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 */
56union conn_context {
57 struct core_conn_context core_ctx;
58 struct eth_conn_context eth_ctx;
59 struct iscsi_conn_context iscsi_ctx;
60 struct fcoe_conn_context fcoe_ctx;
61 struct roce_conn_context roce_ctx;
62};
63
64/* TYPE-0 task context - iSCSI, FCOE */
65union type0_task_context {
66 struct iscsi_task_context iscsi_ctx;
67 struct fcoe_task_context fcoe_ctx;
68};
69
70/* TYPE-1 task context - ROCE */
71union type1_task_context {
72 struct rdma_task_context roce_ctx;
73};
74
75struct 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
95static bool src_proto(enum protocol_type type)
96{
97 return type == PROTOCOLID_TCP_ULP ||
98 type == PROTOCOLID_FCOE ||
99 type == PROTOCOLID_IWARP;
100}
101
102static bool tm_cid_proto(enum protocol_type type)
103{
104 return type == PROTOCOLID_TCP_ULP ||
105 type == PROTOCOLID_FCOE ||
106 type == PROTOCOLID_ROCE ||
107 type == PROTOCOLID_IWARP;
108}
109
110static 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 */
116struct qed_cdu_iids {
117 u32 pf_cids;
118 u32 per_vf_cids;
119};
120
121static 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 */
133struct qed_src_iids {
134 u32 pf_cids;
135 u32 per_vf_cids;
136};
137
138static 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(type: 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 */
156struct 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
164static 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(type: 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(type: 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(type: 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
222static 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
256static 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
271static 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
280u32 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
289static 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_client: ILT_CLI_TSDM);
294 return page_size / XRC_SRQ_CXT_SIZE;
295}
296
297u32 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 */
308static 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
328u32 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
337u32 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
343u32 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
355static 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
368static 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
385static 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
407static 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
427static 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
436static 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
442static 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
456int 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_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, iids: &cdu_iids);
494
495 p_blk = qed_cxt_set_blk(p_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, start_line: curr_line,
500 total_size: total, CONN_CXT_SIZE(p_hwfn));
501
502 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, client_id: 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_client: ILT_CLI_CDUC);
507
508 /* CDUC VF */
509 p_blk = qed_cxt_set_blk(p_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, start_line: curr_line,
513 total_size: total, CONN_CXT_SIZE(p_hwfn));
514
515 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, client_id: 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, p_line: &curr_line,
520 client_id: ILT_CLI_CDUC);
521
522 /* CDUT PF */
523 p_cli = qed_cxt_set_cli(p_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, seg: i);
529 if (!p_seg || p_seg->count == 0)
530 continue;
531
532 p_blk = qed_cxt_set_blk(p_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, start_line: curr_line, total_size: total,
535 elem_size: p_mngr->task_type_size[p_seg->type]);
536
537 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
538 client_id: 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, seg: i);
544 if (!p_seg || p_seg->count == 0)
545 continue;
546
547 p_blk =
548 qed_cxt_set_blk(p_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, start_line: line, total_size: 0, elem_size: 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 start_line: curr_line, total_size: total,
576 elem_size: p_mngr->task_type_size[p_seg->type]);
577
578 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
579 client_id: 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_blk: &p_cli->vf_blks[CDUT_SEG_BLK(0)]);
594 qed_ilt_cli_blk_fill(p_cli, p_blk,
595 start_line: curr_line, total_size: total,
596 elem_size: p_mngr->task_type_size[p_seg->type]);
597
598 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
599 client_id: ILT_CLI_CDUT);
600
601 /* 'init' memory */
602 p_blk =
603 qed_cxt_set_blk(p_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, start_line: line, total_size: 0, elem_size: 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 start_line: curr_line, total_size: total, elem_size: task_size);
612 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
613 client_id: 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, p_line: &curr_line,
622 client_id: 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, p_line: &curr_line,
626 client_id: ILT_CLI_CDUT);
627 }
628 }
629
630 /* QM */
631 p_cli = qed_cxt_set_cli(p_cli: &p_mngr->clients[ILT_CLI_QM]);
632 p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[0]);
633
634 qed_cxt_qm_iids(p_hwfn, iids: &qm_iids);
635 total = qed_qm_pf_mem_size(num_pf_cids: qm_iids.cids,
636 num_vf_cids: qm_iids.vf_cids, num_tids: qm_iids.tids,
637 num_pf_pqs: p_hwfn->qm_info.num_pqs,
638 num_vf_pqs: 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 start_line: curr_line, total_size: total * 0x1000,
650 QM_PQ_ELEMENT_SIZE);
651
652 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, client_id: 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_cli: &p_mngr->clients[ILT_CLI_SRC]);
657 qed_cxt_src_iids(p_mngr, iids: &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_blk: &p_cli->pf_blks[0]);
671 qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line,
672 total_size: total * sizeof(struct src_ent),
673 elem_size: sizeof(struct src_ent));
674
675 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
676 client_id: 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_cli: &p_mngr->clients[ILT_CLI_TM]);
682 qed_cxt_tm_iids(p_hwfn, p_mngr, iids: &tm_iids);
683 total = tm_iids.pf_cids + tm_iids.pf_tids_total;
684 if (total) {
685 p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[0]);
686 qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line,
687 total_size: total * TM_ELEM_SIZE, TM_ELEM_SIZE);
688
689 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
690 client_id: 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_blk: &p_cli->vf_blks[0]);
698 qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line,
699 total_size: total * TM_ELEM_SIZE, TM_ELEM_SIZE);
700
701 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
702 client_id: 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, p_line: &curr_line,
707 client_id: 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_cli: &p_mngr->clients[ILT_CLI_TSDM]);
715 p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[SRQ_BLK]);
716 qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line,
717 total_size: total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
718
719 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line,
720 client_id: 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
733u32 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, seg: 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
771static 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(dev: &p_hwfn->cdev->pdev->dev,
782 size: p_t2->dma_mem[i].size,
783 cpu_addr: p_t2->dma_mem[i].virt_addr,
784 dma_handle: p_t2->dma_mem[i].phys_addr);
785
786 kfree(objp: p_t2->dma_mem);
787 p_t2->dma_mem = NULL;
788}
789
790static int
791qed_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(dev: &p_hwfn->cdev->pdev->dev,
805 size,
806 dma_handle: &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
819static 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, iids: &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(n: p_t2->num_pages, size: 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, page_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
895t2_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 */
907static 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
918static 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(ilt_clients: 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(dev: &p_hwfn->cdev->pdev->dev,
931 size: p_dma->size, cpu_addr: p_dma->virt_addr,
932 dma_handle: p_dma->phys_addr);
933 p_dma->virt_addr = NULL;
934 }
935 kfree(objp: p_mngr->ilt_shadow);
936}
937
938static int qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn,
939 struct qed_ilt_cli_blk *p_blk,
940 enum ilt_clients ilt_client,
941 u32 start_line_offset)
942{
