xref: /illumos-gate/usr/src/uts/common/os/main.c (revision 6a604193b70017bd933cd973200b3f13803674b2)
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 /*	Copyright (c) 1988 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/sysmacros.h>
32 #include <sys/pcb.h>
33 #include <sys/systm.h>
34 #include <sys/signal.h>
35 #include <sys/cred.h>
36 #include <sys/user.h>
37 #include <sys/vfs.h>
38 #include <sys/vnode.h>
39 #include <sys/proc.h>
40 #include <sys/time.h>
41 #include <sys/file.h>
42 #include <sys/priocntl.h>
43 #include <sys/procset.h>
44 #include <sys/disp.h>
45 #include <sys/callo.h>
46 #include <sys/callb.h>
47 #include <sys/debug.h>
48 #include <sys/conf.h>
49 #include <sys/bootconf.h>
50 #include <sys/utsname.h>
51 #include <sys/cmn_err.h>
52 #include <sys/vmparam.h>
53 #include <sys/modctl.h>
54 #include <sys/vm.h>
55 #include <sys/callb.h>
56 #include <sys/ddi_timer.h>
57 #include <sys/kmem.h>
58 #include <sys/vmem.h>
59 #include <sys/cpuvar.h>
60 #include <sys/cladm.h>
61 #include <sys/corectl.h>
62 #include <sys/exec.h>
63 #include <sys/syscall.h>
64 #include <sys/reboot.h>
65 #include <sys/task.h>
66 #include <sys/exacct.h>
67 #include <sys/autoconf.h>
68 #include <sys/errorq.h>
69 #include <sys/class.h>
70 #include <sys/stack.h>
71 #include <sys/brand.h>
72 #include <sys/mmapobj.h>
73 
74 #include <vm/as.h>
75 #include <vm/seg_kmem.h>
76 #include <sys/dc_ki.h>
77 
78 #include <c2/audit.h>
79 #include <sys/bootprops.h>
80 
81 /* well known processes */
82 proc_t *proc_sched;		/* memory scheduler */
83 proc_t *proc_init;		/* init */
84 proc_t *proc_pageout;		/* pageout daemon */
85 proc_t *proc_fsflush;		/* fsflush daemon */
86 
87 pgcnt_t	maxmem;		/* Maximum available memory in pages.	*/
88 pgcnt_t	freemem;	/* Current available memory in pages.	*/
89 int	audit_active;
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  * Process 0's lwp directory and lwpid hash table.
96  */
97 lwpdir_t p0_lwpdir[2];
98 lwpdir_t *p0_tidhash[2];
99 lwpent_t p0_lep;
100 
101 /*
102  * Machine-independent initialization code
103  * Called from cold start routine as
104  * soon as a stack and segmentation
105  * have been established.
106  * Functions:
107  *	clear and free user core
108  *	turn on clock
109  *	hand craft 0th process
110  *	call all initialization routines
111  *	fork	- process 0 to schedule
112  *		- process 1 execute bootstrap
113  *		- process 2 to page out
114  *	create system threads
115  */
116 
117 int cluster_bootflags = 0;
118 
119 void
120 cluster_wrapper(void)
121 {
122 	cluster();
123 	panic("cluster()  returned");
124 }
125 
126 char initname[INITNAME_SZ] = "/sbin/init";	/* also referenced by zone0 */
127 char initargs[BOOTARGS_MAX] = "";		/* also referenced by zone0 */
128 extern int64_t lwp_sigmask(int, uint_t, uint_t);
129 
130 /*
131  * Construct a stack for init containing the arguments to it, then
132  * pass control to exec_common.
133  */
134 int
135 exec_init(const char *initpath, const char *args)
136 {
137 	caddr32_t ucp;
138 	caddr32_t *uap;
139 	caddr32_t *argv;
140 	caddr32_t exec_fnamep;
141 	char *scratchargs;
142 	int i, sarg;
143 	size_t argvlen, alen;
144 	boolean_t in_arg;
145 	int argc = 0;
146 	int error = 0, count = 0;
147 	proc_t *p = ttoproc(curthread);
148 	klwp_t *lwp = ttolwp(curthread);
149 	int brand_action;
150 
151 	if (args == NULL)
152 		args = "";
153 
154 	alen = strlen(initpath) + 1 + strlen(args) + 1;
155 	scratchargs = kmem_alloc(alen, KM_SLEEP);
156 	(void) snprintf(scratchargs, alen, "%s %s", initpath, args);
157 
158 	/*
159 	 * We do a quick two state parse of the string to sort out how big
160 	 * argc should be.
