Home | History | Annotate | Line # | Download | only in rumpkern
emul.c revision 1.169 
      1 /*	$NetBSD: emul.c,v 1.169 2015/01/03 17:23:51 pooka 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: emul.c,v 1.169 2015/01/03 17:23:51 pooka Exp $");
     30 
     31 #include <sys/param.h>
     32 #include <sys/null.h>
     33 #include <sys/vnode.h>
     34 #include <sys/stat.h>
     35 #include <sys/select.h>
     36 #include <sys/syslog.h>
     37 #include <sys/namei.h>
     38 #include <sys/kauth.h>
     39 #include <sys/conf.h>
     40 #include <sys/device.h>
     41 #include <sys/queue.h>
     42 #include <sys/file.h>
     43 #include <sys/filedesc.h>
     44 #include <sys/cpu.h>
     45 #include <sys/kmem.h>
     46 #include <sys/poll.h>
     47 #include <sys/timetc.h>
     48 #include <sys/tprintf.h>
     49 #include <sys/module.h>
     50 #include <sys/tty.h>
     51 #include <sys/reboot.h>
     52 #include <sys/syscall.h>
     53 #include <sys/syscallvar.h>
     54 #include <sys/xcall.h>
     55 #include <sys/sleepq.h>
     56 #include <sys/cprng.h>
     57 
     58 #include <dev/cons.h>
     59 
     60 #include <rump/rumpuser.h>
     61 
     62 #include <uvm/uvm_map.h>
     63 
     64 #include "rump_private.h"
     65 
     66 void (*rump_vfs_fini)(void) = (void *)nullop;
     67 
     68 /*
     69  * physmem is largely unused (except for nmbcluster calculations),
     70  * so pick a default value which suits ZFS.  if an application wants
     71  * a very small memory footprint, it can still adjust this before
     72  * calling rump_init()
     73  */
     74 #define PHYSMEM 512*256
     75 int physmem = PHYSMEM;
     76 int nkmempages = PHYSMEM/2; /* from le chapeau */
     77 #undef PHYSMEM
     78 
     79 struct lwp lwp0 = {
     80 	.l_lid = 1,
     81 	.l_proc = &proc0,
     82 	.l_fd = &filedesc0,
     83 };
     84 struct vnode *rootvp;
     85 dev_t rootdev = NODEV;
     86 
     87 const int schedppq = 1;
     88 int hardclock_ticks;
     89 bool mp_online = false;
     90 struct timeval boottime;
     91 int cold = 1;
     92 int boothowto = AB_SILENT;
     93 struct tty *constty;
     94 
     95 const struct bdevsw *bdevsw0[255];
     96 const struct bdevsw **bdevsw = bdevsw0;
     97 const int sys_cdevsws = 255;
     98 int max_cdevsws = 255;
     99 
    100 const struct cdevsw *cdevsw0[255];
    101 const struct cdevsw **cdevsw = cdevsw0;
    102 const int sys_bdevsws = 255;
    103 int max_bdevsws = 255;
    104 
    105 int mem_no = 2;
    106 
    107 device_t booted_device;
    108 device_t booted_wedge;
    109 int booted_partition;
    110 
    111 /* XXX: unused */
    112 kmutex_t tty_lock;
    113 krwlock_t exec_lock;
    114 
    115 struct lwplist alllwp = LIST_HEAD_INITIALIZER(alllwp);
    116 
    117 /* sparc doesn't sport constant page size, pretend we have 4k pages */
    118 #ifdef __sparc__
    119 int nbpg = 4096;
    120 int pgofset = 4096-1;
    121 int pgshift = 12;
    122 #endif
    123 
    124 /* on sun3 VM_MAX_ADDRESS is a const variable */
    125 /* XXX: should be moved into rump.c and initialize for sun3 and sun3x? */
    126 #ifdef sun3
