp.netdev)
2695 		return 0;
2696 
2697 	spin_lock_init(&nn->mbox_amsg.lock);
2698 	INIT_LIST_HEAD(&nn->mbox_amsg.list);
2699 	INIT_WORK(&nn->mbox_amsg.work, nfp_net_mbox_amsg_work);
2700 
2701 	return register_netdev(nn->dp.netdev);
2702 
2703 err_clean_mbox:
2704 	nfp_ccm_mbox_clean(nn);
2705 	return err;
2706 }
2707 
2708 /**
2709  * nfp_net_clean() - Undo what nfp_net_init() did.
2710  * @nn:		NFP Net device structure
2711  */
2712 void nfp_net_clean(struct nfp_net *nn)
2713 {
2714 	if (!nn->dp.netdev)
2715 		return;
2716 
2717 	unregister_netdev(nn->dp.netdev);
2718 	nfp_net_ipsec_clean(nn);
2719 	nfp_ccm_mbox_clean(nn);
2720 	flush_work(&nn->mbox_amsg.work);
2721 	nfp_net_reconfig_wait_posted(nn);
2722 }
2723