1 /* This file is part of the Emulex RoCE Device Driver for
2  * RoCE (RDMA over Converged Ethernet) adapters.
3  * Copyright (C) 2012-2015 Emulex. All rights reserved.
4  * EMULEX and SLI are trademarks of Emulex.
5  * www.emulex.com
6  *
7  * This software is available to you under a choice of one of two licenses.
8  * You may choose to be licensed under the terms of the GNU General Public
9  * License (GPL) Version 2, available from the file COPYING in the main
10  * directory of this source tree, or the BSD license below:
11  *
12  * Redistribution and use in source and binary forms, with or without
13  * modification, are permitted provided that the following conditions
14  * are met:
15  *
16  * - Redistributions of source code must retain the above copyright notice,
17  *   this list of conditions and the following disclaimer.
18  *
19  * - Redistributions in binary form must reproduce the above copyright
20  *   notice, this list of conditions and the following disclaimer in
21  *   the documentation and/or other materials provided with the distribution.
22  *
23  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
24  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,THE
25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
27  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
28  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
29  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
30  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
31  * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
32  * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF
33  * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34  *
35  * Contact Information:
36  * linux-drivers@emulex.com
37  *
38  * Emulex
39  * 3333 Susan Street
40  * Costa Mesa, CA 92626
41  */
42 
43 #ifndef __OCRDMA_H__
44 #define __OCRDMA_H__
45 
46 #include <linux/mutex.h>
47 #include <linux/list.h>
48 #include <linux/spinlock.h>
49 #include <linux/pci.h>
50 
51 #include <rdma/ib_verbs.h>
52 #include <rdma/ib_user_verbs.h>
53 #include <rdma/ib_addr.h>
54 
55 #include <be_roce.h>
56 #include "ocrdma_sli.h"
57 
58 #define OCRDMA_ROCE_DRV_VERSION "11.0.0.0"
59 
60 #define OCRDMA_ROCE_DRV_DESC "Emulex OneConnect RoCE Driver"
61 #define OCRDMA_NODE_DESC "Emulex OneConnect RoCE HCA"
62 
63 #define OC_NAME_SH	OCRDMA_NODE_DESC "(Skyhawk)"
64 #define OC_NAME_UNKNOWN OCRDMA_NODE_DESC "(Unknown)"
65 
66 #define OC_SKH_DEVICE_PF 0x720
67 #define OC_SKH_DEVICE_VF 0x728
68 #define OCRDMA_MAX_AH 512
69 
70 #define OCRDMA_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
71 
72 #define convert_to_64bit(lo, hi) ((u64)hi << 32 | (u64)lo)
73 #define EQ_INTR_PER_SEC_THRSH_HI 150000
74 #define EQ_INTR_PER_SEC_THRSH_LOW 100000
75 #define EQ_AIC_MAX_EQD 20
76 #define EQ_AIC_MIN_EQD 0
77 
78 void ocrdma_eqd_set_task(struct work_struct *work);
79 
80 struct ocrdma_dev_attr {
81 	u8 fw_ver[32];
82 	u32 vendor_id;
83 	u32 device_id;
84 	u16 max_pd;
85 	u16 max_dpp_pds;
86 	u16 max_cq;
87 	u16 max_cqe;
88 	u16 max_qp;
89 	u16 max_wqe;
90 	u16 max_rqe;
91 	u16 max_srq;
92 	u32 max_inline_data;
93 	int max_send_sge;
94 	int max_recv_sge;
95 	int max_srq_sge;
96 	int max_rdma_sge;
97 	int max_mr;
98 	u64 max_mr_size;
99 	u32 max_num_mr_pbl;
100 	int max_mw;
