xref: /titanic_41/usr/src/uts/common/io/mem.c (revision 9ec394dbf343c1f23c6e13c39df427f238e5a369)
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 #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(__x86)
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 	/* must be character device */
193 	if (typ != OTYP_CHR)
194 		return (EINVAL);
195 	return (0);
196 }
197 
198 struct pollhead	mm_pollhd;
199 
200 /*ARGSUSED*/
201 static int
202 mmchpoll(dev_t dev, short events, int anyyet, short *reventsp,
203     struct pollhead **phpp)
204 {
205 	switch (getminor(dev)) {
206 	case M_NULL:
207 	case M_ZERO:
208 	case M_MEM:
209 	case M_KMEM:
210 	case M_ALLKMEM:
211 		*reventsp = events & (POLLIN | POLLOUT | POLLPRI | POLLRDNORM |
212 		    POLLWRNORM | POLLRDBAND | POLLWRBAND);
213 		/*
214 		 * A non NULL pollhead pointer should be returned in case
215 		 * user polls for 0 events.
216 		 */
217 		*phpp = !anyyet && !*reventsp ?
218 		    &mm_pollhd : (struct pollhead *)NULL;
219 		return (0);
220 	default:
221 		/* no other devices currently support polling */
222 		return (ENXIO);
223 	}
224 }
225 
226 static int
227 mmpropop(dev_t dev, dev_info_t *dip, ddi_prop_op_t prop_op, int flags,
228     char *name, caddr_t valuep, int *lengthp)
229 {
230 	/*
231 	 * implement zero size to reduce overhead (avoid two failing
232 	 * property lookups per stat).
233 	 */
234 	return (ddi_prop_op_size(dev, dip, prop_op,
235 	    flags, name, valuep, lengthp, 0));
236 }
237 
238 static int
239 mmio(struct uio *uio, enum uio_rw rw, pfn_t pfn, off_t pageoff, int allowio)
240 {
241 	int error = 0;
242 	size_t nbytes = MIN((size_t)(PAGESIZE - pageoff),
243 	    (size_t)uio->uio_iov->iov_len);
244 
245 	mutex_enter(&mm_lock);
246 	hat_devload(kas.a_hat, mm_map, PAGESIZE, pfn,
247 	    (uint_t)(rw == UIO_READ ? PROT_READ : PROT_READ | PROT_WRITE),
248 	    HAT_LOAD_NOCONSIST | HAT_LOAD_LOCK);
249 
250 	if (!pf_is_memory(pfn)) {
251 		if (allowio) {
252 			size_t c = uio->uio_iov->iov_len;
253 
254 			if (ddi_peekpokeio(NULL, uio, rw,
255 			    (caddr_t)(uintptr_t)uio->uio_loffset, c,
256 			    sizeof (int32_t)) != DDI_SUCCESS)
257 				error = EFAULT;
258 		} else
259 			error = EIO;
260 	} else
261 		error = uiomove(&mm_map[pageoff], nbytes, rw, uio);
262 
263 	hat_unload(kas.a_hat, mm_map, PAGESIZE, HAT_UNLOAD_UNLOCK);
264 	mutex_exit(&mm_lock);
265 	return (error);
266 }
267 
268 static int
269 mmpagelock(struct as *as, caddr_t va)
270 {
271 	struct seg *seg;
272 	int i;
273 
274 	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
275 	seg = as_segat(as, va);
276 	i = (seg != NULL)? SEGOP_CAPABLE(seg, S_CAPABILITY_NOMINFLT) : 0;
277 	AS_LOCK_EXIT(as, &as->a_lock);
278 
279 	return (i);
280 }
281 
282 #ifdef	__sparc
283 
284 #define	NEED_LOCK_KVADDR(kva)	mmpagelock(&kas, kva)
285 
286 #else	/* __i386, __amd64 */
287 
288 #define	NEED_LOCK_KVADDR(va)	0
289 
290 #endif	/* __sparc */
291 
292 /*ARGSUSED3*/
293 static int
294 mmrw(dev_t dev, struct uio *uio, enum uio_rw rw, cred_t *cred)
295 {
296 	pfn_t v;
297 	struct iovec *iov;
298 	int error = 0;
299 	size_t c;
300 	ssize_t oresid = uio->uio_resid;
301 	minor_t minor = getminor(dev);
302 
303 	while (uio->uio_resid > 0 && error == 0) {
304 		iov = uio->uio_iov;
305 		if (iov->iov_len == 0) {
306 			uio->uio_iov++;
307 			uio->uio_iovcnt--;
308 			if (uio->uio_iovcnt < 0)
309 				panic("mmrw");
310 			continue;
311 		}
312 		switch (minor) {
313 
314 		case M_MEM:
315 			memlist_read_lock();
316 			if (!address_in_memlist(phys_install,
317 			    (uint64_t)uio->uio_loffset, 1)) {
318 				memlist_read_unlock();
319 				error = EFAULT;
320 				break;
321 			}
322 			memlist_read_unlock();
323 
324 			v = BTOP((u_offset_t)uio->uio_loffset);
325 			error = mmio(uio, rw, v,
326 			    uio->uio_loffset & PAGEOFFSET, 0);
327 			break;
328 
329 		case M_KMEM:
330 		case M_ALLKMEM:
331 			{
332 			page_t **ppp;
333 			caddr_t vaddr = (caddr_t)uio->uio_offset;
334 			int try_lock = NEED_LOCK_KVADDR(vaddr);
335 			int locked = 0;
336 
337 			if ((error = plat_mem_do_mmio(uio, rw)) != ENOTSUP)
338 				break;
339 
340 			/*
341 			 * If vaddr does not map a valid page, as_pagelock()
342 			 * will return failure. Hence we can't check the
343 			 * return value and return EFAULT here as we'd like.
