1 /*-
2 * Copyright (C) 2012-2016 Intel Corporation
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 *
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24 * SUCH DAMAGE.
25 */
26
27 #include <sys/cdefs.h>
28 __FBSDID("$FreeBSD$");
29
30 #include <sys/param.h>
31 #include <sys/bus.h>
32 #include <sys/sysctl.h>
33
34 #include "nvme_private.h"
35
36 SYSCTL_NODE(_kern, OID_AUTO, nvme, CTLFLAG_RD, 0, "NVM Express");
37 /*
38 * Intel NVMe controllers have a slow path for I/Os that span a 128KB
39 * stripe boundary but ZFS limits ashift, which is derived from
40 * d_stripesize, to 13 (8KB) so we limit the stripesize reported to
41 * geom(8) to 4KB by default.
42 *
43 * This may result in a small number of additional I/Os to require
44 * splitting in nvme(4), however the NVMe I/O path is very efficient
45 * so these additional I/Os will cause very minimal (if any) difference
46 * in performance or CPU utilisation.
47 */
48 int nvme_max_optimal_sectorsize = 1<<12;
49 SYSCTL_INT(_kern_nvme, OID_AUTO, max_optimal_sectorsize, CTLFLAG_RWTUN,
50 &nvme_max_optimal_sectorsize, 0, "The maximum optimal sectorsize reported");
51
52 /*
53 * CTLTYPE_S64 and sysctl_handle_64 were added in r217616. Define these
54 * explicitly here for older kernels that don't include the r217616
55 * changeset.
56 */
57 #ifndef CTLTYPE_S64
58 #define CTLTYPE_S64 CTLTYPE_QUAD
59 #define sysctl_handle_64 sysctl_handle_quad
60 #endif
61
62 static void
nvme_dump_queue(struct nvme_qpair * qpair)63 nvme_dump_queue(struct nvme_qpair *qpair)
64 {
65 struct nvme_completion *cpl;
66 struct nvme_command *cmd;
67 int i;
68
69 printf("id:%04Xh phase:%d\n", qpair->id, qpair->phase);
70
71 printf("Completion queue:\n");
72 for (i = 0; i < qpair->num_entries; i++) {
73 cpl = &qpair->cpl[i];
74 printf("%05d: ", i);
75 nvme_dump_completion(cpl);
76 }
77
78 printf("Submission queue:\n");
79 for (i = 0; i < qpair->num_entries; i++) {
80 cmd = &qpair->cmd[i];
81 printf("%05d: ", i);
82 nvme_dump_command(cmd);
83 }
84 }
85
86
87 static int
nvme_sysctl_dump_debug(SYSCTL_HANDLER_ARGS)88 nvme_sysctl_dump_debug(SYSCTL_HANDLER_ARGS)
89 {
90 struct nvme_qpair *qpair = arg1;
91 uint32_t val = 0;
92
93 int error = sysctl_handle_int(oidp, &val, 0, req);
94
95 if (error)
96 return (error);
97
98 if (val != 0)
99 nvme_dump_queue(qpair);
100
101 return (0);
102 }
103
104 static int
nvme_sysctl_int_coal_time(SYSCTL_HANDLER_ARGS)105 nvme_sysctl_int_coal_time(SYSCTL_HANDLER_ARGS)
106 {
107 struct nvme_controller *ctrlr = arg1;
108 uint32_t oldval = ctrlr->int_coal_time;
109 int error = sysctl_handle_int(oidp, &ctrlr->int_coal_time, 0,
110 req);
111
112 if (error)
113 return (error);
114
115 if (oldval != ctrlr->int_coal_time)
116 nvme_ctrlr_cmd_set_interrupt_coalescing(ctrlr,
117 ctrlr->int_coal_time, ctrlr->int_coal_threshold, NULL,
118 NULL);
119
120 return (0);
121 }
122
123 static int
nvme_sysctl_int_coal_threshold(SYSCTL_HANDLER_ARGS)124 nvme_sysctl_int_coal_threshold(SYSCTL_HANDLER_ARGS)
125 {
126 struct nvme_controller *ctrlr = arg1;
127 uint32_t oldval = ctrlr->int_coal_threshold;
128 int error = sysctl_handle_int(oidp, &ctrlr->int_coal_threshold, 0,
129 req);
130
131 if (error)
132 return (error);
133
134 if (oldval != ctrlr->int_coal_threshold)
