xref: /titanic_44/usr/src/uts/common/os/main.c (revision 0dc2366f7b9f9f36e10909b1e95edbf2a261c2ac)
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 extern int64_t lwp_sigmask(int, uint_t, uint_t);
138 
139 /*
140  * Construct a stack for init containing the arguments to it, then
141  * pass control to exec_common.
142  */
143 int
144 exec_init(const char *initpath, const char *args)
145 {
146 	caddr32_t ucp;
147 	caddr32_t *uap;
148 	caddr32_t *argv;
149 	caddr32_t exec_fnamep;
150 	char *scratchargs;
151 	int i, sarg;
152 	size_t argvlen, alen;
153 	boolean_t in_arg;
154 	int argc = 0;
155 	int error = 0, count = 0;
156 	proc_t *p = ttoproc(curthread);
157 	klwp_t *lwp = ttolwp(curthread);
158 	int brand_action;
159 
160 	if (args == NULL)
161 		args = "";
162 
163 	alen = strlen(initpath) + 1 + strlen(args) + 1;
164 	scratchargs = kmem_alloc(alen, KM_SLEEP);
165 	(void) snprintf(scratchargs, alen, "%s %s", initpath, args);
166 
167 	/*
168 	 * We do a quick two state parse of the string to sort out how big
169 	 * argc should be.
170 	 */
171 	in_arg = B_FALSE;
172 	for (i = 0; i < strlen(scratchargs); i++) {
173 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
174 			if (in_arg) {
175 				in_arg = B_FALSE;
176 				argc++;
177 			}
178 		} else {
179 			in_arg = B_TRUE;
180 		}
181 	}
182 	argvlen = sizeof (caddr32_t) * (argc + 1);
183 	argv = kmem_zalloc(argvlen, KM_SLEEP);
184 
185 	/*
186 	 * We pull off a bit of a hack here.  We work our way through the
187 	 * args string, putting nulls at the ends of space delimited tokens
188 	 * (boot args don't support quoting at this time).  Then we just
189 	 * copy the whole mess to userland in one go.  In other words, we
190 	 * transform this: "init -s -r\0" into this on the stack:
191 	 *
192 	 *	-0x00 \0
193 	 *	-0x01 r
194 	 *	-0x02 -  <--------.
195 	 *	-0x03 \0	  |
196 	 *	-0x04 s		  |
197 	 *	-0x05 -  <------. |
198 	 *	-0x06 \0	| |
199 	 *	-0x07 t		| |
200 	 *	-0x08 i 	| |
201 	 *	-0x09 n		| |
202 	 *	-0x0a i  <---.  | |
203 	 *	-0x10 NULL   |  | |	(argv[3])
204 	 *	-0x14   -----|--|-'	(argv[2])
205 	 *	-0x18  ------|--'	(argv[1])
206 	 *	-0x1c -------'		(argv[0])
207 	 *
208 	 * Since we know the value of ucp at the beginning of this process,
209 	 * we can trivially compute the argv[] array which we also need to
210 	 * place in userland: argv[i] = ucp - sarg(i), where ucp is the
211 	 * stack ptr, and sarg is the string index of the start of the
212 	 * argument.
213 	 */
214 	ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
215 
216 	argc = 0;
217 	in_arg = B_FALSE;
218 	sarg = 0;
219 
220 	for (i = 0; i < alen; i++) {
221 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
222 			if (in_arg == B_TRUE) {
223 				in_arg = B_FALSE;
224 				scratchargs[i] = '\0';
225 				argv[argc++] = ucp - (alen - sarg);
226 			}
227 		} else if (in_arg == B_FALSE) {
228 			in_arg = B_TRUE;
229 			sarg = i;
230 		}
231 	}
232 	ucp -= alen;
233 	error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
234 
235 	uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
236 	uap--;	/* advance to be below the word we're in */
237 	uap -= (argc + 1);	/* advance argc words down, plus one for NULL */
238 	error |= copyout(argv, uap, argvlen);
239 
240 	if (error != 0) {
241 		zcmn_err(p->p_zone->zone_id, CE_WARN,
242 		    "Could not construct stack for init.\n");
243 		kmem_free(argv, argvlen);
244 		kmem_free(scratchargs, alen);
245 		return (EFAULT);
246 	}
247 
248 	exec_fnamep = argv[0];
249 	kmem_free(argv, argvlen);
250 	kmem_free(scratchargs, alen);
251 
252 	/*
253 	 * Point at the arguments.