344 			 * seg_kp and seg_kpm do not properly support
345 			 * as_pagelock() for this context so we avoid it
346 			 * using the try_lock set check above.  Some day when
347 			 * the kernel page locking gets redesigned all this
348 			 * muck can be cleaned up.
349 			 */
350 			if (try_lock)
351 				locked = (as_pagelock(&kas, &ppp, vaddr,
352 				    PAGESIZE, S_WRITE) == 0);
353 
354 			v = hat_getpfnum(kas.a_hat,
355 			    (caddr_t)(uintptr_t)uio->uio_loffset);
356 			if (v == PFN_INVALID) {
357 				if (locked)
358 					as_pageunlock(&kas, ppp, vaddr,
359 					    PAGESIZE, S_WRITE);
360 				error = EFAULT;
361 				break;
362 			}
363 
364 			error = mmio(uio, rw, v, uio->uio_loffset & PAGEOFFSET,
365 			    minor == M_ALLKMEM || mm_kmem_io_access);
366 			if (locked)
367 				as_pageunlock(&kas, ppp, vaddr, PAGESIZE,
368 				    S_WRITE);
369 			}
370 
371 			break;
372 
373 		case M_ZERO:
374 			if (rw == UIO_READ) {
375 				label_t ljb;
376 
377 				if (on_fault(&ljb)) {
378 					no_fault();
379 					error = EFAULT;
380 					break;
381 				}
382 				uzero(iov->iov_base, iov->iov_len);
383 				no_fault();
384 				uio->uio_resid -= iov->iov_len;
385 				uio->uio_loffset += iov->iov_len;
386 				break;
387 			}
388 			/* else it's a write, fall through to NULL case */
389 			/*FALLTHROUGH*/
390 
391 		case M_NULL:
392 			if (rw == UIO_READ)
393 				return (0);
394 			c = iov->iov_len;
395 			iov->iov_base += c;
396 			iov->iov_len -= c;
397 			uio->uio_loffset += c;
398 			uio->uio_resid -= c;
399 			break;
400 
401 		}
402 	}
403 	return (uio->uio_resid == oresid ? error : 0);
404 }
405 
406 static int
407 mmread(dev_t dev, struct uio *uio, cred_t *cred)
408 {
409 	return (mmrw(dev, uio, UIO_READ, cred));
410 }
411 
412 static int
413 mmwrite(dev_t dev, struct uio *uio, cred_t *cred)
414 {
415 	return (mmrw(dev, uio, UIO_WRITE, cred));
416 }
417 
418 /*
419  * Private ioctl for libkvm to support kvm_physaddr().
420  * Given an address space and a VA, compute the PA.
421  */
422 static int
423 mmioctl_vtop(intptr_t data)
424 {
425 #ifdef _SYSCALL32
426 	mem_vtop32_t vtop32;
427 #endif
428 	mem_vtop_t mem_vtop;
429 	proc_t *p;
430 	pfn_t pfn = (pfn_t)PFN_INVALID;
431 	pid_t pid = 0;
432 	struct as *as;
433 	struct seg *seg;
434 
435 	if (get_udatamodel() == DATAMODEL_NATIVE) {
436 		if (copyin((void *)data, &mem_vtop, sizeof (mem_vtop_t)))
437 			return (EFAULT);
438 	}
439 #ifdef _SYSCALL32
440 	else {
441 		if (copyin((void *)data, &vtop32, sizeof (mem_vtop32_t)))
442 			return (EFAULT);
443 		mem_vtop.m_as = (struct as *)(uintptr_t)vtop32.m_as;
444 		mem_vtop.m_va = (void *)(uintptr_t)vtop32.m_va;
445 
446 		if (mem_vtop.m_as != NULL)
447 			return (EINVAL);
448 	}
449 #endif
450 
451 	if (mem_vtop.m_as == &kas) {
452 		pfn = hat_getpfnum(kas.a_hat, mem_vtop.m_va);
453 	} else {
454 		if (mem_vtop.m_as == NULL) {
455 			/*
456 			 * Assume the calling process's address space if the
457 			 * caller didn't specify one.
