xref: /freebsd/sys/dev/qlnx/qlnxe/ecore_cxt.c (revision eac7052fdebb90caf2f653e06187bdbca837b9c7)
1 /*
2  * Copyright (c) 2017-2018 Cavium, Inc.
3  * All rights reserved.
4  *
5  *  Redistribution and use in source and binary forms, with or without
6  *  modification, are permitted provided that the following conditions
7  *  are met:
8  *
9  *  1. Redistributions of source code must retain the above copyright
10  *     notice, this list of conditions and the following disclaimer.
11  *  2. Redistributions in binary form must reproduce the above copyright
12  *     notice, this list of conditions and the following disclaimer in the
13  *     documentation and/or other materials provided with the distribution.
14  *
15  *  THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
16  *  AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
17  *  IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
18  *  ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
19  *  LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  *  INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  *  CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  *  ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  *  POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 /*
29  * File : ecore_cxt.c
30  */
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "bcm_osal.h"
35 #include "reg_addr.h"
36 #include "common_hsi.h"
37 #include "ecore_hsi_common.h"
38 #include "ecore_hsi_eth.h"
39 #include "tcp_common.h"
40 #include "ecore_hsi_iscsi.h"
41 #include "ecore_hsi_fcoe.h"
42 #include "ecore_hsi_roce.h"
43 #include "ecore_hsi_iwarp.h"
44 #include "ecore_rt_defs.h"
45 #include "ecore_status.h"
46 #include "ecore.h"
47 #include "ecore_init_ops.h"
48 #include "ecore_init_fw_funcs.h"
49 #include "ecore_cxt.h"
50 #include "ecore_hw.h"
51 #include "ecore_dev_api.h"
52 #include "ecore_sriov.h"
53 #include "ecore_rdma.h"
54 #include "ecore_mcp.h"
55 
56 /* Max number of connection types in HW (DQ/CDU etc.) */
57 #define MAX_CONN_TYPES		PROTOCOLID_COMMON
58 #define NUM_TASK_TYPES		2
59 #define NUM_TASK_PF_SEGMENTS	4
60 #define NUM_TASK_VF_SEGMENTS	1
61 
62 /* Doorbell-Queue constants */
63 #define DQ_RANGE_SHIFT	4
64 #define DQ_RANGE_ALIGN	(1 << DQ_RANGE_SHIFT)
65 
66 /* Searcher constants */
67 #define SRC_MIN_NUM_ELEMS 256
68 
69 /* Timers constants */
70 #define TM_SHIFT	7
71 #define TM_ALIGN	(1 << TM_SHIFT)
72 #define TM_ELEM_SIZE	4
73 
74 /* ILT constants */
75 #define ILT_PAGE_IN_BYTES(hw_p_size)	(1U << ((hw_p_size) + 12))
76 #define ILT_CFG_REG(cli, reg)		PSWRQ2_REG_##cli##_##reg##_RT_OFFSET
77 
78 /* ILT entry structure */
79 #define ILT_ENTRY_PHY_ADDR_MASK		0x000FFFFFFFFFFFULL
80 #define ILT_ENTRY_PHY_ADDR_SHIFT	0
81 #define ILT_ENTRY_VALID_MASK		0x1ULL
82 #define ILT_ENTRY_VALID_SHIFT		52
83 #define ILT_ENTRY_IN_REGS		2
84 #define ILT_REG_SIZE_IN_BYTES		4
85 
86 /* connection context union */
87 union conn_context {
88 	struct e4_core_conn_context  core_ctx;
89 	struct e4_eth_conn_context	  eth_ctx;
90 	struct e4_iscsi_conn_context iscsi_ctx;
91 	struct e4_fcoe_conn_context  fcoe_ctx;
92 	struct e4_roce_conn_context  roce_ctx;
93 };
94 
95 /* TYPE-0 task context - iSCSI, FCOE */
96 union type0_task_context {
97 	struct e4_iscsi_task_context iscsi_ctx;
98 	struct e4_fcoe_task_context  fcoe_ctx;
99 };
100 
101 /* TYPE-1 task context - ROCE */
102 union type1_task_context {
103 	struct e4_rdma_task_context roce_ctx;
104 };
105 
106 struct src_ent {
107 	u8  opaque[56];
108 	u64 next;
109 };
110 
111 #define CDUT_SEG_ALIGNMET 3 /* in 4k chunks */
112 #define CDUT_SEG_ALIGNMET_IN_BYTES (1 << (CDUT_SEG_ALIGNMET + 12))
113 
114 #define CONN_CXT_SIZE(p_hwfn) \
115 	ALIGNED_TYPE_SIZE(union conn_context, p_hwfn)
116 
117 #define SRQ_CXT_SIZE (sizeof(struct rdma_srq_context))
118 #define XRC_SRQ_CXT_SIZE (sizeof(struct rdma_xrc_srq_context))
119 
120 #define TYPE0_TASK_CXT_SIZE(p_hwfn) \
121 	ALIGNED_TYPE_SIZE(union type0_task_context, p_hwfn)
122 
123 /* Alignment is inherent to the type1_task_context structure */
124 #define TYPE1_TASK_CXT_SIZE(p_hwfn) sizeof(union type1_task_context)
125 
126 /* PF per protocl configuration object */
127 #define TASK_SEGMENTS   (NUM_TASK_PF_SEGMENTS + NUM_TASK_VF_SEGMENTS)
128 #define TASK_SEGMENT_VF (NUM_TASK_PF_SEGMENTS)
129 
130 struct ecore_tid_seg {
131 	u32	count;
132 	u8	type;
133 	bool	has_fl_mem;
134 };
135 
136 struct ecore_conn_type_cfg {
137 	u32			cid_count;
138 	u32			cids_per_vf;
139 	struct ecore_tid_seg	tid_seg[TASK_SEGMENTS];
140 };
141 
142 /* ILT Client configuration,
143  * Per connection type (protocol) resources (cids, tis, vf cids etc.)
144  * 1 - for connection context (CDUC) and for each task context we need two
145  * values, for regular task context and for force load memory
146  */
147 #define ILT_CLI_PF_BLOCKS	(1 + NUM_TASK_PF_SEGMENTS * 2)
148 #define ILT_CLI_VF_BLOCKS	(1 + NUM_TASK_VF_SEGMENTS * 2)
149 #define CDUC_BLK		(0)
150 #define SRQ_BLK			(0)
151 #define CDUT_SEG_BLK(n)		(1 + (u8)(n))
152 #define CDUT_FL_SEG_BLK(n, X)	(1 + (n) + NUM_TASK_##X##_SEGMENTS)
153 
154 struct ilt_cfg_pair {
155 	u32 reg;
156 	u32 val;
157 };
158 
159 struct ecore_ilt_cli_blk {
160 	u32 total_size; /* 0 means not active */
161 	u32 real_size_in_page;
162 	u32 start_line;
163 	u32 dynamic_line_cnt;
164 };
165 
166 struct ecore_ilt_client_cfg {
167 	bool				active;
168 
169 	/* ILT boundaries */
170 	struct ilt_cfg_pair		first;
171 	struct ilt_cfg_pair		last;
172 	struct ilt_cfg_pair		p_size;
173 
174 	/* ILT client blocks for PF */
175 	struct ecore_ilt_cli_blk	pf_blks[ILT_CLI_PF_BLOCKS];
176 	u32				pf_total_lines;
177 
178 	/* ILT client blocks for VFs */
179 	struct ecore_ilt_cli_blk	vf_blks[ILT_CLI_VF_BLOCKS];
180 	u32				vf_total_lines;
181 };
182 
183 /* Per Path -
184  *      ILT shadow table
185  *      Protocol acquired CID lists
186  *      PF start line in ILT
187  */
188 struct ecore_dma_mem {
189 	dma_addr_t	p_phys;
190 	void		*p_virt;
191 	osal_size_t	size;
192 };
193 
194 #define MAP_WORD_SIZE		sizeof(unsigned long)
195 #define BITS_PER_MAP_WORD	(MAP_WORD_SIZE * 8)
196 
197 struct ecore_cid_acquired_map {
198 	u32		start_cid;
199 	u32		max_count;
200 	unsigned long	*cid_map;
201 };
202 
203 struct ecore_cxt_mngr {
204 	/* Per protocl configuration */
205 	struct ecore_conn_type_cfg	conn_cfg[MAX_CONN_TYPES];
206 
207 	/* computed ILT structure */
208 	struct ecore_ilt_client_cfg	clients[ILT_CLI_MAX];
209 
210 	/* Task type sizes */
211 	u32				task_type_size[NUM_TASK_TYPES];
212 
213 	/* total number of VFs for this hwfn -
214 	 * ALL VFs are symmetric in terms of HW resources
215 	 */
216 	u32				vf_count;
217 
218 	/* Acquired CIDs */
219 	struct ecore_cid_acquired_map acquired[MAX_CONN_TYPES];
220 	/* TBD - do we want this allocated to reserve space? */
221 	struct ecore_cid_acquired_map acquired_vf[MAX_CONN_TYPES][COMMON_MAX_NUM_VFS];
222 
223 	/* ILT shadow table */
224 	struct ecore_dma_mem		*ilt_shadow;
225 	u32				pf_start_line;
226 
227 	/* Mutex for a dynamic ILT allocation */
228 	osal_mutex_t			mutex;
229 
230 	/* SRC T2 */
231 	struct ecore_dma_mem		*t2;
232 	u32				t2_num_pages;
233 	u64				first_free;
234 	u64				last_free;
235 
236 	/* The infrastructure originally was very generic and context/task
237 	 * oriented - per connection-type we would set how many of those
238 	 * are needed, and later when determining how much memory we're
239 	 * needing for a given block we'd iterate over all the relevant
240 	 * connection-types.
241 	 * But since then we've had some additional resources, some of which
242 	 * require memory which is indepent of the general context/task
243 	 * scheme. We add those here explicitly per-feature.
244 	 */
245 
246 	/* total number of SRQ's for this hwfn */
247 	u32				srq_count;
248 	u32				xrc_srq_count;
249 
250 	/* Maximal number of L2 steering filters */
251 	u32				arfs_count;
252 
253 	/* TODO - VF arfs filters ? */
254 };
255 
256 /* check if resources/configuration is required according to protocol type */
257 static bool src_proto(enum protocol_type type)
258 {
259 	return	type == PROTOCOLID_ISCSI	||
260 		type == PROTOCOLID_FCOE		||
261 		type == PROTOCOLID_IWARP;
262 }
263 
264 static bool tm_cid_proto(enum protocol_type type)
265 {
266 	return type == PROTOCOLID_ISCSI ||
267 	       type == PROTOCOLID_FCOE  ||
268 	       type == PROTOCOLID_ROCE  ||
269 	       type == PROTOCOLID_IWARP;
270 }
271 
272 static bool tm_tid_proto(enum protocol_type type)
273 {
274 	return type == PROTOCOLID_FCOE;
275 }
276 
277 /* counts the iids for the CDU/CDUC ILT client configuration */
278 struct ecore_cdu_iids {
279 	u32 pf_cids;
280 	u32 per_vf_cids;
281 };
282 
283 static void ecore_cxt_cdu_iids(struct ecore_cxt_mngr   *p_mngr,
284 			       struct ecore_cdu_iids	*iids)
285 {
286 	u32 type;
287 
288 	for (type = 0; type < MAX_CONN_TYPES; type++) {
289 		iids->pf_cids += p_mngr->conn_cfg[type].cid_count;
290 		iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
291 	}
292 }
293 
294 /* counts the iids for the Searcher block configuration */
295 struct ecore_src_iids {
296 	u32			pf_cids;
297 	u32			per_vf_cids;
298 };
299 
300 static void ecore_cxt_src_iids(struct ecore_cxt_mngr *p_mngr,
301 			       struct ecore_src_iids *iids)
302 {
303 	u32 i;
304 
305 	for (i = 0; i < MAX_CONN_TYPES; i++) {
306 		if (!src_proto(i))
307 			continue;
308 
309 		iids->pf_cids += p_mngr->conn_cfg[i].cid_count;
310 		iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf;
311 	}
312 
313 	/* Add L2 filtering filters in addition */
314 	iids->pf_cids += p_mngr->arfs_count;
315 }
316 
317 /* counts the iids for the Timers block configuration */
318 struct ecore_tm_iids {
319 	u32 pf_cids;
320 	u32 pf_tids[NUM_TASK_PF_SEGMENTS]; /* per segment */
321 	u32 pf_tids_total;
322 	u32 per_vf_cids;
323 	u32 per_vf_tids;
324 };
325 
326 static void ecore_cxt_tm_iids(struct ecore_cxt_mngr *p_mngr,
327 			      struct ecore_tm_iids *iids)
328 {
329 	bool tm_vf_required = false;
330 	bool tm_required = false;
331 	int i, j;
332 
333 	/* Timers is a special case -> we don't count how many cids require
334 	 * timers but what's the max cid that will be used by the timer block.
