xref: /freebsd/sys/dev/qlnx/qlnxe/ecore_cxt.c (revision 405c3050f102b8c74782f0366c8ead927bd07b68)
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 
1066 /* Total number of ILT lines used by this PF */
1067 static u32 ecore_cxt_ilt_shadow_size(struct ecore_ilt_client_cfg *ilt_clients)
1068 {
1069 	u32 size = 0;
1070 	u32 i;
1071 
1072 	for_each_ilt_valid_client(i, ilt_clients)
1073 		size += (ilt_clients[i].last.val -
1074 			 ilt_clients[i].first.val + 1);
1075 
1076 	return size;
1077 }
1078 
1079 static void ecore_ilt_shadow_free(struct ecore_hwfn *p_hwfn)
1080 {
1081 	struct ecore_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
1082 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1083 	u32 ilt_size, i;
1084 
1085 	if (p_mngr->ilt_shadow == OSAL_NULL)
1086 		return;
1087 
1088 	ilt_size = ecore_cxt_ilt_shadow_size(p_cli);
1089 
1090 	for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
1091 		struct ecore_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
1092 
1093 		if (p_dma->p_virt)
1094 			OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
1095 					       p_dma->p_virt,
1096 					       p_dma->p_phys,
1097 					       p_dma->size);
1098 		p_dma->p_virt = OSAL_NULL;
1099 	}
1100 	OSAL_FREE(p_hwfn->p_dev, p_mngr->ilt_shadow);
1101 	p_mngr->ilt_shadow = OSAL_NULL;
1102 }
1103 
1104 static enum _ecore_status_t ecore_ilt_blk_alloc(struct ecore_hwfn *p_hwfn,
1105 						struct ecore_ilt_cli_blk *p_blk,
1106 						enum ilt_clients ilt_client,
1107 						u32 start_line_offset)
1108 {
1109 	struct ecore_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
1110 	u32 lines, line, sz_left, lines_to_skip = 0;
1111 
1112 	/* Special handling for RoCE that supports dynamic allocation */
1113 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn) &&
1114 	    ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
1115 		return ECORE_SUCCESS;
1116 
1117 	lines_to_skip = p_blk->dynamic_line_cnt;
1118 
1119 	if (!p_blk->total_size)
1120 		return ECORE_SUCCESS;
1121 
1122 	sz_left = p_blk->total_size;
1123 	lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) -
1124 		lines_to_skip;
1125 	line = p_blk->start_line + start_line_offset -
1126 	       p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
1127 
1128 	for (; lines; lines--) {
1129 		dma_addr_t p_phys;
1130 		void *p_virt;
1131 		u32 size;
1132 
1133 		size = OSAL_MIN_T(u32, sz_left, p_blk->real_size_in_page);
1134 		p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
1135 						 &p_phys, size);
1136 		if (!p_virt)
1137 			return ECORE_NOMEM;
1138 		OSAL_MEM_ZERO(p_virt, size);
1139 
1140 		ilt_shadow[line].p_phys = p_phys;
1141 		ilt_shadow[line].p_virt = p_virt;
1142 		ilt_shadow[line].size = size;
1143 
1144 		DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1145 			   "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
1146 			   line, (unsigned long long)p_phys, p_virt, size);
1147 
1148 		sz_left -= size;
1149 		line++;
1150 	}
1151 
1152 	return ECORE_SUCCESS;
1153 }
1154 
1155 static enum _ecore_status_t ecore_ilt_shadow_alloc(struct ecore_hwfn *p_hwfn)
1156 {
1157 	struct ecore_cxt_mngr *p_mngr  = p_hwfn->p_cxt_mngr;
1158 	struct ecore_ilt_client_cfg *clients = p_mngr->clients;
1159 	struct ecore_ilt_cli_blk *p_blk;
1160 	u32 size, i, j, k;
1161 	enum _ecore_status_t rc;
1162 
1163 	size = ecore_cxt_ilt_shadow_size(clients);
1164 	p_mngr->ilt_shadow = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL,
1165 					 size * sizeof(struct ecore_dma_mem));
1166 
1167 	if (p_mngr->ilt_shadow == OSAL_NULL) {
1168 		DP_NOTICE(p_hwfn, false, "Failed to allocate ilt shadow table\n");
1169 		rc = ECORE_NOMEM;
1170 		goto ilt_shadow_fail;
1171 	}
1172 
1173 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1174 		   "Allocated 0x%x bytes for ilt shadow\n",
1175 		   (u32)(size * sizeof(struct ecore_dma_mem)));
1176 
1177 	for_each_ilt_valid_client(i, clients) {
1178 		for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1179 			p_blk = &clients[i].pf_blks[j];
1180 			rc = ecore_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1181 			if (rc != ECORE_SUCCESS)
1182 				goto ilt_shadow_fail;
1183 		}
1184 		for (k = 0; k < p_mngr->vf_count; k++) {
1185 			for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1186 				u32 lines = clients[i].vf_total_lines * k;
1187 
1188 				p_blk = &clients[i].vf_blks[j];
1189 				rc = ecore_ilt_blk_alloc(p_hwfn, p_blk,
1190 							 i, lines);
1191 				if (rc != ECORE_SUCCESS)
1192 					goto ilt_shadow_fail;
1193 			}
1194 		}
1195 	}
1196 
1197 	return ECORE_SUCCESS;
1198 
1199 ilt_shadow_fail:
1200 	ecore_ilt_shadow_free(p_hwfn);
1201 	return rc;
1202 }
1203 
1204 static void ecore_cid_map_free(struct ecore_hwfn *p_hwfn)
1205 {
1206 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1207 	u32 type, vf;
1208 
1209 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1210 		OSAL_FREE(p_hwfn->p_dev, p_mngr->acquired[type].cid_map);
1211 		p_mngr->acquired[type].cid_map = OSAL_NULL;
1212 		p_mngr->acquired[type].max_count = 0;
1213 		p_mngr->acquired[type].start_cid = 0;
1214 
1215 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1216 			OSAL_FREE(p_hwfn->p_dev,
1217 				  p_mngr->acquired_vf[type][vf].cid_map);
1218 			p_mngr->acquired_vf[type][vf].cid_map = OSAL_NULL;
1219 			p_mngr->acquired_vf[type][vf].max_count = 0;
1220 			p_mngr->acquired_vf[type][vf].start_cid = 0;
1221 		}
1222 	}
1223 }
1224 
1225 static enum _ecore_status_t
1226 ecore_cid_map_alloc_single(struct ecore_hwfn *p_hwfn, u32 type,
1227 			   u32 cid_start, u32 cid_count,
1228 			   struct ecore_cid_acquired_map *p_map)
1229 {
1230 	u32 size;
1231 
1232 	if (!cid_count)
1233 		return ECORE_SUCCESS;
1234 
1235 	size = MAP_WORD_SIZE * DIV_ROUND_UP(cid_count, BITS_PER_MAP_WORD);
1236 	p_map->cid_map = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, size);
1237 	if (p_map->cid_map == OSAL_NULL)
1238 		return ECORE_NOMEM;
1239 
1240 	p_map->max_count = cid_count;
1241 	p_map->start_cid = cid_start;
1242 
1243 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
1244 		   "Type %08x start: %08x count %08x\n",
1245 		   type, p_map->start_cid, p_map->max_count);
1246 
1247 	return ECORE_SUCCESS;
1248 }
1249 
1250 static enum _ecore_status_t ecore_cid_map_alloc(struct ecore_hwfn *p_hwfn)
1251 {
1252 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1253 	u32 start_cid = 0, vf_start_cid = 0;
1254 	u32 type, vf;
1255 
1256 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1257 		struct ecore_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[type];
1258 		struct ecore_cid_acquired_map *p_map;
1259 
1260 		/* Handle PF maps */
1261 		p_map = &p_mngr->acquired[type];
1262 		if (ecore_cid_map_alloc_single(p_hwfn, type, start_cid,
1263 					       p_cfg->cid_count, p_map))
1264 			goto cid_map_fail;
1265 
1266 		/* Handle VF maps */
1267 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1268 			p_map = &p_mngr->acquired_vf[type][vf];
1269 			if (ecore_cid_map_alloc_single(p_hwfn, type,
1270 						       vf_start_cid,
1271 						       p_cfg->cids_per_vf,
1272 						       p_map))
1273 				goto cid_map_fail;
1274 		}
1275 
1276 		start_cid += p_cfg->cid_count;
1277 		vf_start_cid += p_cfg->cids_per_vf;
1278 	}
1279 
1280 	return ECORE_SUCCESS;
1281 
1282 cid_map_fail:
1283 	ecore_cid_map_free(p_hwfn);
1284 	return ECORE_NOMEM;
