1 | // SPDX-License-Identifier: (GPL-2.0-only OR BSD-3-Clause) |
2 | /* QLogic qed NIC Driver |
3 | * Copyright (c) 2015-2017 QLogic Corporation |
4 | * Copyright (c) 2019-2020 Marvell International Ltd. |
5 | */ |
6 | |
7 | #include <linux/types.h> |
8 | #include <linux/bitops.h> |
9 | #include <linux/dma-mapping.h> |
10 | #include <linux/errno.h> |
11 | #include <linux/kernel.h> |
12 | #include <linux/list.h> |
13 | #include <linux/log2.h> |
14 | #include <linux/pci.h> |
15 | #include <linux/slab.h> |
16 | #include <linux/string.h> |
17 | #include "qed.h" |
18 | #include "qed_cxt.h" |
19 | #include "qed_dev_api.h" |
20 | #include "qed_hsi.h" |
21 | #include "qed_hw.h" |
22 | #include "qed_init_ops.h" |
23 | #include "qed_rdma.h" |
24 | #include "qed_reg_addr.h" |
25 | #include "qed_sriov.h" |
26 | |
27 | /* QM constants */ |
28 | #define QM_PQ_ELEMENT_SIZE 4 /* in bytes */ |
29 | |
30 | /* Doorbell-Queue constants */ |
31 | #define DQ_RANGE_SHIFT 4 |
32 | #define DQ_RANGE_ALIGN BIT(DQ_RANGE_SHIFT) |
33 | |
34 | /* Searcher constants */ |
35 | #define SRC_MIN_NUM_ELEMS 256 |
36 | |
37 | /* Timers constants */ |
38 | #define TM_SHIFT 7 |
39 | #define TM_ALIGN BIT(TM_SHIFT) |
40 | #define TM_ELEM_SIZE 4 |
41 | |
42 | #define ILT_DEFAULT_HW_P_SIZE 4 |
43 | |
44 | #define ILT_PAGE_IN_BYTES(hw_p_size) (1U << ((hw_p_size) + 12)) |
45 | #define ILT_CFG_REG(cli, reg) PSWRQ2_REG_ ## cli ## _ ## reg ## _RT_OFFSET |
46 | |
47 | /* ILT entry structure */ |
48 | #define ILT_ENTRY_PHY_ADDR_MASK (~0ULL >> 12) |
49 | #define ILT_ENTRY_PHY_ADDR_SHIFT 0 |
50 | #define ILT_ENTRY_VALID_MASK 0x1ULL |
51 | #define ILT_ENTRY_VALID_SHIFT 52 |
52 | #define ILT_ENTRY_IN_REGS 2 |
53 | #define ILT_REG_SIZE_IN_BYTES 4 |
54 | |
55 | /* connection context union */ |
56 | union conn_context { |
57 | struct core_conn_context core_ctx; |
58 | struct eth_conn_context eth_ctx; |
59 | struct iscsi_conn_context iscsi_ctx; |
60 | struct fcoe_conn_context fcoe_ctx; |
61 | struct roce_conn_context roce_ctx; |
62 | }; |
63 | |
64 | /* TYPE-0 task context - iSCSI, FCOE */ |
65 | union type0_task_context { |
66 | struct iscsi_task_context iscsi_ctx; |
67 | struct fcoe_task_context fcoe_ctx; |
68 | }; |
69 | |
70 | /* TYPE-1 task context - ROCE */ |
71 | union type1_task_context { |
72 | struct rdma_task_context roce_ctx; |
73 | }; |
74 | |
75 | struct src_ent { |
76 | __u8 opaque[56]; |
77 | __be64 next; |
78 | }; |
79 | |
80 | #define CDUT_SEG_ALIGNMET 3 /* in 4k chunks */ |
81 | #define CDUT_SEG_ALIGNMET_IN_BYTES BIT(CDUT_SEG_ALIGNMET + 12) |
82 | |
83 | #define CONN_CXT_SIZE(p_hwfn) \ |
84 | ALIGNED_TYPE_SIZE(union conn_context, p_hwfn) |
85 | |
86 | #define SRQ_CXT_SIZE (sizeof(struct rdma_srq_context)) |
87 | #define XRC_SRQ_CXT_SIZE (sizeof(struct rdma_xrc_srq_context)) |
88 | |
89 | #define TYPE0_TASK_CXT_SIZE(p_hwfn) \ |
90 | ALIGNED_TYPE_SIZE(union type0_task_context, p_hwfn) |
91 | |
92 | /* Alignment is inherent to the type1_task_context structure */ |
93 | #define TYPE1_TASK_CXT_SIZE(p_hwfn) sizeof(union type1_task_context) |
94 | |
95 | static bool src_proto(enum protocol_type type) |
96 | { |
97 | return type == PROTOCOLID_TCP_ULP || |
98 | type == PROTOCOLID_FCOE || |
99 | type == PROTOCOLID_IWARP; |
100 | } |
101 | |
102 | static bool tm_cid_proto(enum protocol_type type) |
103 | { |
104 | return type == PROTOCOLID_TCP_ULP || |
105 | type == PROTOCOLID_FCOE || |
106 | type == PROTOCOLID_ROCE || |
107 | type == PROTOCOLID_IWARP; |
108 | } |
109 | |
110 | static bool tm_tid_proto(enum protocol_type type) |
111 | { |
112 | return type == PROTOCOLID_FCOE; |
113 | } |
114 | |
115 | /* counts the iids for the CDU/CDUC ILT client configuration */ |
116 | struct qed_cdu_iids { |
117 | u32 pf_cids; |
118 | u32 per_vf_cids; |
119 | }; |
120 | |
121 | static void qed_cxt_cdu_iids(struct qed_cxt_mngr *p_mngr, |
122 | struct qed_cdu_iids *iids) |
123 | { |
124 | u32 type; |
125 | |
126 | for (type = 0; type < MAX_CONN_TYPES; type++) { |
127 | iids->pf_cids += p_mngr->conn_cfg[type].cid_count; |
128 | iids->per_vf_cids += p_mngr->conn_cfg[type].cids_per_vf; |
129 | } |
130 | } |
131 | |
132 | /* counts the iids for the Searcher block configuration */ |
133 | struct qed_src_iids { |
134 | u32 pf_cids; |
135 | u32 per_vf_cids; |
136 | }; |
137 | |
138 | static void qed_cxt_src_iids(struct qed_cxt_mngr *p_mngr, |
139 | struct qed_src_iids *iids) |
140 | { |
141 | u32 i; |
142 | |
143 | for (i = 0; i < MAX_CONN_TYPES; i++) { |
144 | if (!src_proto(type: i)) |
145 | continue; |
146 | |
147 | iids->pf_cids += p_mngr->conn_cfg[i].cid_count; |
148 | iids->per_vf_cids += p_mngr->conn_cfg[i].cids_per_vf; |
149 | } |
150 | |
151 | /* Add L2 filtering filters in addition */ |
152 | iids->pf_cids += p_mngr->arfs_count; |
153 | } |
154 | |
155 | /* counts the iids for the Timers block configuration */ |
156 | struct qed_tm_iids { |
157 | u32 pf_cids; |
158 | u32 pf_tids[NUM_TASK_PF_SEGMENTS]; /* per segment */ |
159 | u32 pf_tids_total; |
160 | u32 per_vf_cids; |
161 | u32 per_vf_tids; |
162 | }; |
163 | |
164 | static void qed_cxt_tm_iids(struct qed_hwfn *p_hwfn, |
165 | struct qed_cxt_mngr *p_mngr, |
166 | struct qed_tm_iids *iids) |
167 | { |
168 | bool tm_vf_required = false; |
169 | bool tm_required = false; |
170 | int i, j; |
171 | |
172 | /* Timers is a special case -> we don't count how many cids require |
173 | * timers but what's the max cid that will be used by the timer block. |
174 | * therefore we traverse in reverse order, and once we hit a protocol |
175 | * that requires the timers memory, we'll sum all the protocols up |
176 | * to that one. |
177 | */ |
178 | for (i = MAX_CONN_TYPES - 1; i >= 0; i--) { |
179 | struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[i]; |
180 | |
181 | if (tm_cid_proto(type: i) || tm_required) { |
182 | if (p_cfg->cid_count) |
183 | tm_required = true; |
184 | |
185 | iids->pf_cids += p_cfg->cid_count; |
186 | } |
187 | |
188 | if (tm_cid_proto(type: i) || tm_vf_required) { |
189 | if (p_cfg->cids_per_vf) |
190 | tm_vf_required = true; |
191 | |
192 | iids->per_vf_cids += p_cfg->cids_per_vf; |
193 | } |
194 | |
195 | if (tm_tid_proto(type: i)) { |
196 | struct qed_tid_seg *segs = p_cfg->tid_seg; |
197 | |
198 | /* for each segment there is at most one |
199 | * protocol for which count is not 0. |
200 | */ |
201 | for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++) |
202 | iids->pf_tids[j] += segs[j].count; |
203 | |
204 | /* The last array elelment is for the VFs. As for PF |
205 | * segments there can be only one protocol for |
206 | * which this value is not 0. |
207 | */ |
208 | iids->per_vf_tids += segs[NUM_TASK_PF_SEGMENTS].count; |
209 | } |
210 | } |
211 | |
212 | iids->pf_cids = roundup(iids->pf_cids, TM_ALIGN); |
213 | iids->per_vf_cids = roundup(iids->per_vf_cids, TM_ALIGN); |
214 | iids->per_vf_tids = roundup(iids->per_vf_tids, TM_ALIGN); |
215 | |
216 | for (iids->pf_tids_total = 0, j = 0; j < NUM_TASK_PF_SEGMENTS; j++) { |
217 | iids->pf_tids[j] = roundup(iids->pf_tids[j], TM_ALIGN); |
218 | iids->pf_tids_total += iids->pf_tids[j]; |
219 | } |
220 | } |
221 | |
222 | static void qed_cxt_qm_iids(struct qed_hwfn *p_hwfn, |
223 | struct qed_qm_iids *iids) |
224 | { |
225 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
226 | struct qed_tid_seg *segs; |
227 | u32 vf_cids = 0, type, j; |
228 | u32 vf_tids = 0; |
229 | |
230 | for (type = 0; type < MAX_CONN_TYPES; type++) { |
231 | iids->cids += p_mngr->conn_cfg[type].cid_count; |
232 | vf_cids += p_mngr->conn_cfg[type].cids_per_vf; |
233 | |
234 | segs = p_mngr->conn_cfg[type].tid_seg; |
235 | /* for each segment there is at most one |
236 | * protocol for which count is not 0. |
237 | */ |
238 | for (j = 0; j < NUM_TASK_PF_SEGMENTS; j++) |
239 | iids->tids += segs[j].count; |
240 | |
241 | /* The last array elelment is for the VFs. As for PF |
242 | * segments there can be only one protocol for |
243 | * which this value is not 0. |
244 | */ |
245 | vf_tids += segs[NUM_TASK_PF_SEGMENTS].count; |
246 | } |
247 | |
248 | iids->vf_cids = vf_cids; |
249 | iids->tids += vf_tids * p_mngr->vf_count; |
250 | |
251 | DP_VERBOSE(p_hwfn, QED_MSG_ILT, |
252 | "iids: CIDS %08x vf_cids %08x tids %08x vf_tids %08x\n" , |
253 | iids->cids, iids->vf_cids, iids->tids, vf_tids); |
254 | } |
255 | |
256 | static struct qed_tid_seg *qed_cxt_tid_seg_info(struct qed_hwfn *p_hwfn, |
257 | u32 seg) |
258 | { |
259 | struct qed_cxt_mngr *p_cfg = p_hwfn->p_cxt_mngr; |
260 | u32 i; |
261 | |
262 | /* Find the protocol with tid count > 0 for this segment. |
263 | * Note: there can only be one and this is already validated. |
264 | */ |
265 | for (i = 0; i < MAX_CONN_TYPES; i++) |
266 | if (p_cfg->conn_cfg[i].tid_seg[seg].count) |
267 | return &p_cfg->conn_cfg[i].tid_seg[seg]; |
268 | return NULL; |
269 | } |
270 | |
271 | static void qed_cxt_set_srq_count(struct qed_hwfn *p_hwfn, |
272 | u32 num_srqs, u32 num_xrc_srqs) |
273 | { |
274 | struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr; |
275 | |
276 | p_mgr->srq_count = num_srqs; |
277 | p_mgr->xrc_srq_count = num_xrc_srqs; |
278 | } |
279 | |
280 | u32 qed_cxt_get_ilt_page_size(struct qed_hwfn *p_hwfn, |
281 | enum ilt_clients ilt_client) |
282 | { |
283 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
284 | struct qed_ilt_client_cfg *p_cli = &p_mngr->clients[ilt_client]; |
285 | |
286 | return ILT_PAGE_IN_BYTES(p_cli->p_size.val); |
287 | } |
288 | |
289 | static u32 qed_cxt_xrc_srqs_per_page(struct qed_hwfn *p_hwfn) |
290 | { |
291 | u32 page_size; |
292 | |
293 | page_size = qed_cxt_get_ilt_page_size(p_hwfn, ilt_client: ILT_CLI_TSDM); |
294 | return page_size / XRC_SRQ_CXT_SIZE; |
295 | } |
296 | |
297 | u32 qed_cxt_get_total_srq_count(struct qed_hwfn *p_hwfn) |
298 | { |
299 | struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr; |
300 | u32 total_srqs; |
301 | |
302 | total_srqs = p_mgr->srq_count + p_mgr->xrc_srq_count; |
303 | |
304 | return total_srqs; |
305 | } |
306 | |
307 | /* set the iids count per protocol */ |
308 | static void qed_cxt_set_proto_cid_count(struct qed_hwfn *p_hwfn, |
309 | enum protocol_type type, |
310 | u32 cid_count, u32 vf_cid_cnt) |
311 | { |
312 | struct qed_cxt_mngr *p_mgr = p_hwfn->p_cxt_mngr; |
313 | struct qed_conn_type_cfg *p_conn = &p_mgr->conn_cfg[type]; |
314 | |
315 | p_conn->cid_count = roundup(cid_count, DQ_RANGE_ALIGN); |
316 | p_conn->cids_per_vf = roundup(vf_cid_cnt, DQ_RANGE_ALIGN); |
317 | |
318 | if (type == PROTOCOLID_ROCE) { |
319 | u32 page_sz = p_mgr->clients[ILT_CLI_CDUC].p_size.val; |
320 | u32 cxt_size = CONN_CXT_SIZE(p_hwfn); |
321 | u32 elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size; |
322 | u32 align = elems_per_page * DQ_RANGE_ALIGN; |
323 | |
324 | p_conn->cid_count = roundup(p_conn->cid_count, align); |
325 | } |
326 | } |
327 | |
328 | u32 qed_cxt_get_proto_cid_count(struct qed_hwfn *p_hwfn, |
329 | enum protocol_type type, u32 *vf_cid) |
330 | { |
331 | if (vf_cid) |
332 | *vf_cid = p_hwfn->p_cxt_mngr->conn_cfg[type].cids_per_vf; |
333 | |
334 | return p_hwfn->p_cxt_mngr->conn_cfg[type].cid_count; |
335 | } |
336 | |
337 | u32 qed_cxt_get_proto_cid_start(struct qed_hwfn *p_hwfn, |
338 | enum protocol_type type) |
339 | { |
340 | return p_hwfn->p_cxt_mngr->acquired[type].start_cid; |
341 | } |
342 | |
343 | u32 qed_cxt_get_proto_tid_count(struct qed_hwfn *p_hwfn, |
344 | enum protocol_type type) |
345 | { |
346 | u32 cnt = 0; |
347 | int i; |
348 | |
349 | for (i = 0; i < TASK_SEGMENTS; i++) |
350 | cnt += p_hwfn->p_cxt_mngr->conn_cfg[type].tid_seg[i].count; |