943 struct phys_mem_desc *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
944 u32 lines, line, sz_left, lines_to_skip = 0;
945
946 /* Special handling for RoCE that supports dynamic allocation */
947 if (QED_IS_RDMA_PERSONALITY(p_hwfn) &&
948 ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
949 return 0;
950
951 lines_to_skip = p_blk->dynamic_line_cnt;
952
953 if (!p_blk->total_size)
954 return 0;
955
956 sz_left = p_blk->total_size;
957 lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) - lines_to_skip;
958 line = p_blk->start_line + start_line_offset -
959 p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
960
961 for (; lines; lines--) {
962 dma_addr_t p_phys;
963 void *p_virt;
964 u32 size;
965
966 size = min_t(u32, sz_left, p_blk->real_size_in_page);
967 p_virt = dma_alloc_coherent(dev: &p_hwfn->cdev->pdev->dev, size,
968 dma_handle: &p_phys, GFP_KERNEL);
969 if (!p_virt)
970 return -ENOMEM;
971
972 ilt_shadow[line].phys_addr = p_phys;
973 ilt_shadow[line].virt_addr = p_virt;
974 ilt_shadow[line].size = size;
975
976 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
977 "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
978 line, (u64)p_phys, p_virt, size);
979
980 sz_left -= size;
981 line++;
982 }
983
984 return 0;
985}
986
987static int qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn)
988{
989 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
990 struct qed_ilt_client_cfg *clients = p_mngr->clients;
991 struct qed_ilt_cli_blk *p_blk;
992 u32 size, i, j, k;
993 int rc;
994
995 size = qed_cxt_ilt_shadow_size(ilt_clients: clients);
996 p_mngr->ilt_shadow = kcalloc(n: size, size: sizeof(struct phys_mem_desc),
997 GFP_KERNEL);
998 if (!p_mngr->ilt_shadow) {
999 rc = -ENOMEM;
1000 goto ilt_shadow_fail;
1001 }
1002
1003 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1004 "Allocated 0x%x bytes for ilt shadow\n",
1005 (u32)(size * sizeof(struct phys_mem_desc)));
1006
1007 for_each_ilt_valid_client(i, clients) {
1008 for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1009 p_blk = &clients[i].pf_blks[j];
1010 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, ilt_client: i, start_line_offset: 0);
1011 if (rc)
1012 goto ilt_shadow_fail;
1013 }
1014 for (k = 0; k < p_mngr->vf_count; k++) {
1015 for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1016 u32 lines = clients[i].vf_total_lines * k;
1017
1018 p_blk = &clients[i].vf_blks[j];
1019 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, ilt_client: i, start_line_offset: lines);
1020 if (rc)
1021 goto ilt_shadow_fail;
1022 }
1023 }
1024 }
1025
1026 return 0;
1027
1028ilt_shadow_fail:
1029 qed_ilt_shadow_free(p_hwfn);
1030 return rc;
1031}
1032
1033static void qed_cid_map_free(struct qed_hwfn *p_hwfn)
1034{
1035 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1036 u32 type, vf;
1037
1038 for (type = 0; type < MAX_CONN_TYPES; type++) {
1039 bitmap_free(bitmap: p_mngr->acquired[type].cid_map);
1040 p_mngr->acquired[type].max_count = 0;
1041 p_mngr->acquired[type].start_cid = 0;
1042
1043 for (vf = 0; vf < MAX_NUM_VFS; vf++) {
1044 bitmap_free(bitmap: p_mngr->acquired_vf[type][vf].cid_map);
1045 p_mngr->acquired_vf[type][vf].max_count = 0;
1046 p_mngr->acquired_vf[type][vf].start_cid = 0;
1047 }
1048 }
1049}
1050
1051static int
1052qed_cid_map_alloc_single(struct qed_hwfn *p_hwfn,
1053 u32 type,
1054 u32 cid_start,
1055 u32 cid_count, struct qed_cid_acquired_map *p_map)
1056{
1057 if (!cid_count)
1058 return 0;
1059
1060 p_map->cid_map = bitmap_zalloc(nbits: cid_count, GFP_KERNEL);
1061 if (!p_map->cid_map)
1062 return -ENOMEM;
1063
1064 p_map->max_count = cid_count;
1065 p_map->start_cid = cid_start;
1066
1067 DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1068 "Type %08x start: %08x count %08x\n",
1069 type, p_map->start_cid, p_map->max_count);
1070
1071 return 0;
1072}
1073
1074static int qed_cid_map_alloc(struct qed_hwfn *p_hwfn)
1075{
1076 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1077 u32 start_cid = 0, vf_start_cid = 0;
1078 u32 type, vf;
1079
1080 for (type = 0; type < MAX_CONN_TYPES; type++) {
1081 struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[type];
1082 struct qed_cid_acquired_map *p_map;
1083
1084 /* Handle PF maps */
1085 p_map = &p_mngr->acquired[type];
1086 if (qed_cid_map_alloc_single(p_hwfn, type, cid_start: start_cid,
1087 cid_count: p_cfg->cid_count, p_map))
1088 goto cid_map_fail;
1089
1090 /* Handle VF maps */
1091 for (vf = 0; vf < MAX_NUM_VFS; vf++) {
1092 p_map = &p_mngr->acquired_vf[type][vf];
1093 if (qed_cid_map_alloc_single(p_hwfn, type,
1094 cid_start: vf_start_cid,
1095 cid_count: p_cfg->cids_per_vf, p_map))
1096 goto cid_map_fail;
1097 }
1098
1099 start_cid += p_cfg->cid_count;
1100 vf_start_cid += p_cfg->cids_per_vf;
1101 }
1102
1103 return 0;
1104
1105cid_map_fail:
1106 qed_cid_map_free(p_hwfn);
1107 return -ENOMEM;
1108}
1109
1110int qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn)
1111{
1112 struct qed_ilt_client_cfg *clients;
1113 struct qed_cxt_mngr *p_mngr;
1114 u32 i;
1115
1116 p_mngr = kzalloc(size: sizeof(*p_mngr), GFP_KERNEL);
1117 if (!p_mngr)
1118 return -ENOMEM;
1119
1120 /* Initialize ILT client registers */
1121 clients = p_mngr->clients;
1122 clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1123 clients[ILT_CLI_CDUC].last.reg = ILT_CFG_REG(CDUC, LAST_ILT);
1124 clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1125
1126 clients[ILT_CLI_QM].first.reg = ILT_CFG_REG(QM, FIRST_ILT);
1127 clients[ILT_CLI_QM].last.reg = ILT_CFG_REG(QM, LAST_ILT);
1128 clients[ILT_CLI_QM].p_size.reg = ILT_CFG_REG(QM, P_SIZE);
1129
1130 clients[ILT_CLI_TM].first.reg = ILT_CFG_REG(TM, FIRST_ILT);
1131 clients[ILT_CLI_TM].last.reg = ILT_CFG_REG(TM, LAST_ILT);
1132 clients[ILT_CLI_TM].p_size.reg = ILT_CFG_REG(TM, P_SIZE);
1133
1134 clients[ILT_CLI_SRC].first.reg = ILT_CFG_REG(SRC, FIRST_ILT);
1135 clients[ILT_CLI_SRC].last.reg = ILT_CFG_REG(SRC, LAST_ILT);
1136 clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1137
1138 clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1139 clients[ILT_CLI_CDUT].last.reg = ILT_CFG_REG(CDUT, LAST_ILT);
1140 clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1141
1142 clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1143 clients[ILT_CLI_TSDM].last.reg = ILT_CFG_REG(TSDM, LAST_ILT);
1144 clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1145 /* default ILT page size for all clients is 64K */
1146 for (i = 0; i < MAX_ILT_CLIENTS; i++)
1147 p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE;
1148
1149 p_mngr->conn_ctx_size = CONN_CXT_SIZE(p_hwfn);
1150
1151 /* Initialize task sizes */
1152 p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1153 p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1154
1155 if (p_hwfn->cdev->p_iov_info) {
1156 p_mngr->vf_count = p_hwfn->cdev->p_iov_info->total_vfs;
1157 p_mngr->first_vf_in_pf =
1158 p_hwfn->cdev->p_iov_info->first_vf_in_pf;
1159 }
1160 /* Initialize the dynamic ILT allocation mutex */
1161 mutex_init(&p_mngr->mutex);
1162
1163 /* Set the cxt mangr pointer priori to further allocations */
1164 p_hwfn->p_cxt_mngr = p_mngr;
1165
1166 return 0;
1167}
1168
1169int qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn)
1170{
1171 int rc;
1172
1173 /* Allocate the ILT shadow table */
1174 rc = qed_ilt_shadow_alloc(p_hwfn);
1175 if (rc)
1176 goto tables_alloc_fail;
1177
1178 /* Allocate the T2 table */
1179 rc = qed_cxt_src_t2_alloc(p_hwfn);
1180 if (rc)
1181 goto tables_alloc_fail;
1182
1183 /* Allocate and initialize the acquired cids bitmaps */
1184 rc = qed_cid_map_alloc(p_hwfn);
1185 if (rc)
1186 goto tables_alloc_fail;
1187
1188 return 0;
1189
1190tables_alloc_fail:
1191 qed_cxt_mngr_free(p_hwfn);
1192 return rc;
1193}
1194
1195void qed_cxt_mngr_free(struct qed_hwfn *p_hwfn)
1196{
1197 if (!p_hwfn->p_cxt_mngr)
1198 return;
1199
1200 qed_cid_map_free(p_hwfn);
1201 qed_cxt_src_t2_free(p_hwfn);
1202 qed_ilt_shadow_free(p_hwfn);
1203 kfree(objp: p_hwfn->p_cxt_mngr);
1204
1205 p_hwfn->p_cxt_mngr = NULL;
1206}
1207
1208void qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn)
1209{
1210 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1211 struct qed_cid_acquired_map *p_map;
1212 struct qed_conn_type_cfg *p_cfg;
1213 int type;
1214
1215 /* Reset acquired cids */
1216 for (type = 0; type < MAX_CONN_TYPES; type++) {
1217 u32 vf;
1218
1219 p_cfg = &p_mngr->conn_cfg[type];
1220 if (p_cfg->cid_count) {
1221 p_map = &p_mngr->acquired[type];
1222 bitmap_zero(dst: p_map->cid_map, nbits: p_map->max_count);
1223 }
1224
1225 if (!p_cfg->cids_per_vf)
1226 continue;
1227
1228 for (vf = 0; vf < MAX_NUM_VFS; vf++) {
1229 p_map = &p_mngr->acquired_vf[type][vf];
1230 bitmap_zero(dst: p_map->cid_map, nbits: p_map->max_count);
1231 }
1232 }
1233}
1234
1235/* CDU Common */
1236#define CDUC_CXT_SIZE_SHIFT \
1237 CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1238
1239#define CDUC_CXT_SIZE_MASK \
1240 (CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1241
1242#define CDUC_BLOCK_WASTE_SHIFT \
1243 CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1244
1245#define CDUC_BLOCK_WASTE_MASK \
1246 (CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1247
1248#define CDUC_NCIB_SHIFT \
