1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Huawei HiNIC PCI Express Linux driver
4 * Copyright(c) 2017 Huawei Technologies Co., Ltd
5 */
6
7#include <linux/kernel.h>
8#include <linux/netdevice.h>
9#include <linux/u64_stats_sync.h>
10#include <linux/errno.h>
11#include <linux/types.h>
12#include <linux/pci.h>
13#include <linux/device.h>
14#include <linux/dma-mapping.h>
15#include <linux/slab.h>
16#include <linux/interrupt.h>
17#include <linux/skbuff.h>
18#include <linux/smp.h>
19#include <asm/byteorder.h>
20#include <linux/ip.h>
21#include <linux/tcp.h>
22#include <linux/sctp.h>
23#include <linux/ipv6.h>
24#include <net/ipv6.h>
25#include <net/checksum.h>
26#include <net/ip6_checksum.h>
27
28#include "hinic_common.h"
29#include "hinic_hw_if.h"
30#include "hinic_hw_wqe.h"
31#include "hinic_hw_wq.h"
32#include "hinic_hw_qp.h"
33#include "hinic_hw_dev.h"
34#include "hinic_dev.h"
35#include "hinic_tx.h"
36
37#define TX_IRQ_NO_PENDING               0
38#define TX_IRQ_NO_COALESC               0
39#define TX_IRQ_NO_LLI_TIMER             0
40#define TX_IRQ_NO_CREDIT                0
41#define TX_IRQ_NO_RESEND_TIMER          0
42
43#define CI_UPDATE_NO_PENDING            0
44#define CI_UPDATE_NO_COALESC            0
45
46#define HW_CONS_IDX(sq)                 be16_to_cpu(*(u16 *)((sq)->hw_ci_addr))
47
48#define MIN_SKB_LEN			32
49
50#define	MAX_PAYLOAD_OFFSET	        221
51#define TRANSPORT_OFFSET(l4_hdr, skb)	((u32)((l4_hdr) - (skb)->data))
52
53union hinic_l3 {
54	struct iphdr *v4;
55	struct ipv6hdr *v6;
56	unsigned char *hdr;
57};
58
59union hinic_l4 {
60	struct tcphdr *tcp;
61	struct udphdr *udp;
62	unsigned char *hdr;
63};
64
65enum hinic_offload_type {
66	TX_OFFLOAD_TSO     = BIT(0),
67	TX_OFFLOAD_CSUM    = BIT(1),
68	TX_OFFLOAD_VLAN    = BIT(2),
69	TX_OFFLOAD_INVALID = BIT(3),
70};
71
72/**
73 * hinic_txq_clean_stats - Clean the statistics of specific queue
74 * @txq: Logical Tx Queue
75 **/
76void hinic_txq_clean_stats(struct hinic_txq *txq)
77{
78	struct hinic_txq_stats *txq_stats = &txq->txq_stats;
79
80	u64_stats_update_begin(&txq_stats->syncp);
81	txq_stats->pkts    = 0;
82	txq_stats->bytes   = 0;
83	txq_stats->tx_busy = 0;
84	txq_stats->tx_wake = 0;
85	txq_stats->tx_dropped = 0;
86	txq_stats->big_frags_pkts = 0;
87	u64_stats_update_end(&txq_stats->syncp);
88}
89
90/**
91 * hinic_txq_get_stats - get statistics of Tx Queue
92 * @txq: Logical Tx Queue
93 * @stats: return updated stats here
94 **/
95void hinic_txq_get_stats(struct hinic_txq *txq, struct hinic_txq_stats *stats)
96{
97	struct hinic_txq_stats *txq_stats = &txq->txq_stats;
98	unsigned int start;
99
100	do {
101		start = u64_stats_fetch_begin_irq(&txq_stats->syncp);
102		stats->pkts    = txq_stats->pkts;
103		stats->bytes   = txq_stats->bytes;
104		stats->tx_busy = txq_stats->tx_busy;
105		stats->tx_wake = txq_stats->tx_wake;
106		stats->tx_dropped = txq_stats->tx_dropped;
107		stats->big_frags_pkts = txq_stats->big_frags_pkts;
108	} while (u64_stats_fetch_retry_irq(&txq_stats->syncp, start));
109}
110
111/**
112 * txq_stats_init - Initialize the statistics of specific queue
113 * @txq: Logical Tx Queue
114 **/
115static void txq_stats_init(struct hinic_txq *txq)
116{
117	struct hinic_txq_stats *txq_stats = &txq->txq_stats;
118
119	u64_stats_init(&txq_stats->syncp);
120	hinic_txq_clean_stats(txq);
121}
122
123/**
124 * tx_map_skb - dma mapping for skb and return sges
125 * @nic_dev: nic device
126 * @skb: the skb
127 * @sges: returned sges
128 *
129 * Return 0 - Success, negative - Failure
130 **/
131static int tx_map_skb(struct hinic_dev *nic_dev, struct sk_buff *skb,
132		      struct hinic_sge *sges)
133{
134	struct hinic_hwdev *hwdev = nic_dev->hwdev;
135	struct hinic_hwif *hwif = hwdev->hwif;
