1 /*
2  * Copyright (C) 2015-2017 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/bitfield.h>
45 #include <linux/bpf.h>
46 #include <linux/bpf_trace.h>
47 #include <linux/module.h>
48 #include <linux/kernel.h>
49 #include <linux/init.h>
50 #include <linux/fs.h>
51 #include <linux/netdevice.h>
52 #include <linux/etherdevice.h>
53 #include <linux/interrupt.h>
54 #include <linux/ip.h>
55 #include <linux/ipv6.h>
56 #include <linux/page_ref.h>
57 #include <linux/pci.h>
58 #include <linux/pci_regs.h>
59 #include <linux/msi.h>
60 #include <linux/ethtool.h>
61 #include <linux/log2.h>
62 #include <linux/if_vlan.h>
63 #include <linux/random.h>
64 #include <linux/vmalloc.h>
65 #include <linux/ktime.h>
66 
67 #include <net/switchdev.h>
68 #include <net/vxlan.h>
69 
70 #include "nfpcore/nfp_nsp.h"
71 #include "nfp_app.h"
72 #include "nfp_net_ctrl.h"
73 #include "nfp_net.h"
74 #include "nfp_net_sriov.h"
75 #include "nfp_port.h"
76 
77 /**
78  * nfp_net_get_fw_version() - Read and parse the FW version
79  * @fw_ver:	Output fw_version structure to read to
80  * @ctrl_bar:	Mapped address of the control BAR
81  */
82 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
83 			    void __iomem *ctrl_bar)
84 {
85 	u32 reg;
86 
87 	reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
88 	put_unaligned_le32(reg, fw_ver);
89 }
90 
91 static dma_addr_t nfp_net_dma_map_rx(struct nfp_net_dp *dp, void *frag)
92 {
93 	return dma_map_single_attrs(dp->dev, frag + NFP_NET_RX_BUF_HEADROOM,
94 				    dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
95 				    dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
96 }
97 
98 static void
99 nfp_net_dma_sync_dev_rx(const struct nfp_net_dp *dp, dma_addr_t dma_addr)
100 {
101 	dma_sync_single_for_device(dp->dev, dma_addr,
102 				   dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
103 				   dp->rx_dma_dir);
104 }
105 
106 static void nfp_net_dma_unmap_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr)
107 {
108 	dma_unmap_single_attrs(dp->dev, dma_addr,
109 			       dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
110 			       dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
111 }
112 
113 static void nfp_net_dma_sync_cpu_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr,
114 				    unsigned int len)
115 {
116 	dma_sync_single_for_cpu(dp->dev, dma_addr - NFP_NET_RX_BUF_HEADROOM,
117 				len, dp->rx_dma_dir);
118 }
119 
120 /* Firmware reconfig
121  *
122  * Firmware reconfig may take a while so we have two versions of it -
123  * synchronous and asynchronous (posted).  All synchronous callers are holding
124  * RTNL so we don't have to worry about serializing them.
125  */
126 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
127 {
128 	nn_writel(nn, NFP_NET_CFG_UPDATE, update);
129 	/* ensure update is written before pinging HW */
130 	nn_pci_flush(nn);
131 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
132 }
133 
134 /* Pass 0 as update to run posted reconfigs. */
135 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
136 {
137 	update |= nn->reconfig_posted;
138 	nn->reconfig_posted = 0;
139 
140 	nfp_net_reconfig_start(nn, update);
141 
142 	nn->reconfig_timer_active = true;
143 	mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
144 }
145 
146 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
147 {
148 	u32 reg;
149 
150 	reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
151 	if (reg == 0)
152 		return true;
153 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
154 		nn_err(nn, "Reconfig error: 0x%08x\n", reg);
155 		return true;
156 	} else if (last_check) {
157 		nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
158 		return true;
159 	}
160 
161 	return false;
162 }
163 
164 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
165 {
166 	bool timed_out = false;
167 
168 	/* Poll update field, waiting for NFP to ack the config */
169 	while (!nfp_net_reconfig_check_done(nn, timed_out)) {
170 		msleep(1);
171 		timed_out = time_is_before_eq_jiffies(deadline);
172 	}
173 
174 	if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
175 		return -EIO;
176 
177 	return timed_out ? -EIO : 0;
178 }
179 
180 static void nfp_net_reconfig_timer(unsigned long data)
181 {
182 	struct nfp_net *nn = (void *)data;
183 
184 	spin_lock_bh(&nn->reconfig_lock);
185 
186 	nn->reconfig_timer_active = false;
187 
188 	/* If sync caller is present it will take over from us */
189 	if (nn->reconfig_sync_present)
190 		goto done;
191 
192 	/* Read reconfig status and report errors */
193 	nfp_net_reconfig_check_done(nn, true);
194 
195 	if (nn->reconfig_posted)
196 		nfp_net_reconfig_start_async(nn, 0);
197 done:
198 	spin_unlock_bh(&nn->reconfig_lock);
199 }
200 
201 /**
202  * nfp_net_reconfig_post() - Post async reconfig request
203  * @nn:      NFP Net device to reconfigure
204  * @update:  The value for the update field in the BAR config
205  *
206  * Record FW reconfiguration request.  Reconfiguration will be kicked off
207  * whenever reconfiguration machinery is idle.  Multiple requests can be
208  * merged together!
209  */
210 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
211 {
212 	spin_lock_bh(&nn->reconfig_lock);
213 
214 	/* Sync caller will kick off async reconf when it's done, just post */
215 	if (nn->reconfig_sync_present) {
216 		nn->reconfig_posted |= update;
217 		goto done;
218 	}
219 
220 	/* Opportunistically check if the previous command is done */
221 	if (!nn->reconfig_timer_active ||
222 	    nfp_net_reconfig_check_done(nn, false))
223 		nfp_net_reconfig_start_async(nn, update);
224 	else
225 		nn->reconfig_posted |= update;
226 done:
227 	spin_unlock_bh(&nn->reconfig_lock);
228 }
229 
230 /**
231  * nfp_net_reconfig() - Reconfigure the firmware
232  * @nn:      NFP Net device to reconfigure
233  * @update:  The value for the update field in the BAR config
234  *
235  * Write the update word to the BAR and ping the reconfig queue.  The
236  * poll until the firmware has acknowledged the update by zeroing the
237  * update word.
238  *
239  * Return: Negative errno on error, 0 on success
240  */
241 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
242 {
243 	bool cancelled_timer = false;
244 	u32 pre_posted_requests;
245 	int ret;
246 
247 	spin_lock_bh(&nn->reconfig_lock);
248 
249 	nn->reconfig_sync_present = true;
250 
251 	if (nn->reconfig_timer_active) {
252 		del_timer(&nn->reconfig_timer);
253 		nn->reconfig_timer_active = false;
254 		cancelled_timer = true;
255 	}
256 	pre_posted_requests = nn->reconfig_posted;
257 	nn->reconfig_posted = 0;
258 
259 	spin_unlock_bh(&nn->reconfig_lock);
260 
261 	if (cancelled_timer)
262 		nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
263 
264 	/* Run the posted reconfigs which were issued before we started */
265 	if (pre_posted_requests) {
266 		nfp_net_reconfig_start(nn, pre_posted_requests);
267 		nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
268 	}
269 
270 	nfp_net_reconfig_start(nn, update);
271 	ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
272 
273 	spin_lock_bh(&nn->reconfig_lock);
274 
275 	if (nn->reconfig_posted)
276 		nfp_net_reconfig_start_async(nn, 0);
277 
278 	nn->reconfig_sync_present = false;
279 
280 	spin_unlock_bh(&nn->reconfig_lock);
281 
282 	return ret;
283 }
284 
285 /**
286  * nfp_net_reconfig_mbox() - Reconfigure the firmware via the mailbox
287  * @nn:        NFP Net device to reconfigure
288  * @mbox_cmd:  The value for the mailbox command
289  *
290  * Helper function for mailbox updates
291  *
292  * Return: Negative errno on error, 0 on success
293  */
294 static int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd)
295 {
296 	int ret;
297 
298 	nn_writeq(nn, NFP_NET_CFG_MBOX_CMD, mbox_cmd);
299 
300 	ret = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
301 	if (ret) {
302 		nn_err(nn, "Mailbox update error\n");
303 		return ret;
304 	}
305 
306 	return -nn_readl(nn, NFP_NET_CFG_MBOX_RET);
307 }
308 
309 /* Interrupt configuration and handling
310  */
311 
312 /**
313  * nfp_net_irq_unmask() - Unmask automasked interrupt
314  * @nn:       NFP Network structure
315  * @entry_nr: MSI-X table entry
316  *
317  * Clear the ICR for the IRQ entry.
318  */
319 static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
320 {
321 	nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
322 	nn_pci_flush(nn);
323 }
324 
325 /**
326  * nfp_net_irqs_alloc() - allocates MSI-X irqs
327  * @pdev:        PCI device structure
328  * @irq_entries: Array to be initialized and used to hold the irq entries
329  * @min_irqs:    Minimal acceptable number of interrupts
330  * @wanted_irqs: Target number of interrupts to allocate
331  *
332  * Return: Number of irqs obtained or 0 on error.
333  */
334 unsigned int
335 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
336 		   unsigned int min_irqs, unsigned int wanted_irqs)
337 {
338 	unsigned int i;
339 	int got_irqs;
340 
341 	for (i = 0; i < wanted_irqs; i++)
342 		irq_entries[i].entry = i;
343 
344 	got_irqs = pci_enable_msix_range(pdev, irq_entries,
345 					 min_irqs, wanted_irqs);
346 	if (got_irqs < 0) {
347 		dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
348 			min_irqs, wanted_irqs, got_irqs);
349 		return 0;
350 	}
351 
352 	if (got_irqs < wanted_irqs)
353 		dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
354 			 wanted_irqs, got_irqs);
355 
356 	return got_irqs;
357 }
358 
359 /**
360  * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
361  * @nn:		 NFP Network structure
362  * @irq_entries: Table of allocated interrupts
363  * @n:		 Size of @irq_entries (number of entries to grab)
364  *
365  * After interrupts are allocated with nfp_net_irqs_alloc() this function
366  * should be called to assign them to a specific netdev (port).
367  */
368 void
369 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
370 		    unsigned int n)
371 {
372 	struct nfp_net_dp *dp = &nn->dp;
373 
374 	nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
375 	dp->num_r_vecs = nn->max_r_vecs;
376 
377 	memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
378 
379 	if (dp->num_rx_rings > dp->num_r_vecs ||
380 	    dp->num_tx_rings > dp->num_r_vecs)
381 		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
382 			 dp->num_rx_rings, dp->num_tx_rings,
383 			 dp->num_r_vecs);
384 
385 	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
386 	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
387 	dp->num_stack_tx_rings = dp->num_tx_rings;
388 }
389 
390 /**
391  * nfp_net_irqs_disable() - Disable interrupts
392  * @pdev:        PCI device structure
393  *
394  * Undoes what @nfp_net_irqs_alloc() does.
395  */
396 void nfp_net_irqs_disable(struct pci_dev *pdev)
397 {
398 	pci_disable_msix(pdev);
399 }
400 
401 /**
402  * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
403  * @irq:      Interrupt
404  * @data:     Opaque data structure
405  *
406  * Return: Indicate if the interrupt has been handled.
407  */
408 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
409 {
410 	struct nfp_net_r_vector *r_vec = data;
411 
412 	napi_schedule_irqoff(&r_vec->napi);
413 
414 	/* The FW auto-masks any interrupt, either via the MASK bit in
415 	 * the MSI-X table or via the per entry ICR field.  So there
416 	 * is no need to disable interrupts here.
417 	 */
418 	return IRQ_HANDLED;
419 }
420 
421 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
422 {
423 	struct nfp_net_r_vector *r_vec = data;
424 
425 	tasklet_schedule(&r_vec->tasklet);
426 
427 	return IRQ_HANDLED;
428 }
429 
430 /**
431  * nfp_net_read_link_status() - Reread link status from control BAR
432  * @nn:       NFP Network structure
433  */
434 static void nfp_net_read_link_status(struct nfp_net *nn)
435 {
436 	unsigned long flags;
437 	bool link_up;
438 	u32 sts;
439 
440 	spin_lock_irqsave(&nn->link_status_lock, flags);
441 
442 	sts = nn_readl(nn, NFP_NET_CFG_STS);
443 	link_up = !!(sts & NFP_NET_CFG_STS_LINK);
444 
445 	if (nn->link_up == link_up)
446 		goto out;
447 
448 	nn->link_up = link_up;
449 	if (nn->port)
450 		set_bit(NFP_PORT_CHANGED, &nn->port->flags);
451 
452 	if (nn->link_up) {
453 		netif_carrier_on(nn->dp.netdev);
454 		netdev_info(nn->dp.netdev, "NIC Link is Up\n");
455 	} else {
456 		netif_carrier_off(nn->dp.netdev);
457 		netdev_info(nn->dp.netdev, "NIC Link is Down\n");
458 	}
459 out:
460 	spin_unlock_irqrestore(&nn->link_status_lock, flags);
461 }
462 
463 /**
464  * nfp_net_irq_lsc() - Interrupt service routine for link state changes
465  * @irq:      Interrupt
466  * @data:     Opaque data structure
467  *
468  * Return: Indicate if the interrupt has been handled.
469  */
470 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
471 {
472 	struct nfp_net *nn = data;
473 	struct msix_entry *entry;
474 
475 	entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
476 
477 	nfp_net_read_link_status(nn);
478 
479 	nfp_net_irq_unmask(nn, entry->entry);
480 
481 	return IRQ_HANDLED;
482 }
483 
484 /**
485  * nfp_net_irq_exn() - Interrupt service routine for exceptions
486  * @irq:      Interrupt
487  * @data:     Opaque data structure
488  *
489  * Return: Indicate if the interrupt has been handled.
490  */
491 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
492 {
493 	struct nfp_net *nn = data;
494 
495 	nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
496 	/* XXX TO BE IMPLEMENTED */
497 	return IRQ_HANDLED;
498 }
499 
500 /**
501  * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
502  * @tx_ring:  TX ring structure
503  * @r_vec:    IRQ vector servicing this ring
504  * @idx:      Ring index
505  * @is_xdp:   Is this an XDP TX ring?
506  */
507 static void
508 nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
509 		     struct nfp_net_r_vector *r_vec, unsigned int idx,
510 		     bool is_xdp)
511 {
512 	struct nfp_net *nn = r_vec->nfp_net;
513 
514 	tx_ring->idx = idx;
515 	tx_ring->r_vec = r_vec;
516 	tx_ring->is_xdp = is_xdp;
517 	u64_stats_init(&tx_ring->r_vec->tx_sync);
518 
519 	tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
520 	tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
521 }
522 
523 /**
524  * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
525  * @rx_ring:  RX ring structure
526  * @r_vec:    IRQ vector servicing this ring
527  * @idx:      Ring index
528  */
529 static void
530 nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
531 		     struct nfp_net_r_vector *r_vec, unsigned int idx)
532 {
533 	struct nfp_net *nn = r_vec->nfp_net;
534 
535 	rx_ring->idx = idx;
536 	rx_ring->r_vec = r_vec;
537 	u64_stats_init(&rx_ring->r_vec->rx_sync);
538 
539 	rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
540 	rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
541 }
542 
543 /**
544  * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
545  * @nn:		NFP Network structure
546  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
547  * @format:	printf-style format to construct the interrupt name
548  * @name:	Pointer to allocated space for interrupt name
549  * @name_sz:	Size of space for interrupt name
550  * @vector_idx:	Index of MSI-X vector used for this interrupt
551  * @handler:	IRQ handler to register for this interrupt
552  */
553 static int
554 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
555 			const char *format, char *name, size_t name_sz,
556 			unsigned int vector_idx, irq_handler_t handler)
557 {
558 	struct msix_entry *entry;
559 	int err;
560 
561 	entry = &nn->irq_entries[vector_idx];
562 
563 	snprintf(name, name_sz, format, nfp_net_name(nn));
564 	err = request_irq(entry->vector, handler, 0, name, nn);
565 	if (err) {
566 		nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
567 		       entry->vector, err);
568 		return err;
569 	}
570 	nn_writeb(nn, ctrl_offset, entry->entry);
571 
572 	return 0;
573 }
574 
575 /**
576  * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
577  * @nn:		NFP Network structure
578  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
579  * @vector_idx:	Index of MSI-X vector used for this interrupt
580  */
581 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
582 				 unsigned int vector_idx)
583 {
584 	nn_writeb(nn, ctrl_offset, 0xff);
585 	free_irq(nn->irq_entries[vector_idx].vector, nn);
586 }
587 
588 /* Transmit
589  *
590  * One queue controller peripheral queue is used for transmit.  The
591  * driver en-queues packets for transmit by advancing the write
592  * pointer.  The device indicates that packets have transmitted by
593  * advancing the read pointer.  The driver maintains a local copy of
594  * the read and write pointer in @struct nfp_net_tx_ring.  The driver
595  * keeps @wr_p in sync with the queue controller write pointer and can
596  * determine how many packets have been transmitted by comparing its
597  * copy of the read pointer @rd_p with the read pointer maintained by
598  * the queue controller peripheral.
