xref: /freebsd/sys/dev/qlnx/qlnxe/ecore_cxt.c (revision a64729f5077d77e13b9497cb33ecb3c82e606ee8)
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 #include "bcm_osal.h"
33 #include "reg_addr.h"
34 #include "common_hsi.h"
35 #include "ecore_hsi_common.h"
36 #include "ecore_hsi_eth.h"
37 #include "tcp_common.h"
38 #include "ecore_hsi_iscsi.h"
39 #include "ecore_hsi_fcoe.h"
40 #include "ecore_hsi_roce.h"
41 #include "ecore_hsi_iwarp.h"
42 #include "ecore_rt_defs.h"
43 #include "ecore_status.h"
44 #include "ecore.h"
45 #include "ecore_init_ops.h"
46 #include "ecore_init_fw_funcs.h"
47 #include "ecore_cxt.h"
48 #include "ecore_hw.h"
49 #include "ecore_dev_api.h"
50 #include "ecore_sriov.h"
51 #include "ecore_rdma.h"
52 #include "ecore_mcp.h"
53 
54 /* Max number of connection types in HW (DQ/CDU etc.) */
55 #define MAX_CONN_TYPES		PROTOCOLID_COMMON
56 #define NUM_TASK_TYPES		2
57 #define NUM_TASK_PF_SEGMENTS	4
58 #define NUM_TASK_VF_SEGMENTS	1
59 
60 /* Doorbell-Queue constants */
61 #define DQ_RANGE_SHIFT	4
62 #define DQ_RANGE_ALIGN	(1 << DQ_RANGE_SHIFT)
63 
64 /* Searcher constants */
65 #define SRC_MIN_NUM_ELEMS 256
66 
67 /* Timers constants */
68 #define TM_SHIFT	7
69 #define TM_ALIGN	(1 << TM_SHIFT)
70 #define TM_ELEM_SIZE	4
71 
72 /* ILT constants */
73 #define ILT_PAGE_IN_BYTES(hw_p_size)	(1U << ((hw_p_size) + 12))
74 #define ILT_CFG_REG(cli, reg)		PSWRQ2_REG_##cli##_##reg##_RT_OFFSET
75 
76 /* ILT entry structure */
77 #define ILT_ENTRY_PHY_ADDR_MASK		0x000FFFFFFFFFFFULL
78 #define ILT_ENTRY_PHY_ADDR_SHIFT	0
79 #define ILT_ENTRY_VALID_MASK		0x1ULL
80 #define ILT_ENTRY_VALID_SHIFT		52
81 #define ILT_ENTRY_IN_REGS		2
82 #define ILT_REG_SIZE_IN_BYTES		4
83 
84 /* connection context union */
85 union conn_context {
86 	struct e4_core_conn_context  core_ctx;
87 	struct e4_eth_conn_context	  eth_ctx;
88 	struct e4_iscsi_conn_context iscsi_ctx;
89 	struct e4_fcoe_conn_context  fcoe_ctx;
90 	struct e4_roce_conn_context  roce_ctx;
91 };
92 
93 /* TYPE-0 task context - iSCSI, FCOE */
94 union type0_task_context {
95 	struct e4_iscsi_task_context iscsi_ctx;
96 	struct e4_fcoe_task_context  fcoe_ctx;
97 };
98 
99 /* TYPE-1 task context - ROCE */
100 union type1_task_context {
101 	struct e4_rdma_task_context roce_ctx;
102 };
103 
104 struct src_ent {
105 	u8  opaque[56];
106 	u64 next;
107 };
108 
109 #define CDUT_SEG_ALIGNMET 3 /* in 4k chunks */
110 #define CDUT_SEG_ALIGNMET_IN_BYTES (1 << (CDUT_SEG_ALIGNMET + 12))
111 
112 #define CONN_CXT_SIZE(p_hwfn) \
113 	ALIGNED_TYPE_SIZE(union conn_context, p_hwfn)
114 
115 #define SRQ_CXT_SIZE (sizeof(struct rdma_srq_context))
116 #define XRC_SRQ_CXT_SIZE (sizeof(struct rdma_xrc_srq_context))
117 
118 #define TYPE0_TASK_CXT_SIZE(p_hwfn) \
119 	ALIGNED_TYPE_SIZE(union type0_task_context, p_hwfn)
120 
121 /* Alignment is inherent to the type1_task_context structure */
122 #define TYPE1_TASK_CXT_SIZE(p_hwfn) sizeof(union type1_task_context)
123 
124 /* PF per protocl configuration object */
125 #define TASK_SEGMENTS   (NUM_TASK_PF_SEGMENTS + NUM_TASK_VF_SEGMENTS)
126 #define TASK_SEGMENT_VF (NUM_TASK_PF_SEGMENTS)
127 
128 struct ecore_tid_seg {
129 	u32	count;
130 	u8	type;
131 	bool	has_fl_mem;
132 };
133 
134 struct ecore_conn_type_cfg {
135 	u32			cid_count;
136 	u32			cids_per_vf;
137 	struct ecore_tid_seg	tid_seg[TASK_SEGMENTS];
138 };
139 
140 /* ILT Client configuration,
141  * Per connection type (protocol) resources (cids, tis, vf cids etc.)
142  * 1 - for connection context (CDUC) and for each task context we need two
143  * values, for regular task context and for force load memory
144  */
145 #define ILT_CLI_PF_BLOCKS	(1 + NUM_TASK_PF_SEGMENTS * 2)
146 #define ILT_CLI_VF_BLOCKS	(1 + NUM_TASK_VF_SEGMENTS * 2)
147 #define CDUC_BLK		(0)
148 #define SRQ_BLK			(0)
149 #define CDUT_SEG_BLK(n)		(1 + (u8)(n))
150 #define CDUT_FL_SEG_BLK(n, X)	(1 + (n) + NUM_TASK_##X##_SEGMENTS)
151 
152 struct ilt_cfg_pair {
153 	u32 reg;
154 	u32 val;
155 };
156 
157 struct ecore_ilt_cli_blk {
158 	u32 total_size; /* 0 means not active */
159 	u32 real_size_in_page;
160 	u32 start_line;
161 	u32 dynamic_line_cnt;
162 };
163 
164 struct ecore_ilt_client_cfg {
165 	bool				active;
166 
167 	/* ILT boundaries */
168 	struct ilt_cfg_pair		first;
169 	struct ilt_cfg_pair		last;
170 	struct ilt_cfg_pair		p_size;
171 
172 	/* ILT client blocks for PF */
173 	struct ecore_ilt_cli_blk	pf_blks[ILT_CLI_PF_BLOCKS];
174 	u32				pf_total_lines;
175 
176 	/* ILT client blocks for VFs */
177 	struct ecore_ilt_cli_blk	vf_blks[ILT_CLI_VF_BLOCKS];
178 	u32				vf_total_lines;
179 };
180 
181 /* Per Path -
182  *      ILT shadow table
183  *      Protocol acquired CID lists
184  *      PF start line in ILT
185  */
186 struct ecore_dma_mem {
187 	dma_addr_t	p_phys;
188 	void		*p_virt;
189 	osal_size_t	size;
190 };
191 
192 #define MAP_WORD_SIZE		sizeof(unsigned long)
193 #define BITS_PER_MAP_WORD	(MAP_WORD_SIZE * 8)
194 
195 struct ecore_cid_acquired_map {
196 	u32		start_cid;
197 	u32		max_count;
198 	unsigned long	*cid_map;
199 };
200 
201 struct ecore_cxt_mngr {
202 	/* Per protocl configuration */
203 	struct ecore_conn_type_cfg	conn_cfg[MAX_CONN_TYPES];
204 
205 	/* computed ILT structure */
206 	struct ecore_ilt_client_cfg	clients[ILT_CLI_MAX];
207 
208 	/* Task type sizes */
209 	u32				task_type_size[NUM_TASK_TYPES];
210 
211 	/* total number of VFs for this hwfn -
212 	 * ALL VFs are symmetric in terms of HW resources
213 	 */
214 	u32				vf_count;
215 
216 	/* Acquired CIDs */
217 	struct ecore_cid_acquired_map acquired[MAX_CONN_TYPES];
218 	/* TBD - do we want this allocated to reserve space? */
219 	struct ecore_cid_acquired_map acquired_vf[MAX_CONN_TYPES][COMMON_MAX_NUM_VFS];
220 
221 	/* ILT shadow table */
222 	struct ecore_dma_mem		*ilt_shadow;
223 	u32				pf_start_line;
224 
225 	/* Mutex for a dynamic ILT allocation */
226 	osal_mutex_t			mutex;
227 
228 	/* SRC T2 */
229 	struct ecore_dma_mem		*t2;
230 	u32				t2_num_pages;
231 	u64				first_free;
232 	u64				last_free;
233 
234 	/* The infrastructure originally was very generic and context/task
235 	 * oriented - per connection-type we would set how many of those
236 	 * are needed, and later when determining how much memory we're
237 	 * needing for a given block we'd iterate over all the relevant
238 	 * connection-types.
239 	 * But since then we've had some additional resources, some of which
240 	 * require memory which is independent of the general context/task
241 	 * scheme. We add those here explicitly per-feature.
242 	 */
243 
244 	/* total number of SRQ's for this hwfn */
245 	u32				srq_count;
246 	u32				xrc_srq_count;
247 
248 	/* Maximal number of L2 steering filters */
249 	u32				arfs_count;
250 
251 	/* TODO - VF arfs filters ? */
252 };
253 
254 /* check if resources/configuration is required according to protocol type */
255 static bool src_proto(enum protocol_type type)
256 {
257 	return	type == PROTOCOLID_ISCSI	||
258 		type == PROTOCOLID_FCOE		||
259 		type == PROTOCOLID_IWARP;
260 }
261 
262 static bool tm_cid_proto(enum protocol_type type)
263 {
264 	return type == PROTOCOLID_ISCSI ||
265 	       type == PROTOCOLID_FCOE  ||
266 	       type == PROTOCOLID_ROCE  ||
267 	       type == PROTOCOLID_IWARP;
268 }
269 
270 static bool tm_tid_proto(enum protocol_type type)
271 {
272 	return type == PROTOCOLID_FCOE;
273 }
274 
275 /* counts the iids for the CDU/CDUC ILT client configuration */
276 struct ecore_cdu_iids {
277 	u32 pf_cids;
278 	u32 per_vf_cids;
279 };
280 
281 static void ecore_cxt_cdu_iids(struct ecore_cxt_mngr   *p_mngr,
282 			       struct ecore_cdu_iids	*iids)
283 {
284 	u32 type;
285 
286 	for (type = 0; type < MAX_CONN_TYPES; type++) {
287 		iids->pf_cids += p_mngr->conn_cfg[type].cid_count;
288 		iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
289 	}
290 }
291 
292 /* counts the iids for the Searcher block configuration */
293 struct ecore_src_iids {
294 	u32			pf_cids;
295 	u32			per_vf_cids;
296 };
297 
298 static void ecore_cxt_src_iids(struct ecore_cxt_mngr *p_mngr,
299 			       struct ecore_src_iids *iids)
300 {
301 	u32 i;
302 
303 	for (i = 0; i < MAX_CONN_TYPES; i++) {
304 		if (!src_proto(i))
305 			continue;
306 
307 		iids->pf_cids += p_mngr->conn_cfg[i].cid_count;
308 		iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf;
309 	}
310 
311 	/* Add L2 filtering filters in addition */
312 	iids->pf_cids += p_mngr->arfs_count;
313 }
314 
315 /* counts the iids for the Timers block configuration */
316 struct ecore_tm_iids {
317 	u32 pf_cids;
318 	u32 pf_tids[NUM_TASK_PF_SEGMENTS]; /* per segment */
319 	u32 pf_tids_total;
320 	u32 per_vf_cids;
321 	u32 per_vf_tids;
322 };
323 
324 static void ecore_cxt_tm_iids(struct ecore_cxt_mngr *p_mngr,
325 			      struct ecore_tm_iids *iids)
326 {
327 	bool tm_vf_required = false;
328 	bool tm_required = false;
329 	int i, j;
330 
331 	/* Timers is a special case -> we don't count how many cids require
332 	 * timers but what's the max cid that will be used by the timer block.
