1// SPDX-License-Identifier: GPL-2.0
2/* Copyright(c) 2013 - 2019 Intel Corporation. */
3
4#include <linux/types.h>
5#include <linux/module.h>
6#include <net/ipv6.h>
7#include <net/ip.h>
8#include <net/tcp.h>
9#include <linux/if_macvlan.h>
10#include <linux/prefetch.h>
11
12#include "fm10k.h"
13
14#define DRV_SUMMARY	"Intel(R) Ethernet Switch Host Interface Driver"
15char fm10k_driver_name[] = "fm10k";
16static const char fm10k_driver_string[] = DRV_SUMMARY;
17static const char fm10k_copyright[] =
18	"Copyright(c) 2013 - 2019 Intel Corporation.";
19
20MODULE_AUTHOR("Intel Corporation, <linux.nics@intel.com>");
21MODULE_DESCRIPTION(DRV_SUMMARY);
22MODULE_LICENSE("GPL v2");
23
24/* single workqueue for entire fm10k driver */
25struct workqueue_struct *fm10k_workqueue;
26
27/**
28 * fm10k_init_module - Driver Registration Routine
29 *
30 * fm10k_init_module is the first routine called when the driver is
31 * loaded.  All it does is register with the PCI subsystem.
32 **/
33static int __init fm10k_init_module(void)
34{
35	int ret;
36
37	pr_info("%s\n", fm10k_driver_string);
38	pr_info("%s\n", fm10k_copyright);
39
40	/* create driver workqueue */
41	fm10k_workqueue = alloc_workqueue("%s", WQ_MEM_RECLAIM, 0,
42					  fm10k_driver_name);
43	if (!fm10k_workqueue)
44		return -ENOMEM;
45
46	fm10k_dbg_init();
47
48	ret = fm10k_register_pci_driver();
49	if (ret) {
50		fm10k_dbg_exit();
51		destroy_workqueue(fm10k_workqueue);
52	}
53
54	return ret;
55}
56module_init(fm10k_init_module);
57
58/**
59 * fm10k_exit_module - Driver Exit Cleanup Routine
60 *
61 * fm10k_exit_module is called just before the driver is removed
62 * from memory.
63 **/
64static void __exit fm10k_exit_module(void)
65{
66	fm10k_unregister_pci_driver();
67
68	fm10k_dbg_exit();
69
70	/* destroy driver workqueue */
71	destroy_workqueue(fm10k_workqueue);
72}
73module_exit(fm10k_exit_module);
74
75static bool fm10k_alloc_mapped_page(struct fm10k_ring *rx_ring,
76				    struct fm10k_rx_buffer *bi)
77{
78	struct page *page = bi->page;
79	dma_addr_t dma;
80
81	/* Only page will be NULL if buffer was consumed */
82	if (likely(page))
83		return true;
84
85	/* alloc new page for storage */
86	page = dev_alloc_page();
87	if (unlikely(!page)) {
88		rx_ring->rx_stats.alloc_failed++;
89		return false;
90	}
91
92	/* map page for use */
93	dma = dma_map_page(rx_ring->dev, page, 0, PAGE_SIZE, DMA_FROM_DEVICE);
94
95	/* if mapping failed free memory back to system since
96	 * there isn't much point in holding memory we can't use
97	 */
98	if (dma_mapping_error(rx_ring->dev, dma)) {
99		__free_page(page);
100
101		rx_ring->rx_stats.alloc_failed++;
102		return false;
103	}
104
105	bi->dma = dma;
106	bi->page = page;
107	bi->page_offset = 0;
108
109	return true;
110}
111
112/**
113 * fm10k_alloc_rx_buffers - Replace used receive buffers
114 * @rx_ring: ring to place buffers on
115 * @cleaned_count: number of buffers to replace
116 **/
117void fm10k_alloc_rx_buffers(struct fm10k_ring *rx_ring, u16 cleaned_count)
118{
119	union fm10k_rx_desc *rx_desc;
120	struct fm10k_rx_buffer *bi;
121	u16 i = rx_ring->next_to_use;
122
123	/* nothing to do */
124	if (!cleaned_count)
125		return;
126
127	rx_desc = FM10K_RX_DESC(rx_ring, i);
128	bi = &rx_ring->rx_buffer[i];
129	i -= rx_ring->count;
130
131	do {
132		if (!fm10k_alloc_mapped_page(rx_ring, bi))
133			break;
134
135		/* Refresh the desc even if buffer_addrs didn't change
136		 * because each write-back erases this info.
137		 */
138		rx_desc->q.pkt_addr = cpu_to_le64(bi->dma + bi->page_offset);
139
140		rx_desc++;
141		bi++;
142		i++;
143		if (unlikely(!i)) {
144			rx_desc = FM10K_RX_DESC(rx_ring, 0);
145			bi = rx_ring->rx_buffer;
146			i -= rx_ring->count;
147		}
148
149		/* clear the status bits for the next_to_use descriptor */
150		rx_desc->d.staterr = 0;
151
152		cleaned_count--;
153	} while (cleaned_count);
154
155	i += rx_ring->count;
156
157	if (rx_ring->next_to_use != i) {
158		/* record the next descriptor to use */
159		rx_ring->next_to_use = i;
160
161		/* update next to alloc since we have filled the ring */
162		rx_ring->next_to_alloc = i;
163
164		/* Force memory writes to complete before letting h/w
165		 * know there are new descriptors to fetch.  (Only
166		 * applicable for weak-ordered memory model archs,
167		 * such as IA-64).
168		 */
169		wmb();
170
171		/* notify hardware of new descriptors */
172		writel(i, rx_ring->tail);
173	}
174}
175
176/**
177 * fm10k_reuse_rx_page - page flip buffer and store it back on the ring
178 * @rx_ring: rx descriptor ring to store buffers on
179 * @old_buff: donor buffer to have page reused
180 *
181 * Synchronizes page for reuse by the interface
182 **/
183static void fm10k_reuse_rx_page(struct fm10k_ring *rx_ring,
184				struct fm10k_rx_buffer *old_buff)
185{
186	struct fm10k_rx_buffer *new_buff;
187	u16 nta = rx_ring->next_to_alloc;
188
189	new_buff = &rx_ring->rx_buffer[nta];
190
191	/* update, and store next to alloc */
192	nta++;
193	rx_ring->next_to_alloc = (nta < rx_ring->count) ? nta : 0;
194
195	/* transfer page from old buffer to new buffer */
196	*new_buff = *old_buff;
197
198	/* sync the buffer for use by the device */
199	dma_sync_single_range_for_device(rx_ring->dev, old_buff->dma,
200					 old_buff->page_offset,
201					 FM10K_RX_BUFSZ,
202					 DMA_FROM_DEVICE);
203}
204
205static inline bool fm10k_page_is_reserved(struct page *page)
206{
207	return (page_to_nid(page) != numa_mem_id()) || page_is_pfmemalloc(page);
208}
209
210static bool fm10k_can_reuse_rx_page(struct fm10k_rx_buffer *rx_buffer,
211				    struct page *page,
212				    unsigned int __maybe_unused truesize)
213{
214	/* avoid re-using remote pages */
215	if (unlikely(fm10k_page_is_reserved(page)))
216		return false;
217
218#if (PAGE_SIZE < 8192)
219	/* if we are only owner of page we can reuse it */
220	if (unlikely(page_count(page) != 1))
221		return false;
222
223	/* flip page offset to other buffer */
224	rx_buffer->page_offset ^= FM10K_RX_BUFSZ;
225#else
226	/* move offset up to the next cache line */
227	rx_buffer->page_offset += truesize;
228
229	if (rx_buffer->page_offset > (PAGE_SIZE - FM10K_RX_BUFSZ))
230		return false;
231#endif
232
233	/* Even if we own the page, we are not allowed to use atomic_set()
234	 * This would break get_page_unless_zero() users.
235	 */
236	page_ref_inc(page);
237
238	return true;
239}
240
241/**
242 * fm10k_add_rx_frag - Add contents of Rx buffer to sk_buff
243 * @rx_buffer: buffer containing page to add
244 * @size: packet size from rx_desc
245 * @rx_desc: descriptor containing length of buffer written by hardware
246 * @skb: sk_buff to place the data into
247 *
248 * This function will add the data contained in rx_buffer->page to the skb.
249 * This is done either through a direct copy if the data in the buffer is
250 * less than the skb header size, otherwise it will just attach the page as
251 * a frag to the skb.
252 *
253 * The function will then update the page offset if necessary and return
254 * true if the buffer can be reused by the interface.
255 **/
256static bool fm10k_add_rx_frag(struct fm10k_rx_buffer *rx_buffer,
257			      unsigned int size,
258			      union fm10k_rx_desc *rx_desc,
259			      struct sk_buff *skb)
260{
261	struct page *page = rx_buffer->page;
262	unsigned char *va = page_address(page) + rx_buffer->page_offset;
263#if (PAGE_SIZE < 8192)
264	unsigned int truesize = FM10K_RX_BUFSZ;
265#else
266	unsigned int truesize = ALIGN(size, 512);
267#endif
268	unsigned int pull_len;
269
270	if (unlikely(skb_is_nonlinear(skb)))
271		goto add_tail_frag;
272
273	if (likely(size <= FM10K_RX_HDR_LEN)) {
274		memcpy(__skb_put(skb, size), va, ALIGN(size, sizeof(long)));
275
276		/* page is not reserved, we can reuse buffer as-is */
277		if (likely(!fm10k_page_is_reserved(page)))
278			return true;
279
280		/* this page cannot be reused so discard it */
281		__free_page(page);
282		return false;
283	}
284
285	/* we need the header to contain the greater of either ETH_HLEN or
286	 * 60 bytes if the skb->len is less than 60 for skb_pad.
