xref: /openbmc/linux/drivers/net/ethernet/netronome/nfp/nfp_net_common.c (revision 28efb0046512e8a13ed9f9bdf0d68d10bbfbe9cf)
1 /*
2  * Copyright (C) 2015-2017 Netronome Systems, Inc.
3  *
4  * This software is dual licensed under the GNU General License Version 2,
5  * June 1991 as shown in the file COPYING in the top-level directory of this
6  * source tree or the BSD 2-Clause License provided below.  You have the
7  * option to license this software under the complete terms of either license.
8  *
9  * The BSD 2-Clause License:
10  *
11  *     Redistribution and use in source and binary forms, with or
12  *     without modification, are permitted provided that the following
13  *     conditions are met:
14  *
15  *      1. Redistributions of source code must retain the above
16  *         copyright notice, this list of conditions and the following
17  *         disclaimer.
18  *
19  *      2. Redistributions in binary form must reproduce the above
20  *         copyright notice, this list of conditions and the following
21  *         disclaimer in the documentation and/or other materials
22  *         provided with the distribution.
23  *
24  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31  * SOFTWARE.
32  */
33 
34 /*
35  * nfp_net_common.c
36  * Netronome network device driver: Common functions between PF and VF
37  * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
38  *          Jason McMullan <jason.mcmullan@netronome.com>
39  *          Rolf Neugebauer <rolf.neugebauer@netronome.com>
40  *          Brad Petrus <brad.petrus@netronome.com>
41  *          Chris Telfer <chris.telfer@netronome.com>
42  */
43 
44 #include <linux/bitfield.h>
45 #include <linux/bpf.h>
46 #include <linux/bpf_trace.h>
47 #include <linux/module.h>
48 #include <linux/kernel.h>
49 #include <linux/init.h>
50 #include <linux/fs.h>
51 #include <linux/netdevice.h>
52 #include <linux/etherdevice.h>
53 #include <linux/interrupt.h>
54 #include <linux/ip.h>
55 #include <linux/ipv6.h>
56 #include <linux/page_ref.h>
57 #include <linux/pci.h>
58 #include <linux/pci_regs.h>
59 #include <linux/msi.h>
60 #include <linux/ethtool.h>
61 #include <linux/log2.h>
62 #include <linux/if_vlan.h>
63 #include <linux/random.h>
64 #include <linux/vmalloc.h>
65 #include <linux/ktime.h>
66 
67 #include <net/switchdev.h>
68 #include <net/vxlan.h>
69 
70 #include "nfpcore/nfp_nsp.h"
71 #include "nfp_app.h"
72 #include "nfp_net_ctrl.h"
73 #include "nfp_net.h"
74 #include "nfp_net_sriov.h"
75 #include "nfp_port.h"
76 
77 /**
78  * nfp_net_get_fw_version() - Read and parse the FW version
79  * @fw_ver:	Output fw_version structure to read to
80  * @ctrl_bar:	Mapped address of the control BAR
81  */
82 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
83 			    void __iomem *ctrl_bar)
84 {
85 	u32 reg;
86 
87 	reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
88 	put_unaligned_le32(reg, fw_ver);
89 }
90 
91 static dma_addr_t nfp_net_dma_map_rx(struct nfp_net_dp *dp, void *frag)
92 {
93 	return dma_map_single_attrs(dp->dev, frag + NFP_NET_RX_BUF_HEADROOM,
94 				    dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
95 				    dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
96 }
97 
98 static void
99 nfp_net_dma_sync_dev_rx(const struct nfp_net_dp *dp, dma_addr_t dma_addr)
100 {
101 	dma_sync_single_for_device(dp->dev, dma_addr,
102 				   dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
103 				   dp->rx_dma_dir);
104 }
105 
106 static void nfp_net_dma_unmap_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr)
107 {
108 	dma_unmap_single_attrs(dp->dev, dma_addr,
109 			       dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
110 			       dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
111 }
112 
113 static void nfp_net_dma_sync_cpu_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr,
114 				    unsigned int len)
115 {
116 	dma_sync_single_for_cpu(dp->dev, dma_addr - NFP_NET_RX_BUF_HEADROOM,
117 				len, dp->rx_dma_dir);
118 }
119 
120 /* Firmware reconfig
121  *
122  * Firmware reconfig may take a while so we have two versions of it -
123  * synchronous and asynchronous (posted).  All synchronous callers are holding
124  * RTNL so we don't have to worry about serializing them.
125  */
126 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
127 {
128 	nn_writel(nn, NFP_NET_CFG_UPDATE, update);
129 	/* ensure update is written before pinging HW */
130 	nn_pci_flush(nn);
131 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
132 }
133 
134 /* Pass 0 as update to run posted reconfigs. */
135 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
136 {
137 	update |= nn->reconfig_posted;
138 	nn->reconfig_posted = 0;
139 
140 	nfp_net_reconfig_start(nn, update);
141 
142 	nn->reconfig_timer_active = true;
143 	mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
144 }
145 
146 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
147 {
148 	u32 reg;
149 
150 	reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
151 	if (reg == 0)
152 		return true;
153 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
154 		nn_err(nn, "Reconfig error: 0x%08x\n", reg);
155 		return true;
156 	} else if (last_check) {
157 		nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
158 		return true;
159 	}
160 
161 	return false;
162 }
163 
164 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
165 {
166 	bool timed_out = false;
167 
168 	/* Poll update field, waiting for NFP to ack the config */
169 	while (!nfp_net_reconfig_check_done(nn, timed_out)) {
170 		msleep(1);
171 		timed_out = time_is_before_eq_jiffies(deadline);
172 	}
173 
174 	if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
175 		return -EIO;
176 
177 	return timed_out ? -EIO : 0;
178 }
179 
180 static void nfp_net_reconfig_timer(unsigned long data)
181 {
182 	struct nfp_net *nn = (void *)data;
183 
184 	spin_lock_bh(&nn->reconfig_lock);
185 
186 	nn->reconfig_timer_active = false;
187 
188 	/* If sync caller is present it will take over from us */
189 	if (nn->reconfig_sync_present)
190 		goto done;
191 
192 	/* Read reconfig status and report errors */
193 	nfp_net_reconfig_check_done(nn, true);
194 
195 	if (nn->reconfig_posted)
196 		nfp_net_reconfig_start_async(nn, 0);
197 done:
198 	spin_unlock_bh(&nn->reconfig_lock);
199 }
200 
201 /**
202  * nfp_net_reconfig_post() - Post async reconfig request
203  * @nn:      NFP Net device to reconfigure
204  * @update:  The value for the update field in the BAR config
205  *
206  * Record FW reconfiguration request.  Reconfiguration will be kicked off
207  * whenever reconfiguration machinery is idle.  Multiple requests can be
208  * merged together!
209  */
210 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
211 {
212 	spin_lock_bh(&nn->reconfig_lock);
213 
214 	/* Sync caller will kick off async reconf when it's done, just post */
215 	if (nn->reconfig_sync_present) {
216 		nn->reconfig_posted |= update;
217 		goto done;
218 	}
219 
220 	/* Opportunistically check if the previous command is done */
221 	if (!nn->reconfig_timer_active ||
222 	    nfp_net_reconfig_check_done(nn, false))
223 		nfp_net_reconfig_start_async(nn, update);
224 	else
225 		nn->reconfig_posted |= update;
226 done:
227 	spin_unlock_bh(&nn->reconfig_lock);
228 }
229 
230 /**
231  * nfp_net_reconfig() - Reconfigure the firmware
232  * @nn:      NFP Net device to reconfigure
233  * @update:  The value for the update field in the BAR config
234  *
235  * Write the update word to the BAR and ping the reconfig queue.  The
236  * poll until the firmware has acknowledged the update by zeroing the
237  * update word.
238  *
239  * Return: Negative errno on error, 0 on success
240  */
241 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
242 {
243 	bool cancelled_timer = false;
244 	u32 pre_posted_requests;
245 	int ret;
246 
247 	spin_lock_bh(&nn->reconfig_lock);
248 
249 	nn->reconfig_sync_present = true;
250 
251 	if (nn->reconfig_timer_active) {
252 		del_timer(&nn->reconfig_timer);
253 		nn->reconfig_timer_active = false;
254 		cancelled_timer = true;
255 	}
256 	pre_posted_requests = nn->reconfig_posted;
257 	nn->reconfig_posted = 0;
258 
259 	spin_unlock_bh(&nn->reconfig_lock);
260 
261 	if (cancelled_timer)
262 		nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
263 
264 	/* Run the posted reconfigs which were issued before we started */
265 	if (pre_posted_requests) {
266 		nfp_net_reconfig_start(nn, pre_posted_requests);
267 		nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
268 	}
269 
270 	nfp_net_reconfig_start(nn, update);
271 	ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
272 
273 	spin_lock_bh(&nn->reconfig_lock);
274 
275 	if (nn->reconfig_posted)
276 		nfp_net_reconfig_start_async(nn, 0);
277 
278 	nn->reconfig_sync_present = false;
279 
280 	spin_unlock_bh(&nn->reconfig_lock);
281 
282 	return ret;
283 }
284 
285 /**
286  * nfp_net_reconfig_mbox() - Reconfigure the firmware via the mailbox
287  * @nn:        NFP Net device to reconfigure
288  * @mbox_cmd:  The value for the mailbox command
289  *
290  * Helper function for mailbox updates
291  *
292  * Return: Negative errno on error, 0 on success
293  */
294 static int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd)
295 {
296 	int ret;
297 
298 	nn_writeq(nn, NFP_NET_CFG_MBOX_CMD, mbox_cmd);
299 
300 	ret = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
301 	if (ret) {
302 		nn_err(nn, "Mailbox update error\n");
303 		return ret;
304 	}
305 
306 	return -nn_readl(nn, NFP_NET_CFG_MBOX_RET);
307 }
308 
309 /* Interrupt configuration and handling
310  */
311 
312 /**
313  * nfp_net_irq_unmask() - Unmask automasked interrupt
314  * @nn:       NFP Network structure
315  * @entry_nr: MSI-X table entry
316  *
317  * Clear the ICR for the IRQ entry.
318  */
319 static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
320 {
321 	nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
322 	nn_pci_flush(nn);
323 }
324 
325 /**
326  * nfp_net_irqs_alloc() - allocates MSI-X irqs
327  * @pdev:        PCI device structure
328  * @irq_entries: Array to be initialized and used to hold the irq entries
329  * @min_irqs:    Minimal acceptable number of interrupts
330  * @wanted_irqs: Target number of interrupts to allocate
331  *
332  * Return: Number of irqs obtained or 0 on error.
333  */
334 unsigned int
335 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
336 		   unsigned int min_irqs, unsigned int wanted_irqs)
337 {
338 	unsigned int i;
339 	int got_irqs;
340 
341 	for (i = 0; i < wanted_irqs; i++)
342 		irq_entries[i].entry = i;
343 
344 	got_irqs = pci_enable_msix_range(pdev, irq_entries,
345 					 min_irqs, wanted_irqs);
346 	if (got_irqs < 0) {
347 		dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
348 			min_irqs, wanted_irqs, got_irqs);
349 		return 0;
350 	}
351 
352 	if (got_irqs < wanted_irqs)
353 		dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
354 			 wanted_irqs, got_irqs);
355 
356 	return got_irqs;
357 }
358 
359 /**
360  * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
361  * @nn:		 NFP Network structure
362  * @irq_entries: Table of allocated interrupts
363  * @n:		 Size of @irq_entries (number of entries to grab)
364  *
365  * After interrupts are allocated with nfp_net_irqs_alloc() this function
366  * should be called to assign them to a specific netdev (port).
367  */
368 void
369 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
370 		    unsigned int n)
371 {
372 	struct nfp_net_dp *dp = &nn->dp;
373 
374 	nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
375 	dp->num_r_vecs = nn->max_r_vecs;
376 
377 	memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
378 
379 	if (dp->num_rx_rings > dp->num_r_vecs ||
380 	    dp->num_tx_rings > dp->num_r_vecs)
381 		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
382 			 dp->num_rx_rings, dp->num_tx_rings,
383 			 dp->num_r_vecs);
384 
385 	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
386 	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
387 	dp->num_stack_tx_rings = dp->num_tx_rings;
388 }
389 
390 /**
391  * nfp_net_irqs_disable() - Disable interrupts
392  * @pdev:        PCI device structure
393  *
394  * Undoes what @nfp_net_irqs_alloc() does.
395  */
396 void nfp_net_irqs_disable(struct pci_dev *pdev)
397 {
398 	pci_disable_msix(pdev);
399 }
400 
401 /**
402  * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
403  * @irq:      Interrupt
404  * @data:     Opaque data structure
405  *
406  * Return: Indicate if the interrupt has been handled.
407  */
408 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
409 {
410 	struct nfp_net_r_vector *r_vec = data;
411 
412 	napi_schedule_irqoff(&r_vec->napi);
413 
414 	/* The FW auto-masks any interrupt, either via the MASK bit in
415 	 * the MSI-X table or via the per entry ICR field.  So there
416 	 * is no need to disable interrupts here.
417 	 */
418 	return IRQ_HANDLED;
419 }
420 
421 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
422 {
423 	struct nfp_net_r_vector *r_vec = data;
424 
425 	tasklet_schedule(&r_vec->tasklet);
426 
427 	return IRQ_HANDLED;
428 }
429 
430 /**
431  * nfp_net_read_link_status() - Reread link status from control BAR
432  * @nn:       NFP Network structure
433  */
434 static void nfp_net_read_link_status(struct nfp_net *nn)
435 {
436 	unsigned long flags;
437 	bool link_up;
438 	u32 sts;
439 
440 	spin_lock_irqsave(&nn->link_status_lock, flags);
441 
442 	sts = nn_readl(nn, NFP_NET_CFG_STS);
443 	link_up = !!(sts & NFP_NET_CFG_STS_LINK);
444 
445 	if (nn->link_up == link_up)
446 		goto out;
447 
448 	nn->link_up = link_up;
449 	if (nn->port)
450 		set_bit(NFP_PORT_CHANGED, &nn->port->flags);
451 
452 	if (nn->link_up) {
453 		netif_carrier_on(nn->dp.netdev);
454 		netdev_info(nn->dp.netdev, "NIC Link is Up\n");
455 	} else {
456 		netif_carrier_off(nn->dp.netdev);
457 		netdev_info(nn->dp.netdev, "NIC Link is Down\n");
458 	}
459 out:
460 	spin_unlock_irqrestore(&nn->link_status_lock, flags);
461 }
462 
463 /**
464  * nfp_net_irq_lsc() - Interrupt service routine for link state changes
465  * @irq:      Interrupt
466  * @data:     Opaque data structure
467  *
468  * Return: Indicate if the interrupt has been handled.
469  */
470 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
471 {
472 	struct nfp_net *nn = data;
473 	struct msix_entry *entry;
474 
475 	entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
476 
477 	nfp_net_read_link_status(nn);
478 
479 	nfp_net_irq_unmask(nn, entry->entry);
480 
481 	return IRQ_HANDLED;
482 }
483 
484 /**
485  * nfp_net_irq_exn() - Interrupt service routine for exceptions
486  * @irq:      Interrupt
487  * @data:     Opaque data structure
488  *
489  * Return: Indicate if the interrupt has been handled.
490  */
491 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
492 {
493 	struct nfp_net *nn = data;
494 
495 	nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
496 	/* XXX TO BE IMPLEMENTED */
497 	return IRQ_HANDLED;
498 }
499 
500 /**
501  * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
502  * @tx_ring:  TX ring structure
503  * @r_vec:    IRQ vector servicing this ring
504  * @idx:      Ring index
505  * @is_xdp:   Is this an XDP TX ring?