599  */
600 
601 /**
602  * nfp_net_tx_full() - Check if the TX ring is full
603  * @tx_ring: TX ring to check
604  * @dcnt:    Number of descriptors that need to be enqueued (must be >= 1)
605  *
606  * This function checks, based on the *host copy* of read/write
607  * pointer if a given TX ring is full.  The real TX queue may have
608  * some newly made available slots.
609  *
610  * Return: True if the ring is full.
611  */
612 static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
613 {
614 	return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
615 }
616 
617 /* Wrappers for deciding when to stop and restart TX queues */
618 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
619 {
620 	return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
621 }
622 
623 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
624 {
625 	return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
626 }
627 
628 /**
629  * nfp_net_tx_ring_stop() - stop tx ring
630  * @nd_q:    netdev queue
631  * @tx_ring: driver tx queue structure
632  *
633  * Safely stop TX ring.  Remember that while we are running .start_xmit()
634  * someone else may be cleaning the TX ring completions so we need to be
635  * extra careful here.
636  */
637 static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
638 				 struct nfp_net_tx_ring *tx_ring)
639 {
640 	netif_tx_stop_queue(nd_q);
641 
642 	/* We can race with the TX completion out of NAPI so recheck */
643 	smp_mb();
644 	if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
645 		netif_tx_start_queue(nd_q);
646 }
647 
648 /**
649  * nfp_net_tx_tso() - Set up Tx descriptor for LSO
650  * @r_vec: per-ring structure
651  * @txbuf: Pointer to driver soft TX descriptor
652  * @txd: Pointer to HW TX descriptor
653  * @skb: Pointer to SKB
654  *
655  * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
656  * Return error on packet header greater than maximum supported LSO header size.
657  */
658 static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
659 			   struct nfp_net_tx_buf *txbuf,
660 			   struct nfp_net_tx_desc *txd, struct sk_buff *skb)
661 {
662 	u32 hdrlen;
663 	u16 mss;
664 
665 	if (!skb_is_gso(skb))
666 		return;
667 
668 	if (!skb->encapsulation) {
669 		txd->l3_offset = skb_network_offset(skb);
670 		txd->l4_offset = skb_transport_offset(skb);
671 		hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
672 	} else {
673 		txd->l3_offset = skb_inner_network_offset(skb);
674 		txd->l4_offset = skb_inner_transport_offset(skb);
675 		hdrlen = skb_inner_transport_header(skb) - skb->data +
676 			inner_tcp_hdrlen(skb);
677 	}
678 
679 	txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
680 	txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
681 
682 	mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
683 	txd->lso_hdrlen = hdrlen;
684 	txd->mss = cpu_to_le16(mss);
685 	txd->flags |= PCIE_DESC_TX_LSO;
686 
687 	u64_stats_update_begin(&r_vec->tx_sync);
688 	r_vec->tx_lso++;
689 	u64_stats_update_end(&r_vec->tx_sync);
690 }
691 
692 /**
693  * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
694  * @dp:  NFP Net data path struct
695  * @r_vec: per-ring structure
696  * @txbuf: Pointer to driver soft TX descriptor
697  * @txd: Pointer to TX descriptor
698  * @skb: Pointer to SKB
699  *
700  * This function sets the TX checksum flags in the TX descriptor based
701  * on the configuration and the protocol of the packet to be transmitted.
702  */
703 static void nfp_net_tx_csum(struct nfp_net_dp *dp,
704 			    struct nfp_net_r_vector *r_vec,
705 			    struct nfp_net_tx_buf *txbuf,
706 			    struct nfp_net_tx_desc *txd, struct sk_buff *skb)
707 {
708 	struct ipv6hdr *ipv6h;
709 	struct iphdr *iph;
710 	u8 l4_hdr;
711 
712 	if (!(dp->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
713 		return;
714 
715 	if (skb->ip_summed != CHECKSUM_PARTIAL)
716 		return;
717 
718 	txd->flags |= PCIE_DESC_TX_CSUM;
719 	if (skb->encapsulation)
720 		txd->flags |= PCIE_DESC_TX_ENCAP;
721 
722 	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
723 	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
724 
725 	if (iph->version == 4) {
726 		txd->flags |= PCIE_DESC_TX_IP4_CSUM;
727 		l4_hdr = iph->protocol;
728 	} else if (ipv6h->version == 6) {
729 		l4_hdr = ipv6h->nexthdr;
730 	} else {
731 		nn_dp_warn(dp, "partial checksum but ipv=%x!\n", iph->version);
732 		return;
733 	}
734 
735 	switch (l4_hdr) {
736 	case IPPROTO_TCP:
737 		txd->flags |= PCIE_DESC_TX_TCP_CSUM;
738 		break;
739 	case IPPROTO_UDP:
740 		txd->flags |= PCIE_DESC_TX_UDP_CSUM;
741 		break;
742 	default:
743 		nn_dp_warn(dp, "partial checksum but l4 proto=%x!\n", l4_hdr);
744 		return;
745 	}
746 
747 	u64_stats_update_begin(&r_vec->tx_sync);
748 	if (skb->encapsulation)
749 		r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
750 	else
751 		r_vec->hw_csum_tx += txbuf->pkt_cnt;
752 	u64_stats_update_end(&r_vec->tx_sync);
753 }
754 
755 static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
756 {
757 	wmb();
758 	nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
759 	tx_ring->wr_ptr_add = 0;
760 }
761 
762 static int nfp_net_prep_port_id(struct sk_buff *skb)
763 {
764 	struct metadata_dst *md_dst = skb_metadata_dst(skb);
765 	unsigned char *data;
766 
767 	if (likely(!md_dst))
768 		return 0;
769 	if (unlikely(md_dst->type != METADATA_HW_PORT_MUX))
770 		return 0;
771 
772 	if (unlikely(skb_cow_head(skb, 8)))
773 		return -ENOMEM;
774 
775 	data = skb_push(skb, 8);
776 	put_unaligned_be32(NFP_NET_META_PORTID, data);
777 	put_unaligned_be32(md_dst->u.port_info.port_id, data + 4);
778 
779 	return 8;
780 }
781 
782 /**
783  * nfp_net_tx() - Main transmit entry point
784  * @skb:    SKB to transmit
785  * @netdev: netdev structure
786  *
787  * Return: NETDEV_TX_OK on success.
788  */
789 static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
790 {
791 	struct nfp_net *nn = netdev_priv(netdev);
792 	const struct skb_frag_struct *frag;
793 	struct nfp_net_tx_desc *txd, txdg;
794 	int f, nr_frags, wr_idx, md_bytes;
795 	struct nfp_net_tx_ring *tx_ring;
796 	struct nfp_net_r_vector *r_vec;
797 	struct nfp_net_tx_buf *txbuf;
798 	struct netdev_queue *nd_q;
799 	struct nfp_net_dp *dp;
800 	dma_addr_t dma_addr;
801 	unsigned int fsize;
802 	u16 qidx;
803 
804 	dp = &nn->dp;
805 	qidx = skb_get_queue_mapping(skb);
806 	tx_ring = &dp->tx_rings[qidx];
807 	r_vec = tx_ring->r_vec;
808 	nd_q = netdev_get_tx_queue(dp->netdev, qidx);
809 
810 	nr_frags = skb_shinfo(skb)->nr_frags;
811 
812 	if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
813 		nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
814 			   qidx, tx_ring->wr_p, tx_ring->rd_p);
815 		netif_tx_stop_queue(nd_q);
816 		nfp_net_tx_xmit_more_flush(tx_ring);
817 		u64_stats_update_begin(&r_vec->tx_sync);
818 		r_vec->tx_busy++;
819 		u64_stats_update_end(&r_vec->tx_sync);
820 		return NETDEV_TX_BUSY;
821 	}
822 
823 	md_bytes = nfp_net_prep_port_id(skb);
824 	if (unlikely(md_bytes < 0)) {
825 		nfp_net_tx_xmit_more_flush(tx_ring);
826 		dev_kfree_skb_any(skb);
827 		return NETDEV_TX_OK;
828 	}
829 
830 	/* Start with the head skbuf */
831 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
832 				  DMA_TO_DEVICE);
833 	if (dma_mapping_error(dp->dev, dma_addr))
834 		goto err_free;
835 
836 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
837 
838 	/* Stash the soft descriptor of the head then initialize it */
839 	txbuf = &tx_ring->txbufs[wr_idx];
840 	txbuf->skb = skb;
841 	txbuf->dma_addr = dma_addr;
842 	txbuf->fidx = -1;
843 	txbuf->pkt_cnt = 1;
844 	txbuf->real_len = skb->len;
845 
846 	/* Build TX descriptor */
847 	txd = &tx_ring->txds[wr_idx];
848 	txd->offset_eop = (nr_frags ? 0 : PCIE_DESC_TX_EOP) | md_bytes;
849 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
850 	nfp_desc_set_dma_addr(txd, dma_addr);
851 	txd->data_len = cpu_to_le16(skb->len);
852 
853 	txd->flags = 0;
854 	txd->mss = 0;
855 	txd->lso_hdrlen = 0;
856 
857 	/* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
858 	nfp_net_tx_tso(r_vec, txbuf, txd, skb);
859 	nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
860 	if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
861 		txd->flags |= PCIE_DESC_TX_VLAN;
862 		txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
863 	}
864 
865 	/* Gather DMA */
866 	if (nr_frags > 0) {
867 		/* all descs must match except for in addr, length and eop */
868 		txdg = *txd;
869 
870 		for (f = 0; f < nr_frags; f++) {
871 			frag = &skb_shinfo(skb)->frags[f];
872 			fsize = skb_frag_size(frag);
873 
874 			dma_addr = skb_frag_dma_map(dp->dev, frag, 0,
875 						    fsize, DMA_TO_DEVICE);
876 			if (dma_mapping_error(dp->dev, dma_addr))
877 				goto err_unmap;
878 
879 			wr_idx = D_IDX(tx_ring, wr_idx + 1);
880 			tx_ring->txbufs[wr_idx].skb = skb;
881 			tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
882 			tx_ring->txbufs[wr_idx].fidx = f;
883 
884 			txd = &tx_ring->txds[wr_idx];
885 			*txd = txdg;
886 			txd->dma_len = cpu_to_le16(fsize);
887 			nfp_desc_set_dma_addr(txd, dma_addr);
888 			txd->offset_eop |=
889 				(f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
890 		}
891 
892 		u64_stats_update_begin(&r_vec->tx_sync);
893 		r_vec->tx_gather++;
894 		u64_stats_update_end(&r_vec->tx_sync);
895 	}
896 
897 	netdev_tx_sent_queue(nd_q, txbuf->real_len);
898 
899 	skb_tx_timestamp(skb);
900 
901 	tx_ring->wr_p += nr_frags + 1;
902 	if (nfp_net_tx_ring_should_stop(tx_ring))
903 		nfp_net_tx_ring_stop(nd_q, tx_ring);
904 
905 	tx_ring->wr_ptr_add += nr_frags + 1;
906 	if (!skb->xmit_more || netif_xmit_stopped(nd_q))
907 		nfp_net_tx_xmit_more_flush(tx_ring);
908 
909 	return NETDEV_TX_OK;
910 
911 err_unmap:
912 	while (--f >= 0) {
913 		frag = &skb_shinfo(skb)->frags[f];
914 		dma_unmap_page(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
915 			       skb_frag_size(frag), DMA_TO_DEVICE);
916 		tx_ring->txbufs[wr_idx].skb = NULL;
917 		tx_ring->txbufs[wr_idx].dma_addr = 0;
918 		tx_ring->txbufs[wr_idx].fidx = -2;
919 		wr_idx = wr_idx - 1;
920 		if (wr_idx < 0)
921 			wr_idx += tx_ring->cnt;
922 	}
923 	dma_unmap_single(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
924 			 skb_headlen(skb), DMA_TO_DEVICE);
925 	tx_ring->txbufs[wr_idx].skb = NULL;
926 	tx_ring->txbufs[wr_idx].dma_addr = 0;
927 	tx_ring->txbufs[wr_idx].fidx = -2;
928 err_free:
929 	nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
930 	nfp_net_tx_xmit_more_flush(tx_ring);
931 	u64_stats_update_begin(&r_vec->tx_sync);
932 	r_vec->tx_errors++;
933 	u64_stats_update_end(&r_vec->tx_sync);
934 	dev_kfree_skb_any(skb);
935 	return NETDEV_TX_OK;
936 }
937 
938 /**
939  * nfp_net_tx_complete() - Handled completed TX packets
940  * @tx_ring:   TX ring structure
941  *
942  * Return: Number of completed TX descriptors
943  */
944 static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring)
945 {
946 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
947 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
948 	const struct skb_frag_struct *frag;
949 	struct netdev_queue *nd_q;
950 	u32 done_pkts = 0, done_bytes = 0;
951 	struct sk_buff *skb;
952 	int todo, nr_frags;
953 	u32 qcp_rd_p;
954 	int fidx;
955 	int idx;
956 
957 	if (tx_ring->wr_p == tx_ring->rd_p)
958 		return;
959 
960 	/* Work out how many descriptors have been transmitted */
961 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
962 
963 	if (qcp_rd_p == tx_ring->qcp_rd_p)
964 		return;
965 
966 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
967 
968 	while (todo--) {
969 		idx = D_IDX(tx_ring, tx_ring->rd_p++);
970 
971 		skb = tx_ring->txbufs[idx].skb;
972 		if (!skb)
973 			continue;
974 
975 		nr_frags = skb_shinfo(skb)->nr_frags;
976 		fidx = tx_ring->txbufs[idx].fidx;
977 
978 		if (fidx == -1) {
979 			/* unmap head */
980 			dma_unmap_single(dp->dev, tx_ring->txbufs[idx].dma_addr,
981 					 skb_headlen(skb), DMA_TO_DEVICE);
982 
983 			done_pkts += tx_ring->txbufs[idx].pkt_cnt;
984 			done_bytes += tx_ring->txbufs[idx].real_len;
985 		} else {
986 			/* unmap fragment */
987 			frag = &skb_shinfo(skb)->frags[fidx];
988 			dma_unmap_page(dp->dev, tx_ring->txbufs[idx].dma_addr,
989 				       skb_frag_size(frag), DMA_TO_DEVICE);
990 		}
991 
992 		/* check for last gather fragment */
993 		if (fidx == nr_frags - 1)
994 			dev_consume_skb_any(skb);
995 
996 		tx_ring->txbufs[idx].dma_addr = 0;
997 		tx_ring->txbufs[idx].skb = NULL;
998 		tx_ring->txbufs[idx].fidx = -2;
999 	}
1000 
1001 	tx_ring->qcp_rd_p = qcp_rd_p;
1002 
1003 	u64_stats_update_begin(&r_vec->tx_sync);
1004 	r_vec->tx_bytes += done_bytes;
1005 	r_vec->tx_pkts += done_pkts;
1006 	u64_stats_update_end(&r_vec->tx_sync);
1007 
1008 	if (!dp->netdev)
1009 		return;
1010 
1011 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1012 	netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
1013 	if (nfp_net_tx_ring_should_wake(tx_ring)) {
1014 		/* Make sure TX thread will see updated tx_ring->rd_p */
1015 		smp_mb();
1016 
1017 		if (unlikely(netif_tx_queue_stopped(nd_q)))
1018 			netif_tx_wake_queue(nd_q);
1019 	}
1020 
1021 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1022 		  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1023 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1024 }
1025 
1026 static bool nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
1027 {
1028 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1029 	u32 done_pkts = 0, done_bytes = 0;
1030 	bool done_all;
1031 	int idx, todo;
1032 	u32 qcp_rd_p;
1033 
1034 	/* Work out how many descriptors have been transmitted */
1035 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1036 
1037 	if (qcp_rd_p == tx_ring->qcp_rd_p)
1038 		return true;
1039 
1040 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1041 
1042 	done_all = todo <= NFP_NET_XDP_MAX_COMPLETE;
1043 	todo = min(todo, NFP_NET_XDP_MAX_COMPLETE);
1044 
1045 	tx_ring->qcp_rd_p = D_IDX(tx_ring, tx_ring->qcp_rd_p + todo);
1046 
1047 	done_pkts = todo;
1048 	while (todo--) {
1049 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1050 		tx_ring->rd_p++;
1051 
1052 		done_bytes += tx_ring->txbufs[idx].real_len;
1053 	}
1054 
1055 	u64_stats_update_begin(&r_vec->tx_sync);