287	 */
288	pull_len = eth_get_headlen(skb->dev, va, FM10K_RX_HDR_LEN);
289
290	/* align pull length to size of long to optimize memcpy performance */
291	memcpy(__skb_put(skb, pull_len), va, ALIGN(pull_len, sizeof(long)));
292
293	/* update all of the pointers */
294	va += pull_len;
295	size -= pull_len;
296
297add_tail_frag:
298	skb_add_rx_frag(skb, skb_shinfo(skb)->nr_frags, page,
299			(unsigned long)va & ~PAGE_MASK, size, truesize);
300
301	return fm10k_can_reuse_rx_page(rx_buffer, page, truesize);
302}
303
304static struct sk_buff *fm10k_fetch_rx_buffer(struct fm10k_ring *rx_ring,
305					     union fm10k_rx_desc *rx_desc,
306					     struct sk_buff *skb)
307{
308	unsigned int size = le16_to_cpu(rx_desc->w.length);
309	struct fm10k_rx_buffer *rx_buffer;
310	struct page *page;
311
312	rx_buffer = &rx_ring->rx_buffer[rx_ring->next_to_clean];
313	page = rx_buffer->page;
314	prefetchw(page);
315
316	if (likely(!skb)) {
317		void *page_addr = page_address(page) +
318				  rx_buffer->page_offset;
319
320		/* prefetch first cache line of first page */
321		net_prefetch(page_addr);
322
323		/* allocate a skb to store the frags */
324		skb = napi_alloc_skb(&rx_ring->q_vector->napi,
325				     FM10K_RX_HDR_LEN);
326		if (unlikely(!skb)) {
327			rx_ring->rx_stats.alloc_failed++;
328			return NULL;
329		}
330
331		/* we will be copying header into skb->data in
332		 * pskb_may_pull so it is in our interest to prefetch
333		 * it now to avoid a possible cache miss
334		 */
335		prefetchw(skb->data);
336	}
337
338	/* we are reusing so sync this buffer for CPU use */
339	dma_sync_single_range_for_cpu(rx_ring->dev,
340				      rx_buffer->dma,
341				      rx_buffer->page_offset,
342				      size,
343				      DMA_FROM_DEVICE);
344
345	/* pull page into skb */
346	if (fm10k_add_rx_frag(rx_buffer, size, rx_desc, skb)) {
347		/* hand second half of page back to the ring */
348		fm10k_reuse_rx_page(rx_ring, rx_buffer);
349	} else {
350		/* we are not reusing the buffer so unmap it */
351		dma_unmap_page(rx_ring->dev, rx_buffer->dma,
352			       PAGE_SIZE, DMA_FROM_DEVICE);
353	}
354
355	/* clear contents of rx_buffer */
356	rx_buffer->page = NULL;
357
358	return skb;
359}
360
361static inline void fm10k_rx_checksum(struct fm10k_ring *ring,
362				     union fm10k_rx_desc *rx_desc,
363				     struct sk_buff *skb)
364{
365	skb_checksum_none_assert(skb);
366
367	/* Rx checksum disabled via ethtool */
368	if (!(ring->netdev->features & NETIF_F_RXCSUM))
369		return;
370
371	/* TCP/UDP checksum error bit is set */
372	if (fm10k_test_staterr(rx_desc,
373			       FM10K_RXD_STATUS_L4E |
374			       FM10K_RXD_STATUS_L4E2 |
375			       FM10K_RXD_STATUS_IPE |
376			       FM10K_RXD_STATUS_IPE2)) {
377		ring->rx_stats.csum_err++;
378		return;
379	}
380
381	/* It must be a TCP or UDP packet with a valid checksum */
382	if (fm10k_test_staterr(rx_desc, FM10K_RXD_STATUS_L4CS2))
383		skb->encapsulation = true;
384	else if (!fm10k_test_staterr(rx_desc, FM10K_RXD_STATUS_L4CS))
385		return;
386
387	skb->ip_summed = CHECKSUM_UNNECESSARY;
388
389	ring->rx_stats.csum_good++;
390}
391
392#define FM10K_RSS_L4_TYPES_MASK \
393	(BIT(FM10K_RSSTYPE_IPV4_TCP) | \
394	 BIT(FM10K_RSSTYPE_IPV4_UDP) | \
395	 BIT(FM10K_RSSTYPE_IPV6_TCP) | \
396	 BIT(FM10K_RSSTYPE_IPV6_UDP))
397
398static inline void fm10k_rx_hash(struct fm10k_ring *ring,
399				 union fm10k_rx_desc *rx_desc,
400				 struct sk_buff *skb)
401{
402	u16 rss_type;
403
404	if (!(ring->netdev->features & NETIF_F_RXHASH))
405		return;
406
407	rss_type = le16_to_cpu(rx_desc->w.pkt_info) & FM10K_RXD_RSSTYPE_MASK;
408	if (!rss_type)
409		return;
410
411	skb_set_hash(skb, le32_to_cpu(rx_desc->d.rss),
412		     (BIT(rss_type) & FM10K_RSS_L4_TYPES_MASK) ?
413		     PKT_HASH_TYPE_L4 : PKT_HASH_TYPE_L3);
414}
415
416static void fm10k_type_trans(struct fm10k_ring *rx_ring,
417			     union fm10k_rx_desc __maybe_unused *rx_desc,
418			     struct sk_buff *skb)
419{
420	struct net_device *dev = rx_ring->netdev;
421	struct fm10k_l2_accel *l2_accel = rcu_dereference_bh(rx_ring->l2_accel);
422
423	/* check to see if DGLORT belongs to a MACVLAN */
424	if (l2_accel) {
425		u16 idx = le16_to_cpu(FM10K_CB(skb)->fi.w.dglort) - 1;
426
427		idx -= l2_accel->dglort;
428		if (idx < l2_accel->size && l2_accel->macvlan[idx])
429			dev = l2_accel->macvlan[idx];
430		else
431			l2_accel = NULL;
432	}
433
434	/* Record Rx queue, or update macvlan statistics */
435	if (!l2_accel)
436		skb_record_rx_queue(skb, rx_ring->queue_index);
437	else
438		macvlan_count_rx(netdev_priv(dev), skb->len + ETH_HLEN, true,
439				 false);
440
441	skb->protocol = eth_type_trans(skb, dev);
442}
443
444/**
445 * fm10k_process_skb_fields - Populate skb header fields from Rx descriptor
446 * @rx_ring: rx descriptor ring packet is being transacted on
447 * @rx_desc: pointer to the EOP Rx descriptor
448 * @skb: pointer to current skb being populated
449 *
450 * This function checks the ring, descriptor, and packet information in
451 * order to populate the hash, checksum, VLAN, timestamp, protocol, and
452 * other fields within the skb.
453 **/
454static unsigned int fm10k_process_skb_fields(struct fm10k_ring *rx_ring,
455					     union fm10k_rx_desc *rx_desc,
456					     struct sk_buff *skb)
457{
458	unsigned int len = skb->len;
459
460	fm10k_rx_hash(rx_ring, rx_desc, skb);
461
462	fm10k_rx_checksum(rx_ring, rx_desc, skb);
463
464	FM10K_CB(skb)->tstamp = rx_desc->q.timestamp;
465
466	FM10K_CB(skb)->fi.w.vlan = rx_desc->w.vlan;
467
468	FM10K_CB(skb)->fi.d.glort = rx_desc->d.glort;
469
470	if (rx_desc->w.vlan) {
471		u16 vid = le16_to_cpu(rx_desc->w.vlan);
472
473		if ((vid & VLAN_VID_MASK) != rx_ring->vid)
474			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vid);
475		else if (vid & VLAN_PRIO_MASK)
476			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
477					       vid & VLAN_PRIO_MASK);
478	}
479
480	fm10k_type_trans(rx_ring, rx_desc, skb);
481
482	return len;
483}
484
485/**
486 * fm10k_is_non_eop - process handling of non-EOP buffers
487 * @rx_ring: Rx ring being processed
488 * @rx_desc: Rx descriptor for current buffer
489 *
490 * This function updates next to clean.  If the buffer is an EOP buffer
491 * this function exits returning false, otherwise it will place the
492 * sk_buff in the next buffer to be chained and return true indicating
493 * that this is in fact a non-EOP buffer.
494 **/
495static bool fm10k_is_non_eop(struct fm10k_ring *rx_ring,
496			     union fm10k_rx_desc *rx_desc)
497{
498	u32 ntc = rx_ring->next_to_clean + 1;
499
500	/* fetch, update, and store next to clean */
501	ntc = (ntc < rx_ring->count) ? ntc : 0;
502	rx_ring->next_to_clean = ntc;
503
504	prefetch(FM10K_RX_DESC(rx_ring, ntc));
505
506	if (likely(fm10k_test_staterr(rx_desc, FM10K_RXD_STATUS_EOP)))
507		return false;
508
509	return true;
510}
511
512/**
513 * fm10k_cleanup_headers - Correct corrupted or empty headers
514 * @rx_ring: rx descriptor ring packet is being transacted on
515 * @rx_desc: pointer to the EOP Rx descriptor
516 * @skb: pointer to current skb being fixed
517 *
518 * Address the case where we are pulling data in on pages only
519 * and as such no data is present in the skb header.
520 *
521 * In addition if skb is not at least 60 bytes we need to pad it so that
522 * it is large enough to qualify as a valid Ethernet frame.
523 *
524 * Returns true if an error was encountered and skb was freed.
