1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause)
2 /* QLogic qed NIC Driver
3  * Copyright (c) 2015-2017  QLogic Corporation
4  * Copyright (c) 2019-2020 Marvell International Ltd.
5  */
6 
7 #include <linux/types.h>
8 #include <linux/bitops.h>
9 #include <linux/dma-mapping.h>
10 #include <linux/errno.h>
11 #include <linux/kernel.h>
12 #include <linux/list.h>
13 #include <linux/log2.h>
14 #include <linux/pci.h>
15 #include <linux/slab.h>
16 #include <linux/string.h>
17 #include "qed.h"
18 #include "qed_cxt.h"
19 #include "qed_dev_api.h"
20 #include "qed_hsi.h"
21 #include "qed_hw.h"
22 #include "qed_init_ops.h"
23 #include "qed_rdma.h"
24 #include "qed_reg_addr.h"
25 #include "qed_sriov.h"
26 
27 /* QM constants */
28 #define QM_PQ_ELEMENT_SIZE	4 /* in bytes */
29 
30 /* Doorbell-Queue constants */
31 #define DQ_RANGE_SHIFT		4
32 #define DQ_RANGE_ALIGN		BIT(DQ_RANGE_SHIFT)
33 
34 /* Searcher constants */
35 #define SRC_MIN_NUM_ELEMS 256
36 
37 /* Timers constants */
38 #define TM_SHIFT        7
39 #define TM_ALIGN        BIT(TM_SHIFT)
40 #define TM_ELEM_SIZE    4
41 
42 #define ILT_DEFAULT_HW_P_SIZE	4
43 
44 #define ILT_PAGE_IN_BYTES(hw_p_size)	(1U << ((hw_p_size) + 12))
45 #define ILT_CFG_REG(cli, reg)	PSWRQ2_REG_ ## cli ## _ ## reg ## _RT_OFFSET
46 
47 /* ILT entry structure */
48 #define ILT_ENTRY_PHY_ADDR_MASK		(~0ULL >> 12)
49 #define ILT_ENTRY_PHY_ADDR_SHIFT	0
50 #define ILT_ENTRY_VALID_MASK		0x1ULL
51 #define ILT_ENTRY_VALID_SHIFT		52
52 #define ILT_ENTRY_IN_REGS		2
53 #define ILT_REG_SIZE_IN_BYTES		4
54 
55 /* connection context union */
56 union conn_context {
57 	struct e4_core_conn_context core_ctx;
58 	struct e4_eth_conn_context eth_ctx;
59 	struct e4_iscsi_conn_context iscsi_ctx;
60 	struct e4_fcoe_conn_context fcoe_ctx;
61 	struct e4_roce_conn_context roce_ctx;
62 };
63 
64 /* TYPE-0 task context - iSCSI, FCOE */
65 union type0_task_context {
66 	struct e4_iscsi_task_context iscsi_ctx;
67 	struct e4_fcoe_task_context fcoe_ctx;
68 };
69 
70 /* TYPE-1 task context - ROCE */
71 union type1_task_context {
72 	struct e4_rdma_task_context roce_ctx;
73 };
74 
75 struct src_ent {
76 	__u8				opaque[56];
77 	__be64				next;
78 };
79 
80 #define CDUT_SEG_ALIGNMET		3 /* in 4k chunks */
81 #define CDUT_SEG_ALIGNMET_IN_BYTES	BIT(CDUT_SEG_ALIGNMET + 12)
82 
83 #define CONN_CXT_SIZE(p_hwfn) \
84 	ALIGNED_TYPE_SIZE(union conn_context, p_hwfn)
85 
86 #define SRQ_CXT_SIZE (sizeof(struct rdma_srq_context))
87 #define XRC_SRQ_CXT_SIZE (sizeof(struct rdma_xrc_srq_context))
88 
89 #define TYPE0_TASK_CXT_SIZE(p_hwfn) \
90 	ALIGNED_TYPE_SIZE(union type0_task_context, p_hwfn)
91 
92 /* Alignment is inherent to the type1_task_context structure */
93 #define TYPE1_TASK_CXT_SIZE(p_hwfn) sizeof(union type1_task_context)
94 
src_proto(enum protocol_type type)95 static bool src_proto(enum protocol_type type)
96 {
97 	return type == PROTOCOLID_ISCSI ||
98 	       type == PROTOCOLID_FCOE ||
99 	       type == PROTOCOLID_IWARP;
100 }
101 
tm_cid_proto(enum protocol_type type)102 static bool tm_cid_proto(enum protocol_type type)
103 {
104 	return type == PROTOCOLID_ISCSI ||
105 	       type == PROTOCOLID_FCOE ||
106 	       type == PROTOCOLID_ROCE ||
107 	       type == PROTOCOLID_IWARP;
108 }
109 
tm_tid_proto(enum protocol_type type)110 static bool tm_tid_proto(enum protocol_type type)
111 {
112 	return type == PROTOCOLID_FCOE;
113 }
114 
115 /* counts the iids for the CDU/CDUC ILT client configuration */
116 struct qed_cdu_iids {
117 	u32 pf_cids;
118 	u32 per_vf_cids;
119 };
120 
qed_cxt_cdu_iids(struct qed_cxt_mngr *p_mngr, struct qed_cdu_iids *iids)121 static void qed_cxt_cdu_iids(struct qed_cxt_mngr *p_mngr,
122 			     struct qed_cdu_iids *iids)
123 {
124 	u32 type;
125 
126 	for (type = 0; type < MAX_CONN_TYPES; type++) {
127 		iids->pf_cids += p_mngr->conn_cfg[type].cid_count;
128 		iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
129 	}
130 }
131 
132 /* counts the iids for the Searcher block configuration */
133 struct qed_src_iids {
134 	u32 pf_cids;
135 	u32 per_vf_cids;
136 };
137 
qed_cxt_src_iids(struct qed_cxt_mngr *p_mngr, struct qed_src_iids *iids)138 static void qed_cxt_src_iids(struct qed_cxt_mngr *p_mngr,
139 			     struct qed_src_iids *iids)
140 {
141 	u32 i;
142 
143 	for (i = 0; i < MAX_CONN_TYPES; i++) {
144 		if (!src_proto(i))
145 			continue;
146 
147 		iids->pf_cids += p_mngr->conn_cfg[i].cid_count;
148 		iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf;
149 	}
150 
151 	/* Add L2 filtering filters in addition */
152 	iids->pf_cids += p_mngr->arfs_count;
153 }
154 
155 /* counts the iids for the Timers block configuration */
156 struct qed_tm_iids {
157 	u32 pf_cids;
158 	u32 pf_tids[NUM_TASK_PF_SEGMENTS];	/* per segment */
159 	u32 pf_tids_total;
160 	u32 per_vf_cids;
161 	u32 per_vf_tids;
162 };
163 
qed_cxt_tm_iids(struct qed_hwfn *p_hwfn, struct qed_cxt_mngr *p_mngr, struct qed_tm_iids *iids)164 static void qed_cxt_tm_iids(struct qed_hwfn *p_hwfn,
165 			    struct qed_cxt_mngr *p_mngr,
166 			    struct qed_tm_iids *iids)
167 {
168 	bool tm_vf_required = false;
169 	bool tm_required = false;
170 	int i, j;
171 
172 	/* Timers is a special case -> we don't count how many cids require
173 	 * timers but what's the max cid that will be used by the timer block.
174 	 * therefore we traverse in reverse order, and once we hit a protocol
175 	 * that requires the timers memory, we'll sum all the protocols up
176 	 * to that one.
177 	 */
178 	for (i = MAX_CONN_TYPES - 1; i >= 0; i--) {
179 		struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
180 
181 		if (tm_cid_proto(i) || tm_required) {
182 			if (p_cfg->cid_count)
183 				tm_required = true;
184 
185 			iids->pf_cids += p_cfg->cid_count;
186 		}
187 
188 		if (tm_cid_proto(i) || tm_vf_required) {
189 			if (p_cfg->cids_per_vf)
190 				tm_vf_required = true;
191 
192 			iids->per_vf_cids += p_cfg->cids_per_vf;
193 		}
194 
195 		if (tm_tid_proto(i)) {
196 			struct qed_tid_seg *segs = p_cfg->tid_seg;
197 
198 			/* for each segment there is at most one
199 			 * protocol for which count is not 0.
200 			 */
201 			for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
202 				iids->pf_tids[j] += segs[j].count;
203 
204 			/* The last array elelment is for the VFs. As for PF
205 			 * segments there can be only one protocol for
206 			 * which this value is not 0.
207 			 */
208 			iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
209 		}
210 	}
211 
212 	iids->pf_cids = roundup(iids->pf_cids, TM_ALIGN);
213 	iids->per_vf_cids = roundup(iids->per_vf_cids, TM_ALIGN);
214 	iids->per_vf_tids = roundup(iids->per_vf_tids, TM_ALIGN);
215 
216 	for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
217 		iids->pf_tids[j] = roundup(iids->pf_tids[j], TM_ALIGN);
218 		iids->pf_tids_total += iids->pf_tids[j];
219 	}
220 }
221 
qed_cxt_qm_iids(struct qed_hwfn *p_hwfn, struct qed_qm_iids *iids)222 static void qed_cxt_qm_iids(struct qed_hwfn *p_hwfn,
223 			    struct qed_qm_iids *iids)
224 {
225 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
226 	struct qed_tid_seg *segs;
227 	u32 vf_cids = 0, type, j;
228 	u32 vf_tids = 0;
229 
230 	for (type = 0; type < MAX_CONN_TYPES; type++) {
231 		iids->cids += p_mngr->conn_cfg[type].cid_count;
232 		vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
233 
234 		segs = p_mngr->conn_cfg[type].tid_seg;
235 		/* for each segment there is at most one
236 		 * protocol for which count is not 0.
237 		 */
238 		for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
239 			iids->tids += segs[j].count;
240 
241 		/* The last array elelment is for the VFs. As for PF
242 		 * segments there can be only one protocol for
243 		 * which this value is not 0.
244 		 */
245 		vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
246 	}
247 
248 	iids->vf_cids = vf_cids;
249 	iids->tids += vf_tids * p_mngr->vf_count;
250 
251 	DP_VERBOSE(p_hwfn, QED_MSG_ILT,
252 		   "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
253 		   iids->cids, iids->vf_cids, iids->tids, vf_tids);
254 }
255 
qed_cxt_tid_seg_info(struct qed_hwfn *p_hwfn, u32 seg)256 static struct qed_tid_seg *qed_cxt_tid_seg_info(struct qed_hwfn *p_hwfn,
257 						u32 seg)
258 {
259 	struct qed_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
260 	u32 i;
261 
262 	/* Find the protocol with tid count > 0 for this segment.
263 	 * Note: there can only be one and this is already validated.
