1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
2 /* Copyright (c) 2021, Microsoft Corporation. */
3
4 #include <uapi/linux/bpf.h>
5
6 #include <linux/debugfs.h>
7 #include <linux/inetdevice.h>
8 #include <linux/etherdevice.h>
9 #include <linux/ethtool.h>
10 #include <linux/filter.h>
11 #include <linux/mm.h>
12 #include <linux/pci.h>
13
14 #include <net/checksum.h>
15 #include <net/ip6_checksum.h>
16 #include <net/page_pool/helpers.h>
17 #include <net/xdp.h>
18
19 #include <net/mana/mana.h>
20 #include <net/mana/mana_auxiliary.h>
21
22 static DEFINE_IDA(mana_adev_ida);
23
mana_adev_idx_alloc(void)24 static int mana_adev_idx_alloc(void)
25 {
26 return ida_alloc(&mana_adev_ida, GFP_KERNEL);
27 }
28
mana_adev_idx_free(int idx)29 static void mana_adev_idx_free(int idx)
30 {
31 ida_free(&mana_adev_ida, idx);
32 }
33
mana_dbg_q_read(struct file * filp,char __user * buf,size_t count,loff_t * pos)34 static ssize_t mana_dbg_q_read(struct file *filp, char __user *buf, size_t count,
35 loff_t *pos)
36 {
37 struct gdma_queue *gdma_q = filp->private_data;
38
39 return simple_read_from_buffer(buf, count, pos, gdma_q->queue_mem_ptr,
40 gdma_q->queue_size);
41 }
42
43 static const struct file_operations mana_dbg_q_fops = {
44 .owner = THIS_MODULE,
45 .open = simple_open,
46 .read = mana_dbg_q_read,
47 };
48
49 /* Microsoft Azure Network Adapter (MANA) functions */
50
mana_open(struct net_device * ndev)51 static int mana_open(struct net_device *ndev)
52 {
53 struct mana_port_context *apc = netdev_priv(ndev);
54 int err;
55
56 err = mana_alloc_queues(ndev);
57 if (err)
58 return err;
59
60 apc->port_is_up = true;
61
62 /* Ensure port state updated before txq state */
63 smp_wmb();
64
65 netif_carrier_on(ndev);
66 netif_tx_wake_all_queues(ndev);
67
68 return 0;
69 }
70
mana_close(struct net_device * ndev)71 static int mana_close(struct net_device *ndev)
72 {
73 struct mana_port_context *apc = netdev_priv(ndev);
74
75 if (!apc->port_is_up)
76 return 0;
77
78 return mana_detach(ndev, true);
79 }
80
mana_can_tx(struct gdma_queue * wq)81 static bool mana_can_tx(struct gdma_queue *wq)
82 {
83 return mana_gd_wq_avail_space(wq) >= MAX_TX_WQE_SIZE;
84 }
85
mana_checksum_info(struct sk_buff * skb)86 static unsigned int mana_checksum_info(struct sk_buff *skb)
87 {
88 if (skb->protocol == htons(ETH_P_IP)) {
89 struct iphdr *ip = ip_hdr(skb);
90
91 if (ip->protocol == IPPROTO_TCP)
92 return IPPROTO_TCP;
93
94 if (ip->protocol == IPPROTO_UDP)
95 return IPPROTO_UDP;
96 } else if (skb->protocol == htons(ETH_P_IPV6)) {
97 struct ipv6hdr *ip6 = ipv6_hdr(skb);
98
99 if (ip6->nexthdr == IPPROTO_TCP)
100 return IPPROTO_TCP;
101
102 if (ip6->nexthdr == IPPROTO_UDP)
103 return IPPROTO_UDP;
104 }
105
106 /* No csum offloading */
107 return 0;
108 }
109
mana_add_sge(struct mana_tx_package * tp,struct mana_skb_head * ash,int sg_i,dma_addr_t da,int sge_len,u32 gpa_mkey)110 static void mana_add_sge(struct mana_tx_package *tp, struct mana_skb_head *ash,
111 int sg_i, dma_addr_t da, int sge_len, u32 gpa_mkey)
112 {
113 ash->dma_handle[sg_i] = da;
114 ash->size[sg_i] = sge_len;
115
116 tp->wqe_req.sgl[sg_i].address = da;
117 tp->wqe_req.sgl[sg_i].mem_key = gpa_mkey;
118 tp->wqe_req.sgl[sg_i].size = sge_len;
119 }
120
mana_map_skb(struct sk_buff * skb,struct mana_port_context * apc,struct mana_tx_package * tp,int gso_hs)121 static int mana_map_skb(struct sk_buff *skb, struct mana_port_context *apc,
122 struct mana_tx_package *tp, int gso_hs)
123 {
124 struct mana_skb_head *ash = (struct mana_skb_head *)skb->head;
125 int hsg = 1; /* num of SGEs of linear part */
126 struct gdma_dev *gd = apc->ac->gdma_dev;
127 int skb_hlen = skb_headlen(skb);
128 int sge0_len, sge1_len = 0;
129 struct gdma_context *gc;
130 struct device *dev;
131 skb_frag_t *frag;
132 dma_addr_t da;
133 int sg_i;
134 int i;
135
136 gc = gd->gdma_context;
137 dev = gc->dev;
138
139 if (gso_hs && gso_hs < skb_hlen) {
140 sge0_len = gso_hs;
141 sge1_len = skb_hlen - gso_hs;
142 } else {
143 sge0_len = skb_hlen;
144 }
145
146 da = dma_map_single(dev, skb->data, sge0_len, DMA_TO_DEVICE);
147 if (dma_mapping_error(dev, da))
148 return -ENOMEM;
149
150 mana_add_sge(tp, ash, 0, da, sge0_len, gd->gpa_mkey);
151
152 if (sge1_len) {
153 sg_i = 1;
154 da = dma_map_single(dev, skb->data + sge0_len, sge1_len,
155 DMA_TO_DEVICE);
156 if (dma_mapping_error(dev, da))
157 goto frag_err;
158
159 mana_add_sge(tp, ash, sg_i, da, sge1_len, gd->gpa_mkey);
160 hsg = 2;
161 }
162
163 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
164 sg_i = hsg + i;
165
166 frag = &skb_shinfo(skb)->frags[i];
167 da = skb_frag_dma_map(dev, frag, 0, skb_frag_size(frag),
168 DMA_TO_DEVICE);
169 if (dma_mapping_error(dev, da))
170 goto frag_err;
171
172 mana_add_sge(tp, ash, sg_i, da, skb_frag_size(frag),
173 gd->gpa_mkey);
174 }
175
176 return 0;
177
178 frag_err:
179 for (i = sg_i - 1; i >= hsg; i--)
180 dma_unmap_page(dev, ash->dma_handle[i], ash->size[i],
181 DMA_TO_DEVICE);
182
183 for (i = hsg - 1; i >= 0; i--)
184 dma_unmap_single(dev, ash->dma_handle[i], ash->size[i],
185 DMA_TO_DEVICE);
186
187 return -ENOMEM;
188 }
189
190 /* Handle the case when GSO SKB linear length is too large.
191 * MANA NIC requires GSO packets to put only the packet header to SGE0.
192 * So, we need 2 SGEs for the skb linear part which contains more than the
193 * header.
194 * Return a positive value for the number of SGEs, or a negative value
195 * for an error.
196 */
mana_fix_skb_head(struct net_device * ndev,struct sk_buff * skb,int gso_hs)197 static int mana_fix_skb_head(struct net_device *ndev, struct sk_buff *skb,
198 int gso_hs)
199 {
200 int num_sge = 1 + skb_shinfo(skb)->nr_frags;
201 int skb_hlen = skb_headlen(skb);
202
203 if (gso_hs < skb_hlen) {
204 num_sge++;
205 } else if (gso_hs > skb_hlen) {
206 if (net_ratelimit())
207 netdev_err(ndev,
208 "TX nonlinear head: hs:%d, skb_hlen:%d\n",
209 gso_hs, skb_hlen);
210
211 return -EINVAL;
212 }
213
214 return num_sge;
215 }
216
217 /* Get the GSO packet's header size */
mana_get_gso_hs(struct sk_buff * skb)218 static int mana_get_gso_hs(struct sk_buff *skb)
219 {
220 int gso_hs;
221
222 if (skb->encapsulation) {
223 gso_hs = skb_inner_tcp_all_headers(skb);
224 } else {
225 if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) {
226 gso_hs = skb_transport_offset(skb) +
227 sizeof(struct udphdr);
228 } else {
229 gso_hs = skb_tcp_all_headers(skb);
230 }
231 }
232
233 return gso_hs;
234 }
235
mana_start_xmit(struct sk_buff * skb,struct net_device * ndev)236 netdev_tx_t mana_start_xmit(struct sk_buff *skb, struct net_device *ndev)
237 {
238 enum mana_tx_pkt_format pkt_fmt = MANA_SHORT_PKT_FMT;
239 struct mana_port_context *apc = netdev_priv(ndev);
240 int gso_hs = 0; /* zero for non-GSO pkts */
241 u16 txq_idx = skb_get_queue_mapping(skb);
242 struct gdma_dev *gd = apc->ac->gdma_dev;
243 bool ipv4 = false, ipv6 = false;
244 struct mana_tx_package pkg = {};
245 struct netdev_queue *net_txq;
246 struct mana_stats_tx *tx_stats;
247 struct gdma_queue *gdma_sq;
248 unsigned int csum_type;
249 struct mana_txq *txq;
250 struct mana_cq *cq;
251 int err, len;
252
253 if (unlikely(!apc->port_is_up))
254 goto tx_drop;
255
256 if (skb_cow_head(skb, MANA_HEADROOM))
257 goto tx_drop_count;
258
259 txq = &apc->tx_qp[txq_idx].txq;
260 gdma_sq = txq->gdma_sq;
261 cq = &apc->tx_qp[txq_idx].tx_cq;
262 tx_stats = &txq->stats;
263
264 pkg.tx_oob.s_oob.vcq_num = cq->gdma_id;
265 pkg.tx_oob.s_oob.vsq_frame = txq->vsq_frame;
266
267 if (txq->vp_offset > MANA_SHORT_VPORT_OFFSET_MAX) {
268 pkg.tx_oob.l_oob.long_vp_offset = txq->vp_offset;
269 pkt_fmt = MANA_LONG_PKT_FMT;
270 } else {
271 pkg.tx_oob.s_oob.short_vp_offset = txq->vp_offset;
272 }
273
274 if (skb_vlan_tag_present(skb)) {
275 pkt_fmt = MANA_LONG_PKT_FMT;
276 pkg.tx_oob.l_oob.inject_vlan_pri_tag = 1;
277 pkg.tx_oob.l_oob.pcp = skb_vlan_tag_get_prio(skb);
278 pkg.tx_oob.l_oob.dei = skb_vlan_tag_get_cfi(skb);
279 pkg.tx_oob.l_oob.vlan_id = skb_vlan_tag_get_id(skb);
280 }
281
282 pkg.tx_oob.s_oob.pkt_fmt = pkt_fmt;
283
284 if (pkt_fmt == MANA_SHORT_PKT_FMT) {
285 pkg.wqe_req.inline_oob_size = sizeof(struct mana_tx_short_oob);
286 u64_stats_update_begin(&tx_stats->syncp);
287 tx_stats->short_pkt_fmt++;
288 u64_stats_update_end(&tx_stats->syncp);
289 } else {
290 pkg.wqe_req.inline_oob_size = sizeof(struct mana_tx_oob);
291 u64_stats_update_begin(&tx_stats->syncp);
292 tx_stats->long_pkt_fmt++;
293 u64_stats_update_end(&tx_stats->syncp);
294 }
295
296 pkg.wqe_req.inline_oob_data = &pkg.tx_oob;
297 pkg.wqe_req.flags = 0;
298 pkg.wqe_req.client_data_unit = 0;
299
300 pkg.wqe_req.num_sge = 1 + skb_shinfo(skb)->nr_frags;
301
302 if (skb->protocol == htons(ETH_P_IP))
303 ipv4 = true;
304 else if (skb->protocol == htons(ETH_P_IPV6))
305 ipv6 = true;
306
307 if (skb_is_gso(skb)) {
308 int num_sge;
309
310 gso_hs = mana_get_gso_hs(skb);
311
312 num_sge = mana_fix_skb_head(ndev, skb, gso_hs);
313 if (num_sge > 0)
314 pkg.wqe_req.num_sge = num_sge;
315 else
316 goto tx_drop_count;
317
318 u64_stats_update_begin(&tx_stats->syncp);
319 if (skb->encapsulation) {
320 tx_stats->tso_inner_packets++;
321 tx_stats->tso_inner_bytes += skb->len - gso_hs;
322 } else {
323 tx_stats->tso_packets++;
324 tx_stats->tso_bytes += skb->len - gso_hs;
325 }
326 u64_stats_update_end(&tx_stats->syncp);
327
328 pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
329 pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
330
331 pkg.tx_oob.s_oob.comp_iphdr_csum = 1;
332 pkg.tx_oob.s_oob.comp_tcp_csum = 1;
333 pkg.tx_oob.s_oob.trans_off = skb_transport_offset(skb);
334
335 pkg.wqe_req.client_data_unit = skb_shinfo(skb)->gso_size;
336 pkg.wqe_req.flags = GDMA_WR_OOB_IN_SGL | GDMA_WR_PAD_BY_SGE0;
337 if (ipv4) {
338 ip_hdr(skb)->tot_len = 0;
339 ip_hdr(skb)->check = 0;
340 tcp_hdr(skb)->check =
341 ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
342 ip_hdr(skb)->daddr, 0,
343 IPPROTO_TCP, 0);
344 } else {
345 ipv6_hdr(skb)->payload_len = 0;
346 tcp_hdr(skb)->check =
347 ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
348 &ipv6_hdr(skb)->daddr, 0,
349 IPPROTO_TCP, 0);
350 }
351 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
352 csum_type = mana_checksum_info(skb);
353
354 u64_stats_update_begin(&tx_stats->syncp);
355 tx_stats->csum_partial++;
356 u64_stats_update_end(&tx_stats->syncp);
357
358 if (csum_type == IPPROTO_TCP) {
359 pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
360 pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
361
362 pkg.tx_oob.s_oob.comp_tcp_csum = 1;
363 pkg.tx_oob.s_oob.trans_off = skb_transport_offset(skb);
364
365 } else if (csum_type == IPPROTO_UDP) {
366 pkg.tx_oob.s_oob.is_outer_ipv4 = ipv4;
367 pkg.tx_oob.s_oob.is_outer_ipv6 = ipv6;
368
369 pkg.tx_oob.s_oob.comp_udp_csum = 1;
370 } else {
371 /* Can't do offload of this type of checksum */
372 if (skb_checksum_help(skb))
373 goto tx_drop_count;
374 }
375 }
376
377 WARN_ON_ONCE(pkg.wqe_req.num_sge > MAX_TX_WQE_SGL_ENTRIES);
378
379 if (pkg.wqe_req.num_sge <= ARRAY_SIZE(pkg.sgl_array)) {
380 pkg.wqe_req.sgl = pkg.sgl_array;
381 } else {
382 pkg.sgl_ptr = kmalloc_array(pkg.wqe_req.num_sge,
383 sizeof(struct gdma_sge),
384 GFP_ATOMIC);
385 if (!pkg.sgl_ptr)
386 goto tx_drop_count;
387
388 pkg.wqe_req.sgl = pkg.sgl_ptr;
389 }
390
391 if (mana_map_skb(skb, apc, &pkg, gso_hs)) {
392 u64_stats_update_begin(&tx_stats->syncp);
393 tx_stats->mana_map_err++;
394 u64_stats_update_end(&tx_stats->syncp);
395 goto free_sgl_ptr;
396 }
397
398 skb_queue_tail(&txq->pending_skbs, skb);
399
400 len = skb->len;
401 net_txq = netdev_get_tx_queue(ndev, txq_idx);
402
403 err = mana_gd_post_work_request(gdma_sq, &pkg.wqe_req,
404 (struct gdma_posted_wqe_info *)skb->cb);
