Home | History | Annotate | Line # | Download | only in rumpkern
      1 /*	$NetBSD: rump.c,v 1.361 2023/10/05 19:41:07 ad Exp $	*/
      2 
      3 /*
      4  * Copyright (c) 2007-2011 Antti Kantee.  All Rights Reserved.
      5  *
      6  * Redistribution and use in source and binary forms, with or without
      7  * modification, are permitted provided that the following conditions
      8  * are met:
      9  * 1. Redistributions of source code must retain the above copyright
     10  *    notice, this list of conditions and the following disclaimer.
     11  * 2. Redistributions in binary form must reproduce the above copyright
     12  *    notice, this list of conditions and the following disclaimer in the
     13  *    documentation and/or other materials provided with the distribution.
     14  *
     15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS
     16  * OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
     17  * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
     18  * DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
     19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
     21  * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     25  * SUCH DAMAGE.
     26  */
     27 
     28 #include <sys/cdefs.h>
     29 __KERNEL_RCSID(0, "$NetBSD: rump.c,v 1.361 2023/10/05 19:41:07 ad Exp $");
     30 
     31 #include <sys/systm.h>
     32 #define ELFSIZE ARCH_ELFSIZE
     33 
     34 #include <sys/param.h>
     35 #include <sys/atomic.h>
     36 #include <sys/buf.h>
     37 #include <sys/callout.h>
     38 #include <sys/conf.h>
     39 #include <sys/cpu.h>
     40 #include <sys/device.h>
     41 #include <sys/device_impl.h>
     42 #include <sys/evcnt.h>
     43 #include <sys/event.h>
     44 #include <sys/exec_elf.h>
     45 #include <sys/filedesc.h>
     46 #include <sys/iostat.h>
     47 #include <sys/kauth.h>
     48 #include <sys/kcpuset.h>
     49 #include <sys/kernel.h>
     50 #include <sys/kmem.h>
     51 #include <sys/kprintf.h>
     52 #include <sys/kthread.h>
     53 #include <sys/ksyms.h>
     54 #include <sys/msgbuf.h>
     55 #include <sys/module.h>
     56 #include <sys/module_hook.h>
     57 #include <sys/namei.h>
     58 #include <sys/once.h>
     59 #include <sys/percpu.h>
     60 #include <sys/pipe.h>
     61 #include <sys/pool.h>
     62 #include <sys/queue.h>
     63 #include <sys/reboot.h>
     64 #include <sys/resourcevar.h>
     65 #include <sys/select.h>
     66 #include <sys/sysctl.h>
     67 #include <sys/syscall.h>
     68 #include <sys/syscallvar.h>
     69 #include <sys/threadpool.h>
     70 #include <sys/timetc.h>
     71 #include <sys/tty.h>
     72 #include <sys/uidinfo.h>
     73 #include <sys/vmem.h>
     74 #include <sys/xcall.h>
     75 #include <sys/cprng.h>
     76 #include <sys/rnd.h>
     77 #include <sys/ktrace.h>
     78 #include <sys/pserialize.h>
     79 #include <sys/psref.h>
     80 
     81 #include <rump-sys/kern.h>
     82 #include <rump-sys/dev.h>
     83 #include <rump-sys/net.h>
     84 #include <rump-sys/vfs.h>
     85 
     86 #include <rump/rumpuser.h>
     87 
     88 #include <prop/proplib.h>
     89 
     90 #include <uvm/uvm_extern.h>
     91 #include <uvm/uvm_readahead.h>
     92 
     93 char machine[] = MACHINE;
     94 char machine_arch[] = MACHINE_ARCH;
     95 
     96 struct proc *initproc;
     97 
     98 struct device rump_rootdev = {
     99 	.dv_class = DV_VIRTUAL
    100 };
    101 
    102 #ifdef RUMP_WITHOUT_THREADS
    103 int rump_threads = 0;
    104 #else
    105 int rump_threads = 1;
    106 #endif
    107 
    108 static void rump_component_addlocal(void);
    109 static struct lwp *bootlwp;
    110 
    111 /* 16k should be enough for std rump needs */
