Linux 4.19.133
[linux/fpc-iii.git] / drivers / net / ethernet / qlogic / qed / qed_iwarp.c
blob39787bb885c86a390c00d46add23c5a5efdd97a0
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
66 #define QED_IWARP_RCV_WND_SIZE_DEF (256 * 1024)
67 #define QED_IWARP_RCV_WND_SIZE_MIN (0xffff)
68 #define TIMESTAMP_HEADER_SIZE (12)
69 #define QED_IWARP_MAX_FIN_RT_DEFAULT (2)
71 #define QED_IWARP_TS_EN BIT(0)
72 #define QED_IWARP_DA_EN BIT(1)
73 #define QED_IWARP_PARAM_CRC_NEEDED (1)
74 #define QED_IWARP_PARAM_P2P (1)
76 #define QED_IWARP_DEF_MAX_RT_TIME (0)
77 #define QED_IWARP_DEF_CWND_FACTOR (4)
78 #define QED_IWARP_DEF_KA_MAX_PROBE_CNT (5)
79 #define QED_IWARP_DEF_KA_TIMEOUT (1200000) /* 20 min */
80 #define QED_IWARP_DEF_KA_INTERVAL (1000) /* 1 sec */
82 static int qed_iwarp_async_event(struct qed_hwfn *p_hwfn,
83 u8 fw_event_code, u16 echo,
84 union event_ring_data *data,
85 u8 fw_return_code);
87 /* Override devinfo with iWARP specific values */
88 void qed_iwarp_init_devinfo(struct qed_hwfn *p_hwfn)
90 struct qed_rdma_device *dev = p_hwfn->p_rdma_info->dev;
92 dev->max_inline = IWARP_REQ_MAX_INLINE_DATA_SIZE;
93 dev->max_qp = min_t(u32,
94 IWARP_MAX_QPS,
95 p_hwfn->p_rdma_info->num_qps) -
96 QED_IWARP_PREALLOC_CNT;
98 dev->max_cq = dev->max_qp;
100 dev->max_qp_resp_rd_atomic_resc = QED_IWARP_IRD_DEFAULT;
101 dev->max_qp_req_rd_atomic_resc = QED_IWARP_ORD_DEFAULT;
104 void qed_iwarp_init_hw(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
106 p_hwfn->rdma_prs_search_reg = PRS_REG_SEARCH_TCP;
107 qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
108 p_hwfn->b_rdma_enabled_in_prs = true;
111 /* We have two cid maps, one for tcp which should be used only from passive
112 * syn processing and replacing a pre-allocated ep in the list. The second
113 * for active tcp and for QPs.
115 static void qed_iwarp_cid_cleaned(struct qed_hwfn *p_hwfn, u32 cid)
117 cid -= qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
119 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
121 if (cid < QED_IWARP_PREALLOC_CNT)
122 qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map,
123 cid);
124 else
125 qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, cid);
127 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
130 void
131 qed_iwarp_init_fw_ramrod(struct qed_hwfn *p_hwfn,
132 struct iwarp_init_func_ramrod_data *p_ramrod)
134 p_ramrod->iwarp.ll2_ooo_q_index =
135 RESC_START(p_hwfn, QED_LL2_QUEUE) +
136 p_hwfn->p_rdma_info->iwarp.ll2_ooo_handle;
138 p_ramrod->tcp.max_fin_rt = QED_IWARP_MAX_FIN_RT_DEFAULT;
140 return;
143 static int qed_iwarp_alloc_cid(struct qed_hwfn *p_hwfn, u32 *cid)
145 int rc;
147 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
148 rc = qed_rdma_bmap_alloc_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, cid);
149 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
150 if (rc) {
151 DP_NOTICE(p_hwfn, "Failed in allocating iwarp cid\n");
152 return rc;
154 *cid += qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
156 rc = qed_cxt_dynamic_ilt_alloc(p_hwfn, QED_ELEM_CXT, *cid);
157 if (rc)
158 qed_iwarp_cid_cleaned(p_hwfn, *cid);
160 return rc;
163 static void qed_iwarp_set_tcp_cid(struct qed_hwfn *p_hwfn, u32 cid)
165 cid -= qed_cxt_get_proto_cid_start(p_hwfn, p_hwfn->p_rdma_info->proto);
167 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
168 qed_bmap_set_id(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, cid);
169 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
172 /* This function allocates a cid for passive tcp (called from syn receive)
173 * the reason it's separate from the regular cid allocation is because it
174 * is assured that these cids already have ilt allocated. They are preallocated
175 * to ensure that we won't need to allocate memory during syn processing
177 static int qed_iwarp_alloc_tcp_cid(struct qed_hwfn *p_hwfn, u32 *cid)
179 int rc;
181 spin_lock_bh(&p_hwfn->p_rdma_info->lock);
183 rc = qed_rdma_bmap_alloc_id(p_hwfn,
184 &p_hwfn->p_rdma_info->tcp_cid_map, cid);
186 spin_unlock_bh(&p_hwfn->p_rdma_info->lock);
188 if (rc) {
189 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
190 "can't allocate iwarp tcp cid max-count=%d\n",
191 p_hwfn->p_rdma_info->tcp_cid_map.max_count);
193 *cid = QED_IWARP_INVALID_TCP_CID;
194 return rc;
197 *cid += qed_cxt_get_proto_cid_start(p_hwfn,
198 p_hwfn->p_rdma_info->proto);
199 return 0;
202 int qed_iwarp_create_qp(struct qed_hwfn *p_hwfn,
203 struct qed_rdma_qp *qp,
204 struct qed_rdma_create_qp_out_params *out_params)
206 struct iwarp_create_qp_ramrod_data *p_ramrod;
207 struct qed_sp_init_data init_data;
208 struct qed_spq_entry *p_ent;
209 u16 physical_queue;
210 u32 cid;
211 int rc;
213 qp->shared_queue = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
214 IWARP_SHARED_QUEUE_PAGE_SIZE,
215 &qp->shared_queue_phys_addr,
216 GFP_KERNEL);
217 if (!qp->shared_queue)
218 return -ENOMEM;
220 out_params->sq_pbl_virt = (u8 *)qp->shared_queue +
221 IWARP_SHARED_QUEUE_PAGE_SQ_PBL_OFFSET;
222 out_params->sq_pbl_phys = qp->shared_queue_phys_addr +
223 IWARP_SHARED_QUEUE_PAGE_SQ_PBL_OFFSET;
224 out_params->rq_pbl_virt = (u8 *)qp->shared_queue +
225 IWARP_SHARED_QUEUE_PAGE_RQ_PBL_OFFSET;
226 out_params->rq_pbl_phys = qp->shared_queue_phys_addr +
227 IWARP_SHARED_QUEUE_PAGE_RQ_PBL_OFFSET;
229 rc = qed_iwarp_alloc_cid(p_hwfn, &cid);
230 if (rc)
231 goto err1;
233 qp->icid = (u16)cid;
235 memset(&init_data, 0, sizeof(init_data));
236 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
237 init_data.cid = qp->icid;
238 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
240 rc = qed_sp_init_request(p_hwfn, &p_ent,
241 IWARP_RAMROD_CMD_ID_CREATE_QP,
242 PROTOCOLID_IWARP, &init_data);
243 if (rc)
244 goto err2;
246 p_ramrod = &p_ent->ramrod.iwarp_create_qp;
248 SET_FIELD(p_ramrod->flags,
249 IWARP_CREATE_QP_RAMROD_DATA_FMR_AND_RESERVED_EN,
250 qp->fmr_and_reserved_lkey);
252 SET_FIELD(p_ramrod->flags,
253 IWARP_CREATE_QP_RAMROD_DATA_SIGNALED_COMP, qp->signal_all);
255 SET_FIELD(p_ramrod->flags,
256 IWARP_CREATE_QP_RAMROD_DATA_RDMA_RD_EN,
257 qp->incoming_rdma_read_en);
259 SET_FIELD(p_ramrod->flags,
260 IWARP_CREATE_QP_RAMROD_DATA_RDMA_WR_EN,
261 qp->incoming_rdma_write_en);
263 SET_FIELD(p_ramrod->flags,
264 IWARP_CREATE_QP_RAMROD_DATA_ATOMIC_EN,
265 qp->incoming_atomic_en);
267 SET_FIELD(p_ramrod->flags,
268 IWARP_CREATE_QP_RAMROD_DATA_SRQ_FLG, qp->use_srq);
270 p_ramrod->pd = qp->pd;
271 p_ramrod->sq_num_pages = qp->sq_num_pages;
272 p_ramrod->rq_num_pages = qp->rq_num_pages;
274 p_ramrod->srq_id.srq_idx = cpu_to_le16(qp->srq_id);
275 p_ramrod->srq_id.opaque_fid = cpu_to_le16(p_hwfn->hw_info.opaque_fid);
276 p_ramrod->qp_handle_for_cqe.hi = cpu_to_le32(qp->qp_handle.hi);
277 p_ramrod->qp_handle_for_cqe.lo = cpu_to_le32(qp->qp_handle.lo);
279 p_ramrod->cq_cid_for_sq =
280 cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) | qp->sq_cq_id);
281 p_ramrod->cq_cid_for_rq =
282 cpu_to_le32((p_hwfn->hw_info.opaque_fid << 16) | qp->rq_cq_id);
284 p_ramrod->dpi = cpu_to_le16(qp->dpi);
286 physical_queue = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_OFLD);
287 p_ramrod->physical_q0 = cpu_to_le16(physical_queue);
288 physical_queue = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_ACK);
289 p_ramrod->physical_q1 = cpu_to_le16(physical_queue);
291 rc = qed_spq_post(p_hwfn, p_ent, NULL);
292 if (rc)
293 goto err2;
295 return rc;
297 err2:
298 qed_iwarp_cid_cleaned(p_hwfn, cid);
299 err1:
300 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
301 IWARP_SHARED_QUEUE_PAGE_SIZE,
302 qp->shared_queue, qp->shared_queue_phys_addr);
304 return rc;
307 static int qed_iwarp_modify_fw(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
309 struct iwarp_modify_qp_ramrod_data *p_ramrod;
310 struct qed_sp_init_data init_data;
311 struct qed_spq_entry *p_ent;
312 int rc;
314 /* Get SPQ entry */
315 memset(&init_data, 0, sizeof(init_data));
316 init_data.cid = qp->icid;
317 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
318 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
320 rc = qed_sp_init_request(p_hwfn, &p_ent,
321 IWARP_RAMROD_CMD_ID_MODIFY_QP,
322 p_hwfn->p_rdma_info->proto, &init_data);
323 if (rc)
324 return rc;
326 p_ramrod = &p_ent->ramrod.iwarp_modify_qp;
327 SET_FIELD(p_ramrod->flags, IWARP_MODIFY_QP_RAMROD_DATA_STATE_TRANS_EN,
328 0x1);
329 if (qp->iwarp_state == QED_IWARP_QP_STATE_CLOSING)
330 p_ramrod->transition_to_state = IWARP_MODIFY_QP_STATE_CLOSING;
331 else
332 p_ramrod->transition_to_state = IWARP_MODIFY_QP_STATE_ERROR;
334 rc = qed_spq_post(p_hwfn, p_ent, NULL);
336 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x)rc=%d\n", qp->icid, rc);
338 return rc;
341 enum qed_iwarp_qp_state qed_roce2iwarp_state(enum qed_roce_qp_state state)
343 switch (state) {
344 case QED_ROCE_QP_STATE_RESET:
345 case QED_ROCE_QP_STATE_INIT:
346 case QED_ROCE_QP_STATE_RTR:
347 return QED_IWARP_QP_STATE_IDLE;
348 case QED_ROCE_QP_STATE_RTS:
349 return QED_IWARP_QP_STATE_RTS;
350 case QED_ROCE_QP_STATE_SQD:
351 return QED_IWARP_QP_STATE_CLOSING;
352 case QED_ROCE_QP_STATE_ERR:
353 return QED_IWARP_QP_STATE_ERROR;
354 case QED_ROCE_QP_STATE_SQE:
355 return QED_IWARP_QP_STATE_TERMINATE;
356 default:
357 return QED_IWARP_QP_STATE_ERROR;
361 static enum qed_roce_qp_state
362 qed_iwarp2roce_state(enum qed_iwarp_qp_state state)
364 switch (state) {
365 case QED_IWARP_QP_STATE_IDLE:
366 return QED_ROCE_QP_STATE_INIT;
367 case QED_IWARP_QP_STATE_RTS:
368 return QED_ROCE_QP_STATE_RTS;
369 case QED_IWARP_QP_STATE_TERMINATE:
370 return QED_ROCE_QP_STATE_SQE;
371 case QED_IWARP_QP_STATE_CLOSING:
372 return QED_ROCE_QP_STATE_SQD;
373 case QED_IWARP_QP_STATE_ERROR:
374 return QED_ROCE_QP_STATE_ERR;
375 default:
376 return QED_ROCE_QP_STATE_ERR;
380 const static char *iwarp_state_names[] = {
381 "IDLE",
382 "RTS",
383 "TERMINATE",
384 "CLOSING",
385 "ERROR",
389 qed_iwarp_modify_qp(struct qed_hwfn *p_hwfn,
390 struct qed_rdma_qp *qp,
391 enum qed_iwarp_qp_state new_state, bool internal)
393 enum qed_iwarp_qp_state prev_iw_state;
394 bool modify_fw = false;
395 int rc = 0;
397 /* modify QP can be called from upper-layer or as a result of async
398 * RST/FIN... therefore need to protect
400 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.qp_lock);
401 prev_iw_state = qp->iwarp_state;
403 if (prev_iw_state == new_state) {
404 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.qp_lock);
405 return 0;
408 switch (prev_iw_state) {
409 case QED_IWARP_QP_STATE_IDLE:
410 switch (new_state) {
411 case QED_IWARP_QP_STATE_RTS:
412 qp->iwarp_state = QED_IWARP_QP_STATE_RTS;
413 break;
414 case QED_IWARP_QP_STATE_ERROR:
415 qp->iwarp_state = QED_IWARP_QP_STATE_ERROR;
416 if (!internal)
417 modify_fw = true;
418 break;
419 default:
420 break;
422 break;
423 case QED_IWARP_QP_STATE_RTS:
424 switch (new_state) {
425 case QED_IWARP_QP_STATE_CLOSING:
426 if (!internal)
427 modify_fw = true;
429 qp->iwarp_state = QED_IWARP_QP_STATE_CLOSING;
430 break;
431 case QED_IWARP_QP_STATE_ERROR:
432 if (!internal)
433 modify_fw = true;
434 qp->iwarp_state = QED_IWARP_QP_STATE_ERROR;
435 break;
436 default:
437 break;
439 break;
440 case QED_IWARP_QP_STATE_ERROR:
441 switch (new_state) {
442 case QED_IWARP_QP_STATE_IDLE:
444 qp->iwarp_state = new_state;
445 break;
446 case QED_IWARP_QP_STATE_CLOSING:
447 /* could happen due to race... do nothing.... */
448 break;
449 default:
450 rc = -EINVAL;
452 break;
453 case QED_IWARP_QP_STATE_TERMINATE:
454 case QED_IWARP_QP_STATE_CLOSING:
455 qp->iwarp_state = new_state;
456 break;
457 default:
458 break;
461 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) %s --> %s%s\n",
462 qp->icid,
463 iwarp_state_names[prev_iw_state],
464 iwarp_state_names[qp->iwarp_state],
465 internal ? "internal" : "");
467 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.qp_lock);
469 if (modify_fw)
470 rc = qed_iwarp_modify_fw(p_hwfn, qp);
472 return rc;
475 int qed_iwarp_fw_destroy(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
477 struct qed_sp_init_data init_data;
478 struct qed_spq_entry *p_ent;
479 int rc;
481 /* Get SPQ entry */
482 memset(&init_data, 0, sizeof(init_data));
483 init_data.cid = qp->icid;
