xref: /illumos-gate/usr/src/uts/common/io/mem.c (revision 507c32411f3f101e90ca2120f042b5ee698ba1d5)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 /*
29  * Memory special file
30  */
31 
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/user.h>
35 #include <sys/buf.h>
36 #include <sys/systm.h>
37 #include <sys/cred.h>
38 #include <sys/vm.h>
39 #include <sys/uio.h>
40 #include <sys/mman.h>
41 #include <sys/kmem.h>
42 #include <vm/seg.h>
43 #include <vm/page.h>
44 #include <sys/stat.h>
45 #include <sys/vmem.h>
46 #include <sys/memlist.h>
47 #include <sys/bootconf.h>
48 
49 #include <vm/seg_vn.h>
50 #include <vm/seg_dev.h>
51 #include <vm/seg_kmem.h>
52 #include <vm/seg_kp.h>
53 #include <vm/seg_kpm.h>
54 #include <vm/hat.h>
55 
56 #include <sys/conf.h>
57 #include <sys/mem.h>
58 #include <sys/types.h>
59 #include <sys/conf.h>
60 #include <sys/param.h>
61 #include <sys/systm.h>
62 #include <sys/errno.h>
63 #include <sys/modctl.h>
64 #include <sys/memlist.h>
65 #include <sys/ddi.h>
66 #include <sys/sunddi.h>
67 #include <sys/debug.h>
68 #include <sys/fm/protocol.h>
69 
70 #if defined(__sparc)
71 extern int cpu_get_mem_name(uint64_t, uint64_t *, uint64_t, char *, int, int *);
72 extern int cpu_get_mem_info(uint64_t, uint64_t, uint64_t *, uint64_t *,
73     uint64_t *, int *, int *, int *);
74 extern size_t cpu_get_name_bufsize(void);
75 extern int cpu_get_mem_sid(char *, char *, int, int *);
76 extern int cpu_get_mem_addr(char *, char *, uint64_t, uint64_t *);
77 #elif defined(__i386) || defined(__amd64)
78 #include <sys/cpu_module.h>
79 #endif	/* __sparc */
80 
81 /*
82  * Turn a byte length into a pagecount.  The DDI btop takes a
83  * 32-bit size on 32-bit machines, this handles 64-bit sizes for
84  * large physical-memory 32-bit machines.
85  */
86 #define	BTOP(x)	((pgcnt_t)((x) >> _pageshift))
87 
88 static kmutex_t mm_lock;
89 static caddr_t mm_map;
90 
91 static dev_info_t *mm_dip;	/* private copy of devinfo pointer */
92 
93 static int mm_kmem_io_access;
94 
95 static int mm_kstat_update(kstat_t *ksp, int rw);
96 static int mm_kstat_snapshot(kstat_t *ksp, void *buf, int rw);
97 
98 static int mm_read_mem_name(intptr_t data, mem_name_t *mem_name);
99 static int mm_read_mem_page(intptr_t data, mem_page_t *mpage);
100 static int mm_get_mem_fmri(mem_page_t *mpage, nvlist_t **nvl);
101 static int mm_get_paddr(nvlist_t *nvl, uint64_t *paddr);
102 
103 /*ARGSUSED1*/
104 static int
105 mm_attach(dev_info_t *devi, ddi_attach_cmd_t cmd)
106 {
107 	int i;
108 	struct mem_minor {
109 		char *name;
110 		minor_t minor;
111 		int privonly;
112 		const char *rdpriv;
113 		const char *wrpriv;
114 		mode_t priv_mode;
115 	} mm[] = {
116 		{ "mem",	M_MEM,		0,	NULL,	"all",	0640 },
117 		{ "kmem",	M_KMEM,		0,	NULL,	"all",	0640 },
118 		{ "allkmem",	M_ALLKMEM,	0,	"all",	"all",	0600 },
119 		{ "null",	M_NULL,	PRIVONLY_DEV,	NULL,	NULL,	0666 },
120 		{ "zero",	M_ZERO, PRIVONLY_DEV,	NULL,	NULL,	0666 },
121 	};
122 	kstat_t *ksp;
123 
124 	mutex_init(&mm_lock, NULL, MUTEX_DEFAULT, NULL);
125 	mm_map = vmem_alloc(heap_arena, PAGESIZE, VM_SLEEP);
126 
127 	for (i = 0; i < (sizeof (mm) / sizeof (mm[0])); i++) {
128 		if (ddi_create_priv_minor_node(devi, mm[i].name, S_IFCHR,
129 		    mm[i].minor, DDI_PSEUDO, mm[i].privonly,
130 		    mm[i].rdpriv, mm[i].wrpriv, mm[i].priv_mode) ==
131 		    DDI_FAILURE) {
132 			ddi_remove_minor_node(devi, NULL);
133 			return (DDI_FAILURE);
134 		}
135 	}
136 
137 	mm_dip = devi;
138 
139 	ksp = kstat_create("mm", 0, "phys_installed", "misc",
140 	    KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VAR_SIZE | KSTAT_FLAG_VIRTUAL);
141 	if (ksp != NULL) {
142 		ksp->ks_update = mm_kstat_update;
143 		ksp->ks_snapshot = mm_kstat_snapshot;
144 		ksp->ks_lock = &mm_lock; /* XXX - not really needed */
145 		kstat_install(ksp);
146 	}
147 
148 	mm_kmem_io_access = ddi_getprop(DDI_DEV_T_ANY, devi, DDI_PROP_DONTPASS,
149 	    "kmem_io_access", 0);
150 
151 	return (DDI_SUCCESS);
152 }
153 
154 /*ARGSUSED*/
155 static int
156 mm_info(dev_info_t *dip, ddi_info_cmd_t infocmd, void *arg, void **result)
157 {
158 	register int error;
159 
160 	switch (infocmd) {
161 	case DDI_INFO_DEVT2DEVINFO:
162 		*result = (void *)mm_dip;
163 		error = DDI_SUCCESS;
164 		break;
165 	case DDI_INFO_DEVT2INSTANCE:
166 		*result = (void *)0;
167 		error = DDI_SUCCESS;