264 	 */
265 	for (i = 0; i < MAX_CONN_TYPES; i++)
266 		if (p_cfg->conn_cfg[i].tid_seg[seg].count)
267 			return &p_cfg->conn_cfg[i].tid_seg[seg];
268 	return NULL;
269 }
270 
qed_cxt_set_srq_count(struct qed_hwfn *p_hwfn, u32 num_srqs, u32 num_xrc_srqs)271 static void qed_cxt_set_srq_count(struct qed_hwfn *p_hwfn,
272 				  u32 num_srqs, u32 num_xrc_srqs)
273 {
274 	struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
275 
276 	p_mgr->srq_count = num_srqs;
277 	p_mgr->xrc_srq_count = num_xrc_srqs;
278 }
279 
qed_cxt_get_ilt_page_size(struct qed_hwfn *p_hwfn, enum ilt_clients ilt_client)280 u32 qed_cxt_get_ilt_page_size(struct qed_hwfn *p_hwfn,
281 			      enum ilt_clients ilt_client)
282 {
283 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
284 	struct qed_ilt_client_cfg *p_cli = &p_mngr->clients[ilt_client];
285 
286 	return ILT_PAGE_IN_BYTES(p_cli->p_size.val);
287 }
288 
qed_cxt_xrc_srqs_per_page(struct qed_hwfn *p_hwfn)289 static u32 qed_cxt_xrc_srqs_per_page(struct qed_hwfn *p_hwfn)
290 {
291 	u32 page_size;
292 
293 	page_size = qed_cxt_get_ilt_page_size(p_hwfn, ILT_CLI_TSDM);
294 	return page_size / XRC_SRQ_CXT_SIZE;
295 }
296 
qed_cxt_get_total_srq_count(struct qed_hwfn *p_hwfn)297 u32 qed_cxt_get_total_srq_count(struct qed_hwfn *p_hwfn)
298 {
299 	struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
300 	u32 total_srqs;
301 
302 	total_srqs = p_mgr->srq_count + p_mgr->xrc_srq_count;
303 
304 	return total_srqs;
305 }
306 
307 /* set the iids count per protocol */
qed_cxt_set_proto_cid_count(struct qed_hwfn *p_hwfn, enum protocol_type type, u32 cid_count, u32 vf_cid_cnt)308 static void qed_cxt_set_proto_cid_count(struct qed_hwfn *p_hwfn,
309 					enum protocol_type type,
310 					u32 cid_count, u32 vf_cid_cnt)
311 {
312 	struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
313 	struct qed_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
314 
315 	p_conn->cid_count = roundup(cid_count, DQ_RANGE_ALIGN);
316 	p_conn->cids_per_vf = roundup(vf_cid_cnt, DQ_RANGE_ALIGN);
317 
318 	if (type == PROTOCOLID_ROCE) {
319 		u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
320 		u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
321 		u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
322 		u32 align = elems_per_page * DQ_RANGE_ALIGN;
323 
324 		p_conn->cid_count = roundup(p_conn->cid_count, align);
325 	}
326 }
327 
qed_cxt_get_proto_cid_count(struct qed_hwfn *p_hwfn, enum protocol_type type, u32 *vf_cid)328 u32 qed_cxt_get_proto_cid_count(struct qed_hwfn *p_hwfn,
329 				enum protocol_type type, u32 *vf_cid)
330 {
331 	if (vf_cid)
332 		*vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
333 
334 	return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
335 }
336 
qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn, enum protocol_type type)337 u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
338 				enum protocol_type type)
339 {
340 	return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
341 }
342 
qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn, enum protocol_type type)343 u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
344 				enum protocol_type type)
345 {
346 	u32 cnt = 0;
347 	int i;
348 
349 	for (i = 0; i < TASK_SEGMENTS; i++)
350 		cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
351 
352 	return cnt;
353 }
354 
qed_cxt_set_proto_tid_count(struct qed_hwfn *p_hwfn, enum protocol_type proto, u8 seg, u8 seg_type, u32 count, bool has_fl)355 static void qed_cxt_set_proto_tid_count(struct qed_hwfn *p_hwfn,
356 					enum protocol_type proto,
357 					u8 seg,
358 					u8 seg_type, u32 count, bool has_fl)
359 {
360 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
361 	struct qed_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
362 
363 	p_seg->count = count;
364 	p_seg->has_fl_mem = has_fl;
365 	p_seg->type = seg_type;
366 }
367 
qed_ilt_cli_blk_fill(struct qed_ilt_client_cfg *p_cli, struct qed_ilt_cli_blk *p_blk, u32 start_line, u32 total_size, u32 elem_size)368 static void qed_ilt_cli_blk_fill(struct qed_ilt_client_cfg *p_cli,
369 				 struct qed_ilt_cli_blk *p_blk,
370 				 u32 start_line, u32 total_size, u32 elem_size)
371 {
372 	u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
373 
374 	/* verify thatits called only once for each block */
375 	if (p_blk->total_size)
376 		return;
377 
378 	p_blk->total_size = total_size;
379 	p_blk->real_size_in_page = 0;
380 	if (elem_size)
381 		p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
382 	p_blk->start_line = start_line;
383 }
384 
qed_ilt_cli_adv_line(struct qed_hwfn *p_hwfn, struct qed_ilt_client_cfg *p_cli, struct qed_ilt_cli_blk *p_blk, u32 *p_line, enum ilt_clients client_id)385 static void qed_ilt_cli_adv_line(struct qed_hwfn *p_hwfn,
386 				 struct qed_ilt_client_cfg *p_cli,
387 				 struct qed_ilt_cli_blk *p_blk,
388 				 u32 *p_line, enum ilt_clients client_id)
389 {
390 	if (!p_blk->total_size)
391 		return;
392 
393 	if (!p_cli->active)
394 		p_cli->first.val = *p_line;
395 
396 	p_cli->active = true;
397 	*p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
398 	p_cli->last.val = *p_line - 1;
399 
400 	DP_VERBOSE(p_hwfn, QED_MSG_ILT,
401 		   "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
402 		   client_id, p_cli->first.val,
403 		   p_cli->last.val, p_blk->total_size,
404 		   p_blk->real_size_in_page, p_blk->start_line);
405 }
406 
qed_ilt_get_dynamic_line_cnt(struct qed_hwfn *p_hwfn, enum ilt_clients ilt_client)407 static u32 qed_ilt_get_dynamic_line_cnt(struct qed_hwfn *p_hwfn,
408 					enum ilt_clients ilt_client)
409 {
410 	u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
411 	struct qed_ilt_client_cfg *p_cli;
412 	u32 lines_to_skip = 0;
413 	u32 cxts_per_p;
414 
415 	if (ilt_client == ILT_CLI_CDUC) {
416 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
417 
418 		cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
419 		    (u32) CONN_CXT_SIZE(p_hwfn);
420 
421 		lines_to_skip = cid_count / cxts_per_p;
422 	}
423 
424 	return lines_to_skip;
425 }
426 
qed_cxt_set_cli(struct qed_ilt_client_cfg *p_cli)427 static struct qed_ilt_client_cfg *qed_cxt_set_cli(struct qed_ilt_client_cfg
428 						  *p_cli)
429 {
430 	p_cli->active = false;
431 	p_cli->first.val = 0;
432 	p_cli->last.val = 0;
433 	return p_cli;
434 }
435 
qed_cxt_set_blk(struct qed_ilt_cli_blk *p_blk)436 static struct qed_ilt_cli_blk *qed_cxt_set_blk(struct qed_ilt_cli_blk *p_blk)
437 {
438 	p_blk->total_size = 0;
439 	return p_blk;
440 }
441 
qed_cxt_ilt_blk_reset(struct qed_hwfn *p_hwfn)442 static void qed_cxt_ilt_blk_reset(struct qed_hwfn *p_hwfn)
443 {
444 	struct qed_ilt_client_cfg *clients = p_hwfn->p_cxt_mngr->clients;
445 	u32 cli_idx, blk_idx;
446 
447 	for (cli_idx = 0; cli_idx < MAX_ILT_CLIENTS; cli_idx++) {
448 		for (blk_idx = 0; blk_idx < ILT_CLI_PF_BLOCKS; blk_idx++)
449 			clients[cli_idx].pf_blks[blk_idx].total_size = 0;
450 
451 		for (blk_idx = 0; blk_idx < ILT_CLI_VF_BLOCKS; blk_idx++)
452 			clients[cli_idx].vf_blks[blk_idx].total_size = 0;
453 	}
454 }
455 
qed_cxt_cfg_ilt_compute(struct qed_hwfn *p_hwfn, u32 *line_count)456 int qed_cxt_cfg_ilt_compute(struct qed_hwfn *p_hwfn, u32 *line_count)
457 {
458 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
459 	u32 curr_line, total, i, task_size, line;
460 	struct qed_ilt_client_cfg *p_cli;
461 	struct qed_ilt_cli_blk *p_blk;
462 	struct qed_cdu_iids cdu_iids;
463 	struct qed_src_iids src_iids;
464 	struct qed_qm_iids qm_iids;
465 	struct qed_tm_iids tm_iids;
466 	struct qed_tid_seg *p_seg;
467 
468 	memset(&qm_iids, 0, sizeof(qm_iids));
469 	memset(&cdu_iids, 0, sizeof(cdu_iids));
470 	memset(&src_iids, 0, sizeof(src_iids));
471 	memset(&tm_iids, 0, sizeof(tm_iids));
472 
473 	p_mngr->pf_start_line = RESC_START(p_hwfn, QED_ILT);
474 
475 	/* Reset all ILT blocks at the beginning of ILT computing in order
476 	 * to prevent memory allocation for irrelevant blocks afterwards.
477 	 */
478 	qed_cxt_ilt_blk_reset(p_hwfn);
479 
480 	DP_VERBOSE(p_hwfn, QED_MSG_ILT,
481 		   "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
482 		   p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
483 
484 	/* CDUC */
485 	p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUC]);
486 
487 	curr_line = p_mngr->pf_start_line;
488 
489 	/* CDUC PF */
490 	p_cli->pf_total_lines = 0;
491 
492 	/* get the counters for the CDUC and QM clients  */
493 	qed_cxt_cdu_iids(p_mngr, &cdu_iids);
494 
495 	p_blk = qed_cxt_set_blk(&p_cli->pf_blks[CDUC_BLK]);
496 
497 	total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
498 
499 	qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
500 			     total, CONN_CXT_SIZE(p_hwfn));
501 
502 	qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
503 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
504 
505 	p_blk->dynamic_line_cnt = qed_ilt_get_dynamic_line_cnt(p_hwfn,
506 							       ILT_CLI_CDUC);
507 
508 	/* CDUC VF */
509 	p_blk = qed_cxt_set_blk(&p_cli->vf_blks[CDUC_BLK]);
510 	total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
511 
512 	qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
513 			     total, CONN_CXT_SIZE(p_hwfn));
514 
515 	qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
516 	p_cli->vf_total_lines = curr_line - p_blk->start_line;
517 
518 	for (i = 1; i < p_mngr->vf_count; i++)
519 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
520 				     ILT_CLI_CDUC);
521 
522 	/* CDUT PF */
523 	p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUT]);
524 	p_cli->first.val = curr_line;
525 
526 	/* first the 'working' task memory */
527 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
528 		p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
529 		if (!p_seg || p_seg->count == 0)
530 			continue;
531 
532 		p_blk = qed_cxt_set_blk(&p_cli->pf_blks[CDUT_SEG_BLK(i)]);
533 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
534 		qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
535 				     p_mngr->task_type_size[p_seg->type]);
536 
537 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
538 				     ILT_CLI_CDUT);
539 	}
540 
541 	/* next the 'init' task memory (forced load memory) */
542 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
543 		p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
544 		if (!p_seg || p_seg->count == 0)
545 			continue;
546 
547 		p_blk =
548 		    qed_cxt_set_blk(&p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)]);
549 
550 		if (!p_seg->has_fl_mem) {
551 			/* The segment is active (total size pf 'working'
552 			 * memory is > 0) but has no FL (forced-load, Init)
553 			 * memory. Thus:
554 			 *
555 			 * 1.   The total-size in the corrsponding FL block of
556 			 *      the ILT client is set to 0 - No ILT line are
557 			 *      provisioned and no ILT memory allocated.
558 			 *
559 			 * 2.   The start-line of said block is set to the
560 			 *      start line of the matching working memory
561 			 *      block in the ILT client. This is later used to
562 			 *      configure the CDU segment offset registers and
563 			 *      results in an FL command for TIDs of this
564 			 *      segement behaves as regular load commands
565 			 *      (loading TIDs from the working memory).