484 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
485 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
487 rc = qed_sp_init_request(p_hwfn, &p_ent,
488 IWARP_RAMROD_CMD_ID_DESTROY_QP,
489 p_hwfn->p_rdma_info->proto, &init_data);
490 if (rc)
491 return rc;
493 rc = qed_spq_post(p_hwfn, p_ent, NULL);
495 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) rc = %d\n", qp->icid, rc);
497 return rc;
500 static void qed_iwarp_destroy_ep(struct qed_hwfn *p_hwfn,
501 struct qed_iwarp_ep *ep,
502 bool remove_from_active_list)
504 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
505 sizeof(*ep->ep_buffer_virt),
506 ep->ep_buffer_virt, ep->ep_buffer_phys);
508 if (remove_from_active_list) {
509 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
510 list_del(&ep->list_entry);
511 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
514 if (ep->qp)
515 ep->qp->ep = NULL;
517 kfree(ep);
520 int qed_iwarp_destroy_qp(struct qed_hwfn *p_hwfn, struct qed_rdma_qp *qp)
522 struct qed_iwarp_ep *ep = qp->ep;
523 int wait_count = 0;
524 int rc = 0;
526 if (qp->iwarp_state != QED_IWARP_QP_STATE_ERROR) {
527 rc = qed_iwarp_modify_qp(p_hwfn, qp,
528 QED_IWARP_QP_STATE_ERROR, false);
529 if (rc)
530 return rc;
533 /* Make sure ep is closed before returning and freeing memory. */
534 if (ep) {
535 while (READ_ONCE(ep->state) != QED_IWARP_EP_CLOSED &&
536 wait_count++ < 200)
537 msleep(100);
539 if (ep->state != QED_IWARP_EP_CLOSED)
540 DP_NOTICE(p_hwfn, "ep state close timeout state=%x\n",
541 ep->state);
543 qed_iwarp_destroy_ep(p_hwfn, ep, false);
546 rc = qed_iwarp_fw_destroy(p_hwfn, qp);
548 if (qp->shared_queue)
549 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
550 IWARP_SHARED_QUEUE_PAGE_SIZE,
551 qp->shared_queue, qp->shared_queue_phys_addr);
553 return rc;
556 static int
557 qed_iwarp_create_ep(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep **ep_out)
559 struct qed_iwarp_ep *ep;
560 int rc;
562 ep = kzalloc(sizeof(*ep), GFP_KERNEL);
563 if (!ep)
564 return -ENOMEM;
566 ep->state = QED_IWARP_EP_INIT;
568 ep->ep_buffer_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
569 sizeof(*ep->ep_buffer_virt),
570 &ep->ep_buffer_phys,
571 GFP_KERNEL);
572 if (!ep->ep_buffer_virt) {
573 rc = -ENOMEM;
574 goto err;
577 ep->sig = QED_EP_SIG;
579 *ep_out = ep;
581 return 0;
583 err:
584 kfree(ep);
585 return rc;
588 static void
589 qed_iwarp_print_tcp_ramrod(struct qed_hwfn *p_hwfn,
590 struct iwarp_tcp_offload_ramrod_data *p_tcp_ramrod)
592 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "local_mac=%x %x %x, remote_mac=%x %x %x\n",
593 p_tcp_ramrod->tcp.local_mac_addr_lo,
594 p_tcp_ramrod->tcp.local_mac_addr_mid,
595 p_tcp_ramrod->tcp.local_mac_addr_hi,
596 p_tcp_ramrod->tcp.remote_mac_addr_lo,
597 p_tcp_ramrod->tcp.remote_mac_addr_mid,
598 p_tcp_ramrod->tcp.remote_mac_addr_hi);
600 if (p_tcp_ramrod->tcp.ip_version == TCP_IPV4) {
601 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
602 "local_ip=%pI4h:%x, remote_ip=%pI4h:%x, vlan=%x\n",
603 p_tcp_ramrod->tcp.local_ip,
604 p_tcp_ramrod->tcp.local_port,
605 p_tcp_ramrod->tcp.remote_ip,
606 p_tcp_ramrod->tcp.remote_port,
607 p_tcp_ramrod->tcp.vlan_id);
608 } else {
609 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
610 "local_ip=%pI6:%x, remote_ip=%pI6:%x, vlan=%x\n",
611 p_tcp_ramrod->tcp.local_ip,
612 p_tcp_ramrod->tcp.local_port,
613 p_tcp_ramrod->tcp.remote_ip,
614 p_tcp_ramrod->tcp.remote_port,
615 p_tcp_ramrod->tcp.vlan_id);
618 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
619 "flow_label=%x, ttl=%x, tos_or_tc=%x, mss=%x, rcv_wnd_scale=%x, connect_mode=%x, flags=%x\n",
620 p_tcp_ramrod->tcp.flow_label,
621 p_tcp_ramrod->tcp.ttl,
622 p_tcp_ramrod->tcp.tos_or_tc,
623 p_tcp_ramrod->tcp.mss,
624 p_tcp_ramrod->tcp.rcv_wnd_scale,
625 p_tcp_ramrod->tcp.connect_mode,
626 p_tcp_ramrod->tcp.flags);
628 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "syn_ip_payload_length=%x, lo=%x, hi=%x\n",
629 p_tcp_ramrod->tcp.syn_ip_payload_length,
630 p_tcp_ramrod->tcp.syn_phy_addr_lo,
631 p_tcp_ramrod->tcp.syn_phy_addr_hi);
634 static int
635 qed_iwarp_tcp_offload(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
637 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
638 struct iwarp_tcp_offload_ramrod_data *p_tcp_ramrod;
639 struct tcp_offload_params_opt2 *tcp;
640 struct qed_sp_init_data init_data;
641 struct qed_spq_entry *p_ent;
642 dma_addr_t async_output_phys;
643 dma_addr_t in_pdata_phys;
644 u16 physical_q;
645 u8 tcp_flags;
646 int rc;
647 int i;
649 memset(&init_data, 0, sizeof(init_data));
650 init_data.cid = ep->tcp_cid;
651 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
652 if (ep->connect_mode == TCP_CONNECT_PASSIVE)
653 init_data.comp_mode = QED_SPQ_MODE_CB;
654 else
655 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
657 rc = qed_sp_init_request(p_hwfn, &p_ent,
658 IWARP_RAMROD_CMD_ID_TCP_OFFLOAD,
659 PROTOCOLID_IWARP, &init_data);
660 if (rc)
661 return rc;
663 p_tcp_ramrod = &p_ent->ramrod.iwarp_tcp_offload;
665 in_pdata_phys = ep->ep_buffer_phys +
666 offsetof(struct qed_iwarp_ep_memory, in_pdata);
667 DMA_REGPAIR_LE(p_tcp_ramrod->iwarp.incoming_ulp_buffer.addr,
668 in_pdata_phys);
670 p_tcp_ramrod->iwarp.incoming_ulp_buffer.len =
671 cpu_to_le16(sizeof(ep->ep_buffer_virt->in_pdata));
673 async_output_phys = ep->ep_buffer_phys +
674 offsetof(struct qed_iwarp_ep_memory, async_output);
675 DMA_REGPAIR_LE(p_tcp_ramrod->iwarp.async_eqe_output_buf,
676 async_output_phys);
678 p_tcp_ramrod->iwarp.handle_for_async.hi = cpu_to_le32(PTR_HI(ep));
679 p_tcp_ramrod->iwarp.handle_for_async.lo = cpu_to_le32(PTR_LO(ep));
681 physical_q = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_OFLD);
682 p_tcp_ramrod->iwarp.physical_q0 = cpu_to_le16(physical_q);
683 physical_q = qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_ACK);
684 p_tcp_ramrod->iwarp.physical_q1 = cpu_to_le16(physical_q);
685 p_tcp_ramrod->iwarp.mpa_mode = iwarp_info->mpa_rev;
687 tcp = &p_tcp_ramrod->tcp;
688 qed_set_fw_mac_addr(&tcp->remote_mac_addr_hi,
689 &tcp->remote_mac_addr_mid,
690 &tcp->remote_mac_addr_lo, ep->remote_mac_addr);
691 qed_set_fw_mac_addr(&tcp->local_mac_addr_hi, &tcp->local_mac_addr_mid,
692 &tcp->local_mac_addr_lo, ep->local_mac_addr);
694 tcp->vlan_id = cpu_to_le16(ep->cm_info.vlan);
696 tcp_flags = p_hwfn->p_rdma_info->iwarp.tcp_flags;
697 tcp->flags = 0;
698 SET_FIELD(tcp->flags, TCP_OFFLOAD_PARAMS_OPT2_TS_EN,
699 !!(tcp_flags & QED_IWARP_TS_EN));
701 SET_FIELD(tcp->flags, TCP_OFFLOAD_PARAMS_OPT2_DA_EN,
702 !!(tcp_flags & QED_IWARP_DA_EN));
704 tcp->ip_version = ep->cm_info.ip_version;
706 for (i = 0; i < 4; i++) {
707 tcp->remote_ip[i] = cpu_to_le32(ep->cm_info.remote_ip[i]);
708 tcp->local_ip[i] = cpu_to_le32(ep->cm_info.local_ip[i]);
711 tcp->remote_port = cpu_to_le16(ep->cm_info.remote_port);
712 tcp->local_port = cpu_to_le16(ep->cm_info.local_port);
713 tcp->mss = cpu_to_le16(ep->mss);
714 tcp->flow_label = 0;
715 tcp->ttl = 0x40;
716 tcp->tos_or_tc = 0;
718 tcp->max_rt_time = QED_IWARP_DEF_MAX_RT_TIME;
719 tcp->cwnd = QED_IWARP_DEF_CWND_FACTOR * tcp->mss;
720 tcp->ka_max_probe_cnt = QED_IWARP_DEF_KA_MAX_PROBE_CNT;
721 tcp->ka_timeout = QED_IWARP_DEF_KA_TIMEOUT;
722 tcp->ka_interval = QED_IWARP_DEF_KA_INTERVAL;
724 tcp->rcv_wnd_scale = (u8)p_hwfn->p_rdma_info->iwarp.rcv_wnd_scale;
725 tcp->connect_mode = ep->connect_mode;
727 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
728 tcp->syn_ip_payload_length =
729 cpu_to_le16(ep->syn_ip_payload_length);
730 tcp->syn_phy_addr_hi = DMA_HI_LE(ep->syn_phy_addr);
731 tcp->syn_phy_addr_lo = DMA_LO_LE(ep->syn_phy_addr);
734 qed_iwarp_print_tcp_ramrod(p_hwfn, p_tcp_ramrod);
736 rc = qed_spq_post(p_hwfn, p_ent, NULL);
738 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
739 "EP(0x%x) Offload completed rc=%d\n", ep->tcp_cid, rc);
741 return rc;
744 static void
745 qed_iwarp_mpa_received(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
747 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
748 struct qed_iwarp_cm_event_params params;
749 struct mpa_v2_hdr *mpa_v2;
750 union async_output *async_data;
751 u16 mpa_ord, mpa_ird;
752 u8 mpa_hdr_size = 0;
753 u8 mpa_rev;
755 async_data = &ep->ep_buffer_virt->async_output;
757 mpa_rev = async_data->mpa_request.mpa_handshake_mode;
758 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
759 "private_data_len=%x handshake_mode=%x private_data=(%x)\n",
760 async_data->mpa_request.ulp_data_len,
761 mpa_rev, *((u32 *)(ep->ep_buffer_virt->in_pdata)));
763 if (mpa_rev == MPA_NEGOTIATION_TYPE_ENHANCED) {
764 /* Read ord/ird values from private data buffer */
765 mpa_v2 = (struct mpa_v2_hdr *)ep->ep_buffer_virt->in_pdata;
766 mpa_hdr_size = sizeof(*mpa_v2);
768 mpa_ord = ntohs(mpa_v2->ord);
769 mpa_ird = ntohs(mpa_v2->ird);
771 /* Temprary store in cm_info incoming ord/ird requested, later
772 * replace with negotiated value during accept
774 ep->cm_info.ord = (u8)min_t(u16,
775 (mpa_ord & MPA_V2_IRD_ORD_MASK),
776 QED_IWARP_ORD_DEFAULT);
778 ep->cm_info.ird = (u8)min_t(u16,
779 (mpa_ird & MPA_V2_IRD_ORD_MASK),
780 QED_IWARP_IRD_DEFAULT);
782 /* Peer2Peer negotiation */
783 ep->rtr_type = MPA_RTR_TYPE_NONE;
784 if (mpa_ird & MPA_V2_PEER2PEER_MODEL) {
785 if (mpa_ord & MPA_V2_WRITE_RTR)
786 ep->rtr_type |= MPA_RTR_TYPE_ZERO_WRITE;
788 if (mpa_ord & MPA_V2_READ_RTR)
789 ep->rtr_type |= MPA_RTR_TYPE_ZERO_READ;
791 if (mpa_ird & MPA_V2_SEND_RTR)
792 ep->rtr_type |= MPA_RTR_TYPE_ZERO_SEND;
794 ep->rtr_type &= iwarp_info->rtr_type;
796 /* if we're left with no match send our capabilities */
797 if (ep->rtr_type == MPA_RTR_TYPE_NONE)
798 ep->rtr_type = iwarp_info->rtr_type;
801 ep->mpa_rev = MPA_NEGOTIATION_TYPE_ENHANCED;
802 } else {
803 ep->cm_info.ord = QED_IWARP_ORD_DEFAULT;
804 ep->cm_info.ird = QED_IWARP_IRD_DEFAULT;
805 ep->mpa_rev = MPA_NEGOTIATION_TYPE_BASIC;
808 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
809 "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x rtr:0x%x ulp_data_len = %x mpa_hdr_size = %x\n",
810 mpa_rev, ep->cm_info.ord, ep->cm_info.ird, ep->rtr_type,
811 async_data->mpa_request.ulp_data_len, mpa_hdr_size);
813 /* Strip mpa v2 hdr from private data before sending to upper layer */
814 ep->cm_info.private_data = ep->ep_buffer_virt->in_pdata + mpa_hdr_size;
816 ep->cm_info.private_data_len = async_data->mpa_request.ulp_data_len -
817 mpa_hdr_size;
819 params.event = QED_IWARP_EVENT_MPA_REQUEST;
820 params.cm_info = &ep->cm_info;
821 params.ep_context = ep;
822 params.status = 0;
824 ep->state = QED_IWARP_EP_MPA_REQ_RCVD;
825 ep->event_cb(ep->cb_context, &params);
828 static int
829 qed_iwarp_mpa_offload(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
831 struct iwarp_mpa_offload_ramrod_data *p_mpa_ramrod;
832 struct qed_iwarp_info *iwarp_info;
833 struct qed_sp_init_data init_data;
834 dma_addr_t async_output_phys;
835 struct qed_spq_entry *p_ent;
836 dma_addr_t out_pdata_phys;
837 dma_addr_t in_pdata_phys;
838 struct qed_rdma_qp *qp;
839 bool reject;
840 int rc;
842 if (!ep)
843 return -EINVAL;
845 qp = ep->qp;
846 reject = !qp;
848 memset(&init_data, 0, sizeof(init_data));
849 init_data.cid = reject ? ep->tcp_cid : qp->icid;
850 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
852 if (ep->connect_mode == TCP_CONNECT_ACTIVE)
853 init_data.comp_mode = QED_SPQ_MODE_CB;
854 else
855 init_data.comp_mode = QED_SPQ_MODE_EBLOCK;
857 rc = qed_sp_init_request(p_hwfn, &p_ent,
858 IWARP_RAMROD_CMD_ID_MPA_OFFLOAD,
859 PROTOCOLID_IWARP, &init_data);
860 if (rc)
861 return rc;
863 p_mpa_ramrod = &p_ent->ramrod.iwarp_mpa_offload;
864 out_pdata_phys = ep->ep_buffer_phys +
865 offsetof(struct qed_iwarp_ep_memory, out_pdata);
866 DMA_REGPAIR_LE(p_mpa_ramrod->common.outgoing_ulp_buffer.addr,
867 out_pdata_phys);
868 p_mpa_ramrod->common.outgoing_ulp_buffer.len =
869 ep->cm_info.private_data_len;
870 p_mpa_ramrod->common.crc_needed = p_hwfn->p_rdma_info->iwarp.crc_needed;
872 p_mpa_ramrod->common.out_rq.ord = ep->cm_info.ord;
873 p_mpa_ramrod->common.out_rq.ird = ep->cm_info.ird;
875 p_mpa_ramrod->tcp_cid = p_hwfn->hw_info.opaque_fid << 16 | ep->tcp_cid;
877 in_pdata_phys = ep->ep_buffer_phys +
