xref: /illumos-gate/usr/src/uts/common/io/mem.c (revision f3324781c875e2f9865c291e43f86ee710b0c145)
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 2007 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 	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 static int
266 mmpagelock(struct as *as, caddr_t va)
267 {
268 	struct seg *seg;
269 	int i;
270 
271 	AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
272 	seg = as_segat(as, va);
273 	i = (seg != NULL)? SEGOP_CAPABLE(seg, S_CAPABILITY_NOMINFLT) : 0;
274 	AS_LOCK_EXIT(as, &as->a_lock);
275 
276 	return (i);
277 }
278 
279 #ifdef	__sparc
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 			if ((error = plat_mem_do_mmio(uio, rw)) != ENOTSUP)
335 				break;
336 
337 			/*
338 			 * If vaddr does not map a valid page, as_pagelock()
339 			 * will return failure. Hence we can't check the
340 			 * return value and return EFAULT here as we'd like.
341 			 * seg_kp and seg_kpm do not properly support
342 			 * as_pagelock() for this context so we avoid it
343 			 * using the try_lock set check above.  Some day when
344 			 * the kernel page locking gets redesigned all this
345 			 * muck can be cleaned up.
346 			 */
347 			if (try_lock)
348 				locked = (as_pagelock(&kas, &ppp, vaddr,
349 				    PAGESIZE, S_WRITE) == 0);
350 
351 			v = hat_getpfnum(kas.a_hat,
352 			    (caddr_t)(uintptr_t)uio->uio_loffset);
353 			if (v == PFN_INVALID) {
354 				if (locked)
355 					as_pageunlock(&kas, ppp, vaddr,
356 					    PAGESIZE, S_WRITE);
357 				error = EFAULT;
358 				break;
359 			}
360 
361 			error = mmio(uio, rw, v, uio->uio_loffset & PAGEOFFSET,
362 			    minor == M_ALLKMEM || mm_kmem_io_access);
363 			if (locked)
364 				as_pageunlock(&kas, ppp, vaddr, PAGESIZE,
365 				    S_WRITE);
366 			}
367 
368 			break;
369 
370 		case M_ZERO:
371 			if (rw == UIO_READ) {
372 				label_t ljb;
373 
374 				if (on_fault(&ljb)) {
375 					no_fault();
376 					error = EFAULT;
377 					break;
378 				}
379 				uzero(iov->iov_base, iov->iov_len);
380 				no_fault();
381 				uio->uio_resid -= iov->iov_len;
382 				uio->uio_loffset += iov->iov_len;
383 				break;
384 			}
385 			/* else it's a write, fall through to NULL case */
386 			/*FALLTHROUGH*/
387 
388 		case M_NULL:
389 			if (rw == UIO_READ)
390 				return (0);
391 			c = iov->iov_len;
392 			iov->iov_base += c;
393 			iov->iov_len -= c;
394 			uio->uio_loffset += c;
395 			uio->uio_resid -= c;
396 			break;
397 
398 		}
399 	}
400 	return (uio->uio_resid == oresid ? error : 0);
401 }
402 
403 static int
404 mmread(dev_t dev, struct uio *uio, cred_t *cred)
405 {
406 	return (mmrw(dev, uio, UIO_READ, cred));
407 }
408 
409 static int
410 mmwrite(dev_t dev, struct uio *uio, cred_t *cred)
411 {
412 	return (mmrw(dev, uio, UIO_WRITE, cred));
413 }
414 
415 /*
416  * Private ioctl for libkvm to support kvm_physaddr().
417  * Given an address space and a VA, compute the PA.