1056 	r_vec->tx_bytes += done_bytes;
1057 	r_vec->tx_pkts += done_pkts;
1058 	u64_stats_update_end(&r_vec->tx_sync);
1059 
1060 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1061 		  "XDP TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1062 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1063 
1064 	return done_all;
1065 }
1066 
1067 /**
1068  * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
1069  * @dp:		NFP Net data path struct
1070  * @tx_ring:	TX ring structure
1071  *
1072  * Assumes that the device is stopped
1073  */
1074 static void
1075 nfp_net_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
1076 {
1077 	const struct skb_frag_struct *frag;
1078 	struct netdev_queue *nd_q;
1079 
1080 	while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
1081 		struct nfp_net_tx_buf *tx_buf;
1082 		struct sk_buff *skb;
1083 		int idx, nr_frags;
1084 
1085 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1086 		tx_buf = &tx_ring->txbufs[idx];
1087 
1088 		skb = tx_ring->txbufs[idx].skb;
1089 		nr_frags = skb_shinfo(skb)->nr_frags;
1090 
1091 		if (tx_buf->fidx == -1) {
1092 			/* unmap head */
1093 			dma_unmap_single(dp->dev, tx_buf->dma_addr,
1094 					 skb_headlen(skb), DMA_TO_DEVICE);
1095 		} else {
1096 			/* unmap fragment */
1097 			frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1098 			dma_unmap_page(dp->dev, tx_buf->dma_addr,
1099 				       skb_frag_size(frag), DMA_TO_DEVICE);
1100 		}
1101 
1102 		/* check for last gather fragment */
1103 		if (tx_buf->fidx == nr_frags - 1)
1104 			dev_kfree_skb_any(skb);
1105 
1106 		tx_buf->dma_addr = 0;
1107 		tx_buf->skb = NULL;
1108 		tx_buf->fidx = -2;
1109 
1110 		tx_ring->qcp_rd_p++;
1111 		tx_ring->rd_p++;
1112 	}
1113 
1114 	memset(tx_ring->txds, 0, sizeof(*tx_ring->txds) * tx_ring->cnt);
1115 	tx_ring->wr_p = 0;
1116 	tx_ring->rd_p = 0;
1117 	tx_ring->qcp_rd_p = 0;
1118 	tx_ring->wr_ptr_add = 0;
1119 
1120 	if (tx_ring->is_xdp || !dp->netdev)
1121 		return;
1122 
1123 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1124 	netdev_tx_reset_queue(nd_q);
1125 }
1126 
1127 static void nfp_net_tx_timeout(struct net_device *netdev)
1128 {
1129 	struct nfp_net *nn = netdev_priv(netdev);
1130 	int i;
1131 
1132 	for (i = 0; i < nn->dp.netdev->real_num_tx_queues; i++) {
1133 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i)))
1134 			continue;
1135 		nn_warn(nn, "TX timeout on ring: %d\n", i);
1136 	}
1137 	nn_warn(nn, "TX watchdog timeout\n");
1138 }
1139 
1140 /* Receive processing
1141  */
1142 static unsigned int
1143 nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
1144 {
1145 	unsigned int fl_bufsz;
1146 
1147 	fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
1148 	fl_bufsz += dp->rx_dma_off;
1149 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1150 		fl_bufsz += NFP_NET_MAX_PREPEND;
1151 	else
1152 		fl_bufsz += dp->rx_offset;
1153 	fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
1154 
1155 	fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1156 	fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1157 
1158 	return fl_bufsz;
1159 }
1160 
1161 static void
1162 nfp_net_free_frag(void *frag, bool xdp)
1163 {
1164 	if (!xdp)
1165 		skb_free_frag(frag);
1166 	else
1167 		__free_page(virt_to_page(frag));
1168 }
1169 
1170 /**
1171  * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
1172  * @dp:		NFP Net data path struct
1173  * @dma_addr:	Pointer to storage for DMA address (output param)
1174  *
1175  * This function will allcate a new page frag, map it for DMA.
1176  *
1177  * Return: allocated page frag or NULL on failure.
1178  */
1179 static void *nfp_net_rx_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1180 {
1181 	void *frag;
1182 
1183 	if (!dp->xdp_prog) {
1184 		frag = netdev_alloc_frag(dp->fl_bufsz);
1185 	} else {
1186 		struct page *page;
1187 
1188 		page = alloc_page(GFP_KERNEL | __GFP_COLD);
1189 		frag = page ? page_address(page) : NULL;
1190 	}
1191 	if (!frag) {
1192 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1193 		return NULL;
1194 	}
1195 
1196 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1197 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1198 		nfp_net_free_frag(frag, dp->xdp_prog);
1199 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1200 		return NULL;
1201 	}
1202 
1203 	return frag;
1204 }
1205 
1206 static void *nfp_net_napi_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1207 {
1208 	void *frag;
1209 
1210 	if (!dp->xdp_prog) {
1211 		frag = napi_alloc_frag(dp->fl_bufsz);
1212 	} else {
1213 		struct page *page;
1214 
1215 		page = alloc_page(GFP_ATOMIC | __GFP_COLD);
1216 		frag = page ? page_address(page) : NULL;
1217 	}
1218 	if (!frag) {
1219 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1220 		return NULL;
1221 	}
1222 
1223 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1224 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1225 		nfp_net_free_frag(frag, dp->xdp_prog);
1226 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1227 		return NULL;
1228 	}
1229 
1230 	return frag;
1231 }
1232 
1233 /**
1234  * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1235  * @dp:		NFP Net data path struct
1236  * @rx_ring:	RX ring structure
1237  * @frag:	page fragment buffer
1238  * @dma_addr:	DMA address of skb mapping
1239  */
1240 static void nfp_net_rx_give_one(const struct nfp_net_dp *dp,
1241 				struct nfp_net_rx_ring *rx_ring,
1242 				void *frag, dma_addr_t dma_addr)
1243 {
1244 	unsigned int wr_idx;
1245 
1246 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1247 
1248 	nfp_net_dma_sync_dev_rx(dp, dma_addr);
1249 
1250 	/* Stash SKB and DMA address away */
1251 	rx_ring->rxbufs[wr_idx].frag = frag;
1252 	rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1253 
1254 	/* Fill freelist descriptor */
1255 	rx_ring->rxds[wr_idx].fld.reserved = 0;
1256 	rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1257 	nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld,
1258 			      dma_addr + dp->rx_dma_off);
1259 
1260 	rx_ring->wr_p++;
1261 	if (!(rx_ring->wr_p % NFP_NET_FL_BATCH)) {
1262 		/* Update write pointer of the freelist queue. Make
1263 		 * sure all writes are flushed before telling the hardware.
1264 		 */
1265 		wmb();
1266 		nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, NFP_NET_FL_BATCH);
1267 	}
1268 }
1269 
1270 /**
1271  * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1272  * @rx_ring:	RX ring structure
1273  *
1274  * Warning: Do *not* call if ring buffers were never put on the FW freelist
1275  *	    (i.e. device was not enabled)!
1276  */
1277 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1278 {
1279 	unsigned int wr_idx, last_idx;
1280 
1281 	/* Move the empty entry to the end of the list */
1282 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1283 	last_idx = rx_ring->cnt - 1;
1284 	rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1285 	rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1286 	rx_ring->rxbufs[last_idx].dma_addr = 0;
1287 	rx_ring->rxbufs[last_idx].frag = NULL;
1288 
1289 	memset(rx_ring->rxds, 0, sizeof(*rx_ring->rxds) * rx_ring->cnt);
1290 	rx_ring->wr_p = 0;
1291 	rx_ring->rd_p = 0;
1292 }
1293 
1294 /**
1295  * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1296  * @dp:		NFP Net data path struct
1297  * @rx_ring:	RX ring to remove buffers from
1298  *
1299  * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1300  * entries.  After device is disabled nfp_net_rx_ring_reset() must be called
1301  * to restore required ring geometry.
1302  */
1303 static void
1304 nfp_net_rx_ring_bufs_free(struct nfp_net_dp *dp,
1305 			  struct nfp_net_rx_ring *rx_ring)
1306 {
1307 	unsigned int i;
1308 
1309 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1310 		/* NULL skb can only happen when initial filling of the ring
1311 		 * fails to allocate enough buffers and calls here to free
1312 		 * already allocated ones.
1313 		 */
1314 		if (!rx_ring->rxbufs[i].frag)
1315 			continue;
1316 
1317 		nfp_net_dma_unmap_rx(dp, rx_ring->rxbufs[i].dma_addr);
1318 		nfp_net_free_frag(rx_ring->rxbufs[i].frag, dp->xdp_prog);
1319 		rx_ring->rxbufs[i].dma_addr = 0;
1320 		rx_ring->rxbufs[i].frag = NULL;
1321 	}
1322 }
1323 
1324 /**
1325  * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1326  * @dp:		NFP Net data path struct
1327  * @rx_ring:	RX ring to remove buffers from
1328  */
1329 static int
1330 nfp_net_rx_ring_bufs_alloc(struct nfp_net_dp *dp,
1331 			   struct nfp_net_rx_ring *rx_ring)
1332 {
1333 	struct nfp_net_rx_buf *rxbufs;
1334 	unsigned int i;
1335 
1336 	rxbufs = rx_ring->rxbufs;
1337 
1338 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1339 		rxbufs[i].frag = nfp_net_rx_alloc_one(dp, &rxbufs[i].dma_addr);
1340 		if (!rxbufs[i].frag) {
1341 			nfp_net_rx_ring_bufs_free(dp, rx_ring);
1342 			return -ENOMEM;
1343 		}
1344 	}
1345 
1346 	return 0;
1347 }
1348 
1349 /**
1350  * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1351  * @dp:	     NFP Net data path struct
1352  * @rx_ring: RX ring to fill
1353  */
1354 static void
1355 nfp_net_rx_ring_fill_freelist(struct nfp_net_dp *dp,
1356 			      struct nfp_net_rx_ring *rx_ring)
1357 {
1358 	unsigned int i;
1359 
1360 	for (i = 0; i < rx_ring->cnt - 1; i++)
1361 		nfp_net_rx_give_one(dp, rx_ring, rx_ring->rxbufs[i].frag,
1362 				    rx_ring->rxbufs[i].dma_addr);
1363 }
1364 
1365 /**
1366  * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1367  * @flags: RX descriptor flags field in CPU byte order
1368  */
1369 static int nfp_net_rx_csum_has_errors(u16 flags)
1370 {
1371 	u16 csum_all_checked, csum_all_ok;
1372 
1373 	csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1374 	csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1375 
1376 	return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1377 }
1378 
1379 /**
1380  * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1381  * @dp:  NFP Net data path struct
1382  * @r_vec: per-ring structure
1383  * @rxd: Pointer to RX descriptor
1384  * @meta: Parsed metadata prepend
1385  * @skb: Pointer to SKB
1386  */
1387 static void nfp_net_rx_csum(struct nfp_net_dp *dp,
1388 			    struct nfp_net_r_vector *r_vec,
1389 			    struct nfp_net_rx_desc *rxd,
1390 			    struct nfp_meta_parsed *meta, struct sk_buff *skb)
1391 {
1392 	skb_checksum_none_assert(skb);
1393 
1394 	if (!(dp->netdev->features & NETIF_F_RXCSUM))
1395 		return;
1396 
1397 	if (meta->csum_type) {
1398 		skb->ip_summed = meta->csum_type;
1399 		skb->csum = meta->csum;
1400 		u64_stats_update_begin(&r_vec->rx_sync);
1401 		r_vec->hw_csum_rx_ok++;
1402 		u64_stats_update_end(&r_vec->rx_sync);
1403 		return;
1404 	}
1405 
1406 	if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1407 		u64_stats_update_begin(&r_vec->rx_sync);
1408 		r_vec->hw_csum_rx_error++;
1409 		u64_stats_update_end(&r_vec->rx_sync);
1410 		return;
1411 	}
1412 
1413 	/* Assume that the firmware will never report inner CSUM_OK unless outer
1414 	 * L4 headers were successfully parsed. FW will always report zero UDP
1415 	 * checksum as CSUM_OK.
1416 	 */
1417 	if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1418 	    rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1419 		__skb_incr_checksum_unnecessary(skb);
1420 		u64_stats_update_begin(&r_vec->rx_sync);
1421 		r_vec->hw_csum_rx_ok++;
1422 		u64_stats_update_end(&r_vec->rx_sync);
1423 	}
1424 
1425 	if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1426 	    rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1427 		__skb_incr_checksum_unnecessary(skb);
1428 		u64_stats_update_begin(&r_vec->rx_sync);
1429 		r_vec->hw_csum_rx_inner_ok++;
1430 		u64_stats_update_end(&r_vec->rx_sync);
1431 	}
1432 }
1433 
1434 static void
1435 nfp_net_set_hash(struct net_device *netdev, struct nfp_meta_parsed *meta,
1436 		 unsigned int type, __be32 *hash)
1437 {
1438 	if (!(netdev->features & NETIF_F_RXHASH))
1439 		return;
1440 
1441 	switch (type) {
1442 	case NFP_NET_RSS_IPV4:
1443 	case NFP_NET_RSS_IPV6:
1444 	case NFP_NET_RSS_IPV6_EX:
1445 		meta->hash_type = PKT_HASH_TYPE_L3;
1446 		break;
1447 	default:
1448 		meta->hash_type = PKT_HASH_TYPE_L4;
1449 		break;
1450 	}
1451 
1452 	meta->hash = get_unaligned_be32(hash);
1453 }
1454 
1455 static void
1456 nfp_net_set_hash_desc(struct net_device *netdev, struct nfp_meta_parsed *meta,
1457 		      void *data, struct nfp_net_rx_desc *rxd)
1458 {
1459 	struct nfp_net_rx_hash *rx_hash = data;
1460 
1461 	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1462 		return;
1463 
1464 	nfp_net_set_hash(netdev, meta, get_unaligned_be32(&rx_hash->hash_type),
1465 			 &rx_hash->hash);
1466 }
1467 
1468 static void *
1469 nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
1470 		   void *data, int meta_len)
1471 {
1472 	u32 meta_info;
1473 
1474 	meta_info = get_unaligned_be32(data);
1475 	data += 4;
1476 
1477 	while (meta_info) {
1478 		switch (meta_info & NFP_NET_META_FIELD_MASK) {
1479 		case NFP_NET_META_HASH:
1480 			meta_info >>= NFP_NET_META_FIELD_SIZE;
1481 			nfp_net_set_hash(netdev, meta,
1482 					 meta_info & NFP_NET_META_FIELD_MASK,
1483 					 (__be32 *)data);
1484 			data += 4;
1485 			break;
1486 		case NFP_NET_META_MARK:
1487 			meta->mark = get_unaligned_be32(data);
1488 			data += 4;
1489 			break;
1490 		case NFP_NET_META_PORTID:
1491 			meta->portid = get_unaligned_be32(data);
1492 			data += 4;
1493 			break;
1494 		case NFP_NET_META_CSUM:
1495 			meta->csum_type = CHECKSUM_COMPLETE;
1496 			meta->csum =
1497 				(__force __wsum)__get_unaligned_cpu32(data);
1498 			data += 4;
1499 			break;
1500 		default:
1501 			return NULL;
1502 		}
1503 
1504 		meta_info >>= NFP_NET_META_FIELD_SIZE;
1505 	}
1506 
1507 	return data;
1508 }
1509 
1510 static void
1511 nfp_net_rx_drop(const struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
1512 		struct nfp_net_rx_ring *rx_ring, struct nfp_net_rx_buf *rxbuf,
1513 		struct sk_buff *skb)
1514 {
1515 	u64_stats_update_begin(&r_vec->rx_sync);
1516 	r_vec->rx_drops++;
1517 	u64_stats_update_end(&r_vec->rx_sync);
1518 
1519 	/* skb is build based on the frag, free_skb() would free the frag
1520 	 * so to be able to reuse it we need an extra ref.