351 | |
352 | return cnt; |
353 | } |
354 | |
355 | static void qed_cxt_set_proto_tid_count(struct qed_hwfn *p_hwfn, |
356 | enum protocol_type proto, |
357 | u8 seg, |
358 | u8 seg_type, u32 count, bool has_fl) |
359 | { |
360 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
361 | struct qed_tid_seg *p_seg = &p_mngr->conn_cfg[proto].tid_seg[seg]; |
362 | |
363 | p_seg->count = count; |
364 | p_seg->has_fl_mem = has_fl; |
365 | p_seg->type = seg_type; |
366 | } |
367 | |
368 | static void qed_ilt_cli_blk_fill(struct qed_ilt_client_cfg *p_cli, |
369 | struct qed_ilt_cli_blk *p_blk, |
370 | u32 start_line, u32 total_size, u32 elem_size) |
371 | { |
372 | u32 ilt_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val); |
373 | |
374 | /* verify thatits called only once for each block */ |
375 | if (p_blk->total_size) |
376 | return; |
377 | |
378 | p_blk->total_size = total_size; |
379 | p_blk->real_size_in_page = 0; |
380 | if (elem_size) |
381 | p_blk->real_size_in_page = (ilt_size / elem_size) * elem_size; |
382 | p_blk->start_line = start_line; |
383 | } |
384 | |
385 | static void qed_ilt_cli_adv_line(struct qed_hwfn *p_hwfn, |
386 | struct qed_ilt_client_cfg *p_cli, |
387 | struct qed_ilt_cli_blk *p_blk, |
388 | u32 *p_line, enum ilt_clients client_id) |
389 | { |
390 | if (!p_blk->total_size) |
391 | return; |
392 | |
393 | if (!p_cli->active) |
394 | p_cli->first.val = *p_line; |
395 | |
396 | p_cli->active = true; |
397 | *p_line += DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page); |
398 | p_cli->last.val = *p_line - 1; |
399 | |
400 | DP_VERBOSE(p_hwfn, QED_MSG_ILT, |
401 | "ILT[Client %d] - Lines: [%08x - %08x]. Block - Size %08x [Real %08x] Start line %d\n" , |
402 | client_id, p_cli->first.val, |
403 | p_cli->last.val, p_blk->total_size, |
404 | p_blk->real_size_in_page, p_blk->start_line); |
405 | } |
406 | |
407 | static u32 qed_ilt_get_dynamic_line_cnt(struct qed_hwfn *p_hwfn, |
408 | enum ilt_clients ilt_client) |
409 | { |
410 | u32 cid_count = p_hwfn->p_cxt_mngr->conn_cfg[PROTOCOLID_ROCE].cid_count; |
411 | struct qed_ilt_client_cfg *p_cli; |
412 | u32 lines_to_skip = 0; |
413 | u32 cxts_per_p; |
414 | |
415 | if (ilt_client == ILT_CLI_CDUC) { |
416 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC]; |
417 | |
418 | cxts_per_p = ILT_PAGE_IN_BYTES(p_cli->p_size.val) / |
419 | (u32) CONN_CXT_SIZE(p_hwfn); |
420 | |
421 | lines_to_skip = cid_count / cxts_per_p; |
422 | } |
423 | |
424 | return lines_to_skip; |
425 | } |
426 | |
427 | static struct qed_ilt_client_cfg *qed_cxt_set_cli(struct qed_ilt_client_cfg |
428 | *p_cli) |
429 | { |
430 | p_cli->active = false; |
431 | p_cli->first.val = 0; |
432 | p_cli->last.val = 0; |
433 | return p_cli; |
434 | } |
435 | |
436 | static struct qed_ilt_cli_blk *qed_cxt_set_blk(struct qed_ilt_cli_blk *p_blk) |
437 | { |
438 | p_blk->total_size = 0; |
439 | return p_blk; |
440 | } |
441 | |
442 | static void qed_cxt_ilt_blk_reset(struct qed_hwfn *p_hwfn) |
443 | { |
444 | struct qed_ilt_client_cfg *clients = p_hwfn->p_cxt_mngr->clients; |
445 | u32 cli_idx, blk_idx; |
446 | |
447 | for (cli_idx = 0; cli_idx < MAX_ILT_CLIENTS; cli_idx++) { |
448 | for (blk_idx = 0; blk_idx < ILT_CLI_PF_BLOCKS; blk_idx++) |
449 | clients[cli_idx].pf_blks[blk_idx].total_size = 0; |
450 | |
451 | for (blk_idx = 0; blk_idx < ILT_CLI_VF_BLOCKS; blk_idx++) |
452 | clients[cli_idx].vf_blks[blk_idx].total_size = 0; |
453 | } |
454 | } |
455 | |
456 | int qed_cxt_cfg_ilt_compute(struct qed_hwfn *p_hwfn, u32 *line_count) |
457 | { |
458 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
459 | u32 curr_line, total, i, task_size, line; |
460 | struct qed_ilt_client_cfg *p_cli; |
461 | struct qed_ilt_cli_blk *p_blk; |
462 | struct qed_cdu_iids cdu_iids; |
463 | struct qed_src_iids src_iids; |
464 | struct qed_qm_iids qm_iids; |
465 | struct qed_tm_iids tm_iids; |
466 | struct qed_tid_seg *p_seg; |
467 | |
468 | memset(&qm_iids, 0, sizeof(qm_iids)); |
469 | memset(&cdu_iids, 0, sizeof(cdu_iids)); |
470 | memset(&src_iids, 0, sizeof(src_iids)); |
471 | memset(&tm_iids, 0, sizeof(tm_iids)); |
472 | |
473 | p_mngr->pf_start_line = RESC_START(p_hwfn, QED_ILT); |
474 | |
475 | /* Reset all ILT blocks at the beginning of ILT computing in order |
476 | * to prevent memory allocation for irrelevant blocks afterwards. |
477 | */ |
478 | qed_cxt_ilt_blk_reset(p_hwfn); |
479 | |
480 | DP_VERBOSE(p_hwfn, QED_MSG_ILT, |
481 | "hwfn [%d] - Set context manager starting line to be 0x%08x\n" , |
482 | p_hwfn->my_id, p_hwfn->p_cxt_mngr->pf_start_line); |
483 | |
484 | /* CDUC */ |
485 | p_cli = qed_cxt_set_cli(p_cli: &p_mngr->clients[ILT_CLI_CDUC]); |
486 | |
487 | curr_line = p_mngr->pf_start_line; |
488 | |
489 | /* CDUC PF */ |
490 | p_cli->pf_total_lines = 0; |
491 | |
492 | /* get the counters for the CDUC and QM clients */ |
493 | qed_cxt_cdu_iids(p_mngr, iids: &cdu_iids); |
494 | |
495 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[CDUC_BLK]); |
496 | |
497 | total = cdu_iids.pf_cids * CONN_CXT_SIZE(p_hwfn); |
498 | |
499 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line, |
500 | total_size: total, CONN_CXT_SIZE(p_hwfn)); |
501 | |
502 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, client_id: ILT_CLI_CDUC); |
503 | p_cli->pf_total_lines = curr_line - p_blk->start_line; |
504 | |
505 | p_blk->dynamic_line_cnt = qed_ilt_get_dynamic_line_cnt(p_hwfn, |
506 | ilt_client: ILT_CLI_CDUC); |
507 | |
508 | /* CDUC VF */ |
509 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->vf_blks[CDUC_BLK]); |
510 | total = cdu_iids.per_vf_cids * CONN_CXT_SIZE(p_hwfn); |
511 | |
512 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line, |
513 | total_size: total, CONN_CXT_SIZE(p_hwfn)); |
514 | |
515 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, client_id: ILT_CLI_CDUC); |
516 | p_cli->vf_total_lines = curr_line - p_blk->start_line; |
517 | |
518 | for (i = 1; i < p_mngr->vf_count; i++) |
519 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
520 | client_id: ILT_CLI_CDUC); |
521 | |
522 | /* CDUT PF */ |
523 | p_cli = qed_cxt_set_cli(p_cli: &p_mngr->clients[ILT_CLI_CDUT]); |
524 | p_cli->first.val = curr_line; |
525 | |
526 | /* first the 'working' task memory */ |
527 | for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) { |
528 | p_seg = qed_cxt_tid_seg_info(p_hwfn, seg: i); |
529 | if (!p_seg || p_seg->count == 0) |
530 | continue; |
531 | |
532 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[CDUT_SEG_BLK(i)]); |
533 | total = p_seg->count * p_mngr->task_type_size[p_seg->type]; |
534 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line, total_size: total, |
535 | elem_size: p_mngr->task_type_size[p_seg->type]); |
536 | |
537 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
538 | client_id: ILT_CLI_CDUT); |
539 | } |
540 | |
541 | /* next the 'init' task memory (forced load memory) */ |
542 | for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) { |
543 | p_seg = qed_cxt_tid_seg_info(p_hwfn, seg: i); |
544 | if (!p_seg || p_seg->count == 0) |
545 | continue; |
546 | |
547 | p_blk = |
548 | qed_cxt_set_blk(p_blk: &p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)]); |
549 | |
550 | if (!p_seg->has_fl_mem) { |
551 | /* The segment is active (total size pf 'working' |
552 | * memory is > 0) but has no FL (forced-load, Init) |
553 | * memory. Thus: |
554 | * |
555 | * 1. The total-size in the corrsponding FL block of |
556 | * the ILT client is set to 0 - No ILT line are |
557 | * provisioned and no ILT memory allocated. |
558 | * |
559 | * 2. The start-line of said block is set to the |
560 | * start line of the matching working memory |
561 | * block in the ILT client. This is later used to |
562 | * configure the CDU segment offset registers and |
563 | * results in an FL command for TIDs of this |
564 | * segement behaves as regular load commands |
565 | * (loading TIDs from the working memory). |
566 | */ |
567 | line = p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line; |
568 | |
569 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: line, total_size: 0, elem_size: 0); |
570 | continue; |
571 | } |
572 | total = p_seg->count * p_mngr->task_type_size[p_seg->type]; |
573 | |
574 | qed_ilt_cli_blk_fill(p_cli, p_blk, |
575 | start_line: curr_line, total_size: total, |
576 | elem_size: p_mngr->task_type_size[p_seg->type]); |
577 | |
578 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
579 | client_id: ILT_CLI_CDUT); |
580 | } |
581 | p_cli->pf_total_lines = curr_line - p_cli->pf_blks[0].start_line; |
582 | |
583 | /* CDUT VF */ |
584 | p_seg = qed_cxt_tid_seg_info(p_hwfn, TASK_SEGMENT_VF); |
585 | if (p_seg && p_seg->count) { |
586 | /* Stricly speaking we need to iterate over all VF |
587 | * task segment types, but a VF has only 1 segment |
588 | */ |
589 | |
590 | /* 'working' memory */ |
591 | total = p_seg->count * p_mngr->task_type_size[p_seg->type]; |
592 | |
593 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->vf_blks[CDUT_SEG_BLK(0)]); |
594 | qed_ilt_cli_blk_fill(p_cli, p_blk, |
595 | start_line: curr_line, total_size: total, |
596 | elem_size: p_mngr->task_type_size[p_seg->type]); |
597 | |
598 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
599 | client_id: ILT_CLI_CDUT); |
600 | |
601 | /* 'init' memory */ |
602 | p_blk = |
603 | qed_cxt_set_blk(p_blk: &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)]); |
604 | if (!p_seg->has_fl_mem) { |
605 | /* see comment above */ |
606 | line = p_cli->vf_blks[CDUT_SEG_BLK(0)].start_line; |
607 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: line, total_size: 0, elem_size: 0); |
608 | } else { |
609 | task_size = p_mngr->task_type_size[p_seg->type]; |
610 | qed_ilt_cli_blk_fill(p_cli, p_blk, |
611 | start_line: curr_line, total_size: total, elem_size: task_size); |
612 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
613 | client_id: ILT_CLI_CDUT); |
614 | } |
615 | p_cli->vf_total_lines = curr_line - |
616 | p_cli->vf_blks[0].start_line; |
617 | |
618 | /* Now for the rest of the VFs */ |
619 | for (i = 1; i < p_mngr->vf_count; i++) { |
620 | p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(0)]; |
621 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
622 | client_id: ILT_CLI_CDUT); |
623 | |
624 | p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(0, VF)]; |
625 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
626 | client_id: ILT_CLI_CDUT); |
627 | } |
628 | } |
629 | |
630 | /* QM */ |
631 | p_cli = qed_cxt_set_cli(p_cli: &p_mngr->clients[ILT_CLI_QM]); |
632 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[0]); |
633 | |
634 | qed_cxt_qm_iids(p_hwfn, iids: &qm_iids); |
635 | total = qed_qm_pf_mem_size(num_pf_cids: qm_iids.cids, |
636 | num_vf_cids: qm_iids.vf_cids, num_tids: qm_iids.tids, |
637 | num_pf_pqs: p_hwfn->qm_info.num_pqs, |
638 | num_vf_pqs: p_hwfn->qm_info.num_vf_pqs); |
639 | |
640 | DP_VERBOSE(p_hwfn, |
641 | QED_MSG_ILT, |
642 | "QM ILT Info, (cids=%d, vf_cids=%d, tids=%d, num_pqs=%d, num_vf_pqs=%d, memory_size=%d)\n" , |
643 | qm_iids.cids, |
644 | qm_iids.vf_cids, |
645 | qm_iids.tids, |
646 | p_hwfn->qm_info.num_pqs, p_hwfn->qm_info.num_vf_pqs, total); |
647 | |
648 | qed_ilt_cli_blk_fill(p_cli, p_blk, |
649 | start_line: curr_line, total_size: total * 0x1000, |
650 | QM_PQ_ELEMENT_SIZE); |
651 | |
652 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, client_id: ILT_CLI_QM); |
653 | p_cli->pf_total_lines = curr_line - p_blk->start_line; |
654 | |
655 | /* SRC */ |
656 | p_cli = qed_cxt_set_cli(p_cli: &p_mngr->clients[ILT_CLI_SRC]); |
