hns_roce_srq.c 11.4 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11
// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
/*
 * Copyright (c) 2018 Hisilicon Limited.
 */

#include <rdma/ib_umem.h>
#include <rdma/hns-abi.h>
#include "hns_roce_device.h"
#include "hns_roce_cmd.h"
#include "hns_roce_hem.h"

12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34
void hns_roce_srq_event(struct hns_roce_dev *hr_dev, u32 srqn, int event_type)
{
	struct hns_roce_srq_table *srq_table = &hr_dev->srq_table;
	struct hns_roce_srq *srq;

	xa_lock(&srq_table->xa);
	srq = xa_load(&srq_table->xa, srqn & (hr_dev->caps.num_srqs - 1));
	if (srq)
		atomic_inc(&srq->refcount);
	xa_unlock(&srq_table->xa);

	if (!srq) {
		dev_warn(hr_dev->dev, "Async event for bogus SRQ %08x\n", srqn);
		return;
	}

	srq->event(srq, event_type);

	if (atomic_dec_and_test(&srq->refcount))
		complete(&srq->free);
}
EXPORT_SYMBOL_GPL(hns_roce_srq_event);

35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212
static void hns_roce_ib_srq_event(struct hns_roce_srq *srq,
				  enum hns_roce_event event_type)
{
	struct hns_roce_dev *hr_dev = to_hr_dev(srq->ibsrq.device);
	struct ib_srq *ibsrq = &srq->ibsrq;
	struct ib_event event;

	if (ibsrq->event_handler) {
		event.device      = ibsrq->device;
		event.element.srq = ibsrq;
		switch (event_type) {
		case HNS_ROCE_EVENT_TYPE_SRQ_LIMIT_REACH:
			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
			break;
		case HNS_ROCE_EVENT_TYPE_SRQ_CATAS_ERROR:
			event.event = IB_EVENT_SRQ_ERR;
			break;
		default:
			dev_err(hr_dev->dev,
			   "hns_roce:Unexpected event type 0x%x on SRQ %06lx\n",
			   event_type, srq->srqn);
			return;
		}

		ibsrq->event_handler(&event, ibsrq->srq_context);
	}
}

static int hns_roce_sw2hw_srq(struct hns_roce_dev *dev,
			      struct hns_roce_cmd_mailbox *mailbox,
			      unsigned long srq_num)
{
	return hns_roce_cmd_mbox(dev, mailbox->dma, 0, srq_num, 0,
				 HNS_ROCE_CMD_SW2HW_SRQ,
				 HNS_ROCE_CMD_TIMEOUT_MSECS);
}

static int hns_roce_hw2sw_srq(struct hns_roce_dev *dev,
			     struct hns_roce_cmd_mailbox *mailbox,
			     unsigned long srq_num)
{
	return hns_roce_cmd_mbox(dev, 0, mailbox ? mailbox->dma : 0, srq_num,
				 mailbox ? 0 : 1, HNS_ROCE_CMD_HW2SW_SRQ,
				 HNS_ROCE_CMD_TIMEOUT_MSECS);
}

int hns_roce_srq_alloc(struct hns_roce_dev *hr_dev, u32 pdn, u32 cqn, u16 xrcd,
		       struct hns_roce_mtt *hr_mtt, u64 db_rec_addr,
		       struct hns_roce_srq *srq)
{
	struct hns_roce_srq_table *srq_table = &hr_dev->srq_table;
	struct hns_roce_cmd_mailbox *mailbox;
	dma_addr_t dma_handle_wqe;
	dma_addr_t dma_handle_idx;
	u64 *mtts_wqe;
	u64 *mtts_idx;
	int ret;

	/* Get the physical address of srq buf */
	mtts_wqe = hns_roce_table_find(hr_dev,
				       &hr_dev->mr_table.mtt_srqwqe_table,
				       srq->mtt.first_seg,
				       &dma_handle_wqe);
	if (!mtts_wqe) {
		dev_err(hr_dev->dev,
			"SRQ alloc.Failed to find srq buf addr.\n");
		return -EINVAL;
	}