335 	 * therefore we traverse in reverse order, and once we hit a protocol
336 	 * that requires the timers memory, we'll sum all the protocols up
337 	 * to that one.
338 	 */
339 	for (i = MAX_CONN_TYPES - 1; i >= 0; i--) {
340 		struct ecore_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
341 
342 		if (tm_cid_proto(i) || tm_required) {
343 			if (p_cfg->cid_count)
344 				tm_required = true;
345 
346 			iids->pf_cids += p_cfg->cid_count;
347 		}
348 
349 		if (tm_cid_proto(i) || tm_vf_required) {
350 			if (p_cfg->cids_per_vf)
351 				tm_vf_required = true;
352 
353 			iids->per_vf_cids += p_cfg->cids_per_vf;
354 		}
355 
356 		if (tm_tid_proto(i)) {
357 			struct ecore_tid_seg *segs = p_cfg->tid_seg;
358 
359 			/* for each segment there is at most one
360 			 * protocol for which count is not 0.
361 			 */
362 			for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
363 				iids->pf_tids[j] += segs[j].count;
364 
365 			/* The last array elelment is for the VFs. As for PF
366 			 * segments there can be only one protocol for
367 			 * which this value is not 0.
368 			 */
369 			iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
370 		}
371 	}
372 
373 	iids->pf_cids = ROUNDUP(iids->pf_cids, TM_ALIGN);
374 	iids->per_vf_cids = ROUNDUP(iids->per_vf_cids, TM_ALIGN);
375 	iids->per_vf_tids = ROUNDUP(iids->per_vf_tids, TM_ALIGN);
376 
377 	for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
378 		iids->pf_tids[j] = ROUNDUP(iids->pf_tids[j], TM_ALIGN);
379 		iids->pf_tids_total += iids->pf_tids[j];
380 	}
381 }
382 
383 static void ecore_cxt_qm_iids(struct ecore_hwfn *p_hwfn,
384 			      struct ecore_qm_iids *iids)
385 {
386 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
387 	struct ecore_tid_seg *segs;
388 	u32 vf_cids = 0, type, j;
389 	u32 vf_tids = 0;
390 
391 	for (type = 0; type < MAX_CONN_TYPES; type++) {
392 		iids->cids += p_mngr->conn_cfg[type].cid_count;
393 		vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
394 
395 		segs = p_mngr->conn_cfg[type].tid_seg;
396 		/* for each segment there is at most one
397 		 * protocol for which count is not 0.
398 		 */
399 		for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
400 			iids->tids += segs[j].count;
401 
402 		/* The last array elelment is for the VFs. As for PF
403 		 * segments there can be only one protocol for
404 		 * which this value is not 0.
405 		 */
406 		vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
407 	}
408 
409 	iids->vf_cids += vf_cids * p_mngr->vf_count;
410 	iids->tids += vf_tids * p_mngr->vf_count;
411 
412 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
413 		   "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
414 		   iids->cids, iids->vf_cids, iids->tids, vf_tids);
415 }
416 
417 static struct ecore_tid_seg *ecore_cxt_tid_seg_info(struct ecore_hwfn   *p_hwfn,
418 						    u32			seg)
419 {
420 	struct ecore_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
421 	u32 i;
422 
423 	/* Find the protocol with tid count > 0 for this segment.
424 	   Note: there can only be one and this is already validated.
425 	 */
426 	for (i = 0; i < MAX_CONN_TYPES; i++) {
427 		if (p_cfg->conn_cfg[i].tid_seg[seg].count)
428 			return &p_cfg->conn_cfg[i].tid_seg[seg];
429 	}
430 	return OSAL_NULL;
431 }
432 
433 static void ecore_cxt_set_srq_count(struct ecore_hwfn *p_hwfn,
434 				    u32 num_srqs, u32 num_xrc_srqs)
435 {
436 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
437 
438 	p_mgr->srq_count = num_srqs;
439 	p_mgr->xrc_srq_count = num_xrc_srqs;
440 }
441 
442 u32 ecore_cxt_get_srq_count(struct ecore_hwfn *p_hwfn)
443 {
444 	return p_hwfn->p_cxt_mngr->srq_count;
445 }
446 
447 u32 ecore_cxt_get_xrc_srq_count(struct ecore_hwfn *p_hwfn)
448 {
449 	return p_hwfn->p_cxt_mngr->xrc_srq_count;
450 }
451 
452 u32 ecore_cxt_get_ilt_page_size(struct ecore_hwfn *p_hwfn,
453 				enum ilt_clients ilt_client)
454 {
455 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
456 	struct ecore_ilt_client_cfg *p_cli = &p_mngr->clients[ilt_client];
457 
458 	return ILT_PAGE_IN_BYTES(p_cli->p_size.val);
459 }
460 
461 static u32 ecore_cxt_srqs_per_page(struct ecore_hwfn *p_hwfn)
462 {
463 	u32 page_size;
464 
465 	page_size = ecore_cxt_get_ilt_page_size(p_hwfn, ILT_CLI_TSDM);
466 	return page_size / SRQ_CXT_SIZE;
467 }
468 
469 u32 ecore_cxt_get_total_srq_count(struct ecore_hwfn *p_hwfn)
470 {
471 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
472 	u32 total_srqs;
473 
474 	total_srqs = p_mgr->srq_count;
475 
476 	/* XRC SRQs use the first and only the first SRQ ILT page. So if XRC
477 	 * SRQs are requested we need to allocate an extra SRQ ILT page for
478 	 * them. For that We increase the number of regular SRQs to cause the
479 	 * allocation of that extra page.
480 	 */
481 	if (p_mgr->xrc_srq_count)
482 		total_srqs += ecore_cxt_srqs_per_page(p_hwfn);
483 
484 	return total_srqs;
485 }
486 
487 /* set the iids (cid/tid) count per protocol */
488 static void ecore_cxt_set_proto_cid_count(struct ecore_hwfn *p_hwfn,
489 					  enum protocol_type type,
490 					  u32 cid_count, u32 vf_cid_cnt)
491 {
492 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
493 	struct ecore_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
494 
495 	p_conn->cid_count = ROUNDUP(cid_count, DQ_RANGE_ALIGN);
496 	p_conn->cids_per_vf = ROUNDUP(vf_cid_cnt, DQ_RANGE_ALIGN);
497 
498 	if (type == PROTOCOLID_ROCE) {
499 		u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
500 		u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
501 		u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
502 		u32 align = elems_per_page * DQ_RANGE_ALIGN;
503 
504 		p_conn->cid_count = ROUNDUP(p_conn->cid_count, align);
505 	}
506 }
507 
508 u32 ecore_cxt_get_proto_cid_count(struct ecore_hwfn	*p_hwfn,
509 				  enum protocol_type	type,
510 				  u32			*vf_cid)
511 {
512 	if (vf_cid)
513 		*vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
514 
515 	return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
516 }
517 
518 u32 ecore_cxt_get_proto_cid_start(struct ecore_hwfn	*p_hwfn,
519 				  enum protocol_type	type)
520 {
521 	return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
522 }
523 
524 u32 ecore_cxt_get_proto_tid_count(struct ecore_hwfn *p_hwfn,
525 				  enum protocol_type type)
526 {
527 	u32 cnt = 0;
528 	int i;
529 
530 	for (i = 0; i < TASK_SEGMENTS; i++)
531 		cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
532 
533 	return cnt;
534 }
535 
536 static void ecore_cxt_set_proto_tid_count(struct ecore_hwfn *p_hwfn,
537 					  enum protocol_type proto,
538 					  u8 seg,
539 					  u8 seg_type,
540 					  u32 count,
541 					  bool has_fl)
542 {
543 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
544 	struct ecore_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
545 
546 	p_seg->count = count;
547 	p_seg->has_fl_mem = has_fl;
548 	p_seg->type = seg_type;
549 }
550 
551 /* the *p_line parameter must be either 0 for the first invocation or the
552    value returned in the previous invocation.
553  */
554 static void ecore_ilt_cli_blk_fill(struct ecore_ilt_client_cfg	*p_cli,
555 				   struct ecore_ilt_cli_blk	*p_blk,
556 				   u32				start_line,
557 				   u32				total_size,
558 				   u32				elem_size)
559 {
560 	u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
561 
562 	/* verify that it's called once for each block */
563 	if (p_blk->total_size)
564 		return;
565 
566 	p_blk->total_size = total_size;
567 	p_blk->real_size_in_page = 0;
568 	if (elem_size)
569 		p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
570 	p_blk->start_line = start_line;
571 }
572 
573 static void ecore_ilt_cli_adv_line(struct ecore_hwfn		*p_hwfn,
574 				    struct ecore_ilt_client_cfg	*p_cli,
575 				    struct ecore_ilt_cli_blk	*p_blk,
576 				    u32				*p_line,
577 				    enum ilt_clients		client_id)
578 {
579 	if (!p_blk->total_size)
580 		return;
581 
582 	if (!p_cli->active)
583 		p_cli->first.val = *p_line;
584 
585 	p_cli->active = true;
586 	*p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
587 	p_cli->last.val = *p_line-1;
588 
589 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
590 		   "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
591 		   client_id, p_cli->first.val, p_cli->last.val,
592 		   p_blk->total_size, p_blk->real_size_in_page,
593 		   p_blk->start_line);
594 }
595 
596 static u32 ecore_ilt_get_dynamic_line_cnt(struct ecore_hwfn *p_hwfn,
597 					  enum ilt_clients ilt_client)
598 {
599 	u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
600 	struct ecore_ilt_client_cfg *p_cli;
601 	u32 lines_to_skip = 0;
602 	u32 cxts_per_p;
603 
604 	/* TBD MK: ILT code should be simplified once PROTO enum is changed */
605 
606 	if (ilt_client == ILT_CLI_CDUC) {
607 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
608 
609 		cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
610 			     (u32)CONN_CXT_SIZE(p_hwfn);
611 
612 		lines_to_skip = cid_count / cxts_per_p;
613 	}
614 
615 	return lines_to_skip;
616 }
617 
618 static struct ecore_ilt_client_cfg *
619 ecore_cxt_set_cli(struct ecore_ilt_client_cfg *p_cli)
620 {
621 	p_cli->active = false;
622 	p_cli->first.val = 0;
623 	p_cli->last.val = 0;
624 	return p_cli;
625 }
626 
627 static struct ecore_ilt_cli_blk *
628 ecore_cxt_set_blk(struct ecore_ilt_cli_blk *p_blk)
629 {
630 	p_blk->total_size = 0;
631 	return p_blk;
632 }
633 
634 enum _ecore_status_t ecore_cxt_cfg_ilt_compute(struct ecore_hwfn *p_hwfn,
635 					       u32 *line_count)
636 {
637 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
638 	u32 curr_line, total, i, task_size, line;
639 	struct ecore_ilt_client_cfg *p_cli;
640 	struct ecore_ilt_cli_blk *p_blk;
641 	struct ecore_cdu_iids cdu_iids;
642 	struct ecore_src_iids src_iids;
643 	struct ecore_qm_iids qm_iids;
644 	struct ecore_tm_iids tm_iids;
645 	struct ecore_tid_seg *p_seg;
646 
647 	OSAL_MEM_ZERO(&qm_iids, sizeof(qm_iids));
648 	OSAL_MEM_ZERO(&cdu_iids, sizeof(cdu_iids));
649 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
650 	OSAL_MEM_ZERO(&tm_iids, sizeof(tm_iids));
651 
652 	p_mngr->pf_start_line = RESC_START(p_hwfn, ECORE_ILT);
653 
654 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
655 		   "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
656 		   p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
657 
658 	/* CDUC */
659 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUC]);
660 
661 	curr_line = p_mngr->pf_start_line;
662 
663 	/* CDUC PF */
664 	p_cli->pf_total_lines = 0;
665 
666 	/* get the counters for the CDUC,CDUC and QM clients  */
667 	ecore_cxt_cdu_iids(p_mngr, &cdu_iids);
668 
669 	p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[CDUC_BLK]);
670 
671 	total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
672 
673 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
674 			       total, CONN_CXT_SIZE(p_hwfn));
675 
676 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
677 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
678 
679 	p_blk->dynamic_line_cnt = ecore_ilt_get_dynamic_line_cnt(p_hwfn,
680 								 ILT_CLI_CDUC);
681 
682 	/* CDUC VF */
683 	p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[CDUC_BLK]);
684 	total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
685 
686 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
687 			       total, CONN_CXT_SIZE(p_hwfn));
688 
689 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
690 	p_cli->vf_total_lines = curr_line - p_blk->start_line;
691 
692 	for (i = 1; i < p_mngr->vf_count; i++)
693 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
694 				       ILT_CLI_CDUC);
695 
696 	/* CDUT PF */
697 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUT]);
698 	p_cli->first.val = curr_line;
699 
700 	/* first the 'working' task memory */
701 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
702 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
703 		if (!p_seg || p_seg->count == 0)
704 			continue;
705 
706 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[CDUT_SEG_BLK(i)]);
707 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
708 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
709 				       p_mngr->task_type_size[p_seg->type]);
710 
711 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
712 				       ILT_CLI_CDUT);
713 	}
714 
715 	/* next the 'init' task memory (forced load memory) */
716 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
717 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
718 		if (!p_seg || p_seg->count == 0)
719 			continue;
720 
721 		p_blk = ecore_cxt_set_blk(
722 				&p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)]);
723 
724 		if (!p_seg->has_fl_mem) {
725 			/* The segment is active (total size pf 'working'
726 			 * memory is > 0) but has no FL (forced-load, Init)
727 			 * memory. Thus:
728 			 *
729 			 * 1.   The total-size in the corrsponding FL block of
730 			 *      the ILT client is set to 0 - No ILT line are
731 			 *      provisioned and no ILT memory allocated.
