xref: /titanic_44/usr/src/uts/common/cpr/cpr_mod.c (revision d0538f66491267879b7418b21ad78e3dcc2dcc83)
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 2007 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 		/*
194 		 * if MP then do not support suspend to RAM, however override
195 		 * the MP restriction if cpr_mp_enable has been set
196 		 */
197 		if (ncpus > 1 && cpr_mp_enable == 0)
198 			return (ENOTSUP);
199 		else
200 			cpr_sleeptype = CPR_TORAM;
201 		break;
202 #endif
203 	}
204 #if defined(__sparc)
205 	/*
206 	 * Need to know if we're in reusable mode, but we will likely have
207 	 * rebooted since REUSEINIT, so we have to get the info from the
208 	 * file system
209 	 */
210 	if (!cpr_reusable_mode)
211 		cpr_reusable_mode = cpr_get_reusable_mode();
212 
213 	cpr_forget_cprconfig();
214 #endif
215 
216 	switch (fcn) {
217 
218 #if defined(__sparc)
219 	case AD_CPR_REUSEINIT:
220 		if (!i_cpr_reusable_supported())
221 			return (ENOTSUP);
222 		if (!cpr_statefile_is_spec()) {
223 			cpr_err(CE_CONT, blockstr);
224 			return (EINVAL);
225 		}
226 		if ((rc = cpr_check_spec_statefile()) != 0)
227 			return (rc);
228 		if (swapinfo) {
229 			cpr_err(CE_CONT, noswapstr);
230 			return (EINVAL);
231 		}
232 		cpr_test_mode = 0;
233 		break;
234 
235 	case AD_CPR_NOCOMPRESS:
236 	case AD_CPR_COMPRESS:
237 	case AD_CPR_FORCE:
238 		if (cpr_reusable_mode) {
239 			cpr_err(CE_CONT, normalfmt, A_FREEZE, AD_REUSEFINI);
240 			return (ENOTSUP);
241 		}
242 		cpr_test_mode = 0;
243 		break;
244 
245 	case AD_CPR_REUSABLE:
246 		if (!i_cpr_reusable_supported())
247 			return (ENOTSUP);
248 		if (!cpr_statefile_is_spec()) {
249 			cpr_err(CE_CONT, blockstr);
250 			return (EINVAL);
251 		}
252 		if ((rc = cpr_check_spec_statefile()) != 0)
253 			return (rc);
254 		if (swapinfo) {
255 			cpr_err(CE_CONT, noswapstr);
256 			return (EINVAL);
257 		}
258 		if ((rc = cpr_reusable_mount_check()) != 0)
259 			return (rc);
260 		cpr_test_mode = 0;
261 		break;
262 
263 	case AD_CPR_REUSEFINI:
264 		if (!i_cpr_reusable_supported())
265 			return (ENOTSUP);
266 		cpr_test_mode = 0;
267 		break;
268 
269 	case AD_CPR_TESTZ:
270 	case AD_CPR_TESTNOZ:
271 	case AD_CPR_TESTHALT:
272 		if (cpr_reusable_mode) {
273 			cpr_err(CE_CONT, normalfmt, A_FREEZE, AD_REUSEFINI);
274 			return (ENOTSUP);
275 		}
276 		cpr_test_mode = 1;
277 		break;
278 
279 	case AD_CPR_CHECK:
280 		if (!i_cpr_is_supported(cpr_sleeptype) || cpr_reusable_mode)
281 			return (ENOTSUP);
282 		return (0);
283 
284 	case AD_CPR_PRINT:
285 		CPR_STAT_EVENT_END("POST CPR DELAY");
286 		cpr_stat_event_print();
287 		return (0);
288 #endif
289 
290 	case AD_CPR_DEBUG0:
291 		cpr_debug = 0;
292 		return (0);
293 
294 	case AD_CPR_DEBUG1:
295 	case AD_CPR_DEBUG2:
296 	case AD_CPR_DEBUG3:
297 	case AD_CPR_DEBUG4:
298 	case AD_CPR_DEBUG5:
299 	case AD_CPR_DEBUG7:
300 	case AD_CPR_DEBUG8:
301 		cpr_debug |= CPR_DEBUG_BIT(fcn);
302 		return (0);
303 
304 	case AD_CPR_DEBUG9:
305 		cpr_debug |= CPR_DEBUG6;
306 		return (0);
307 
308 	/* The DEV_* values need to be removed after sys-syspend is fixed */
309 	case DEV_CHECK_SUSPEND_TO_RAM:
310 	case DEV_SUSPEND_TO_RAM:
311 	case AD_CHECK_SUSPEND_TO_RAM:
312 	case AD_SUSPEND_TO_RAM:
313 		cpr_test_point = LOOP_BACK_NONE;
314 		break;
315 
316 	case AD_LOOPBACK_SUSPEND_TO_RAM_PASS:
317 		cpr_test_point = LOOP_BACK_PASS;
318 		break;
319 
320 	case AD_LOOPBACK_SUSPEND_TO_RAM_FAIL:
321 		cpr_test_point = LOOP_BACK_FAIL;
322 		break;
323 
324 	case AD_FORCE_SUSPEND_TO_RAM:
325 		cpr_test_point = FORCE_SUSPEND_TO_RAM;
326 		break;
327 
328 	case AD_DEVICE_SUSPEND_TO_RAM:
329 		if (mdep == NULL) {
330 			/* Didn't pass enough arguments */
331 			return (EINVAL);
332 		}
333 		cpr_test_point = DEVICE_SUSPEND_TO_RAM;
334 		cpr_device = (major_t)atoi((char *)mdep);
335 		break;
336 
337 	case AD_CPR_SUSP_DEVICES:
338 		cpr_test_point = FORCE_SUSPEND_TO_RAM;
339 		if (cpr_suspend_devices(ddi_root_node()) != DDI_SUCCESS)