506  */
507 static void
508 nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
509 		     struct nfp_net_r_vector *r_vec, unsigned int idx,
510 		     bool is_xdp)
511 {
512 	struct nfp_net *nn = r_vec->nfp_net;
513 
514 	tx_ring->idx = idx;
515 	tx_ring->r_vec = r_vec;
516 	tx_ring->is_xdp = is_xdp;
517 	u64_stats_init(&tx_ring->r_vec->tx_sync);
518 
519 	tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
520 	tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
521 }
522 
523 /**
524  * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
525  * @rx_ring:  RX ring structure
526  * @r_vec:    IRQ vector servicing this ring
527  * @idx:      Ring index
528  */
529 static void
530 nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
531 		     struct nfp_net_r_vector *r_vec, unsigned int idx)
532 {
533 	struct nfp_net *nn = r_vec->nfp_net;
534 
535 	rx_ring->idx = idx;
536 	rx_ring->r_vec = r_vec;
537 	u64_stats_init(&rx_ring->r_vec->rx_sync);
538 
539 	rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
540 	rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
541 }
542 
543 /**
544  * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
545  * @nn:		NFP Network structure
546  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
547  * @format:	printf-style format to construct the interrupt name
548  * @name:	Pointer to allocated space for interrupt name
549  * @name_sz:	Size of space for interrupt name
550  * @vector_idx:	Index of MSI-X vector used for this interrupt
551  * @handler:	IRQ handler to register for this interrupt
552  */
553 static int
554 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
555 			const char *format, char *name, size_t name_sz,
556 			unsigned int vector_idx, irq_handler_t handler)
557 {
558 	struct msix_entry *entry;
559 	int err;
560 
561 	entry = &nn->irq_entries[vector_idx];
562 
563 	snprintf(name, name_sz, format, nfp_net_name(nn));
564 	err = request_irq(entry->vector, handler, 0, name, nn);
565 	if (err) {
566 		nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
567 		       entry->vector, err);
568 		return err;
569 	}
570 	nn_writeb(nn, ctrl_offset, entry->entry);
571 
572 	return 0;
573 }
574 
575 /**
576  * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
577  * @nn:		NFP Network structure
578  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
579  * @vector_idx:	Index of MSI-X vector used for this interrupt
580  */
581 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
582 				 unsigned int vector_idx)
583 {
584 	nn_writeb(nn, ctrl_offset, 0xff);
585 	free_irq(nn->irq_entries[vector_idx].vector, nn);
586 }
587 
588 /* Transmit
589  *
590  * One queue controller peripheral queue is used for transmit.  The
591  * driver en-queues packets for transmit by advancing the write
592  * pointer.  The device indicates that packets have transmitted by
593  * advancing the read pointer.  The driver maintains a local copy of
594  * the read and write pointer in @struct nfp_net_tx_ring.  The driver
595  * keeps @wr_p in sync with the queue controller write pointer and can
596  * determine how many packets have been transmitted by comparing its
597  * copy of the read pointer @rd_p with the read pointer maintained by
598  * the queue controller peripheral.
599  */
600 
601 /**
602  * nfp_net_tx_full() - Check if the TX ring is full
603  * @tx_ring: TX ring to check
604  * @dcnt:    Number of descriptors that need to be enqueued (must be >= 1)
605  *
606  * This function checks, based on the *host copy* of read/write
607  * pointer if a given TX ring is full.  The real TX queue may have
608  * some newly made available slots.
609  *
610  * Return: True if the ring is full.
611  */
612 static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
613 {
614 	return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
615 }
616 
617 /* Wrappers for deciding when to stop and restart TX queues */
618 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
619 {
620 	return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
621 }
622 
623 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
624 {
625 	return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
626 }
627 
628 /**
629  * nfp_net_tx_ring_stop() - stop tx ring
630  * @nd_q:    netdev queue
631  * @tx_ring: driver tx queue structure
632  *
633  * Safely stop TX ring.  Remember that while we are running .start_xmit()
634  * someone else may be cleaning the TX ring completions so we need to be
635  * extra careful here.
636  */
637 static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
638 				 struct nfp_net_tx_ring *tx_ring)
639 {
640 	netif_tx_stop_queue(nd_q);
641 
642 	/* We can race with the TX completion out of NAPI so recheck */
643 	smp_mb();
644 	if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
645 		netif_tx_start_queue(nd_q);
646 }
647 
648 /**
649  * nfp_net_tx_tso() - Set up Tx descriptor for LSO
650  * @r_vec: per-ring structure
651  * @txbuf: Pointer to driver soft TX descriptor
652  * @txd: Pointer to HW TX descriptor
653  * @skb: Pointer to SKB
654  *
655  * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
656  * Return error on packet header greater than maximum supported LSO header size.
657  */
658 static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
659 			   struct nfp_net_tx_buf *txbuf,
660 			   struct nfp_net_tx_desc *txd, struct sk_buff *skb)
661 {
662 	u32 hdrlen;
663 	u16 mss;
664 
665 	if (!skb_is_gso(skb))
666 		return;
667 
668 	if (!skb->encapsulation) {
669 		txd->l3_offset = skb_network_offset(skb);
670 		txd->l4_offset = skb_transport_offset(skb);
671 		hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
672 	} else {
673 		txd->l3_offset = skb_inner_network_offset(skb);
674 		txd->l4_offset = skb_inner_transport_offset(skb);
675 		hdrlen = skb_inner_transport_header(skb) - skb->data +
676 			inner_tcp_hdrlen(skb);
677 	}
678 
679 	txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
680 	txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
681 
682 	mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
683 	txd->lso_hdrlen = hdrlen;
684 	txd->mss = cpu_to_le16(mss);
685 	txd->flags |= PCIE_DESC_TX_LSO;
686 
687 	u64_stats_update_begin(&r_vec->tx_sync);
688 	r_vec->tx_lso++;
689 	u64_stats_update_end(&r_vec->tx_sync);
690 }
691 
692 /**
693  * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
694  * @dp:  NFP Net data path struct
695  * @r_vec: per-ring structure
696  * @txbuf: Pointer to driver soft TX descriptor
697  * @txd: Pointer to TX descriptor
698  * @skb: Pointer to SKB
699  *
700  * This function sets the TX checksum flags in the TX descriptor based
701  * on the configuration and the protocol of the packet to be transmitted.
702  */
703 static void nfp_net_tx_csum(struct nfp_net_dp *dp,
704 			    struct nfp_net_r_vector *r_vec,
705 			    struct nfp_net_tx_buf *txbuf,
706 			    struct nfp_net_tx_desc *txd, struct sk_buff *skb)
707 {
708 	struct ipv6hdr *ipv6h;
709 	struct iphdr *iph;
710 	u8 l4_hdr;
711 
712 	if (!(dp->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
713 		return;
714 
715 	if (skb->ip_summed != CHECKSUM_PARTIAL)
716 		return;
717 
718 	txd->flags |= PCIE_DESC_TX_CSUM;
719 	if (skb->encapsulation)
720 		txd->flags |= PCIE_DESC_TX_ENCAP;
721 
722 	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
723 	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
724 
725 	if (iph->version == 4) {
726 		txd->flags |= PCIE_DESC_TX_IP4_CSUM;
727 		l4_hdr = iph->protocol;
728 	} else if (ipv6h->version == 6) {
729 		l4_hdr = ipv6h->nexthdr;
730 	} else {
731 		nn_dp_warn(dp, "partial checksum but ipv=%x!\n", iph->version);
732 		return;
733 	}
734 
735 	switch (l4_hdr) {
736 	case IPPROTO_TCP:
737 		txd->flags |= PCIE_DESC_TX_TCP_CSUM;
738 		break;
739 	case IPPROTO_UDP:
740 		txd->flags |= PCIE_DESC_TX_UDP_CSUM;
741 		break;
742 	default:
743 		nn_dp_warn(dp, "partial checksum but l4 proto=%x!\n", l4_hdr);
744 		return;
745 	}
746 
747 	u64_stats_update_begin(&r_vec->tx_sync);
748 	if (skb->encapsulation)
749 		r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
750 	else
751 		r_vec->hw_csum_tx += txbuf->pkt_cnt;
752 	u64_stats_update_end(&r_vec->tx_sync);
753 }
754 
755 static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
756 {
757 	wmb();
758 	nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
759 	tx_ring->wr_ptr_add = 0;
760 }
761 
762 static int nfp_net_prep_port_id(struct sk_buff *skb)
763 {
764 	struct metadata_dst *md_dst = skb_metadata_dst(skb);
765 	unsigned char *data;
766 
767 	if (likely(!md_dst))
768 		return 0;
769 	if (unlikely(md_dst->type != METADATA_HW_PORT_MUX))
770 		return 0;
771 
772 	if (unlikely(skb_cow_head(skb, 8)))
773 		return -ENOMEM;
774 
775 	data = skb_push(skb, 8);
776 	put_unaligned_be32(NFP_NET_META_PORTID, data);
777 	put_unaligned_be32(md_dst->u.port_info.port_id, data + 4);
778 
779 	return 8;
780 }
781 
782 /**
783  * nfp_net_tx() - Main transmit entry point
784  * @skb:    SKB to transmit
785  * @netdev: netdev structure
786  *
787  * Return: NETDEV_TX_OK on success.
788  */
789 static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
790 {
791 	struct nfp_net *nn = netdev_priv(netdev);
792 	const struct skb_frag_struct *frag;
793 	struct nfp_net_tx_desc *txd, txdg;
794 	int f, nr_frags, wr_idx, md_bytes;
795 	struct nfp_net_tx_ring *tx_ring;
796 	struct nfp_net_r_vector *r_vec;
797 	struct nfp_net_tx_buf *txbuf;
798 	struct netdev_queue *nd_q;
799 	struct nfp_net_dp *dp;
800 	dma_addr_t dma_addr;
801 	unsigned int fsize;
802 	u16 qidx;
803 
804 	dp = &nn->dp;
805 	qidx = skb_get_queue_mapping(skb);
806 	tx_ring = &dp->tx_rings[qidx];
807 	r_vec = tx_ring->r_vec;
808 	nd_q = netdev_get_tx_queue(dp->netdev, qidx);
809 
810 	nr_frags = skb_shinfo(skb)->nr_frags;
811 
812 	if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
813 		nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
814 			   qidx, tx_ring->wr_p, tx_ring->rd_p);
815 		netif_tx_stop_queue(nd_q);
816 		nfp_net_tx_xmit_more_flush(tx_ring);
817 		u64_stats_update_begin(&r_vec->tx_sync);
818 		r_vec->tx_busy++;
819 		u64_stats_update_end(&r_vec->tx_sync);
820 		return NETDEV_TX_BUSY;
821 	}
822 
823 	md_bytes = nfp_net_prep_port_id(skb);
824 	if (unlikely(md_bytes < 0)) {
825 		nfp_net_tx_xmit_more_flush(tx_ring);
826 		dev_kfree_skb_any(skb);
827 		return NETDEV_TX_OK;
828 	}
829 
830 	/* Start with the head skbuf */
831 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
832 				  DMA_TO_DEVICE);
833 	if (dma_mapping_error(dp->dev, dma_addr))
834 		goto err_free;
835 
836 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
837 
838 	/* Stash the soft descriptor of the head then initialize it */
839 	txbuf = &tx_ring->txbufs[wr_idx];
840 	txbuf->skb = skb;
841 	txbuf->dma_addr = dma_addr;
842 	txbuf->fidx = -1;
843 	txbuf->pkt_cnt = 1;
844 	txbuf->real_len = skb->len;
845 
846 	/* Build TX descriptor */
847 	txd = &tx_ring->txds[wr_idx];
848 	txd->offset_eop = (nr_frags ? 0 : PCIE_DESC_TX_EOP) | md_bytes;
849 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
850 	nfp_desc_set_dma_addr(txd, dma_addr);
851 	txd->data_len = cpu_to_le16(skb->len);
852 
853 	txd->flags = 0;
854 	txd->mss = 0;
855 	txd->lso_hdrlen = 0;
856 
857 	/* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
858 	nfp_net_tx_tso(r_vec, txbuf, txd, skb);
859 	nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
860 	if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
861 		txd->flags |= PCIE_DESC_TX_VLAN;
862 		txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
863 	}
864 
865 	/* Gather DMA */
866 	if (nr_frags > 0) {
867 		/* all descs must match except for in addr, length and eop */
868 		txdg = *txd;
869 
870 		for (f = 0; f < nr_frags; f++) {
871 			frag = &skb_shinfo(skb)->frags[f];
872 			fsize = skb_frag_size(frag);
873 
874 			dma_addr = skb_frag_dma_map(dp->dev, frag, 0,
875 						    fsize, DMA_TO_DEVICE);
876 			if (dma_mapping_error(dp->dev, dma_addr))
877 				goto err_unmap;
878 
879 			wr_idx = D_IDX(tx_ring, wr_idx + 1);
880 			tx_ring->txbufs[wr_idx].skb = skb;
881 			tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
882 			tx_ring->txbufs[wr_idx].fidx = f;
883 
884 			txd = &tx_ring->txds[wr_idx];
885 			*txd = txdg;
886 			txd->dma_len = cpu_to_le16(fsize);
887 			nfp_desc_set_dma_addr(txd, dma_addr);
888 			txd->offset_eop |=
889 				(f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
890 		}
891 
892 		u64_stats_update_begin(&r_vec->tx_sync);
893 		r_vec->tx_gather++;
894 		u64_stats_update_end(&r_vec->tx_sync);
895 	}
896 
897 	netdev_tx_sent_queue(nd_q, txbuf->real_len);
898 
899 	skb_tx_timestamp(skb);
900 
901 	tx_ring->wr_p += nr_frags + 1;
902 	if (nfp_net_tx_ring_should_stop(tx_ring))
903 		nfp_net_tx_ring_stop(nd_q, tx_ring);
904 
905 	tx_ring->wr_ptr_add += nr_frags + 1;
906 	if (!skb->xmit_more || netif_xmit_stopped(nd_q))
907 		nfp_net_tx_xmit_more_flush(tx_ring);
908 
909 	return NETDEV_TX_OK;
910 
911 err_unmap:
912 	while (--f >= 0) {
913 		frag = &skb_shinfo(skb)->frags[f];
914 		dma_unmap_page(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
915 			       skb_frag_size(frag), DMA_TO_DEVICE);
916 		tx_ring->txbufs[wr_idx].skb = NULL;
917 		tx_ring->txbufs[wr_idx].dma_addr = 0;
918 		tx_ring->txbufs[wr_idx].fidx = -2;
919 		wr_idx = wr_idx - 1;
920 		if (wr_idx < 0)
921 			wr_idx += tx_ring->cnt;
922 	}
923 	dma_unmap_single(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
924 			 skb_headlen(skb), DMA_TO_DEVICE);
925 	tx_ring->txbufs[wr_idx].skb = NULL;
926 	tx_ring->txbufs[wr_idx].dma_addr = 0;
927 	tx_ring->txbufs[wr_idx].fidx = -2;
928 err_free:
929 	nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
930 	nfp_net_tx_xmit_more_flush(tx_ring);
931 	u64_stats_update_begin(&r_vec->tx_sync);
932 	r_vec->tx_errors++;
933 	u64_stats_update_end(&r_vec->tx_sync);
934 	dev_kfree_skb_any(skb);
935 	return NETDEV_TX_OK;
936 }
937 
938 /**
939  * nfp_net_tx_complete() - Handled completed TX packets
940  * @tx_ring:   TX ring structure
941  *
942  * Return: Number of completed TX descriptors
943  */
944 static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring)
945 {
946 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
