treewide: remove redundant IS_ERR() before error code check
[linux/fpc-iii.git] / drivers / net / ethernet / qlogic / qed / qed_iwarp.c
blobd2fe61a5cf567a58ebc4b7ec2bc7d72b78e623c9
1 /* QLogic qed NIC Driver
2 * Copyright (c) 2015-2017 QLogic Corporation
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
12 * conditions are met:
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
16 * disclaimer.
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and /or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
30 * SOFTWARE.
32 #include <linux/if_ether.h>
33 #include <linux/if_vlan.h>
34 #include <linux/ip.h>
35 #include <linux/ipv6.h>
36 #include <linux/spinlock.h>
37 #include <linux/tcp.h>
38 #include "qed_cxt.h"
39 #include "qed_hw.h"
40 #include "qed_ll2.h"
41 #include "qed_rdma.h"
42 #include "qed_reg_addr.h"
43 #include "qed_sp.h"
44 #include "qed_ooo.h"
46 #define QED_IWARP_ORD_DEFAULT 32
47 #define QED_IWARP_IRD_DEFAULT 32
48 #define QED_IWARP_MAX_FW_MSS 4120
50 #define QED_EP_SIG 0xecabcdef
52 struct mpa_v2_hdr {
53 __be16 ird;
54 __be16 ord;
57 #define MPA_V2_PEER2PEER_MODEL 0x8000
58 #define MPA_V2_SEND_RTR 0x4000 /* on ird */
59 #define MPA_V2_READ_RTR 0x4000 /* on ord */
60 #define MPA_V2_WRITE_RTR 0x8000
61 #define MPA_V2_IRD_ORD_MASK 0x3FFF
63 #define MPA_REV2(_mpa_rev) ((_mpa_rev) == MPA_NEGOTIATION_TYPE_ENHANCED)
65 #define QED_IWARP_INVALID_TCP_CID 0xffffffff
67 #define QED_IWARP_RCV_WND_SIZE_DEF_BB_2P (200 * 1024)
68 #define QED_IWARP_RCV_WND_SIZE_DEF_BB_4P (100 * 1024)
69 #define QED_IWARP_RCV_WND_SIZE_DEF_AH_2P (150 * 1024)
70 #define QED_IWARP_RCV_WND_SIZE_DEF_AH_4P (90 * 1024)
72 #define QED_IWARP_RCV_WND_SIZE_MIN (0xffff)
73 #define TIMESTAMP_HEADER_SIZE (12)
74 #define QED_IWARP_MAX_FIN_RT_DEFAULT (2)
76 #define QED_IWARP_TS_EN BIT(0)
77 #define QED_IWARP_DA_EN BIT(1)
78 #define QED_IWARP_PARAM_CRC_NEEDED (1)
79 #define QED_IWARP_PARAM_P2P (1)
81 #define QED_IWARP_DEF_MAX_RT_TIME (0)
82 #define QED_IWARP_DEF_CWND_FACTOR (4)
83 #define QED_IWARP_DEF_KA_MAX_PROBE_CNT (5)
84 #define QED_IWARP_DEF_KA_TIMEOUT (1200000) /* 20 min */
85 #define QED_IWARP_DEF_KA_INTERVAL (1000) /* 1 sec */
87 static int qed_iwarp_async_event(struct qed_hwfn *p_hwfn,
88 u8 fw_event_code, u16 echo,
89 union event_ring_data *data,
90 u8 fw_return_code);
92 /* Override devinfo with iWARP specific values */
93 void qed_iwarp_init_devinfo(struct qed_hwfn *p_hwfn)
95 struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
97 dev->max_inline = IWARP_REQ_MAX_INLINE_DATA_SIZE;
98 dev->max_qp = min_t(u32,
99 IWARP_MAX_QPS,
100 p_hwfn->p_rdma_info->num_qps) -
101 QED_IWARP_PREALLOC_CNT;
103 dev->max_cq = dev->max_qp;
105 dev->max_qp_resp_rd_atomic_resc = QED_IWARP_IRD_DEFAULT;
106 dev->max_qp_req_rd_atomic_resc = QED_IWARP_ORD_DEFAULT;
109 void qed_iwarp_init_hw(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
111 p_hwfn->rdma_prs_search_reg = PRS_REG_SEARCH_TCP;
112 qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
113 p_hwfn->b_rdma_enabled_in_prs = true;
116 /* We have two cid maps, one for tcp which should be used only from passive
117 * syn processing and replacing a pre-allocated ep in the list. The second
118 * for active tcp and for QPs.
120 static void qed_iwarp_cid_cleaned(struct qed_hwfn *p_hwfn, u32 cid)
122 cid -= qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
124 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
126 if (cid < QED_IWARP_PREALLOC_CNT)
127 qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map,
128 cid);
129 else
130 qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, cid);
132 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
135 void
136 qed_iwarp_init_fw_ramrod(struct qed_hwfn *p_hwfn,
137 struct iwarp_init_func_ramrod_data *p_ramrod)
139 p_ramrod->iwarp.ll2_ooo_q_index =
140 RESC_START(p_hwfn, QED_LL2_RAM_QUEUE) +
141 p_hwfn->p_rdma_info->iwarp.ll2_ooo_handle;
143 p_ramrod->tcp.max_fin_rt = QED_IWARP_MAX_FIN_RT_DEFAULT;
145 return;
148 static int qed_iwarp_alloc_cid(struct qed_hwfn *p_hwfn, u32 *cid)
150 int rc;
152 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
153 rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, cid);
154 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
155 if (rc) {
156 DP_NOTICE(p_hwfn, "Failed in allocating iwarp cid\n");
157 return rc;
159 *cid += qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
161 rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, *cid);
162 if (rc)
163 qed_iwarp_cid_cleaned(p_hwfn, *cid);
165 return rc;
168 static void qed_iwarp_set_tcp_cid(struct qed_hwfn *p_hwfn, u32 cid)
170 cid -= qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
172 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
173 qed_bmap_set_id(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, cid);
174 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
177 /* This function allocates a cid for passive tcp (called from syn receive)
178 * the reason it's separate from the regular cid allocation is because it
179 * is assured that these cids already have ilt allocated. They are preallocated
180 * to ensure that we won't need to allocate memory during syn processing
182 static int qed_iwarp_alloc_tcp_cid(struct qed_hwfn *p_hwfn, u32 *cid)
184 int rc;
186 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
188 rc = qed_rdma_bmap_alloc_id(p_hwfn,
189 &p_hwfn->p_rdma_info->tcp_cid_map, cid);
191 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
193 if (rc) {
194 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
195 "can't allocate iwarp tcp cid max-count=%d\n",
196 p_hwfn->p_rdma_info->tcp_cid_map.max_count);
198 *cid = QED_IWARP_INVALID_TCP_CID;
199 return rc;
202 *cid += qed_cxt_get_proto_cid_start(p_hwfn,
203 p_hwfn->p_rdma_info->proto);
204 return 0;
207 int qed_iwarp_create_qp(struct qed_hwfn *p_hwfn,
208 struct qed_rdma_qp *qp,
209 struct qed_rdma_create_qp_out_params *out_params)
211 struct iwarp_create_qp_ramrod_data *p_ramrod;
212 struct qed_sp_init_data init_data;
213 struct qed_spq_entry *p_ent;
214 u16 physical_queue;
215 u32 cid;
216 int rc;
218 qp->shared_queue = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
219 IWARP_SHARED_QUEUE_PAGE_SIZE,
220 &qp->shared_queue_phys_addr,
221 GFP_KERNEL);
222 if (!qp->shared_queue)
223 return -ENOMEM;
225 out_params->sq_pbl_virt = (u8 *)qp->shared_queue +
226 IWARP_SHARED_QUEUE_PAGE_SQ_PBL_OFFSET;
227 out_params->sq_pbl_phys = qp->shared_queue_phys_addr +
228 IWARP_SHARED_QUEUE_PAGE_SQ_PBL_OFFSET;
229 out_params->rq_pbl_virt = (u8 *)qp->shared_queue +
230 IWARP_SHARED_QUEUE_PAGE_RQ_PBL_OFFSET;
231 out_params->rq_pbl_phys = qp->shared_queue_phys_addr +
232 IWARP_SHARED_QUEUE_PAGE_RQ_PBL_OFFSET;
234 rc = qed_iwarp_alloc_cid(p_hwfn, &cid);
235 if (rc)
236 goto err1;
238 qp->icid = (u16)cid;
240 memset(&init_data, 0, sizeof(init_data));
241 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
242 init_data.cid = qp->icid;
243 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
245 rc = qed_sp_init_request(p_hwfn, &p_ent,
246 IWARP_RAMROD_CMD_ID_CREATE_QP,
247 PROTOCOLID_IWARP, &init_data);
248 if (rc)
249 goto err2;
251 p_ramrod = &p_ent->ramrod.iwarp_create_qp;
253 SET_FIELD(p_ramrod->flags,
254 IWARP_CREATE_QP_RAMROD_DATA_FMR_AND_RESERVED_EN,
255 qp->fmr_and_reserved_lkey);
257 SET_FIELD(p_ramrod->flags,
258 IWARP_CREATE_QP_RAMROD_DATA_SIGNALED_COMP, qp->signal_all);
260 SET_FIELD(p_ramrod->flags,
261 IWARP_CREATE_QP_RAMROD_DATA_RDMA_RD_EN,
262 qp->incoming_rdma_read_en);
264 SET_FIELD(p_ramrod->flags,
265 IWARP_CREATE_QP_RAMROD_DATA_RDMA_WR_EN,
266 qp->incoming_rdma_write_en);
268 SET_FIELD(p_ramrod->flags,
269 IWARP_CREATE_QP_RAMROD_DATA_ATOMIC_EN,
270 qp->incoming_atomic_en);
272 SET_FIELD(p_ramrod->flags,
273 IWARP_CREATE_QP_RAMROD_DATA_SRQ_FLG, qp->use_srq);
275 p_ramrod->pd = qp->pd;
276 p_ramrod->sq_num_pages = qp->sq_num_pages;
277 p_ramrod->rq_num_pages = qp->rq_num_pages;
279 p_ramrod->srq_id.srq_idx = cpu_to_le16(qp->srq_id);
280 p_ramrod->srq_id.opaque_fid = cpu_to_le16(p_hwfn->hw_info.opaque_fid);
281 p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
282 p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
284 p_ramrod->cq_cid_for_sq =
285 cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) | qp->sq_cq_id);
286 p_ramrod->cq_cid_for_rq =
287 cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) | qp->rq_cq_id);
289 p_ramrod->dpi = cpu_to_le16(qp->dpi);
291 physical_queue = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_OFLD);
292 p_ramrod->physical_q0 = cpu_to_le16(physical_queue);
293 physical_queue = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_ACK);
294 p_ramrod->physical_q1 = cpu_to_le16(physical_queue);
296 rc = qed_spq_post(p_hwfn, p_ent, NULL);
297 if (rc)
298 goto err2;
300 return rc;
302 err2:
303 qed_iwarp_cid_cleaned(p_hwfn, cid);
304 err1:
305 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
306 IWARP_SHARED_QUEUE_PAGE_SIZE,
307 qp->shared_queue, qp->shared_queue_phys_addr);
309 return rc;
312 static int qed_iwarp_modify_fw(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
314 struct iwarp_modify_qp_ramrod_data *p_ramrod;
315 struct qed_sp_init_data init_data;
316 struct qed_spq_entry *p_ent;
317 int rc;
319 /* Get SPQ entry */
320 memset(&init_data, 0, sizeof(init_data));
321 init_data.cid = qp->icid;
322 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
323 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
325 rc = qed_sp_init_request(p_hwfn, &p_ent,
326 IWARP_RAMROD_CMD_ID_MODIFY_QP,
327 p_hwfn->p_rdma_info->proto, &init_data);
328 if (rc)
329 return rc;
331 p_ramrod = &p_ent->ramrod.iwarp_modify_qp;
332 SET_FIELD(p_ramrod->flags, IWARP_MODIFY_QP_RAMROD_DATA_STATE_TRANS_EN,
333 0x1);
334 if (qp->iwarp_state == QED_IWARP_QP_STATE_CLOSING)
335 p_ramrod->transition_to_state = IWARP_MODIFY_QP_STATE_CLOSING;
336 else
337 p_ramrod->transition_to_state = IWARP_MODIFY_QP_STATE_ERROR;
339 rc = qed_spq_post(p_hwfn, p_ent, NULL);
341 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x)rc=%d\n", qp->icid, rc);
343 return rc;
346 enum qed_iwarp_qp_state qed_roce2iwarp_state(enum qed_roce_qp_state state)
348 switch (state) {
349 case QED_ROCE_QP_STATE_RESET:
350 case QED_ROCE_QP_STATE_INIT:
351 case QED_ROCE_QP_STATE_RTR:
352 return QED_IWARP_QP_STATE_IDLE;
353 case QED_ROCE_QP_STATE_RTS:
354 return QED_IWARP_QP_STATE_RTS;
355 case QED_ROCE_QP_STATE_SQD:
356 return QED_IWARP_QP_STATE_CLOSING;
357 case QED_ROCE_QP_STATE_ERR:
358 return QED_IWARP_QP_STATE_ERROR;
359 case QED_ROCE_QP_STATE_SQE:
360 return QED_IWARP_QP_STATE_TERMINATE;
361 default:
362 return QED_IWARP_QP_STATE_ERROR;
366 static enum qed_roce_qp_state
367 qed_iwarp2roce_state(enum qed_iwarp_qp_state state)
369 switch (state) {
370 case QED_IWARP_QP_STATE_IDLE:
371 return QED_ROCE_QP_STATE_INIT;
372 case QED_IWARP_QP_STATE_RTS:
373 return QED_ROCE_QP_STATE_RTS;
374 case QED_IWARP_QP_STATE_TERMINATE:
375 return QED_ROCE_QP_STATE_SQE;
376 case QED_IWARP_QP_STATE_CLOSING:
377 return QED_ROCE_QP_STATE_SQD;
378 case QED_IWARP_QP_STATE_ERROR:
379 return QED_ROCE_QP_STATE_ERR;
380 default:
381 return QED_ROCE_QP_STATE_ERR;
385 static const char * const iwarp_state_names[] = {
386 "IDLE",
387 "RTS",
388 "TERMINATE",
389 "CLOSING",
390 "ERROR",
394 qed_iwarp_modify_qp(struct qed_hwfn *p_hwfn,
395 struct qed_rdma_qp *qp,
396 enum qed_iwarp_qp_state new_state, bool internal)
398 enum qed_iwarp_qp_state prev_iw_state;
399 bool modify_fw = false;
400 int rc = 0;
402 /* modify QP can be called from upper-layer or as a result of async
403 * RST/FIN... therefore need to protect
405 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.qp_lock);
406 prev_iw_state = qp->iwarp_state;
408 if (prev_iw_state == new_state) {
409 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.qp_lock);
410 return 0;
413 switch (prev_iw_state) {
414 case QED_IWARP_QP_STATE_IDLE:
415 switch (new_state) {
416 case QED_IWARP_QP_STATE_RTS:
417 qp->iwarp_state = QED_IWARP_QP_STATE_RTS;
418 break;
419 case QED_IWARP_QP_STATE_ERROR:
420 qp->iwarp_state = QED_IWARP_QP_STATE_ERROR;
421 if (!internal)
422 modify_fw = true;
423 break;
424 default:
425 break;
427 break;
428 case QED_IWARP_QP_STATE_RTS:
429 switch (new_state) {
430 case QED_IWARP_QP_STATE_CLOSING:
431 if (!internal)
432 modify_fw = true;
434 qp->iwarp_state = QED_IWARP_QP_STATE_CLOSING;
435 break;
436 case QED_IWARP_QP_STATE_ERROR:
437 if (!internal)
438 modify_fw = true;
439 qp->iwarp_state = QED_IWARP_QP_STATE_ERROR;
440 break;
441 default:
442 break;
444 break;
445 case QED_IWARP_QP_STATE_ERROR:
446 switch (new_state) {
447 case QED_IWARP_QP_STATE_IDLE:
449 qp->iwarp_state = new_state;
450 break;