135 nvme_ctrlr_cmd_set_interrupt_coalescing(ctrlr,
136 ctrlr->int_coal_time, ctrlr->int_coal_threshold, NULL,
137 NULL);
138
139 return (0);
140 }
141
142 static int
nvme_sysctl_timeout_period(SYSCTL_HANDLER_ARGS)143 nvme_sysctl_timeout_period(SYSCTL_HANDLER_ARGS)
144 {
145 struct nvme_controller *ctrlr = arg1;
146 uint32_t oldval = ctrlr->timeout_period;
147 int error = sysctl_handle_int(oidp, &ctrlr->timeout_period, 0, req);
148
149 if (error)
150 return (error);
151
152 if (ctrlr->timeout_period > NVME_MAX_TIMEOUT_PERIOD ||
153 ctrlr->timeout_period < NVME_MIN_TIMEOUT_PERIOD) {
154 ctrlr->timeout_period = oldval;
155 return (EINVAL);
156 }
157
158 return (0);
159 }
160
161 static void
nvme_qpair_reset_stats(struct nvme_qpair * qpair)162 nvme_qpair_reset_stats(struct nvme_qpair *qpair)
163 {
164
165 qpair->num_cmds = 0;
166 qpair->num_intr_handler_calls = 0;
167 }
168
169 static int
nvme_sysctl_num_cmds(SYSCTL_HANDLER_ARGS)170 nvme_sysctl_num_cmds(SYSCTL_HANDLER_ARGS)
171 {
172 struct nvme_controller *ctrlr = arg1;
173 int64_t num_cmds = 0;
174 int i;
175
176 num_cmds = ctrlr->adminq.num_cmds;
177
178 for (i = 0; i < ctrlr->num_io_queues; i++)
179 num_cmds += ctrlr->ioq[i].num_cmds;
180
181 return (sysctl_handle_64(oidp, &num_cmds, 0, req));
182 }
183
184 static int
nvme_sysctl_num_intr_handler_calls(SYSCTL_HANDLER_ARGS)185 nvme_sysctl_num_intr_handler_calls(SYSCTL_HANDLER_ARGS)
186 {
187 struct nvme_controller *ctrlr = arg1;
188 int64_t num_intr_handler_calls = 0;
189 int i;
190
191 num_intr_handler_calls = ctrlr->adminq.num_intr_handler_calls;
192
193 for (i = 0; i < ctrlr->num_io_queues; i++)
194 num_intr_handler_calls += ctrlr->ioq[i].num_intr_handler_calls;
195
196 return (sysctl_handle_64(oidp, &num_intr_handler_calls, 0, req));
197 }
198
199 static int
nvme_sysctl_reset_stats(SYSCTL_HANDLER_ARGS)200 nvme_sysctl_reset_stats(SYSCTL_HANDLER_ARGS)
201 {
202 struct nvme_controller *ctrlr = arg1;
203 uint32_t i, val = 0;
204
205 int error = sysctl_handle_int(oidp, &val, 0, req);
206
207 if (error)
208 return (error);
209
210 if (val != 0) {
211 nvme_qpair_reset_stats(&ctrlr->adminq);
212
213 for (i = 0; i < ctrlr->num_io_queues; i++)
214 nvme_qpair_reset_stats(&ctrlr->ioq[i]);
215 }
216
217 return (0);
218 }
219
220
221 static void
nvme_sysctl_initialize_queue(struct nvme_qpair * qpair,struct sysctl_ctx_list * ctrlr_ctx,struct sysctl_oid * que_tree)222 nvme_sysctl_initialize_queue(struct nvme_qpair *qpair,
223 struct sysctl_ctx_list *ctrlr_ctx, struct sysctl_oid *que_tree)
224 {
225 struct sysctl_oid_list *que_list = SYSCTL_CHILDREN(que_tree);
226
227 SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "num_entries",
228 CTLFLAG_RD, &qpair->num_entries, 0,
229 "Number of entries in hardware queue");
230 SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "num_trackers",
231 CTLFLAG_RD, &qpair->num_trackers, 0,
232 "Number of trackers pre-allocated for this queue pair");
233 SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "sq_head",
234 CTLFLAG_RD, &qpair->sq_head, 0,
235 "Current head of submission queue (as observed by driver)");
236 SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "sq_tail",
237 CTLFLAG_RD, &qpair->sq_tail, 0,
238 "Current tail of submission queue (as observed by driver)");