1521 	 */
1522 	if (skb && rxbuf && skb->head == rxbuf->frag)
1523 		page_ref_inc(virt_to_head_page(rxbuf->frag));
1524 	if (rxbuf)
1525 		nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag, rxbuf->dma_addr);
1526 	if (skb)
1527 		dev_kfree_skb_any(skb);
1528 }
1529 
1530 static bool
1531 nfp_net_tx_xdp_buf(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring,
1532 		   struct nfp_net_tx_ring *tx_ring,
1533 		   struct nfp_net_rx_buf *rxbuf, unsigned int dma_off,
1534 		   unsigned int pkt_len, bool *completed)
1535 {
1536 	struct nfp_net_tx_buf *txbuf;
1537 	struct nfp_net_tx_desc *txd;
1538 	int wr_idx;
1539 
1540 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1541 		if (!*completed) {
1542 			nfp_net_xdp_complete(tx_ring);
1543 			*completed = true;
1544 		}
1545 
1546 		if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1547 			nfp_net_rx_drop(dp, rx_ring->r_vec, rx_ring, rxbuf,
1548 					NULL);
1549 			return false;
1550 		}
1551 	}
1552 
1553 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1554 
1555 	/* Stash the soft descriptor of the head then initialize it */
1556 	txbuf = &tx_ring->txbufs[wr_idx];
1557 
1558 	nfp_net_rx_give_one(dp, rx_ring, txbuf->frag, txbuf->dma_addr);
1559 
1560 	txbuf->frag = rxbuf->frag;
1561 	txbuf->dma_addr = rxbuf->dma_addr;
1562 	txbuf->fidx = -1;
1563 	txbuf->pkt_cnt = 1;
1564 	txbuf->real_len = pkt_len;
1565 
1566 	dma_sync_single_for_device(dp->dev, rxbuf->dma_addr + dma_off,
1567 				   pkt_len, DMA_BIDIRECTIONAL);
1568 
1569 	/* Build TX descriptor */
1570 	txd = &tx_ring->txds[wr_idx];
1571 	txd->offset_eop = PCIE_DESC_TX_EOP;
1572 	txd->dma_len = cpu_to_le16(pkt_len);
1573 	nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + dma_off);
1574 	txd->data_len = cpu_to_le16(pkt_len);
1575 
1576 	txd->flags = 0;
1577 	txd->mss = 0;
1578 	txd->lso_hdrlen = 0;
1579 
1580 	tx_ring->wr_p++;
1581 	tx_ring->wr_ptr_add++;
1582 	return true;
1583 }
1584 
1585 static int nfp_net_run_xdp(struct bpf_prog *prog, void *data, void *hard_start,
1586 			   unsigned int *off, unsigned int *len)
1587 {
1588 	struct xdp_buff xdp;
1589 	void *orig_data;
1590 	int ret;
1591 
1592 	xdp.data_hard_start = hard_start;
1593 	xdp.data = data + *off;
1594 	xdp.data_end = data + *off + *len;
1595 
1596 	orig_data = xdp.data;
1597 	ret = bpf_prog_run_xdp(prog, &xdp);
1598 
1599 	*len -= xdp.data - orig_data;
1600 	*off += xdp.data - orig_data;
1601 
1602 	return ret;
1603 }
1604 
1605 /**
1606  * nfp_net_rx() - receive up to @budget packets on @rx_ring
1607  * @rx_ring:   RX ring to receive from
1608  * @budget:    NAPI budget
1609  *
1610  * Note, this function is separated out from the napi poll function to
1611  * more cleanly separate packet receive code from other bookkeeping
1612  * functions performed in the napi poll function.
1613  *
1614  * Return: Number of packets received.
1615  */
1616 static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1617 {
1618 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1619 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1620 	struct nfp_net_tx_ring *tx_ring;
1621 	struct bpf_prog *xdp_prog;
1622 	bool xdp_tx_cmpl = false;
1623 	unsigned int true_bufsz;
1624 	struct sk_buff *skb;
1625 	int pkts_polled = 0;
1626 	int idx;
1627 
1628 	rcu_read_lock();
1629 	xdp_prog = READ_ONCE(dp->xdp_prog);
1630 	true_bufsz = xdp_prog ? PAGE_SIZE : dp->fl_bufsz;
1631 	tx_ring = r_vec->xdp_ring;
1632 
1633 	while (pkts_polled < budget) {
1634 		unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1635 		struct nfp_net_rx_buf *rxbuf;
1636 		struct nfp_net_rx_desc *rxd;
1637 		struct nfp_meta_parsed meta;
1638 		struct net_device *netdev;
1639 		dma_addr_t new_dma_addr;
1640 		void *new_frag;
1641 
1642 		idx = D_IDX(rx_ring, rx_ring->rd_p);
1643 
1644 		rxd = &rx_ring->rxds[idx];
1645 		if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1646 			break;
1647 
1648 		/* Memory barrier to ensure that we won't do other reads
1649 		 * before the DD bit.
1650 		 */
1651 		dma_rmb();
1652 
1653 		memset(&meta, 0, sizeof(meta));
1654 
1655 		rx_ring->rd_p++;
1656 		pkts_polled++;
1657 
1658 		rxbuf =	&rx_ring->rxbufs[idx];
1659 		/*         < meta_len >
1660 		 *  <-- [rx_offset] -->
1661 		 *  ---------------------------------------------------------
1662 		 * | [XX] |  metadata  |             packet           | XXXX |
1663 		 *  ---------------------------------------------------------
1664 		 *         <---------------- data_len --------------->
1665 		 *
1666 		 * The rx_offset is fixed for all packets, the meta_len can vary
1667 		 * on a packet by packet basis. If rx_offset is set to zero
1668 		 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1669 		 * buffer and is immediately followed by the packet (no [XX]).
1670 		 */
1671 		meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1672 		data_len = le16_to_cpu(rxd->rxd.data_len);
1673 		pkt_len = data_len - meta_len;
1674 
1675 		pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1676 		if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1677 			pkt_off += meta_len;
1678 		else
1679 			pkt_off += dp->rx_offset;
1680 		meta_off = pkt_off - meta_len;
1681 
1682 		/* Stats update */
1683 		u64_stats_update_begin(&r_vec->rx_sync);
1684 		r_vec->rx_pkts++;
1685 		r_vec->rx_bytes += pkt_len;
1686 		u64_stats_update_end(&r_vec->rx_sync);
1687 
1688 		if (unlikely(meta_len > NFP_NET_MAX_PREPEND ||
1689 			     (dp->rx_offset && meta_len > dp->rx_offset))) {
1690 			nn_dp_warn(dp, "oversized RX packet metadata %u\n",
1691 				   meta_len);
1692 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1693 			continue;
1694 		}
1695 
1696 		nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,
1697 					data_len);
1698 
1699 		if (!dp->chained_metadata_format) {
1700 			nfp_net_set_hash_desc(dp->netdev, &meta,
1701 					      rxbuf->frag + meta_off, rxd);
1702 		} else if (meta_len) {
1703 			void *end;
1704 
1705 			end = nfp_net_parse_meta(dp->netdev, &meta,
1706 						 rxbuf->frag + meta_off,
1707 						 meta_len);
1708 			if (unlikely(end != rxbuf->frag + pkt_off)) {
1709 				nn_dp_warn(dp, "invalid RX packet metadata\n");
1710 				nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1711 						NULL);
1712 				continue;
1713 			}
1714 		}
1715 
1716 		if (xdp_prog && !(rxd->rxd.flags & PCIE_DESC_RX_BPF &&
1717 				  dp->bpf_offload_xdp) && !meta.portid) {
1718 			unsigned int dma_off;
1719 			void *hard_start;
1720 			int act;
1721 
1722 			hard_start = rxbuf->frag + NFP_NET_RX_BUF_HEADROOM;
1723 
1724 			act = nfp_net_run_xdp(xdp_prog, rxbuf->frag, hard_start,
1725 					      &pkt_off, &pkt_len);
1726 			switch (act) {
1727 			case XDP_PASS:
1728 				break;
1729 			case XDP_TX:
1730 				dma_off = pkt_off - NFP_NET_RX_BUF_HEADROOM;
1731 				if (unlikely(!nfp_net_tx_xdp_buf(dp, rx_ring,
1732 								 tx_ring, rxbuf,
1733 								 dma_off,
1734 								 pkt_len,
1735 								 &xdp_tx_cmpl)))
1736 					trace_xdp_exception(dp->netdev,
1737 							    xdp_prog, act);
1738 				continue;
1739 			default:
1740 				bpf_warn_invalid_xdp_action(act);
1741 				/* fall through */
1742 			case XDP_ABORTED:
1743 				trace_xdp_exception(dp->netdev, xdp_prog, act);
1744 				/* fall through */
1745 			case XDP_DROP:
1746 				nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1747 						    rxbuf->dma_addr);
1748 				continue;
1749 			}
1750 		}
1751 
1752 		skb = build_skb(rxbuf->frag, true_bufsz);
1753 		if (unlikely(!skb)) {
1754 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1755 			continue;
1756 		}
1757 		new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
1758 		if (unlikely(!new_frag)) {
1759 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
1760 			continue;
1761 		}
1762 
1763 		nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
1764 
1765 		nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
1766 
1767 		if (likely(!meta.portid)) {
1768 			netdev = dp->netdev;
1769 		} else {
1770 			struct nfp_net *nn;
1771 
1772 			nn = netdev_priv(dp->netdev);
1773 			netdev = nfp_app_repr_get(nn->app, meta.portid);
1774 			if (unlikely(!netdev)) {
1775 				nfp_net_rx_drop(dp, r_vec, rx_ring, NULL, skb);
1776 				continue;
1777 			}
1778 			nfp_repr_inc_rx_stats(netdev, pkt_len);
1779 		}
1780 
1781 		skb_reserve(skb, pkt_off);
1782 		skb_put(skb, pkt_len);
1783 
1784 		skb->mark = meta.mark;
1785 		skb_set_hash(skb, meta.hash, meta.hash_type);
1786 
1787 		skb_record_rx_queue(skb, rx_ring->idx);
1788 		skb->protocol = eth_type_trans(skb, netdev);
1789 
1790 		nfp_net_rx_csum(dp, r_vec, rxd, &meta, skb);
1791 
1792 		if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
1793 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1794 					       le16_to_cpu(rxd->rxd.vlan));
1795 
1796 		napi_gro_receive(&rx_ring->r_vec->napi, skb);
1797 	}
1798 
1799 	if (xdp_prog) {
1800 		if (tx_ring->wr_ptr_add)
1801 			nfp_net_tx_xmit_more_flush(tx_ring);
1802 		else if (unlikely(tx_ring->wr_p != tx_ring->rd_p) &&
1803 			 !xdp_tx_cmpl)
1804 			if (!nfp_net_xdp_complete(tx_ring))
1805 				pkts_polled = budget;
1806 	}
1807 	rcu_read_unlock();
1808 
1809 	return pkts_polled;
1810 }
1811 
1812 /**
1813  * nfp_net_poll() - napi poll function
1814  * @napi:    NAPI structure
1815  * @budget:  NAPI budget
1816  *
1817  * Return: number of packets polled.