451 case QED_IWARP_QP_STATE_CLOSING:
452 /* could happen due to race... do nothing.... */
453 break;
454 default:
455 rc = -EINVAL;
457 break;
458 case QED_IWARP_QP_STATE_TERMINATE:
459 case QED_IWARP_QP_STATE_CLOSING:
460 qp->iwarp_state = new_state;
461 break;
462 default:
463 break;
466 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) %s --> %s%s\n",
467 qp->icid,
468 iwarp_state_names[prev_iw_state],
469 iwarp_state_names[qp->iwarp_state],
470 internal ? "internal" : "");
472 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.qp_lock);
474 if (modify_fw)
475 rc = qed_iwarp_modify_fw(p_hwfn, qp);
477 return rc;
480 int qed_iwarp_fw_destroy(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
482 struct qed_sp_init_data init_data;
483 struct qed_spq_entry *p_ent;
484 int rc;
486 /* Get SPQ entry */
487 memset(&init_data, 0, sizeof(init_data));
488 init_data.cid = qp->icid;
489 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
490 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
492 rc = qed_sp_init_request(p_hwfn, &p_ent,
493 IWARP_RAMROD_CMD_ID_DESTROY_QP,
494 p_hwfn->p_rdma_info->proto, &init_data);
495 if (rc)
496 return rc;
498 rc = qed_spq_post(p_hwfn, p_ent, NULL);
500 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) rc = %d\n", qp->icid, rc);
502 return rc;
505 static void qed_iwarp_destroy_ep(struct qed_hwfn *p_hwfn,
506 struct qed_iwarp_ep *ep,
507 bool remove_from_active_list)
509 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
510 sizeof(*ep->ep_buffer_virt),
511 ep->ep_buffer_virt, ep->ep_buffer_phys);
513 if (remove_from_active_list) {
514 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
515 list_del(&ep->list_entry);
516 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
519 if (ep->qp)
520 ep->qp->ep = NULL;
522 kfree(ep);
525 int qed_iwarp_destroy_qp(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
527 struct qed_iwarp_ep *ep = qp->ep;
528 int wait_count = 0;
529 int rc = 0;
531 if (qp->iwarp_state != QED_IWARP_QP_STATE_ERROR) {
532 rc = qed_iwarp_modify_qp(p_hwfn, qp,
533 QED_IWARP_QP_STATE_ERROR, false);
534 if (rc)
535 return rc;
538 /* Make sure ep is closed before returning and freeing memory. */
539 if (ep) {
540 while (READ_ONCE(ep->state) != QED_IWARP_EP_CLOSED &&
541 wait_count++ < 200)
542 msleep(100);
544 if (ep->state != QED_IWARP_EP_CLOSED)
545 DP_NOTICE(p_hwfn, "ep state close timeout state=%x\n",
546 ep->state);
548 qed_iwarp_destroy_ep(p_hwfn, ep, false);
551 rc = qed_iwarp_fw_destroy(p_hwfn, qp);
553 if (qp->shared_queue)
554 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
555 IWARP_SHARED_QUEUE_PAGE_SIZE,
556 qp->shared_queue, qp->shared_queue_phys_addr);
558 return rc;
561 static int
562 qed_iwarp_create_ep(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep **ep_out)
564 struct qed_iwarp_ep *ep;
565 int rc;
567 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
568 if (!ep)
569 return -ENOMEM;
571 ep->state = QED_IWARP_EP_INIT;
573 ep->ep_buffer_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
574 sizeof(*ep->ep_buffer_virt),
575 &ep->ep_buffer_phys,
576 GFP_KERNEL);
577 if (!ep->ep_buffer_virt) {
578 rc = -ENOMEM;
579 goto err;
582 ep->sig = QED_EP_SIG;
584 *ep_out = ep;
586 return 0;
588 err:
589 kfree(ep);
590 return rc;
593 static void
594 qed_iwarp_print_tcp_ramrod(struct qed_hwfn *p_hwfn,
595 struct iwarp_tcp_offload_ramrod_data *p_tcp_ramrod)
597 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "local_mac=%x %x %x, remote_mac=%x %x %x\n",
598 p_tcp_ramrod->tcp.local_mac_addr_lo,
599 p_tcp_ramrod->tcp.local_mac_addr_mid,
600 p_tcp_ramrod->tcp.local_mac_addr_hi,
601 p_tcp_ramrod->tcp.remote_mac_addr_lo,
602 p_tcp_ramrod->tcp.remote_mac_addr_mid,
603 p_tcp_ramrod->tcp.remote_mac_addr_hi);
605 if (p_tcp_ramrod->tcp.ip_version == TCP_IPV4) {
606 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
607 "local_ip=%pI4h:%x, remote_ip=%pI4h:%x, vlan=%x\n",
608 p_tcp_ramrod->tcp.local_ip,
609 p_tcp_ramrod->tcp.local_port,
610 p_tcp_ramrod->tcp.remote_ip,
611 p_tcp_ramrod->tcp.remote_port,
612 p_tcp_ramrod->tcp.vlan_id);
613 } else {
614 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
615 "local_ip=%pI6:%x, remote_ip=%pI6:%x, vlan=%x\n",
616 p_tcp_ramrod->tcp.local_ip,
617 p_tcp_ramrod->tcp.local_port,
618 p_tcp_ramrod->tcp.remote_ip,
619 p_tcp_ramrod->tcp.remote_port,
620 p_tcp_ramrod->tcp.vlan_id);
623 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
624 "flow_label=%x, ttl=%x, tos_or_tc=%x, mss=%x, rcv_wnd_scale=%x, connect_mode=%x, flags=%x\n",
625 p_tcp_ramrod->tcp.flow_label,
626 p_tcp_ramrod->tcp.ttl,
627 p_tcp_ramrod->tcp.tos_or_tc,
628 p_tcp_ramrod->tcp.mss,
629 p_tcp_ramrod->tcp.rcv_wnd_scale,
630 p_tcp_ramrod->tcp.connect_mode,
631 p_tcp_ramrod->tcp.flags);
633 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "syn_ip_payload_length=%x, lo=%x, hi=%x\n",
634 p_tcp_ramrod->tcp.syn_ip_payload_length,
635 p_tcp_ramrod->tcp.syn_phy_addr_lo,
636 p_tcp_ramrod->tcp.syn_phy_addr_hi);
639 static int
640 qed_iwarp_tcp_offload(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
642 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
643 struct iwarp_tcp_offload_ramrod_data *p_tcp_ramrod;
644 struct tcp_offload_params_opt2 *tcp;
645 struct qed_sp_init_data init_data;
646 struct qed_spq_entry *p_ent;
647 dma_addr_t async_output_phys;
648 dma_addr_t in_pdata_phys;
649 u16 physical_q;
650 u8 tcp_flags;
651 int rc;
652 int i;
654 memset(&init_data, 0, sizeof(init_data));
655 init_data.cid = ep->tcp_cid;
656 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
657 if (ep->connect_mode == TCP_CONNECT_PASSIVE)
658 init_data.comp_mode = QED_SPQ_MODE_CB;
659 else
660 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
662 rc = qed_sp_init_request(p_hwfn, &p_ent,
663 IWARP_RAMROD_CMD_ID_TCP_OFFLOAD,
664 PROTOCOLID_IWARP, &init_data);
665 if (rc)
666 return rc;
668 p_tcp_ramrod = &p_ent->ramrod.iwarp_tcp_offload;
670 in_pdata_phys = ep->ep_buffer_phys +
671 offsetof(struct qed_iwarp_ep_memory, in_pdata);
672 DMA_REGPAIR_LE(p_tcp_ramrod->iwarp.incoming_ulp_buffer.addr,
673 in_pdata_phys);
675 p_tcp_ramrod->iwarp.incoming_ulp_buffer.len =
676 cpu_to_le16(sizeof(ep->ep_buffer_virt->in_pdata));
678 async_output_phys = ep->ep_buffer_phys +
679 offsetof(struct qed_iwarp_ep_memory, async_output);
680 DMA_REGPAIR_LE(p_tcp_ramrod->iwarp.async_eqe_output_buf,
681 async_output_phys);
683 p_tcp_ramrod->iwarp.handle_for_async.hi = cpu_to_le32(PTR_HI(ep));
684 p_tcp_ramrod->iwarp.handle_for_async.lo = cpu_to_le32(PTR_LO(ep));
686 physical_q = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_OFLD);
687 p_tcp_ramrod->iwarp.physical_q0 = cpu_to_le16(physical_q);
688 physical_q = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_ACK);
689 p_tcp_ramrod->iwarp.physical_q1 = cpu_to_le16(physical_q);
690 p_tcp_ramrod->iwarp.mpa_mode = iwarp_info->mpa_rev;
692 tcp = &p_tcp_ramrod->tcp;
693 qed_set_fw_mac_addr(&tcp->remote_mac_addr_hi,
694 &tcp->remote_mac_addr_mid,
695 &tcp->remote_mac_addr_lo, ep->remote_mac_addr);
696 qed_set_fw_mac_addr(&tcp->local_mac_addr_hi, &tcp->local_mac_addr_mid,
697 &tcp->local_mac_addr_lo, ep->local_mac_addr);
699 tcp->vlan_id = cpu_to_le16(ep->cm_info.vlan);
701 tcp_flags = p_hwfn->p_rdma_info->iwarp.tcp_flags;
702 tcp->flags = 0;
703 SET_FIELD(tcp->flags, TCP_OFFLOAD_PARAMS_OPT2_TS_EN,
704 !!(tcp_flags & QED_IWARP_TS_EN));
706 SET_FIELD(tcp->flags, TCP_OFFLOAD_PARAMS_OPT2_DA_EN,
707 !!(tcp_flags & QED_IWARP_DA_EN));
709 tcp->ip_version = ep->cm_info.ip_version;
711 for (i = 0; i < 4; i++) {
712 tcp->remote_ip[i] = cpu_to_le32(ep->cm_info.remote_ip[i]);
713 tcp->local_ip[i] = cpu_to_le32(ep->cm_info.local_ip[i]);
716 tcp->remote_port = cpu_to_le16(ep->cm_info.remote_port);
717 tcp->local_port = cpu_to_le16(ep->cm_info.local_port);
718 tcp->mss = cpu_to_le16(ep->mss);
719 tcp->flow_label = 0;
720 tcp->ttl = 0x40;
721 tcp->tos_or_tc = 0;
723 tcp->max_rt_time = QED_IWARP_DEF_MAX_RT_TIME;
724 tcp->cwnd = QED_IWARP_DEF_CWND_FACTOR * tcp->mss;
725 tcp->ka_max_probe_cnt = QED_IWARP_DEF_KA_MAX_PROBE_CNT;
726 tcp->ka_timeout = QED_IWARP_DEF_KA_TIMEOUT;
727 tcp->ka_interval = QED_IWARP_DEF_KA_INTERVAL;
729 tcp->rcv_wnd_scale = (u8)p_hwfn->p_rdma_info->iwarp.rcv_wnd_scale;
730 tcp->connect_mode = ep->connect_mode;
732 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
733 tcp->syn_ip_payload_length =
734 cpu_to_le16(ep->syn_ip_payload_length);
735 tcp->syn_phy_addr_hi = DMA_HI_LE(ep->syn_phy_addr);
736 tcp->syn_phy_addr_lo = DMA_LO_LE(ep->syn_phy_addr);
739 qed_iwarp_print_tcp_ramrod(p_hwfn, p_tcp_ramrod);
741 rc = qed_spq_post(p_hwfn, p_ent, NULL);
743 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
744 "EP(0x%x) Offload completed rc=%d\n", ep->tcp_cid, rc);
746 return rc;
749 static void
750 qed_iwarp_mpa_received(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
752 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
753 struct qed_iwarp_cm_event_params params;
754 struct mpa_v2_hdr *mpa_v2;
755 union async_output *async_data;
756 u16 mpa_ord, mpa_ird;
757 u8 mpa_hdr_size = 0;
758 u8 mpa_rev;
760 async_data = &ep->ep_buffer_virt->async_output;
762 mpa_rev = async_data->mpa_request.mpa_handshake_mode;
763 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
764 "private_data_len=%x handshake_mode=%x private_data=(%x)\n",
765 async_data->mpa_request.ulp_data_len,
766 mpa_rev, *((u32 *)(ep->ep_buffer_virt->in_pdata)));
768 if (mpa_rev == MPA_NEGOTIATION_TYPE_ENHANCED) {
769 /* Read ord/ird values from private data buffer */
770 mpa_v2 = (struct mpa_v2_hdr *)ep->ep_buffer_virt->in_pdata;
771 mpa_hdr_size = sizeof(*mpa_v2);
773 mpa_ord = ntohs(mpa_v2->ord);
774 mpa_ird = ntohs(mpa_v2->ird);
776 /* Temprary store in cm_info incoming ord/ird requested, later
777 * replace with negotiated value during accept
779 ep->cm_info.ord = (u8)min_t(u16,
780 (mpa_ord & MPA_V2_IRD_ORD_MASK),
781 QED_IWARP_ORD_DEFAULT);
783 ep->cm_info.ird = (u8)min_t(u16,
784 (mpa_ird & MPA_V2_IRD_ORD_MASK),
785 QED_IWARP_IRD_DEFAULT);
787 /* Peer2Peer negotiation */
788 ep->rtr_type = MPA_RTR_TYPE_NONE;
789 if (mpa_ird & MPA_V2_PEER2PEER_MODEL) {
790 if (mpa_ord & MPA_V2_WRITE_RTR)
791 ep->rtr_type |= MPA_RTR_TYPE_ZERO_WRITE;
793 if (mpa_ord & MPA_V2_READ_RTR)
794 ep->rtr_type |= MPA_RTR_TYPE_ZERO_READ;
796 if (mpa_ird & MPA_V2_SEND_RTR)
797 ep->rtr_type |= MPA_RTR_TYPE_ZERO_SEND;
799 ep->rtr_type &= iwarp_info->rtr_type;
801 /* if we're left with no match send our capabilities */
802 if (ep->rtr_type == MPA_RTR_TYPE_NONE)
803 ep->rtr_type = iwarp_info->rtr_type;
806 ep->mpa_rev = MPA_NEGOTIATION_TYPE_ENHANCED;
807 } else {
808 ep->cm_info.ord = QED_IWARP_ORD_DEFAULT;
809 ep->cm_info.ird = QED_IWARP_IRD_DEFAULT;
810 ep->mpa_rev = MPA_NEGOTIATION_TYPE_BASIC;
813 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
814 "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x rtr:0x%x ulp_data_len = %x mpa_hdr_size = %x\n",
815 mpa_rev, ep->cm_info.ord, ep->cm_info.ird, ep->rtr_type,
816 async_data->mpa_request.ulp_data_len, mpa_hdr_size);
818 /* Strip mpa v2 hdr from private data before sending to upper layer */
819 ep->cm_info.private_data = ep->ep_buffer_virt->in_pdata + mpa_hdr_size;
821 ep->cm_info.private_data_len = async_data->mpa_request.ulp_data_len -
822 mpa_hdr_size;
824 params.event = QED_IWARP_EVENT_MPA_REQUEST;
825 params.cm_info = &ep->cm_info;
826 params.ep_context = ep;
827 params.status = 0;
829 ep->state = QED_IWARP_EP_MPA_REQ_RCVD;
830 ep->event_cb(ep->cb_context, &params);
833 static int
834 qed_iwarp_mpa_offload(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
836 struct iwarp_mpa_offload_ramrod_data *p_mpa_ramrod;
837 struct qed_iwarp_info *iwarp_info;
838 struct qed_sp_init_data init_data;
839 dma_addr_t async_output_phys;
840 struct qed_spq_entry *p_ent;
841 dma_addr_t out_pdata_phys;
842 dma_addr_t in_pdata_phys;
843 struct qed_rdma_qp *qp;
844 bool reject;
845 int rc;
847 if (!ep)
848 return -EINVAL;
850 qp = ep->qp;
851 reject = !qp;
853 memset(&init_data, 0, sizeof(init_data));
854 init_data.cid = reject ? ep->tcp_cid : qp->icid;
855 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
857 if (ep->connect_mode == TCP_CONNECT_ACTIVE)
858 init_data.comp_mode = QED_SPQ_MODE_CB;
859 else
860 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
862 rc = qed_sp_init_request(p_hwfn, &p_ent,
863 IWARP_RAMROD_CMD_ID_MPA_OFFLOAD,
864 PROTOCOLID_IWARP, &init_data);
865 if (rc)
866 return rc;
868 p_mpa_ramrod = &p_ent->ramrod.iwarp_mpa_offload;
869 out_pdata_phys = ep->ep_buffer_phys +
870 offsetof(struct qed_iwarp_ep_memory, out_pdata);
871 DMA_REGPAIR_LE(p_mpa_ramrod->common.outgoing_ulp_buffer.addr,
