xref: /titanic_50/usr/src/uts/common/os/main.c (revision 71269a2275bf5a143dad6461eee2710a344e7261)
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 
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 
74 #include <vm/as.h>
75 #include <vm/seg_kmem.h>
76 #include <sys/dc_ki.h>
77 
78 #include <c2/audit.h>
79 
80 /* well known processes */
81 proc_t *proc_sched;		/* memory scheduler */
82 proc_t *proc_init;		/* init */
83 proc_t *proc_pageout;		/* pageout daemon */
84 proc_t *proc_fsflush;		/* fsflush daemon */
85 
86 pgcnt_t	maxmem;		/* Maximum available memory in pages.	*/
87 pgcnt_t	freemem;	/* Current available memory in pages.	*/
88 int	audit_active;
89 int	interrupts_unleashed;	/* set when we do the first spl0() */
90 
91 kmem_cache_t *process_cache;	/* kmem cache for proc structures */
92 
93 /*
94  * Process 0's lwp directory and lwpid hash table.
95  */
96 lwpdir_t p0_lwpdir[2];
97 lwpdir_t *p0_tidhash[2];
98 lwpent_t p0_lep;
99 
100 /*
101  * Machine-independent initialization code
102  * Called from cold start routine as
103  * soon as a stack and segmentation
104  * have been established.
105  * Functions:
106  *	clear and free user core
107  *	turn on clock
108  *	hand craft 0th process
109  *	call all initialization routines
110  *	fork	- process 0 to schedule
111  *		- process 1 execute bootstrap
112  *		- process 2 to page out
113  *	create system threads
114  */
115 
116 int cluster_bootflags = 0;
117 
118 void
119 cluster_wrapper(void)
120 {
121 	cluster();
122 	panic("cluster()  returned");
123 }
124 
125 char initname[INITNAME_SZ] = "/sbin/init";	/* also referenced by zone0 */
126 char initargs[BOOTARGS_MAX] = "";		/* also referenced by zone0 */
127 extern int64_t lwp_sigmask(int, uint_t, uint_t);
128 
129 /*
130  * Construct a stack for init containing the arguments to it, then
131  * pass control to exec_common.
132  */
133 int
134 exec_init(const char *initpath, const char *args)
135 {
136 	caddr32_t ucp;
137 	caddr32_t *uap;
138 	caddr32_t *argv;
139 	caddr32_t exec_fnamep;
140 	char *scratchargs;
141 	int i, sarg;
142 	size_t argvlen, alen;
143 	boolean_t in_arg;
144 	int argc = 0;
145 	int error = 0, count = 0;
146 	proc_t *p = ttoproc(curthread);
147 	klwp_t *lwp = ttolwp(curthread);
148 	int brand_action;
149 
150 	if (args == NULL)
151 		args = "";
152 
153 	alen = strlen(initpath) + 1 + strlen(args) + 1;
154 	scratchargs = kmem_alloc(alen, KM_SLEEP);
155 	(void) snprintf(scratchargs, alen, "%s %s", initpath, args);
156 
157 	/*
158 	 * We do a quick two state parse of the string to sort out how big
159 	 * argc should be.
160 	 */
161 	in_arg = B_FALSE;
162 	for (i = 0; i < strlen(scratchargs); i++) {
163 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
164 			if (in_arg) {
165 				in_arg = B_FALSE;
166 				argc++;
167 			}
168 		} else {
169 			in_arg = B_TRUE;
170 		}
171 	}
172 	argvlen = sizeof (caddr32_t) * (argc + 1);
173 	argv = kmem_zalloc(argvlen, KM_SLEEP);
174 
175 	/*
176 	 * We pull off a bit of a hack here.  We work our way through the
177 	 * args string, putting nulls at the ends of space delimited tokens
178 	 * (boot args don't support quoting at this time).  Then we just
179 	 * copy the whole mess to userland in one go.  In other words, we
180 	 * transform this: "init -s -r\0" into this on the stack:
181 	 *
182 	 *	-0x00 \0
183 	 *	-0x01 r
184 	 *	-0x02 -  <--------.
185 	 *	-0x03 \0	  |
186 	 *	-0x04 s		  |
187 	 *	-0x05 -  <------. |
188 	 *	-0x06 \0	| |
189 	 *	-0x07 t		| |
190 	 *	-0x08 i 	| |
191 	 *	-0x09 n		| |
192 	 *	-0x0a i  <---.  | |
193 	 *	-0x10 NULL   |  | |	(argv[3])
194 	 *	-0x14   -----|--|-'	(argv[2])
195 	 *	-0x18  ------|--'	(argv[1])
196 	 *	-0x1c -------'		(argv[0])
197 	 *
198 	 * Since we know the value of ucp at the beginning of this process,
199 	 * we can trivially compute the argv[] array which we also need to
200 	 * place in userland: argv[i] = ucp - sarg(i), where ucp is the
201 	 * stack ptr, and sarg is the string index of the start of the
202 	 * argument.