657 | qed_cxt_src_iids(p_mngr, iids: &src_iids); |
658 | |
659 | /* Both the PF and VFs searcher connections are stored in the per PF |
660 | * database. Thus sum the PF searcher cids and all the VFs searcher |
661 | * cids. |
662 | */ |
663 | total = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count; |
664 | if (total) { |
665 | u32 local_max = max_t(u32, total, |
666 | SRC_MIN_NUM_ELEMS); |
667 | |
668 | total = roundup_pow_of_two(local_max); |
669 | |
670 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[0]); |
671 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line, |
672 | total_size: total * sizeof(struct src_ent), |
673 | elem_size: sizeof(struct src_ent)); |
674 | |
675 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
676 | client_id: ILT_CLI_SRC); |
677 | p_cli->pf_total_lines = curr_line - p_blk->start_line; |
678 | } |
679 | |
680 | /* TM PF */ |
681 | p_cli = qed_cxt_set_cli(p_cli: &p_mngr->clients[ILT_CLI_TM]); |
682 | qed_cxt_tm_iids(p_hwfn, p_mngr, iids: &tm_iids); |
683 | total = tm_iids.pf_cids + tm_iids.pf_tids_total; |
684 | if (total) { |
685 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[0]); |
686 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line, |
687 | total_size: total * TM_ELEM_SIZE, TM_ELEM_SIZE); |
688 | |
689 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
690 | client_id: ILT_CLI_TM); |
691 | p_cli->pf_total_lines = curr_line - p_blk->start_line; |
692 | } |
693 | |
694 | /* TM VF */ |
695 | total = tm_iids.per_vf_cids + tm_iids.per_vf_tids; |
696 | if (total) { |
697 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->vf_blks[0]); |
698 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line, |
699 | total_size: total * TM_ELEM_SIZE, TM_ELEM_SIZE); |
700 | |
701 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
702 | client_id: ILT_CLI_TM); |
703 | |
704 | p_cli->vf_total_lines = curr_line - p_blk->start_line; |
705 | for (i = 1; i < p_mngr->vf_count; i++) |
706 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
707 | client_id: ILT_CLI_TM); |
708 | } |
709 | |
710 | /* TSDM (SRQ CONTEXT) */ |
711 | total = qed_cxt_get_total_srq_count(p_hwfn); |
712 | |
713 | if (total) { |
714 | p_cli = qed_cxt_set_cli(p_cli: &p_mngr->clients[ILT_CLI_TSDM]); |
715 | p_blk = qed_cxt_set_blk(p_blk: &p_cli->pf_blks[SRQ_BLK]); |
716 | qed_ilt_cli_blk_fill(p_cli, p_blk, start_line: curr_line, |
717 | total_size: total * SRQ_CXT_SIZE, SRQ_CXT_SIZE); |
718 | |
719 | qed_ilt_cli_adv_line(p_hwfn, p_cli, p_blk, p_line: &curr_line, |
720 | client_id: ILT_CLI_TSDM); |
721 | p_cli->pf_total_lines = curr_line - p_blk->start_line; |
722 | } |
723 | |
724 | *line_count = curr_line - p_hwfn->p_cxt_mngr->pf_start_line; |
725 | |
726 | if (curr_line - p_hwfn->p_cxt_mngr->pf_start_line > |
727 | RESC_NUM(p_hwfn, QED_ILT)) |
728 | return -EINVAL; |
729 | |
730 | return 0; |
731 | } |
732 | |
733 | u32 qed_cxt_cfg_ilt_compute_excess(struct qed_hwfn *p_hwfn, u32 used_lines) |
734 | { |
735 | struct qed_ilt_client_cfg *p_cli; |
736 | u32 excess_lines, available_lines; |
737 | struct qed_cxt_mngr *p_mngr; |
738 | u32 ilt_page_size, elem_size; |
739 | struct qed_tid_seg *p_seg; |
740 | int i; |
741 | |
742 | available_lines = RESC_NUM(p_hwfn, QED_ILT); |
743 | excess_lines = used_lines - available_lines; |
744 | |
745 | if (!excess_lines) |
746 | return 0; |
747 | |
748 | if (!QED_IS_RDMA_PERSONALITY(p_hwfn)) |
749 | return 0; |
750 | |
751 | p_mngr = p_hwfn->p_cxt_mngr; |
752 | p_cli = &p_mngr->clients[ILT_CLI_CDUT]; |
753 | ilt_page_size = ILT_PAGE_IN_BYTES(p_cli->p_size.val); |
754 | |
755 | for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) { |
756 | p_seg = qed_cxt_tid_seg_info(p_hwfn, seg: i); |
757 | if (!p_seg || p_seg->count == 0) |
758 | continue; |
759 | |
760 | elem_size = p_mngr->task_type_size[p_seg->type]; |
761 | if (!elem_size) |
762 | continue; |
763 | |
764 | return (ilt_page_size / elem_size) * excess_lines; |
765 | } |
766 | |
767 | DP_NOTICE(p_hwfn, "failed computing excess ILT lines\n" ); |
768 | return 0; |
769 | } |
770 | |
771 | static void qed_cxt_src_t2_free(struct qed_hwfn *p_hwfn) |
772 | { |
773 | struct qed_src_t2 *p_t2 = &p_hwfn->p_cxt_mngr->src_t2; |
774 | u32 i; |
775 | |
776 | if (!p_t2 || !p_t2->dma_mem) |
777 | return; |
778 | |
779 | for (i = 0; i < p_t2->num_pages; i++) |
780 | if (p_t2->dma_mem[i].virt_addr) |
781 | dma_free_coherent(dev: &p_hwfn->cdev->pdev->dev, |
782 | size: p_t2->dma_mem[i].size, |
783 | cpu_addr: p_t2->dma_mem[i].virt_addr, |
784 | dma_handle: p_t2->dma_mem[i].phys_addr); |
785 | |
786 | kfree(objp: p_t2->dma_mem); |
787 | p_t2->dma_mem = NULL; |
788 | } |
789 | |
790 | static int |
791 | qed_cxt_t2_alloc_pages(struct qed_hwfn *p_hwfn, |
792 | struct qed_src_t2 *p_t2, u32 total_size, u32 page_size) |
793 | { |
794 | void **p_virt; |
795 | u32 size, i; |
796 | |
797 | if (!p_t2 || !p_t2->dma_mem) |
798 | return -EINVAL; |
799 | |
800 | for (i = 0; i < p_t2->num_pages; i++) { |
801 | size = min_t(u32, total_size, page_size); |
802 | p_virt = &p_t2->dma_mem[i].virt_addr; |
803 | |
804 | *p_virt = dma_alloc_coherent(dev: &p_hwfn->cdev->pdev->dev, |
805 | size, |
806 | dma_handle: &p_t2->dma_mem[i].phys_addr, |
807 | GFP_KERNEL); |
808 | if (!p_t2->dma_mem[i].virt_addr) |
809 | return -ENOMEM; |
810 | |
811 | memset(*p_virt, 0, size); |
812 | p_t2->dma_mem[i].size = size; |
813 | total_size -= size; |
814 | } |
815 | |
816 | return 0; |
817 | } |
818 | |
819 | static int qed_cxt_src_t2_alloc(struct qed_hwfn *p_hwfn) |
820 | { |
821 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
822 | u32 conn_num, total_size, ent_per_page, psz, i; |
823 | struct phys_mem_desc *p_t2_last_page; |
824 | struct qed_ilt_client_cfg *p_src; |
825 | struct qed_src_iids src_iids; |
826 | struct qed_src_t2 *p_t2; |
827 | int rc; |
828 | |
829 | memset(&src_iids, 0, sizeof(src_iids)); |
830 | |
831 | /* if the SRC ILT client is inactive - there are no connection |
832 | * requiring the searcer, leave. |
833 | */ |
834 | p_src = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_SRC]; |
835 | if (!p_src->active) |
836 | return 0; |
837 | |
838 | qed_cxt_src_iids(p_mngr, iids: &src_iids); |
839 | conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count; |
840 | total_size = conn_num * sizeof(struct src_ent); |
841 | |
842 | /* use the same page size as the SRC ILT client */ |
843 | psz = ILT_PAGE_IN_BYTES(p_src->p_size.val); |
844 | p_t2 = &p_mngr->src_t2; |
845 | p_t2->num_pages = DIV_ROUND_UP(total_size, psz); |
846 | |
847 | /* allocate t2 */ |
848 | p_t2->dma_mem = kcalloc(n: p_t2->num_pages, size: sizeof(struct phys_mem_desc), |
849 | GFP_KERNEL); |
850 | if (!p_t2->dma_mem) { |
851 | DP_NOTICE(p_hwfn, "Failed to allocate t2 table\n" ); |
852 | rc = -ENOMEM; |
853 | goto t2_fail; |
854 | } |
855 | |
856 | rc = qed_cxt_t2_alloc_pages(p_hwfn, p_t2, total_size, page_size: psz); |
857 | if (rc) |
858 | goto t2_fail; |
859 | |
860 | /* Set the t2 pointers */ |
861 | |
862 | /* entries per page - must be a power of two */ |
863 | ent_per_page = psz / sizeof(struct src_ent); |
864 | |
865 | p_t2->first_free = (u64)p_t2->dma_mem[0].phys_addr; |
866 | |
867 | p_t2_last_page = &p_t2->dma_mem[(conn_num - 1) / ent_per_page]; |
868 | p_t2->last_free = (u64)p_t2_last_page->phys_addr + |
869 | ((conn_num - 1) & (ent_per_page - 1)) * sizeof(struct src_ent); |
870 | |
871 | for (i = 0; i < p_t2->num_pages; i++) { |
872 | u32 ent_num = min_t(u32, |
873 | ent_per_page, |
874 | conn_num); |
875 | struct src_ent *entries = p_t2->dma_mem[i].virt_addr; |
876 | u64 p_ent_phys = (u64)p_t2->dma_mem[i].phys_addr, val; |
877 | u32 j; |
878 | |
879 | for (j = 0; j < ent_num - 1; j++) { |
880 | val = p_ent_phys + (j + 1) * sizeof(struct src_ent); |
881 | entries[j].next = cpu_to_be64(val); |
882 | } |
883 | |
884 | if (i < p_t2->num_pages - 1) |
885 | val = (u64)p_t2->dma_mem[i + 1].phys_addr; |
886 | else |
887 | val = 0; |
888 | entries[j].next = cpu_to_be64(val); |
889 | |
890 | conn_num -= ent_num; |
891 | } |
892 | |
893 | return 0; |
894 | |
895 | t2_fail: |
896 | qed_cxt_src_t2_free(p_hwfn); |
897 | return rc; |
898 | } |
899 | |
900 | #define for_each_ilt_valid_client(pos, clients) \ |
901 | for (pos = 0; pos < MAX_ILT_CLIENTS; pos++) \ |
902 | if (!clients[pos].active) { \ |
903 | continue; \ |
904 | } else \ |
905 | |
906 | /* Total number of ILT lines used by this PF */ |
907 | static u32 qed_cxt_ilt_shadow_size(struct qed_ilt_client_cfg *ilt_clients) |
908 | { |
909 | u32 size = 0; |
910 | u32 i; |
911 | |
912 | for_each_ilt_valid_client(i, ilt_clients) |
913 | size += (ilt_clients[i].last.val - ilt_clients[i].first.val + 1); |
914 | |
915 | return size; |
916 | } |
917 | |
918 | static void qed_ilt_shadow_free(struct qed_hwfn *p_hwfn) |
919 | { |
920 | struct qed_ilt_client_cfg *p_cli = p_hwfn->p_cxt_mngr->clients; |
921 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
922 | u32 ilt_size, i; |
923 | |
924 | ilt_size = qed_cxt_ilt_shadow_size(ilt_clients: p_cli); |
925 | |
926 | for (i = 0; p_mngr->ilt_shadow && i < ilt_size; i++) { |
927 | struct phys_mem_desc *p_dma = &p_mngr->ilt_shadow[i]; |
928 | |
929 | if (p_dma->virt_addr) |
930 | dma_free_coherent(dev: &p_hwfn->cdev->pdev->dev, |
931 | size: p_dma->size, cpu_addr: p_dma->virt_addr, |
932 | dma_handle: p_dma->phys_addr); |
933 | p_dma->virt_addr = NULL; |
934 | } |
935 | kfree(objp: p_mngr->ilt_shadow); |
936 | p_mngr->ilt_shadow = NULL; |
937 | } |
938 | |
939 | static int qed_ilt_blk_alloc(struct qed_hwfn *p_hwfn, |
940 | struct qed_ilt_cli_blk *p_blk, |
941 | enum ilt_clients ilt_client, |
942 | u32 start_line_offset) |
943 | { |
944 | struct phys_mem_desc *ilt_shadow = p_hwfn->p_cxt_mngr->ilt_shadow; |
945 | u32 lines, line, sz_left, lines_to_skip = 0; |
946 | |
947 | /* Special handling for RoCE that supports dynamic allocation */ |
948 | if (QED_IS_RDMA_PERSONALITY(p_hwfn) && |
949 | ((ilt_client == ILT_CLI_CDUT) || ilt_client == ILT_CLI_TSDM)) |
950 | return 0; |
951 | |
952 | lines_to_skip = p_blk->dynamic_line_cnt; |
953 | |
954 | if (!p_blk->total_size) |
955 | return 0; |
956 | |
957 | sz_left = p_blk->total_size; |
958 | lines = DIV_ROUND_UP(sz_left, p_blk->real_size_in_page) - lines_to_skip; |
959 | line = p_blk->start_line + start_line_offset - |
960 | p_hwfn->p_cxt_mngr->pf_start_line + lines_to_skip; |
961 | |
962 | for (; lines; lines--) { |
963 | dma_addr_t p_phys; |
964 | void *p_virt; |
965 | u32 size; |
966 | |
967 | size = min_t(u32, sz_left, p_blk->real_size_in_page); |
968 | p_virt = dma_alloc_coherent(dev: &p_hwfn->cdev->pdev->dev, size, |
969 | dma_handle: &p_phys, GFP_KERNEL); |
970 | if (!p_virt) |
971 | return -ENOMEM; |
972 | |
973 | ilt_shadow[line].phys_addr = p_phys; |
974 | ilt_shadow[line].virt_addr = p_virt; |
975 | ilt_shadow[line].size = size; |
976 | |
977 | DP_VERBOSE(p_hwfn, QED_MSG_ILT, |
978 | "ILT shadow: Line [%d] Physical 0x%llx Virtual %p Size %d\n" , |
979 | line, (u64)p_phys, p_virt, size); |
980 | |
981 | sz_left -= size; |
982 | line++; |
983 | } |
984 | |
985 | return 0; |
986 | } |
987 | |
988 | static int qed_ilt_shadow_alloc(struct qed_hwfn *p_hwfn) |
989 | { |
990 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
991 | struct qed_ilt_client_cfg *clients = p_mngr->clients; |
992 | struct qed_ilt_cli_blk *p_blk; |
993 | u32 size, i, j, k; |
994 | int rc; |
995 | |
996 | size = qed_cxt_ilt_shadow_size(ilt_clients: clients); |
997 | p_mngr->ilt_shadow = kcalloc(n: size, size: sizeof(struct phys_mem_desc), |
998 | GFP_KERNEL); |