333 	 * therefore we traverse in reverse order, and once we hit a protocol
334 	 * that requires the timers memory, we'll sum all the protocols up
335 	 * to that one.
336 	 */
337 	for (i = MAX_CONN_TYPES - 1; i >= 0; i--) {
338 		struct ecore_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
339 
340 		if (tm_cid_proto(i) || tm_required) {
341 			if (p_cfg->cid_count)
342 				tm_required = true;
343 
344 			iids->pf_cids += p_cfg->cid_count;
345 		}
346 
347 		if (tm_cid_proto(i) || tm_vf_required) {
348 			if (p_cfg->cids_per_vf)
349 				tm_vf_required = true;
350 
351 			iids->per_vf_cids += p_cfg->cids_per_vf;
352 		}
353 
354 		if (tm_tid_proto(i)) {
355 			struct ecore_tid_seg *segs = p_cfg->tid_seg;
356 
357 			/* for each segment there is at most one
358 			 * protocol for which count is not 0.
359 			 */
360 			for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
361 				iids->pf_tids[j] += segs[j].count;
362 
363 			/* The last array elelment is for the VFs. As for PF
364 			 * segments there can be only one protocol for
365 			 * which this value is not 0.
366 			 */
367 			iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
368 		}
369 	}
370 
371 	iids->pf_cids = ROUNDUP(iids->pf_cids, TM_ALIGN);
372 	iids->per_vf_cids = ROUNDUP(iids->per_vf_cids, TM_ALIGN);
373 	iids->per_vf_tids = ROUNDUP(iids->per_vf_tids, TM_ALIGN);
374 
375 	for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
376 		iids->pf_tids[j] = ROUNDUP(iids->pf_tids[j], TM_ALIGN);
377 		iids->pf_tids_total += iids->pf_tids[j];
378 	}
379 }
380 
381 static void ecore_cxt_qm_iids(struct ecore_hwfn *p_hwfn,
382 			      struct ecore_qm_iids *iids)
383 {
384 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
385 	struct ecore_tid_seg *segs;
386 	u32 vf_cids = 0, type, j;
387 	u32 vf_tids = 0;
388 
389 	for (type = 0; type < MAX_CONN_TYPES; type++) {
390 		iids->cids += p_mngr->conn_cfg[type].cid_count;
391 		vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
392 
393 		segs = p_mngr->conn_cfg[type].tid_seg;
394 		/* for each segment there is at most one
395 		 * protocol for which count is not 0.
396 		 */
397 		for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
398 			iids->tids += segs[j].count;
399 
400 		/* The last array elelment is for the VFs. As for PF
401 		 * segments there can be only one protocol for
402 		 * which this value is not 0.
403 		 */
404 		vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
405 	}
406 
407 	iids->vf_cids += vf_cids * p_mngr->vf_count;
408 	iids->tids += vf_tids * p_mngr->vf_count;
409 
410 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
411 		   "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
412 		   iids->cids, iids->vf_cids, iids->tids, vf_tids);
413 }
414 
415 static struct ecore_tid_seg *ecore_cxt_tid_seg_info(struct ecore_hwfn   *p_hwfn,
416 						    u32			seg)
417 {
418 	struct ecore_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
419 	u32 i;
420 
421 	/* Find the protocol with tid count > 0 for this segment.
422 	   Note: there can only be one and this is already validated.
423 	 */
424 	for (i = 0; i < MAX_CONN_TYPES; i++) {
425 		if (p_cfg->conn_cfg[i].tid_seg[seg].count)
426 			return &p_cfg->conn_cfg[i].tid_seg[seg];
427 	}
428 	return OSAL_NULL;
429 }
430 
431 static void ecore_cxt_set_srq_count(struct ecore_hwfn *p_hwfn,
432 				    u32 num_srqs, u32 num_xrc_srqs)
433 {
434 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
435 
436 	p_mgr->srq_count = num_srqs;
437 	p_mgr->xrc_srq_count = num_xrc_srqs;
438 }
439 
440 u32 ecore_cxt_get_srq_count(struct ecore_hwfn *p_hwfn)
441 {
442 	return p_hwfn->p_cxt_mngr->srq_count;
443 }
444 
445 u32 ecore_cxt_get_xrc_srq_count(struct ecore_hwfn *p_hwfn)
446 {
447 	return p_hwfn->p_cxt_mngr->xrc_srq_count;
448 }
449 
450 u32 ecore_cxt_get_ilt_page_size(struct ecore_hwfn *p_hwfn,
451 				enum ilt_clients ilt_client)
452 {
453 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
454 	struct ecore_ilt_client_cfg *p_cli = &p_mngr->clients[ilt_client];
455 
456 	return ILT_PAGE_IN_BYTES(p_cli->p_size.val);
457 }
458 
459 static u32 ecore_cxt_srqs_per_page(struct ecore_hwfn *p_hwfn)
460 {
461 	u32 page_size;
462 
463 	page_size = ecore_cxt_get_ilt_page_size(p_hwfn, ILT_CLI_TSDM);
464 	return page_size / SRQ_CXT_SIZE;
465 }
466 
467 u32 ecore_cxt_get_total_srq_count(struct ecore_hwfn *p_hwfn)
468 {
469 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
470 	u32 total_srqs;
471 
472 	total_srqs = p_mgr->srq_count;
473 
474 	/* XRC SRQs use the first and only the first SRQ ILT page. So if XRC
475 	 * SRQs are requested we need to allocate an extra SRQ ILT page for
476 	 * them. For that We increase the number of regular SRQs to cause the
477 	 * allocation of that extra page.
478 	 */
479 	if (p_mgr->xrc_srq_count)
480 		total_srqs += ecore_cxt_srqs_per_page(p_hwfn);
481 
482 	return total_srqs;
483 }
484 
485 /* set the iids (cid/tid) count per protocol */
486 static void ecore_cxt_set_proto_cid_count(struct ecore_hwfn *p_hwfn,
487 					  enum protocol_type type,
488 					  u32 cid_count, u32 vf_cid_cnt)
489 {
490 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
491 	struct ecore_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
492 
493 	p_conn->cid_count = ROUNDUP(cid_count, DQ_RANGE_ALIGN);
494 	p_conn->cids_per_vf = ROUNDUP(vf_cid_cnt, DQ_RANGE_ALIGN);
495 
496 	if (type == PROTOCOLID_ROCE) {
497 		u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
498 		u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
499 		u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
500 		u32 align = elems_per_page * DQ_RANGE_ALIGN;
501 
502 		p_conn->cid_count = ROUNDUP(p_conn->cid_count, align);
503 	}
504 }
505 
506 u32 ecore_cxt_get_proto_cid_count(struct ecore_hwfn	*p_hwfn,
507 				  enum protocol_type	type,
508 				  u32			*vf_cid)
509 {
510 	if (vf_cid)
511 		*vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
512 
513 	return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
514 }
515 
516 u32 ecore_cxt_get_proto_cid_start(struct ecore_hwfn	*p_hwfn,
517 				  enum protocol_type	type)
518 {
519 	return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
520 }
521 
522 u32 ecore_cxt_get_proto_tid_count(struct ecore_hwfn *p_hwfn,
523 				  enum protocol_type type)
524 {
525 	u32 cnt = 0;
526 	int i;
527 
528 	for (i = 0; i < TASK_SEGMENTS; i++)
529 		cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
530 
531 	return cnt;
532 }
533 
534 static void ecore_cxt_set_proto_tid_count(struct ecore_hwfn *p_hwfn,
535 					  enum protocol_type proto,
536 					  u8 seg,
537 					  u8 seg_type,
538 					  u32 count,
539 					  bool has_fl)
540 {
541 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
542 	struct ecore_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
543 
544 	p_seg->count = count;
545 	p_seg->has_fl_mem = has_fl;
546 	p_seg->type = seg_type;
547 }
548 
549 /* the *p_line parameter must be either 0 for the first invocation or the
550    value returned in the previous invocation.
551  */
552 static void ecore_ilt_cli_blk_fill(struct ecore_ilt_client_cfg	*p_cli,
553 				   struct ecore_ilt_cli_blk	*p_blk,
554 				   u32				start_line,
555 				   u32				total_size,
556 				   u32				elem_size)
557 {
558 	u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
559 
560 	/* verify that it's called once for each block */
561 	if (p_blk->total_size)
562 		return;
563 
564 	p_blk->total_size = total_size;
565 	p_blk->real_size_in_page = 0;
566 	if (elem_size)
567 		p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
568 	p_blk->start_line = start_line;
569 }
570 
571 static void ecore_ilt_cli_adv_line(struct ecore_hwfn		*p_hwfn,
572 				    struct ecore_ilt_client_cfg	*p_cli,
573 				    struct ecore_ilt_cli_blk	*p_blk,
574 				    u32				*p_line,
575 				    enum ilt_clients		client_id)
576 {
577 	if (!p_blk->total_size)
578 		return;
579 
580 	if (!p_cli->active)
581 		p_cli->first.val = *p_line;
582 
583 	p_cli->active = true;
584 	*p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
585 	p_cli->last.val = *p_line-1;
586 
587 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
588 		   "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
589 		   client_id, p_cli->first.val, p_cli->last.val,
590 		   p_blk->total_size, p_blk->real_size_in_page,
591 		   p_blk->start_line);
592 }
593 
594 static u32 ecore_ilt_get_dynamic_line_cnt(struct ecore_hwfn *p_hwfn,
595 					  enum ilt_clients ilt_client)
596 {
597 	u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
598 	struct ecore_ilt_client_cfg *p_cli;
599 	u32 lines_to_skip = 0;
600 	u32 cxts_per_p;
601 
602 	/* TBD MK: ILT code should be simplified once PROTO enum is changed */
603 
604 	if (ilt_client == ILT_CLI_CDUC) {
605 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
606 
607 		cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
608 			     (u32)CONN_CXT_SIZE(p_hwfn);
609 
610 		lines_to_skip = cid_count / cxts_per_p;
611 	}
612 
613 	return lines_to_skip;
614 }
615 
616 static struct ecore_ilt_client_cfg *
617 ecore_cxt_set_cli(struct ecore_ilt_client_cfg *p_cli)
618 {
619 	p_cli->active = false;
620 	p_cli->first.val = 0;
621 	p_cli->last.val = 0;
622 	return p_cli;
623 }
624 
625 static struct ecore_ilt_cli_blk *
626 ecore_cxt_set_blk(struct ecore_ilt_cli_blk *p_blk)
627 {
628 	p_blk->total_size = 0;
629 	return p_blk;
630 }
631 
632 enum _ecore_status_t ecore_cxt_cfg_ilt_compute(struct ecore_hwfn *p_hwfn,
633 					       u32 *line_count)
634 {
635 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
636 	u32 curr_line, total, i, task_size, line;
637 	struct ecore_ilt_client_cfg *p_cli;
638 	struct ecore_ilt_cli_blk *p_blk;
639 	struct ecore_cdu_iids cdu_iids;
640 	struct ecore_src_iids src_iids;
641 	struct ecore_qm_iids qm_iids;
642 	struct ecore_tm_iids tm_iids;
643 	struct ecore_tid_seg *p_seg;
644 
645 	OSAL_MEM_ZERO(&qm_iids, sizeof(qm_iids));
646 	OSAL_MEM_ZERO(&cdu_iids, sizeof(cdu_iids));
647 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
648 	OSAL_MEM_ZERO(&tm_iids, sizeof(tm_iids));
649 
650 	p_mngr->pf_start_line = RESC_START(p_hwfn, ECORE_ILT);
651 
652 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
653 		   "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
654 		   p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
655 
656 	/* CDUC */
657 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUC]);
658 
659 	curr_line = p_mngr->pf_start_line;
660 
661 	/* CDUC PF */
662 	p_cli->pf_total_lines = 0;
663 
664 	/* get the counters for the CDUC,CDUC and QM clients  */
665 	ecore_cxt_cdu_iids(p_mngr, &cdu_iids);
666 
667 	p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[CDUC_BLK]);
668 
669 	total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
670 
671 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
672 			       total, CONN_CXT_SIZE(p_hwfn));
673 
674 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
675 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
676 
677 	p_blk->dynamic_line_cnt = ecore_ilt_get_dynamic_line_cnt(p_hwfn,
678 								 ILT_CLI_CDUC);
679 
680 	/* CDUC VF */
681 	p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[CDUC_BLK]);
682 	total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
683 
684 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
685 			       total, CONN_CXT_SIZE(p_hwfn));
686 
687 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
688 	p_cli->vf_total_lines = curr_line - p_blk->start_line;
689 
690 	for (i = 1; i < p_mngr->vf_count; i++)
691 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
692 				       ILT_CLI_CDUC);
693 
694 	/* CDUT PF */
695 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUT]);
696 	p_cli->first.val = curr_line;
697 
698 	/* first the 'working' task memory */
699 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
700 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
701 		if (!p_seg || p_seg->count == 0)
702 			continue;
703 
704 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[CDUT_SEG_BLK(i)]);
705 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
706 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
707 				       p_mngr->task_type_size[p_seg->type]);
708 
709 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
710 				       ILT_CLI_CDUT);
711 	}
712 
713 	/* next the 'init' task memory (forced load memory) */
714 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
715 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
716 		if (!p_seg || p_seg->count == 0)
717 			continue;
718 
719 		p_blk = ecore_cxt_set_blk(
720 				&p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)]);
721 
722 		if (!p_seg->has_fl_mem) {
723 			/* The segment is active (total size pf 'working'
724 			 * memory is > 0) but has no FL (forced-load, Init)
725 			 * memory. Thus:
726 			 *
727 			 * 1.   The total-size in the corrsponding FL block of
728 			 *      the ILT client is set to 0 - No ILT line are
729 			 *      provisioned and no ILT memory allocated.