878 offsetof(struct qed_iwarp_ep_memory, in_pdata);
879 p_mpa_ramrod->tcp_connect_side = ep->connect_mode;
880 DMA_REGPAIR_LE(p_mpa_ramrod->incoming_ulp_buffer.addr,
881 in_pdata_phys);
882 p_mpa_ramrod->incoming_ulp_buffer.len =
883 cpu_to_le16(sizeof(ep->ep_buffer_virt->in_pdata));
884 async_output_phys = ep->ep_buffer_phys +
885 offsetof(struct qed_iwarp_ep_memory, async_output);
886 DMA_REGPAIR_LE(p_mpa_ramrod->async_eqe_output_buf,
887 async_output_phys);
888 p_mpa_ramrod->handle_for_async.hi = cpu_to_le32(PTR_HI(ep));
889 p_mpa_ramrod->handle_for_async.lo = cpu_to_le32(PTR_LO(ep));
891 if (!reject) {
892 DMA_REGPAIR_LE(p_mpa_ramrod->shared_queue_addr,
893 qp->shared_queue_phys_addr);
894 p_mpa_ramrod->stats_counter_id =
895 RESC_START(p_hwfn, QED_RDMA_STATS_QUEUE) + qp->stats_queue;
896 } else {
897 p_mpa_ramrod->common.reject = 1;
900 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
901 p_mpa_ramrod->rcv_wnd = iwarp_info->rcv_wnd_size;
902 p_mpa_ramrod->mode = ep->mpa_rev;
903 SET_FIELD(p_mpa_ramrod->rtr_pref,
904 IWARP_MPA_OFFLOAD_RAMROD_DATA_RTR_SUPPORTED, ep->rtr_type);
906 ep->state = QED_IWARP_EP_MPA_OFFLOADED;
907 rc = qed_spq_post(p_hwfn, p_ent, NULL);
908 if (!reject)
909 ep->cid = qp->icid; /* Now they're migrated. */
911 DP_VERBOSE(p_hwfn,
912 QED_MSG_RDMA,
913 "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",
914 reject ? 0xffff : qp->icid,
915 ep->tcp_cid,
917 ep->cm_info.ird,
918 ep->cm_info.ord, ep->rtr_type, ep->mpa_rev, reject);
919 return rc;
922 static void
923 qed_iwarp_return_ep(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
925 ep->state = QED_IWARP_EP_INIT;
926 if (ep->qp)
927 ep->qp->ep = NULL;
928 ep->qp = NULL;
929 memset(&ep->cm_info, 0, sizeof(ep->cm_info));
931 if (ep->tcp_cid == QED_IWARP_INVALID_TCP_CID) {
932 /* We don't care about the return code, it's ok if tcp_cid
933 * remains invalid...in this case we'll defer allocation
935 qed_iwarp_alloc_tcp_cid(p_hwfn, &ep->tcp_cid);
937 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
939 list_del(&ep->list_entry);
940 list_add_tail(&ep->list_entry,
941 &p_hwfn->p_rdma_info->iwarp.ep_free_list);
943 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
946 static void
947 qed_iwarp_parse_private_data(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
949 struct mpa_v2_hdr *mpa_v2_params;
950 union async_output *async_data;
951 u16 mpa_ird, mpa_ord;
952 u8 mpa_data_size = 0;
954 if (MPA_REV2(p_hwfn->p_rdma_info->iwarp.mpa_rev)) {
955 mpa_v2_params =
956 (struct mpa_v2_hdr *)(ep->ep_buffer_virt->in_pdata);
957 mpa_data_size = sizeof(*mpa_v2_params);
958 mpa_ird = ntohs(mpa_v2_params->ird);
959 mpa_ord = ntohs(mpa_v2_params->ord);
961 ep->cm_info.ird = (u8)(mpa_ord & MPA_V2_IRD_ORD_MASK);
962 ep->cm_info.ord = (u8)(mpa_ird & MPA_V2_IRD_ORD_MASK);
964 async_data = &ep->ep_buffer_virt->async_output;
966 ep->cm_info.private_data = ep->ep_buffer_virt->in_pdata + mpa_data_size;
967 ep->cm_info.private_data_len = async_data->mpa_response.ulp_data_len -
968 mpa_data_size;
971 static void
972 qed_iwarp_mpa_reply_arrived(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
974 struct qed_iwarp_cm_event_params params;
976 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
977 DP_NOTICE(p_hwfn,
978 "MPA reply event not expected on passive side!\n");
979 return;
982 params.event = QED_IWARP_EVENT_ACTIVE_MPA_REPLY;
984 qed_iwarp_parse_private_data(p_hwfn, ep);
986 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
987 "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x\n",
988 ep->mpa_rev, ep->cm_info.ord, ep->cm_info.ird);
990 params.cm_info = &ep->cm_info;
991 params.ep_context = ep;
992 params.status = 0;
994 ep->mpa_reply_processed = true;
996 ep->event_cb(ep->cb_context, &params);
999 #define QED_IWARP_CONNECT_MODE_STRING(ep) \
1000 ((ep)->connect_mode == TCP_CONNECT_PASSIVE) ? "Passive" : "Active"
1002 /* Called as a result of the event:
1003 * IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE
1005 static void
1006 qed_iwarp_mpa_complete(struct qed_hwfn *p_hwfn,
1007 struct qed_iwarp_ep *ep, u8 fw_return_code)
1009 struct qed_iwarp_cm_event_params params;
1011 if (ep->connect_mode == TCP_CONNECT_ACTIVE)
1012 params.event = QED_IWARP_EVENT_ACTIVE_COMPLETE;
1013 else
1014 params.event = QED_IWARP_EVENT_PASSIVE_COMPLETE;
1016 if (ep->connect_mode == TCP_CONNECT_ACTIVE && !ep->mpa_reply_processed)
1017 qed_iwarp_parse_private_data(p_hwfn, ep);
1019 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1020 "MPA_NEGOTIATE (v%d): ORD: 0x%x IRD: 0x%x\n",
1021 ep->mpa_rev, ep->cm_info.ord, ep->cm_info.ird);
1023 params.cm_info = &ep->cm_info;
1025 params.ep_context = ep;
1027 switch (fw_return_code) {
1028 case RDMA_RETURN_OK:
1029 ep->qp->max_rd_atomic_req = ep->cm_info.ord;
1030 ep->qp->max_rd_atomic_resp = ep->cm_info.ird;
1031 qed_iwarp_modify_qp(p_hwfn, ep->qp, QED_IWARP_QP_STATE_RTS, 1);
1032 ep->state = QED_IWARP_EP_ESTABLISHED;
1033 params.status = 0;
1034 break;
1035 case IWARP_CONN_ERROR_MPA_TIMEOUT:
1036 DP_NOTICE(p_hwfn, "%s(0x%x) MPA timeout\n",
1037 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1038 params.status = -EBUSY;
1039 break;
1040 case IWARP_CONN_ERROR_MPA_ERROR_REJECT:
1041 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Reject\n",
1042 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1043 params.status = -ECONNREFUSED;
1044 break;
1045 case IWARP_CONN_ERROR_MPA_RST:
1046 DP_NOTICE(p_hwfn, "%s(0x%x) MPA reset(tcp cid: 0x%x)\n",
1047 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid,
1048 ep->tcp_cid);
1049 params.status = -ECONNRESET;
1050 break;
1051 case IWARP_CONN_ERROR_MPA_FIN:
1052 DP_NOTICE(p_hwfn, "%s(0x%x) MPA received FIN\n",
1053 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1054 params.status = -ECONNREFUSED;
1055 break;
1056 case IWARP_CONN_ERROR_MPA_INSUF_IRD:
1057 DP_NOTICE(p_hwfn, "%s(0x%x) MPA insufficient ird\n",
1058 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1059 params.status = -ECONNREFUSED;
1060 break;
1061 case IWARP_CONN_ERROR_MPA_RTR_MISMATCH:
1062 DP_NOTICE(p_hwfn, "%s(0x%x) MPA RTR MISMATCH\n",
1063 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1064 params.status = -ECONNREFUSED;
1065 break;
1066 case IWARP_CONN_ERROR_MPA_INVALID_PACKET:
1067 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Invalid Packet\n",
1068 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1069 params.status = -ECONNREFUSED;
1070 break;
1071 case IWARP_CONN_ERROR_MPA_LOCAL_ERROR:
1072 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Local Error\n",
1073 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1074 params.status = -ECONNREFUSED;
1075 break;
1076 case IWARP_CONN_ERROR_MPA_TERMINATE:
1077 DP_NOTICE(p_hwfn, "%s(0x%x) MPA TERMINATE\n",
1078 QED_IWARP_CONNECT_MODE_STRING(ep), ep->cid);
1079 params.status = -ECONNREFUSED;
1080 break;
1081 default:
1082 params.status = -ECONNRESET;
1083 break;
1086 if (fw_return_code != RDMA_RETURN_OK)
1087 /* paired with READ_ONCE in destroy_qp */
1088 smp_store_release(&ep->state, QED_IWARP_EP_CLOSED);
1090 ep->event_cb(ep->cb_context, &params);
1092 /* on passive side, if there is no associated QP (REJECT) we need to
1093 * return the ep to the pool, (in the regular case we add an element
1094 * in accept instead of this one.
1095 * In both cases we need to remove it from the ep_list.
1097 if (fw_return_code != RDMA_RETURN_OK) {
1098 ep->tcp_cid = QED_IWARP_INVALID_TCP_CID;
1099 if ((ep->connect_mode == TCP_CONNECT_PASSIVE) &&
1100 (!ep->qp)) { /* Rejected */
1101 qed_iwarp_return_ep(p_hwfn, ep);
1102 } else {
1103 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1104 list_del(&ep->list_entry);
1105 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1110 static void
1111 qed_iwarp_mpa_v2_set_private(struct qed_hwfn *p_hwfn,
1112 struct qed_iwarp_ep *ep, u8 *mpa_data_size)
1114 struct mpa_v2_hdr *mpa_v2_params;
1115 u16 mpa_ird, mpa_ord;
1117 *mpa_data_size = 0;
1118 if (MPA_REV2(ep->mpa_rev)) {
1119 mpa_v2_params =
1120 (struct mpa_v2_hdr *)ep->ep_buffer_virt->out_pdata;
1121 *mpa_data_size = sizeof(*mpa_v2_params);
1123 mpa_ird = (u16)ep->cm_info.ird;
1124 mpa_ord = (u16)ep->cm_info.ord;
1126 if (ep->rtr_type != MPA_RTR_TYPE_NONE) {
1127 mpa_ird |= MPA_V2_PEER2PEER_MODEL;
1129 if (ep->rtr_type & MPA_RTR_TYPE_ZERO_SEND)
1130 mpa_ird |= MPA_V2_SEND_RTR;
1132 if (ep->rtr_type & MPA_RTR_TYPE_ZERO_WRITE)
1133 mpa_ord |= MPA_V2_WRITE_RTR;
1135 if (ep->rtr_type & MPA_RTR_TYPE_ZERO_READ)
1136 mpa_ord |= MPA_V2_READ_RTR;
1139 mpa_v2_params->ird = htons(mpa_ird);
1140 mpa_v2_params->ord = htons(mpa_ord);
1142 DP_VERBOSE(p_hwfn,
1143 QED_MSG_RDMA,
1144 "MPA_NEGOTIATE Header: [%x ord:%x ird] %x ord:%x ird:%x peer2peer:%x rtr_send:%x rtr_write:%x rtr_read:%x\n",
1145 mpa_v2_params->ird,
1146 mpa_v2_params->ord,
1147 *((u32 *)mpa_v2_params),
1148 mpa_ord & MPA_V2_IRD_ORD_MASK,
1149 mpa_ird & MPA_V2_IRD_ORD_MASK,
1150 !!(mpa_ird & MPA_V2_PEER2PEER_MODEL),
1151 !!(mpa_ird & MPA_V2_SEND_RTR),
1152 !!(mpa_ord & MPA_V2_WRITE_RTR),
1153 !!(mpa_ord & MPA_V2_READ_RTR));
1157 int qed_iwarp_connect(void *rdma_cxt,
1158 struct qed_iwarp_connect_in *iparams,
1159 struct qed_iwarp_connect_out *oparams)
1161 struct qed_hwfn *p_hwfn = rdma_cxt;
1162 struct qed_iwarp_info *iwarp_info;
1163 struct qed_iwarp_ep *ep;
1164 u8 mpa_data_size = 0;
1165 u32 cid;
1166 int rc;
1168 if ((iparams->cm_info.ord > QED_IWARP_ORD_DEFAULT) ||
1169 (iparams->cm_info.ird > QED_IWARP_IRD_DEFAULT)) {
1170 DP_NOTICE(p_hwfn,
1171 "QP(0x%x) ERROR: Invalid ord(0x%x)/ird(0x%x)\n",
1172 iparams->qp->icid, iparams->cm_info.ord,
1173 iparams->cm_info.ird);
1175 return -EINVAL;
1178 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
1180 /* Allocate ep object */
1181 rc = qed_iwarp_alloc_cid(p_hwfn, &cid);
1182 if (rc)
1183 return rc;
1185 rc = qed_iwarp_create_ep(p_hwfn, &ep);
1186 if (rc)
1187 goto err;
1189 ep->tcp_cid = cid;
1191 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1192 list_add_tail(&ep->list_entry, &p_hwfn->p_rdma_info->iwarp.ep_list);
1193 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1195 ep->qp = iparams->qp;
1196 ep->qp->ep = ep;
1197 ether_addr_copy(ep->remote_mac_addr, iparams->remote_mac_addr);
1198 ether_addr_copy(ep->local_mac_addr, iparams->local_mac_addr);
1199 memcpy(&ep->cm_info, &iparams->cm_info, sizeof(ep->cm_info));
1201 ep->cm_info.ord = iparams->cm_info.ord;
1202 ep->cm_info.ird = iparams->cm_info.ird;
1204 ep->rtr_type = iwarp_info->rtr_type;
1205 if (!iwarp_info->peer2peer)
1206 ep->rtr_type = MPA_RTR_TYPE_NONE;
1208 if ((ep->rtr_type & MPA_RTR_TYPE_ZERO_READ) && (ep->cm_info.ord == 0))
1209 ep->cm_info.ord = 1;
1211 ep->mpa_rev = iwarp_info->mpa_rev;
1213 qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
1215 ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
1216 ep->cm_info.private_data_len = iparams->cm_info.private_data_len +
1217 mpa_data_size;
1219 memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
1220 iparams->cm_info.private_data,
1221 iparams->cm_info.private_data_len);
1223 ep->mss = iparams->mss;
1224 ep->mss = min_t(u16, QED_IWARP_MAX_FW_MSS, ep->mss);
1226 ep->event_cb = iparams->event_cb;
1227 ep->cb_context = iparams->cb_context;
1228 ep->connect_mode = TCP_CONNECT_ACTIVE;
1230 oparams->ep_context = ep;
1232 rc = qed_iwarp_tcp_offload(p_hwfn, ep);
1234 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) EP(0x%x) rc = %d\n",
1235 iparams->qp->icid, ep->tcp_cid, rc);
1237 if (rc) {
1238 qed_iwarp_destroy_ep(p_hwfn, ep, true);
1239 goto err;
1242 return rc;
1243 err:
1244 qed_iwarp_cid_cleaned(p_hwfn, cid);
1246 return rc;
1249 static struct qed_iwarp_ep *qed_iwarp_get_free_ep(struct qed_hwfn *p_hwfn)
1251 struct qed_iwarp_ep *ep = NULL;
1252 int rc;
1254 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1256 if (list_empty(&p_hwfn->p_rdma_info->iwarp.ep_free_list)) {
1257 DP_ERR(p_hwfn, "Ep list is empty\n");
1258 goto out;
1261 ep = list_first_entry(&p_hwfn->p_rdma_info->iwarp.ep_free_list,
1262 struct qed_iwarp_ep, list_entry);
1264 /* in some cases we could have failed allocating a tcp cid when added
1265 * from accept / failure... retry now..this is not the common case.