458 			 */
459 			p = curthread->t_procp;
460 			if (p == NULL)
461 				return (EIO);
462 			mem_vtop.m_as = p->p_as;
463 		}
464 
465 		mutex_enter(&pidlock);
466 		for (p = practive; p != NULL; p = p->p_next) {
467 			if (p->p_as == mem_vtop.m_as) {
468 				pid = p->p_pid;
469 				break;
470 			}
471 		}
472 		mutex_exit(&pidlock);
473 		if (p == NULL)
474 			return (EIO);
475 		p = sprlock(pid);
476 		if (p == NULL)
477 			return (EIO);
478 		as = p->p_as;
479 		if (as == mem_vtop.m_as) {
480 			mutex_exit(&p->p_lock);
481 			AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
482 			for (seg = AS_SEGFIRST(as); seg != NULL;
483 			    seg = AS_SEGNEXT(as, seg))
484 				if ((uintptr_t)mem_vtop.m_va -
485 				    (uintptr_t)seg->s_base < seg->s_size)
486 					break;
487 			if (seg != NULL)
488 				pfn = hat_getpfnum(as->a_hat, mem_vtop.m_va);
489 			AS_LOCK_EXIT(as, &as->a_lock);
490 			mutex_enter(&p->p_lock);
491 		}
492 		sprunlock(p);
493 	}
494 	mem_vtop.m_pfn = pfn;
495 	if (pfn == PFN_INVALID)
496 		return (EIO);
497 
498 	if (get_udatamodel() == DATAMODEL_NATIVE) {
499 		if (copyout(&mem_vtop, (void *)data, sizeof (mem_vtop_t)))
500 			return (EFAULT);
501 	}
502 #ifdef _SYSCALL32
503 	else {
504 		vtop32.m_pfn = mem_vtop.m_pfn;
505 		if (copyout(&vtop32, (void *)data, sizeof (mem_vtop32_t)))
506 			return (EFAULT);
507 	}
508 #endif
509 
510 	return (0);
511 }
512 
513 /*
514  * Given a PA, execute the given page retire command on it.
515  */
516 static int
517 mmioctl_page_retire(int cmd, intptr_t data)
518 {
519 	extern int page_retire_test(void);
520 	uint64_t pa;
521 
522 	if (copyin((void *)data, &pa, sizeof (uint64_t))) {
523 		return (EFAULT);
524 	}
525 
526 	switch (cmd) {
527 	case MEM_PAGE_ISRETIRED:
528 		return (page_retire_check(pa, NULL));
529 
530 	case MEM_PAGE_UNRETIRE:
531 		return (page_unretire(pa));
532 
533 	case MEM_PAGE_RETIRE:
534 		return (page_retire(pa, PR_FMA));
535 
536 	case MEM_PAGE_RETIRE_MCE:
537 		return (page_retire(pa, PR_MCE));
538 
539 	case MEM_PAGE_RETIRE_UE:
540 		return (page_retire(pa, PR_UE));
541 
542 	case MEM_PAGE_GETERRORS:
543 		{
544 			uint64_t page_errors;
545 			int rc = page_retire_check(pa, &page_errors);
546 			if (copyout(&page_errors, (void *)data,
547 			    sizeof (uint64_t))) {
548 				return (EFAULT);
549 			}
550 			return (rc);
551 		}
552 
553 	case MEM_PAGE_RETIRE_TEST:
554 		return (page_retire_test());
555 
556 	}
557 
558 	return (EINVAL);
559 }
560 
561 /*
562  * Given a mem-scheme FMRI for a page, execute the given page retire
563  * command on it.