1285 }
1286 
1287 enum _ecore_status_t ecore_cxt_mngr_alloc(struct ecore_hwfn *p_hwfn)
1288 {
1289 	struct ecore_ilt_client_cfg *clients;
1290 	struct ecore_cxt_mngr *p_mngr;
1291 	u32 i;
1292 
1293 	p_mngr = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, sizeof(*p_mngr));
1294 	if (!p_mngr) {
1295 		DP_NOTICE(p_hwfn, false, "Failed to allocate `struct ecore_cxt_mngr'\n");
1296 		return ECORE_NOMEM;
1297 	}
1298 
1299 	/* Initialize ILT client registers */
1300 	clients = p_mngr->clients;
1301 	clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1302 	clients[ILT_CLI_CDUC].last.reg  = ILT_CFG_REG(CDUC, LAST_ILT);
1303 	clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1304 
1305 	clients[ILT_CLI_QM].first.reg   = ILT_CFG_REG(QM, FIRST_ILT);
1306 	clients[ILT_CLI_QM].last.reg    = ILT_CFG_REG(QM, LAST_ILT);
1307 	clients[ILT_CLI_QM].p_size.reg  = ILT_CFG_REG(QM, P_SIZE);
1308 
1309 	clients[ILT_CLI_TM].first.reg   = ILT_CFG_REG(TM, FIRST_ILT);
1310 	clients[ILT_CLI_TM].last.reg    = ILT_CFG_REG(TM, LAST_ILT);
1311 	clients[ILT_CLI_TM].p_size.reg  = ILT_CFG_REG(TM, P_SIZE);
1312 
1313 	clients[ILT_CLI_SRC].first.reg  = ILT_CFG_REG(SRC, FIRST_ILT);
1314 	clients[ILT_CLI_SRC].last.reg   = ILT_CFG_REG(SRC, LAST_ILT);
1315 	clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1316 
1317 	clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1318 	clients[ILT_CLI_CDUT].last.reg  = ILT_CFG_REG(CDUT, LAST_ILT);
1319 	clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1320 
1321 	clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1322 	clients[ILT_CLI_TSDM].last.reg  = ILT_CFG_REG(TSDM, LAST_ILT);
1323 	clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1324 
1325 	/* default ILT page size for all clients is 64K */
1326 	for (i = 0; i < ILT_CLI_MAX; i++)
1327 		p_mngr->clients[i].p_size.val = p_hwfn->p_dev->ilt_page_size;
1328 
1329 	/* Initialize task sizes */
1330 	p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1331 	p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1332 
1333 	if (p_hwfn->p_dev->p_iov_info)
1334 		p_mngr->vf_count = p_hwfn->p_dev->p_iov_info->total_vfs;
1335 
1336 	/* Initialize the dynamic ILT allocation mutex */
1337 #ifdef CONFIG_ECORE_LOCK_ALLOC
1338 	OSAL_MUTEX_ALLOC(p_hwfn, &p_mngr->mutex);
1339 #endif
1340 	OSAL_MUTEX_INIT(&p_mngr->mutex);
1341 
1342 	/* Set the cxt mangr pointer priori to further allocations */
1343 	p_hwfn->p_cxt_mngr = p_mngr;
1344 
1345 	return ECORE_SUCCESS;
1346 }
1347 
1348 enum _ecore_status_t ecore_cxt_tables_alloc(struct ecore_hwfn *p_hwfn)
1349 {
1350 	enum _ecore_status_t    rc;
1351 
1352 	/* Allocate the ILT shadow table */
1353 	rc = ecore_ilt_shadow_alloc(p_hwfn);
1354 	if (rc) {
1355 		DP_NOTICE(p_hwfn, false, "Failed to allocate ilt memory\n");
1356 		goto tables_alloc_fail;
1357 	}
1358 
1359 	/* Allocate the T2  table */
1360 	rc = ecore_cxt_src_t2_alloc(p_hwfn);
1361 	if (rc) {
1362 		DP_NOTICE(p_hwfn, false, "Failed to allocate T2 memory\n");
1363 		goto tables_alloc_fail;
1364 	}
1365 
1366 	/* Allocate and initialize the acquired cids bitmaps */
1367 	rc = ecore_cid_map_alloc(p_hwfn);
1368 	if (rc) {
1369 		DP_NOTICE(p_hwfn, false, "Failed to allocate cid maps\n");
1370 		goto tables_alloc_fail;
1371 	}
1372 
1373 	return ECORE_SUCCESS;
1374 
1375 tables_alloc_fail:
1376 	ecore_cxt_mngr_free(p_hwfn);
1377 	return rc;
1378 }
1379 void ecore_cxt_mngr_free(struct ecore_hwfn *p_hwfn)
1380 {
1381 	if (!p_hwfn->p_cxt_mngr)
1382 		return;
1383 
1384 	ecore_cid_map_free(p_hwfn);
1385 	ecore_cxt_src_t2_free(p_hwfn);
1386 	ecore_ilt_shadow_free(p_hwfn);
1387 #ifdef CONFIG_ECORE_LOCK_ALLOC
1388 	OSAL_MUTEX_DEALLOC(&p_hwfn->p_cxt_mngr->mutex);
1389 #endif
1390 	OSAL_FREE(p_hwfn->p_dev, p_hwfn->p_cxt_mngr);
1391 
1392 	p_hwfn->p_cxt_mngr = OSAL_NULL;
1393 }
1394 
1395 void ecore_cxt_mngr_setup(struct ecore_hwfn *p_hwfn)
1396 {
1397 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1398 	struct ecore_cid_acquired_map *p_map;
1399 	struct ecore_conn_type_cfg *p_cfg;
1400 	int type;
1401 	u32 len;
1402 
1403 	/* Reset acquired cids */
1404 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1405 		u32 vf;
1406 
1407 		p_cfg = &p_mngr->conn_cfg[type];
1408 		if (p_cfg->cid_count) {
1409 			p_map = &p_mngr->acquired[type];
1410 			len = DIV_ROUND_UP(p_map->max_count,
1411 					   BITS_PER_MAP_WORD) *
1412 			      MAP_WORD_SIZE;
1413 			OSAL_MEM_ZERO(p_map->cid_map, len);
1414 		}
1415 
1416 		if (!p_cfg->cids_per_vf)
1417 			continue;
1418 
1419 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1420 			p_map = &p_mngr->acquired_vf[type][vf];
1421 			len = DIV_ROUND_UP(p_map->max_count,
1422 					   BITS_PER_MAP_WORD) *
1423 			      MAP_WORD_SIZE;
1424 			OSAL_MEM_ZERO(p_map->cid_map, len);
1425 		}
1426 	}
1427 }
1428 
1429 /* HW initialization helper (per Block, per phase) */
1430 
1431 /* CDU Common */
1432 #define CDUC_CXT_SIZE_SHIFT						\
1433 	CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1434 
1435 #define CDUC_CXT_SIZE_MASK						\
1436 	(CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1437 
1438 #define CDUC_BLOCK_WASTE_SHIFT						\
1439 	CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1440 
1441 #define CDUC_BLOCK_WASTE_MASK						\
1442 	(CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1443 
1444 #define CDUC_NCIB_SHIFT							\
1445 	CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1446 
1447 #define CDUC_NCIB_MASK							\
1448 	(CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1449 
1450 #define CDUT_TYPE0_CXT_SIZE_SHIFT					\
1451 	CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1452 
1453 #define CDUT_TYPE0_CXT_SIZE_MASK					\
1454 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >>				\
1455 	CDUT_TYPE0_CXT_SIZE_SHIFT)
1456 
1457 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT					\
1458 	CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1459 
1460 #define CDUT_TYPE0_BLOCK_WASTE_MASK					\
1461 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >>			\
1462 	CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1463 
1464 #define CDUT_TYPE0_NCIB_SHIFT						\
1465 	CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1466 
1467 #define CDUT_TYPE0_NCIB_MASK						\
1468 	(CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >>		\
1469 	CDUT_TYPE0_NCIB_SHIFT)
1470 
1471 #define CDUT_TYPE1_CXT_SIZE_SHIFT					\
1472 	CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1473 
1474 #define CDUT_TYPE1_CXT_SIZE_MASK					\
1475 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >>				\
1476 	CDUT_TYPE1_CXT_SIZE_SHIFT)
1477 
1478 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT					\
1479 	CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1480 
1481 #define CDUT_TYPE1_BLOCK_WASTE_MASK					\
1482 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >>			\
1483 	CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1484 
1485 #define CDUT_TYPE1_NCIB_SHIFT						\
1486 	CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1487 
1488 #define CDUT_TYPE1_NCIB_MASK						\
1489 	(CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >>		\
1490 	CDUT_TYPE1_NCIB_SHIFT)
1491 
1492 static void ecore_cdu_init_common(struct ecore_hwfn *p_hwfn)
1493 {
1494 	u32 page_sz, elems_per_page, block_waste,  cxt_size, cdu_params = 0;
1495 
1496 	/* CDUC - connection configuration */