168 		break;
169 	default:
170 		error = DDI_FAILURE;
171 	}
172 	return (error);
173 }
174 
175 /*ARGSUSED1*/
176 static int
177 mmopen(dev_t *devp, int flag, int typ, struct cred *cred)
178 {
179 	switch (getminor(*devp)) {
180 	case M_NULL:
181 	case M_ZERO:
182 	case M_MEM:
183 	case M_KMEM:
184 	case M_ALLKMEM:
185 		/* standard devices */
186 		break;
187 
188 	default:
189 		/* Unsupported or unknown type */
190 		return (EINVAL);
191 	}
192 	return (0);
193 }
194 
195 struct pollhead	mm_pollhd;
196 
197 /*ARGSUSED*/
198 static int
199 mmchpoll(dev_t dev, short events, int anyyet, short *reventsp,
200     struct pollhead **phpp)
201 {
202 	switch (getminor(dev)) {
203 	case M_NULL:
204 	case M_ZERO:
205 	case M_MEM:
206 	case M_KMEM:
207 	case M_ALLKMEM:
208 		*reventsp = events & (POLLIN | POLLOUT | POLLPRI | POLLRDNORM |
209 			POLLWRNORM | POLLRDBAND | POLLWRBAND);
210 		/*
211 		 * A non NULL pollhead pointer should be returned in case
212 		 * user polls for 0 events.
213 		 */
214 		*phpp = !anyyet && !*reventsp ?
215 		    &mm_pollhd : (struct pollhead *)NULL;
216 		return (0);
217 	default:
218 		/* no other devices currently support polling */
219 		return (ENXIO);
220 	}
221 }
222 
223 static int
224 mmpropop(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int flags,
225     char *name, caddr_t valuep, int *lengthp)
226 {
227 	/*
228 	 * implement zero size to reduce overhead (avoid two failing
229 	 * property lookups per stat).
230 	 */
231 	return (ddi_prop_op_size(dev, dip, prop_op,
232 	    flags, name, valuep, lengthp, 0));
233 }
234 
235 static int
236 mmio(struct uio *uio, enum uio_rw rw, pfn_t pfn, off_t pageoff, int allowio)
237 {
238 	int error = 0;
239 	size_t nbytes = MIN((size_t)(PAGESIZE - pageoff),
240 	    (size_t)uio->uio_iov->iov_len);
241 
242 	mutex_enter(&mm_lock);
243 	hat_devload(kas.a_hat, mm_map, PAGESIZE, pfn,
244 	    (uint_t)(rw == UIO_READ ? PROT_READ : PROT_READ | PROT_WRITE),
245 	    HAT_LOAD_NOCONSIST | HAT_LOAD_LOCK);
246 
247 	if (!pf_is_memory(pfn)) {
248 		if (allowio) {
249 			size_t c = uio->uio_iov->iov_len;
250 
251 			if (ddi_peekpokeio(NULL, uio, rw,
252 			    (caddr_t)(uintptr_t)uio->uio_loffset, c,
253 			    sizeof (int32_t)) != DDI_SUCCESS)
254 				error = EFAULT;
255 		} else
256 			error = EIO;
257 	} else
258 		error = uiomove(&mm_map[pageoff], nbytes, rw, uio);
259 
260 	hat_unload(kas.a_hat, mm_map, PAGESIZE, HAT_UNLOAD_UNLOCK);
261 	mutex_exit(&mm_lock);
262 	return (error);
263 }
264 
265 #ifdef	__sparc
266 
267 static int
268 mmpagelock(struct as *as, caddr_t va)
269 {
270 	struct seg *seg;
271 	int i;
272 
273 	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
274 	seg = as_segat(as, va);
275 	i = (seg != NULL)? SEGOP_CAPABLE(seg, S_CAPABILITY_NOMINFLT) : 0;
276 	AS_LOCK_EXIT(as, &as->a_lock);
277 
278 	return (i);
279 }
280 
281 #define	NEED_LOCK_KVADDR(kva)	mmpagelock(&kas, kva)
282 
283 #else	/* __i386, __amd64 */
284 
285 #define	NEED_LOCK_KVADDR(va)	0
286 
287 #endif	/* __sparc */
288 
289 /*ARGSUSED3*/
290 static int
291 mmrw(dev_t dev, struct uio *uio, enum uio_rw rw, cred_t *cred)
292 {
293 	pfn_t v;
294 	struct iovec *iov;
295 	int error = 0;
296 	size_t c;
297 	ssize_t oresid = uio->uio_resid;
298 	minor_t minor = getminor(dev);
299 
300 	while (uio->uio_resid > 0 && error == 0) {
301 		iov = uio->uio_iov;
302 		if (iov->iov_len == 0) {
303 			uio->uio_iov++;
304 			uio->uio_iovcnt--;
305 			if (uio->uio_iovcnt < 0)
306 				panic("mmrw");
307 			continue;
308 		}
309 		switch (minor) {
310 
311 		case M_MEM:
312 			memlist_read_lock();
313 			if (!address_in_memlist(phys_install,
314 			    (uint64_t)uio->uio_loffset, 1)) {
315 				memlist_read_unlock();
316 				error = EFAULT;
317 				break;
318 			}
319 			memlist_read_unlock();
320 
321 			v = BTOP((u_offset_t)uio->uio_loffset);
322 			error = mmio(uio, rw, v,
323 			    uio->uio_loffset & PAGEOFFSET, 0);
324 			break;
325 
326 		case M_KMEM:
327 		case M_ALLKMEM:
328 			{
329 			page_t **ppp;
330 			caddr_t vaddr = (caddr_t)uio->uio_offset;
331 			int try_lock = NEED_LOCK_KVADDR(vaddr);
332 			int locked = 0;
333 
334 			/*
335 			 * If vaddr does not map a valid page, as_pagelock()
336 			 * will return failure. Hence we can't check the
337 			 * return value and return EFAULT here as we'd like.