405 if (!mana_can_tx(gdma_sq)) {
406 netif_tx_stop_queue(net_txq);
407 apc->eth_stats.stop_queue++;
408 }
409
410 if (err) {
411 (void)skb_dequeue_tail(&txq->pending_skbs);
412 netdev_warn(ndev, "Failed to post TX OOB: %d\n", err);
413 err = NETDEV_TX_BUSY;
414 goto tx_busy;
415 }
416
417 err = NETDEV_TX_OK;
418 atomic_inc(&txq->pending_sends);
419
420 mana_gd_wq_ring_doorbell(gd->gdma_context, gdma_sq);
421
422 /* skb may be freed after mana_gd_post_work_request. Do not use it. */
423 skb = NULL;
424
425 tx_stats = &txq->stats;
426 u64_stats_update_begin(&tx_stats->syncp);
427 tx_stats->packets++;
428 tx_stats->bytes += len;
429 u64_stats_update_end(&tx_stats->syncp);
430
431 tx_busy:
432 if (netif_tx_queue_stopped(net_txq) && mana_can_tx(gdma_sq)) {
433 netif_tx_wake_queue(net_txq);
434 apc->eth_stats.wake_queue++;
435 }
436
437 kfree(pkg.sgl_ptr);
438 return err;
439
440 free_sgl_ptr:
441 kfree(pkg.sgl_ptr);
442 tx_drop_count:
443 ndev->stats.tx_dropped++;
444 tx_drop:
445 dev_kfree_skb_any(skb);
446 return NETDEV_TX_OK;
447 }
448
mana_get_stats64(struct net_device * ndev,struct rtnl_link_stats64 * st)449 static void mana_get_stats64(struct net_device *ndev,
450 struct rtnl_link_stats64 *st)
451 {
452 struct mana_port_context *apc = netdev_priv(ndev);
453 unsigned int num_queues = apc->num_queues;
454 struct mana_stats_rx *rx_stats;
455 struct mana_stats_tx *tx_stats;
456 unsigned int start;
457 u64 packets, bytes;
458 int q;
459
460 if (!apc->port_is_up)
461 return;
462
463 netdev_stats_to_stats64(st, &ndev->stats);
464
465 for (q = 0; q < num_queues; q++) {
466 rx_stats = &apc->rxqs[q]->stats;
467
468 do {
469 start = u64_stats_fetch_begin(&rx_stats->syncp);
470 packets = rx_stats->packets;
471 bytes = rx_stats->bytes;
472 } while (u64_stats_fetch_retry(&rx_stats->syncp, start));
473
474 st->rx_packets += packets;
475 st->rx_bytes += bytes;
476 }
477
478 for (q = 0; q < num_queues; q++) {
479 tx_stats = &apc->tx_qp[q].txq.stats;
480
481 do {
482 start = u64_stats_fetch_begin(&tx_stats->syncp);
483 packets = tx_stats->packets;
484 bytes = tx_stats->bytes;
485 } while (u64_stats_fetch_retry(&tx_stats->syncp, start));
486
487 st->tx_packets += packets;
488 st->tx_bytes += bytes;
489 }
490 }
491
mana_get_tx_queue(struct net_device * ndev,struct sk_buff * skb,int old_q)492 static int mana_get_tx_queue(struct net_device *ndev, struct sk_buff *skb,
493 int old_q)
494 {
495 struct mana_port_context *apc = netdev_priv(ndev);
496 u32 hash = skb_get_hash(skb);
497 struct sock *sk = skb->sk;
498 int txq;
499
500 txq = apc->indir_table[hash & (apc->indir_table_sz - 1)];
501
502 if (txq != old_q && sk && sk_fullsock(sk) &&
503 rcu_access_pointer(sk->sk_dst_cache))
504 sk_tx_queue_set(sk, txq);
505
506 return txq;
507 }
508
mana_select_queue(struct net_device * ndev,struct sk_buff * skb,struct net_device * sb_dev)509 static u16 mana_select_queue(struct net_device *ndev, struct sk_buff *skb,
510 struct net_device *sb_dev)
511 {
512 int txq;
513
514 if (ndev->real_num_tx_queues == 1)
515 return 0;
516
517 txq = sk_tx_queue_get(skb->sk);
518
519 if (txq < 0 || skb->ooo_okay || txq >= ndev->real_num_tx_queues) {
520 if (skb_rx_queue_recorded(skb))
521 txq = skb_get_rx_queue(skb);
522 else
523 txq = mana_get_tx_queue(ndev, skb, txq);
524 }
525
526 return txq;
527 }
528
529 /* Release pre-allocated RX buffers */
mana_pre_dealloc_rxbufs(struct mana_port_context * mpc)530 void mana_pre_dealloc_rxbufs(struct mana_port_context *mpc)
531 {
532 struct device *dev;
533 int i;
534
535 dev = mpc->ac->gdma_dev->gdma_context->dev;
536
537 if (!mpc->rxbufs_pre)
538 goto out1;
539
540 if (!mpc->das_pre)
541 goto out2;
542
543 while (mpc->rxbpre_total) {
544 i = --mpc->rxbpre_total;
545 dma_unmap_single(dev, mpc->das_pre[i], mpc->rxbpre_datasize,
546 DMA_FROM_DEVICE);
547 put_page(virt_to_head_page(mpc->rxbufs_pre[i]));
548 }
549
550 kfree(mpc->das_pre);
551 mpc->das_pre = NULL;
552
553 out2:
554 kfree(mpc->rxbufs_pre);
555 mpc->rxbufs_pre = NULL;
556
557 out1:
558 mpc->rxbpre_datasize = 0;
559 mpc->rxbpre_alloc_size = 0;
560 mpc->rxbpre_headroom = 0;
561 }
562
563 /* Get a buffer from the pre-allocated RX buffers */
mana_get_rxbuf_pre(struct mana_rxq * rxq,dma_addr_t * da)564 static void *mana_get_rxbuf_pre(struct mana_rxq *rxq, dma_addr_t *da)
565 {
566 struct net_device *ndev = rxq->ndev;
567 struct mana_port_context *mpc;
568 void *va;
569
570 mpc = netdev_priv(ndev);
571
572 if (!mpc->rxbufs_pre || !mpc->das_pre || !mpc->rxbpre_total) {
573 netdev_err(ndev, "No RX pre-allocated bufs\n");
574 return NULL;
575 }
576
577 /* Check sizes to catch unexpected coding error */
578 if (mpc->rxbpre_datasize != rxq->datasize) {
579 netdev_err(ndev, "rxbpre_datasize mismatch: %u: %u\n",
580 mpc->rxbpre_datasize, rxq->datasize);
581 return NULL;
582 }
583
584 if (mpc->rxbpre_alloc_size != rxq->alloc_size) {
585 netdev_err(ndev, "rxbpre_alloc_size mismatch: %u: %u\n",
586 mpc->rxbpre_alloc_size, rxq->alloc_size);
587 return NULL;
588 }
589
590 if (mpc->rxbpre_headroom != rxq->headroom) {
591 netdev_err(ndev, "rxbpre_headroom mismatch: %u: %u\n",
592 mpc->rxbpre_headroom, rxq->headroom);
593 return NULL;
594 }
595
596 mpc->rxbpre_total--;
597
598 *da = mpc->das_pre[mpc->rxbpre_total];
599 va = mpc->rxbufs_pre[mpc->rxbpre_total];
600 mpc->rxbufs_pre[mpc->rxbpre_total] = NULL;
601
602 /* Deallocate the array after all buffers are gone */
603 if (!mpc->rxbpre_total)
604 mana_pre_dealloc_rxbufs(mpc);
605
606 return va;
607 }
608
609 /* Get RX buffer's data size, alloc size, XDP headroom based on MTU */
mana_get_rxbuf_cfg(int mtu,u32 * datasize,u32 * alloc_size,u32 * headroom)610 static void mana_get_rxbuf_cfg(int mtu, u32 *datasize, u32 *alloc_size,
611 u32 *headroom)
612 {
613 if (mtu > MANA_XDP_MTU_MAX)
614 *headroom = 0; /* no support for XDP */
615 else
616 *headroom = XDP_PACKET_HEADROOM;
617
618 *alloc_size = SKB_DATA_ALIGN(mtu + MANA_RXBUF_PAD + *headroom);
619
620 /* Using page pool in this case, so alloc_size is PAGE_SIZE */
621 if (*alloc_size < PAGE_SIZE)
622 *alloc_size = PAGE_SIZE;
623
624 *datasize = mtu + ETH_HLEN;
625 }
626
mana_pre_alloc_rxbufs(struct mana_port_context * mpc,int new_mtu,int num_queues)627 int mana_pre_alloc_rxbufs(struct mana_port_context *mpc, int new_mtu, int num_queues)
628 {
629 struct device *dev;
630 struct page *page;
631 dma_addr_t da;
632 int num_rxb;
633 void *va;
634 int i;
635
636 mana_get_rxbuf_cfg(new_mtu, &mpc->rxbpre_datasize,
637 &mpc->rxbpre_alloc_size, &mpc->rxbpre_headroom);
638
639 dev = mpc->ac->gdma_dev->gdma_context->dev;
640
641 num_rxb = num_queues * mpc->rx_queue_size;
642
643 WARN(mpc->rxbufs_pre, "mana rxbufs_pre exists\n");
644 mpc->rxbufs_pre = kmalloc_array(num_rxb, sizeof(void *), GFP_KERNEL);
645 if (!mpc->rxbufs_pre)
646 goto error;
647
648 mpc->das_pre = kmalloc_array(num_rxb, sizeof(dma_addr_t), GFP_KERNEL);
649 if (!mpc->das_pre)
650 goto error;
651
652 mpc->rxbpre_total = 0;
653
654 for (i = 0; i < num_rxb; i++) {
655 page = dev_alloc_pages(get_order(mpc->rxbpre_alloc_size));
656 if (!page)
657 goto error;
658
659 va = page_to_virt(page);
660
661 da = dma_map_single(dev, va + mpc->rxbpre_headroom,
662 mpc->rxbpre_datasize, DMA_FROM_DEVICE);
663 if (dma_mapping_error(dev, da)) {
664 put_page(page);
665 goto error;
666 }
667
668 mpc->rxbufs_pre[i] = va;
669 mpc->das_pre[i] = da;
670 mpc->rxbpre_total = i + 1;
671 }
672
673 return 0;
674
675 error:
676 mana_pre_dealloc_rxbufs(mpc);
677 return -ENOMEM;
678 }
679
mana_change_mtu(struct net_device * ndev,int new_mtu)680 static int mana_change_mtu(struct net_device *ndev, int new_mtu)
681 {
682 struct mana_port_context *mpc = netdev_priv(ndev);
683 unsigned int old_mtu = ndev->mtu;
684 int err;
685
686 /* Pre-allocate buffers to prevent failure in mana_attach later */
687 err = mana_pre_alloc_rxbufs(mpc, new_mtu, mpc->num_queues);
688 if (err) {
689 netdev_err(ndev, "Insufficient memory for new MTU\n");
690 return err;
691 }
692
693 err = mana_detach(ndev, false);
694 if (err) {
695 netdev_err(ndev, "mana_detach failed: %d\n", err);
696 goto out;
697 }
698
699 WRITE_ONCE(ndev->mtu, new_mtu);
700
701 err = mana_attach(ndev);
702 if (err) {
703 netdev_err(ndev, "mana_attach failed: %d\n", err);
704 WRITE_ONCE(ndev->mtu, old_mtu);
705 }
706
707 out:
708 mana_pre_dealloc_rxbufs(mpc);
709 return err;
710 }
711
712 static const struct net_device_ops mana_devops = {
713 .ndo_open = mana_open,
714 .ndo_stop = mana_close,
715 .ndo_select_queue = mana_select_queue,
716 .ndo_start_xmit = mana_start_xmit,
717 .ndo_validate_addr = eth_validate_addr,
718 .ndo_get_stats64 = mana_get_stats64,
719 .ndo_bpf = mana_bpf,
720 .ndo_xdp_xmit = mana_xdp_xmit,
721 .ndo_change_mtu = mana_change_mtu,
722 };
723
mana_cleanup_port_context(struct mana_port_context * apc)724 static void mana_cleanup_port_context(struct mana_port_context *apc)
725 {
726 /*
727 * make sure subsequent cleanup attempts don't end up removing already
728 * cleaned dentry pointer
729 */
730 debugfs_remove(apc->mana_port_debugfs);
731 apc->mana_port_debugfs = NULL;
732 kfree(apc->rxqs);
733 apc->rxqs = NULL;
734 }
735
mana_cleanup_indir_table(struct mana_port_context * apc)736 static void mana_cleanup_indir_table(struct mana_port_context *apc)
737 {
738 apc->indir_table_sz = 0;
739 kfree(apc->indir_table);
740 kfree(apc->rxobj_table);
741 }
742
mana_init_port_context(struct mana_port_context * apc)743 static int mana_init_port_context(struct mana_port_context *apc)
744 {
745 apc->rxqs = kcalloc(apc->num_queues, sizeof(struct mana_rxq *),
746 GFP_KERNEL);
747
748 return !apc->rxqs ? -ENOMEM : 0;
749 }
750
mana_send_request(struct mana_context * ac,void * in_buf,u32 in_len,void * out_buf,u32 out_len)751 static int mana_send_request(struct mana_context *ac, void *in_buf,
752 u32 in_len, void *out_buf, u32 out_len)
753 {
754 struct gdma_context *gc = ac->gdma_dev->gdma_context;
755 struct gdma_resp_hdr *resp = out_buf;
756 struct gdma_req_hdr *req = in_buf;
757 struct device *dev = gc->dev;
758 static atomic_t activity_id;
759 int err;
760
761 req->dev_id = gc->mana.dev_id;
762 req->activity_id = atomic_inc_return(&activity_id);
763
764 err = mana_gd_send_request(gc, in_len, in_buf, out_len,
765 out_buf);
766 if (err || resp->status) {
767 dev_err(dev, "Failed to send mana message: %d, 0x%x\n",
768 err, resp->status);
769 return err ? err : -EPROTO;
770 }
771
772 if (req->dev_id.as_uint32 != resp->dev_id.as_uint32 ||
773 req->activity_id != resp->activity_id) {
774 dev_err(dev, "Unexpected mana message response: %x,%x,%x,%x\n",
775 req->dev_id.as_uint32, resp->dev_id.as_uint32,
776 req->activity_id, resp->activity_id);
777 return -EPROTO;
778 }
779
780 return 0;
781 }
782
mana_verify_resp_hdr(const struct gdma_resp_hdr * resp_hdr,const enum mana_command_code expected_code,const u32 min_size)783 static int mana_verify_resp_hdr(const struct gdma_resp_hdr *resp_hdr,
784 const enum mana_command_code expected_code,
785 const u32 min_size)
786 {
787 if (resp_hdr->response.msg_type != expected_code)
788 return -EPROTO;
789
790 if (resp_hdr->response.msg_version < GDMA_MESSAGE_V1)
791 return -EPROTO;
792
793 if (resp_hdr->response.msg_size < min_size)
794 return -EPROTO;
795
796 return 0;
797 }
798
mana_pf_register_hw_vport(struct mana_port_context * apc)799 static int mana_pf_register_hw_vport(struct mana_port_context *apc)
800 {
801 struct mana_register_hw_vport_resp resp = {};
802 struct mana_register_hw_vport_req req = {};
803 int err;
804
805 mana_gd_init_req_hdr(&req.hdr, MANA_REGISTER_HW_PORT,
806 sizeof(req), sizeof(resp));
807 req.attached_gfid = 1;
808 req.is_pf_default_vport = 1;
809 req.allow_all_ether_types = 1;
810
811 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
