xref: /linux/drivers/net/ethernet/netronome/nfp/nfp_net_common.c (revision e9f0878c4b2004ac19581274c1ae4c61ae3ca70e)
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/mm.h>
57 #include <linux/overflow.h>
58 #include <linux/page_ref.h>
59 #include <linux/pci.h>
60 #include <linux/pci_regs.h>
61 #include <linux/msi.h>
62 #include <linux/ethtool.h>
63 #include <linux/log2.h>
64 #include <linux/if_vlan.h>
65 #include <linux/random.h>
66 #include <linux/vmalloc.h>
67 #include <linux/ktime.h>
68 
69 #include <net/switchdev.h>
70 #include <net/vxlan.h>
71 
72 #include "nfpcore/nfp_nsp.h"
73 #include "nfp_app.h"
74 #include "nfp_net_ctrl.h"
75 #include "nfp_net.h"
76 #include "nfp_net_sriov.h"
77 #include "nfp_port.h"
78 
79 /**
80  * nfp_net_get_fw_version() - Read and parse the FW version
81  * @fw_ver:	Output fw_version structure to read to
82  * @ctrl_bar:	Mapped address of the control BAR
83  */
84 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
85 			    void __iomem *ctrl_bar)
86 {
87 	u32 reg;
88 
89 	reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
90 	put_unaligned_le32(reg, fw_ver);
91 }
92 
93 static dma_addr_t nfp_net_dma_map_rx(struct nfp_net_dp *dp, void *frag)
94 {
95 	return dma_map_single_attrs(dp->dev, frag + NFP_NET_RX_BUF_HEADROOM,
96 				    dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
97 				    dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
98 }
99 
100 static void
101 nfp_net_dma_sync_dev_rx(const struct nfp_net_dp *dp, dma_addr_t dma_addr)
102 {
103 	dma_sync_single_for_device(dp->dev, dma_addr,
104 				   dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
105 				   dp->rx_dma_dir);
106 }
107 
108 static void nfp_net_dma_unmap_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr)
109 {
110 	dma_unmap_single_attrs(dp->dev, dma_addr,
111 			       dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
112 			       dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
113 }
114 
115 static void nfp_net_dma_sync_cpu_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr,
116 				    unsigned int len)
117 {
118 	dma_sync_single_for_cpu(dp->dev, dma_addr - NFP_NET_RX_BUF_HEADROOM,
119 				len, dp->rx_dma_dir);
120 }
121 
122 /* Firmware reconfig
123  *
124  * Firmware reconfig may take a while so we have two versions of it -
125  * synchronous and asynchronous (posted).  All synchronous callers are holding
126  * RTNL so we don't have to worry about serializing them.
127  */
128 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
129 {
130 	nn_writel(nn, NFP_NET_CFG_UPDATE, update);
131 	/* ensure update is written before pinging HW */
132 	nn_pci_flush(nn);
133 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
134 }
135 
136 /* Pass 0 as update to run posted reconfigs. */
137 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
138 {
139 	update |= nn->reconfig_posted;
140 	nn->reconfig_posted = 0;
141 
142 	nfp_net_reconfig_start(nn, update);
143 
144 	nn->reconfig_timer_active = true;
145 	mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
146 }
147 
148 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
149 {
150 	u32 reg;
151 
152 	reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
153 	if (reg == 0)
154 		return true;
155 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
156 		nn_err(nn, "Reconfig error: 0x%08x\n", reg);
157 		return true;
158 	} else if (last_check) {
159 		nn_err(nn, "Reconfig timeout: 0x%08x\n", reg);
160 		return true;
161 	}
162 
163 	return false;
164 }
165 
166 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
167 {
168 	bool timed_out = false;
169 
170 	/* Poll update field, waiting for NFP to ack the config */
171 	while (!nfp_net_reconfig_check_done(nn, timed_out)) {
172 		msleep(1);
173 		timed_out = time_is_before_eq_jiffies(deadline);
174 	}
175 
176 	if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
177 		return -EIO;
178 
179 	return timed_out ? -EIO : 0;
180 }
181 
182 static void nfp_net_reconfig_timer(struct timer_list *t)
183 {
184 	struct nfp_net *nn = from_timer(nn, t, reconfig_timer);
185 
186 	spin_lock_bh(&nn->reconfig_lock);
187 
188 	nn->reconfig_timer_active = false;
189 
190 	/* If sync caller is present it will take over from us */
191 	if (nn->reconfig_sync_present)
192 		goto done;
193 
194 	/* Read reconfig status and report errors */
195 	nfp_net_reconfig_check_done(nn, true);
196 
197 	if (nn->reconfig_posted)
198 		nfp_net_reconfig_start_async(nn, 0);
199 done:
200 	spin_unlock_bh(&nn->reconfig_lock);
201 }
202 
203 /**
204  * nfp_net_reconfig_post() - Post async reconfig request
205  * @nn:      NFP Net device to reconfigure
206  * @update:  The value for the update field in the BAR config
207  *
208  * Record FW reconfiguration request.  Reconfiguration will be kicked off
209  * whenever reconfiguration machinery is idle.  Multiple requests can be
210  * merged together!
211  */
212 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
213 {
214 	spin_lock_bh(&nn->reconfig_lock);
215 
216 	/* Sync caller will kick off async reconf when it's done, just post */
217 	if (nn->reconfig_sync_present) {
218 		nn->reconfig_posted |= update;
219 		goto done;
220 	}
221 
222 	/* Opportunistically check if the previous command is done */
223 	if (!nn->reconfig_timer_active ||
224 	    nfp_net_reconfig_check_done(nn, false))
225 		nfp_net_reconfig_start_async(nn, update);
226 	else
227 		nn->reconfig_posted |= update;
228 done:
229 	spin_unlock_bh(&nn->reconfig_lock);
230 }
231 
232 /**
233  * nfp_net_reconfig() - Reconfigure the firmware
234  * @nn:      NFP Net device to reconfigure
235  * @update:  The value for the update field in the BAR config
236  *
237  * Write the update word to the BAR and ping the reconfig queue.  The
238  * poll until the firmware has acknowledged the update by zeroing the
239  * update word.
240  *
241  * Return: Negative errno on error, 0 on success
242  */
243 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
244 {
245 	bool cancelled_timer = false;
246 	u32 pre_posted_requests;
247 	int ret;
248 
249 	spin_lock_bh(&nn->reconfig_lock);
250 
251 	nn->reconfig_sync_present = true;
252 
253 	if (nn->reconfig_timer_active) {
254 		del_timer(&nn->reconfig_timer);
255 		nn->reconfig_timer_active = false;
256 		cancelled_timer = true;
257 	}
258 	pre_posted_requests = nn->reconfig_posted;
259 	nn->reconfig_posted = 0;
260 
261 	spin_unlock_bh(&nn->reconfig_lock);
262 
263 	if (cancelled_timer)
264 		nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
265 
266 	/* Run the posted reconfigs which were issued before we started */
267 	if (pre_posted_requests) {
268 		nfp_net_reconfig_start(nn, pre_posted_requests);
269 		nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
270 	}
271 
272 	nfp_net_reconfig_start(nn, update);
273 	ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
274 
275 	spin_lock_bh(&nn->reconfig_lock);
276 
277 	if (nn->reconfig_posted)
278 		nfp_net_reconfig_start_async(nn, 0);
279 
280 	nn->reconfig_sync_present = false;
281 
282 	spin_unlock_bh(&nn->reconfig_lock);
283 
284 	return ret;
285 }
286 
287 /**
288  * nfp_net_reconfig_mbox() - Reconfigure the firmware via the mailbox
289  * @nn:        NFP Net device to reconfigure
290  * @mbox_cmd:  The value for the mailbox command
291  *
292  * Helper function for mailbox updates
293  *
294  * Return: Negative errno on error, 0 on success
295  */
296 static int nfp_net_reconfig_mbox(struct nfp_net *nn, u32 mbox_cmd)
297 {
298 	u32 mbox = nn->tlv_caps.mbox_off;
299 	int ret;
300 
301 	if (!nfp_net_has_mbox(&nn->tlv_caps)) {
302 		nn_err(nn, "no mailbox present, command: %u\n", mbox_cmd);
303 		return -EIO;
304 	}
305 
306 	nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
307 
308 	ret = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
309 	if (ret) {
310 		nn_err(nn, "Mailbox update error\n");
311 		return ret;
312 	}
313 
314 	return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
315 }
316 
317 /* Interrupt configuration and handling
318  */
319 
320 /**
321  * nfp_net_irq_unmask() - Unmask automasked interrupt
322  * @nn:       NFP Network structure
323  * @entry_nr: MSI-X table entry
324  *
325  * Clear the ICR for the IRQ entry.
326  */
327 static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
328 {
329 	nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
330 	nn_pci_flush(nn);
331 }
332 
333 /**
334  * nfp_net_irqs_alloc() - allocates MSI-X irqs
335  * @pdev:        PCI device structure
336  * @irq_entries: Array to be initialized and used to hold the irq entries
337  * @min_irqs:    Minimal acceptable number of interrupts
338  * @wanted_irqs: Target number of interrupts to allocate
339  *
340  * Return: Number of irqs obtained or 0 on error.
341  */
342 unsigned int
343 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
344 		   unsigned int min_irqs, unsigned int wanted_irqs)
345 {
346 	unsigned int i;
347 	int got_irqs;
348 
349 	for (i = 0; i < wanted_irqs; i++)
350 		irq_entries[i].entry = i;
351 
352 	got_irqs = pci_enable_msix_range(pdev, irq_entries,
353 					 min_irqs, wanted_irqs);
354 	if (got_irqs < 0) {
355 		dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
356 			min_irqs, wanted_irqs, got_irqs);
357 		return 0;
358 	}
359 
360 	if (got_irqs < wanted_irqs)
361 		dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
362 			 wanted_irqs, got_irqs);
363 
364 	return got_irqs;
365 }
366 
367 /**
368  * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
369  * @nn:		 NFP Network structure
370  * @irq_entries: Table of allocated interrupts
371  * @n:		 Size of @irq_entries (number of entries to grab)
372  *
373  * After interrupts are allocated with nfp_net_irqs_alloc() this function
374  * should be called to assign them to a specific netdev (port).
375  */
376 void
377 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
378 		    unsigned int n)
379 {
380 	struct nfp_net_dp *dp = &nn->dp;
381 
382 	nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
383 	dp->num_r_vecs = nn->max_r_vecs;
384 
385 	memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
386 
387 	if (dp->num_rx_rings > dp->num_r_vecs ||
388 	    dp->num_tx_rings > dp->num_r_vecs)
389 		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
390 			 dp->num_rx_rings, dp->num_tx_rings,
391 			 dp->num_r_vecs);
392 
393 	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
394 	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
395 	dp->num_stack_tx_rings = dp->num_tx_rings;
396 }
397 
398 /**
399  * nfp_net_irqs_disable() - Disable interrupts
400  * @pdev:        PCI device structure
401  *
402  * Undoes what @nfp_net_irqs_alloc() does.
403  */
404 void nfp_net_irqs_disable(struct pci_dev *pdev)
405 {
406 	pci_disable_msix(pdev);
407 }
408 
409 /**
410  * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
411  * @irq:      Interrupt
412  * @data:     Opaque data structure
413  *
414  * Return: Indicate if the interrupt has been handled.
415  */
416 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
417 {
418 	struct nfp_net_r_vector *r_vec = data;
419 
420 	napi_schedule_irqoff(&r_vec->napi);
421 
422 	/* The FW auto-masks any interrupt, either via the MASK bit in
423 	 * the MSI-X table or via the per entry ICR field.  So there
424 	 * is no need to disable interrupts here.
425 	 */
426 	return IRQ_HANDLED;
427 }
428 
429 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
430 {
431 	struct nfp_net_r_vector *r_vec = data;
432 
433 	tasklet_schedule(&r_vec->tasklet);
434 
435 	return IRQ_HANDLED;
436 }
437 
438 /**
439  * nfp_net_read_link_status() - Reread link status from control BAR
440  * @nn:       NFP Network structure
441  */
442 static void nfp_net_read_link_status(struct nfp_net *nn)
443 {
444 	unsigned long flags;
445 	bool link_up;
446 	u32 sts;
447 
448 	spin_lock_irqsave(&nn->link_status_lock, flags);
449 
450 	sts = nn_readl(nn, NFP_NET_CFG_STS);
451 	link_up = !!(sts & NFP_NET_CFG_STS_LINK);
452 
453 	if (nn->link_up == link_up)
454 		goto out;
455 
456 	nn->link_up = link_up;
457 	if (nn->port)
458 		set_bit(NFP_PORT_CHANGED, &nn->port->flags);
459 
460 	if (nn->link_up) {
461 		netif_carrier_on(nn->dp.netdev);
462 		netdev_info(nn->dp.netdev, "NIC Link is Up\n");
463 	} else {
464 		netif_carrier_off(nn->dp.netdev);
465 		netdev_info(nn->dp.netdev, "NIC Link is Down\n");
466 	}
467 out:
468 	spin_unlock_irqrestore(&nn->link_status_lock, flags);
469 }
470 
471 /**
472  * nfp_net_irq_lsc() - Interrupt service routine for link state changes
473  * @irq:      Interrupt
474  * @data:     Opaque data structure
475  *
476  * Return: Indicate if the interrupt has been handled.
477  */
478 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
479 {
480 	struct nfp_net *nn = data;
481 	struct msix_entry *entry;
482 
483 	entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
484 
485 	nfp_net_read_link_status(nn);
486 
487 	nfp_net_irq_unmask(nn, entry->entry);
488 
489 	return IRQ_HANDLED;
490 }
491 
492 /**
493  * nfp_net_irq_exn() - Interrupt service routine for exceptions
494  * @irq:      Interrupt
495  * @data:     Opaque data structure
496  *
497  * Return: Indicate if the interrupt has been handled.
498  */
499 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
500 {
501 	struct nfp_net *nn = data;
502 
503 	nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
504 	/* XXX TO BE IMPLEMENTED */
505 	return IRQ_HANDLED;
506 }
507 
508 /**
509  * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
510  * @tx_ring:  TX ring structure
511  * @r_vec:    IRQ vector servicing this ring
512  * @idx:      Ring index
513  * @is_xdp:   Is this an XDP TX ring?
514  */
515 static void
516 nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
517 		     struct nfp_net_r_vector *r_vec, unsigned int idx,
518 		     bool is_xdp)
519 {
520 	struct nfp_net *nn = r_vec->nfp_net;
521 
522 	tx_ring->idx = idx;
523 	tx_ring->r_vec = r_vec;
524 	tx_ring->is_xdp = is_xdp;
525 	u64_stats_init(&tx_ring->r_vec->tx_sync);
526 
527 	tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
528 	tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
529 }
530 
531 /**
532  * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
533  * @rx_ring:  RX ring structure
534  * @r_vec:    IRQ vector servicing this ring
535  * @idx:      Ring index
536  */
537 static void
538 nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
539 		     struct nfp_net_r_vector *r_vec, unsigned int idx)
540 {
541 	struct nfp_net *nn = r_vec->nfp_net;
542 
543 	rx_ring->idx = idx;
544 	rx_ring->r_vec = r_vec;
545 	u64_stats_init(&rx_ring->r_vec->rx_sync);
546 
547 	rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
548 	rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
549 }
550 
551 /**
552  * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
553  * @nn:		NFP Network structure
554  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
555  * @format:	printf-style format to construct the interrupt name
556  * @name:	Pointer to allocated space for interrupt name
557  * @name_sz:	Size of space for interrupt name
558  * @vector_idx:	Index of MSI-X vector used for this interrupt
559  * @handler:	IRQ handler to register for this interrupt
560  */
561 static int
562 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
563 			const char *format, char *name, size_t name_sz,
564 			unsigned int vector_idx, irq_handler_t handler)
565 {
566 	struct msix_entry *entry;
567 	int err;
568 
569 	entry = &nn->irq_entries[vector_idx];
570 
571 	snprintf(name, name_sz, format, nfp_net_name(nn));
572 	err = request_irq(entry->vector, handler, 0, name, nn);
573 	if (err) {
574 		nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
575 		       entry->vector, err);
576 		return err;
577 	}
578 	nn_writeb(nn, ctrl_offset, entry->entry);
579 	nfp_net_irq_unmask(nn, entry->entry);
580 
581 	return 0;
582 }
583 
584 /**
585  * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
586  * @nn:		NFP Network structure
587  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
588  * @vector_idx:	Index of MSI-X vector used for this interrupt
589  */
590 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
591 				 unsigned int vector_idx)
592 {
593 	nn_writeb(nn, ctrl_offset, 0xff);
594 	nn_pci_flush(nn);
595 	free_irq(nn->irq_entries[vector_idx].vector, nn);
596 }
597 
598 /* Transmit
599  *
600  * One queue controller peripheral queue is used for transmit.  The
601  * driver en-queues packets for transmit by advancing the write
602  * pointer.  The device indicates that packets have transmitted by
603  * advancing the read pointer.  The driver maintains a local copy of
604  * the read and write pointer in @struct nfp_net_tx_ring.  The driver
605  * keeps @wr_p in sync with the queue controller write pointer and can
606  * determine how many packets have been transmitted by comparing its
607  * copy of the read pointer @rd_p with the read pointer maintained by
608  * the queue controller peripheral.
