1 /* QLogic qed NIC Driver
2 * Copyright (c) 2015 QLogic Corporation
4 * This software is available under the terms of the GNU General Public License
5 * (GPL) Version 2, available from the file COPYING in the main directory of
9 #include <linux/types.h>
10 #include <linux/bitops.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/errno.h>
13 #include <linux/kernel.h>
14 #include <linux/list.h>
15 #include <linux/log2.h>
16 #include <linux/pci.h>
17 #include <linux/slab.h>
18 #include <linux/string.h>
19 #include <linux/bitops.h>
22 #include "qed_dev_api.h"
25 #include "qed_init_ops.h"
26 #include "qed_reg_addr.h"
27 #include "qed_sriov.h"
29 /* Max number of connection types in HW (DQ/CDU etc.) */
30 #define MAX_CONN_TYPES PROTOCOLID_COMMON
31 #define NUM_TASK_TYPES 2
32 #define NUM_TASK_PF_SEGMENTS 4
33 #define NUM_TASK_VF_SEGMENTS 1
36 #define QM_PQ_ELEMENT_SIZE 4 /* in bytes */
38 /* Doorbell-Queue constants */
39 #define DQ_RANGE_SHIFT 4
40 #define DQ_RANGE_ALIGN BIT(DQ_RANGE_SHIFT)
42 /* Searcher constants */
43 #define SRC_MIN_NUM_ELEMS 256
45 /* Timers constants */
47 #define TM_ALIGN BIT(TM_SHIFT)
48 #define TM_ELEM_SIZE 4
51 #define ILT_DEFAULT_HW_P_SIZE 3
52 #define ILT_PAGE_IN_BYTES(hw_p_size) (1U << ((hw_p_size) + 12))
53 #define ILT_CFG_REG(cli, reg) PSWRQ2_REG_ ## cli ## _ ## reg ## _RT_OFFSET
55 /* ILT entry structure */
56 #define ILT_ENTRY_PHY_ADDR_MASK 0x000FFFFFFFFFFFULL
57 #define ILT_ENTRY_PHY_ADDR_SHIFT 0
58 #define ILT_ENTRY_VALID_MASK 0x1ULL
59 #define ILT_ENTRY_VALID_SHIFT 52
60 #define ILT_ENTRY_IN_REGS 2
61 #define ILT_REG_SIZE_IN_BYTES 4
63 /* connection context union */
65 struct core_conn_context core_ctx;
66 struct eth_conn_context eth_ctx;
67 struct iscsi_conn_context iscsi_ctx;
68 struct roce_conn_context roce_ctx;
71 /* TYPE-0 task context - iSCSI */
72 union type0_task_context {
73 struct iscsi_task_context iscsi_ctx;
76 /* TYPE-1 task context - ROCE */
77 union type1_task_context {
78 struct rdma_task_context roce_ctx;
86 #define CDUT_SEG_ALIGNMET 3 /* in 4k chunks */
87 #define CDUT_SEG_ALIGNMET_IN_BYTES (1 << (CDUT_SEG_ALIGNMET + 12))
89 #define CONN_CXT_SIZE(p_hwfn) \
90 ALIGNED_TYPE_SIZE(union conn_context, p_hwfn)
92 #define SRQ_CXT_SIZE (sizeof(struct rdma_srq_context))
94 #define TYPE0_TASK_CXT_SIZE(p_hwfn) \
95 ALIGNED_TYPE_SIZE(union type0_task_context, p_hwfn)
97 /* Alignment is inherent to the type1_task_context structure */
98 #define TYPE1_TASK_CXT_SIZE(p_hwfn) sizeof(union type1_task_context)
100 /* PF per protocl configuration object */
101 #define TASK_SEGMENTS (NUM_TASK_PF_SEGMENTS + NUM_TASK_VF_SEGMENTS)
102 #define TASK_SEGMENT_VF (NUM_TASK_PF_SEGMENTS)
110 struct qed_conn_type_cfg {
114 struct qed_tid_seg tid_seg[TASK_SEGMENTS];
117 /* ILT Client configuration, Per connection type (protocol) resources. */
118 #define ILT_CLI_PF_BLOCKS (1 + NUM_TASK_PF_SEGMENTS * 2)
119 #define ILT_CLI_VF_BLOCKS (1 + NUM_TASK_VF_SEGMENTS * 2)
122 #define CDUT_SEG_BLK(n) (1 + (u8)(n))
123 #define CDUT_FL_SEG_BLK(n, X) (1 + (n) + NUM_TASK_ ## X ## _SEGMENTS)
135 struct ilt_cfg_pair {
140 struct qed_ilt_cli_blk {
141 u32 total_size; /* 0 means not active */
142 u32 real_size_in_page;
144 u32 dynamic_line_cnt;
147 struct qed_ilt_client_cfg {
151 struct ilt_cfg_pair first;
152 struct ilt_cfg_pair last;
153 struct ilt_cfg_pair p_size;
155 /* ILT client blocks for PF */
156 struct qed_ilt_cli_blk pf_blks[ILT_CLI_PF_BLOCKS];
159 /* ILT client blocks for VFs */
160 struct qed_ilt_cli_blk vf_blks[ILT_CLI_VF_BLOCKS];
166 * Protocol acquired CID lists
167 * PF start line in ILT
175 struct qed_cid_acquired_map {
178 unsigned long *cid_map;
181 struct qed_cxt_mngr {
182 /* Per protocl configuration */
183 struct qed_conn_type_cfg conn_cfg[MAX_CONN_TYPES];
185 /* computed ILT structure */
186 struct qed_ilt_client_cfg clients[ILT_CLI_MAX];
188 /* Task type sizes */
189 u32 task_type_size[NUM_TASK_TYPES];
191 /* total number of VFs for this hwfn -
192 * ALL VFs are symmetric in terms of HW resources
196 /* total number of SRQ's for this hwfn */
200 struct qed_cid_acquired_map acquired[MAX_CONN_TYPES];
202 /* ILT shadow table */
203 struct qed_dma_mem *ilt_shadow;
206 /* Mutex for a dynamic ILT allocation */
210 struct qed_dma_mem *t2;
215 static bool src_proto(enum protocol_type type)
217 return type == PROTOCOLID_ISCSI ||
218 type == PROTOCOLID_ROCE;
221 static bool tm_cid_proto(enum protocol_type type)
223 return type == PROTOCOLID_ISCSI ||
224 type == PROTOCOLID_ROCE;
227 /* counts the iids for the CDU/CDUC ILT client configuration */
228 struct qed_cdu_iids {
233 static void qed_cxt_cdu_iids(struct qed_cxt_mngr *p_mngr,
234 struct qed_cdu_iids *iids)
238 for (type = 0; type < MAX_CONN_TYPES; type++) {
239 iids->pf_cids += p_mngr->conn_cfg[type].cid_count;
240 iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
244 /* counts the iids for the Searcher block configuration */
245 struct qed_src_iids {
250 static void qed_cxt_src_iids(struct qed_cxt_mngr *p_mngr,
251 struct qed_src_iids *iids)
255 for (i = 0; i < MAX_CONN_TYPES; i++) {
259 iids->pf_cids += p_mngr->conn_cfg[i].cid_count;
260 iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf;
264 /* counts the iids for the Timers block configuration */
267 u32 pf_tids[NUM_TASK_PF_SEGMENTS]; /* per segment */
273 static void qed_cxt_tm_iids(struct qed_cxt_mngr *p_mngr,
274 struct qed_tm_iids *iids)
278 for (i = 0; i < MAX_CONN_TYPES; i++) {
279 struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
281 if (tm_cid_proto(i)) {
282 iids->pf_cids += p_cfg->cid_count;
283 iids->per_vf_cids += p_cfg->cids_per_vf;
287 iids->pf_cids = roundup(iids->pf_cids, TM_ALIGN);
288 iids->per_vf_cids = roundup(iids->per_vf_cids, TM_ALIGN);
289 iids->per_vf_tids = roundup(iids->per_vf_tids, TM_ALIGN);
291 for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
292 iids->pf_tids[j] = roundup(iids->pf_tids[j], TM_ALIGN);
293 iids->pf_tids_total += iids->pf_tids[j];
297 static void qed_cxt_qm_iids(struct qed_hwfn *p_hwfn,
298 struct qed_qm_iids *iids)
300 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
301 struct qed_tid_seg *segs;
302 u32 vf_cids = 0, type, j;
305 for (type = 0; type < MAX_CONN_TYPES; type++) {
306 iids->cids += p_mngr->conn_cfg[type].cid_count;
307 vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
309 segs = p_mngr->conn_cfg[type].tid_seg;
310 /* for each segment there is at most one
311 * protocol for which count is not 0.