999 | if (!p_mngr->ilt_shadow) { |
1000 | rc = -ENOMEM; |
1001 | goto ilt_shadow_fail; |
1002 | } |
1003 | |
1004 | DP_VERBOSE(p_hwfn, QED_MSG_ILT, |
1005 | "Allocated 0x%x bytes for ilt shadow\n" , |
1006 | (u32)(size * sizeof(struct phys_mem_desc))); |
1007 | |
1008 | for_each_ilt_valid_client(i, clients) { |
1009 | for (j = 0; j < ILT_CLI_PF_BLOCKS; j++) { |
1010 | p_blk = &clients[i].pf_blks[j]; |
1011 | rc = qed_ilt_blk_alloc(p_hwfn, p_blk, ilt_client: i, start_line_offset: 0); |
1012 | if (rc) |
1013 | goto ilt_shadow_fail; |
1014 | } |
1015 | for (k = 0; k < p_mngr->vf_count; k++) { |
1016 | for (j = 0; j < ILT_CLI_VF_BLOCKS; j++) { |
1017 | u32 lines = clients[i].vf_total_lines * k; |
1018 | |
1019 | p_blk = &clients[i].vf_blks[j]; |
1020 | rc = qed_ilt_blk_alloc(p_hwfn, p_blk, ilt_client: i, start_line_offset: lines); |
1021 | if (rc) |
1022 | goto ilt_shadow_fail; |
1023 | } |
1024 | } |
1025 | } |
1026 | |
1027 | return 0; |
1028 | |
1029 | ilt_shadow_fail: |
1030 | qed_ilt_shadow_free(p_hwfn); |
1031 | return rc; |
1032 | } |
1033 | |
1034 | static void qed_cid_map_free(struct qed_hwfn *p_hwfn) |
1035 | { |
1036 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1037 | u32 type, vf; |
1038 | |
1039 | for (type = 0; type < MAX_CONN_TYPES; type++) { |
1040 | bitmap_free(bitmap: p_mngr->acquired[type].cid_map); |
1041 | p_mngr->acquired[type].max_count = 0; |
1042 | p_mngr->acquired[type].start_cid = 0; |
1043 | |
1044 | for (vf = 0; vf < MAX_NUM_VFS; vf++) { |
1045 | bitmap_free(bitmap: p_mngr->acquired_vf[type][vf].cid_map); |
1046 | p_mngr->acquired_vf[type][vf].max_count = 0; |
1047 | p_mngr->acquired_vf[type][vf].start_cid = 0; |
1048 | } |
1049 | } |
1050 | } |
1051 | |
1052 | static int |
1053 | qed_cid_map_alloc_single(struct qed_hwfn *p_hwfn, |
1054 | u32 type, |
1055 | u32 cid_start, |
1056 | u32 cid_count, struct qed_cid_acquired_map *p_map) |
1057 | { |
1058 | if (!cid_count) |
1059 | return 0; |
1060 | |
1061 | p_map->cid_map = bitmap_zalloc(nbits: cid_count, GFP_KERNEL); |
1062 | if (!p_map->cid_map) |
1063 | return -ENOMEM; |
1064 | |
1065 | p_map->max_count = cid_count; |
1066 | p_map->start_cid = cid_start; |
1067 | |
1068 | DP_VERBOSE(p_hwfn, QED_MSG_CXT, |
1069 | "Type %08x start: %08x count %08x\n" , |
1070 | type, p_map->start_cid, p_map->max_count); |
1071 | |
1072 | return 0; |
1073 | } |
1074 | |
1075 | static int qed_cid_map_alloc(struct qed_hwfn *p_hwfn) |
1076 | { |
1077 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1078 | u32 start_cid = 0, vf_start_cid = 0; |
1079 | u32 type, vf; |
1080 | |
1081 | for (type = 0; type < MAX_CONN_TYPES; type++) { |
1082 | struct qed_conn_type_cfg *p_cfg = &p_mngr->conn_cfg[type]; |
1083 | struct qed_cid_acquired_map *p_map; |
1084 | |
1085 | /* Handle PF maps */ |
1086 | p_map = &p_mngr->acquired[type]; |
1087 | if (qed_cid_map_alloc_single(p_hwfn, type, cid_start: start_cid, |
1088 | cid_count: p_cfg->cid_count, p_map)) |
1089 | goto cid_map_fail; |
1090 | |
1091 | /* Handle VF maps */ |
1092 | for (vf = 0; vf < MAX_NUM_VFS; vf++) { |
1093 | p_map = &p_mngr->acquired_vf[type][vf]; |
1094 | if (qed_cid_map_alloc_single(p_hwfn, type, |
1095 | cid_start: vf_start_cid, |
1096 | cid_count: p_cfg->cids_per_vf, p_map)) |
1097 | goto cid_map_fail; |
1098 | } |
1099 | |
1100 | start_cid += p_cfg->cid_count; |
1101 | vf_start_cid += p_cfg->cids_per_vf; |
1102 | } |
1103 | |
1104 | return 0; |
1105 | |
1106 | cid_map_fail: |
1107 | qed_cid_map_free(p_hwfn); |
1108 | return -ENOMEM; |
1109 | } |
1110 | |
1111 | int qed_cxt_mngr_alloc(struct qed_hwfn *p_hwfn) |
1112 | { |
1113 | struct qed_ilt_client_cfg *clients; |
1114 | struct qed_cxt_mngr *p_mngr; |
1115 | u32 i; |
1116 | |
1117 | p_mngr = kzalloc(size: sizeof(*p_mngr), GFP_KERNEL); |
1118 | if (!p_mngr) |
1119 | return -ENOMEM; |
1120 | |
1121 | /* Initialize ILT client registers */ |
1122 | clients = p_mngr->clients; |
1123 | clients[ILT_CLI_CDUC].first.reg = ILT_CFG_REG(CDUC, FIRST_ILT); |
1124 | clients[ILT_CLI_CDUC].last.reg = ILT_CFG_REG(CDUC, LAST_ILT); |
1125 | clients[ILT_CLI_CDUC].p_size.reg = ILT_CFG_REG(CDUC, P_SIZE); |
1126 | |
1127 | clients[ILT_CLI_QM].first.reg = ILT_CFG_REG(QM, FIRST_ILT); |
1128 | clients[ILT_CLI_QM].last.reg = ILT_CFG_REG(QM, LAST_ILT); |
1129 | clients[ILT_CLI_QM].p_size.reg = ILT_CFG_REG(QM, P_SIZE); |
1130 | |
1131 | clients[ILT_CLI_TM].first.reg = ILT_CFG_REG(TM, FIRST_ILT); |
1132 | clients[ILT_CLI_TM].last.reg = ILT_CFG_REG(TM, LAST_ILT); |
1133 | clients[ILT_CLI_TM].p_size.reg = ILT_CFG_REG(TM, P_SIZE); |
1134 | |
1135 | clients[ILT_CLI_SRC].first.reg = ILT_CFG_REG(SRC, FIRST_ILT); |
1136 | clients[ILT_CLI_SRC].last.reg = ILT_CFG_REG(SRC, LAST_ILT); |
1137 | clients[ILT_CLI_SRC].p_size.reg = ILT_CFG_REG(SRC, P_SIZE); |
1138 | |
1139 | clients[ILT_CLI_CDUT].first.reg = ILT_CFG_REG(CDUT, FIRST_ILT); |
1140 | clients[ILT_CLI_CDUT].last.reg = ILT_CFG_REG(CDUT, LAST_ILT); |
1141 | clients[ILT_CLI_CDUT].p_size.reg = ILT_CFG_REG(CDUT, P_SIZE); |
1142 | |
1143 | clients[ILT_CLI_TSDM].first.reg = ILT_CFG_REG(TSDM, FIRST_ILT); |
1144 | clients[ILT_CLI_TSDM].last.reg = ILT_CFG_REG(TSDM, LAST_ILT); |
1145 | clients[ILT_CLI_TSDM].p_size.reg = ILT_CFG_REG(TSDM, P_SIZE); |
1146 | /* default ILT page size for all clients is 64K */ |
1147 | for (i = 0; i < MAX_ILT_CLIENTS; i++) |
1148 | p_mngr->clients[i].p_size.val = ILT_DEFAULT_HW_P_SIZE; |
1149 | |
1150 | p_mngr->conn_ctx_size = CONN_CXT_SIZE(p_hwfn); |
1151 | |
1152 | /* Initialize task sizes */ |
1153 | p_mngr->task_type_size[0] = TYPE0_TASK_CXT_SIZE(p_hwfn); |
1154 | p_mngr->task_type_size[1] = TYPE1_TASK_CXT_SIZE(p_hwfn); |
1155 | |
1156 | if (p_hwfn->cdev->p_iov_info) { |
1157 | p_mngr->vf_count = p_hwfn->cdev->p_iov_info->total_vfs; |
1158 | p_mngr->first_vf_in_pf = |
1159 | p_hwfn->cdev->p_iov_info->first_vf_in_pf; |
1160 | } |
1161 | /* Initialize the dynamic ILT allocation mutex */ |
1162 | mutex_init(&p_mngr->mutex); |
1163 | |
1164 | /* Set the cxt mangr pointer priori to further allocations */ |
1165 | p_hwfn->p_cxt_mngr = p_mngr; |
1166 | |
1167 | return 0; |
1168 | } |
1169 | |
1170 | int qed_cxt_tables_alloc(struct qed_hwfn *p_hwfn) |
1171 | { |
1172 | int rc; |
1173 | |
1174 | /* Allocate the ILT shadow table */ |
1175 | rc = qed_ilt_shadow_alloc(p_hwfn); |
1176 | if (rc) |
1177 | goto tables_alloc_fail; |
1178 | |
1179 | /* Allocate the T2 table */ |
1180 | rc = qed_cxt_src_t2_alloc(p_hwfn); |
1181 | if (rc) |
1182 | goto tables_alloc_fail; |
1183 | |
1184 | /* Allocate and initialize the acquired cids bitmaps */ |
1185 | rc = qed_cid_map_alloc(p_hwfn); |
1186 | if (rc) |
1187 | goto tables_alloc_fail; |
1188 | |
1189 | return 0; |
1190 | |
1191 | tables_alloc_fail: |
1192 | qed_cxt_mngr_free(p_hwfn); |
1193 | return rc; |
1194 | } |
1195 | |
1196 | void qed_cxt_mngr_free(struct qed_hwfn *p_hwfn) |
1197 | { |
1198 | if (!p_hwfn->p_cxt_mngr) |
1199 | return; |
1200 | |
1201 | qed_cid_map_free(p_hwfn); |
1202 | qed_cxt_src_t2_free(p_hwfn); |
1203 | qed_ilt_shadow_free(p_hwfn); |
1204 | kfree(objp: p_hwfn->p_cxt_mngr); |
1205 | |
1206 | p_hwfn->p_cxt_mngr = NULL; |
1207 | } |
1208 | |
1209 | void qed_cxt_mngr_setup(struct qed_hwfn *p_hwfn) |
1210 | { |
1211 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1212 | struct qed_cid_acquired_map *p_map; |
1213 | struct qed_conn_type_cfg *p_cfg; |
1214 | int type; |
1215 | |
1216 | /* Reset acquired cids */ |
1217 | for (type = 0; type < MAX_CONN_TYPES; type++) { |
1218 | u32 vf; |
1219 | |
1220 | p_cfg = &p_mngr->conn_cfg[type]; |
1221 | if (p_cfg->cid_count) { |
1222 | p_map = &p_mngr->acquired[type]; |
1223 | bitmap_zero(dst: p_map->cid_map, nbits: p_map->max_count); |
1224 | } |
1225 | |
1226 | if (!p_cfg->cids_per_vf) |
1227 | continue; |
1228 | |
1229 | for (vf = 0; vf < MAX_NUM_VFS; vf++) { |
1230 | p_map = &p_mngr->acquired_vf[type][vf]; |
1231 | bitmap_zero(dst: p_map->cid_map, nbits: p_map->max_count); |
1232 | } |
1233 | } |
1234 | } |
1235 | |
1236 | /* CDU Common */ |
1237 | #define CDUC_CXT_SIZE_SHIFT \ |
1238 | CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE_SHIFT |
1239 | |
1240 | #define CDUC_CXT_SIZE_MASK \ |
1241 | (CDU_REG_CID_ADDR_PARAMS_CONTEXT_SIZE >> CDUC_CXT_SIZE_SHIFT) |
1242 | |
1243 | #define CDUC_BLOCK_WASTE_SHIFT \ |
1244 | CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE_SHIFT |
1245 | |
1246 | #define CDUC_BLOCK_WASTE_MASK \ |
1247 | (CDU_REG_CID_ADDR_PARAMS_BLOCK_WASTE >> CDUC_BLOCK_WASTE_SHIFT) |
1248 | |
1249 | #define CDUC_NCIB_SHIFT \ |
1250 | CDU_REG_CID_ADDR_PARAMS_NCIB_SHIFT |
1251 | |
1252 | #define CDUC_NCIB_MASK \ |
1253 | (CDU_REG_CID_ADDR_PARAMS_NCIB >> CDUC_NCIB_SHIFT) |
1254 | |
1255 | #define CDUT_TYPE0_CXT_SIZE_SHIFT \ |
1256 | CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE_SHIFT |
1257 | |
1258 | #define CDUT_TYPE0_CXT_SIZE_MASK \ |
1259 | (CDU_REG_SEGMENT0_PARAMS_T0_TID_SIZE >> \ |
1260 | CDUT_TYPE0_CXT_SIZE_SHIFT) |
1261 | |
1262 | #define CDUT_TYPE0_BLOCK_WASTE_SHIFT \ |
1263 | CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE_SHIFT |
1264 | |
1265 | #define CDUT_TYPE0_BLOCK_WASTE_MASK \ |
1266 | (CDU_REG_SEGMENT0_PARAMS_T0_TID_BLOCK_WASTE >> \ |
1267 | CDUT_TYPE0_BLOCK_WASTE_SHIFT) |
1268 | |
1269 | #define CDUT_TYPE0_NCIB_SHIFT \ |
1270 | CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK_SHIFT |
1271 | |
1272 | #define CDUT_TYPE0_NCIB_MASK \ |
1273 | (CDU_REG_SEGMENT0_PARAMS_T0_NUM_TIDS_IN_BLOCK >> \ |
1274 | CDUT_TYPE0_NCIB_SHIFT) |
1275 | |
1276 | #define CDUT_TYPE1_CXT_SIZE_SHIFT \ |
1277 | CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE_SHIFT |
1278 | |
1279 | #define CDUT_TYPE1_CXT_SIZE_MASK \ |
1280 | (CDU_REG_SEGMENT1_PARAMS_T1_TID_SIZE >> \ |
1281 | CDUT_TYPE1_CXT_SIZE_SHIFT) |
1282 | |
1283 | #define CDUT_TYPE1_BLOCK_WASTE_SHIFT \ |
1284 | CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE_SHIFT |
1285 | |
1286 | #define CDUT_TYPE1_BLOCK_WASTE_MASK \ |
1287 | (CDU_REG_SEGMENT1_PARAMS_T1_TID_BLOCK_WASTE >> \ |
1288 | CDUT_TYPE1_BLOCK_WASTE_SHIFT) |
1289 | |
1290 | #define CDUT_TYPE1_NCIB_SHIFT \ |
1291 | CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK_SHIFT |
1292 | |
1293 | #define CDUT_TYPE1_NCIB_MASK \ |
1294 | (CDU_REG_SEGMENT1_PARAMS_T1_NUM_TIDS_IN_BLOCK >> \ |
1295 | CDUT_TYPE1_NCIB_SHIFT) |
1296 | |
1297 | static void qed_cdu_init_common(struct qed_hwfn *p_hwfn) |
1298 | { |
1299 | u32 page_sz, elems_per_page, block_waste, cxt_size, cdu_params = 0; |
1300 | |
1301 | /* CDUC - connection configuration */ |
1302 | page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val; |
1303 | cxt_size = CONN_CXT_SIZE(p_hwfn); |
1304 | elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size; |
1305 | block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size; |
1306 | |
1307 | SET_FIELD(cdu_params, CDUC_CXT_SIZE, cxt_size); |
1308 | SET_FIELD(cdu_params, CDUC_BLOCK_WASTE, block_waste); |
1309 | SET_FIELD(cdu_params, CDUC_NCIB, elems_per_page); |
1310 | STORE_RT_REG(p_hwfn, CDU_REG_CID_ADDR_PARAMS_RT_OFFSET, cdu_params); |
1311 | |
1312 | /* CDUT - type-0 tasks configuration */ |
1313 | page_sz = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT].p_size.val; |
1314 | cxt_size = p_hwfn->p_cxt_mngr->task_type_size[0]; |
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; |
1317 | |
1318 | /* cxt size and block-waste are multipes of 8 */ |
1319 | cdu_params = 0; |
1320 | SET_FIELD(cdu_params, CDUT_TYPE0_CXT_SIZE, (cxt_size >> 3)); |
1321 | SET_FIELD(cdu_params, CDUT_TYPE0_BLOCK_WASTE, (block_waste >> 3)); |
1322 | SET_FIELD(cdu_params, CDUT_TYPE0_NCIB, elems_per_page); |