1249 CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1250
1251#define CDUC_NCIB_MASK \
1252 (CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1253
1254#define CDUT_TYPE0_CXT_SIZE_SHIFT \
1255 CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1256
1257#define CDUT_TYPE0_CXT_SIZE_MASK \
1258 (CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >> \
1259 CDUT_TYPE0_CXT_SIZE_SHIFT)
1260
1261#define CDUT_TYPE0_BLOCK_WASTE_SHIFT \
1262 CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1263
1264#define CDUT_TYPE0_BLOCK_WASTE_MASK \
1265 (CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >> \
1266 CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1267
1268#define CDUT_TYPE0_NCIB_SHIFT \
1269 CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1270
1271#define CDUT_TYPE0_NCIB_MASK \
1272 (CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >> \
1273 CDUT_TYPE0_NCIB_SHIFT)
1274
1275#define CDUT_TYPE1_CXT_SIZE_SHIFT \
1276 CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1277
1278#define CDUT_TYPE1_CXT_SIZE_MASK \
1279 (CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >> \
1280 CDUT_TYPE1_CXT_SIZE_SHIFT)
1281
1282#define CDUT_TYPE1_BLOCK_WASTE_SHIFT \
1283 CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1284
1285#define CDUT_TYPE1_BLOCK_WASTE_MASK \
1286 (CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >> \
1287 CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1288
1289#define CDUT_TYPE1_NCIB_SHIFT \
1290 CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1291
1292#define CDUT_TYPE1_NCIB_MASK \
1293 (CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >> \
1294 CDUT_TYPE1_NCIB_SHIFT)
1295
1296static void qed_cdu_init_common(struct qed_hwfn *p_hwfn)
1297{
1298 u32 page_sz, elems_per_page, block_waste, cxt_size, cdu_params = 0;
1299
1300 /* CDUC - connection configuration */
1301 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1302 cxt_size = CONN_CXT_SIZE(p_hwfn);
1303 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1304 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1305
1306 SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1307 SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1308 SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
1309 STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1310
1311 /* CDUT - type-0 tasks configuration */
1312 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1313 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1314 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1315 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1316
1317 /* cxt size and block-waste are multipes of 8 */
1318 cdu_params = 0;
1319 SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1320 SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1321 SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1322 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1323
1324 /* CDUT - type-1 tasks configuration */
1325 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1326 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1327 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1328
1329 /* cxt size and block-waste are multipes of 8 */
1330 cdu_params = 0;
1331 SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1332 SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1333 SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1334 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1335}
1336
1337/* CDU PF */
1338#define CDU_SEG_REG_TYPE_SHIFT CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1339#define CDU_SEG_REG_TYPE_MASK 0x1
1340#define CDU_SEG_REG_OFFSET_SHIFT 0
1341#define CDU_SEG_REG_OFFSET_MASK CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1342
1343static void qed_cdu_init_pf(struct qed_hwfn *p_hwfn)
1344{
1345 struct qed_ilt_client_cfg *p_cli;
1346 struct qed_tid_seg *p_seg;
1347 u32 cdu_seg_params, offset;
1348 int i;
1349
1350 static const u32 rt_type_offset_arr[] = {
1351 CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1352 CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1353 CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1354 CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1355 };
1356
1357 static const u32 rt_type_offset_fl_arr[] = {
1358 CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1359 CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1360 CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1361 CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1362 };
1363
1364 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1365
1366 /* There are initializations only for CDUT during pf Phase */
1367 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1368 /* Segment 0 */
1369 p_seg = qed_cxt_tid_seg_info(p_hwfn, seg: i);
1370 if (!p_seg)
1371 continue;
1372
1373 /* Note: start_line is already adjusted for the CDU
1374 * segment register granularity, so we just need to
1375 * divide. Adjustment is implicit as we assume ILT
1376 * Page size is larger than 32K!
1377 */
1378 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1379 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1380 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1381
1382 cdu_seg_params = 0;
1383 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1384 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1385 STORE_RT_REG(p_hwfn, rt_type_offset_arr[i], cdu_seg_params);
1386
1387 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1388 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1389 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1390
1391 cdu_seg_params = 0;
1392 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1393 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1394 STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i], cdu_seg_params);
1395 }
1396}
1397
1398void qed_qm_init_pf(struct qed_hwfn *p_hwfn,
1399 struct qed_ptt *p_ptt, bool is_pf_loading)
1400{
1401 struct qed_qm_info *qm_info = &p_hwfn->qm_info;
1402 struct qed_qm_pf_rt_init_params params;
1403 struct qed_qm_iids iids;
1404
1405 memset(&iids, 0, sizeof(iids));
1406 qed_cxt_qm_iids(p_hwfn, iids: &iids);
1407
1408 memset(&params, 0, sizeof(params));
1409 params.port_id = p_hwfn->port_id;
1410 params.pf_id = p_hwfn->rel_pf_id;
1411 params.max_phys_tcs_per_port = qm_info->max_phys_tcs_per_port;
1412 params.is_pf_loading = is_pf_loading;
1413 params.num_pf_cids = iids.cids;
1414 params.num_vf_cids = iids.vf_cids;
1415 params.num_tids = iids.tids;
1416 params.start_pq = qm_info->start_pq;
1417 params.num_pf_pqs = qm_info->num_pqs - qm_info->num_vf_pqs;
1418 params.num_vf_pqs = qm_info->num_vf_pqs;
1419 params.start_vport = qm_info->start_vport;
1420 params.num_vports = qm_info->num_vports;
1421 params.pf_wfq = qm_info->pf_wfq;
1422 params.pf_rl = qm_info->pf_rl;
1423 params.pq_params = qm_info->qm_pq_params;
1424 params.vport_params = qm_info->qm_vport_params;
1425
1426 qed_qm_pf_rt_init(p_hwfn, p_ptt, p_params: &params);
1427}
1428
1429/* CM PF */
1430static void qed_cm_init_pf(struct qed_hwfn *p_hwfn)
1431{
1432 /* XCM pure-LB queue */
1433 STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET,
1434 qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1435}
1436
1437/* DQ PF */
1438static void qed_dq_init_pf(struct qed_hwfn *p_hwfn)
1439{
1440 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1441 u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1442
1443 dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1444 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1445
1446 dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1447 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1448
1449 dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1450 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1451
1452 dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1453 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1454
1455 dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1456 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1457
1458 dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1459 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1460
1461 dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1462 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1463
1464 dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1465 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1466
1467 dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1468 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1469
1470 dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1471 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1472
1473 dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1474 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1475
1476 dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1477 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1478
1479 /* Connection types 6 & 7 are not in use, yet they must be configured
1480 * as the highest possible connection. Not configuring them means the
1481 * defaults will be used, and with a large number of cids a bug may
1482 * occur, if the defaults will be smaller than dq_pf_max_cid /
1483 * dq_vf_max_cid.