	/* Get physical address of idx que buf */
	mtts_idx = hns_roce_table_find(hr_dev, &hr_dev->mr_table.mtt_idx_table,
				       srq->idx_que.mtt.first_seg,
				       &dma_handle_idx);
	if (!mtts_idx) {
		dev_err(hr_dev->dev,
			"SRQ alloc.Failed to find idx que buf addr.\n");
		return -EINVAL;
	}

	ret = hns_roce_bitmap_alloc(&srq_table->bitmap, &srq->srqn);
	if (ret == -1) {
		dev_err(hr_dev->dev, "SRQ alloc.Failed to alloc index.\n");
		return -ENOMEM;
	}

	ret = hns_roce_table_get(hr_dev, &srq_table->table, srq->srqn);
	if (ret)
		goto err_out;

	ret = xa_err(xa_store(&srq_table->xa, srq->srqn, srq, GFP_KERNEL));
	if (ret)
		goto err_put;

	mailbox = hns_roce_alloc_cmd_mailbox(hr_dev);
	if (IS_ERR(mailbox)) {
		ret = PTR_ERR(mailbox);
		goto err_xa;
	}

	hr_dev->hw->write_srqc(hr_dev, srq, pdn, xrcd, cqn, mailbox->buf,
			       mtts_wqe, mtts_idx, dma_handle_wqe,
			       dma_handle_idx);

	ret = hns_roce_sw2hw_srq(hr_dev, mailbox, srq->srqn);
	hns_roce_free_cmd_mailbox(hr_dev, mailbox);
	if (ret)
		goto err_xa;

	atomic_set(&srq->refcount, 1);
	init_completion(&srq->free);
	return ret;

err_xa:
	xa_erase(&srq_table->xa, srq->srqn);

err_put:
	hns_roce_table_put(hr_dev, &srq_table->table, srq->srqn);

err_out:
	hns_roce_bitmap_free(&srq_table->bitmap, srq->srqn, BITMAP_NO_RR);
	return ret;
}

void hns_roce_srq_free(struct hns_roce_dev *hr_dev, struct hns_roce_srq *srq)
{
	struct hns_roce_srq_table *srq_table = &hr_dev->srq_table;
	int ret;

	ret = hns_roce_hw2sw_srq(hr_dev, NULL, srq->srqn);
	if (ret)
		dev_err(hr_dev->dev, "HW2SW_SRQ failed (%d) for CQN %06lx\n",
			ret, srq->srqn);

	xa_erase(&srq_table->xa, srq->srqn);

	if (atomic_dec_and_test(&srq->refcount))
		complete(&srq->free);
	wait_for_completion(&srq->free);

	hns_roce_table_put(hr_dev, &srq_table->table, srq->srqn);
	hns_roce_bitmap_free(&srq_table->bitmap, srq->srqn, BITMAP_NO_RR);
}

static int hns_roce_create_idx_que(struct ib_pd *pd, struct hns_roce_srq *srq,
				   u32 page_shift)
{
	struct hns_roce_dev *hr_dev = to_hr_dev(pd->device);
	struct hns_roce_idx_que *idx_que = &srq->idx_que;
	u32 bitmap_num;
	int i;

	bitmap_num = HNS_ROCE_ALOGN_UP(srq->max, 8 * sizeof(u64));

	idx_que->bitmap = kcalloc(1, bitmap_num / 8, GFP_KERNEL);
	if (!idx_que->bitmap)
		return -ENOMEM;

	bitmap_num = bitmap_num / (8 * sizeof(u64));

	idx_que->buf_size = srq->idx_que.buf_size;

	if (hns_roce_buf_alloc(hr_dev, idx_que->buf_size, (1 << page_shift) * 2,
			       &idx_que->idx_buf, page_shift)) {
		kfree(idx_que->bitmap);
		return -ENOMEM;
	}

	for (i = 0; i < bitmap_num; i++)
		idx_que->bitmap[i] = ~(0UL);

	return 0;
}

struct ib_srq *hns_roce_create_srq(struct ib_pd *pd,
				   struct ib_srq_init_attr *srq_init_attr,
				   struct ib_udata *udata)
{
	struct hns_roce_dev *hr_dev = to_hr_dev(pd->device);
213
	struct hns_roce_ib_create_srq_resp resp = {};
214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255
	struct hns_roce_srq *srq;
	int srq_desc_size;
	int srq_buf_size;
	u32 page_shift;
	int ret = 0;
	u32 npages;
	u32 cqn;