418  */
419 static int
420 mmioctl_vtop(intptr_t data)
421 {
422 #ifdef _SYSCALL32
423 	mem_vtop32_t vtop32;
424 #endif
425 	mem_vtop_t mem_vtop;
426 	proc_t *p;
427 	pfn_t pfn = (pfn_t)PFN_INVALID;
428 	pid_t pid = 0;
429 	struct as *as;
430 	struct seg *seg;
431 
432 	if (get_udatamodel() == DATAMODEL_NATIVE) {
433 		if (copyin((void *)data, &mem_vtop, sizeof (mem_vtop_t)))
434 			return (EFAULT);
435 	}
436 #ifdef _SYSCALL32
437 	else {
438 		if (copyin((void *)data, &vtop32, sizeof (mem_vtop32_t)))
439 			return (EFAULT);
440 		mem_vtop.m_as = (struct as *)(uintptr_t)vtop32.m_as;
441 		mem_vtop.m_va = (void *)(uintptr_t)vtop32.m_va;
442 
443 		if (mem_vtop.m_as != NULL)
444 			return (EINVAL);
445 	}
446 #endif
447 
448 	if (mem_vtop.m_as == &kas) {
449 		pfn = hat_getpfnum(kas.a_hat, mem_vtop.m_va);
450 	} else {
451 		if (mem_vtop.m_as == NULL) {
452 			/*
453 			 * Assume the calling process's address space if the
454 			 * caller didn't specify one.
455 			 */
456 			p = curthread->t_procp;
457 			if (p == NULL)
458 				return (EIO);
459 			mem_vtop.m_as = p->p_as;
460 		}
461 
462 		mutex_enter(&pidlock);
463 		for (p = practive; p != NULL; p = p->p_next) {
464 			if (p->p_as == mem_vtop.m_as) {
465 				pid = p->p_pid;
466 				break;
467 			}
468 		}
469 		mutex_exit(&pidlock);
470 		if (p == NULL)
471 			return (EIO);
472 		p = sprlock(pid);
473 		if (p == NULL)
474 			return (EIO);
475 		as = p->p_as;
476 		if (as == mem_vtop.m_as) {
477 			mutex_exit(&p->p_lock);
478 			AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
479 			for (seg = AS_SEGFIRST(as); seg != NULL;
480 			    seg = AS_SEGNEXT(as, seg))
481 				if ((uintptr_t)mem_vtop.m_va -
482 				    (uintptr_t)seg->s_base < seg->s_size)
483 					break;
484 			if (seg != NULL)
485 				pfn = hat_getpfnum(as->a_hat, mem_vtop.m_va);
486 			AS_LOCK_EXIT(as, &as->a_lock);
487 			mutex_enter(&p->p_lock);
488 		}
489 		sprunlock(p);
490 	}
491 	mem_vtop.m_pfn = pfn;
492 	if (pfn == PFN_INVALID)
493 		return (EIO);
494 
495 	if (get_udatamodel() == DATAMODEL_NATIVE) {
496 		if (copyout(&mem_vtop, (void *)data, sizeof (mem_vtop_t)))
497 			return (EFAULT);
498 	}
499 #ifdef _SYSCALL32
500 	else {
501 		vtop32.m_pfn = mem_vtop.m_pfn;
502 		if (copyout(&vtop32, (void *)data, sizeof (mem_vtop32_t)))
503 			return (EFAULT);
504 	}
505 #endif
506 
507 	return (0);
508 }
509 
510 /*
511  * Given a PA, execute the given page retire command on it.
512  */
513 static int
514 mmioctl_page_retire(int cmd, intptr_t data)
515 {
516 	extern int page_retire_test(void);
517 	uint64_t pa;
518 
519 	if (copyin((void *)data, &pa, sizeof (uint64_t))) {
520 		return (EFAULT);
521 	}
522 
523 	switch (cmd) {
524 	case MEM_PAGE_ISRETIRED:
525 		return (page_retire_check(pa, NULL));
526 
527 	case MEM_PAGE_UNRETIRE:
528 		return (page_unretire(pa));
529 
530 	case MEM_PAGE_RETIRE:
531 		return (page_retire(pa, PR_FMA));
532 
533 	case MEM_PAGE_RETIRE_MCE:
534 		return (page_retire(pa, PR_MCE));
535 
536 	case MEM_PAGE_RETIRE_UE:
537 		return (page_retire(pa, PR_UE));
538 
539 	case MEM_PAGE_GETERRORS:
540 		{
541 			uint64_t page_errors;
542 			int rc = page_retire_check(pa, &page_errors);
543 			if (copyout(&page_errors, (void *)data,
544 			    sizeof (uint64_t))) {
545 				return (EFAULT);
546 			}
547 			return (rc);
548 		}
549 
550 	case MEM_PAGE_RETIRE_TEST:
551 		return (page_retire_test());
552 
553 	}
554 
555 	return (EINVAL);
556 }
557 
558 /*
559  * Given a mem-scheme FMRI for a page, execute the given page retire
560  * command on it.
