| 1 | // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause |
| 2 | /* |
| 3 | * Copyright (c) 2015-2018 Oracle. All rights reserved. |
| 4 | * Copyright (c) 2005-2006 Network Appliance, 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 BSD-type |
| 10 | * license below: |
| 11 | * |
| 12 | * Redistribution and use in source and binary forms, with or without |
| 13 | * modification, are permitted provided that the following conditions |
| 14 | * are met: |
| 15 | * |
| 16 | * Redistributions of source code must retain the above copyright |
| 17 | * notice, this list of conditions and the following 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 provided |
| 22 | * with the distribution. |
| 23 | * |
| 24 | * Neither the name of the Network Appliance, Inc. nor the names of |
| 25 | * its contributors may be used to endorse or promote products |
| 26 | * derived from this software without specific prior written |
| 27 | * permission. |
| 28 | * |
| 29 | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 30 | * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 31 | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 32 | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT |
| 33 | * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, |
| 34 | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT |
| 35 | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, |
| 36 | * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY |
| 37 | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT |
| 38 | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE |
| 39 | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 40 | * |
| 41 | * Author: Tom Tucker <tom@opengridcomputing.com> |
| 42 | */ |
| 43 | |
| 44 | #include <linux/slab.h> |
| 45 | #include <linux/fs.h> |
| 46 | #include <linux/sysctl.h> |
| 47 | #include <linux/workqueue.h> |
| 48 | #include <linux/sunrpc/clnt.h> |
| 49 | #include <linux/sunrpc/sched.h> |
| 50 | #include <linux/sunrpc/svc_rdma.h> |
| 51 | |
| 52 | #define RPCDBG_FACILITY RPCDBG_SVCXPRT |
| 53 | |
| 54 | /* RPC/RDMA parameters */ |
| 55 | unsigned int svcrdma_ord = 16; /* historical default */ |
| 56 | static unsigned int min_ord = 1; |
| 57 | static unsigned int max_ord = 255; |
| 58 | unsigned int svcrdma_max_requests = RPCRDMA_MAX_REQUESTS; |
| 59 | unsigned int svcrdma_max_bc_requests = RPCRDMA_MAX_BC_REQUESTS; |
| 60 | static unsigned int min_max_requests = 4; |
| 61 | static unsigned int max_max_requests = 16384; |
| 62 | unsigned int svcrdma_max_req_size = RPCRDMA_DEF_INLINE_THRESH; |
| 63 | static unsigned int min_max_inline = RPCRDMA_DEF_INLINE_THRESH; |
| 64 | static unsigned int max_max_inline = RPCRDMA_MAX_INLINE_THRESH; |
| 65 | static unsigned int svcrdma_stat_unused; |
| 66 | static unsigned int zero; |
| 67 | |
| 68 | struct percpu_counter svcrdma_stat_read; |
| 69 | struct percpu_counter svcrdma_stat_recv; |
| 70 | struct percpu_counter svcrdma_stat_sq_starve; |
| 71 | struct percpu_counter svcrdma_stat_write; |
| 72 | |
| 73 | enum { |
| 74 | SVCRDMA_COUNTER_BUFSIZ = sizeof(unsigned long long), |
| 75 | }; |
| 76 | |
| 77 | static int svcrdma_counter_handler(const struct ctl_table *table, int write, |
| 78 | void *buffer, size_t *lenp, loff_t *ppos) |
| 79 | { |
| 80 | struct percpu_counter *stat = (struct percpu_counter *)table->data; |
| 81 | char tmp[SVCRDMA_COUNTER_BUFSIZ + 1]; |
| 82 | int len; |
| 83 | |
| 84 | if (write) { |
