1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 * Copyright 2022 Oxide Computer Company
25 */
26
27
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/stat.h>
31 #include <sys/open.h>
32 #include <sys/file.h>
33 #include <sys/conf.h>
34 #include <sys/modctl.h>
35 #include <sys/cmn_err.h>
36 #include <sys/bitmap.h>
37 #include <sys/debug.h>
38 #include <sys/kmem.h>
39 #include <sys/errno.h>
40 #include <sys/sysmacros.h>
41 #include <sys/lockstat.h>
42 #include <sys/atomic.h>
43 #include <sys/dtrace.h>
44
45 #include <sys/ddi.h>
46 #include <sys/sunddi.h>
47
48 typedef struct lockstat_probe {
49 const char *lsp_func;
50 const char *lsp_name;
51 int lsp_probe;
52 dtrace_id_t lsp_id;
53 } lockstat_probe_t;
54
55 lockstat_probe_t lockstat_probes[] =
56 {
57 { LS_MUTEX_ENTER, LSA_ACQUIRE, LS_MUTEX_ENTER_ACQUIRE },
58 { LS_MUTEX_ENTER, LSA_BLOCK, LS_MUTEX_ENTER_BLOCK },
59 { LS_MUTEX_ENTER, LSA_SPIN, LS_MUTEX_ENTER_SPIN },
60 { LS_MUTEX_EXIT, LSA_RELEASE, LS_MUTEX_EXIT_RELEASE },
61 { LS_MUTEX_DESTROY, LSA_RELEASE, LS_MUTEX_DESTROY_RELEASE },
62 { LS_MUTEX_TRYENTER, LSA_ACQUIRE, LS_MUTEX_TRYENTER_ACQUIRE },
63 { LS_LOCK_SET, LSS_ACQUIRE, LS_LOCK_SET_ACQUIRE },
64 { LS_LOCK_SET, LSS_SPIN, LS_LOCK_SET_SPIN },
65 { LS_LOCK_SET_SPL, LSS_ACQUIRE, LS_LOCK_SET_SPL_ACQUIRE },
66 { LS_LOCK_SET_SPL, LSS_SPIN, LS_LOCK_SET_SPL_SPIN },
67 { LS_LOCK_TRY, LSS_ACQUIRE, LS_LOCK_TRY_ACQUIRE },
68 { LS_LOCK_CLEAR, LSS_RELEASE, LS_LOCK_CLEAR_RELEASE },
69 { LS_LOCK_CLEAR_SPLX, LSS_RELEASE, LS_LOCK_CLEAR_SPLX_RELEASE },
70 { LS_CLOCK_UNLOCK, LSS_RELEASE, LS_CLOCK_UNLOCK_RELEASE },
71 { LS_RW_ENTER, LSR_ACQUIRE, LS_RW_ENTER_ACQUIRE },
72 { LS_RW_ENTER, LSR_BLOCK, LS_RW_ENTER_BLOCK },
73 { LS_RW_EXIT, LSR_RELEASE, LS_RW_EXIT_RELEASE },
74 { LS_RW_TRYENTER, LSR_ACQUIRE, LS_RW_TRYENTER_ACQUIRE },
75 { LS_RW_TRYUPGRADE, LSR_UPGRADE, LS_RW_TRYUPGRADE_UPGRADE },
76 { LS_RW_DOWNGRADE, LSR_DOWNGRADE, LS_RW_DOWNGRADE_DOWNGRADE },
77 { LS_THREAD_LOCK, LST_SPIN, LS_THREAD_LOCK_SPIN },
78 { LS_THREAD_LOCK_HIGH, LST_SPIN, LS_THREAD_LOCK_HIGH_SPIN },
79 { NULL }
80 };
81
82 static dev_info_t *lockstat_devi; /* saved in xxattach() for xxinfo() */
83 static kmutex_t lockstat_test; /* for testing purposes only */
84 static dtrace_provider_id_t lockstat_id;
85
86 /*ARGSUSED*/
87 static int
lockstat_enable(void * arg,dtrace_id_t id,void * parg)88 lockstat_enable(void *arg, dtrace_id_t id, void *parg)
89 {
90 lockstat_probe_t *probe = parg;
91
92 ASSERT(!lockstat_probemap[probe->lsp_probe]);
93
94 lockstat_probemap[probe->lsp_probe] = id;
95 membar_producer();
96
97 lockstat_hotpatch_probe(probe->lsp_probe);
98 membar_producer();
99
100 /*
101 * Immediately generate a record for the lockstat_test mutex
102 * to verify that the mutex hot-patch code worked as expected.