313 for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
314 iids->tids += segs[j].count;
316 /* The last array elelment is for the VFs. As for PF
317 * segments there can be only one protocol for
318 * which this value is not 0.
320 vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
323 iids->vf_cids += vf_cids * p_mngr->vf_count;
324 iids->tids += vf_tids * p_mngr->vf_count;
326 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
327 "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
328 iids->cids, iids->vf_cids, iids->tids, vf_tids);
331 static struct qed_tid_seg *qed_cxt_tid_seg_info(struct qed_hwfn *p_hwfn,
334 struct qed_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
337 /* Find the protocol with tid count > 0 for this segment.
338 * Note: there can only be one and this is already validated.
340 for (i = 0; i < MAX_CONN_TYPES; i++)
341 if (p_cfg->conn_cfg[i].tid_seg[seg].count)
342 return &p_cfg->conn_cfg[i].tid_seg[seg];
346 void qed_cxt_set_srq_count(struct qed_hwfn *p_hwfn, u32 num_srqs)
348 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
350 p_mgr->srq_count = num_srqs;
353 u32 qed_cxt_get_srq_count(struct qed_hwfn *p_hwfn)
355 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
357 return p_mgr->srq_count;
360 /* set the iids count per protocol */
361 static void qed_cxt_set_proto_cid_count(struct qed_hwfn *p_hwfn,
362 enum protocol_type type,
363 u32 cid_count, u32 vf_cid_cnt)
365 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
366 struct qed_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
368 p_conn->cid_count = roundup(cid_count, DQ_RANGE_ALIGN);
369 p_conn->cids_per_vf = roundup(vf_cid_cnt, DQ_RANGE_ALIGN);
371 if (type == PROTOCOLID_ROCE) {
372 u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
373 u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
374 u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
376 p_conn->cid_count = roundup(p_conn->cid_count, elems_per_page);
380 u32 qed_cxt_get_proto_cid_count(struct qed_hwfn *p_hwfn,
381 enum protocol_type type,
385 *vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
387 return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
390 u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
391 enum protocol_type type)
393 return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
396 u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
397 enum protocol_type type)
402 for (i = 0; i < TASK_SEGMENTS; i++)
403 cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
409 qed_cxt_set_proto_tid_count(struct qed_hwfn *p_hwfn,
410 enum protocol_type proto,
411 u8 seg, u8 seg_type, u32 count, bool has_fl)
413 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
414 struct qed_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
416 p_seg->count = count;
417 p_seg->has_fl_mem = has_fl;
418 p_seg->type = seg_type;
421 static void qed_ilt_cli_blk_fill(struct qed_ilt_client_cfg *p_cli,
422 struct qed_ilt_cli_blk *p_blk,
423 u32 start_line, u32 total_size,
426 u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
428 /* verify thatits called only once for each block */
429 if (p_blk->total_size)
432 p_blk->total_size = total_size;
433 p_blk->real_size_in_page = 0;
435 p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
436 p_blk->start_line = start_line;
439 static void qed_ilt_cli_adv_line(struct qed_hwfn *p_hwfn,
440 struct qed_ilt_client_cfg *p_cli,
441 struct qed_ilt_cli_blk *p_blk,
442 u32 *p_line, enum ilt_clients client_id)
444 if (!p_blk->total_size)
448 p_cli->first.val = *p_line;
450 p_cli->active = true;
451 *p_line += DIV_ROUND_UP(p_blk->total_size,
452 p_blk->real_size_in_page);
453 p_cli->last.val = *p_line - 1;
455 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
456 "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
457 client_id, p_cli->first.val,
458 p_cli->last.val, p_blk->total_size,
459 p_blk->real_size_in_page, p_blk->start_line);
462 static u32 qed_ilt_get_dynamic_line_cnt(struct qed_hwfn *p_hwfn,
463 enum ilt_clients ilt_client)
465 u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
466 struct qed_ilt_client_cfg *p_cli;
467 u32 lines_to_skip = 0;
470 if (ilt_client == ILT_CLI_CDUC) {
471 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
473 cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
474 (u32) CONN_CXT_SIZE(p_hwfn);
476 lines_to_skip = cid_count / cxts_per_p;
479 return lines_to_skip;
482 int qed_cxt_cfg_ilt_compute(struct qed_hwfn *p_hwfn)
484 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
485 u32 curr_line, total, i, task_size, line;
486 struct qed_ilt_client_cfg *p_cli;
487 struct qed_ilt_cli_blk *p_blk;
488 struct qed_cdu_iids cdu_iids;
489 struct qed_src_iids src_iids;
490 struct qed_qm_iids qm_iids;
491 struct qed_tm_iids tm_iids;
492 struct qed_tid_seg *p_seg;
494 memset(&qm_iids, 0, sizeof(qm_iids));
495 memset(&cdu_iids, 0, sizeof(cdu_iids));
496 memset(&src_iids, 0, sizeof(src_iids));
497 memset(&tm_iids, 0, sizeof(tm_iids));
499 p_mngr->pf_start_line = RESC_START(p_hwfn, QED_ILT);
501 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
502 "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
503 p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
506 p_cli = &p_mngr->clients[ILT_CLI_CDUC];
507 curr_line = p_mngr->pf_start_line;
510 p_cli->pf_total_lines = 0;
512 /* get the counters for the CDUC and QM clients */
513 qed_cxt_cdu_iids(p_mngr, &cdu_iids);
515 p_blk = &p_cli->pf_blks[CDUC_BLK];
517 total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
519 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
520 total, CONN_CXT_SIZE(p_hwfn));
522 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
523 p_cli->pf_total_lines = curr_line - p_blk->start_line;
525 p_blk->dynamic_line_cnt = qed_ilt_get_dynamic_line_cnt(p_hwfn,
529 p_blk = &p_cli->vf_blks[CDUC_BLK];
530 total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
532 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
533 total, CONN_CXT_SIZE(p_hwfn));
535 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
536 p_cli->vf_total_lines = curr_line - p_blk->start_line;
538 for (i = 1; i < p_mngr->vf_count; i++)
539 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
543 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
544 p_cli->first.val = curr_line;
546 /* first the 'working' task memory */
547 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
548 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
549 if (!p_seg || p_seg->count == 0)
552 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(i)];
553 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
554 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
555 p_mngr->task_type_size[p_seg->type]);
557 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
561 /* next the 'init' task memory (forced load memory) */
562 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
563 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
564 if (!p_seg || p_seg->count == 0)
567 p_blk = &p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)];
569 if (!p_seg->has_fl_mem) {
570 /* The segment is active (total size pf 'working'
571 * memory is > 0) but has no FL (forced-load, Init)
574 * 1. The total-size in the corrsponding FL block of
575 * the ILT client is set to 0 - No ILT line are
576 * provisioned and no ILT memory allocated.