872 out_pdata_phys);
873 p_mpa_ramrod->common.outgoing_ulp_buffer.len =
874 ep->cm_info.private_data_len;
875 p_mpa_ramrod->common.crc_needed = p_hwfn->p_rdma_info->iwarp.crc_needed;
877 p_mpa_ramrod->common.out_rq.ord = ep->cm_info.ord;
878 p_mpa_ramrod->common.out_rq.ird = ep->cm_info.ird;
880 p_mpa_ramrod->tcp_cid = p_hwfn->hw_info.opaque_fid << 16 | ep->tcp_cid;
882 in_pdata_phys = ep->ep_buffer_phys +
883 offsetof(struct qed_iwarp_ep_memory, in_pdata);
884 p_mpa_ramrod->tcp_connect_side = ep->connect_mode;
885 DMA_REGPAIR_LE(p_mpa_ramrod->incoming_ulp_buffer.addr,
886 in_pdata_phys);
887 p_mpa_ramrod->incoming_ulp_buffer.len =
888 cpu_to_le16(sizeof(ep->ep_buffer_virt->in_pdata));
889 async_output_phys = ep->ep_buffer_phys +
890 offsetof(struct qed_iwarp_ep_memory, async_output);
891 DMA_REGPAIR_LE(p_mpa_ramrod->async_eqe_output_buf,
892 async_output_phys);
893 p_mpa_ramrod->handle_for_async.hi = cpu_to_le32(PTR_HI(ep));
894 p_mpa_ramrod->handle_for_async.lo = cpu_to_le32(PTR_LO(ep));
896 if (!reject) {
897 DMA_REGPAIR_LE(p_mpa_ramrod->shared_queue_addr,
898 qp->shared_queue_phys_addr);
899 p_mpa_ramrod->stats_counter_id =
900 RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) + qp->stats_queue;
901 } else {
902 p_mpa_ramrod->common.reject = 1;
905 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
906 p_mpa_ramrod->rcv_wnd = iwarp_info->rcv_wnd_size;
907 p_mpa_ramrod->mode = ep->mpa_rev;
908 SET_FIELD(p_mpa_ramrod->rtr_pref,
909 IWARP_MPA_OFFLOAD_RAMROD_DATA_RTR_SUPPORTED, ep->rtr_type);
911 ep->state = QED_IWARP_EP_MPA_OFFLOADED;
912 rc = qed_spq_post(p_hwfn, p_ent, NULL);
913 if (!reject)
914 ep->cid = qp->icid; /* Now they're migrated. */
916 DP_VERBOSE(p_hwfn,
917 QED_MSG_RDMA,
918 "QP(0x%x) EP(0x%x) MPA Offload rc = %d IRD=0x%x ORD=0x%x rtr_type=%d mpa_rev=%d reject=%d\n",
919 reject ? 0xffff : qp->icid,
920 ep->tcp_cid,
922 ep->cm_info.ird,
923 ep->cm_info.ord, ep->rtr_type, ep->mpa_rev, reject);
924 return rc;
927 static void
928 qed_iwarp_return_ep(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
930 ep->state = QED_IWARP_EP_INIT;
931 if (ep->qp)
932 ep->qp->ep = NULL;
933 ep->qp = NULL;
934 memset(&ep->cm_info, 0, sizeof(ep->cm_info));
936 if (ep->tcp_cid == QED_IWARP_INVALID_TCP_CID) {
937 /* We don't care about the return code, it's ok if tcp_cid
938 * remains invalid...in this case we'll defer allocation
940 qed_iwarp_alloc_tcp_cid(p_hwfn, &ep->tcp_cid);
942 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
944 list_move_tail(&ep->list_entry,
945 &p_hwfn->p_rdma_info->iwarp.ep_free_list);
947 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
950 static void
951 qed_iwarp_parse_private_data(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
953 struct mpa_v2_hdr *mpa_v2_params;
954 union async_output *async_data;
955 u16 mpa_ird, mpa_ord;
956 u8 mpa_data_size = 0;
958 if (MPA_REV2(p_hwfn->p_rdma_info->iwarp.mpa_rev)) {
959 mpa_v2_params =
960 (struct mpa_v2_hdr *)(ep->ep_buffer_virt->in_pdata);
961 mpa_data_size = sizeof(*mpa_v2_params);
962 mpa_ird = ntohs(mpa_v2_params->ird);
963 mpa_ord = ntohs(mpa_v2_params->ord);
965 ep->cm_info.ird = (u8)(mpa_ord & MPA_V2_IRD_ORD_MASK);
966 ep->cm_info.ord = (u8)(mpa_ird & MPA_V2_IRD_ORD_MASK);
968 async_data = &ep->ep_buffer_virt->async_output;
970 ep->cm_info.private_data = ep->ep_buffer_virt->in_pdata + mpa_data_size;
971 ep->cm_info.private_data_len = async_data->mpa_response.ulp_data_len -
972 mpa_data_size;
975 static void
976 qed_iwarp_mpa_reply_arrived(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
978 struct qed_iwarp_cm_event_params params;
980 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
981 DP_NOTICE(p_hwfn,
982 "MPA reply event not expected on passive side!\n");
983 return;
986 params.event = QED_IWARP_EVENT_ACTIVE_MPA_REPLY;
988 qed_iwarp_parse_private_data(p_hwfn, ep);
990 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
991 "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x\n",
992 ep->mpa_rev, ep->cm_info.ord, ep->cm_info.ird);
994 params.cm_info = &ep->cm_info;
995 params.ep_context = ep;
996 params.status = 0;
998 ep->mpa_reply_processed = true;
1000 ep->event_cb(ep->cb_context, &params);
1003 #define QED_IWARP_CONNECT_MODE_STRING(ep) \
1004 ((ep)->connect_mode == TCP_CONNECT_PASSIVE) ? "Passive" : "Active"
1006 /* Called as a result of the event:
1007 * IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE
1009 static void
1010 qed_iwarp_mpa_complete(struct qed_hwfn *p_hwfn,
1011 struct qed_iwarp_ep *ep, u8 fw_return_code)
1013 struct qed_iwarp_cm_event_params params;
1015 if (ep->connect_mode == TCP_CONNECT_ACTIVE)
1016 params.event = QED_IWARP_EVENT_ACTIVE_COMPLETE;
1017 else
1018 params.event = QED_IWARP_EVENT_PASSIVE_COMPLETE;
1020 if (ep->connect_mode == TCP_CONNECT_ACTIVE && !ep->mpa_reply_processed)
1021 qed_iwarp_parse_private_data(p_hwfn, ep);
1023 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1024 "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x\n",
1025 ep->mpa_rev, ep->cm_info.ord, ep->cm_info.ird);
1027 params.cm_info = &ep->cm_info;
1029 params.ep_context = ep;
1031 switch (fw_return_code) {
1032 case RDMA_RETURN_OK:
1033 ep->qp->max_rd_atomic_req = ep->cm_info.ord;
1034 ep->qp->max_rd_atomic_resp = ep->cm_info.ird;
1035 qed_iwarp_modify_qp(p_hwfn, ep->qp, QED_IWARP_QP_STATE_RTS, 1);
1036 ep->state = QED_IWARP_EP_ESTABLISHED;
1037 params.status = 0;
1038 break;
1039 case IWARP_CONN_ERROR_MPA_TIMEOUT:
1040 DP_NOTICE(p_hwfn, "%s(0x%x) MPA timeout\n",
1041 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1042 params.status = -EBUSY;
1043 break;
1044 case IWARP_CONN_ERROR_MPA_ERROR_REJECT:
1045 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Reject\n",
1046 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1047 params.status = -ECONNREFUSED;
1048 break;
1049 case IWARP_CONN_ERROR_MPA_RST:
1050 DP_NOTICE(p_hwfn, "%s(0x%x) MPA reset(tcp cid: 0x%x)\n",
1051 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid,
1052 ep->tcp_cid);
1053 params.status = -ECONNRESET;
1054 break;
1055 case IWARP_CONN_ERROR_MPA_FIN:
1056 DP_NOTICE(p_hwfn, "%s(0x%x) MPA received FIN\n",
1057 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1058 params.status = -ECONNREFUSED;
1059 break;
1060 case IWARP_CONN_ERROR_MPA_INSUF_IRD:
1061 DP_NOTICE(p_hwfn, "%s(0x%x) MPA insufficient ird\n",
1062 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1063 params.status = -ECONNREFUSED;
1064 break;
1065 case IWARP_CONN_ERROR_MPA_RTR_MISMATCH:
1066 DP_NOTICE(p_hwfn, "%s(0x%x) MPA RTR MISMATCH\n",
1067 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1068 params.status = -ECONNREFUSED;
1069 break;
1070 case IWARP_CONN_ERROR_MPA_INVALID_PACKET:
1071 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Invalid Packet\n",
1072 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1073 params.status = -ECONNREFUSED;
1074 break;
1075 case IWARP_CONN_ERROR_MPA_LOCAL_ERROR:
1076 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Local Error\n",
1077 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1078 params.status = -ECONNREFUSED;
1079 break;
1080 case IWARP_CONN_ERROR_MPA_TERMINATE:
1081 DP_NOTICE(p_hwfn, "%s(0x%x) MPA TERMINATE\n",
1082 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1083 params.status = -ECONNREFUSED;
1084 break;
1085 default:
1086 params.status = -ECONNRESET;
1087 break;
1090 if (fw_return_code != RDMA_RETURN_OK)
1091 /* paired with READ_ONCE in destroy_qp */
1092 smp_store_release(&ep->state, QED_IWARP_EP_CLOSED);
1094 ep->event_cb(ep->cb_context, &params);
1096 /* on passive side, if there is no associated QP (REJECT) we need to
1097 * return the ep to the pool, (in the regular case we add an element
1098 * in accept instead of this one.
1099 * In both cases we need to remove it from the ep_list.
1101 if (fw_return_code != RDMA_RETURN_OK) {
1102 ep->tcp_cid = QED_IWARP_INVALID_TCP_CID;
1103 if ((ep->connect_mode == TCP_CONNECT_PASSIVE) &&
1104 (!ep->qp)) { /* Rejected */
1105 qed_iwarp_return_ep(p_hwfn, ep);
1106 } else {
1107 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1108 list_del(&ep->list_entry);
1109 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1114 static void
1115 qed_iwarp_mpa_v2_set_private(struct qed_hwfn *p_hwfn,
1116 struct qed_iwarp_ep *ep, u8 *mpa_data_size)
1118 struct mpa_v2_hdr *mpa_v2_params;
1119 u16 mpa_ird, mpa_ord;
1121 *mpa_data_size = 0;
1122 if (MPA_REV2(ep->mpa_rev)) {
1123 mpa_v2_params =
1124 (struct mpa_v2_hdr *)ep->ep_buffer_virt->out_pdata;
1125 *mpa_data_size = sizeof(*mpa_v2_params);
1127 mpa_ird = (u16)ep->cm_info.ird;
1128 mpa_ord = (u16)ep->cm_info.ord;
1130 if (ep->rtr_type != MPA_RTR_TYPE_NONE) {
1131 mpa_ird |= MPA_V2_PEER2PEER_MODEL;
1133 if (ep->rtr_type & MPA_RTR_TYPE_ZERO_SEND)
1134 mpa_ird |= MPA_V2_SEND_RTR;
1136 if (ep->rtr_type & MPA_RTR_TYPE_ZERO_WRITE)
1137 mpa_ord |= MPA_V2_WRITE_RTR;
1139 if (ep->rtr_type & MPA_RTR_TYPE_ZERO_READ)
1140 mpa_ord |= MPA_V2_READ_RTR;
1143 mpa_v2_params->ird = htons(mpa_ird);
1144 mpa_v2_params->ord = htons(mpa_ord);
1146 DP_VERBOSE(p_hwfn,
1147 QED_MSG_RDMA,
1148 "MPA_NEGOTIATE Header: [%x ord:%x ird] %x ord:%x ird:%x peer2peer:%x rtr_send:%x rtr_write:%x rtr_read:%x\n",
1149 mpa_v2_params->ird,
1150 mpa_v2_params->ord,
1151 *((u32 *)mpa_v2_params),
1152 mpa_ord & MPA_V2_IRD_ORD_MASK,
1153 mpa_ird & MPA_V2_IRD_ORD_MASK,
1154 !!(mpa_ird & MPA_V2_PEER2PEER_MODEL),
1155 !!(mpa_ird & MPA_V2_SEND_RTR),
1156 !!(mpa_ord & MPA_V2_WRITE_RTR),
1157 !!(mpa_ord & MPA_V2_READ_RTR));
1161 int qed_iwarp_connect(void *rdma_cxt,
1162 struct qed_iwarp_connect_in *iparams,
1163 struct qed_iwarp_connect_out *oparams)
1165 struct qed_hwfn *p_hwfn = rdma_cxt;
1166 struct qed_iwarp_info *iwarp_info;
1167 struct qed_iwarp_ep *ep;
1168 u8 mpa_data_size = 0;
1169 u32 cid;
1170 int rc;
1172 if ((iparams->cm_info.ord > QED_IWARP_ORD_DEFAULT) ||
1173 (iparams->cm_info.ird > QED_IWARP_IRD_DEFAULT)) {
1174 DP_NOTICE(p_hwfn,
1175 "QP(0x%x) ERROR: Invalid ord(0x%x)/ird(0x%x)\n",
1176 iparams->qp->icid, iparams->cm_info.ord,
1177 iparams->cm_info.ird);
1179 return -EINVAL;
1182 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
1184 /* Allocate ep object */
1185 rc = qed_iwarp_alloc_cid(p_hwfn, &cid);
1186 if (rc)
1187 return rc;
1189 rc = qed_iwarp_create_ep(p_hwfn, &ep);
1190 if (rc)
1191 goto err;
1193 ep->tcp_cid = cid;
1195 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1196 list_add_tail(&ep->list_entry, &p_hwfn->p_rdma_info->iwarp.ep_list);
1197 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1199 ep->qp = iparams->qp;
1200 ep->qp->ep = ep;
1201 ether_addr_copy(ep->remote_mac_addr, iparams->remote_mac_addr);
1202 ether_addr_copy(ep->local_mac_addr, iparams->local_mac_addr);
1203 memcpy(&ep->cm_info, &iparams->cm_info, sizeof(ep->cm_info));
1205 ep->cm_info.ord = iparams->cm_info.ord;
1206 ep->cm_info.ird = iparams->cm_info.ird;
1208 ep->rtr_type = iwarp_info->rtr_type;
1209 if (!iwarp_info->peer2peer)
1210 ep->rtr_type = MPA_RTR_TYPE_NONE;
1212 if ((ep->rtr_type & MPA_RTR_TYPE_ZERO_READ) && (ep->cm_info.ord == 0))
1213 ep->cm_info.ord = 1;
1215 ep->mpa_rev = iwarp_info->mpa_rev;
1217 qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
1219 ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
1220 ep->cm_info.private_data_len = iparams->cm_info.private_data_len +
1221 mpa_data_size;
1223 memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
1224 iparams->cm_info.private_data,
1225 iparams->cm_info.private_data_len);
1227 ep->mss = iparams->mss;
1228 ep->mss = min_t(u16, QED_IWARP_MAX_FW_MSS, ep->mss);
1230 ep->event_cb = iparams->event_cb;
1231 ep->cb_context = iparams->cb_context;
1232 ep->connect_mode = TCP_CONNECT_ACTIVE;
1234 oparams->ep_context = ep;
1236 rc = qed_iwarp_tcp_offload(p_hwfn, ep);
1238 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) EP(0x%x) rc = %d\n",
1239 iparams->qp->icid, ep->tcp_cid, rc);
1241 if (rc) {
1242 qed_iwarp_destroy_ep(p_hwfn, ep, true);
1243 goto err;
1246 return rc;
1247 err:
1248 qed_iwarp_cid_cleaned(p_hwfn, cid);
1250 return rc;
1253 static struct qed_iwarp_ep *qed_iwarp_get_free_ep(struct qed_hwfn *p_hwfn)
1255 struct qed_iwarp_ep *ep = NULL;
1256 int rc;
1258 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1260 if (list_empty(&p_hwfn->p_rdma_info->iwarp.ep_free_list)) {
1261 DP_ERR(p_hwfn, "Ep list is empty\n");
1262 goto out;
1265 ep = list_first_entry(&p_hwfn->p_rdma_info->iwarp.ep_free_list,
1266 struct qed_iwarp_ep, list_entry);
1268 /* in some cases we could have failed allocating a tcp cid when added
1269 * from accept / failure... retry now..this is not the common case.