1497 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1498 	cxt_size = CONN_CXT_SIZE(p_hwfn);
1499 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1500 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1501 
1502 	SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1503 	SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1504 	SET_FIELD(cdu_params, (u32)CDUC_NCIB, elems_per_page);
1505 	STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1506 
1507 	/* CDUT - type-0 tasks configuration */
1508 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1509 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1510 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1511 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1512 
1513 	/* cxt size and block-waste are multipes of 8 */
1514 	cdu_params = 0;
1515 	SET_FIELD(cdu_params, (u32)CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1516 	SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1517 	SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1518 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1519 
1520 	/* CDUT - type-1 tasks configuration */
1521 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1522 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1523 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1524 
1525 	/* cxt size and block-waste are multipes of 8 */
1526 	cdu_params = 0;
1527 	SET_FIELD(cdu_params, (u32)CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1528 	SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1529 	SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1530 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1531 }
1532 
1533 /* CDU PF */
1534 #define CDU_SEG_REG_TYPE_SHIFT		CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1535 #define CDU_SEG_REG_TYPE_MASK		0x1
1536 #define CDU_SEG_REG_OFFSET_SHIFT	0
1537 #define CDU_SEG_REG_OFFSET_MASK		CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1538 
1539 static void ecore_cdu_init_pf(struct ecore_hwfn *p_hwfn)
1540 {
1541 	struct ecore_ilt_client_cfg *p_cli;
1542 	struct ecore_tid_seg *p_seg;
1543 	u32 cdu_seg_params, offset;
1544 	int i;
1545 
1546 	static const u32 rt_type_offset_arr[] = {
1547 		CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1548 		CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1549 		CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1550 		CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1551 	};
1552 
1553 	static const u32 rt_type_offset_fl_arr[] = {
1554 		CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1555 		CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1556 		CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1557 		CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1558 	};
1559 
1560 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1561 
1562 	/* There are initializations only for CDUT during pf Phase */
1563 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1564 		/* Segment 0*/
1565 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
1566 		if (!p_seg)
1567 			continue;
1568 
1569 		/* Note: start_line is already adjusted for the CDU
1570 		 * segment register granularity, so we just need to
1571 		 * divide. Adjustment is implicit as we assume ILT
1572 		 * Page size is larger than 32K!
1573 		 */
1574 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1575 			 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1576 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1577 
1578 		cdu_seg_params = 0;
1579 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1580 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1581 		STORE_RT_REG(p_hwfn, rt_type_offset_arr[i],
1582 			     cdu_seg_params);
1583 
1584 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1585 			 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1586 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1587 
1588 		cdu_seg_params = 0;
1589 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1590 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1591 		STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i],
1592 			     cdu_seg_params);
1593 
1594 	}
1595 }
1596 
1597 void ecore_qm_init_pf(struct ecore_hwfn *p_hwfn, struct ecore_ptt *p_ptt,
1598 		      bool is_pf_loading)
1599 {
1600 	struct ecore_qm_info *qm_info = &p_hwfn->qm_info;
1601 	struct ecore_mcp_link_state *p_link;
1602 	struct ecore_qm_iids iids;
1603 
1604 	OSAL_MEM_ZERO(&iids, sizeof(iids));
1605 	ecore_cxt_qm_iids(p_hwfn, &iids);
1606 
1607 	p_link = &ECORE_LEADING_HWFN(p_hwfn->p_dev)->mcp_info->link_output;
1608 
1609 	ecore_qm_pf_rt_init(p_hwfn, p_ptt, p_hwfn->port_id,
1610 			    p_hwfn->rel_pf_id, qm_info->max_phys_tcs_per_port,
1611 			    is_pf_loading,
1612 			    iids.cids, iids.vf_cids, iids.tids,
1613 			    qm_info->start_pq,
1614 			    qm_info->num_pqs - qm_info->num_vf_pqs,
1615 			    qm_info->num_vf_pqs,
1616 			    qm_info->start_vport,
1617 			    qm_info->num_vports, qm_info->pf_wfq,
1618 			    qm_info->pf_rl, p_link->speed,
1619 			    p_hwfn->qm_info.qm_pq_params,
1620 			    p_hwfn->qm_info.qm_vport_params);
1621 }
1622 
1623 /* CM PF */
1624 static void ecore_cm_init_pf(struct ecore_hwfn *p_hwfn)
1625 {
1626 	STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET, ecore_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1627 }
1628 
1629 /* DQ PF */
1630 static void ecore_dq_init_pf(struct ecore_hwfn *p_hwfn)
1631 {
1632 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1633 	u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1634 
1635 	dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1636 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1637 
1638 	dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1639 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1640 
1641 	dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1642 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1643 
1644 	dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1645 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1646 
1647 	dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1648 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1649 
1650 	dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1651 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1652 
1653 	dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1654 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1655 
1656 	dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1657 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1658 
1659 	dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1660 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1661 
1662 	dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1663 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1664 
1665 	dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1666 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1667 
1668 	dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1669 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1670 
1671 	/* Connection types 6 & 7 are not in use, yet they must be configured
1672 	 * as the highest possible connection. Not configuring them means the
1673 	 * defaults will be  used, and with a large number of cids a bug may
1674 	 * occur, if the defaults will be smaller than dq_pf_max_cid /
1675 	 * dq_vf_max_cid.
