1 /*
2  * Copyright (C) 2015 Netronome Systems, Inc.
3  *
4  * This software is dual licensed under the GNU General License Version 2,
5  * June 1991 as shown in the file COPYING in the top-level directory of this
6  * source tree or the BSD 2-Clause License provided below.  You have the
7  * option to license this software under the complete terms of either license.
8  *
9  * The BSD 2-Clause License:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      1. Redistributions of source code must retain the above
16  *         copyright notice, this list of conditions and the following
17  *         disclaimer.
18  *
19  *      2. Redistributions in binary form must reproduce the above
20  *         copyright notice, this list of conditions and the following
21  *         disclaimer in the documentation and/or other materials
22  *         provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 /*
35  * nfp_net_common.c
36  * Netronome network device driver: Common functions between PF and VF
37  * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
38  *          Jason McMullan <jason.mcmullan@netronome.com>
39  *          Rolf Neugebauer <rolf.neugebauer@netronome.com>
40  *          Brad Petrus <brad.petrus@netronome.com>
41  *          Chris Telfer <chris.telfer@netronome.com>
42  */
43 
44 #include <linux/bpf.h>
45 #include <linux/module.h>
46 #include <linux/kernel.h>
47 #include <linux/init.h>
48 #include <linux/fs.h>
49 #include <linux/netdevice.h>
50 #include <linux/etherdevice.h>
51 #include <linux/interrupt.h>
52 #include <linux/ip.h>
53 #include <linux/ipv6.h>
54 #include <linux/page_ref.h>
55 #include <linux/pci.h>
56 #include <linux/pci_regs.h>
57 #include <linux/msi.h>
58 #include <linux/ethtool.h>
59 #include <linux/log2.h>
60 #include <linux/if_vlan.h>
61 #include <linux/random.h>
62 
63 #include <linux/ktime.h>
64 
65 #include <net/pkt_cls.h>
66 #include <net/vxlan.h>
67 
68 #include "nfp_net_ctrl.h"
69 #include "nfp_net.h"
70 
71 /**
72  * nfp_net_get_fw_version() - Read and parse the FW version
73  * @fw_ver:	Output fw_version structure to read to
74  * @ctrl_bar:	Mapped address of the control BAR
75  */
76 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
77 			    void __iomem *ctrl_bar)
78 {
79 	u32 reg;
80 
81 	reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
82 	put_unaligned_le32(reg, fw_ver);
83 }
84 
85 static dma_addr_t
86 nfp_net_dma_map_rx(struct nfp_net *nn, void *frag, unsigned int bufsz,
87 		   int direction)
88 {
89 	return dma_map_single(&nn->pdev->dev, frag + NFP_NET_RX_BUF_HEADROOM,
90 			      bufsz - NFP_NET_RX_BUF_NON_DATA, direction);
91 }
92 
93 static void
94 nfp_net_dma_unmap_rx(struct nfp_net *nn, dma_addr_t dma_addr,
95 		     unsigned int bufsz, int direction)
96 {
97 	dma_unmap_single(&nn->pdev->dev, dma_addr,
98 			 bufsz - NFP_NET_RX_BUF_NON_DATA, direction);
99 }
100 
101 /* Firmware reconfig
102  *
103  * Firmware reconfig may take a while so we have two versions of it -
104  * synchronous and asynchronous (posted).  All synchronous callers are holding
105  * RTNL so we don't have to worry about serializing them.
106  */
107 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
108 {
109 	nn_writel(nn, NFP_NET_CFG_UPDATE, update);
110 	/* ensure update is written before pinging HW */
111 	nn_pci_flush(nn);
112 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
113 }
114 
115 /* Pass 0 as update to run posted reconfigs. */
116 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
117 {
118 	update |= nn->reconfig_posted;
119 	nn->reconfig_posted = 0;
120 
121 	nfp_net_reconfig_start(nn, update);
122 
123 	nn->reconfig_timer_active = true;
124 	mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
125 }
126 
127 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
128 {
129 	u32 reg;
130 
131 	reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
132 	if (reg == 0)
133 		return true;
134 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
135 		nn_err(nn, "Reconfig error: 0x%08x\n", reg);
136 		return true;
137 	} else if (last_check) {
138 		nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
139 		return true;
140 	}
141 
142 	return false;
143 }
144 
145 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
146 {
147 	bool timed_out = false;
148 
149 	/* Poll update field, waiting for NFP to ack the config */
150 	while (!nfp_net_reconfig_check_done(nn, timed_out)) {
151 		msleep(1);
152 		timed_out = time_is_before_eq_jiffies(deadline);
153 	}
154 
155 	if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
156 		return -EIO;
157 
158 	return timed_out ? -EIO : 0;
159 }
160 
161 static void nfp_net_reconfig_timer(unsigned long data)
162 {
163 	struct nfp_net *nn = (void *)data;
164 
165 	spin_lock_bh(&nn->reconfig_lock);
166 
167 	nn->reconfig_timer_active = false;
168 
169 	/* If sync caller is present it will take over from us */
170 	if (nn->reconfig_sync_present)
171 		goto done;
172 
173 	/* Read reconfig status and report errors */
174 	nfp_net_reconfig_check_done(nn, true);
175 
176 	if (nn->reconfig_posted)
177 		nfp_net_reconfig_start_async(nn, 0);
178 done:
179 	spin_unlock_bh(&nn->reconfig_lock);
180 }
181 
182 /**
183  * nfp_net_reconfig_post() - Post async reconfig request
184  * @nn:      NFP Net device to reconfigure
185  * @update:  The value for the update field in the BAR config
186  *
187  * Record FW reconfiguration request.  Reconfiguration will be kicked off
188  * whenever reconfiguration machinery is idle.  Multiple requests can be
189  * merged together!
190  */
191 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
192 {
193 	spin_lock_bh(&nn->reconfig_lock);
194 
195 	/* Sync caller will kick off async reconf when it's done, just post */
196 	if (nn->reconfig_sync_present) {
197 		nn->reconfig_posted |= update;
198 		goto done;
199 	}
200 
201 	/* Opportunistically check if the previous command is done */
202 	if (!nn->reconfig_timer_active ||
203 	    nfp_net_reconfig_check_done(nn, false))
204 		nfp_net_reconfig_start_async(nn, update);
205 	else
206 		nn->reconfig_posted |= update;
207 done:
208 	spin_unlock_bh(&nn->reconfig_lock);
209 }
210 
211 /**
212  * nfp_net_reconfig() - Reconfigure the firmware
213  * @nn:      NFP Net device to reconfigure
214  * @update:  The value for the update field in the BAR config
215  *
216  * Write the update word to the BAR and ping the reconfig queue.  The
217  * poll until the firmware has acknowledged the update by zeroing the
218  * update word.
219  *
220  * Return: Negative errno on error, 0 on success
221  */
222 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
223 {
224 	bool cancelled_timer = false;
225 	u32 pre_posted_requests;
226 	int ret;
227 
228 	spin_lock_bh(&nn->reconfig_lock);
229 
230 	nn->reconfig_sync_present = true;
231 
232 	if (nn->reconfig_timer_active) {
233 		del_timer(&nn->reconfig_timer);
234 		nn->reconfig_timer_active = false;
235 		cancelled_timer = true;
236 	}
237 	pre_posted_requests = nn->reconfig_posted;
238 	nn->reconfig_posted = 0;
239 
240 	spin_unlock_bh(&nn->reconfig_lock);
241 
242 	if (cancelled_timer)
243 		nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
244 
245 	/* Run the posted reconfigs which were issued before we started */
246 	if (pre_posted_requests) {
247 		nfp_net_reconfig_start(nn, pre_posted_requests);
248 		nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
249 	}
250 
251 	nfp_net_reconfig_start(nn, update);
252 	ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
253 
254 	spin_lock_bh(&nn->reconfig_lock);
255 
256 	if (nn->reconfig_posted)
257 		nfp_net_reconfig_start_async(nn, 0);
258 
259 	nn->reconfig_sync_present = false;
260 
261 	spin_unlock_bh(&nn->reconfig_lock);
262 
263 	return ret;
264 }
265 
266 /* Interrupt configuration and handling
267  */
268 
269 /**
270  * nfp_net_irq_unmask() - Unmask automasked interrupt
271  * @nn:       NFP Network structure
272  * @entry_nr: MSI-X table entry
273  *
274  * Clear the ICR for the IRQ entry.
275  */
276 static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
277 {
278 	nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
279 	nn_pci_flush(nn);
280 }
281 
282 /**
283  * nfp_net_msix_alloc() - Try to allocate MSI-X irqs
284  * @nn:       NFP Network structure
285  * @nr_vecs:  Number of MSI-X vectors to allocate
286  *
287  * For MSI-X we want at least NFP_NET_NON_Q_VECTORS + 1 vectors.
288  *
289  * Return: Number of MSI-X vectors obtained or 0 on error.
290  */
291 static int nfp_net_msix_alloc(struct nfp_net *nn, int nr_vecs)
292 {
293 	struct pci_dev *pdev = nn->pdev;
294 	int nvecs;
295 	int i;
296 
297 	for (i = 0; i < nr_vecs; i++)
298 		nn->irq_entries[i].entry = i;
299 
300 	nvecs = pci_enable_msix_range(pdev, nn->irq_entries,
301 				      NFP_NET_NON_Q_VECTORS + 1, nr_vecs);
302 	if (nvecs < 0) {
303 		nn_warn(nn, "Failed to enable MSI-X. Wanted %d-%d (err=%d)\n",
304 			NFP_NET_NON_Q_VECTORS + 1, nr_vecs, nvecs);
305 		return 0;
306 	}
307 
308 	return nvecs;
309 }
310 
311 /**
312  * nfp_net_irqs_alloc() - allocates MSI-X irqs
313  * @nn:       NFP Network structure
314  *
315  * Return: Number of irqs obtained or 0 on error.
316  */
317 int nfp_net_irqs_alloc(struct nfp_net *nn)
318 {
319 	int wanted_irqs;
320 	unsigned int n;
321 
322 	wanted_irqs = nn->num_r_vecs + NFP_NET_NON_Q_VECTORS;
323 
324 	n = nfp_net_msix_alloc(nn, wanted_irqs);
325 	if (n == 0) {
326 		nn_err(nn, "Failed to allocate MSI-X IRQs\n");
327 		return 0;
328 	}
329 
330 	nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
331 	nn->num_r_vecs = nn->max_r_vecs;
332 
333 	if (n < wanted_irqs)
334 		nn_warn(nn, "Unable to allocate %d vectors. Got %d instead\n",
335 			wanted_irqs, n);
336 
337 	return n;
338 }
339 
340 /**
341  * nfp_net_irqs_disable() - Disable interrupts
342  * @nn:       NFP Network structure
343  *
344  * Undoes what @nfp_net_irqs_alloc() does.
345  */
346 void nfp_net_irqs_disable(struct nfp_net *nn)
347 {
348 	pci_disable_msix(nn->pdev);
349 }
350 
351 /**
352  * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
353  * @irq:      Interrupt
354  * @data:     Opaque data structure
355  *
356  * Return: Indicate if the interrupt has been handled.
357  */
358 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
359 {
360 	struct nfp_net_r_vector *r_vec = data;
361 
362 	napi_schedule_irqoff(&r_vec->napi);
363 
364 	/* The FW auto-masks any interrupt, either via the MASK bit in
365 	 * the MSI-X table or via the per entry ICR field.  So there
366 	 * is no need to disable interrupts here.
367 	 */
368 	return IRQ_HANDLED;
369 }
370 
371 /**
372  * nfp_net_read_link_status() - Reread link status from control BAR
373  * @nn:       NFP Network structure
374  */
375 static void nfp_net_read_link_status(struct nfp_net *nn)
376 {
377 	unsigned long flags;
378 	bool link_up;
379 	u32 sts;
380 
381 	spin_lock_irqsave(&nn->link_status_lock, flags);
382 
383 	sts = nn_readl(nn, NFP_NET_CFG_STS);
384 	link_up = !!(sts & NFP_NET_CFG_STS_LINK);
385 
386 	if (nn->link_up == link_up)
387 		goto out;
388 
389 	nn->link_up = link_up;
390 
391 	if (nn->link_up) {
392 		netif_carrier_on(nn->netdev);
393 		netdev_info(nn->netdev, "NIC Link is Up\n");
394 	} else {
395 		netif_carrier_off(nn->netdev);
396 		netdev_info(nn->netdev, "NIC Link is Down\n");
397 	}
398 out:
399 	spin_unlock_irqrestore(&nn->link_status_lock, flags);
400 }
401 
402 /**
403  * nfp_net_irq_lsc() - Interrupt service routine for link state changes
404  * @irq:      Interrupt
405  * @data:     Opaque data structure
406  *
407  * Return: Indicate if the interrupt has been handled.
408  */
409 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
410 {
411 	struct nfp_net *nn = data;
412 
413 	nfp_net_read_link_status(nn);
414 
415 	nfp_net_irq_unmask(nn, NFP_NET_IRQ_LSC_IDX);
416 
417 	return IRQ_HANDLED;
418 }
419 
420 /**
421  * nfp_net_irq_exn() - Interrupt service routine for exceptions
422  * @irq:      Interrupt
423  * @data:     Opaque data structure
424  *
425  * Return: Indicate if the interrupt has been handled.
426  */
427 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
428 {
429 	struct nfp_net *nn = data;
430 
431 	nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
432 	/* XXX TO BE IMPLEMENTED */
433 	return IRQ_HANDLED;
434 }
435 
436 /**
437  * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
438  * @tx_ring:  TX ring structure
439  * @r_vec:    IRQ vector servicing this ring
440  * @idx:      Ring index
441  */
442 static void
443 nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
444 		     struct nfp_net_r_vector *r_vec, unsigned int idx)
445 {
446 	struct nfp_net *nn = r_vec->nfp_net;
447 
448 	tx_ring->idx = idx;
449 	tx_ring->r_vec = r_vec;
450 
451 	tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
452 	tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
453 }
454 
455 /**
456  * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
457  * @rx_ring:  RX ring structure
458  * @r_vec:    IRQ vector servicing this ring
459  * @idx:      Ring index
460  */
461 static void
462 nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
463 		     struct nfp_net_r_vector *r_vec, unsigned int idx)
464 {
465 	struct nfp_net *nn = r_vec->nfp_net;
466 
467 	rx_ring->idx = idx;
468 	rx_ring->r_vec = r_vec;
469 
470 	rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
471 	rx_ring->rx_qcidx = rx_ring->fl_qcidx + (nn->stride_rx - 1);
472 
473 	rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
474 	rx_ring->qcp_rx = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->rx_qcidx);
475 }
476 
477 /**
478  * nfp_net_irqs_assign() - Assign IRQs and setup rvecs.