578 * 2. The start-line of said block is set to the
579 * start line of the matching working memory
580 * block in the ILT client. This is later used to
581 * configure the CDU segment offset registers and
582 * results in an FL command for TIDs of this
583 * segement behaves as regular load commands
584 * (loading TIDs from the working memory).
586 line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
588 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
591 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
593 qed_ilt_cli_blk_fill(p_cli, p_blk,
595 p_mngr->task_type_size[p_seg->type]);
597 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
600 p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
603 p_seg = qed_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
604 if (p_seg && p_seg->count) {
605 /* Stricly speaking we need to iterate over all VF
606 * task segment types, but a VF has only 1 segment
609 /* 'working' memory */
610 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
612 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
613 qed_ilt_cli_blk_fill(p_cli, p_blk,
615 p_mngr->task_type_size[p_seg->type]);
617 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
621 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
622 if (!p_seg->has_fl_mem) {
623 /* see comment above */
624 line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
625 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
627 task_size = p_mngr->task_type_size[p_seg->type];
628 qed_ilt_cli_blk_fill(p_cli, p_blk,
629 curr_line, total, task_size);
630 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
633 p_cli->vf_total_lines = curr_line -
634 p_cli->vf_blks[0].start_line;
636 /* Now for the rest of the VFs */
637 for (i = 1; i < p_mngr->vf_count; i++) {
638 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
639 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
642 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
643 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
649 p_cli = &p_mngr->clients[ILT_CLI_QM];
650 p_blk = &p_cli->pf_blks[0];
652 qed_cxt_qm_iids(p_hwfn, &qm_iids);
653 total = qed_qm_pf_mem_size(p_hwfn->rel_pf_id, qm_iids.cids,
654 qm_iids.vf_cids, qm_iids.tids,
655 p_hwfn->qm_info.num_pqs,
656 p_hwfn->qm_info.num_vf_pqs);
660 "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
664 p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
666 qed_ilt_cli_blk_fill(p_cli, p_blk,
667 curr_line, total * 0x1000,
670 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
671 p_cli->pf_total_lines = curr_line - p_blk->start_line;
674 p_cli = &p_mngr->clients[ILT_CLI_SRC];
675 qed_cxt_src_iids(p_mngr, &src_iids);
677 /* Both the PF and VFs searcher connections are stored in the per PF
678 * database. Thus sum the PF searcher cids and all the VFs searcher
681 total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
683 u32 local_max = max_t(u32, total,
686 total = roundup_pow_of_two(local_max);
688 p_blk = &p_cli->pf_blks[0];
689 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
690 total * sizeof(struct src_ent),
691 sizeof(struct src_ent));
693 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
695 p_cli->pf_total_lines = curr_line - p_blk->start_line;
699 p_cli = &p_mngr->clients[ILT_CLI_TM];
700 qed_cxt_tm_iids(p_mngr, &tm_iids);
701 total = tm_iids.pf_cids + tm_iids.pf_tids_total;
703 p_blk = &p_cli->pf_blks[0];
704 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
705 total * TM_ELEM_SIZE, TM_ELEM_SIZE);
707 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
709 p_cli->pf_total_lines = curr_line - p_blk->start_line;
713 total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
715 p_blk = &p_cli->vf_blks[0];
716 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
717 total * TM_ELEM_SIZE, TM_ELEM_SIZE);
719 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
721 p_cli->pf_total_lines = curr_line - p_blk->start_line;
723 for (i = 1; i < p_mngr->vf_count; i++)
724 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
728 /* TSDM (SRQ CONTEXT) */
729 total = qed_cxt_get_srq_count(p_hwfn);
732 p_cli = &p_mngr->clients[ILT_CLI_TSDM];
733 p_blk = &p_cli->pf_blks[SRQ_BLK];
734 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
735 total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
737 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
739 p_cli->pf_total_lines = curr_line - p_blk->start_line;
742 if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
743 RESC_NUM(p_hwfn, QED_ILT)) {
744 DP_ERR(p_hwfn, "too many ilt lines...#lines=%d\n",
745 curr_line - p_hwfn->p_cxt_mngr->pf_start_line);
752 static void qed_cxt_src_t2_free(struct qed_hwfn *p_hwfn)
754 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
760 for (i = 0; i < p_mngr->t2_num_pages; i++)
761 if (p_mngr->t2[i].p_virt)
762 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
764 p_mngr->t2[i].p_virt,
765 p_mngr->t2[i].p_phys);
771 static int qed_cxt_src_t2_alloc(struct qed_hwfn *p_hwfn)
773 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
774 u32 conn_num, total_size, ent_per_page, psz, i;
775 struct qed_ilt_client_cfg *p_src;
776 struct qed_src_iids src_iids;
777 struct qed_dma_mem *p_t2;
780 memset(&src_iids, 0, sizeof(src_iids));
782 /* if the SRC ILT client is inactive - there are no connection
783 * requiring the searcer, leave.