947 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
948 	const struct skb_frag_struct *frag;
949 	struct netdev_queue *nd_q;
950 	u32 done_pkts = 0, done_bytes = 0;
951 	struct sk_buff *skb;
952 	int todo, nr_frags;
953 	u32 qcp_rd_p;
954 	int fidx;
955 	int idx;
956 
957 	if (tx_ring->wr_p == tx_ring->rd_p)
958 		return;
959 
960 	/* Work out how many descriptors have been transmitted */
961 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
962 
963 	if (qcp_rd_p == tx_ring->qcp_rd_p)
964 		return;
965 
966 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
967 
968 	while (todo--) {
969 		idx = D_IDX(tx_ring, tx_ring->rd_p++);
970 
971 		skb = tx_ring->txbufs[idx].skb;
972 		if (!skb)
973 			continue;
974 
975 		nr_frags = skb_shinfo(skb)->nr_frags;
976 		fidx = tx_ring->txbufs[idx].fidx;
977 
978 		if (fidx == -1) {
979 			/* unmap head */
980 			dma_unmap_single(dp->dev, tx_ring->txbufs[idx].dma_addr,
981 					 skb_headlen(skb), DMA_TO_DEVICE);
982 
983 			done_pkts += tx_ring->txbufs[idx].pkt_cnt;
984 			done_bytes += tx_ring->txbufs[idx].real_len;
985 		} else {
986 			/* unmap fragment */
987 			frag = &skb_shinfo(skb)->frags[fidx];
988 			dma_unmap_page(dp->dev, tx_ring->txbufs[idx].dma_addr,
989 				       skb_frag_size(frag), DMA_TO_DEVICE);
990 		}
991 
992 		/* check for last gather fragment */
993 		if (fidx == nr_frags - 1)
994 			dev_consume_skb_any(skb);
995 
996 		tx_ring->txbufs[idx].dma_addr = 0;
997 		tx_ring->txbufs[idx].skb = NULL;
998 		tx_ring->txbufs[idx].fidx = -2;
999 	}
1000 
1001 	tx_ring->qcp_rd_p = qcp_rd_p;
1002 
1003 	u64_stats_update_begin(&r_vec->tx_sync);
1004 	r_vec->tx_bytes += done_bytes;
1005 	r_vec->tx_pkts += done_pkts;
1006 	u64_stats_update_end(&r_vec->tx_sync);
1007 
1008 	if (!dp->netdev)
1009 		return;
1010 
1011 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1012 	netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
1013 	if (nfp_net_tx_ring_should_wake(tx_ring)) {
1014 		/* Make sure TX thread will see updated tx_ring->rd_p */
1015 		smp_mb();
1016 
1017 		if (unlikely(netif_tx_queue_stopped(nd_q)))
1018 			netif_tx_wake_queue(nd_q);
1019 	}
1020 
1021 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1022 		  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1023 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1024 }
1025 
1026 static bool nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
1027 {
1028 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1029 	u32 done_pkts = 0, done_bytes = 0;
1030 	bool done_all;
1031 	int idx, todo;
1032 	u32 qcp_rd_p;
1033 
1034 	/* Work out how many descriptors have been transmitted */
1035 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1036 
1037 	if (qcp_rd_p == tx_ring->qcp_rd_p)
1038 		return true;
1039 
1040 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1041 
1042 	done_all = todo <= NFP_NET_XDP_MAX_COMPLETE;
1043 	todo = min(todo, NFP_NET_XDP_MAX_COMPLETE);
1044 
1045 	tx_ring->qcp_rd_p = D_IDX(tx_ring, tx_ring->qcp_rd_p + todo);
1046 
1047 	done_pkts = todo;
1048 	while (todo--) {
1049 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1050 		tx_ring->rd_p++;
1051 
1052 		done_bytes += tx_ring->txbufs[idx].real_len;
1053 	}
1054 
1055 	u64_stats_update_begin(&r_vec->tx_sync);
1056 	r_vec->tx_bytes += done_bytes;
1057 	r_vec->tx_pkts += done_pkts;
1058 	u64_stats_update_end(&r_vec->tx_sync);
1059 
1060 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1061 		  "XDP TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1062 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1063 
1064 	return done_all;
1065 }
1066 
1067 /**
1068  * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
1069  * @dp:		NFP Net data path struct
1070  * @tx_ring:	TX ring structure
1071  *
1072  * Assumes that the device is stopped
1073  */
1074 static void
1075 nfp_net_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
1076 {
1077 	const struct skb_frag_struct *frag;
1078 	struct netdev_queue *nd_q;
1079 
1080 	while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
1081 		struct nfp_net_tx_buf *tx_buf;
1082 		struct sk_buff *skb;
1083 		int idx, nr_frags;
1084 
1085 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1086 		tx_buf = &tx_ring->txbufs[idx];
1087 
1088 		skb = tx_ring->txbufs[idx].skb;
1089 		nr_frags = skb_shinfo(skb)->nr_frags;
1090 
1091 		if (tx_buf->fidx == -1) {
1092 			/* unmap head */
1093 			dma_unmap_single(dp->dev, tx_buf->dma_addr,
1094 					 skb_headlen(skb), DMA_TO_DEVICE);
1095 		} else {
1096 			/* unmap fragment */
1097 			frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1098 			dma_unmap_page(dp->dev, tx_buf->dma_addr,
1099 				       skb_frag_size(frag), DMA_TO_DEVICE);
1100 		}
1101 
1102 		/* check for last gather fragment */
1103 		if (tx_buf->fidx == nr_frags - 1)
1104 			dev_kfree_skb_any(skb);
1105 
1106 		tx_buf->dma_addr = 0;
1107 		tx_buf->skb = NULL;
1108 		tx_buf->fidx = -2;
1109 
1110 		tx_ring->qcp_rd_p++;
1111 		tx_ring->rd_p++;
1112 	}
1113 
1114 	memset(tx_ring->txds, 0, sizeof(*tx_ring->txds) * tx_ring->cnt);
1115 	tx_ring->wr_p = 0;
1116 	tx_ring->rd_p = 0;
1117 	tx_ring->qcp_rd_p = 0;
1118 	tx_ring->wr_ptr_add = 0;
1119 
1120 	if (tx_ring->is_xdp || !dp->netdev)
1121 		return;
1122 
1123 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1124 	netdev_tx_reset_queue(nd_q);
1125 }
1126 
1127 static void nfp_net_tx_timeout(struct net_device *netdev)
1128 {
1129 	struct nfp_net *nn = netdev_priv(netdev);
1130 	int i;
1131 
1132 	for (i = 0; i < nn->dp.netdev->real_num_tx_queues; i++) {
1133 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i)))
1134 			continue;
1135 		nn_warn(nn, "TX timeout on ring: %d\n", i);
1136 	}
1137 	nn_warn(nn, "TX watchdog timeout\n");
1138 }
1139 
1140 /* Receive processing
1141  */
1142 static unsigned int
1143 nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
1144 {
1145 	unsigned int fl_bufsz;
1146 
1147 	fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
1148 	fl_bufsz += dp->rx_dma_off;
1149 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1150 		fl_bufsz += NFP_NET_MAX_PREPEND;
1151 	else
1152 		fl_bufsz += dp->rx_offset;
1153 	fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
1154 
1155 	fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1156 	fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1157 
1158 	return fl_bufsz;
1159 }
1160 
1161 static void
1162 nfp_net_free_frag(void *frag, bool xdp)
1163 {
1164 	if (!xdp)
1165 		skb_free_frag(frag);
1166 	else
1167 		__free_page(virt_to_page(frag));
1168 }
1169 
1170 /**
1171  * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
1172  * @dp:		NFP Net data path struct
1173  * @dma_addr:	Pointer to storage for DMA address (output param)
1174  *
1175  * This function will allcate a new page frag, map it for DMA.
1176  *
1177  * Return: allocated page frag or NULL on failure.
1178  */
1179 static void *nfp_net_rx_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1180 {
1181 	void *frag;
1182 
1183 	if (!dp->xdp_prog)
1184 		frag = netdev_alloc_frag(dp->fl_bufsz);
1185 	else
1186 		frag = page_address(alloc_page(GFP_KERNEL | __GFP_COLD));
1187 	if (!frag) {
1188 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1189 		return NULL;
1190 	}
1191 
1192 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1193 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1194 		nfp_net_free_frag(frag, dp->xdp_prog);
1195 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1196 		return NULL;
1197 	}
1198 
1199 	return frag;
1200 }
1201 
1202 static void *nfp_net_napi_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1203 {
1204 	void *frag;
1205 
1206 	if (!dp->xdp_prog)
1207 		frag = napi_alloc_frag(dp->fl_bufsz);
1208 	else
1209 		frag = page_address(alloc_page(GFP_ATOMIC | __GFP_COLD));
1210 	if (!frag) {
1211 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1212 		return NULL;
1213 	}
1214 
1215 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1216 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1217 		nfp_net_free_frag(frag, dp->xdp_prog);
1218 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1219 		return NULL;
1220 	}
1221 
1222 	return frag;
1223 }
1224 
1225 /**
1226  * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1227  * @dp:		NFP Net data path struct
1228  * @rx_ring:	RX ring structure
1229  * @frag:	page fragment buffer
1230  * @dma_addr:	DMA address of skb mapping
1231  */
1232 static void nfp_net_rx_give_one(const struct nfp_net_dp *dp,
1233 				struct nfp_net_rx_ring *rx_ring,
1234 				void *frag, dma_addr_t dma_addr)
1235 {
1236 	unsigned int wr_idx;
1237 
1238 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1239 
1240 	nfp_net_dma_sync_dev_rx(dp, dma_addr);
1241 
1242 	/* Stash SKB and DMA address away */
1243 	rx_ring->rxbufs[wr_idx].frag = frag;
1244 	rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1245 
1246 	/* Fill freelist descriptor */
1247 	rx_ring->rxds[wr_idx].fld.reserved = 0;
1248 	rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1249 	nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld,
1250 			      dma_addr + dp->rx_dma_off);
1251 
1252 	rx_ring->wr_p++;
1253 	if (!(rx_ring->wr_p % NFP_NET_FL_BATCH)) {
1254 		/* Update write pointer of the freelist queue. Make
1255 		 * sure all writes are flushed before telling the hardware.
1256 		 */
1257 		wmb();
1258 		nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, NFP_NET_FL_BATCH);
1259 	}
1260 }
1261 
1262 /**
1263  * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1264  * @rx_ring:	RX ring structure
1265  *
1266  * Warning: Do *not* call if ring buffers were never put on the FW freelist
1267  *	    (i.e. device was not enabled)!
1268  */
1269 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1270 {
1271 	unsigned int wr_idx, last_idx;
1272 
1273 	/* Move the empty entry to the end of the list */
1274 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1275 	last_idx = rx_ring->cnt - 1;
1276 	rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1277 	rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1278 	rx_ring->rxbufs[last_idx].dma_addr = 0;
1279 	rx_ring->rxbufs[last_idx].frag = NULL;
1280 
1281 	memset(rx_ring->rxds, 0, sizeof(*rx_ring->rxds) * rx_ring->cnt);
1282 	rx_ring->wr_p = 0;
1283 	rx_ring->rd_p = 0;
1284 }
1285 
1286 /**
1287  * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1288  * @dp:		NFP Net data path struct
1289  * @rx_ring:	RX ring to remove buffers from
1290  *
1291  * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1292  * entries.  After device is disabled nfp_net_rx_ring_reset() must be called
1293  * to restore required ring geometry.
1294  */
1295 static void
1296 nfp_net_rx_ring_bufs_free(struct nfp_net_dp *dp,
1297 			  struct nfp_net_rx_ring *rx_ring)
1298 {
1299 	unsigned int i;
1300 
1301 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1302 		/* NULL skb can only happen when initial filling of the ring
1303 		 * fails to allocate enough buffers and calls here to free
1304 		 * already allocated ones.
1305 		 */
1306 		if (!rx_ring->rxbufs[i].frag)
1307 			continue;
1308 
1309 		nfp_net_dma_unmap_rx(dp, rx_ring->rxbufs[i].dma_addr);
1310 		nfp_net_free_frag(rx_ring->rxbufs[i].frag, dp->xdp_prog);
1311 		rx_ring->rxbufs[i].dma_addr = 0;
1312 		rx_ring->rxbufs[i].frag = NULL;
1313 	}
1314 }
1315 
1316 /**
1317  * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1318  * @dp:		NFP Net data path struct
1319  * @rx_ring:	RX ring to remove buffers from
1320  */
1321 static int
1322 nfp_net_rx_ring_bufs_alloc(struct nfp_net_dp *dp,
1323 			   struct nfp_net_rx_ring *rx_ring)
1324 {
1325 	struct nfp_net_rx_buf *rxbufs;
1326 	unsigned int i;
1327 
1328 	rxbufs = rx_ring->rxbufs;
1329 
1330 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1331 		rxbufs[i].frag = nfp_net_rx_alloc_one(dp, &rxbufs[i].dma_addr);
1332 		if (!rxbufs[i].frag) {
1333 			nfp_net_rx_ring_bufs_free(dp, rx_ring);
1334 			return -ENOMEM;
1335 		}
1336 	}
1337 
1338 	return 0;
1339 }
1340 
1341 /**
1342  * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1343  * @dp:	     NFP Net data path struct
1344  * @rx_ring: RX ring to fill
1345  */
1346 static void
1347 nfp_net_rx_ring_fill_freelist(struct nfp_net_dp *dp,
1348 			      struct nfp_net_rx_ring *rx_ring)
1349 {
1350 	unsigned int i;
1351 
1352 	for (i = 0; i < rx_ring->cnt - 1; i++)
1353 		nfp_net_rx_give_one(dp, rx_ring, rx_ring->rxbufs[i].frag,
1354 				    rx_ring->rxbufs[i].dma_addr);
1355 }
1356 
1357 /**
1358  * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1359  * @flags: RX descriptor flags field in CPU byte order
1360  */
1361 static int nfp_net_rx_csum_has_errors(u16 flags)
1362 {
1363 	u16 csum_all_checked, csum_all_ok;
1364 
1365 	csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1366 	csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1367 
1368 	return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1369 }
1370 
1371 /**
1372  * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1373  * @dp:  NFP Net data path struct
1374  * @r_vec: per-ring structure
1375  * @rxd: Pointer to RX descriptor
1376  * @meta: Parsed metadata prepend
1377  * @skb: Pointer to SKB
1378  */
1379 static void nfp_net_rx_csum(struct nfp_net_dp *dp,
1380 			    struct nfp_net_r_vector *r_vec,
1381 			    struct nfp_net_rx_desc *rxd,
1382 			    struct nfp_meta_parsed *meta, struct sk_buff *skb)
1383 {
1384 	skb_checksum_none_assert(skb);
1385 
1386 	if (!(dp->netdev->features & NETIF_F_RXCSUM))
1387 		return;
1388 
1389 	if (meta->csum_type) {
1390 		skb->ip_summed = meta->csum_type;
1391 		skb->csum = meta->csum;
1392 		u64_stats_update_begin(&r_vec->rx_sync);
1393 		r_vec->hw_csum_rx_ok++;
1394 		u64_stats_update_end(&r_vec->rx_sync);
1395 		return;
1396 	}
1397 
1398 	if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1399 		u64_stats_update_begin(&r_vec->rx_sync);
1400 		r_vec->hw_csum_rx_error++;
1401 		u64_stats_update_end(&r_vec->rx_sync);
1402 		return;
1403 	}
1404 
1405 	/* Assume that the firmware will never report inner CSUM_OK unless outer
1406 	 * L4 headers were successfully parsed. FW will always report zero UDP
1407 	 * checksum as CSUM_OK.