    112 static  char rump_msgbuf[16*1024] __aligned(256);
    113 
    114 bool rump_ttycomponent = false;
    115 
    116 extern pool_cache_t pnbuf_cache;
    117 
    118 static int rump_inited;
    119 
    120 void (*rump_vfs_drainbufs)(int) = (void *)nullop;
    121 int  (*rump_vfs_makeonedevnode)(dev_t, const char *,
    122 				devmajor_t, devminor_t) = (void *)nullop;
    123 int  (*rump_vfs_makedevnodes)(dev_t, const char *, char,
    124 			      devmajor_t, devminor_t, int) = (void *)nullop;
    125 int  (*rump_vfs_makesymlink)(const char *, const char *) = (void *)nullop;
    126 
    127 rump_proc_vfs_init_fn rump_proc_vfs_init = (void *)nullop;
    128 rump_proc_vfs_release_fn rump_proc_vfs_release = (void *)nullop;
    129 
    130 static void add_linkedin_modules(const struct modinfo *const *, size_t);
    131 static void add_static_evcnt(struct evcnt *);
    132 
    133 static pid_t rspo_wrap_getpid(void) {
    134 	return rump_sysproxy_hyp_getpid();
    135 }
    136 static int rspo_wrap_syscall(int num, void *arg, long *retval) {
    137 	return rump_sysproxy_hyp_syscall(num, arg, retval);
    138 }
    139 static int rspo_wrap_rfork(void *priv, int flag, const char *comm) {
    140 	return rump_sysproxy_hyp_rfork(priv, flag, comm);
    141 }
    142 static void rspo_wrap_lwpexit(void) {
    143 	rump_sysproxy_hyp_lwpexit();
    144 }
    145 static void rspo_wrap_execnotify(const char *comm) {
    146 	rump_sysproxy_hyp_execnotify(comm);
    147 }
    148 static const struct rumpuser_hyperup hyp = {
    149 	.hyp_schedule		= rump_schedule,
    150 	.hyp_unschedule		= rump_unschedule,
    151 	.hyp_backend_unschedule	= rump_user_unschedule,
    152 	.hyp_backend_schedule	= rump_user_schedule,
    153 	.hyp_lwproc_switch	= rump_lwproc_switch,
    154 	.hyp_lwproc_release	= rump_lwproc_releaselwp,
    155 	.hyp_lwproc_newlwp	= rump_lwproc_newlwp,
    156 	.hyp_lwproc_curlwp	= rump_lwproc_curlwp,
    157 
    158 	.hyp_getpid		= rspo_wrap_getpid,
    159 	.hyp_syscall		= rspo_wrap_syscall,
    160 	.hyp_lwproc_rfork	= rspo_wrap_rfork,
    161 	.hyp_lwpexit		= rspo_wrap_lwpexit,
    162 	.hyp_execnotify		= rspo_wrap_execnotify,
    163 };
    164 struct rump_sysproxy_ops rump_sysproxy_ops = {
    165 	.rspo_copyin		= (void *)enxio,
    166 	.rspo_copyinstr 	= (void *)enxio,
    167 	.rspo_copyout	 	= (void *)enxio,
    168 	.rspo_copyoutstr 	= (void *)enxio,
    169 	.rspo_anonmmap 		= (void *)enxio,
    170 	.rspo_raise 		= (void *)enxio,
    171 	.rspo_fini 		= (void *)enxio,
    172 	.rspo_hyp_getpid 	= (void *)enxio,
    173 	.rspo_hyp_syscall 	= (void *)enxio,
    174 	.rspo_hyp_rfork 	= (void *)enxio,
    175 	.rspo_hyp_lwpexit 	= (void *)enxio,
    176 	.rspo_hyp_execnotify 	= (void *)enxio,
    177 };
    178 
    179 int
    180 rump_daemonize_begin(void)
    181 {
    182 
    183 	if (rump_inited)
    184 		return EALREADY;
    185 
    186 	return rumpuser_daemonize_begin();
    187 }
    188 
    189 int
    190 rump_daemonize_done(int error)
    191 {
    192 
    193 	return rumpuser_daemonize_done(error);
    194 }
    195 
    196 #ifdef RUMP_USE_CTOR
    197 
    198 /* sysctl bootstrap handling */
    199 struct sysctl_boot_chain sysctl_boot_chain \
    200     = LIST_HEAD_INITIALIZER(sysctl_boot_chain);
    201 __link_set_add_text(sysctl_funcs,voidop); /* ensure linkset is non-empty */
    202 
    203 #else /* RUMP_USE_CTOR */
    204 
    205 RUMP_COMPONENT(RUMP_COMPONENT_POSTINIT)
    206 {
    207 	__link_set_decl(rump_components, struct rump_component);
    208 
    209 	/*
    210 	 * Trick compiler into generating references so that statically
    211 	 * linked rump kernels are generated with the link set symbols.