1323 | STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT0_PARAMS_RT_OFFSET, cdu_params); |
1324 | |
1325 | /* CDUT - type-1 tasks configuration */ |
1326 | cxt_size = p_hwfn->p_cxt_mngr->task_type_size[1]; |
1327 | elems_per_page = ILT_PAGE_IN_BYTES(page_sz) / cxt_size; |
1328 | block_waste = ILT_PAGE_IN_BYTES(page_sz) - elems_per_page * cxt_size; |
1329 | |
1330 | /* cxt size and block-waste are multipes of 8 */ |
1331 | cdu_params = 0; |
1332 | SET_FIELD(cdu_params, CDUT_TYPE1_CXT_SIZE, (cxt_size >> 3)); |
1333 | SET_FIELD(cdu_params, CDUT_TYPE1_BLOCK_WASTE, (block_waste >> 3)); |
1334 | SET_FIELD(cdu_params, CDUT_TYPE1_NCIB, elems_per_page); |
1335 | STORE_RT_REG(p_hwfn, CDU_REG_SEGMENT1_PARAMS_RT_OFFSET, cdu_params); |
1336 | } |
1337 | |
1338 | /* CDU PF */ |
1339 | #define CDU_SEG_REG_TYPE_SHIFT CDU_SEG_TYPE_OFFSET_REG_TYPE_SHIFT |
1340 | #define CDU_SEG_REG_TYPE_MASK 0x1 |
1341 | #define CDU_SEG_REG_OFFSET_SHIFT 0 |
1342 | #define CDU_SEG_REG_OFFSET_MASK CDU_SEG_TYPE_OFFSET_REG_OFFSET_MASK |
1343 | |
1344 | static void qed_cdu_init_pf(struct qed_hwfn *p_hwfn) |
1345 | { |
1346 | struct qed_ilt_client_cfg *p_cli; |
1347 | struct qed_tid_seg *p_seg; |
1348 | u32 cdu_seg_params, offset; |
1349 | int i; |
1350 | |
1351 | static const u32 rt_type_offset_arr[] = { |
1352 | CDU_REG_PF_SEG0_TYPE_OFFSET_RT_OFFSET, |
1353 | CDU_REG_PF_SEG1_TYPE_OFFSET_RT_OFFSET, |
1354 | CDU_REG_PF_SEG2_TYPE_OFFSET_RT_OFFSET, |
1355 | CDU_REG_PF_SEG3_TYPE_OFFSET_RT_OFFSET |
1356 | }; |
1357 | |
1358 | static const u32 rt_type_offset_fl_arr[] = { |
1359 | CDU_REG_PF_FL_SEG0_TYPE_OFFSET_RT_OFFSET, |
1360 | CDU_REG_PF_FL_SEG1_TYPE_OFFSET_RT_OFFSET, |
1361 | CDU_REG_PF_FL_SEG2_TYPE_OFFSET_RT_OFFSET, |
1362 | CDU_REG_PF_FL_SEG3_TYPE_OFFSET_RT_OFFSET |
1363 | }; |
1364 | |
1365 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
1366 | |
1367 | /* There are initializations only for CDUT during pf Phase */ |
1368 | for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) { |
1369 | /* Segment 0 */ |
1370 | p_seg = qed_cxt_tid_seg_info(p_hwfn, seg: i); |
1371 | if (!p_seg) |
1372 | continue; |
1373 | |
1374 | /* Note: start_line is already adjusted for the CDU |
1375 | * segment register granularity, so we just need to |
1376 | * divide. Adjustment is implicit as we assume ILT |
1377 | * Page size is larger than 32K! |
1378 | */ |
1379 | offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) * |
1380 | (p_cli->pf_blks[CDUT_SEG_BLK(i)].start_line - |
1381 | p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES; |
1382 | |
1383 | cdu_seg_params = 0; |
1384 | SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type); |
1385 | SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset); |
1386 | STORE_RT_REG(p_hwfn, rt_type_offset_arr[i], cdu_seg_params); |
1387 | |
1388 | offset = (ILT_PAGE_IN_BYTES(p_cli->p_size.val) * |
1389 | (p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)].start_line - |
1390 | p_cli->first.val)) / CDUT_SEG_ALIGNMET_IN_BYTES; |
1391 | |
1392 | cdu_seg_params = 0; |
1393 | SET_FIELD(cdu_seg_params, CDU_SEG_REG_TYPE, p_seg->type); |
1394 | SET_FIELD(cdu_seg_params, CDU_SEG_REG_OFFSET, offset); |
1395 | STORE_RT_REG(p_hwfn, rt_type_offset_fl_arr[i], cdu_seg_params); |
1396 | } |
1397 | } |
1398 | |
1399 | void qed_qm_init_pf(struct qed_hwfn *p_hwfn, |
1400 | struct qed_ptt *p_ptt, bool is_pf_loading) |
1401 | { |
1402 | struct qed_qm_info *qm_info = &p_hwfn->qm_info; |
1403 | struct qed_qm_pf_rt_init_params params; |
1404 | struct qed_qm_iids iids; |
1405 | |
1406 | memset(&iids, 0, sizeof(iids)); |
1407 | qed_cxt_qm_iids(p_hwfn, iids: &iids); |
1408 | |
1409 | memset(¶ms, 0, sizeof(params)); |
1410 | params.port_id = p_hwfn->port_id; |
1411 | params.pf_id = p_hwfn->rel_pf_id; |
1412 | params.max_phys_tcs_per_port = qm_info->max_phys_tcs_per_port; |
1413 | params.is_pf_loading = is_pf_loading; |
1414 | params.num_pf_cids = iids.cids; |
1415 | params.num_vf_cids = iids.vf_cids; |
1416 | params.num_tids = iids.tids; |
1417 | params.start_pq = qm_info->start_pq; |
1418 | params.num_pf_pqs = qm_info->num_pqs - qm_info->num_vf_pqs; |
1419 | params.num_vf_pqs = qm_info->num_vf_pqs; |
1420 | params.start_vport = qm_info->start_vport; |
1421 | params.num_vports = qm_info->num_vports; |
1422 | params.pf_wfq = qm_info->pf_wfq; |
1423 | params.pf_rl = qm_info->pf_rl; |
1424 | params.pq_params = qm_info->qm_pq_params; |
1425 | params.vport_params = qm_info->qm_vport_params; |
1426 | |
1427 | qed_qm_pf_rt_init(p_hwfn, p_ptt, p_params: ¶ms); |
1428 | } |
1429 | |
1430 | /* CM PF */ |
1431 | static void qed_cm_init_pf(struct qed_hwfn *p_hwfn) |
1432 | { |
1433 | /* XCM pure-LB queue */ |
1434 | STORE_RT_REG(p_hwfn, XCM_REG_CON_PHY_Q3_RT_OFFSET, |
1435 | qed_get_cm_pq_idx(p_hwfn, PQ_FLAGS_LB)); |
1436 | } |
1437 | |
1438 | /* DQ PF */ |
1439 | static void qed_dq_init_pf(struct qed_hwfn *p_hwfn) |
1440 | { |
1441 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1442 | u32 dq_pf_max_cid = 0, dq_vf_max_cid = 0; |
1443 | |
1444 | dq_pf_max_cid += (p_mngr->conn_cfg[0].cid_count >> DQ_RANGE_SHIFT); |
1445 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_0_RT_OFFSET, dq_pf_max_cid); |
1446 | |
1447 | dq_vf_max_cid += (p_mngr->conn_cfg[0].cids_per_vf >> DQ_RANGE_SHIFT); |
1448 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_0_RT_OFFSET, dq_vf_max_cid); |
1449 | |
1450 | dq_pf_max_cid += (p_mngr->conn_cfg[1].cid_count >> DQ_RANGE_SHIFT); |
1451 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_1_RT_OFFSET, dq_pf_max_cid); |
1452 | |
1453 | dq_vf_max_cid += (p_mngr->conn_cfg[1].cids_per_vf >> DQ_RANGE_SHIFT); |
1454 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_1_RT_OFFSET, dq_vf_max_cid); |
1455 | |
1456 | dq_pf_max_cid += (p_mngr->conn_cfg[2].cid_count >> DQ_RANGE_SHIFT); |
1457 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_2_RT_OFFSET, dq_pf_max_cid); |
1458 | |
1459 | dq_vf_max_cid += (p_mngr->conn_cfg[2].cids_per_vf >> DQ_RANGE_SHIFT); |
1460 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_2_RT_OFFSET, dq_vf_max_cid); |
1461 | |
1462 | dq_pf_max_cid += (p_mngr->conn_cfg[3].cid_count >> DQ_RANGE_SHIFT); |
1463 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_3_RT_OFFSET, dq_pf_max_cid); |
1464 | |
1465 | dq_vf_max_cid += (p_mngr->conn_cfg[3].cids_per_vf >> DQ_RANGE_SHIFT); |
1466 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_3_RT_OFFSET, dq_vf_max_cid); |
1467 | |
1468 | dq_pf_max_cid += (p_mngr->conn_cfg[4].cid_count >> DQ_RANGE_SHIFT); |
1469 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_4_RT_OFFSET, dq_pf_max_cid); |
1470 | |
1471 | dq_vf_max_cid += (p_mngr->conn_cfg[4].cids_per_vf >> DQ_RANGE_SHIFT); |
1472 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_4_RT_OFFSET, dq_vf_max_cid); |
1473 | |
1474 | dq_pf_max_cid += (p_mngr->conn_cfg[5].cid_count >> DQ_RANGE_SHIFT); |
1475 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_5_RT_OFFSET, dq_pf_max_cid); |
1476 | |
1477 | dq_vf_max_cid += (p_mngr->conn_cfg[5].cids_per_vf >> DQ_RANGE_SHIFT); |
1478 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_5_RT_OFFSET, dq_vf_max_cid); |
1479 | |
1480 | /* Connection types 6 & 7 are not in use, yet they must be configured |
1481 | * as the highest possible connection. Not configuring them means the |
1482 | * defaults will be used, and with a large number of cids a bug may |
1483 | * occur, if the defaults will be smaller than dq_pf_max_cid / |
1484 | * dq_vf_max_cid. |
1485 | */ |
1486 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_6_RT_OFFSET, dq_pf_max_cid); |
1487 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_6_RT_OFFSET, dq_vf_max_cid); |
1488 | |
1489 | STORE_RT_REG(p_hwfn, DORQ_REG_PF_MAX_ICID_7_RT_OFFSET, dq_pf_max_cid); |
1490 | STORE_RT_REG(p_hwfn, DORQ_REG_VF_MAX_ICID_7_RT_OFFSET, dq_vf_max_cid); |
1491 | } |
1492 | |
1493 | static void qed_ilt_bounds_init(struct qed_hwfn *p_hwfn) |
1494 | { |
1495 | struct qed_ilt_client_cfg *ilt_clients; |
1496 | int i; |
1497 | |
1498 | ilt_clients = p_hwfn->p_cxt_mngr->clients; |
1499 | for_each_ilt_valid_client(i, ilt_clients) { |
1500 | STORE_RT_REG(p_hwfn, |
1501 | ilt_clients[i].first.reg, |
1502 | ilt_clients[i].first.val); |
1503 | STORE_RT_REG(p_hwfn, |
1504 | ilt_clients[i].last.reg, ilt_clients[i].last.val); |
1505 | STORE_RT_REG(p_hwfn, |
1506 | ilt_clients[i].p_size.reg, |
1507 | ilt_clients[i].p_size.val); |
1508 | } |
1509 | } |
1510 | |
1511 | static void qed_ilt_vf_bounds_init(struct qed_hwfn *p_hwfn) |
1512 | { |
1513 | struct qed_ilt_client_cfg *p_cli; |
1514 | u32 blk_factor; |
1515 | |
1516 | /* For simplicty we set the 'block' to be an ILT page */ |
1517 | if (p_hwfn->cdev->p_iov_info) { |
1518 | struct qed_hw_sriov_info *p_iov = p_hwfn->cdev->p_iov_info; |
1519 | |
1520 | STORE_RT_REG(p_hwfn, |
1521 | PSWRQ2_REG_VF_BASE_RT_OFFSET, |
1522 | p_iov->first_vf_in_pf); |
1523 | STORE_RT_REG(p_hwfn, |
1524 | PSWRQ2_REG_VF_LAST_ILT_RT_OFFSET, |
1525 | p_iov->first_vf_in_pf + p_iov->total_vfs); |
1526 | } |
1527 | |
1528 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC]; |
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_CDUC_BLOCKS_FACTOR_RT_OFFSET, |
1533 | blk_factor); |
1534 | STORE_RT_REG(p_hwfn, |
1535 | PSWRQ2_REG_CDUC_NUMBER_OF_PF_BLOCKS_RT_OFFSET, |
1536 | p_cli->pf_total_lines); |
1537 | STORE_RT_REG(p_hwfn, |
1538 | PSWRQ2_REG_CDUC_VF_BLOCKS_RT_OFFSET, |
1539 | p_cli->vf_total_lines); |
1540 | } |
1541 | |
1542 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
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_CDUT_BLOCKS_FACTOR_RT_OFFSET, |
1547 | blk_factor); |
1548 | STORE_RT_REG(p_hwfn, |
1549 | PSWRQ2_REG_CDUT_NUMBER_OF_PF_BLOCKS_RT_OFFSET, |
1550 | p_cli->pf_total_lines); |
1551 | STORE_RT_REG(p_hwfn, |
1552 | PSWRQ2_REG_CDUT_VF_BLOCKS_RT_OFFSET, |
1553 | p_cli->vf_total_lines); |
1554 | } |
1555 | |
1556 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TM]; |
1557 | blk_factor = ilog2(ILT_PAGE_IN_BYTES(p_cli->p_size.val) >> 10); |
1558 | if (p_cli->active) { |
1559 | STORE_RT_REG(p_hwfn, |
1560 | PSWRQ2_REG_TM_BLOCKS_FACTOR_RT_OFFSET, blk_factor); |
1561 | STORE_RT_REG(p_hwfn, |
1562 | PSWRQ2_REG_TM_NUMBER_OF_PF_BLOCKS_RT_OFFSET, |
1563 | p_cli->pf_total_lines); |
1564 | STORE_RT_REG(p_hwfn, |
1565 | PSWRQ2_REG_TM_VF_BLOCKS_RT_OFFSET, |
1566 | p_cli->vf_total_lines); |
1567 | } |
1568 | } |
1569 | |
1570 | /* ILT (PSWRQ2) PF */ |
1571 | static void qed_ilt_init_pf(struct qed_hwfn *p_hwfn) |
1572 | { |
1573 | struct qed_ilt_client_cfg *clients; |
1574 | struct qed_cxt_mngr *p_mngr; |
1575 | struct phys_mem_desc *p_shdw; |
1576 | u32 line, rt_offst, i; |
1577 | |
1578 | qed_ilt_bounds_init(p_hwfn); |
1579 | qed_ilt_vf_bounds_init(p_hwfn); |
1580 | |
1581 | p_mngr = p_hwfn->p_cxt_mngr; |
1582 | p_shdw = p_mngr->ilt_shadow; |
1583 | clients = p_hwfn->p_cxt_mngr->clients; |
1584 | |
1585 | for_each_ilt_valid_client(i, clients) { |
1586 | /** Client's 1st val and RT array are absolute, ILT shadows' |
1587 | * lines are relative. |
1588 | */ |
1589 | line = clients[i].first.val - p_mngr->pf_start_line; |
1590 | rt_offst = PSWRQ2_REG_ILT_MEMORY_RT_OFFSET + |
1591 | clients[i].first.val * ILT_ENTRY_IN_REGS; |
1592 | |
1593 | for (; line <= clients[i].last.val - p_mngr->pf_start_line; |
1594 | line++, rt_offst += ILT_ENTRY_IN_REGS) { |
1595 | u64 ilt_hw_entry = 0; |
1596 | |
1597 | /** p_virt could be NULL incase of dynamic |
1598 | * allocation |
1599 | */ |
1600 | if (p_shdw[line].virt_addr) { |
1601 | SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL); |