561  */
562 static int
563 mmioctl_page_fmri_retire(int cmd, intptr_t data)
564 {
565 	mem_page_t mpage;
566 	uint64_t pa;
567 	nvlist_t *nvl;
568 	int err;
569 
570 	if ((err = mm_read_mem_page(data, &mpage)) < 0)
571 		return (err);
572 
573 	if ((err = mm_get_mem_fmri(&mpage, &nvl)) != 0)
574 		return (err);
575 
576 	if ((err = mm_get_paddr(nvl, &pa)) != 0) {
577 		nvlist_free(nvl);
578 		return (err);
579 	}
580 
581 	nvlist_free(nvl);
582 
583 	switch (cmd) {
584 	case MEM_PAGE_FMRI_ISRETIRED:
585 		return (page_retire_check(pa, NULL));
586 
587 	case MEM_PAGE_FMRI_RETIRE:
588 		return (page_retire(pa, PR_FMA));
589 	}
590 
591 	return (EINVAL);
592 }
593 
594 #ifdef __sparc
595 /*
596  * Given a syndrome, syndrome type, and address return the
597  * associated memory name in the provided data buffer.
598  */
599 static int
600 mmioctl_get_mem_name(intptr_t data)
601 {
602 	mem_name_t mem_name;
603 	void *buf;
604 	size_t bufsize;
605 	int len, err;
606 
607 	if ((bufsize = cpu_get_name_bufsize()) == 0)
608 		return (ENOTSUP);
609 
610 	if ((err = mm_read_mem_name(data, &mem_name)) < 0)
611 		return (err);
612 
613 	buf = kmem_alloc(bufsize, KM_SLEEP);
614 
615 	/*
616 	 * Call into cpu specific code to do the lookup.
617 	 */
618 	if ((err = cpu_get_mem_name(mem_name.m_synd, mem_name.m_type,
619 	    mem_name.m_addr, buf, bufsize, &len)) != 0) {
620 		kmem_free(buf, bufsize);
621 		return (err);
622 	}
623 
624 	if (len >= mem_name.m_namelen) {
625 		kmem_free(buf, bufsize);
626 		return (ENAMETOOLONG);
627 	}
628 
629 	if (copyoutstr(buf, (char *)mem_name.m_name,
630 	    mem_name.m_namelen, NULL) != 0) {
631 		kmem_free(buf, bufsize);
632 		return (EFAULT);
633 	}
634 
635 	kmem_free(buf, bufsize);
636 	return (0);
637 }
638 
639 /*
640  * Given a syndrome and address return information about the associated memory.