1271 if (ep->tcp_cid == QED_IWARP_INVALID_TCP_CID) {
1272 rc = qed_iwarp_alloc_tcp_cid(p_hwfn, &ep->tcp_cid);
1274 /* if we fail we could look for another entry with a valid
1275 * tcp_cid, but since we don't expect to reach this anyway
1276 * it's not worth the handling
1278 if (rc) {
1279 ep->tcp_cid = QED_IWARP_INVALID_TCP_CID;
1280 ep = NULL;
1281 goto out;
1285 list_del(&ep->list_entry);
1287 out:
1288 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1289 return ep;
1292 #define QED_IWARP_MAX_CID_CLEAN_TIME 100
1293 #define QED_IWARP_MAX_NO_PROGRESS_CNT 5
1295 /* This function waits for all the bits of a bmap to be cleared, as long as
1296 * there is progress ( i.e. the number of bits left to be cleared decreases )
1297 * the function continues.
1299 static int
1300 qed_iwarp_wait_cid_map_cleared(struct qed_hwfn *p_hwfn, struct qed_bmap *bmap)
1302 int prev_weight = 0;
1303 int wait_count = 0;
1304 int weight = 0;
1306 weight = bitmap_weight(bmap->bitmap, bmap->max_count);
1307 prev_weight = weight;
1309 while (weight) {
1310 msleep(QED_IWARP_MAX_CID_CLEAN_TIME);
1312 weight = bitmap_weight(bmap->bitmap, bmap->max_count);
1314 if (prev_weight == weight) {
1315 wait_count++;
1316 } else {
1317 prev_weight = weight;
1318 wait_count = 0;
1321 if (wait_count > QED_IWARP_MAX_NO_PROGRESS_CNT) {
1322 DP_NOTICE(p_hwfn,
1323 "%s bitmap wait timed out (%d cids pending)\n",
1324 bmap->name, weight);
1325 return -EBUSY;
1328 return 0;
1331 static int qed_iwarp_wait_for_all_cids(struct qed_hwfn *p_hwfn)
1333 int rc;
1334 int i;
1336 rc = qed_iwarp_wait_cid_map_cleared(p_hwfn,
1337 &p_hwfn->p_rdma_info->tcp_cid_map);
1338 if (rc)
1339 return rc;
1341 /* Now free the tcp cids from the main cid map */
1342 for (i = 0; i < QED_IWARP_PREALLOC_CNT; i++)
1343 qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, i);
1345 /* Now wait for all cids to be completed */
1346 return qed_iwarp_wait_cid_map_cleared(p_hwfn,
1347 &p_hwfn->p_rdma_info->cid_map);
1350 static void qed_iwarp_free_prealloc_ep(struct qed_hwfn *p_hwfn)
1352 struct qed_iwarp_ep *ep;
1354 while (!list_empty(&p_hwfn->p_rdma_info->iwarp.ep_free_list)) {
1355 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1357 ep = list_first_entry(&p_hwfn->p_rdma_info->iwarp.ep_free_list,
1358 struct qed_iwarp_ep, list_entry);
1360 if (!ep) {
1361 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1362 break;
1364 list_del(&ep->list_entry);
1366 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1368 if (ep->tcp_cid != QED_IWARP_INVALID_TCP_CID)
1369 qed_iwarp_cid_cleaned(p_hwfn, ep->tcp_cid);
1371 qed_iwarp_destroy_ep(p_hwfn, ep, false);
1375 static int qed_iwarp_prealloc_ep(struct qed_hwfn *p_hwfn, bool init)
1377 struct qed_iwarp_ep *ep;
1378 int rc = 0;
1379 int count;
1380 u32 cid;
1381 int i;
1383 count = init ? QED_IWARP_PREALLOC_CNT : 1;
1384 for (i = 0; i < count; i++) {
1385 rc = qed_iwarp_create_ep(p_hwfn, &ep);
1386 if (rc)
1387 return rc;
1389 /* During initialization we allocate from the main pool,
1390 * afterwards we allocate only from the tcp_cid.
1392 if (init) {
1393 rc = qed_iwarp_alloc_cid(p_hwfn, &cid);
1394 if (rc)
1395 goto err;
1396 qed_iwarp_set_tcp_cid(p_hwfn, cid);
1397 } else {
1398 /* We don't care about the return code, it's ok if
1399 * tcp_cid remains invalid...in this case we'll
1400 * defer allocation
1402 qed_iwarp_alloc_tcp_cid(p_hwfn, &cid);
1405 ep->tcp_cid = cid;
1407 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1408 list_add_tail(&ep->list_entry,
1409 &p_hwfn->p_rdma_info->iwarp.ep_free_list);
1410 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1413 return rc;
1415 err:
1416 qed_iwarp_destroy_ep(p_hwfn, ep, false);
1418 return rc;
1421 int qed_iwarp_alloc(struct qed_hwfn *p_hwfn)
1423 int rc;
1425 /* Allocate bitmap for tcp cid. These are used by passive side
1426 * to ensure it can allocate a tcp cid during dpc that was
1427 * pre-acquired and doesn't require dynamic allocation of ilt
1429 rc = qed_rdma_bmap_alloc(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map,
1430 QED_IWARP_PREALLOC_CNT, "TCP_CID");
1431 if (rc) {
1432 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1433 "Failed to allocate tcp cid, rc = %d\n", rc);
1434 return rc;
1437 INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.ep_free_list);
1438 spin_lock_init(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1440 rc = qed_iwarp_prealloc_ep(p_hwfn, true);
1441 if (rc)
1442 return rc;
1444 return qed_ooo_alloc(p_hwfn);
1447 void qed_iwarp_resc_free(struct qed_hwfn *p_hwfn)
1449 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
1451 qed_ooo_free(p_hwfn);
1452 qed_rdma_bmap_free(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, 1);
1453 kfree(iwarp_info->mpa_bufs);
1454 kfree(iwarp_info->partial_fpdus);
1455 kfree(iwarp_info->mpa_intermediate_buf);
1458 int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams)
1460 struct qed_hwfn *p_hwfn = rdma_cxt;
1461 struct qed_iwarp_ep *ep;
1462 u8 mpa_data_size = 0;
1463 int rc;
1465 ep = iparams->ep_context;
1466 if (!ep) {
1467 DP_ERR(p_hwfn, "Ep Context receive in accept is NULL\n");
1468 return -EINVAL;
1471 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) EP(0x%x)\n",
1472 iparams->qp->icid, ep->tcp_cid);
1474 if ((iparams->ord > QED_IWARP_ORD_DEFAULT) ||
1475 (iparams->ird > QED_IWARP_IRD_DEFAULT)) {
1476 DP_VERBOSE(p_hwfn,
1477 QED_MSG_RDMA,
1478 "QP(0x%x) EP(0x%x) ERROR: Invalid ord(0x%x)/ird(0x%x)\n",
1479 iparams->qp->icid,
1480 ep->tcp_cid, iparams->ord, iparams->ord);
1481 return -EINVAL;
1484 qed_iwarp_prealloc_ep(p_hwfn, false);
1486 ep->cb_context = iparams->cb_context;
1487 ep->qp = iparams->qp;
1488 ep->qp->ep = ep;
1490 if (ep->mpa_rev == MPA_NEGOTIATION_TYPE_ENHANCED) {
1491 /* Negotiate ord/ird: if upperlayer requested ord larger than
1492 * ird advertised by remote, we need to decrease our ord
1494 if (iparams->ord > ep->cm_info.ird)
1495 iparams->ord = ep->cm_info.ird;
1497 if ((ep->rtr_type & MPA_RTR_TYPE_ZERO_READ) &&
1498 (iparams->ird == 0))
1499 iparams->ird = 1;
1502 /* Update cm_info ord/ird to be negotiated values */
1503 ep->cm_info.ord = iparams->ord;
1504 ep->cm_info.ird = iparams->ird;
1506 qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
1508 ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
1509 ep->cm_info.private_data_len = iparams->private_data_len +
1510 mpa_data_size;
1512 memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
1513 iparams->private_data, iparams->private_data_len);
1515 rc = qed_iwarp_mpa_offload(p_hwfn, ep);
1516 if (rc)
1517 qed_iwarp_modify_qp(p_hwfn,
1518 iparams->qp, QED_IWARP_QP_STATE_ERROR, 1);
1520 return rc;
1523 int qed_iwarp_reject(void *rdma_cxt, struct qed_iwarp_reject_in *iparams)
1525 struct qed_hwfn *p_hwfn = rdma_cxt;
1526 struct qed_iwarp_ep *ep;
1527 u8 mpa_data_size = 0;
1529 ep = iparams->ep_context;
1530 if (!ep) {
1531 DP_ERR(p_hwfn, "Ep Context receive in reject is NULL\n");
1532 return -EINVAL;
1535 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "EP(0x%x)\n", ep->tcp_cid);
1537 ep->cb_context = iparams->cb_context;
1538 ep->qp = NULL;
1540 qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
1542 ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
1543 ep->cm_info.private_data_len = iparams->private_data_len +
1544 mpa_data_size;
1546 memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
1547 iparams->private_data, iparams->private_data_len);
1549 return qed_iwarp_mpa_offload(p_hwfn, ep);
1552 static void
1553 qed_iwarp_print_cm_info(struct qed_hwfn *p_hwfn,
1554 struct qed_iwarp_cm_info *cm_info)
1556 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "ip_version = %d\n",
1557 cm_info->ip_version);
1559 if (cm_info->ip_version == QED_TCP_IPV4)
1560 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1561 "remote_ip %pI4h:%x, local_ip %pI4h:%x vlan=%x\n",
1562 cm_info->remote_ip, cm_info->remote_port,
1563 cm_info->local_ip, cm_info->local_port,
1564 cm_info->vlan);
1565 else
1566 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1567 "remote_ip %pI6:%x, local_ip %pI6:%x vlan=%x\n",
1568 cm_info->remote_ip, cm_info->remote_port,
1569 cm_info->local_ip, cm_info->local_port,
1570 cm_info->vlan);
1572 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1573 "private_data_len = %x ord = %d, ird = %d\n",
1574 cm_info->private_data_len, cm_info->ord, cm_info->ird);
1577 static int
1578 qed_iwarp_ll2_post_rx(struct qed_hwfn *p_hwfn,
1579 struct qed_iwarp_ll2_buff *buf, u8 handle)
1581 int rc;
1583 rc = qed_ll2_post_rx_buffer(p_hwfn, handle, buf->data_phys_addr,
1584 (u16)buf->buff_size, buf, 1);
1585 if (rc) {
1586 DP_NOTICE(p_hwfn,
1587 "Failed to repost rx buffer to ll2 rc = %d, handle=%d\n",
1588 rc, handle);
1589 dma_free_coherent(&p_hwfn->cdev->pdev->dev, buf->buff_size,
1590 buf->data, buf->data_phys_addr);
1591 kfree(buf);
1594 return rc;
1597 static bool
1598 qed_iwarp_ep_exists(struct qed_hwfn *p_hwfn, struct qed_iwarp_cm_info *cm_info)
1600 struct qed_iwarp_ep *ep = NULL;
1601 bool found = false;
1603 list_for_each_entry(ep,
1604 &p_hwfn->p_rdma_info->iwarp.ep_list,
1605 list_entry) {
1606 if ((ep->cm_info.local_port == cm_info->local_port) &&
1607 (ep->cm_info.remote_port == cm_info->remote_port) &&
1608 (ep->cm_info.vlan == cm_info->vlan) &&
1609 !memcmp(&ep->cm_info.local_ip, cm_info->local_ip,
1610 sizeof(cm_info->local_ip)) &&
1611 !memcmp(&ep->cm_info.remote_ip, cm_info->remote_ip,
1612 sizeof(cm_info->remote_ip))) {
1613 found = true;
1614 break;
1618 if (found) {
1619 DP_NOTICE(p_hwfn,
1620 "SYN received on active connection - dropping\n");
1621 qed_iwarp_print_cm_info(p_hwfn, cm_info);
1623 return true;
1626 return false;
1629 static struct qed_iwarp_listener *
1630 qed_iwarp_get_listener(struct qed_hwfn *p_hwfn,
1631 struct qed_iwarp_cm_info *cm_info)
1633 struct qed_iwarp_listener *listener = NULL;
1634 static const u32 ip_zero[4] = { 0, 0, 0, 0 };
1635 bool found = false;
1637 qed_iwarp_print_cm_info(p_hwfn, cm_info);
1639 list_for_each_entry(listener,
1640 &p_hwfn->p_rdma_info->iwarp.listen_list,
1641 list_entry) {
1642 if (listener->port == cm_info->local_port) {
1643 if (!memcmp(listener->ip_addr,
1644 ip_zero, sizeof(ip_zero))) {
1645 found = true;
1646 break;
1649 if (!memcmp(listener->ip_addr,
1650 cm_info->local_ip,
1651 sizeof(cm_info->local_ip)) &&
1652 (listener->vlan == cm_info->vlan)) {
1653 found = true;
1654 break;
1659 if (found) {
1660 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "listener found = %p\n",
1661 listener);
1662 return listener;
1665 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "listener not found\n");
1666 return NULL;
1669 static int
1670 qed_iwarp_parse_rx_pkt(struct qed_hwfn *p_hwfn,
1671 struct qed_iwarp_cm_info *cm_info,
1672 void *buf,
1673 u8 *remote_mac_addr,
1674 u8 *local_mac_addr,
1675 int *payload_len, int *tcp_start_offset)
1677 struct vlan_ethhdr *vethh;
1678 bool vlan_valid = false;
1679 struct ipv6hdr *ip6h;
1680 struct ethhdr *ethh;
1681 struct tcphdr *tcph;
1682 struct iphdr *iph;
1683 int eth_hlen;
1684 int ip_hlen;
1685 int eth_type;
1686 int i;
1688 ethh = buf;
1689 eth_type = ntohs(ethh->h_proto);
1690 if (eth_type == ETH_P_8021Q) {
1691 vlan_valid = true;
1692 vethh = (struct vlan_ethhdr *)ethh;
1693 cm_info->vlan = ntohs(vethh->h_vlan_TCI) & VLAN_VID_MASK;
1694 eth_type = ntohs(vethh->h_vlan_encapsulated_proto);
1697 eth_hlen = ETH_HLEN + (vlan_valid ? sizeof(u32) : 0);
1699 if (!ether_addr_equal(ethh->h_dest,
1700 p_hwfn->p_rdma_info->iwarp.mac_addr)) {
1701 DP_VERBOSE(p_hwfn,
1702 QED_MSG_RDMA,
1703 "Got unexpected mac %pM instead of %pM\n",
1704 ethh->h_dest, p_hwfn->p_rdma_info->iwarp.mac_addr);
1705 return -EINVAL;
1708 ether_addr_copy(remote_mac_addr, ethh->h_source);
1709 ether_addr_copy(local_mac_addr, ethh->h_dest);