338 			 * seg_kp and seg_kpm do not properly support
339 			 * as_pagelock() for this context so we avoid it
340 			 * using the try_lock set check above.  Some day when
341 			 * the kernel page locking gets redesigned all this
342 			 * muck can be cleaned up.
343 			 */
344 			if (try_lock)
345 				locked = (as_pagelock(&kas, &ppp, vaddr,
346 				    PAGESIZE, S_WRITE) == 0);
347 
348 			v = hat_getpfnum(kas.a_hat,
349 			    (caddr_t)(uintptr_t)uio->uio_loffset);
350 			if (v == PFN_INVALID) {
351 				if (locked)
352 					as_pageunlock(&kas, ppp, vaddr,
353 					    PAGESIZE, S_WRITE);
354 				error = EFAULT;
355 				break;
356 			}
357 
358 			error = mmio(uio, rw, v, uio->uio_loffset & PAGEOFFSET,
359 			    minor == M_ALLKMEM || mm_kmem_io_access);
360 			if (locked)
361 				as_pageunlock(&kas, ppp, vaddr, PAGESIZE,
362 				    S_WRITE);
363 			}
364 
365 			break;
366 
367 		case M_ZERO:
368 			if (rw == UIO_READ) {
369 				label_t ljb;
370 
371 				if (on_fault(&ljb)) {
372 					no_fault();
373 					error = EFAULT;
374 					break;
375 				}
376 				uzero(iov->iov_base, iov->iov_len);
377 				no_fault();
378 				uio->uio_resid -= iov->iov_len;
379 				uio->uio_loffset += iov->iov_len;
380 				break;
381 			}
382 			/* else it's a write, fall through to NULL case */
383 			/*FALLTHROUGH*/
384 
385 		case M_NULL:
386 			if (rw == UIO_READ)
387 				return (0);
388 			c = iov->iov_len;
389 			iov->iov_base += c;
390 			iov->iov_len -= c;
391 			uio->uio_loffset += c;
392 			uio->uio_resid -= c;
393 			break;
394 
395 		}
396 	}
397 	return (uio->uio_resid == oresid ? error : 0);
398 }
399 
400 static int
401 mmread(dev_t dev, struct uio *uio, cred_t *cred)
402 {
403 	return (mmrw(dev, uio, UIO_READ, cred));
404 }
405 
406 static int
407 mmwrite(dev_t dev, struct uio *uio, cred_t *cred)
408 {
409 	return (mmrw(dev, uio, UIO_WRITE, cred));
410 }
411 
412 /*
413  * Private ioctl for libkvm to support kvm_physaddr().
414  * Given an address space and a VA, compute the PA.
415  */
416 static int
417 mmioctl_vtop(intptr_t data)
418 {
419 #ifdef _SYSCALL32
420 	mem_vtop32_t vtop32;
421 #endif
422 	mem_vtop_t mem_vtop;
423 	proc_t *p;
424 	pfn_t pfn = (pfn_t)PFN_INVALID;
425 	pid_t pid = 0;
426 	struct as *as;
427 	struct seg *seg;
428 
429 	if (get_udatamodel() == DATAMODEL_NATIVE) {
430 		if (copyin((void *)data, &mem_vtop, sizeof (mem_vtop_t)))
431 			return (EFAULT);
432 	}
433 #ifdef _SYSCALL32
434 	else {
435 		if (copyin((void *)data, &vtop32, sizeof (mem_vtop32_t)))
436 			return (EFAULT);
437 		mem_vtop.m_as = (struct as *)vtop32.m_as;
438 		mem_vtop.m_va = (void *)vtop32.m_va;
439 
440 		if (mem_vtop.m_as != NULL)
441 			return (EINVAL);
442 	}
443 #endif
444 
445 	if (mem_vtop.m_as == &kas) {
446 		pfn = hat_getpfnum(kas.a_hat, mem_vtop.m_va);
447 	} else {
448 		if (mem_vtop.m_as == NULL) {
449 			/*
450 			 * Assume the calling process's address space if the
451 			 * caller didn't specify one.