564  */
565 static int
566 mmioctl_page_fmri_retire(int cmd, intptr_t data)
567 {
568 	mem_page_t mpage;
569 	uint64_t pa;
570 	nvlist_t *nvl;
571 	int err;
572 
573 	if ((err = mm_read_mem_page(data, &mpage)) < 0)
574 		return (err);
575 
576 	if ((err = mm_get_mem_fmri(&mpage, &nvl)) != 0)
577 		return (err);
578 
579 	if ((err = mm_get_paddr(nvl, &pa)) != 0) {
580 		nvlist_free(nvl);
581 		return (err);
582 	}
583 
584 	nvlist_free(nvl);
585 
586 	switch (cmd) {
587 	case MEM_PAGE_FMRI_ISRETIRED:
588 		return (page_retire_check(pa, NULL));
589 
590 	case MEM_PAGE_FMRI_RETIRE:
591 		return (page_retire(pa, PR_FMA));
592 
593 	case MEM_PAGE_FMRI_UNRETIRE:
594 		return (page_unretire(pa));
595 	}
596 
597 	return (EINVAL);
598 }
599 
600 #ifdef __sparc
601 /*
602  * Given a syndrome, syndrome type, and address return the
603  * associated memory name in the provided data buffer.
604  */
605 static int
606 mmioctl_get_mem_name(intptr_t data)
607 {
608 	mem_name_t mem_name;
609 	void *buf;
610 	size_t bufsize;
611 	int len, err;
612 
613 	if ((bufsize = cpu_get_name_bufsize()) == 0)
614 		return (ENOTSUP);
615 
616 	if ((err = mm_read_mem_name(data, &mem_name)) < 0)
617 		return (err);
618 
619 	buf = kmem_alloc(bufsize, KM_SLEEP);
620 
621 	/*
622 	 * Call into cpu specific code to do the lookup.
623 	 */
624 	if ((err = cpu_get_mem_name(mem_name.m_synd, mem_name.m_type,
625 	    mem_name.m_addr, buf, bufsize, &len)) != 0) {
626 		kmem_free(buf, bufsize);
627 		return (err);
628 	}
629 
630 	if (len >= mem_name.m_namelen) {
631 		kmem_free(buf, bufsize);
632 		return (ENOSPC);
633 	}
634 
635 	if (copyoutstr(buf, (char *)mem_name.m_name,
636 	    mem_name.m_namelen, NULL) != 0) {
637 		kmem_free(buf, bufsize);
638 		return (EFAULT);
639 	}
640 
641 	kmem_free(buf, bufsize);
642 	return (0);
643 }
644 
645 /*
646  * Given a syndrome and address return information about the associated memory.
647  */
648 static int
649 mmioctl_get_mem_info(intptr_t data)
650 {
651 	mem_info_t mem_info;
652 	int err;
653 
654 	if (copyin((void *)data, &mem_info, sizeof (mem_info_t)))
655 		return (EFAULT);
656 
657 	if ((err = cpu_get_mem_info(mem_info.m_synd, mem_info.m_addr,
658 	    &mem_info.m_mem_size, &mem_info.m_seg_size, &mem_info.m_bank_size,
659 	    &mem_info.m_segments, &mem_info.m_banks, &mem_info.m_mcid)) != 0)
660 		return (err);
661 
662 	if (copyout(&mem_info, (void *)data, sizeof (mem_info_t)) != 0)
663 		return (EFAULT);
664 
665 	return (0);
666 }
667 
668 /*
669  * Given a memory name, return its associated serial id
670  */
671 static int
672 mmioctl_get_mem_sid(intptr_t data)
673 {
674 	mem_name_t mem_name;
675 	void *buf;
676 	void *name;
677 	size_t	name_len;
678 	size_t bufsize;
679 	int len, err;
680 
681 	if ((bufsize = cpu_get_name_bufsize()) == 0)
682 		return (ENOTSUP);
683 
684 	if ((err = mm_read_mem_name(data, &mem_name)) < 0)
685 		return (err);
686 
687 	buf = kmem_alloc(bufsize, KM_SLEEP);
688 
689 	if (mem_name.m_namelen > 1024)
690 		mem_name.m_namelen = 1024; /* cap at 1024 bytes */
691 
692 	name = kmem_alloc(mem_name.m_namelen, KM_SLEEP);
693 
694 	if ((err = copyinstr((char *)mem_name.m_name, (char *)name,
695 	    mem_name.m_namelen, &name_len)) != 0) {
696 		kmem_free(buf, bufsize);
697 		kmem_free(name, mem_name.m_namelen);
698 		return (err);
699 	}
700 
701 	/*
702 	 * Call into cpu specific code to do the lookup.