479  * @netdev:   netdev structure
480  */
481 static void nfp_net_irqs_assign(struct net_device *netdev)
482 {
483 	struct nfp_net *nn = netdev_priv(netdev);
484 	struct nfp_net_r_vector *r_vec;
485 	int r;
486 
487 	if (nn->num_rx_rings > nn->num_r_vecs ||
488 	    nn->num_tx_rings > nn->num_r_vecs)
489 		nn_warn(nn, "More rings (%d,%d) than vectors (%d).\n",
490 			nn->num_rx_rings, nn->num_tx_rings, nn->num_r_vecs);
491 
492 	nn->num_rx_rings = min(nn->num_r_vecs, nn->num_rx_rings);
493 	nn->num_tx_rings = min(nn->num_r_vecs, nn->num_tx_rings);
494 	nn->num_stack_tx_rings = nn->num_tx_rings;
495 
496 	nn->lsc_handler = nfp_net_irq_lsc;
497 	nn->exn_handler = nfp_net_irq_exn;
498 
499 	for (r = 0; r < nn->max_r_vecs; r++) {
500 		r_vec = &nn->r_vecs[r];
501 		r_vec->nfp_net = nn;
502 		r_vec->handler = nfp_net_irq_rxtx;
503 		r_vec->irq_idx = NFP_NET_NON_Q_VECTORS + r;
504 
505 		cpumask_set_cpu(r, &r_vec->affinity_mask);
506 	}
507 }
508 
509 /**
510  * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
511  * @nn:		NFP Network structure
512  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
513  * @format:	printf-style format to construct the interrupt name
514  * @name:	Pointer to allocated space for interrupt name
515  * @name_sz:	Size of space for interrupt name
516  * @vector_idx:	Index of MSI-X vector used for this interrupt
517  * @handler:	IRQ handler to register for this interrupt
518  */
519 static int
520 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
521 			const char *format, char *name, size_t name_sz,
522 			unsigned int vector_idx, irq_handler_t handler)
523 {
524 	struct msix_entry *entry;
525 	int err;
526 
527 	entry = &nn->irq_entries[vector_idx];
528 
529 	snprintf(name, name_sz, format, netdev_name(nn->netdev));
530 	err = request_irq(entry->vector, handler, 0, name, nn);
531 	if (err) {
532 		nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
533 		       entry->vector, err);
534 		return err;
535 	}
536 	nn_writeb(nn, ctrl_offset, vector_idx);
537 
538 	return 0;
539 }
540 
541 /**
542  * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
543  * @nn:		NFP Network structure
544  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
545  * @vector_idx:	Index of MSI-X vector used for this interrupt
546  */
547 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
548 				 unsigned int vector_idx)
549 {
550 	nn_writeb(nn, ctrl_offset, 0xff);
551 	free_irq(nn->irq_entries[vector_idx].vector, nn);
552 }
553 
554 /* Transmit
555  *
556  * One queue controller peripheral queue is used for transmit.  The
557  * driver en-queues packets for transmit by advancing the write
558  * pointer.  The device indicates that packets have transmitted by
559  * advancing the read pointer.  The driver maintains a local copy of
560  * the read and write pointer in @struct nfp_net_tx_ring.  The driver
561  * keeps @wr_p in sync with the queue controller write pointer and can
562  * determine how many packets have been transmitted by comparing its
563  * copy of the read pointer @rd_p with the read pointer maintained by
564  * the queue controller peripheral.
565  */
566 
567 /**
568  * nfp_net_tx_full() - Check if the TX ring is full
569  * @tx_ring: TX ring to check
570  * @dcnt:    Number of descriptors that need to be enqueued (must be >= 1)
571  *
572  * This function checks, based on the *host copy* of read/write
573  * pointer if a given TX ring is full.  The real TX queue may have
574  * some newly made available slots.
575  *
576  * Return: True if the ring is full.
577  */
578 static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
579 {
580 	return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
581 }
582 
583 /* Wrappers for deciding when to stop and restart TX queues */
584 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
585 {
586 	return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
587 }
588 
589 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
590 {
591 	return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
592 }
593 
594 /**
595  * nfp_net_tx_ring_stop() - stop tx ring
596  * @nd_q:    netdev queue
597  * @tx_ring: driver tx queue structure
598  *
599  * Safely stop TX ring.  Remember that while we are running .start_xmit()
600  * someone else may be cleaning the TX ring completions so we need to be
601  * extra careful here.
602  */
603 static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
604 				 struct nfp_net_tx_ring *tx_ring)
605 {
606 	netif_tx_stop_queue(nd_q);
607 
608 	/* We can race with the TX completion out of NAPI so recheck */
609 	smp_mb();
610 	if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
611 		netif_tx_start_queue(nd_q);
612 }
613 
614 /**
615  * nfp_net_tx_tso() - Set up Tx descriptor for LSO
616  * @nn:  NFP Net device
617  * @r_vec: per-ring structure
618  * @txbuf: Pointer to driver soft TX descriptor
619  * @txd: Pointer to HW TX descriptor
620  * @skb: Pointer to SKB
621  *
622  * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
623  * Return error on packet header greater than maximum supported LSO header size.
624  */
625 static void nfp_net_tx_tso(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
626 			   struct nfp_net_tx_buf *txbuf,
627 			   struct nfp_net_tx_desc *txd, struct sk_buff *skb)
628 {
629 	u32 hdrlen;
630 	u16 mss;
631 
632 	if (!skb_is_gso(skb))
633 		return;
634 
635 	if (!skb->encapsulation)
636 		hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
637 	else
638 		hdrlen = skb_inner_transport_header(skb) - skb->data +
639 			inner_tcp_hdrlen(skb);
640 
641 	txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
642 	txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
643 
644 	mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
645 	txd->l4_offset = hdrlen;
646 	txd->mss = cpu_to_le16(mss);
647 	txd->flags |= PCIE_DESC_TX_LSO;
648 
649 	u64_stats_update_begin(&r_vec->tx_sync);
650 	r_vec->tx_lso++;
651 	u64_stats_update_end(&r_vec->tx_sync);
652 }
653 
654 /**
655  * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
656  * @nn:  NFP Net device
657  * @r_vec: per-ring structure
658  * @txbuf: Pointer to driver soft TX descriptor
659  * @txd: Pointer to TX descriptor
660  * @skb: Pointer to SKB
661  *
662  * This function sets the TX checksum flags in the TX descriptor based
663  * on the configuration and the protocol of the packet to be transmitted.
664  */
665 static void nfp_net_tx_csum(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
666 			    struct nfp_net_tx_buf *txbuf,
667 			    struct nfp_net_tx_desc *txd, struct sk_buff *skb)
668 {
669 	struct ipv6hdr *ipv6h;
670 	struct iphdr *iph;
671 	u8 l4_hdr;
672 
673 	if (!(nn->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
674 		return;
675 
676 	if (skb->ip_summed != CHECKSUM_PARTIAL)
677 		return;
678 
679 	txd->flags |= PCIE_DESC_TX_CSUM;
680 	if (skb->encapsulation)
681 		txd->flags |= PCIE_DESC_TX_ENCAP;
682 
683 	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
684 	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
685 
686 	if (iph->version == 4) {
687 		txd->flags |= PCIE_DESC_TX_IP4_CSUM;
688 		l4_hdr = iph->protocol;
689 	} else if (ipv6h->version == 6) {
690 		l4_hdr = ipv6h->nexthdr;
691 	} else {
692 		nn_warn_ratelimit(nn, "partial checksum but ipv=%x!\n",
693 				  iph->version);
694 		return;
695 	}
696 
697 	switch (l4_hdr) {
698 	case IPPROTO_TCP:
699 		txd->flags |= PCIE_DESC_TX_TCP_CSUM;
700 		break;
701 	case IPPROTO_UDP:
702 		txd->flags |= PCIE_DESC_TX_UDP_CSUM;
703 		break;
704 	default:
705 		nn_warn_ratelimit(nn, "partial checksum but l4 proto=%x!\n",
706 				  l4_hdr);
707 		return;
708 	}
709 
710 	u64_stats_update_begin(&r_vec->tx_sync);
711 	if (skb->encapsulation)
712 		r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
713 	else
714 		r_vec->hw_csum_tx += txbuf->pkt_cnt;
715 	u64_stats_update_end(&r_vec->tx_sync);
716 }
717 
718 static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
719 {
720 	wmb();
721 	nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
722 	tx_ring->wr_ptr_add = 0;
723 }
724 
725 /**
726  * nfp_net_tx() - Main transmit entry point
727  * @skb:    SKB to transmit
728  * @netdev: netdev structure
729  *
730  * Return: NETDEV_TX_OK on success.
731  */
732 static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
733 {
734 	struct nfp_net *nn = netdev_priv(netdev);
735 	const struct skb_frag_struct *frag;
736 	struct nfp_net_r_vector *r_vec;
737 	struct nfp_net_tx_desc *txd, txdg;
738 	struct nfp_net_tx_buf *txbuf;
739 	struct nfp_net_tx_ring *tx_ring;
740 	struct netdev_queue *nd_q;
741 	dma_addr_t dma_addr;
742 	unsigned int fsize;
743 	int f, nr_frags;
744 	int wr_idx;
745 	u16 qidx;
746 
747 	qidx = skb_get_queue_mapping(skb);
748 	tx_ring = &nn->tx_rings[qidx];
749 	r_vec = tx_ring->r_vec;
750 	nd_q = netdev_get_tx_queue(nn->netdev, qidx);
751 
752 	nr_frags = skb_shinfo(skb)->nr_frags;
753 
754 	if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
755 		nn_warn_ratelimit(nn, "TX ring %d busy. wrp=%u rdp=%u\n",
756 				  qidx, tx_ring->wr_p, tx_ring->rd_p);
757 		netif_tx_stop_queue(nd_q);
758 		u64_stats_update_begin(&r_vec->tx_sync);
759 		r_vec->tx_busy++;
760 		u64_stats_update_end(&r_vec->tx_sync);
761 		return NETDEV_TX_BUSY;
762 	}
763 
764 	/* Start with the head skbuf */
765 	dma_addr = dma_map_single(&nn->pdev->dev, skb->data, skb_headlen(skb),
766 				  DMA_TO_DEVICE);
767 	if (dma_mapping_error(&nn->pdev->dev, dma_addr))
768 		goto err_free;
769 
770 	wr_idx = tx_ring->wr_p & (tx_ring->cnt - 1);
771 
772 	/* Stash the soft descriptor of the head then initialize it */
773 	txbuf = &tx_ring->txbufs[wr_idx];
774 	txbuf->skb = skb;
775 	txbuf->dma_addr = dma_addr;
776 	txbuf->fidx = -1;
777 	txbuf->pkt_cnt = 1;
778 	txbuf->real_len = skb->len;
779 
780 	/* Build TX descriptor */
781 	txd = &tx_ring->txds[wr_idx];
782 	txd->offset_eop = (nr_frags == 0) ? PCIE_DESC_TX_EOP : 0;
783 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
784 	nfp_desc_set_dma_addr(txd, dma_addr);
785 	txd->data_len = cpu_to_le16(skb->len);
786 
787 	txd->flags = 0;
788 	txd->mss = 0;
789 	txd->l4_offset = 0;
790 
791 	nfp_net_tx_tso(nn, r_vec, txbuf, txd, skb);
792 
793 	nfp_net_tx_csum(nn, r_vec, txbuf, txd, skb);
794 
795 	if (skb_vlan_tag_present(skb) && nn->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
796 		txd->flags |= PCIE_DESC_TX_VLAN;
797 		txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
798 	}
799 
800 	/* Gather DMA */
801 	if (nr_frags > 0) {
802 		/* all descs must match except for in addr, length and eop */
803 		txdg = *txd;
804 
805 		for (f = 0; f < nr_frags; f++) {
806 			frag = &skb_shinfo(skb)->frags[f];
807 			fsize = skb_frag_size(frag);
808 
809 			dma_addr = skb_frag_dma_map(&nn->pdev->dev, frag, 0,
810 						    fsize, DMA_TO_DEVICE);
811 			if (dma_mapping_error(&nn->pdev->dev, dma_addr))
812 				goto err_unmap;
813 
814 			wr_idx = (wr_idx + 1) & (tx_ring->cnt - 1);
815 			tx_ring->txbufs[wr_idx].skb = skb;
816 			tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
817 			tx_ring->txbufs[wr_idx].fidx = f;
818 
819 			txd = &tx_ring->txds[wr_idx];
820 			*txd = txdg;
821 			txd->dma_len = cpu_to_le16(fsize);
822 			nfp_desc_set_dma_addr(txd, dma_addr);
823 			txd->offset_eop =
824 				(f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
825 		}
826 
827 		u64_stats_update_begin(&r_vec->tx_sync);
828 		r_vec->tx_gather++;
829 		u64_stats_update_end(&r_vec->tx_sync);
830 	}
831 
832 	netdev_tx_sent_queue(nd_q, txbuf->real_len);
833 
834 	tx_ring->wr_p += nr_frags + 1;
835 	if (nfp_net_tx_ring_should_stop(tx_ring))
836 		nfp_net_tx_ring_stop(nd_q, tx_ring);
837 
838 	tx_ring->wr_ptr_add += nr_frags + 1;
839 	if (!skb->xmit_more || netif_xmit_stopped(nd_q))
840 		nfp_net_tx_xmit_more_flush(tx_ring);
841 
842 	skb_tx_timestamp(skb);
843 
844 	return NETDEV_TX_OK;
845 
846 err_unmap:
847 	--f;
848 	while (f >= 0) {
849 		frag = &skb_shinfo(skb)->frags[f];
850 		dma_unmap_page(&nn->pdev->dev,
851 			       tx_ring->txbufs[wr_idx].dma_addr,
852 			       skb_frag_size(frag), DMA_TO_DEVICE);
853 		tx_ring->txbufs[wr_idx].skb = NULL;
854 		tx_ring->txbufs[wr_idx].dma_addr = 0;
855 		tx_ring->txbufs[wr_idx].fidx = -2;
856 		wr_idx = wr_idx - 1;
857 		if (wr_idx < 0)
858 			wr_idx += tx_ring->cnt;
859 	}
860 	dma_unmap_single(&nn->pdev->dev, tx_ring->txbufs[wr_idx].dma_addr,
861 			 skb_headlen(skb), DMA_TO_DEVICE);
862 	tx_ring->txbufs[wr_idx].skb = NULL;
863 	tx_ring->txbufs[wr_idx].dma_addr = 0;
864 	tx_ring->txbufs[wr_idx].fidx = -2;
865 err_free:
866 	nn_warn_ratelimit(nn, "Failed to map DMA TX buffer\n");
867 	u64_stats_update_begin(&r_vec->tx_sync);
868 	r_vec->tx_errors++;
869 	u64_stats_update_end(&r_vec->tx_sync);
870 	dev_kfree_skb_any(skb);
871 	return NETDEV_TX_OK;
872 }
873 
874 /**
875  * nfp_net_tx_complete() - Handled completed TX packets
876  * @tx_ring:   TX ring structure
877  *
878  * Return: Number of completed TX descriptors
879  */
880 static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring)
881 {
882 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
883 	struct nfp_net *nn = r_vec->nfp_net;
884 	const struct skb_frag_struct *frag;
885 	struct netdev_queue *nd_q;
886 	u32 done_pkts = 0, done_bytes = 0;
887 	struct sk_buff *skb;
888 	int todo, nr_frags;
889 	u32 qcp_rd_p;
890 	int fidx;
891 	int idx;
892 
893 	/* Work out how many descriptors have been transmitted */
894 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
895 
896 	if (qcp_rd_p == tx_ring->qcp_rd_p)
897 		return;
898 
899 	if (qcp_rd_p > tx_ring->qcp_rd_p)
900 		todo = qcp_rd_p - tx_ring->qcp_rd_p;
901 	else
902 		todo = qcp_rd_p + tx_ring->cnt - tx_ring->qcp_rd_p;
903 
904 	while (todo--) {
905 		idx = tx_ring->rd_p & (tx_ring->cnt - 1);
906 		tx_ring->rd_p++;
907 
908 		skb = tx_ring->txbufs[idx].skb;
909 		if (!skb)
910 			continue;
911 
912 		nr_frags = skb_shinfo(skb)->nr_frags;
913 		fidx = tx_ring->txbufs[idx].fidx;
914 
915 		if (fidx == -1) {
916 			/* unmap head */
917 			dma_unmap_single(&nn->pdev->dev,
918 					 tx_ring->txbufs[idx].dma_addr,
919 					 skb_headlen(skb), DMA_TO_DEVICE);
920 
921 			done_pkts += tx_ring->txbufs[idx].pkt_cnt;
922 			done_bytes += tx_ring->txbufs[idx].real_len;
923 		} else {
924 			/* unmap fragment */
925 			frag = &skb_shinfo(skb)->frags[fidx];
926 			dma_unmap_page(&nn->pdev->dev,
927 				       tx_ring->txbufs[idx].dma_addr,
928 				       skb_frag_size(frag), DMA_TO_DEVICE);
929 		}
930 
931 		/* check for last gather fragment */
932 		if (fidx == nr_frags - 1)
933 			dev_kfree_skb_any(skb);
934 
935 		tx_ring->txbufs[idx].dma_addr = 0;
936 		tx_ring->txbufs[idx].skb = NULL;
937 		tx_ring->txbufs[idx].fidx = -2;
938 	}
939 
940 	tx_ring->qcp_rd_p = qcp_rd_p;
941 
942 	u64_stats_update_begin(&r_vec->tx_sync);
943 	r_vec->tx_bytes += done_bytes;
944 	r_vec->tx_pkts += done_pkts;
945 	u64_stats_update_end(&r_vec->tx_sync);
946 
947 	nd_q = netdev_get_tx_queue(nn->netdev, tx_ring->idx);
948 	netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
949 	if (nfp_net_tx_ring_should_wake(tx_ring)) {
950 		/* Make sure TX thread will see updated tx_ring->rd_p */
951 		smp_mb();
952 
953 		if (unlikely(netif_tx_queue_stopped(nd_q)))
954 			netif_tx_wake_queue(nd_q);
955 	}
956 
957 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
958 		  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
959 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
960 }
961 
962 static void nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
963 {
964 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
965 	struct nfp_net *nn = r_vec->nfp_net;
966 	u32 done_pkts = 0, done_bytes = 0;
967 	int idx, todo;
968 	u32 qcp_rd_p;
969 
970 	/* Work out how many descriptors have been transmitted */
971 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
972 
973 	if (qcp_rd_p == tx_ring->qcp_rd_p)
974 		return;
975 
976 	if (qcp_rd_p > tx_ring->qcp_rd_p)
977 		todo = qcp_rd_p - tx_ring->qcp_rd_p;
978 	else
979 		todo = qcp_rd_p + tx_ring->cnt - tx_ring->qcp_rd_p;
980 
981 	while (todo--) {
982 		idx = tx_ring->rd_p & (tx_ring->cnt - 1);
983 		tx_ring->rd_p++;
984 
985 		if (!tx_ring->txbufs[idx].frag)
986 			continue;
987 
988 		nfp_net_dma_unmap_rx(nn, tx_ring->txbufs[idx].dma_addr,
989 				     nn->fl_bufsz, DMA_BIDIRECTIONAL);
990 		__free_page(virt_to_page(tx_ring->txbufs[idx].frag));
991 
992 		done_pkts++;
993 		done_bytes += tx_ring->txbufs[idx].real_len;
994 
995 		tx_ring->txbufs[idx].dma_addr = 0;
996 		tx_ring->txbufs[idx].frag = NULL;
997 		tx_ring->txbufs[idx].fidx = -2;
998 	}
999 
1000 	tx_ring->qcp_rd_p = qcp_rd_p;
1001 
1002 	u64_stats_update_begin(&r_vec->tx_sync);
1003 	r_vec->tx_bytes += done_bytes;
1004 	r_vec->tx_pkts += done_pkts;
1005 	u64_stats_update_end(&r_vec->tx_sync);
1006 
1007 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1008 		  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1009 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1010 }
1011 
1012 /**
1013  * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
1014  * @nn:		NFP Net device
1015  * @tx_ring:	TX ring structure
1016  *
1017  * Assumes that the device is stopped
1018  */
1019 static void
1020 nfp_net_tx_ring_reset(struct nfp_net *nn, struct nfp_net_tx_ring *tx_ring)
1021 {
1022 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1023 	const struct skb_frag_struct *frag;
1024 	struct pci_dev *pdev = nn->pdev;
1025 	struct netdev_queue *nd_q;
1026 
1027 	while (tx_ring->rd_p != tx_ring->wr_p) {
1028 		struct nfp_net_tx_buf *tx_buf;
1029 		int idx;
1030 
1031 		idx = tx_ring->rd_p & (tx_ring->cnt - 1);
1032 		tx_buf = &tx_ring->txbufs[idx];
1033 
1034 		if (tx_ring == r_vec->xdp_ring) {
1035 			nfp_net_dma_unmap_rx(nn, tx_buf->dma_addr,
1036 					     nn->fl_bufsz, DMA_BIDIRECTIONAL);
1037 			__free_page(virt_to_page(tx_ring->txbufs[idx].frag));
1038 		} else {
1039 			struct sk_buff *skb = tx_ring->txbufs[idx].skb;
1040 			int nr_frags = skb_shinfo(skb)->nr_frags;
1041 
1042 			if (tx_buf->fidx == -1) {
1043 				/* unmap head */
1044 				dma_unmap_single(&pdev->dev, tx_buf->dma_addr,
1045 						 skb_headlen(skb),
1046 						 DMA_TO_DEVICE);
1047 			} else {
1048 				/* unmap fragment */
1049 				frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1050 				dma_unmap_page(&pdev->dev, tx_buf->dma_addr,
1051 					       skb_frag_size(frag),
1052 					       DMA_TO_DEVICE);
1053 			}
1054 
1055 			/* check for last gather fragment */
1056 			if (tx_buf->fidx == nr_frags - 1)
1057 				dev_kfree_skb_any(skb);
1058 		}
1059 
1060 		tx_buf->dma_addr = 0;
1061 		tx_buf->skb = NULL;
1062 		tx_buf->fidx = -2;
1063 
1064 		tx_ring->qcp_rd_p++;
1065 		tx_ring->rd_p++;
1066 	}
1067 
1068 	memset(tx_ring->txds, 0, sizeof(*tx_ring->txds) * tx_ring->cnt);
1069 	tx_ring->wr_p = 0;
1070 	tx_ring->rd_p = 0;
1071 	tx_ring->qcp_rd_p = 0;
1072 	tx_ring->wr_ptr_add = 0;
1073 
1074 	if (tx_ring == r_vec->xdp_ring)
1075 		return;
1076 
1077 	nd_q = netdev_get_tx_queue(nn->netdev, tx_ring->idx);
1078 	netdev_tx_reset_queue(nd_q);
1079 }
1080 
1081 static void nfp_net_tx_timeout(struct net_device *netdev)
1082 {
1083 	struct nfp_net *nn = netdev_priv(netdev);
1084 	int i;
1085 
1086 	for (i = 0; i < nn->netdev->real_num_tx_queues; i++) {
1087 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i)))
1088 			continue;
1089 		nn_warn(nn, "TX timeout on ring: %d\n", i);
1090 	}
1091 	nn_warn(nn, "TX watchdog timeout\n");
1092 }
1093 
1094 /* Receive processing
1095  */
1096 static unsigned int
1097 nfp_net_calc_fl_bufsz(struct nfp_net *nn, unsigned int mtu)
1098 {
1099 	unsigned int fl_bufsz;
1100 
1101 	fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
1102 	if (nn->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1103 		fl_bufsz += NFP_NET_MAX_PREPEND;
1104 	else
1105 		fl_bufsz += nn->rx_offset;
1106 	fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + mtu;
1107 
1108 	fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1109 	fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1110 
1111 	return fl_bufsz;
1112 }
1113 
1114 static void
1115 nfp_net_free_frag(void *frag, bool xdp)
1116 {
1117 	if (!xdp)
1118 		skb_free_frag(frag);
1119 	else
1120 		__free_page(virt_to_page(frag));
1121 }
1122 
1123 /**
1124  * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
1125  * @rx_ring:	RX ring structure of the skb
1126  * @dma_addr:	Pointer to storage for DMA address (output param)
1127  * @fl_bufsz:	size of freelist buffers
1128  * @xdp:	Whether XDP is enabled
1129  *
1130  * This function will allcate a new page frag, map it for DMA.
1131  *
1132  * Return: allocated page frag or NULL on failure.
1133  */
1134 static void *
1135 nfp_net_rx_alloc_one(struct nfp_net_rx_ring *rx_ring, dma_addr_t *dma_addr,
1136 		     unsigned int fl_bufsz, bool xdp)
1137 {
1138 	struct nfp_net *nn = rx_ring->r_vec->nfp_net;
1139 	int direction;
1140 	void *frag;
1141 
1142 	if (!xdp)
1143 		frag = netdev_alloc_frag(fl_bufsz);
1144 	else
1145 		frag = page_address(alloc_page(GFP_KERNEL | __GFP_COLD));
1146 	if (!frag) {
1147 		nn_warn_ratelimit(nn, "Failed to alloc receive page frag\n");
1148 		return NULL;
1149 	}
1150 
1151 	direction = xdp ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
1152 
1153 	*dma_addr = nfp_net_dma_map_rx(nn, frag, fl_bufsz, direction);
1154 	if (dma_mapping_error(&nn->pdev->dev, *dma_addr)) {
1155 		nfp_net_free_frag(frag, xdp);
1156 		nn_warn_ratelimit(nn, "Failed to map DMA RX buffer\n");
1157 		return NULL;
1158 	}
1159 
1160 	return frag;
1161 }
1162 
1163 static void *
1164 nfp_net_napi_alloc_one(struct nfp_net *nn, int direction, dma_addr_t *dma_addr)
1165 {
1166 	void *frag;
1167 
1168 	if (!nn->xdp_prog)
1169 		frag = napi_alloc_frag(nn->fl_bufsz);
1170 	else
1171 		frag = page_address(alloc_page(GFP_ATOMIC | __GFP_COLD));
1172 	if (!frag) {
1173 		nn_warn_ratelimit(nn, "Failed to alloc receive page frag\n");
1174 		return NULL;
1175 	}
1176 
1177 	*dma_addr = nfp_net_dma_map_rx(nn, frag, nn->fl_bufsz, direction);
1178 	if (dma_mapping_error(&nn->pdev->dev, *dma_addr)) {
1179 		nfp_net_free_frag(frag, nn->xdp_prog);
1180 		nn_warn_ratelimit(nn, "Failed to map DMA RX buffer\n");
1181 		return NULL;
1182 	}
1183 
1184 	return frag;
1185 }
1186 
1187 /**
1188  * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1189  * @rx_ring:	RX ring structure
1190  * @frag:	page fragment buffer
1191  * @dma_addr:	DMA address of skb mapping
1192  */
1193 static void nfp_net_rx_give_one(struct nfp_net_rx_ring *rx_ring,
1194 				void *frag, dma_addr_t dma_addr)
1195 {
1196 	unsigned int wr_idx;
1197 
1198 	wr_idx = rx_ring->wr_p & (rx_ring->cnt - 1);
1199 
1200 	/* Stash SKB and DMA address away */
1201 	rx_ring->rxbufs[wr_idx].frag = frag;
1202 	rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1203 
1204 	/* Fill freelist descriptor */
1205 	rx_ring->rxds[wr_idx].fld.reserved = 0;
1206 	rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1207 	nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld, dma_addr);
1208 
1209 	rx_ring->wr_p++;
1210 	rx_ring->wr_ptr_add++;
1211 	if (rx_ring->wr_ptr_add >= NFP_NET_FL_BATCH) {
1212 		/* Update write pointer of the freelist queue. Make
1213 		 * sure all writes are flushed before telling the hardware.
1214 		 */
1215 		wmb();
1216 		nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, rx_ring->wr_ptr_add);
1217 		rx_ring->wr_ptr_add = 0;
1218 	}
1219 }
1220 
1221 /**
1222  * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1223  * @rx_ring:	RX ring structure
1224  *
1225  * Warning: Do *not* call if ring buffers were never put on the FW freelist
1226  *	    (i.e. device was not enabled)!
1227  */
1228 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1229 {
1230 	unsigned int wr_idx, last_idx;
1231 
1232 	/* Move the empty entry to the end of the list */
1233 	wr_idx = rx_ring->wr_p & (rx_ring->cnt - 1);
1234 	last_idx = rx_ring->cnt - 1;
1235 	rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1236 	rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1237 	rx_ring->rxbufs[last_idx].dma_addr = 0;
1238 	rx_ring->rxbufs[last_idx].frag = NULL;
1239 
1240 	memset(rx_ring->rxds, 0, sizeof(*rx_ring->rxds) * rx_ring->cnt);
1241 	rx_ring->wr_p = 0;
1242 	rx_ring->rd_p = 0;
1243 	rx_ring->wr_ptr_add = 0;
1244 }
1245 
1246 /**
1247  * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1248  * @nn:		NFP Net device
1249  * @rx_ring:	RX ring to remove buffers from
1250  * @xdp:	Whether XDP is enabled
1251  *
1252  * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1253  * entries.  After device is disabled nfp_net_rx_ring_reset() must be called
1254  * to restore required ring geometry.
1255  */
1256 static void
1257 nfp_net_rx_ring_bufs_free(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring,
1258 			  bool xdp)
1259 {
1260 	int direction = xdp ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
1261 	unsigned int i;
1262 
1263 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1264 		/* NULL skb can only happen when initial filling of the ring
1265 		 * fails to allocate enough buffers and calls here to free
1266 		 * already allocated ones.