452 			 */
453 			p = curthread->t_procp;
454 			if (p == NULL)
455 				return (EIO);
456 			mem_vtop.m_as = p->p_as;
457 		}
458 
459 		mutex_enter(&pidlock);
460 		for (p = practive; p != NULL; p = p->p_next) {
461 			if (p->p_as == mem_vtop.m_as) {
462 				pid = p->p_pid;
463 				break;
464 			}
465 		}
466 		mutex_exit(&pidlock);
467 		if (p == NULL)
468 			return (EIO);
469 		p = sprlock(pid);
470 		if (p == NULL)
471 			return (EIO);
472 		as = p->p_as;
473 		if (as == mem_vtop.m_as) {
474 			mutex_exit(&p->p_lock);
475 			AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
476 			for (seg = AS_SEGFIRST(as); seg != NULL;
477 			    seg = AS_SEGNEXT(as, seg))
478 				if ((uintptr_t)mem_vtop.m_va -
479 				    (uintptr_t)seg->s_base < seg->s_size)
480 					break;
481 			if (seg != NULL)
482 				pfn = hat_getpfnum(as->a_hat, mem_vtop.m_va);
483 			AS_LOCK_EXIT(as, &as->a_lock);
484 			mutex_enter(&p->p_lock);
485 		}
486 		sprunlock(p);
487 	}
488 	mem_vtop.m_pfn = pfn;
489 	if (pfn == PFN_INVALID)
490 		return (EIO);
491 
492 	if (get_udatamodel() == DATAMODEL_NATIVE) {
493 		if (copyout(&mem_vtop, (void *)data, sizeof (mem_vtop_t)))
494 			return (EFAULT);
495 	}
496 #ifdef _SYSCALL32
497 	else {
498 		vtop32.m_pfn = mem_vtop.m_pfn;
499 		if (copyout(&vtop32, (void *)data, sizeof (mem_vtop32_t)))
500 			return (EFAULT);
501 	}
502 #endif
503 
504 	return (0);
505 }
506 
507 /*
508  * Given a PA, execute the given page retire command on it.
509  */
510 static int
511 mmioctl_page_retire(int cmd, intptr_t data)
512 {
513 	extern int page_retire_test(void);
514 	uint64_t pa;
515 
516 	if (copyin((void *)data, &pa, sizeof (uint64_t))) {
517 		return (EFAULT);
518 	}
519 
520 	switch (cmd) {
521 	case MEM_PAGE_ISRETIRED:
522 		return (page_retire_check(pa, NULL));
523 
524 	case MEM_PAGE_UNRETIRE:
525 		return (page_unretire(pa));
526 
527 	case MEM_PAGE_RETIRE:
528 		return (page_retire(pa, PR_FMA));
529 
530 	case MEM_PAGE_RETIRE_MCE:
531 		return (page_retire(pa, PR_MCE));
532 
533 	case MEM_PAGE_RETIRE_UE:
534 		return (page_retire(pa, PR_UE));
535 
536 	case MEM_PAGE_GETERRORS:
537 		{
538 			uint64_t page_errors;
539 			int rc = page_retire_check(pa, &page_errors);
540 			if (copyout(&page_errors, (void *)data,
541 			    sizeof (uint64_t))) {
542 				return (EFAULT);
543 			}
544 			return (rc);
545 		}
546 
547 	case MEM_PAGE_RETIRE_TEST:
548 		return (page_retire_test());
549 
550 	}
551 
552 	return (EINVAL);
553 }
554 
555 /*
556  * Given a mem-scheme FMRI for a page, execute the given page retire
557  * command on it.
558  */
559 static int
560 mmioctl_page_fmri_retire(int cmd, intptr_t data)
561 {
562 	mem_page_t mpage;
563 	uint64_t pa;
564 	nvlist_t *nvl;
565 	int err;
566 
567 	if ((err = mm_read_mem_page(data, &mpage)) < 0)
568 		return (err);
569 
570 	if ((err = mm_get_mem_fmri(&mpage, &nvl)) < 0)
571 		return (err);
572 
573 	if ((err = mm_get_paddr(nvl, &pa)) != 0) {
574 		nvlist_free(nvl);
575 		return (err);
576 	}
577 
578 	nvlist_free(nvl);
579 
580 	switch (cmd) {
581 	case MEM_PAGE_FMRI_ISRETIRED:
582 		return (page_retire_check(pa, NULL));
583 
584 	case MEM_PAGE_FMRI_RETIRE:
585 		return (page_retire(pa, PR_FMA));
586 	}
587 
588 	return (EINVAL);
589 }
590 
591 #ifdef __sparc
592 /*
593  * Given a syndrome, syndrome type, and address return the
594  * associated memory name in the provided data buffer.
595  */
596 static int
597 mmioctl_get_mem_name(intptr_t data)
598 {
599 	mem_name_t mem_name;
600 	void *buf;
601 	size_t bufsize;
602 	int len, err;
603 
604 	if ((bufsize = cpu_get_name_bufsize()) == 0)
605 		return (ENOTSUP);
606 
607 	if ((err = mm_read_mem_name(data, &mem_name)) < 0)
608 		return (err);
609 
610 	buf = kmem_alloc(bufsize, KM_SLEEP);
611 
612 	/*
613 	 * Call into cpu specific code to do the lookup.