254 	 */
255 	lwp->lwp_ap = lwp->lwp_arg;
256 	lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
257 	lwp->lwp_arg[1] = (uintptr_t)uap;
258 	lwp->lwp_arg[2] = NULL;
259 	curthread->t_post_sys = 1;
260 	curthread->t_sysnum = SYS_execve;
261 
262 	/*
263 	 * If we are executing init from zsched, we may have inherited its
264 	 * parent process's signal mask.  Clear it now so that we behave in
265 	 * the same way as when started from the global zone.
266 	 */
267 	(void) lwp_sigmask(SIG_UNBLOCK, 0xffffffff, 0xffffffff);
268 
269 	brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
270 again:
271 	error = exec_common((const char *)(uintptr_t)exec_fnamep,
272 	    (const char **)(uintptr_t)uap, NULL, brand_action);
273 
274 	/*
275 	 * Normally we would just set lwp_argsaved and t_post_sys and
276 	 * let post_syscall reset lwp_ap for us.  Unfortunately,
277 	 * exec_init isn't always called from a system call.  Instead
278 	 * of making a mess of trap_cleanup, we just reset the args
279 	 * pointer here.
280 	 */
281 	reset_syscall_args();
282 
283 	switch (error) {
284 	case 0:
285 		return (0);
286 
287 	case ENOENT:
288 		zcmn_err(p->p_zone->zone_id, CE_WARN,
289 		    "exec(%s) failed (file not found).\n", initpath);
290 		return (ENOENT);
291 
292 	case EAGAIN:
293 	case EINTR:
294 		++count;
295 		if (count < 5) {
296 			zcmn_err(p->p_zone->zone_id, CE_WARN,
297 			    "exec(%s) failed with errno %d.  Retrying...\n",
298 			    initpath, error);
299 			goto again;
300 		}
301 	}
302 
303 	zcmn_err(p->p_zone->zone_id, CE_WARN,
304 	    "exec(%s) failed with errno %d.", initpath, error);
305 	return (error);
306 }
307 
308 /*
309  * This routine does all of the common setup for invoking init; global
310  * and non-global zones employ this routine for the functionality which is
311  * in common.
312  *
313  * This program (init, presumably) must be a 32-bit process.
314  */
315 int
316 start_init_common()
317 {
318 	proc_t *p = curproc;
319 	ASSERT_STACK_ALIGNED();
320 	p->p_zone->zone_proc_initpid = p->p_pid;
321 
322 	p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
323 	p->p_usrstack = (caddr_t)USRSTACK32;
324 	p->p_model = DATAMODEL_ILP32;
325 	p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
326 	p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
327 	p->p_stk_ctl = INT32_MAX;
328 
329 	p->p_as = as_alloc();
330 	p->p_as->a_proc = p;
331 	p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
332 	(void) hat_setup(p->p_as->a_hat, HAT_INIT);
333 
334 	init_core();
335 
336 	init_mstate(curthread, LMS_SYSTEM);
337 	return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
338 }
339 
340 /*
341  * Start the initial user process for the global zone; once running, if
342  * init should subsequently fail, it will be automatically be caught in the
343  * exit(2) path, and restarted by restart_init().