641  */
642 static int
643 mmioctl_get_mem_info(intptr_t data)
644 {
645 	mem_info_t mem_info;
646 	int err;
647 
648 	if (copyin((void *)data, &mem_info, sizeof (mem_info_t)))
649 		return (EFAULT);
650 
651 	if ((err = cpu_get_mem_info(mem_info.m_synd, mem_info.m_addr,
652 	    &mem_info.m_mem_size, &mem_info.m_seg_size, &mem_info.m_bank_size,
653 	    &mem_info.m_segments, &mem_info.m_banks, &mem_info.m_mcid)) != 0)
654 		return (err);
655 
656 	if (copyout(&mem_info, (void *)data, sizeof (mem_info_t)) != 0)
657 		return (EFAULT);
658 
659 	return (0);
660 }
661 
662 /*
663  * Given a memory name, return its associated serial id
664  */
665 static int
666 mmioctl_get_mem_sid(intptr_t data)
667 {
668 	mem_name_t mem_name;
669 	void *buf;
670 	void *name;
671 	size_t	name_len;
672 	size_t bufsize;
673 	int len, err;
674 
675 	if ((bufsize = cpu_get_name_bufsize()) == 0)
676 		return (ENOTSUP);
677 
678 	if ((err = mm_read_mem_name(data, &mem_name)) < 0)
679 		return (err);
680 
681 	buf = kmem_alloc(bufsize, KM_SLEEP);
682 
683 	if (mem_name.m_namelen > 1024)
684 		mem_name.m_namelen = 1024; /* cap at 1024 bytes */
685 
686 	name = kmem_alloc(mem_name.m_namelen, KM_SLEEP);
687 
688 	if ((err = copyinstr((char *)mem_name.m_name, (char *)name,
689 	    mem_name.m_namelen, &name_len)) != 0) {
690 		kmem_free(buf, bufsize);
691 		kmem_free(name, mem_name.m_namelen);
692 		return (err);
693 	}
694 
695 	/*
696 	 * Call into cpu specific code to do the lookup.
697 	 */
698 	if ((err = cpu_get_mem_sid(name, buf, bufsize, &len)) != 0) {
699 		kmem_free(buf, bufsize);
700 		kmem_free(name, mem_name.m_namelen);
701 		return (err);
702 	}
703 
704 	if (len > mem_name.m_sidlen) {
705 		kmem_free(buf, bufsize);
706 		kmem_free(name, mem_name.m_namelen);
707 		return (ENAMETOOLONG);
708 	}
709 
710 	if (copyoutstr(buf, (char *)mem_name.m_sid,
711 	    mem_name.m_sidlen, NULL) != 0) {
712 		kmem_free(buf, bufsize);
713 		kmem_free(name, mem_name.m_namelen);
714 		return (EFAULT);
715 	}
716 
717 	kmem_free(buf, bufsize);
718 	kmem_free(name, mem_name.m_namelen);
719 	return (0);
720 }
721 #endif	/* __sparc */
722 
723 /*
724  * Private ioctls for
725  *	libkvm to support kvm_physaddr().
726  *	FMA support for page_retire() and memory attribute information.
727  */
728 /*ARGSUSED*/
729 static int
730 mmioctl(dev_t dev, int cmd, intptr_t data, int flag, cred_t *cred, int *rvalp)
731 {
732 	if ((cmd == MEM_VTOP && getminor(dev) != M_KMEM) ||
733 	    (cmd != MEM_VTOP && getminor(dev) != M_MEM))
734 		return (ENXIO);
735 
736 	switch (cmd) {
737 	case MEM_VTOP:
738 		return (mmioctl_vtop(data));
739 
740 	case MEM_PAGE_RETIRE:
741 	case MEM_PAGE_ISRETIRED:
742 	case MEM_PAGE_UNRETIRE:
743 	case MEM_PAGE_RETIRE_MCE:
744 	case MEM_PAGE_RETIRE_UE:
745 	case MEM_PAGE_GETERRORS:
746 	case MEM_PAGE_RETIRE_TEST:
747 		return (mmioctl_page_retire(cmd, data));
748 
749 	case MEM_PAGE_FMRI_RETIRE:
750 	case MEM_PAGE_FMRI_ISRETIRED:
751 		return (mmioctl_page_fmri_retire(cmd, data));
752 
753 #ifdef __sparc
754 	case MEM_NAME:
755 		return (mmioctl_get_mem_name(data));
756 
757 	case MEM_INFO:
758 		return (mmioctl_get_mem_info(data));
759 
760 	case MEM_SID:
761 		return (mmioctl_get_mem_sid(data));
762 #else
763 	case MEM_NAME:
764 	case MEM_INFO:
765 	case MEM_SID:
766 		return (ENOTSUP);
767 #endif	/* __sparc */
768 	}
769 	return (ENXIO);
770 }
771 
772 /*ARGSUSED2*/
773 static int
774 mmmmap(dev_t dev, off_t off, int prot)
775 {
776 	pfn_t pf;
777 	struct memlist *pmem;
778 	minor_t minor = getminor(dev);
779 
780 	switch (minor) {
781 	case M_MEM:
782 		pf = btop(off);
783 		memlist_read_lock();
784 		for (pmem = phys_install; pmem != NULL; pmem = pmem->next) {
785 			if (pf >= BTOP(pmem->address) &&
786 			    pf < BTOP(pmem->address + pmem->size)) {
787 				memlist_read_unlock();
788 				return (impl_obmem_pfnum(pf));
789 			}
790 		}
791 		memlist_read_unlock();
792 		break;
793 
794 	case M_KMEM:
795 	case M_ALLKMEM:
796 		/* no longer supported with KPR */
797 		return (-1);
798 
799 	case M_ZERO:
800 		/*
801 		 * We shouldn't be mmap'ing to /dev/zero here as
802 		 * mmsegmap() should have already converted
803 		 * a mapping request for this device to a mapping
804 		 * using seg_vn for anonymous memory.
805 		 */
806 		break;
807 
808 	}
809 	return (-1);
810 }
811 
812 /*
813  * This function is called when a memory device is mmap'ed.
814  * Set up the mapping to the correct device driver.
815  */
816 static int
817 mmsegmap(dev_t dev, off_t off, struct as *as, caddr_t *addrp, off_t len,
818     uint_t prot, uint_t maxprot, uint_t flags, struct cred *cred)
819 {
820 	struct segvn_crargs vn_a;
821 	struct segdev_crargs dev_a;
822 	int error;
823 	minor_t minor;
824 	off_t i;
825 
826 	minor = getminor(dev);
827 
828 	as_rangelock(as);
829 	if ((flags & MAP_FIXED) == 0) {
830 		/*
831 		 * No need to worry about vac alignment on /dev/zero
832 		 * since this is a "clone" object that doesn't yet exist.
833 		 */
834 		map_addr(addrp, len, (offset_t)off,
835 		    (minor == M_MEM) || (minor == M_KMEM), flags);
836 
837 		if (*addrp == NULL) {
838 			as_rangeunlock(as);
839 			return (ENOMEM);
840 		}
841 	} else {
842 		/*
843 		 * User specified address -
844 		 * Blow away any previous mappings.
845 		 */
846 		(void) as_unmap(as, *addrp, len);
847 	}
848 
849 	switch (minor) {
850 	case M_MEM:
851 		/* /dev/mem cannot be mmap'ed with MAP_PRIVATE */
852 		if ((flags & MAP_TYPE) != MAP_SHARED) {
853 			as_rangeunlock(as);
854 			return (EINVAL);
855 		}
856 
857 		/*
858 		 * Check to ensure that the entire range is
859 		 * legal and we are not trying to map in
860 		 * more than the device will let us.
861 		 */
862 		for (i = 0; i < len; i += PAGESIZE) {
863 			if (mmmmap(dev, off + i, maxprot) == -1) {
864 				as_rangeunlock(as);
865 				return (ENXIO);
866 			}
867 		}
868 
869 		/*
870 		 * Use seg_dev segment driver for /dev/mem mapping.
871 		 */
872 		dev_a.mapfunc = mmmmap;
873 		dev_a.dev = dev;
874 		dev_a.offset = off;
875 		dev_a.type = (flags & MAP_TYPE);
876 		dev_a.prot = (uchar_t)prot;
877 		dev_a.maxprot = (uchar_t)maxprot;
878 		dev_a.hat_attr = 0;
879 
880 		/*
881 		 * Make /dev/mem mappings non-consistent since we can't
882 		 * alias pages that don't have page structs behind them,
883 		 * such as kernel stack pages. If someone mmap()s a kernel
884 		 * stack page and if we give him a tte with cv, a line from
885 		 * that page can get into both pages of the spitfire d$.
