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_hwfn *p_hwfn,
307 struct qed_cxt_mngr *p_mngr,
308 struct qed_tm_iids *iids)
310 bool tm_vf_required = false;
311 bool tm_required = false;
314 /* Timers is a special case -> we don't count how many cids require
315 * timers but what's the max cid that will be used by the timer block.
316 * therefore we traverse in reverse order, and once we hit a protocol
317 * that requires the timers memory, we'll sum all the protocols up
320 for (i = MAX_CONN_TYPES - 1; i >= 0; i--) {
321 struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i];
323 if (tm_cid_proto(i) || tm_required) {
324 if (p_cfg->cid_count)
327 iids->pf_cids += p_cfg->cid_count;
330 if (tm_cid_proto(i) || tm_vf_required) {
331 if (p_cfg->cids_per_vf)
332 tm_vf_required = true;
334 iids->per_vf_cids += p_cfg->cids_per_vf;
337 if (tm_tid_proto(i)) {
338 struct qed_tid_seg *segs = p_cfg->tid_seg;
340 /* for each segment there is at most one
341 * protocol for which count is not 0.
343 for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
344 iids->pf_tids[j] += segs[j].count;
346 /* The last array elelment is for the VFs. As for PF
347 * segments there can be only one protocol for
348 * which this value is not 0.
350 iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
354 iids->pf_cids = roundup(iids->pf_cids, TM_ALIGN);
355 iids->per_vf_cids = roundup(iids->per_vf_cids, TM_ALIGN);
356 iids->per_vf_tids = roundup(iids->per_vf_tids, TM_ALIGN);
358 for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) {
359 iids->pf_tids[j] = roundup(iids->pf_tids[j], TM_ALIGN);
360 iids->pf_tids_total += iids->pf_tids[j];
364 static void qed_cxt_qm_iids(struct qed_hwfn *p_hwfn,
365 struct qed_qm_iids *iids)
367 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
368 struct qed_tid_seg *segs;
369 u32 vf_cids = 0, type, j;
372 for (type = 0; type < MAX_CONN_TYPES; type++) {
373 iids->cids += p_mngr->conn_cfg[type].cid_count;
374 vf_cids += p_mngr->conn_cfg[type].cids_per_vf;
376 segs = p_mngr->conn_cfg[type].tid_seg;
377 /* for each segment there is at most one
378 * protocol for which count is not 0.
380 for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++)
381 iids->tids += segs[j].count;
383 /* The last array elelment is for the VFs. As for PF
384 * segments there can be only one protocol for
385 * which this value is not 0.
387 vf_tids += segs[NUM_TASK_PF_SEGMENTS].count;
390 iids->vf_cids += vf_cids * p_mngr->vf_count;
391 iids->tids += vf_tids * p_mngr->vf_count;
393 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
394 "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n",
395 iids->cids, iids->vf_cids, iids->tids, vf_tids);
398 static struct qed_tid_seg *qed_cxt_tid_seg_info(struct qed_hwfn *p_hwfn,
401 struct qed_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr;
404 /* Find the protocol with tid count > 0 for this segment.
405 * Note: there can only be one and this is already validated.
407 for (i = 0; i < MAX_CONN_TYPES; i++)
408 if (p_cfg->conn_cfg[i].tid_seg[seg].count)
409 return &p_cfg->conn_cfg[i].tid_seg[seg];
413 static void qed_cxt_set_srq_count(struct qed_hwfn *p_hwfn, u32 num_srqs)
415 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
417 p_mgr->srq_count = num_srqs;
420 static u32 qed_cxt_get_srq_count(struct qed_hwfn *p_hwfn)
422 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
424 return p_mgr->srq_count;
427 /* set the iids count per protocol */
428 static void qed_cxt_set_proto_cid_count(struct qed_hwfn *p_hwfn,
429 enum protocol_type type,
430 u32 cid_count, u32 vf_cid_cnt)
432 struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr;
433 struct qed_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type];
435 p_conn->cid_count = roundup(cid_count, DQ_RANGE_ALIGN);
436 p_conn->cids_per_vf = roundup(vf_cid_cnt, DQ_RANGE_ALIGN);
438 if (type == PROTOCOLID_ROCE) {
439 u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val;
440 u32 cxt_size = CONN_CXT_SIZE(p_hwfn);
441 u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
442 u32 align = elems_per_page * DQ_RANGE_ALIGN;
444 p_conn->cid_count = roundup(p_conn->cid_count, align);
448 u32 qed_cxt_get_proto_cid_count(struct qed_hwfn *p_hwfn,
449 enum protocol_type type, u32 *vf_cid)
452 *vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf;
454 return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
457 u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn,
458 enum protocol_type type)
460 return p_hwfn->p_cxt_mngr->acquired[type].start_cid;
463 u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn,
464 enum protocol_type type)
469 for (i = 0; i < TASK_SEGMENTS; i++)
470 cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count;
475 static void qed_cxt_set_proto_tid_count(struct qed_hwfn *p_hwfn,
476 enum protocol_type proto,
478 u8 seg_type, u32 count, bool has_fl)
480 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
481 struct qed_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg];
483 p_seg->count = count;
484 p_seg->has_fl_mem = has_fl;
485 p_seg->type = seg_type;
488 static void qed_ilt_cli_blk_fill(struct qed_ilt_client_cfg *p_cli,
489 struct qed_ilt_cli_blk *p_blk,
490 u32 start_line, u32 total_size, u32 elem_size)
492 u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val);
494 /* verify thatits called only once for each block */
495 if (p_blk->total_size)
498 p_blk->total_size = total_size;
499 p_blk->real_size_in_page = 0;
501 p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size;
502 p_blk->start_line = start_line;
505 static void qed_ilt_cli_adv_line(struct qed_hwfn *p_hwfn,
506 struct qed_ilt_client_cfg *p_cli,
507 struct qed_ilt_cli_blk *p_blk,
508 u32 *p_line, enum ilt_clients client_id)
510 if (!p_blk->total_size)
514 p_cli->first.val = *p_line;
516 p_cli->active = true;
517 *p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page);
518 p_cli->last.val = *p_line - 1;
520 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
521 "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n",
522 client_id, p_cli->first.val,
523 p_cli->last.val, p_blk->total_size,
524 p_blk->real_size_in_page, p_blk->start_line);
527 static u32 qed_ilt_get_dynamic_line_cnt(struct qed_hwfn *p_hwfn,
528 enum ilt_clients ilt_client)
530 u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count;
531 struct qed_ilt_client_cfg *p_cli;
532 u32 lines_to_skip = 0;
535 if (ilt_client == ILT_CLI_CDUC) {
536 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
538 cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) /
539 (u32) CONN_CXT_SIZE(p_hwfn);
541 lines_to_skip = cid_count / cxts_per_p;
544 return lines_to_skip;
547 int qed_cxt_cfg_ilt_compute(struct qed_hwfn *p_hwfn)
549 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
550 u32 curr_line, total, i, task_size, line;
551 struct qed_ilt_client_cfg *p_cli;
552 struct qed_ilt_cli_blk *p_blk;
553 struct qed_cdu_iids cdu_iids;
554 struct qed_src_iids src_iids;
555 struct qed_qm_iids qm_iids;
556 struct qed_tm_iids tm_iids;
557 struct qed_tid_seg *p_seg;
559 memset(&qm_iids, 0, sizeof(qm_iids));
560 memset(&cdu_iids, 0, sizeof(cdu_iids));
561 memset(&src_iids, 0, sizeof(src_iids));
562 memset(&tm_iids, 0, sizeof(tm_iids));
564 p_mngr->pf_start_line = RESC_START(p_hwfn, QED_ILT);
566 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
567 "hwfn [%d] - Set context manager starting line to be 0x%08x\n",
568 p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line);
