xref: /titanic_41/usr/src/uts/common/os/main.c (revision e802abbda8c322f24d47835734f4a793ef15ddc8)
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 2009 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	audit_active;
91 int	interrupts_unleashed;	/* set when we do the first spl0() */
92 
93 kmem_cache_t *process_cache;	/* kmem cache for proc structures */
94 
95 /*
96  * Process 0's lwp directory and lwpid hash table.
97  */
98 lwpdir_t p0_lwpdir[2];
99 tidhash_t p0_tidhash[2];
100 lwpent_t p0_lep;
101 
102 /*
103  * Machine-independent initialization code
104  * Called from cold start routine as
105  * soon as a stack and segmentation
106  * have been established.
107  * Functions:
108  *	clear and free user core
109  *	turn on clock
110  *	hand craft 0th process
111  *	call all initialization routines
112  *	fork	- process 0 to schedule
113  *		- process 1 execute bootstrap
114  *		- process 2 to page out
115  *	create system threads
116  */
117 
118 int cluster_bootflags = 0;
119 
120 void
121 cluster_wrapper(void)
122 {
123 	cluster();
124 	panic("cluster()  returned");
125 }
126 
127 char initname[INITNAME_SZ] = "/sbin/init";	/* also referenced by zone0 */
128 char initargs[BOOTARGS_MAX] = "";		/* also referenced by zone0 */
129 extern int64_t lwp_sigmask(int, uint_t, uint_t);
130 
131 /*
132  * Construct a stack for init containing the arguments to it, then
133  * pass control to exec_common.
134  */
135 int
136 exec_init(const char *initpath, const char *args)
137 {
138 	caddr32_t ucp;
139 	caddr32_t *uap;
140 	caddr32_t *argv;
141 	caddr32_t exec_fnamep;
142 	char *scratchargs;
143 	int i, sarg;
144 	size_t argvlen, alen;
145 	boolean_t in_arg;
146 	int argc = 0;
147 	int error = 0, count = 0;
148 	proc_t *p = ttoproc(curthread);
149 	klwp_t *lwp = ttolwp(curthread);
150 	int brand_action;
151 
152 	if (args == NULL)
153 		args = "";
154 
155 	alen = strlen(initpath) + 1 + strlen(args) + 1;
156 	scratchargs = kmem_alloc(alen, KM_SLEEP);
157 	(void) snprintf(scratchargs, alen, "%s %s", initpath, args);
158 
159 	/*
160 	 * We do a quick two state parse of the string to sort out how big
161 	 * argc should be.
162 	 */
163 	in_arg = B_FALSE;
164 	for (i = 0; i < strlen(scratchargs); i++) {
165 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
166 			if (in_arg) {
167 				in_arg = B_FALSE;
168 				argc++;
169 			}
170 		} else {
171 			in_arg = B_TRUE;
172 		}
173 	}
174 	argvlen = sizeof (caddr32_t) * (argc + 1);
175 	argv = kmem_zalloc(argvlen, KM_SLEEP);
176 
177 	/*
178 	 * We pull off a bit of a hack here.  We work our way through the
179 	 * args string, putting nulls at the ends of space delimited tokens
180 	 * (boot args don't support quoting at this time).  Then we just
181 	 * copy the whole mess to userland in one go.  In other words, we
182 	 * transform this: "init -s -r\0" into this on the stack:
183 	 *
184 	 *	-0x00 \0
185 	 *	-0x01 r
186 	 *	-0x02 -  <--------.
187 	 *	-0x03 \0	  |
188 	 *	-0x04 s		  |
189 	 *	-0x05 -  <------. |
190 	 *	-0x06 \0	| |
191 	 *	-0x07 t		| |
192 	 *	-0x08 i 	| |
193 	 *	-0x09 n		| |
194 	 *	-0x0a i  <---.  | |
195 	 *	-0x10 NULL   |  | |	(argv[3])
196 	 *	-0x14   -----|--|-'	(argv[2])
197 	 *	-0x18  ------|--'	(argv[1])
198 	 *	-0x1c -------'		(argv[0])
199 	 *
200 	 * Since we know the value of ucp at the beginning of this process,
201 	 * we can trivially compute the argv[] array which we also need to
202 	 * place in userland: argv[i] = ucp - sarg(i), where ucp is the
203 	 * stack ptr, and sarg is the string index of the start of the
204 	 * argument.