136	struct pci_dev *pdev = hwif->pdev;
137	skb_frag_t *frag;
138	dma_addr_t dma_addr;
139	int i, j;
140
141	dma_addr = dma_map_single(&pdev->dev, skb->data, skb_headlen(skb),
142				  DMA_TO_DEVICE);
143	if (dma_mapping_error(&pdev->dev, dma_addr)) {
144		dev_err(&pdev->dev, "Failed to map Tx skb data\n");
145		return -EFAULT;
146	}
147
148	hinic_set_sge(&sges[0], dma_addr, skb_headlen(skb));
149
150	for (i = 0 ; i < skb_shinfo(skb)->nr_frags; i++) {
151		frag = &skb_shinfo(skb)->frags[i];
152
153		dma_addr = skb_frag_dma_map(&pdev->dev, frag, 0,
154					    skb_frag_size(frag),
155					    DMA_TO_DEVICE);
156		if (dma_mapping_error(&pdev->dev, dma_addr)) {
157			dev_err(&pdev->dev, "Failed to map Tx skb frag\n");
158			goto err_tx_map;
159		}
160
161		hinic_set_sge(&sges[i + 1], dma_addr, skb_frag_size(frag));
162	}
163
164	return 0;
165
166err_tx_map:
167	for (j = 0; j < i; j++)
168		dma_unmap_page(&pdev->dev, hinic_sge_to_dma(&sges[j + 1]),
169			       sges[j + 1].len, DMA_TO_DEVICE);
170
171	dma_unmap_single(&pdev->dev, hinic_sge_to_dma(&sges[0]), sges[0].len,
172			 DMA_TO_DEVICE);
173	return -EFAULT;
174}
175
176/**
177 * tx_unmap_skb - unmap the dma address of the skb
178 * @nic_dev: nic device
179 * @skb: the skb
180 * @sges: the sges that are connected to the skb
181 **/
182static void tx_unmap_skb(struct hinic_dev *nic_dev, struct sk_buff *skb,
183			 struct hinic_sge *sges)
184{
185	struct hinic_hwdev *hwdev = nic_dev->hwdev;
186	struct hinic_hwif *hwif = hwdev->hwif;
187	struct pci_dev *pdev = hwif->pdev;
188	int i;
189
190	for (i = 0; i < skb_shinfo(skb)->nr_frags ; i++)
191		dma_unmap_page(&pdev->dev, hinic_sge_to_dma(&sges[i + 1]),
192			       sges[i + 1].len, DMA_TO_DEVICE);
193
194	dma_unmap_single(&pdev->dev, hinic_sge_to_dma(&sges[0]), sges[0].len,
195			 DMA_TO_DEVICE);
196}
197
198static void get_inner_l3_l4_type(struct sk_buff *skb, union hinic_l3 *ip,
199				 union hinic_l4 *l4,
200				 enum hinic_offload_type offload_type,
201				 enum hinic_l3_offload_type *l3_type,
202				 u8 *l4_proto)
203{
204	u8 *exthdr;
205
206	if (ip->v4->version == 4) {
207		*l3_type = (offload_type == TX_OFFLOAD_CSUM) ?
208			   IPV4_PKT_NO_CHKSUM_OFFLOAD :
209			   IPV4_PKT_WITH_CHKSUM_OFFLOAD;
210		*l4_proto = ip->v4->protocol;
211	} else if (ip->v4->version == 6) {
212		*l3_type = IPV6_PKT;
213		exthdr = ip->hdr + sizeof(*ip->v6);
214		*l4_proto = ip->v6->nexthdr;
215		if (exthdr != l4->hdr) {
216			int start = exthdr - skb->data;
217			__be16 frag_off;
218
219			ipv6_skip_exthdr(skb, start, l4_proto, &frag_off);
220		}
221	} else {
222		*l3_type = L3TYPE_UNKNOWN;
223		*l4_proto = 0;
224	}
225}
226
227static void get_inner_l4_info(struct sk_buff *skb, union hinic_l4 *l4,
228			      enum hinic_offload_type offload_type, u8 l4_proto,
229			      enum hinic_l4_offload_type *l4_offload,
230			      u32 *l4_len, u32 *offset)
231{
232	*l4_offload = OFFLOAD_DISABLE;
233	*offset = 0;
234	*l4_len = 0;
235
236	switch (l4_proto) {
237	case IPPROTO_TCP:
238		*l4_offload = TCP_OFFLOAD_ENABLE;
239		/* doff in unit of 4B */
240		*l4_len = l4->tcp->doff * 4;
241		*offset = *l4_len + TRANSPORT_OFFSET(l4->hdr, skb);
242		break;
243
244	case IPPROTO_UDP:
245		*l4_offload = UDP_OFFLOAD_ENABLE;
246		*l4_len = sizeof(struct udphdr);
247		*offset = TRANSPORT_OFFSET(l4->hdr, skb);
248		break;
249
250	case IPPROTO_SCTP:
251		/* only csum offload support sctp */
252		if (offload_type != TX_OFFLOAD_CSUM)
253			break;
254
255		*l4_offload = SCTP_OFFLOAD_ENABLE;
256		*l4_len = sizeof(struct sctphdr);
257		*offset = TRANSPORT_OFFSET(l4->hdr, skb);
258		break;
259
260	default:
261		break;
262	}
263}
264
265static __sum16 csum_magic(union hinic_l3 *ip, unsigned short proto)
266{
267	return (ip->v4->version == 4) ?