614 	 */
615 	if ((err = cpu_get_mem_name(mem_name.m_synd, mem_name.m_type,
616 	    mem_name.m_addr, buf, bufsize, &len)) != 0) {
617 		kmem_free(buf, bufsize);
618 		return (err);
619 	}
620 
621 	if (len >= mem_name.m_namelen) {
622 		kmem_free(buf, bufsize);
623 		return (ENAMETOOLONG);
624 	}
625 
626 	if (copyoutstr(buf, (char *)mem_name.m_name,
627 	    mem_name.m_namelen, NULL) != 0) {
628 		kmem_free(buf, bufsize);
629 		return (EFAULT);
630 	}
631 
632 	kmem_free(buf, bufsize);
633 	return (0);
634 }
635 
636 /*
637  * Given a syndrome and address return information about the associated memory.
638  */
639 static int
640 mmioctl_get_mem_info(intptr_t data)
641 {
642 	mem_info_t mem_info;
643 	int err;
644 
645 	if (copyin((void *)data, &mem_info, sizeof (mem_info_t)))
646 		return (EFAULT);
647 
648 	if ((err = cpu_get_mem_info(mem_info.m_synd, mem_info.m_addr,
649 	    &mem_info.m_mem_size, &mem_info.m_seg_size, &mem_info.m_bank_size,
650 	    &mem_info.m_segments, &mem_info.m_banks, &mem_info.m_mcid)) != 0)
651 		return (err);
652 
653 	if (copyout(&mem_info, (void *)data, sizeof (mem_info_t)) != 0)
654 		return (EFAULT);
655 
656 	return (0);
657 }
658 
659 /*
660  * Given a memory name, return its associated serial id
661  */
662 static int
663 mmioctl_get_mem_sid(intptr_t data)
664 {
665 	mem_name_t mem_name;
666 	void *buf;
667 	void *name;
668 	size_t	name_len;
669 	size_t bufsize;
670 	int len, err;
671 
672 	if ((bufsize = cpu_get_name_bufsize()) == 0)
673 		return (ENOTSUP);
674 
675 	if ((err = mm_read_mem_name(data, &mem_name)) < 0)
676 		return (err);
677 
678 	buf = kmem_alloc(bufsize, KM_SLEEP);
679 
680 	if (mem_name.m_namelen > 1024)
681 		mem_name.m_namelen = 1024; /* cap at 1024 bytes */
682 
683 	name = kmem_alloc(mem_name.m_namelen, KM_SLEEP);
684 
685 	if ((err = copyinstr((char *)mem_name.m_name, (char *)name,
686 	    mem_name.m_namelen, &name_len)) != 0) {
687 		kmem_free(buf, bufsize);
688 		kmem_free(name, mem_name.m_namelen);
689 		return (err);
690 	}
691 
692 	/*
693 	 * Call into cpu specific code to do the lookup.
694 	 */
695 	if ((err = cpu_get_mem_sid(name, buf, bufsize, &len)) != 0) {
696 		kmem_free(buf, bufsize);
697 		kmem_free(name, mem_name.m_namelen);
698 		return (err);
699 	}
700 
701 	if (len > mem_name.m_sidlen) {
702 		kmem_free(buf, bufsize);
703 		kmem_free(name, mem_name.m_namelen);
704 		return (ENAMETOOLONG);
705 	}
706 
707 	if (copyoutstr(buf, (char *)mem_name.m_sid,
708 	    mem_name.m_sidlen, NULL) != 0) {
709 		kmem_free(buf, bufsize);
710 		kmem_free(name, mem_name.m_namelen);
711 		return (EFAULT);
712 	}
713 
714 	kmem_free(buf, bufsize);
715 	kmem_free(name, mem_name.m_namelen);
716 	return (0);
717 }
718 #endif	/* __sparc */
719 
720 /*
721  * Private ioctls for
722  *	libkvm to support kvm_physaddr().
723  *	FMA support for page_retire() and memory attribute information.
724  */
725 /*ARGSUSED*/
726 static int
727 mmioctl(dev_t dev, int cmd, intptr_t data, int flag, cred_t *cred, int *rvalp)
728 {
729 	if ((cmd == MEM_VTOP && getminor(dev) != M_KMEM) ||
730 	    (cmd != MEM_VTOP && getminor(dev) != M_MEM))
731 		return (ENXIO);
732 
733 	switch (cmd) {
734 	case MEM_VTOP:
735 		return (mmioctl_vtop(data));
736 
737 	case MEM_PAGE_RETIRE:
738 	case MEM_PAGE_ISRETIRED:
739 	case MEM_PAGE_UNRETIRE:
740 	case MEM_PAGE_RETIRE_MCE:
741 	case MEM_PAGE_RETIRE_UE:
742 	case MEM_PAGE_GETERRORS:
743 	case MEM_PAGE_RETIRE_TEST:
744 		return (mmioctl_page_retire(cmd, data));
745 
746 	case MEM_PAGE_FMRI_RETIRE:
747 	case MEM_PAGE_FMRI_ISRETIRED:
748 		return (mmioctl_page_fmri_retire(cmd, data));
749 
750 #ifdef __sparc
751 	case MEM_NAME:
752 		return (mmioctl_get_mem_name(data));
753 
754 	case MEM_INFO:
755 		return (mmioctl_get_mem_info(data));
756 
757 	case MEM_SID:
758 		return (mmioctl_get_mem_sid(data));
759 #else
760 	case MEM_NAME:
761 	case MEM_INFO:
762 	case MEM_SID:
763 		return (ENOTSUP);
764 #endif	/* __sparc */
765 	}
766 	return (ENXIO);
767 }
768 
769 /*ARGSUSED2*/
770 static int
771 mmmmap(dev_t dev, off_t off, int prot)
772 {
773 	pfn_t pf;
774 	struct memlist *pmem;
775 	minor_t minor = getminor(dev);
776 
777 	switch (minor) {
778 	case M_MEM:
779 		pf = btop(off);
780 		memlist_read_lock();
781 		for (pmem = phys_install; pmem != NULL; pmem = pmem->next) {
782 			if (pf >= BTOP(pmem->address) &&
783 			    pf < BTOP(pmem->address + pmem->size)) {
784 				memlist_read_unlock();
785 				return (impl_obmem_pfnum(pf));
786 			}
787 		}
788 		memlist_read_unlock();
789 		break;
790 
791 	case M_KMEM:
792 	case M_ALLKMEM:
793 		/* no longer supported with KPR */
794 		return (-1);
795 
796 	case M_ZERO:
797 		/*
798 		 * We shouldn't be mmap'ing to /dev/zero here as
799 		 * mmsegmap() should have already converted
800 		 * a mapping request for this device to a mapping
801 		 * using seg_vn for anonymous memory.
