1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License, v.1,  (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://opensource.org/licenses/CDDL-1.0.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 
22 /*
23 * Copyright 2014-2017 Cavium, Inc.
24 * The contents of this file are subject to the terms of the Common Development
25 * and Distribution License, v.1,  (the "License").
26 
27 * You may not use this file except in compliance with the License.
28 
29 * You can obtain a copy of the License at available
30 * at http://opensource.org/licenses/CDDL-1.0
31 
32 * See the License for the specific language governing permissions and
33 * limitations under the License.
34 */
35 
36 #include "bcm_osal.h"
37 #include "reg_addr.h"
38 #include "common_hsi.h"
39 #include "ecore_hsi_common.h"
40 #include "ecore_hsi_eth.h"
41 #include "tcp_common.h"
42 #include "ecore_hsi_iscsi.h"
43 #include "ecore_hsi_fcoe.h"
44 #include "ecore_hsi_roce.h"
45 #include "ecore_hsi_iwarp.h"
46 #include "ecore_rt_defs.h"
47 #include "ecore_status.h"
48 #include "ecore.h"
49 #include "ecore_init_ops.h"
50 #include "ecore_init_fw_funcs.h"
51 #include "ecore_cxt.h"
52 #include "ecore_hw.h"
53 #include "ecore_dev_api.h"
54 #include "ecore_sriov.h"
55 #include "ecore_roce.h"
56 #include "ecore_mcp.h"
57 
58 /* Max number of connection types in HW (DQ/CDU etc.) */
59 #define MAX_CONN_TYPES		PROTOCOLID_COMMON
60 #define NUM_TASK_TYPES		2
61 #define NUM_TASK_PF_SEGMENTS	4
62 #define NUM_TASK_VF_SEGMENTS	1
63 
64 /* Doorbell-Queue constants */
65 #define DQ_RANGE_SHIFT	4
66 #define DQ_RANGE_ALIGN	(1 << DQ_RANGE_SHIFT)
67 
68 /* Searcher constants */
69 #define SRC_MIN_NUM_ELEMS 256
70 
71 /* Timers constants */
72 #define TM_SHIFT	7
73 #define TM_ALIGN	(1 << TM_SHIFT)
74 #define TM_ELEM_SIZE	4
75 
76 /* ILT constants */
77 /* If for some reason, HW P size is modified to be less than 32K,
78  * special handling needs to be made for CDU initialization
79  */
80 #ifdef CONFIG_ECORE_ROCE
81 /* For RoCE we configure to 64K to cover for RoCE max tasks 256K purpose. Can be
82  * optimized with resource management scheme
83  */
84 #define ILT_DEFAULT_HW_P_SIZE	4
85 #else
86 #define ILT_DEFAULT_HW_P_SIZE	3
87 #endif
88 
89 #define ILT_PAGE_IN_BYTES(hw_p_size)	(1U << ((hw_p_size) + 12))
90 #define ILT_CFG_REG(cli, reg)		PSWRQ2_REG_##cli##_##reg##_RT_OFFSET
91 
92 /* ILT entry structure */
93 #define ILT_ENTRY_PHY_ADDR_MASK		0x000FFFFFFFFFFFULL
94 #define ILT_ENTRY_PHY_ADDR_SHIFT	0
95 #define ILT_ENTRY_VALID_MASK		0x1ULL
96 #define ILT_ENTRY_VALID_SHIFT		52
97 #define ILT_ENTRY_IN_REGS		2
98 #define ILT_REG_SIZE_IN_BYTES		4
99 
100 /* connection context union */
101 union conn_context {
102 	struct core_conn_context  core_ctx;
103 	struct eth_conn_context	  eth_ctx;
104 	struct iscsi_conn_context iscsi_ctx;
105 	struct fcoe_conn_context  fcoe_ctx;
106 	struct roce_conn_context  roce_ctx;
107 };
108 
109 /* TYPE-0 task context - iSCSI, FCOE */
110 union type0_task_context {
111 	struct iscsi_task_context iscsi_ctx;
112 	struct fcoe_task_context  fcoe_ctx;
113 };
114 
115 /* TYPE-1 task context - ROCE */
116 union type1_task_context {
117 	struct rdma_task_context roce_ctx;
118 };
119 
120 struct src_ent {
121 	u8  opaque[56];
122 	u64 next;
123 };
124 
125 #define CDUT_SEG_ALIGNMET 3 /* in 4k chunks */
126 #define CDUT_SEG_ALIGNMET_IN_BYTES (1 << (CDUT_SEG_ALIGNMET + 12))
127 
128 #define CONN_CXT_SIZE(p_hwfn) \
129 	ALIGNED_TYPE_SIZE(union conn_context, p_hwfn)
130 
131 #define SRQ_CXT_SIZE (sizeof(struct rdma_srq_context))
132 
133 #define TYPE0_TASK_CXT_SIZE(p_hwfn) \
134 	ALIGNED_TYPE_SIZE(union type0_task_context, p_hwfn)
135 
136 /* Alignment is inherent to the type1_task_context structure */
137 #define TYPE1_TASK_CXT_SIZE(p_hwfn) sizeof(union type1_task_context)
138 
139 /* PF per protocl configuration object */
140 #define TASK_SEGMENTS   (NUM_TASK_PF_SEGMENTS + NUM_TASK_VF_SEGMENTS)
141 #define TASK_SEGMENT_VF (NUM_TASK_PF_SEGMENTS)
142 
143 struct ecore_tid_seg {
144 	u32	count;
145 	u8	type;
146 	bool	has_fl_mem;
147 };
148 
149 struct ecore_conn_type_cfg {
150 	u32			cid_count;
151 	u32			cids_per_vf;
152 	struct ecore_tid_seg	tid_seg[TASK_SEGMENTS];
153 };
154 
155 /* ILT Client configuration,
156  * Per connection type (protocol) resources (cids, tis, vf cids etc.)
157  * 1 - for connection context (CDUC) and for each task context we need two
158  * values, for regular task context and for force load memory
159  */
160 #define ILT_CLI_PF_BLOCKS	(1 + NUM_TASK_PF_SEGMENTS * 2)
161 #define ILT_CLI_VF_BLOCKS	(1 + NUM_TASK_VF_SEGMENTS * 2)
162 #define CDUC_BLK		(0)
163 #define SRQ_BLK			(0)
164 #define CDUT_SEG_BLK(n)		(1 + (u8)(n))
165 #define CDUT_FL_SEG_BLK(n, X)	(1 + (n) + NUM_TASK_##X##_SEGMENTS)
166 
167 enum ilt_clients {
168 	ILT_CLI_CDUC,
169 	ILT_CLI_CDUT,
170 	ILT_CLI_QM,
171 	ILT_CLI_TM,
172 	ILT_CLI_SRC,
173 	ILT_CLI_TSDM,
174 	ILT_CLI_MAX
175 };
176 
177 struct ilt_cfg_pair {
178 	u32 reg;
179 	u32 val;
180 };
181 
182 struct ecore_ilt_cli_blk {
183 	u32 total_size; /* 0 means not active */
184 	u32 real_size_in_page;
185 	u32 start_line;
186 	u32 dynamic_line_cnt;
187 };
188 
189 struct ecore_ilt_client_cfg {
190 	bool				active;
191 
192 	/* ILT boundaries */
193 	struct ilt_cfg_pair		first;
194 	struct ilt_cfg_pair		last;
195 	struct ilt_cfg_pair		p_size;
196 
197 	/* ILT client blocks for PF */
198 	struct ecore_ilt_cli_blk	pf_blks[ILT_CLI_PF_BLOCKS];
199 	u32				pf_total_lines;
200 
201 	/* ILT client blocks for VFs */
202 	struct ecore_ilt_cli_blk	vf_blks[ILT_CLI_VF_BLOCKS];
203 	u32				vf_total_lines;
204 };
205 
206 /* Per Path -
207  *      ILT shadow table
208  *      Protocol acquired CID lists
209  *      PF start line in ILT
210  */
211 struct ecore_dma_mem {
212 	dma_addr_t	p_phys;
213 	void		*p_virt;
214 	osal_size_t	size;
215 };
216 
217 #define MAP_WORD_SIZE		sizeof(unsigned long)
218 #define BITS_PER_MAP_WORD	(MAP_WORD_SIZE * 8)
219 
220 struct ecore_cid_acquired_map {
221 	u32		start_cid;
222 	u32		max_count;
223 	unsigned long	*cid_map;
224 };
225 
226 struct ecore_cxt_mngr {
227 	/* Per protocl configuration */
228 	struct ecore_conn_type_cfg	conn_cfg[MAX_CONN_TYPES];
229 
230 	/* computed ILT structure */
231 	struct ecore_ilt_client_cfg	clients[ILT_CLI_MAX];
232 
233 	/* Task type sizes */
234 	u32				task_type_size[NUM_TASK_TYPES];
235 
236 	/* total number of VFs for this hwfn -
237 	 * ALL VFs are symmetric in terms of HW resources
238 	 */
239 	u32				vf_count;
240 
241 	/* Acquired CIDs */
242 	struct ecore_cid_acquired_map acquired[MAX_CONN_TYPES];
243 	/* TBD - do we want this allocated to reserve space? */
244 	struct ecore_cid_acquired_map acquired_vf[MAX_CONN_TYPES][COMMON_MAX_NUM_VFS];
245 
246 	/* ILT shadow table */
247 	struct ecore_dma_mem		*ilt_shadow;
248 	u32				pf_start_line;
249 
250 	/* Mutex for a dynamic ILT allocation */
251 	osal_mutex_t			mutex;
252 
253 	/* SRC T2 */
254 	struct ecore_dma_mem		*t2;
255 	u32				t2_num_pages;
256 	u64				first_free;
257 	u64				last_free;
258 
259 	/* The infrastructure originally was very generic and context/task
260 	 * oriented - per connection-type we would set how many of those
261 	 * are needed, and later when determining how much memory we're
262 	 * needing for a given block we'd iterate over all the relevant
263 	 * connection-types.
264 	 * But since then we've had some additional resources, some of which
265 	 * require memory which is indepent of the general context/task
266 	 * scheme. We add those here explicitly per-feature.
267 	 */
268 
269 	/* total number of SRQ's for this hwfn */
270 	u32				srq_count;
271 
272 	/* Maximal number of L2 steering filters */
273 	u32				arfs_count;
274 
275 	/* TODO - VF arfs filters ? */
276 };
277 
278 /* check if resources/configuration is required according to protocol type */
src_proto(struct ecore_hwfn * p_hwfn,enum protocol_type type)279 static bool src_proto(struct ecore_hwfn *p_hwfn,
280 		      enum protocol_type type)
281 {
282 	return	type == PROTOCOLID_ISCSI	||
283 		type == PROTOCOLID_FCOE		||
284 		type == PROTOCOLID_TOE		||
285 		type == PROTOCOLID_IWARP;
286 }
287 
tm_cid_proto(enum protocol_type type)288 static bool tm_cid_proto(enum protocol_type type)
289 {
290 	return type == PROTOCOLID_ISCSI ||
291 	       type == PROTOCOLID_FCOE  ||
292 	       type == PROTOCOLID_ROCE  ||
293 	       type == PROTOCOLID_IWARP;
294 }
295 
tm_tid_proto(enum protocol_type type)296 static bool tm_tid_proto(enum protocol_type type)
297 {
298 	return type == PROTOCOLID_FCOE;
299 }
300 
301 /* counts the iids for the CDU/CDUC ILT client configuration */
302 struct ecore_cdu_iids {
303 	u32 pf_cids;
304 	u32 per_vf_cids;
305 };
306 
ecore_cxt_cdu_iids(struct ecore_cxt_mngr * p_mngr,struct ecore_cdu_iids * iids)307 static void ecore_cxt_cdu_iids(struct ecore_cxt_mngr   *p_mngr,
308 			       struct ecore_cdu_iids	*iids)
309 {
310 	u32 type;
311 
312 	for (type = 0; type < MAX_CONN_TYPES; type++) {
313 		iids->pf_cids += p_mngr->conn_cfg[type].cid_count;
314 		iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
315 	}
316 }
317 
318 /* counts the iids for the Searcher block configuration */
319 struct ecore_src_iids {
320 	u32			pf_cids;
321 	u32			per_vf_cids;
322 };
323 
ecore_cxt_src_iids(struct ecore_hwfn * p_hwfn,struct ecore_cxt_mngr * p_mngr,struct ecore_src_iids * iids)324 static void ecore_cxt_src_iids(struct ecore_hwfn *p_hwfn,
325 			       struct ecore_cxt_mngr *p_mngr,
326 			       struct ecore_src_iids *iids)
327 {
328 	u32 i;
329 
330 	for (i = 0; i < MAX_CONN_TYPES; i++) {
331 		if (!src_proto(p_hwfn, i))
332 			continue;
333 
334 		iids->pf_cids += p_mngr->conn_cfg[i].cid_count;
335 		iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf;
336 	}
337 
338 	/* Add L2 filtering filters in addition */
339 	iids->pf_cids += p_mngr->arfs_count;
340 }
341 
342 /* counts the iids for the Timers block configuration */
343 struct ecore_tm_iids {
344 	u32 pf_cids;
345 	u32 pf_tids[NUM_TASK_PF_SEGMENTS]; /* per segment */
346 	u32 pf_tids_total;
347 	u32 per_vf_cids;
348 	u32 per_vf_tids;
349 };
350 
ecore_cxt_tm_iids(struct ecore_hwfn * p_hwfn,struct ecore_cxt_mngr * p_mngr,struct ecore_tm_iids * iids)351 static void ecore_cxt_tm_iids(struct ecore_hwfn *p_hwfn,
352 			      struct ecore_cxt_mngr *p_mngr,
353 			      struct ecore_tm_iids *iids)
354 {
355 	bool tm_vf_required = false;
356 	bool tm_required = false;
357 	int i, j;
358 
359 	/* Timers is a special case -> we don't count how many cids require
360 	 * timers but what's the max cid that will be used by the timer block.