1267 		 */
1268 		if (!rx_ring->rxbufs[i].frag)
1269 			continue;
1270 
1271 		nfp_net_dma_unmap_rx(nn, rx_ring->rxbufs[i].dma_addr,
1272 				     rx_ring->bufsz, direction);
1273 		nfp_net_free_frag(rx_ring->rxbufs[i].frag, xdp);
1274 		rx_ring->rxbufs[i].dma_addr = 0;
1275 		rx_ring->rxbufs[i].frag = NULL;
1276 	}
1277 }
1278 
1279 /**
1280  * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1281  * @nn:		NFP Net device
1282  * @rx_ring:	RX ring to remove buffers from
1283  * @xdp:	Whether XDP is enabled
1284  */
1285 static int
1286 nfp_net_rx_ring_bufs_alloc(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring,
1287 			   bool xdp)
1288 {
1289 	struct nfp_net_rx_buf *rxbufs;
1290 	unsigned int i;
1291 
1292 	rxbufs = rx_ring->rxbufs;
1293 
1294 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1295 		rxbufs[i].frag =
1296 			nfp_net_rx_alloc_one(rx_ring, &rxbufs[i].dma_addr,
1297 					     rx_ring->bufsz, xdp);
1298 		if (!rxbufs[i].frag) {
1299 			nfp_net_rx_ring_bufs_free(nn, rx_ring, xdp);
1300 			return -ENOMEM;
1301 		}
1302 	}
1303 
1304 	return 0;
1305 }
1306 
1307 /**
1308  * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1309  * @rx_ring: RX ring to fill
1310  */
1311 static void nfp_net_rx_ring_fill_freelist(struct nfp_net_rx_ring *rx_ring)
1312 {
1313 	unsigned int i;
1314 
1315 	for (i = 0; i < rx_ring->cnt - 1; i++)
1316 		nfp_net_rx_give_one(rx_ring, rx_ring->rxbufs[i].frag,
1317 				    rx_ring->rxbufs[i].dma_addr);
1318 }
1319 
1320 /**
1321  * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1322  * @flags: RX descriptor flags field in CPU byte order
1323  */
1324 static int nfp_net_rx_csum_has_errors(u16 flags)
1325 {
1326 	u16 csum_all_checked, csum_all_ok;
1327 
1328 	csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1329 	csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1330 
1331 	return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1332 }
1333 
1334 /**
1335  * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1336  * @nn:  NFP Net device
1337  * @r_vec: per-ring structure
1338  * @rxd: Pointer to RX descriptor
1339  * @skb: Pointer to SKB
1340  */
1341 static void nfp_net_rx_csum(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1342 			    struct nfp_net_rx_desc *rxd, struct sk_buff *skb)
1343 {
1344 	skb_checksum_none_assert(skb);
1345 
1346 	if (!(nn->netdev->features & NETIF_F_RXCSUM))
1347 		return;
1348 
1349 	if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1350 		u64_stats_update_begin(&r_vec->rx_sync);
1351 		r_vec->hw_csum_rx_error++;
1352 		u64_stats_update_end(&r_vec->rx_sync);
1353 		return;
1354 	}
1355 
1356 	/* Assume that the firmware will never report inner CSUM_OK unless outer
1357 	 * L4 headers were successfully parsed. FW will always report zero UDP
1358 	 * checksum as CSUM_OK.
1359 	 */
1360 	if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1361 	    rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1362 		__skb_incr_checksum_unnecessary(skb);
1363 		u64_stats_update_begin(&r_vec->rx_sync);
1364 		r_vec->hw_csum_rx_ok++;
1365 		u64_stats_update_end(&r_vec->rx_sync);
1366 	}
1367 
1368 	if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1369 	    rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1370 		__skb_incr_checksum_unnecessary(skb);
1371 		u64_stats_update_begin(&r_vec->rx_sync);
1372 		r_vec->hw_csum_rx_inner_ok++;
1373 		u64_stats_update_end(&r_vec->rx_sync);
1374 	}
1375 }
1376 
1377 static void nfp_net_set_hash(struct net_device *netdev, struct sk_buff *skb,
1378 			     unsigned int type, __be32 *hash)
1379 {
1380 	if (!(netdev->features & NETIF_F_RXHASH))
1381 		return;
1382 
1383 	switch (type) {
1384 	case NFP_NET_RSS_IPV4:
1385 	case NFP_NET_RSS_IPV6:
1386 	case NFP_NET_RSS_IPV6_EX:
1387 		skb_set_hash(skb, get_unaligned_be32(hash), PKT_HASH_TYPE_L3);
1388 		break;
1389 	default:
1390 		skb_set_hash(skb, get_unaligned_be32(hash), PKT_HASH_TYPE_L4);
1391 		break;
1392 	}
1393 }
1394 
1395 static void
1396 nfp_net_set_hash_desc(struct net_device *netdev, struct sk_buff *skb,
1397 		      struct nfp_net_rx_desc *rxd)
1398 {
1399 	struct nfp_net_rx_hash *rx_hash;
1400 
1401 	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1402 		return;
1403 
1404 	rx_hash = (struct nfp_net_rx_hash *)(skb->data - sizeof(*rx_hash));
1405 
1406 	nfp_net_set_hash(netdev, skb, get_unaligned_be32(&rx_hash->hash_type),
1407 			 &rx_hash->hash);
1408 }
1409 
1410 static void *
1411 nfp_net_parse_meta(struct net_device *netdev, struct sk_buff *skb,
1412 		   int meta_len)
1413 {
1414 	u8 *data = skb->data - meta_len;
1415 	u32 meta_info;
1416 
1417 	meta_info = get_unaligned_be32(data);
1418 	data += 4;
1419 
1420 	while (meta_info) {
1421 		switch (meta_info & NFP_NET_META_FIELD_MASK) {
1422 		case NFP_NET_META_HASH:
1423 			meta_info >>= NFP_NET_META_FIELD_SIZE;
1424 			nfp_net_set_hash(netdev, skb,
1425 					 meta_info & NFP_NET_META_FIELD_MASK,
1426 					 (__be32 *)data);
1427 			data += 4;
1428 			break;
1429 		case NFP_NET_META_MARK:
1430 			skb->mark = get_unaligned_be32(data);
1431 			data += 4;
1432 			break;
1433 		default:
1434 			return NULL;
1435 		}
1436 
1437 		meta_info >>= NFP_NET_META_FIELD_SIZE;
1438 	}
1439 
1440 	return data;
1441 }
1442 
1443 static void
1444 nfp_net_rx_drop(struct nfp_net_r_vector *r_vec, struct nfp_net_rx_ring *rx_ring,
1445 		struct nfp_net_rx_buf *rxbuf, struct sk_buff *skb)
1446 {
1447 	u64_stats_update_begin(&r_vec->rx_sync);
1448 	r_vec->rx_drops++;
1449 	u64_stats_update_end(&r_vec->rx_sync);
1450 
1451 	/* skb is build based on the frag, free_skb() would free the frag
1452 	 * so to be able to reuse it we need an extra ref.
1453 	 */
1454 	if (skb && rxbuf && skb->head == rxbuf->frag)
1455 		page_ref_inc(virt_to_head_page(rxbuf->frag));
1456 	if (rxbuf)
1457 		nfp_net_rx_give_one(rx_ring, rxbuf->frag, rxbuf->dma_addr);
1458 	if (skb)
1459 		dev_kfree_skb_any(skb);
1460 }
1461 
1462 static void
1463 nfp_net_tx_xdp_buf(struct nfp_net *nn, struct nfp_net_rx_ring *rx_ring,
1464 		   struct nfp_net_tx_ring *tx_ring,
1465 		   struct nfp_net_rx_buf *rxbuf, unsigned int pkt_off,
1466 		   unsigned int pkt_len)
1467 {
1468 	struct nfp_net_tx_buf *txbuf;
1469 	struct nfp_net_tx_desc *txd;
1470 	dma_addr_t new_dma_addr;
1471 	void *new_frag;
1472 	int wr_idx;
1473 
1474 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1475 		nfp_net_rx_drop(rx_ring->r_vec, rx_ring, rxbuf, NULL);
1476 		return;
1477 	}
1478 
1479 	new_frag = nfp_net_napi_alloc_one(nn, DMA_BIDIRECTIONAL, &new_dma_addr);
1480 	if (unlikely(!new_frag)) {
1481 		nfp_net_rx_drop(rx_ring->r_vec, rx_ring, rxbuf, NULL);
1482 		return;
1483 	}
1484 	nfp_net_rx_give_one(rx_ring, new_frag, new_dma_addr);
1485 
1486 	wr_idx = tx_ring->wr_p & (tx_ring->cnt - 1);
1487 
1488 	/* Stash the soft descriptor of the head then initialize it */
1489 	txbuf = &tx_ring->txbufs[wr_idx];
1490 	txbuf->frag = rxbuf->frag;
1491 	txbuf->dma_addr = rxbuf->dma_addr;
1492 	txbuf->fidx = -1;
1493 	txbuf->pkt_cnt = 1;
1494 	txbuf->real_len = pkt_len;
1495 
1496 	dma_sync_single_for_device(&nn->pdev->dev, rxbuf->dma_addr + pkt_off,
1497 				   pkt_len, DMA_TO_DEVICE);
1498 
1499 	/* Build TX descriptor */
1500 	txd = &tx_ring->txds[wr_idx];
1501 	txd->offset_eop = PCIE_DESC_TX_EOP;
1502 	txd->dma_len = cpu_to_le16(pkt_len);
1503 	nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + pkt_off);
1504 	txd->data_len = cpu_to_le16(pkt_len);
1505 
1506 	txd->flags = 0;
1507 	txd->mss = 0;
1508 	txd->l4_offset = 0;
1509 
1510 	tx_ring->wr_p++;
1511 	tx_ring->wr_ptr_add++;
1512 }
1513 
1514 static int nfp_net_run_xdp(struct bpf_prog *prog, void *data, unsigned int len)
1515 {
1516 	struct xdp_buff xdp;
1517 
1518 	xdp.data = data;
1519 	xdp.data_end = data + len;
1520 
1521 	return bpf_prog_run_xdp(prog, &xdp);
1522 }
1523 
1524 /**
1525  * nfp_net_rx() - receive up to @budget packets on @rx_ring
1526  * @rx_ring:   RX ring to receive from
1527  * @budget:    NAPI budget
1528  *
1529  * Note, this function is separated out from the napi poll function to
1530  * more cleanly separate packet receive code from other bookkeeping
1531  * functions performed in the napi poll function.
1532  *
1533  * Return: Number of packets received.
1534  */
1535 static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1536 {
1537 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1538 	struct nfp_net *nn = r_vec->nfp_net;
1539 	struct nfp_net_tx_ring *tx_ring;
1540 	struct bpf_prog *xdp_prog;
1541 	unsigned int true_bufsz;
1542 	struct sk_buff *skb;
1543 	int pkts_polled = 0;
1544 	int rx_dma_map_dir;
1545 	int idx;
1546 
1547 	rcu_read_lock();
1548 	xdp_prog = READ_ONCE(nn->xdp_prog);
1549 	rx_dma_map_dir = xdp_prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
1550 	true_bufsz = xdp_prog ? PAGE_SIZE : nn->fl_bufsz;
1551 	tx_ring = r_vec->xdp_ring;
1552 
1553 	while (pkts_polled < budget) {
1554 		unsigned int meta_len, data_len, data_off, pkt_len, pkt_off;
1555 		struct nfp_net_rx_buf *rxbuf;
1556 		struct nfp_net_rx_desc *rxd;
1557 		dma_addr_t new_dma_addr;
1558 		void *new_frag;
1559 
1560 		idx = rx_ring->rd_p & (rx_ring->cnt - 1);
1561 
1562 		rxd = &rx_ring->rxds[idx];
1563 		if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1564 			break;
1565 
1566 		/* Memory barrier to ensure that we won't do other reads
1567 		 * before the DD bit.
1568 		 */
1569 		dma_rmb();
1570 
1571 		rx_ring->rd_p++;
1572 		pkts_polled++;
1573 
1574 		rxbuf =	&rx_ring->rxbufs[idx];
1575 		/*         < meta_len >
1576 		 *  <-- [rx_offset] -->
1577 		 *  ---------------------------------------------------------
1578 		 * | [XX] |  metadata  |             packet           | XXXX |
1579 		 *  ---------------------------------------------------------
1580 		 *         <---------------- data_len --------------->
1581 		 *
1582 		 * The rx_offset is fixed for all packets, the meta_len can vary
1583 		 * on a packet by packet basis. If rx_offset is set to zero
1584 		 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1585 		 * buffer and is immediately followed by the packet (no [XX]).
1586 		 */
1587 		meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1588 		data_len = le16_to_cpu(rxd->rxd.data_len);
1589 		pkt_len = data_len - meta_len;
1590 
1591 		if (nn->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1592 			pkt_off = meta_len;
1593 		else
1594 			pkt_off = nn->rx_offset;
1595 		data_off = NFP_NET_RX_BUF_HEADROOM + pkt_off;
1596 
1597 		/* Stats update */
1598 		u64_stats_update_begin(&r_vec->rx_sync);
1599 		r_vec->rx_pkts++;
1600 		r_vec->rx_bytes += pkt_len;
1601 		u64_stats_update_end(&r_vec->rx_sync);
1602 
1603 		if (xdp_prog && !(rxd->rxd.flags & PCIE_DESC_RX_BPF &&
1604 				  nn->bpf_offload_xdp)) {
1605 			int act;
1606 
1607 			dma_sync_single_for_cpu(&nn->pdev->dev,
1608 						rxbuf->dma_addr + pkt_off,
1609 						pkt_len, DMA_FROM_DEVICE);
1610 			act = nfp_net_run_xdp(xdp_prog, rxbuf->frag + data_off,
1611 					      pkt_len);
1612 			switch (act) {
1613 			case XDP_PASS:
1614 				break;
1615 			case XDP_TX:
1616 				nfp_net_tx_xdp_buf(nn, rx_ring, tx_ring, rxbuf,
1617 						   pkt_off, pkt_len);
1618 				continue;
1619 			default:
1620 				bpf_warn_invalid_xdp_action(act);
1621 			case XDP_ABORTED:
1622 			case XDP_DROP:
1623 				nfp_net_rx_give_one(rx_ring, rxbuf->frag,
1624 						    rxbuf->dma_addr);
1625 				continue;
1626 			}
1627 		}
1628 
1629 		skb = build_skb(rxbuf->frag, true_bufsz);
1630 		if (unlikely(!skb)) {
1631 			nfp_net_rx_drop(r_vec, rx_ring, rxbuf, NULL);
1632 			continue;
1633 		}
1634 		new_frag = nfp_net_napi_alloc_one(nn, rx_dma_map_dir,
1635 						  &new_dma_addr);
1636 		if (unlikely(!new_frag)) {
1637 			nfp_net_rx_drop(r_vec, rx_ring, rxbuf, skb);
1638 			continue;
1639 		}
1640 
1641 		nfp_net_dma_unmap_rx(nn, rxbuf->dma_addr, nn->fl_bufsz,
1642 				     rx_dma_map_dir);
1643 
1644 		nfp_net_rx_give_one(rx_ring, new_frag, new_dma_addr);
1645 
1646 		skb_reserve(skb, data_off);
1647 		skb_put(skb, pkt_len);
1648 
1649 		if (nn->fw_ver.major <= 3) {
1650 			nfp_net_set_hash_desc(nn->netdev, skb, rxd);
1651 		} else if (meta_len) {
1652 			void *end;
1653 
1654 			end = nfp_net_parse_meta(nn->netdev, skb, meta_len);
1655 			if (unlikely(end != skb->data)) {
1656 				nn_warn_ratelimit(nn, "invalid RX packet metadata\n");
1657 				nfp_net_rx_drop(r_vec, rx_ring, NULL, skb);
1658 				continue;
1659 			}
1660 		}
1661 
1662 		skb_record_rx_queue(skb, rx_ring->idx);
1663 		skb->protocol = eth_type_trans(skb, nn->netdev);
1664 
1665 		nfp_net_rx_csum(nn, r_vec, rxd, skb);
1666 
1667 		if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
1668 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1669 					       le16_to_cpu(rxd->rxd.vlan));
1670 
1671 		napi_gro_receive(&rx_ring->r_vec->napi, skb);
1672 	}
1673 
1674 	if (xdp_prog && tx_ring->wr_ptr_add)
1675 		nfp_net_tx_xmit_more_flush(tx_ring);
1676 	rcu_read_unlock();
1677 
1678 	return pkts_polled;
1679 }
1680 
1681 /**
1682  * nfp_net_poll() - napi poll function
1683  * @napi:    NAPI structure
1684  * @budget:  NAPI budget
1685  *
1686  * Return: number of packets polled.
