xref: /illumos-gate/usr/src/uts/common/cpr/cpr_mod.c (revision 3dec9fcdd56adf1b4a563137b4915c8f2d83b881)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * System call to checkpoint and resume the currently running kernel
30  */
31 #include <sys/types.h>
32 #include <sys/errno.h>
33 #include <sys/modctl.h>
34 #include <sys/syscall.h>
35 #include <sys/cred.h>
36 #include <sys/uadmin.h>
37 #include <sys/cmn_err.h>
38 #include <sys/systm.h>
39 #include <sys/cpr.h>
40 #include <sys/swap.h>
41 #include <sys/vfs.h>
42 #include <sys/autoconf.h>
43 #include <sys/machsystm.h>
44 
45 extern int i_cpr_is_supported(int sleeptype);
46 extern int cpr_is_ufs(struct vfs *);
47 extern int cpr_check_spec_statefile(void);
48 extern int cpr_reusable_mount_check(void);
49 extern int i_cpr_reusable_supported(void);
50 extern int i_cpr_reusefini(void);
51 extern struct mod_ops mod_miscops;
52 
53 extern int cpr_init(int);
54 extern void cpr_done(void);
55 extern void i_cpr_stop_other_cpus(void);
56 extern int i_cpr_power_down();
57 
58 #if defined(__sparc)
59 extern void cpr_forget_cprconfig(void);
60 #endif
61 
62 static struct modlmisc modlmisc = {
63 	&mod_miscops, "checkpoint resume"
64 };
65 
66 static struct modlinkage modlinkage = {
67 	MODREV_1, (void *)&modlmisc, NULL
68 };
69 
70 char _depends_on[] = "misc/bootdev";	/* i_devname_to_promname() */
71 
72 int cpr_reusable_mode;
73 
74 kmutex_t	cpr_slock;	/* cpr serial lock */
75 cpr_t		cpr_state;
76 int		cpr_debug;
77 int		cpr_test_mode; /* true if called via uadmin testmode */
78 int		cpr_test_point = LOOP_BACK_NONE;	/* cpr test point */
79 int		cpr_mp_enable = 0;	/* set to 1 to enable MP suspend */
80 major_t		cpr_device = 0;		/* major number for S3 on one device */
81 
82 /*
83  * All the loadable module related code follows
84  */
85 int
86 _init(void)
87 {
88 	register int e;
89 
90 	if ((e = mod_install(&modlinkage)) == 0) {
91 		mutex_init(&cpr_slock, NULL, MUTEX_DEFAULT, NULL);
92 	}
93 	return (e);
94 }
95 
96 int
97 _fini(void)
98 {
99 	register int e;
100 
101 	if ((e = mod_remove(&modlinkage)) == 0) {
102 		mutex_destroy(&cpr_slock);
103 	}
104 	return (e);
105 }
106 
107 int
108 _info(struct modinfo *modinfop)
109 {
110 	return (mod_info(&modlinkage, modinfop));
111 }
112 
113 static
114 int
115 atoi(char *p)
116 {
117 	int	i;
118 
119 	i = (*p++ - '0');
120 
121 	while (*p != '\0')
122 		i = 10 * i + (*p++ - '0');
123 
124 	return (i);
125 }
126 
127 int
128 cpr(int fcn, void *mdep)
129 {
130 
131 #if defined(__sparc)
132 	static const char noswapstr[] = "reusable statefile requires "
133 	    "that no swap area be configured.\n";
134 	static const char blockstr[] = "reusable statefile must be "
135 	    "a block device.  See power.conf(4) and pmconfig(1M).\n";
136 	static const char normalfmt[] = "cannot run normal "
137 	    "checkpoint/resume when in reusable statefile mode. "
138 	    "use uadmin A_FREEZE AD_REUSEFINI (uadmin %d %d) "
139 	    "to exit reusable statefile mode.\n";
140 	static const char modefmt[] = "%s in reusable mode.\n";
141 #endif
142 	register int rc = 0;
143 	int cpr_sleeptype;
144 
145 	/*
146 	 * First, reject commands that we don't (yet) support on this arch.
147 	 * This is easier to understand broken out like this than grotting
148 	 * through the second switch below.
