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