1687  */
1688 static int nfp_net_poll(struct napi_struct *napi, int budget)
1689 {
1690 	struct nfp_net_r_vector *r_vec =
1691 		container_of(napi, struct nfp_net_r_vector, napi);
1692 	unsigned int pkts_polled = 0;
1693 
1694 	if (r_vec->tx_ring)
1695 		nfp_net_tx_complete(r_vec->tx_ring);
1696 	if (r_vec->rx_ring) {
1697 		pkts_polled = nfp_net_rx(r_vec->rx_ring, budget);
1698 		if (r_vec->xdp_ring)
1699 			nfp_net_xdp_complete(r_vec->xdp_ring);
1700 	}
1701 
1702 	if (pkts_polled < budget) {
1703 		napi_complete_done(napi, pkts_polled);
1704 		nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_idx);
1705 	}
1706 
1707 	return pkts_polled;
1708 }
1709 
1710 /* Setup and Configuration
1711  */
1712 
1713 /**
1714  * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
1715  * @tx_ring:   TX ring to free
1716  */
1717 static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
1718 {
1719 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1720 	struct nfp_net *nn = r_vec->nfp_net;
1721 	struct pci_dev *pdev = nn->pdev;
1722 
1723 	kfree(tx_ring->txbufs);
1724 
1725 	if (tx_ring->txds)
1726 		dma_free_coherent(&pdev->dev, tx_ring->size,
1727 				  tx_ring->txds, tx_ring->dma);
1728 
1729 	tx_ring->cnt = 0;
1730 	tx_ring->txbufs = NULL;
1731 	tx_ring->txds = NULL;
1732 	tx_ring->dma = 0;
1733 	tx_ring->size = 0;
1734 }
1735 
1736 /**
1737  * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
1738  * @tx_ring:   TX Ring structure to allocate
1739  * @cnt:       Ring buffer count
1740  * @is_xdp:    True if ring will be used for XDP
1741  *
1742  * Return: 0 on success, negative errno otherwise.
1743  */
1744 static int
1745 nfp_net_tx_ring_alloc(struct nfp_net_tx_ring *tx_ring, u32 cnt, bool is_xdp)
1746 {
1747 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1748 	struct nfp_net *nn = r_vec->nfp_net;
1749 	struct pci_dev *pdev = nn->pdev;
1750 	int sz;
1751 
1752 	tx_ring->cnt = cnt;
1753 
1754 	tx_ring->size = sizeof(*tx_ring->txds) * tx_ring->cnt;
1755 	tx_ring->txds = dma_zalloc_coherent(&pdev->dev, tx_ring->size,
1756 					    &tx_ring->dma, GFP_KERNEL);
1757 	if (!tx_ring->txds)
1758 		goto err_alloc;
1759 
1760 	sz = sizeof(*tx_ring->txbufs) * tx_ring->cnt;
1761 	tx_ring->txbufs = kzalloc(sz, GFP_KERNEL);
1762 	if (!tx_ring->txbufs)
1763 		goto err_alloc;
1764 
1765 	if (!is_xdp)
1766 		netif_set_xps_queue(nn->netdev, &r_vec->affinity_mask,
1767 				    tx_ring->idx);
1768 
1769 	nn_dbg(nn, "TxQ%02d: QCidx=%02d cnt=%d dma=%#llx host=%p %s\n",
1770 	       tx_ring->idx, tx_ring->qcidx,
1771 	       tx_ring->cnt, (unsigned long long)tx_ring->dma, tx_ring->txds,
1772 	       is_xdp ? "XDP" : "");
1773 
1774 	return 0;
1775 
1776 err_alloc:
1777 	nfp_net_tx_ring_free(tx_ring);
1778 	return -ENOMEM;
1779 }
1780 
1781 static struct nfp_net_tx_ring *
1782 nfp_net_tx_ring_set_prepare(struct nfp_net *nn, struct nfp_net_ring_set *s,
1783 			    unsigned int num_stack_tx_rings)
1784 {
1785 	struct nfp_net_tx_ring *rings;
1786 	unsigned int r;
1787 
1788 	rings = kcalloc(s->n_rings, sizeof(*rings), GFP_KERNEL);
1789 	if (!rings)
1790 		return NULL;
1791 
1792 	for (r = 0; r < s->n_rings; r++) {
1793 		int bias = 0;
1794 
1795 		if (r >= num_stack_tx_rings)
1796 			bias = num_stack_tx_rings;
1797 
1798 		nfp_net_tx_ring_init(&rings[r], &nn->r_vecs[r - bias], r);
1799 
1800 		if (nfp_net_tx_ring_alloc(&rings[r], s->dcnt, bias))
1801 			goto err_free_prev;
1802 	}
1803 
1804 	return s->rings = rings;
1805 
1806 err_free_prev:
1807 	while (r--)
1808 		nfp_net_tx_ring_free(&rings[r]);
1809 	kfree(rings);
1810 	return NULL;
1811 }
1812 
1813 static void
1814 nfp_net_tx_ring_set_swap(struct nfp_net *nn, struct nfp_net_ring_set *s)
1815 {
1816 	struct nfp_net_ring_set new = *s;
1817 
1818 	s->dcnt = nn->txd_cnt;
1819 	s->rings = nn->tx_rings;
1820 	s->n_rings = nn->num_tx_rings;
1821 
1822 	nn->txd_cnt = new.dcnt;
1823 	nn->tx_rings = new.rings;
1824 	nn->num_tx_rings = new.n_rings;
1825 }
1826 
1827 static void
1828 nfp_net_tx_ring_set_free(struct nfp_net *nn, struct nfp_net_ring_set *s)
1829 {
1830 	struct nfp_net_tx_ring *rings = s->rings;
1831 	unsigned int r;
1832 
1833 	for (r = 0; r < s->n_rings; r++)
1834 		nfp_net_tx_ring_free(&rings[r]);
1835 
1836 	kfree(rings);
1837 }
1838 
1839 /**
1840  * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
1841  * @rx_ring:  RX ring to free
1842  */
1843 static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
1844 {
1845 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1846 	struct nfp_net *nn = r_vec->nfp_net;
1847 	struct pci_dev *pdev = nn->pdev;
1848 
1849 	kfree(rx_ring->rxbufs);
1850 
1851 	if (rx_ring->rxds)
1852 		dma_free_coherent(&pdev->dev, rx_ring->size,
1853 				  rx_ring->rxds, rx_ring->dma);
1854 
1855 	rx_ring->cnt = 0;
1856 	rx_ring->rxbufs = NULL;
1857 	rx_ring->rxds = NULL;
1858 	rx_ring->dma = 0;
1859 	rx_ring->size = 0;
1860 }
1861 
1862 /**
1863  * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
1864  * @rx_ring:  RX ring to allocate
1865  * @fl_bufsz: Size of buffers to allocate
1866  * @cnt:      Ring buffer count
1867  *
1868  * Return: 0 on success, negative errno otherwise.
1869  */
1870 static int
1871 nfp_net_rx_ring_alloc(struct nfp_net_rx_ring *rx_ring, unsigned int fl_bufsz,
1872 		      u32 cnt)
1873 {
1874 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1875 	struct nfp_net *nn = r_vec->nfp_net;
1876 	struct pci_dev *pdev = nn->pdev;
1877 	int sz;
1878 
1879 	rx_ring->cnt = cnt;
1880 	rx_ring->bufsz = fl_bufsz;
1881 
1882 	rx_ring->size = sizeof(*rx_ring->rxds) * rx_ring->cnt;
1883 	rx_ring->rxds = dma_zalloc_coherent(&pdev->dev, rx_ring->size,
1884 					    &rx_ring->dma, GFP_KERNEL);
1885 	if (!rx_ring->rxds)
1886 		goto err_alloc;
1887 
1888 	sz = sizeof(*rx_ring->rxbufs) * rx_ring->cnt;
1889 	rx_ring->rxbufs = kzalloc(sz, GFP_KERNEL);
1890 	if (!rx_ring->rxbufs)
1891 		goto err_alloc;
1892 
1893 	nn_dbg(nn, "RxQ%02d: FlQCidx=%02d RxQCidx=%02d cnt=%d dma=%#llx host=%p\n",
1894 	       rx_ring->idx, rx_ring->fl_qcidx, rx_ring->rx_qcidx,
1895 	       rx_ring->cnt, (unsigned long long)rx_ring->dma, rx_ring->rxds);
1896 
1897 	return 0;
1898 
1899 err_alloc:
1900 	nfp_net_rx_ring_free(rx_ring);
1901 	return -ENOMEM;
1902 }
1903 
1904 static struct nfp_net_rx_ring *
1905 nfp_net_rx_ring_set_prepare(struct nfp_net *nn, struct nfp_net_ring_set *s,
1906 			    bool xdp)
1907 {
1908 	unsigned int fl_bufsz =	nfp_net_calc_fl_bufsz(nn, s->mtu);
1909 	struct nfp_net_rx_ring *rings;
1910 	unsigned int r;
1911 
1912 	rings = kcalloc(s->n_rings, sizeof(*rings), GFP_KERNEL);
1913 	if (!rings)
1914 		return NULL;
1915 
1916 	for (r = 0; r < s->n_rings; r++) {
1917 		nfp_net_rx_ring_init(&rings[r], &nn->r_vecs[r], r);
1918 
1919 		if (nfp_net_rx_ring_alloc(&rings[r], fl_bufsz, s->dcnt))
1920 			goto err_free_prev;
1921 
1922 		if (nfp_net_rx_ring_bufs_alloc(nn, &rings[r], xdp))
1923 			goto err_free_ring;
1924 	}
1925 
1926 	return s->rings = rings;
1927 
1928 err_free_prev:
1929 	while (r--) {
1930 		nfp_net_rx_ring_bufs_free(nn, &rings[r], xdp);
1931 err_free_ring:
1932 		nfp_net_rx_ring_free(&rings[r]);
1933 	}
1934 	kfree(rings);
1935 	return NULL;
1936 }
1937 
1938 static void
1939 nfp_net_rx_ring_set_swap(struct nfp_net *nn, struct nfp_net_ring_set *s)
1940 {
1941 	struct nfp_net_ring_set new = *s;
1942 
1943 	s->mtu = nn->netdev->mtu;
1944 	s->dcnt = nn->rxd_cnt;
1945 	s->rings = nn->rx_rings;
1946 	s->n_rings = nn->num_rx_rings;
1947 
1948 	nn->netdev->mtu = new.mtu;
1949 	nn->fl_bufsz = nfp_net_calc_fl_bufsz(nn, new.mtu);
1950 	nn->rxd_cnt = new.dcnt;
1951 	nn->rx_rings = new.rings;
1952 	nn->num_rx_rings = new.n_rings;
1953 }
1954 
1955 static void
1956 nfp_net_rx_ring_set_free(struct nfp_net *nn, struct nfp_net_ring_set *s,
1957 			 bool xdp)
1958 {
1959 	struct nfp_net_rx_ring *rings = s->rings;
1960 	unsigned int r;
1961 
1962 	for (r = 0; r < s->n_rings; r++) {
1963 		nfp_net_rx_ring_bufs_free(nn, &rings[r], xdp);
1964 		nfp_net_rx_ring_free(&rings[r]);
1965 	}
1966 
1967 	kfree(rings);
1968 }
1969 
1970 static void
1971 nfp_net_vector_assign_rings(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1972 			    int idx)
1973 {
1974 	r_vec->rx_ring = idx < nn->num_rx_rings ? &nn->rx_rings[idx] : NULL;
1975 	r_vec->tx_ring =
1976 		idx < nn->num_stack_tx_rings ? &nn->tx_rings[idx] : NULL;
1977 
1978 	r_vec->xdp_ring = idx < nn->num_tx_rings - nn->num_stack_tx_rings ?
1979 		&nn->tx_rings[nn->num_stack_tx_rings + idx] : NULL;
1980 }
1981 
1982 static int
1983 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1984 		       int idx)
1985 {
1986 	struct msix_entry *entry = &nn->irq_entries[r_vec->irq_idx];
1987 	int err;
1988 
1989 	/* Setup NAPI */
1990 	netif_napi_add(nn->netdev, &r_vec->napi,
1991 		       nfp_net_poll, NAPI_POLL_WEIGHT);
1992 
1993 	snprintf(r_vec->name, sizeof(r_vec->name),
1994 		 "%s-rxtx-%d", nn->netdev->name, idx);
1995 	err = request_irq(entry->vector, r_vec->handler, 0, r_vec->name, r_vec);
1996 	if (err) {
1997 		netif_napi_del(&r_vec->napi);
1998 		nn_err(nn, "Error requesting IRQ %d\n", entry->vector);
1999 		return err;
2000 	}
2001 	disable_irq(entry->vector);
2002 
2003 	irq_set_affinity_hint(entry->vector, &r_vec->affinity_mask);
2004 
2005 	nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, entry->vector, entry->entry);
2006 
2007 	return 0;
2008 }
2009 
2010 static void
2011 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
2012 {
2013 	struct msix_entry *entry = &nn->irq_entries[r_vec->irq_idx];
2014 
2015 	irq_set_affinity_hint(entry->vector, NULL);
2016 	netif_napi_del(&r_vec->napi);
2017 	free_irq(entry->vector, r_vec);
2018 }
2019 
2020 /**
2021  * nfp_net_rss_write_itbl() - Write RSS indirection table to device
2022  * @nn:      NFP Net device to reconfigure
2023  */
2024 void nfp_net_rss_write_itbl(struct nfp_net *nn)
2025 {
2026 	int i;
2027 
2028 	for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
2029 		nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
2030 			  get_unaligned_le32(nn->rss_itbl + i));
2031 }
2032 
2033 /**
2034  * nfp_net_rss_write_key() - Write RSS hash key to device
2035  * @nn:      NFP Net device to reconfigure
2036  */
2037 void nfp_net_rss_write_key(struct nfp_net *nn)
2038 {
2039 	int i;
2040 
2041 	for (i = 0; i < NFP_NET_CFG_RSS_KEY_SZ; i += 4)
2042 		nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
2043 			  get_unaligned_le32(nn->rss_key + i));
2044 }
2045 
2046 /**
2047  * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
2048  * @nn:      NFP Net device to reconfigure
2049  */
2050 void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
2051 {
2052 	u8 i;
2053 	u32 factor;
2054 	u32 value;
2055 
2056 	/* Compute factor used to convert coalesce '_usecs' parameters to
2057 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
2058 	 * count.
2059 	 */
2060 	factor = nn->me_freq_mhz / 16;
2061 
2062 	/* copy RX interrupt coalesce parameters */
2063 	value = (nn->rx_coalesce_max_frames << 16) |
2064 		(factor * nn->rx_coalesce_usecs);
2065 	for (i = 0; i < nn->num_rx_rings; i++)
2066 		nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
2067 
2068 	/* copy TX interrupt coalesce parameters */
2069 	value = (nn->tx_coalesce_max_frames << 16) |
2070 		(factor * nn->tx_coalesce_usecs);
2071 	for (i = 0; i < nn->num_tx_rings; i++)
2072 		nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
2073 }
2074 
2075 /**
2076  * nfp_net_write_mac_addr() - Write mac address to the device control BAR
2077  * @nn:      NFP Net device to reconfigure
2078  *
2079  * Writes the MAC address from the netdev to the device control BAR.  Does not
2080  * perform the required reconfig.  We do a bit of byte swapping dance because
2081  * firmware is LE.