566 			 */
567 			line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
568 
569 			qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
570 			continue;
571 		}
572 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
573 
574 		qed_ilt_cli_blk_fill(p_cli, p_blk,
575 				     curr_line, total,
576 				     p_mngr->task_type_size[p_seg->type]);
577 
578 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
579 				     ILT_CLI_CDUT);
580 	}
581 	p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
582 
583 	/* CDUT VF */
584 	p_seg = qed_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
585 	if (p_seg && p_seg->count) {
586 		/* Stricly speaking we need to iterate over all VF
587 		 * task segment types, but a VF has only 1 segment
588 		 */
589 
590 		/* 'working' memory */
591 		total = p_seg->count * p_mngr->task_type_size[p_seg->type];
592 
593 		p_blk = qed_cxt_set_blk(&p_cli->vf_blks[CDUT_SEG_BLK(0)]);
594 		qed_ilt_cli_blk_fill(p_cli, p_blk,
595 				     curr_line, total,
596 				     p_mngr->task_type_size[p_seg->type]);
597 
598 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
599 				     ILT_CLI_CDUT);
600 
601 		/* 'init' memory */
602 		p_blk =
603 		    qed_cxt_set_blk(&p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)]);
604 		if (!p_seg->has_fl_mem) {
605 			/* see comment above */
606 			line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
607 			qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
608 		} else {
609 			task_size = p_mngr->task_type_size[p_seg->type];
610 			qed_ilt_cli_blk_fill(p_cli, p_blk,
611 					     curr_line, total, task_size);
612 			qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
613 					     ILT_CLI_CDUT);
614 		}
615 		p_cli->vf_total_lines = curr_line -
616 		    p_cli->vf_blks[0].start_line;
617 
618 		/* Now for the rest of the VFs */
619 		for (i = 1; i < p_mngr->vf_count; i++) {
620 			p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
621 			qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
622 					     ILT_CLI_CDUT);
623 
624 			p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
625 			qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
626 					     ILT_CLI_CDUT);
627 		}
628 	}
629 
630 	/* QM */
631 	p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_QM]);
632 	p_blk = qed_cxt_set_blk(&p_cli->pf_blks[0]);
633 
634 	qed_cxt_qm_iids(p_hwfn, &qm_iids);
635 	total = qed_qm_pf_mem_size(qm_iids.cids,
636 				   qm_iids.vf_cids, qm_iids.tids,
637 				   p_hwfn->qm_info.num_pqs,
638 				   p_hwfn->qm_info.num_vf_pqs);
639 
640 	DP_VERBOSE(p_hwfn,
641 		   QED_MSG_ILT,
642 		   "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
643 		   qm_iids.cids,
644 		   qm_iids.vf_cids,
645 		   qm_iids.tids,
646 		   p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
647 
648 	qed_ilt_cli_blk_fill(p_cli, p_blk,
649 			     curr_line, total * 0x1000,
650 			     QM_PQ_ELEMENT_SIZE);
651 
652 	qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
653 	p_cli->pf_total_lines = curr_line - p_blk->start_line;
654 
655 	/* SRC */
656 	p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_SRC]);
657 	qed_cxt_src_iids(p_mngr, &src_iids);
658 
659 	/* Both the PF and VFs searcher connections are stored in the per PF
660 	 * database. Thus sum the PF searcher cids and all the VFs searcher
661 	 * cids.
662 	 */
663 	total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
664 	if (total) {
665 		u32 local_max = max_t(u32, total,
666 				      SRC_MIN_NUM_ELEMS);
667 
668 		total = roundup_pow_of_two(local_max);
669 
670 		p_blk = qed_cxt_set_blk(&p_cli->pf_blks[0]);
671 		qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
672 				     total * sizeof(struct src_ent),
673 				     sizeof(struct src_ent));
674 
675 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
676 				     ILT_CLI_SRC);
677 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
678 	}
679 
680 	/* TM PF */
681 	p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_TM]);
682 	qed_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
683 	total = tm_iids.pf_cids + tm_iids.pf_tids_total;
684 	if (total) {
685 		p_blk = qed_cxt_set_blk(&p_cli->pf_blks[0]);
686 		qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
687 				     total * TM_ELEM_SIZE, TM_ELEM_SIZE);
688 
689 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
690 				     ILT_CLI_TM);
691 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
692 	}
693 
694 	/* TM VF */
695 	total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
696 	if (total) {
697 		p_blk = qed_cxt_set_blk(&p_cli->vf_blks[0]);
698 		qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
699 				     total * TM_ELEM_SIZE, TM_ELEM_SIZE);
700 
701 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
702 				     ILT_CLI_TM);
703 
704 		p_cli->vf_total_lines = curr_line - p_blk->start_line;
705 		for (i = 1; i < p_mngr->vf_count; i++)
706 			qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
707 					     ILT_CLI_TM);
708 	}
709 
710 	/* TSDM (SRQ CONTEXT) */
711 	total = qed_cxt_get_total_srq_count(p_hwfn);
712 
713 	if (total) {
714 		p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_TSDM]);
715 		p_blk = qed_cxt_set_blk(&p_cli->pf_blks[SRQ_BLK]);
716 		qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
717 				     total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
718 
719 		qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
720 				     ILT_CLI_TSDM);
721 		p_cli->pf_total_lines = curr_line - p_blk->start_line;
722 	}
723 
724 	*line_count = curr_line - p_hwfn->p_cxt_mngr->pf_start_line;
725 
726 	if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
727 	    RESC_NUM(p_hwfn, QED_ILT))
728 		return -EINVAL;
729 
730 	return 0;
731 }
732 
qed_cxt_cfg_ilt_compute_excess(struct qed_hwfn *p_hwfn, u32 used_lines)733 u32 qed_cxt_cfg_ilt_compute_excess(struct qed_hwfn *p_hwfn, u32 used_lines)
734 {
735 	struct qed_ilt_client_cfg *p_cli;
736 	u32 excess_lines, available_lines;
737 	struct qed_cxt_mngr *p_mngr;
738 	u32 ilt_page_size, elem_size;
739 	struct qed_tid_seg *p_seg;
740 	int i;
741 
742 	available_lines = RESC_NUM(p_hwfn, QED_ILT);
743 	excess_lines = used_lines - available_lines;
744 
745 	if (!excess_lines)
746 		return 0;
747 
748 	if (!QED_IS_RDMA_PERSONALITY(p_hwfn))
749 		return 0;
750 
751 	p_mngr = p_hwfn->p_cxt_mngr;
752 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
753 	ilt_page_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
754 
755 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
756 		p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
757 		if (!p_seg || p_seg->count == 0)
758 			continue;
759 
760 		elem_size = p_mngr->task_type_size[p_seg->type];
761 		if (!elem_size)
762 			continue;
763 
764 		return (ilt_page_size / elem_size) * excess_lines;
765 	}
766 
767 	DP_NOTICE(p_hwfn, "failed computing excess ILT lines\n");
768 	return 0;
769 }
770 
qed_cxt_src_t2_free(struct qed_hwfn *p_hwfn)771 static void qed_cxt_src_t2_free(struct qed_hwfn *p_hwfn)
772 {
773 	struct qed_src_t2 *p_t2 = &p_hwfn->p_cxt_mngr->src_t2;
774 	u32 i;
775 
776 	if (!p_t2 || !p_t2->dma_mem)
777 		return;
778 
779 	for (i = 0; i < p_t2->num_pages; i++)
780 		if (p_t2->dma_mem[i].virt_addr)
781 			dma_free_coherent(&p_hwfn->cdev->pdev->dev,
782 					  p_t2->dma_mem[i].size,
783 					  p_t2->dma_mem[i].virt_addr,
784 					  p_t2->dma_mem[i].phys_addr);
785 
786 	kfree(p_t2->dma_mem);
787 	p_t2->dma_mem = NULL;
788 }
789 
790 static int
qed_cxt_t2_alloc_pages(struct qed_hwfn *p_hwfn, struct qed_src_t2 *p_t2, u32 total_size, u32 page_size)791 qed_cxt_t2_alloc_pages(struct qed_hwfn *p_hwfn,
792 		       struct qed_src_t2 *p_t2, u32 total_size, u32 page_size)
793 {
794 	void **p_virt;
795 	u32 size, i;
796 
797 	if (!p_t2 || !p_t2->dma_mem)
798 		return -EINVAL;
799 
800 	for (i = 0; i < p_t2->num_pages; i++) {
801 		size = min_t(u32, total_size, page_size);
802 		p_virt = &p_t2->dma_mem[i].virt_addr;
803 
804 		*p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
805 					     size,
806 					     &p_t2->dma_mem[i].phys_addr,
807 					     GFP_KERNEL);
808 		if (!p_t2->dma_mem[i].virt_addr)
809 			return -ENOMEM;
810 
811 		memset(*p_virt, 0, size);
812 		p_t2->dma_mem[i].size = size;
813 		total_size -= size;
814 	}
815 
816 	return 0;
817 }
818 
qed_cxt_src_t2_alloc(struct qed_hwfn *p_hwfn)819 static int qed_cxt_src_t2_alloc(struct qed_hwfn *p_hwfn)
820 {
821 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
822 	u32 conn_num, total_size, ent_per_page, psz, i;
823 	struct phys_mem_desc *p_t2_last_page;
824 	struct qed_ilt_client_cfg *p_src;
825 	struct qed_src_iids src_iids;
826 	struct qed_src_t2 *p_t2;
827 	int rc;
828 
829 	memset(&src_iids, 0, sizeof(src_iids));
830 
831 	/* if the SRC ILT client is inactive - there are no connection
832 	 * requiring the searcer, leave.
