profile.c revision 272407
1/* 2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved. 3 * Copyright (c) 2005, 2014 Mellanox Technologies. All rights reserved. 4 * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the 10 * OpenIB.org BSD license below: 11 * 12 * Redistribution and use in source and binary forms, with or 13 * without modification, are permitted provided that the following 14 * conditions are met: 15 * 16 * - Redistributions of source code must retain the above 17 * copyright notice, this list of conditions and the following 18 * disclaimer. 19 * 20 * - Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials 23 * provided with the distribution. 24 * 25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 32 * SOFTWARE. 33 */ 34 35#include <linux/slab.h> 36 37#include "mlx4.h" 38#include "fw.h" 39 40enum { 41 MLX4_RES_QP, 42 MLX4_RES_RDMARC, 43 MLX4_RES_ALTC, 44 MLX4_RES_AUXC, 45 MLX4_RES_SRQ, 46 MLX4_RES_CQ, 47 MLX4_RES_EQ, 48 MLX4_RES_DMPT, 49 MLX4_RES_CMPT, 50 MLX4_RES_MTT, 51 MLX4_RES_MCG, 52 MLX4_RES_NUM 53}; 54 55static const char *res_name[] = { 56 [MLX4_RES_QP] = "QP", 57 [MLX4_RES_RDMARC] = "RDMARC", 58 [MLX4_RES_ALTC] = "ALTC", 59 [MLX4_RES_AUXC] = "AUXC", 60 [MLX4_RES_SRQ] = "SRQ", 61 [MLX4_RES_CQ] = "CQ", 62 [MLX4_RES_EQ] = "EQ", 63 [MLX4_RES_DMPT] = "DMPT", 64 [MLX4_RES_CMPT] = "CMPT", 65 [MLX4_RES_MTT] = "MTT", 66 [MLX4_RES_MCG] = "MCG", 67}; 68 69u64 mlx4_make_profile(struct mlx4_dev *dev, 70 struct mlx4_profile *request, 71 struct mlx4_dev_cap *dev_cap, 72 struct mlx4_init_hca_param *init_hca) 73{ 74 struct mlx4_priv *priv = mlx4_priv(dev); 75 struct mlx4_resource { 76 u64 size; 77 u64 start; 78 int type; 79 u64 num; 80 int log_num; 81 }; 82 83 u64 total_size = 0; 84 struct mlx4_resource *profile; 85 struct mlx4_resource tmp; 86 int i, j; 87 88 profile = kcalloc(MLX4_RES_NUM, sizeof(*profile), GFP_KERNEL); 89 if (!profile) 90 return -ENOMEM; 91 92 profile[MLX4_RES_QP].size = dev_cap->qpc_entry_sz; 93 profile[MLX4_RES_RDMARC].size = dev_cap->rdmarc_entry_sz; 94 profile[MLX4_RES_ALTC].size = dev_cap->altc_entry_sz; 95 profile[MLX4_RES_AUXC].size = dev_cap->aux_entry_sz; 96 profile[MLX4_RES_SRQ].size = dev_cap->srq_entry_sz; 97 profile[MLX4_RES_CQ].size = dev_cap->cqc_entry_sz; 98 profile[MLX4_RES_EQ].size = dev_cap->eqc_entry_sz; 99 profile[MLX4_RES_DMPT].size = dev_cap->dmpt_entry_sz; 100 profile[MLX4_RES_CMPT].size = dev_cap->cmpt_entry_sz; 101 profile[MLX4_RES_MTT].size = dev_cap->mtt_entry_sz; 102 profile[MLX4_RES_MCG].size = mlx4_get_mgm_entry_size(dev); 103 104 profile[MLX4_RES_QP].num = request->num_qp; 105 profile[MLX4_RES_RDMARC].num = request->num_qp * request->rdmarc_per_qp; 106 profile[MLX4_RES_ALTC].num = request->num_qp; 107 profile[MLX4_RES_AUXC].num = request->num_qp; 108 profile[MLX4_RES_SRQ].num = request->num_srq; 109 profile[MLX4_RES_CQ].num = request->num_cq; 110 profile[MLX4_RES_EQ].num = mlx4_is_mfunc(dev) ? 111 dev->phys_caps.num_phys_eqs : 112 min_t(unsigned, dev_cap->max_eqs, MAX_MSIX); 113 profile[MLX4_RES_DMPT].num = request->num_mpt; 114 profile[MLX4_RES_CMPT].num = MLX4_NUM_CMPTS; 115 profile[MLX4_RES_MTT].num = ((u64)request->num_mtt_segs) * 116 (1 << log_mtts_per_seg); 117 profile[MLX4_RES_MCG].num = request->num_mcg; 118 119 for (i = 0; i < MLX4_RES_NUM; ++i) { 120 profile[i].type = i; 121 profile[i].num = roundup_pow_of_two(profile[i].num); 122 profile[i].log_num = ilog2(profile[i].num); 123 profile[i].size *= profile[i].num; 124 profile[i].size = max(profile[i].size, (u64) PAGE_SIZE); 125 } 126 127 /* 128 * Sort the resources in decreasing order of size. Since they 129 * all have sizes that are powers of 2, we'll be able to keep 130 * resources aligned to their size and pack them without gaps 131 * using the sorted order. 