1818  */
1819 static int nfp_net_poll(struct napi_struct *napi, int budget)
1820 {
1821 	struct nfp_net_r_vector *r_vec =
1822 		container_of(napi, struct nfp_net_r_vector, napi);
1823 	unsigned int pkts_polled = 0;
1824 
1825 	if (r_vec->tx_ring)
1826 		nfp_net_tx_complete(r_vec->tx_ring);
1827 	if (r_vec->rx_ring)
1828 		pkts_polled = nfp_net_rx(r_vec->rx_ring, budget);
1829 
1830 	if (pkts_polled < budget)
1831 		if (napi_complete_done(napi, pkts_polled))
1832 			nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
1833 
1834 	return pkts_polled;
1835 }
1836 
1837 /* Control device data path
1838  */
1839 
1840 static bool
1841 nfp_ctrl_tx_one(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1842 		struct sk_buff *skb, bool old)
1843 {
1844 	unsigned int real_len = skb->len, meta_len = 0;
1845 	struct nfp_net_tx_ring *tx_ring;
1846 	struct nfp_net_tx_buf *txbuf;
1847 	struct nfp_net_tx_desc *txd;
1848 	struct nfp_net_dp *dp;
1849 	dma_addr_t dma_addr;
1850 	int wr_idx;
1851 
1852 	dp = &r_vec->nfp_net->dp;
1853 	tx_ring = r_vec->tx_ring;
1854 
1855 	if (WARN_ON_ONCE(skb_shinfo(skb)->nr_frags)) {
1856 		nn_dp_warn(dp, "Driver's CTRL TX does not implement gather\n");
1857 		goto err_free;
1858 	}
1859 
1860 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1861 		u64_stats_update_begin(&r_vec->tx_sync);
1862 		r_vec->tx_busy++;
1863 		u64_stats_update_end(&r_vec->tx_sync);
1864 		if (!old)
1865 			__skb_queue_tail(&r_vec->queue, skb);
1866 		else
1867 			__skb_queue_head(&r_vec->queue, skb);
1868 		return true;
1869 	}
1870 
1871 	if (nfp_app_ctrl_has_meta(nn->app)) {
1872 		if (unlikely(skb_headroom(skb) < 8)) {
1873 			nn_dp_warn(dp, "CTRL TX on skb without headroom\n");
1874 			goto err_free;
1875 		}
1876 		meta_len = 8;
1877 		put_unaligned_be32(NFP_META_PORT_ID_CTRL, skb_push(skb, 4));
1878 		put_unaligned_be32(NFP_NET_META_PORTID, skb_push(skb, 4));
1879 	}
1880 
1881 	/* Start with the head skbuf */
1882 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
1883 				  DMA_TO_DEVICE);
1884 	if (dma_mapping_error(dp->dev, dma_addr))
1885 		goto err_dma_warn;
1886 
1887 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1888 
1889 	/* Stash the soft descriptor of the head then initialize it */
1890 	txbuf = &tx_ring->txbufs[wr_idx];
1891 	txbuf->skb = skb;
1892 	txbuf->dma_addr = dma_addr;
1893 	txbuf->fidx = -1;
1894 	txbuf->pkt_cnt = 1;
1895 	txbuf->real_len = real_len;
1896 
1897 	/* Build TX descriptor */
1898 	txd = &tx_ring->txds[wr_idx];
1899 	txd->offset_eop = meta_len | PCIE_DESC_TX_EOP;
1900 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
1901 	nfp_desc_set_dma_addr(txd, dma_addr);
1902 	txd->data_len = cpu_to_le16(skb->len);
1903 
1904 	txd->flags = 0;
1905 	txd->mss = 0;
1906 	txd->lso_hdrlen = 0;
1907 
1908 	tx_ring->wr_p++;
1909 	tx_ring->wr_ptr_add++;
1910 	nfp_net_tx_xmit_more_flush(tx_ring);
1911 
1912 	return false;
1913 
1914 err_dma_warn:
1915 	nn_dp_warn(dp, "Failed to DMA map TX CTRL buffer\n");
1916 err_free:
1917 	u64_stats_update_begin(&r_vec->tx_sync);
1918 	r_vec->tx_errors++;
1919 	u64_stats_update_end(&r_vec->tx_sync);
1920 	dev_kfree_skb_any(skb);
1921 	return false;
1922 }
1923 
1924 bool nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
1925 {
1926 	struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
1927 	bool ret;
1928 
1929 	spin_lock_bh(&r_vec->lock);
1930 	ret = nfp_ctrl_tx_one(nn, r_vec, skb, false);
1931 	spin_unlock_bh(&r_vec->lock);
1932 
1933 	return ret;
1934 }
1935 
1936 static void __nfp_ctrl_tx_queued(struct nfp_net_r_vector *r_vec)
1937 {
1938 	struct sk_buff *skb;
1939 
1940 	while ((skb = __skb_dequeue(&r_vec->queue)))
1941 		if (nfp_ctrl_tx_one(r_vec->nfp_net, r_vec, skb, true))
1942 			return;
1943 }
1944 
1945 static bool
1946 nfp_ctrl_meta_ok(struct nfp_net *nn, void *data, unsigned int meta_len)
1947 {
1948 	u32 meta_type, meta_tag;
1949 
1950 	if (!nfp_app_ctrl_has_meta(nn->app))
1951 		return !meta_len;
1952 
1953 	if (meta_len != 8)
1954 		return false;
1955 
1956 	meta_type = get_unaligned_be32(data);
1957 	meta_tag = get_unaligned_be32(data + 4);
1958 
1959 	return (meta_type == NFP_NET_META_PORTID &&
1960 		meta_tag == NFP_META_PORT_ID_CTRL);
1961 }
1962 
1963 static bool
1964 nfp_ctrl_rx_one(struct nfp_net *nn, struct nfp_net_dp *dp,
1965 		struct nfp_net_r_vector *r_vec, struct nfp_net_rx_ring *rx_ring)
1966 {
1967 	unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1968 	struct nfp_net_rx_buf *rxbuf;
1969 	struct nfp_net_rx_desc *rxd;
1970 	dma_addr_t new_dma_addr;
1971 	struct sk_buff *skb;
1972 	void *new_frag;
1973 	int idx;
1974 
1975 	idx = D_IDX(rx_ring, rx_ring->rd_p);
1976 
1977 	rxd = &rx_ring->rxds[idx];
1978 	if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1979 		return false;
1980 
1981 	/* Memory barrier to ensure that we won't do other reads
1982 	 * before the DD bit.
1983 	 */
1984 	dma_rmb();
1985 
1986 	rx_ring->rd_p++;
1987 
1988 	rxbuf =	&rx_ring->rxbufs[idx];
1989 	meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1990 	data_len = le16_to_cpu(rxd->rxd.data_len);
1991 	pkt_len = data_len - meta_len;
1992 
1993 	pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1994 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1995 		pkt_off += meta_len;
1996 	else
1997 		pkt_off += dp->rx_offset;
1998 	meta_off = pkt_off - meta_len;
1999 
2000 	/* Stats update */
2001 	u64_stats_update_begin(&r_vec->rx_sync);
2002 	r_vec->rx_pkts++;
2003 	r_vec->rx_bytes += pkt_len;
2004 	u64_stats_update_end(&r_vec->rx_sync);
2005 
2006 	nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,	data_len);
2007 
2008 	if (unlikely(!nfp_ctrl_meta_ok(nn, rxbuf->frag + meta_off, meta_len))) {
2009 		nn_dp_warn(dp, "incorrect metadata for ctrl packet (%d)\n",
2010 			   meta_len);
2011 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2012 		return true;
2013 	}
2014 
2015 	skb = build_skb(rxbuf->frag, dp->fl_bufsz);
2016 	if (unlikely(!skb)) {
2017 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2018 		return true;
2019 	}
2020 	new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
2021 	if (unlikely(!new_frag)) {
2022 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
2023 		return true;
2024 	}
2025 
2026 	nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
2027 
2028 	nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
2029 
2030 	skb_reserve(skb, pkt_off);
2031 	skb_put(skb, pkt_len);
2032 
2033 	nfp_app_ctrl_rx(nn->app, skb);
2034 
2035 	return true;
2036 }
2037 
2038 static void nfp_ctrl_rx(struct nfp_net_r_vector *r_vec)
2039 {
2040 	struct nfp_net_rx_ring *rx_ring = r_vec->rx_ring;
2041 	struct nfp_net *nn = r_vec->nfp_net;
2042 	struct nfp_net_dp *dp = &nn->dp;
2043 
2044 	while (nfp_ctrl_rx_one(nn, dp, r_vec, rx_ring))
2045 		continue;
2046 }
2047 
2048 static void nfp_ctrl_poll(unsigned long arg)
2049 {
2050 	struct nfp_net_r_vector *r_vec = (void *)arg;
2051 
2052 	spin_lock_bh(&r_vec->lock);
2053 	nfp_net_tx_complete(r_vec->tx_ring);
2054 	__nfp_ctrl_tx_queued(r_vec);
2055 	spin_unlock_bh(&r_vec->lock);
2056 
2057 	nfp_ctrl_rx(r_vec);
2058 
2059 	nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2060 }
2061 
2062 /* Setup and Configuration
2063  */
2064 
2065 /**
2066  * nfp_net_vecs_init() - Assign IRQs and setup rvecs.
2067  * @nn:		NFP Network structure
2068  */
2069 static void nfp_net_vecs_init(struct nfp_net *nn)
2070 {
2071 	struct nfp_net_r_vector *r_vec;
2072 	int r;
2073 
2074 	nn->lsc_handler = nfp_net_irq_lsc;
2075 	nn->exn_handler = nfp_net_irq_exn;
2076 
2077 	for (r = 0; r < nn->max_r_vecs; r++) {
2078 		struct msix_entry *entry;
2079 
2080 		entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r];
2081 
2082 		r_vec = &nn->r_vecs[r];
2083 		r_vec->nfp_net = nn;
2084 		r_vec->irq_entry = entry->entry;
2085 		r_vec->irq_vector = entry->vector;
2086 
2087 		if (nn->dp.netdev) {
2088 			r_vec->handler = nfp_net_irq_rxtx;
2089 		} else {
2090 			r_vec->handler = nfp_ctrl_irq_rxtx;
2091 
2092 			__skb_queue_head_init(&r_vec->queue);
2093 			spin_lock_init(&r_vec->lock);
2094 			tasklet_init(&r_vec->tasklet, nfp_ctrl_poll,
2095 				     (unsigned long)r_vec);
2096 			tasklet_disable(&r_vec->tasklet);
2097 		}
2098 
2099 		cpumask_set_cpu(r, &r_vec->affinity_mask);
2100 	}
2101 }
2102 
2103 /**
2104  * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
2105  * @tx_ring:   TX ring to free
2106  */
2107 static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
2108 {
2109 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2110 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2111 
2112 	kfree(tx_ring->txbufs);
2113 
2114 	if (tx_ring->txds)
2115 		dma_free_coherent(dp->dev, tx_ring->size,
2116 				  tx_ring->txds, tx_ring->dma);
2117 
2118 	tx_ring->cnt = 0;
2119 	tx_ring->txbufs = NULL;
2120 	tx_ring->txds = NULL;
2121 	tx_ring->dma = 0;
2122 	tx_ring->size = 0;
2123 }
2124 
2125 /**
2126  * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
2127  * @dp:        NFP Net data path struct
2128  * @tx_ring:   TX Ring structure to allocate
2129  *
2130  * Return: 0 on success, negative errno otherwise.
2131  */
2132 static int
2133 nfp_net_tx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
2134 {
2135 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2136 	int sz;
2137 
2138 	tx_ring->cnt = dp->txd_cnt;
2139 
2140 	tx_ring->size = sizeof(*tx_ring->txds) * tx_ring->cnt;
2141 	tx_ring->txds = dma_zalloc_coherent(dp->dev, tx_ring->size,
2142 					    &tx_ring->dma, GFP_KERNEL);
2143 	if (!tx_ring->txds)
2144 		goto err_alloc;
2145 
2146 	sz = sizeof(*tx_ring->txbufs) * tx_ring->cnt;
2147 	tx_ring->txbufs = kzalloc(sz, GFP_KERNEL);
2148 	if (!tx_ring->txbufs)
2149 		goto err_alloc;
2150 
2151 	if (!tx_ring->is_xdp && dp->netdev)
2152 		netif_set_xps_queue(dp->netdev, &r_vec->affinity_mask,
2153 				    tx_ring->idx);
2154 
2155 	return 0;
2156 
2157 err_alloc:
2158 	nfp_net_tx_ring_free(tx_ring);
2159 	return -ENOMEM;
2160 }
2161 
2162 static void
2163 nfp_net_tx_ring_bufs_free(struct nfp_net_dp *dp,
2164 			  struct nfp_net_tx_ring *tx_ring)
2165 {
2166 	unsigned int i;
2167 
2168 	if (!tx_ring->is_xdp)
2169 		return;
2170 
2171 	for (i = 0; i < tx_ring->cnt; i++) {
2172 		if (!tx_ring->txbufs[i].frag)
2173 			return;
2174 
2175 		nfp_net_dma_unmap_rx(dp, tx_ring->txbufs[i].dma_addr);
2176 		__free_page(virt_to_page(tx_ring->txbufs[i].frag));
2177 	}
2178 }
2179 
2180 static int
2181 nfp_net_tx_ring_bufs_alloc(struct nfp_net_dp *dp,
2182 			   struct nfp_net_tx_ring *tx_ring)
2183 {
2184 	struct nfp_net_tx_buf *txbufs = tx_ring->txbufs;
2185 	unsigned int i;
2186 
2187 	if (!tx_ring->is_xdp)
2188 		return 0;
2189 
2190 	for (i = 0; i < tx_ring->cnt; i++) {
2191 		txbufs[i].frag = nfp_net_rx_alloc_one(dp, &txbufs[i].dma_addr);
2192 		if (!txbufs[i].frag) {
2193 			nfp_net_tx_ring_bufs_free(dp, tx_ring);
2194 			return -ENOMEM;
2195 		}
2196 	}
2197 
2198 	return 0;
2199 }
2200 
2201 static int nfp_net_tx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2202 {
2203 	unsigned int r;
2204 
2205 	dp->tx_rings = kcalloc(dp->num_tx_rings, sizeof(*dp->tx_rings),
2206 			       GFP_KERNEL);
2207 	if (!dp->tx_rings)
2208 		return -ENOMEM;
2209 
2210 	for (r = 0; r < dp->num_tx_rings; r++) {
2211 		int bias = 0;
2212 
2213 		if (r >= dp->num_stack_tx_rings)
2214 			bias = dp->num_stack_tx_rings;
2215 
2216 		nfp_net_tx_ring_init(&dp->tx_rings[r], &nn->r_vecs[r - bias],
2217 				     r, bias);
2218 
2219 		if (nfp_net_tx_ring_alloc(dp, &dp->tx_rings[r]))
2220 			goto err_free_prev;
2221 
2222 		if (nfp_net_tx_ring_bufs_alloc(dp, &dp->tx_rings[r]))
2223 			goto err_free_ring;
2224 	}
2225 
2226 	return 0;
2227 
2228 err_free_prev:
2229 	while (r--) {
2230 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2231 err_free_ring:
2232 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2233 	}
2234 	kfree(dp->tx_rings);
2235 	return -ENOMEM;
2236 }
2237 
2238 static void nfp_net_tx_rings_free(struct nfp_net_dp *dp)
2239 {
2240 	unsigned int r;
2241 
2242 	for (r = 0; r < dp->num_tx_rings; r++) {
2243 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2244 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2245 	}
2246 
2247 	kfree(dp->tx_rings);
2248 }
2249 
2250 /**
2251  * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
2252  * @rx_ring:  RX ring to free
2253  */
2254 static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
2255 {
2256 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
2257 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2258 
2259 	kfree(rx_ring->rxbufs);
2260 
2261 	if (rx_ring->rxds)
2262 		dma_free_coherent(dp->dev, rx_ring->size,
2263 				  rx_ring->rxds, rx_ring->dma);
2264 
2265 	rx_ring->cnt = 0;
2266 	rx_ring->rxbufs = NULL;
2267 	rx_ring->rxds = NULL;
2268 	rx_ring->dma = 0;
2269 	rx_ring->size = 0;
2270 }
2271 
2272 /**
2273  * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
2274  * @dp:	      NFP Net data path struct
2275  * @rx_ring:  RX ring to allocate
2276  *
2277  * Return: 0 on success, negative errno otherwise.
2278  */
2279 static int
2280 nfp_net_rx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring)
2281 {
2282 	int sz;
2283 
2284 	rx_ring->cnt = dp->rxd_cnt;
2285 	rx_ring->size = sizeof(*rx_ring->rxds) * rx_ring->cnt;
2286 	rx_ring->rxds = dma_zalloc_coherent(dp->dev, rx_ring->size,
2287 					    &rx_ring->dma, GFP_KERNEL);
2288 	if (!rx_ring->rxds)
2289 		goto err_alloc;
2290 
2291 	sz = sizeof(*rx_ring->rxbufs) * rx_ring->cnt;
2292 	rx_ring->rxbufs = kzalloc(sz, GFP_KERNEL);
2293 	if (!rx_ring->rxbufs)
2294 		goto err_alloc;
2295 
2296 	return 0;
2297 
2298 err_alloc:
2299 	nfp_net_rx_ring_free(rx_ring);
2300 	return -ENOMEM;
2301 }
2302 
2303 static int nfp_net_rx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2304 {
2305 	unsigned int r;
2306 
2307 	dp->rx_rings = kcalloc(dp->num_rx_rings, sizeof(*dp->rx_rings),
2308 			       GFP_KERNEL);
2309 	if (!dp->rx_rings)
2310 		return -ENOMEM;
2311 
2312 	for (r = 0; r < dp->num_rx_rings; r++) {
2313 		nfp_net_rx_ring_init(&dp->rx_rings[r], &nn->r_vecs[r], r);
2314 
2315 		if (nfp_net_rx_ring_alloc(dp, &dp->rx_rings[r]))
2316 			goto err_free_prev;
2317 
2318 		if (nfp_net_rx_ring_bufs_alloc(dp, &dp->rx_rings[r]))
2319 			goto err_free_ring;
2320 	}
2321 
2322 	return 0;
2323 
2324 err_free_prev:
2325 	while (r--) {
2326 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2327 err_free_ring:
2328 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2329 	}
2330 	kfree(dp->rx_rings);
2331 	return -ENOMEM;
2332 }
2333 
2334 static void nfp_net_rx_rings_free(struct nfp_net_dp *dp)
2335 {
2336 	unsigned int r;
2337 
2338 	for (r = 0; r < dp->num_rx_rings; r++) {
2339 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2340 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2341 	}
2342 
2343 	kfree(dp->rx_rings);
2344 }
2345 
2346 static void
2347 nfp_net_vector_assign_rings(struct nfp_net_dp *dp,
2348 			    struct nfp_net_r_vector *r_vec, int idx)
2349 {
2350 	r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL;
2351 	r_vec->tx_ring =
2352 		idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL;
2353 
2354 	r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ?