1711 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "eth_type =%d source mac: %pM\n",
1712 eth_type, ethh->h_source);
1714 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "eth_hlen=%d destination mac: %pM\n",
1715 eth_hlen, ethh->h_dest);
1717 iph = (struct iphdr *)((u8 *)(ethh) + eth_hlen);
1719 if (eth_type == ETH_P_IP) {
1720 if (iph->protocol != IPPROTO_TCP) {
1721 DP_NOTICE(p_hwfn,
1722 "Unexpected ip protocol on ll2 %x\n",
1723 iph->protocol);
1724 return -EINVAL;
1727 cm_info->local_ip[0] = ntohl(iph->daddr);
1728 cm_info->remote_ip[0] = ntohl(iph->saddr);
1729 cm_info->ip_version = QED_TCP_IPV4;
1731 ip_hlen = (iph->ihl) * sizeof(u32);
1732 *payload_len = ntohs(iph->tot_len) - ip_hlen;
1733 } else if (eth_type == ETH_P_IPV6) {
1734 ip6h = (struct ipv6hdr *)iph;
1736 if (ip6h->nexthdr != IPPROTO_TCP) {
1737 DP_NOTICE(p_hwfn,
1738 "Unexpected ip protocol on ll2 %x\n",
1739 iph->protocol);
1740 return -EINVAL;
1743 for (i = 0; i < 4; i++) {
1744 cm_info->local_ip[i] =
1745 ntohl(ip6h->daddr.in6_u.u6_addr32[i]);
1746 cm_info->remote_ip[i] =
1747 ntohl(ip6h->saddr.in6_u.u6_addr32[i]);
1749 cm_info->ip_version = QED_TCP_IPV6;
1751 ip_hlen = sizeof(*ip6h);
1752 *payload_len = ntohs(ip6h->payload_len);
1753 } else {
1754 DP_NOTICE(p_hwfn, "Unexpected ethertype on ll2 %x\n", eth_type);
1755 return -EINVAL;
1758 tcph = (struct tcphdr *)((u8 *)iph + ip_hlen);
1760 if (!tcph->syn) {
1761 DP_NOTICE(p_hwfn,
1762 "Only SYN type packet expected on this ll2 conn, iph->ihl=%d source=%d dest=%d\n",
1763 iph->ihl, tcph->source, tcph->dest);
1764 return -EINVAL;
1767 cm_info->local_port = ntohs(tcph->dest);
1768 cm_info->remote_port = ntohs(tcph->source);
1770 qed_iwarp_print_cm_info(p_hwfn, cm_info);
1772 *tcp_start_offset = eth_hlen + ip_hlen;
1774 return 0;
1777 static struct qed_iwarp_fpdu *qed_iwarp_get_curr_fpdu(struct qed_hwfn *p_hwfn,
1778 u16 cid)
1780 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
1781 struct qed_iwarp_fpdu *partial_fpdu;
1782 u32 idx;
1784 idx = cid - qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_IWARP);
1785 if (idx >= iwarp_info->max_num_partial_fpdus) {
1786 DP_ERR(p_hwfn, "Invalid cid %x max_num_partial_fpdus=%x\n", cid,
1787 iwarp_info->max_num_partial_fpdus);
1788 return NULL;
1791 partial_fpdu = &iwarp_info->partial_fpdus[idx];
1793 return partial_fpdu;
1796 enum qed_iwarp_mpa_pkt_type {
1797 QED_IWARP_MPA_PKT_PACKED,
1798 QED_IWARP_MPA_PKT_PARTIAL,
1799 QED_IWARP_MPA_PKT_UNALIGNED
1802 #define QED_IWARP_INVALID_FPDU_LENGTH 0xffff
1803 #define QED_IWARP_MPA_FPDU_LENGTH_SIZE (2)
1804 #define QED_IWARP_MPA_CRC32_DIGEST_SIZE (4)
1806 /* Pad to multiple of 4 */
1807 #define QED_IWARP_PDU_DATA_LEN_WITH_PAD(data_len) ALIGN(data_len, 4)
1808 #define QED_IWARP_FPDU_LEN_WITH_PAD(_mpa_len) \
1809 (QED_IWARP_PDU_DATA_LEN_WITH_PAD((_mpa_len) + \
1810 QED_IWARP_MPA_FPDU_LENGTH_SIZE) + \
1811 QED_IWARP_MPA_CRC32_DIGEST_SIZE)
1813 /* fpdu can be fragmented over maximum 3 bds: header, partial mpa, unaligned */
1814 #define QED_IWARP_MAX_BDS_PER_FPDU 3
1816 static const char * const pkt_type_str[] = {
1817 "QED_IWARP_MPA_PKT_PACKED",
1818 "QED_IWARP_MPA_PKT_PARTIAL",
1819 "QED_IWARP_MPA_PKT_UNALIGNED"
1822 static int
1823 qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn,
1824 struct qed_iwarp_fpdu *fpdu,
1825 struct qed_iwarp_ll2_buff *buf);
1827 static enum qed_iwarp_mpa_pkt_type
1828 qed_iwarp_mpa_classify(struct qed_hwfn *p_hwfn,
1829 struct qed_iwarp_fpdu *fpdu,
1830 u16 tcp_payload_len, u8 *mpa_data)
1832 enum qed_iwarp_mpa_pkt_type pkt_type;
1833 u16 mpa_len;
1835 if (fpdu->incomplete_bytes) {
1836 pkt_type = QED_IWARP_MPA_PKT_UNALIGNED;
1837 goto out;
1840 /* special case of one byte remaining...
1841 * lower byte will be read next packet
1843 if (tcp_payload_len == 1) {
1844 fpdu->fpdu_length = *mpa_data << BITS_PER_BYTE;
1845 pkt_type = QED_IWARP_MPA_PKT_PARTIAL;
1846 goto out;
1849 mpa_len = ntohs(*((u16 *)(mpa_data)));
1850 fpdu->fpdu_length = QED_IWARP_FPDU_LEN_WITH_PAD(mpa_len);
1852 if (fpdu->fpdu_length <= tcp_payload_len)
1853 pkt_type = QED_IWARP_MPA_PKT_PACKED;
1854 else
1855 pkt_type = QED_IWARP_MPA_PKT_PARTIAL;
1857 out:
1858 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1859 "MPA_ALIGN: %s: fpdu_length=0x%x tcp_payload_len:0x%x\n",
1860 pkt_type_str[pkt_type], fpdu->fpdu_length, tcp_payload_len);
1862 return pkt_type;
1865 static void
1866 qed_iwarp_init_fpdu(struct qed_iwarp_ll2_buff *buf,
1867 struct qed_iwarp_fpdu *fpdu,
1868 struct unaligned_opaque_data *pkt_data,
1869 u16 tcp_payload_size, u8 placement_offset)
1871 fpdu->mpa_buf = buf;
1872 fpdu->pkt_hdr = buf->data_phys_addr + placement_offset;
1873 fpdu->pkt_hdr_size = pkt_data->tcp_payload_offset;
1874 fpdu->mpa_frag = buf->data_phys_addr + pkt_data->first_mpa_offset;
1875 fpdu->mpa_frag_virt = (u8 *)(buf->data) + pkt_data->first_mpa_offset;
1877 if (tcp_payload_size == 1)
1878 fpdu->incomplete_bytes = QED_IWARP_INVALID_FPDU_LENGTH;
1879 else if (tcp_payload_size < fpdu->fpdu_length)
1880 fpdu->incomplete_bytes = fpdu->fpdu_length - tcp_payload_size;
1881 else
1882 fpdu->incomplete_bytes = 0; /* complete fpdu */
1884 fpdu->mpa_frag_len = fpdu->fpdu_length - fpdu->incomplete_bytes;
1887 static int
1888 qed_iwarp_cp_pkt(struct qed_hwfn *p_hwfn,
1889 struct qed_iwarp_fpdu *fpdu,
1890 struct unaligned_opaque_data *pkt_data,
1891 struct qed_iwarp_ll2_buff *buf, u16 tcp_payload_size)
1893 u8 *tmp_buf = p_hwfn->p_rdma_info->iwarp.mpa_intermediate_buf;
1894 int rc;
1896 /* need to copy the data from the partial packet stored in fpdu
1897 * to the new buf, for this we also need to move the data currently
1898 * placed on the buf. The assumption is that the buffer is big enough
1899 * since fpdu_length <= mss, we use an intermediate buffer since
1900 * we may need to copy the new data to an overlapping location
1902 if ((fpdu->mpa_frag_len + tcp_payload_size) > (u16)buf->buff_size) {
1903 DP_ERR(p_hwfn,
1904 "MPA ALIGN: Unexpected: buffer is not large enough for split fpdu buff_size = %d mpa_frag_len = %d, tcp_payload_size = %d, incomplete_bytes = %d\n",
1905 buf->buff_size, fpdu->mpa_frag_len,
1906 tcp_payload_size, fpdu->incomplete_bytes);
1907 return -EINVAL;
1910 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1911 "MPA ALIGN Copying fpdu: [%p, %d] [%p, %d]\n",
1912 fpdu->mpa_frag_virt, fpdu->mpa_frag_len,
1913 (u8 *)(buf->data) + pkt_data->first_mpa_offset,
1914 tcp_payload_size);
1916 memcpy(tmp_buf, fpdu->mpa_frag_virt, fpdu->mpa_frag_len);
1917 memcpy(tmp_buf + fpdu->mpa_frag_len,
1918 (u8 *)(buf->data) + pkt_data->first_mpa_offset,
1919 tcp_payload_size);
1921 rc = qed_iwarp_recycle_pkt(p_hwfn, fpdu, fpdu->mpa_buf);
1922 if (rc)
1923 return rc;
1925 /* If we managed to post the buffer copy the data to the new buffer
1926 * o/w this will occur in the next round...
1928 memcpy((u8 *)(buf->data), tmp_buf,
1929 fpdu->mpa_frag_len + tcp_payload_size);
1931 fpdu->mpa_buf = buf;
1932 /* fpdu->pkt_hdr remains as is */
1933 /* fpdu->mpa_frag is overridden with new buf */
1934 fpdu->mpa_frag = buf->data_phys_addr;
1935 fpdu->mpa_frag_virt = buf->data;
1936 fpdu->mpa_frag_len += tcp_payload_size;
1938 fpdu->incomplete_bytes -= tcp_payload_size;
1940 DP_VERBOSE(p_hwfn,
1941 QED_MSG_RDMA,
1942 "MPA ALIGN: split fpdu buff_size = %d mpa_frag_len = %d, tcp_payload_size = %d, incomplete_bytes = %d\n",
1943 buf->buff_size, fpdu->mpa_frag_len, tcp_payload_size,
1944 fpdu->incomplete_bytes);
1946 return 0;
1949 static void
1950 qed_iwarp_update_fpdu_length(struct qed_hwfn *p_hwfn,
1951 struct qed_iwarp_fpdu *fpdu, u8 *mpa_data)
1953 u16 mpa_len;
1955 /* Update incomplete packets if needed */
1956 if (fpdu->incomplete_bytes == QED_IWARP_INVALID_FPDU_LENGTH) {
1957 /* Missing lower byte is now available */
1958 mpa_len = fpdu->fpdu_length | *mpa_data;
1959 fpdu->fpdu_length = QED_IWARP_FPDU_LEN_WITH_PAD(mpa_len);
1960 /* one byte of hdr */
1961 fpdu->mpa_frag_len = 1;
1962 fpdu->incomplete_bytes = fpdu->fpdu_length - 1;
1963 DP_VERBOSE(p_hwfn,
1964 QED_MSG_RDMA,
1965 "MPA_ALIGN: Partial header mpa_len=%x fpdu_length=%x incomplete_bytes=%x\n",
1966 mpa_len, fpdu->fpdu_length, fpdu->incomplete_bytes);
1970 #define QED_IWARP_IS_RIGHT_EDGE(_curr_pkt) \
1971 (GET_FIELD((_curr_pkt)->flags, \
1972 UNALIGNED_OPAQUE_DATA_PKT_REACHED_WIN_RIGHT_EDGE))
1974 /* This function is used to recycle a buffer using the ll2 drop option. It
1975 * uses the mechanism to ensure that all buffers posted to tx before this one
1976 * were completed. The buffer sent here will be sent as a cookie in the tx
1977 * completion function and can then be reposted to rx chain when done. The flow
1978 * that requires this is the flow where a FPDU splits over more than 3 tcp
1979 * segments. In this case the driver needs to re-post a rx buffer instead of
1980 * the one received, but driver can't simply repost a buffer it copied from
1981 * as there is a case where the buffer was originally a packed FPDU, and is
1982 * partially posted to FW. Driver needs to ensure FW is done with it.
1984 static int
1985 qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn,
1986 struct qed_iwarp_fpdu *fpdu,
1987 struct qed_iwarp_ll2_buff *buf)
1989 struct qed_ll2_tx_pkt_info tx_pkt;
1990 u8 ll2_handle;
1991 int rc;
1993 memset(&tx_pkt, 0, sizeof(tx_pkt));
1994 tx_pkt.num_of_bds = 1;
1995 tx_pkt.tx_dest = QED_LL2_TX_DEST_DROP;
1996 tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2;
1997 tx_pkt.first_frag = fpdu->pkt_hdr;
1998 tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
1999 buf->piggy_buf = NULL;
2000 tx_pkt.cookie = buf;
2002 ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
2004 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
2005 if (rc)
2006 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2007 "Can't drop packet rc=%d\n", rc);
2009 DP_VERBOSE(p_hwfn,
2010 QED_MSG_RDMA,
2011 "MPA_ALIGN: send drop tx packet [%lx, 0x%x], buf=%p, rc=%d\n",
2012 (unsigned long int)tx_pkt.first_frag,
2013 tx_pkt.first_frag_len, buf, rc);
2015 return rc;
2018 static int
2019 qed_iwarp_win_right_edge(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu)
2021 struct qed_ll2_tx_pkt_info tx_pkt;
2022 u8 ll2_handle;
2023 int rc;
2025 memset(&tx_pkt, 0, sizeof(tx_pkt));
2026 tx_pkt.num_of_bds = 1;
2027 tx_pkt.tx_dest = QED_LL2_TX_DEST_LB;
2028 tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2;
2030 tx_pkt.first_frag = fpdu->pkt_hdr;
2031 tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
2032 tx_pkt.enable_ip_cksum = true;
2033 tx_pkt.enable_l4_cksum = true;
2034 tx_pkt.calc_ip_len = true;
2035 /* vlan overload with enum iwarp_ll2_tx_queues */
2036 tx_pkt.vlan = IWARP_LL2_ALIGNED_RIGHT_TRIMMED_TX_QUEUE;
2038 ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
2040 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
2041 if (rc)
2042 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2043 "Can't send right edge rc=%d\n", rc);
2044 DP_VERBOSE(p_hwfn,
2045 QED_MSG_RDMA,
2046 "MPA_ALIGN: Sent right edge FPDU num_bds=%d [%lx, 0x%x], rc=%d\n",
2047 tx_pkt.num_of_bds,
2048 (unsigned long int)tx_pkt.first_frag,
2049 tx_pkt.first_frag_len, rc);
2051 return rc;
2054 static int
2055 qed_iwarp_send_fpdu(struct qed_hwfn *p_hwfn,
2056 struct qed_iwarp_fpdu *fpdu,
2057 struct unaligned_opaque_data *curr_pkt,
2058 struct qed_iwarp_ll2_buff *buf,
2059 u16 tcp_payload_size, enum qed_iwarp_mpa_pkt_type pkt_type)
2061 struct qed_ll2_tx_pkt_info tx_pkt;
2062 u8 ll2_handle;
2063 int rc;
2065 memset(&tx_pkt, 0, sizeof(tx_pkt));
2067 /* An unaligned packet means it's split over two tcp segments. So the
2068 * complete packet requires 3 bds, one for the header, one for the
2069 * part of the fpdu of the first tcp segment, and the last fragment
2070 * will point to the remainder of the fpdu. A packed pdu, requires only
2071 * two bds, one for the header and one for the data.