1408 	 */
1409 	if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1410 	    rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1411 		__skb_incr_checksum_unnecessary(skb);
1412 		u64_stats_update_begin(&r_vec->rx_sync);
1413 		r_vec->hw_csum_rx_ok++;
1414 		u64_stats_update_end(&r_vec->rx_sync);
1415 	}
1416 
1417 	if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1418 	    rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1419 		__skb_incr_checksum_unnecessary(skb);
1420 		u64_stats_update_begin(&r_vec->rx_sync);
1421 		r_vec->hw_csum_rx_inner_ok++;
1422 		u64_stats_update_end(&r_vec->rx_sync);
1423 	}
1424 }
1425 
1426 static void
1427 nfp_net_set_hash(struct net_device *netdev, struct nfp_meta_parsed *meta,
1428 		 unsigned int type, __be32 *hash)
1429 {
1430 	if (!(netdev->features & NETIF_F_RXHASH))
1431 		return;
1432 
1433 	switch (type) {
1434 	case NFP_NET_RSS_IPV4:
1435 	case NFP_NET_RSS_IPV6:
1436 	case NFP_NET_RSS_IPV6_EX:
1437 		meta->hash_type = PKT_HASH_TYPE_L3;
1438 		break;
1439 	default:
1440 		meta->hash_type = PKT_HASH_TYPE_L4;
1441 		break;
1442 	}
1443 
1444 	meta->hash = get_unaligned_be32(hash);
1445 }
1446 
1447 static void
1448 nfp_net_set_hash_desc(struct net_device *netdev, struct nfp_meta_parsed *meta,
1449 		      void *data, struct nfp_net_rx_desc *rxd)
1450 {
1451 	struct nfp_net_rx_hash *rx_hash = data;
1452 
1453 	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1454 		return;
1455 
1456 	nfp_net_set_hash(netdev, meta, get_unaligned_be32(&rx_hash->hash_type),
1457 			 &rx_hash->hash);
1458 }
1459 
1460 static void *
1461 nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
1462 		   void *data, int meta_len)
1463 {
1464 	u32 meta_info;
1465 
1466 	meta_info = get_unaligned_be32(data);
1467 	data += 4;
1468 
1469 	while (meta_info) {
1470 		switch (meta_info & NFP_NET_META_FIELD_MASK) {
1471 		case NFP_NET_META_HASH:
1472 			meta_info >>= NFP_NET_META_FIELD_SIZE;
1473 			nfp_net_set_hash(netdev, meta,
1474 					 meta_info & NFP_NET_META_FIELD_MASK,
1475 					 (__be32 *)data);
1476 			data += 4;
1477 			break;
1478 		case NFP_NET_META_MARK:
1479 			meta->mark = get_unaligned_be32(data);
1480 			data += 4;
1481 			break;
1482 		case NFP_NET_META_PORTID:
1483 			meta->portid = get_unaligned_be32(data);
1484 			data += 4;
1485 			break;
1486 		case NFP_NET_META_CSUM:
1487 			meta->csum_type = CHECKSUM_COMPLETE;
1488 			meta->csum =
1489 				(__force __wsum)__get_unaligned_cpu32(data);
1490 			data += 4;
1491 			break;
1492 		default:
1493 			return NULL;
1494 		}
1495 
1496 		meta_info >>= NFP_NET_META_FIELD_SIZE;
1497 	}
1498 
1499 	return data;
1500 }
1501 
1502 static void
1503 nfp_net_rx_drop(const struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
1504 		struct nfp_net_rx_ring *rx_ring, struct nfp_net_rx_buf *rxbuf,
1505 		struct sk_buff *skb)
1506 {
1507 	u64_stats_update_begin(&r_vec->rx_sync);
1508 	r_vec->rx_drops++;
1509 	u64_stats_update_end(&r_vec->rx_sync);
1510 
1511 	/* skb is build based on the frag, free_skb() would free the frag
1512 	 * so to be able to reuse it we need an extra ref.
1513 	 */
1514 	if (skb && rxbuf && skb->head == rxbuf->frag)
1515 		page_ref_inc(virt_to_head_page(rxbuf->frag));
1516 	if (rxbuf)
1517 		nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag, rxbuf->dma_addr);
1518 	if (skb)
1519 		dev_kfree_skb_any(skb);
1520 }
1521 
1522 static bool
1523 nfp_net_tx_xdp_buf(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring,
1524 		   struct nfp_net_tx_ring *tx_ring,
1525 		   struct nfp_net_rx_buf *rxbuf, unsigned int dma_off,
1526 		   unsigned int pkt_len, bool *completed)
1527 {
1528 	struct nfp_net_tx_buf *txbuf;
1529 	struct nfp_net_tx_desc *txd;
1530 	int wr_idx;
1531 
1532 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1533 		if (!*completed) {
1534 			nfp_net_xdp_complete(tx_ring);
1535 			*completed = true;
1536 		}
1537 
1538 		if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1539 			nfp_net_rx_drop(dp, rx_ring->r_vec, rx_ring, rxbuf,
1540 					NULL);
1541 			return false;
1542 		}
1543 	}
1544 
1545 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1546 
1547 	/* Stash the soft descriptor of the head then initialize it */
1548 	txbuf = &tx_ring->txbufs[wr_idx];
1549 
1550 	nfp_net_rx_give_one(dp, rx_ring, txbuf->frag, txbuf->dma_addr);
1551 
1552 	txbuf->frag = rxbuf->frag;
1553 	txbuf->dma_addr = rxbuf->dma_addr;
1554 	txbuf->fidx = -1;
1555 	txbuf->pkt_cnt = 1;
1556 	txbuf->real_len = pkt_len;
1557 
1558 	dma_sync_single_for_device(dp->dev, rxbuf->dma_addr + dma_off,
1559 				   pkt_len, DMA_BIDIRECTIONAL);
1560 
1561 	/* Build TX descriptor */
1562 	txd = &tx_ring->txds[wr_idx];
1563 	txd->offset_eop = PCIE_DESC_TX_EOP;
1564 	txd->dma_len = cpu_to_le16(pkt_len);
1565 	nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + dma_off);
1566 	txd->data_len = cpu_to_le16(pkt_len);
1567 
1568 	txd->flags = 0;
1569 	txd->mss = 0;
1570 	txd->lso_hdrlen = 0;
1571 
1572 	tx_ring->wr_p++;
1573 	tx_ring->wr_ptr_add++;
1574 	return true;
1575 }
1576 
1577 /**
1578  * nfp_net_rx() - receive up to @budget packets on @rx_ring
1579  * @rx_ring:   RX ring to receive from
1580  * @budget:    NAPI budget
1581  *
1582  * Note, this function is separated out from the napi poll function to
1583  * more cleanly separate packet receive code from other bookkeeping
1584  * functions performed in the napi poll function.
1585  *
1586  * Return: Number of packets received.
1587  */
1588 static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1589 {
1590 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1591 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1592 	struct nfp_net_tx_ring *tx_ring;
1593 	struct bpf_prog *xdp_prog;
1594 	bool xdp_tx_cmpl = false;
1595 	unsigned int true_bufsz;
1596 	struct sk_buff *skb;
1597 	int pkts_polled = 0;
1598 	int idx;
1599 
1600 	rcu_read_lock();
1601 	xdp_prog = READ_ONCE(dp->xdp_prog);
1602 	true_bufsz = xdp_prog ? PAGE_SIZE : dp->fl_bufsz;
1603 	tx_ring = r_vec->xdp_ring;
1604 
1605 	while (pkts_polled < budget) {
1606 		unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1607 		struct nfp_net_rx_buf *rxbuf;
1608 		struct nfp_net_rx_desc *rxd;
1609 		struct nfp_meta_parsed meta;
1610 		struct net_device *netdev;
1611 		dma_addr_t new_dma_addr;
1612 		u32 meta_len_xdp = 0;
1613 		void *new_frag;
1614 
1615 		idx = D_IDX(rx_ring, rx_ring->rd_p);
1616 
1617 		rxd = &rx_ring->rxds[idx];
1618 		if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1619 			break;
1620 
1621 		/* Memory barrier to ensure that we won't do other reads
1622 		 * before the DD bit.
1623 		 */
1624 		dma_rmb();
1625 
1626 		memset(&meta, 0, sizeof(meta));
1627 
1628 		rx_ring->rd_p++;
1629 		pkts_polled++;
1630 
1631 		rxbuf =	&rx_ring->rxbufs[idx];
1632 		/*         < meta_len >
1633 		 *  <-- [rx_offset] -->
1634 		 *  ---------------------------------------------------------
1635 		 * | [XX] |  metadata  |             packet           | XXXX |
1636 		 *  ---------------------------------------------------------
1637 		 *         <---------------- data_len --------------->
1638 		 *
1639 		 * The rx_offset is fixed for all packets, the meta_len can vary
1640 		 * on a packet by packet basis. If rx_offset is set to zero
1641 		 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1642 		 * buffer and is immediately followed by the packet (no [XX]).
1643 		 */
1644 		meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1645 		data_len = le16_to_cpu(rxd->rxd.data_len);
1646 		pkt_len = data_len - meta_len;
1647 
1648 		pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1649 		if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1650 			pkt_off += meta_len;
1651 		else
1652 			pkt_off += dp->rx_offset;
1653 		meta_off = pkt_off - meta_len;
1654 
1655 		/* Stats update */
1656 		u64_stats_update_begin(&r_vec->rx_sync);
1657 		r_vec->rx_pkts++;
1658 		r_vec->rx_bytes += pkt_len;
1659 		u64_stats_update_end(&r_vec->rx_sync);
1660 
1661 		if (unlikely(meta_len > NFP_NET_MAX_PREPEND ||
1662 			     (dp->rx_offset && meta_len > dp->rx_offset))) {
1663 			nn_dp_warn(dp, "oversized RX packet metadata %u\n",
1664 				   meta_len);
1665 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1666 			continue;
1667 		}
1668 
1669 		nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,
1670 					data_len);
1671 
1672 		if (!dp->chained_metadata_format) {
1673 			nfp_net_set_hash_desc(dp->netdev, &meta,
1674 					      rxbuf->frag + meta_off, rxd);
1675 		} else if (meta_len) {
1676 			void *end;
1677 
1678 			end = nfp_net_parse_meta(dp->netdev, &meta,
1679 						 rxbuf->frag + meta_off,
1680 						 meta_len);
1681 			if (unlikely(end != rxbuf->frag + pkt_off)) {
1682 				nn_dp_warn(dp, "invalid RX packet metadata\n");
1683 				nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1684 						NULL);
1685 				continue;
1686 			}
1687 		}
1688 
1689 		if (xdp_prog && !(rxd->rxd.flags & PCIE_DESC_RX_BPF &&
1690 				  dp->bpf_offload_xdp) && !meta.portid) {
1691 			void *orig_data = rxbuf->frag + pkt_off;
1692 			unsigned int dma_off;
1693 			struct xdp_buff xdp;
1694 			int act;
1695 
1696 			xdp.data_hard_start = rxbuf->frag + NFP_NET_RX_BUF_HEADROOM;
1697 			xdp.data = orig_data;
1698 			xdp.data_meta = orig_data;
1699 			xdp.data_end = orig_data + pkt_len;
1700 
1701 			act = bpf_prog_run_xdp(xdp_prog, &xdp);
1702 
1703 			pkt_len -= xdp.data - orig_data;
1704 			pkt_off += xdp.data - orig_data;
1705 
1706 			switch (act) {
1707 			case XDP_PASS:
1708 				meta_len_xdp = xdp.data - xdp.data_meta;
1709 				break;
1710 			case XDP_TX:
1711 				dma_off = pkt_off - NFP_NET_RX_BUF_HEADROOM;
1712 				if (unlikely(!nfp_net_tx_xdp_buf(dp, rx_ring,
1713 								 tx_ring, rxbuf,
1714 								 dma_off,
1715 								 pkt_len,
1716 								 &xdp_tx_cmpl)))
1717 					trace_xdp_exception(dp->netdev,
1718 							    xdp_prog, act);
1719 				continue;
1720 			default:
1721 				bpf_warn_invalid_xdp_action(act);
1722 				/* fall through */
1723 			case XDP_ABORTED:
1724 				trace_xdp_exception(dp->netdev, xdp_prog, act);
1725 				/* fall through */
1726 			case XDP_DROP:
1727 				nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1728 						    rxbuf->dma_addr);
1729 				continue;
1730 			}
1731 		}
1732 
1733 		skb = build_skb(rxbuf->frag, true_bufsz);
1734 		if (unlikely(!skb)) {
1735 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1736 			continue;
1737 		}
1738 		new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
1739 		if (unlikely(!new_frag)) {
1740 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
1741 			continue;
1742 		}
1743 
1744 		nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
1745 
1746 		nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
1747 
1748 		if (likely(!meta.portid)) {
1749 			netdev = dp->netdev;
1750 		} else {
1751 			struct nfp_net *nn;
1752 
1753 			nn = netdev_priv(dp->netdev);
1754 			netdev = nfp_app_repr_get(nn->app, meta.portid);
1755 			if (unlikely(!netdev)) {
1756 				nfp_net_rx_drop(dp, r_vec, rx_ring, NULL, skb);
1757 				continue;
1758 			}
1759 			nfp_repr_inc_rx_stats(netdev, pkt_len);
1760 		}
1761 
1762 		skb_reserve(skb, pkt_off);
1763 		skb_put(skb, pkt_len);
1764 
1765 		skb->mark = meta.mark;
1766 		skb_set_hash(skb, meta.hash, meta.hash_type);
1767 
1768 		skb_record_rx_queue(skb, rx_ring->idx);
1769 		skb->protocol = eth_type_trans(skb, netdev);
1770 
1771 		nfp_net_rx_csum(dp, r_vec, rxd, &meta, skb);
1772 
1773 		if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
1774 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1775 					       le16_to_cpu(rxd->rxd.vlan));
1776 		if (meta_len_xdp)
1777 			skb_metadata_set(skb, meta_len_xdp);
1778 
1779 		napi_gro_receive(&rx_ring->r_vec->napi, skb);
1780 	}
1781 
1782 	if (xdp_prog) {
1783 		if (tx_ring->wr_ptr_add)
1784 			nfp_net_tx_xmit_more_flush(tx_ring);
1785 		else if (unlikely(tx_ring->wr_p != tx_ring->rd_p) &&
1786 			 !xdp_tx_cmpl)
1787 			if (!nfp_net_xdp_complete(tx_ring))
1788 				pkts_polled = budget;
1789 	}
1790 	rcu_read_unlock();
1791 
1792 	return pkts_polled;
1793 }
1794 
1795 /**
1796  * nfp_net_poll() - napi poll function
1797  * @napi:    NAPI structure
1798  * @budget:  NAPI budget
1799  *
1800  * Return: number of packets polled.
