Commit | Line | Data |
---|---|---|
9a443537 | 1 | /* |
2 | * Copyright (c) 2016 Hisilicon Limited. | |
3 | * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved. | |
4 | * | |
5 | * This software is available to you under a choice of one of two | |
6 | * licenses. You may choose to be licensed under the terms of the GNU | |
7 | * General Public License (GPL) Version 2, available from the file | |
8 | * COPYING in the main directory of this source tree, or the | |
9 | * OpenIB.org BSD license below: | |
10 | * | |
11 | * Redistribution and use in source and binary forms, with or | |
12 | * without modification, are permitted provided that the following | |
13 | * conditions are met: | |
14 | * | |
15 | * - Redistributions of source code must retain the above | |
16 | * copyright notice, this list of conditions and the following | |
17 | * disclaimer. | |
18 | * | |
19 | * - Redistributions in binary form must reproduce the above | |
20 | * copyright notice, this list of conditions and the following | |
21 | * disclaimer in the documentation and/or other materials | |
22 | * provided with the distribution. | |
23 | * | |
24 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
25 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
26 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND | |
27 | * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS | |
28 | * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN | |
29 | * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN | |
30 | * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE | |
31 | * SOFTWARE. | |
32 | */ | |
33 | ||
9a443537 | 34 | #include "hns_roce_device.h" |
35 | #include "hns_roce_hem.h" | |
36 | #include "hns_roce_common.h" | |
37 | ||
2f49de21 XW |
38 | #define HEM_INDEX_BUF BIT(0) |
39 | #define HEM_INDEX_L0 BIT(1) | |
40 | #define HEM_INDEX_L1 BIT(2) | |
41 | struct hns_roce_hem_index { | |
42 | u64 buf; | |
43 | u64 l0; | |
44 | u64 l1; | |
45 | u32 inited; /* indicate which index is available */ | |
46 | }; | |
47 | ||
a25d13cb SX |
48 | bool hns_roce_check_whether_mhop(struct hns_roce_dev *hr_dev, u32 type) |
49 | { | |
d7019c0f LO |
50 | int hop_num = 0; |
51 | ||
52 | switch (type) { | |
53 | case HEM_TYPE_QPC: | |
54 | hop_num = hr_dev->caps.qpc_hop_num; | |
55 | break; | |
56 | case HEM_TYPE_MTPT: | |
57 | hop_num = hr_dev->caps.mpt_hop_num; | |
58 | break; | |
59 | case HEM_TYPE_CQC: | |
60 | hop_num = hr_dev->caps.cqc_hop_num; | |
61 | break; | |
62 | case HEM_TYPE_SRQC: | |
63 | hop_num = hr_dev->caps.srqc_hop_num; | |
64 | break; | |
65 | case HEM_TYPE_SCCC: | |
66 | hop_num = hr_dev->caps.sccc_hop_num; | |
67 | break; | |
68 | case HEM_TYPE_QPC_TIMER: | |
69 | hop_num = hr_dev->caps.qpc_timer_hop_num; | |
70 | break; | |
71 | case HEM_TYPE_CQC_TIMER: | |
72 | hop_num = hr_dev->caps.cqc_timer_hop_num; | |
73 | break; | |
d6d91e46 WL |
74 | case HEM_TYPE_GMV: |
75 | hop_num = hr_dev->caps.gmv_hop_num; | |
76 | break; | |
d7019c0f LO |
77 | default: |
78 | return false; | |
79 | } | |
80 | ||
272bba19 | 81 | return hop_num; |
a25d13cb | 82 | } |
a25d13cb | 83 | |
38dcb350 XW |
84 | static bool hns_roce_check_hem_null(struct hns_roce_hem **hem, u64 hem_idx, |
85 | u32 bt_chunk_num, u64 hem_max_num) | |
a25d13cb | 86 | { |
38dcb350 | 87 | u64 start_idx = round_down(hem_idx, bt_chunk_num); |
9bba3f0c XW |
88 | u64 check_max_num = start_idx + bt_chunk_num; |
89 | u64 i; | |
a25d13cb | 90 | |
9bba3f0c | 91 | for (i = start_idx; (i < check_max_num) && (i < hem_max_num); i++) |
38dcb350 | 92 | if (i != hem_idx && hem[i]) |
a25d13cb SX |
93 | return false; |
94 | ||
95 | return true; | |
96 | } | |
97 | ||
38dcb350 | 98 | static bool hns_roce_check_bt_null(u64 **bt, u64 ba_idx, u32 bt_chunk_num) |
a25d13cb | 99 | { |
38dcb350 | 100 | u64 start_idx = round_down(ba_idx, bt_chunk_num); |
a25d13cb SX |
101 | int i; |
102 | ||
103 | for (i = 0; i < bt_chunk_num; i++) | |
38dcb350 | 104 | if (i != ba_idx && bt[start_idx + i]) |
a25d13cb SX |
105 | return false; |
106 | ||
107 | return true; | |
108 | } | |
109 | ||
110 | static int hns_roce_get_bt_num(u32 table_type, u32 hop_num) | |
111 | { | |
112 | if (check_whether_bt_num_3(table_type, hop_num)) | |
113 | return 3; | |
114 | else if (check_whether_bt_num_2(table_type, hop_num)) | |
115 | return 2; | |
116 | else if (check_whether_bt_num_1(table_type, hop_num)) | |
117 | return 1; | |
118 | else | |
119 | return 0; | |
120 | } | |
121 | ||
d7019c0f LO |
122 | static int get_hem_table_config(struct hns_roce_dev *hr_dev, |
123 | struct hns_roce_hem_mhop *mhop, | |
124 | u32 type) | |
a25d13cb SX |
125 | { |
126 | struct device *dev = hr_dev->dev; | |
a25d13cb | 127 | |
d7019c0f | 128 | switch (type) { |
a25d13cb SX |
129 | case HEM_TYPE_QPC: |
130 | mhop->buf_chunk_size = 1 << (hr_dev->caps.qpc_buf_pg_sz | |
131 | + PAGE_SHIFT); | |
132 | mhop->bt_chunk_size = 1 << (hr_dev->caps.qpc_ba_pg_sz | |
133 | + PAGE_SHIFT); | |
134 | mhop->ba_l0_num = hr_dev->caps.qpc_bt_num; | |
135 | mhop->hop_num = hr_dev->caps.qpc_hop_num; | |
136 | break; | |
137 | case HEM_TYPE_MTPT: | |
138 | mhop->buf_chunk_size = 1 << (hr_dev->caps.mpt_buf_pg_sz | |
139 | + PAGE_SHIFT); | |
140 | mhop->bt_chunk_size = 1 << (hr_dev->caps.mpt_ba_pg_sz | |
141 | + PAGE_SHIFT); | |
142 | mhop->ba_l0_num = hr_dev->caps.mpt_bt_num; | |
143 | mhop->hop_num = hr_dev->caps.mpt_hop_num; | |
144 | break; | |
145 | case HEM_TYPE_CQC: | |
146 | mhop->buf_chunk_size = 1 << (hr_dev->caps.cqc_buf_pg_sz | |
147 | + PAGE_SHIFT); | |
148 | mhop->bt_chunk_size = 1 << (hr_dev->caps.cqc_ba_pg_sz | |
149 | + PAGE_SHIFT); | |
150 | mhop->ba_l0_num = hr_dev->caps.cqc_bt_num; | |
151 | mhop->hop_num = hr_dev->caps.cqc_hop_num; | |
152 | break; | |
6a157f7d YL |
153 | case HEM_TYPE_SCCC: |
154 | mhop->buf_chunk_size = 1 << (hr_dev->caps.sccc_buf_pg_sz | |
155 | + PAGE_SHIFT); | |
156 | mhop->bt_chunk_size = 1 << (hr_dev->caps.sccc_ba_pg_sz | |
157 | + PAGE_SHIFT); | |
158 | mhop->ba_l0_num = hr_dev->caps.sccc_bt_num; | |
159 | mhop->hop_num = hr_dev->caps.sccc_hop_num; | |
160 | break; | |
0e40dc2f YL |
161 | case HEM_TYPE_QPC_TIMER: |
162 | mhop->buf_chunk_size = 1 << (hr_dev->caps.qpc_timer_buf_pg_sz | |
163 | + PAGE_SHIFT); | |
164 | mhop->bt_chunk_size = 1 << (hr_dev->caps.qpc_timer_ba_pg_sz | |
165 | + PAGE_SHIFT); | |
166 | mhop->ba_l0_num = hr_dev->caps.qpc_timer_bt_num; | |
167 | mhop->hop_num = hr_dev->caps.qpc_timer_hop_num; | |
168 | break; | |
169 | case HEM_TYPE_CQC_TIMER: | |
170 | mhop->buf_chunk_size = 1 << (hr_dev->caps.cqc_timer_buf_pg_sz | |
171 | + PAGE_SHIFT); | |