| 85 | percpu_counter_set(fbc: stat, amount: 0); |
| 86 | return 0; |
| 87 | } |
| 88 | |
| 89 | len = snprintf(buf: tmp, size: SVCRDMA_COUNTER_BUFSIZ, fmt: "%lld\n" , |
| 90 | percpu_counter_sum_positive(fbc: stat)); |
| 91 | if (len >= SVCRDMA_COUNTER_BUFSIZ) |
| 92 | return -EFAULT; |
| 93 | len = strlen(tmp); |
| 94 | if (*ppos > len) { |
| 95 | *lenp = 0; |
| 96 | return 0; |
| 97 | } |
| 98 | len -= *ppos; |
| 99 | if (len > *lenp) |
| 100 | len = *lenp; |
| 101 | if (len) |
| 102 | memcpy(buffer, tmp, len); |
| 103 | *lenp = len; |
| 104 | *ppos += len; |
| 105 | |
| 106 | return 0; |
| 107 | } |
| 108 | |
| 109 | static struct ctl_table_header *; |
| 110 | static struct ctl_table svcrdma_parm_table[] = { |
| 111 | { |
| 112 | .procname = "max_requests" , |
| 113 | .data = &svcrdma_max_requests, |
| 114 | .maxlen = sizeof(unsigned int), |
| 115 | .mode = 0644, |
| 116 | .proc_handler = proc_dointvec_minmax, |
| 117 | .extra1 = &min_max_requests, |
| 118 | .extra2 = &max_max_requests |
| 119 | }, |
| 120 | { |
| 121 | .procname = "max_req_size" , |
| 122 | .data = &svcrdma_max_req_size, |
| 123 | .maxlen = sizeof(unsigned int), |
| 124 | .mode = 0644, |
| 125 | .proc_handler = proc_dointvec_minmax, |
| 126 | .extra1 = &min_max_inline, |
| 127 | .extra2 = &max_max_inline |
| 128 | }, |
| 129 | { |
| 130 | .procname = "max_outbound_read_requests" , |
| 131 | .data = &svcrdma_ord, |
| 132 | .maxlen = sizeof(unsigned int), |
| 133 | .mode = 0644, |
| 134 | .proc_handler = proc_dointvec_minmax, |
| 135 | .extra1 = &min_ord, |
| 136 | .extra2 = &max_ord, |
| 137 | }, |
| 138 | |
| 139 | { |
| 140 | .procname = "rdma_stat_read" , |
| 141 | .data = &svcrdma_stat_read, |
| 142 | .maxlen = SVCRDMA_COUNTER_BUFSIZ, |
| 143 | .mode = 0644, |
| 144 | .proc_handler = svcrdma_counter_handler, |
| 145 | }, |
| 146 | { |
| 147 | .procname = "rdma_stat_recv" , |
| 148 | .data = &svcrdma_stat_recv, |
| 149 | .maxlen = SVCRDMA_COUNTER_BUFSIZ, |
| 150 | .mode = 0644, |
| 151 | .proc_handler = svcrdma_counter_handler, |
| 152 | }, |
| 153 | { |
| 154 | .procname = "rdma_stat_write" , |
| 155 | .data = &svcrdma_stat_write, |
| 156 | .maxlen = SVCRDMA_COUNTER_BUFSIZ, |
| 157 | .mode = 0644, |
| 158 | .proc_handler = svcrdma_counter_handler, |
| 159 | }, |
| 160 | { |
| 161 | .procname = "rdma_stat_sq_starve" , |
| 162 | .data = &svcrdma_stat_sq_starve, |
| 163 | .maxlen = SVCRDMA_COUNTER_BUFSIZ, |
| 164 | .mode = 0644, |
| 165 | .proc_handler = svcrdma_counter_handler, |
| 166 | }, |
| 167 | { |
| 168 | .procname = "rdma_stat_rq_starve" , |
| 169 | .data = &svcrdma_stat_unused, |
| 170 | .maxlen = sizeof(unsigned int), |
| 171 | .mode = 0644, |
| 172 | .proc_handler = proc_dointvec_minmax, |
| 173 | .extra1 = &zero, |
| 174 | .extra2 = &zero, |
| 175 | }, |
| 176 | { |
| 177 | .procname = "rdma_stat_rq_poll" , |
| 178 | .data = &svcrdma_stat_unused, |
| 179 | .maxlen = sizeof(unsigned int), |
| 180 | .mode = 0644, |
| 181 | .proc_handler = proc_dointvec_minmax, |
| 182 | .extra1 = &zero, |
| 183 | .extra2 = &zero, |
| 184 | }, |
| 185 | { |
| 186 | .procname = "rdma_stat_rq_prod" , |
| 187 | .data = &svcrdma_stat_unused, |
| 188 | .maxlen = sizeof(unsigned int), |
| 189 | .mode = 0644, |
| 190 | .proc_handler = proc_dointvec_minmax, |
| 191 | .extra1 = &zero, |
| 192 | .extra2 = &zero, |
| 193 | }, |
| 194 | { |
| 195 | .procname = "rdma_stat_sq_poll" , |
| 196 | .data = &svcrdma_stat_unused, |
| 197 | .maxlen = sizeof(unsigned int), |