1801  */
1802 static int nfp_net_poll(struct napi_struct *napi, int budget)
1803 {
1804 	struct nfp_net_r_vector *r_vec =
1805 		container_of(napi, struct nfp_net_r_vector, napi);
1806 	unsigned int pkts_polled = 0;
1807 
1808 	if (r_vec->tx_ring)
1809 		nfp_net_tx_complete(r_vec->tx_ring);
1810 	if (r_vec->rx_ring)
1811 		pkts_polled = nfp_net_rx(r_vec->rx_ring, budget);
1812 
1813 	if (pkts_polled < budget)
1814 		if (napi_complete_done(napi, pkts_polled))
1815 			nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
1816 
1817 	return pkts_polled;
1818 }
1819 
1820 /* Control device data path
1821  */
1822 
1823 static bool
1824 nfp_ctrl_tx_one(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1825 		struct sk_buff *skb, bool old)
1826 {
1827 	unsigned int real_len = skb->len, meta_len = 0;
1828 	struct nfp_net_tx_ring *tx_ring;
1829 	struct nfp_net_tx_buf *txbuf;
1830 	struct nfp_net_tx_desc *txd;
1831 	struct nfp_net_dp *dp;
1832 	dma_addr_t dma_addr;
1833 	int wr_idx;
1834 
1835 	dp = &r_vec->nfp_net->dp;
1836 	tx_ring = r_vec->tx_ring;
1837 
1838 	if (WARN_ON_ONCE(skb_shinfo(skb)->nr_frags)) {
1839 		nn_dp_warn(dp, "Driver's CTRL TX does not implement gather\n");
1840 		goto err_free;
1841 	}
1842 
1843 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1844 		u64_stats_update_begin(&r_vec->tx_sync);
1845 		r_vec->tx_busy++;
1846 		u64_stats_update_end(&r_vec->tx_sync);
1847 		if (!old)
1848 			__skb_queue_tail(&r_vec->queue, skb);
1849 		else
1850 			__skb_queue_head(&r_vec->queue, skb);
1851 		return true;
1852 	}
1853 
1854 	if (nfp_app_ctrl_has_meta(nn->app)) {
1855 		if (unlikely(skb_headroom(skb) < 8)) {
1856 			nn_dp_warn(dp, "CTRL TX on skb without headroom\n");
1857 			goto err_free;
1858 		}
1859 		meta_len = 8;
1860 		put_unaligned_be32(NFP_META_PORT_ID_CTRL, skb_push(skb, 4));
1861 		put_unaligned_be32(NFP_NET_META_PORTID, skb_push(skb, 4));
1862 	}
1863 
1864 	/* Start with the head skbuf */
1865 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
1866 				  DMA_TO_DEVICE);
1867 	if (dma_mapping_error(dp->dev, dma_addr))
1868 		goto err_dma_warn;
1869 
1870 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1871 
1872 	/* Stash the soft descriptor of the head then initialize it */
1873 	txbuf = &tx_ring->txbufs[wr_idx];
1874 	txbuf->skb = skb;
1875 	txbuf->dma_addr = dma_addr;
1876 	txbuf->fidx = -1;
1877 	txbuf->pkt_cnt = 1;
1878 	txbuf->real_len = real_len;
1879 
1880 	/* Build TX descriptor */
1881 	txd = &tx_ring->txds[wr_idx];
1882 	txd->offset_eop = meta_len | PCIE_DESC_TX_EOP;
1883 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
1884 	nfp_desc_set_dma_addr(txd, dma_addr);
1885 	txd->data_len = cpu_to_le16(skb->len);
1886 
1887 	txd->flags = 0;
1888 	txd->mss = 0;
1889 	txd->lso_hdrlen = 0;
1890 
1891 	tx_ring->wr_p++;
1892 	tx_ring->wr_ptr_add++;
1893 	nfp_net_tx_xmit_more_flush(tx_ring);
1894 
1895 	return false;
1896 
1897 err_dma_warn:
1898 	nn_dp_warn(dp, "Failed to DMA map TX CTRL buffer\n");
1899 err_free:
1900 	u64_stats_update_begin(&r_vec->tx_sync);
1901 	r_vec->tx_errors++;
1902 	u64_stats_update_end(&r_vec->tx_sync);
1903 	dev_kfree_skb_any(skb);
1904 	return false;
1905 }
1906 
1907 bool nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
1908 {
1909 	struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
1910 	bool ret;
1911 
1912 	spin_lock_bh(&r_vec->lock);
1913 	ret = nfp_ctrl_tx_one(nn, r_vec, skb, false);
1914 	spin_unlock_bh(&r_vec->lock);
1915 
1916 	return ret;
1917 }
1918 
1919 static void __nfp_ctrl_tx_queued(struct nfp_net_r_vector *r_vec)
1920 {
1921 	struct sk_buff *skb;
1922 
1923 	while ((skb = __skb_dequeue(&r_vec->queue)))
1924 		if (nfp_ctrl_tx_one(r_vec->nfp_net, r_vec, skb, true))
1925 			return;
1926 }
1927 
1928 static bool
1929 nfp_ctrl_meta_ok(struct nfp_net *nn, void *data, unsigned int meta_len)
1930 {
1931 	u32 meta_type, meta_tag;
1932 
1933 	if (!nfp_app_ctrl_has_meta(nn->app))
1934 		return !meta_len;
1935 
1936 	if (meta_len != 8)
1937 		return false;
1938 
1939 	meta_type = get_unaligned_be32(data);
1940 	meta_tag = get_unaligned_be32(data + 4);
1941 
1942 	return (meta_type == NFP_NET_META_PORTID &&
1943 		meta_tag == NFP_META_PORT_ID_CTRL);
1944 }
1945 
1946 static bool
1947 nfp_ctrl_rx_one(struct nfp_net *nn, struct nfp_net_dp *dp,
1948 		struct nfp_net_r_vector *r_vec, struct nfp_net_rx_ring *rx_ring)
1949 {
1950 	unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1951 	struct nfp_net_rx_buf *rxbuf;
1952 	struct nfp_net_rx_desc *rxd;
1953 	dma_addr_t new_dma_addr;
1954 	struct sk_buff *skb;
1955 	void *new_frag;
1956 	int idx;
1957 
1958 	idx = D_IDX(rx_ring, rx_ring->rd_p);
1959 
1960 	rxd = &rx_ring->rxds[idx];
1961 	if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1962 		return false;
1963 
1964 	/* Memory barrier to ensure that we won't do other reads
1965 	 * before the DD bit.
1966 	 */
1967 	dma_rmb();
1968 
1969 	rx_ring->rd_p++;
1970 
1971 	rxbuf =	&rx_ring->rxbufs[idx];
1972 	meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1973 	data_len = le16_to_cpu(rxd->rxd.data_len);
1974 	pkt_len = data_len - meta_len;
1975 
1976 	pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1977 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1978 		pkt_off += meta_len;
1979 	else
1980 		pkt_off += dp->rx_offset;
1981 	meta_off = pkt_off - meta_len;
1982 
1983 	/* Stats update */
1984 	u64_stats_update_begin(&r_vec->rx_sync);
1985 	r_vec->rx_pkts++;
1986 	r_vec->rx_bytes += pkt_len;
1987 	u64_stats_update_end(&r_vec->rx_sync);
1988 
1989 	nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,	data_len);
1990 
1991 	if (unlikely(!nfp_ctrl_meta_ok(nn, rxbuf->frag + meta_off, meta_len))) {
1992 		nn_dp_warn(dp, "incorrect metadata for ctrl packet (%d)\n",
1993 			   meta_len);
1994 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1995 		return true;
1996 	}
1997 
1998 	skb = build_skb(rxbuf->frag, dp->fl_bufsz);
1999 	if (unlikely(!skb)) {
2000 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2001 		return true;
2002 	}
2003 	new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
2004 	if (unlikely(!new_frag)) {
2005 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
2006 		return true;
2007 	}
2008 
2009 	nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
2010 
2011 	nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
2012 
2013 	skb_reserve(skb, pkt_off);
2014 	skb_put(skb, pkt_len);
2015 
2016 	nfp_app_ctrl_rx(nn->app, skb);
2017 
2018 	return true;
2019 }
2020 
2021 static void nfp_ctrl_rx(struct nfp_net_r_vector *r_vec)
2022 {
2023 	struct nfp_net_rx_ring *rx_ring = r_vec->rx_ring;
2024 	struct nfp_net *nn = r_vec->nfp_net;
2025 	struct nfp_net_dp *dp = &nn->dp;
2026 
2027 	while (nfp_ctrl_rx_one(nn, dp, r_vec, rx_ring))
2028 		continue;
2029 }
2030 
2031 static void nfp_ctrl_poll(unsigned long arg)
2032 {
2033 	struct nfp_net_r_vector *r_vec = (void *)arg;
2034 
2035 	spin_lock_bh(&r_vec->lock);
2036 	nfp_net_tx_complete(r_vec->tx_ring);
2037 	__nfp_ctrl_tx_queued(r_vec);
2038 	spin_unlock_bh(&r_vec->lock);
2039 
2040 	nfp_ctrl_rx(r_vec);
2041 
2042 	nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2043 }
2044 
2045 /* Setup and Configuration
2046  */
2047 
2048 /**
2049  * nfp_net_vecs_init() - Assign IRQs and setup rvecs.
2050  * @nn:		NFP Network structure
2051  */
2052 static void nfp_net_vecs_init(struct nfp_net *nn)
2053 {
2054 	struct nfp_net_r_vector *r_vec;
2055 	int r;
2056 
2057 	nn->lsc_handler = nfp_net_irq_lsc;
2058 	nn->exn_handler = nfp_net_irq_exn;
2059 
2060 	for (r = 0; r < nn->max_r_vecs; r++) {
2061 		struct msix_entry *entry;
2062 
2063 		entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r];
2064 
2065 		r_vec = &nn->r_vecs[r];
2066 		r_vec->nfp_net = nn;
2067 		r_vec->irq_entry = entry->entry;
2068 		r_vec->irq_vector = entry->vector;
2069 
2070 		if (nn->dp.netdev) {
2071 			r_vec->handler = nfp_net_irq_rxtx;
2072 		} else {
2073 			r_vec->handler = nfp_ctrl_irq_rxtx;
2074 
2075 			__skb_queue_head_init(&r_vec->queue);
2076 			spin_lock_init(&r_vec->lock);
2077 			tasklet_init(&r_vec->tasklet, nfp_ctrl_poll,
2078 				     (unsigned long)r_vec);
2079 			tasklet_disable(&r_vec->tasklet);
2080 		}
2081 
2082 		cpumask_set_cpu(r, &r_vec->affinity_mask);
2083 	}
2084 }
2085 
2086 /**
2087  * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
2088  * @tx_ring:   TX ring to free
2089  */
2090 static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
2091 {
2092 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2093 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2094 
2095 	kfree(tx_ring->txbufs);
2096 
2097 	if (tx_ring->txds)
2098 		dma_free_coherent(dp->dev, tx_ring->size,
2099 				  tx_ring->txds, tx_ring->dma);
2100 
2101 	tx_ring->cnt = 0;
2102 	tx_ring->txbufs = NULL;
2103 	tx_ring->txds = NULL;
2104 	tx_ring->dma = 0;
2105 	tx_ring->size = 0;
2106 }
2107 
2108 /**
2109  * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
2110  * @dp:        NFP Net data path struct
2111  * @tx_ring:   TX Ring structure to allocate
2112  *
2113  * Return: 0 on success, negative errno otherwise.
2114  */
2115 static int
2116 nfp_net_tx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
2117 {
2118 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2119 	int sz;
2120 
2121 	tx_ring->cnt = dp->txd_cnt;
2122 
2123 	tx_ring->size = sizeof(*tx_ring->txds) * tx_ring->cnt;
2124 	tx_ring->txds = dma_zalloc_coherent(dp->dev, tx_ring->size,
2125 					    &tx_ring->dma, GFP_KERNEL);
2126 	if (!tx_ring->txds)
2127 		goto err_alloc;
2128 
2129 	sz = sizeof(*tx_ring->txbufs) * tx_ring->cnt;
2130 	tx_ring->txbufs = kzalloc(sz, GFP_KERNEL);
2131 	if (!tx_ring->txbufs)
2132 		goto err_alloc;
2133 
2134 	if (!tx_ring->is_xdp && dp->netdev)
2135 		netif_set_xps_queue(dp->netdev, &r_vec->affinity_mask,
2136 				    tx_ring->idx);
2137 
2138 	return 0;
2139 
2140 err_alloc:
2141 	nfp_net_tx_ring_free(tx_ring);
2142 	return -ENOMEM;
2143 }
2144 
2145 static void
2146 nfp_net_tx_ring_bufs_free(struct nfp_net_dp *dp,
2147 			  struct nfp_net_tx_ring *tx_ring)
2148 {
2149 	unsigned int i;
2150 
2151 	if (!tx_ring->is_xdp)
2152 		return;
2153 
2154 	for (i = 0; i < tx_ring->cnt; i++) {
2155 		if (!tx_ring->txbufs[i].frag)
2156 			return;
2157 
2158 		nfp_net_dma_unmap_rx(dp, tx_ring->txbufs[i].dma_addr);
2159 		__free_page(virt_to_page(tx_ring->txbufs[i].frag));
2160 	}
2161 }
2162 
2163 static int
2164 nfp_net_tx_ring_bufs_alloc(struct nfp_net_dp *dp,
2165 			   struct nfp_net_tx_ring *tx_ring)
2166 {
2167 	struct nfp_net_tx_buf *txbufs = tx_ring->txbufs;
2168 	unsigned int i;
2169 
2170 	if (!tx_ring->is_xdp)
2171 		return 0;
2172 
2173 	for (i = 0; i < tx_ring->cnt; i++) {
2174 		txbufs[i].frag = nfp_net_rx_alloc_one(dp, &txbufs[i].dma_addr);
2175 		if (!txbufs[i].frag) {
2176 			nfp_net_tx_ring_bufs_free(dp, tx_ring);
2177 			return -ENOMEM;
2178 		}
2179 	}
2180 
2181 	return 0;
2182 }
2183 
2184 static int nfp_net_tx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2185 {
2186 	unsigned int r;
2187 
2188 	dp->tx_rings = kcalloc(dp->num_tx_rings, sizeof(*dp->tx_rings),
2189 			       GFP_KERNEL);
2190 	if (!dp->tx_rings)
2191 		return -ENOMEM;
2192 
2193 	for (r = 0; r < dp->num_tx_rings; r++) {
2194 		int bias = 0;
2195 
2196 		if (r >= dp->num_stack_tx_rings)
2197 			bias = dp->num_stack_tx_rings;
2198 
2199 		nfp_net_tx_ring_init(&dp->tx_rings[r], &nn->r_vecs[r - bias],
2200 				     r, bias);
2201 
2202 		if (nfp_net_tx_ring_alloc(dp, &dp->tx_rings[r]))
2203 			goto err_free_prev;
2204 
2205 		if (nfp_net_tx_ring_bufs_alloc(dp, &dp->tx_rings[r]))
2206 			goto err_free_ring;
2207 	}
2208 
2209 	return 0;
2210 
2211 err_free_prev:
2212 	while (r--) {
2213 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2214 err_free_ring:
2215 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2216 	}
2217 	kfree(dp->tx_rings);
2218 	return -ENOMEM;
2219 }
2220 
2221 static void nfp_net_tx_rings_free(struct nfp_net_dp *dp)
2222 {
2223 	unsigned int r;
2224 
2225 	for (r = 0; r < dp->num_tx_rings; r++) {
2226 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2227 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2228 	}
2229 
2230 	kfree(dp->tx_rings);
2231 }
2232 
2233 /**
2234  * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
2235  * @rx_ring:  RX ring to free
2236  */
2237 static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
2238 {
2239 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
2240 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2241 
2242 	kfree(rx_ring->rxbufs);
2243 
2244 	if (rx_ring->rxds)
2245 		dma_free_coherent(dp->dev, rx_ring->size,
2246 				  rx_ring->rxds, rx_ring->dma);
2247 
2248 	rx_ring->cnt = 0;
2249 	rx_ring->rxbufs = NULL;
2250 	rx_ring->rxds = NULL;
2251 	rx_ring->dma = 0;
2252 	rx_ring->size = 0;
2253 }
2254 
2255 /**
2256  * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
2257  * @dp:	      NFP Net data path struct
2258  * @rx_ring:  RX ring to allocate
2259  *
2260  * Return: 0 on success, negative errno otherwise.
2261  */
2262 static int
2263 nfp_net_rx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring)
2264 {
2265 	int sz;
2266 
2267 	rx_ring->cnt = dp->rxd_cnt;
2268 	rx_ring->size = sizeof(*rx_ring->rxds) * rx_ring->cnt;
2269 	rx_ring->rxds = dma_zalloc_coherent(dp->dev, rx_ring->size,
2270 					    &rx_ring->dma, GFP_KERNEL);
2271 	if (!rx_ring->rxds)
2272 		goto err_alloc;
2273 
2274 	sz = sizeof(*rx_ring->rxbufs) * rx_ring->cnt;
2275 	rx_ring->rxbufs = kzalloc(sz, GFP_KERNEL);
2276 	if (!rx_ring->rxbufs)
2277 		goto err_alloc;
2278 
2279 	return 0;
2280 
2281 err_alloc:
2282 	nfp_net_rx_ring_free(rx_ring);
2283 	return -ENOMEM;
2284 }
2285 
2286 static int nfp_net_rx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2287 {
2288 	unsigned int r;
2289 
2290 	dp->rx_rings = kcalloc(dp->num_rx_rings, sizeof(*dp->rx_rings),
2291 			       GFP_KERNEL);
2292 	if (!dp->rx_rings)
2293 		return -ENOMEM;
2294 
2295 	for (r = 0; r < dp->num_rx_rings; r++) {
2296 		nfp_net_rx_ring_init(&dp->rx_rings[r], &nn->r_vecs[r], r);
2297 
2298 		if (nfp_net_rx_ring_alloc(dp, &dp->rx_rings[r]))
2299 			goto err_free_prev;
2300 
2301 		if (nfp_net_rx_ring_bufs_alloc(dp, &dp->rx_rings[r]))
2302 			goto err_free_ring;
2303 	}
2304 
2305 	return 0;
2306 
2307 err_free_prev:
2308 	while (r--) {
2309 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2310 err_free_ring:
2311 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2312 	}
2313 	kfree(dp->rx_rings);
2314 	return -ENOMEM;
2315 }
2316 
2317 static void nfp_net_rx_rings_free(struct nfp_net_dp *dp)
2318 {
2319 	unsigned int r;
2320 
2321 	for (r = 0; r < dp->num_rx_rings; r++) {
2322 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2323 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2324 	}
2325 
2326 	kfree(dp->rx_rings);
2327 }
2328 
2329 static void
2330 nfp_net_vector_assign_rings(struct nfp_net_dp *dp,
2331 			    struct nfp_net_r_vector *r_vec, int idx)
2332 {
2333 	r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL;
2334 	r_vec->tx_ring =
2335 		idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL;
2336 
2337 	r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ?