703 	 */
704 	if ((err = cpu_get_mem_sid(name, buf, bufsize, &len)) != 0) {
705 		kmem_free(buf, bufsize);
706 		kmem_free(name, mem_name.m_namelen);
707 		return (err);
708 	}
709 
710 	if (len > mem_name.m_sidlen) {
711 		kmem_free(buf, bufsize);
712 		kmem_free(name, mem_name.m_namelen);
713 		return (ENAMETOOLONG);
714 	}
715 
716 	if (copyoutstr(buf, (char *)mem_name.m_sid,
717 	    mem_name.m_sidlen, NULL) != 0) {
718 		kmem_free(buf, bufsize);
719 		kmem_free(name, mem_name.m_namelen);
720 		return (EFAULT);
721 	}
722 
723 	kmem_free(buf, bufsize);
724 	kmem_free(name, mem_name.m_namelen);
725 	return (0);
726 }
727 #endif	/* __sparc */
728 
729 /*
730  * Private ioctls for
731  *	libkvm to support kvm_physaddr().
732  *	FMA support for page_retire() and memory attribute information.
733  */
734 /*ARGSUSED*/
735 static int
736 mmioctl(dev_t dev, int cmd, intptr_t data, int flag, cred_t *cred, int *rvalp)
737 {
738 	if ((cmd == MEM_VTOP && getminor(dev) != M_KMEM) ||
739 	    (cmd != MEM_VTOP && getminor(dev) != M_MEM))
740 		return (ENXIO);
741 
742 	switch (cmd) {
743 	case MEM_VTOP:
744 		return (mmioctl_vtop(data));
745 
746 	case MEM_PAGE_RETIRE:
747 	case MEM_PAGE_ISRETIRED:
748 	case MEM_PAGE_UNRETIRE:
749 	case MEM_PAGE_RETIRE_MCE:
750 	case MEM_PAGE_RETIRE_UE:
751 	case MEM_PAGE_GETERRORS:
752 	case MEM_PAGE_RETIRE_TEST:
753 		return (mmioctl_page_retire(cmd, data));
754 
755 	case MEM_PAGE_FMRI_RETIRE:
756 	case MEM_PAGE_FMRI_ISRETIRED:
757 	case MEM_PAGE_FMRI_UNRETIRE:
758 		return (mmioctl_page_fmri_retire(cmd, data));
759 
760 #ifdef __sparc
761 	case MEM_NAME:
762 		return (mmioctl_get_mem_name(data));
763 
764 	case MEM_INFO:
765 		return (mmioctl_get_mem_info(data));
766 
767 	case MEM_SID:
768 		return (mmioctl_get_mem_sid(data));
769 #else
770 	case MEM_NAME:
771 	case MEM_INFO:
772 	case MEM_SID:
773 		return (ENOTSUP);
774 #endif	/* __sparc */
775 	}
776 	return (ENXIO);
777 }
778 
779 /*ARGSUSED2*/
780 static int
781 mmmmap(dev_t dev, off_t off, int prot)
782 {
783 	pfn_t pf;
784 	struct memlist *pmem;
785 	minor_t minor = getminor(dev);
786 
787 	switch (minor) {
788 	case M_MEM:
789 		pf = btop(off);
790 		memlist_read_lock();
791 		for (pmem = phys_install; pmem != NULL; pmem = pmem->next) {
792 			if (pf >= BTOP(pmem->address) &&
793 			    pf < BTOP(pmem->address + pmem->size)) {
794 				memlist_read_unlock();
795 				return (impl_obmem_pfnum(pf));
796 			}
797 		}
798 		memlist_read_unlock();
799 		break;
800 
801 	case M_KMEM:
802 	case M_ALLKMEM:
803 		/* no longer supported with KPR */
804 		return (-1);
805 
806 	case M_ZERO:
807 		/*
808 		 * We shouldn't be mmap'ing to /dev/zero here as
809 		 * mmsegmap() should have already converted
810 		 * a mapping request for this device to a mapping
811 		 * using seg_vn for anonymous memory.
812 		 */
813 		break;
814 
815 	}
816 	return (-1);
817 }
818 
819 /*
820  * This function is called when a memory device is mmap'ed.
821  * Set up the mapping to the correct device driver.
822  */
823 static int
824 mmsegmap(dev_t dev, off_t off, struct as *as, caddr_t *addrp, off_t len,
825     uint_t prot, uint_t maxprot, uint_t flags, struct cred *cred)
826 {
827 	struct segvn_crargs vn_a;
828 	struct segdev_crargs dev_a;
829 	int error;
830 	minor_t minor;
831 	off_t i;
832 
833 	minor = getminor(dev);
834 
835 	as_rangelock(as);
836 	/*
837 	 * No need to worry about vac alignment on /dev/zero
838 	 * since this is a "clone" object that doesn't yet exist.
