1 /* QLogic qed NIC Driver
2 * Copyright (c) 2015-2017 QLogic Corporation
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and /or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <linux/types.h>
34 #include <linux/bitops.h>
35 #include <linux/dma-mapping.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/list.h>
39 #include <linux/log2.h>
40 #include <linux/pci.h>
41 #include <linux/slab.h>
42 #include <linux/string.h>
43 #include <linux/bitops.h>
46 #include "qed_dev_api.h"
49 #include "qed_init_ops.h"
50 #include "qed_reg_addr.h"
51 #include "qed_sriov.h"
53 /* Max number of connection types in HW (DQ/CDU etc.) */
54 #define MAX_CONN_TYPES PROTOCOLID_COMMON
55 #define NUM_TASK_TYPES 2
56 #define NUM_TASK_PF_SEGMENTS 4
57 #define NUM_TASK_VF_SEGMENTS 1
60 #define QM_PQ_ELEMENT_SIZE 4 /* in bytes */
62 /* Doorbell-Queue constants */
63 #define DQ_RANGE_SHIFT 4
64 #define DQ_RANGE_ALIGN BIT(DQ_RANGE_SHIFT)
66 /* Searcher constants */
67 #define SRC_MIN_NUM_ELEMS 256
69 /* Timers constants */
71 #define TM_ALIGN BIT(TM_SHIFT)
72 #define TM_ELEM_SIZE 4
74 #define ILT_DEFAULT_HW_P_SIZE 4
76 #define ILT_PAGE_IN_BYTES(hw_p_size) (1U << ((hw_p_size) + 12))
77 #define ILT_CFG_REG(cli, reg) PSWRQ2_REG_ ## cli ## _ ## reg ## _RT_OFFSET
79 /* ILT entry structure */
80 #define ILT_ENTRY_PHY_ADDR_MASK 0x000FFFFFFFFFFFULL
81 #define ILT_ENTRY_PHY_ADDR_SHIFT 0
82 #define ILT_ENTRY_VALID_MASK 0x1ULL
83 #define ILT_ENTRY_VALID_SHIFT 52
84 #define ILT_ENTRY_IN_REGS 2
85 #define ILT_REG_SIZE_IN_BYTES 4
87 /* connection context union */
89 struct core_conn_context core_ctx;
90 struct eth_conn_context eth_ctx;
91 struct iscsi_conn_context iscsi_ctx;
92 struct fcoe_conn_context fcoe_ctx;
93 struct roce_conn_context roce_ctx;
96 /* TYPE-0 task context - iSCSI, FCOE */
97 union type0_task_context {
98 struct iscsi_task_context iscsi_ctx;
99 struct fcoe_task_context fcoe_ctx;
102 /* TYPE-1 task context - ROCE */
103 union type1_task_context {
104 struct rdma_task_context roce_ctx;
112 #define CDUT_SEG_ALIGNMET 3 /* in 4k chunks */
113 #define CDUT_SEG_ALIGNMET_IN_BYTES (1 << (CDUT_SEG_ALIGNMET + 12))
115 #define CONN_CXT_SIZE(p_hwfn) \
116 ALIGNED_TYPE_SIZE(union conn_context, p_hwfn)
118 #define SRQ_CXT_SIZE (sizeof(struct rdma_srq_context))
120 #define TYPE0_TASK_CXT_SIZE(p_hwfn) \
121 ALIGNED_TYPE_SIZE(union type0_task_context, p_hwfn)
123 /* Alignment is inherent to the type1_task_context structure */
124 #define TYPE1_TASK_CXT_SIZE(p_hwfn) sizeof(union type1_task_context)
126 /* PF per protocl configuration object */
127 #define TASK_SEGMENTS (NUM_TASK_PF_SEGMENTS + NUM_TASK_VF_SEGMENTS)
128 #define TASK_SEGMENT_VF (NUM_TASK_PF_SEGMENTS)
136 struct qed_conn_type_cfg {
140 struct qed_tid_seg tid_seg[TASK_SEGMENTS];
143 /* ILT Client configuration, Per connection type (protocol) resources. */
144 #define ILT_CLI_PF_BLOCKS (1 + NUM_TASK_PF_SEGMENTS * 2)
145 #define ILT_CLI_VF_BLOCKS (1 + NUM_TASK_VF_SEGMENTS * 2)
148 #define CDUT_SEG_BLK(n) (1 + (u8)(n))
149 #define CDUT_FL_SEG_BLK(n, X) (1 + (n) + NUM_TASK_ ## X ## _SEGMENTS)
161 struct ilt_cfg_pair {
166 struct qed_ilt_cli_blk {
167 u32 total_size; /* 0 means not active */
168 u32 real_size_in_page;
170 u32 dynamic_line_cnt;
173 struct qed_ilt_client_cfg {
177 struct ilt_cfg_pair first;
178 struct ilt_cfg_pair last;
179 struct ilt_cfg_pair p_size;
181 /* ILT client blocks for PF */
182 struct qed_ilt_cli_blk pf_blks[ILT_CLI_PF_BLOCKS];
185 /* ILT client blocks for VFs */
186 struct qed_ilt_cli_blk vf_blks[ILT_CLI_VF_BLOCKS];
192 * Protocol acquired CID lists
193 * PF start line in ILT
201 struct qed_cid_acquired_map {
204 unsigned long *cid_map;
207 struct qed_cxt_mngr {
208 /* Per protocl configuration */
209 struct qed_conn_type_cfg conn_cfg[MAX_CONN_TYPES];
211 /* computed ILT structure */
212 struct qed_ilt_client_cfg clients[ILT_CLI_MAX];
214 /* Task type sizes */
215 u32 task_type_size[NUM_TASK_TYPES];
217 /* total number of VFs for this hwfn -
218 * ALL VFs are symmetric in terms of HW resources
223 struct qed_cid_acquired_map acquired[MAX_CONN_TYPES];
225 /* ILT shadow table */
226 struct qed_dma_mem *ilt_shadow;
229 /* Mutex for a dynamic ILT allocation */
233 struct qed_dma_mem *t2;
238 /* total number of SRQ's for this hwfn */
241 /* Maximal number of L2 steering filters */
244 static bool src_proto(enum protocol_type type)
246 return type == PROTOCOLID_ISCSI ||
247 type == PROTOCOLID_FCOE;
250 static bool tm_cid_proto(enum protocol_type type)
252 return type == PROTOCOLID_ISCSI ||
253 type == PROTOCOLID_FCOE ||
254 type == PROTOCOLID_ROCE;
257 static bool tm_tid_proto(enum protocol_type type)
259 return type == PROTOCOLID_FCOE;
262 /* counts the iids for the CDU/CDUC ILT client configuration */
263 struct qed_cdu_iids {
268 static void qed_cxt_cdu_iids(struct qed_cxt_mngr *p_mngr,
269 struct qed_cdu_iids *iids)
273 for (type = 0; type < MAX_CONN_TYPES; type++) {
274 iids->pf_cids += p_mngr->conn_cfg[type].cid_count;
275 iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
279 /* counts the iids for the Searcher block configuration */
280 struct qed_src_iids {
285 static void qed_cxt_src_iids(struct qed_cxt_mngr *p_mngr,
286 struct qed_src_iids *iids)
290 for (i = 0; i < MAX_CONN_TYPES; i++) {
294 iids->pf_cids += p_mngr->conn_cfg[i].cid_count;
295 iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf;
298 /* Add L2 filtering filters in addition */
299 iids->pf_cids += p_mngr->arfs_count;
302 /* counts the iids for the Timers block configuration */
305 u32 pf_tids[NUM_TASK_PF_SEGMENTS]; /* per segment */
311 static void qed_cxt_tm_iids(struct qed_hwfn *p_hwfn,
312 struct qed_cxt_mngr *p_mngr,
313 struct qed_tm_iids *iids)
315 bool tm_vf_required = false;
316 bool tm_required = false;
319 /* Timers is a special case -> we don't count how many cids require
320 * timers but what's the max cid that will be used by the timer block.
