xref: /linux/drivers/net/ethernet/netronome/nfp/nfp_net_common.c (revision 995231c820e3bd3633cb38bf4ea6f2541e1da331)
1 /*
2  * Copyright (C) 2015-2017 Netronome Systems, Inc.
3  *
4  * This software is dual licensed under the GNU General License Version 2,
5  * June 1991 as shown in the file COPYING in the top-level directory of this
6  * source tree or the BSD 2-Clause License provided below.  You have the
7  * option to license this software under the complete terms of either license.
8  *
9  * The BSD 2-Clause License:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      1. Redistributions of source code must retain the above
16  *         copyright notice, this list of conditions and the following
17  *         disclaimer.
18  *
19  *      2. Redistributions in binary form must reproduce the above
20  *         copyright notice, this list of conditions and the following
21  *         disclaimer in the documentation and/or other materials
22  *         provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 /*
35  * nfp_net_common.c
36  * Netronome network device driver: Common functions between PF and VF
37  * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
38  *          Jason McMullan <jason.mcmullan@netronome.com>
39  *          Rolf Neugebauer <rolf.neugebauer@netronome.com>
40  *          Brad Petrus <brad.petrus@netronome.com>
41  *          Chris Telfer <chris.telfer@netronome.com>
42  */
43 
44 #include <linux/bitfield.h>
45 #include <linux/bpf.h>
46 #include <linux/bpf_trace.h>
47 #include <linux/module.h>
48 #include <linux/kernel.h>
49 #include <linux/init.h>
50 #include <linux/fs.h>
51 #include <linux/netdevice.h>
52 #include <linux/etherdevice.h>
53 #include <linux/interrupt.h>
54 #include <linux/ip.h>
55 #include <linux/ipv6.h>
56 #include <linux/page_ref.h>
57 #include <linux/pci.h>
58 #include <linux/pci_regs.h>
59 #include <linux/msi.h>
60 #include <linux/ethtool.h>
61 #include <linux/log2.h>
62 #include <linux/if_vlan.h>
63 #include <linux/random.h>
64 #include <linux/vmalloc.h>
65 #include <linux/ktime.h>
66 
67 #include <net/switchdev.h>
68 #include <net/vxlan.h>
69 
70 #include "nfpcore/nfp_nsp.h"
71 #include "nfp_app.h"
72 #include "nfp_net_ctrl.h"
73 #include "nfp_net.h"
74 #include "nfp_net_sriov.h"
75 #include "nfp_port.h"
76 
77 /**
78  * nfp_net_get_fw_version() - Read and parse the FW version
79  * @fw_ver:	Output fw_version structure to read to
80  * @ctrl_bar:	Mapped address of the control BAR
81  */
82 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
83 			    void __iomem *ctrl_bar)
84 {
85 	u32 reg;
86 
87 	reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
88 	put_unaligned_le32(reg, fw_ver);
89 }
90 
91 static dma_addr_t nfp_net_dma_map_rx(struct nfp_net_dp *dp, void *frag)
92 {
93 	return dma_map_single_attrs(dp->dev, frag + NFP_NET_RX_BUF_HEADROOM,
94 				    dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
95 				    dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
96 }
97 
98 static void
99 nfp_net_dma_sync_dev_rx(const struct nfp_net_dp *dp, dma_addr_t dma_addr)
100 {
101 	dma_sync_single_for_device(dp->dev, dma_addr,
102 				   dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
103 				   dp->rx_dma_dir);
104 }
105 
106 static void nfp_net_dma_unmap_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr)
107 {
108 	dma_unmap_single_attrs(dp->dev, dma_addr,
109 			       dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
110 			       dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
111 }
112 
113 static void nfp_net_dma_sync_cpu_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr,
114 				    unsigned int len)
115 {
116 	dma_sync_single_for_cpu(dp->dev, dma_addr - NFP_NET_RX_BUF_HEADROOM,
117 				len, dp->rx_dma_dir);
118 }
119 
120 /* Firmware reconfig
121  *
122  * Firmware reconfig may take a while so we have two versions of it -
123  * synchronous and asynchronous (posted).  All synchronous callers are holding
124  * RTNL so we don't have to worry about serializing them.
125  */
126 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
127 {
128 	nn_writel(nn, NFP_NET_CFG_UPDATE, update);
129 	/* ensure update is written before pinging HW */
130 	nn_pci_flush(nn);
131 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
132 }
133 
134 /* Pass 0 as update to run posted reconfigs. */
135 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
136 {
137 	update |= nn->reconfig_posted;
138 	nn->reconfig_posted = 0;
139 
140 	nfp_net_reconfig_start(nn, update);
141 
142 	nn->reconfig_timer_active = true;
143 	mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
144 }
145 
146 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
147 {
148 	u32 reg;
149 
150 	reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
151 	if (reg == 0)
152 		return true;
153 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
154 		nn_err(nn, "Reconfig error: 0x%08x\n", reg);
155 		return true;
156 	} else if (last_check) {
157 		nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
158 		return true;
159 	}
160 
161 	return false;
162 }
163 
164 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
165 {
166 	bool timed_out = false;
167 
168 	/* Poll update field, waiting for NFP to ack the config */
169 	while (!nfp_net_reconfig_check_done(nn, timed_out)) {
170 		msleep(1);
171 		timed_out = time_is_before_eq_jiffies(deadline);
172 	}
173 
174 	if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
175 		return -EIO;
176 
177 	return timed_out ? -EIO : 0;
178 }
179 
180 static void nfp_net_reconfig_timer(struct timer_list *t)
181 {
182 	struct nfp_net *nn = from_timer(nn, t, reconfig_timer);
183 
184 	spin_lock_bh(&nn->reconfig_lock);
185 
186 	nn->reconfig_timer_active = false;
187 
188 	/* If sync caller is present it will take over from us */
189 	if (nn->reconfig_sync_present)
190 		goto done;
191 
192 	/* Read reconfig status and report errors */
193 	nfp_net_reconfig_check_done(nn, true);
194 
195 	if (nn->reconfig_posted)
196 		nfp_net_reconfig_start_async(nn, 0);
197 done:
198 	spin_unlock_bh(&nn->reconfig_lock);
199 }
200 
201 /**
202  * nfp_net_reconfig_post() - Post async reconfig request
203  * @nn:      NFP Net device to reconfigure
204  * @update:  The value for the update field in the BAR config
205  *
206  * Record FW reconfiguration request.  Reconfiguration will be kicked off
207  * whenever reconfiguration machinery is idle.  Multiple requests can be
208  * merged together!
209  */
210 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
211 {
212 	spin_lock_bh(&nn->reconfig_lock);
213 
214 	/* Sync caller will kick off async reconf when it's done, just post */
215 	if (nn->reconfig_sync_present) {
216 		nn->reconfig_posted |= update;
217 		goto done;
218 	}
219 
220 	/* Opportunistically check if the previous command is done */
221 	if (!nn->reconfig_timer_active ||
222 	    nfp_net_reconfig_check_done(nn, false))
223 		nfp_net_reconfig_start_async(nn, update);
224 	else
225 		nn->reconfig_posted |= update;
226 done:
227 	spin_unlock_bh(&nn->reconfig_lock);
228 }
229 
230 /**
231  * nfp_net_reconfig() - Reconfigure the firmware
232  * @nn:      NFP Net device to reconfigure
233  * @update:  The value for the update field in the BAR config
234  *
235  * Write the update word to the BAR and ping the reconfig queue.  The
236  * poll until the firmware has acknowledged the update by zeroing the
237  * update word.
238  *
239  * Return: Negative errno on error, 0 on success
240  */
241 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
242 {
243 	bool cancelled_timer = false;
244 	u32 pre_posted_requests;
245 	int ret;
246 
247 	spin_lock_bh(&nn->reconfig_lock);
248 
249 	nn->reconfig_sync_present = true;
250 
251 	if (nn->reconfig_timer_active) {
252 		del_timer(&nn->reconfig_timer);
253 		nn->reconfig_timer_active = false;
254 		cancelled_timer = true;
255 	}
256 	pre_posted_requests = nn->reconfig_posted;
257 	nn->reconfig_posted = 0;
258 
259 	spin_unlock_bh(&nn->reconfig_lock);
260 
261 	if (cancelled_timer)
262 		nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
263 
264 	/* Run the posted reconfigs which were issued before we started */
265 	if (pre_posted_requests) {
266 		nfp_net_reconfig_start(nn, pre_posted_requests);
267 		nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
268 	}
269 
270 	nfp_net_reconfig_start(nn, update);
271 	ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
272 
273 	spin_lock_bh(&nn->reconfig_lock);
274 
275 	if (nn->reconfig_posted)
276 		nfp_net_reconfig_start_async(nn, 0);
277 
278 	nn->reconfig_sync_present = false;
279 
280 	spin_unlock_bh(&nn->reconfig_lock);
281 
282 	return ret;
283 }
284 
285 /**
286  * nfp_net_reconfig_mbox() - Reconfigure the firmware via the mailbox
287  * @nn:        NFP Net device to reconfigure
288  * @mbox_cmd:  The value for the mailbox command
289  *
290  * Helper function for mailbox updates
291  *
292  * Return: Negative errno on error, 0 on success
293  */
294 static int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd)
295 {
296 	int ret;
297 
298 	nn_writeq(nn, NFP_NET_CFG_MBOX_CMD, mbox_cmd);
299 
300 	ret = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
301 	if (ret) {
302 		nn_err(nn, "Mailbox update error\n");
303 		return ret;
304 	}
305 
306 	return -nn_readl(nn, NFP_NET_CFG_MBOX_RET);
307 }
308 
309 /* Interrupt configuration and handling
310  */
311 
312 /**
313  * nfp_net_irq_unmask() - Unmask automasked interrupt
314  * @nn:       NFP Network structure
315  * @entry_nr: MSI-X table entry
316  *
317  * Clear the ICR for the IRQ entry.
318  */
319 static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
320 {
321 	nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
322 	nn_pci_flush(nn);
323 }
324 
325 /**
326  * nfp_net_irqs_alloc() - allocates MSI-X irqs
327  * @pdev:        PCI device structure
328  * @irq_entries: Array to be initialized and used to hold the irq entries
329  * @min_irqs:    Minimal acceptable number of interrupts
330  * @wanted_irqs: Target number of interrupts to allocate
331  *
332  * Return: Number of irqs obtained or 0 on error.
333  */
334 unsigned int
335 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
336 		   unsigned int min_irqs, unsigned int wanted_irqs)
337 {
338 	unsigned int i;
339 	int got_irqs;
340 
341 	for (i = 0; i < wanted_irqs; i++)
342 		irq_entries[i].entry = i;
343 
344 	got_irqs = pci_enable_msix_range(pdev, irq_entries,
345 					 min_irqs, wanted_irqs);
346 	if (got_irqs < 0) {
347 		dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
348 			min_irqs, wanted_irqs, got_irqs);
349 		return 0;
350 	}
351 
352 	if (got_irqs < wanted_irqs)
353 		dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
354 			 wanted_irqs, got_irqs);
355 
356 	return got_irqs;
357 }
358 
359 /**
360  * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
361  * @nn:		 NFP Network structure
362  * @irq_entries: Table of allocated interrupts
363  * @n:		 Size of @irq_entries (number of entries to grab)
364  *
365  * After interrupts are allocated with nfp_net_irqs_alloc() this function
366  * should be called to assign them to a specific netdev (port).
367  */
368 void
369 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
370 		    unsigned int n)
371 {
372 	struct nfp_net_dp *dp = &nn->dp;
373 
374 	nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
375 	dp->num_r_vecs = nn->max_r_vecs;
376 
377 	memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
378 
379 	if (dp->num_rx_rings > dp->num_r_vecs ||
380 	    dp->num_tx_rings > dp->num_r_vecs)
381 		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
382 			 dp->num_rx_rings, dp->num_tx_rings,
383 			 dp->num_r_vecs);
384 
385 	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
386 	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
387 	dp->num_stack_tx_rings = dp->num_tx_rings;
388 }
389 
390 /**
391  * nfp_net_irqs_disable() - Disable interrupts
392  * @pdev:        PCI device structure
393  *
394  * Undoes what @nfp_net_irqs_alloc() does.
395  */
396 void nfp_net_irqs_disable(struct pci_dev *pdev)
397 {
398 	pci_disable_msix(pdev);
399 }
400 
401 /**
402  * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
403  * @irq:      Interrupt
404  * @data:     Opaque data structure
405  *
406  * Return: Indicate if the interrupt has been handled.