571 p_cli = &p_mngr->clients[ILT_CLI_CDUC];
572 curr_line = p_mngr->pf_start_line;
575 p_cli->pf_total_lines = 0;
577 /* get the counters for the CDUC and QM clients */
578 qed_cxt_cdu_iids(p_mngr, &cdu_iids);
580 p_blk = &p_cli->pf_blks[CDUC_BLK];
582 total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn);
584 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
585 total, CONN_CXT_SIZE(p_hwfn));
587 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
588 p_cli->pf_total_lines = curr_line - p_blk->start_line;
590 p_blk->dynamic_line_cnt = qed_ilt_get_dynamic_line_cnt(p_hwfn,
594 p_blk = &p_cli->vf_blks[CDUC_BLK];
595 total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn);
597 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
598 total, CONN_CXT_SIZE(p_hwfn));
600 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_CDUC);
601 p_cli->vf_total_lines = curr_line - p_blk->start_line;
603 for (i = 1; i < p_mngr->vf_count; i++)
604 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
608 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
609 p_cli->first.val = curr_line;
611 /* first the 'working' task memory */
612 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
613 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
614 if (!p_seg || p_seg->count == 0)
617 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(i)];
618 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
619 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line, total,
620 p_mngr->task_type_size[p_seg->type]);
622 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
626 /* next the 'init' task memory (forced load memory) */
627 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
628 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
629 if (!p_seg || p_seg->count == 0)
632 p_blk = &p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)];
634 if (!p_seg->has_fl_mem) {
635 /* The segment is active (total size pf 'working'
636 * memory is > 0) but has no FL (forced-load, Init)
639 * 1. The total-size in the corrsponding FL block of
640 * the ILT client is set to 0 - No ILT line are
641 * provisioned and no ILT memory allocated.
643 * 2. The start-line of said block is set to the
644 * start line of the matching working memory
645 * block in the ILT client. This is later used to
646 * configure the CDU segment offset registers and
647 * results in an FL command for TIDs of this
648 * segement behaves as regular load commands
649 * (loading TIDs from the working memory).
651 line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line;
653 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
656 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
658 qed_ilt_cli_blk_fill(p_cli, p_blk,
660 p_mngr->task_type_size[p_seg->type]);
662 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
665 p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line;
668 p_seg = qed_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF);
669 if (p_seg && p_seg->count) {
670 /* Stricly speaking we need to iterate over all VF
671 * task segment types, but a VF has only 1 segment
674 /* 'working' memory */
675 total = p_seg->count * p_mngr->task_type_size[p_seg->type];
677 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
678 qed_ilt_cli_blk_fill(p_cli, p_blk,
680 p_mngr->task_type_size[p_seg->type]);
682 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
686 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
687 if (!p_seg->has_fl_mem) {
688 /* see comment above */
689 line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line;
690 qed_ilt_cli_blk_fill(p_cli, p_blk, line, 0, 0);
692 task_size = p_mngr->task_type_size[p_seg->type];
693 qed_ilt_cli_blk_fill(p_cli, p_blk,
694 curr_line, total, task_size);
695 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
698 p_cli->vf_total_lines = curr_line -
699 p_cli->vf_blks[0].start_line;
701 /* Now for the rest of the VFs */
702 for (i = 1; i < p_mngr->vf_count; i++) {
703 p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)];
704 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
707 p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)];
708 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
714 p_cli = &p_mngr->clients[ILT_CLI_QM];
715 p_blk = &p_cli->pf_blks[0];
717 qed_cxt_qm_iids(p_hwfn, &qm_iids);
718 total = qed_qm_pf_mem_size(p_hwfn->rel_pf_id, qm_iids.cids,
719 qm_iids.vf_cids, qm_iids.tids,
720 p_hwfn->qm_info.num_pqs,
721 p_hwfn->qm_info.num_vf_pqs);
725 "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n",
729 p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total);
731 qed_ilt_cli_blk_fill(p_cli, p_blk,
732 curr_line, total * 0x1000,
735 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line, ILT_CLI_QM);
736 p_cli->pf_total_lines = curr_line - p_blk->start_line;
739 p_cli = &p_mngr->clients[ILT_CLI_SRC];
740 qed_cxt_src_iids(p_mngr, &src_iids);
742 /* Both the PF and VFs searcher connections are stored in the per PF
743 * database. Thus sum the PF searcher cids and all the VFs searcher
746 total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
748 u32 local_max = max_t(u32, total,
751 total = roundup_pow_of_two(local_max);
753 p_blk = &p_cli->pf_blks[0];
754 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
755 total * sizeof(struct src_ent),
756 sizeof(struct src_ent));
758 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
760 p_cli->pf_total_lines = curr_line - p_blk->start_line;
764 p_cli = &p_mngr->clients[ILT_CLI_TM];
765 qed_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
766 total = tm_iids.pf_cids + tm_iids.pf_tids_total;
768 p_blk = &p_cli->pf_blks[0];
769 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
770 total * TM_ELEM_SIZE, TM_ELEM_SIZE);
772 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
774 p_cli->pf_total_lines = curr_line - p_blk->start_line;
778 total = tm_iids.per_vf_cids + tm_iids.per_vf_tids;
780 p_blk = &p_cli->vf_blks[0];
781 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
782 total * TM_ELEM_SIZE, TM_ELEM_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;
788 for (i = 1; i < p_mngr->vf_count; i++)
789 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
793 /* TSDM (SRQ CONTEXT) */
794 total = qed_cxt_get_srq_count(p_hwfn);
797 p_cli = &p_mngr->clients[ILT_CLI_TSDM];
798 p_blk = &p_cli->pf_blks[SRQ_BLK];
799 qed_ilt_cli_blk_fill(p_cli, p_blk, curr_line,
800 total * SRQ_CXT_SIZE, SRQ_CXT_SIZE);
802 qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, &curr_line,
804 p_cli->pf_total_lines = curr_line - p_blk->start_line;
807 if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line >
808 RESC_NUM(p_hwfn, QED_ILT)) {
809 DP_ERR(p_hwfn, "too many ilt lines...#lines=%d\n",
810 curr_line - p_hwfn->p_cxt_mngr->pf_start_line);
817 static void qed_cxt_src_t2_free(struct qed_hwfn *p_hwfn)
819 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
825 for (i = 0; i < p_mngr->t2_num_pages; i++)
826 if (p_mngr->t2[i].p_virt)
827 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
829 p_mngr->t2[i].p_virt,
830 p_mngr->t2[i].p_phys);
836 static int qed_cxt_src_t2_alloc(struct qed_hwfn *p_hwfn)
838 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
839 u32 conn_num, total_size, ent_per_page, psz, i;
840 struct qed_ilt_client_cfg *p_src;
841 struct qed_src_iids src_iids;
842 struct qed_dma_mem *p_t2;
845 memset(&src_iids, 0, sizeof(src_iids));
847 /* if the SRC ILT client is inactive - there are no connection
848 * requiring the searcer, leave.