103 */
104 mutex_enter(&lockstat_test);
105 mutex_exit(&lockstat_test);
106 return (0);
107 }
108
109 /*ARGSUSED*/
110 static void
lockstat_disable(void * arg,dtrace_id_t id,void * parg)111 lockstat_disable(void *arg, dtrace_id_t id, void *parg)
112 {
113 lockstat_probe_t *probe = parg;
114 int i;
115
116 ASSERT(lockstat_probemap[probe->lsp_probe]);
117
118 lockstat_probemap[probe->lsp_probe] = 0;
119 lockstat_hotpatch_probe(probe->lsp_probe);
120 membar_producer();
121
122 /*
123 * See if we have any probes left enabled.
124 */
125 for (i = 0; i < LS_NPROBES; i++) {
126 if (lockstat_probemap[i]) {
127 /*
128 * This probe is still enabled. We don't need to deal
129 * with waiting for all threads to be out of the
130 * lockstat critical sections; just return.
131 */
132 return;
133 }
134 }
135
136 /*
137 * The delay() here isn't as cheesy as you might think. We don't
138 * want to busy-loop in the kernel, so we have to give up the
139 * CPU between calls to lockstat_active_threads(); that much is
140 * obvious. But the reason it's a do..while loop rather than a
141 * while loop is subtle. The memory barrier above guarantees that
142 * no threads will enter the lockstat code from this point forward.
143 * However, another thread could already be executing lockstat code
144 * without our knowledge if the update to its t_lockstat field hasn't
145 * cleared its CPU's store buffer. Delaying for one clock tick
146 * guarantees that either (1) the thread will have *ample* time to
147 * complete its work, or (2) the thread will be preempted, in which
148 * case it will have to grab and release a dispatcher lock, which
149 * will flush that CPU's store buffer. Either way we're covered.
150 */
151 do {
152 delay(1);
153 } while (lockstat_active_threads());
154 }
155
156 /*ARGSUSED*/
157 static int
lockstat_open(dev_t * devp,int flag,int otyp,cred_t * cred_p)158 lockstat_open(dev_t *devp, int flag, int otyp, cred_t *cred_p)
159 {
160 return (0);
161 }
162
163 /* ARGSUSED */
164 static int
lockstat_info(dev_info_t * dip,ddi_info_cmd_t infocmd,void * arg,void ** result)165 lockstat_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
166 {
167 int error;
168
169 switch (infocmd) {
170 case DDI_INFO_DEVT2DEVINFO:
171 *result = (void *) lockstat_devi;
172 error = DDI_SUCCESS;
173 break;
174 case DDI_INFO_DEVT2INSTANCE:
175 *result = (void *)0;
176 error = DDI_SUCCESS;
177 break;
178 default:
179 error = DDI_FAILURE;
180 }
181 return (error);
182 }
183
184 /*ARGSUSED*/
185 static void
lockstat_provide(void * arg,const dtrace_probedesc_t * desc)186 lockstat_provide(void *arg, const dtrace_probedesc_t *desc)
187 {
188 int i = 0;
189
190 for (i = 0; lockstat_probes[i].lsp_func != NULL; i++) {
191 lockstat_probe_t *probe = &lockstat_probes[i];
192
193 if (dtrace_probe_lookup(lockstat_id, "genunix",
194 probe->lsp_func, probe->lsp_name) != 0)
195 continue;
196
197 ASSERT(!probe->lsp_id);
198 probe->lsp_id = dtrace_probe_create(lockstat_id,
199 "genunix", probe->lsp_func, probe->lsp_name,
200 1, probe);
201 }
202 }
203
204 /*ARGSUSED*/
205 static void
lockstat_destroy(void * arg,dtrace_id_t id,void * parg)206 lockstat_destroy(void *arg, dtrace_id_t id, void *parg)
207 {
208 lockstat_probe_t *probe = parg;
209
210 ASSERT(!lockstat_probemap[probe->lsp_probe]);