407  */
408 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
409 {
410 	struct nfp_net_r_vector *r_vec = data;
411 
412 	napi_schedule_irqoff(&r_vec->napi);
413 
414 	/* The FW auto-masks any interrupt, either via the MASK bit in
415 	 * the MSI-X table or via the per entry ICR field.  So there
416 	 * is no need to disable interrupts here.
417 	 */
418 	return IRQ_HANDLED;
419 }
420 
421 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
422 {
423 	struct nfp_net_r_vector *r_vec = data;
424 
425 	tasklet_schedule(&r_vec->tasklet);
426 
427 	return IRQ_HANDLED;
428 }
429 
430 /**
431  * nfp_net_read_link_status() - Reread link status from control BAR
432  * @nn:       NFP Network structure
433  */
434 static void nfp_net_read_link_status(struct nfp_net *nn)
435 {
436 	unsigned long flags;
437 	bool link_up;
438 	u32 sts;
439 
440 	spin_lock_irqsave(&nn->link_status_lock, flags);
441 
442 	sts = nn_readl(nn, NFP_NET_CFG_STS);
443 	link_up = !!(sts & NFP_NET_CFG_STS_LINK);
444 
445 	if (nn->link_up == link_up)
446 		goto out;
447 
448 	nn->link_up = link_up;
449 	if (nn->port)
450 		set_bit(NFP_PORT_CHANGED, &nn->port->flags);
451 
452 	if (nn->link_up) {
453 		netif_carrier_on(nn->dp.netdev);
454 		netdev_info(nn->dp.netdev, "NIC Link is Up\n");
455 	} else {
456 		netif_carrier_off(nn->dp.netdev);
457 		netdev_info(nn->dp.netdev, "NIC Link is Down\n");
458 	}
459 out:
460 	spin_unlock_irqrestore(&nn->link_status_lock, flags);
461 }
462 
463 /**
464  * nfp_net_irq_lsc() - Interrupt service routine for link state changes
465  * @irq:      Interrupt
466  * @data:     Opaque data structure
467  *
468  * Return: Indicate if the interrupt has been handled.
469  */
470 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
471 {
472 	struct nfp_net *nn = data;
473 	struct msix_entry *entry;
474 
475 	entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
476 
477 	nfp_net_read_link_status(nn);
478 
479 	nfp_net_irq_unmask(nn, entry->entry);
480 
481 	return IRQ_HANDLED;
482 }
483 
484 /**
485  * nfp_net_irq_exn() - Interrupt service routine for exceptions
486  * @irq:      Interrupt
487  * @data:     Opaque data structure
488  *
489  * Return: Indicate if the interrupt has been handled.
490  */
491 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
492 {
493 	struct nfp_net *nn = data;
494 
495 	nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
496 	/* XXX TO BE IMPLEMENTED */
497 	return IRQ_HANDLED;
498 }
499 
500 /**
501  * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
502  * @tx_ring:  TX ring structure
503  * @r_vec:    IRQ vector servicing this ring
504  * @idx:      Ring index
505  * @is_xdp:   Is this an XDP TX ring?
506  */
507 static void
508 nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
509 		     struct nfp_net_r_vector *r_vec, unsigned int idx,
510 		     bool is_xdp)
511 {
512 	struct nfp_net *nn = r_vec->nfp_net;
513 
514 	tx_ring->idx = idx;
515 	tx_ring->r_vec = r_vec;
516 	tx_ring->is_xdp = is_xdp;
517 	u64_stats_init(&tx_ring->r_vec->tx_sync);
518 
519 	tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
520 	tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
521 }
522 
523 /**
524  * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
525  * @rx_ring:  RX ring structure
526  * @r_vec:    IRQ vector servicing this ring
527  * @idx:      Ring index
528  */
529 static void
530 nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
531 		     struct nfp_net_r_vector *r_vec, unsigned int idx)
532 {
533 	struct nfp_net *nn = r_vec->nfp_net;
534 
535 	rx_ring->idx = idx;
536 	rx_ring->r_vec = r_vec;
537 	u64_stats_init(&rx_ring->r_vec->rx_sync);
538 
539 	rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
540 	rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
541 }
542 
543 /**
544  * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
545  * @nn:		NFP Network structure
546  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
547  * @format:	printf-style format to construct the interrupt name
548  * @name:	Pointer to allocated space for interrupt name
549  * @name_sz:	Size of space for interrupt name
550  * @vector_idx:	Index of MSI-X vector used for this interrupt
551  * @handler:	IRQ handler to register for this interrupt
552  */
553 static int
554 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
555 			const char *format, char *name, size_t name_sz,
556 			unsigned int vector_idx, irq_handler_t handler)
557 {
558 	struct msix_entry *entry;
559 	int err;
560 
561 	entry = &nn->irq_entries[vector_idx];
562 
563 	snprintf(name, name_sz, format, nfp_net_name(nn));
564 	err = request_irq(entry->vector, handler, 0, name, nn);
565 	if (err) {
566 		nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
567 		       entry->vector, err);
568 		return err;
569 	}
570 	nn_writeb(nn, ctrl_offset, entry->entry);
571 
572 	return 0;
573 }
574 
575 /**
576  * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
577  * @nn:		NFP Network structure
578  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
579  * @vector_idx:	Index of MSI-X vector used for this interrupt
580  */
581 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
582 				 unsigned int vector_idx)
583 {
584 	nn_writeb(nn, ctrl_offset, 0xff);
585 	free_irq(nn->irq_entries[vector_idx].vector, nn);
586 }
587 
588 /* Transmit
589  *
590  * One queue controller peripheral queue is used for transmit.  The
591  * driver en-queues packets for transmit by advancing the write
592  * pointer.  The device indicates that packets have transmitted by
593  * advancing the read pointer.  The driver maintains a local copy of
594  * the read and write pointer in @struct nfp_net_tx_ring.  The driver
595  * keeps @wr_p in sync with the queue controller write pointer and can
596  * determine how many packets have been transmitted by comparing its
597  * copy of the read pointer @rd_p with the read pointer maintained by
598  * the queue controller peripheral.
599  */
600 
601 /**
602  * nfp_net_tx_full() - Check if the TX ring is full
603  * @tx_ring: TX ring to check
604  * @dcnt:    Number of descriptors that need to be enqueued (must be >= 1)
605  *
606  * This function checks, based on the *host copy* of read/write
607  * pointer if a given TX ring is full.  The real TX queue may have
608  * some newly made available slots.
609  *
610  * Return: True if the ring is full.
611  */
612 static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
613 {
614 	return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
615 }
616 
617 /* Wrappers for deciding when to stop and restart TX queues */
618 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
619 {
620 	return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
621 }
622 
623 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
624 {
625 	return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
626 }
627 
628 /**
629  * nfp_net_tx_ring_stop() - stop tx ring
630  * @nd_q:    netdev queue
631  * @tx_ring: driver tx queue structure
632  *
633  * Safely stop TX ring.  Remember that while we are running .start_xmit()
634  * someone else may be cleaning the TX ring completions so we need to be
635  * extra careful here.
636  */
637 static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
638 				 struct nfp_net_tx_ring *tx_ring)
639 {
640 	netif_tx_stop_queue(nd_q);
641 
642 	/* We can race with the TX completion out of NAPI so recheck */
643 	smp_mb();
644 	if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
645 		netif_tx_start_queue(nd_q);
646 }
647 
648 /**
649  * nfp_net_tx_tso() - Set up Tx descriptor for LSO
650  * @r_vec: per-ring structure
651  * @txbuf: Pointer to driver soft TX descriptor
652  * @txd: Pointer to HW TX descriptor
653  * @skb: Pointer to SKB
654  *
655  * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
656  * Return error on packet header greater than maximum supported LSO header size.
657  */
658 static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
659 			   struct nfp_net_tx_buf *txbuf,
660 			   struct nfp_net_tx_desc *txd, struct sk_buff *skb)
661 {
662 	u32 hdrlen;
663 	u16 mss;
664 
665 	if (!skb_is_gso(skb))
666 		return;
667 
668 	if (!skb->encapsulation) {
669 		txd->l3_offset = skb_network_offset(skb);
670 		txd->l4_offset = skb_transport_offset(skb);
671 		hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
672 	} else {
673 		txd->l3_offset = skb_inner_network_offset(skb);
674 		txd->l4_offset = skb_inner_transport_offset(skb);
675 		hdrlen = skb_inner_transport_header(skb) - skb->data +
676 			inner_tcp_hdrlen(skb);
677 	}
678 
679 	txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
680 	txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
681 
682 	mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
683 	txd->lso_hdrlen = hdrlen;
684 	txd->mss = cpu_to_le16(mss);
685 	txd->flags |= PCIE_DESC_TX_LSO;
686 
687 	u64_stats_update_begin(&r_vec->tx_sync);
688 	r_vec->tx_lso++;
689 	u64_stats_update_end(&r_vec->tx_sync);
690 }
691 
692 /**
693  * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
694  * @dp:  NFP Net data path struct
695  * @r_vec: per-ring structure
696  * @txbuf: Pointer to driver soft TX descriptor
697  * @txd: Pointer to TX descriptor
698  * @skb: Pointer to SKB
699  *
700  * This function sets the TX checksum flags in the TX descriptor based
701  * on the configuration and the protocol of the packet to be transmitted.
702  */
703 static void nfp_net_tx_csum(struct nfp_net_dp *dp,
704 			    struct nfp_net_r_vector *r_vec,
705 			    struct nfp_net_tx_buf *txbuf,
706 			    struct nfp_net_tx_desc *txd, struct sk_buff *skb)
707 {
708 	struct ipv6hdr *ipv6h;
709 	struct iphdr *iph;
710 	u8 l4_hdr;
711 
712 	if (!(dp->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
713 		return;
714 
715 	if (skb->ip_summed != CHECKSUM_PARTIAL)
716 		return;
717 
718 	txd->flags |= PCIE_DESC_TX_CSUM;
719 	if (skb->encapsulation)
720 		txd->flags |= PCIE_DESC_TX_ENCAP;
721 
722 	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
723 	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
724 
725 	if (iph->version == 4) {
726 		txd->flags |= PCIE_DESC_TX_IP4_CSUM;
727 		l4_hdr = iph->protocol;
728 	} else if (ipv6h->version == 6) {
729 		l4_hdr = ipv6h->nexthdr;
730 	} else {
731 		nn_dp_warn(dp, "partial checksum but ipv=%x!\n", iph->version);
732 		return;
733 	}
734 
735 	switch (l4_hdr) {
736 	case IPPROTO_TCP:
737 		txd->flags |= PCIE_DESC_TX_TCP_CSUM;
738 		break;
739 	case IPPROTO_UDP:
740 		txd->flags |= PCIE_DESC_TX_UDP_CSUM;
741 		break;
742 	default:
743 		nn_dp_warn(dp, "partial checksum but l4 proto=%x!\n", l4_hdr);
744 		return;
745 	}
746 
747 	u64_stats_update_begin(&r_vec->tx_sync);
748 	if (skb->encapsulation)
749 		r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
750 	else
751 		r_vec->hw_csum_tx += txbuf->pkt_cnt;
752 	u64_stats_update_end(&r_vec->tx_sync);
753 }
754 
755 static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
756 {
757 	wmb();
758 	nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
759 	tx_ring->wr_ptr_add = 0;
760 }
761 
762 static int nfp_net_prep_port_id(struct sk_buff *skb)
763 {
764 	struct metadata_dst *md_dst = skb_metadata_dst(skb);
765 	unsigned char *data;
766 
767 	if (likely(!md_dst))
768 		return 0;
769 	if (unlikely(md_dst->type != METADATA_HW_PORT_MUX))
770 		return 0;
771 
772 	if (unlikely(skb_cow_head(skb, 8)))
773 		return -ENOMEM;
774 
775 	data = skb_push(skb, 8);
776 	put_unaligned_be32(NFP_NET_META_PORTID, data);
777 	put_unaligned_be32(md_dst->u.port_info.port_id, data + 4);
778 
779 	return 8;
780 }
781 
782 /**
783  * nfp_net_tx() - Main transmit entry point
784  * @skb:    SKB to transmit
785  * @netdev: netdev structure
786  *
787  * Return: NETDEV_TX_OK on success.