850 p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC];
854 qed_cxt_src_iids(p_mngr, &src_iids);
855 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
856 total_size = conn_num * sizeof(struct src_ent);
858 /* use the same page size as the SRC ILT client */
859 psz = ILT_PAGE_IN_BYTES(p_src->p_size.val);
860 p_mngr->t2_num_pages = DIV_ROUND_UP(total_size, psz);
863 p_mngr->t2 = kcalloc(p_mngr->t2_num_pages, sizeof(struct qed_dma_mem),
870 /* allocate t2 pages */
871 for (i = 0; i < p_mngr->t2_num_pages; i++) {
872 u32 size = min_t(u32, total_size, psz);
873 void **p_virt = &p_mngr->t2[i].p_virt;
875 *p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
877 &p_mngr->t2[i].p_phys, GFP_KERNEL);
878 if (!p_mngr->t2[i].p_virt) {
882 memset(*p_virt, 0, size);
883 p_mngr->t2[i].size = size;
887 /* Set the t2 pointers */
889 /* entries per page - must be a power of two */
890 ent_per_page = psz / sizeof(struct src_ent);
892 p_mngr->first_free = (u64) p_mngr->t2[0].p_phys;
894 p_t2 = &p_mngr->t2[(conn_num - 1) / ent_per_page];
895 p_mngr->last_free = (u64) p_t2->p_phys +
896 ((conn_num - 1) & (ent_per_page - 1)) * sizeof(struct src_ent);
898 for (i = 0; i < p_mngr->t2_num_pages; i++) {
899 u32 ent_num = min_t(u32,
902 struct src_ent *entries = p_mngr->t2[i].p_virt;
903 u64 p_ent_phys = (u64) p_mngr->t2[i].p_phys, val;
906 for (j = 0; j < ent_num - 1; j++) {
907 val = p_ent_phys + (j + 1) * sizeof(struct src_ent);
908 entries[j].next = cpu_to_be64(val);
911 if (i < p_mngr->t2_num_pages - 1)
912 val = (u64) p_mngr->t2[i + 1].p_phys;
915 entries[j].next = cpu_to_be64(val);
923 qed_cxt_src_t2_free(p_hwfn);
927 #define for_each_ilt_valid_client(pos, clients) \
928 for (pos = 0; pos < ILT_CLI_MAX; pos++) \
929 if (!clients[pos].active) { \
933 /* Total number of ILT lines used by this PF */
934 static u32 qed_cxt_ilt_shadow_size(struct qed_ilt_client_cfg *ilt_clients)
939 for_each_ilt_valid_client(i, ilt_clients)
940 size += (ilt_clients[i].last.val - ilt_clients[i].first.val + 1);
945 static void qed_ilt_shadow_free(struct qed_hwfn *p_hwfn)
947 struct qed_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients;
948 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
951 ilt_size = qed_cxt_ilt_shadow_size(p_cli);
953 for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) {
954 struct qed_dma_mem *p_dma = &p_mngr->ilt_shadow[i];
957 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
958 p_dma->size, p_dma->p_virt,
960 p_dma->p_virt = NULL;
962 kfree(p_mngr->ilt_shadow);
965 static int qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn,
966 struct qed_ilt_cli_blk *p_blk,
967 enum ilt_clients ilt_client,
968 u32 start_line_offset)
970 struct qed_dma_mem *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow;
971 u32 lines, line, sz_left, lines_to_skip = 0;
973 /* Special handling for RoCE that supports dynamic allocation */
974 if ((p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE) &&
975 ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM))
978 lines_to_skip = p_blk->dynamic_line_cnt;
980 if (!p_blk->total_size)
983 sz_left = p_blk->total_size;
984 lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) - lines_to_skip;
985 line = p_blk->start_line + start_line_offset -
986 p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip;
988 for (; lines; lines--) {
993 size = min_t(u32, sz_left, p_blk->real_size_in_page);
994 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
995 size, &p_phys, GFP_KERNEL);
998 memset(p_virt, 0, size);
1000 ilt_shadow[line].p_phys = p_phys;
1001 ilt_shadow[line].p_virt = p_virt;
1002 ilt_shadow[line].size = size;
1004 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1005 "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n",
1006 line, (u64)p_phys, p_virt, size);
1015 static int qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn)
1017 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1018 struct qed_ilt_client_cfg *clients = p_mngr->clients;
1019 struct qed_ilt_cli_blk *p_blk;
1023 size = qed_cxt_ilt_shadow_size(clients);
1024 p_mngr->ilt_shadow = kcalloc(size, sizeof(struct qed_dma_mem),
1026 if (!p_mngr->ilt_shadow) {
1028 goto ilt_shadow_fail;
1031 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1032 "Allocated 0x%x bytes for ilt shadow\n",
1033 (u32)(size * sizeof(struct qed_dma_mem)));
1035 for_each_ilt_valid_client(i, clients) {
1036 for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) {
1037 p_blk = &clients[i].pf_blks[j];
1038 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, 0);
1040 goto ilt_shadow_fail;
1042 for (k = 0; k < p_mngr->vf_count; k++) {
1043 for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) {
1044 u32 lines = clients[i].vf_total_lines * k;
1046 p_blk = &clients[i].vf_blks[j];
1047 rc = qed_ilt_blk_alloc(p_hwfn, p_blk, i, lines);
1049 goto ilt_shadow_fail;
1057 qed_ilt_shadow_free(p_hwfn);
1061 static void qed_cid_map_free(struct qed_hwfn *p_hwfn)
1063 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1066 for (type = 0; type < MAX_CONN_TYPES; type++) {
1067 kfree(p_mngr->acquired[type].cid_map);
1068 p_mngr->acquired[type].max_count = 0;
1069 p_mngr->acquired[type].start_cid = 0;
1073 static int qed_cid_map_alloc(struct qed_hwfn *p_hwfn)
1075 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1079 for (type = 0; type < MAX_CONN_TYPES; type++) {
1080 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1086 size = DIV_ROUND_UP(cid_cnt,
1087 sizeof(unsigned long) * BITS_PER_BYTE) *
1088 sizeof(unsigned long);
1089 p_mngr->acquired[type].cid_map = kzalloc(size, GFP_KERNEL);
1090 if (!p_mngr->acquired[type].cid_map)
1093 p_mngr->acquired[type].max_count = cid_cnt;
1094 p_mngr->acquired[type].start_cid = start_cid;
1096 p_hwfn->p_cxt_mngr->conn_cfg[type].cid_start = start_cid;
1098 DP_VERBOSE(p_hwfn, QED_MSG_CXT,
1099 "Type %08x start: %08x count %08x\n",
1100 type, p_mngr->acquired[type].start_cid,
1101 p_mngr->acquired[type].max_count);
1102 start_cid += cid_cnt;
1108 qed_cid_map_free(p_hwfn);
1112 int qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn)
1114 struct qed_ilt_client_cfg *clients;
1115 struct qed_cxt_mngr *p_mngr;
1118 p_mngr = kzalloc(sizeof(*p_mngr), GFP_KERNEL);
1122 /* Initialize ILT client registers */
1123 clients = p_mngr->clients;
1124 clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT);
1125 clients[ILT_CLI_CDUC].last.reg = ILT_CFG_REG(CDUC, LAST_ILT);
1126 clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE);
1128 clients[ILT_CLI_QM].first.reg = ILT_CFG_REG(QM, FIRST_ILT);
1129 clients[ILT_CLI_QM].last.reg = ILT_CFG_REG(QM, LAST_ILT);
1130 clients[ILT_CLI_QM].p_size.reg = ILT_CFG_REG(QM, P_SIZE);
1132 clients[ILT_CLI_TM].first.reg = ILT_CFG_REG(TM, FIRST_ILT);
1133 clients[ILT_CLI_TM].last.reg = ILT_CFG_REG(TM, LAST_ILT);
1134 clients[ILT_CLI_TM].p_size.reg = ILT_CFG_REG(TM, P_SIZE);
1136 clients[ILT_CLI_SRC].first.reg = ILT_CFG_REG(SRC, FIRST_ILT);
1137 clients[ILT_CLI_SRC].last.reg = ILT_CFG_REG(SRC, LAST_ILT);
1138 clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE);
1140 clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT);
1141 clients[ILT_CLI_CDUT].last.reg = ILT_CFG_REG(CDUT, LAST_ILT);
1142 clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE);