161 	 */
162 	in_arg = B_FALSE;
163 	for (i = 0; i < strlen(scratchargs); i++) {
164 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
165 			if (in_arg) {
166 				in_arg = B_FALSE;
167 				argc++;
168 			}
169 		} else {
170 			in_arg = B_TRUE;
171 		}
172 	}
173 	argvlen = sizeof (caddr32_t) * (argc + 1);
174 	argv = kmem_zalloc(argvlen, KM_SLEEP);
175 
176 	/*
177 	 * We pull off a bit of a hack here.  We work our way through the
178 	 * args string, putting nulls at the ends of space delimited tokens
179 	 * (boot args don't support quoting at this time).  Then we just
180 	 * copy the whole mess to userland in one go.  In other words, we
181 	 * transform this: "init -s -r\0" into this on the stack:
182 	 *
183 	 *	-0x00 \0
184 	 *	-0x01 r
185 	 *	-0x02 -  <--------.
186 	 *	-0x03 \0	  |
187 	 *	-0x04 s		  |
188 	 *	-0x05 -  <------. |
189 	 *	-0x06 \0	| |
190 	 *	-0x07 t		| |
191 	 *	-0x08 i 	| |
192 	 *	-0x09 n		| |
193 	 *	-0x0a i  <---.  | |
194 	 *	-0x10 NULL   |  | |	(argv[3])
195 	 *	-0x14   -----|--|-'	(argv[2])
196 	 *	-0x18  ------|--'	(argv[1])
197 	 *	-0x1c -------'		(argv[0])
198 	 *
199 	 * Since we know the value of ucp at the beginning of this process,
200 	 * we can trivially compute the argv[] array which we also need to
201 	 * place in userland: argv[i] = ucp - sarg(i), where ucp is the
202 	 * stack ptr, and sarg is the string index of the start of the
203 	 * argument.
204 	 */
205 	ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
206 
207 	argc = 0;
208 	in_arg = B_FALSE;
209 	sarg = 0;
210 
211 	for (i = 0; i < alen; i++) {
212 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
213 			if (in_arg == B_TRUE) {
214 				in_arg = B_FALSE;
215 				scratchargs[i] = '\0';
216 				argv[argc++] = ucp - (alen - sarg);
217 			}
218 		} else if (in_arg == B_FALSE) {
219 			in_arg = B_TRUE;
220 			sarg = i;
221 		}
222 	}
223 	ucp -= alen;
224 	error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
225 
226 	uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
227 	uap--;	/* advance to be below the word we're in */
228 	uap -= (argc + 1);	/* advance argc words down, plus one for NULL */
229 	error |= copyout(argv, uap, argvlen);
230 
231 	if (error != 0) {
232 		zcmn_err(p->p_zone->zone_id, CE_WARN,
233 		    "Could not construct stack for init.\n");
234 		kmem_free(argv, argvlen);
235 		kmem_free(scratchargs, alen);
236 		return (EFAULT);
237 	}
238 
239 	exec_fnamep = argv[0];
240 	kmem_free(argv, argvlen);
241 	kmem_free(scratchargs, alen);
242 
243 	/*
244 	 * Point at the arguments.
245 	 */
246 	lwp->lwp_ap = lwp->lwp_arg;
247 	lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
248 	lwp->lwp_arg[1] = (uintptr_t)uap;
249 	lwp->lwp_arg[2] = NULL;
250 	curthread->t_post_sys = 1;
251 	curthread->t_sysnum = SYS_execve;
252 
253 	/*
254 	 * If we are executing init from zsched, we may have inherited its
255 	 * parent process's signal mask.  Clear it now so that we behave in
256 	 * the same way as when started from the global zone.