101 	int max_map_per_fmr;
102 	int max_pages_per_frmr;
103 	u16 max_ord_per_qp;
104 	u16 max_ird_per_qp;
105 
106 	int device_cap_flags;
107 	u8 cq_overflow_detect;
108 	u8 srq_supported;
109 
110 	u32 wqe_size;
111 	u32 rqe_size;
112 	u32 ird_page_size;
113 	u8 local_ca_ack_delay;
114 	u8 ird;
115 	u8 num_ird_pages;
116 	u8 udp_encap;
117 };
118 
119 struct ocrdma_dma_mem {
120 	void *va;
121 	dma_addr_t pa;
122 	u32 size;
123 };
124 
125 struct ocrdma_pbl {
126 	void *va;
127 	dma_addr_t pa;
128 };
129 
130 struct ocrdma_queue_info {
131 	void *va;
132 	dma_addr_t dma;
133 	u32 size;
134 	u16 len;
135 	u16 entry_size;		/* Size of an element in the queue */
136 	u16 id;			/* qid, where to ring the doorbell. */
137 	u16 head, tail;
138 	bool created;
139 };
140 
141 struct ocrdma_aic_obj {         /* Adaptive interrupt coalescing (AIC) info */
142 	u32 prev_eqd;
143 	u64 eq_intr_cnt;
144 	u64 prev_eq_intr_cnt;
145 };
146 
147 struct ocrdma_eq {
148 	struct ocrdma_queue_info q;
149 	u32 vector;
150 	int cq_cnt;
151 	struct ocrdma_dev *dev;
152 	char irq_name[32];
153 	struct ocrdma_aic_obj aic_obj;
154 };
155 
156 struct ocrdma_mq {
157 	struct ocrdma_queue_info sq;
158 	struct ocrdma_queue_info cq;
159 	bool rearm_cq;
160 };
161 
162 struct mqe_ctx {
163 	struct mutex lock; /* for serializing mailbox commands on MQ */
164 	wait_queue_head_t cmd_wait;
165 	u32 tag;
166 	u16 cqe_status;
167 	u16 ext_status;
168 	bool cmd_done;
169 	bool fw_error_state;
170 };
171 
172 struct ocrdma_hw_mr {
173 	u32 lkey;
174 	u8 fr_mr;
175 	u8 remote_atomic;
176 	u8 remote_rd;
177 	u8 remote_wr;
178 	u8 local_rd;
179 	u8 local_wr;
180 	u8 mw_bind;
181 	u8 rsvd;
182 	u64 len;
183 	struct ocrdma_pbl *pbl_table;
184 	u32 num_pbls;
185 	u32 num_pbes;
186 	u32 pbl_size;
187 	u32 pbe_size;
188 	u64 va;
189 };
190 
191 struct ocrdma_mr {
192 	struct ib_mr ibmr;
193 	struct ib_umem *umem;
194 	struct ocrdma_hw_mr hwmr;
195 	u64 *pages;
196 	u32 npages;
197 };
198 
199 struct ocrdma_stats {
200 	u8 type;
201 	struct ocrdma_dev *dev;
202 };
203 
204 struct ocrdma_pd_resource_mgr {
205 	u32 pd_norm_start;
206 	u16 pd_norm_count;
207 	u16 pd_norm_thrsh;
208 	u16 max_normal_pd;
209 	u32 pd_dpp_start;
210 	u16 pd_dpp_count;
211 	u16 pd_dpp_thrsh;
212 	u16 max_dpp_pd;
213 	u16 dpp_page_index;
214 	unsigned long *pd_norm_bitmap;
215 	unsigned long *pd_dpp_bitmap;
216 	bool pd_prealloc_valid;
217 };
218 
219 struct stats_mem {
220 	struct ocrdma_mqe mqe;
221 	void *va;
222 	dma_addr_t pa;
223 	u32 size;
224 	char *debugfs_mem;
225 };
226 
227 struct phy_info {
228 	u16 auto_speeds_supported;
229 	u16 fixed_speeds_supported;
230 	u16 phy_type;
231 	u16 interface_type;
232 };
233 
234 enum ocrdma_flags {
235 	OCRDMA_FLAGS_LINK_STATUS_INIT = 0x01
236 };
237 
238 struct ocrdma_dev {
239 	struct ib_device ibdev;
240 	struct ocrdma_dev_attr attr;
241 
242 	struct mutex dev_lock; /* provides syncronise access to device data */
243 	spinlock_t flush_q_lock ____cacheline_aligned;
244 
245 	struct ocrdma_cq **cq_tbl;
246 	struct ocrdma_qp **qp_tbl;
247 
248 	struct ocrdma_eq *eq_tbl;
249 	int eq_cnt;
250 	struct delayed_work eqd_work;
251 	u16 base_eqid;
252 	u16 max_eq;
253 
254 	/* provided synchronization to sgid table for
255 	 * updating gid entries triggered by notifier.