609  */
610 
611 /**
612  * nfp_net_tx_full() - Check if the TX ring is full
613  * @tx_ring: TX ring to check
614  * @dcnt:    Number of descriptors that need to be enqueued (must be >= 1)
615  *
616  * This function checks, based on the *host copy* of read/write
617  * pointer if a given TX ring is full.  The real TX queue may have
618  * some newly made available slots.
619  *
620  * Return: True if the ring is full.
621  */
622 static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
623 {
624 	return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
625 }
626 
627 /* Wrappers for deciding when to stop and restart TX queues */
628 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
629 {
630 	return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
631 }
632 
633 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
634 {
635 	return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
636 }
637 
638 /**
639  * nfp_net_tx_ring_stop() - stop tx ring
640  * @nd_q:    netdev queue
641  * @tx_ring: driver tx queue structure
642  *
643  * Safely stop TX ring.  Remember that while we are running .start_xmit()
644  * someone else may be cleaning the TX ring completions so we need to be
645  * extra careful here.
646  */
647 static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
648 				 struct nfp_net_tx_ring *tx_ring)
649 {
650 	netif_tx_stop_queue(nd_q);
651 
652 	/* We can race with the TX completion out of NAPI so recheck */
653 	smp_mb();
654 	if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
655 		netif_tx_start_queue(nd_q);
656 }
657 
658 /**
659  * nfp_net_tx_tso() - Set up Tx descriptor for LSO
660  * @r_vec: per-ring structure
661  * @txbuf: Pointer to driver soft TX descriptor
662  * @txd: Pointer to HW TX descriptor
663  * @skb: Pointer to SKB
664  *
665  * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
666  * Return error on packet header greater than maximum supported LSO header size.
667  */
668 static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
669 			   struct nfp_net_tx_buf *txbuf,
670 			   struct nfp_net_tx_desc *txd, struct sk_buff *skb)
671 {
672 	u32 hdrlen;
673 	u16 mss;
674 
675 	if (!skb_is_gso(skb))
676 		return;
677 
678 	if (!skb->encapsulation) {
679 		txd->l3_offset = skb_network_offset(skb);
680 		txd->l4_offset = skb_transport_offset(skb);
681 		hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
682 	} else {
683 		txd->l3_offset = skb_inner_network_offset(skb);
684 		txd->l4_offset = skb_inner_transport_offset(skb);
685 		hdrlen = skb_inner_transport_header(skb) - skb->data +
686 			inner_tcp_hdrlen(skb);
687 	}
688 
689 	txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
690 	txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
691 
692 	mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
693 	txd->lso_hdrlen = hdrlen;
694 	txd->mss = cpu_to_le16(mss);
695 	txd->flags |= PCIE_DESC_TX_LSO;
696 
697 	u64_stats_update_begin(&r_vec->tx_sync);
698 	r_vec->tx_lso++;
699 	u64_stats_update_end(&r_vec->tx_sync);
700 }
701 
702 /**
703  * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
704  * @dp:  NFP Net data path struct
705  * @r_vec: per-ring structure
706  * @txbuf: Pointer to driver soft TX descriptor
707  * @txd: Pointer to TX descriptor
708  * @skb: Pointer to SKB
709  *
710  * This function sets the TX checksum flags in the TX descriptor based
711  * on the configuration and the protocol of the packet to be transmitted.
712  */
713 static void nfp_net_tx_csum(struct nfp_net_dp *dp,
714 			    struct nfp_net_r_vector *r_vec,
715 			    struct nfp_net_tx_buf *txbuf,
716 			    struct nfp_net_tx_desc *txd, struct sk_buff *skb)
717 {
718 	struct ipv6hdr *ipv6h;
719 	struct iphdr *iph;
720 	u8 l4_hdr;
721 
722 	if (!(dp->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
723 		return;
724 
725 	if (skb->ip_summed != CHECKSUM_PARTIAL)
726 		return;
727 
728 	txd->flags |= PCIE_DESC_TX_CSUM;
729 	if (skb->encapsulation)
730 		txd->flags |= PCIE_DESC_TX_ENCAP;
731 
732 	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
733 	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
734 
735 	if (iph->version == 4) {
736 		txd->flags |= PCIE_DESC_TX_IP4_CSUM;
737 		l4_hdr = iph->protocol;
738 	} else if (ipv6h->version == 6) {
739 		l4_hdr = ipv6h->nexthdr;
740 	} else {
741 		nn_dp_warn(dp, "partial checksum but ipv=%x!\n", iph->version);
742 		return;
743 	}
744 
745 	switch (l4_hdr) {
746 	case IPPROTO_TCP:
747 		txd->flags |= PCIE_DESC_TX_TCP_CSUM;
748 		break;
749 	case IPPROTO_UDP:
750 		txd->flags |= PCIE_DESC_TX_UDP_CSUM;
751 		break;
752 	default:
753 		nn_dp_warn(dp, "partial checksum but l4 proto=%x!\n", l4_hdr);
754 		return;
755 	}
756 
757 	u64_stats_update_begin(&r_vec->tx_sync);
758 	if (skb->encapsulation)
759 		r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
760 	else
761 		r_vec->hw_csum_tx += txbuf->pkt_cnt;
762 	u64_stats_update_end(&r_vec->tx_sync);
763 }
764 
765 static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
766 {
767 	wmb();
768 	nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
769 	tx_ring->wr_ptr_add = 0;
770 }
771 
772 static int nfp_net_prep_port_id(struct sk_buff *skb)
773 {
774 	struct metadata_dst *md_dst = skb_metadata_dst(skb);
775 	unsigned char *data;
776 
777 	if (likely(!md_dst))
778 		return 0;
779 	if (unlikely(md_dst->type != METADATA_HW_PORT_MUX))
780 		return 0;
781 
782 	if (unlikely(skb_cow_head(skb, 8)))
783 		return -ENOMEM;
784 
785 	data = skb_push(skb, 8);
786 	put_unaligned_be32(NFP_NET_META_PORTID, data);
787 	put_unaligned_be32(md_dst->u.port_info.port_id, data + 4);
788 
789 	return 8;
790 }
791 
792 /**
793  * nfp_net_tx() - Main transmit entry point
794  * @skb:    SKB to transmit
795  * @netdev: netdev structure
796  *
797  * Return: NETDEV_TX_OK on success.
798  */
799 static int nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
800 {
801 	struct nfp_net *nn = netdev_priv(netdev);
802 	const struct skb_frag_struct *frag;
803 	struct nfp_net_tx_desc *txd, txdg;
804 	int f, nr_frags, wr_idx, md_bytes;
805 	struct nfp_net_tx_ring *tx_ring;
806 	struct nfp_net_r_vector *r_vec;
807 	struct nfp_net_tx_buf *txbuf;
808 	struct netdev_queue *nd_q;
809 	struct nfp_net_dp *dp;
810 	dma_addr_t dma_addr;
811 	unsigned int fsize;
812 	u16 qidx;
813 
814 	dp = &nn->dp;
815 	qidx = skb_get_queue_mapping(skb);
816 	tx_ring = &dp->tx_rings[qidx];
817 	r_vec = tx_ring->r_vec;
818 	nd_q = netdev_get_tx_queue(dp->netdev, qidx);
819 
820 	nr_frags = skb_shinfo(skb)->nr_frags;
821 
822 	if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
823 		nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
824 			   qidx, tx_ring->wr_p, tx_ring->rd_p);
825 		netif_tx_stop_queue(nd_q);
826 		nfp_net_tx_xmit_more_flush(tx_ring);
827 		u64_stats_update_begin(&r_vec->tx_sync);
828 		r_vec->tx_busy++;
829 		u64_stats_update_end(&r_vec->tx_sync);
830 		return NETDEV_TX_BUSY;
831 	}
832 
833 	md_bytes = nfp_net_prep_port_id(skb);
834 	if (unlikely(md_bytes < 0)) {
835 		nfp_net_tx_xmit_more_flush(tx_ring);
836 		dev_kfree_skb_any(skb);
837 		return NETDEV_TX_OK;
838 	}
839 
840 	/* Start with the head skbuf */
841 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
842 				  DMA_TO_DEVICE);
843 	if (dma_mapping_error(dp->dev, dma_addr))
844 		goto err_free;
845 
846 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
847 
848 	/* Stash the soft descriptor of the head then initialize it */
849 	txbuf = &tx_ring->txbufs[wr_idx];
850 	txbuf->skb = skb;
851 	txbuf->dma_addr = dma_addr;
852 	txbuf->fidx = -1;
853 	txbuf->pkt_cnt = 1;
854 	txbuf->real_len = skb->len;
855 
856 	/* Build TX descriptor */
857 	txd = &tx_ring->txds[wr_idx];
858 	txd->offset_eop = (nr_frags ? 0 : PCIE_DESC_TX_EOP) | md_bytes;
859 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
860 	nfp_desc_set_dma_addr(txd, dma_addr);
861 	txd->data_len = cpu_to_le16(skb->len);
862 
863 	txd->flags = 0;
864 	txd->mss = 0;
865 	txd->lso_hdrlen = 0;
866 
867 	/* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
868 	nfp_net_tx_tso(r_vec, txbuf, txd, skb);
869 	nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
870 	if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
871 		txd->flags |= PCIE_DESC_TX_VLAN;
872 		txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
873 	}
874 
875 	/* Gather DMA */
876 	if (nr_frags > 0) {
877 		/* all descs must match except for in addr, length and eop */
878 		txdg = *txd;
879 
880 		for (f = 0; f < nr_frags; f++) {
881 			frag = &skb_shinfo(skb)->frags[f];
882 			fsize = skb_frag_size(frag);
883 
884 			dma_addr = skb_frag_dma_map(dp->dev, frag, 0,
885 						    fsize, DMA_TO_DEVICE);
886 			if (dma_mapping_error(dp->dev, dma_addr))
887 				goto err_unmap;
888 
889 			wr_idx = D_IDX(tx_ring, wr_idx + 1);
890 			tx_ring->txbufs[wr_idx].skb = skb;
891 			tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
892 			tx_ring->txbufs[wr_idx].fidx = f;
893 
894 			txd = &tx_ring->txds[wr_idx];
895 			*txd = txdg;
896 			txd->dma_len = cpu_to_le16(fsize);
897 			nfp_desc_set_dma_addr(txd, dma_addr);
898 			txd->offset_eop |=
899 				(f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0;
900 		}
901 
902 		u64_stats_update_begin(&r_vec->tx_sync);
903 		r_vec->tx_gather++;
904 		u64_stats_update_end(&r_vec->tx_sync);
905 	}
906 
907 	netdev_tx_sent_queue(nd_q, txbuf->real_len);
908 
909 	skb_tx_timestamp(skb);
910 
911 	tx_ring->wr_p += nr_frags + 1;
912 	if (nfp_net_tx_ring_should_stop(tx_ring))
913 		nfp_net_tx_ring_stop(nd_q, tx_ring);
914 
915 	tx_ring->wr_ptr_add += nr_frags + 1;
916 	if (!skb->xmit_more || netif_xmit_stopped(nd_q))
917 		nfp_net_tx_xmit_more_flush(tx_ring);
918 
919 	return NETDEV_TX_OK;
920 
921 err_unmap:
922 	while (--f >= 0) {
923 		frag = &skb_shinfo(skb)->frags[f];
924 		dma_unmap_page(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
925 			       skb_frag_size(frag), DMA_TO_DEVICE);
926 		tx_ring->txbufs[wr_idx].skb = NULL;
927 		tx_ring->txbufs[wr_idx].dma_addr = 0;
928 		tx_ring->txbufs[wr_idx].fidx = -2;
929 		wr_idx = wr_idx - 1;
930 		if (wr_idx < 0)
931 			wr_idx += tx_ring->cnt;
932 	}
933 	dma_unmap_single(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
934 			 skb_headlen(skb), DMA_TO_DEVICE);
935 	tx_ring->txbufs[wr_idx].skb = NULL;
936 	tx_ring->txbufs[wr_idx].dma_addr = 0;
937 	tx_ring->txbufs[wr_idx].fidx = -2;
938 err_free:
939 	nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
940 	nfp_net_tx_xmit_more_flush(tx_ring);
941 	u64_stats_update_begin(&r_vec->tx_sync);
942 	r_vec->tx_errors++;
943 	u64_stats_update_end(&r_vec->tx_sync);
944 	dev_kfree_skb_any(skb);
945 	return NETDEV_TX_OK;
946 }
947 
948 /**
949  * nfp_net_tx_complete() - Handled completed TX packets
950  * @tx_ring:	TX ring structure
951  * @budget:	NAPI budget (only used as bool to determine if in NAPI context)
952  */
953 static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget)
954 {
955 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
956 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
957 	const struct skb_frag_struct *frag;
958 	struct netdev_queue *nd_q;
959 	u32 done_pkts = 0, done_bytes = 0;
960 	struct sk_buff *skb;
961 	int todo, nr_frags;
962 	u32 qcp_rd_p;
963 	int fidx;
964 	int idx;
965 
966 	if (tx_ring->wr_p == tx_ring->rd_p)
967 		return;
968 
969 	/* Work out how many descriptors have been transmitted */
970 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
971 
972 	if (qcp_rd_p == tx_ring->qcp_rd_p)
973 		return;
974 
975 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
976 
977 	while (todo--) {
978 		idx = D_IDX(tx_ring, tx_ring->rd_p++);
979 
980 		skb = tx_ring->txbufs[idx].skb;
981 		if (!skb)
982 			continue;
983 
984 		nr_frags = skb_shinfo(skb)->nr_frags;
985 		fidx = tx_ring->txbufs[idx].fidx;
986 
987 		if (fidx == -1) {
988 			/* unmap head */
989 			dma_unmap_single(dp->dev, tx_ring->txbufs[idx].dma_addr,
990 					 skb_headlen(skb), DMA_TO_DEVICE);
991 
992 			done_pkts += tx_ring->txbufs[idx].pkt_cnt;
993 			done_bytes += tx_ring->txbufs[idx].real_len;
994 		} else {
995 			/* unmap fragment */
996 			frag = &skb_shinfo(skb)->frags[fidx];
997 			dma_unmap_page(dp->dev, tx_ring->txbufs[idx].dma_addr,
998 				       skb_frag_size(frag), DMA_TO_DEVICE);
999 		}
1000 
1001 		/* check for last gather fragment */
1002 		if (fidx == nr_frags - 1)
1003 			napi_consume_skb(skb, budget);
1004 
1005 		tx_ring->txbufs[idx].dma_addr = 0;
1006 		tx_ring->txbufs[idx].skb = NULL;
1007 		tx_ring->txbufs[idx].fidx = -2;
1008 	}
1009 
1010 	tx_ring->qcp_rd_p = qcp_rd_p;
1011 
1012 	u64_stats_update_begin(&r_vec->tx_sync);
1013 	r_vec->tx_bytes += done_bytes;
1014 	r_vec->tx_pkts += done_pkts;
1015 	u64_stats_update_end(&r_vec->tx_sync);
1016 
1017 	if (!dp->netdev)
1018 		return;
1019 
1020 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1021 	netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
1022 	if (nfp_net_tx_ring_should_wake(tx_ring)) {
1023 		/* Make sure TX thread will see updated tx_ring->rd_p */
1024 		smp_mb();
1025 
1026 		if (unlikely(netif_tx_queue_stopped(nd_q)))
1027 			netif_tx_wake_queue(nd_q);
1028 	}
1029 
1030 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1031 		  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1032 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1033 }
1034 
1035 static bool nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
1036 {
1037 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1038 	u32 done_pkts = 0, done_bytes = 0;
1039 	bool done_all;
1040 	int idx, todo;
1041 	u32 qcp_rd_p;
1042 
1043 	/* Work out how many descriptors have been transmitted */
1044 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1045 
1046 	if (qcp_rd_p == tx_ring->qcp_rd_p)
1047 		return true;
1048 
1049 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1050 
1051 	done_all = todo <= NFP_NET_XDP_MAX_COMPLETE;
1052 	todo = min(todo, NFP_NET_XDP_MAX_COMPLETE);
1053 
1054 	tx_ring->qcp_rd_p = D_IDX(tx_ring, tx_ring->qcp_rd_p + todo);
1055 
1056 	done_pkts = todo;
1057 	while (todo--) {
1058 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1059 		tx_ring->rd_p++;
1060 
1061 		done_bytes += tx_ring->txbufs[idx].real_len;
1062 	}
1063 
1064 	u64_stats_update_begin(&r_vec->tx_sync);
1065 	r_vec->tx_bytes += done_bytes;
1066 	r_vec->tx_pkts += done_pkts;
1067 	u64_stats_update_end(&r_vec->tx_sync);
1068 
1069 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1070 		  "XDP TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1071 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1072 
1073 	return done_all;
1074 }
1075 
1076 /**
1077  * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
1078  * @dp:		NFP Net data path struct
1079  * @tx_ring:	TX ring structure
1080  *
1081  * Assumes that the device is stopped, must be idempotent.