1267 if (ep->tcp_cid == QED_IWARP_INVALID_TCP_CID) {
1268 rc = qed_iwarp_alloc_tcp_cid(p_hwfn, &ep->tcp_cid);
1270 /* if we fail we could look for another entry with a valid
1271 * tcp_cid, but since we don't expect to reach this anyway
1272 * it's not worth the handling
1274 if (rc) {
1275 ep->tcp_cid = QED_IWARP_INVALID_TCP_CID;
1276 ep = NULL;
1277 goto out;
1281 list_del(&ep->list_entry);
1283 out:
1284 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1285 return ep;
1288 #define QED_IWARP_MAX_CID_CLEAN_TIME 100
1289 #define QED_IWARP_MAX_NO_PROGRESS_CNT 5
1291 /* This function waits for all the bits of a bmap to be cleared, as long as
1292 * there is progress ( i.e. the number of bits left to be cleared decreases )
1293 * the function continues.
1295 static int
1296 qed_iwarp_wait_cid_map_cleared(struct qed_hwfn *p_hwfn, struct qed_bmap *bmap)
1298 int prev_weight = 0;
1299 int wait_count = 0;
1300 int weight = 0;
1302 weight = bitmap_weight(bmap->bitmap, bmap->max_count);
1303 prev_weight = weight;
1305 while (weight) {
1306 msleep(QED_IWARP_MAX_CID_CLEAN_TIME);
1308 weight = bitmap_weight(bmap->bitmap, bmap->max_count);
1310 if (prev_weight == weight) {
1311 wait_count++;
1312 } else {
1313 prev_weight = weight;
1314 wait_count = 0;
1317 if (wait_count > QED_IWARP_MAX_NO_PROGRESS_CNT) {
1318 DP_NOTICE(p_hwfn,
1319 "%s bitmap wait timed out (%d cids pending)\n",
1320 bmap->name, weight);
1321 return -EBUSY;
1324 return 0;
1327 static int qed_iwarp_wait_for_all_cids(struct qed_hwfn *p_hwfn)
1329 int rc;
1330 int i;
1332 rc = qed_iwarp_wait_cid_map_cleared(p_hwfn,
1333 &p_hwfn->p_rdma_info->tcp_cid_map);
1334 if (rc)
1335 return rc;
1337 /* Now free the tcp cids from the main cid map */
1338 for (i = 0; i < QED_IWARP_PREALLOC_CNT; i++)
1339 qed_bmap_release_id(p_hwfn, &p_hwfn->p_rdma_info->cid_map, i);
1341 /* Now wait for all cids to be completed */
1342 return qed_iwarp_wait_cid_map_cleared(p_hwfn,
1343 &p_hwfn->p_rdma_info->cid_map);
1346 static void qed_iwarp_free_prealloc_ep(struct qed_hwfn *p_hwfn)
1348 struct qed_iwarp_ep *ep;
1350 while (!list_empty(&p_hwfn->p_rdma_info->iwarp.ep_free_list)) {
1351 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1353 ep = list_first_entry(&p_hwfn->p_rdma_info->iwarp.ep_free_list,
1354 struct qed_iwarp_ep, list_entry);
1356 if (!ep) {
1357 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1358 break;
1360 list_del(&ep->list_entry);
1362 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1364 if (ep->tcp_cid != QED_IWARP_INVALID_TCP_CID)
1365 qed_iwarp_cid_cleaned(p_hwfn, ep->tcp_cid);
1367 qed_iwarp_destroy_ep(p_hwfn, ep, false);
1371 static int qed_iwarp_prealloc_ep(struct qed_hwfn *p_hwfn, bool init)
1373 struct qed_iwarp_ep *ep;
1374 int rc = 0;
1375 int count;
1376 u32 cid;
1377 int i;
1379 count = init ? QED_IWARP_PREALLOC_CNT : 1;
1380 for (i = 0; i < count; i++) {
1381 rc = qed_iwarp_create_ep(p_hwfn, &ep);
1382 if (rc)
1383 return rc;
1385 /* During initialization we allocate from the main pool,
1386 * afterwards we allocate only from the tcp_cid.
1388 if (init) {
1389 rc = qed_iwarp_alloc_cid(p_hwfn, &cid);
1390 if (rc)
1391 goto err;
1392 qed_iwarp_set_tcp_cid(p_hwfn, cid);
1393 } else {
1394 /* We don't care about the return code, it's ok if
1395 * tcp_cid remains invalid...in this case we'll
1396 * defer allocation
1398 qed_iwarp_alloc_tcp_cid(p_hwfn, &cid);
1401 ep->tcp_cid = cid;
1403 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1404 list_add_tail(&ep->list_entry,
1405 &p_hwfn->p_rdma_info->iwarp.ep_free_list);
1406 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1409 return rc;
1411 err:
1412 qed_iwarp_destroy_ep(p_hwfn, ep, false);
1414 return rc;
1417 int qed_iwarp_alloc(struct qed_hwfn *p_hwfn)
1419 int rc;
1421 /* Allocate bitmap for tcp cid. These are used by passive side
1422 * to ensure it can allocate a tcp cid during dpc that was
1423 * pre-acquired and doesn't require dynamic allocation of ilt
1425 rc = qed_rdma_bmap_alloc(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map,
1426 QED_IWARP_PREALLOC_CNT, "TCP_CID");
1427 if (rc) {
1428 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1429 "Failed to allocate tcp cid, rc = %d\n", rc);
1430 return rc;
1433 INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.ep_free_list);
1434 spin_lock_init(&p_hwfn->p_rdma_info->iwarp.iw_lock);
1436 rc = qed_iwarp_prealloc_ep(p_hwfn, true);
1437 if (rc)
1438 return rc;
1440 return qed_ooo_alloc(p_hwfn);
1443 void qed_iwarp_resc_free(struct qed_hwfn *p_hwfn)
1445 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
1447 qed_ooo_free(p_hwfn);
1448 qed_rdma_bmap_free(p_hwfn, &p_hwfn->p_rdma_info->tcp_cid_map, 1);
1449 kfree(iwarp_info->mpa_bufs);
1450 kfree(iwarp_info->partial_fpdus);
1451 kfree(iwarp_info->mpa_intermediate_buf);
1454 int qed_iwarp_accept(void *rdma_cxt, struct qed_iwarp_accept_in *iparams)
1456 struct qed_hwfn *p_hwfn = rdma_cxt;
1457 struct qed_iwarp_ep *ep;
1458 u8 mpa_data_size = 0;
1459 int rc;
1461 ep = iparams->ep_context;
1462 if (!ep) {
1463 DP_ERR(p_hwfn, "Ep Context receive in accept is NULL\n");
1464 return -EINVAL;
1467 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) EP(0x%x)\n",
1468 iparams->qp->icid, ep->tcp_cid);
1470 if ((iparams->ord > QED_IWARP_ORD_DEFAULT) ||
1471 (iparams->ird > QED_IWARP_IRD_DEFAULT)) {
1472 DP_VERBOSE(p_hwfn,
1473 QED_MSG_RDMA,
1474 "QP(0x%x) EP(0x%x) ERROR: Invalid ord(0x%x)/ird(0x%x)\n",
1475 iparams->qp->icid,
1476 ep->tcp_cid, iparams->ord, iparams->ord);
1477 return -EINVAL;
1480 qed_iwarp_prealloc_ep(p_hwfn, false);
1482 ep->cb_context = iparams->cb_context;
1483 ep->qp = iparams->qp;
1484 ep->qp->ep = ep;
1486 if (ep->mpa_rev == MPA_NEGOTIATION_TYPE_ENHANCED) {
1487 /* Negotiate ord/ird: if upperlayer requested ord larger than
1488 * ird advertised by remote, we need to decrease our ord
1490 if (iparams->ord > ep->cm_info.ird)
1491 iparams->ord = ep->cm_info.ird;
1493 if ((ep->rtr_type & MPA_RTR_TYPE_ZERO_READ) &&
1494 (iparams->ird == 0))
1495 iparams->ird = 1;
1498 /* Update cm_info ord/ird to be negotiated values */
1499 ep->cm_info.ord = iparams->ord;
1500 ep->cm_info.ird = iparams->ird;
1502 qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
1504 ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
1505 ep->cm_info.private_data_len = iparams->private_data_len +
1506 mpa_data_size;
1508 memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
1509 iparams->private_data, iparams->private_data_len);
1511 rc = qed_iwarp_mpa_offload(p_hwfn, ep);
1512 if (rc)
1513 qed_iwarp_modify_qp(p_hwfn,
1514 iparams->qp, QED_IWARP_QP_STATE_ERROR, 1);
1516 return rc;
1519 int qed_iwarp_reject(void *rdma_cxt, struct qed_iwarp_reject_in *iparams)
1521 struct qed_hwfn *p_hwfn = rdma_cxt;
1522 struct qed_iwarp_ep *ep;
1523 u8 mpa_data_size = 0;
1525 ep = iparams->ep_context;
1526 if (!ep) {
1527 DP_ERR(p_hwfn, "Ep Context receive in reject is NULL\n");
1528 return -EINVAL;
1531 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "EP(0x%x)\n", ep->tcp_cid);
1533 ep->cb_context = iparams->cb_context;
1534 ep->qp = NULL;
1536 qed_iwarp_mpa_v2_set_private(p_hwfn, ep, &mpa_data_size);
1538 ep->cm_info.private_data = ep->ep_buffer_virt->out_pdata;
1539 ep->cm_info.private_data_len = iparams->private_data_len +
1540 mpa_data_size;
1542 memcpy((u8 *)ep->ep_buffer_virt->out_pdata + mpa_data_size,
1543 iparams->private_data, iparams->private_data_len);
1545 return qed_iwarp_mpa_offload(p_hwfn, ep);
1548 static void
1549 qed_iwarp_print_cm_info(struct qed_hwfn *p_hwfn,
1550 struct qed_iwarp_cm_info *cm_info)
1552 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "ip_version = %d\n",
1553 cm_info->ip_version);
1555 if (cm_info->ip_version == QED_TCP_IPV4)
1556 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1557 "remote_ip %pI4h:%x, local_ip %pI4h:%x vlan=%x\n",
1558 cm_info->remote_ip, cm_info->remote_port,
1559 cm_info->local_ip, cm_info->local_port,
1560 cm_info->vlan);
1561 else
1562 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1563 "remote_ip %pI6:%x, local_ip %pI6:%x vlan=%x\n",
1564 cm_info->remote_ip, cm_info->remote_port,
1565 cm_info->local_ip, cm_info->local_port,
1566 cm_info->vlan);
1568 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1569 "private_data_len = %x ord = %d, ird = %d\n",
1570 cm_info->private_data_len, cm_info->ord, cm_info->ird);
1573 static int
1574 qed_iwarp_ll2_post_rx(struct qed_hwfn *p_hwfn,
1575 struct qed_iwarp_ll2_buff *buf, u8 handle)
1577 int rc;
1579 rc = qed_ll2_post_rx_buffer(p_hwfn, handle, buf->data_phys_addr,
1580 (u16)buf->buff_size, buf, 1);
1581 if (rc) {
1582 DP_NOTICE(p_hwfn,
1583 "Failed to repost rx buffer to ll2 rc = %d, handle=%d\n",
1584 rc, handle);
1585 dma_free_coherent(&p_hwfn->cdev->pdev->dev, buf->buff_size,
1586 buf->data, buf->data_phys_addr);
1587 kfree(buf);
1590 return rc;
1593 static bool
1594 qed_iwarp_ep_exists(struct qed_hwfn *p_hwfn, struct qed_iwarp_cm_info *cm_info)
1596 struct qed_iwarp_ep *ep = NULL;
1597 bool found = false;
1599 list_for_each_entry(ep,
1600 &p_hwfn->p_rdma_info->iwarp.ep_list,
1601 list_entry) {
1602 if ((ep->cm_info.local_port == cm_info->local_port) &&
1603 (ep->cm_info.remote_port == cm_info->remote_port) &&
1604 (ep->cm_info.vlan == cm_info->vlan) &&
1605 !memcmp(&ep->cm_info.local_ip, cm_info->local_ip,
1606 sizeof(cm_info->local_ip)) &&
1607 !memcmp(&ep->cm_info.remote_ip, cm_info->remote_ip,
1608 sizeof(cm_info->remote_ip))) {
1609 found = true;
1610 break;
1614 if (found) {
1615 DP_NOTICE(p_hwfn,
1616 "SYN received on active connection - dropping\n");
1617 qed_iwarp_print_cm_info(p_hwfn, cm_info);
1619 return true;
1622 return false;
1625 static struct qed_iwarp_listener *
1626 qed_iwarp_get_listener(struct qed_hwfn *p_hwfn,
1627 struct qed_iwarp_cm_info *cm_info)
1629 struct qed_iwarp_listener *listener = NULL;
1630 static const u32 ip_zero[4] = { 0, 0, 0, 0 };
1631 bool found = false;
1633 qed_iwarp_print_cm_info(p_hwfn, cm_info);
1635 list_for_each_entry(listener,
1636 &p_hwfn->p_rdma_info->iwarp.listen_list,
1637 list_entry) {
1638 if (listener->port == cm_info->local_port) {
1639 if (!memcmp(listener->ip_addr,
1640 ip_zero, sizeof(ip_zero))) {
1641 found = true;
1642 break;
1645 if (!memcmp(listener->ip_addr,
1646 cm_info->local_ip,
1647 sizeof(cm_info->local_ip)) &&
1648 (listener->vlan == cm_info->vlan)) {
1649 found = true;
1650 break;
1655 if (found) {
1656 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "listener found = %p\n",
1657 listener);
1658 return listener;
1661 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "listener not found\n");
1662 return NULL;
1665 static int
1666 qed_iwarp_parse_rx_pkt(struct qed_hwfn *p_hwfn,
1667 struct qed_iwarp_cm_info *cm_info,
1668 void *buf,
1669 u8 *remote_mac_addr,
1670 u8 *local_mac_addr,
1671 int *payload_len, int *tcp_start_offset)
1673 struct vlan_ethhdr *vethh;
1674 bool vlan_valid = false;
1675 struct ipv6hdr *ip6h;
1676 struct ethhdr *ethh;
1677 struct tcphdr *tcph;
1678 struct iphdr *iph;
1679 int eth_hlen;
1680 int ip_hlen;
1681 int eth_type;
1682 int i;
1684 ethh = buf;
1685 eth_type = ntohs(ethh->h_proto);
1686 if (eth_type == ETH_P_8021Q) {
1687 vlan_valid = true;
1688 vethh = (struct vlan_ethhdr *)ethh;
1689 cm_info->vlan = ntohs(vethh->h_vlan_TCI) & VLAN_VID_MASK;
1690 eth_type = ntohs(vethh->h_vlan_encapsulated_proto);
1693 eth_hlen = ETH_HLEN + (vlan_valid ? sizeof(u32) : 0);
1695 if (!ether_addr_equal(ethh->h_dest,
1696 p_hwfn->p_rdma_info->iwarp.mac_addr)) {
1697 DP_VERBOSE(p_hwfn,
1698 QED_MSG_RDMA,
1699 "Got unexpected mac %pM instead of %pM\n",
1700 ethh->h_dest, p_hwfn->p_rdma_info->iwarp.mac_addr);
1701 return -EINVAL;
1704 ether_addr_copy(remote_mac_addr, ethh->h_source);
1705 ether_addr_copy(local_mac_addr, ethh->h_dest);
1707 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "eth_type =%d source mac: %pM\n",
1708 eth_type, ethh->h_source);
1710 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "eth_hlen=%d destination mac: %pM\n",
1711 eth_hlen, ethh->h_dest);
1713 iph = (struct iphdr *)((u8 *)(ethh) + eth_hlen);
1715 if (eth_type == ETH_P_IP) {
1716 if (iph->protocol != IPPROTO_TCP) {
1717 DP_NOTICE(p_hwfn,
1718 "Unexpected ip protocol on ll2 %x\n",
1719 iph->protocol);
1720 return -EINVAL;
1723 cm_info->local_ip[0] = ntohl(iph->daddr);
1724 cm_info->remote_ip[0] = ntohl(iph->saddr);
1725 cm_info->ip_version = QED_TCP_IPV4;
1727 ip_hlen = (iph->ihl) * sizeof(u32);
1728 *payload_len = ntohs(iph->tot_len) - ip_hlen;
1729 } else if (eth_type == ETH_P_IPV6) {
1730 ip6h = (struct ipv6hdr *)iph;
1732 if (ip6h->nexthdr != IPPROTO_TCP) {
1733 DP_NOTICE(p_hwfn,
1734 "Unexpected ip protocol on ll2 %x\n",
1735 iph->protocol);
1736 return -EINVAL;
1739 for (i = 0; i < 4; i++) {
1740 cm_info->local_ip[i] =
1741 ntohl(ip6h->daddr.in6_u.u6_addr32[i]);
1742 cm_info->remote_ip[i] =
1743 ntohl(ip6h->saddr.in6_u.u6_addr32[i]);
1745 cm_info->ip_version = QED_TCP_IPV6;
1747 ip_hlen = sizeof(*ip6h);
1748 *payload_len = ntohs(ip6h->payload_len);
1749 } else {
1750 DP_NOTICE(p_hwfn, "Unexpected ethertype on ll2 %x\n", eth_type);
1751 return -EINVAL;
1754 tcph = (struct tcphdr *)((u8 *)iph + ip_hlen);
1756 if (!tcph->syn) {
1757 DP_NOTICE(p_hwfn,
1758 "Only SYN type packet expected on this ll2 conn, iph->ihl=%d source=%d dest=%d\n",
1759 iph->ihl, tcph->source, tcph->dest);
1760 return -EINVAL;
1763 cm_info->local_port = ntohs(tcph->dest);
1764 cm_info->remote_port = ntohs(tcph->source);
1766 qed_iwarp_print_cm_info(p_hwfn, cm_info);
1768 *tcp_start_offset = eth_hlen + ip_hlen;
1770 return 0;
1773 static struct qed_iwarp_fpdu *qed_iwarp_get_curr_fpdu(struct qed_hwfn *p_hwfn,
1774 u16 cid)
1776 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
1777 struct qed_iwarp_fpdu *partial_fpdu;
1778 u32 idx;
1780 idx = cid - qed_cxt_get_proto_cid_start(p_hwfn, PROTOCOLID_IWARP);
1781 if (idx >= iwarp_info->max_num_partial_fpdus) {
1782 DP_ERR(p_hwfn, "Invalid cid %x max_num_partial_fpdus=%x\n", cid,
1783 iwarp_info->max_num_partial_fpdus);
1784 return NULL;
1787 partial_fpdu = &iwarp_info->partial_fpdus[idx];
1789 return partial_fpdu;
1792 enum qed_iwarp_mpa_pkt_type {
1793 QED_IWARP_MPA_PKT_PACKED,
1794 QED_IWARP_MPA_PKT_PARTIAL,
1795 QED_IWARP_MPA_PKT_UNALIGNED
1798 #define QED_IWARP_INVALID_FPDU_LENGTH 0xffff
1799 #define QED_IWARP_MPA_FPDU_LENGTH_SIZE (2)
1800 #define QED_IWARP_MPA_CRC32_DIGEST_SIZE (4)
1802 /* Pad to multiple of 4 */
1803 #define QED_IWARP_PDU_DATA_LEN_WITH_PAD(data_len) ALIGN(data_len, 4)
1804 #define QED_IWARP_FPDU_LEN_WITH_PAD(_mpa_len) \
1805 (QED_IWARP_PDU_DATA_LEN_WITH_PAD((_mpa_len) + \
1806 QED_IWARP_MPA_FPDU_LENGTH_SIZE) + \
1807 QED_IWARP_MPA_CRC32_DIGEST_SIZE)
1809 /* fpdu can be fragmented over maximum 3 bds: header, partial mpa, unaligned */
1810 #define QED_IWARP_MAX_BDS_PER_FPDU 3
1812 static const char * const pkt_type_str[] = {
1813 "QED_IWARP_MPA_PKT_PACKED",
1814 "QED_IWARP_MPA_PKT_PARTIAL",
1815 "QED_IWARP_MPA_PKT_UNALIGNED"
1818 static int
1819 qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn,
1820 struct qed_iwarp_fpdu *fpdu,
1821 struct qed_iwarp_ll2_buff *buf);
1823 static enum qed_iwarp_mpa_pkt_type
1824 qed_iwarp_mpa_classify(struct qed_hwfn *p_hwfn,
1825 struct qed_iwarp_fpdu *fpdu,
1826 u16 tcp_payload_len, u8 *mpa_data)
1828 enum qed_iwarp_mpa_pkt_type pkt_type;
1829 u16 mpa_len;
1831 if (fpdu->incomplete_bytes) {
1832 pkt_type = QED_IWARP_MPA_PKT_UNALIGNED;
1833 goto out;
1836 /* special case of one byte remaining...