1484 */
1485 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1486 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1487
1488 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1489 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1490}
1491
1492static void qed_ilt_bounds_init(struct qed_hwfn *p_hwfn)
1493{
1494 struct qed_ilt_client_cfg *ilt_clients;
1495 int i;
1496
1497 ilt_clients = p_hwfn->p_cxt_mngr->clients;
1498 for_each_ilt_valid_client(i, ilt_clients) {
1499 STORE_RT_REG(p_hwfn,
1500 ilt_clients[i].first.reg,
1501 ilt_clients[i].first.val);
1502 STORE_RT_REG(p_hwfn,
1503 ilt_clients[i].last.reg, ilt_clients[i].last.val);
1504 STORE_RT_REG(p_hwfn,
1505 ilt_clients[i].p_size.reg,
1506 ilt_clients[i].p_size.val);
1507 }
1508}
1509
1510static void qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn)
1511{
1512 struct qed_ilt_client_cfg *p_cli;
1513 u32 blk_factor;
1514
1515 /* For simplicty we set the 'block' to be an ILT page */
1516 if (p_hwfn->cdev->p_iov_info) {
1517 struct qed_hw_sriov_info *p_iov = p_hwfn->cdev->p_iov_info;
1518
1519 STORE_RT_REG(p_hwfn,
1520 PSWRQ2_REG_VF_BASE_RT_OFFSET,
1521 p_iov->first_vf_in_pf);
1522 STORE_RT_REG(p_hwfn,
1523 PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1524 p_iov->first_vf_in_pf + p_iov->total_vfs);
1525 }
1526
1527 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1528 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1529 if (p_cli->active) {
1530 STORE_RT_REG(p_hwfn,
1531 PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1532 blk_factor);
1533 STORE_RT_REG(p_hwfn,
1534 PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1535 p_cli->pf_total_lines);
1536 STORE_RT_REG(p_hwfn,
1537 PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1538 p_cli->vf_total_lines);
1539 }
1540
1541 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1542 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1543 if (p_cli->active) {
1544 STORE_RT_REG(p_hwfn,
1545 PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1546 blk_factor);
1547 STORE_RT_REG(p_hwfn,
1548 PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1549 p_cli->pf_total_lines);
1550 STORE_RT_REG(p_hwfn,
1551 PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1552 p_cli->vf_total_lines);
1553 }
1554
1555 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1556 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1557 if (p_cli->active) {
1558 STORE_RT_REG(p_hwfn,
1559 PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET, blk_factor);
1560 STORE_RT_REG(p_hwfn,
1561 PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1562 p_cli->pf_total_lines);
1563 STORE_RT_REG(p_hwfn,
1564 PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1565 p_cli->vf_total_lines);
1566 }
1567}
1568
1569/* ILT (PSWRQ2) PF */
1570static void qed_ilt_init_pf(struct qed_hwfn *p_hwfn)
1571{
1572 struct qed_ilt_client_cfg *clients;
1573 struct qed_cxt_mngr *p_mngr;
1574 struct phys_mem_desc *p_shdw;
1575 u32 line, rt_offst, i;
1576
1577 qed_ilt_bounds_init(p_hwfn);
1578 qed_ilt_vf_bounds_init(p_hwfn);
1579
1580 p_mngr = p_hwfn->p_cxt_mngr;
1581 p_shdw = p_mngr->ilt_shadow;
1582 clients = p_hwfn->p_cxt_mngr->clients;
1583
1584 for_each_ilt_valid_client(i, clients) {
1585 /** Client's 1st val and RT array are absolute, ILT shadows'
1586 * lines are relative.
1587 */
1588 line = clients[i].first.val - p_mngr->pf_start_line;
1589 rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1590 clients[i].first.val * ILT_ENTRY_IN_REGS;
1591
1592 for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1593 line++, rt_offst += ILT_ENTRY_IN_REGS) {
1594 u64 ilt_hw_entry = 0;
1595
1596 /** p_virt could be NULL incase of dynamic
1597 * allocation
1598 */
1599 if (p_shdw[line].virt_addr) {
1600 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1601 SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1602 (p_shdw[line].phys_addr >> 12));
1603
1604 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1605 "Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1606 rt_offst, line, i,
1607 (u64)(p_shdw[line].phys_addr >> 12));
1608 }
1609
1610 STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1611 }
1612 }
1613}
1614
1615/* SRC (Searcher) PF */
1616static void qed_src_init_pf(struct qed_hwfn *p_hwfn)
1617{
1618 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1619 u32 rounded_conn_num, conn_num, conn_max;
1620 struct qed_src_iids src_iids;
1621
1622 memset(&src_iids, 0, sizeof(src_iids));
1623 qed_cxt_src_iids(p_mngr, iids: &src_iids);
1624 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1625 if (!conn_num)
1626 return;
1627
1628 conn_max = max_t(u32, conn_num, SRC_MIN_NUM_ELEMS);
1629 rounded_conn_num = roundup_pow_of_two(conn_max);
1630
1631 STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1632 STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1633 ilog2(rounded_conn_num));
1634
1635 STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1636 p_hwfn->p_cxt_mngr->src_t2.first_free);
1637 STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1638 p_hwfn->p_cxt_mngr->src_t2.last_free);
1639}
1640
1641/* Timers PF */
1642#define TM_CFG_NUM_IDS_SHIFT 0
1643#define TM_CFG_NUM_IDS_MASK 0xFFFFULL
1644#define TM_CFG_PRE_SCAN_OFFSET_SHIFT 16
1645#define TM_CFG_PRE_SCAN_OFFSET_MASK 0x1FFULL
1646#define TM_CFG_PARENT_PF_SHIFT 25
1647#define TM_CFG_PARENT_PF_MASK 0x7ULL
1648
1649#define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT 30
1650#define TM_CFG_CID_PRE_SCAN_ROWS_MASK 0x1FFULL
1651
1652#define TM_CFG_TID_OFFSET_SHIFT 30
1653#define TM_CFG_TID_OFFSET_MASK 0x7FFFFULL
1654#define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT 49
1655#define TM_CFG_TID_PRE_SCAN_ROWS_MASK 0x1FFULL
1656
1657static void qed_tm_init_pf(struct qed_hwfn *p_hwfn)
1658{
1659 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1660 u32 active_seg_mask = 0, tm_offset, rt_reg;
1661 struct qed_tm_iids tm_iids;
1662 u64 cfg_word;
1663 u8 i;
1664
1665 memset(&tm_iids, 0, sizeof(tm_iids));
1666 qed_cxt_tm_iids(p_hwfn, p_mngr, iids: &tm_iids);
1667
1668 /* @@@TBD No pre-scan for now */
1669
1670 /* Note: We assume consecutive VFs for a PF */
1671 for (i = 0; i < p_mngr->vf_count; i++) {
1672 cfg_word = 0;
1673 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1674 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1675 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1676 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0);
1677 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1678 (sizeof(cfg_word) / sizeof(u32)) *
1679 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1680 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1681 }
1682
1683 cfg_word = 0;
1684 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1685 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1686 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0); /* n/a for PF */
1687 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1688
1689 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1690 (sizeof(cfg_word) / sizeof(u32)) *
1691 (NUM_OF_VFS(p_hwfn->cdev) + p_hwfn->rel_pf_id);
1692 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1693
1694 /* enale scan */
1695 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1696 tm_iids.pf_cids ? 0x1 : 0x0);
1697
1698 /* @@@TBD how to enable the scan for the VFs */
1699
1700 tm_offset = tm_iids.per_vf_cids;
1701
1702 /* Note: We assume consecutive VFs for a PF */