1082  */
1083 static void
1084 nfp_net_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
1085 {
1086 	const struct skb_frag_struct *frag;
1087 	struct netdev_queue *nd_q;
1088 
1089 	while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
1090 		struct nfp_net_tx_buf *tx_buf;
1091 		struct sk_buff *skb;
1092 		int idx, nr_frags;
1093 
1094 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1095 		tx_buf = &tx_ring->txbufs[idx];
1096 
1097 		skb = tx_ring->txbufs[idx].skb;
1098 		nr_frags = skb_shinfo(skb)->nr_frags;
1099 
1100 		if (tx_buf->fidx == -1) {
1101 			/* unmap head */
1102 			dma_unmap_single(dp->dev, tx_buf->dma_addr,
1103 					 skb_headlen(skb), DMA_TO_DEVICE);
1104 		} else {
1105 			/* unmap fragment */
1106 			frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1107 			dma_unmap_page(dp->dev, tx_buf->dma_addr,
1108 				       skb_frag_size(frag), DMA_TO_DEVICE);
1109 		}
1110 
1111 		/* check for last gather fragment */
1112 		if (tx_buf->fidx == nr_frags - 1)
1113 			dev_kfree_skb_any(skb);
1114 
1115 		tx_buf->dma_addr = 0;
1116 		tx_buf->skb = NULL;
1117 		tx_buf->fidx = -2;
1118 
1119 		tx_ring->qcp_rd_p++;
1120 		tx_ring->rd_p++;
1121 	}
1122 
1123 	memset(tx_ring->txds, 0, tx_ring->size);
1124 	tx_ring->wr_p = 0;
1125 	tx_ring->rd_p = 0;
1126 	tx_ring->qcp_rd_p = 0;
1127 	tx_ring->wr_ptr_add = 0;
1128 
1129 	if (tx_ring->is_xdp || !dp->netdev)
1130 		return;
1131 
1132 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1133 	netdev_tx_reset_queue(nd_q);
1134 }
1135 
1136 static void nfp_net_tx_timeout(struct net_device *netdev)
1137 {
1138 	struct nfp_net *nn = netdev_priv(netdev);
1139 	int i;
1140 
1141 	for (i = 0; i < nn->dp.netdev->real_num_tx_queues; i++) {
1142 		if (!netif_tx_queue_stopped(netdev_get_tx_queue(netdev, i)))
1143 			continue;
1144 		nn_warn(nn, "TX timeout on ring: %d\n", i);
1145 	}
1146 	nn_warn(nn, "TX watchdog timeout\n");
1147 }
1148 
1149 /* Receive processing
1150  */
1151 static unsigned int
1152 nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
1153 {
1154 	unsigned int fl_bufsz;
1155 
1156 	fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
1157 	fl_bufsz += dp->rx_dma_off;
1158 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1159 		fl_bufsz += NFP_NET_MAX_PREPEND;
1160 	else
1161 		fl_bufsz += dp->rx_offset;
1162 	fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
1163 
1164 	fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1165 	fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1166 
1167 	return fl_bufsz;
1168 }
1169 
1170 static void
1171 nfp_net_free_frag(void *frag, bool xdp)
1172 {
1173 	if (!xdp)
1174 		skb_free_frag(frag);
1175 	else
1176 		__free_page(virt_to_page(frag));
1177 }
1178 
1179 /**
1180  * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
1181  * @dp:		NFP Net data path struct
1182  * @dma_addr:	Pointer to storage for DMA address (output param)
1183  *
1184  * This function will allcate a new page frag, map it for DMA.
1185  *
1186  * Return: allocated page frag or NULL on failure.
1187  */
1188 static void *nfp_net_rx_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1189 {
1190 	void *frag;
1191 
1192 	if (!dp->xdp_prog) {
1193 		frag = netdev_alloc_frag(dp->fl_bufsz);
1194 	} else {
1195 		struct page *page;
1196 
1197 		page = alloc_page(GFP_KERNEL);
1198 		frag = page ? page_address(page) : NULL;
1199 	}
1200 	if (!frag) {
1201 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1202 		return NULL;
1203 	}
1204 
1205 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1206 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1207 		nfp_net_free_frag(frag, dp->xdp_prog);
1208 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1209 		return NULL;
1210 	}
1211 
1212 	return frag;
1213 }
1214 
1215 static void *nfp_net_napi_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1216 {
1217 	void *frag;
1218 
1219 	if (!dp->xdp_prog) {
1220 		frag = napi_alloc_frag(dp->fl_bufsz);
1221 		if (unlikely(!frag))
1222 			return NULL;
1223 	} else {
1224 		struct page *page;
1225 
1226 		page = dev_alloc_page();
1227 		if (unlikely(!page))
1228 			return NULL;
1229 		frag = page_address(page);
1230 	}
1231 
1232 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1233 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1234 		nfp_net_free_frag(frag, dp->xdp_prog);
1235 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1236 		return NULL;
1237 	}
1238 
1239 	return frag;
1240 }
1241 
1242 /**
1243  * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1244  * @dp:		NFP Net data path struct
1245  * @rx_ring:	RX ring structure
1246  * @frag:	page fragment buffer
1247  * @dma_addr:	DMA address of skb mapping
1248  */
1249 static void nfp_net_rx_give_one(const struct nfp_net_dp *dp,
1250 				struct nfp_net_rx_ring *rx_ring,
1251 				void *frag, dma_addr_t dma_addr)
1252 {
1253 	unsigned int wr_idx;
1254 
1255 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1256 
1257 	nfp_net_dma_sync_dev_rx(dp, dma_addr);
1258 
1259 	/* Stash SKB and DMA address away */
1260 	rx_ring->rxbufs[wr_idx].frag = frag;
1261 	rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1262 
1263 	/* Fill freelist descriptor */
1264 	rx_ring->rxds[wr_idx].fld.reserved = 0;
1265 	rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1266 	nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld,
1267 			      dma_addr + dp->rx_dma_off);
1268 
1269 	rx_ring->wr_p++;
1270 	if (!(rx_ring->wr_p % NFP_NET_FL_BATCH)) {
1271 		/* Update write pointer of the freelist queue. Make
1272 		 * sure all writes are flushed before telling the hardware.
1273 		 */
1274 		wmb();
1275 		nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, NFP_NET_FL_BATCH);
1276 	}
1277 }
1278 
1279 /**
1280  * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1281  * @rx_ring:	RX ring structure
1282  *
1283  * Assumes that the device is stopped, must be idempotent.
1284  */
1285 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1286 {
1287 	unsigned int wr_idx, last_idx;
1288 
1289 	/* wr_p == rd_p means ring was never fed FL bufs.  RX rings are always
1290 	 * kept at cnt - 1 FL bufs.
1291 	 */
1292 	if (rx_ring->wr_p == 0 && rx_ring->rd_p == 0)
1293 		return;
1294 
1295 	/* Move the empty entry to the end of the list */
1296 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1297 	last_idx = rx_ring->cnt - 1;
1298 	rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1299 	rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1300 	rx_ring->rxbufs[last_idx].dma_addr = 0;
1301 	rx_ring->rxbufs[last_idx].frag = NULL;
1302 
1303 	memset(rx_ring->rxds, 0, rx_ring->size);
1304 	rx_ring->wr_p = 0;
1305 	rx_ring->rd_p = 0;
1306 }
1307 
1308 /**
1309  * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1310  * @dp:		NFP Net data path struct
1311  * @rx_ring:	RX ring to remove buffers from
1312  *
1313  * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1314  * entries.  After device is disabled nfp_net_rx_ring_reset() must be called
1315  * to restore required ring geometry.
1316  */
1317 static void
1318 nfp_net_rx_ring_bufs_free(struct nfp_net_dp *dp,
1319 			  struct nfp_net_rx_ring *rx_ring)
1320 {
1321 	unsigned int i;
1322 
1323 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1324 		/* NULL skb can only happen when initial filling of the ring
1325 		 * fails to allocate enough buffers and calls here to free
1326 		 * already allocated ones.
1327 		 */
1328 		if (!rx_ring->rxbufs[i].frag)
1329 			continue;
1330 
1331 		nfp_net_dma_unmap_rx(dp, rx_ring->rxbufs[i].dma_addr);
1332 		nfp_net_free_frag(rx_ring->rxbufs[i].frag, dp->xdp_prog);
1333 		rx_ring->rxbufs[i].dma_addr = 0;
1334 		rx_ring->rxbufs[i].frag = NULL;
1335 	}
1336 }
1337 
1338 /**
1339  * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1340  * @dp:		NFP Net data path struct
1341  * @rx_ring:	RX ring to remove buffers from
1342  */
1343 static int
1344 nfp_net_rx_ring_bufs_alloc(struct nfp_net_dp *dp,
1345 			   struct nfp_net_rx_ring *rx_ring)
1346 {
1347 	struct nfp_net_rx_buf *rxbufs;
1348 	unsigned int i;
1349 
1350 	rxbufs = rx_ring->rxbufs;
1351 
1352 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1353 		rxbufs[i].frag = nfp_net_rx_alloc_one(dp, &rxbufs[i].dma_addr);
1354 		if (!rxbufs[i].frag) {
1355 			nfp_net_rx_ring_bufs_free(dp, rx_ring);
1356 			return -ENOMEM;
1357 		}
1358 	}
1359 
1360 	return 0;
1361 }
1362 
1363 /**
1364  * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1365  * @dp:	     NFP Net data path struct
1366  * @rx_ring: RX ring to fill
1367  */
1368 static void
1369 nfp_net_rx_ring_fill_freelist(struct nfp_net_dp *dp,
1370 			      struct nfp_net_rx_ring *rx_ring)
1371 {
1372 	unsigned int i;
1373 
1374 	for (i = 0; i < rx_ring->cnt - 1; i++)
1375 		nfp_net_rx_give_one(dp, rx_ring, rx_ring->rxbufs[i].frag,
1376 				    rx_ring->rxbufs[i].dma_addr);
1377 }
1378 
1379 /**
1380  * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1381  * @flags: RX descriptor flags field in CPU byte order
1382  */
1383 static int nfp_net_rx_csum_has_errors(u16 flags)
1384 {
1385 	u16 csum_all_checked, csum_all_ok;
1386 
1387 	csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1388 	csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1389 
1390 	return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1391 }
1392 
1393 /**
1394  * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1395  * @dp:  NFP Net data path struct
1396  * @r_vec: per-ring structure
1397  * @rxd: Pointer to RX descriptor
1398  * @meta: Parsed metadata prepend
1399  * @skb: Pointer to SKB
1400  */
1401 static void nfp_net_rx_csum(struct nfp_net_dp *dp,
1402 			    struct nfp_net_r_vector *r_vec,
1403 			    struct nfp_net_rx_desc *rxd,
1404 			    struct nfp_meta_parsed *meta, struct sk_buff *skb)
1405 {
1406 	skb_checksum_none_assert(skb);
1407 
1408 	if (!(dp->netdev->features & NETIF_F_RXCSUM))
1409 		return;
1410 
1411 	if (meta->csum_type) {
1412 		skb->ip_summed = meta->csum_type;
1413 		skb->csum = meta->csum;
1414 		u64_stats_update_begin(&r_vec->rx_sync);
1415 		r_vec->hw_csum_rx_complete++;
1416 		u64_stats_update_end(&r_vec->rx_sync);
1417 		return;
1418 	}
1419 
1420 	if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1421 		u64_stats_update_begin(&r_vec->rx_sync);
1422 		r_vec->hw_csum_rx_error++;
1423 		u64_stats_update_end(&r_vec->rx_sync);
1424 		return;
1425 	}
1426 
1427 	/* Assume that the firmware will never report inner CSUM_OK unless outer
1428 	 * L4 headers were successfully parsed. FW will always report zero UDP
1429 	 * checksum as CSUM_OK.
1430 	 */
1431 	if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1432 	    rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1433 		__skb_incr_checksum_unnecessary(skb);
1434 		u64_stats_update_begin(&r_vec->rx_sync);
1435 		r_vec->hw_csum_rx_ok++;
1436 		u64_stats_update_end(&r_vec->rx_sync);
1437 	}
1438 
1439 	if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1440 	    rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1441 		__skb_incr_checksum_unnecessary(skb);
1442 		u64_stats_update_begin(&r_vec->rx_sync);
1443 		r_vec->hw_csum_rx_inner_ok++;
1444 		u64_stats_update_end(&r_vec->rx_sync);
1445 	}
1446 }
1447 
1448 static void
1449 nfp_net_set_hash(struct net_device *netdev, struct nfp_meta_parsed *meta,
1450 		 unsigned int type, __be32 *hash)
1451 {
1452 	if (!(netdev->features & NETIF_F_RXHASH))
1453 		return;
1454 
1455 	switch (type) {
1456 	case NFP_NET_RSS_IPV4:
1457 	case NFP_NET_RSS_IPV6:
1458 	case NFP_NET_RSS_IPV6_EX:
1459 		meta->hash_type = PKT_HASH_TYPE_L3;
1460 		break;
1461 	default:
1462 		meta->hash_type = PKT_HASH_TYPE_L4;
1463 		break;
1464 	}
1465 
1466 	meta->hash = get_unaligned_be32(hash);
1467 }
1468 
1469 static void
1470 nfp_net_set_hash_desc(struct net_device *netdev, struct nfp_meta_parsed *meta,
1471 		      void *data, struct nfp_net_rx_desc *rxd)
1472 {
1473 	struct nfp_net_rx_hash *rx_hash = data;
1474 
1475 	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1476 		return;
1477 
1478 	nfp_net_set_hash(netdev, meta, get_unaligned_be32(&rx_hash->hash_type),
1479 			 &rx_hash->hash);
1480 }
1481 
1482 static void *
1483 nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
1484 		   void *data, int meta_len)
1485 {
1486 	u32 meta_info;
1487 
1488 	meta_info = get_unaligned_be32(data);
1489 	data += 4;
1490 
1491 	while (meta_info) {
1492 		switch (meta_info & NFP_NET_META_FIELD_MASK) {
1493 		case NFP_NET_META_HASH:
1494 			meta_info >>= NFP_NET_META_FIELD_SIZE;
1495 			nfp_net_set_hash(netdev, meta,
1496 					 meta_info & NFP_NET_META_FIELD_MASK,
1497 					 (__be32 *)data);
1498 			data += 4;
1499 			break;
1500 		case NFP_NET_META_MARK:
1501 			meta->mark = get_unaligned_be32(data);
1502 			data += 4;
1503 			break;
1504 		case NFP_NET_META_PORTID:
1505 			meta->portid = get_unaligned_be32(data);
1506 			data += 4;
1507 			break;
1508 		case NFP_NET_META_CSUM:
1509 			meta->csum_type = CHECKSUM_COMPLETE;
1510 			meta->csum =
1511 				(__force __wsum)__get_unaligned_cpu32(data);
1512 			data += 4;
1513 			break;
1514 		default:
1515 			return NULL;
1516 		}
1517 
1518 		meta_info >>= NFP_NET_META_FIELD_SIZE;
1519 	}
1520 
1521 	return data;
1522 }
1523 
1524 static void
1525 nfp_net_rx_drop(const struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
1526 		struct nfp_net_rx_ring *rx_ring, struct nfp_net_rx_buf *rxbuf,
1527 		struct sk_buff *skb)
1528 {
1529 	u64_stats_update_begin(&r_vec->rx_sync);
1530 	r_vec->rx_drops++;
1531 	/* If we have both skb and rxbuf the replacement buffer allocation
1532 	 * must have failed, count this as an alloc failure.