    127 const vaddr_t kernbase = KERNBASE3;
    128 #endif
    129 
    130 struct loadavg averunnable = {
    131 	{ 0 * FSCALE,
    132 	  1 * FSCALE,
    133 	  11 * FSCALE, },
    134 	FSCALE,
    135 };
    136 
    137 struct emul emul_netbsd = {
    138 	.e_name = "netbsd-rump",
    139 	.e_sysent = rump_sysent,
    140 #ifndef __HAVE_MINIMAL_EMUL
    141 	.e_nsysent = SYS_NSYSENT,
    142 #endif
    143 	.e_vm_default_addr = uvm_default_mapaddr,
    144 #ifdef __HAVE_SYSCALL_INTERN
    145 	.e_syscall_intern = syscall_intern,
    146 #endif
    147 };
    148 
    149 u_int nprocs = 1;
    150 
    151 cprng_strong_t *kern_cprng;
    152 
    153 /* not used, but need the symbols for pointer comparisons */
    154 syncobj_t mutex_syncobj, rw_syncobj;
    155 
    156 int
    157 kpause(const char *wmesg, bool intr, int timeo, kmutex_t *mtx)
    158 {
    159 	extern int hz;
    160 	int rv __diagused;
    161 	uint64_t sec, nsec;
    162 
    163 	if (mtx)
    164 		mutex_exit(mtx);
    165 
    166 	sec = timeo / hz;
    167 	nsec = (timeo % hz) * (1000000000 / hz);
    168 	rv = rumpuser_clock_sleep(RUMPUSER_CLOCK_RELWALL, sec, nsec);
    169 	KASSERT(rv == 0);
    170 
    171 	if (mtx)
    172 		mutex_enter(mtx);
    173 
    174 	return 0;
    175 }
    176 
    177 void
    178 lwp_unsleep(lwp_t *l, bool cleanup)
    179 {
    180 
    181 	KASSERT(mutex_owned(l->l_mutex));
    182 
    183 	(*l->l_syncobj->sobj_unsleep)(l, cleanup);
    184 }
    185 
    186 void
    187 lwp_update_creds(struct lwp *l)
    188 {
    189 	struct proc *p;
    190 	kauth_cred_t oldcred;
    191 
    192 	p = l->l_proc;
    193 	oldcred = l->l_cred;
    194 	l->l_prflag &= ~LPR_CRMOD;
    195 
    196 	mutex_enter(p->p_lock);
    197 	kauth_cred_hold(p->p_cred);
    198 	l->l_cred = p->p_cred;
    199 	mutex_exit(p->p_lock);
    200 
    201 	if (oldcred != NULL)
    202 		kauth_cred_free(oldcred);
    203 }
    204 
    205 vaddr_t
    206 calc_cache_size(vsize_t vasz, int pct, int va_pct)
    207 {
    208 	paddr_t t;
    209 
    210 	t = (paddr_t)physmem * pct / 100 * PAGE_SIZE;
    211 	if ((vaddr_t)t != t) {
    212 		panic("%s: needs tweak", __func__);
    213 	}
    214 	return t;
    215 }
    216 
    217 void
    218 assert_sleepable(void)
    219 {
    220 
    221 	/* always sleepable, although we should improve this */
    222 }
    223 
    224 void
    225 module_init_md(void)
    226 {
    227 
    228 	/*
    229 	 * Nothing for now.  However, we should load the librump
    230 	 * symbol table.
    231 	 */
    232 }
    233 
    234 /*
    235  * Try to emulate all the MD definitions of DELAY() / delay().
    236  * Would be nice to fix the #defines in MD headers, but this quicker.
    237  *
    238  * XXX: we'd need a rumpuser_clock_sleep_nowrap() here.  Since we
    239  * don't have it in the current hypercall revision, busyloop.
    240  * Note that rather than calibrate a loop delay and work with that,
    241  * get call gettime (which does not block) in a loop to make sure
    242  * we didn't get virtual ghosttime.  That might be slightly inaccurate
    243  * for very small delays ...
