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
222 /* total number of SRQ's for this hwfn */
226 struct qed_cid_acquired_map acquired[MAX_CONN_TYPES];
228 /* ILT shadow table */
229 struct qed_dma_mem *ilt_shadow;
232 /* Mutex for a dynamic ILT allocation */
236 struct qed_dma_mem *t2;
241 static bool src_proto(enum protocol_type type)
243 return type == PROTOCOLID_ISCSI ||
244 type == PROTOCOLID_FCOE ||
245 type == PROTOCOLID_ROCE;
248 static bool tm_cid_proto(enum protocol_type type)
250 return type == PROTOCOLID_ISCSI ||
251 type == PROTOCOLID_FCOE ||
252 type == PROTOCOLID_ROCE;
255 static bool tm_tid_proto(enum protocol_type type)
257 return type == PROTOCOLID_FCOE;
260 /* counts the iids for the CDU/CDUC ILT client configuration */
261 struct qed_cdu_iids {
266 static void qed_cxt_cdu_iids(struct qed_cxt_mngr *p_mngr,
267 struct qed_cdu_iids *iids)
271 for (type = 0; type < MAX_CONN_TYPES; type++) {
272 iids->pf_cids += p_mngr->conn_cfg[type].cid_count;
273 iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
277 /* counts the iids for the Searcher block configuration */
278 struct qed_src_iids {
283 static void qed_cxt_src_iids(struct qed_cxt_mngr *p_mngr,
284 struct qed_src_iids *iids)
288 for (i = 0; i < MAX_CONN_TYPES; i++) {
292 iids->pf_cids += p_mngr->conn_cfg[i].cid_count;
293 iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf;
297 /* counts the iids for the Timers block configuration */
300 u32 pf_tids[NUM_TASK_PF_SEGMENTS]; /* per segment */
306 static void qed_cxt_tm_iids(struct qed_cxt_mngr *p_mngr,
307 struct qed_tm_iids *iids)
311 for (i = 0; i < MAX_CONN_TYPES; i++) {
312 struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
314 if (tm_cid_proto(i)) {
315 iids->pf_cids += p_cfg->cid_count;
316 iids->per_vf_cids += p_cfg->cids_per_vf;
319 if (tm_tid_proto(i)) {
320 struct qed_tid_seg *segs = p_cfg->tid_seg;
322 /* for each segment there is at most one
323 * protocol for which count is not 0.
325 for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
326 iids->pf_tids[j] += segs[j].count;
328 /* The last array elelment is for the VFs. As for PF
329 * segments there can be only one protocol for
330 * which this value is not 0.
332 iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
336 iids->pf_cids = roundup(iids->pf_cids, TM_ALIGN);
337 iids->per_vf_cids = roundup(iids->per_vf_cids, TM_ALIGN);
338 iids->per_vf_tids = roundup(iids->per_vf_tids, TM_ALIGN);
340 for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
341 iids->pf_tids[j] = roundup(iids->pf_tids[j], TM_ALIGN);
342 iids->pf_tids_total += iids->pf_tids[j];
346 static void qed_cxt_qm_iids(struct qed_hwfn *p_hwfn,
347 struct qed_qm_iids *iids)
349 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
350 struct qed_tid_seg *segs;
351 u32 vf_cids = 0, type, j;
354 for (type = 0; type < MAX_CONN_TYPES; type++) {
355 iids->cids += p_mngr->conn_cfg[type].cid_count;
356 vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
358 segs = p_mngr->conn_cfg[type].tid_seg;
359 /* for each segment there is at most one
360 * protocol for which count is not 0.
362 for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
363 iids->tids += segs[j].count;
365 /* The last array elelment is for the VFs. As for PF
366 * segments there can be only one protocol for
367 * which this value is not 0.
369 vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
372 iids->vf_cids += vf_cids * p_mngr->vf_count;
373 iids->tids += vf_tids * p_mngr->vf_count;
375 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
376 "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
377 iids->cids, iids->vf_cids, iids->tids, vf_tids);
380 static struct qed_tid_seg *qed_cxt_tid_seg_info(struct qed_hwfn *p_hwfn,
383 struct qed_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
386 /* Find the protocol with tid count > 0 for this segment.
387 * Note: there can only be one and this is already validated.
389 for (i = 0; i < MAX_CONN_TYPES; i++)
390 if (p_cfg->conn_cfg[i].tid_seg[seg].count)
391 return &p_cfg->conn_cfg[i].tid_seg[seg];
395 static void qed_cxt_set_srq_count(struct qed_hwfn *p_hwfn, u32 num_srqs)
397 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
399 p_mgr->srq_count = num_srqs;
402 static u32 qed_cxt_get_srq_count(struct qed_hwfn *p_hwfn)
404 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
406 return p_mgr->srq_count;
409 /* set the iids count per protocol */
410 static void qed_cxt_set_proto_cid_count(struct qed_hwfn *p_hwfn,
411 enum protocol_type type,
412 u32 cid_count, u32 vf_cid_cnt)
414 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
415 struct qed_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
417 p_conn->cid_count = roundup(cid_count, DQ_RANGE_ALIGN);
418 p_conn->cids_per_vf = roundup(vf_cid_cnt, DQ_RANGE_ALIGN);
420 if (type == PROTOCOLID_ROCE) {
421 u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
422 u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
423 u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
424 u32 align = elems_per_page * DQ_RANGE_ALIGN;
426 p_conn->cid_count = roundup(p_conn->cid_count, align);
430 u32 qed_cxt_get_proto_cid_count(struct qed_hwfn *p_hwfn,
431 enum protocol_type type, u32 *vf_cid)
434 *vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
436 return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
439 u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
440 enum protocol_type type)
442 return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
445 u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
446 enum protocol_type type)
451 for (i = 0; i < TASK_SEGMENTS; i++)
452 cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
457 static void qed_cxt_set_proto_tid_count(struct qed_hwfn *p_hwfn,
458 enum protocol_type proto,
460 u8 seg_type, u32 count, bool has_fl)
462 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
463 struct qed_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
465 p_seg->count = count;
466 p_seg->has_fl_mem = has_fl;
467 p_seg->type = seg_type;
470 static void qed_ilt_cli_blk_fill(struct qed_ilt_client_cfg *p_cli,
471 struct qed_ilt_cli_blk *p_blk,
472 u32 start_line, u32 total_size, u32 elem_size)
474 u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
476 /* verify thatits called only once for each block */
477 if (p_blk->total_size)
480 p_blk->total_size = total_size;
481 p_blk->real_size_in_page = 0;
483 p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
484 p_blk->start_line = start_line;
487 static void qed_ilt_cli_adv_line(struct qed_hwfn *p_hwfn,
488 struct qed_ilt_client_cfg *p_cli,
489 struct qed_ilt_cli_blk *p_blk,
490 u32 *p_line, enum ilt_clients client_id)
492 if (!p_blk->total_size)
496 p_cli->first.val = *p_line;
498 p_cli->active = true;
499 *p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
500 p_cli->last.val = *p_line - 1;
502 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
503 "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
504 client_id, p_cli->first.val,
505 p_cli->last.val, p_blk->total_size,
506 p_blk->real_size_in_page, p_blk->start_line);
509 static u32 qed_ilt_get_dynamic_line_cnt(struct qed_hwfn *p_hwfn,
510 enum ilt_clients ilt_client)
512 u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
513 struct qed_ilt_client_cfg *p_cli;
514 u32 lines_to_skip = 0;
517 if (ilt_client == ILT_CLI_CDUC) {
518 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
520 cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
521 (u32) CONN_CXT_SIZE(p_hwfn);
523 lines_to_skip = cid_count / cxts_per_p;
526 return lines_to_skip;
529 int qed_cxt_cfg_ilt_compute(struct qed_hwfn *p_hwfn)
531 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
532 u32 curr_line, total, i, task_size, line;
533 struct qed_ilt_client_cfg *p_cli;
534 struct qed_ilt_cli_blk *p_blk;
535 struct qed_cdu_iids cdu_iids;
536 struct qed_src_iids src_iids;
537 struct qed_qm_iids qm_iids;
538 struct qed_tm_iids tm_iids;