    212 	 */
    213 	asm("" :: "r"(__start_link_set_rump_components));
    214 	asm("" :: "r"(__stop_link_set_rump_components));
    215 }
    216 
    217 #endif /* RUMP_USE_CTOR */
    218 
    219 int
    220 rump_init_callback(void (*cpuinit_callback) (void))
    221 {
    222 	char buf[256];
    223 	struct timespec bts;
    224 	int64_t sec;
    225 	long nsec;
    226 	struct lwp *l, *initlwp;
    227 	int i, numcpu;
    228 
    229 	/* not reentrant */
    230 	if (rump_inited)
    231 		return 0;
    232 	else if (rump_inited == -1)
    233 		panic("%s: host process restart required", __func__);
    234 	else
    235 		rump_inited = 1;
    236 
    237 	/* initialize hypervisor */
    238 	if (rumpuser_init(RUMPUSER_VERSION, &hyp) != 0) {
    239 		rumpuser_dprintf("rumpuser init failed\n");
    240 		return EINVAL;
    241 	}
    242 
    243 	/* init minimal lwp/cpu context */
    244 	rump_lwproc_init();
    245 	l = &lwp0;
    246 	l->l_cpu = l->l_target_cpu = &rump_bootcpu;
    247 	rump_lwproc_curlwp_set(l);
    248 
    249 	/* retrieve env vars which affect the early stage of bootstrap */
    250 	if (rumpuser_getparam("RUMP_THREADS", buf, sizeof(buf)) == 0) {
    251 		rump_threads = *buf != '0';
    252 	}
    253 	if (rumpuser_getparam("RUMP_VERBOSE", buf, sizeof(buf)) == 0) {
    254 		if (*buf != '0')
    255 			boothowto = AB_VERBOSE;
    256 	}
    257 
    258 	if (rumpuser_getparam(RUMPUSER_PARAM_NCPU, buf, sizeof(buf)) != 0)
    259 		panic("%s: mandatory hypervisor configuration (NCPU) missing",
    260 		    __func__);
    261 	numcpu = strtoll(buf, NULL, 10);
    262 	if (numcpu < 1) {
    263 		panic("%s: rump kernels are not lightweight enough for %d CPUs",
    264 		    __func__, numcpu);
    265 	}
    266 
    267 	rump_thread_init();
    268 	rump_cpus_bootstrap(&numcpu);
    269 
    270 	rumpuser_clock_gettime(RUMPUSER_CLOCK_RELWALL, &sec, &nsec);
    271 	bts.tv_sec = sec;
    272 	bts.tv_nsec = nsec;
    273 
    274 	initmsgbuf(rump_msgbuf, sizeof(rump_msgbuf));
    275 	aprint_verbose("%s%s", copyright, version);
    276 
    277 	rump_intr_init(numcpu);
    278 
    279 	rump_tsleep_init();
    280 
    281 	rumpuser_mutex_init(&rump_giantlock, RUMPUSER_MTX_SPIN);
    282 	ksyms_init();
    283 	uvm_init();
    284 	evcnt_init();
    285 
    286 	kcpuset_sysinit();
    287 	once_init();
    288 	kernconfig_lock_init();
    289 	prop_kern_init();
    290 
    291 	kmem_init();
    292 
    293 	uvm_ra_init();
    294 	uao_init();
    295 
    296 	callout_startup();
    297 
    298 	kprintf_init();
    299 	percpu_init();
    300 	pserialize_init();
    301 
    302 	kauth_init();
    303 
    304 	secmodel_init();
    305 	sysctl_init();
    306 	/*
    307 	 * The above call to sysctl_init() only initializes sysctl nodes
    308 	 * from link sets.  Initialize sysctls in case we used ctors.
