xref: /titanic_41/usr/src/uts/common/os/main.c (revision 9fb67ea305c66b6a297583b9b0db6796b0dfe497)
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 /*
23  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 /*	Copyright (c) 1988 AT&T	*/
28 /*	  All Rights Reserved  	*/
29 
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/sysmacros.h>
33 #include <sys/pcb.h>
34 #include <sys/systm.h>
35 #include <sys/signal.h>
36 #include <sys/cred.h>
37 #include <sys/user.h>
38 #include <sys/vfs.h>
39 #include <sys/vnode.h>
40 #include <sys/proc.h>
41 #include <sys/time.h>
42 #include <sys/file.h>
43 #include <sys/priocntl.h>
44 #include <sys/procset.h>
45 #include <sys/disp.h>
46 #include <sys/callo.h>
47 #include <sys/callb.h>
48 #include <sys/debug.h>
49 #include <sys/conf.h>
50 #include <sys/bootconf.h>
51 #include <sys/utsname.h>
52 #include <sys/cmn_err.h>
53 #include <sys/vmparam.h>
54 #include <sys/modctl.h>
55 #include <sys/vm.h>
56 #include <sys/callb.h>
57 #include <sys/ddi_timer.h>
58 #include <sys/kmem.h>
59 #include <sys/vmem.h>
60 #include <sys/cpuvar.h>
61 #include <sys/cladm.h>
62 #include <sys/corectl.h>
63 #include <sys/exec.h>
64 #include <sys/syscall.h>
65 #include <sys/reboot.h>
66 #include <sys/task.h>
67 #include <sys/exacct.h>
68 #include <sys/autoconf.h>
69 #include <sys/errorq.h>
70 #include <sys/class.h>
71 #include <sys/stack.h>
72 #include <sys/brand.h>
73 #include <sys/mmapobj.h>
74 
75 #include <vm/as.h>
76 #include <vm/seg_kmem.h>
77 #include <sys/dc_ki.h>
78 
79 #include <c2/audit.h>
80 #include <sys/bootprops.h>
81 
82 /* well known processes */
83 proc_t *proc_sched;		/* memory scheduler */
84 proc_t *proc_init;		/* init */
85 proc_t *proc_pageout;		/* pageout daemon */
86 proc_t *proc_fsflush;		/* fsflush daemon */
87 
88 pgcnt_t	maxmem;		/* Maximum available memory in pages.	*/
89 pgcnt_t	freemem;	/* Current available memory in pages.	*/
90 int	interrupts_unleashed;	/* set when we do the first spl0() */
91 
92 kmem_cache_t *process_cache;	/* kmem cache for proc structures */
93 
94 /*
95  * Indicates whether the auditing module (c2audit) is loaded. Possible
96  * values are:
97  * 0 - c2audit module is excluded in /etc/system and cannot be loaded
98  * 1 - c2audit module is not loaded but can be anytime
99  * 2 - c2audit module is loaded
100  */
101 int audit_active = C2AUDIT_DISABLED;
102 
103 /*
104  * Process 0's lwp directory and lwpid hash table.
105  */
106 lwpdir_t p0_lwpdir[2];
107 tidhash_t p0_tidhash[2];
108 lwpent_t p0_lep;
109 
110 /*
111  * Machine-independent initialization code
112  * Called from cold start routine as
113  * soon as a stack and segmentation
114  * have been established.
115  * Functions:
116  *	clear and free user core
117  *	turn on clock
118  *	hand craft 0th process
119  *	call all initialization routines
120  *	fork	- process 0 to schedule
121  *		- process 1 execute bootstrap
122  *		- process 2 to page out
123  *	create system threads
124  */
125 
126 int cluster_bootflags = 0;
127 
128 void
129 cluster_wrapper(void)
130 {
131 	cluster();
132 	panic("cluster()  returned");
133 }
134 
135 char initname[INITNAME_SZ] = "/sbin/init";	/* also referenced by zone0 */
136 char initargs[BOOTARGS_MAX] = "";		/* also referenced by zone0 */
137 
138 /*
139  * Construct a stack for init containing the arguments to it, then
140  * pass control to exec_common.