257 	 */
258 	(void) lwp_sigmask(SIG_UNBLOCK, 0xffffffff, 0xffffffff);
259 
260 	brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
261 again:
262 	error = exec_common((const char *)(uintptr_t)exec_fnamep,
263 	    (const char **)(uintptr_t)uap, NULL, brand_action);
264 
265 	/*
266 	 * Normally we would just set lwp_argsaved and t_post_sys and
267 	 * let post_syscall reset lwp_ap for us.  Unfortunately,
268 	 * exec_init isn't always called from a system call.  Instead
269 	 * of making a mess of trap_cleanup, we just reset the args
270 	 * pointer here.
271 	 */
272 	reset_syscall_args();
273 
274 	switch (error) {
275 	case 0:
276 		return (0);
277 
278 	case ENOENT:
279 		zcmn_err(p->p_zone->zone_id, CE_WARN,
280 		    "exec(%s) failed (file not found).\n", initpath);
281 		return (ENOENT);
282 
283 	case EAGAIN:
284 	case EINTR:
285 		++count;
286 		if (count < 5) {
287 			zcmn_err(p->p_zone->zone_id, CE_WARN,
288 			    "exec(%s) failed with errno %d.  Retrying...\n",
289 			    initpath, error);
290 			goto again;
291 		}
292 	}
293 
294 	zcmn_err(p->p_zone->zone_id, CE_WARN,
295 	    "exec(%s) failed with errno %d.", initpath, error);
296 	return (error);
297 }
298 
299 /*
300  * This routine does all of the common setup for invoking init; global
301  * and non-global zones employ this routine for the functionality which is
302  * in common.
303  *
304  * This program (init, presumably) must be a 32-bit process.
305  */
306 int
307 start_init_common()
308 {
309 	proc_t *p = curproc;
310 	ASSERT_STACK_ALIGNED();
311 	p->p_zone->zone_proc_initpid = p->p_pid;
312 
313 	p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
314 	p->p_usrstack = (caddr_t)USRSTACK32;
315 	p->p_model = DATAMODEL_ILP32;
316 	p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
317 	p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
318 	p->p_stk_ctl = INT32_MAX;
319 
320 	p->p_as = as_alloc();
321 	p->p_as->a_proc = p;
322 	p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
323 	(void) hat_setup(p->p_as->a_hat, HAT_INIT);
324 
325 	init_core();
326 
327 	init_mstate(curthread, LMS_SYSTEM);
328 	return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
329 }
330 
331 /*
332  * Start the initial user process for the global zone; once running, if
333  * init should subsequently fail, it will be automatically be caught in the
334  * exit(2) path, and restarted by restart_init().
335  */
336 static void
337 start_init(void)
338 {
339 	proc_init = curproc;
340 
341 	ASSERT(curproc->p_zone->zone_initname != NULL);
342 
343 	if (start_init_common() != 0)
344 		halt("unix: Could not start init");
345 	lwp_rtt();
346 }
347 
348 #if defined(__i386) || defined(__amd64)
349 extern void return_instr(void);
350 void (*rootnex_iommu_add_intr)(void) = (void (*)(void))return_instr;
351 #endif
352 
353 void
354 main(void)
355 {
356 	proc_t		*p = ttoproc(curthread);	/* &p0 */
357 	int		(**initptr)();
358 	extern void	sched();
359 	extern void	fsflush();
360 	extern int	(*init_tbl[])();
361 	extern int	(*mp_init_tbl[])();
362 	extern id_t	syscid, defaultcid;
363 	extern int	swaploaded;
364 	extern int	netboot;
365 	extern ib_boot_prop_t *iscsiboot_prop;
366 	extern void	vm_init(void);
367 	extern void	cbe_init_pre(void);
368 	extern void	cbe_init(void);
369 	extern void	clock_tick_init_pre(void);
370 	extern void	clock_tick_init_post(void);
371 	extern void	clock_init(void);
372 	extern void	physio_bufs_init(void);
373 	extern void	pm_cfb_setup_intr(void);
374 	extern int	pm_adjust_timestamps(dev_info_t *, void *);
375 	extern void	start_other_cpus(int);
376 	extern void	sysevent_evc_thrinit();
377 	extern void	lgrp_main_init(void);
378 	extern void	lgrp_main_mp_init(void);
379 #if defined(__x86)
380 	extern void	cpupm_post_startup(void);
381 #endif
382 	/*
383 	 * In the horrible world of x86 in-lines, you can't get symbolic
384 	 * structure offsets a la genassym.  This assertion is here so
385 	 * that the next poor slob who innocently changes the offset of
386 	 * cpu_thread doesn't waste as much time as I just did finding
387 	 * out that it's hard-coded in i86/ml/i86.il.  Similarly for
388 	 * curcpup.  You're welcome.