361 	 * therefore we traverse in reverse order, and once we hit a protocol
362 	 * that requires the timers memory, we'll sum all the protocols up
363 	 * to that one.
364 	 */
365 	for (i = MAX_CONN_TYPES - 1; i >= 0; i--) {
366 		struct ecore_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
367 
368 		if (tm_cid_proto(i) || tm_required) {
369 			if (p_cfg->cid_count)
370 				tm_required = true;
371 
372 			iids->pf_cids += p_cfg->cid_count;
373 		}
374 
375 		if (tm_cid_proto(i) || tm_vf_required) {
376 			if (p_cfg->cids_per_vf)
377 				tm_vf_required = true;
378 
379 			iids->per_vf_cids += p_cfg->cids_per_vf;
380 		}
381 
382 		if (tm_tid_proto(i)) {
383 			struct ecore_tid_seg *segs = p_cfg->tid_seg;
384 
385 			/* for each segment there is at most one
386 			 * protocol for which count is not 0.
387 			 */
388 			for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
389 				iids->pf_tids[j] += segs[j].count;
390 
391 			/* The last array elelment is for the VFs. As for PF
392 			 * segments there can be only one protocol for
393 			 * which this value is not 0.
394 			 */
395 			iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
396 		}
397 	}
398 
399 	iids->pf_cids = ROUNDUP(iids->pf_cids, TM_ALIGN);
400 	iids->per_vf_cids = ROUNDUP(iids->per_vf_cids, TM_ALIGN);
401 	iids->per_vf_tids = ROUNDUP(iids->per_vf_tids, TM_ALIGN);
402 
403 	for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
404 		iids->pf_tids[j] = ROUNDUP(iids->pf_tids[j], TM_ALIGN);
405 		iids->pf_tids_total += iids->pf_tids[j];
406 	}
407 }
408 
ecore_cxt_qm_iids(struct ecore_hwfn * p_hwfn,struct ecore_qm_iids * iids)409 static void ecore_cxt_qm_iids(struct ecore_hwfn *p_hwfn,
410 			      struct ecore_qm_iids *iids)
411 {
412 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
413 	struct ecore_tid_seg *segs;
414 	u32 vf_cids = 0, type, j;
415 	u32 vf_tids = 0;
416 
417 	for (type = 0; type < MAX_CONN_TYPES; type++) {
418 		iids->cids += p_mngr->conn_cfg[type].cid_count;
419 		vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
420 
421 		segs = p_mngr->conn_cfg[type].tid_seg;
422 		/* for each segment there is at most one
423 		 * protocol for which count is not 0.
424 		 */
425 		for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
426 			iids->tids += segs[j].count;
427 
428 		/* The last array elelment is for the VFs. As for PF
429 		 * segments there can be only one protocol for
430 		 * which this value is not 0.
431 		 */
432 		vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
433 	}
434 
435 	iids->vf_cids += vf_cids * p_mngr->vf_count;
436 	iids->tids += vf_tids * p_mngr->vf_count;
437 
438 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
439 		   "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
440 		   iids->cids, iids->vf_cids, iids->tids, vf_tids);
441 }
442 
ecore_cxt_tid_seg_info(struct ecore_hwfn * p_hwfn,u32 seg)443 static struct ecore_tid_seg *ecore_cxt_tid_seg_info(struct ecore_hwfn   *p_hwfn,
444 						    u32			seg)
445 {
446 	struct ecore_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
447 	u32 i;
448 
449 	/* Find the protocol with tid count > 0 for this segment.
450 	   Note: there can only be one and this is already validated.
451 	 */
452 	for (i = 0; i < MAX_CONN_TYPES; i++) {
453 		if (p_cfg->conn_cfg[i].tid_seg[seg].count)
454 			return &p_cfg->conn_cfg[i].tid_seg[seg];
455 	}
456 	return OSAL_NULL;
457 }
458 
459 /* set the iids (cid/tid) count per protocol */
ecore_cxt_set_proto_cid_count(struct ecore_hwfn * p_hwfn,enum protocol_type type,u32 cid_count,u32 vf_cid_cnt)460 static void ecore_cxt_set_proto_cid_count(struct ecore_hwfn *p_hwfn,
461 					  enum protocol_type type,
462 					  u32 cid_count, u32 vf_cid_cnt)
463 {
464 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
465 	struct ecore_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
466 
467 	p_conn->cid_count = ROUNDUP(cid_count, DQ_RANGE_ALIGN);
468 	p_conn->cids_per_vf = ROUNDUP(vf_cid_cnt, DQ_RANGE_ALIGN);
469 
470 	if (type == PROTOCOLID_ROCE) {
471 		u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
472 		u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
473 		u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
474 		u32 align = elems_per_page * DQ_RANGE_ALIGN;
475 
476 		p_conn->cid_count = ROUNDUP(p_conn->cid_count, align);
477 	}
478 }
479 
ecore_cxt_get_proto_cid_count(struct ecore_hwfn * p_hwfn,enum protocol_type type,u32 * vf_cid)480 u32 ecore_cxt_get_proto_cid_count(struct ecore_hwfn	*p_hwfn,
481 				  enum protocol_type	type,
482 				  u32			*vf_cid)
483 {
484 	if (vf_cid)
485 		*vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
486 
487 	return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
488 }
489 
ecore_cxt_get_proto_cid_start(struct ecore_hwfn * p_hwfn,enum protocol_type type)490 u32 ecore_cxt_get_proto_cid_start(struct ecore_hwfn	*p_hwfn,
491 				  enum protocol_type	type)
492 {
493 	return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
494 }
495 
ecore_cxt_get_proto_tid_count(struct ecore_hwfn * p_hwfn,enum protocol_type type)496 u32 ecore_cxt_get_proto_tid_count(struct ecore_hwfn *p_hwfn,
497 				  enum protocol_type type)
498 {
499 	u32 cnt = 0;
500 	int i;
501 
502 	for (i = 0; i < TASK_SEGMENTS; i++)
503 		cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
504 
505 	return cnt;
506 }
507 
ecore_cxt_set_proto_tid_count(struct ecore_hwfn * p_hwfn,enum protocol_type proto,u8 seg,u8 seg_type,u32 count,bool has_fl)508 static void ecore_cxt_set_proto_tid_count(struct ecore_hwfn *p_hwfn,
509 					  enum protocol_type proto,
510 					  u8 seg,
511 					  u8 seg_type,
512 					  u32 count,
513 					  bool has_fl)
514 {
515 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
516 	struct ecore_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
517 
518 	p_seg->count = count;
519 	p_seg->has_fl_mem = has_fl;
520 	p_seg->type = seg_type;
521 }
522 
523 /* the *p_line parameter must be either 0 for the first invocation or the
524    value returned in the previous invocation.
525  */
ecore_ilt_cli_blk_fill(struct ecore_ilt_client_cfg * p_cli,struct ecore_ilt_cli_blk * p_blk,u32 start_line,u32 total_size,u32 elem_size)526 static void ecore_ilt_cli_blk_fill(struct ecore_ilt_client_cfg	*p_cli,
527 				   struct ecore_ilt_cli_blk	*p_blk,
528 				   u32				start_line,
529 				   u32				total_size,
530 				   u32				elem_size)
531 {
532 	u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
533 
534 	/* verify that it's called once for each block */
535 	if (p_blk->total_size)
536 		return;
537 
538 	p_blk->total_size = total_size;
539 	p_blk->real_size_in_page = 0;
540 	if (elem_size)
541 		p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
542 	p_blk->start_line = start_line;
543 }
544 
ecore_ilt_cli_adv_line(struct ecore_hwfn * p_hwfn,struct ecore_ilt_client_cfg * p_cli,struct ecore_ilt_cli_blk * p_blk,u32 * p_line,enum ilt_clients client_id)545 static void ecore_ilt_cli_adv_line(struct ecore_hwfn		*p_hwfn,
546 				    struct ecore_ilt_client_cfg	*p_cli,
547 				    struct ecore_ilt_cli_blk	*p_blk,
548 				    u32				*p_line,
549 				    enum ilt_clients		client_id)
550 {
551 	if (!p_blk->total_size)
552 		return;
553 
554 	if (!p_cli->active)
555 		p_cli->first.val = *p_line;
556 
557 	p_cli->active = true;
558 	*p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
559 	p_cli->last.val = *p_line-1;
560 
561 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
562 		   "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
563 		   client_id, p_cli->first.val, p_cli->last.val,
564 		   p_blk->total_size, p_blk->real_size_in_page,
565 		   p_blk->start_line);
566 }
567 
ecore_ilt_get_dynamic_line_cnt(struct ecore_hwfn * p_hwfn,enum ilt_clients ilt_client)568 static u32 ecore_ilt_get_dynamic_line_cnt(struct ecore_hwfn *p_hwfn,
569 					  enum ilt_clients ilt_client)
570 {
571 	u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
572 	struct ecore_ilt_client_cfg *p_cli;
573 	u32 lines_to_skip = 0;
574 	u32 cxts_per_p;
575 
576 	/* TBD MK: ILT code should be simplified once PROTO enum is changed */
577 
578 	if (ilt_client == ILT_CLI_CDUC) {
579 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
580 
581 		cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
582 			     (u32)CONN_CXT_SIZE(p_hwfn);
583 
584 		lines_to_skip = cid_count / cxts_per_p;
585 	}
586 
587 	return lines_to_skip;
588 }
589 
590 static struct ecore_ilt_client_cfg *
ecore_cxt_set_cli(struct ecore_ilt_client_cfg * p_cli)591 ecore_cxt_set_cli(struct ecore_ilt_client_cfg *p_cli)
592 {
593 	p_cli->active = false;
594 	p_cli->first.val = 0;
595 	p_cli->last.val = 0;
596 	return p_cli;
597 }
598 
599 static struct ecore_ilt_cli_blk *
ecore_cxt_set_blk(struct ecore_ilt_cli_blk * p_blk)600 ecore_cxt_set_blk(struct ecore_ilt_cli_blk *p_blk)
601 {
602 	p_blk->total_size = 0;
603 	return p_blk;
604 }
605 
ecore_cxt_cfg_ilt_compute(struct ecore_hwfn * p_hwfn,u32 * line_count)606 enum _ecore_status_t ecore_cxt_cfg_ilt_compute(struct ecore_hwfn *p_hwfn,
607 					       u32 *line_count)
608 {
609 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
610 	u32 curr_line, total, i, task_size, line;
611 	struct ecore_ilt_client_cfg *p_cli;
612 	struct ecore_ilt_cli_blk *p_blk;
613 	struct ecore_cdu_iids cdu_iids;
614 	struct ecore_src_iids src_iids;
615 	struct ecore_qm_iids qm_iids;
616 	struct ecore_tm_iids tm_iids;
617 	struct ecore_tid_seg *p_seg;
618 
619 	OSAL_MEM_ZERO(&qm_iids, sizeof(qm_iids));
620 	OSAL_MEM_ZERO(&cdu_iids, sizeof(cdu_iids));
621 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
622 	OSAL_MEM_ZERO(&tm_iids, sizeof(tm_iids));
623 
624 	p_mngr->pf_start_line = RESC_START(p_hwfn, ECORE_ILT);
625 
626 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
627 		   "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
628 		   p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
629 
630 	/* CDUC */
631 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUC]);
632 
633 	curr_line = p_mngr->pf_start_line;
634 
635 	/* CDUC PF */
636 	p_cli->pf_total_lines = 0;
637 
638 	/* get the counters for the CDUC,CDUC and QM clients  */
639 	ecore_cxt_cdu_iids(p_mngr, &cdu_iids);
640 
641 	p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[CDUC_BLK]);
642 
643 	total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
644 
645 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
646 			       total, CONN_CXT_SIZE(p_hwfn));
647 
648 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
649 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
650 
651 	p_blk->dynamic_line_cnt = ecore_ilt_get_dynamic_line_cnt(p_hwfn,
652 								 ILT_CLI_CDUC);
653 
654 	/* CDUC VF */
655 	p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[CDUC_BLK]);
656 	total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
657 
658 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
659 			       total, CONN_CXT_SIZE(p_hwfn));
660 
661 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
662 	p_cli->vf_total_lines = curr_line - p_blk->start_line;
663 
664 	for (i = 1; i < p_mngr->vf_count; i++)
665 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
666 				       ILT_CLI_CDUC);
667 
668 	/* CDUT PF */
669 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUT]);
670 	p_cli->first.val = curr_line;
671 
672 	/* first the 'working' task memory */
673 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
674 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
675 		if (!p_seg || p_seg->count == 0)
676 			continue;
677 
678 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[CDUT_SEG_BLK(i)]);
679 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
680 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
681 				       p_mngr->task_type_size[p_seg->type]);
682 
683 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
684 				       ILT_CLI_CDUT);
685 	}
686 
687 	/* next the 'init' task memory (forced load memory) */
688 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
689 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
690 		if (!p_seg || p_seg->count == 0)
691 			continue;
692 
693 		p_blk = ecore_cxt_set_blk(
694 				&p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)]);
695 
696 		if (!p_seg->has_fl_mem) {
697 			/* The segment is active (total size pf 'working'
698 			 * memory is > 0) but has no FL (forced-load, Init)
699 			 * memory. Thus:
700 			 *
701 			 * 1.   The total-size in the corrsponding FL block of
702 			 *      the ILT client is set to 0 - No ILT line are
703 			 *      provisioned and no ILT memory allocated.