344  */
345 static void
346 start_init(void)
347 {
348 	proc_init = curproc;
349 
350 	ASSERT(curproc->p_zone->zone_initname != NULL);
351 
352 	if (start_init_common() != 0)
353 		halt("unix: Could not start init");
354 	lwp_rtt();
355 }
356 
357 void
358 main(void)
359 {
360 	proc_t		*p = ttoproc(curthread);	/* &p0 */
361 	int		(**initptr)();
362 	extern void	sched();
363 	extern void	fsflush();
364 	extern int	(*init_tbl[])();
365 	extern int	(*mp_init_tbl[])();
366 	extern id_t	syscid, defaultcid;
367 	extern int	swaploaded;
368 	extern int	netboot;
369 	extern ib_boot_prop_t *iscsiboot_prop;
370 	extern void	vm_init(void);
371 	extern void	cbe_init_pre(void);
372 	extern void	cbe_init(void);
373 	extern void	clock_tick_init_pre(void);
374 	extern void	clock_tick_init_post(void);
375 	extern void	clock_init(void);
376 	extern void	physio_bufs_init(void);
377 	extern void	pm_cfb_setup_intr(void);
378 	extern int	pm_adjust_timestamps(dev_info_t *, void *);
379 	extern void	start_other_cpus(int);
380 	extern void	sysevent_evc_thrinit();
381 #if defined(__x86)
382 	extern void	fastboot_post_startup(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 	/*
417 	 * On some platforms, clkinitf() changes the timing source that
418 	 * gethrtime_unscaled() uses to generate timestamps.  cbe_init() calls
419 	 * clkinitf(), so re-initialize the microstate counters after the
420 	 * timesource has been chosen.
421 	 */
422 	init_mstate(&t0, LMS_SYSTEM);
423 	init_cpu_mstate(CPU, CMS_SYSTEM);
424 
425 	/*
426 	 * May need to probe to determine latencies from CPU 0 after
427 	 * gethrtime() comes alive in cbe_init() and before enabling interrupts
428 	 * and copy and release any temporary memory allocated with BOP_ALLOC()
429 	 * before release_bootstrap() frees boot memory
430 	 */
431 	lgrp_init(LGRP_INIT_STAGE3);
432 
433 	/*
434 	 * Call all system initialization functions.
435 	 */
436 	for (initptr = &init_tbl[0]; *initptr; initptr++)
437 		(**initptr)();
438 	/*
439 	 * Load iSCSI boot properties
440 	 */
441 	ld_ib_prop();
442 	/*
443 	 * initialize vm related stuff.
444 	 */
445 	vm_init();
446 
447 	/*
448 	 * initialize buffer pool for raw I/O requests
449 	 */
450 	physio_bufs_init();
451 
452 	ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
453 
454 	/*
455 	 * Drop the interrupt level and allow interrupts.  At this point
456 	 * the DDI guarantees that interrupts are enabled.
457 	 */
458 	(void) spl0();
459 	interrupts_unleashed = 1;
460 
461 	/*
462 	 * Create kmem cache for proc structures
463 	 */
464 	process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
465 	    0, NULL, NULL, NULL, NULL, NULL, 0);
466 
467 	vfs_mountroot();	/* Mount the root file system */
468 	errorq_init();		/* after vfs_mountroot() so DDI root is ready */
469 	cpu_kstat_init(CPU);	/* after vfs_mountroot() so TOD is valid */
470 	ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
471 				/* after vfs_mountroot() so hrestime is valid */
472 
473 	post_startup();
474 	swaploaded = 1;
475 
476 	/*
477 	 * Initialize Solaris Audit Subsystem
478 	 */
479 	audit_init();
480 
481 	/*
482 	 * Plumb the protocol modules and drivers only if we are not
483 	 * networked booted, in this case we already did it in rootconf().
484 	 */
485 	if (netboot == 0 && iscsiboot_prop == NULL)
486 		(void) strplumb();
487 
488 	gethrestime(&PTOU(curproc)->u_start);
489 	curthread->t_start = PTOU(curproc)->u_start.tv_sec;
490 	p->p_mstart = gethrtime();
491 
492 	/*
493 	 * Perform setup functions that can only be done after root
494 	 * and swap have been set up.
495 	 */
496 	consconfig();
497 #ifndef	__sparc
498 	release_bootstrap();
499 #endif
500 
501 	/*
502 	 * attach drivers with ddi-forceattach prop
503 	 * It must be done early enough to load hotplug drivers (e.g.
504 	 * pcmcia nexus) so that devices enumerated via hotplug is
505 	 * available before I/O subsystem is fully initialized.
506 	 */
507 	i_ddi_forceattach_drivers();
508 
509 	/*
510 	 * Set the scan rate and other parameters of the paging subsystem.