732 			 *
733 			 * 2.   The start-line of said block is set to the
734 			 *      start line of the matching working memory
735 			 *      block in the ILT client. This is later used to
736 			 *      configure the CDU segment offset registers and
737 			 *      results in an FL command for TIDs of this
738 			 *      segement behaves as regular load commands
739 			 *      (loading TIDs from the working memory).
740 			 */
741 			line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
742 
743 			ecore_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
744 			continue;
745 		}
746 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
747 
748 		ecore_ilt_cli_blk_fill(p_cli, p_blk,
749 				       curr_line, total,
750 				       p_mngr->task_type_size[p_seg->type]);
751 
752 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
753 				       ILT_CLI_CDUT);
754 	}
755 	p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
756 
757 	/* CDUT VF */
758 	p_seg = ecore_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
759 	if (p_seg && p_seg->count) {
760 		/* Stricly speaking we need to iterate over all VF
761 		 * task segment types, but a VF has only 1 segment
762 		 */
763 
764 		/* 'working' memory */
765 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
766 
767 		p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[CDUT_SEG_BLK(0)]);
768 		ecore_ilt_cli_blk_fill(p_cli, p_blk,
769 				       curr_line, total,
770 				       p_mngr->task_type_size[p_seg->type]);
771 
772 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
773 				       ILT_CLI_CDUT);
774 
775 		/* 'init' memory */
776 		p_blk = ecore_cxt_set_blk(
777 				&p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)]);
778 		if (!p_seg->has_fl_mem) {
779 			/* see comment above */
780 			line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
781 			ecore_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
782 		} else {
783 			task_size = p_mngr->task_type_size[p_seg->type];
784 			ecore_ilt_cli_blk_fill(p_cli, p_blk,
785 					       curr_line, total,
786 					       task_size);
787 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
788 					       ILT_CLI_CDUT);
789 		}
790 		p_cli->vf_total_lines = curr_line -
791 					p_cli->vf_blks[0].start_line;
792 
793 		/* Now for the rest of the VFs */
794 		for (i = 1; i < p_mngr->vf_count; i++) {
795 			/* don't set p_blk i.e. don't clear total_size */
796 			p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
797 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
798 					       ILT_CLI_CDUT);
799 
800 			/* don't set p_blk i.e. don't clear total_size */
801 			p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
802 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
803 					       ILT_CLI_CDUT);
804 		}
805 	}
806 
807 	/* QM */
808 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_QM]);
809 	p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
810 
811 	ecore_cxt_qm_iids(p_hwfn, &qm_iids);
812 	total = ecore_qm_pf_mem_size(qm_iids.cids,
813 				     qm_iids.vf_cids, qm_iids.tids,
814 				     p_hwfn->qm_info.num_pqs,
815 				     p_hwfn->qm_info.num_vf_pqs);
816 
817 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
818 		   "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
819 		   qm_iids.cids, qm_iids.vf_cids, qm_iids.tids,
820 		   p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
821 
822 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total * 0x1000,
823 			       QM_PQ_ELEMENT_SIZE);
824 
825 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
826 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
827 
828 	/* SRC */
829 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_SRC]);
830 	ecore_cxt_src_iids(p_mngr, &src_iids);
831 
832 	/* Both the PF and VFs searcher connections are stored in the per PF
833 	 * database. Thus sum the PF searcher cids and all the VFs searcher
834 	 * cids.
835 	 */
836 	total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
837 	if (total) {
838 		u32 local_max = OSAL_MAX_T(u32, total,
839 					   SRC_MIN_NUM_ELEMS);
840 
841 		total = OSAL_ROUNDUP_POW_OF_TWO(local_max);
842 
843 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
844 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
845 				       total * sizeof(struct src_ent),
846 				       sizeof(struct src_ent));
847 
848 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
849 				       ILT_CLI_SRC);
850 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
851 	}
852 
853 	/* TM PF */
854 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_TM]);
855 	ecore_cxt_tm_iids(p_mngr, &tm_iids);
856 	total = tm_iids.pf_cids + tm_iids.pf_tids_total;
857 	if (total) {
858 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
859 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
860 				       total * TM_ELEM_SIZE,
861 				       TM_ELEM_SIZE);
862 
863 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
864 				       ILT_CLI_TM);
865 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
866 	}
867 
868 	/* TM VF */
869 	total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
870 	if (total) {
871 		p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[0]);
872 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
873 				       total * TM_ELEM_SIZE,
874 				       TM_ELEM_SIZE);
875 
876 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
877 				       ILT_CLI_TM);
878 
879 		p_cli->vf_total_lines = curr_line - p_blk->start_line;
880 		for (i = 1; i < p_mngr->vf_count; i++) {
881 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
882 					       ILT_CLI_TM);
883 		}
884 	}
885 
886 	/* TSDM (SRQ CONTEXT) */
887 	total = ecore_cxt_get_total_srq_count(p_hwfn);
888 	if (total) {
889 		p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_TSDM]);
890 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[SRQ_BLK]);
891 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
892 				       total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
893 
894 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
895 				       ILT_CLI_TSDM);
896 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
897 	}
898 
899 	*line_count = curr_line - p_hwfn->p_cxt_mngr->pf_start_line;
900 
901 	if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
902 	    RESC_NUM(p_hwfn, ECORE_ILT)) {
903 		return ECORE_INVAL;
904 	}
905 
906 	return ECORE_SUCCESS;
907 }
908 
909 u32 ecore_cxt_cfg_ilt_compute_excess(struct ecore_hwfn *p_hwfn, u32 used_lines)
910 {
911 	struct ecore_ilt_client_cfg *p_cli;
912 	u32 excess_lines, available_lines;
913 	struct ecore_cxt_mngr *p_mngr;
914 	u32 ilt_page_size, elem_size;
915 	struct ecore_tid_seg *p_seg;
916 	int i;
917 
918 	available_lines = RESC_NUM(p_hwfn, ECORE_ILT);
919 	excess_lines = used_lines - available_lines;
920 
921 	if (!excess_lines)
922 		return 0;
923 
924 	if (!ECORE_IS_RDMA_PERSONALITY(p_hwfn))
925 		return 0;
926 
927 	p_mngr = p_hwfn->p_cxt_mngr;
928 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
929 	ilt_page_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
930 
931 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
932 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
933 		if (!p_seg || p_seg->count == 0)
934 			continue;
935 
936 		elem_size = p_mngr->task_type_size[p_seg->type];
937 		if (!elem_size)
938 			continue;
939 
940 		return (ilt_page_size / elem_size) * excess_lines;
941 	}
942 
943 	DP_ERR(p_hwfn, "failed computing excess ILT lines\n");
944 	return 0;
945 }
946 
947 static void ecore_cxt_src_t2_free(struct ecore_hwfn *p_hwfn)
948 {
949 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
950 	u32 i;
951 
952 	if (!p_mngr->t2)
953 		return;
954 
955 	for (i = 0; i < p_mngr->t2_num_pages; i++)
956 		if (p_mngr->t2[i].p_virt)
957 			OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
958 					       p_mngr->t2[i].p_virt,
959 					       p_mngr->t2[i].p_phys,
960 					       p_mngr->t2[i].size);
961 
962 	OSAL_FREE(p_hwfn->p_dev, p_mngr->t2);
963 	p_mngr->t2 = OSAL_NULL;
964 }
965 
966 static enum _ecore_status_t ecore_cxt_src_t2_alloc(struct ecore_hwfn *p_hwfn)
967 {
968 	struct ecore_cxt_mngr *p_mngr  = p_hwfn->p_cxt_mngr;
969 	u32 conn_num, total_size, ent_per_page, psz, i;
970 	struct ecore_ilt_client_cfg *p_src;
971 	struct ecore_src_iids src_iids;
972 	struct ecore_dma_mem *p_t2;
973 	enum _ecore_status_t rc;
974 
975 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
976 
977 	/* if the SRC ILT client is inactive - there are no connection
978 	 * requiring the searcer, leave.