1533 	 */
1534 	if (skb && rxbuf)
1535 		r_vec->rx_replace_buf_alloc_fail++;
1536 	u64_stats_update_end(&r_vec->rx_sync);
1537 
1538 	/* skb is build based on the frag, free_skb() would free the frag
1539 	 * so to be able to reuse it we need an extra ref.
1540 	 */
1541 	if (skb && rxbuf && skb->head == rxbuf->frag)
1542 		page_ref_inc(virt_to_head_page(rxbuf->frag));
1543 	if (rxbuf)
1544 		nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag, rxbuf->dma_addr);
1545 	if (skb)
1546 		dev_kfree_skb_any(skb);
1547 }
1548 
1549 static bool
1550 nfp_net_tx_xdp_buf(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring,
1551 		   struct nfp_net_tx_ring *tx_ring,
1552 		   struct nfp_net_rx_buf *rxbuf, unsigned int dma_off,
1553 		   unsigned int pkt_len, bool *completed)
1554 {
1555 	struct nfp_net_tx_buf *txbuf;
1556 	struct nfp_net_tx_desc *txd;
1557 	int wr_idx;
1558 
1559 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1560 		if (!*completed) {
1561 			nfp_net_xdp_complete(tx_ring);
1562 			*completed = true;
1563 		}
1564 
1565 		if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1566 			nfp_net_rx_drop(dp, rx_ring->r_vec, rx_ring, rxbuf,
1567 					NULL);
1568 			return false;
1569 		}
1570 	}
1571 
1572 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1573 
1574 	/* Stash the soft descriptor of the head then initialize it */
1575 	txbuf = &tx_ring->txbufs[wr_idx];
1576 
1577 	nfp_net_rx_give_one(dp, rx_ring, txbuf->frag, txbuf->dma_addr);
1578 
1579 	txbuf->frag = rxbuf->frag;
1580 	txbuf->dma_addr = rxbuf->dma_addr;
1581 	txbuf->fidx = -1;
1582 	txbuf->pkt_cnt = 1;
1583 	txbuf->real_len = pkt_len;
1584 
1585 	dma_sync_single_for_device(dp->dev, rxbuf->dma_addr + dma_off,
1586 				   pkt_len, DMA_BIDIRECTIONAL);
1587 
1588 	/* Build TX descriptor */
1589 	txd = &tx_ring->txds[wr_idx];
1590 	txd->offset_eop = PCIE_DESC_TX_EOP;
1591 	txd->dma_len = cpu_to_le16(pkt_len);
1592 	nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + dma_off);
1593 	txd->data_len = cpu_to_le16(pkt_len);
1594 
1595 	txd->flags = 0;
1596 	txd->mss = 0;
1597 	txd->lso_hdrlen = 0;
1598 
1599 	tx_ring->wr_p++;
1600 	tx_ring->wr_ptr_add++;
1601 	return true;
1602 }
1603 
1604 /**
1605  * nfp_net_rx() - receive up to @budget packets on @rx_ring
1606  * @rx_ring:   RX ring to receive from
1607  * @budget:    NAPI budget
1608  *
1609  * Note, this function is separated out from the napi poll function to
1610  * more cleanly separate packet receive code from other bookkeeping
1611  * functions performed in the napi poll function.
1612  *
1613  * Return: Number of packets received.
1614  */
1615 static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1616 {
1617 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1618 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1619 	struct nfp_net_tx_ring *tx_ring;
1620 	struct bpf_prog *xdp_prog;
1621 	bool xdp_tx_cmpl = false;
1622 	unsigned int true_bufsz;
1623 	struct sk_buff *skb;
1624 	int pkts_polled = 0;
1625 	struct xdp_buff xdp;
1626 	int idx;
1627 
1628 	rcu_read_lock();
1629 	xdp_prog = READ_ONCE(dp->xdp_prog);
1630 	true_bufsz = xdp_prog ? PAGE_SIZE : dp->fl_bufsz;
1631 	xdp.rxq = &rx_ring->xdp_rxq;
1632 	tx_ring = r_vec->xdp_ring;
1633 
1634 	while (pkts_polled < budget) {
1635 		unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1636 		struct nfp_net_rx_buf *rxbuf;
1637 		struct nfp_net_rx_desc *rxd;
1638 		struct nfp_meta_parsed meta;
1639 		struct net_device *netdev;
1640 		dma_addr_t new_dma_addr;
1641 		u32 meta_len_xdp = 0;
1642 		void *new_frag;
1643 
1644 		idx = D_IDX(rx_ring, rx_ring->rd_p);
1645 
1646 		rxd = &rx_ring->rxds[idx];
1647 		if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1648 			break;
1649 
1650 		/* Memory barrier to ensure that we won't do other reads
1651 		 * before the DD bit.
1652 		 */
1653 		dma_rmb();
1654 
1655 		memset(&meta, 0, sizeof(meta));
1656 
1657 		rx_ring->rd_p++;
1658 		pkts_polled++;
1659 
1660 		rxbuf =	&rx_ring->rxbufs[idx];
1661 		/*         < meta_len >
1662 		 *  <-- [rx_offset] -->
1663 		 *  ---------------------------------------------------------
1664 		 * | [XX] |  metadata  |             packet           | XXXX |
1665 		 *  ---------------------------------------------------------
1666 		 *         <---------------- data_len --------------->
1667 		 *
1668 		 * The rx_offset is fixed for all packets, the meta_len can vary
1669 		 * on a packet by packet basis. If rx_offset is set to zero
1670 		 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1671 		 * buffer and is immediately followed by the packet (no [XX]).
1672 		 */
1673 		meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1674 		data_len = le16_to_cpu(rxd->rxd.data_len);
1675 		pkt_len = data_len - meta_len;
1676 
1677 		pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1678 		if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1679 			pkt_off += meta_len;
1680 		else
1681 			pkt_off += dp->rx_offset;
1682 		meta_off = pkt_off - meta_len;
1683 
1684 		/* Stats update */
1685 		u64_stats_update_begin(&r_vec->rx_sync);
1686 		r_vec->rx_pkts++;
1687 		r_vec->rx_bytes += pkt_len;
1688 		u64_stats_update_end(&r_vec->rx_sync);
1689 
1690 		if (unlikely(meta_len > NFP_NET_MAX_PREPEND ||
1691 			     (dp->rx_offset && meta_len > dp->rx_offset))) {
1692 			nn_dp_warn(dp, "oversized RX packet metadata %u\n",
1693 				   meta_len);
1694 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1695 			continue;
1696 		}
1697 
1698 		nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,
1699 					data_len);
1700 
1701 		if (!dp->chained_metadata_format) {
1702 			nfp_net_set_hash_desc(dp->netdev, &meta,
1703 					      rxbuf->frag + meta_off, rxd);
1704 		} else if (meta_len) {
1705 			void *end;
1706 
1707 			end = nfp_net_parse_meta(dp->netdev, &meta,
1708 						 rxbuf->frag + meta_off,
1709 						 meta_len);
1710 			if (unlikely(end != rxbuf->frag + pkt_off)) {
1711 				nn_dp_warn(dp, "invalid RX packet metadata\n");
1712 				nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1713 						NULL);
1714 				continue;
1715 			}
1716 		}
1717 
1718 		if (xdp_prog && !meta.portid) {
1719 			void *orig_data = rxbuf->frag + pkt_off;
1720 			unsigned int dma_off;
1721 			int act;
1722 
1723 			xdp.data_hard_start = rxbuf->frag + NFP_NET_RX_BUF_HEADROOM;
1724 			xdp.data = orig_data;
1725 			xdp.data_meta = orig_data;
1726 			xdp.data_end = orig_data + pkt_len;
1727 
1728 			act = bpf_prog_run_xdp(xdp_prog, &xdp);
1729 
1730 			pkt_len = xdp.data_end - xdp.data;
1731 			pkt_off += xdp.data - orig_data;
1732 
1733 			switch (act) {
1734 			case XDP_PASS:
1735 				meta_len_xdp = xdp.data - xdp.data_meta;
1736 				break;
1737 			case XDP_TX:
1738 				dma_off = pkt_off - NFP_NET_RX_BUF_HEADROOM;
1739 				if (unlikely(!nfp_net_tx_xdp_buf(dp, rx_ring,
1740 								 tx_ring, rxbuf,
1741 								 dma_off,
1742 								 pkt_len,
1743 								 &xdp_tx_cmpl)))
1744 					trace_xdp_exception(dp->netdev,
1745 							    xdp_prog, act);
1746 				continue;
1747 			default:
1748 				bpf_warn_invalid_xdp_action(act);
1749 				/* fall through */
1750 			case XDP_ABORTED:
1751 				trace_xdp_exception(dp->netdev, xdp_prog, act);
1752 				/* fall through */
1753 			case XDP_DROP:
1754 				nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1755 						    rxbuf->dma_addr);
1756 				continue;
1757 			}
1758 		}
1759 
1760 		if (likely(!meta.portid)) {
1761 			netdev = dp->netdev;
1762 		} else if (meta.portid == NFP_META_PORT_ID_CTRL) {
1763 			struct nfp_net *nn = netdev_priv(dp->netdev);
1764 
1765 			nfp_app_ctrl_rx_raw(nn->app, rxbuf->frag + pkt_off,
1766 					    pkt_len);
1767 			nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1768 					    rxbuf->dma_addr);
1769 			continue;
1770 		} else {
1771 			struct nfp_net *nn;
1772 
1773 			nn = netdev_priv(dp->netdev);
1774 			netdev = nfp_app_repr_get(nn->app, meta.portid);
1775 			if (unlikely(!netdev)) {
1776 				nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1777 						NULL);
1778 				continue;
1779 			}
1780 			nfp_repr_inc_rx_stats(netdev, pkt_len);
1781 		}
1782 
1783 		skb = build_skb(rxbuf->frag, true_bufsz);
1784 		if (unlikely(!skb)) {
1785 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1786 			continue;
1787 		}
1788 		new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
1789 		if (unlikely(!new_frag)) {
1790 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
1791 			continue;
1792 		}
1793 
1794 		nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
1795 
1796 		nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
1797 
1798 		skb_reserve(skb, pkt_off);
1799 		skb_put(skb, pkt_len);
1800 
1801 		skb->mark = meta.mark;
1802 		skb_set_hash(skb, meta.hash, meta.hash_type);
1803 
1804 		skb_record_rx_queue(skb, rx_ring->idx);
1805 		skb->protocol = eth_type_trans(skb, netdev);
1806 
1807 		nfp_net_rx_csum(dp, r_vec, rxd, &meta, skb);
1808 
1809 		if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
1810 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
1811 					       le16_to_cpu(rxd->rxd.vlan));
1812 		if (meta_len_xdp)
1813 			skb_metadata_set(skb, meta_len_xdp);
1814 
1815 		napi_gro_receive(&rx_ring->r_vec->napi, skb);
1816 	}
1817 
1818 	if (xdp_prog) {
1819 		if (tx_ring->wr_ptr_add)
1820 			nfp_net_tx_xmit_more_flush(tx_ring);
1821 		else if (unlikely(tx_ring->wr_p != tx_ring->rd_p) &&
1822 			 !xdp_tx_cmpl)
1823 			if (!nfp_net_xdp_complete(tx_ring))
1824 				pkts_polled = budget;
1825 	}
1826 	rcu_read_unlock();
1827 
1828 	return pkts_polled;
1829 }
1830 
1831 /**
1832  * nfp_net_poll() - napi poll function
1833  * @napi:    NAPI structure
1834  * @budget:  NAPI budget
1835  *
1836  * Return: number of packets polled.
1837  */
1838 static int nfp_net_poll(struct napi_struct *napi, int budget)
1839 {
1840 	struct nfp_net_r_vector *r_vec =
1841 		container_of(napi, struct nfp_net_r_vector, napi);
1842 	unsigned int pkts_polled = 0;
1843 
1844 	if (r_vec->tx_ring)
1845 		nfp_net_tx_complete(r_vec->tx_ring, budget);
1846 	if (r_vec->rx_ring)
1847 		pkts_polled = nfp_net_rx(r_vec->rx_ring, budget);
1848 
1849 	if (pkts_polled < budget)
1850 		if (napi_complete_done(napi, pkts_polled))
1851 			nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
1852 
1853 	return pkts_polled;
1854 }
1855 
1856 /* Control device data path
1857  */
1858 
1859 static bool
1860 nfp_ctrl_tx_one(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
1861 		struct sk_buff *skb, bool old)
1862 {
1863 	unsigned int real_len = skb->len, meta_len = 0;
1864 	struct nfp_net_tx_ring *tx_ring;
1865 	struct nfp_net_tx_buf *txbuf;
1866 	struct nfp_net_tx_desc *txd;
1867 	struct nfp_net_dp *dp;
1868 	dma_addr_t dma_addr;
1869 	int wr_idx;
1870 
1871 	dp = &r_vec->nfp_net->dp;
1872 	tx_ring = r_vec->tx_ring;
1873 
1874 	if (WARN_ON_ONCE(skb_shinfo(skb)->nr_frags)) {
1875 		nn_dp_warn(dp, "Driver's CTRL TX does not implement gather\n");
1876 		goto err_free;
1877 	}
1878 
1879 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1880 		u64_stats_update_begin(&r_vec->tx_sync);
1881 		r_vec->tx_busy++;
1882 		u64_stats_update_end(&r_vec->tx_sync);
1883 		if (!old)
1884 			__skb_queue_tail(&r_vec->queue, skb);
1885 		else
1886 			__skb_queue_head(&r_vec->queue, skb);
1887 		return true;
1888 	}
1889 
1890 	if (nfp_app_ctrl_has_meta(nn->app)) {
1891 		if (unlikely(skb_headroom(skb) < 8)) {
1892 			nn_dp_warn(dp, "CTRL TX on skb without headroom\n");
1893 			goto err_free;
1894 		}
1895 		meta_len = 8;
1896 		put_unaligned_be32(NFP_META_PORT_ID_CTRL, skb_push(skb, 4));
1897 		put_unaligned_be32(NFP_NET_META_PORTID, skb_push(skb, 4));
1898 	}
1899 
1900 	/* Start with the head skbuf */
1901 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
1902 				  DMA_TO_DEVICE);
1903 	if (dma_mapping_error(dp->dev, dma_addr))
1904 		goto err_dma_warn;
1905 
1906 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1907 
1908 	/* Stash the soft descriptor of the head then initialize it */
1909 	txbuf = &tx_ring->txbufs[wr_idx];
1910 	txbuf->skb = skb;
1911 	txbuf->dma_addr = dma_addr;
1912 	txbuf->fidx = -1;
1913 	txbuf->pkt_cnt = 1;
1914 	txbuf->real_len = real_len;
1915 
1916 	/* Build TX descriptor */
1917 	txd = &tx_ring->txds[wr_idx];
1918 	txd->offset_eop = meta_len | PCIE_DESC_TX_EOP;
1919 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
1920 	nfp_desc_set_dma_addr(txd, dma_addr);
1921 	txd->data_len = cpu_to_le16(skb->len);
1922 
1923 	txd->flags = 0;
1924 	txd->mss = 0;
1925 	txd->lso_hdrlen = 0;
1926 
1927 	tx_ring->wr_p++;
1928 	tx_ring->wr_ptr_add++;
1929 	nfp_net_tx_xmit_more_flush(tx_ring);
1930 
1931 	return false;
1932 
1933 err_dma_warn:
1934 	nn_dp_warn(dp, "Failed to DMA map TX CTRL buffer\n");
1935 err_free:
1936 	u64_stats_update_begin(&r_vec->tx_sync);
1937 	r_vec->tx_errors++;
1938 	u64_stats_update_end(&r_vec->tx_sync);
1939 	dev_kfree_skb_any(skb);
1940 	return false;
1941 }
1942 
1943 bool __nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
1944 {
1945 	struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
1946 
1947 	return nfp_ctrl_tx_one(nn, r_vec, skb, false);
1948 }
1949 
1950 bool nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
1951 {
1952 	struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
1953 	bool ret;
1954 
1955 	spin_lock_bh(&r_vec->lock);
1956 	ret = nfp_ctrl_tx_one(nn, r_vec, skb, false);
1957 	spin_unlock_bh(&r_vec->lock);
1958 
1959 	return ret;
1960 }
1961 
1962 static void __nfp_ctrl_tx_queued(struct nfp_net_r_vector *r_vec)
1963 {
1964 	struct sk_buff *skb;
1965 
1966 	while ((skb = __skb_dequeue(&r_vec->queue)))
1967 		if (nfp_ctrl_tx_one(r_vec->nfp_net, r_vec, skb, true))
1968 			return;
1969 }
1970 
1971 static bool
1972 nfp_ctrl_meta_ok(struct nfp_net *nn, void *data, unsigned int meta_len)
1973 {
1974 	u32 meta_type, meta_tag;
1975 
1976 	if (!nfp_app_ctrl_has_meta(nn->app))
1977 		return !meta_len;
1978 
1979 	if (meta_len != 8)
1980 		return false;
1981 
1982 	meta_type = get_unaligned_be32(data);
1983 	meta_tag = get_unaligned_be32(data + 4);
1984 
1985 	return (meta_type == NFP_NET_META_PORTID &&
1986 		meta_tag == NFP_META_PORT_ID_CTRL);
1987 }
1988 
1989 static bool
1990 nfp_ctrl_rx_one(struct nfp_net *nn, struct nfp_net_dp *dp,
1991 		struct nfp_net_r_vector *r_vec, struct nfp_net_rx_ring *rx_ring)
1992 {
1993 	unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1994 	struct nfp_net_rx_buf *rxbuf;
1995 	struct nfp_net_rx_desc *rxd;
1996 	dma_addr_t new_dma_addr;
1997 	struct sk_buff *skb;
1998 	void *new_frag;
1999 	int idx;
2000 
2001 	idx = D_IDX(rx_ring, rx_ring->rd_p);
2002 
2003 	rxd = &rx_ring->rxds[idx];
2004 	if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
2005 		return false;
2006 
2007 	/* Memory barrier to ensure that we won't do other reads
2008 	 * before the DD bit.