2073 tx_pkt.num_of_bds = (pkt_type == QED_IWARP_MPA_PKT_UNALIGNED) ? 3 : 2;
2074 tx_pkt.tx_dest = QED_LL2_TX_DEST_LB;
2075 tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2; /* offset in words */
2077 /* Send the mpa_buf only with the last fpdu (in case of packed) */
2078 if (pkt_type == QED_IWARP_MPA_PKT_UNALIGNED ||
2079 tcp_payload_size <= fpdu->fpdu_length)
2080 tx_pkt.cookie = fpdu->mpa_buf;
2082 tx_pkt.first_frag = fpdu->pkt_hdr;
2083 tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
2084 tx_pkt.enable_ip_cksum = true;
2085 tx_pkt.enable_l4_cksum = true;
2086 tx_pkt.calc_ip_len = true;
2087 /* vlan overload with enum iwarp_ll2_tx_queues */
2088 tx_pkt.vlan = IWARP_LL2_ALIGNED_TX_QUEUE;
2090 /* special case of unaligned packet and not packed, need to send
2091 * both buffers as cookie to release.
2093 if (tcp_payload_size == fpdu->incomplete_bytes)
2094 fpdu->mpa_buf->piggy_buf = buf;
2096 ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
2098 /* Set first fragment to header */
2099 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
2100 if (rc)
2101 goto out;
2103 /* Set second fragment to first part of packet */
2104 rc = qed_ll2_set_fragment_of_tx_packet(p_hwfn, ll2_handle,
2105 fpdu->mpa_frag,
2106 fpdu->mpa_frag_len);
2107 if (rc)
2108 goto out;
2110 if (!fpdu->incomplete_bytes)
2111 goto out;
2113 /* Set third fragment to second part of the packet */
2114 rc = qed_ll2_set_fragment_of_tx_packet(p_hwfn,
2115 ll2_handle,
2116 buf->data_phys_addr +
2117 curr_pkt->first_mpa_offset,
2118 fpdu->incomplete_bytes);
2119 out:
2120 DP_VERBOSE(p_hwfn,
2121 QED_MSG_RDMA,
2122 "MPA_ALIGN: Sent FPDU num_bds=%d first_frag_len=%x, mpa_frag_len=0x%x, incomplete_bytes:0x%x rc=%d\n",
2123 tx_pkt.num_of_bds,
2124 tx_pkt.first_frag_len,
2125 fpdu->mpa_frag_len,
2126 fpdu->incomplete_bytes, rc);
2128 return rc;
2131 static void
2132 qed_iwarp_mpa_get_data(struct qed_hwfn *p_hwfn,
2133 struct unaligned_opaque_data *curr_pkt,
2134 u32 opaque_data0, u32 opaque_data1)
2136 u64 opaque_data;
2138 opaque_data = HILO_64(opaque_data1, opaque_data0);
2139 *curr_pkt = *((struct unaligned_opaque_data *)&opaque_data);
2141 curr_pkt->first_mpa_offset = curr_pkt->tcp_payload_offset +
2142 le16_to_cpu(curr_pkt->first_mpa_offset);
2143 curr_pkt->cid = le32_to_cpu(curr_pkt->cid);
2146 /* This function is called when an unaligned or incomplete MPA packet arrives
2147 * driver needs to align the packet, perhaps using previous data and send
2148 * it down to FW once it is aligned.
2150 static int
2151 qed_iwarp_process_mpa_pkt(struct qed_hwfn *p_hwfn,
2152 struct qed_iwarp_ll2_mpa_buf *mpa_buf)
2154 struct unaligned_opaque_data *curr_pkt = &mpa_buf->data;
2155 struct qed_iwarp_ll2_buff *buf = mpa_buf->ll2_buf;
2156 enum qed_iwarp_mpa_pkt_type pkt_type;
2157 struct qed_iwarp_fpdu *fpdu;
2158 int rc = -EINVAL;
2159 u8 *mpa_data;
2161 fpdu = qed_iwarp_get_curr_fpdu(p_hwfn, curr_pkt->cid & 0xffff);
2162 if (!fpdu) { /* something corrupt with cid, post rx back */
2163 DP_ERR(p_hwfn, "Invalid cid, drop and post back to rx cid=%x\n",
2164 curr_pkt->cid);
2165 goto err;
2168 do {
2169 mpa_data = ((u8 *)(buf->data) + curr_pkt->first_mpa_offset);
2171 pkt_type = qed_iwarp_mpa_classify(p_hwfn, fpdu,
2172 mpa_buf->tcp_payload_len,
2173 mpa_data);
2175 switch (pkt_type) {
2176 case QED_IWARP_MPA_PKT_PARTIAL:
2177 qed_iwarp_init_fpdu(buf, fpdu,
2178 curr_pkt,
2179 mpa_buf->tcp_payload_len,
2180 mpa_buf->placement_offset);
2182 if (!QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) {
2183 mpa_buf->tcp_payload_len = 0;
2184 break;
2187 rc = qed_iwarp_win_right_edge(p_hwfn, fpdu);
2189 if (rc) {
2190 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2191 "Can't send FPDU:reset rc=%d\n", rc);
2192 memset(fpdu, 0, sizeof(*fpdu));
2193 break;
2196 mpa_buf->tcp_payload_len = 0;
2197 break;
2198 case QED_IWARP_MPA_PKT_PACKED:
2199 qed_iwarp_init_fpdu(buf, fpdu,
2200 curr_pkt,
2201 mpa_buf->tcp_payload_len,
2202 mpa_buf->placement_offset);
2204 rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf,
2205 mpa_buf->tcp_payload_len,
2206 pkt_type);
2207 if (rc) {
2208 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2209 "Can't send FPDU:reset rc=%d\n", rc);
2210 memset(fpdu, 0, sizeof(*fpdu));
2211 break;
2214 mpa_buf->tcp_payload_len -= fpdu->fpdu_length;
2215 curr_pkt->first_mpa_offset += fpdu->fpdu_length;
2216 break;
2217 case QED_IWARP_MPA_PKT_UNALIGNED:
2218 qed_iwarp_update_fpdu_length(p_hwfn, fpdu, mpa_data);
2219 if (mpa_buf->tcp_payload_len < fpdu->incomplete_bytes) {
2220 /* special handling of fpdu split over more
2221 * than 2 segments
2223 if (QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) {
2224 rc = qed_iwarp_win_right_edge(p_hwfn,
2225 fpdu);
2226 /* packet will be re-processed later */
2227 if (rc)
2228 return rc;
2231 rc = qed_iwarp_cp_pkt(p_hwfn, fpdu, curr_pkt,
2232 buf,
2233 mpa_buf->tcp_payload_len);
2234 if (rc) /* packet will be re-processed later */
2235 return rc;
2237 mpa_buf->tcp_payload_len = 0;
2238 break;
2241 rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf,
2242 mpa_buf->tcp_payload_len,
2243 pkt_type);
2244 if (rc) {
2245 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2246 "Can't send FPDU:delay rc=%d\n", rc);
2247 /* don't reset fpdu -> we need it for next
2248 * classify
2250 break;
2253 mpa_buf->tcp_payload_len -= fpdu->incomplete_bytes;
2254 curr_pkt->first_mpa_offset += fpdu->incomplete_bytes;
2255 /* The framed PDU was sent - no more incomplete bytes */
2256 fpdu->incomplete_bytes = 0;
2257 break;
2259 } while (mpa_buf->tcp_payload_len && !rc);
2261 return rc;
2263 err:
2264 qed_iwarp_ll2_post_rx(p_hwfn,
2265 buf,
2266 p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle);
2267 return rc;
2270 static void qed_iwarp_process_pending_pkts(struct qed_hwfn *p_hwfn)
2272 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2273 struct qed_iwarp_ll2_mpa_buf *mpa_buf = NULL;
2274 int rc;
2276 while (!list_empty(&iwarp_info->mpa_buf_pending_list)) {
2277 mpa_buf = list_first_entry(&iwarp_info->mpa_buf_pending_list,
2278 struct qed_iwarp_ll2_mpa_buf,
2279 list_entry);
2281 rc = qed_iwarp_process_mpa_pkt(p_hwfn, mpa_buf);
2283 /* busy means break and continue processing later, don't
2284 * remove the buf from the pending list.
2286 if (rc == -EBUSY)
2287 break;
2289 list_move_tail(&mpa_buf->list_entry,
2290 &iwarp_info->mpa_buf_list);
2292 if (rc) { /* different error, don't continue */
2293 DP_NOTICE(p_hwfn, "process pkts failed rc=%d\n", rc);
2294 break;
2299 static void
2300 qed_iwarp_ll2_comp_mpa_pkt(void *cxt, struct qed_ll2_comp_rx_data *data)
2302 struct qed_iwarp_ll2_mpa_buf *mpa_buf;
2303 struct qed_iwarp_info *iwarp_info;
2304 struct qed_hwfn *p_hwfn = cxt;
2306 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2307 mpa_buf = list_first_entry(&iwarp_info->mpa_buf_list,
2308 struct qed_iwarp_ll2_mpa_buf, list_entry);
2309 if (!mpa_buf) {
2310 DP_ERR(p_hwfn, "No free mpa buf\n");
2311 goto err;
2314 list_del(&mpa_buf->list_entry);
2315 qed_iwarp_mpa_get_data(p_hwfn, &mpa_buf->data,
2316 data->opaque_data_0, data->opaque_data_1);
2318 DP_VERBOSE(p_hwfn,
2319 QED_MSG_RDMA,
2320 "LL2 MPA CompRx payload_len:0x%x\tfirst_mpa_offset:0x%x\ttcp_payload_offset:0x%x\tflags:0x%x\tcid:0x%x\n",
2321 data->length.packet_length, mpa_buf->data.first_mpa_offset,
2322 mpa_buf->data.tcp_payload_offset, mpa_buf->data.flags,
2323 mpa_buf->data.cid);
2325 mpa_buf->ll2_buf = data->cookie;
2326 mpa_buf->tcp_payload_len = data->length.packet_length -
2327 mpa_buf->data.first_mpa_offset;
2328 mpa_buf->data.first_mpa_offset += data->u.placement_offset;
2329 mpa_buf->placement_offset = data->u.placement_offset;
2331 list_add_tail(&mpa_buf->list_entry, &iwarp_info->mpa_buf_pending_list);
2333 qed_iwarp_process_pending_pkts(p_hwfn);
2334 return;
2335 err:
2336 qed_iwarp_ll2_post_rx(p_hwfn, data->cookie,
2337 iwarp_info->ll2_mpa_handle);
2340 static void
2341 qed_iwarp_ll2_comp_syn_pkt(void *cxt, struct qed_ll2_comp_rx_data *data)
2343 struct qed_iwarp_ll2_buff *buf = data->cookie;
2344 struct qed_iwarp_listener *listener;
2345 struct qed_ll2_tx_pkt_info tx_pkt;
2346 struct qed_iwarp_cm_info cm_info;
2347 struct qed_hwfn *p_hwfn = cxt;
2348 u8 remote_mac_addr[ETH_ALEN];
2349 u8 local_mac_addr[ETH_ALEN];
2350 struct qed_iwarp_ep *ep;
2351 int tcp_start_offset;
2352 u8 ll2_syn_handle;
2353 int payload_len;
2354 u32 hdr_size;
2355 int rc;
2357 memset(&cm_info, 0, sizeof(cm_info));
2358 ll2_syn_handle = p_hwfn->p_rdma_info->iwarp.ll2_syn_handle;
2360 /* Check if packet was received with errors... */
2361 if (data->err_flags) {
2362 DP_NOTICE(p_hwfn, "Error received on SYN packet: 0x%x\n",
2363 data->err_flags);
2364 goto err;
2367 if (GET_FIELD(data->parse_flags,
2368 PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED) &&
2369 GET_FIELD(data->parse_flags, PARSING_AND_ERR_FLAGS_L4CHKSMERROR)) {
2370 DP_NOTICE(p_hwfn, "Syn packet received with checksum error\n");
2371 goto err;
2374 rc = qed_iwarp_parse_rx_pkt(p_hwfn, &cm_info, (u8 *)(buf->data) +
2375 data->u.placement_offset, remote_mac_addr,
2376 local_mac_addr, &payload_len,
2377 &tcp_start_offset);
2378 if (rc)
2379 goto err;
2381 /* Check if there is a listener for this 4-tuple+vlan */
2382 listener = qed_iwarp_get_listener(p_hwfn, &cm_info);
2383 if (!listener) {
2384 DP_VERBOSE(p_hwfn,
2385 QED_MSG_RDMA,
2386 "SYN received on tuple not listened on parse_flags=%d packet len=%d\n",
2387 data->parse_flags, data->length.packet_length);
2389 memset(&tx_pkt, 0, sizeof(tx_pkt));
2390 tx_pkt.num_of_bds = 1;
2391 tx_pkt.l4_hdr_offset_w = (data->length.packet_length) >> 2;
2392 tx_pkt.tx_dest = QED_LL2_TX_DEST_LB;
2393 tx_pkt.first_frag = buf->data_phys_addr +
2394 data->u.placement_offset;
2395 tx_pkt.first_frag_len = data->length.packet_length;
2396 tx_pkt.cookie = buf;
2398 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_syn_handle,
2399 &tx_pkt, true);
2401 if (rc) {
2402 DP_NOTICE(p_hwfn,
2403 "Can't post SYN back to chip rc=%d\n", rc);
2404 goto err;
2406 return;
2409 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Received syn on listening port\n");
2410 /* There may be an open ep on this connection if this is a syn
2411 * retrasnmit... need to make sure there isn't...
2413 if (qed_iwarp_ep_exists(p_hwfn, &cm_info))
2414 goto err;
2416 ep = qed_iwarp_get_free_ep(p_hwfn);
2417 if (!ep)
2418 goto err;
2420 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2421 list_add_tail(&ep->list_entry, &p_hwfn->p_rdma_info->iwarp.ep_list);
2422 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2424 ether_addr_copy(ep->remote_mac_addr, remote_mac_addr);
2425 ether_addr_copy(ep->local_mac_addr, local_mac_addr);
2427 memcpy(&ep->cm_info, &cm_info, sizeof(ep->cm_info));
2429 hdr_size = ((cm_info.ip_version == QED_TCP_IPV4) ? 40 : 60);
2430 ep->mss = p_hwfn->p_rdma_info->iwarp.max_mtu - hdr_size;
2431 ep->mss = min_t(u16, QED_IWARP_MAX_FW_MSS, ep->mss);
2433 ep->event_cb = listener->event_cb;
2434 ep->cb_context = listener->cb_context;
2435 ep->connect_mode = TCP_CONNECT_PASSIVE;
2437 ep->syn = buf;
2438 ep->syn_ip_payload_length = (u16)payload_len;
2439 ep->syn_phy_addr = buf->data_phys_addr + data->u.placement_offset +
2440 tcp_start_offset;
2442 rc = qed_iwarp_tcp_offload(p_hwfn, ep);
2443 if (rc) {
2444 qed_iwarp_return_ep(p_hwfn, ep);
2445 goto err;
2448 return;
2449 err:
2450 qed_iwarp_ll2_post_rx(p_hwfn, buf, ll2_syn_handle);
2453 static void qed_iwarp_ll2_rel_rx_pkt(void *cxt, u8 connection_handle,
2454 void *cookie, dma_addr_t rx_buf_addr,
2455 bool b_last_packet)
2457 struct qed_iwarp_ll2_buff *buffer = cookie;
2458 struct qed_hwfn *p_hwfn = cxt;
2460 dma_free_coherent(&p_hwfn->cdev->pdev->dev, buffer->buff_size,
2461 buffer->data, buffer->data_phys_addr);
2462 kfree(buffer);
2465 static void qed_iwarp_ll2_comp_tx_pkt(void *cxt, u8 connection_handle,
2466 void *cookie, dma_addr_t first_frag_addr,
2467 bool b_last_fragment, bool b_last_packet)
2469 struct qed_iwarp_ll2_buff *buffer = cookie;
2470 struct qed_iwarp_ll2_buff *piggy;
2471 struct qed_hwfn *p_hwfn = cxt;
2473 if (!buffer) /* can happen in packed mpa unaligned... */
2474 return;
2476 /* this was originally an rx packet, post it back */
2477 piggy = buffer->piggy_buf;
2478 if (piggy) {
2479 buffer->piggy_buf = NULL;
2480 qed_iwarp_ll2_post_rx(p_hwfn, piggy, connection_handle);
2483 qed_iwarp_ll2_post_rx(p_hwfn, buffer, connection_handle);
2485 if (connection_handle == p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle)
2486 qed_iwarp_process_pending_pkts(p_hwfn);
2488 return;
2491 static void qed_iwarp_ll2_rel_tx_pkt(void *cxt, u8 connection_handle,
2492 void *cookie, dma_addr_t first_frag_addr,
2493 bool b_last_fragment, bool b_last_packet)
2495 struct qed_iwarp_ll2_buff *buffer = cookie;
2496 struct qed_hwfn *p_hwfn = cxt;
2498 if (!buffer)
2499 return;
2501 if (buffer->piggy_buf) {
2502 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
2503 buffer->piggy_buf->buff_size,
2504 buffer->piggy_buf->data,
2505 buffer->piggy_buf->data_phys_addr);
2507 kfree(buffer->piggy_buf);
2510 dma_free_coherent(&p_hwfn->cdev->pdev->dev, buffer->buff_size,
2511 buffer->data, buffer->data_phys_addr);
2513 kfree(buffer);
2516 /* The only slowpath for iwarp ll2 is unalign flush. When this completion
2517 * is received, need to reset the FPDU.