785 p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
789 qed_cxt_src_iids(p_mngr, &src_iids);
790 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
791 total_size = conn_num * sizeof(struct src_ent);
793 /* use the same page size as the SRC ILT client */
794 psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
795 p_mngr->t2_num_pages = DIV_ROUND_UP(total_size, psz);
798 p_mngr->t2 = kzalloc(p_mngr->t2_num_pages * sizeof(struct qed_dma_mem),
801 DP_NOTICE(p_hwfn, "Failed to allocate t2 table\n");
806 /* allocate t2 pages */
807 for (i = 0; i < p_mngr->t2_num_pages; i++) {
808 u32 size = min_t(u32, total_size, psz);
809 void **p_virt = &p_mngr->t2[i].p_virt;
811 *p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
813 &p_mngr->t2[i].p_phys, GFP_KERNEL);
814 if (!p_mngr->t2[i].p_virt) {
818 memset(*p_virt, 0, size);
819 p_mngr->t2[i].size = size;
823 /* Set the t2 pointers */
825 /* entries per page - must be a power of two */
826 ent_per_page = psz / sizeof(struct src_ent);
828 p_mngr->first_free = (u64) p_mngr->t2[0].p_phys;
830 p_t2 = &p_mngr->t2[(conn_num - 1) / ent_per_page];
831 p_mngr->last_free = (u64) p_t2->p_phys +
832 ((conn_num - 1) & (ent_per_page - 1)) * sizeof(struct src_ent);
834 for (i = 0; i < p_mngr->t2_num_pages; i++) {
835 u32 ent_num = min_t(u32,
838 struct src_ent *entries = p_mngr->t2[i].p_virt;
839 u64 p_ent_phys = (u64) p_mngr->t2[i].p_phys, val;
842 for (j = 0; j < ent_num - 1; j++) {
843 val = p_ent_phys + (j + 1) * sizeof(struct src_ent);
844 entries[j].next = cpu_to_be64(val);
847 if (i < p_mngr->t2_num_pages - 1)
848 val = (u64) p_mngr->t2[i + 1].p_phys;
851 entries[j].next = cpu_to_be64(val);
859 qed_cxt_src_t2_free(p_hwfn);
863 #define for_each_ilt_valid_client(pos, clients) \
864 for (pos = 0; pos < ILT_CLI_MAX; pos++) \
865 if (!clients[pos].active) { \
869 /* Total number of ILT lines used by this PF */
870 static u32 qed_cxt_ilt_shadow_size(struct qed_ilt_client_cfg *ilt_clients)
875 for_each_ilt_valid_client(i, ilt_clients)
876 size += (ilt_clients[i].last.val - ilt_clients[i].first.val + 1);
881 static void qed_ilt_shadow_free(struct qed_hwfn *p_hwfn)
883 struct qed_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
884 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
887 ilt_size = qed_cxt_ilt_shadow_size(p_cli);
889 for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
890 struct qed_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
893 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
894 p_dma->size, p_dma->p_virt,
896 p_dma->p_virt = NULL;
898 kfree(p_mngr->ilt_shadow);
901 static int qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn,
902 struct qed_ilt_cli_blk *p_blk,
903 enum ilt_clients ilt_client,
904 u32 start_line_offset)
906 struct qed_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
907 u32 lines, line, sz_left, lines_to_skip = 0;
909 /* Special handling for RoCE that supports dynamic allocation */
910 if ((p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE) &&
911 ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
914 lines_to_skip = p_blk->dynamic_line_cnt;
916 if (!p_blk->total_size)
919 sz_left = p_blk->total_size;
920 lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) - lines_to_skip;
921 line = p_blk->start_line + start_line_offset -
922 p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
924 for (; lines; lines--) {
929 size = min_t(u32, sz_left,
930 p_blk->real_size_in_page);
931 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
937 memset(p_virt, 0, size);
939 ilt_shadow[line].p_phys = p_phys;
940 ilt_shadow[line].p_virt = p_virt;
941 ilt_shadow[line].size = size;
943 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
944 "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
945 line, (u64)p_phys, p_virt, size);
954 static int qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn)
956 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
957 struct qed_ilt_client_cfg *clients = p_mngr->clients;
958 struct qed_ilt_cli_blk *p_blk;
962 size = qed_cxt_ilt_shadow_size(clients);
963 p_mngr->ilt_shadow = kcalloc(size, sizeof(struct qed_dma_mem),
965 if (!p_mngr->ilt_shadow) {
966 DP_NOTICE(p_hwfn, "Failed to allocate ilt shadow table\n");
968 goto ilt_shadow_fail;
971 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
972 "Allocated 0x%x bytes for ilt shadow\n",
973 (u32)(size * sizeof(struct qed_dma_mem)));
975 for_each_ilt_valid_client(i, clients) {
976 for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
977 p_blk = &clients[i].pf_blks[j];
978 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
980 goto ilt_shadow_fail;
982 for (k = 0; k < p_mngr->vf_count; k++) {
983 for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
984 u32 lines = clients[i].vf_total_lines * k;
986 p_blk = &clients[i].vf_blks[j];
987 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, lines);
989 goto ilt_shadow_fail;
997 qed_ilt_shadow_free(p_hwfn);
1001 static void qed_cid_map_free(struct qed_hwfn *p_hwfn)
1003 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1006 for (type = 0; type < MAX_CONN_TYPES; type++) {
1007 kfree(p_mngr->acquired[type].cid_map);
1008 p_mngr->acquired[type].max_count = 0;
1009 p_mngr->acquired[type].start_cid = 0;
1013 static int qed_cid_map_alloc(struct qed_hwfn *p_hwfn)
1015 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1019 for (type = 0; type < MAX_CONN_TYPES; type++) {
1020 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1026 size = DIV_ROUND_UP(cid_cnt,
1027 sizeof(unsigned long) * BITS_PER_BYTE) *
1028 sizeof(unsigned long);
1029 p_mngr->acquired[type].cid_map = kzalloc(size, GFP_KERNEL);
1030 if (!p_mngr->acquired[type].cid_map)
1033 p_mngr->acquired[type].max_count = cid_cnt;
1034 p_mngr->acquired[type].start_cid = start_cid;
1036 p_hwfn->p_cxt_mngr->conn_cfg[type].cid_start = start_cid;
1038 DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1039 "Type %08x start: %08x count %08x\n",
1040 type, p_mngr->acquired[type].start_cid,
1041 p_mngr->acquired[type].max_count);
1042 start_cid += cid_cnt;
1048 qed_cid_map_free(p_hwfn);
1052 int qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn)
1054 struct qed_ilt_client_cfg *clients;
1055 struct qed_cxt_mngr *p_mngr;
1058 p_mngr = kzalloc(sizeof(*p_mngr), GFP_KERNEL);
1060 DP_NOTICE(p_hwfn, "Failed to allocate `struct qed_cxt_mngr'\n");
1064 /* Initialize ILT client registers */
1065 clients = p_mngr->clients;
1066 clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1067 clients[ILT_CLI_CDUC].last.reg = ILT_CFG_REG(CDUC, LAST_ILT);
1068 clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1070 clients[ILT_CLI_QM].first.reg = ILT_CFG_REG(QM, FIRST_ILT);
1071 clients[ILT_CLI_QM].last.reg = ILT_CFG_REG(QM, LAST_ILT);
1072 clients[ILT_CLI_QM].p_size.reg = ILT_CFG_REG(QM, P_SIZE);
1074 clients[ILT_CLI_TM].first.reg = ILT_CFG_REG(TM, FIRST_ILT);
1075 clients[ILT_CLI_TM].last.reg = ILT_CFG_REG(TM, LAST_ILT);
1076 clients[ILT_CLI_TM].p_size.reg = ILT_CFG_REG(TM, P_SIZE);
1078 clients[ILT_CLI_SRC].first.reg = ILT_CFG_REG(SRC, FIRST_ILT);
1079 clients[ILT_CLI_SRC].last.reg = ILT_CFG_REG(SRC, LAST_ILT);
1080 clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1082 clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1083 clients[ILT_CLI_CDUT].last.reg = ILT_CFG_REG(CDUT, LAST_ILT);