2009 	 */
2010 	dma_rmb();
2011 
2012 	rx_ring->rd_p++;
2013 
2014 	rxbuf =	&rx_ring->rxbufs[idx];
2015 	meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
2016 	data_len = le16_to_cpu(rxd->rxd.data_len);
2017 	pkt_len = data_len - meta_len;
2018 
2019 	pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
2020 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
2021 		pkt_off += meta_len;
2022 	else
2023 		pkt_off += dp->rx_offset;
2024 	meta_off = pkt_off - meta_len;
2025 
2026 	/* Stats update */
2027 	u64_stats_update_begin(&r_vec->rx_sync);
2028 	r_vec->rx_pkts++;
2029 	r_vec->rx_bytes += pkt_len;
2030 	u64_stats_update_end(&r_vec->rx_sync);
2031 
2032 	nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,	data_len);
2033 
2034 	if (unlikely(!nfp_ctrl_meta_ok(nn, rxbuf->frag + meta_off, meta_len))) {
2035 		nn_dp_warn(dp, "incorrect metadata for ctrl packet (%d)\n",
2036 			   meta_len);
2037 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2038 		return true;
2039 	}
2040 
2041 	skb = build_skb(rxbuf->frag, dp->fl_bufsz);
2042 	if (unlikely(!skb)) {
2043 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2044 		return true;
2045 	}
2046 	new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
2047 	if (unlikely(!new_frag)) {
2048 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
2049 		return true;
2050 	}
2051 
2052 	nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
2053 
2054 	nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
2055 
2056 	skb_reserve(skb, pkt_off);
2057 	skb_put(skb, pkt_len);
2058 
2059 	nfp_app_ctrl_rx(nn->app, skb);
2060 
2061 	return true;
2062 }
2063 
2064 static void nfp_ctrl_rx(struct nfp_net_r_vector *r_vec)
2065 {
2066 	struct nfp_net_rx_ring *rx_ring = r_vec->rx_ring;
2067 	struct nfp_net *nn = r_vec->nfp_net;
2068 	struct nfp_net_dp *dp = &nn->dp;
2069 
2070 	while (nfp_ctrl_rx_one(nn, dp, r_vec, rx_ring))
2071 		continue;
2072 }
2073 
2074 static void nfp_ctrl_poll(unsigned long arg)
2075 {
2076 	struct nfp_net_r_vector *r_vec = (void *)arg;
2077 
2078 	spin_lock_bh(&r_vec->lock);
2079 	nfp_net_tx_complete(r_vec->tx_ring, 0);
2080 	__nfp_ctrl_tx_queued(r_vec);
2081 	spin_unlock_bh(&r_vec->lock);
2082 
2083 	nfp_ctrl_rx(r_vec);
2084 
2085 	nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2086 }
2087 
2088 /* Setup and Configuration
2089  */
2090 
2091 /**
2092  * nfp_net_vecs_init() - Assign IRQs and setup rvecs.
2093  * @nn:		NFP Network structure
2094  */
2095 static void nfp_net_vecs_init(struct nfp_net *nn)
2096 {
2097 	struct nfp_net_r_vector *r_vec;
2098 	int r;
2099 
2100 	nn->lsc_handler = nfp_net_irq_lsc;
2101 	nn->exn_handler = nfp_net_irq_exn;
2102 
2103 	for (r = 0; r < nn->max_r_vecs; r++) {
2104 		struct msix_entry *entry;
2105 
2106 		entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r];
2107 
2108 		r_vec = &nn->r_vecs[r];
2109 		r_vec->nfp_net = nn;
2110 		r_vec->irq_entry = entry->entry;
2111 		r_vec->irq_vector = entry->vector;
2112 
2113 		if (nn->dp.netdev) {
2114 			r_vec->handler = nfp_net_irq_rxtx;
2115 		} else {
2116 			r_vec->handler = nfp_ctrl_irq_rxtx;
2117 
2118 			__skb_queue_head_init(&r_vec->queue);
2119 			spin_lock_init(&r_vec->lock);
2120 			tasklet_init(&r_vec->tasklet, nfp_ctrl_poll,
2121 				     (unsigned long)r_vec);
2122 			tasklet_disable(&r_vec->tasklet);
2123 		}
2124 
2125 		cpumask_set_cpu(r, &r_vec->affinity_mask);
2126 	}
2127 }
2128 
2129 /**
2130  * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
2131  * @tx_ring:   TX ring to free
2132  */
2133 static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
2134 {
2135 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2136 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2137 
2138 	kvfree(tx_ring->txbufs);
2139 
2140 	if (tx_ring->txds)
2141 		dma_free_coherent(dp->dev, tx_ring->size,
2142 				  tx_ring->txds, tx_ring->dma);
2143 
2144 	tx_ring->cnt = 0;
2145 	tx_ring->txbufs = NULL;
2146 	tx_ring->txds = NULL;
2147 	tx_ring->dma = 0;
2148 	tx_ring->size = 0;
2149 }
2150 
2151 /**
2152  * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
2153  * @dp:        NFP Net data path struct
2154  * @tx_ring:   TX Ring structure to allocate
2155  *
2156  * Return: 0 on success, negative errno otherwise.
2157  */
2158 static int
2159 nfp_net_tx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
2160 {
2161 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2162 
2163 	tx_ring->cnt = dp->txd_cnt;
2164 
2165 	tx_ring->size = array_size(tx_ring->cnt, sizeof(*tx_ring->txds));
2166 	tx_ring->txds = dma_zalloc_coherent(dp->dev, tx_ring->size,
2167 					    &tx_ring->dma, GFP_KERNEL);
2168 	if (!tx_ring->txds)
2169 		goto err_alloc;
2170 
2171 	tx_ring->txbufs = kvcalloc(tx_ring->cnt, sizeof(*tx_ring->txbufs),
2172 				   GFP_KERNEL);
2173 	if (!tx_ring->txbufs)
2174 		goto err_alloc;
2175 
2176 	if (!tx_ring->is_xdp && dp->netdev)
2177 		netif_set_xps_queue(dp->netdev, &r_vec->affinity_mask,
2178 				    tx_ring->idx);
2179 
2180 	return 0;
2181 
2182 err_alloc:
2183 	nfp_net_tx_ring_free(tx_ring);
2184 	return -ENOMEM;
2185 }
2186 
2187 static void
2188 nfp_net_tx_ring_bufs_free(struct nfp_net_dp *dp,
2189 			  struct nfp_net_tx_ring *tx_ring)
2190 {
2191 	unsigned int i;
2192 
2193 	if (!tx_ring->is_xdp)
2194 		return;
2195 
2196 	for (i = 0; i < tx_ring->cnt; i++) {
2197 		if (!tx_ring->txbufs[i].frag)
2198 			return;
2199 
2200 		nfp_net_dma_unmap_rx(dp, tx_ring->txbufs[i].dma_addr);
2201 		__free_page(virt_to_page(tx_ring->txbufs[i].frag));
2202 	}
2203 }
2204 
2205 static int
2206 nfp_net_tx_ring_bufs_alloc(struct nfp_net_dp *dp,
2207 			   struct nfp_net_tx_ring *tx_ring)
2208 {
2209 	struct nfp_net_tx_buf *txbufs = tx_ring->txbufs;
2210 	unsigned int i;
2211 
2212 	if (!tx_ring->is_xdp)
2213 		return 0;
2214 
2215 	for (i = 0; i < tx_ring->cnt; i++) {
2216 		txbufs[i].frag = nfp_net_rx_alloc_one(dp, &txbufs[i].dma_addr);
2217 		if (!txbufs[i].frag) {
2218 			nfp_net_tx_ring_bufs_free(dp, tx_ring);
2219 			return -ENOMEM;
2220 		}
2221 	}
2222 
2223 	return 0;
2224 }
2225 
2226 static int nfp_net_tx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2227 {
2228 	unsigned int r;
2229 
2230 	dp->tx_rings = kcalloc(dp->num_tx_rings, sizeof(*dp->tx_rings),
2231 			       GFP_KERNEL);
2232 	if (!dp->tx_rings)
2233 		return -ENOMEM;
2234 
2235 	for (r = 0; r < dp->num_tx_rings; r++) {
2236 		int bias = 0;
2237 
2238 		if (r >= dp->num_stack_tx_rings)
2239 			bias = dp->num_stack_tx_rings;
2240 
2241 		nfp_net_tx_ring_init(&dp->tx_rings[r], &nn->r_vecs[r - bias],
2242 				     r, bias);
2243 
2244 		if (nfp_net_tx_ring_alloc(dp, &dp->tx_rings[r]))
2245 			goto err_free_prev;
2246 
2247 		if (nfp_net_tx_ring_bufs_alloc(dp, &dp->tx_rings[r]))
2248 			goto err_free_ring;
2249 	}
2250 
2251 	return 0;
2252 
2253 err_free_prev:
2254 	while (r--) {
2255 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2256 err_free_ring:
2257 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2258 	}
2259 	kfree(dp->tx_rings);
2260 	return -ENOMEM;
2261 }
2262 
2263 static void nfp_net_tx_rings_free(struct nfp_net_dp *dp)
2264 {
2265 	unsigned int r;
2266 
2267 	for (r = 0; r < dp->num_tx_rings; r++) {
2268 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2269 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2270 	}
2271 
2272 	kfree(dp->tx_rings);
2273 }
2274 
2275 /**
2276  * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
2277  * @rx_ring:  RX ring to free
2278  */
2279 static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
2280 {
2281 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
2282 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2283 
2284 	if (dp->netdev)
2285 		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
2286 	kvfree(rx_ring->rxbufs);
2287 
2288 	if (rx_ring->rxds)
2289 		dma_free_coherent(dp->dev, rx_ring->size,
2290 				  rx_ring->rxds, rx_ring->dma);
2291 
2292 	rx_ring->cnt = 0;
2293 	rx_ring->rxbufs = NULL;
2294 	rx_ring->rxds = NULL;
2295 	rx_ring->dma = 0;
2296 	rx_ring->size = 0;
2297 }
2298 
2299 /**
2300  * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
2301  * @dp:	      NFP Net data path struct
2302  * @rx_ring:  RX ring to allocate
2303  *
2304  * Return: 0 on success, negative errno otherwise.
2305  */
2306 static int
2307 nfp_net_rx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring)
2308 {
2309 	int err;
2310 
2311 	if (dp->netdev) {
2312 		err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, dp->netdev,
2313 				       rx_ring->idx);
2314 		if (err < 0)
2315 			return err;
2316 	}
2317 
2318 	rx_ring->cnt = dp->rxd_cnt;
2319 	rx_ring->size = array_size(rx_ring->cnt, sizeof(*rx_ring->rxds));
2320 	rx_ring->rxds = dma_zalloc_coherent(dp->dev, rx_ring->size,
2321 					    &rx_ring->dma, GFP_KERNEL);
2322 	if (!rx_ring->rxds)
2323 		goto err_alloc;
2324 
2325 	rx_ring->rxbufs = kvcalloc(rx_ring->cnt, sizeof(*rx_ring->rxbufs),
2326 				   GFP_KERNEL);
2327 	if (!rx_ring->rxbufs)
2328 		goto err_alloc;
2329 
2330 	return 0;
2331 
2332 err_alloc:
2333 	nfp_net_rx_ring_free(rx_ring);
2334 	return -ENOMEM;
2335 }
2336 
2337 static int nfp_net_rx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2338 {
2339 	unsigned int r;
2340 
2341 	dp->rx_rings = kcalloc(dp->num_rx_rings, sizeof(*dp->rx_rings),
2342 			       GFP_KERNEL);
2343 	if (!dp->rx_rings)
2344 		return -ENOMEM;
2345 
2346 	for (r = 0; r < dp->num_rx_rings; r++) {
2347 		nfp_net_rx_ring_init(&dp->rx_rings[r], &nn->r_vecs[r], r);
2348 
2349 		if (nfp_net_rx_ring_alloc(dp, &dp->rx_rings[r]))
2350 			goto err_free_prev;
2351 
2352 		if (nfp_net_rx_ring_bufs_alloc(dp, &dp->rx_rings[r]))
2353 			goto err_free_ring;
2354 	}
2355 
2356 	return 0;
2357 
2358 err_free_prev:
2359 	while (r--) {
2360 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2361 err_free_ring:
2362 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2363 	}
2364 	kfree(dp->rx_rings);
2365 	return -ENOMEM;
2366 }
2367 
2368 static void nfp_net_rx_rings_free(struct nfp_net_dp *dp)
2369 {
2370 	unsigned int r;
2371 
2372 	for (r = 0; r < dp->num_rx_rings; r++) {
2373 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2374 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2375 	}
2376 
2377 	kfree(dp->rx_rings);
2378 }
2379 
2380 static void
2381 nfp_net_vector_assign_rings(struct nfp_net_dp *dp,
2382 			    struct nfp_net_r_vector *r_vec, int idx)
2383 {
2384 	r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL;
2385 	r_vec->tx_ring =
2386 		idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL;
2387 
2388 	r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ?