    309 	 */
    310 #ifdef RUMP_USE_CTOR
    311 	{
    312 		struct sysctl_setup_chain *ssc;
    313 
    314 		while ((ssc = LIST_FIRST(&sysctl_boot_chain)) != NULL) {
    315 			LIST_REMOVE(ssc, ssc_entries);
    316 			ssc->ssc_func(NULL);
    317 		}
    318 	}
    319 #endif /* RUMP_USE_CTOR */
    320 
    321 	rnd_init();
    322 	rump_hyperentropy_init();
    323 
    324 	procinit();
    325 	proc0_init();
    326 	uid_init();
    327 	chgproccnt(0, 1);
    328 
    329 	l->l_proc = &proc0;
    330 	l->l_cred = kauth_cred_hold(l->l_proc->p_cred);
    331 
    332 	lwpinit_specificdata();
    333 	lwp_initspecific(&lwp0);
    334 
    335 	/* Must be called after lwpinit_specificdata */
    336 	psref_init();
    337 
    338 	threadpools_init();
    339 
    340 	loginit();
    341 
    342 	rump_biglock_init();
    343 
    344 	rump_scheduler_init(numcpu);
    345 	/* revert temporary context and schedule a semireal context */
    346 	rump_lwproc_curlwp_clear(l);
    347 	initproc = &proc0; /* borrow proc0 before we get initproc started */
    348 	rump_schedule();
    349 	bootlwp = curlwp;
    350 
    351 	inittimecounter();
    352 	ntp_init();
    353 
    354 #ifdef KTRACE
    355 	ktrinit();
    356 #endif
    357 
    358 	tc_setclock(&bts);
    359 
    360 	extern krwlock_t exec_lock;
    361 	rw_init(&exec_lock);
    362 
    363 	/* we are mostly go.  do per-cpu subsystem init */
    364 	for (i = 0; i < numcpu; i++) {
    365 		struct cpu_info *ci = cpu_lookup(i);
    366 
    367 		/* attach non-bootstrap CPUs */
    368 		if (i > 0) {
    369 			rump_cpu_attach(ci);
    370 			ncpu++;
    371 		}
    372 		snprintf(ci->ci_cpuname, sizeof ci->ci_cpuname, "cpu%d", i);
    373 
    374 		callout_init_cpu(ci);
    375 		softint_init(ci);
    376 		xc_init_cpu(ci);
    377 		pool_cache_cpu_init(ci);
    378 		selsysinit(ci);
    379 		percpu_init_cpu(ci);
    380 
    381 		TAILQ_INIT(&ci->ci_data.cpu_ld_locks);
    382 		__cpu_simple_lock_init(&ci->ci_data.cpu_ld_lock);
    383 
    384 		aprint_verbose("cpu%d at thinair0: rump virtual cpu\n", i);
    385 	}
    386 	ncpuonline = ncpu;
    387 
    388 	/* Once all CPUs are detected, initialize the per-CPU cprng_fast.  */
    389 	cprng_init();
    390 	cprng_fast_init();
    391 
    392 	mp_online = true;
    393 
    394 	if (cpuinit_callback)
    395 		(*cpuinit_callback)();
    396 
    397 	/* CPUs are up.  allow kernel threads to run */
    398 	rump_thread_allow(NULL);