1602 | SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR, |
1603 | (p_shdw[line].phys_addr >> 12)); |
1604 | |
1605 | DP_VERBOSE(p_hwfn, QED_MSG_ILT, |
1606 | "Setting RT[0x%08x] from ILT[0x%08x] [Client is %d] to Physical addr: 0x%llx\n" , |
1607 | rt_offst, line, i, |
1608 | (u64)(p_shdw[line].phys_addr >> 12)); |
1609 | } |
1610 | |
1611 | STORE_RT_REG_AGG(p_hwfn, rt_offst, ilt_hw_entry); |
1612 | } |
1613 | } |
1614 | } |
1615 | |
1616 | /* SRC (Searcher) PF */ |
1617 | static void qed_src_init_pf(struct qed_hwfn *p_hwfn) |
1618 | { |
1619 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1620 | u32 rounded_conn_num, conn_num, conn_max; |
1621 | struct qed_src_iids src_iids; |
1622 | |
1623 | memset(&src_iids, 0, sizeof(src_iids)); |
1624 | qed_cxt_src_iids(p_mngr, iids: &src_iids); |
1625 | conn_num = src_iids.pf_cids + src_iids.per_vf_cids * p_mngr->vf_count; |
1626 | if (!conn_num) |
1627 | return; |
1628 | |
1629 | conn_max = max_t(u32, conn_num, SRC_MIN_NUM_ELEMS); |
1630 | rounded_conn_num = roundup_pow_of_two(conn_max); |
1631 | |
1632 | STORE_RT_REG(p_hwfn, SRC_REG_COUNTFREE_RT_OFFSET, conn_num); |
1633 | STORE_RT_REG(p_hwfn, SRC_REG_NUMBER_HASH_BITS_RT_OFFSET, |
1634 | ilog2(rounded_conn_num)); |
1635 | |
1636 | STORE_RT_REG_AGG(p_hwfn, SRC_REG_FIRSTFREE_RT_OFFSET, |
1637 | p_hwfn->p_cxt_mngr->src_t2.first_free); |
1638 | STORE_RT_REG_AGG(p_hwfn, SRC_REG_LASTFREE_RT_OFFSET, |
1639 | p_hwfn->p_cxt_mngr->src_t2.last_free); |
1640 | } |
1641 | |
1642 | /* Timers PF */ |
1643 | #define TM_CFG_NUM_IDS_SHIFT 0 |
1644 | #define TM_CFG_NUM_IDS_MASK 0xFFFFULL |
1645 | #define TM_CFG_PRE_SCAN_OFFSET_SHIFT 16 |
1646 | #define TM_CFG_PRE_SCAN_OFFSET_MASK 0x1FFULL |
1647 | #define TM_CFG_PARENT_PF_SHIFT 25 |
1648 | #define TM_CFG_PARENT_PF_MASK 0x7ULL |
1649 | |
1650 | #define TM_CFG_CID_PRE_SCAN_ROWS_SHIFT 30 |
1651 | #define TM_CFG_CID_PRE_SCAN_ROWS_MASK 0x1FFULL |
1652 | |
1653 | #define TM_CFG_TID_OFFSET_SHIFT 30 |
1654 | #define TM_CFG_TID_OFFSET_MASK 0x7FFFFULL |
1655 | #define TM_CFG_TID_PRE_SCAN_ROWS_SHIFT 49 |
1656 | #define TM_CFG_TID_PRE_SCAN_ROWS_MASK 0x1FFULL |
1657 | |
1658 | static void qed_tm_init_pf(struct qed_hwfn *p_hwfn) |
1659 | { |
1660 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1661 | u32 active_seg_mask = 0, tm_offset, rt_reg; |
1662 | struct qed_tm_iids tm_iids; |
1663 | u64 cfg_word; |
1664 | u8 i; |
1665 | |
1666 | memset(&tm_iids, 0, sizeof(tm_iids)); |
1667 | qed_cxt_tm_iids(p_hwfn, p_mngr, iids: &tm_iids); |
1668 | |
1669 | /* @@@TBD No pre-scan for now */ |
1670 | |
1671 | /* Note: We assume consecutive VFs for a PF */ |
1672 | for (i = 0; i < p_mngr->vf_count; i++) { |
1673 | cfg_word = 0; |
1674 | SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_cids); |
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_CID_PRE_SCAN_ROWS, 0); |
1678 | rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET + |
1679 | (sizeof(cfg_word) / sizeof(u32)) * |
1680 | (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i); |
1681 | STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word); |
1682 | } |
1683 | |
1684 | cfg_word = 0; |
1685 | SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_cids); |
1686 | SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0); |
1687 | SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0); /* n/a for PF */ |
1688 | SET_FIELD(cfg_word, TM_CFG_CID_PRE_SCAN_ROWS, 0); /* scan all */ |
1689 | |
1690 | rt_reg = TM_REG_CONFIG_CONN_MEM_RT_OFFSET + |
1691 | (sizeof(cfg_word) / sizeof(u32)) * |
1692 | (NUM_OF_VFS(p_hwfn->cdev) + p_hwfn->rel_pf_id); |
1693 | STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word); |
1694 | |
1695 | /* enale scan */ |
1696 | STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_CONN_RT_OFFSET, |
1697 | tm_iids.pf_cids ? 0x1 : 0x0); |
1698 | |
1699 | /* @@@TBD how to enable the scan for the VFs */ |
1700 | |
1701 | tm_offset = tm_iids.per_vf_cids; |
1702 | |
1703 | /* Note: We assume consecutive VFs for a PF */ |
1704 | for (i = 0; i < p_mngr->vf_count; i++) { |
1705 | cfg_word = 0; |
1706 | SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.per_vf_tids); |
1707 | SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0); |
1708 | SET_FIELD(cfg_word, TM_CFG_PARENT_PF, p_hwfn->rel_pf_id); |
1709 | SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset); |
1710 | SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0); |
1711 | |
1712 | rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET + |
1713 | (sizeof(cfg_word) / sizeof(u32)) * |
1714 | (p_hwfn->cdev->p_iov_info->first_vf_in_pf + i); |
1715 | |
1716 | STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word); |
1717 | } |
1718 | |
1719 | tm_offset = tm_iids.pf_cids; |
1720 | for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) { |
1721 | cfg_word = 0; |
1722 | SET_FIELD(cfg_word, TM_CFG_NUM_IDS, tm_iids.pf_tids[i]); |
1723 | SET_FIELD(cfg_word, TM_CFG_PRE_SCAN_OFFSET, 0); |
1724 | SET_FIELD(cfg_word, TM_CFG_PARENT_PF, 0); |
1725 | SET_FIELD(cfg_word, TM_CFG_TID_OFFSET, tm_offset); |
1726 | SET_FIELD(cfg_word, TM_CFG_TID_PRE_SCAN_ROWS, (u64) 0); |
1727 | |
1728 | rt_reg = TM_REG_CONFIG_TASK_MEM_RT_OFFSET + |
1729 | (sizeof(cfg_word) / sizeof(u32)) * |
1730 | (NUM_OF_VFS(p_hwfn->cdev) + |
1731 | p_hwfn->rel_pf_id * NUM_TASK_PF_SEGMENTS + i); |
1732 | |
1733 | STORE_RT_REG_AGG(p_hwfn, rt_reg, cfg_word); |
1734 | active_seg_mask |= (tm_iids.pf_tids[i] ? BIT(i) : 0); |
1735 | |
1736 | tm_offset += tm_iids.pf_tids[i]; |
1737 | } |
1738 | |
1739 | if (QED_IS_RDMA_PERSONALITY(p_hwfn)) |
1740 | active_seg_mask = 0; |
1741 | |
1742 | STORE_RT_REG(p_hwfn, TM_REG_PF_ENABLE_TASK_RT_OFFSET, active_seg_mask); |
1743 | |
1744 | /* @@@TBD how to enable the scan for the VFs */ |
1745 | } |
1746 | |
1747 | static void qed_prs_init_common(struct qed_hwfn *p_hwfn) |
1748 | { |
1749 | if ((p_hwfn->hw_info.personality == QED_PCI_FCOE) && |
1750 | p_hwfn->pf_params.fcoe_pf_params.is_target) |
1751 | STORE_RT_REG(p_hwfn, |
1752 | PRS_REG_SEARCH_RESP_INITIATOR_TYPE_RT_OFFSET, 0); |
1753 | } |
1754 | |
1755 | static void qed_prs_init_pf(struct qed_hwfn *p_hwfn) |
1756 | { |
1757 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1758 | struct qed_conn_type_cfg *p_fcoe; |
1759 | struct qed_tid_seg *p_tid; |
1760 | |
1761 | p_fcoe = &p_mngr->conn_cfg[PROTOCOLID_FCOE]; |
1762 | |
1763 | /* If FCoE is active set the MAX OX_ID (tid) in the Parser */ |
1764 | if (!p_fcoe->cid_count) |
1765 | return; |
1766 | |
1767 | p_tid = &p_fcoe->tid_seg[QED_CXT_FCOE_TID_SEG]; |
1768 | if (p_hwfn->pf_params.fcoe_pf_params.is_target) { |
1769 | STORE_RT_REG_AGG(p_hwfn, |
1770 | PRS_REG_TASK_ID_MAX_TARGET_PF_RT_OFFSET, |
1771 | p_tid->count); |
1772 | } else { |
1773 | STORE_RT_REG_AGG(p_hwfn, |
1774 | PRS_REG_TASK_ID_MAX_INITIATOR_PF_RT_OFFSET, |
1775 | p_tid->count); |
1776 | } |
1777 | } |
1778 | |
1779 | void qed_cxt_hw_init_common(struct qed_hwfn *p_hwfn) |
1780 | { |
1781 | qed_cdu_init_common(p_hwfn); |
1782 | qed_prs_init_common(p_hwfn); |
1783 | } |
1784 | |
1785 | void qed_cxt_hw_init_pf(struct qed_hwfn *p_hwfn, struct qed_ptt *p_ptt) |
1786 | { |
1787 | qed_qm_init_pf(p_hwfn, p_ptt, is_pf_loading: true); |
1788 | qed_cm_init_pf(p_hwfn); |
1789 | qed_dq_init_pf(p_hwfn); |
1790 | qed_cdu_init_pf(p_hwfn); |
1791 | qed_ilt_init_pf(p_hwfn); |
1792 | qed_src_init_pf(p_hwfn); |
1793 | qed_tm_init_pf(p_hwfn); |
1794 | qed_prs_init_pf(p_hwfn); |
1795 | } |
1796 | |
1797 | int _qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn, |
1798 | enum protocol_type type, u32 *p_cid, u8 vfid) |
1799 | { |
1800 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1801 | struct qed_cid_acquired_map *p_map; |
1802 | u32 rel_cid; |
1803 | |
1804 | if (type >= MAX_CONN_TYPES) { |
1805 | DP_NOTICE(p_hwfn, "Invalid protocol type %d" , type); |
1806 | return -EINVAL; |
1807 | } |
1808 | |
1809 | if (vfid >= MAX_NUM_VFS && vfid != QED_CXT_PF_CID) { |
1810 | DP_NOTICE(p_hwfn, "VF [%02x] is out of range\n" , vfid); |
1811 | return -EINVAL; |
1812 | } |
1813 | |
1814 | /* Determine the right map to take this CID from */ |
1815 | if (vfid == QED_CXT_PF_CID) |
1816 | p_map = &p_mngr->acquired[type]; |
1817 | else |
1818 | p_map = &p_mngr->acquired_vf[type][vfid]; |
1819 | |
1820 | if (!p_map->cid_map) { |
1821 | DP_NOTICE(p_hwfn, "Invalid protocol type %d" , type); |
1822 | return -EINVAL; |
1823 | } |
1824 | |
1825 | rel_cid = find_first_zero_bit(addr: p_map->cid_map, size: p_map->max_count); |
1826 | |
1827 | if (rel_cid >= p_map->max_count) { |
1828 | DP_NOTICE(p_hwfn, "no CID available for protocol %d\n" , type); |
1829 | return -EINVAL; |
1830 | } |
1831 | |
1832 | __set_bit(rel_cid, p_map->cid_map); |
1833 | |
1834 | *p_cid = rel_cid + p_map->start_cid; |
1835 | |
1836 | DP_VERBOSE(p_hwfn, QED_MSG_CXT, |
1837 | "Acquired cid 0x%08x [rel. %08x] vfid %02x type %d\n" , |
1838 | *p_cid, rel_cid, vfid, type); |
1839 | |
1840 | return 0; |
1841 | } |
1842 | |
1843 | int qed_cxt_acquire_cid(struct qed_hwfn *p_hwfn, |
1844 | enum protocol_type type, u32 *p_cid) |
1845 | { |
1846 | return _qed_cxt_acquire_cid(p_hwfn, type, p_cid, QED_CXT_PF_CID); |
1847 | } |
1848 | |
1849 | static bool qed_cxt_test_cid_acquired(struct qed_hwfn *p_hwfn, |
1850 | u32 cid, |
1851 | u8 vfid, |
1852 | enum protocol_type *p_type, |
1853 | struct qed_cid_acquired_map **pp_map) |
1854 | { |
1855 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1856 | u32 rel_cid; |
1857 | |
1858 | /* Iterate over protocols and find matching cid range */ |
1859 | for (*p_type = 0; *p_type < MAX_CONN_TYPES; (*p_type)++) { |
1860 | if (vfid == QED_CXT_PF_CID) |
1861 | *pp_map = &p_mngr->acquired[*p_type]; |
1862 | else |
1863 | *pp_map = &p_mngr->acquired_vf[*p_type][vfid]; |
1864 | |
1865 | if (!((*pp_map)->cid_map)) |
1866 | continue; |
1867 | if (cid >= (*pp_map)->start_cid && |
1868 | cid < (*pp_map)->start_cid + (*pp_map)->max_count) |
1869 | break; |
1870 | } |
1871 | |
1872 | if (*p_type == MAX_CONN_TYPES) { |
1873 | DP_NOTICE(p_hwfn, "Invalid CID %d vfid %02x" , cid, vfid); |
1874 | goto fail; |
1875 | } |
1876 | |
1877 | rel_cid = cid - (*pp_map)->start_cid; |
1878 | if (!test_bit(rel_cid, (*pp_map)->cid_map)) { |
1879 | DP_NOTICE(p_hwfn, "CID %d [vifd %02x] not acquired" , |
1880 | cid, vfid); |
1881 | goto fail; |
1882 | } |
1883 | |
1884 | return true; |
1885 | fail: |
1886 | *p_type = MAX_CONN_TYPES; |
1887 | *pp_map = NULL; |
1888 | return false; |
1889 | } |
1890 | |
1891 | void _qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid, u8 vfid) |
1892 | { |
1893 | struct qed_cid_acquired_map *p_map = NULL; |
1894 | enum protocol_type type; |
1895 | bool b_acquired; |
1896 | u32 rel_cid; |
1897 | |
1898 | if (vfid != QED_CXT_PF_CID && vfid > MAX_NUM_VFS) { |
1899 | DP_NOTICE(p_hwfn, |
1900 | "Trying to return incorrect CID belonging to VF %02x\n" , |
1901 | vfid); |
1902 | return; |
1903 | } |
1904 | |
1905 | /* Test acquired and find matching per-protocol map */ |
1906 | b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid, vfid, |
1907 | p_type: &type, pp_map: &p_map); |
1908 | |
1909 | if (!b_acquired) |
1910 | return; |
1911 | |
1912 | rel_cid = cid - p_map->start_cid; |
1913 | clear_bit(nr: rel_cid, addr: p_map->cid_map); |
1914 | |
1915 | DP_VERBOSE(p_hwfn, QED_MSG_CXT, |
1916 | "Released CID 0x%08x [rel. %08x] vfid %02x type %d\n" , |
1917 | cid, rel_cid, vfid, type); |
1918 | } |
1919 | |
1920 | void qed_cxt_release_cid(struct qed_hwfn *p_hwfn, u32 cid) |
1921 | { |
1922 | _qed_cxt_release_cid(p_hwfn, cid, QED_CXT_PF_CID); |
1923 | } |
1924 | |