525 **/
526static bool fm10k_cleanup_headers(struct fm10k_ring *rx_ring,
527				  union fm10k_rx_desc *rx_desc,
528				  struct sk_buff *skb)
529{
530	if (unlikely((fm10k_test_staterr(rx_desc,
531					 FM10K_RXD_STATUS_RXE)))) {
532#define FM10K_TEST_RXD_BIT(rxd, bit) \
533	((rxd)->w.csum_err & cpu_to_le16(bit))
534		if (FM10K_TEST_RXD_BIT(rx_desc, FM10K_RXD_ERR_SWITCH_ERROR))
535			rx_ring->rx_stats.switch_errors++;
536		if (FM10K_TEST_RXD_BIT(rx_desc, FM10K_RXD_ERR_NO_DESCRIPTOR))
537			rx_ring->rx_stats.drops++;
538		if (FM10K_TEST_RXD_BIT(rx_desc, FM10K_RXD_ERR_PP_ERROR))
539			rx_ring->rx_stats.pp_errors++;
540		if (FM10K_TEST_RXD_BIT(rx_desc, FM10K_RXD_ERR_SWITCH_READY))
541			rx_ring->rx_stats.link_errors++;
542		if (FM10K_TEST_RXD_BIT(rx_desc, FM10K_RXD_ERR_TOO_BIG))
543			rx_ring->rx_stats.length_errors++;
544		dev_kfree_skb_any(skb);
545		rx_ring->rx_stats.errors++;
546		return true;
547	}
548
549	/* if eth_skb_pad returns an error the skb was freed */
550	if (eth_skb_pad(skb))
551		return true;
552
553	return false;
554}
555
556/**
557 * fm10k_receive_skb - helper function to handle rx indications
558 * @q_vector: structure containing interrupt and ring information
559 * @skb: packet to send up
560 **/
561static void fm10k_receive_skb(struct fm10k_q_vector *q_vector,
562			      struct sk_buff *skb)
563{
564	napi_gro_receive(&q_vector->napi, skb);
565}
566
567static int fm10k_clean_rx_irq(struct fm10k_q_vector *q_vector,
568			      struct fm10k_ring *rx_ring,
569			      int budget)
570{
571	struct sk_buff *skb = rx_ring->skb;
572	unsigned int total_bytes = 0, total_packets = 0;
573	u16 cleaned_count = fm10k_desc_unused(rx_ring);
574
575	while (likely(total_packets < budget)) {
576		union fm10k_rx_desc *rx_desc;
577
578		/* return some buffers to hardware, one at a time is too slow */
579		if (cleaned_count >= FM10K_RX_BUFFER_WRITE) {
580			fm10k_alloc_rx_buffers(rx_ring, cleaned_count);
581			cleaned_count = 0;
582		}
583
584		rx_desc = FM10K_RX_DESC(rx_ring, rx_ring->next_to_clean);
585
586		if (!rx_desc->d.staterr)
587			break;
588
589		/* This memory barrier is needed to keep us from reading
590		 * any other fields out of the rx_desc until we know the
591		 * descriptor has been written back
592		 */
593		dma_rmb();
594
595		/* retrieve a buffer from the ring */
596		skb = fm10k_fetch_rx_buffer(rx_ring, rx_desc, skb);
597
598		/* exit if we failed to retrieve a buffer */
599		if (!skb)
600			break;
601
602		cleaned_count++;
603
604		/* fetch next buffer in frame if non-eop */
605		if (fm10k_is_non_eop(rx_ring, rx_desc))
606			continue;
607
608		/* verify the packet layout is correct */
609		if (fm10k_cleanup_headers(rx_ring, rx_desc, skb)) {
610			skb = NULL;
611			continue;
612		}
613
614		/* populate checksum, timestamp, VLAN, and protocol */
615		total_bytes += fm10k_process_skb_fields(rx_ring, rx_desc, skb);
616
617		fm10k_receive_skb(q_vector, skb);
618
619		/* reset skb pointer */
620		skb = NULL;
621
622		/* update budget accounting */
623		total_packets++;
624	}
625
626	/* place incomplete frames back on ring for completion */
627	rx_ring->skb = skb;
628
629	u64_stats_update_begin(&rx_ring->syncp);
630	rx_ring->stats.packets += total_packets;
631	rx_ring->stats.bytes += total_bytes;
632	u64_stats_update_end(&rx_ring->syncp);
633	q_vector->rx.total_packets += total_packets;
634	q_vector->rx.total_bytes += total_bytes;
635
636	return total_packets;
637}
638
639#define VXLAN_HLEN (sizeof(struct udphdr) + 8)
640static struct ethhdr *fm10k_port_is_vxlan(struct sk_buff *skb)
641{
642	struct fm10k_intfc *interface = netdev_priv(skb->dev);
643
644	if (interface->vxlan_port != udp_hdr(skb)->dest)
645		return NULL;
646
647	/* return offset of udp_hdr plus 8 bytes for VXLAN header */
648	return (struct ethhdr *)(skb_transport_header(skb) + VXLAN_HLEN);
649}
650
651#define FM10K_NVGRE_RESERVED0_FLAGS htons(0x9FFF)
652#define NVGRE_TNI htons(0x2000)
653struct fm10k_nvgre_hdr {
654	__be16 flags;
655	__be16 proto;
656	__be32 tni;
657};
658
659static struct ethhdr *fm10k_gre_is_nvgre(struct sk_buff *skb)
660{
661	struct fm10k_nvgre_hdr *nvgre_hdr;
662	int hlen = ip_hdrlen(skb);
663
664	/* currently only IPv4 is supported due to hlen above */
665	if (vlan_get_protocol(skb) != htons(ETH_P_IP))
666		return NULL;
667
668	/* our transport header should be NVGRE */
669	nvgre_hdr = (struct fm10k_nvgre_hdr *)(skb_network_header(skb) + hlen);
670
671	/* verify all reserved flags are 0 */
672	if (nvgre_hdr->flags & FM10K_NVGRE_RESERVED0_FLAGS)
673		return NULL;
674
675	/* report start of ethernet header */
676	if (nvgre_hdr->flags & NVGRE_TNI)
677		return (struct ethhdr *)(nvgre_hdr + 1);
678
679	return (struct ethhdr *)(&nvgre_hdr->tni);
680}
681
682__be16 fm10k_tx_encap_offload(struct sk_buff *skb)
683{
684	u8 l4_hdr = 0, inner_l4_hdr = 0, inner_l4_hlen;
685	struct ethhdr *eth_hdr;
686
687	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
688	    skb->inner_protocol != htons(ETH_P_TEB))
689		return 0;
690
691	switch (vlan_get_protocol(skb)) {
692	case htons(ETH_P_IP):
693		l4_hdr = ip_hdr(skb)->protocol;
694		break;
695	case htons(ETH_P_IPV6):
696		l4_hdr = ipv6_hdr(skb)->nexthdr;
697		break;
698	default:
699		return 0;
700	}
701
702	switch (l4_hdr) {
703	case IPPROTO_UDP:
704		eth_hdr = fm10k_port_is_vxlan(skb);
705		break;
706	case IPPROTO_GRE:
707		eth_hdr = fm10k_gre_is_nvgre(skb);
708		break;
709	default:
710		return 0;
711	}
712
713	if (!eth_hdr)
714		return 0;
715
716	switch (eth_hdr->h_proto) {
717	case htons(ETH_P_IP):
718		inner_l4_hdr = inner_ip_hdr(skb)->protocol;
719		break;
720	case htons(ETH_P_IPV6):
721		inner_l4_hdr = inner_ipv6_hdr(skb)->nexthdr;
722		break;
723	default:
724		return 0;
725	}
726
727	switch (inner_l4_hdr) {
728	case IPPROTO_TCP:
729		inner_l4_hlen = inner_tcp_hdrlen(skb);
730		break;
731	case IPPROTO_UDP:
732		inner_l4_hlen = 8;
733		break;
734	default:
735		return 0;
736	}
737
738	/* The hardware allows tunnel offloads only if the combined inner and
739	 * outer header is 184 bytes or less
740	 */
741	if (skb_inner_transport_header(skb) + inner_l4_hlen -
742	    skb_mac_header(skb) > FM10K_TUNNEL_HEADER_LENGTH)
743		return 0;
744
745	return eth_hdr->h_proto;
746}
747
748static int fm10k_tso(struct fm10k_ring *tx_ring,
749		     struct fm10k_tx_buffer *first)
750{
751	struct sk_buff *skb = first->skb;
752	struct fm10k_tx_desc *tx_desc;
753	unsigned char *th;
754	u8 hdrlen;
755
756	if (skb->ip_summed != CHECKSUM_PARTIAL)
757		return 0;
758
759	if (!skb_is_gso(skb))
760		return 0;
761
762	/* compute header lengths */
763	if (skb->encapsulation) {
764		if (!fm10k_tx_encap_offload(skb))
765			goto err_vxlan;
766		th = skb_inner_transport_header(skb);
767	} else {
768		th = skb_transport_header(skb);
769	}
770
771	/* compute offset from SOF to transport header and add header len */
772	hdrlen = (th - skb->data) + (((struct tcphdr *)th)->doff << 2);
773
774	first->tx_flags |= FM10K_TX_FLAGS_CSUM;
775
776	/* update gso size and bytecount with header size */
777	first->gso_segs = skb_shinfo(skb)->gso_segs;
778	first->bytecount += (first->gso_segs - 1) * hdrlen;