704 			 *
705 			 * 2.   The start-line of said block is set to the
706 			 *      start line of the matching working memory
707 			 *      block in the ILT client. This is later used to
708 			 *      configure the CDU segment offset registers and
709 			 *      results in an FL command for TIDs of this
710 			 *      segement behaves as regular load commands
711 			 *      (loading TIDs from the working memory).
712 			 */
713 			line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
714 
715 			ecore_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
716 			continue;
717 		}
718 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
719 
720 		ecore_ilt_cli_blk_fill(p_cli, p_blk,
721 				       curr_line, total,
722 				       p_mngr->task_type_size[p_seg->type]);
723 
724 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
725 				       ILT_CLI_CDUT);
726 	}
727 	p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
728 
729 	/* CDUT VF */
730 	p_seg = ecore_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
731 	if (p_seg && p_seg->count) {
732 		/* Stricly speaking we need to iterate over all VF
733 		 * task segment types, but a VF has only 1 segment
734 		 */
735 
736 		/* 'working' memory */
737 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
738 
739 		p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[CDUT_SEG_BLK(0)]);
740 		ecore_ilt_cli_blk_fill(p_cli, p_blk,
741 				       curr_line, total,
742 				       p_mngr->task_type_size[p_seg->type]);
743 
744 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
745 				       ILT_CLI_CDUT);
746 
747 		/* 'init' memory */
748 		p_blk = ecore_cxt_set_blk(
749 				&p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)]);
750 		if (!p_seg->has_fl_mem) {
751 			/* see comment above */
752 			line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
753 			ecore_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
754 		} else {
755 			task_size = p_mngr->task_type_size[p_seg->type];
756 			ecore_ilt_cli_blk_fill(p_cli, p_blk,
757 					       curr_line, total,
758 					       task_size);
759 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
760 					       ILT_CLI_CDUT);
761 		}
762 		p_cli->vf_total_lines = curr_line -
763 					p_cli->vf_blks[0].start_line;
764 
765 		/* Now for the rest of the VFs */
766 		for (i = 1; i < p_mngr->vf_count; i++) {
767 			/* don't set p_blk i.e. don't clear total_size */
768 			p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
769 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
770 					       ILT_CLI_CDUT);
771 
772 			/* don't set p_blk i.e. don't clear total_size */
773 			p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
774 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
775 					       ILT_CLI_CDUT);
776 		}
777 	}
778 
779 	/* QM */
780 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_QM]);
781 	p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
782 
783 	ecore_cxt_qm_iids(p_hwfn, &qm_iids);
784 	total = ecore_qm_pf_mem_size(p_hwfn->rel_pf_id, qm_iids.cids,
785 				     qm_iids.vf_cids, qm_iids.tids,
786 				     p_hwfn->qm_info.num_pqs,
787 				     p_hwfn->qm_info.num_vf_pqs);
788 
789 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
790 		   "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
791 		   qm_iids.cids, qm_iids.vf_cids, qm_iids.tids,
792 		   p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
793 
794 	ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total * 0x1000,
795 			       QM_PQ_ELEMENT_SIZE);
796 
797 	ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
798 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
799 
800 	/* SRC */
801 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_SRC]);
802 	ecore_cxt_src_iids(p_hwfn, p_mngr, &src_iids);
803 
804 	/* Both the PF and VFs searcher connections are stored in the per PF
805 	 * database. Thus sum the PF searcher cids and all the VFs searcher
806 	 * cids.
807 	 */
808 	total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
809 	if (total) {
810 		u32 local_max = OSAL_MAX_T(u32, total,
811 					   SRC_MIN_NUM_ELEMS);
812 
813 		total = OSAL_ROUNDUP_POW_OF_TWO(local_max);
814 
815 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
816 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
817 				       total * sizeof(struct src_ent),
818 				       sizeof(struct src_ent));
819 
820 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
821 				       ILT_CLI_SRC);
822 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
823 	}
824 
825 	/* TM PF */
826 	p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_TM]);
827 	ecore_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
828 	total = tm_iids.pf_cids + tm_iids.pf_tids_total;
829 	if (total) {
830 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[0]);
831 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
832 				       total * TM_ELEM_SIZE,
833 				       TM_ELEM_SIZE);
834 
835 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
836 				       ILT_CLI_TM);
837 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
838 	}
839 
840 	/* TM VF */
841 	total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
842 	if (total) {
843 		p_blk = ecore_cxt_set_blk(&p_cli->vf_blks[0]);
844 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
845 				       total * TM_ELEM_SIZE,
846 				       TM_ELEM_SIZE);
847 
848 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
849 				       ILT_CLI_TM);
850 
851 		p_cli->vf_total_lines = curr_line - p_blk->start_line;
852 		for (i = 1; i < p_mngr->vf_count; i++) {
853 			ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
854 					       ILT_CLI_TM);
855 		}
856 	}
857 
858 	/* TSDM (SRQ CONTEXT) */
859 	total = ecore_cxt_get_srq_count(p_hwfn);
860 
861 	if (total) {
862 		p_cli = ecore_cxt_set_cli(&p_mngr->clients[ILT_CLI_TSDM]);
863 		p_blk = ecore_cxt_set_blk(&p_cli->pf_blks[SRQ_BLK]);
864 		ecore_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
865 				       total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
866 
867 		ecore_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
868 				       ILT_CLI_TSDM);
869 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
870 	}
871 
872 	*line_count = curr_line - p_hwfn->p_cxt_mngr->pf_start_line;
873 
874 	if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
875 	    RESC_NUM(p_hwfn, ECORE_ILT)) {
876 		return ECORE_INVAL;
877 	}
878 
879 	return ECORE_SUCCESS;
880 }
881 
ecore_cxt_cfg_ilt_compute_excess(struct ecore_hwfn * p_hwfn,u32 used_lines)882 u32 ecore_cxt_cfg_ilt_compute_excess(struct ecore_hwfn *p_hwfn, u32 used_lines)
883 {
884 	struct ecore_ilt_client_cfg *p_cli;
885 	u32 excess_lines, available_lines;
886 	struct ecore_cxt_mngr *p_mngr;
887 	u32 ilt_page_size, elem_size;
888 	struct ecore_tid_seg *p_seg;
889 	int i;
890 
891 	available_lines = RESC_NUM(p_hwfn, ECORE_ILT);
892 	excess_lines = used_lines - available_lines;
893 
894 	if (!excess_lines)
895 		return 0;
896 
897 	if (!ECORE_IS_RDMA_PERSONALITY(p_hwfn))
898 		return 0;
899 
900 	p_mngr = p_hwfn->p_cxt_mngr;
901 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
902 	ilt_page_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
903 
904 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
905 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
906 		if (!p_seg || p_seg->count == 0)
907 			continue;
908 
909 		elem_size = p_mngr->task_type_size[p_seg->type];
910 		if (!elem_size)
911 			continue;
912 
913 		return (ilt_page_size / elem_size) * excess_lines;
914 	}
915 
916 	DP_ERR(p_hwfn, "failed computing excess ILT lines\n");
917 	return 0;
918 }
919 
ecore_cxt_src_t2_free(struct ecore_hwfn * p_hwfn)920 static void ecore_cxt_src_t2_free(struct ecore_hwfn *p_hwfn)
921 {
922 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
923 	u32 i;
924 
925 	if (!p_mngr->t2)
926 		return;
927 
928 	for (i = 0; i < p_mngr->t2_num_pages; i++)
929 		if (p_mngr->t2[i].p_virt)
930 			OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
931 					       p_mngr->t2[i].p_virt,
932 					       p_mngr->t2[i].p_phys,
933 					       p_mngr->t2[i].size);
934 
935 	OSAL_FREE(p_hwfn->p_dev, p_mngr->t2);
936 	p_mngr->t2 = OSAL_NULL;
937 }
938 
ecore_cxt_src_t2_alloc(struct ecore_hwfn * p_hwfn)939 static enum _ecore_status_t ecore_cxt_src_t2_alloc(struct ecore_hwfn *p_hwfn)
940 {
941 	struct ecore_cxt_mngr *p_mngr  = p_hwfn->p_cxt_mngr;
942 	u32 conn_num, total_size, ent_per_page, psz, i;
943 	struct ecore_ilt_client_cfg *p_src;
944 	struct ecore_src_iids src_iids;
945 	struct ecore_dma_mem *p_t2;
946 	enum _ecore_status_t rc;
947 
948 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
949 
950 	/* if the SRC ILT client is inactive - there are no connection
951 	 * requiring the searcer, leave.