539 struct qed_tid_seg *p_seg;
541 memset(&qm_iids, 0, sizeof(qm_iids));
542 memset(&cdu_iids, 0, sizeof(cdu_iids));
543 memset(&src_iids, 0, sizeof(src_iids));
544 memset(&tm_iids, 0, sizeof(tm_iids));
546 p_mngr->pf_start_line = RESC_START(p_hwfn, QED_ILT);
548 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
549 "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
550 p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
553 p_cli = &p_mngr->clients[ILT_CLI_CDUC];
554 curr_line = p_mngr->pf_start_line;
557 p_cli->pf_total_lines = 0;
559 /* get the counters for the CDUC and QM clients */
560 qed_cxt_cdu_iids(p_mngr, &cdu_iids);
562 p_blk = &p_cli->pf_blks[CDUC_BLK];
564 total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
566 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
567 total, CONN_CXT_SIZE(p_hwfn));
569 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
570 p_cli->pf_total_lines = curr_line - p_blk->start_line;
572 p_blk->dynamic_line_cnt = qed_ilt_get_dynamic_line_cnt(p_hwfn,
576 p_blk = &p_cli->vf_blks[CDUC_BLK];
577 total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
579 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
580 total, CONN_CXT_SIZE(p_hwfn));
582 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
583 p_cli->vf_total_lines = curr_line - p_blk->start_line;
585 for (i = 1; i < p_mngr->vf_count; i++)
586 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
590 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
591 p_cli->first.val = curr_line;
593 /* first the 'working' task memory */
594 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
595 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
596 if (!p_seg || p_seg->count == 0)
599 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(i)];
600 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
601 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
602 p_mngr->task_type_size[p_seg->type]);
604 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
608 /* next the 'init' task memory (forced load memory) */
609 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
610 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
611 if (!p_seg || p_seg->count == 0)
614 p_blk = &p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)];
616 if (!p_seg->has_fl_mem) {
617 /* The segment is active (total size pf 'working'
618 * memory is > 0) but has no FL (forced-load, Init)
621 * 1. The total-size in the corrsponding FL block of
622 * the ILT client is set to 0 - No ILT line are
623 * provisioned and no ILT memory allocated.
625 * 2. The start-line of said block is set to the
626 * start line of the matching working memory
627 * block in the ILT client. This is later used to
628 * configure the CDU segment offset registers and
629 * results in an FL command for TIDs of this
630 * segement behaves as regular load commands
631 * (loading TIDs from the working memory).
633 line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
635 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
638 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
640 qed_ilt_cli_blk_fill(p_cli, p_blk,
642 p_mngr->task_type_size[p_seg->type]);
644 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
647 p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
650 p_seg = qed_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
651 if (p_seg && p_seg->count) {
652 /* Stricly speaking we need to iterate over all VF
653 * task segment types, but a VF has only 1 segment
656 /* 'working' memory */
657 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
659 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
660 qed_ilt_cli_blk_fill(p_cli, p_blk,
662 p_mngr->task_type_size[p_seg->type]);
664 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
668 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
669 if (!p_seg->has_fl_mem) {
670 /* see comment above */
671 line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
672 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
674 task_size = p_mngr->task_type_size[p_seg->type];
675 qed_ilt_cli_blk_fill(p_cli, p_blk,
676 curr_line, total, task_size);
677 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
680 p_cli->vf_total_lines = curr_line -
681 p_cli->vf_blks[0].start_line;
683 /* Now for the rest of the VFs */
684 for (i = 1; i < p_mngr->vf_count; i++) {
685 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
686 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
689 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
690 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
696 p_cli = &p_mngr->clients[ILT_CLI_QM];
697 p_blk = &p_cli->pf_blks[0];
699 qed_cxt_qm_iids(p_hwfn, &qm_iids);
700 total = qed_qm_pf_mem_size(p_hwfn->rel_pf_id, qm_iids.cids,
701 qm_iids.vf_cids, qm_iids.tids,
702 p_hwfn->qm_info.num_pqs,
703 p_hwfn->qm_info.num_vf_pqs);
707 "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
711 p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
713 qed_ilt_cli_blk_fill(p_cli, p_blk,
714 curr_line, total * 0x1000,
717 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
718 p_cli->pf_total_lines = curr_line - p_blk->start_line;
721 p_cli = &p_mngr->clients[ILT_CLI_SRC];
722 qed_cxt_src_iids(p_mngr, &src_iids);
724 /* Both the PF and VFs searcher connections are stored in the per PF
725 * database. Thus sum the PF searcher cids and all the VFs searcher
728 total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
730 u32 local_max = max_t(u32, total,
733 total = roundup_pow_of_two(local_max);
735 p_blk = &p_cli->pf_blks[0];
736 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
737 total * sizeof(struct src_ent),
738 sizeof(struct src_ent));
740 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
742 p_cli->pf_total_lines = curr_line - p_blk->start_line;
746 p_cli = &p_mngr->clients[ILT_CLI_TM];
747 qed_cxt_tm_iids(p_mngr, &tm_iids);
748 total = tm_iids.pf_cids + tm_iids.pf_tids_total;
750 p_blk = &p_cli->pf_blks[0];
751 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
752 total * TM_ELEM_SIZE, TM_ELEM_SIZE);
754 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
756 p_cli->pf_total_lines = curr_line - p_blk->start_line;
760 total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
762 p_blk = &p_cli->vf_blks[0];
763 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
764 total * TM_ELEM_SIZE, TM_ELEM_SIZE);
766 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
768 p_cli->pf_total_lines = curr_line - p_blk->start_line;
770 for (i = 1; i < p_mngr->vf_count; i++)
771 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
775 /* TSDM (SRQ CONTEXT) */
776 total = qed_cxt_get_srq_count(p_hwfn);
779 p_cli = &p_mngr->clients[ILT_CLI_TSDM];
780 p_blk = &p_cli->pf_blks[SRQ_BLK];
781 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
782 total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
784 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
786 p_cli->pf_total_lines = curr_line - p_blk->start_line;
789 if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
790 RESC_NUM(p_hwfn, QED_ILT)) {
791 DP_ERR(p_hwfn, "too many ilt lines...#lines=%d\n",
792 curr_line - p_hwfn->p_cxt_mngr->pf_start_line);
799 static void qed_cxt_src_t2_free(struct qed_hwfn *p_hwfn)
801 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
807 for (i = 0; i < p_mngr->t2_num_pages; i++)
808 if (p_mngr->t2[i].p_virt)
809 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
811 p_mngr->t2[i].p_virt,
812 p_mngr->t2[i].p_phys);
818 static int qed_cxt_src_t2_alloc(struct qed_hwfn *p_hwfn)
820 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
821 u32 conn_num, total_size, ent_per_page, psz, i;
822 struct qed_ilt_client_cfg *p_src;
823 struct qed_src_iids src_iids;
824 struct qed_dma_mem *p_t2;
827 memset(&src_iids, 0, sizeof(src_iids));
829 /* if the SRC ILT client is inactive - there are no connection
830 * requiring the searcer, leave.