1703 for (i = 0; i < p_mngr->vf_count; i++) {
1704 cfg_word = 0;
1705 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1706 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1707 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1708 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1709 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1710
1711 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1712 (sizeof(cfg_word) / sizeof(u32)) *
1713 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1714
1715 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1716 }
1717
1718 tm_offset = tm_iids.pf_cids;
1719 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1720 cfg_word = 0;
1721 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1722 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1723 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1724 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1725 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1726
1727 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1728 (sizeof(cfg_word) / sizeof(u32)) *
1729 (NUM_OF_VFS(p_hwfn->cdev) +
1730 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1731
1732 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1733 active_seg_mask |= (tm_iids.pf_tids[i] ? BIT(i) : 0);
1734
1735 tm_offset += tm_iids.pf_tids[i];
1736 }
1737
1738 if (QED_IS_RDMA_PERSONALITY(p_hwfn))
1739 active_seg_mask = 0;
1740
1741 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1742
1743 /* @@@TBD how to enable the scan for the VFs */
1744}
1745
1746static void qed_prs_init_common(struct qed_hwfn *p_hwfn)
1747{
1748 if ((p_hwfn->hw_info.personality == QED_PCI_FCOE) &&
1749 p_hwfn->pf_params.fcoe_pf_params.is_target)
1750 STORE_RT_REG(p_hwfn,
1751 PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1752}
1753
1754static void qed_prs_init_pf(struct qed_hwfn *p_hwfn)
1755{
1756 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1757 struct qed_conn_type_cfg *p_fcoe;
1758 struct qed_tid_seg *p_tid;
1759
1760 p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1761
1762 /* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1763 if (!p_fcoe->cid_count)
1764 return;
1765
1766 p_tid = &p_fcoe->tid_seg[QED_CXT_FCOE_TID_SEG];
1767 if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1768 STORE_RT_REG_AGG(p_hwfn,
1769 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1770 p_tid->count);
1771 } else {
1772 STORE_RT_REG_AGG(p_hwfn,
1773 PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1774 p_tid->count);
1775 }
1776}
1777
1778void qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn)
1779{
1780 qed_cdu_init_common(p_hwfn);
1781 qed_prs_init_common(p_hwfn);
1782}
1783
1784void qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
1785{
1786 qed_qm_init_pf(p_hwfn, p_ptt, is_pf_loading: true);
1787 qed_cm_init_pf(p_hwfn);
1788 qed_dq_init_pf(p_hwfn);
1789 qed_cdu_init_pf(p_hwfn);
1790 qed_ilt_init_pf(p_hwfn);
1791 qed_src_init_pf(p_hwfn);
1792 qed_tm_init_pf(p_hwfn);
1793 qed_prs_init_pf(p_hwfn);
1794}
1795
1796int _qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1797 enum protocol_type type, u32 *p_cid, u8 vfid)
1798{
1799 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1800 struct qed_cid_acquired_map *p_map;
1801 u32 rel_cid;
1802
1803 if (type >= MAX_CONN_TYPES) {
1804 DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1805 return -EINVAL;
1806 }
1807
1808 if (vfid >= MAX_NUM_VFS && vfid != QED_CXT_PF_CID) {
1809 DP_NOTICE(p_hwfn, "VF [%02x] is out of range\n", vfid);
1810 return -EINVAL;
1811 }
1812
1813 /* Determine the right map to take this CID from */
1814 if (vfid == QED_CXT_PF_CID)
1815 p_map = &p_mngr->acquired[type];
1816 else
1817 p_map = &p_mngr->acquired_vf[type][vfid];
1818
1819 if (!p_map->cid_map) {
1820 DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1821 return -EINVAL;
1822 }
1823
1824 rel_cid = find_first_zero_bit(addr: p_map->cid_map, size: p_map->max_count);
1825
1826 if (rel_cid >= p_map->max_count) {
1827 DP_NOTICE(p_hwfn, "no CID available for protocol %d\n", type);
1828 return -EINVAL;
1829 }
1830
1831 __set_bit(rel_cid, p_map->cid_map);
1832
1833 *p_cid = rel_cid + p_map->start_cid;
1834
1835 DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1836 "Acquired cid 0x%08x [rel. %08x] vfid %02x type %d\n",
1837 *p_cid, rel_cid, vfid, type);
1838
1839 return 0;
1840}
1841
1842int qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1843 enum protocol_type type, u32 *p_cid)
1844{
1845 return _qed_cxt_acquire_cid(p_hwfn, type, p_cid, QED_CXT_PF_CID);
1846}
1847
1848static bool qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn,
1849 u32 cid,
1850 u8 vfid,
1851 enum protocol_type *p_type,
1852 struct qed_cid_acquired_map **pp_map)
1853{
1854 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1855 u32 rel_cid;
1856
1857 /* Iterate over protocols and find matching cid range */
1858 for (*p_type = 0; *p_type < MAX_CONN_TYPES; (*p_type)++) {
1859 if (vfid == QED_CXT_PF_CID)
1860 *pp_map = &p_mngr->acquired[*p_type];
1861 else
1862 *pp_map = &p_mngr->acquired_vf[*p_type][vfid];
1863
1864 if (!((*pp_map)->cid_map))
1865 continue;
1866 if (cid >= (*pp_map)->start_cid &&
1867 cid < (*pp_map)->start_cid + (*pp_map)->max_count)
1868 break;
1869 }
1870
1871 if (*p_type == MAX_CONN_TYPES) {
1872 DP_NOTICE(p_hwfn, "Invalid CID %d vfid %02x", cid, vfid);
1873 goto fail;
1874 }
1875
1876 rel_cid = cid - (*pp_map)->start_cid;
1877 if (!test_bit(rel_cid, (*pp_map)->cid_map)) {
1878 DP_NOTICE(p_hwfn, "CID %d [vifd %02x] not acquired",
1879 cid, vfid);
1880 goto fail;
1881 }
1882
1883 return true;
1884fail:
1885 *p_type = MAX_CONN_TYPES;
1886 *pp_map = NULL;
1887 return false;
1888}
1889
1890void _qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid, u8 vfid)
1891{
1892 struct qed_cid_acquired_map *p_map = NULL;
1893 enum protocol_type type;
1894 bool b_acquired;
1895 u32 rel_cid;
1896
1897 if (vfid != QED_CXT_PF_CID && vfid > MAX_NUM_VFS) {
1898 DP_NOTICE(p_hwfn,
1899 "Trying to return incorrect CID belonging to VF %02x\n",
1900 vfid);
1901 return;
1902 }
1903
1904 /* Test acquired and find matching per-protocol map */
1905 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid, vfid,
1906 p_type: &type, pp_map: &p_map);
1907
1908 if (!b_acquired)
1909 return;
1910
1911 rel_cid = cid - p_map->start_cid;
1912 clear_bit(nr: rel_cid, addr: p_map->cid_map);
1913
1914 DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1915 "Released CID 0x%08x [rel. %08x] vfid %02x type %d\n",
1916 cid, rel_cid, vfid, type);
1917}
1918
1919void qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid)
1920{
1921 _qed_cxt_release_cid(p_hwfn, cid, QED_CXT_PF_CID);
1922}
1923
1924int qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn, struct qed_cxt_info *p_info)
1925{
1926 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1927 struct qed_cid_acquired_map *p_map = NULL;
1928 u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
1929 enum protocol_type type;
1930 bool b_acquired;
1931
1932 /* Test acquired and find matching per-protocol map */
1933 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid: p_info->iid,
1934 QED_CXT_PF_CID, p_type: &type, pp_map: &p_map);
1935
1936 if (!b_acquired)
1937 return -EINVAL;
1938
1939 /* set the protocl type */
1940 p_info->type = type;
1941
1942 /* compute context virtual pointer */