952 	 */
953 	p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
954 	if (!p_src->active)
955 		return ECORE_SUCCESS;
956 
957 	ecore_cxt_src_iids(p_hwfn, p_mngr, &src_iids);
958 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
959 	total_size = conn_num * sizeof(struct src_ent);
960 
961 	/* use the same page size as the SRC ILT client */
962 	psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
963 	p_mngr->t2_num_pages = DIV_ROUND_UP(total_size, psz);
964 
965 	/* allocate t2 */
966 	p_mngr->t2 = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL,
967 				 p_mngr->t2_num_pages *
968 				 sizeof(struct ecore_dma_mem));
969 	if (!p_mngr->t2) {
970 		DP_NOTICE(p_hwfn, true, "Failed to allocate t2 table\n");
971 		rc = ECORE_NOMEM;
972 		goto t2_fail;
973 	}
974 
975 	/* allocate t2 pages */
976 	for (i = 0; i < p_mngr->t2_num_pages; i++) {
977 		u32 size = OSAL_MIN_T(u32, total_size, psz);
978 		void **p_virt = &p_mngr->t2[i].p_virt;
979 
980 		*p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
981 						  &p_mngr->t2[i].p_phys,
982 						  size);
983 		if (!p_mngr->t2[i].p_virt) {
984 			rc = ECORE_NOMEM;
985 			goto t2_fail;
986 		}
987 		OSAL_MEM_ZERO(*p_virt, size);
988 		p_mngr->t2[i].size = size;
989 		total_size -= size;
990 	}
991 
992 	/* Set the t2 pointers */
993 
994 	/* entries per page - must be a power of two */
995 	ent_per_page = psz / sizeof(struct src_ent);
996 
997 	p_mngr->first_free = (u64)p_mngr->t2[0].p_phys;
998 
999 	p_t2 = &p_mngr->t2[(conn_num - 1) / ent_per_page];
1000 	p_mngr->last_free = (u64)p_t2->p_phys +
1001 				 ((conn_num - 1) & (ent_per_page - 1)) *
1002 				 sizeof(struct src_ent);
1003 
1004 	for (i = 0; i < p_mngr->t2_num_pages; i++) {
1005 		u32 ent_num = OSAL_MIN_T(u32, ent_per_page, conn_num);
1006 		struct src_ent *entries = p_mngr->t2[i].p_virt;
1007 		u64 p_ent_phys = (u64)p_mngr->t2[i].p_phys, val;
1008 		u32 j;
1009 
1010 		for (j = 0; j < ent_num - 1; j++) {
1011 			val = p_ent_phys +
1012 			      (j + 1) * sizeof(struct src_ent);
1013 			entries[j].next = OSAL_CPU_TO_BE64(val);
1014 		}
1015 
1016 		if (i < p_mngr->t2_num_pages - 1)
1017 			val = (u64)p_mngr->t2[i + 1].p_phys;
1018 		else
1019 			val = 0;
1020 		entries[j].next = OSAL_CPU_TO_BE64(val);
1021 
1022 		conn_num -= ent_num;
1023 	}
1024 
1025 	return ECORE_SUCCESS;
1026 
1027 t2_fail:
1028 	ecore_cxt_src_t2_free(p_hwfn);
1029 	return rc;
1030 }
1031 
1032 #define for_each_ilt_valid_client(pos, clients)	\
1033 	for (pos = 0; pos < ILT_CLI_MAX; pos++)	\
1034 		if (!clients[pos].active) {	\
1035 			continue;		\
1036 		} else				\
1037 
1038 
1039 /* Total number of ILT lines used by this PF */
ecore_cxt_ilt_shadow_size(struct ecore_ilt_client_cfg * ilt_clients)1040 static u32 ecore_cxt_ilt_shadow_size(struct ecore_ilt_client_cfg *ilt_clients)
1041 {
1042 	u32 size = 0;
1043 	u32 i;
1044 
1045 	for_each_ilt_valid_client(i, ilt_clients)
1046 		size += (ilt_clients[i].last.val -
1047 			 ilt_clients[i].first.val + 1);
1048 
1049 	return size;
1050 }
1051 
ecore_ilt_shadow_free(struct ecore_hwfn * p_hwfn)1052 static void ecore_ilt_shadow_free(struct ecore_hwfn *p_hwfn)
1053 {
1054 	struct ecore_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
1055 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1056 	u32 ilt_size, i;
1057 
1058 	ilt_size = ecore_cxt_ilt_shadow_size(p_cli);
1059 
1060 	for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
1061 		struct ecore_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
1062 
1063 		if (p_dma->p_virt)
1064 			OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
1065 					       p_dma->p_virt,
1066 					       p_dma->p_phys,
1067 					       p_dma->size);
1068 		p_dma->p_virt = OSAL_NULL;
1069 	}
1070 	OSAL_FREE(p_hwfn->p_dev, p_mngr->ilt_shadow);
1071 }
1072 
ecore_ilt_blk_alloc(struct ecore_hwfn * p_hwfn,struct ecore_ilt_cli_blk * p_blk,enum ilt_clients ilt_client,u32 start_line_offset)1073 static enum _ecore_status_t ecore_ilt_blk_alloc(struct ecore_hwfn *p_hwfn,
1074 						struct ecore_ilt_cli_blk *p_blk,
1075 						enum ilt_clients ilt_client,
1076 						u32 start_line_offset)
1077 {
1078 	struct ecore_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
1079 	u32 lines, line, sz_left, lines_to_skip = 0;
1080 
1081 	/* Special handling for RoCE that supports dynamic allocation */
1082 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn) &&
1083 	    ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
1084 		return ECORE_SUCCESS;
1085 
1086 	lines_to_skip = p_blk->dynamic_line_cnt;
1087 
1088 	if (!p_blk->total_size)
1089 		return ECORE_SUCCESS;
1090 
1091 	sz_left = p_blk->total_size;
1092 	lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) -
1093 		lines_to_skip;
1094 	line = p_blk->start_line + start_line_offset -
1095 	       p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
1096 
1097 	for (; lines; lines--) {
1098 		dma_addr_t p_phys;
1099 		void *p_virt;
1100 		u32 size;
1101 
1102 		size = OSAL_MIN_T(u32, sz_left, p_blk->real_size_in_page);
1103 		p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
1104 						 &p_phys, size);
1105 		if (!p_virt)
1106 			return ECORE_NOMEM;
1107 		OSAL_MEM_ZERO(p_virt, size);
1108 
1109 		ilt_shadow[line].p_phys = p_phys;
1110 		ilt_shadow[line].p_virt = p_virt;
1111 		ilt_shadow[line].size = size;
1112 
1113 		DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1114 			   "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
1115 			   line, (u64)p_phys, p_virt, size);
1116 
1117 		sz_left -= size;
1118 		line++;
1119 	}
1120 
1121 	return ECORE_SUCCESS;
1122 }
1123 
ecore_ilt_shadow_alloc(struct ecore_hwfn * p_hwfn)1124 static enum _ecore_status_t ecore_ilt_shadow_alloc(struct ecore_hwfn *p_hwfn)
1125 {
1126 	struct ecore_cxt_mngr *p_mngr  = p_hwfn->p_cxt_mngr;
1127 	struct ecore_ilt_client_cfg *clients = p_mngr->clients;
1128 	struct ecore_ilt_cli_blk *p_blk;
1129 	u32 size, i, j, k;
1130 	enum _ecore_status_t rc;
1131 
1132 	size = ecore_cxt_ilt_shadow_size(clients);
1133 	p_mngr->ilt_shadow = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL,
1134 					 size * sizeof(struct ecore_dma_mem));
1135 
1136 	if (!p_mngr->ilt_shadow) {
1137 		DP_NOTICE(p_hwfn, true, "Failed to allocate ilt shadow table\n");
1138 		rc = ECORE_NOMEM;
1139 		goto ilt_shadow_fail;
1140 	}
1141 
1142 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1143 		   "Allocated 0x%x bytes for ilt shadow\n",
1144 		   (u32)(size * sizeof(struct ecore_dma_mem)));
1145 
1146 	for_each_ilt_valid_client(i, clients) {
1147 		for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1148 			p_blk = &clients[i].pf_blks[j];
1149 			rc = ecore_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1150 			if (rc != ECORE_SUCCESS)
1151 				goto ilt_shadow_fail;
1152 		}
1153 		for (k = 0; k < p_mngr->vf_count; k++) {
1154 			for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1155 				u32 lines = clients[i].vf_total_lines * k;
1156 
1157 				p_blk = &clients[i].vf_blks[j];
1158 				rc = ecore_ilt_blk_alloc(p_hwfn, p_blk,
1159 							 i, lines);
1160 				if (rc != ECORE_SUCCESS)
1161 					goto ilt_shadow_fail;
1162 			}
1163 		}
1164 	}
1165 
1166 	return ECORE_SUCCESS;
1167 
1168 ilt_shadow_fail:
1169 	ecore_ilt_shadow_free(p_hwfn);
1170 	return rc;
1171 }
1172 
ecore_cid_map_free(struct ecore_hwfn * p_hwfn)1173 static void ecore_cid_map_free(struct ecore_hwfn *p_hwfn)
1174 {
1175 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1176 	u32 type, vf;
1177 
1178 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1179 		OSAL_FREE(p_hwfn->p_dev, p_mngr->acquired[type].cid_map);
1180 		p_mngr->acquired[type].max_count = 0;
1181 		p_mngr->acquired[type].start_cid = 0;
1182 
1183 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1184 			OSAL_FREE(p_hwfn->p_dev,
1185 				  p_mngr->acquired_vf[type][vf].cid_map);
1186 			p_mngr->acquired_vf[type][vf].max_count = 0;
1187 			p_mngr->acquired_vf[type][vf].start_cid = 0;
1188 		}
1189 	}
1190 }
1191 
1192 static enum _ecore_status_t
ecore_cid_map_alloc_single(struct ecore_hwfn * p_hwfn,u32 type,u32 cid_start,u32 cid_count,struct ecore_cid_acquired_map * p_map)1193 ecore_cid_map_alloc_single(struct ecore_hwfn *p_hwfn, u32 type,
1194 			   u32 cid_start, u32 cid_count,
1195 			   struct ecore_cid_acquired_map *p_map)
1196 {
1197 	u32 size;
1198 
1199 	if (!cid_count)
1200 		return ECORE_SUCCESS;
1201 
1202 	size = MAP_WORD_SIZE * DIV_ROUND_UP(cid_count, BITS_PER_MAP_WORD);
1203 	p_map->cid_map = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, size);
1204 	if (p_map->cid_map == OSAL_NULL)
1205 		return ECORE_NOMEM;
1206 
1207 	p_map->max_count = cid_count;
1208 	p_map->start_cid = cid_start;
1209 
1210 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
1211 		   "Type %08x start: %08x count %08x\n",
1212 		   type, p_map->start_cid, p_map->max_count);
1213 
1214 	return ECORE_SUCCESS;
1215 }
1216 
ecore_cid_map_alloc(struct ecore_hwfn * p_hwfn)1217 static enum _ecore_status_t ecore_cid_map_alloc(struct ecore_hwfn *p_hwfn)
1218 {
1219 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1220 	u32 start_cid = 0, vf_start_cid = 0;
1221 	u32 type, vf;
1222 
1223 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1224 		struct ecore_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[type];
1225 		struct ecore_cid_acquired_map *p_map;
1226 
1227 		/* Handle PF maps */
1228 		p_map = &p_mngr->acquired[type];
1229 		if (ecore_cid_map_alloc_single(p_hwfn, type, start_cid,
1230 					       p_cfg->cid_count, p_map))
1231 			goto cid_map_fail;
1232 
1233 		/* Handle VF maps */
1234 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1235 			p_map = &p_mngr->acquired_vf[type][vf];
1236 			if (ecore_cid_map_alloc_single(p_hwfn, type,
1237 						       vf_start_cid,
1238 						       p_cfg->cids_per_vf,
1239 						       p_map))
1240 				goto cid_map_fail;
1241 		}
1242 
1243 		start_cid += p_cfg->cid_count;
1244 		vf_start_cid += p_cfg->cids_per_vf;
1245 	}
1246 
1247 	return ECORE_SUCCESS;
1248 
1249 cid_map_fail:
1250 	ecore_cid_map_free(p_hwfn);
1251 	return ECORE_NOMEM;
1252 }
1253 
ecore_cxt_mngr_alloc(struct ecore_hwfn * p_hwfn)1254 enum _ecore_status_t ecore_cxt_mngr_alloc(struct ecore_hwfn *p_hwfn)
1255 {
1256 	struct ecore_ilt_client_cfg *clients;
1257 	struct ecore_cxt_mngr *p_mngr;
1258 	u32 i;
1259 
1260 	p_mngr = OSAL_ZALLOC(p_hwfn->p_dev, GFP_KERNEL, sizeof(*p_mngr));
1261 	if (!p_mngr) {
1262 		DP_NOTICE(p_hwfn, true, "Failed to allocate `struct ecore_cxt_mngr'\n");
1263 		return ECORE_NOMEM;
1264 	}
1265 
1266 	/* Initialize ILT client registers */
1267 	clients = p_mngr->clients;
1268 	clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1269 	clients[ILT_CLI_CDUC].last.reg  = ILT_CFG_REG(CDUC, LAST_ILT);
1270 	clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1271 
1272 	clients[ILT_CLI_QM].first.reg   = ILT_CFG_REG(QM, FIRST_ILT);
1273 	clients[ILT_CLI_QM].last.reg    = ILT_CFG_REG(QM, LAST_ILT);
1274 	clients[ILT_CLI_QM].p_size.reg  = ILT_CFG_REG(QM, P_SIZE);
1275 
1276 	clients[ILT_CLI_TM].first.reg   = ILT_CFG_REG(TM, FIRST_ILT);
1277 	clients[ILT_CLI_TM].last.reg    = ILT_CFG_REG(TM, LAST_ILT);
1278 	clients[ILT_CLI_TM].p_size.reg  = ILT_CFG_REG(TM, P_SIZE);
1279 
1280 	clients[ILT_CLI_SRC].first.reg  = ILT_CFG_REG(SRC, FIRST_ILT);
1281 	clients[ILT_CLI_SRC].last.reg   = ILT_CFG_REG(SRC, LAST_ILT);
1282 	clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1283 
1284 	clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1285 	clients[ILT_CLI_CDUT].last.reg  = ILT_CFG_REG(CDUT, LAST_ILT);
1286 	clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1287 
1288 	clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1289 	clients[ILT_CLI_TSDM].last.reg  = ILT_CFG_REG(TSDM, LAST_ILT);
1290 	clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1291 
1292 	/* default ILT page size for all clients is 32K */
1293 	for (i = 0; i < ILT_CLI_MAX; i++)
1294 		p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE;
1295 
1296 	/* Initialize task sizes */
1297 	p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1298 	p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1299 
1300 	if (p_hwfn->p_dev->p_iov_info)
1301 		p_mngr->vf_count = p_hwfn->p_dev->p_iov_info->total_vfs;
1302 
1303 	/* Initialize the dynamic ILT allocation mutex */
1304 	OSAL_MUTEX_ALLOC(p_hwfn, &p_mngr->mutex);
1305 	OSAL_MUTEX_INIT(&p_mngr->mutex);
1306 
1307 	/* Set the cxt mangr pointer priori to further allocations */
1308 	p_hwfn->p_cxt_mngr = p_mngr;
1309 
1310 	return ECORE_SUCCESS;
1311 }
1312 
ecore_cxt_tables_alloc(struct ecore_hwfn * p_hwfn)1313 enum _ecore_status_t ecore_cxt_tables_alloc(struct ecore_hwfn *p_hwfn)
1314 {
1315 	enum _ecore_status_t    rc;
1316 
1317 	/* Allocate the ILT shadow table */
1318 	rc = ecore_ilt_shadow_alloc(p_hwfn);
1319 	if (rc) {
1320 		DP_NOTICE(p_hwfn, true, "Failed to allocate ilt memory\n");
1321 		goto tables_alloc_fail;
1322 	}
1323 
1324 	/* Allocate the T2  table */
1325 	rc = ecore_cxt_src_t2_alloc(p_hwfn);
1326 	if (rc) {
1327 		DP_NOTICE(p_hwfn, true, "Failed to allocate T2 memory\n");
1328 		goto tables_alloc_fail;
1329 	}
1330 
1331 	/* Allocate and initalize the acquired cids bitmaps */
1332 	rc = ecore_cid_map_alloc(p_hwfn);
1333 	if (rc) {
1334 		DP_NOTICE(p_hwfn, true, "Failed to allocate cid maps\n");
1335 		goto tables_alloc_fail;
1336 	}
1337 
1338 	return ECORE_SUCCESS;
1339 
1340 tables_alloc_fail:
1341 	ecore_cxt_mngr_free(p_hwfn);
1342 	return rc;
1343 }
ecore_cxt_mngr_free(struct ecore_hwfn * p_hwfn)1344 void ecore_cxt_mngr_free(struct ecore_hwfn *p_hwfn)
1345 {
1346 	if (!p_hwfn->p_cxt_mngr)
1347 		return;
1348 
1349 	ecore_cid_map_free(p_hwfn);
1350 	ecore_cxt_src_t2_free(p_hwfn);
1351 	ecore_ilt_shadow_free(p_hwfn);
1352 	OSAL_MUTEX_DEALLOC(&p_hwfn->p_cxt_mngr->mutex);
1353 	OSAL_FREE(p_hwfn->p_dev, p_hwfn->p_cxt_mngr);
1354 
1355 	p_hwfn->p_cxt_mngr = OSAL_NULL;
1356 }
1357 
ecore_cxt_mngr_setup(struct ecore_hwfn * p_hwfn)1358 void ecore_cxt_mngr_setup(struct ecore_hwfn *p_hwfn)
1359 {
1360 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1361 	struct ecore_cid_acquired_map *p_map;
1362 	struct ecore_conn_type_cfg *p_cfg;
1363 	int type;
1364 	u32 len;
1365 
1366 	/* Reset acquired cids */
1367 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1368 		u32 vf;
1369 