1144 clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT);
1145 clients[ILT_CLI_TSDM].last.reg = ILT_CFG_REG(TSDM, LAST_ILT);
1146 clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE);
1147 /* default ILT page size for all clients is 64K */
1148 for (i = 0; i < ILT_CLI_MAX; i++)
1149 p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE;
1151 /* Initialize task sizes */
1152 p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn);
1153 p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn);
1155 if (p_hwfn->cdev->p_iov_info)
1156 p_mngr->vf_count = p_hwfn->cdev->p_iov_info->total_vfs;
1157 /* Initialize the dynamic ILT allocation mutex */
1158 mutex_init(&p_mngr->mutex);
1160 /* Set the cxt mangr pointer priori to further allocations */
1161 p_hwfn->p_cxt_mngr = p_mngr;
1166 int qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn)
1170 /* Allocate the ILT shadow table */
1171 rc = qed_ilt_shadow_alloc(p_hwfn);
1173 goto tables_alloc_fail;
1175 /* Allocate the T2 table */
1176 rc = qed_cxt_src_t2_alloc(p_hwfn);
1178 goto tables_alloc_fail;
1180 /* Allocate and initialize the acquired cids bitmaps */
1181 rc = qed_cid_map_alloc(p_hwfn);
1183 goto tables_alloc_fail;
1188 qed_cxt_mngr_free(p_hwfn);
1192 void qed_cxt_mngr_free(struct qed_hwfn *p_hwfn)
1194 if (!p_hwfn->p_cxt_mngr)
1197 qed_cid_map_free(p_hwfn);
1198 qed_cxt_src_t2_free(p_hwfn);
1199 qed_ilt_shadow_free(p_hwfn);
1200 kfree(p_hwfn->p_cxt_mngr);
1202 p_hwfn->p_cxt_mngr = NULL;
1205 void qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn)
1207 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1210 /* Reset acquired cids */
1211 for (type = 0; type < MAX_CONN_TYPES; type++) {
1212 u32 cid_cnt = p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count;
1217 memset(p_mngr->acquired[type].cid_map, 0,
1218 DIV_ROUND_UP(cid_cnt,
1219 sizeof(unsigned long) * BITS_PER_BYTE) *
1220 sizeof(unsigned long));
1225 #define CDUC_CXT_SIZE_SHIFT \
1226 CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT
1228 #define CDUC_CXT_SIZE_MASK \
1229 (CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT)
1231 #define CDUC_BLOCK_WASTE_SHIFT \
1232 CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT
1234 #define CDUC_BLOCK_WASTE_MASK \
1235 (CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT)
1237 #define CDUC_NCIB_SHIFT \
1238 CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT
1240 #define CDUC_NCIB_MASK \
1241 (CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT)
1243 #define CDUT_TYPE0_CXT_SIZE_SHIFT \
1244 CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT
1246 #define CDUT_TYPE0_CXT_SIZE_MASK \
1247 (CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >> \
1248 CDUT_TYPE0_CXT_SIZE_SHIFT)
1250 #define CDUT_TYPE0_BLOCK_WASTE_SHIFT \
1251 CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT
1253 #define CDUT_TYPE0_BLOCK_WASTE_MASK \
1254 (CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >> \
1255 CDUT_TYPE0_BLOCK_WASTE_SHIFT)
1257 #define CDUT_TYPE0_NCIB_SHIFT \
1258 CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT
1260 #define CDUT_TYPE0_NCIB_MASK \
1261 (CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >> \
1262 CDUT_TYPE0_NCIB_SHIFT)
1264 #define CDUT_TYPE1_CXT_SIZE_SHIFT \
1265 CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT
1267 #define CDUT_TYPE1_CXT_SIZE_MASK \
1268 (CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >> \
1269 CDUT_TYPE1_CXT_SIZE_SHIFT)
1271 #define CDUT_TYPE1_BLOCK_WASTE_SHIFT \
1272 CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT
1274 #define CDUT_TYPE1_BLOCK_WASTE_MASK \
1275 (CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >> \
1276 CDUT_TYPE1_BLOCK_WASTE_SHIFT)
1278 #define CDUT_TYPE1_NCIB_SHIFT \
1279 CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT
1281 #define CDUT_TYPE1_NCIB_MASK \
1282 (CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >> \
1283 CDUT_TYPE1_NCIB_SHIFT)
1285 static void qed_cdu_init_common(struct qed_hwfn *p_hwfn)
1287 u32 page_sz, elems_per_page, block_waste, cxt_size, cdu_params = 0;
1289 /* CDUC - connection configuration */
1290 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1291 cxt_size = CONN_CXT_SIZE(p_hwfn);
1292 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1293 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1295 SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size);
1296 SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste);
1297 SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page);
1298 STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params);
1300 /* CDUT - type-0 tasks configuration */
1301 page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val;
1302 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0];
1303 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1304 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1306 /* cxt size and block-waste are multipes of 8 */
1308 SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3));
1309 SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3));
1310 SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page);
1311 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params);
1313 /* CDUT - type-1 tasks configuration */
1314 cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1];
1315 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size;
1316 block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size;
1318 /* cxt size and block-waste are multipes of 8 */
1320 SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3));
1321 SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3));
1322 SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page);
1323 STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params);
1327 #define CDU_SEG_REG_TYPE_SHIFT CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT
1328 #define CDU_SEG_REG_TYPE_MASK 0x1
1329 #define CDU_SEG_REG_OFFSET_SHIFT 0
1330 #define CDU_SEG_REG_OFFSET_MASK CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK
1332 static void qed_cdu_init_pf(struct qed_hwfn *p_hwfn)
1334 struct qed_ilt_client_cfg *p_cli;
1335 struct qed_tid_seg *p_seg;
1336 u32 cdu_seg_params, offset;
1339 static const u32 rt_type_offset_arr[] = {
1340 CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET,
1341 CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET,
1342 CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET,
1343 CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET
1346 static const u32 rt_type_offset_fl_arr[] = {
1347 CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET,
1348 CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET,
1349 CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET,
1350 CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET
1353 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1355 /* There are initializations only for CDUT during pf Phase */
1356 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1358 p_seg = qed_cxt_tid_seg_info(p_hwfn, i);
1362 /* Note: start_line is already adjusted for the CDU
1363 * segment register granularity, so we just need to
1364 * divide. Adjustment is implicit as we assume ILT
1365 * Page size is larger than 32K!
