2 * Copyright (c) 2013-2015, Mellanox Technologies. All rights reserved.
4 * This software is available to you under a choice of one of two
5 * licenses. You may choose to be licensed under the terms of the GNU
6 * General Public License (GPL) Version 2, available from the file
7 * COPYING in the main directory of this source tree, or the
8 * OpenIB.org BSD license below:
10 * Redistribution and use in source and binary forms, with or
11 * without modification, are permitted provided that the following
14 * - Redistributions of source code must retain the above
15 * copyright notice, this list of conditions and the following
18 * - Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials
21 * provided with the distribution.
23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
33 #include <rdma/ib_umem.h>
34 #include <rdma/ib_umem_odp.h>
36 #include <linux/jiffies.h>
39 * Fill in a physical address list. ib_umem_num_dma_blocks() entries will be
40 * filled in the pas array.
42 void mlx5_ib_populate_pas(struct ib_umem *umem, size_t page_size, __be64 *pas,
45 struct ib_block_iter biter;
47 rdma_umem_for_each_dma_block (umem, &biter, page_size) {
48 *pas = cpu_to_be64(rdma_block_iter_dma_address(&biter) |
55 * Compute the page shift and page_offset for mailboxes that use a quantized
56 * page_offset. The granulatity of the page offset scales according to page
59 unsigned long __mlx5_umem_find_best_quantized_pgoff(
60 struct ib_umem *umem, unsigned long pgsz_bitmap,
61 unsigned int page_offset_bits, u64 pgoff_bitmask, unsigned int scale,
62 unsigned int *page_offset_quantized)
64 const u64 page_offset_mask = (1UL << page_offset_bits) - 1;
65 unsigned long page_size;
68 page_size = ib_umem_find_best_pgoff(umem, pgsz_bitmap, pgoff_bitmask);
73 * page size is the largest possible page size.
75 * Reduce the page_size, and thus the page_offset and quanta, until the
76 * page_offset fits into the mailbox field. Once page_size < scale this
77 * loop is guaranteed to terminate.
79 page_offset = ib_umem_dma_offset(umem, page_size);
80 while (page_offset & ~(u64)(page_offset_mask * (page_size / scale))) {
82 page_offset = ib_umem_dma_offset(umem, page_size);
86 * The address is not aligned, or otherwise cannot be represented by the
89 if (!(pgsz_bitmap & page_size))
92 *page_offset_quantized =
93 (unsigned long)page_offset / (page_size / scale);
94 if (WARN_ON(*page_offset_quantized > page_offset_mask))
100 #define WR_ID_END 0xBAD
101 #define TEST_WC_NUM_WQES 255
102 #define TEST_WC_POLLING_MAX_TIME_JIFFIES msecs_to_jiffies(100)
103 static int post_send_nop(struct mlx5_ib_dev *dev, struct ib_qp *ibqp, u64 wr_id,
106 struct mlx5_ib_qp *qp = to_mqp(ibqp);
107 struct mlx5_wqe_ctrl_seg *ctrl;
108 struct mlx5_bf *bf = &qp->bf;
109 __be32 mmio_wqe[16] = {};
114 if (unlikely(dev->mdev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR))
117 spin_lock_irqsave(&qp->sq.lock, flags);
119 idx = qp->sq.cur_post & (qp->sq.wqe_cnt - 1);
120 ctrl = mlx5_frag_buf_get_wqe(&qp->sq.fbc, idx);
122 memset(ctrl, 0, sizeof(struct mlx5_wqe_ctrl_seg));