812 sizeof(resp));
813 if (err) {
814 netdev_err(apc->ndev, "Failed to register hw vPort: %d\n", err);
815 return err;
816 }
817
818 err = mana_verify_resp_hdr(&resp.hdr, MANA_REGISTER_HW_PORT,
819 sizeof(resp));
820 if (err || resp.hdr.status) {
821 netdev_err(apc->ndev, "Failed to register hw vPort: %d, 0x%x\n",
822 err, resp.hdr.status);
823 return err ? err : -EPROTO;
824 }
825
826 apc->port_handle = resp.hw_vport_handle;
827 return 0;
828 }
829
mana_pf_deregister_hw_vport(struct mana_port_context * apc)830 static void mana_pf_deregister_hw_vport(struct mana_port_context *apc)
831 {
832 struct mana_deregister_hw_vport_resp resp = {};
833 struct mana_deregister_hw_vport_req req = {};
834 int err;
835
836 mana_gd_init_req_hdr(&req.hdr, MANA_DEREGISTER_HW_PORT,
837 sizeof(req), sizeof(resp));
838 req.hw_vport_handle = apc->port_handle;
839
840 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
841 sizeof(resp));
842 if (err) {
843 netdev_err(apc->ndev, "Failed to unregister hw vPort: %d\n",
844 err);
845 return;
846 }
847
848 err = mana_verify_resp_hdr(&resp.hdr, MANA_DEREGISTER_HW_PORT,
849 sizeof(resp));
850 if (err || resp.hdr.status)
851 netdev_err(apc->ndev,
852 "Failed to deregister hw vPort: %d, 0x%x\n",
853 err, resp.hdr.status);
854 }
855
mana_pf_register_filter(struct mana_port_context * apc)856 static int mana_pf_register_filter(struct mana_port_context *apc)
857 {
858 struct mana_register_filter_resp resp = {};
859 struct mana_register_filter_req req = {};
860 int err;
861
862 mana_gd_init_req_hdr(&req.hdr, MANA_REGISTER_FILTER,
863 sizeof(req), sizeof(resp));
864 req.vport = apc->port_handle;
865 memcpy(req.mac_addr, apc->mac_addr, ETH_ALEN);
866
867 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
868 sizeof(resp));
869 if (err) {
870 netdev_err(apc->ndev, "Failed to register filter: %d\n", err);
871 return err;
872 }
873
874 err = mana_verify_resp_hdr(&resp.hdr, MANA_REGISTER_FILTER,
875 sizeof(resp));
876 if (err || resp.hdr.status) {
877 netdev_err(apc->ndev, "Failed to register filter: %d, 0x%x\n",
878 err, resp.hdr.status);
879 return err ? err : -EPROTO;
880 }
881
882 apc->pf_filter_handle = resp.filter_handle;
883 return 0;
884 }
885
mana_pf_deregister_filter(struct mana_port_context * apc)886 static void mana_pf_deregister_filter(struct mana_port_context *apc)
887 {
888 struct mana_deregister_filter_resp resp = {};
889 struct mana_deregister_filter_req req = {};
890 int err;
891
892 mana_gd_init_req_hdr(&req.hdr, MANA_DEREGISTER_FILTER,
893 sizeof(req), sizeof(resp));
894 req.filter_handle = apc->pf_filter_handle;
895
896 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
897 sizeof(resp));
898 if (err) {
899 netdev_err(apc->ndev, "Failed to unregister filter: %d\n",
900 err);
901 return;
902 }
903
904 err = mana_verify_resp_hdr(&resp.hdr, MANA_DEREGISTER_FILTER,
905 sizeof(resp));
906 if (err || resp.hdr.status)
907 netdev_err(apc->ndev,
908 "Failed to deregister filter: %d, 0x%x\n",
909 err, resp.hdr.status);
910 }
911
mana_query_device_cfg(struct mana_context * ac,u32 proto_major_ver,u32 proto_minor_ver,u32 proto_micro_ver,u16 * max_num_vports)912 static int mana_query_device_cfg(struct mana_context *ac, u32 proto_major_ver,
913 u32 proto_minor_ver, u32 proto_micro_ver,
914 u16 *max_num_vports)
915 {
916 struct gdma_context *gc = ac->gdma_dev->gdma_context;
917 struct mana_query_device_cfg_resp resp = {};
918 struct mana_query_device_cfg_req req = {};
919 struct device *dev = gc->dev;
920 int err = 0;
921
922 mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_DEV_CONFIG,
923 sizeof(req), sizeof(resp));
924
925 req.hdr.resp.msg_version = GDMA_MESSAGE_V2;
926
927 req.proto_major_ver = proto_major_ver;
928 req.proto_minor_ver = proto_minor_ver;
929 req.proto_micro_ver = proto_micro_ver;
930
931 err = mana_send_request(ac, &req, sizeof(req), &resp, sizeof(resp));
932 if (err) {
933 dev_err(dev, "Failed to query config: %d", err);
934 return err;
935 }
936
937 err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_DEV_CONFIG,
938 sizeof(resp));
939 if (err || resp.hdr.status) {
940 dev_err(dev, "Invalid query result: %d, 0x%x\n", err,
941 resp.hdr.status);
942 if (!err)
943 err = -EPROTO;
944 return err;
945 }
946
947 *max_num_vports = resp.max_num_vports;
948
949 if (resp.hdr.response.msg_version == GDMA_MESSAGE_V2)
950 gc->adapter_mtu = resp.adapter_mtu;
951 else
952 gc->adapter_mtu = ETH_FRAME_LEN;
953
954 debugfs_create_u16("adapter-MTU", 0400, gc->mana_pci_debugfs, &gc->adapter_mtu);
955
956 return 0;
957 }
958
mana_query_vport_cfg(struct mana_port_context * apc,u32 vport_index,u32 * max_sq,u32 * max_rq,u32 * num_indir_entry)959 static int mana_query_vport_cfg(struct mana_port_context *apc, u32 vport_index,
960 u32 *max_sq, u32 *max_rq, u32 *num_indir_entry)
961 {
962 struct mana_query_vport_cfg_resp resp = {};
963 struct mana_query_vport_cfg_req req = {};
964 int err;
965
966 mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_VPORT_CONFIG,
967 sizeof(req), sizeof(resp));
968
969 req.vport_index = vport_index;
970
971 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
972 sizeof(resp));
973 if (err)
974 return err;
975
976 err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_VPORT_CONFIG,
977 sizeof(resp));
978 if (err)
979 return err;
980
981 if (resp.hdr.status)
982 return -EPROTO;
983
984 *max_sq = resp.max_num_sq;
985 *max_rq = resp.max_num_rq;
986 if (resp.num_indirection_ent > 0 &&
987 resp.num_indirection_ent <= MANA_INDIRECT_TABLE_MAX_SIZE &&
988 is_power_of_2(resp.num_indirection_ent)) {
989 *num_indir_entry = resp.num_indirection_ent;
990 } else {
991 netdev_warn(apc->ndev,
992 "Setting indirection table size to default %d for vPort %d\n",
993 MANA_INDIRECT_TABLE_DEF_SIZE, apc->port_idx);
994 *num_indir_entry = MANA_INDIRECT_TABLE_DEF_SIZE;
995 }
996
997 apc->port_handle = resp.vport;
998 ether_addr_copy(apc->mac_addr, resp.mac_addr);
999
1000 return 0;
1001 }
1002
mana_uncfg_vport(struct mana_port_context * apc)1003 void mana_uncfg_vport(struct mana_port_context *apc)
1004 {
1005 mutex_lock(&apc->vport_mutex);
1006 apc->vport_use_count--;
1007 WARN_ON(apc->vport_use_count < 0);
1008 mutex_unlock(&apc->vport_mutex);
1009 }
1010 EXPORT_SYMBOL_NS(mana_uncfg_vport, "NET_MANA");
1011
mana_cfg_vport(struct mana_port_context * apc,u32 protection_dom_id,u32 doorbell_pg_id)1012 int mana_cfg_vport(struct mana_port_context *apc, u32 protection_dom_id,
1013 u32 doorbell_pg_id)
1014 {
1015 struct mana_config_vport_resp resp = {};
1016 struct mana_config_vport_req req = {};
1017 int err;
1018
1019 /* This function is used to program the Ethernet port in the hardware
1020 * table. It can be called from the Ethernet driver or the RDMA driver.
1021 *
1022 * For Ethernet usage, the hardware supports only one active user on a
1023 * physical port. The driver checks on the port usage before programming
1024 * the hardware when creating the RAW QP (RDMA driver) or exposing the
1025 * device to kernel NET layer (Ethernet driver).
1026 *
1027 * Because the RDMA driver doesn't know in advance which QP type the
1028 * user will create, it exposes the device with all its ports. The user
1029 * may not be able to create RAW QP on a port if this port is already
1030 * in used by the Ethernet driver from the kernel.
1031 *
1032 * This physical port limitation only applies to the RAW QP. For RC QP,
1033 * the hardware doesn't have this limitation. The user can create RC
1034 * QPs on a physical port up to the hardware limits independent of the
1035 * Ethernet usage on the same port.
1036 */
1037 mutex_lock(&apc->vport_mutex);
1038 if (apc->vport_use_count > 0) {
1039 mutex_unlock(&apc->vport_mutex);
1040 return -EBUSY;
1041 }
1042 apc->vport_use_count++;
1043 mutex_unlock(&apc->vport_mutex);
1044
1045 mana_gd_init_req_hdr(&req.hdr, MANA_CONFIG_VPORT_TX,
1046 sizeof(req), sizeof(resp));
1047 req.vport = apc->port_handle;
1048 req.pdid = protection_dom_id;
1049 req.doorbell_pageid = doorbell_pg_id;
1050
1051 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1052 sizeof(resp));
1053 if (err) {
1054 netdev_err(apc->ndev, "Failed to configure vPort: %d\n", err);
1055 goto out;
1056 }
1057
1058 err = mana_verify_resp_hdr(&resp.hdr, MANA_CONFIG_VPORT_TX,
1059 sizeof(resp));
1060 if (err || resp.hdr.status) {
1061 netdev_err(apc->ndev, "Failed to configure vPort: %d, 0x%x\n",
1062 err, resp.hdr.status);
1063 if (!err)
1064 err = -EPROTO;
1065
1066 goto out;
1067 }
1068
1069 apc->tx_shortform_allowed = resp.short_form_allowed;
1070 apc->tx_vp_offset = resp.tx_vport_offset;
1071
1072 netdev_info(apc->ndev, "Configured vPort %llu PD %u DB %u\n",
1073 apc->port_handle, protection_dom_id, doorbell_pg_id);
1074 out:
1075 if (err)
1076 mana_uncfg_vport(apc);
1077
1078 return err;
1079 }
1080 EXPORT_SYMBOL_NS(mana_cfg_vport, "NET_MANA");
1081
mana_cfg_vport_steering(struct mana_port_context * apc,enum TRI_STATE rx,bool update_default_rxobj,bool update_key,bool update_tab)1082 static int mana_cfg_vport_steering(struct mana_port_context *apc,
1083 enum TRI_STATE rx,
1084 bool update_default_rxobj, bool update_key,
1085 bool update_tab)
1086 {
1087 struct mana_cfg_rx_steer_req_v2 *req;
1088 struct mana_cfg_rx_steer_resp resp = {};
1089 struct net_device *ndev = apc->ndev;
1090 u32 req_buf_size;
1091 int err;
1092
1093 req_buf_size = struct_size(req, indir_tab, apc->indir_table_sz);
1094 req = kzalloc(req_buf_size, GFP_KERNEL);
1095 if (!req)
1096 return -ENOMEM;
1097
1098 mana_gd_init_req_hdr(&req->hdr, MANA_CONFIG_VPORT_RX, req_buf_size,
1099 sizeof(resp));
1100
1101 req->hdr.req.msg_version = GDMA_MESSAGE_V2;
1102
1103 req->vport = apc->port_handle;
1104 req->num_indir_entries = apc->indir_table_sz;
1105 req->indir_tab_offset = offsetof(struct mana_cfg_rx_steer_req_v2,
1106 indir_tab);
1107 req->rx_enable = rx;
1108 req->rss_enable = apc->rss_state;
1109 req->update_default_rxobj = update_default_rxobj;
1110 req->update_hashkey = update_key;
1111 req->update_indir_tab = update_tab;
1112 req->default_rxobj = apc->default_rxobj;
1113 req->cqe_coalescing_enable = 0;
1114
1115 if (update_key)
1116 memcpy(&req->hashkey, apc->hashkey, MANA_HASH_KEY_SIZE);
1117
1118 if (update_tab)
1119 memcpy(req->indir_tab, apc->rxobj_table,
1120 flex_array_size(req, indir_tab, req->num_indir_entries));
1121
1122 err = mana_send_request(apc->ac, req, req_buf_size, &resp,
1123 sizeof(resp));
1124 if (err) {
1125 netdev_err(ndev, "Failed to configure vPort RX: %d\n", err);
1126 goto out;
1127 }
1128
1129 err = mana_verify_resp_hdr(&resp.hdr, MANA_CONFIG_VPORT_RX,
1130 sizeof(resp));
1131 if (err) {
1132 netdev_err(ndev, "vPort RX configuration failed: %d\n", err);
1133 goto out;
1134 }
1135
1136 if (resp.hdr.status) {
1137 netdev_err(ndev, "vPort RX configuration failed: 0x%x\n",
1138 resp.hdr.status);
1139 err = -EPROTO;
1140 }
1141
1142 netdev_info(ndev, "Configured steering vPort %llu entries %u\n",
1143 apc->port_handle, apc->indir_table_sz);
1144 out:
1145 kfree(req);
1146 return err;
1147 }
1148
mana_create_wq_obj(struct mana_port_context * apc,mana_handle_t vport,u32 wq_type,struct mana_obj_spec * wq_spec,struct mana_obj_spec * cq_spec,mana_handle_t * wq_obj)1149 int mana_create_wq_obj(struct mana_port_context *apc,
1150 mana_handle_t vport,
1151 u32 wq_type, struct mana_obj_spec *wq_spec,
1152 struct mana_obj_spec *cq_spec,
1153 mana_handle_t *wq_obj)
1154 {
1155 struct mana_create_wqobj_resp resp = {};
1156 struct mana_create_wqobj_req req = {};
1157 struct net_device *ndev = apc->ndev;
1158 int err;
1159
1160 mana_gd_init_req_hdr(&req.hdr, MANA_CREATE_WQ_OBJ,
1161 sizeof(req), sizeof(resp));
1162 req.vport = vport;
1163 req.wq_type = wq_type;
1164 req.wq_gdma_region = wq_spec->gdma_region;