1676 	 */
1677 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1678 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1679 
1680 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1681 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1682 }
1683 
1684 static void ecore_ilt_bounds_init(struct ecore_hwfn *p_hwfn)
1685 {
1686 	struct ecore_ilt_client_cfg *ilt_clients;
1687 	int i;
1688 
1689 	ilt_clients = p_hwfn->p_cxt_mngr->clients;
1690 	for_each_ilt_valid_client(i, ilt_clients) {
1691 		STORE_RT_REG(p_hwfn,
1692 			     ilt_clients[i].first.reg,
1693 			     ilt_clients[i].first.val);
1694 		STORE_RT_REG(p_hwfn,
1695 			     ilt_clients[i].last.reg,
1696 			     ilt_clients[i].last.val);
1697 		STORE_RT_REG(p_hwfn,
1698 			     ilt_clients[i].p_size.reg,
1699 			     ilt_clients[i].p_size.val);
1700 	}
1701 }
1702 
1703 static void ecore_ilt_vf_bounds_init(struct ecore_hwfn *p_hwfn)
1704 {
1705 	struct ecore_ilt_client_cfg *p_cli;
1706 	u32 blk_factor;
1707 
1708 	/* For simplicty  we set the 'block' to be an ILT page */
1709 	if (p_hwfn->p_dev->p_iov_info) {
1710 		struct ecore_hw_sriov_info *p_iov = p_hwfn->p_dev->p_iov_info;
1711 
1712 		STORE_RT_REG(p_hwfn,
1713 			     PSWRQ2_REG_VF_BASE_RT_OFFSET,
1714 			     p_iov->first_vf_in_pf);
1715 		STORE_RT_REG(p_hwfn,
1716 			     PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1717 			     p_iov->first_vf_in_pf + p_iov->total_vfs);
1718 	}
1719 
1720 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1721 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1722 	if (p_cli->active) {
1723 		STORE_RT_REG(p_hwfn,
1724 			     PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1725 			     blk_factor);
1726 		STORE_RT_REG(p_hwfn,
1727 			     PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1728 			     p_cli->pf_total_lines);
1729 		STORE_RT_REG(p_hwfn,
1730 			     PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1731 			     p_cli->vf_total_lines);
1732 	}
1733 
1734 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1735 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1736 	if (p_cli->active) {
1737 		STORE_RT_REG(p_hwfn,
1738 			     PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1739 			     blk_factor);
1740 		STORE_RT_REG(p_hwfn,
1741 			     PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1742 			     p_cli->pf_total_lines);
1743 		STORE_RT_REG(p_hwfn,
1744 			     PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1745 			     p_cli->vf_total_lines);
1746 	}
1747 
1748 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1749 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1750 	if (p_cli->active) {
1751 		STORE_RT_REG(p_hwfn,
1752 			     PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET,
1753 			     blk_factor);
1754 		STORE_RT_REG(p_hwfn,
1755 			     PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1756 			     p_cli->pf_total_lines);
1757 		STORE_RT_REG(p_hwfn,
1758 			     PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1759 			     p_cli->vf_total_lines);
1760 	}
1761 }
1762 
1763 /* ILT (PSWRQ2) PF */
1764 static void ecore_ilt_init_pf(struct ecore_hwfn *p_hwfn)
1765 {
1766 	struct ecore_ilt_client_cfg *clients;
1767 	struct ecore_cxt_mngr *p_mngr;
1768 	struct ecore_dma_mem *p_shdw;
1769 	u32 line, rt_offst, i;
1770 
1771 	ecore_ilt_bounds_init(p_hwfn);
1772 	ecore_ilt_vf_bounds_init(p_hwfn);
1773 
1774 	p_mngr  = p_hwfn->p_cxt_mngr;
1775 	p_shdw  = p_mngr->ilt_shadow;
1776 	clients = p_hwfn->p_cxt_mngr->clients;
1777 
1778 	for_each_ilt_valid_client(i, clients) {
1779 		/* Client's 1st val and RT array are absolute, ILT shadows'
1780 		 * lines are relative.
1781 		 */
1782 		line = clients[i].first.val - p_mngr->pf_start_line;
1783 		rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1784 			   clients[i].first.val * ILT_ENTRY_IN_REGS;
1785 
1786 		for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1787 		     line++, rt_offst += ILT_ENTRY_IN_REGS) {
1788 			u64 ilt_hw_entry = 0;
1789 
1790 			/** p_virt could be OSAL_NULL incase of dynamic
1791 			 *  allocation
1792 			 */
1793 			if (p_shdw[line].p_virt != OSAL_NULL) {
1794 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1795 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1796 					  (unsigned long long)(p_shdw[line].p_phys >> 12));
1797 
1798 				DP_VERBOSE(
1799 					p_hwfn, ECORE_MSG_ILT,
1800 					"Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1801 					rt_offst, line, i,
1802 					(unsigned long long)(p_shdw[line].p_phys >> 12));
1803 			}
1804 
1805 			STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1806 		}
1807 	}
1808 }
1809 
1810 /* SRC (Searcher) PF */
1811 static void ecore_src_init_pf(struct ecore_hwfn *p_hwfn)
1812 {
1813 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1814 	u32 rounded_conn_num, conn_num, conn_max;
1815 	struct ecore_src_iids src_iids;
1816 
1817 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
1818 	ecore_cxt_src_iids(p_mngr, &src_iids);
1819 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1820 	if (!conn_num)
1821 		return;
1822 
1823 	conn_max = OSAL_MAX_T(u32, conn_num, SRC_MIN_NUM_ELEMS);
1824 	rounded_conn_num = OSAL_ROUNDUP_POW_OF_TWO(conn_max);
1825 
1826 	STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1827 	STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1828 		     OSAL_LOG2(rounded_conn_num));
1829 
1830 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1831 			 p_hwfn->p_cxt_mngr->first_free);
1832 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1833 			 p_hwfn->p_cxt_mngr->last_free);
1834 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1835 		   "Configured SEARCHER for 0x%08x connections\n",
1836 		   conn_num);
1837 }
1838 
1839 /* Timers PF */
1840 #define TM_CFG_NUM_IDS_SHIFT		0
1841 #define TM_CFG_NUM_IDS_MASK		0xFFFFULL
1842 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT	16
1843 #define TM_CFG_PRE_SCAN_OFFSET_MASK	0x1FFULL
1844 #define TM_CFG_PARENT_PF_SHIFT		25
1845 #define TM_CFG_PARENT_PF_MASK		0x7ULL
1846 
1847 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT	30
1848 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK	0x1FFULL
1849 
1850 #define TM_CFG_TID_OFFSET_SHIFT		30
1851 #define TM_CFG_TID_OFFSET_MASK		0x7FFFFULL
1852 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT	49
1853 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK	0x1FFULL