149 	 */
150 
151 	switch (fcn) {
152 #if defined(__sparc)
153 	case AD_CHECK_SUSPEND_TO_RAM:
154 	case AD_SUSPEND_TO_RAM:
155 		return (ENOTSUP);
156 	case AD_CHECK_SUSPEND_TO_DISK:
157 	case AD_SUSPEND_TO_DISK:
158 	case AD_CPR_REUSEINIT:
159 	case AD_CPR_NOCOMPRESS:
160 	case AD_CPR_FORCE:
161 	case AD_CPR_REUSABLE:
162 	case AD_CPR_REUSEFINI:
163 	case AD_CPR_TESTZ:
164 	case AD_CPR_TESTNOZ:
165 	case AD_CPR_TESTHALT:
166 	case AD_CPR_SUSP_DEVICES:
167 		cpr_sleeptype = CPR_TODISK;
168 		break;
169 #endif
170 #if defined(__x86)
171 	case AD_CHECK_SUSPEND_TO_DISK:
172 	case AD_SUSPEND_TO_DISK:
173 	case AD_CPR_REUSEINIT:
174 	case AD_CPR_NOCOMPRESS:
175 	case AD_CPR_FORCE:
176 	case AD_CPR_REUSABLE:
177 	case AD_CPR_REUSEFINI:
178 	case AD_CPR_TESTZ:
179 	case AD_CPR_TESTNOZ:
180 	case AD_CPR_TESTHALT:
181 	case AD_CPR_PRINT:
182 		return (ENOTSUP);
183 	/* The DEV_* values need to be removed after sys-syspend is fixed */
184 	case DEV_CHECK_SUSPEND_TO_RAM:
185 	case DEV_SUSPEND_TO_RAM:
186 	case AD_CPR_SUSP_DEVICES:
187 	case AD_CHECK_SUSPEND_TO_RAM:
188 	case AD_SUSPEND_TO_RAM:
189 	case AD_LOOPBACK_SUSPEND_TO_RAM_PASS:
190 	case AD_LOOPBACK_SUSPEND_TO_RAM_FAIL:
191 	case AD_FORCE_SUSPEND_TO_RAM:
192 	case AD_DEVICE_SUSPEND_TO_RAM:
193 		cpr_sleeptype = CPR_TORAM;
194 		break;
195 #endif
196 	}
197 #if defined(__sparc)
198 	/*
199 	 * Need to know if we're in reusable mode, but we will likely have
200 	 * rebooted since REUSEINIT, so we have to get the info from the
201 	 * file system
202 	 */
203 	if (!cpr_reusable_mode)
204 		cpr_reusable_mode = cpr_get_reusable_mode();
205 
206 	cpr_forget_cprconfig();
207 #endif
208 
209 	switch (fcn) {
210 
211 #if defined(__sparc)
212 	case AD_CPR_REUSEINIT:
213 		if (!i_cpr_reusable_supported())
214 			return (ENOTSUP);
215 		if (!cpr_statefile_is_spec()) {
216 			cpr_err(CE_CONT, blockstr);
217 			return (EINVAL);
218 		}
219 		if ((rc = cpr_check_spec_statefile()) != 0)
220 			return (rc);
221 		if (swapinfo) {
222 			cpr_err(CE_CONT, noswapstr);
223 			return (EINVAL);
224 		}
225 		cpr_test_mode = 0;
226 		break;
227 
228 	case AD_CPR_NOCOMPRESS:
229 	case AD_CPR_COMPRESS:
230 	case AD_CPR_FORCE:
231 		if (cpr_reusable_mode) {
232 			cpr_err(CE_CONT, normalfmt, A_FREEZE, AD_REUSEFINI);
233 			return (ENOTSUP);
234 		}
235 		cpr_test_mode = 0;
236 		break;
237 
238 	case AD_CPR_REUSABLE:
239 		if (!i_cpr_reusable_supported())
240 			return (ENOTSUP);
241 		if (!cpr_statefile_is_spec()) {
242 			cpr_err(CE_CONT, blockstr);
243 			return (EINVAL);
244 		}
245 		if ((rc = cpr_check_spec_statefile()) != 0)
246 			return (rc);
247 		if (swapinfo) {
248 			cpr_err(CE_CONT, noswapstr);
249 			return (EINVAL);
250 		}
251 		if ((rc = cpr_reusable_mount_check()) != 0)
252 			return (rc);
253 		cpr_test_mode = 0;
254 		break;
255 
256 	case AD_CPR_REUSEFINI:
257 		if (!i_cpr_reusable_supported())
258 			return (ENOTSUP);
259 		cpr_test_mode = 0;
260 		break;
261 
262 	case AD_CPR_TESTZ:
263 	case AD_CPR_TESTNOZ:
264 	case AD_CPR_TESTHALT:
265 		if (cpr_reusable_mode) {