    399 
    400 	rnd_init_softint();
    401 
    402 	kqueue_init();
    403 	iostat_init();
    404 	fd_sys_init();
    405 	module_init();
    406 	module_hook_init();
    407 	devsw_init();
    408 	pipe_init();
    409 	resource_init();
    410 	procinit_sysctl();
    411 	time_init();
    412 	config_init();
    413 
    414 	/* start page baroness */
    415 	if (rump_threads) {
    416 		if (kthread_create(PRI_PGDAEMON, KTHREAD_MPSAFE, NULL,
    417 		    uvm_pageout, NULL, &uvm.pagedaemon_lwp, "pdaemon") != 0)
    418 			panic("%s: pagedaemon create failed", __func__);
    419 	} else
    420 		uvm.pagedaemon_lwp = NULL; /* doesn't match curlwp */
    421 
    422 	/* process dso's */
    423 	rumpuser_dl_bootstrap(add_linkedin_modules,
    424 	    rump_kernelfsym_load, rump_component_load, add_static_evcnt);
    425 
    426 	rump_component_addlocal();
    427 	rump_component_init(RUMP_COMPONENT_KERN);
    428 
    429 	/* initialize factions, if present */
    430 	rump_component_init(RUMP__FACTION_VFS);
    431 	/* pnbuf_cache is used even without vfs */
    432 	if (rump_component_count(RUMP__FACTION_VFS) == 0) {
    433 		pnbuf_cache = pool_cache_init(MAXPATHLEN, 0, 0, 0, "pnbufpl",
    434 		    NULL, IPL_NONE, NULL, NULL, NULL);
    435 	}
    436 	rump_component_init(RUMP__FACTION_NET);
    437 	rump_component_init(RUMP__FACTION_DEV);
    438 	KASSERT(rump_component_count(RUMP__FACTION_VFS) <= 1
    439 	    && rump_component_count(RUMP__FACTION_NET) <= 1
    440 	    && rump_component_count(RUMP__FACTION_DEV) <= 1);
    441 
    442 	rump_component_init(RUMP_COMPONENT_KERN_VFS);
    443 
    444 	/*
    445 	 * if we initialized the tty component above, the tyttymtx is
    446 	 * now initialized.  otherwise, we need to initialize it.
    447 	 */
    448 	if (!rump_ttycomponent)
    449 		mutex_init(&tty_lock, MUTEX_DEFAULT, IPL_VM);
    450 
    451 	cold = 0;
    452 
    453 	sysctl_finalize();
    454 
    455 	module_init_class(MODULE_CLASS_ANY);
    456 
    457 	if (rumpuser_getparam(RUMPUSER_PARAM_HOSTNAME,
    458 	    hostname, MAXHOSTNAMELEN) != 0) {
    459 		panic(
    460 		    "%s: mandatory hypervisor configuration (HOSTNAME) missing",
    461 		    __func__);
    462 	}
    463 	hostnamelen = strlen(hostname);
    464 
    465 	sigemptyset(&sigcantmask);
    466 
    467 	if (rump_threads)
    468 		vmem_rehash_start();
    469 
    470 	/*
    471 	 * Create init (proc 1), used to attach implicit threads in rump.