172 | mhop->bt_chunk_size = 1 << (hr_dev->caps.cqc_timer_ba_pg_sz | |
173 | + PAGE_SHIFT); | |
174 | mhop->ba_l0_num = hr_dev->caps.cqc_timer_bt_num; | |
175 | mhop->hop_num = hr_dev->caps.cqc_timer_hop_num; | |
176 | break; | |
a25d13cb SX |
177 | case HEM_TYPE_SRQC: |
178 | mhop->buf_chunk_size = 1 << (hr_dev->caps.srqc_buf_pg_sz | |
179 | + PAGE_SHIFT); | |
180 | mhop->bt_chunk_size = 1 << (hr_dev->caps.srqc_ba_pg_sz | |
181 | + PAGE_SHIFT); | |
182 | mhop->ba_l0_num = hr_dev->caps.srqc_bt_num; | |
183 | mhop->hop_num = hr_dev->caps.srqc_hop_num; | |
184 | break; | |
d6d91e46 WL |
185 | case HEM_TYPE_GMV: |
186 | mhop->buf_chunk_size = 1 << (hr_dev->caps.gmv_buf_pg_sz + | |
187 | PAGE_SHIFT); | |
188 | mhop->bt_chunk_size = 1 << (hr_dev->caps.gmv_ba_pg_sz + | |
189 | PAGE_SHIFT); | |
190 | mhop->ba_l0_num = hr_dev->caps.gmv_bt_num; | |
191 | mhop->hop_num = hr_dev->caps.gmv_hop_num; | |
192 | break; | |
a25d13cb | 193 | default: |
61918e9b | 194 | dev_err(dev, "table %u not support multi-hop addressing!\n", |
d7019c0f | 195 | type); |
a25d13cb SX |
196 | return -EINVAL; |
197 | } | |
198 | ||
d7019c0f LO |
199 | return 0; |
200 | } | |
201 | ||
202 | int hns_roce_calc_hem_mhop(struct hns_roce_dev *hr_dev, | |
203 | struct hns_roce_hem_table *table, unsigned long *obj, | |
204 | struct hns_roce_hem_mhop *mhop) | |
205 | { | |
206 | struct device *dev = hr_dev->dev; | |
207 | u32 chunk_ba_num; | |
dc93a0d9 | 208 | u32 chunk_size; |
d7019c0f LO |
209 | u32 table_idx; |
210 | u32 bt_num; | |
d7019c0f LO |
211 | |
212 | if (get_hem_table_config(hr_dev, mhop, table->type)) | |
213 | return -EINVAL; | |
214 | ||
a25d13cb SX |
215 | if (!obj) |
216 | return 0; | |
217 | ||
6a93c77a | 218 | /* |
6a157f7d | 219 | * QPC/MTPT/CQC/SRQC/SCCC alloc hem for buffer pages. |
6a93c77a SX |
220 | * MTT/CQE alloc hem for bt pages. |
221 | */ | |
a25d13cb | 222 | bt_num = hns_roce_get_bt_num(table->type, mhop->hop_num); |
2a3d923f | 223 | chunk_ba_num = mhop->bt_chunk_size / BA_BYTE_LEN; |
6a93c77a SX |
224 | chunk_size = table->type < HEM_TYPE_MTT ? mhop->buf_chunk_size : |
225 | mhop->bt_chunk_size; | |
61b460d1 | 226 | table_idx = *obj / (chunk_size / table->obj_size); |
a25d13cb SX |
227 | switch (bt_num) { |
228 | case 3: | |
229 | mhop->l2_idx = table_idx & (chunk_ba_num - 1); | |
230 | mhop->l1_idx = table_idx / chunk_ba_num & (chunk_ba_num - 1); | |
73b4e1f4 | 231 | mhop->l0_idx = (table_idx / chunk_ba_num) / chunk_ba_num; |
a25d13cb SX |
232 | break; |
233 | case 2: | |
234 | mhop->l1_idx = table_idx & (chunk_ba_num - 1); | |
235 | mhop->l0_idx = table_idx / chunk_ba_num; | |
236 | break; | |
237 | case 1: | |
238 | mhop->l0_idx = table_idx; | |
239 | break; | |
240 | default: | |
61918e9b YL |
241 | dev_err(dev, "table %u not support hop_num = %u!\n", |
242 | table->type, mhop->hop_num); | |
a25d13cb SX |
243 | return -EINVAL; |
244 | } | |
245 | if (mhop->l0_idx >= mhop->ba_l0_num) | |
246 | mhop->l0_idx %= mhop->ba_l0_num; | |
247 | ||
248 | return 0; | |
249 | } | |
a25d13cb SX |
250 | |
251 | static struct hns_roce_hem *hns_roce_alloc_hem(struct hns_roce_dev *hr_dev, | |
a25d13cb SX |
252 | unsigned long hem_alloc_size, |
253 | gfp_t gfp_mask) | |
9a443537 | 254 | { |
9a443537 | 255 | struct hns_roce_hem *hem; |
9a443537 | 256 | int order; |
257 | void *buf; | |
258 | ||
259 | WARN_ON(gfp_mask & __GFP_HIGHMEM); | |
260 | ||
c00743cb YL |
261 | order = get_order(hem_alloc_size); |
262 | if (PAGE_SIZE << order != hem_alloc_size) { | |
263 | dev_err(hr_dev->dev, "invalid hem_alloc_size: %lu!\n", | |
264 | hem_alloc_size); | |
265 | return NULL; | |
266 | } | |
267 | ||
9a443537 | 268 | hem = kmalloc(sizeof(*hem), |
269 | gfp_mask & ~(__GFP_HIGHMEM | __GFP_NOWARN)); | |
270 | if (!hem) | |
271 | return NULL; | |
272 | ||
c00743cb YL |
273 | buf = dma_alloc_coherent(hr_dev->dev, hem_alloc_size, |
274 | &hem->dma, gfp_mask); | |
275 | if (!buf) | |
276 | goto fail; | |
9a443537 | 277 | |
c00743cb YL |
278 | hem->buf = buf; |
279 | hem->size = hem_alloc_size; | |
9a443537 | 280 | |
281 | return hem; | |
282 | ||
283 | fail: | |
dc3bda6e | 284 | kfree(hem); |
9a443537 | 285 | return NULL; |
286 | } | |
287 | ||
288 | void hns_roce_free_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem *hem) | |
289 | { | |
9a443537 | 290 | if (!hem) |
291 | return; | |
292 | ||
c00743cb | 293 | dma_free_coherent(hr_dev->dev, hem->size, hem->buf, hem->dma); |
9a443537 | 294 | |
295 | kfree(hem); | |
296 | } | |
297 | ||
2f49de21 XW |
298 | static int calc_hem_config(struct hns_roce_dev *hr_dev, |
299 | struct hns_roce_hem_table *table, unsigned long obj, | |
300 | struct hns_roce_hem_mhop *mhop, | |
301 | struct hns_roce_hem_index *index) | |
a25d13cb | 302 | { |
2f49de21 XW |
303 | struct ib_device *ibdev = &hr_dev->ib_dev; |
304 | unsigned long mhop_obj = obj; | |
305 | u32 l0_idx, l1_idx, l2_idx; | |
a25d13cb | 306 | u32 chunk_ba_num; |
a25d13cb | 307 | u32 bt_num; |
a25d13cb SX |
308 | int ret; |
309 | ||
2f49de21 | 310 | ret = hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, mhop); |
a25d13cb SX |
311 | if (ret) |
312 | return ret; | |
313 | ||
2f49de21 XW |
314 | l0_idx = mhop->l0_idx; |
315 | l1_idx = mhop->l1_idx; | |
316 | l2_idx = mhop->l2_idx; | |
317 | chunk_ba_num = mhop->bt_chunk_size / BA_BYTE_LEN; | |
318 | bt_num = hns_roce_get_bt_num(table->type, mhop->hop_num); | |
a25d13cb SX |
319 | switch (bt_num) { |
320 | case 3: | |
2f49de21 XW |
321 | index->l1 = l0_idx * chunk_ba_num + l1_idx; |
322 | index->l0 = l0_idx; | |
323 | index->buf = l0_idx * chunk_ba_num * chunk_ba_num + | |
324 | l1_idx * chunk_ba_num + l2_idx; | |
a25d13cb SX |
325 | break; |
326 | case 2: | |
2f49de21 XW |
327 | index->l0 = l0_idx; |
328 | index->buf = l0_idx * chunk_ba_num + l1_idx; | |
a25d13cb SX |
329 | break; |
330 | case 1: | |
2f49de21 | 331 | index->buf = l0_idx; |
a25d13cb SX |
332 | break; |
333 | default: | |
61918e9b | 334 | ibdev_err(ibdev, "table %u not support mhop.hop_num = %u!\n", |
2f49de21 | 335 | table->type, mhop->hop_num); |
a25d13cb SX |
336 | return -EINVAL; |
337 | } | |
338 | ||
2f49de21 | 339 | if (unlikely(index->buf >= table->num_hem)) { |
61918e9b | 340 | ibdev_err(ibdev, "table %u exceed hem limt idx %llu, max %lu!\n", |
2f49de21 | 341 | table->type, index->buf, table->num_hem); |
9bba3f0c XW |
342 | return -EINVAL; |
343 | } | |
344 | ||
2f49de21 XW |
345 | return 0; |
346 | } | |
a25d13cb | 347 | |
2f49de21 XW |
348 | static void free_mhop_hem(struct hns_roce_dev *hr_dev, |
349 | struct hns_roce_hem_table *table, | |
350 | struct hns_roce_hem_mhop *mhop, | |
351 | struct hns_roce_hem_index *index) | |