321 * therefore we traverse in reverse order, and once we hit a protocol
322 * that requires the timers memory, we'll sum all the protocols up
325 for (i = MAX_CONN_TYPES - 1; i >= 0; i--) {
326 struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
328 if (tm_cid_proto(i) || tm_required) {
329 if (p_cfg->cid_count)
332 iids->pf_cids += p_cfg->cid_count;
335 if (tm_cid_proto(i) || tm_vf_required) {
336 if (p_cfg->cids_per_vf)
337 tm_vf_required = true;
339 iids->per_vf_cids += p_cfg->cids_per_vf;
342 if (tm_tid_proto(i)) {
343 struct qed_tid_seg *segs = p_cfg->tid_seg;
345 /* for each segment there is at most one
346 * protocol for which count is not 0.
348 for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
349 iids->pf_tids[j] += segs[j].count;
351 /* The last array elelment is for the VFs. As for PF
352 * segments there can be only one protocol for
353 * which this value is not 0.
355 iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
359 iids->pf_cids = roundup(iids->pf_cids, TM_ALIGN);
360 iids->per_vf_cids = roundup(iids->per_vf_cids, TM_ALIGN);
361 iids->per_vf_tids = roundup(iids->per_vf_tids, TM_ALIGN);
363 for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
364 iids->pf_tids[j] = roundup(iids->pf_tids[j], TM_ALIGN);
365 iids->pf_tids_total += iids->pf_tids[j];
369 static void qed_cxt_qm_iids(struct qed_hwfn *p_hwfn,
370 struct qed_qm_iids *iids)
372 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
373 struct qed_tid_seg *segs;
374 u32 vf_cids = 0, type, j;
377 for (type = 0; type < MAX_CONN_TYPES; type++) {
378 iids->cids += p_mngr->conn_cfg[type].cid_count;
379 vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
381 segs = p_mngr->conn_cfg[type].tid_seg;
382 /* for each segment there is at most one
383 * protocol for which count is not 0.
385 for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
386 iids->tids += segs[j].count;
388 /* The last array elelment is for the VFs. As for PF
389 * segments there can be only one protocol for
390 * which this value is not 0.
392 vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
395 iids->vf_cids += vf_cids * p_mngr->vf_count;
396 iids->tids += vf_tids * p_mngr->vf_count;
398 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
399 "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
400 iids->cids, iids->vf_cids, iids->tids, vf_tids);
403 static struct qed_tid_seg *qed_cxt_tid_seg_info(struct qed_hwfn *p_hwfn,
406 struct qed_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
409 /* Find the protocol with tid count > 0 for this segment.
410 * Note: there can only be one and this is already validated.
412 for (i = 0; i < MAX_CONN_TYPES; i++)
413 if (p_cfg->conn_cfg[i].tid_seg[seg].count)
414 return &p_cfg->conn_cfg[i].tid_seg[seg];
418 static void qed_cxt_set_srq_count(struct qed_hwfn *p_hwfn, u32 num_srqs)
420 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
422 p_mgr->srq_count = num_srqs;
425 static u32 qed_cxt_get_srq_count(struct qed_hwfn *p_hwfn)
427 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
429 return p_mgr->srq_count;
432 /* set the iids count per protocol */
433 static void qed_cxt_set_proto_cid_count(struct qed_hwfn *p_hwfn,
434 enum protocol_type type,
435 u32 cid_count, u32 vf_cid_cnt)
437 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
438 struct qed_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
440 p_conn->cid_count = roundup(cid_count, DQ_RANGE_ALIGN);
441 p_conn->cids_per_vf = roundup(vf_cid_cnt, DQ_RANGE_ALIGN);
443 if (type == PROTOCOLID_ROCE) {
444 u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
445 u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
446 u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
447 u32 align = elems_per_page * DQ_RANGE_ALIGN;
449 p_conn->cid_count = roundup(p_conn->cid_count, align);
453 u32 qed_cxt_get_proto_cid_count(struct qed_hwfn *p_hwfn,
454 enum protocol_type type, u32 *vf_cid)
457 *vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
459 return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
462 u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
463 enum protocol_type type)
465 return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
468 u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
469 enum protocol_type type)
474 for (i = 0; i < TASK_SEGMENTS; i++)
475 cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
480 static void qed_cxt_set_proto_tid_count(struct qed_hwfn *p_hwfn,
481 enum protocol_type proto,
483 u8 seg_type, u32 count, bool has_fl)
485 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
486 struct qed_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
488 p_seg->count = count;
489 p_seg->has_fl_mem = has_fl;
490 p_seg->type = seg_type;
493 static void qed_ilt_cli_blk_fill(struct qed_ilt_client_cfg *p_cli,
494 struct qed_ilt_cli_blk *p_blk,
495 u32 start_line, u32 total_size, u32 elem_size)
497 u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
499 /* verify thatits called only once for each block */
500 if (p_blk->total_size)
503 p_blk->total_size = total_size;
504 p_blk->real_size_in_page = 0;
506 p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
507 p_blk->start_line = start_line;
510 static void qed_ilt_cli_adv_line(struct qed_hwfn *p_hwfn,
511 struct qed_ilt_client_cfg *p_cli,
512 struct qed_ilt_cli_blk *p_blk,
513 u32 *p_line, enum ilt_clients client_id)
515 if (!p_blk->total_size)
519 p_cli->first.val = *p_line;
521 p_cli->active = true;
522 *p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
523 p_cli->last.val = *p_line - 1;
525 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
526 "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
527 client_id, p_cli->first.val,
528 p_cli->last.val, p_blk->total_size,
529 p_blk->real_size_in_page, p_blk->start_line);
532 static u32 qed_ilt_get_dynamic_line_cnt(struct qed_hwfn *p_hwfn,
533 enum ilt_clients ilt_client)
535 u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
536 struct qed_ilt_client_cfg *p_cli;
537 u32 lines_to_skip = 0;
540 if (ilt_client == ILT_CLI_CDUC) {
541 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
543 cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
544 (u32) CONN_CXT_SIZE(p_hwfn);
546 lines_to_skip = cid_count / cxts_per_p;
549 return lines_to_skip;
552 static struct qed_ilt_client_cfg *qed_cxt_set_cli(struct qed_ilt_client_cfg
555 p_cli->active = false;
556 p_cli->first.val = 0;
561 static struct qed_ilt_cli_blk *qed_cxt_set_blk(struct qed_ilt_cli_blk *p_blk)
563 p_blk->total_size = 0;
567 int qed_cxt_cfg_ilt_compute(struct qed_hwfn *p_hwfn, u32 *line_count)
569 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
570 u32 curr_line, total, i, task_size, line;
571 struct qed_ilt_client_cfg *p_cli;
572 struct qed_ilt_cli_blk *p_blk;
573 struct qed_cdu_iids cdu_iids;
574 struct qed_src_iids src_iids;
575 struct qed_qm_iids qm_iids;
576 struct qed_tm_iids tm_iids;
577 struct qed_tid_seg *p_seg;
579 memset(&qm_iids, 0, sizeof(qm_iids));
580 memset(&cdu_iids, 0, sizeof(cdu_iids));
581 memset(&src_iids, 0, sizeof(src_iids));
582 memset(&tm_iids, 0, sizeof(tm_iids));
584 p_mngr->pf_start_line = RESC_START(p_hwfn, QED_ILT);
586 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
587 "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
588 p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
591 p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUC]);
593 curr_line = p_mngr->pf_start_line;
596 p_cli->pf_total_lines = 0;
598 /* get the counters for the CDUC and QM clients */
599 qed_cxt_cdu_iids(p_mngr, &cdu_iids);
601 p_blk = qed_cxt_set_blk(&p_cli->pf_blks[CDUC_BLK]);
603 total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
605 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
606 total, CONN_CXT_SIZE(p_hwfn));
608 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
609 p_cli->pf_total_lines = curr_line - p_blk->start_line;
611 p_blk->dynamic_line_cnt = qed_ilt_get_dynamic_line_cnt(p_hwfn,
615 p_blk = qed_cxt_set_blk(&p_cli->vf_blks[CDUC_BLK]);
616 total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
618 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
619 total, CONN_CXT_SIZE(p_hwfn));
621 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
622 p_cli->vf_total_lines = curr_line - p_blk->start_line;
624 for (i = 1; i < p_mngr->vf_count; i++)
625 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
629 p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_CDUT]);
630 p_cli->first.val = curr_line;
632 /* first the 'working' task memory */
633 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
634 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
635 if (!p_seg || p_seg->count == 0)
638 p_blk = qed_cxt_set_blk(&p_cli->pf_blks[CDUT_SEG_BLK(i)]);
639 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
640 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
641 p_mngr->task_type_size[p_seg->type]);
643 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
647 /* next the 'init' task memory (forced load memory) */
648 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
649 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
650 if (!p_seg || p_seg->count == 0)
654 qed_cxt_set_blk(&p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)]);
656 if (!p_seg->has_fl_mem) {
657 /* The segment is active (total size pf 'working'
658 * memory is > 0) but has no FL (forced-load, Init)
661 * 1. The total-size in the corrsponding FL block of
662 * the ILT client is set to 0 - No ILT line are
663 * provisioned and no ILT memory allocated.