2389 		&dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL;
2390 }
2391 
2392 static int
2393 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2394 		       int idx)
2395 {
2396 	int err;
2397 
2398 	/* Setup NAPI */
2399 	if (nn->dp.netdev)
2400 		netif_napi_add(nn->dp.netdev, &r_vec->napi,
2401 			       nfp_net_poll, NAPI_POLL_WEIGHT);
2402 	else
2403 		tasklet_enable(&r_vec->tasklet);
2404 
2405 	snprintf(r_vec->name, sizeof(r_vec->name),
2406 		 "%s-rxtx-%d", nfp_net_name(nn), idx);
2407 	err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name,
2408 			  r_vec);
2409 	if (err) {
2410 		if (nn->dp.netdev)
2411 			netif_napi_del(&r_vec->napi);
2412 		else
2413 			tasklet_disable(&r_vec->tasklet);
2414 
2415 		nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector);
2416 		return err;
2417 	}
2418 	disable_irq(r_vec->irq_vector);
2419 
2420 	irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask);
2421 
2422 	nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector,
2423 	       r_vec->irq_entry);
2424 
2425 	return 0;
2426 }
2427 
2428 static void
2429 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
2430 {
2431 	irq_set_affinity_hint(r_vec->irq_vector, NULL);
2432 	if (nn->dp.netdev)
2433 		netif_napi_del(&r_vec->napi);
2434 	else
2435 		tasklet_disable(&r_vec->tasklet);
2436 
2437 	free_irq(r_vec->irq_vector, r_vec);
2438 }
2439 
2440 /**
2441  * nfp_net_rss_write_itbl() - Write RSS indirection table to device
2442  * @nn:      NFP Net device to reconfigure
2443  */
2444 void nfp_net_rss_write_itbl(struct nfp_net *nn)
2445 {
2446 	int i;
2447 
2448 	for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
2449 		nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
2450 			  get_unaligned_le32(nn->rss_itbl + i));
2451 }
2452 
2453 /**
2454  * nfp_net_rss_write_key() - Write RSS hash key to device
2455  * @nn:      NFP Net device to reconfigure
2456  */
2457 void nfp_net_rss_write_key(struct nfp_net *nn)
2458 {
2459 	int i;
2460 
2461 	for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4)
2462 		nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
2463 			  get_unaligned_le32(nn->rss_key + i));
2464 }
2465 
2466 /**
2467  * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
2468  * @nn:      NFP Net device to reconfigure
2469  */
2470 void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
2471 {
2472 	u8 i;
2473 	u32 factor;
2474 	u32 value;
2475 
2476 	/* Compute factor used to convert coalesce '_usecs' parameters to
2477 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
2478 	 * count.
2479 	 */
2480 	factor = nn->tlv_caps.me_freq_mhz / 16;
2481 
2482 	/* copy RX interrupt coalesce parameters */
2483 	value = (nn->rx_coalesce_max_frames << 16) |
2484 		(factor * nn->rx_coalesce_usecs);
2485 	for (i = 0; i < nn->dp.num_rx_rings; i++)
2486 		nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
2487 
2488 	/* copy TX interrupt coalesce parameters */
2489 	value = (nn->tx_coalesce_max_frames << 16) |
2490 		(factor * nn->tx_coalesce_usecs);
2491 	for (i = 0; i < nn->dp.num_tx_rings; i++)
2492 		nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
2493 }
2494 
2495 /**
2496  * nfp_net_write_mac_addr() - Write mac address to the device control BAR
2497  * @nn:      NFP Net device to reconfigure
2498  * @addr:    MAC address to write
2499  *
2500  * Writes the MAC address from the netdev to the device control BAR.  Does not
2501  * perform the required reconfig.  We do a bit of byte swapping dance because
2502  * firmware is LE.
2503  */
2504 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr)
2505 {
2506 	nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr));
2507 	nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4));
2508 }
2509 
2510 static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
2511 {
2512 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
2513 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
2514 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
2515 
2516 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
2517 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
2518 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
2519 }
2520 
2521 /**
2522  * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
2523  * @nn:      NFP Net device to reconfigure
2524  *
2525  * Warning: must be fully idempotent.
2526  */
2527 static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
2528 {
2529 	u32 new_ctrl, update;
2530 	unsigned int r;
2531 	int err;
2532 
2533 	new_ctrl = nn->dp.ctrl;
2534 	new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
2535 	update = NFP_NET_CFG_UPDATE_GEN;
2536 	update |= NFP_NET_CFG_UPDATE_MSIX;
2537 	update |= NFP_NET_CFG_UPDATE_RING;
2538 
2539 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2540 		new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
2541 
2542 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2543 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2544 
2545 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2546 	err = nfp_net_reconfig(nn, update);
2547 	if (err)
2548 		nn_err(nn, "Could not disable device: %d\n", err);
2549 
2550 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2551 		nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]);
2552 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2553 		nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]);
2554 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2555 		nfp_net_vec_clear_ring_data(nn, r);
2556 
2557 	nn->dp.ctrl = new_ctrl;
2558 }
2559 
2560 static void
2561 nfp_net_rx_ring_hw_cfg_write(struct nfp_net *nn,
2562 			     struct nfp_net_rx_ring *rx_ring, unsigned int idx)
2563 {
2564 	/* Write the DMA address, size and MSI-X info to the device */
2565 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), rx_ring->dma);
2566 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(rx_ring->cnt));
2567 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), rx_ring->r_vec->irq_entry);
2568 }
2569 
2570 static void
2571 nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
2572 			     struct nfp_net_tx_ring *tx_ring, unsigned int idx)
2573 {
2574 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), tx_ring->dma);
2575 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(tx_ring->cnt));
2576 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_entry);
2577 }
2578 
2579 /**
2580  * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2581  * @nn:      NFP Net device to reconfigure
2582  */
2583 static int nfp_net_set_config_and_enable(struct nfp_net *nn)
2584 {
2585 	u32 bufsz, new_ctrl, update = 0;
2586 	unsigned int r;
2587 	int err;
2588 
2589 	new_ctrl = nn->dp.ctrl;
2590 
2591 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) {
2592 		nfp_net_rss_write_key(nn);
2593 		nfp_net_rss_write_itbl(nn);
2594 		nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
2595 		update |= NFP_NET_CFG_UPDATE_RSS;
2596 	}
2597 
2598 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) {
2599 		nfp_net_coalesce_write_cfg(nn);
2600 		update |= NFP_NET_CFG_UPDATE_IRQMOD;
2601 	}
2602 
2603 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2604 		nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r);
2605 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2606 		nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r);
2607 
2608 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->dp.num_tx_rings == 64 ?
2609 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_tx_rings) - 1);
2610 
2611 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->dp.num_rx_rings == 64 ?
2612 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_rx_rings) - 1);
2613 
2614 	if (nn->dp.netdev)
2615 		nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
2616 
2617 	nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu);
2618 
2619 	bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA;
2620 	nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz);
2621 
2622 	/* Enable device */
2623 	new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2624 	update |= NFP_NET_CFG_UPDATE_GEN;
2625 	update |= NFP_NET_CFG_UPDATE_MSIX;
2626 	update |= NFP_NET_CFG_UPDATE_RING;
2627 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2628 		new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2629 
2630 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2631 	err = nfp_net_reconfig(nn, update);
2632 	if (err) {
2633 		nfp_net_clear_config_and_disable(nn);
2634 		return err;
2635 	}
2636 
2637 	nn->dp.ctrl = new_ctrl;
2638 
2639 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2640 		nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]);
2641 
2642 	/* Since reconfiguration requests while NFP is down are ignored we
2643 	 * have to wipe the entire VXLAN configuration and reinitialize it.
2644 	 */
2645 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN) {
2646 		memset(&nn->vxlan_ports, 0, sizeof(nn->vxlan_ports));
2647 		memset(&nn->vxlan_usecnt, 0, sizeof(nn->vxlan_usecnt));
2648 		udp_tunnel_get_rx_info(nn->dp.netdev);
2649 	}
2650 
2651 	return 0;
2652 }
2653 
2654 /**
2655  * nfp_net_close_stack() - Quiesce the stack (part of close)
2656  * @nn:	     NFP Net device to reconfigure
2657  */
2658 static void nfp_net_close_stack(struct nfp_net *nn)
2659 {
2660 	unsigned int r;
2661 
2662 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2663 	netif_carrier_off(nn->dp.netdev);
2664 	nn->link_up = false;
2665 
2666 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2667 		disable_irq(nn->r_vecs[r].irq_vector);
2668 		napi_disable(&nn->r_vecs[r].napi);
2669 	}
2670 
2671 	netif_tx_disable(nn->dp.netdev);
2672 }
2673 
2674 /**
2675  * nfp_net_close_free_all() - Free all runtime resources
2676  * @nn:      NFP Net device to reconfigure
2677  */
2678 static void nfp_net_close_free_all(struct nfp_net *nn)
2679 {
2680 	unsigned int r;
2681 
2682 	nfp_net_tx_rings_free(&nn->dp);
2683 	nfp_net_rx_rings_free(&nn->dp);
2684 
2685 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2686 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2687 
2688 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2689 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2690 }
2691 
2692 /**
2693  * nfp_net_netdev_close() - Called when the device is downed
2694  * @netdev:      netdev structure
2695  */
2696 static int nfp_net_netdev_close(struct net_device *netdev)
2697 {
2698 	struct nfp_net *nn = netdev_priv(netdev);
2699 
2700 	/* Step 1: Disable RX and TX rings from the Linux kernel perspective
2701 	 */
2702 	nfp_net_close_stack(nn);
2703 
2704 	/* Step 2: Tell NFP
2705 	 */
2706 	nfp_net_clear_config_and_disable(nn);
2707 	nfp_port_configure(netdev, false);
2708 
2709 	/* Step 3: Free resources
2710 	 */
2711 	nfp_net_close_free_all(nn);
2712 
2713 	nn_dbg(nn, "%s down", netdev->name);
2714 	return 0;
2715 }
2716 
2717 void nfp_ctrl_close(struct nfp_net *nn)
2718 {
2719 	int r;
2720 
2721 	rtnl_lock();
2722 
2723 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2724 		disable_irq(nn->r_vecs[r].irq_vector);
2725 		tasklet_disable(&nn->r_vecs[r].tasklet);
2726 	}
2727 
2728 	nfp_net_clear_config_and_disable(nn);
2729 
2730 	nfp_net_close_free_all(nn);
2731 
2732 	rtnl_unlock();
2733 }
2734 
2735 /**
2736  * nfp_net_open_stack() - Start the device from stack's perspective
2737  * @nn:      NFP Net device to reconfigure
2738  */
2739 static void nfp_net_open_stack(struct nfp_net *nn)
2740 {
2741 	unsigned int r;
2742 
2743 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2744 		napi_enable(&nn->r_vecs[r].napi);
2745 		enable_irq(nn->r_vecs[r].irq_vector);
2746 	}
2747 
2748 	netif_tx_wake_all_queues(nn->dp.netdev);
2749 
2750 	enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2751 	nfp_net_read_link_status(nn);
2752 }
2753 
2754 static int nfp_net_open_alloc_all(struct nfp_net *nn)
2755 {
2756 	int err, r;
2757 
2758 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2759 				      nn->exn_name, sizeof(nn->exn_name),
2760 				      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2761 	if (err)
2762 		return err;
2763 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2764 				      nn->lsc_name, sizeof(nn->lsc_name),
2765 				      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2766 	if (err)
2767 		goto err_free_exn;
2768 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2769 
2770 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2771 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2772 		if (err)
2773 			goto err_cleanup_vec_p;
2774 	}
2775 
2776 	err = nfp_net_rx_rings_prepare(nn, &nn->dp);
2777 	if (err)
2778 		goto err_cleanup_vec;
2779 
2780 	err = nfp_net_tx_rings_prepare(nn, &nn->dp);
2781 	if (err)
2782 		goto err_free_rx_rings;
2783 
2784 	for (r = 0; r < nn->max_r_vecs; r++)
2785 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2786 
2787 	return 0;
2788 
2789 err_free_rx_rings:
2790 	nfp_net_rx_rings_free(&nn->dp);
2791 err_cleanup_vec:
2792 	r = nn->dp.num_r_vecs;
2793 err_cleanup_vec_p:
2794 	while (r--)
2795 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2796 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2797 err_free_exn:
2798 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2799 	return err;
2800 }
2801 
2802 static int nfp_net_netdev_open(struct net_device *netdev)
2803 {
2804 	struct nfp_net *nn = netdev_priv(netdev);
2805 	int err;
2806 
2807 	/* Step 1: Allocate resources for rings and the like
2808 	 * - Request interrupts
2809 	 * - Allocate RX and TX ring resources
2810 	 * - Setup initial RSS table
2811 	 */
2812 	err = nfp_net_open_alloc_all(nn);
2813 	if (err)
2814 		return err;
2815 
2816 	err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
2817 	if (err)
2818 		goto err_free_all;
2819 
2820 	err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
2821 	if (err)
2822 		goto err_free_all;
2823 
2824 	/* Step 2: Configure the NFP
2825 	 * - Ifup the physical interface if it exists
2826 	 * - Enable rings from 0 to tx_rings/rx_rings - 1.