832 p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
836 qed_cxt_src_iids(p_mngr, &src_iids);
837 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
838 total_size = conn_num * sizeof(struct src_ent);
840 /* use the same page size as the SRC ILT client */
841 psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
842 p_mngr->t2_num_pages = DIV_ROUND_UP(total_size, psz);
845 p_mngr->t2 = kcalloc(p_mngr->t2_num_pages, sizeof(struct qed_dma_mem),
852 /* allocate t2 pages */
853 for (i = 0; i < p_mngr->t2_num_pages; i++) {
854 u32 size = min_t(u32, total_size, psz);
855 void **p_virt = &p_mngr->t2[i].p_virt;
857 *p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
859 &p_mngr->t2[i].p_phys, GFP_KERNEL);
860 if (!p_mngr->t2[i].p_virt) {
864 memset(*p_virt, 0, size);
865 p_mngr->t2[i].size = size;
869 /* Set the t2 pointers */
871 /* entries per page - must be a power of two */
872 ent_per_page = psz / sizeof(struct src_ent);
874 p_mngr->first_free = (u64) p_mngr->t2[0].p_phys;
876 p_t2 = &p_mngr->t2[(conn_num - 1) / ent_per_page];
877 p_mngr->last_free = (u64) p_t2->p_phys +
878 ((conn_num - 1) & (ent_per_page - 1)) * sizeof(struct src_ent);
880 for (i = 0; i < p_mngr->t2_num_pages; i++) {
881 u32 ent_num = min_t(u32,
884 struct src_ent *entries = p_mngr->t2[i].p_virt;
885 u64 p_ent_phys = (u64) p_mngr->t2[i].p_phys, val;
888 for (j = 0; j < ent_num - 1; j++) {
889 val = p_ent_phys + (j + 1) * sizeof(struct src_ent);
890 entries[j].next = cpu_to_be64(val);
893 if (i < p_mngr->t2_num_pages - 1)
894 val = (u64) p_mngr->t2[i + 1].p_phys;
897 entries[j].next = cpu_to_be64(val);
905 qed_cxt_src_t2_free(p_hwfn);
909 #define for_each_ilt_valid_client(pos, clients) \
910 for (pos = 0; pos < ILT_CLI_MAX; pos++) \
911 if (!clients[pos].active) { \
915 /* Total number of ILT lines used by this PF */
916 static u32 qed_cxt_ilt_shadow_size(struct qed_ilt_client_cfg *ilt_clients)
921 for_each_ilt_valid_client(i, ilt_clients)
922 size += (ilt_clients[i].last.val - ilt_clients[i].first.val + 1);
927 static void qed_ilt_shadow_free(struct qed_hwfn *p_hwfn)
929 struct qed_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
930 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
933 ilt_size = qed_cxt_ilt_shadow_size(p_cli);
935 for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
936 struct qed_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
939 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
940 p_dma->size, p_dma->p_virt,
942 p_dma->p_virt = NULL;
944 kfree(p_mngr->ilt_shadow);
947 static int qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn,
948 struct qed_ilt_cli_blk *p_blk,
949 enum ilt_clients ilt_client,
950 u32 start_line_offset)
952 struct qed_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
953 u32 lines, line, sz_left, lines_to_skip = 0;
955 /* Special handling for RoCE that supports dynamic allocation */
956 if ((p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE) &&
957 ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
960 lines_to_skip = p_blk->dynamic_line_cnt;
962 if (!p_blk->total_size)
965 sz_left = p_blk->total_size;
966 lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) - lines_to_skip;
967 line = p_blk->start_line + start_line_offset -
968 p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
970 for (; lines; lines--) {
975 size = min_t(u32, sz_left, p_blk->real_size_in_page);
976 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
977 size, &p_phys, GFP_KERNEL);
980 memset(p_virt, 0, size);
982 ilt_shadow[line].p_phys = p_phys;
983 ilt_shadow[line].p_virt = p_virt;
984 ilt_shadow[line].size = size;
986 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
987 "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
988 line, (u64)p_phys, p_virt, size);
997 static int qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn)
999 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1000 struct qed_ilt_client_cfg *clients = p_mngr->clients;
1001 struct qed_ilt_cli_blk *p_blk;
1005 size = qed_cxt_ilt_shadow_size(clients);
1006 p_mngr->ilt_shadow = kcalloc(size, sizeof(struct qed_dma_mem),
1008 if (!p_mngr->ilt_shadow) {
1010 goto ilt_shadow_fail;
1013 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1014 "Allocated 0x%x bytes for ilt shadow\n",
1015 (u32)(size * sizeof(struct qed_dma_mem)));
1017 for_each_ilt_valid_client(i, clients) {
1018 for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1019 p_blk = &clients[i].pf_blks[j];
1020 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1022 goto ilt_shadow_fail;
1024 for (k = 0; k < p_mngr->vf_count; k++) {
1025 for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1026 u32 lines = clients[i].vf_total_lines * k;
1028 p_blk = &clients[i].vf_blks[j];
1029 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, lines);
1031 goto ilt_shadow_fail;
1039 qed_ilt_shadow_free(p_hwfn);
1043 static void qed_cid_map_free(struct qed_hwfn *p_hwfn)
1045 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1048 for (type = 0; type < MAX_CONN_TYPES; type++) {
1049 kfree(p_mngr->acquired[type].cid_map);
1050 p_mngr->acquired[type].max_count = 0;
1051 p_mngr->acquired[type].start_cid = 0;
1055 static int qed_cid_map_alloc(struct qed_hwfn *p_hwfn)
1057 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1061 for (type = 0; type < MAX_CONN_TYPES; type++) {
1062 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1068 size = DIV_ROUND_UP(cid_cnt,
1069 sizeof(unsigned long) * BITS_PER_BYTE) *
1070 sizeof(unsigned long);
1071 p_mngr->acquired[type].cid_map = kzalloc(size, GFP_KERNEL);
1072 if (!p_mngr->acquired[type].cid_map)
1075 p_mngr->acquired[type].max_count = cid_cnt;
1076 p_mngr->acquired[type].start_cid = start_cid;
1078 p_hwfn->p_cxt_mngr->conn_cfg[type].cid_start = start_cid;
1080 DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1081 "Type %08x start: %08x count %08x\n",
1082 type, p_mngr->acquired[type].start_cid,
1083 p_mngr->acquired[type].max_count);
1084 start_cid += cid_cnt;
1090 qed_cid_map_free(p_hwfn);
1094 int qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn)
1096 struct qed_ilt_client_cfg *clients;
1097 struct qed_cxt_mngr *p_mngr;
1100 p_mngr = kzalloc(sizeof(*p_mngr), GFP_KERNEL);
1104 /* Initialize ILT client registers */
1105 clients = p_mngr->clients;
1106 clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1107 clients[ILT_CLI_CDUC].last.reg = ILT_CFG_REG(CDUC, LAST_ILT);
1108 clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1110 clients[ILT_CLI_QM].first.reg = ILT_CFG_REG(QM, FIRST_ILT);
1111 clients[ILT_CLI_QM].last.reg = ILT_CFG_REG(QM, LAST_ILT);
1112 clients[ILT_CLI_QM].p_size.reg = ILT_CFG_REG(QM, P_SIZE);
1114 clients[ILT_CLI_TM].first.reg = ILT_CFG_REG(TM, FIRST_ILT);
1115 clients[ILT_CLI_TM].last.reg = ILT_CFG_REG(TM, LAST_ILT);
1116 clients[ILT_CLI_TM].p_size.reg = ILT_CFG_REG(TM, P_SIZE);
1118 clients[ILT_CLI_SRC].first.reg = ILT_CFG_REG(SRC, FIRST_ILT);
1119 clients[ILT_CLI_SRC].last.reg = ILT_CFG_REG(SRC, LAST_ILT);
1120 clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1122 clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1123 clients[ILT_CLI_CDUT].last.reg = ILT_CFG_REG(CDUT, LAST_ILT);