268		csum_tcpudp_magic(ip->v4->saddr, ip->v4->daddr, 0, proto, 0) :
269		csum_ipv6_magic(&ip->v6->saddr, &ip->v6->daddr, 0, proto, 0);
270}
271
272static int offload_tso(struct hinic_sq_task *task, u32 *queue_info,
273		       struct sk_buff *skb)
274{
275	u32 offset, l4_len, ip_identify, network_hdr_len;
276	enum hinic_l3_offload_type l3_offload;
277	enum hinic_l4_offload_type l4_offload;
278	union hinic_l3 ip;
279	union hinic_l4 l4;
280	u8 l4_proto;
281
282	if (!skb_is_gso(skb))
283		return 0;
284
285	if (skb_cow_head(skb, 0) < 0)
286		return -EPROTONOSUPPORT;
287
288	if (skb->encapsulation) {
289		u32 gso_type = skb_shinfo(skb)->gso_type;
290		u32 tunnel_type = 0;
291		u32 l4_tunnel_len;
292
293		ip.hdr = skb_network_header(skb);
294		l4.hdr = skb_transport_header(skb);
295		network_hdr_len = skb_inner_network_header_len(skb);
296
297		if (ip.v4->version == 4) {
298			ip.v4->tot_len = 0;
299			l3_offload = IPV4_PKT_WITH_CHKSUM_OFFLOAD;
300		} else if (ip.v4->version == 6) {
301			l3_offload = IPV6_PKT;
302		} else {
303			l3_offload = 0;
304		}
305
306		hinic_task_set_outter_l3(task, l3_offload,
307					 skb_network_header_len(skb));
308
309		if (gso_type & SKB_GSO_UDP_TUNNEL_CSUM) {
310			l4.udp->check = ~csum_magic(&ip, IPPROTO_UDP);
311			tunnel_type = TUNNEL_UDP_CSUM;
312		} else if (gso_type & SKB_GSO_UDP_TUNNEL) {
313			tunnel_type = TUNNEL_UDP_NO_CSUM;
314		}
315
316		l4_tunnel_len = skb_inner_network_offset(skb) -
317				skb_transport_offset(skb);
318		hinic_task_set_tunnel_l4(task, tunnel_type, l4_tunnel_len);
319
320		ip.hdr = skb_inner_network_header(skb);
321		l4.hdr = skb_inner_transport_header(skb);
322	} else {
323		ip.hdr = skb_network_header(skb);
324		l4.hdr = skb_transport_header(skb);
325		network_hdr_len = skb_network_header_len(skb);
326	}
327
328	/* initialize inner IP header fields */
329	if (ip.v4->version == 4)
330		ip.v4->tot_len = 0;
331	else
332		ip.v6->payload_len = 0;
333
334	get_inner_l3_l4_type(skb, &ip, &l4, TX_OFFLOAD_TSO, &l3_offload,
335			     &l4_proto);
336
337	hinic_task_set_inner_l3(task, l3_offload, network_hdr_len);
338
339	ip_identify = 0;
340	if (l4_proto == IPPROTO_TCP)
341		l4.tcp->check = ~csum_magic(&ip, IPPROTO_TCP);
342
343	get_inner_l4_info(skb, &l4, TX_OFFLOAD_TSO, l4_proto, &l4_offload,
344			  &l4_len, &offset);
345
346	hinic_set_tso_inner_l4(task, queue_info, l4_offload, l4_len, offset,
347			       ip_identify, skb_shinfo(skb)->gso_size);
348
349	return 1;
350}
351
352static int offload_csum(struct hinic_sq_task *task, u32 *queue_info,
353			struct sk_buff *skb)
354{
355	enum hinic_l4_offload_type l4_offload;
356	u32 offset, l4_len, network_hdr_len;
357	enum hinic_l3_offload_type l3_type;
358	u32 tunnel_type = NOT_TUNNEL;
359	union hinic_l3 ip;
360	union hinic_l4 l4;
361	u8 l4_proto;
362
363	if (skb->ip_summed != CHECKSUM_PARTIAL)
364		return 0;
365
366	if (skb->encapsulation) {
367		u32 l4_tunnel_len;
368
369		tunnel_type = TUNNEL_UDP_NO_CSUM;
370		ip.hdr = skb_network_header(skb);
371
372		if (ip.v4->version == 4) {
373			l3_type = IPV4_PKT_NO_CHKSUM_OFFLOAD;
374			l4_proto = ip.v4->protocol;
375		} else if (ip.v4->version == 6) {
376			unsigned char *exthdr;
377			__be16 frag_off;
378			l3_type = IPV6_PKT;