340 			cmn_err(CE_WARN,
341 			    "Some devices did not suspend "
342 			    "and may be unusable");
343 		(void) cpr_resume_devices(ddi_root_node(), 0);
344 		return (0);
345 
346 	default:
347 		return (ENOTSUP);
348 	}
349 
350 	if (!i_cpr_is_supported(cpr_sleeptype) ||
351 	    (cpr_sleeptype == CPR_TODISK && !cpr_is_ufs(rootvfs)))
352 		return (ENOTSUP);
353 
354 	if (fcn == AD_CHECK_SUSPEND_TO_RAM ||
355 	    fcn == DEV_CHECK_SUSPEND_TO_RAM) {
356 		ASSERT(i_cpr_is_supported(cpr_sleeptype));
357 		return (0);
358 	}
359 
360 #if defined(__sparc)
361 	if (fcn == AD_CPR_REUSEINIT) {
362 		if (mutex_tryenter(&cpr_slock) == 0)
363 			return (EBUSY);
364 		if (cpr_reusable_mode) {
365 			cpr_err(CE_CONT, modefmt, "already");
366 			mutex_exit(&cpr_slock);
367 			return (EBUSY);
368 		}
369 		rc = i_cpr_reuseinit();
370 		mutex_exit(&cpr_slock);
371 		return (rc);
372 	}
373 
374 	if (fcn == AD_CPR_REUSEFINI) {
375 		if (mutex_tryenter(&cpr_slock) == 0)
376 			return (EBUSY);
377 		if (!cpr_reusable_mode) {
378 			cpr_err(CE_CONT, modefmt, "not");
379 			mutex_exit(&cpr_slock);
380 			return (EINVAL);
381 		}
382 		rc = i_cpr_reusefini();
383 		mutex_exit(&cpr_slock);
384 		return (rc);
385 	}
386 #endif
387 
388 	/*
389 	 * acquire cpr serial lock and init cpr state structure.
390 	 */
391 	if (rc = cpr_init(fcn))
392 		return (rc);
393 
394 #if defined(__sparc)
395 	if (fcn == AD_CPR_REUSABLE) {
396 		if ((rc = i_cpr_check_cprinfo()) != 0)  {
397 			mutex_exit(&cpr_slock);
398 			return (rc);
399 		}
400 	}
401 #endif
402 
403 	/*
404 	 * Call the main cpr routine. If we are successful, we will be coming
405 	 * down from the resume side, otherwise we are still in suspend.
406 	 */
407 	cpr_err(CE_CONT, "System is being suspended");
408 	if (rc = cpr_main(cpr_sleeptype)) {
409 		CPR->c_flags |= C_ERROR;
410 		PMD(PMD_SX, ("cpr: Suspend operation failed.\n"))
411 		cpr_err(CE_NOTE, "Suspend operation failed.");
412 	} else if (CPR->c_flags & C_SUSPENDING) {
413 
414 		/*
415 		 * In the suspend to RAM case, by the time we get
416 		 * control back we're already resumed
417 		 */
418 		if (cpr_sleeptype == CPR_TORAM) {
419 			PMD(PMD_SX, ("cpr: cpr CPR_TORAM done\n"))
420 			cpr_done();
421 			return (rc);
422 		}
423 
424 #if defined(__sparc)
425 
426 		PMD(PMD_SX, ("cpr: Suspend operation succeeded.\n"))
427 		/*
428 		 * Back from a successful checkpoint
429 		 */
430 		if (fcn == AD_CPR_TESTZ || fcn == AD_CPR_TESTNOZ) {
431 			mdboot(0, AD_BOOT, "", B_FALSE);
432 			/* NOTREACHED */
433 		}
434 
435 		/* make sure there are no more changes to the device tree */
436 		PMD(PMD_SX, ("cpr: dev tree freeze\n"))
437 		devtree_freeze();
438 
439 		/*
440 		 * stop other cpus and raise our priority.  since there is only
441 		 * one active cpu after this, and our priority will be too high
442 		 * for us to be preempted, we're essentially single threaded
443 		 * from here on out.
444 		 */
445 		PMD(PMD_SX, ("cpr: stop other cpus\n"))
446 		i_cpr_stop_other_cpus();
447 		PMD(PMD_SX, ("cpr: spl6\n"))
448 		(void) spl6();
449 
450 		/*
451 		 * try and reset leaf devices.  reset_leaves() should only
452 		 * be called when there are no other threads that could be
453 		 * accessing devices
454 		 */
455 		PMD(PMD_SX, ("cpr: reset leaves\n"))
456 		reset_leaves();
457 
458 		/*
459 		 * If i_cpr_power_down() succeeds, it'll not return
460 		 *
461 		 * Drives with write-cache enabled need to flush
462 		 * their cache.
463 		 */
464 		if (fcn != AD_CPR_TESTHALT) {
465 			PMD(PMD_SX, ("cpr: power down\n"))
466 			(void) i_cpr_power_down(cpr_sleeptype);
467 		}
468 		ASSERT(cpr_sleeptype == CPR_TODISK);
469 		/* currently CPR_TODISK comes back via a boot path */
470 		CPR_DEBUG(CPR_DEBUG1, "(Done. Please Switch Off)\n");
471 		halt(NULL);
472 		/* NOTREACHED */
473 #endif
474 	}
475 	PMD(PMD_SX, ("cpr: cpr done\n"))
476 	cpr_done();
477 	return (rc);
478 }
479