1943 hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1944
1945 conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
1946 cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
1947 line = p_info->iid / cxts_per_p;
1948
1949 /* Make sure context is allocated (dynamic allocation) */
1950 if (!p_mngr->ilt_shadow[line].virt_addr)
1951 return -EINVAL;
1952
1953 p_info->p_cxt = p_mngr->ilt_shadow[line].virt_addr +
1954 p_info->iid % cxts_per_p * conn_cxt_size;
1955
1956 DP_VERBOSE(p_hwfn, (QED_MSG_ILT | QED_MSG_CXT),
1957 "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
1958 p_info->iid / cxts_per_p, p_info->p_cxt, p_info->iid);
1959
1960 return 0;
1961}
1962
1963static void qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn,
1964 struct qed_rdma_pf_params *p_params,
1965 u32 num_tasks)
1966{
1967 u32 num_cons, num_qps;
1968 enum protocol_type proto;
1969
1970 if (p_hwfn->mcp_info->func_info.protocol == QED_PCI_ETH_RDMA) {
1971 DP_VERBOSE(p_hwfn, QED_MSG_SP,
1972 "Current day drivers don't support RoCE & iWARP simultaneously on the same PF. Default to RoCE-only\n");
1973 p_hwfn->hw_info.personality = QED_PCI_ETH_ROCE;
1974 }
1975
1976 switch (p_hwfn->hw_info.personality) {
1977 case QED_PCI_ETH_IWARP:
1978 /* Each QP requires one connection */
1979 num_cons = min_t(u32, IWARP_MAX_QPS, p_params->num_qps);
1980 proto = PROTOCOLID_IWARP;
1981 break;
1982 case QED_PCI_ETH_ROCE:
1983 num_qps = min_t(u32, ROCE_MAX_QPS, p_params->num_qps);
1984 num_cons = num_qps * 2; /* each QP requires two connections */
1985 proto = PROTOCOLID_ROCE;
1986 break;
1987 default:
1988 return;
1989 }
1990
1991 if (num_cons && num_tasks) {
1992 u32 num_srqs, num_xrc_srqs;
1993
1994 qed_cxt_set_proto_cid_count(p_hwfn, type: proto, cid_count: num_cons, vf_cid_cnt: 0);
1995
1996 /* Deliberatly passing ROCE for tasks id. This is because
1997 * iWARP / RoCE share the task id.
1998 */
1999 qed_cxt_set_proto_tid_count(p_hwfn, proto: PROTOCOLID_ROCE,
2000 QED_CXT_ROCE_TID_SEG, seg_type: 1,
2001 count: num_tasks, has_fl: false);
2002
2003 num_srqs = min_t(u32, QED_RDMA_MAX_SRQS, p_params->num_srqs);
2004
2005 /* XRC SRQs populate a single ILT page */
2006 num_xrc_srqs = qed_cxt_xrc_srqs_per_page(p_hwfn);
2007
2008 qed_cxt_set_srq_count(p_hwfn, num_srqs, num_xrc_srqs);
2009 } else {
2010 DP_INFO(p_hwfn->cdev,
2011 "RDMA personality used without setting params!\n");
2012 }
2013}
2014
2015int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn, u32 rdma_tasks)
2016{
2017 /* Set the number of required CORE connections */
2018 u32 core_cids = 1; /* SPQ */
2019
2020 if (p_hwfn->using_ll2)
2021 core_cids += 4;
2022 qed_cxt_set_proto_cid_count(p_hwfn, type: PROTOCOLID_CORE, cid_count: core_cids, vf_cid_cnt: 0);
2023
2024 switch (p_hwfn->hw_info.personality) {
2025 case QED_PCI_ETH_RDMA:
2026 case QED_PCI_ETH_IWARP:
2027 case QED_PCI_ETH_ROCE:
2028 {
2029 qed_rdma_set_pf_params(p_hwfn,
2030 p_params: &p_hwfn->
2031 pf_params.rdma_pf_params,
2032 num_tasks: rdma_tasks);
2033 /* no need for break since RoCE coexist with Ethernet */
2034 }
2035 fallthrough;
2036 case QED_PCI_ETH:
2037 {
2038 struct qed_eth_pf_params *p_params =
2039 &p_hwfn->pf_params.eth_pf_params;
2040
2041 if (!p_params->num_vf_cons)
2042 p_params->num_vf_cons =
2043 ETH_PF_PARAMS_VF_CONS_DEFAULT;
2044 qed_cxt_set_proto_cid_count(p_hwfn, type: PROTOCOLID_ETH,
2045 cid_count: p_params->num_cons,
2046 vf_cid_cnt: p_params->num_vf_cons);
2047 p_hwfn->p_cxt_mngr->arfs_count = p_params->num_arfs_filters;
2048 break;
2049 }
2050 case QED_PCI_FCOE:
2051 {
2052 struct qed_fcoe_pf_params *p_params;
2053
2054 p_params = &p_hwfn->pf_params.fcoe_pf_params;
2055
2056 if (p_params->num_cons && p_params->num_tasks) {
2057 qed_cxt_set_proto_cid_count(p_hwfn,
2058 type: PROTOCOLID_FCOE,
2059 cid_count: p_params->num_cons,
2060 vf_cid_cnt: 0);
2061 qed_cxt_set_proto_tid_count(p_hwfn, proto: PROTOCOLID_FCOE,
2062 QED_CXT_FCOE_TID_SEG, seg_type: 0,
2063 count: p_params->num_tasks, has_fl: true);
2064 } else {
2065 DP_INFO(p_hwfn->cdev,
2066 "Fcoe personality used without setting params!\n");
2067 }
2068 break;
2069 }
2070 case QED_PCI_ISCSI:
2071 {
2072 struct qed_iscsi_pf_params *p_params;
2073
2074 p_params = &p_hwfn->pf_params.iscsi_pf_params;
2075
2076 if (p_params->num_cons && p_params->num_tasks) {
2077 qed_cxt_set_proto_cid_count(p_hwfn,
2078 type: PROTOCOLID_TCP_ULP,
2079 cid_count: p_params->num_cons,
2080 vf_cid_cnt: 0);
2081 qed_cxt_set_proto_tid_count(p_hwfn,
2082 proto: PROTOCOLID_TCP_ULP,
2083 QED_CXT_TCP_ULP_TID_SEG,
2084 seg_type: 0,
2085 count: p_params->num_tasks,
2086 has_fl: true);
2087 } else {
2088 DP_INFO(p_hwfn->cdev,
2089 "Iscsi personality used without setting params!\n");
2090 }
2091 break;
2092 }
2093 case QED_PCI_NVMETCP:
2094 {
2095 struct qed_nvmetcp_pf_params *p_params;
2096
2097 p_params = &p_hwfn->pf_params.nvmetcp_pf_params;
2098
2099 if (p_params->num_cons && p_params->num_tasks) {
2100 qed_cxt_set_proto_cid_count(p_hwfn,
2101 type: PROTOCOLID_TCP_ULP,
2102 cid_count: p_params->num_cons,
2103 vf_cid_cnt: 0);
2104 qed_cxt_set_proto_tid_count(p_hwfn,
2105 proto: PROTOCOLID_TCP_ULP,
2106 QED_CXT_TCP_ULP_TID_SEG,
2107 seg_type: 0,
2108 count: p_params->num_tasks,
2109 has_fl: true);
2110 } else {
2111 DP_INFO(p_hwfn->cdev,
2112 "NvmeTCP personality used without setting params!\n");
2113 }
2114 break;
2115 }
2116 default:
2117 return -EINVAL;
2118 }
2119
2120 return 0;
2121}
2122
2123int qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn,
2124 struct qed_tid_mem *p_info)
2125{
2126 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2127 u32 proto, seg, total_lines, i, shadow_line;
2128 struct qed_ilt_client_cfg *p_cli;
2129 struct qed_ilt_cli_blk *p_fl_seg;
2130 struct qed_tid_seg *p_seg_info;
2131
2132 /* Verify the personality */
2133 switch (p_hwfn->hw_info.personality) {
2134 case QED_PCI_FCOE:
2135 proto = PROTOCOLID_FCOE;
2136 seg = QED_CXT_FCOE_TID_SEG;
2137 break;
2138 case QED_PCI_ISCSI:
2139 case QED_PCI_NVMETCP:
2140 proto = PROTOCOLID_TCP_ULP;
2141 seg = QED_CXT_TCP_ULP_TID_SEG;
2142 break;
2143 default:
2144 return -EINVAL;
2145 }
2146
2147 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2148 if (!p_cli->active)
2149 return -EINVAL;
2150
2151 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2152 if (!p_seg_info->has_fl_mem)
2153 return -EINVAL;
2154
2155 p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2156 total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2157 p_fl_seg->real_size_in_page);
2158
2159 for (i = 0; i < total_lines; i++) {
2160 shadow_line = i + p_fl_seg->start_line -
2161 p_hwfn->p_cxt_mngr->pf_start_line;
2162 p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].virt_addr;
2163 }
2164 p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2165 p_fl_seg->real_size_in_page;
2166 p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2167 p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2168 p_info->tid_size;
2169
2170 return 0;
2171}
2172
2173/* This function is very RoCE oriented, if another protocol in the future
2174 * will want this feature we'll need to modify the function to be more generic
2175 */
2176int
2177qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
2178 enum qed_cxt_elem_type elem_type, u32 iid)
2179{
2180 u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2181 struct tdif_task_context *tdif_context;
2182 struct qed_ilt_client_cfg *p_cli;
2183 struct qed_ilt_cli_blk *p_blk;