239 SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "cq_head",
240 CTLFLAG_RD, &qpair->cq_head, 0,
241 "Current head of completion queue (as observed by driver)");
242
243 SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_cmds",
244 CTLFLAG_RD, &qpair->num_cmds, "Number of commands submitted");
245 SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_intr_handler_calls",
246 CTLFLAG_RD, &qpair->num_intr_handler_calls,
247 "Number of times interrupt handler was invoked (will typically be "
248 "less than number of actual interrupts generated due to "
249 "coalescing)");
250
251 SYSCTL_ADD_PROC(ctrlr_ctx, que_list, OID_AUTO,
252 "dump_debug", CTLTYPE_UINT | CTLFLAG_RW, qpair, 0,
253 nvme_sysctl_dump_debug, "IU", "Dump debug data");
254 }
255
256 void
nvme_sysctl_initialize_ctrlr(struct nvme_controller * ctrlr)257 nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr)
258 {
259 struct sysctl_ctx_list *ctrlr_ctx;
260 struct sysctl_oid *ctrlr_tree, *que_tree;
261 struct sysctl_oid_list *ctrlr_list;
262 #define QUEUE_NAME_LENGTH 16
263 char queue_name[QUEUE_NAME_LENGTH];
264 int i;
265
266 ctrlr_ctx = device_get_sysctl_ctx(ctrlr->dev);
267 ctrlr_tree = device_get_sysctl_tree(ctrlr->dev);
268 ctrlr_list = SYSCTL_CHILDREN(ctrlr_tree);
269
270 SYSCTL_ADD_UINT(ctrlr_ctx, ctrlr_list, OID_AUTO, "num_cpus_per_ioq",
271 CTLFLAG_RD, &ctrlr->num_cpus_per_ioq, 0,
272 "Number of CPUs assigned per I/O queue pair");
273
274 SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
275 "int_coal_time", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
276 nvme_sysctl_int_coal_time, "IU",
277 "Interrupt coalescing timeout (in microseconds)");
278
279 SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
280 "int_coal_threshold", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
281 nvme_sysctl_int_coal_threshold, "IU",
282 "Interrupt coalescing threshold");
283
284 SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
285 "timeout_period", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
286 nvme_sysctl_timeout_period, "IU",
287 "Timeout period (in seconds)");
288
289 SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
290 "num_cmds", CTLTYPE_S64 | CTLFLAG_RD,
291 ctrlr, 0, nvme_sysctl_num_cmds, "IU",
292 "Number of commands submitted");
293
294 SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
295 "num_intr_handler_calls", CTLTYPE_S64 | CTLFLAG_RD,
296 ctrlr, 0, nvme_sysctl_num_intr_handler_calls, "IU",
297 "Number of times interrupt handler was invoked (will "
298 "typically be less than number of actual interrupts "
299 "generated due to coalescing)");
300
301 SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO,
302 "reset_stats", CTLTYPE_UINT | CTLFLAG_RW, ctrlr, 0,
303 nvme_sysctl_reset_stats, "IU", "Reset statistics to zero");
304
305 que_tree = SYSCTL_ADD_NODE(ctrlr_ctx, ctrlr_list, OID_AUTO, "adminq",
306 CTLFLAG_RD, NULL, "Admin Queue");
307
308 nvme_sysctl_initialize_queue(&ctrlr->adminq, ctrlr_ctx, que_tree);
309
310 for (i = 0; i < ctrlr->num_io_queues; i++) {
311 snprintf(queue_name, QUEUE_NAME_LENGTH, "ioq%d", i);
312 que_tree = SYSCTL_ADD_NODE(ctrlr_ctx, ctrlr_list, OID_AUTO,
313 queue_name, CTLFLAG_RD, NULL, "IO Queue");
314 nvme_sysctl_initialize_queue(&ctrlr->ioq[i], ctrlr_ctx,
315 que_tree);
316 }
317 }
318