779
780	/* populate Tx descriptor header size and mss */
781	tx_desc = FM10K_TX_DESC(tx_ring, tx_ring->next_to_use);
782	tx_desc->hdrlen = hdrlen;
783	tx_desc->mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
784
785	return 1;
786
787err_vxlan:
788	tx_ring->netdev->features &= ~NETIF_F_GSO_UDP_TUNNEL;
789	if (net_ratelimit())
790		netdev_err(tx_ring->netdev,
791			   "TSO requested for unsupported tunnel, disabling offload\n");
792	return -1;
793}
794
795static void fm10k_tx_csum(struct fm10k_ring *tx_ring,
796			  struct fm10k_tx_buffer *first)
797{
798	struct sk_buff *skb = first->skb;
799	struct fm10k_tx_desc *tx_desc;
800	union {
801		struct iphdr *ipv4;
802		struct ipv6hdr *ipv6;
803		u8 *raw;
804	} network_hdr;
805	u8 *transport_hdr;
806	__be16 frag_off;
807	__be16 protocol;
808	u8 l4_hdr = 0;
809
810	if (skb->ip_summed != CHECKSUM_PARTIAL)
811		goto no_csum;
812
813	if (skb->encapsulation) {
814		protocol = fm10k_tx_encap_offload(skb);
815		if (!protocol) {
816			if (skb_checksum_help(skb)) {
817				dev_warn(tx_ring->dev,
818					 "failed to offload encap csum!\n");
819				tx_ring->tx_stats.csum_err++;
820			}
821			goto no_csum;
822		}
823		network_hdr.raw = skb_inner_network_header(skb);
824		transport_hdr = skb_inner_transport_header(skb);
825	} else {
826		protocol = vlan_get_protocol(skb);
827		network_hdr.raw = skb_network_header(skb);
828		transport_hdr = skb_transport_header(skb);
829	}
830
831	switch (protocol) {
832	case htons(ETH_P_IP):
833		l4_hdr = network_hdr.ipv4->protocol;
834		break;
835	case htons(ETH_P_IPV6):
836		l4_hdr = network_hdr.ipv6->nexthdr;
837		if (likely((transport_hdr - network_hdr.raw) ==
838			   sizeof(struct ipv6hdr)))
839			break;
840		ipv6_skip_exthdr(skb, network_hdr.raw - skb->data +
841				      sizeof(struct ipv6hdr),
842				 &l4_hdr, &frag_off);
843		if (unlikely(frag_off))
844			l4_hdr = NEXTHDR_FRAGMENT;
845		break;
846	default:
847		break;
848	}
849
850	switch (l4_hdr) {
851	case IPPROTO_TCP:
852	case IPPROTO_UDP:
853		break;
854	case IPPROTO_GRE:
855		if (skb->encapsulation)
856			break;
857		fallthrough;
858	default:
859		if (unlikely(net_ratelimit())) {
860			dev_warn(tx_ring->dev,
861				 "partial checksum, version=%d l4 proto=%x\n",
862				 protocol, l4_hdr);
863		}
864		skb_checksum_help(skb);
865		tx_ring->tx_stats.csum_err++;
866		goto no_csum;
867	}
868
869	/* update TX checksum flag */
870	first->tx_flags |= FM10K_TX_FLAGS_CSUM;
871	tx_ring->tx_stats.csum_good++;
872
873no_csum:
874	/* populate Tx descriptor header size and mss */
875	tx_desc = FM10K_TX_DESC(tx_ring, tx_ring->next_to_use);
876	tx_desc->hdrlen = 0;
877	tx_desc->mss = 0;
878}
879
880#define FM10K_SET_FLAG(_input, _flag, _result) \
881	((_flag <= _result) ? \
882	 ((u32)(_input & _flag) * (_result / _flag)) : \
883	 ((u32)(_input & _flag) / (_flag / _result)))
884
885static u8 fm10k_tx_desc_flags(struct sk_buff *skb, u32 tx_flags)
886{
887	/* set type for advanced descriptor with frame checksum insertion */
888	u32 desc_flags = 0;
889
890	/* set checksum offload bits */
891	desc_flags |= FM10K_SET_FLAG(tx_flags, FM10K_TX_FLAGS_CSUM,
892				     FM10K_TXD_FLAG_CSUM);
893
894	return desc_flags;
895}
896
897static bool fm10k_tx_desc_push(struct fm10k_ring *tx_ring,
898			       struct fm10k_tx_desc *tx_desc, u16 i,
899			       dma_addr_t dma, unsigned int size, u8 desc_flags)
900{
901	/* set RS and INT for last frame in a cache line */
902	if ((++i & (FM10K_TXD_WB_FIFO_SIZE - 1)) == 0)
903		desc_flags |= FM10K_TXD_FLAG_RS | FM10K_TXD_FLAG_INT;
904
905	/* record values to descriptor */
906	tx_desc->buffer_addr = cpu_to_le64(dma);
907	tx_desc->flags = desc_flags;
908	tx_desc->buflen = cpu_to_le16(size);
909
910	/* return true if we just wrapped the ring */
911	return i == tx_ring->count;
912}
913
914static int __fm10k_maybe_stop_tx(struct fm10k_ring *tx_ring, u16 size)
915{
916	netif_stop_subqueue(tx_ring->netdev, tx_ring->queue_index);
917
918	/* Memory barrier before checking head and tail */
919	smp_mb();
920
921	/* Check again in a case another CPU has just made room available */
922	if (likely(fm10k_desc_unused(tx_ring) < size))
923		return -EBUSY;
924
925	/* A reprieve! - use start_queue because it doesn't call schedule */
926	netif_start_subqueue(tx_ring->netdev, tx_ring->queue_index);
927	++tx_ring->tx_stats.restart_queue;
928	return 0;
929}
930
931static inline int fm10k_maybe_stop_tx(struct fm10k_ring *tx_ring, u16 size)
932{
933	if (likely(fm10k_desc_unused(tx_ring) >= size))
934		return 0;
935	return __fm10k_maybe_stop_tx(tx_ring, size);
936}
937
938static void fm10k_tx_map(struct fm10k_ring *tx_ring,
939			 struct fm10k_tx_buffer *first)
940{
941	struct sk_buff *skb = first->skb;
942	struct fm10k_tx_buffer *tx_buffer;
943	struct fm10k_tx_desc *tx_desc;
944	skb_frag_t *frag;
945	unsigned char *data;
946	dma_addr_t dma;
947	unsigned int data_len, size;
948	u32 tx_flags = first->tx_flags;
949	u16 i = tx_ring->next_to_use;
950	u8 flags = fm10k_tx_desc_flags(skb, tx_flags);
951
952	tx_desc = FM10K_TX_DESC(tx_ring, i);
953
954	/* add HW VLAN tag */
955	if (skb_vlan_tag_present(skb))
956		tx_desc->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
957	else
958		tx_desc->vlan = 0;
959
960	size = skb_headlen(skb);
961	data = skb->data;
962
963	dma = dma_map_single(tx_ring->dev, data, size, DMA_TO_DEVICE);
964
965	data_len = skb->data_len;
966	tx_buffer = first;
967
968	for (frag = &skb_shinfo(skb)->frags[0];; frag++) {
969		if (dma_mapping_error(tx_ring->dev, dma))
970			goto dma_error;
971
972		/* record length, and DMA address */
973		dma_unmap_len_set(tx_buffer, len, size);
974		dma_unmap_addr_set(tx_buffer, dma, dma);
975
976		while (unlikely(size > FM10K_MAX_DATA_PER_TXD)) {
977			if (fm10k_tx_desc_push(tx_ring, tx_desc++, i++, dma,
978					       FM10K_MAX_DATA_PER_TXD, flags)) {
979				tx_desc = FM10K_TX_DESC(tx_ring, 0);
980				i = 0;
981			}
982
983			dma += FM10K_MAX_DATA_PER_TXD;
984			size -= FM10K_MAX_DATA_PER_TXD;
985		}
986
987		if (likely(!data_len))
988			break;
989
990		if (fm10k_tx_desc_push(tx_ring, tx_desc++, i++,
991				       dma, size, flags)) {
992			tx_desc = FM10K_TX_DESC(tx_ring, 0);
993			i = 0;
994		}
995
996		size = skb_frag_size(frag);
997		data_len -= size;
998
999		dma = skb_frag_dma_map(tx_ring->dev, frag, 0, size,
1000				       DMA_TO_DEVICE);
1001
1002		tx_buffer = &tx_ring->tx_buffer[i];
1003	}
1004
1005	/* write last descriptor with LAST bit set */
1006	flags |= FM10K_TXD_FLAG_LAST;
1007
1008	if (fm10k_tx_desc_push(tx_ring, tx_desc, i++, dma, size, flags))
1009		i = 0;
1010
1011	/* record bytecount for BQL */
1012	netdev_tx_sent_queue(txring_txq(tx_ring), first->bytecount);
1013
1014	/* record SW timestamp if HW timestamp is not available */
1015	skb_tx_timestamp(first->skb);
1016
1017	/* Force memory writes to complete before letting h/w know there
1018	 * are new descriptors to fetch.  (Only applicable for weak-ordered
1019	 * memory model archs, such as IA-64).
1020	 *
1021	 * We also need this memory barrier to make certain all of the
1022	 * status bits have been updated before next_to_watch is written.