979 	 */
980 	p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
981 	if (!p_src->active)
982 		return ECORE_SUCCESS;
983 
984 	ecore_cxt_src_iids(p_mngr, &src_iids);
985 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
986 	total_size = conn_num * sizeof(struct src_ent);
987 
988 	/* use the same page size as the SRC ILT client */
989 	psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
990 	p_mngr->t2_num_pages = DIV_ROUND_UP(total_size, psz);
991 
992 	/* allocate t2 */
993 	p_mngr->t2 = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL,
994 				 p_mngr->t2_num_pages *
995 				 sizeof(struct ecore_dma_mem));
996 	if (!p_mngr->t2) {
997 		DP_NOTICE(p_hwfn, false, "Failed to allocate t2 table\n");
998 		rc = ECORE_NOMEM;
999 		goto t2_fail;
1000 	}
1001 
1002 	/* allocate t2 pages */
1003 	for (i = 0; i < p_mngr->t2_num_pages; i++) {
1004 		u32 size = OSAL_MIN_T(u32, total_size, psz);
1005 		void **p_virt = &p_mngr->t2[i].p_virt;
1006 
1007 		*p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
1008 						  &p_mngr->t2[i].p_phys,
1009 						  size);
1010 		if (!p_mngr->t2[i].p_virt) {
1011 			rc = ECORE_NOMEM;
1012 			goto t2_fail;
1013 		}
1014 		OSAL_MEM_ZERO(*p_virt, size);
1015 		p_mngr->t2[i].size = size;
1016 		total_size -= size;
1017 	}
1018 
1019 	/* Set the t2 pointers */
1020 
1021 	/* entries per page - must be a power of two */
1022 	ent_per_page = psz / sizeof(struct src_ent);
1023 
1024 	p_mngr->first_free = (u64)p_mngr->t2[0].p_phys;
1025 
1026 	p_t2 = &p_mngr->t2[(conn_num - 1) / ent_per_page];
1027 	p_mngr->last_free = (u64)p_t2->p_phys +
1028 				 ((conn_num - 1) & (ent_per_page - 1)) *
1029 				 sizeof(struct src_ent);
1030 
1031 	for (i = 0; i < p_mngr->t2_num_pages; i++) {
1032 		u32 ent_num = OSAL_MIN_T(u32, ent_per_page, conn_num);
1033 		struct src_ent *entries = p_mngr->t2[i].p_virt;
1034 		u64 p_ent_phys = (u64)p_mngr->t2[i].p_phys, val;
1035 		u32 j;
1036 
1037 		for (j = 0; j < ent_num - 1; j++) {
1038 			val = p_ent_phys +
1039 			      (j + 1) * sizeof(struct src_ent);
1040 			entries[j].next = OSAL_CPU_TO_BE64(val);
1041 		}
1042 
1043 		if (i < p_mngr->t2_num_pages - 1)
1044 			val = (u64)p_mngr->t2[i + 1].p_phys;
1045 		else
1046 			val = 0;
1047 		entries[j].next = OSAL_CPU_TO_BE64(val);
1048 
1049 		conn_num -= ent_num;
1050 	}
1051 
1052 	return ECORE_SUCCESS;
1053 
1054 t2_fail:
1055 	ecore_cxt_src_t2_free(p_hwfn);
1056 	return rc;
1057 }
1058 
1059 #define for_each_ilt_valid_client(pos, clients)	\
1060 	for (pos = 0; pos < ILT_CLI_MAX; pos++)	\
1061 		if (!clients[pos].active) {	\
1062 			continue;		\
1063 		} else				\
1064 
1065 /* Total number of ILT lines used by this PF */
1066 static u32 ecore_cxt_ilt_shadow_size(struct ecore_ilt_client_cfg *ilt_clients)
1067 {
1068 	u32 size = 0;
1069 	u32 i;
1070 
1071 	for_each_ilt_valid_client(i, ilt_clients)
1072 		size += (ilt_clients[i].last.val -
1073 			 ilt_clients[i].first.val + 1);
1074 
1075 	return size;
1076 }
1077 
1078 static void ecore_ilt_shadow_free(struct ecore_hwfn *p_hwfn)
1079 {
1080 	struct ecore_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
1081 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1082 	u32 ilt_size, i;
1083 
1084 	if (p_mngr->ilt_shadow == OSAL_NULL)
1085 		return;
1086 
1087 	ilt_size = ecore_cxt_ilt_shadow_size(p_cli);
1088 
1089 	for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
1090 		struct ecore_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
1091 
1092 		if (p_dma->p_virt)
1093 			OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
1094 					       p_dma->p_virt,
1095 					       p_dma->p_phys,
1096 					       p_dma->size);
1097 		p_dma->p_virt = OSAL_NULL;
1098 	}
1099 	OSAL_FREE(p_hwfn->p_dev, p_mngr->ilt_shadow);
1100 	p_mngr->ilt_shadow = OSAL_NULL;
1101 }
1102 
1103 static enum _ecore_status_t ecore_ilt_blk_alloc(struct ecore_hwfn *p_hwfn,
1104 						struct ecore_ilt_cli_blk *p_blk,
1105 						enum ilt_clients ilt_client,
1106 						u32 start_line_offset)
1107 {
1108 	struct ecore_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
1109 	u32 lines, line, sz_left, lines_to_skip = 0;
1110 
1111 	/* Special handling for RoCE that supports dynamic allocation */
1112 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn) &&
1113 	    ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
1114 		return ECORE_SUCCESS;
1115 
1116 	lines_to_skip = p_blk->dynamic_line_cnt;
1117 
1118 	if (!p_blk->total_size)
1119 		return ECORE_SUCCESS;
1120 
1121 	sz_left = p_blk->total_size;
1122 	lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) -
1123 		lines_to_skip;
1124 	line = p_blk->start_line + start_line_offset -
1125 	       p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
1126 
1127 	for (; lines; lines--) {
1128 		dma_addr_t p_phys;
1129 		void *p_virt;
1130 		u32 size;
1131 
1132 		size = OSAL_MIN_T(u32, sz_left, p_blk->real_size_in_page);
1133 		p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
1134 						 &p_phys, size);
1135 		if (!p_virt)
1136 			return ECORE_NOMEM;
1137 		OSAL_MEM_ZERO(p_virt, size);
1138 
1139 		ilt_shadow[line].p_phys = p_phys;
1140 		ilt_shadow[line].p_virt = p_virt;
1141 		ilt_shadow[line].size = size;
1142 
1143 		DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1144 			   "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
1145 			   line, (unsigned long long)p_phys, p_virt, size);
1146 
1147 		sz_left -= size;
1148 		line++;
1149 	}
1150 
1151 	return ECORE_SUCCESS;
1152 }
1153 
1154 static enum _ecore_status_t ecore_ilt_shadow_alloc(struct ecore_hwfn *p_hwfn)
1155 {
1156 	struct ecore_cxt_mngr *p_mngr  = p_hwfn->p_cxt_mngr;
1157 	struct ecore_ilt_client_cfg *clients = p_mngr->clients;
1158 	struct ecore_ilt_cli_blk *p_blk;
1159 	u32 size, i, j, k;
1160 	enum _ecore_status_t rc;
1161 
1162 	size = ecore_cxt_ilt_shadow_size(clients);
1163 	p_mngr->ilt_shadow = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL,
1164 					 size * sizeof(struct ecore_dma_mem));
1165 
1166 	if (p_mngr->ilt_shadow == OSAL_NULL) {
1167 		DP_NOTICE(p_hwfn, false, "Failed to allocate ilt shadow table\n");
1168 		rc = ECORE_NOMEM;
1169 		goto ilt_shadow_fail;
1170 	}
1171 
1172 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1173 		   "Allocated 0x%x bytes for ilt shadow\n",
1174 		   (u32)(size * sizeof(struct ecore_dma_mem)));
1175 
1176 	for_each_ilt_valid_client(i, clients) {
1177 		for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1178 			p_blk = &clients[i].pf_blks[j];
1179 			rc = ecore_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1180 			if (rc != ECORE_SUCCESS)
1181 				goto ilt_shadow_fail;
1182 		}
1183 		for (k = 0; k < p_mngr->vf_count; k++) {
1184 			for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1185 				u32 lines = clients[i].vf_total_lines * k;
1186 
1187 				p_blk = &clients[i].vf_blks[j];
1188 				rc = ecore_ilt_blk_alloc(p_hwfn, p_blk,
1189 							 i, lines);
1190 				if (rc != ECORE_SUCCESS)
1191 					goto ilt_shadow_fail;
1192 			}
1193 		}
1194 	}
1195 
1196 	return ECORE_SUCCESS;
1197 
1198 ilt_shadow_fail:
1199 	ecore_ilt_shadow_free(p_hwfn);
1200 	return rc;
1201 }
1202 
1203 static void ecore_cid_map_free(struct ecore_hwfn *p_hwfn)
1204 {
1205 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1206 	u32 type, vf;
1207 
1208 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1209 		OSAL_FREE(p_hwfn->p_dev, p_mngr->acquired[type].cid_map);
1210 		p_mngr->acquired[type].cid_map = OSAL_NULL;
1211 		p_mngr->acquired[type].max_count = 0;
1212 		p_mngr->acquired[type].start_cid = 0;
1213 
1214 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1215 			OSAL_FREE(p_hwfn->p_dev,
1216 				  p_mngr->acquired_vf[type][vf].cid_map);
1217 			p_mngr->acquired_vf[type][vf].cid_map = OSAL_NULL;
1218 			p_mngr->acquired_vf[type][vf].max_count = 0;
1219 			p_mngr->acquired_vf[type][vf].start_cid = 0;
1220 		}
1221 	}
1222 }
1223 
1224 static enum _ecore_status_t
1225 ecore_cid_map_alloc_single(struct ecore_hwfn *p_hwfn, u32 type,
1226 			   u32 cid_start, u32 cid_count,
1227 			   struct ecore_cid_acquired_map *p_map)
1228 {
1229 	u32 size;
1230 
1231 	if (!cid_count)
1232 		return ECORE_SUCCESS;
1233 
1234 	size = MAP_WORD_SIZE * DIV_ROUND_UP(cid_count, BITS_PER_MAP_WORD);
1235 	p_map->cid_map = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, size);
1236 	if (p_map->cid_map == OSAL_NULL)
1237 		return ECORE_NOMEM;
1238 
1239 	p_map->max_count = cid_count;
1240 	p_map->start_cid = cid_start;
1241 
1242 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
1243 		   "Type %08x start: %08x count %08x\n",
1244 		   type, p_map->start_cid, p_map->max_count);
1245 
1246 	return ECORE_SUCCESS;
1247 }
1248 
1249 static enum _ecore_status_t ecore_cid_map_alloc(struct ecore_hwfn *p_hwfn)
1250 {
1251 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1252 	u32 start_cid = 0, vf_start_cid = 0;
1253 	u32 type, vf;
1254 
1255 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1256 		struct ecore_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[type];
1257 		struct ecore_cid_acquired_map *p_map;
1258 
1259 		/* Handle PF maps */
1260 		p_map = &p_mngr->acquired[type];
1261 		if (ecore_cid_map_alloc_single(p_hwfn, type, start_cid,
1262 					       p_cfg->cid_count, p_map))
1263 			goto cid_map_fail;
1264 
1265 		/* Handle VF maps */
1266 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1267 			p_map = &p_mngr->acquired_vf[type][vf];
1268 			if (ecore_cid_map_alloc_single(p_hwfn, type,
1269 						       vf_start_cid,
1270 						       p_cfg->cids_per_vf,
1271 						       p_map))
1272 				goto cid_map_fail;
1273 		}
1274 
1275 		start_cid += p_cfg->cid_count;
1276 		vf_start_cid += p_cfg->cids_per_vf;
1277 	}
1278 
1279 	return ECORE_SUCCESS;
1280 
1281 cid_map_fail:
1282 	ecore_cid_map_free(p_hwfn);
1283 	return ECORE_NOMEM;
1284 }
1285 
1286 enum _ecore_status_t ecore_cxt_mngr_alloc(struct ecore_hwfn *p_hwfn)
1287 {
1288 	struct ecore_ilt_client_cfg *clients;
1289 	struct ecore_cxt_mngr *p_mngr;
1290 	u32 i;
1291 
1292 	p_mngr = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, sizeof(*p_mngr));
1293 	if (!p_mngr) {
1294 		DP_NOTICE(p_hwfn, false, "Failed to allocate `struct ecore_cxt_mngr'\n");
1295 		return ECORE_NOMEM;
1296 	}
1297 
1298 	/* Initialize ILT client registers */
1299 	clients = p_mngr->clients;
1300 	clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1301 	clients[ILT_CLI_CDUC].last.reg  = ILT_CFG_REG(CDUC, LAST_ILT);
1302 	clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1303 
1304 	clients[ILT_CLI_QM].first.reg   = ILT_CFG_REG(QM, FIRST_ILT);
1305 	clients[ILT_CLI_QM].last.reg    = ILT_CFG_REG(QM, LAST_ILT);
1306 	clients[ILT_CLI_QM].p_size.reg  = ILT_CFG_REG(QM, P_SIZE);
1307 
1308 	clients[ILT_CLI_TM].first.reg   = ILT_CFG_REG(TM, FIRST_ILT);
1309 	clients[ILT_CLI_TM].last.reg    = ILT_CFG_REG(TM, LAST_ILT);
1310 	clients[ILT_CLI_TM].p_size.reg  = ILT_CFG_REG(TM, P_SIZE);
1311 
1312 	clients[ILT_CLI_SRC].first.reg  = ILT_CFG_REG(SRC, FIRST_ILT);
1313 	clients[ILT_CLI_SRC].last.reg   = ILT_CFG_REG(SRC, LAST_ILT);
1314 	clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1315 
1316 	clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1317 	clients[ILT_CLI_CDUT].last.reg  = ILT_CFG_REG(CDUT, LAST_ILT);
1318 	clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1319 
1320 	clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1321 	clients[ILT_CLI_TSDM].last.reg  = ILT_CFG_REG(TSDM, LAST_ILT);
1322 	clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1323 
1324 	/* default ILT page size for all clients is 64K */
1325 	for (i = 0; i < ILT_CLI_MAX; i++)
1326 		p_mngr->clients[i].p_size.val = p_hwfn->p_dev->ilt_page_size;
1327 
1328 	/* Initialize task sizes */
1329 	p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1330 	p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1331 
1332 	if (p_hwfn->p_dev->p_iov_info)
1333 		p_mngr->vf_count = p_hwfn->p_dev->p_iov_info->total_vfs;
1334 
1335 	/* Initialize the dynamic ILT allocation mutex */
1336 #ifdef CONFIG_ECORE_LOCK_ALLOC
1337 	OSAL_MUTEX_ALLOC(p_hwfn, &p_mngr->mutex);
1338 #endif
1339 	OSAL_MUTEX_INIT(&p_mngr->mutex);
1340 
1341 	/* Set the cxt mangr pointer priori to further allocations */
1342 	p_hwfn->p_cxt_mngr = p_mngr;