2082  */
2083 static void nfp_net_write_mac_addr(struct nfp_net *nn)
2084 {
2085 	nn_writel(nn, NFP_NET_CFG_MACADDR + 0,
2086 		  get_unaligned_be32(nn->netdev->dev_addr));
2087 	nn_writew(nn, NFP_NET_CFG_MACADDR + 6,
2088 		  get_unaligned_be16(nn->netdev->dev_addr + 4));
2089 }
2090 
2091 static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
2092 {
2093 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
2094 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
2095 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
2096 
2097 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
2098 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
2099 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
2100 }
2101 
2102 /**
2103  * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
2104  * @nn:      NFP Net device to reconfigure
2105  */
2106 static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
2107 {
2108 	u32 new_ctrl, update;
2109 	unsigned int r;
2110 	int err;
2111 
2112 	new_ctrl = nn->ctrl;
2113 	new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
2114 	update = NFP_NET_CFG_UPDATE_GEN;
2115 	update |= NFP_NET_CFG_UPDATE_MSIX;
2116 	update |= NFP_NET_CFG_UPDATE_RING;
2117 
2118 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2119 		new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
2120 
2121 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2122 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2123 
2124 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2125 	err = nfp_net_reconfig(nn, update);
2126 	if (err)
2127 		nn_err(nn, "Could not disable device: %d\n", err);
2128 
2129 	for (r = 0; r < nn->num_rx_rings; r++)
2130 		nfp_net_rx_ring_reset(&nn->rx_rings[r]);
2131 	for (r = 0; r < nn->num_tx_rings; r++)
2132 		nfp_net_tx_ring_reset(nn, &nn->tx_rings[r]);
2133 	for (r = 0; r < nn->num_r_vecs; r++)
2134 		nfp_net_vec_clear_ring_data(nn, r);
2135 
2136 	nn->ctrl = new_ctrl;
2137 }
2138 
2139 static void
2140 nfp_net_rx_ring_hw_cfg_write(struct nfp_net *nn,
2141 			     struct nfp_net_rx_ring *rx_ring, unsigned int idx)
2142 {
2143 	/* Write the DMA address, size and MSI-X info to the device */
2144 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), rx_ring->dma);
2145 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(rx_ring->cnt));
2146 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), rx_ring->r_vec->irq_idx);
2147 }
2148 
2149 static void
2150 nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
2151 			     struct nfp_net_tx_ring *tx_ring, unsigned int idx)
2152 {
2153 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), tx_ring->dma);
2154 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(tx_ring->cnt));
2155 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_idx);
2156 }
2157 
2158 static int __nfp_net_set_config_and_enable(struct nfp_net *nn)
2159 {
2160 	u32 new_ctrl, update = 0;
2161 	unsigned int r;
2162 	int err;
2163 
2164 	new_ctrl = nn->ctrl;
2165 
2166 	if (nn->cap & NFP_NET_CFG_CTRL_RSS) {
2167 		nfp_net_rss_write_key(nn);
2168 		nfp_net_rss_write_itbl(nn);
2169 		nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
2170 		update |= NFP_NET_CFG_UPDATE_RSS;
2171 	}
2172 
2173 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
2174 		nfp_net_coalesce_write_cfg(nn);
2175 
2176 		new_ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
2177 		update |= NFP_NET_CFG_UPDATE_IRQMOD;
2178 	}
2179 
2180 	for (r = 0; r < nn->num_tx_rings; r++)
2181 		nfp_net_tx_ring_hw_cfg_write(nn, &nn->tx_rings[r], r);
2182 	for (r = 0; r < nn->num_rx_rings; r++)
2183 		nfp_net_rx_ring_hw_cfg_write(nn, &nn->rx_rings[r], r);
2184 
2185 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->num_tx_rings == 64 ?
2186 		  0xffffffffffffffffULL : ((u64)1 << nn->num_tx_rings) - 1);
2187 
2188 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->num_rx_rings == 64 ?
2189 		  0xffffffffffffffffULL : ((u64)1 << nn->num_rx_rings) - 1);
2190 
2191 	nfp_net_write_mac_addr(nn);
2192 
2193 	nn_writel(nn, NFP_NET_CFG_MTU, nn->netdev->mtu);
2194 	nn_writel(nn, NFP_NET_CFG_FLBUFSZ, nn->fl_bufsz);
2195 
2196 	/* Enable device */
2197 	new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2198 	update |= NFP_NET_CFG_UPDATE_GEN;
2199 	update |= NFP_NET_CFG_UPDATE_MSIX;
2200 	update |= NFP_NET_CFG_UPDATE_RING;
2201 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2202 		new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2203 
2204 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2205 	err = nfp_net_reconfig(nn, update);
2206 
2207 	nn->ctrl = new_ctrl;
2208 
2209 	for (r = 0; r < nn->num_rx_rings; r++)
2210 		nfp_net_rx_ring_fill_freelist(&nn->rx_rings[r]);
2211 
2212 	/* Since reconfiguration requests while NFP is down are ignored we
2213 	 * have to wipe the entire VXLAN configuration and reinitialize it.
2214 	 */
2215 	if (nn->ctrl & NFP_NET_CFG_CTRL_VXLAN) {
2216 		memset(&nn->vxlan_ports, 0, sizeof(nn->vxlan_ports));
2217 		memset(&nn->vxlan_usecnt, 0, sizeof(nn->vxlan_usecnt));
2218 		udp_tunnel_get_rx_info(nn->netdev);
2219 	}
2220 
2221 	return err;
2222 }
2223 
2224 /**
2225  * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2226  * @nn:      NFP Net device to reconfigure
2227  */
2228 static int nfp_net_set_config_and_enable(struct nfp_net *nn)
2229 {
2230 	int err;
2231 
2232 	err = __nfp_net_set_config_and_enable(nn);
2233 	if (err)
2234 		nfp_net_clear_config_and_disable(nn);
2235 
2236 	return err;
2237 }
2238 
2239 /**
2240  * nfp_net_open_stack() - Start the device from stack's perspective
2241  * @nn:      NFP Net device to reconfigure
2242  */
2243 static void nfp_net_open_stack(struct nfp_net *nn)
2244 {
2245 	unsigned int r;
2246 
2247 	for (r = 0; r < nn->num_r_vecs; r++) {
2248 		napi_enable(&nn->r_vecs[r].napi);
2249 		enable_irq(nn->irq_entries[nn->r_vecs[r].irq_idx].vector);
2250 	}
2251 
2252 	netif_tx_wake_all_queues(nn->netdev);
2253 
2254 	enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2255 	nfp_net_read_link_status(nn);
2256 }
2257 
2258 static int nfp_net_netdev_open(struct net_device *netdev)
2259 {
2260 	struct nfp_net *nn = netdev_priv(netdev);
2261 	struct nfp_net_ring_set rx = {
2262 		.n_rings = nn->num_rx_rings,
2263 		.mtu = nn->netdev->mtu,
2264 		.dcnt = nn->rxd_cnt,
2265 	};
2266 	struct nfp_net_ring_set tx = {
2267 		.n_rings = nn->num_tx_rings,
2268 		.dcnt = nn->txd_cnt,
2269 	};
2270 	int err, r;
2271 
2272 	if (nn->ctrl & NFP_NET_CFG_CTRL_ENABLE) {
2273 		nn_err(nn, "Dev is already enabled: 0x%08x\n", nn->ctrl);
2274 		return -EBUSY;
2275 	}
2276 
2277 	/* Step 1: Allocate resources for rings and the like
2278 	 * - Request interrupts
2279 	 * - Allocate RX and TX ring resources
2280 	 * - Setup initial RSS table
2281 	 */
2282 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2283 				      nn->exn_name, sizeof(nn->exn_name),
2284 				      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2285 	if (err)
2286 		return err;
2287 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2288 				      nn->lsc_name, sizeof(nn->lsc_name),
2289 				      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2290 	if (err)
2291 		goto err_free_exn;
2292 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2293 
2294 	for (r = 0; r < nn->num_r_vecs; r++) {
2295 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2296 		if (err)
2297 			goto err_cleanup_vec_p;
2298 	}
2299 
2300 	nn->rx_rings = nfp_net_rx_ring_set_prepare(nn, &rx, nn->xdp_prog);
2301 	if (!nn->rx_rings) {
2302 		err = -ENOMEM;
2303 		goto err_cleanup_vec;
2304 	}
2305 
2306 	nn->tx_rings = nfp_net_tx_ring_set_prepare(nn, &tx,
2307 						   nn->num_stack_tx_rings);
2308 	if (!nn->tx_rings) {
2309 		err = -ENOMEM;
2310 		goto err_free_rx_rings;
2311 	}
2312 
2313 	for (r = 0; r < nn->max_r_vecs; r++)
2314 		nfp_net_vector_assign_rings(nn, &nn->r_vecs[r], r);
2315 
2316 	err = netif_set_real_num_tx_queues(netdev, nn->num_stack_tx_rings);
2317 	if (err)
2318 		goto err_free_rings;
2319 
2320 	err = netif_set_real_num_rx_queues(netdev, nn->num_rx_rings);
2321 	if (err)
2322 		goto err_free_rings;
2323 
2324 	/* Step 2: Configure the NFP
2325 	 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2326 	 * - Write MAC address (in case it changed)
2327 	 * - Set the MTU
2328 	 * - Set the Freelist buffer size
2329 	 * - Enable the FW
2330 	 */
2331 	err = nfp_net_set_config_and_enable(nn);
2332 	if (err)
2333 		goto err_free_rings;
2334 
2335 	/* Step 3: Enable for kernel
2336 	 * - put some freelist descriptors on each RX ring
2337 	 * - enable NAPI on each ring
2338 	 * - enable all TX queues
2339 	 * - set link state
2340 	 */
2341 	nfp_net_open_stack(nn);
2342 
2343 	return 0;
2344 
2345 err_free_rings:
2346 	nfp_net_tx_ring_set_free(nn, &tx);
2347 err_free_rx_rings:
2348 	nfp_net_rx_ring_set_free(nn, &rx, nn->xdp_prog);
2349 err_cleanup_vec:
2350 	r = nn->num_r_vecs;
2351 err_cleanup_vec_p:
2352 	while (r--)
2353 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2354 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2355 err_free_exn:
2356 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2357 	return err;
2358 }
2359 
2360 /**
2361  * nfp_net_close_stack() - Quiescent the stack (part of close)
2362  * @nn:	     NFP Net device to reconfigure
2363  */
2364 static void nfp_net_close_stack(struct nfp_net *nn)
2365 {
2366 	unsigned int r;
2367 
2368 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2369 	netif_carrier_off(nn->netdev);
2370 	nn->link_up = false;
2371 
2372 	for (r = 0; r < nn->num_r_vecs; r++) {
2373 		disable_irq(nn->irq_entries[nn->r_vecs[r].irq_idx].vector);
2374 		napi_disable(&nn->r_vecs[r].napi);
2375 	}
2376 
2377 	netif_tx_disable(nn->netdev);
2378 }
2379 
2380 /**
2381  * nfp_net_close_free_all() - Free all runtime resources
2382  * @nn:      NFP Net device to reconfigure
2383  */
2384 static void nfp_net_close_free_all(struct nfp_net *nn)
2385 {
2386 	unsigned int r;
2387 
2388 	for (r = 0; r < nn->num_rx_rings; r++) {
2389 		nfp_net_rx_ring_bufs_free(nn, &nn->rx_rings[r], nn->xdp_prog);
2390 		nfp_net_rx_ring_free(&nn->rx_rings[r]);
2391 	}
2392 	for (r = 0; r < nn->num_tx_rings; r++)
2393 		nfp_net_tx_ring_free(&nn->tx_rings[r]);
2394 	for (r = 0; r < nn->num_r_vecs; r++)
2395 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2396 
2397 	kfree(nn->rx_rings);
2398 	kfree(nn->tx_rings);
2399 
2400 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2401 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2402 }
2403 
2404 /**
2405  * nfp_net_netdev_close() - Called when the device is downed
2406  * @netdev:      netdev structure
2407  */
2408 static int nfp_net_netdev_close(struct net_device *netdev)
2409 {
2410 	struct nfp_net *nn = netdev_priv(netdev);
2411 
2412 	if (!(nn->ctrl & NFP_NET_CFG_CTRL_ENABLE)) {
2413 		nn_err(nn, "Dev is not up: 0x%08x\n", nn->ctrl);
2414 		return 0;
2415 	}
2416 
2417 	/* Step 1: Disable RX and TX rings from the Linux kernel perspective
2418 	 */
2419 	nfp_net_close_stack(nn);
2420 
2421 	/* Step 2: Tell NFP
2422 	 */
2423 	nfp_net_clear_config_and_disable(nn);
2424 
2425 	/* Step 3: Free resources
2426 	 */
2427 	nfp_net_close_free_all(nn);
2428 
2429 	nn_dbg(nn, "%s down", netdev->name);
2430 	return 0;
2431 }
2432 
2433 static void nfp_net_set_rx_mode(struct net_device *netdev)
2434 {
2435 	struct nfp_net *nn = netdev_priv(netdev);
2436 	u32 new_ctrl;
2437 
2438 	new_ctrl = nn->ctrl;
2439 
2440 	if (netdev->flags & IFF_PROMISC) {
2441 		if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
2442 			new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
2443 		else
2444 			nn_warn(nn, "FW does not support promiscuous mode\n");
2445 	} else {
2446 		new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
2447 	}
2448 
2449 	if (new_ctrl == nn->ctrl)
2450 		return;
2451 
2452 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2453 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
2454 
2455 	nn->ctrl = new_ctrl;
2456 }
2457 
2458 static void nfp_net_rss_init_itbl(struct nfp_net *nn)
2459 {
2460 	int i;
2461 
2462 	for (i = 0; i < sizeof(nn->rss_itbl); i++)
2463 		nn->rss_itbl[i] =
2464 			ethtool_rxfh_indir_default(i, nn->num_rx_rings);
2465 }
2466 
2467 static int