730 			 *
731 			 * 2.   The start-line of said block is set to the
732 			 *      start line of the matching working memory
733 			 *      block in the ILT client. This is later used to
734 			 *      configure the CDU segment offset registers and
735 			 *      results in an FL command for TIDs of this
736 			 *      segement behaves as regular load commands
737 			 *      (loading TIDs from the working memory).
738 			 */
739 			line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
740 
741 			ecore_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
742 			continue;
743 		}
744 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
745 
746 		ecore_ilt_cli_blk_fill(p_cli, p_blk,
747 				       curr_line, total,
748 				       p_mngr->task_type_size[p_seg->type]);
749 
750 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
751 				       ILT_CLI_CDUT);
752 	}
753 	p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
754 
755 	/* CDUT VF */
756 	p_seg = ecore_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
757 	if (p_seg && p_seg->count) {
758 		/* Stricly speaking we need to iterate over all VF
759 		 * task segment types, but a VF has only 1 segment
760 		 */
761 
762 		/* 'working' memory */
763 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
764 
765 		p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[CDUT_SEG_BLK(0)]);
766 		ecore_ilt_cli_blk_fill(p_cli, p_blk,
767 				       curr_line, total,
768 				       p_mngr->task_type_size[p_seg->type]);
769 
770 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
771 				       ILT_CLI_CDUT);
772 
773 		/* 'init' memory */
774 		p_blk = ecore_cxt_set_blk(
775 				&p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)]);
776 		if (!p_seg->has_fl_mem) {
777 			/* see comment above */
778 			line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
779 			ecore_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
780 		} else {
781 			task_size = p_mngr->task_type_size[p_seg->type];
782 			ecore_ilt_cli_blk_fill(p_cli, p_blk,
783 					       curr_line, total,
784 					       task_size);
785 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
786 					       ILT_CLI_CDUT);
787 		}
788 		p_cli->vf_total_lines = curr_line -
789 					p_cli->vf_blks[0].start_line;
790 
791 		/* Now for the rest of the VFs */
792 		for (i = 1; i < p_mngr->vf_count; i++) {
793 			/* don't set p_blk i.e. don't clear total_size */
794 			p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
795 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
796 					       ILT_CLI_CDUT);
797 
798 			/* don't set p_blk i.e. don't clear total_size */
799 			p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
800 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
801 					       ILT_CLI_CDUT);
802 		}
803 	}
804 
805 	/* QM */
806 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_QM]);
807 	p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
808 
809 	ecore_cxt_qm_iids(p_hwfn, &qm_iids);
810 	total = ecore_qm_pf_mem_size(qm_iids.cids,
811 				     qm_iids.vf_cids, qm_iids.tids,
812 				     p_hwfn->qm_info.num_pqs,
813 				     p_hwfn->qm_info.num_vf_pqs);
814 
815 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
816 		   "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
817 		   qm_iids.cids, qm_iids.vf_cids, qm_iids.tids,
818 		   p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
819 
820 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total * 0x1000,
821 			       QM_PQ_ELEMENT_SIZE);
822 
823 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
824 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
825 
826 	/* SRC */
827 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_SRC]);
828 	ecore_cxt_src_iids(p_mngr, &src_iids);
829 
830 	/* Both the PF and VFs searcher connections are stored in the per PF
831 	 * database. Thus sum the PF searcher cids and all the VFs searcher
832 	 * cids.
833 	 */
834 	total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
835 	if (total) {
836 		u32 local_max = OSAL_MAX_T(u32, total,
837 					   SRC_MIN_NUM_ELEMS);
838 
839 		total = OSAL_ROUNDUP_POW_OF_TWO(local_max);
840 
841 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
842 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
843 				       total * sizeof(struct src_ent),
844 				       sizeof(struct src_ent));
845 
846 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
847 				       ILT_CLI_SRC);
848 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
849 	}
850 
851 	/* TM PF */
852 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_TM]);
853 	ecore_cxt_tm_iids(p_mngr, &tm_iids);
854 	total = tm_iids.pf_cids + tm_iids.pf_tids_total;
855 	if (total) {
856 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
857 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
858 				       total * TM_ELEM_SIZE,
859 				       TM_ELEM_SIZE);
860 
861 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
862 				       ILT_CLI_TM);
863 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
864 	}
865 
866 	/* TM VF */
867 	total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
868 	if (total) {
869 		p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[0]);
870 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
871 				       total * TM_ELEM_SIZE,
872 				       TM_ELEM_SIZE);
873 
874 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
875 				       ILT_CLI_TM);
876 
877 		p_cli->vf_total_lines = curr_line - p_blk->start_line;
878 		for (i = 1; i < p_mngr->vf_count; i++) {
879 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
880 					       ILT_CLI_TM);
881 		}
882 	}
883 
884 	/* TSDM (SRQ CONTEXT) */
885 	total = ecore_cxt_get_total_srq_count(p_hwfn);
886 	if (total) {
887 		p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_TSDM]);
888 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[SRQ_BLK]);
889 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
890 				       total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
891 
892 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
893 				       ILT_CLI_TSDM);
894 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
895 	}
896 
897 	*line_count = curr_line - p_hwfn->p_cxt_mngr->pf_start_line;
898 
899 	if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
900 	    RESC_NUM(p_hwfn, ECORE_ILT)) {
901 		return ECORE_INVAL;
902 	}
903 
904 	return ECORE_SUCCESS;
905 }
906 
907 u32 ecore_cxt_cfg_ilt_compute_excess(struct ecore_hwfn *p_hwfn, u32 used_lines)
908 {
909 	struct ecore_ilt_client_cfg *p_cli;
910 	u32 excess_lines, available_lines;
911 	struct ecore_cxt_mngr *p_mngr;
912 	u32 ilt_page_size, elem_size;
913 	struct ecore_tid_seg *p_seg;
914 	int i;
915 
916 	available_lines = RESC_NUM(p_hwfn, ECORE_ILT);
917 	excess_lines = used_lines - available_lines;
918 
919 	if (!excess_lines)
920 		return 0;
921 
922 	if (!ECORE_IS_RDMA_PERSONALITY(p_hwfn))
923 		return 0;
924 
925 	p_mngr = p_hwfn->p_cxt_mngr;
926 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
927 	ilt_page_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
928 
929 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
930 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
931 		if (!p_seg || p_seg->count == 0)
932 			continue;
933 
934 		elem_size = p_mngr->task_type_size[p_seg->type];
935 		if (!elem_size)
936 			continue;
937 
938 		return (ilt_page_size / elem_size) * excess_lines;
939 	}
940 
941 	DP_ERR(p_hwfn, "failed computing excess ILT lines\n");
942 	return 0;
943 }
944 
945 static void ecore_cxt_src_t2_free(struct ecore_hwfn *p_hwfn)
946 {
947 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
948 	u32 i;
949 
950 	if (!p_mngr->t2)
951 		return;
952 
953 	for (i = 0; i < p_mngr->t2_num_pages; i++)
954 		if (p_mngr->t2[i].p_virt)
955 			OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
956 					       p_mngr->t2[i].p_virt,
957 					       p_mngr->t2[i].p_phys,
958 					       p_mngr->t2[i].size);
959 
960 	OSAL_FREE(p_hwfn->p_dev, p_mngr->t2);
961 	p_mngr->t2 = OSAL_NULL;
962 }
963 
964 static enum _ecore_status_t ecore_cxt_src_t2_alloc(struct ecore_hwfn *p_hwfn)
965 {
966 	struct ecore_cxt_mngr *p_mngr  = p_hwfn->p_cxt_mngr;
967 	u32 conn_num, total_size, ent_per_page, psz, i;
968 	struct ecore_ilt_client_cfg *p_src;
969 	struct ecore_src_iids src_iids;
970 	struct ecore_dma_mem *p_t2;
971 	enum _ecore_status_t rc;
972 
973 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
974 
975 	/* if the SRC ILT client is inactive - there are no connection
976 	 * requiring the searcer, leave.