802 		 */
803 		break;
804 
805 	}
806 	return (-1);
807 }
808 
809 /*
810  * This function is called when a memory device is mmap'ed.
811  * Set up the mapping to the correct device driver.
812  */
813 static int
814 mmsegmap(dev_t dev, off_t off, struct as *as, caddr_t *addrp, off_t len,
815     uint_t prot, uint_t maxprot, uint_t flags, struct cred *cred)
816 {
817 	struct segvn_crargs vn_a;
818 	struct segdev_crargs dev_a;
819 	int error;
820 	minor_t minor;
821 	off_t i;
822 
823 	minor = getminor(dev);
824 
825 	as_rangelock(as);
826 	if ((flags & MAP_FIXED) == 0) {
827 		/*
828 		 * No need to worry about vac alignment on /dev/zero
829 		 * since this is a "clone" object that doesn't yet exist.
830 		 */
831 		map_addr(addrp, len, (offset_t)off,
832 				(minor == M_MEM) || (minor == M_KMEM), flags);
833 
834 		if (*addrp == NULL) {
835 			as_rangeunlock(as);
836 			return (ENOMEM);
837 		}
838 	} else {
839 		/*
840 		 * User specified address -
841 		 * Blow away any previous mappings.
842 		 */
843 		(void) as_unmap(as, *addrp, len);
844 	}
845 
846 	switch (minor) {
847 	case M_MEM:
848 		/* /dev/mem cannot be mmap'ed with MAP_PRIVATE */
849 		if ((flags & MAP_TYPE) != MAP_SHARED) {
850 			as_rangeunlock(as);
851 			return (EINVAL);
852 		}
853 
854 		/*
855 		 * Check to ensure that the entire range is
856 		 * legal and we are not trying to map in
857 		 * more than the device will let us.
858 		 */
859 		for (i = 0; i < len; i += PAGESIZE) {
860 			if (mmmmap(dev, off + i, maxprot) == -1) {
861 				as_rangeunlock(as);
862 				return (ENXIO);
863 			}
864 		}
865 
866 		/*
867 		 * Use seg_dev segment driver for /dev/mem mapping.
868 		 */
869 		dev_a.mapfunc = mmmmap;
870 		dev_a.dev = dev;
871 		dev_a.offset = off;
872 		dev_a.type = (flags & MAP_TYPE);
873 		dev_a.prot = (uchar_t)prot;
874 		dev_a.maxprot = (uchar_t)maxprot;
875 		dev_a.hat_attr = 0;
876 
877 		/*
878 		 * Make /dev/mem mappings non-consistent since we can't
879 		 * alias pages that don't have page structs behind them,
880 		 * such as kernel stack pages. If someone mmap()s a kernel
881 		 * stack page and if we give him a tte with cv, a line from
882 		 * that page can get into both pages of the spitfire d$.
883 		 * But snoop from another processor will only invalidate
884 		 * the first page. This later caused kernel (xc_attention)
885 		 * to go into an infinite loop at pil 13 and no interrupts
886 		 * could come in. See 1203630.
887 		 *
888 		 */
889 		dev_a.hat_flags = HAT_LOAD_NOCONSIST;
890 		dev_a.devmap_data = NULL;
891 
892 		error = as_map(as, *addrp, len, segdev_create, &dev_a);
893 		break;
894 
895 	case M_ZERO:
896 		/*
897 		 * Use seg_vn segment driver for /dev/zero mapping.
898 		 * Passing in a NULL amp gives us the "cloning" effect.
899 		 */
900 		vn_a.vp = NULL;
901 		vn_a.offset = 0;
902 		vn_a.type = (flags & MAP_TYPE);
903 		vn_a.prot = prot;
904 		vn_a.maxprot = maxprot;
905 		vn_a.flags = flags & ~MAP_TYPE;
906 		vn_a.cred = cred;
907 		vn_a.amp = NULL;
908 		vn_a.szc = 0;
909 		vn_a.lgrp_mem_policy_flags = 0;
910 		error = as_map(as, *addrp, len, segvn_create, &vn_a);
911 		break;
912 
913 	case M_KMEM:
914 	case M_ALLKMEM:
915 		/* No longer supported with KPR. */
916 		error = ENXIO;
917 		break;
918 
919 	case M_NULL:
920 		/*
921 		 * Use seg_dev segment driver for /dev/null mapping.