1837 * lower byte will be read next packet
1839 if (tcp_payload_len == 1) {
1840 fpdu->fpdu_length = *mpa_data << BITS_PER_BYTE;
1841 pkt_type = QED_IWARP_MPA_PKT_PARTIAL;
1842 goto out;
1845 mpa_len = ntohs(*((u16 *)(mpa_data)));
1846 fpdu->fpdu_length = QED_IWARP_FPDU_LEN_WITH_PAD(mpa_len);
1848 if (fpdu->fpdu_length <= tcp_payload_len)
1849 pkt_type = QED_IWARP_MPA_PKT_PACKED;
1850 else
1851 pkt_type = QED_IWARP_MPA_PKT_PARTIAL;
1853 out:
1854 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1855 "MPA_ALIGN: %s: fpdu_length=0x%x tcp_payload_len:0x%x\n",
1856 pkt_type_str[pkt_type], fpdu->fpdu_length, tcp_payload_len);
1858 return pkt_type;
1861 static void
1862 qed_iwarp_init_fpdu(struct qed_iwarp_ll2_buff *buf,
1863 struct qed_iwarp_fpdu *fpdu,
1864 struct unaligned_opaque_data *pkt_data,
1865 u16 tcp_payload_size, u8 placement_offset)
1867 fpdu->mpa_buf = buf;
1868 fpdu->pkt_hdr = buf->data_phys_addr + placement_offset;
1869 fpdu->pkt_hdr_size = pkt_data->tcp_payload_offset;
1870 fpdu->mpa_frag = buf->data_phys_addr + pkt_data->first_mpa_offset;
1871 fpdu->mpa_frag_virt = (u8 *)(buf->data) + pkt_data->first_mpa_offset;
1873 if (tcp_payload_size == 1)
1874 fpdu->incomplete_bytes = QED_IWARP_INVALID_FPDU_LENGTH;
1875 else if (tcp_payload_size < fpdu->fpdu_length)
1876 fpdu->incomplete_bytes = fpdu->fpdu_length - tcp_payload_size;
1877 else
1878 fpdu->incomplete_bytes = 0; /* complete fpdu */
1880 fpdu->mpa_frag_len = fpdu->fpdu_length - fpdu->incomplete_bytes;
1883 static int
1884 qed_iwarp_cp_pkt(struct qed_hwfn *p_hwfn,
1885 struct qed_iwarp_fpdu *fpdu,
1886 struct unaligned_opaque_data *pkt_data,
1887 struct qed_iwarp_ll2_buff *buf, u16 tcp_payload_size)
1889 u8 *tmp_buf = p_hwfn->p_rdma_info->iwarp.mpa_intermediate_buf;
1890 int rc;
1892 /* need to copy the data from the partial packet stored in fpdu
1893 * to the new buf, for this we also need to move the data currently
1894 * placed on the buf. The assumption is that the buffer is big enough
1895 * since fpdu_length <= mss, we use an intermediate buffer since
1896 * we may need to copy the new data to an overlapping location
1898 if ((fpdu->mpa_frag_len + tcp_payload_size) > (u16)buf->buff_size) {
1899 DP_ERR(p_hwfn,
1900 "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",
1901 buf->buff_size, fpdu->mpa_frag_len,
1902 tcp_payload_size, fpdu->incomplete_bytes);
1903 return -EINVAL;
1906 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
1907 "MPA ALIGN Copying fpdu: [%p, %d] [%p, %d]\n",
1908 fpdu->mpa_frag_virt, fpdu->mpa_frag_len,
1909 (u8 *)(buf->data) + pkt_data->first_mpa_offset,
1910 tcp_payload_size);
1912 memcpy(tmp_buf, fpdu->mpa_frag_virt, fpdu->mpa_frag_len);
1913 memcpy(tmp_buf + fpdu->mpa_frag_len,
1914 (u8 *)(buf->data) + pkt_data->first_mpa_offset,
1915 tcp_payload_size);
1917 rc = qed_iwarp_recycle_pkt(p_hwfn, fpdu, fpdu->mpa_buf);
1918 if (rc)
1919 return rc;
1921 /* If we managed to post the buffer copy the data to the new buffer
1922 * o/w this will occur in the next round...
1924 memcpy((u8 *)(buf->data), tmp_buf,
1925 fpdu->mpa_frag_len + tcp_payload_size);
1927 fpdu->mpa_buf = buf;
1928 /* fpdu->pkt_hdr remains as is */
1929 /* fpdu->mpa_frag is overridden with new buf */
1930 fpdu->mpa_frag = buf->data_phys_addr;
1931 fpdu->mpa_frag_virt = buf->data;
1932 fpdu->mpa_frag_len += tcp_payload_size;
1934 fpdu->incomplete_bytes -= tcp_payload_size;
1936 DP_VERBOSE(p_hwfn,
1937 QED_MSG_RDMA,
1938 "MPA ALIGN: split fpdu buff_size = %d mpa_frag_len = %d, tcp_payload_size = %d, incomplete_bytes = %d\n",
1939 buf->buff_size, fpdu->mpa_frag_len, tcp_payload_size,
1940 fpdu->incomplete_bytes);
1942 return 0;
1945 static void
1946 qed_iwarp_update_fpdu_length(struct qed_hwfn *p_hwfn,
1947 struct qed_iwarp_fpdu *fpdu, u8 *mpa_data)
1949 u16 mpa_len;
1951 /* Update incomplete packets if needed */
1952 if (fpdu->incomplete_bytes == QED_IWARP_INVALID_FPDU_LENGTH) {
1953 /* Missing lower byte is now available */
1954 mpa_len = fpdu->fpdu_length | *mpa_data;
1955 fpdu->fpdu_length = QED_IWARP_FPDU_LEN_WITH_PAD(mpa_len);
1956 /* one byte of hdr */
1957 fpdu->mpa_frag_len = 1;
1958 fpdu->incomplete_bytes = fpdu->fpdu_length - 1;
1959 DP_VERBOSE(p_hwfn,
1960 QED_MSG_RDMA,
1961 "MPA_ALIGN: Partial header mpa_len=%x fpdu_length=%x incomplete_bytes=%x\n",
1962 mpa_len, fpdu->fpdu_length, fpdu->incomplete_bytes);
1966 #define QED_IWARP_IS_RIGHT_EDGE(_curr_pkt) \
1967 (GET_FIELD((_curr_pkt)->flags, \
1968 UNALIGNED_OPAQUE_DATA_PKT_REACHED_WIN_RIGHT_EDGE))
1970 /* This function is used to recycle a buffer using the ll2 drop option. It
1971 * uses the mechanism to ensure that all buffers posted to tx before this one
1972 * were completed. The buffer sent here will be sent as a cookie in the tx
1973 * completion function and can then be reposted to rx chain when done. The flow
1974 * that requires this is the flow where a FPDU splits over more than 3 tcp
1975 * segments. In this case the driver needs to re-post a rx buffer instead of
1976 * the one received, but driver can't simply repost a buffer it copied from
1977 * as there is a case where the buffer was originally a packed FPDU, and is
1978 * partially posted to FW. Driver needs to ensure FW is done with it.
1980 static int
1981 qed_iwarp_recycle_pkt(struct qed_hwfn *p_hwfn,
1982 struct qed_iwarp_fpdu *fpdu,
1983 struct qed_iwarp_ll2_buff *buf)
1985 struct qed_ll2_tx_pkt_info tx_pkt;
1986 u8 ll2_handle;
1987 int rc;
1989 memset(&tx_pkt, 0, sizeof(tx_pkt));
1990 tx_pkt.num_of_bds = 1;
1991 tx_pkt.tx_dest = QED_LL2_TX_DEST_DROP;
1992 tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2;
1993 tx_pkt.first_frag = fpdu->pkt_hdr;
1994 tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
1995 buf->piggy_buf = NULL;
1996 tx_pkt.cookie = buf;
1998 ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
2000 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
2001 if (rc)
2002 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2003 "Can't drop packet rc=%d\n", rc);
2005 DP_VERBOSE(p_hwfn,
2006 QED_MSG_RDMA,
2007 "MPA_ALIGN: send drop tx packet [%lx, 0x%x], buf=%p, rc=%d\n",
2008 (unsigned long int)tx_pkt.first_frag,
2009 tx_pkt.first_frag_len, buf, rc);
2011 return rc;
2014 static int
2015 qed_iwarp_win_right_edge(struct qed_hwfn *p_hwfn, struct qed_iwarp_fpdu *fpdu)
2017 struct qed_ll2_tx_pkt_info tx_pkt;
2018 u8 ll2_handle;
2019 int rc;
2021 memset(&tx_pkt, 0, sizeof(tx_pkt));
2022 tx_pkt.num_of_bds = 1;
2023 tx_pkt.tx_dest = QED_LL2_TX_DEST_LB;
2024 tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2;
2026 tx_pkt.first_frag = fpdu->pkt_hdr;
2027 tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
2028 tx_pkt.enable_ip_cksum = true;
2029 tx_pkt.enable_l4_cksum = true;
2030 tx_pkt.calc_ip_len = true;
2031 /* vlan overload with enum iwarp_ll2_tx_queues */
2032 tx_pkt.vlan = IWARP_LL2_ALIGNED_RIGHT_TRIMMED_TX_QUEUE;
2034 ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
2036 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
2037 if (rc)
2038 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2039 "Can't send right edge rc=%d\n", rc);
2040 DP_VERBOSE(p_hwfn,
2041 QED_MSG_RDMA,
2042 "MPA_ALIGN: Sent right edge FPDU num_bds=%d [%lx, 0x%x], rc=%d\n",
2043 tx_pkt.num_of_bds,
2044 (unsigned long int)tx_pkt.first_frag,
2045 tx_pkt.first_frag_len, rc);
2047 return rc;
2050 static int
2051 qed_iwarp_send_fpdu(struct qed_hwfn *p_hwfn,
2052 struct qed_iwarp_fpdu *fpdu,
2053 struct unaligned_opaque_data *curr_pkt,
2054 struct qed_iwarp_ll2_buff *buf,
2055 u16 tcp_payload_size, enum qed_iwarp_mpa_pkt_type pkt_type)
2057 struct qed_ll2_tx_pkt_info tx_pkt;
2058 u8 ll2_handle;
2059 int rc;
2061 memset(&tx_pkt, 0, sizeof(tx_pkt));
2063 /* An unaligned packet means it's split over two tcp segments. So the
2064 * complete packet requires 3 bds, one for the header, one for the
2065 * part of the fpdu of the first tcp segment, and the last fragment
2066 * will point to the remainder of the fpdu. A packed pdu, requires only
2067 * two bds, one for the header and one for the data.
2069 tx_pkt.num_of_bds = (pkt_type == QED_IWARP_MPA_PKT_UNALIGNED) ? 3 : 2;
2070 tx_pkt.tx_dest = QED_LL2_TX_DEST_LB;
2071 tx_pkt.l4_hdr_offset_w = fpdu->pkt_hdr_size >> 2; /* offset in words */
2073 /* Send the mpa_buf only with the last fpdu (in case of packed) */
2074 if (pkt_type == QED_IWARP_MPA_PKT_UNALIGNED ||
2075 tcp_payload_size <= fpdu->fpdu_length)
2076 tx_pkt.cookie = fpdu->mpa_buf;
2078 tx_pkt.first_frag = fpdu->pkt_hdr;
2079 tx_pkt.first_frag_len = fpdu->pkt_hdr_size;
2080 tx_pkt.enable_ip_cksum = true;
2081 tx_pkt.enable_l4_cksum = true;
2082 tx_pkt.calc_ip_len = true;
2083 /* vlan overload with enum iwarp_ll2_tx_queues */
2084 tx_pkt.vlan = IWARP_LL2_ALIGNED_TX_QUEUE;
2086 /* special case of unaligned packet and not packed, need to send
2087 * both buffers as cookie to release.