2827 	 * - Write MAC address (in case it changed)
2828 	 * - Set the MTU
2829 	 * - Set the Freelist buffer size
2830 	 * - Enable the FW
2831 	 */
2832 	err = nfp_port_configure(netdev, true);
2833 	if (err)
2834 		goto err_free_all;
2835 
2836 	err = nfp_net_set_config_and_enable(nn);
2837 	if (err)
2838 		goto err_port_disable;
2839 
2840 	/* Step 3: Enable for kernel
2841 	 * - put some freelist descriptors on each RX ring
2842 	 * - enable NAPI on each ring
2843 	 * - enable all TX queues
2844 	 * - set link state
2845 	 */
2846 	nfp_net_open_stack(nn);
2847 
2848 	return 0;
2849 
2850 err_port_disable:
2851 	nfp_port_configure(netdev, false);
2852 err_free_all:
2853 	nfp_net_close_free_all(nn);
2854 	return err;
2855 }
2856 
2857 int nfp_ctrl_open(struct nfp_net *nn)
2858 {
2859 	int err, r;
2860 
2861 	/* ring dumping depends on vNICs being opened/closed under rtnl */
2862 	rtnl_lock();
2863 
2864 	err = nfp_net_open_alloc_all(nn);
2865 	if (err)
2866 		goto err_unlock;
2867 
2868 	err = nfp_net_set_config_and_enable(nn);
2869 	if (err)
2870 		goto err_free_all;
2871 
2872 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2873 		enable_irq(nn->r_vecs[r].irq_vector);
2874 
2875 	rtnl_unlock();
2876 
2877 	return 0;
2878 
2879 err_free_all:
2880 	nfp_net_close_free_all(nn);
2881 err_unlock:
2882 	rtnl_unlock();
2883 	return err;
2884 }
2885 
2886 static void nfp_net_set_rx_mode(struct net_device *netdev)
2887 {
2888 	struct nfp_net *nn = netdev_priv(netdev);
2889 	u32 new_ctrl;
2890 
2891 	new_ctrl = nn->dp.ctrl;
2892 
2893 	if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI)
2894 		new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC;
2895 	else
2896 		new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC;
2897 
2898 	if (netdev->flags & IFF_PROMISC) {
2899 		if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
2900 			new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
2901 		else
2902 			nn_warn(nn, "FW does not support promiscuous mode\n");
2903 	} else {
2904 		new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
2905 	}
2906 
2907 	if (new_ctrl == nn->dp.ctrl)
2908 		return;
2909 
2910 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2911 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
2912 
2913 	nn->dp.ctrl = new_ctrl;
2914 }
2915 
2916 static void nfp_net_rss_init_itbl(struct nfp_net *nn)
2917 {
2918 	int i;
2919 
2920 	for (i = 0; i < sizeof(nn->rss_itbl); i++)
2921 		nn->rss_itbl[i] =
2922 			ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
2923 }
2924 
2925 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
2926 {
2927 	struct nfp_net_dp new_dp = *dp;
2928 
2929 	*dp = nn->dp;
2930 	nn->dp = new_dp;
2931 
2932 	nn->dp.netdev->mtu = new_dp.mtu;
2933 
2934 	if (!netif_is_rxfh_configured(nn->dp.netdev))
2935 		nfp_net_rss_init_itbl(nn);
2936 }
2937 
2938 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
2939 {
2940 	unsigned int r;
2941 	int err;
2942 
2943 	nfp_net_dp_swap(nn, dp);
2944 
2945 	for (r = 0; r <	nn->max_r_vecs; r++)
2946 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
2947 
2948 	err = netif_set_real_num_rx_queues(nn->dp.netdev, nn->dp.num_rx_rings);
2949 	if (err)
2950 		return err;
2951 
2952 	if (nn->dp.netdev->real_num_tx_queues != nn->dp.num_stack_tx_rings) {
2953 		err = netif_set_real_num_tx_queues(nn->dp.netdev,
2954 						   nn->dp.num_stack_tx_rings);
2955 		if (err)
2956 			return err;
2957 	}
2958 
2959 	return nfp_net_set_config_and_enable(nn);
2960 }
2961 
2962 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
2963 {
2964 	struct nfp_net_dp *new;
2965 
2966 	new = kmalloc(sizeof(*new), GFP_KERNEL);
2967 	if (!new)
2968 		return NULL;
2969 
2970 	*new = nn->dp;
2971 
2972 	/* Clear things which need to be recomputed */
2973 	new->fl_bufsz = 0;
2974 	new->tx_rings = NULL;
2975 	new->rx_rings = NULL;
2976 	new->num_r_vecs = 0;
2977 	new->num_stack_tx_rings = 0;
2978 
2979 	return new;
2980 }
2981 
2982 static int
2983 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
2984 		     struct netlink_ext_ack *extack)
2985 {
2986 	/* XDP-enabled tests */
2987 	if (!dp->xdp_prog)
2988 		return 0;
2989 	if (dp->fl_bufsz > PAGE_SIZE) {
2990 		NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
2991 		return -EINVAL;
2992 	}
2993 	if (dp->num_tx_rings > nn->max_tx_rings) {
2994 		NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
2995 		return -EINVAL;
2996 	}
2997 
2998 	return 0;
2999 }
3000 
3001 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
3002 			  struct netlink_ext_ack *extack)
3003 {
3004 	int r, err;
3005 
3006 	dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
3007 
3008 	dp->num_stack_tx_rings = dp->num_tx_rings;
3009 	if (dp->xdp_prog)
3010 		dp->num_stack_tx_rings -= dp->num_rx_rings;
3011 
3012 	dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
3013 
3014 	err = nfp_net_check_config(nn, dp, extack);
3015 	if (err)
3016 		goto exit_free_dp;
3017 
3018 	if (!netif_running(dp->netdev)) {
3019 		nfp_net_dp_swap(nn, dp);
3020 		err = 0;
3021 		goto exit_free_dp;
3022 	}
3023 
3024 	/* Prepare new rings */
3025 	for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
3026 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
3027 		if (err) {
3028 			dp->num_r_vecs = r;
3029 			goto err_cleanup_vecs;
3030 		}
3031 	}
3032 
3033 	err = nfp_net_rx_rings_prepare(nn, dp);
3034 	if (err)
3035 		goto err_cleanup_vecs;
3036 
3037 	err = nfp_net_tx_rings_prepare(nn, dp);
3038 	if (err)
3039 		goto err_free_rx;
3040 
3041 	/* Stop device, swap in new rings, try to start the firmware */
3042 	nfp_net_close_stack(nn);
3043 	nfp_net_clear_config_and_disable(nn);
3044 
3045 	err = nfp_net_dp_swap_enable(nn, dp);
3046 	if (err) {
3047 		int err2;
3048 
3049 		nfp_net_clear_config_and_disable(nn);
3050 
3051 		/* Try with old configuration and old rings */
3052 		err2 = nfp_net_dp_swap_enable(nn, dp);
3053 		if (err2)
3054 			nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
3055 			       err, err2);
3056 	}
3057 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3058 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3059 
3060 	nfp_net_rx_rings_free(dp);
3061 	nfp_net_tx_rings_free(dp);
3062 
3063 	nfp_net_open_stack(nn);
3064 exit_free_dp:
3065 	kfree(dp);
3066 
3067 	return err;
3068 
3069 err_free_rx:
3070 	nfp_net_rx_rings_free(dp);
3071 err_cleanup_vecs:
3072 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3073 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3074 	kfree(dp);
3075 	return err;
3076 }
3077 
3078 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
3079 {
3080 	struct nfp_net *nn = netdev_priv(netdev);
3081 	struct nfp_net_dp *dp;
3082 	int err;
3083 
3084 	err = nfp_app_check_mtu(nn->app, netdev, new_mtu);
3085 	if (err)
3086 		return err;
3087 
3088 	dp = nfp_net_clone_dp(nn);
3089 	if (!dp)
3090 		return -ENOMEM;
3091 
3092 	dp->mtu = new_mtu;
3093 
3094 	return nfp_net_ring_reconfig(nn, dp, NULL);
3095 }
3096 
3097 static int
3098 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
3099 {
3100 	struct nfp_net *nn = netdev_priv(netdev);
3101 
3102 	/* Priority tagged packets with vlan id 0 are processed by the
3103 	 * NFP as untagged packets
3104 	 */
3105 	if (!vid)
3106 		return 0;
3107 
3108 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3109 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3110 		  ETH_P_8021Q);
3111 
3112 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD);
3113 }
3114 
3115 static int
3116 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
3117 {
3118 	struct nfp_net *nn = netdev_priv(netdev);
3119 
3120 	/* Priority tagged packets with vlan id 0 are processed by the
3121 	 * NFP as untagged packets
3122 	 */
3123 	if (!vid)
3124 		return 0;
3125 
3126 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3127 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3128 		  ETH_P_8021Q);
3129 
3130 	return nfp_net_reconfig_mbox(nn, NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL);
3131 }
3132 
3133 #ifdef CONFIG_NET_POLL_CONTROLLER
3134 static void nfp_net_netpoll(struct net_device *netdev)
3135 {
3136 	struct nfp_net *nn = netdev_priv(netdev);
3137 	int i;
3138 
3139 	/* nfp_net's NAPIs are statically allocated so even if there is a race
3140 	 * with reconfig path this will simply try to schedule some disabled
3141 	 * NAPI instances.
3142 	 */
3143 	for (i = 0; i < nn->dp.num_stack_tx_rings; i++)
3144 		napi_schedule_irqoff(&nn->r_vecs[i].napi);
3145 }
3146 #endif
3147 
3148 static void nfp_net_stat64(struct net_device *netdev,
3149 			   struct rtnl_link_stats64 *stats)
3150 {
3151 	struct nfp_net *nn = netdev_priv(netdev);
3152 	int r;
3153 
3154 	for (r = 0; r < nn->max_r_vecs; r++) {
3155 		struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
3156 		u64 data[3];
3157 		unsigned int start;
3158 
3159 		do {
3160 			start = u64_stats_fetch_begin(&r_vec->rx_sync);
3161 			data[0] = r_vec->rx_pkts;
3162 			data[1] = r_vec->rx_bytes;
3163 			data[2] = r_vec->rx_drops;
3164 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
3165 		stats->rx_packets += data[0];
3166 		stats->rx_bytes += data[1];
3167 		stats->rx_dropped += data[2];
3168 
3169 		do {
3170 			start = u64_stats_fetch_begin(&r_vec->tx_sync);
3171 			data[0] = r_vec->tx_pkts;
3172 			data[1] = r_vec->tx_bytes;
3173 			data[2] = r_vec->tx_errors;
3174 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
3175 		stats->tx_packets += data[0];
3176 		stats->tx_bytes += data[1];
3177 		stats->tx_errors += data[2];
3178 	}
3179 }
3180 
3181 static int nfp_net_set_features(struct net_device *netdev,
3182 				netdev_features_t features)
3183 {
3184 	netdev_features_t changed = netdev->features ^ features;
3185 	struct nfp_net *nn = netdev_priv(netdev);
3186 	u32 new_ctrl;
3187 	int err;
3188 
3189 	/* Assume this is not called with features we have not advertised */
3190 
3191 	new_ctrl = nn->dp.ctrl;
3192 
3193 	if (changed & NETIF_F_RXCSUM) {
3194 		if (features & NETIF_F_RXCSUM)
3195 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3196 		else
3197 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
3198 	}
3199 
3200 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3201 		if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
3202 			new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3203 		else
3204 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
3205 	}
3206 
3207 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
3208 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
3209 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3210 					      NFP_NET_CFG_CTRL_LSO;
3211 		else
3212 			new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3213 	}
3214 
3215 	if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
3216 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3217 			new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3218 		else
3219 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
3220 	}
3221 
3222 	if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
3223 		if (features & NETIF_F_HW_VLAN_CTAG_TX)
3224 			new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3225 		else
3226 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
3227 	}
3228 
3229 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
3230 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
3231 			new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3232 		else
3233 			new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
3234 	}
3235 
3236 	if (changed & NETIF_F_SG) {
3237 		if (features & NETIF_F_SG)
3238 			new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
3239 		else
3240 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
3241 	}
3242 
3243 	err = nfp_port_set_features(netdev, features);
3244 	if (err)
3245 		return err;
3246 
3247 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
3248 	       netdev->features, features, changed);
3249 
3250 	if (new_ctrl == nn->dp.ctrl)
3251 		return 0;
3252 
3253 	nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
3254 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3255 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
3256 	if (err)
3257 		return err;
3258 
3259 	nn->dp.ctrl = new_ctrl;
3260 
3261 	return 0;
3262 }
3263 
3264 static netdev_features_t
3265 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
3266 		       netdev_features_t features)
3267 {
3268 	u8 l4_hdr;
3269 
3270 	/* We can't do TSO over double tagged packets (802.1AD) */
3271 	features &= vlan_features_check(skb, features);
3272 
3273 	if (!skb->encapsulation)
3274 		return features;
3275 
3276 	/* Ensure that inner L4 header offset fits into TX descriptor field */
3277 	if (skb_is_gso(skb)) {
3278 		u32 hdrlen;
3279 
3280 		hdrlen = skb_inner_transport_header(skb) - skb->data +
3281 			inner_tcp_hdrlen(skb);
3282 
3283 		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ))
3284 			features &= ~NETIF_F_GSO_MASK;
3285 	}
3286 
3287 	/* VXLAN/GRE check */
3288 	switch (vlan_get_protocol(skb)) {
3289 	case htons(ETH_P_IP):
3290 		l4_hdr = ip_hdr(skb)->protocol;
3291 		break;
3292 	case htons(ETH_P_IPV6):
3293 		l4_hdr = ipv6_hdr(skb)->nexthdr;
3294 		break;
3295 	default:
3296 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3297 	}
3298 
3299 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
3300 	    skb->inner_protocol != htons(ETH_P_TEB) ||
3301 	    (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
3302 	    (l4_hdr == IPPROTO_UDP &&
3303 	     (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
3304 	      sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
3305 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3306 
3307 	return features;
3308 }
3309 
3310 static int
3311 nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len)
3312 {
3313 	struct nfp_net *nn = netdev_priv(netdev);
3314 	int n;
3315 
3316 	if (nn->port)
3317 		return nfp_port_get_phys_port_name(netdev, name, len);
3318 
3319 	if (nn->dp.is_vf || nn->vnic_no_name)
3320 		return -EOPNOTSUPP;
3321 
3322 	n = snprintf(name, len, "n%d", nn->id);
3323 	if (n >= len)
3324 		return -EINVAL;
3325 
3326 	return 0;
3327 }
3328 
3329 /**
3330  * nfp_net_set_vxlan_port() - set vxlan port in SW and reconfigure HW
3331  * @nn:   NFP Net device to reconfigure
3332  * @idx:  Index into the port table where new port should be written
3333  * @port: UDP port to configure (pass zero to remove VXLAN port)
3334  */
3335 static void nfp_net_set_vxlan_port(struct nfp_net *nn, int idx, __be16 port)
3336 {
3337 	int i;
3338 
3339 	nn->vxlan_ports[idx] = port;
3340 
3341 	if (!(nn->dp.ctrl & NFP_NET_CFG_CTRL_VXLAN))
3342 		return;
3343 
3344 	BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
3345 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2)
3346 		nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(port),
3347 			  be16_to_cpu(nn->vxlan_ports[i + 1]) << 16 |
3348 			  be16_to_cpu(nn->vxlan_ports[i]));
3349 
3350 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_VXLAN);
3351 }
3352 
3353 /**
3354  * nfp_net_find_vxlan_idx() - find table entry of the port or a free one
3355  * @nn:   NFP Network structure
3356  * @port: UDP port to look for
3357  *
3358  * Return: if the port is already in the table -- it's position;
3359  *	   if the port is not in the table -- free position to use;
3360  *	   if the table is full -- -ENOSPC.
3361  */
3362 static int nfp_net_find_vxlan_idx(struct nfp_net *nn, __be16 port)
3363 {
3364 	int i, free_idx = -ENOSPC;
3365 
3366 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i++) {
3367 		if (nn->vxlan_ports[i] == port)
3368 			return i;
3369 		if (!nn->vxlan_usecnt[i])
3370 			free_idx = i;
3371 	}
3372 
3373 	return free_idx;
3374 }
3375 
3376 static void nfp_net_add_vxlan_port(struct net_device *netdev,
3377 				   struct udp_tunnel_info *ti)
3378 {
3379 	struct nfp_net *nn = netdev_priv(netdev);
3380 	int idx;
3381 
3382 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3383 		return;
3384 
3385 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3386 	if (idx == -ENOSPC)
3387 		return;
3388 
3389 	if (!nn->vxlan_usecnt[idx]++)
3390 		nfp_net_set_vxlan_port(nn, idx, ti->port);
3391 }
3392 
3393 static void nfp_net_del_vxlan_port(struct net_device *netdev,
3394 				   struct udp_tunnel_info *ti)
3395 {
3396 	struct nfp_net *nn = netdev_priv(netdev);
3397 	int idx;
3398 
3399 	if (ti->type != UDP_TUNNEL_TYPE_VXLAN)
3400 		return;
3401 
3402 	idx = nfp_net_find_vxlan_idx(nn, ti->port);
3403 	if (idx == -ENOSPC || !nn->vxlan_usecnt[idx])
3404 		return;
3405 
3406 	if (!--nn->vxlan_usecnt[idx])
3407 		nfp_net_set_vxlan_port(nn, idx, 0);
3408 }
3409 
3410 static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf)
3411 {
3412 	struct bpf_prog *prog = bpf->prog;
3413 	struct nfp_net_dp *dp;
3414 	int err;
3415 
3416 	if (!xdp_attachment_flags_ok(&nn->xdp, bpf))
3417 		return -EBUSY;
3418 
3419 	if (!prog == !nn->dp.xdp_prog) {
3420 		WRITE_ONCE(nn->dp.xdp_prog, prog);
3421 		xdp_attachment_setup(&nn->xdp, bpf);
3422 		return 0;
3423 	}
3424 
3425 	dp = nfp_net_clone_dp(nn);
3426 	if (!dp)
3427 		return -ENOMEM;
3428 
3429 	dp->xdp_prog = prog;
3430 	dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
3431 	dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
3432 	dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
3433 
3434 	/* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
3435 	err = nfp_net_ring_reconfig(nn, dp, bpf->extack);
3436 	if (err)
3437 		return err;
3438 
3439 	xdp_attachment_setup(&nn->xdp, bpf);
3440 	return 0;
3441 }
3442 
3443 static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf)
3444 {
3445 	int err;
3446 
3447 	if (!xdp_attachment_flags_ok(&nn->xdp_hw, bpf))
3448 		return -EBUSY;
3449 
3450 	err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack);
3451 	if (err)
3452 		return err;
3453 
3454 	xdp_attachment_setup(&nn->xdp_hw, bpf);
3455 	return 0;
3456 }
3457 
3458 static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
3459 {
3460 	struct nfp_net *nn = netdev_priv(netdev);
3461 
3462 	switch (xdp->command) {
3463 	case XDP_SETUP_PROG:
3464 		return nfp_net_xdp_setup_drv(nn, xdp);
3465 	case XDP_SETUP_PROG_HW:
3466 		return nfp_net_xdp_setup_hw(nn, xdp);
3467 	case XDP_QUERY_PROG:
3468 		return xdp_attachment_query(&nn->xdp, xdp);
3469 	case XDP_QUERY_PROG_HW:
3470 		return xdp_attachment_query(&nn->xdp_hw, xdp);
3471 	default:
3472 		return nfp_app_bpf(nn->app, nn, xdp);
3473 	}
3474 }
3475 
3476 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
3477 {
3478 	struct nfp_net *nn = netdev_priv(netdev);
3479 	struct sockaddr *saddr = addr;
3480 	int err;
3481 
3482 	err = eth_prepare_mac_addr_change(netdev, addr);
3483 	if (err)
3484 		return err;
3485 
3486 	nfp_net_write_mac_addr(nn, saddr->sa_data);
3487 
3488 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
3489 	if (err)
3490 		return err;
3491 
3492 	eth_commit_mac_addr_change(netdev, addr);
3493 
3494 	return 0;
3495 }
3496 
3497 const struct net_device_ops nfp_net_netdev_ops = {
3498 	.ndo_init		= nfp_app_ndo_init,
3499 	.ndo_uninit		= nfp_app_ndo_uninit,
3500 	.ndo_open		= nfp_net_netdev_open,
3501 	.ndo_stop		= nfp_net_netdev_close,
3502 	.ndo_start_xmit		= nfp_net_tx,
3503 	.ndo_get_stats64	= nfp_net_stat64,
3504 	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
3505 	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
3506 #ifdef CONFIG_NET_POLL_CONTROLLER
3507 	.ndo_poll_controller	= nfp_net_netpoll,
3508 #endif
3509 	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
3510 	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
3511 	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
3512 	.ndo_get_vf_config	= nfp_app_get_vf_config,
3513 	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
3514 	.ndo_setup_tc		= nfp_port_setup_tc,
3515 	.ndo_tx_timeout		= nfp_net_tx_timeout,
3516 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
3517 	.ndo_change_mtu		= nfp_net_change_mtu,
3518 	.ndo_set_mac_address	= nfp_net_set_mac_address,
3519 	.ndo_set_features	= nfp_net_set_features,
3520 	.ndo_features_check	= nfp_net_features_check,
3521 	.ndo_get_phys_port_name	= nfp_net_get_phys_port_name,
3522 	.ndo_udp_tunnel_add	= nfp_net_add_vxlan_port,
3523 	.ndo_udp_tunnel_del	= nfp_net_del_vxlan_port,
3524 	.ndo_bpf		= nfp_net_xdp,
3525 };
3526 
3527 /**
3528  * nfp_net_info() - Print general info about the NIC
3529  * @nn:      NFP Net device to reconfigure
3530  */
3531 void nfp_net_info(struct nfp_net *nn)
3532 {
3533 	nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
3534 		nn->dp.is_vf ? "VF " : "",
3535 		nn->dp.num_tx_rings, nn->max_tx_rings,
3536 		nn->dp.num_rx_rings, nn->max_rx_rings);
3537 	nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3538 		nn->fw_ver.resv, nn->fw_ver.class,
3539 		nn->fw_ver.major, nn->fw_ver.minor,
3540 		nn->max_mtu);
3541 	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",
3542 		nn->cap,
3543 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
3544 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
3545 		nn->cap & NFP_NET_CFG_CTRL_L2MC     ? "L2MCFILT " : "",
3546 		nn->cap & NFP_NET_CFG_CTRL_RXCSUM   ? "RXCSUM "   : "",
3547 		nn->cap & NFP_NET_CFG_CTRL_TXCSUM   ? "TXCSUM "   : "",
3548 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN   ? "RXVLAN "   : "",
3549 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN   ? "TXVLAN "   : "",
3550 		nn->cap & NFP_NET_CFG_CTRL_SCATTER  ? "SCATTER "  : "",
3551 		nn->cap & NFP_NET_CFG_CTRL_GATHER   ? "GATHER "   : "",
3552 		nn->cap & NFP_NET_CFG_CTRL_LSO      ? "TSO1 "     : "",
3553 		nn->cap & NFP_NET_CFG_CTRL_LSO2     ? "TSO2 "     : "",
3554 		nn->cap & NFP_NET_CFG_CTRL_RSS      ? "RSS1 "     : "",
3555 		nn->cap & NFP_NET_CFG_CTRL_RSS2     ? "RSS2 "     : "",
3556 		nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "",
3557 		nn->cap & NFP_NET_CFG_CTRL_L2SWITCH ? "L2SWITCH " : "",
3558 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
3559 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
3560 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
3561 		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
3562 		nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ?