352 | { | |
353 | u32 bt_size = mhop->bt_chunk_size; | |
354 | struct device *dev = hr_dev->dev; | |
355 | ||
356 | if (index->inited & HEM_INDEX_BUF) { | |
357 | hns_roce_free_hem(hr_dev, table->hem[index->buf]); | |
358 | table->hem[index->buf] = NULL; | |
359 | } | |
360 | ||
361 | if (index->inited & HEM_INDEX_L1) { | |
362 | dma_free_coherent(dev, bt_size, table->bt_l1[index->l1], | |
363 | table->bt_l1_dma_addr[index->l1]); | |
364 | table->bt_l1[index->l1] = NULL; | |
365 | } | |
366 | ||
367 | if (index->inited & HEM_INDEX_L0) { | |
368 | dma_free_coherent(dev, bt_size, table->bt_l0[index->l0], | |
369 | table->bt_l0_dma_addr[index->l0]); | |
370 | table->bt_l0[index->l0] = NULL; | |
a25d13cb | 371 | } |
2f49de21 XW |
372 | } |
373 | ||
374 | static int alloc_mhop_hem(struct hns_roce_dev *hr_dev, | |
375 | struct hns_roce_hem_table *table, | |
376 | struct hns_roce_hem_mhop *mhop, | |
377 | struct hns_roce_hem_index *index) | |
378 | { | |
379 | u32 bt_size = mhop->bt_chunk_size; | |
380 | struct device *dev = hr_dev->dev; | |
2f49de21 XW |
381 | gfp_t flag; |
382 | u64 bt_ba; | |
383 | u32 size; | |
384 | int ret; | |
a25d13cb SX |
385 | |
386 | /* alloc L1 BA's chunk */ | |
2f49de21 XW |
387 | if ((check_whether_bt_num_3(table->type, mhop->hop_num) || |
388 | check_whether_bt_num_2(table->type, mhop->hop_num)) && | |
389 | !table->bt_l0[index->l0]) { | |
390 | table->bt_l0[index->l0] = dma_alloc_coherent(dev, bt_size, | |
391 | &table->bt_l0_dma_addr[index->l0], | |
a25d13cb | 392 | GFP_KERNEL); |
2f49de21 | 393 | if (!table->bt_l0[index->l0]) { |
a25d13cb SX |
394 | ret = -ENOMEM; |
395 | goto out; | |
396 | } | |
2f49de21 | 397 | index->inited |= HEM_INDEX_L0; |
a25d13cb SX |
398 | } |
399 | ||
400 | /* alloc L2 BA's chunk */ | |
2f49de21 XW |
401 | if (check_whether_bt_num_3(table->type, mhop->hop_num) && |
402 | !table->bt_l1[index->l1]) { | |
403 | table->bt_l1[index->l1] = dma_alloc_coherent(dev, bt_size, | |
404 | &table->bt_l1_dma_addr[index->l1], | |
a25d13cb | 405 | GFP_KERNEL); |
2f49de21 | 406 | if (!table->bt_l1[index->l1]) { |
a25d13cb | 407 | ret = -ENOMEM; |
2f49de21 | 408 | goto err_alloc_hem; |
a25d13cb | 409 | } |
2f49de21 XW |
410 | index->inited |= HEM_INDEX_L1; |
411 | *(table->bt_l0[index->l0] + mhop->l1_idx) = | |
412 | table->bt_l1_dma_addr[index->l1]; | |
a25d13cb SX |
413 | } |
414 | ||
6a93c77a | 415 | /* |
6a157f7d | 416 | * alloc buffer space chunk for QPC/MTPT/CQC/SRQC/SCCC. |
6a93c77a SX |
417 | * alloc bt space chunk for MTT/CQE. |
418 | */ | |
2f49de21 | 419 | size = table->type < HEM_TYPE_MTT ? mhop->buf_chunk_size : bt_size; |
29dc0635 | 420 | flag = GFP_KERNEL | __GFP_NOWARN; |
c00743cb | 421 | table->hem[index->buf] = hns_roce_alloc_hem(hr_dev, size, flag); |
2f49de21 | 422 | if (!table->hem[index->buf]) { |
a25d13cb | 423 | ret = -ENOMEM; |
2f49de21 | 424 | goto err_alloc_hem; |
a25d13cb SX |
425 | } |
426 | ||
2f49de21 | 427 | index->inited |= HEM_INDEX_BUF; |
c00743cb YL |
428 | bt_ba = table->hem[index->buf]->dma; |
429 | ||
a25d13cb | 430 | if (table->type < HEM_TYPE_MTT) { |
2f49de21 XW |
431 | if (mhop->hop_num == 2) |
432 | *(table->bt_l1[index->l1] + mhop->l2_idx) = bt_ba; | |
433 | else if (mhop->hop_num == 1) | |
434 | *(table->bt_l0[index->l0] + mhop->l1_idx) = bt_ba; | |
435 | } else if (mhop->hop_num == 2) { | |
436 | *(table->bt_l0[index->l0] + mhop->l1_idx) = bt_ba; | |
437 | } | |
438 | ||
439 | return 0; | |
440 | err_alloc_hem: | |
441 | free_mhop_hem(hr_dev, table, mhop, index); | |
442 | out: | |
443 | return ret; | |
444 | } | |
445 | ||
446 | static int set_mhop_hem(struct hns_roce_dev *hr_dev, | |
447 | struct hns_roce_hem_table *table, unsigned long obj, | |
448 | struct hns_roce_hem_mhop *mhop, | |
449 | struct hns_roce_hem_index *index) | |
450 | { | |
451 | struct ib_device *ibdev = &hr_dev->ib_dev; | |
e50cda2b | 452 | u32 step_idx; |
d35dc58d | 453 | int ret = 0; |
2f49de21 XW |
454 | |
455 | if (index->inited & HEM_INDEX_L0) { | |
456 | ret = hr_dev->hw->set_hem(hr_dev, table, obj, 0); | |
457 | if (ret) { | |
458 | ibdev_err(ibdev, "set HEM step 0 failed!\n"); | |
459 | goto out; | |
a25d13cb | 460 | } |
2f49de21 | 461 | } |
a25d13cb | 462 | |
2f49de21 XW |
463 | if (index->inited & HEM_INDEX_L1) { |
464 | ret = hr_dev->hw->set_hem(hr_dev, table, obj, 1); | |
465 | if (ret) { | |
466 | ibdev_err(ibdev, "set HEM step 1 failed!\n"); | |
467 | goto out; | |
a25d13cb SX |
468 | } |
469 | } | |
470 | ||
2f49de21 XW |
471 | if (index->inited & HEM_INDEX_BUF) { |
472 | if (mhop->hop_num == HNS_ROCE_HOP_NUM_0) | |
473 | step_idx = 0; | |
474 | else | |
475 | step_idx = mhop->hop_num; | |
476 | ret = hr_dev->hw->set_hem(hr_dev, table, obj, step_idx); | |
477 | if (ret) | |
478 | ibdev_err(ibdev, "set HEM step last failed!\n"); | |
479 | } | |
480 | out: | |
481 | return ret; | |
482 | } | |
483 | ||
484 | static int hns_roce_table_mhop_get(struct hns_roce_dev *hr_dev, | |
485 | struct hns_roce_hem_table *table, | |
486 | unsigned long obj) | |
487 | { | |
488 | struct ib_device *ibdev = &hr_dev->ib_dev; | |
489 | struct hns_roce_hem_index index = {}; | |
490 | struct hns_roce_hem_mhop mhop = {}; | |
491 | int ret; | |
a25d13cb | 492 | |
2f49de21 XW |
493 | ret = calc_hem_config(hr_dev, table, obj, &mhop, &index); |
494 | if (ret) { | |
495 | ibdev_err(ibdev, "calc hem config failed!\n"); | |
496 | return ret; | |
a25d13cb SX |
497 | } |
498 | ||
2f49de21 XW |
499 | mutex_lock(&table->mutex); |
500 | if (table->hem[index.buf]) { | |
82eb481d | 501 | refcount_inc(&table->hem[index.buf]->refcount); |
2f49de21 | 502 | goto out; |
a25d13cb SX |
503 | } |
504 | ||
2f49de21 XW |
505 | ret = alloc_mhop_hem(hr_dev, table, &mhop, &index); |
506 | if (ret) { | |
507 | ibdev_err(ibdev, "alloc mhop hem failed!\n"); | |
508 | goto out; | |
509 | } | |
510 | ||
511 | /* set HEM base address to hardware */ | |
512 | if (table->type < HEM_TYPE_MTT) { | |
513 | ret = set_mhop_hem(hr_dev, table, obj, &mhop, &index); | |
514 | if (ret) { | |
515 | ibdev_err(ibdev, "set HEM address to HW failed!\n"); | |
516 | goto err_alloc; | |
517 | } | |
518 | } | |
519 | ||
82eb481d | 520 | refcount_set(&table->hem[index.buf]->refcount, 1); |
2f49de21 XW |
521 | goto out; |
522 | ||
523 | err_alloc: | |
524 | free_mhop_hem(hr_dev, table, &mhop, &index); | |
a25d13cb SX |
525 | out: |
526 | mutex_unlock(&table->mutex); | |
527 | return ret; | |
528 | } | |
529 | ||
9a443537 | 530 | int hns_roce_table_get(struct hns_roce_dev *hr_dev, |
531 | struct hns_roce_hem_table *table, unsigned long obj) | |
532 | { | |
13ca970e | 533 | struct device *dev = hr_dev->dev; |