1023	 */
1024	wmb();
1025
1026	/* set next_to_watch value indicating a packet is present */
1027	first->next_to_watch = tx_desc;
1028
1029	tx_ring->next_to_use = i;
1030
1031	/* Make sure there is space in the ring for the next send. */
1032	fm10k_maybe_stop_tx(tx_ring, DESC_NEEDED);
1033
1034	/* notify HW of packet */
1035	if (netif_xmit_stopped(txring_txq(tx_ring)) || !netdev_xmit_more()) {
1036		writel(i, tx_ring->tail);
1037	}
1038
1039	return;
1040dma_error:
1041	dev_err(tx_ring->dev, "TX DMA map failed\n");
1042
1043	/* clear dma mappings for failed tx_buffer map */
1044	for (;;) {
1045		tx_buffer = &tx_ring->tx_buffer[i];
1046		fm10k_unmap_and_free_tx_resource(tx_ring, tx_buffer);
1047		if (tx_buffer == first)
1048			break;
1049		if (i == 0)
1050			i = tx_ring->count;
1051		i--;
1052	}
1053
1054	tx_ring->next_to_use = i;
1055}
1056
1057netdev_tx_t fm10k_xmit_frame_ring(struct sk_buff *skb,
1058				  struct fm10k_ring *tx_ring)
1059{
1060	u16 count = TXD_USE_COUNT(skb_headlen(skb));
1061	struct fm10k_tx_buffer *first;
1062	unsigned short f;
1063	u32 tx_flags = 0;
1064	int tso;
1065
1066	/* need: 1 descriptor per page * PAGE_SIZE/FM10K_MAX_DATA_PER_TXD,
1067	 *       + 1 desc for skb_headlen/FM10K_MAX_DATA_PER_TXD,
1068	 *       + 2 desc gap to keep tail from touching head
1069	 * otherwise try next time
1070	 */
1071	for (f = 0; f < skb_shinfo(skb)->nr_frags; f++) {
1072		skb_frag_t *frag = &skb_shinfo(skb)->frags[f];
1073
1074		count += TXD_USE_COUNT(skb_frag_size(frag));
1075	}
1076
1077	if (fm10k_maybe_stop_tx(tx_ring, count + 3)) {
1078		tx_ring->tx_stats.tx_busy++;
1079		return NETDEV_TX_BUSY;
1080	}
1081
1082	/* record the location of the first descriptor for this packet */
1083	first = &tx_ring->tx_buffer[tx_ring->next_to_use];
1084	first->skb = skb;
1085	first->bytecount = max_t(unsigned int, skb->len, ETH_ZLEN);
1086	first->gso_segs = 1;
1087
1088	/* record initial flags and protocol */
1089	first->tx_flags = tx_flags;
1090
1091	tso = fm10k_tso(tx_ring, first);
1092	if (tso < 0)
1093		goto out_drop;
1094	else if (!tso)
1095		fm10k_tx_csum(tx_ring, first);
1096
1097	fm10k_tx_map(tx_ring, first);
1098
1099	return NETDEV_TX_OK;
1100
1101out_drop:
1102	dev_kfree_skb_any(first->skb);
1103	first->skb = NULL;
1104
1105	return NETDEV_TX_OK;
1106}
1107
1108static u64 fm10k_get_tx_completed(struct fm10k_ring *ring)
1109{
1110	return ring->stats.packets;
1111}
1112
1113/**
1114 * fm10k_get_tx_pending - how many Tx descriptors not processed
1115 * @ring: the ring structure
1116 * @in_sw: is tx_pending being checked in SW or in HW?
1117 */
1118u64 fm10k_get_tx_pending(struct fm10k_ring *ring, bool in_sw)
1119{
1120	struct fm10k_intfc *interface = ring->q_vector->interface;
1121	struct fm10k_hw *hw = &interface->hw;
1122	u32 head, tail;
1123
1124	if (likely(in_sw)) {
1125		head = ring->next_to_clean;
1126		tail = ring->next_to_use;
1127	} else {
1128		head = fm10k_read_reg(hw, FM10K_TDH(ring->reg_idx));
1129		tail = fm10k_read_reg(hw, FM10K_TDT(ring->reg_idx));
1130	}
1131
1132	return ((head <= tail) ? tail : tail + ring->count) - head;
1133}
1134
1135bool fm10k_check_tx_hang(struct fm10k_ring *tx_ring)
1136{
1137	u32 tx_done = fm10k_get_tx_completed(tx_ring);
1138	u32 tx_done_old = tx_ring->tx_stats.tx_done_old;
1139	u32 tx_pending = fm10k_get_tx_pending(tx_ring, true);
1140
1141	clear_check_for_tx_hang(tx_ring);
1142
1143	/* Check for a hung queue, but be thorough. This verifies
1144	 * that a transmit has been completed since the previous
1145	 * check AND there is at least one packet pending. By
1146	 * requiring this to fail twice we avoid races with
1147	 * clearing the ARMED bit and conditions where we
1148	 * run the check_tx_hang logic with a transmit completion
1149	 * pending but without time to complete it yet.
1150	 */
1151	if (!tx_pending || (tx_done_old != tx_done)) {
1152		/* update completed stats and continue */
1153		tx_ring->tx_stats.tx_done_old = tx_done;
1154		/* reset the countdown */
1155		clear_bit(__FM10K_HANG_CHECK_ARMED, tx_ring->state);
1156
1157		return false;
1158	}
1159
1160	/* make sure it is true for two checks in a row */
1161	return test_and_set_bit(__FM10K_HANG_CHECK_ARMED, tx_ring->state);
1162}
1163
1164/**
1165 * fm10k_tx_timeout_reset - initiate reset due to Tx timeout
1166 * @interface: driver private struct
1167 **/
1168void fm10k_tx_timeout_reset(struct fm10k_intfc *interface)
1169{
1170	/* Do the reset outside of interrupt context */
1171	if (!test_bit(__FM10K_DOWN, interface->state)) {
1172		interface->tx_timeout_count++;
1173		set_bit(FM10K_FLAG_RESET_REQUESTED, interface->flags);
1174		fm10k_service_event_schedule(interface);
1175	}
1176}
1177
1178/**
1179 * fm10k_clean_tx_irq - Reclaim resources after transmit completes
1180 * @q_vector: structure containing interrupt and ring information
1181 * @tx_ring: tx ring to clean
1182 * @napi_budget: Used to determine if we are in netpoll
1183 **/
1184static bool fm10k_clean_tx_irq(struct fm10k_q_vector *q_vector,
1185			       struct fm10k_ring *tx_ring, int napi_budget)
1186{
1187	struct fm10k_intfc *interface = q_vector->interface;
1188	struct fm10k_tx_buffer *tx_buffer;
1189	struct fm10k_tx_desc *tx_desc;
1190	unsigned int total_bytes = 0, total_packets = 0;
1191	unsigned int budget = q_vector->tx.work_limit;
1192	unsigned int i = tx_ring->next_to_clean;
1193
1194	if (test_bit(__FM10K_DOWN, interface->state))
1195		return true;
1196
1197	tx_buffer = &tx_ring->tx_buffer[i];
1198	tx_desc = FM10K_TX_DESC(tx_ring, i);
1199	i -= tx_ring->count;
1200
1201	do {
1202		struct fm10k_tx_desc *eop_desc = tx_buffer->next_to_watch;
1203
1204		/* if next_to_watch is not set then there is no work pending */
1205		if (!eop_desc)
1206			break;
1207
1208		/* prevent any other reads prior to eop_desc */
1209		smp_rmb();
1210
1211		/* if DD is not set pending work has not been completed */
1212		if (!(eop_desc->flags & FM10K_TXD_FLAG_DONE))
1213			break;
1214
1215		/* clear next_to_watch to prevent false hangs */
1216		tx_buffer->next_to_watch = NULL;
1217
1218		/* update the statistics for this packet */
1219		total_bytes += tx_buffer->bytecount;
1220		total_packets += tx_buffer->gso_segs;
1221
1222		/* free the skb */
1223		napi_consume_skb(tx_buffer->skb, napi_budget);
1224
1225		/* unmap skb header data */
1226		dma_unmap_single(tx_ring->dev,
1227				 dma_unmap_addr(tx_buffer, dma),
1228				 dma_unmap_len(tx_buffer, len),
1229				 DMA_TO_DEVICE);
1230
1231		/* clear tx_buffer data */
1232		tx_buffer->skb = NULL;
1233		dma_unmap_len_set(tx_buffer, len, 0);
1234
1235		/* unmap remaining buffers */
1236		while (tx_desc != eop_desc) {
1237			tx_buffer++;
1238			tx_desc++;
1239			i++;
1240			if (unlikely(!i)) {
1241				i -= tx_ring->count;
1242				tx_buffer = tx_ring->tx_buffer;
1243				tx_desc = FM10K_TX_DESC(tx_ring, 0);
1244			}
1245
1246			/* unmap any remaining paged data */
1247			if (dma_unmap_len(tx_buffer, len)) {
1248				dma_unmap_page(tx_ring->dev,
1249					       dma_unmap_addr(tx_buffer, dma),
1250					       dma_unmap_len(tx_buffer, len),
1251					       DMA_TO_DEVICE);
1252				dma_unmap_len_set(tx_buffer, len, 0);
1253			}
1254		}
1255
1256		/* move us one more past the eop_desc for start of next pkt */
1257		tx_buffer++;
1258		tx_desc++;
1259		i++;
1260		if (unlikely(!i)) {
1261			i -= tx_ring->count;
1262			tx_buffer = tx_ring->tx_buffer;
1263			tx_desc = FM10K_TX_DESC(tx_ring, 0);
1264		}
1265
1266		/* issue prefetch for next Tx descriptor */
1267		prefetch(tx_desc);
1268
1269		/* update budget accounting */
1270		budget--;
1271	} while (likely(budget));
1272
1273	i += tx_ring->count;
1274	tx_ring->next_to_clean = i;
1275	u64_stats_update_begin(&tx_ring->syncp);
1276	tx_ring->stats.bytes += total_bytes;
1277	tx_ring->stats.packets += total_packets;
1278	u64_stats_update_end(&tx_ring->syncp);
1279	q_vector->tx.total_bytes += total_bytes;
1280	q_vector->tx.total_packets += total_packets;
1281
1282	if (check_for_tx_hang(tx_ring) && fm10k_check_tx_hang(tx_ring)) {
1283		/* schedule immediate reset if we believe we hung */
1284		struct fm10k_hw *hw = &interface->hw;
1285
1286		netif_err(interface, drv, tx_ring->netdev,
1287			  "Detected Tx Unit Hang\n"
1288			  "  Tx Queue             <%d>\n"
1289			  "  TDH, TDT             <%x>, <%x>\n"
1290			  "  next_to_use          <%x>\n"
1291			  "  next_to_clean        <%x>\n",
1292			  tx_ring->queue_index,
1293			  fm10k_read_reg(hw, FM10K_TDH(tx_ring->reg_idx)),
1294			  fm10k_read_reg(hw, FM10K_TDT(tx_ring->reg_idx)),
1295			  tx_ring->next_to_use, i);
1296
1297		netif_stop_subqueue(tx_ring->netdev,
1298				    tx_ring->queue_index);
1299
1300		netif_info(interface, probe, tx_ring->netdev,
1301			   "tx hang %d detected on queue %d, resetting interface\n",
1302			   interface->tx_timeout_count + 1,
1303			   tx_ring->queue_index);
1304
1305		fm10k_tx_timeout_reset(interface);
1306
1307		/* the netdev is about to reset, no point in enabling stuff */
1308		return true;
1309	}
1310
1311	/* notify netdev of completed buffers */
1312	netdev_tx_completed_queue(txring_txq(tx_ring),
1313				  total_packets, total_bytes);
1314
1315#define TX_WAKE_THRESHOLD min_t(u16, FM10K_MIN_TXD - 1, DESC_NEEDED * 2)
1316	if (unlikely(total_packets && netif_carrier_ok(tx_ring->netdev) &&
1317		     (fm10k_desc_unused(tx_ring) >= TX_WAKE_THRESHOLD))) {
1318		/* Make sure that anybody stopping the queue after this
1319		 * sees the new next_to_clean.