141  */
142 int
143 exec_init(const char *initpath, const char *args)
144 {
145 	caddr32_t ucp;
146 	caddr32_t *uap;
147 	caddr32_t *argv;
148 	caddr32_t exec_fnamep;
149 	char *scratchargs;
150 	int i, sarg;
151 	size_t argvlen, alen;
152 	boolean_t in_arg;
153 	int argc = 0;
154 	int error = 0, count = 0;
155 	proc_t *p = ttoproc(curthread);
156 	klwp_t *lwp = ttolwp(curthread);
157 	int brand_action;
158 
159 	if (args == NULL)
160 		args = "";
161 
162 	alen = strlen(initpath) + 1 + strlen(args) + 1;
163 	scratchargs = kmem_alloc(alen, KM_SLEEP);
164 	(void) snprintf(scratchargs, alen, "%s %s", initpath, args);
165 
166 	/*
167 	 * We do a quick two state parse of the string to sort out how big
168 	 * argc should be.
169 	 */
170 	in_arg = B_FALSE;
171 	for (i = 0; i < strlen(scratchargs); i++) {
172 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
173 			if (in_arg) {
174 				in_arg = B_FALSE;
175 				argc++;
176 			}
177 		} else {
178 			in_arg = B_TRUE;
179 		}
180 	}
181 	argvlen = sizeof (caddr32_t) * (argc + 1);
182 	argv = kmem_zalloc(argvlen, KM_SLEEP);
183 
184 	/*
185 	 * We pull off a bit of a hack here.  We work our way through the
186 	 * args string, putting nulls at the ends of space delimited tokens
187 	 * (boot args don't support quoting at this time).  Then we just
188 	 * copy the whole mess to userland in one go.  In other words, we
189 	 * transform this: "init -s -r\0" into this on the stack:
190 	 *
191 	 *	-0x00 \0
192 	 *	-0x01 r
193 	 *	-0x02 -  <--------.
194 	 *	-0x03 \0	  |
195 	 *	-0x04 s		  |
196 	 *	-0x05 -  <------. |
197 	 *	-0x06 \0	| |
198 	 *	-0x07 t		| |
199 	 *	-0x08 i 	| |
200 	 *	-0x09 n		| |
201 	 *	-0x0a i  <---.  | |
202 	 *	-0x10 NULL   |  | |	(argv[3])
203 	 *	-0x14   -----|--|-'	(argv[2])
204 	 *	-0x18  ------|--'	(argv[1])
205 	 *	-0x1c -------'		(argv[0])
206 	 *
207 	 * Since we know the value of ucp at the beginning of this process,
208 	 * we can trivially compute the argv[] array which we also need to
209 	 * place in userland: argv[i] = ucp - sarg(i), where ucp is the
210 	 * stack ptr, and sarg is the string index of the start of the
211 	 * argument.
212 	 */
213 	ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
214 
215 	argc = 0;
216 	in_arg = B_FALSE;
217 	sarg = 0;
218 
219 	for (i = 0; i < alen; i++) {
220 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
221 			if (in_arg == B_TRUE) {
222 				in_arg = B_FALSE;
223 				scratchargs[i] = '\0';
224 				argv[argc++] = ucp - (alen - sarg);
225 			}
226 		} else if (in_arg == B_FALSE) {
227 			in_arg = B_TRUE;
228 			sarg = i;
229 		}
230 	}
231 	ucp -= alen;
232 	error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
233 
234 	uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
235 	uap--;	/* advance to be below the word we're in */
236 	uap -= (argc + 1);	/* advance argc words down, plus one for NULL */
237 	error |= copyout(argv, uap, argvlen);
238 
239 	if (error != 0) {
240 		zcmn_err(p->p_zone->zone_id, CE_WARN,
241 		    "Could not construct stack for init.\n");
242 		kmem_free(argv, argvlen);
243 		kmem_free(scratchargs, alen);
244 		return (EFAULT);
245 	}
246 
247 	exec_fnamep = argv[0];
248 	kmem_free(argv, argvlen);
249 	kmem_free(scratchargs, alen);
250 
251 	/*
252 	 * Point at the arguments.
253 	 */
254 	lwp->lwp_ap = lwp->lwp_arg;
255 	lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
256 	lwp->lwp_arg[1] = (uintptr_t)uap;
257 	lwp->lwp_arg[2] = NULL;
258 	curthread->t_post_sys = 1;
259 	curthread->t_sysnum = SYS_execve;
260 
261 	/*
262 	 * If we are executing init from zsched, we may have inherited its
263 	 * parent process's signal mask.  Clear it now so that we behave in
264 	 * the same way as when started from the global zone.