1084 clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1086 clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1087 clients[ILT_CLI_TSDM].last.reg = ILT_CFG_REG(TSDM, LAST_ILT);
1088 clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1089 /* default ILT page size for all clients is 32K */
1090 for (i = 0; i < ILT_CLI_MAX; i++)
1091 p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE;
1093 /* Initialize task sizes */
1094 p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1095 p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1097 if (p_hwfn->cdev->p_iov_info)
1098 p_mngr->vf_count = p_hwfn->cdev->p_iov_info->total_vfs;
1099 /* Initialize the dynamic ILT allocation mutex */
1100 mutex_init(&p_mngr->mutex);
1102 /* Set the cxt mangr pointer priori to further allocations */
1103 p_hwfn->p_cxt_mngr = p_mngr;
1108 int qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn)
1112 /* Allocate the ILT shadow table */
1113 rc = qed_ilt_shadow_alloc(p_hwfn);
1115 DP_NOTICE(p_hwfn, "Failed to allocate ilt memory\n");
1116 goto tables_alloc_fail;
1119 /* Allocate the T2 table */
1120 rc = qed_cxt_src_t2_alloc(p_hwfn);
1122 DP_NOTICE(p_hwfn, "Failed to allocate T2 memory\n");
1123 goto tables_alloc_fail;
1126 /* Allocate and initialize the acquired cids bitmaps */
1127 rc = qed_cid_map_alloc(p_hwfn);
1129 DP_NOTICE(p_hwfn, "Failed to allocate cid maps\n");
1130 goto tables_alloc_fail;
1136 qed_cxt_mngr_free(p_hwfn);
1140 void qed_cxt_mngr_free(struct qed_hwfn *p_hwfn)
1142 if (!p_hwfn->p_cxt_mngr)
1145 qed_cid_map_free(p_hwfn);
1146 qed_cxt_src_t2_free(p_hwfn);
1147 qed_ilt_shadow_free(p_hwfn);
1148 kfree(p_hwfn->p_cxt_mngr);
1150 p_hwfn->p_cxt_mngr = NULL;
1153 void qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn)
1155 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1158 /* Reset acquired cids */
1159 for (type = 0; type < MAX_CONN_TYPES; type++) {
1160 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1165 memset(p_mngr->acquired[type].cid_map, 0,
1166 DIV_ROUND_UP(cid_cnt,
1167 sizeof(unsigned long) * BITS_PER_BYTE) *
1168 sizeof(unsigned long));
1173 #define CDUC_CXT_SIZE_SHIFT \
1174 CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1176 #define CDUC_CXT_SIZE_MASK \
1177 (CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1179 #define CDUC_BLOCK_WASTE_SHIFT \
1180 CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1182 #define CDUC_BLOCK_WASTE_MASK \
1183 (CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1185 #define CDUC_NCIB_SHIFT \
1186 CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1188 #define CDUC_NCIB_MASK \
1189 (CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1191 #define CDUT_TYPE0_CXT_SIZE_SHIFT \
1192 CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1194 #define CDUT_TYPE0_CXT_SIZE_MASK \
1195 (CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >> \
1196 CDUT_TYPE0_CXT_SIZE_SHIFT)
1198 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT \
1199 CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1201 #define CDUT_TYPE0_BLOCK_WASTE_MASK \
1202 (CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >> \
1203 CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1205 #define CDUT_TYPE0_NCIB_SHIFT \
1206 CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1208 #define CDUT_TYPE0_NCIB_MASK \
1209 (CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >> \
1210 CDUT_TYPE0_NCIB_SHIFT)
1212 #define CDUT_TYPE1_CXT_SIZE_SHIFT \
1213 CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1215 #define CDUT_TYPE1_CXT_SIZE_MASK \
1216 (CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >> \
1217 CDUT_TYPE1_CXT_SIZE_SHIFT)
1219 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT \
1220 CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1222 #define CDUT_TYPE1_BLOCK_WASTE_MASK \
1223 (CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >> \
1224 CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1226 #define CDUT_TYPE1_NCIB_SHIFT \
1227 CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1229 #define CDUT_TYPE1_NCIB_MASK \
1230 (CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >> \
1231 CDUT_TYPE1_NCIB_SHIFT)
1233 static void qed_cdu_init_common(struct qed_hwfn *p_hwfn)
1235 u32 page_sz, elems_per_page, block_waste, cxt_size, cdu_params = 0;
1237 /* CDUC - connection configuration */
1238 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1239 cxt_size = CONN_CXT_SIZE(p_hwfn);
1240 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1241 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1243 SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1244 SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1245 SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
1246 STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1248 /* CDUT - type-0 tasks configuration */
1249 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1250 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1251 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1252 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1254 /* cxt size and block-waste are multipes of 8 */
1256 SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1257 SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1258 SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1259 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1261 /* CDUT - type-1 tasks configuration */
1262 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1263 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1264 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1266 /* cxt size and block-waste are multipes of 8 */
1268 SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1269 SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1270 SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1271 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1275 #define CDU_SEG_REG_TYPE_SHIFT CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1276 #define CDU_SEG_REG_TYPE_MASK 0x1
1277 #define CDU_SEG_REG_OFFSET_SHIFT 0
1278 #define CDU_SEG_REG_OFFSET_MASK CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1280 static void qed_cdu_init_pf(struct qed_hwfn *p_hwfn)
1282 struct qed_ilt_client_cfg *p_cli;
1283 struct qed_tid_seg *p_seg;
1284 u32 cdu_seg_params, offset;
1287 static const u32 rt_type_offset_arr[] = {
1288 CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1289 CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1290 CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1291 CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1294 static const u32 rt_type_offset_fl_arr[] = {
1295 CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1296 CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1297 CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1298 CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1301 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1303 /* There are initializations only for CDUT during pf Phase */
1304 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1306 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
1310 /* Note: start_line is already adjusted for the CDU
1311 * segment register granularity, so we just need to
1312 * divide. Adjustment is implicit as we assume ILT
1313 * Page size is larger than 32K!