2184 struct qed_ptt *p_ptt;
2185 dma_addr_t p_phys;
2186 u64 ilt_hw_entry;
2187 void *p_virt;
2188 u32 flags1;
2189 int rc = 0;
2190
2191 switch (elem_type) {
2192 case QED_ELEM_CXT:
2193 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2194 elem_size = CONN_CXT_SIZE(p_hwfn);
2195 p_blk = &p_cli->pf_blks[CDUC_BLK];
2196 break;
2197 case QED_ELEM_SRQ:
2198 /* The first ILT page is not used for regular SRQs. Skip it. */
2199 iid += p_hwfn->p_cxt_mngr->xrc_srq_count;
2200 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2201 elem_size = SRQ_CXT_SIZE;
2202 p_blk = &p_cli->pf_blks[SRQ_BLK];
2203 break;
2204 case QED_ELEM_XRC_SRQ:
2205 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2206 elem_size = XRC_SRQ_CXT_SIZE;
2207 p_blk = &p_cli->pf_blks[SRQ_BLK];
2208 break;
2209 case QED_ELEM_TASK:
2210 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2211 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2212 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2213 break;
2214 default:
2215 DP_NOTICE(p_hwfn, "-EOPNOTSUPP elem type = %d", elem_type);
2216 return -EOPNOTSUPP;
2217 }
2218
2219 /* Calculate line in ilt */
2220 hw_p_size = p_cli->p_size.val;
2221 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2222 line = p_blk->start_line + (iid / elems_per_p);
2223 shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2224
2225 /* If line is already allocated, do nothing, otherwise allocate it and
2226 * write it to the PSWRQ2 registers.
2227 * This section can be run in parallel from different contexts and thus
2228 * a mutex protection is needed.
2229 */
2230
2231 mutex_lock(&p_hwfn->p_cxt_mngr->mutex);
2232
2233 if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].virt_addr)
2234 goto out0;
2235
2236 p_ptt = qed_ptt_acquire(p_hwfn);
2237 if (!p_ptt) {
2238 DP_NOTICE(p_hwfn,
2239 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2240 rc = -EBUSY;
2241 goto out0;
2242 }
2243
2244 p_virt = dma_alloc_coherent(dev: &p_hwfn->cdev->pdev->dev,
2245 size: p_blk->real_size_in_page, dma_handle: &p_phys,
2246 GFP_KERNEL);
2247 if (!p_virt) {
2248 rc = -ENOMEM;
2249 goto out1;
2250 }
2251
2252 /* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2253 * to compensate for a HW bug, but it is configured even if DIF is not
2254 * enabled. This is harmless and allows us to avoid a dedicated API. We
2255 * configure the field for all of the contexts on the newly allocated
2256 * page.
2257 */
2258 if (elem_type == QED_ELEM_TASK) {
2259 u32 elem_i;
2260 u8 *elem_start = (u8 *)p_virt;
2261 union type1_task_context *elem;
2262
2263 for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2264 elem = (union type1_task_context *)elem_start;
2265 tdif_context = &elem->roce_ctx.tdif_context;
2266
2267 flags1 = le32_to_cpu(tdif_context->flags1);
2268 SET_FIELD(flags1, TDIF_TASK_CONTEXT_REF_TAG_MASK, 0xf);
2269 tdif_context->flags1 = cpu_to_le32(flags1);
2270
2271 elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2272 }
2273 }
2274
2275 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].virt_addr = p_virt;
2276 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].phys_addr = p_phys;
2277 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2278 p_blk->real_size_in_page;
2279
2280 /* compute absolute offset */
2281 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2282 (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2283
2284 ilt_hw_entry = 0;
2285 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2286 SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
2287 (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].phys_addr
2288 >> 12));
2289
2290 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2291 qed_dmae_host2grc(p_hwfn, p_ptt, source_addr: (u64) (uintptr_t)&ilt_hw_entry,
2292 grc_addr: reg_offset, size_in_dwords: sizeof(ilt_hw_entry) / sizeof(u32),
2293 NULL);
2294
2295 if (elem_type == QED_ELEM_CXT) {
2296 u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2297 elems_per_p;
2298
2299 /* Update the relevant register in the parser */
2300 qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2301 val: last_cid_allocated - 1);
2302
2303 if (!p_hwfn->b_rdma_enabled_in_prs) {
2304 /* Enable RDMA search */
2305 qed_wr(p_hwfn, p_ptt, hw_addr: p_hwfn->rdma_prs_search_reg, val: 1);
2306 p_hwfn->b_rdma_enabled_in_prs = true;
2307 }
2308 }
2309
2310out1:
2311 qed_ptt_release(p_hwfn, p_ptt);
2312out0:
2313 mutex_unlock(lock: &p_hwfn->p_cxt_mngr->mutex);
2314
2315 return rc;
2316}
2317
2318/* This function is very RoCE oriented, if another protocol in the future
2319 * will want this feature we'll need to modify the function to be more generic
2320 */
2321static int
2322qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn,
2323 enum qed_cxt_elem_type elem_type,
2324 u32 start_iid, u32 count)
2325{
2326 u32 start_line, end_line, shadow_start_line, shadow_end_line;
2327 u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2328 struct qed_ilt_client_cfg *p_cli;
2329 struct qed_ilt_cli_blk *p_blk;
2330 u32 end_iid = start_iid + count;
2331 struct qed_ptt *p_ptt;
2332 u64 ilt_hw_entry = 0;
2333 u32 i;
2334
2335 switch (elem_type) {
2336 case QED_ELEM_CXT:
2337 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2338 elem_size = CONN_CXT_SIZE(p_hwfn);
2339 p_blk = &p_cli->pf_blks[CDUC_BLK];
2340 break;
2341 case QED_ELEM_SRQ:
2342 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2343 elem_size = SRQ_CXT_SIZE;
2344 p_blk = &p_cli->pf_blks[SRQ_BLK];
2345 break;
2346 case QED_ELEM_XRC_SRQ:
2347 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2348 elem_size = XRC_SRQ_CXT_SIZE;
2349 p_blk = &p_cli->pf_blks[SRQ_BLK];
2350 break;
2351 case QED_ELEM_TASK:
2352 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2353 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2354 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2355 break;
2356 default:
2357 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2358 return -EINVAL;
2359 }
2360
2361 /* Calculate line in ilt */
2362 hw_p_size = p_cli->p_size.val;
2363 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2364 start_line = p_blk->start_line + (start_iid / elems_per_p);
2365 end_line = p_blk->start_line + (end_iid / elems_per_p);
2366 if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2367 end_line--;
2368
2369 shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2370 shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2371
2372 p_ptt = qed_ptt_acquire(p_hwfn);
2373 if (!p_ptt) {
2374 DP_NOTICE(p_hwfn,
2375 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2376 return -EBUSY;
2377 }
2378
2379 for (i = shadow_start_line; i < shadow_end_line; i++) {
2380 if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr)
2381 continue;
2382
2383 dma_free_coherent(dev: &p_hwfn->cdev->pdev->dev,
2384 size: p_hwfn->p_cxt_mngr->ilt_shadow[i].size,
2385 cpu_addr: p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr,
2386 dma_handle: p_hwfn->p_cxt_mngr->ilt_shadow[i].phys_addr);
2387
2388 p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr = NULL;
2389 p_hwfn->p_cxt_mngr->ilt_shadow[i].phys_addr = 0;
2390 p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2391
2392 /* compute absolute offset */
2393 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2394 ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2395 ILT_ENTRY_IN_REGS);
2396
2397 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2398 * wide-bus.