2355 		&dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL;
2356 }
2357 
2358 static int
2359 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2360 		       int idx)
2361 {
2362 	int err;
2363 
2364 	/* Setup NAPI */
2365 	if (nn->dp.netdev)
2366 		netif_napi_add(nn->dp.netdev, &r_vec->napi,
2367 			       nfp_net_poll, NAPI_POLL_WEIGHT);
2368 	else
2369 		tasklet_enable(&r_vec->tasklet);
2370 
2371 	snprintf(r_vec->name, sizeof(r_vec->name),
2372 		 "%s-rxtx-%d", nfp_net_name(nn), idx);
2373 	err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name,
2374 			  r_vec);
2375 	if (err) {
2376 		if (nn->dp.netdev)
2377 			netif_napi_del(&r_vec->napi);
2378 		else
2379 			tasklet_disable(&r_vec->tasklet);
2380 
2381 		nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector);
2382 		return err;
2383 	}
2384 	disable_irq(r_vec->irq_vector);
2385 
2386 	irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask);
2387 
2388 	nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector,
2389 	       r_vec->irq_entry);
2390 
2391 	return 0;
2392 }
2393 
2394 static void
2395 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
2396 {
2397 	irq_set_affinity_hint(r_vec->irq_vector, NULL);
2398 	if (nn->dp.netdev)
2399 		netif_napi_del(&r_vec->napi);
2400 	else
2401 		tasklet_disable(&r_vec->tasklet);
2402 
2403 	free_irq(r_vec->irq_vector, r_vec);
2404 }
2405 
2406 /**
2407  * nfp_net_rss_write_itbl() - Write RSS indirection table to device
2408  * @nn:      NFP Net device to reconfigure
2409  */
2410 void nfp_net_rss_write_itbl(struct nfp_net *nn)
2411 {
2412 	int i;
2413 
2414 	for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
2415 		nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
2416 			  get_unaligned_le32(nn->rss_itbl + i));
2417 }
2418 
2419 /**
2420  * nfp_net_rss_write_key() - Write RSS hash key to device
2421  * @nn:      NFP Net device to reconfigure
2422  */
2423 void nfp_net_rss_write_key(struct nfp_net *nn)
2424 {
2425 	int i;
2426 
2427 	for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4)
2428 		nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
2429 			  get_unaligned_le32(nn->rss_key + i));
2430 }
2431 
2432 /**
2433  * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
2434  * @nn:      NFP Net device to reconfigure
2435  */
2436 void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
2437 {
2438 	u8 i;
2439 	u32 factor;
2440 	u32 value;
2441 
2442 	/* Compute factor used to convert coalesce '_usecs' parameters to
2443 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
2444 	 * count.
2445 	 */
2446 	factor = nn->me_freq_mhz / 16;
2447 
2448 	/* copy RX interrupt coalesce parameters */
2449 	value = (nn->rx_coalesce_max_frames << 16) |
2450 		(factor * nn->rx_coalesce_usecs);
2451 	for (i = 0; i < nn->dp.num_rx_rings; i++)
2452 		nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
2453 
2454 	/* copy TX interrupt coalesce parameters */
2455 	value = (nn->tx_coalesce_max_frames << 16) |
2456 		(factor * nn->tx_coalesce_usecs);
2457 	for (i = 0; i < nn->dp.num_tx_rings; i++)
2458 		nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
2459 }
2460 
2461 /**
2462  * nfp_net_write_mac_addr() - Write mac address to the device control BAR
2463  * @nn:      NFP Net device to reconfigure
2464  * @addr:    MAC address to write
2465  *
2466  * Writes the MAC address from the netdev to the device control BAR.  Does not
2467  * perform the required reconfig.  We do a bit of byte swapping dance because
2468  * firmware is LE.
2469  */
2470 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr)
2471 {
2472 	nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr));
2473 	nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4));
2474 }
2475 
2476 static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
2477 {
2478 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
2479 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
2480 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
2481 
2482 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
2483 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
2484 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
2485 }
2486 
2487 /**
2488  * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
2489  * @nn:      NFP Net device to reconfigure
2490  */
2491 static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
2492 {
2493 	u32 new_ctrl, update;
2494 	unsigned int r;
2495 	int err;
2496 
2497 	new_ctrl = nn->dp.ctrl;
2498 	new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
2499 	update = NFP_NET_CFG_UPDATE_GEN;
2500 	update |= NFP_NET_CFG_UPDATE_MSIX;
2501 	update |= NFP_NET_CFG_UPDATE_RING;
2502 
2503 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2504 		new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
2505 
2506 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2507 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2508 
2509 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2510 	err = nfp_net_reconfig(nn, update);
2511 	if (err)
2512 		nn_err(nn, "Could not disable device: %d\n", err);
2513 
2514 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2515 		nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]);
2516 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2517 		nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]);
2518 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2519 		nfp_net_vec_clear_ring_data(nn, r);
2520 
2521 	nn->dp.ctrl = new_ctrl;
2522 }
2523 
2524 static void
2525 nfp_net_rx_ring_hw_cfg_write(struct nfp_net *nn,
2526 			     struct nfp_net_rx_ring *rx_ring, unsigned int idx)
2527 {
2528 	/* Write the DMA address, size and MSI-X info to the device */
2529 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), rx_ring->dma);
2530 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(rx_ring->cnt));
2531 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), rx_ring->r_vec->irq_entry);
2532 }
2533 
2534 static void
2535 nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
2536 			     struct nfp_net_tx_ring *tx_ring, unsigned int idx)
2537 {
2538 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), tx_ring->dma);
2539 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(tx_ring->cnt));
2540 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_entry);
2541 }
2542 
2543 /**
2544  * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2545  * @nn:      NFP Net device to reconfigure
2546  */
2547 static int nfp_net_set_config_and_enable(struct nfp_net *nn)
2548 {
2549 	u32 bufsz, new_ctrl, update = 0;
2550 	unsigned int r;
2551 	int err;
2552 
2553 	new_ctrl = nn->dp.ctrl;
2554 
2555 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) {
2556 		nfp_net_rss_write_key(nn);
2557 		nfp_net_rss_write_itbl(nn);
2558 		nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
2559 		update |= NFP_NET_CFG_UPDATE_RSS;
2560 	}
2561 
2562 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) {
2563 		nfp_net_coalesce_write_cfg(nn);
2564 		update |= NFP_NET_CFG_UPDATE_IRQMOD;
2565 	}
2566 
2567 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2568 		nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r);
2569 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2570 		nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r);
2571 
2572 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->dp.num_tx_rings == 64 ?
2573 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_tx_rings) - 1);
2574 
2575 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->dp.num_rx_rings == 64 ?
2576 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_rx_rings) - 1);
2577 
2578 	if (nn->dp.netdev)
2579 		nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
2580 
2581 	nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu);
2582 
2583 	bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA;
2584 	nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz);
2585 
2586 	/* Enable device */
2587 	new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2588 	update |= NFP_NET_CFG_UPDATE_GEN;
2589 	update |= NFP_NET_CFG_UPDATE_MSIX;
2590 	update |= NFP_NET_CFG_UPDATE_RING;
2591 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2592 		new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2593 
2594 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2595 	err = nfp_net_reconfig(nn, update);
2596 	if (err) {
2597 		nfp_net_clear_config_and_disable(nn);
2598 		return err;
2599 	}
2600 
2601 	nn->dp.ctrl = new_ctrl;
2602 
2603 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2604 		nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]);
2605 
2606 	/* Since reconfiguration requests while NFP is down are ignored we
2607 	 * have to wipe the entire VXLAN configuration and reinitialize it.
2608 	 */
2609 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN) {
2610 		memset(&nn->vxlan_ports, 0, sizeof(nn->vxlan_ports));
2611 		memset(&nn->vxlan_usecnt, 0, sizeof(nn->vxlan_usecnt));
2612 		udp_tunnel_get_rx_info(nn->dp.netdev);
2613 	}
2614 
2615 	return 0;
2616 }
2617 
2618 /**
2619  * nfp_net_close_stack() - Quiesce the stack (part of close)
2620  * @nn:	     NFP Net device to reconfigure
2621  */
2622 static void nfp_net_close_stack(struct nfp_net *nn)
2623 {
2624 	unsigned int r;
2625 
2626 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2627 	netif_carrier_off(nn->dp.netdev);
2628 	nn->link_up = false;
2629 
2630 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2631 		disable_irq(nn->r_vecs[r].irq_vector);
2632 		napi_disable(&nn->r_vecs[r].napi);
2633 	}
2634 
2635 	netif_tx_disable(nn->dp.netdev);
2636 }
2637 
2638 /**
2639  * nfp_net_close_free_all() - Free all runtime resources
2640  * @nn:      NFP Net device to reconfigure
2641  */
2642 static void nfp_net_close_free_all(struct nfp_net *nn)
2643 {
2644 	unsigned int r;
2645 
2646 	nfp_net_tx_rings_free(&nn->dp);
2647 	nfp_net_rx_rings_free(&nn->dp);
2648 
2649 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2650 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2651 
2652 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2653 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2654 }
2655 
2656 /**
2657  * nfp_net_netdev_close() - Called when the device is downed
2658  * @netdev:      netdev structure
2659  */
2660 static int nfp_net_netdev_close(struct net_device *netdev)
2661 {
2662 	struct nfp_net *nn = netdev_priv(netdev);
2663 
2664 	/* Step 1: Disable RX and TX rings from the Linux kernel perspective
2665 	 */
2666 	nfp_net_close_stack(nn);
2667 
2668 	/* Step 2: Tell NFP
2669 	 */
2670 	nfp_net_clear_config_and_disable(nn);
2671 	nfp_port_configure(netdev, false);
2672 
2673 	/* Step 3: Free resources
2674 	 */
2675 	nfp_net_close_free_all(nn);
2676 
2677 	nn_dbg(nn, "%s down", netdev->name);
2678 	return 0;
2679 }
2680 
2681 void nfp_ctrl_close(struct nfp_net *nn)
2682 {
2683 	int r;
2684 
2685 	rtnl_lock();
2686 
2687 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2688 		disable_irq(nn->r_vecs[r].irq_vector);
2689 		tasklet_disable(&nn->r_vecs[r].tasklet);
2690 	}
2691 
2692 	nfp_net_clear_config_and_disable(nn);
2693 
2694 	nfp_net_close_free_all(nn);
2695 
2696 	rtnl_unlock();
2697 }
2698 
2699 /**
2700  * nfp_net_open_stack() - Start the device from stack's perspective
2701  * @nn:      NFP Net device to reconfigure
2702  */
2703 static void nfp_net_open_stack(struct nfp_net *nn)
2704 {
2705 	unsigned int r;
2706 
2707 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2708 		napi_enable(&nn->r_vecs[r].napi);
2709 		enable_irq(nn->r_vecs[r].irq_vector);
2710 	}
2711 
2712 	netif_tx_wake_all_queues(nn->dp.netdev);
2713 
2714 	enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2715 	nfp_net_read_link_status(nn);
2716 }
2717 
2718 static int nfp_net_open_alloc_all(struct nfp_net *nn)
2719 {
2720 	int err, r;
2721 
2722 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2723 				      nn->exn_name, sizeof(nn->exn_name),
2724 				      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2725 	if (err)
2726 		return err;
2727 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2728 				      nn->lsc_name, sizeof(nn->lsc_name),
2729 				      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2730 	if (err)
2731 		goto err_free_exn;
2732 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2733 
2734 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2735 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2736 		if (err)
2737 			goto err_cleanup_vec_p;
2738 	}
2739 
2740 	err = nfp_net_rx_rings_prepare(nn, &nn->dp);
2741 	if (err)
2742 		goto err_cleanup_vec;
2743 
2744 	err = nfp_net_tx_rings_prepare(nn, &nn->dp);
2745 	if (err)
2746 		goto err_free_rx_rings;
2747 
2748 	for (r = 0; r < nn->max_r_vecs; r++)
2749 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2750 
2751 	return 0;
2752 
2753 err_free_rx_rings:
2754 	nfp_net_rx_rings_free(&nn->dp);
2755 err_cleanup_vec:
2756 	r = nn->dp.num_r_vecs;
2757 err_cleanup_vec_p:
2758 	while (r--)
2759 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2760 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2761 err_free_exn:
2762 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2763 	return err;
2764 }
2765 
2766 static int nfp_net_netdev_open(struct net_device *netdev)
2767 {
2768 	struct nfp_net *nn = netdev_priv(netdev);
2769 	int err;
2770 
2771 	/* Step 1: Allocate resources for rings and the like
2772 	 * - Request interrupts
2773 	 * - Allocate RX and TX ring resources
2774 	 * - Setup initial RSS table
2775 	 */
2776 	err = nfp_net_open_alloc_all(nn);
2777 	if (err)
2778 		return err;
2779 
2780 	err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
2781 	if (err)
2782 		goto err_free_all;
2783 
2784 	err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
2785 	if (err)
2786 		goto err_free_all;
2787 
2788 	/* Step 2: Configure the NFP
2789 	 * - Ifup the physical interface if it exists
2790 	 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2791 	 * - Write MAC address (in case it changed)
2792 	 * - Set the MTU
2793 	 * - Set the Freelist buffer size
2794 	 * - Enable the FW
2795 	 */
2796 	err = nfp_port_configure(netdev, true);
2797 	if (err)
2798 		goto err_free_all;
2799 
2800 	err = nfp_net_set_config_and_enable(nn);
2801 	if (err)
2802 		goto err_port_disable;
2803 
2804 	/* Step 3: Enable for kernel
2805 	 * - put some freelist descriptors on each RX ring
2806 	 * - enable NAPI on each ring
2807 	 * - enable all TX queues
2808 	 * - set link state
2809 	 */
2810 	nfp_net_open_stack(nn);
2811 
2812 	return 0;
2813 
2814 err_port_disable:
2815 	nfp_port_configure(netdev, false);
2816 err_free_all:
2817 	nfp_net_close_free_all(nn);
2818 	return err;
2819 }
2820 
2821 int nfp_ctrl_open(struct nfp_net *nn)
2822 {
2823 	int err, r;
2824 
2825 	/* ring dumping depends on vNICs being opened/closed under rtnl */
2826 	rtnl_lock();
2827 
2828 	err = nfp_net_open_alloc_all(nn);
2829 	if (err)
2830 		goto err_unlock;
2831 
2832 	err = nfp_net_set_config_and_enable(nn);
2833 	if (err)
2834 		goto err_free_all;
2835 
2836 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2837 		enable_irq(nn->r_vecs[r].irq_vector);
2838 
2839 	rtnl_unlock();
2840 
2841 	return 0;
2842 
2843 err_free_all:
2844 	nfp_net_close_free_all(nn);
2845 err_unlock:
2846 	rtnl_unlock();