2468 nfp_net_ring_swap_enable(struct nfp_net *nn, unsigned int *num_vecs,
2469 			 unsigned int *stack_tx_rings,
2470 			 struct bpf_prog **xdp_prog,
2471 			 struct nfp_net_ring_set *rx,
2472 			 struct nfp_net_ring_set *tx)
2473 {
2474 	unsigned int r;
2475 	int err;
2476 
2477 	if (rx)
2478 		nfp_net_rx_ring_set_swap(nn, rx);
2479 	if (tx)
2480 		nfp_net_tx_ring_set_swap(nn, tx);
2481 
2482 	swap(*num_vecs, nn->num_r_vecs);
2483 	swap(*stack_tx_rings, nn->num_stack_tx_rings);
2484 	*xdp_prog = xchg(&nn->xdp_prog, *xdp_prog);
2485 
2486 	for (r = 0; r <	nn->max_r_vecs; r++)
2487 		nfp_net_vector_assign_rings(nn, &nn->r_vecs[r], r);
2488 
2489 	if (!netif_is_rxfh_configured(nn->netdev))
2490 		nfp_net_rss_init_itbl(nn);
2491 
2492 	err = netif_set_real_num_rx_queues(nn->netdev,
2493 					   nn->num_rx_rings);
2494 	if (err)
2495 		return err;
2496 
2497 	if (nn->netdev->real_num_tx_queues != nn->num_stack_tx_rings) {
2498 		err = netif_set_real_num_tx_queues(nn->netdev,
2499 						   nn->num_stack_tx_rings);
2500 		if (err)
2501 			return err;
2502 	}
2503 
2504 	return __nfp_net_set_config_and_enable(nn);
2505 }
2506 
2507 static int
2508 nfp_net_check_config(struct nfp_net *nn, struct bpf_prog *xdp_prog,
2509 		     struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx)
2510 {
2511 	/* XDP-enabled tests */
2512 	if (!xdp_prog)
2513 		return 0;
2514 	if (rx && nfp_net_calc_fl_bufsz(nn, rx->mtu) > PAGE_SIZE) {
2515 		nn_warn(nn, "MTU too large w/ XDP enabled\n");
2516 		return -EINVAL;
2517 	}
2518 	if (tx && tx->n_rings > nn->max_tx_rings) {
2519 		nn_warn(nn, "Insufficient number of TX rings w/ XDP enabled\n");
2520 		return -EINVAL;
2521 	}
2522 
2523 	return 0;
2524 }
2525 
2526 static void
2527 nfp_net_ring_reconfig_down(struct nfp_net *nn, struct bpf_prog **xdp_prog,
2528 			   struct nfp_net_ring_set *rx,
2529 			   struct nfp_net_ring_set *tx,
2530 			   unsigned int stack_tx_rings, unsigned int num_vecs)
2531 {
2532 	nn->netdev->mtu = rx ? rx->mtu : nn->netdev->mtu;
2533 	nn->fl_bufsz = nfp_net_calc_fl_bufsz(nn, nn->netdev->mtu);
2534 	nn->rxd_cnt = rx ? rx->dcnt : nn->rxd_cnt;
2535 	nn->txd_cnt = tx ? tx->dcnt : nn->txd_cnt;
2536 	nn->num_rx_rings = rx ? rx->n_rings : nn->num_rx_rings;
2537 	nn->num_tx_rings = tx ? tx->n_rings : nn->num_tx_rings;
2538 	nn->num_stack_tx_rings = stack_tx_rings;
2539 	nn->num_r_vecs = num_vecs;
2540 	*xdp_prog = xchg(&nn->xdp_prog, *xdp_prog);
2541 
2542 	if (!netif_is_rxfh_configured(nn->netdev))
2543 		nfp_net_rss_init_itbl(nn);
2544 }
2545 
2546 int
2547 nfp_net_ring_reconfig(struct nfp_net *nn, struct bpf_prog **xdp_prog,
2548 		      struct nfp_net_ring_set *rx, struct nfp_net_ring_set *tx)
2549 {
2550 	unsigned int stack_tx_rings, num_vecs, r;
2551 	int err;
2552 
2553 	stack_tx_rings = tx ? tx->n_rings : nn->num_tx_rings;
2554 	if (*xdp_prog)
2555 		stack_tx_rings -= rx ? rx->n_rings : nn->num_rx_rings;
2556 
2557 	num_vecs = max(rx ? rx->n_rings : nn->num_rx_rings, stack_tx_rings);
2558 
2559 	err = nfp_net_check_config(nn, *xdp_prog, rx, tx);
2560 	if (err)
2561 		return err;
2562 
2563 	if (!netif_running(nn->netdev)) {
2564 		nfp_net_ring_reconfig_down(nn, xdp_prog, rx, tx,
2565 					   stack_tx_rings, num_vecs);
2566 		return 0;
2567 	}
2568 
2569 	/* Prepare new rings */
2570 	for (r = nn->num_r_vecs; r < num_vecs; r++) {
2571 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2572 		if (err) {
2573 			num_vecs = r;
2574 			goto err_cleanup_vecs;
2575 		}
2576 	}
2577 	if (rx) {
2578 		if (!nfp_net_rx_ring_set_prepare(nn, rx, *xdp_prog)) {
2579 			err = -ENOMEM;
2580 			goto err_cleanup_vecs;
2581 		}
2582 	}
2583 	if (tx) {
2584 		if (!nfp_net_tx_ring_set_prepare(nn, tx, stack_tx_rings)) {
2585 			err = -ENOMEM;
2586 			goto err_free_rx;
2587 		}
2588 	}
2589 
2590 	/* Stop device, swap in new rings, try to start the firmware */
2591 	nfp_net_close_stack(nn);
2592 	nfp_net_clear_config_and_disable(nn);
2593 
2594 	err = nfp_net_ring_swap_enable(nn, &num_vecs, &stack_tx_rings,
2595 				       xdp_prog, rx, tx);
2596 	if (err) {
2597 		int err2;
2598 
2599 		nfp_net_clear_config_and_disable(nn);
2600 
2601 		/* Try with old configuration and old rings */
2602 		err2 = nfp_net_ring_swap_enable(nn, &num_vecs, &stack_tx_rings,
2603 						xdp_prog, rx, tx);
2604 		if (err2)
2605 			nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
2606 			       err, err2);
2607 	}
2608 	for (r = num_vecs - 1; r >= nn->num_r_vecs; r--)
2609 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2610 
2611 	if (rx)
2612 		nfp_net_rx_ring_set_free(nn, rx, *xdp_prog);
2613 	if (tx)
2614 		nfp_net_tx_ring_set_free(nn, tx);
2615 
2616 	nfp_net_open_stack(nn);
2617 
2618 	return err;
2619 
2620 err_free_rx:
2621 	if (rx)
2622 		nfp_net_rx_ring_set_free(nn, rx, *xdp_prog);
2623 err_cleanup_vecs:
2624 	for (r = num_vecs - 1; r >= nn->num_r_vecs; r--)
2625 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2626 	return err;
2627 }
2628 
2629 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
2630 {
2631 	struct nfp_net *nn = netdev_priv(netdev);
2632 	struct nfp_net_ring_set rx = {
2633 		.n_rings = nn->num_rx_rings,
2634 		.mtu = new_mtu,
2635 		.dcnt = nn->rxd_cnt,
2636 	};
2637 
2638 	return nfp_net_ring_reconfig(nn, &nn->xdp_prog, &rx, NULL);
2639 }
2640 
2641 static struct rtnl_link_stats64 *nfp_net_stat64(struct net_device *netdev,
2642 						struct rtnl_link_stats64 *stats)
2643 {
2644 	struct nfp_net *nn = netdev_priv(netdev);
2645 	int r;
2646 
2647 	for (r = 0; r < nn->num_r_vecs; r++) {
2648 		struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
2649 		u64 data[3];
2650 		unsigned int start;
2651 
2652 		do {
2653 			start = u64_stats_fetch_begin(&r_vec->rx_sync);
2654 			data[0] = r_vec->rx_pkts;
2655 			data[1] = r_vec->rx_bytes;
2656 			data[2] = r_vec->rx_drops;
2657 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
2658 		stats->rx_packets += data[0];
2659 		stats->rx_bytes += data[1];
2660 		stats->rx_dropped += data[2];
2661 
2662 		do {
2663 			start = u64_stats_fetch_begin(&r_vec->tx_sync);
2664 			data[0] = r_vec->tx_pkts;
2665 			data[1] = r_vec->tx_bytes;
2666 			data[2] = r_vec->tx_errors;
2667 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
2668 		stats->tx_packets += data[0];
2669 		stats->tx_bytes += data[1];
2670 		stats->tx_errors += data[2];
2671 	}
2672 
2673 	return stats;
2674 }
2675 
2676 static bool nfp_net_ebpf_capable(struct nfp_net *nn)
2677 {
2678 	if (nn->cap & NFP_NET_CFG_CTRL_BPF &&
2679 	    nn_readb(nn, NFP_NET_CFG_BPF_ABI) == NFP_NET_BPF_ABI)
2680 		return true;
2681 	return false;
2682 }
2683 
2684 static int
2685 nfp_net_setup_tc(struct net_device *netdev, u32 handle, __be16 proto,
2686 		 struct tc_to_netdev *tc)
2687 {
2688 	struct nfp_net *nn = netdev_priv(netdev);
2689 
2690 	if (TC_H_MAJ(handle) != TC_H_MAJ(TC_H_INGRESS))
2691 		return -ENOTSUPP;
2692 	if (proto != htons(ETH_P_ALL))
2693 		return -ENOTSUPP;
2694 
2695 	if (tc->type == TC_SETUP_CLSBPF && nfp_net_ebpf_capable(nn)) {
2696 		if (!nn->bpf_offload_xdp)
2697 			return nfp_net_bpf_offload(nn, tc->cls_bpf);
2698 		else
2699 			return -EBUSY;
2700 	}
2701 
2702 	return -EINVAL;
2703 }
2704 
2705 static int nfp_net_set_features(struct net_device *netdev,
2706 				netdev_features_t features)
2707 {
2708 	netdev_features_t changed = netdev->features ^ features;
2709 	struct nfp_net *nn = netdev_priv(netdev);
2710 	u32 new_ctrl;
2711 	int err;
2712 
2713 	/* Assume this is not called with features we have not advertised */
2714 
2715 	new_ctrl = nn->ctrl;
2716 
2717 	if (changed & NETIF_F_RXCSUM) {
2718 		if (features & NETIF_F_RXCSUM)
2719 			new_ctrl |= NFP_NET_CFG_CTRL_RXCSUM;
2720 		else
2721 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM;
2722 	}
2723 
2724 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
2725 		if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
2726 			new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
2727 		else
2728 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
2729 	}
2730 
2731 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
2732 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
2733 			new_ctrl |= NFP_NET_CFG_CTRL_LSO;
2734 		else
2735 			new_ctrl &= ~NFP_NET_CFG_CTRL_LSO;
2736 	}
2737 
2738 	if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
2739 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
2740 			new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
2741 		else
2742 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
2743 	}
2744 
2745 	if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
2746 		if (features & NETIF_F_HW_VLAN_CTAG_TX)
2747 			new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
2748 		else
2749 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
2750 	}
2751 
2752 	if (changed & NETIF_F_SG) {
2753 		if (features & NETIF_F_SG)
2754 			new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
2755 		else
2756 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
2757 	}
2758 
2759 	if (changed & NETIF_F_HW_TC && nn->ctrl & NFP_NET_CFG_CTRL_BPF) {
2760 		nn_err(nn, "Cannot disable HW TC offload while in use\n");
2761 		return -EBUSY;
2762 	}
2763 
2764 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
2765 	       netdev->features, features, changed);
2766 
2767 	if (new_ctrl == nn->ctrl)
2768 		return 0;
2769 
2770 	nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->ctrl, new_ctrl);
2771 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2772 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
2773 	if (err)
2774 		return err;
2775 
2776 	nn->ctrl = new_ctrl;
2777 
2778 	return 0;
2779 }
2780 
2781 static netdev_features_t
2782 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
2783 		       netdev_features_t features)
2784 {
2785 	u8 l4_hdr;
2786 
2787 	/* We can't do TSO over double tagged packets (802.1AD) */
2788 	features &= vlan_features_check(skb, features);
2789 
2790 	if (!skb->encapsulation)
2791 		return features;
2792 
2793 	/* Ensure that inner L4 header offset fits into TX descriptor field */
2794 	if (skb_is_gso(skb)) {
2795 		u32 hdrlen;
2796 
2797 		hdrlen = skb_inner_transport_header(skb) - skb->data +
2798 			inner_tcp_hdrlen(skb);
2799 
2800 		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
2801 			features &= ~NETIF_F_GSO_MASK;
2802 	}
2803 
2804 	/* VXLAN/GRE check */
2805 	switch (vlan_get_protocol(skb)) {
2806 	case htons(ETH_P_IP):
2807 		l4_hdr = ip_hdr(skb)->protocol;
2808 		break;
2809 	case htons(ETH_P_IPV6):
2810 		l4_hdr = ipv6_hdr(skb)->nexthdr;
2811 		break;
2812 	default:
2813 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2814 	}
2815 
2816 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
2817 	    skb->inner_protocol != htons(ETH_P_TEB) ||
2818 	    (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
2819 	    (l4_hdr == IPPROTO_UDP &&
2820 	     (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
2821 	      sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
2822 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
2823 
2824 	return features;
2825 }
2826 
2827 /**
2828  * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
2829  * @nn:   NFP Net device to reconfigure
2830  * @idx:  Index into the port table where new port should be written
2831  * @port: UDP port to configure (pass zero to remove VXLAN port)
2832  */
2833 static void nfp_net_set_vxlan_port(struct nfp_net *nn, int idx, __be16 port)
2834 {
2835 	int i;
2836 
2837 	nn->vxlan_ports[idx] = port;
2838 
2839 	if (!(nn->ctrl & NFP_NET_CFG_CTRL_VXLAN))
2840 		return;
2841 
2842 	BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
2843 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2)
2844 		nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(port),
2845 			  be16_to_cpu(nn->vxlan_ports[i + 1]) << 16 |
2846 			  be16_to_cpu(nn->vxlan_ports[i]));
2847 
2848 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_VXLAN);
2849 }
2850 
2851 /**
2852  * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
2853  * @nn:   NFP Network structure
2854  * @port: UDP port to look for
2855  *
2856  * Return: if the port is already in the table -- it's position;
2857  *	   if the port is not in the table -- free position to use;
2858  *	   if the table is full -- -ENOSPC.