132 */ 133 for (i = MLX4_RES_NUM; i > 0; --i) 134 for (j = 1; j < i; ++j) { 135 if (profile[j].size > profile[j - 1].size) { 136 tmp = profile[j]; 137 profile[j] = profile[j - 1]; 138 profile[j - 1] = tmp; 139 } 140 } 141 142 for (i = 0; i < MLX4_RES_NUM; ++i) { 143 if (profile[i].size) { 144 profile[i].start = total_size; 145 total_size += profile[i].size; 146 } 147 148 if (total_size > dev_cap->max_icm_sz) { 149 mlx4_err(dev, "Profile requires 0x%llx bytes; " 150 "won't fit in 0x%llx bytes of context memory.\n", 151 (unsigned long long) total_size, 152 (unsigned long long) dev_cap->max_icm_sz); 153 kfree(profile); 154 return -ENOMEM; 155 } 156 157 if (profile[i].size) 158 mlx4_dbg(dev, " profile[%2d] (%6s): 2^%02d entries @ 0x%10llx, " 159 "size 0x%10llx\n", 160 i, res_name[profile[i].type], profile[i].log_num, 161 (unsigned long long) profile[i].start, 162 (unsigned long long) profile[i].size); 163 } 164 165 mlx4_dbg(dev, "HCA context memory: reserving %d KB\n", 166 (int) (total_size >> 10)); 167 168 for (i = 0; i < MLX4_RES_NUM; ++i) { 169 switch (profile[i].type) { 170 case MLX4_RES_QP: 171 dev->caps.num_qps = profile[i].num; 172 init_hca->qpc_base = profile[i].start; 173 init_hca->log_num_qps = profile[i].log_num; 174 break; 175 case MLX4_RES_RDMARC: 176 for (priv->qp_table.rdmarc_shift = 0; 177 request->num_qp << priv->qp_table.rdmarc_shift < profile[i].num; 178 ++priv->qp_table.rdmarc_shift) 179 ; /* nothing */ 180 dev->caps.max_qp_dest_rdma = 1 << priv->qp_table.rdmarc_shift; 181 priv->qp_table.rdmarc_base = (u32) profile[i].start; 182 init_hca->rdmarc_base = profile[i].start; 183 init_hca->log_rd_per_qp = priv->qp_table.rdmarc_shift; 184 break; 185 case MLX4_RES_ALTC: 186 init_hca->altc_base = profile[i].start; 187 break; 188 case MLX4_RES_AUXC: 189 init_hca->auxc_base = profile[i].start; 190 break; 191 case MLX4_RES_SRQ: 192 dev->caps.num_srqs = profile[i].num; 193 init_hca->srqc_base = profile[i].start; 194 init_hca->log_num_srqs = profile[i].log_num; 195 break; 196 case MLX4_RES_CQ: 197 dev->caps.num_cqs = profile[i].num; 198 init_hca->cqc_base = profile[i].start; 199 init_hca->log_num_cqs = profile[i].log_num; 200 break; 201 case MLX4_RES_EQ: 202 dev->caps.num_eqs = roundup_pow_of_two(min_t(unsigned, dev_cap->max_eqs, 203 MAX_MSIX)); 204 init_hca->eqc_base = profile[i].start; 205 init_hca->log_num_eqs = ilog2(dev->caps.num_eqs); 206 break; 207 case MLX4_RES_DMPT: 208 dev->caps.num_mpts = profile[i].num; 209 priv->mr_table.mpt_base = profile[i].start; 210 init_hca->dmpt_base = profile[i].start; 211 init_hca->log_mpt_sz = profile[i].log_num; 212 break; 213 case MLX4_RES_CMPT: 214 init_hca->cmpt_base = profile[i].start; 215 break; 216 case MLX4_RES_MTT: 217 dev->caps.num_mtts = profile[i].num; 218 priv->mr_table.mtt_base = profile[i].start; 219 init_hca->mtt_base = profile[i].start; 220 break; 221 case MLX4_RES_MCG: 222 init_hca->mc_base = profile[i].start; 223 init_hca->log_mc_entry_sz = 224 ilog2(mlx4_get_mgm_entry_size(dev)); 225 init_hca->log_mc_table_sz = profile[i].log_num; 226 if (dev->caps.steering_mode == 227 MLX4_STEERING_MODE_DEVICE_MANAGED) { 228 dev->caps.num_mgms = profile[i].num; 229 } else { 230 init_hca->log_mc_hash_sz = 231 profile[i].log_num - 1; 232 dev->caps.num_mgms = profile[i].num >> 1; 233 dev->caps.num_amgms = profile[i].num >> 1; 234 } 235 break; 236 default: 237 break; 238 } 239 } 240 241 /* 242 * PDs don't take any HCA memory, but we assign them as part 243 * of the HCA profile anyway. 244 */ 245 dev->caps.num_pds = MLX4_NUM_PDS; 246 247 kfree(profile); 248 return total_size; 249} 250