1315 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1316 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1317 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1320 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1321 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1322 STORE_RT_REG(p_hwfn, rt_type_offset_arr[i], cdu_seg_params);
1324 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1325 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1326 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1329 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1330 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1331 STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i], cdu_seg_params);
1335 void qed_qm_init_pf(struct qed_hwfn *p_hwfn)
1337 struct qed_qm_pf_rt_init_params params;
1338 struct qed_qm_info *qm_info = &p_hwfn->qm_info;
1339 struct qed_qm_iids iids;
1341 memset(&iids, 0, sizeof(iids));
1342 qed_cxt_qm_iids(p_hwfn, &iids);
1344 memset(¶ms, 0, sizeof(params));
1345 params.port_id = p_hwfn->port_id;
1346 params.pf_id = p_hwfn->rel_pf_id;
1347 params.max_phys_tcs_per_port = qm_info->max_phys_tcs_per_port;
1348 params.is_first_pf = p_hwfn->first_on_engine;
1349 params.num_pf_cids = iids.cids;
1350 params.num_vf_cids = iids.vf_cids;
1351 params.start_pq = qm_info->start_pq;
1352 params.num_pf_pqs = qm_info->num_pqs - qm_info->num_vf_pqs;
1353 params.num_vf_pqs = qm_info->num_vf_pqs;
1354 params.start_vport = qm_info->start_vport;
1355 params.num_vports = qm_info->num_vports;
1356 params.pf_wfq = qm_info->pf_wfq;
1357 params.pf_rl = qm_info->pf_rl;
1358 params.pq_params = qm_info->qm_pq_params;
1359 params.vport_params = qm_info->qm_vport_params;
1361 qed_qm_pf_rt_init(p_hwfn, p_hwfn->p_main_ptt, ¶ms);
1365 static int qed_cm_init_pf(struct qed_hwfn *p_hwfn)
1367 union qed_qm_pq_params pq_params;
1370 /* XCM pure-LB queue */
1371 memset(&pq_params, 0, sizeof(pq_params));
1372 pq_params.core.tc = LB_TC;
1373 pq = qed_get_qm_pq(p_hwfn, PROTOCOLID_CORE, &pq_params);
1374 STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET, pq);
1380 static void qed_dq_init_pf(struct qed_hwfn *p_hwfn)
1382 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1383 u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1385 dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1386 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1388 dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1389 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1391 dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1392 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1394 dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1395 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1397 dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1398 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1400 dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1401 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1403 dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1404 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1406 dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1407 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1409 dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1410 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1412 dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1413 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1415 dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1416 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1418 dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1419 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1421 /* Connection types 6 & 7 are not in use, yet they must be configured
1422 * as the highest possible connection. Not configuring them means the
1423 * defaults will be used, and with a large number of cids a bug may
1424 * occur, if the defaults will be smaller than dq_pf_max_cid /
1427 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1428 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1430 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1431 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1434 static void qed_ilt_bounds_init(struct qed_hwfn *p_hwfn)
1436 struct qed_ilt_client_cfg *ilt_clients;
1439 ilt_clients = p_hwfn->p_cxt_mngr->clients;
1440 for_each_ilt_valid_client(i, ilt_clients) {
1441 STORE_RT_REG(p_hwfn,
1442 ilt_clients[i].first.reg,
1443 ilt_clients[i].first.val);
1444 STORE_RT_REG(p_hwfn,
1445 ilt_clients[i].last.reg, ilt_clients[i].last.val);
1446 STORE_RT_REG(p_hwfn,
1447 ilt_clients[i].p_size.reg,
1448 ilt_clients[i].p_size.val);
1452 static void qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn)
1454 struct qed_ilt_client_cfg *p_cli;
1457 /* For simplicty we set the 'block' to be an ILT page */
1458 if (p_hwfn->cdev->p_iov_info) {
1459 struct qed_hw_sriov_info *p_iov = p_hwfn->cdev->p_iov_info;
1461 STORE_RT_REG(p_hwfn,
1462 PSWRQ2_REG_VF_BASE_RT_OFFSET,
1463 p_iov->first_vf_in_pf);
1464 STORE_RT_REG(p_hwfn,
1465 PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1466 p_iov->first_vf_in_pf + p_iov->total_vfs);
1469 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1470 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1471 if (p_cli->active) {
1472 STORE_RT_REG(p_hwfn,
1473 PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1475 STORE_RT_REG(p_hwfn,
1476 PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1477 p_cli->pf_total_lines);
1478 STORE_RT_REG(p_hwfn,
1479 PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1480 p_cli->vf_total_lines);
1483 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1484 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1485 if (p_cli->active) {
1486 STORE_RT_REG(p_hwfn,
1487 PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1489 STORE_RT_REG(p_hwfn,
1490 PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1491 p_cli->pf_total_lines);
1492 STORE_RT_REG(p_hwfn,
1493 PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1494 p_cli->vf_total_lines);
1497 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1498 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1499 if (p_cli->active) {
1500 STORE_RT_REG(p_hwfn,
1501 PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET, blk_factor);
1502 STORE_RT_REG(p_hwfn,
1503 PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1504 p_cli->pf_total_lines);
1505 STORE_RT_REG(p_hwfn,
1506 PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1507 p_cli->vf_total_lines);
1511 /* ILT (PSWRQ2) PF */
1512 static void qed_ilt_init_pf(struct qed_hwfn *p_hwfn)
1514 struct qed_ilt_client_cfg *clients;
1515 struct qed_cxt_mngr *p_mngr;
1516 struct qed_dma_mem *p_shdw;
1517 u32 line, rt_offst, i;
1519 qed_ilt_bounds_init(p_hwfn);
1520 qed_ilt_vf_bounds_init(p_hwfn);
1522 p_mngr = p_hwfn->p_cxt_mngr;
1523 p_shdw = p_mngr->ilt_shadow;
1524 clients = p_hwfn->p_cxt_mngr->clients;
1526 for_each_ilt_valid_client(i, clients) {
1527 /** Client's 1st val and RT array are absolute, ILT shadows'
1528 * lines are relative.
1530 line = clients[i].first.val - p_mngr->pf_start_line;
1531 rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1532 clients[i].first.val * ILT_ENTRY_IN_REGS;
1534 for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1535 line++, rt_offst += ILT_ENTRY_IN_REGS) {
1536 u64 ilt_hw_entry = 0;
1538 /** p_virt could be NULL incase of dynamic
1541 if (p_shdw[line].p_virt) {
1542 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1543 SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1544 (p_shdw[line].p_phys >> 12));
1546 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1547 "Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1549 (u64)(p_shdw[line].p_phys >> 12));
1552 STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1557 /* SRC (Searcher) PF */
1558 static void qed_src_init_pf(struct qed_hwfn *p_hwfn)
1560 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1561 u32 rounded_conn_num, conn_num, conn_max;
1562 struct qed_src_iids src_iids;
1564 memset(&src_iids, 0, sizeof(src_iids));
1565 qed_cxt_src_iids(p_mngr, &src_iids);
1566 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1570 conn_max = max_t(u32, conn_num, SRC_MIN_NUM_ELEMS);
1571 rounded_conn_num = roundup_pow_of_two(conn_max);
1573 STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1574 STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1575 ilog2(rounded_conn_num));
1577 STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1578 p_hwfn->p_cxt_mngr->first_free);
1579 STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1580 p_hwfn->p_cxt_mngr->last_free);
1584 #define TM_CFG_NUM_IDS_SHIFT 0
1585 #define TM_CFG_NUM_IDS_MASK 0xFFFFULL
1586 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT 16
1587 #define TM_CFG_PRE_SCAN_OFFSET_MASK 0x1FFULL
1588 #define TM_CFG_PARENT_PF_SHIFT 25
1589 #define TM_CFG_PARENT_PF_MASK 0x7ULL
1591 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT 30
1592 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK 0x1FFULL
1594 #define TM_CFG_TID_OFFSET_SHIFT 30
1595 #define TM_CFG_TID_OFFSET_MASK 0x7FFFFULL
1596 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT 49
1597 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK 0x1FFULL
1599 static void qed_tm_init_pf(struct qed_hwfn *p_hwfn)
1601 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1602 u32 active_seg_mask = 0, tm_offset, rt_reg;
1603 struct qed_tm_iids tm_iids;
1607 memset(&tm_iids, 0, sizeof(tm_iids));
1608 qed_cxt_tm_iids(p_mngr, &tm_iids);
1610 /* @@@TBD No pre-scan for now */
1612 /* Note: We assume consecutive VFs for a PF */
1613 for (i = 0; i < p_mngr->vf_count; i++) {
1615 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1616 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1617 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1618 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0);
1619 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1620 (sizeof(cfg_word) / sizeof(u32)) *