1367 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1368 (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line -
1369 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1372 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1373 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1374 STORE_RT_REG(p_hwfn, rt_type_offset_arr[i], cdu_seg_params);
1376 offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) *
1377 (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line -
1378 p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES;
1381 SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type);
1382 SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset);
1383 STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i], cdu_seg_params);
1387 void qed_qm_init_pf(struct qed_hwfn *p_hwfn)
1389 struct qed_qm_pf_rt_init_params params;
1390 struct qed_qm_info *qm_info = &p_hwfn->qm_info;
1391 struct qed_qm_iids iids;
1393 memset(&iids, 0, sizeof(iids));
1394 qed_cxt_qm_iids(p_hwfn, &iids);
1396 memset(¶ms, 0, sizeof(params));
1397 params.port_id = p_hwfn->port_id;
1398 params.pf_id = p_hwfn->rel_pf_id;
1399 params.max_phys_tcs_per_port = qm_info->max_phys_tcs_per_port;
1400 params.is_first_pf = p_hwfn->first_on_engine;
1401 params.num_pf_cids = iids.cids;
1402 params.num_vf_cids = iids.vf_cids;
1403 params.start_pq = qm_info->start_pq;
1404 params.num_pf_pqs = qm_info->num_pqs - qm_info->num_vf_pqs;
1405 params.num_vf_pqs = qm_info->num_vf_pqs;
1406 params.start_vport = qm_info->start_vport;
1407 params.num_vports = qm_info->num_vports;
1408 params.pf_wfq = qm_info->pf_wfq;
1409 params.pf_rl = qm_info->pf_rl;
1410 params.pq_params = qm_info->qm_pq_params;
1411 params.vport_params = qm_info->qm_vport_params;
1413 qed_qm_pf_rt_init(p_hwfn, p_hwfn->p_main_ptt, ¶ms);
1417 void qed_cm_init_pf(struct qed_hwfn *p_hwfn)
1419 /* XCM pure-LB queue */
1420 STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET,
1421 qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB));
1425 static void qed_dq_init_pf(struct qed_hwfn *p_hwfn)
1427 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1428 u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0;
1430 dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT);
1431 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid);
1433 dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT);
1434 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid);
1436 dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT);
1437 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid);
1439 dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT);
1440 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid);
1442 dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT);
1443 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid);
1445 dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT);
1446 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid);
1448 dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT);
1449 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid);
1451 dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT);
1452 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid);
1454 dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT);
1455 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid);
1457 dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT);
1458 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid);
1460 dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT);
1461 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid);
1463 dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT);
1464 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid);
1466 /* Connection types 6 & 7 are not in use, yet they must be configured
1467 * as the highest possible connection. Not configuring them means the
1468 * defaults will be used, and with a large number of cids a bug may
1469 * occur, if the defaults will be smaller than dq_pf_max_cid /
1472 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid);
1473 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid);
1475 STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid);
1476 STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid);
1479 static void qed_ilt_bounds_init(struct qed_hwfn *p_hwfn)
1481 struct qed_ilt_client_cfg *ilt_clients;
1484 ilt_clients = p_hwfn->p_cxt_mngr->clients;
1485 for_each_ilt_valid_client(i, ilt_clients) {
1486 STORE_RT_REG(p_hwfn,
1487 ilt_clients[i].first.reg,
1488 ilt_clients[i].first.val);
1489 STORE_RT_REG(p_hwfn,
1490 ilt_clients[i].last.reg, ilt_clients[i].last.val);
1491 STORE_RT_REG(p_hwfn,
1492 ilt_clients[i].p_size.reg,
1493 ilt_clients[i].p_size.val);
1497 static void qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn)
1499 struct qed_ilt_client_cfg *p_cli;
1502 /* For simplicty we set the 'block' to be an ILT page */
1503 if (p_hwfn->cdev->p_iov_info) {
1504 struct qed_hw_sriov_info *p_iov = p_hwfn->cdev->p_iov_info;
1506 STORE_RT_REG(p_hwfn,
1507 PSWRQ2_REG_VF_BASE_RT_OFFSET,
1508 p_iov->first_vf_in_pf);
1509 STORE_RT_REG(p_hwfn,
1510 PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET,
1511 p_iov->first_vf_in_pf + p_iov->total_vfs);
1514 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
1515 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1516 if (p_cli->active) {
1517 STORE_RT_REG(p_hwfn,
1518 PSWRQ2_REG_CDUC_BLOCKS_FACTOR_RT_OFFSET,
1520 STORE_RT_REG(p_hwfn,
1521 PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1522 p_cli->pf_total_lines);
1523 STORE_RT_REG(p_hwfn,
1524 PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET,
1525 p_cli->vf_total_lines);
1528 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
1529 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1530 if (p_cli->active) {
1531 STORE_RT_REG(p_hwfn,
1532 PSWRQ2_REG_CDUT_BLOCKS_FACTOR_RT_OFFSET,
1534 STORE_RT_REG(p_hwfn,
1535 PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1536 p_cli->pf_total_lines);
1537 STORE_RT_REG(p_hwfn,
1538 PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET,
1539 p_cli->vf_total_lines);
1542 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM];
1543 blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10);
1544 if (p_cli->active) {
1545 STORE_RT_REG(p_hwfn,
1546 PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET, blk_factor);
1547 STORE_RT_REG(p_hwfn,
1548 PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET,
1549 p_cli->pf_total_lines);
1550 STORE_RT_REG(p_hwfn,
1551 PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET,
1552 p_cli->vf_total_lines);
1556 /* ILT (PSWRQ2) PF */
1557 static void qed_ilt_init_pf(struct qed_hwfn *p_hwfn)
1559 struct qed_ilt_client_cfg *clients;
1560 struct qed_cxt_mngr *p_mngr;
1561 struct qed_dma_mem *p_shdw;
1562 u32 line, rt_offst, i;
1564 qed_ilt_bounds_init(p_hwfn);
1565 qed_ilt_vf_bounds_init(p_hwfn);
1567 p_mngr = p_hwfn->p_cxt_mngr;
1568 p_shdw = p_mngr->ilt_shadow;
1569 clients = p_hwfn->p_cxt_mngr->clients;
1571 for_each_ilt_valid_client(i, clients) {
1572 /** Client's 1st val and RT array are absolute, ILT shadows'
1573 * lines are relative.