265 	 */
266 	sigemptyset(&curthread->t_hold);
267 
268 	brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
269 again:
270 	error = exec_common((const char *)(uintptr_t)exec_fnamep,
271 	    (const char **)(uintptr_t)uap, NULL, brand_action);
272 
273 	/*
274 	 * Normally we would just set lwp_argsaved and t_post_sys and
275 	 * let post_syscall reset lwp_ap for us.  Unfortunately,
276 	 * exec_init isn't always called from a system call.  Instead
277 	 * of making a mess of trap_cleanup, we just reset the args
278 	 * pointer here.
279 	 */
280 	reset_syscall_args();
281 
282 	switch (error) {
283 	case 0:
284 		return (0);
285 
286 	case ENOENT:
287 		zcmn_err(p->p_zone->zone_id, CE_WARN,
288 		    "exec(%s) failed (file not found).\n", initpath);
289 		return (ENOENT);
290 
291 	case EAGAIN:
292 	case EINTR:
293 		++count;
294 		if (count < 5) {
295 			zcmn_err(p->p_zone->zone_id, CE_WARN,
296 			    "exec(%s) failed with errno %d.  Retrying...\n",
297 			    initpath, error);
298 			goto again;
299 		}
300 	}
301 
302 	zcmn_err(p->p_zone->zone_id, CE_WARN,
303 	    "exec(%s) failed with errno %d.", initpath, error);
304 	return (error);
305 }
306 
307 /*
308  * This routine does all of the common setup for invoking init; global
309  * and non-global zones employ this routine for the functionality which is
310  * in common.
311  *
312  * This program (init, presumably) must be a 32-bit process.
313  */
314 int
315 start_init_common()
316 {
317 	proc_t *p = curproc;
318 	ASSERT_STACK_ALIGNED();
319 	p->p_zone->zone_proc_initpid = p->p_pid;
320 
321 	p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
322 	p->p_usrstack = (caddr_t)USRSTACK32;
323 	p->p_model = DATAMODEL_ILP32;
324 	p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
325 	p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
326 	p->p_stk_ctl = INT32_MAX;
327 
328 	p->p_as = as_alloc();
329 	p->p_as->a_proc = p;
330 	p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
331 	(void) hat_setup(p->p_as->a_hat, HAT_INIT);
332 
333 	init_core();
334 
335 	init_mstate(curthread, LMS_SYSTEM);
336 	return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
337 }
338 
339 /*
340  * Start the initial user process for the global zone; once running, if
341  * init should subsequently fail, it will be automatically be caught in the
342  * exit(2) path, and restarted by restart_init().
343  */
344 static void
345 start_init(void)
346 {
347 	proc_init = curproc;
348 
349 	ASSERT(curproc->p_zone->zone_initname != NULL);
350 
351 	if (start_init_common() != 0)
352 		halt("unix: Could not start init");
353 	lwp_rtt();
354 }
355 
356 void
357 main(void)
358 {
359 	proc_t		*p = ttoproc(curthread);	/* &p0 */
360 	int		(**initptr)();
361 	extern void	sched();
362 	extern void	fsflush();
363 	extern int	(*init_tbl[])();
364 	extern int	(*mp_init_tbl[])();
365 	extern id_t	syscid, defaultcid;
366 	extern int	swaploaded;
367 	extern int	netboot;
368 	extern ib_boot_prop_t *iscsiboot_prop;
369 	extern void	vm_init(void);
370 	extern void	cbe_init_pre(void);
371 	extern void	cbe_init(void);
372 	extern void	clock_tick_init_pre(void);
373 	extern void	clock_tick_init_post(void);
374 	extern void	clock_init(void);
375 	extern void	physio_bufs_init(void);
376 	extern void	pm_cfb_setup_intr(void);
377 	extern int	pm_adjust_timestamps(dev_info_t *, void *);
378 	extern void	start_other_cpus(int);
379 	extern void	sysevent_evc_thrinit();
380 #if defined(__x86)
381 	extern void	fastboot_post_startup(void);
382 	extern void	progressbar_start(void);
383 #endif
384 	/*
385 	 * In the horrible world of x86 in-lines, you can't get symbolic
386 	 * structure offsets a la genassym.  This assertion is here so
387 	 * that the next poor slob who innocently changes the offset of
388 	 * cpu_thread doesn't waste as much time as I just did finding
389 	 * out that it's hard-coded in i86/ml/i86.il.  Similarly for
390 	 * curcpup.  You're welcome.