1124 clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1126 clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1127 clients[ILT_CLI_TSDM].last.reg = ILT_CFG_REG(TSDM, LAST_ILT);
1128 clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1129 /* default ILT page size for all clients is 64K */
1130 for (i = 0; i < ILT_CLI_MAX; i++)
1131 p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE;
1133 /* Initialize task sizes */
1134 p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1135 p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1137 if (p_hwfn->cdev->p_iov_info)
1138 p_mngr->vf_count = p_hwfn->cdev->p_iov_info->total_vfs;
1139 /* Initialize the dynamic ILT allocation mutex */
1140 mutex_init(&p_mngr->mutex);
1142 /* Set the cxt mangr pointer priori to further allocations */
1143 p_hwfn->p_cxt_mngr = p_mngr;
1148 int qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn)
1152 /* Allocate the ILT shadow table */
1153 rc = qed_ilt_shadow_alloc(p_hwfn);
1155 goto tables_alloc_fail;
1157 /* Allocate the T2 table */
1158 rc = qed_cxt_src_t2_alloc(p_hwfn);
1160 goto tables_alloc_fail;
1162 /* Allocate and initialize the acquired cids bitmaps */
1163 rc = qed_cid_map_alloc(p_hwfn);
1165 goto tables_alloc_fail;
1170 qed_cxt_mngr_free(p_hwfn);
1174 void qed_cxt_mngr_free(struct qed_hwfn *p_hwfn)
1176 if (!p_hwfn->p_cxt_mngr)
1179 qed_cid_map_free(p_hwfn);
1180 qed_cxt_src_t2_free(p_hwfn);
1181 qed_ilt_shadow_free(p_hwfn);
1182 kfree(p_hwfn->p_cxt_mngr);
1184 p_hwfn->p_cxt_mngr = NULL;
1187 void qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn)
1189 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1192 /* Reset acquired cids */
1193 for (type = 0; type < MAX_CONN_TYPES; type++) {
1194 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1199 memset(p_mngr->acquired[type].cid_map, 0,
1200 DIV_ROUND_UP(cid_cnt,
1201 sizeof(unsigned long) * BITS_PER_BYTE) *
1202 sizeof(unsigned long));
1207 #define CDUC_CXT_SIZE_SHIFT \
1208 CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1210 #define CDUC_CXT_SIZE_MASK \
1211 (CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1213 #define CDUC_BLOCK_WASTE_SHIFT \
1214 CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1216 #define CDUC_BLOCK_WASTE_MASK \
1217 (CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1219 #define CDUC_NCIB_SHIFT \
1220 CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1222 #define CDUC_NCIB_MASK \
1223 (CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1225 #define CDUT_TYPE0_CXT_SIZE_SHIFT \
1226 CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1228 #define CDUT_TYPE0_CXT_SIZE_MASK \
1229 (CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >> \
1230 CDUT_TYPE0_CXT_SIZE_SHIFT)
1232 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT \
1233 CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1235 #define CDUT_TYPE0_BLOCK_WASTE_MASK \
1236 (CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >> \
1237 CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1239 #define CDUT_TYPE0_NCIB_SHIFT \
1240 CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1242 #define CDUT_TYPE0_NCIB_MASK \
1243 (CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >> \
1244 CDUT_TYPE0_NCIB_SHIFT)
1246 #define CDUT_TYPE1_CXT_SIZE_SHIFT \
1247 CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1249 #define CDUT_TYPE1_CXT_SIZE_MASK \
1250 (CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >> \
1251 CDUT_TYPE1_CXT_SIZE_SHIFT)
1253 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT \
1254 CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1256 #define CDUT_TYPE1_BLOCK_WASTE_MASK \
1257 (CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >> \
1258 CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1260 #define CDUT_TYPE1_NCIB_SHIFT \
1261 CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1263 #define CDUT_TYPE1_NCIB_MASK \
1264 (CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >> \
1265 CDUT_TYPE1_NCIB_SHIFT)
1267 static void qed_cdu_init_common(struct qed_hwfn *p_hwfn)
1269 u32 page_sz, elems_per_page, block_waste, cxt_size, cdu_params = 0;
1271 /* CDUC - connection configuration */
1272 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1273 cxt_size = CONN_CXT_SIZE(p_hwfn);
1274 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1275 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1277 SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1278 SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1279 SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
1280 STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1282 /* CDUT - type-0 tasks configuration */
1283 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1284 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1285 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1286 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1288 /* cxt size and block-waste are multipes of 8 */
1290 SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1291 SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1292 SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1293 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1295 /* CDUT - type-1 tasks configuration */
1296 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1297 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1298 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1300 /* cxt size and block-waste are multipes of 8 */
1302 SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1303 SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1304 SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1305 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1309 #define CDU_SEG_REG_TYPE_SHIFT CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1310 #define CDU_SEG_REG_TYPE_MASK 0x1
1311 #define CDU_SEG_REG_OFFSET_SHIFT 0
1312 #define CDU_SEG_REG_OFFSET_MASK CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1314 static void qed_cdu_init_pf(struct qed_hwfn *p_hwfn)
1316 struct qed_ilt_client_cfg *p_cli;
1317 struct qed_tid_seg *p_seg;
1318 u32 cdu_seg_params, offset;
1321 static const u32 rt_type_offset_arr[] = {
1322 CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1323 CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1324 CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1325 CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1328 static const u32 rt_type_offset_fl_arr[] = {
1329 CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1330 CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1331 CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1332 CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1335 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1337 /* There are initializations only for CDUT during pf Phase */
1338 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1340 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
1344 /* Note: start_line is already adjusted for the CDU