788  */
789 static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
790 {
791 	struct nfp_net *nn = netdev_priv(netdev);
792 	const struct skb_frag_struct *frag;
793 	struct nfp_net_tx_desc *txd, txdg;
794 	int f, nr_frags, wr_idx, md_bytes;
795 	struct nfp_net_tx_ring *tx_ring;
796 	struct nfp_net_r_vector *r_vec;
797 	struct nfp_net_tx_buf *txbuf;
798 	struct netdev_queue *nd_q;
799 	struct nfp_net_dp *dp;
800 	dma_addr_t dma_addr;
801 	unsigned int fsize;
802 	u16 qidx;
803 
804 	dp = &nn->dp;
805 	qidx = skb_get_queue_mapping(skb);
806 	tx_ring = &dp->tx_rings[qidx];
807 	r_vec = tx_ring->r_vec;
808 	nd_q = netdev_get_tx_queue(dp->netdev, qidx);
809 
810 	nr_frags = skb_shinfo(skb)->nr_frags;
811 
812 	if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
813 		nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
814 			   qidx, tx_ring->wr_p, tx_ring->rd_p);
815 		netif_tx_stop_queue(nd_q);
816 		nfp_net_tx_xmit_more_flush(tx_ring);
817 		u64_stats_update_begin(&r_vec->tx_sync);
818 		r_vec->tx_busy++;
819 		u64_stats_update_end(&r_vec->tx_sync);
820 		return NETDEV_TX_BUSY;
821 	}
822 
823 	md_bytes = nfp_net_prep_port_id(skb);
824 	if (unlikely(md_bytes < 0)) {
825 		nfp_net_tx_xmit_more_flush(tx_ring);
826 		dev_kfree_skb_any(skb);
827 		return NETDEV_TX_OK;
828 	}
829 
830 	/* Start with the head skbuf */
831 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
832 				  DMA_TO_DEVICE);
833 	if (dma_mapping_error(dp->dev, dma_addr))
834 		goto err_free;
835 
836 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
837 
838 	/* Stash the soft descriptor of the head then initialize it */
839 	txbuf = &tx_ring->txbufs[wr_idx];
840 	txbuf->skb = skb;
841 	txbuf->dma_addr = dma_addr;
842 	txbuf->fidx = -1;
843 	txbuf->pkt_cnt = 1;
844 	txbuf->real_len = skb->len;
845 
846 	/* Build TX descriptor */
847 	txd = &tx_ring->txds[wr_idx];
848 	txd->offset_eop = (nr_frags ? 0 : PCIE_DESC_TX_EOP) | md_bytes;
849 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
850 	nfp_desc_set_dma_addr(txd, dma_addr);
851 	txd->data_len = cpu_to_le16(skb->len);
852 
853 	txd->flags = 0;
854 	txd->mss = 0;
855 	txd->lso_hdrlen = 0;
856 
857 	/* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
858 	nfp_net_tx_tso(r_vec, txbuf, txd, skb);
859 	nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
860 	if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
861 		txd->flags |= PCIE_DESC_TX_VLAN;
862 		txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
863 	}
864 
865 	/* Gather DMA */
866 	if (nr_frags > 0) {
867 		/* all descs must match except for in addr, length and eop */
868 		txdg = *txd;
869 
870 		for (f = 0; f < nr_frags; f++) {
871 			frag = &skb_shinfo(skb)->frags[f];
872 			fsize = skb_frag_size(frag);
873 
874 			dma_addr = skb_frag_dma_map(dp->dev, frag, 0,
875 						    fsize, DMA_TO_DEVICE);
876 			if (dma_mapping_error(dp->dev, dma_addr))
877 				goto err_unmap;
878 
879 			wr_idx = D_IDX(tx_ring, wr_idx + 1);
880 			tx_ring->txbufs[wr_idx].skb = skb;
881 			tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
882 			tx_ring->txbufs[wr_idx].fidx = f;
883 
884 			txd = &tx_ring->txds[wr_idx];
885 			*txd = txdg;
886 			txd->dma_len = cpu_to_le16(fsize);
887 			nfp_desc_set_dma_addr(txd, dma_addr);
888 			txd->offset_eop |=
889 				(f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
890 		}
891 
892 		u64_stats_update_begin(&r_vec->tx_sync);
893 		r_vec->tx_gather++;
894 		u64_stats_update_end(&r_vec->tx_sync);
895 	}
896 
897 	netdev_tx_sent_queue(nd_q, txbuf->real_len);
898 
899 	skb_tx_timestamp(skb);
900 
901 	tx_ring->wr_p += nr_frags + 1;
902 	if (nfp_net_tx_ring_should_stop(tx_ring))
903 		nfp_net_tx_ring_stop(nd_q, tx_ring);
904 
905 	tx_ring->wr_ptr_add += nr_frags + 1;
906 	if (!skb->xmit_more || netif_xmit_stopped(nd_q))
907 		nfp_net_tx_xmit_more_flush(tx_ring);
908 
909 	return NETDEV_TX_OK;
910 
911 err_unmap:
912 	while (--f >= 0) {
913 		frag = &skb_shinfo(skb)->frags[f];
914 		dma_unmap_page(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
915 			       skb_frag_size(frag), DMA_TO_DEVICE);
916 		tx_ring->txbufs[wr_idx].skb = NULL;
917 		tx_ring->txbufs[wr_idx].dma_addr = 0;
918 		tx_ring->txbufs[wr_idx].fidx = -2;
919 		wr_idx = wr_idx - 1;
920 		if (wr_idx < 0)
921 			wr_idx += tx_ring->cnt;
922 	}
923 	dma_unmap_single(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
924 			 skb_headlen(skb), DMA_TO_DEVICE);
925 	tx_ring->txbufs[wr_idx].skb = NULL;
926 	tx_ring->txbufs[wr_idx].dma_addr = 0;
927 	tx_ring->txbufs[wr_idx].fidx = -2;
928 err_free:
929 	nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
930 	nfp_net_tx_xmit_more_flush(tx_ring);
931 	u64_stats_update_begin(&r_vec->tx_sync);
932 	r_vec->tx_errors++;
933 	u64_stats_update_end(&r_vec->tx_sync);
934 	dev_kfree_skb_any(skb);
935 	return NETDEV_TX_OK;
936 }
937 
938 /**
939  * nfp_net_tx_complete() - Handled completed TX packets
940  * @tx_ring:   TX ring structure
941  *
942  * Return: Number of completed TX descriptors
943  */
944 static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring)
945 {
946 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
947 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
948 	const struct skb_frag_struct *frag;
949 	struct netdev_queue *nd_q;
950 	u32 done_pkts = 0, done_bytes = 0;
951 	struct sk_buff *skb;
952 	int todo, nr_frags;
953 	u32 qcp_rd_p;
954 	int fidx;
955 	int idx;
956 
957 	if (tx_ring->wr_p == tx_ring->rd_p)
958 		return;
959 
960 	/* Work out how many descriptors have been transmitted */
961 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
962 
963 	if (qcp_rd_p == tx_ring->qcp_rd_p)
964 		return;
965 
966 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
967 
968 	while (todo--) {
969 		idx = D_IDX(tx_ring, tx_ring->rd_p++);
970 
971 		skb = tx_ring->txbufs[idx].skb;
972 		if (!skb)
973 			continue;
974 
975 		nr_frags = skb_shinfo(skb)->nr_frags;
976 		fidx = tx_ring->txbufs[idx].fidx;
977 
978 		if (fidx == -1) {
979 			/* unmap head */
980 			dma_unmap_single(dp->dev, tx_ring->txbufs[idx].dma_addr,
981 					 skb_headlen(skb), DMA_TO_DEVICE);
982 
983 			done_pkts += tx_ring->txbufs[idx].pkt_cnt;
984 			done_bytes += tx_ring->txbufs[idx].real_len;
985 		} else {
986 			/* unmap fragment */
987 			frag = &skb_shinfo(skb)->frags[fidx];
988 			dma_unmap_page(dp->dev, tx_ring->txbufs[idx].dma_addr,
989 				       skb_frag_size(frag), DMA_TO_DEVICE);
990 		}
991 
992 		/* check for last gather fragment */
993 		if (fidx == nr_frags - 1)
994 			dev_consume_skb_any(skb);
995 
996 		tx_ring->txbufs[idx].dma_addr = 0;
997 		tx_ring->txbufs[idx].skb = NULL;
998 		tx_ring->txbufs[idx].fidx = -2;
999 	}
1000 
1001 	tx_ring->qcp_rd_p = qcp_rd_p;
1002 
1003 	u64_stats_update_begin(&r_vec->tx_sync);
1004 	r_vec->tx_bytes += done_bytes;
1005 	r_vec->tx_pkts += done_pkts;
1006 	u64_stats_update_end(&r_vec->tx_sync);
1007 
1008 	if (!dp->netdev)
1009 		return;
1010 
1011 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1012 	netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
1013 	if (nfp_net_tx_ring_should_wake(tx_ring)) {
1014 		/* Make sure TX thread will see updated tx_ring->rd_p */
1015 		smp_mb();
1016 
1017 		if (unlikely(netif_tx_queue_stopped(nd_q)))
1018 			netif_tx_wake_queue(nd_q);
1019 	}
1020 
1021 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1022 		  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1023 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1024 }
1025 
1026 static bool nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
1027 {
1028 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1029 	u32 done_pkts = 0, done_bytes = 0;
1030 	bool done_all;
1031 	int idx, todo;
1032 	u32 qcp_rd_p;
1033 
1034 	/* Work out how many descriptors have been transmitted */
1035 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1036 
1037 	if (qcp_rd_p == tx_ring->qcp_rd_p)
1038 		return true;
1039 
1040 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1041 
1042 	done_all = todo <= NFP_NET_XDP_MAX_COMPLETE;
1043 	todo = min(todo, NFP_NET_XDP_MAX_COMPLETE);
1044 
1045 	tx_ring->qcp_rd_p = D_IDX(tx_ring, tx_ring->qcp_rd_p + todo);
1046 
1047 	done_pkts = todo;
1048 	while (todo--) {
1049 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1050 		tx_ring->rd_p++;
1051 
1052 		done_bytes += tx_ring->txbufs[idx].real_len;
1053 	}
1054 
1055 	u64_stats_update_begin(&r_vec->tx_sync);
1056 	r_vec->tx_bytes += done_bytes;
1057 	r_vec->tx_pkts += done_pkts;
1058 	u64_stats_update_end(&r_vec->tx_sync);
1059 
1060 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1061 		  "XDP TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1062 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1063 
1064 	return done_all;
1065 }
1066 
1067 /**
1068  * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
1069  * @dp:		NFP Net data path struct
1070  * @tx_ring:	TX ring structure
1071  *
1072  * Assumes that the device is stopped
1073  */
1074 static void
1075 nfp_net_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
1076 {
1077 	const struct skb_frag_struct *frag;
1078 	struct netdev_queue *nd_q;
1079 
1080 	while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
1081 		struct nfp_net_tx_buf *tx_buf;
1082 		struct sk_buff *skb;
1083 		int idx, nr_frags;
1084 
1085 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1086 		tx_buf = &tx_ring->txbufs[idx];
1087 
1088 		skb = tx_ring->txbufs[idx].skb;
1089 		nr_frags = skb_shinfo(skb)->nr_frags;
1090 
1091 		if (tx_buf->fidx == -1) {
1092 			/* unmap head */
1093 			dma_unmap_single(dp->dev, tx_buf->dma_addr,
1094 					 skb_headlen(skb), DMA_TO_DEVICE);
1095 		} else {
1096 			/* unmap fragment */
1097 			frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1098 			dma_unmap_page(dp->dev, tx_buf->dma_addr,
1099 				       skb_frag_size(frag), DMA_TO_DEVICE);
1100 		}
1101 
1102 		/* check for last gather fragment */
1103 		if (tx_buf->fidx == nr_frags - 1)
1104 			dev_kfree_skb_any(skb);
1105 
1106 		tx_buf->dma_addr = 0;
1107 		tx_buf->skb = NULL;
1108 		tx_buf->fidx = -2;
1109 
1110 		tx_ring->qcp_rd_p++;
1111 		tx_ring->rd_p++;
1112 	}
1113 
1114 	memset(tx_ring->txds, 0, sizeof(*tx_ring->txds) * tx_ring->cnt);
1115 	tx_ring->wr_p = 0;
1116 	tx_ring->rd_p = 0;
1117 	tx_ring->qcp_rd_p = 0;
1118 	tx_ring->wr_ptr_add = 0;
1119 
1120 	if (tx_ring->is_xdp || !dp->netdev)
1121 		return;
1122 
1123 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1124 	netdev_tx_reset_queue(nd_q);
1125 }
1126 
1127 static void nfp_net_tx_timeout(struct net_device *netdev)
1128 {
1129 	struct nfp_net *nn = netdev_priv(netdev);
1130 	int i;
1131 
1132 	for (i = 0; i < nn->dp.netdev->real_num_tx_queues; i++) {
1133 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i)))
1134 			continue;
1135 		nn_warn(nn, "TX timeout on ring: %d\n", i);
1136 	}
1137 	nn_warn(nn, "TX watchdog timeout\n");
1138 }
1139 
1140 /* Receive processing
1141  */
1142 static unsigned int
1143 nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
1144 {
1145 	unsigned int fl_bufsz;
1146 
1147 	fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
1148 	fl_bufsz += dp->rx_dma_off;
1149 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1150 		fl_bufsz += NFP_NET_MAX_PREPEND;
1151 	else
1152 		fl_bufsz += dp->rx_offset;
1153 	fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
1154 
1155 	fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1156 	fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1157 
1158 	return fl_bufsz;
1159 }
1160 
1161 static void
1162 nfp_net_free_frag(void *frag, bool xdp)
1163 {
1164 	if (!xdp)
1165 		skb_free_frag(frag);
1166 	else
1167 		__free_page(virt_to_page(frag));
1168 }
1169 
1170 /**
1171  * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
1172  * @dp:		NFP Net data path struct
1173  * @dma_addr:	Pointer to storage for DMA address (output param)
1174  *
1175  * This function will allcate a new page frag, map it for DMA.