1854 
1855 static void ecore_tm_init_pf(struct ecore_hwfn *p_hwfn)
1856 {
1857 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1858 	u32 active_seg_mask = 0, tm_offset, rt_reg;
1859 	struct ecore_tm_iids tm_iids;
1860 	u64 cfg_word;
1861 	u8 i;
1862 
1863 	OSAL_MEM_ZERO(&tm_iids, sizeof(tm_iids));
1864 	ecore_cxt_tm_iids(p_mngr, &tm_iids);
1865 
1866 	/* @@@TBD No pre-scan for now */
1867 
1868 	/* Note: We assume consecutive VFs for a PF */
1869 	for (i = 0; i < p_mngr->vf_count; i++) {
1870 		cfg_word = 0;
1871 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1872 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1873 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1874 		SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1875 
1876 		rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1877 			 (sizeof(cfg_word) / sizeof(u32)) *
1878 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1879 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1880 	}
1881 
1882 	cfg_word = 0;
1883 	SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1884 	SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1885 	SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);	  /* n/a for PF */
1886 	SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all   */
1887 
1888 	rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1889 		 (sizeof(cfg_word) / sizeof(u32)) *
1890 		 (NUM_OF_VFS(p_hwfn->p_dev) + p_hwfn->rel_pf_id);
1891 	STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1892 
1893 	/* enale scan */
1894 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1895 		     tm_iids.pf_cids  ? 0x1 : 0x0);
1896 
1897 	/* @@@TBD how to enable the scan for the VFs */
1898 
1899 	tm_offset = tm_iids.per_vf_cids;
1900 
1901 	/* Note: We assume consecutive VFs for a PF */
1902 	for (i = 0; i < p_mngr->vf_count; i++) {
1903 		cfg_word = 0;
1904 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1905 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1906 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1907 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1908 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1909 
1910 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1911 			 (sizeof(cfg_word) / sizeof(u32)) *
1912 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1913 
1914 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1915 	}
1916 
1917 	tm_offset = tm_iids.pf_cids;
1918 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1919 		cfg_word = 0;
1920 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1921 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1922 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1923 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1924 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1925 
1926 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1927 			 (sizeof(cfg_word) / sizeof(u32)) *
1928 			 (NUM_OF_VFS(p_hwfn->p_dev) +
1929 			 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1930 
1931 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1932 		active_seg_mask |= (tm_iids.pf_tids[i] ? (1 << i) : 0);
1933 
1934 		tm_offset += tm_iids.pf_tids[i];
1935 	}
1936 
1937 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn))
1938 		active_seg_mask = 0;
1939 
1940 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1941 
1942 	/* @@@TBD how to enable the scan for the VFs */
1943 }
1944 
1945 static void ecore_prs_init_common(struct ecore_hwfn *p_hwfn)
1946 {
1947 	if ((p_hwfn->hw_info.personality == ECORE_PCI_FCOE) &&
1948 	    p_hwfn->pf_params.fcoe_pf_params.is_target)
1949 		STORE_RT_REG(p_hwfn, PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1950 }
1951 
1952 static void ecore_prs_init_pf(struct ecore_hwfn *p_hwfn)
1953 {
1954 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1955 	struct ecore_conn_type_cfg *p_fcoe;
1956 	struct ecore_tid_seg *p_tid;
1957 
1958 	p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1959 
1960 	/* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1961 	if (!p_fcoe->cid_count)
1962 		return;
1963 
1964 	p_tid = &p_fcoe->tid_seg[ECORE_CXT_FCOE_TID_SEG];
1965 	if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1966 		STORE_RT_REG_AGG(p_hwfn,
1967 				 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1968 				 p_tid->count);
1969 	} else {
1970 		STORE_RT_REG_AGG(p_hwfn,
1971 				PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1972 				p_tid->count);
1973 	}
1974 }
1975 
1976 void ecore_cxt_hw_init_common(struct ecore_hwfn *p_hwfn)
1977 {
1978 	/* CDU configuration */
1979 	ecore_cdu_init_common(p_hwfn);
1980 	ecore_prs_init_common(p_hwfn);
1981 }
1982 
1983 void ecore_cxt_hw_init_pf(struct ecore_hwfn *p_hwfn, struct ecore_ptt *p_ptt)
1984 {
1985 	ecore_qm_init_pf(p_hwfn, p_ptt, true);
1986 	ecore_cm_init_pf(p_hwfn);
1987 	ecore_dq_init_pf(p_hwfn);
1988 	ecore_cdu_init_pf(p_hwfn);
1989 	ecore_ilt_init_pf(p_hwfn);
1990 	ecore_src_init_pf(p_hwfn);
1991 	ecore_tm_init_pf(p_hwfn);
1992 	ecore_prs_init_pf(p_hwfn);
1993 }
1994 
1995 enum _ecore_status_t _ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
1996 					    enum protocol_type type,
1997 					    u32 *p_cid, u8 vfid)
1998 {
1999 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2000 	struct ecore_cid_acquired_map *p_map;
2001 	u32 rel_cid;
2002 
2003 	if (type >= MAX_CONN_TYPES) {
2004 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
2005 		return ECORE_INVAL;
2006 	}
2007 
2008 	if (vfid >= COMMON_MAX_NUM_VFS && vfid != ECORE_CXT_PF_CID) {
2009 		DP_NOTICE(p_hwfn, true, "VF [%02x] is out of range\n", vfid);
2010 		return ECORE_INVAL;
2011 	}
2012 
2013 	/* Determine the right map to take this CID from */
2014 	if (vfid == ECORE_CXT_PF_CID)
2015 		p_map = &p_mngr->acquired[type];
2016 	else
2017 		p_map = &p_mngr->acquired_vf[type][vfid];
2018 
2019 	if (p_map->cid_map == OSAL_NULL) {
2020 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
2021 		return ECORE_INVAL;
2022 	}
2023 
2024 	rel_cid = OSAL_FIND_FIRST_ZERO_BIT(p_map->cid_map,
2025 					   p_map->max_count);
2026 
2027 	if (rel_cid >= p_map->max_count) {
2028 		DP_NOTICE(p_hwfn, false, "no CID available for protocol %d\n",
2029 			  type);
2030 		return ECORE_NORESOURCES;
2031 	}
2032 
2033 	OSAL_SET_BIT(rel_cid, p_map->cid_map);
2034 
2035 	*p_cid = rel_cid + p_map->start_cid;
2036 
2037 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