3563 						      "RXCSUM_COMPLETE " : "",
3564 		nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "",
3565 		nfp_app_extra_cap(nn->app, nn));
3566 }
3567 
3568 /**
3569  * nfp_net_alloc() - Allocate netdev and related structure
3570  * @pdev:         PCI device
3571  * @needs_netdev: Whether to allocate a netdev for this vNIC
3572  * @max_tx_rings: Maximum number of TX rings supported by device
3573  * @max_rx_rings: Maximum number of RX rings supported by device
3574  *
3575  * This function allocates a netdev device and fills in the initial
3576  * part of the @struct nfp_net structure.  In case of control device
3577  * nfp_net structure is allocated without the netdev.
3578  *
3579  * Return: NFP Net device structure, or ERR_PTR on error.
3580  */
3581 struct nfp_net *nfp_net_alloc(struct pci_dev *pdev, bool needs_netdev,
3582 			      unsigned int max_tx_rings,
3583 			      unsigned int max_rx_rings)
3584 {
3585 	struct nfp_net *nn;
3586 
3587 	if (needs_netdev) {
3588 		struct net_device *netdev;
3589 
3590 		netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
3591 					    max_tx_rings, max_rx_rings);
3592 		if (!netdev)
3593 			return ERR_PTR(-ENOMEM);
3594 
3595 		SET_NETDEV_DEV(netdev, &pdev->dev);
3596 		nn = netdev_priv(netdev);
3597 		nn->dp.netdev = netdev;
3598 	} else {
3599 		nn = vzalloc(sizeof(*nn));
3600 		if (!nn)
3601 			return ERR_PTR(-ENOMEM);
3602 	}
3603 
3604 	nn->dp.dev = &pdev->dev;
3605 	nn->pdev = pdev;
3606 
3607 	nn->max_tx_rings = max_tx_rings;
3608 	nn->max_rx_rings = max_rx_rings;
3609 
3610 	nn->dp.num_tx_rings = min_t(unsigned int,
3611 				    max_tx_rings, num_online_cpus());
3612 	nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings,
3613 				 netif_get_num_default_rss_queues());
3614 
3615 	nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings);
3616 	nn->dp.num_r_vecs = min_t(unsigned int,
3617 				  nn->dp.num_r_vecs, num_online_cpus());
3618 
3619 	nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
3620 	nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
3621 
3622 	spin_lock_init(&nn->reconfig_lock);
3623 	spin_lock_init(&nn->link_status_lock);
3624 
3625 	timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0);
3626 
3627 	return nn;
3628 }
3629 
3630 /**
3631  * nfp_net_free() - Undo what @nfp_net_alloc() did
3632  * @nn:      NFP Net device to reconfigure
3633  */
3634 void nfp_net_free(struct nfp_net *nn)
3635 {
3636 	if (nn->dp.netdev)
3637 		free_netdev(nn->dp.netdev);
3638 	else
3639 		vfree(nn);
3640 }
3641 
3642 /**
3643  * nfp_net_rss_key_sz() - Get current size of the RSS key
3644  * @nn:		NFP Net device instance
3645  *
3646  * Return: size of the RSS key for currently selected hash function.
3647  */
3648 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn)
3649 {
3650 	switch (nn->rss_hfunc) {
3651 	case ETH_RSS_HASH_TOP:
3652 		return NFP_NET_CFG_RSS_KEY_SZ;
3653 	case ETH_RSS_HASH_XOR:
3654 		return 0;
3655 	case ETH_RSS_HASH_CRC32:
3656 		return 4;
3657 	}
3658 
3659 	nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc);
3660 	return 0;
3661 }
3662 
3663 /**
3664  * nfp_net_rss_init() - Set the initial RSS parameters
3665  * @nn:	     NFP Net device to reconfigure
3666  */
3667 static void nfp_net_rss_init(struct nfp_net *nn)
3668 {
3669 	unsigned long func_bit, rss_cap_hfunc;
3670 	u32 reg;
3671 
3672 	/* Read the RSS function capability and select first supported func */
3673 	reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP);
3674 	rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg);
3675 	if (!rss_cap_hfunc)
3676 		rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC,
3677 					  NFP_NET_CFG_RSS_TOEPLITZ);
3678 
3679 	func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS);
3680 	if (func_bit == NFP_NET_CFG_RSS_HFUNCS) {
3681 		dev_warn(nn->dp.dev,
3682 			 "Bad RSS config, defaulting to Toeplitz hash\n");
3683 		func_bit = ETH_RSS_HASH_TOP_BIT;
3684 	}
3685 	nn->rss_hfunc = 1 << func_bit;
3686 
3687 	netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn));
3688 
3689 	nfp_net_rss_init_itbl(nn);
3690 
3691 	/* Enable IPv4/IPv6 TCP by default */
3692 	nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
3693 		      NFP_NET_CFG_RSS_IPV6_TCP |
3694 		      FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) |
3695 		      NFP_NET_CFG_RSS_MASK;
3696 }
3697 
3698 /**
3699  * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
3700  * @nn:	     NFP Net device to reconfigure
3701  */
3702 static void nfp_net_irqmod_init(struct nfp_net *nn)
3703 {
3704 	nn->rx_coalesce_usecs      = 50;
3705 	nn->rx_coalesce_max_frames = 64;
3706 	nn->tx_coalesce_usecs      = 50;
3707 	nn->tx_coalesce_max_frames = 64;
3708 }
3709 
3710 static void nfp_net_netdev_init(struct nfp_net *nn)
3711 {
3712 	struct net_device *netdev = nn->dp.netdev;
3713 
3714 	nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
3715 
3716 	netdev->mtu = nn->dp.mtu;
3717 
3718 	/* Advertise/enable offloads based on capabilities
3719 	 *
3720 	 * Note: netdev->features show the currently enabled features
3721 	 * and netdev->hw_features advertises which features are
3722 	 * supported.  By default we enable most features.
3723 	 */
3724 	if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
3725 		netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3726 
3727 	netdev->hw_features = NETIF_F_HIGHDMA;
3728 	if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) {
3729 		netdev->hw_features |= NETIF_F_RXCSUM;
3730 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3731 	}
3732 	if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
3733 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3734 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3735 	}
3736 	if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
3737 		netdev->hw_features |= NETIF_F_SG;
3738 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER;
3739 	}
3740 	if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
3741 	    nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3742 		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3743 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3744 					 NFP_NET_CFG_CTRL_LSO;
3745 	}
3746 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY)
3747 		netdev->hw_features |= NETIF_F_RXHASH;
3748 	if (nn->cap & NFP_NET_CFG_CTRL_VXLAN &&
3749 	    nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
3750 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3751 			netdev->hw_features |= NETIF_F_GSO_GRE |
3752 					       NETIF_F_GSO_UDP_TUNNEL;
3753 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE;
3754 
3755 		netdev->hw_enc_features = netdev->hw_features;
3756 	}
3757 
3758 	netdev->vlan_features = netdev->hw_features;
3759 
3760 	if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
3761 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3762 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3763 	}
3764 	if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
3765 		if (nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3766 			nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
3767 		} else {
3768 			netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3769 			nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3770 		}
3771 	}
3772 	if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) {
3773 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3774 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3775 	}
3776 
3777 	netdev->features = netdev->hw_features;
3778 
3779 	if (nfp_app_has_tc(nn->app) && nn->port)
3780 		netdev->hw_features |= NETIF_F_HW_TC;
3781 
3782 	/* Advertise but disable TSO by default. */
3783 	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
3784 	nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3785 
3786 	/* Finalise the netdev setup */
3787 	netdev->netdev_ops = &nfp_net_netdev_ops;
3788 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
3789 
3790 	SWITCHDEV_SET_OPS(netdev, &nfp_port_switchdev_ops);
3791 
3792 	/* MTU range: 68 - hw-specific max */
3793 	netdev->min_mtu = ETH_MIN_MTU;
3794 	netdev->max_mtu = nn->max_mtu;
3795 
3796 	netdev->gso_max_segs = NFP_NET_LSO_MAX_SEGS;
3797 
3798 	netif_carrier_off(netdev);
3799 
3800 	nfp_net_set_ethtool_ops(netdev);
3801 }
3802 
3803 static int nfp_net_read_caps(struct nfp_net *nn)
3804 {
3805 	/* Get some of the read-only fields from the BAR */
3806 	nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
3807 	nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
3808 
3809 	/* ABI 4.x and ctrl vNIC always use chained metadata, in other cases
3810 	 * we allow use of non-chained metadata if RSS(v1) is the only
3811 	 * advertised capability requiring metadata.
3812 	 */
3813 	nn->dp.chained_metadata_format = nn->fw_ver.major == 4 ||
3814 					 !nn->dp.netdev ||
3815 					 !(nn->cap & NFP_NET_CFG_CTRL_RSS) ||
3816 					 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META;
3817 	/* RSS(v1) uses non-chained metadata format, except in ABI 4.x where
3818 	 * it has the same meaning as RSSv2.
3819 	 */
3820 	if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4)
3821 		nn->cap &= ~NFP_NET_CFG_CTRL_RSS;
3822 
3823 	/* Determine RX packet/metadata boundary offset */
3824 	if (nn->fw_ver.major >= 2) {
3825 		u32 reg;
3826 
3827 		reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
3828 		if (reg > NFP_NET_MAX_PREPEND) {
3829 			nn_err(nn, "Invalid rx offset: %d\n", reg);
3830 			return -EINVAL;
3831 		}
3832 		nn->dp.rx_offset = reg;
3833 	} else {
3834 		nn->dp.rx_offset = NFP_NET_RX_OFFSET;
3835 	}
3836 
3837 	/* For control vNICs mask out the capabilities app doesn't want. */
3838 	if (!nn->dp.netdev)
3839 		nn->cap &= nn->app->type->ctrl_cap_mask;
3840 
3841 	return 0;
3842 }
3843 
3844 /**
3845  * nfp_net_init() - Initialise/finalise the nfp_net structure
3846  * @nn:		NFP Net device structure
3847  *
3848  * Return: 0 on success or negative errno on error.
3849  */
3850 int nfp_net_init(struct nfp_net *nn)
3851 {
3852 	int err;
3853 
3854 	nn->dp.rx_dma_dir = DMA_FROM_DEVICE;
3855 
3856 	err = nfp_net_read_caps(nn);
3857 	if (err)
3858 		return err;
3859 
3860 	/* Set default MTU and Freelist buffer size */
3861 	if (nn->max_mtu < NFP_NET_DEFAULT_MTU)
3862 		nn->dp.mtu = nn->max_mtu;
3863 	else
3864 		nn->dp.mtu = NFP_NET_DEFAULT_MTU;
3865 	nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp);
3866 
3867 	if (nfp_app_ctrl_uses_data_vnics(nn->app))
3868 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_CMSG_DATA;
3869 
3870 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) {
3871 		nfp_net_rss_init(nn);
3872 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?:
3873 					 NFP_NET_CFG_CTRL_RSS;
3874 	}
3875 
3876 	/* Allow L2 Broadcast and Multicast through by default, if supported */
3877 	if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
3878 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC;
3879 
3880 	/* Allow IRQ moderation, if supported */
3881 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
3882 		nfp_net_irqmod_init(nn);
3883 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
3884 	}
3885 
3886 	err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar,
3887 				     &nn->tlv_caps);
3888 	if (err)
3889 		return err;
3890 
3891 	if (nn->dp.netdev)
3892 		nfp_net_netdev_init(nn);
3893 
3894 	/* Stash the re-configuration queue away.  First odd queue in TX Bar */
3895 	nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
3896 
3897 	/* Make sure the FW knows the netdev is supposed to be disabled here */
3898 	nn_writel(nn, NFP_NET_CFG_CTRL, 0);
3899 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
3900 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
3901 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
3902 				   NFP_NET_CFG_UPDATE_GEN);
3903 	if (err)
3904 		return err;
3905 
3906 	nfp_net_vecs_init(nn);
3907 
3908 	if (!nn->dp.netdev)
3909 		return 0;
3910 	return register_netdev(nn->dp.netdev);
3911 }
3912 
3913 /**
3914  * nfp_net_clean() - Undo what nfp_net_init() did.
3915  * @nn:		NFP Net device structure
3916  */
3917 void nfp_net_clean(struct nfp_net *nn)
3918 {
3919 	if (!nn->dp.netdev)
3920 		return;
3921 
3922 	unregister_netdev(nn->dp.netdev);
3923 }
3924