1176  *
1177  * Return: allocated page frag or NULL on failure.
1178  */
1179 static void *nfp_net_rx_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1180 {
1181 	void *frag;
1182 
1183 	if (!dp->xdp_prog) {
1184 		frag = netdev_alloc_frag(dp->fl_bufsz);
1185 	} else {
1186 		struct page *page;
1187 
1188 		page = alloc_page(GFP_KERNEL | __GFP_COLD);
1189 		frag = page ? page_address(page) : NULL;
1190 	}
1191 	if (!frag) {
1192 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1193 		return NULL;
1194 	}
1195 
1196 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1197 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1198 		nfp_net_free_frag(frag, dp->xdp_prog);
1199 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1200 		return NULL;
1201 	}
1202 
1203 	return frag;
1204 }
1205 
1206 static void *nfp_net_napi_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1207 {
1208 	void *frag;
1209 
1210 	if (!dp->xdp_prog) {
1211 		frag = napi_alloc_frag(dp->fl_bufsz);
1212 	} else {
1213 		struct page *page;
1214 
1215 		page = alloc_page(GFP_ATOMIC | __GFP_COLD);
1216 		frag = page ? page_address(page) : NULL;
1217 	}
1218 	if (!frag) {
1219 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1220 		return NULL;
1221 	}
1222 
1223 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1224 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1225 		nfp_net_free_frag(frag, dp->xdp_prog);
1226 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1227 		return NULL;
1228 	}
1229 
1230 	return frag;
1231 }
1232 
1233 /**
1234  * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1235  * @dp:		NFP Net data path struct
1236  * @rx_ring:	RX ring structure
1237  * @frag:	page fragment buffer
1238  * @dma_addr:	DMA address of skb mapping
1239  */
1240 static void nfp_net_rx_give_one(const struct nfp_net_dp *dp,
1241 				struct nfp_net_rx_ring *rx_ring,
1242 				void *frag, dma_addr_t dma_addr)
1243 {
1244 	unsigned int wr_idx;
1245 
1246 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1247 
1248 	nfp_net_dma_sync_dev_rx(dp, dma_addr);
1249 
1250 	/* Stash SKB and DMA address away */
1251 	rx_ring->rxbufs[wr_idx].frag = frag;
1252 	rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1253 
1254 	/* Fill freelist descriptor */
1255 	rx_ring->rxds[wr_idx].fld.reserved = 0;
1256 	rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1257 	nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld,
1258 			      dma_addr + dp->rx_dma_off);
1259 
1260 	rx_ring->wr_p++;
1261 	if (!(rx_ring->wr_p % NFP_NET_FL_BATCH)) {
1262 		/* Update write pointer of the freelist queue. Make
1263 		 * sure all writes are flushed before telling the hardware.
1264 		 */
1265 		wmb();
1266 		nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, NFP_NET_FL_BATCH);
1267 	}
1268 }
1269 
1270 /**
1271  * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1272  * @rx_ring:	RX ring structure
1273  *
1274  * Warning: Do *not* call if ring buffers were never put on the FW freelist
1275  *	    (i.e. device was not enabled)!
1276  */
1277 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1278 {
1279 	unsigned int wr_idx, last_idx;
1280 
1281 	/* Move the empty entry to the end of the list */
1282 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1283 	last_idx = rx_ring->cnt - 1;
1284 	rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1285 	rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1286 	rx_ring->rxbufs[last_idx].dma_addr = 0;
1287 	rx_ring->rxbufs[last_idx].frag = NULL;
1288 
1289 	memset(rx_ring->rxds, 0, sizeof(*rx_ring->rxds) * rx_ring->cnt);
1290 	rx_ring->wr_p = 0;
1291 	rx_ring->rd_p = 0;
1292 }
1293 
1294 /**
1295  * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1296  * @dp:		NFP Net data path struct
1297  * @rx_ring:	RX ring to remove buffers from
1298  *
1299  * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1300  * entries.  After device is disabled nfp_net_rx_ring_reset() must be called
1301  * to restore required ring geometry.
1302  */
1303 static void
1304 nfp_net_rx_ring_bufs_free(struct nfp_net_dp *dp,
1305 			  struct nfp_net_rx_ring *rx_ring)
1306 {
1307 	unsigned int i;
1308 
1309 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1310 		/* NULL skb can only happen when initial filling of the ring
1311 		 * fails to allocate enough buffers and calls here to free
1312 		 * already allocated ones.
1313 		 */
1314 		if (!rx_ring->rxbufs[i].frag)
1315 			continue;
1316 
1317 		nfp_net_dma_unmap_rx(dp, rx_ring->rxbufs[i].dma_addr);
1318 		nfp_net_free_frag(rx_ring->rxbufs[i].frag, dp->xdp_prog);
1319 		rx_ring->rxbufs[i].dma_addr = 0;
1320 		rx_ring->rxbufs[i].frag = NULL;
1321 	}
1322 }
1323 
1324 /**
1325  * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1326  * @dp:		NFP Net data path struct
1327  * @rx_ring:	RX ring to remove buffers from
1328  */
1329 static int
1330 nfp_net_rx_ring_bufs_alloc(struct nfp_net_dp *dp,
1331 			   struct nfp_net_rx_ring *rx_ring)
1332 {
1333 	struct nfp_net_rx_buf *rxbufs;
1334 	unsigned int i;
1335 
1336 	rxbufs = rx_ring->rxbufs;
1337 
1338 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1339 		rxbufs[i].frag = nfp_net_rx_alloc_one(dp, &rxbufs[i].dma_addr);
1340 		if (!rxbufs[i].frag) {
1341 			nfp_net_rx_ring_bufs_free(dp, rx_ring);
1342 			return -ENOMEM;
1343 		}
1344 	}
1345 
1346 	return 0;
1347 }
1348 
1349 /**
1350  * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1351  * @dp:	     NFP Net data path struct
1352  * @rx_ring: RX ring to fill
1353  */
1354 static void
1355 nfp_net_rx_ring_fill_freelist(struct nfp_net_dp *dp,
1356 			      struct nfp_net_rx_ring *rx_ring)
1357 {
1358 	unsigned int i;
1359 
1360 	for (i = 0; i < rx_ring->cnt - 1; i++)
1361 		nfp_net_rx_give_one(dp, rx_ring, rx_ring->rxbufs[i].frag,
1362 				    rx_ring->rxbufs[i].dma_addr);
1363 }
1364 
1365 /**
1366  * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1367  * @flags: RX descriptor flags field in CPU byte order
1368  */
1369 static int nfp_net_rx_csum_has_errors(u16 flags)
1370 {
1371 	u16 csum_all_checked, csum_all_ok;
1372 
1373 	csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1374 	csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1375 
1376 	return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1377 }
1378 
1379 /**
1380  * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1381  * @dp:  NFP Net data path struct
1382  * @r_vec: per-ring structure
1383  * @rxd: Pointer to RX descriptor
1384  * @meta: Parsed metadata prepend
1385  * @skb: Pointer to SKB
1386  */
1387 static void nfp_net_rx_csum(struct nfp_net_dp *dp,
1388 			    struct nfp_net_r_vector *r_vec,
1389 			    struct nfp_net_rx_desc *rxd,
1390 			    struct nfp_meta_parsed *meta, struct sk_buff *skb)
1391 {
1392 	skb_checksum_none_assert(skb);
1393 
1394 	if (!(dp->netdev->features & NETIF_F_RXCSUM))
1395 		return;
1396 
1397 	if (meta->csum_type) {
1398 		skb->ip_summed = meta->csum_type;
1399 		skb->csum = meta->csum;
1400 		u64_stats_update_begin(&r_vec->rx_sync);
1401 		r_vec->hw_csum_rx_ok++;
1402 		u64_stats_update_end(&r_vec->rx_sync);
1403 		return;
1404 	}
1405 
1406 	if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1407 		u64_stats_update_begin(&r_vec->rx_sync);
1408 		r_vec->hw_csum_rx_error++;
1409 		u64_stats_update_end(&r_vec->rx_sync);
1410 		return;
1411 	}
1412 
1413 	/* Assume that the firmware will never report inner CSUM_OK unless outer
1414 	 * L4 headers were successfully parsed. FW will always report zero UDP
1415 	 * checksum as CSUM_OK.
1416 	 */
1417 	if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1418 	    rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1419 		__skb_incr_checksum_unnecessary(skb);
1420 		u64_stats_update_begin(&r_vec->rx_sync);
1421 		r_vec->hw_csum_rx_ok++;
1422 		u64_stats_update_end(&r_vec->rx_sync);
1423 	}
1424 
1425 	if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1426 	    rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1427 		__skb_incr_checksum_unnecessary(skb);
1428 		u64_stats_update_begin(&r_vec->rx_sync);
1429 		r_vec->hw_csum_rx_inner_ok++;
1430 		u64_stats_update_end(&r_vec->rx_sync);
1431 	}
1432 }
1433 
1434 static void
1435 nfp_net_set_hash(struct net_device *netdev, struct nfp_meta_parsed *meta,
1436 		 unsigned int type, __be32 *hash)
1437 {
1438 	if (!(netdev->features & NETIF_F_RXHASH))
1439 		return;
1440 
1441 	switch (type) {
1442 	case NFP_NET_RSS_IPV4:
1443 	case NFP_NET_RSS_IPV6:
1444 	case NFP_NET_RSS_IPV6_EX:
1445 		meta->hash_type = PKT_HASH_TYPE_L3;
1446 		break;
1447 	default:
1448 		meta->hash_type = PKT_HASH_TYPE_L4;
1449 		break;
1450 	}
1451 
1452 	meta->hash = get_unaligned_be32(hash);
1453 }
1454 
1455 static void
1456 nfp_net_set_hash_desc(struct net_device *netdev, struct nfp_meta_parsed *meta,
1457 		      void *data, struct nfp_net_rx_desc *rxd)
1458 {
1459 	struct nfp_net_rx_hash *rx_hash = data;
1460 
1461 	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1462 		return;
1463 
1464 	nfp_net_set_hash(netdev, meta, get_unaligned_be32(&rx_hash->hash_type),
1465 			 &rx_hash->hash);
1466 }
1467 
1468 static void *
1469 nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
1470 		   void *data, int meta_len)
1471 {
1472 	u32 meta_info;
1473 
1474 	meta_info = get_unaligned_be32(data);
1475 	data += 4;
1476 
1477 	while (meta_info) {
1478 		switch (meta_info & NFP_NET_META_FIELD_MASK) {
1479 		case NFP_NET_META_HASH:
1480 			meta_info >>= NFP_NET_META_FIELD_SIZE;
1481 			nfp_net_set_hash(netdev, meta,
1482 					 meta_info & NFP_NET_META_FIELD_MASK,
1483 					 (__be32 *)data);
1484 			data += 4;
1485 			break;
1486 		case NFP_NET_META_MARK:
1487 			meta->mark = get_unaligned_be32(data);
1488 			data += 4;
1489 			break;
1490 		case NFP_NET_META_PORTID:
1491 			meta->portid = get_unaligned_be32(data);
1492 			data += 4;
1493 			break;
1494 		case NFP_NET_META_CSUM:
1495 			meta->csum_type = CHECKSUM_COMPLETE;
1496 			meta->csum =
1497 				(__force __wsum)__get_unaligned_cpu32(data);
1498 			data += 4;
1499 			break;
1500 		default:
1501 			return NULL;
1502 		}
1503 
1504 		meta_info >>= NFP_NET_META_FIELD_SIZE;
1505 	}
1506 
1507 	return data;
1508 }
1509 
1510 static void
1511 nfp_net_rx_drop(const struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
1512 		struct nfp_net_rx_ring *rx_ring, struct nfp_net_rx_buf *rxbuf,
1513 		struct sk_buff *skb)
1514 {
1515 	u64_stats_update_begin(&r_vec->rx_sync);
1516 	r_vec->rx_drops++;
1517 	u64_stats_update_end(&r_vec->rx_sync);
1518 
1519 	/* skb is build based on the frag, free_skb() would free the frag
1520 	 * so to be able to reuse it we need an extra ref.