977 	 */
978 	p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
979 	if (!p_src->active)
980 		return ECORE_SUCCESS;
981 
982 	ecore_cxt_src_iids(p_mngr, &src_iids);
983 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
984 	total_size = conn_num * sizeof(struct src_ent);
985 
986 	/* use the same page size as the SRC ILT client */
987 	psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
988 	p_mngr->t2_num_pages = DIV_ROUND_UP(total_size, psz);
989 
990 	/* allocate t2 */
991 	p_mngr->t2 = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL,
992 				 p_mngr->t2_num_pages *
993 				 sizeof(struct ecore_dma_mem));
994 	if (!p_mngr->t2) {
995 		DP_NOTICE(p_hwfn, false, "Failed to allocate t2 table\n");
996 		rc = ECORE_NOMEM;
997 		goto t2_fail;
998 	}
999 
1000 	/* allocate t2 pages */
1001 	for (i = 0; i < p_mngr->t2_num_pages; i++) {
1002 		u32 size = OSAL_MIN_T(u32, total_size, psz);
1003 		void **p_virt = &p_mngr->t2[i].p_virt;
1004 
1005 		*p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
1006 						  &p_mngr->t2[i].p_phys,
1007 						  size);
1008 		if (!p_mngr->t2[i].p_virt) {
1009 			rc = ECORE_NOMEM;
1010 			goto t2_fail;
1011 		}
1012 		OSAL_MEM_ZERO(*p_virt, size);
1013 		p_mngr->t2[i].size = size;
1014 		total_size -= size;
1015 	}
1016 
1017 	/* Set the t2 pointers */
1018 
1019 	/* entries per page - must be a power of two */
1020 	ent_per_page = psz / sizeof(struct src_ent);
1021 
1022 	p_mngr->first_free = (u64)p_mngr->t2[0].p_phys;
1023 
1024 	p_t2 = &p_mngr->t2[(conn_num - 1) / ent_per_page];
1025 	p_mngr->last_free = (u64)p_t2->p_phys +
1026 				 ((conn_num - 1) & (ent_per_page - 1)) *
1027 				 sizeof(struct src_ent);
1028 
1029 	for (i = 0; i < p_mngr->t2_num_pages; i++) {
1030 		u32 ent_num = OSAL_MIN_T(u32, ent_per_page, conn_num);
1031 		struct src_ent *entries = p_mngr->t2[i].p_virt;
1032 		u64 p_ent_phys = (u64)p_mngr->t2[i].p_phys, val;
1033 		u32 j;
1034 
1035 		for (j = 0; j < ent_num - 1; j++) {
1036 			val = p_ent_phys +
1037 			      (j + 1) * sizeof(struct src_ent);
1038 			entries[j].next = OSAL_CPU_TO_BE64(val);
1039 		}
1040 
1041 		if (i < p_mngr->t2_num_pages - 1)
1042 			val = (u64)p_mngr->t2[i + 1].p_phys;
1043 		else
1044 			val = 0;
1045 		entries[j].next = OSAL_CPU_TO_BE64(val);
1046 
1047 		conn_num -= ent_num;
1048 	}
1049 
1050 	return ECORE_SUCCESS;
1051 
1052 t2_fail:
1053 	ecore_cxt_src_t2_free(p_hwfn);
1054 	return rc;
1055 }
1056 
1057 #define for_each_ilt_valid_client(pos, clients)	\
1058 	for (pos = 0; pos < ILT_CLI_MAX; pos++)	\
1059 		if (!clients[pos].active) {	\
1060 			continue;		\
1061 		} else				\
1062 
1063 /* Total number of ILT lines used by this PF */
1064 static u32 ecore_cxt_ilt_shadow_size(struct ecore_ilt_client_cfg *ilt_clients)
1065 {
1066 	u32 size = 0;
1067 	u32 i;
1068 
1069 	for_each_ilt_valid_client(i, ilt_clients)
1070 		size += (ilt_clients[i].last.val -
1071 			 ilt_clients[i].first.val + 1);
1072 
1073 	return size;
1074 }
1075 
1076 static void ecore_ilt_shadow_free(struct ecore_hwfn *p_hwfn)
1077 {
1078 	struct ecore_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
1079 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1080 	u32 ilt_size, i;
1081 
1082 	if (p_mngr->ilt_shadow == OSAL_NULL)
1083 		return;
1084 
1085 	ilt_size = ecore_cxt_ilt_shadow_size(p_cli);
1086 
1087 	for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
1088 		struct ecore_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
1089 
1090 		if (p_dma->p_virt)
1091 			OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
1092 					       p_dma->p_virt,
1093 					       p_dma->p_phys,
1094 					       p_dma->size);
1095 		p_dma->p_virt = OSAL_NULL;
1096 	}
1097 	OSAL_FREE(p_hwfn->p_dev, p_mngr->ilt_shadow);
1098 	p_mngr->ilt_shadow = OSAL_NULL;
1099 }
1100 
1101 static enum _ecore_status_t ecore_ilt_blk_alloc(struct ecore_hwfn *p_hwfn,
1102 						struct ecore_ilt_cli_blk *p_blk,
1103 						enum ilt_clients ilt_client,
1104 						u32 start_line_offset)
1105 {
1106 	struct ecore_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
1107 	u32 lines, line, sz_left, lines_to_skip = 0;
1108 
1109 	/* Special handling for RoCE that supports dynamic allocation */
1110 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn) &&
1111 	    ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
1112 		return ECORE_SUCCESS;
1113 
1114 	lines_to_skip = p_blk->dynamic_line_cnt;
1115 
1116 	if (!p_blk->total_size)
1117 		return ECORE_SUCCESS;
1118 
1119 	sz_left = p_blk->total_size;
1120 	lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) -
1121 		lines_to_skip;
1122 	line = p_blk->start_line + start_line_offset -
1123 	       p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
1124 
1125 	for (; lines; lines--) {
1126 		dma_addr_t p_phys;
1127 		void *p_virt;
1128 		u32 size;
1129 
1130 		size = OSAL_MIN_T(u32, sz_left, p_blk->real_size_in_page);
1131 		p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
1132 						 &p_phys, size);
1133 		if (!p_virt)
1134 			return ECORE_NOMEM;
1135 		OSAL_MEM_ZERO(p_virt, size);
1136 
1137 		ilt_shadow[line].p_phys = p_phys;
1138 		ilt_shadow[line].p_virt = p_virt;
1139 		ilt_shadow[line].size = size;
1140 
1141 		DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1142 			   "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
1143 			   line, (unsigned long long)p_phys, p_virt, size);
1144 
1145 		sz_left -= size;
1146 		line++;
1147 	}
1148 
1149 	return ECORE_SUCCESS;
1150 }
1151 
1152 static enum _ecore_status_t ecore_ilt_shadow_alloc(struct ecore_hwfn *p_hwfn)
1153 {
1154 	struct ecore_cxt_mngr *p_mngr  = p_hwfn->p_cxt_mngr;
1155 	struct ecore_ilt_client_cfg *clients = p_mngr->clients;
1156 	struct ecore_ilt_cli_blk *p_blk;
1157 	u32 size, i, j, k;
1158 	enum _ecore_status_t rc;
1159 
1160 	size = ecore_cxt_ilt_shadow_size(clients);
1161 	p_mngr->ilt_shadow = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL,
1162 					 size * sizeof(struct ecore_dma_mem));
1163 
1164 	if (p_mngr->ilt_shadow == OSAL_NULL) {
1165 		DP_NOTICE(p_hwfn, false, "Failed to allocate ilt shadow table\n");
1166 		rc = ECORE_NOMEM;
1167 		goto ilt_shadow_fail;
1168 	}
1169 
1170 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1171 		   "Allocated 0x%x bytes for ilt shadow\n",
1172 		   (u32)(size * sizeof(struct ecore_dma_mem)));
1173 
1174 	for_each_ilt_valid_client(i, clients) {
1175 		for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1176 			p_blk = &clients[i].pf_blks[j];
1177 			rc = ecore_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1178 			if (rc != ECORE_SUCCESS)
1179 				goto ilt_shadow_fail;
1180 		}
1181 		for (k = 0; k < p_mngr->vf_count; k++) {
1182 			for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1183 				u32 lines = clients[i].vf_total_lines * k;
1184 
1185 				p_blk = &clients[i].vf_blks[j];
1186 				rc = ecore_ilt_blk_alloc(p_hwfn, p_blk,
1187 							 i, lines);
1188 				if (rc != ECORE_SUCCESS)
1189 					goto ilt_shadow_fail;
1190 			}
1191 		}
1192 	}
1193 
1194 	return ECORE_SUCCESS;
1195 
1196 ilt_shadow_fail:
1197 	ecore_ilt_shadow_free(p_hwfn);
1198 	return rc;
1199 }
1200 
1201 static void ecore_cid_map_free(struct ecore_hwfn *p_hwfn)
1202 {
1203 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1204 	u32 type, vf;
1205 
1206 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1207 		OSAL_FREE(p_hwfn->p_dev, p_mngr->acquired[type].cid_map);
1208 		p_mngr->acquired[type].cid_map = OSAL_NULL;
1209 		p_mngr->acquired[type].max_count = 0;
1210 		p_mngr->acquired[type].start_cid = 0;
1211 
1212 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1213 			OSAL_FREE(p_hwfn->p_dev,
1214 				  p_mngr->acquired_vf[type][vf].cid_map);
1215 			p_mngr->acquired_vf[type][vf].cid_map = OSAL_NULL;
1216 			p_mngr->acquired_vf[type][vf].max_count = 0;
1217 			p_mngr->acquired_vf[type][vf].start_cid = 0;
1218 		}
1219 	}
1220 }
1221 
1222 static enum _ecore_status_t
1223 ecore_cid_map_alloc_single(struct ecore_hwfn *p_hwfn, u32 type,
1224 			   u32 cid_start, u32 cid_count,
1225 			   struct ecore_cid_acquired_map *p_map)
1226 {
1227 	u32 size;
1228 
1229 	if (!cid_count)
1230 		return ECORE_SUCCESS;
1231 
1232 	size = MAP_WORD_SIZE * DIV_ROUND_UP(cid_count, BITS_PER_MAP_WORD);
1233 	p_map->cid_map = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, size);
1234 	if (p_map->cid_map == OSAL_NULL)
1235 		return ECORE_NOMEM;
1236 
1237 	p_map->max_count = cid_count;
1238 	p_map->start_cid = cid_start;
1239 
1240 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
1241 		   "Type %08x start: %08x count %08x\n",
1242 		   type, p_map->start_cid, p_map->max_count);
1243 
1244 	return ECORE_SUCCESS;
1245 }
1246 
1247 static enum _ecore_status_t ecore_cid_map_alloc(struct ecore_hwfn *p_hwfn)
1248 {
1249 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1250 	u32 start_cid = 0, vf_start_cid = 0;
1251 	u32 type, vf;
1252 
1253 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1254 		struct ecore_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[type];
1255 		struct ecore_cid_acquired_map *p_map;
1256 
1257 		/* Handle PF maps */
1258 		p_map = &p_mngr->acquired[type];
1259 		if (ecore_cid_map_alloc_single(p_hwfn, type, start_cid,
1260 					       p_cfg->cid_count, p_map))
1261 			goto cid_map_fail;
1262 
1263 		/* Handle VF maps */
1264 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1265 			p_map = &p_mngr->acquired_vf[type][vf];
1266 			if (ecore_cid_map_alloc_single(p_hwfn, type,
1267 						       vf_start_cid,
1268 						       p_cfg->cids_per_vf,
1269 						       p_map))
1270 				goto cid_map_fail;
1271 		}
1272 
1273 		start_cid += p_cfg->cid_count;
1274 		vf_start_cid += p_cfg->cids_per_vf;
1275 	}
1276 
1277 	return ECORE_SUCCESS;
1278 
1279 cid_map_fail:
1280 	ecore_cid_map_free(p_hwfn);
1281 	return ECORE_NOMEM;
1282 }
1283 
1284 enum _ecore_status_t ecore_cxt_mngr_alloc(struct ecore_hwfn *p_hwfn)
1285 {
1286 	struct ecore_ilt_client_cfg *clients;
1287 	struct ecore_cxt_mngr *p_mngr;
1288 	u32 i;
1289 
1290 	p_mngr = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, sizeof(*p_mngr));
1291 	if (!p_mngr) {
1292 		DP_NOTICE(p_hwfn, false, "Failed to allocate `struct ecore_cxt_mngr'\n");
1293 		return ECORE_NOMEM;
1294 	}
1295 
1296 	/* Initialize ILT client registers */
1297 	clients = p_mngr->clients;
1298 	clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1299 	clients[ILT_CLI_CDUC].last.reg  = ILT_CFG_REG(CDUC, LAST_ILT);
1300 	clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1301 
1302 	clients[ILT_CLI_QM].first.reg   = ILT_CFG_REG(QM, FIRST_ILT);
1303 	clients[ILT_CLI_QM].last.reg    = ILT_CFG_REG(QM, LAST_ILT);
1304 	clients[ILT_CLI_QM].p_size.reg  = ILT_CFG_REG(QM, P_SIZE);
1305 
1306 	clients[ILT_CLI_TM].first.reg   = ILT_CFG_REG(TM, FIRST_ILT);
1307 	clients[ILT_CLI_TM].last.reg    = ILT_CFG_REG(TM, LAST_ILT);
1308 	clients[ILT_CLI_TM].p_size.reg  = ILT_CFG_REG(TM, P_SIZE);
1309 
1310 	clients[ILT_CLI_SRC].first.reg  = ILT_CFG_REG(SRC, FIRST_ILT);
1311 	clients[ILT_CLI_SRC].last.reg   = ILT_CFG_REG(SRC, LAST_ILT);
1312 	clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1313 
1314 	clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1315 	clients[ILT_CLI_CDUT].last.reg  = ILT_CFG_REG(CDUT, LAST_ILT);
1316 	clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1317 
1318 	clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1319 	clients[ILT_CLI_TSDM].last.reg  = ILT_CFG_REG(TSDM, LAST_ILT);
1320 	clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1321 
1322 	/* default ILT page size for all clients is 64K */
1323 	for (i = 0; i < ILT_CLI_MAX; i++)
1324 		p_mngr->clients[i].p_size.val = p_hwfn->p_dev->ilt_page_size;
1325 
1326 	/* Initialize task sizes */
1327 	p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1328 	p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1329 
1330 	if (p_hwfn->p_dev->p_iov_info)
1331 		p_mngr->vf_count = p_hwfn->p_dev->p_iov_info->total_vfs;
1332 
1333 	/* Initialize the dynamic ILT allocation mutex */
1334 #ifdef CONFIG_ECORE_LOCK_ALLOC
1335 	OSAL_MUTEX_ALLOC(p_hwfn, &p_mngr->mutex);
1336 #endif
1337 	OSAL_MUTEX_INIT(&p_mngr->mutex);
1338 
1339 	/* Set the cxt mangr pointer priori to further allocations */
1340 	p_hwfn->p_cxt_mngr = p_mngr;