379			tunnel_type = TUNNEL_UDP_CSUM;
380			exthdr = ip.hdr + sizeof(*ip.v6);
381			l4_proto = ip.v6->nexthdr;
382			l4.hdr = skb_transport_header(skb);
383			if (l4.hdr != exthdr)
384				ipv6_skip_exthdr(skb, exthdr - skb->data,
385						 &l4_proto, &frag_off);
386		} else {
387			l3_type = L3TYPE_UNKNOWN;
388			l4_proto = IPPROTO_RAW;
389		}
390
391		hinic_task_set_outter_l3(task, l3_type,
392					 skb_network_header_len(skb));
393
394		switch (l4_proto) {
395		case IPPROTO_UDP:
396			l4_tunnel_len = skb_inner_network_offset(skb) -
397					skb_transport_offset(skb);
398			ip.hdr = skb_inner_network_header(skb);
399			l4.hdr = skb_inner_transport_header(skb);
400			network_hdr_len = skb_inner_network_header_len(skb);
401			break;
402		case IPPROTO_IPIP:
403		case IPPROTO_IPV6:
404			tunnel_type = NOT_TUNNEL;
405			l4_tunnel_len = 0;
406
407			ip.hdr = skb_inner_network_header(skb);
408			l4.hdr = skb_transport_header(skb);
409			network_hdr_len = skb_network_header_len(skb);
410			break;
411		default:
412			/* Unsupported tunnel packet, disable csum offload */
413			skb_checksum_help(skb);
414			return 0;
415		}
416
417		hinic_task_set_tunnel_l4(task, tunnel_type, l4_tunnel_len);
418	} else {
419		ip.hdr = skb_network_header(skb);
420		l4.hdr = skb_transport_header(skb);
421		network_hdr_len = skb_network_header_len(skb);
422	}
423
424	get_inner_l3_l4_type(skb, &ip, &l4, TX_OFFLOAD_CSUM, &l3_type,
425			     &l4_proto);
426
427	hinic_task_set_inner_l3(task, l3_type, network_hdr_len);
428
429	get_inner_l4_info(skb, &l4, TX_OFFLOAD_CSUM, l4_proto, &l4_offload,
430			  &l4_len, &offset);
431
432	hinic_set_cs_inner_l4(task, queue_info, l4_offload, l4_len, offset);
433
434	return 1;
435}
436
437static void offload_vlan(struct hinic_sq_task *task, u32 *queue_info,
438			 u16 vlan_tag, u16 vlan_pri)
439{
440	task->pkt_info0 |= HINIC_SQ_TASK_INFO0_SET(vlan_tag, VLAN_TAG) |
441				HINIC_SQ_TASK_INFO0_SET(1U, VLAN_OFFLOAD);
442
443	*queue_info |= HINIC_SQ_CTRL_SET(vlan_pri, QUEUE_INFO_PRI);
444}
445
446static int hinic_tx_offload(struct sk_buff *skb, struct hinic_sq_task *task,
447			    u32 *queue_info)
448{
449	enum hinic_offload_type offload = 0;
450	u16 vlan_tag;
451	int enabled;
452
453	enabled = offload_tso(task, queue_info, skb);
454	if (enabled > 0) {
455		offload |= TX_OFFLOAD_TSO;
456	} else if (enabled == 0) {
457		enabled = offload_csum(task, queue_info, skb);
458		if (enabled)
459			offload |= TX_OFFLOAD_CSUM;
460	} else {
461		return -EPROTONOSUPPORT;
462	}
463
464	if (unlikely(skb_vlan_tag_present(skb))) {
465		vlan_tag = skb_vlan_tag_get(skb);
466		offload_vlan(task, queue_info, vlan_tag,
467			     vlan_tag >> VLAN_PRIO_SHIFT);
468		offload |= TX_OFFLOAD_VLAN;
469	}
470
471	if (offload)
472		hinic_task_set_l2hdr(task, skb_network_offset(skb));
473
474	/* payload offset should not more than 221 */
475	if (HINIC_SQ_CTRL_GET(*queue_info, QUEUE_INFO_PLDOFF) >
476	    MAX_PAYLOAD_OFFSET) {
477		return -EPROTONOSUPPORT;
478	}
479
480	/* mss should not less than 80 */
481	if (HINIC_SQ_CTRL_GET(*queue_info, QUEUE_INFO_MSS) < HINIC_MSS_MIN) {
482		*queue_info = HINIC_SQ_CTRL_CLEAR(*queue_info, QUEUE_INFO_MSS);
483		*queue_info |= HINIC_SQ_CTRL_SET(HINIC_MSS_MIN, QUEUE_INFO_MSS);