| 198 | .mode = 0644, |
| 199 | .proc_handler = proc_dointvec_minmax, |
| 200 | .extra1 = &zero, |
| 201 | .extra2 = &zero, |
| 202 | }, |
| 203 | { |
| 204 | .procname = "rdma_stat_sq_prod" , |
| 205 | .data = &svcrdma_stat_unused, |
| 206 | .maxlen = sizeof(unsigned int), |
| 207 | .mode = 0644, |
| 208 | .proc_handler = proc_dointvec_minmax, |
| 209 | .extra1 = &zero, |
| 210 | .extra2 = &zero, |
| 211 | }, |
| 212 | }; |
| 213 | |
| 214 | static void svc_rdma_proc_cleanup(void) |
| 215 | { |
| 216 | if (!svcrdma_table_header) |
| 217 | return; |
| 218 | unregister_sysctl_table(table: svcrdma_table_header); |
| 219 | svcrdma_table_header = NULL; |
| 220 | |
| 221 | percpu_counter_destroy(fbc: &svcrdma_stat_write); |
| 222 | percpu_counter_destroy(fbc: &svcrdma_stat_sq_starve); |
| 223 | percpu_counter_destroy(fbc: &svcrdma_stat_recv); |
| 224 | percpu_counter_destroy(fbc: &svcrdma_stat_read); |
| 225 | } |
| 226 | |
| 227 | static int svc_rdma_proc_init(void) |
| 228 | { |
| 229 | int rc; |
| 230 | |
| 231 | if (svcrdma_table_header) |
| 232 | return 0; |
| 233 | |
| 234 | rc = percpu_counter_init(&svcrdma_stat_read, 0, GFP_KERNEL); |
| 235 | if (rc) |
| 236 | goto err; |
| 237 | rc = percpu_counter_init(&svcrdma_stat_recv, 0, GFP_KERNEL); |
| 238 | if (rc) |
| 239 | goto err_read; |
| 240 | rc = percpu_counter_init(&svcrdma_stat_sq_starve, 0, GFP_KERNEL); |
| 241 | if (rc) |
| 242 | goto err_recv; |
| 243 | rc = percpu_counter_init(&svcrdma_stat_write, 0, GFP_KERNEL); |
| 244 | if (rc) |
| 245 | goto err_sq; |
| 246 | |
| 247 | svcrdma_table_header = register_sysctl("sunrpc/svc_rdma" , |
| 248 | svcrdma_parm_table); |
| 249 | if (!svcrdma_table_header) |
| 250 | goto err_write; |
| 251 | |
| 252 | return 0; |
| 253 | |
| 254 | err_write: |
| 255 | rc = -ENOMEM; |
| 256 | percpu_counter_destroy(fbc: &svcrdma_stat_write); |
| 257 | err_sq: |
| 258 | percpu_counter_destroy(fbc: &svcrdma_stat_sq_starve); |
| 259 | err_recv: |
| 260 | percpu_counter_destroy(fbc: &svcrdma_stat_recv); |
| 261 | err_read: |
| 262 | percpu_counter_destroy(fbc: &svcrdma_stat_read); |
| 263 | err: |
| 264 | return rc; |
| 265 | } |
| 266 | |
| 267 | struct workqueue_struct *svcrdma_wq; |
| 268 | |
| 269 | void svc_rdma_cleanup(void) |
| 270 | { |
| 271 | svc_unreg_xprt_class(&svc_rdma_class); |
| 272 | svc_rdma_proc_cleanup(); |
| 273 | if (svcrdma_wq) { |
| 274 | struct workqueue_struct *wq = svcrdma_wq; |
| 275 | |
| 276 | svcrdma_wq = NULL; |
| 277 | destroy_workqueue(wq); |
| 278 | } |
| 279 | |
| 280 | dprintk("SVCRDMA Module Removed, deregister RPC RDMA transport\n" ); |
| 281 | } |
| 282 | |
| 283 | int svc_rdma_init(void) |
| 284 | { |
| 285 | struct workqueue_struct *wq; |
| 286 | int rc; |
| 287 | |
| 288 | wq = alloc_workqueue("svcrdma" , WQ_UNBOUND, 0); |
| 289 | if (!wq) |
| 290 | return -ENOMEM; |
| 291 | |
| 292 | rc = svc_rdma_proc_init(); |
| 293 | if (rc) { |
| 294 | destroy_workqueue(wq); |
| 295 | return rc; |
| 296 | } |
| 297 | |
| 298 | svcrdma_wq = wq; |
| 299 | svc_reg_xprt_class(&svc_rdma_class); |
| 300 | |
| 301 | dprintk("SVCRDMA Module Init, register RPC RDMA transport\n" ); |
| 302 | dprintk("\tsvcrdma_ord : %d\n" , svcrdma_ord); |
| 303 | dprintk("\tmax_requests : %u\n" , svcrdma_max_requests); |
| 304 | dprintk("\tmax_bc_requests : %u\n" , svcrdma_max_bc_requests); |
| 305 | dprintk("\tmax_inline : %d\n" , svcrdma_max_req_size); |
| 306 | return 0; |
| 307 | } |
| 308 | |