1341 
1342 	return ECORE_SUCCESS;
1343 }
1344 
1345 enum _ecore_status_t ecore_cxt_tables_alloc(struct ecore_hwfn *p_hwfn)
1346 {
1347 	enum _ecore_status_t    rc;
1348 
1349 	/* Allocate the ILT shadow table */
1350 	rc = ecore_ilt_shadow_alloc(p_hwfn);
1351 	if (rc) {
1352 		DP_NOTICE(p_hwfn, false, "Failed to allocate ilt memory\n");
1353 		goto tables_alloc_fail;
1354 	}
1355 
1356 	/* Allocate the T2  table */
1357 	rc = ecore_cxt_src_t2_alloc(p_hwfn);
1358 	if (rc) {
1359 		DP_NOTICE(p_hwfn, false, "Failed to allocate T2 memory\n");
1360 		goto tables_alloc_fail;
1361 	}
1362 
1363 	/* Allocate and initialize the acquired cids bitmaps */
1364 	rc = ecore_cid_map_alloc(p_hwfn);
1365 	if (rc) {
1366 		DP_NOTICE(p_hwfn, false, "Failed to allocate cid maps\n");
1367 		goto tables_alloc_fail;
1368 	}
1369 
1370 	return ECORE_SUCCESS;
1371 
1372 tables_alloc_fail:
1373 	ecore_cxt_mngr_free(p_hwfn);
1374 	return rc;
1375 }
1376 void ecore_cxt_mngr_free(struct ecore_hwfn *p_hwfn)
1377 {
1378 	if (!p_hwfn->p_cxt_mngr)
1379 		return;
1380 
1381 	ecore_cid_map_free(p_hwfn);
1382 	ecore_cxt_src_t2_free(p_hwfn);
1383 	ecore_ilt_shadow_free(p_hwfn);
1384 #ifdef CONFIG_ECORE_LOCK_ALLOC
1385 	OSAL_MUTEX_DEALLOC(&p_hwfn->p_cxt_mngr->mutex);
1386 #endif
1387 	OSAL_FREE(p_hwfn->p_dev, p_hwfn->p_cxt_mngr);
1388 
1389 	p_hwfn->p_cxt_mngr = OSAL_NULL;
1390 }
1391 
1392 void ecore_cxt_mngr_setup(struct ecore_hwfn *p_hwfn)
1393 {
1394 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1395 	struct ecore_cid_acquired_map *p_map;
1396 	struct ecore_conn_type_cfg *p_cfg;
1397 	int type;
1398 	u32 len;
1399 
1400 	/* Reset acquired cids */
1401 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1402 		u32 vf;
1403 
1404 		p_cfg = &p_mngr->conn_cfg[type];
1405 		if (p_cfg->cid_count) {
1406 			p_map = &p_mngr->acquired[type];
1407 			len = DIV_ROUND_UP(p_map->max_count,
1408 					   BITS_PER_MAP_WORD) *
1409 			      MAP_WORD_SIZE;
1410 			OSAL_MEM_ZERO(p_map->cid_map, len);
1411 		}
1412 
1413 		if (!p_cfg->cids_per_vf)
1414 			continue;
1415 
1416 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1417 			p_map = &p_mngr->acquired_vf[type][vf];
1418 			len = DIV_ROUND_UP(p_map->max_count,
1419 					   BITS_PER_MAP_WORD) *
1420 			      MAP_WORD_SIZE;
1421 			OSAL_MEM_ZERO(p_map->cid_map, len);
1422 		}
1423 	}
1424 }
1425 
1426 /* HW initialization helper (per Block, per phase) */
1427 
1428 /* CDU Common */
1429 #define CDUC_CXT_SIZE_SHIFT						\
1430 	CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1431 
1432 #define CDUC_CXT_SIZE_MASK						\
1433 	(CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1434 
1435 #define CDUC_BLOCK_WASTE_SHIFT						\
1436 	CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1437 
1438 #define CDUC_BLOCK_WASTE_MASK						\
1439 	(CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1440 
1441 #define CDUC_NCIB_SHIFT							\
1442 	CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1443 
1444 #define CDUC_NCIB_MASK							\
1445 	(CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1446 
1447 #define CDUT_TYPE0_CXT_SIZE_SHIFT					\
1448 	CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1449 
1450 #define CDUT_TYPE0_CXT_SIZE_MASK					\
1451 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >>				\
1452 	CDUT_TYPE0_CXT_SIZE_SHIFT)
1453 
1454 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT					\
1455 	CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1456 
1457 #define CDUT_TYPE0_BLOCK_WASTE_MASK					\
1458 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >>			\
1459 	CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1460 
1461 #define CDUT_TYPE0_NCIB_SHIFT						\
1462 	CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1463 
1464 #define CDUT_TYPE0_NCIB_MASK						\
1465 	(CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >>		\
1466 	CDUT_TYPE0_NCIB_SHIFT)
1467 
1468 #define CDUT_TYPE1_CXT_SIZE_SHIFT					\
1469 	CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1470 
1471 #define CDUT_TYPE1_CXT_SIZE_MASK					\
1472 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >>				\
1473 	CDUT_TYPE1_CXT_SIZE_SHIFT)
1474 
1475 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT					\
1476 	CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1477 
1478 #define CDUT_TYPE1_BLOCK_WASTE_MASK					\
1479 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >>			\
1480 	CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1481 
1482 #define CDUT_TYPE1_NCIB_SHIFT						\
1483 	CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1484 
1485 #define CDUT_TYPE1_NCIB_MASK						\
1486 	(CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >>		\
1487 	CDUT_TYPE1_NCIB_SHIFT)
1488 
1489 static void ecore_cdu_init_common(struct ecore_hwfn *p_hwfn)
1490 {
1491 	u32 page_sz, elems_per_page, block_waste,  cxt_size, cdu_params = 0;
1492 
1493 	/* CDUC - connection configuration */
1494 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1495 	cxt_size = CONN_CXT_SIZE(p_hwfn);
1496 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1497 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1498 
1499 	SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1500 	SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1501 	SET_FIELD(cdu_params, (u32)CDUC_NCIB, elems_per_page);
1502 	STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1503 
1504 	/* CDUT - type-0 tasks configuration */
1505 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1506 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1507 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1508 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1509 
1510 	/* cxt size and block-waste are multipes of 8 */
1511 	cdu_params = 0;
1512 	SET_FIELD(cdu_params, (u32)CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1513 	SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1514 	SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1515 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1516 
1517 	/* CDUT - type-1 tasks configuration */
1518 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1519 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1520 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1521 
1522 	/* cxt size and block-waste are multipes of 8 */
1523 	cdu_params = 0;
1524 	SET_FIELD(cdu_params, (u32)CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1525 	SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1526 	SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1527 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1528 }
1529 
1530 /* CDU PF */
1531 #define CDU_SEG_REG_TYPE_SHIFT		CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1532 #define CDU_SEG_REG_TYPE_MASK		0x1
1533 #define CDU_SEG_REG_OFFSET_SHIFT	0
1534 #define CDU_SEG_REG_OFFSET_MASK		CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1535 
1536 static void ecore_cdu_init_pf(struct ecore_hwfn *p_hwfn)
1537 {
1538 	struct ecore_ilt_client_cfg *p_cli;
1539 	struct ecore_tid_seg *p_seg;
1540 	u32 cdu_seg_params, offset;
1541 	int i;
1542 
1543 	static const u32 rt_type_offset_arr[] = {
1544 		CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1545 		CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1546 		CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1547 		CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1548 	};
1549 
1550 	static const u32 rt_type_offset_fl_arr[] = {
1551 		CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1552 		CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1553 		CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1554 		CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1555 	};
1556 
1557 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1558 
1559 	/* There are initializations only for CDUT during pf Phase */
1560 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1561 		/* Segment 0*/
1562 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
1563 		if (!p_seg)
1564 			continue;
1565 
1566 		/* Note: start_line is already adjusted for the CDU
1567 		 * segment register granularity, so we just need to
1568 		 * divide. Adjustment is implicit as we assume ILT
1569 		 * Page size is larger than 32K!
1570 		 */
1571 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1572 			 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1573 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1574 
1575 		cdu_seg_params = 0;
1576 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1577 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1578 		STORE_RT_REG(p_hwfn, rt_type_offset_arr[i],
1579 			     cdu_seg_params);
1580 
1581 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1582 			 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1583 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1584 
1585 		cdu_seg_params = 0;
1586 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1587 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1588 		STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i],
1589 			     cdu_seg_params);
1590 	}
1591 }
1592 
1593 void ecore_qm_init_pf(struct ecore_hwfn *p_hwfn, struct ecore_ptt *p_ptt,
1594 		      bool is_pf_loading)
1595 {
1596 	struct ecore_qm_info *qm_info = &p_hwfn->qm_info;
1597 	struct ecore_mcp_link_state *p_link;
1598 	struct ecore_qm_iids iids;
1599 
1600 	OSAL_MEM_ZERO(&iids, sizeof(iids));
1601 	ecore_cxt_qm_iids(p_hwfn, &iids);
1602 
1603 	p_link = &ECORE_LEADING_HWFN(p_hwfn->p_dev)->mcp_info->link_output;
1604 
1605 	ecore_qm_pf_rt_init(p_hwfn, p_ptt, p_hwfn->port_id,
1606 			    p_hwfn->rel_pf_id, qm_info->max_phys_tcs_per_port,
1607 			    is_pf_loading,
1608 			    iids.cids, iids.vf_cids, iids.tids,
1609 			    qm_info->start_pq,
1610 			    qm_info->num_pqs - qm_info->num_vf_pqs,
1611 			    qm_info->num_vf_pqs,
1612 			    qm_info->start_vport,
1613 			    qm_info->num_vports, qm_info->pf_wfq,
1614 			    qm_info->pf_rl, p_link->speed,
1615 			    p_hwfn->qm_info.qm_pq_params,
1616 			    p_hwfn->qm_info.qm_vport_params);
1617 }
1618 
1619 /* CM PF */
1620 static void ecore_cm_init_pf(struct ecore_hwfn *p_hwfn)
1621 {
1622 	STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET, ecore_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1623 }
1624 
1625 /* DQ PF */
1626 static void ecore_dq_init_pf(struct ecore_hwfn *p_hwfn)
1627 {
1628 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1629 	u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1630 
1631 	dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1632 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1633 
1634 	dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1635 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1636 
1637 	dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1638 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1639 
1640 	dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1641 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1642 
1643 	dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1644 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1645 
1646 	dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1647 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1648 
1649 	dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1650 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1651 
1652 	dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1653 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1654 
1655 	dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1656 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1657 
1658 	dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1659 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1660 
1661 	dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1662 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1663 
1664 	dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1665 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1666 
1667 	/* Connection types 6 & 7 are not in use, yet they must be configured
1668 	 * as the highest possible connection. Not configuring them means the
1669 	 * defaults will be  used, and with a large number of cids a bug may
1670 	 * occur, if the defaults will be smaller than dq_pf_max_cid /
1671 	 * dq_vf_max_cid.