1345 * segment register granularity, so we just need to
1346 * divide. Adjustment is implicit as we assume ILT
1347 * Page size is larger than 32K!
1349 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1350 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1351 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1354 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1355 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1356 STORE_RT_REG(p_hwfn, rt_type_offset_arr[i], cdu_seg_params);
1358 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1359 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1360 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1363 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1364 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1365 STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i], cdu_seg_params);
1369 void qed_qm_init_pf(struct qed_hwfn *p_hwfn)
1371 struct qed_qm_pf_rt_init_params params;
1372 struct qed_qm_info *qm_info = &p_hwfn->qm_info;
1373 struct qed_qm_iids iids;
1375 memset(&iids, 0, sizeof(iids));
1376 qed_cxt_qm_iids(p_hwfn, &iids);
1378 memset(¶ms, 0, sizeof(params));
1379 params.port_id = p_hwfn->port_id;
1380 params.pf_id = p_hwfn->rel_pf_id;
1381 params.max_phys_tcs_per_port = qm_info->max_phys_tcs_per_port;
1382 params.is_first_pf = p_hwfn->first_on_engine;
1383 params.num_pf_cids = iids.cids;
1384 params.num_vf_cids = iids.vf_cids;
1385 params.start_pq = qm_info->start_pq;
1386 params.num_pf_pqs = qm_info->num_pqs - qm_info->num_vf_pqs;
1387 params.num_vf_pqs = qm_info->num_vf_pqs;
1388 params.start_vport = qm_info->start_vport;
1389 params.num_vports = qm_info->num_vports;
1390 params.pf_wfq = qm_info->pf_wfq;
1391 params.pf_rl = qm_info->pf_rl;
1392 params.pq_params = qm_info->qm_pq_params;
1393 params.vport_params = qm_info->qm_vport_params;
1395 qed_qm_pf_rt_init(p_hwfn, p_hwfn->p_main_ptt, ¶ms);
1399 void qed_cm_init_pf(struct qed_hwfn *p_hwfn)
1401 /* XCM pure-LB queue */
1402 STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET,
1403 qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1407 static void qed_dq_init_pf(struct qed_hwfn *p_hwfn)
1409 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1410 u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1412 dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1413 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1415 dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1416 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1418 dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1419 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1421 dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1422 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1424 dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1425 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1427 dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1428 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1430 dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1431 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1433 dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1434 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1436 dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1437 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1439 dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1440 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1442 dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1443 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1445 dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1446 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1448 /* Connection types 6 & 7 are not in use, yet they must be configured
1449 * as the highest possible connection. Not configuring them means the
1450 * defaults will be used, and with a large number of cids a bug may
1451 * occur, if the defaults will be smaller than dq_pf_max_cid /
1454 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1455 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1457 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1458 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1461 static void qed_ilt_bounds_init(struct qed_hwfn *p_hwfn)
1463 struct qed_ilt_client_cfg *ilt_clients;
1466 ilt_clients = p_hwfn->p_cxt_mngr->clients;
1467 for_each_ilt_valid_client(i, ilt_clients) {
1468 STORE_RT_REG(p_hwfn,
1469 ilt_clients[i].first.reg,
1470 ilt_clients[i].first.val);
1471 STORE_RT_REG(p_hwfn,
1472 ilt_clients[i].last.reg, ilt_clients[i].last.val);
1473 STORE_RT_REG(p_hwfn,
1474 ilt_clients[i].p_size.reg,
1475 ilt_clients[i].p_size.val);
1479 static void qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn)
1481 struct qed_ilt_client_cfg *p_cli;
1484 /* For simplicty we set the 'block' to be an ILT page */
1485 if (p_hwfn->cdev->p_iov_info) {
1486 struct qed_hw_sriov_info *p_iov = p_hwfn->cdev->p_iov_info;
1488 STORE_RT_REG(p_hwfn,
1489 PSWRQ2_REG_VF_BASE_RT_OFFSET,
1490 p_iov->first_vf_in_pf);
1491 STORE_RT_REG(p_hwfn,
1492 PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1493 p_iov->first_vf_in_pf + p_iov->total_vfs);
1496 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1497 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1498 if (p_cli->active) {
1499 STORE_RT_REG(p_hwfn,
1500 PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1502 STORE_RT_REG(p_hwfn,
1503 PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1504 p_cli->pf_total_lines);
1505 STORE_RT_REG(p_hwfn,
1506 PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1507 p_cli->vf_total_lines);
1510 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1511 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1512 if (p_cli->active) {
1513 STORE_RT_REG(p_hwfn,
1514 PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1516 STORE_RT_REG(p_hwfn,
1517 PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1518 p_cli->pf_total_lines);
1519 STORE_RT_REG(p_hwfn,
1520 PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1521 p_cli->vf_total_lines);
1524 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1525 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1526 if (p_cli->active) {
1527 STORE_RT_REG(p_hwfn,
1528 PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET, blk_factor);
1529 STORE_RT_REG(p_hwfn,
1530 PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1531 p_cli->pf_total_lines);
1532 STORE_RT_REG(p_hwfn,
1533 PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1534 p_cli->vf_total_lines);
1538 /* ILT (PSWRQ2) PF */
1539 static void qed_ilt_init_pf(struct qed_hwfn *p_hwfn)
1541 struct qed_ilt_client_cfg *clients;
1542 struct qed_cxt_mngr *p_mngr;
1543 struct qed_dma_mem *p_shdw;
1544 u32 line, rt_offst, i;
1546 qed_ilt_bounds_init(p_hwfn);
1547 qed_ilt_vf_bounds_init(p_hwfn);
1549 p_mngr = p_hwfn->p_cxt_mngr;
1550 p_shdw = p_mngr->ilt_shadow;
1551 clients = p_hwfn->p_cxt_mngr->clients;
1553 for_each_ilt_valid_client(i, clients) {
1554 /** Client's 1st val and RT array are absolute, ILT shadows'
1555 * lines are relative.