833 	 */
834 	p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
835 	if (!p_src->active)
836 		return 0;
837 
838 	qed_cxt_src_iids(p_mngr, &src_iids);
839 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
840 	total_size = conn_num * sizeof(struct src_ent);
841 
842 	/* use the same page size as the SRC ILT client */
843 	psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
844 	p_t2 = &p_mngr->src_t2;
845 	p_t2->num_pages = DIV_ROUND_UP(total_size, psz);
846 
847 	/* allocate t2 */
848 	p_t2->dma_mem = kcalloc(p_t2->num_pages, sizeof(struct phys_mem_desc),
849 				GFP_KERNEL);
850 	if (!p_t2->dma_mem) {
851 		DP_NOTICE(p_hwfn, "Failed to allocate t2 table\n");
852 		rc = -ENOMEM;
853 		goto t2_fail;
854 	}
855 
856 	rc = qed_cxt_t2_alloc_pages(p_hwfn, p_t2, total_size, psz);
857 	if (rc)
858 		goto t2_fail;
859 
860 	/* Set the t2 pointers */
861 
862 	/* entries per page - must be a power of two */
863 	ent_per_page = psz / sizeof(struct src_ent);
864 
865 	p_t2->first_free = (u64)p_t2->dma_mem[0].phys_addr;
866 
867 	p_t2_last_page = &p_t2->dma_mem[(conn_num - 1) / ent_per_page];
868 	p_t2->last_free = (u64)p_t2_last_page->phys_addr +
869 	    ((conn_num - 1) & (ent_per_page - 1)) * sizeof(struct src_ent);
870 
871 	for (i = 0; i < p_t2->num_pages; i++) {
872 		u32 ent_num = min_t(u32,
873 				    ent_per_page,
874 				    conn_num);
875 		struct src_ent *entries = p_t2->dma_mem[i].virt_addr;
876 		u64 p_ent_phys = (u64)p_t2->dma_mem[i].phys_addr, val;
877 		u32 j;
878 
879 		for (j = 0; j < ent_num - 1; j++) {
880 			val = p_ent_phys + (j + 1) * sizeof(struct src_ent);
881 			entries[j].next = cpu_to_be64(val);
882 		}
883 
884 		if (i < p_t2->num_pages - 1)
885 			val = (u64)p_t2->dma_mem[i + 1].phys_addr;
886 		else
887 			val = 0;
888 		entries[j].next = cpu_to_be64(val);
889 
890 		conn_num -= ent_num;
891 	}
892 
893 	return 0;
894 
895 t2_fail:
896 	qed_cxt_src_t2_free(p_hwfn);
897 	return rc;
898 }
899 
900 #define for_each_ilt_valid_client(pos, clients)	\
901 	for (pos = 0; pos < MAX_ILT_CLIENTS; pos++)	\
902 		if (!clients[pos].active) {	\
903 			continue;		\
904 		} else				\
905 
906 /* Total number of ILT lines used by this PF */
qed_cxt_ilt_shadow_size(struct qed_ilt_client_cfg *ilt_clients)907 static u32 qed_cxt_ilt_shadow_size(struct qed_ilt_client_cfg *ilt_clients)
908 {
909 	u32 size = 0;
910 	u32 i;
911 
912 	for_each_ilt_valid_client(i, ilt_clients)
913 	    size += (ilt_clients[i].last.val - ilt_clients[i].first.val + 1);
914 
915 	return size;
916 }
917 
qed_ilt_shadow_free(struct qed_hwfn *p_hwfn)918 static void qed_ilt_shadow_free(struct qed_hwfn *p_hwfn)
919 {
920 	struct qed_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
921 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
922 	u32 ilt_size, i;
923 
924 	ilt_size = qed_cxt_ilt_shadow_size(p_cli);
925 
926 	for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
927 		struct phys_mem_desc *p_dma = &p_mngr->ilt_shadow[i];
928 
929 		if (p_dma->virt_addr)
930 			dma_free_coherent(&p_hwfn->cdev->pdev->dev,
931 					  p_dma->size, p_dma->virt_addr,
932 					  p_dma->phys_addr);
933 		p_dma->virt_addr = NULL;
934 	}
935 	kfree(p_mngr->ilt_shadow);
936 	p_mngr->ilt_shadow = NULL;
937 }
938 
qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn, struct qed_ilt_cli_blk *p_blk, enum ilt_clients ilt_client, u32 start_line_offset)939 static int qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn,
940 			     struct qed_ilt_cli_blk *p_blk,
941 			     enum ilt_clients ilt_client,
942 			     u32 start_line_offset)
943 {
944 	struct phys_mem_desc *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
945 	u32 lines, line, sz_left, lines_to_skip = 0;
946 
947 	/* Special handling for RoCE that supports dynamic allocation */
948 	if (QED_IS_RDMA_PERSONALITY(p_hwfn) &&
949 	    ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
950 		return 0;
951 
952 	lines_to_skip = p_blk->dynamic_line_cnt;
953 
954 	if (!p_blk->total_size)
955 		return 0;
956 
957 	sz_left = p_blk->total_size;
958 	lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) - lines_to_skip;
959 	line = p_blk->start_line + start_line_offset -
960 	    p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
961 
962 	for (; lines; lines--) {
963 		dma_addr_t p_phys;
964 		void *p_virt;
965 		u32 size;
966 
967 		size = min_t(u32, sz_left, p_blk->real_size_in_page);
968 		p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev, size,
969 					    &p_phys, GFP_KERNEL);
970 		if (!p_virt)
971 			return -ENOMEM;
972 
973 		ilt_shadow[line].phys_addr = p_phys;
974 		ilt_shadow[line].virt_addr = p_virt;
975 		ilt_shadow[line].size = size;
976 
977 		DP_VERBOSE(p_hwfn, QED_MSG_ILT,
978 			   "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
979 			    line, (u64)p_phys, p_virt, size);
980 
981 		sz_left -= size;
982 		line++;
983 	}
984 
985 	return 0;
986 }
987 
qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn)988 static int qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn)
989 {
990 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
991 	struct qed_ilt_client_cfg *clients = p_mngr->clients;
992 	struct qed_ilt_cli_blk *p_blk;
993 	u32 size, i, j, k;
994 	int rc;
995 
996 	size = qed_cxt_ilt_shadow_size(clients);
997 	p_mngr->ilt_shadow = kcalloc(size, sizeof(struct phys_mem_desc),
998 				     GFP_KERNEL);
999 	if (!p_mngr->ilt_shadow) {
1000 		rc = -ENOMEM;
1001 		goto ilt_shadow_fail;
1002 	}
1003 
1004 	DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1005 		   "Allocated 0x%x bytes for ilt shadow\n",
1006 		   (u32)(size * sizeof(struct phys_mem_desc)));
1007 
1008 	for_each_ilt_valid_client(i, clients) {
1009 		for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1010 			p_blk = &clients[i].pf_blks[j];
1011 			rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1012 			if (rc)
1013 				goto ilt_shadow_fail;
1014 		}
1015 		for (k = 0; k < p_mngr->vf_count; k++) {
1016 			for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1017 				u32 lines = clients[i].vf_total_lines * k;
1018 
1019 				p_blk = &clients[i].vf_blks[j];
1020 				rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, lines);
1021 				if (rc)
1022 					goto ilt_shadow_fail;
1023 			}
1024 		}
1025 	}
1026 
1027 	return 0;
1028 
1029 ilt_shadow_fail:
1030 	qed_ilt_shadow_free(p_hwfn);
1031 	return rc;
1032 }
1033 
qed_cid_map_free(struct qed_hwfn *p_hwfn)1034 static void qed_cid_map_free(struct qed_hwfn *p_hwfn)
1035 {
1036 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1037 	u32 type, vf;
1038 
1039 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1040 		kfree(p_mngr->acquired[type].cid_map);
1041 		p_mngr->acquired[type].max_count = 0;
1042 		p_mngr->acquired[type].start_cid = 0;
1043 
1044 		for (vf = 0; vf < MAX_NUM_VFS; vf++) {
1045 			kfree(p_mngr->acquired_vf[type][vf].cid_map);
1046 			p_mngr->acquired_vf[type][vf].max_count = 0;
1047 			p_mngr->acquired_vf[type][vf].start_cid = 0;
1048 		}
1049 	}
1050 }
1051 
1052 static int
qed_cid_map_alloc_single(struct qed_hwfn *p_hwfn, u32 type, u32 cid_start, u32 cid_count, struct qed_cid_acquired_map *p_map)1053 qed_cid_map_alloc_single(struct qed_hwfn *p_hwfn,
1054 			 u32 type,
1055 			 u32 cid_start,
1056 			 u32 cid_count, struct qed_cid_acquired_map *p_map)
1057 {
1058 	u32 size;
1059 
1060 	if (!cid_count)
1061 		return 0;
1062 
1063 	size = DIV_ROUND_UP(cid_count,
1064 			    sizeof(unsigned long) * BITS_PER_BYTE) *
1065 	       sizeof(unsigned long);
1066 	p_map->cid_map = kzalloc(size, GFP_KERNEL);
1067 	if (!p_map->cid_map)
1068 		return -ENOMEM;
1069 
1070 	p_map->max_count = cid_count;
1071 	p_map->start_cid = cid_start;
1072 
1073 	DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1074 		   "Type %08x start: %08x count %08x\n",
1075 		   type, p_map->start_cid, p_map->max_count);
1076 
1077 	return 0;
1078 }
1079 
qed_cid_map_alloc(struct qed_hwfn *p_hwfn)1080 static int qed_cid_map_alloc(struct qed_hwfn *p_hwfn)
1081 {
1082 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1083 	u32 start_cid = 0, vf_start_cid = 0;
1084 	u32 type, vf;
1085 
1086 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1087 		struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[type];
1088 		struct qed_cid_acquired_map *p_map;
1089 
1090 		/* Handle PF maps */
1091 		p_map = &p_mngr->acquired[type];
1092 		if (qed_cid_map_alloc_single(p_hwfn, type, start_cid,
1093 					     p_cfg->cid_count, p_map))
1094 			goto cid_map_fail;
1095 
1096 		/* Handle VF maps */
1097 		for (vf = 0; vf < MAX_NUM_VFS; vf++) {
1098 			p_map = &p_mngr->acquired_vf[type][vf];
1099 			if (qed_cid_map_alloc_single(p_hwfn, type,
1100 						     vf_start_cid,
1101 						     p_cfg->cids_per_vf, p_map))
1102 				goto cid_map_fail;
1103 		}
1104 
1105 		start_cid += p_cfg->cid_count;
1106 		vf_start_cid += p_cfg->cids_per_vf;
1107 	}
1108 
1109 	return 0;
1110 
1111 cid_map_fail:
1112 	qed_cid_map_free(p_hwfn);
1113 	return -ENOMEM;
1114 }
1115 
qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn)1116 int qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn)
1117 {
1118 	struct qed_ilt_client_cfg *clients;
1119 	struct qed_cxt_mngr *p_mngr;
1120 	u32 i;
1121 
1122 	p_mngr = kzalloc(sizeof(*p_mngr), GFP_KERNEL);
1123 	if (!p_mngr)
1124 		return -ENOMEM;
1125 
1126 	/* Initialize ILT client registers */
1127 	clients = p_mngr->clients;
1128 	clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1129 	clients[ILT_CLI_CDUC].last.reg = ILT_CFG_REG(CDUC, LAST_ILT);
1130 	clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1131 
1132 	clients[ILT_CLI_QM].first.reg = ILT_CFG_REG(QM, FIRST_ILT);
1133 	clients[ILT_CLI_QM].last.reg = ILT_CFG_REG(QM, LAST_ILT);
1134 	clients[ILT_CLI_QM].p_size.reg = ILT_CFG_REG(QM, P_SIZE);
1135 
1136 	clients[ILT_CLI_TM].first.reg = ILT_CFG_REG(TM, FIRST_ILT);
1137 	clients[ILT_CLI_TM].last.reg = ILT_CFG_REG(TM, LAST_ILT);
1138 	clients[ILT_CLI_TM].p_size.reg = ILT_CFG_REG(TM, P_SIZE);
1139 
1140 	clients[ILT_CLI_SRC].first.reg = ILT_CFG_REG(SRC, FIRST_ILT);
1141 	clients[ILT_CLI_SRC].last.reg = ILT_CFG_REG(SRC, LAST_ILT);
1142 	clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1143 
1144 	clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1145 	clients[ILT_CLI_CDUT].last.reg = ILT_CFG_REG(CDUT, LAST_ILT);
1146 	clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1147 
1148 	clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1149 	clients[ILT_CLI_TSDM].last.reg = ILT_CFG_REG(TSDM, LAST_ILT);
1150 	clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1151 	/* default ILT page size for all clients is 64K */
1152 	for (i = 0; i < MAX_ILT_CLIENTS; i++)
1153 		p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE;
1154 
1155 	p_mngr->conn_ctx_size = CONN_CXT_SIZE(p_hwfn);
1156 
1157 	/* Initialize task sizes */
1158 	p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1159 	p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1160 
1161 	if (p_hwfn->cdev->p_iov_info) {
1162 		p_mngr->vf_count = p_hwfn->cdev->p_iov_info->total_vfs;
1163 		p_mngr->first_vf_in_pf =
1164 			p_hwfn->cdev->p_iov_info->first_vf_in_pf;
1165 	}
1166 	/* Initialize the dynamic ILT allocation mutex */
1167 	mutex_init(&p_mngr->mutex);
1168 
1169 	/* Set the cxt mangr pointer priori to further allocations */
1170 	p_hwfn->p_cxt_mngr = p_mngr;
1171 
1172 	return 0;
1173 }
1174 
qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn)1175 int qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn)
1176 {
1177 	int rc;
1178 
1179 	/* Allocate the ILT shadow table */
1180 	rc = qed_ilt_shadow_alloc(p_hwfn);
1181 	if (rc)
1182 		goto tables_alloc_fail;
1183 
1184 	/* Allocate the T2  table */
1185 	rc = qed_cxt_src_t2_alloc(p_hwfn);
1186 	if (rc)
1187 		goto tables_alloc_fail;
1188 
1189 	/* Allocate and initialize the acquired cids bitmaps */
1190 	rc = qed_cid_map_alloc(p_hwfn);
1191 	if (rc)
1192 		goto tables_alloc_fail;
1193 
1194 	return 0;
1195 
1196 tables_alloc_fail:
1197 	qed_cxt_mngr_free(p_hwfn);
1198 	return rc;
1199 }
1200 
qed_cxt_mngr_free(struct qed_hwfn *p_hwfn)1201 void qed_cxt_mngr_free(struct qed_hwfn *p_hwfn)
1202 {
1203 	if (!p_hwfn->p_cxt_mngr)
1204 		return;
1205 
1206 	qed_cid_map_free(p_hwfn);
1207 	qed_cxt_src_t2_free(p_hwfn);
1208 	qed_ilt_shadow_free(p_hwfn);
1209 	kfree(p_hwfn->p_cxt_mngr);
1210 
1211 	p_hwfn->p_cxt_mngr = NULL;
1212 }
1213 
qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn)1214 void qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn)
1215 {
1216 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1217 	struct qed_cid_acquired_map *p_map;
1218 	struct qed_conn_type_cfg *p_cfg;
1219 	int type;
1220 	u32 len;
1221 
1222 	/* Reset acquired cids */
1223 	for (type = 0; type < MAX_CONN_TYPES; type++) {
1224 		u32 vf;
1225 
1226 		p_cfg = &p_mngr->conn_cfg[type];
1227 		if (p_cfg->cid_count) {
1228 			p_map = &p_mngr->acquired[type];
1229 			len = DIV_ROUND_UP(p_map->max_count,
1230 					   sizeof(unsigned long) *
1231 					   BITS_PER_BYTE) *
1232 			      sizeof(unsigned long);
1233 			memset(p_map->cid_map, 0, len);
1234 		}
1235 
1236 		if (!p_cfg->cids_per_vf)
1237 			continue;
1238 
1239 		for (vf = 0; vf < MAX_NUM_VFS; vf++) {
1240 			p_map = &p_mngr->acquired_vf[type][vf];
1241 			len = DIV_ROUND_UP(p_map->max_count,
1242 					   sizeof(unsigned long) *
1243 					   BITS_PER_BYTE) *
1244 			      sizeof(unsigned long);
1245 			memset(p_map->cid_map, 0, len);
1246 		}
1247 	}
1248 }
1249 
1250 /* CDU Common */
1251 #define CDUC_CXT_SIZE_SHIFT \
1252 	CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1253 
1254 #define CDUC_CXT_SIZE_MASK \
1255 	(CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1256 
1257 #define CDUC_BLOCK_WASTE_SHIFT \