205 	 */
206 	ucp = (caddr32_t)(uintptr_t)p->p_usrstack;
207 
208 	argc = 0;
209 	in_arg = B_FALSE;
210 	sarg = 0;
211 
212 	for (i = 0; i < alen; i++) {
213 		if (scratchargs[i] == ' ' || scratchargs[i] == '\0') {
214 			if (in_arg == B_TRUE) {
215 				in_arg = B_FALSE;
216 				scratchargs[i] = '\0';
217 				argv[argc++] = ucp - (alen - sarg);
218 			}
219 		} else if (in_arg == B_FALSE) {
220 			in_arg = B_TRUE;
221 			sarg = i;
222 		}
223 	}
224 	ucp -= alen;
225 	error |= copyout(scratchargs, (caddr_t)(uintptr_t)ucp, alen);
226 
227 	uap = (caddr32_t *)P2ALIGN((uintptr_t)ucp, sizeof (caddr32_t));
228 	uap--;	/* advance to be below the word we're in */
229 	uap -= (argc + 1);	/* advance argc words down, plus one for NULL */
230 	error |= copyout(argv, uap, argvlen);
231 
232 	if (error != 0) {
233 		zcmn_err(p->p_zone->zone_id, CE_WARN,
234 		    "Could not construct stack for init.\n");
235 		kmem_free(argv, argvlen);
236 		kmem_free(scratchargs, alen);
237 		return (EFAULT);
238 	}
239 
240 	exec_fnamep = argv[0];
241 	kmem_free(argv, argvlen);
242 	kmem_free(scratchargs, alen);
243 
244 	/*
245 	 * Point at the arguments.
246 	 */
247 	lwp->lwp_ap = lwp->lwp_arg;
248 	lwp->lwp_arg[0] = (uintptr_t)exec_fnamep;
249 	lwp->lwp_arg[1] = (uintptr_t)uap;
250 	lwp->lwp_arg[2] = NULL;
251 	curthread->t_post_sys = 1;
252 	curthread->t_sysnum = SYS_execve;
253 
254 	/*
255 	 * If we are executing init from zsched, we may have inherited its
256 	 * parent process's signal mask.  Clear it now so that we behave in
257 	 * the same way as when started from the global zone.
258 	 */
259 	(void) lwp_sigmask(SIG_UNBLOCK, 0xffffffff, 0xffffffff);
260 
261 	brand_action = ZONE_IS_BRANDED(p->p_zone) ? EBA_BRAND : EBA_NONE;
262 again:
263 	error = exec_common((const char *)(uintptr_t)exec_fnamep,
264 	    (const char **)(uintptr_t)uap, NULL, brand_action);
265 
266 	/*
267 	 * Normally we would just set lwp_argsaved and t_post_sys and
268 	 * let post_syscall reset lwp_ap for us.  Unfortunately,
269 	 * exec_init isn't always called from a system call.  Instead
270 	 * of making a mess of trap_cleanup, we just reset the args
271 	 * pointer here.
272 	 */
273 	reset_syscall_args();
274 
275 	switch (error) {
276 	case 0:
277 		return (0);
278 
279 	case ENOENT:
280 		zcmn_err(p->p_zone->zone_id, CE_WARN,
281 		    "exec(%s) failed (file not found).\n", initpath);
282 		return (ENOENT);
283 
284 	case EAGAIN:
285 	case EINTR:
286 		++count;
287 		if (count < 5) {
288 			zcmn_err(p->p_zone->zone_id, CE_WARN,
289 			    "exec(%s) failed with errno %d.  Retrying...\n",
290 			    initpath, error);
291 			goto again;
292 		}
293 	}
294 
295 	zcmn_err(p->p_zone->zone_id, CE_WARN,
296 	    "exec(%s) failed with errno %d.", initpath, error);
297 	return (error);
298 }
299 
300 /*
301  * This routine does all of the common setup for invoking init; global
302  * and non-global zones employ this routine for the functionality which is
303  * in common.
304  *
305  * This program (init, presumably) must be a 32-bit process.
306  */
307 int
308 start_init_common()
309 {
310 	proc_t *p = curproc;
311 	ASSERT_STACK_ALIGNED();
312 	p->p_zone->zone_proc_initpid = p->p_pid;
313 
314 	p->p_cstime = p->p_stime = p->p_cutime = p->p_utime = 0;
315 	p->p_usrstack = (caddr_t)USRSTACK32;
316 	p->p_model = DATAMODEL_ILP32;
317 	p->p_stkprot = PROT_ZFOD & ~PROT_EXEC;
318 	p->p_datprot = PROT_ZFOD & ~PROT_EXEC;
319 	p->p_stk_ctl = INT32_MAX;
320 
321 	p->p_as = as_alloc();
322 	p->p_as->a_proc = p;
323 	p->p_as->a_userlimit = (caddr_t)USERLIMIT32;
324 	(void) hat_setup(p->p_as->a_hat, HAT_INIT);
325 
326 	init_core();
327 
328 	init_mstate(curthread, LMS_SYSTEM);
329 	return (exec_init(p->p_zone->zone_initname, p->p_zone->zone_bootargs));
330 }
331 
332 /*
333  * Start the initial user process for the global zone; once running, if
334  * init should subsequently fail, it will be automatically be caught in the
335  * exit(2) path, and restarted by restart_init().