1557 line = clients[i].first.val - p_mngr->pf_start_line;
1558 rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1559 clients[i].first.val * ILT_ENTRY_IN_REGS;
1561 for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1562 line++, rt_offst += ILT_ENTRY_IN_REGS) {
1563 u64 ilt_hw_entry = 0;
1565 /** p_virt could be NULL incase of dynamic
1568 if (p_shdw[line].p_virt) {
1569 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1570 SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1571 (p_shdw[line].p_phys >> 12));
1573 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1574 "Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1576 (u64)(p_shdw[line].p_phys >> 12));
1579 STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1584 /* SRC (Searcher) PF */
1585 static void qed_src_init_pf(struct qed_hwfn *p_hwfn)
1587 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1588 u32 rounded_conn_num, conn_num, conn_max;
1589 struct qed_src_iids src_iids;
1591 memset(&src_iids, 0, sizeof(src_iids));
1592 qed_cxt_src_iids(p_mngr, &src_iids);
1593 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1597 conn_max = max_t(u32, conn_num, SRC_MIN_NUM_ELEMS);
1598 rounded_conn_num = roundup_pow_of_two(conn_max);
1600 STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1601 STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1602 ilog2(rounded_conn_num));
1604 STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1605 p_hwfn->p_cxt_mngr->first_free);
1606 STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1607 p_hwfn->p_cxt_mngr->last_free);
1611 #define TM_CFG_NUM_IDS_SHIFT 0
1612 #define TM_CFG_NUM_IDS_MASK 0xFFFFULL
1613 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT 16
1614 #define TM_CFG_PRE_SCAN_OFFSET_MASK 0x1FFULL
1615 #define TM_CFG_PARENT_PF_SHIFT 25
1616 #define TM_CFG_PARENT_PF_MASK 0x7ULL
1618 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT 30
1619 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK 0x1FFULL
1621 #define TM_CFG_TID_OFFSET_SHIFT 30
1622 #define TM_CFG_TID_OFFSET_MASK 0x7FFFFULL
1623 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT 49
1624 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK 0x1FFULL
1626 static void qed_tm_init_pf(struct qed_hwfn *p_hwfn)
1628 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1629 u32 active_seg_mask = 0, tm_offset, rt_reg;
1630 struct qed_tm_iids tm_iids;
1634 memset(&tm_iids, 0, sizeof(tm_iids));
1635 qed_cxt_tm_iids(p_mngr, &tm_iids);
1637 /* @@@TBD No pre-scan for now */
1639 /* Note: We assume consecutive VFs for a PF */
1640 for (i = 0; i < p_mngr->vf_count; i++) {
1642 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1643 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1644 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1645 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0);
1646 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1647 (sizeof(cfg_word) / sizeof(u32)) *
1648 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1649 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1653 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1654 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1655 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0); /* n/a for PF */
1656 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1658 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1659 (sizeof(cfg_word) / sizeof(u32)) *
1660 (NUM_OF_VFS(p_hwfn->cdev) + p_hwfn->rel_pf_id);
1661 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1664 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1665 tm_iids.pf_cids ? 0x1 : 0x0);
1667 /* @@@TBD how to enable the scan for the VFs */
1669 tm_offset = tm_iids.per_vf_cids;
1671 /* Note: We assume consecutive VFs for a PF */
1672 for (i = 0; i < p_mngr->vf_count; i++) {
1674 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1675 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1676 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1677 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1678 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1680 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1681 (sizeof(cfg_word) / sizeof(u32)) *
1682 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1684 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1687 tm_offset = tm_iids.pf_cids;
1688 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1690 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1691 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1692 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1693 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1694 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1696 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1697 (sizeof(cfg_word) / sizeof(u32)) *
1698 (NUM_OF_VFS(p_hwfn->cdev) +
1699 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1701 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1702 active_seg_mask |= (tm_iids.pf_tids[i] ? BIT(i) : 0);
1704 tm_offset += tm_iids.pf_tids[i];
1707 if (p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE)
1708 active_seg_mask = 0;
1710 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1712 /* @@@TBD how to enable the scan for the VFs */
1715 static void qed_prs_init_common(struct qed_hwfn *p_hwfn)
1717 if ((p_hwfn->hw_info.personality == QED_PCI_FCOE) &&
1718 p_hwfn->pf_params.fcoe_pf_params.is_target)
1719 STORE_RT_REG(p_hwfn,
1720 PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1723 static void qed_prs_init_pf(struct qed_hwfn *p_hwfn)
1725 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1726 struct qed_conn_type_cfg *p_fcoe;
1727 struct qed_tid_seg *p_tid;
1729 p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1731 /* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1732 if (!p_fcoe->cid_count)
1735 p_tid = &p_fcoe->tid_seg[QED_CXT_FCOE_TID_SEG];
1736 if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1737 STORE_RT_REG_AGG(p_hwfn,
1738 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1741 STORE_RT_REG_AGG(p_hwfn,
1742 PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1747 void qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn)
1749 qed_cdu_init_common(p_hwfn);
1750 qed_prs_init_common(p_hwfn);
1753 void qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn)
1755 qed_qm_init_pf(p_hwfn);
1756 qed_cm_init_pf(p_hwfn);
1757 qed_dq_init_pf(p_hwfn);
1758 qed_cdu_init_pf(p_hwfn);
1759 qed_ilt_init_pf(p_hwfn);
1760 qed_src_init_pf(p_hwfn);
1761 qed_tm_init_pf(p_hwfn);
1762 qed_prs_init_pf(p_hwfn);
1765 int qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1766 enum protocol_type type, u32 *p_cid)
1768 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1771 if (type >= MAX_CONN_TYPES || !p_mngr->acquired[type].cid_map) {
1772 DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1776 rel_cid = find_first_zero_bit(p_mngr->acquired[type].cid_map,
1777 p_mngr->acquired[type].max_count);
1779 if (rel_cid >= p_mngr->acquired[type].max_count) {
1780 DP_NOTICE(p_hwfn, "no CID available for protocol %d\n", type);
1784 __set_bit(rel_cid, p_mngr->acquired[type].cid_map);
1786 *p_cid = rel_cid + p_mngr->acquired[type].start_cid;