1575 line = clients[i].first.val - p_mngr->pf_start_line;
1576 rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET +
1577 clients[i].first.val * ILT_ENTRY_IN_REGS;
1579 for (; line <= clients[i].last.val - p_mngr->pf_start_line;
1580 line++, rt_offst += ILT_ENTRY_IN_REGS) {
1581 u64 ilt_hw_entry = 0;
1583 /** p_virt could be NULL incase of dynamic
1586 if (p_shdw[line].p_virt) {
1587 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
1588 SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR,
1589 (p_shdw[line].p_phys >> 12));
1591 DP_VERBOSE(p_hwfn, QED_MSG_ILT,
1592 "Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n",
1594 (u64)(p_shdw[line].p_phys >> 12));
1597 STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry);
1602 /* SRC (Searcher) PF */
1603 static void qed_src_init_pf(struct qed_hwfn *p_hwfn)
1605 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1606 u32 rounded_conn_num, conn_num, conn_max;
1607 struct qed_src_iids src_iids;
1609 memset(&src_iids, 0, sizeof(src_iids));
1610 qed_cxt_src_iids(p_mngr, &src_iids);
1611 conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count;
1615 conn_max = max_t(u32, conn_num, SRC_MIN_NUM_ELEMS);
1616 rounded_conn_num = roundup_pow_of_two(conn_max);
1618 STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num);
1619 STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET,
1620 ilog2(rounded_conn_num));
1622 STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET,
1623 p_hwfn->p_cxt_mngr->first_free);
1624 STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET,
1625 p_hwfn->p_cxt_mngr->last_free);
1629 #define TM_CFG_NUM_IDS_SHIFT 0
1630 #define TM_CFG_NUM_IDS_MASK 0xFFFFULL
1631 #define TM_CFG_PRE_SCAN_OFFSET_SHIFT 16
1632 #define TM_CFG_PRE_SCAN_OFFSET_MASK 0x1FFULL
1633 #define TM_CFG_PARENT_PF_SHIFT 25
1634 #define TM_CFG_PARENT_PF_MASK 0x7ULL
1636 #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT 30
1637 #define TM_CFG_CID_PRE_SCAN_ROWS_MASK 0x1FFULL
1639 #define TM_CFG_TID_OFFSET_SHIFT 30
1640 #define TM_CFG_TID_OFFSET_MASK 0x7FFFFULL
1641 #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT 49
1642 #define TM_CFG_TID_PRE_SCAN_ROWS_MASK 0x1FFULL
1644 static void qed_tm_init_pf(struct qed_hwfn *p_hwfn)
1646 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1647 u32 active_seg_mask = 0, tm_offset, rt_reg;
1648 struct qed_tm_iids tm_iids;
1652 memset(&tm_iids, 0, sizeof(tm_iids));
1653 qed_cxt_tm_iids(p_hwfn, p_mngr, &tm_iids);
1655 /* @@@TBD No pre-scan for now */
1657 /* Note: We assume consecutive VFs for a PF */
1658 for (i = 0; i < p_mngr->vf_count; i++) {
1660 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids);
1661 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1662 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1663 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0);
1664 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1665 (sizeof(cfg_word) / sizeof(u32)) *
1666 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1667 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1671 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids);
1672 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1673 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0); /* n/a for PF */
1674 SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */
1676 rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET +
1677 (sizeof(cfg_word) / sizeof(u32)) *
1678 (NUM_OF_VFS(p_hwfn->cdev) + p_hwfn->rel_pf_id);
1679 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1682 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET,
1683 tm_iids.pf_cids ? 0x1 : 0x0);
1685 /* @@@TBD how to enable the scan for the VFs */
1687 tm_offset = tm_iids.per_vf_cids;
1689 /* Note: We assume consecutive VFs for a PF */
1690 for (i = 0; i < p_mngr->vf_count; i++) {
1692 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids);
1693 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1694 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id);
1695 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1696 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1698 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1699 (sizeof(cfg_word) / sizeof(u32)) *
1700 (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i);
1702 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1705 tm_offset = tm_iids.pf_cids;
1706 for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) {
1708 SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]);
1709 SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0);
1710 SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0);
1711 SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset);
1712 SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0);
1714 rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET +
1715 (sizeof(cfg_word) / sizeof(u32)) *
1716 (NUM_OF_VFS(p_hwfn->cdev) +
1717 p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i);
1719 STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word);
1720 active_seg_mask |= (tm_iids.pf_tids[i] ? BIT(i) : 0);
1722 tm_offset += tm_iids.pf_tids[i];
1725 if (p_hwfn->hw_info.personality == QED_PCI_ETH_ROCE)
1726 active_seg_mask = 0;
1728 STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask);
1730 /* @@@TBD how to enable the scan for the VFs */
1733 static void qed_prs_init_common(struct qed_hwfn *p_hwfn)
1735 if ((p_hwfn->hw_info.personality == QED_PCI_FCOE) &&
1736 p_hwfn->pf_params.fcoe_pf_params.is_target)
1737 STORE_RT_REG(p_hwfn,
1738 PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0);
1741 static void qed_prs_init_pf(struct qed_hwfn *p_hwfn)
1743 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1744 struct qed_conn_type_cfg *p_fcoe;
1745 struct qed_tid_seg *p_tid;
1747 p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE];
1749 /* If FCoE is active set the MAX OX_ID (tid) in the Parser */
1750 if (!p_fcoe->cid_count)
1753 p_tid = &p_fcoe->tid_seg[QED_CXT_FCOE_TID_SEG];
1754 if (p_hwfn->pf_params.fcoe_pf_params.is_target) {
1755 STORE_RT_REG_AGG(p_hwfn,
1756 PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET,
1759 STORE_RT_REG_AGG(p_hwfn,
1760 PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET,
1765 void qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn)
1767 qed_cdu_init_common(p_hwfn);
1768 qed_prs_init_common(p_hwfn);
1771 void qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn)
1773 qed_qm_init_pf(p_hwfn);
1774 qed_cm_init_pf(p_hwfn);
1775 qed_dq_init_pf(p_hwfn);
1776 qed_cdu_init_pf(p_hwfn);
1777 qed_ilt_init_pf(p_hwfn);
1778 qed_src_init_pf(p_hwfn);
1779 qed_tm_init_pf(p_hwfn);
1780 qed_prs_init_pf(p_hwfn);
1783 int qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn,
1784 enum protocol_type type, u32 *p_cid)
1786 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1789 if (type >= MAX_CONN_TYPES || !p_mngr->acquired[type].cid_map) {
1790 DP_NOTICE(p_hwfn, "Invalid protocol type %d", type);
1794 rel_cid = find_first_zero_bit(p_mngr->acquired[type].cid_map,
1795 p_mngr->acquired[type].max_count);
1797 if (rel_cid >= p_mngr->acquired[type].max_count) {
1798 DP_NOTICE(p_hwfn, "no CID available for protocol %d\n", type);
1802 __set_bit(rel_cid, p_mngr->acquired[type].cid_map);