665 * 2. The start-line of said block is set to the
666 * start line of the matching working memory
667 * block in the ILT client. This is later used to
668 * configure the CDU segment offset registers and
669 * results in an FL command for TIDs of this
670 * segement behaves as regular load commands
671 * (loading TIDs from the working memory).
673 line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
675 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
678 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
680 qed_ilt_cli_blk_fill(p_cli, p_blk,
682 p_mngr->task_type_size[p_seg->type]);
684 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
687 p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
690 p_seg = qed_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
691 if (p_seg && p_seg->count) {
692 /* Stricly speaking we need to iterate over all VF
693 * task segment types, but a VF has only 1 segment
696 /* 'working' memory */
697 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
699 p_blk = qed_cxt_set_blk(&p_cli->vf_blks[CDUT_SEG_BLK(0)]);
700 qed_ilt_cli_blk_fill(p_cli, p_blk,
702 p_mngr->task_type_size[p_seg->type]);
704 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
709 qed_cxt_set_blk(&p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)]);
710 if (!p_seg->has_fl_mem) {
711 /* see comment above */
712 line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
713 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
715 task_size = p_mngr->task_type_size[p_seg->type];
716 qed_ilt_cli_blk_fill(p_cli, p_blk,
717 curr_line, total, task_size);
718 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
721 p_cli->vf_total_lines = curr_line -
722 p_cli->vf_blks[0].start_line;
724 /* Now for the rest of the VFs */
725 for (i = 1; i < p_mngr->vf_count; i++) {
726 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
727 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
730 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
731 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
737 p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_QM]);
738 p_blk = qed_cxt_set_blk(&p_cli->pf_blks[0]);
740 qed_cxt_qm_iids(p_hwfn, &qm_iids);
741 total = qed_qm_pf_mem_size(p_hwfn->rel_pf_id, qm_iids.cids,
742 qm_iids.vf_cids, qm_iids.tids,
743 p_hwfn->qm_info.num_pqs,
744 p_hwfn->qm_info.num_vf_pqs);
748 "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
752 p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
754 qed_ilt_cli_blk_fill(p_cli, p_blk,
755 curr_line, total * 0x1000,
758 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
759 p_cli->pf_total_lines = curr_line - p_blk->start_line;
762 p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_SRC]);
763 qed_cxt_src_iids(p_mngr, &src_iids);
765 /* Both the PF and VFs searcher connections are stored in the per PF
766 * database. Thus sum the PF searcher cids and all the VFs searcher
769 total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
771 u32 local_max = max_t(u32, total,
774 total = roundup_pow_of_two(local_max);
776 p_blk = qed_cxt_set_blk(&p_cli->pf_blks[0]);
777 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
778 total * sizeof(struct src_ent),
779 sizeof(struct src_ent));
781 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
783 p_cli->pf_total_lines = curr_line - p_blk->start_line;
787 p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_TM]);
788 qed_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
789 total = tm_iids.pf_cids + tm_iids.pf_tids_total;
791 p_blk = qed_cxt_set_blk(&p_cli->pf_blks[0]);
792 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
793 total * TM_ELEM_SIZE, TM_ELEM_SIZE);
795 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
797 p_cli->pf_total_lines = curr_line - p_blk->start_line;
801 total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
803 p_blk = qed_cxt_set_blk(&p_cli->vf_blks[0]);
804 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
805 total * TM_ELEM_SIZE, TM_ELEM_SIZE);
807 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
810 p_cli->vf_total_lines = curr_line - p_blk->start_line;
811 for (i = 1; i < p_mngr->vf_count; i++)
812 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
816 /* TSDM (SRQ CONTEXT) */
817 total = qed_cxt_get_srq_count(p_hwfn);
820 p_cli = qed_cxt_set_cli(&p_mngr->clients[ILT_CLI_TSDM]);
821 p_blk = qed_cxt_set_blk(&p_cli->pf_blks[SRQ_BLK]);
822 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
823 total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
825 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
827 p_cli->pf_total_lines = curr_line - p_blk->start_line;
830 *line_count = curr_line - p_hwfn->p_cxt_mngr->pf_start_line;
832 if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
833 RESC_NUM(p_hwfn, QED_ILT))
839 u32 qed_cxt_cfg_ilt_compute_excess(struct qed_hwfn *p_hwfn, u32 used_lines)
841 struct qed_ilt_client_cfg *p_cli;
842 u32 excess_lines, available_lines;
843 struct qed_cxt_mngr *p_mngr;
844 u32 ilt_page_size, elem_size;
845 struct qed_tid_seg *p_seg;
848 available_lines = RESC_NUM(p_hwfn, QED_ILT);
849 excess_lines = used_lines - available_lines;
854 if (p_hwfn->hw_info.personality != QED_PCI_ETH_ROCE)
857 p_mngr = p_hwfn->p_cxt_mngr;
858 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
859 ilt_page_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
861 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
862 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
863 if (!p_seg || p_seg->count == 0)
866 elem_size = p_mngr->task_type_size[p_seg->type];
870 return (ilt_page_size / elem_size) * excess_lines;
873 DP_NOTICE(p_hwfn, "failed computing excess ILT lines\n");
877 static void qed_cxt_src_t2_free(struct qed_hwfn *p_hwfn)
879 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
885 for (i = 0; i < p_mngr->t2_num_pages; i++)
886 if (p_mngr->t2[i].p_virt)
887 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
889 p_mngr->t2[i].p_virt,
890 p_mngr->t2[i].p_phys);
896 static int qed_cxt_src_t2_alloc(struct qed_hwfn *p_hwfn)
898 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
899 u32 conn_num, total_size, ent_per_page, psz, i;
900 struct qed_ilt_client_cfg *p_src;
901 struct qed_src_iids src_iids;
902 struct qed_dma_mem *p_t2;
905 memset(&src_iids, 0, sizeof(src_iids));
907 /* if the SRC ILT client is inactive - there are no connection
908 * requiring the searcer, leave.