2338 		&dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL;
2339 }
2340 
2341 static int
2342 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2343 		       int idx)
2344 {
2345 	int err;
2346 
2347 	/* Setup NAPI */
2348 	if (nn->dp.netdev)
2349 		netif_napi_add(nn->dp.netdev, &r_vec->napi,
2350 			       nfp_net_poll, NAPI_POLL_WEIGHT);
2351 	else
2352 		tasklet_enable(&r_vec->tasklet);
2353 
2354 	snprintf(r_vec->name, sizeof(r_vec->name),
2355 		 "%s-rxtx-%d", nfp_net_name(nn), idx);
2356 	err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name,
2357 			  r_vec);
2358 	if (err) {
2359 		if (nn->dp.netdev)
2360 			netif_napi_del(&r_vec->napi);
2361 		else
2362 			tasklet_disable(&r_vec->tasklet);
2363 
2364 		nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector);
2365 		return err;
2366 	}
2367 	disable_irq(r_vec->irq_vector);
2368 
2369 	irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask);
2370 
2371 	nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector,
2372 	       r_vec->irq_entry);
2373 
2374 	return 0;
2375 }
2376 
2377 static void
2378 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
2379 {
2380 	irq_set_affinity_hint(r_vec->irq_vector, NULL);
2381 	if (nn->dp.netdev)
2382 		netif_napi_del(&r_vec->napi);
2383 	else
2384 		tasklet_disable(&r_vec->tasklet);
2385 
2386 	free_irq(r_vec->irq_vector, r_vec);
2387 }
2388 
2389 /**
2390  * nfp_net_rss_write_itbl() - Write RSS indirection table to device
2391  * @nn:      NFP Net device to reconfigure
2392  */
2393 void nfp_net_rss_write_itbl(struct nfp_net *nn)
2394 {
2395 	int i;
2396 
2397 	for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
2398 		nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
2399 			  get_unaligned_le32(nn->rss_itbl + i));
2400 }
2401 
2402 /**
2403  * nfp_net_rss_write_key() - Write RSS hash key to device
2404  * @nn:      NFP Net device to reconfigure
2405  */
2406 void nfp_net_rss_write_key(struct nfp_net *nn)
2407 {
2408 	int i;
2409 
2410 	for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4)
2411 		nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
2412 			  get_unaligned_le32(nn->rss_key + i));
2413 }
2414 
2415 /**
2416  * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
2417  * @nn:      NFP Net device to reconfigure
2418  */
2419 void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
2420 {
2421 	u8 i;
2422 	u32 factor;
2423 	u32 value;
2424 
2425 	/* Compute factor used to convert coalesce '_usecs' parameters to
2426 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
2427 	 * count.
2428 	 */
2429 	factor = nn->me_freq_mhz / 16;
2430 
2431 	/* copy RX interrupt coalesce parameters */
2432 	value = (nn->rx_coalesce_max_frames << 16) |
2433 		(factor * nn->rx_coalesce_usecs);
2434 	for (i = 0; i < nn->dp.num_rx_rings; i++)
2435 		nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
2436 
2437 	/* copy TX interrupt coalesce parameters */
2438 	value = (nn->tx_coalesce_max_frames << 16) |
2439 		(factor * nn->tx_coalesce_usecs);
2440 	for (i = 0; i < nn->dp.num_tx_rings; i++)
2441 		nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
2442 }
2443 
2444 /**
2445  * nfp_net_write_mac_addr() - Write mac address to the device control BAR
2446  * @nn:      NFP Net device to reconfigure
2447  * @addr:    MAC address to write
2448  *
2449  * Writes the MAC address from the netdev to the device control BAR.  Does not
2450  * perform the required reconfig.  We do a bit of byte swapping dance because
2451  * firmware is LE.
2452  */
2453 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr)
2454 {
2455 	nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr));
2456 	nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4));
2457 }
2458 
2459 static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
2460 {
2461 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
2462 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
2463 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
2464 
2465 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
2466 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
2467 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
2468 }
2469 
2470 /**
2471  * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
2472  * @nn:      NFP Net device to reconfigure
2473  */
2474 static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
2475 {
2476 	u32 new_ctrl, update;
2477 	unsigned int r;
2478 	int err;
2479 
2480 	new_ctrl = nn->dp.ctrl;
2481 	new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
2482 	update = NFP_NET_CFG_UPDATE_GEN;
2483 	update |= NFP_NET_CFG_UPDATE_MSIX;
2484 	update |= NFP_NET_CFG_UPDATE_RING;
2485 
2486 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2487 		new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
2488 
2489 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2490 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2491 
2492 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2493 	err = nfp_net_reconfig(nn, update);
2494 	if (err)
2495 		nn_err(nn, "Could not disable device: %d\n", err);
2496 
2497 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2498 		nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]);
2499 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2500 		nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]);
2501 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2502 		nfp_net_vec_clear_ring_data(nn, r);
2503 
2504 	nn->dp.ctrl = new_ctrl;
2505 }
2506 
2507 static void
2508 nfp_net_rx_ring_hw_cfg_write(struct nfp_net *nn,
2509 			     struct nfp_net_rx_ring *rx_ring, unsigned int idx)
2510 {
2511 	/* Write the DMA address, size and MSI-X info to the device */
2512 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), rx_ring->dma);
2513 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(rx_ring->cnt));
2514 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), rx_ring->r_vec->irq_entry);
2515 }
2516 
2517 static void
2518 nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
2519 			     struct nfp_net_tx_ring *tx_ring, unsigned int idx)
2520 {
2521 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), tx_ring->dma);
2522 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(tx_ring->cnt));
2523 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_entry);
2524 }
2525 
2526 /**
2527  * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2528  * @nn:      NFP Net device to reconfigure
2529  */
2530 static int nfp_net_set_config_and_enable(struct nfp_net *nn)
2531 {
2532 	u32 bufsz, new_ctrl, update = 0;
2533 	unsigned int r;
2534 	int err;
2535 
2536 	new_ctrl = nn->dp.ctrl;
2537 
2538 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) {
2539 		nfp_net_rss_write_key(nn);
2540 		nfp_net_rss_write_itbl(nn);
2541 		nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
2542 		update |= NFP_NET_CFG_UPDATE_RSS;
2543 	}
2544 
2545 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) {
2546 		nfp_net_coalesce_write_cfg(nn);
2547 		update |= NFP_NET_CFG_UPDATE_IRQMOD;
2548 	}
2549 
2550 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2551 		nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r);
2552 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2553 		nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r);
2554 
2555 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->dp.num_tx_rings == 64 ?
2556 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_tx_rings) - 1);
2557 
2558 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->dp.num_rx_rings == 64 ?
2559 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_rx_rings) - 1);
2560 
2561 	if (nn->dp.netdev)
2562 		nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
2563 
2564 	nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu);
2565 
2566 	bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA;
2567 	nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz);
2568 
2569 	/* Enable device */
2570 	new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2571 	update |= NFP_NET_CFG_UPDATE_GEN;
2572 	update |= NFP_NET_CFG_UPDATE_MSIX;
2573 	update |= NFP_NET_CFG_UPDATE_RING;
2574 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2575 		new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2576 
2577 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2578 	err = nfp_net_reconfig(nn, update);
2579 	if (err) {
2580 		nfp_net_clear_config_and_disable(nn);
2581 		return err;
2582 	}
2583 
2584 	nn->dp.ctrl = new_ctrl;
2585 
2586 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2587 		nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]);
2588 
2589 	/* Since reconfiguration requests while NFP is down are ignored we
2590 	 * have to wipe the entire VXLAN configuration and reinitialize it.
2591 	 */
2592 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN) {
2593 		memset(&nn->vxlan_ports, 0, sizeof(nn->vxlan_ports));
2594 		memset(&nn->vxlan_usecnt, 0, sizeof(nn->vxlan_usecnt));
2595 		udp_tunnel_get_rx_info(nn->dp.netdev);
2596 	}
2597 
2598 	return 0;
2599 }
2600 
2601 /**
2602  * nfp_net_close_stack() - Quiesce the stack (part of close)
2603  * @nn:	     NFP Net device to reconfigure
2604  */
2605 static void nfp_net_close_stack(struct nfp_net *nn)
2606 {
2607 	unsigned int r;
2608 
2609 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2610 	netif_carrier_off(nn->dp.netdev);
2611 	nn->link_up = false;
2612 
2613 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2614 		disable_irq(nn->r_vecs[r].irq_vector);
2615 		napi_disable(&nn->r_vecs[r].napi);
2616 	}
2617 
2618 	netif_tx_disable(nn->dp.netdev);
2619 }
2620 
2621 /**
2622  * nfp_net_close_free_all() - Free all runtime resources
2623  * @nn:      NFP Net device to reconfigure
2624  */
2625 static void nfp_net_close_free_all(struct nfp_net *nn)
2626 {
2627 	unsigned int r;
2628 
2629 	nfp_net_tx_rings_free(&nn->dp);
2630 	nfp_net_rx_rings_free(&nn->dp);
2631 
2632 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2633 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2634 
2635 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2636 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2637 }
2638 
2639 /**
2640  * nfp_net_netdev_close() - Called when the device is downed
2641  * @netdev:      netdev structure
2642  */
2643 static int nfp_net_netdev_close(struct net_device *netdev)
2644 {
2645 	struct nfp_net *nn = netdev_priv(netdev);
2646 
2647 	/* Step 1: Disable RX and TX rings from the Linux kernel perspective
2648 	 */
2649 	nfp_net_close_stack(nn);
2650 
2651 	/* Step 2: Tell NFP
2652 	 */
2653 	nfp_net_clear_config_and_disable(nn);
2654 	nfp_port_configure(netdev, false);
2655 
2656 	/* Step 3: Free resources
2657 	 */
2658 	nfp_net_close_free_all(nn);
2659 
2660 	nn_dbg(nn, "%s down", netdev->name);
2661 	return 0;
2662 }
2663 
2664 void nfp_ctrl_close(struct nfp_net *nn)
2665 {
2666 	int r;
2667 
2668 	rtnl_lock();
2669 
2670 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2671 		disable_irq(nn->r_vecs[r].irq_vector);
2672 		tasklet_disable(&nn->r_vecs[r].tasklet);
2673 	}
2674 
2675 	nfp_net_clear_config_and_disable(nn);
2676 
2677 	nfp_net_close_free_all(nn);
2678 
2679 	rtnl_unlock();
2680 }
2681 
2682 /**
2683  * nfp_net_open_stack() - Start the device from stack's perspective
2684  * @nn:      NFP Net device to reconfigure
2685  */
2686 static void nfp_net_open_stack(struct nfp_net *nn)
2687 {
2688 	unsigned int r;
2689 
2690 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2691 		napi_enable(&nn->r_vecs[r].napi);
2692 		enable_irq(nn->r_vecs[r].irq_vector);
2693 	}
2694 
2695 	netif_tx_wake_all_queues(nn->dp.netdev);
2696 
2697 	enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2698 	nfp_net_read_link_status(nn);
2699 }
2700 
2701 static int nfp_net_open_alloc_all(struct nfp_net *nn)
2702 {
2703 	int err, r;
2704 
2705 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2706 				      nn->exn_name, sizeof(nn->exn_name),
2707 				      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2708 	if (err)
2709 		return err;
2710 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2711 				      nn->lsc_name, sizeof(nn->lsc_name),
2712 				      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2713 	if (err)
2714 		goto err_free_exn;
2715 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2716 
2717 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2718 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2719 		if (err)
2720 			goto err_cleanup_vec_p;
2721 	}
2722 
2723 	err = nfp_net_rx_rings_prepare(nn, &nn->dp);
2724 	if (err)
2725 		goto err_cleanup_vec;
2726 
2727 	err = nfp_net_tx_rings_prepare(nn, &nn->dp);
2728 	if (err)
2729 		goto err_free_rx_rings;
2730 
2731 	for (r = 0; r < nn->max_r_vecs; r++)
2732 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2733 
2734 	return 0;
2735 
2736 err_free_rx_rings:
2737 	nfp_net_rx_rings_free(&nn->dp);
2738 err_cleanup_vec:
2739 	r = nn->dp.num_r_vecs;
2740 err_cleanup_vec_p:
2741 	while (r--)
2742 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2743 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2744 err_free_exn:
2745 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2746 	return err;
2747 }
2748 
2749 static int nfp_net_netdev_open(struct net_device *netdev)
2750 {
2751 	struct nfp_net *nn = netdev_priv(netdev);
2752 	int err;
2753 
2754 	/* Step 1: Allocate resources for rings and the like
2755 	 * - Request interrupts
2756 	 * - Allocate RX and TX ring resources
2757 	 * - Setup initial RSS table
2758 	 */
2759 	err = nfp_net_open_alloc_all(nn);
2760 	if (err)
2761 		return err;
2762 
2763 	err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
2764 	if (err)
2765 		goto err_free_all;
2766 
2767 	err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
2768 	if (err)
2769 		goto err_free_all;
2770 
2771 	/* Step 2: Configure the NFP
2772 	 * - Ifup the physical interface if it exists
2773 	 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2774 	 * - Write MAC address (in case it changed)
2775 	 * - Set the MTU
2776 	 * - Set the Freelist buffer size
2777 	 * - Enable the FW
2778 	 */
2779 	err = nfp_port_configure(netdev, true);
2780 	if (err)
2781 		goto err_free_all;
2782 
2783 	err = nfp_net_set_config_and_enable(nn);
2784 	if (err)
2785 		goto err_port_disable;
2786 
2787 	/* Step 3: Enable for kernel
2788 	 * - put some freelist descriptors on each RX ring
2789 	 * - enable NAPI on each ring
2790 	 * - enable all TX queues
2791 	 * - set link state
2792 	 */
2793 	nfp_net_open_stack(nn);
2794 
2795 	return 0;
2796 
2797 err_port_disable:
2798 	nfp_port_configure(netdev, false);
2799 err_free_all:
2800 	nfp_net_close_free_all(nn);
2801 	return err;
2802 }
2803 
2804 int nfp_ctrl_open(struct nfp_net *nn)
2805 {
2806 	int err, r;
2807 
2808 	/* ring dumping depends on vNICs being opened/closed under rtnl */
2809 	rtnl_lock();
2810 
2811 	err = nfp_net_open_alloc_all(nn);
2812 	if (err)
2813 		goto err_unlock;
2814 
2815 	err = nfp_net_set_config_and_enable(nn);
2816 	if (err)
2817 		goto err_free_all;
2818 
2819 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2820 		enable_irq(nn->r_vecs[r].irq_vector);
2821 
2822 	rtnl_unlock();
2823 
2824 	return 0;
2825 
2826 err_free_all:
2827 	nfp_net_close_free_all(nn);
2828 err_unlock:
2829 	rtnl_unlock();