1370 		p_cfg = &p_mngr->conn_cfg[type];
1371 		if (p_cfg->cid_count) {
1372 			p_map = &p_mngr->acquired[type];
1373 			len = DIV_ROUND_UP(p_map->max_count,
1374 					   BITS_PER_MAP_WORD) *
1375 			      MAP_WORD_SIZE;
1376 			OSAL_MEM_ZERO(p_map->cid_map, len);
1377 		}
1378 
1379 		if (!p_cfg->cids_per_vf)
1380 			continue;
1381 
1382 		for (vf = 0; vf < COMMON_MAX_NUM_VFS; vf++) {
1383 			p_map = &p_mngr->acquired_vf[type][vf];
1384 			len = DIV_ROUND_UP(p_map->max_count,
1385 					   BITS_PER_MAP_WORD) *
1386 			      MAP_WORD_SIZE;
1387 			OSAL_MEM_ZERO(p_map->cid_map, len);
1388 		}
1389 	}
1390 }
1391 
1392 /* HW initialization helper (per Block, per phase) */
1393 
1394 /* CDU Common */
1395 #define CDUC_CXT_SIZE_SHIFT						\
1396 	CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1397 
1398 #define CDUC_CXT_SIZE_MASK						\
1399 	(CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1400 
1401 #define CDUC_BLOCK_WASTE_SHIFT						\
1402 	CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1403 
1404 #define CDUC_BLOCK_WASTE_MASK						\
1405 	(CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1406 
1407 #define CDUC_NCIB_SHIFT							\
1408 	CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1409 
1410 #define CDUC_NCIB_MASK							\
1411 	(CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1412 
1413 #define CDUT_TYPE0_CXT_SIZE_SHIFT					\
1414 	CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1415 
1416 #define CDUT_TYPE0_CXT_SIZE_MASK					\
1417 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >>				\
1418 	CDUT_TYPE0_CXT_SIZE_SHIFT)
1419 
1420 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT					\
1421 	CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1422 
1423 #define CDUT_TYPE0_BLOCK_WASTE_MASK					\
1424 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >>			\
1425 	CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1426 
1427 #define CDUT_TYPE0_NCIB_SHIFT						\
1428 	CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1429 
1430 #define CDUT_TYPE0_NCIB_MASK						\
1431 	(CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >>		\
1432 	CDUT_TYPE0_NCIB_SHIFT)
1433 
1434 #define CDUT_TYPE1_CXT_SIZE_SHIFT					\
1435 	CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1436 
1437 #define CDUT_TYPE1_CXT_SIZE_MASK					\
1438 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >>				\
1439 	CDUT_TYPE1_CXT_SIZE_SHIFT)
1440 
1441 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT					\
1442 	CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1443 
1444 #define CDUT_TYPE1_BLOCK_WASTE_MASK					\
1445 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >>			\
1446 	CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1447 
1448 #define CDUT_TYPE1_NCIB_SHIFT						\
1449 	CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1450 
1451 #define CDUT_TYPE1_NCIB_MASK						\
1452 	(CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >>		\
1453 	CDUT_TYPE1_NCIB_SHIFT)
1454 
ecore_cdu_init_common(struct ecore_hwfn * p_hwfn)1455 static void ecore_cdu_init_common(struct ecore_hwfn *p_hwfn)
1456 {
1457 	u32 page_sz, elems_per_page, block_waste,  cxt_size, cdu_params = 0;
1458 
1459 	/* CDUC - connection configuration */
1460 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1461 	cxt_size = CONN_CXT_SIZE(p_hwfn);
1462 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1463 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1464 
1465 	SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1466 	SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1467 	SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
1468 	STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1469 
1470 	/* CDUT - type-0 tasks configuration */
1471 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1472 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1473 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1474 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1475 
1476 	/* cxt size and block-waste are multipes of 8 */
1477 	cdu_params = 0;
1478 	SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1479 	SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1480 	SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1481 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1482 
1483 	/* CDUT - type-1 tasks configuration */
1484 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1485 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1486 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1487 
1488 	/* cxt size and block-waste are multipes of 8 */
1489 	cdu_params = 0;
1490 	SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1491 	SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1492 	SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1493 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1494 }
1495 
1496 /* CDU PF */
1497 #define CDU_SEG_REG_TYPE_SHIFT		CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1498 #define CDU_SEG_REG_TYPE_MASK		0x1
1499 #define CDU_SEG_REG_OFFSET_SHIFT	0
1500 #define CDU_SEG_REG_OFFSET_MASK		CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1501 
ecore_cdu_init_pf(struct ecore_hwfn * p_hwfn)1502 static void ecore_cdu_init_pf(struct ecore_hwfn *p_hwfn)
1503 {
1504 	struct ecore_ilt_client_cfg *p_cli;
1505 	struct ecore_tid_seg *p_seg;
1506 	u32 cdu_seg_params, offset;
1507 	int i;
1508 
1509 	static const u32 rt_type_offset_arr[] = {
1510 		CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1511 		CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1512 		CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1513 		CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1514 	};
1515 
1516 	static const u32 rt_type_offset_fl_arr[] = {
1517 		CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1518 		CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1519 		CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1520 		CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1521 	};
1522 
1523 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1524 
1525 	/* There are initializations only for CDUT during pf Phase */
1526 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1527 		/* Segment 0*/
1528 		p_seg = ecore_cxt_tid_seg_info(p_hwfn, i);
1529 		if (!p_seg)
1530 			continue;
1531 
1532 		/* Note: start_line is already adjusted for the CDU
1533 		 * segment register granularity, so we just need to
1534 		 * divide. Adjustment is implicit as we assume ILT
1535 		 * Page size is larger than 32K!
1536 		 */
1537 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1538 			 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1539 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1540 
1541 		cdu_seg_params = 0;
1542 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1543 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1544 		STORE_RT_REG(p_hwfn, rt_type_offset_arr[i],
1545 			     cdu_seg_params);
1546 
1547 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1548 			 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1549 			  p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1550 
1551 		cdu_seg_params = 0;
1552 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1553 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1554 		STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i],
1555 			     cdu_seg_params);
1556 
1557 	}
1558 }
1559 
ecore_qm_init_pf(struct ecore_hwfn * p_hwfn)1560 void ecore_qm_init_pf(struct ecore_hwfn *p_hwfn)
1561 {
1562 	struct ecore_qm_info *qm_info = &p_hwfn->qm_info;
1563 	struct ecore_qm_iids iids;
1564 
1565 	OSAL_MEM_ZERO(&iids, sizeof(iids));
1566 	ecore_cxt_qm_iids(p_hwfn, &iids);
1567 
1568 	ecore_qm_pf_rt_init(p_hwfn, p_hwfn->p_main_ptt, p_hwfn->port_id,
1569 			    p_hwfn->rel_pf_id, qm_info->max_phys_tcs_per_port,
1570 			    p_hwfn->first_on_engine,
1571 			    iids.cids, iids.vf_cids, iids.tids,
1572 			    qm_info->start_pq,
1573 			    qm_info->num_pqs - qm_info->num_vf_pqs,
1574 			    qm_info->num_vf_pqs,
1575 			    qm_info->start_vport,
1576 			    qm_info->num_vports, qm_info->pf_wfq, qm_info->pf_rl,
1577 			    p_hwfn->qm_info.qm_pq_params,
1578 			    p_hwfn->qm_info.qm_vport_params);
1579 }
1580 
1581 /* CM PF */
ecore_cm_init_pf(struct ecore_hwfn * p_hwfn)1582 void ecore_cm_init_pf(struct ecore_hwfn *p_hwfn)
1583 {
1584 	STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET, ecore_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1585 }
1586 
1587 /* DQ PF */
ecore_dq_init_pf(struct ecore_hwfn * p_hwfn)1588 static void ecore_dq_init_pf(struct ecore_hwfn *p_hwfn)
1589 {
1590 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1591 	u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1592 
1593 	dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1594 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1595 
1596 	dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1597 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1598 
1599 	dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1600 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1601 
1602 	dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1603 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1604 
1605 	dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1606 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1607 
1608 	dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1609 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1610 
1611 	dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1612 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1613 
1614 	dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1615 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1616 
1617 	dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1618 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1619 
1620 	dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1621 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1622 
1623 	dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1624 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1625 
1626 	dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1627 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1628 
1629 	/* Connection types 6 & 7 are not in use, yet they must be configured
1630 	 * as the highest possible connection. Not configuring them means the
1631 	 * defaults will be  used, and with a large number of cids a bug may
1632 	 * occur, if the defaults will be smaller than dq_pf_max_cid /
1633 	 * dq_vf_max_cid.
1634 	 */
1635 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1636 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1637 
1638 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1639 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1640 }
1641 
ecore_ilt_bounds_init(struct ecore_hwfn * p_hwfn)1642 static void ecore_ilt_bounds_init(struct ecore_hwfn *p_hwfn)
1643 {
1644 	struct ecore_ilt_client_cfg *ilt_clients;
1645 	int i;
1646 
1647 	ilt_clients = p_hwfn->p_cxt_mngr->clients;
1648 	for_each_ilt_valid_client(i, ilt_clients) {
1649 		STORE_RT_REG(p_hwfn,
1650 			     ilt_clients[i].first.reg,
1651 			     ilt_clients[i].first.val);
1652 		STORE_RT_REG(p_hwfn,
1653 			     ilt_clients[i].last.reg,
1654 			     ilt_clients[i].last.val);
1655 		STORE_RT_REG(p_hwfn,
1656 			     ilt_clients[i].p_size.reg,
1657 			     ilt_clients[i].p_size.val);
1658 	}
1659 }
1660 
ecore_ilt_vf_bounds_init(struct ecore_hwfn * p_hwfn)1661 static void ecore_ilt_vf_bounds_init(struct ecore_hwfn *p_hwfn)
1662 {
1663 	struct ecore_ilt_client_cfg *p_cli;
1664 	u32 blk_factor;
1665 
1666 	/* For simplicty  we set the 'block' to be an ILT page */
1667 	if (p_hwfn->p_dev->p_iov_info) {
1668 		struct ecore_hw_sriov_info *p_iov = p_hwfn->p_dev->p_iov_info;
1669 
1670 		STORE_RT_REG(p_hwfn,
1671 			     PSWRQ2_REG_VF_BASE_RT_OFFSET,
1672 			     p_iov->first_vf_in_pf);
1673 		STORE_RT_REG(p_hwfn,
1674 			     PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1675 			     p_iov->first_vf_in_pf + p_iov->total_vfs);
1676 	}
1677 
1678 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1679 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1680 	if (p_cli->active) {
1681 		STORE_RT_REG(p_hwfn,
1682 			     PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1683 			     blk_factor);
1684 		STORE_RT_REG(p_hwfn,
1685 			     PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1686 			     p_cli->pf_total_lines);
1687 		STORE_RT_REG(p_hwfn,
1688 			     PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1689 			     p_cli->vf_total_lines);
1690 	}
1691 
1692 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1693 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1694 	if (p_cli->active) {
1695 		STORE_RT_REG(p_hwfn,
1696 			     PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1697 			     blk_factor);
1698 		STORE_RT_REG(p_hwfn,
1699 			     PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1700 			     p_cli->pf_total_lines);
1701 		STORE_RT_REG(p_hwfn,
1702 			     PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1703 			     p_cli->vf_total_lines);
1704 	}
1705 
1706 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1707 	blk_factor = OSAL_LOG2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1708 	if (p_cli->active) {
1709 		STORE_RT_REG(p_hwfn,
1710 			     PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET,
1711 			     blk_factor);
1712 		STORE_RT_REG(p_hwfn,
1713 			     PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1714 			     p_cli->pf_total_lines);
1715 		STORE_RT_REG(p_hwfn,
1716 			     PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1717 			     p_cli->vf_total_lines);
1718 	}
1719 }
1720 
1721 /* ILT (PSWRQ2) PF */
ecore_ilt_init_pf(struct ecore_hwfn * p_hwfn)1722 static void ecore_ilt_init_pf(struct ecore_hwfn *p_hwfn)
1723 {
1724 	struct ecore_ilt_client_cfg *clients;
1725 	struct ecore_cxt_mngr *p_mngr;
1726 	struct ecore_dma_mem *p_shdw;
1727 	u32 line, rt_offst, i;
1728 
1729 	ecore_ilt_bounds_init(p_hwfn);
1730 	ecore_ilt_vf_bounds_init(p_hwfn);
1731 
1732 	p_mngr  = p_hwfn->p_cxt_mngr;
1733 	p_shdw  = p_mngr->ilt_shadow;
1734 	clients = p_hwfn->p_cxt_mngr->clients;
1735 
1736 	for_each_ilt_valid_client(i, clients) {
1737 		/* Client's 1st val and RT array are absolute, ILT shadows'
1738 		 * lines are relative.