484	}
485
486	return 0;
487}
488
489netdev_tx_t hinic_lb_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
490{
491	struct hinic_dev *nic_dev = netdev_priv(netdev);
492	u16 prod_idx, q_id = skb->queue_mapping;
493	struct netdev_queue *netdev_txq;
494	int nr_sges, err = NETDEV_TX_OK;
495	struct hinic_sq_wqe *sq_wqe;
496	unsigned int wqe_size;
497	struct hinic_txq *txq;
498	struct hinic_qp *qp;
499
500	txq = &nic_dev->txqs[q_id];
501	qp = container_of(txq->sq, struct hinic_qp, sq);
502	nr_sges = skb_shinfo(skb)->nr_frags + 1;
503
504	err = tx_map_skb(nic_dev, skb, txq->sges);
505	if (err)
506		goto skb_error;
507
508	wqe_size = HINIC_SQ_WQE_SIZE(nr_sges);
509
510	sq_wqe = hinic_sq_get_wqe(txq->sq, wqe_size, &prod_idx);
511	if (!sq_wqe) {
512		netif_stop_subqueue(netdev, qp->q_id);
513
514		sq_wqe = hinic_sq_get_wqe(txq->sq, wqe_size, &prod_idx);
515		if (sq_wqe) {
516			netif_wake_subqueue(nic_dev->netdev, qp->q_id);
517			goto process_sq_wqe;
518		}
519
520		tx_unmap_skb(nic_dev, skb, txq->sges);
521
522		u64_stats_update_begin(&txq->txq_stats.syncp);
523		txq->txq_stats.tx_busy++;
524		u64_stats_update_end(&txq->txq_stats.syncp);
525		err = NETDEV_TX_BUSY;
526		wqe_size = 0;
527		goto flush_skbs;
528	}
529
530process_sq_wqe:
531	hinic_sq_prepare_wqe(txq->sq, prod_idx, sq_wqe, txq->sges, nr_sges);
532	hinic_sq_write_wqe(txq->sq, prod_idx, sq_wqe, skb, wqe_size);
533
534flush_skbs:
535	netdev_txq = netdev_get_tx_queue(netdev, q_id);
536	if ((!netdev_xmit_more()) || (netif_xmit_stopped(netdev_txq)))
537		hinic_sq_write_db(txq->sq, prod_idx, wqe_size, 0);
538
539	return err;
540
541skb_error:
542	dev_kfree_skb_any(skb);
543	u64_stats_update_begin(&txq->txq_stats.syncp);
544	txq->txq_stats.tx_dropped++;
545	u64_stats_update_end(&txq->txq_stats.syncp);
546
547	return NETDEV_TX_OK;
548}
549
550netdev_tx_t hinic_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
551{
552	struct hinic_dev *nic_dev = netdev_priv(netdev);
553	u16 prod_idx, q_id = skb->queue_mapping;
554	struct netdev_queue *netdev_txq;
555	int nr_sges, err = NETDEV_TX_OK;
556	struct hinic_sq_wqe *sq_wqe;
557	unsigned int wqe_size;
558	struct hinic_txq *txq;
559	struct hinic_qp *qp;
560
561	txq = &nic_dev->txqs[q_id];
562	qp = container_of(txq->sq, struct hinic_qp, sq);
563
564	if (skb->len < MIN_SKB_LEN) {
565		if (skb_pad(skb, MIN_SKB_LEN - skb->len)) {
566			netdev_err(netdev, "Failed to pad skb\n");
567			goto update_error_stats;
568		}
569
570		skb->len = MIN_SKB_LEN;
571	}
572
573	nr_sges = skb_shinfo(skb)->nr_frags + 1;
574	if (nr_sges > 17) {
575		u64_stats_update_begin(&txq->txq_stats.syncp);
576		txq->txq_stats.big_frags_pkts++;
577		u64_stats_update_end(&txq->txq_stats.syncp);
578	}
579
580	if (nr_sges > txq->max_sges) {
581		netdev_err(netdev, "Too many Tx sges\n");
582		goto skb_error;
583	}
584
585	err = tx_map_skb(nic_dev, skb, txq->sges);
586	if (err)
587		goto skb_error;
588
589	wqe_size = HINIC_SQ_WQE_SIZE(nr_sges);
590
591	sq_wqe = hinic_sq_get_wqe(txq->sq, wqe_size, &prod_idx);
592	if (!sq_wqe) {
593		netif_stop_subqueue(netdev, qp->q_id);
594
595		/* Check for the case free_tx_poll is called in another cpu
596		 * and we stopped the subqueue after free_tx_poll check.