839 	 */
840 	error = choose_addr(as, addrp, len, off,
841 	    (minor == M_MEM) || (minor == M_KMEM), flags);
842 	if (error != 0) {
843 		as_rangeunlock(as);
844 		return (error);
845 	}
846 
847 	switch (minor) {
848 	case M_MEM:
849 		/* /dev/mem cannot be mmap'ed with MAP_PRIVATE */
850 		if ((flags & MAP_TYPE) != MAP_SHARED) {
851 			as_rangeunlock(as);
852 			return (EINVAL);
853 		}
854 
855 		/*
856 		 * Check to ensure that the entire range is
857 		 * legal and we are not trying to map in
858 		 * more than the device will let us.
859 		 */
860 		for (i = 0; i < len; i += PAGESIZE) {
861 			if (mmmmap(dev, off + i, maxprot) == -1) {
862 				as_rangeunlock(as);
863 				return (ENXIO);
864 			}
865 		}
866 
867 		/*
868 		 * Use seg_dev segment driver for /dev/mem mapping.
869 		 */
870 		dev_a.mapfunc = mmmmap;
871 		dev_a.dev = dev;
872 		dev_a.offset = off;
873 		dev_a.type = (flags & MAP_TYPE);
874 		dev_a.prot = (uchar_t)prot;
875 		dev_a.maxprot = (uchar_t)maxprot;
876 		dev_a.hat_attr = 0;
877 
878 		/*
879 		 * Make /dev/mem mappings non-consistent since we can't
880 		 * alias pages that don't have page structs behind them,
881 		 * such as kernel stack pages. If someone mmap()s a kernel
882 		 * stack page and if we give him a tte with cv, a line from
883 		 * that page can get into both pages of the spitfire d$.
884 		 * But snoop from another processor will only invalidate
885 		 * the first page. This later caused kernel (xc_attention)
886 		 * to go into an infinite loop at pil 13 and no interrupts
887 		 * could come in. See 1203630.
888 		 *
889 		 */
890 		dev_a.hat_flags = HAT_LOAD_NOCONSIST;
891 		dev_a.devmap_data = NULL;
892 
893 		error = as_map(as, *addrp, len, segdev_create, &dev_a);
894 		break;
895 
896 	case M_ZERO:
897 		/*
898 		 * Use seg_vn segment driver for /dev/zero mapping.
899 		 * Passing in a NULL amp gives us the "cloning" effect.
900 		 */
901 		vn_a.vp = NULL;
902 		vn_a.offset = 0;
903 		vn_a.type = (flags & MAP_TYPE);
904 		vn_a.prot = prot;
905 		vn_a.maxprot = maxprot;
906 		vn_a.flags = flags & ~MAP_TYPE;
907 		vn_a.cred = cred;
908 		vn_a.amp = NULL;
909 		vn_a.szc = 0;
910 		vn_a.lgrp_mem_policy_flags = 0;
911 		error = as_map(as, *addrp, len, segvn_create, &vn_a);
912 		break;
913 
914 	case M_KMEM:
915 	case M_ALLKMEM:
916 		/* No longer supported with KPR. */
917 		error = ENXIO;
918 		break;
919 
920 	case M_NULL:
921 		/*
922 		 * Use seg_dev segment driver for /dev/null mapping.