256 	 */
257 	spinlock_t sgid_lock;
258 
259 	int gsi_qp_created;
260 	struct ocrdma_cq *gsi_sqcq;
261 	struct ocrdma_cq *gsi_rqcq;
262 
263 	struct {
264 		struct ocrdma_av *va;
265 		dma_addr_t pa;
266 		u32 size;
267 		u32 num_ah;
268 		/* provide synchronization for av
269 		 * entry allocations.
270 		 */
271 		spinlock_t lock;
272 		u32 ahid;
273 		struct ocrdma_pbl pbl;
274 	} av_tbl;
275 
276 	void *mbx_cmd;
277 	struct ocrdma_mq mq;
278 	struct mqe_ctx mqe_ctx;
279 
280 	struct be_dev_info nic_info;
281 	struct phy_info phy;
282 	char model_number[32];
283 	u32 hba_port_num;
284 
285 	struct list_head entry;
286 	int id;
287 	u64 *stag_arr;
288 	u8 sl; /* service level */
289 	bool pfc_state;
290 	atomic_t update_sl;
291 	u16 pvid;
292 	u32 asic_id;
293 	u32 flags;
294 
295 	ulong last_stats_time;
296 	struct mutex stats_lock; /* provide synch for debugfs operations */
297 	struct stats_mem stats_mem;
298 	struct ocrdma_stats rsrc_stats;
299 	struct ocrdma_stats rx_stats;
300 	struct ocrdma_stats wqe_stats;
301 	struct ocrdma_stats tx_stats;
302 	struct ocrdma_stats db_err_stats;
303 	struct ocrdma_stats tx_qp_err_stats;
304 	struct ocrdma_stats rx_qp_err_stats;
305 	struct ocrdma_stats tx_dbg_stats;
306 	struct ocrdma_stats rx_dbg_stats;
307 	struct ocrdma_stats driver_stats;
308 	struct ocrdma_stats reset_stats;
309 	struct dentry *dir;
310 	atomic_t async_err_stats[OCRDMA_MAX_ASYNC_ERRORS];
311 	atomic_t cqe_err_stats[OCRDMA_MAX_CQE_ERR];
312 	struct ocrdma_pd_resource_mgr *pd_mgr;
313 };
314 
315 struct ocrdma_cq {
316 	struct ib_cq ibcq;
317 	struct ocrdma_cqe *va;
318 	u32 phase;
319 	u32 getp;	/* pointer to pending wrs to
320 			 * return to stack, wrap arounds
321 			 * at max_hw_cqe
322 			 */
323 	u32 max_hw_cqe;
324 	bool phase_change;
325 	spinlock_t cq_lock ____cacheline_aligned; /* provide synchronization
326 						   * to cq polling
327 						   */
328 	/* syncronizes cq completion handler invoked from multiple context */
329 	spinlock_t comp_handler_lock ____cacheline_aligned;
330 	u16 id;
331 	u16 eqn;
332 
333 	struct ocrdma_ucontext *ucontext;
334 	dma_addr_t pa;
335 	u32 len;
336 	u32 cqe_cnt;
337 
338 	/* head of all qp's sq and rq for which cqes need to be flushed
339 	 * by the software.