336  */
337 static void
338 start_init(void)
339 {
340 	proc_init = curproc;
341 
342 	ASSERT(curproc->p_zone->zone_initname != NULL);
343 
344 	if (start_init_common() != 0)
345 		halt("unix: Could not start init");
346 	lwp_rtt();
347 }
348 
349 #if defined(__i386) || defined(__amd64)
350 extern void return_instr(void);
351 void (*rootnex_iommu_add_intr)(void) = (void (*)(void))return_instr;
352 #endif
353 
354 void
355 main(void)
356 {
357 	proc_t		*p = ttoproc(curthread);	/* &p0 */
358 	int		(**initptr)();
359 	extern void	sched();
360 	extern void	fsflush();
361 	extern int	(*init_tbl[])();
362 	extern int	(*mp_init_tbl[])();
363 	extern id_t	syscid, defaultcid;
364 	extern int	swaploaded;
365 	extern int	netboot;
366 	extern ib_boot_prop_t *iscsiboot_prop;
367 	extern void	vm_init(void);
368 	extern void	cbe_init_pre(void);
369 	extern void	cbe_init(void);
370 	extern void	clock_tick_init_pre(void);
371 	extern void	clock_tick_init_post(void);
372 	extern void	clock_init(void);
373 	extern void	physio_bufs_init(void);
374 	extern void	pm_cfb_setup_intr(void);
375 	extern int	pm_adjust_timestamps(dev_info_t *, void *);
376 	extern void	start_other_cpus(int);
377 	extern void	sysevent_evc_thrinit();
378 	extern void	lgrp_main_init(void);
379 	extern void	lgrp_main_mp_init(void);
380 #if defined(__x86)
381 	extern void	cpupm_post_startup(void);
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 the first lgroup, and home t0
398 	 */
399 	lgrp_setup();
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 	 */
429 	lgrp_plat_probe();
430 
431 	/*
432 	 * Call all system initialization functions.
433 	 */
434 	for (initptr = &init_tbl[0]; *initptr; initptr++)
435 		(**initptr)();
436 	/*
437 	 * Load iSCSI boot properties
438 	 */
439 	ld_ib_prop();
440 	/*
441 	 * initialize vm related stuff.
442 	 */
443 	vm_init();
444 
445 	/*
446 	 * initialize buffer pool for raw I/O requests
447 	 */
448 	physio_bufs_init();
449 
450 	ttolwp(curthread)->lwp_error = 0; /* XXX kludge for SCSI driver */
451 
452 	/*
453 	 * Drop the interrupt level and allow interrupts.  At this point
454 	 * the DDI guarantees that interrupts are enabled.
455 	 */
456 	(void) spl0();
457 	interrupts_unleashed = 1;
458 
459 #if defined(__i386) || defined(__amd64)
460 	/*
461 	 * add intel iommu fault event handler
462 	 */
463 	rootnex_iommu_add_intr();
464 #endif
465 
466 	vfs_mountroot();	/* Mount the root file system */
467 	errorq_init();		/* after vfs_mountroot() so DDI root is ready */
468 	cpu_kstat_init(CPU);	/* after vfs_mountroot() so TOD is valid */
469 	ddi_walk_devs(ddi_root_node(), pm_adjust_timestamps, NULL);
470 				/* after vfs_mountroot() so hrestime is valid */
471 
472 	post_startup();
473 	swaploaded = 1;
474 
475 	/*
476 	 * Initialize Solaris Audit Subsystem
477 	 */
478 	audit_init();
479 
480 	/*
481 	 * Plumb the protocol modules and drivers only if we are not
482 	 * networked booted, in this case we already did it in rootconf().
483 	 */
484 	if (netboot == 0 && iscsiboot_prop == NULL)
485 		(void) strplumb();
486 
487 	gethrestime(&PTOU(curproc)->u_start);
488 	curthread->t_start = PTOU(curproc)->u_start.tv_sec;
489 	p->p_mstart = gethrtime();
490 
491 	/*
492 	 * Perform setup functions that can only be done after root
493 	 * and swap have been set up.