910 p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
914 qed_cxt_src_iids(p_mngr, &src_iids);
915 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
916 total_size = conn_num * sizeof(struct src_ent);
918 /* use the same page size as the SRC ILT client */
919 psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
920 p_mngr->t2_num_pages = DIV_ROUND_UP(total_size, psz);
923 p_mngr->t2 = kcalloc(p_mngr->t2_num_pages, sizeof(struct qed_dma_mem),
930 /* allocate t2 pages */
931 for (i = 0; i < p_mngr->t2_num_pages; i++) {
932 u32 size = min_t(u32, total_size, psz);
933 void **p_virt = &p_mngr->t2[i].p_virt;
935 *p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
937 &p_mngr->t2[i].p_phys, GFP_KERNEL);
938 if (!p_mngr->t2[i].p_virt) {
942 memset(*p_virt, 0, size);
943 p_mngr->t2[i].size = size;
947 /* Set the t2 pointers */
949 /* entries per page - must be a power of two */
950 ent_per_page = psz / sizeof(struct src_ent);
952 p_mngr->first_free = (u64) p_mngr->t2[0].p_phys;
954 p_t2 = &p_mngr->t2[(conn_num - 1) / ent_per_page];
955 p_mngr->last_free = (u64) p_t2->p_phys +
956 ((conn_num - 1) & (ent_per_page - 1)) * sizeof(struct src_ent);
958 for (i = 0; i < p_mngr->t2_num_pages; i++) {
959 u32 ent_num = min_t(u32,
962 struct src_ent *entries = p_mngr->t2[i].p_virt;
963 u64 p_ent_phys = (u64) p_mngr->t2[i].p_phys, val;
966 for (j = 0; j < ent_num - 1; j++) {
967 val = p_ent_phys + (j + 1) * sizeof(struct src_ent);
968 entries[j].next = cpu_to_be64(val);
971 if (i < p_mngr->t2_num_pages - 1)
972 val = (u64) p_mngr->t2[i + 1].p_phys;
975 entries[j].next = cpu_to_be64(val);
983 qed_cxt_src_t2_free(p_hwfn);
987 #define for_each_ilt_valid_client(pos, clients) \
988 for (pos = 0; pos < ILT_CLI_MAX; pos++) \
989 if (!clients[pos].active) { \
993 /* Total number of ILT lines used by this PF */
994 static u32 qed_cxt_ilt_shadow_size(struct qed_ilt_client_cfg *ilt_clients)
999 for_each_ilt_valid_client(i, ilt_clients)
1000 size += (ilt_clients[i].last.val - ilt_clients[i].first.val + 1);
1005 static void qed_ilt_shadow_free(struct qed_hwfn *p_hwfn)
1007 struct qed_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
1008 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1011 ilt_size = qed_cxt_ilt_shadow_size(p_cli);
1013 for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
1014 struct qed_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
1017 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
1018 p_dma->size, p_dma->p_virt,
1020 p_dma->p_virt = NULL;
1022 kfree(p_mngr->ilt_shadow);
1025 static int qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn,
1026 struct qed_ilt_cli_blk *p_blk,
1027 enum ilt_clients ilt_client,
1028 u32 start_line_offset)
1030 struct qed_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
1031 u32 lines, line, sz_left, lines_to_skip = 0;
1033 /* Special handling for RoCE that supports dynamic allocation */
1034 if ((p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE) &&
1035 ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
1038 lines_to_skip = p_blk->dynamic_line_cnt;
1040 if (!p_blk->total_size)
1043 sz_left = p_blk->total_size;
1044 lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) - lines_to_skip;
1045 line = p_blk->start_line + start_line_offset -
1046 p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
1048 for (; lines; lines--) {
1053 size = min_t(u32, sz_left, p_blk->real_size_in_page);
1054 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
1055 size, &p_phys, GFP_KERNEL);
1058 memset(p_virt, 0, size);
1060 ilt_shadow[line].p_phys = p_phys;
1061 ilt_shadow[line].p_virt = p_virt;
1062 ilt_shadow[line].size = size;
1064 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1065 "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
1066 line, (u64)p_phys, p_virt, size);
1075 static int qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn)
1077 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1078 struct qed_ilt_client_cfg *clients = p_mngr->clients;
1079 struct qed_ilt_cli_blk *p_blk;
1083 size = qed_cxt_ilt_shadow_size(clients);
1084 p_mngr->ilt_shadow = kcalloc(size, sizeof(struct qed_dma_mem),
1086 if (!p_mngr->ilt_shadow) {
1088 goto ilt_shadow_fail;
1091 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1092 "Allocated 0x%x bytes for ilt shadow\n",
1093 (u32)(size * sizeof(struct qed_dma_mem)));
1095 for_each_ilt_valid_client(i, clients) {
1096 for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1097 p_blk = &clients[i].pf_blks[j];
1098 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1100 goto ilt_shadow_fail;
1102 for (k = 0; k < p_mngr->vf_count; k++) {
1103 for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1104 u32 lines = clients[i].vf_total_lines * k;
1106 p_blk = &clients[i].vf_blks[j];
1107 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, lines);
1109 goto ilt_shadow_fail;
1117 qed_ilt_shadow_free(p_hwfn);
1121 static void qed_cid_map_free(struct qed_hwfn *p_hwfn)
1123 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1126 for (type = 0; type < MAX_CONN_TYPES; type++) {
1127 kfree(p_mngr->acquired[type].cid_map);
1128 p_mngr->acquired[type].max_count = 0;
1129 p_mngr->acquired[type].start_cid = 0;
1133 static int qed_cid_map_alloc(struct qed_hwfn *p_hwfn)
1135 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1139 for (type = 0; type < MAX_CONN_TYPES; type++) {
1140 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1146 size = DIV_ROUND_UP(cid_cnt,
1147 sizeof(unsigned long) * BITS_PER_BYTE) *
1148 sizeof(unsigned long);
1149 p_mngr->acquired[type].cid_map = kzalloc(size, GFP_KERNEL);
1150 if (!p_mngr->acquired[type].cid_map)
1153 p_mngr->acquired[type].max_count = cid_cnt;
1154 p_mngr->acquired[type].start_cid = start_cid;
1156 p_hwfn->p_cxt_mngr->conn_cfg[type].cid_start = start_cid;
1158 DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1159 "Type %08x start: %08x count %08x\n",
1160 type, p_mngr->acquired[type].start_cid,
1161 p_mngr->acquired[type].max_count);
1162 start_cid += cid_cnt;
1168 qed_cid_map_free(p_hwfn);
1172 int qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn)
1174 struct qed_ilt_client_cfg *clients;
1175 struct qed_cxt_mngr *p_mngr;
1178 p_mngr = kzalloc(sizeof(*p_mngr), GFP_KERNEL);
1182 /* Initialize ILT client registers */
1183 clients = p_mngr->clients;
1184 clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1185 clients[ILT_CLI_CDUC].last.reg = ILT_CFG_REG(CDUC, LAST_ILT);
1186 clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1188 clients[ILT_CLI_QM].first.reg = ILT_CFG_REG(QM, FIRST_ILT);
1189 clients[ILT_CLI_QM].last.reg = ILT_CFG_REG(QM, LAST_ILT);
1190 clients[ILT_CLI_QM].p_size.reg = ILT_CFG_REG(QM, P_SIZE);
1192 clients[ILT_CLI_TM].first.reg = ILT_CFG_REG(TM, FIRST_ILT);
1193 clients[ILT_CLI_TM].last.reg = ILT_CFG_REG(TM, LAST_ILT);
1194 clients[ILT_CLI_TM].p_size.reg = ILT_CFG_REG(TM, P_SIZE);
1196 clients[ILT_CLI_SRC].first.reg = ILT_CFG_REG(SRC, FIRST_ILT);
1197 clients[ILT_CLI_SRC].last.reg = ILT_CFG_REG(SRC, LAST_ILT);
1198 clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1200 clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1201 clients[ILT_CLI_CDUT].last.reg = ILT_CFG_REG(CDUT, LAST_ILT);