2038 		   "Acquired cid 0x%08x [rel. %08x] vfid %02x type %d\n",
2039 		   *p_cid, rel_cid, vfid, type);
2040 
2041 	return ECORE_SUCCESS;
2042 }
2043 
2044 enum _ecore_status_t ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
2045 					   enum protocol_type type,
2046 					   u32 *p_cid)
2047 {
2048 	return _ecore_cxt_acquire_cid(p_hwfn, type, p_cid, ECORE_CXT_PF_CID);
2049 }
2050 
2051 static bool ecore_cxt_test_cid_acquired(struct ecore_hwfn *p_hwfn,
2052 					u32 cid, u8 vfid,
2053 					enum protocol_type *p_type,
2054 					struct ecore_cid_acquired_map **pp_map)
2055 {
2056 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2057 	u32 rel_cid;
2058 
2059 	/* Iterate over protocols and find matching cid range */
2060 	for (*p_type = 0; *p_type < MAX_CONN_TYPES; (*p_type)++) {
2061 		if (vfid == ECORE_CXT_PF_CID)
2062 			*pp_map = &p_mngr->acquired[*p_type];
2063 		else
2064 			*pp_map = &p_mngr->acquired_vf[*p_type][vfid];
2065 
2066 		if (!((*pp_map)->cid_map))
2067 			continue;
2068 		if (cid >= (*pp_map)->start_cid &&
2069 		    cid < (*pp_map)->start_cid + (*pp_map)->max_count) {
2070 			break;
2071 		}
2072 	}
2073 
2074 	if (*p_type == MAX_CONN_TYPES) {
2075 		DP_NOTICE(p_hwfn, true, "Invalid CID %d vfid %02x", cid, vfid);
2076 		goto fail;
2077 	}
2078 
2079 	rel_cid = cid - (*pp_map)->start_cid;
2080 	if (!OSAL_TEST_BIT(rel_cid, (*pp_map)->cid_map)) {
2081 		DP_NOTICE(p_hwfn, true,
2082 			  "CID %d [vifd %02x] not acquired", cid, vfid);
2083 		goto fail;
2084 	}
2085 
2086 	return true;
2087 fail:
2088 	*p_type = MAX_CONN_TYPES;
2089 	*pp_map = OSAL_NULL;
2090 	return false;
2091 }
2092 
2093 void _ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid, u8 vfid)
2094 {
2095 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2096 	enum protocol_type type;
2097 	bool b_acquired;
2098 	u32 rel_cid;
2099 
2100 	if (vfid != ECORE_CXT_PF_CID && vfid > COMMON_MAX_NUM_VFS) {
2101 		DP_NOTICE(p_hwfn, true,
2102 			  "Trying to return incorrect CID belonging to VF %02x\n",
2103 			  vfid);
2104 		return;
2105 	}
2106 
2107 	/* Test acquired and find matching per-protocol map */
2108 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, cid, vfid,
2109 						 &type, &p_map);
2110 
2111 	if (!b_acquired)
2112 		return;
2113 
2114 	rel_cid = cid - p_map->start_cid;
2115 	OSAL_CLEAR_BIT(rel_cid, p_map->cid_map);
2116 
2117 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
2118 		   "Released CID 0x%08x [rel. %08x] vfid %02x type %d\n",
2119 		   cid, rel_cid, vfid, type);
2120 }
2121 
2122 void ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid)
2123 {
2124 	_ecore_cxt_release_cid(p_hwfn, cid, ECORE_CXT_PF_CID);
2125 }
2126 
2127 enum _ecore_status_t ecore_cxt_get_cid_info(struct ecore_hwfn *p_hwfn,
2128 					    struct ecore_cxt_info *p_info)
2129 {
2130 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2131 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2132 	u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
2133 	enum protocol_type type;
2134 	bool b_acquired;
2135 
2136 	/* Test acquired and find matching per-protocol map */
2137 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, p_info->iid,
2138 						 ECORE_CXT_PF_CID,
2139 						 &type, &p_map);
2140 
2141 	if (!b_acquired)
2142 		return ECORE_INVAL;
2143 
2144 	/* set the protocl type */
2145 	p_info->type = type;
2146 
2147 	/* compute context virtual pointer */
2148 	hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
2149 
2150 	conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
2151 	cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
2152 	line = p_info->iid / cxts_per_p;
2153 
2154 	/* Make sure context is allocated (dynamic allocation) */
2155 	if (!p_mngr->ilt_shadow[line].p_virt)
2156 		return ECORE_INVAL;
2157 
2158 	p_info->p_cxt = (u8 *)p_mngr->ilt_shadow[line].p_virt +
2159 			      p_info->iid % cxts_per_p * conn_cxt_size;
2160 
2161 	DP_VERBOSE(p_hwfn, (ECORE_MSG_ILT | ECORE_MSG_CXT),
2162 		   "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
2163 		   (p_info->iid / cxts_per_p), p_info->p_cxt, p_info->iid);
2164 
2165 	return ECORE_SUCCESS;
2166 }
2167 
2168 static void ecore_rdma_set_pf_params(struct ecore_hwfn *p_hwfn,
2169 				     struct ecore_rdma_pf_params *p_params,
2170 				     u32 num_tasks)
2171 {
2172 	u32 num_cons, num_qps;
2173 	enum protocol_type proto;
2174 
2175 	/* The only case RDMA personality can be overriden is if NVRAM is
2176 	 * configured with ETH_RDMA or if no rdma protocol was requested
2177 	 */
2178 	switch (p_params->rdma_protocol) {
2179 	case ECORE_RDMA_PROTOCOL_DEFAULT:
2180 		if (p_hwfn->mcp_info->func_info.protocol ==
2181 		    ECORE_PCI_ETH_RDMA) {
2182 			DP_NOTICE(p_hwfn, false,
2183 				  "Current day drivers don't support RoCE & iWARP. Default to RoCE-only\n");
2184 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2185 		}
2186 		break;
2187 	case ECORE_RDMA_PROTOCOL_NONE:
2188 		p_hwfn->hw_info.personality = ECORE_PCI_ETH;
2189 		return; /* intentional... nothing left to do... */
2190 	case ECORE_RDMA_PROTOCOL_ROCE:
2191 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2192 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2193 		break;
2194 	case ECORE_RDMA_PROTOCOL_IWARP:
2195 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2196 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_IWARP;
2197 		break;
2198 	}
2199 
2200 	switch (p_hwfn->hw_info.personality) {
2201 	case ECORE_PCI_ETH_IWARP:
2202 		/* Each QP requires one connection */
2203 		num_cons = OSAL_MIN_T(u32, IWARP_MAX_QPS, p_params->num_qps);
2204 #ifdef CONFIG_ECORE_IWARP /* required for the define */
2205 		/* additional connections required for passive tcp handling */
2206 		num_cons += ECORE_IWARP_PREALLOC_CNT;
2207 #endif
2208 		proto = PROTOCOLID_IWARP;
2209 		break;
2210 	case ECORE_PCI_ETH_ROCE:
2211 		num_qps = OSAL_MIN_T(u32, ROCE_MAX_QPS, p_params->num_qps);
2212 		num_cons = num_qps * 2; /* each QP requires two connections */
2213 		proto = PROTOCOLID_ROCE;
2214 		break;
2215 	default:
2216 		return;
2217 	}
2218 
2219 	if (num_cons && num_tasks) {
2220 		u32 num_srqs, num_xrc_srqs, max_xrc_srqs, page_size;
2221 
2222 		ecore_cxt_set_proto_cid_count(p_hwfn, proto,
2223 					      num_cons, 0);
2224 
2225 		/* Deliberatly passing ROCE for tasks id. This is because
2226 		 * iWARP / RoCE share the task id.