1804 *p_cid = rel_cid + p_mngr->acquired[type].start_cid;
1809 static bool qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn,
1810 u32 cid, enum protocol_type *p_type)
1812 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1813 struct qed_cid_acquired_map *p_map;
1814 enum protocol_type p;
1817 /* Iterate over protocols and find matching cid range */
1818 for (p = 0; p < MAX_CONN_TYPES; p++) {
1819 p_map = &p_mngr->acquired[p];
1821 if (!p_map->cid_map)
1823 if (cid >= p_map->start_cid &&
1824 cid < p_map->start_cid + p_map->max_count)
1829 if (p == MAX_CONN_TYPES) {
1830 DP_NOTICE(p_hwfn, "Invalid CID %d", cid);
1834 rel_cid = cid - p_map->start_cid;
1835 if (!test_bit(rel_cid, p_map->cid_map)) {
1836 DP_NOTICE(p_hwfn, "CID %d not acquired", cid);
1842 void qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid)
1844 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1845 enum protocol_type type;
1849 /* Test acquired and find matching per-protocol map */
1850 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid, &type);
1855 rel_cid = cid - p_mngr->acquired[type].start_cid;
1856 __clear_bit(rel_cid, p_mngr->acquired[type].cid_map);
1859 int qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn, struct qed_cxt_info *p_info)
1861 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
1862 u32 conn_cxt_size, hw_p_size, cxts_per_p, line;
1863 enum protocol_type type;
1866 /* Test acquired and find matching per-protocol map */
1867 b_acquired = qed_cxt_test_cid_acquired(p_hwfn, p_info->iid, &type);
1872 /* set the protocl type */
1873 p_info->type = type;
1875 /* compute context virtual pointer */
1876 hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val;
1878 conn_cxt_size = CONN_CXT_SIZE(p_hwfn);
1879 cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size;
1880 line = p_info->iid / cxts_per_p;
1882 /* Make sure context is allocated (dynamic allocation) */
1883 if (!p_mngr->ilt_shadow[line].p_virt)
1886 p_info->p_cxt = p_mngr->ilt_shadow[line].p_virt +
1887 p_info->iid % cxts_per_p * conn_cxt_size;
1889 DP_VERBOSE(p_hwfn, (QED_MSG_ILT | QED_MSG_CXT),
1890 "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n",
1891 p_info->iid / cxts_per_p, p_info->p_cxt, p_info->iid);
1896 static void qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn,
1897 struct qed_rdma_pf_params *p_params)
1899 u32 num_cons, num_tasks, num_qps, num_mrs, num_srqs;
1900 enum protocol_type proto;
1902 num_mrs = min_t(u32, RDMA_MAX_TIDS, p_params->num_mrs);
1903 num_tasks = num_mrs; /* each mr uses a single task id */
1904 num_srqs = min_t(u32, 32 * 1024, p_params->num_srqs);
1906 switch (p_hwfn->hw_info.personality) {
1907 case QED_PCI_ETH_ROCE:
1908 num_qps = min_t(u32, ROCE_MAX_QPS, p_params->num_qps);
1909 num_cons = num_qps * 2; /* each QP requires two connections */
1910 proto = PROTOCOLID_ROCE;
1916 if (num_cons && num_tasks) {
1917 qed_cxt_set_proto_cid_count(p_hwfn, proto, num_cons, 0);
1919 /* Deliberatly passing ROCE for tasks id. This is because
1920 * iWARP / RoCE share the task id.
1922 qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_ROCE,
1923 QED_CXT_ROCE_TID_SEG, 1,
1925 qed_cxt_set_srq_count(p_hwfn, num_srqs);
1927 DP_INFO(p_hwfn->cdev,
1928 "RDMA personality used without setting params!\n");
1932 int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn)
1934 /* Set the number of required CORE connections */
1935 u32 core_cids = 1; /* SPQ */
1937 if (p_hwfn->using_ll2)
1939 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_CORE, core_cids, 0);
1941 switch (p_hwfn->hw_info.personality) {
1942 case QED_PCI_ETH_ROCE:
1944 qed_rdma_set_pf_params(p_hwfn,
1946 pf_params.rdma_pf_params);
1947 /* no need for break since RoCE coexist with Ethernet */
1951 struct qed_eth_pf_params *p_params =
1952 &p_hwfn->pf_params.eth_pf_params;
1954 qed_cxt_set_proto_cid_count(p_hwfn, PROTOCOLID_ETH,
1955 p_params->num_cons, 1);
1960 struct qed_fcoe_pf_params *p_params;
1962 p_params = &p_hwfn->pf_params.fcoe_pf_params;
1964 if (p_params->num_cons && p_params->num_tasks) {
1965 qed_cxt_set_proto_cid_count(p_hwfn,
1970 qed_cxt_set_proto_tid_count(p_hwfn, PROTOCOLID_FCOE,
1971 QED_CXT_FCOE_TID_SEG, 0,
1972 p_params->num_tasks, true);
1974 DP_INFO(p_hwfn->cdev,
1975 "Fcoe personality used without setting params!\n");
1981 struct qed_iscsi_pf_params *p_params;
1983 p_params = &p_hwfn->pf_params.iscsi_pf_params;
1985 if (p_params->num_cons && p_params->num_tasks) {
1986 qed_cxt_set_proto_cid_count(p_hwfn,
1991 qed_cxt_set_proto_tid_count(p_hwfn,
1993 QED_CXT_ISCSI_TID_SEG,
1995 p_params->num_tasks,
1998 DP_INFO(p_hwfn->cdev,
1999 "Iscsi personality used without setting params!\n");
2010 int qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn,
2011 struct qed_tid_mem *p_info)
2013 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2014 u32 proto, seg, total_lines, i, shadow_line;
2015 struct qed_ilt_client_cfg *p_cli;
2016 struct qed_ilt_cli_blk *p_fl_seg;
2017 struct qed_tid_seg *p_seg_info;
2019 /* Verify the personality */
2020 switch (p_hwfn->hw_info.personality) {
2022 proto = PROTOCOLID_FCOE;
2023 seg = QED_CXT_FCOE_TID_SEG;
2026 proto = PROTOCOLID_ISCSI;
2027 seg = QED_CXT_ISCSI_TID_SEG;
2033 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2037 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2038 if (!p_seg_info->has_fl_mem)
2041 p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2042 total_lines = DIV_ROUND_UP(p_fl_seg->total_size,
2043 p_fl_seg->real_size_in_page);
2045 for (i = 0; i < total_lines; i++) {
2046 shadow_line = i + p_fl_seg->start_line -
2047 p_hwfn->p_cxt_mngr->pf_start_line;
2048 p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].p_virt;
2050 p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) -
2051 p_fl_seg->real_size_in_page;
2052 p_info->tid_size = p_mngr->task_type_size[p_seg_info->type];
2053 p_info->num_tids_per_block = p_fl_seg->real_size_in_page /
2059 /* This function is very RoCE oriented, if another protocol in the future
2060 * will want this feature we'll need to modify the function to be more generic
2063 qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn,
2064 enum qed_cxt_elem_type elem_type, u32 iid)
2066 u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line;
2067 struct qed_ilt_client_cfg *p_cli;
2068 struct qed_ilt_cli_blk *p_blk;
2069 struct qed_ptt *p_ptt;
2075 switch (elem_type) {
2077 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2078 elem_size = CONN_CXT_SIZE(p_hwfn);
2079 p_blk = &p_cli->pf_blks[CDUC_BLK];
2082 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2083 elem_size = SRQ_CXT_SIZE;
2084 p_blk = &p_cli->pf_blks[SRQ_BLK];
2087 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2088 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2089 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2092 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2096 /* Calculate line in ilt */
2097 hw_p_size = p_cli->p_size.val;
2098 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2099 line = p_blk->start_line + (iid / elems_per_p);
2100 shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line;
2102 /* If line is already allocated, do nothing, otherwise allocate it and
2103 * write it to the PSWRQ2 registers.