340 	 */
341 	struct list_head sq_head, rq_head;
342 };
343 
344 struct ocrdma_pd {
345 	struct ib_pd ibpd;
346 	struct ocrdma_ucontext *uctx;
347 	u32 id;
348 	int num_dpp_qp;
349 	u32 dpp_page;
350 	bool dpp_enabled;
351 };
352 
353 struct ocrdma_ah {
354 	struct ib_ah ibah;
355 	struct ocrdma_av *av;
356 	u16 sgid_index;
357 	u32 id;
358 	u8 hdr_type;
359 };
360 
361 struct ocrdma_qp_hwq_info {
362 	u8 *va;			/* virtual address */
363 	u32 max_sges;
364 	u32 head, tail;
365 	u32 entry_size;
366 	u32 max_cnt;
367 	u32 max_wqe_idx;
368 	u16 dbid;		/* qid, where to ring the doorbell. */
369 	u32 len;
370 	dma_addr_t pa;
371 };
372 
373 struct ocrdma_srq {
374 	struct ib_srq ibsrq;
375 	u8 __iomem *db;
376 	struct ocrdma_qp_hwq_info rq;
377 	u64 *rqe_wr_id_tbl;
378 	u32 *idx_bit_fields;
379 	u32 bit_fields_len;
380 
381 	/* provide synchronization to multiple context(s) posting rqe */
382 	spinlock_t q_lock ____cacheline_aligned;
383 
384 	struct ocrdma_pd *pd;
385 	u32 id;
386 };
387 
388 struct ocrdma_qp {
389 	struct ib_qp ibqp;
390 
391 	u8 __iomem *sq_db;
392 	struct ocrdma_qp_hwq_info sq;
393 	struct {
394 		uint64_t wrid;
395 		uint16_t dpp_wqe_idx;
396 		uint16_t dpp_wqe;
397 		uint8_t  signaled;
398 		uint8_t  rsvd[3];
399 	} *wqe_wr_id_tbl;
400 	u32 max_inline_data;
401 
402 	/* provide synchronization to multiple context(s) posting wqe, rqe */
403 	spinlock_t q_lock ____cacheline_aligned;
404 	struct ocrdma_cq *sq_cq;
405 	/* list maintained per CQ to flush SQ errors */
406 	struct list_head sq_entry;
407 
408 	u8 __iomem *rq_db;
409 	struct ocrdma_qp_hwq_info rq;
410 	u64 *rqe_wr_id_tbl;
411 	struct ocrdma_cq *rq_cq;
412 	struct ocrdma_srq *srq;
413 	/* list maintained per CQ to flush RQ errors */
414 	struct list_head rq_entry;
415 
416 	enum ocrdma_qp_state state;	/*  QP state */
417 	int cap_flags;
418 	u32 max_ord, max_ird;
419 
420 	u32 id;
421 	struct ocrdma_pd *pd;
422 
423 	enum ib_qp_type qp_type;
424 
425 	int sgid_idx;
426 	u32 qkey;
427 	bool dpp_enabled;
428 	u8 *ird_q_va;
429 	bool signaled;
430 };
431 
432 struct ocrdma_ucontext {
433 	struct ib_ucontext ibucontext;
434 
435 	struct list_head mm_head;
436 	struct mutex mm_list_lock; /* protects list entries of mm type */
437 	struct ocrdma_pd *cntxt_pd;
438 	int pd_in_use;
439 
440 	struct {
441 		u32 *va;
442 		dma_addr_t pa;
443 		u32 len;
444 	} ah_tbl;
445 };
446 
447 struct ocrdma_mm {
448 	struct {
449 		u64 phy_addr;
450 		unsigned long len;
451 	} key;
452 	struct list_head entry;
453 };
454 
get_ocrdma_dev(struct ib_device * ibdev)455 static inline struct ocrdma_dev *get_ocrdma_dev(struct ib_device *ibdev)