597		 */
598		sq_wqe = hinic_sq_get_wqe(txq->sq, wqe_size, &prod_idx);
599		if (sq_wqe) {
600			netif_wake_subqueue(nic_dev->netdev, qp->q_id);
601			goto process_sq_wqe;
602		}
603
604		tx_unmap_skb(nic_dev, skb, txq->sges);
605
606		u64_stats_update_begin(&txq->txq_stats.syncp);
607		txq->txq_stats.tx_busy++;
608		u64_stats_update_end(&txq->txq_stats.syncp);
609		err = NETDEV_TX_BUSY;
610		wqe_size = 0;
611		goto flush_skbs;
612	}
613
614process_sq_wqe:
615	hinic_sq_prepare_wqe(txq->sq, prod_idx, sq_wqe, txq->sges, nr_sges);
616
617	err = hinic_tx_offload(skb, &sq_wqe->task, &sq_wqe->ctrl.queue_info);
618	if (err)
619		goto offload_error;
620
621	hinic_sq_write_wqe(txq->sq, prod_idx, sq_wqe, skb, wqe_size);
622
623flush_skbs:
624	netdev_txq = netdev_get_tx_queue(netdev, q_id);
625	if ((!netdev_xmit_more()) || (netif_xmit_stopped(netdev_txq)))
626		hinic_sq_write_db(txq->sq, prod_idx, wqe_size, 0);
627
628	return err;
629
630offload_error:
631	hinic_sq_return_wqe(txq->sq, wqe_size);
632	tx_unmap_skb(nic_dev, skb, txq->sges);
633
634skb_error:
635	dev_kfree_skb_any(skb);
636
637update_error_stats:
638	u64_stats_update_begin(&txq->txq_stats.syncp);
639	txq->txq_stats.tx_dropped++;
640	u64_stats_update_end(&txq->txq_stats.syncp);
641
642	return NETDEV_TX_OK;
643}
644
645/**
646 * tx_free_skb - unmap and free skb
647 * @nic_dev: nic device
648 * @skb: the skb
649 * @sges: the sges that are connected to the skb
650 **/
651static void tx_free_skb(struct hinic_dev *nic_dev, struct sk_buff *skb,
652			struct hinic_sge *sges)
653{
654	tx_unmap_skb(nic_dev, skb, sges);
655
656	dev_kfree_skb_any(skb);
657}
658
659/**
660 * free_all_rx_skbs - free all skbs in tx queue
661 * @txq: tx queue
662 **/
663static void free_all_tx_skbs(struct hinic_txq *txq)
664{
665	struct hinic_dev *nic_dev = netdev_priv(txq->netdev);
666	struct hinic_sq *sq = txq->sq;
667	struct hinic_sq_wqe *sq_wqe;
668	unsigned int wqe_size;
669	struct sk_buff *skb;
670	int nr_sges;
671	u16 ci;
672
673	while ((sq_wqe = hinic_sq_read_wqebb(sq, &skb, &wqe_size, &ci))) {
674		sq_wqe = hinic_sq_read_wqe(sq, &skb, wqe_size, &ci);
675		if (!sq_wqe)
676			break;
677
678		nr_sges = skb_shinfo(skb)->nr_frags + 1;
679
680		hinic_sq_get_sges(sq_wqe, txq->free_sges, nr_sges);
681
682		hinic_sq_put_wqe(sq, wqe_size);
683
684		tx_free_skb(nic_dev, skb, txq->free_sges);
685	}
686}
687
688/**
689 * free_tx_poll - free finished tx skbs in tx queue that connected to napi
690 * @napi: napi
691 * @budget: number of tx
692 *
693 * Return 0 - Success, negative - Failure
694 **/
695static int free_tx_poll(struct napi_struct *napi, int budget)
696{
697	struct hinic_txq *txq = container_of(napi, struct hinic_txq, napi);
698	struct hinic_qp *qp = container_of(txq->sq, struct hinic_qp, sq);
699	struct hinic_dev *nic_dev = netdev_priv(txq->netdev);
700	struct netdev_queue *netdev_txq;
701	struct hinic_sq *sq = txq->sq;
702	struct hinic_wq *wq = sq->wq;
703	struct hinic_sq_wqe *sq_wqe;
704	unsigned int wqe_size;
705	int nr_sges, pkts = 0;
706	struct sk_buff *skb;
707	u64 tx_bytes = 0;
708	u16 hw_ci, sw_ci;
709
710	do {
711		hw_ci = HW_CONS_IDX(sq) & wq->mask;
712
713		dma_rmb();
714
715		/* Reading a WQEBB to get real WQE size and consumer index. */
716		sq_wqe = hinic_sq_read_wqebb(sq, &skb, &wqe_size, &sw_ci);
717		if ((!sq_wqe) ||
718		    (((hw_ci - sw_ci) & wq->mask) * wq->wqebb_size < wqe_size))
719			break;
720
721		/* If this WQE have multiple WQEBBs, we will read again to get
722		 * full size WQE.