1258 	CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1259 
1260 #define CDUC_BLOCK_WASTE_MASK \
1261 	(CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1262 
1263 #define CDUC_NCIB_SHIFT	\
1264 	CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1265 
1266 #define CDUC_NCIB_MASK \
1267 	(CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1268 
1269 #define CDUT_TYPE0_CXT_SIZE_SHIFT \
1270 	CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1271 
1272 #define CDUT_TYPE0_CXT_SIZE_MASK		\
1273 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >>	\
1274 	 CDUT_TYPE0_CXT_SIZE_SHIFT)
1275 
1276 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT \
1277 	CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1278 
1279 #define CDUT_TYPE0_BLOCK_WASTE_MASK		       \
1280 	(CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >> \
1281 	 CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1282 
1283 #define CDUT_TYPE0_NCIB_SHIFT \
1284 	CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1285 
1286 #define CDUT_TYPE0_NCIB_MASK				 \
1287 	(CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >> \
1288 	 CDUT_TYPE0_NCIB_SHIFT)
1289 
1290 #define CDUT_TYPE1_CXT_SIZE_SHIFT \
1291 	CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1292 
1293 #define CDUT_TYPE1_CXT_SIZE_MASK		\
1294 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >>	\
1295 	 CDUT_TYPE1_CXT_SIZE_SHIFT)
1296 
1297 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT \
1298 	CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1299 
1300 #define CDUT_TYPE1_BLOCK_WASTE_MASK		       \
1301 	(CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >> \
1302 	 CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1303 
1304 #define CDUT_TYPE1_NCIB_SHIFT \
1305 	CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1306 
1307 #define CDUT_TYPE1_NCIB_MASK				 \
1308 	(CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >> \
1309 	 CDUT_TYPE1_NCIB_SHIFT)
1310 
qed_cdu_init_common(struct qed_hwfn *p_hwfn)1311 static void qed_cdu_init_common(struct qed_hwfn *p_hwfn)
1312 {
1313 	u32 page_sz, elems_per_page, block_waste, cxt_size, cdu_params = 0;
1314 
1315 	/* CDUC - connection configuration */
1316 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1317 	cxt_size = CONN_CXT_SIZE(p_hwfn);
1318 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1319 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1320 
1321 	SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1322 	SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1323 	SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
1324 	STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1325 
1326 	/* CDUT - type-0 tasks configuration */
1327 	page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1328 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1329 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1330 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1331 
1332 	/* cxt size and block-waste are multipes of 8 */
1333 	cdu_params = 0;
1334 	SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1335 	SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1336 	SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1337 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1338 
1339 	/* CDUT - type-1 tasks configuration */
1340 	cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1341 	elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1342 	block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1343 
1344 	/* cxt size and block-waste are multipes of 8 */
1345 	cdu_params = 0;
1346 	SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1347 	SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1348 	SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1349 	STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1350 }
1351 
1352 /* CDU PF */
1353 #define CDU_SEG_REG_TYPE_SHIFT          CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1354 #define CDU_SEG_REG_TYPE_MASK           0x1
1355 #define CDU_SEG_REG_OFFSET_SHIFT        0
1356 #define CDU_SEG_REG_OFFSET_MASK         CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1357 
qed_cdu_init_pf(struct qed_hwfn *p_hwfn)1358 static void qed_cdu_init_pf(struct qed_hwfn *p_hwfn)
1359 {
1360 	struct qed_ilt_client_cfg *p_cli;
1361 	struct qed_tid_seg *p_seg;
1362 	u32 cdu_seg_params, offset;
1363 	int i;
1364 
1365 	static const u32 rt_type_offset_arr[] = {
1366 		CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1367 		CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1368 		CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1369 		CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1370 	};
1371 
1372 	static const u32 rt_type_offset_fl_arr[] = {
1373 		CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1374 		CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1375 		CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1376 		CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1377 	};
1378 
1379 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1380 
1381 	/* There are initializations only for CDUT during pf Phase */
1382 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1383 		/* Segment 0 */
1384 		p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
1385 		if (!p_seg)
1386 			continue;
1387 
1388 		/* Note: start_line is already adjusted for the CDU
1389 		 * segment register granularity, so we just need to
1390 		 * divide. Adjustment is implicit as we assume ILT
1391 		 * Page size is larger than 32K!
1392 		 */
1393 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1394 			  (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1395 			   p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1396 
1397 		cdu_seg_params = 0;
1398 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1399 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1400 		STORE_RT_REG(p_hwfn, rt_type_offset_arr[i], cdu_seg_params);
1401 
1402 		offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1403 			  (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1404 			   p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1405 
1406 		cdu_seg_params = 0;
1407 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1408 		SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1409 		STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i], cdu_seg_params);
1410 	}
1411 }
1412 
qed_qm_init_pf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt, bool is_pf_loading)1413 void qed_qm_init_pf(struct qed_hwfn *p_hwfn,
1414 		    struct qed_ptt *p_ptt, bool is_pf_loading)
1415 {
1416 	struct qed_qm_info *qm_info = &p_hwfn->qm_info;
1417 	struct qed_qm_pf_rt_init_params params;
1418 	struct qed_qm_iids iids;
1419 
1420 	memset(&iids, 0, sizeof(iids));
1421 	qed_cxt_qm_iids(p_hwfn, &iids);
1422 
1423 	memset(&params, 0, sizeof(params));
1424 	params.port_id = p_hwfn->port_id;
1425 	params.pf_id = p_hwfn->rel_pf_id;
1426 	params.max_phys_tcs_per_port = qm_info->max_phys_tcs_per_port;
1427 	params.is_pf_loading = is_pf_loading;
1428 	params.num_pf_cids = iids.cids;
1429 	params.num_vf_cids = iids.vf_cids;
1430 	params.num_tids = iids.tids;
1431 	params.start_pq = qm_info->start_pq;
1432 	params.num_pf_pqs = qm_info->num_pqs - qm_info->num_vf_pqs;
1433 	params.num_vf_pqs = qm_info->num_vf_pqs;
1434 	params.start_vport = qm_info->start_vport;
1435 	params.num_vports = qm_info->num_vports;
1436 	params.pf_wfq = qm_info->pf_wfq;
1437 	params.pf_rl = qm_info->pf_rl;
1438 	params.pq_params = qm_info->qm_pq_params;
1439 	params.vport_params = qm_info->qm_vport_params;
1440 
1441 	qed_qm_pf_rt_init(p_hwfn, p_ptt, &params);
1442 }
1443 
1444 /* CM PF */
qed_cm_init_pf(struct qed_hwfn *p_hwfn)1445 static void qed_cm_init_pf(struct qed_hwfn *p_hwfn)
1446 {
1447 	/* XCM pure-LB queue */
1448 	STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET,
1449 		     qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1450 }
1451 
1452 /* DQ PF */
qed_dq_init_pf(struct qed_hwfn *p_hwfn)1453 static void qed_dq_init_pf(struct qed_hwfn *p_hwfn)
1454 {
1455 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1456 	u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1457 
1458 	dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1459 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1460 
1461 	dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1462 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1463 
1464 	dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1465 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1466 
1467 	dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1468 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1469 
1470 	dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1471 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1472 
1473 	dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1474 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1475 
1476 	dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1477 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1478 
1479 	dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1480 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1481 
1482 	dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1483 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1484 
1485 	dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1486 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1487 
1488 	dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1489 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1490 
1491 	dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1492 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1493 
1494 	/* Connection types 6 & 7 are not in use, yet they must be configured
1495 	 * as the highest possible connection. Not configuring them means the
1496 	 * defaults will be  used, and with a large number of cids a bug may
1497 	 * occur, if the defaults will be smaller than dq_pf_max_cid /
1498 	 * dq_vf_max_cid.
1499 	 */
1500 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1501 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1502 
1503 	STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1504 	STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1505 }
1506 
qed_ilt_bounds_init(struct qed_hwfn *p_hwfn)1507 static void qed_ilt_bounds_init(struct qed_hwfn *p_hwfn)
1508 {
1509 	struct qed_ilt_client_cfg *ilt_clients;
1510 	int i;
1511 
1512 	ilt_clients = p_hwfn->p_cxt_mngr->clients;
1513 	for_each_ilt_valid_client(i, ilt_clients) {
1514 		STORE_RT_REG(p_hwfn,
1515 			     ilt_clients[i].first.reg,
1516 			     ilt_clients[i].first.val);
1517 		STORE_RT_REG(p_hwfn,
1518 			     ilt_clients[i].last.reg, ilt_clients[i].last.val);
1519 		STORE_RT_REG(p_hwfn,
1520 			     ilt_clients[i].p_size.reg,
1521 			     ilt_clients[i].p_size.val);
1522 	}
1523 }
1524 
qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn)1525 static void qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn)
1526 {
1527 	struct qed_ilt_client_cfg *p_cli;
1528 	u32 blk_factor;
1529 
1530 	/* For simplicty  we set the 'block' to be an ILT page */
1531 	if (p_hwfn->cdev->p_iov_info) {
1532 		struct qed_hw_sriov_info *p_iov = p_hwfn->cdev->p_iov_info;
1533 
1534 		STORE_RT_REG(p_hwfn,
1535 			     PSWRQ2_REG_VF_BASE_RT_OFFSET,
1536 			     p_iov->first_vf_in_pf);
1537 		STORE_RT_REG(p_hwfn,
1538 			     PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1539 			     p_iov->first_vf_in_pf + p_iov->total_vfs);
1540 	}
1541 
1542 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1543 	blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1544 	if (p_cli->active) {
1545 		STORE_RT_REG(p_hwfn,
1546 			     PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1547 			     blk_factor);
1548 		STORE_RT_REG(p_hwfn,
1549 			     PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1550 			     p_cli->pf_total_lines);
1551 		STORE_RT_REG(p_hwfn,
1552 			     PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1553 			     p_cli->vf_total_lines);
1554 	}
1555 
1556 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1557 	blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1558 	if (p_cli->active) {
1559 		STORE_RT_REG(p_hwfn,
1560 			     PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1561 			     blk_factor);
1562 		STORE_RT_REG(p_hwfn,
1563 			     PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1564 			     p_cli->pf_total_lines);
1565 		STORE_RT_REG(p_hwfn,
1566 			     PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1567 			     p_cli->vf_total_lines);
1568 	}
1569 
1570 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1571 	blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1572 	if (p_cli->active) {
1573 		STORE_RT_REG(p_hwfn,
1574 			     PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET, blk_factor);
1575 		STORE_RT_REG(p_hwfn,
1576 			     PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1577 			     p_cli->pf_total_lines);
1578 		STORE_RT_REG(p_hwfn,
1579 			     PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1580 			     p_cli->vf_total_lines);
1581 	}
1582 }
1583 
1584 /* ILT (PSWRQ2) PF */
qed_ilt_init_pf(struct qed_hwfn *p_hwfn)1585 static void qed_ilt_init_pf(struct qed_hwfn *p_hwfn)
1586 {
1587 	struct qed_ilt_client_cfg *clients;
1588 	struct qed_cxt_mngr *p_mngr;
1589 	struct phys_mem_desc *p_shdw;
1590 	u32 line, rt_offst, i;
1591 
1592 	qed_ilt_bounds_init(p_hwfn);
1593 	qed_ilt_vf_bounds_init(p_hwfn);
1594 
1595 	p_mngr = p_hwfn->p_cxt_mngr;
1596 	p_shdw = p_mngr->ilt_shadow;
1597 	clients = p_hwfn->p_cxt_mngr->clients;
1598 
1599 	for_each_ilt_valid_client(i, clients) {
1600 		/** Client's 1st val and RT array are absolute, ILT shadows'
1601 		 *  lines are relative.