391 	 */
392 	ASSERT(CPU == CPU->cpu_self);
393 	ASSERT(curthread == CPU->cpu_thread);
394 	ASSERT_STACK_ALIGNED();
395 
396 	/*
397 	 * Setup root lgroup and leaf lgroup for CPU 0
398 	 */
399 	lgrp_init(LGRP_INIT_STAGE2);
400 
401 	/*
402 	 * Once 'startup()' completes, the thread_reaper() daemon would be
403 	 * created(in thread_init()). After that, it is safe to create threads
404 	 * that could exit. These exited threads will get reaped.
405 	 */
406 	startup();
407 	segkmem_gc();
408 	callb_init();
409 	cbe_init_pre();	/* x86 must initialize gethrtimef before timer_init */
410 	timer_init();	/* timer must be initialized before cyclic starts */
411 	cbe_init();
412 	callout_init();	/* callout table MUST be init'd after cyclics */
413 	clock_tick_init_pre();
414 	clock_init();
415 
416 #if defined(__x86)
417 	/*
418 	 * The progressbar thread uses cv_reltimedwait() and hence needs to be
419 	 * started after the callout mechanism has been initialized.
420 	 */
421 	progressbar_start();
422 #endif
423 	/*
424 	 * On some platforms, clkinitf() changes the timing source that
425 	 * gethrtime_unscaled() uses to generate timestamps.  cbe_init() calls
426 	 * clkinitf(), so re-initialize the microstate counters after the
427 	 * timesource has been chosen.
428 	 */
429 	init_mstate(&t0, LMS_SYSTEM);
430 	init_cpu_mstate(CPU, CMS_SYSTEM);
431 
432 	/*
433 	 * May need to probe to determine latencies from CPU 0 after
434 	 * gethrtime() comes alive in cbe_init() and before enabling interrupts
435 	 * and copy and release any temporary memory allocated with BOP_ALLOC()
436 	 * before release_bootstrap() frees boot memory
437 	 */
438 	lgrp_init(LGRP_INIT_STAGE3);
439 
440 	/*
441 	 * Call all system initialization functions.
442 	 */
443 	for (initptr = &init_tbl[0]; *initptr; initptr++)
444 		(**initptr)();
445 	/*
446 	 * Load iSCSI boot properties
447 	 */
448 	ld_ib_prop();
449 	/*
450 	 * initialize vm related stuff.
451 	 */
452 	vm_init();
453 
454 	/*
455 	 * initialize buffer pool for raw I/O requests
456 	 */
457 	physio_bufs_init();
458 
459 	ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
460 
461 	/*
462 	 * Drop the interrupt level and allow interrupts.  At this point
463 	 * the DDI guarantees that interrupts are enabled.
464 	 */
465 	(void) spl0();
466 	interrupts_unleashed = 1;
467 
468 	/*
469 	 * Create kmem cache for proc structures
470 	 */
471 	process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
472 	    0, NULL, NULL, NULL, NULL, NULL, 0);
473 
474 	vfs_mountroot();	/* Mount the root file system */
475 	errorq_init();		/* after vfs_mountroot() so DDI root is ready */
476 	cpu_kstat_init(CPU);	/* after vfs_mountroot() so TOD is valid */
477 	ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
478 				/* after vfs_mountroot() so hrestime is valid */
479 
480 	post_startup();
481 	swaploaded = 1;
482 
483 	/*
484 	 * Initialize Solaris Audit Subsystem
485 	 */
486 	audit_init();
487 
488 	/*
489 	 * Plumb the protocol modules and drivers only if we are not
490 	 * networked booted, in this case we already did it in rootconf().
491 	 */
492 	if (netboot == 0 && iscsiboot_prop == NULL)
493 		(void) strplumb();
494 
495 	gethrestime(&PTOU(curproc)->u_start);
496 	curthread->t_start = PTOU(curproc)->u_start.tv_sec;
497 	p->p_mstart = gethrtime();
498 
499 	/*
500 	 * Perform setup functions that can only be done after root
501 	 * and swap have been set up.
502 	 */
503 	consconfig();
504 #ifndef	__sparc
505 	release_bootstrap();
506 #endif
507 
508 	/*
509 	 * attach drivers with ddi-forceattach prop
510 	 * It must be done early enough to load hotplug drivers (e.g.