723		 */
724		if (wqe_size > wq->wqebb_size) {
725			sq_wqe = hinic_sq_read_wqe(sq, &skb, wqe_size, &sw_ci);
726			if (unlikely(!sq_wqe))
727				break;
728		}
729
730		tx_bytes += skb->len;
731		pkts++;
732
733		nr_sges = skb_shinfo(skb)->nr_frags + 1;
734
735		hinic_sq_get_sges(sq_wqe, txq->free_sges, nr_sges);
736
737		hinic_sq_put_wqe(sq, wqe_size);
738
739		tx_free_skb(nic_dev, skb, txq->free_sges);
740	} while (pkts < budget);
741
742	if (__netif_subqueue_stopped(nic_dev->netdev, qp->q_id) &&
743	    hinic_get_sq_free_wqebbs(sq) >= HINIC_MIN_TX_NUM_WQEBBS(sq)) {
744		netdev_txq = netdev_get_tx_queue(txq->netdev, qp->q_id);
745
746		__netif_tx_lock(netdev_txq, smp_processor_id());
747		if (!netif_testing(nic_dev->netdev))
748			netif_wake_subqueue(nic_dev->netdev, qp->q_id);
749
750		__netif_tx_unlock(netdev_txq);
751
752		u64_stats_update_begin(&txq->txq_stats.syncp);
753		txq->txq_stats.tx_wake++;
754		u64_stats_update_end(&txq->txq_stats.syncp);
755	}
756
757	u64_stats_update_begin(&txq->txq_stats.syncp);
758	txq->txq_stats.bytes += tx_bytes;
759	txq->txq_stats.pkts += pkts;
760	u64_stats_update_end(&txq->txq_stats.syncp);
761
762	if (pkts < budget) {
763		napi_complete(napi);
764		if (!HINIC_IS_VF(nic_dev->hwdev->hwif))
765			hinic_hwdev_set_msix_state(nic_dev->hwdev,
766						   sq->msix_entry,
767						   HINIC_MSIX_ENABLE);
768
769		return pkts;
770	}
771
772	return budget;
773}
774
775static irqreturn_t tx_irq(int irq, void *data)
776{
777	struct hinic_txq *txq = data;
778	struct hinic_dev *nic_dev;
779
780	nic_dev = netdev_priv(txq->netdev);
781
782	if (!HINIC_IS_VF(nic_dev->hwdev->hwif))
783		/* Disable the interrupt until napi will be completed */
784		hinic_hwdev_set_msix_state(nic_dev->hwdev,
785					   txq->sq->msix_entry,
786					   HINIC_MSIX_DISABLE);
787
788	hinic_hwdev_msix_cnt_set(nic_dev->hwdev, txq->sq->msix_entry);
789
790	napi_schedule(&txq->napi);
791	return IRQ_HANDLED;
792}
793
794static int tx_request_irq(struct hinic_txq *txq)
795{
796	struct hinic_dev *nic_dev = netdev_priv(txq->netdev);
797	struct hinic_msix_config interrupt_info = {0};
798	struct hinic_intr_coal_info *intr_coal = NULL;
799	struct hinic_hwdev *hwdev = nic_dev->hwdev;
800	struct hinic_hwif *hwif = hwdev->hwif;
801	struct pci_dev *pdev = hwif->pdev;
802	struct hinic_sq *sq = txq->sq;
803	struct hinic_qp *qp;
804	int err;
805
806	qp = container_of(sq, struct hinic_qp, sq);
807
808	netif_napi_add(txq->netdev, &txq->napi, free_tx_poll, nic_dev->tx_weight);
809
810	hinic_hwdev_msix_set(nic_dev->hwdev, sq->msix_entry,
811			     TX_IRQ_NO_PENDING, TX_IRQ_NO_COALESC,
812			     TX_IRQ_NO_LLI_TIMER, TX_IRQ_NO_CREDIT,
813			     TX_IRQ_NO_RESEND_TIMER);
814
815	intr_coal = &nic_dev->tx_intr_coalesce[qp->q_id];
816	interrupt_info.msix_index = sq->msix_entry;
817	interrupt_info.coalesce_timer_cnt = intr_coal->coalesce_timer_cfg;
818	interrupt_info.pending_cnt = intr_coal->pending_limt;