2227 		 */
2228 		ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
2229 					      ECORE_CXT_ROCE_TID_SEG,
2230 					      1, /* RoCE segment type */
2231 					      num_tasks,
2232 					      false); /* !force load */
2233 
2234 		num_srqs = OSAL_MIN_T(u32, ECORE_RDMA_MAX_SRQS,
2235 				      p_params->num_srqs);
2236 
2237 		/* XRC SRQs populate a single ILT page */
2238 		page_size = ecore_cxt_get_ilt_page_size(p_hwfn, ILT_CLI_TSDM);
2239 		max_xrc_srqs =  page_size / XRC_SRQ_CXT_SIZE;
2240 		max_xrc_srqs = OSAL_MIN_T(u32, max_xrc_srqs, ECORE_RDMA_MAX_XRC_SRQS);
2241 
2242 		num_xrc_srqs = OSAL_MIN_T(u32, p_params->num_xrc_srqs,
2243 					  max_xrc_srqs);
2244 		ecore_cxt_set_srq_count(p_hwfn, num_srqs, num_xrc_srqs);
2245 
2246 	} else {
2247 		DP_INFO(p_hwfn->p_dev,
2248 			"RDMA personality used without setting params!\n");
2249 	}
2250 }
2251 
2252 enum _ecore_status_t ecore_cxt_set_pf_params(struct ecore_hwfn *p_hwfn,
2253 					     u32 rdma_tasks)
2254 {
2255 	/* Set the number of required CORE connections */
2256 	u32 core_cids = 1; /* SPQ */
2257 
2258 	if (p_hwfn->using_ll2)
2259 		core_cids += 4; /* @@@TBD Use the proper #define */
2260 
2261 	ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
2262 
2263 	switch (p_hwfn->hw_info.personality) {
2264 	case ECORE_PCI_ETH_RDMA:
2265 	case ECORE_PCI_ETH_IWARP:
2266 	case ECORE_PCI_ETH_ROCE:
2267 	{
2268 		ecore_rdma_set_pf_params(p_hwfn,
2269 					 &p_hwfn->pf_params.rdma_pf_params,
2270 					 rdma_tasks);
2271 
2272 		/* no need for break since RoCE coexist with Ethernet */
2273 	}
2274 	case ECORE_PCI_ETH:
2275 	{
2276 		u32 count = 0;
2277 
2278 		struct ecore_eth_pf_params *p_params =
2279 					&p_hwfn->pf_params.eth_pf_params;
2280 
2281 		if (!p_params->num_vf_cons)
2282 			p_params->num_vf_cons = ETH_PF_PARAMS_VF_CONS_DEFAULT;
2283 		ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
2284 					      p_params->num_cons,
2285 					      p_params->num_vf_cons);
2286 
2287 		count = p_params->num_arfs_filters;
2288 
2289 		if (!OSAL_TEST_BIT(ECORE_MF_DISABLE_ARFS,
2290 				   &p_hwfn->p_dev->mf_bits))
2291 			p_hwfn->p_cxt_mngr->arfs_count = count;
2292 
2293 		break;
2294 	}
2295 	case ECORE_PCI_FCOE:
2296 	{
2297 		struct ecore_fcoe_pf_params *p_params;
2298 
2299 		p_params = &p_hwfn->pf_params.fcoe_pf_params;
2300 
2301 		if (p_params->num_cons && p_params->num_tasks) {
2302 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_FCOE,
2303 						      p_params->num_cons, 0);
2304 
2305 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
2306 						      ECORE_CXT_FCOE_TID_SEG,
2307 						      0, /* segment type */
2308 						      p_params->num_tasks,
2309 						      true);
2310 		} else {
2311 			DP_INFO(p_hwfn->p_dev,
2312 				"Fcoe personality used without setting params!\n");
2313 		}
2314 		break;
2315 	}
2316 	case ECORE_PCI_ISCSI:
2317 	{
2318 		struct ecore_iscsi_pf_params *p_params;
2319 
2320 		p_params = &p_hwfn->pf_params.iscsi_pf_params;
2321 
2322 		if (p_params->num_cons && p_params->num_tasks) {
2323 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ISCSI,
2324 						      p_params->num_cons, 0);
2325 
2326 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ISCSI,
2327 						      ECORE_CXT_ISCSI_TID_SEG,
2328 						      0, /* segment type */
2329 						      p_params->num_tasks,
2330 						      true);
2331 		} else {
2332 			DP_INFO(p_hwfn->p_dev,
2333 				"Iscsi personality used without setting params!\n");
2334 		}
2335 		break;
2336 	}
2337 	default:
2338 		return ECORE_INVAL;
2339 	}
2340 
2341 	return ECORE_SUCCESS;
2342 }
2343 
2344 enum _ecore_status_t ecore_cxt_get_tid_mem_info(struct ecore_hwfn *p_hwfn,
2345 						struct ecore_tid_mem *p_info)
2346 {
2347 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2348 	u32 proto, seg, total_lines, i, shadow_line;
2349 	struct ecore_ilt_client_cfg *p_cli;
2350 	struct ecore_ilt_cli_blk *p_fl_seg;
2351 	struct ecore_tid_seg *p_seg_info;
2352 
2353 	/* Verify the personality */
2354 	switch (p_hwfn->hw_info.personality) {
2355 	case ECORE_PCI_FCOE:
2356 		proto = PROTOCOLID_FCOE;
2357 		seg = ECORE_CXT_FCOE_TID_SEG;
2358 		break;
2359 	case ECORE_PCI_ISCSI:
2360 		proto = PROTOCOLID_ISCSI;
2361 		seg = ECORE_CXT_ISCSI_TID_SEG;
2362 		break;
2363 	default:
2364 		return ECORE_INVAL;
2365 	}
2366 
2367 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2368 	if (!p_cli->active) {
2369 		return ECORE_INVAL;
2370 	}
2371 
2372 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2373 	if (!p_seg_info->has_fl_mem)
2374 		return ECORE_INVAL;
2375 
2376 	p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2377 	total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2378 				   p_fl_seg->real_size_in_page);
2379 
2380 	for (i = 0; i < total_lines; i++) {
2381 		shadow_line = i + p_fl_seg->start_line -
2382 			      p_hwfn->p_cxt_mngr->pf_start_line;
2383 		p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
2384 	}
2385 	p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2386 			p_fl_seg->real_size_in_page;
2387 	p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2388 	p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2389 				     p_info->tid_size;
2390 
2391 	return ECORE_SUCCESS;
2392 }
2393 
2394 /* This function is very RoCE oriented, if another protocol in the future
2395  * will want this feature we'll need to modify the function to be more generic
2396  */
2397 enum _ecore_status_t
2398 ecore_cxt_dynamic_ilt_alloc(struct ecore_hwfn *p_hwfn,
2399 			    enum ecore_cxt_elem_type elem_type,
2400 			    u32 iid)
2401 {
2402 	u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2403 	struct ecore_ilt_client_cfg *p_cli;
2404 	struct ecore_ilt_cli_blk *p_blk;
2405 	struct ecore_ptt *p_ptt;
2406 	dma_addr_t p_phys;
2407 	u64 ilt_hw_entry;
2408 	void *p_virt;
2409 	enum _ecore_status_t rc = ECORE_SUCCESS;
2410 
2411 	switch (elem_type) {
2412 	case ECORE_ELEM_CXT:
2413 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2414 		elem_size = CONN_CXT_SIZE(p_hwfn);
2415 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2416 		break;
2417 	case ECORE_ELEM_SRQ:
2418 		/* The first ILT page is not used for regular SRQs. Skip it. */
2419 		iid += ecore_cxt_srqs_per_page(p_hwfn);
2420 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2421 		elem_size = SRQ_CXT_SIZE;
2422 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2423 		break;
2424 	case ECORE_ELEM_XRC_SRQ:
2425 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2426 		elem_size = XRC_SRQ_CXT_SIZE;
2427 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2428 		break;
2429 	case ECORE_ELEM_TASK:
2430 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2431 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2432 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2433 		break;
2434 	default:
2435 		DP_NOTICE(p_hwfn, false,
2436 			  "ECORE_INVALID elem type = %d", elem_type);
2437 		return ECORE_INVAL;
2438 	}
2439 
2440 	/* Calculate line in ilt */
2441 	hw_p_size = p_cli->p_size.val;
2442 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2443 	line = p_blk->start_line + (iid / elems_per_p);
2444 	shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2445 
2446 	/* If line is already allocated, do nothing, otherwise allocate it and
2447 	 * write it to the PSWRQ2 registers.
2448 	 * This section can be run in parallel from different contexts and thus
2449 	 * a mutex protection is needed.
2450 	 */
2451 #ifdef _NTDDK_
2452 #pragma warning(suppress : 28121)
2453 #endif
2454 	OSAL_MUTEX_ACQUIRE(&p_hwfn->p_cxt_mngr->mutex);
2455 
2456 	if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2457 		goto out0;
2458 
2459 	p_ptt = ecore_ptt_acquire(p_hwfn);
2460 	if (!p_ptt) {
2461 		DP_NOTICE(p_hwfn, false,
2462 			  "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2463 		rc = ECORE_TIMEOUT;
2464 		goto out0;
2465 	}
2466 
2467 	p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
2468 					 &p_phys,
2469 					 p_blk->real_size_in_page);
2470 	if (!p_virt) {
2471 		rc = ECORE_NOMEM;
2472 		goto out1;
2473 	}
2474 	OSAL_MEM_ZERO(p_virt, p_blk->real_size_in_page);
2475 
2476 	/* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2477 	 * to compensate for a HW bug, but it is configured even if DIF is not
2478 	 * enabled. This is harmless and allows us to avoid a dedicated API. We
2479 	 * configure the field for all of the contexts on the newly allocated
2480 	 * page.