2089 if (tcp_payload_size == fpdu->incomplete_bytes)
2090 fpdu->mpa_buf->piggy_buf = buf;
2092 ll2_handle = p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle;
2094 /* Set first fragment to header */
2095 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_handle, &tx_pkt, true);
2096 if (rc)
2097 goto out;
2099 /* Set second fragment to first part of packet */
2100 rc = qed_ll2_set_fragment_of_tx_packet(p_hwfn, ll2_handle,
2101 fpdu->mpa_frag,
2102 fpdu->mpa_frag_len);
2103 if (rc)
2104 goto out;
2106 if (!fpdu->incomplete_bytes)
2107 goto out;
2109 /* Set third fragment to second part of the packet */
2110 rc = qed_ll2_set_fragment_of_tx_packet(p_hwfn,
2111 ll2_handle,
2112 buf->data_phys_addr +
2113 curr_pkt->first_mpa_offset,
2114 fpdu->incomplete_bytes);
2115 out:
2116 DP_VERBOSE(p_hwfn,
2117 QED_MSG_RDMA,
2118 "MPA_ALIGN: Sent FPDU num_bds=%d first_frag_len=%x, mpa_frag_len=0x%x, incomplete_bytes:0x%x rc=%d\n",
2119 tx_pkt.num_of_bds,
2120 tx_pkt.first_frag_len,
2121 fpdu->mpa_frag_len,
2122 fpdu->incomplete_bytes, rc);
2124 return rc;
2127 static void
2128 qed_iwarp_mpa_get_data(struct qed_hwfn *p_hwfn,
2129 struct unaligned_opaque_data *curr_pkt,
2130 u32 opaque_data0, u32 opaque_data1)
2132 u64 opaque_data;
2134 opaque_data = HILO_64(opaque_data1, opaque_data0);
2135 *curr_pkt = *((struct unaligned_opaque_data *)&opaque_data);
2137 curr_pkt->first_mpa_offset = curr_pkt->tcp_payload_offset +
2138 le16_to_cpu(curr_pkt->first_mpa_offset);
2139 curr_pkt->cid = le32_to_cpu(curr_pkt->cid);
2142 /* This function is called when an unaligned or incomplete MPA packet arrives
2143 * driver needs to align the packet, perhaps using previous data and send
2144 * it down to FW once it is aligned.
2146 static int
2147 qed_iwarp_process_mpa_pkt(struct qed_hwfn *p_hwfn,
2148 struct qed_iwarp_ll2_mpa_buf *mpa_buf)
2150 struct unaligned_opaque_data *curr_pkt = &mpa_buf->data;
2151 struct qed_iwarp_ll2_buff *buf = mpa_buf->ll2_buf;
2152 enum qed_iwarp_mpa_pkt_type pkt_type;
2153 struct qed_iwarp_fpdu *fpdu;
2154 int rc = -EINVAL;
2155 u8 *mpa_data;
2157 fpdu = qed_iwarp_get_curr_fpdu(p_hwfn, curr_pkt->cid & 0xffff);
2158 if (!fpdu) { /* something corrupt with cid, post rx back */
2159 DP_ERR(p_hwfn, "Invalid cid, drop and post back to rx cid=%x\n",
2160 curr_pkt->cid);
2161 goto err;
2164 do {
2165 mpa_data = ((u8 *)(buf->data) + curr_pkt->first_mpa_offset);
2167 pkt_type = qed_iwarp_mpa_classify(p_hwfn, fpdu,
2168 mpa_buf->tcp_payload_len,
2169 mpa_data);
2171 switch (pkt_type) {
2172 case QED_IWARP_MPA_PKT_PARTIAL:
2173 qed_iwarp_init_fpdu(buf, fpdu,
2174 curr_pkt,
2175 mpa_buf->tcp_payload_len,
2176 mpa_buf->placement_offset);
2178 if (!QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) {
2179 mpa_buf->tcp_payload_len = 0;
2180 break;
2183 rc = qed_iwarp_win_right_edge(p_hwfn, fpdu);
2185 if (rc) {
2186 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2187 "Can't send FPDU:reset rc=%d\n", rc);
2188 memset(fpdu, 0, sizeof(*fpdu));
2189 break;
2192 mpa_buf->tcp_payload_len = 0;
2193 break;
2194 case QED_IWARP_MPA_PKT_PACKED:
2195 qed_iwarp_init_fpdu(buf, fpdu,
2196 curr_pkt,
2197 mpa_buf->tcp_payload_len,
2198 mpa_buf->placement_offset);
2200 rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf,
2201 mpa_buf->tcp_payload_len,
2202 pkt_type);
2203 if (rc) {
2204 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2205 "Can't send FPDU:reset rc=%d\n", rc);
2206 memset(fpdu, 0, sizeof(*fpdu));
2207 break;
2210 mpa_buf->tcp_payload_len -= fpdu->fpdu_length;
2211 curr_pkt->first_mpa_offset += fpdu->fpdu_length;
2212 break;
2213 case QED_IWARP_MPA_PKT_UNALIGNED:
2214 qed_iwarp_update_fpdu_length(p_hwfn, fpdu, mpa_data);
2215 if (mpa_buf->tcp_payload_len < fpdu->incomplete_bytes) {
2216 /* special handling of fpdu split over more
2217 * than 2 segments
2219 if (QED_IWARP_IS_RIGHT_EDGE(curr_pkt)) {
2220 rc = qed_iwarp_win_right_edge(p_hwfn,
2221 fpdu);
2222 /* packet will be re-processed later */
2223 if (rc)
2224 return rc;
2227 rc = qed_iwarp_cp_pkt(p_hwfn, fpdu, curr_pkt,
2228 buf,
2229 mpa_buf->tcp_payload_len);
2230 if (rc) /* packet will be re-processed later */
2231 return rc;
2233 mpa_buf->tcp_payload_len = 0;
2234 break;
2237 rc = qed_iwarp_send_fpdu(p_hwfn, fpdu, curr_pkt, buf,
2238 mpa_buf->tcp_payload_len,
2239 pkt_type);
2240 if (rc) {
2241 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2242 "Can't send FPDU:delay rc=%d\n", rc);
2243 /* don't reset fpdu -> we need it for next
2244 * classify
2246 break;
2249 mpa_buf->tcp_payload_len -= fpdu->incomplete_bytes;
2250 curr_pkt->first_mpa_offset += fpdu->incomplete_bytes;
2251 /* The framed PDU was sent - no more incomplete bytes */
2252 fpdu->incomplete_bytes = 0;
2253 break;
2255 } while (mpa_buf->tcp_payload_len && !rc);
2257 return rc;
2259 err:
2260 qed_iwarp_ll2_post_rx(p_hwfn,
2261 buf,
2262 p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle);
2263 return rc;
2266 static void qed_iwarp_process_pending_pkts(struct qed_hwfn *p_hwfn)
2268 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2269 struct qed_iwarp_ll2_mpa_buf *mpa_buf = NULL;
2270 int rc;
2272 while (!list_empty(&iwarp_info->mpa_buf_pending_list)) {
2273 mpa_buf = list_first_entry(&iwarp_info->mpa_buf_pending_list,
2274 struct qed_iwarp_ll2_mpa_buf,
2275 list_entry);
2277 rc = qed_iwarp_process_mpa_pkt(p_hwfn, mpa_buf);
2279 /* busy means break and continue processing later, don't
2280 * remove the buf from the pending list.
2282 if (rc == -EBUSY)
2283 break;
2285 list_del(&mpa_buf->list_entry);
2286 list_add_tail(&mpa_buf->list_entry, &iwarp_info->mpa_buf_list);
2288 if (rc) { /* different error, don't continue */
2289 DP_NOTICE(p_hwfn, "process pkts failed rc=%d\n", rc);
2290 break;
2295 static void
2296 qed_iwarp_ll2_comp_mpa_pkt(void *cxt, struct qed_ll2_comp_rx_data *data)
2298 struct qed_iwarp_ll2_mpa_buf *mpa_buf;
2299 struct qed_iwarp_info *iwarp_info;
2300 struct qed_hwfn *p_hwfn = cxt;
2302 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2303 mpa_buf = list_first_entry(&iwarp_info->mpa_buf_list,
2304 struct qed_iwarp_ll2_mpa_buf, list_entry);
2305 if (!mpa_buf) {
2306 DP_ERR(p_hwfn, "No free mpa buf\n");
2307 goto err;
2310 list_del(&mpa_buf->list_entry);
2311 qed_iwarp_mpa_get_data(p_hwfn, &mpa_buf->data,
2312 data->opaque_data_0, data->opaque_data_1);
2314 DP_VERBOSE(p_hwfn,
2315 QED_MSG_RDMA,
2316 "LL2 MPA CompRx payload_len:0x%x\tfirst_mpa_offset:0x%x\ttcp_payload_offset:0x%x\tflags:0x%x\tcid:0x%x\n",
2317 data->length.packet_length, mpa_buf->data.first_mpa_offset,
2318 mpa_buf->data.tcp_payload_offset, mpa_buf->data.flags,
2319 mpa_buf->data.cid);
2321 mpa_buf->ll2_buf = data->cookie;
2322 mpa_buf->tcp_payload_len = data->length.packet_length -
2323 mpa_buf->data.first_mpa_offset;
2324 mpa_buf->data.first_mpa_offset += data->u.placement_offset;
2325 mpa_buf->placement_offset = data->u.placement_offset;
2327 list_add_tail(&mpa_buf->list_entry, &iwarp_info->mpa_buf_pending_list);
2329 qed_iwarp_process_pending_pkts(p_hwfn);
2330 return;
2331 err:
2332 qed_iwarp_ll2_post_rx(p_hwfn, data->cookie,
2333 iwarp_info->ll2_mpa_handle);
2336 static void
2337 qed_iwarp_ll2_comp_syn_pkt(void *cxt, struct qed_ll2_comp_rx_data *data)
2339 struct qed_iwarp_ll2_buff *buf = data->cookie;
2340 struct qed_iwarp_listener *listener;
2341 struct qed_ll2_tx_pkt_info tx_pkt;
2342 struct qed_iwarp_cm_info cm_info;
2343 struct qed_hwfn *p_hwfn = cxt;
2344 u8 remote_mac_addr[ETH_ALEN];
2345 u8 local_mac_addr[ETH_ALEN];
2346 struct qed_iwarp_ep *ep;
2347 int tcp_start_offset;
2348 u8 ll2_syn_handle;
2349 int payload_len;
2350 u32 hdr_size;
2351 int rc;
2353 memset(&cm_info, 0, sizeof(cm_info));
2354 ll2_syn_handle = p_hwfn->p_rdma_info->iwarp.ll2_syn_handle;
2356 /* Check if packet was received with errors... */
2357 if (data->err_flags) {
2358 DP_NOTICE(p_hwfn, "Error received on SYN packet: 0x%x\n",
2359 data->err_flags);
2360 goto err;
2363 if (GET_FIELD(data->parse_flags,
2364 PARSING_AND_ERR_FLAGS_L4CHKSMWASCALCULATED) &&
2365 GET_FIELD(data->parse_flags, PARSING_AND_ERR_FLAGS_L4CHKSMERROR)) {
2366 DP_NOTICE(p_hwfn, "Syn packet received with checksum error\n");
2367 goto err;
2370 rc = qed_iwarp_parse_rx_pkt(p_hwfn, &cm_info, (u8 *)(buf->data) +
2371 data->u.placement_offset, remote_mac_addr,
2372 local_mac_addr, &payload_len,
2373 &tcp_start_offset);
2374 if (rc)
2375 goto err;
2377 /* Check if there is a listener for this 4-tuple+vlan */
2378 listener = qed_iwarp_get_listener(p_hwfn, &cm_info);
2379 if (!listener) {
2380 DP_VERBOSE(p_hwfn,
2381 QED_MSG_RDMA,
2382 "SYN received on tuple not listened on parse_flags=%d packet len=%d\n",
2383 data->parse_flags, data->length.packet_length);
2385 memset(&tx_pkt, 0, sizeof(tx_pkt));
2386 tx_pkt.num_of_bds = 1;
2387 tx_pkt.l4_hdr_offset_w = (data->length.packet_length) >> 2;
2388 tx_pkt.tx_dest = QED_LL2_TX_DEST_LB;
2389 tx_pkt.first_frag = buf->data_phys_addr +
2390 data->u.placement_offset;
2391 tx_pkt.first_frag_len = data->length.packet_length;
2392 tx_pkt.cookie = buf;
2394 rc = qed_ll2_prepare_tx_packet(p_hwfn, ll2_syn_handle,
2395 &tx_pkt, true);
2397 if (rc) {
2398 DP_NOTICE(p_hwfn,
2399 "Can't post SYN back to chip rc=%d\n", rc);
2400 goto err;
2402 return;
2405 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "Received syn on listening port\n");
2406 /* There may be an open ep on this connection if this is a syn
2407 * retrasnmit... need to make sure there isn't...
2409 if (qed_iwarp_ep_exists(p_hwfn, &cm_info))
2410 goto err;
2412 ep = qed_iwarp_get_free_ep(p_hwfn);
2413 if (!ep)
2414 goto err;
2416 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2417 list_add_tail(&ep->list_entry, &p_hwfn->p_rdma_info->iwarp.ep_list);
2418 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2420 ether_addr_copy(ep->remote_mac_addr, remote_mac_addr);
2421 ether_addr_copy(ep->local_mac_addr, local_mac_addr);
2423 memcpy(&ep->cm_info, &cm_info, sizeof(ep->cm_info));
2425 hdr_size = ((cm_info.ip_version == QED_TCP_IPV4) ? 40 : 60);
2426 ep->mss = p_hwfn->p_rdma_info->iwarp.max_mtu - hdr_size;
2427 ep->mss = min_t(u16, QED_IWARP_MAX_FW_MSS, ep->mss);
2429 ep->event_cb = listener->event_cb;
2430 ep->cb_context = listener->cb_context;
2431 ep->connect_mode = TCP_CONNECT_PASSIVE;
2433 ep->syn = buf;
2434 ep->syn_ip_payload_length = (u16)payload_len;
2435 ep->syn_phy_addr = buf->data_phys_addr + data->u.placement_offset +
2436 tcp_start_offset;
2438 rc = qed_iwarp_tcp_offload(p_hwfn, ep);
2439 if (rc) {
2440 qed_iwarp_return_ep(p_hwfn, ep);
2441 goto err;
2444 return;
2445 err:
2446 qed_iwarp_ll2_post_rx(p_hwfn, buf, ll2_syn_handle);
2449 static void qed_iwarp_ll2_rel_rx_pkt(void *cxt, u8 connection_handle,
2450 void *cookie, dma_addr_t rx_buf_addr,
2451 bool b_last_packet)
2453 struct qed_iwarp_ll2_buff *buffer = cookie;
2454 struct qed_hwfn *p_hwfn = cxt;
2456 dma_free_coherent(&p_hwfn->cdev->pdev->dev, buffer->buff_size,
2457 buffer->data, buffer->data_phys_addr);
2458 kfree(buffer);
2461 static void qed_iwarp_ll2_comp_tx_pkt(void *cxt, u8 connection_handle,
2462 void *cookie, dma_addr_t first_frag_addr,
2463 bool b_last_fragment, bool b_last_packet)
2465 struct qed_iwarp_ll2_buff *buffer = cookie;
2466 struct qed_iwarp_ll2_buff *piggy;
2467 struct qed_hwfn *p_hwfn = cxt;
2469 if (!buffer) /* can happen in packed mpa unaligned... */
2470 return;
2472 /* this was originally an rx packet, post it back */
2473 piggy = buffer->piggy_buf;
2474 if (piggy) {
2475 buffer->piggy_buf = NULL;
2476 qed_iwarp_ll2_post_rx(p_hwfn, piggy, connection_handle);
2479 qed_iwarp_ll2_post_rx(p_hwfn, buffer, connection_handle);
2481 if (connection_handle == p_hwfn->p_rdma_info->iwarp.ll2_mpa_handle)
2482 qed_iwarp_process_pending_pkts(p_hwfn);
2484 return;
2487 static void qed_iwarp_ll2_rel_tx_pkt(void *cxt, u8 connection_handle,
2488 void *cookie, dma_addr_t first_frag_addr,
2489 bool b_last_fragment, bool b_last_packet)
2491 struct qed_iwarp_ll2_buff *buffer = cookie;
2492 struct qed_hwfn *p_hwfn = cxt;
2494 if (!buffer)
2495 return;
2497 if (buffer->piggy_buf) {
2498 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
2499 buffer->piggy_buf->buff_size,
2500 buffer->piggy_buf->data,
2501 buffer->piggy_buf->data_phys_addr);
2503 kfree(buffer->piggy_buf);
2506 dma_free_coherent(&p_hwfn->cdev->pdev->dev, buffer->buff_size,
2507 buffer->data, buffer->data_phys_addr);
2509 kfree(buffer);
2512 /* The only slowpath for iwarp ll2 is unalign flush. When this completion
2513 * is received, need to reset the FPDU.