1521 	 */
1522 	if (skb && rxbuf && skb->head == rxbuf->frag)
1523 		page_ref_inc(virt_to_head_page(rxbuf->frag));
1524 	if (rxbuf)
1525 		nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag, rxbuf->dma_addr);
1526 	if (skb)
1527 		dev_kfree_skb_any(skb);
1528 }
1529 
1530 static bool
1531 nfp_net_tx_xdp_buf(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring,
1532 		   struct nfp_net_tx_ring *tx_ring,
1533 		   struct nfp_net_rx_buf *rxbuf, unsigned int dma_off,
1534 		   unsigned int pkt_len, bool *completed)
1535 {
1536 	struct nfp_net_tx_buf *txbuf;
1537 	struct nfp_net_tx_desc *txd;
1538 	int wr_idx;
1539 
1540 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1541 		if (!*completed) {
1542 			nfp_net_xdp_complete(tx_ring);
1543 			*completed = true;
1544 		}
1545 
1546 		if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1547 			nfp_net_rx_drop(dp, rx_ring->r_vec, rx_ring, rxbuf,
1548 					NULL);
1549 			return false;
1550 		}
1551 	}
1552 
1553 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1554 
1555 	/* Stash the soft descriptor of the head then initialize it */
1556 	txbuf = &tx_ring->txbufs[wr_idx];
1557 
1558 	nfp_net_rx_give_one(dp, rx_ring, txbuf->frag, txbuf->dma_addr);
1559 
1560 	txbuf->frag = rxbuf->frag;
1561 	txbuf->dma_addr = rxbuf->dma_addr;
1562 	txbuf->fidx = -1;
1563 	txbuf->pkt_cnt = 1;
1564 	txbuf->real_len = pkt_len;
1565 
1566 	dma_sync_single_for_device(dp->dev, rxbuf->dma_addr + dma_off,
1567 				   pkt_len, DMA_BIDIRECTIONAL);
1568 
1569 	/* Build TX descriptor */
1570 	txd = &tx_ring->txds[wr_idx];
1571 	txd->offset_eop = PCIE_DESC_TX_EOP;
1572 	txd->dma_len = cpu_to_le16(pkt_len);
1573 	nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + dma_off);
1574 	txd->data_len = cpu_to_le16(pkt_len);
1575 
1576 	txd->flags = 0;
1577 	txd->mss = 0;
1578 	txd->lso_hdrlen = 0;
1579 
1580 	tx_ring->wr_p++;
1581 	tx_ring->wr_ptr_add++;
1582 	return true;
1583 }
1584 
1585 /**
1586  * nfp_net_rx() - receive up to @budget packets on @rx_ring
1587  * @rx_ring:   RX ring to receive from
1588  * @budget:    NAPI budget
1589  *
1590  * Note, this function is separated out from the napi poll function to
1591  * more cleanly separate packet receive code from other bookkeeping
1592  * functions performed in the napi poll function.
1593  *
1594  * Return: Number of packets received.
1595  */
1596 static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1597 {
1598 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1599 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1600 	struct nfp_net_tx_ring *tx_ring;
1601 	struct bpf_prog *xdp_prog;
1602 	bool xdp_tx_cmpl = false;
1603 	unsigned int true_bufsz;
1604 	struct sk_buff *skb;
1605 	int pkts_polled = 0;
1606 	int idx;
1607 
1608 	rcu_read_lock();
1609 	xdp_prog = READ_ONCE(dp->xdp_prog);
1610 	true_bufsz = xdp_prog ? PAGE_SIZE : dp->fl_bufsz;
1611 	tx_ring = r_vec->xdp_ring;
1612 
1613 	while (pkts_polled < budget) {
1614 		unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1615 		struct nfp_net_rx_buf *rxbuf;
1616 		struct nfp_net_rx_desc *rxd;
1617 		struct nfp_meta_parsed meta;
1618 		struct net_device *netdev;
1619 		dma_addr_t new_dma_addr;
1620 		u32 meta_len_xdp = 0;
1621 		void *new_frag;
1622 
1623 		idx = D_IDX(rx_ring, rx_ring->rd_p);
1624 
1625 		rxd = &rx_ring->rxds[idx];
1626 		if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1627 			break;
1628 
1629 		/* Memory barrier to ensure that we won't do other reads
1630 		 * before the DD bit.
1631 		 */
1632 		dma_rmb();
1633 
1634 		memset(&meta, 0, sizeof(meta));
1635 
1636 		rx_ring->rd_p++;
1637 		pkts_polled++;
1638 
1639 		rxbuf =	&rx_ring->rxbufs[idx];
1640 		/*         < meta_len >
1641 		 *  <-- [rx_offset] -->
1642 		 *  ---------------------------------------------------------
1643 		 * | [XX] |  metadata  |             packet           | XXXX |
1644 		 *  ---------------------------------------------------------
1645 		 *         <---------------- data_len --------------->
1646 		 *
1647 		 * The rx_offset is fixed for all packets, the meta_len can vary
1648 		 * on a packet by packet basis. If rx_offset is set to zero
1649 		 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1650 		 * buffer and is immediately followed by the packet (no [XX]).
1651 		 */
1652 		meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1653 		data_len = le16_to_cpu(rxd->rxd.data_len);
1654 		pkt_len = data_len - meta_len;
1655 
1656 		pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1657 		if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1658 			pkt_off += meta_len;
1659 		else
1660 			pkt_off += dp->rx_offset;
1661 		meta_off = pkt_off - meta_len;
1662 
1663 		/* Stats update */
1664 		u64_stats_update_begin(&r_vec->rx_sync);
1665 		r_vec->rx_pkts++;
1666 		r_vec->rx_bytes += pkt_len;
1667 		u64_stats_update_end(&r_vec->rx_sync);
1668 
1669 		if (unlikely(meta_len > NFP_NET_MAX_PREPEND ||
1670 			     (dp->rx_offset && meta_len > dp->rx_offset))) {
1671 			nn_dp_warn(dp, "oversized RX packet metadata %u\n",
1672 				   meta_len);
1673 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1674 			continue;
1675 		}
1676 
1677 		nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,
1678 					data_len);
1679 
1680 		if (!dp->chained_metadata_format) {
1681 			nfp_net_set_hash_desc(dp->netdev, &meta,
1682 					      rxbuf->frag + meta_off, rxd);
1683 		} else if (meta_len) {
1684 			void *end;
1685 
1686 			end = nfp_net_parse_meta(dp->netdev, &meta,
1687 						 rxbuf->frag + meta_off,
1688 						 meta_len);
1689 			if (unlikely(end != rxbuf->frag + pkt_off)) {
1690 				nn_dp_warn(dp, "invalid RX packet metadata\n");
1691 				nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1692 						NULL);
1693 				continue;
1694 			}
1695 		}
1696 
1697 		if (xdp_prog && !(rxd->rxd.flags & PCIE_DESC_RX_BPF &&
1698 				  dp->bpf_offload_xdp) && !meta.portid) {
1699 			void *orig_data = rxbuf->frag + pkt_off;
1700 			unsigned int dma_off;
1701 			struct xdp_buff xdp;
1702 			int act;
1703 
1704 			xdp.data_hard_start = rxbuf->frag + NFP_NET_RX_BUF_HEADROOM;
1705 			xdp.data = orig_data;
1706 			xdp.data_meta = orig_data;
1707 			xdp.data_end = orig_data + pkt_len;
1708 
1709 			act = bpf_prog_run_xdp(xdp_prog, &xdp);
1710 
1711 			pkt_len -= xdp.data - orig_data;
1712 			pkt_off += xdp.data - orig_data;
1713 
1714 			switch (act) {
1715 			case XDP_PASS:
1716 				meta_len_xdp = xdp.data - xdp.data_meta;
1717 				break;
1718 			case XDP_TX:
1719 				dma_off = pkt_off - NFP_NET_RX_BUF_HEADROOM;
1720 				if (unlikely(!nfp_net_tx_xdp_buf(dp, rx_ring,
1721 								 tx_ring, rxbuf,
1722 								 dma_off,
1723 								 pkt_len,
1724 								 &xdp_tx_cmpl)))
1725 					trace_xdp_exception(dp->netdev,
1726 							    xdp_prog, act);
1727 				continue;
1728 			default:
1729 				bpf_warn_invalid_xdp_action(act);
1730 				/* fall through */
1731 			case XDP_ABORTED:
1732 				trace_xdp_exception(dp->netdev, xdp_prog, act);
1733 				/* fall through */
1734 			case XDP_DROP:
1735 				nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1736 						    rxbuf->dma_addr);
1737 				continue;
1738 			}
1739 		}
1740 
1741 		skb = build_skb(rxbuf->frag, true_bufsz);
1742 		if (unlikely(!skb)) {
1743 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1744 			continue;
1745 		}
1746 		new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
1747 		if (unlikely(!new_frag)) {
1748 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
1749 			continue;
1750 		}
1751 
1752 		nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
1753 
1754 		nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
1755 
1756 		if (likely(!meta.portid)) {
1757 			netdev = dp->netdev;
1758 		} else {
1759 			struct nfp_net *nn;
1760 
1761 			nn = netdev_priv(dp->netdev);
1762 			netdev = nfp_app_repr_get(nn->app, meta.portid);
1763 			if (unlikely(!netdev)) {
1764 				nfp_net_rx_drop(dp, r_vec, rx_ring, NULL, skb);
1765 				continue;
1766 			}
1767 			nfp_repr_inc_rx_stats(netdev, pkt_len);
1768 		}
1769 
1770 		skb_reserve(skb, pkt_off);
1771 		skb_put(skb, pkt_len);
1772 
1773 		skb->mark = meta.mark;
1774 		skb_set_hash(skb, meta.hash, meta.hash_type);
1775 
1776 		skb_record_rx_queue(skb, rx_ring->idx);
1777 		skb->protocol = eth_type_trans(skb, netdev);
1778 
1779 		nfp_net_rx_csum(dp, r_vec, rxd, &meta, skb);
1780 
1781 		if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
1782 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1783 					       le16_to_cpu(rxd->rxd.vlan));
1784 		if (meta_len_xdp)
1785 			skb_metadata_set(skb, meta_len_xdp);
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 	nfp_port_configure(netdev, false);
2663 
2664 	/* Step 3: Free resources
2665 	 */
2666 	nfp_net_close_free_all(nn);
2667 
2668 	nn_dbg(nn, "%s down", netdev->name);
2669 	return 0;
2670 }
2671 
2672 void nfp_ctrl_close(struct nfp_net *nn)
2673 {
2674 	int r;
2675 
2676 	rtnl_lock();
2677 
2678 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2679 		disable_irq(nn->r_vecs[r].irq_vector);
2680 		tasklet_disable(&nn->r_vecs[r].tasklet);
2681 	}
2682 
2683 	nfp_net_clear_config_and_disable(nn);
2684 
2685 	nfp_net_close_free_all(nn);
2686 
2687 	rtnl_unlock();
2688 }
2689 
2690 /**
2691  * nfp_net_open_stack() - Start the device from stack's perspective
2692  * @nn:      NFP Net device to reconfigure
2693  */
2694 static void nfp_net_open_stack(struct nfp_net *nn)
2695 {
2696 	unsigned int r;
2697 
2698 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2699 		napi_enable(&nn->r_vecs[r].napi);
2700 		enable_irq(nn->r_vecs[r].irq_vector);
2701 	}
2702 
2703 	netif_tx_wake_all_queues(nn->dp.netdev);
2704 
2705 	enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2706 	nfp_net_read_link_status(nn);
2707 }
2708 
2709 static int nfp_net_open_alloc_all(struct nfp_net *nn)
2710 {
2711 	int err, r;
2712 
2713 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2714 				      nn->exn_name, sizeof(nn->exn_name),
2715 				      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2716 	if (err)
2717 		return err;
2718 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2719 				      nn->lsc_name, sizeof(nn->lsc_name),
2720 				      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2721 	if (err)
2722 		goto err_free_exn;
2723 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2724 
2725 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2726 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2727 		if (err)
2728 			goto err_cleanup_vec_p;
2729 	}
2730 
2731 	err = nfp_net_rx_rings_prepare(nn, &nn->dp);
2732 	if (err)
2733 		goto err_cleanup_vec;
2734 
2735 	err = nfp_net_tx_rings_prepare(nn, &nn->dp);
2736 	if (err)
2737 		goto err_free_rx_rings;
2738 
2739 	for (r = 0; r < nn->max_r_vecs; r++)
2740 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2741 
2742 	return 0;
2743 
2744 err_free_rx_rings:
2745 	nfp_net_rx_rings_free(&nn->dp);
2746 err_cleanup_vec:
2747 	r = nn->dp.num_r_vecs;
2748 err_cleanup_vec_p:
2749 	while (r--)
2750 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2751 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2752 err_free_exn:
2753 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2754 	return err;
2755 }
2756 
2757 static int nfp_net_netdev_open(struct net_device *netdev)
2758 {
2759 	struct nfp_net *nn = netdev_priv(netdev);
2760 	int err;
2761 
2762 	/* Step 1: Allocate resources for rings and the like
2763 	 * - Request interrupts
2764 	 * - Allocate RX and TX ring resources
2765 	 * - Setup initial RSS table
2766 	 */
2767 	err = nfp_net_open_alloc_all(nn);
2768 	if (err)
2769 		return err;
2770 
2771 	err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
2772 	if (err)
2773 		goto err_free_all;
2774 
2775 	err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
2776 	if (err)
2777 		goto err_free_all;
2778 
2779 	/* Step 2: Configure the NFP
2780 	 * - Ifup the physical interface if it exists
2781 	 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2782 	 * - Write MAC address (in case it changed)
2783 	 * - Set the MTU
2784 	 * - Set the Freelist buffer size
2785 	 * - Enable the FW
2786 	 */
2787 	err = nfp_port_configure(netdev, true);
2788 	if (err)
2789 		goto err_free_all;
2790 
2791 	err = nfp_net_set_config_and_enable(nn);
2792 	if (err)
2793 		goto err_port_disable;
2794 
2795 	/* Step 3: Enable for kernel
2796 	 * - put some freelist descriptors on each RX ring
2797 	 * - enable NAPI on each ring
2798 	 * - enable all TX queues
2799 	 * - set link state
2800 	 */
2801 	nfp_net_open_stack(nn);
2802 
2803 	return 0;
2804 
2805 err_port_disable:
2806 	nfp_port_configure(netdev, false);
2807 err_free_all:
2808 	nfp_net_close_free_all(nn);
2809 	return err;
2810 }
2811 
2812 int nfp_ctrl_open(struct nfp_net *nn)
2813 {
2814 	int err, r;
2815 
2816 	/* ring dumping depends on vNICs being opened/closed under rtnl */
2817 	rtnl_lock();
2818 
2819 	err = nfp_net_open_alloc_all(nn);
2820 	if (err)
2821 		goto err_unlock;
2822 
2823 	err = nfp_net_set_config_and_enable(nn);
2824 	if (err)
2825 		goto err_free_all;
2826 
2827 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2828 		enable_irq(nn->r_vecs[r].irq_vector);
2829 
2830 	rtnl_unlock();
2831 
2832 	return 0;
2833 
2834 err_free_all:
2835 	nfp_net_close_free_all(nn);
2836 err_unlock:
2837 	rtnl_unlock();
2838 	return err;
2839 }
2840 
2841 static void nfp_net_set_rx_mode(struct net_device *netdev)
2842 {
2843 	struct nfp_net *nn = netdev_priv(netdev);
2844 	u32 new_ctrl;
2845 
2846 	new_ctrl = nn->dp.ctrl;
2847 
2848 	if (netdev->flags & IFF_PROMISC) {
2849 		if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
2850 			new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
2851 		else
2852 			nn_warn(nn, "FW does not support promiscuous mode\n");
2853 	} else {
2854 		new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
2855 	}
2856 
2857 	if (new_ctrl == nn->dp.ctrl)
2858 		return;
2859 
2860 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2861 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
2862 
2863 	nn->dp.ctrl = new_ctrl;
2864 }
2865 
2866 static void nfp_net_rss_init_itbl(struct nfp_net *nn)
2867 {
2868 	int i;
2869 
2870 	for (i = 0; i < sizeof(nn->rss_itbl); i++)
2871 		nn->rss_itbl[i] =
2872 			ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
2873 }
2874 
2875 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
2876 {
2877 	struct nfp_net_dp new_dp = *dp;
2878 
2879 	*dp = nn->dp;
2880 	nn->dp = new_dp;
2881 
2882 	nn->dp.netdev->mtu = new_dp.mtu;
2883 
2884 	if (!netif_is_rxfh_configured(nn->dp.netdev))
2885 		nfp_net_rss_init_itbl(nn);
2886 }
2887 
2888 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
2889 {
2890 	unsigned int r;
2891 	int err;
2892 
2893 	nfp_net_dp_swap(nn, dp);
2894 
2895 	for (r = 0; r <	nn->max_r_vecs; r++)
2896 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2897 
2898 	err = netif_set_real_num_rx_queues(nn->dp.netdev, nn->dp.num_rx_rings);
2899 	if (err)
2900 		return err;
2901 
2902 	if (nn->dp.netdev->real_num_tx_queues != nn->dp.num_stack_tx_rings) {
2903 		err = netif_set_real_num_tx_queues(nn->dp.netdev,
2904 						   nn->dp.num_stack_tx_rings);
2905 		if (err)
2906 			return err;
2907 	}
2908 
2909 	return nfp_net_set_config_and_enable(nn);
2910 }
2911 
2912 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
2913 {
2914 	struct nfp_net_dp *new;
2915 
2916 	new = kmalloc(sizeof(*new), GFP_KERNEL);
2917 	if (!new)
2918 		return NULL;
2919 
2920 	*new = nn->dp;
2921 
2922 	/* Clear things which need to be recomputed */
2923 	new->fl_bufsz = 0;
2924 	new->tx_rings = NULL;
2925 	new->rx_rings = NULL;
2926 	new->num_r_vecs = 0;
2927 	new->num_stack_tx_rings = 0;
2928 
2929 	return new;
2930 }
2931 
2932 static int
2933 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
2934 		     struct netlink_ext_ack *extack)
2935 {
2936 	/* XDP-enabled tests */
2937 	if (!dp->xdp_prog)
2938 		return 0;
2939 	if (dp->fl_bufsz > PAGE_SIZE) {
2940 		NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
2941 		return -EINVAL;
2942 	}
2943 	if (dp->num_tx_rings > nn->max_tx_rings) {
2944 		NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
2945 		return -EINVAL;
2946 	}
2947 
2948 	return 0;
2949 }
2950 
2951 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
2952 			  struct netlink_ext_ack *extack)
2953 {
2954 	int r, err;
2955 
2956 	dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
2957 
2958 	dp->num_stack_tx_rings = dp->num_tx_rings;
2959 	if (dp->xdp_prog)
2960 		dp->num_stack_tx_rings -= dp->num_rx_rings;
2961 
2962 	dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
2963 
2964 	err = nfp_net_check_config(nn, dp, extack);
2965 	if (err)
2966 		goto exit_free_dp;
2967 
2968 	if (!netif_running(dp->netdev)) {
2969 		nfp_net_dp_swap(nn, dp);
2970 		err = 0;
2971 		goto exit_free_dp;
2972 	}
2973 
2974 	/* Prepare new rings */
2975 	for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
2976 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2977 		if (err) {
2978 			dp->num_r_vecs = r;
2979 			goto err_cleanup_vecs;
2980 		}
2981 	}
2982 
2983 	err = nfp_net_rx_rings_prepare(nn, dp);
2984 	if (err)
2985 		goto err_cleanup_vecs;
2986 
2987 	err = nfp_net_tx_rings_prepare(nn, dp);
2988 	if (err)
2989 		goto err_free_rx;
2990 
2991 	/* Stop device, swap in new rings, try to start the firmware */
2992 	nfp_net_close_stack(nn);
2993 	nfp_net_clear_config_and_disable(nn);
2994 
2995 	err = nfp_net_dp_swap_enable(nn, dp);
2996 	if (err) {
2997 		int err2;
2998 
2999 		nfp_net_clear_config_and_disable(nn);
3000 
3001 		/* Try with old configuration and old rings */
3002 		err2 = nfp_net_dp_swap_enable(nn, dp);
3003 		if (err2)
3004 			nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
3005 			       err, err2);
3006 	}
3007 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3008 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3009 
3010 	nfp_net_rx_rings_free(dp);
3011 	nfp_net_tx_rings_free(dp);
3012 
3013 	nfp_net_open_stack(nn);
3014 exit_free_dp:
3015 	kfree(dp);
3016 
3017 	return err;
3018 
3019 err_free_rx:
3020 	nfp_net_rx_rings_free(dp);
3021 err_cleanup_vecs:
3022 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3023 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3024 	kfree(dp);
3025 	return err;
3026 }
3027 
3028 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
3029 {
3030 	struct nfp_net *nn = netdev_priv(netdev);
3031 	struct nfp_net_dp *dp;
3032 
3033 	dp = nfp_net_clone_dp(nn);
3034 	if (!dp)
3035 		return -ENOMEM;
3036 
3037 	dp->mtu = new_mtu;
3038 
3039 	return nfp_net_ring_reconfig(nn, dp, NULL);
3040 }
3041 
3042 static int
3043 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
3044 {
3045 	struct nfp_net *nn = netdev_priv(netdev);
3046 
3047 	/* Priority tagged packets with vlan id 0 are processed by the
3048 	 * NFP as untagged packets
3049 	 */
3050 	if (!vid)
3051 		return 0;
3052 
3053 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_VID, vid);
3054 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_PROTO, ETH_P_8021Q);
3055 
3056 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD);
3057 }
3058 
3059 static int
3060 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
3061 {
3062 	struct nfp_net *nn = netdev_priv(netdev);
3063 
3064 	/* Priority tagged packets with vlan id 0 are processed by the
3065 	 * NFP as untagged packets
3066 	 */
3067 	if (!vid)
3068 		return 0;
3069 
3070 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_VID, vid);
3071 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_PROTO, ETH_P_8021Q);
3072 
3073 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL);
3074 }
3075 
3076 static void nfp_net_stat64(struct net_device *netdev,
3077 			   struct rtnl_link_stats64 *stats)
3078 {
3079 	struct nfp_net *nn = netdev_priv(netdev);
3080 	int r;
3081 
3082 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
3083 		struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
3084 		u64 data[3];
3085 		unsigned int start;
3086 
3087 		do {
3088 			start = u64_stats_fetch_begin(&r_vec->rx_sync);
3089 			data[0] = r_vec->rx_pkts;
3090 			data[1] = r_vec->rx_bytes;
3091 			data[2] = r_vec->rx_drops;
3092 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
3093 		stats->rx_packets += data[0];
3094 		stats->rx_bytes += data[1];
3095 		stats->rx_dropped += data[2];
3096 
3097 		do {
3098 			start = u64_stats_fetch_begin(&r_vec->tx_sync);
3099 			data[0] = r_vec->tx_pkts;
3100 			data[1] = r_vec->tx_bytes;
3101 			data[2] = r_vec->tx_errors;
3102 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
3103 		stats->tx_packets += data[0];
3104 		stats->tx_bytes += data[1];
3105 		stats->tx_errors += data[2];
3106 	}
3107 }
3108 
3109 static int nfp_net_set_features(struct net_device *netdev,
3110 				netdev_features_t features)
3111 {
3112 	netdev_features_t changed = netdev->features ^ features;
3113 	struct nfp_net *nn = netdev_priv(netdev);
3114 	u32 new_ctrl;
3115 	int err;
3116 
3117 	/* Assume this is not called with features we have not advertised */
3118 
3119 	new_ctrl = nn->dp.ctrl;
3120 
3121 	if (changed & NETIF_F_RXCSUM) {
3122 		if (features & NETIF_F_RXCSUM)
3123 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3124 		else
3125 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
3126 	}
3127 
3128 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3129 		if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
3130 			new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3131 		else
3132 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
3133 	}
3134 
3135 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
3136 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
3137 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3138 					      NFP_NET_CFG_CTRL_LSO;
3139 		else
3140 			new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3141 	}
3142 
3143 	if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
3144 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3145 			new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3146 		else
3147 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
3148 	}
3149 
3150 	if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