1202 clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1204 clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1205 clients[ILT_CLI_TSDM].last.reg = ILT_CFG_REG(TSDM, LAST_ILT);
1206 clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1207 /* default ILT page size for all clients is 64K */
1208 for (i = 0; i < ILT_CLI_MAX; i++)
1209 p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE;
1211 /* Initialize task sizes */
1212 p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1213 p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1215 if (p_hwfn->cdev->p_iov_info)
1216 p_mngr->vf_count = p_hwfn->cdev->p_iov_info->total_vfs;
1217 /* Initialize the dynamic ILT allocation mutex */
1218 mutex_init(&p_mngr->mutex);
1220 /* Set the cxt mangr pointer priori to further allocations */
1221 p_hwfn->p_cxt_mngr = p_mngr;
1226 int qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn)
1230 /* Allocate the ILT shadow table */
1231 rc = qed_ilt_shadow_alloc(p_hwfn);
1233 goto tables_alloc_fail;
1235 /* Allocate the T2 table */
1236 rc = qed_cxt_src_t2_alloc(p_hwfn);
1238 goto tables_alloc_fail;
1240 /* Allocate and initialize the acquired cids bitmaps */
1241 rc = qed_cid_map_alloc(p_hwfn);
1243 goto tables_alloc_fail;
1248 qed_cxt_mngr_free(p_hwfn);
1252 void qed_cxt_mngr_free(struct qed_hwfn *p_hwfn)
1254 if (!p_hwfn->p_cxt_mngr)
1257 qed_cid_map_free(p_hwfn);
1258 qed_cxt_src_t2_free(p_hwfn);
1259 qed_ilt_shadow_free(p_hwfn);
1260 kfree(p_hwfn->p_cxt_mngr);
1262 p_hwfn->p_cxt_mngr = NULL;
1265 void qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn)
1267 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1270 /* Reset acquired cids */
1271 for (type = 0; type < MAX_CONN_TYPES; type++) {
1272 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1277 memset(p_mngr->acquired[type].cid_map, 0,
1278 DIV_ROUND_UP(cid_cnt,
1279 sizeof(unsigned long) * BITS_PER_BYTE) *
1280 sizeof(unsigned long));
1285 #define CDUC_CXT_SIZE_SHIFT \
1286 CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1288 #define CDUC_CXT_SIZE_MASK \
1289 (CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1291 #define CDUC_BLOCK_WASTE_SHIFT \
1292 CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1294 #define CDUC_BLOCK_WASTE_MASK \
1295 (CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1297 #define CDUC_NCIB_SHIFT \
1298 CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1300 #define CDUC_NCIB_MASK \
1301 (CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1303 #define CDUT_TYPE0_CXT_SIZE_SHIFT \
1304 CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1306 #define CDUT_TYPE0_CXT_SIZE_MASK \
1307 (CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >> \
1308 CDUT_TYPE0_CXT_SIZE_SHIFT)
1310 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT \
1311 CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1313 #define CDUT_TYPE0_BLOCK_WASTE_MASK \
1314 (CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >> \
1315 CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1317 #define CDUT_TYPE0_NCIB_SHIFT \
1318 CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1320 #define CDUT_TYPE0_NCIB_MASK \
1321 (CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >> \
1322 CDUT_TYPE0_NCIB_SHIFT)
1324 #define CDUT_TYPE1_CXT_SIZE_SHIFT \
1325 CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1327 #define CDUT_TYPE1_CXT_SIZE_MASK \
1328 (CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >> \
1329 CDUT_TYPE1_CXT_SIZE_SHIFT)
1331 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT \
1332 CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1334 #define CDUT_TYPE1_BLOCK_WASTE_MASK \
1335 (CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >> \
1336 CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1338 #define CDUT_TYPE1_NCIB_SHIFT \
1339 CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1341 #define CDUT_TYPE1_NCIB_MASK \
1342 (CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >> \
1343 CDUT_TYPE1_NCIB_SHIFT)
1345 static void qed_cdu_init_common(struct qed_hwfn *p_hwfn)
1347 u32 page_sz, elems_per_page, block_waste, cxt_size, cdu_params = 0;
1349 /* CDUC - connection configuration */
1350 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1351 cxt_size = CONN_CXT_SIZE(p_hwfn);
1352 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1353 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1355 SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1356 SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1357 SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
1358 STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1360 /* CDUT - type-0 tasks configuration */
1361 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1362 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1363 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1364 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1366 /* cxt size and block-waste are multipes of 8 */
1368 SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1369 SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1370 SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1371 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1373 /* CDUT - type-1 tasks configuration */
1374 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1375 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1376 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1378 /* cxt size and block-waste are multipes of 8 */
1380 SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1381 SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1382 SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1383 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1387 #define CDU_SEG_REG_TYPE_SHIFT CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1388 #define CDU_SEG_REG_TYPE_MASK 0x1
1389 #define CDU_SEG_REG_OFFSET_SHIFT 0
1390 #define CDU_SEG_REG_OFFSET_MASK CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1392 static void qed_cdu_init_pf(struct qed_hwfn *p_hwfn)
1394 struct qed_ilt_client_cfg *p_cli;
1395 struct qed_tid_seg *p_seg;
1396 u32 cdu_seg_params, offset;
1399 static const u32 rt_type_offset_arr[] = {
1400 CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1401 CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1402 CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1403 CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1406 static const u32 rt_type_offset_fl_arr[] = {
1407 CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1408 CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1409 CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1410 CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1413 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1415 /* There are initializations only for CDUT during pf Phase */
1416 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1418 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
1422 /* Note: start_line is already adjusted for the CDU
1423 * segment register granularity, so we just need to
1424 * divide. Adjustment is implicit as we assume ILT
1425 * Page size is larger than 32K!