1739 		 */
1740 		line = clients[i].first.val - p_mngr->pf_start_line;
1741 		rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1742 			   clients[i].first.val * ILT_ENTRY_IN_REGS;
1743 
1744 		for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1745 		     line++, rt_offst += ILT_ENTRY_IN_REGS) {
1746 			u64 ilt_hw_entry = 0;
1747 
1748 			/** p_virt could be OSAL_NULL incase of dynamic
1749 			 *  allocation
1750 			 */
1751 			if (p_shdw[line].p_virt != OSAL_NULL) {
1752 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1753 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1754 					  (p_shdw[line].p_phys >> 12));
1755 
1756 				DP_VERBOSE(
1757 					p_hwfn, ECORE_MSG_ILT,
1758 					"Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1759 					rt_offst, line, i,
1760 					(u64)(p_shdw[line].p_phys >> 12));
1761 			}
1762 
1763 			STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1764 		}
1765 	}
1766 }
1767 
1768 /* SRC (Searcher) PF */
ecore_src_init_pf(struct ecore_hwfn * p_hwfn)1769 static void ecore_src_init_pf(struct ecore_hwfn *p_hwfn)
1770 {
1771 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1772 	u32 rounded_conn_num, conn_num, conn_max;
1773 	struct ecore_src_iids src_iids;
1774 
1775 	OSAL_MEM_ZERO(&src_iids, sizeof(src_iids));
1776 	ecore_cxt_src_iids(p_hwfn, p_mngr, &src_iids);
1777 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1778 	if (!conn_num)
1779 		return;
1780 
1781 	conn_max = OSAL_MAX_T(u32, conn_num, SRC_MIN_NUM_ELEMS);
1782 	rounded_conn_num = OSAL_ROUNDUP_POW_OF_TWO(conn_max);
1783 
1784 	STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1785 	STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1786 		     OSAL_LOG2(rounded_conn_num));
1787 
1788 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1789 			 p_hwfn->p_cxt_mngr->first_free);
1790 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1791 			 p_hwfn->p_cxt_mngr->last_free);
1792 	DP_VERBOSE(p_hwfn, ECORE_MSG_ILT,
1793 		   "Configured SEARCHER for 0x%08x connections\n",
1794 		   conn_num);
1795 }
1796 
1797 /* Timers PF */
1798 #define TM_CFG_NUM_IDS_SHIFT		0
1799 #define TM_CFG_NUM_IDS_MASK		0xFFFFULL
1800 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT	16
1801 #define TM_CFG_PRE_SCAN_OFFSET_MASK	0x1FFULL
1802 #define TM_CFG_PARENT_PF_SHIFT		25
1803 #define TM_CFG_PARENT_PF_MASK		0x7ULL
1804 
1805 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT	30
1806 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK	0x1FFULL
1807 
1808 #define TM_CFG_TID_OFFSET_SHIFT		30
1809 #define TM_CFG_TID_OFFSET_MASK		0x7FFFFULL
1810 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT	49
1811 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK	0x1FFULL
1812 
ecore_tm_init_pf(struct ecore_hwfn * p_hwfn)1813 static void ecore_tm_init_pf(struct ecore_hwfn *p_hwfn)
1814 {
1815 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1816 	u32 active_seg_mask = 0, tm_offset, rt_reg;
1817 	struct ecore_tm_iids tm_iids;
1818 	u64 cfg_word;
1819 	u8 i;
1820 
1821 	OSAL_MEM_ZERO(&tm_iids, sizeof(tm_iids));
1822 	ecore_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
1823 
1824 	/* @@@TBD No pre-scan for now */
1825 
1826 	/* Note: We assume consecutive VFs for a PF */
1827 	for (i = 0; i < p_mngr->vf_count; i++) {
1828 		cfg_word = 0;
1829 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1830 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1831 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1832 		SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1833 
1834 		rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1835 			 (sizeof(cfg_word) / sizeof(u32)) *
1836 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1837 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1838 	}
1839 
1840 	cfg_word = 0;
1841 	SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1842 	SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1843 	SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);	  /* n/a for PF */
1844 	SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all   */
1845 
1846 	rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1847 		 (sizeof(cfg_word) / sizeof(u32)) *
1848 		 (NUM_OF_VFS(p_hwfn->p_dev) + p_hwfn->rel_pf_id);
1849 	STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1850 
1851 	/* enale scan */
1852 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1853 		     tm_iids.pf_cids  ? 0x1 : 0x0);
1854 
1855 	/* @@@TBD how to enable the scan for the VFs */
1856 
1857 	tm_offset = tm_iids.per_vf_cids;
1858 
1859 	/* Note: We assume consecutive VFs for a PF */
1860 	for (i = 0; i < p_mngr->vf_count; i++) {
1861 		cfg_word = 0;
1862 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1863 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1864 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1865 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1866 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1867 
1868 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1869 			 (sizeof(cfg_word) / sizeof(u32)) *
1870 			 (p_hwfn->p_dev->p_iov_info->first_vf_in_pf + i);
1871 
1872 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1873 	}
1874 
1875 	tm_offset = tm_iids.pf_cids;
1876 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1877 		cfg_word = 0;
1878 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1879 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1880 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1881 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1882 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64)0);
1883 
1884 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1885 			 (sizeof(cfg_word) / sizeof(u32)) *
1886 			 (NUM_OF_VFS(p_hwfn->p_dev) +
1887 			 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1888 
1889 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1890 		active_seg_mask |= (tm_iids.pf_tids[i] ? (1 << i) : 0);
1891 
1892 		tm_offset += tm_iids.pf_tids[i];
1893 	}
1894 
1895 	if (ECORE_IS_RDMA_PERSONALITY(p_hwfn))
1896 		active_seg_mask = 0;
1897 
1898 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1899 
1900 	/* @@@TBD how to enable the scan for the VFs */
1901 }
1902 
ecore_prs_init_common(struct ecore_hwfn * p_hwfn)1903 static void ecore_prs_init_common(struct ecore_hwfn *p_hwfn)
1904 {
1905 	if ((p_hwfn->hw_info.personality == ECORE_PCI_FCOE) &&
1906 	    p_hwfn->pf_params.fcoe_pf_params.is_target)
1907 		STORE_RT_REG(p_hwfn, PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1908 }
1909 
ecore_prs_init_pf(struct ecore_hwfn * p_hwfn)1910 static void ecore_prs_init_pf(struct ecore_hwfn *p_hwfn)
1911 {
1912 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1913 	struct ecore_conn_type_cfg *p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1914 	struct ecore_tid_seg *p_tid;
1915 
1916 	/* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1917 	if (!p_fcoe->cid_count)
1918 		return;
1919 
1920 	p_tid = &p_fcoe->tid_seg[ECORE_CXT_FCOE_TID_SEG];
1921 	if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1922 		STORE_RT_REG_AGG(p_hwfn,
1923 				 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1924 				 p_tid->count);
1925 	} else {
1926 		STORE_RT_REG_AGG(p_hwfn,
1927 				PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1928 				p_tid->count);
1929 	}
1930 }
1931 
ecore_cxt_hw_init_common(struct ecore_hwfn * p_hwfn)1932 void ecore_cxt_hw_init_common(struct ecore_hwfn *p_hwfn)
1933 {
1934 	/* CDU configuration */
1935 	ecore_cdu_init_common(p_hwfn);
1936 	ecore_prs_init_common(p_hwfn);
1937 }
1938 
ecore_cxt_hw_init_pf(struct ecore_hwfn * p_hwfn)1939 void ecore_cxt_hw_init_pf(struct ecore_hwfn *p_hwfn)
1940 {
1941 	ecore_qm_init_pf(p_hwfn);
1942 	ecore_cm_init_pf(p_hwfn);
1943 	ecore_dq_init_pf(p_hwfn);
1944 	ecore_cdu_init_pf(p_hwfn);
1945 	ecore_ilt_init_pf(p_hwfn);
1946 	ecore_src_init_pf(p_hwfn);
1947 	ecore_tm_init_pf(p_hwfn);
1948 	ecore_prs_init_pf(p_hwfn);
1949 }
1950 
_ecore_cxt_acquire_cid(struct ecore_hwfn * p_hwfn,enum protocol_type type,u32 * p_cid,u8 vfid)1951 enum _ecore_status_t _ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
1952 					    enum protocol_type type,
1953 					    u32 *p_cid, u8 vfid)
1954 {
1955 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1956 	struct ecore_cid_acquired_map *p_map;
1957 	u32 rel_cid;
1958 
1959 	if (type >= MAX_CONN_TYPES) {
1960 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
1961 		return ECORE_INVAL;
1962 	}
1963 
1964 	if (vfid >= COMMON_MAX_NUM_VFS && vfid != ECORE_CXT_PF_CID) {
1965 		DP_NOTICE(p_hwfn, true, "VF [%02x] is out of range\n", vfid);
1966 		return ECORE_INVAL;
1967 	}
1968 
1969 	/* Determine the right map to take this CID from */
1970 	if (vfid == ECORE_CXT_PF_CID)
1971 		p_map = &p_mngr->acquired[type];
1972 	else
1973 		p_map = &p_mngr->acquired_vf[type][vfid];
1974 
1975 	if (p_map->cid_map == OSAL_NULL) {
1976 		DP_NOTICE(p_hwfn, true, "Invalid protocol type %d", type);
1977 		return ECORE_INVAL;
1978 	}
1979 
1980 	rel_cid = OSAL_FIND_FIRST_ZERO_BIT(p_map->cid_map,
1981 					   p_map->max_count);
1982 
1983 	if (rel_cid >= p_map->max_count) {
1984 		DP_NOTICE(p_hwfn, false, "no CID available for protocol %d\n",
1985 			  type);
1986 		return ECORE_NORESOURCES;
1987 	}
1988 
1989 	OSAL_SET_BIT(rel_cid, p_map->cid_map);
1990 
1991 	*p_cid = rel_cid + p_map->start_cid;
1992 
1993 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
1994 		   "Acquired cid 0x%08x [rel. %08x] vfid %02x type %d\n",
1995 		   *p_cid, rel_cid, vfid, type);
1996 
1997 	return ECORE_SUCCESS;
1998 }
1999 
ecore_cxt_acquire_cid(struct ecore_hwfn * p_hwfn,enum protocol_type type,u32 * p_cid)2000 enum _ecore_status_t ecore_cxt_acquire_cid(struct ecore_hwfn *p_hwfn,
2001 					   enum protocol_type type,
2002 					   u32 *p_cid)
2003 {
2004 	return _ecore_cxt_acquire_cid(p_hwfn, type, p_cid, ECORE_CXT_PF_CID);
2005 }
2006 
ecore_cxt_test_cid_acquired(struct ecore_hwfn * p_hwfn,u32 cid,u8 vfid,enum protocol_type * p_type,struct ecore_cid_acquired_map ** pp_map)2007 static bool ecore_cxt_test_cid_acquired(struct ecore_hwfn *p_hwfn,
2008 					u32 cid, u8 vfid,
2009 					enum protocol_type *p_type,
2010 					struct ecore_cid_acquired_map **pp_map)
2011 {
2012 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2013 	u32 rel_cid;
2014 
2015 	/* Iterate over protocols and find matching cid range */
2016 	for (*p_type = 0; *p_type < MAX_CONN_TYPES; (*p_type)++) {
2017 		if (vfid == ECORE_CXT_PF_CID)
2018 			*pp_map = &p_mngr->acquired[*p_type];
2019 		else
2020 			*pp_map = &p_mngr->acquired_vf[*p_type][vfid];
2021 
2022 		if (!((*pp_map)->cid_map))
2023 			continue;
2024 		if (cid >= (*pp_map)->start_cid &&
2025 		    cid < (*pp_map)->start_cid + (*pp_map)->max_count) {
2026 			break;
2027 		}
2028 	}
2029 
2030 	if (*p_type == MAX_CONN_TYPES) {
2031 		DP_NOTICE(p_hwfn, true, "Invalid CID %d vfid %02x", cid, vfid);
2032 		goto fail;
2033 	}
2034 
2035 	rel_cid = cid - (*pp_map)->start_cid;
2036 	if (!OSAL_TEST_BIT(rel_cid, (*pp_map)->cid_map)) {
2037 		DP_NOTICE(p_hwfn, true,
2038 			  "CID %d [vifd %02x] not acquired", cid, vfid);
2039 		goto fail;
2040 	}
2041 
2042 	return true;
2043 fail:
2044 	*p_type = MAX_CONN_TYPES;
2045 	*pp_map = OSAL_NULL;
2046 	return false;
2047 }
2048 
_ecore_cxt_release_cid(struct ecore_hwfn * p_hwfn,u32 cid,u8 vfid)2049 void _ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid, u8 vfid)
2050 {
2051 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2052 	enum protocol_type type;
2053 	bool b_acquired;
2054 	u32 rel_cid;
2055 
2056 	if (vfid != ECORE_CXT_PF_CID && vfid > COMMON_MAX_NUM_VFS) {
2057 		DP_NOTICE(p_hwfn, true,
2058 			  "Trying to return incorrect CID belonging to VF %02x\n",
2059 			  vfid);
2060 		return;
2061 	}
2062 
2063 	/* Test acquired and find matching per-protocol map */