1343 
1344 	return ECORE_SUCCESS;
1345 }
1346 
1347 enum _ecore_status_t ecore_cxt_tables_alloc(struct ecore_hwfn *p_hwfn)
1348 {
1349 	enum _ecore_status_t    rc;
1350 
1351 	/* Allocate the ILT shadow table */
1352 	rc = ecore_ilt_shadow_alloc(p_hwfn);
1353 	if (rc) {
1354 		DP_NOTICE(p_hwfn, false, "Failed to allocate ilt memory\n");
1355 		goto tables_alloc_fail;
1356 	}
1357 
1358 	/* Allocate the T2  table */
1359 	rc = ecore_cxt_src_t2_alloc(p_hwfn);
1360 	if (rc) {
1361 		DP_NOTICE(p_hwfn, false, "Failed to allocate T2 memory\n");
1362 		goto tables_alloc_fail;
1363 	}
1364 
1365 	/* Allocate and initialize the acquired cids bitmaps */
1366 	rc = ecore_cid_map_alloc(p_hwfn);
1367 	if (rc) {
1368 		DP_NOTICE(p_hwfn, false, "Failed to allocate cid maps\n");
1369 		goto tables_alloc_fail;
1370 	}
1371 
1372 	return ECORE_SUCCESS;
1373 
1374 tables_alloc_fail:
1375 	ecore_cxt_mngr_free(p_hwfn);
1376 	return rc;
1377 }
1378 void ecore_cxt_mngr_free(struct ecore_hwfn *p_hwfn)
1379 {
1380 	if (!p_hwfn->p_cxt_mngr)
1381 		return;
1382 
1383 	ecore_cid_map_free(p_hwfn);
1384 	ecore_cxt_src_t2_free(p_hwfn);
1385 	ecore_ilt_shadow_free(p_hwfn);
1386 #ifdef CONFIG_ECORE_LOCK_ALLOC
1387 	OSAL_MUTEX_DEALLOC(&p_hwfn->p_cxt_mngr->mutex);
1388 #endif
1389 	OSAL_FREE(p_hwfn->p_dev, p_hwfn->p_cxt_mngr);
1390 
1391 	p_hwfn->p_cxt_mngr = OSAL_NULL;
1392 }
1393 
1394 void ecore_cxt_mngr_setup(struct ecore_hwfn *p_hwfn)
1395 {
1396 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1397 	struct ecore_cid_acquired_map *p_map;
1398 	struct ecore_conn_type_cfg *p_cfg;
1399 	int type;
1400 	u32 len;
1401 
1402 	/* Reset acquired cids */
1403 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1404 		u32 vf;
1405 
1406 		p_cfg = &p_mngr->conn_cfg[type];
1407 		if (p_cfg->cid_count) {
1408 			p_map = &p_mngr->acquired[type];
1409 			len = DIV_ROUND_UP(p_map->max_count,
1410 					   BITS_PER_MAP_WORD) *
1411 			      MAP_WORD_SIZE;
1412 			OSAL_MEM_ZERO(p_map->cid_map, len);
1413 		}
1414 
1415 		if (!p_cfg->cids_per_vf)
1416 			continue;
1417 
1418 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1419 			p_map = &p_mngr->acquired_vf[type][vf];
1420 			len = DIV_ROUND_UP(p_map->max_count,
1421 					   BITS_PER_MAP_WORD) *
1422 			      MAP_WORD_SIZE;
1423 			OSAL_MEM_ZERO(p_map->cid_map, len);
1424 		}
1425 	}
1426 }
1427 
1428 /* HW initialization helper (per Block, per phase) */
1429 
1430 /* CDU Common */
1431 #define CDUC_CXT_SIZE_SHIFT						\
1432 	CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1433 
1434 #define CDUC_CXT_SIZE_MASK						\
1435 	(CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1436 
1437 #define CDUC_BLOCK_WASTE_SHIFT						\
1438 	CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1439 
1440 #define CDUC_BLOCK_WASTE_MASK						\
1441 	(CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1442 
1443 #define CDUC_NCIB_SHIFT							\
1444 	CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1445 
1446 #define CDUC_NCIB_MASK							\
1447 	(CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1448 
1449 #define CDUT_TYPE0_CXT_SIZE_SHIFT					\
1450 	CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1451 
1452 #define CDUT_TYPE0_CXT_SIZE_MASK					\
1453 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >>				\
1454 	CDUT_TYPE0_CXT_SIZE_SHIFT)
1455 
1456 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT					\
1457 	CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1458 
1459 #define CDUT_TYPE0_BLOCK_WASTE_MASK					\
1460 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >>			\
1461 	CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1462 
1463 #define CDUT_TYPE0_NCIB_SHIFT						\
1464 	CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1465 
1466 #define CDUT_TYPE0_NCIB_MASK						\
1467 	(CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >>		\
1468 	CDUT_TYPE0_NCIB_SHIFT)
1469 
1470 #define CDUT_TYPE1_CXT_SIZE_SHIFT					\
1471 	CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1472 
1473 #define CDUT_TYPE1_CXT_SIZE_MASK					\
1474 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >>				\
1475 	CDUT_TYPE1_CXT_SIZE_SHIFT)
1476 
1477 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT					\
1478 	CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1479 
1480 #define CDUT_TYPE1_BLOCK_WASTE_MASK					\
1481 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >>			\
1482 	CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1483 
1484 #define CDUT_TYPE1_NCIB_SHIFT						\
1485 	CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1486 
1487 #define CDUT_TYPE1_NCIB_MASK						\
1488 	(CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >>		\
1489 	CDUT_TYPE1_NCIB_SHIFT)
1490 
1491 static void ecore_cdu_init_common(struct ecore_hwfn *p_hwfn)
1492 {
1493 	u32 page_sz, elems_per_page, block_waste,  cxt_size, cdu_params = 0;
1494 
1495 	/* CDUC - connection configuration */
1496 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1497 	cxt_size = CONN_CXT_SIZE(p_hwfn);
1498 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1499 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1500 
1501 	SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1502 	SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1503 	SET_FIELD(cdu_params, (u32)CDUC_NCIB, elems_per_page);
1504 	STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1505 
1506 	/* CDUT - type-0 tasks configuration */
1507 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1508 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1509 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1510 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1511 
1512 	/* cxt size and block-waste are multipes of 8 */
1513 	cdu_params = 0;
1514 	SET_FIELD(cdu_params, (u32)CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1515 	SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1516 	SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1517 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1518 
1519 	/* CDUT - type-1 tasks configuration */
1520 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1521 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1522 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1523 
1524 	/* cxt size and block-waste are multipes of 8 */
1525 	cdu_params = 0;
1526 	SET_FIELD(cdu_params, (u32)CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1527 	SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1528 	SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1529 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1530 }
1531 
1532 /* CDU PF */
1533 #define CDU_SEG_REG_TYPE_SHIFT		CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1534 #define CDU_SEG_REG_TYPE_MASK		0x1
1535 #define CDU_SEG_REG_OFFSET_SHIFT	0
1536 #define CDU_SEG_REG_OFFSET_MASK		CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1537 
1538 static void ecore_cdu_init_pf(struct ecore_hwfn *p_hwfn)
1539 {
1540 	struct ecore_ilt_client_cfg *p_cli;
1541 	struct ecore_tid_seg *p_seg;
1542 	u32 cdu_seg_params, offset;
1543 	int i;
1544 
1545 	static const u32 rt_type_offset_arr[] = {
1546 		CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1547 		CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1548 		CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1549 		CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1550 	};
1551 
1552 	static const u32 rt_type_offset_fl_arr[] = {
1553 		CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1554 		CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1555 		CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1556 		CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1557 	};
1558 
1559 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1560 
1561 	/* There are initializations only for CDUT during pf Phase */
1562 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1563 		/* Segment 0*/
1564 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
1565 		if (!p_seg)
1566 			continue;
1567 
1568 		/* Note: start_line is already adjusted for the CDU
1569 		 * segment register granularity, so we just need to
1570 		 * divide. Adjustment is implicit as we assume ILT
1571 		 * Page size is larger than 32K!
1572 		 */
1573 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1574 			 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1575 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1576 
1577 		cdu_seg_params = 0;
1578 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1579 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1580 		STORE_RT_REG(p_hwfn, rt_type_offset_arr[i],
1581 			     cdu_seg_params);
1582 
1583 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1584 			 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1585 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1586 
1587 		cdu_seg_params = 0;
1588 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1589 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1590 		STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i],
1591 			     cdu_seg_params);
1592 	}
1593 }
1594 
1595 void ecore_qm_init_pf(struct ecore_hwfn *p_hwfn, struct ecore_ptt *p_ptt,
1596 		      bool is_pf_loading)
1597 {
1598 	struct ecore_qm_info *qm_info = &p_hwfn->qm_info;
1599 	struct ecore_mcp_link_state *p_link;
1600 	struct ecore_qm_iids iids;
1601 
1602 	OSAL_MEM_ZERO(&iids, sizeof(iids));
1603 	ecore_cxt_qm_iids(p_hwfn, &iids);
1604 
1605 	p_link = &ECORE_LEADING_HWFN(p_hwfn->p_dev)->mcp_info->link_output;
1606 
1607 	ecore_qm_pf_rt_init(p_hwfn, p_ptt, p_hwfn->port_id,
1608 			    p_hwfn->rel_pf_id, qm_info->max_phys_tcs_per_port,
1609 			    is_pf_loading,
1610 			    iids.cids, iids.vf_cids, iids.tids,
1611 			    qm_info->start_pq,
1612 			    qm_info->num_pqs - qm_info->num_vf_pqs,
1613 			    qm_info->num_vf_pqs,
1614 			    qm_info->start_vport,
1615 			    qm_info->num_vports, qm_info->pf_wfq,
1616 			    qm_info->pf_rl, p_link->speed,
1617 			    p_hwfn->qm_info.qm_pq_params,
1618 			    p_hwfn->qm_info.qm_vport_params);
1619 }
1620 
1621 /* CM PF */
1622 static void ecore_cm_init_pf(struct ecore_hwfn *p_hwfn)
1623 {
1624 	STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET, ecore_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1625 }
1626 
1627 /* DQ PF */
1628 static void ecore_dq_init_pf(struct ecore_hwfn *p_hwfn)
1629 {
1630 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1631 	u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1632 
1633 	dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1634 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1635 
1636 	dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1637 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1638 
1639 	dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1640 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1641 
1642 	dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1643 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1644 
1645 	dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1646 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1647 
1648 	dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1649 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1650 
1651 	dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1652 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1653 
1654 	dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1655 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1656 
1657 	dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1658 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1659 
1660 	dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1661 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1662 
1663 	dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1664 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1665 
1666 	dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1667 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1668 
1669 	/* Connection types 6 & 7 are not in use, yet they must be configured
1670 	 * as the highest possible connection. Not configuring them means the
1671 	 * defaults will be  used, and with a large number of cids a bug may
1672 	 * occur, if the defaults will be smaller than dq_pf_max_cid /
1673 	 * dq_vf_max_cid.