203 	 */
204 	ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
205 
206 	argc = 0;
207 	in_arg = B_FALSE;
208 	sarg = 0;
209 
210 	for (i = 0; i < alen; i++) {
211 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
212 			if (in_arg == B_TRUE) {
213 				in_arg = B_FALSE;
214 				scratchargs[i] = '\0';
215 				argv[argc++] = ucp - (alen - sarg);
216 			}
217 		} else if (in_arg == B_FALSE) {
218 			in_arg = B_TRUE;
219 			sarg = i;
220 		}
221 	}
222 	ucp -= alen;
223 	error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
224 
225 	uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
226 	uap--;	/* advance to be below the word we're in */
227 	uap -= (argc + 1);	/* advance argc words down, plus one for NULL */
228 	error |= copyout(argv, uap, argvlen);
229 
230 	if (error != 0) {
231 		zcmn_err(p->p_zone->zone_id, CE_WARN,
232 		    "Could not construct stack for init.\n");
233 		kmem_free(argv, argvlen);
234 		kmem_free(scratchargs, alen);
235 		return (EFAULT);
236 	}
237 
238 	exec_fnamep = argv[0];
239 	kmem_free(argv, argvlen);
240 	kmem_free(scratchargs, alen);
241 
242 	/*
243 	 * Point at the arguments.
244 	 */
245 	lwp->lwp_ap = lwp->lwp_arg;
246 	lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
247 	lwp->lwp_arg[1] = (uintptr_t)uap;
248 	lwp->lwp_arg[2] = NULL;
249 	curthread->t_post_sys = 1;
250 	curthread->t_sysnum = SYS_execve;
251 
252 	/*
253 	 * If we are executing init from zsched, we may have inherited its
254 	 * parent process's signal mask.  Clear it now so that we behave in
255 	 * the same way as when started from the global zone.
256 	 */
257 	(void) lwp_sigmask(SIG_UNBLOCK, 0xffffffff, 0xffffffff);
258 
259 	brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
260 again:
261 	error = exec_common((const char *)(uintptr_t)exec_fnamep,
262 	    (const char **)(uintptr_t)uap, NULL, brand_action);
263 
264 	/*
265 	 * Normally we would just set lwp_argsaved and t_post_sys and
266 	 * let post_syscall reset lwp_ap for us.  Unfortunately,
267 	 * exec_init isn't always called from a system call.  Instead
268 	 * of making a mess of trap_cleanup, we just reset the args
269 	 * pointer here.
270 	 */
271 	reset_syscall_args();
272 
273 	switch (error) {
274 	case 0:
275 		return (0);
276 
277 	case ENOENT:
278 		zcmn_err(p->p_zone->zone_id, CE_WARN,
279 		    "exec(%s) failed (file not found).\n", initpath);
280 		return (ENOENT);
281 
282 	case EAGAIN:
283 	case EINTR:
284 		++count;
285 		if (count < 5) {
286 			zcmn_err(p->p_zone->zone_id, CE_WARN,
287 			    "exec(%s) failed with errno %d.  Retrying...\n",
288 			    initpath, error);
289 			goto again;
290 		}
291 	}
292 
293 	zcmn_err(p->p_zone->zone_id, CE_WARN,
294 	    "exec(%s) failed with errno %d.", initpath, error);
295 	return (error);
296 }
297 
298 /*
299  * This routine does all of the common setup for invoking init; global
300  * and non-global zones employ this routine for the functionality which is
301  * in common.
302  *
303  * This program (init, presumably) must be a 32-bit process.
304  */
305 int
306 start_init_common()
307 {
308 	proc_t *p = curproc;
309 	ASSERT_STACK_ALIGNED();
310 	p->p_zone->zone_proc_initpid = p->p_pid;
311 
312 	p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
313 	p->p_usrstack = (caddr_t)USRSTACK32;
314 	p->p_model = DATAMODEL_ILP32;
315 	p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
316 	p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
317 	p->p_stk_ctl = INT32_MAX;
318 
319 	p->p_as = as_alloc();
320 	p->p_as->a_proc = p;
321 	p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
322 	(void) hat_setup(p->p_as->a_hat, HAT_INIT);
323 
324 	init_core();
325 
326 	init_mstate(curthread, LMS_SYSTEM);
327 	return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
328 }
329 
330 /*
331  * Start the initial user process for the global zone; once running, if
332  * init should subsequently fail, it will be automatically be caught in the
333  * exit(2) path, and restarted by restart_init().