2830 	return err;
2831 }
2832 
2833 static void nfp_net_set_rx_mode(struct net_device *netdev)
2834 {
2835 	struct nfp_net *nn = netdev_priv(netdev);
2836 	u32 new_ctrl;
2837 
2838 	new_ctrl = nn->dp.ctrl;
2839 
2840 	if (netdev->flags & IFF_PROMISC) {
2841 		if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
2842 			new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
2843 		else
2844 			nn_warn(nn, "FW does not support promiscuous mode\n");
2845 	} else {
2846 		new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
2847 	}
2848 
2849 	if (new_ctrl == nn->dp.ctrl)
2850 		return;
2851 
2852 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2853 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
2854 
2855 	nn->dp.ctrl = new_ctrl;
2856 }
2857 
2858 static void nfp_net_rss_init_itbl(struct nfp_net *nn)
2859 {
2860 	int i;
2861 
2862 	for (i = 0; i < sizeof(nn->rss_itbl); i++)
2863 		nn->rss_itbl[i] =
2864 			ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
2865 }
2866 
2867 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
2868 {
2869 	struct nfp_net_dp new_dp = *dp;
2870 
2871 	*dp = nn->dp;
2872 	nn->dp = new_dp;
2873 
2874 	nn->dp.netdev->mtu = new_dp.mtu;
2875 
2876 	if (!netif_is_rxfh_configured(nn->dp.netdev))
2877 		nfp_net_rss_init_itbl(nn);
2878 }
2879 
2880 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
2881 {
2882 	unsigned int r;
2883 	int err;
2884 
2885 	nfp_net_dp_swap(nn, dp);
2886 
2887 	for (r = 0; r <	nn->max_r_vecs; r++)
2888 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2889 
2890 	err = netif_set_real_num_rx_queues(nn->dp.netdev, nn->dp.num_rx_rings);
2891 	if (err)
2892 		return err;
2893 
2894 	if (nn->dp.netdev->real_num_tx_queues != nn->dp.num_stack_tx_rings) {
2895 		err = netif_set_real_num_tx_queues(nn->dp.netdev,
2896 						   nn->dp.num_stack_tx_rings);
2897 		if (err)
2898 			return err;
2899 	}
2900 
2901 	return nfp_net_set_config_and_enable(nn);
2902 }
2903 
2904 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
2905 {
2906 	struct nfp_net_dp *new;
2907 
2908 	new = kmalloc(sizeof(*new), GFP_KERNEL);
2909 	if (!new)
2910 		return NULL;
2911 
2912 	*new = nn->dp;
2913 
2914 	/* Clear things which need to be recomputed */
2915 	new->fl_bufsz = 0;
2916 	new->tx_rings = NULL;
2917 	new->rx_rings = NULL;
2918 	new->num_r_vecs = 0;
2919 	new->num_stack_tx_rings = 0;
2920 
2921 	return new;
2922 }
2923 
2924 static int
2925 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
2926 		     struct netlink_ext_ack *extack)
2927 {
2928 	/* XDP-enabled tests */
2929 	if (!dp->xdp_prog)
2930 		return 0;
2931 	if (dp->fl_bufsz > PAGE_SIZE) {
2932 		NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
2933 		return -EINVAL;
2934 	}
2935 	if (dp->num_tx_rings > nn->max_tx_rings) {
2936 		NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
2937 		return -EINVAL;
2938 	}
2939 
2940 	return 0;
2941 }
2942 
2943 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
2944 			  struct netlink_ext_ack *extack)
2945 {
2946 	int r, err;
2947 
2948 	dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
2949 
2950 	dp->num_stack_tx_rings = dp->num_tx_rings;
2951 	if (dp->xdp_prog)
2952 		dp->num_stack_tx_rings -= dp->num_rx_rings;
2953 
2954 	dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
2955 
2956 	err = nfp_net_check_config(nn, dp, extack);
2957 	if (err)
2958 		goto exit_free_dp;
2959 
2960 	if (!netif_running(dp->netdev)) {
2961 		nfp_net_dp_swap(nn, dp);
2962 		err = 0;
2963 		goto exit_free_dp;
2964 	}
2965 
2966 	/* Prepare new rings */
2967 	for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
2968 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2969 		if (err) {
2970 			dp->num_r_vecs = r;
2971 			goto err_cleanup_vecs;
2972 		}
2973 	}
2974 
2975 	err = nfp_net_rx_rings_prepare(nn, dp);
2976 	if (err)
2977 		goto err_cleanup_vecs;
2978 
2979 	err = nfp_net_tx_rings_prepare(nn, dp);
2980 	if (err)
2981 		goto err_free_rx;
2982 
2983 	/* Stop device, swap in new rings, try to start the firmware */
2984 	nfp_net_close_stack(nn);
2985 	nfp_net_clear_config_and_disable(nn);
2986 
2987 	err = nfp_net_dp_swap_enable(nn, dp);
2988 	if (err) {
2989 		int err2;
2990 
2991 		nfp_net_clear_config_and_disable(nn);
2992 
2993 		/* Try with old configuration and old rings */
2994 		err2 = nfp_net_dp_swap_enable(nn, dp);
2995 		if (err2)
2996 			nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
2997 			       err, err2);
2998 	}
2999 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3000 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3001 
3002 	nfp_net_rx_rings_free(dp);
3003 	nfp_net_tx_rings_free(dp);
3004 
3005 	nfp_net_open_stack(nn);
3006 exit_free_dp:
3007 	kfree(dp);
3008 
3009 	return err;
3010 
3011 err_free_rx:
3012 	nfp_net_rx_rings_free(dp);
3013 err_cleanup_vecs:
3014 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3015 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3016 	kfree(dp);
3017 	return err;
3018 }
3019 
3020 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
3021 {
3022 	struct nfp_net *nn = netdev_priv(netdev);
3023 	struct nfp_net_dp *dp;
3024 
3025 	dp = nfp_net_clone_dp(nn);
3026 	if (!dp)
3027 		return -ENOMEM;
3028 
3029 	dp->mtu = new_mtu;
3030 
3031 	return nfp_net_ring_reconfig(nn, dp, NULL);
3032 }
3033 
3034 static int
3035 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
3036 {
3037 	struct nfp_net *nn = netdev_priv(netdev);
3038 
3039 	/* Priority tagged packets with vlan id 0 are processed by the
3040 	 * NFP as untagged packets
3041 	 */
3042 	if (!vid)
3043 		return 0;
3044 
3045 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_VID, vid);
3046 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_PROTO, ETH_P_8021Q);
3047 
3048 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD);
3049 }
3050 
3051 static int
3052 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
3053 {
3054 	struct nfp_net *nn = netdev_priv(netdev);
3055 
3056 	/* Priority tagged packets with vlan id 0 are processed by the
3057 	 * NFP as untagged packets
3058 	 */
3059 	if (!vid)
3060 		return 0;
3061 
3062 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_VID, vid);
3063 	nn_writew(nn, NFP_NET_CFG_VLAN_FILTER_PROTO, ETH_P_8021Q);
3064 
3065 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL);
3066 }
3067 
3068 static void nfp_net_stat64(struct net_device *netdev,
3069 			   struct rtnl_link_stats64 *stats)
3070 {
3071 	struct nfp_net *nn = netdev_priv(netdev);
3072 	int r;
3073 
3074 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
3075 		struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
3076 		u64 data[3];
3077 		unsigned int start;
3078 
3079 		do {
3080 			start = u64_stats_fetch_begin(&r_vec->rx_sync);
3081 			data[0] = r_vec->rx_pkts;
3082 			data[1] = r_vec->rx_bytes;
3083 			data[2] = r_vec->rx_drops;
3084 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
3085 		stats->rx_packets += data[0];
3086 		stats->rx_bytes += data[1];
3087 		stats->rx_dropped += data[2];
3088 
3089 		do {
3090 			start = u64_stats_fetch_begin(&r_vec->tx_sync);
3091 			data[0] = r_vec->tx_pkts;
3092 			data[1] = r_vec->tx_bytes;
3093 			data[2] = r_vec->tx_errors;
3094 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
3095 		stats->tx_packets += data[0];
3096 		stats->tx_bytes += data[1];
3097 		stats->tx_errors += data[2];
3098 	}
3099 }
3100 
3101 static int nfp_net_set_features(struct net_device *netdev,
3102 				netdev_features_t features)
3103 {
3104 	netdev_features_t changed = netdev->features ^ features;
3105 	struct nfp_net *nn = netdev_priv(netdev);
3106 	u32 new_ctrl;
3107 	int err;
3108 
3109 	/* Assume this is not called with features we have not advertised */
3110 
3111 	new_ctrl = nn->dp.ctrl;
3112 
3113 	if (changed & NETIF_F_RXCSUM) {
3114 		if (features & NETIF_F_RXCSUM)
3115 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3116 		else
3117 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
3118 	}
3119 
3120 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3121 		if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
3122 			new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3123 		else
3124 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
3125 	}
3126 
3127 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
3128 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
3129 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3130 					      NFP_NET_CFG_CTRL_LSO;
3131 		else
3132 			new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3133 	}
3134 
3135 	if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
3136 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3137 			new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3138 		else
3139 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
3140 	}
3141 
3142 	if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
3143 		if (features & NETIF_F_HW_VLAN_CTAG_TX)
3144 			new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3145 		else
3146 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
3147 	}
3148 
3149 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
3150 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
3151 			new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3152 		else
3153 			new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
3154 	}
3155 
3156 	if (changed & NETIF_F_SG) {
3157 		if (features & NETIF_F_SG)
3158 			new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
3159 		else
3160 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
3161 	}
3162 
3163 	if (changed & NETIF_F_HW_TC && nfp_app_tc_busy(nn->app, nn)) {
3164 		nn_err(nn, "Cannot disable HW TC offload while in use\n");
3165 		return -EBUSY;
3166 	}
3167 
3168 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
3169 	       netdev->features, features, changed);
3170 
3171 	if (new_ctrl == nn->dp.ctrl)
3172 		return 0;
3173 
3174 	nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
3175 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3176 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
3177 	if (err)
3178 		return err;
3179 
3180 	nn->dp.ctrl = new_ctrl;
3181 
3182 	return 0;
3183 }
3184 
3185 static netdev_features_t
3186 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
3187 		       netdev_features_t features)
3188 {
3189 	u8 l4_hdr;
3190 
3191 	/* We can't do TSO over double tagged packets (802.1AD) */
3192 	features &= vlan_features_check(skb, features);
3193 
3194 	if (!skb->encapsulation)
3195 		return features;
3196 
3197 	/* Ensure that inner L4 header offset fits into TX descriptor field */
3198 	if (skb_is_gso(skb)) {
3199 		u32 hdrlen;
3200 
3201 		hdrlen = skb_inner_transport_header(skb) - skb->data +
3202 			inner_tcp_hdrlen(skb);
3203 
3204 		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
3205 			features &= ~NETIF_F_GSO_MASK;
3206 	}
3207 
3208 	/* VXLAN/GRE check */
3209 	switch (vlan_get_protocol(skb)) {
3210 	case htons(ETH_P_IP):
3211 		l4_hdr = ip_hdr(skb)->protocol;
3212 		break;
3213 	case htons(ETH_P_IPV6):
3214 		l4_hdr = ipv6_hdr(skb)->nexthdr;
3215 		break;
3216 	default:
3217 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3218 	}
3219 
3220 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
3221 	    skb->inner_protocol != htons(ETH_P_TEB) ||
3222 	    (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
3223 	    (l4_hdr == IPPROTO_UDP &&
3224 	     (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
3225 	      sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
3226 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3227 
3228 	return features;
3229 }
3230 
3231 /**
3232  * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
3233  * @nn:   NFP Net device to reconfigure
3234  * @idx:  Index into the port table where new port should be written
3235  * @port: UDP port to configure (pass zero to remove VXLAN port)
3236  */
3237 static void nfp_net_set_vxlan_port(struct nfp_net *nn, int idx, __be16 port)
3238 {
3239 	int i;
3240 
3241 	nn->vxlan_ports[idx] = port;
3242 
3243 	if (!(nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN))
3244 		return;
3245 
3246 	BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
3247 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2)
3248 		nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(port),
3249 			  be16_to_cpu(nn->vxlan_ports[i + 1]) << 16 |
3250 			  be16_to_cpu(nn->vxlan_ports[i]));
3251 
3252 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_VXLAN);
3253 }
3254 
3255 /**
3256  * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
3257  * @nn:   NFP Network structure
3258  * @port: UDP port to look for
3259  *
3260  * Return: if the port is already in the table -- it's position;
3261  *	   if the port is not in the table -- free position to use;
3262  *	   if the table is full -- -ENOSPC.
3263  */
3264 static int nfp_net_find_vxlan_idx(struct nfp_net *nn, __be16 port)
3265 {
3266 	int i, free_idx = -ENOSPC;
3267 
3268 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i++) {
3269 		if (nn->vxlan_ports[i] == port)
3270 			return i;
3271 		if (!nn->vxlan_usecnt[i])
3272 			free_idx = i;
3273 	}
3274 
3275 	return free_idx;
3276 }
3277 
3278 static void nfp_net_add_vxlan_port(struct net_device *netdev,
3279 				   struct udp_tunnel_info *ti)
3280 {
3281 	struct nfp_net *nn = netdev_priv(netdev);
3282 	int idx;
3283 
3284 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3285 		return;
3286 
3287 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3288 	if (idx == -ENOSPC)
3289 		return;
3290 
3291 	if (!nn->vxlan_usecnt[idx]++)
3292 		nfp_net_set_vxlan_port(nn, idx, ti->port);
3293 }
3294 
3295 static void nfp_net_del_vxlan_port(struct net_device *netdev,
3296 				   struct udp_tunnel_info *ti)
3297 {
3298 	struct nfp_net *nn = netdev_priv(netdev);
3299 	int idx;
3300 
3301 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3302 		return;
3303 
3304 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3305 	if (idx == -ENOSPC || !nn->vxlan_usecnt[idx])
3306 		return;
3307 
3308 	if (!--nn->vxlan_usecnt[idx])
3309 		nfp_net_set_vxlan_port(nn, idx, 0);
3310 }
3311 
3312 static int
3313 nfp_net_xdp_setup_drv(struct nfp_net *nn, struct bpf_prog *prog,
3314 		      struct netlink_ext_ack *extack)
3315 {
3316 	struct nfp_net_dp *dp;
3317 
3318 	if (!prog == !nn->dp.xdp_prog) {
3319 		WRITE_ONCE(nn->dp.xdp_prog, prog);
3320 		return 0;
3321 	}
3322 
3323 	dp = nfp_net_clone_dp(nn);
3324 	if (!dp)
3325 		return -ENOMEM;
3326 
3327 	dp->xdp_prog = prog;
3328 	dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
3329 	dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
3330 	dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
3331 
3332 	/* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
3333 	return nfp_net_ring_reconfig(nn, dp, extack);
3334 }
3335 
3336 static int
3337 nfp_net_xdp_setup(struct nfp_net *nn, struct bpf_prog *prog, u32 flags,
3338 		  struct netlink_ext_ack *extack)
3339 {
3340 	struct bpf_prog *drv_prog, *offload_prog;
3341 	int err;
3342 
3343 	if (nn->xdp_prog && (flags ^ nn->xdp_flags) & XDP_FLAGS_MODES)
3344 		return -EBUSY;
3345 
3346 	/* Load both when no flags set to allow easy activation of driver path
3347 	 * when program is replaced by one which can't be offloaded.