1165 req.cq_gdma_region = cq_spec->gdma_region;
1166 req.wq_size = wq_spec->queue_size;
1167 req.cq_size = cq_spec->queue_size;
1168 req.cq_moderation_ctx_id = cq_spec->modr_ctx_id;
1169 req.cq_parent_qid = cq_spec->attached_eq;
1170
1171 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1172 sizeof(resp));
1173 if (err) {
1174 netdev_err(ndev, "Failed to create WQ object: %d\n", err);
1175 goto out;
1176 }
1177
1178 err = mana_verify_resp_hdr(&resp.hdr, MANA_CREATE_WQ_OBJ,
1179 sizeof(resp));
1180 if (err || resp.hdr.status) {
1181 netdev_err(ndev, "Failed to create WQ object: %d, 0x%x\n", err,
1182 resp.hdr.status);
1183 if (!err)
1184 err = -EPROTO;
1185 goto out;
1186 }
1187
1188 if (resp.wq_obj == INVALID_MANA_HANDLE) {
1189 netdev_err(ndev, "Got an invalid WQ object handle\n");
1190 err = -EPROTO;
1191 goto out;
1192 }
1193
1194 *wq_obj = resp.wq_obj;
1195 wq_spec->queue_index = resp.wq_id;
1196 cq_spec->queue_index = resp.cq_id;
1197
1198 return 0;
1199 out:
1200 return err;
1201 }
1202 EXPORT_SYMBOL_NS(mana_create_wq_obj, "NET_MANA");
1203
mana_destroy_wq_obj(struct mana_port_context * apc,u32 wq_type,mana_handle_t wq_obj)1204 void mana_destroy_wq_obj(struct mana_port_context *apc, u32 wq_type,
1205 mana_handle_t wq_obj)
1206 {
1207 struct mana_destroy_wqobj_resp resp = {};
1208 struct mana_destroy_wqobj_req req = {};
1209 struct net_device *ndev = apc->ndev;
1210 int err;
1211
1212 mana_gd_init_req_hdr(&req.hdr, MANA_DESTROY_WQ_OBJ,
1213 sizeof(req), sizeof(resp));
1214 req.wq_type = wq_type;
1215 req.wq_obj_handle = wq_obj;
1216
1217 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1218 sizeof(resp));
1219 if (err) {
1220 netdev_err(ndev, "Failed to destroy WQ object: %d\n", err);
1221 return;
1222 }
1223
1224 err = mana_verify_resp_hdr(&resp.hdr, MANA_DESTROY_WQ_OBJ,
1225 sizeof(resp));
1226 if (err || resp.hdr.status)
1227 netdev_err(ndev, "Failed to destroy WQ object: %d, 0x%x\n", err,
1228 resp.hdr.status);
1229 }
1230 EXPORT_SYMBOL_NS(mana_destroy_wq_obj, "NET_MANA");
1231
mana_destroy_eq(struct mana_context * ac)1232 static void mana_destroy_eq(struct mana_context *ac)
1233 {
1234 struct gdma_context *gc = ac->gdma_dev->gdma_context;
1235 struct gdma_queue *eq;
1236 int i;
1237
1238 if (!ac->eqs)
1239 return;
1240
1241 debugfs_remove_recursive(ac->mana_eqs_debugfs);
1242 ac->mana_eqs_debugfs = NULL;
1243
1244 for (i = 0; i < gc->max_num_queues; i++) {
1245 eq = ac->eqs[i].eq;
1246 if (!eq)
1247 continue;
1248
1249 mana_gd_destroy_queue(gc, eq);
1250 }
1251
1252 kfree(ac->eqs);
1253 ac->eqs = NULL;
1254 }
1255
mana_create_eq_debugfs(struct mana_context * ac,int i)1256 static void mana_create_eq_debugfs(struct mana_context *ac, int i)
1257 {
1258 struct mana_eq eq = ac->eqs[i];
1259 char eqnum[32];
1260
1261 sprintf(eqnum, "eq%d", i);
1262 eq.mana_eq_debugfs = debugfs_create_dir(eqnum, ac->mana_eqs_debugfs);
1263 debugfs_create_u32("head", 0400, eq.mana_eq_debugfs, &eq.eq->head);
1264 debugfs_create_u32("tail", 0400, eq.mana_eq_debugfs, &eq.eq->tail);
1265 debugfs_create_file("eq_dump", 0400, eq.mana_eq_debugfs, eq.eq, &mana_dbg_q_fops);
1266 }
1267
mana_create_eq(struct mana_context * ac)1268 static int mana_create_eq(struct mana_context *ac)
1269 {
1270 struct gdma_dev *gd = ac->gdma_dev;
1271 struct gdma_context *gc = gd->gdma_context;
1272 struct gdma_queue_spec spec = {};
1273 int err;
1274 int i;
1275
1276 ac->eqs = kcalloc(gc->max_num_queues, sizeof(struct mana_eq),
1277 GFP_KERNEL);
1278 if (!ac->eqs)
1279 return -ENOMEM;
1280
1281 spec.type = GDMA_EQ;
1282 spec.monitor_avl_buf = false;
1283 spec.queue_size = EQ_SIZE;
1284 spec.eq.callback = NULL;
1285 spec.eq.context = ac->eqs;
1286 spec.eq.log2_throttle_limit = LOG2_EQ_THROTTLE;
1287
1288 ac->mana_eqs_debugfs = debugfs_create_dir("EQs", gc->mana_pci_debugfs);
1289
1290 for (i = 0; i < gc->max_num_queues; i++) {
1291 spec.eq.msix_index = (i + 1) % gc->num_msix_usable;
1292 err = mana_gd_create_mana_eq(gd, &spec, &ac->eqs[i].eq);
1293 if (err)
1294 goto out;
1295 mana_create_eq_debugfs(ac, i);
1296 }
1297
1298 return 0;
1299 out:
1300 mana_destroy_eq(ac);
1301 return err;
1302 }
1303
mana_fence_rq(struct mana_port_context * apc,struct mana_rxq * rxq)1304 static int mana_fence_rq(struct mana_port_context *apc, struct mana_rxq *rxq)
1305 {
1306 struct mana_fence_rq_resp resp = {};
1307 struct mana_fence_rq_req req = {};
1308 int err;
1309
1310 init_completion(&rxq->fence_event);
1311
1312 mana_gd_init_req_hdr(&req.hdr, MANA_FENCE_RQ,
1313 sizeof(req), sizeof(resp));
1314 req.wq_obj_handle = rxq->rxobj;
1315
1316 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
1317 sizeof(resp));
1318 if (err) {
1319 netdev_err(apc->ndev, "Failed to fence RQ %u: %d\n",
1320 rxq->rxq_idx, err);
1321 return err;
1322 }
1323
1324 err = mana_verify_resp_hdr(&resp.hdr, MANA_FENCE_RQ, sizeof(resp));
1325 if (err || resp.hdr.status) {
1326 netdev_err(apc->ndev, "Failed to fence RQ %u: %d, 0x%x\n",
1327 rxq->rxq_idx, err, resp.hdr.status);
1328 if (!err)
1329 err = -EPROTO;
1330
1331 return err;
1332 }
1333
1334 if (wait_for_completion_timeout(&rxq->fence_event, 10 * HZ) == 0) {
1335 netdev_err(apc->ndev, "Failed to fence RQ %u: timed out\n",
1336 rxq->rxq_idx);
1337 return -ETIMEDOUT;
1338 }
1339
1340 return 0;
1341 }
1342
mana_fence_rqs(struct mana_port_context * apc)1343 static void mana_fence_rqs(struct mana_port_context *apc)
1344 {
1345 unsigned int rxq_idx;
1346 struct mana_rxq *rxq;
1347 int err;
1348
1349 for (rxq_idx = 0; rxq_idx < apc->num_queues; rxq_idx++) {
1350 rxq = apc->rxqs[rxq_idx];
1351 err = mana_fence_rq(apc, rxq);
1352
1353 /* In case of any error, use sleep instead. */
1354 if (err)
1355 msleep(100);
1356 }
1357 }
1358
mana_move_wq_tail(struct gdma_queue * wq,u32 num_units)1359 static int mana_move_wq_tail(struct gdma_queue *wq, u32 num_units)
1360 {
1361 u32 used_space_old;
1362 u32 used_space_new;
1363
1364 used_space_old = wq->head - wq->tail;
1365 used_space_new = wq->head - (wq->tail + num_units);
1366
1367 if (WARN_ON_ONCE(used_space_new > used_space_old))
1368 return -ERANGE;
1369
1370 wq->tail += num_units;
1371 return 0;
1372 }
1373
mana_unmap_skb(struct sk_buff * skb,struct mana_port_context * apc)1374 static void mana_unmap_skb(struct sk_buff *skb, struct mana_port_context *apc)
1375 {
1376 struct mana_skb_head *ash = (struct mana_skb_head *)skb->head;
1377 struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
1378 struct device *dev = gc->dev;
1379 int hsg, i;
1380
1381 /* Number of SGEs of linear part */
1382 hsg = (skb_is_gso(skb) && skb_headlen(skb) > ash->size[0]) ? 2 : 1;
1383
1384 for (i = 0; i < hsg; i++)
1385 dma_unmap_single(dev, ash->dma_handle[i], ash->size[i],
1386 DMA_TO_DEVICE);
1387
1388 for (i = hsg; i < skb_shinfo(skb)->nr_frags + hsg; i++)
1389 dma_unmap_page(dev, ash->dma_handle[i], ash->size[i],
1390 DMA_TO_DEVICE);
1391 }
1392
mana_poll_tx_cq(struct mana_cq * cq)1393 static void mana_poll_tx_cq(struct mana_cq *cq)
1394 {
1395 struct gdma_comp *completions = cq->gdma_comp_buf;
1396 struct gdma_posted_wqe_info *wqe_info;
1397 unsigned int pkt_transmitted = 0;
1398 unsigned int wqe_unit_cnt = 0;
1399 struct mana_txq *txq = cq->txq;
1400 struct mana_port_context *apc;
1401 struct netdev_queue *net_txq;
1402 struct gdma_queue *gdma_wq;
1403 unsigned int avail_space;
1404 struct net_device *ndev;
1405 struct sk_buff *skb;
1406 bool txq_stopped;
1407 int comp_read;
1408 int i;
1409
1410 ndev = txq->ndev;
1411 apc = netdev_priv(ndev);
1412
1413 comp_read = mana_gd_poll_cq(cq->gdma_cq, completions,
1414 CQE_POLLING_BUFFER);
1415
1416 if (comp_read < 1)
1417 return;
1418
1419 for (i = 0; i < comp_read; i++) {
1420 struct mana_tx_comp_oob *cqe_oob;
1421
1422 if (WARN_ON_ONCE(!completions[i].is_sq))
1423 return;
1424
1425 cqe_oob = (struct mana_tx_comp_oob *)completions[i].cqe_data;
1426 if (WARN_ON_ONCE(cqe_oob->cqe_hdr.client_type !=
1427 MANA_CQE_COMPLETION))
1428 return;
1429
1430 switch (cqe_oob->cqe_hdr.cqe_type) {
1431 case CQE_TX_OKAY:
1432 break;
1433
1434 case CQE_TX_SA_DROP:
1435 case CQE_TX_MTU_DROP:
1436 case CQE_TX_INVALID_OOB:
1437 case CQE_TX_INVALID_ETH_TYPE:
1438 case CQE_TX_HDR_PROCESSING_ERROR:
1439 case CQE_TX_VF_DISABLED:
1440 case CQE_TX_VPORT_IDX_OUT_OF_RANGE:
1441 case CQE_TX_VPORT_DISABLED:
1442 case CQE_TX_VLAN_TAGGING_VIOLATION:
1443 if (net_ratelimit())
1444 netdev_err(ndev, "TX: CQE error %d\n",
1445 cqe_oob->cqe_hdr.cqe_type);
1446
1447 apc->eth_stats.tx_cqe_err++;
1448 break;
1449
1450 default:
1451 /* If the CQE type is unknown, log an error,
1452 * and still free the SKB, update tail, etc.
1453 */
1454 if (net_ratelimit())
1455 netdev_err(ndev, "TX: unknown CQE type %d\n",
1456 cqe_oob->cqe_hdr.cqe_type);
1457
1458 apc->eth_stats.tx_cqe_unknown_type++;
1459 break;
1460 }
1461
1462 if (WARN_ON_ONCE(txq->gdma_txq_id != completions[i].wq_num))
1463 return;
1464
1465 skb = skb_dequeue(&txq->pending_skbs);
1466 if (WARN_ON_ONCE(!skb))
1467 return;
1468
1469 wqe_info = (struct gdma_posted_wqe_info *)skb->cb;
1470 wqe_unit_cnt += wqe_info->wqe_size_in_bu;
1471
1472 mana_unmap_skb(skb, apc);
1473
1474 napi_consume_skb(skb, cq->budget);
1475
1476 pkt_transmitted++;
1477 }
1478
1479 if (WARN_ON_ONCE(wqe_unit_cnt == 0))
1480 return;
1481
1482 mana_move_wq_tail(txq->gdma_sq, wqe_unit_cnt);
1483
1484 gdma_wq = txq->gdma_sq;
1485 avail_space = mana_gd_wq_avail_space(gdma_wq);
1486
1487 /* Ensure tail updated before checking q stop */
1488 smp_mb();
1489
1490 net_txq = txq->net_txq;
1491 txq_stopped = netif_tx_queue_stopped(net_txq);
1492
1493 /* Ensure checking txq_stopped before apc->port_is_up. */
1494 smp_rmb();
1495
1496 if (txq_stopped && apc->port_is_up && avail_space >= MAX_TX_WQE_SIZE) {
1497 netif_tx_wake_queue(net_txq);
1498 apc->eth_stats.wake_queue++;
1499 }
1500
1501 if (atomic_sub_return(pkt_transmitted, &txq->pending_sends) < 0)
1502 WARN_ON_ONCE(1);
1503
1504 cq->work_done = pkt_transmitted;
1505 }
1506
mana_post_pkt_rxq(struct mana_rxq * rxq)1507 static void mana_post_pkt_rxq(struct mana_rxq *rxq)
1508 {
1509 struct mana_recv_buf_oob *recv_buf_oob;
1510 u32 curr_index;
1511 int err;
1512
1513 curr_index = rxq->buf_index++;
1514 if (rxq->buf_index == rxq->num_rx_buf)
1515 rxq->buf_index = 0;
1516
1517 recv_buf_oob = &rxq->rx_oobs[curr_index];
1518
1519 err = mana_gd_post_work_request(rxq->gdma_rq, &recv_buf_oob->wqe_req,
1520 &recv_buf_oob->wqe_inf);
1521 if (WARN_ON_ONCE(err))
1522 return;
1523
1524 WARN_ON_ONCE(recv_buf_oob->wqe_inf.wqe_size_in_bu != 1);
1525 }
1526
mana_build_skb(struct mana_rxq * rxq,void * buf_va,uint pkt_len,struct xdp_buff * xdp)1527 static struct sk_buff *mana_build_skb(struct mana_rxq *rxq, void *buf_va,
1528 uint pkt_len, struct xdp_buff *xdp)
1529 {
1530 struct sk_buff *skb = napi_build_skb(buf_va, rxq->alloc_size);
1531
1532 if (!skb)
1533 return NULL;
1534
1535 if (xdp->data_hard_start) {
1536 skb_reserve(skb, xdp->data - xdp->data_hard_start);
1537 skb_put(skb, xdp->data_end - xdp->data);
1538 return skb;
1539 }
1540
1541 skb_reserve(skb, rxq->headroom);
1542 skb_put(skb, pkt_len);
1543
1544 return skb;
1545 }
1546
mana_rx_skb(void * buf_va,bool from_pool,struct mana_rxcomp_oob * cqe,struct mana_rxq * rxq)1547 static void mana_rx_skb(void *buf_va, bool from_pool,
1548 struct mana_rxcomp_oob *cqe, struct mana_rxq *rxq)
1549 {
1550 struct mana_stats_rx *rx_stats = &rxq->stats;
1551 struct net_device *ndev = rxq->ndev;
1552 uint pkt_len = cqe->ppi[0].pkt_len;
1553 u16 rxq_idx = rxq->rxq_idx;
1554 struct napi_struct *napi;
1555 struct xdp_buff xdp = {};
1556 struct sk_buff *skb;
1557 u32 hash_value;
1558 u32 act;
1559
1560 rxq->rx_cq.work_done++;
1561 napi = &rxq->rx_cq.napi;
1562
1563 if (!buf_va) {