2859  */
2860 static int nfp_net_find_vxlan_idx(struct nfp_net *nn, __be16 port)
2861 {
2862 	int i, free_idx = -ENOSPC;
2863 
2864 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i++) {
2865 		if (nn->vxlan_ports[i] == port)
2866 			return i;
2867 		if (!nn->vxlan_usecnt[i])
2868 			free_idx = i;
2869 	}
2870 
2871 	return free_idx;
2872 }
2873 
2874 static void nfp_net_add_vxlan_port(struct net_device *netdev,
2875 				   struct udp_tunnel_info *ti)
2876 {
2877 	struct nfp_net *nn = netdev_priv(netdev);
2878 	int idx;
2879 
2880 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2881 		return;
2882 
2883 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
2884 	if (idx == -ENOSPC)
2885 		return;
2886 
2887 	if (!nn->vxlan_usecnt[idx]++)
2888 		nfp_net_set_vxlan_port(nn, idx, ti->port);
2889 }
2890 
2891 static void nfp_net_del_vxlan_port(struct net_device *netdev,
2892 				   struct udp_tunnel_info *ti)
2893 {
2894 	struct nfp_net *nn = netdev_priv(netdev);
2895 	int idx;
2896 
2897 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
2898 		return;
2899 
2900 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
2901 	if (idx == -ENOSPC || !nn->vxlan_usecnt[idx])
2902 		return;
2903 
2904 	if (!--nn->vxlan_usecnt[idx])
2905 		nfp_net_set_vxlan_port(nn, idx, 0);
2906 }
2907 
2908 static int nfp_net_xdp_offload(struct nfp_net *nn, struct bpf_prog *prog)
2909 {
2910 	struct tc_cls_bpf_offload cmd = {
2911 		.prog = prog,
2912 	};
2913 	int ret;
2914 
2915 	if (!nfp_net_ebpf_capable(nn))
2916 		return -EINVAL;
2917 
2918 	if (nn->ctrl & NFP_NET_CFG_CTRL_BPF) {
2919 		if (!nn->bpf_offload_xdp)
2920 			return prog ? -EBUSY : 0;
2921 		cmd.command = prog ? TC_CLSBPF_REPLACE : TC_CLSBPF_DESTROY;
2922 	} else {
2923 		if (!prog)
2924 			return 0;
2925 		cmd.command = TC_CLSBPF_ADD;
2926 	}
2927 
2928 	ret = nfp_net_bpf_offload(nn, &cmd);
2929 	/* Stop offload if replace not possible */
2930 	if (ret && cmd.command == TC_CLSBPF_REPLACE)
2931 		nfp_net_xdp_offload(nn, NULL);
2932 	nn->bpf_offload_xdp = prog && !ret;
2933 	return ret;
2934 }
2935 
2936 static int nfp_net_xdp_setup(struct nfp_net *nn, struct bpf_prog *prog)
2937 {
2938 	struct nfp_net_ring_set rx = {
2939 		.n_rings = nn->num_rx_rings,
2940 		.mtu = nn->netdev->mtu,
2941 		.dcnt = nn->rxd_cnt,
2942 	};
2943 	struct nfp_net_ring_set tx = {
2944 		.n_rings = nn->num_tx_rings,
2945 		.dcnt = nn->txd_cnt,
2946 	};
2947 	int err;
2948 
2949 	if (prog && prog->xdp_adjust_head) {
2950 		nn_err(nn, "Does not support bpf_xdp_adjust_head()\n");
2951 		return -EOPNOTSUPP;
2952 	}
2953 	if (!prog && !nn->xdp_prog)
2954 		return 0;
2955 	if (prog && nn->xdp_prog) {
2956 		prog = xchg(&nn->xdp_prog, prog);
2957 		bpf_prog_put(prog);
2958 		nfp_net_xdp_offload(nn, nn->xdp_prog);
2959 		return 0;
2960 	}
2961 
2962 	tx.n_rings += prog ? nn->num_rx_rings : -nn->num_rx_rings;
2963 
2964 	/* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
2965 	err = nfp_net_ring_reconfig(nn, &prog, &rx, &tx);
2966 	if (err)
2967 		return err;
2968 
2969 	/* @prog got swapped and is now the old one */
2970 	if (prog)
2971 		bpf_prog_put(prog);
2972 
2973 	nfp_net_xdp_offload(nn, nn->xdp_prog);
2974 
2975 	return 0;
2976 }
2977 
2978 static int nfp_net_xdp(struct net_device *netdev, struct netdev_xdp *xdp)
2979 {
2980 	struct nfp_net *nn = netdev_priv(netdev);
2981 
2982 	switch (xdp->command) {
2983 	case XDP_SETUP_PROG:
2984 		return nfp_net_xdp_setup(nn, xdp->prog);
2985 	case XDP_QUERY_PROG:
2986 		xdp->prog_attached = !!nn->xdp_prog;
2987 		return 0;
2988 	default:
2989 		return -EINVAL;
2990 	}
2991 }
2992 
2993 static const struct net_device_ops nfp_net_netdev_ops = {
2994 	.ndo_open		= nfp_net_netdev_open,
2995 	.ndo_stop		= nfp_net_netdev_close,
2996 	.ndo_start_xmit		= nfp_net_tx,
2997 	.ndo_get_stats64	= nfp_net_stat64,
2998 	.ndo_setup_tc		= nfp_net_setup_tc,
2999 	.ndo_tx_timeout		= nfp_net_tx_timeout,
3000 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
3001 	.ndo_change_mtu		= nfp_net_change_mtu,
3002 	.ndo_set_mac_address	= eth_mac_addr,
3003 	.ndo_set_features	= nfp_net_set_features,
3004 	.ndo_features_check	= nfp_net_features_check,
3005 	.ndo_udp_tunnel_add	= nfp_net_add_vxlan_port,
3006 	.ndo_udp_tunnel_del	= nfp_net_del_vxlan_port,
3007 	.ndo_xdp		= nfp_net_xdp,
3008 };
3009 
3010 /**
3011  * nfp_net_info() - Print general info about the NIC
3012  * @nn:      NFP Net device to reconfigure
3013  */
3014 void nfp_net_info(struct nfp_net *nn)
3015 {
3016 	nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
3017 		nn->is_vf ? "VF " : "",
3018 		nn->num_tx_rings, nn->max_tx_rings,
3019 		nn->num_rx_rings, nn->max_rx_rings);
3020 	nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3021 		nn->fw_ver.resv, nn->fw_ver.class,
3022 		nn->fw_ver.major, nn->fw_ver.minor,
3023 		nn->max_mtu);
3024 	nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
3025 		nn->cap,
3026 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
3027 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
3028 		nn->cap & NFP_NET_CFG_CTRL_L2MC     ? "L2MCFILT " : "",
3029 		nn->cap & NFP_NET_CFG_CTRL_RXCSUM   ? "RXCSUM "   : "",
3030 		nn->cap & NFP_NET_CFG_CTRL_TXCSUM   ? "TXCSUM "   : "",
3031 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN   ? "RXVLAN "   : "",
3032 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN   ? "TXVLAN "   : "",
3033 		nn->cap & NFP_NET_CFG_CTRL_SCATTER  ? "SCATTER "  : "",
3034 		nn->cap & NFP_NET_CFG_CTRL_GATHER   ? "GATHER "   : "",
3035 		nn->cap & NFP_NET_CFG_CTRL_LSO      ? "TSO "      : "",
3036 		nn->cap & NFP_NET_CFG_CTRL_RSS      ? "RSS "      : "",
3037 		nn->cap & NFP_NET_CFG_CTRL_L2SWITCH ? "L2SWITCH " : "",
3038 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
3039 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
3040 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
3041 		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
3042 		nfp_net_ebpf_capable(nn)            ? "BPF "	  : "");
3043 }
3044 
3045 /**
3046  * nfp_net_netdev_alloc() - Allocate netdev and related structure
3047  * @pdev:         PCI device
3048  * @max_tx_rings: Maximum number of TX rings supported by device
3049  * @max_rx_rings: Maximum number of RX rings supported by device
3050  *
3051  * This function allocates a netdev device and fills in the initial
3052  * part of the @struct nfp_net structure.
3053  *
3054  * Return: NFP Net device structure, or ERR_PTR on error.
3055  */
3056 struct nfp_net *nfp_net_netdev_alloc(struct pci_dev *pdev,
3057 				     unsigned int max_tx_rings,
3058 				     unsigned int max_rx_rings)
3059 {
3060 	struct net_device *netdev;
3061 	struct nfp_net *nn;
3062 
3063 	netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
3064 				    max_tx_rings, max_rx_rings);
3065 	if (!netdev)
3066 		return ERR_PTR(-ENOMEM);
3067 
3068 	SET_NETDEV_DEV(netdev, &pdev->dev);
3069 	nn = netdev_priv(netdev);
3070 
3071 	nn->netdev = netdev;
3072 	nn->pdev = pdev;
3073 
3074 	nn->max_tx_rings = max_tx_rings;
3075 	nn->max_rx_rings = max_rx_rings;
3076 
3077 	nn->num_tx_rings = min_t(unsigned int, max_tx_rings, num_online_cpus());
3078 	nn->num_rx_rings = min_t(unsigned int, max_rx_rings,
3079 				 netif_get_num_default_rss_queues());
3080 
3081 	nn->num_r_vecs = max(nn->num_tx_rings, nn->num_rx_rings);
3082 	nn->num_r_vecs = min_t(unsigned int, nn->num_r_vecs, num_online_cpus());
3083 
3084 	nn->txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
3085 	nn->rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
3086 
3087 	spin_lock_init(&nn->reconfig_lock);
3088 	spin_lock_init(&nn->rx_filter_lock);
3089 	spin_lock_init(&nn->link_status_lock);
3090 
3091 	setup_timer(&nn->reconfig_timer,
3092 		    nfp_net_reconfig_timer, (unsigned long)nn);
3093 	setup_timer(&nn->rx_filter_stats_timer,
3094 		    nfp_net_filter_stats_timer, (unsigned long)nn);
3095 
3096 	return nn;
3097 }
3098 
3099 /**
3100  * nfp_net_netdev_free() - Undo what @nfp_net_netdev_alloc() did
3101  * @nn:      NFP Net device to reconfigure
3102  */
3103 void nfp_net_netdev_free(struct nfp_net *nn)
3104 {
3105 	free_netdev(nn->netdev);
3106 }
3107 
3108 /**
3109  * nfp_net_rss_init() - Set the initial RSS parameters
3110  * @nn:	     NFP Net device to reconfigure
3111  */
3112 static void nfp_net_rss_init(struct nfp_net *nn)
3113 {
3114 	netdev_rss_key_fill(nn->rss_key, NFP_NET_CFG_RSS_KEY_SZ);
3115 
3116 	nfp_net_rss_init_itbl(nn);
3117 
3118 	/* Enable IPv4/IPv6 TCP by default */
3119 	nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
3120 		      NFP_NET_CFG_RSS_IPV6_TCP |
3121 		      NFP_NET_CFG_RSS_TOEPLITZ |
3122 		      NFP_NET_CFG_RSS_MASK;
3123 }
3124 
3125 /**
3126  * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
3127  * @nn:	     NFP Net device to reconfigure
3128  */
3129 static void nfp_net_irqmod_init(struct nfp_net *nn)
3130 {
3131 	nn->rx_coalesce_usecs      = 50;
3132 	nn->rx_coalesce_max_frames = 64;
3133 	nn->tx_coalesce_usecs      = 50;
3134 	nn->tx_coalesce_max_frames = 64;
3135 }
3136 
3137 /**
3138  * nfp_net_netdev_init() - Initialise/finalise the netdev structure
3139  * @netdev:      netdev structure
3140  *
3141  * Return: 0 on success or negative errno on error.
3142  */
3143 int nfp_net_netdev_init(struct net_device *netdev)
3144 {
3145 	struct nfp_net *nn = netdev_priv(netdev);
3146 	int err;
3147 
3148 	/* Get some of the read-only fields from the BAR */
3149 	nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
3150 	nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
3151 
3152 	nfp_net_write_mac_addr(nn);
3153 
3154 	/* Determine RX packet/metadata boundary offset */
3155 	if (nn->fw_ver.major >= 2)
3156 		nn->rx_offset = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
3157 	else
3158 		nn->rx_offset = NFP_NET_RX_OFFSET;
3159 
3160 	/* Set default MTU and Freelist buffer size */
3161 	if (nn->max_mtu < NFP_NET_DEFAULT_MTU)
3162 		netdev->mtu = nn->max_mtu;
3163 	else
3164 		netdev->mtu = NFP_NET_DEFAULT_MTU;
3165 	nn->fl_bufsz = nfp_net_calc_fl_bufsz(nn, netdev->mtu);
3166 
3167 	/* Advertise/enable offloads based on capabilities
3168 	 *
3169 	 * Note: netdev->features show the currently enabled features
3170 	 * and netdev->hw_features advertises which features are
3171 	 * supported.  By default we enable most features.
3172 	 */
3173 	netdev->hw_features = NETIF_F_HIGHDMA;
3174 	if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM) {
3175 		netdev->hw_features |= NETIF_F_RXCSUM;
3176 		nn->ctrl |= NFP_NET_CFG_CTRL_RXCSUM;
3177 	}
3178 	if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
3179 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3180 		nn->ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3181 	}
3182 	if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
3183 		netdev->hw_features |= NETIF_F_SG;
3184 		nn->ctrl |= NFP_NET_CFG_CTRL_GATHER;
3185 	}
3186 	if ((nn->cap & NFP_NET_CFG_CTRL_LSO) && nn->fw_ver.major > 2) {
3187 		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3188 		nn->ctrl |= NFP_NET_CFG_CTRL_LSO;
3189 	}
3190 	if (nn->cap & NFP_NET_CFG_CTRL_RSS) {
3191 		netdev->hw_features |= NETIF_F_RXHASH;
3192 		nfp_net_rss_init(nn);
3193 		nn->ctrl |= NFP_NET_CFG_CTRL_RSS;
3194 	}
3195 	if (nn->cap & NFP_NET_CFG_CTRL_VXLAN &&
3196 	    nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
3197 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3198 			netdev->hw_features |= NETIF_F_GSO_GRE |
3199 					       NETIF_F_GSO_UDP_TUNNEL;
3200 		nn->ctrl |= NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE;
3201 
3202 		netdev->hw_enc_features = netdev->hw_features;
3203 	}
3204 
3205 	netdev->vlan_features = netdev->hw_features;
3206 
3207 	if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
3208 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3209 		nn->ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3210 	}
3211 	if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
3212 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3213 		nn->ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3214 	}
3215 
3216 	netdev->features = netdev->hw_features;
3217 
3218 	if (nfp_net_ebpf_capable(nn))
3219 		netdev->hw_features |= NETIF_F_HW_TC;
3220 
3221 	/* Advertise but disable TSO by default. */
3222 	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
3223 
3224 	/* Allow L2 Broadcast and Multicast through by default, if supported */
3225 	if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
3226 		nn->ctrl |= NFP_NET_CFG_CTRL_L2BC;
3227 	if (nn->cap & NFP_NET_CFG_CTRL_L2MC)
3228 		nn->ctrl |= NFP_NET_CFG_CTRL_L2MC;
3229 
3230 	/* Allow IRQ moderation, if supported */
3231 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
3232 		nfp_net_irqmod_init(nn);
3233 		nn->ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
3234 	}
3235 
3236 	/* Stash the re-configuration queue away.  First odd queue in TX Bar */
3237 	nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
3238 
3239 	/* Make sure the FW knows the netdev is supposed to be disabled here */
3240 	nn_writel(nn, NFP_NET_CFG_CTRL, 0);
3241 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
3242 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
3243 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
3244 				   NFP_NET_CFG_UPDATE_GEN);
3245 	if (err)
3246 		return err;
3247 
3248 	/* Finalise the netdev setup */
3249 	netdev->netdev_ops = &nfp_net_netdev_ops;
3250 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
3251 
3252 	/* MTU range: 68 - hw-specific max */
3253 	netdev->min_mtu = ETH_MIN_MTU;
3254 	netdev->max_mtu = nn->max_mtu;
3255 
3256 	netif_carrier_off(netdev);
3257 
3258 	nfp_net_set_ethtool_ops(netdev);
3259 	nfp_net_irqs_assign(netdev);
3260 
3261 	return register_netdev(netdev);
3262 }
3263 
3264 /**
3265  * nfp_net_netdev_clean() - Undo what nfp_net_netdev_init() did.
3266  * @netdev:      netdev structure
3267  */
3268 void nfp_net_netdev_clean(struct net_device *netdev)
3269 {
3270 	struct nfp_net *nn = netdev_priv(netdev);
3271 
3272 	if (nn->xdp_prog)
3273 		bpf_prog_put(nn->xdp_prog);
3274 	if (nn->bpf_offload_xdp)
3275 		nfp_net_xdp_offload(nn, NULL);
3276 	unregister_netdev(nn->netdev);
3277 }
3278