2064 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, cid, vfid,
2065 						 &type, &p_map);
2066 
2067 	if (!b_acquired)
2068 		return;
2069 
2070 	rel_cid = cid - p_map->start_cid;
2071 	OSAL_CLEAR_BIT(rel_cid, p_map->cid_map);
2072 
2073 	DP_VERBOSE(p_hwfn, ECORE_MSG_CXT,
2074 		   "Released CID 0x%08x [rel. %08x] vfid %02x type %d\n",
2075 		   cid, rel_cid, vfid, type);
2076 }
2077 
ecore_cxt_release_cid(struct ecore_hwfn * p_hwfn,u32 cid)2078 void ecore_cxt_release_cid(struct ecore_hwfn *p_hwfn, u32 cid)
2079 {
2080 	_ecore_cxt_release_cid(p_hwfn, cid, ECORE_CXT_PF_CID);
2081 }
2082 
ecore_cxt_get_cid_info(struct ecore_hwfn * p_hwfn,struct ecore_cxt_info * p_info)2083 enum _ecore_status_t ecore_cxt_get_cid_info(struct ecore_hwfn *p_hwfn,
2084 					    struct ecore_cxt_info *p_info)
2085 {
2086 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2087 	struct ecore_cid_acquired_map *p_map = OSAL_NULL;
2088 	u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
2089 	enum protocol_type type;
2090 	bool b_acquired;
2091 
2092 	/* Test acquired and find matching per-protocol map */
2093 	b_acquired = ecore_cxt_test_cid_acquired(p_hwfn, p_info->iid,
2094 						 ECORE_CXT_PF_CID,
2095 						 &type, &p_map);
2096 
2097 	if (!b_acquired)
2098 		return ECORE_INVAL;
2099 
2100 	/* set the protocl type */
2101 	p_info->type = type;
2102 
2103 	/* compute context virtual pointer */
2104 	hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
2105 
2106 	conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
2107 	cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
2108 	line = p_info->iid / cxts_per_p;
2109 
2110 	/* Make sure context is allocated (dynamic allocation) */
2111 	if (!p_mngr->ilt_shadow[line].p_virt)
2112 		return ECORE_INVAL;
2113 
2114 	p_info->p_cxt = (u8 *)p_mngr->ilt_shadow[line].p_virt +
2115 			      p_info->iid % cxts_per_p * conn_cxt_size;
2116 
2117 	DP_VERBOSE(p_hwfn, (ECORE_MSG_ILT | ECORE_MSG_CXT),
2118 		   "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
2119 		   (p_info->iid / cxts_per_p), p_info->p_cxt, p_info->iid);
2120 
2121 	return ECORE_SUCCESS;
2122 }
2123 
ecore_cxt_set_srq_count(struct ecore_hwfn * p_hwfn,u32 num_srqs)2124 static void ecore_cxt_set_srq_count(struct ecore_hwfn *p_hwfn, u32 num_srqs)
2125 {
2126 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
2127 
2128 	p_mgr->srq_count = num_srqs;
2129 }
2130 
ecore_cxt_get_srq_count(struct ecore_hwfn * p_hwfn)2131 u32 ecore_cxt_get_srq_count(struct ecore_hwfn *p_hwfn)
2132 {
2133 	struct ecore_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
2134 
2135 	return p_mgr->srq_count;
2136 }
2137 
ecore_rdma_set_pf_params(struct ecore_hwfn * p_hwfn,struct ecore_rdma_pf_params * p_params,u32 num_tasks)2138 static void ecore_rdma_set_pf_params(struct ecore_hwfn *p_hwfn,
2139 				     struct ecore_rdma_pf_params *p_params,
2140 				     u32 num_tasks)
2141 {
2142 	u32 num_cons, num_qps, num_srqs;
2143 	enum protocol_type proto;
2144 
2145 	/* Override personality with rdma flavor */
2146 	num_srqs = OSAL_MIN_T(u32, ECORE_RDMA_MAX_SRQS, p_params->num_srqs);
2147 
2148 	/* The only case RDMA personality can be overriden is if NVRAM is
2149 	 * configured with ETH_RDMA or if no rdma protocol was requested
2150 	 */
2151 	switch (p_params->rdma_protocol) {
2152 	case ECORE_RDMA_PROTOCOL_DEFAULT:
2153 		if (p_hwfn->mcp_info->func_info.protocol ==
2154 		    ECORE_PCI_ETH_RDMA) {
2155 			DP_NOTICE(p_hwfn, false,
2156 				  "Current day drivers don't support RoCE & iWARP. Default to RoCE-only\n");
2157 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2158 		}
2159 		break;
2160 	case ECORE_RDMA_PROTOCOL_NONE:
2161 		p_hwfn->hw_info.personality = ECORE_PCI_ETH;
2162 		return; /* intentional... nothing left to do... */
2163 	case ECORE_RDMA_PROTOCOL_ROCE:
2164 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2165 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_ROCE;
2166 		break;
2167 	case ECORE_RDMA_PROTOCOL_IWARP:
2168 		if (p_hwfn->mcp_info->func_info.protocol == ECORE_PCI_ETH_RDMA)
2169 			p_hwfn->hw_info.personality = ECORE_PCI_ETH_IWARP;
2170 		break;
2171 	}
2172 
2173 	switch (p_hwfn->hw_info.personality) {
2174 	case ECORE_PCI_ETH_IWARP:
2175 		/* Each QP requires one connection */
2176 		num_cons = OSAL_MIN_T(u32, IWARP_MAX_QPS, p_params->num_qps);
2177 #ifdef CONFIG_ECORE_IWARP /* required for the define */
2178 		/* additional connections required for passive tcp handling */
2179 		num_cons += ECORE_IWARP_PREALLOC_CNT;
2180 #endif
2181 		proto = PROTOCOLID_IWARP;
2182 		p_params->roce_edpm_mode = false;
2183 		break;
2184 	case ECORE_PCI_ETH_ROCE:
2185 		num_qps = OSAL_MIN_T(u32, ROCE_MAX_QPS, p_params->num_qps);
2186 		num_cons = num_qps * 2; /* each QP requires two connections */
2187 		proto = PROTOCOLID_ROCE;
2188 		break;
2189 	default:
2190 		return;
2191 	}
2192 
2193 	if (num_cons && num_tasks) {
2194 		ecore_cxt_set_proto_cid_count(p_hwfn, proto,
2195 					      num_cons, 0);
2196 
2197 		/* Deliberatly passing ROCE for tasks id. This is because
2198 		 * iWARP / RoCE share the task id.
2199 		 */
2200 		ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
2201 					      ECORE_CXT_ROCE_TID_SEG,
2202 					      1, /* RoCE segment type */
2203 					      num_tasks,
2204 					      false); /* !force load */
2205 		ecore_cxt_set_srq_count(p_hwfn, num_srqs);
2206 
2207 	} else {
2208 		DP_INFO(p_hwfn->p_dev,
2209 			"RDMA personality used without setting params!\n");
2210 	}
2211 }
2212 
ecore_cxt_set_pf_params(struct ecore_hwfn * p_hwfn,u32 rdma_tasks)2213 enum _ecore_status_t ecore_cxt_set_pf_params(struct ecore_hwfn *p_hwfn,
2214 					     u32 rdma_tasks)
2215 {
2216 	/* Set the number of required CORE connections */
2217 	u32 core_cids = 1; /* SPQ */
2218 
2219 	if (p_hwfn->using_ll2)
2220 		core_cids += 4; /* @@@TBD Use the proper #define */
2221 
2222 	ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
2223 
2224 	switch (p_hwfn->hw_info.personality) {
2225 	case ECORE_PCI_ETH_RDMA:
2226 	case ECORE_PCI_ETH_IWARP:
2227 	case ECORE_PCI_ETH_ROCE:
2228 		ecore_rdma_set_pf_params(p_hwfn,
2229 					 &p_hwfn->pf_params.rdma_pf_params,
2230 					 rdma_tasks);
2231 
2232 		/* no need for break since RoCE coexist with Ethernet */
2233 		/* FALLTHROUGH */
2234 	case ECORE_PCI_ETH:
2235 	{
2236 		struct ecore_eth_pf_params *p_params =
2237 					&p_hwfn->pf_params.eth_pf_params;
2238 
2239 		if (!p_params->num_vf_cons)
2240 			p_params->num_vf_cons = ETH_PF_PARAMS_VF_CONS_DEFAULT;
2241 		ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
2242 					      p_params->num_cons,
2243 					      p_params->num_vf_cons);
2244 		p_hwfn->p_cxt_mngr->arfs_count = p_params->num_arfs_filters;
2245 
2246 		break;
2247 	}
2248 	case ECORE_PCI_FCOE:
2249 	{
2250 		struct ecore_fcoe_pf_params *p_params;
2251 
2252 		p_params = &p_hwfn->pf_params.fcoe_pf_params;
2253 
2254 		if (p_params->num_cons && p_params->num_tasks) {
2255 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_FCOE,
2256 						      p_params->num_cons, 0);
2257 
2258 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
2259 						      ECORE_CXT_FCOE_TID_SEG,
2260 						      0, /* segment type */
2261 						      p_params->num_tasks,
2262 						      true);
2263 		} else {
2264 			DP_INFO(p_hwfn->p_dev,
2265 				"Fcoe personality used without setting params!\n");
2266 		}
2267 		break;
2268 	}
2269 	case ECORE_PCI_ISCSI:
2270 	{
2271 		struct ecore_iscsi_pf_params *p_params;
2272 
2273 		p_params = &p_hwfn->pf_params.iscsi_pf_params;
2274 
2275 		if (p_params->num_cons && p_params->num_tasks) {
2276 			ecore_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ISCSI,
2277 						      p_params->num_cons, 0);
2278 
2279 			ecore_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ISCSI,
2280 						      ECORE_CXT_ISCSI_TID_SEG,
2281 						      0, /* segment type */
2282 						      p_params->num_tasks,
2283 						      true);
2284 		} else {
2285 			DP_INFO(p_hwfn->p_dev,
2286 				"Iscsi personality used without setting params!\n");
2287 		}
2288 		break;
2289 	}
2290 	default:
2291 		return ECORE_INVAL;
2292 	}
2293 
2294 	return ECORE_SUCCESS;
2295 }
2296 
ecore_cxt_get_tid_mem_info(struct ecore_hwfn * p_hwfn,struct ecore_tid_mem * p_info)2297 enum _ecore_status_t ecore_cxt_get_tid_mem_info(struct ecore_hwfn *p_hwfn,
2298 						struct ecore_tid_mem *p_info)
2299 {
2300 	struct ecore_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2301 	u32 proto, seg, total_lines, i, shadow_line;
2302 	struct ecore_ilt_client_cfg *p_cli;
2303 	struct ecore_ilt_cli_blk *p_fl_seg;
2304 	struct ecore_tid_seg *p_seg_info;
2305 
2306 	/* Verify the personality */
2307 	switch (p_hwfn->hw_info.personality) {
2308 	case ECORE_PCI_FCOE:
2309 		proto = PROTOCOLID_FCOE;
2310 		seg = ECORE_CXT_FCOE_TID_SEG;
2311 		break;
2312 	case ECORE_PCI_ISCSI:
2313 		proto = PROTOCOLID_ISCSI;
2314 		seg = ECORE_CXT_ISCSI_TID_SEG;
2315 		break;
2316 	default:
2317 		return ECORE_INVAL;
2318 	}
2319 
2320 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2321 	if (!p_cli->active) {
2322 		return ECORE_INVAL;
2323 	}
2324 
2325 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2326 	if (!p_seg_info->has_fl_mem)
2327 		return ECORE_INVAL;
2328 
2329 	p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2330 	total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2331 				   p_fl_seg->real_size_in_page);
2332 
2333 	for (i = 0; i < total_lines; i++) {
2334 		shadow_line = i + p_fl_seg->start_line -
2335 			      p_hwfn->p_cxt_mngr->pf_start_line;
2336 		p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
2337 	}
2338 	p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2339 			p_fl_seg->real_size_in_page;
2340 	p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2341 	p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2342 				     p_info->tid_size;
2343 
2344 	return ECORE_SUCCESS;
2345 }
2346 
2347 /* This function is very RoCE oriented, if another protocol in the future
2348  * will want this feature we'll need to modify the function to be more generic
2349  */
2350 enum _ecore_status_t
ecore_cxt_dynamic_ilt_alloc(struct ecore_hwfn * p_hwfn,enum ecore_cxt_elem_type elem_type,u32 iid)2351 ecore_cxt_dynamic_ilt_alloc(struct ecore_hwfn *p_hwfn,
2352 			    enum ecore_cxt_elem_type elem_type,
2353 			    u32 iid)
2354 {
2355 	u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2356 	struct ecore_ilt_client_cfg *p_cli;
2357 	struct ecore_ilt_cli_blk *p_blk;
2358 	struct ecore_ptt *p_ptt;
2359 	dma_addr_t p_phys;
2360 	u64 ilt_hw_entry;
2361 	void *p_virt;
2362 	enum _ecore_status_t rc = ECORE_SUCCESS;
2363 
2364 	switch (elem_type) {
2365 	case ECORE_ELEM_CXT:
2366 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2367 		elem_size = CONN_CXT_SIZE(p_hwfn);
2368 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2369 		break;
2370 	case ECORE_ELEM_SRQ:
2371 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2372 		elem_size = SRQ_CXT_SIZE;
2373 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2374 		break;
2375 	case ECORE_ELEM_TASK:
2376 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2377 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2378 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2379 		break;
2380 	default:
2381 		DP_NOTICE(p_hwfn, false,
2382 			  "ECORE_INVALID elem type = %d", elem_type);
2383 		return ECORE_INVAL;
2384 	}
2385 
2386 	/* Calculate line in ilt */
2387 	hw_p_size = p_cli->p_size.val;
2388 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2389 	line = p_blk->start_line + (iid / elems_per_p);
2390 	shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2391 
2392 	/* If line is already allocated, do nothing, otherwise allocate it and
2393 	 * write it to the PSWRQ2 registers.