1427 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1428 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1429 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1432 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1433 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1434 STORE_RT_REG(p_hwfn, rt_type_offset_arr[i], cdu_seg_params);
1436 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1437 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1438 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1441 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1442 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1443 STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i], cdu_seg_params);
1447 void qed_qm_init_pf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
1449 struct qed_qm_pf_rt_init_params params;
1450 struct qed_qm_info *qm_info = &p_hwfn->qm_info;
1451 struct qed_qm_iids iids;
1453 memset(&iids, 0, sizeof(iids));
1454 qed_cxt_qm_iids(p_hwfn, &iids);
1456 memset(¶ms, 0, sizeof(params));
1457 params.port_id = p_hwfn->port_id;
1458 params.pf_id = p_hwfn->rel_pf_id;
1459 params.max_phys_tcs_per_port = qm_info->max_phys_tcs_per_port;
1460 params.is_first_pf = p_hwfn->first_on_engine;
1461 params.num_pf_cids = iids.cids;
1462 params.num_vf_cids = iids.vf_cids;
1463 params.num_tids = iids.tids;
1464 params.start_pq = qm_info->start_pq;
1465 params.num_pf_pqs = qm_info->num_pqs - qm_info->num_vf_pqs;
1466 params.num_vf_pqs = qm_info->num_vf_pqs;
1467 params.start_vport = qm_info->start_vport;
1468 params.num_vports = qm_info->num_vports;
1469 params.pf_wfq = qm_info->pf_wfq;
1470 params.pf_rl = qm_info->pf_rl;
1471 params.pq_params = qm_info->qm_pq_params;
1472 params.vport_params = qm_info->qm_vport_params;
1474 qed_qm_pf_rt_init(p_hwfn, p_ptt, ¶ms);
1478 void qed_cm_init_pf(struct qed_hwfn *p_hwfn)
1480 /* XCM pure-LB queue */
1481 STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET,
1482 qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1486 static void qed_dq_init_pf(struct qed_hwfn *p_hwfn)
1488 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1489 u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1491 dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1492 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1494 dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1495 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1497 dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1498 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1500 dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1501 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1503 dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1504 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1506 dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1507 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1509 dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1510 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1512 dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1513 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1515 dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1516 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1518 dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1519 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1521 dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1522 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1524 dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1525 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1527 /* Connection types 6 & 7 are not in use, yet they must be configured
1528 * as the highest possible connection. Not configuring them means the
1529 * defaults will be used, and with a large number of cids a bug may
1530 * occur, if the defaults will be smaller than dq_pf_max_cid /
1533 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1534 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1536 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1537 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1540 static void qed_ilt_bounds_init(struct qed_hwfn *p_hwfn)
1542 struct qed_ilt_client_cfg *ilt_clients;
1545 ilt_clients = p_hwfn->p_cxt_mngr->clients;
1546 for_each_ilt_valid_client(i, ilt_clients) {
1547 STORE_RT_REG(p_hwfn,
1548 ilt_clients[i].first.reg,
1549 ilt_clients[i].first.val);
1550 STORE_RT_REG(p_hwfn,
1551 ilt_clients[i].last.reg, ilt_clients[i].last.val);
1552 STORE_RT_REG(p_hwfn,
1553 ilt_clients[i].p_size.reg,
1554 ilt_clients[i].p_size.val);
1558 static void qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn)
1560 struct qed_ilt_client_cfg *p_cli;
1563 /* For simplicty we set the 'block' to be an ILT page */
1564 if (p_hwfn->cdev->p_iov_info) {
1565 struct qed_hw_sriov_info *p_iov = p_hwfn->cdev->p_iov_info;
1567 STORE_RT_REG(p_hwfn,
1568 PSWRQ2_REG_VF_BASE_RT_OFFSET,
1569 p_iov->first_vf_in_pf);
1570 STORE_RT_REG(p_hwfn,
1571 PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1572 p_iov->first_vf_in_pf + p_iov->total_vfs);
1575 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1576 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1577 if (p_cli->active) {
1578 STORE_RT_REG(p_hwfn,
1579 PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1581 STORE_RT_REG(p_hwfn,
1582 PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1583 p_cli->pf_total_lines);
1584 STORE_RT_REG(p_hwfn,
1585 PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1586 p_cli->vf_total_lines);
1589 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1590 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1591 if (p_cli->active) {
1592 STORE_RT_REG(p_hwfn,
1593 PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1595 STORE_RT_REG(p_hwfn,
1596 PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1597 p_cli->pf_total_lines);
1598 STORE_RT_REG(p_hwfn,
1599 PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1600 p_cli->vf_total_lines);
1603 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1604 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1605 if (p_cli->active) {
1606 STORE_RT_REG(p_hwfn,
1607 PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET, blk_factor);
1608 STORE_RT_REG(p_hwfn,
1609 PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1610 p_cli->pf_total_lines);
1611 STORE_RT_REG(p_hwfn,
1612 PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1613 p_cli->vf_total_lines);
1617 /* ILT (PSWRQ2) PF */
1618 static void qed_ilt_init_pf(struct qed_hwfn *p_hwfn)
1620 struct qed_ilt_client_cfg *clients;
1621 struct qed_cxt_mngr *p_mngr;
1622 struct qed_dma_mem *p_shdw;
1623 u32 line, rt_offst, i;
1625 qed_ilt_bounds_init(p_hwfn);
1626 qed_ilt_vf_bounds_init(p_hwfn);
1628 p_mngr = p_hwfn->p_cxt_mngr;
1629 p_shdw = p_mngr->ilt_shadow;
1630 clients = p_hwfn->p_cxt_mngr->clients;
1632 for_each_ilt_valid_client(i, clients) {
1633 /** Client's 1st val and RT array are absolute, ILT shadows'
1634 * lines are relative.
1636 line = clients[i].first.val - p_mngr->pf_start_line;
1637 rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1638 clients[i].first.val * ILT_ENTRY_IN_REGS;
1640 for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1641 line++, rt_offst += ILT_ENTRY_IN_REGS) {
1642 u64 ilt_hw_entry = 0;
1644 /** p_virt could be NULL incase of dynamic
1647 if (p_shdw[line].p_virt) {
1648 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1649 SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1650 (p_shdw[line].p_phys >> 12));
1652 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1653 "Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1655 (u64)(p_shdw[line].p_phys >> 12));
1658 STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1663 /* SRC (Searcher) PF */
1664 static void qed_src_init_pf(struct qed_hwfn *p_hwfn)
1666 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1667 u32 rounded_conn_num, conn_num, conn_max;
1668 struct qed_src_iids src_iids;
1670 memset(&src_iids, 0, sizeof(src_iids));
1671 qed_cxt_src_iids(p_mngr, &src_iids);
1672 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1676 conn_max = max_t(u32, conn_num, SRC_MIN_NUM_ELEMS);
1677 rounded_conn_num = roundup_pow_of_two(conn_max);
1679 STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1680 STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1681 ilog2(rounded_conn_num));
1683 STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1684 p_hwfn->p_cxt_mngr->first_free);
1685 STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1686 p_hwfn->p_cxt_mngr->last_free);
1690 #define TM_CFG_NUM_IDS_SHIFT 0
1691 #define TM_CFG_NUM_IDS_MASK 0xFFFFULL
1692 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT 16
1693 #define TM_CFG_PRE_SCAN_OFFSET_MASK 0x1FFULL
1694 #define TM_CFG_PARENT_PF_SHIFT 25
1695 #define TM_CFG_PARENT_PF_MASK 0x7ULL
1697 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT 30
1698 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK 0x1FFULL
1700 #define TM_CFG_TID_OFFSET_SHIFT 30
1701 #define TM_CFG_TID_OFFSET_MASK 0x7FFFFULL
1702 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT 49
1703 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK 0x1FFULL
1705 static void qed_tm_init_pf(struct qed_hwfn *p_hwfn)
1707 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1708 u32 active_seg_mask = 0, tm_offset, rt_reg;
1709 struct qed_tm_iids tm_iids;
1713 memset(&tm_iids, 0, sizeof(tm_iids));
1714 qed_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
1716 /* @@@TBD No pre-scan for now */
1718 /* Note: We assume consecutive VFs for a PF */
1719 for (i = 0; i < p_mngr->vf_count; i++) {
1721 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1722 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1723 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1724 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0);
1725 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1726 (sizeof(cfg_word) / sizeof(u32)) *
1727 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1728 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1732 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1733 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1734 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0); /* n/a for PF */
1735 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1737 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1738 (sizeof(cfg_word) / sizeof(u32)) *
1739 (NUM_OF_VFS(p_hwfn->cdev) + p_hwfn->rel_pf_id);
1740 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1743 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1744 tm_iids.pf_cids ? 0x1 : 0x0);