1672 	 */
1673 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1674 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1675 
1676 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1677 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1678 }
1679 
1680 static void ecore_ilt_bounds_init(struct ecore_hwfn *p_hwfn)
1681 {
1682 	struct ecore_ilt_client_cfg *ilt_clients;
1683 	int i;
1684 
1685 	ilt_clients = p_hwfn->p_cxt_mngr->clients;
1686 	for_each_ilt_valid_client(i, ilt_clients) {
1687 		STORE_RT_REG(p_hwfn,
1688 			     ilt_clients[i].first.reg,
1689 			     ilt_clients[i].first.val);
1690 		STORE_RT_REG(p_hwfn,
1691 			     ilt_clients[i].last.reg,
1692 			     ilt_clients[i].last.val);
1693 		STORE_RT_REG(p_hwfn,
1694 			     ilt_clients[i].p_size.reg,
1695 			     ilt_clients[i].p_size.val);
1696 	}
1697 }
1698 
1699 static void ecore_ilt_vf_bounds_init(struct ecore_hwfn *p_hwfn)
1700 {
1701 	struct ecore_ilt_client_cfg *p_cli;
1702 	u32 blk_factor;
1703 
1704 	/* For simplicty  we set the 'block' to be an ILT page */
1705 	if (p_hwfn->p_dev->p_iov_info) {
1706 		struct ecore_hw_sriov_info *p_iov = p_hwfn->p_dev->p_iov_info;
1707 
1708 		STORE_RT_REG(p_hwfn,
1709 			     PSWRQ2_REG_VF_BASE_RT_OFFSET,
1710 			     p_iov->first_vf_in_pf);
1711 		STORE_RT_REG(p_hwfn,
1712 			     PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1713 			     p_iov->first_vf_in_pf + p_iov->total_vfs);
1714 	}
1715 
1716 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1717 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1718 	if (p_cli->active) {
1719 		STORE_RT_REG(p_hwfn,
1720 			     PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1721 			     blk_factor);
1722 		STORE_RT_REG(p_hwfn,
1723 			     PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1724 			     p_cli->pf_total_lines);
1725 		STORE_RT_REG(p_hwfn,
1726 			     PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1727 			     p_cli->vf_total_lines);
1728 	}
1729 
1730 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1731 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1732 	if (p_cli->active) {
1733 		STORE_RT_REG(p_hwfn,
1734 			     PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1735 			     blk_factor);
1736 		STORE_RT_REG(p_hwfn,
1737 			     PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1738 			     p_cli->pf_total_lines);
1739 		STORE_RT_REG(p_hwfn,
1740 			     PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1741 			     p_cli->vf_total_lines);
1742 	}
1743 
1744 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1745 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1746 	if (p_cli->active) {
1747 		STORE_RT_REG(p_hwfn,
1748 			     PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET,
1749 			     blk_factor);
1750 		STORE_RT_REG(p_hwfn,
1751 			     PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1752 			     p_cli->pf_total_lines);
1753 		STORE_RT_REG(p_hwfn,
1754 			     PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1755 			     p_cli->vf_total_lines);
1756 	}
1757 }
1758 
1759 /* ILT (PSWRQ2) PF */
1760 static void ecore_ilt_init_pf(struct ecore_hwfn *p_hwfn)
1761 {
1762 	struct ecore_ilt_client_cfg *clients;
1763 	struct ecore_cxt_mngr *p_mngr;
1764 	struct ecore_dma_mem *p_shdw;
1765 	u32 line, rt_offst, i;
1766 
1767 	ecore_ilt_bounds_init(p_hwfn);
1768 	ecore_ilt_vf_bounds_init(p_hwfn);
1769 
1770 	p_mngr  = p_hwfn->p_cxt_mngr;
1771 	p_shdw  = p_mngr->ilt_shadow;
1772 	clients = p_hwfn->p_cxt_mngr->clients;
1773 
1774 	for_each_ilt_valid_client(i, clients) {
1775 		/* Client's 1st val and RT array are absolute, ILT shadows'
1776 		 * lines are relative.
1777 		 */
1778 		line = clients[i].first.val - p_mngr->pf_start_line;
1779 		rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1780 			   clients[i].first.val * ILT_ENTRY_IN_REGS;
1781 
1782 		for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1783 		     line++, rt_offst += ILT_ENTRY_IN_REGS) {
1784 			u64 ilt_hw_entry = 0;
1785 
1786 			/** p_virt could be OSAL_NULL incase of dynamic
1787 			 *  allocation
1788 			 */
1789 			if (p_shdw[line].p_virt != OSAL_NULL) {
1790 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1791 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1792 					  (unsigned long long)(p_shdw[line].p_phys >> 12));
1793 
1794 				DP_VERBOSE(
1795 					p_hwfn, ECORE_MSG_ILT,
1796 					"Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1797 					rt_offst, line, i,
1798 					(unsigned long long)(p_shdw[line].p_phys >> 12));
1799 			}
1800 
1801 			STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1802 		}
1803 	}
1804 }
1805 
1806 /* SRC (Searcher) PF */
1807 static void ecore_src_init_pf(struct ecore_hwfn *p_hwfn)
1808 {
1809 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1810 	u32 rounded_conn_num, conn_num, conn_max;
1811 	struct ecore_src_iids src_iids;
1812 
1813 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
1814 	ecore_cxt_src_iids(p_mngr, &src_iids);
1815 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1816 	if (!conn_num)
1817 		return;
1818 
1819 	conn_max = OSAL_MAX_T(u32, conn_num, SRC_MIN_NUM_ELEMS);
1820 	rounded_conn_num = OSAL_ROUNDUP_POW_OF_TWO(conn_max);
1821 
1822 	STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1823 	STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1824 		     OSAL_LOG2(rounded_conn_num));
1825 
1826 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1827 			 p_hwfn->p_cxt_mngr->first_free);
1828 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1829 			 p_hwfn->p_cxt_mngr->last_free);
1830 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1831 		   "Configured SEARCHER for 0x%08x connections\n",
1832 		   conn_num);
1833 }
1834 
1835 /* Timers PF */
1836 #define TM_CFG_NUM_IDS_SHIFT		0
1837 #define TM_CFG_NUM_IDS_MASK		0xFFFFULL
1838 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT	16
1839 #define TM_CFG_PRE_SCAN_OFFSET_MASK	0x1FFULL
1840 #define TM_CFG_PARENT_PF_SHIFT		25
1841 #define TM_CFG_PARENT_PF_MASK		0x7ULL
1842 
1843 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT	30
1844 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK	0x1FFULL
1845 
1846 #define TM_CFG_TID_OFFSET_SHIFT		30
1847 #define TM_CFG_TID_OFFSET_MASK		0x7FFFFULL
1848 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT	49
1849 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK	0x1FFULL
1850 
1851 static void ecore_tm_init_pf(struct ecore_hwfn *p_hwfn)
1852 {
1853 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1854 	u32 active_seg_mask = 0, tm_offset, rt_reg;
1855 	struct ecore_tm_iids tm_iids;
1856 	u64 cfg_word;
1857 	u8 i;
1858 
1859 	OSAL_MEM_ZERO(&tm_iids, sizeof(tm_iids));
1860 	ecore_cxt_tm_iids(p_mngr, &tm_iids);
1861 
1862 	/* @@@TBD No pre-scan for now */
1863 
1864 	/* Note: We assume consecutive VFs for a PF */
1865 	for (i = 0; i < p_mngr->vf_count; i++) {
1866 		cfg_word = 0;
1867 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1868 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1869 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1870 		SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1871 
1872 		rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1873 			 (sizeof(cfg_word) / sizeof(u32)) *
1874 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1875 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1876 	}
1877 
1878 	cfg_word = 0;
1879 	SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1880 	SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1881 	SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);	  /* n/a for PF */
1882 	SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all   */
1883 
1884 	rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1885 		 (sizeof(cfg_word) / sizeof(u32)) *
1886 		 (NUM_OF_VFS(p_hwfn->p_dev) + p_hwfn->rel_pf_id);
1887 	STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1888 
1889 	/* enale scan */
1890 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1891 		     tm_iids.pf_cids  ? 0x1 : 0x0);
1892 
1893 	/* @@@TBD how to enable the scan for the VFs */
1894 
1895 	tm_offset = tm_iids.per_vf_cids;
1896 
1897 	/* Note: We assume consecutive VFs for a PF */
1898 	for (i = 0; i < p_mngr->vf_count; i++) {
1899 		cfg_word = 0;
1900 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1901 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1902 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1903 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1904 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1905 
1906 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1907 			 (sizeof(cfg_word) / sizeof(u32)) *
1908 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1909 
1910 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1911 	}
1912 
1913 	tm_offset = tm_iids.pf_cids;
1914 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1915 		cfg_word = 0;
1916 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1917 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1918 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1919 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1920 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1921 
1922 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1923 			 (sizeof(cfg_word) / sizeof(u32)) *
1924 			 (NUM_OF_VFS(p_hwfn->p_dev) +
1925 			 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1926 
1927 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1928 		active_seg_mask |= (tm_iids.pf_tids[i] ? (1 << i) : 0);
1929 
1930 		tm_offset += tm_iids.pf_tids[i];
1931 	}
1932 
1933 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn))
1934 		active_seg_mask = 0;
1935 
1936 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1937 
1938 	/* @@@TBD how to enable the scan for the VFs */
1939 }
1940 
1941 static void ecore_prs_init_common(struct ecore_hwfn *p_hwfn)
1942 {
1943 	if ((p_hwfn->hw_info.personality == ECORE_PCI_FCOE) &&
1944 	    p_hwfn->pf_params.fcoe_pf_params.is_target)
1945 		STORE_RT_REG(p_hwfn, PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1946 }
1947 
1948 static void ecore_prs_init_pf(struct ecore_hwfn *p_hwfn)
1949 {
1950 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1951 	struct ecore_conn_type_cfg *p_fcoe;
1952 	struct ecore_tid_seg *p_tid;
1953 
1954 	p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1955 
1956 	/* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1957 	if (!p_fcoe->cid_count)
1958 		return;
1959 
1960 	p_tid = &p_fcoe->tid_seg[ECORE_CXT_FCOE_TID_SEG];
1961 	if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1962 		STORE_RT_REG_AGG(p_hwfn,
1963 				 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1964 				 p_tid->count);
1965 	} else {
1966 		STORE_RT_REG_AGG(p_hwfn,
1967 				PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1968 				p_tid->count);
1969 	}
1970 }
1971 
1972 void ecore_cxt_hw_init_common(struct ecore_hwfn *p_hwfn)
1973 {
1974 	/* CDU configuration */
1975 	ecore_cdu_init_common(p_hwfn);
1976 	ecore_prs_init_common(p_hwfn);
1977 }
1978 
1979 void ecore_cxt_hw_init_pf(struct ecore_hwfn *p_hwfn, struct ecore_ptt *p_ptt)
1980 {
1981 	ecore_qm_init_pf(p_hwfn, p_ptt, true);
1982 	ecore_cm_init_pf(p_hwfn);
1983 	ecore_dq_init_pf(p_hwfn);
1984 	ecore_cdu_init_pf(p_hwfn);
1985 	ecore_ilt_init_pf(p_hwfn);
1986 	ecore_src_init_pf(p_hwfn);
1987 	ecore_tm_init_pf(p_hwfn);
1988 	ecore_prs_init_pf(p_hwfn);
1989 }
1990 
1991 enum _ecore_status_t _ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
1992 					    enum protocol_type type,
1993 					    u32 *p_cid, u8 vfid)
1994 {
1995 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1996 	struct ecore_cid_acquired_map *p_map;
1997 	u32 rel_cid;
1998 
1999 	if (type >= MAX_CONN_TYPES) {
2000 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
2001 		return ECORE_INVAL;
2002 	}
2003 
2004 	if (vfid >= COMMON_MAX_NUM_VFS && vfid != ECORE_CXT_PF_CID) {
2005 		DP_NOTICE(p_hwfn, true, "VF [%02x] is out of range\n", vfid);
2006 		return ECORE_INVAL;
2007 	}
2008 
2009 	/* Determine the right map to take this CID from */
2010 	if (vfid == ECORE_CXT_PF_CID)
2011 		p_map = &p_mngr->acquired[type];
2012 	else
2013 		p_map = &p_mngr->acquired_vf[type][vfid];
2014 
2015 	if (p_map->cid_map == OSAL_NULL) {
2016 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
2017 		return ECORE_INVAL;
2018 	}
2019 
2020 	rel_cid = OSAL_FIND_FIRST_ZERO_BIT(p_map->cid_map,
2021 					   p_map->max_count);
2022 
2023 	if (rel_cid >= p_map->max_count) {
2024 		DP_NOTICE(p_hwfn, false, "no CID available for protocol %d\n",
2025 			  type);
2026 		return ECORE_NORESOURCES;
2027 	}
2028 
2029 	OSAL_SET_BIT(rel_cid, p_map->cid_map);
2030 
2031 	*p_cid = rel_cid + p_map->start_cid;
2032 
2033 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
2034 		   "Acquired cid 0x%08x [rel. %08x] vfid %02x type %d\n",
2035 		   *p_cid, rel_cid, vfid, type);
2036 
2037 	return ECORE_SUCCESS;
2038 }
2039 
2040 enum _ecore_status_t ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
2041 					   enum protocol_type type,
2042 					   u32 *p_cid)
2043 {
2044 	return _ecore_cxt_acquire_cid(p_hwfn, type, p_cid, ECORE_CXT_PF_CID);
2045 }
2046 
2047 static bool ecore_cxt_test_cid_acquired(struct ecore_hwfn *p_hwfn,