456 {
457 	return container_of(ibdev, struct ocrdma_dev, ibdev);
458 }
459 
get_ocrdma_ucontext(struct ib_ucontext * ibucontext)460 static inline struct ocrdma_ucontext *get_ocrdma_ucontext(struct ib_ucontext
461 							  *ibucontext)
462 {
463 	return container_of(ibucontext, struct ocrdma_ucontext, ibucontext);
464 }
465 
get_ocrdma_pd(struct ib_pd * ibpd)466 static inline struct ocrdma_pd *get_ocrdma_pd(struct ib_pd *ibpd)
467 {
468 	return container_of(ibpd, struct ocrdma_pd, ibpd);
469 }
470 
get_ocrdma_cq(struct ib_cq * ibcq)471 static inline struct ocrdma_cq *get_ocrdma_cq(struct ib_cq *ibcq)
472 {
473 	return container_of(ibcq, struct ocrdma_cq, ibcq);
474 }
475 
get_ocrdma_qp(struct ib_qp * ibqp)476 static inline struct ocrdma_qp *get_ocrdma_qp(struct ib_qp *ibqp)
477 {
478 	return container_of(ibqp, struct ocrdma_qp, ibqp);
479 }
480 
get_ocrdma_mr(struct ib_mr * ibmr)481 static inline struct ocrdma_mr *get_ocrdma_mr(struct ib_mr *ibmr)
482 {
483 	return container_of(ibmr, struct ocrdma_mr, ibmr);
484 }
485 
get_ocrdma_ah(struct ib_ah * ibah)486 static inline struct ocrdma_ah *get_ocrdma_ah(struct ib_ah *ibah)
487 {
488 	return container_of(ibah, struct ocrdma_ah, ibah);
489 }
490 
get_ocrdma_srq(struct ib_srq * ibsrq)491 static inline struct ocrdma_srq *get_ocrdma_srq(struct ib_srq *ibsrq)
492 {
493 	return container_of(ibsrq, struct ocrdma_srq, ibsrq);
494 }
495 
is_cqe_valid(struct ocrdma_cq * cq,struct ocrdma_cqe * cqe)496 static inline int is_cqe_valid(struct ocrdma_cq *cq, struct ocrdma_cqe *cqe)
497 {
498 	int cqe_valid;
499 	cqe_valid = le32_to_cpu(cqe->flags_status_srcqpn) & OCRDMA_CQE_VALID;
500 	return (cqe_valid == cq->phase);
501 }
502 
is_cqe_for_sq(struct ocrdma_cqe * cqe)503 static inline int is_cqe_for_sq(struct ocrdma_cqe *cqe)
504 {
505 	return (le32_to_cpu(cqe->flags_status_srcqpn) &
506 		OCRDMA_CQE_QTYPE) ? 0 : 1;
507 }
508 
is_cqe_invalidated(struct ocrdma_cqe * cqe)509 static inline int is_cqe_invalidated(struct ocrdma_cqe *cqe)
510 {
511 	return (le32_to_cpu(cqe->flags_status_srcqpn) &
512 		OCRDMA_CQE_INVALIDATE) ? 1 : 0;
513 }
514 
is_cqe_imm(struct ocrdma_cqe * cqe)515 static inline int is_cqe_imm(struct ocrdma_cqe *cqe)
516 {
517 	return (le32_to_cpu(cqe->flags_status_srcqpn) &
518 		OCRDMA_CQE_IMM) ? 1 : 0;
519 }
520 
is_cqe_wr_imm(struct ocrdma_cqe * cqe)521 static inline int is_cqe_wr_imm(struct ocrdma_cqe *cqe)
522 {
523 	return (le32_to_cpu(cqe->flags_status_srcqpn) &
524 		OCRDMA_CQE_WRITE_IMM) ? 1 : 0;
525 }
526 