2394 	 * This section can be run in parallel from different contexts and thus
2395 	 * a mutex protection is needed.
2396 	 */
2397 
2398 	OSAL_MUTEX_ACQUIRE(&p_hwfn->p_cxt_mngr->mutex);
2399 
2400 	if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2401 		goto out0;
2402 
2403 	p_ptt = ecore_ptt_acquire(p_hwfn);
2404 	if (!p_ptt) {
2405 		DP_NOTICE(p_hwfn, false,
2406 			  "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2407 		rc = ECORE_TIMEOUT;
2408 		goto out0;
2409 	}
2410 
2411 	p_virt = OSAL_DMA_ALLOC_COHERENT(p_hwfn->p_dev,
2412 					 &p_phys,
2413 					 p_blk->real_size_in_page);
2414 	if (!p_virt) {
2415 		rc = ECORE_NOMEM;
2416 		goto out1;
2417 	}
2418 	OSAL_MEM_ZERO(p_virt, p_blk->real_size_in_page);
2419 
2420 	/* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2421 	 * to compensate for a HW bug, but it is configured even if DIF is not
2422 	 * enabled. This is harmless and allows us to avoid a dedicated API. We
2423 	 * configure the field for all of the contexts on the newly allocated
2424 	 * page.
2425 	 */
2426 	if (elem_type == ECORE_ELEM_TASK) {
2427 		u32 elem_i;
2428 		u8 *elem_start = (u8 *)p_virt;
2429 		union type1_task_context *elem;
2430 
2431 		for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2432 			elem = (union type1_task_context *)elem_start;
2433 			SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2434 				  TDIF_TASK_CONTEXT_REFTAGMASK , 0xf);
2435 			elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2436 		}
2437 	}
2438 
2439 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2440 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2441 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2442 		p_blk->real_size_in_page;
2443 
2444 	/* compute absolute offset */
2445 	reg_offset = PSWRQ2_REG_ILT_MEMORY +
2446 		     (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2447 
2448 	ilt_hw_entry = 0;
2449 	SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2450 	SET_FIELD(ilt_hw_entry,
2451 		  ILT_ENTRY_PHY_ADDR,
2452 		  (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2453 
2454 	/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2455 	ecore_dmae_host2grc(p_hwfn, p_ptt, (u64)(osal_uintptr_t)&ilt_hw_entry,
2456 			    reg_offset, sizeof(ilt_hw_entry) / sizeof(u32),
2457 			    0 /* no flags */);
2458 
2459 	if (elem_type == ECORE_ELEM_CXT) {
2460 		u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2461 					 elems_per_p;
2462 
2463 		/* Update the relevant register in the parser */
2464 		ecore_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2465 			 last_cid_allocated - 1);
2466 
2467 		/* RoCE w/a -> we don't write to the prs search reg until first
2468 		 * cid is allocated. This is because the prs checks
2469 		 * last_cid-1 >=0 making 0 a valid value... this will cause
2470 		 * the a context load to occur on a RoCE packet received with
2471 		 * cid=0 even before context was initialized, can happen with a
2472 		 * stray packet from switch or a packet with crc-error
2473 		 */
2474 
2475 		if (!p_hwfn->b_rdma_enabled_in_prs) {
2476 			/* Enable Rdma search */
2477 			ecore_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2478 			p_hwfn->b_rdma_enabled_in_prs = true;
2479 		}
2480 	}
2481 
2482 out1:
2483 	ecore_ptt_release(p_hwfn, p_ptt);
2484 out0:
2485 	OSAL_MUTEX_RELEASE(&p_hwfn->p_cxt_mngr->mutex);
2486 
2487 	return rc;
2488 }
2489 
2490 /* This function is very RoCE oriented, if another protocol in the future
2491  * will want this feature we'll need to modify the function to be more generic
2492  */
2493 enum _ecore_status_t
ecore_cxt_free_ilt_range(struct ecore_hwfn * p_hwfn,enum ecore_cxt_elem_type elem_type,u32 start_iid,u32 count)2494 ecore_cxt_free_ilt_range(struct ecore_hwfn *p_hwfn,
2495 			 enum ecore_cxt_elem_type elem_type,
2496 			 u32 start_iid, u32 count)
2497 {
2498 	u32 start_line, end_line, shadow_start_line, shadow_end_line;
2499 	u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2500 	struct ecore_ilt_client_cfg *p_cli;
2501 	struct ecore_ilt_cli_blk *p_blk;
2502 	u32 end_iid = start_iid + count;
2503 	struct ecore_ptt *p_ptt;
2504 	u64 ilt_hw_entry = 0;
2505 	u32 i;
2506 
2507 	switch (elem_type) {
2508 	case ECORE_ELEM_CXT:
2509 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2510 		elem_size = CONN_CXT_SIZE(p_hwfn);
2511 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2512 		break;
2513 	case ECORE_ELEM_SRQ:
2514 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2515 		elem_size = SRQ_CXT_SIZE;
2516 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2517 		break;
2518 	case ECORE_ELEM_TASK:
2519 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2520 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2521 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(ECORE_CXT_ROCE_TID_SEG)];
2522 		break;
2523 	default:
2524 		DP_NOTICE(p_hwfn, false,
2525 			  "ECORE_INVALID elem type = %d", elem_type);
2526 		return ECORE_INVAL;
2527 	}
2528 
2529 	/* Calculate line in ilt */
2530 	hw_p_size = p_cli->p_size.val;
2531 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2532 	start_line = p_blk->start_line + (start_iid / elems_per_p);
2533 	end_line = p_blk->start_line + (end_iid / elems_per_p);
2534 	if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2535 		end_line--;
2536 
2537 	shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2538 	shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2539 
2540 	p_ptt = ecore_ptt_acquire(p_hwfn);
2541 	if (!p_ptt) {
2542 		DP_NOTICE(p_hwfn, false, "ECORE_TIME_OUT on ptt acquire - dynamic allocation");
2543 		return ECORE_TIMEOUT;
2544 	}
2545 
2546 	for (i = shadow_start_line; i < shadow_end_line; i++) {
2547 		if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2548 			continue;
2549 
2550 		OSAL_DMA_FREE_COHERENT(p_hwfn->p_dev,
2551 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2552 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys,
2553 				       p_hwfn->p_cxt_mngr->ilt_shadow[i].size);
2554 
2555 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = OSAL_NULL;
2556 		p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2557 		p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2558 
2559 		/* compute absolute offset */
2560 		reg_offset = PSWRQ2_REG_ILT_MEMORY +
2561 			     ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2562 			      ILT_ENTRY_IN_REGS);
2563 
2564 		/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2565 		 * wide-bus.
2566 		 */
2567 		ecore_dmae_host2grc(p_hwfn, p_ptt,
2568 				    (u64)(osal_uintptr_t)&ilt_hw_entry,
2569 				    reg_offset,
2570 				    sizeof(ilt_hw_entry) / sizeof(u32),
2571 				    0 /* no flags */);
2572 	}
2573 
2574 	ecore_ptt_release(p_hwfn, p_ptt);
2575 
2576 	return ECORE_SUCCESS;
2577 }
2578 
ecore_cxt_get_task_ctx(struct ecore_hwfn * p_hwfn,u32 tid,u8 ctx_type,void ** pp_task_ctx)2579 enum _ecore_status_t ecore_cxt_get_task_ctx(struct ecore_hwfn *p_hwfn,
2580 					    u32 tid,
2581 					    u8 ctx_type,
2582 					    void **pp_task_ctx)
2583 {
2584 	struct ecore_cxt_mngr		*p_mngr = p_hwfn->p_cxt_mngr;
2585 	struct ecore_ilt_client_cfg     *p_cli;
2586 	struct ecore_ilt_cli_blk	*p_seg;
2587 	struct ecore_tid_seg		*p_seg_info;
2588 	u32				proto, seg;
2589 	u32				total_lines;
2590 	u32				tid_size, ilt_idx;
2591 	u32				num_tids_per_block;
2592 
2593 	/* Verify the personality */
2594 	switch (p_hwfn->hw_info.personality) {
2595 	case ECORE_PCI_FCOE:
2596 		proto = PROTOCOLID_FCOE;
2597 		seg = ECORE_CXT_FCOE_TID_SEG;
2598 		break;
2599 	case ECORE_PCI_ISCSI:
2600 		proto = PROTOCOLID_ISCSI;
2601 		seg = ECORE_CXT_ISCSI_TID_SEG;
2602 		break;
2603 	default:
2604 		return ECORE_INVAL;
2605 	}
2606 
2607 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2608 	if (!p_cli->active) {
2609 		return ECORE_INVAL;
2610 	}
2611 
2612 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2613 
2614 	if (ctx_type == ECORE_CTX_WORKING_MEM) {
2615 		p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2616 	} else if (ctx_type == ECORE_CTX_FL_MEM) {
2617 		if (!p_seg_info->has_fl_mem) {
2618 			return ECORE_INVAL;
2619 		}
2620 		p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2621 	} else {
2622 		return ECORE_INVAL;
2623 	}
2624 	total_lines = DIV_ROUND_UP(p_seg->total_size,
2625 				   p_seg->real_size_in_page);
2626 	tid_size = p_mngr->task_type_size[p_seg_info->type];
2627 	num_tids_per_block = p_seg->real_size_in_page / tid_size;
2628 
2629 	if (total_lines < tid/num_tids_per_block)
2630 		return ECORE_INVAL;
2631 
2632 	ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2633 		  p_mngr->pf_start_line;
2634 	*pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2635 			     (tid % num_tids_per_block) * tid_size;
2636 
2637 	return ECORE_SUCCESS;
2638 }
2639