1925 | int qed_cxt_get_cid_info(struct qed_hwfn *p_hwfn, struct qed_cxt_info *p_info) |
1926 | { |
1927 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
1928 | struct qed_cid_acquired_map *p_map = NULL; |
1929 | u32 conn_cxt_size, hw_p_size, cxts_per_p, line; |
1930 | enum protocol_type type; |
1931 | bool b_acquired; |
1932 | |
1933 | /* Test acquired and find matching per-protocol map */ |
1934 | b_acquired = qed_cxt_test_cid_acquired(p_hwfn, cid: p_info->iid, |
1935 | QED_CXT_PF_CID, p_type: &type, pp_map: &p_map); |
1936 | |
1937 | if (!b_acquired) |
1938 | return -EINVAL; |
1939 | |
1940 | /* set the protocl type */ |
1941 | p_info->type = type; |
1942 | |
1943 | /* compute context virtual pointer */ |
1944 | hw_p_size = p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC].p_size.val; |
1945 | |
1946 | conn_cxt_size = CONN_CXT_SIZE(p_hwfn); |
1947 | cxts_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / conn_cxt_size; |
1948 | line = p_info->iid / cxts_per_p; |
1949 | |
1950 | /* Make sure context is allocated (dynamic allocation) */ |
1951 | if (!p_mngr->ilt_shadow[line].virt_addr) |
1952 | return -EINVAL; |
1953 | |
1954 | p_info->p_cxt = p_mngr->ilt_shadow[line].virt_addr + |
1955 | p_info->iid % cxts_per_p * conn_cxt_size; |
1956 | |
1957 | DP_VERBOSE(p_hwfn, (QED_MSG_ILT | QED_MSG_CXT), |
1958 | "Accessing ILT shadow[%d]: CXT pointer is at %p (for iid %d)\n" , |
1959 | p_info->iid / cxts_per_p, p_info->p_cxt, p_info->iid); |
1960 | |
1961 | return 0; |
1962 | } |
1963 | |
1964 | static void qed_rdma_set_pf_params(struct qed_hwfn *p_hwfn, |
1965 | struct qed_rdma_pf_params *p_params, |
1966 | u32 num_tasks) |
1967 | { |
1968 | u32 num_cons, num_qps; |
1969 | enum protocol_type proto; |
1970 | |
1971 | if (p_hwfn->mcp_info->func_info.protocol == QED_PCI_ETH_RDMA) { |
1972 | DP_VERBOSE(p_hwfn, QED_MSG_SP, |
1973 | "Current day drivers don't support RoCE & iWARP simultaneously on the same PF. Default to RoCE-only\n" ); |
1974 | p_hwfn->hw_info.personality = QED_PCI_ETH_ROCE; |
1975 | } |
1976 | |
1977 | switch (p_hwfn->hw_info.personality) { |
1978 | case QED_PCI_ETH_IWARP: |
1979 | /* Each QP requires one connection */ |
1980 | num_cons = min_t(u32, IWARP_MAX_QPS, p_params->num_qps); |
1981 | proto = PROTOCOLID_IWARP; |
1982 | break; |
1983 | case QED_PCI_ETH_ROCE: |
1984 | num_qps = min_t(u32, ROCE_MAX_QPS, p_params->num_qps); |
1985 | num_cons = num_qps * 2; /* each QP requires two connections */ |
1986 | proto = PROTOCOLID_ROCE; |
1987 | break; |
1988 | default: |
1989 | return; |
1990 | } |
1991 | |
1992 | if (num_cons && num_tasks) { |
1993 | u32 num_srqs, num_xrc_srqs; |
1994 | |
1995 | qed_cxt_set_proto_cid_count(p_hwfn, type: proto, cid_count: num_cons, vf_cid_cnt: 0); |
1996 | |
1997 | /* Deliberatly passing ROCE for tasks id. This is because |
1998 | * iWARP / RoCE share the task id. |
1999 | */ |
2000 | qed_cxt_set_proto_tid_count(p_hwfn, proto: PROTOCOLID_ROCE, |
2001 | QED_CXT_ROCE_TID_SEG, seg_type: 1, |
2002 | count: num_tasks, has_fl: false); |
2003 | |
2004 | num_srqs = min_t(u32, QED_RDMA_MAX_SRQS, p_params->num_srqs); |
2005 | |
2006 | /* XRC SRQs populate a single ILT page */ |
2007 | num_xrc_srqs = qed_cxt_xrc_srqs_per_page(p_hwfn); |
2008 | |
2009 | qed_cxt_set_srq_count(p_hwfn, num_srqs, num_xrc_srqs); |
2010 | } else { |
2011 | DP_INFO(p_hwfn->cdev, |
2012 | "RDMA personality used without setting params!\n" ); |
2013 | } |
2014 | } |
2015 | |
2016 | int qed_cxt_set_pf_params(struct qed_hwfn *p_hwfn, u32 rdma_tasks) |
2017 | { |
2018 | /* Set the number of required CORE connections */ |
2019 | u32 core_cids = 1; /* SPQ */ |
2020 | |
2021 | if (p_hwfn->using_ll2) |
2022 | core_cids += 4; |
2023 | qed_cxt_set_proto_cid_count(p_hwfn, type: PROTOCOLID_CORE, cid_count: core_cids, vf_cid_cnt: 0); |
2024 | |
2025 | switch (p_hwfn->hw_info.personality) { |
2026 | case QED_PCI_ETH_RDMA: |
2027 | case QED_PCI_ETH_IWARP: |
2028 | case QED_PCI_ETH_ROCE: |
2029 | { |
2030 | qed_rdma_set_pf_params(p_hwfn, |
2031 | p_params: &p_hwfn-> |
2032 | pf_params.rdma_pf_params, |
2033 | num_tasks: rdma_tasks); |
2034 | /* no need for break since RoCE coexist with Ethernet */ |
2035 | } |
2036 | fallthrough; |
2037 | case QED_PCI_ETH: |
2038 | { |
2039 | struct qed_eth_pf_params *p_params = |
2040 | &p_hwfn->pf_params.eth_pf_params; |
2041 | |
2042 | if (!p_params->num_vf_cons) |
2043 | p_params->num_vf_cons = |
2044 | ETH_PF_PARAMS_VF_CONS_DEFAULT; |
2045 | qed_cxt_set_proto_cid_count(p_hwfn, type: PROTOCOLID_ETH, |
2046 | cid_count: p_params->num_cons, |
2047 | vf_cid_cnt: p_params->num_vf_cons); |
2048 | p_hwfn->p_cxt_mngr->arfs_count = p_params->num_arfs_filters; |
2049 | break; |
2050 | } |
2051 | case QED_PCI_FCOE: |
2052 | { |
2053 | struct qed_fcoe_pf_params *p_params; |
2054 | |
2055 | p_params = &p_hwfn->pf_params.fcoe_pf_params; |
2056 | |
2057 | if (p_params->num_cons && p_params->num_tasks) { |
2058 | qed_cxt_set_proto_cid_count(p_hwfn, |
2059 | type: PROTOCOLID_FCOE, |
2060 | cid_count: p_params->num_cons, |
2061 | vf_cid_cnt: 0); |
2062 | qed_cxt_set_proto_tid_count(p_hwfn, proto: PROTOCOLID_FCOE, |
2063 | QED_CXT_FCOE_TID_SEG, seg_type: 0, |
2064 | count: p_params->num_tasks, has_fl: true); |
2065 | } else { |
2066 | DP_INFO(p_hwfn->cdev, |
2067 | "Fcoe personality used without setting params!\n" ); |
2068 | } |
2069 | break; |
2070 | } |
2071 | case QED_PCI_ISCSI: |
2072 | { |
2073 | struct qed_iscsi_pf_params *p_params; |
2074 | |
2075 | p_params = &p_hwfn->pf_params.iscsi_pf_params; |
2076 | |
2077 | if (p_params->num_cons && p_params->num_tasks) { |
2078 | qed_cxt_set_proto_cid_count(p_hwfn, |
2079 | type: PROTOCOLID_TCP_ULP, |
2080 | cid_count: p_params->num_cons, |
2081 | vf_cid_cnt: 0); |
2082 | qed_cxt_set_proto_tid_count(p_hwfn, |
2083 | proto: PROTOCOLID_TCP_ULP, |
2084 | QED_CXT_TCP_ULP_TID_SEG, |
2085 | seg_type: 0, |
2086 | count: p_params->num_tasks, |
2087 | has_fl: true); |
2088 | } else { |
2089 | DP_INFO(p_hwfn->cdev, |
2090 | "Iscsi personality used without setting params!\n" ); |
2091 | } |
2092 | break; |
2093 | } |
2094 | case QED_PCI_NVMETCP: |
2095 | { |
2096 | struct qed_nvmetcp_pf_params *p_params; |
2097 | |
2098 | p_params = &p_hwfn->pf_params.nvmetcp_pf_params; |
2099 | |
2100 | if (p_params->num_cons && p_params->num_tasks) { |
2101 | qed_cxt_set_proto_cid_count(p_hwfn, |
2102 | type: PROTOCOLID_TCP_ULP, |
2103 | cid_count: p_params->num_cons, |
2104 | vf_cid_cnt: 0); |
2105 | qed_cxt_set_proto_tid_count(p_hwfn, |
2106 | proto: PROTOCOLID_TCP_ULP, |
2107 | QED_CXT_TCP_ULP_TID_SEG, |
2108 | seg_type: 0, |
2109 | count: p_params->num_tasks, |
2110 | has_fl: true); |
2111 | } else { |
2112 | DP_INFO(p_hwfn->cdev, |
2113 | "NvmeTCP personality used without setting params!\n" ); |
2114 | } |
2115 | break; |
2116 | } |
2117 | default: |
2118 | return -EINVAL; |
2119 | } |
2120 | |
2121 | return 0; |
2122 | } |
2123 | |
2124 | int qed_cxt_get_tid_mem_info(struct qed_hwfn *p_hwfn, |
2125 | struct qed_tid_mem *p_info) |
2126 | { |
2127 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
2128 | u32 proto, seg, total_lines, i, shadow_line; |
2129 | struct qed_ilt_client_cfg *p_cli; |
2130 | struct qed_ilt_cli_blk *p_fl_seg; |
2131 | struct qed_tid_seg *p_seg_info; |
2132 | |
2133 | /* Verify the personality */ |
2134 | switch (p_hwfn->hw_info.personality) { |
2135 | case QED_PCI_FCOE: |
2136 | proto = PROTOCOLID_FCOE; |
2137 | seg = QED_CXT_FCOE_TID_SEG; |
2138 | break; |
2139 | case QED_PCI_ISCSI: |
2140 | case QED_PCI_NVMETCP: |
2141 | proto = PROTOCOLID_TCP_ULP; |
2142 | seg = QED_CXT_TCP_ULP_TID_SEG; |
2143 | break; |
2144 | default: |
2145 | return -EINVAL; |
2146 | } |
2147 | |
2148 | p_cli = &p_mngr->clients[ILT_CLI_CDUT]; |
2149 | if (!p_cli->active) |
2150 | return -EINVAL; |
2151 | |
2152 | p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg]; |
2153 | if (!p_seg_info->has_fl_mem) |
2154 | return -EINVAL; |
2155 | |
2156 | p_fl_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)]; |
2157 | total_lines = DIV_ROUND_UP(p_fl_seg->total_size, |
2158 | p_fl_seg->real_size_in_page); |
2159 | |
2160 | for (i = 0; i < total_lines; i++) { |
2161 | shadow_line = i + p_fl_seg->start_line - |
2162 | p_hwfn->p_cxt_mngr->pf_start_line; |
2163 | p_info->blocks[i] = p_mngr->ilt_shadow[shadow_line].virt_addr; |
2164 | } |
2165 | p_info->waste = ILT_PAGE_IN_BYTES(p_cli->p_size.val) - |
2166 | p_fl_seg->real_size_in_page; |
2167 | p_info->tid_size = p_mngr->task_type_size[p_seg_info->type]; |
2168 | p_info->num_tids_per_block = p_fl_seg->real_size_in_page / |
2169 | p_info->tid_size; |
2170 | |
2171 | return 0; |
2172 | } |
2173 | |
2174 | /* This function is very RoCE oriented, if another protocol in the future |
2175 | * will want this feature we'll need to modify the function to be more generic |
2176 | */ |
2177 | int |
2178 | qed_cxt_dynamic_ilt_alloc(struct qed_hwfn *p_hwfn, |
2179 | enum qed_cxt_elem_type elem_type, u32 iid) |
2180 | { |
2181 | u32 reg_offset, shadow_line, elem_size, hw_p_size, elems_per_p, line; |
2182 | struct tdif_task_context *tdif_context; |
2183 | struct qed_ilt_client_cfg *p_cli; |
2184 | struct qed_ilt_cli_blk *p_blk; |
2185 | struct qed_ptt *p_ptt; |
2186 | dma_addr_t p_phys; |
2187 | u64 ilt_hw_entry; |
2188 | void *p_virt; |
2189 | u32 flags1; |
2190 | int rc = 0; |
2191 | |
2192 | switch (elem_type) { |
2193 | case QED_ELEM_CXT: |
2194 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC]; |
2195 | elem_size = CONN_CXT_SIZE(p_hwfn); |
2196 | p_blk = &p_cli->pf_blks[CDUC_BLK]; |
2197 | break; |
2198 | case QED_ELEM_SRQ: |
2199 | /* The first ILT page is not used for regular SRQs. Skip it. */ |
2200 | iid += p_hwfn->p_cxt_mngr->xrc_srq_count; |
2201 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM]; |
2202 | elem_size = SRQ_CXT_SIZE; |
2203 | p_blk = &p_cli->pf_blks[SRQ_BLK]; |
2204 | break; |
2205 | case QED_ELEM_XRC_SRQ: |
2206 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM]; |
2207 | elem_size = XRC_SRQ_CXT_SIZE; |
2208 | p_blk = &p_cli->pf_blks[SRQ_BLK]; |
2209 | break; |
2210 | case QED_ELEM_TASK: |
2211 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
2212 | elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn); |
2213 | p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)]; |
2214 | break; |
2215 | default: |
2216 | DP_NOTICE(p_hwfn, "-EOPNOTSUPP elem type = %d" , elem_type); |
2217 | return -EOPNOTSUPP; |
2218 | } |
2219 | |
2220 | /* Calculate line in ilt */ |
2221 | hw_p_size = p_cli->p_size.val; |
2222 | elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size; |
2223 | line = p_blk->start_line + (iid / elems_per_p); |
2224 | shadow_line = line - p_hwfn->p_cxt_mngr->pf_start_line; |
2225 | |
2226 | /* If line is already allocated, do nothing, otherwise allocate it and |
2227 | * write it to the PSWRQ2 registers. |
2228 | * This section can be run in parallel from different contexts and thus |
2229 | * a mutex protection is needed. |
2230 | */ |
2231 | |
2232 | mutex_lock(&p_hwfn->p_cxt_mngr->mutex); |
2233 | |
2234 | if (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].virt_addr) |
2235 | goto out0; |
2236 | |
2237 | p_ptt = qed_ptt_acquire(p_hwfn); |
2238 | if (!p_ptt) { |
2239 | DP_NOTICE(p_hwfn, |
2240 | "QED_TIME_OUT on ptt acquire - dynamic allocation" ); |
2241 | rc = -EBUSY; |
2242 | goto out0; |
2243 | } |
2244 | |
2245 | p_virt = dma_alloc_coherent(dev: &p_hwfn->cdev->pdev->dev, |
2246 | size: p_blk->real_size_in_page, dma_handle: &p_phys, |
2247 | GFP_KERNEL); |
2248 | if (!p_virt) { |
2249 | rc = -ENOMEM; |
2250 | goto out1; |