1602 		 */
1603 		line = clients[i].first.val - p_mngr->pf_start_line;
1604 		rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1605 			   clients[i].first.val * ILT_ENTRY_IN_REGS;
1606 
1607 		for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1608 		     line++, rt_offst += ILT_ENTRY_IN_REGS) {
1609 			u64 ilt_hw_entry = 0;
1610 
1611 			/** p_virt could be NULL incase of dynamic
1612 			 *  allocation
1613 			 */
1614 			if (p_shdw[line].virt_addr) {
1615 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1616 				SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1617 					  (p_shdw[line].phys_addr >> 12));
1618 
1619 				DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1620 					   "Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1621 					   rt_offst, line, i,
1622 					   (u64)(p_shdw[line].phys_addr >> 12));
1623 			}
1624 
1625 			STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1626 		}
1627 	}
1628 }
1629 
1630 /* SRC (Searcher) PF */
qed_src_init_pf(struct qed_hwfn *p_hwfn)1631 static void qed_src_init_pf(struct qed_hwfn *p_hwfn)
1632 {
1633 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1634 	u32 rounded_conn_num, conn_num, conn_max;
1635 	struct qed_src_iids src_iids;
1636 
1637 	memset(&src_iids, 0, sizeof(src_iids));
1638 	qed_cxt_src_iids(p_mngr, &src_iids);
1639 	conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1640 	if (!conn_num)
1641 		return;
1642 
1643 	conn_max = max_t(u32, conn_num, SRC_MIN_NUM_ELEMS);
1644 	rounded_conn_num = roundup_pow_of_two(conn_max);
1645 
1646 	STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1647 	STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1648 		     ilog2(rounded_conn_num));
1649 
1650 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1651 			 p_hwfn->p_cxt_mngr->src_t2.first_free);
1652 	STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1653 			 p_hwfn->p_cxt_mngr->src_t2.last_free);
1654 }
1655 
1656 /* Timers PF */
1657 #define TM_CFG_NUM_IDS_SHIFT            0
1658 #define TM_CFG_NUM_IDS_MASK             0xFFFFULL
1659 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT    16
1660 #define TM_CFG_PRE_SCAN_OFFSET_MASK     0x1FFULL
1661 #define TM_CFG_PARENT_PF_SHIFT          25
1662 #define TM_CFG_PARENT_PF_MASK           0x7ULL
1663 
1664 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT  30
1665 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK   0x1FFULL
1666 
1667 #define TM_CFG_TID_OFFSET_SHIFT         30
1668 #define TM_CFG_TID_OFFSET_MASK          0x7FFFFULL
1669 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT  49
1670 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK   0x1FFULL
1671 
qed_tm_init_pf(struct qed_hwfn *p_hwfn)1672 static void qed_tm_init_pf(struct qed_hwfn *p_hwfn)
1673 {
1674 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1675 	u32 active_seg_mask = 0, tm_offset, rt_reg;
1676 	struct qed_tm_iids tm_iids;
1677 	u64 cfg_word;
1678 	u8 i;
1679 
1680 	memset(&tm_iids, 0, sizeof(tm_iids));
1681 	qed_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
1682 
1683 	/* @@@TBD No pre-scan for now */
1684 
1685 	/* Note: We assume consecutive VFs for a PF */
1686 	for (i = 0; i < p_mngr->vf_count; i++) {
1687 		cfg_word = 0;
1688 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1689 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1690 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1691 		SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0);
1692 		rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1693 		    (sizeof(cfg_word) / sizeof(u32)) *
1694 		    (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1695 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1696 	}
1697 
1698 	cfg_word = 0;
1699 	SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1700 	SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1701 	SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);	/* n/a for PF */
1702 	SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0);	/* scan all   */
1703 
1704 	rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1705 	    (sizeof(cfg_word) / sizeof(u32)) *
1706 	    (NUM_OF_VFS(p_hwfn->cdev) + p_hwfn->rel_pf_id);
1707 	STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1708 
1709 	/* enale scan */
1710 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1711 		     tm_iids.pf_cids ? 0x1 : 0x0);
1712 
1713 	/* @@@TBD how to enable the scan for the VFs */
1714 
1715 	tm_offset = tm_iids.per_vf_cids;
1716 
1717 	/* Note: We assume consecutive VFs for a PF */
1718 	for (i = 0; i < p_mngr->vf_count; i++) {
1719 		cfg_word = 0;
1720 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1721 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1722 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1723 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1724 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1725 
1726 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1727 		    (sizeof(cfg_word) / sizeof(u32)) *
1728 		    (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1729 
1730 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1731 	}
1732 
1733 	tm_offset = tm_iids.pf_cids;
1734 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1735 		cfg_word = 0;
1736 		SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1737 		SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1738 		SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1739 		SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1740 		SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1741 
1742 		rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1743 		    (sizeof(cfg_word) / sizeof(u32)) *
1744 		    (NUM_OF_VFS(p_hwfn->cdev) +
1745 		     p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1746 
1747 		STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1748 		active_seg_mask |= (tm_iids.pf_tids[i] ? BIT(i) : 0);
1749 
1750 		tm_offset += tm_iids.pf_tids[i];
1751 	}
1752 
1753 	if (QED_IS_RDMA_PERSONALITY(p_hwfn))
1754 		active_seg_mask = 0;
1755 
1756 	STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1757 
1758 	/* @@@TBD how to enable the scan for the VFs */
1759 }
1760 
qed_prs_init_common(struct qed_hwfn *p_hwfn)1761 static void qed_prs_init_common(struct qed_hwfn *p_hwfn)
1762 {
1763 	if ((p_hwfn->hw_info.personality == QED_PCI_FCOE) &&
1764 	    p_hwfn->pf_params.fcoe_pf_params.is_target)
1765 		STORE_RT_REG(p_hwfn,
1766 			     PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1767 }
1768 
qed_prs_init_pf(struct qed_hwfn *p_hwfn)1769 static void qed_prs_init_pf(struct qed_hwfn *p_hwfn)
1770 {
1771 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1772 	struct qed_conn_type_cfg *p_fcoe;
1773 	struct qed_tid_seg *p_tid;
1774 
1775 	p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1776 
1777 	/* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1778 	if (!p_fcoe->cid_count)
1779 		return;
1780 
1781 	p_tid = &p_fcoe->tid_seg[QED_CXT_FCOE_TID_SEG];
1782 	if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1783 		STORE_RT_REG_AGG(p_hwfn,
1784 				 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1785 				 p_tid->count);
1786 	} else {
1787 		STORE_RT_REG_AGG(p_hwfn,
1788 				 PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1789 				 p_tid->count);
1790 	}
1791 }
1792 
qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn)1793 void qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn)
1794 {
1795 	qed_cdu_init_common(p_hwfn);
1796 	qed_prs_init_common(p_hwfn);
1797 }
1798 
qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)1799 void qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
1800 {
1801 	qed_qm_init_pf(p_hwfn, p_ptt, true);
1802 	qed_cm_init_pf(p_hwfn);
1803 	qed_dq_init_pf(p_hwfn);
1804 	qed_cdu_init_pf(p_hwfn);
1805 	qed_ilt_init_pf(p_hwfn);
1806 	qed_src_init_pf(p_hwfn);
1807 	qed_tm_init_pf(p_hwfn);
1808 	qed_prs_init_pf(p_hwfn);
1809 }
1810 
_qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn, enum protocol_type type, u32 *p_cid, u8 vfid)1811 int _qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1812 			 enum protocol_type type, u32 *p_cid, u8 vfid)
1813 {
1814 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1815 	struct qed_cid_acquired_map *p_map;
1816 	u32 rel_cid;
1817 
1818 	if (type >= MAX_CONN_TYPES) {
1819 		DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1820 		return -EINVAL;
1821 	}
1822 
1823 	if (vfid >= MAX_NUM_VFS && vfid != QED_CXT_PF_CID) {
1824 		DP_NOTICE(p_hwfn, "VF [%02x] is out of range\n", vfid);
1825 		return -EINVAL;
1826 	}
1827 
1828 	/* Determine the right map to take this CID from */
1829 	if (vfid == QED_CXT_PF_CID)
1830 		p_map = &p_mngr->acquired[type];
1831 	else
1832 		p_map = &p_mngr->acquired_vf[type][vfid];
1833 
1834 	if (!p_map->cid_map) {
1835 		DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1836 		return -EINVAL;
1837 	}
1838 
1839 	rel_cid = find_first_zero_bit(p_map->cid_map, p_map->max_count);
1840 
1841 	if (rel_cid >= p_map->max_count) {
1842 		DP_NOTICE(p_hwfn, "no CID available for protocol %d\n", type);
1843 		return -EINVAL;
1844 	}
1845 
1846 	__set_bit(rel_cid, p_map->cid_map);
1847 
1848 	*p_cid = rel_cid + p_map->start_cid;
1849 
1850 	DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1851 		   "Acquired cid 0x%08x [rel. %08x] vfid %02x type %d\n",
1852 		   *p_cid, rel_cid, vfid, type);
1853 
1854 	return 0;
1855 }
1856 
qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn, enum protocol_type type, u32 *p_cid)1857 int qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1858 			enum protocol_type type, u32 *p_cid)
1859 {
1860 	return _qed_cxt_acquire_cid(p_hwfn, type, p_cid, QED_CXT_PF_CID);
1861 }
1862 
qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn, u32 cid, u8 vfid, enum protocol_type *p_type, struct qed_cid_acquired_map **pp_map)1863 static bool qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn,
1864 				      u32 cid,
1865 				      u8 vfid,
1866 				      enum protocol_type *p_type,
1867 				      struct qed_cid_acquired_map **pp_map)
1868 {
1869 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1870 	u32 rel_cid;
1871 
1872 	/* Iterate over protocols and find matching cid range */
1873 	for (*p_type = 0; *p_type < MAX_CONN_TYPES; (*p_type)++) {
1874 		if (vfid == QED_CXT_PF_CID)
1875 			*pp_map = &p_mngr->acquired[*p_type];
1876 		else
1877 			*pp_map = &p_mngr->acquired_vf[*p_type][vfid];
1878 
1879 		if (!((*pp_map)->cid_map))
1880 			continue;
1881 		if (cid >= (*pp_map)->start_cid &&
1882 		    cid < (*pp_map)->start_cid + (*pp_map)->max_count)
1883 			break;
1884 	}
1885 
1886 	if (*p_type == MAX_CONN_TYPES) {
1887 		DP_NOTICE(p_hwfn, "Invalid CID %d vfid %02x", cid, vfid);
1888 		goto fail;
1889 	}
1890 
1891 	rel_cid = cid - (*pp_map)->start_cid;
1892 	if (!test_bit(rel_cid, (*pp_map)->cid_map)) {
1893 		DP_NOTICE(p_hwfn, "CID %d [vifd %02x] not acquired",
1894 			  cid, vfid);
1895 		goto fail;
1896 	}
1897 
1898 	return true;
1899 fail:
1900 	*p_type = MAX_CONN_TYPES;
1901 	*pp_map = NULL;
1902 	return false;
1903 }
1904 
_qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid, u8 vfid)1905 void _qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid, u8 vfid)
1906 {
1907 	struct qed_cid_acquired_map *p_map = NULL;
1908 	enum protocol_type type;
1909 	bool b_acquired;
1910 	u32 rel_cid;
1911 
1912 	if (vfid != QED_CXT_PF_CID && vfid > MAX_NUM_VFS) {
1913 		DP_NOTICE(p_hwfn,