1564 ++ndev->stats.rx_dropped;
1565 return;
1566 }
1567
1568 act = mana_run_xdp(ndev, rxq, &xdp, buf_va, pkt_len);
1569
1570 if (act == XDP_REDIRECT && !rxq->xdp_rc)
1571 return;
1572
1573 if (act != XDP_PASS && act != XDP_TX)
1574 goto drop_xdp;
1575
1576 skb = mana_build_skb(rxq, buf_va, pkt_len, &xdp);
1577
1578 if (!skb)
1579 goto drop;
1580
1581 if (from_pool)
1582 skb_mark_for_recycle(skb);
1583
1584 skb->dev = napi->dev;
1585
1586 skb->protocol = eth_type_trans(skb, ndev);
1587 skb_checksum_none_assert(skb);
1588 skb_record_rx_queue(skb, rxq_idx);
1589
1590 if ((ndev->features & NETIF_F_RXCSUM) && cqe->rx_iphdr_csum_succeed) {
1591 if (cqe->rx_tcp_csum_succeed || cqe->rx_udp_csum_succeed)
1592 skb->ip_summed = CHECKSUM_UNNECESSARY;
1593 }
1594
1595 if (cqe->rx_hashtype != 0 && (ndev->features & NETIF_F_RXHASH)) {
1596 hash_value = cqe->ppi[0].pkt_hash;
1597
1598 if (cqe->rx_hashtype & MANA_HASH_L4)
1599 skb_set_hash(skb, hash_value, PKT_HASH_TYPE_L4);
1600 else
1601 skb_set_hash(skb, hash_value, PKT_HASH_TYPE_L3);
1602 }
1603
1604 if (cqe->rx_vlantag_present) {
1605 u16 vlan_tci = cqe->rx_vlan_id;
1606
1607 __vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q), vlan_tci);
1608 }
1609
1610 u64_stats_update_begin(&rx_stats->syncp);
1611 rx_stats->packets++;
1612 rx_stats->bytes += pkt_len;
1613
1614 if (act == XDP_TX)
1615 rx_stats->xdp_tx++;
1616 u64_stats_update_end(&rx_stats->syncp);
1617
1618 if (act == XDP_TX) {
1619 skb_set_queue_mapping(skb, rxq_idx);
1620 mana_xdp_tx(skb, ndev);
1621 return;
1622 }
1623
1624 napi_gro_receive(napi, skb);
1625
1626 return;
1627
1628 drop_xdp:
1629 u64_stats_update_begin(&rx_stats->syncp);
1630 rx_stats->xdp_drop++;
1631 u64_stats_update_end(&rx_stats->syncp);
1632
1633 drop:
1634 if (from_pool) {
1635 page_pool_recycle_direct(rxq->page_pool,
1636 virt_to_head_page(buf_va));
1637 } else {
1638 WARN_ON_ONCE(rxq->xdp_save_va);
1639 /* Save for reuse */
1640 rxq->xdp_save_va = buf_va;
1641 }
1642
1643 ++ndev->stats.rx_dropped;
1644
1645 return;
1646 }
1647
mana_get_rxfrag(struct mana_rxq * rxq,struct device * dev,dma_addr_t * da,bool * from_pool)1648 static void *mana_get_rxfrag(struct mana_rxq *rxq, struct device *dev,
1649 dma_addr_t *da, bool *from_pool)
1650 {
1651 struct page *page;
1652 void *va;
1653
1654 *from_pool = false;
1655
1656 /* Reuse XDP dropped page if available */
1657 if (rxq->xdp_save_va) {
1658 va = rxq->xdp_save_va;
1659 rxq->xdp_save_va = NULL;
1660 } else {
1661 page = page_pool_dev_alloc_pages(rxq->page_pool);
1662 if (!page)
1663 return NULL;
1664
1665 *from_pool = true;
1666 va = page_to_virt(page);
1667 }
1668
1669 *da = dma_map_single(dev, va + rxq->headroom, rxq->datasize,
1670 DMA_FROM_DEVICE);
1671 if (dma_mapping_error(dev, *da)) {
1672 if (*from_pool)
1673 page_pool_put_full_page(rxq->page_pool, page, false);
1674 else
1675 put_page(virt_to_head_page(va));
1676
1677 return NULL;
1678 }
1679
1680 return va;
1681 }
1682
1683 /* Allocate frag for rx buffer, and save the old buf */
mana_refill_rx_oob(struct device * dev,struct mana_rxq * rxq,struct mana_recv_buf_oob * rxoob,void ** old_buf,bool * old_fp)1684 static void mana_refill_rx_oob(struct device *dev, struct mana_rxq *rxq,
1685 struct mana_recv_buf_oob *rxoob, void **old_buf,
1686 bool *old_fp)
1687 {
1688 bool from_pool;
1689 dma_addr_t da;
1690 void *va;
1691
1692 va = mana_get_rxfrag(rxq, dev, &da, &from_pool);
1693 if (!va)
1694 return;
1695
1696 dma_unmap_single(dev, rxoob->sgl[0].address, rxq->datasize,
1697 DMA_FROM_DEVICE);
1698 *old_buf = rxoob->buf_va;
1699 *old_fp = rxoob->from_pool;
1700
1701 rxoob->buf_va = va;
1702 rxoob->sgl[0].address = da;
1703 rxoob->from_pool = from_pool;
1704 }
1705
mana_process_rx_cqe(struct mana_rxq * rxq,struct mana_cq * cq,struct gdma_comp * cqe)1706 static void mana_process_rx_cqe(struct mana_rxq *rxq, struct mana_cq *cq,
1707 struct gdma_comp *cqe)
1708 {
1709 struct mana_rxcomp_oob *oob = (struct mana_rxcomp_oob *)cqe->cqe_data;
1710 struct gdma_context *gc = rxq->gdma_rq->gdma_dev->gdma_context;
1711 struct net_device *ndev = rxq->ndev;
1712 struct mana_recv_buf_oob *rxbuf_oob;
1713 struct mana_port_context *apc;
1714 struct device *dev = gc->dev;
1715 void *old_buf = NULL;
1716 u32 curr, pktlen;
1717 bool old_fp;
1718
1719 apc = netdev_priv(ndev);
1720
1721 switch (oob->cqe_hdr.cqe_type) {
1722 case CQE_RX_OKAY:
1723 break;
1724
1725 case CQE_RX_TRUNCATED:
1726 ++ndev->stats.rx_dropped;
1727 rxbuf_oob = &rxq->rx_oobs[rxq->buf_index];
1728 netdev_warn_once(ndev, "Dropped a truncated packet\n");
1729 goto drop;
1730
1731 case CQE_RX_COALESCED_4:
1732 netdev_err(ndev, "RX coalescing is unsupported\n");
1733 apc->eth_stats.rx_coalesced_err++;
1734 return;
1735
1736 case CQE_RX_OBJECT_FENCE:
1737 complete(&rxq->fence_event);
1738 return;
1739
1740 default:
1741 netdev_err(ndev, "Unknown RX CQE type = %d\n",
1742 oob->cqe_hdr.cqe_type);
1743 apc->eth_stats.rx_cqe_unknown_type++;
1744 return;
1745 }
1746
1747 pktlen = oob->ppi[0].pkt_len;
1748
1749 if (pktlen == 0) {
1750 /* data packets should never have packetlength of zero */
1751 netdev_err(ndev, "RX pkt len=0, rq=%u, cq=%u, rxobj=0x%llx\n",
1752 rxq->gdma_id, cq->gdma_id, rxq->rxobj);
1753 return;
1754 }
1755
1756 curr = rxq->buf_index;
1757 rxbuf_oob = &rxq->rx_oobs[curr];
1758 WARN_ON_ONCE(rxbuf_oob->wqe_inf.wqe_size_in_bu != 1);
1759
1760 mana_refill_rx_oob(dev, rxq, rxbuf_oob, &old_buf, &old_fp);
1761
1762 /* Unsuccessful refill will have old_buf == NULL.
1763 * In this case, mana_rx_skb() will drop the packet.
1764 */
1765 mana_rx_skb(old_buf, old_fp, oob, rxq);
1766
1767 drop:
1768 mana_move_wq_tail(rxq->gdma_rq, rxbuf_oob->wqe_inf.wqe_size_in_bu);
1769
1770 mana_post_pkt_rxq(rxq);
1771 }
1772
mana_poll_rx_cq(struct mana_cq * cq)1773 static void mana_poll_rx_cq(struct mana_cq *cq)
1774 {
1775 struct gdma_comp *comp = cq->gdma_comp_buf;
1776 struct mana_rxq *rxq = cq->rxq;
1777 int comp_read, i;
1778
1779 comp_read = mana_gd_poll_cq(cq->gdma_cq, comp, CQE_POLLING_BUFFER);
1780 WARN_ON_ONCE(comp_read > CQE_POLLING_BUFFER);
1781
1782 rxq->xdp_flush = false;
1783
1784 for (i = 0; i < comp_read; i++) {
1785 if (WARN_ON_ONCE(comp[i].is_sq))
1786 return;
1787
1788 /* verify recv cqe references the right rxq */
1789 if (WARN_ON_ONCE(comp[i].wq_num != cq->rxq->gdma_id))
1790 return;
1791
1792 mana_process_rx_cqe(rxq, cq, &comp[i]);
1793 }
1794
1795 if (comp_read > 0) {
1796 struct gdma_context *gc = rxq->gdma_rq->gdma_dev->gdma_context;
1797
1798 mana_gd_wq_ring_doorbell(gc, rxq->gdma_rq);
1799 }
1800
1801 if (rxq->xdp_flush)
1802 xdp_do_flush();
1803 }
1804
mana_cq_handler(void * context,struct gdma_queue * gdma_queue)1805 static int mana_cq_handler(void *context, struct gdma_queue *gdma_queue)
1806 {
1807 struct mana_cq *cq = context;
1808 int w;
1809
1810 WARN_ON_ONCE(cq->gdma_cq != gdma_queue);
1811
1812 if (cq->type == MANA_CQ_TYPE_RX)
1813 mana_poll_rx_cq(cq);
1814 else
1815 mana_poll_tx_cq(cq);
1816
1817 w = cq->work_done;
1818 cq->work_done_since_doorbell += w;
1819
1820 if (w < cq->budget) {
1821 mana_gd_ring_cq(gdma_queue, SET_ARM_BIT);
1822 cq->work_done_since_doorbell = 0;
1823 napi_complete_done(&cq->napi, w);
1824 } else if (cq->work_done_since_doorbell >
1825 cq->gdma_cq->queue_size / COMP_ENTRY_SIZE * 4) {
1826 /* MANA hardware requires at least one doorbell ring every 8
1827 * wraparounds of CQ even if there is no need to arm the CQ.
1828 * This driver rings the doorbell as soon as we have exceeded
1829 * 4 wraparounds.
1830 */
1831 mana_gd_ring_cq(gdma_queue, 0);
1832 cq->work_done_since_doorbell = 0;
1833 }
1834
1835 return w;
1836 }
1837
mana_poll(struct napi_struct * napi,int budget)1838 static int mana_poll(struct napi_struct *napi, int budget)
1839 {
1840 struct mana_cq *cq = container_of(napi, struct mana_cq, napi);
1841 int w;
1842
1843 cq->work_done = 0;
1844 cq->budget = budget;
1845
1846 w = mana_cq_handler(cq, cq->gdma_cq);
1847
1848 return min(w, budget);
1849 }
1850
mana_schedule_napi(void * context,struct gdma_queue * gdma_queue)1851 static void mana_schedule_napi(void *context, struct gdma_queue *gdma_queue)
1852 {
1853 struct mana_cq *cq = context;
1854
1855 napi_schedule_irqoff(&cq->napi);
1856 }
1857
mana_deinit_cq(struct mana_port_context * apc,struct mana_cq * cq)1858 static void mana_deinit_cq(struct mana_port_context *apc, struct mana_cq *cq)
1859 {
1860 struct gdma_dev *gd = apc->ac->gdma_dev;
1861
1862 if (!cq->gdma_cq)
1863 return;
1864
1865 mana_gd_destroy_queue(gd->gdma_context, cq->gdma_cq);
1866 }
1867
mana_deinit_txq(struct mana_port_context * apc,struct mana_txq * txq)1868 static void mana_deinit_txq(struct mana_port_context *apc, struct mana_txq *txq)
1869 {
1870 struct gdma_dev *gd = apc->ac->gdma_dev;
1871
1872 if (!txq->gdma_sq)
1873 return;
1874
1875 mana_gd_destroy_queue(gd->gdma_context, txq->gdma_sq);
1876 }
1877
mana_destroy_txq(struct mana_port_context * apc)1878 static void mana_destroy_txq(struct mana_port_context *apc)
1879 {
1880 struct napi_struct *napi;
1881 int i;
1882
1883 if (!apc->tx_qp)
1884 return;
1885
1886 for (i = 0; i < apc->num_queues; i++) {
1887 debugfs_remove_recursive(apc->tx_qp[i].mana_tx_debugfs);
1888 apc->tx_qp[i].mana_tx_debugfs = NULL;
1889
1890 napi = &apc->tx_qp[i].tx_cq.napi;
1891 if (apc->tx_qp[i].txq.napi_initialized) {
1892 napi_synchronize(napi);
1893 napi_disable(napi);
1894 netif_napi_del(napi);
1895 apc->tx_qp[i].txq.napi_initialized = false;
1896 }
1897 mana_destroy_wq_obj(apc, GDMA_SQ, apc->tx_qp[i].tx_object);
1898
1899 mana_deinit_cq(apc, &apc->tx_qp[i].tx_cq);
1900
1901 mana_deinit_txq(apc, &apc->tx_qp[i].txq);
1902 }
1903
1904 kfree(apc->tx_qp);
1905 apc->tx_qp = NULL;
1906 }
1907
mana_create_txq_debugfs(struct mana_port_context * apc,int idx)1908 static void mana_create_txq_debugfs(struct mana_port_context *apc, int idx)
1909 {
1910 struct mana_tx_qp *tx_qp = &apc->tx_qp[idx];
1911 char qnum[32];
1912
1913 sprintf(qnum, "TX-%d", idx);
1914 tx_qp->mana_tx_debugfs = debugfs_create_dir(qnum, apc->mana_port_debugfs);
1915 debugfs_create_u32("sq_head", 0400, tx_qp->mana_tx_debugfs,
1916 &tx_qp->txq.gdma_sq->head);
1917 debugfs_create_u32("sq_tail", 0400, tx_qp->mana_tx_debugfs,
1918 &tx_qp->txq.gdma_sq->tail);
1919 debugfs_create_u32("sq_pend_skb_qlen", 0400, tx_qp->mana_tx_debugfs,
1920 &tx_qp->txq.pending_skbs.qlen);
1921 debugfs_create_u32("cq_head", 0400, tx_qp->mana_tx_debugfs,
1922 &tx_qp->tx_cq.gdma_cq->head);
1923 debugfs_create_u32("cq_tail", 0400, tx_qp->mana_tx_debugfs,
1924 &tx_qp->tx_cq.gdma_cq->tail);
1925 debugfs_create_u32("cq_budget", 0400, tx_qp->mana_tx_debugfs,
1926 &tx_qp->tx_cq.budget);
1927 debugfs_create_file("txq_dump", 0400, tx_qp->mana_tx_debugfs,
1928 tx_qp->txq.gdma_sq, &mana_dbg_q_fops);
1929 debugfs_create_file("cq_dump", 0400, tx_qp->mana_tx_debugfs,
1930 tx_qp->tx_cq.gdma_cq, &mana_dbg_q_fops);
1931 }
1932
mana_create_txq(struct mana_port_context * apc,struct net_device * net)1933 static int mana_create_txq(struct mana_port_context *apc,
1934 struct net_device *net)
1935 {
1936 struct mana_context *ac = apc->ac;
1937 struct gdma_dev *gd = ac->gdma_dev;
1938 struct mana_obj_spec wq_spec;
1939 struct mana_obj_spec cq_spec;
1940 struct gdma_queue_spec spec;
1941 struct gdma_context *gc;
1942 struct mana_txq *txq;
1943 struct mana_cq *cq;
1944 u32 txq_size;
1945 u32 cq_size;
1946 int err;
1947 int i;
1948
1949 apc->tx_qp = kcalloc(apc->num_queues, sizeof(struct mana_tx_qp),
1950 GFP_KERNEL);
1951 if (!apc->tx_qp)
1952 return -ENOMEM;
1953
1954 /* The minimum size of the WQE is 32 bytes, hence
1955 * apc->tx_queue_size represents the maximum number of WQEs
1956 * the SQ can store. This value is then used to size other queues
1957 * to prevent overflow.