3151 		if (features & NETIF_F_HW_VLAN_CTAG_TX)
3152 			new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3153 		else
3154 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
3155 	}
3156 
3157 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
3158 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
3159 			new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3160 		else
3161 			new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
3162 	}
3163 
3164 	if (changed & NETIF_F_SG) {
3165 		if (features & NETIF_F_SG)
3166 			new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
3167 		else
3168 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
3169 	}
3170 
3171 	if (changed & NETIF_F_HW_TC && nfp_app_tc_busy(nn->app, nn)) {
3172 		nn_err(nn, "Cannot disable HW TC offload while in use\n");
3173 		return -EBUSY;
3174 	}
3175 
3176 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
3177 	       netdev->features, features, changed);
3178 
3179 	if (new_ctrl == nn->dp.ctrl)
3180 		return 0;
3181 
3182 	nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
3183 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3184 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
3185 	if (err)
3186 		return err;
3187 
3188 	nn->dp.ctrl = new_ctrl;
3189 
3190 	return 0;
3191 }
3192 
3193 static netdev_features_t
3194 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
3195 		       netdev_features_t features)
3196 {
3197 	u8 l4_hdr;
3198 
3199 	/* We can't do TSO over double tagged packets (802.1AD) */
3200 	features &= vlan_features_check(skb, features);
3201 
3202 	if (!skb->encapsulation)
3203 		return features;
3204 
3205 	/* Ensure that inner L4 header offset fits into TX descriptor field */
3206 	if (skb_is_gso(skb)) {
3207 		u32 hdrlen;
3208 
3209 		hdrlen = skb_inner_transport_header(skb) - skb->data +
3210 			inner_tcp_hdrlen(skb);
3211 
3212 		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
3213 			features &= ~NETIF_F_GSO_MASK;
3214 	}
3215 
3216 	/* VXLAN/GRE check */
3217 	switch (vlan_get_protocol(skb)) {
3218 	case htons(ETH_P_IP):
3219 		l4_hdr = ip_hdr(skb)->protocol;
3220 		break;
3221 	case htons(ETH_P_IPV6):
3222 		l4_hdr = ipv6_hdr(skb)->nexthdr;
3223 		break;
3224 	default:
3225 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3226 	}
3227 
3228 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
3229 	    skb->inner_protocol != htons(ETH_P_TEB) ||
3230 	    (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
3231 	    (l4_hdr == IPPROTO_UDP &&
3232 	     (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
3233 	      sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
3234 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3235 
3236 	return features;
3237 }
3238 
3239 /**
3240  * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
3241  * @nn:   NFP Net device to reconfigure
3242  * @idx:  Index into the port table where new port should be written
3243  * @port: UDP port to configure (pass zero to remove VXLAN port)
3244  */
3245 static void nfp_net_set_vxlan_port(struct nfp_net *nn, int idx, __be16 port)
3246 {
3247 	int i;
3248 
3249 	nn->vxlan_ports[idx] = port;
3250 
3251 	if (!(nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN))
3252 		return;
3253 
3254 	BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
3255 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2)
3256 		nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(port),
3257 			  be16_to_cpu(nn->vxlan_ports[i + 1]) << 16 |
3258 			  be16_to_cpu(nn->vxlan_ports[i]));
3259 
3260 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_VXLAN);
3261 }
3262 
3263 /**
3264  * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
3265  * @nn:   NFP Network structure
3266  * @port: UDP port to look for
3267  *
3268  * Return: if the port is already in the table -- it's position;
3269  *	   if the port is not in the table -- free position to use;
3270  *	   if the table is full -- -ENOSPC.
3271  */
3272 static int nfp_net_find_vxlan_idx(struct nfp_net *nn, __be16 port)
3273 {
3274 	int i, free_idx = -ENOSPC;
3275 
3276 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i++) {
3277 		if (nn->vxlan_ports[i] == port)
3278 			return i;
3279 		if (!nn->vxlan_usecnt[i])
3280 			free_idx = i;
3281 	}
3282 
3283 	return free_idx;
3284 }
3285 
3286 static void nfp_net_add_vxlan_port(struct net_device *netdev,
3287 				   struct udp_tunnel_info *ti)
3288 {
3289 	struct nfp_net *nn = netdev_priv(netdev);
3290 	int idx;
3291 
3292 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3293 		return;
3294 
3295 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3296 	if (idx == -ENOSPC)
3297 		return;
3298 
3299 	if (!nn->vxlan_usecnt[idx]++)
3300 		nfp_net_set_vxlan_port(nn, idx, ti->port);
3301 }
3302 
3303 static void nfp_net_del_vxlan_port(struct net_device *netdev,
3304 				   struct udp_tunnel_info *ti)
3305 {
3306 	struct nfp_net *nn = netdev_priv(netdev);
3307 	int idx;
3308 
3309 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3310 		return;
3311 
3312 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3313 	if (idx == -ENOSPC || !nn->vxlan_usecnt[idx])
3314 		return;
3315 
3316 	if (!--nn->vxlan_usecnt[idx])
3317 		nfp_net_set_vxlan_port(nn, idx, 0);
3318 }
3319 
3320 static int
3321 nfp_net_xdp_setup_drv(struct nfp_net *nn, struct bpf_prog *prog,
3322 		      struct netlink_ext_ack *extack)
3323 {
3324 	struct nfp_net_dp *dp;
3325 
3326 	if (!prog == !nn->dp.xdp_prog) {
3327 		WRITE_ONCE(nn->dp.xdp_prog, prog);
3328 		return 0;
3329 	}
3330 
3331 	dp = nfp_net_clone_dp(nn);
3332 	if (!dp)
3333 		return -ENOMEM;
3334 
3335 	dp->xdp_prog = prog;
3336 	dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
3337 	dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
3338 	dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
3339 
3340 	/* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
3341 	return nfp_net_ring_reconfig(nn, dp, extack);
3342 }
3343 
3344 static int
3345 nfp_net_xdp_setup(struct nfp_net *nn, struct bpf_prog *prog, u32 flags,
3346 		  struct netlink_ext_ack *extack)
3347 {
3348 	struct bpf_prog *drv_prog, *offload_prog;
3349 	int err;
3350 
3351 	if (nn->xdp_prog && (flags ^ nn->xdp_flags) & XDP_FLAGS_MODES)
3352 		return -EBUSY;
3353 
3354 	/* Load both when no flags set to allow easy activation of driver path
3355 	 * when program is replaced by one which can't be offloaded.
3356 	 */
3357 	drv_prog     = flags & XDP_FLAGS_HW_MODE  ? NULL : prog;
3358 	offload_prog = flags & XDP_FLAGS_DRV_MODE ? NULL : prog;
3359 
3360 	err = nfp_net_xdp_setup_drv(nn, drv_prog, extack);
3361 	if (err)
3362 		return err;
3363 
3364 	err = nfp_app_xdp_offload(nn->app, nn, offload_prog);
3365 	if (err && flags & XDP_FLAGS_HW_MODE)
3366 		return err;
3367 
3368 	if (nn->xdp_prog)
3369 		bpf_prog_put(nn->xdp_prog);
3370 	nn->xdp_prog = prog;
3371 	nn->xdp_flags = flags;
3372 
3373 	return 0;
3374 }
3375 
3376 static int nfp_net_xdp(struct net_device *netdev, struct netdev_xdp *xdp)
3377 {
3378 	struct nfp_net *nn = netdev_priv(netdev);
3379 
3380 	switch (xdp->command) {
3381 	case XDP_SETUP_PROG:
3382 	case XDP_SETUP_PROG_HW:
3383 		return nfp_net_xdp_setup(nn, xdp->prog, xdp->flags,
3384 					 xdp->extack);
3385 	case XDP_QUERY_PROG:
3386 		xdp->prog_attached = !!nn->xdp_prog;
3387 		if (nn->dp.bpf_offload_xdp)
3388 			xdp->prog_attached = XDP_ATTACHED_HW;
3389 		xdp->prog_id = nn->xdp_prog ? nn->xdp_prog->aux->id : 0;
3390 		return 0;
3391 	default:
3392 		return -EINVAL;
3393 	}
3394 }
3395 
3396 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
3397 {
3398 	struct nfp_net *nn = netdev_priv(netdev);
3399 	struct sockaddr *saddr = addr;
3400 	int err;
3401 
3402 	err = eth_prepare_mac_addr_change(netdev, addr);
3403 	if (err)
3404 		return err;
3405 
3406 	nfp_net_write_mac_addr(nn, saddr->sa_data);
3407 
3408 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
3409 	if (err)
3410 		return err;
3411 
3412 	eth_commit_mac_addr_change(netdev, addr);
3413 
3414 	return 0;
3415 }
3416 
3417 const struct net_device_ops nfp_net_netdev_ops = {
3418 	.ndo_open		= nfp_net_netdev_open,
3419 	.ndo_stop		= nfp_net_netdev_close,
3420 	.ndo_start_xmit		= nfp_net_tx,
3421 	.ndo_get_stats64	= nfp_net_stat64,
3422 	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
3423 	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
3424 	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
3425 	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
3426 	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
3427 	.ndo_get_vf_config	= nfp_app_get_vf_config,
3428 	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
3429 	.ndo_setup_tc		= nfp_port_setup_tc,
3430 	.ndo_tx_timeout		= nfp_net_tx_timeout,
3431 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
3432 	.ndo_change_mtu		= nfp_net_change_mtu,
3433 	.ndo_set_mac_address	= nfp_net_set_mac_address,
3434 	.ndo_set_features	= nfp_net_set_features,
3435 	.ndo_features_check	= nfp_net_features_check,
3436 	.ndo_get_phys_port_name	= nfp_port_get_phys_port_name,
3437 	.ndo_udp_tunnel_add	= nfp_net_add_vxlan_port,
3438 	.ndo_udp_tunnel_del	= nfp_net_del_vxlan_port,
3439 	.ndo_xdp		= nfp_net_xdp,
3440 };
3441 
3442 /**
3443  * nfp_net_info() - Print general info about the NIC
3444  * @nn:      NFP Net device to reconfigure
3445  */
3446 void nfp_net_info(struct nfp_net *nn)
3447 {
3448 	nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
3449 		nn->dp.is_vf ? "VF " : "",
3450 		nn->dp.num_tx_rings, nn->max_tx_rings,
3451 		nn->dp.num_rx_rings, nn->max_rx_rings);
3452 	nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3453 		nn->fw_ver.resv, nn->fw_ver.class,
3454 		nn->fw_ver.major, nn->fw_ver.minor,
3455 		nn->max_mtu);
3456 	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",
3457 		nn->cap,
3458 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
3459 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
3460 		nn->cap & NFP_NET_CFG_CTRL_L2MC     ? "L2MCFILT " : "",
3461 		nn->cap & NFP_NET_CFG_CTRL_RXCSUM   ? "RXCSUM "   : "",
3462 		nn->cap & NFP_NET_CFG_CTRL_TXCSUM   ? "TXCSUM "   : "",
3463 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN   ? "RXVLAN "   : "",
3464 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN   ? "TXVLAN "   : "",
3465 		nn->cap & NFP_NET_CFG_CTRL_SCATTER  ? "SCATTER "  : "",
3466 		nn->cap & NFP_NET_CFG_CTRL_GATHER   ? "GATHER "   : "",
3467 		nn->cap & NFP_NET_CFG_CTRL_LSO      ? "TSO1 "     : "",
3468 		nn->cap & NFP_NET_CFG_CTRL_LSO2     ? "TSO2 "     : "",
3469 		nn->cap & NFP_NET_CFG_CTRL_RSS      ? "RSS1 "     : "",
3470 		nn->cap & NFP_NET_CFG_CTRL_RSS2     ? "RSS2 "     : "",
3471 		nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "",
3472 		nn->cap & NFP_NET_CFG_CTRL_L2SWITCH ? "L2SWITCH " : "",
3473 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
3474 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
3475 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
3476 		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
3477 		nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ?