1791 static bool qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn,
1792 u32 cid, enum protocol_type *p_type)
1794 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1795 struct qed_cid_acquired_map *p_map;
1796 enum protocol_type p;
1799 /* Iterate over protocols and find matching cid range */
1800 for (p = 0; p < MAX_CONN_TYPES; p++) {
1801 p_map = &p_mngr->acquired[p];
1803 if (!p_map->cid_map)
1805 if (cid >= p_map->start_cid &&
1806 cid < p_map->start_cid + p_map->max_count)
1811 if (p == MAX_CONN_TYPES) {
1812 DP_NOTICE(p_hwfn, "Invalid CID %d", cid);
1816 rel_cid = cid - p_map->start_cid;
1817 if (!test_bit(rel_cid, p_map->cid_map)) {
1818 DP_NOTICE(p_hwfn, "CID %d not acquired", cid);
1824 void qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid)
1826 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1827 enum protocol_type type;
1831 /* Test acquired and find matching per-protocol map */
1832 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid, &type);
1837 rel_cid = cid - p_mngr->acquired[type].start_cid;
1838 __clear_bit(rel_cid, p_mngr->acquired[type].cid_map);
1841 int qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn, struct qed_cxt_info *p_info)
1843 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1844 u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
1845 enum protocol_type type;
1848 /* Test acquired and find matching per-protocol map */
1849 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, p_info->iid, &type);
1854 /* set the protocl type */
1855 p_info->type = type;
1857 /* compute context virtual pointer */
1858 hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1860 conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
1861 cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
1862 line = p_info->iid / cxts_per_p;
1864 /* Make sure context is allocated (dynamic allocation) */
1865 if (!p_mngr->ilt_shadow[line].p_virt)
1868 p_info->p_cxt = p_mngr->ilt_shadow[line].p_virt +
1869 p_info->iid % cxts_per_p * conn_cxt_size;
1871 DP_VERBOSE(p_hwfn, (QED_MSG_ILT | QED_MSG_CXT),
1872 "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
1873 p_info->iid / cxts_per_p, p_info->p_cxt, p_info->iid);
1878 static void qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn,
1879 struct qed_rdma_pf_params *p_params)
1881 u32 num_cons, num_tasks, num_qps, num_mrs, num_srqs;
1882 enum protocol_type proto;
1884 num_mrs = min_t(u32, RDMA_MAX_TIDS, p_params->num_mrs);
1885 num_tasks = num_mrs; /* each mr uses a single task id */
1886 num_srqs = min_t(u32, 32 * 1024, p_params->num_srqs);
1888 switch (p_hwfn->hw_info.personality) {
1889 case QED_PCI_ETH_ROCE:
1890 num_qps = min_t(u32, ROCE_MAX_QPS, p_params->num_qps);
1891 num_cons = num_qps * 2; /* each QP requires two connections */
1892 proto = PROTOCOLID_ROCE;
1898 if (num_cons && num_tasks) {
1899 qed_cxt_set_proto_cid_count(p_hwfn, proto, num_cons, 0);
1901 /* Deliberatly passing ROCE for tasks id. This is because
1902 * iWARP / RoCE share the task id.
1904 qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
1905 QED_CXT_ROCE_TID_SEG, 1,
1907 qed_cxt_set_srq_count(p_hwfn, num_srqs);
1909 DP_INFO(p_hwfn->cdev,
1910 "RDMA personality used without setting params!\n");
1914 int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn)
1916 /* Set the number of required CORE connections */
1917 u32 core_cids = 1; /* SPQ */
1919 if (p_hwfn->using_ll2)
1921 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
1923 switch (p_hwfn->hw_info.personality) {
1924 case QED_PCI_ETH_ROCE:
1926 qed_rdma_set_pf_params(p_hwfn,
1928 pf_params.rdma_pf_params);
1929 /* no need for break since RoCE coexist with Ethernet */
1933 struct qed_eth_pf_params *p_params =
1934 &p_hwfn->pf_params.eth_pf_params;
1936 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
1937 p_params->num_cons, 1);
1942 struct qed_fcoe_pf_params *p_params;
1944 p_params = &p_hwfn->pf_params.fcoe_pf_params;
1946 if (p_params->num_cons && p_params->num_tasks) {
1947 qed_cxt_set_proto_cid_count(p_hwfn,
1952 qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
1953 QED_CXT_FCOE_TID_SEG, 0,
1954 p_params->num_tasks, true);
1956 DP_INFO(p_hwfn->cdev,
1957 "Fcoe personality used without setting params!\n");
1963 struct qed_iscsi_pf_params *p_params;
1965 p_params = &p_hwfn->pf_params.iscsi_pf_params;
1967 if (p_params->num_cons && p_params->num_tasks) {
1968 qed_cxt_set_proto_cid_count(p_hwfn,
1973 qed_cxt_set_proto_tid_count(p_hwfn,
1975 QED_CXT_ISCSI_TID_SEG,
1977 p_params->num_tasks,
1980 DP_INFO(p_hwfn->cdev,
1981 "Iscsi personality used without setting params!\n");
1992 int qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn,
1993 struct qed_tid_mem *p_info)
1995 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1996 u32 proto, seg, total_lines, i, shadow_line;
1997 struct qed_ilt_client_cfg *p_cli;
1998 struct qed_ilt_cli_blk *p_fl_seg;
1999 struct qed_tid_seg *p_seg_info;
2001 /* Verify the personality */
2002 switch (p_hwfn->hw_info.personality) {
2004 proto = PROTOCOLID_FCOE;
2005 seg = QED_CXT_FCOE_TID_SEG;
2008 proto = PROTOCOLID_ISCSI;
2009 seg = QED_CXT_ISCSI_TID_SEG;
2015 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2019 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2020 if (!p_seg_info->has_fl_mem)
2023 p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2024 total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2025 p_fl_seg->real_size_in_page);
2027 for (i = 0; i < total_lines; i++) {
2028 shadow_line = i + p_fl_seg->start_line -
2029 p_hwfn->p_cxt_mngr->pf_start_line;
2030 p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
2032 p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2033 p_fl_seg->real_size_in_page;
2034 p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2035 p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2041 /* This function is very RoCE oriented, if another protocol in the future
2042 * will want this feature we'll need to modify the function to be more generic
2045 qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
2046 enum qed_cxt_elem_type elem_type, u32 iid)
2048 u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2049 struct qed_ilt_client_cfg *p_cli;
2050 struct qed_ilt_cli_blk *p_blk;
2051 struct qed_ptt *p_ptt;
2057 switch (elem_type) {
2059 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2060 elem_size = CONN_CXT_SIZE(p_hwfn);
2061 p_blk = &p_cli->pf_blks[CDUC_BLK];
2064 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2065 elem_size = SRQ_CXT_SIZE;
2066 p_blk = &p_cli->pf_blks[SRQ_BLK];
2069 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2070 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2071 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2074 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2078 /* Calculate line in ilt */
2079 hw_p_size = p_cli->p_size.val;
2080 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2081 line = p_blk->start_line + (iid / elems_per_p);
2082 shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2084 /* If line is already allocated, do nothing, otherwise allocate it and
2085 * write it to the PSWRQ2 registers.