923 		 */
924 		dev_a.mapfunc = mmmmap;
925 		dev_a.dev = dev;
926 		dev_a.offset = off;
927 		dev_a.type = 0;		/* neither PRIVATE nor SHARED */
928 		dev_a.prot = dev_a.maxprot = (uchar_t)PROT_NONE;
929 		dev_a.hat_attr = 0;
930 		dev_a.hat_flags = 0;
931 		error = as_map(as, *addrp, len, segdev_create, &dev_a);
932 		break;
933 
934 	default:
935 		error = ENXIO;
936 	}
937 
938 	as_rangeunlock(as);
939 	return (error);
940 }
941 
942 static struct cb_ops mm_cb_ops = {
943 	mmopen,			/* open */
944 	nulldev,		/* close */
945 	nodev,			/* strategy */
946 	nodev,			/* print */
947 	nodev,			/* dump */
948 	mmread,			/* read */
949 	mmwrite,		/* write */
950 	mmioctl,		/* ioctl */
951 	nodev,			/* devmap */
952 	mmmmap,			/* mmap */
953 	mmsegmap,		/* segmap */
954 	mmchpoll,		/* poll */
955 	mmpropop,		/* prop_op */
956 	0,			/* streamtab  */
957 	D_NEW | D_MP | D_64BIT | D_U64BIT
958 };
959 
960 static struct dev_ops mm_ops = {
961 	DEVO_REV,		/* devo_rev, */
962 	0,			/* refcnt  */
963 	mm_info,		/* get_dev_info */
964 	nulldev,		/* identify */
965 	nulldev,		/* probe */
966 	mm_attach,		/* attach */
967 	nodev,			/* detach */
968 	nodev,			/* reset */
969 	&mm_cb_ops,		/* driver operations */
970 	(struct bus_ops *)0	/* bus operations */
971 };
972 
973 static struct modldrv modldrv = {
974 	&mod_driverops, "memory driver %I%", &mm_ops,
975 };
976 
977 static struct modlinkage modlinkage = {
978 	MODREV_1, &modldrv, NULL
979 };
980 
981 int
982 _init(void)
983 {
984 	return (mod_install(&modlinkage));
985 }
986 
987 int
988 _info(struct modinfo *modinfop)
989 {
990 	return (mod_info(&modlinkage, modinfop));
991 }
992 
993 int
994 _fini(void)
995 {
996 	return (mod_remove(&modlinkage));
997 }
998 
999 static int
1000 mm_kstat_update(kstat_t *ksp, int rw)
1001 {
1002 	struct memlist *pmem;
1003 	uint_t count;
1004 
1005 	if (rw == KSTAT_WRITE)
1006 		return (EACCES);
1007 
1008 	count = 0;
1009 	memlist_read_lock();
1010 	for (pmem = phys_install; pmem != NULL; pmem = pmem->next) {
1011 		count++;
1012 	}
1013 	memlist_read_unlock();
1014 
1015 	ksp->ks_ndata = count;
1016 	ksp->ks_data_size = count * 2 * sizeof (uint64_t);
1017 
1018 	return (0);
1019 }
1020 
1021 static int
1022 mm_kstat_snapshot(kstat_t *ksp, void *buf, int rw)
1023 {
1024 	struct memlist *pmem;
1025 	struct memunit {
1026 		uint64_t address;
1027 		uint64_t size;
1028 	} *kspmem;
1029 
1030 	if (rw == KSTAT_WRITE)
1031 		return (EACCES);
1032 
1033 	ksp->ks_snaptime = gethrtime();
1034 
1035 	kspmem = (struct memunit *)buf;
1036 	memlist_read_lock();
1037 	for (pmem = phys_install; pmem != NULL; pmem = pmem->next, kspmem++) {
1038 		if ((caddr_t)kspmem >= (caddr_t)buf + ksp->ks_data_size)
1039 			break;
1040 		kspmem->address = pmem->address;
1041 		kspmem->size = pmem->size;
1042 	}
1043 	memlist_read_unlock();
1044 
1045 	return (0);
1046 }
1047 
1048 /*
1049  * Read a mem_name_t from user-space and store it in the mem_name_t
1050  * pointed to by the mem_name argument.
1051  */
1052 static int
1053 mm_read_mem_name(intptr_t data, mem_name_t *mem_name)
1054 {
1055 	if (get_udatamodel() == DATAMODEL_NATIVE) {
1056 		if (copyin((void *)data, mem_name, sizeof (mem_name_t)))
1057 			return (EFAULT);
1058 	}
1059 #ifdef	_SYSCALL32
1060 	else {
1061 		mem_name32_t mem_name32;
1062 
1063 		if (copyin((void *)data, &mem_name32, sizeof (mem_name32_t)))
1064 			return (EFAULT);
1065 		mem_name->m_addr = mem_name32.m_addr;
1066 		mem_name->m_synd = mem_name32.m_synd;
1067 		mem_name->m_type[0] = mem_name32.m_type[0];
1068 		mem_name->m_type[1] = mem_name32.m_type[1];
1069 		mem_name->m_name = (caddr_t)(uintptr_t)mem_name32.m_name;
1070 		mem_name->m_namelen = (size_t)mem_name32.m_namelen;
1071 		mem_name->m_sid = (caddr_t)(uintptr_t)mem_name32.m_sid;
1072 		mem_name->m_sidlen = (size_t)mem_name32.m_sidlen;
1073 	}
1074 #endif	/* _SYSCALL32 */
1075 
1076 	return (0);
1077 }
1078 
1079 /*
1080  * Read a mem_page_t from user-space and store it in the mem_page_t
1081  * pointed to by the mpage argument.