1958 * Also note that the txq_size is always going to be MANA_PAGE_ALIGNED,
1959 * as min val of apc->tx_queue_size is 128 and that would make
1960 * txq_size 128*32 = 4096 and the other higher values of apc->tx_queue_size
1961 * are always power of two
1962 */
1963 txq_size = apc->tx_queue_size * 32;
1964
1965 cq_size = apc->tx_queue_size * COMP_ENTRY_SIZE;
1966
1967 gc = gd->gdma_context;
1968
1969 for (i = 0; i < apc->num_queues; i++) {
1970 apc->tx_qp[i].tx_object = INVALID_MANA_HANDLE;
1971
1972 /* Create SQ */
1973 txq = &apc->tx_qp[i].txq;
1974
1975 u64_stats_init(&txq->stats.syncp);
1976 txq->ndev = net;
1977 txq->net_txq = netdev_get_tx_queue(net, i);
1978 txq->vp_offset = apc->tx_vp_offset;
1979 txq->napi_initialized = false;
1980 skb_queue_head_init(&txq->pending_skbs);
1981
1982 memset(&spec, 0, sizeof(spec));
1983 spec.type = GDMA_SQ;
1984 spec.monitor_avl_buf = true;
1985 spec.queue_size = txq_size;
1986 err = mana_gd_create_mana_wq_cq(gd, &spec, &txq->gdma_sq);
1987 if (err)
1988 goto out;
1989
1990 /* Create SQ's CQ */
1991 cq = &apc->tx_qp[i].tx_cq;
1992 cq->type = MANA_CQ_TYPE_TX;
1993
1994 cq->txq = txq;
1995
1996 memset(&spec, 0, sizeof(spec));
1997 spec.type = GDMA_CQ;
1998 spec.monitor_avl_buf = false;
1999 spec.queue_size = cq_size;
2000 spec.cq.callback = mana_schedule_napi;
2001 spec.cq.parent_eq = ac->eqs[i].eq;
2002 spec.cq.context = cq;
2003 err = mana_gd_create_mana_wq_cq(gd, &spec, &cq->gdma_cq);
2004 if (err)
2005 goto out;
2006
2007 memset(&wq_spec, 0, sizeof(wq_spec));
2008 memset(&cq_spec, 0, sizeof(cq_spec));
2009
2010 wq_spec.gdma_region = txq->gdma_sq->mem_info.dma_region_handle;
2011 wq_spec.queue_size = txq->gdma_sq->queue_size;
2012
2013 cq_spec.gdma_region = cq->gdma_cq->mem_info.dma_region_handle;
2014 cq_spec.queue_size = cq->gdma_cq->queue_size;
2015 cq_spec.modr_ctx_id = 0;
2016 cq_spec.attached_eq = cq->gdma_cq->cq.parent->id;
2017
2018 err = mana_create_wq_obj(apc, apc->port_handle, GDMA_SQ,
2019 &wq_spec, &cq_spec,
2020 &apc->tx_qp[i].tx_object);
2021
2022 if (err)
2023 goto out;
2024
2025 txq->gdma_sq->id = wq_spec.queue_index;
2026 cq->gdma_cq->id = cq_spec.queue_index;
2027
2028 txq->gdma_sq->mem_info.dma_region_handle =
2029 GDMA_INVALID_DMA_REGION;
2030 cq->gdma_cq->mem_info.dma_region_handle =
2031 GDMA_INVALID_DMA_REGION;
2032
2033 txq->gdma_txq_id = txq->gdma_sq->id;
2034
2035 cq->gdma_id = cq->gdma_cq->id;
2036
2037 if (WARN_ON(cq->gdma_id >= gc->max_num_cqs)) {
2038 err = -EINVAL;
2039 goto out;
2040 }
2041
2042 gc->cq_table[cq->gdma_id] = cq->gdma_cq;
2043
2044 mana_create_txq_debugfs(apc, i);
2045
2046 netif_napi_add_tx(net, &cq->napi, mana_poll);
2047 napi_enable(&cq->napi);
2048 txq->napi_initialized = true;
2049
2050 mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
2051 }
2052
2053 return 0;
2054 out:
2055 mana_destroy_txq(apc);
2056 return err;
2057 }
2058
mana_destroy_rxq(struct mana_port_context * apc,struct mana_rxq * rxq,bool napi_initialized)2059 static void mana_destroy_rxq(struct mana_port_context *apc,
2060 struct mana_rxq *rxq, bool napi_initialized)
2061
2062 {
2063 struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
2064 struct mana_recv_buf_oob *rx_oob;
2065 struct device *dev = gc->dev;
2066 struct napi_struct *napi;
2067 struct page *page;
2068 int i;
2069
2070 if (!rxq)
2071 return;
2072
2073 debugfs_remove_recursive(rxq->mana_rx_debugfs);
2074 rxq->mana_rx_debugfs = NULL;
2075
2076 napi = &rxq->rx_cq.napi;
2077
2078 if (napi_initialized) {
2079 napi_synchronize(napi);
2080
2081 napi_disable(napi);
2082
2083 netif_napi_del(napi);
2084 }
2085 xdp_rxq_info_unreg(&rxq->xdp_rxq);
2086
2087 mana_destroy_wq_obj(apc, GDMA_RQ, rxq->rxobj);
2088
2089 mana_deinit_cq(apc, &rxq->rx_cq);
2090
2091 if (rxq->xdp_save_va)
2092 put_page(virt_to_head_page(rxq->xdp_save_va));
2093
2094 for (i = 0; i < rxq->num_rx_buf; i++) {
2095 rx_oob = &rxq->rx_oobs[i];
2096
2097 if (!rx_oob->buf_va)
2098 continue;
2099
2100 dma_unmap_single(dev, rx_oob->sgl[0].address,
2101 rx_oob->sgl[0].size, DMA_FROM_DEVICE);
2102
2103 page = virt_to_head_page(rx_oob->buf_va);
2104
2105 if (rx_oob->from_pool)
2106 page_pool_put_full_page(rxq->page_pool, page, false);
2107 else
2108 put_page(page);
2109
2110 rx_oob->buf_va = NULL;
2111 }
2112
2113 page_pool_destroy(rxq->page_pool);
2114
2115 if (rxq->gdma_rq)
2116 mana_gd_destroy_queue(gc, rxq->gdma_rq);
2117
2118 kfree(rxq);
2119 }
2120
mana_fill_rx_oob(struct mana_recv_buf_oob * rx_oob,u32 mem_key,struct mana_rxq * rxq,struct device * dev)2121 static int mana_fill_rx_oob(struct mana_recv_buf_oob *rx_oob, u32 mem_key,
2122 struct mana_rxq *rxq, struct device *dev)
2123 {
2124 struct mana_port_context *mpc = netdev_priv(rxq->ndev);
2125 bool from_pool = false;
2126 dma_addr_t da;
2127 void *va;
2128
2129 if (mpc->rxbufs_pre)
2130 va = mana_get_rxbuf_pre(rxq, &da);
2131 else
2132 va = mana_get_rxfrag(rxq, dev, &da, &from_pool);
2133
2134 if (!va)
2135 return -ENOMEM;
2136
2137 rx_oob->buf_va = va;
2138 rx_oob->from_pool = from_pool;
2139
2140 rx_oob->sgl[0].address = da;
2141 rx_oob->sgl[0].size = rxq->datasize;
2142 rx_oob->sgl[0].mem_key = mem_key;
2143
2144 return 0;
2145 }
2146
2147 #define MANA_WQE_HEADER_SIZE 16
2148 #define MANA_WQE_SGE_SIZE 16
2149
mana_alloc_rx_wqe(struct mana_port_context * apc,struct mana_rxq * rxq,u32 * rxq_size,u32 * cq_size)2150 static int mana_alloc_rx_wqe(struct mana_port_context *apc,
2151 struct mana_rxq *rxq, u32 *rxq_size, u32 *cq_size)
2152 {
2153 struct gdma_context *gc = apc->ac->gdma_dev->gdma_context;
2154 struct mana_recv_buf_oob *rx_oob;
2155 struct device *dev = gc->dev;
2156 u32 buf_idx;
2157 int ret;
2158
2159 WARN_ON(rxq->datasize == 0);
2160
2161 *rxq_size = 0;
2162 *cq_size = 0;
2163
2164 for (buf_idx = 0; buf_idx < rxq->num_rx_buf; buf_idx++) {
2165 rx_oob = &rxq->rx_oobs[buf_idx];
2166 memset(rx_oob, 0, sizeof(*rx_oob));
2167
2168 rx_oob->num_sge = 1;
2169
2170 ret = mana_fill_rx_oob(rx_oob, apc->ac->gdma_dev->gpa_mkey, rxq,
2171 dev);
2172 if (ret)
2173 return ret;
2174
2175 rx_oob->wqe_req.sgl = rx_oob->sgl;
2176 rx_oob->wqe_req.num_sge = rx_oob->num_sge;
2177 rx_oob->wqe_req.inline_oob_size = 0;
2178 rx_oob->wqe_req.inline_oob_data = NULL;
2179 rx_oob->wqe_req.flags = 0;
2180 rx_oob->wqe_req.client_data_unit = 0;
2181
2182 *rxq_size += ALIGN(MANA_WQE_HEADER_SIZE +
2183 MANA_WQE_SGE_SIZE * rx_oob->num_sge, 32);
2184 *cq_size += COMP_ENTRY_SIZE;
2185 }
2186
2187 return 0;
2188 }
2189
mana_push_wqe(struct mana_rxq * rxq)2190 static int mana_push_wqe(struct mana_rxq *rxq)
2191 {
2192 struct mana_recv_buf_oob *rx_oob;
2193 u32 buf_idx;
2194 int err;
2195
2196 for (buf_idx = 0; buf_idx < rxq->num_rx_buf; buf_idx++) {
2197 rx_oob = &rxq->rx_oobs[buf_idx];
2198
2199 err = mana_gd_post_and_ring(rxq->gdma_rq, &rx_oob->wqe_req,
2200 &rx_oob->wqe_inf);
2201 if (err)
2202 return -ENOSPC;
2203 }
2204
2205 return 0;
2206 }
2207
mana_create_page_pool(struct mana_rxq * rxq,struct gdma_context * gc)2208 static int mana_create_page_pool(struct mana_rxq *rxq, struct gdma_context *gc)
2209 {
2210 struct mana_port_context *mpc = netdev_priv(rxq->ndev);
2211 struct page_pool_params pprm = {};
2212 int ret;
2213
2214 pprm.pool_size = mpc->rx_queue_size;
2215 pprm.nid = gc->numa_node;
2216 pprm.napi = &rxq->rx_cq.napi;
2217 pprm.netdev = rxq->ndev;
2218 pprm.order = get_order(rxq->alloc_size);
2219
2220 rxq->page_pool = page_pool_create(&pprm);
2221
2222 if (IS_ERR(rxq->page_pool)) {
2223 ret = PTR_ERR(rxq->page_pool);
2224 rxq->page_pool = NULL;
2225 return ret;
2226 }
2227
2228 return 0;
2229 }
2230
mana_create_rxq(struct mana_port_context * apc,u32 rxq_idx,struct mana_eq * eq,struct net_device * ndev)2231 static struct mana_rxq *mana_create_rxq(struct mana_port_context *apc,
2232 u32 rxq_idx, struct mana_eq *eq,
2233 struct net_device *ndev)
2234 {
2235 struct gdma_dev *gd = apc->ac->gdma_dev;
2236 struct mana_obj_spec wq_spec;
2237 struct mana_obj_spec cq_spec;
2238 struct gdma_queue_spec spec;
2239 struct mana_cq *cq = NULL;
2240 struct gdma_context *gc;
2241 u32 cq_size, rq_size;
2242 struct mana_rxq *rxq;
2243 int err;
2244
2245 gc = gd->gdma_context;
2246
2247 rxq = kzalloc(struct_size(rxq, rx_oobs, apc->rx_queue_size),
2248 GFP_KERNEL);
2249 if (!rxq)
2250 return NULL;
2251
2252 rxq->ndev = ndev;
2253 rxq->num_rx_buf = apc->rx_queue_size;
2254 rxq->rxq_idx = rxq_idx;
2255 rxq->rxobj = INVALID_MANA_HANDLE;
2256
2257 mana_get_rxbuf_cfg(ndev->mtu, &rxq->datasize, &rxq->alloc_size,
2258 &rxq->headroom);
2259
2260 /* Create page pool for RX queue */
2261 err = mana_create_page_pool(rxq, gc);
2262 if (err) {
2263 netdev_err(ndev, "Create page pool err:%d\n", err);
2264 goto out;
2265 }
2266
2267 err = mana_alloc_rx_wqe(apc, rxq, &rq_size, &cq_size);
2268 if (err)
2269 goto out;
2270
2271 rq_size = MANA_PAGE_ALIGN(rq_size);
2272 cq_size = MANA_PAGE_ALIGN(cq_size);
2273
2274 /* Create RQ */
2275 memset(&spec, 0, sizeof(spec));
2276 spec.type = GDMA_RQ;
2277 spec.monitor_avl_buf = true;
2278 spec.queue_size = rq_size;
2279 err = mana_gd_create_mana_wq_cq(gd, &spec, &rxq->gdma_rq);
2280 if (err)
2281 goto out;
2282
2283 /* Create RQ's CQ */
2284 cq = &rxq->rx_cq;
2285 cq->type = MANA_CQ_TYPE_RX;
2286 cq->rxq = rxq;
2287
2288 memset(&spec, 0, sizeof(spec));
2289 spec.type = GDMA_CQ;
2290 spec.monitor_avl_buf = false;
2291 spec.queue_size = cq_size;
2292 spec.cq.callback = mana_schedule_napi;
2293 spec.cq.parent_eq = eq->eq;
2294 spec.cq.context = cq;
2295 err = mana_gd_create_mana_wq_cq(gd, &spec, &cq->gdma_cq);
2296 if (err)
2297 goto out;
2298
2299 memset(&wq_spec, 0, sizeof(wq_spec));
2300 memset(&cq_spec, 0, sizeof(cq_spec));
2301 wq_spec.gdma_region = rxq->gdma_rq->mem_info.dma_region_handle;
2302 wq_spec.queue_size = rxq->gdma_rq->queue_size;
2303
2304 cq_spec.gdma_region = cq->gdma_cq->mem_info.dma_region_handle;
2305 cq_spec.queue_size = cq->gdma_cq->queue_size;
2306 cq_spec.modr_ctx_id = 0;
2307 cq_spec.attached_eq = cq->gdma_cq->cq.parent->id;
2308
2309 err = mana_create_wq_obj(apc, apc->port_handle, GDMA_RQ,
2310 &wq_spec, &cq_spec, &rxq->rxobj);
2311 if (err)
2312 goto out;
2313
2314 rxq->gdma_rq->id = wq_spec.queue_index;
2315 cq->gdma_cq->id = cq_spec.queue_index;
2316
2317 rxq->gdma_rq->mem_info.dma_region_handle = GDMA_INVALID_DMA_REGION;
2318 cq->gdma_cq->mem_info.dma_region_handle = GDMA_INVALID_DMA_REGION;
2319
2320 rxq->gdma_id = rxq->gdma_rq->id;
2321 cq->gdma_id = cq->gdma_cq->id;
2322
2323 err = mana_push_wqe(rxq);
2324 if (err)
2325 goto out;
2326
2327 if (WARN_ON(cq->gdma_id >= gc->max_num_cqs)) {
2328 err = -EINVAL;
2329 goto out;
2330 }
2331
2332 gc->cq_table[cq->gdma_id] = cq->gdma_cq;
2333
2334 netif_napi_add_weight(ndev, &cq->napi, mana_poll, 1);
2335
2336 WARN_ON(xdp_rxq_info_reg(&rxq->xdp_rxq, ndev, rxq_idx,
2337 cq->napi.napi_id));
2338 WARN_ON(xdp_rxq_info_reg_mem_model(&rxq->xdp_rxq, MEM_TYPE_PAGE_POOL,
2339 rxq->page_pool));
2340
2341 napi_enable(&cq->napi);
2342
2343 mana_gd_ring_cq(cq->gdma_cq, SET_ARM_BIT);
2344 out:
2345 if (!err)
2346 return rxq;
2347
2348 netdev_err(ndev, "Failed to create RXQ: err = %d\n", err);
2349
2350 mana_destroy_rxq(apc, rxq, false);
2351
2352 if (cq)
2353 mana_deinit_cq(apc, cq);
2354
2355 return NULL;
2356 }
2357
mana_create_rxq_debugfs(struct mana_port_context * apc,int idx)2358 static void mana_create_rxq_debugfs(struct mana_port_context *apc, int idx)
2359 {
2360 struct mana_rxq *rxq;
2361 char qnum[32];
2362
2363 rxq = apc->rxqs[idx];
2364
2365 sprintf(qnum, "RX-%d", idx);
2366 rxq->mana_rx_debugfs = debugfs_create_dir(qnum, apc->mana_port_debugfs);