886 		 * But snoop from another processor will only invalidate
887 		 * the first page. This later caused kernel (xc_attention)
888 		 * to go into an infinite loop at pil 13 and no interrupts
889 		 * could come in. See 1203630.
890 		 *
891 		 */
892 		dev_a.hat_flags = HAT_LOAD_NOCONSIST;
893 		dev_a.devmap_data = NULL;
894 
895 		error = as_map(as, *addrp, len, segdev_create, &dev_a);
896 		break;
897 
898 	case M_ZERO:
899 		/*
900 		 * Use seg_vn segment driver for /dev/zero mapping.
901 		 * Passing in a NULL amp gives us the "cloning" effect.
902 		 */
903 		vn_a.vp = NULL;
904 		vn_a.offset = 0;
905 		vn_a.type = (flags & MAP_TYPE);
906 		vn_a.prot = prot;
907 		vn_a.maxprot = maxprot;
908 		vn_a.flags = flags & ~MAP_TYPE;
909 		vn_a.cred = cred;
910 		vn_a.amp = NULL;
911 		vn_a.szc = 0;
912 		vn_a.lgrp_mem_policy_flags = 0;
913 		error = as_map(as, *addrp, len, segvn_create, &vn_a);
914 		break;
915 
916 	case M_KMEM:
917 	case M_ALLKMEM:
918 		/* No longer supported with KPR. */
919 		error = ENXIO;
920 		break;
921 
922 	case M_NULL:
923 		/*
924 		 * Use seg_dev segment driver for /dev/null mapping.
925 		 */
926 		dev_a.mapfunc = mmmmap;
927 		dev_a.dev = dev;
928 		dev_a.offset = off;
929 		dev_a.type = 0;		/* neither PRIVATE nor SHARED */
930 		dev_a.prot = dev_a.maxprot = (uchar_t)PROT_NONE;
931 		dev_a.hat_attr = 0;
932 		dev_a.hat_flags = 0;
933 		error = as_map(as, *addrp, len, segdev_create, &dev_a);
934 		break;
935 
936 	default:
937 		error = ENXIO;
938 	}
939 
940 	as_rangeunlock(as);
941 	return (error);
942 }
943 
944 static struct cb_ops mm_cb_ops = {
945 	mmopen,			/* open */
946 	nulldev,		/* close */
947 	nodev,			/* strategy */
948 	nodev,			/* print */
949 	nodev,			/* dump */
950 	mmread,			/* read */
951 	mmwrite,		/* write */
952 	mmioctl,		/* ioctl */
953 	nodev,			/* devmap */
954 	mmmmap,			/* mmap */
955 	mmsegmap,		/* segmap */
956 	mmchpoll,		/* poll */
957 	mmpropop,		/* prop_op */
958 	0,			/* streamtab  */
959 	D_NEW | D_MP | D_64BIT | D_U64BIT
960 };
961 
962 static struct dev_ops mm_ops = {
963 	DEVO_REV,		/* devo_rev, */
964 	0,			/* refcnt  */
965 	mm_info,		/* get_dev_info */
966 	nulldev,		/* identify */
967 	nulldev,		/* probe */
968 	mm_attach,		/* attach */
969 	nodev,			/* detach */
970 	nodev,			/* reset */
971 	&mm_cb_ops,		/* driver operations */
972 	(struct bus_ops *)0	/* bus operations */
973 };
974 
975 static struct modldrv modldrv = {
976 	&mod_driverops, "memory driver %I%", &mm_ops,
977 };
978 
979 static struct modlinkage modlinkage = {
980 	MODREV_1, &modldrv, NULL
981 };
982 
983 int
984 _init(void)
985 {
986 	return (mod_install(&modlinkage));
987 }
988 
989 int
990 _info(struct modinfo *modinfop)
991 {
992 	return (mod_info(&modlinkage, modinfop));
993 }
994 
995 int
996 _fini(void)
997 {
998 	return (mod_remove(&modlinkage));
999 }
1000 
1001 static int
1002 mm_kstat_update(kstat_t *ksp, int rw)