266 			cpr_err(CE_CONT, normalfmt, A_FREEZE, AD_REUSEFINI);
267 			return (ENOTSUP);
268 		}
269 		cpr_test_mode = 1;
270 		break;
271 
272 	case AD_CPR_CHECK:
273 		if (!i_cpr_is_supported(cpr_sleeptype) || cpr_reusable_mode)
274 			return (ENOTSUP);
275 		return (0);
276 
277 	case AD_CPR_PRINT:
278 		CPR_STAT_EVENT_END("POST CPR DELAY");
279 		cpr_stat_event_print();
280 		return (0);
281 #endif
282 
283 	case AD_CPR_DEBUG0:
284 		cpr_debug = 0;
285 		return (0);
286 
287 	case AD_CPR_DEBUG1:
288 	case AD_CPR_DEBUG2:
289 	case AD_CPR_DEBUG3:
290 	case AD_CPR_DEBUG4:
291 	case AD_CPR_DEBUG5:
292 	case AD_CPR_DEBUG7:
293 	case AD_CPR_DEBUG8:
294 		cpr_debug |= CPR_DEBUG_BIT(fcn);
295 		return (0);
296 
297 	case AD_CPR_DEBUG9:
298 		cpr_debug |= CPR_DEBUG6;
299 		return (0);
300 
301 	/* The DEV_* values need to be removed after sys-syspend is fixed */
302 	case DEV_CHECK_SUSPEND_TO_RAM:
303 	case DEV_SUSPEND_TO_RAM:
304 	case AD_CHECK_SUSPEND_TO_RAM:
305 	case AD_SUSPEND_TO_RAM:
306 		cpr_test_point = LOOP_BACK_NONE;
307 		break;
308 
309 	case AD_LOOPBACK_SUSPEND_TO_RAM_PASS:
310 		cpr_test_point = LOOP_BACK_PASS;
311 		break;
312 
313 	case AD_LOOPBACK_SUSPEND_TO_RAM_FAIL:
314 		cpr_test_point = LOOP_BACK_FAIL;
315 		break;
316 
317 	case AD_FORCE_SUSPEND_TO_RAM:
318 		cpr_test_point = FORCE_SUSPEND_TO_RAM;
319 		break;
320 
321 	case AD_DEVICE_SUSPEND_TO_RAM:
322 		if (mdep == NULL) {
323 			/* Didn't pass enough arguments */
324 			return (EINVAL);
325 		}
326 		cpr_test_point = DEVICE_SUSPEND_TO_RAM;
327 		cpr_device = (major_t)atoi((char *)mdep);
328 		break;
329 
330 	case AD_CPR_SUSP_DEVICES:
331 		cpr_test_point = FORCE_SUSPEND_TO_RAM;
332 		if (cpr_suspend_devices(ddi_root_node()) != DDI_SUCCESS)
333 			cmn_err(CE_WARN,
334 			    "Some devices did not suspend "
335 			    "and may be unusable");
336 		(void) cpr_resume_devices(ddi_root_node(), 0);
337 		return (0);
338 
339 	default:
340 		return (ENOTSUP);
341 	}
342 
343 	if (!i_cpr_is_supported(cpr_sleeptype) ||
344 	    (cpr_sleeptype == CPR_TODISK && !cpr_is_ufs(rootvfs)))
345 		return (ENOTSUP);
346 
347 	if (fcn == AD_CHECK_SUSPEND_TO_RAM ||
348 	    fcn == DEV_CHECK_SUSPEND_TO_RAM) {
349 		ASSERT(i_cpr_is_supported(cpr_sleeptype));
350 		return (0);
351 	}
352 
353 #if defined(__sparc)
354 	if (fcn == AD_CPR_REUSEINIT) {
355 		if (mutex_tryenter(&cpr_slock) == 0)
356 			return (EBUSY);
357 		if (cpr_reusable_mode) {
358 			cpr_err(CE_CONT, modefmt, "already");
359 			mutex_exit(&cpr_slock);
360 			return (EBUSY);
361 		}
362 		rc = i_cpr_reuseinit();
363 		mutex_exit(&cpr_slock);
364 		return (rc);
365 	}
366 
367 	if (fcn == AD_CPR_REUSEFINI) {
368 		if (mutex_tryenter(&cpr_slock) == 0)
369 			return (EBUSY);
370 		if (!cpr_reusable_mode) {
371 			cpr_err(CE_CONT, modefmt, "not");
372 			mutex_exit(&cpr_slock);
373 			return (EINVAL);
374 		}
375 		rc = i_cpr_reusefini();
376 		mutex_exit(&cpr_slock);
377 		return (rc);
378 	}
379 #endif
380 
381 	/*
382 	 * acquire cpr serial lock and init cpr state structure.