3478 						      "RXCSUM_COMPLETE " : "",
3479 		nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "",
3480 		nfp_app_extra_cap(nn->app, nn));
3481 }
3482 
3483 /**
3484  * nfp_net_alloc() - Allocate netdev and related structure
3485  * @pdev:         PCI device
3486  * @needs_netdev: Whether to allocate a netdev for this vNIC
3487  * @max_tx_rings: Maximum number of TX rings supported by device
3488  * @max_rx_rings: Maximum number of RX rings supported by device
3489  *
3490  * This function allocates a netdev device and fills in the initial
3491  * part of the @struct nfp_net structure.  In case of control device
3492  * nfp_net structure is allocated without the netdev.
3493  *
3494  * Return: NFP Net device structure, or ERR_PTR on error.
3495  */
3496 struct nfp_net *nfp_net_alloc(struct pci_dev *pdev, bool needs_netdev,
3497 			      unsigned int max_tx_rings,
3498 			      unsigned int max_rx_rings)
3499 {
3500 	struct nfp_net *nn;
3501 
3502 	if (needs_netdev) {
3503 		struct net_device *netdev;
3504 
3505 		netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
3506 					    max_tx_rings, max_rx_rings);
3507 		if (!netdev)
3508 			return ERR_PTR(-ENOMEM);
3509 
3510 		SET_NETDEV_DEV(netdev, &pdev->dev);
3511 		nn = netdev_priv(netdev);
3512 		nn->dp.netdev = netdev;
3513 	} else {
3514 		nn = vzalloc(sizeof(*nn));
3515 		if (!nn)
3516 			return ERR_PTR(-ENOMEM);
3517 	}
3518 
3519 	nn->dp.dev = &pdev->dev;
3520 	nn->pdev = pdev;
3521 
3522 	nn->max_tx_rings = max_tx_rings;
3523 	nn->max_rx_rings = max_rx_rings;
3524 
3525 	nn->dp.num_tx_rings = min_t(unsigned int,
3526 				    max_tx_rings, num_online_cpus());
3527 	nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings,
3528 				 netif_get_num_default_rss_queues());
3529 
3530 	nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings);
3531 	nn->dp.num_r_vecs = min_t(unsigned int,
3532 				  nn->dp.num_r_vecs, num_online_cpus());
3533 
3534 	nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
3535 	nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
3536 
3537 	spin_lock_init(&nn->reconfig_lock);
3538 	spin_lock_init(&nn->link_status_lock);
3539 
3540 	timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0);
3541 
3542 	return nn;
3543 }
3544 
3545 /**
3546  * nfp_net_free() - Undo what @nfp_net_alloc() did
3547  * @nn:      NFP Net device to reconfigure
3548  */
3549 void nfp_net_free(struct nfp_net *nn)
3550 {
3551 	if (nn->xdp_prog)
3552 		bpf_prog_put(nn->xdp_prog);
3553 
3554 	if (nn->dp.netdev)
3555 		free_netdev(nn->dp.netdev);
3556 	else
3557 		vfree(nn);
3558 }
3559 
3560 /**
3561  * nfp_net_rss_key_sz() - Get current size of the RSS key
3562  * @nn:		NFP Net device instance
3563  *
3564  * Return: size of the RSS key for currently selected hash function.
3565  */
3566 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn)
3567 {
3568 	switch (nn->rss_hfunc) {
3569 	case ETH_RSS_HASH_TOP:
3570 		return NFP_NET_CFG_RSS_KEY_SZ;
3571 	case ETH_RSS_HASH_XOR:
3572 		return 0;
3573 	case ETH_RSS_HASH_CRC32:
3574 		return 4;
3575 	}
3576 
3577 	nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc);
3578 	return 0;
3579 }
3580 
3581 /**
3582  * nfp_net_rss_init() - Set the initial RSS parameters
3583  * @nn:	     NFP Net device to reconfigure
3584  */
3585 static void nfp_net_rss_init(struct nfp_net *nn)
3586 {
3587 	unsigned long func_bit, rss_cap_hfunc;
3588 	u32 reg;
3589 
3590 	/* Read the RSS function capability and select first supported func */
3591 	reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP);
3592 	rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg);
3593 	if (!rss_cap_hfunc)
3594 		rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC,
3595 					  NFP_NET_CFG_RSS_TOEPLITZ);
3596 
3597 	func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS);
3598 	if (func_bit == NFP_NET_CFG_RSS_HFUNCS) {
3599 		dev_warn(nn->dp.dev,
3600 			 "Bad RSS config, defaulting to Toeplitz hash\n");
3601 		func_bit = ETH_RSS_HASH_TOP_BIT;
3602 	}
3603 	nn->rss_hfunc = 1 << func_bit;
3604 
3605 	netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn));
3606 
3607 	nfp_net_rss_init_itbl(nn);
3608 
3609 	/* Enable IPv4/IPv6 TCP by default */
3610 	nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
3611 		      NFP_NET_CFG_RSS_IPV6_TCP |
3612 		      FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) |
3613 		      NFP_NET_CFG_RSS_MASK;
3614 }
3615 
3616 /**
3617  * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
3618  * @nn:	     NFP Net device to reconfigure
3619  */
3620 static void nfp_net_irqmod_init(struct nfp_net *nn)
3621 {
3622 	nn->rx_coalesce_usecs      = 50;
3623 	nn->rx_coalesce_max_frames = 64;
3624 	nn->tx_coalesce_usecs      = 50;
3625 	nn->tx_coalesce_max_frames = 64;
3626 }
3627 
3628 static void nfp_net_netdev_init(struct nfp_net *nn)
3629 {
3630 	struct net_device *netdev = nn->dp.netdev;
3631 
3632 	nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
3633 
3634 	netdev->mtu = nn->dp.mtu;
3635 
3636 	/* Advertise/enable offloads based on capabilities
3637 	 *
3638 	 * Note: netdev->features show the currently enabled features
3639 	 * and netdev->hw_features advertises which features are
3640 	 * supported.  By default we enable most features.
3641 	 */
3642 	if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
3643 		netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3644 
3645 	netdev->hw_features = NETIF_F_HIGHDMA;
3646 	if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) {
3647 		netdev->hw_features |= NETIF_F_RXCSUM;
3648 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3649 	}
3650 	if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
3651 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3652 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3653 	}
3654 	if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
3655 		netdev->hw_features |= NETIF_F_SG;
3656 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER;
3657 	}
3658 	if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
3659 	    nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3660 		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3661 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3662 					 NFP_NET_CFG_CTRL_LSO;
3663 	}
3664 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY)
3665 		netdev->hw_features |= NETIF_F_RXHASH;
3666 	if (nn->cap & NFP_NET_CFG_CTRL_VXLAN &&
3667 	    nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
3668 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3669 			netdev->hw_features |= NETIF_F_GSO_GRE |
3670 					       NETIF_F_GSO_UDP_TUNNEL;
3671 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE;
3672 
3673 		netdev->hw_enc_features = netdev->hw_features;
3674 	}
3675 
3676 	netdev->vlan_features = netdev->hw_features;
3677 
3678 	if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
3679 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3680 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3681 	}
3682 	if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
3683 		if (nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3684 			nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
3685 		} else {
3686 			netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3687 			nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3688 		}
3689 	}
3690 	if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) {
3691 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3692 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3693 	}
3694 
3695 	netdev->features = netdev->hw_features;
3696 
3697 	if (nfp_app_has_tc(nn->app))
3698 		netdev->hw_features |= NETIF_F_HW_TC;
3699 
3700 	/* Advertise but disable TSO by default. */
3701 	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
3702 	nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3703 
3704 	/* Finalise the netdev setup */
3705 	netdev->netdev_ops = &nfp_net_netdev_ops;
3706 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
3707 
3708 	SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
3709 
3710 	/* MTU range: 68 - hw-specific max */
3711 	netdev->min_mtu = ETH_MIN_MTU;
3712 	netdev->max_mtu = nn->max_mtu;
3713 
3714 	netif_carrier_off(netdev);
3715 
3716 	nfp_net_set_ethtool_ops(netdev);
3717 }
3718 
3719 /**
3720  * nfp_net_init() - Initialise/finalise the nfp_net structure
3721  * @nn:		NFP Net device structure
3722  *
3723  * Return: 0 on success or negative errno on error.
3724  */
3725 int nfp_net_init(struct nfp_net *nn)
3726 {
3727 	int err;
3728 
3729 	nn->dp.rx_dma_dir = DMA_FROM_DEVICE;
3730 
3731 	/* Get some of the read-only fields from the BAR */
3732 	nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
3733 	nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
3734 
3735 	/* ABI 4.x and ctrl vNIC always use chained metadata, in other cases
3736 	 * we allow use of non-chained metadata if RSS(v1) is the only
3737 	 * advertised capability requiring metadata.
3738 	 */
3739 	nn->dp.chained_metadata_format = nn->fw_ver.major == 4 ||
3740 					 !nn->dp.netdev ||
3741 					 !(nn->cap & NFP_NET_CFG_CTRL_RSS) ||
3742 					 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META;
3743 	/* RSS(v1) uses non-chained metadata format, except in ABI 4.x where
3744 	 * it has the same meaning as RSSv2.
3745 	 */
3746 	if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4)
3747 		nn->cap &= ~NFP_NET_CFG_CTRL_RSS;
3748 
3749 	/* Determine RX packet/metadata boundary offset */
3750 	if (nn->fw_ver.major >= 2) {
3751 		u32 reg;
3752 
3753 		reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
3754 		if (reg > NFP_NET_MAX_PREPEND) {
3755 			nn_err(nn, "Invalid rx offset: %d\n", reg);
3756 			return -EINVAL;
3757 		}
3758 		nn->dp.rx_offset = reg;
3759 	} else {
3760 		nn->dp.rx_offset = NFP_NET_RX_OFFSET;
3761 	}
3762 
3763 	/* Set default MTU and Freelist buffer size */
3764 	if (nn->max_mtu < NFP_NET_DEFAULT_MTU)
3765 		nn->dp.mtu = nn->max_mtu;
3766 	else
3767 		nn->dp.mtu = NFP_NET_DEFAULT_MTU;
3768 	nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp);
3769 
3770 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) {
3771 		nfp_net_rss_init(nn);
3772 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?:
3773 					 NFP_NET_CFG_CTRL_RSS;
3774 	}
3775 
3776 	/* Allow L2 Broadcast and Multicast through by default, if supported */
3777 	if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
3778 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC;
3779 	if (nn->cap & NFP_NET_CFG_CTRL_L2MC)
3780 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2MC;
3781 
3782 	/* Allow IRQ moderation, if supported */
3783 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
3784 		nfp_net_irqmod_init(nn);
3785 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
3786 	}
3787 
3788 	if (nn->dp.netdev)
3789 		nfp_net_netdev_init(nn);
3790 
3791 	/* Stash the re-configuration queue away.  First odd queue in TX Bar */
3792 	nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
3793 
3794 	/* Make sure the FW knows the netdev is supposed to be disabled here */
3795 	nn_writel(nn, NFP_NET_CFG_CTRL, 0);
3796 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
3797 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
3798 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
3799 				   NFP_NET_CFG_UPDATE_GEN);
3800 	if (err)
3801 		return err;
3802 
3803 	nfp_net_vecs_init(nn);
3804 
3805 	if (!nn->dp.netdev)
3806 		return 0;
3807 	return register_netdev(nn->dp.netdev);
3808 }
3809 
3810 /**
3811  * nfp_net_clean() - Undo what nfp_net_init() did.
3812  * @nn:		NFP Net device structure
3813  */
3814 void nfp_net_clean(struct nfp_net *nn)
3815 {
3816 	if (!nn->dp.netdev)
3817 		return;
3818 
3819 	unregister_netdev(nn->dp.netdev);
3820 }
3821