334  */
335 static void
336 start_init(void)
337 {
338 	proc_init = curproc;
339 
340 	ASSERT(curproc->p_zone->zone_initname != NULL);
341 
342 	if (start_init_common() != 0)
343 		halt("unix: Could not start init");
344 	lwp_rtt();
345 }
346 
347 #if defined(__i386) || defined(__amd64)
348 extern void return_instr(void);
349 void (*rootnex_iommu_add_intr)(void) = (void (*)(void))return_instr;
350 #endif
351 
352 void
353 main(void)
354 {
355 	proc_t		*p = ttoproc(curthread);	/* &p0 */
356 	int		(**initptr)();
357 	extern void	sched();
358 	extern void	fsflush();
359 	extern int	(*init_tbl[])();
360 	extern int	(*mp_init_tbl[])();
361 	extern id_t	syscid, defaultcid;
362 	extern int	swaploaded;
363 	extern int	netboot;
364 	extern void	vm_init(void);
365 	extern void	cbe_init_pre(void);
366 	extern void	cbe_init(void);
367 	extern void	clock_tick_init_pre(void);
368 	extern void	clock_tick_init_post(void);
369 	extern void	clock_init(void);
370 	extern void	physio_bufs_init(void);
371 	extern void	pm_cfb_setup_intr(void);
372 	extern int	pm_adjust_timestamps(dev_info_t *, void *);
373 	extern void	start_other_cpus(int);
374 	extern void	sysevent_evc_thrinit();
375 	extern void	lgrp_main_init(void);
376 	extern void	lgrp_main_mp_init(void);
377 #if defined(__x86)
378 	extern void	cpupm_post_startup(void);
379 #endif
380 	/*
381 	 * In the horrible world of x86 in-lines, you can't get symbolic
382 	 * structure offsets a la genassym.  This assertion is here so
383 	 * that the next poor slob who innocently changes the offset of
384 	 * cpu_thread doesn't waste as much time as I just did finding
385 	 * out that it's hard-coded in i86/ml/i86.il.  Similarly for
386 	 * curcpup.  You're welcome.
387 	 */
388 	ASSERT(CPU == CPU->cpu_self);
389 	ASSERT(curthread == CPU->cpu_thread);
390 	ASSERT_STACK_ALIGNED();
391 
392 	/*
393 	 * Setup the first lgroup, and home t0
394 	 */
395 	lgrp_setup();
396 
397 	/*
398 	 * Once 'startup()' completes, the thread_reaper() daemon would be
399 	 * created(in thread_init()). After that, it is safe to create threads
400 	 * that could exit. These exited threads will get reaped.
401 	 */
402 	startup();
403 	segkmem_gc();
404 	callb_init();
405 	cbe_init_pre();	/* x86 must initialize gethrtimef before timer_init */
406 	timer_init();	/* timer must be initialized before cyclic starts */
407 	cbe_init();
408 	callout_init();	/* callout table MUST be init'd after cyclics */
409 	clock_tick_init_pre();
410 	clock_init();
411 
412 	/*
413 	 * On some platforms, clkinitf() changes the timing source that
414 	 * gethrtime_unscaled() uses to generate timestamps.  cbe_init() calls
415 	 * clkinitf(), so re-initialize the microstate counters after the
416 	 * timesource has been chosen.
417 	 */
418 	init_mstate(&t0, LMS_SYSTEM);
419 	init_cpu_mstate(CPU, CMS_SYSTEM);
420 
421 	/*
422 	 * May need to probe to determine latencies from CPU 0 after
423 	 * gethrtime() comes alive in cbe_init() and before enabling interrupts
424 	 */
425 	lgrp_plat_probe();
426 
427 	/*
428 	 * Call all system initialization functions.
429 	 */
430 	for (initptr = &init_tbl[0]; *initptr; initptr++)
431 		(**initptr)();
432 
433 	/*
434 	 * initialize vm related stuff.
435 	 */
436 	vm_init();
437 
438 	/*
439 	 * initialize buffer pool for raw I/O requests
440 	 */
441 	physio_bufs_init();
442 
443 	ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
444 
445 	/*
446 	 * Drop the interrupt level and allow interrupts.  At this point
447 	 * the DDI guarantees that interrupts are enabled.
448 	 */
449 	(void) spl0();
450 	interrupts_unleashed = 1;
451 
452 #if defined(__i386) || defined(__amd64)
453 	/*
454 	 * add intel iommu fault event handler
455 	 */
456 	rootnex_iommu_add_intr();
457 #endif
458 
459 	vfs_mountroot();	/* Mount the root file system */
460 	errorq_init();		/* after vfs_mountroot() so DDI root is ready */
461 	cpu_kstat_init(CPU);	/* after vfs_mountroot() so TOD is valid */
462 	ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
463 				/* after vfs_mountroot() so hrestime is valid */
464 
465 	post_startup();
466 	swaploaded = 1;
467 
468 	/*
469 	 * Initialize Solaris Audit Subsystem
470 	 */
471 	audit_init();
472 
473 	/*
474 	 * Plumb the protocol modules and drivers only if we are not
475 	 * networked booted, in this case we already did it in rootconf().