9a443537 | 534 | unsigned long i; |
dc93a0d9 | 535 | int ret = 0; |
9a443537 | 536 | |
a25d13cb SX |
537 | if (hns_roce_check_whether_mhop(hr_dev, table->type)) |
538 | return hns_roce_table_mhop_get(hr_dev, table, obj); | |
539 | ||
61b460d1 | 540 | i = obj / (table->table_chunk_size / table->obj_size); |
9a443537 | 541 | |
542 | mutex_lock(&table->mutex); | |
543 | ||
544 | if (table->hem[i]) { | |
82eb481d | 545 | refcount_inc(&table->hem[i]->refcount); |
9a443537 | 546 | goto out; |
547 | } | |
548 | ||
549 | table->hem[i] = hns_roce_alloc_hem(hr_dev, | |
29a1fe5d | 550 | table->table_chunk_size, |
29dc0635 | 551 | GFP_KERNEL | __GFP_NOWARN); |
9a443537 | 552 | if (!table->hem[i]) { |
553 | ret = -ENOMEM; | |
554 | goto out; | |
555 | } | |
556 | ||
557 | /* Set HEM base address(128K/page, pa) to Hardware */ | |
cf5b608f CT |
558 | ret = hr_dev->hw->set_hem(hr_dev, table, obj, HEM_HOP_STEP_DIRECT); |
559 | if (ret) { | |
08eb3018 WHX |
560 | hns_roce_free_hem(hr_dev, table->hem[i]); |
561 | table->hem[i] = NULL; | |
cf5b608f CT |
562 | dev_err(dev, "set HEM base address to HW failed, ret = %d.\n", |
563 | ret); | |
9a443537 | 564 | goto out; |
565 | } | |
566 | ||
82eb481d | 567 | refcount_set(&table->hem[i]->refcount, 1); |
9a443537 | 568 | out: |
569 | mutex_unlock(&table->mutex); | |
570 | return ret; | |
571 | } | |
572 | ||
38dcb350 XW |
573 | static void clear_mhop_hem(struct hns_roce_dev *hr_dev, |
574 | struct hns_roce_hem_table *table, unsigned long obj, | |
575 | struct hns_roce_hem_mhop *mhop, | |
576 | struct hns_roce_hem_index *index) | |
577 | { | |
578 | struct ib_device *ibdev = &hr_dev->ib_dev; | |
579 | u32 hop_num = mhop->hop_num; | |
580 | u32 chunk_ba_num; | |
e50cda2b | 581 | u32 step_idx; |
a519a612 | 582 | int ret; |
38dcb350 XW |
583 | |
584 | index->inited = HEM_INDEX_BUF; | |
585 | chunk_ba_num = mhop->bt_chunk_size / BA_BYTE_LEN; | |
586 | if (check_whether_bt_num_2(table->type, hop_num)) { | |
587 | if (hns_roce_check_hem_null(table->hem, index->buf, | |
588 | chunk_ba_num, table->num_hem)) | |
589 | index->inited |= HEM_INDEX_L0; | |
590 | } else if (check_whether_bt_num_3(table->type, hop_num)) { | |
591 | if (hns_roce_check_hem_null(table->hem, index->buf, | |
592 | chunk_ba_num, table->num_hem)) { | |
593 | index->inited |= HEM_INDEX_L1; | |
594 | if (hns_roce_check_bt_null(table->bt_l1, index->l1, | |
595 | chunk_ba_num)) | |
596 | index->inited |= HEM_INDEX_L0; | |
597 | } | |
598 | } | |
599 | ||
600 | if (table->type < HEM_TYPE_MTT) { | |
601 | if (hop_num == HNS_ROCE_HOP_NUM_0) | |
602 | step_idx = 0; | |
603 | else | |
604 | step_idx = hop_num; | |
605 | ||
a519a612 CT |
606 | ret = hr_dev->hw->clear_hem(hr_dev, table, obj, step_idx); |
607 | if (ret) | |
608 | ibdev_warn(ibdev, "failed to clear hop%u HEM, ret = %d.\n", | |
609 | hop_num, ret); | |
610 | ||
611 | if (index->inited & HEM_INDEX_L1) { | |
612 | ret = hr_dev->hw->clear_hem(hr_dev, table, obj, 1); | |
613 | if (ret) | |
614 | ibdev_warn(ibdev, "failed to clear HEM step 1, ret = %d.\n", | |
615 | ret); | |
616 | } | |
38dcb350 | 617 | |
a519a612 CT |
618 | if (index->inited & HEM_INDEX_L0) { |
619 | ret = hr_dev->hw->clear_hem(hr_dev, table, obj, 0); | |
620 | if (ret) | |
621 | ibdev_warn(ibdev, "failed to clear HEM step 0, ret = %d.\n", | |
622 | ret); | |
623 | } | |
38dcb350 XW |
624 | } |
625 | } | |
626 | ||
281d0ccf CIK |
627 | static void hns_roce_table_mhop_put(struct hns_roce_dev *hr_dev, |
628 | struct hns_roce_hem_table *table, | |
629 | unsigned long obj, | |
630 | int check_refcount) | |
a25d13cb | 631 | { |
38dcb350 XW |
632 | struct ib_device *ibdev = &hr_dev->ib_dev; |
633 | struct hns_roce_hem_index index = {}; | |
634 | struct hns_roce_hem_mhop mhop = {}; | |
a25d13cb SX |
635 | int ret; |
636 | ||
38dcb350 XW |
637 | ret = calc_hem_config(hr_dev, table, obj, &mhop, &index); |
638 | if (ret) { | |
639 | ibdev_err(ibdev, "calc hem config failed!\n"); | |
a25d13cb SX |
640 | return; |
641 | } | |
642 | ||
82eb481d WL |
643 | if (!check_refcount) |
644 | mutex_lock(&table->mutex); | |
645 | else if (!refcount_dec_and_mutex_lock(&table->hem[index.buf]->refcount, | |
646 | &table->mutex)) | |
a25d13cb | 647 | return; |
a25d13cb | 648 | |
38dcb350 XW |
649 | clear_mhop_hem(hr_dev, table, obj, &mhop, &index); |
650 | free_mhop_hem(hr_dev, table, &mhop, &index); | |
a25d13cb SX |
651 | |
652 | mutex_unlock(&table->mutex); | |
653 | } | |
654 | ||
9a443537 | 655 | void hns_roce_table_put(struct hns_roce_dev *hr_dev, |
656 | struct hns_roce_hem_table *table, unsigned long obj) | |
657 | { | |
13ca970e | 658 | struct device *dev = hr_dev->dev; |
9a443537 | 659 | unsigned long i; |
a519a612 | 660 | int ret; |
9a443537 | 661 | |
a25d13cb SX |
662 | if (hns_roce_check_whether_mhop(hr_dev, table->type)) { |
663 | hns_roce_table_mhop_put(hr_dev, table, obj, 1); | |
664 | return; | |
665 | } | |
666 | ||
61b460d1 | 667 | i = obj / (table->table_chunk_size / table->obj_size); |
9a443537 | 668 | |
82eb481d WL |
669 | if (!refcount_dec_and_mutex_lock(&table->hem[i]->refcount, |
670 | &table->mutex)) | |
671 | return; | |
9a443537 | 672 | |
a519a612 CT |
673 | ret = hr_dev->hw->clear_hem(hr_dev, table, obj, HEM_HOP_STEP_DIRECT); |
674 | if (ret) | |
675 | dev_warn(dev, "failed to clear HEM base address, ret = %d.\n", | |
676 | ret); | |
9a443537 | 677 | |
82eb481d WL |
678 | hns_roce_free_hem(hr_dev, table->hem[i]); |
679 | table->hem[i] = NULL; | |
9a443537 | 680 | |
681 | mutex_unlock(&table->mutex); | |
682 | } | |
683 | ||
6a93c77a SX |
684 | void *hns_roce_table_find(struct hns_roce_dev *hr_dev, |
685 | struct hns_roce_hem_table *table, | |
686 | unsigned long obj, dma_addr_t *dma_handle) | |
9a443537 | 687 | { |
6a93c77a | 688 | struct hns_roce_hem_mhop mhop; |
9a443537 | 689 | struct hns_roce_hem *hem; |
6a93c77a | 690 | unsigned long mhop_obj = obj; |
0203b14c | 691 | unsigned long obj_per_chunk; |
692 | unsigned long idx_offset; | |
6a93c77a | 693 | int offset, dma_offset; |
dc93a0d9 LC |
694 | void *addr = NULL; |
695 | u32 hem_idx = 0; | |
6a93c77a | 696 | int i, j; |
9a443537 | 697 | |
9a443537 | 698 | mutex_lock(&table->mutex); |
6a93c77a SX |
699 | |
700 | if (!hns_roce_check_whether_mhop(hr_dev, table->type)) { | |
0203b14c | 701 | obj_per_chunk = table->table_chunk_size / table->obj_size; |
61b460d1 XW |
702 | hem = table->hem[obj / obj_per_chunk]; |
703 | idx_offset = obj % obj_per_chunk; | |
0203b14c | 704 | dma_offset = offset = idx_offset * table->obj_size; |
6a93c77a | 705 | } else { |
4772e03d LO |
706 | u32 seg_size = 64; /* 8 bytes per BA and 8 BA per segment */ |