922 		 */
923 		dev_a.mapfunc = mmmmap;
924 		dev_a.dev = dev;
925 		dev_a.offset = off;
926 		dev_a.type = 0;		/* neither PRIVATE nor SHARED */
927 		dev_a.prot = dev_a.maxprot = (uchar_t)PROT_NONE;
928 		dev_a.hat_attr = 0;
929 		dev_a.hat_flags = 0;
930 		error = as_map(as, *addrp, len, segdev_create, &dev_a);
931 		break;
932 
933 	default:
934 		error = ENXIO;
935 	}
936 
937 	as_rangeunlock(as);
938 	return (error);
939 }
940 
941 static struct cb_ops mm_cb_ops = {
942 	mmopen,			/* open */
943 	nulldev,		/* close */
944 	nodev,			/* strategy */
945 	nodev,			/* print */
946 	nodev,			/* dump */
947 	mmread,			/* read */
948 	mmwrite,		/* write */
949 	mmioctl,		/* ioctl */
950 	nodev,			/* devmap */
951 	mmmmap,			/* mmap */
952 	mmsegmap,		/* segmap */
953 	mmchpoll,		/* poll */
954 	mmpropop,		/* prop_op */
955 	0,			/* streamtab  */
956 	D_NEW | D_MP | D_64BIT | D_U64BIT
957 };
958 
959 static struct dev_ops mm_ops = {
960 	DEVO_REV,		/* devo_rev, */
961 	0,			/* refcnt  */
962 	mm_info,		/* get_dev_info */
963 	nulldev,		/* identify */
964 	nulldev,		/* probe */
965 	mm_attach,		/* attach */
966 	nodev,			/* detach */
967 	nodev,			/* reset */
968 	&mm_cb_ops,		/* driver operations */
969 	(struct bus_ops *)0	/* bus operations */
970 };
971 
972 static struct modldrv modldrv = {
973 	&mod_driverops, "memory driver %I%", &mm_ops,
974 };
975 
976 static struct modlinkage modlinkage = {
977 	MODREV_1, &modldrv, NULL
978 };
979 
980 int
981 _init(void)
982 {
983 	return (mod_install(&modlinkage));
984 }
985 
986 int
987 _info(struct modinfo *modinfop)
988 {
989 	return (mod_info(&modlinkage, modinfop));
990 }
991 
992 int
993 _fini(void)
994 {
995 	return (mod_remove(&modlinkage));
996 }
997 
998 static int
999 mm_kstat_update(kstat_t *ksp, int rw)
1000 {
1001 	struct memlist *pmem;
1002 	uint_t count;
1003 
1004 	if (rw == KSTAT_WRITE)
1005 		return (EACCES);
1006 
1007 	count = 0;
1008 	memlist_read_lock();
1009 	for (pmem = phys_install; pmem != NULL; pmem = pmem->next) {
1010 		count++;
1011 	}
1012 	memlist_read_unlock();
1013 
1014 	ksp->ks_ndata = count;
1015 	ksp->ks_data_size = count * 2 * sizeof (uint64_t);
1016 
1017 	return (0);
1018 }
1019 
1020 static int
1021 mm_kstat_snapshot(kstat_t *ksp, void *buf, int rw)
1022 {
1023 	struct memlist *pmem;
1024 	struct memunit {
1025 		uint64_t address;
1026 		uint64_t size;
1027 	} *kspmem;
1028 
1029 	if (rw == KSTAT_WRITE)
1030 		return (EACCES);
1031 
1032 	ksp->ks_snaptime = gethrtime();
1033 
1034 	kspmem = (struct memunit *)buf;
1035 	memlist_read_lock();
1036 	for (pmem = phys_install; pmem != NULL; pmem = pmem->next, kspmem++) {
1037 		if ((caddr_t)kspmem >= (caddr_t)buf + ksp->ks_data_size)
1038 			break;
1039 		kspmem->address = pmem->address;
1040 		kspmem->size = pmem->size;
1041 	}
1042 	memlist_read_unlock();
1043 
1044 	return (0);
1045 }
1046 
1047 /*
1048  * Read a mem_name_t from user-space and store it in the mem_name_t
1049  * pointed to by the mem_name argument.
1050  */
1051 static int
1052 mm_read_mem_name(intptr_t data, mem_name_t *mem_name)
1053 {
1054 	if (get_udatamodel() == DATAMODEL_NATIVE) {
1055 		if (copyin((void *)data, mem_name, sizeof (mem_name_t)))
1056 			return (EFAULT);
1057 	}
1058 #ifdef	_SYSCALL32
1059 	else {
1060 		mem_name32_t mem_name32;
1061 
1062 		if (copyin((void *)data, &mem_name32, sizeof (mem_name32_t)))
1063 			return (EFAULT);
1064 		mem_name->m_addr = mem_name32.m_addr;
1065 		mem_name->m_synd = mem_name32.m_synd;
1066 		mem_name->m_type[0] = mem_name32.m_type[0];
1067 		mem_name->m_type[1] = mem_name32.m_type[1];
1068 		mem_name->m_name = (caddr_t)(uintptr_t)mem_name32.m_name;
1069 		mem_name->m_namelen = (size_t)mem_name32.m_namelen;
1070 		mem_name->m_sid = (caddr_t)(uintptr_t)mem_name32.m_sid;
1071 		mem_name->m_sidlen = (size_t)mem_name32.m_sidlen;
1072 	}
1073 #endif	/* _SYSCALL32 */
1074 
1075 	return (0);
1076 }
1077 
1078 /*
1079  * Read a mem_page_t from user-space and store it in the mem_page_t
1080  * pointed to by the mpage argument.