    472 	 * (note: must be done after vfsinit to get cwdi)
    473 	 */
    474 	initlwp = rump__lwproc_alloclwp(NULL);
    475 	mutex_enter(&proc_lock);
    476 	initproc = proc_find_raw(1);
    477 	mutex_exit(&proc_lock);
    478 	if (initproc == NULL)
    479 		panic("%s: where in the world is initproc?", __func__);
    480 	strlcpy(initproc->p_comm, "rumplocal", sizeof(initproc->p_comm));
    481 
    482 	rump_component_init(RUMP_COMPONENT_POSTINIT);
    483 
    484 	/* load syscalls */
    485 	rump_component_init(RUMP_COMPONENT_SYSCALL);
    486 
    487 	/* component inits done */
    488 	bootlwp = NULL;
    489 
    490 	/* open 0/1/2 for init */
    491 	KASSERT(rump_lwproc_curlwp() == NULL);
    492 	rump_lwproc_switch(initlwp);
    493 	rump_consdev_init();
    494 	rump_lwproc_switch(NULL);
    495 
    496 	/* release cpu */
    497 	rump_unschedule();
    498 
    499 	return 0;
    500 }
    501 
    502 int
    503 rump_init(void)
    504 {
    505 	return rump_init_callback(NULL);
    506 }
    507 
    508 /* historic compat */
    509 __strong_alias(rump__init,rump_init);
    510 
    511 static int compcounter[RUMP_COMPONENT_MAX];
    512 static int compinited[RUMP_COMPONENT_MAX];
    513 
    514 /*
    515  * Yea, this is O(n^2), but we're only looking at a handful of components.
    516  * Components are always initialized from the thread that called rump_init().
    517  */
    518 static LIST_HEAD(, rump_component) rchead = LIST_HEAD_INITIALIZER(rchead);
    519 
    520 #ifdef RUMP_USE_CTOR
    521 struct modinfo_boot_chain modinfo_boot_chain \
    522     = LIST_HEAD_INITIALIZER(modinfo_boot_chain);
    523 
    524 static void
    525 rump_component_addlocal(void)
    526 {
    527 	struct modinfo_chain *mc;
    528 
    529 	while ((mc = LIST_FIRST(&modinfo_boot_chain)) != NULL) {
    530 		LIST_REMOVE(mc, mc_entries);
    531 		module_builtin_add(&mc->mc_info, 1, false);
    532 	}
    533 }
    534 
    535 #else /* RUMP_USE_CTOR */
    536 
    537 static void
    538 rump_component_addlocal(void)
    539 {
    540 	__link_set_decl(rump_components, struct rump_component);
    541 	struct rump_component *const *rc;
    542 
    543 	__link_set_foreach(rc, rump_components) {
    544 		rump_component_load(*rc);
    545 	}
    546 }
    547 #endif /* RUMP_USE_CTOR */
    548 
    549 void
    550 rump_component_load(const struct rump_component *rc_const)
    551 {
    552 	struct rump_component *rc, *rc_iter;
    553 
    554 	/* time for rump component loading and unloading has passed */
    555 	if (!cold)
    556 		return;
    557 
    558 	/*
    559 	 * XXX: this is ok since the "const" was removed from the
    560 	 * definition of RUMP_COMPONENT().
    561 	 *
    562 	 * However, to preserve the hypercall interface, the const
    563 	 * remains here.  This can be fixed in the next hypercall revision.
    564 	 */
    565 	rc = __UNCONST(rc_const);
    566 
    567 	KASSERT(!rump_inited || curlwp == bootlwp);
    568 
    569 	LIST_FOREACH(rc_iter, &rchead, rc_entries) {
    570 		if (rc_iter == rc)
    571 			return;
    572 	}
    573 
    574 	LIST_INSERT_HEAD(&rchead, rc, rc_entries);
    575 	KASSERT(rc->rc_type < RUMP_COMPONENT_MAX);
    576 	compcounter[rc->rc_type]++;
    577 }
    578 
    579 void
    580 rump_component_unload(struct rump_component *rc)
    581 {
    582 
    583 	/*
    584 	 * Checking for cold is enough because rump_init() both
    585 	 * flips it and handles component loading.