1320		 */
1321		smp_mb();
1322		if (__netif_subqueue_stopped(tx_ring->netdev,
1323					     tx_ring->queue_index) &&
1324		    !test_bit(__FM10K_DOWN, interface->state)) {
1325			netif_wake_subqueue(tx_ring->netdev,
1326					    tx_ring->queue_index);
1327			++tx_ring->tx_stats.restart_queue;
1328		}
1329	}
1330
1331	return !!budget;
1332}
1333
1334/**
1335 * fm10k_update_itr - update the dynamic ITR value based on packet size
1336 *
1337 *      Stores a new ITR value based on strictly on packet size.  The
1338 *      divisors and thresholds used by this function were determined based
1339 *      on theoretical maximum wire speed and testing data, in order to
1340 *      minimize response time while increasing bulk throughput.
1341 *
1342 * @ring_container: Container for rings to have ITR updated
1343 **/
1344static void fm10k_update_itr(struct fm10k_ring_container *ring_container)
1345{
1346	unsigned int avg_wire_size, packets, itr_round;
1347
1348	/* Only update ITR if we are using adaptive setting */
1349	if (!ITR_IS_ADAPTIVE(ring_container->itr))
1350		goto clear_counts;
1351
1352	packets = ring_container->total_packets;
1353	if (!packets)
1354		goto clear_counts;
1355
1356	avg_wire_size = ring_container->total_bytes / packets;
1357
1358	/* The following is a crude approximation of:
1359	 *  wmem_default / (size + overhead) = desired_pkts_per_int
1360	 *  rate / bits_per_byte / (size + ethernet overhead) = pkt_rate
1361	 *  (desired_pkt_rate / pkt_rate) * usecs_per_sec = ITR value
1362	 *
1363	 * Assuming wmem_default is 212992 and overhead is 640 bytes per
1364	 * packet, (256 skb, 64 headroom, 320 shared info), we can reduce the
1365	 * formula down to
1366	 *
1367	 *  (34 * (size + 24)) / (size + 640) = ITR
1368	 *
1369	 * We first do some math on the packet size and then finally bitshift
1370	 * by 8 after rounding up. We also have to account for PCIe link speed
1371	 * difference as ITR scales based on this.
1372	 */
1373	if (avg_wire_size <= 360) {
1374		/* Start at 250K ints/sec and gradually drop to 77K ints/sec */
1375		avg_wire_size *= 8;
1376		avg_wire_size += 376;
1377	} else if (avg_wire_size <= 1152) {
1378		/* 77K ints/sec to 45K ints/sec */
1379		avg_wire_size *= 3;
1380		avg_wire_size += 2176;
1381	} else if (avg_wire_size <= 1920) {
1382		/* 45K ints/sec to 38K ints/sec */
1383		avg_wire_size += 4480;
1384	} else {
1385		/* plateau at a limit of 38K ints/sec */
1386		avg_wire_size = 6656;
1387	}
1388
1389	/* Perform final bitshift for division after rounding up to ensure
1390	 * that the calculation will never get below a 1. The bit shift
1391	 * accounts for changes in the ITR due to PCIe link speed.
1392	 */
1393	itr_round = READ_ONCE(ring_container->itr_scale) + 8;
1394	avg_wire_size += BIT(itr_round) - 1;
1395	avg_wire_size >>= itr_round;
1396
1397	/* write back value and retain adaptive flag */
1398	ring_container->itr = avg_wire_size | FM10K_ITR_ADAPTIVE;
1399
1400clear_counts:
1401	ring_container->total_bytes = 0;
1402	ring_container->total_packets = 0;
1403}
1404
1405static void fm10k_qv_enable(struct fm10k_q_vector *q_vector)
1406{
1407	/* Enable auto-mask and clear the current mask */
1408	u32 itr = FM10K_ITR_ENABLE;
1409
1410	/* Update Tx ITR */
1411	fm10k_update_itr(&q_vector->tx);
1412
1413	/* Update Rx ITR */
1414	fm10k_update_itr(&q_vector->rx);
1415
1416	/* Store Tx itr in timer slot 0 */
1417	itr |= (q_vector->tx.itr & FM10K_ITR_MAX);
1418
1419	/* Shift Rx itr to timer slot 1 */
1420	itr |= (q_vector->rx.itr & FM10K_ITR_MAX) << FM10K_ITR_INTERVAL1_SHIFT;
1421
1422	/* Write the final value to the ITR register */
1423	writel(itr, q_vector->itr);
1424}
1425
1426static int fm10k_poll(struct napi_struct *napi, int budget)
1427{
1428	struct fm10k_q_vector *q_vector =
1429			       container_of(napi, struct fm10k_q_vector, napi);
1430	struct fm10k_ring *ring;
1431	int per_ring_budget, work_done = 0;
1432	bool clean_complete = true;
1433
1434	fm10k_for_each_ring(ring, q_vector->tx) {
1435		if (!fm10k_clean_tx_irq(q_vector, ring, budget))
1436			clean_complete = false;
1437	}
1438
1439	/* Handle case where we are called by netpoll with a budget of 0 */
1440	if (budget <= 0)
1441		return budget;
1442
1443	/* attempt to distribute budget to each queue fairly, but don't
1444	 * allow the budget to go below 1 because we'll exit polling
1445	 */
1446	if (q_vector->rx.count > 1)
1447		per_ring_budget = max(budget / q_vector->rx.count, 1);
1448	else
1449		per_ring_budget = budget;
1450
1451	fm10k_for_each_ring(ring, q_vector->rx) {
1452		int work = fm10k_clean_rx_irq(q_vector, ring, per_ring_budget);
1453
1454		work_done += work;
1455		if (work >= per_ring_budget)
1456			clean_complete = false;
1457	}
1458
1459	/* If all work not completed, return budget and keep polling */
1460	if (!clean_complete)
1461		return budget;
1462
1463	/* Exit the polling mode, but don't re-enable interrupts if stack might
1464	 * poll us due to busy-polling
1465	 */
1466	if (likely(napi_complete_done(napi, work_done)))
1467		fm10k_qv_enable(q_vector);
1468
1469	return min(work_done, budget - 1);
1470}
1471
1472/**
1473 * fm10k_set_qos_queues: Allocate queues for a QOS-enabled device
1474 * @interface: board private structure to initialize
1475 *
1476 * When QoS (Quality of Service) is enabled, allocate queues for
1477 * each traffic class.  If multiqueue isn't available,then abort QoS
1478 * initialization.
1479 *
1480 * This function handles all combinations of Qos and RSS.