2367 debugfs_create_u32("rq_head", 0400, rxq->mana_rx_debugfs, &rxq->gdma_rq->head);
2368 debugfs_create_u32("rq_tail", 0400, rxq->mana_rx_debugfs, &rxq->gdma_rq->tail);
2369 debugfs_create_u32("rq_nbuf", 0400, rxq->mana_rx_debugfs, &rxq->num_rx_buf);
2370 debugfs_create_u32("cq_head", 0400, rxq->mana_rx_debugfs,
2371 &rxq->rx_cq.gdma_cq->head);
2372 debugfs_create_u32("cq_tail", 0400, rxq->mana_rx_debugfs,
2373 &rxq->rx_cq.gdma_cq->tail);
2374 debugfs_create_u32("cq_budget", 0400, rxq->mana_rx_debugfs, &rxq->rx_cq.budget);
2375 debugfs_create_file("rxq_dump", 0400, rxq->mana_rx_debugfs, rxq->gdma_rq, &mana_dbg_q_fops);
2376 debugfs_create_file("cq_dump", 0400, rxq->mana_rx_debugfs, rxq->rx_cq.gdma_cq,
2377 &mana_dbg_q_fops);
2378 }
2379
mana_add_rx_queues(struct mana_port_context * apc,struct net_device * ndev)2380 static int mana_add_rx_queues(struct mana_port_context *apc,
2381 struct net_device *ndev)
2382 {
2383 struct mana_context *ac = apc->ac;
2384 struct mana_rxq *rxq;
2385 int err = 0;
2386 int i;
2387
2388 for (i = 0; i < apc->num_queues; i++) {
2389 rxq = mana_create_rxq(apc, i, &ac->eqs[i], ndev);
2390 if (!rxq) {
2391 err = -ENOMEM;
2392 goto out;
2393 }
2394
2395 u64_stats_init(&rxq->stats.syncp);
2396
2397 apc->rxqs[i] = rxq;
2398
2399 mana_create_rxq_debugfs(apc, i);
2400 }
2401
2402 apc->default_rxobj = apc->rxqs[0]->rxobj;
2403 out:
2404 return err;
2405 }
2406
mana_destroy_vport(struct mana_port_context * apc)2407 static void mana_destroy_vport(struct mana_port_context *apc)
2408 {
2409 struct gdma_dev *gd = apc->ac->gdma_dev;
2410 struct mana_rxq *rxq;
2411 u32 rxq_idx;
2412
2413 for (rxq_idx = 0; rxq_idx < apc->num_queues; rxq_idx++) {
2414 rxq = apc->rxqs[rxq_idx];
2415 if (!rxq)
2416 continue;
2417
2418 mana_destroy_rxq(apc, rxq, true);
2419 apc->rxqs[rxq_idx] = NULL;
2420 }
2421
2422 mana_destroy_txq(apc);
2423 mana_uncfg_vport(apc);
2424
2425 if (gd->gdma_context->is_pf)
2426 mana_pf_deregister_hw_vport(apc);
2427 }
2428
mana_create_vport(struct mana_port_context * apc,struct net_device * net)2429 static int mana_create_vport(struct mana_port_context *apc,
2430 struct net_device *net)
2431 {
2432 struct gdma_dev *gd = apc->ac->gdma_dev;
2433 int err;
2434
2435 apc->default_rxobj = INVALID_MANA_HANDLE;
2436
2437 if (gd->gdma_context->is_pf) {
2438 err = mana_pf_register_hw_vport(apc);
2439 if (err)
2440 return err;
2441 }
2442
2443 err = mana_cfg_vport(apc, gd->pdid, gd->doorbell);
2444 if (err)
2445 return err;
2446
2447 return mana_create_txq(apc, net);
2448 }
2449
mana_rss_table_alloc(struct mana_port_context * apc)2450 static int mana_rss_table_alloc(struct mana_port_context *apc)
2451 {
2452 if (!apc->indir_table_sz) {
2453 netdev_err(apc->ndev,
2454 "Indirection table size not set for vPort %d\n",
2455 apc->port_idx);
2456 return -EINVAL;
2457 }
2458
2459 apc->indir_table = kcalloc(apc->indir_table_sz, sizeof(u32), GFP_KERNEL);
2460 if (!apc->indir_table)
2461 return -ENOMEM;
2462
2463 apc->rxobj_table = kcalloc(apc->indir_table_sz, sizeof(mana_handle_t), GFP_KERNEL);
2464 if (!apc->rxobj_table) {
2465 kfree(apc->indir_table);
2466 return -ENOMEM;
2467 }
2468
2469 return 0;
2470 }
2471
mana_rss_table_init(struct mana_port_context * apc)2472 static void mana_rss_table_init(struct mana_port_context *apc)
2473 {
2474 int i;
2475
2476 for (i = 0; i < apc->indir_table_sz; i++)
2477 apc->indir_table[i] =
2478 ethtool_rxfh_indir_default(i, apc->num_queues);
2479 }
2480
mana_config_rss(struct mana_port_context * apc,enum TRI_STATE rx,bool update_hash,bool update_tab)2481 int mana_config_rss(struct mana_port_context *apc, enum TRI_STATE rx,
2482 bool update_hash, bool update_tab)
2483 {
2484 u32 queue_idx;
2485 int err;
2486 int i;
2487
2488 if (update_tab) {
2489 for (i = 0; i < apc->indir_table_sz; i++) {
2490 queue_idx = apc->indir_table[i];
2491 apc->rxobj_table[i] = apc->rxqs[queue_idx]->rxobj;
2492 }
2493 }
2494
2495 err = mana_cfg_vport_steering(apc, rx, true, update_hash, update_tab);
2496 if (err)
2497 return err;
2498
2499 mana_fence_rqs(apc);
2500
2501 return 0;
2502 }
2503
mana_query_gf_stats(struct mana_port_context * apc)2504 void mana_query_gf_stats(struct mana_port_context *apc)
2505 {
2506 struct mana_query_gf_stat_resp resp = {};
2507 struct mana_query_gf_stat_req req = {};
2508 struct net_device *ndev = apc->ndev;
2509 int err;
2510
2511 mana_gd_init_req_hdr(&req.hdr, MANA_QUERY_GF_STAT,
2512 sizeof(req), sizeof(resp));
2513 req.hdr.resp.msg_version = GDMA_MESSAGE_V2;
2514 req.req_stats = STATISTICS_FLAGS_RX_DISCARDS_NO_WQE |
2515 STATISTICS_FLAGS_RX_ERRORS_VPORT_DISABLED |
2516 STATISTICS_FLAGS_HC_RX_BYTES |
2517 STATISTICS_FLAGS_HC_RX_UCAST_PACKETS |
2518 STATISTICS_FLAGS_HC_RX_UCAST_BYTES |
2519 STATISTICS_FLAGS_HC_RX_MCAST_PACKETS |
2520 STATISTICS_FLAGS_HC_RX_MCAST_BYTES |
2521 STATISTICS_FLAGS_HC_RX_BCAST_PACKETS |
2522 STATISTICS_FLAGS_HC_RX_BCAST_BYTES |
2523 STATISTICS_FLAGS_TX_ERRORS_GF_DISABLED |
2524 STATISTICS_FLAGS_TX_ERRORS_VPORT_DISABLED |
2525 STATISTICS_FLAGS_TX_ERRORS_INVAL_VPORT_OFFSET_PACKETS |
2526 STATISTICS_FLAGS_TX_ERRORS_VLAN_ENFORCEMENT |
2527 STATISTICS_FLAGS_TX_ERRORS_ETH_TYPE_ENFORCEMENT |
2528 STATISTICS_FLAGS_TX_ERRORS_SA_ENFORCEMENT |
2529 STATISTICS_FLAGS_TX_ERRORS_SQPDID_ENFORCEMENT |
2530 STATISTICS_FLAGS_TX_ERRORS_CQPDID_ENFORCEMENT |
2531 STATISTICS_FLAGS_TX_ERRORS_MTU_VIOLATION |
2532 STATISTICS_FLAGS_TX_ERRORS_INVALID_OOB |
2533 STATISTICS_FLAGS_HC_TX_BYTES |
2534 STATISTICS_FLAGS_HC_TX_UCAST_PACKETS |
2535 STATISTICS_FLAGS_HC_TX_UCAST_BYTES |
2536 STATISTICS_FLAGS_HC_TX_MCAST_PACKETS |
2537 STATISTICS_FLAGS_HC_TX_MCAST_BYTES |
2538 STATISTICS_FLAGS_HC_TX_BCAST_PACKETS |
2539 STATISTICS_FLAGS_HC_TX_BCAST_BYTES |
2540 STATISTICS_FLAGS_TX_ERRORS_GDMA_ERROR;
2541
2542 err = mana_send_request(apc->ac, &req, sizeof(req), &resp,
2543 sizeof(resp));
2544 if (err) {
2545 netdev_err(ndev, "Failed to query GF stats: %d\n", err);
2546 return;
2547 }
2548 err = mana_verify_resp_hdr(&resp.hdr, MANA_QUERY_GF_STAT,
2549 sizeof(resp));
2550 if (err || resp.hdr.status) {
2551 netdev_err(ndev, "Failed to query GF stats: %d, 0x%x\n", err,
2552 resp.hdr.status);
2553 return;
2554 }
2555
2556 apc->eth_stats.hc_rx_discards_no_wqe = resp.rx_discards_nowqe;
2557 apc->eth_stats.hc_rx_err_vport_disabled = resp.rx_err_vport_disabled;
2558 apc->eth_stats.hc_rx_bytes = resp.hc_rx_bytes;
2559 apc->eth_stats.hc_rx_ucast_pkts = resp.hc_rx_ucast_pkts;
2560 apc->eth_stats.hc_rx_ucast_bytes = resp.hc_rx_ucast_bytes;
2561 apc->eth_stats.hc_rx_bcast_pkts = resp.hc_rx_bcast_pkts;
2562 apc->eth_stats.hc_rx_bcast_bytes = resp.hc_rx_bcast_bytes;
2563 apc->eth_stats.hc_rx_mcast_pkts = resp.hc_rx_mcast_pkts;
2564 apc->eth_stats.hc_rx_mcast_bytes = resp.hc_rx_mcast_bytes;
2565 apc->eth_stats.hc_tx_err_gf_disabled = resp.tx_err_gf_disabled;
2566 apc->eth_stats.hc_tx_err_vport_disabled = resp.tx_err_vport_disabled;
2567 apc->eth_stats.hc_tx_err_inval_vportoffset_pkt =
2568 resp.tx_err_inval_vport_offset_pkt;
2569 apc->eth_stats.hc_tx_err_vlan_enforcement =
2570 resp.tx_err_vlan_enforcement;
2571 apc->eth_stats.hc_tx_err_eth_type_enforcement =
2572 resp.tx_err_ethtype_enforcement;
2573 apc->eth_stats.hc_tx_err_sa_enforcement = resp.tx_err_SA_enforcement;
2574 apc->eth_stats.hc_tx_err_sqpdid_enforcement =
2575 resp.tx_err_SQPDID_enforcement;
2576 apc->eth_stats.hc_tx_err_cqpdid_enforcement =
2577 resp.tx_err_CQPDID_enforcement;
2578 apc->eth_stats.hc_tx_err_mtu_violation = resp.tx_err_mtu_violation;
2579 apc->eth_stats.hc_tx_err_inval_oob = resp.tx_err_inval_oob;
2580 apc->eth_stats.hc_tx_bytes = resp.hc_tx_bytes;
2581 apc->eth_stats.hc_tx_ucast_pkts = resp.hc_tx_ucast_pkts;
2582 apc->eth_stats.hc_tx_ucast_bytes = resp.hc_tx_ucast_bytes;
2583 apc->eth_stats.hc_tx_bcast_pkts = resp.hc_tx_bcast_pkts;
2584 apc->eth_stats.hc_tx_bcast_bytes = resp.hc_tx_bcast_bytes;
2585 apc->eth_stats.hc_tx_mcast_pkts = resp.hc_tx_mcast_pkts;
2586 apc->eth_stats.hc_tx_mcast_bytes = resp.hc_tx_mcast_bytes;
2587 apc->eth_stats.hc_tx_err_gdma = resp.tx_err_gdma;
2588 }
2589
mana_init_port(struct net_device * ndev)2590 static int mana_init_port(struct net_device *ndev)
2591 {
2592 struct mana_port_context *apc = netdev_priv(ndev);
2593 struct gdma_dev *gd = apc->ac->gdma_dev;
2594 u32 max_txq, max_rxq, max_queues;
2595 int port_idx = apc->port_idx;
2596 struct gdma_context *gc;
2597 char vport[32];
2598 int err;
2599
2600 err = mana_init_port_context(apc);
2601 if (err)
2602 return err;
2603
2604 gc = gd->gdma_context;
2605
2606 err = mana_query_vport_cfg(apc, port_idx, &max_txq, &max_rxq,
2607 &apc->indir_table_sz);
2608 if (err) {
2609 netdev_err(ndev, "Failed to query info for vPort %d\n",
2610 port_idx);
2611 goto reset_apc;
2612 }
2613
2614 max_queues = min_t(u32, max_txq, max_rxq);
2615 if (apc->max_queues > max_queues)
2616 apc->max_queues = max_queues;
2617
2618 if (apc->num_queues > apc->max_queues)
2619 apc->num_queues = apc->max_queues;
2620
2621 eth_hw_addr_set(ndev, apc->mac_addr);
2622 sprintf(vport, "vport%d", port_idx);
2623 apc->mana_port_debugfs = debugfs_create_dir(vport, gc->mana_pci_debugfs);
2624 return 0;
2625
2626 reset_apc:
2627 mana_cleanup_port_context(apc);
2628 return err;
2629 }
2630
mana_alloc_queues(struct net_device * ndev)2631 int mana_alloc_queues(struct net_device *ndev)
2632 {
2633 struct mana_port_context *apc = netdev_priv(ndev);
2634 struct gdma_dev *gd = apc->ac->gdma_dev;
2635 int err;
2636
2637 err = mana_create_vport(apc, ndev);
2638 if (err)
2639 return err;
2640
2641 err = netif_set_real_num_tx_queues(ndev, apc->num_queues);
2642 if (err)
2643 goto destroy_vport;
2644
2645 err = mana_add_rx_queues(apc, ndev);
2646 if (err)
2647 goto destroy_vport;
2648
2649 apc->rss_state = apc->num_queues > 1 ? TRI_STATE_TRUE : TRI_STATE_FALSE;
2650
2651 err = netif_set_real_num_rx_queues(ndev, apc->num_queues);
2652 if (err)
2653 goto destroy_vport;
2654
2655 mana_rss_table_init(apc);
2656
2657 err = mana_config_rss(apc, TRI_STATE_TRUE, true, true);
2658 if (err)
2659 goto destroy_vport;
2660
2661 if (gd->gdma_context->is_pf) {
2662 err = mana_pf_register_filter(apc);
2663 if (err)
2664 goto destroy_vport;
2665 }
2666
2667 mana_chn_setxdp(apc, mana_xdp_get(apc));
2668
2669 return 0;
2670
2671 destroy_vport:
2672 mana_destroy_vport(apc);
2673 return err;
2674 }
2675
mana_attach(struct net_device * ndev)2676 int mana_attach(struct net_device *ndev)
2677 {
2678 struct mana_port_context *apc = netdev_priv(ndev);
2679 int err;
2680
2681 ASSERT_RTNL();
2682
2683 err = mana_init_port(ndev);
2684 if (err)
2685 return err;
2686
2687 if (apc->port_st_save) {
2688 err = mana_alloc_queues(ndev);
2689 if (err) {
2690 mana_cleanup_port_context(apc);
2691 return err;
2692 }
2693 }
2694
2695 apc->port_is_up = apc->port_st_save;
2696
2697 /* Ensure port state updated before txq state */
2698 smp_wmb();
2699
2700 if (apc->port_is_up)
2701 netif_carrier_on(ndev);
2702
2703 netif_device_attach(ndev);
2704
2705 return 0;
2706 }
2707
mana_dealloc_queues(struct net_device * ndev)2708 static int mana_dealloc_queues(struct net_device *ndev)
2709 {
2710 struct mana_port_context *apc = netdev_priv(ndev);
2711 unsigned long timeout = jiffies + 120 * HZ;
2712 struct gdma_dev *gd = apc->ac->gdma_dev;
2713 struct mana_txq *txq;
2714 struct sk_buff *skb;
2715 int i, err;
2716 u32 tsleep;
2717
2718 if (apc->port_is_up)
2719 return -EINVAL;
2720
2721 mana_chn_setxdp(apc, NULL);
2722
2723 if (gd->gdma_context->is_pf)
2724 mana_pf_deregister_filter(apc);
2725
2726 /* No packet can be transmitted now since apc->port_is_up is false.