1914 			  "Trying to return incorrect CID belonging to VF %02x\n",
1915 			  vfid);
1916 		return;
1917 	}
1918 
1919 	/* Test acquired and find matching per-protocol map */
1920 	b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid, vfid,
1921 					       &type, &p_map);
1922 
1923 	if (!b_acquired)
1924 		return;
1925 
1926 	rel_cid = cid - p_map->start_cid;
1927 	clear_bit(rel_cid, p_map->cid_map);
1928 
1929 	DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1930 		   "Released CID 0x%08x [rel. %08x] vfid %02x type %d\n",
1931 		   cid, rel_cid, vfid, type);
1932 }
1933 
qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid)1934 void qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid)
1935 {
1936 	_qed_cxt_release_cid(p_hwfn, cid, QED_CXT_PF_CID);
1937 }
1938 
qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn, struct qed_cxt_info *p_info)1939 int qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn, struct qed_cxt_info *p_info)
1940 {
1941 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1942 	struct qed_cid_acquired_map *p_map = NULL;
1943 	u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
1944 	enum protocol_type type;
1945 	bool b_acquired;
1946 
1947 	/* Test acquired and find matching per-protocol map */
1948 	b_acquired = qed_cxt_test_cid_acquired(p_hwfn, p_info->iid,
1949 					       QED_CXT_PF_CID, &type, &p_map);
1950 
1951 	if (!b_acquired)
1952 		return -EINVAL;
1953 
1954 	/* set the protocl type */
1955 	p_info->type = type;
1956 
1957 	/* compute context virtual pointer */
1958 	hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1959 
1960 	conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
1961 	cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
1962 	line = p_info->iid / cxts_per_p;
1963 
1964 	/* Make sure context is allocated (dynamic allocation) */
1965 	if (!p_mngr->ilt_shadow[line].virt_addr)
1966 		return -EINVAL;
1967 
1968 	p_info->p_cxt = p_mngr->ilt_shadow[line].virt_addr +
1969 			p_info->iid % cxts_per_p * conn_cxt_size;
1970 
1971 	DP_VERBOSE(p_hwfn, (QED_MSG_ILT | QED_MSG_CXT),
1972 		   "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
1973 		   p_info->iid / cxts_per_p, p_info->p_cxt, p_info->iid);
1974 
1975 	return 0;
1976 }
1977 
qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn, struct qed_rdma_pf_params *p_params, u32 num_tasks)1978 static void qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn,
1979 				   struct qed_rdma_pf_params *p_params,
1980 				   u32 num_tasks)
1981 {
1982 	u32 num_cons, num_qps;
1983 	enum protocol_type proto;
1984 
1985 	if (p_hwfn->mcp_info->func_info.protocol == QED_PCI_ETH_RDMA) {
1986 		DP_VERBOSE(p_hwfn, QED_MSG_SP,
1987 			   "Current day drivers don't support RoCE & iWARP simultaneously on the same PF. Default to RoCE-only\n");
1988 		p_hwfn->hw_info.personality = QED_PCI_ETH_ROCE;
1989 	}
1990 
1991 	switch (p_hwfn->hw_info.personality) {
1992 	case QED_PCI_ETH_IWARP:
1993 		/* Each QP requires one connection */
1994 		num_cons = min_t(u32, IWARP_MAX_QPS, p_params->num_qps);
1995 		proto = PROTOCOLID_IWARP;
1996 		break;
1997 	case QED_PCI_ETH_ROCE:
1998 		num_qps = min_t(u32, ROCE_MAX_QPS, p_params->num_qps);
1999 		num_cons = num_qps * 2;	/* each QP requires two connections */
2000 		proto = PROTOCOLID_ROCE;
2001 		break;
2002 	default:
2003 		return;
2004 	}
2005 
2006 	if (num_cons && num_tasks) {
2007 		u32 num_srqs, num_xrc_srqs;
2008 
2009 		qed_cxt_set_proto_cid_count(p_hwfn, proto, num_cons, 0);
2010 
2011 		/* Deliberatly passing ROCE for tasks id. This is because
2012 		 * iWARP / RoCE share the task id.
2013 		 */
2014 		qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
2015 					    QED_CXT_ROCE_TID_SEG, 1,
2016 					    num_tasks, false);
2017 
2018 		num_srqs = min_t(u32, QED_RDMA_MAX_SRQS, p_params->num_srqs);
2019 
2020 		/* XRC SRQs populate a single ILT page */
2021 		num_xrc_srqs = qed_cxt_xrc_srqs_per_page(p_hwfn);
2022 
2023 		qed_cxt_set_srq_count(p_hwfn, num_srqs, num_xrc_srqs);
2024 	} else {
2025 		DP_INFO(p_hwfn->cdev,
2026 			"RDMA personality used without setting params!\n");
2027 	}
2028 }
2029 
qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn, u32 rdma_tasks)2030 int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn, u32 rdma_tasks)
2031 {
2032 	/* Set the number of required CORE connections */
2033 	u32 core_cids = 1; /* SPQ */
2034 
2035 	if (p_hwfn->using_ll2)
2036 		core_cids += 4;
2037 	qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
2038 
2039 	switch (p_hwfn->hw_info.personality) {
2040 	case QED_PCI_ETH_RDMA:
2041 	case QED_PCI_ETH_IWARP:
2042 	case QED_PCI_ETH_ROCE:
2043 	{
2044 			qed_rdma_set_pf_params(p_hwfn,
2045 					       &p_hwfn->
2046 					       pf_params.rdma_pf_params,
2047 					       rdma_tasks);
2048 		/* no need for break since RoCE coexist with Ethernet */
2049 	}
2050 		fallthrough;
2051 	case QED_PCI_ETH:
2052 	{
2053 		struct qed_eth_pf_params *p_params =
2054 		    &p_hwfn->pf_params.eth_pf_params;
2055 
2056 		if (!p_params->num_vf_cons)
2057 			p_params->num_vf_cons =
2058 			    ETH_PF_PARAMS_VF_CONS_DEFAULT;
2059 		qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
2060 					    p_params->num_cons,
2061 					    p_params->num_vf_cons);
2062 		p_hwfn->p_cxt_mngr->arfs_count = p_params->num_arfs_filters;
2063 		break;
2064 	}
2065 	case QED_PCI_FCOE:
2066 	{
2067 		struct qed_fcoe_pf_params *p_params;
2068 
2069 		p_params = &p_hwfn->pf_params.fcoe_pf_params;
2070 
2071 		if (p_params->num_cons && p_params->num_tasks) {
2072 			qed_cxt_set_proto_cid_count(p_hwfn,
2073 						    PROTOCOLID_FCOE,
2074 						    p_params->num_cons,
2075 						    0);
2076 
2077 			qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
2078 						    QED_CXT_FCOE_TID_SEG, 0,
2079 						    p_params->num_tasks, true);
2080 		} else {
2081 			DP_INFO(p_hwfn->cdev,
2082 				"Fcoe personality used without setting params!\n");
2083 		}
2084 		break;
2085 	}
2086 	case QED_PCI_ISCSI:
2087 	{
2088 		struct qed_iscsi_pf_params *p_params;
2089 
2090 		p_params = &p_hwfn->pf_params.iscsi_pf_params;
2091 
2092 		if (p_params->num_cons && p_params->num_tasks) {
2093 			qed_cxt_set_proto_cid_count(p_hwfn,
2094 						    PROTOCOLID_ISCSI,
2095 						    p_params->num_cons,
2096 						    0);
2097 
2098 			qed_cxt_set_proto_tid_count(p_hwfn,
2099 						    PROTOCOLID_ISCSI,
2100 						    QED_CXT_ISCSI_TID_SEG,
2101 						    0,
2102 						    p_params->num_tasks,
2103 						    true);
2104 		} else {
2105 			DP_INFO(p_hwfn->cdev,
2106 				"Iscsi personality used without setting params!\n");
2107 		}
2108 		break;
2109 	}
2110 	default:
2111 		return -EINVAL;
2112 	}
2113 
2114 	return 0;
2115 }
2116 
qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn, struct qed_tid_mem *p_info)2117 int qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn,
2118 			     struct qed_tid_mem *p_info)
2119 {
2120 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2121 	u32 proto, seg, total_lines, i, shadow_line;
2122 	struct qed_ilt_client_cfg *p_cli;
2123 	struct qed_ilt_cli_blk *p_fl_seg;
2124 	struct qed_tid_seg *p_seg_info;
2125 
2126 	/* Verify the personality */
2127 	switch (p_hwfn->hw_info.personality) {
2128 	case QED_PCI_FCOE:
2129 		proto = PROTOCOLID_FCOE;
2130 		seg = QED_CXT_FCOE_TID_SEG;
2131 		break;
2132 	case QED_PCI_ISCSI:
2133 		proto = PROTOCOLID_ISCSI;
2134 		seg = QED_CXT_ISCSI_TID_SEG;
2135 		break;
2136 	default:
2137 		return -EINVAL;
2138 	}
2139 
2140 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2141 	if (!p_cli->active)
2142 		return -EINVAL;
2143 
2144 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2145 	if (!p_seg_info->has_fl_mem)
2146 		return -EINVAL;
2147 
2148 	p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2149 	total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2150 				   p_fl_seg->real_size_in_page);
2151 
2152 	for (i = 0; i < total_lines; i++) {
2153 		shadow_line = i + p_fl_seg->start_line -
2154 		    p_hwfn->p_cxt_mngr->pf_start_line;
2155 		p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].virt_addr;
2156 	}
2157 	p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2158 	    p_fl_seg->real_size_in_page;
2159 	p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2160 	p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2161 	    p_info->tid_size;
2162 
2163 	return 0;
2164 }
2165 
2166 /* This function is very RoCE oriented, if another protocol in the future
2167  * will want this feature we'll need to modify the function to be more generic
2168  */
2169 int
qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn, enum qed_cxt_elem_type elem_type, u32 iid)2170 qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
2171 			  enum qed_cxt_elem_type elem_type, u32 iid)
2172 {
2173 	u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2174 	struct tdif_task_context *tdif_context;
2175 	struct qed_ilt_client_cfg *p_cli;
2176 	struct qed_ilt_cli_blk *p_blk;
2177 	struct qed_ptt *p_ptt;
2178 	dma_addr_t p_phys;
2179 	u64 ilt_hw_entry;
2180 	void *p_virt;
2181 	u32 flags1;
2182 	int rc = 0;
2183 
2184 	switch (elem_type) {
2185 	case QED_ELEM_CXT:
2186 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2187 		elem_size = CONN_CXT_SIZE(p_hwfn);
2188 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2189 		break;
2190 	case QED_ELEM_SRQ:
2191 		/* The first ILT page is not used for regular SRQs. Skip it. */
2192 		iid += p_hwfn->p_cxt_mngr->xrc_srq_count;
2193 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2194 		elem_size = SRQ_CXT_SIZE;
2195 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2196 		break;
2197 	case QED_ELEM_XRC_SRQ:
2198 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2199 		elem_size = XRC_SRQ_CXT_SIZE;
2200 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2201 		break;
2202 	case QED_ELEM_TASK:
2203 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2204 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2205 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2206 		break;
2207 	default:
2208 		DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2209 		return -EINVAL;
2210 	}
2211 
2212 	/* Calculate line in ilt */
2213 	hw_p_size = p_cli->p_size.val;
2214 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2215 	line = p_blk->start_line + (iid / elems_per_p);
2216 	shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2217 
2218 	/* If line is already allocated, do nothing, otherwise allocate it and
2219 	 * write it to the PSWRQ2 registers.
2220 	 * This section can be run in parallel from different contexts and thus
2221 	 * a mutex protection is needed.
2222 	 */
2223 
2224 	mutex_lock(&p_hwfn->p_cxt_mngr->mutex);
2225 
2226 	if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].virt_addr)
2227 		goto out0;
2228 
2229 	p_ptt = qed_ptt_acquire(p_hwfn);
2230 	if (!p_ptt) {
2231 		DP_NOTICE(p_hwfn,
2232 			  "QED_TIME_OUT on ptt acquire - dynamic allocation");
2233 		rc = -EBUSY;
2234 		goto out0;
2235 	}
2236 
2237 	p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2238 				    p_blk->real_size_in_page, &p_phys,
2239 				    GFP_KERNEL);
2240 	if (!p_virt) {
2241 		rc = -ENOMEM;
2242 		goto out1;
2243 	}
2244 
2245 	/* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2246 	 * to compensate for a HW bug, but it is configured even if DIF is not
2247 	 * enabled. This is harmless and allows us to avoid a dedicated API. We
2248 	 * configure the field for all of the contexts on the newly allocated
2249 	 * page.