1481 *
1482 **/
1483static bool fm10k_set_qos_queues(struct fm10k_intfc *interface)
1484{
1485	struct net_device *dev = interface->netdev;
1486	struct fm10k_ring_feature *f;
1487	int rss_i, i;
1488	int pcs;
1489
1490	/* Map queue offset and counts onto allocated tx queues */
1491	pcs = netdev_get_num_tc(dev);
1492
1493	if (pcs <= 1)
1494		return false;
1495
1496	/* set QoS mask and indices */
1497	f = &interface->ring_feature[RING_F_QOS];
1498	f->indices = pcs;
1499	f->mask = BIT(fls(pcs - 1)) - 1;
1500
1501	/* determine the upper limit for our current DCB mode */
1502	rss_i = interface->hw.mac.max_queues / pcs;
1503	rss_i = BIT(fls(rss_i) - 1);
1504
1505	/* set RSS mask and indices */
1506	f = &interface->ring_feature[RING_F_RSS];
1507	rss_i = min_t(u16, rss_i, f->limit);
1508	f->indices = rss_i;
1509	f->mask = BIT(fls(rss_i - 1)) - 1;
1510
1511	/* configure pause class to queue mapping */
1512	for (i = 0; i < pcs; i++)
1513		netdev_set_tc_queue(dev, i, rss_i, rss_i * i);
1514
1515	interface->num_rx_queues = rss_i * pcs;
1516	interface->num_tx_queues = rss_i * pcs;
1517
1518	return true;
1519}
1520
1521/**
1522 * fm10k_set_rss_queues: Allocate queues for RSS
1523 * @interface: board private structure to initialize
1524 *
1525 * This is our "base" multiqueue mode.  RSS (Receive Side Scaling) will try
1526 * to allocate one Rx queue per CPU, and if available, one Tx queue per CPU.
1527 *
1528 **/
1529static bool fm10k_set_rss_queues(struct fm10k_intfc *interface)
1530{
1531	struct fm10k_ring_feature *f;
1532	u16 rss_i;
1533
1534	f = &interface->ring_feature[RING_F_RSS];
1535	rss_i = min_t(u16, interface->hw.mac.max_queues, f->limit);
1536
1537	/* record indices and power of 2 mask for RSS */
1538	f->indices = rss_i;
1539	f->mask = BIT(fls(rss_i - 1)) - 1;
1540
1541	interface->num_rx_queues = rss_i;
1542	interface->num_tx_queues = rss_i;
1543
1544	return true;
1545}
1546
1547/**
1548 * fm10k_set_num_queues: Allocate queues for device, feature dependent
1549 * @interface: board private structure to initialize
1550 *
1551 * This is the top level queue allocation routine.  The order here is very
1552 * important, starting with the "most" number of features turned on at once,
1553 * and ending with the smallest set of features.  This way large combinations
1554 * can be allocated if they're turned on, and smaller combinations are the
1555 * fall through conditions.
1556 *
1557 **/
1558static void fm10k_set_num_queues(struct fm10k_intfc *interface)
1559{
1560	/* Attempt to setup QoS and RSS first */
1561	if (fm10k_set_qos_queues(interface))
1562		return;
1563
1564	/* If we don't have QoS, just fallback to only RSS. */
1565	fm10k_set_rss_queues(interface);
1566}
1567
1568/**
1569 * fm10k_reset_num_queues - Reset the number of queues to zero
1570 * @interface: board private structure
1571 *
1572 * This function should be called whenever we need to reset the number of
1573 * queues after an error condition.
1574 */
1575static void fm10k_reset_num_queues(struct fm10k_intfc *interface)
1576{
1577	interface->num_tx_queues = 0;
1578	interface->num_rx_queues = 0;
1579	interface->num_q_vectors = 0;
1580}
1581
1582/**
1583 * fm10k_alloc_q_vector - Allocate memory for a single interrupt vector
1584 * @interface: board private structure to initialize
1585 * @v_count: q_vectors allocated on interface, used for ring interleaving
1586 * @v_idx: index of vector in interface struct
1587 * @txr_count: total number of Tx rings to allocate
1588 * @txr_idx: index of first Tx ring to allocate
1589 * @rxr_count: total number of Rx rings to allocate
1590 * @rxr_idx: index of first Rx ring to allocate
1591 *
1592 * We allocate one q_vector.  If allocation fails we return -ENOMEM.
1593 **/
1594static int fm10k_alloc_q_vector(struct fm10k_intfc *interface,
1595				unsigned int v_count, unsigned int v_idx,
1596				unsigned int txr_count, unsigned int txr_idx,
1597				unsigned int rxr_count, unsigned int rxr_idx)
1598{
1599	struct fm10k_q_vector *q_vector;
1600	struct fm10k_ring *ring;
1601	int ring_count;
1602
1603	ring_count = txr_count + rxr_count;
1604
1605	/* allocate q_vector and rings */
1606	q_vector = kzalloc(struct_size(q_vector, ring, ring_count), GFP_KERNEL);
1607	if (!q_vector)
1608		return -ENOMEM;
1609
1610	/* initialize NAPI */
1611	netif_napi_add(interface->netdev, &q_vector->napi,
1612		       fm10k_poll, NAPI_POLL_WEIGHT);
1613
1614	/* tie q_vector and interface together */
1615	interface->q_vector[v_idx] = q_vector;
1616	q_vector->interface = interface;
1617	q_vector->v_idx = v_idx;
1618
1619	/* initialize pointer to rings */
1620	ring = q_vector->ring;
1621
1622	/* save Tx ring container info */
1623	q_vector->tx.ring = ring;
1624	q_vector->tx.work_limit = FM10K_DEFAULT_TX_WORK;
1625	q_vector->tx.itr = interface->tx_itr;
1626	q_vector->tx.itr_scale = interface->hw.mac.itr_scale;
1627	q_vector->tx.count = txr_count;
1628
1629	while (txr_count) {
1630		/* assign generic ring traits */
1631		ring->dev = &interface->pdev->dev;
1632		ring->netdev = interface->netdev;
1633
1634		/* configure backlink on ring */
1635		ring->q_vector = q_vector;
1636
1637		/* apply Tx specific ring traits */
1638		ring->count = interface->tx_ring_count;
1639		ring->queue_index = txr_idx;
1640
1641		/* assign ring to interface */
1642		interface->tx_ring[txr_idx] = ring;
1643
1644		/* update count and index */
1645		txr_count--;
1646		txr_idx += v_count;
1647
1648		/* push pointer to next ring */
1649		ring++;
1650	}
1651
1652	/* save Rx ring container info */
1653	q_vector->rx.ring = ring;
1654	q_vector->rx.itr = interface->rx_itr;
1655	q_vector->rx.itr_scale = interface->hw.mac.itr_scale;
1656	q_vector->rx.count = rxr_count;
1657
1658	while (rxr_count) {
1659		/* assign generic ring traits */
1660		ring->dev = &interface->pdev->dev;
1661		ring->netdev = interface->netdev;
1662		rcu_assign_pointer(ring->l2_accel, interface->l2_accel);
1663
1664		/* configure backlink on ring */
1665		ring->q_vector = q_vector;
1666
1667		/* apply Rx specific ring traits */
1668		ring->count = interface->rx_ring_count;
1669		ring->queue_index = rxr_idx;
1670
1671		/* assign ring to interface */
1672		interface->rx_ring[rxr_idx] = ring;
1673
1674		/* update count and index */
1675		rxr_count--;
1676		rxr_idx += v_count;
1677
1678		/* push pointer to next ring */
1679		ring++;
1680	}
1681
1682	fm10k_dbg_q_vector_init(q_vector);
1683
1684	return 0;
1685}
1686
1687/**
1688 * fm10k_free_q_vector - Free memory allocated for specific interrupt vector
1689 * @interface: board private structure to initialize
1690 * @v_idx: Index of vector to be freed
1691 *
1692 * This function frees the memory allocated to the q_vector.  In addition if
1693 * NAPI is enabled it will delete any references to the NAPI struct prior
1694 * to freeing the q_vector.
1695 **/
1696static void fm10k_free_q_vector(struct fm10k_intfc *interface, int v_idx)
1697{
1698	struct fm10k_q_vector *q_vector = interface->q_vector[v_idx];
1699	struct fm10k_ring *ring;
1700
1701	fm10k_dbg_q_vector_exit(q_vector);
1702
1703	fm10k_for_each_ring(ring, q_vector->tx)
1704		interface->tx_ring[ring->queue_index] = NULL;
1705
1706	fm10k_for_each_ring(ring, q_vector->rx)
1707		interface->rx_ring[ring->queue_index] = NULL;
1708
1709	interface->q_vector[v_idx] = NULL;
1710	netif_napi_del(&q_vector->napi);
1711	kfree_rcu(q_vector, rcu);
1712}
1713
1714/**
1715 * fm10k_alloc_q_vectors - Allocate memory for interrupt vectors
1716 * @interface: board private structure to initialize
1717 *
1718 * We allocate one q_vector per queue interrupt.  If allocation fails we
1719 * return -ENOMEM.
1720 **/
1721static int fm10k_alloc_q_vectors(struct fm10k_intfc *interface)
1722{
1723	unsigned int q_vectors = interface->num_q_vectors;
1724	unsigned int rxr_remaining = interface->num_rx_queues;
1725	unsigned int txr_remaining = interface->num_tx_queues;
1726	unsigned int rxr_idx = 0, txr_idx = 0, v_idx = 0;
1727	int err;
1728
1729	if (q_vectors >= (rxr_remaining + txr_remaining)) {
1730		for (; rxr_remaining; v_idx++) {
1731			err = fm10k_alloc_q_vector(interface, q_vectors, v_idx,
1732						   0, 0, 1, rxr_idx);
1733			if (err)
1734				goto err_out;
1735
1736			/* update counts and index */
1737			rxr_remaining--;
1738			rxr_idx++;
1739		}
1740	}
1741
1742	for (; v_idx < q_vectors; v_idx++) {
1743		int rqpv = DIV_ROUND_UP(rxr_remaining, q_vectors - v_idx);
1744		int tqpv = DIV_ROUND_UP(txr_remaining, q_vectors - v_idx);
1745
1746		err = fm10k_alloc_q_vector(interface, q_vectors, v_idx,
1747					   tqpv, txr_idx,
1748					   rqpv, rxr_idx);
1749
1750		if (err)
1751			goto err_out;
1752
1753		/* update counts and index */
1754		rxr_remaining -= rqpv;
1755		txr_remaining -= tqpv;
1756		rxr_idx++;
1757		txr_idx++;
1758	}
1759
1760	return 0;
1761
1762err_out:
1763	fm10k_reset_num_queues(interface);
1764
1765	while (v_idx--)
1766		fm10k_free_q_vector(interface, v_idx);
1767
1768	return -ENOMEM;
1769}
1770
1771/**
1772 * fm10k_free_q_vectors - Free memory allocated for interrupt vectors
1773 * @interface: board private structure to initialize
1774 *
1775 * This function frees the memory allocated to the q_vectors.  In addition if
1776 * NAPI is enabled it will delete any references to the NAPI struct prior
1777 * to freeing the q_vector.