	/* Check the actual SRQ wqe and SRQ sge num */
	if (srq_init_attr->attr.max_wr >= hr_dev->caps.max_srq_wrs ||
	    srq_init_attr->attr.max_sge > hr_dev->caps.max_srq_sges)
		return ERR_PTR(-EINVAL);

	srq = kzalloc(sizeof(*srq), GFP_KERNEL);
	if (!srq)
		return ERR_PTR(-ENOMEM);

	mutex_init(&srq->mutex);
	spin_lock_init(&srq->lock);

	srq->max = roundup_pow_of_two(srq_init_attr->attr.max_wr + 1);
	srq->max_gs = srq_init_attr->attr.max_sge;

	srq_desc_size = max(16, 16 * srq->max_gs);

	srq->wqe_shift = ilog2(srq_desc_size);

	srq_buf_size = srq->max * srq_desc_size;

	srq->idx_que.entry_sz = HNS_ROCE_IDX_QUE_ENTRY_SZ;
	srq->idx_que.buf_size = srq->max * srq->idx_que.entry_sz;
	srq->mtt.mtt_type = MTT_TYPE_SRQWQE;
	srq->idx_que.mtt.mtt_type = MTT_TYPE_IDX;

	if (udata) {
		struct hns_roce_ib_create_srq  ucmd;

		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
			ret = -EFAULT;
			goto err_srq;
		}

256 257
		srq->umem =
			ib_umem_get(udata, ucmd.buf_addr, srq_buf_size, 0, 0);
258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283
		if (IS_ERR(srq->umem)) {
			ret = PTR_ERR(srq->umem);
			goto err_srq;
		}

		if (hr_dev->caps.srqwqe_buf_pg_sz) {
			npages = (ib_umem_page_count(srq->umem) +
				  (1 << hr_dev->caps.srqwqe_buf_pg_sz) - 1) /
				  (1 << hr_dev->caps.srqwqe_buf_pg_sz);
			page_shift = PAGE_SHIFT + hr_dev->caps.srqwqe_buf_pg_sz;
			ret = hns_roce_mtt_init(hr_dev, npages,
						page_shift,
						&srq->mtt);
		} else
			ret = hns_roce_mtt_init(hr_dev,
						ib_umem_page_count(srq->umem),
						srq->umem->page_shift,
						&srq->mtt);
		if (ret)
			goto err_buf;

		ret = hns_roce_ib_umem_write_mtt(hr_dev, &srq->mtt, srq->umem);
		if (ret)
			goto err_srq_mtt;

		/* config index queue BA */
284
		srq->idx_que.umem = ib_umem_get(udata, ucmd.que_addr,
285 286 287 288
						srq->idx_que.buf_size, 0, 0);
		if (IS_ERR(srq->idx_que.umem)) {
			dev_err(hr_dev->dev,
				"ib_umem_get error for index queue\n");
289
			ret = PTR_ERR(srq->idx_que.umem);
290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
			goto err_srq_mtt;
		}

		if (hr_dev->caps.idx_buf_pg_sz) {
			npages = (ib_umem_page_count(srq->idx_que.umem) +
				  (1 << hr_dev->caps.idx_buf_pg_sz) - 1) /
				  (1 << hr_dev->caps.idx_buf_pg_sz);
			page_shift = PAGE_SHIFT + hr_dev->caps.idx_buf_pg_sz;
			ret = hns_roce_mtt_init(hr_dev, npages,
						page_shift, &srq->idx_que.mtt);
		} else {
			ret = hns_roce_mtt_init(hr_dev,
				       ib_umem_page_count(srq->idx_que.umem),
				       srq->idx_que.umem->page_shift,
				       &srq->idx_que.mtt);
		}

		if (ret) {
			dev_err(hr_dev->dev,
				"hns_roce_mtt_init error for idx que\n");
			goto err_idx_mtt;
		}

		ret = hns_roce_ib_umem_write_mtt(hr_dev, &srq->idx_que.mtt,
						 srq->idx_que.umem);
		if (ret) {
			dev_err(hr_dev->dev,
			      "hns_roce_ib_umem_write_mtt error for idx que\n");
			goto err_idx_buf;
		}
	} else {
		page_shift = PAGE_SHIFT + hr_dev->caps.srqwqe_buf_pg_sz;
		if (hns_roce_buf_alloc(hr_dev, srq_buf_size,
				      (1 << page_shift) * 2,
				      &srq->buf, page_shift)) {
			ret = -ENOMEM;
			goto err_srq;
		}

		srq->head = 0;
		srq->tail = srq->max - 1;

		ret = hns_roce_mtt_init(hr_dev, srq->buf.npages,
					srq->buf.page_shift, &srq->mtt);
		if (ret)
			goto err_buf;

		ret = hns_roce_buf_write_mtt(hr_dev, &srq->mtt, &srq->buf);
		if (ret)
			goto err_srq_mtt;

		page_shift = PAGE_SHIFT + hr_dev->caps.idx_buf_pg_sz;
		ret = hns_roce_create_idx_que(pd, srq, page_shift);
		if (ret) {
			dev_err(hr_dev->dev, "Create idx queue fail(%d)!\n",
				ret);
			goto err_srq_mtt;
		}