476 	 */
477 	if (netboot == 0)
478 		(void) strplumb();
479 
480 	gethrestime(&PTOU(curproc)->u_start);
481 	curthread->t_start = PTOU(curproc)->u_start.tv_sec;
482 	p->p_mstart = gethrtime();
483 
484 	/*
485 	 * Perform setup functions that can only be done after root
486 	 * and swap have been set up.
487 	 */
488 	consconfig();
489 	release_bootstrap();
490 
491 	/*
492 	 * attach drivers with ddi-forceattach prop
493 	 * This must be done after consconfig() to prevent usb key/mouse
494 	 * from attaching before the upper console stream is plumbed.
495 	 * It must be done early enough to load hotplug drivers (e.g.
496 	 * pcmcia nexus) so that devices enumerated via hotplug is
497 	 * available before I/O subsystem is fully initialized.
498 	 */
499 	i_ddi_forceattach_drivers();
500 
501 	/*
502 	 * Set the scan rate and other parameters of the paging subsystem.
503 	 */
504 	setupclock(0);
505 
506 	/*
507 	 * Create kmem cache for proc structures
508 	 */
509 	process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
510 	    0, NULL, NULL, NULL, NULL, NULL, 0);
511 
512 	/*
513 	 * Initialize process 0's lwp directory and lwpid hash table.
514 	 */
515 	p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
516 	p->p_lwpdir->ld_next = p->p_lwpdir + 1;
517 	p->p_lwpdir_sz = 2;
518 	p->p_tidhash = p0_tidhash;
519 	p->p_tidhash_sz = 2;
520 	p0_lep.le_thread = curthread;
521 	p0_lep.le_lwpid = curthread->t_tid;
522 	p0_lep.le_start = curthread->t_start;
523 	lwp_hash_in(p, &p0_lep);
524 
525 	/*
526 	 * Initialize extended accounting.
527 	 */
528 	exacct_init();
529 
530 	/*
531 	 * Initialize threads of sysevent event channels
532 	 */
533 	sysevent_evc_thrinit();
534 
535 	/*
536 	 * main lgroup initialization
537 	 * This must be done after post_startup(), but before
538 	 * start_other_cpus()
539 	 */
540 	lgrp_main_init();
541 
542 	/*
543 	 * Perform MP initialization, if any.
544 	 */
545 	start_other_cpus(0);
546 
547 	/*
548 	 * Finish lgrp initialization after all CPUS are brought online.
549 	 */
550 	lgrp_main_mp_init();
551 
552 	/*
553 	 * After mp_init(), number of cpus are known (this is
554 	 * true for the time being, when there are actually
555 	 * hot pluggable cpus then this scheme  would not do).
556 	 * Any per cpu initialization is done here.
557 	 */
558 	kmem_mp_init();
559 	vmem_update(NULL);
560 
561 	clock_tick_init_post();
562 
563 	for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
564 		(**initptr)();
565 
566 	/*
567 	 * These must be called after start_other_cpus
568 	 */
569 	pm_cfb_setup_intr();
570 #if defined(__x86)
571 	cpupm_post_startup();
572 #endif
573 
574 	/*
575 	 * Make init process; enter scheduling loop with system process.
576 	 */
577 
578 	/* create init process */
579 	if (newproc(start_init, NULL, defaultcid, 59, NULL))
580 		panic("main: unable to fork init.");
581 
582 	/* create pageout daemon */
583 	if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL))
584 		panic("main: unable to fork pageout()");
585 
586 	/* create fsflush daemon */
587 	if (newproc(fsflush, NULL, syscid, minclsyspri, NULL))
588 		panic("main: unable to fork fsflush()");
589 
590 	/* create cluster process if we're a member of one */
591 	if (cluster_bootflags & CLUSTER_BOOTED) {
592 		if (newproc(cluster_wrapper, NULL, syscid, minclsyspri, NULL))
593 			panic("main: unable to fork cluster()");
594 	}
595 
596 	/*
597 	 * Create system threads (threads are associated with p0)
598 	 */
599 
600 	/* create module uninstall daemon */
601 	/* BugID 1132273. If swapping over NFS need a bigger stack */
602 	(void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
603 	    NULL, 0, &p0, TS_RUN, minclsyspri);
604 
605 	(void) thread_create(NULL, 0, seg_pasync_thread,
606 	    NULL, 0, &p0, TS_RUN, minclsyspri);
607 
608 	pid_setmin();
609 
610 	bcopy("sched", PTOU(curproc)->u_psargs, 6);
611 	bcopy("sched", PTOU(curproc)->u_comm, 5);
612 	sched();
613 	/* NOTREACHED */
614 }
615