707 | ||
0e20ebf8 LC |
708 | if (hns_roce_calc_hem_mhop(hr_dev, table, &mhop_obj, &mhop)) |
709 | goto out; | |
6a93c77a SX |
710 | /* mtt mhop */ |
711 | i = mhop.l0_idx; | |
712 | j = mhop.l1_idx; | |
713 | if (mhop.hop_num == 2) | |
2a3d923f | 714 | hem_idx = i * (mhop.bt_chunk_size / BA_BYTE_LEN) + j; |
6a93c77a SX |
715 | else if (mhop.hop_num == 1 || |
716 | mhop.hop_num == HNS_ROCE_HOP_NUM_0) | |
717 | hem_idx = i; | |
718 | ||
719 | hem = table->hem[hem_idx]; | |
61b460d1 | 720 | dma_offset = offset = obj * seg_size % mhop.bt_chunk_size; |
6a93c77a SX |
721 | if (mhop.hop_num == 2) |
722 | dma_offset = offset = 0; | |
723 | } | |
9a443537 | 724 | |
725 | if (!hem) | |
726 | goto out; | |
727 | ||
c00743cb YL |
728 | *dma_handle = hem->dma + dma_offset; |
729 | addr = hem->buf + offset; | |
9a443537 | 730 | |
731 | out: | |
732 | mutex_unlock(&table->mutex); | |
378efe79 | 733 | return addr; |
9a443537 | 734 | } |
735 | ||
9a443537 | 736 | int hns_roce_init_hem_table(struct hns_roce_dev *hr_dev, |
737 | struct hns_roce_hem_table *table, u32 type, | |
29dc0635 | 738 | unsigned long obj_size, unsigned long nobj) |
9a443537 | 739 | { |
740 | unsigned long obj_per_chunk; | |
741 | unsigned long num_hem; | |
742 | ||
a25d13cb | 743 | if (!hns_roce_check_whether_mhop(hr_dev, type)) { |
29a1fe5d WHX |
744 | table->table_chunk_size = hr_dev->caps.chunk_sz; |
745 | obj_per_chunk = table->table_chunk_size / obj_size; | |
61b460d1 | 746 | num_hem = DIV_ROUND_UP(nobj, obj_per_chunk); |
a25d13cb SX |
747 | |
748 | table->hem = kcalloc(num_hem, sizeof(*table->hem), GFP_KERNEL); | |
749 | if (!table->hem) | |
750 | return -ENOMEM; | |
751 | } else { | |
d7019c0f | 752 | struct hns_roce_hem_mhop mhop = {}; |
a25d13cb SX |
753 | unsigned long buf_chunk_size; |
754 | unsigned long bt_chunk_size; | |
755 | unsigned long bt_chunk_num; | |
62f3b70e | 756 | unsigned long num_bt_l0; |
a25d13cb SX |
757 | u32 hop_num; |
758 | ||
d7019c0f | 759 | if (get_hem_table_config(hr_dev, &mhop, type)) |
a25d13cb | 760 | return -EINVAL; |
d7019c0f LO |
761 | |
762 | buf_chunk_size = mhop.buf_chunk_size; | |
763 | bt_chunk_size = mhop.bt_chunk_size; | |
764 | num_bt_l0 = mhop.ba_l0_num; | |
765 | hop_num = mhop.hop_num; | |
766 | ||
a25d13cb | 767 | obj_per_chunk = buf_chunk_size / obj_size; |
61b460d1 | 768 | num_hem = DIV_ROUND_UP(nobj, obj_per_chunk); |
2a3d923f | 769 | bt_chunk_num = bt_chunk_size / BA_BYTE_LEN; |
61b460d1 | 770 | |
215a8c09 | 771 | if (type >= HEM_TYPE_MTT) |
6a93c77a | 772 | num_bt_l0 = bt_chunk_num; |
a25d13cb SX |
773 | |
774 | table->hem = kcalloc(num_hem, sizeof(*table->hem), | |
775 | GFP_KERNEL); | |
776 | if (!table->hem) | |
777 | goto err_kcalloc_hem_buf; | |
778 | ||
215a8c09 | 779 | if (check_whether_bt_num_3(type, hop_num)) { |
a25d13cb SX |
780 | unsigned long num_bt_l1; |
781 | ||
61b460d1 | 782 | num_bt_l1 = DIV_ROUND_UP(num_hem, bt_chunk_num); |
a25d13cb SX |
783 | table->bt_l1 = kcalloc(num_bt_l1, |
784 | sizeof(*table->bt_l1), | |
785 | GFP_KERNEL); | |
786 | if (!table->bt_l1) | |
787 | goto err_kcalloc_bt_l1; | |
788 | ||
789 | table->bt_l1_dma_addr = kcalloc(num_bt_l1, | |
790 | sizeof(*table->bt_l1_dma_addr), | |
791 | GFP_KERNEL); | |
792 | ||
793 | if (!table->bt_l1_dma_addr) | |
794 | goto err_kcalloc_l1_dma; | |
795 | } | |
9a443537 | 796 | |
215a8c09 | 797 | if (check_whether_bt_num_2(type, hop_num) || |
798 | check_whether_bt_num_3(type, hop_num)) { | |
a25d13cb SX |
799 | table->bt_l0 = kcalloc(num_bt_l0, sizeof(*table->bt_l0), |
800 | GFP_KERNEL); | |
801 | if (!table->bt_l0) | |
802 | goto err_kcalloc_bt_l0; | |
803 | ||
804 | table->bt_l0_dma_addr = kcalloc(num_bt_l0, | |
805 | sizeof(*table->bt_l0_dma_addr), | |
806 | GFP_KERNEL); | |
807 | if (!table->bt_l0_dma_addr) | |
808 | goto err_kcalloc_l0_dma; | |
809 | } | |
810 | } | |
9a443537 | 811 | |
812 | table->type = type; | |
813 | table->num_hem = num_hem; | |
9a443537 | 814 | table->obj_size = obj_size; |
9a443537 | 815 | mutex_init(&table->mutex); |
816 | ||
817 | return 0; | |
a25d13cb SX |
818 | |
819 | err_kcalloc_l0_dma: | |
820 | kfree(table->bt_l0); | |
821 | table->bt_l0 = NULL; | |
822 | ||
823 | err_kcalloc_bt_l0: | |
824 | kfree(table->bt_l1_dma_addr); | |
825 | table->bt_l1_dma_addr = NULL; | |
826 | ||
827 | err_kcalloc_l1_dma: | |
828 | kfree(table->bt_l1); | |
829 | table->bt_l1 = NULL; | |
830 | ||
831 | err_kcalloc_bt_l1: | |
832 | kfree(table->hem); | |
833 | table->hem = NULL; | |
834 | ||
835 | err_kcalloc_hem_buf: | |
836 | return -ENOMEM; | |
837 | } | |
838 | ||
281d0ccf CIK |
839 | static void hns_roce_cleanup_mhop_hem_table(struct hns_roce_dev *hr_dev, |
840 | struct hns_roce_hem_table *table) | |
a25d13cb SX |
841 | { |
842 | struct hns_roce_hem_mhop mhop; | |
843 | u32 buf_chunk_size; | |
a25d13cb | 844 | u64 obj; |
dc93a0d9 | 845 | int i; |
a25d13cb | 846 | |
0e20ebf8 LC |
847 | if (hns_roce_calc_hem_mhop(hr_dev, table, NULL, &mhop)) |
848 | return; | |
6a93c77a SX |
849 | buf_chunk_size = table->type < HEM_TYPE_MTT ? mhop.buf_chunk_size : |
850 | mhop.bt_chunk_size; | |
a25d13cb SX |
851 | |
852 | for (i = 0; i < table->num_hem; ++i) { | |
853 | obj = i * buf_chunk_size / table->obj_size; | |
854 | if (table->hem[i]) | |
855 | hns_roce_table_mhop_put(hr_dev, table, obj, 0); | |
856 | } | |
857 | ||
858 | kfree(table->hem); | |
859 | table->hem = NULL; | |
860 | kfree(table->bt_l1); | |
861 | table->bt_l1 = NULL; | |
862 | kfree(table->bt_l1_dma_addr); | |
863 | table->bt_l1_dma_addr = NULL; | |
864 | kfree(table->bt_l0); | |
865 | table->bt_l0 = NULL; | |
866 | kfree(table->bt_l0_dma_addr); | |
867 | table->bt_l0_dma_addr = NULL; | |
9a443537 | 868 | } |
869 | ||
870 | void hns_roce_cleanup_hem_table(struct hns_roce_dev *hr_dev, | |
871 | struct hns_roce_hem_table *table) | |
872 | { | |
13ca970e | 873 | struct device *dev = hr_dev->dev; |
9a443537 | 874 | unsigned long i; |
a519a612 CT |
875 | int obj; |
876 | int ret; | |
9a443537 | 877 | |
a25d13cb SX |
878 | if (hns_roce_check_whether_mhop(hr_dev, table->type)) { |
879 | hns_roce_cleanup_mhop_hem_table(hr_dev, table); | |
9a84848d | 880 | mutex_destroy(&table->mutex); |
a25d13cb SX |
881 | return; |
882 | } | |
883 | ||
9a443537 | 884 | for (i = 0; i < table->num_hem; ++i) |
885 | if (table->hem[i]) { | |
a519a612 CT |
886 | obj = i * table->table_chunk_size / table->obj_size; |
887 | ret = hr_dev->hw->clear_hem(hr_dev, table, obj, 0); | |
888 | if (ret) | |
889 | dev_err(dev, "clear HEM base address failed, ret = %d.\n", | |
890 | ret); | |
9a443537 | 891 | |
892 | hns_roce_free_hem(hr_dev, table->hem[i]); | |