1746 /* @@@TBD how to enable the scan for the VFs */
1748 tm_offset = tm_iids.per_vf_cids;
1750 /* Note: We assume consecutive VFs for a PF */
1751 for (i = 0; i < p_mngr->vf_count; i++) {
1753 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1754 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1755 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1756 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1757 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1759 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1760 (sizeof(cfg_word) / sizeof(u32)) *
1761 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1763 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1766 tm_offset = tm_iids.pf_cids;
1767 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1769 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1770 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1771 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1772 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1773 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1775 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1776 (sizeof(cfg_word) / sizeof(u32)) *
1777 (NUM_OF_VFS(p_hwfn->cdev) +
1778 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1780 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1781 active_seg_mask |= (tm_iids.pf_tids[i] ? BIT(i) : 0);
1783 tm_offset += tm_iids.pf_tids[i];
1786 if (p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE)
1787 active_seg_mask = 0;
1789 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1791 /* @@@TBD how to enable the scan for the VFs */
1794 static void qed_prs_init_common(struct qed_hwfn *p_hwfn)
1796 if ((p_hwfn->hw_info.personality == QED_PCI_FCOE) &&
1797 p_hwfn->pf_params.fcoe_pf_params.is_target)
1798 STORE_RT_REG(p_hwfn,
1799 PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1802 static void qed_prs_init_pf(struct qed_hwfn *p_hwfn)
1804 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1805 struct qed_conn_type_cfg *p_fcoe;
1806 struct qed_tid_seg *p_tid;
1808 p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1810 /* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1811 if (!p_fcoe->cid_count)
1814 p_tid = &p_fcoe->tid_seg[QED_CXT_FCOE_TID_SEG];
1815 if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1816 STORE_RT_REG_AGG(p_hwfn,
1817 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1820 STORE_RT_REG_AGG(p_hwfn,
1821 PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1826 void qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn)
1828 qed_cdu_init_common(p_hwfn);
1829 qed_prs_init_common(p_hwfn);
1832 void qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt)
1834 qed_qm_init_pf(p_hwfn, p_ptt);
1835 qed_cm_init_pf(p_hwfn);
1836 qed_dq_init_pf(p_hwfn);
1837 qed_cdu_init_pf(p_hwfn);
1838 qed_ilt_init_pf(p_hwfn);
1839 qed_src_init_pf(p_hwfn);
1840 qed_tm_init_pf(p_hwfn);
1841 qed_prs_init_pf(p_hwfn);
1844 int qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1845 enum protocol_type type, u32 *p_cid)
1847 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1850 if (type >= MAX_CONN_TYPES || !p_mngr->acquired[type].cid_map) {
1851 DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1855 rel_cid = find_first_zero_bit(p_mngr->acquired[type].cid_map,
1856 p_mngr->acquired[type].max_count);
1858 if (rel_cid >= p_mngr->acquired[type].max_count) {
1859 DP_NOTICE(p_hwfn, "no CID available for protocol %d\n", type);
1863 __set_bit(rel_cid, p_mngr->acquired[type].cid_map);
1865 *p_cid = rel_cid + p_mngr->acquired[type].start_cid;
1870 static bool qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn,
1871 u32 cid, enum protocol_type *p_type)
1873 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1874 struct qed_cid_acquired_map *p_map;
1875 enum protocol_type p;
1878 /* Iterate over protocols and find matching cid range */
1879 for (p = 0; p < MAX_CONN_TYPES; p++) {
1880 p_map = &p_mngr->acquired[p];
1882 if (!p_map->cid_map)
1884 if (cid >= p_map->start_cid &&
1885 cid < p_map->start_cid + p_map->max_count)
1890 if (p == MAX_CONN_TYPES) {
1891 DP_NOTICE(p_hwfn, "Invalid CID %d", cid);
1895 rel_cid = cid - p_map->start_cid;
1896 if (!test_bit(rel_cid, p_map->cid_map)) {
1897 DP_NOTICE(p_hwfn, "CID %d not acquired", cid);
1903 void qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid)
1905 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1906 enum protocol_type type;
1910 /* Test acquired and find matching per-protocol map */
1911 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid, &type);
1916 rel_cid = cid - p_mngr->acquired[type].start_cid;
1917 __clear_bit(rel_cid, p_mngr->acquired[type].cid_map);
1920 int qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn, struct qed_cxt_info *p_info)
1922 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1923 u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
1924 enum protocol_type type;
1927 /* Test acquired and find matching per-protocol map */
1928 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, p_info->iid, &type);
1933 /* set the protocl type */
1934 p_info->type = type;
1936 /* compute context virtual pointer */
1937 hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1939 conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
1940 cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
1941 line = p_info->iid / cxts_per_p;
1943 /* Make sure context is allocated (dynamic allocation) */
1944 if (!p_mngr->ilt_shadow[line].p_virt)
1947 p_info->p_cxt = p_mngr->ilt_shadow[line].p_virt +
1948 p_info->iid % cxts_per_p * conn_cxt_size;
1950 DP_VERBOSE(p_hwfn, (QED_MSG_ILT | QED_MSG_CXT),
1951 "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
1952 p_info->iid / cxts_per_p, p_info->p_cxt, p_info->iid);
1957 static void qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn,
1958 struct qed_rdma_pf_params *p_params,
1961 u32 num_cons, num_qps, num_srqs;
1962 enum protocol_type proto;
1964 num_srqs = min_t(u32, 32 * 1024, p_params->num_srqs);
1966 switch (p_hwfn->hw_info.personality) {
1967 case QED_PCI_ETH_ROCE:
1968 num_qps = min_t(u32, ROCE_MAX_QPS, p_params->num_qps);
1969 num_cons = num_qps * 2; /* each QP requires two connections */
1970 proto = PROTOCOLID_ROCE;
1976 if (num_cons && num_tasks) {
1977 qed_cxt_set_proto_cid_count(p_hwfn, proto, num_cons, 0);
1979 /* Deliberatly passing ROCE for tasks id. This is because
1980 * iWARP / RoCE share the task id.
1982 qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
1983 QED_CXT_ROCE_TID_SEG, 1,
1985 qed_cxt_set_srq_count(p_hwfn, num_srqs);
1987 DP_INFO(p_hwfn->cdev,
1988 "RDMA personality used without setting params!\n");
1992 int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn, u32 rdma_tasks)
1994 /* Set the number of required CORE connections */
1995 u32 core_cids = 1; /* SPQ */
1997 if (p_hwfn->using_ll2)
1999 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
2001 switch (p_hwfn->hw_info.personality) {
2002 case QED_PCI_ETH_ROCE:
2004 qed_rdma_set_pf_params(p_hwfn,
2006 pf_params.rdma_pf_params,
2008 /* no need for break since RoCE coexist with Ethernet */
2012 struct qed_eth_pf_params *p_params =
2013 &p_hwfn->pf_params.eth_pf_params;
2015 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
2016 p_params->num_cons, 1);
2017 p_hwfn->p_cxt_mngr->arfs_count = p_params->num_arfs_filters;
2022 struct qed_fcoe_pf_params *p_params;
2024 p_params = &p_hwfn->pf_params.fcoe_pf_params;
2026 if (p_params->num_cons && p_params->num_tasks) {
2027 qed_cxt_set_proto_cid_count(p_hwfn,
2032 qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
2033 QED_CXT_FCOE_TID_SEG, 0,
2034 p_params->num_tasks, true);
2036 DP_INFO(p_hwfn->cdev,
2037 "Fcoe personality used without setting params!\n");
2043 struct qed_iscsi_pf_params *p_params;
2045 p_params = &p_hwfn->pf_params.iscsi_pf_params;
2047 if (p_params->num_cons && p_params->num_tasks) {
2048 qed_cxt_set_proto_cid_count(p_hwfn,
2053 qed_cxt_set_proto_tid_count(p_hwfn,
2055 QED_CXT_ISCSI_TID_SEG,
2057 p_params->num_tasks,
2060 DP_INFO(p_hwfn->cdev,
2061 "Iscsi personality used without setting params!\n");
2072 int qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn,
2073 struct qed_tid_mem *p_info)
2075 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2076 u32 proto, seg, total_lines, i, shadow_line;
2077 struct qed_ilt_client_cfg *p_cli;
2078 struct qed_ilt_cli_blk *p_fl_seg;
2079 struct qed_tid_seg *p_seg_info;
2081 /* Verify the personality */
2082 switch (p_hwfn->hw_info.personality) {
2084 proto = PROTOCOLID_FCOE;
2085 seg = QED_CXT_FCOE_TID_SEG;
2088 proto = PROTOCOLID_ISCSI;
2089 seg = QED_CXT_ISCSI_TID_SEG;
2095 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2099 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2100 if (!p_seg_info->has_fl_mem)
2103 p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2104 total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2105 p_fl_seg->real_size_in_page);
2107 for (i = 0; i < total_lines; i++) {
2108 shadow_line = i + p_fl_seg->start_line -
2109 p_hwfn->p_cxt_mngr->pf_start_line;
2110 p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
2112 p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2113 p_fl_seg->real_size_in_page;
2114 p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2115 p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2121 /* This function is very RoCE oriented, if another protocol in the future
2122 * will want this feature we'll need to modify the function to be more generic
2125 qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
2126 enum qed_cxt_elem_type elem_type, u32 iid)
2128 u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2129 struct qed_ilt_client_cfg *p_cli;
2130 struct qed_ilt_cli_blk *p_blk;
2131 struct qed_ptt *p_ptt;
2137 switch (elem_type) {
2139 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2140 elem_size = CONN_CXT_SIZE(p_hwfn);
2141 p_blk = &p_cli->pf_blks[CDUC_BLK];
2144 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2145 elem_size = SRQ_CXT_SIZE;
2146 p_blk = &p_cli->pf_blks[SRQ_BLK];
2149 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2150 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2151 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2154 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2158 /* Calculate line in ilt */
2159 hw_p_size = p_cli->p_size.val;
2160 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2161 line = p_blk->start_line + (iid / elems_per_p);
2162 shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2164 /* If line is already allocated, do nothing, otherwise allocate it and
2165 * write it to the PSWRQ2 registers.
