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 34 #include <linux/vmalloc.h> 35 #include <rdma/ib_umem.h> 36 #include "hns_roce_device.h" 37 38 void hns_roce_buf_free(struct hns_roce_dev *hr_dev, struct hns_roce_buf *buf) 39 { 40 struct hns_roce_buf_list *trunks; 41 u32 i; 42 43 if (!buf) 44 return; 45 46 trunks = buf->trunk_list; 47 if (trunks) { 48 buf->trunk_list = NULL; 49 for (i = 0; i < buf->ntrunks; i++) 50 dma_free_coherent(hr_dev->dev, 1 << buf->trunk_shift, 51 trunks[i].buf, trunks[i].map); 52 53 kfree(trunks); 54 } 55 56 kfree(buf); 57 } 58 59 /* 60 * Allocate the dma buffer for storing ROCEE table entries 61 * 62 * @size: required size 63 * @page_shift: the unit size in a continuous dma address range 64 * @flags: HNS_ROCE_BUF_ flags to control the allocation flow. 65 */ 66 struct hns_roce_buf *hns_roce_buf_alloc(struct hns_roce_dev *hr_dev, u32 size, 67 u32 page_shift, u32 flags) 68 { 69 u32 trunk_size, page_size, alloced_size; 70 struct hns_roce_buf_list *trunks; 71 struct hns_roce_buf *buf; 72 gfp_t gfp_flags; 73 u32 ntrunk, i; 74 75 /* The minimum shift of the page accessed by hw is HNS_HW_PAGE_SHIFT */ 76 if (WARN_ON(page_shift < HNS_HW_PAGE_SHIFT)) 77 return ERR_PTR(-EINVAL); 78 79 gfp_flags = (flags & HNS_ROCE_BUF_NOSLEEP) ? GFP_ATOMIC : GFP_KERNEL; 80 buf = kzalloc(sizeof(*buf), gfp_flags); 81 if (!buf) 82 return ERR_PTR(-ENOMEM); 83 84 buf->page_shift = page_shift; 85 page_size = 1 << buf->page_shift; 86 87 /* Calc the trunk size and num by required size and page_shift */ 88 if (flags & HNS_ROCE_BUF_DIRECT) { 89 buf->trunk_shift = order_base_2(ALIGN(size, PAGE_SIZE)); 90 ntrunk = 1; 91 } else { 92 buf->trunk_shift = order_base_2(ALIGN(page_size, PAGE_SIZE)); 93 ntrunk = DIV_ROUND_UP(size, 1 << buf->trunk_shift); 94 } 95 96 trunks = kcalloc(ntrunk, sizeof(*trunks), gfp_flags); 97 if (!trunks) { 98 kfree(buf); 99 return ERR_PTR(-ENOMEM); 100 } 101 102 trunk_size = 1 << buf->trunk_shift; 103 alloced_size = 0; 104 for (i = 0; i < ntrunk; i++) { 105 trunks[i].buf = dma_alloc_coherent(hr_dev->dev, trunk_size, 106 &trunks[i].map, gfp_flags); 107 if (!trunks[i].buf) 108 break; 109 110 alloced_size += trunk_size; 111 } 112 113 buf->ntrunks = i; 114 115 /* In nofail mode, it's only failed when the alloced size is 0 */ 116 if ((flags & HNS_ROCE_BUF_NOFAIL) ? i == 0 : i != ntrunk) { 117 for (i = 0; i < buf->ntrunks; i++) 118 dma_free_coherent(hr_dev->dev, trunk_size, 119 trunks[i].buf, trunks[i].map); 120 121 kfree(trunks); 122 kfree(buf); 123 return ERR_PTR(-ENOMEM); 124 } 125 126 buf->npages = DIV_ROUND_UP(alloced_size, page_size); 127 buf->trunk_list = trunks; 128 129 return buf; 130 } 131 132 int hns_roce_get_kmem_bufs(struct hns_roce_dev *hr_dev, dma_addr_t *bufs, 133 int buf_cnt, struct hns_roce_buf *buf, 134 unsigned int page_shift) 135 { 136 unsigned int offset, max_size; 137 int total = 0; 138 int i; 139 140 if (page_shift > buf->trunk_shift) { 141 dev_err(hr_dev->dev, "failed to check kmem buf shift %u > %u\n", 142 page_shift, buf->trunk_shift); 143 return -EINVAL; 144 } 145 146 offset = 0; 147 max_size = buf->ntrunks << buf->trunk_shift; 148 for (i = 0; i < buf_cnt && offset < max_size; i++) { 149 bufs[total++] = hns_roce_buf_dma_addr(buf, offset); 150 offset += (1 << page_shift); 151 } 152 153 return total; 154 } 155 156 int hns_roce_get_umem_bufs(struct hns_roce_dev *hr_dev, dma_addr_t *bufs, 157 int buf_cnt, struct ib_umem *umem, 158 unsigned int page_shift) 159 { 160 struct ib_block_iter biter; 161 int total = 0; 162 163 /* convert system page cnt to hw page cnt */ 164 rdma_umem_for_each_dma_block(umem, &biter, 1 << page_shift) { 165 bufs[total++] = rdma_block_iter_dma_address(&biter); 166 if (total >= buf_cnt) 167 goto done; 168 } 169 170 done: 171 return total; 172 } 173 174 void hns_roce_cleanup_bitmap(struct hns_roce_dev *hr_dev) 175 { 176 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC) 177 ida_destroy(&hr_dev->xrcd_ida.ida); 178 179 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) 180 ida_destroy(&hr_dev->srq_table.srq_ida.ida); 181 hns_roce_cleanup_qp_table(hr_dev); 182 hns_roce_cleanup_cq_table(hr_dev); 183 ida_destroy(&hr_dev->mr_table.mtpt_ida.ida); 184 ida_destroy(&hr_dev->pd_ida.ida); 185 ida_destroy(&hr_dev->uar_ida.ida); 186 } 187