2048 					u32 cid, u8 vfid,
2049 					enum protocol_type *p_type,
2050 					struct ecore_cid_acquired_map **pp_map)
2051 {
2052 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2053 	u32 rel_cid;
2054 
2055 	/* Iterate over protocols and find matching cid range */
2056 	for (*p_type = 0; *p_type < MAX_CONN_TYPES; (*p_type)++) {
2057 		if (vfid == ECORE_CXT_PF_CID)
2058 			*pp_map = &p_mngr->acquired[*p_type];
2059 		else
2060 			*pp_map = &p_mngr->acquired_vf[*p_type][vfid];
2061 
2062 		if (!((*pp_map)->cid_map))
2063 			continue;
2064 		if (cid >= (*pp_map)->start_cid &&
2065 		    cid < (*pp_map)->start_cid + (*pp_map)->max_count) {
2066 			break;
2067 		}
2068 	}
2069 
2070 	if (*p_type == MAX_CONN_TYPES) {
2071 		DP_NOTICE(p_hwfn, true, "Invalid CID %d vfid %02x", cid, vfid);
2072 		goto fail;
2073 	}
2074 
2075 	rel_cid = cid - (*pp_map)->start_cid;
2076 	if (!OSAL_TEST_BIT(rel_cid, (*pp_map)->cid_map)) {
2077 		DP_NOTICE(p_hwfn, true,
2078 			  "CID %d [vifd %02x] not acquired", cid, vfid);
2079 		goto fail;
2080 	}
2081 
2082 	return true;
2083 fail:
2084 	*p_type = MAX_CONN_TYPES;
2085 	*pp_map = OSAL_NULL;
2086 	return false;
2087 }
2088 
2089 void _ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid, u8 vfid)
2090 {
2091 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2092 	enum protocol_type type;
2093 	bool b_acquired;
2094 	u32 rel_cid;
2095 
2096 	if (vfid != ECORE_CXT_PF_CID && vfid > COMMON_MAX_NUM_VFS) {
2097 		DP_NOTICE(p_hwfn, true,
2098 			  "Trying to return incorrect CID belonging to VF %02x\n",
2099 			  vfid);
2100 		return;
2101 	}
2102 
2103 	/* Test acquired and find matching per-protocol map */
2104 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, cid, vfid,
2105 						 &type, &p_map);
2106 
2107 	if (!b_acquired)
2108 		return;
2109 
2110 	rel_cid = cid - p_map->start_cid;
2111 	OSAL_CLEAR_BIT(rel_cid, p_map->cid_map);
2112 
2113 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
2114 		   "Released CID 0x%08x [rel. %08x] vfid %02x type %d\n",
2115 		   cid, rel_cid, vfid, type);
2116 }
2117 
2118 void ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid)
2119 {
2120 	_ecore_cxt_release_cid(p_hwfn, cid, ECORE_CXT_PF_CID);
2121 }
2122 
2123 enum _ecore_status_t ecore_cxt_get_cid_info(struct ecore_hwfn *p_hwfn,
2124 					    struct ecore_cxt_info *p_info)
2125 {
2126 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2127 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2128 	u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
2129 	enum protocol_type type;
2130 	bool b_acquired;
2131 
2132 	/* Test acquired and find matching per-protocol map */
2133 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, p_info->iid,
2134 						 ECORE_CXT_PF_CID,
2135 						 &type, &p_map);
2136 
2137 	if (!b_acquired)
2138 		return ECORE_INVAL;
2139 
2140 	/* set the protocl type */
2141 	p_info->type = type;
2142 
2143 	/* compute context virtual pointer */
2144 	hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
2145 
2146 	conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
2147 	cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
2148 	line = p_info->iid / cxts_per_p;
2149 
2150 	/* Make sure context is allocated (dynamic allocation) */
2151 	if (!p_mngr->ilt_shadow[line].p_virt)
2152 		return ECORE_INVAL;
2153 
2154 	p_info->p_cxt = (u8 *)p_mngr->ilt_shadow[line].p_virt +
2155 			      p_info->iid % cxts_per_p * conn_cxt_size;
2156 
2157 	DP_VERBOSE(p_hwfn, (ECORE_MSG_ILT | ECORE_MSG_CXT),
2158 		   "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
2159 		   (p_info->iid / cxts_per_p), p_info->p_cxt, p_info->iid);
2160 
2161 	return ECORE_SUCCESS;
2162 }
2163 
2164 static void ecore_rdma_set_pf_params(struct ecore_hwfn *p_hwfn,
2165 				     struct ecore_rdma_pf_params *p_params,
2166 				     u32 num_tasks)
2167 {
2168 	u32 num_cons, num_qps;
2169 	enum protocol_type proto;
2170 
2171 	/* The only case RDMA personality can be overriden is if NVRAM is
2172 	 * configured with ETH_RDMA or if no rdma protocol was requested
2173 	 */
2174 	switch (p_params->rdma_protocol) {
2175 	case ECORE_RDMA_PROTOCOL_DEFAULT:
2176 		if (p_hwfn->mcp_info->func_info.protocol ==
2177 		    ECORE_PCI_ETH_RDMA) {
2178 			DP_NOTICE(p_hwfn, false,
2179 				  "Current day drivers don't support RoCE & iWARP. Default to RoCE-only\n");
2180 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2181 		}
2182 		break;
2183 	case ECORE_RDMA_PROTOCOL_NONE:
2184 		p_hwfn->hw_info.personality = ECORE_PCI_ETH;
2185 		return; /* intentional... nothing left to do... */
2186 	case ECORE_RDMA_PROTOCOL_ROCE:
2187 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2188 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2189 		break;
2190 	case ECORE_RDMA_PROTOCOL_IWARP:
2191 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2192 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_IWARP;
2193 		break;
2194 	}
2195 
2196 	switch (p_hwfn->hw_info.personality) {
2197 	case ECORE_PCI_ETH_IWARP:
2198 		/* Each QP requires one connection */
2199 		num_cons = OSAL_MIN_T(u32, IWARP_MAX_QPS, p_params->num_qps);
2200 #ifdef CONFIG_ECORE_IWARP /* required for the define */
2201 		/* additional connections required for passive tcp handling */
2202 		num_cons += ECORE_IWARP_PREALLOC_CNT;
2203 #endif
2204 		proto = PROTOCOLID_IWARP;
2205 		break;
2206 	case ECORE_PCI_ETH_ROCE:
2207 		num_qps = OSAL_MIN_T(u32, ROCE_MAX_QPS, p_params->num_qps);
2208 		num_cons = num_qps * 2; /* each QP requires two connections */
2209 		proto = PROTOCOLID_ROCE;
2210 		break;
2211 	default:
2212 		return;
2213 	}
2214 
2215 	if (num_cons && num_tasks) {
2216 		u32 num_srqs, num_xrc_srqs, max_xrc_srqs, page_size;
2217 
2218 		ecore_cxt_set_proto_cid_count(p_hwfn, proto,
2219 					      num_cons, 0);
2220 
2221 		/* Deliberatly passing ROCE for tasks id. This is because
2222 		 * iWARP / RoCE share the task id.
2223 		 */
2224 		ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
2225 					      ECORE_CXT_ROCE_TID_SEG,
2226 					      1, /* RoCE segment type */
2227 					      num_tasks,
2228 					      false); /* !force load */
2229 
2230 		num_srqs = OSAL_MIN_T(u32, ECORE_RDMA_MAX_SRQS,
2231 				      p_params->num_srqs);
2232 
2233 		/* XRC SRQs populate a single ILT page */
2234 		page_size = ecore_cxt_get_ilt_page_size(p_hwfn, ILT_CLI_TSDM);
2235 		max_xrc_srqs =  page_size / XRC_SRQ_CXT_SIZE;
2236 		max_xrc_srqs = OSAL_MIN_T(u32, max_xrc_srqs, ECORE_RDMA_MAX_XRC_SRQS);
2237 
2238 		num_xrc_srqs = OSAL_MIN_T(u32, p_params->num_xrc_srqs,
2239 					  max_xrc_srqs);
2240 		ecore_cxt_set_srq_count(p_hwfn, num_srqs, num_xrc_srqs);
2241 
2242 	} else {
2243 		DP_INFO(p_hwfn->p_dev,
2244 			"RDMA personality used without setting params!\n");
2245 	}
2246 }
2247 
2248 enum _ecore_status_t ecore_cxt_set_pf_params(struct ecore_hwfn *p_hwfn,
2249 					     u32 rdma_tasks)
2250 {
2251 	/* Set the number of required CORE connections */
2252 	u32 core_cids = 1; /* SPQ */
2253 
2254 	if (p_hwfn->using_ll2)
2255 		core_cids += 4; /* @@@TBD Use the proper #define */
2256 
2257 	ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
2258 
2259 	switch (p_hwfn->hw_info.personality) {
2260 	case ECORE_PCI_ETH_RDMA:
2261 	case ECORE_PCI_ETH_IWARP:
2262 	case ECORE_PCI_ETH_ROCE:
2263 	{
2264 		ecore_rdma_set_pf_params(p_hwfn,
2265 					 &p_hwfn->pf_params.rdma_pf_params,
2266 					 rdma_tasks);
2267 
2268 		/* no need for break since RoCE coexist with Ethernet */
2269 	}
2270 	case ECORE_PCI_ETH:
2271 	{
2272 		u32 count = 0;
2273 
2274 		struct ecore_eth_pf_params *p_params =
2275 					&p_hwfn->pf_params.eth_pf_params;
2276 
2277 		if (!p_params->num_vf_cons)
2278 			p_params->num_vf_cons = ETH_PF_PARAMS_VF_CONS_DEFAULT;
2279 		ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
2280 					      p_params->num_cons,
2281 					      p_params->num_vf_cons);
2282 
2283 		count = p_params->num_arfs_filters;
2284 
2285 		if (!OSAL_TEST_BIT(ECORE_MF_DISABLE_ARFS,
2286 				   &p_hwfn->p_dev->mf_bits))
2287 			p_hwfn->p_cxt_mngr->arfs_count = count;
2288 
2289 		break;
2290 	}
2291 	case ECORE_PCI_FCOE:
2292 	{
2293 		struct ecore_fcoe_pf_params *p_params;
2294 
2295 		p_params = &p_hwfn->pf_params.fcoe_pf_params;
2296 
2297 		if (p_params->num_cons && p_params->num_tasks) {
2298 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_FCOE,
2299 						      p_params->num_cons, 0);
2300 
2301 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
2302 						      ECORE_CXT_FCOE_TID_SEG,
2303 						      0, /* segment type */
2304 						      p_params->num_tasks,
2305 						      true);
2306 		} else {
2307 			DP_INFO(p_hwfn->p_dev,
2308 				"Fcoe personality used without setting params!\n");
2309 		}
2310 		break;
2311 	}
2312 	case ECORE_PCI_ISCSI:
2313 	{
2314 		struct ecore_iscsi_pf_params *p_params;
2315 
2316 		p_params = &p_hwfn->pf_params.iscsi_pf_params;
2317 
2318 		if (p_params->num_cons && p_params->num_tasks) {
2319 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ISCSI,
2320 						      p_params->num_cons, 0);
2321 
2322 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ISCSI,
2323 						      ECORE_CXT_ISCSI_TID_SEG,
2324 						      0, /* segment type */
2325 						      p_params->num_tasks,
2326 						      true);
2327 		} else {
2328 			DP_INFO(p_hwfn->p_dev,
2329 				"Iscsi personality used without setting params!\n");
2330 		}
2331 		break;
2332 	}
2333 	default:
2334 		return ECORE_INVAL;
2335 	}
2336 
2337 	return ECORE_SUCCESS;
2338 }
2339 
2340 enum _ecore_status_t ecore_cxt_get_tid_mem_info(struct ecore_hwfn *p_hwfn,
2341 						struct ecore_tid_mem *p_info)
2342 {
2343 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2344 	u32 proto, seg, total_lines, i, shadow_line;
2345 	struct ecore_ilt_client_cfg *p_cli;
2346 	struct ecore_ilt_cli_blk *p_fl_seg;
2347 	struct ecore_tid_seg *p_seg_info;
2348 
2349 	/* Verify the personality */
2350 	switch (p_hwfn->hw_info.personality) {
2351 	case ECORE_PCI_FCOE:
2352 		proto = PROTOCOLID_FCOE;
2353 		seg = ECORE_CXT_FCOE_TID_SEG;
2354 		break;
2355 	case ECORE_PCI_ISCSI:
2356 		proto = PROTOCOLID_ISCSI;
2357 		seg = ECORE_CXT_ISCSI_TID_SEG;
2358 		break;
2359 	default:
2360 		return ECORE_INVAL;
2361 	}
2362 
2363 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2364 	if (!p_cli->active) {
2365 		return ECORE_INVAL;
2366 	}
2367 
2368 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2369 	if (!p_seg_info->has_fl_mem)
2370 		return ECORE_INVAL;
2371 
2372 	p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2373 	total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2374 				   p_fl_seg->real_size_in_page);
2375 
2376 	for (i = 0; i < total_lines; i++) {
2377 		shadow_line = i + p_fl_seg->start_line -
2378 			      p_hwfn->p_cxt_mngr->pf_start_line;
2379 		p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
2380 	}
2381 	p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2382 			p_fl_seg->real_size_in_page;
2383 	p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2384 	p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2385 				     p_info->tid_size;
2386 
2387 	return ECORE_SUCCESS;
2388 }
2389 
2390 /* This function is very RoCE oriented, if another protocol in the future
2391  * will want this feature we'll need to modify the function to be more generic
2392  */
2393 enum _ecore_status_t
2394 ecore_cxt_dynamic_ilt_alloc(struct ecore_hwfn *p_hwfn,
2395 			    enum ecore_cxt_elem_type elem_type,
2396 			    u32 iid)
2397 {
2398 	u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2399 	struct ecore_ilt_client_cfg *p_cli;
2400 	struct ecore_ilt_cli_blk *p_blk;
2401 	struct ecore_ptt *p_ptt;
2402 	dma_addr_t p_phys;
2403 	u64 ilt_hw_entry;
2404 	void *p_virt;
2405 	enum _ecore_status_t rc = ECORE_SUCCESS;
2406 
2407 	switch (elem_type) {
2408 	case ECORE_ELEM_CXT:
2409 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2410 		elem_size = CONN_CXT_SIZE(p_hwfn);
2411 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2412 		break;
2413 	case ECORE_ELEM_SRQ:
2414 		/* The first ILT page is not used for regular SRQs. Skip it. */
2415 		iid += ecore_cxt_srqs_per_page(p_hwfn);
2416 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2417 		elem_size = SRQ_CXT_SIZE;
2418 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2419 		break;
2420 	case ECORE_ELEM_XRC_SRQ:
2421 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2422 		elem_size = XRC_SRQ_CXT_SIZE;
2423 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2424 		break;
2425 	case ECORE_ELEM_TASK:
2426 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2427 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2428 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2429 		break;
2430 	default:
2431 		DP_NOTICE(p_hwfn, false,
2432 			  "ECORE_INVALID elem type = %d", elem_type);
2433 		return ECORE_INVAL;
2434 	}
2435 
2436 	/* Calculate line in ilt */
2437 	hw_p_size = p_cli->p_size.val;
2438 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2439 	line = p_blk->start_line + (iid / elems_per_p);
2440 	shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2441 
2442 	/* If line is already allocated, do nothing, otherwise allocate it and
2443 	 * write it to the PSWRQ2 registers.