1621 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1622 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1626 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1627 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1628 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0); /* n/a for PF */
1629 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1631 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1632 (sizeof(cfg_word) / sizeof(u32)) *
1633 (NUM_OF_VFS(p_hwfn->cdev) + p_hwfn->rel_pf_id);
1634 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1637 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1638 tm_iids.pf_cids ? 0x1 : 0x0);
1640 /* @@@TBD how to enable the scan for the VFs */
1642 tm_offset = tm_iids.per_vf_cids;
1644 /* Note: We assume consecutive VFs for a PF */
1645 for (i = 0; i < p_mngr->vf_count; i++) {
1647 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1648 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1649 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1650 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1651 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1653 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1654 (sizeof(cfg_word) / sizeof(u32)) *
1655 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1657 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1660 tm_offset = tm_iids.pf_cids;
1661 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1663 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1664 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1665 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1666 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1667 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1669 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1670 (sizeof(cfg_word) / sizeof(u32)) *
1671 (NUM_OF_VFS(p_hwfn->cdev) +
1672 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1674 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1675 active_seg_mask |= (tm_iids.pf_tids[i] ? (1 << i) : 0);
1677 tm_offset += tm_iids.pf_tids[i];
1680 if (p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE)
1681 active_seg_mask = 0;
1683 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1685 /* @@@TBD how to enable the scan for the VFs */
1688 void qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn)
1690 qed_cdu_init_common(p_hwfn);
1693 void qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn)
1695 qed_qm_init_pf(p_hwfn);
1696 qed_cm_init_pf(p_hwfn);
1697 qed_dq_init_pf(p_hwfn);
1698 qed_cdu_init_pf(p_hwfn);
1699 qed_ilt_init_pf(p_hwfn);
1700 qed_src_init_pf(p_hwfn);
1701 qed_tm_init_pf(p_hwfn);
1704 int qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1705 enum protocol_type type,
1708 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1711 if (type >= MAX_CONN_TYPES || !p_mngr->acquired[type].cid_map) {
1712 DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1716 rel_cid = find_first_zero_bit(p_mngr->acquired[type].cid_map,
1717 p_mngr->acquired[type].max_count);
1719 if (rel_cid >= p_mngr->acquired[type].max_count) {
1720 DP_NOTICE(p_hwfn, "no CID available for protocol %d\n",
1725 __set_bit(rel_cid, p_mngr->acquired[type].cid_map);
1727 *p_cid = rel_cid + p_mngr->acquired[type].start_cid;
1732 static bool qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn,
1734 enum protocol_type *p_type)
1736 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1737 struct qed_cid_acquired_map *p_map;
1738 enum protocol_type p;
1741 /* Iterate over protocols and find matching cid range */
1742 for (p = 0; p < MAX_CONN_TYPES; p++) {
1743 p_map = &p_mngr->acquired[p];
1745 if (!p_map->cid_map)
1747 if (cid >= p_map->start_cid &&
1748 cid < p_map->start_cid + p_map->max_count)
1753 if (p == MAX_CONN_TYPES) {
1754 DP_NOTICE(p_hwfn, "Invalid CID %d", cid);
1758 rel_cid = cid - p_map->start_cid;
1759 if (!test_bit(rel_cid, p_map->cid_map)) {
1760 DP_NOTICE(p_hwfn, "CID %d not acquired", cid);
1766 void qed_cxt_release_cid(struct qed_hwfn *p_hwfn,
1769 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1770 enum protocol_type type;
1774 /* Test acquired and find matching per-protocol map */
1775 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid, &type);
1780 rel_cid = cid - p_mngr->acquired[type].start_cid;
1781 __clear_bit(rel_cid, p_mngr->acquired[type].cid_map);
1784 int qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn,
1785 struct qed_cxt_info *p_info)
1787 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1788 u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
1789 enum protocol_type type;
1792 /* Test acquired and find matching per-protocol map */
1793 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, p_info->iid, &type);
1798 /* set the protocl type */
1799 p_info->type = type;
1801 /* compute context virtual pointer */
1802 hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1804 conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
1805 cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
1806 line = p_info->iid / cxts_per_p;
1808 /* Make sure context is allocated (dynamic allocation) */
1809 if (!p_mngr->ilt_shadow[line].p_virt)
1812 p_info->p_cxt = p_mngr->ilt_shadow[line].p_virt +
1813 p_info->iid % cxts_per_p * conn_cxt_size;
1815 DP_VERBOSE(p_hwfn, (QED_MSG_ILT | QED_MSG_CXT),
1816 "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
1817 p_info->iid / cxts_per_p, p_info->p_cxt, p_info->iid);
1822 void qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn,
1823 struct qed_rdma_pf_params *p_params)
1825 u32 num_cons, num_tasks, num_qps, num_mrs, num_srqs;
1826 enum protocol_type proto;
1828 num_mrs = min_t(u32, RDMA_MAX_TIDS, p_params->num_mrs);
1829 num_tasks = num_mrs; /* each mr uses a single task id */
1830 num_srqs = min_t(u32, 32 * 1024, p_params->num_srqs);
1832 switch (p_hwfn->hw_info.personality) {
1833 case QED_PCI_ETH_ROCE:
1834 num_qps = min_t(u32, ROCE_MAX_QPS, p_params->num_qps);
1835 num_cons = num_qps * 2; /* each QP requires two connections */
1836 proto = PROTOCOLID_ROCE;
1842 if (num_cons && num_tasks) {
1843 qed_cxt_set_proto_cid_count(p_hwfn, proto, num_cons, 0);
1845 /* Deliberatly passing ROCE for tasks id. This is because
1846 * iWARP / RoCE share the task id.
1848 qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
1849 QED_CXT_ROCE_TID_SEG, 1,
1851 qed_cxt_set_srq_count(p_hwfn, num_srqs);
1853 DP_INFO(p_hwfn->cdev,
1854 "RDMA personality used without setting params!\n");
1858 int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn)
1860 /* Set the number of required CORE connections */
1861 u32 core_cids = 1; /* SPQ */
1863 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
1865 switch (p_hwfn->hw_info.personality) {
1866 case QED_PCI_ETH_ROCE:
1868 qed_rdma_set_pf_params(p_hwfn,
1870 pf_params.rdma_pf_params);
1871 /* no need for break since RoCE coexist with Ethernet */
1875 struct qed_eth_pf_params *p_params =
1876 &p_hwfn->pf_params.eth_pf_params;
1878 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
1879 p_params->num_cons, 1);
1884 struct qed_iscsi_pf_params *p_params;
1886 p_params = &p_hwfn->pf_params.iscsi_pf_params;
1888 if (p_params->num_cons && p_params->num_tasks) {
1889 qed_cxt_set_proto_cid_count(p_hwfn,
1894 qed_cxt_set_proto_tid_count(p_hwfn,
1896 QED_CXT_ISCSI_TID_SEG,
1898 p_params->num_tasks,
1901 DP_INFO(p_hwfn->cdev,
1902 "Iscsi personality used without setting params!\n");
1913 int qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn,
1914 struct qed_tid_mem *p_info)
1916 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1917 u32 proto, seg, total_lines, i, shadow_line;
1918 struct qed_ilt_client_cfg *p_cli;
1919 struct qed_ilt_cli_blk *p_fl_seg;
1920 struct qed_tid_seg *p_seg_info;
1922 /* Verify the personality */
1923 switch (p_hwfn->hw_info.personality) {
1925 proto = PROTOCOLID_ISCSI;
1926 seg = QED_CXT_ISCSI_TID_SEG;
1932 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
1936 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
1937 if (!p_seg_info->has_fl_mem)
1940 p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
1941 total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
1942 p_fl_seg->real_size_in_page);
1944 for (i = 0; i < total_lines; i++) {
1945 shadow_line = i + p_fl_seg->start_line -
1946 p_hwfn->p_cxt_mngr->pf_start_line;
1947 p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
1949 p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
1950 p_fl_seg->real_size_in_page;
1951 p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
1952 p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
1958 /* This function is very RoCE oriented, if another protocol in the future
1959 * will want this feature we'll need to modify the function to be more generic
1962 qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
1963 enum qed_cxt_elem_type elem_type, u32 iid)
1965 u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
1966 struct qed_ilt_client_cfg *p_cli;
1967 struct qed_ilt_cli_blk *p_blk;
1968 struct qed_ptt *p_ptt;
1974 switch (elem_type) {
1976 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1977 elem_size = CONN_CXT_SIZE(p_hwfn);
1978 p_blk = &p_cli->pf_blks[CDUC_BLK];
1981 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
1982 elem_size = SRQ_CXT_SIZE;
1983 p_blk = &p_cli->pf_blks[SRQ_BLK];
1986 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1987 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
1988 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
1991 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
1995 /* Calculate line in ilt */
1996 hw_p_size = p_cli->p_size.val;
1997 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
1998 line = p_blk->start_line + (iid / elems_per_p);
1999 shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2001 /* If line is already allocated, do nothing, otherwise allocate it and
2002 * write it to the PSWRQ2 registers.