893 | } | |
894 | ||
9a84848d | 895 | mutex_destroy(&table->mutex); |
9a443537 | 896 | kfree(table->hem); |
897 | } | |
898 | ||
899 | void hns_roce_cleanup_hem(struct hns_roce_dev *hr_dev) | |
900 | { | |
4ddeacf6 | 901 | if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) |
5c1f167a LO |
902 | hns_roce_cleanup_hem_table(hr_dev, |
903 | &hr_dev->srq_table.table); | |
9a443537 | 904 | hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table); |
0e40dc2f YL |
905 | if (hr_dev->caps.qpc_timer_entry_sz) |
906 | hns_roce_cleanup_hem_table(hr_dev, | |
907 | &hr_dev->qpc_timer_table); | |
908 | if (hr_dev->caps.cqc_timer_entry_sz) | |
909 | hns_roce_cleanup_hem_table(hr_dev, | |
910 | &hr_dev->cqc_timer_table); | |
4ddeacf6 | 911 | if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) |
6a157f7d YL |
912 | hns_roce_cleanup_hem_table(hr_dev, |
913 | &hr_dev->qp_table.sccc_table); | |
e92f2c18 | 914 | if (hr_dev->caps.trrl_entry_sz) |
915 | hns_roce_cleanup_hem_table(hr_dev, | |
916 | &hr_dev->qp_table.trrl_table); | |
d6d91e46 WL |
917 | |
918 | if (hr_dev->caps.gmv_entry_sz) | |
919 | hns_roce_cleanup_hem_table(hr_dev, &hr_dev->gmv_table); | |
920 | ||
ae25db00 | 921 | hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table); |
9a443537 | 922 | hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table); |
923 | hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table); | |
9a443537 | 924 | } |
38389eaa | 925 | |
1f704d8c | 926 | struct hns_roce_hem_item { |
38389eaa LO |
927 | struct list_head list; /* link all hems in the same bt level */ |
928 | struct list_head sibling; /* link all hems in last hop for mtt */ | |
929 | void *addr; | |
930 | dma_addr_t dma_addr; | |
931 | size_t count; /* max ba numbers */ | |
932 | int start; /* start buf offset in this hem */ | |
933 | int end; /* end buf offset in this hem */ | |
934 | }; | |
935 | ||
1f704d8c XW |
936 | /* All HEM items are linked in a tree structure */ |
937 | struct hns_roce_hem_head { | |
938 | struct list_head branch[HNS_ROCE_MAX_BT_REGION]; | |
939 | struct list_head root; | |
940 | struct list_head leaf; | |
941 | }; | |
942 | ||
943 | static struct hns_roce_hem_item * | |
944 | hem_list_alloc_item(struct hns_roce_dev *hr_dev, int start, int end, int count, | |
be1eeb66 | 945 | bool exist_bt) |
38389eaa | 946 | { |
1f704d8c | 947 | struct hns_roce_hem_item *hem; |
38389eaa LO |
948 | |
949 | hem = kzalloc(sizeof(*hem), GFP_KERNEL); | |
950 | if (!hem) | |
951 | return NULL; | |
952 | ||
953 | if (exist_bt) { | |
9ea9a53e XW |
954 | hem->addr = dma_alloc_coherent(hr_dev->dev, count * BA_BYTE_LEN, |
955 | &hem->dma_addr, GFP_KERNEL); | |
38389eaa LO |
956 | if (!hem->addr) { |
957 | kfree(hem); | |
958 | return NULL; | |
959 | } | |
960 | } | |
961 | ||
962 | hem->count = count; | |
963 | hem->start = start; | |
964 | hem->end = end; | |
965 | INIT_LIST_HEAD(&hem->list); | |
966 | INIT_LIST_HEAD(&hem->sibling); | |
967 | ||
968 | return hem; | |
969 | } | |
970 | ||
971 | static void hem_list_free_item(struct hns_roce_dev *hr_dev, | |
1f704d8c | 972 | struct hns_roce_hem_item *hem, bool exist_bt) |
38389eaa LO |
973 | { |
974 | if (exist_bt) | |
975 | dma_free_coherent(hr_dev->dev, hem->count * BA_BYTE_LEN, | |
976 | hem->addr, hem->dma_addr); | |
977 | kfree(hem); | |
978 | } | |
979 | ||
980 | static void hem_list_free_all(struct hns_roce_dev *hr_dev, | |
981 | struct list_head *head, bool exist_bt) | |
982 | { | |
1f704d8c | 983 | struct hns_roce_hem_item *hem, *temp_hem; |
38389eaa LO |
984 | |
985 | list_for_each_entry_safe(hem, temp_hem, head, list) { | |
986 | list_del(&hem->list); | |
987 | hem_list_free_item(hr_dev, hem, exist_bt); | |
988 | } | |
989 | } | |
990 | ||
f4caa864 | 991 | static void hem_list_link_bt(void *base_addr, u64 table_addr) |
38389eaa LO |
992 | { |
993 | *(u64 *)(base_addr) = table_addr; | |
994 | } | |
995 | ||
996 | /* assign L0 table address to hem from root bt */ | |
f4caa864 | 997 | static void hem_list_assign_bt(struct hns_roce_hem_item *hem, void *cpu_addr, |
38389eaa LO |
998 | u64 phy_addr) |
999 | { | |
1000 | hem->addr = cpu_addr; | |
1001 | hem->dma_addr = (dma_addr_t)phy_addr; | |
1002 | } | |
1003 | ||
1f704d8c | 1004 | static inline bool hem_list_page_is_in_range(struct hns_roce_hem_item *hem, |
38389eaa LO |
1005 | int offset) |
1006 | { | |
1007 | return (hem->start <= offset && offset <= hem->end); | |
1008 | } | |
1009 | ||
1f704d8c XW |
1010 | static struct hns_roce_hem_item *hem_list_search_item(struct list_head *ba_list, |
1011 | int page_offset) | |
38389eaa | 1012 | { |
1f704d8c XW |
1013 | struct hns_roce_hem_item *hem, *temp_hem; |
1014 | struct hns_roce_hem_item *found = NULL; | |
38389eaa LO |
1015 | |
1016 | list_for_each_entry_safe(hem, temp_hem, ba_list, list) { | |
1017 | if (hem_list_page_is_in_range(hem, page_offset)) { | |
1018 | found = hem; | |
1019 | break; | |
1020 | } | |
1021 | } | |
1022 | ||
1023 | return found; | |
1024 | } | |
1025 | ||
1026 | static bool hem_list_is_bottom_bt(int hopnum, int bt_level) | |
1027 | { | |
1028 | /* | |
1029 | * hopnum base address table levels | |
1030 | * 0 L0(buf) | |
1031 | * 1 L0 -> buf | |
1032 | * 2 L0 -> L1 -> buf | |
1033 | * 3 L0 -> L1 -> L2 -> buf | |
1034 | */ | |
1035 | return bt_level >= (hopnum ? hopnum - 1 : hopnum); | |
1036 | } | |
1037 | ||
bf194997 | 1038 | /* |
38389eaa LO |
1039 | * calc base address entries num |
1040 | * @hopnum: num of mutihop addressing | |
1041 | * @bt_level: base address table level | |
1042 | * @unit: ba entries per bt page | |
1043 | */ | |
1044 | static u32 hem_list_calc_ba_range(int hopnum, int bt_level, int unit) | |
1045 | { | |
1046 | u32 step; | |
1047 | int max; | |
1048 | int i; | |
1049 | ||
1050 | if (hopnum <= bt_level) | |
1051 | return 0; | |
1052 | /* | |
1053 | * hopnum bt_level range | |
1054 | * 1 0 unit | |
1055 | * ------------ | |
1056 | * 2 0 unit * unit | |
1057 | * 2 1 unit | |
1058 | * ------------ | |
1059 | * 3 0 unit * unit * unit | |
1060 | * 3 1 unit * unit | |
1061 | * 3 2 unit | |
1062 | */ | |
1063 | step = 1; | |
1064 | max = hopnum - bt_level; | |
1065 | for (i = 0; i < max; i++) | |
1066 | step = step * unit; | |
1067 | ||
1068 | return step; | |
1069 | } | |
1070 | ||
bf194997 | 1071 | /* |
38389eaa LO |
1072 | * calc the root ba entries which could cover all regions |
1073 | * @regions: buf region array | |
1074 | * @region_cnt: array size of @regions | |
1075 | * @unit: ba entries per bt page | |
1076 | */ | |
1077 | int hns_roce_hem_list_calc_root_ba(const struct hns_roce_buf_region *regions, | |