1674 	 */
1675 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1676 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1677 
1678 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1679 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1680 }
1681 
1682 static void ecore_ilt_bounds_init(struct ecore_hwfn *p_hwfn)
1683 {
1684 	struct ecore_ilt_client_cfg *ilt_clients;
1685 	int i;
1686 
1687 	ilt_clients = p_hwfn->p_cxt_mngr->clients;
1688 	for_each_ilt_valid_client(i, ilt_clients) {
1689 		STORE_RT_REG(p_hwfn,
1690 			     ilt_clients[i].first.reg,
1691 			     ilt_clients[i].first.val);
1692 		STORE_RT_REG(p_hwfn,
1693 			     ilt_clients[i].last.reg,
1694 			     ilt_clients[i].last.val);
1695 		STORE_RT_REG(p_hwfn,
1696 			     ilt_clients[i].p_size.reg,
1697 			     ilt_clients[i].p_size.val);
1698 	}
1699 }
1700 
1701 static void ecore_ilt_vf_bounds_init(struct ecore_hwfn *p_hwfn)
1702 {
1703 	struct ecore_ilt_client_cfg *p_cli;
1704 	u32 blk_factor;
1705 
1706 	/* For simplicty  we set the 'block' to be an ILT page */
1707 	if (p_hwfn->p_dev->p_iov_info) {
1708 		struct ecore_hw_sriov_info *p_iov = p_hwfn->p_dev->p_iov_info;
1709 
1710 		STORE_RT_REG(p_hwfn,
1711 			     PSWRQ2_REG_VF_BASE_RT_OFFSET,
1712 			     p_iov->first_vf_in_pf);
1713 		STORE_RT_REG(p_hwfn,
1714 			     PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1715 			     p_iov->first_vf_in_pf + p_iov->total_vfs);
1716 	}
1717 
1718 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1719 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1720 	if (p_cli->active) {
1721 		STORE_RT_REG(p_hwfn,
1722 			     PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1723 			     blk_factor);
1724 		STORE_RT_REG(p_hwfn,
1725 			     PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1726 			     p_cli->pf_total_lines);
1727 		STORE_RT_REG(p_hwfn,
1728 			     PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1729 			     p_cli->vf_total_lines);
1730 	}
1731 
1732 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1733 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1734 	if (p_cli->active) {
1735 		STORE_RT_REG(p_hwfn,
1736 			     PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1737 			     blk_factor);
1738 		STORE_RT_REG(p_hwfn,
1739 			     PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1740 			     p_cli->pf_total_lines);
1741 		STORE_RT_REG(p_hwfn,
1742 			     PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1743 			     p_cli->vf_total_lines);
1744 	}
1745 
1746 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1747 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1748 	if (p_cli->active) {
1749 		STORE_RT_REG(p_hwfn,
1750 			     PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET,
1751 			     blk_factor);
1752 		STORE_RT_REG(p_hwfn,
1753 			     PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1754 			     p_cli->pf_total_lines);
1755 		STORE_RT_REG(p_hwfn,
1756 			     PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1757 			     p_cli->vf_total_lines);
1758 	}
1759 }
1760 
1761 /* ILT (PSWRQ2) PF */
1762 static void ecore_ilt_init_pf(struct ecore_hwfn *p_hwfn)
1763 {
1764 	struct ecore_ilt_client_cfg *clients;
1765 	struct ecore_cxt_mngr *p_mngr;
1766 	struct ecore_dma_mem *p_shdw;
1767 	u32 line, rt_offst, i;
1768 
1769 	ecore_ilt_bounds_init(p_hwfn);
1770 	ecore_ilt_vf_bounds_init(p_hwfn);
1771 
1772 	p_mngr  = p_hwfn->p_cxt_mngr;
1773 	p_shdw  = p_mngr->ilt_shadow;
1774 	clients = p_hwfn->p_cxt_mngr->clients;
1775 
1776 	for_each_ilt_valid_client(i, clients) {
1777 		/* Client's 1st val and RT array are absolute, ILT shadows'
1778 		 * lines are relative.
1779 		 */
1780 		line = clients[i].first.val - p_mngr->pf_start_line;
1781 		rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1782 			   clients[i].first.val * ILT_ENTRY_IN_REGS;
1783 
1784 		for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1785 		     line++, rt_offst += ILT_ENTRY_IN_REGS) {
1786 			u64 ilt_hw_entry = 0;
1787 
1788 			/** p_virt could be OSAL_NULL incase of dynamic
1789 			 *  allocation
1790 			 */
1791 			if (p_shdw[line].p_virt != OSAL_NULL) {
1792 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1793 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1794 					  (unsigned long long)(p_shdw[line].p_phys >> 12));
1795 
1796 				DP_VERBOSE(
1797 					p_hwfn, ECORE_MSG_ILT,
1798 					"Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1799 					rt_offst, line, i,
1800 					(unsigned long long)(p_shdw[line].p_phys >> 12));
1801 			}
1802 
1803 			STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1804 		}
1805 	}
1806 }
1807 
1808 /* SRC (Searcher) PF */
1809 static void ecore_src_init_pf(struct ecore_hwfn *p_hwfn)
1810 {
1811 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1812 	u32 rounded_conn_num, conn_num, conn_max;
1813 	struct ecore_src_iids src_iids;
1814 
1815 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
1816 	ecore_cxt_src_iids(p_mngr, &src_iids);
1817 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1818 	if (!conn_num)
1819 		return;
1820 
1821 	conn_max = OSAL_MAX_T(u32, conn_num, SRC_MIN_NUM_ELEMS);
1822 	rounded_conn_num = OSAL_ROUNDUP_POW_OF_TWO(conn_max);
1823 
1824 	STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1825 	STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1826 		     OSAL_LOG2(rounded_conn_num));
1827 
1828 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1829 			 p_hwfn->p_cxt_mngr->first_free);
1830 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1831 			 p_hwfn->p_cxt_mngr->last_free);
1832 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1833 		   "Configured SEARCHER for 0x%08x connections\n",
1834 		   conn_num);
1835 }
1836 
1837 /* Timers PF */
1838 #define TM_CFG_NUM_IDS_SHIFT		0
1839 #define TM_CFG_NUM_IDS_MASK		0xFFFFULL
1840 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT	16
1841 #define TM_CFG_PRE_SCAN_OFFSET_MASK	0x1FFULL
1842 #define TM_CFG_PARENT_PF_SHIFT		25
1843 #define TM_CFG_PARENT_PF_MASK		0x7ULL
1844 
1845 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT	30
1846 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK	0x1FFULL
1847 
1848 #define TM_CFG_TID_OFFSET_SHIFT		30
1849 #define TM_CFG_TID_OFFSET_MASK		0x7FFFFULL
1850 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT	49
1851 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK	0x1FFULL
1852 
1853 static void ecore_tm_init_pf(struct ecore_hwfn *p_hwfn)
1854 {
1855 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1856 	u32 active_seg_mask = 0, tm_offset, rt_reg;
1857 	struct ecore_tm_iids tm_iids;
1858 	u64 cfg_word;
1859 	u8 i;
1860 
1861 	OSAL_MEM_ZERO(&tm_iids, sizeof(tm_iids));
1862 	ecore_cxt_tm_iids(p_mngr, &tm_iids);
1863 
1864 	/* @@@TBD No pre-scan for now */
1865 
1866 	/* Note: We assume consecutive VFs for a PF */
1867 	for (i = 0; i < p_mngr->vf_count; i++) {
1868 		cfg_word = 0;
1869 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1870 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1871 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1872 		SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1873 
1874 		rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1875 			 (sizeof(cfg_word) / sizeof(u32)) *
1876 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1877 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1878 	}
1879 
1880 	cfg_word = 0;
1881 	SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1882 	SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1883 	SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);	  /* n/a for PF */
1884 	SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all   */
1885 
1886 	rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1887 		 (sizeof(cfg_word) / sizeof(u32)) *
1888 		 (NUM_OF_VFS(p_hwfn->p_dev) + p_hwfn->rel_pf_id);
1889 	STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1890 
1891 	/* enale scan */
1892 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1893 		     tm_iids.pf_cids  ? 0x1 : 0x0);
1894 
1895 	/* @@@TBD how to enable the scan for the VFs */
1896 
1897 	tm_offset = tm_iids.per_vf_cids;
1898 
1899 	/* Note: We assume consecutive VFs for a PF */
1900 	for (i = 0; i < p_mngr->vf_count; i++) {
1901 		cfg_word = 0;
1902 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1903 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1904 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1905 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1906 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1907 
1908 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1909 			 (sizeof(cfg_word) / sizeof(u32)) *
1910 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1911 
1912 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1913 	}
1914 
1915 	tm_offset = tm_iids.pf_cids;
1916 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1917 		cfg_word = 0;
1918 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1919 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1920 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1921 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1922 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1923 
1924 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1925 			 (sizeof(cfg_word) / sizeof(u32)) *
1926 			 (NUM_OF_VFS(p_hwfn->p_dev) +
1927 			 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1928 
1929 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1930 		active_seg_mask |= (tm_iids.pf_tids[i] ? (1 << i) : 0);
1931 
1932 		tm_offset += tm_iids.pf_tids[i];
1933 	}
1934 
1935 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn))
1936 		active_seg_mask = 0;
1937 
1938 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1939 
1940 	/* @@@TBD how to enable the scan for the VFs */
1941 }
1942 
1943 static void ecore_prs_init_common(struct ecore_hwfn *p_hwfn)
1944 {
1945 	if ((p_hwfn->hw_info.personality == ECORE_PCI_FCOE) &&
1946 	    p_hwfn->pf_params.fcoe_pf_params.is_target)
1947 		STORE_RT_REG(p_hwfn, PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1948 }
1949 
1950 static void ecore_prs_init_pf(struct ecore_hwfn *p_hwfn)
1951 {
1952 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1953 	struct ecore_conn_type_cfg *p_fcoe;
1954 	struct ecore_tid_seg *p_tid;
1955 
1956 	p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1957 
1958 	/* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1959 	if (!p_fcoe->cid_count)
1960 		return;
1961 
1962 	p_tid = &p_fcoe->tid_seg[ECORE_CXT_FCOE_TID_SEG];
1963 	if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1964 		STORE_RT_REG_AGG(p_hwfn,
1965 				 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1966 				 p_tid->count);
1967 	} else {
1968 		STORE_RT_REG_AGG(p_hwfn,
1969 				PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1970 				p_tid->count);
1971 	}
1972 }
1973 
1974 void ecore_cxt_hw_init_common(struct ecore_hwfn *p_hwfn)
1975 {
1976 	/* CDU configuration */
1977 	ecore_cdu_init_common(p_hwfn);
1978 	ecore_prs_init_common(p_hwfn);
1979 }
1980 
1981 void ecore_cxt_hw_init_pf(struct ecore_hwfn *p_hwfn, struct ecore_ptt *p_ptt)
1982 {
1983 	ecore_qm_init_pf(p_hwfn, p_ptt, true);
1984 	ecore_cm_init_pf(p_hwfn);
1985 	ecore_dq_init_pf(p_hwfn);
1986 	ecore_cdu_init_pf(p_hwfn);
1987 	ecore_ilt_init_pf(p_hwfn);
1988 	ecore_src_init_pf(p_hwfn);
1989 	ecore_tm_init_pf(p_hwfn);
1990 	ecore_prs_init_pf(p_hwfn);
1991 }
1992 
1993 enum _ecore_status_t _ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
1994 					    enum protocol_type type,
1995 					    u32 *p_cid, u8 vfid)
1996 {
1997 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1998 	struct ecore_cid_acquired_map *p_map;
1999 	u32 rel_cid;
2000 
2001 	if (type >= MAX_CONN_TYPES) {
2002 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
2003 		return ECORE_INVAL;
2004 	}
2005 
2006 	if (vfid >= COMMON_MAX_NUM_VFS && vfid != ECORE_CXT_PF_CID) {
2007 		DP_NOTICE(p_hwfn, true, "VF [%02x] is out of range\n", vfid);
2008 		return ECORE_INVAL;
2009 	}
2010 
2011 	/* Determine the right map to take this CID from */
2012 	if (vfid == ECORE_CXT_PF_CID)
2013 		p_map = &p_mngr->acquired[type];
2014 	else
2015 		p_map = &p_mngr->acquired_vf[type][vfid];
2016 
2017 	if (p_map->cid_map == OSAL_NULL) {
2018 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
2019 		return ECORE_INVAL;
2020 	}
2021 
2022 	rel_cid = OSAL_FIND_FIRST_ZERO_BIT(p_map->cid_map,
2023 					   p_map->max_count);
2024 
2025 	if (rel_cid >= p_map->max_count) {
2026 		DP_NOTICE(p_hwfn, false, "no CID available for protocol %d\n",
2027 			  type);
2028 		return ECORE_NORESOURCES;
2029 	}
2030 
2031 	OSAL_SET_BIT(rel_cid, p_map->cid_map);
2032 
2033 	*p_cid = rel_cid + p_map->start_cid;
2034 
2035 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
2036 		   "Acquired cid 0x%08x [rel. %08x] vfid %02x type %d\n",
2037 		   *p_cid, rel_cid, vfid, type);
2038 
2039 	return ECORE_SUCCESS;
2040 }
2041 
2042 enum _ecore_status_t ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
2043 					   enum protocol_type type,
2044 					   u32 *p_cid)
2045 {
2046 	return _ecore_cxt_acquire_cid(p_hwfn, type, p_cid, ECORE_CXT_PF_CID);
2047 }
2048 
2049 static bool ecore_cxt_test_cid_acquired(struct ecore_hwfn *p_hwfn,