2847 	return err;
2848 }
2849 
2850 static void nfp_net_set_rx_mode(struct net_device *netdev)
2851 {
2852 	struct nfp_net *nn = netdev_priv(netdev);
2853 	u32 new_ctrl;
2854 
2855 	new_ctrl = nn->dp.ctrl;
2856 
2857 	if (netdev->flags & IFF_PROMISC) {
2858 		if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
2859 			new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
2860 		else
2861 			nn_warn(nn, "FW does not support promiscuous mode\n");
2862 	} else {
2863 		new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
2864 	}
2865 
2866 	if (new_ctrl == nn->dp.ctrl)
2867 		return;
2868 
2869 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2870 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
2871 
2872 	nn->dp.ctrl = new_ctrl;
2873 }
2874 
2875 static void nfp_net_rss_init_itbl(struct nfp_net *nn)
2876 {
2877 	int i;
2878 
2879 	for (i = 0; i < sizeof(nn->rss_itbl); i++)
2880 		nn->rss_itbl[i] =
2881 			ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
2882 }
2883 
2884 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
2885 {
2886 	struct nfp_net_dp new_dp = *dp;
2887 
2888 	*dp = nn->dp;
2889 	nn->dp = new_dp;
2890 
2891 	nn->dp.netdev->mtu = new_dp.mtu;
2892 
2893 	if (!netif_is_rxfh_configured(nn->dp.netdev))
2894 		nfp_net_rss_init_itbl(nn);
2895 }
2896 
2897 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
2898 {
2899 	unsigned int r;
2900 	int err;
2901 
2902 	nfp_net_dp_swap(nn, dp);
2903 
2904 	for (r = 0; r <	nn->max_r_vecs; r++)
2905 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2906 
2907 	err = netif_set_real_num_rx_queues(nn->dp.netdev, nn->dp.num_rx_rings);
2908 	if (err)
2909 		return err;
2910 
2911 	if (nn->dp.netdev->real_num_tx_queues != nn->dp.num_stack_tx_rings) {
2912 		err = netif_set_real_num_tx_queues(nn->dp.netdev,
2913 						   nn->dp.num_stack_tx_rings);
2914 		if (err)
2915 			return err;
2916 	}
2917 
2918 	return nfp_net_set_config_and_enable(nn);
2919 }
2920 
2921 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
2922 {
2923 	struct nfp_net_dp *new;
2924 
2925 	new = kmalloc(sizeof(*new), GFP_KERNEL);
2926 	if (!new)
2927 		return NULL;
2928 
2929 	*new = nn->dp;
2930 
2931 	/* Clear things which need to be recomputed */
2932 	new->fl_bufsz = 0;
2933 	new->tx_rings = NULL;
2934 	new->rx_rings = NULL;
2935 	new->num_r_vecs = 0;
2936 	new->num_stack_tx_rings = 0;
2937 
2938 	return new;
2939 }
2940 
2941 static int
2942 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
2943 		     struct netlink_ext_ack *extack)
2944 {
2945 	/* XDP-enabled tests */
2946 	if (!dp->xdp_prog)
2947 		return 0;
2948 	if (dp->fl_bufsz > PAGE_SIZE) {
2949 		NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
2950 		return -EINVAL;
2951 	}
2952 	if (dp->num_tx_rings > nn->max_tx_rings) {
2953 		NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
2954 		return -EINVAL;
2955 	}
2956 
2957 	return 0;
2958 }
2959 
2960 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
2961 			  struct netlink_ext_ack *extack)
2962 {
2963 	int r, err;
2964 
2965 	dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
2966 
2967 	dp->num_stack_tx_rings = dp->num_tx_rings;
2968 	if (dp->xdp_prog)
2969 		dp->num_stack_tx_rings -= dp->num_rx_rings;
2970 
2971 	dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
2972 
2973 	err = nfp_net_check_config(nn, dp, extack);
2974 	if (err)
2975 		goto exit_free_dp;
2976 
2977 	if (!netif_running(dp->netdev)) {
2978 		nfp_net_dp_swap(nn, dp);
2979 		err = 0;
2980 		goto exit_free_dp;
2981 	}
2982 
2983 	/* Prepare new rings */
2984 	for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
2985 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2986 		if (err) {
2987 			dp->num_r_vecs = r;
2988 			goto err_cleanup_vecs;
2989 		}
2990 	}
2991 
2992 	err = nfp_net_rx_rings_prepare(nn, dp);
2993 	if (err)
2994 		goto err_cleanup_vecs;
2995 
2996 	err = nfp_net_tx_rings_prepare(nn, dp);
2997 	if (err)
2998 		goto err_free_rx;
2999 
3000 	/* Stop device, swap in new rings, try to start the firmware */
3001 	nfp_net_close_stack(nn);
3002 	nfp_net_clear_config_and_disable(nn);
3003 
3004 	err = nfp_net_dp_swap_enable(nn, dp);
3005 	if (err) {
3006 		int err2;
3007 
3008 		nfp_net_clear_config_and_disable(nn);
3009 
3010 		/* Try with old configuration and old rings */
3011 		err2 = nfp_net_dp_swap_enable(nn, dp);
3012 		if (err2)
3013 			nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
3014 			       err, err2);
3015 	}
3016 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3017 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3018 
3019 	nfp_net_rx_rings_free(dp);
3020 	nfp_net_tx_rings_free(dp);
3021 
3022 	nfp_net_open_stack(nn);
3023 exit_free_dp:
3024 	kfree(dp);
3025 
3026 	return err;
3027 
3028 err_free_rx:
3029 	nfp_net_rx_rings_free(dp);
3030 err_cleanup_vecs:
3031 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3032 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3033 	kfree(dp);
3034 	return err;
3035 }
3036 
3037 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
3038 {
3039 	struct nfp_net *nn = netdev_priv(netdev);
3040 	struct nfp_net_dp *dp;
3041 
3042 	dp = nfp_net_clone_dp(nn);
3043 	if (!dp)
3044 		return -ENOMEM;
3045 
3046 	dp->mtu = new_mtu;
3047 
3048 	return nfp_net_ring_reconfig(nn, dp, NULL);
3049 }
3050 
3051 static int
3052 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
3053 {
3054 	struct nfp_net *nn = netdev_priv(netdev);
3055 
3056 	/* Priority tagged packets with vlan id 0 are processed by the
3057 	 * NFP as untagged packets
3058 	 */
3059 	if (!vid)
3060 		return 0;
3061 
3062 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_VID, vid);
3063 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_PROTO, ETH_P_8021Q);
3064 
3065 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD);
3066 }
3067 
3068 static int
3069 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
3070 {
3071 	struct nfp_net *nn = netdev_priv(netdev);
3072 
3073 	/* Priority tagged packets with vlan id 0 are processed by the
3074 	 * NFP as untagged packets
3075 	 */
3076 	if (!vid)
3077 		return 0;
3078 
3079 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_VID, vid);
3080 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_PROTO, ETH_P_8021Q);
3081 
3082 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL);
3083 }
3084 
3085 static void nfp_net_stat64(struct net_device *netdev,
3086 			   struct rtnl_link_stats64 *stats)
3087 {
3088 	struct nfp_net *nn = netdev_priv(netdev);
3089 	int r;
3090 
3091 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
3092 		struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
3093 		u64 data[3];
3094 		unsigned int start;
3095 
3096 		do {
3097 			start = u64_stats_fetch_begin(&r_vec->rx_sync);
3098 			data[0] = r_vec->rx_pkts;
3099 			data[1] = r_vec->rx_bytes;
3100 			data[2] = r_vec->rx_drops;
3101 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
3102 		stats->rx_packets += data[0];
3103 		stats->rx_bytes += data[1];
3104 		stats->rx_dropped += data[2];
3105 
3106 		do {
3107 			start = u64_stats_fetch_begin(&r_vec->tx_sync);
3108 			data[0] = r_vec->tx_pkts;
3109 			data[1] = r_vec->tx_bytes;
3110 			data[2] = r_vec->tx_errors;
3111 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
3112 		stats->tx_packets += data[0];
3113 		stats->tx_bytes += data[1];
3114 		stats->tx_errors += data[2];
3115 	}
3116 }
3117 
3118 static int nfp_net_set_features(struct net_device *netdev,
3119 				netdev_features_t features)
3120 {
3121 	netdev_features_t changed = netdev->features ^ features;
3122 	struct nfp_net *nn = netdev_priv(netdev);
3123 	u32 new_ctrl;
3124 	int err;
3125 
3126 	/* Assume this is not called with features we have not advertised */
3127 
3128 	new_ctrl = nn->dp.ctrl;
3129 
3130 	if (changed & NETIF_F_RXCSUM) {
3131 		if (features & NETIF_F_RXCSUM)
3132 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3133 		else
3134 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
3135 	}
3136 
3137 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3138 		if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
3139 			new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3140 		else
3141 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
3142 	}
3143 
3144 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
3145 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
3146 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3147 					      NFP_NET_CFG_CTRL_LSO;
3148 		else
3149 			new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3150 	}
3151 
3152 	if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
3153 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3154 			new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3155 		else
3156 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
3157 	}
3158 
3159 	if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
3160 		if (features & NETIF_F_HW_VLAN_CTAG_TX)
3161 			new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3162 		else
3163 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
3164 	}
3165 
3166 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
3167 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
3168 			new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3169 		else
3170 			new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
3171 	}
3172 
3173 	if (changed & NETIF_F_SG) {
3174 		if (features & NETIF_F_SG)
3175 			new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
3176 		else
3177 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
3178 	}
3179 
3180 	if (changed & NETIF_F_HW_TC && nfp_app_tc_busy(nn->app, nn)) {
3181 		nn_err(nn, "Cannot disable HW TC offload while in use\n");
3182 		return -EBUSY;
3183 	}
3184 
3185 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
3186 	       netdev->features, features, changed);
3187 
3188 	if (new_ctrl == nn->dp.ctrl)
3189 		return 0;
3190 
3191 	nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
3192 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3193 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
3194 	if (err)
3195 		return err;
3196 
3197 	nn->dp.ctrl = new_ctrl;
3198 
3199 	return 0;
3200 }
3201 
3202 static netdev_features_t
3203 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
3204 		       netdev_features_t features)
3205 {
3206 	u8 l4_hdr;
3207 
3208 	/* We can't do TSO over double tagged packets (802.1AD) */
3209 	features &= vlan_features_check(skb, features);
3210 
3211 	if (!skb->encapsulation)
3212 		return features;
3213 
3214 	/* Ensure that inner L4 header offset fits into TX descriptor field */
3215 	if (skb_is_gso(skb)) {
3216 		u32 hdrlen;
3217 
3218 		hdrlen = skb_inner_transport_header(skb) - skb->data +
3219 			inner_tcp_hdrlen(skb);
3220 
3221 		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
3222 			features &= ~NETIF_F_GSO_MASK;
3223 	}
3224 
3225 	/* VXLAN/GRE check */
3226 	switch (vlan_get_protocol(skb)) {
3227 	case htons(ETH_P_IP):
3228 		l4_hdr = ip_hdr(skb)->protocol;
3229 		break;
3230 	case htons(ETH_P_IPV6):
3231 		l4_hdr = ipv6_hdr(skb)->nexthdr;
3232 		break;
3233 	default:
3234 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3235 	}
3236 
3237 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
3238 	    skb->inner_protocol != htons(ETH_P_TEB) ||
3239 	    (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
3240 	    (l4_hdr == IPPROTO_UDP &&
3241 	     (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
3242 	      sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
3243 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3244 
3245 	return features;
3246 }
3247 
3248 /**
3249  * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
3250  * @nn:   NFP Net device to reconfigure
3251  * @idx:  Index into the port table where new port should be written
3252  * @port: UDP port to configure (pass zero to remove VXLAN port)
3253  */
3254 static void nfp_net_set_vxlan_port(struct nfp_net *nn, int idx, __be16 port)
3255 {
3256 	int i;
3257 
3258 	nn->vxlan_ports[idx] = port;
3259 
3260 	if (!(nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN))
3261 		return;
3262 
3263 	BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
3264 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2)
3265 		nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(port),
3266 			  be16_to_cpu(nn->vxlan_ports[i + 1]) << 16 |
3267 			  be16_to_cpu(nn->vxlan_ports[i]));
3268 
3269 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_VXLAN);
3270 }
3271 
3272 /**
3273  * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
3274  * @nn:   NFP Network structure
3275  * @port: UDP port to look for
3276  *
3277  * Return: if the port is already in the table -- it's position;
3278  *	   if the port is not in the table -- free position to use;
3279  *	   if the table is full -- -ENOSPC.
3280  */
3281 static int nfp_net_find_vxlan_idx(struct nfp_net *nn, __be16 port)
3282 {
3283 	int i, free_idx = -ENOSPC;
3284 
3285 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i++) {
3286 		if (nn->vxlan_ports[i] == port)
3287 			return i;
3288 		if (!nn->vxlan_usecnt[i])
3289 			free_idx = i;
3290 	}
3291 
3292 	return free_idx;
3293 }
3294 
3295 static void nfp_net_add_vxlan_port(struct net_device *netdev,
3296 				   struct udp_tunnel_info *ti)
3297 {
3298 	struct nfp_net *nn = netdev_priv(netdev);
3299 	int idx;
3300 
3301 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3302 		return;
3303 
3304 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3305 	if (idx == -ENOSPC)
3306 		return;
3307 
3308 	if (!nn->vxlan_usecnt[idx]++)
3309 		nfp_net_set_vxlan_port(nn, idx, ti->port);
3310 }
3311 
3312 static void nfp_net_del_vxlan_port(struct net_device *netdev,
3313 				   struct udp_tunnel_info *ti)
3314 {
3315 	struct nfp_net *nn = netdev_priv(netdev);
3316 	int idx;
3317 
3318 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3319 		return;
3320 
3321 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3322 	if (idx == -ENOSPC || !nn->vxlan_usecnt[idx])
3323 		return;
3324 
3325 	if (!--nn->vxlan_usecnt[idx])
3326 		nfp_net_set_vxlan_port(nn, idx, 0);
3327 }
3328 
3329 static int
3330 nfp_net_xdp_setup_drv(struct nfp_net *nn, struct bpf_prog *prog,
3331 		      struct netlink_ext_ack *extack)
3332 {
3333 	struct nfp_net_dp *dp;
3334 
3335 	if (!prog == !nn->dp.xdp_prog) {
3336 		WRITE_ONCE(nn->dp.xdp_prog, prog);
3337 		return 0;
3338 	}
3339 
3340 	dp = nfp_net_clone_dp(nn);
3341 	if (!dp)
3342 		return -ENOMEM;
3343 
3344 	dp->xdp_prog = prog;
3345 	dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
3346 	dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
3347 	dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
3348 
3349 	/* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
3350 	return nfp_net_ring_reconfig(nn, dp, extack);
3351 }
3352 
3353 static int
3354 nfp_net_xdp_setup(struct nfp_net *nn, struct bpf_prog *prog, u32 flags,
3355 		  struct netlink_ext_ack *extack)
3356 {
3357 	struct bpf_prog *drv_prog, *offload_prog;
3358 	int err;
3359 
3360 	if (nn->xdp_prog && (flags ^ nn->xdp_flags) & XDP_FLAGS_MODES)
3361 		return -EBUSY;
3362 
3363 	/* Load both when no flags set to allow easy activation of driver path
3364 	 * when program is replaced by one which can't be offloaded.