ocrdma_resolve_dmac(struct ocrdma_dev * dev,struct rdma_ah_attr * ah_attr,u8 * mac_addr)527 static inline int ocrdma_resolve_dmac(struct ocrdma_dev *dev,
528 		struct rdma_ah_attr *ah_attr, u8 *mac_addr)
529 {
530 	struct in6_addr in6;
531 
532 	memcpy(&in6, rdma_ah_read_grh(ah_attr)->dgid.raw, sizeof(in6));
533 	if (rdma_is_multicast_addr(&in6))
534 		rdma_get_mcast_mac(&in6, mac_addr);
535 	else if (rdma_link_local_addr(&in6))
536 		rdma_get_ll_mac(&in6, mac_addr);
537 	else
538 		memcpy(mac_addr, ah_attr->roce.dmac, ETH_ALEN);
539 	return 0;
540 }
541 
hca_name(struct ocrdma_dev * dev)542 static inline char *hca_name(struct ocrdma_dev *dev)
543 {
544 	switch (dev->nic_info.pdev->device) {
545 	case OC_SKH_DEVICE_PF:
546 	case OC_SKH_DEVICE_VF:
547 		return OC_NAME_SH;
548 	default:
549 		return OC_NAME_UNKNOWN;
550 	}
551 }
552 
ocrdma_get_eq_table_index(struct ocrdma_dev * dev,int eqid)553 static inline int ocrdma_get_eq_table_index(struct ocrdma_dev *dev,
554 		int eqid)
555 {
556 	int indx;
557 
558 	for (indx = 0; indx < dev->eq_cnt; indx++) {
559 		if (dev->eq_tbl[indx].q.id == eqid)
560 			return indx;
561 	}
562 
563 	return -EINVAL;
564 }
565 
ocrdma_get_asic_type(struct ocrdma_dev * dev)566 static inline u8 ocrdma_get_asic_type(struct ocrdma_dev *dev)
567 {
568 	if (dev->nic_info.dev_family == 0xF && !dev->asic_id) {
569 		pci_read_config_dword(
570 			dev->nic_info.pdev,
571 			OCRDMA_SLI_ASIC_ID_OFFSET, &dev->asic_id);
572 	}
573 
574 	return (dev->asic_id & OCRDMA_SLI_ASIC_GEN_NUM_MASK) >>
575 				OCRDMA_SLI_ASIC_GEN_NUM_SHIFT;
576 }
577 
ocrdma_get_pfc_prio(u8 * pfc,u8 prio)578 static inline u8 ocrdma_get_pfc_prio(u8 *pfc, u8 prio)
579 {
580 	return *(pfc + prio);
581 }
582 
ocrdma_get_app_prio(u8 * app_prio,u8 prio)583 static inline u8 ocrdma_get_app_prio(u8 *app_prio, u8 prio)
584 {
585 	return *(app_prio + prio);
586 }
587 
ocrdma_is_enabled_and_synced(u32 state)588 static inline u8 ocrdma_is_enabled_and_synced(u32 state)
589 {	/* May also be used to interpret TC-state, QCN-state
590 	 * Appl-state and Logical-link-state in future.
591 	 */
592 	return (state & OCRDMA_STATE_FLAG_ENABLED) &&
593 		(state & OCRDMA_STATE_FLAG_SYNC);
594 }
595 
ocrdma_get_ae_link_state(u32 ae_state)596 static inline u8 ocrdma_get_ae_link_state(u32 ae_state)
597 {
598 	return ((ae_state & OCRDMA_AE_LSC_LS_MASK) >> OCRDMA_AE_LSC_LS_SHIFT);
599 }
600 
ocrdma_is_udp_encap_supported(struct ocrdma_dev * dev)601 static inline bool ocrdma_is_udp_encap_supported(struct ocrdma_dev *dev)
602 {
603 	return (dev->attr.udp_encap & OCRDMA_L3_TYPE_IPV4) ||
604 	       (dev->attr.udp_encap & OCRDMA_L3_TYPE_IPV6);
605 }
606 
607 #endif
608