    244  *
    245  * The other option would be to run a thread in the hypervisor which
    246  * sleeps for us and we can wait for it using rumpuser_cv_wait_nowrap()
    247  * Probably too fussy.  Better just wait for hypercall rev 18 ;)
    248  */
    249 static void
    250 rump_delay(unsigned int us)
    251 {
    252 	struct timespec target, tmp;
    253 	uint64_t sec, sec_ini, sec_now;
    254 	long nsec, nsec_ini, nsec_now;
    255 	int loops;
    256 
    257 	rumpuser_clock_gettime(RUMPUSER_CLOCK_ABSMONO, &sec_ini, &nsec_ini);
    258 
    259 #ifdef __mac68k__
    260 	sec = us / 1000;
    261 	nsec = (us % 1000) * 1000000;
    262 #else
    263 	sec = us / 1000000;
    264 	nsec = (us % 1000000) * 1000;
    265 #endif
    266 
    267 	target.tv_sec = sec_ini;
    268 	tmp.tv_sec = sec;
    269 	target.tv_nsec = nsec_ini;
    270 	tmp.tv_nsec = nsec;
    271 	timespecadd(&target, &tmp, &target);
    272 
    273 	if (__predict_false(sec != 0))
    274 		printf("WARNING: over 1s delay\n");
    275 
    276 	for (loops = 0; loops < 1000*1000*100; loops++) {
    277 		struct timespec cur;
    278 
    279 		rumpuser_clock_gettime(RUMPUSER_CLOCK_ABSMONO,
    280 		    &sec_now, &nsec_now);
    281 		cur.tv_sec = sec_now;
    282 		cur.tv_nsec = nsec_now;
    283 		if (timespeccmp(&cur, &target, >=)) {
    284 			return;
    285 		}
    286 	}
    287 	printf("WARNING: DELAY ESCAPED\n");
    288 }
    289 void (*delay_func)(unsigned int) = rump_delay;
    290 __strong_alias(delay,rump_delay);
    291 __strong_alias(_delay,rump_delay);
    292 
    293 /*
    294  * Provide weak aliases for tty routines used by printf.
    295  * They will be used unless the rumpkern_tty component is present.
    296  */
    297 
    298 int rump_ttycheckoutq(struct tty *, int);
    299 int
    300 rump_ttycheckoutq(struct tty *tp, int wait)
    301 {
    302 
    303 	return 1;
    304 }
    305 __weak_alias(ttycheckoutq,rump_ttycheckoutq);
    306 
    307 int rump_tputchar(int, int, struct tty *);
    308 int
    309 rump_tputchar(int c, int flags, struct tty *tp)
    310 {
    311 
    312 	cnputc(c);
    313 	return 0;
    314 }
    315 __weak_alias(tputchar,rump_tputchar);
    316 
    317 void
    318 cnputc(int c)
    319 {
    320 
    321 	rumpuser_putchar(c);
    322 }
    323 
    324 void
    325 cnflush(void)
    326 {
    327 
    328 	/* done */
    329 }
    330 
    331 void
    332 resettodr(void)
    333 {
    334 
    335 	/* setting clocks is not in the jurisdiction of rump kernels */
    336 }
    337 
    338 #ifdef __HAVE_SYSCALL_INTERN
    339 void
    340 syscall_intern(struct proc *p)
    341 {
    342 
    343 	p->p_emuldata = NULL;
    344 }
    345 #endif
    346 
    347 #ifdef LOCKDEBUG
    348 void
    349 turnstile_print(volatile void *obj, void (*pr)(const char *, ...))
    350 {
    351 
    352 	/* nada */
    353 }
    354 #endif
    355 
    356 void
    357 cpu_reboot(int howto, char *bootstr)
    358 {
    359 	int ruhow = 0;
    360 	void *finiarg;
    361 
    362 	printf("rump kernel halting...\n");
    363 
    364 	if (!RUMP_LOCALPROC_P(curproc))
    365 		finiarg = curproc->p_vmspace->vm_map.pmap;
    366 	else
    367 		finiarg = NULL;
    368 
    369 	/* dump means we really take the dive here */
    370 	if ((howto & RB_DUMP) || panicstr) {
    371 		ruhow = RUMPUSER_PANIC;
    372 		goto out;
    373 	}
    374 
    375 	/* try to sync */
    376 	if (!((howto & RB_NOSYNC) || panicstr)) {
    377 		rump_vfs_fini();
    378 	}
    379 
    380 	doshutdownhooks();
    381 
    382 	/* your wish is my command */
    383 	if (howto & RB_HALT) {
    384 		printf("rump kernel halted\n");
    385 		rump_sysproxy_fini(finiarg);
    386 		for (;;) {
    387 			rumpuser_clock_sleep(RUMPUSER_CLOCK_RELWALL, 10, 0);
    388 		}
    389 	}
    390 
    391 	/* this function is __dead, we must exit */
    392  out:
    393 	printf("halted\n");
    394 	rump_sysproxy_fini(finiarg);
    395 	rumpuser_exit(ruhow);
    396 }
    397