2250 	 */
2251 	if (elem_type == QED_ELEM_TASK) {
2252 		u32 elem_i;
2253 		u8 *elem_start = (u8 *)p_virt;
2254 		union type1_task_context *elem;
2255 
2256 		for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2257 			elem = (union type1_task_context *)elem_start;
2258 			tdif_context = &elem->roce_ctx.tdif_context;
2259 
2260 			flags1 = le32_to_cpu(tdif_context->flags1);
2261 			SET_FIELD(flags1, TDIF_TASK_CONTEXT_REF_TAG_MASK, 0xf);
2262 			tdif_context->flags1 = cpu_to_le32(flags1);
2263 
2264 			elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2265 		}
2266 	}
2267 
2268 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].virt_addr = p_virt;
2269 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].phys_addr = p_phys;
2270 	p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2271 	    p_blk->real_size_in_page;
2272 
2273 	/* compute absolute offset */
2274 	reg_offset = PSWRQ2_REG_ILT_MEMORY +
2275 	    (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2276 
2277 	ilt_hw_entry = 0;
2278 	SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2279 	SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
2280 		  (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].phys_addr
2281 		   >> 12));
2282 
2283 	/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2284 	qed_dmae_host2grc(p_hwfn, p_ptt, (u64) (uintptr_t)&ilt_hw_entry,
2285 			  reg_offset, sizeof(ilt_hw_entry) / sizeof(u32),
2286 			  NULL);
2287 
2288 	if (elem_type == QED_ELEM_CXT) {
2289 		u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2290 		    elems_per_p;
2291 
2292 		/* Update the relevant register in the parser */
2293 		qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2294 		       last_cid_allocated - 1);
2295 
2296 		if (!p_hwfn->b_rdma_enabled_in_prs) {
2297 			/* Enable RDMA search */
2298 			qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2299 			p_hwfn->b_rdma_enabled_in_prs = true;
2300 		}
2301 	}
2302 
2303 out1:
2304 	qed_ptt_release(p_hwfn, p_ptt);
2305 out0:
2306 	mutex_unlock(&p_hwfn->p_cxt_mngr->mutex);
2307 
2308 	return rc;
2309 }
2310 
2311 /* This function is very RoCE oriented, if another protocol in the future
2312  * will want this feature we'll need to modify the function to be more generic
2313  */
2314 static int
qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn, enum qed_cxt_elem_type elem_type, u32 start_iid, u32 count)2315 qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn,
2316 		       enum qed_cxt_elem_type elem_type,
2317 		       u32 start_iid, u32 count)
2318 {
2319 	u32 start_line, end_line, shadow_start_line, shadow_end_line;
2320 	u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2321 	struct qed_ilt_client_cfg *p_cli;
2322 	struct qed_ilt_cli_blk *p_blk;
2323 	u32 end_iid = start_iid + count;
2324 	struct qed_ptt *p_ptt;
2325 	u64 ilt_hw_entry = 0;
2326 	u32 i;
2327 
2328 	switch (elem_type) {
2329 	case QED_ELEM_CXT:
2330 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2331 		elem_size = CONN_CXT_SIZE(p_hwfn);
2332 		p_blk = &p_cli->pf_blks[CDUC_BLK];
2333 		break;
2334 	case QED_ELEM_SRQ:
2335 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2336 		elem_size = SRQ_CXT_SIZE;
2337 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2338 		break;
2339 	case QED_ELEM_XRC_SRQ:
2340 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2341 		elem_size = XRC_SRQ_CXT_SIZE;
2342 		p_blk = &p_cli->pf_blks[SRQ_BLK];
2343 		break;
2344 	case QED_ELEM_TASK:
2345 		p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2346 		elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2347 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2348 		break;
2349 	default:
2350 		DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2351 		return -EINVAL;
2352 	}
2353 
2354 	/* Calculate line in ilt */
2355 	hw_p_size = p_cli->p_size.val;
2356 	elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2357 	start_line = p_blk->start_line + (start_iid / elems_per_p);
2358 	end_line = p_blk->start_line + (end_iid / elems_per_p);
2359 	if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2360 		end_line--;
2361 
2362 	shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2363 	shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2364 
2365 	p_ptt = qed_ptt_acquire(p_hwfn);
2366 	if (!p_ptt) {
2367 		DP_NOTICE(p_hwfn,
2368 			  "QED_TIME_OUT on ptt acquire - dynamic allocation");
2369 		return -EBUSY;
2370 	}
2371 
2372 	for (i = shadow_start_line; i < shadow_end_line; i++) {
2373 		if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr)
2374 			continue;
2375 
2376 		dma_free_coherent(&p_hwfn->cdev->pdev->dev,
2377 				  p_hwfn->p_cxt_mngr->ilt_shadow[i].size,
2378 				  p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr,
2379 				  p_hwfn->p_cxt_mngr->ilt_shadow[i].phys_addr);
2380 
2381 		p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr = NULL;
2382 		p_hwfn->p_cxt_mngr->ilt_shadow[i].phys_addr = 0;
2383 		p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2384 
2385 		/* compute absolute offset */
2386 		reg_offset = PSWRQ2_REG_ILT_MEMORY +
2387 		    ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2388 		     ILT_ENTRY_IN_REGS);
2389 
2390 		/* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2391 		 * wide-bus.
2392 		 */
2393 		qed_dmae_host2grc(p_hwfn, p_ptt,
2394 				  (u64) (uintptr_t) &ilt_hw_entry,
2395 				  reg_offset,
2396 				  sizeof(ilt_hw_entry) / sizeof(u32),
2397 				  NULL);
2398 	}
2399 
2400 	qed_ptt_release(p_hwfn, p_ptt);
2401 
2402 	return 0;
2403 }
2404 
qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto)2405 int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto)
2406 {
2407 	int rc;
2408 	u32 cid;
2409 
2410 	/* Free Connection CXT */
2411 	rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_CXT,
2412 				    qed_cxt_get_proto_cid_start(p_hwfn,
2413 								proto),
2414 				    qed_cxt_get_proto_cid_count(p_hwfn,
2415 								proto, &cid));
2416 
2417 	if (rc)
2418 		return rc;
2419 
2420 	/* Free Task CXT ( Intentionally RoCE as task-id is shared between
2421 	 * RoCE and iWARP )
2422 	 */
2423 	proto = PROTOCOLID_ROCE;
2424 	rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_TASK, 0,
2425 				    qed_cxt_get_proto_tid_count(p_hwfn, proto));
2426 	if (rc)
2427 		return rc;
2428 
2429 	/* Free TSDM CXT */
2430 	rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_XRC_SRQ, 0,
2431 				    p_hwfn->p_cxt_mngr->xrc_srq_count);
2432 
2433 	rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_SRQ,
2434 				    p_hwfn->p_cxt_mngr->xrc_srq_count,
2435 				    p_hwfn->p_cxt_mngr->srq_count);
2436 
2437 	return rc;
2438 }
2439 
qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn, u32 tid, u8 ctx_type, void **pp_task_ctx)2440 int qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn,
2441 			 u32 tid, u8 ctx_type, void **pp_task_ctx)
2442 {
2443 	struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2444 	struct qed_ilt_client_cfg *p_cli;
2445 	struct qed_tid_seg *p_seg_info;
2446 	struct qed_ilt_cli_blk *p_seg;
2447 	u32 num_tids_per_block;
2448 	u32 tid_size, ilt_idx;
2449 	u32 total_lines;
2450 	u32 proto, seg;
2451 
2452 	/* Verify the personality */
2453 	switch (p_hwfn->hw_info.personality) {
2454 	case QED_PCI_FCOE:
2455 		proto = PROTOCOLID_FCOE;
2456 		seg = QED_CXT_FCOE_TID_SEG;
2457 		break;
2458 	case QED_PCI_ISCSI:
2459 		proto = PROTOCOLID_ISCSI;
2460 		seg = QED_CXT_ISCSI_TID_SEG;
2461 		break;
2462 	default:
2463 		return -EINVAL;
2464 	}
2465 
2466 	p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2467 	if (!p_cli->active)
2468 		return -EINVAL;
2469 
2470 	p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2471 
2472 	if (ctx_type == QED_CTX_WORKING_MEM) {
2473 		p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2474 	} else if (ctx_type == QED_CTX_FL_MEM) {
2475 		if (!p_seg_info->has_fl_mem)
2476 			return -EINVAL;
2477 		p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2478 	} else {
2479 		return -EINVAL;
2480 	}
2481 	total_lines = DIV_ROUND_UP(p_seg->total_size, p_seg->real_size_in_page);
2482 	tid_size = p_mngr->task_type_size[p_seg_info->type];
2483 	num_tids_per_block = p_seg->real_size_in_page / tid_size;
2484 
2485 	if (total_lines < tid / num_tids_per_block)
2486 		return -EINVAL;
2487 
2488 	ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2489 		  p_mngr->pf_start_line;
2490 	*pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].virt_addr +
2491 		       (tid % num_tids_per_block) * tid_size;
2492 
2493 	return 0;
2494 }
2495 
qed_blk_calculate_pages(struct qed_ilt_cli_blk *p_blk)2496 static u16 qed_blk_calculate_pages(struct qed_ilt_cli_blk *p_blk)
2497 {
2498 	if (p_blk->real_size_in_page == 0)
2499 		return 0;
2500 
2501 	return DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
2502 }
2503 
qed_get_cdut_num_pf_init_pages(struct qed_hwfn *p_hwfn)2504 u16 qed_get_cdut_num_pf_init_pages(struct qed_hwfn *p_hwfn)
2505 {
2506 	struct qed_ilt_client_cfg *p_cli;
2507 	struct qed_ilt_cli_blk *p_blk;
2508 	u16 i, pages = 0;
2509 
2510 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2511 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
2512 		p_blk = &p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)];
2513 		pages += qed_blk_calculate_pages(p_blk);
2514 	}
2515 
2516 	return pages;
2517 }
2518 
qed_get_cdut_num_vf_init_pages(struct qed_hwfn *p_hwfn)2519 u16 qed_get_cdut_num_vf_init_pages(struct qed_hwfn *p_hwfn)
2520 {
2521 	struct qed_ilt_client_cfg *p_cli;
2522 	struct qed_ilt_cli_blk *p_blk;
2523 	u16 i, pages = 0;
2524 
2525 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2526 	for (i = 0; i < NUM_TASK_VF_SEGMENTS; i++) {
2527 		p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(i, VF)];
2528 		pages += qed_blk_calculate_pages(p_blk);
2529 	}
2530 
2531 	return pages;
2532 }
2533 
qed_get_cdut_num_pf_work_pages(struct qed_hwfn *p_hwfn)2534 u16 qed_get_cdut_num_pf_work_pages(struct qed_hwfn *p_hwfn)
2535 {
2536 	struct qed_ilt_client_cfg *p_cli;
2537 	struct qed_ilt_cli_blk *p_blk;
2538 	u16 i, pages = 0;
2539 
2540 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2541 	for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
2542 		p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(i)];
2543 		pages += qed_blk_calculate_pages(p_blk);
2544 	}
2545 
2546 	return pages;
2547 }
2548 
qed_get_cdut_num_vf_work_pages(struct qed_hwfn *p_hwfn)2549 u16 qed_get_cdut_num_vf_work_pages(struct qed_hwfn *p_hwfn)
2550 {
2551 	struct qed_ilt_client_cfg *p_cli;
2552 	struct qed_ilt_cli_blk *p_blk;
2553 	u16 pages = 0, i;
2554 
2555 	p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2556 	for (i = 0; i < NUM_TASK_VF_SEGMENTS; i++) {
2557 		p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(i)];
2558 		pages += qed_blk_calculate_pages(p_blk);
2559 	}
2560 
2561 	return pages;
2562 }
2563