2444 	 * This section can be run in parallel from different contexts and thus
2445 	 * a mutex protection is needed.
2446 	 */
2447 #ifdef _NTDDK_
2448 #pragma warning(suppress : 28121)
2449 #endif
2450 	OSAL_MUTEX_ACQUIRE(&p_hwfn->p_cxt_mngr->mutex);
2451 
2452 	if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2453 		goto out0;
2454 
2455 	p_ptt = ecore_ptt_acquire(p_hwfn);
2456 	if (!p_ptt) {
2457 		DP_NOTICE(p_hwfn, false,
2458 			  "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2459 		rc = ECORE_TIMEOUT;
2460 		goto out0;
2461 	}
2462 
2463 	p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
2464 					 &p_phys,
2465 					 p_blk->real_size_in_page);
2466 	if (!p_virt) {
2467 		rc = ECORE_NOMEM;
2468 		goto out1;
2469 	}
2470 	OSAL_MEM_ZERO(p_virt, p_blk->real_size_in_page);
2471 
2472 	/* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2473 	 * to compensate for a HW bug, but it is configured even if DIF is not
2474 	 * enabled. This is harmless and allows us to avoid a dedicated API. We
2475 	 * configure the field for all of the contexts on the newly allocated
2476 	 * page.
2477 	 */
2478 	if (elem_type == ECORE_ELEM_TASK) {
2479 		u32 elem_i;
2480 		u8 *elem_start = (u8 *)p_virt;
2481 		union type1_task_context *elem;
2482 
2483 		for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2484 			elem = (union type1_task_context *)elem_start;
2485 			SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2486 				  TDIF_TASK_CONTEXT_REF_TAG_MASK , 0xf);
2487 			elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2488 		}
2489 	}
2490 
2491 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2492 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2493 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2494 		p_blk->real_size_in_page;
2495 
2496 	/* compute absolute offset */
2497 	reg_offset = PSWRQ2_REG_ILT_MEMORY +
2498 		     (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2499 
2500 	ilt_hw_entry = 0;
2501 	SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2502 	SET_FIELD(ilt_hw_entry,
2503 		  ILT_ENTRY_PHY_ADDR,
2504 		  (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2505 
2506 	/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2507 	ecore_dmae_host2grc(p_hwfn, p_ptt, (u64)(osal_uintptr_t)&ilt_hw_entry,
2508 			    reg_offset, sizeof(ilt_hw_entry) / sizeof(u32),
2509 			    OSAL_NULL /* default parameters */);
2510 
2511 	if (elem_type == ECORE_ELEM_CXT) {
2512 		u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2513 					 elems_per_p;
2514 
2515 		/* Update the relevant register in the parser */
2516 		ecore_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2517 			 last_cid_allocated - 1);
2518 
2519 		/* RoCE w/a -> we don't write to the prs search reg until first
2520 		 * cid is allocated. This is because the prs checks
2521 		 * last_cid-1 >=0 making 0 a valid value... this will cause
2522 		 * the a context load to occur on a RoCE packet received with
2523 		 * cid=0 even before context was initialized, can happen with a
2524 		 * stray packet from switch or a packet with crc-error
2525 		 */
2526 
2527 		if (!p_hwfn->b_rdma_enabled_in_prs) {
2528 			/* Enable Rdma search */
2529 			ecore_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2530 			p_hwfn->b_rdma_enabled_in_prs = true;
2531 		}
2532 	}
2533 
2534 out1:
2535 	ecore_ptt_release(p_hwfn, p_ptt);
2536 out0:
2537 	OSAL_MUTEX_RELEASE(&p_hwfn->p_cxt_mngr->mutex);
2538 
2539 	return rc;
2540 }
2541 
2542 /* This function is very RoCE oriented, if another protocol in the future
2543  * will want this feature we'll need to modify the function to be more generic
2544  */
2545 enum _ecore_status_t
2546 ecore_cxt_free_ilt_range(struct ecore_hwfn *p_hwfn,
2547 			 enum ecore_cxt_elem_type elem_type,
2548 			 u32 start_iid, u32 count)
2549 {
2550 	u32 start_line, end_line, shadow_start_line, shadow_end_line;
2551 	u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2552 	struct ecore_ilt_client_cfg *p_cli;
2553 	struct ecore_ilt_cli_blk *p_blk;
2554 	u32 end_iid = start_iid + count;
2555 	struct ecore_ptt *p_ptt;
2556 	u64 ilt_hw_entry = 0;
2557 	u32 i;
2558 
2559 	switch (elem_type) {
2560 	case ECORE_ELEM_CXT:
2561 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2562 		elem_size = CONN_CXT_SIZE(p_hwfn);
2563 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2564 		break;
2565 	case ECORE_ELEM_SRQ:
2566 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2567 		elem_size = SRQ_CXT_SIZE;
2568 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2569 		break;
2570 	case ECORE_ELEM_TASK:
2571 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2572 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2573 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2574 		break;
2575 	default:
2576 		DP_NOTICE(p_hwfn, false,
2577 			  "ECORE_INVALID elem type = %d", elem_type);
2578 		return ECORE_INVAL;
2579 	}
2580 
2581 	/* Calculate line in ilt */
2582 	hw_p_size = p_cli->p_size.val;
2583 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2584 	start_line = p_blk->start_line + (start_iid / elems_per_p);
2585 	end_line = p_blk->start_line + (end_iid / elems_per_p);
2586 	if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2587 		end_line--;
2588 
2589 	shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2590 	shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2591 
2592 	p_ptt = ecore_ptt_acquire(p_hwfn);
2593 	if (!p_ptt) {
2594 		DP_NOTICE(p_hwfn, false, "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2595 		return ECORE_TIMEOUT;
2596 	}
2597 
2598 	for (i = shadow_start_line; i < shadow_end_line; i++) {
2599 		if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2600 			continue;
2601 
2602 		OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
2603 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2604 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys,
2605 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].size);
2606 
2607 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = OSAL_NULL;
2608 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2609 		p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2610 
2611 		/* compute absolute offset */
2612 		reg_offset = PSWRQ2_REG_ILT_MEMORY +
2613 			     ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2614 			      ILT_ENTRY_IN_REGS);
2615 
2616 		/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2617 		 * wide-bus.
2618 		 */
2619 		ecore_dmae_host2grc(p_hwfn, p_ptt,
2620 				    (u64)(osal_uintptr_t)&ilt_hw_entry,
2621 				    reg_offset,
2622 				    sizeof(ilt_hw_entry) / sizeof(u32),
2623 				    OSAL_NULL /* default parameters */);
2624 	}
2625 
2626 	ecore_ptt_release(p_hwfn, p_ptt);
2627 
2628 	return ECORE_SUCCESS;
2629 }
2630 
2631 enum _ecore_status_t ecore_cxt_get_task_ctx(struct ecore_hwfn *p_hwfn,
2632 					    u32 tid,
2633 					    u8 ctx_type,
2634 					    void **pp_task_ctx)
2635 {
2636 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2637 	struct ecore_ilt_client_cfg *p_cli;
2638 	struct ecore_tid_seg *p_seg_info;
2639 	struct ecore_ilt_cli_blk *p_seg;
2640 	u32 num_tids_per_block;
2641 	u32 tid_size, ilt_idx;
2642 	u32 total_lines;
2643 	u32 proto, seg;
2644 
2645 	/* Verify the personality */
2646 	switch (p_hwfn->hw_info.personality) {
2647 	case ECORE_PCI_FCOE:
2648 		proto = PROTOCOLID_FCOE;
2649 		seg = ECORE_CXT_FCOE_TID_SEG;
2650 		break;
2651 	case ECORE_PCI_ISCSI:
2652 		proto = PROTOCOLID_ISCSI;
2653 		seg = ECORE_CXT_ISCSI_TID_SEG;
2654 		break;
2655 	default:
2656 		return ECORE_INVAL;
2657 	}
2658 
2659 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2660 	if (!p_cli->active) {
2661 		return ECORE_INVAL;
2662 	}
2663 
2664 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2665 
2666 	if (ctx_type == ECORE_CTX_WORKING_MEM) {
2667 		p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2668 	} else if (ctx_type == ECORE_CTX_FL_MEM) {
2669 		if (!p_seg_info->has_fl_mem) {
2670 			return ECORE_INVAL;
2671 		}
2672 		p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2673 	} else {
2674 		return ECORE_INVAL;
2675 	}
2676 	total_lines = DIV_ROUND_UP(p_seg->total_size,
2677 				   p_seg->real_size_in_page);
2678 	tid_size = p_mngr->task_type_size[p_seg_info->type];
2679 	num_tids_per_block = p_seg->real_size_in_page / tid_size;
2680 
2681 	if (total_lines < tid/num_tids_per_block)
2682 		return ECORE_INVAL;
2683 
2684 	ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2685 		  p_mngr->pf_start_line;
2686 	*pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2687 			     (tid % num_tids_per_block) * tid_size;
2688 
2689 	return ECORE_SUCCESS;
2690 }
2691