2003 * This section can be run in parallel from different contexts and thus
2004 * a mutex protection is needed.
2007 mutex_lock(&p_hwfn->p_cxt_mngr->mutex);
2009 if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2012 p_ptt = qed_ptt_acquire(p_hwfn);
2015 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2020 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2021 p_blk->real_size_in_page,
2022 &p_phys, GFP_KERNEL);
2027 memset(p_virt, 0, p_blk->real_size_in_page);
2029 /* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2030 * to compensate for a HW bug, but it is configured even if DIF is not
2031 * enabled. This is harmless and allows us to avoid a dedicated API. We
2032 * configure the field for all of the contexts on the newly allocated
2035 if (elem_type == QED_ELEM_TASK) {
2037 u8 *elem_start = (u8 *)p_virt;
2038 union type1_task_context *elem;
2040 for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2041 elem = (union type1_task_context *)elem_start;
2042 SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2043 TDIF_TASK_CONTEXT_REFTAGMASK, 0xf);
2044 elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2048 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2049 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2050 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2051 p_blk->real_size_in_page;
2053 /* compute absolute offset */
2054 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2055 (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2058 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2059 SET_FIELD(ilt_hw_entry,
2061 (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2063 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2064 qed_dmae_host2grc(p_hwfn, p_ptt, (u64) (uintptr_t)&ilt_hw_entry,
2065 reg_offset, sizeof(ilt_hw_entry) / sizeof(u32), 0);
2067 if (elem_type == QED_ELEM_CXT) {
2068 u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2071 /* Update the relevant register in the parser */
2072 qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2073 last_cid_allocated - 1);
2075 if (!p_hwfn->b_rdma_enabled_in_prs) {
2076 /* Enable RoCE search */
2077 qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2078 p_hwfn->b_rdma_enabled_in_prs = true;
2083 qed_ptt_release(p_hwfn, p_ptt);
2085 mutex_unlock(&p_hwfn->p_cxt_mngr->mutex);
2090 /* This function is very RoCE oriented, if another protocol in the future
2091 * will want this feature we'll need to modify the function to be more generic
2094 qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn,
2095 enum qed_cxt_elem_type elem_type,
2096 u32 start_iid, u32 count)
2098 u32 start_line, end_line, shadow_start_line, shadow_end_line;
2099 u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2100 struct qed_ilt_client_cfg *p_cli;
2101 struct qed_ilt_cli_blk *p_blk;
2102 u32 end_iid = start_iid + count;
2103 struct qed_ptt *p_ptt;
2104 u64 ilt_hw_entry = 0;
2107 switch (elem_type) {
2109 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2110 elem_size = CONN_CXT_SIZE(p_hwfn);
2111 p_blk = &p_cli->pf_blks[CDUC_BLK];
2114 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2115 elem_size = SRQ_CXT_SIZE;
2116 p_blk = &p_cli->pf_blks[SRQ_BLK];
2119 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2120 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2121 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2124 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2128 /* Calculate line in ilt */
2129 hw_p_size = p_cli->p_size.val;
2130 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2131 start_line = p_blk->start_line + (start_iid / elems_per_p);
2132 end_line = p_blk->start_line + (end_iid / elems_per_p);
2133 if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2136 shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2137 shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2139 p_ptt = qed_ptt_acquire(p_hwfn);
2142 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2146 for (i = shadow_start_line; i < shadow_end_line; i++) {
2147 if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2150 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
2151 p_hwfn->p_cxt_mngr->ilt_shadow[i].size,
2152 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2153 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys);
2155 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = NULL;
2156 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2157 p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2159 /* compute absolute offset */
2160 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2161 ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2164 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2167 qed_dmae_host2grc(p_hwfn, p_ptt,
2168 (u64) (uintptr_t) &ilt_hw_entry,
2170 sizeof(ilt_hw_entry) / sizeof(u32),
2174 qed_ptt_release(p_hwfn, p_ptt);
2179 int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto)
2184 /* Free Connection CXT */
2185 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_CXT,
2186 qed_cxt_get_proto_cid_start(p_hwfn,
2188 qed_cxt_get_proto_cid_count(p_hwfn,
2195 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_TASK, 0,
2196 qed_cxt_get_proto_tid_count(p_hwfn, proto));
2201 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_SRQ, 0,
2202 qed_cxt_get_srq_count(p_hwfn));
2207 int qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn,
2208 u32 tid, u8 ctx_type, void **pp_task_ctx)
2210 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2211 struct qed_ilt_client_cfg *p_cli;
2212 struct qed_ilt_cli_blk *p_seg;
2213 struct qed_tid_seg *p_seg_info;
2216 u32 tid_size, ilt_idx;
2217 u32 num_tids_per_block;
2219 /* Verify the personality */
2220 switch (p_hwfn->hw_info.personality) {
2222 proto = PROTOCOLID_ISCSI;
2223 seg = QED_CXT_ISCSI_TID_SEG;
2229 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2233 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2235 if (ctx_type == QED_CTX_WORKING_MEM) {
2236 p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2237 } else if (ctx_type == QED_CTX_FL_MEM) {
2238 if (!p_seg_info->has_fl_mem)
2240 p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2244 total_lines = DIV_ROUND_UP(p_seg->total_size, p_seg->real_size_in_page);
2245 tid_size = p_mngr->task_type_size[p_seg_info->type];
2246 num_tids_per_block = p_seg->real_size_in_page / tid_size;
2248 if (total_lines < tid / num_tids_per_block)
2251 ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2252 p_mngr->pf_start_line;
2253 *pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2254 (tid % num_tids_per_block) * tid_size;