2519 static void
2520 qed_iwarp_ll2_slowpath(void *cxt,
2521 u8 connection_handle,
2522 u32 opaque_data_0, u32 opaque_data_1)
2524 struct unaligned_opaque_data unalign_data;
2525 struct qed_hwfn *p_hwfn = cxt;
2526 struct qed_iwarp_fpdu *fpdu;
2528 qed_iwarp_mpa_get_data(p_hwfn, &unalign_data,
2529 opaque_data_0, opaque_data_1);
2531 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "(0x%x) Flush fpdu\n",
2532 unalign_data.cid);
2534 fpdu = qed_iwarp_get_curr_fpdu(p_hwfn, (u16)unalign_data.cid);
2535 if (fpdu)
2536 memset(fpdu, 0, sizeof(*fpdu));
2539 static int qed_iwarp_ll2_stop(struct qed_hwfn *p_hwfn)
2541 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2542 int rc = 0;
2544 if (iwarp_info->ll2_syn_handle != QED_IWARP_HANDLE_INVAL) {
2545 rc = qed_ll2_terminate_connection(p_hwfn,
2546 iwarp_info->ll2_syn_handle);
2547 if (rc)
2548 DP_INFO(p_hwfn, "Failed to terminate syn connection\n");
2550 qed_ll2_release_connection(p_hwfn, iwarp_info->ll2_syn_handle);
2551 iwarp_info->ll2_syn_handle = QED_IWARP_HANDLE_INVAL;
2554 if (iwarp_info->ll2_ooo_handle != QED_IWARP_HANDLE_INVAL) {
2555 rc = qed_ll2_terminate_connection(p_hwfn,
2556 iwarp_info->ll2_ooo_handle);
2557 if (rc)
2558 DP_INFO(p_hwfn, "Failed to terminate ooo connection\n");
2560 qed_ll2_release_connection(p_hwfn, iwarp_info->ll2_ooo_handle);
2561 iwarp_info->ll2_ooo_handle = QED_IWARP_HANDLE_INVAL;
2564 if (iwarp_info->ll2_mpa_handle != QED_IWARP_HANDLE_INVAL) {
2565 rc = qed_ll2_terminate_connection(p_hwfn,
2566 iwarp_info->ll2_mpa_handle);
2567 if (rc)
2568 DP_INFO(p_hwfn, "Failed to terminate mpa connection\n");
2570 qed_ll2_release_connection(p_hwfn, iwarp_info->ll2_mpa_handle);
2571 iwarp_info->ll2_mpa_handle = QED_IWARP_HANDLE_INVAL;
2574 qed_llh_remove_mac_filter(p_hwfn->cdev, 0,
2575 p_hwfn->p_rdma_info->iwarp.mac_addr);
2577 return rc;
2580 static int
2581 qed_iwarp_ll2_alloc_buffers(struct qed_hwfn *p_hwfn,
2582 int num_rx_bufs, int buff_size, u8 ll2_handle)
2584 struct qed_iwarp_ll2_buff *buffer;
2585 int rc = 0;
2586 int i;
2588 for (i = 0; i < num_rx_bufs; i++) {
2589 buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
2590 if (!buffer) {
2591 rc = -ENOMEM;
2592 break;
2595 buffer->data = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2596 buff_size,
2597 &buffer->data_phys_addr,
2598 GFP_KERNEL);
2599 if (!buffer->data) {
2600 kfree(buffer);
2601 rc = -ENOMEM;
2602 break;
2605 buffer->buff_size = buff_size;
2606 rc = qed_iwarp_ll2_post_rx(p_hwfn, buffer, ll2_handle);
2607 if (rc)
2608 /* buffers will be deallocated by qed_ll2 */
2609 break;
2611 return rc;
2614 #define QED_IWARP_MAX_BUF_SIZE(mtu) \
2615 ALIGN((mtu) + ETH_HLEN + 2 * VLAN_HLEN + 2 + ETH_CACHE_LINE_SIZE, \
2616 ETH_CACHE_LINE_SIZE)
2618 static int
2619 qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn,
2620 struct qed_rdma_start_in_params *params,
2621 u32 rcv_wnd_size)
2623 struct qed_iwarp_info *iwarp_info;
2624 struct qed_ll2_acquire_data data;
2625 struct qed_ll2_cbs cbs;
2626 u32 buff_size;
2627 u16 n_ooo_bufs;
2628 int rc = 0;
2629 int i;
2631 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2632 iwarp_info->ll2_syn_handle = QED_IWARP_HANDLE_INVAL;
2633 iwarp_info->ll2_ooo_handle = QED_IWARP_HANDLE_INVAL;
2634 iwarp_info->ll2_mpa_handle = QED_IWARP_HANDLE_INVAL;
2636 iwarp_info->max_mtu = params->max_mtu;
2638 ether_addr_copy(p_hwfn->p_rdma_info->iwarp.mac_addr, params->mac_addr);
2640 rc = qed_llh_add_mac_filter(p_hwfn->cdev, 0, params->mac_addr);
2641 if (rc)
2642 return rc;
2644 /* Start SYN connection */
2645 cbs.rx_comp_cb = qed_iwarp_ll2_comp_syn_pkt;
2646 cbs.rx_release_cb = qed_iwarp_ll2_rel_rx_pkt;
2647 cbs.tx_comp_cb = qed_iwarp_ll2_comp_tx_pkt;
2648 cbs.tx_release_cb = qed_iwarp_ll2_rel_tx_pkt;
2649 cbs.slowpath_cb = NULL;
2650 cbs.cookie = p_hwfn;
2652 memset(&data, 0, sizeof(data));
2653 data.input.conn_type = QED_LL2_TYPE_IWARP;
2654 /* SYN will use ctx based queues */
2655 data.input.rx_conn_type = QED_LL2_RX_TYPE_CTX;
2656 data.input.mtu = params->max_mtu;
2657 data.input.rx_num_desc = QED_IWARP_LL2_SYN_RX_SIZE;
2658 data.input.tx_num_desc = QED_IWARP_LL2_SYN_TX_SIZE;
2659 data.input.tx_max_bds_per_packet = 1; /* will never be fragmented */
2660 data.input.tx_tc = PKT_LB_TC;
2661 data.input.tx_dest = QED_LL2_TX_DEST_LB;
2662 data.p_connection_handle = &iwarp_info->ll2_syn_handle;
2663 data.cbs = &cbs;
2665 rc = qed_ll2_acquire_connection(p_hwfn, &data);
2666 if (rc) {
2667 DP_NOTICE(p_hwfn, "Failed to acquire LL2 connection\n");
2668 qed_llh_remove_mac_filter(p_hwfn->cdev, 0, params->mac_addr);
2669 return rc;
2672 rc = qed_ll2_establish_connection(p_hwfn, iwarp_info->ll2_syn_handle);
2673 if (rc) {
2674 DP_NOTICE(p_hwfn, "Failed to establish LL2 connection\n");
2675 goto err;
2678 buff_size = QED_IWARP_MAX_BUF_SIZE(params->max_mtu);
2679 rc = qed_iwarp_ll2_alloc_buffers(p_hwfn,
2680 QED_IWARP_LL2_SYN_RX_SIZE,
2681 buff_size,
2682 iwarp_info->ll2_syn_handle);
2683 if (rc)
2684 goto err;
2686 /* Start OOO connection */
2687 data.input.conn_type = QED_LL2_TYPE_OOO;
2688 /* OOO/unaligned will use legacy ll2 queues (ram based) */
2689 data.input.rx_conn_type = QED_LL2_RX_TYPE_LEGACY;
2690 data.input.mtu = params->max_mtu;
2692 n_ooo_bufs = (QED_IWARP_MAX_OOO * rcv_wnd_size) /
2693 iwarp_info->max_mtu;
2694 n_ooo_bufs = min_t(u32, n_ooo_bufs, QED_IWARP_LL2_OOO_MAX_RX_SIZE);
2696 data.input.rx_num_desc = n_ooo_bufs;
2697 data.input.rx_num_ooo_buffers = n_ooo_bufs;
2699 data.input.tx_max_bds_per_packet = 1; /* will never be fragmented */
2700 data.input.tx_num_desc = QED_IWARP_LL2_OOO_DEF_TX_SIZE;
2701 data.p_connection_handle = &iwarp_info->ll2_ooo_handle;
2703 rc = qed_ll2_acquire_connection(p_hwfn, &data);
2704 if (rc)
2705 goto err;
2707 rc = qed_ll2_establish_connection(p_hwfn, iwarp_info->ll2_ooo_handle);
2708 if (rc)
2709 goto err;
2711 /* Start Unaligned MPA connection */
2712 cbs.rx_comp_cb = qed_iwarp_ll2_comp_mpa_pkt;
2713 cbs.slowpath_cb = qed_iwarp_ll2_slowpath;
2715 memset(&data, 0, sizeof(data));
2716 data.input.conn_type = QED_LL2_TYPE_IWARP;
2717 data.input.mtu = params->max_mtu;
2718 /* FW requires that once a packet arrives OOO, it must have at
2719 * least 2 rx buffers available on the unaligned connection
2720 * for handling the case that it is a partial fpdu.
2722 data.input.rx_num_desc = n_ooo_bufs * 2;
2723 data.input.tx_num_desc = data.input.rx_num_desc;
2724 data.input.tx_max_bds_per_packet = QED_IWARP_MAX_BDS_PER_FPDU;
2725 data.input.tx_tc = PKT_LB_TC;
2726 data.input.tx_dest = QED_LL2_TX_DEST_LB;
2727 data.p_connection_handle = &iwarp_info->ll2_mpa_handle;
2728 data.input.secondary_queue = true;
2729 data.cbs = &cbs;
2731 rc = qed_ll2_acquire_connection(p_hwfn, &data);
2732 if (rc)
2733 goto err;
2735 rc = qed_ll2_establish_connection(p_hwfn, iwarp_info->ll2_mpa_handle);
2736 if (rc)
2737 goto err;
2739 rc = qed_iwarp_ll2_alloc_buffers(p_hwfn,
2740 data.input.rx_num_desc,
2741 buff_size,
2742 iwarp_info->ll2_mpa_handle);
2743 if (rc)
2744 goto err;
2746 iwarp_info->partial_fpdus = kcalloc((u16)p_hwfn->p_rdma_info->num_qps,
2747 sizeof(*iwarp_info->partial_fpdus),
2748 GFP_KERNEL);
2749 if (!iwarp_info->partial_fpdus)
2750 goto err;
2752 iwarp_info->max_num_partial_fpdus = (u16)p_hwfn->p_rdma_info->num_qps;
2754 iwarp_info->mpa_intermediate_buf = kzalloc(buff_size, GFP_KERNEL);
2755 if (!iwarp_info->mpa_intermediate_buf)
2756 goto err;
2758 /* The mpa_bufs array serves for pending RX packets received on the
2759 * mpa ll2 that don't have place on the tx ring and require later
2760 * processing. We can't fail on allocation of such a struct therefore
2761 * we allocate enough to take care of all rx packets
2763 iwarp_info->mpa_bufs = kcalloc(data.input.rx_num_desc,
2764 sizeof(*iwarp_info->mpa_bufs),
2765 GFP_KERNEL);
2766 if (!iwarp_info->mpa_bufs)
2767 goto err;
2769 INIT_LIST_HEAD(&iwarp_info->mpa_buf_pending_list);
2770 INIT_LIST_HEAD(&iwarp_info->mpa_buf_list);
2771 for (i = 0; i < data.input.rx_num_desc; i++)
2772 list_add_tail(&iwarp_info->mpa_bufs[i].list_entry,
2773 &iwarp_info->mpa_buf_list);
2774 return rc;
2775 err:
2776 qed_iwarp_ll2_stop(p_hwfn);
2778 return rc;
2781 static struct {
2782 u32 two_ports;
2783 u32 four_ports;
2784 } qed_iwarp_rcv_wnd_size[MAX_CHIP_IDS] = {
2785 {QED_IWARP_RCV_WND_SIZE_DEF_BB_2P, QED_IWARP_RCV_WND_SIZE_DEF_BB_4P},
2786 {QED_IWARP_RCV_WND_SIZE_DEF_AH_2P, QED_IWARP_RCV_WND_SIZE_DEF_AH_4P}
2789 int qed_iwarp_setup(struct qed_hwfn *p_hwfn,
2790 struct qed_rdma_start_in_params *params)
2792 struct qed_dev *cdev = p_hwfn->cdev;
2793 struct qed_iwarp_info *iwarp_info;
2794 enum chip_ids chip_id;
2795 u32 rcv_wnd_size;
2797 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2799 iwarp_info->tcp_flags = QED_IWARP_TS_EN;
2801 chip_id = QED_IS_BB(cdev) ? CHIP_BB : CHIP_K2;
2802 rcv_wnd_size = (qed_device_num_ports(cdev) == 4) ?
2803 qed_iwarp_rcv_wnd_size[chip_id].four_ports :
2804 qed_iwarp_rcv_wnd_size[chip_id].two_ports;
2806 /* value 0 is used for ilog2(QED_IWARP_RCV_WND_SIZE_MIN) */
2807 iwarp_info->rcv_wnd_scale = ilog2(rcv_wnd_size) -
2808 ilog2(QED_IWARP_RCV_WND_SIZE_MIN);
2809 iwarp_info->rcv_wnd_size = rcv_wnd_size >> iwarp_info->rcv_wnd_scale;
2810 iwarp_info->crc_needed = QED_IWARP_PARAM_CRC_NEEDED;
2811 iwarp_info->mpa_rev = MPA_NEGOTIATION_TYPE_ENHANCED;
2813 iwarp_info->peer2peer = QED_IWARP_PARAM_P2P;
2815 iwarp_info->rtr_type = MPA_RTR_TYPE_ZERO_SEND |
2816 MPA_RTR_TYPE_ZERO_WRITE |
2817 MPA_RTR_TYPE_ZERO_READ;
2819 spin_lock_init(&p_hwfn->p_rdma_info->iwarp.qp_lock);
2820 INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.ep_list);
2821 INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.listen_list);
2823 qed_spq_register_async_cb(p_hwfn, PROTOCOLID_IWARP,
2824 qed_iwarp_async_event);
2825 qed_ooo_setup(p_hwfn);
2827 return qed_iwarp_ll2_start(p_hwfn, params, rcv_wnd_size);
2830 int qed_iwarp_stop(struct qed_hwfn *p_hwfn)
2832 int rc;
2834 qed_iwarp_free_prealloc_ep(p_hwfn);
2835 rc = qed_iwarp_wait_for_all_cids(p_hwfn);
2836 if (rc)
2837 return rc;
2839 qed_spq_unregister_async_cb(p_hwfn, PROTOCOLID_IWARP);
2841 return qed_iwarp_ll2_stop(p_hwfn);
2844 static void qed_iwarp_qp_in_error(struct qed_hwfn *p_hwfn,
2845 struct qed_iwarp_ep *ep,
2846 u8 fw_return_code)
2848 struct qed_iwarp_cm_event_params params;
2850 qed_iwarp_modify_qp(p_hwfn, ep->qp, QED_IWARP_QP_STATE_ERROR, true);
2852 params.event = QED_IWARP_EVENT_CLOSE;
2853 params.ep_context = ep;
2854 params.cm_info = &ep->cm_info;
2855 params.status = (fw_return_code == IWARP_QP_IN_ERROR_GOOD_CLOSE) ?