389 	 */
390 	ASSERT(CPU == CPU->cpu_self);
391 	ASSERT(curthread == CPU->cpu_thread);
392 	ASSERT_STACK_ALIGNED();
393 
394 	/*
395 	 * Setup the first lgroup, and home t0
396 	 */
397 	lgrp_setup();
398 
399 	/*
400 	 * Once 'startup()' completes, the thread_reaper() daemon would be
401 	 * created(in thread_init()). After that, it is safe to create threads
402 	 * that could exit. These exited threads will get reaped.
403 	 */
404 	startup();
405 	segkmem_gc();
406 	callb_init();
407 	cbe_init_pre();	/* x86 must initialize gethrtimef before timer_init */
408 	timer_init();	/* timer must be initialized before cyclic starts */
409 	cbe_init();
410 	callout_init();	/* callout table MUST be init'd after cyclics */
411 	clock_tick_init_pre();
412 	clock_init();
413 
414 	/*
415 	 * On some platforms, clkinitf() changes the timing source that
416 	 * gethrtime_unscaled() uses to generate timestamps.  cbe_init() calls
417 	 * clkinitf(), so re-initialize the microstate counters after the
418 	 * timesource has been chosen.
419 	 */
420 	init_mstate(&t0, LMS_SYSTEM);
421 	init_cpu_mstate(CPU, CMS_SYSTEM);
422 
423 	/*
424 	 * May need to probe to determine latencies from CPU 0 after
425 	 * gethrtime() comes alive in cbe_init() and before enabling interrupts
426 	 */
427 	lgrp_plat_probe();
428 
429 	/*
430 	 * Call all system initialization functions.
431 	 */
432 	for (initptr = &init_tbl[0]; *initptr; initptr++)
433 		(**initptr)();
434 	/*
435 	 * Load iSCSI boot properties
436 	 */
437 	ld_ib_prop();
438 	/*
439 	 * initialize vm related stuff.
440 	 */
441 	vm_init();
442 
443 	/*
444 	 * initialize buffer pool for raw I/O requests
445 	 */
446 	physio_bufs_init();
447 
448 	ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
449 
450 	/*
451 	 * Drop the interrupt level and allow interrupts.  At this point
452 	 * the DDI guarantees that interrupts are enabled.
453 	 */
454 	(void) spl0();
455 	interrupts_unleashed = 1;
456 
457 #if defined(__i386) || defined(__amd64)
458 	/*
459 	 * add intel iommu fault event handler
460 	 */
461 	rootnex_iommu_add_intr();
462 #endif
463 
464 	vfs_mountroot();	/* Mount the root file system */
465 	errorq_init();		/* after vfs_mountroot() so DDI root is ready */
466 	cpu_kstat_init(CPU);	/* after vfs_mountroot() so TOD is valid */
467 	ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
468 				/* after vfs_mountroot() so hrestime is valid */
469 
470 	post_startup();
471 	swaploaded = 1;
472 
473 	/*
474 	 * Initialize Solaris Audit Subsystem
475 	 */
476 	audit_init();
477 
478 	/*
479 	 * Plumb the protocol modules and drivers only if we are not
480 	 * networked booted, in this case we already did it in rootconf().
481 	 */
482 	if (netboot == 0 && iscsiboot_prop == NULL)
483 		(void) strplumb();
484 
485 	gethrestime(&PTOU(curproc)->u_start);
486 	curthread->t_start = PTOU(curproc)->u_start.tv_sec;
487 	p->p_mstart = gethrtime();
488 
489 	/*
490 	 * Perform setup functions that can only be done after root
491 	 * and swap have been set up.
492 	 */
493 	consconfig();
494 	release_bootstrap();
495 
496 	/*
497 	 * attach drivers with ddi-forceattach prop
498 	 * This must be done after consconfig() to prevent usb key/mouse
499 	 * from attaching before the upper console stream is plumbed.
500 	 * It must be done early enough to load hotplug drivers (e.g.