1778 **/
1779static void fm10k_free_q_vectors(struct fm10k_intfc *interface)
1780{
1781	int v_idx = interface->num_q_vectors;
1782
1783	fm10k_reset_num_queues(interface);
1784
1785	while (v_idx--)
1786		fm10k_free_q_vector(interface, v_idx);
1787}
1788
1789/**
1790 * f10k_reset_msix_capability - reset MSI-X capability
1791 * @interface: board private structure to initialize
1792 *
1793 * Reset the MSI-X capability back to its starting state
1794 **/
1795static void fm10k_reset_msix_capability(struct fm10k_intfc *interface)
1796{
1797	pci_disable_msix(interface->pdev);
1798	kfree(interface->msix_entries);
1799	interface->msix_entries = NULL;
1800}
1801
1802/**
1803 * f10k_init_msix_capability - configure MSI-X capability
1804 * @interface: board private structure to initialize
1805 *
1806 * Attempt to configure the interrupts using the best available
1807 * capabilities of the hardware and the kernel.
1808 **/
1809static int fm10k_init_msix_capability(struct fm10k_intfc *interface)
1810{
1811	struct fm10k_hw *hw = &interface->hw;
1812	int v_budget, vector;
1813
1814	/* It's easy to be greedy for MSI-X vectors, but it really
1815	 * doesn't do us much good if we have a lot more vectors
1816	 * than CPU's.  So let's be conservative and only ask for
1817	 * (roughly) the same number of vectors as there are CPU's.
1818	 * the default is to use pairs of vectors
1819	 */
1820	v_budget = max(interface->num_rx_queues, interface->num_tx_queues);
1821	v_budget = min_t(u16, v_budget, num_online_cpus());
1822
1823	/* account for vectors not related to queues */
1824	v_budget += NON_Q_VECTORS;
1825
1826	/* At the same time, hardware can only support a maximum of
1827	 * hw.mac->max_msix_vectors vectors.  With features
1828	 * such as RSS and VMDq, we can easily surpass the number of Rx and Tx
1829	 * descriptor queues supported by our device.  Thus, we cap it off in
1830	 * those rare cases where the cpu count also exceeds our vector limit.
1831	 */
1832	v_budget = min_t(int, v_budget, hw->mac.max_msix_vectors);
1833
1834	/* A failure in MSI-X entry allocation is fatal. */
1835	interface->msix_entries = kcalloc(v_budget, sizeof(struct msix_entry),
1836					  GFP_KERNEL);
1837	if (!interface->msix_entries)
1838		return -ENOMEM;
1839
1840	/* populate entry values */
1841	for (vector = 0; vector < v_budget; vector++)
1842		interface->msix_entries[vector].entry = vector;
1843
1844	/* Attempt to enable MSI-X with requested value */
1845	v_budget = pci_enable_msix_range(interface->pdev,
1846					 interface->msix_entries,
1847					 MIN_MSIX_COUNT(hw),
1848					 v_budget);
1849	if (v_budget < 0) {
1850		kfree(interface->msix_entries);
1851		interface->msix_entries = NULL;
1852		return v_budget;
1853	}
1854
1855	/* record the number of queues available for q_vectors */
1856	interface->num_q_vectors = v_budget - NON_Q_VECTORS;
1857
1858	return 0;
1859}
1860
1861/**
1862 * fm10k_cache_ring_qos - Descriptor ring to register mapping for QoS
1863 * @interface: Interface structure continaining rings and devices
1864 *
1865 * Cache the descriptor ring offsets for Qos
1866 **/
1867static bool fm10k_cache_ring_qos(struct fm10k_intfc *interface)
1868{
1869	struct net_device *dev = interface->netdev;
1870	int pc, offset, rss_i, i;
1871	u16 pc_stride = interface->ring_feature[RING_F_QOS].mask + 1;
1872	u8 num_pcs = netdev_get_num_tc(dev);
1873
1874	if (num_pcs <= 1)
1875		return false;
1876
1877	rss_i = interface->ring_feature[RING_F_RSS].indices;
1878
1879	for (pc = 0, offset = 0; pc < num_pcs; pc++, offset += rss_i) {
1880		int q_idx = pc;
1881
1882		for (i = 0; i < rss_i; i++) {
1883			interface->tx_ring[offset + i]->reg_idx = q_idx;
1884			interface->tx_ring[offset + i]->qos_pc = pc;
1885			interface->rx_ring[offset + i]->reg_idx = q_idx;
1886			interface->rx_ring[offset + i]->qos_pc = pc;
1887			q_idx += pc_stride;
1888		}
1889	}
1890
1891	return true;
1892}
1893
1894/**
1895 * fm10k_cache_ring_rss - Descriptor ring to register mapping for RSS
1896 * @interface: Interface structure continaining rings and devices
1897 *
1898 * Cache the descriptor ring offsets for RSS
1899 **/
1900static void fm10k_cache_ring_rss(struct fm10k_intfc *interface)
1901{
1902	int i;
1903
1904	for (i = 0; i < interface->num_rx_queues; i++)
1905		interface->rx_ring[i]->reg_idx = i;
1906
1907	for (i = 0; i < interface->num_tx_queues; i++)
1908		interface->tx_ring[i]->reg_idx = i;
1909}
1910
1911/**
1912 * fm10k_assign_rings - Map rings to network devices
1913 * @interface: Interface structure containing rings and devices
1914 *
1915 * This function is meant to go though and configure both the network
1916 * devices so that they contain rings, and configure the rings so that
1917 * they function with their network devices.
1918 **/
1919static void fm10k_assign_rings(struct fm10k_intfc *interface)
1920{
1921	if (fm10k_cache_ring_qos(interface))
1922		return;
1923
1924	fm10k_cache_ring_rss(interface);
1925}
1926
1927static void fm10k_init_reta(struct fm10k_intfc *interface)
1928{
1929	u16 i, rss_i = interface->ring_feature[RING_F_RSS].indices;
1930	u32 reta;
1931
1932	/* If the Rx flow indirection table has been configured manually, we
1933	 * need to maintain it when possible.
1934	 */
1935	if (netif_is_rxfh_configured(interface->netdev)) {
1936		for (i = FM10K_RETA_SIZE; i--;) {
1937			reta = interface->reta[i];
1938			if ((((reta << 24) >> 24) < rss_i) &&
1939			    (((reta << 16) >> 24) < rss_i) &&
1940			    (((reta <<  8) >> 24) < rss_i) &&
1941			    (((reta)       >> 24) < rss_i))
1942				continue;
1943
1944			/* this should never happen */
1945			dev_err(&interface->pdev->dev,
1946				"RSS indirection table assigned flows out of queue bounds. Reconfiguring.\n");
1947			goto repopulate_reta;
1948		}
1949
1950		/* do nothing if all of the elements are in bounds */
1951		return;
1952	}
1953
1954repopulate_reta:
1955	fm10k_write_reta(interface, NULL);
1956}
1957
1958/**
1959 * fm10k_init_queueing_scheme - Determine proper queueing scheme
1960 * @interface: board private structure to initialize
1961 *
1962 * We determine which queueing scheme to use based on...
1963 * - Hardware queue count (num_*_queues)
1964 *   - defined by miscellaneous hardware support/features (RSS, etc.)
1965 **/
1966int fm10k_init_queueing_scheme(struct fm10k_intfc *interface)
1967{
1968	int err;
1969
1970	/* Number of supported queues */
1971	fm10k_set_num_queues(interface);
1972
1973	/* Configure MSI-X capability */
1974	err = fm10k_init_msix_capability(interface);
1975	if (err) {
1976		dev_err(&interface->pdev->dev,
1977			"Unable to initialize MSI-X capability\n");
1978		goto err_init_msix;
1979	}
1980
1981	/* Allocate memory for queues */
1982	err = fm10k_alloc_q_vectors(interface);
1983	if (err) {
1984		dev_err(&interface->pdev->dev,
1985			"Unable to allocate queue vectors\n");
1986		goto err_alloc_q_vectors;
1987	}
1988
1989	/* Map rings to devices, and map devices to physical queues */
1990	fm10k_assign_rings(interface);
1991
1992	/* Initialize RSS redirection table */
1993	fm10k_init_reta(interface);
1994
1995	return 0;
1996
1997err_alloc_q_vectors:
1998	fm10k_reset_msix_capability(interface);
1999err_init_msix:
2000	fm10k_reset_num_queues(interface);
2001	return err;
2002}
2003
2004/**
2005 * fm10k_clear_queueing_scheme - Clear the current queueing scheme settings
2006 * @interface: board private structure to clear queueing scheme on
2007 *
2008 * We go through and clear queueing specific resources and reset the structure
2009 * to pre-load conditions
2010 **/
2011void fm10k_clear_queueing_scheme(struct fm10k_intfc *interface)
2012{
2013	fm10k_free_q_vectors(interface);
2014	fm10k_reset_msix_capability(interface);
2015}
2016