2050 					u32 cid, u8 vfid,
2051 					enum protocol_type *p_type,
2052 					struct ecore_cid_acquired_map **pp_map)
2053 {
2054 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2055 	u32 rel_cid;
2056 
2057 	/* Iterate over protocols and find matching cid range */
2058 	for (*p_type = 0; *p_type < MAX_CONN_TYPES; (*p_type)++) {
2059 		if (vfid == ECORE_CXT_PF_CID)
2060 			*pp_map = &p_mngr->acquired[*p_type];
2061 		else
2062 			*pp_map = &p_mngr->acquired_vf[*p_type][vfid];
2063 
2064 		if (!((*pp_map)->cid_map))
2065 			continue;
2066 		if (cid >= (*pp_map)->start_cid &&
2067 		    cid < (*pp_map)->start_cid + (*pp_map)->max_count) {
2068 			break;
2069 		}
2070 	}
2071 
2072 	if (*p_type == MAX_CONN_TYPES) {
2073 		DP_NOTICE(p_hwfn, true, "Invalid CID %d vfid %02x", cid, vfid);
2074 		goto fail;
2075 	}
2076 
2077 	rel_cid = cid - (*pp_map)->start_cid;
2078 	if (!OSAL_TEST_BIT(rel_cid, (*pp_map)->cid_map)) {
2079 		DP_NOTICE(p_hwfn, true,
2080 			  "CID %d [vifd %02x] not acquired", cid, vfid);
2081 		goto fail;
2082 	}
2083 
2084 	return true;
2085 fail:
2086 	*p_type = MAX_CONN_TYPES;
2087 	*pp_map = OSAL_NULL;
2088 	return false;
2089 }
2090 
2091 void _ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid, u8 vfid)
2092 {
2093 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2094 	enum protocol_type type;
2095 	bool b_acquired;
2096 	u32 rel_cid;
2097 
2098 	if (vfid != ECORE_CXT_PF_CID && vfid > COMMON_MAX_NUM_VFS) {
2099 		DP_NOTICE(p_hwfn, true,
2100 			  "Trying to return incorrect CID belonging to VF %02x\n",
2101 			  vfid);
2102 		return;
2103 	}
2104 
2105 	/* Test acquired and find matching per-protocol map */
2106 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, cid, vfid,
2107 						 &type, &p_map);
2108 
2109 	if (!b_acquired)
2110 		return;
2111 
2112 	rel_cid = cid - p_map->start_cid;
2113 	OSAL_CLEAR_BIT(rel_cid, p_map->cid_map);
2114 
2115 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
2116 		   "Released CID 0x%08x [rel. %08x] vfid %02x type %d\n",
2117 		   cid, rel_cid, vfid, type);
2118 }
2119 
2120 void ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid)
2121 {
2122 	_ecore_cxt_release_cid(p_hwfn, cid, ECORE_CXT_PF_CID);
2123 }
2124 
2125 enum _ecore_status_t ecore_cxt_get_cid_info(struct ecore_hwfn *p_hwfn,
2126 					    struct ecore_cxt_info *p_info)
2127 {
2128 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2129 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2130 	u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
2131 	enum protocol_type type;
2132 	bool b_acquired;
2133 
2134 	/* Test acquired and find matching per-protocol map */
2135 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, p_info->iid,
2136 						 ECORE_CXT_PF_CID,
2137 						 &type, &p_map);
2138 
2139 	if (!b_acquired)
2140 		return ECORE_INVAL;
2141 
2142 	/* set the protocl type */
2143 	p_info->type = type;
2144 
2145 	/* compute context virtual pointer */
2146 	hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
2147 
2148 	conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
2149 	cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
2150 	line = p_info->iid / cxts_per_p;
2151 
2152 	/* Make sure context is allocated (dynamic allocation) */
2153 	if (!p_mngr->ilt_shadow[line].p_virt)
2154 		return ECORE_INVAL;
2155 
2156 	p_info->p_cxt = (u8 *)p_mngr->ilt_shadow[line].p_virt +
2157 			      p_info->iid % cxts_per_p * conn_cxt_size;
2158 
2159 	DP_VERBOSE(p_hwfn, (ECORE_MSG_ILT | ECORE_MSG_CXT),
2160 		   "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
2161 		   (p_info->iid / cxts_per_p), p_info->p_cxt, p_info->iid);
2162 
2163 	return ECORE_SUCCESS;
2164 }
2165 
2166 static void ecore_rdma_set_pf_params(struct ecore_hwfn *p_hwfn,
2167 				     struct ecore_rdma_pf_params *p_params,
2168 				     u32 num_tasks)
2169 {
2170 	u32 num_cons, num_qps;
2171 	enum protocol_type proto;
2172 
2173 	/* The only case RDMA personality can be overriden is if NVRAM is
2174 	 * configured with ETH_RDMA or if no rdma protocol was requested
2175 	 */
2176 	switch (p_params->rdma_protocol) {
2177 	case ECORE_RDMA_PROTOCOL_DEFAULT:
2178 		if (p_hwfn->mcp_info->func_info.protocol ==
2179 		    ECORE_PCI_ETH_RDMA) {
2180 			DP_NOTICE(p_hwfn, false,
2181 				  "Current day drivers don't support RoCE & iWARP. Default to RoCE-only\n");
2182 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2183 		}
2184 		break;
2185 	case ECORE_RDMA_PROTOCOL_NONE:
2186 		p_hwfn->hw_info.personality = ECORE_PCI_ETH;
2187 		return; /* intentional... nothing left to do... */
2188 	case ECORE_RDMA_PROTOCOL_ROCE:
2189 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2190 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2191 		break;
2192 	case ECORE_RDMA_PROTOCOL_IWARP:
2193 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2194 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_IWARP;
2195 		break;
2196 	}
2197 
2198 	switch (p_hwfn->hw_info.personality) {
2199 	case ECORE_PCI_ETH_IWARP:
2200 		/* Each QP requires one connection */
2201 		num_cons = OSAL_MIN_T(u32, IWARP_MAX_QPS, p_params->num_qps);
2202 #ifdef CONFIG_ECORE_IWARP /* required for the define */
2203 		/* additional connections required for passive tcp handling */
2204 		num_cons += ECORE_IWARP_PREALLOC_CNT;
2205 #endif
2206 		proto = PROTOCOLID_IWARP;
2207 		break;
2208 	case ECORE_PCI_ETH_ROCE:
2209 		num_qps = OSAL_MIN_T(u32, ROCE_MAX_QPS, p_params->num_qps);
2210 		num_cons = num_qps * 2; /* each QP requires two connections */
2211 		proto = PROTOCOLID_ROCE;
2212 		break;
2213 	default:
2214 		return;
2215 	}
2216 
2217 	if (num_cons && num_tasks) {
2218 		u32 num_srqs, num_xrc_srqs, max_xrc_srqs, page_size;
2219 
2220 		ecore_cxt_set_proto_cid_count(p_hwfn, proto,
2221 					      num_cons, 0);
2222 
2223 		/* Deliberatly passing ROCE for tasks id. This is because
2224 		 * iWARP / RoCE share the task id.
2225 		 */
2226 		ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
2227 					      ECORE_CXT_ROCE_TID_SEG,
2228 					      1, /* RoCE segment type */
2229 					      num_tasks,
2230 					      false); /* !force load */
2231 
2232 		num_srqs = OSAL_MIN_T(u32, ECORE_RDMA_MAX_SRQS,
2233 				      p_params->num_srqs);
2234 
2235 		/* XRC SRQs populate a single ILT page */
2236 		page_size = ecore_cxt_get_ilt_page_size(p_hwfn, ILT_CLI_TSDM);
2237 		max_xrc_srqs =  page_size / XRC_SRQ_CXT_SIZE;
2238 		max_xrc_srqs = OSAL_MIN_T(u32, max_xrc_srqs, ECORE_RDMA_MAX_XRC_SRQS);
2239 
2240 		num_xrc_srqs = OSAL_MIN_T(u32, p_params->num_xrc_srqs,
2241 					  max_xrc_srqs);
2242 		ecore_cxt_set_srq_count(p_hwfn, num_srqs, num_xrc_srqs);
2243 
2244 	} else {
2245 		DP_INFO(p_hwfn->p_dev,
2246 			"RDMA personality used without setting params!\n");
2247 	}
2248 }
2249 
2250 enum _ecore_status_t ecore_cxt_set_pf_params(struct ecore_hwfn *p_hwfn,
2251 					     u32 rdma_tasks)
2252 {
2253 	/* Set the number of required CORE connections */
2254 	u32 core_cids = 1; /* SPQ */
2255 
2256 	if (p_hwfn->using_ll2)
2257 		core_cids += 4; /* @@@TBD Use the proper #define */
2258 
2259 	ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
2260 
2261 	switch (p_hwfn->hw_info.personality) {
2262 	case ECORE_PCI_ETH_RDMA:
2263 	case ECORE_PCI_ETH_IWARP:
2264 	case ECORE_PCI_ETH_ROCE:
2265 	{
2266 		ecore_rdma_set_pf_params(p_hwfn,
2267 					 &p_hwfn->pf_params.rdma_pf_params,
2268 					 rdma_tasks);
2269 
2270 		/* no need for break since RoCE coexist with Ethernet */
2271 	}
2272 	case ECORE_PCI_ETH:
2273 	{
2274 		u32 count = 0;
2275 
2276 		struct ecore_eth_pf_params *p_params =
2277 					&p_hwfn->pf_params.eth_pf_params;
2278 
2279 		if (!p_params->num_vf_cons)
2280 			p_params->num_vf_cons = ETH_PF_PARAMS_VF_CONS_DEFAULT;
2281 		ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
2282 					      p_params->num_cons,
2283 					      p_params->num_vf_cons);
2284 
2285 		count = p_params->num_arfs_filters;
2286 
2287 		if (!OSAL_TEST_BIT(ECORE_MF_DISABLE_ARFS,
2288 				   &p_hwfn->p_dev->mf_bits))
2289 			p_hwfn->p_cxt_mngr->arfs_count = count;
2290 
2291 		break;
2292 	}
2293 	case ECORE_PCI_FCOE:
2294 	{
2295 		struct ecore_fcoe_pf_params *p_params;
2296 
2297 		p_params = &p_hwfn->pf_params.fcoe_pf_params;
2298 
2299 		if (p_params->num_cons && p_params->num_tasks) {
2300 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_FCOE,
2301 						      p_params->num_cons, 0);
2302 
2303 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
2304 						      ECORE_CXT_FCOE_TID_SEG,
2305 						      0, /* segment type */
2306 						      p_params->num_tasks,
2307 						      true);
2308 		} else {
2309 			DP_INFO(p_hwfn->p_dev,
2310 				"Fcoe personality used without setting params!\n");
2311 		}
2312 		break;
2313 	}
2314 	case ECORE_PCI_ISCSI:
2315 	{
2316 		struct ecore_iscsi_pf_params *p_params;
2317 
2318 		p_params = &p_hwfn->pf_params.iscsi_pf_params;
2319 
2320 		if (p_params->num_cons && p_params->num_tasks) {
2321 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ISCSI,
2322 						      p_params->num_cons, 0);
2323 
2324 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ISCSI,
2325 						      ECORE_CXT_ISCSI_TID_SEG,
2326 						      0, /* segment type */
2327 						      p_params->num_tasks,
2328 						      true);
2329 		} else {
2330 			DP_INFO(p_hwfn->p_dev,
2331 				"Iscsi personality used without setting params!\n");
2332 		}
2333 		break;
2334 	}
2335 	default:
2336 		return ECORE_INVAL;
2337 	}
2338 
2339 	return ECORE_SUCCESS;
2340 }
2341 
2342 enum _ecore_status_t ecore_cxt_get_tid_mem_info(struct ecore_hwfn *p_hwfn,
2343 						struct ecore_tid_mem *p_info)
2344 {
2345 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2346 	u32 proto, seg, total_lines, i, shadow_line;
2347 	struct ecore_ilt_client_cfg *p_cli;
2348 	struct ecore_ilt_cli_blk *p_fl_seg;
2349 	struct ecore_tid_seg *p_seg_info;
2350 
2351 	/* Verify the personality */
2352 	switch (p_hwfn->hw_info.personality) {
2353 	case ECORE_PCI_FCOE:
2354 		proto = PROTOCOLID_FCOE;
2355 		seg = ECORE_CXT_FCOE_TID_SEG;
2356 		break;
2357 	case ECORE_PCI_ISCSI:
2358 		proto = PROTOCOLID_ISCSI;
2359 		seg = ECORE_CXT_ISCSI_TID_SEG;
2360 		break;
2361 	default:
2362 		return ECORE_INVAL;
2363 	}
2364 
2365 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2366 	if (!p_cli->active) {
2367 		return ECORE_INVAL;
2368 	}
2369 
2370 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2371 	if (!p_seg_info->has_fl_mem)
2372 		return ECORE_INVAL;
2373 
2374 	p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2375 	total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2376 				   p_fl_seg->real_size_in_page);
2377 
2378 	for (i = 0; i < total_lines; i++) {
2379 		shadow_line = i + p_fl_seg->start_line -
2380 			      p_hwfn->p_cxt_mngr->pf_start_line;
2381 		p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
2382 	}
2383 	p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2384 			p_fl_seg->real_size_in_page;
2385 	p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2386 	p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2387 				     p_info->tid_size;
2388 
2389 	return ECORE_SUCCESS;
2390 }
2391 
2392 /* This function is very RoCE oriented, if another protocol in the future
2393  * will want this feature we'll need to modify the function to be more generic
2394  */
2395 enum _ecore_status_t
2396 ecore_cxt_dynamic_ilt_alloc(struct ecore_hwfn *p_hwfn,
2397 			    enum ecore_cxt_elem_type elem_type,
2398 			    u32 iid)
2399 {
2400 	u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2401 	struct ecore_ilt_client_cfg *p_cli;
2402 	struct ecore_ilt_cli_blk *p_blk;
2403 	struct ecore_ptt *p_ptt;
2404 	dma_addr_t p_phys;
2405 	u64 ilt_hw_entry;
2406 	void *p_virt;
2407 	enum _ecore_status_t rc = ECORE_SUCCESS;
2408 
2409 	switch (elem_type) {
2410 	case ECORE_ELEM_CXT:
2411 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2412 		elem_size = CONN_CXT_SIZE(p_hwfn);
2413 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2414 		break;
2415 	case ECORE_ELEM_SRQ:
2416 		/* The first ILT page is not used for regular SRQs. Skip it. */
2417 		iid += ecore_cxt_srqs_per_page(p_hwfn);
2418 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2419 		elem_size = SRQ_CXT_SIZE;
2420 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2421 		break;
2422 	case ECORE_ELEM_XRC_SRQ:
2423 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2424 		elem_size = XRC_SRQ_CXT_SIZE;
2425 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2426 		break;
2427 	case ECORE_ELEM_TASK:
2428 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2429 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2430 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2431 		break;
2432 	default:
2433 		DP_NOTICE(p_hwfn, false,
2434 			  "ECORE_INVALID elem type = %d", elem_type);
2435 		return ECORE_INVAL;
2436 	}
2437 
2438 	/* Calculate line in ilt */
2439 	hw_p_size = p_cli->p_size.val;
2440 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2441 	line = p_blk->start_line + (iid / elems_per_p);
2442 	shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2443 
2444 	/* If line is already allocated, do nothing, otherwise allocate it and
2445 	 * write it to the PSWRQ2 registers.