2166 * This section can be run in parallel from different contexts and thus
2167 * a mutex protection is needed.
2170 mutex_lock(&p_hwfn->p_cxt_mngr->mutex);
2172 if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2175 p_ptt = qed_ptt_acquire(p_hwfn);
2178 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2183 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2184 p_blk->real_size_in_page,
2185 &p_phys, GFP_KERNEL);
2190 memset(p_virt, 0, p_blk->real_size_in_page);
2192 /* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2193 * to compensate for a HW bug, but it is configured even if DIF is not
2194 * enabled. This is harmless and allows us to avoid a dedicated API. We
2195 * configure the field for all of the contexts on the newly allocated
2198 if (elem_type == QED_ELEM_TASK) {
2200 u8 *elem_start = (u8 *)p_virt;
2201 union type1_task_context *elem;
2203 for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2204 elem = (union type1_task_context *)elem_start;
2205 SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2206 TDIF_TASK_CONTEXT_REFTAGMASK, 0xf);
2207 elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2211 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2212 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2213 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2214 p_blk->real_size_in_page;
2216 /* compute absolute offset */
2217 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2218 (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2221 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2222 SET_FIELD(ilt_hw_entry,
2224 (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2226 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2227 qed_dmae_host2grc(p_hwfn, p_ptt, (u64) (uintptr_t)&ilt_hw_entry,
2228 reg_offset, sizeof(ilt_hw_entry) / sizeof(u32), 0);
2230 if (elem_type == QED_ELEM_CXT) {
2231 u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2234 /* Update the relevant register in the parser */
2235 qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2236 last_cid_allocated - 1);
2238 if (!p_hwfn->b_rdma_enabled_in_prs) {
2239 /* Enable RoCE search */
2240 qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2241 p_hwfn->b_rdma_enabled_in_prs = true;
2246 qed_ptt_release(p_hwfn, p_ptt);
2248 mutex_unlock(&p_hwfn->p_cxt_mngr->mutex);
2253 /* This function is very RoCE oriented, if another protocol in the future
2254 * will want this feature we'll need to modify the function to be more generic
2257 qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn,
2258 enum qed_cxt_elem_type elem_type,
2259 u32 start_iid, u32 count)
2261 u32 start_line, end_line, shadow_start_line, shadow_end_line;
2262 u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2263 struct qed_ilt_client_cfg *p_cli;
2264 struct qed_ilt_cli_blk *p_blk;
2265 u32 end_iid = start_iid + count;
2266 struct qed_ptt *p_ptt;
2267 u64 ilt_hw_entry = 0;
2270 switch (elem_type) {
2272 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2273 elem_size = CONN_CXT_SIZE(p_hwfn);
2274 p_blk = &p_cli->pf_blks[CDUC_BLK];
2277 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2278 elem_size = SRQ_CXT_SIZE;
2279 p_blk = &p_cli->pf_blks[SRQ_BLK];
2282 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2283 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2284 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2287 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2291 /* Calculate line in ilt */
2292 hw_p_size = p_cli->p_size.val;
2293 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2294 start_line = p_blk->start_line + (start_iid / elems_per_p);
2295 end_line = p_blk->start_line + (end_iid / elems_per_p);
2296 if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2299 shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2300 shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2302 p_ptt = qed_ptt_acquire(p_hwfn);
2305 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2309 for (i = shadow_start_line; i < shadow_end_line; i++) {
2310 if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2313 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
2314 p_hwfn->p_cxt_mngr->ilt_shadow[i].size,
2315 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2316 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys);
2318 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = NULL;
2319 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2320 p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2322 /* compute absolute offset */
2323 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2324 ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2327 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2330 qed_dmae_host2grc(p_hwfn, p_ptt,
2331 (u64) (uintptr_t) &ilt_hw_entry,
2333 sizeof(ilt_hw_entry) / sizeof(u32),
2337 qed_ptt_release(p_hwfn, p_ptt);
2342 int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto)
2347 /* Free Connection CXT */
2348 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_CXT,
2349 qed_cxt_get_proto_cid_start(p_hwfn,
2351 qed_cxt_get_proto_cid_count(p_hwfn,
2358 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_TASK, 0,
2359 qed_cxt_get_proto_tid_count(p_hwfn, proto));
2364 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_SRQ, 0,
2365 qed_cxt_get_srq_count(p_hwfn));
2370 int qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn,
2371 u32 tid, u8 ctx_type, void **pp_task_ctx)
2373 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2374 struct qed_ilt_client_cfg *p_cli;
2375 struct qed_tid_seg *p_seg_info;
2376 struct qed_ilt_cli_blk *p_seg;
2377 u32 num_tids_per_block;
2378 u32 tid_size, ilt_idx;
2382 /* Verify the personality */
2383 switch (p_hwfn->hw_info.personality) {
2385 proto = PROTOCOLID_FCOE;
2386 seg = QED_CXT_FCOE_TID_SEG;
2389 proto = PROTOCOLID_ISCSI;
2390 seg = QED_CXT_ISCSI_TID_SEG;
2396 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2400 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2402 if (ctx_type == QED_CTX_WORKING_MEM) {
2403 p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2404 } else if (ctx_type == QED_CTX_FL_MEM) {
2405 if (!p_seg_info->has_fl_mem)
2407 p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2411 total_lines = DIV_ROUND_UP(p_seg->total_size, p_seg->real_size_in_page);
2412 tid_size = p_mngr->task_type_size[p_seg_info->type];
2413 num_tids_per_block = p_seg->real_size_in_page / tid_size;
2415 if (total_lines < tid / num_tids_per_block)
2418 ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2419 p_mngr->pf_start_line;
2420 *pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2421 (tid % num_tids_per_block) * tid_size;