1003 {
1004 	struct memlist *pmem;
1005 	uint_t count;
1006 
1007 	if (rw == KSTAT_WRITE)
1008 		return (EACCES);
1009 
1010 	count = 0;
1011 	memlist_read_lock();
1012 	for (pmem = phys_install; pmem != NULL; pmem = pmem->next) {
1013 		count++;
1014 	}
1015 	memlist_read_unlock();
1016 
1017 	ksp->ks_ndata = count;
1018 	ksp->ks_data_size = count * 2 * sizeof (uint64_t);
1019 
1020 	return (0);
1021 }
1022 
1023 static int
1024 mm_kstat_snapshot(kstat_t *ksp, void *buf, int rw)
1025 {
1026 	struct memlist *pmem;
1027 	struct memunit {
1028 		uint64_t address;
1029 		uint64_t size;
1030 	} *kspmem;
1031 
1032 	if (rw == KSTAT_WRITE)
1033 		return (EACCES);
1034 
1035 	ksp->ks_snaptime = gethrtime();
1036 
1037 	kspmem = (struct memunit *)buf;
1038 	memlist_read_lock();
1039 	for (pmem = phys_install; pmem != NULL; pmem = pmem->next, kspmem++) {
1040 		if ((caddr_t)kspmem >= (caddr_t)buf + ksp->ks_data_size)
1041 			break;
1042 		kspmem->address = pmem->address;
1043 		kspmem->size = pmem->size;
1044 	}
1045 	memlist_read_unlock();
1046 
1047 	return (0);
1048 }
1049 
1050 /*
1051  * Read a mem_name_t from user-space and store it in the mem_name_t
1052  * pointed to by the mem_name argument.
1053  */
1054 static int
1055 mm_read_mem_name(intptr_t data, mem_name_t *mem_name)
1056 {
1057 	if (get_udatamodel() == DATAMODEL_NATIVE) {
1058 		if (copyin((void *)data, mem_name, sizeof (mem_name_t)))
1059 			return (EFAULT);
1060 	}
1061 #ifdef	_SYSCALL32
1062 	else {
1063 		mem_name32_t mem_name32;
1064 
1065 		if (copyin((void *)data, &mem_name32, sizeof (mem_name32_t)))
1066 			return (EFAULT);
1067 		mem_name->m_addr = mem_name32.m_addr;
1068 		mem_name->m_synd = mem_name32.m_synd;
1069 		mem_name->m_type[0] = mem_name32.m_type[0];
1070 		mem_name->m_type[1] = mem_name32.m_type[1];
1071 		mem_name->m_name = (caddr_t)(uintptr_t)mem_name32.m_name;
1072 		mem_name->m_namelen = (size_t)mem_name32.m_namelen;
1073 		mem_name->m_sid = (caddr_t)(uintptr_t)mem_name32.m_sid;
1074 		mem_name->m_sidlen = (size_t)mem_name32.m_sidlen;
1075 	}
1076 #endif	/* _SYSCALL32 */
1077 
1078 	return (0);
1079 }
1080 
1081 /*
1082  * Read a mem_page_t from user-space and store it in the mem_page_t
1083  * pointed to by the mpage argument.
1084  */
1085 static int
1086 mm_read_mem_page(intptr_t data, mem_page_t *mpage)
1087 {
1088 	if (get_udatamodel() == DATAMODEL_NATIVE) {
1089 		if (copyin((void *)data, mpage, sizeof (mem_page_t)) != 0)
1090 			return (EFAULT);
1091 	}
1092 #ifdef _SYSCALL32
1093 	else {
1094 		mem_page32_t	mpage32;
1095 
1096 		if (copyin((void *)data, &mpage32, sizeof (mem_page32_t)) != 0)
1097 			return (EFAULT);
1098 
1099 		mpage->m_fmri = (caddr_t)(uintptr_t)mpage32.m_fmri;
1100 		mpage->m_fmrisz = mpage32.m_fmrisz;
1101 	}
1102 #endif	/* _SYSCALL32 */
1103 
1104 	return (0);
1105 }
1106 
1107 /*
1108  * Expand an FMRI from a mem_page_t.