2104 * This section can be run in parallel from different contexts and thus
2105 * a mutex protection is needed.
2108 mutex_lock(&p_hwfn->p_cxt_mngr->mutex);
2110 if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt)
2113 p_ptt = qed_ptt_acquire(p_hwfn);
2116 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2121 p_virt = dma_alloc_coherent(&p_hwfn->cdev->pdev->dev,
2122 p_blk->real_size_in_page,
2123 &p_phys, GFP_KERNEL);
2128 memset(p_virt, 0, p_blk->real_size_in_page);
2130 /* configuration of refTagMask to 0xF is required for RoCE DIF MR only,
2131 * to compensate for a HW bug, but it is configured even if DIF is not
2132 * enabled. This is harmless and allows us to avoid a dedicated API. We
2133 * configure the field for all of the contexts on the newly allocated
2136 if (elem_type == QED_ELEM_TASK) {
2138 u8 *elem_start = (u8 *)p_virt;
2139 union type1_task_context *elem;
2141 for (elem_i = 0; elem_i < elems_per_p; elem_i++) {
2142 elem = (union type1_task_context *)elem_start;
2143 SET_FIELD(elem->roce_ctx.tdif_context.flags1,
2144 TDIF_TASK_CONTEXT_REFTAGMASK, 0xf);
2145 elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn);
2149 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_virt = p_virt;
2150 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys = p_phys;
2151 p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size =
2152 p_blk->real_size_in_page;
2154 /* compute absolute offset */
2155 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2156 (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS);
2159 SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL);
2160 SET_FIELD(ilt_hw_entry,
2162 (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].p_phys >> 12));
2164 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */
2165 qed_dmae_host2grc(p_hwfn, p_ptt, (u64) (uintptr_t)&ilt_hw_entry,
2166 reg_offset, sizeof(ilt_hw_entry) / sizeof(u32), 0);
2168 if (elem_type == QED_ELEM_CXT) {
2169 u32 last_cid_allocated = (1 + (iid / elems_per_p)) *
2172 /* Update the relevant register in the parser */
2173 qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF,
2174 last_cid_allocated - 1);
2176 if (!p_hwfn->b_rdma_enabled_in_prs) {
2177 /* Enable RoCE search */
2178 qed_wr(p_hwfn, p_ptt, p_hwfn->rdma_prs_search_reg, 1);
2179 p_hwfn->b_rdma_enabled_in_prs = true;
2184 qed_ptt_release(p_hwfn, p_ptt);
2186 mutex_unlock(&p_hwfn->p_cxt_mngr->mutex);
2191 /* This function is very RoCE oriented, if another protocol in the future
2192 * will want this feature we'll need to modify the function to be more generic
2195 qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn,
2196 enum qed_cxt_elem_type elem_type,
2197 u32 start_iid, u32 count)
2199 u32 start_line, end_line, shadow_start_line, shadow_end_line;
2200 u32 reg_offset, elem_size, hw_p_size, elems_per_p;
2201 struct qed_ilt_client_cfg *p_cli;
2202 struct qed_ilt_cli_blk *p_blk;
2203 u32 end_iid = start_iid + count;
2204 struct qed_ptt *p_ptt;
2205 u64 ilt_hw_entry = 0;
2208 switch (elem_type) {
2210 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC];
2211 elem_size = CONN_CXT_SIZE(p_hwfn);
2212 p_blk = &p_cli->pf_blks[CDUC_BLK];
2215 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM];
2216 elem_size = SRQ_CXT_SIZE;
2217 p_blk = &p_cli->pf_blks[SRQ_BLK];
2220 p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT];
2221 elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn);
2222 p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)];
2225 DP_NOTICE(p_hwfn, "-EINVALID elem type = %d", elem_type);
2229 /* Calculate line in ilt */
2230 hw_p_size = p_cli->p_size.val;
2231 elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size;
2232 start_line = p_blk->start_line + (start_iid / elems_per_p);
2233 end_line = p_blk->start_line + (end_iid / elems_per_p);
2234 if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p))
2237 shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line;
2238 shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line;
2240 p_ptt = qed_ptt_acquire(p_hwfn);
2243 "QED_TIME_OUT on ptt acquire - dynamic allocation");
2247 for (i = shadow_start_line; i < shadow_end_line; i++) {
2248 if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt)
2251 dma_free_coherent(&p_hwfn->cdev->pdev->dev,
2252 p_hwfn->p_cxt_mngr->ilt_shadow[i].size,
2253 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt,
2254 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys);
2256 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_virt = NULL;
2257 p_hwfn->p_cxt_mngr->ilt_shadow[i].p_phys = 0;
2258 p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0;
2260 /* compute absolute offset */
2261 reg_offset = PSWRQ2_REG_ILT_MEMORY +
2262 ((start_line++) * ILT_REG_SIZE_IN_BYTES *
2265 /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a
2268 qed_dmae_host2grc(p_hwfn, p_ptt,
2269 (u64) (uintptr_t) &ilt_hw_entry,
2271 sizeof(ilt_hw_entry) / sizeof(u32),
2275 qed_ptt_release(p_hwfn, p_ptt);
2280 int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto)
2285 /* Free Connection CXT */
2286 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_CXT,
2287 qed_cxt_get_proto_cid_start(p_hwfn,
2289 qed_cxt_get_proto_cid_count(p_hwfn,
2296 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_TASK, 0,
2297 qed_cxt_get_proto_tid_count(p_hwfn, proto));
2302 rc = qed_cxt_free_ilt_range(p_hwfn, QED_ELEM_SRQ, 0,
2303 qed_cxt_get_srq_count(p_hwfn));
2308 int qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn,
2309 u32 tid, u8 ctx_type, void **pp_task_ctx)
2311 struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr;
2312 struct qed_ilt_client_cfg *p_cli;
2313 struct qed_tid_seg *p_seg_info;
2314 struct qed_ilt_cli_blk *p_seg;
2315 u32 num_tids_per_block;
2316 u32 tid_size, ilt_idx;
2320 /* Verify the personality */
2321 switch (p_hwfn->hw_info.personality) {
2323 proto = PROTOCOLID_FCOE;
2324 seg = QED_CXT_FCOE_TID_SEG;
2327 proto = PROTOCOLID_ISCSI;
2328 seg = QED_CXT_ISCSI_TID_SEG;
2334 p_cli = &p_mngr->clients[ILT_CLI_CDUT];
2338 p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg];
2340 if (ctx_type == QED_CTX_WORKING_MEM) {
2341 p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)];
2342 } else if (ctx_type == QED_CTX_FL_MEM) {
2343 if (!p_seg_info->has_fl_mem)
2345 p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)];
2349 total_lines = DIV_ROUND_UP(p_seg->total_size, p_seg->real_size_in_page);
2350 tid_size = p_mngr->task_type_size[p_seg_info->type];
2351 num_tids_per_block = p_seg->real_size_in_page / tid_size;
2353 if (total_lines < tid / num_tids_per_block)
2356 ilt_idx = tid / num_tids_per_block + p_seg->start_line -
2357 p_mngr->pf_start_line;
2358 *pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].p_virt +
2359 (tid % num_tids_per_block) * tid_size;