2856 0 : -ECONNRESET;
2858 /* paired with READ_ONCE in destroy_qp */
2859 smp_store_release(&ep->state, QED_IWARP_EP_CLOSED);
2861 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2862 list_del(&ep->list_entry);
2863 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2865 ep->event_cb(ep->cb_context, &params);
2868 static void qed_iwarp_exception_received(struct qed_hwfn *p_hwfn,
2869 struct qed_iwarp_ep *ep,
2870 int fw_ret_code)
2872 struct qed_iwarp_cm_event_params params;
2873 bool event_cb = false;
2875 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "EP(0x%x) fw_ret_code=%d\n",
2876 ep->cid, fw_ret_code);
2878 switch (fw_ret_code) {
2879 case IWARP_EXCEPTION_DETECTED_LLP_CLOSED:
2880 params.status = 0;
2881 params.event = QED_IWARP_EVENT_DISCONNECT;
2882 event_cb = true;
2883 break;
2884 case IWARP_EXCEPTION_DETECTED_LLP_RESET:
2885 params.status = -ECONNRESET;
2886 params.event = QED_IWARP_EVENT_DISCONNECT;
2887 event_cb = true;
2888 break;
2889 case IWARP_EXCEPTION_DETECTED_RQ_EMPTY:
2890 params.event = QED_IWARP_EVENT_RQ_EMPTY;
2891 event_cb = true;
2892 break;
2893 case IWARP_EXCEPTION_DETECTED_IRQ_FULL:
2894 params.event = QED_IWARP_EVENT_IRQ_FULL;
2895 event_cb = true;
2896 break;
2897 case IWARP_EXCEPTION_DETECTED_LLP_TIMEOUT:
2898 params.event = QED_IWARP_EVENT_LLP_TIMEOUT;
2899 event_cb = true;
2900 break;
2901 case IWARP_EXCEPTION_DETECTED_REMOTE_PROTECTION_ERROR:
2902 params.event = QED_IWARP_EVENT_REMOTE_PROTECTION_ERROR;
2903 event_cb = true;
2904 break;
2905 case IWARP_EXCEPTION_DETECTED_CQ_OVERFLOW:
2906 params.event = QED_IWARP_EVENT_CQ_OVERFLOW;
2907 event_cb = true;
2908 break;
2909 case IWARP_EXCEPTION_DETECTED_LOCAL_CATASTROPHIC:
2910 params.event = QED_IWARP_EVENT_QP_CATASTROPHIC;
2911 event_cb = true;
2912 break;
2913 case IWARP_EXCEPTION_DETECTED_LOCAL_ACCESS_ERROR:
2914 params.event = QED_IWARP_EVENT_LOCAL_ACCESS_ERROR;
2915 event_cb = true;
2916 break;
2917 case IWARP_EXCEPTION_DETECTED_REMOTE_OPERATION_ERROR:
2918 params.event = QED_IWARP_EVENT_REMOTE_OPERATION_ERROR;
2919 event_cb = true;
2920 break;
2921 case IWARP_EXCEPTION_DETECTED_TERMINATE_RECEIVED:
2922 params.event = QED_IWARP_EVENT_TERMINATE_RECEIVED;
2923 event_cb = true;
2924 break;
2925 default:
2926 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2927 "Unhandled exception received...fw_ret_code=%d\n",
2928 fw_ret_code);
2929 break;
2932 if (event_cb) {
2933 params.ep_context = ep;
2934 params.cm_info = &ep->cm_info;
2935 ep->event_cb(ep->cb_context, &params);
2939 static void
2940 qed_iwarp_tcp_connect_unsuccessful(struct qed_hwfn *p_hwfn,
2941 struct qed_iwarp_ep *ep, u8 fw_return_code)
2943 struct qed_iwarp_cm_event_params params;
2945 memset(&params, 0, sizeof(params));
2946 params.event = QED_IWARP_EVENT_ACTIVE_COMPLETE;
2947 params.ep_context = ep;
2948 params.cm_info = &ep->cm_info;
2949 /* paired with READ_ONCE in destroy_qp */
2950 smp_store_release(&ep->state, QED_IWARP_EP_CLOSED);
2952 switch (fw_return_code) {
2953 case IWARP_CONN_ERROR_TCP_CONNECT_INVALID_PACKET:
2954 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2955 "%s(0x%x) TCP connect got invalid packet\n",
2956 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2957 params.status = -ECONNRESET;
2958 break;
2959 case IWARP_CONN_ERROR_TCP_CONNECTION_RST:
2960 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2961 "%s(0x%x) TCP Connection Reset\n",
2962 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2963 params.status = -ECONNRESET;
2964 break;
2965 case IWARP_CONN_ERROR_TCP_CONNECT_TIMEOUT:
2966 DP_NOTICE(p_hwfn, "%s(0x%x) TCP timeout\n",
2967 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2968 params.status = -EBUSY;
2969 break;
2970 case IWARP_CONN_ERROR_MPA_NOT_SUPPORTED_VER:
2971 DP_NOTICE(p_hwfn, "%s(0x%x) MPA not supported VER\n",
2972 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2973 params.status = -ECONNREFUSED;
2974 break;
2975 case IWARP_CONN_ERROR_MPA_INVALID_PACKET:
2976 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Invalid Packet\n",
2977 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2978 params.status = -ECONNRESET;
2979 break;
2980 default:
2981 DP_ERR(p_hwfn,
2982 "%s(0x%x) Unexpected return code tcp connect: %d\n",
2983 QED_IWARP_CONNECT_MODE_STRING(ep),
2984 ep->tcp_cid, fw_return_code);
2985 params.status = -ECONNRESET;
2986 break;
2989 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
2990 ep->tcp_cid = QED_IWARP_INVALID_TCP_CID;
2991 qed_iwarp_return_ep(p_hwfn, ep);
2992 } else {
2993 ep->event_cb(ep->cb_context, &params);
2994 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2995 list_del(&ep->list_entry);
2996 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3000 static void
3001 qed_iwarp_connect_complete(struct qed_hwfn *p_hwfn,
3002 struct qed_iwarp_ep *ep, u8 fw_return_code)
3004 u8 ll2_syn_handle = p_hwfn->p_rdma_info->iwarp.ll2_syn_handle;
3006 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
3007 /* Done with the SYN packet, post back to ll2 rx */
3008 qed_iwarp_ll2_post_rx(p_hwfn, ep->syn, ll2_syn_handle);
3010 ep->syn = NULL;
3012 /* If connect failed - upper layer doesn't know about it */
3013 if (fw_return_code == RDMA_RETURN_OK)
3014 qed_iwarp_mpa_received(p_hwfn, ep);
3015 else
3016 qed_iwarp_tcp_connect_unsuccessful(p_hwfn, ep,
3017 fw_return_code);
3018 } else {
3019 if (fw_return_code == RDMA_RETURN_OK)
3020 qed_iwarp_mpa_offload(p_hwfn, ep);
3021 else
3022 qed_iwarp_tcp_connect_unsuccessful(p_hwfn, ep,
3023 fw_return_code);
3027 static inline bool
3028 qed_iwarp_check_ep_ok(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
3030 if (!ep || (ep->sig != QED_EP_SIG)) {
3031 DP_ERR(p_hwfn, "ERROR ON ASYNC ep=%p\n", ep);
3032 return false;
3035 return true;
3038 static int qed_iwarp_async_event(struct qed_hwfn *p_hwfn,
3039 u8 fw_event_code, u16 echo,
3040 union event_ring_data *data,
3041 u8 fw_return_code)
3043 struct qed_rdma_events events = p_hwfn->p_rdma_info->events;
3044 struct regpair *fw_handle = &data->rdma_data.async_handle;
3045 struct qed_iwarp_ep *ep = NULL;
3046 u16 srq_offset;
3047 u16 srq_id;
3048 u16 cid;
3050 ep = (struct qed_iwarp_ep *)(uintptr_t)HILO_64(fw_handle->hi,
3051 fw_handle->lo);
3053 switch (fw_event_code) {
3054 case IWARP_EVENT_TYPE_ASYNC_CONNECT_COMPLETE:
3055 /* Async completion after TCP 3-way handshake */
3056 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3057 return -EINVAL;
3058 DP_VERBOSE(p_hwfn,
3059 QED_MSG_RDMA,
3060 "EP(0x%x) IWARP_EVENT_TYPE_ASYNC_CONNECT_COMPLETE fw_ret_code=%d\n",
3061 ep->tcp_cid, fw_return_code);
3062 qed_iwarp_connect_complete(p_hwfn, ep, fw_return_code);
3063 break;
3064 case IWARP_EVENT_TYPE_ASYNC_EXCEPTION_DETECTED:
3065 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3066 return -EINVAL;
3067 DP_VERBOSE(p_hwfn,
3068 QED_MSG_RDMA,
3069 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_EXCEPTION_DETECTED fw_ret_code=%d\n",
3070 ep->cid, fw_return_code);
3071 qed_iwarp_exception_received(p_hwfn, ep, fw_return_code);
3072 break;
3073 case IWARP_EVENT_TYPE_ASYNC_QP_IN_ERROR_STATE:
3074 /* Async completion for Close Connection ramrod */
3075 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3076 return -EINVAL;
3077 DP_VERBOSE(p_hwfn,
3078 QED_MSG_RDMA,
3079 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_QP_IN_ERROR_STATE fw_ret_code=%d\n",
3080 ep->cid, fw_return_code);
3081 qed_iwarp_qp_in_error(p_hwfn, ep, fw_return_code);
3082 break;
3083 case IWARP_EVENT_TYPE_ASYNC_ENHANCED_MPA_REPLY_ARRIVED:
3084 /* Async event for active side only */
3085 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3086 return -EINVAL;
3087 DP_VERBOSE(p_hwfn,
3088 QED_MSG_RDMA,
3089 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_MPA_REPLY_ARRIVED fw_ret_code=%d\n",
3090 ep->cid, fw_return_code);
3091 qed_iwarp_mpa_reply_arrived(p_hwfn, ep);
3092 break;
3093 case IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE:
3094 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3095 return -EINVAL;
3096 DP_VERBOSE(p_hwfn,
3097 QED_MSG_RDMA,
3098 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE fw_ret_code=%d\n",
3099 ep->cid, fw_return_code);
3100 qed_iwarp_mpa_complete(p_hwfn, ep, fw_return_code);
3101 break;
3102 case IWARP_EVENT_TYPE_ASYNC_CID_CLEANED:
3103 cid = (u16)le32_to_cpu(fw_handle->lo);
3104 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
3105 "(0x%x)IWARP_EVENT_TYPE_ASYNC_CID_CLEANED\n", cid);
3106 qed_iwarp_cid_cleaned(p_hwfn, cid);
3108 break;
3109 case IWARP_EVENT_TYPE_ASYNC_SRQ_EMPTY:
3110 DP_NOTICE(p_hwfn, "IWARP_EVENT_TYPE_ASYNC_SRQ_EMPTY\n");
3111 srq_offset = p_hwfn->p_rdma_info->srq_id_offset;
3112 /* FW assigns value that is no greater than u16 */
3113 srq_id = ((u16)le32_to_cpu(fw_handle->lo)) - srq_offset;
3114 events.affiliated_event(events.context,
3115 QED_IWARP_EVENT_SRQ_EMPTY,
3116 &srq_id);
3117 break;
3118 case IWARP_EVENT_TYPE_ASYNC_SRQ_LIMIT:
3119 DP_NOTICE(p_hwfn, "IWARP_EVENT_TYPE_ASYNC_SRQ_LIMIT\n");
3120 srq_offset = p_hwfn->p_rdma_info->srq_id_offset;
3121 /* FW assigns value that is no greater than u16 */
3122 srq_id = ((u16)le32_to_cpu(fw_handle->lo)) - srq_offset;
3123 events.affiliated_event(events.context,
3124 QED_IWARP_EVENT_SRQ_LIMIT,
3125 &srq_id);
3126 break;
3127 case IWARP_EVENT_TYPE_ASYNC_CQ_OVERFLOW:
3128 DP_NOTICE(p_hwfn, "IWARP_EVENT_TYPE_ASYNC_CQ_OVERFLOW\n");
3130 p_hwfn->p_rdma_info->events.affiliated_event(
3131 p_hwfn->p_rdma_info->events.context,
3132 QED_IWARP_EVENT_CQ_OVERFLOW,
3133 (void *)fw_handle);
3134 break;
3135 default:
3136 DP_ERR(p_hwfn, "Received unexpected async iwarp event %d\n",
3137 fw_event_code);
3138 return -EINVAL;
3140 return 0;
3144 qed_iwarp_create_listen(void *rdma_cxt,
3145 struct qed_iwarp_listen_in *iparams,
3146 struct qed_iwarp_listen_out *oparams)
3148 struct qed_hwfn *p_hwfn = rdma_cxt;
3149 struct qed_iwarp_listener *listener;
3151 listener = kzalloc(sizeof(*listener), GFP_KERNEL);
3152 if (!listener)
3153 return -ENOMEM;
3155 listener->ip_version = iparams->ip_version;
3156 memcpy(listener->ip_addr, iparams->ip_addr, sizeof(listener->ip_addr));
3157 listener->port = iparams->port;
3158 listener->vlan = iparams->vlan;
3160 listener->event_cb = iparams->event_cb;
3161 listener->cb_context = iparams->cb_context;
3162 listener->max_backlog = iparams->max_backlog;
3163 oparams->handle = listener;
3165 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3166 list_add_tail(&listener->list_entry,
3167 &p_hwfn->p_rdma_info->iwarp.listen_list);
3168 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3170 DP_VERBOSE(p_hwfn,
3171 QED_MSG_RDMA,
3172 "callback=%p handle=%p ip=%x:%x:%x:%x port=0x%x vlan=0x%x\n",
3173 listener->event_cb,
3174 listener,
3175 listener->ip_addr[0],
3176 listener->ip_addr[1],
3177 listener->ip_addr[2],
3178 listener->ip_addr[3], listener->port, listener->vlan);
3180 return 0;
3183 int qed_iwarp_destroy_listen(void *rdma_cxt, void *handle)
3185 struct qed_iwarp_listener *listener = handle;
3186 struct qed_hwfn *p_hwfn = rdma_cxt;
3188 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "handle=%p\n", handle);
3190 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3191 list_del(&listener->list_entry);
3192 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3194 kfree(listener);
3196 return 0;
3199 int qed_iwarp_send_rtr(void *rdma_cxt, struct qed_iwarp_send_rtr_in *iparams)
3201 struct qed_hwfn *p_hwfn = rdma_cxt;
3202 struct qed_sp_init_data init_data;
3203 struct qed_spq_entry *p_ent;
3204 struct qed_iwarp_ep *ep;
3205 struct qed_rdma_qp *qp;
3206 int rc;
3208 ep = iparams->ep_context;
3209 if (!ep) {
3210 DP_ERR(p_hwfn, "Ep Context receive in send_rtr is NULL\n");
3211 return -EINVAL;
3214 qp = ep->qp;
3216 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) EP(0x%x)\n",
3217 qp->icid, ep->tcp_cid);
3219 memset(&init_data, 0, sizeof(init_data));
3220 init_data.cid = qp->icid;
3221 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
3222 init_data.comp_mode = QED_SPQ_MODE_CB;
3224 rc = qed_sp_init_request(p_hwfn, &p_ent,
3225 IWARP_RAMROD_CMD_ID_MPA_OFFLOAD_SEND_RTR,
3226 PROTOCOLID_IWARP, &init_data);
3228 if (rc)
3229 return rc;
3231 rc = qed_spq_post(p_hwfn, p_ent, NULL);
3233 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = 0x%x\n", rc);
3235 return rc;
3238 void
3239 qed_iwarp_query_qp(struct qed_rdma_qp *qp,
3240 struct qed_rdma_query_qp_out_params *out_params)
3242 out_params->state = qed_iwarp2roce_state(qp->iwarp_state);