494 	 */
495 	consconfig();
496 	release_bootstrap();
497 
498 	/*
499 	 * attach drivers with ddi-forceattach prop
500 	 * This must be done after consconfig() to prevent usb key/mouse
501 	 * from attaching before the upper console stream is plumbed.
502 	 * It must be done early enough to load hotplug drivers (e.g.
503 	 * pcmcia nexus) so that devices enumerated via hotplug is
504 	 * available before I/O subsystem is fully initialized.
505 	 */
506 	i_ddi_forceattach_drivers();
507 
508 	/*
509 	 * Set the scan rate and other parameters of the paging subsystem.
510 	 */
511 	setupclock(0);
512 
513 	/*
514 	 * Create kmem cache for proc structures
515 	 */
516 	process_cache = kmem_cache_create("process_cache", sizeof (proc_t),
517 	    0, NULL, NULL, NULL, NULL, NULL, 0);
518 
519 	/*
520 	 * Initialize process 0's lwp directory and lwpid hash table.
521 	 */
522 	p->p_lwpdir = p->p_lwpfree = p0_lwpdir;
523 	p->p_lwpdir->ld_next = p->p_lwpdir + 1;
524 	p->p_lwpdir_sz = 2;
525 	p->p_tidhash = p0_tidhash;
526 	p->p_tidhash_sz = 2;
527 	p0_lep.le_thread = curthread;
528 	p0_lep.le_lwpid = curthread->t_tid;
529 	p0_lep.le_start = curthread->t_start;
530 	lwp_hash_in(p, &p0_lep, p0_tidhash, 2, 0);
531 
532 	/*
533 	 * Initialize extended accounting.
534 	 */
535 	exacct_init();
536 
537 	/*
538 	 * Initialize threads of sysevent event channels
539 	 */
540 	sysevent_evc_thrinit();
541 
542 	/*
543 	 * main lgroup initialization
544 	 * This must be done after post_startup(), but before
545 	 * start_other_cpus()
546 	 */
547 	lgrp_main_init();
548 
549 	/*
550 	 * Perform MP initialization, if any.
551 	 */
552 	start_other_cpus(0);
553 
554 	/*
555 	 * Finish lgrp initialization after all CPUS are brought online.
556 	 */
557 	lgrp_main_mp_init();
558 
559 	/*
560 	 * After mp_init(), number of cpus are known (this is
561 	 * true for the time being, when there are actually
562 	 * hot pluggable cpus then this scheme  would not do).
563 	 * Any per cpu initialization is done here.
564 	 */
565 	kmem_mp_init();
566 	vmem_update(NULL);
567 
568 	clock_tick_init_post();
569 
570 	for (initptr = &mp_init_tbl[0]; *initptr; initptr++)
571 		(**initptr)();
572 
573 	/*
574 	 * These must be called after start_other_cpus
575 	 */
576 	pm_cfb_setup_intr();
577 #if defined(__x86)
578 	cpupm_post_startup();
579 	fastboot_post_startup();
580 #endif
581 
582 	/*
583 	 * Make init process; enter scheduling loop with system process.
584 	 */
585 
586 	/* create init process */
587 	if (newproc(start_init, NULL, defaultcid, 59, NULL))
588 		panic("main: unable to fork init.");
589 
590 	/* create pageout daemon */
591 	if (newproc(pageout, NULL, syscid, maxclsyspri - 1, NULL))
592 		panic("main: unable to fork pageout()");
593 
594 	/* create fsflush daemon */
595 	if (newproc(fsflush, NULL, syscid, minclsyspri, NULL))
596 		panic("main: unable to fork fsflush()");
597 
598 	/* create cluster process if we're a member of one */
599 	if (cluster_bootflags & CLUSTER_BOOTED) {
600 		if (newproc(cluster_wrapper, NULL, syscid, minclsyspri, NULL))
601 			panic("main: unable to fork cluster()");
602 	}
603 
604 	/*
605 	 * Create system threads (threads are associated with p0)
606 	 */
607 
608 	/* create module uninstall daemon */
609 	/* BugID 1132273. If swapping over NFS need a bigger stack */
610 	(void) thread_create(NULL, 0, (void (*)())mod_uninstall_daemon,
611 	    NULL, 0, &p0, TS_RUN, minclsyspri);
612 
613 	(void) thread_create(NULL, 0, seg_pasync_thread,
614 	    NULL, 0, &p0, TS_RUN, minclsyspri);
615 
616 	pid_setmin();
617 
618 	bcopy("sched", PTOU(curproc)->u_psargs, 6);
619 	bcopy("sched", PTOU(curproc)->u_comm, 5);
620 	sched();
621 	/* NOTREACHED */
622 }
623