    586 	 */
    587 	if (!cold)
    588 		return;
    589 
    590 	LIST_REMOVE(rc, rc_entries);
    591 }
    592 
    593 int
    594 rump_component_count(enum rump_component_type type)
    595 {
    596 
    597 	KASSERT(curlwp == bootlwp);
    598 	KASSERT(type < RUMP_COMPONENT_MAX);
    599 	return compcounter[type];
    600 }
    601 
    602 void
    603 rump_component_init(enum rump_component_type type)
    604 {
    605 	struct rump_component *rc, *rc_next, rc_marker;
    606 
    607 	KASSERT(curlwp == bootlwp);
    608 	KASSERT(!compinited[type]);
    609 
    610 	rc_marker.rc_type = RUMP_COMPONENT_MAX;
    611 	rc_marker.rc_init = NULL;
    612 	for (rc = LIST_FIRST(&rchead); rc != NULL; rc = rc_next) {
    613 		if (rc->rc_type == type) {
    614 			LIST_INSERT_AFTER(rc, &rc_marker, rc_entries);
    615 			rc->rc_init();
    616 			LIST_REMOVE(rc, rc_entries);
    617 			rc_next = LIST_NEXT(&rc_marker, rc_entries);
    618 			LIST_REMOVE(&rc_marker, rc_entries);
    619 		} else {
    620 			rc_next = LIST_NEXT(rc, rc_entries);
    621 		}
    622 	}
    623 	compinited[type] = 1;
    624 }
    625 
    626 /*
    627  * Initialize a module which has already been loaded and linked
    628  * with dlopen(). This is fundamentally the same as a builtin module.
    629  *
    630  * XXX: this interface does not really work in the RUMP_USE_CTOR case,
    631  * but I'm not sure it's anything to cry about.  In feeling blue,
    632  * things could somehow be handled via modinfo_boot_chain.
    633  */
    634 int
    635 rump_module_init(const struct modinfo * const *mip, size_t nmodinfo)
    636 {
    637 
    638 	return module_builtin_add(mip, nmodinfo, true);
    639 }
    640 
    641 /*
    642  * Finish module (flawless victory, fatality!).
    643  */
    644 int
    645 rump_module_fini(const struct modinfo *mi)
    646 {
    647 
    648 	return module_builtin_remove(mi, true);
    649 }
    650 
    651 /*
    652  * Add loaded and linked module to the builtin list.  It will
    653  * later be initialized with module_init_class().
    654  */
    655 
    656 static void
    657 add_linkedin_modules(const struct modinfo * const *mip, size_t nmodinfo)
    658 {
    659 
    660 	module_builtin_add(mip, nmodinfo, false);
    661 }
    662 
    663 /*
    664  * Add an evcnt.
    665  */
    666 static void
    667 add_static_evcnt(struct evcnt *ev)
    668 {
    669 
    670 	evcnt_attach_static(ev);
    671 }
    672 
    673 int
    674 rump_kernelfsym_load(void *symtab, uint64_t symsize,
    675 	char *strtab, uint64_t strsize)
    676 {
    677 	static int inited = 0;
    678 	Elf64_Ehdr ehdr;
    679 
    680 	if (inited)
    681 		return EBUSY;
    682 	inited = 1;
    683 
    684 	/*
    685 	 * Use 64bit header since it's bigger.  Shouldn't make a
    686 	 * difference, since we're passing in all zeroes anyway.
    687 	 */
    688 	memset(&ehdr, 0, sizeof(ehdr));
    689 	ksyms_addsyms_explicit(&ehdr, symtab, symsize, strtab, strsize);
    690 
    691 	return 0;
    692 }
    693 
    694 int
    695 rump_boot_gethowto()
    696 {
    697 
    698 	return boothowto;
    699 }
    700 
    701 void
    702 rump_boot_sethowto(int howto)
    703 {
    704 
    705 	boothowto = howto;
    706 }
    707 
    708 int
    709 rump_getversion(void)
    710 {
    711 
    712 	return __NetBSD_Version__;
    713 }
    714 /* compat */
    715 __strong_alias(rump_pub_getversion,rump_getversion);
    716 
    717 /*
    718  * Note: may be called unscheduled.  Not fully safe since no locking
    719  * of allevents (currently that's not even available).
    720  */
    721 void
    722 rump_printevcnts()
    723 {
    724 	struct evcnt *ev;
    725 
    726 	TAILQ_FOREACH(ev, &allevents, ev_list)
    727 		rumpuser_dprintf("%s / %s: %" PRIu64 "\n",
    728 		    ev->ev_group, ev->ev_name, ev->ev_count);
    729 }
    730 
    731 /*
    732  * If you use this interface ... well ... all bets are off.