2086 * This section can be run in parallel from different contexts and thus
2087 * a mutex protection is needed.
2090 mutex_lock(&p_hwfn->p_cxt_mngr->mutex);
2092 if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2095 p_ptt = qed_ptt_acquire(p_hwfn);
2098 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2103 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2104 p_blk->real_size_in_page,
2105 &p_phys, GFP_KERNEL);
2110 memset(p_virt, 0, p_blk->real_size_in_page);
2112 /* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2113 * to compensate for a HW bug, but it is configured even if DIF is not
2114 * enabled. This is harmless and allows us to avoid a dedicated API. We
2115 * configure the field for all of the contexts on the newly allocated
2118 if (elem_type == QED_ELEM_TASK) {
2120 u8 *elem_start = (u8 *)p_virt;
2121 union type1_task_context *elem;
2123 for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2124 elem = (union type1_task_context *)elem_start;
2125 SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2126 TDIF_TASK_CONTEXT_REFTAGMASK, 0xf);
2127 elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2131 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2132 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2133 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2134 p_blk->real_size_in_page;
2136 /* compute absolute offset */
2137 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2138 (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2141 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2142 SET_FIELD(ilt_hw_entry,
2144 (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2146 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2147 qed_dmae_host2grc(p_hwfn, p_ptt, (u64) (uintptr_t)&ilt_hw_entry,
2148 reg_offset, sizeof(ilt_hw_entry) / sizeof(u32), 0);
2150 if (elem_type == QED_ELEM_CXT) {
2151 u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2154 /* Update the relevant register in the parser */
2155 qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2156 last_cid_allocated - 1);
2158 if (!p_hwfn->b_rdma_enabled_in_prs) {
2159 /* Enable RoCE search */
2160 qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2161 p_hwfn->b_rdma_enabled_in_prs = true;
2166 qed_ptt_release(p_hwfn, p_ptt);
2168 mutex_unlock(&p_hwfn->p_cxt_mngr->mutex);
2173 /* This function is very RoCE oriented, if another protocol in the future
2174 * will want this feature we'll need to modify the function to be more generic
2177 qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn,
2178 enum qed_cxt_elem_type elem_type,
2179 u32 start_iid, u32 count)
2181 u32 start_line, end_line, shadow_start_line, shadow_end_line;
2182 u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2183 struct qed_ilt_client_cfg *p_cli;
2184 struct qed_ilt_cli_blk *p_blk;
2185 u32 end_iid = start_iid + count;
2186 struct qed_ptt *p_ptt;
2187 u64 ilt_hw_entry = 0;
2190 switch (elem_type) {
2192 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2193 elem_size = CONN_CXT_SIZE(p_hwfn);
2194 p_blk = &p_cli->pf_blks[CDUC_BLK];
2197 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2198 elem_size = SRQ_CXT_SIZE;
2199 p_blk = &p_cli->pf_blks[SRQ_BLK];
2202 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2203 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2204 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2207 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2211 /* Calculate line in ilt */
2212 hw_p_size = p_cli->p_size.val;
2213 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2214 start_line = p_blk->start_line + (start_iid / elems_per_p);
2215 end_line = p_blk->start_line + (end_iid / elems_per_p);
2216 if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2219 shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2220 shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2222 p_ptt = qed_ptt_acquire(p_hwfn);
2225 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2229 for (i = shadow_start_line; i < shadow_end_line; i++) {
2230 if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2233 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
2234 p_hwfn->p_cxt_mngr->ilt_shadow[i].size,
2235 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2236 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys);
2238 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = NULL;
2239 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2240 p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2242 /* compute absolute offset */
2243 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2244 ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2247 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2250 qed_dmae_host2grc(p_hwfn, p_ptt,
2251 (u64) (uintptr_t) &ilt_hw_entry,
2253 sizeof(ilt_hw_entry) / sizeof(u32),
2257 qed_ptt_release(p_hwfn, p_ptt);
2262 int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto)
2267 /* Free Connection CXT */
2268 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_CXT,
2269 qed_cxt_get_proto_cid_start(p_hwfn,
2271 qed_cxt_get_proto_cid_count(p_hwfn,
2278 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_TASK, 0,
2279 qed_cxt_get_proto_tid_count(p_hwfn, proto));
2284 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_SRQ, 0,
2285 qed_cxt_get_srq_count(p_hwfn));
2290 int qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn,
2291 u32 tid, u8 ctx_type, void **pp_task_ctx)
2293 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2294 struct qed_ilt_client_cfg *p_cli;
2295 struct qed_tid_seg *p_seg_info;
2296 struct qed_ilt_cli_blk *p_seg;
2297 u32 num_tids_per_block;
2298 u32 tid_size, ilt_idx;
2302 /* Verify the personality */
2303 switch (p_hwfn->hw_info.personality) {
2305 proto = PROTOCOLID_FCOE;
2306 seg = QED_CXT_FCOE_TID_SEG;
2309 proto = PROTOCOLID_ISCSI;
2310 seg = QED_CXT_ISCSI_TID_SEG;
2316 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2320 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2322 if (ctx_type == QED_CTX_WORKING_MEM) {
2323 p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2324 } else if (ctx_type == QED_CTX_FL_MEM) {
2325 if (!p_seg_info->has_fl_mem)
2327 p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2331 total_lines = DIV_ROUND_UP(p_seg->total_size, p_seg->real_size_in_page);
2332 tid_size = p_mngr->task_type_size[p_seg_info->type];
2333 num_tids_per_block = p_seg->real_size_in_page / tid_size;
2335 if (total_lines < tid / num_tids_per_block)
2338 ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2339 p_mngr->pf_start_line;
2340 *pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2341 (tid % num_tids_per_block) * tid_size;