123 ctrl->fm_ce_se = signaled ? MLX5_WQE_CTRL_CQ_UPDATE : 0;
124 ctrl->opmod_idx_opcode =
125 cpu_to_be32(((u32)(qp->sq.cur_post) << 8) | MLX5_OPCODE_NOP);
126 ctrl->qpn_ds = cpu_to_be32((sizeof(struct mlx5_wqe_ctrl_seg) / 16) |
127 (qp->trans_qp.base.mqp.qpn << 8));
129 qp->sq.wrid[idx] = wr_id;
130 qp->sq.w_list[idx].opcode = MLX5_OPCODE_NOP;
131 qp->sq.wqe_head[idx] = qp->sq.head + 1;
132 qp->sq.cur_post += DIV_ROUND_UP(sizeof(struct mlx5_wqe_ctrl_seg),
134 qp->sq.w_list[idx].next = qp->sq.cur_post;
137 memcpy(mmio_wqe, ctrl, sizeof(*ctrl));
138 ((struct mlx5_wqe_ctrl_seg *)&mmio_wqe)->fm_ce_se |=
139 MLX5_WQE_CTRL_CQ_UPDATE;
141 /* Make sure that descriptors are written before
142 * updating doorbell record and ringing the doorbell
146 qp->db.db[MLX5_SND_DBR] = cpu_to_be32(qp->sq.cur_post);
148 /* Make sure doorbell record is visible to the HCA before
152 for (i = 0; i < 8; i++)
153 mlx5_write64(&mmio_wqe[i * 2],
154 bf->bfreg->map + bf->offset + i * 8);
156 bf->offset ^= bf->buf_size;
158 spin_unlock_irqrestore(&qp->sq.lock, flags);
163 static int test_wc_poll_cq_result(struct mlx5_ib_dev *dev, struct ib_cq *cq)
166 struct ib_wc wc = {};
167 unsigned long end = jiffies + TEST_WC_POLLING_MAX_TIME_JIFFIES;
170 ret = ib_poll_cq(cq, 1, &wc);
171 if (ret < 0 || wc.status)
172 return ret < 0 ? ret : -EINVAL;
175 } while (!time_after(jiffies, end));
180 if (wc.wr_id != WR_ID_BF)
186 static int test_wc_do_send(struct mlx5_ib_dev *dev, struct ib_qp *qp)
190 for (i = 0; i < TEST_WC_NUM_WQES; i++) {
191 err = post_send_nop(dev, qp, WR_ID_BF, false);
196 return post_send_nop(dev, qp, WR_ID_END, true);
199 int mlx5_ib_test_wc(struct mlx5_ib_dev *dev)
201 struct ib_cq_init_attr cq_attr = { .cqe = TEST_WC_NUM_WQES + 1 };
202 int port_type_cap = MLX5_CAP_GEN(dev->mdev, port_type);
203 struct ib_qp_init_attr qp_init_attr = {
204 .cap = { .max_send_wr = TEST_WC_NUM_WQES },
205 .qp_type = IB_QPT_UD,
206 .sq_sig_type = IB_SIGNAL_REQ_WR,
207 .create_flags = MLX5_IB_QP_CREATE_WC_TEST,
209 struct ib_qp_attr qp_attr = { .port_num = 1 };
210 struct ib_device *ibdev = &dev->ib_dev;
216 if (!MLX5_CAP_GEN(dev->mdev, bf))
219 if (!dev->mdev->roce.roce_en &&
220 port_type_cap == MLX5_CAP_PORT_TYPE_ETH) {
221 if (mlx5_core_is_pf(dev->mdev))
222 dev->wc_support = arch_can_pci_mmap_wc();
226 ret = mlx5_alloc_bfreg(dev->mdev, &dev->wc_bfreg, true, false);
230 if (!dev->wc_bfreg.wc)
233 pd = ib_alloc_pd(ibdev, 0);
239 cq = ib_create_cq(ibdev, NULL, NULL, NULL, &cq_attr);
245 qp_init_attr.recv_cq = cq;
246 qp_init_attr.send_cq = cq;
247 qp = ib_create_qp(pd, &qp_init_attr);
253 qp_attr.qp_state = IB_QPS_INIT;
254 ret = ib_modify_qp(qp, &qp_attr,
255 IB_QP_STATE | IB_QP_PORT | IB_QP_PKEY_INDEX |
260 qp_attr.qp_state = IB_QPS_RTR;
261 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE);
265 qp_attr.qp_state = IB_QPS_RTS;
266 ret = ib_modify_qp(qp, &qp_attr, IB_QP_STATE | IB_QP_SQ_PSN);
270 ret = test_wc_do_send(dev, qp);
274 ret = test_wc_poll_cq_result(dev, cq);
276 dev->wc_support = true;
287 mlx5_free_bfreg(dev->mdev, &dev->wc_bfreg);
292 "Error %d while trying to test write-combining support\n",