819	interrupt_info.resend_timer_cnt = intr_coal->resend_timer_cfg;
820
821	err = hinic_set_interrupt_cfg(hwdev, &interrupt_info);
822	if (err) {
823		netif_err(nic_dev, drv, txq->netdev,
824			  "Failed to set TX interrupt coalescing attribute\n");
825		netif_napi_del(&txq->napi);
826		return err;
827	}
828
829	err = request_irq(sq->irq, tx_irq, 0, txq->irq_name, txq);
830	if (err) {
831		dev_err(&pdev->dev, "Failed to request Tx irq\n");
832		netif_napi_del(&txq->napi);
833		return err;
834	}
835
836	return 0;
837}
838
839static void tx_free_irq(struct hinic_txq *txq)
840{
841	struct hinic_sq *sq = txq->sq;
842
843	free_irq(sq->irq, txq);
844	netif_napi_del(&txq->napi);
845}
846
847/**
848 * hinic_init_txq - Initialize the Tx Queue
849 * @txq: Logical Tx Queue
850 * @sq: Hardware Tx Queue to connect the Logical queue with
851 * @netdev: network device to connect the Logical queue with
852 *
853 * Return 0 - Success, negative - Failure
854 **/
855int hinic_init_txq(struct hinic_txq *txq, struct hinic_sq *sq,
856		   struct net_device *netdev)
857{
858	struct hinic_qp *qp = container_of(sq, struct hinic_qp, sq);
859	struct hinic_dev *nic_dev = netdev_priv(netdev);
860	struct hinic_hwdev *hwdev = nic_dev->hwdev;
861	int err, irqname_len;
862
863	txq->netdev = netdev;
864	txq->sq = sq;
865
866	txq_stats_init(txq);
867
868	txq->max_sges = HINIC_MAX_SQ_BUFDESCS;
869
870	txq->sges = devm_kcalloc(&netdev->dev, txq->max_sges,
871				 sizeof(*txq->sges), GFP_KERNEL);
872	if (!txq->sges)
873		return -ENOMEM;
874
875	txq->free_sges = devm_kcalloc(&netdev->dev, txq->max_sges,
876				      sizeof(*txq->free_sges), GFP_KERNEL);
877	if (!txq->free_sges) {
878		err = -ENOMEM;
879		goto err_alloc_free_sges;
880	}
881
882	irqname_len = snprintf(NULL, 0, "%s_txq%d", netdev->name, qp->q_id) + 1;
883	txq->irq_name = devm_kzalloc(&netdev->dev, irqname_len, GFP_KERNEL);
884	if (!txq->irq_name) {
885		err = -ENOMEM;
886		goto err_alloc_irqname;
887	}
888
889	sprintf(txq->irq_name, "%s_txq%d", netdev->name, qp->q_id);
890
891	err = hinic_hwdev_hw_ci_addr_set(hwdev, sq, CI_UPDATE_NO_PENDING,
892					 CI_UPDATE_NO_COALESC);
893	if (err)
894		goto err_hw_ci;
895
896	err = tx_request_irq(txq);
897	if (err) {
898		netdev_err(netdev, "Failed to request Tx irq\n");
899		goto err_req_tx_irq;
900	}
901
902	return 0;
903
904err_req_tx_irq:
905err_hw_ci:
906	devm_kfree(&netdev->dev, txq->irq_name);
907
908err_alloc_irqname:
909	devm_kfree(&netdev->dev, txq->free_sges);
910
911err_alloc_free_sges:
912	devm_kfree(&netdev->dev, txq->sges);
913	return err;
914}
915
916/**
917 * hinic_clean_txq - Clean the Tx Queue
918 * @txq: Logical Tx Queue
919 **/
920void hinic_clean_txq(struct hinic_txq *txq)
921{
922	struct net_device *netdev = txq->netdev;
923
924	tx_free_irq(txq);
925
926	free_all_tx_skbs(txq);
927
928	devm_kfree(&netdev->dev, txq->irq_name);
929	devm_kfree(&netdev->dev, txq->free_sges);
930	devm_kfree(&netdev->dev, txq->sges);
931}
932