511 	 * pcmcia nexus) so that devices enumerated via hotplug is
512 	 * available before I/O subsystem is fully initialized.
513 	 */
514 	i_ddi_forceattach_drivers();
515 
516 	/*
517 	 * Set the scan rate and other parameters of the paging subsystem.
518 	 */
519 	setupclock(0);
520 
521 	/*
522 	 * Initialize process 0's lwp directory and lwpid hash table.
523 	 */
524 	p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
525 	p->p_lwpdir->ld_next = p->p_lwpdir + 1;
526 	p->p_lwpdir_sz = 2;
527 	p->p_tidhash = p0_tidhash;
528 	p->p_tidhash_sz = 2;
529 	p0_lep.le_thread = curthread;
530 	p0_lep.le_lwpid = curthread->t_tid;
531 	p0_lep.le_start = curthread->t_start;
532 	lwp_hash_in(p, &p0_lep, p0_tidhash, 2, 0);
533 
534 	/*
535 	 * Initialize extended accounting.
536 	 */
537 	exacct_init();
538 
539 	/*
540 	 * Initialize threads of sysevent event channels
541 	 */
542 	sysevent_evc_thrinit();
543 
544 	/*
545 	 * This must be done after post_startup() but before
546 	 * start_other_cpus()
547 	 */
548 	lgrp_init(LGRP_INIT_STAGE4);
549 
550 	/*
551 	 * Perform MP initialization, if any.
552 	 */
553 	start_other_cpus(0);
554 
555 #ifdef	__sparc
556 	/*
557 	 * Release bootstrap here since PROM interfaces are
558 	 * used to start other CPUs above.
559 	 */
560 	release_bootstrap();
561 #endif
562 
563 	/*
564 	 * Finish lgrp initialization after all CPUS are brought online.
565 	 */
566 	lgrp_init(LGRP_INIT_STAGE5);
567 
568 	/*
569 	 * After mp_init(), number of cpus are known (this is
570 	 * true for the time being, when there are actually
571 	 * hot pluggable cpus then this scheme  would not do).
572 	 * Any per cpu initialization is done here.
573 	 */
574 	kmem_mp_init();
575 	vmem_update(NULL);
576 
577 	clock_tick_init_post();
578 
579 	for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
580 		(**initptr)();
581 
582 	/*
583 	 * These must be called after start_other_cpus
584 	 */
585 	pm_cfb_setup_intr();
586 #if defined(__x86)
587 	fastboot_post_startup();
588 #endif
589 
590 	/*
591 	 * Make init process; enter scheduling loop with system process.
592 	 *
593 	 * Note that we manually assign the pids for these processes, for
594 	 * historical reasons.  If more pre-assigned pids are needed,
595 	 * FAMOUS_PIDS will have to be updated.
596 	 */
597 
598 	/* create init process */
599 	if (newproc(start_init, NULL, defaultcid, 59, NULL,
600 	    FAMOUS_PID_INIT))
601 		panic("main: unable to fork init.");
602 
603 	/* create pageout daemon */
604 	if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL,
605 	    FAMOUS_PID_PAGEOUT))
606 		panic("main: unable to fork pageout()");
607 
608 	/* create fsflush daemon */
609 	if (newproc(fsflush, NULL, syscid, minclsyspri, NULL,
610 	    FAMOUS_PID_FSFLUSH))
611 		panic("main: unable to fork fsflush()");
612 
613 	/* create cluster process if we're a member of one */
614 	if (cluster_bootflags & CLUSTER_BOOTED) {
615 		if (newproc(cluster_wrapper, NULL, syscid, minclsyspri,
616 		    NULL, 0)) {
617 			panic("main: unable to fork cluster()");
618 		}
619 	}
620 
621 	/*
622 	 * Create system threads (threads are associated with p0)
623 	 */
624 
625 	/* create module uninstall daemon */
626 	/* BugID 1132273. If swapping over NFS need a bigger stack */
627 	(void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
628 	    NULL, 0, &p0, TS_RUN, minclsyspri);
629 
630 	(void) thread_create(NULL, 0, seg_pasync_thread,
631 	    NULL, 0, &p0, TS_RUN, minclsyspri);
632 
633 	pid_setmin();
634 
635 	bcopy("sched", PTOU(curproc)->u_psargs, 6);
636 	bcopy("sched", PTOU(curproc)->u_comm, 5);
637 	sched();
638 	/* NOTREACHED */
639 }
640