2515 static void
2516 qed_iwarp_ll2_slowpath(void *cxt,
2517 u8 connection_handle,
2518 u32 opaque_data_0, u32 opaque_data_1)
2520 struct unaligned_opaque_data unalign_data;
2521 struct qed_hwfn *p_hwfn = cxt;
2522 struct qed_iwarp_fpdu *fpdu;
2524 qed_iwarp_mpa_get_data(p_hwfn, &unalign_data,
2525 opaque_data_0, opaque_data_1);
2527 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "(0x%x) Flush fpdu\n",
2528 unalign_data.cid);
2530 fpdu = qed_iwarp_get_curr_fpdu(p_hwfn, (u16)unalign_data.cid);
2531 if (fpdu)
2532 memset(fpdu, 0, sizeof(*fpdu));
2535 static int qed_iwarp_ll2_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
2537 struct qed_iwarp_info *iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2538 int rc = 0;
2540 if (iwarp_info->ll2_syn_handle != QED_IWARP_HANDLE_INVAL) {
2541 rc = qed_ll2_terminate_connection(p_hwfn,
2542 iwarp_info->ll2_syn_handle);
2543 if (rc)
2544 DP_INFO(p_hwfn, "Failed to terminate syn connection\n");
2546 qed_ll2_release_connection(p_hwfn, iwarp_info->ll2_syn_handle);
2547 iwarp_info->ll2_syn_handle = QED_IWARP_HANDLE_INVAL;
2550 if (iwarp_info->ll2_ooo_handle != QED_IWARP_HANDLE_INVAL) {
2551 rc = qed_ll2_terminate_connection(p_hwfn,
2552 iwarp_info->ll2_ooo_handle);
2553 if (rc)
2554 DP_INFO(p_hwfn, "Failed to terminate ooo connection\n");
2556 qed_ll2_release_connection(p_hwfn, iwarp_info->ll2_ooo_handle);
2557 iwarp_info->ll2_ooo_handle = QED_IWARP_HANDLE_INVAL;
2560 if (iwarp_info->ll2_mpa_handle != QED_IWARP_HANDLE_INVAL) {
2561 rc = qed_ll2_terminate_connection(p_hwfn,
2562 iwarp_info->ll2_mpa_handle);
2563 if (rc)
2564 DP_INFO(p_hwfn, "Failed to terminate mpa connection\n");
2566 qed_ll2_release_connection(p_hwfn, iwarp_info->ll2_mpa_handle);
2567 iwarp_info->ll2_mpa_handle = QED_IWARP_HANDLE_INVAL;
2570 qed_llh_remove_mac_filter(p_hwfn,
2571 p_ptt, p_hwfn->p_rdma_info->iwarp.mac_addr);
2572 return rc;
2575 static int
2576 qed_iwarp_ll2_alloc_buffers(struct qed_hwfn *p_hwfn,
2577 int num_rx_bufs, int buff_size, u8 ll2_handle)
2579 struct qed_iwarp_ll2_buff *buffer;
2580 int rc = 0;
2581 int i;
2583 for (i = 0; i < num_rx_bufs; i++) {
2584 buffer = kzalloc(sizeof(*buffer), GFP_KERNEL);
2585 if (!buffer) {
2586 rc = -ENOMEM;
2587 break;
2590 buffer->data = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2591 buff_size,
2592 &buffer->data_phys_addr,
2593 GFP_KERNEL);
2594 if (!buffer->data) {
2595 kfree(buffer);
2596 rc = -ENOMEM;
2597 break;
2600 buffer->buff_size = buff_size;
2601 rc = qed_iwarp_ll2_post_rx(p_hwfn, buffer, ll2_handle);
2602 if (rc)
2603 /* buffers will be deallocated by qed_ll2 */
2604 break;
2606 return rc;
2609 #define QED_IWARP_MAX_BUF_SIZE(mtu) \
2610 ALIGN((mtu) + ETH_HLEN + 2 * VLAN_HLEN + 2 + ETH_CACHE_LINE_SIZE, \
2611 ETH_CACHE_LINE_SIZE)
2613 static int
2614 qed_iwarp_ll2_start(struct qed_hwfn *p_hwfn,
2615 struct qed_rdma_start_in_params *params,
2616 struct qed_ptt *p_ptt)
2618 struct qed_iwarp_info *iwarp_info;
2619 struct qed_ll2_acquire_data data;
2620 struct qed_ll2_cbs cbs;
2621 u32 buff_size;
2622 u16 n_ooo_bufs;
2623 int rc = 0;
2624 int i;
2626 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2627 iwarp_info->ll2_syn_handle = QED_IWARP_HANDLE_INVAL;
2628 iwarp_info->ll2_ooo_handle = QED_IWARP_HANDLE_INVAL;
2629 iwarp_info->ll2_mpa_handle = QED_IWARP_HANDLE_INVAL;
2631 iwarp_info->max_mtu = params->max_mtu;
2633 ether_addr_copy(p_hwfn->p_rdma_info->iwarp.mac_addr, params->mac_addr);
2635 rc = qed_llh_add_mac_filter(p_hwfn, p_ptt, params->mac_addr);
2636 if (rc)
2637 return rc;
2639 /* Start SYN connection */
2640 cbs.rx_comp_cb = qed_iwarp_ll2_comp_syn_pkt;
2641 cbs.rx_release_cb = qed_iwarp_ll2_rel_rx_pkt;
2642 cbs.tx_comp_cb = qed_iwarp_ll2_comp_tx_pkt;
2643 cbs.tx_release_cb = qed_iwarp_ll2_rel_tx_pkt;
2644 cbs.slowpath_cb = NULL;
2645 cbs.cookie = p_hwfn;
2647 memset(&data, 0, sizeof(data));
2648 data.input.conn_type = QED_LL2_TYPE_IWARP;
2649 data.input.mtu = params->max_mtu;
2650 data.input.rx_num_desc = QED_IWARP_LL2_SYN_RX_SIZE;
2651 data.input.tx_num_desc = QED_IWARP_LL2_SYN_TX_SIZE;
2652 data.input.tx_max_bds_per_packet = 1; /* will never be fragmented */
2653 data.input.tx_tc = PKT_LB_TC;
2654 data.input.tx_dest = QED_LL2_TX_DEST_LB;
2655 data.p_connection_handle = &iwarp_info->ll2_syn_handle;
2656 data.cbs = &cbs;
2658 rc = qed_ll2_acquire_connection(p_hwfn, &data);
2659 if (rc) {
2660 DP_NOTICE(p_hwfn, "Failed to acquire LL2 connection\n");
2661 qed_llh_remove_mac_filter(p_hwfn, p_ptt, params->mac_addr);
2662 return rc;
2665 rc = qed_ll2_establish_connection(p_hwfn, iwarp_info->ll2_syn_handle);
2666 if (rc) {
2667 DP_NOTICE(p_hwfn, "Failed to establish LL2 connection\n");
2668 goto err;
2671 buff_size = QED_IWARP_MAX_BUF_SIZE(params->max_mtu);
2672 rc = qed_iwarp_ll2_alloc_buffers(p_hwfn,
2673 QED_IWARP_LL2_SYN_RX_SIZE,
2674 buff_size,
2675 iwarp_info->ll2_syn_handle);
2676 if (rc)
2677 goto err;
2679 /* Start OOO connection */
2680 data.input.conn_type = QED_LL2_TYPE_OOO;
2681 data.input.mtu = params->max_mtu;
2683 n_ooo_bufs = (QED_IWARP_MAX_OOO * QED_IWARP_RCV_WND_SIZE_DEF) /
2684 iwarp_info->max_mtu;
2685 n_ooo_bufs = min_t(u32, n_ooo_bufs, QED_IWARP_LL2_OOO_MAX_RX_SIZE);
2687 data.input.rx_num_desc = n_ooo_bufs;
2688 data.input.rx_num_ooo_buffers = n_ooo_bufs;
2690 data.input.tx_max_bds_per_packet = 1; /* will never be fragmented */
2691 data.input.tx_num_desc = QED_IWARP_LL2_OOO_DEF_TX_SIZE;
2692 data.p_connection_handle = &iwarp_info->ll2_ooo_handle;
2694 rc = qed_ll2_acquire_connection(p_hwfn, &data);
2695 if (rc)
2696 goto err;
2698 rc = qed_ll2_establish_connection(p_hwfn, iwarp_info->ll2_ooo_handle);
2699 if (rc)
2700 goto err;
2702 /* Start Unaligned MPA connection */
2703 cbs.rx_comp_cb = qed_iwarp_ll2_comp_mpa_pkt;
2704 cbs.slowpath_cb = qed_iwarp_ll2_slowpath;
2706 memset(&data, 0, sizeof(data));
2707 data.input.conn_type = QED_LL2_TYPE_IWARP;
2708 data.input.mtu = params->max_mtu;
2709 /* FW requires that once a packet arrives OOO, it must have at
2710 * least 2 rx buffers available on the unaligned connection
2711 * for handling the case that it is a partial fpdu.
2713 data.input.rx_num_desc = n_ooo_bufs * 2;
2714 data.input.tx_num_desc = data.input.rx_num_desc;
2715 data.input.tx_max_bds_per_packet = QED_IWARP_MAX_BDS_PER_FPDU;
2716 data.input.tx_tc = PKT_LB_TC;
2717 data.input.tx_dest = QED_LL2_TX_DEST_LB;
2718 data.p_connection_handle = &iwarp_info->ll2_mpa_handle;
2719 data.input.secondary_queue = true;
2720 data.cbs = &cbs;
2722 rc = qed_ll2_acquire_connection(p_hwfn, &data);
2723 if (rc)
2724 goto err;
2726 rc = qed_ll2_establish_connection(p_hwfn, iwarp_info->ll2_mpa_handle);
2727 if (rc)
2728 goto err;
2730 rc = qed_iwarp_ll2_alloc_buffers(p_hwfn,
2731 data.input.rx_num_desc,
2732 buff_size,
2733 iwarp_info->ll2_mpa_handle);
2734 if (rc)
2735 goto err;
2737 iwarp_info->partial_fpdus = kcalloc((u16)p_hwfn->p_rdma_info->num_qps,
2738 sizeof(*iwarp_info->partial_fpdus),
2739 GFP_KERNEL);
2740 if (!iwarp_info->partial_fpdus)
2741 goto err;
2743 iwarp_info->max_num_partial_fpdus = (u16)p_hwfn->p_rdma_info->num_qps;
2745 iwarp_info->mpa_intermediate_buf = kzalloc(buff_size, GFP_KERNEL);
2746 if (!iwarp_info->mpa_intermediate_buf)
2747 goto err;
2749 /* The mpa_bufs array serves for pending RX packets received on the
2750 * mpa ll2 that don't have place on the tx ring and require later
2751 * processing. We can't fail on allocation of such a struct therefore
2752 * we allocate enough to take care of all rx packets
2754 iwarp_info->mpa_bufs = kcalloc(data.input.rx_num_desc,
2755 sizeof(*iwarp_info->mpa_bufs),
2756 GFP_KERNEL);
2757 if (!iwarp_info->mpa_bufs)
2758 goto err;
2760 INIT_LIST_HEAD(&iwarp_info->mpa_buf_pending_list);
2761 INIT_LIST_HEAD(&iwarp_info->mpa_buf_list);
2762 for (i = 0; i < data.input.rx_num_desc; i++)
2763 list_add_tail(&iwarp_info->mpa_bufs[i].list_entry,
2764 &iwarp_info->mpa_buf_list);
2765 return rc;
2766 err:
2767 qed_iwarp_ll2_stop(p_hwfn, p_ptt);
2769 return rc;
2772 int qed_iwarp_setup(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt,
2773 struct qed_rdma_start_in_params *params)
2775 struct qed_iwarp_info *iwarp_info;
2776 u32 rcv_wnd_size;
2778 iwarp_info = &p_hwfn->p_rdma_info->iwarp;
2780 iwarp_info->tcp_flags = QED_IWARP_TS_EN;
2781 rcv_wnd_size = QED_IWARP_RCV_WND_SIZE_DEF;
2783 /* value 0 is used for ilog2(QED_IWARP_RCV_WND_SIZE_MIN) */
2784 iwarp_info->rcv_wnd_scale = ilog2(rcv_wnd_size) -
2785 ilog2(QED_IWARP_RCV_WND_SIZE_MIN);
2786 iwarp_info->rcv_wnd_size = rcv_wnd_size >> iwarp_info->rcv_wnd_scale;
2787 iwarp_info->crc_needed = QED_IWARP_PARAM_CRC_NEEDED;
2788 iwarp_info->mpa_rev = MPA_NEGOTIATION_TYPE_ENHANCED;
2790 iwarp_info->peer2peer = QED_IWARP_PARAM_P2P;
2792 iwarp_info->rtr_type = MPA_RTR_TYPE_ZERO_SEND |
2793 MPA_RTR_TYPE_ZERO_WRITE |
2794 MPA_RTR_TYPE_ZERO_READ;
2796 spin_lock_init(&p_hwfn->p_rdma_info->iwarp.qp_lock);
2797 INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.ep_list);
2798 INIT_LIST_HEAD(&p_hwfn->p_rdma_info->iwarp.listen_list);
2800 qed_spq_register_async_cb(p_hwfn, PROTOCOLID_IWARP,
2801 qed_iwarp_async_event);
2802 qed_ooo_setup(p_hwfn);
2804 return qed_iwarp_ll2_start(p_hwfn, params, p_ptt);
2807 int qed_iwarp_stop(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
2809 int rc;
2811 qed_iwarp_free_prealloc_ep(p_hwfn);
2812 rc = qed_iwarp_wait_for_all_cids(p_hwfn);
2813 if (rc)
2814 return rc;
2816 qed_spq_unregister_async_cb(p_hwfn, PROTOCOLID_IWARP);
2818 return qed_iwarp_ll2_stop(p_hwfn, p_ptt);
2821 static void qed_iwarp_qp_in_error(struct qed_hwfn *p_hwfn,
2822 struct qed_iwarp_ep *ep,
2823 u8 fw_return_code)
2825 struct qed_iwarp_cm_event_params params;
2827 qed_iwarp_modify_qp(p_hwfn, ep->qp, QED_IWARP_QP_STATE_ERROR, true);
2829 params.event = QED_IWARP_EVENT_CLOSE;
2830 params.ep_context = ep;
2831 params.cm_info = &ep->cm_info;
2832 params.status = (fw_return_code == IWARP_QP_IN_ERROR_GOOD_CLOSE) ?