1082  */
1083 static int
1084 mm_read_mem_page(intptr_t data, mem_page_t *mpage)
1085 {
1086 	if (get_udatamodel() == DATAMODEL_NATIVE) {
1087 		if (copyin((void *)data, mpage, sizeof (mem_page_t)) != 0)
1088 			return (EFAULT);
1089 	}
1090 #ifdef _SYSCALL32
1091 	else {
1092 		mem_page32_t	mpage32;
1093 
1094 		if (copyin((void *)data, &mpage32, sizeof (mem_page32_t)) != 0)
1095 			return (EFAULT);
1096 
1097 		mpage->m_fmri = (caddr_t)(uintptr_t)mpage32.m_fmri;
1098 		mpage->m_fmrisz = mpage32.m_fmrisz;
1099 	}
1100 #endif	/* _SYSCALL32 */
1101 
1102 	return (0);
1103 }
1104 
1105 /*
1106  * Expand an FMRI from a mem_page_t.
1107  */
1108 static int
1109 mm_get_mem_fmri(mem_page_t *mpage, nvlist_t **nvl)
1110 {
1111 	char *buf;
1112 	int err;
1113 
1114 	if (mpage->m_fmri == NULL || mpage->m_fmrisz > MEM_FMRI_MAX_BUFSIZE)
1115 		return (EINVAL);
1116 
1117 	buf = kmem_alloc(mpage->m_fmrisz, KM_SLEEP);
1118 	if (copyin(mpage->m_fmri, buf, mpage->m_fmrisz) != 0) {
1119 		kmem_free(buf, mpage->m_fmrisz);
1120 		return (EFAULT);
1121 	}
1122 
1123 	err = nvlist_unpack(buf, mpage->m_fmrisz, nvl, KM_SLEEP);
1124 	kmem_free(buf, mpage->m_fmrisz);
1125 
1126 	return (err);
1127 }
1128 
1129 static int
1130 mm_get_paddr(nvlist_t *nvl, uint64_t *paddr)
1131 {
1132 	uint8_t version;
1133 	uint64_t pa;
1134 	char *scheme;
1135 	int err;
1136 #ifdef __sparc
1137 	uint64_t offset;
1138 	char *unum;
1139 	char **serids;
1140 	uint_t nserids;
1141 #endif
1142 
1143 	/* Verify FMRI scheme name and version number */
1144 	if ((nvlist_lookup_string(nvl, FM_FMRI_SCHEME, &scheme) != 0) ||
1145 	    (strcmp(scheme, FM_FMRI_SCHEME_MEM) != 0) ||
1146 	    (nvlist_lookup_uint8(nvl, FM_VERSION, &version) != 0) ||
1147 	    version > FM_MEM_SCHEME_VERSION) {
1148 		return (EINVAL);
1149 	}
1150 
1151 	/*
1152 	 * There are two ways a physical address can be  obtained from a mem
1153 	 * scheme FMRI.  One way is to use the "offset" and  "serial"
1154 	 * members, if they are present, together with the "unum" member to
1155 	 * calculate a physical address.  This is the preferred way since
1156 	 * it is independent of possible changes to the programming of
1157 	 * underlying hardware registers that may change the physical address.
1158 	 * If the "offset" member is not present, then the address is
1159 	 * retrieved from the "physaddr" member.
1160 	 */
1161 #if defined(__sparc)
1162 	if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_OFFSET, &offset) != 0) {
1163 		if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_PHYSADDR, &pa) !=
1164 		    0) {
1165 			return (EINVAL);
1166 		}
1167 	} else if (nvlist_lookup_string(nvl, FM_FMRI_MEM_UNUM, &unum) != 0 ||
1168 	    nvlist_lookup_string_array(nvl, FM_FMRI_MEM_SERIAL_ID, &serids,
1169 	    &nserids) != 0) {
1170 		return (EINVAL);
1171 	} else {
1172 		err = cpu_get_mem_addr(unum, serids[0], offset, &pa);
1173 		if (err != 0) {
1174 			if (err == ENOTSUP) {
1175 				/* Fall back to physaddr */
1176 				if (nvlist_lookup_uint64(nvl,
1177 				    FM_FMRI_MEM_PHYSADDR, &pa) != 0)
1178 					return (EINVAL);
1179 			} else
1180 				return (err);
1181 		}
1182 	}
1183 #elif defined(__x86)
1184 	if ((err = cmi_mc_unumtopa(NULL, nvl, &pa)) != CMI_SUCCESS &&
1185 	    err != CMIERR_MC_PARTIALUNUMTOPA)
1186 		return (EINVAL);
1187 #else
1188 #error "port me"
1189 #endif /* __sparc */
1190 
1191 	*paddr = pa;
1192 	return (0);
1193 }
1194