501 	 * pcmcia nexus) so that devices enumerated via hotplug is
502 	 * available before I/O subsystem is fully initialized.
503 	 */
504 	i_ddi_forceattach_drivers();
505 
506 	/*
507 	 * Set the scan rate and other parameters of the paging subsystem.
508 	 */
509 	setupclock(0);
510 
511 	/*
512 	 * Create kmem cache for proc structures
513 	 */
514 	process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
515 	    0, NULL, NULL, NULL, NULL, NULL, 0);
516 
517 	/*
518 	 * Initialize process 0's lwp directory and lwpid hash table.
519 	 */
520 	p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
521 	p->p_lwpdir->ld_next = p->p_lwpdir + 1;
522 	p->p_lwpdir_sz = 2;
523 	p->p_tidhash = p0_tidhash;
524 	p->p_tidhash_sz = 2;
525 	p0_lep.le_thread = curthread;
526 	p0_lep.le_lwpid = curthread->t_tid;
527 	p0_lep.le_start = curthread->t_start;
528 	lwp_hash_in(p, &p0_lep);
529 
530 	/*
531 	 * Initialize extended accounting.
532 	 */
533 	exacct_init();
534 
535 	/*
536 	 * Initialize threads of sysevent event channels
537 	 */
538 	sysevent_evc_thrinit();
539 
540 	/*
541 	 * main lgroup initialization
542 	 * This must be done after post_startup(), but before
543 	 * start_other_cpus()
544 	 */
545 	lgrp_main_init();
546 
547 	/*
548 	 * Perform MP initialization, if any.
549 	 */
550 	start_other_cpus(0);
551 
552 	/*
553 	 * Finish lgrp initialization after all CPUS are brought online.
554 	 */
555 	lgrp_main_mp_init();
556 
557 	/*
558 	 * Initialize lib_va arenas.  Needs to be done after start_other_cpus
559 	 * so that USERLIMIT32 represents the final value after any
560 	 * workarounds have been applied. Also need to be done before we
561 	 * create any processes so that all libs can be cached.
562 	 */
563 	lib_va_init();
564 
565 	/*
566 	 * After mp_init(), number of cpus are known (this is
567 	 * true for the time being, when there are actually
568 	 * hot pluggable cpus then this scheme  would not do).
569 	 * Any per cpu initialization is done here.
570 	 */
571 	kmem_mp_init();
572 	vmem_update(NULL);
573 
574 	clock_tick_init_post();
575 
576 	for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
577 		(**initptr)();
578 
579 	/*
580 	 * These must be called after start_other_cpus
581 	 */
582 	pm_cfb_setup_intr();
583 #if defined(__x86)
584 	cpupm_post_startup();
585 #endif
586 
587 	/*
588 	 * Make init process; enter scheduling loop with system process.
589 	 */
590 
591 	/* create init process */
592 	if (newproc(start_init, NULL, defaultcid, 59, NULL))
593 		panic("main: unable to fork init.");
594 
595 	/* create pageout daemon */
596 	if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL))
597 		panic("main: unable to fork pageout()");
598 
599 	/* create fsflush daemon */
600 	if (newproc(fsflush, NULL, syscid, minclsyspri, NULL))
601 		panic("main: unable to fork fsflush()");
602 
603 	/* create cluster process if we're a member of one */
604 	if (cluster_bootflags & CLUSTER_BOOTED) {
605 		if (newproc(cluster_wrapper, NULL, syscid, minclsyspri, NULL))
606 			panic("main: unable to fork cluster()");
607 	}
608 
609 	/*
610 	 * Create system threads (threads are associated with p0)
611 	 */
612 
613 	/* create module uninstall daemon */
614 	/* BugID 1132273. If swapping over NFS need a bigger stack */
615 	(void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
616 	    NULL, 0, &p0, TS_RUN, minclsyspri);
617 
618 	(void) thread_create(NULL, 0, seg_pasync_thread,
619 	    NULL, 0, &p0, TS_RUN, minclsyspri);
620 
621 	pid_setmin();
622 
623 	bcopy("sched", PTOU(curproc)->u_psargs, 6);
624 	bcopy("sched", PTOU(curproc)->u_comm, 5);
625 	sched();
626 	/* NOTREACHED */
627 }
628