2727 * There is still a tiny chance that mana_poll_tx_cq() can re-enable
2728 * a txq because it may not timely see apc->port_is_up being cleared
2729 * to false, but it doesn't matter since mana_start_xmit() drops any
2730 * new packets due to apc->port_is_up being false.
2731 *
2732 * Drain all the in-flight TX packets.
2733 * A timeout of 120 seconds for all the queues is used.
2734 * This will break the while loop when h/w is not responding.
2735 * This value of 120 has been decided here considering max
2736 * number of queues.
2737 */
2738
2739 for (i = 0; i < apc->num_queues; i++) {
2740 txq = &apc->tx_qp[i].txq;
2741 tsleep = 1000;
2742 while (atomic_read(&txq->pending_sends) > 0 &&
2743 time_before(jiffies, timeout)) {
2744 usleep_range(tsleep, tsleep + 1000);
2745 tsleep <<= 1;
2746 }
2747 if (atomic_read(&txq->pending_sends)) {
2748 err = pcie_flr(to_pci_dev(gd->gdma_context->dev));
2749 if (err) {
2750 netdev_err(ndev, "flr failed %d with %d pkts pending in txq %u\n",
2751 err, atomic_read(&txq->pending_sends),
2752 txq->gdma_txq_id);
2753 }
2754 break;
2755 }
2756 }
2757
2758 for (i = 0; i < apc->num_queues; i++) {
2759 txq = &apc->tx_qp[i].txq;
2760 while ((skb = skb_dequeue(&txq->pending_skbs))) {
2761 mana_unmap_skb(skb, apc);
2762 dev_kfree_skb_any(skb);
2763 }
2764 atomic_set(&txq->pending_sends, 0);
2765 }
2766 /* We're 100% sure the queues can no longer be woken up, because
2767 * we're sure now mana_poll_tx_cq() can't be running.
2768 */
2769
2770 apc->rss_state = TRI_STATE_FALSE;
2771 err = mana_config_rss(apc, TRI_STATE_FALSE, false, false);
2772 if (err) {
2773 netdev_err(ndev, "Failed to disable vPort: %d\n", err);
2774 return err;
2775 }
2776
2777 mana_destroy_vport(apc);
2778
2779 return 0;
2780 }
2781
mana_detach(struct net_device * ndev,bool from_close)2782 int mana_detach(struct net_device *ndev, bool from_close)
2783 {
2784 struct mana_port_context *apc = netdev_priv(ndev);
2785 int err;
2786
2787 ASSERT_RTNL();
2788
2789 apc->port_st_save = apc->port_is_up;
2790 apc->port_is_up = false;
2791
2792 /* Ensure port state updated before txq state */
2793 smp_wmb();
2794
2795 netif_tx_disable(ndev);
2796 netif_carrier_off(ndev);
2797
2798 if (apc->port_st_save) {
2799 err = mana_dealloc_queues(ndev);
2800 if (err)
2801 return err;
2802 }
2803
2804 if (!from_close) {
2805 netif_device_detach(ndev);
2806 mana_cleanup_port_context(apc);
2807 }
2808
2809 return 0;
2810 }
2811
mana_probe_port(struct mana_context * ac,int port_idx,struct net_device ** ndev_storage)2812 static int mana_probe_port(struct mana_context *ac, int port_idx,
2813 struct net_device **ndev_storage)
2814 {
2815 struct gdma_context *gc = ac->gdma_dev->gdma_context;
2816 struct mana_port_context *apc;
2817 struct net_device *ndev;
2818 int err;
2819
2820 ndev = alloc_etherdev_mq(sizeof(struct mana_port_context),
2821 gc->max_num_queues);
2822 if (!ndev)
2823 return -ENOMEM;
2824
2825 *ndev_storage = ndev;
2826
2827 apc = netdev_priv(ndev);
2828 apc->ac = ac;
2829 apc->ndev = ndev;
2830 apc->max_queues = gc->max_num_queues;
2831 apc->num_queues = gc->max_num_queues;
2832 apc->tx_queue_size = DEF_TX_BUFFERS_PER_QUEUE;
2833 apc->rx_queue_size = DEF_RX_BUFFERS_PER_QUEUE;
2834 apc->port_handle = INVALID_MANA_HANDLE;
2835 apc->pf_filter_handle = INVALID_MANA_HANDLE;
2836 apc->port_idx = port_idx;
2837
2838 mutex_init(&apc->vport_mutex);
2839 apc->vport_use_count = 0;
2840
2841 ndev->netdev_ops = &mana_devops;
2842 ndev->ethtool_ops = &mana_ethtool_ops;
2843 ndev->mtu = ETH_DATA_LEN;
2844 ndev->max_mtu = gc->adapter_mtu - ETH_HLEN;
2845 ndev->min_mtu = ETH_MIN_MTU;
2846 ndev->needed_headroom = MANA_HEADROOM;
2847 ndev->dev_port = port_idx;
2848 SET_NETDEV_DEV(ndev, gc->dev);
2849
2850 netif_carrier_off(ndev);
2851
2852 netdev_rss_key_fill(apc->hashkey, MANA_HASH_KEY_SIZE);
2853
2854 err = mana_init_port(ndev);
2855 if (err)
2856 goto free_net;
2857
2858 err = mana_rss_table_alloc(apc);
2859 if (err)
2860 goto reset_apc;
2861
2862 netdev_lockdep_set_classes(ndev);
2863
2864 ndev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
2865 ndev->hw_features |= NETIF_F_RXCSUM;
2866 ndev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
2867 ndev->hw_features |= NETIF_F_RXHASH;
2868 ndev->features = ndev->hw_features | NETIF_F_HW_VLAN_CTAG_TX |
2869 NETIF_F_HW_VLAN_CTAG_RX;
2870 ndev->vlan_features = ndev->features;
2871 xdp_set_features_flag(ndev, NETDEV_XDP_ACT_BASIC |
2872 NETDEV_XDP_ACT_REDIRECT |
2873 NETDEV_XDP_ACT_NDO_XMIT);
2874
2875 err = register_netdev(ndev);
2876 if (err) {
2877 netdev_err(ndev, "Unable to register netdev.\n");
2878 goto free_indir;
2879 }
2880
2881 return 0;
2882
2883 free_indir:
2884 mana_cleanup_indir_table(apc);
2885 reset_apc:
2886 mana_cleanup_port_context(apc);
2887 free_net:
2888 *ndev_storage = NULL;
2889 netdev_err(ndev, "Failed to probe vPort %d: %d\n", port_idx, err);
2890 free_netdev(ndev);
2891 return err;
2892 }
2893
adev_release(struct device * dev)2894 static void adev_release(struct device *dev)
2895 {
2896 struct mana_adev *madev = container_of(dev, struct mana_adev, adev.dev);
2897
2898 kfree(madev);
2899 }
2900
remove_adev(struct gdma_dev * gd)2901 static void remove_adev(struct gdma_dev *gd)
2902 {
2903 struct auxiliary_device *adev = gd->adev;
2904 int id = adev->id;
2905
2906 auxiliary_device_delete(adev);
2907 auxiliary_device_uninit(adev);
2908
2909 mana_adev_idx_free(id);
2910 gd->adev = NULL;
2911 }
2912
add_adev(struct gdma_dev * gd)2913 static int add_adev(struct gdma_dev *gd)
2914 {
2915 struct auxiliary_device *adev;
2916 struct mana_adev *madev;
2917 int ret;
2918
2919 madev = kzalloc(sizeof(*madev), GFP_KERNEL);
2920 if (!madev)
2921 return -ENOMEM;
2922
2923 adev = &madev->adev;
2924 ret = mana_adev_idx_alloc();
2925 if (ret < 0)
2926 goto idx_fail;
2927 adev->id = ret;
2928
2929 adev->name = "rdma";
2930 adev->dev.parent = gd->gdma_context->dev;
2931 adev->dev.release = adev_release;
2932 madev->mdev = gd;
2933
2934 ret = auxiliary_device_init(adev);
2935 if (ret)
2936 goto init_fail;
2937
2938 /* madev is owned by the auxiliary device */
2939 madev = NULL;
2940 ret = auxiliary_device_add(adev);
2941 if (ret)
2942 goto add_fail;
2943
2944 gd->adev = adev;
2945 return 0;
2946
2947 add_fail:
2948 auxiliary_device_uninit(adev);
2949
2950 init_fail:
2951 mana_adev_idx_free(adev->id);
2952
2953 idx_fail:
2954 kfree(madev);
2955
2956 return ret;
2957 }
2958
mana_probe(struct gdma_dev * gd,bool resuming)2959 int mana_probe(struct gdma_dev *gd, bool resuming)
2960 {
2961 struct gdma_context *gc = gd->gdma_context;
2962 struct mana_context *ac = gd->driver_data;
2963 struct device *dev = gc->dev;
2964 u16 num_ports = 0;
2965 int err;
2966 int i;
2967
2968 dev_info(dev,
2969 "Microsoft Azure Network Adapter protocol version: %d.%d.%d\n",
2970 MANA_MAJOR_VERSION, MANA_MINOR_VERSION, MANA_MICRO_VERSION);
2971
2972 err = mana_gd_register_device(gd);
2973 if (err)
2974 return err;
2975
2976 if (!resuming) {
2977 ac = kzalloc(sizeof(*ac), GFP_KERNEL);
2978 if (!ac)
2979 return -ENOMEM;
2980
2981 ac->gdma_dev = gd;
2982 gd->driver_data = ac;
2983 }
2984
2985 err = mana_create_eq(ac);
2986 if (err)
2987 goto out;
2988
2989 err = mana_query_device_cfg(ac, MANA_MAJOR_VERSION, MANA_MINOR_VERSION,
2990 MANA_MICRO_VERSION, &num_ports);
2991 if (err)
2992 goto out;
2993
2994 if (!resuming) {
2995 ac->num_ports = num_ports;
2996 } else {
2997 if (ac->num_ports != num_ports) {
2998 dev_err(dev, "The number of vPorts changed: %d->%d\n",
2999 ac->num_ports, num_ports);
3000 err = -EPROTO;
3001 goto out;
3002 }
3003 }
3004
3005 if (ac->num_ports == 0)
3006 dev_err(dev, "Failed to detect any vPort\n");
3007
3008 if (ac->num_ports > MAX_PORTS_IN_MANA_DEV)
3009 ac->num_ports = MAX_PORTS_IN_MANA_DEV;
3010
3011 if (!resuming) {
3012 for (i = 0; i < ac->num_ports; i++) {
3013 err = mana_probe_port(ac, i, &ac->ports[i]);
3014 /* we log the port for which the probe failed and stop
3015 * probes for subsequent ports.
3016 * Note that we keep running ports, for which the probes
3017 * were successful, unless add_adev fails too
3018 */
3019 if (err) {
3020 dev_err(dev, "Probe Failed for port %d\n", i);
3021 break;
3022 }
3023 }
3024 } else {
3025 for (i = 0; i < ac->num_ports; i++) {
3026 rtnl_lock();
3027 err = mana_attach(ac->ports[i]);
3028 rtnl_unlock();
3029 /* we log the port for which the attach failed and stop
3030 * attach for subsequent ports
3031 * Note that we keep running ports, for which the attach
3032 * were successful, unless add_adev fails too
3033 */
3034 if (err) {
3035 dev_err(dev, "Attach Failed for port %d\n", i);
3036 break;
3037 }
3038 }
3039 }
3040
3041 err = add_adev(gd);
3042 out:
3043 if (err)
3044 mana_remove(gd, false);
3045
3046 return err;
3047 }
3048
mana_remove(struct gdma_dev * gd,bool suspending)3049 void mana_remove(struct gdma_dev *gd, bool suspending)
3050 {
3051 struct gdma_context *gc = gd->gdma_context;
3052 struct mana_context *ac = gd->driver_data;
3053 struct mana_port_context *apc;
3054 struct device *dev = gc->dev;
3055 struct net_device *ndev;
3056 int err;
3057 int i;
3058
3059 /* adev currently doesn't support suspending, always remove it */
3060 if (gd->adev)
3061 remove_adev(gd);
3062
3063 for (i = 0; i < ac->num_ports; i++) {
3064 ndev = ac->ports[i];
3065 apc = netdev_priv(ndev);
3066 if (!ndev) {
3067 if (i == 0)
3068 dev_err(dev, "No net device to remove\n");
3069 goto out;
3070 }
3071
3072 /* All cleanup actions should stay after rtnl_lock(), otherwise
3073 * other functions may access partially cleaned up data.
3074 */
3075 rtnl_lock();
3076
3077 err = mana_detach(ndev, false);
3078 if (err)
3079 netdev_err(ndev, "Failed to detach vPort %d: %d\n",
3080 i, err);
3081
3082 if (suspending) {
3083 /* No need to unregister the ndev. */
3084 rtnl_unlock();
3085 continue;
3086 }
3087
3088 unregister_netdevice(ndev);
3089 mana_cleanup_indir_table(apc);
3090
3091 rtnl_unlock();
3092
3093 free_netdev(ndev);
3094 }
3095
3096 mana_destroy_eq(ac);
3097 out:
3098 mana_gd_deregister_device(gd);
3099
3100 if (suspending)
3101 return;
3102
3103 gd->driver_data = NULL;
3104 gd->gdma_context = NULL;
3105 kfree(ac);
3106 }
3107
mana_get_primary_netdev_rcu(struct mana_context * ac,u32 port_index)3108 struct net_device *mana_get_primary_netdev_rcu(struct mana_context *ac, u32 port_index)
3109 {
3110 struct net_device *ndev;
3111
3112 RCU_LOCKDEP_WARN(!rcu_read_lock_held(),
3113 "Taking primary netdev without holding the RCU read lock");
3114 if (port_index >= ac->num_ports)
3115 return NULL;
3116
3117 /* When mana is used in netvsc, the upper netdevice should be returned. */
3118 if (ac->ports[port_index]->flags & IFF_SLAVE)
3119 ndev = netdev_master_upper_dev_get_rcu(ac->ports[port_index]);
3120 else
3121 ndev = ac->ports[port_index];
3122
3123 return ndev;
3124 }
3125 EXPORT_SYMBOL_NS(mana_get_primary_netdev_rcu, "NET_MANA");
3126