2481 	 */
2482 	if (elem_type == ECORE_ELEM_TASK) {
2483 		u32 elem_i;
2484 		u8 *elem_start = (u8 *)p_virt;
2485 		union type1_task_context *elem;
2486 
2487 		for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2488 			elem = (union type1_task_context *)elem_start;
2489 			SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2490 				  TDIF_TASK_CONTEXT_REF_TAG_MASK , 0xf);
2491 			elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2492 		}
2493 	}
2494 
2495 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2496 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2497 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2498 		p_blk->real_size_in_page;
2499 
2500 	/* compute absolute offset */
2501 	reg_offset = PSWRQ2_REG_ILT_MEMORY +
2502 		     (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2503 
2504 	ilt_hw_entry = 0;
2505 	SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2506 	SET_FIELD(ilt_hw_entry,
2507 		  ILT_ENTRY_PHY_ADDR,
2508 		  (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2509 
2510 	/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2511 	ecore_dmae_host2grc(p_hwfn, p_ptt, (u64)(osal_uintptr_t)&ilt_hw_entry,
2512 			    reg_offset, sizeof(ilt_hw_entry) / sizeof(u32),
2513 			    OSAL_NULL /* default parameters */);
2514 
2515 	if (elem_type == ECORE_ELEM_CXT) {
2516 		u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2517 					 elems_per_p;
2518 
2519 		/* Update the relevant register in the parser */
2520 		ecore_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2521 			 last_cid_allocated - 1);
2522 
2523 		/* RoCE w/a -> we don't write to the prs search reg until first
2524 		 * cid is allocated. This is because the prs checks
2525 		 * last_cid-1 >=0 making 0 a valid value... this will cause
2526 		 * the a context load to occur on a RoCE packet received with
2527 		 * cid=0 even before context was initialized, can happen with a
2528 		 * stray packet from switch or a packet with crc-error
2529 		 */
2530 
2531 		if (!p_hwfn->b_rdma_enabled_in_prs) {
2532 			/* Enable Rdma search */
2533 			ecore_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2534 			p_hwfn->b_rdma_enabled_in_prs = true;
2535 		}
2536 	}
2537 
2538 out1:
2539 	ecore_ptt_release(p_hwfn, p_ptt);
2540 out0:
2541 	OSAL_MUTEX_RELEASE(&p_hwfn->p_cxt_mngr->mutex);
2542 
2543 	return rc;
2544 }
2545 
2546 /* This function is very RoCE oriented, if another protocol in the future
2547  * will want this feature we'll need to modify the function to be more generic
2548  */
2549 enum _ecore_status_t
2550 ecore_cxt_free_ilt_range(struct ecore_hwfn *p_hwfn,
2551 			 enum ecore_cxt_elem_type elem_type,
2552 			 u32 start_iid, u32 count)
2553 {
2554 	u32 start_line, end_line, shadow_start_line, shadow_end_line;
2555 	u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2556 	struct ecore_ilt_client_cfg *p_cli;
2557 	struct ecore_ilt_cli_blk *p_blk;
2558 	u32 end_iid = start_iid + count;
2559 	struct ecore_ptt *p_ptt;
2560 	u64 ilt_hw_entry = 0;
2561 	u32 i;
2562 
2563 	switch (elem_type) {
2564 	case ECORE_ELEM_CXT:
2565 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2566 		elem_size = CONN_CXT_SIZE(p_hwfn);
2567 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2568 		break;
2569 	case ECORE_ELEM_SRQ:
2570 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2571 		elem_size = SRQ_CXT_SIZE;
2572 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2573 		break;
2574 	case ECORE_ELEM_TASK:
2575 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2576 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2577 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2578 		break;
2579 	default:
2580 		DP_NOTICE(p_hwfn, false,
2581 			  "ECORE_INVALID elem type = %d", elem_type);
2582 		return ECORE_INVAL;
2583 	}
2584 
2585 	/* Calculate line in ilt */
2586 	hw_p_size = p_cli->p_size.val;
2587 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2588 	start_line = p_blk->start_line + (start_iid / elems_per_p);
2589 	end_line = p_blk->start_line + (end_iid / elems_per_p);
2590 	if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2591 		end_line--;
2592 
2593 	shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2594 	shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2595 
2596 	p_ptt = ecore_ptt_acquire(p_hwfn);
2597 	if (!p_ptt) {
2598 		DP_NOTICE(p_hwfn, false, "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2599 		return ECORE_TIMEOUT;
2600 	}
2601 
2602 	for (i = shadow_start_line; i < shadow_end_line; i++) {
2603 		if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2604 			continue;
2605 
2606 		OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
2607 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2608 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys,
2609 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].size);
2610 
2611 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = OSAL_NULL;
2612 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2613 		p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2614 
2615 		/* compute absolute offset */
2616 		reg_offset = PSWRQ2_REG_ILT_MEMORY +
2617 			     ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2618 			      ILT_ENTRY_IN_REGS);
2619 
2620 		/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2621 		 * wide-bus.
2622 		 */
2623 		ecore_dmae_host2grc(p_hwfn, p_ptt,
2624 				    (u64)(osal_uintptr_t)&ilt_hw_entry,
2625 				    reg_offset,
2626 				    sizeof(ilt_hw_entry) / sizeof(u32),
2627 				    OSAL_NULL /* default parameters */);
2628 	}
2629 
2630 	ecore_ptt_release(p_hwfn, p_ptt);
2631 
2632 	return ECORE_SUCCESS;
2633 }
2634 
2635 enum _ecore_status_t ecore_cxt_get_task_ctx(struct ecore_hwfn *p_hwfn,
2636 					    u32 tid,
2637 					    u8 ctx_type,
2638 					    void **pp_task_ctx)
2639 {
2640 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2641 	struct ecore_ilt_client_cfg *p_cli;
2642 	struct ecore_tid_seg *p_seg_info;
2643 	struct ecore_ilt_cli_blk *p_seg;
2644 	u32 num_tids_per_block;
2645 	u32 tid_size, ilt_idx;
2646 	u32 total_lines;
2647 	u32 proto, seg;
2648 
2649 	/* Verify the personality */
2650 	switch (p_hwfn->hw_info.personality) {
2651 	case ECORE_PCI_FCOE:
2652 		proto = PROTOCOLID_FCOE;
2653 		seg = ECORE_CXT_FCOE_TID_SEG;
2654 		break;
2655 	case ECORE_PCI_ISCSI:
2656 		proto = PROTOCOLID_ISCSI;
2657 		seg = ECORE_CXT_ISCSI_TID_SEG;
2658 		break;
2659 	default:
2660 		return ECORE_INVAL;
2661 	}
2662 
2663 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2664 	if (!p_cli->active) {
2665 		return ECORE_INVAL;
2666 	}
2667 
2668 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2669 
2670 	if (ctx_type == ECORE_CTX_WORKING_MEM) {
2671 		p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2672 	} else if (ctx_type == ECORE_CTX_FL_MEM) {
2673 		if (!p_seg_info->has_fl_mem) {
2674 			return ECORE_INVAL;
2675 		}
2676 		p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2677 	} else {
2678 		return ECORE_INVAL;
2679 	}
2680 	total_lines = DIV_ROUND_UP(p_seg->total_size,
2681 				   p_seg->real_size_in_page);
2682 	tid_size = p_mngr->task_type_size[p_seg_info->type];
2683 	num_tids_per_block = p_seg->real_size_in_page / tid_size;
2684 
2685 	if (total_lines < tid/num_tids_per_block)
2686 		return ECORE_INVAL;
2687 
2688 	ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2689 		  p_mngr->pf_start_line;
2690 	*pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2691 			     (tid % num_tids_per_block) * tid_size;
2692 
2693 	return ECORE_SUCCESS;
2694 }
2695