2251 | } |
2252 | |
2253 | /* configuration of refTagMask to 0xF is required for RoCE DIF MR only, |
2254 | * to compensate for a HW bug, but it is configured even if DIF is not |
2255 | * enabled. This is harmless and allows us to avoid a dedicated API. We |
2256 | * configure the field for all of the contexts on the newly allocated |
2257 | * page. |
2258 | */ |
2259 | if (elem_type == QED_ELEM_TASK) { |
2260 | u32 elem_i; |
2261 | u8 *elem_start = (u8 *)p_virt; |
2262 | union type1_task_context *elem; |
2263 | |
2264 | for (elem_i = 0; elem_i < elems_per_p; elem_i++) { |
2265 | elem = (union type1_task_context *)elem_start; |
2266 | tdif_context = &elem->roce_ctx.tdif_context; |
2267 | |
2268 | flags1 = le32_to_cpu(tdif_context->flags1); |
2269 | SET_FIELD(flags1, TDIF_TASK_CONTEXT_REF_TAG_MASK, 0xf); |
2270 | tdif_context->flags1 = cpu_to_le32(flags1); |
2271 | |
2272 | elem_start += TYPE1_TASK_CXT_SIZE(p_hwfn); |
2273 | } |
2274 | } |
2275 | |
2276 | p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].virt_addr = p_virt; |
2277 | p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].phys_addr = p_phys; |
2278 | p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].size = |
2279 | p_blk->real_size_in_page; |
2280 | |
2281 | /* compute absolute offset */ |
2282 | reg_offset = PSWRQ2_REG_ILT_MEMORY + |
2283 | (line * ILT_REG_SIZE_IN_BYTES * ILT_ENTRY_IN_REGS); |
2284 | |
2285 | ilt_hw_entry = 0; |
2286 | SET_FIELD(ilt_hw_entry, ILT_ENTRY_VALID, 1ULL); |
2287 | SET_FIELD(ilt_hw_entry, ILT_ENTRY_PHY_ADDR, |
2288 | (p_hwfn->p_cxt_mngr->ilt_shadow[shadow_line].phys_addr |
2289 | >> 12)); |
2290 | |
2291 | /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a wide-bus */ |
2292 | qed_dmae_host2grc(p_hwfn, p_ptt, source_addr: (u64) (uintptr_t)&ilt_hw_entry, |
2293 | grc_addr: reg_offset, size_in_dwords: sizeof(ilt_hw_entry) / sizeof(u32), |
2294 | NULL); |
2295 | |
2296 | if (elem_type == QED_ELEM_CXT) { |
2297 | u32 last_cid_allocated = (1 + (iid / elems_per_p)) * |
2298 | elems_per_p; |
2299 | |
2300 | /* Update the relevant register in the parser */ |
2301 | qed_wr(p_hwfn, p_ptt, PRS_REG_ROCE_DEST_QP_MAX_PF, |
2302 | val: last_cid_allocated - 1); |
2303 | |
2304 | if (!p_hwfn->b_rdma_enabled_in_prs) { |
2305 | /* Enable RDMA search */ |
2306 | qed_wr(p_hwfn, p_ptt, hw_addr: p_hwfn->rdma_prs_search_reg, val: 1); |
2307 | p_hwfn->b_rdma_enabled_in_prs = true; |
2308 | } |
2309 | } |
2310 | |
2311 | out1: |
2312 | qed_ptt_release(p_hwfn, p_ptt); |
2313 | out0: |
2314 | mutex_unlock(lock: &p_hwfn->p_cxt_mngr->mutex); |
2315 | |
2316 | return rc; |
2317 | } |
2318 | |
2319 | /* This function is very RoCE oriented, if another protocol in the future |
2320 | * will want this feature we'll need to modify the function to be more generic |
2321 | */ |
2322 | static int |
2323 | qed_cxt_free_ilt_range(struct qed_hwfn *p_hwfn, |
2324 | enum qed_cxt_elem_type elem_type, |
2325 | u32 start_iid, u32 count) |
2326 | { |
2327 | u32 start_line, end_line, shadow_start_line, shadow_end_line; |
2328 | u32 reg_offset, elem_size, hw_p_size, elems_per_p; |
2329 | struct qed_ilt_client_cfg *p_cli; |
2330 | struct qed_ilt_cli_blk *p_blk; |
2331 | u32 end_iid = start_iid + count; |
2332 | struct qed_ptt *p_ptt; |
2333 | u64 ilt_hw_entry = 0; |
2334 | u32 i; |
2335 | |
2336 | switch (elem_type) { |
2337 | case QED_ELEM_CXT: |
2338 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUC]; |
2339 | elem_size = CONN_CXT_SIZE(p_hwfn); |
2340 | p_blk = &p_cli->pf_blks[CDUC_BLK]; |
2341 | break; |
2342 | case QED_ELEM_SRQ: |
2343 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM]; |
2344 | elem_size = SRQ_CXT_SIZE; |
2345 | p_blk = &p_cli->pf_blks[SRQ_BLK]; |
2346 | break; |
2347 | case QED_ELEM_XRC_SRQ: |
2348 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_TSDM]; |
2349 | elem_size = XRC_SRQ_CXT_SIZE; |
2350 | p_blk = &p_cli->pf_blks[SRQ_BLK]; |
2351 | break; |
2352 | case QED_ELEM_TASK: |
2353 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
2354 | elem_size = TYPE1_TASK_CXT_SIZE(p_hwfn); |
2355 | p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(QED_CXT_ROCE_TID_SEG)]; |
2356 | break; |
2357 | default: |
2358 | DP_NOTICE(p_hwfn, "-EINVALID elem type = %d" , elem_type); |
2359 | return -EINVAL; |
2360 | } |
2361 | |
2362 | /* Calculate line in ilt */ |
2363 | hw_p_size = p_cli->p_size.val; |
2364 | elems_per_p = ILT_PAGE_IN_BYTES(hw_p_size) / elem_size; |
2365 | start_line = p_blk->start_line + (start_iid / elems_per_p); |
2366 | end_line = p_blk->start_line + (end_iid / elems_per_p); |
2367 | if (((end_iid + 1) / elems_per_p) != (end_iid / elems_per_p)) |
2368 | end_line--; |
2369 | |
2370 | shadow_start_line = start_line - p_hwfn->p_cxt_mngr->pf_start_line; |
2371 | shadow_end_line = end_line - p_hwfn->p_cxt_mngr->pf_start_line; |
2372 | |
2373 | p_ptt = qed_ptt_acquire(p_hwfn); |
2374 | if (!p_ptt) { |
2375 | DP_NOTICE(p_hwfn, |
2376 | "QED_TIME_OUT on ptt acquire - dynamic allocation" ); |
2377 | return -EBUSY; |
2378 | } |
2379 | |
2380 | for (i = shadow_start_line; i < shadow_end_line; i++) { |
2381 | if (!p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr) |
2382 | continue; |
2383 | |
2384 | dma_free_coherent(dev: &p_hwfn->cdev->pdev->dev, |
2385 | size: p_hwfn->p_cxt_mngr->ilt_shadow[i].size, |
2386 | cpu_addr: p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr, |
2387 | dma_handle: p_hwfn->p_cxt_mngr->ilt_shadow[i].phys_addr); |
2388 | |
2389 | p_hwfn->p_cxt_mngr->ilt_shadow[i].virt_addr = NULL; |
2390 | p_hwfn->p_cxt_mngr->ilt_shadow[i].phys_addr = 0; |
2391 | p_hwfn->p_cxt_mngr->ilt_shadow[i].size = 0; |
2392 | |
2393 | /* compute absolute offset */ |
2394 | reg_offset = PSWRQ2_REG_ILT_MEMORY + |
2395 | ((start_line++) * ILT_REG_SIZE_IN_BYTES * |
2396 | ILT_ENTRY_IN_REGS); |
2397 | |
2398 | /* Write via DMAE since the PSWRQ2_REG_ILT_MEMORY line is a |
2399 | * wide-bus. |
2400 | */ |
2401 | qed_dmae_host2grc(p_hwfn, p_ptt, |
2402 | source_addr: (u64) (uintptr_t) &ilt_hw_entry, |
2403 | grc_addr: reg_offset, |
2404 | size_in_dwords: sizeof(ilt_hw_entry) / sizeof(u32), |
2405 | NULL); |
2406 | } |
2407 | |
2408 | qed_ptt_release(p_hwfn, p_ptt); |
2409 | |
2410 | return 0; |
2411 | } |
2412 | |
2413 | int qed_cxt_free_proto_ilt(struct qed_hwfn *p_hwfn, enum protocol_type proto) |
2414 | { |
2415 | int rc; |
2416 | u32 cid; |
2417 | |
2418 | /* Free Connection CXT */ |
2419 | rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_CXT, |
2420 | start_iid: qed_cxt_get_proto_cid_start(p_hwfn, |
2421 | type: proto), |
2422 | count: qed_cxt_get_proto_cid_count(p_hwfn, |
2423 | type: proto, vf_cid: &cid)); |
2424 | |
2425 | if (rc) |
2426 | return rc; |
2427 | |
2428 | /* Free Task CXT ( Intentionally RoCE as task-id is shared between |
2429 | * RoCE and iWARP ) |
2430 | */ |
2431 | proto = PROTOCOLID_ROCE; |
2432 | rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_TASK, start_iid: 0, |
2433 | count: qed_cxt_get_proto_tid_count(p_hwfn, type: proto)); |
2434 | if (rc) |
2435 | return rc; |
2436 | |
2437 | /* Free TSDM CXT */ |
2438 | rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_XRC_SRQ, start_iid: 0, |
2439 | count: p_hwfn->p_cxt_mngr->xrc_srq_count); |
2440 | |
2441 | rc = qed_cxt_free_ilt_range(p_hwfn, elem_type: QED_ELEM_SRQ, |
2442 | start_iid: p_hwfn->p_cxt_mngr->xrc_srq_count, |
2443 | count: p_hwfn->p_cxt_mngr->srq_count); |
2444 | |
2445 | return rc; |
2446 | } |
2447 | |
2448 | int qed_cxt_get_task_ctx(struct qed_hwfn *p_hwfn, |
2449 | u32 tid, u8 ctx_type, void **pp_task_ctx) |
2450 | { |
2451 | struct qed_cxt_mngr *p_mngr = p_hwfn->p_cxt_mngr; |
2452 | struct qed_ilt_client_cfg *p_cli; |
2453 | struct qed_tid_seg *p_seg_info; |
2454 | struct qed_ilt_cli_blk *p_seg; |
2455 | u32 num_tids_per_block; |
2456 | u32 tid_size, ilt_idx; |
2457 | u32 total_lines; |
2458 | u32 proto, seg; |
2459 | |
2460 | /* Verify the personality */ |
2461 | switch (p_hwfn->hw_info.personality) { |
2462 | case QED_PCI_FCOE: |
2463 | proto = PROTOCOLID_FCOE; |
2464 | seg = QED_CXT_FCOE_TID_SEG; |
2465 | break; |
2466 | case QED_PCI_ISCSI: |
2467 | case QED_PCI_NVMETCP: |
2468 | proto = PROTOCOLID_TCP_ULP; |
2469 | seg = QED_CXT_TCP_ULP_TID_SEG; |
2470 | break; |
2471 | default: |
2472 | return -EINVAL; |
2473 | } |
2474 | |
2475 | p_cli = &p_mngr->clients[ILT_CLI_CDUT]; |
2476 | if (!p_cli->active) |
2477 | return -EINVAL; |
2478 | |
2479 | p_seg_info = &p_mngr->conn_cfg[proto].tid_seg[seg]; |
2480 | |
2481 | if (ctx_type == QED_CTX_WORKING_MEM) { |
2482 | p_seg = &p_cli->pf_blks[CDUT_SEG_BLK(seg)]; |
2483 | } else if (ctx_type == QED_CTX_FL_MEM) { |
2484 | if (!p_seg_info->has_fl_mem) |
2485 | return -EINVAL; |
2486 | p_seg = &p_cli->pf_blks[CDUT_FL_SEG_BLK(seg, PF)]; |
2487 | } else { |
2488 | return -EINVAL; |
2489 | } |
2490 | total_lines = DIV_ROUND_UP(p_seg->total_size, p_seg->real_size_in_page); |
2491 | tid_size = p_mngr->task_type_size[p_seg_info->type]; |
2492 | num_tids_per_block = p_seg->real_size_in_page / tid_size; |
2493 | |
2494 | if (total_lines < tid / num_tids_per_block) |
2495 | return -EINVAL; |
2496 | |
2497 | ilt_idx = tid / num_tids_per_block + p_seg->start_line - |
2498 | p_mngr->pf_start_line; |
2499 | *pp_task_ctx = (u8 *)p_mngr->ilt_shadow[ilt_idx].virt_addr + |
2500 | (tid % num_tids_per_block) * tid_size; |
2501 | |
2502 | return 0; |
2503 | } |
2504 | |
2505 | static u16 qed_blk_calculate_pages(struct qed_ilt_cli_blk *p_blk) |
2506 | { |
2507 | if (p_blk->real_size_in_page == 0) |
2508 | return 0; |
2509 | |
2510 | return DIV_ROUND_UP(p_blk->total_size, p_blk->real_size_in_page); |
2511 | } |
2512 | |
2513 | u16 qed_get_cdut_num_pf_init_pages(struct qed_hwfn *p_hwfn) |
2514 | { |
2515 | struct qed_ilt_client_cfg *p_cli; |
2516 | struct qed_ilt_cli_blk *p_blk; |
2517 | u16 i, pages = 0; |
2518 | |
2519 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
2520 | for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) { |
2521 | p_blk = &p_cli->pf_blks[CDUT_FL_SEG_BLK(i, PF)]; |
2522 | pages += qed_blk_calculate_pages(p_blk); |
2523 | } |
2524 | |
2525 | return pages; |
2526 | } |
2527 | |
2528 | u16 qed_get_cdut_num_vf_init_pages(struct qed_hwfn *p_hwfn) |
2529 | { |
2530 | struct qed_ilt_client_cfg *p_cli; |
2531 | struct qed_ilt_cli_blk *p_blk; |
2532 | u16 i, pages = 0; |
2533 | |
2534 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
2535 | for (i = 0; i < NUM_TASK_VF_SEGMENTS; i++) { |
2536 | p_blk = &p_cli->vf_blks[CDUT_FL_SEG_BLK(i, VF)]; |
2537 | pages += qed_blk_calculate_pages(p_blk); |
2538 | } |
2539 | |
2540 | return pages; |
2541 | } |
2542 | |
2543 | u16 qed_get_cdut_num_pf_work_pages(struct qed_hwfn *p_hwfn) |
2544 | { |
2545 | struct qed_ilt_client_cfg *p_cli; |
2546 | struct qed_ilt_cli_blk *p_blk; |
2547 | u16 i, pages = 0; |
2548 | |
2549 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
2550 | for (i = 0; i < NUM_TASK_PF_SEGMENTS; i++) { |
2551 | p_blk = &p_cli->pf_blks[CDUT_SEG_BLK(i)]; |
2552 | pages += qed_blk_calculate_pages(p_blk); |
2553 | } |
2554 | |
2555 | return pages; |
2556 | } |
2557 | |
2558 | u16 qed_get_cdut_num_vf_work_pages(struct qed_hwfn *p_hwfn) |
2559 | { |
2560 | struct qed_ilt_client_cfg *p_cli; |
2561 | struct qed_ilt_cli_blk *p_blk; |
2562 | u16 pages = 0, i; |
2563 | |
2564 | p_cli = &p_hwfn->p_cxt_mngr->clients[ILT_CLI_CDUT]; |
2565 | for (i = 0; i < NUM_TASK_VF_SEGMENTS; i++) { |
2566 | p_blk = &p_cli->vf_blks[CDUT_SEG_BLK(i)]; |
2567 | pages += qed_blk_calculate_pages(p_blk); |
2568 | } |
2569 | |
2570 | return pages; |
2571 | } |
2572 | |