211 probe->lsp_id = 0;
212 }
213
214 static dtrace_pattr_t lockstat_attr = {
215 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
216 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
217 { DTRACE_STABILITY_PRIVATE, DTRACE_STABILITY_PRIVATE, DTRACE_CLASS_UNKNOWN },
218 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
219 { DTRACE_STABILITY_EVOLVING, DTRACE_STABILITY_EVOLVING, DTRACE_CLASS_COMMON },
220 };
221
222 static dtrace_pops_t lockstat_pops = {
223 lockstat_provide,
224 NULL,
225 lockstat_enable,
226 lockstat_disable,
227 NULL,
228 NULL,
229 NULL,
230 NULL,
231 NULL,
232 lockstat_destroy
233 };
234
235 static int
lockstat_attach(dev_info_t * devi,ddi_attach_cmd_t cmd)236 lockstat_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
237 {
238 switch (cmd) {
239 case DDI_ATTACH:
240 break;
241 case DDI_RESUME:
242 return (DDI_SUCCESS);
243 default:
244 return (DDI_FAILURE);
245 }
246
247 if (ddi_create_minor_node(devi, "lockstat", S_IFCHR, 0,
248 DDI_PSEUDO, 0) == DDI_FAILURE ||
249 dtrace_register("lockstat", &lockstat_attr, DTRACE_PRIV_KERNEL,
250 NULL, &lockstat_pops, NULL, &lockstat_id) != 0) {
251 ddi_remove_minor_node(devi, NULL);
252 return (DDI_FAILURE);
253 }
254
255 lockstat_probe = dtrace_probe;
256 membar_producer();
257
258 ddi_report_dev(devi);
259 lockstat_devi = devi;
260 return (DDI_SUCCESS);
261 }
262
263 static int
lockstat_detach(dev_info_t * devi,ddi_detach_cmd_t cmd)264 lockstat_detach(dev_info_t *devi, ddi_detach_cmd_t cmd)
265 {
266 switch (cmd) {
267 case DDI_DETACH:
268 break;
269 case DDI_SUSPEND:
270 return (DDI_SUCCESS);
271 default:
272 return (DDI_FAILURE);
273 }
274
275 if (dtrace_unregister(lockstat_id) != 0)
276 return (DDI_FAILURE);
277
278 ddi_remove_minor_node(devi, NULL);
279 return (DDI_SUCCESS);
280 }
281
282 /*
283 * Configuration data structures
284 */
285 static struct cb_ops lockstat_cb_ops = {
286 lockstat_open, /* open */
287 nodev, /* close */
288 nulldev, /* strategy */
289 nulldev, /* print */
290 nodev, /* dump */
291 nodev, /* read */
292 nodev, /* write */
293 nodev, /* ioctl */
294 nodev, /* devmap */
295 nodev, /* mmap */
296 nodev, /* segmap */
297 nochpoll, /* poll */
298 ddi_prop_op, /* cb_prop_op */
299 0, /* streamtab */
300 D_MP | D_NEW /* Driver compatibility flag */
301 };
302
303 static struct dev_ops lockstat_ops = {
304 DEVO_REV, /* devo_rev, */
305 0, /* refcnt */
306 lockstat_info, /* getinfo */
307 nulldev, /* identify */
308 nulldev, /* probe */
309 lockstat_attach, /* attach */
310 lockstat_detach, /* detach */
311 nulldev, /* reset */
312 &lockstat_cb_ops, /* cb_ops */
313 NULL, /* bus_ops */
314 NULL, /* power */
315 ddi_quiesce_not_needed, /* quiesce */
316 };
317
318 static struct modldrv modldrv = {
319 &mod_driverops, /* Type of module. This one is a driver */
320 "Lock Statistics", /* name of module */
321 &lockstat_ops, /* driver ops */
322 };
323
324 static struct modlinkage modlinkage = {
325 MODREV_1, (void *)&modldrv, NULL
326 };
327
328 int
_init(void)329 _init(void)
330 {
331 return (mod_install(&modlinkage));
332 }
333
334 int
_fini(void)335 _fini(void)
336 {
337 return (mod_remove(&modlinkage));
338 }
339
340 int
_info(struct modinfo * modinfop)341 _info(struct modinfo *modinfop)
342 {
343 return (mod_info(&modlinkage, modinfop));
344 }
345