511 	 */
512 	setupclock(0);
513 
514 	/*
515 	 * Initialize process 0's lwp directory and lwpid hash table.
516 	 */
517 	p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
518 	p->p_lwpdir->ld_next = p->p_lwpdir + 1;
519 	p->p_lwpdir_sz = 2;
520 	p->p_tidhash = p0_tidhash;
521 	p->p_tidhash_sz = 2;
522 	p0_lep.le_thread = curthread;
523 	p0_lep.le_lwpid = curthread->t_tid;
524 	p0_lep.le_start = curthread->t_start;
525 	lwp_hash_in(p, &p0_lep, p0_tidhash, 2, 0);
526 
527 	/*
528 	 * Initialize extended accounting.
529 	 */
530 	exacct_init();
531 
532 	/*
533 	 * Initialize threads of sysevent event channels
534 	 */
535 	sysevent_evc_thrinit();
536 
537 	/*
538 	 * This must be done after post_startup() but before
539 	 * start_other_cpus()
540 	 */
541 	lgrp_init(LGRP_INIT_STAGE4);
542 
543 	/*
544 	 * Perform MP initialization, if any.
545 	 */
546 	start_other_cpus(0);
547 
548 #ifdef	__sparc
549 	/*
550 	 * Release bootstrap here since PROM interfaces are
551 	 * used to start other CPUs above.
552 	 */
553 	release_bootstrap();
554 #endif
555 
556 	/*
557 	 * Finish lgrp initialization after all CPUS are brought online.
558 	 */
559 	lgrp_init(LGRP_INIT_STAGE5);
560 
561 	/*
562 	 * After mp_init(), number of cpus are known (this is
563 	 * true for the time being, when there are actually
564 	 * hot pluggable cpus then this scheme  would not do).
565 	 * Any per cpu initialization is done here.
566 	 */
567 	kmem_mp_init();
568 	vmem_update(NULL);
569 
570 	clock_tick_init_post();
571 
572 	for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
573 		(**initptr)();
574 
575 	/*
576 	 * These must be called after start_other_cpus
577 	 */
578 	pm_cfb_setup_intr();
579 #if defined(__x86)
580 	fastboot_post_startup();
581 #endif
582 
583 	/*
584 	 * Make init process; enter scheduling loop with system process.
585 	 *
586 	 * Note that we manually assign the pids for these processes, for
587 	 * historical reasons.  If more pre-assigned pids are needed,
588 	 * FAMOUS_PIDS will have to be updated.
589 	 */
590 
591 	/* create init process */
592 	if (newproc(start_init, NULL, defaultcid, 59, NULL,
593 	    FAMOUS_PID_INIT))
594 		panic("main: unable to fork init.");
595 
596 	/* create pageout daemon */
597 	if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL,
598 	    FAMOUS_PID_PAGEOUT))
599 		panic("main: unable to fork pageout()");
600 
601 	/* create fsflush daemon */
602 	if (newproc(fsflush, NULL, syscid, minclsyspri, NULL,
603 	    FAMOUS_PID_FSFLUSH))
604 		panic("main: unable to fork fsflush()");
605 
606 	/* create cluster process if we're a member of one */
607 	if (cluster_bootflags & CLUSTER_BOOTED) {
608 		if (newproc(cluster_wrapper, NULL, syscid, minclsyspri,
609 		    NULL, 0)) {
610 			panic("main: unable to fork cluster()");
611 		}
612 	}
613 
614 	/*
615 	 * Create system threads (threads are associated with p0)
616 	 */
617 
618 	/* create module uninstall daemon */
619 	/* BugID 1132273. If swapping over NFS need a bigger stack */
620 	(void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
621 	    NULL, 0, &p0, TS_RUN, minclsyspri);
622 
623 	(void) thread_create(NULL, 0, seg_pasync_thread,
624 	    NULL, 0, &p0, TS_RUN, minclsyspri);
625 
626 	pid_setmin();
627 
628 	bcopy("sched", PTOU(curproc)->u_psargs, 6);
629 	bcopy("sched", PTOU(curproc)->u_comm, 5);
630 	sched();
631 	/* NOTREACHED */
632 }
633