		/* Init mtt table for idx_que */
		ret = hns_roce_mtt_init(hr_dev, srq->idx_que.idx_buf.npages,
					srq->idx_que.idx_buf.page_shift,
					&srq->idx_que.mtt);
		if (ret)
			goto err_create_idx;

		/* Write buffer address into the mtt table */
		ret = hns_roce_buf_write_mtt(hr_dev, &srq->idx_que.mtt,
					     &srq->idx_que.idx_buf);
		if (ret)
			goto err_idx_buf;

		srq->wrid = kvmalloc_array(srq->max, sizeof(u64), GFP_KERNEL);
		if (!srq->wrid) {
			ret = -ENOMEM;
			goto err_idx_buf;
		}
	}

	cqn = ib_srq_has_cq(srq_init_attr->srq_type) ?
	      to_hr_cq(srq_init_attr->ext.cq)->cqn : 0;

	srq->db_reg_l = hr_dev->reg_base + SRQ_DB_REG;

	ret = hns_roce_srq_alloc(hr_dev, to_hr_pd(pd)->pdn, cqn, 0,
				 &srq->mtt, 0, srq);
	if (ret)
		goto err_wrid;

	srq->event = hns_roce_ib_srq_event;
	srq->ibsrq.ext.xrc.srq_num = srq->srqn;
381
	resp.srqn = srq->srqn;
382

383
	if (udata) {
384 385
		if (ib_copy_to_udata(udata, &resp,
				     min(udata->outlen, sizeof(resp)))) {
386
			ret = -EFAULT;
387
			goto err_srqc_alloc;
388 389 390 391 392
		}
	}

	return &srq->ibsrq;

393 394 395
err_srqc_alloc:
	hns_roce_srq_free(hr_dev, srq);

396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447
err_wrid:
	kvfree(srq->wrid);

err_idx_buf:
	hns_roce_mtt_cleanup(hr_dev, &srq->idx_que.mtt);

err_idx_mtt:
	if (udata)
		ib_umem_release(srq->idx_que.umem);

err_create_idx:
	hns_roce_buf_free(hr_dev, srq->idx_que.buf_size,
			  &srq->idx_que.idx_buf);
	kfree(srq->idx_que.bitmap);

err_srq_mtt:
	hns_roce_mtt_cleanup(hr_dev, &srq->mtt);

err_buf:
	if (udata)
		ib_umem_release(srq->umem);
	else
		hns_roce_buf_free(hr_dev, srq_buf_size, &srq->buf);

err_srq:
	kfree(srq);
	return ERR_PTR(ret);
}

int hns_roce_destroy_srq(struct ib_srq *ibsrq)
{
	struct hns_roce_dev *hr_dev = to_hr_dev(ibsrq->device);
	struct hns_roce_srq *srq = to_hr_srq(ibsrq);

	hns_roce_srq_free(hr_dev, srq);
	hns_roce_mtt_cleanup(hr_dev, &srq->mtt);

	if (ibsrq->uobject) {
		hns_roce_mtt_cleanup(hr_dev, &srq->idx_que.mtt);
		ib_umem_release(srq->idx_que.umem);
		ib_umem_release(srq->umem);
	} else {
		kvfree(srq->wrid);
		hns_roce_buf_free(hr_dev, srq->max << srq->wqe_shift,
				  &srq->buf);
	}

	kfree(srq);

	return 0;
}

448 449 450 451 452 453 454 455 456 457 458 459 460 461 462
int hns_roce_init_srq_table(struct hns_roce_dev *hr_dev)
{
	struct hns_roce_srq_table *srq_table = &hr_dev->srq_table;

	xa_init(&srq_table->xa);

	return hns_roce_bitmap_init(&srq_table->bitmap, hr_dev->caps.num_srqs,
				    hr_dev->caps.num_srqs - 1,
				    hr_dev->caps.reserved_srqs, 0);
}

void hns_roce_cleanup_srq_table(struct hns_roce_dev *hr_dev)
{
	hns_roce_bitmap_cleanup(&hr_dev->srq_table.bitmap);
}
反馈
建议
客服 返回
顶部