    733  * The original purpose is for the p2k fs server library to be
    734  * able to use the same pid/lid for VOPs as the host kernel.
    735  */
    736 void
    737 rump_allbetsareoff_setid(pid_t pid, int lid)
    738 {
    739 	struct lwp *l = curlwp;
    740 	struct proc *p = l->l_proc;
    741 
    742 	l->l_lid = lid;
    743 	p->p_pid = pid;
    744 }
    745 
    746 static void
    747 ipiemu(void *a1, void *a2)
    748 {
    749 
    750 	xc__highpri_intr(NULL);
    751 }
    752 
    753 void
    754 rump_xc_highpri(struct cpu_info *ci)
    755 {
    756 
    757 	if (ci)
    758 		xc_unicast(0, ipiemu, NULL, NULL, ci);
    759 	else
    760 		xc_broadcast(0, ipiemu, NULL, NULL);
    761 }
    762 
    763 int
    764 rump_syscall(int num, void *data, size_t dlen, register_t *retval)
    765 {
    766 	struct proc *p;
    767 	struct emul *e;
    768 	struct sysent *callp;
    769 	const int *etrans = NULL;
    770 	int rv;
    771 
    772 	rump_schedule();
    773 	p = curproc;
    774 	e = p->p_emul;
    775 #ifndef __HAVE_MINIMAL_EMUL
    776 	num &= e->e_nsysent - 1;
    777 #else
    778 	num &= SYS_NSYSENT - 1;
    779 #endif
    780 	callp = e->e_sysent + num;
    781 
    782 	rv = sy_invoke(callp, curlwp, data, retval, num);
    783 
    784 	/*
    785 	 * I hope that (!__HAVE_MINIMAL_EMUL || __HAVE_SYSCALL_INTERN) is
    786 	 * an invariant ...
    787 	 */
    788 #if !defined(__HAVE_MINIMAL_EMUL)
    789 	etrans = e->e_errno;
    790 #elif defined(__HAVE_SYSCALL_INTERN)
    791 	etrans = p->p_emuldata;
    792 #endif
    793 
    794 	if (etrans) {
    795 		rv = etrans[rv];
    796 		/*
    797 		 * XXX: small hack since Linux etrans vectors on some
    798 		 * archs contain negative errnos, but rump_syscalls
    799 		 * uses the -1 + errno ABI.  Note that these
    800 		 * negative values are always the result of translation,
    801 		 * otherwise the above translation method would not
    802 		 * work very well.
    803 		 */
    804 		if (rv < 0)
    805 			rv = -rv;
    806 	}
    807 	rump_unschedule();
    808 
    809 	return rv;
    810 }
    811 
    812 void
    813 rump_syscall_boot_establish(const struct rump_onesyscall *calls, size_t ncall)
    814 {
    815 	struct sysent *callp;
    816 	size_t i;
    817 
    818 	for (i = 0; i < ncall; i++) {
    819 		callp = rump_sysent + calls[i].ros_num;
    820 		KASSERT(bootlwp != NULL
    821 		    && callp->sy_call == (sy_call_t *)(void *)enosys);
    822 		callp->sy_call = calls[i].ros_handler;
    823 	}
    824 }
    825 
    826 struct rump_boot_etfs *ebstart;
    827 void
    828 rump_boot_etfs_register(struct rump_boot_etfs *eb)
    829 {
    830 
    831 	/*
    832 	 * Could use atomics, but, since caller would need to synchronize
    833 	 * against calling rump_init() anyway, easier to just specify the
    834 	 * interface as "caller serializes".  This solve-by-specification
    835 	 * approach avoids the grey area of using atomics before rump_init()
    836 	 * runs.
    837 	 */
    838 	eb->_eb_next = ebstart;
    839 	eb->eb_status = -1;
    840 	ebstart = eb;
    841 }
    842