1078 | int region_cnt, int unit) | |
1079 | { | |
1080 | struct hns_roce_buf_region *r; | |
1081 | int total = 0; | |
1082 | int step; | |
1083 | int i; | |
1084 | ||
1085 | for (i = 0; i < region_cnt; i++) { | |
1086 | r = (struct hns_roce_buf_region *)®ions[i]; | |
1087 | if (r->hopnum > 1) { | |
1088 | step = hem_list_calc_ba_range(r->hopnum, 1, unit); | |
1089 | if (step > 0) | |
1090 | total += (r->count + step - 1) / step; | |
1091 | } else { | |
1092 | total += r->count; | |
1093 | } | |
1094 | } | |
1095 | ||
1096 | return total; | |
1097 | } | |
1098 | ||
1099 | static int hem_list_alloc_mid_bt(struct hns_roce_dev *hr_dev, | |
1100 | const struct hns_roce_buf_region *r, int unit, | |
1101 | int offset, struct list_head *mid_bt, | |
1102 | struct list_head *btm_bt) | |
1103 | { | |
1f704d8c | 1104 | struct hns_roce_hem_item *hem_ptrs[HNS_ROCE_MAX_BT_LEVEL] = { NULL }; |
38389eaa | 1105 | struct list_head temp_list[HNS_ROCE_MAX_BT_LEVEL]; |
1f704d8c | 1106 | struct hns_roce_hem_item *cur, *pre; |
38389eaa LO |
1107 | const int hopnum = r->hopnum; |
1108 | int start_aligned; | |
1109 | int distance; | |
1110 | int ret = 0; | |
1111 | int max_ofs; | |
1112 | int level; | |
1113 | u32 step; | |
1114 | int end; | |
1115 | ||
1116 | if (hopnum <= 1) | |
1117 | return 0; | |
1118 | ||
1119 | if (hopnum > HNS_ROCE_MAX_BT_LEVEL) { | |
1120 | dev_err(hr_dev->dev, "invalid hopnum %d!\n", hopnum); | |
1121 | return -EINVAL; | |
1122 | } | |
1123 | ||
1124 | if (offset < r->offset) { | |
61918e9b | 1125 | dev_err(hr_dev->dev, "invalid offset %d, min %u!\n", |
38389eaa LO |
1126 | offset, r->offset); |
1127 | return -EINVAL; | |
1128 | } | |
1129 | ||
1130 | distance = offset - r->offset; | |
1131 | max_ofs = r->offset + r->count - 1; | |
1132 | for (level = 0; level < hopnum; level++) | |
1133 | INIT_LIST_HEAD(&temp_list[level]); | |
1134 | ||
1135 | /* config L1 bt to last bt and link them to corresponding parent */ | |
1136 | for (level = 1; level < hopnum; level++) { | |
1137 | cur = hem_list_search_item(&mid_bt[level], offset); | |
1138 | if (cur) { | |
1139 | hem_ptrs[level] = cur; | |
1140 | continue; | |
1141 | } | |
1142 | ||
1143 | step = hem_list_calc_ba_range(hopnum, level, unit); | |
1144 | if (step < 1) { | |
1145 | ret = -EINVAL; | |
1146 | goto err_exit; | |
1147 | } | |
1148 | ||
1149 | start_aligned = (distance / step) * step + r->offset; | |
1150 | end = min_t(int, start_aligned + step - 1, max_ofs); | |
1151 | cur = hem_list_alloc_item(hr_dev, start_aligned, end, unit, | |
be1eeb66 | 1152 | true); |
38389eaa LO |
1153 | if (!cur) { |
1154 | ret = -ENOMEM; | |
1155 | goto err_exit; | |
1156 | } | |
1157 | hem_ptrs[level] = cur; | |
1158 | list_add(&cur->list, &temp_list[level]); | |
1159 | if (hem_list_is_bottom_bt(hopnum, level)) | |
1160 | list_add(&cur->sibling, &temp_list[0]); | |
1161 | ||
1162 | /* link bt to parent bt */ | |
1163 | if (level > 1) { | |
1164 | pre = hem_ptrs[level - 1]; | |
1165 | step = (cur->start - pre->start) / step * BA_BYTE_LEN; | |
f4caa864 | 1166 | hem_list_link_bt(pre->addr + step, cur->dma_addr); |
38389eaa LO |
1167 | } |
1168 | } | |
1169 | ||
1170 | list_splice(&temp_list[0], btm_bt); | |
1171 | for (level = 1; level < hopnum; level++) | |
1172 | list_splice(&temp_list[level], &mid_bt[level]); | |
1173 | ||
1174 | return 0; | |
1175 | ||
1176 | err_exit: | |
1177 | for (level = 1; level < hopnum; level++) | |
1178 | hem_list_free_all(hr_dev, &temp_list[level], true); | |
1179 | ||
1180 | return ret; | |
1181 | } | |
1182 | ||
1f704d8c XW |
1183 | static struct hns_roce_hem_item * |
1184 | alloc_root_hem(struct hns_roce_dev *hr_dev, int unit, int *max_ba_num, | |
1185 | const struct hns_roce_buf_region *regions, int region_cnt) | |
38389eaa | 1186 | { |
38389eaa | 1187 | const struct hns_roce_buf_region *r; |
1f704d8c | 1188 | struct hns_roce_hem_item *hem; |
11334014 | 1189 | int ba_num; |
38389eaa | 1190 | int offset; |
38389eaa | 1191 | |
11334014 XW |
1192 | ba_num = hns_roce_hem_list_calc_root_ba(regions, region_cnt, unit); |
1193 | if (ba_num < 1) | |
1f704d8c | 1194 | return ERR_PTR(-ENOMEM); |
11334014 | 1195 | |
9ea9a53e | 1196 | if (ba_num > unit) |
1f704d8c | 1197 | return ERR_PTR(-ENOBUFS); |
9ea9a53e | 1198 | |
1f704d8c | 1199 | offset = regions[0].offset; |
38389eaa LO |
1200 | /* indicate to last region */ |
1201 | r = ®ions[region_cnt - 1]; | |
1f704d8c | 1202 | hem = hem_list_alloc_item(hr_dev, offset, r->offset + r->count - 1, |
be1eeb66 | 1203 | ba_num, true); |
1f704d8c XW |
1204 | if (!hem) |
1205 | return ERR_PTR(-ENOMEM); | |
1206 | ||
1207 | *max_ba_num = ba_num; | |
1208 | ||
1209 | return hem; | |
1210 | } | |
1211 | ||
1212 | static int alloc_fake_root_bt(struct hns_roce_dev *hr_dev, void *cpu_base, | |
1213 | u64 phy_base, const struct hns_roce_buf_region *r, | |
1214 | struct list_head *branch_head, | |
1215 | struct list_head *leaf_head) | |
1216 | { | |
1217 | struct hns_roce_hem_item *hem; | |
1218 | ||
1219 | hem = hem_list_alloc_item(hr_dev, r->offset, r->offset + r->count - 1, | |
be1eeb66 | 1220 | r->count, false); |
1f704d8c | 1221 | if (!hem) |
38389eaa | 1222 | return -ENOMEM; |
38389eaa | 1223 | |
f4caa864 | 1224 | hem_list_assign_bt(hem, cpu_base, phy_base); |
1f704d8c XW |
1225 | list_add(&hem->list, branch_head); |
1226 | list_add(&hem->sibling, leaf_head); | |
38389eaa | 1227 | |
1f704d8c XW |
1228 | return r->count; |
1229 | } | |
1230 | ||
1231 | static int setup_middle_bt(struct hns_roce_dev *hr_dev, void *cpu_base, | |
1232 | int unit, const struct hns_roce_buf_region *r, | |
1233 | const struct list_head *branch_head) | |
1234 | { | |
1235 | struct hns_roce_hem_item *hem, *temp_hem; | |
1236 | int total = 0; | |
1237 | int offset; | |
1238 | int step; | |
1239 | ||
1240 | step = hem_list_calc_ba_range(r->hopnum, 1, unit); | |
1241 | if (step < 1) | |
1242 | return -EINVAL; | |
1243 | ||
1244 | /* if exist mid bt, link L1 to L0 */ | |
1245 | list_for_each_entry_safe(hem, temp_hem, branch_head, list) { | |
1246 | offset = (hem->start - r->offset) / step * BA_BYTE_LEN; | |
f4caa864 | 1247 | hem_list_link_bt(cpu_base + offset, hem->dma_addr); |
1f704d8c XW |
1248 | total++; |
1249 | } | |
1250 | ||
1251 | return total; | |
1252 | } | |
1253 | ||
1254 | static int | |
1255 | setup_root_hem(struct hns_roce_dev *hr_dev, struct hns_roce_hem_list *hem_list, | |
1256 | int unit, int max_ba_num, struct hns_roce_hem_head *head, | |
1257 | const struct hns_roce_buf_region *regions, int region_cnt) | |
1258 | { | |
1259 | const struct hns_roce_buf_region *r; | |
1260 | struct hns_roce_hem_item *root_hem; | |