2446 	 * This section can be run in parallel from different contexts and thus
2447 	 * a mutex protection is needed.
2448 	 */
2449 #ifdef _NTDDK_
2450 #pragma warning(suppress : 28121)
2451 #endif
2452 	OSAL_MUTEX_ACQUIRE(&p_hwfn->p_cxt_mngr->mutex);
2453 
2454 	if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2455 		goto out0;
2456 
2457 	p_ptt = ecore_ptt_acquire(p_hwfn);
2458 	if (!p_ptt) {
2459 		DP_NOTICE(p_hwfn, false,
2460 			  "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2461 		rc = ECORE_TIMEOUT;
2462 		goto out0;
2463 	}
2464 
2465 	p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
2466 					 &p_phys,
2467 					 p_blk->real_size_in_page);
2468 	if (!p_virt) {
2469 		rc = ECORE_NOMEM;
2470 		goto out1;
2471 	}
2472 	OSAL_MEM_ZERO(p_virt, p_blk->real_size_in_page);
2473 
2474 	/* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2475 	 * to compensate for a HW bug, but it is configured even if DIF is not
2476 	 * enabled. This is harmless and allows us to avoid a dedicated API. We
2477 	 * configure the field for all of the contexts on the newly allocated
2478 	 * page.
2479 	 */
2480 	if (elem_type == ECORE_ELEM_TASK) {
2481 		u32 elem_i;
2482 		u8 *elem_start = (u8 *)p_virt;
2483 		union type1_task_context *elem;
2484 
2485 		for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2486 			elem = (union type1_task_context *)elem_start;
2487 			SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2488 				  TDIF_TASK_CONTEXT_REF_TAG_MASK , 0xf);
2489 			elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2490 		}
2491 	}
2492 
2493 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2494 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2495 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2496 		p_blk->real_size_in_page;
2497 
2498 	/* compute absolute offset */
2499 	reg_offset = PSWRQ2_REG_ILT_MEMORY +
2500 		     (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2501 
2502 	ilt_hw_entry = 0;
2503 	SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2504 	SET_FIELD(ilt_hw_entry,
2505 		  ILT_ENTRY_PHY_ADDR,
2506 		  (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2507 
2508 	/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2509 	ecore_dmae_host2grc(p_hwfn, p_ptt, (u64)(osal_uintptr_t)&ilt_hw_entry,
2510 			    reg_offset, sizeof(ilt_hw_entry) / sizeof(u32),
2511 			    OSAL_NULL /* default parameters */);
2512 
2513 	if (elem_type == ECORE_ELEM_CXT) {
2514 		u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2515 					 elems_per_p;
2516 
2517 		/* Update the relevant register in the parser */
2518 		ecore_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2519 			 last_cid_allocated - 1);
2520 
2521 		/* RoCE w/a -> we don't write to the prs search reg until first
2522 		 * cid is allocated. This is because the prs checks
2523 		 * last_cid-1 >=0 making 0 a valid value... this will cause
2524 		 * the a context load to occur on a RoCE packet received with
2525 		 * cid=0 even before context was initialized, can happen with a
2526 		 * stray packet from switch or a packet with crc-error
2527 		 */
2528 
2529 		if (!p_hwfn->b_rdma_enabled_in_prs) {
2530 			/* Enable Rdma search */
2531 			ecore_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2532 			p_hwfn->b_rdma_enabled_in_prs = true;
2533 		}
2534 	}
2535 
2536 out1:
2537 	ecore_ptt_release(p_hwfn, p_ptt);
2538 out0:
2539 	OSAL_MUTEX_RELEASE(&p_hwfn->p_cxt_mngr->mutex);
2540 
2541 	return rc;
2542 }
2543 
2544 /* This function is very RoCE oriented, if another protocol in the future
2545  * will want this feature we'll need to modify the function to be more generic
2546  */
2547 enum _ecore_status_t
2548 ecore_cxt_free_ilt_range(struct ecore_hwfn *p_hwfn,
2549 			 enum ecore_cxt_elem_type elem_type,
2550 			 u32 start_iid, u32 count)
2551 {
2552 	u32 start_line, end_line, shadow_start_line, shadow_end_line;
2553 	u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2554 	struct ecore_ilt_client_cfg *p_cli;
2555 	struct ecore_ilt_cli_blk *p_blk;
2556 	u32 end_iid = start_iid + count;
2557 	struct ecore_ptt *p_ptt;
2558 	u64 ilt_hw_entry = 0;
2559 	u32 i;
2560 
2561 	switch (elem_type) {
2562 	case ECORE_ELEM_CXT:
2563 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2564 		elem_size = CONN_CXT_SIZE(p_hwfn);
2565 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2566 		break;
2567 	case ECORE_ELEM_SRQ:
2568 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2569 		elem_size = SRQ_CXT_SIZE;
2570 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2571 		break;
2572 	case ECORE_ELEM_TASK:
2573 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2574 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2575 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2576 		break;
2577 	default:
2578 		DP_NOTICE(p_hwfn, false,
2579 			  "ECORE_INVALID elem type = %d", elem_type);
2580 		return ECORE_INVAL;
2581 	}
2582 
2583 	/* Calculate line in ilt */
2584 	hw_p_size = p_cli->p_size.val;
2585 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2586 	start_line = p_blk->start_line + (start_iid / elems_per_p);
2587 	end_line = p_blk->start_line + (end_iid / elems_per_p);
2588 	if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2589 		end_line--;
2590 
2591 	shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2592 	shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2593 
2594 	p_ptt = ecore_ptt_acquire(p_hwfn);
2595 	if (!p_ptt) {
2596 		DP_NOTICE(p_hwfn, false, "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2597 		return ECORE_TIMEOUT;
2598 	}
2599 
2600 	for (i = shadow_start_line; i < shadow_end_line; i++) {
2601 		if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2602 			continue;
2603 
2604 		OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
2605 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2606 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys,
2607 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].size);
2608 
2609 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = OSAL_NULL;
2610 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2611 		p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2612 
2613 		/* compute absolute offset */
2614 		reg_offset = PSWRQ2_REG_ILT_MEMORY +
2615 			     ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2616 			      ILT_ENTRY_IN_REGS);
2617 
2618 		/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2619 		 * wide-bus.
2620 		 */
2621 		ecore_dmae_host2grc(p_hwfn, p_ptt,
2622 				    (u64)(osal_uintptr_t)&ilt_hw_entry,
2623 				    reg_offset,
2624 				    sizeof(ilt_hw_entry) / sizeof(u32),
2625 				    OSAL_NULL /* default parameters */);
2626 	}
2627 
2628 	ecore_ptt_release(p_hwfn, p_ptt);
2629 
2630 	return ECORE_SUCCESS;
2631 }
2632 
2633 enum _ecore_status_t ecore_cxt_get_task_ctx(struct ecore_hwfn *p_hwfn,
2634 					    u32 tid,
2635 					    u8 ctx_type,
2636 					    void **pp_task_ctx)
2637 {
2638 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2639 	struct ecore_ilt_client_cfg *p_cli;
2640 	struct ecore_tid_seg *p_seg_info;
2641 	struct ecore_ilt_cli_blk *p_seg;
2642 	u32 num_tids_per_block;
2643 	u32 tid_size, ilt_idx;
2644 	u32 total_lines;
2645 	u32 proto, seg;
2646 
2647 	/* Verify the personality */
2648 	switch (p_hwfn->hw_info.personality) {
2649 	case ECORE_PCI_FCOE:
2650 		proto = PROTOCOLID_FCOE;
2651 		seg = ECORE_CXT_FCOE_TID_SEG;
2652 		break;
2653 	case ECORE_PCI_ISCSI:
2654 		proto = PROTOCOLID_ISCSI;
2655 		seg = ECORE_CXT_ISCSI_TID_SEG;
2656 		break;
2657 	default:
2658 		return ECORE_INVAL;
2659 	}
2660 
2661 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2662 	if (!p_cli->active) {
2663 		return ECORE_INVAL;
2664 	}
2665 
2666 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2667 
2668 	if (ctx_type == ECORE_CTX_WORKING_MEM) {
2669 		p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2670 	} else if (ctx_type == ECORE_CTX_FL_MEM) {
2671 		if (!p_seg_info->has_fl_mem) {
2672 			return ECORE_INVAL;
2673 		}
2674 		p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2675 	} else {
2676 		return ECORE_INVAL;
2677 	}
2678 	total_lines = DIV_ROUND_UP(p_seg->total_size,
2679 				   p_seg->real_size_in_page);
2680 	tid_size = p_mngr->task_type_size[p_seg_info->type];
2681 	num_tids_per_block = p_seg->real_size_in_page / tid_size;
2682 
2683 	if (total_lines < tid/num_tids_per_block)
2684 		return ECORE_INVAL;
2685 
2686 	ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2687 		  p_mngr->pf_start_line;
2688 	*pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2689 			     (tid % num_tids_per_block) * tid_size;
2690 
2691 	return ECORE_SUCCESS;
2692 }
2693