2399 */
2400 qed_dmae_host2grc(p_hwfn, p_ptt,
2401 source_addr: (u64) (uintptr_t) &ilt_hw_entry,
2402 grc_addr: reg_offset,
2403 size_in_dwords: sizeof(ilt_hw_entry) / sizeof(u32),
2404 NULL);
2405 }
2406
2407 qed_ptt_release(p_hwfn, p_ptt);
2408
2409 return 0;
2410}
2411
2412int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto)
2413{
2414 int rc;
2415 u32 cid;
2416
2417 /* Free Connection CXT */
2418 rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_CXT,
2419 start_iid: qed_cxt_get_proto_cid_start(p_hwfn,
2420 type: proto),
2421 count: qed_cxt_get_proto_cid_count(p_hwfn,
2422 type: proto, vf_cid: &cid));
2423
2424 if (rc)
2425 return rc;
2426
2427 /* Free Task CXT ( Intentionally RoCE as task-id is shared between
2428 * RoCE and iWARP )
2429 */
2430 proto = PROTOCOLID_ROCE;
2431 rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_TASK, start_iid: 0,
2432 count: qed_cxt_get_proto_tid_count(p_hwfn, type: proto));
2433 if (rc)
2434 return rc;
2435
2436 /* Free TSDM CXT */
2437 rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_XRC_SRQ, start_iid: 0,
2438 count: p_hwfn->p_cxt_mngr->xrc_srq_count);
2439
2440 rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_SRQ,
2441 start_iid: p_hwfn->p_cxt_mngr->xrc_srq_count,
2442 count: p_hwfn->p_cxt_mngr->srq_count);
2443
2444 return rc;
2445}
2446
2447int qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn,
2448 u32 tid, u8 ctx_type, void **pp_task_ctx)
2449{
2450 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2451 struct qed_ilt_client_cfg *p_cli;
2452 struct qed_tid_seg *p_seg_info;
2453 struct qed_ilt_cli_blk *p_seg;
2454 u32 num_tids_per_block;
2455 u32 tid_size, ilt_idx;
2456 u32 total_lines;
2457 u32 proto, seg;
2458
2459 /* Verify the personality */
2460 switch (p_hwfn->hw_info.personality) {
2461 case QED_PCI_FCOE:
2462 proto = PROTOCOLID_FCOE;
2463 seg = QED_CXT_FCOE_TID_SEG;
2464 break;
2465 case QED_PCI_ISCSI:
2466 case QED_PCI_NVMETCP:
2467 proto = PROTOCOLID_TCP_ULP;
2468 seg = QED_CXT_TCP_ULP_TID_SEG;
2469 break;
2470 default:
2471 return -EINVAL;
2472 }
2473
2474 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2475 if (!p_cli->active)
2476 return -EINVAL;
2477
2478 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2479
2480 if (ctx_type == QED_CTX_WORKING_MEM) {
2481 p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2482 } else if (ctx_type == QED_CTX_FL_MEM) {
2483 if (!p_seg_info->has_fl_mem)
2484 return -EINVAL;
2485 p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2486 } else {
2487 return -EINVAL;
2488 }
2489 total_lines = DIV_ROUND_UP(p_seg->total_size, p_seg->real_size_in_page);
2490 tid_size = p_mngr->task_type_size[p_seg_info->type];
2491 num_tids_per_block = p_seg->real_size_in_page / tid_size;
2492
2493 if (total_lines < tid / num_tids_per_block)
2494 return -EINVAL;
2495
2496 ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2497 p_mngr->pf_start_line;
2498 *pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].virt_addr +
2499 (tid % num_tids_per_block) * tid_size;
2500
2501 return 0;
2502}
2503
2504static u16 qed_blk_calculate_pages(struct qed_ilt_cli_blk *p_blk)
2505{
2506 if (p_blk->real_size_in_page == 0)
2507 return 0;
2508
2509 return DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
2510}
2511
2512u16 qed_get_cdut_num_pf_init_pages(struct qed_hwfn *p_hwfn)
2513{
2514 struct qed_ilt_client_cfg *p_cli;
2515 struct qed_ilt_cli_blk *p_blk;
2516 u16 i, pages = 0;
2517
2518 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2519 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
2520 p_blk = &p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)];
2521 pages += qed_blk_calculate_pages(p_blk);
2522 }
2523
2524 return pages;
2525}
2526
2527u16 qed_get_cdut_num_vf_init_pages(struct qed_hwfn *p_hwfn)
2528{
2529 struct qed_ilt_client_cfg *p_cli;
2530 struct qed_ilt_cli_blk *p_blk;
2531 u16 i, pages = 0;
2532
2533 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2534 for (i = 0; i < NUM_TASK_VF_SEGMENTS; i++) {
2535 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(i, VF)];
2536 pages += qed_blk_calculate_pages(p_blk);
2537 }
2538
2539 return pages;
2540}
2541
2542u16 qed_get_cdut_num_pf_work_pages(struct qed_hwfn *p_hwfn)
2543{
2544 struct qed_ilt_client_cfg *p_cli;
2545 struct qed_ilt_cli_blk *p_blk;
2546 u16 i, pages = 0;
2547
2548 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2549 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
2550 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(i)];
2551 pages += qed_blk_calculate_pages(p_blk);
2552 }
2553
2554 return pages;
2555}
2556
2557u16 qed_get_cdut_num_vf_work_pages(struct qed_hwfn *p_hwfn)
2558{
2559 struct qed_ilt_client_cfg *p_cli;
2560 struct qed_ilt_cli_blk *p_blk;
2561 u16 pages = 0, i;
2562
2563 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2564 for (i = 0; i < NUM_TASK_VF_SEGMENTS; i++) {
2565 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(i)];
2566 pages += qed_blk_calculate_pages(p_blk);
2567 }
2568
2569 return pages;
2570}
2571

source code of linux/drivers/net/ethernet/qlogic/qed/qed_cxt.c