383 	 */
384 	if (rc = cpr_init(fcn))
385 		return (rc);
386 
387 #if defined(__sparc)
388 	if (fcn == AD_CPR_REUSABLE) {
389 		if ((rc = i_cpr_check_cprinfo()) != 0)  {
390 			mutex_exit(&cpr_slock);
391 			return (rc);
392 		}
393 	}
394 #endif
395 
396 	/*
397 	 * Call the main cpr routine. If we are successful, we will be coming
398 	 * down from the resume side, otherwise we are still in suspend.
399 	 */
400 	cpr_err(CE_CONT, "System is being suspended");
401 	if (rc = cpr_main(cpr_sleeptype)) {
402 		CPR->c_flags |= C_ERROR;
403 		PMD(PMD_SX, ("cpr: Suspend operation failed.\n"))
404 		cpr_err(CE_NOTE, "Suspend operation failed.");
405 	} else if (CPR->c_flags & C_SUSPENDING) {
406 
407 		/*
408 		 * In the suspend to RAM case, by the time we get
409 		 * control back we're already resumed
410 		 */
411 		if (cpr_sleeptype == CPR_TORAM) {
412 			PMD(PMD_SX, ("cpr: cpr CPR_TORAM done\n"))
413 			cpr_done();
414 			return (rc);
415 		}
416 
417 #if defined(__sparc)
418 
419 		PMD(PMD_SX, ("cpr: Suspend operation succeeded.\n"))
420 		/*
421 		 * Back from a successful checkpoint
422 		 */
423 		if (fcn == AD_CPR_TESTZ || fcn == AD_CPR_TESTNOZ) {
424 			mdboot(0, AD_BOOT, "", B_FALSE);
425 			/* NOTREACHED */
426 		}
427 
428 		/* make sure there are no more changes to the device tree */
429 		PMD(PMD_SX, ("cpr: dev tree freeze\n"))
430 		devtree_freeze();
431 
432 		/*
433 		 * stop other cpus and raise our priority.  since there is only
434 		 * one active cpu after this, and our priority will be too high
435 		 * for us to be preempted, we're essentially single threaded
436 		 * from here on out.
437 		 */
438 		PMD(PMD_SX, ("cpr: stop other cpus\n"))
439 		i_cpr_stop_other_cpus();
440 		PMD(PMD_SX, ("cpr: spl6\n"))
441 		(void) spl6();
442 
443 		/*
444 		 * try and reset leaf devices.  reset_leaves() should only
445 		 * be called when there are no other threads that could be
446 		 * accessing devices
447 		 */
448 		PMD(PMD_SX, ("cpr: reset leaves\n"))
449 		reset_leaves();
450 
451 		/*
452 		 * If i_cpr_power_down() succeeds, it'll not return
453 		 *
454 		 * Drives with write-cache enabled need to flush
455 		 * their cache.
456 		 */
457 		if (fcn != AD_CPR_TESTHALT) {
458 			PMD(PMD_SX, ("cpr: power down\n"))
459 			(void) i_cpr_power_down(cpr_sleeptype);
460 		}
461 		ASSERT(cpr_sleeptype == CPR_TODISK);
462 		/* currently CPR_TODISK comes back via a boot path */
463 		CPR_DEBUG(CPR_DEBUG1, "(Done. Please Switch Off)\n");
464 		halt(NULL);
465 		/* NOTREACHED */
466 #endif
467 	}
468 	PMD(PMD_SX, ("cpr: cpr done\n"))
469 	cpr_done();
470 	return (rc);
471 }
472