1109  */
1110 static int
1111 mm_get_mem_fmri(mem_page_t *mpage, nvlist_t **nvl)
1112 {
1113 	char *buf;
1114 	int err;
1115 
1116 	if (mpage->m_fmri == NULL || mpage->m_fmrisz > MEM_FMRI_MAX_BUFSIZE)
1117 		return (EINVAL);
1118 
1119 	buf = kmem_alloc(mpage->m_fmrisz, KM_SLEEP);
1120 	if (copyin(mpage->m_fmri, buf, mpage->m_fmrisz) != 0) {
1121 		kmem_free(buf, mpage->m_fmrisz);
1122 		return (EFAULT);
1123 	}
1124 
1125 	err = nvlist_unpack(buf, mpage->m_fmrisz, nvl, KM_SLEEP);
1126 	kmem_free(buf, mpage->m_fmrisz);
1127 
1128 	return (err);
1129 }
1130 
1131 static int
1132 mm_get_paddr(nvlist_t *nvl, uint64_t *paddr)
1133 {
1134 	uint8_t version;
1135 	uint64_t pa;
1136 	char *scheme;
1137 #ifdef __sparc
1138 	uint64_t offset;
1139 	char *unum;
1140 	char **serids;
1141 	uint_t nserids;
1142 	int err;
1143 #endif
1144 
1145 	/* Verify FMRI scheme name and version number */
1146 	if ((nvlist_lookup_string(nvl, FM_FMRI_SCHEME, &scheme) != 0) ||
1147 	    (strcmp(scheme, FM_FMRI_SCHEME_MEM) != 0) ||
1148 	    (nvlist_lookup_uint8(nvl, FM_VERSION, &version) != 0) ||
1149 	    version > FM_MEM_SCHEME_VERSION) {
1150 		return (EINVAL);
1151 	}
1152 
1153 	/*
1154 	 * There are two ways a physical address can be  obtained from a mem
1155 	 * scheme FMRI.  One way is to use the "offset" and  "serial"
1156 	 * members, if they are present, together with the "unum" member to
1157 	 * calculate a physical address.  This is the preferred way since
1158 	 * it is independent of possible changes to the programming of
1159 	 * underlying hardware registers that may change the physical address.
1160 	 * If the "offset" member is not present, then the address is
1161 	 * retrieved from the "physaddr" member.
1162 	 */
1163 #if defined(__sparc)
1164 	if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_OFFSET, &offset) != 0) {
1165 		if (nvlist_lookup_uint64(nvl, FM_FMRI_MEM_PHYSADDR, &pa) !=
1166 		    0) {
1167 			return (EINVAL);
1168 		}
1169 	} else if (nvlist_lookup_string(nvl, FM_FMRI_MEM_UNUM, &unum) != 0 ||
1170 	    nvlist_lookup_string_array(nvl, FM_FMRI_MEM_SERIAL_ID, &serids,
1171 	    &nserids) != 0) {
1172 		return (EINVAL);
1173 	} else {
1174 		err = cpu_get_mem_addr(unum, serids[0], offset, &pa);
1175 		if (err != 0) {
1176 			if (err == ENOTSUP) {
1177 				/* Fall back to physaddr */
1178 				if (nvlist_lookup_uint64(nvl,
1179 				    FM_FMRI_MEM_PHYSADDR, &pa) != 0)
1180 					return (EINVAL);
1181 			} else
1182 				return (err);
1183 		}
1184 	}
1185 #elif defined(__x86)
1186 	if (cmi_mc_unumtopa(NULL, nvl, &pa) == 0)
1187 		return (EINVAL);
1188 #else
1189 #error "port me"
1190 #endif /* __sparc */
1191 
1192 	*paddr = pa;
1193 	return (0);
1194 }
1195