1261 | void *cpu_base; | |
1262 | u64 phy_base; | |
1263 | int i, total; | |
1264 | int ret; | |
1265 | ||
1266 | root_hem = list_first_entry(&head->root, | |
1267 | struct hns_roce_hem_item, list); | |
1268 | if (!root_hem) | |
1269 | return -ENOMEM; | |
38389eaa LO |
1270 | |
1271 | total = 0; | |
1f704d8c | 1272 | for (i = 0; i < region_cnt && total < max_ba_num; i++) { |
38389eaa LO |
1273 | r = ®ions[i]; |
1274 | if (!r->count) | |
1275 | continue; | |
1276 | ||
1277 | /* all regions's mid[x][0] shared the root_bt's trunk */ | |
1278 | cpu_base = root_hem->addr + total * BA_BYTE_LEN; | |
1279 | phy_base = root_hem->dma_addr + total * BA_BYTE_LEN; | |
1280 | ||
1281 | /* if hopnum is 0 or 1, cut a new fake hem from the root bt | |
1282 | * which's address share to all regions. | |
1283 | */ | |
1f704d8c XW |
1284 | if (hem_list_is_bottom_bt(r->hopnum, 0)) |
1285 | ret = alloc_fake_root_bt(hr_dev, cpu_base, phy_base, r, | |
1286 | &head->branch[i], &head->leaf); | |
1287 | else | |
1288 | ret = setup_middle_bt(hr_dev, cpu_base, unit, r, | |
1289 | &hem_list->mid_bt[i][1]); | |
1290 | ||
1291 | if (ret < 0) | |
1292 | return ret; | |
1293 | ||
1294 | total += ret; | |
38389eaa LO |
1295 | } |
1296 | ||
1f704d8c XW |
1297 | list_splice(&head->leaf, &hem_list->btm_bt); |
1298 | list_splice(&head->root, &hem_list->root_bt); | |
38389eaa | 1299 | for (i = 0; i < region_cnt; i++) |
1f704d8c | 1300 | list_splice(&head->branch[i], &hem_list->mid_bt[i][0]); |
38389eaa LO |
1301 | |
1302 | return 0; | |
1f704d8c | 1303 | } |
38389eaa | 1304 | |
1f704d8c XW |
1305 | static int hem_list_alloc_root_bt(struct hns_roce_dev *hr_dev, |
1306 | struct hns_roce_hem_list *hem_list, int unit, | |
1307 | const struct hns_roce_buf_region *regions, | |
1308 | int region_cnt) | |
1309 | { | |
1310 | struct hns_roce_hem_item *root_hem; | |
1311 | struct hns_roce_hem_head head; | |
1312 | int max_ba_num; | |
1313 | int ret; | |
1314 | int i; | |
1315 | ||
1316 | root_hem = hem_list_search_item(&hem_list->root_bt, regions[0].offset); | |
1317 | if (root_hem) | |
1318 | return 0; | |
1319 | ||
1320 | max_ba_num = 0; | |
1321 | root_hem = alloc_root_hem(hr_dev, unit, &max_ba_num, regions, | |
1322 | region_cnt); | |
1323 | if (IS_ERR(root_hem)) | |
1324 | return PTR_ERR(root_hem); | |
1325 | ||
1326 | /* List head for storing all allocated HEM items */ | |
1327 | INIT_LIST_HEAD(&head.root); | |
1328 | INIT_LIST_HEAD(&head.leaf); | |
38389eaa | 1329 | for (i = 0; i < region_cnt; i++) |
1f704d8c | 1330 | INIT_LIST_HEAD(&head.branch[i]); |
38389eaa | 1331 | |
1f704d8c XW |
1332 | hem_list->root_ba = root_hem->dma_addr; |
1333 | list_add(&root_hem->list, &head.root); | |
1334 | ret = setup_root_hem(hr_dev, hem_list, unit, max_ba_num, &head, regions, | |
1335 | region_cnt); | |
1336 | if (ret) { | |
1337 | for (i = 0; i < region_cnt; i++) | |
1338 | hem_list_free_all(hr_dev, &head.branch[i], false); | |
1339 | ||
1340 | hem_list_free_all(hr_dev, &head.root, true); | |
1341 | } | |
38389eaa LO |
1342 | |
1343 | return ret; | |
1344 | } | |
1345 | ||
1346 | /* construct the base address table and link them by address hop config */ | |
1347 | int hns_roce_hem_list_request(struct hns_roce_dev *hr_dev, | |
1348 | struct hns_roce_hem_list *hem_list, | |
1349 | const struct hns_roce_buf_region *regions, | |
82d07a4e | 1350 | int region_cnt, unsigned int bt_pg_shift) |
38389eaa LO |
1351 | { |
1352 | const struct hns_roce_buf_region *r; | |
1353 | int ofs, end; | |
38389eaa | 1354 | int unit; |
dc93a0d9 | 1355 | int ret; |
38389eaa LO |
1356 | int i; |
1357 | ||
1358 | if (region_cnt > HNS_ROCE_MAX_BT_REGION) { | |
1359 | dev_err(hr_dev->dev, "invalid region region_cnt %d!\n", | |
1360 | region_cnt); | |
1361 | return -EINVAL; | |
1362 | } | |
1363 | ||
3c873161 | 1364 | unit = (1 << bt_pg_shift) / BA_BYTE_LEN; |
38389eaa LO |
1365 | for (i = 0; i < region_cnt; i++) { |
1366 | r = ®ions[i]; | |
1367 | if (!r->count) | |
1368 | continue; | |
1369 | ||
1370 | end = r->offset + r->count; | |
1371 | for (ofs = r->offset; ofs < end; ofs += unit) { | |
1372 | ret = hem_list_alloc_mid_bt(hr_dev, r, unit, ofs, | |
1373 | hem_list->mid_bt[i], | |
1374 | &hem_list->btm_bt); | |
1375 | if (ret) { | |
1376 | dev_err(hr_dev->dev, | |
f0588567 | 1377 | "alloc hem trunk fail ret = %d!\n", ret); |
38389eaa LO |
1378 | goto err_alloc; |
1379 | } | |
1380 | } | |
1381 | } | |
1382 | ||
1383 | ret = hem_list_alloc_root_bt(hr_dev, hem_list, unit, regions, | |
1384 | region_cnt); | |
1385 | if (ret) | |
f0588567 | 1386 | dev_err(hr_dev->dev, "alloc hem root fail ret = %d!\n", ret); |
38389eaa LO |
1387 | else |
1388 | return 0; | |
1389 | ||
1390 | err_alloc: | |
1391 | hns_roce_hem_list_release(hr_dev, hem_list); | |
1392 | ||
1393 | return ret; | |
1394 | } | |
1395 | ||
1396 | void hns_roce_hem_list_release(struct hns_roce_dev *hr_dev, | |
1397 | struct hns_roce_hem_list *hem_list) | |
1398 | { | |
1399 | int i, j; | |
1400 | ||
1401 | for (i = 0; i < HNS_ROCE_MAX_BT_REGION; i++) | |
1402 | for (j = 0; j < HNS_ROCE_MAX_BT_LEVEL; j++) | |
1403 | hem_list_free_all(hr_dev, &hem_list->mid_bt[i][j], | |
1404 | j != 0); | |
1405 | ||
1406 | hem_list_free_all(hr_dev, &hem_list->root_bt, true); | |
1407 | INIT_LIST_HEAD(&hem_list->btm_bt); | |
1408 | hem_list->root_ba = 0; | |
1409 | } | |
1410 | ||
3c873161 | 1411 | void hns_roce_hem_list_init(struct hns_roce_hem_list *hem_list) |
38389eaa LO |
1412 | { |
1413 | int i, j; | |
1414 | ||
1415 | INIT_LIST_HEAD(&hem_list->root_bt); | |
1416 | INIT_LIST_HEAD(&hem_list->btm_bt); | |
1417 | for (i = 0; i < HNS_ROCE_MAX_BT_REGION; i++) | |
1418 | for (j = 0; j < HNS_ROCE_MAX_BT_LEVEL; j++) | |
1419 | INIT_LIST_HEAD(&hem_list->mid_bt[i][j]); | |
38389eaa LO |
1420 | } |
1421 | ||
1422 | void *hns_roce_hem_list_find_mtt(struct hns_roce_dev *hr_dev, | |
1423 | struct hns_roce_hem_list *hem_list, | |
5f652387 | 1424 | int offset, int *mtt_cnt) |
38389eaa LO |
1425 | { |
1426 | struct list_head *head = &hem_list->btm_bt; | |
1f704d8c | 1427 | struct hns_roce_hem_item *hem, *temp_hem; |
38389eaa | 1428 | void *cpu_base = NULL; |
38389eaa LO |
1429 | int nr = 0; |
1430 | ||
1431 | list_for_each_entry_safe(hem, temp_hem, head, sibling) { | |
1432 | if (hem_list_page_is_in_range(hem, offset)) { | |
1433 | nr = offset - hem->start; | |
1434 | cpu_base = hem->addr + nr * BA_BYTE_LEN; | |
38389eaa LO |
1435 | nr = hem->end + 1 - offset; |
1436 | break; | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | if (mtt_cnt) | |
1441 | *mtt_cnt = nr; | |
1442 | ||
38389eaa LO |
1443 | return cpu_base; |
1444 | } |