1081  */
1082 static int
1083 mm_read_mem_page(intptr_t data, mem_page_t *mpage)
1084 {
1085 	if (get_udatamodel() == DATAMODEL_NATIVE) {
1086 		if (copyin((void *)data, mpage, sizeof (mem_page_t)) != 0)
1087 			return (EFAULT);
1088 	}
1089 #ifdef _SYSCALL32
1090 	else {
1091 		mem_page32_t	mpage32;
1092 
1093 		if (copyin((void *)data, &mpage32, sizeof (mem_page32_t)) != 0)
1094 			return (EFAULT);
1095 
1096 		mpage->m_fmri = (caddr_t)(uintptr_t)mpage32.m_fmri;
1097 		mpage->m_fmrisz = mpage32.m_fmrisz;
1098 	}
1099 #endif	/* _SYSCALL32 */
1100 
1101 	return (0);
1102 }
1103 
1104 /*
1105  * Expand an FMRI from a mem_page_t.
1106  */
1107 static int
1108 mm_get_mem_fmri(mem_page_t *mpage, nvlist_t **nvl)
1109 {
1110 	char *buf;
1111 	int err;
1112 
1113 	if (mpage->m_fmri == NULL || mpage->m_fmrisz > MEM_FMRI_MAX_BUFSIZE)
1114 		return (EINVAL);
1115 
1116 	buf = kmem_alloc(mpage->m_fmrisz, KM_SLEEP);
1117 	if (copyin(mpage->m_fmri, buf, mpage->m_fmrisz) != 0) {
1118 		kmem_free(buf, mpage->m_fmrisz);
1119 		return (EFAULT);
1120 	}
1121 
1122 	err = nvlist_unpack(buf, mpage->m_fmrisz, nvl, KM_SLEEP);
1123 	kmem_free(buf, mpage->m_fmrisz);
1124 
1125 	return (err);
1126 }
1127 
1128 static int
1129 mm_get_paddr(nvlist_t *nvl, uint64_t *paddr)
1130 {
1131 	uint8_t version;
1132 	uint64_t pa;
1133 	char *scheme;
1134 #ifdef __sparc
1135 	uint64_t offset;
1136 	char *unum;
1137 	char **serids;
1138 	uint_t nserids;
1139 	int err;
1140 #endif
1141 
1142 	/* Verify FMRI scheme name and version number */
1143 	if ((nvlist_lookup_string(nvl, FM_FMRI_SCHEME, &scheme) != 0) ||
1144 	    (strcmp(scheme, FM_FMRI_SCHEME_MEM) != 0) ||
1145 	    (nvlist_lookup_uint8(nvl, FM_VERSION, &version) != 0) ||
1146 	    version > FM_MEM_SCHEME_VERSION) {
1147 		return (EINVAL);
1148 	}
1149 
1150 	/*
1151 	 * There are two ways a physical address can be  obtained from a mem
1152 	 * scheme FMRI.  One way is to use the "offset" and  "serial"
1153 	 * members, if they are present, together with the "unum" member to
1154 	 * calculate a physical address.  This is the preferred way since
1155 	 * it is independent of possible changes to the programming of
1156 	 * underlying hardware registers that may change the physical address.
1157 	 * If the "offset" member is not present, then the address is
1158 	 * retrieved from the "physaddr" member.
1159 	 */
1160 #if defined(__sparc)
1161 	if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_OFFSET, &offset) != 0) {
1162 		if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_PHYSADDR, &pa) !=
1163 		    0) {
1164 			return (EINVAL);
1165 		}
1166 	} else if (nvlist_lookup_string(nvl, FM_FMRI_MEM_UNUM, &unum) != 0 ||
1167 	    nvlist_lookup_string_array(nvl, FM_FMRI_MEM_SERIAL_ID, &serids,
1168 	    &nserids) != 0) {
1169 		return (EINVAL);
1170 	} else {
1171 		err = cpu_get_mem_addr(unum, serids[0], offset, &pa);
1172 		if (err != 0) {
1173 			if (err == ENOTSUP) {
1174 				/* Fall back to physaddr */
1175 				if (nvlist_lookup_uint64(nvl,
1176 				    FM_FMRI_MEM_PHYSADDR, &pa) != 0)
1177 					return (EINVAL);
1178 			} else
1179 				return (err);
1180 		}
1181 	}
1182 #elif defined(__i386) || defined(__amd64)
1183 	if (cmi_mc_unumtopa(NULL, nvl, &pa) == 0)
1184 		return (EINVAL);
1185 #else
1186 #error "port me"
1187 #endif /* __sparc */
1188 
1189 	*paddr = pa;
1190 	return (0);
1191 }
1192