2833 0 : -ECONNRESET;
2835 /* paired with READ_ONCE in destroy_qp */
2836 smp_store_release(&ep->state, QED_IWARP_EP_CLOSED);
2838 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2839 list_del(&ep->list_entry);
2840 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2842 ep->event_cb(ep->cb_context, &params);
2845 static void qed_iwarp_exception_received(struct qed_hwfn *p_hwfn,
2846 struct qed_iwarp_ep *ep,
2847 int fw_ret_code)
2849 struct qed_iwarp_cm_event_params params;
2850 bool event_cb = false;
2852 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "EP(0x%x) fw_ret_code=%d\n",
2853 ep->cid, fw_ret_code);
2855 switch (fw_ret_code) {
2856 case IWARP_EXCEPTION_DETECTED_LLP_CLOSED:
2857 params.status = 0;
2858 params.event = QED_IWARP_EVENT_DISCONNECT;
2859 event_cb = true;
2860 break;
2861 case IWARP_EXCEPTION_DETECTED_LLP_RESET:
2862 params.status = -ECONNRESET;
2863 params.event = QED_IWARP_EVENT_DISCONNECT;
2864 event_cb = true;
2865 break;
2866 case IWARP_EXCEPTION_DETECTED_RQ_EMPTY:
2867 params.event = QED_IWARP_EVENT_RQ_EMPTY;
2868 event_cb = true;
2869 break;
2870 case IWARP_EXCEPTION_DETECTED_IRQ_FULL:
2871 params.event = QED_IWARP_EVENT_IRQ_FULL;
2872 event_cb = true;
2873 break;
2874 case IWARP_EXCEPTION_DETECTED_LLP_TIMEOUT:
2875 params.event = QED_IWARP_EVENT_LLP_TIMEOUT;
2876 event_cb = true;
2877 break;
2878 case IWARP_EXCEPTION_DETECTED_REMOTE_PROTECTION_ERROR:
2879 params.event = QED_IWARP_EVENT_REMOTE_PROTECTION_ERROR;
2880 event_cb = true;
2881 break;
2882 case IWARP_EXCEPTION_DETECTED_CQ_OVERFLOW:
2883 params.event = QED_IWARP_EVENT_CQ_OVERFLOW;
2884 event_cb = true;
2885 break;
2886 case IWARP_EXCEPTION_DETECTED_LOCAL_CATASTROPHIC:
2887 params.event = QED_IWARP_EVENT_QP_CATASTROPHIC;
2888 event_cb = true;
2889 break;
2890 case IWARP_EXCEPTION_DETECTED_LOCAL_ACCESS_ERROR:
2891 params.event = QED_IWARP_EVENT_LOCAL_ACCESS_ERROR;
2892 event_cb = true;
2893 break;
2894 case IWARP_EXCEPTION_DETECTED_REMOTE_OPERATION_ERROR:
2895 params.event = QED_IWARP_EVENT_REMOTE_OPERATION_ERROR;
2896 event_cb = true;
2897 break;
2898 case IWARP_EXCEPTION_DETECTED_TERMINATE_RECEIVED:
2899 params.event = QED_IWARP_EVENT_TERMINATE_RECEIVED;
2900 event_cb = true;
2901 break;
2902 default:
2903 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2904 "Unhandled exception received...fw_ret_code=%d\n",
2905 fw_ret_code);
2906 break;
2909 if (event_cb) {
2910 params.ep_context = ep;
2911 params.cm_info = &ep->cm_info;
2912 ep->event_cb(ep->cb_context, &params);
2916 static void
2917 qed_iwarp_tcp_connect_unsuccessful(struct qed_hwfn *p_hwfn,
2918 struct qed_iwarp_ep *ep, u8 fw_return_code)
2920 struct qed_iwarp_cm_event_params params;
2922 memset(&params, 0, sizeof(params));
2923 params.event = QED_IWARP_EVENT_ACTIVE_COMPLETE;
2924 params.ep_context = ep;
2925 params.cm_info = &ep->cm_info;
2926 /* paired with READ_ONCE in destroy_qp */
2927 smp_store_release(&ep->state, QED_IWARP_EP_CLOSED);
2929 switch (fw_return_code) {
2930 case IWARP_CONN_ERROR_TCP_CONNECT_INVALID_PACKET:
2931 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2932 "%s(0x%x) TCP connect got invalid packet\n",
2933 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2934 params.status = -ECONNRESET;
2935 break;
2936 case IWARP_CONN_ERROR_TCP_CONNECTION_RST:
2937 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
2938 "%s(0x%x) TCP Connection Reset\n",
2939 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2940 params.status = -ECONNRESET;
2941 break;
2942 case IWARP_CONN_ERROR_TCP_CONNECT_TIMEOUT:
2943 DP_NOTICE(p_hwfn, "%s(0x%x) TCP timeout\n",
2944 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2945 params.status = -EBUSY;
2946 break;
2947 case IWARP_CONN_ERROR_MPA_NOT_SUPPORTED_VER:
2948 DP_NOTICE(p_hwfn, "%s(0x%x) MPA not supported VER\n",
2949 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2950 params.status = -ECONNREFUSED;
2951 break;
2952 case IWARP_CONN_ERROR_MPA_INVALID_PACKET:
2953 DP_NOTICE(p_hwfn, "%s(0x%x) MPA Invalid Packet\n",
2954 QED_IWARP_CONNECT_MODE_STRING(ep), ep->tcp_cid);
2955 params.status = -ECONNRESET;
2956 break;
2957 default:
2958 DP_ERR(p_hwfn,
2959 "%s(0x%x) Unexpected return code tcp connect: %d\n",
2960 QED_IWARP_CONNECT_MODE_STRING(ep),
2961 ep->tcp_cid, fw_return_code);
2962 params.status = -ECONNRESET;
2963 break;
2966 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
2967 ep->tcp_cid = QED_IWARP_INVALID_TCP_CID;
2968 qed_iwarp_return_ep(p_hwfn, ep);
2969 } else {
2970 ep->event_cb(ep->cb_context, &params);
2971 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2972 list_del(&ep->list_entry);
2973 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
2977 static void
2978 qed_iwarp_connect_complete(struct qed_hwfn *p_hwfn,
2979 struct qed_iwarp_ep *ep, u8 fw_return_code)
2981 u8 ll2_syn_handle = p_hwfn->p_rdma_info->iwarp.ll2_syn_handle;
2983 if (ep->connect_mode == TCP_CONNECT_PASSIVE) {
2984 /* Done with the SYN packet, post back to ll2 rx */
2985 qed_iwarp_ll2_post_rx(p_hwfn, ep->syn, ll2_syn_handle);
2987 ep->syn = NULL;
2989 /* If connect failed - upper layer doesn't know about it */
2990 if (fw_return_code == RDMA_RETURN_OK)
2991 qed_iwarp_mpa_received(p_hwfn, ep);
2992 else
2993 qed_iwarp_tcp_connect_unsuccessful(p_hwfn, ep,
2994 fw_return_code);
2995 } else {
2996 if (fw_return_code == RDMA_RETURN_OK)
2997 qed_iwarp_mpa_offload(p_hwfn, ep);
2998 else
2999 qed_iwarp_tcp_connect_unsuccessful(p_hwfn, ep,
3000 fw_return_code);
3004 static inline bool
3005 qed_iwarp_check_ep_ok(struct qed_hwfn *p_hwfn, struct qed_iwarp_ep *ep)
3007 if (!ep || (ep->sig != QED_EP_SIG)) {
3008 DP_ERR(p_hwfn, "ERROR ON ASYNC ep=%p\n", ep);
3009 return false;
3012 return true;
3015 static int qed_iwarp_async_event(struct qed_hwfn *p_hwfn,
3016 u8 fw_event_code, u16 echo,
3017 union event_ring_data *data,
3018 u8 fw_return_code)
3020 struct qed_rdma_events events = p_hwfn->p_rdma_info->events;
3021 struct regpair *fw_handle = &data->rdma_data.async_handle;
3022 struct qed_iwarp_ep *ep = NULL;
3023 u16 srq_offset;
3024 u16 srq_id;
3025 u16 cid;
3027 ep = (struct qed_iwarp_ep *)(uintptr_t)HILO_64(fw_handle->hi,
3028 fw_handle->lo);
3030 switch (fw_event_code) {
3031 case IWARP_EVENT_TYPE_ASYNC_CONNECT_COMPLETE:
3032 /* Async completion after TCP 3-way handshake */
3033 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3034 return -EINVAL;
3035 DP_VERBOSE(p_hwfn,
3036 QED_MSG_RDMA,
3037 "EP(0x%x) IWARP_EVENT_TYPE_ASYNC_CONNECT_COMPLETE fw_ret_code=%d\n",
3038 ep->tcp_cid, fw_return_code);
3039 qed_iwarp_connect_complete(p_hwfn, ep, fw_return_code);
3040 break;
3041 case IWARP_EVENT_TYPE_ASYNC_EXCEPTION_DETECTED:
3042 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3043 return -EINVAL;
3044 DP_VERBOSE(p_hwfn,
3045 QED_MSG_RDMA,
3046 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_EXCEPTION_DETECTED fw_ret_code=%d\n",
3047 ep->cid, fw_return_code);
3048 qed_iwarp_exception_received(p_hwfn, ep, fw_return_code);
3049 break;
3050 case IWARP_EVENT_TYPE_ASYNC_QP_IN_ERROR_STATE:
3051 /* Async completion for Close Connection ramrod */
3052 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3053 return -EINVAL;
3054 DP_VERBOSE(p_hwfn,
3055 QED_MSG_RDMA,
3056 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_QP_IN_ERROR_STATE fw_ret_code=%d\n",
3057 ep->cid, fw_return_code);
3058 qed_iwarp_qp_in_error(p_hwfn, ep, fw_return_code);
3059 break;
3060 case IWARP_EVENT_TYPE_ASYNC_ENHANCED_MPA_REPLY_ARRIVED:
3061 /* Async event for active side only */
3062 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3063 return -EINVAL;
3064 DP_VERBOSE(p_hwfn,
3065 QED_MSG_RDMA,
3066 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_MPA_REPLY_ARRIVED fw_ret_code=%d\n",
3067 ep->cid, fw_return_code);
3068 qed_iwarp_mpa_reply_arrived(p_hwfn, ep);
3069 break;
3070 case IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE:
3071 if (!qed_iwarp_check_ep_ok(p_hwfn, ep))
3072 return -EINVAL;
3073 DP_VERBOSE(p_hwfn,
3074 QED_MSG_RDMA,
3075 "QP(0x%x) IWARP_EVENT_TYPE_ASYNC_MPA_HANDSHAKE_COMPLETE fw_ret_code=%d\n",
3076 ep->cid, fw_return_code);
3077 qed_iwarp_mpa_complete(p_hwfn, ep, fw_return_code);
3078 break;
3079 case IWARP_EVENT_TYPE_ASYNC_CID_CLEANED:
3080 cid = (u16)le32_to_cpu(fw_handle->lo);
3081 DP_VERBOSE(p_hwfn, QED_MSG_RDMA,
3082 "(0x%x)IWARP_EVENT_TYPE_ASYNC_CID_CLEANED\n", cid);
3083 qed_iwarp_cid_cleaned(p_hwfn, cid);
3085 break;
3086 case IWARP_EVENT_TYPE_ASYNC_SRQ_EMPTY:
3087 DP_NOTICE(p_hwfn, "IWARP_EVENT_TYPE_ASYNC_SRQ_EMPTY\n");
3088 srq_offset = p_hwfn->p_rdma_info->srq_id_offset;
3089 /* FW assigns value that is no greater than u16 */
3090 srq_id = ((u16)le32_to_cpu(fw_handle->lo)) - srq_offset;
3091 events.affiliated_event(events.context,
3092 QED_IWARP_EVENT_SRQ_EMPTY,
3093 &srq_id);
3094 break;
3095 case IWARP_EVENT_TYPE_ASYNC_SRQ_LIMIT:
3096 DP_NOTICE(p_hwfn, "IWARP_EVENT_TYPE_ASYNC_SRQ_LIMIT\n");
3097 srq_offset = p_hwfn->p_rdma_info->srq_id_offset;
3098 /* FW assigns value that is no greater than u16 */
3099 srq_id = ((u16)le32_to_cpu(fw_handle->lo)) - srq_offset;
3100 events.affiliated_event(events.context,
3101 QED_IWARP_EVENT_SRQ_LIMIT,
3102 &srq_id);
3103 break;
3104 case IWARP_EVENT_TYPE_ASYNC_CQ_OVERFLOW:
3105 DP_NOTICE(p_hwfn, "IWARP_EVENT_TYPE_ASYNC_CQ_OVERFLOW\n");
3107 p_hwfn->p_rdma_info->events.affiliated_event(
3108 p_hwfn->p_rdma_info->events.context,
3109 QED_IWARP_EVENT_CQ_OVERFLOW,
3110 (void *)fw_handle);
3111 break;
3112 default:
3113 DP_ERR(p_hwfn, "Received unexpected async iwarp event %d\n",
3114 fw_event_code);
3115 return -EINVAL;
3117 return 0;
3121 qed_iwarp_create_listen(void *rdma_cxt,
3122 struct qed_iwarp_listen_in *iparams,
3123 struct qed_iwarp_listen_out *oparams)
3125 struct qed_hwfn *p_hwfn = rdma_cxt;
3126 struct qed_iwarp_listener *listener;
3128 listener = kzalloc(sizeof(*listener), GFP_KERNEL);
3129 if (!listener)
3130 return -ENOMEM;
3132 listener->ip_version = iparams->ip_version;
3133 memcpy(listener->ip_addr, iparams->ip_addr, sizeof(listener->ip_addr));
3134 listener->port = iparams->port;
3135 listener->vlan = iparams->vlan;
3137 listener->event_cb = iparams->event_cb;
3138 listener->cb_context = iparams->cb_context;
3139 listener->max_backlog = iparams->max_backlog;
3140 oparams->handle = listener;
3142 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3143 list_add_tail(&listener->list_entry,
3144 &p_hwfn->p_rdma_info->iwarp.listen_list);
3145 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3147 DP_VERBOSE(p_hwfn,
3148 QED_MSG_RDMA,
3149 "callback=%p handle=%p ip=%x:%x:%x:%x port=0x%x vlan=0x%x\n",
3150 listener->event_cb,
3151 listener,
3152 listener->ip_addr[0],
3153 listener->ip_addr[1],
3154 listener->ip_addr[2],
3155 listener->ip_addr[3], listener->port, listener->vlan);
3157 return 0;
3160 int qed_iwarp_destroy_listen(void *rdma_cxt, void *handle)
3162 struct qed_iwarp_listener *listener = handle;
3163 struct qed_hwfn *p_hwfn = rdma_cxt;
3165 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "handle=%p\n", handle);
3167 spin_lock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3168 list_del(&listener->list_entry);
3169 spin_unlock_bh(&p_hwfn->p_rdma_info->iwarp.iw_lock);
3171 kfree(listener);
3173 return 0;
3176 int qed_iwarp_send_rtr(void *rdma_cxt, struct qed_iwarp_send_rtr_in *iparams)
3178 struct qed_hwfn *p_hwfn = rdma_cxt;
3179 struct qed_sp_init_data init_data;
3180 struct qed_spq_entry *p_ent;
3181 struct qed_iwarp_ep *ep;
3182 struct qed_rdma_qp *qp;
3183 int rc;
3185 ep = iparams->ep_context;
3186 if (!ep) {
3187 DP_ERR(p_hwfn, "Ep Context receive in send_rtr is NULL\n");
3188 return -EINVAL;
3191 qp = ep->qp;
3193 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "QP(0x%x) EP(0x%x)\n",
3194 qp->icid, ep->tcp_cid);
3196 memset(&init_data, 0, sizeof(init_data));
3197 init_data.cid = qp->icid;
3198 init_data.opaque_fid = p_hwfn->hw_info.opaque_fid;
3199 init_data.comp_mode = QED_SPQ_MODE_CB;
3201 rc = qed_sp_init_request(p_hwfn, &p_ent,
3202 IWARP_RAMROD_CMD_ID_MPA_OFFLOAD_SEND_RTR,
3203 PROTOCOLID_IWARP, &init_data);
3205 if (rc)
3206 return rc;
3208 rc = qed_spq_post(p_hwfn, p_ent, NULL);
3210 DP_VERBOSE(p_hwfn, QED_MSG_RDMA, "rc = 0x%x\n", rc);
3212 return rc;
3215 void
3216 qed_iwarp_query_qp(struct qed_rdma_qp *qp,
3217 struct qed_rdma_query_qp_out_params *out_params)
3219 out_params->state = qed_iwarp2roce_state(qp->iwarp_state);