3348 	 */
3349 	drv_prog     = flags & XDP_FLAGS_HW_MODE  ? NULL : prog;
3350 	offload_prog = flags & XDP_FLAGS_DRV_MODE ? NULL : prog;
3351 
3352 	err = nfp_net_xdp_setup_drv(nn, drv_prog, extack);
3353 	if (err)
3354 		return err;
3355 
3356 	err = nfp_app_xdp_offload(nn->app, nn, offload_prog);
3357 	if (err && flags & XDP_FLAGS_HW_MODE)
3358 		return err;
3359 
3360 	if (nn->xdp_prog)
3361 		bpf_prog_put(nn->xdp_prog);
3362 	nn->xdp_prog = prog;
3363 	nn->xdp_flags = flags;
3364 
3365 	return 0;
3366 }
3367 
3368 static int nfp_net_xdp(struct net_device *netdev, struct netdev_xdp *xdp)
3369 {
3370 	struct nfp_net *nn = netdev_priv(netdev);
3371 
3372 	switch (xdp->command) {
3373 	case XDP_SETUP_PROG:
3374 	case XDP_SETUP_PROG_HW:
3375 		return nfp_net_xdp_setup(nn, xdp->prog, xdp->flags,
3376 					 xdp->extack);
3377 	case XDP_QUERY_PROG:
3378 		xdp->prog_attached = !!nn->xdp_prog;
3379 		if (nn->dp.bpf_offload_xdp)
3380 			xdp->prog_attached = XDP_ATTACHED_HW;
3381 		xdp->prog_id = nn->xdp_prog ? nn->xdp_prog->aux->id : 0;
3382 		return 0;
3383 	default:
3384 		return -EINVAL;
3385 	}
3386 }
3387 
3388 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
3389 {
3390 	struct nfp_net *nn = netdev_priv(netdev);
3391 	struct sockaddr *saddr = addr;
3392 	int err;
3393 
3394 	err = eth_prepare_mac_addr_change(netdev, addr);
3395 	if (err)
3396 		return err;
3397 
3398 	nfp_net_write_mac_addr(nn, saddr->sa_data);
3399 
3400 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
3401 	if (err)
3402 		return err;
3403 
3404 	eth_commit_mac_addr_change(netdev, addr);
3405 
3406 	return 0;
3407 }
3408 
3409 const struct net_device_ops nfp_net_netdev_ops = {
3410 	.ndo_open		= nfp_net_netdev_open,
3411 	.ndo_stop		= nfp_net_netdev_close,
3412 	.ndo_start_xmit		= nfp_net_tx,
3413 	.ndo_get_stats64	= nfp_net_stat64,
3414 	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
3415 	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
3416 	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
3417 	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
3418 	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
3419 	.ndo_get_vf_config	= nfp_app_get_vf_config,
3420 	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
3421 	.ndo_setup_tc		= nfp_port_setup_tc,
3422 	.ndo_tx_timeout		= nfp_net_tx_timeout,
3423 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
3424 	.ndo_change_mtu		= nfp_net_change_mtu,
3425 	.ndo_set_mac_address	= nfp_net_set_mac_address,
3426 	.ndo_set_features	= nfp_net_set_features,
3427 	.ndo_features_check	= nfp_net_features_check,
3428 	.ndo_get_phys_port_name	= nfp_port_get_phys_port_name,
3429 	.ndo_udp_tunnel_add	= nfp_net_add_vxlan_port,
3430 	.ndo_udp_tunnel_del	= nfp_net_del_vxlan_port,
3431 	.ndo_xdp		= nfp_net_xdp,
3432 };
3433 
3434 /**
3435  * nfp_net_info() - Print general info about the NIC
3436  * @nn:      NFP Net device to reconfigure
3437  */
3438 void nfp_net_info(struct nfp_net *nn)
3439 {
3440 	nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
3441 		nn->dp.is_vf ? "VF " : "",
3442 		nn->dp.num_tx_rings, nn->max_tx_rings,
3443 		nn->dp.num_rx_rings, nn->max_rx_rings);
3444 	nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3445 		nn->fw_ver.resv, nn->fw_ver.class,
3446 		nn->fw_ver.major, nn->fw_ver.minor,
3447 		nn->max_mtu);
3448 	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",
3449 		nn->cap,
3450 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
3451 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
3452 		nn->cap & NFP_NET_CFG_CTRL_L2MC     ? "L2MCFILT " : "",
3453 		nn->cap & NFP_NET_CFG_CTRL_RXCSUM   ? "RXCSUM "   : "",
3454 		nn->cap & NFP_NET_CFG_CTRL_TXCSUM   ? "TXCSUM "   : "",
3455 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN   ? "RXVLAN "   : "",
3456 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN   ? "TXVLAN "   : "",
3457 		nn->cap & NFP_NET_CFG_CTRL_SCATTER  ? "SCATTER "  : "",
3458 		nn->cap & NFP_NET_CFG_CTRL_GATHER   ? "GATHER "   : "",
3459 		nn->cap & NFP_NET_CFG_CTRL_LSO      ? "TSO1 "     : "",
3460 		nn->cap & NFP_NET_CFG_CTRL_LSO2     ? "TSO2 "     : "",
3461 		nn->cap & NFP_NET_CFG_CTRL_RSS      ? "RSS1 "     : "",
3462 		nn->cap & NFP_NET_CFG_CTRL_RSS2     ? "RSS2 "     : "",
3463 		nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "",
3464 		nn->cap & NFP_NET_CFG_CTRL_L2SWITCH ? "L2SWITCH " : "",
3465 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
3466 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
3467 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
3468 		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
3469 		nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ?
3470 						      "RXCSUM_COMPLETE " : "",
3471 		nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "",
3472 		nfp_app_extra_cap(nn->app, nn));
3473 }
3474 
3475 /**
3476  * nfp_net_alloc() - Allocate netdev and related structure
3477  * @pdev:         PCI device
3478  * @needs_netdev: Whether to allocate a netdev for this vNIC
3479  * @max_tx_rings: Maximum number of TX rings supported by device
3480  * @max_rx_rings: Maximum number of RX rings supported by device
3481  *
3482  * This function allocates a netdev device and fills in the initial
3483  * part of the @struct nfp_net structure.  In case of control device
3484  * nfp_net structure is allocated without the netdev.
3485  *
3486  * Return: NFP Net device structure, or ERR_PTR on error.
3487  */
3488 struct nfp_net *nfp_net_alloc(struct pci_dev *pdev, bool needs_netdev,
3489 			      unsigned int max_tx_rings,
3490 			      unsigned int max_rx_rings)
3491 {
3492 	struct nfp_net *nn;
3493 
3494 	if (needs_netdev) {
3495 		struct net_device *netdev;
3496 
3497 		netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
3498 					    max_tx_rings, max_rx_rings);
3499 		if (!netdev)
3500 			return ERR_PTR(-ENOMEM);
3501 
3502 		SET_NETDEV_DEV(netdev, &pdev->dev);
3503 		nn = netdev_priv(netdev);
3504 		nn->dp.netdev = netdev;
3505 	} else {
3506 		nn = vzalloc(sizeof(*nn));
3507 		if (!nn)
3508 			return ERR_PTR(-ENOMEM);
3509 	}
3510 
3511 	nn->dp.dev = &pdev->dev;
3512 	nn->pdev = pdev;
3513 
3514 	nn->max_tx_rings = max_tx_rings;
3515 	nn->max_rx_rings = max_rx_rings;
3516 
3517 	nn->dp.num_tx_rings = min_t(unsigned int,
3518 				    max_tx_rings, num_online_cpus());
3519 	nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings,
3520 				 netif_get_num_default_rss_queues());
3521 
3522 	nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings);
3523 	nn->dp.num_r_vecs = min_t(unsigned int,
3524 				  nn->dp.num_r_vecs, num_online_cpus());
3525 
3526 	nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
3527 	nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
3528 
3529 	spin_lock_init(&nn->reconfig_lock);
3530 	spin_lock_init(&nn->link_status_lock);
3531 
3532 	setup_timer(&nn->reconfig_timer,
3533 		    nfp_net_reconfig_timer, (unsigned long)nn);
3534 
3535 	return nn;
3536 }
3537 
3538 /**
3539  * nfp_net_free() - Undo what @nfp_net_alloc() did
3540  * @nn:      NFP Net device to reconfigure
3541  */
3542 void nfp_net_free(struct nfp_net *nn)
3543 {
3544 	if (nn->xdp_prog)
3545 		bpf_prog_put(nn->xdp_prog);
3546 
3547 	if (nn->dp.netdev)
3548 		free_netdev(nn->dp.netdev);
3549 	else
3550 		vfree(nn);
3551 }
3552 
3553 /**
3554  * nfp_net_rss_key_sz() - Get current size of the RSS key
3555  * @nn:		NFP Net device instance
3556  *
3557  * Return: size of the RSS key for currently selected hash function.
3558  */
3559 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn)
3560 {
3561 	switch (nn->rss_hfunc) {
3562 	case ETH_RSS_HASH_TOP:
3563 		return NFP_NET_CFG_RSS_KEY_SZ;
3564 	case ETH_RSS_HASH_XOR:
3565 		return 0;
3566 	case ETH_RSS_HASH_CRC32:
3567 		return 4;
3568 	}
3569 
3570 	nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc);
3571 	return 0;
3572 }
3573 
3574 /**
3575  * nfp_net_rss_init() - Set the initial RSS parameters
3576  * @nn:	     NFP Net device to reconfigure
3577  */
3578 static void nfp_net_rss_init(struct nfp_net *nn)
3579 {
3580 	unsigned long func_bit, rss_cap_hfunc;
3581 	u32 reg;
3582 
3583 	/* Read the RSS function capability and select first supported func */
3584 	reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP);
3585 	rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg);
3586 	if (!rss_cap_hfunc)
3587 		rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC,
3588 					  NFP_NET_CFG_RSS_TOEPLITZ);
3589 
3590 	func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS);
3591 	if (func_bit == NFP_NET_CFG_RSS_HFUNCS) {
3592 		dev_warn(nn->dp.dev,
3593 			 "Bad RSS config, defaulting to Toeplitz hash\n");
3594 		func_bit = ETH_RSS_HASH_TOP_BIT;
3595 	}
3596 	nn->rss_hfunc = 1 << func_bit;
3597 
3598 	netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn));
3599 
3600 	nfp_net_rss_init_itbl(nn);
3601 
3602 	/* Enable IPv4/IPv6 TCP by default */
3603 	nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
3604 		      NFP_NET_CFG_RSS_IPV6_TCP |
3605 		      FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) |
3606 		      NFP_NET_CFG_RSS_MASK;
3607 }
3608 
3609 /**
3610  * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
3611  * @nn:	     NFP Net device to reconfigure
3612  */
3613 static void nfp_net_irqmod_init(struct nfp_net *nn)
3614 {
3615 	nn->rx_coalesce_usecs      = 50;
3616 	nn->rx_coalesce_max_frames = 64;
3617 	nn->tx_coalesce_usecs      = 50;
3618 	nn->tx_coalesce_max_frames = 64;
3619 }
3620 
3621 static void nfp_net_netdev_init(struct nfp_net *nn)
3622 {
3623 	struct net_device *netdev = nn->dp.netdev;
3624 
3625 	nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
3626 
3627 	netdev->mtu = nn->dp.mtu;
3628 
3629 	/* Advertise/enable offloads based on capabilities
3630 	 *
3631 	 * Note: netdev->features show the currently enabled features
3632 	 * and netdev->hw_features advertises which features are
3633 	 * supported.  By default we enable most features.
3634 	 */
3635 	if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
3636 		netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3637 
3638 	netdev->hw_features = NETIF_F_HIGHDMA;
3639 	if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) {
3640 		netdev->hw_features |= NETIF_F_RXCSUM;
3641 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3642 	}
3643 	if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
3644 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3645 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3646 	}
3647 	if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
3648 		netdev->hw_features |= NETIF_F_SG;
3649 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER;
3650 	}
3651 	if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
3652 	    nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3653 		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3654 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3655 					 NFP_NET_CFG_CTRL_LSO;
3656 	}
3657 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY)
3658 		netdev->hw_features |= NETIF_F_RXHASH;
3659 	if (nn->cap & NFP_NET_CFG_CTRL_VXLAN &&
3660 	    nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
3661 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3662 			netdev->hw_features |= NETIF_F_GSO_GRE |
3663 					       NETIF_F_GSO_UDP_TUNNEL;
3664 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE;
3665 
3666 		netdev->hw_enc_features = netdev->hw_features;
3667 	}
3668 
3669 	netdev->vlan_features = netdev->hw_features;
3670 
3671 	if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
3672 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3673 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3674 	}
3675 	if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
3676 		if (nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3677 			nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
3678 		} else {
3679 			netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3680 			nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3681 		}
3682 	}
3683 	if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) {
3684 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3685 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3686 	}
3687 
3688 	netdev->features = netdev->hw_features;
3689 
3690 	if (nfp_app_has_tc(nn->app))
3691 		netdev->hw_features |= NETIF_F_HW_TC;
3692 
3693 	/* Advertise but disable TSO by default. */
3694 	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
3695 	nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3696 
3697 	/* Finalise the netdev setup */
3698 	netdev->netdev_ops = &nfp_net_netdev_ops;
3699 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
3700 
3701 	SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
3702 
3703 	/* MTU range: 68 - hw-specific max */
3704 	netdev->min_mtu = ETH_MIN_MTU;
3705 	netdev->max_mtu = nn->max_mtu;
3706 
3707 	netif_carrier_off(netdev);
3708 
3709 	nfp_net_set_ethtool_ops(netdev);
3710 }
3711 
3712 /**
3713  * nfp_net_init() - Initialise/finalise the nfp_net structure
3714  * @nn:		NFP Net device structure
3715  *
3716  * Return: 0 on success or negative errno on error.
3717  */
3718 int nfp_net_init(struct nfp_net *nn)
3719 {
3720 	int err;
3721 
3722 	nn->dp.rx_dma_dir = DMA_FROM_DEVICE;
3723 
3724 	/* Get some of the read-only fields from the BAR */
3725 	nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
3726 	nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
3727 
3728 	/* ABI 4.x and ctrl vNIC always use chained metadata, in other cases
3729 	 * we allow use of non-chained metadata if RSS(v1) is the only
3730 	 * advertised capability requiring metadata.
3731 	 */
3732 	nn->dp.chained_metadata_format = nn->fw_ver.major == 4 ||
3733 					 !nn->dp.netdev ||
3734 					 !(nn->cap & NFP_NET_CFG_CTRL_RSS) ||
3735 					 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META;
3736 	/* RSS(v1) uses non-chained metadata format, except in ABI 4.x where
3737 	 * it has the same meaning as RSSv2.
3738 	 */
3739 	if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4)
3740 		nn->cap &= ~NFP_NET_CFG_CTRL_RSS;
3741 
3742 	/* Determine RX packet/metadata boundary offset */
3743 	if (nn->fw_ver.major >= 2) {
3744 		u32 reg;
3745 
3746 		reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
3747 		if (reg > NFP_NET_MAX_PREPEND) {
3748 			nn_err(nn, "Invalid rx offset: %d\n", reg);
3749 			return -EINVAL;
3750 		}
3751 		nn->dp.rx_offset = reg;
3752 	} else {
3753 		nn->dp.rx_offset = NFP_NET_RX_OFFSET;
3754 	}
3755 
3756 	/* Set default MTU and Freelist buffer size */
3757 	if (nn->max_mtu < NFP_NET_DEFAULT_MTU)
3758 		nn->dp.mtu = nn->max_mtu;
3759 	else
3760 		nn->dp.mtu = NFP_NET_DEFAULT_MTU;
3761 	nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp);
3762 
3763 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) {
3764 		nfp_net_rss_init(nn);
3765 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?:
3766 					 NFP_NET_CFG_CTRL_RSS;
3767 	}
3768 
3769 	/* Allow L2 Broadcast and Multicast through by default, if supported */
3770 	if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
3771 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC;
3772 	if (nn->cap & NFP_NET_CFG_CTRL_L2MC)
3773 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2MC;
3774 
3775 	/* Allow IRQ moderation, if supported */
3776 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
3777 		nfp_net_irqmod_init(nn);
3778 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
3779 	}
3780 
3781 	if (nn->dp.netdev)
3782 		nfp_net_netdev_init(nn);
3783 
3784 	/* Stash the re-configuration queue away.  First odd queue in TX Bar */
3785 	nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
3786 
3787 	/* Make sure the FW knows the netdev is supposed to be disabled here */
3788 	nn_writel(nn, NFP_NET_CFG_CTRL, 0);
3789 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
3790 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
3791 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
3792 				   NFP_NET_CFG_UPDATE_GEN);
3793 	if (err)
3794 		return err;
3795 
3796 	nfp_net_vecs_init(nn);
3797 
3798 	if (!nn->dp.netdev)
3799 		return 0;
3800 	return register_netdev(nn->dp.netdev);
3801 }
3802 
3803 /**
3804  * nfp_net_clean() - Undo what nfp_net_init() did.
3805  * @nn:		NFP Net device structure
3806  */
3807 void nfp_net_clean(struct nfp_net *nn)
3808 {
3809 	if (!nn->dp.netdev)
3810 		return;
3811 
3812 	unregister_netdev(nn->dp.netdev);
3813 }
3814