3365 	 */
3366 	drv_prog     = flags & XDP_FLAGS_HW_MODE  ? NULL : prog;
3367 	offload_prog = flags & XDP_FLAGS_DRV_MODE ? NULL : prog;
3368 
3369 	err = nfp_net_xdp_setup_drv(nn, drv_prog, extack);
3370 	if (err)
3371 		return err;
3372 
3373 	err = nfp_app_xdp_offload(nn->app, nn, offload_prog);
3374 	if (err && flags & XDP_FLAGS_HW_MODE)
3375 		return err;
3376 
3377 	if (nn->xdp_prog)
3378 		bpf_prog_put(nn->xdp_prog);
3379 	nn->xdp_prog = prog;
3380 	nn->xdp_flags = flags;
3381 
3382 	return 0;
3383 }
3384 
3385 static int nfp_net_xdp(struct net_device *netdev, struct netdev_xdp *xdp)
3386 {
3387 	struct nfp_net *nn = netdev_priv(netdev);
3388 
3389 	switch (xdp->command) {
3390 	case XDP_SETUP_PROG:
3391 	case XDP_SETUP_PROG_HW:
3392 		return nfp_net_xdp_setup(nn, xdp->prog, xdp->flags,
3393 					 xdp->extack);
3394 	case XDP_QUERY_PROG:
3395 		xdp->prog_attached = !!nn->xdp_prog;
3396 		if (nn->dp.bpf_offload_xdp)
3397 			xdp->prog_attached = XDP_ATTACHED_HW;
3398 		xdp->prog_id = nn->xdp_prog ? nn->xdp_prog->aux->id : 0;
3399 		return 0;
3400 	default:
3401 		return -EINVAL;
3402 	}
3403 }
3404 
3405 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
3406 {
3407 	struct nfp_net *nn = netdev_priv(netdev);
3408 	struct sockaddr *saddr = addr;
3409 	int err;
3410 
3411 	err = eth_prepare_mac_addr_change(netdev, addr);
3412 	if (err)
3413 		return err;
3414 
3415 	nfp_net_write_mac_addr(nn, saddr->sa_data);
3416 
3417 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
3418 	if (err)
3419 		return err;
3420 
3421 	eth_commit_mac_addr_change(netdev, addr);
3422 
3423 	return 0;
3424 }
3425 
3426 const struct net_device_ops nfp_net_netdev_ops = {
3427 	.ndo_open		= nfp_net_netdev_open,
3428 	.ndo_stop		= nfp_net_netdev_close,
3429 	.ndo_start_xmit		= nfp_net_tx,
3430 	.ndo_get_stats64	= nfp_net_stat64,
3431 	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
3432 	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
3433 	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
3434 	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
3435 	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
3436 	.ndo_get_vf_config	= nfp_app_get_vf_config,
3437 	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
3438 	.ndo_setup_tc		= nfp_port_setup_tc,
3439 	.ndo_tx_timeout		= nfp_net_tx_timeout,
3440 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
3441 	.ndo_change_mtu		= nfp_net_change_mtu,
3442 	.ndo_set_mac_address	= nfp_net_set_mac_address,
3443 	.ndo_set_features	= nfp_net_set_features,
3444 	.ndo_features_check	= nfp_net_features_check,
3445 	.ndo_get_phys_port_name	= nfp_port_get_phys_port_name,
3446 	.ndo_udp_tunnel_add	= nfp_net_add_vxlan_port,
3447 	.ndo_udp_tunnel_del	= nfp_net_del_vxlan_port,
3448 	.ndo_xdp		= nfp_net_xdp,
3449 };
3450 
3451 /**
3452  * nfp_net_info() - Print general info about the NIC
3453  * @nn:      NFP Net device to reconfigure
3454  */
3455 void nfp_net_info(struct nfp_net *nn)
3456 {
3457 	nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
3458 		nn->dp.is_vf ? "VF " : "",
3459 		nn->dp.num_tx_rings, nn->max_tx_rings,
3460 		nn->dp.num_rx_rings, nn->max_rx_rings);
3461 	nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3462 		nn->fw_ver.resv, nn->fw_ver.class,
3463 		nn->fw_ver.major, nn->fw_ver.minor,
3464 		nn->max_mtu);
3465 	nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
3466 		nn->cap,
3467 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
3468 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
3469 		nn->cap & NFP_NET_CFG_CTRL_L2MC     ? "L2MCFILT " : "",
3470 		nn->cap & NFP_NET_CFG_CTRL_RXCSUM   ? "RXCSUM "   : "",
3471 		nn->cap & NFP_NET_CFG_CTRL_TXCSUM   ? "TXCSUM "   : "",
3472 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN   ? "RXVLAN "   : "",
3473 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN   ? "TXVLAN "   : "",
3474 		nn->cap & NFP_NET_CFG_CTRL_SCATTER  ? "SCATTER "  : "",
3475 		nn->cap & NFP_NET_CFG_CTRL_GATHER   ? "GATHER "   : "",
3476 		nn->cap & NFP_NET_CFG_CTRL_LSO      ? "TSO1 "     : "",
3477 		nn->cap & NFP_NET_CFG_CTRL_LSO2     ? "TSO2 "     : "",
3478 		nn->cap & NFP_NET_CFG_CTRL_RSS      ? "RSS1 "     : "",
3479 		nn->cap & NFP_NET_CFG_CTRL_RSS2     ? "RSS2 "     : "",
3480 		nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "",
3481 		nn->cap & NFP_NET_CFG_CTRL_L2SWITCH ? "L2SWITCH " : "",
3482 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
3483 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
3484 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
3485 		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
3486 		nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ?
3487 						      "RXCSUM_COMPLETE " : "",
3488 		nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "",
3489 		nfp_app_extra_cap(nn->app, nn));
3490 }
3491 
3492 /**
3493  * nfp_net_alloc() - Allocate netdev and related structure
3494  * @pdev:         PCI device
3495  * @needs_netdev: Whether to allocate a netdev for this vNIC
3496  * @max_tx_rings: Maximum number of TX rings supported by device
3497  * @max_rx_rings: Maximum number of RX rings supported by device
3498  *
3499  * This function allocates a netdev device and fills in the initial
3500  * part of the @struct nfp_net structure.  In case of control device
3501  * nfp_net structure is allocated without the netdev.
3502  *
3503  * Return: NFP Net device structure, or ERR_PTR on error.
3504  */
3505 struct nfp_net *nfp_net_alloc(struct pci_dev *pdev, bool needs_netdev,
3506 			      unsigned int max_tx_rings,
3507 			      unsigned int max_rx_rings)
3508 {
3509 	struct nfp_net *nn;
3510 
3511 	if (needs_netdev) {
3512 		struct net_device *netdev;
3513 
3514 		netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
3515 					    max_tx_rings, max_rx_rings);
3516 		if (!netdev)
3517 			return ERR_PTR(-ENOMEM);
3518 
3519 		SET_NETDEV_DEV(netdev, &pdev->dev);
3520 		nn = netdev_priv(netdev);
3521 		nn->dp.netdev = netdev;
3522 	} else {
3523 		nn = vzalloc(sizeof(*nn));
3524 		if (!nn)
3525 			return ERR_PTR(-ENOMEM);
3526 	}
3527 
3528 	nn->dp.dev = &pdev->dev;
3529 	nn->pdev = pdev;
3530 
3531 	nn->max_tx_rings = max_tx_rings;
3532 	nn->max_rx_rings = max_rx_rings;
3533 
3534 	nn->dp.num_tx_rings = min_t(unsigned int,
3535 				    max_tx_rings, num_online_cpus());
3536 	nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings,
3537 				 netif_get_num_default_rss_queues());
3538 
3539 	nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings);
3540 	nn->dp.num_r_vecs = min_t(unsigned int,
3541 				  nn->dp.num_r_vecs, num_online_cpus());
3542 
3543 	nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
3544 	nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
3545 
3546 	spin_lock_init(&nn->reconfig_lock);
3547 	spin_lock_init(&nn->link_status_lock);
3548 
3549 	setup_timer(&nn->reconfig_timer,
3550 		    nfp_net_reconfig_timer, (unsigned long)nn);
3551 
3552 	return nn;
3553 }
3554 
3555 /**
3556  * nfp_net_free() - Undo what @nfp_net_alloc() did
3557  * @nn:      NFP Net device to reconfigure
3558  */
3559 void nfp_net_free(struct nfp_net *nn)
3560 {
3561 	if (nn->xdp_prog)
3562 		bpf_prog_put(nn->xdp_prog);
3563 
3564 	if (nn->dp.netdev)
3565 		free_netdev(nn->dp.netdev);
3566 	else
3567 		vfree(nn);
3568 }
3569 
3570 /**
3571  * nfp_net_rss_key_sz() - Get current size of the RSS key
3572  * @nn:		NFP Net device instance
3573  *
3574  * Return: size of the RSS key for currently selected hash function.
3575  */
3576 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn)
3577 {
3578 	switch (nn->rss_hfunc) {
3579 	case ETH_RSS_HASH_TOP:
3580 		return NFP_NET_CFG_RSS_KEY_SZ;
3581 	case ETH_RSS_HASH_XOR:
3582 		return 0;
3583 	case ETH_RSS_HASH_CRC32:
3584 		return 4;
3585 	}
3586 
3587 	nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc);
3588 	return 0;
3589 }
3590 
3591 /**
3592  * nfp_net_rss_init() - Set the initial RSS parameters
3593  * @nn:	     NFP Net device to reconfigure
3594  */
3595 static void nfp_net_rss_init(struct nfp_net *nn)
3596 {
3597 	unsigned long func_bit, rss_cap_hfunc;
3598 	u32 reg;
3599 
3600 	/* Read the RSS function capability and select first supported func */
3601 	reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP);
3602 	rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg);
3603 	if (!rss_cap_hfunc)
3604 		rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC,
3605 					  NFP_NET_CFG_RSS_TOEPLITZ);
3606 
3607 	func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS);
3608 	if (func_bit == NFP_NET_CFG_RSS_HFUNCS) {
3609 		dev_warn(nn->dp.dev,
3610 			 "Bad RSS config, defaulting to Toeplitz hash\n");
3611 		func_bit = ETH_RSS_HASH_TOP_BIT;
3612 	}
3613 	nn->rss_hfunc = 1 << func_bit;
3614 
3615 	netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn));
3616 
3617 	nfp_net_rss_init_itbl(nn);
3618 
3619 	/* Enable IPv4/IPv6 TCP by default */
3620 	nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
3621 		      NFP_NET_CFG_RSS_IPV6_TCP |
3622 		      FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) |
3623 		      NFP_NET_CFG_RSS_MASK;
3624 }
3625 
3626 /**
3627  * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
3628  * @nn:	     NFP Net device to reconfigure
3629  */
3630 static void nfp_net_irqmod_init(struct nfp_net *nn)
3631 {
3632 	nn->rx_coalesce_usecs      = 50;
3633 	nn->rx_coalesce_max_frames = 64;
3634 	nn->tx_coalesce_usecs      = 50;
3635 	nn->tx_coalesce_max_frames = 64;
3636 }
3637 
3638 static void nfp_net_netdev_init(struct nfp_net *nn)
3639 {
3640 	struct net_device *netdev = nn->dp.netdev;
3641 
3642 	nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
3643 
3644 	netdev->mtu = nn->dp.mtu;
3645 
3646 	/* Advertise/enable offloads based on capabilities
3647 	 *
3648 	 * Note: netdev->features show the currently enabled features
3649 	 * and netdev->hw_features advertises which features are
3650 	 * supported.  By default we enable most features.
3651 	 */
3652 	if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
3653 		netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3654 
3655 	netdev->hw_features = NETIF_F_HIGHDMA;
3656 	if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) {
3657 		netdev->hw_features |= NETIF_F_RXCSUM;
3658 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3659 	}
3660 	if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
3661 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3662 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3663 	}
3664 	if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
3665 		netdev->hw_features |= NETIF_F_SG;
3666 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER;
3667 	}
3668 	if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
3669 	    nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3670 		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3671 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3672 					 NFP_NET_CFG_CTRL_LSO;
3673 	}
3674 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY)
3675 		netdev->hw_features |= NETIF_F_RXHASH;
3676 	if (nn->cap & NFP_NET_CFG_CTRL_VXLAN &&
3677 	    nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
3678 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3679 			netdev->hw_features |= NETIF_F_GSO_GRE |
3680 					       NETIF_F_GSO_UDP_TUNNEL;
3681 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE;
3682 
3683 		netdev->hw_enc_features = netdev->hw_features;
3684 	}
3685 
3686 	netdev->vlan_features = netdev->hw_features;
3687 
3688 	if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
3689 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3690 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3691 	}
3692 	if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
3693 		if (nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3694 			nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
3695 		} else {
3696 			netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3697 			nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3698 		}
3699 	}
3700 	if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) {
3701 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3702 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3703 	}
3704 
3705 	netdev->features = netdev->hw_features;
3706 
3707 	if (nfp_app_has_tc(nn->app))
3708 		netdev->hw_features |= NETIF_F_HW_TC;
3709 
3710 	/* Advertise but disable TSO by default. */
3711 	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
3712 	nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3713 
3714 	/* Finalise the netdev setup */
3715 	netdev->netdev_ops = &nfp_net_netdev_ops;
3716 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
3717 
3718 	SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
3719 
3720 	/* MTU range: 68 - hw-specific max */
3721 	netdev->min_mtu = ETH_MIN_MTU;
3722 	netdev->max_mtu = nn->max_mtu;
3723 
3724 	netif_carrier_off(netdev);
3725 
3726 	nfp_net_set_ethtool_ops(netdev);
3727 }
3728 
3729 /**
3730  * nfp_net_init() - Initialise/finalise the nfp_net structure
3731  * @nn:		NFP Net device structure
3732  *
3733  * Return: 0 on success or negative errno on error.
3734  */
3735 int nfp_net_init(struct nfp_net *nn)
3736 {
3737 	int err;
3738 
3739 	nn->dp.rx_dma_dir = DMA_FROM_DEVICE;
3740 
3741 	/* Get some of the read-only fields from the BAR */
3742 	nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
3743 	nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
3744 
3745 	/* ABI 4.x and ctrl vNIC always use chained metadata, in other cases
3746 	 * we allow use of non-chained metadata if RSS(v1) is the only
3747 	 * advertised capability requiring metadata.
3748 	 */
3749 	nn->dp.chained_metadata_format = nn->fw_ver.major == 4 ||
3750 					 !nn->dp.netdev ||
3751 					 !(nn->cap & NFP_NET_CFG_CTRL_RSS) ||
3752 					 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META;
3753 	/* RSS(v1) uses non-chained metadata format, except in ABI 4.x where
3754 	 * it has the same meaning as RSSv2.
3755 	 */
3756 	if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4)
3757 		nn->cap &= ~NFP_NET_CFG_CTRL_RSS;
3758 
3759 	/* Determine RX packet/metadata boundary offset */
3760 	if (nn->fw_ver.major >= 2) {
3761 		u32 reg;
3762 
3763 		reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
3764 		if (reg > NFP_NET_MAX_PREPEND) {
3765 			nn_err(nn, "Invalid rx offset: %d\n", reg);
3766 			return -EINVAL;
3767 		}
3768 		nn->dp.rx_offset = reg;
3769 	} else {
3770 		nn->dp.rx_offset = NFP_NET_RX_OFFSET;
3771 	}
3772 
3773 	/* Set default MTU and Freelist buffer size */
3774 	if (nn->max_mtu < NFP_NET_DEFAULT_MTU)
3775 		nn->dp.mtu = nn->max_mtu;
3776 	else
3777 		nn->dp.mtu = NFP_NET_DEFAULT_MTU;
3778 	nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp);
3779 
3780 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) {
3781 		nfp_net_rss_init(nn);
3782 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?:
3783 					 NFP_NET_CFG_CTRL_RSS;
3784 	}
3785 
3786 	/* Allow L2 Broadcast and Multicast through by default, if supported */
3787 	if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
3788 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC;
3789 	if (nn->cap & NFP_NET_CFG_CTRL_L2MC)
3790 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2MC;
3791 
3792 	/* Allow IRQ moderation, if supported */
3793 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
3794 		nfp_net_irqmod_init(nn);
3795 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
3796 	}
3797 
3798 	if (nn->dp.netdev)
3799 		nfp_net_netdev_init(nn);
3800 
3801 	/* Stash the re-configuration queue away.  First odd queue in TX Bar */
3802 	nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
3803 
3804 	/* Make sure the FW knows the netdev is supposed to be disabled here */
3805 	nn_writel(nn, NFP_NET_CFG_CTRL, 0);
3806 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
3807 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
3808 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
3809 				   NFP_NET_CFG_UPDATE_GEN);
3810 	if (err)
3811 		return err;
3812 
3813 	nfp_net_vecs_init(nn);
3814 
3815 	if (!nn->dp.netdev)
3816 		return 0;
3817 	return register_netdev(nn->dp.netdev);
3818 }
3819 
3820 /**
3821  * nfp_net_clean() - Undo what nfp_net_init() did.
3822  * @nn:		NFP Net device structure
3823  */
3824 void nfp_net_clean(struct nfp_net *nn)
3825 {
3826 	if (!nn->dp.netdev)
3827 		return;
3828 
3829 	unregister_netdev(nn->dp.netdev);
3830 }
3831